Analysis of Functions of a Single Variable

By: Lawrence Baggett

Analysis of Functions of a Single Variable

By: Lawrence Baggett

Online: < http://cnx.org/content/col11249/1.1/ >

CONNEXIONS
Rice University, Houston, Texas

This selection and arrangement of content as a collection is copyrighted by Lawrence Baggett. It is licensed under the Creative Commons Attribution 3.0 license (http://creativecommons.org/licenses/by/3.0/). Collection structure revised: December 11, 2010 PDF generated: October 29, 2012 For copyright and attribution information for the modules contained in this collection, see p. 240.

Table of Contents
Preface to Analysis of Functions of a Single Variable: A Detailed Development
................... 1

1 The Real and Complex Numbers 1.1 Denition of the Numbers 1, i, and the square root of 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.2 The Natural Numbers and the Integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.3 The Rational Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.4 The Real Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.5 Properties of the Real Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.6 Intervals and Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 1.7 The Geometric Progression and the Binomial Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 1.8 The Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2 The Limit of a Sequence of Numbers 2.1 Denition of the Number e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 2.2 Sequences and Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 2.3 Existence of Certain Fundamental Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 2.4 Denition of e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 2.5 Properties of Convergent Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 2.6 Subsequences and Cluster Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 2.7 A Little Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 2.8 Innite Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 3 Functions and Continuity 3.1 Functions and Continuity Denition of the Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 3.2 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 3.3 Polynomial Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 3.4 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 3.5 Continuity and Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 3.6 Deeper Analytic Properties of Continuous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 3.7 Power Series Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 3.8 The Elementary Transcendental Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 3.9 Analytic Functions and Taylor Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 3.10 Uniform Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 4 Dierentiation, Local Behavior 4.1 Dierentiation, Local Behavior E^i = -1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 4.2 The Limit of a Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 4.3 The Derivative of a Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 4.4 Consequences of Dierentiability, the Mean Value Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 4.5 The Exponential and Logarithm Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 4.6 The Trigonometric and Hyperbolic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 4.7 L'Hopital's Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 4.8 Higher Order Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 4.9 Taylor Polynomials and Taylor's Remainder Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 4.10 The General Binomial Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 4.11 More on Partial Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 5 Integration, Average Behavior 5.1 Integration, Average Behavior A= r^2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 117 5.2 Integrals of Step Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

iv

5.3 Integrable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 5.4 The Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 5.5 Consequences of the Fundamental Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 5.6 Area of Regions in the Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 5.7 Extending the Denition of Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140 5.8 Integration in the Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 6 Integration over Smooth Curves in the Plane 6.1 Integration Over Smooth Curves in the Plane C=2 r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 6.2 Smooth Curves in the Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 6.3 Arc Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 6.4 Integration with Respect to Arc Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164 6.5 Contour Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 168 6.6 Vector Fields, Dierential Forms, and Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 6.7 Integration Around Closed Curves, and Green's Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174 7 The Fundamental Theorem of Algebra, and The Fundamental Theorem of Analysis 7.1 The Fundamental Theorem of Algebra, and the Fundamental Theorem of Analy7.2 Cauchy's Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 186 7.3 Basic Applications of the Cauchy Integral Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192 7.4 The Fundamental Theorem of Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 7.5 The Maximum Modulus Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 7.6 The Open Mapping Theorem and the Inverse Function Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 7.7 Uniform Convergence of Analytic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 7.8 Isolated Singularities, and the Residue Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 8 Appendix: Existence and Uniqueness of a Complete Ordered Field . . . . . . . . . . . . . . . . . . . . . . . 211 Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 Attributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
sis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

Preface to Analysis of Functions of a Single Variable: A Detailed Development

1

For Christy My Light I have written this book primarily for serious and talented mathematics scholars, seniors or rst-year graduate students, who by the time they nish their schooling should have had the opportunity to study in some detail the great discoveries of our subject. What did we know and how and when did we know it? I hope this book is useful toward that goal, especially when it comes to the great achievements of that part of mathematics known as analysis. I have tried to write a complete and thorough account of the elementary theories of functions of a single real variable and functions of a single complex variable. Separating these two subjects does not at all jive with their development historically, and to me it seems unnecessary and potentially confusing to do so. On the other hand, functions of several variables seems to me to be a very dierent kettle of sh, so I have decided to limit this book by concentrating on one variable at a time. Everyone is taught (told) in school that the area of a circle is given by the formula 2 is irrational. Students of natural sciences learn that

A = r2 .

We are also

told that the product of two negatives is a positive, that you cant trisect an angle, and that the square root of

ei = 1 and that sin2 + cos2 = 1. More sophisticated

students are taught the Fundamental Theorem of calculus and the Fundamental Theorem of Algebra. Some are also told that it is impossible to solve a general fth degree polynomial equation by radicals. On the other hand, very few people indeed have the opportunity to nd out precisely why these things are really true, and at the same time to realize just how intellectually deep and profound these facts are. Indeed, we mathematicians believe that these facts are among the most marvelous accomplishments of the human mind. Engineers and scientists can and do commit such mathematical facts to memory, and quite often combine them to useful purposes. However, it is left to us mathematicians to share the basic knowledge of why and how, and happily to us this is more a privilege than a chore. A large part of what makes the verication of such simple sounding and elementary truths so dicult is that we of necessity must spend quite a lot of energy determining what the relevant words themselves really mean. That is, to be quite careful about studying mathematics, we need to ask very basic questions: What is a circle? What are numbers? What is the denition of the area of a set in the Euclidean plane? What is the precise denition of numbers like

,i,

and

e?

We surely cannot prove that

ei = 1

without a clear denition of these particular numbers.

The mathematical analysis story is a long one, beginning with the early civilizations, and in some sense only coming to a satisfactory completion in the late nineteenth century. It is a story of ideas, well worth learning. There are many many fantastic mathematical truths (facts), and it seems to me that some of them are so beautiful and fundamental to human intellectual development, that a student who wants to be called a mathematician, ought to know how to explain them, or at the very least should have known how to explain them at some point. Each professor might make up a slightly dierent list of such truths. Here is mine: 1. The square root of 2 is a real number but is not a rational number.

1 This

content is available online at <http://cnx.org/content/m36084/1.3/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

2. The formula for the area of a circle of radius 4.

is

3. The formula for the circumference of a circle of radius

A = r2 . r is C = 2r.

5. The Fundamental Theorem of Calculus, complex numbers.

b f (t) dt = F (b) F (a) . a 6. The Fundamental Theorem of Algebra, every nonconstant polynomial has at least one root in the
7. It is impossible to trisect an arbitrary angle using only a compass and straight edge.

ei = 1.

Other mathematical marvels, such as the fact that there are more real numbers than there are rationals, the set of all sets is not a set, an arbitrary fth degree polynomial equation can not be solved in terms of radicals, a simple closed curve divides the plain into exactly two components, there are an innite number of primes, etc., are clearly wonderful results, but the seven in the list above are really of a more primary nature to me, an analyst, for they stem from the work of ancient mathematicians and except for number 7, which continues to this day to evoke so-called disproofs, have been accepted as true by most people even in the absence of precise arguments for hundreds if not thousands of years. Perhaps one should ruminate on why it took so long for us to formulate precise denitions of things like numbers and areas? Only with the advent of calculus in the seventeenth century, together with the contributions of people like Euler, Cauchy, and Weierstrass during the next two hundred years, were the rst six items above really proved, and only with the contributions of Galois in the early nineteenth century was the last one truly understood. This text, while including a traditional treatment of introductory analysis, specically addresses, as kinds of milestones, the rst six of these truths and gives careful derivations of them. The seventh, which looks like an assertion from geometry, turns out to be an algebraic result that is not appropriate for this course in analysis, but in my opinion it should denitely be presented in an undergraduate algebra course. As for the rst six, I insist here on developing precise mathematical denitions of all the relevant notions, and moving

step by step through their derivations. Specically, what are the denitions of

2, A, , r, r2 , C, 2, e, i, ,

and

1? My feeling is that mathematicians should understand exactly where these concepts come from in precise
mathematical terms, why it took so long to discover these denitions, and why the various relations among them hold. The numbers

1, 2,

and

can be disposed of fairly quickly by a discussion of what exactly is meant by

the real and complex number systems. Of course, this is in fact no trivial matter, having had to wait until the end of the nineteenth century for a clear explanation, and in fact I leave the actual proof of the existence of the real numbers to an appendix. However, a complete mathematics education ought to include a study of this proof, and if one nds the time in this analysis course, it really should be included here. Having a denition of the real numbers to work with, i.e., having introduced the notion of least upper bound, one can relatively easily prove that there is a real number whose square is 2, and that this number can not be a rational number, thereby disposing of the rst of our goals. All this is done in Section 1.1. Maintaining the attitude that we should not distinguish between functions of a real variable and functions of a complex variable, at least at the beginning of the development, Section 1.1 concludes with a careful introduction of the basic properties of the eld of complex numbers. unlike the elementary numbers

1, 2,

and

i,

the denitions of the real numbers

and

are quite a

dierent story. In fact, one cannot make sense of either to dene

or

until a substantial amount of analysis has

1 n }, in some ways the rst nontrivial n example of a convergent sequence, and this is presented in Section 2.1. Its relation to logarithms and

been developed, for they both are necessarily dened somehow in terms of a limit process. I have chosen

here as the limit of the rather intriguing sequence

{ 1+

exponentials, whatever they are, has to be postponed to Section 4.1. Section 2.1 also contains a section on the elementary topological properties (compactness, limit points, etc.) of the real and complex numbers as well as a thorough development of innite series. To dene

as the ratio of the circumference of a circle to its diameter is attractive, indeed was quite

acceptable to Euclid, but is dangerously imprecise unless we have at the outset a clear denition of what is meant by the length of a curve, e.g., the circumference of a circle. That notion is by no means trivial, and in fact it only can be carefully treated in a development of analysis well after other concepts. Rather, I have

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

chosen to dene

here as the smallest positive zero of the sine function. Of course, I have to dene the

sine function rst, and this is itself quite deep. I do it using power series functions, choosing to avoid the common denition of the trigonometric functions in terms of  wrapping the real line around a circle, for that notion again requires a precise denition of arc length before it would make sense. I get to arc length eventually, but not until Section 6.1. In Section 3.1 I introduce power series functions as generalizations of polynomials, specically the three power series functions that turn out to be the exponential, sine, and cosine functions. From these denitions it follows directly that

expiz = cosz + isinz

for every complex number

z.

Here is a place where allowing the

variable to be complex is critical, and it has cost us nothing. However, even after establishing that there is in fact a smallest positive zero of the sine function (which we decide to call things to work out), one cannot at this point deduce that of the function

since we know how we want

cos = 1,

so that the equality

ei = 1

also has

to wait for its derivation until Section 4.1. In fact, more serious, we have no knowledge at all at this point

ez

for a complex exponent

z.

What does it mean to raise a real number, or even an integer,

to a complex exponent? The very denition of such a function has to wait. Section 3.1 also contains all the standard theorems about continuous functions, culminating with a lengthy section on uniform convergence, and nally Abel's fantastic theorem on the continuity of a power series function on the boundary of its disk of convergence. The fourth chapter begins with all the usual theorems from calculus, Mean Value Theorem, Chain Rule, First Derivative Test, and so on. Power series functions are shown to be dierentiable, from which the Immediately then, all of the trigonometric for every rational number law of exponents emerges for the power series function exp. and exponential identities are also derived. We observe that we at last can dene consistently number

er = exp (r)

r,

and

ez

to be the value of the power series function

exp (z )

for any complex

z. From that, we establish the equation ei = 1. Careful proofs of Taylor's Remainder Theorem and

L'Hopital's Rule are given, as well as an initial approach to the general Binomial Theorem for non-integer exponents. It is in Section 4.1 that the rst glimpse of a dierence between functions of a real variable and functions of a complex variable emerges. For example, one of the results in this chapter is that every dierentiable, real-valued function of a complex variable must be a constant function, something that is certainly not true for functions of a real variable. At the end of this chapter, I briey slip into the realm of real-valued functions of two real variables. I introduce the denition of dierentiability of such a function of two real variables, and then derive the initial relationships among the partial derivatives of such a function and the derivative of that function thought of as a function of a complex variable. This is obviously done in preparation for Chapter VII where holomorphic functions are central. Perhaps most well-understood by math majors is that computing the area under a curve requires Newton's calculus, i.e., integration theory. What is often overlooked by students is that the very denition of the concept of area is intimately tied up with this integration theory. My treatment here of integration diers from most others in that the class of functions dened as integrable are those that are uniform limits of step functions. This is a smaller collection of functions than those that are Riemann-integrable, but they suce for my purposes, and this approach serves to emphasize the importance of uniform convergence. In particular, I include careful proofs of the Fundamental Theorem of Calculus, the integration by substitution theorem, the integral form of Taylor's Remainder Theorem, and the complete proof of the general Binomial Theorem. Not wishing to delve into the set-theoretic complications of measure theory, I have chosen only to dene the area for certain geometric subsets of the plane. examples of such geometric sets, so that the formula of radius variables. Having developed the notions of arc length in the early part of Section 6.1, including the derivation of the formula for the circumference of a circle, I introduce the idea of a contour integral, i.e., integrating a These are those subsets bounded above and below by graphs of continuous functions. Of course these suce for most purposes, and in particular circles are

A = r2

can be established for the area of a circle

r.

Section 5.1 concludes with a development of integration over geometric subsets of the plane.

Once again, anticipating later needs, we have again strayed into some investigations of functions of two real

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

function around a curve in the complex plane. The Fundamental Theorem of Calculus has generalizations to higher dimensions, and it becomes Green's Theorem in 2 dimensions. I give a careful proof in Section 6.1, just over geometric sets, of this rather complicated theorem. Perhaps the main application of Green's Theorem is the Cauchy Integral Theorem, a result about complex-valued functions of a complex variable, that is often called the Fundamental Theorem of Analysis. I prove this theorem in Section 7.1. From this Cauchy theorem one can deduce the usual marvelous theorems of a rst course in complex variables, e.g., the Identity Theorem, Liouville's Theorem, the Maximum Modulus Principle, the Open Mapping Theorem, the Residue Theorem, and last but not least our mathematical truth number 6, the Fundamental Theorem of Algebra. That so much mathematical analysis is used to prove the fundamental theorem of algebra does make me smile. I will leave it to my algebraist colleagues to point out how some of the fundamental results in analysis require substantial algebraic arguments. The overriding philosophical point of this book is that many analytic assertions in mathematics are intellectually very deep; they require years of study for most people to understand; they demonstrate how intricate mathematical thought is and how far it has come over the years. Graduates in mathematics should be proud of the degree they have earned, and they should be proud of the depth of their understanding and the extremes to which they have pushed their own intellect. I love teaching these students, that is to say, I love sharing this marvelous material with them.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

Chapter 1

The Real and Complex Numbers

1.1 Denition of the Numbers 1, i, and the square root of 2
1
In order to make precise sense out of the concepts we study in mathematical analysis, we must rst come to terms with what the "real numbers" are. Everything in mathematical analysis is based on these numbers, and their very denition and existence is quite deep. We will, in fact, not attempt to demonstrate (prove) the existence of the real numbers in the body of this text, but will content ourselves with a careful delineation of their properties, referring the interested reader to an appendix for the existence and uniqueness proofs. Although people may always have had an intuitive idea of what these real numbers were, it was not until the nineteenth century that mathematically precise denitions were given. The history of how mathematicians came to realize the necessity for such precision in their denitions is fascinating from a philosophical point of view as much as from a mathematical one. However, we will not pursue the philosophical aspects of the subject in this book, but will be content to concentrate our attention just on the mathematical facts. These precise denitions are quite complicated, but the powerful possibilities within mathematical analysis rely heavily on this precision, so we must pursue them. Toward our primary goals, we will in this chapter give denitions of the symbols (numbers)

1, i,

and

2.

The main points of this chapter are the following: 1. The notions of numbers,

least upper bound (supremum) and greatest lower bound (inmum) of a set of
19),

2. The denition of the 4. 5.

3. the formula for the sum of a

real numbersR, geometric progression (Theorem 1.9, Geometric Progression, p. the Binomial Theorem (Theorem 1.10, p. 20), and the triangle inequality for complex numbers (Theorem 1.15, Triangle Inequality, p. 26).
2

1.2 The Natural Numbers and the Integers

set

We will take for granted that we understand the existence of what we call the

natural numbers,

i.e., the

whose elements are the numbers

1, 2, 3, 4, ....

Indeed, the two salient properties of this set are that

(a) there is a frist element (the natural number 1), and (b) for each element

of this set there is a very

next one, i.e., an immediate successor. We assume that the algebraic notions of sum and product of natural numbers is well-dened and familiar. These operations satisfy three basic relations:

Basic Algebraic Relations.

1. (Commutativity)

n+m=m+n

and

nm=mn

for all

n, m N.

1 This 2 This

content is available online at <http://cnx.org/content/m36082/1.3/>. content is available online at <http://cnx.org/content/m36075/1.2/>. Available for free at Connexions <http://cnx.org/content/col11249/1.1>

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

2. (Associativity) 3. (Distributivity)

n + (m + k ) = (n + m) + k and n (m k ) = (n m) k n (m + k ) = n m + n k for all n, m, k N.

for all

n, m, k N.

We also take as given the notion of one natural number being larger than another one. etc. We will accept as true the

1.1: AXIOM OF MATHEMATICAL INDUCTION. Let S be a subset of the set N of natural

numbers. Suppose that
1. 2.

axiom of mathematical induction, that is, the following statement:

2 > 1,5 > 3,n +1 > n,

If a natural number k is in S, then the natural number k + 1 also is in S.

n
belongs to

1 S.

Then S = N.
That is, every natural number

S.

1.2: REMARK

The axiom of mathematical induction is for our purposes frequently employed as That is, if we wish to show that a certain proposition holds for all natural

a method of proof. numbers, then we let

denote the set of numbers for which the proposition is true, and then, using

the axiom of mathematical induction, we verify that using it, we can dene an innite number of objects Think of

is all of

by showing that

satises both of

the above conditions. Mathematical induction can also be used as a method of denition. That is,

{On } that are indexed by the natural numbers. On is dened. We check rst to see that the object O1 is dened. We check next that, if the object Ok is dened for a natural number k, then there is a prescribed procedure for dening the object Ok+1 . So, by the axiom of S
as the set of natural numbers for which the object mathematical induction, the object is dened for all natural numbers. This method of dening an innite set of objects is often referred to as sl recursive denition, or

denition by recursion. As an example of recursive denition, let us carefully dene exponentiation.

Denition 1.1:
Let

be a natural number.

We dene inductively natural numbers

an

as follows:

a1 = a,

and,

whenever The set

ak

is dened, then

ak+1

is dened to be

a ak .

an is dened is therefore all of N. For, a1 is dened, and if a is dened there is a prescription for dening ak+1 . This careful denition of an may seem unnecessarily n detailed. Why not simply dene a as a a a a... an times? The answer is that the ..., though suggestive enough, is just not mathematically precise. After all, how would you explain what ... means? The answer to that is that you invent a recursive denition to make the intuitive meaning of the ... mathematically precise. We will of course use the symbol ... to simplify and shorten our notation, but keep in mind that, if pressed, S
of all natural numbers for which

we should be able to provide a careful denition.

Exercise 1.2.1
and and

a. Derive the three laws of exponents for the natural numbers:

an+m = an am . HINT: Fix a n = (am ) . HINT: Fix a and m m and use the axiom of mathematical induction. a n n n use the axiom of mathematical induction. (a b) = a b . HINT: Fix a and b and
nm

use the axiom of mathematical induction. b. Dene inductively numbers to be

{Si } as follows: S1 = 1, and if Sk is dened, then Sk+1 is dened Sk + k + 1. Prove, by induction, that Sn = n (n + 1) /2. Note that we could have dened Sn using the ... notation by Sn = 1 + 2 + 3 + ... + n. 1 + 4 + 9 + 16 + ... + n2 = n (n + 1) (2n + 1) . 6
is called (1.1)

c. Prove that

d. Make a recursive denition of

n! = 1 2 3 ... n.n!

factorial.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

There is a slightly more general statement of the axiom of mathematical induction, which is sometimes of use.

N of natural numbers, and suppose that S satises the following conditions

1. 2.

1.3: GENERAL AXIOM OF MATHEMATICAL INDUCTION Let S be a subset of the set

There exists a natural number k0 such that k0 S. If S contains a natural number k, then S contains the natural number k + 1.
N
of natural numbers, we construct the set

Then S contains every natural number n that is larger than or equal to k0 .

From the fundamental set create an additional number called 0 that satises the equations

Z of all integers. First, we simply 0 + n = n for all n N and 0 n = 0 for all

n N.

The word create is, for some mathematicians, a little unsettling. In fact, the idea of zero did not

appear in mathematics until around the year 900. It is easy to see how the so-called natural numbers came by their name. Fingers, toes, trees, sh, etc., can all be counted, and the very concept of counting is what the natural numbers are about. On the other hand, one never needed to count zero ngers or sh, so that the notion of zero as a number easily could have only come into mathematics at a later time, a time when arithmetic was becoming more sophisticated. In any case, from our twenty-rst century viewpoint, 0 seems very understandable, and we won't belabor the fundamental question of its existence any further here. Next, we introduce the so-called For every natural number

negative numbers.

This is again quite reasonable from our point of view.

n,

we let

be a number which, when added to

n,

give 0. Again, the question of

whether or not such negative numbers exist will not concern us here. We simply create them. In short, we will take as given the existence of a set natural numbers, the additional number

Z, called the integers, which comprises the set N of 0, and the set N of all negative numbers. We assume that addition

and multiplication of integers satisfy the three basic algebraic relations of commutativity, associativity, and distributivity stated above. We also assume that the following additional relations hold:

(n) (k ) = n k,
for all natural numbers

and

(n) k = n (k ) = (n k )

(1.2)

and

k.
3

1.3 The Rational Numbers

Next, we discuss the set

Q of rational numbers, which we ordinarily think of as quotients k/n of integers. Of n of the quotient k/n to be 0. Also, we must remember that there isn't a 1-1 correspondence between the set Q of all rational numbers and the set of all such quotients k/n. Indeed, the two distinct quotients 2/3 and 6/9 represent the same rational number. To be precise, the set Q is a collection of equivalence classes of ordered pairs (k, n) of integers, for which the second component
course, we do not allow the second element of the pair is not 0. The equivalence relation among these ordered pairs is this:

(k, n) k ' , n'

if

k n' = n k ' .

(1.3)

We will not dwell on this possibly subtle denition, but will rather accept the usual understanding of the rational numbers and their arithmetic properties. In particular, we will represent them as quotients rather than as ordered pairs, and, if

is a rational number, we will write

r = k/n, k/n

instead of writing as the

as the of the

equivalence class containing the ordered pair

k/n

as the

numerator

(k, n) .

As usual, we refer to the rst integer in the quotient

and the second (nonzero) integer in the quotient

denominator

quotient. The familiar denitions of sum and product for rational numbers are these:

k' kn' + nk ' k + ' = n n nn'

3 This
content is available online at <http://cnx.org/content/m36061/1.2/>.

(1.4)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

and

kk ' k' k . ' = n n nn'

associativity and distributivity stated earlier. We note that the integers

(1.5)

Addition and multiplication of rational numbers satisfy the three basic algebraic relations of commutativity,

can be identied in an obvious way as a subset of the rational numbers

Q.

Indeed, we identify the integer

numbers

0 0/1

and

k = 0,

and any other

k with the quotient k/1. In this way, we note that Q contains the two 1 1/1. Notice that any other quotient k/n that is equivalent to 0/1 must satisfy ' ' quotient k/n that is equivalent to 1/1 must satisfy k = n. Remember, k/n k /n if

and only if element we may

The set

kn' = k ' n. Q has an additional property not shared by the set of integers Z. It is this: For each nonzero r Q, there exists an element r' Q for which r r' = 1. Indeed, if r = k/n = 0, then k = 0, and ' dene r = n/k. Consequently, the set Q of all rational numbers is what is known in mathematics

as a eld.

Denition 1.2:
A

eld

is a nonempty set

on which there are dened two binary operations, addition (+) and

multiplication (), such that the following six axioms hold: 1. Both addition and multiplication are commutative and associative. 2. Multiplication is distributive over addition; i.e.,

x (y + z ) = x y + x z
for all

(1.6)

x, y, z F. F,
which we will denote by

3. There exists an element in

0,

that is an identity for addition; i.e.,

x+0=x

for all

4. There exists a 5. 6.

F, which we will denote by 1, that is an identity for x F. If x F, then there exists a unique element y F such that x + y = 0. This element y is called the additive inverse of x and is denoted by x. If x F and x = 0, then there exists a unique element y F such that x y = 1. This 1 element y is called the multiplicative inverse of x and is denoted by x .
element in for all multiplication; i.e.,

nonzero

x F.

x1=x

1.4: REMARK.
F,

There are many examples of elds.

(See Exercise 1.3.1.)

They all share certain is an element of a eld

arithmetic properties, which can be derived from the axioms above. If then according to one of the axioms above, we have that

1 x = x.

(Note that this 1 is the

multiplicative identity of the eld

and not the natural number 1.) However, it is tempting to an element of

write x + x = 2 x in the eld F. The 2 here is not priori x + x = 2 x is not really justied. This is an example of denition can be useful.

F,

so that the equation

a situation where a careful recursive

Denition 1.3:
If

is an element of a eld

is dened, set therefore, by

F, dene inductively elements n x nx of F by 1 x = x, and, if k x (k + 1) x = x + k x. The set S of all natural numbers n for which n x is dened is the axiom of mathematical induction, all of N.
Of course,

Usually we will write itself

Exercise 1.3.1
a. Justify for yourself that the set that all six of the axioms hold.

times:

nx instead of n x. nx = x + x + x + ... + x.

nx

is just the element of

obtained by adding

to

of all rational numbers is a eld. That is, carefully verify

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

b. Let

F7

denote the seven elements

{0, 1, 2, 3, 4, 5, 6}.

Dene addition and multiplication on

F7

as ordinary addition and multiplication mod 7. Prove that for every c. Let

F7

is a eld. (You may assume

that axioms (1) and (2) hold. Check only conditions (3)(6).) Show in addition that

7x = 0

x F7 . {0, 1, 2, 3, 4, 5, 6, 7, 8}.
Dene addition

F9

denote the set consisting of the nine elements

and multiplication on d. Show that the set Show that the set

F9

to be ordinary addition and multiplication mod 9. Show that

F9

is

not a eld. Which of the axioms fail to hold?

N Z

of natural numbers is not a eld. Which of the eld axioms fail to hold? of all integers is not a eld. Which of the eld axioms fail to hold?

Exercise 1.3.2
Let a.

be any eld. Verify that the following arithmetic properties hold in

F.

b. If c. d. e. f.

0 x = 0 for all x F. HINT: Use the distributive law and the fact that 0 = 0 + 0. x and y are nonzero elements of F, then x y is nonzero. And, the multiplicative 1 of x y satises (x y ) = x1 y 1 . (1) x = (x) for all x F. (x) (y ) = x y for all x, y F. x x y y = (x y ) (x + y ) . (x + y ) (x + y ) = x x + 2 x y + y y.

inverse

Denition 1.4:
Let if

be a eld, and let

be a nonzero element of Of course,

F. xn
in

For each natural number

is dened, set

Dene

x0

to be

x 1.

k+1

n, we = x xk . n,

dene inductively an element

is just the product of

F as nx's.

follows:

x1 = x,

and,

For each natural number

dene

x n

to be the multiplicative inverse

(xn ) 0n

of the element

xn .
Finally, we dene

0m

to be 0 for every positive integer

m,

and we leave

and

00

undened.

We have therefore dened

Exercise 1.3.3
Let a.

for every nonzero

and every integer

m Z.

be a eld. Derive the following laws of exponents:

b. c.

xn+m = xn xm for all nonzero elements x F and all integers n and m. HINT: Fix x F and m Z and use induction to derive this law for all natural numbers n. Then use the fact 1 that in any eld (x y ) = x1 y 1 . nm m n x = (x ) for all nonzero x F and all n, m Z. n (x y ) = xn y n for all nonzero x, y F and all n Z. xy
will be denoted simply as

From now on, we will indicate multiplication in a eld by juxtaposition; i.e.,

xy.

Also, we will use the standard fractional notation to indicate multiplicative inverses. For instance,

xy 1 = x

x 1 = . y y

(1.7)

1.4 The Real Numbers

What are the real numbers?

From a geometric point of view (and a historical one as well) real numbers

are quantities, i.e., lengths of segments, areas of surfaces, volumes of solids, etc. For example, once we have

4 This

content is available online at <http://cnx.org/content/m36069/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

10

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

k/n.
In some obvious sense then, the rational numbers are real

settled on a unit of length, i.e., a segment whose length we call 1, we can, using a compass and straightedge, construct segments of any rational length numbers. Apparently it was an intellectual shock to the Pythagoreans to discover that there are some other real numbers, the so-called irrational ones. Indeed, the square root of 2 is a real number, since we can construct a segment the square of whose length is 2 by making a right triangle each of whose legs has length 1. (By the Pythagorean Theorem of plane geometry, the square of the hypotenuse of this triangle must equal 2.) And, Pythagoras proved that there is no rational number whose square is 2, thereby establishing that there are real numbers tha are not rational. See part (c) of Exercise 1.4.5. Similarly, the area of a circle of radius 1 should be a real number; i.e., wasn't until the late 1800's that Hermite showed that dene

should be a real number. It

is not a rational number. One diculty is that to

as the area of a circle of radius 1 we must rst dene what is meant by the  area" of a circle,

and this turns out to be no easy task. In fact, this naive, geometric approach to the denition of the real numbers turns out to be unsatisfactory in the sense that we are not able to prove or derive from these rst principles certain intuitively obvious arithmetic results. For instance, how can we multiply or divide an area by a volume? How can we construct a segment of length the cube root of 2? And, what about negative numbers? Let us begin by presenting two properties we expect any set that we call the real numbers ought to possess.

Algebraic Properties
We should be able to add, multiply, divide, etc., real numbers. In short, we require the set of real numbers to be a eld.

Positivity Properties
The second aspect of any set we think of as the real numbers is that it has some notion of direction, some notion of positivity. It is this aspect that will allow us to compare numbers, e.g., one number is larger than another. The mathematically precise way to discuss this notion is the following.

Denition 1.5:
A eld

is called an

ordered eld

if there exists a subset

P F

that satises the following two

properties: 1. If 2. If

x, y P, then x + y and xy are in P. x F, then one and only one of the following
i. ii.

three statements is true.

iii.

x P, x P, and x = 0. (This property P

is known as the

law of tricotomy.)
F,
and the elements

The elements of the set

are called

negative

are called

positive
P,

elements of

for which

belong to

elements of

F.
we may introduce in

As a consequence of these properties of

Denition 1.6:
If

a notion of order.

is an ordered eld, and

that

xy

if either

We say that

x and y are elements x < y or x = y. x > y if y < x, and x y if y x.

of

F,

we say that

x<y

if

y x P.

We say

An ordered eld satises the familiar laws of inequalities. They are consequences of the two properties of the set

P. F,
together with the axioms for a eld,

Exercise 1.4.1
Using the positivity properties above for an ordered eld

derive the familiar laws of inequalities: a. (Transitivity) If c. If

x<y

and

b. (Adding like inequalities) If

x<y

and

a > 0,

then

y < z, then x < z. x < y and z < w, ax < ay.

then

x + z < y + w.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

11

d. If f. g. h.

e. If

x < y and a < 0, then ay < ax. 0 < a < b and 0 < c < d, then ac < bd. Verify parts (a) through (e) with < replaced by . If x and y are elements of F, show that one and only one of the following three relations can hold: (i) x < y, (ii) x > y, (iii) x = y. Suppose x and y are elements of F, and assume that x y and y x. Prove that x = y.

Exercise 1.4.2
a. If

is an ordered eld, show that

1 P;

i.e., that

0 < 1.

HINT: By the law of tricotomy,

only one of the three possibilities holds for b. Show that c. Prove that

1.

Rule out the last two.

F7 Q

of Exercise 1.3.1 is not an ordered eld; i.e., there is no subset is an ordered eld, where the set

P F7

such

that the two positivity properties can hold. HINT: Use part (a) and positivity property (1).

is taken to be the usual set of positive

rational numbers. That is, are natural numbers. d. Suppose numbers

consists of those rational numbers

a/b F.

for which both

and

F is an ordered nnx = 0.

eld and that

is a nonzero element of

Show that for all natural

e. (e) Show that, in an ordered eld, every nonzero square is positive; i.e., if

x = 0, then x2 P.

We remarked earlier that there are many dierent examples of elds, and many of these are also ordered elds. Some elds, though technically dierent from each other, are really indistinguishable from the algebraic point of view, and we make this mathematically precise with the following denition.

Denition 1.7:
Let and 1. 2. 3.

F1 and F2 be two ordered elds, and write P1 and P2 for the set of positive elements F2 respectively. A 1-1 correspondence J between F1 and F2 is called an isomorphism if J (x + y ) = J (x) + J (y ) for all x, y F1 . J (xy ) = J (x) J (y ) for all x, y F1 . x P1 if and only if J (x) P2 .

in

F1

1.5: REMARK.
between

In general, if

A1

A1

and

A2

is called an

isomorphism
A2 .

and

A2

are two algebraic systems, then a 1-1 correspondence if it converts the algebraic structure on

A1

into the

Exercise 1.4.3
a. Let

corresponding algebraic structure on

be an ordered eld. Dene a function

J :N F

by

J (n) = n 1.

Prove that

is an

isomorphism of

onto a subset

of

F.

That is, show that this correspondence is one-to-one

b.

into addition and multiplication in F. Give an F is merely a eld and not an ordered eld. 1 Let F be an ordered eld. Dene a function J : Q F by J (k/n) = k 1 (n 1) . of the ordered eld F. Prove that J is an isomorphism of the ordered eld Q onto a subset Q Conclude that every ordered eld F contains a subset that is isomorphic to the ordered eld Q. and converts addition and multiplication in example to show that this result is not true if

1.6: REMARK. Part (b) of list, p.

11 shows that the ordered eld

is the smallest possible ordered

eld, in the sense that every other ordered eld contains an isomorphic copy of mentioned earlier, the ordered eld below.

Q.

However, as

Q cannot suce as the set of real numbers. F

There is no rational

number whose square is 2, and we want the square root of 2 to be a real number. See Exercise 1.4.5 What extra property is there about an ordered eld that will allow us to prove that

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

12

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

numbers like

2,,

and so on are elements of

F?

It turns out that the extra property we need is

related to a quite subtle point concerning upper and lower bounds of sets. It gives us some initial indication that the known-to-be subtle concept of a of what the real numbers are.

limit may be fundamental to our very notion

Denition 1.8:
bounded below
a lower bound. for An element

If S is a subset of an ordered eld F, then an element x F is called an upper bound for S if x y for every y S. An element z is called a lower bound for S if z y for every y S. A subset S of an ordered eld F is called bounded above if it has an upper bound; it is called if it has a lower bound; and it is called

bounded

if it has both an upper bound and

M is called the least upper bound or supremum of a set S if it is an upper bound M x for every other upper bound x of S. That is, M is less than or equal to any other upper bound of S. Similarly, an element m is called the greatest lower bound or inmum of S if it is a lower bound for S and if z m for every other lower bound z of S. That is, m is greater than or equal to any other lower bound of S. S
and if

Clearly, the supremum and inmum of a set upper bounds of a set

are unique. For instance, if

S,

then they are both upper bounds of

S.

We would then have

M ' are both least M M ' and M ' M.

and

Therefore, by part (h) of Exercise 1.4.1, particular, the least upper bound of If

M = M '. S
need not be an element of

It is important to keep in mind that an upper bound of a set

may or may not actually belong to S.

M
by

S,

and in

is the supremum of a set

S,

we denote

supS.

If

is the inmum of a set

S,

we denote it by

inf S.

Exercise 1.4.4
a. Suppose b. Let for c. If

is a nonempty subset of an ordered eld

and that

is an element of

F.

What that

does it mean to say that  x is not an upper bound for

S ?'' F. Prove yF
is a lower bound

be an ordered eld, and let

be the empty set, thought of as a subset of

every element

xF

is an upper bound for

and that every element

S. HINT: If not, then what? S = , show that S has no least

upper bound and no greatest lower bound.

1.7: REMARK.
happen when

The preceding exercise shows that peculiar things about upper and lower bounds

is the empty set. One point is that just because a set has an upper bound does

not mean it has to have a least upper bound. That is, no matter which upper bound we choose, there is always another one that is strictly smaller. This is a very subtle point, and it is in fact quite dicult to give a simple concrete example of this phenomenon. See the remark following Theorem 1.6, p. 14. However, part (d) of Exercise 1.4.5 contains the seed of an example.

Exercise 1.4.5
called

A natural number

odd

is called

even

if there exists a natural number

such that

a = 2c,

and

is

if there exists a natural number

such that

a = 2c + 1.

a. Prove by induction that every natural number is either odd or even. b. Prove that a natural number c. Prove that there is no which d.

a is even if and only if a2 = a a is even. element x of Q whose square is 2. That is, the square

root of 2 is not a

rational number. HINT: Argue by contradiction. Suppose there is a rational number

k/n for k 2 /n2 = 2, and assume, as we may, that the natural numbers k and n have no common factor. Observe that k must be even, and then observe that n also must be even. 2 Let S be the set of all positive rational numbers x for which x = x x < 2. Prove that S has an upper bound and a lower bound. Can you determine whether or not S has a least upper
bound?

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

13

The existence of least upper bounds and greatest lower bounds of bounded sets turns out to be the critical idea in dening the real numbers. It is precisely the existence of such suprema and inmas that enables us to dene as real numbers quantities such as

2,,e, S

Denition 1.9:

and so on.

An ordered eld

is called

complete

if every nonempty subset

of

that has an upper bound

has a least upper bound.

1.8: REMARK. Although Q is an ordered eld, we will see that it is not a complete ordered eld.

In fact, the answer to part (d) of Exercise 1.4.5 is no. The set described there, though bounded above, has no least upper bound. In fact, it was one of nineteenth century mathematicians' major achievements to prove the following theorem.

Theorem 1.1:

There exists a complete ordered eld. We leave the proof of this theorem to the appendix. Perhaps the most reassuring result along these lines is the following companion theorem, whose proof we also leave to the appendix.

Theorem 1.2:
If

F1

and

F2

are two complete ordered elds, then they are isomorphic.

Taken together, the content of the two preceding theorems is that, up to isomorphism, there exists one and only one complete ordered eld. For no other reason that that, this special eld should be an important object in mathematics. Our denition of the real numbers is then the following:

Denition 1.10:
By the set

of

real numbers

we mean the (unique) complete ordered eld.

1.5 Properties of the Real Numbers

Theorem 1.3:
The set

contains a subset that is isomorphic to the ordered eld

of rational numbers, and

hence subsets that are isomorphic to

1.9: REMARK. The proof of Statement of Theorem 1.3, p.

and

Z.
13 is immediate from part (b) of Exercise

1.7. In view of this theorem, we will simply think of the natural numbers, the integers, and the rational numbers as subsets of the real numbers. Having made a denition of the set of real numbers, it is incumbent upon us now to verify that this set

satises our intuitive notions about the reals. Indeed, we will show that

is an element of

and hence

is a real number (as plane geometry indicates it should be), and we will show in later chapters that there are elements of

that agree with our intuition about

and

Before we can proceed to these tasks, we must

establish some special properties of the eld

R.

The rst, the next theorem, is simply an analog for lower

bounds of the least upper bound condition that comes from the completeness property.

If S is a nonempty subset of R that is bounded below, then there exists a greatest lower bound for
S.

Theorem 1.4:

Proof:
Dene

to be the set of all real numbers

rst that

is bounded above. Thus, let

x for which x S. That is, T is the set S. We claim m be a lower bound for the set S, and let us show that the

5 This

content is available online at <http://cnx.org/content/m36085/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

14

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

number

is an upper bound for

T.

If

Since this is true for all

x T,

the number

x T, then x S. So, m x, m is an upper bound for T.

implying that

m x.

the greatest lower bound for

T has a least upper bound M0 . We claim that the number M0 is S. To prove this, we must check two things. First, we must show that M0 is a lower bound for S. Thus, let y be an element of S. Then y T, and therefore y M0 . Hence, M0 y, showing that M0 is a lower bound for S. Finally, we must show that M0 is the greatest lower bound for S. Thus, let m be a lower bound for S. We saw above that this implies that m is an upper bound for T. Hence, because M0 is the least upper bound for T, we have that m M0 , implying that m M0 , and this proves that M0 is the inmum of the set S.
Now, by the completeness assumption, The following is the most basic and frequently used property of least upper bounds. It is our rst glimpse of  limits. Though the argument is remarkably short and sweet, it will provide the mechanism for many of our later proofs, so master this one.

Let S be a nonempty subset of R that is bounded above, and Let M0 denote the least upper bound of S ; i.e., M0 = supS. Then, for any positive real number there exists an element t of S such that

Theorem 1.5:

Proof:
Let

t > M0 . >0
be given. Since

M0 < M0 ,

it must be that

is necessarily less than or equal to any other upper bound of

M0 is not an upper bound for S. (M0 S.) Therefore, there exists an element

tS

for which

t > M0 .

This is exactly what the theorem asserts.

Exercise 1.5.1
a. Let

S be a nonempty subset of R which is bounded below, and let m0 denote the inmum of S. Prove that, for every positive , there exists an element s of S such that s < m0 + . Mimic S
be any bounded subset of

the proof to Theorem 1.5, p. 14. b. Let

S.

Prove that

R, sup (S ) = inf S.

and write

for the set of negatives of the elements of

c. Use part (b) to give an alternate proof of part (a) by using Theorem 1.5, p. 14 and a minus sign.

Exercise 1.5.2
a. Let for b.

S S. Let S

be the set of all real numbers be the set of all

What is the least upper bound of

xR

for

What is the least upper bound of We show now that

x < 1. Give an example of an S ? Is supS an element of S ? 2 which x 4. Give an example of an upper S ? Does supS belong to S ?
for which

upper bound bound for

S.

contains elements other than the rational numbers in

Q.

Of course this holds for any

complete ordered eld. The next theorem makes this quite explicit.

If x is a positive real number, then there exists a positive real number y such that y 2 = x. That is, every positive real number x has a positive square root in R. Moreover, there is only one positive square root of x.
t for which t2 x. Then S is nonempty Indeed, If x > 1, S because 12 = 1 1 < 1 x = x. And, if x 1, then x itself is in S , because x2 = x x 1 x = x.
Let

Theorem 1.6:

Proof:
S

be the set of positive real numbers

then 1 is in

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

15

Also,

is bounded above. In fact, the number

by contradiction, suppose there were a

in

such that

1 + x/2 is an upper bound t > 1 + x/2. Then

of

S.

Indeed, arguing

x t2 > (1 + x/2) = 1 + x + x2 /4 > x,

which is a contradiction. Therefore, Now let show that

(1.8) and so

1 + x/2

is an upper bound of

S,

y = supS. We y 2 x. It will

wish to show that

y 2 = x.

We show rst that

then follow from the tricotomy law that and write

S is bounded above. y 2 x, and then we will y 2 = x. We prove both these

inequalities by contradiction. So, assume rst that number and

y 2 > x,

for the positive number

/ (2y ) , and, using y t < = /2y. So, = = = < < = y2 x

Theorem 1.5, choose a

tS

such that

y 2 x. Let be the positive t > y . Then y + t (2y ) ,

y 2 t2 + t2 x y 2 t2 (y + t) (y t) 2y (y t) 2y 2y , y2
is not greater than

2y

which is a contradiction. Therefore Now we show that

x.
also less than

y2

is not less than

x.

Again, arguing by contradiction, suppose it is, and let

be the positive number x y 2 . Choose a positive number that is less than y and / (3y ) . Let s = y + . Then s is not in S, whence s2 > x, so that we must have = = = = < < x y2 x s2 + s2 y 2 s2 y 2 (s + y ) (s y ) (2y + ) 3y , y 2 = x;
i.e., that

which again is a contradiction. This completes the proof that Finally, if

has a positive square root.

'

were another positive number for which

the other two cases: giving that

y < y'

and

y > y' .

For instance, if

y ' = x, we show that y = y ' by ruling out 2 y < y ' , then we would have that y 2 < y ' ,

x = y 2 < y ' = x, implying that x < x,

and this is a contradiction.

Denition 1.11:
If

is a positive real number, then the symbol

which

y = x.

1.10: REMARK Part (c) of Exercise 1.4.5 shows that the eld Q contains no number whose square is
R
does contain a number whose square is 2. We

Of course,

will denote the unique positive number

for

denotes the number 0.

2, and Theorem 1.6, p. 14 shows that the eld

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

16

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

have therefore proved that the real numbers is a larger set than the rational numbers. It may come as a surprise to learn that we only now have been able to prove that. Look back through the chapter to be sure. It follows also that be isomorphic to does not.

itself is not a complete ordered eld. If it were, it would

R,

by Theorem 1.2, so that it would have to contain a square root of 2, which it

Denition 1.12:
A real number called an

Exercise 1.5.3

irrational number.

that is not a rational number, i.e., is not an element of the subset

of

R,

is

a. Prove that every positive real number has exactly 2 square roots, one positive ( other negative b. Prove that if

( x).

x)

and the

is a negative real number, then there is no real number

such that

y 2 = x.

c. Prove that the product of a nonzero rational number and an arbitrary irrational number must be irrational. Show by example that the sum and product of irrational numbers can be rational.

1.6 Intervals and Approximation

Denition 1.13:
1. (i) 3.

We introduce next into the set of real numbers some geometric concepts, namely, a notion of distance between numbers. Of course this had to happen, for geometry is the very basis of mathematics.

The absolute value of a real number

is denoted by

|x|

and is dened as follows:

2. (ii) If

|0| = 0. x > 0 then |x| = x. (iii) If x < 0 (x > 0) then |x| = x.

We dene the

distance d (x, y ) between two real numbers x and y

be real numbers.

by

d (x, y ) = |x y |.

Obviously, such denitions of absolute value and distance can be made in any ordered eld.

Exercise 1.6.1
Let

and

a. Show that

|x| 0 ,

and that

x |x|. |x + y | |x| + |y |.

b. Prove the Triangle Inequality for absolute values. (1.9)

HINT: Check the three cases

x + y > 0,x + y < 0,

and

x + y = 0.

c. Prove the so-called ` ` backward triangle inequality.

|x y | ||x| |y ||.
HINT: Write d. Prove that e. Prove that

(1.10)

|x| = | (x y ) + y |, and use part |xy | = |x||y |. |x| = x2 for all real numbers x.

(b).

6 This

content is available online at <http://cnx.org/content/m36094/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

17

f. Prove the Triangle Inequality for the distance function. That is, show that

d (x, y ) d (x, z ) + d (z, y )

for all

(1.11)

x, y, z R.

Exercise 1.6.2
a. Prove that Suppose b. Prove

x = y if |x y | < for every positive number . HINT: Argue by x = y, and take = |x y |/2. that x = y if and only if x y and y x for every positive .

contradiction.

Denition 1.14:
Let

numbers

x
By

for

b be real numbers for which a < b. By the open interval (a,b) we mean the set of all real x for which a < x < b, and by the closed interval [a,b] we mean the set of all real numbers which a x b.
and we mean the set of all real numbers

(a, )

for which

a < x,

and by

[a, )

we mean the set of all

real numbers

for which

Analogously, we

a x. dene (, b)

and

(, b]

to be respectively the set of all real numbers

for which

x<b

Exercise 1.6.3
interval.

and the set of all real numbers

for which

x b.

a. Show that the intersection of two open intervals either is the empty set or it is again an open b. Show that c. Let

(a, b) = (, b) (a, ) . y be a xed real number, and let be a positive |x y | < is equivalent to the pair of inequalities

number.

Show that the inequality

y < x and x < y + ;

i.e., show that Deduce that

(1.12) latter ones.

x satises the rst inequality if and only if it satises the two |x y | < if and only if x is in the open interval (y , y + ) .

Here is one of those assertions that seems like an obvious fact. However, it requires a proof which we only now can give, for it depends on the completeness axiom, and in fact is false in some ordered elds.

Let N denote the set of natural Numbers, thought of as a subset of R. Then N is not bounded above.

Theorem 1.7:

Proof:

Suppose false. Let bound for

be an upper bound for the nonempty set

N,

and let

M0

be the least upper

N.

Taking So

to be the positive number 1/2, and applying Theorem 1.5, we have that

there exists an element

M0 + 1/2 < k + 1.

k of N such that M0 1/2 < k. But then M0 1/2 + 1 < k + 1, or, M0 < k + 1. But M0 k + 1 because M0 is an upper bound for N. We have

thus arrived at a contradiction, and the theorem is proved.

1.11: REMARK As mentioned above, there do exist ordered elds F

by Abraham Robinson in 1966. The fact that

in which the subset

N is bounded

above. Such elds give rise to what is called nonstandard analysis, and they were rst introduced

is a complete ordered eld is apparently crucial to

be able to conclude the intuitively clear fact that the natural numbers have no upper bound. Exercise 1.6.4 presents another intuitively obvious fact, and this one is in some real sense the basis for many of our upcoming arguments about limits. It relies on the preceding theorem, is in fact just a corollary,

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

18

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

so it has to be considered as a rather deep property of the real numbers; it is not something that works in every ordered eld.

Exercise 1.6.4
Prove that if

is a

positive real number, then there exists a natural number

such that

1/N < .

Theorem 1.8, p. 18 and Exercise 1.6.5 show that the set eld

of rational numbers is everywhere dense in the

R.

That is, every real number can be approximated arbitrarily closely by rational numbers. Again, we

point out that this result holds in any complete ordered eld, and it is the completeness that is critical.

Let a < b be two real numbers. Then there exists a rational number r = p/q in the open interval (a, b) . In fact, there exist innitely many rational numbers in the interval (a, b) .

Theorem 1.8:

Proof:

If a < 0 and b > 0, then taking r = 0 satises the rst statement of the theorem. Assume rst that a 0 and b > a. Let n be a natural number for which 1/n is less than the positive number b a. (Here, we are using the completeness of the eld, because we are referring to Theorem 1.7, where completeness was vital.) If Suppose then that

a = 0,

then

b = b a.

Setting

r = 1/n, q

we would have that

a < r < b.

So, again, the rst part of the theorem would be proved in that case.

a > 0, N,

and choose the natural number

to be such that

the minimum of the two positive numbers upper bound for the set we may let

and

b a.

Now, because the number

be the smallest natural number that is

1/q is less than aq is not an larger than aq. Set

r = p/q. aq < p, implying that a < p/q = r. Also, because p is the smallest natural aq, we must have that p 1 aq. Therefore, (p 1) /q < a, or (p/q ) (1/q ) < a, implying that r = p/q a + 1/q < a + (b a) = b. Hence, a < r and r < b, and the rst statement of the theorem is proved when both a and b are nonnegative. If both a and b are nonpositive, then both b and a are nonnegative, and, using the rst part of the proof, we can nd a rational number r such that b < r < a. So, a < r < b, and the
We have rst that number larger than rst part of the theorem is proved in this case as well. Clearly, we may replace b by r and repeat the argument to obtain another rational r1 such that a < r1 < r < b. Then, replacing b by r1 and repeating the argument, we get a third rational r2 such that a < r2 < r1 < r < b. Continuing this procedure would lead to an innite number of rationals, all between a and b. This proves the second statement of the theorem.

Exercise 1.6.5
a. Let number

> 0 be given, and p/q such that

let

be a nonnegative integer. Prove that there exists a rational

k < p/q < (k + 1) .

b. Let

(1.13)

be a positive real number and let

a rational number negative numbers

p/q x.

such that

be a positive real number. x < p/q < x. State and prove

Prove that there exists an analogous result for

Exercise 1.6.6
a. If

and

are real numbers with

a < b,

show that there is an irrational number

(not a

rational number) between the numbers

a 2

and

a b 2.

and

b,

i.e., with

a < x < b.

HINT: Apply Theorem 1.8, p. 18 to

b. Conclude that within every open interval

(a, b)

there is a rational number and an irrational

number. Are there necessarily innitely many rationals and irrationals in

(a, b)?

The preceding exercise shows the denseness of the rationals and the irrationals in the reals. It is essentially clear from this that every real number is arbitrarily close to a rational number and an irrational one.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

19

1.7 The Geometric Progression and the Binomial Theorem

There are two special algebraic identities that hold in They are both proved by mathematical induction. progression.

R (in fact in any eld F

whatsoever) that we emphasize.

The rst is the formula for the sum of a geometric

Theorem 1.9:
Let

Geometric Progression

be a real number, and let

be a natural number. Then,

1. If

x = 1,

then

xj =
j =0
2. If

1 xn+1 . 1x

(1.14)

x = 1,

then

xj = n + 1 .
j =0

(1.15)

Proof:
The second claim is clear, since there are

n+1

summands and each is equal to 1.

We prove the rst claim by induction. Thus, if

n = 1,

then the assertion is true, since

xj = x0 + x1 = 1 + x = (1 + x)
j =0

1x 1 x2 = . 1x 1x k,
i.e., that

(1.16)

Now, supposing that the assertion is true for the natural number

xj =
j =0

1 xk+1 , 1x k + 1.
Thus

(1.17)

let us show that the assertion holds for the natural number

k+1 j =0

xj

= = = =

k j k+1 j =0 x + x k+1 1x k+1 1x + x k+1 k+1 1x +x xk+2 1x 1xk+1+1 , 1x

(1.18)

which completes the proof.

The second algebraic formula we wish to emphasize is the Binomial Theorem. Before stating it, we must introduce some useful notation.

Denition 1.15:
Let

be a natural number. As earlier in this chapter, we dene

n!

as follows: (1.19)

n! = n (n 1) (n 2) ... 2 1.
For later notational convenience, we also dene If

0!

to be 1.

is any integer for which

0 k n,

we dene the

binomial coecient

n k

by

n n! n (n 1) (n 2) ... (n k + 1) = = . k k ! (n k )! k!
7 This
content is available online at <http://cnx.org/content/m36104/1.2/>.

(1.20)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

20

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

Exercise 1.7.1
a. Prove that b. Prove that

n 0

= 1,

n 1

=n

and

n n

= 1.
(1.21)

for all natural numbers c. Prove that

and all

2nk n k k 2 integers 0 k n. n+1 k = n + k n k1

(1.22)

for all natural numbers

and all integers

1 k n.

Theorem 1.10:
If

x, y R

and

is a natural number, then

(x + y ) =
k=0

n k nk x y . k
then the assertion is true, for

(1.23)

Proof:
We shall prove this theorem by induction. If and

n = 1,

(x + y ) = x + y

k=0

1 k

xk y 1k =

1 0

x0 y 1 +

1 1

x1 y 0 = x + y. j;
i.e.,

(1.24)

Now, assume that the assertion holds for the natural number

(x + y ) =
k=0

j k

x k y j k , j + 1.

(1.25)

and let us prove that the assertion holds for the natural number (c) of Exercise 1.7.1. We have that

We will make use of part

(x + y )

j +1

= = = = = x

(x + y ) (x + y ) (x + y )
j k=0 k j k j k

xk y j k

= = = =

j j j j + y k=0 k x k y j k k=0 k x y j j j j k+1 j k y + k=0 k xk y j +1k k=0 k x j 1 j k+1 j k j +1 0 y + j y k=0 k x j x j j j k j +1k 0 j +1 + k=1 k x y + 0 x y j j xj +1 + k=1 k1 xk y j +1k j j + k=1 k xk y j +1k + y j +1 j j j xj +1 + k=1 xk y j +1k + y j +1 k 1 + k j xj +1 + k=1 j +1 xk y j +1k + y j +1 k j j +1 j +1 0 y + k=1 j +1 xk y j +1k + j +1 x0 y j +1 j +1 x k 0 j +1 j +1 xk y j +1k , k=0 k

(1.26)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

21

which shows that the assertion of the theorem holds for the natural number the proof.

j + 1.

This completes

The next exercise is valid in any ordered eld, but, since we are mainly interested in the order eld we state everything in terms of that eld.

R,

Exercise 1.7.2
a. If

that b. For c. For

y are positive real numbers, and if n and k are natural numbers with k n, show n k nk (x + y ) n . k x y n any positive real number x and natural number n, show that (1 + x) 1 + nx. n any real number x > 1 and natural number n, prove that (1 + x) 1 + nx. HINT:
and

Do not try to use the binomial theorem as in part (b); it won't work because the terms are not all positive; prove this directly by induction. There is one more important algebraic identity, which again can be proved by induction. It is actually just a corollary of the geometric progression formula.

Theorem 1.11:
If

x, y R

and

is a natural number, then

xn y n = (x y )

n1

xj y n1j .(1.27)
j =0

Proof:
If for the natural number

n = 1 the theorem is clear. Suppose it holds for a natural number k, and let us prove the identity k + 1. We have xk+1 y k+1 = = = = = = xk+1 xk y + xk y y k+1 (x y ) xk + y xk y k (x y ) xk + y (x y ) (x y ) xk + (x y ) (x y ) xk y kk + (x y )
k j =0 k1 j k1j j =0 x y k1 j kj j =0 x y k1 j kj j =0 x y
(1.28)

x j y k j k + 1. So,
by induction, the theorem

, which shows that the assertion holds for the natural number is proved.

Exercise 1.7.3
Let

and

be real numbers.

a. Let

be an odd natural number; i.e.,

n = 2k + 1
n1 j =0

for some natural number

k.

Show that

xn + y n = (x + y ) x n + y n = x n ( y ) . 2 Show that x + y can not be factored choices of real numbers a, b, c, and d.

HINT: Write

(1) xj y n1j .(1.29)

b.

into a product of the form

(ax + by ) (cx + dy )

for any

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

22

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

This next theorem is denitely not valid in any ordered eld, for it again depends

Using the Binomial Theorem together with the preceding theorem, we may now investigate the existence of

nth roots of real numbers.

on the completeness property.

Theorem 1.12:

Let

real number

Proof:

n be a natural number and let x be a positive real number. Then there exists a unique positive y such that y n = x; i.e., x has a unique positive nth root. 0 t < s, t
then

Note rst that if

tn < sn .

(To see this, argue by induction, and use part (e) of

S be the set of tn x. Then S is nonempty and bounded above. Indeed, if x 1, then 1 S, while if x < 1, then x itself is in S. Therefore, S is nonempty. Also, using part (b) of Exercise 1.7.2, we see that 1 + (x/n) is an upper bound for S. For, if t > 1 + x/n, then
Exercise 1.4.1.) Using this, we mimic the proof of Theorem 1.6, p. 14. Thus, let for which all positive real numbers

tn > (1 + (x/n)) 1 + n (x/n) > x.

Now let

(1.30)

y = supS, and let us show that y n = x. We rule out the other two possibilities. First, if y > x, let be the positive number y n x, and dene ' to be the positive number / ny n1 . ' Then, using Theorem 1.5, p. 14, choose t S so that y < t y. (Theorem 1.5, p. 14 is where the completeness of the ordered eld R is crucial.) We have
n

= = = = < = ( y t) (y t) ( y t)

yn x y n tn + tn x y n tn
n1 j n1j j =0 y t n1 j n1j j =0 y y n1 n1 j =0 y
(1.31)

' ny n1 , yn
is not greater than

and this is a contradiction. Therefore,

x. >0
such that

y < x, let be n < /(y + 1) . Then, using

Now, if

the positive number

x yn ,

and choose a

<1

and

the Binomial Theorem, we have that

(y + )

= = = < = = < =

n n k=0 k n1 y n + k=0 n1 y n + k=0 n y n + k=0 n

y k nk
n k n k n k

y k nk y k n1k y k 1nk
n n
(1.32)

y + (y + 1) x+ x,

x + (y + 1)

implying that

x,

and so

y + S. y = x.
n

But this is a contradiction, since

y = supS.

Therefore,

yn

is not less than

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

23

We have shown the existence of a positive

nth

root of

x.

To see the uniqueness, suppose

and

y'

are two positive

nth

roots of

x. 0

Then

= = y y'
n1

yn y'
n1 j =0

n nj 1

yj y'

(1.33)

which implies that either positive terms, it cannot

y y ' = 0 or j =0 y j y ' ' be 0, whence y = y . This shows

nj 1

= 0.

Since this latter sum consists of

that there is but one positive

nth

root of

x,

and the theorem is proved.

Exercise 1.7.4
a. Show that if two distinct b. If c. If root.

n = 2k is an even natural number, then every positive real number has exactly nth roots. n = 2k + 1 is an odd natural number, show that every real number has exactly one nth n
is a natural number greater than 1, prove that there is no rational number whose

nth

power equals 2, i.e., the

nth

root of 2 is not a rational number.

1.8 The Complex Numbers

the real numbers

It is useful to build from the real numbers another number system called the

complex numbers.

Although

have many of the properties we expect, i.e., every positive number has a positive square

root, every number has a cube root, and so on, there are somewhat less prominent properties that

fails to

possess. For instance, negative numbers do not have square roots. This is actually a property that is missing in any ordered eld, since every square is positive in an ordered eld. See part (e) of Exercise 1.4.2. One way of describing this shortcoming on the part of the real numbers is to note that the equation has no solution in the real numbers. Any solution would have to be a number whose square is system in which this equation has a solution? We faced a similar kind of problem earlier on. In the set for all

1 + x2 = 0 1, and no

real number has that property. As an initial extension of the set of real numbers, why not build a number

there is no element

such that

j+n = n

n N.

That is, there was no element like 0 in the natural numbers. The solution to the problem

in that case was simply to create something called zero, and just adjoin it to our set property that its square

of solution exists for us now. Let us invent an additional number, this time denoted by

N. The same kind i, which has the

is 1. Because the square of any nonzero real number is positive, this new number i was traditionally referred to as an imaginary number. We simply adjoin this number to the set R, and we will then have a number whose square is negative, i.e., 1. Of course, we will require that our new number system should still be a eld; we don't want to give up our basic algebraic operations. There are several

i2

y is any real number, then we must also adjoin to R the y i yi, for our new number system should be closed under multiplication. Of course the square 2 2 2 of iy will equal i y = y , and therefore this new number iy must also be imaginary, i.e., not a real number. Secondly, if x and y are any two real numbers, we must have in our new system a number called x + yi,
implications of this requirement: First of all, if number because our new system should be closed under addition.

Denition 1.16:
Let in

i denote an object whose square i2 = 1. Let C be the set of all objects that can be represented the form z = x + yi, where both x and y are real numbers.
content is available online at <http://cnx.org/content/m36113/1.2/>.

8 This

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

24

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

Dene two operations

and

on

as follows: (1.34)

(x + yi) + x' + y ' i = x + x' + y + y ' i,

and

(x + iy ) x' + iy ' = xx' + xiy ' + iyx' + iyiy ' = xx' yy ' + xy ' + yx' i.

(1.35)

Theorem 1.13:
1. The two operations

and

dened above are commutative and associative, and multiplica-

tion is distributive over addition. 2. Each operation has an identity: for multiplication. 3. The set

(0 + 0i) is the identity for addition, and (1 + 0i) is the identity

with these operations is a eld.

Proof:
C is a eld, we need z = x + yi = 0 = 0 + 0i, then there exists a w = u + vi such that z w = 1 = 1 + 0i. Thus, suppose z = x + yi = 0. Then at least one of the 2 2 two real numbers x and y must be nonzero, so that x + y > 0. Dene a complex number w by
We leave the proofs of Parts (1) and (2) to the following exercise. To see that to verify one nal condition, and that is to show that if

w=
We then have

y x + 2 i. x2 + y 2 x + y2

(1.36)

zw

= = = = =

(x + yi)
x2 x2 +y 2

y 2 x2 +y 2 + x2 +y 2 x2 +y 2

y x x2 +y 2 + x2 +y 2 i y x x x2 +y 2 + y x2 +y 2

i
(1.37)

+ 1,

0 x2 +y 2 i

1 + 0i

as desired.

Exercise 1.8.1
Prove parts (1) and (2) of Theorem 1.13, p. 24. One might think that these kinds of improvements of the real numbers will go on and on. For instance, we might next have to create and adjoin another object the equation

so that the number

has a square root; i.e., so that

i z2 = 0

has a solution. Fortunately and surprisingly, this is not necessary, as we will see

when we nally come to the Fundamental Theorem of Algebra in Theorem 7.7, Fundamental Theorem of Algebra, p. 195. The subset of

consisting of the pairs

x + 0i

is a perfect (isomorphic) copy of the real number system

R.

We are justied then in saying that the complex number system extends the real number system, and we

will say that a real number

is the same as the complex number

kinds of complex numbers. The complex numbers of the form Obviously, the only complex number that is both real and set

x + 0i. That is, real numbers are special 0 + yi are called purely imaginary numbers. purely imaginary is the number 0 = 0 + 0i. The

can also be regarded as a 2-dimensional space, a plane, and it is also helpful to realize that the complex

numbers form a 2-dimensional vector space over the eld of real numbers.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

25

Denition 1.17:
If

z = x + yi,
If

we say that the real number

the real number

is the

imaginary part
2

of

x z

is the

real part of z
y=

and write of

x=
by

(z ) .

We say that

and write

z = x + yi

is a complex number, dene the

complex conjugate z

(z ) .

z = x yi.

The complex number

satises

or equivalently that the equation extending the real numbers. roots,

i = 1, showing that the negative number 1 has a square root in C, 1 + z 2 = 0 has a solution in C. We have thus satised our initial goal of
Do they have square roots, cube

But what about other complex numbers?

Exercise 1.8.2
satises

nth

roots? What about solutions to other kinds of equations than

1 + z2?

a. Prove that every complex number has a square root. HINT: Let

z = a + bi. Assume w = x + yi
and

w2 = z, C.

and just solve the two equations in two unknowns that arise.

b. Prove that every quadratic equation a solution in HINT: If

az 2 + bz + c = 0,

for

a, b,

a = 0,

it is easy to nd a solution. If

c complex numbers, has a = 0, we need only nd a

solution to the equivalent equation

c b z 2 + z + = 0. a a
Justify the following algebraic manipulations, and then solve the equation.

(1.38)

b z2 + a z+

c a

b = z2 + a z+

=
What about this new eld equation have a solution in

z+

b2 b2 c 4a2 4a2 + a 2 2 b c 4ba2 + a . 2a

(1.39)

C? C?

Does every complex number have a cube root, a fourth root, does every A natural instinct would be to suspect that

takes care of square roots,

but that it probably does not necessarily have higher order roots. However, the content of the Fundamental Theorem of Algebra, to be proved in Section 7.4, is that every equation of the form nonconstant polynomial, has a solution in

P (z ) = 0,

where

is a

nth

root, for any solution of the equation

C. This immediately implies that every z n c = 0 would be an nth root of c.

complex number

has an

The fact that the Fundamental Theorem of Algebra is true is a good indication that the eld good eld. But it's not perfect.

is a

Theorem 1.14:

In no way can the eld

be made into an ordered eld. That is, there exists no subset

of

that satises the two positivity axioms.

Proof:

Suppose

were an ordered eld, and write

for its set of positive elements. Then, since every

square in an ordered eld must be in must be in

(part (e) of Exercise 1.4.2), we must have that

1 = i2
is not an

P.

But, by part (a) of Exercise 1.4.2, we also must have that 1 is in

P,

and this leads to

a contradiction of the law of tricotomy. We can't have both 1 and ordered eld.

in

P.

Therefore,

Although we may not dene when one complex number is smaller than another, we can dene the absolute value of a complex number and the distance between two of them.

Denition 1.18:
If

z = x + yi

is in

C,

we dene the

absolute value
|z | =

of

by (1.40) and

x2 + y 2 . w
by

We dene the

distance d (z, w) between two complex numbers z

d (z, w) = |z w|.
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

26

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

If

cC

and

r > 0,

we dene the

open disk of radius r around c, and denote it by Br (c) , by

(1.41)

Br (c) = {z C : |z c| < r}.

The

closed disk

of radius

around

is denoted by

B r (c)

and is dened by (1.42) around

B r (c) = {z C : |z c| r}.
We also dene open and closed

punctured

disks

' Br (c)

and

B r (c)

'

by (1.43)

' Br (c) = {z : 0 < |z c| < r}

and

B r (c) = {z : 0 < |z c r}.

point

'

(1.44)

These punctured disks are just like the regular disks, except that they do not contain the central

c. S
is any subset of

Exercise 1.8.3
b. Let

around S, |z w| < r. That is, Nr (S ) is the set of all complex numbers that are within a distance of r of the set S. We dene the closed neighborhood of radius r around S, and denote it by N r (S ) , to be the set of all z C for which there exists a w S such that |z w | r. More generally, if we dene the denoted by

C,

open neighborhood of radius r

wS
for which

Nr (S ) ,

to be the set of all

such that there exists a

a. Prove that the absolute value of a complex number addition that

|z | = zz. x
is the same whether we think of Note that this just amounts to (1.45)

is a nonnegative real number. Show in

be a real number. Show that the absolute value of

as a real number or as a complex number.

c. Prove that

max (| (z ) |, | (z ) |) |z | | (z ) | + | (z ) |. max (|x|, |y |)

verifying that

x2 + y 2 |x| + |y |

d. For any complex numbers e. f.

x and y. z and w, show that z + w = z + w, and that z = z. Show that z + z = 2 (z ) and z z = 2i (z ) . ' ' If z = a + bi and w = a + b i, prove that |zw | = |z ||w |. HINT: Just ' ' 2 | (a + bi) a + b i | .
for any two real numbers

compute

The next theorem is in a true sense the most often used inequality of mathematical analysis. We have already proved the triangle inequality for the absolute value of real numbers, and the proof was not very dicult in that case. For complex numbers, it is not at all simple, and this should be taken as a good indication that it is a deep result.

Theorem 1.15:
z
and

Triangle Inequality

If

z'

are two complex numbers, then

|z + z ' | |z | + |z ' |
and

(1.46)

|z z ' | ||z | |z ' ||.

(1.47)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

27

Proof:
We use the results contained in Exercise 1.8.3.

|z + z ' |2

= = = = = = =
2

z + z ' (z + z ' ) z + z' z + z'

2
(1.48)

zz + z ' z + zz ' + z ' z ' |z | + z ' z + z ' z + |z ' | |z | + 2 |z | + 2|

2 2 2 2

z ' z + |z ' |

z ' z | + |z ' |2

|z | + 2|z ' z | + |z ' |2 |z | + 2|z ' ||z | + |z ' |2 |z | + |z ' |

2

The Triangle Inequality follows now by taking square roots.

1.12: REMARK The Triangle Inequality is often used in conjunction with what's called the add and
|z w|, and we can often

subtract trick. Frequently we want to estimate the size of a quantity like bars:

accomplish this estimation by adding and subtracting the same thing within the absolute value

|z w| = |z v + v w| |z v | + |v w|.
The point is that we have replaced the estimation problem of the possibly unknown quantity by the estimation problems of two other quantities

(1.49)

|z w|

|z v |

and

|v w|.

It is often easier to estimate

these latter two quantities, usually by an ingenious choice of

Exercise 1.8.4

of course.

a. Prove the second assertion of the preceding theorem. b. Prove the Triangle Inequality for the distance function. That is, prove that

d (z, w) d (z, v ) + d (v, w)

for all

(1.50)

z, w, v C.
n n

c. Use mathematical induction to prove that

|
i=1

ai |
i=1

|ai |.

(1.51)

It may not be necessary to point out that part (b) of the preceding exercise provides a justication for the name triangle inequality. Indeed, part (b) of that exercise is just the assertion that the length of one side of a triangle in the plane is less than or equal to the sum of the lengths of the other two sides. Plot the three points

z, w,

Denition 1.19:
A subset in

and

v,
of

and see that this interpretation is correct.

is called

Bounded

if there exists a real number

such that

|z | M

for every

S.

Exercise 1.8.5
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

28

CHAPTER 1. THE REAL AND COMPLEX NUMBERS

a. Let of

be a subset of

numbers in

S,

and let

S.

Prove that

C . Let S1 be the subset of R consisting of the real parts of the complex S2 be the subset of R consisting of the imaginary parts of the elements S is bounded if and only if S1 and S2 are both bounded.
i.e., the set of all complex numbers of part (a).

HINT: Use Part (c) of Exercise 1.8.3.. b. Let

S be the unit circle in the plane, |z | = 1. Compute the sets S1 and S2

z = x + iy

for which

Exercise 1.8.6
a. Verify that the formulas for the sum of a geometric progression and the binomial theorem (Theorem 1.9, Geometric Progression, p. 19 and Theorem 1.10, p. 20) are valid for complex numbers eld. b. Prove Theorem 1.11, p. 21 for complex numbers

and

z'.

HINT: Check that, as claimed, the proofs of those theorems work in any

and

z'.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

Chapter 2

The Limit of a Sequence of Numbers

2.1 Denition of the Number e
1
This chapter contains the beginnings of the most important, and probably the most subtle, notion in mathematical analysis, i.e., the concept of a limit. Though Newton and Leibniz discovered the calculus with its tangent lines described as limits of secant lines, and though the Greeks were already estimating areas of regions by a kind of limiting process, the precise notion of limit that we use today was not formulated until the 19th century by Cauchy and Weierstrass. The main results of this chapter are the following: 1. The denition of the 3. The 4.

limit of a sequence,
e
(Theorem 2.3, Denition of

2. The denition of the real number

Squeeze Theorem (Theorem 2.5, Squeeze Theorem, p. 37), the Bolzano Weierstrass Theorem (Theorem 2.8, Bolzano-Weierstrass, p. Cauchy Criterion (Theorem 2.9, Cauchy Criterion, p. 43), innite series, the Comparison Test (Theorem 2.17, Comparison Test, p. 49), and the Alternating Series Test (Theorem 2.18, Alternating Series Test, p.
p. 45),

e.,

p. 35),

40 and Theorem 2.10,

5. The 7. 8.

6. the denition of an

51).

These are powerful basic results about limits that will serve us well in later chapters.

2.2 Sequences and Limits

Denition 2.1:
A

sequence

of real or complex numbers is dened to be a function from the set

numbers into the setR or ordinarily use the

N of natural C. Instead of referring to such a function as an assignment n f (n) , we notation {an },{an }1 , or {a1 , a2 , a3 , ...}. Here, of course, an denotes the number

f (n) .

2.1: REMARK

We expand this denition slightly on occasion to make some of our notation more

indicative. That is, we sometimes than 1. For example, we write

{an } 0 ,{a0 , a1 , ...},

index the terms of a sequence beginning with an integer other

or even

{an } 3 .

We give next what is the most signicant denition in the whole of mathematical analysis, i.e., what it means for a sequence to converge or to have a limit.

1 This 2 This

content is available online at <http://cnx.org/content/m36117/1.2/>. content is available online at <http://cnx.org/content/m36118/1.2/>. Available for free at Connexions <http://cnx.org/content/col11249/1.1>

29

30

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

Denition 2.2:
Let to

L be a real number. The sequence {an } is said L, or that L is the limit of {an }, if the following condition is satised. For every positive number , there exists a natural number N such that if n N, then |an L| < . In symbols, we say L = liman or

converge

{an }

be a sequence of real numbers and let to

L = lim an .
n
We also may write If a sequence

(2.1)

{an }

is

convergent.

an L. {an } of real or complex numbers converges to a number L, we say that the sequence {an } N
of real numbers such that if

We say that a sequence

diverges
then

to

there exists a natural number

n N,

+ if for every positive number M, an M. Note that we do not say that

real number the same as

such a sequence is convergent. Similarly, we say that a sequence

{an }

of real numbers

M,

there exists a natural number

such that if

The denition of convergence for a sequence for a sequence of real numbers. Thus, let complex number. The sequence that if

diverges to if for every n N, then an M. {zn } of complex numbers is exactly

to

{zn }

{zn }

is said to

converge

be a sequence of complex numbers and let

following condition is satised. For every positive number

L be a L, or that L is the limit of {zn }, if the , there exists a natural number N such

n N,

N ' very likely will need to be larger than N. Sometimes we will indicate this dependence by writing N () instead of simply N. It is always wise to remember that N depends on . On the other hand, the N or N () in this denition is not unique. It should be clear that if a natural number N satises this denition, then any larger natural number M will also satisfy the denition. So, in fact, if there exists one natural
number

2.2: REMARKS The natural number N

.
If

then

|zn L| < .

of the preceding denition surely depends on the positive

'

is a smaller positive number than

then the corresponding

number that works, then there exist innitely many such natural numbers. It is clear, too, from the denition that whether or not a sequence is convergent only depends on the tail of the sequence. Specically, for any positive integer concerned with

K,

the numbers

a1 , a2 , ..., aK

can

take on any value whatsoever without aecting the convergence of the entire sequence. We are only

an 's

for

n N,

and as soon as

is chosen to be greater than

K,

the rst part of

the sequence is irrelevant. The denition of convergence is given as a fairly complicated sentence, and there are several other ways of saying the same thing. Here are two: For every whenever

for

n N,|an L| < . And, which n N. It's a good idea

given an

> 0,

there exists a

> 0, there exists a N such that, N such that |an L| < for all

to think about these two sentences and convince yourself that

they really do mean the same thing as the one dening convergence. It is clear from this denition that we can't check whether a sequence converges or not unless we know the limit value

L.

The whole thrust of this denition has to do with estimating the quantity

|an L|.

We will see later that there are ways to tell in advance that a sequence converges without

knowing the value of the limit.

Example 2.1

Let an = 1/n, and let us show that liman = 0. Given 1/N < . (How do we know we can nd such a N ?) Now,

an if

> 0, let us choose n N, then we have

such that

which is exactly what we needed

1 1 1 < , |an 0| = | | = n n N to show to conclude that 0 = liman .

(2.2)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

31

an = (2n + 1) / (1 3n) , and let L = 2/3. Let us show that L = liman . Indeed, if > 0 is N, such that if n N then |an + (2/3) | < . Let us examine the quantity |an + 2/3|. Maybe we can make some estimates on it, in such a way that it becomes clear how to nd the natural number N.
Let given, we must nd a

Example 2.2

|an + (2/3) |

= = = = = <

2 2n+1 |1 3n + 3 | 6n | 6n+3+2 | 39n 5 | 3 9n | 5 9n3 5 6n+3n3 5 6n 1 n,

(2.3)

for all

n 1.

Therefore, if

is an integer for which

N > 1/,

then (2.4)

|an + 2/3| < 1/n 1/N < ,

whenever number

n N, 1/?)

as desired. (How do we know that there exists a

which is larger than the

Example 2.3
Let

an = 1/ n,
2

and let us show that

liman = 0.

Given an

> 0,

we must nd an integer

N N

that such

satises the requirements of the denition. It's a little trickier this time to choose this the positive number that

N.

Consider

2 .

We know, from Exercise 1.16, that there exists a natural number

1/N < .

Now, if

n N,

then

2.3: REMARK A good way to attack a limit problem is to immediately examine the quantity |an L|,
which is what we did in Example 2.2 above. This is the quantity we eventually wish to show is less than

which shows that

1 1 |an 0| = = n N 0 = lim1/ n.

1 < 2 = , N

(2.5)

when

n N,

and determining which

to use is always the hard part. Ordinarily, some

algebraic manipulations can be performed on the expression

|an L|

that can help us gure out

Exercise 2.2.1
c.

exactly how to choose

N.

Just know that this process takes some getting used to, so practice!

a. Using the basic denition, prove that b. Using the basic denition, prove that

lim3/ (2n + 7) = 0. lim1/n2 = 0. 2 2 Using the basic denition, prove that lim n + 1 / n + 100n = 1. from the remark above; i.e., examine the quantity |an L|. n + n2 i = 1. n n2 i

HINT: Use the idea

d. Again, using the basic denition, prove that

lim

(2.6)

Remember the denition of the absolute value of a complex number.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

32

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

e. Using the basic denition, prove that

lim
n

n3 + n2 i = i. 1 n3 i

(2.7)

f. Let

an = (1) . Prove that 1 is not the limit of the sequence {an }. HINT: Suppose the sequence {an } does converge to 1. Use = 1, let N be the corresponding integer that exists in the denition, satisfying |an 1| < 1 for all n N, and then examine the quantity |an 1| for various n's to get a contradiction.

Exercise 2.2.2
a. Let

b.

{an } be a sequence of (real or complex) numbers, and let L be a number. Prove that L = liman if and only if for every positive integer k there exists an integer N, such that if n N then |an L| < 1/k. Let {cn } be a sequence of complex numbers, and suppose that cn L. If cn = an + bn i and L = a + bi, show that a = liman and b = limbn . Conversely, if a = liman and b = limbn , show that a + bi = lim (an + bn i) . That is, a sequence {cn = an + bn i} of complex numbers converges if and only if the sequence {an } of the real parts converges and the sequence {bn } of
the imaginary parts converges. HINT: You need to show that, given some hypotheses, certain quantities are less than

Part (c) of Exercise 1.8.3 should be of help.

Exercise 2.2.3
a. Prove that a constant sequence (an b. Prove that the sequence c.

c) converges to c. n2 +1 } diverges to . {2 13n n Prove that the sequence {(1) } does not converge to any number L. HINT: Argue by contradiction. Suppose it does converge to a number L. Use = 1/2, let N be the corresponding integer that exists in the denition, and then examine |an an+1 | for n N. Use the following
useful add and subtract trick:

|an an+1 | = |an L + L an+1 | |an L| + |L an+1 |.

(2.8)

2.3 Existence of Certain Fundamental Limits

evaluate limits of certain sequences.

We have, in the preceding exercises, seen that certain specic sequences converge. It's time to develop some general theory, something that will apply to lots of sequences, and something that will help us actually

Denition 2.3:
A sequence

nonincreasing if an an+1 for all n. It strictly decreasing if an > an+1 for all n.
A sequence number

{an }

of real numbers is called

is called

nondecreasing if an an+1 for all n, and it is called strictly increasing if an < an+1 for all n, and eventually nondecreasing
if there exists a natural

{an }

such that

exists a natural

an an+1 for all n N, and it is called eventually nonincreasing if there number N such that an an+1 for all n N. We make analogous denitions of

of real numbers is called

eventually strictly increasing and eventually strictly decreasing. It is ordinarily very dicult to tell whether a given sequence converges or not; and even if we know in theory that a sequence converges, it is still frequently dicult to tell what the limit is. The next theorem is

3 This

content is available online at <http://cnx.org/content/m36120/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

33

therefore very useful. It is also very fundamental, for it makes explicit use of the existence of a least upper bound.

{an } be a nondecreasing sequence of real numbers. Suppose that the set S of elements of the {an } is bounded above. Then the sequence {an } is convergent, and the limit L is given byL = supS = supan . Analogously, if {an } is a nonincreasing sequence that is bounded below, then {an } converges to inf an .
Let sequence

Theorem 2.1:

Proof:
L

We prove the rst statement. The second is done analogously, and we leave it to an exercise. Write for the supremum such that that that

supan . Let be a positive number. By Theorem 1.5, there exists an integer N aN > L , which implies that L aN < . Since {an } is nondecreasing, we then have an aN > L for all n N. Since L is an upper bound for the entire sequence, we know L an for every n, and so we have that |L an | = L an L aN <
(2.9)

for all

n N.

This completes the proof of the rst assertion.

Exercise 2.3.1
a. Prove the second assertion of the preceding theorem. b. Show that Theorem 2.1, p. 33 holds for sequences that are eventually nondecreasing or eventually nonincreasing. (Re-read the remark following the denition of the limit of a sequence.) The next exercise again demonstrates the denseness of the rational and irrational numbers in the set all real numbers.

of

Exercise 2.3.2
a. Let

b. c.

x be a real number. Prove that there exists a sequence {rn } of rational numbers such that x = limrn . In fact, show that the sequence {rn } can be chosen to be nondecreasing. HINT: For example, for each n, use Theorem 1.8, p. 18 to choose a rational number rn between x 1/n and x. ' Let x be a real number. Prove that there exists a sequence {r n } of irrational numbers such ' that x = limrn . Let z = x + iy be a complex number. Prove that there exists a sequence {n } = {n + in } of complex numbers that converges to z, such that each n and each n is a rational number. {an } and {bn } are two convergent sequences, and suppose that liman = a and limbn = b. {an + bn } is convergent and that lim (an + bn ) = a + b.
(2.10)

Exercise 2.3.3
Suppose

Prove that the sequence

HINT: Use an

/2

argument. That is, choose a natural number

N1

so that

n N1 ,

and choose a natural number

N2

so that

|bn b| < /2

for all

|an a| < /2 for all n N2 . Then let N be the

larger of the two numbers

N1

and

N2 . N.

The next theorem establishes the existence of four nontrivial and important limits. This time, the proofs are more tricky. Some clever idea will have to be used before we can tell how to choose the

Theorem 2.2:
1. Let

z C satisfy |z | < 1, n write limz = 0.

and dene

an = z n .

then the sequence

{an }

converges to

0.

We

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

34

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

2. Let 3. 4.

be a xed positive number greater than 1, and dene

an = b1/n .

See Theorem 1.11, p.

21. Then

limb1/n = 1. 1/n Let b be a positive number less than 1. Then limb = 1. 1/n If an = n , then liman = limn1/n = 1. liman = 1.
Again, we write

Proof:
z = 0, z = 0, and let > 0 be given. Let w = 1/|z |, and observe that w > 1. So, we may write w = 1 + h for some positive h. (That step is the clever idea for this n n argument.) Then, using the Binomial Theorem, w > nh, and so 1/w < 1/ (nh) . See part (a) of
We prove parts (1) and (2) and leave the rest of the proof to the exercise that follows. If claim (1) is obvious. Assume then that Exercise 1.7.2. But then

|z n 0| = |z n | = |z | = (1/w) = 1/wn < 1/ (nh) .

So, if

(2.11)

is any natural number larger than

1/ (h) ,
n

then

|z n 0| = |z n | = |z | <
for all

1 1 < nh Nh

(2.12)

n N.

This completes the proof of the rst assertion of the theorem.

an = b1/n = 1+xn , i.e., xn = b1/n 1, and observe rst that xn > 0. Indeed, 1/n 1/n since b > 1, it must be that the nth root b is also > 1. (Why?) Therefore, xn = b 1 > 0. n 1/n 1/n n (Again, writing b as 1 + xn is the clever idea.) Now, b = b = (1 + xn ) , which, again by the Binomial Theorem, implies that b > 1 + nxn . So, xn < (b 1) /n, and therefore
To see part (2), write

|b1/n 1| = b1/n 1 = xn <

whenever

b1 < n

(2.13)

n > / (b 1) ,

and this proves part (2).

Exercise 2.3.4
a. Prove part (3) of the preceding theorem. calculation: HINT: For

b 1,

use the following algebraic

|b1/n 1| = b1/n |1 (1/b)

1/n

| |1 (1/b) 1/b.

1/n

|,

(2.14)

and then use part (2) as applied to the positive number (2). HINT: Write

b. Prove part (4) of the preceding theorem. Explain why it does not follow directly from part

n1/n = 1 + hn .

Observe that

binomial theorem in the expansion sequence

n = (1 + hn ) .

hn > 0.

Then use the third term of the

c. Construct an alternate proof to part (2) of the preceding theorem as follows: Show that the

{b1/n }

is nonincreasing and bounded below by 1. Deduce, from Theorem 2.1, p. 33,

that the sequence converges to a number

L.

Now prove that

must be

1.

2.4 Denition of e

Part (4) of Theorem 2.2, p. 33 raises an interesting point. Suppose we have a sequence diverging to innity, and suppose we have another sequence can be said about the sequence
n {ab n }?

{an }, like {n}, that is bn

is going to 0. In

{bn }, an

like

{1/n},

that is converging to 0. What

The base

an

is blowing up, while the exponent

other words, there are two competing processes going on. If

is blowing up, then its powers ought to be

blowing up as well. On the other hand, anything to the 0 power should be 1, so that, as the exponents of the

4 This

content is available online at <http://cnx.org/content/m36124/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

35

elements of a sequence converge to 0, the sequence ought to converge to 1. This competition between the convergence of the base to innity and the convergence of the exponent to 0 makes it subtle, if not impossibly dicult, to tell what the combination does. For the special case of part (4) of Theorem 2.2, p. 33, the answer was 1, indicating that, in that case at least, the exponents going to 0 seem to be more important than the base going to innity. One can think up all kinds of such examples: on. We will see later that all sorts of things can happen. Of course there is the reverse situation. Suppose suppose

{(2n )

1/n

},{(n!)

1/n

},{(n!)

1/n2

},

and so

{an } is a sequence of numbers that decreases to 1, and {bn } is a sequence of numbers that diverges to innity. What can we say about the sequence {an bn }?

The base is tending to 1, so that one might expect that the whole sequence also would be converging to 1. On the other hand the exponents are blowing up, so that one might think that the whole sequence should blow up as well. Again, there are lots of examples, and they don't all work the same way. Here is perhaps the most famous such example.

Theorem 2.3:
For

Denition of

n 1,

dene

e. n an = (1 + 1/n) .

Then the sequence

{an }

is nondecreasing and bounded above,

Proof:

whence it is convergent. (We will denote the limit of this special sequence by the letter To see that

e.)

{an }

is nondecreasing, it will suce to prove that

computation below, we will use the fact (part (c)of Exercise 1.7.2) that if

an+1 /an 1 for all n. In the n x > 1 then (1 + x)

1 + nx.

So,
n+1 rcl aa n 1 (1+ n+1 ) 1 n 1+ ) ( n n+2 n+1 (n +1 ) +1 n ( nn ) n+1

= = =

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

n+1 n+1 n
(2.2)

n+1 n

n2 +2n n2 +2n+1

n+1

n+1 n

1 (n+1)2

n+1

n+1 n

1 (n + 1)

1 n+1

n+1 n

1 n+1

n+1 n n n+1

= 1,

as desired. We show next that

{an }

is bounded above.

This time, we use the binomial theorem, the geometric

progression, and Exercise 1.7.1.

rclan

= = < = = <

1 n n n n 1 k k=0 k n n nk 1 k k=0 2 2k n k n 2 k=0 1 2 n+1 1( 1 2) 2 1 1 2

1+

(2.16)

4,

as desired. That the sequence

{an }

converges is now a consequence of Theorem 2.1, p. 33.

2.4: REMARK We have now dened the real number e. Its central role in mathematics is not at all
evident yet; at this point we have no denition of exponential function, logarithm, or trigonometric functions. It does follow from the proof above that

is between 2 and 4, and with a little more

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

36

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

careful estimates we can show that actually

e 3.

For the moment, we will omit any further

discussion of its precise value. Later, in Exercise 4.5.4, we will show that it is an irrational number.

2.5 Properties of Convergent Sequences

just such a result.

Often, our goal is to show that a given sequence is convergent. However, as we study convergent sequences, we would like to establish various properties that they have in common. The rst theorem of this section is

Theorem 2.4:

Suppose

{an } is a convergent sequence of real or complex numbers.

Then the sequence

{an } forms

a bounded set.

Proof:

Write L = liman . Let be the positive number 1. Then, there exists a natural number N such that |an L| < 1 for all n N. By the backward triangle inequality, this implies that ||an | |L|| < 1 for all n N, which implies that |an | |L| + 1 for all n N. This shows that at least the tail of the sequence is bounded by the constant |L| + 1. Next, let K be a number larger than the nitely many numbers |a1 |, ..., |aN 1 |. Then, for any n,|an | is either less than K or |L| + 1. Let M be the larger of the two numbers K and |L| + 1. Then |an | < M for all n. Hence, the sequence {an } is bounded.

Note that the preceding theorem is a partial converse to Theorem 2.1, p. 33; i.e., a convergent sequence is necessarily bounded. Of course, not every convergent sequence must be either nondecreasing or nonincreasing, so that a full converse to Theorem 2.1, p. 33 is not true. For instance, take

z = 1/2

in part (1)

of Theorem 2.2, p. 33. It converges to 0 all right, but it is neither nondecreasing nor nonincreasing.

Exercise 2.5.1
a. Suppose

b.

{an } is a sequence of real numbers that converges to a number a, and assume that an c for all n. Prove that a c. HINT: Suppose not, and let be the positive number c a. Let N be a natural number corresponding to this choice of , and derive a contradiction. If {an } is a sequence of real numbers for which liman = a, and if a = 0, then prove that an = 0 for all large enough n. Show in fact that there exists an N such that |an | > |a|/2 for all n N. HINT: Make use of the positive number = |a|/2.

Exercise 2.5.2
a. If b.

numbers for which liman = a > 0, prove that lim an = {an } is a sequence of positive real a. HINT: Multiply the expression an a above and below by an + a. If {an } is a sequence of complex numbers, and liman = a, prove that lim|an | = |a|. HINT:
Use the backward triangle inequality.

Exercise 2.5.3
Suppose such that

{an } is a sequence of real numbers and that L = liman . Let M1 and M2 be real numbers M1 an M2 for all n. Prove that M1 L M2 . HINT: Suppose, for instance, that L > M2 . Make use of the positive number L M2 to derive

a contradiction. We are often able to show that a sequence converges by comparing it to another sequence that we already know converges. The following exercise demonstrates some of these techniques.

Exercise 2.5.4
Let

{an }

be a sequence of complex numbers.

5 This

content is available online at <http://cnx.org/content/m36126/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

37

a. Suppose that, for each b. Suppose Prove

n,|an | < 1/n. Prove that 0 = liman . {bn } is a sequence that converges to 0, and suppose that 0 = liman .

that, for each

n,|an | < |bn |.

The next result is perhaps the most powerful technique we have for showing that a given sequence converges to a given number.

Theorem 2.5:
{an }

Squeeze Theorem is a sequence of real numbers and that

Suppose that sequence

{an }
also

{bn }

and

numbers for which

Proof:

bn an cn for converges to L. |an L, |

all

n.

Suppose further that

{cn } are two sequences of real limbn = limcn = L. Then the

We examine the quantity computations:

employ some add and subtract tricks, and make the following

rcl|an L|

= =

|an bn + bn L| |an bn | + |bn L| an bn + |bn L| cn bn + |bn L| |cn bn | + |bn L| |cn L| + |L bn | + |bn L|.
(2.17)

|an L| < by making |cn L| < /3 and |bn L| < /3. So, let N1 be a positive |cn L| < /3 if n N1 , and let N2 be a positive integer so that |bn L| < /3 if n N2 . Then set N = max (N1 , N2 ) . Clearly, if n N, then both inequalities |cn L| < /3 and |bn L| < /3, and hence |an L| < . This nishes the proof.
So, we can make integer such that The next result establishes what are frequently called the limit theorems. Basically, these results show how convergence interacts with algebraic operations.

Theorem 2.6:
Let

{an }

and

{bn }

be two sequences of complex numbers with

a = liman

and

b = limbn .

Then

1. The sequence

{an + bn }

converges, and

lim (an + bn ) = liman + limbn = a + b.

2. The sequence

(2.18)

{an bn }

is convergent, and

lim (an bn ) = liman limbn = ab.

3. If all the

(2.19)

bn 's

as well as

are nonzero, then the sequence

{an /bn }

is convergent, and

lim

liman a an = = .(2.20) bn limbn b

Proof:
Part (1) is exactly the same as Exercise 2.3.3. Let us prove part (2). 36, both sequences {an } and {bn } are bounded. Therefore, let M be a |an | M and |bn | M for all n. Now, let > 0 be given. There exists an N1 such that |an a| < / (2M ) whenever n N1 , and there exists an N2 such that |bn b| < / (2M ) By Theorem 2.4, p. number such that

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

38

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

whenever again.

n N2 .

Let

be the maximum of

N1

and

N2 .

Here comes the add and subtract trick

rcl|an bn ab|

= = <

|an bn abn + abn ab| |an bn abn | + |abn ab| |an a||bn | + |a||b bn | |an a|M + M |bn b|
(2.21)

lim (an bn ) = ab. M be as in the previous paragraph, and let > 0 be given. There exists 2 an N1 such that |an a| < |b| / (4M ) whenever n N1 ; there also exists an N2 such that 2 |bn b| < |b| / (4M ) whenever n N2 ; and there exists an N3 such that |bn | > |b|/2 whenever n N3 . (See Exercise 2.5.1.) Let N be the maximum of the three numbers N1 , N2 and N3 . Then:
if which shows that To prove part (3), let

n N,

a n rcl| a bn b |

= = < < <

bn a | an b | bn b 1 |an b bn a| |bn b|

|an b bn a| |b|1 2 /2 (|an a||b| + |a||bn b|) |b2 |2 (M |an a| + M |bn b|) |b2 |2

(2.22)

if

n N.

This completes the proof.

2.5: REMARK

The proof of part (3) of the preceding theorem may look mysterious.

Where, for

|b| /4M come from? The answer is that one begins such a proof by |an /bn a/b| to see if by some algebraic manipulation one can discover how to control its size by using the quantities |an a| and |bn b|. The assumption that a = liman and b = limbn mean exactly that the quantities |an a| and |bn b| can be controlled by requiring n
instance, does this number examining the quantity to be large enough. The algebraic computation in the proof above shows that

an a 2 | (M |an a| + M |bn b|) 2 , bn b |b| |an a|

and

(2.23) so that

and one can then see exactly how small to make Indeed, this is the way most limit proofs work.

|bn b|

|an /bn a/b| < .

Exercise 2.5.5

If possible, determine the limits of the following sequences by using Theorem 2.2, p. 33, Theorem 2.3, Denition of Theorem, p. 37. a. b. c. d. e. f.

e., p.

35, Theorem 2.6, p. 37, and the squeeze theorem Theorem 2.5, Squeeze

{n1/n }. 1/n { n2 }. 1/n {(1 + n) }. { 1 + n2 }. 2/n {(1 + 1/n) }. 2n {(1 + 1/n) }.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

1/n3

39

g. h.

{(1 + 1/n) }. n {(1 1/n) }. HINT:

n2

Note that

1 1/n = {(1 1/ (2n)) }. 1/n {(n!) }.

3n

n1 = n

1
n n1

1
n1+1 n1

1 1 . 1 + n 1

(2.24)

i. j.

2.6 Subsequences and Cluster Points

Denition 2.4:
Let is determined by the sequence

{an } be a sequence of real or complex numbers. A subsequence of {an } is a sequence {bk } that {an } together with a strictly increasing sequence {nk } of natural numbers. The sequence {bk } is dened by bk = ank . That is, the k th term of the sequence {bk } is the nk th term of the original sequence {an }.

Exercise 2.6.1

be a sequence of numbers, and let

{an } is itself a subsequence of {an }. Thus, let {an } {bk } = {ank } be a subsequence of {an }. Suppose {cj } = {bkj } is a subsequence of the sequence {bk }. Prove that {cj } is a subsequence of {an }. What is the strictly increasing sequence {mj } of natural numbers for which cj = amj ?
Prove that a subsequence of a subsequence of

Here is an interesting generalization of the notion of the limit of a sequence.

Denition 2.5:
Let sequence

{an } be a sequence of real or complex numbers. A number x is called a cluster point of the {an } if there exists a subsequence {bk } of {an } such that x = limbk . The set of all cluster points of a sequence {an } is called the cluster set of the sequence.

Exercise 2.6.2

a. Give an example of a sequence whose cluster set contains two points. a sequence whose cluster set contains exactly cluster set is innite? b. Let c. If

Give an example of

points. Can you think of a sequence whose

{an }

be a sequence with cluster set

What is the cluster set for the sequence

{bn }

is a sequence for which

S. What is the cluster set for the sequence {an }? {a2 n }? b = limbn , and {an } is another sequence, what is the cluster

set of the sequence

{an bn }? {an }
is bounded above, then the cluster set

d. Give an example of a sequence whose cluster set is empty. e. Show that if the sequence Show also that if

is bounded above.

{an }

is bounded below, then

is bounded below.

f. Give an example of a sequence whose cluster set

is bounded above but not bounded below.

g. Give an example of a sequence that is not bounded, and which has exactly one cluster point.

Theorem 2.7:
Suppose

{an }

is a sequence of real or complex numbers.

1. (Uniqueness of limits) Suppose

liman = L,

and

liman = M.

Then

L = M.

That is, if the

limit of a sequence exists, it is unique.

6 This

content is available online at <http://cnx.org/content/m36129/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

40

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

2. If

and

L = liman , and if {bk } is a subsequence of {an }, then the sequence {bk } is convergent, limbk = L. That is, if a sequence has a limit, then every subsequence is convergent and

converges to that same limit.

Proof:
if Then

Suppose

liman = Landliman = M. Let be a positive number, and choose N1 so that |an L| < /2 n N1 , and choose N2 so that |an M | < /2 if n N2 . Choose an n larger than both N1 andN2 .

|L M | = |L an + an M | |L an | + |an M | < .
Therefore, since part (1).

(2.25)

|L M | <

for every positive

it follows that

LM = 0

or

L = M.

This proves

Next, suppose liman = L and let {bk } be a subsequence of {an }. We wish to show that limbk = L. Let > 0 be given, and choose an N such that |an L| < if n N. Choose a K so that nK N . (How?) Then, if k K, we have nk nK N, whence |bk L| = |ank L| < , which shows that limbk = L. This proves part (2).

2.6: REMARK The

number is

preceding theorem has the following interpretation. It says that if a sequence

converges to a number

L.

Indeed, if

that converges to

L, then the cluster set of the sequence contains only one number, and that x is a cluster point of the sequence, then there must be some subsequence x. But, by part (2), every subsequence converges to L. Then, by part (1), x = L.

Part (g) of Exercise 2.6.2 shows that the converse of this theorem is not valid. that is, the cluster set may contain only one point, and yet the sequence is not convergent. We give next what is probably the most useful fundamental result about sequences, the BolzanoWeierstrass Theorem. It is this theorem that will enable us to derive many of the important properties of continuity, dierentiability, and integrability.

Theorem 2.8:

Bolzano-Weierstrass

Every bounded sequence

{an }

of real or complex numbers has a cluster point.

In other words,

every bounded sequence has a convergent subsequence. The Bolzano-Weierstrass Theorem is, perhaps not surprisingly, a very dicult theorem to prove. We begin with a technical, but very helpful, lemma.

{an } be a bounded sequence of real numbers; i.e., assume that there exists an M such that |an | M for all n. For each n 1, let Sn be the set whose elements are {an , an+1 , an+2 , ...}. That is, Sn is just the elements of the tail of the sequence from n on. Dene xn = supSn = supkn ak .
Let Then 1. The sequence 2. The sequence 3. The sequence 4. The limit 5. If

Lemma 2.1:

{ xn } { xn } { xn } {bk }

is bounded (above and below). is non-increasing. converges to a number

x. {an }.
That is, there exists

of the sequence

{xn }

is a cluster point of the sequence

a subsequence

of the sequence

is any cluster point of the

{an } that converges to x. sequence {an }, then y x, where x

is the cluster point of part

(4). That is,

is the maximum of all cluster points of the sequence

{an }.

Proof:
Since

xn

is the supremum of the set

Sn ,

and since each element of that set is bounded between

and

M,

part (1) is immediate.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

41

Since

Sn+1 Sn ,

it is clear that

xn+1 = supSn+1 supSn = xn ,

showing part (2). The fact that the sequence p. 33.

(2.26)

{xn } converges to a number x is then a consequence of Theorem 2.1,

We have to show that the limit x of the sequence {xn } is a cluster point of {an }. Notice that {xn } may not itself be a subsequence of {an }, each xn may or may not be one of the numbers ak , so that there really is something to prove. In fact, this is the hard part of this lemma. To nish the proof of part (4), we must dene an increasing sequence {nk } of natural numbers for which the corresponding subsequence {bk } = {ank } of {an } converges to x. We will choose these natural numbers {nk } so that |x ank | < 1/k. Once we have accomplished this, the fact that the corresponding subsequence {ank } converges to x will be clear. We choose the nk 's inductively. First, using the fact that x = limxn , choose an n so that |xn x| = xn x < 1/1. Then, because xn = supSn , we may choose by Theorem 1.5, p. 14 some m n such that xn am > xn 1/1. But then |am x| < 1/1. (Why?) This m we call n1 . We have that |an1 x| < 1/1. Next, again using the fact that x = limxn , choose another n so that n > n1 and so that |xn x| = xn x < 1/2. Then, since this xn = supSn , we may choose another m n such that xn am > xn 1/2. This m we call n2 . Note that we have |an2 x| < 1/2. Arguing by induction, if we have found an increasing set n1 < n2 < ... < nj , for which |ani x| < 1/i for 1 i j, choose an n larger than nj such that |xn x| < 1/ (j + 1) . Then, since xn = supSn , choose an m n so that xn am > xn 1/ (j + 1) . Then |am x| < 1/ (j + 1), and we let nj +1 be this m. It follows that |anj +1 x| < 1/ (j + 1) . So, by recursive denition, we have constructed a subsequence of {an } that converges to x, and this completes the proof of part (4) of the lemma. Finally, if implying that (5). Now, using the lemma, we can give the proof of the Bolzano-Weierstrass Theorem.

y is any cluster point of {an }, and if y = limank , then nk k, and so ank xk , xk ank 0. Hence, taking limits on k, we see that x y 0, and this proves part

Proof:
If If

{an }

is a sequence of real numbers, this theorem is an immediate consequence of part (4) of the is a sequence of complex numbers, and if

preceding lemma.

an = bn + cn i

{an }

is bounded, then

{bn }

and

{cn }

are both bounded sequences of real numbers. See Exercise 1.8.5. So, by the preceding paragraph, there exists a subsequence

{bnk }

of

{bn }

that converges to a real number

b.

Now, the subsequence

{cnk }

is itself a bounded sequence of real numbers, so there is a subsequence

{cnkj }

that converges

to a real number converges to

b + ci;

i.e.,

c. By part (2) of Theorem 2.7, p. 39, we also have that the subsequence {bnkj } b. So the subsequence {ankj } = {bnkj + cnkj i} of {an } converges to the complex number {an } has a cluster point. This completes the proof.

There is an important result that is analogous to the Lemma above, and its proof is easily adapted from the proof of that lemma.

Exercise 2.6.3
Let that: a. b.

{an } be a bounded sequence of real numbers. {yn } is nondecreasing y = limyn is a cluster

and bounded above. point of

Dene a sequence

{yn } by yn = infkn ak . Prove

{an }.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

42

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

c. If

is any cluster point of

points of the sequence

{an }, then y z. That is, y is the minimum of all the cluster {an }. HINT: Let {n } = {an }, and apply the preceding lemma to {an }

{n }.

This exercise will then follow from that. is nonempty. It is

The Bolzano-Wierstrass Theorem shows that the cluster set of a bounded sequence also a bounded set itself. The following denition is only for sequences of real numbers.

However, like the Bolzano-Weierstrass

Theorem, it is of very basic importance and will be used several times in the sequel.

Denition 2.6:
Let

{an } S is If S is
If

be a sequence of real numbers and let

denote its cluster set.

nonempty and bounded above, we dene nonempty and bounded below, we dene

lim supan lim inf an

to be the supremum to be the inmum

If the sequence {an } of real numbers is not bounded above, we dene {an } is not bounded below, we dene lim inf an to be . If {an } diverges to , then we dene lim supan and lim inf an both to be . And, if {an } diverges to , we dene lim supan and lim inf an both to be . We call lim supan the limit superior of the sequence {an }, and lim inf an the limit inferior of {an }. if

supS of S. inf S of S. lim supan to be , and

Exercise 2.6.4
a. Suppose

{an }

is a bounded sequence of real numbers. Prove that the sequence

lemma following Theorem 2.8, Bolzano-Weierstrass, p. 40 converges to that the sequence b. Let

{xn } of the lim supan . Show also

{yn }

of Exercise 2.6.3 converges to

lim inf an .
(2.27)

{an }

be a not necessarily bounded sequence of real numbers. Prove that

lim supan = inf supak = limsupak .

n k n
and

n k n

lim inf an = supinf ak = liminf k nak .

n k n n
HINT: Check all cases, and use Lemma 2.1, p. 40 and Exercise 2.6.3. c. Let

(2.28)

{an }

be a sequence of real numbers. Prove that

lim supan = lim inf (an ) .

d. Give examples to show that all four of the following possibilities can happen. a. b. c. d.

(2.29)

lim supan is nite, and lim inf an = . lim supan = and lim inf an is nite. lim supan = and lim inf an = . both lim supan and lim inf an are nite.
The

The notions of limsup and liminf are perhaps mysterious, and they are in fact dicult to grasp.

previous exercise describes them as the resultof a kind of two-level process, and there are occasions when this description is a great help. However, the limsup and liminf can also be characterized in other ways that are more reminiscent of the denition of a limit. These other ways are indicated in the next exercise.

Exercise 2.6.5
Let a. For each that

{an } be a bounded sequence of real numbers with lim supan = L and lim inf an = l. Prove that L and l > 0, there exists an N such lim supan = L is the number x of limit of a specic sequence {xn }.
that

satisfy the following properties. for all

an < L +

n N.

HINT: Use the fact

the lemma following Theorem 2.8, and that

is the

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

43

b. For each c. d. e.

f.

> 0, and any natural number k, there exists a natural number j k such that aj > L . Same hint as for part (a). For each > 0, there exists an N such that an > l for all n N. For each > 0, and any natural number k, there exists a natural number j > k such that aj < l + . ' ' Suppose L is a number that satises parts (a) and (b). Prove that L is the limsup of {an }. ' HINT: Use part (a) to show that L is greater than or equal to every cluster point of {an }. ' Then use part (b) to show that L is less than or equal to some cluster point. ' ' If l is any number that satises parts (c) and (d), show that l is the liminf of the sequence {an }.

Exercise 2.6.6
a. Let

b.

{an } and {bn } be two bounded sequences of real numbers, and write L = lim supan M = lim supbn . Prove that lim sup (an + bn ) lim supan + lim supbn . HINT: Using part (a) of the preceding exercise, show that for every > 0 there exists a N such that an + bn < L + M + for all n N, and conclude from this that every cluster point y of the sequence {an + bn } is less than or equal to L + M. This will nish the proof, since lim sup (an + bn ) is a cluster point of that sequence. Again, let {an } and {bn } be two bounded sequences of real numbers, and write l = lim inf an and m = lim inf bn . Prove that lim inf (an + bn ) lim inf an + lim inf bn . HINT: Use part
and (c) of the previous exercise.

c. Find examples of sequences

{an }

and

{bn }

for which

lim supan = lim supbn = 1,

but

lim sup (an + bn ) = 0.

We introduce next another property that a sequence can possess. of the sequence It looks very like the denition of a

convergent sequence, but it diers in a crucial way, and that is that this denition only concerns the elements

Denition 2.7:
A sequence

{an }

and not the limit

L.

{an }

of real or complex numbers is a

Cauchy
then

sequence if for every

> 0,

there exists

a natural number

2.7: REMARK No doubt, this denition has something to do with limits.

and an

such that if

nN

and

mN

|an am | < .
Any time there is a positive

N,

we must be near some kind of limit notion. The point of the denition of a Cauchy It isn't that the terms of the

sequence is that there is no explicit mention of what the limit is. sequence are getting closer and closer to some number

L,

it's that the terms of the sequence are

getting closer and closer to each other. This subtle dierence is worth some thought.

Exercise 2.6.7

Prove that a Cauchy sequence is bounded. (Try to adjust the proof of Theorem 2.4, p. 36 to work for this situation.) The next theorem, like the Bolzano-Weierstrass Theorem, seems to be quite abstract, but it also turns out to be a very useful tool for proving theorems about continity, dierentiability, etc. not complete. In the proof, the completeness of the set of real numbers will be crucial. This theorem is not true in ordered elds that are

Theorem 2.9:
A sequence If

Cauchy Criterion of real or complex numbers is convergent if and only if it is a Cauchy sequence.

Proof:

{an }

liman = a {an }

then given

> 0,

choose

inequality, and by adding and subtracting Hence, if is convergent, then

N so that |ak a| < /2 if k N. From the triangle a, we obtain that |an am | < if n N and m N.

{an }

is a Cauchy sequence.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

44

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

Conversely, if

we use the fact that

{an } is a cauchy sequence, then {an } is bounded by {an } is a sequence of real or complex numbers.

the previous exercise. Now Let

be a cluster point of

{an }.

We know that one exists by the Bolzano-Weierstrass Theorem. Let us show that in fact this

number

x not only is a cluster point but that it is in fact the limit of the sequence {an }. Given > 0, choose N so that |an am | < /2 whenever both n and m N. Let {ank } be a subsequence of {an } that converges to x. Because {nk } is strictly increasing, we may choose a k so that nk > N and also so that |ank x| < /2. Then, if n N, then both n and this particular nk are larger than or equal to N. Therefore, |an x| |an ank | + |ank x| < . this completes the proof that x = liman .

2.7 A Little Topology

Denition 2.8:

We now investigate some properties that subsets of

R and C

may possess. We will dene closed sets, open

sets, and limit points of sets. These notions are the rudimentary notions of what is called topology. As in earlier denitions, these topological ones will be enlightening when we come to continuity. Let S be a subset of C. A complex number x is called a limit point of S {xn } of elements of S such that x = limxn . A set S C is called closed if every limit point of S belongs to S.

if there exists a sequence

Every limit point of a set of real numbers is a real number. Closed intervals sets in

[a, b] r

are examples of closed

R,

while open intervals and half-open intervals may not be closed sets. Similarly, closed disks

of radius

around a point

in

C,

and closed neighborhoods

N r (S )

of radius

around a set

Br (c) S C, are

closed sets, while the open disks or open neighborhoods are not closed sets. As a rst example of a limit point of a set, we give the following exercise.

Exercise 2.7.1
Let

be a nonempty bounded set of real numbers, and let

M = supS.

Prove that there exists

a sequence infs.

{an }

of elements of

such that

M = liman .

That is, prove that the supremum of a

bounded set of real numbers is a limit point of that set. State and prove an analogous result for HINT: Use Theorem 1.5, p. 14, and let

Exercise 2.7.2
a. Suppose

run through the numbers

1/n.

is a set of real numbers, and that

z = a + bi C

with

b = 0.

Show that

is not a

limit point of b. Let

S.

That is, every limit point of a set of real numbers is a real number. HINT:

Suppose false; write

a + bi = limxn , C.

and make use of the positive number

|b|. |z c| r. Br (0)
that

be a complex number, and let

S = B r (c)

be the set of all

zC

for which

Show that is not in

is a closed subset of

HINT: Use part (b) of Exercise 2.5.2.

c. Show that the open disk

Br (0)

is not a closed set in

by nding a limit point of

Br (0) . R. x
of a set

d. State and prove results analogous to parts b and c for intervals in e. Show that every element f. Let

is a limit point of

S. x is not a limit point of |y x| < , then y is not in S.

use the

be a subset of

C,

and let

be a complex number. Show that

if and only if there exists a positive number

such that if

That is, g.

the sequence

S B (x) = . HINT: To prove the  only if  part, argue by contradiction, and {1/n} as 's. Let {an } be a sequence of complex numbers, and let S be the set of all the an 's. What is dierence between a cluster point of the sequence {an } and a limit point of the set S ?

7 This

content is available online at <http://cnx.org/content/m36157/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

45

h. (h) Prove that the cluster set of a sequence is a closed set. HINT: Use parts (e) and (f ).

Exercise 2.7.3
a. Show that the set b. Show that if

of all rational numbers is not a closed set. Show also that the set of all

irrational numbers is not a closed set.

is a closed subset of

that contains

Q,

then

must equal all of

R.

Here is another version of the Bolzano-Weierstrass Theorem, this time stated in terms of closed sets rather than bounded sequences.

Theorem 2.10:
Let

be a bounded and closed subset of

Proof:
Let

subsequence that converges to an element of

C . Then S.

every sequence

{xn }

of elements of

has a

{xn }

be a sequence in

S.

Since

is bounded, we know by Theorem 2.8, Bolzano-Weierstrass,

p. 40 that there exists a subsequence

{xnk }

of

{xn }

that converges to some number

x.

Since each

x nk

belongs to

S,

it follows that

is a limit point of

S.

Finally, because

is a closed subset of

C,

it then follows that

x S.

We have dened the concept of a closed set. Now let's give the denition of an open set.

C. A point x S is called an interior point of S if there exists an > 0 such B (x) of radius around x is entirely contained in S. The set of all interior 0 0 points of S is denoted by S and we call S the interior of S. A subset S of C is called an open subset of C if every point of S is an interior point of S ; i.e., 0 if S = S . Analogously, let S be a subset of R. A point x S is called an interior point of S if there exists an > 0 such that the open interval (x , x + ) is entirely contained in S. Again, we denote the 0 0 set of all interior points of S by S and call S the interior of S. A subset S of R is called an open subset of R if every point of S is an interior point of S ; i.e., 0 if S = S .
Let

Denition 2.9:
S

be a subset of

that the open disk

Exercise 2.7.4
(a, b)

a. Prove that an open interval is an interior point of b. Prove that any disk an open set, where central point

' Br (c) is an open subset of C. Show also that the punctured disk Br (c) is ' Br (c) = {z : 0 < |z c| < r}, i.e., evrything in the disk Br (c) except the

(a, b) in R (a, b) .

is an open subset of

R;

i.e., show that every point of

c. Nr (S ) of R is Q
radius

c. Prove that the neighborhood e. (e) Prove that the set

around a set

is an open subset of

C.

d. Prove that no nonempty subset of

an open subset of

C. R.
We have seen in

of all rational numbers is not an open subset of

part (a) of Exercise 2.7.3 that nor closed.

is not a closed set. Consequently it is an example of a set

that is neither open nor closed. Show that the set of all irrational numbers is neither open

We give next a useful application of the Bolzano-Weierstrass Theorem, or more precisely an application of Theorem 2.10, p. 45. This also provides some insight into the structure of open sets.

Theorem 2.11:
Let

be a closed and bounded subset of

C,

and suppose

is a subset of an open set

U.

Then

there exists an

r>0

such that the neighborhood

Nr (S )

is contained in

U.

That is, every open set

containing a closed and bounded set

actually contains a neighborhood of

S.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

46

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

Proof:
If

{x}, then this theorem is asserting nothing more than the fact that x is in U, which it is if U is an open set. However, when S is an innite set, then the result is more subtle. We argue by contradiction. Thus, suppose there is no such r > 0 for which Nr (S ) U. then for each positive integer n there must be a point xn that is not in U, and a corresponding point yn S, such that |xn yn | < 1/n. Otherwise, the number r = 1/n would satisfy the claim of the theorem. Now, because the yn 's all belong to S, we know from Theorem 2.10, p. 45 that a subsequence {ynk } of the sequence {yn } must converge to a number y S. Next, we see that S
is just a singleton the interior of

|xnk y | |xnk ynk | + |ynk y |, <

and this quantity tends to 0. Hence, the subsequence to

1 + |ynk y |, nk
of the sequence

(2.30)

{xnk }

{xn }

also converges

y.

Finally, because y belongs to S and hence to the open set U, we know that there must exist an > 0 such that the entire disk B (y ) U. Then, since the subsequence {xnk } converges to y, there must exist ank such that |xnk y | < , implying that xnk B (y ) , and hence belongs to U. But this is our contradiction, because all of the xn 's were not in U. So, the theorem is proved.

We give next a result that claries to some extent the connection between open sets and closed sets. Always remember that there are sets that are neither open nor closed, and just because a set is not open

does not mean that it is closed. Theorem 2.12: Proof:

A subset

of

C (R ) S

is open if and only if its complement is open, and let us show that

= C \ S (R \ S ) S

is closed.

is closed. Suppose not. We will derive a S that converges to a contradiction. Suppose then that there is a sequence {xn } of elements of S number x that is not in S ; i.e., x is an element of S. Since every element of S is an interior point of S, there must exist an > 0 such that the entire disk B (x) (or interval (x , x + )) is a subset of S. Now, since x = limxn , there must exist anN such that |xn x| < for every n N. In particular, |xN x| < ; i.e., xN belongs to B (x) (or (x , x + )). This implies that xN S. and this is a contradiction. Hence, if S is open, then S is closed. But xN S, Conversely, assume that S is closed, and let us show that S must be open. Again we argue by contradiction. Thus, assuming that S is not open, there must exist a point x S that is not an interior point of S. Hence, for every > 0 the disk B (x) (or interval (x , x + )) is not entirely contained in S. So, for each positive integer n, there must exist a point xn such that |xn x| < 1/n Since S is a closed set, we must and xn / S. It follows then that x = limxn , and that each xn S. is closed, have that x S. But x S, and we have arrived at the desired contradiction. Hence, if S then S is open, and the theorem is proved.
First, assume that The theorem below, the famous Heine-Borel Theorem, gives an equivalent and dierent description of closed and bounded sets. This description is in terms of open sets, whereas the original denitions were interms of limit points. Any time we can nd two very dierent descriptions of the same phenomenon, we have found something useful.

Denition 2.10:
Let

be a subset of

(respectively

R.

By an

open cover

of

we mean a sequence

subsets of

(respectively

R)

such that

S Un ;

i.e., for every

x S there

exists an

{Un } of open n such that

x Un .
A subset

of

(respectively

if every open cover of

has a nite subcover. That is, if

R) is called compact, or is said to satisfy the Heine-Borel property, {Un } is an open cover of S, then there

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

47

exists an integer

such that

S N n=1 Un .

In other words, only a nite number of the open sets

are necessary to cover

S.

2.8: REMARK

The denition we have given here for a set being compact is a little less general

from the one found in books on topology. We have restricted the notion of an open cover to be a sequence of open sets, while in the general setting an open cover is just a collection of open sets. The distinction between a sequence of open sets and a collection of open sets is genuine in general topology, but it can be disregarded in the case of the topological spaces

Theorem 2.13:
S
of

and

C.

Heine-Borel Theorem (respectively

Proof:
integer

A subset

R)

is compact if and only if it is a closed and bounded set.

We prove this theorem for subsets Suppose rst that

of

C,

and leave the proof for subsets of

to the exercises.

SC

is compact, i.e., satises the Heine-Borel property. For each positive

Bn (0) . Then S Un , because C = Un . Hence, by N such that S N n=1 Un . But then S BN (0) , implying that S is bounded. Indeed, |x| N for all x S. is open. Next, still assuming that S is compact, we will show that S is closed by showing that S Thus, let x be an element of S. For each positive integer n, dene Un to be the complement of the closed set B1/n (x). Then each Un is an open set by Theorem 2.12, and we claim that {Un } is an open cover of S. Indeed, if y S, then y = x, and |y x| > 0. Choose an n so that 1/n < |y x|. Then y / B1/n (x), implying that y Un . This proves our claim that {Un } is an open cover of S. N Now, by the Heine-Borel property, there exists an N such that S n=1 Un . But this implies that for every z S we must have |z x| 1/N, and this implies that the disk B1/N (x) is entirely Therefore, every element x of S is an interior point of S. So, S is open, whence S contained in S. n,
dene

Un

to be the open set

the Heine-Borel property, there must exist an

is closed. This nishes the proof that compact sets are necessarily closed and bounded.

S satises the Heine{Un } that has no nite subcover. So, for each positive integer n there must exist an element xn S for which xn / n k=1 Uk . Otherwise, there would be a nite subcover. By Theorem 2.10, p. 45, there exists a subsequence {xnj } of {xn } that converges to an element x of S. Now, because {Un } is an open cover of S, there must exist an N such that x UN . Because UN is open, there exists an > 0 so that the entire disk B (x) is contained in UN . Since x = limxnj , there exists a J so that |xnj x| < if j J. Therefore, if j J, then xnj UN . But the sequence {nj } is strictly increasing, so that there exists a j ' J such that nj ' > N, and by the choice of the point xn ' , we know that xn ' / N k=1 Uk . We have j j
Conversely, assume that is both closed and bounded. We must show that Borel property. Suppose not. Then, there exists an open cover arrived at a contradiction, and so the second half of the theorem is proved.

Exercise 2.7.5
a. Prove that the union b. Prove that the union HINT: Use

A B of two open sets is open and the intersection A B is also open. A B of two closed sets is closed and the intersection A B is also closed. B=A B, and A B=A B. Theorem 2.12, p. 46 and the set equations A AB AB
of two bounded sets is bounded and the intersection of two compact sets is compact and the intersection

These set equations are known as Demorgan's Laws. c. Prove that the union bounded. d. Prove that the union compact. e. Prove that the intersection of a compact set and a closed set is compact. f. Suppose

AB AB

is also is also

is a compact set in

and

is a positive real number.

Prove that the closed

neighborhood

N r (S )

of radius

around

is compact. HINT: To see that this set is closed,

show that its coplement is open.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

48

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

8

2.8 Innite Series

and exponential functions.

Probably the most interesting and important examples of sequences are those that arise as the partial sums of an innite series. In fact, it will be innite series that allow us to explain such things as trigonometric

Denition 2.11:
Let

{an } 0 be a sequence sequence {SN } dened by

of real or complex numbers. By the

innite series

an

we mean the

SN =
n=0
The sequence

an .

(2.31)

{SN }

is called the

innite series is said to be partial sums converges to An innite series series If

S.The sum of an innite series is the limit of its partial sums. an is called absolutely summable or absolutely convergent if the innite
If it is convergent but not absolutely convergent,

sequence of partial sums of the innite series an , and the summable to a number S, or to be convergent, if the sequence {SN } of

|an | is convergent. an is not convergent, it is called divergent.

it is called

conditionally convergent.

A few simple formulas relating the

an 's

and the

SN 's

are useful: (2.32)

SN = a0 + a1 + a2 + ... + aN , SN +1 = SN + aN +1 ,
and

(2.33)

SM SK =
n=K +1
for

an = aK +1 + aK +2 + ... + AM ,

(2.34)

M > K.

2.9: REMARK Determining whether or not a given innite series converges is one of the most im-

portant and subtle parts of analysis. Even the rst few elementary theorems depend in deep ways on our previous development, particularly the Cauchy criterion.

Theorem 2.14:
{an }

Let

be a sequence of nonnegative real numbers. Then the innite series

an

is summable if

and only if the sequence

Proof:
If

{SN }

of partial sums is bounded. is convergent, whence bounded according to Theorem 2.4, p.

an

is summable, then So, if

{SN }

36. Conversely, we see from the hypothesis that each

an 0

that

{SN }

is nondecreasing (SN +1

SN + aN +1 SN ).

{SN }

is bounded, then it automatically converges by Theorem 2.1, p. 33,

and hence the innite series

an

is summable.

The next theorem is the rst one most calculus students learn about innite series. Unfortunately, it is often misinterpreted, so be careful! Both of the proofs to the next two theorems use Theorem 2.9, Cauchy Criterion, p. 43, which again is a serious and fundamental result about the real numbers. Therefore, these two theorems must be deep results themselves.

8 This

content is available online at <http://cnx.org/content/m36135/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

49

Theorem 2.15: Proof:

Let

an

be a convergent innite series. Then the sequence

{an }

is convergent, and

liman = 0.

Because

an

is summable, the sequence exists an

{SN }

is convergent and so is a Cauchy sequence. Thereand

fore, given an

n > N0 ,

let

> 0, there m = n 1. We

N0

so that

have then that

|Sn Sm | < whenever both n |an | = |Sn Sm | < , which completes

m N0 .

If

the proof.

2.10: REMARK Note that this theorem is not an if and only if  theorem.
to show that an innite series is

The harmonic series (part

(b) of Exercise 2.8.2 below) is the standard counterexample. The theorem above is mainly used

not summable.
an

If we can prove that the sequence

{an }

does not

converge to 0, then the innite series

does not converge. The misinterpretation of this result

referred to above is exactly in trying to apply the (false) converse of this theorem.

Theorem 2.16:
If

an

is an absolutely convergent innite series of complex numbers, then it is a convergent

innite series. (Absolute convergence implies convergence.)

Proof:
If

sums for

{SN } denotes the sequence of partial sums for |an |, then

M

an , and if {TN } denotes the sequence of partial

M

|SM SN | = |
n=N +1
for all

an |
n=N +1

|an | = |TM TN |

(2.35)

and

the inequality above, as well.) This

{TN } is convergent and hence it is a Cauchy sequence. So, by {SN } must also be a Cauchy sequence. (If |TN TM | < , then |SN SM | < implies that an is convergent. M.
We are given that

Exercise 2.8.1: The Innite Geometric Series

Let

z be a complex number, and dene a sequence {an } by an = z n . Consider the innite series an . Show that n=0 an converges to a number S if and only if |z | < 1. Show in fact that S = 1/ (1 z ) , when |z | < 1. HINT: Evaluate explicitly the partial sums SN , and then take their limit. Show that SN = 1z N +1 1z .

Exercise 2.8.2

1 n=1 n(n+1) converges to 1, by computing explicit formulas for the partial sums. HINT: Use a partial fraction decomposition for the an 's. b. (The Harmonic Series.) Show that n=1 1/n diverges by verifying that S2k > k/2. HINT:
a. Show that Group the terms in the sum as follows,

1+

1 + 2

1 1 + 3 4

1 1 1 1 + + + 5 6 7 8

1 1 1 + + ... + 9 10 16

+ ...,

(2.36)

and then estimate the sum of each group. Remember this example as an innite series that diverges, despite the fact that is terms tend to 0. The next theorem is the most important one we have concerning innite series of numbers.

Theorem 2.17:
Suppose

Comparison Test

{an }

and

{bn }

are two sequences of nonnegative real numbers for which there exists a

positive integer

and a constant

such that

bn Can

for all

n M.

If the innite series

an

converges, so must the innite series

bn .

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

50

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

Proof:
We will show that the sequence Write

{TN }

of partial sums of the innite series

bn

is a bounded

sequence. Then, by Theorem 2.14, p. 48, the innite series

bn

must be summable.

SN

for the

N th

partial sum of the convergent innite series

an .

Because this series

is summable, its sequence of partial sums is a bounded sequence. Let B be a number such that

SN B

for all

N.

We have for all

N >M = = =

that

rclTN

N n=1 bn M N n=1 bn + n=M +1 bn M N n=1 bn + n=M +1 Can M N n=1 bn + C n=M +1 an M N n=1 bn + C n=1 an M n=1 bn + CSN M n=1 bn + CB,

(2.37)

which completes the proof, since this nal quantity is a xed constant.

Exercise 2.8.3
a. Let

{an }

and

{bn }

be as in the preceding theorem. Show that if

bn

diverges, then

an

also must diverge. b. Show by example that the hypothesis that the nonnegative can not be dropped.

an 's

and

bn 's

of the Comparison Test are

Exercise 2.8.4: The Ratio Test

Let

{an }

be a sequence of positive numbers. show that

a. If

b. c. d.

an converges. HINT: If lim supan+1 /an = < 1, let < < 1. Using part (a) of Exercise 2.6.5, show that there exists an N such that for all n > N we must have an+1 /an < , or equivalently an+1 < an , and k therefore aN +k < aN . Now use the comparison test with the geometric series k. If lim inf an+1 /an > 1, show that an diverges. As special cases of parts (a) and (b), show that {an } converges if limn an+1 /an < 1, and diverges if limn an+1 /an > 1. Find two examples of innite series' an of positive numbers, such that liman+1 /an = 1 for
be a number for which both examples, and such that one innite series converges and the other diverges.

lim supan+1 /an < 1,

Exercise 2.8.5
a. Derive the Root Test: If

b.

{an } is a sequence of positive numbers for which lim supan < 1, 1/n then an converges. And, if lim inf an > 1, then an diverges. Let r be a positive integer. Show that 1/nr converges if and only if r 2. HINT: Use Exercise 2.8.2 and the Comparison Test for r = 2. 1/ n2 + 1 ,
for

1/n

c. Show that the following innite series are summable.

n/2n ,

an /n!,

(2.38)

any complex number.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

51

Exercise 2.8.6
Let

{an }

and

{bn }

be sequences of complex numbers, and let

{SN }

denote the sequence of partial

sums of the innite series

an .
N

Derive the Abel Summation Formula:

N 1

an bn = SN bN +
n=1 n=1

Sn (bn bn+1 ) .

(2.39)

The Comparison Test is the most powerful theorem we have about innite series of positive terms. Of course, most series do not consist entirely of positive terms, so that the Comparison Test is not enough. The next theorem is therefore of much importance.

Theorem 2.18:
Suppose

Alternating Series Test is an alternating sequence of real numbers; i.e., their signs alternate. As-

{a1 , a2 , a3 , ...}

sume further that the sequence

{|an |}

is nonincreasing with

0 = lim|an |.

Then the innite series

Proof:

an

converges.

Assume, without loss of generality, that the odd terms the even terms

a2n+1

of the sequence

a2n

are negative. We collect some facts about the partial sums

{an } are positive and SN = a1 + a2 + ... + aN

of the innite series

an . S2N
is less than the following odd partial sum

1. Every even partial sum

And every odd partial sum

S2N +1

is greater than the following even partial sum

S2N +1 = S2N + a2N +1 , S2N +2 =

S2N +1 + a2N +2 .
2. Every even partial sum

S2N is less than or equal to the next even partial sum S2N +2 = S2N + a2N +1 + a2N +2 , implying that the sequence of even partial sums {S2N } is nondecreasing. S2N +1 is greater than or equal to the next odd partial sum S2N +3 = S2N +1 + a2N +2 + a2N +3 , implying that the sequence of odd partial sums {S2N +1 } is noninS2N +1
is bounded below by

3. Every odd partial sum creasing.

4. Every odd partial sum even partial sum

S2N

is bounded above by

S1 .

For,

S2 . For, S2N +1 > S2N S2 . S2N < S2N +1 S1 .

And, every

5. Therefore, the sequence

sequence must then have a limit, which we denote by must have a limit, which we denote by Now

{S2N } of even partial sums is nondecreasing and bounded above. That Se . Similarly, the sequence {S2N +1 } So .

of odd partial sums is nonincreasing and bounded below. This sequence of partial sums also

So Se = limS2N +1 limS2N = lim (S2N +1 S2N ) = lima2N +1 = 0,

showing that exists an

(2.40)

Se = So , and we denote this common limit by S. Finally, given an > 0, there N1 so that |S2N S | < if 2N N1 , and there exists an N2 so that |S2N +1 S | < if 2N + 1 N2 . Therefore, if N max (N1 , N2 ) , then |SN S | < , and this proves that the innite
series converges.

Exercise 2.8.7: The Alternating Harmonic Series

a. Show that b. Let

{an }

n n=1 (1) /n converges, but that it is not absolutely convergent. be an alternating series, as in the preceding theorem. Show that the sum

S=

an

is trapped between d. Show that

SN

and

SN +1 ,

and that

|S SN | |aN |.

c. State and prove a theorem about eventually alternating innite series.

z n /n converges if and only if |z | 1, and z = 1. HINT: Use the Abel Summation

Formula to evaluate the partial sums.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

52

CHAPTER 2. THE LIMIT OF A SEQUENCE OF NUMBERS

Exercise 2.8.8
Let

s = p/q

be a positive rational number.

a. For each b. Prove

x > 0, show that there exists a unique y > 0 such that y s = x; i.e., y p = xq . that 1/ns converges if s > 1 and diverges if s 1. HINT: Group the terms

as in

part (b) of Exercise 2.8.2.

x be a real number, and suppose that {pN /qN } is a sequence of rational numbers for x = limpN /qN and x = pN /qN for any N. If limqN |x pN /qN | = 0, then x is irrational.
Let

Theorem 2.19:

Test for Irrationality which

Proof:

We prove the contrapositive statement; i.e., if

x = p/q

is a rational number, then

limqN |x

pN /qN | = 0.

We have

x pN /qN = p/q pN /qN =

Now the numerator

pqN qpN . qqN x = p/q = pN /qN ,

is an integer, we have that

(2.41) which

pqN qpN

is not 0 for any

N.

For, if it were, then

we have assumed not to be the case. Therefore, since

pqN qpN

|x pN /qN | = |
So,

pqN qpN 1 . | qqN |qqN |

(2.42)

qN |x pN /qN |
and this clearly does not converge to 0.

1 , |q |

(2.43)

Exercise 2.8.9
n n n=0 (1) /2 . Prove that x is a rational number. n n2 b. Let n=0 (1) /2 . Prove that y is an irrational number. HINT: The partial sums of this series are rational numbers. Now use the preceding theorem and part (b) of Exercise 2.8.7
a. Let

x= y=

(The Alternating Harmonic Series).

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

Chapter 3

Functions and Continuity

3.1 Functions and Continuity Denition of the Number
functions.

The concept of a function is perhaps the most basic one in mathematical analysis. The objects of interest in our subject can often be represented as functions, and the  unknowns in our equations are frequently Therefore, we will spend some time developing and understanding various kinds of functions, including functions dened by polynomials, by power series, and as limits of other functions. In particular, we introduce in this chapter the elementary transcendental functions. We begin with the familiar set theoretical notion of a function, and then move quickly to their analytical properties, specically that of continuity. The main theorems of this chapter include: 1. The

2. the theorem that asserts that a 3. 4.

Intermediate Value Theorem (Theorem 3.6, Intermediate Value Theorem, p. 64), continuous real-valued function on a compact set attains a maximum and minimum value (Theorem 3.8, p. 65), A continuous function on a compact set is uniformly continuous (), The Identity Theorem for Power Series Functions (Theorem 3.14, Identity Theorem, p. 71), uniform limit of a sequence of continuous functions is continuous (Theorem 3.18, The uniform limit of continuous functions is continuous., p. 76), and the Weierstrass M -Test (Theorem 3.19, Weierstrass M-Test, p. 77).
2

5. The denition of the real number

6. The theorem that asserts that the 7.

3.2 Functions
Denition 3.1:
Let

and

be sets. A

function from S

into

element

in

a unique element denoted by

T (notation f : S T ) f (x) in T. T.

is a rule that assigns to each

It is useful to think of a function as a mechanism or black box. We use the elements of inputs to the function, and the outputs are elements of the set If

as

f :ST
The

f. an x
of

range
f

is a function, then or

image
S

is called the

domain of f, and the set T

y
in the codomain

is called the

codomain

of

is the set of all elements

for which there exists

in the domain

such that

y = f ( x) . S

We denote the range by

f (S ) .

The codomain is the set

of all potential outputs, while the range is the set of actual outputs. Suppose is a function from a set into a set

T.

If

containing all the elements

tT

for which there exists an

A S, we write f (A) for the subset of T s A such that t = f (s) . We call f (A)

1 This 2 This

content is available online at <http://cnx.org/content/m36131/1.2/>. content is available online at <http://cnx.org/content/m36141/1.2/>. Available for free at Connexions <http://cnx.org/content/col11249/1.1>

53

54

CHAPTER 3. FUNCTIONS AND CONTINUITY

the

A under f. Similarly, if B T, we write f 1 (B ) for the subset of S containing all s S such that f (s) B, and we call the set f 1 (B ) the inverse image or preimage 1 of B. The symbol f (B ) is a little confusing, since it could be misinterpreted as the image of the 1 set B under a function called f . We will discuss inverse functions later on, but this notation is not meant to imply that the function f has an inverse. If f : S T, then the graph of f is the subset G of the Cartesian product S T consisting of all the pairs of the form (x, f (x)) . If f : S R is a function, then we call f a real-valued function, and if f : S C, then we call f a complex-valued function. If f : S C is a complex-valued function, then for each x S the complex number f (x) can be written as u (x) + iv (x) , where u (x) and v (x) are the real and imaginary parts of the complex number f (x) . The two real-valued functions u : S R and v : S R are called respectively the real and imaginary parts of the complex-valued function f. If f : S T and S R, then f is called a function of a real variable, and if S C, then f is
of the elements called a function of a If the range of The function

image

complex variable.

f equals f :ST

the codomain, then

is called

onto.
implies that

is called one-to-one if

f (x1 ) = f (x2 )
a subset

x1 = x2 . f : S T,
we

The domain of are free to regard

f is the set of x's for which f (x) f as having a smaller domain, i.e.,

is dened. If we are given a function

S'

of

S.

Although this restricted function is

in reality a dierent function, we usually continue to call it by the same name The codomain of

f.

Enlarging the domain of a

function, in some consistent manner, is often impossible, but is nevertheless frequently of great importance.

is distinguished from the range of f, which is frequently a proper subset of the codomain.

For example, since every real number is a complex number, any real-valued function (special kind of ) complex-valued function.

f :S R

is also a

We consider in this book functions either of a real variable or of complex variable. that is, the domains of functions here will be subsets either of about this, for this distinction is thinking of by denoting real variables with the letter

not always made.

R or of C. Frequently, we will indicate what kind of variable we are x and complex variables with the letter z. Be careful

Many functions, though not all by any means, are dened by a single equation:

y = 3x 7, y = x2 + x + 1 x2 + y 2 = 4,
(How does this last equation dene a function?)

(3.1)

2/3

(3.2)

(3.3)

1 x7 y 11

2/3

= (x/ (1 y ))

8/17

(3.4)

(How does this equation determine a function?) There are various types of functions, and they can be combined in a variety of ways to produce other functions. It is necessary therefore to spend a fair amount of time at the beginning of this chapter to present these denitions.

Denition 3.2:
If

and

and if

is a complex number, we dene

g are two complex-valued functions with the same domain S, i.e., f : S C f + g, f g, f /g (if g (x) is never 0), and cf

and

g : S C,

by the familiar

formulas:

(f + g ) (x) = f (x) + g (x) , (f g ) (x) = f (x) g (x) ,

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

(3.5)

(3.6)

55

(f /g ) (x) = f (x) /g (x) ,

and

(3.7)

(cf ) (x) = cf (x) .

If

(3.8) and

and

are real-valued functions, we dene functions

max (f, g )

min (f, g )

by (3.9)

[max (f, g )] (x) = max (f (x) , g (x))

(the maximum of the numbers

f ( x)

and

g (x)),

and (3.10)

[min (f, g )] (x) = min (f (x) , g (x)) ,

(the minimum of the two numbers

f ( x)

and

g (x)).
such that

bounded
domain

If

is either a real-valued or a complex-valued function on a domain if there exists a positive number

|f (x) | M

for all

S, then x S.

we say that

is

There are two special types of functions of a real or complex variable, the even functions and the odd functions. In fact, every function that is dened on all of

equals

R or C (or, more generally, S ) can be written uniquely as a sum of an even part and an odd part.

any function whose This decomposition

of a general function into simpler parts is frequently helpful.

Denition 3.3:
A function

whose domain

its domain. It is called an

odd

equals

function if

S, is called an even function if f (z ) = f (z ) f (z ) = f (z ) for all z in its domain.

for all

in

We next give the denition for perhaps the most familiar kinds of functions.

Denition 3.4:
A nonzero

polynomial

or

polynomial function is a complex-valued function of a complex variable,

n

p : C C,

that is dened by a formula of the form

p (z ) =
k=0

ak z k = a0 + a1 z + a2 z 2 + ... + an z n ,

(3.11)

where the ak 's are complex numbers and an = 0. The integer n is called the degree of the polynomial p and is denoted by deg (p) . The numbers a0 , a1 , ..., an are called the coecients of the polynomial. The domain of a polynomial function is all of C ; i.e., p (z ) is dened for every complex number z. For technical reasons of consistency, the identically 0 function is called the of its coecients are 0 and its degree is dened to be A where

zero polynomial.

All

rational function is a function r

q
is a nonzero polynomial and

. r (z ) = p (z ) /q (z ) ,

that is given by an equation of the form

is a (possibly zero) polynomial. The domain of a rational

function is the set

of all

zC

for which

q (z ) = 0,

i.e., for which

r (z )

is dened.

Two other kinds of functions that are simple and important are step functions and polygonal functions. Let [a, b] be a closed bounded interval of real numbers. By a partition of [a, b] we mean a nite set P = {x0 < x1 < ... < xn } of n + 1 points, where x0 = a and xn = b. The n intervals {[xi1 , xi ]}, for 1 i n, are called the closed subintervals of the partition P, and the n intervals {(xi1 , xi )} are called the open subintervals of P. We write P for the maximum of the numbers (lengths of the subintervals) {xi xi1 }, and call the number P the mesh size of the partition P. A function h : [a, b] C is called a step function if there exists a partition P = {x0 < x1 < ... < xn } of [a, b] and n numbers {a1 , a2 , ..., an } such that h (x) = ai if xi1 < x < xi . That is, h is a step function if it is a constant function on each of the (open) subintervals (xi1 , xi ) determined

Denition 3.5:

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

56

CHAPTER 3. FUNCTIONS AND CONTINUITY

by a partition A function each

P.

Note that the values of a step function at the points

{ xi }

of the partition are not

restricted in any way. exists a partition

l : [a, b] R is called a polygonal function, or a piecewise linear function, if there P = {x0 < x1 < ... < xn } of [a, b] and n + 1 numbers {y0 , y1 , ..., yn } such that for x [xi1 , xi ] ,l (x) is given by the linear equation l (x) = yi1 + mi (x xi1 ) ,
(3.12)

mi = (yi yi1 ) / (xi xi1 ) . That is, l is a polygonal function if it is a linear function on [xi1 , xi ] determined by a partition P. Note that the values of a piecewise linear function at the points {xi } of the partition P are the same, whether we think of xi in the interval [xi1 , xi ] or [xi , xi+1 ] . (Check the two formulas for l (xi ) .) The graph of a piecewise linear function is the polygonal line joining the n + 1 points {(xi , yi )}. There is a natural generalization of the notion of a step function that works for any domain S, e.g., a rectangle in the plane C. Thus, if S is a set, we dene a partition of S to be a nite collection {E1 , E2 , ..., En } of subsets of S for which
where each of the closed subintervals 1. 2.

n i=1 Ei = S, and Ei Ej = if i = j.

Then, a

step function

on

would be a function

that is constant on each subset

Ei .

We will

encounter an even more elaborate generalized notion of a step function in Chapter V, but for now we will restrict our attention to step functions dened on intervals

[a, b] .

The set of polynomials and the set of step functions are both closed under addition and multiplication, and the set of rational functions is closed under addition, multiplication, and division.

Exercise 3.2.1
deg (p

a. Prove that the sum and product of two polynomials is again a polynomial.

Show that

+ q ) max (deg (p) , deg (q ))

and deg (pq )

deg (p)

+ deg (q ) .

Show that a constant

function is a polynomial, and that the degree of a nonzero constant function is 0. b. Show that the set of step functions is closed under addition and multiplication. Show also that the maximum and minimum of two step functions is again a step function. (Be careful to note that dierent step functions may be determined by dierent partitions. For instance, a partition determining the sum of two step functions may be dierent from the partitions determining the two individual step functions.) Note, in fact, that a step function can be determined by innitely many dierent partitions. Prove that the sum, the maximum, and the minimum of two piecewise linear functions is again a piecewise linear function. Show by example that the product of two piecewise linear functions need not be piecewise linear. c. Prove that the sum, product, and quotient of two rational functions is again a rational function. d. Prove the

ak z k is a nonzero polynomial of degree n, and if c is a complex number for which p (c) = 0, then there exists a nonzero polynomial n1 q (z ) = j =0 bj z j of degree n 1 such that p (z ) = (z c) q (z ) for all z. That is, if c is a root of p, then z c is a factor of p. Show also that the leading coecient bn1 of q equals the leading coecient an of p. HINT: Write
If

Root Theorem:

p (z ) =

n k=0

p (z ) = p (z ) p (c) =
k=0
e. Let

ak z k ck = ....

(3.13)

be a function whose domain

equals

S.
and

Dene functions

fe

and

fo

by the formulas

fe ( z ) =

f (z ) + f (z ) 2

fo (z ) =

f (z ) f (z ) . 2

(3.14)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

57

also that, if function.

fe is an even function, that fo is an odd function, and that f = fe + fo . Show f = g + h, where g is an even function and h is an odd function, then g = fe and h = fo . That is, there is only one way to write f as the sum of an even function and an odd
Show that

f. Use part (e) to show that a polynomial coecients are even ones, i.e., the g. Suppose

p is an a2k 's. Show

even function if and only if its only nonzero also that a polynomial is an odd function if

and only if its only nonzero coecients are odd ones, i.e., the

p (z ) =

n k=0

a2k+1 's.

a2k z 2k

is a polynomial that is an even function. Show that

p (iz ) =
k=0
where h. If

(1) a2k z 2k = pa (z ) ,

(3.15)

pa

is the polynomial obtained from

p by alternating the signs of its nonzero coecients.

q (z ) =

n k=0

a2k+1 z 2k+1

is a polynomial that is an odd function, show that

q (iz ) = i
k=0
where again coecients. i. If

(1) a2k+1 z 2k+1 = iq a (z ) , q

(3.16)

qa

is the polynomial obtained from

by alternating the signs of its nonzero

is any polynomial, show that

a p (iz ) = pe (iz ) + po (iz ) = pa e (z ) + ipo (z ) ,

and hence that

(3.17)

pe (iz ) = pa e (z )

and

po (iz ) = ipa o (z ) .

3.3 Polynomial Functions

If

n m k j k=0 ak z and q (z ) = j =0 bj z are two polynomials, it certainly seems clear that they determine the same function only if they have identical coecients. This is true, but by no means an obvious fact.

p (z ) =

Also, it seems clear that, as its leading term

|z | gets larger and larger, a polynomial function is more and more comparable to

an z n .

We collect in the next theorem some elementary properties of polynomial functions,

and in particular we verify the above uniqueness of coecients result and the behavior at innity result.

Theorem 3.1:
1. Suppose at most 2. If

p (z ) = n distinct

n k k=0 ak z is a nonconstant polynomial of degree complex numbers.

n > 0.

Then

p (z ) = 0 r

for

is a polynomial for which

r (z ) = 0

for an innite number of distinct points, then

is the

zero polynomial. That is, all of its coecients are 0. 3. Suppose of distinct points. Then 4. Let

p and q are nonzero polynomials, and assume that p (z ) = q (z ) for an innite number p (z ) = q (z ) for all z, and p and q have the same coecients. That
n
be a polynomial of degree

is, they are the same polynomial.

|cn | n n |z | |p (z ) | M |z | (3.18) 2 for all complex numbers z for which |z | B. That is, For all complex numbers z with |z | B, n the numbers |p (z ) | and |z | are comparable.
3 This
content is available online at <http://cnx.org/content/m36147/1.2/>.

p (z ) = j =0 cj z j m and B such that

n > 0.

Then there exist positive constants

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

58

CHAPTER 3. FUNCTIONS AND CONTINUITY

5. If

f : [0, ) C is p (x) for all x 0.

dened by

f (x) =

x,

then there is no polynomial

for which

f (x) =

That is, the square root function does not agree with any polynomial

function.

Proof:
We prove part (1) using an argument by contradiction. Thus, suppose there does exist a counterexample to the claim, i.e., a nonzero polynomial for which

p (cj ) = 0 n0 .

for all

1 j n + 1.

From the set of all such counterexamples, let

p of degree n and n +1 distinct points {c1 , c2 , ..., cn+1 } p0 be one

with minimum degree smaller than

n0 .

That is, the claim in part (1) is true for any polynomial whose degree is

We write

n0

p0 (z ) =
k=0
and we suppose that

ak z k ,

(3.19)

We use next the Root Theorem (part (d) of Exercise 3.2.1) to write

p0 (cj ) = 0 for j = 1 to n0 + 1, where these ck 's are distinct complex numbers. p0 (z ) = (z cn0 +1 ) q (z ) , where n0 1 q (z ) = k=0 bk z k . We have that q is a polynomial of degree n0 1 and the leading coecient an0 of p0 equals the leading coecient bn0 1 of q. Note that for 1 j n0 we have 0 = p0 (cj ) = (cj cn0 +1 ) q (cj ) ,
which implies that nonzero polynomial (3.20)

q (cj ) = 0 for 1 j n0 , since cj cn0 +1 = 0. But, since deg (q ) < n0 , the q can not be a counterexample to part (1), implying that q (z ) = 0 for at most r (z ) = 0
for an innite number of distinct points. It

n0 1

distinct points. We have arrived at a contradiction, and part (1) is proved.

Next, let

be a polynomial for which

follows from part (1) that

cannot be a nonzero polynomial, for in that case it would have a degree

n0

and could be 0 for at most

distinct points. Hence,

is the zero polynomial, and part (2)

is proved. Now, to see part (3), set r = p q. Then r is a polynomial for which r (z ) = 0 for innitely many z 's. By part (2), it follows then that r (z ) = 0 for all z, whence p (z ) = q (z ) for all z. Moreover, p q is the zero polynomial, all of whose coecients are 0, and this implies that the coecients for p and q are identical. To prove the rst inequality in part (4), suppose that |z | > 1, and from the backwards triangle inequality, note that

|p (z ) | = = =
Set by

n k k=0 ck z | n n ck |z | | k=0 zn k | n n1 ck |z | | k=0 z nk + cn | n n1 ck |z | |cn | | k=0 zn k | n1 n k| |z | |cn | k=0 |z||c nk n1 k | n |z | |cn | k=0 ||cz | n1 n 1 |z | |cn | |z| k=0 |ck |

(3.21)

B equal to the constant (2/|cn |) 1/B, we obtain

n1 j =0

|cj |. Then, replacing the 1/|z | in the preceding calculation

n

|p (z ) | m|z |
for every

(3.22)

for which

|z | B.

This proves the rst half of part (4).

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

59

To get the other half of part (4), suppose again that

|z | > 1.
n

We have

|p (z ) |
k=0

|ck ||z |
k=0

|ck ||z | ,

(3.23)

n k=0 |ck |. Finally, to see part (5), suppose that there does exist a polynomial p of degree n such that 2 x = p (x) for all x 0. Then x = (p (x)) for all x 0. Now p2 is a polynomial of degree 2n. By 2 part (2), the two polynomials q (x) = x and (p (x)) must be the same, implying that they have the 2 same degree. However, the degree of q is 1, which is odd, and the degree of p is 2n, which is even.
so that we get the other half of part (4) by setting

M=

Hence, we have arrived at a contradiction.

Exercise 3.3.1
a. Let
'

b.

c.

r (z ) = p (z ) /q (z ) and r' (z ) = p' (z ) /q ' (z ) be two rational functions. Suppose r (z ) = r (z ) for innitely many z 's. Prove that r (z ) = r' (z ) for all z in the intersection of their ' ' domains. Is it true that p = p and q = q ? Let p and q be polynomials of degree n and m respectively, and dene a rational function r nm by r = p/q. Prove that there exist positive constants C and B such that |r (z ) | < C |z | for all complex numbers z for which |z | > B. Dene f : [0, ) R by f (x) = x. Show that there is no rational function r such that f (x) = r (x) for all x 0. That is, the square root function does not agree with a rational
function.

d. Dene the real-valued function mial

on

by

r ( x) = 1 / 1 + x2 .

Prove that there is no polyno-

e. If

f.

x. f (x) = |x|, show that f is not a rational function. HINT: Suppose |x| = p (x) /q (x) . Then |x|q (x) = p (x) implying that |x|q (x) is a polynomial s (x) . Now use Theorem 3.1 to conclude that p (x) = xq (x) for all x and that p (x) = xq (x) for all x. Let f be any complex-valued function of a complex variable, and let c1 , ..., cn be n distinct complex numbers that belong to the domain of f. Show that there does exist a polynomial p of degree n such that p (cj ) = f (cj ) for all 1 j n. HINT: Describe p in factored form.
such that for innitely many real numbers

p ( x) = r ( x)

is the real-valued function of a real variable given by

g. Give examples to show that the maximum and minimum of two polynomials need not be a polynomial or even a rational function. Very important is the denition of the

Denition 3.6:
Let codomain

compositiong f

of two functions

and

g.

Exercise 3.3.2
a. Suppose b. Suppose

f : S T and g : T U be functions. We dene a function g f, with domain S and U, by (g f ) (x) = g (f (x)) . If f : S T,g : T S, and g f (x) = x for all x S, then g is called a left inverse of f. If f g (y ) = y for all y T, then g is called a right inverse for f. If g is both a left inverse and a right 1 inverse, then g is called an inverse for f,f is called invertible, and we denote g by f . f :ST f :ST
has a left inverse. Prove that

is 1-1.

has a right inverse. Prove that

is onto.

c. Show that the composition of two polynomials is a polynomial and that the composition of two rational functions is a rational function. HINT: If that

is a polynomial, show by induction

pn

is a polynomial. Now use Exercise 3.2.1.

d. Find formulas for

g f

and

f g for the following.

What are the domains of these compositions?

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

60

CHAPTER 3. FUNCTIONS AND CONTINUITY

i. ii. iii.

f (x) = 1 + x2 and g (x) = 1/(1 + x) . f (x) = x/ (x + 1) and g (x) = x/ (1 x) . f (x) = ax + b and g (x) = cx + d.

1/2

3.4 Continuity
Denition 3.7:
pointc
Let

Next, we come to the denition of continuity. Unlike the preceding discussion, which can be viewed as being related primarily to the algebraic properties of functions, this one is an analytic notion.

T be sets of complex numbers, and let f : S T. Then f is said to be continuous at a S if for every positive , there exists a positive such that if x S satises |x c| < , then |f (x) f (c) | < . The function f is called continuous on S if it is continuous at every point c of S. If the domain S of f consists of real numbers, then the function f is called right continuous at c if for every > 0 there exists a > 0 such that |f (x) f (c) | < whenever x S and 0 x c < , and is called left continuous at c if for every > 0 there exists a > 0 such that |f (x) f (c) | < whenever x S and 0 x c > . S
and of

3.1: REMARK If f

is continuous at a point

c,

then the positive number

of the preceding denition

is not unique (any smaller number would work as well), but it does depend both on the number and on the point

c.

Sometimes we will write

(, c)

to make this dependence explicit. Later, we

will introduce a notion of uniform continuity in which the particular point

only depends on the number

and not on

c.

The next theorem indicates the interaction between the algebraic properties of functions and continuity.

Theorem 3.2:
Let

and

be subsets of

C,

let

are both continuous at a point 1. There exists a 2. 3. 4. 5. 6.

of

f and g be S. Then

functions from

into

T,

and suppose that

and

> 0 and a positive number M such that if |y c| < and y S then |f (y ) | M. That is, if f is continuous at c, then it is bounded near c. f + g is continuous at c. f g is continuous at c. |f | is continuous at c. If g (c) = 0, then f /g is continuous at c. If f is a complex-valued function, and u and v are the real and imaginary parts of f, then f is continuous at c if and only if u and v are continuous at c.

Proof:
We prove parts (1) and (5), and leave the remaining parts to the exercise that follows. To see part (1), let = 1. Then, since f is continuous at c, there exists a > 0 such that if |y c| < and y S then |f (y ) f (c) | < 1. Since |z w| ||z | |w|| for any two complex numbers z and w (backwards Triangle Inequality), it then follows that ||f (y ) | |f (c) || < 1, from which it follows that if |y c| < then |f (y ) | < |f (c) | + 1. Hence, setting M = |f (c) | + 1, we have that if |y c| < and y S, then |f (y ) | M as desired. To prove part (5), we rst make use of part 1. Let 1 , M1 and 2 , M2 be chosen so that if |y c| < 1 and y S then

|f (y ) | < M1
4 This
content is available online at <http://cnx.org/content/m36150/1.2/>.

(3.24)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

61

and if

|y c| < 2

and

yS

then

|g (y ) | < M2
Next, let

(3.25)

'

yS
that

then

|g (c) |/2. Then, there exists a ' > 0 such that if |y c| < ' and |g (y ) g (c) | < = |g (c) |/2. It then follows from the backwards triangle inequality
be the positive number
'

|g (y ) | > ' = |g (c) |/2 so

Now, to nish the proof of part (5), let

that

|1/g (y ) | < 2/|g (c) | |y c| < min 1 , 2 ,

'

(3.26) and

>0

be given. If

y S,

then

from Inequalities (3.1), (3.2), and (3.3) we obtain

(y ) |f g (y )

f (c) g (c) |

= = <

|f (y )g (c)f (c)g (y )| |g (y )g (c)| |f (y )g (c)f (c)g (c)+f (c)g (c)f (c)g (y )| |g (y )||g (c)| |f (y )f (c)||g (c)|+|f (c)||g (c)g (y )| |g (y )||g (c)|

(3.27)

(|f (y ) f (c) |M2 + M1 |g (c) g (y ) |) f

and

2 . |g (c)|2

Finally, using the continuity of both and

applied to the positive numbers

1 = / 4M2 |g (c) | |y c| < |y c| <

2 = / 4M1 |g (c) |

choose

> 0,

with

< min 1 , 2 , ' ,

and such that if

and and

y S then |f (y ) f (c) | < y S we have that

and 4M2 /|g (c)|2

|g (c) g (y ) | <

. Then, if 4M1 /|g (c)|2

|
as desired.

f (y ) f (c) |< g (y ) g (c)

(3.28)

Exercise 3.4.1
a. Prove part (2) of the preceding theorem. (It's an c. Prove part (4) of the preceding theorem. d. Prove part (6) of the preceding theorem. e. Suppose f. If

/2

argument.)

b. Prove part (3) of the preceding theorem. (It's similar to the proof of part (5) only easier.)

is a subset of

R.

Verify the above theorem replacing  continuity with left conti-

nuity and right continuity.

is a subset of

R,

show that

is continuous at a point

cS

if and only if it is both right

continuous and left continuous at

c.

Theorem 3.3:
Let

The composition of continuous functions is continuous. be subsets of

S, T,

and

C,

and let

f :S T

and

continuous at a point

gf

Proof:

is continuous at

cS c.

and that

is continuous at the point

g : T U be functions. Suppose f is f (c) T. Then the composition

Let > 0 be given. Because g is continuous at the point f (c) , there exists an > 0 such that |g (t) g (f (c)) | < if |t f (c) | < . Now, using this positive number , and using the fact that f is continuous at the point c, there exists a > 0 so that |f (s) f (c) | < if |s c| < . Therefore, if |s c| < , then |f (s) f (c) | < , and hence |g (f (s)) g (f (c)) | = |g f (s) g f (c) | < , which completes the proof.

Exercise 3.4.2
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

62

CHAPTER 3. FUNCTIONS AND CONTINUITY

a. If

f :CC C.
domain.

is the function dened by

f (z ) = z,

prove that

is continuous at each point of

b. Use part (a) and Theorem 3.2 to conclude that every rational function is continuous on its c. Prove that a step function the points of the partition

h : [a, b] C is continuous P that determines h.

everywhere on

[a, b]

except possibly at

Exercise 3.4.3
a. Let

S be that f is

the set of nonnegative real numbers, and dene continuous at each point of

S.

HINT: For

f : S S by f (x) = x. Prove c = 0, use = 2 . For c = 0, use the

(3.29)

identity

c=

y+ c yc yc = . c y+ c y+ c c f (z ) = |z |,
show that

b. If

f :CR

is the function dened by

is continuous at every point

of its domain.

Exercise 3.4.4
Using the previous theorems and exercises, explain why the following functions on their domains. Describe the domains as well. a. b. c.

are continuous

f (z ) = 1 z 2 / 1 + z 2 . f (z ) = |1 + z + z 2 + z 3 (1/z ) |. f (z ) = 1+ 1 |z | .
2

Exercise 3.4.5
a. If b. If c.

c and d are real numbers, show that max (c, d) = (c + d) /2 + |c d|/2. f and g are functions from S into R, show that max (f, g ) = (f + g ) /2 + |f g |/2. If f and g are real-valued functions that are both continuous at a point c, show that max (f, g ) and min (f, g ) are both continuous at c.

Exercise 3.4.7: Negations

b. Negate the c. Negate the

N be the set of natural numbers, let P be the set of positive real numbers, and dene f : N P f (n) = 1 + n. Prove that f is continuous at each point of N. Show in fact that every function f : N C is continuous on this domain N. HINT: Show that for any > 0, the choice of = 1 will work.
Let by

Exercise 3.4.6

a. Negate the statement: For every

> 0,|x| < .'' statement: For every > 0, there exists '' statement that  f is continuous at c.

an

for which

|x| < .''

The next result establishes an equivalence between the basic the

denition of continuity and a sequential

formulation. In many cases, maybe most, this sequential version of continuity is easier to work with than

Theorem 3.4:
Let at

version.

f :SC c,

be a complex-valued function on the sequence

S,

and let

be a point in

if and only if the following condition holds: For every sequence

converges to

{f (xn )}

converges to

f (c) .

Or, said a

S. Then f is continuous {xn } of elements of S that dierent way, if {xn } converges

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

63

to

c,

then

{f (xn )}

converges to

f (c) .

And, said yet a third (somewhat less precise) way, the

function

Proof:

converts convergent sequences to convergent sequences.

Suppose rst that f is continuous at c, and let {xn } be a sequence of elements of S that converges to c. Let > 0 be given. We must nd a natural number N such that if n N then |f (xn ) f (c) | < . First, choose > 0 so that |f (y ) f (c) | < whenever y S and |y c| < . Now, choose N so that |xn c| < whenever n N. Then if n N, we have that |xn c| < , whence |f (xn ) f (c) | < . This shows that the sequence {f (xn )} converges to f (c) , as desired. We prove the converse by proving the contrapositive statement; i.e., we will show that if f is not continuous at c, then there does exist a sequence {xn } that converges to c but for which the sequence {f (xn )} does not converge to f (c) . Thus, suppose f is not continuous at c. Then there exists an 0 > 0 such that for every > 0 there is a y S such that |y c| < but |f (y ) f (c) | 0 . To obtain a sequence, we apply this statement to 's of the form = 1/n. Hence, for every natural number n there exists a point xn S such that |xn c| < 1/n but |f (xn ) f (c) | 0 . Clearly, the sequence {xn } converges to c since |xn c| < 1/n. On the other hand, the sequence {f (xn )} cannot be converging to f (c) , because |f (xn ) f (c) | is always 0 . This completes the proof of the theorem.

3.5 Continuity and Topology

Let

T. Recall that f 1 (A) denotes the subset of the domain S consisting of all those x S for which f (x) A. Our original denition of continuity was in terms of 's and 's. Theorem 3.4, p. 62 established an f :S T
be a function, and let

be a subset of the codomain

equivalent form of continuity, often called sequential continuity, that involves convergence of sequences. The next result shows a connection between continuity and topology, i.e., open and closed sets.

Theorem 3.5:
1. Suppose Then of

is a closed subset of

is continuous on

C. That is, U

f : S C is a complex-valued function on S. f 1 (A) is a closed set whenever A is a closed subset f is continuous on a closed set S if and only if the inverse image of every closed C
and that

if and only if

set is closed. 2. Suppose Then of is an open subset of

and that

is continuous on

if and only if

C.

That is,

is continuous on an

f : U C is a complex-valued function on U. f 1 (A) is an open set whenever A is an open subset open set U if and only if the inverse image of every open

set is open.

Proof:
C. We wish to f 1 (A) that converges to a 1 is a closed set, we know that c S, but in order to see that f (A) is closed, we 1 need to show that c f (A) . That is, we need to show that f (c) A. Now, f (xn ) A for every n, and, because f is continuous at c, we have by Theorem 3.4, p. 62 that f (c) = limf (xn ) . Hence, f (c) is a limit point of A, and so f (c) A because A is a closed set. Therefore, c f 1 (A) , and f 1 (A) is closed. Conversely, still supposing that S is a closed set, suppose f is not continuous on S, and let c be a point of S at which f fails to be continuous. Then, there exists an > 0 and a sequence {xn } of elements of S such that c = limxn but such that |f (c) f (xn ) | for all n. (Why? See the proof of Theorem 3.4.) Let A be the complement of the open disk B (f (c)) . Then A is a closed 1 subset of C. We have that f (xn ) A for all n, but f (c) is not in A. So, xn f (A) for all n, but
Suppose rst that is continuous on a closed set and that is a closed subset of

f 1 (A) point c. Because S

show that

is closed. Thus, let

{xn }

be a sequence of points in

5 This

content is available online at <http://cnx.org/content/m36152/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

64

CHAPTER 3. FUNCTIONS AND CONTINUITY

c = limxn
is not in is not closed. Hence, if Next, suppose and let

f 1 (A) . Hence, f 1 (A) does not contain all of its limit points, and so f 1 (A) f is not continuous on S, then there exists a closed set A such that f 1 (A)
is an open set, and assume that

is not closed. This completes the proof of the second half of part (1).

f is continuous on U. Let A be an open set f 1 (A) . In order to prove thatf 1 (A) is open, we need to show 1 that c belongs to the interior of f (A) . Now, f (c) A,A is open, and so there exists an > 0 such that the entire disk B (f (c)) A. Then, because f is continuous at the point c, there exists a > 0 such that if |x c| < then |f (x) f (c) | < . In other words, if x B (c) , then f (x) B (f (c)) A. This means that B (c) is contained in f 1 (A) , and hence c belongs to the 1 interior of f (A) . Hence, if f is continuous on an open set U, then f 1 (A) is open whenever A
in

C,

be an element of

is open. This proves half of part (2).

f 1 (A) is open whenever A is open, let c be a point of S, and let us prove that f is continuous at c. Thus, let > 0 be given, and let A be the 1 open set A = B (f (c)) . Then, by our assumption, f (A) is an open set. Also, c belongs to this 1 open set f (A) , and hence c belongs to the interior of f 1 (A) . Therefore, there exists a > 0 1 such that the entire disk b (c) f (A) . But this means that if S satises |x c| < , then 1 x B (c) f (A) , and so f (x) A = B (f (c)) . Therefore, if |xc| < , then |f (x)f (c) | < , which proves that f is continuous at c, and the theorem is completely proved.
Finally, still assuming that

is open, suppose

3.6 Deeper Analytic Properties of Continuous Functions

We collect here some theorems that show some of the consequences of continuity. a real variable.

Some of the theorems

apply to functions either of a real variable or of a complex variable, while others apply only to functions of a real variable. We begin with what may be the most famous such result, and this one is about functions of

f : [a, b] R is a real-valued function that is continuous at each point of the closed interval [a, b] , and if v is a number (value) between the numbers f (a) and f (b) , then there exists a point c between a and b such that f (c) = v.
If

Theorem 3.6:

Intermediate Value Theorem

Proof:
If

Let

v = f (a) or f (b) , we are done. Suppose then, without loss of generality, that f (a) < v < f (b) . S be the set of all x [a, b] such that f (x) v, and note that S is nonempty and bounded above. (a S, and b is an upper bound for S.) Let c = supS. Then there exists a sequence {xn } of elements of S that converges to c. (See Exercise 2.7.1.) So, f (c) = limf (xn ) by Theorem 3.4, p. 62. Hence, f (c) v. (Why?) Now, arguing by contradiction, if f (c) < v, let be the positive number v f (c) . Because f is continuous at c, there must exist a > 0 such that |f (y ) f (c) | < whenever |y c| < and y [a, b] . Since any smaller satises the same condition, we may also assume that < b c. Consider y = c + /2. Then y [a, b] , |y c| < , and so |f (y ) f (c) | < . Hence f (y ) < f (c)+ = v, which implies that y S. But, since c = supS,c must satisfy c y = c + /2. This is a contradiction, so f (c) = v, and the theorem is proved.

The Intermediate Value Theorem tells us something qualitative about the range of a continuous function on an interval

[a, b] .

It tells us that the range is connected; i.e., if the range contains two points

and

d,

then the range contains all the points between

and

d.

It is dicult to think what the analogous assertion

would be for functions of a complex variable, since between doesn't mean anything for complex numbers. We will eventually prove something called the Open Mapping Theorem in Section 7.6 that could be regarded as the complex analog of the Intermediate Value Theorem.

6 This

content is available online at <http://cnx.org/content/m36167/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

65

The next theorem is about functions of either a real or a complex variable.

Theorem 3.7:
Let

f :SC

be a continuous function, and let

be a compact (closed and bounded) subset of

S.

Then the image

f (C )

of

is also compact. That is, the continuous image of a compact set is

compact.

Proof:

for each

f (C ) is not bounded. Thus, let {xn } be a sequence of elements of C such that, n,|f (xn ) | > n. By the Bolzano-Weierstrass Theorem, the sequence {xn } has a convergent subsequence {xnk }. Let x = limxnk . Then x C because C is a closed subset of C. Co, f (x) = limf (xnk ) by Exercise 2.7.1. But since |f (xnk ) | > nk , the sequence {f (xnk )} is not bounded, so cannot be convergent. Hence, we have arrived at a contradiction, and the set f (C ) must be
First, suppose bounded. of

f (C ) is closed. Thus, let y be a limit point of the image f (C ) yn f (C ) . For each n, let xn C satisfy f (xn ) = yn . Again, using the Bolzano-Weierstrass Theorem, let {xnk } be a convergent subsequence of the bounded sequence {xn }, and write x = limxnk . Then x C, since C is closed, and from Exercise 2.7.1
Now, we must show that the image and let

C,

y = limyn

where each

y = limf (xn ) = limf (xnk ) = f (x) ,

showing that

(3.30)

y f (C ) ,

implying that

f (C )

is closed.

This theorem tells us something about the range of a continuous function of a real or complex variable. It says that if a subset of the domain is closed and bounded, so is the image of that subset. The next theorem is about continuous real-valued functions of a complex variable, and it is one of the theorems to remember.

Theorem 3.8:
Let

be a continuous real-valued function on a compact subset

of

a maximum and a minimum value on

S.

That is, there exist points

C. Then f attains both z1 and z2 in S such that

Proof:

f (z1 ) f (z ) f (z2 )
We prove that

for all

z S. f
attains a minimum value to the

attains a maximum value, leaving the fact that

know that this supremum exists?) We will show that there exists an

f (x) for x S. (How do we z2 S such that f (z2 ) = M0 . This will nish the proof, since we would then have f (z2 ) = M0 f (z ) for all z S. Thus, let {yn } be a sequence of elements in the range of f for which the sequence {yn } converges to M0 . (This is Exercise 2.20 again.) For each n, let xn be an element of S such that yn = f (xn ) . Then the sequence {f (xn )} converges to M0 . Let {xnk } be a convergent subsequence of {xn }. (How?) Let z2 = limxnk . Then z2 S, because S is closed, and f (z2 ) = limf (xnk ) , because f is continuous. Hence, f (z2 ) = M0 , as desired.
exercise that follows. Let

M0

be the supremum of the set of all numbers

Exercise 3.6.1
a. Prove that the

of the preceding theorem attains a minimum value on

S.

b. Give an alternate proof of Theorem 3.8, p. 65 by using Theorem 3.7, p. 65, and then proving that a closed and bounded subset of c. Let

R contains both its supremum and its inmum. C, and let c be a point of C that is not in S. Prove that there is a closest point to c in S. That is, show that there exists a point w S such that |w c| |z c| for all points z S. HINT: The function z |z c| is continuous on the set S. S
be a compact subset of

Exercise 3.6.2
Let

f : [a, b] R

be a real-valued function that is continuous at each point of

[a, b] .

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

66

CHAPTER 3. FUNCTIONS AND CONTINUITY

a. Prove that the range of b.

f is a closed interval a' , b' . Show by example that the four numbers f (a) , f (b) ,a' and b' can be distinct. Suppose f is 1-1. Show that, if c is in the open interval (a, b) , then f (c) is in the open interval a' , b' .

We introduce next a dierent kind of continuity called uniform continuity. The dierence between regular continuity and uniform continuity is a bit subtle, and well worth some thought.

Denition 3.8:
A function

f : S C

is called

uniformly continuous
|f (x) f (y ) | <

on

if for each positive number

exists a positive number

such that

for all

x, y S

satisfying

, there |x y | < .
works for

Basically, the dierence between regular continuity and uniform conintuity is that the same all points in

S.

Here is another theorem worth remembering.

Theorem 3.9:

Proof:

A continuous complex-valued function on a compact subset We argue by contradiction.

of

is uniformly continuous.

Thus, suppose f is continuous on S but not uniformly continuous. > 0 for which no positive number satises the uniform continuity denition. Therefore, thinking of the 's as ranging through the numbers 1/n, we know that for each positive integer n, there exist two points xn and yn in S so that Then, there exists an 1. 2.

|yn xn | < 1/n, and |f (yn ) f (xn ) | . 1/n

would suce for a

Otherwise, some limit

{xn } with x is also the limit of the corresponding subsequence {ynk } of {yn }. But then f (x) = limf (xnk ) = limf (ynk ) , implying that 0 = lim|f (ynk ) f (xnk ) |, which implies that |f (ynk ) f (xnk ) | < for all large enough k. But .
Let be a convergent subsequence of

{ x nk }

x.

By (1) and the triangle inequality, we deduce that

that contradicts (2), and this completes the proof.

Continuous functions whose domains are not compact sets may or may not be uniformly continuous, as the next exercise shows.

Exercise 3.6.3
a. Let

b.

f : (0, 1) R be dened by f (x) = 1/x. Prove that f is continuous at each x in its f is not uniformly continuous there. HINT: Set = 1, and consider the pairs of points xn = 1/n and yn = 1/ (n + 1) . x. Prove that f is not bounded, but is Let f : [1, ) [1, ) be dened by f (x) = nevertheless uniformly continuous on its domain. HINT: Take = .
domain but that

Theorem 3.10:
Let

f :ST

be a continuous 1-1 function from a compact (closed and bounded) subset of

onto

the (compact) set

T.

Let

g :T S

denote the inverse function

f 1

of

f.

Then

is continuous.

The inverse of a continuous function, that has a compact domain, is also continuous.

Proof:

We prove that whenever

is continuous by using Theorem 3.5, p. 63; i.e., we will show that

is a closed subset of

C.

But this is easy, since

(A) = g

g 1 (A) is closed (A S ) = f (A S ) , and

this is a closed set by Theorem 3.7, p. 65, because

AS

is compact. See part (e) of Exercise 2.7.5.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

67

3.2: REMARK Using the preceding theorem, and the exercise below, we will show that taking nth
roots is a continuous function. that is, the function

Exercise 3.6.4

dened by

f (x) = x1/n

is continuous.

Use the preceding theorem to show the continuity of the following functions. a. Show that if all of

b. c.

n is an odd positive integer, then there exists a continuous function g dened on n R such that g (x) is an nth root of x for all real numbers x. That is, (g (x)) = x for all n real x. (The function f (x) = x is 1-1 and continuous.) Show that if n is any positive integer then there exists a unique continuous function g dened on [0, ) such that g (x) is an nth root of x for all nonnegative x. Let r = p/q be a rational number. Prove that there exists a continuous function g : [0, ) q [0, ) such that g (x) = xp for all x 0; i.e., g (x) = xr for all x 0.

Theorem 3.11:
Let

be a continuous 1-1 function from the interval

[a, b]

onto the interval

[c, d] .

Then

must be

strictly monotonic, i.e., strictly increasing everywhere or strictly decreasing everywhere.

Proof:
Since that

f is 1-1, we clearly have that f (a) = f (b) , and, without loss of generality, let us assume c = f (a) < f (b) = d. It will suce to show that if and belong to the open interval (a, b) , and < , then f () f ( ) . (Why will this suce?) Suppose by way of contradiction that there exists < in (a, b) for which f () > f ( ) . We use the intermediate value theorem to derive a contradiction. Consider the four points a < < < b. Either f (a) < f () or f ( ) < f (b) . (Why?) In the rst case (f (a) < f ()), f ([a, ]) contains every value between f (a) and f () . And, f ([, ]) contains every value between f () and f ( ) . So, let v be a number such that f (a) < v,f ( ) < v, and v < f () (why does such a number v exist?). By the Intermediate Value Theorem, there exists x1 (a, ) such that v = f (x1 ) , and there exists an x2 (, ) such that v = f (x2 ) . But this contradicts the hypothesis that f is 1-1, since x1 = x2 . A similar argument leads to a contradiction in the second case f ( ) < f (b) . (See the following exercise.) Hence, there can exist no such and , implying that f is strictly increasing on [a, b] .

Exercise 3.6.5
Derive a contradiction from the assumption that

f ( ) < f (b)

in the preceding proof.

3.7 Power Series Functions

functions, and the

The class of functions that we know are continuous includes, among others, the polynomials, the rational

nth root functions.

We can combine these functions in various ways, e.g., sums, products,

quotients, and so on. We also can combine continuous functions using composition, so that we know that

nth

roots of rational functions are also continuous. The set of all functions obtained in this manner is called

the class of algebraic functions. Now that we also have developed a notion of limit, or innite sum, we can construct other continuous functions. We introduce next a new kind of function. It is a natural generalization of a polynomial function. Among these will be the exponential function and the trigonometric functions. We begin by discussing functions of a complex varible, although totally analogous denitions and theorems hold for functions of a real variable.

Denition 3.9:
Let

n=0

{an } 0 be a sequence of real or complex numbers. By the power series functionf (z ) = an z n we mean the function f : S C where the domain S is the set of all z C for which

7 This

content is available online at <http://cnx.org/content/m36165/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

68

CHAPTER 3. FUNCTIONS AND CONTINUITY

the innite series of the series. The numbers We write

an z n {an }

converges, and where

is the rule that assigns to such a

zS

the sum

dening a power series function are called the

We associate to a power series function

f (z ) =
N

n=0

an z n

coecients

of the function. of partial sums.

its sequence

{SN }

SN (z ) =
n=0

an z n .

(3.31)

Notice that polynomial functions are very special cases of power series functions. They are the power series functions for which the coecients each partial sum

{an } are all 0 beyond some point. Note also that SN for any power series function is itself a polynomial function of degree less than or equal to N. Moreover, if f is a power series function, then for each z in its domain we have f (z ) = limN SN (z ) . Evidently, every power series function is a limit of a sequence of polynomials. Obviously, the domain S Sf of a power series function f depends on the coecients {an }
determining the function. Our rst goal is to describe this domain.

Theorem 3.12:
Let 1. 3.

be a power series function: 0 belongs to

f (z ) =

n=0

an z n

with domain

S.

Then:

S.
belongs to

2. If a number

4.

S, then every number u, for which |u| < |t|, also belongs to S. r around 0 in C (possibly open, possibly closed, possibly neither, possibly innite). That is, S consists of the disk Br (0) = {z : |z | < r } possibly together with some of the points z for which |z | = r. The radius r of the disk in part (3) is given by the Cauchy-Hadamard formula: t S
is a disk of radius

r=
which we interpret to imply that

1 , lim sup|an |1/n

(3.32)

r=

if and only if that limsup is

r=0 0.

if and only if the limsup on the right is innite, and

Proof:
Part (1) is clear. To see part 2, assume that series

belongs to

and that

an un

converges.

In fact, we will show that

|u| < |t|. We |an un | is

wish to show that the innites convergent, i.e., that

an un

is absolutely convergent. that the terms

We are given that the innite series an t converges, which implies an tn tend to 0. Hence, let B be a number such that |an z n | B for all n, and set = |u|/|t|. Then < 1, and therefore the innite series Bn is convergent. Finally, n n n n |an u | = |an t | B , which, by the Comparison Test, implies that |an un | is convergent, as desired. Part (3) follows, with just a little thought, from part 2.

lim sup|an |1/n either is nite or it is innite. assume rst that the sequence {|an | } is not bounded; i.e., that lim sup|an |1/n = . Then, given any number p, there 1/n that are larger than p. So, for any z = 0, there exist innitely are innitely many terms |an | 1/n n many terms |an | that are larger than 1/|z |. But then |an z | > 1 for all such terms. Therefore the n n innite series an z is not convergent, since liman z is not zero. So no such z is in the domain S. This shows that if lim sup|an |1/n = , then r = 0 = 1/lim sup|an |1/n . 1/n Now, suppose the sequence {|an | } is bounded, and let L denote its limsup. We must show that 1/r = L. We will show the following two claims: (a) if 1/|z | > L, then z S, and (b) if 1/|z | < L, then z / S. (Why will these two claims complete the proof ?) Thus, suppose that 1/|z | > L. Let be a number satisfying L < < 1/|z |, and let = |z |. Then 0 < < 1. Now there exists a 1/n natural number N so that |an | < for all n N, or equivalently |an | n for all n N. (See
To prove part (4), note that

1/n

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

69

part (a) of Exercise 2.17. ) This means that for all convergent. Hence,

nN

we have

|an z n | = |an / n ||z | n .

an z n is absolutely convergent, whence z S, and this proves claim (a) above. Incidentally, note also that if L = 0, this argument shows that r = , as desired. To verify claim (b), suppose that 1/|z | < L. Then there are innitely many terms of the sequence {|an |1/n } that are greater than 1/|z |. (Why?) For each such term, we would then have |an z n | 1. This means that the innite series an z n is not convergent and z / S, which shows claim b. 1/n Hence, in all cases, we have that r = 1/lim sup|an | , as desired.
This implies by the Comparison Test that the power series

Denition 3.10:
convergence of the power series. disk of convergence.
If

is a power series function, the number The disk

of the preceding theorem is called the

radius of

of radius

around 0, denoted by

Br (0) ,

is called the

Exercise 3.7.1

Compute directly the radii of convergence for the following power series functions, i.e., without using the Cauchy-Hadamard formula. formula agrees with your computation. a. b. c. d. e. Then, when possible, verify that the Cauchy-Hadamard

f (z ) = f (z ) = f (z ) = f (z ) = f (z ) =

zn. n2 z n . n (1) (1/ (n + 1)) z n . (1/ (n + 1)) z 3n+1 . n n=0 z /n!.

Exercise 3.7.2
a. Use part (e) of Exercise 3.2.1 to show that a power series function and only if its only nonzero coecients are even ones, i.e., the the

is an even function if

a2k 's.

Show also that a power

series function is an odd function if and only if its only nonzero coecients are odd ones, i.e., b. Suppose

a2k+1 's. f (z ) =

k=0

a2k z 2k

is a power series function that is an even function. Show that

f (iz ) =
k=0
where c. If

(1) a2k z 2k = f a (z ) , f

(3.33)

fa

is the power series function obtained from

by alternating the signs of its coe-

cients. We call this function

g (z ) =

k=0

fa

the alternating version of

f.

a2k+1 z 2k+1

is a power series function that is an odd function, show that

g (iz ) = i
k=0
where again coecients. d. If

(1) a2k+1 z 2k+1 = ig a (z ) , g

(3.34)

ga

is the power series function obtained from

by alternating the signs of its

is any power series function, show that

a a f (iz ) = fe (iz ) + fo (iz ) = fe (z ) + ifo (z ) ,

and hence that

(3.35)

a fe (iz ) = fe (z )

and

fo (iz ) = ipa o (z ) .

The next theorem will not come as a shock, but its proof is not so simple.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

70

CHAPTER 3. FUNCTIONS AND CONTINUITY

Theorem 3.13:
Let

f (z ) =

an z n

be a power series function with radius of convergence

at each point in the open disk

Proof:
Let

Br (0) ,

i.e., at each point

for which

r. |z | < r.

Then

is continuous

z Br (0) be given. We must make some auxiliary constructions before we can show that f is z. First, choose a z ' such that |z | < |z ' | < r. Next, set bn = |nan |, and dene g (z ) = bn z n . By the Cauchy-Hadamard formula, we see that the power series function g has the same 1/n radius of convergence as the power series function f. Indeed, lim sup|bn | = lim supn1/n |an | = 1/n ' limn lim sup|an |. Therefore, z belongs to the domain of g. Let M be a number such that each n N ' partial sum of the series g z = n=0 bn z ' is bounded by M. ' Now, let > 0 be given, and choose to be the minimum of the two positive numbers |z |/M ' ' and |z | |z |. We consider any y for which |y z | < . Then y Br (0) ,|y | < |z |, and
continuous at

|f (y ) f (z ) |

= = = < lim
N

lim|SN (y ) SN (z ) | lim| lim

N N n=1 N n=0 N n=0

an (y n z n ) | |an ||y n z n |
n1 j =0

|an ||y z | |an ||y z |

|y j ||z n1j | |z ' |

n1
(3.36)

lim
N N

N n=1

n1 j =0 N n=0

lim|y z | 1/|z ' |

N

n|an ||z ' |n

|y z |lim |M z' | lim |M z' |

N

This completes the proof.

Exercise 3.7.3
a. Let for

m n k n=0 an z be a power series function, and let p (z ) = k=0 bk z be a polynomial function. Prove that f + p and f p are both power series functions. Express the coecients

f (z ) =

b. Suppose

f p in terms of the an 's and bk 's. g are power series functions. Prove that f + g is a power series function. What is its radius of convergence? What about cf ? What about f g ? What about f /g ? What about |f |?
and

f +p

and

Exercise 3.7.4
a. Prove that every polynomial is a power series function with innite radius of convergence. b. Prove that right.) c. Dene

1/z

and

(1/ (z 1) (z + 2))

are not power series functions. (Their domains aren't

n 2n+1 . Prove that the radius of convergence of this power series n=0 (1) z z function is 1, and that f (z ) = 1+z 2 for all z B1 (0) . Conclude that the rational function 2 z/ 1 + z agrees with a power series function on the disk B1 (0) . But, they are the same

f (z ) =

not

function. HINT: Use the innite geometric series. Theorem 3.13, p. 69 and Exercise 3.7.3 and Exercise 3.7.4 raise a very interesting and subtle point. Suppose

f (z ) =
says that

an z n f

is a power series function having nite radius of convergence

r > 0.

Theorem 3.13, p. 69

is continuous on the open disk, but it does not say anything about the continuity of

at points

on the boundary of this disk that are in the domain of

f,

i,e., at points

z0

for which

|z0 | = r.

and

n an z0

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

71

converges. Suppose point

g (z )

is a continuous function whose domain contains the open disk

z0 ,

and assume that

f (z ) = g (z )

for all

z Br (0) .

Does

f (z0 )

have to agree with

Br (0) and g (z0 )? It's

also a worth

some thought to understand just what this question means. It amounts to a question of the equality of two dierent kinds of limits. because

is continuous at

f (z0 ) is the sum of an innite series, the limit of a sequence of partial sums, while, z0 ,g (z0 = limzz0 g (z ) . At the end of this chapter, we include a theorem of Abel

that answers this question. The next theorem is the analog for power series functions of part (2) of Theorem 3.1, p. 57 for polynomials. We call it the Identity Theorem, but it equally well could be known as the Uniqueness of Coecients Theorem, for it implies that dierent coecients mean dierent functions.

Theorem 3.14:
f (z ) =

Identity Theorem be a power series function with positive radius of convergence

Let

an z n

r.

Suppose

{zk }

is a sequence of nonzero distinct numbers in the domain of 1. 2. Then

such that:

limzk = 0. f (zk ) = 0 for f

all

k. (z ) 0
for all

Proof:
at

is identically 0 (f

z S ).

Moreover, each coecient

an

of

equals

0.

Arguing by induction on

0,

and since

Assume

n, let us prove that all the coecients an are 0. First, since f limzk = 0, we have that a0 , which equals f (0) , = limf (zk ) = 0. then that a0 = a1 = ... = an1 = 0. Then f (z ) = an z n + an+1 z n+1 + ... = zn
j j =0 bj z ,

is continuous

(3.37)

g (z ) = bj z j , then, by the Cauchyg is the same as that for f. (Why 1/k 1/j n does lim sup|bj | = lim sup|ak | ?) We have that f (z ) = z g (z ) for all z in the common disk n of convergence of these functions f and g. Since, for each k, zk = 0 and f (zk ) = zk g (zk ) = 0, it follows that g (zk ) = 0 for every k. Since g is continuous at 0, it then follows as in the argument above that g (0) = 0. But, g (0) = b0 = an . Hence an = 0, and so by induction all the coecients of the power series function f are 0. Clearly this implies that f (z ) is identically 0.
where

bj = an+j .

If

is the power series function dened by

Hadamard Formula, we have that the radius of convergence for

Rule 3.1:
Suppose

and

are two power series functions, that

converges to 0, and that

Exercise 3.7.5

radius of convergence,

{zk } is a sequence of nonzero points that f (zk ) = g (zk ) for all k. Then f and g have the same coecients, the same and hence f (z ) = g (z ) for all z in their common domain.

a. Prove the preceding corollary. (Compare with the proof of Theorem 3.1, p. 57.) b. Use the corollary, and the power series function power series function. c. Show that there are power series functions that are not polynomial functions. d. Let

g (z ) = z,

to prove that

f (z ) = |z |

is not a

f (z ) =

an z n

be a power series function with innite radius of convergence, all of

whose coecients are positive.

Show that there is no rational function r = p/q for which f (z ) = r (z ) for all complex numbers z. Conclude that the collection of power series functions n provides some new functions. HINT: Use the fact that for any n we have that f (x) > an x for all positive x. Then, by choosing n appropriately, derive a contradiction to the resulting n fact that |p (x) /q (x) | > an x for all positive x. See part (b) of Exercise 3.3.1.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

72

CHAPTER 3. FUNCTIONS AND CONTINUITY

8

3.8 The Elementary Transcendental Functions

Having introduced a class of new functions (power series functions), we might well expect that some of these will have interesting and unexpected properties. So, which sets of coecients might give us an exotic new function? Unfortunately, at this point in our development, we haven't much insight into this question. It is true, see Exercise 3.7.1, that most power series functions that we naturally write down have nite radii of convergence. Such functions may well be new and fascinating, but as a rst example, we would prefer to consider a power series function that is dened everywhere, i.e., one with an innite radius of convergence. Again revisiting Exercise 3.7.1, let us consider the coecients let's have a look.

an = 1/n!.

This may seem a bit ad hoc, but

Denition 3.11:
Dene a power series function, denoted by exp, as follows:

exp (z ) =

zn . n! n=0

(3.38)

We will call this function, with 20-20 hindsight, the

exponential function.
ez ; that will come in the next chapter.

What do we know about this function, apart from the fact that it is dened for all complex numbers? We certainly do not know that it has anything to do with the function We do know what the number

is, but we do not know how to raise that number to a complex exponent.

All of the exponential function's coecients are positive, and so by part (d) of Exercise 3.7.5 exp is not a rational function; it really is something new. It is natural to consider the even and odd parts versions

expe

and

expo

of this new function. And then, considering the constructions in Exercise 3.7.2, to introduce the alternating

Denition 3.12:

expa e

and

expa o

of them.

Dene two power series functions cosh (hyperbolic cosine) and sinh (hyperbolic sine) by

cosh (z ) =

exp (z ) + exp (z ) 2

and

sinh (z ) =

exp (z ) exp (z ) , 2

(3.39)

and two other power series functions cos (cosine) and sin (sine) by

cos (z ) = cosh (iz ) =

and

exp (iz ) + exp (iz ) 2 exp (iz ) exp (iz ) . 2i

(3.40)

sin (z ) = isinh (iz ) =

(3.41)

The ve functions just dened are called the elementary transcendental functions, the sinh and cosh functions are called the basic hyperbolic functions, and the sine and cosine functions are called the basic

trigonometric

or

circular functions.

The connections between the hyperbolic functions

and hyperbolic geometry, and the connection between the trigonometric functions and circles and triangles, will only emerge in the next chapter. From the very denitions, however, we can see a connection between the hyperbolic functions and the trigonometric functions. It's something like interchanging the roles of the real and imaginary axes. This is probably worth some more thought.

Exercise 3.8.1
8 This

content is available online at <http://cnx.org/content/m36160/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

73

a. Verify the following equations:

exp (z )

= = = 1+z+
z2 2!

zn n=0 n! z3 3! + ...

zk k!

+ ...,

(3.42)

cosh (z ) + sinh (z ) .
z3 3!

sin (z )

= z =

z5 5!

k z 2k+1 z7 7! + ... + (1) (2k+1)! k z 2k+1 k=0 (1) (2k+1)! , k z 2k z6 6! + ... + (1) (2k)! k z 2k k=0 (1) (2k)! ,

+ ...

(3.43)

cos (z )

= =

z2 2!

z4 4!

+ ...

(3.44)

sinh (z ) =
k=0
and

z 2k+1 , (2k + 1)! z 2k . (2k )!

(3.45)

cosh (z ) =
k=0

(3.46)

(These expressions for the elementary transcendental functions are perhaps the more familiar ones from a calculus course.) b. Compute the radii of convergence for the elementary transcendental functions. HINT: Do not use the Cauchy-Hadamard formula. Just gure out for which c. Verify that

exp (0) = 1,sin (0) = sinh (0) = 0, sin2 (z ) + cos2 (z ) 1?

and

z 's the functions cos (0) = cosh (0) = 1.

are dened.

d. Prove that all ve of the elementary transcendental functions are not rational functions. e. Can you explain why What about the  Addition Formula (3.47)

sin (z + w) = sin (z ) cos (w) + cos (z ) sin (w) .

Exercise 3.8.2
a. Show that the elementary transcendental functions map real numbers to real numbers. That is, as functions of a real variable, they are real-valued functions. b. Show that the exponential function exp is not bounded above. Show in fact that, for each nonnegative integer fact, does

mean

n,exp (x) /xn is unbounded. Can you show if x is an irrational or complex number?

that

exp (x) = ex ?

What, in

At this point, we probably need a little fanfare!

Theorem 3.15: THEOREM 3.14159 (Denition of )

x
for which

There exists a smallest positive number

sin (x) = 0.

We will denote this distinguished

Proof:

number

by the symbol

.
is positive. Indeed, the innite series for

First we observe that

sin (1)

follows from the alternating series test (Theorem 2.18) that Next, again using the alternating series test, we observe

sin (1) is alternating. sin (1) > 1 1/6 = 5/6. that sin (4) < 0. Indeed,

It

43 45 47 49 + + 0.4553 < 0. (3.48) 3! 5! 7! 9! Hence, by the intermediate value theorem, there must exist a number c between 1 and 4 such that sin (c) = 0. So, there is at least one positive number x such that sin (x) = 0. However, we must sin (4) < 4
show that there is a smallest positive number satisfying this equation.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

74

CHAPTER 3. FUNCTIONS AND CONTINUITY

for which sin (x) = 0. Then A is a nonempty set of real numbers = inf A. We need to prove that sin ( ) = 0, and that > 0. Clearly then it will be the smallest positive number x for which sin (x) = 0. By Exercise 2.7.1, there exists a sequence {xk } of elements of A such that = limxk . Since sin is continuous at , it follows that sin ( ) = limsin (xk ) = lim0 = 0. Finally, if were equal to 0, then by the Identity Theorem, Theorem 3.14, we would have that sinx = 0 for all x. Since this is clearly not the case, we must have that > 0. Hence, is the smallest (minimum) positive number x for which sin (x) = 0. Let be the set of all that is bounded below. Dene

x>0

As hinted at earlier, the connection between this number

and circles is not at all evident at the moment.

For instance, you probably will not be able to answer the questions in the next exercise.

Exercise 3.8.3

a. Can you see why

sin (x + 2 ) sin (x)? That is, cos ( ) = 1?

is it obvious that

sin is a periodic function?

b. Can you prove that

3.3: REMARK Dening to be the smallest positive zero of the sine function may strike many people
as very much out of the blue. However, the zeroes of a function are often important numbers. For instance, a zero of the function of the function

x2 2

is a square root of 2, and that number we know was A zero

exztremely important to the Greeks as they began the study of what real numbers are.

z2 + 1

is something whose square is -1, i.e., negative. The idea of a square being

negative was implausible at rst, but is fundamental now, so that the zero of this particular function is critical for our understanding to numbers. Very likely, then, the zeroes of any new function will be worth studying. For instance, we will soon see that, perhaps disappointingly, there are no zeroes for the exponential function:

exp (z )

is never 0. Maybe it's even more interesting then that

there are zeroes of the sine function. The next theorem establishes some familiar facts about the trigonometric functions.

Theorem 3.16:
1. 2. Let

exp (iz ) = cos (z ) + isin (z ) for all z C. {zk } be a sequence of complex numbers lim {zk }

that converges to 0. Then

sin (zk ) = 0. zk

(3.49)

3. Let

be a sequence of complex numbers that converges to 0. Then

lim

1 cos (zk ) 1 = . 2 zk 2

(3.50)

Exercise 3.8.4
Prove Theorem 3.16, p. 74. HINT: For parts (2) and (3), use Theorem 3.13, p. 69.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

75

3.9 Analytic Functions and Taylor Series

Denition 3.13:
Let Then

S be a subset of C, let f : S C be a complex-valued function, and let c be a point of S. f is said to be expandable in a Taylor series around c with radius of convergence r if there exists an r > 0 such that Br (c) S, and f (z ) is given by the formula

f (z ) =
n=0
for all Let Then

an (z c)

(3.51)

z Br (c) . S be a subset
is said to be

expandable in a Taylor series around c

(c r, c + r) S,

and

of

R,

let

f :SR

be a real-valued function on

S,

and let

be a point of

S.

with radius of convergence

if there

exists an

r>0

such that the interval

f ( x)

is given by the formula

f (x) =
n=0
for all

an (x c)

(3.52)

x (c r, c + r) . S is an open subset of C. A function f : S C is called analytic on S if it is expandable in a Taylor series around every point c of S. Suppose S is an open subset of R. A function f : S C is called real analytic on S if it is expandable in a Taylor series around every point c of S.
Suppose

Theorem 3.17:
Suppose

is a subset of

C,

that

f :SC

is a complex-valued function and that

belongs to

S.
is

Assume that Suppose continuous

is expandable in a Taylor series around

with radius of convergence

r.

Then

continuous at each Assume that

Proof:
by

S is a that f : S R f is expandable in a Taylor series at each x (c r, c + r ) .

z Br (c) . subset of R,

is a real-valued function and that around

belongs to

S.
is

with radius of convergence

r.

Then

If we let

g be the power series function given by g (z ) = T (z ) = z c, then f (z ) = g (T (z )) , and this theorem

an z n ,

and

be the function dened

is a consequence of Theorem 3.3, The

composition of continuous functions is continuous., p. 61 and Theorem 3.13, p. 69.

Exercise 3.9.1
Prove that

Exercise 3.9.2

HINT: Use

f (z ) = 1/z is analytic on its domain. r = |c|, and then use the innite geometric

series.

State and prove an Identity Theorem, analogous to Theorem 3.14, Identity Theorem, p. 71, for functions that are expandable in a Taylor series around a point

c.

Exercise 3.9.3

a. Prove that every polynomial is expandable in a Taylor series around every point Use the binomial theorem. b. Is the exponential function expandable in a Taylor series around the number

c.

HINT:

1?

9 This

content is available online at <http://cnx.org/content/m36168/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

76

CHAPTER 3. FUNCTIONS AND CONTINUITY

10

3.10 Uniform Convergence

numbers, and let

We introduce now two dierent notions of the limit of a sequence of functions. Let

be a set of complex

{fn }converges or converges pointwise to a function f : S C if for x S and every > 0 there exists a natural number N, depending on x and , such that for every n N,|fn (x) f (x) | < . That is, equivalently, {fn } converges pointwise to f if for every x S the sequence {fn (x)} of numbers converges to the number f (x) . We say that the sequence {fn }converges uniformly to a function f if for every > 0, there exists an N, depending only on , such that for every n N and every x S,|fn (x) f (x) | < . If {un } is a sequence of functions dened on S, we say that the innite series un converges uniformly if the sequence {SN = N u } of partial sums converges uniformly. n n=0
We say that the sequence every These two denitions of convergence of a sequence of functions dier in subtle ways. order in the denitions. Study the word

Denition 3.14:

{fn }

be a sequence of complex-valued functions each having domain

S.

Exercise 3.10.1
b.

a. Prove that if a sequence

{fn } of functions converges uniformly on a set S to a function f then f. n Let S = (0, 1) , and for each n dene fn (x) = x . Prove that {fn } converges pointwise to the zero function, but that {fn } does not converge uniformly to the zero function. Conclude that
it converges pointwise to pointwise convergence does the form

not

imply uniform convergence. let

HINT: Suppose the sequence

does converge uniformly. Take

= 1/2,

be a corresponding integer, and consider

x's

of

x=1h

for tiny

h's.

c. Suppose the sequence uniformly to d. e.

{fn } converges uniformly to f on S, and the sequence {gn } converges g on S. Prove that the sequence {fn + gn } converges uniformly to f + g on S. Suppose {fn } converges uniformly to f on S, and let c be a constant. Show that {cfn } converges uniformly to cf on S. 2 Let S = R, and set fn (x) = x + (1/n) . Does {fn } converge uniformly on S ? Does {fn } converge uniformly on S ? What does this say about the limit of a product of uniformly
convergent sequences versus the product of the limits?

f. Suppose g.

h.

a and b are nonnegative real numbers and that |a b| < 2 . Prove that | a b| < 2. HINT: Break this into cases, the rst one being when both a and b are less than . Suppose {fn } is a sequence of nonnegative real-valued functions that converges uniformly to f on S. Use part (f ) to prove that the sequence { fn } converges uniformly to f . 1+1/n For each positive integer n, dene fn on (1, 1) by fn (x) = |x| . Prove that the sequence {fn } converges uniformly on (1, 1) to the function f (x) = |x|. HINT: Let > 0 be given. Consider |x|'s that are < and |x|'s that are . For |x| < , show that |fn (x) f (x) | < 1/n for all n. For |x| , choose N so that | 1| < . How?

Exercise 3.10.2
Let

{fn }

be a sequence of functions on a set

n
to

we have

|f (x) fn (x) | < 1/n

for all

S, let f be a function on S, and suppose that for each x S. Prove that the sequence {fn } converges uniformly
The rst of these theorems is

f. f
is the uniform

We give next four important theorems concerning uniform convergence. limit of a sequence of continuous functions, then

frequently used to prove that a given function is continuous. The theorem asserts that if

is itself continuous.

10 This

content is available online at <http://cnx.org/content/m36178/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

77

Theorem 3.18:
Suppose

The uniform limit of continuous functions is continuous.

{fn }

is a sequence of continuous functions on a set

Proof:

{fn }

converges uniformly to a function

f.

Then

is continuous on

S C, and S.

assume that the sequence

This proof is an example of what is called by mathematicians a  3 argument. Fix an x S and |f (y ) f (x) | < . an

> 0.

We wish to nd a

>0

such that if

yS

and

|y x| <

then

there exists a natural number

Thus, given this > 0, n N then |f (z ) fn (z ) | < /3 for all z S. Now, because fN is continuous at x, there exists a > 0 such that if y S and |y x| < then |fN (y ) fN (x) | < /3. So, if y S and |y x| < , then We use rst the hypothesis that the sequence converges uniformly.

such that if

|f (y ) f (x) |

= < =

|f (y ) fN (y ) + fN (y ) fN (x) + fN (x) f (x) | |f (y ) fN (y ) | + |fN (y ) fN (x) | + |fN (x) f (x) |

3

(3.53)

This completes the proof.

3.4: REMARK Many properties of functions are preserved under the taking of uniform limits, e.g.,
continuity, as we have just seen. However, not all properties are preserved under this limit process. Dierentiability is not, integrability is sometimes, being a power series function is, and so on. We must be alert to be aware of when it works and when it does not.

Theorem 3.19:
Let

Weierstrass M-Test

{un }

be a sequence of complex-valued functions dened on a set

partial sum

SN (x) = |un (x) | Mn for all x

1. If

N n=0 un (x) . Suppose that, for each S. Then

n,

there

S C. Write SN for the exists an Mn > 0 for which

Mn

converges, then the sequence

innite series 2. If each function continuous.

{SN } converges uniformly to a function S. That is, the un converges uniformly. un is continuous, and Mn converges, then the function S of part (1) is

Proof:
series

Mn is convergent, it follows from the Comparison Test that for each x S the innite un (x) is absolutely convergent, hence convergent. Dene a function S by S (x) = u ( x ) = limSN (x) . n=0 n To show that {SN } converges uniformly to S, let > 0 be given, and choose a natural number N such that n=N +1 Mn < . This can be done because Mn converges. Now, for any x S
Because

n=0

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

78

CHAPTER 3. FUNCTIONS AND CONTINUITY

and any

m N,

we have

|S (x) Sm (x) |

= = = = = <

| lim Sk (x) Sm (x) |

k

| lim (Sk (x) Sm (x)) |

k k k

lim |Sk (x) Sm (x) |

k n=m+1 k n=m+1

lim | lim

un (x) |
(3.54)

k k

|un (x) | Mn

lim

k n=m+1

n=m+1 Mn n=N Mn

This proves part (1). Part (2) now follows from part (1) and Theorem 3.18, The uniform limit of continuous functions is continuous., p. 76, since the

SN 's

are continuous.

n f (z ) = n=0 an z be a power series function with {SN (z )} denote the sequence of partial sums of this series:
Let

Theorem 3.20:

radius of convergence

r > 0,

and let

SN (z ) =
n=0

an z n . f
on the diskBr '

(3.55)

Proof:

If

0 < r' < r,

then the sequence

{SN }
by

converges uniformly to

(0) .

n n=0 |an |z , and note that the radius of convergence ' for g is the same as that for f, i.e., r. Choose t so that r < t < r. Then, since t belongs to the disk of n n convergence of the power series function g, we know that n=0 |an |t converges. Set mn = |an |t ,
Dene a power series function

g (z ) =

and note that

mn

converges. Now, for each

z Br' (0) ,
n

we have that (3.56) by the Weierstrass M-Test.

|an z n | |an |r' |an |tn = mn ,

so that the innite series

an z n

converges uniformly on

Br' (0)

Exercise 3.10.3
Let

f (z ) = n=0 z n . Recall that the radius of convergence for f is 1. Verify that the sequence {SN } of partial sums of this power series function fails to converge uniformly on the full open disk ' of convergence B1 (0) , so that the requirement that r < r is necessary in the preceding theorem.
The next theorem shows that continuous, real-valued functions on closed bounded intervals are uniform limits of step functions. Step functions have not been mentioned lately, since they aren't continuous functions, but this next theorem will be crucial for us when we study integration in Section 5.1.

Theorem 3.21:
Let

be a continuous real-valued function on the closed and bounded interval

[a, b] .

Then there

exists a sequence

Proof:
p. 66).

{hn }

of step functions on

[a, b]

that converges uniformly to

f.

We use the fact that a continuous function on a compact set is uniformly continuous (Theorem 3.9,

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

79

For each positive integer n, let n be a positive number satisfying |f (x) f (y ) | < 1/n if |xy | < n . Such a n exists by the uniform continuity of f on [a, b] . Let Pn = {x0 < x1 < ... < xmn } be a partition of [a, b] for which xi xi1 < n for all 1 i mn . Dene a step function hn on [a, b] as follows: If xi1 x < xi , then hn (x) = f (xi1 ) . This denes hn (x) for every x [a, b) , and we complete the denition of hn by setting hn (b) = f (b) . It follows immediately that hn is a step function. Now, we claim that since Then

f (b) = hn (b)

for all

|f (x) hn (x) | < 1/n n. For any other x,

for all let

x [a, b] .

This is clearly the case for

be the unique index such that

x = b, xi1 x < xi .

|f (x) hn (x) | = |f (x) f (xi1 ) | < 1/n

because

(3.57)

|x xi1 | < n . {hn }

of step functions, and the sequence

So, we have dened a sequence formly to

{hn }

converges uni-

by Exercise 3.10.2.

We close this chapter with a famous theorem of Abel concerning the behavior of a power series function on the boundary of its disk of convergence. See the comments following Exercise 3.7.4.

n n=0 an z is a power series function having nite radius of convergence r > 0, and suppose there exists a point z0 on the boundary of Br (0) that is in the domain of f ; i.e., n an z0 converges to f (z0 ) . Suppose g is a continuous function whose domain contains the open
Suppose

Theorem 3.22:
f (z ) =

Abel

disk

Then

Proof:
Sn

Br (0) as well as the point z0 , f (z0 ) must equal g (z0 ) .

and assume that

f (z ) = g (z )

for all

in the open disk

Br (0) .

For simplicity, assume that

for the partial sum of the

r = 1 and that z0 = 1. See the exercise that follows this proof. Write n an 's: Sn = n=0 an . In the following computation, we will use the
N N 1

Abel Summation Formula in the form

an z n = SN z N +
n=0
See Exercise 2.8.6. Let

Sn z n z n+1 .
n=0

(3.58)

be a positive number. Then, for any

0<t<1

and any positive integer

N,

we have

N n n |g (1) f (1) | = |g (1) f (t) + f (t) N n=0 an t + n=0 an t f (1) | N N n n |g (1) g (t) | + |f (t) f (1) | |g (1) n=0 an t | + | n=0 an t N N 1 n N n n+1 g (t) | + |f (t) a t | + | S t + S ) f (1) | = N n (t t n=0 n n=0 N 1 N n N |g (1) g (t) | + |f (t) n=0 an t + |SN t + n=0 (Sn SN ) (tn tn+1 ) + 1 n N n+1 n SN N ) f (1) | = |g (1) g (t) | + |f (t) + n=0 (t t n=0 an t

(3.59)

N 1 n=0

(Sn SN ) (tn tn+1 ) + SN tN +

N 1 n=0

(tn tn+1 ) f (1) | |g (1)

1 n n n+1 g (t) | + |f (t) N +| N ) | + |SN f (1) | n=0 an t | n=0 (Sn SN ) (t t P N n |g (1) g (t) | + |f (t) + | n=0 (Sn SN ) (tn tn+1 ) | + n=0 an t | N 1 | n=P +1 (Sn SN ) (tn tn+1 ) | + |SN f (1) | |g (1) g (t) | + |f (t) N 1 n n n+1 n n+1 +| P )| + N )+ n=0 an t | n=0 (Sn SN ) (t t n=P +1 |Sn SN | (t t |SN f (1) | = t1 + t2 + t3 + t4 + t5 .

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

80

CHAPTER 3. FUNCTIONS AND CONTINUITY

N are both larger than M1 , then t4 < . (The tk tk+1 is telescoping.) Fix such a P > M1 . Then choose a > 0 so that if 1 > t > 1 , then both t1 and t3 < . How? Fix such a t. Finally, choose a N, greater than M1 , and also large enough so that both t2 and t5 are less than . (How?) Now, |g (1) f (1) | < 5. Since this is true for every > 0, it follows that f (1) = g (1) , and
First, choose an integer so that if and sequence

M1

{Sk }

is a Cauchy sequence, and

the theorem is proved.

Exercise 3.10.4
Let ^

f, g, r,

and

z0

be as in the statement of the preceding theorem. Dene

f (z ) = f (z0 z )

and

g (z ) = g (z0 z ) .
a. Prove that

^ is a power series function

f (z ) = f (1) ;
^

to 1, and such that b. Show that point

n=0 bn converges to

n n=0 bn z , with radius of convergence equal ^ i.e., 1 is in the domain of f .

g z = 1.

is a continuous function whose domain contains the open disk ^ ^

B1 (0)

and the

c. Show that, if

f (1) = g (1) ,

then

f (z0 ) = g (z0 ) .

Deduce that the simplication in the

preceding proof is justied. d. State and prove the generalization of Abel's Theorem to a function Taylor series around a point

that is expandable in a

c.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

Chapter 4

Dierentiation, Local Behavior

4.1 Dierentiation, Local Behavior E^i = -1.
In this chapter we will nally see why

ei

is

1.

Along the way, we will give careful proofs of all the

standard theorems of Dierential Calculus, and in the process we will discover all the familiar facts about the trigonometric and exponential functions. At this point, we only know their denitions as power series functions. The fact that yet dened what is

sin2 + cos2 = 1 or that ex+y = ex ey x meant by e for an arbitrary number x.

are not at all obvious. In fact, we haven't even

The main theorems of this chapter include: 1. The 3. 4. 5.

2. The

Chain Rule (Theorem 4.7, Chain Rule, p. 89), Mean Value Theorem (Theorem 4.9, Mean Value Theorem, p. 93), The Inverse Function Theorem (Theorem 4.10, Inverse Function Theorem, p. 95), The Laws of Exponents ( and Corollary 4.1, Law of Exponents, p. 97), and Taylor's Remainder Theorem (Theorem 4.19, Taylor's Remainder Theorem, p. 107).
2

4.2 The Limit of a Function

Denition 4.1:
Let

The concept of the derivative of a function is what most people think of as the beginning of calculus. However, before we can even dene the derivative we must introduce a kind of generalization of the notion of continuity. That is, we must begin with the denition of the limit of a function.

an element of

f : S C be a function, where S C, and let c be a limit point of S that S. We say that f has a limit L as z approaches c, and we write
z c

is not necessarily

limf (z ) = L,

(4.1) and

if for every

>0

there exists a

>0

such that if

If the domain

is unbounded, we say that

f has a limit L as z approaches , and we write

zS

0 < |z c| < ,

then

|f (z ) L| < .

L = lim f (z ) ,
z
if for every

(4.2)

> 0 there exists a positive number B such that if z S and |z | B, then |f (z )L| < .

1 This 2 This

content is available online at <http://cnx.org/content/m36173/1.2/>. content is available online at <http://cnx.org/content/m36185/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

81

82

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

B S R, we say limx f (x) = L if for every > 0 there exists a real number x S and x B, then |f (x) L| < . And we say that limx f (x) = L if for every > 0 there exists a real number B such that if x S and x B, then |f (x) L| < . Finally, for f : (a, b) C a function of a real variable, and for c [a, b] , we dene the onesided (left and right) limits of f at c. We say that f has a left hand limit of L at c, and we write L = limxc0 f (x) , if for every > 0 there exists a > 0 such that if x (a, b) and 0 < c x < then |f (x) L| < . We say that f has a right hand limit of L at c, and write L = limxc+0 f (x) , if for every > 0 there exists a > 0 such that if x S and 0 < x c < then |f (x) L| < .
Analogously, if such that if

The rst few results about limits of functions are not surprising. The analogy between functions having limits and functions being continuous is very close, so that for every elementary result about continuous functions there will be a companion result about limits of functions.

Theorem 4.1:
Let

be a complex number. Let

f :SC

and

g:SC

be functions. Assume that both

f and

have limits as

approaches

c.

Then:

1. There exists a

>0

and a positive number

such that if

|f (z ) | < M.
2.

That is, if

has a limit as

approaches

c,

then

z S and 0 < |z c| < f is bounded near c.

then

z c
3.

lim (f (z ) + g (z )) = limf (z ) + limg (z ) .

z c z c

(4.3)

z c
4. If

lim (f (z ) g (z )) = limf (z ) limg (z ) .

z c z c

(4.4)

limzc g (z ) = 0,

then

z c g (z )
5. If

lim

f (z )

limzc f (z ) , limzc g (z )

(4.5)

limits as

u and v are the real and imaginary parts of a complex-valued function f, then u and v z approaches c if and only if f has a limit as z approaches c. And,
z c

have

limf (z ) = limu (z ) + ilimv (z ) .

z c z c

(4.6)

Exercise 4.2.1
a. Prove Theorem 4.1, p. 82. HINT: Compare with Theorem 3.2, p. 60. b. Prove that c.

limxc f (x) = L if and only if, for every sequence {xn } of elements of S that c, we have limf (xn ) = L. HINT: Compare with Theorem 3.4, p. 62. Prove the analog of Theorem 4.1, p. 82 replacing the limit as z approaches c by the limit as z approaches .
converges to

Exercise 4.2.2
a. Prove that a function

f :SC

is continuous at a point

of

if and only if

limxc f (x) =

f (c) .
same. b. Let

HINT: Carefully write down both denitions, and observe that they are verbetim the be a function with domain

is a function with domain

at

c.

Prove that

S, and let c be a limit point of S that is not in S. Suppose S {c}, that f (x) = g (x) for all x S, and that g is continuous limxc f (x) = g (c) .

Exercise 4.2.3
Prove that the following functions

have the specied limits

at the given points

c.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

83

a. b. c. d.

f (x) = x3 8 / x2 4 , c = 2, and L = 3. f (x) = x2 + 1 / x3 + 1 , c = 1, and L = 1. f (x) = x8 1 / x6 + 1 , c = i, and L = 4/3. f (x) = (sin (x) + cos (x) exp (x)) / x2 , c = 0,

and

L = 1.

Exercise 4.2.4
S of all nonzero real numbers by f (x) = c if x < 0 and f (x) = d if x > 0. limx0 f (x) exists if and only if c = d. (b) Let f : (a, b) C be a complex-valued function on the open interval (a, b) . Suppose c is a point of (a, b) . Prove that limxc f (x) exists if and only if the two one-sided limits limxc0 f (x) and limxc+0 f (x) exist and are equal.
Dene on the set Show that Suppose

Exercise 4.2.5: Change of variable in a limit

f : S C
is a function, and that

limxc f (x) = L.

Dene a function

by

g (y ) =

f ( y + c) .
a. What is the domain of c. Suppose

g?
and that

b. Show that 0 is a limit point of the domain of

T C,

that

h : T S,

g and that limy0 g (y ) = limxc f (x) . limyd h (y ) = c. Prove that

xc
(4.7)

y d

lim f (h (y )) = lim f (x) = L.

4.1: REMARK When we use the word  interior in connection with a set S, it is obviously important
S being thought of as a set of real numbers or as a set of complex c is in the interior of a set S of complex numbers if the entire disk B (c) of radius around c is contained in S. While, a point c belongs to the interior of a set S of real numbers if the entire interval (c , c + ) is contained in S. Hence, in the following denition, we will be careful to distinguish between the cases that f is a function of a real variable or is a function of a
to understand the context; i.e., is numbers. A point complex variable.

4.3 The Derivative of a Function

Denition 4.2:
Let

Now begins what is ordinarily thought of as the rst main subject of calculus, the derivative.

be a subset of

R,

let

be an element of the interior of

f : S C be a complex-valued function (of S. We say that f is dierentiable at c if

h0

a real variable), and let

lim

f (c + h) f (c) h f :SC S.

(4.8)

exists. (Here, the number Analogously, let variable), and let

is a real number.)

be a subset of

C,

let

be a complex-valued function (of a complex We say that

be an element of the interior of

is

dierentiable at c

if

h0
exists. (Here, the number

lim

f (c + h) f (c) h

(4.9)

is a complex number.)

3 This

content is available online at <http://cnx.org/content/m36186/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

84

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

If

subset of

f : S C is a function either of a real variable or a complex variable, and if S ' denotes the S consisting of the points c where f is dierentiable, we dene a function f ' : S ' C by f ' (x) = lim f (x + h) f (x) . h
(4.10)

h0

derivative of f. f : [a, b] C that is dierentiable at each point x (a, b) , and whose the ' derivative f is continuous on (a, b) , is called a smooth function on [a, b] . If there exists a partition {a = x0 < x1 < ... < xn = b} of [a, b] such that f is smooth on each subinterval [xi1 , xi ] , then f is called piecewise smooth on [a, b] . ' ' ' Higher order derivatives are dened inductively. That is, f is the derivative of f , and so on. (n) We use the symbol f for the nth derivative of f.
The function

f'

is called the

A continuous function

4.2: REMARK In the denition of the derivative of a function f, we are interested in the limit, as h
f
but of the quotient

approaches 0, not of of the function

q,

f (c+h)f (c) . Notice that 0 is not in the domain h but 0 is a limit point of that domain. This is the reason why we had to make

q (h) =

quotient.

such a big deal above out of the limit of a function. The function

is often called the

dierential

4.3: REMARK

As mentioned in Section 3.1, we are often interested in solving for unknowns that The most common such problem is to solve a dierential equation. In such a

are functions.

problem, there is an unknown function for which there is some kind of relationship between it and its derivatives. Dierential equations can be extremely complicated, and many are unsolvable. However, we will have to consider certain relatively simple ones in this chapter, e.g., and

f ' = f,f ' = f,

f ' = f.

'

There are various equivalent ways to formulate the denition of dierentiable, and each of these ways has its advantages. The next theorem presents one of those alternative ways.

Theorem 4.2:
c

Let

belong to the interior of a set

(either in

or in

C ),

and let

f :SC

be a function. Then

the following are equivalent. 1.

is dierentiable at

c.

That is,

h0

lim

f (c + h) f (c) exists. h

(4.11)

2.

xc
3. There exists a number

lim

f (x) f (c) exists. xc

such that the following two conditions hold:

(4.12)

and a function

f (c + h) f (c) = Lh + (h)
and

(4.13)

(h) = 0. h0 h lim
is unique and equals
'

(4.14)

L is unique and equals f ' (c) , and the function f (c) f (c) h.
In this case,

f (c + h)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

85

Proof:
That (1) and (2) are equivalent follows from by writing Suppose next that

as

c + h.

is dierentiable at

c,

and dene

L = f ' (c) = lim

Set

h0

f (c + h) f (c) . h

(4.15)

(h) = f (c + h) f (c) f ' (c) h.

Then clearly

(4.16)

f (c + h) f (c) = Lh + (h) ,
which is (4.13). Also

(4.17)

h) | ( h | =

(c)f | f (c+h)fh )f (c) | f (c+hh

'

(c)h

=
which tends to 0 as

f ' (c) |, c.

(4.18)

approaches 0 because

is dierentiable at

Hence, we have established

(4.13) and (4.14), showing that (1) implies (3). Finally, suppose there is a number

and a function

satisfying (4.13) and (4.14). Then

f (c + h) f (c) (h) =L+ , h h

which converges to

(4.19) 82. Hence,

L = f ' (c) ,
proved.

and so

L as h approaches 0 by (4.14) (h) = f (c + h) f (c) f ' (c) h.

and part (2) of Theorem 4.1, p.

Therefore, (3) implies (1), and the theorem is

4.4: REMARK Though it seems articial and awkward, Condition (3) of this theorem is very convenient for many proofs. One should remember it.

Exercise 4.3.1
b. Let

a. What is the domain of the function

of condition (3) in the preceding theorem? Is 0 in this

domain? Are there any points in the interior of this domain?

and

exists a

be as > 0 such

in part (3) of the preceding theorem. Prove that, given an that if

>0

there

|h| <

then

| (h) | < |h|.

Theorem 4.3:
If at a point

limh0 (f (c + h) f (c)) /h = L. Hence, there exists a positive number 0 f (c+h)f (c) such that | L| < 1 if |h| < 0 , implying that |f (c + h) f (c) | < |h| (|L| + 1)whenever h |h| < 0 . So, if > 0 is given, let be the minimum of 0 and / (|L| + 1) . If y S and |y c| < ,
We are assuming that then, thinking of

Proof:

f : S C is a function, either of a real variable or a complex variable, and if f is dierentiable c of S, then f is continuous at c. That is, dierentiability implies continuity.

as being

c + h,
(4.20)

|f (y ) f (c) | = |f (c + h) f (c) | < |h| (|L| + 1) = |y c| (|L| + 1) < .

(Every

can be written as

c+h

for some

h,

and

|y c| = |h|.)

Exercise 4.3.2
Dene

f (z ) = |z |

for

z C.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

86

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

a. Prove that

is continuous at every point of

b. Show that, if

is dierentiable at a point

C. c,

then

f ' (c) = 0.

HINT: Using part (b) of

Theorem 4.1, p. 82, evaluate

f ' (c)

in the following two ways.

f ' (c) = lim

and

1 |c + n | | c| 1 n

(4.21)

f ' (c) = lim

i |c + n | |c| i n

(4.22)

Show that the only way these two limits can be equal is for them to be 0. c. Conclude that d. Suppose

is not dierentiable anywhere. Indeed, if it were, what would the function is the function of a real variable that is dened by

have to be, and why wouldn't it satisfy (4.14)?

f :RR

f (x) = |x|.

Show that

is dierentiable at every point

x = 0.

How does this result not contradict part (c)?

The following theorem generalizes the preceding exercise.

Theorem 4.4:
Suppose

tiable at a point

Proof:

complex

f : S R is a real-valued function of a complex variable, and assume that f is dierenc S. Then f ' (c) = 0. That is, every real-valued, dierentiable function f of a ' ' variable satises f (c) = 0 for all c in the domain of f . f ' (c)
in two ways.

We compute

f ' (c) = lim

n

f c+
i n i n

1 n 1 n

f (c)

is a real number..

(4.23)

f ' (c) = lim

n
Hence,

f c+

f (c)

is a purely imaginary number.

(4.24)

f ' (c)

must be 0, as claimed.

4.5: REMARK This theorem may come as a surprise, for it shows that there are very few real-valued
dierentiable functions of a complex variable. For this reason, whenever dierentiable function, we will presume that

f : S R is a real-valued,

is a function of a real variable; i.e., that the domain and

S R.
Evaluating

limh0 q (h) in the two dierent ways, h real,

h pure imaginary,

led to the proof

of the last theorem. It also leads us to make denitions of what are called partial derivatives of real-valued functions whose domains are subsets of

C R2 .

As the next exercise will show, the

theory of partial derivatives of real-valued functions is a much richer theory than that of standard derivatives of real-valued functions of a single complex variable.

f : S R be dened on a set S C R2 , and let c = (a, b) = + + bi be a point in interior of S. We dene the partial derivative of f with respect to x at the point c = (a, b) by
Let formula

Denition 4.3:

the the

tialf f (a + h, b) f (a, b) (a, b) = lim , h0 tialx h

(4.25)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

87

and the

partial derivative of f with respect to y

at

c = (a, b)

by the formula

tialf f (a, b + h) f (a, b) (a, b) = lim , h0 tialy h

whenever these limits exist. (In both these limits, the variable

(4.26)

is a real variable.)(

It is clear that the partial derivatives of a function arise when we x either the real part of the variable or the imaginary part of the variable to be a constant, and then consider the resulting function of the other (real) variable. We will see in Exercise 4.3.3 that there is a denite dierence between a function's being dierentiable at a point having partial derivatives at the point

c = (a + bi)

in the complex plane

versus its

(a, b)

Exercise 4.3.3
a. Suppose that

in

R2 .

is a complex-valued function of a complex variable, and assume that both the real

b.

f are dierentiable at a point c. Show that f is dierentiable at c and f ' (c) = 0. Let f = u + iv be a complex-valued function of a complex variable that is dierentiable at a point c. Prove that both partial derivatives of u and v exist at c = (a, b) , and in fact that
and imaginary parts of

tialv tialu (c) + i (c) = f ' (c) tialx tialx

and

(4.27)

tialv tialu (c) + i (c) = if ' (c) . tialy tialy f

on

(4.28)

c. Dene a complex-valued function

C R2 by f (z ) = f (x + iy ) = x iy. Write f = u + iv, and show that both partial derivatives of u and v exist at every point, but that f is not a dierentiable function of the complex variable z.

The next theorem is, in part, what we call in calculus the dierentiation formulas.

Theorem 4.5:
f
and

Let

be functions (either of a real variable or a complex variable), which are both dieren-

tiable at a point 1. 2. 3.

c.

Let

and

be complex numbers. Then:

(af + bg ) (c) = af ' (c) + bg ' (c) . ' ' ' (Product Formula) f g is dierentiable at c, and (f g ) (c) = f (c) g (c) + f (c) g (c) . (Quotient Formula) f /g is dierentiable at c (providing that g (c) = 0), and af + bg
is dierentiable at

c,

and

'

f g
4. If

'

(c) =

g (c) f ' (c) f (c) g ' (c) (g (c))

2

(4.29)

and

f = u + iv is a complex-valued function, then f is dierentiable v are dierentiable at c, and f ' (c) = u' (c) + iv ' (c) .

at a point

if and only if

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

88

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

Proof:
We prove part (2) and leave parts (1), (3), and (4) for the exercises. We have

h0

)(f g )(c) lim (f g)(c+hh

= =

h)f (c)g (c) lim f (c+h)g(c+ h h0 )f (c)g (c+h) lim f (c+h)g(c+hh h0 )f (c)g (c) + lim f (c)g(c+hh h0
(4.30)

h0

)f (c) lim f (c+hh lim g (c + h) h0 )g (c) + lim f (c) lim g(c+hh h0 h0

f (c) g (c) + f (c) g ' (c) ,

'

where we have used Theorem 4.1, p. 82, Theorem 4.2, p. 84, and Theorem 4.3, p. 85.

Exercise 4.3.4
a. Prove parts (1), (3), and (4) of Theorem 4.5, p. 87. b. If

and

are real-valued functions that are dierentiable at a point

c, what can be said about

the dierentiability of c. Let d. e.

f be a constant Prove that f is dierentiable everywhere and that f ' (z ) = 0 for all z. ' Dene a function f by f (z ) = z. Prove that f is dierentiable everywhere and that f (z ) = 1 for all z. n k Verify the usual derivative formulas for polynomial functions: If p (z ) = k=0 ak z , then n ' k1 p (z ) = k=1 kak z .
Are they dierentiable functions? If so, are their derivatives again

max (f, g )? function f (z ) k.

What about power series functions?

power series functions? In fact, everything works as expected.

Theorem 4.6:
Let

be a power series function

is dierentiable at each point

n n=0 an z having radius of convergence in its open disk Br (0) of convergence, and

f (z ) =

r > 0.

Then

f ' (z ) =
n=0

nan z n1 =
n=1

nan z n1 .

(4.31)

Proof:
The proof will use part (3) of Theorem 4.2, p.
' '

r so that n |z | < r < r, and write for r |z |, i.e., |z | + = r . Note rst that the innite series n=0 |an |r' converges to a positive number we will call M. Also, from the Cauchy-Hadamard Formula, we know that the power series function nan wn has the same radius of convergence as does f, and hence n1 the innite series nan z converges to a number we will denote by L. We dene a function
84. Fix an with Choose
'

|z | < r.

'

by

(h) = f (z + h) f (z ) Lh

from which it follows immediately that

f (z + h) f (z ) = Lh + (h) ,
which establishes (4.13). To complete the proof that

(4.32)

is dierentiable at

z,

it will suce to

establish (4.14), i.e., to show that

h0

lim

(h) = 0. h

(4.33)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

89

That is, given

>0

we must show that there exists a

>0

such that if

0 < |h| <

then

| (h) /h| = |
Assuming, without loss of generality,

f (z + h) f (z ) L| < . h that |h| < , we have that |

P
n=0

(4.34)

)f (z ) | f (z+hh L| =

= = = = = | = = = =
so that if

P n an (z +h)n n=0 an z L| h P Pn P n nk k n h ) n=0 an ( k=0 ( k )z n=0 an z | L| h P Pn n nk k n a z h z (( ( ) ) ) n k=0 k | n=0 L| P Pn h n nk k h ) n=1 an ( k=1 ( k )z | L| h n n nk k1 h n=1 nan z n1 | n=1 an k=1 k z n n nk k 1 n1 h n=1 n | k=1 k z n=1 an 1 an z n n nk k1 | n=2 an h | k=2 k z n nk k 1 n |h| n=2 k=2 |an | k |z | n nk k 2 n |h| n=2 |an | k=2 k |z | |h| nk k 2 n |h| n=2 |an | k=2 n || k |z | nk k n n 1 |h| 2 n=0 |an | k=0 k |z | n 1 |h| 2 n=0 |an | (|z | + ) n 1 ' |h| 2 n=0 |an |r M |h| 2,
whenever

(4.35)

M = / 2,

then

| (h) /h| < ,

|h| < ,

as desired.

4.6: REMARK Theorem 4.6, p.

88 shows that indeed power series functions are dierentiable, and

in fact their derivatives can be computed, just like polynomials, by dierentiating term by term. This is certainly a result we would have hoped was true, but the proof is

not trivial.

The next theorem, the Chain Rule, is another nontrivial one. It deals with the dierentiability of the composition of two dierentiable functions. Again, the result is what we would have wanted, the composition of two dierentiable functions is itself dierentiable, but the argument required to prove it is tricky.

Theorem 4.7:
Let

Chain Rule

f :SC

be a function, and assume that

is a function, that composition

T C, that

the number

f is dierentiable at a point c. Suppose g : T C f (c) T, and that g is dierentiable at f (c) . Then the

gf

is dierentiable at

and
'

(g f ) (c) = g ' (f (c)) f ' (c) .

(4.36)

Proof:
Using part (3) of Theorem 4.2, p. 84, write

g (f (c) + k ) g (f (c)) = Lg k + g (k )
and

(4.37)

f (c + h) f (c) = Lf h + f (h) .

(4.38)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

90

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

We know from that theorem that

Lg = g ' (f (c)) g (k ) =0 k

and

Lf = f ' (c) . f (h) = 0. h

And, we also know that

k0

lim

and

h0

lim

(4.39)

Dene a function k (h) = f (c + h) f (c) . Then, by Theorem 4.3, p. 85, we have that limh0 k (h) = 0. We will show that g f is dierentiable at c by showing that there exists a number L and a function satisfying the two conditions of part (3) of Theorem 4.2, p. 84. Thus, we have that

g f (c + h) g f (c) g (f (c) + k (h)) g (f (c))

= g (f (c + h)) g (f (c)) = =

Lg k (h) + g (k (h)) = Lg (f (c + h) f (c)) + g (k (h)) = Lg (Lf h + f (h)) + g (k (h)) = Lg Lf h + Lg f (h) + g (k (h)) .

We dene (4.40)

L = Lg lf = g ' (f (c)) f ' (c) ,

and we dene the function

by (4.41)

(h) = Lg f (h) + g (k (h)) .

By our denitions, we have established (4.13)

g f (c + h) g f (c) = Lh + (h) ,
so that it remains to verify (4.14). We must show that, given First, choose an

(4.42)

> 0, there exists a > 0 such that if 0 < |h| <

2
'

then

| (h) /h| < .

' > 0

so that

|Lg |' + |Lf |' + ' <

Next, using part (b) of Exercise 4.3.1, choose a Finally, choose and

(4.3) .
' '

(4.43)

> 0 such that if |k | < then |g (k ) | < |k |. > 0 so that if 0 < |h| < , then the following two inequalities hold. |k (h) | < ' |f (h) | < ' |h|. The rst can be satised because f is continuous at c, and the second is a

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

91

consequence of part (b) of Exercise 4.3.1. Then: if

0 < |h| < , = |Lg f (h) + g (k (h)) |

| (h) | |Lg ||f (h) | + |g (k (h)) |

< |Lg | |h| + |k (h) | = |Lg | |h| + |f (c + h) f (c) | = |Lg | |h| + |Lf h + f (h) | |Lg |' |h| + ' |Lf ||h| + ' |f (h) | < |Lg | |h| + |Lf ||h| + |h| = |Lg | + |Lf | +
whence
' ' ' ' ' ' ' ' ' ' ' ' '

(4.44)

|h|,

| (h) /h| <

as desired.

|Lg |' + |Lf |' + '

< ,

(4.45)

Exercise 4.3.5
a. Derive the familiar formulas for the derivatives of the elementary transcendental functions:

exp' = exp, sin' = cos, , sinh' = cosh, cosh' = sinh and cos' = sin.

(4.46)

b. Dene a function
'

c.

f as follows. f (z ) = cos2 (z ) + sin2 (z ) . Use part (a) and the Chain Rule to 2 2 show that f (z ) = 0 for all z C. Does this imply that cos (z ) + sin (z ) = 1 for all complex numbers z ? n Suppose f is expandable in a Taylor series around the point c :f (z ) = n=0 an (z c) for all z Br (c) . Prove that f is dierentiable at each point of the open disk Br (c) , and show
that

f ' (z ) =
n=1

nan (z c)

n1

(4.47)

HINT: Use Theorem 4.6, p. 88 and the chain rule.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

92

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

4

4.4 Consequences of Dierentiability, the Mean Value Theorem

Denition 4.4:
Let

S.

f : S R be a real-valued function of a real variable, and let c be an element of the interior of f is said to attain a local maximum at c if there exists a > 0 such that (c , c + ) S and f (c) f (x) for all x (c , c + ) . The function f is said to attain a local minimum at c if there exists an interval (c , c + ) S such that f (c) f (x) for all x (c , c + ) .
Then

The next theorem should be a familiar result from calculus.

Theorem 4.8:
Let

First Derivative Test for Extreme Values be a real-valued function of a real variable, and let

f :S R

attains a local maximum or a local minimum. If

Proof:

is dierentiable at

c S be a point at which f c, then f ' (c) must be 0.

attains

We prove the theorem when Thus, let

attains a local maximum at

c. The proof for the case when f

a local minimum is completely analogous.

> 0

be such that

suciently large, then both in two ways. First,

1 c+ n

and

f (c) f (x) for all x such that |x c| < . Note that, if n is 1 ' c n belong to the interval (c , c + ) . We evaluate f (c) f c+
1 n 1 n

f ' (c) = lim

n

f (c)

(4.48)

because the numerator is always nonpositive and the denominator is always positive. On the other hand,

f ' (c) = lim

n

f c

1 n 1 n

f (c)

0 f ' (c)
must be 0, as desired.

(4.49)

since both numerator and denominator are nonpositive. Therefore,

Of course we do not need a result like Theorem 4.8, First Derivative Test for Extreme Values, p. necessarily is 0, independent of whether or not the function attains an extreme value.

92

for functions of a complex variable, since the derivative of every real-valued function of a complex variable

4.7: REMARK

As mentioned earlier, the zeroes of a function are often important numbers.

The

preceding theorem shows that the zeroes of the derivative to nding the extreme values of the function for

f'

of a function

are intimately related

f.

The zeroes of

f'

are often called the

critical points

f. Part (a) of the Exercise 4.4.1 establishes the familiar procedure from calculus for determining

the maximum and minimum of a continuous real-valued function on a closed interval.

Exercise 4.4.1
a. Let

b. c.

[a, b] , and assume that f is M be the maximum value of f on this interval, and m be its minimum value on this interval. Write S for the set of all x (a, b) for which f ' (x) = 0. Suppose x is a point of [a, b] for which f (x) is either M or m. Prove that x either is an element of the set S, or x is one of the endpoints a or b. Let f be the function dened on [0, 1/2) by f (t) = t/ (1 t) . Show that f (t) < 1 for all t [0, 1/2) . Let t (1/2, 1/2) be given. Prove that there exists an r < 1, depending on t, such that |t/ (1 + y ) | < r for all y between 0 and t.
be a continuous real-valued function on a closed interval dierentiable at each point

in the open interval

(a, b) .

Let

4 This

content is available online at <http://cnx.org/content/m36203/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

93

d. Let

be a xed number for which

0 < t < 1.

Show that, for all

0 s t,(t s) / (1 + s) t.

Probably the most powerful theorem about dierentiation is the next one. It is stated as an equation, but its power is usually as an inequality; i.e., the absolute value of the left hand side is less than or equal to the absolute value of the right hand side.

Theorem 4.9:
f

Mean Value Theorem

Let that

be a real-valued continuous function on a closed bounded interval

[a, b] ,

and assume that

is dierentiable at each point

in the open interval

(a, b) .

Then there exists a point

c (a, b)

such

f (b) f (a) = f ' (c) (b a) .

(4.50)

Proof:
This proof is tricky. Dene a function

on

[a, b]

by (4.51) Indeed, (4.52)

h (x) = x (f (b) f (a)) f (x) (b a) .

Clearly,

is continuous on

[a, b]

and is dierentiable at each point

x (a, b) .

h' (x) = f (b) f (a) f ' (x) (b a) .

point

It follows from this equation that the theorem will be proved if we can show that there exists a

c (a, b)

for which

h' (c) = 0.

Note also that (4.53)

h (a) = a (f (b) f (a)) f (a) (b a) = af (b) bf (a)

and

h (b) = b (f (b) f (a)) f (b) (b a) = af (b) bf (a) ,

showing that Let

(4.54)

h (a) = h (b) .

h on the compact interval [a, b] and let M be the maximum value attained by h on [a, b] . If m = M, then h is a constant on [a, b] and h' (c) = 0 for all c (a, b) . Hence, the theorem is true if M = m, and we could use any c (a, b) . If m = M, then at least one of these two extreme values is not equal to h (a) . Suppose m = h (a) . Of course, m is also not equal to h (b) . Let c [a, b] be such that h (c) = m. Then, in ' fact, c (a, b) . By Theorem 4.8, First Derivative Test for Extreme Values, p. 92, h (c) = 0. ' We have then that in every case there exists a point c (a, b) for which h (c) = 0. This completes
be the minimum value attained by the continuous function the proof.

4.8: REMARK The Mean Value Theorem is a theorem about real-valued functions of a real variable,
and we will see later that it fails for complex-valued functions of a complex variable. (See part (f ) of Exercise 4.5.1.) In fact, it can fail for a complex-valued function of a real variable. Indeed, if

f (x) = u (x)+ iv (x) is a continuous complex-valued function on the interval [a, b] , and dierentiable on the open interval (a, b) , then the Mean Value Theorem certainly holds for the two real-valued functions u and v, so that we would have f (b) f (a) = u (b) u (a) + i (v (b) v (a)) = u' (c1 ) (b a) + iv ' (c2 ) (b a) ,
which is not (4.55)

f (c) (b a)

'

unless we can be sure that the two points

equal. This simply is not always possible. Look at the function

c1 and c2 can be chosen to be f (x) = x2 + ix3 on the interval

[0, 1] .

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

94

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

On the other hand, if

is a real-valued function of a complex variable (two real variables), then

a generalized version of the Mean Value Theorem does hold. See part (c) of Exercise 4.11.4. One of the rst applications of the Mean Value Theorem is to show that a function whose derivative is identically 0 is necessarily a constant function. This seemingly obvious fact is just variable, even though the Mean Value Theorem does not.

not

obvious. The next exercise shows that this result holds for complex-valued functions of a complex

Exercise 4.4.2
a. Suppose

b.

c.

f ' (x) = 0 for all x (a, b) . Prove that f is a constant function on (a, b) . HINT: Show that f (x) = f (a) for all x [a, b] by using the Mean Value Theorem applied to the interval [a, x] . Let f be a complex-valued function of a real variable. Suppose f is dierentiable at each ' point x in an open interval (a, b) , and assume that f (x) = 0 for all x (a, b) . Prove that f is a constant function. HINT: Use the real and imaginary parts of f. Let f be a complex-valued function of a complex variable, and suppose that f is dierentiable ' on a disk Br (c) C, and that f (z ) = 0 for all z Br (c) . Prove that f (z ) is constant on Br (c) . HINT: Let z be an arbitrary point in Br (c) , and dene a function h : [0, 1] C by h (t) = f ((1 t) c + tz ) . Apply part (b) to h. f
is a continuous real-valued function on

(a, b)

and that

The next exercise establishes, at last, two important identities.

Exercise 4.4.3
(cos

+ sin2 = 1

and

exp (i = 1. )

a. Prove that b.

c. d. e.

cos2 (z ) + sin2 (z ) = 1 for all complex numbers z. Prove that cos ( ) = 1. HINT: We know from part (a) that cos ( ) = 1. Using the Mean Value Theorem for the cosine function on the interval [0, ] , derive a contradiction from the assumption that cos ( ) = 1. Prove that exp (i ) = 1. HINT: Recall that exp (iz ) = cos (z ) + isin (z ) for all complex z. i (Note that this does not yet tell us that e = 1. We do not yet know that exp (z ) = ez .) 2 2 Prove that cosh z sinh z = 1 for all complex numbers z. Compute the derivatives of the tangent and hyperbolic tangent functions tan = sin/cos and tanh = sinh/cosh. Show in fact that tan' = 1 cos2
and

tanh' =

1 . cosh2

(4.56)

Here are two more elementary consequences of the Mean Value Theorem.

Exercise 4.4.4
a. Suppose that

b. c.

d.

f and g are two complex-valued functions of a real (or complex) variable, and suppose f ' (x) = g ' (x) for all x (a, b) (or x Br (c) .) Prove that there exists a constant k such that f (x) = g (x) + k for all x (a, b) (or x Br (c) .) ' Suppose f (z ) = cexp (az ) for all z, where c and a are complex constants with a = 0. Prove c ' ' that there exists a constant c such that f (z ) = exp (az ) + c . What if a = 0? a (A generalization of part (a)) Suppose f and g are continuous real-valued functions on the closed interval [a, b] , and suppose there exists a partition {x0 < x1 < ... < xn } of [a, b] such that both f and g are dierentiable on each subinterval (xi1 , xi ) . (That is, we do not ' ' assume that f and g are dierentiable at the endpoints.) Suppose that f (x) = g (x) for every x in each open subinterval (xi1 , xi ) . Prove that there exists a constant k such that f (x) = g (x) + k for all x [a, b] . HINT: Use part (a) to conclude that f = g + h where h is a step function, and then observe that h must be continuous and hence a constant. ' Suppose f is a dierentiable real-valued function on (a, b) and assume that f (x) = 0 for all x (a, b) . Prove that f is 1-1 on (a, b) .
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

95

Exercise 4.4.5
Let

f : [a, b] R

be a function that is continuous on its domain

[a, b]

and dierentiable on

(a, b) .

(We do not suppose that a. Prove that b.

f'

is continuous on

(a, b) .)
if and only if
'

f ' (x) 0 for all x (a, b) . Show that f is nonincreasing on [a, b] if and only if f (x) 0 for all x (a, b) . ' Conclude that, if f takes on both positive and negative values on (a, b) , then f is not f
is nondecreasing on

[a, b]

also 1-1.

(See the proof of Theorem 3.11, p. 67.) c. Show that, if

d.

f ' takes on both positive and negative values on (a, b) , then there must exist ' ' a point c (a, b) for which f (c) = 0. (If f were continuous, this would follow from the ' Intermediate Value Theorem. But, we are not assuming here that f is continuous.) Prove the Intermediate Value Theorem for Derivatives: Suppose f is continuous on the closed ' bounded interval [a, b] and dierentiable on the open interval (a, b) . If f attains two distinct ' ' ' values v1 = f (x1 ) < v2 = f (x2 ) , then f attains every value v between v1 and v2 . HINT: Suppose v is a value between v1 and v2 . Dene a function g on [a, b] by g (x) = f (x) vx. Now apply part (c) to g.

Here is another perfectly reasonable and expected theorem, but one whose proof is tough.

f : (a, b) R is a function that is continuous and 1-1 from (a, b) onto the interval a' , b' . ' 1 Assume that f is dierentiable at a point c (a, b) and that f (c) = 0. Then f is dierentiable at the point f (c) , and
Suppose

Theorem 4.10:

Inverse Function Theorem

f 1 (f (c)) =

'

1 . f ' (c)

(4.57)

Proof:
For, if

The formula

f 1 (f (c)) = 1/f ' (c) is f 1 (f (x)) = x, and f and f 1

'

no surprise.

This follows directly from the Chain Rule.

are both dierentiable, then

f 1 (f (c)) f ' (c) = 1,

'

which

gives the formula. The diculty with this theorem is in proving that the inverse function

f 1

of

f (c) . In fact, the rst thing to check is that the point f (c) belongs to the 1 interior of the domain of f , for that is essential if f 1 is to be dierentiable there, and here is where the hypothesis that f is a real-valued function of a real variable is important. According to ' ' Exercise 3.6.2, the 1-1 continuous function f maps [a, b] onto an interval a , b , and f (c) is in the ' ' 1 open interval a , b , i.e., is in the interior of the domain of f . 1 According to part (2) of Theorem 4.2, p. 84, we can prove that f is dierentiable at f (c) by
is dierentiable at showing that

f 1 (x) f 1 (f (c)) 1 . = ' x f (c) f (c) xf (c) lim

That is, we need to show that, given an then

(4.58) such that if

> 0,

there exists a

>0

0 < |x f (c) | <

f 1 (x) f 1 (f (c)) 1 ' | < . x f (c) f (c) 1/q

is continuous at the point

(4.59) there exists an

First of all, because the function that if

f ' (c) ,

' > 0

such

|q f (c) | < ,

'

'

then

1 1 |< | ' q f (c)

(4.60)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

96

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

Next, because

is dierentiable at

c,

there exists a

' > 0

such that if

0 < |y c| < '

then

|
Now, by Theorem 3.10, p. 66,

f (y ) f (c) f ' (c) | < ' . yc

continuous at the point

(4.61)

>0

such that if

f 1 is |x f (c) | < then

f (c) ,

and therefore there exists a

|f 1 (x) f 1 f (c) | < ' (4.62)

So, if

|x f (c) | < ,

then

|f 1 (x) c| = |f 1 (x) f 1 (f (c)) | < ' .

But then, by (4.61),

(4.63)

f f 1 (x) f (c) f ' (c) | < ' , f 1 (x) c

(4.64)

from which it follows, using (4.60), that

|
as desired.

f 1 (x) f 1 (f (c)) 1 ' | < , x f (c) f (c)

(4.65)

4.9: REMARK A result very like Theorem 4.10, Inverse Function Theorem, p.
the domain

95 is actually true for

complex-valued functions of a complex variable. We will have to show that if

is in the interior of

of a one-to-one, continuously dierentiable, complex-valued function

of a complex

variable, then

f (c)

is in the interior of the domain

f (S )

of

f 1 .

But, in the complex variable case,

this requires a somewhat more dicult argument.

Once that fact is established, the proof that

f 1

is dierentiable at

f (c)

will be the same for complex-valued functions of complex variables as

it is here for real-valued functions of a real variable. Though the proof of Theorem 4.10, Inverse Function Theorem, p. 95 is reasonably complicated for real-valued functions of a real variable, the corresponding result for complex functions is much more deep, and that proof will have to be postponed to a later chapter. See Theorem 7.10, Open Mapping Theorem, p. 198.

4.5 The Exponential and Logarithm Functions

exponential function, we mean the power series function logarithm function.

We derive next the elementary properties of the exponential and logarithmic functions.

Of course, by

exp.

And, as yet, we have not even dened a

Exercise 4.5.1
a. Dene a complex-valued function for all

f :CC

by

f (z ) = exp (z ) exp (z ) . Prove that f (z ) = 1 exp (z ) =

z C.

b. Conclude from part (a) that the exponential function is never 0, and that

1/exp (z ) .
c. Show that the exponential function is always positive on d. Prove that

exp

is continuous and 1-1 from

e. Show that the exponential function is

not

(, ) 1-1 on C.

onto

R, and that limx exp (x) = 0. (0, ) .

5 This

content is available online at <http://cnx.org/content/m36199/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

97

f. Use parts b and e to show that the Mean Value Theorem is not in any way valid for complexvalued functions of a complex variable. Using part (d) of the preceding exercise, we make the following important denition.

Denition 4.5:

function, and we denote this function by ln.

We call the inverse

exp1

of the restriction of the exponential function to

R the (natural) logarithm

The properties of the exponential and logarithm functions are strongly tied to the simplest kinds of dierential equations. The connection is suggested by the fact, we have already observed, that The next theorem, corollary, and exercises make these remarks more precise.

exp' = exp.

Theorem 4.11:
Suppose

f : C C

is dierentiable everywhere and satises the dierential equation

f ' = af,

where

Proof:

is a complex number. Then

f (z ) = cexp (az ) ,

where

c = f (0) .

Consider the function

h (z ) = f (z ) /exp (az ) .

Using the Quotient Formula, we have that

h' (z ) =

exp (az ) f ' (z ) aexp (az ) f (z ) [exp (az )] c

gives

2
such that

exp (az ) f ' (z ) af (z ) [exp (z )]

for all

= 0.

(4.66)

Hence, there exists a complex number for all

h (z ) = c

z.

Therefore,

f (z ) = cexp (az )

z.

Setting

z=0

f (0) = c,

as desired.

Corollary 4.1: Proof:

Fix

Law of Exponents

For all complex numbers

and

w,exp (z + w) = exp (z ) exp (w) .

and apply the preceding theorem. We have

w, dene f (z ) = exp (z + w) , exp (z + w) = f (z ) , so we get

f ' (z ) =

exp (z + w) = f (z ) = f (0) exp (z ) = exp (w) exp (z ) .

(4.67)

Exercise 4.5.2
a. If b. If

n is a positive integer and z is any complex number, show that exp (nz ) = (exp (z )) . r r is a rational number and x is any real number, show that exp (rx) = (exp (x)) .

Exercise 4.5.3
a. Show that b. Prove that the logarithm function c. d.

(0, ) onto R. ln is dierentiable at each point y (0, ) and that ln' (y ) = 1/y. HINT: Write y = exp (c) and use Theorem 4.10. Derive the rst law of logarithms: ln (xy ) = ln (x) + ln (y ) . Derive the second law of logarithms: That is, if r is a rational number and x is a positive real r number, show that ln (x ) = rln (x) . ln
is continuous and 1-1 from

We are about to make the connection between the number is the rst step.

e and the exponential function.

The next theorem

Theorem 4.12:
ln (1) = 0
and

Proof:

ln (e) = 1.
then

If we write

1 = exp (t) ,

t = ln (1) .

But

exp (0) = 1,

so that

ln (1) = 0,

which establishes the

rst assertion.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

98

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

Recall that

e = lim 1 +
n
Therefore,

1 n

(4.68)

ln (e)

= ln lim 1 +
n

= = = = = = =

limln
n

1+

1 n n 1 n n

1 limnln 1 + n n 1 ln(1+ n ) lim 1 n n 1 ln(1+ n )ln(1) lim 1 n

n

(4.69)

ln' (1) 1/1 1.

This establishes the second assertion of the theorem.

Exercise 4.5.4
a. Prove that

e=

1 . n ! n=0

(4.70)

HINT: Use the fact that the logarithm function is 1-1. b. For c. If

a rational number, show that

is a positive number and

exp (r) = er . r = p/q is a rational

number, show that (4.71)

ar = exp (rln (a)) .

d. Prove that

is irrational.

HINT: Let

(a). Show that

qn n!,

and that

pn /qn be the nth limqn (e pn /qn ) = 0.

partial sum of the series in part Then use Theorem 2.19, Test for

Irrationality, p. 52. We have nally reached a point in our development where we can make sense of raising any positive number to an arbitrary complex exponent. Of course this includes raising positive numbers to irrational powers. We make our general denition based on part (c) of the preceding exercise.

Denition 4.6:
For

a positive real number and

an arbitrary complex number, dene

az

by (4.72)

az = exp (zln (a)) .

4.10: REMARK The point is that our old understanding of what ar

rational number, coincides with the function arbitrary complex exponent.

means, where

exp (rln (a)) .

So, this new denition of

and is consistent with our old denition. And, it now allows us to raies a

a > 0 and r is a az coincides positive number a to an

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

99

4.11: REMARK Let the bugles sound!!

that if want:

Now, having made all the appropriate denitions and derived

all the relevant theorems, we can nally prove that

ei = 1.

From the denition above, we see

a = e,

then we have

ez = exp (z ) .

Then, from part (c) of Exercise 4.4.3, we have what we

ei = 1.

(4.73)

Exercise 4.5.5
a. Prove that, for all complex numbers b. If

and

w,ez+w = ez ew .
z

is a real number and

is any complex number, show that

(ex ) = exz . f : C C by f (z ) = az . z that f (z ) = ln (a) a . If a and b are positive real numbers and z
'

(4.74)

c. Let

be a xed positive number, and dene a function

Show that and

is dierentiable at every

zC

and

d. Prove the general laws of exponents: complex numbers,

are (4.75)

az+w = az aw , az bz = (ab) ,
and, if

(4.76)

is real,

axw = (ax ) .
e. If f. g.

(4.77)

y is a real number, show that |eiy | = 1. If z = x + iy is a complex number, show that |e | = ex . Let = a + bi be a complex number, and dene a function f : (0, ) C by f (x) = x = ln(x) ' 1 e . Prove that f is dierentiable at each point x of (0, ) and that f (x) = x . a Let = a + bi be as in part (f ). For x > 0, show that |x | = x .
z

4.6 The Trigonometric and Hyperbolic Functions

and hyperbolic functions, give the following addition formulas:

The laws of exponents and the algebraic connections between the exponential function and the trigonometric

Theorem 4.13:

The following identities hold for all complex numbers

and

w.
(4.78)

sin (z + w) = sin (z ) cos (w) + cos (z ) sin (w) . cos (z + w) = cos (z ) cos (w) sin (z ) sin (w) . sinh (z + w) = sinh (z ) cosh (w) + cosh (z ) sinh (w) .
6 This
content is available online at <http://cnx.org/content/m36196/1.2/>.

(4.79)

(4.80)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

100

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

cosh (z + w) = cosh (z ) cosh (w) + sinh (z ) sinh (w) .

(4.81)

Proof:
We derive the rst formula and leave the others to an exercise. First, for any two real numbers

and

y,

we have

cos (x + y ) + isin (x + y )

= = = =

ei(x+y) eix eiy (cosx + isinx) (cosy + isiny ) cosxcosy sinxsiny + i (cosxsiny + sinxcosy ) ,
(4.82)

which, equating real and imaginary parts, gives that

cos (x + y ) = cosxcosy sinxsiny

and

(4.83)

sin (x + y ) = sinxcosy + cosxsiny.

holds for real numbers and let

(4.84)

The second of these equations is exactly what we want, but this calculation only shows that it

and

holds for all complex numbers

y. We can use the Identity Theorem to show that in fact this formula z and w. Thus, x a real number y. Let f (z ) = sinzcosy + coszsiny, 1 2i 1 iz iy e e eiz eiy .(4.85) 2i

g (z ) = sin (z + y ) =
Then both calculation,

ei(z+y) ei(z+y) =

g (z )

for all

f and g are power series functions of the variable z. Furthermore, by the previous f (1/k ) = g (1/k ) for all positive integers k. Hence, by the Identity Theorem, f (z ) = complex z. Hence we have the formula we want for all complex numbers z and all real z.
Let

numbers

y. f (w ) =
and let

To nish the proof, we do the same trick one more time. Fix a complex number

sinzcosw + coszsinw,

g (w) = sin (z + w) =
Again, both

1 2i

ei(z+w) ei(z+w) =

1 iz iw e e eiz eiw .(4.86) 2i w,

and they agree on the sequence

and

are power series functions of the variable

{1/k }.

Hence they agree everywhere, and this completes the proof of the rst addition formula.

Exercise 4.6.1
a. Derive the remaining three addition formulas of the preceding theorem. b. From the addition formulas, derive the two half angle formulas for the trigonometric functions:

sin2 (z ) =
and

1 cos (2z ) , 2 1 + cos (2z ) . 2

(4.87)

cos2 (z ) =

(4.88)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

101

Theorem 4.14:
The trigonometric functions

sin

and

cos

are periodic with period

2 ;

i.e.,

sin (z + 2 ) = sin (z )

Proof:

and

cos (z + 2 ) = cos (z )

for all complex numbers

z.

We have from the preceding exercise that

that the periodicity assertion for the sine function will follow if we show that

sin (z + 2 ) = sin (z ) cos (2 ) + cos (z ) sin (2 ) , so cos (2 ) = 1 and 1 cos (2 ) 2

it then follows that

sin (2 ) = 0.

From part (b) of the preceding exercise, we have that

0 = sin2 ( ) =
which shows that

(4.89)

cos (2 ) = 1.

Since

cos2 + sin2 = 1,

sin (2 ) = 0.

The periodicity of the cosine function is proved similarly.

Exercise 4.6.2
a. Prove that the hyperbolic functions b. Prove that the hyperbolic cosine

c.

sinh and cosh are periodic. What is the period? cosh (x) is never 0 for x a real number, that the hyperbolic tangent tanh (x) = sinh (x) /cosh (x) is bounded and increasing from R onto (1, 1) , and 1 ' that the inverse hyperbolic tangent has derivative given by tanh (y ) = 1 / 1 y 2 . Verify that for all y (1, 1) tanh1 (y ) = ln 1+y 1y .
(4.90)

Exercise 4.6.3: Polar coordinates

Let

be a nonzero complex number. Prove that there exists a unique real number

0 < 2 1,
and

such that

z = rei , where r = |z |. HINT: If z = a + bi, then z = r +

b 2 r

a 2 r

a r

b +r i.

Observe that

= 1.

Show that there exists a unique

0 < 2

a r a such that r

1,1 = cos

b r b and r

= sin.

4.7 L'Hopital's Rule

Many limits of certain combinations of functions are dicult to evaluate because they lead to what's known as indeterminate forms. These are expressions of the form very denition of the derivative itself is such a case:

0/0,/,00 , ,1 , and the like.

They are

precisely combinations of functions that are not covered by our limit theorems. See Theorem 4.1, p. 82. The

limh0 (f (c + h) f (c)) = 0,limh0 h = 0, and we are lim (1 + 1/n) = 1,limn = , and we are interested 1 . L'Hopital's Rule, Theorem 4.16, p.

interested in the limit of the quotient of these two functions, which would lead us to the indeterminate form

0/0.

The denition of the number

in the limit of

(1 + 1/n) ,

is another example:

which leads to the indeterminate form

102 below, is our strongest tool for handling such indeterminate forms. To begin with, here is a useful generalization of the Mean Value Theorem.

Theorem 4.15:
Let

Cauchy Mean Value Theorem

and

be continuous real-valued functions on a closed interval

and assume that both point

and

are dierentiable on the open interval

[a, b] , suppose g (a) = g (b) , (a, b) . Then there exists a

c (a, b)

such that

f ' (c) f (b) f (a) = ' . g (b) g (a) g (c)

7 This
content is available online at <http://cnx.org/content/m36201/1.2/>.

(4.91)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

102

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

Exercise 4.7.1
Prove the preceding theorem. HINT: Dene an auxiliary function Theorem.

as was done in the proof of the original Mean Value

Theorem 4.16:
Suppose

The following theorem and exercise comprise what is called L'Hopital's Rule.

and

are dierentiable real-valued functions on the bounded open interval

(a, b)

and

assume that

f ' ( x) = L, xa+0 g ' (x) lim L is a real number. (a, a + ) .) Suppose further
where (Implicit in this hypothesis is that that either

(4.92)

g ' (x) = 0

for

in some interval

xa+0
or

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

xa+0

(4.93)

xa+0
then

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

xa+0

(4.94)

xa+0 g (x)

lim

f (x)

= L.

(4.95)

Proof:
Suppose rst that

xa+0
Observe rst that, because the function

lim f (x) = lim g (x) = 0.

xa+0
for all

(4.96)

g ' ( x) = 0

in some interval

(a, a + ) ,g ' (x)

is either always

positive or always negative on that interval. (This follows from part (d) of Exercise 4.4.5.) Therefore

(a, a + ) . Hence, since limxa+0 g (x) = g (x) = 0 on the interval (a, a + ) . f ' (c) Now, given an > 0, choose a positive < such that if a < c < a + then | ' L| < . g (c) Then, for every natural number n for which 1/n < , and every a < x < a + , we have by the Cauchy Mean Value Theorem that there exists a point c between a + 1/n and x such that
must be strictly monotonic on the interval

0,

we must have that

f ' (c) f (x) f (a + 1/n) L| = | ' L| < . g (x) g (a + 1/n) g (c) n

approaches

(4.97)

Therefore, taking the limit as

we obtain

|
for all

f (x) f (x) f (a + 1/n) L| = lim | L| n g (x) g (a + 1/n) g (x)

This proves the theorem in this rst case.

(4.98)

for which

a < x < a + .

Next, suppose that

xa+0

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

xa+0

(4.99)

This part of the theorem is a bit more complicated to prove. First, choose a positive

so that

f (x)

and

g (x)

are both positive on the interval

(a, a + ) .

This is possible because both functions

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

103

are tending to innity as such that

approaches

a.

Now, given an

> 0,

choose a positive number

<

|
for all

f ' (c) L| < g ' (c) 2

(4.100)

a < c < a + .

We express this absolute value inequality as the following pair of ordinary

inequalities:

L
Set

f ' (c) < ' <L+ . 2 g (c) 2

(4.101)

that for all

y = a + . Using a<x<y

the Cauchy Mean Value Theorem, and the preceding inequalities, we have

L
implying that

f (x) f (y ) < <L+ , 2 g ( x) g ( y ) 2 (g (x) g (y )) + f (y ) . 2

(4.102)

Dividing

(g (x) g (y )) + f (y ) < f (x) < 2 through by g (x) and simplifying we obtain L L

L+

(4.103)

g (y ) f (y ) g (y ) f (y ) L 2 L+ 2 f (x) + < <L+ + . 2 g ( x) g ( x) g (x) 2 g ( x) g ( x)

(4.104)

Finally, using the hypothesis that all constants, choose a

> 0,

with

limxa+0 g (x) = , and the fact that L, , g (y ) , < , such that if a < x < a + , then
L 2 g (y ) f (y ) + |< g (x) g (x) 2

and

f (y )

are

|
and

(4.105)

|
Then, for all

L+ 2 g (y ) f (y ) + |< . g (x) g (x) 2

(4.106)

a < x < a + ,

we would have

L<
implying that

f (x) < L + , g (x)

(4.107)

|
and the theorem is proved.

f ( x) L| < , g (x)

(4.108)

Exercise 4.7.2
a. Show that the conclusions of the preceding theorem also hold if we assume that

f ' ( x) = . xa+0 g ' (x) lim

HINT: Replace

(4.109)

by a large real number

and show that

f (x) /g (x) > B

if

0 < x a < .

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

104

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

b. Show that the preceding theorem, as well as part (a) of this exercise, also holds if we replace the (nite) endpoint

by

HINT: Replace the

's

by negative numbers

B.

c. Show that the preceding theorem, as well as parts a and b of this exercise, hold if the limit as

approaches

from the right is replaced by the limit as

approaches

from the left. HINT:

Replace

f (x) f

by

f (x)

and

g (x)

by

g (x) .

d. Give an example to show that the converse of L'Hopital's Rule need not hold; i.e., nd functions and

for which

limxa+0 f (x) = limxa+0 g (x) = 0,

exists, but

f (x) xa+0 g (x) lim

f ' (x) xa+0 g ' (x) lim

does not exist.

(4.110)

e. Deduce from the proof given above that if

f.

limxa+0 f ' (x) /g ' (x) = L and limxa+0 g (x) = , then limxa+0 f (x) /g (x) = L independent of the behavior of f. 1/x 1/x Evaluate limx x , and limx0 (1 x) . HINT: Take logarithms.

4.8 Higher Order Derivatives

Denition 4.7:
Let

say that

R (or C ), and Let f : S C be a function of a real (or complex) variable. We f is continuously dierentiable on S 0 if f is dierentiable at each point x of S 0 and the ' 0 1 function f is continuous on S . We say that f C (S ) if f is continuous on S and continuously 0 0 dierentiable on S . We say that f is 2-times continuously dierentiable on S if the rst derivative ' 0 f is itself continuously dierentiable on S . And, inductively, we say that f is k-times continuously dierentiable on S 0 if the k 1st derivative of f is itself continuously dierentiable on S 0 . We write f (k) for the k th derivative of f, and we write f C k (S ) if f is continuous on S and is k times 0 k (j ) continuously dierentiable on S . Of course, if f C (S ) , then all the derivatives f , for j k, 0 exist nd are continuous on S . (Why?) (0) For completeness, we dene f to be f itself, and we say that f C (S ) if f is continuous 0 on S and has innitely many continuous derivatives on S ; i.e., all of its derivatives exist and are 0 continuous on S . As in Section 3.1, we say that f is real-analytic (or complex-analytic ) on S if it is expandable 0 in a Taylor series around each point c S S
be a subset of

4.12: REMARK Keep in mind that the denition above, as applied to functions whose domain S is a

nontrivial subset of entiable on the set partial derivatives

Theorem 4.17:
Let

C, has to do with functions of a complex variable that are continuously dierS 0 . We have seen that this is quite dierent from a function having continuous 0 on S . We will return to partial derivatives at the end of this chapter. R
(or

be an open subset of

C ).

1. Suppose 2. 3.

W S is a subset of R. Then, for each k 1, there exists a function in C k (S ) that is not in C (S ) . That is, C k+1 (S ) is a proper subset of C k (S ) . If f is real-analytic (or complex-analytic) on S, then f C (S ) . There exists a function in C (R) that is not real-analytic on R. That is, the set of real analytic functions on R is a proper subset of the set C (R ) .
k+1

8 This

content is available online at <http://cnx.org/content/m36192/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

105

REMARK Suppose S
Variables that if

is an open subset of

C.

It is a famous result from the Theory of Complex

is in

C 1 (S ) ,

then

is necessarily complex analytic on

S.

We will prove this

amazing result in Theorem 7.5, p. 192. Part (3) of the theorem shows that the situation is quite dierent for real-valued functions of a real variable.

Proof:

For part (1), see the exercise below. Part (2) is immediate from part (c) of Exercise 4.3.5. Before nishing the proof of part (3), we present the following lemma:

Lemma 4.1:
Let

be the function dened on all of

as follows.

f (x) = {0x 0
where

p (x) e1/x x>0 xn

Then

(4.111)

at each point

Proof:
that

p (x) is a xed polynomial function and n is a xed nonnegative integer. x of R. x = 0. To see that f

is continuous

The assertion of the lemma is clear if

is continuous at 0, it will suce to prove

p (x) e1/x = 0. x0+0 xn lim

(Why?) But, for implying that

(4.112) 3.8.2 that

1/x

x > 0, we know from part (b) of Exercise < xn+1 (n + 1)!. Hence, for x > 0, |f (x) | =

e1/x > 1/ xn+1 (n + 1)! ,

|p (x) |e1/x < (n + 1)!x|p (x) |, xn

(4.113)

and this tends to 0 as

approaches 0 from the right, as desired.

Returning to the proof of Theorem 4.17, p. the preceding lemma then (Why?)

104, we verify part (3) by observing that if

is as in

is actually dierentiable, and its derivative

f'

is a function of the same sort.

It follows that any such function belongs to

C (R) .

On the other hand, a nontrivial such

cannot be expandable in a Taylor series around 0 because of the Identity Theorem. (Take completes the proof.

xk = 1/k.)

This

Exercise 4.8.1
a. Prove part (1) of Theorem 4.17, p. 104. Use functions of the form b. Prove that any function of the form of the on

xn sin (1/x) .

in the lemma above is everywhere dierentiable Conclude that any such function belongs to

R, and C (R) .

its derivative has the same form.

c. For each positive integer Prove that each

. (1, 1) , including 0. Prove also that the sequence {fn } converges uniformly to the function f (x) = |x|. (See part (h) of Exercise 3.10.1.)
dene a function on the interval by

n,

fn

(1, 1

fn (x) = |x|

1+1/n

fn

is dierentiable at every point in

Conclude that the uniform limit of dierentiable functions of a real variable need not be dierentiable. (Again, for functions of a complex variable, the situation is very dierent. In that case, the uniform limit of dierentiable functions 199.)

is dierentiable.

See Theorem 7.11, p.

Exercise 4.8.2: A smooth approximation to a step function.

Suppose

a < b < c < d are real numbers. Show that there exists a function in C (R) such that 0 (x) 1 for all x, (x) 1 for x [b, c] , and (x) 0 for x / (a, d) . (If a is close to b and c

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

106

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

is close to

d, then this function is a C

approximation to the step function that is 1 on the interval

[b, c]

and 0 elsewhere.)

a. Let

f be a function like the one in the lemma. Think about the graphs of the functions f (x c) and f (b x) . Construct a C function g that is 0 between b and c and positive
everywhere else.

b. Construct a c. Let d.

function h that is positive between a and d and 0 everywhere else. g and h be as in parts (a) and (b). If j = g + h, show that j is never 0, and write k the C function k = 1/j. Examine the function hk, and show that it is the desired function . Formula for the coecients of a Taylor Series function

for

Theorem 4.18:
Let

be expandable in a Taylor series around a point

c:
n
(4.114)

f (x) =

an (x c) .

Proof:

Then for each

n,an = f (n) (c) /n!.

Because each derivative of a Taylor series function is again a Taylor series function, and because the value of a Taylor series function at the point

is equal to its constant term

a1 = f ' (c) .

Computing the derivative of the derivative, we see that

a0 , we have that ' 2a2 = f ' (c) = f (2) (c) .

Continuing this, i.e., arguing by induction, we nd that

n!an = f (n) (c) ,

which proves the theorem.

4.9 Taylor Polynomials and Taylor's Remainder Theorem

Denition 4.8:
Let

Taylor polynomial

be in

C n (Br (c))

for

of degree

c a xed complex number, r > 0, and n a positive integer. n for f at c to be the polynomial T n T(n f,c) given by the
n

Dene the formula:

T(n f,c) (z ) =
j =0
where

aj (z c) ,

(4.115)

4.13: REMARK If f
n

aj = f (j ) (c) /j !.
is expandable in a Taylor series on

Br (c) ,

then the Taylor polynomial for

of

degree

is nothing but the

nth

partial sum of the Taylor series for

on

Br (c) .

However, any

function that is

times dierentiable at a point

has a Taylor polynomial of order

n.

Functions

that are innitely dierentiable have Taylor polynomials of all orders, and we might suspect that these polynomials are some kind of good approximation to the function itself.

Exercise 4.9.1
Prove that

is expandable in a Taylor series function around a point

r > 0)

if and only if the sequence

{T(n f,c) }

of Taylor polynomials converges pointwise to

c (with radius of convergence f ; i.e.,

(4.116)

f (z ) = lim T(n f,c) (z )

Exercise 4.9.2
Let

for all

in

Br (c) .
Prove that

f C n (Br (c)) .

f ' C n1 (Br (c)) .

Prove also that

T(n f,c)

'

1 = T(n . f ' ,c)

9 This

content is available online at <http://cnx.org/content/m36204/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

107

The next theorem is, in many ways, the fundamental theorem of numerical analysis. It clearly has to do with approximating a general function by polynomials. It is a generalization of the Mean Value Theorem, and as in that case this theorem holds only for real-valued functions of a real variable.

Theorem 4.19:
f f is c and x
(n)

Taylor's Remainder Theorem

Let

be a real-valued function on an interval dierentiable on such that

and that between

(c r, c + r) , and assume that f C n ((c r, c + r)) , (c r, c + r) . Then, for each x in (c r, c + r) there exists a y f (n+1) (y ) n+1 (x c) . (n + 1)!

f (x) T(n f,c) (x) =

(4.7)

(4.117)

n REMARK If we write f (x) = Tf,c (x) + Rn+1 (x) , where Rn+1 (x) is the error or remainder term,
then this theorem gives a formula, and hence an estimate, for that remainder term. This is the evident connection with Numerical Analysis.

Proof:
Thus,

We prove this theorem by induction on

n.

For

n = 0,

this is precisely the Mean Value Theorem.

0 f (x) Tf,c (x) = f (x) f (c) = f ' (y ) (x c.(4.118)

Now, assuming the theorem is true for all functions in the given function

C n1 ((c r, c + r)) , let us show it is true for

n+1

f C n ((c r, c + r)) . Set g (x) = f (x) T(n f,c) (x) and let h (x) = (x c) g (c) = 0
and

Observe that both

h (c) = 0.

Also, if

x = c,

then

h (x) = 0.

So, by the Cauchy Mean

Value Theorem, we have that

g (x) g (c) g ' (w) g (x) = = ' h (x) h (x) h (c) h (w )

for some

(4.119)

between

and

x.

Now
'

g ' (w) = f ' (w) tn (f,c)

(See the preceding exercise.), and

1 (w ) (w) = f ' (w) T(n f ' ,c) n

Therefore,

(4.120)

h' (w) = (n + 1) (w c) . = = =

n f (x)(T( f,c) )(x)

(xc)n+1

g (x) h(x) g ' (w) ' h (w ) 1 f ' (w) T n (w) (f ' ,c) . (n+1)(wc)n

(4.121)

We apply the inductive hypotheses to the function

f'

(which is in

C n1 ((c r, c + r))

and obtain

n f (x)(T( f,c) )(x)

(xc)n+1

= = = =

1 f ' (w) T n (w) (f ' ,c) (n+1)(wc)n

f'

(wc)n n! (n+1)(wc)n (n) f' (y ) (n+1)! f (n+1) (y ) (n+1)!

(n)

(y )

(4.122)

for some

between

and

w.

But this implies that

f (x) T(n f,c) (x) =

f (n+1) (y ) (x c) (n + 1)!

n+1

(4.123)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

108

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

for some

between

and

x,

which nishes the proof of the theorem.

Exercise 4.9.3
Dene for all

f (x) = 0 for x 0 n, and yet f is not

and

f (x) = e1/x

for

x > 0.

Verify that

f C (R ) ,

that

f (n) (0) = 0

expandable in a Taylor series around

Theorem for this function. That is, describe the remainder

0. Interpret Rn+1 (x) .

Taylor's Remainder

As a rst application of Taylor's Remainder Theorem we give the following result, which should be familiar from calculus. It is the generalized version of what's ordinarily called the second derivative test.

C n (c r, c + r) , suppose that the n + 1st derivative f (n+1) of f exists everywhere on (c r, c + r) and is continuous at c, and suppose that f (k) (c) = 0 for all 1 k n and that f (n+1) (c) = 0. Then:
Let

Theorem 4.20:
f

Test for Local Maxima and Minima

be a real-valued function in

1. If 2. If 3. If

n n n

is even,

called an

inection point.

attains neither a local maximum nor a local minimum at

c.

In this case,

is

is odd and is odd and

f (n+1) (c) < 0, f (n+1) (c) > 0,

then then

f f

attains a local maximum at attains a local minimum at

c. c.

Proof:
Since

(n+1)

f (n+1) (c) for

there exists a

c, there exists a > 0 such that f (n+1) (y ) has the same sign as all y (c , c + ) . We have by Taylor's Theorem that if x (c , c + ) then y between x and c such that
is continuous at

f (x) = T(n f,c) (x) +

from which it follows that

f (n+1) (y ) n+1 (x c) , (n + 1)!

(4.124)

f (x) f (c)

= =

n k=1

f (k) (c) k !(x c) +

f (n+1) (y ) (n+1)! (x

c)

f (n+1) (y ) (n+1)! (x n+1

c)

n+1
(4.125)

n is even. It follows then that if x < c, the sign of (x c) is negative, so that (n+1) the sign of f (x) f (c) is the opposite of the sign of f (c) . On the other hand, if x > c, then n+1 (x c) > 0, so that the sign of f (x) f (c) is the same as the sign of f (n+1) (c) . So, f (x) > f (c) for all nearby x on one side of c, while f (x) < f (c) for all nearby x on the other side of c. Therefore, f attains neither a local maximum nor a local minimum at c. This proves part (1). (n+1) Now, if n is odd, the sign of f (x) f (c) is the same as the sign of f (y ) , which is the same (n+1) as the sign of f (c) , for all x (c , c + ) . Hence, if f (n+1) (c) < 0, then f (x) f (c) < 0 (n+1) for all x (c , c + ) , showing that f attains a local maximum at c. And, if f (c) > 0, then the sign of f (x) f (c) is positive for all x (c , c + ) , showing that f attains a local minimum at c. This proves parts (2) and (3).
Suppose

n+1

4.10 The General Binomial Theorem

10 This

10

We use Taylor's Remainder Theorem to derive a generalization of the Binomial Theorem to nonintegral exponents. First we must generalize the denition of binomial coecient.
content is available online at <http://cnx.org/content/m36205/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

109

Denition 4.9:
coecient
Let

be a complex number, and let

be a nonnegative integer. We dene the general

binomial
(4.126)

by

( 1) ... ( k + 1) = . k! k is itself a positive integer and k , then coecient, and k = 0 when k > . However, number, then every k = 0.
If

k agrees with the earlier denition of the binomial if is not an integer, but just an arbitrary complex

Exercise 4.10.1
a. Show that

Estimates for the size of binomial coecients. Let

be a xed complex number.

|
for all nonnegative integers

| | | 1+ k j j =1

(4.127)

k.

HINT: Note that

|| (|alpha| + 1) (|alpha| + 2) ... (|| + k 1) . | k! k C

such that

(4.128)

b. Use part (a) to prove that there exists a constant

|
for all nonnegative integers c. Show in fact that for each

| C 2k k
for all that

(4.129)

k. HINT: Note that (1 + ||/j ) < 2 > 0 there exists a constant C such | k | C (1 + ) k
k=0 k

j > ||.

(4.130)

d. Let

k. h (t) be the power series function given by h (t) = that the radius of convergence for h equals 1.
for all nonnegative integers

tk . Use the ratio test to show

4.14: REMARK The general Binomial Theorem, if there is one, should be something like the following:

(x + y ) =
k=0

k k x y . k x = 0,

(4.131)

The problem is to determine when this innite series converges, i.e., for what values of the three variables that

x, y, x = 0, in

and

does it converge. It certainly is correct if

so we may as well assume

which case we are considering the validity of the formula

(x + y ) = x (1 + t) = x
k=0
where series

k t , k t
and

(4.132)

t = y/x.

Therefore, it will suce to determine for what values of

does the innite

k=0

k t k

(4.133)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

110

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

equal

(1 + t) .
The answer is that, for n arbitrary complex number all

(4.134)

, this series converges to the correct value for t (1, 1) . (Of course, t must be larger than 1 for the expression (1 + t) even to be dened.) However, the next theorem only establishes this equality for t's in the subinterval (1/2, 1/2) .
As mentioned earlier, its proof is based on Taylor's Remainder Theorem. We must postpone the complete proof to Section 5.1, where we will have a better version of Taylor's Theorem.

Theorem 4.21:
Let

= a + bi

be a xed complex number. Then

(1 + t) =
k=0

k t k

(4.135)

Proof:

for all

t (1/2, 1/2) .

is a nonnegative integer, for it is then just the original Binomial t. For a general complex number , we have only dened x for positive x's, so that (1 + t) is not even dened for t < 1. Now, for a general = a + bi, consider the function g : (1/2, 1/2) C dened by g (t) = (1 + t) . Observe that the nth derivative of g is given by
Of course, this theorem is true if Theorem, and in fact in that case it holds for every complex number

g (n) (t) =
Then

( 1) ... ( n + 1) (1 + t)
n
is actually in

(4.136) but the present theorem is

g C ((1/2, 1/2)) . (Of course, g only concerned with t's in (1/2, 1/2) .) For each nonnegative integer k dene ak = g (k) (0) /k ! =
and set

C (1, 1) ,

( 1) ... ( k + 1) = , k! k h (t) =

(4.137)

k k=0 ak t . According to part (d) of the preceding exercise, the radius of convergence for the power series ak tk is 1. The aim of this

equal to the power series function given by

theorem is to show that

g (t) = h (t)

for all

1/2 < t < 1/2.

In other words, we wish to show that

(1/2, 1/2) . It will suce to show {Sn } of partial sums of the power series function h converges to the function g, at least on (1/2, 1/2) . We note also that the nth partial sum of this power series is just the nth n Taylor polynomial Tg for g.
agrees with this power series function at least on the interval that the sequence

Sn (t) =
k=0
Now, x a

k t = k

k=0

g (k) (0) k t . k!

(4.138)

That is,

t strictly between 1/2 and 1/2, and let r < 1 be as in part (c) of Exercise 4.10.1. |t/ (1 + y ) | < r for every y between 0 and t. (This is an important inequality for our proof, and this is one place where the hypothesis that t (1/2, 1/2) is necessary.) Note also that, for a a a any y (1/2, 1/2) , we have | (1 + y ) | = (1 + y ) , and this is trapped between (1/2) and (3/2) . Hence, there exists a number M such that | (1 + y ) | M for all y (1/2, 1/2) . Next, choose an > 0 for which = (1 + ) r < 1. We let C be a constant satisfying the
inequality in Part (c) of Exercise 4.10.1. So, using Taylor's Remainder Theorem, we have that

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

111

there exists a

between 0 and

for which

|g (t)

n k=0

ak tk | = = = |

|g (t) T(n g,0) (t) |

(y ) n+1 t | | g (n+1)! (1)...(n) n+1 | (n t | +1)!(1+y )n+1 n+1 t n+1 || (1 + y ) || 1+ y| n+1 t M | 1+ y| n+1
(4.139)
(n+1)

C (1 + )

C (1 + )

n+1

M rn+1

C M n+1 , . < 1,
shows that

Taking the limit as

tends to

and recalling that

g (t) = h (t)

for all

1/2 < t < 1/2,

which completes the proof.

4.11 More on Partial Derivatives

variable and p.

11

We close the chapter with a little more concerning partial derivatives. Thus far, we have discussed functions of a single variable, either real or complex. However, it is dicult not to think of a function of one complex

z = x + iy a + bi to mean

as equally well being a function of the two real variables the same point in

C R2 ,

and we will write

quantity, i.e., the absolute value of the complex number

x and y. We will write (a, b) | (a, b) | and |a + bi| to indicate the same a + bi (a, b) . We have seen in Theorem 4.4,

86 that the only real-valued, dierentiable functions of a complex variable are the constant functions.

However, this is far from the case if we consider real-valued functions of two real variables, as is indicated in Exercise 4.3.3. Consequently, we make the following denition of dierentiability of a real-valued function of two real variables. Note that it is clearly dierent from the denition of dierentiability of a function of a single complex variable, and though the various notations for these two kinds of dierentiability are clearly ambiguous, we will leave it to the context to indicate which kind we are using.

f : S R be a function whose domain is a subset S of R2 , and let c = (a, b) be a point in 0 the interior S of S. We say that f is dierentiable, as a function of two real variables, at the point (a, b) if there exists a pair of real numbers L1 and L2 and a function such that
Let

Denition 4.10:

f (a + h1 , b + h2 ) f (a, b) = L1 h1 + L2 h2 + (h1 , h2 )
and

(4.140)

(h1 , h2 ) = 0. |(h1 ,h2 )|0 | (h1 , h2 ) | lim

One should compare this denition with part (3) of Theorem 4.2, p. 84. Each partial derivative of a function

(4.141)

is again a real-valued function of two real variables, and

so it can have partial derivatives of its own. We use simplifying notation like to indicate higher order mixed partial derivatives. For instance, derivative of

fxyxx

and

fyyyxyy... y,

fxxyx

denotes the fourth partial

f,

rst with respect to

x,

second with respect to

again, third with respect to

and nally fourth with respect to

derivatives.

x.

These higher order partial derivatives are called

mixed partial

11 This

content is available online at <http://cnx.org/content/m36206/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

112

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

Denition 4.11:
Suppose

is a subset of

R2 ,

and that

is a continuous real-valued function on

S.

If both partial

derivatives of

is said to belong to

S 0 , then f k th order mixed partial derivatives exist at each point of S 0 , and 0 k all of them are continuous on S , then f is said to belong to C (S ) . Finally, if all mixed partial 0 derivatives, of arbitrary orders, exist and are continuous on S , then f is said to belong to C (S ) . f
exist at each point of the interior of

S0

S,

and both are continuous on

C 1 (S ) .

If all

Exercise 4.11.1
a. Suppose

is a real-valued function of two real variables and that it is dierentiable, as a

function of two real variables, at the point

denition are exactly the partial derivatives of

(a, b) . Show that the numbers L1 f at (a, b) . That is,

and

L2

in the

L1 =
and

f (a + h, b) f (a, b) tialf (a, b) = lim h0 tialx h f (a, b + h) f (a, b) tialf (a, b) = lim . h0 tialy h

(4.142)

L2 =
b. Dene

(4.143)

f (0, 0) = 0, and if (x, y ) = (0, 0) , then f (x, y ) = xy/ x2 + y 2 . Show that both partial derivatives of f at (0, 0) exist and are 0. Show also that f is not, as a function of two real variables, dierentiable at (0, 0) . HINT: Let h and k run through the numbers 1/n. f
on as follows:

R2

c. What do parts (a) and (b) tell about the relationship between a function of two real variables being dierentiable at a point d. Suppose

(a, b)

and its having both partial derivatives exist at

f = u + iv

is a complex-valued function of a complex variable, and assume that

(a, b)? f

is

dierentiable, as a function of a complex variable, at a point the real and imaginary parts Relate the ve quantities

c = a + bi (a, b) .

Prove that

u and v

of

are dierentiable, as functions of two real variables.

tialu tialu tialv tialv (a, b) , (a, b) , (a, b) , (a, b) , tialx tialy tialx tialy

and

f ' (c) .

(4.144)

Perhaps the most interesting theorem about partial derivatives is the mixed partials are equal theorem. That is,

Theorem 4.22:
Let

fxy = fyx .

The point is that this is

not always the case.

An extra hypothesis is necessary.

Theorem on mixed partials

f :S R

be such that both second order partials derivatives

fxy

and

fyx

exist at a point

S, and assume in addition at every point in a disk Br (a, b) around (a, b) and fxy (a, b) = fyx (a, b) .
of the interior of

(a, b)

that one of these second order partials exists that it is continuous at the point

(a, b) .

Then

Proof:
if then

Suppose that it is

fyx that is continuous at (a, b) . Let > 0 | (c, d) (a, b) | < 1 then |fyx (c, d) fyx (a, b) | < . Next,

be given, and let choose a

1 > 0

be such that

such that if

0 < |k | < 2 ,

|fxy (a, b)
and x such a

k.

We may also

fx (a, b + k ) fx (a, b) | < , k assume that |k | < 1 /2. Finally, choose f (a + h, b + k ) f (a, b + k ) | < |k |, h

(4.145) a

3 > 0

such that if

0 < |h| < 3 ,

then

|fx (a, b + k )

(4.146)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

113

and

|fx (a, b)
and x such an

h.

Again, we may also

f (a + h, b) f (a, b) | < |k |, h assume that |h| < 1 /2.

(4.147)

In the following calculation we will use the Mean Value Theorem twice.

|fxy (a, b) fyx (a, b) |

fx (a,b+k)fx (a,b) | k fx (a,b+k)fx (a,b) +| fyx (a, b) | k f (a+h,b+k)f (a,b+k) fx (a,b+k) h +| | k f (a+h,b)f (a,b) f ( a,b ) x h +| | k f (a+h,b+k)f (a,b+k)+(f (a+h,b)f (a,b)) fyx (a, b) | +| hk f (a+h,b+k)f (a,b+k)+(f (a+h,b)f (a,b)) +| fyx (a, b) | hk fy (a+h,b' )fy (a,b' ) fyx (a, b) | 3 + | h

|fxy (a, b)

(4.148)

< = = <
because

3 + |fyx a' , b' fyx (a, b) | 4,

b'

1 .

Hence,

b and b + k, and a' is between a and a + h, so that | a' , b' (a, b) | < 1 / 2 < |fxy (a, b) fyx (a, b) < 4, for an arbitrary , and so the theorem is proved.
is between

Exercise 4.11.2
Let

be dened on

R2

by

f (0, 0) = 0

and, for

(x, y ) = (0, 0) ,f (x, y ) = x3 y/ x2 + y 2 .

a. Prove that both partial derivatives b. Show that c. Show that

fx and fy exist at each point in the plane. fyx (0, 0) = 1 and fxy (0, 0) = 0. fxy exists at each point in the plane, but that it is not continuous at (0, 0) .

The following exercise is an obvious generalization of the First Derivative Test for Extreme Values, Theorem 4.8, First Derivative Test for Extreme Values, p. 92, to real-valued functions of two real variables.

Exercise 4.11.3
Let

f : S R be a real-valued function of two real variables, and let c = (a, b) S 0 be a point at which f attains a local maximum or a local minimum. Show that if either of the partial derivatives tialf /tialx or tialf /tialy exists at c, then it must be equal to 0. HINT: Just consider real-valued functions of a real variable like x f (x, b) or y f (a, y ) ,

and use . Whenever we make a new denition about functions, the question arises of how the denition ts with algebraic combinations of functions and how it ts with the operation of composition. In that light, the next theorem is an expected one.

Theorem 4.23:

that (a, b) is a point in the interior of S, and f : S R is a real-valued function that is dierentiable, as a function of two real variables, at the point (a, b) . Suppose that T is a subset of R, that c belongs to the interior of T, and that : T R2 is dierentiable at the point c and (c) = (a, b) . Write (t) = (x (t) , y (t)) . Then the composition f is dierentiable at c and (Chain Rule again) Suppose is a subset of that

R2 ,

f ' (c) =

tialf tialf tialf tialf (a, b) x' (c) + (a, b) y ' (c) = ( (c)) x' (c) + ( (c)) y ' (c) . tialx tialy tialx tialy
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

(4.149)

114

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

Proof:
From the denition of dierentiability of a real-valued function of two real variables, write

f (a + h1 , b + h2 ) f (a, b) = L1 h1 + L2 h2 + f (H1 , h2 ) .
and from part (3) of Theorem 4.2, p. 84, write

(4.150)

(c + h) (c) = ' (c) h + (h) ,

or, in component form,

(4.151)

x (c + h) x (c) = x (c + h) a = x' (c) h + x (h)

and

(4.152)

y (c + h) y (c) = y (c + h) b = y ' (c) h + y (h) .

We also have that

(4.153)

f ((h1 , h2 )) = 0, |(h1 ,h2 )|0 | (h1 , h2 ) | lim

h0
and

(4.154)

lim

x (h) = 0, h

(4.155)

h0
We will show that

lim

y (h) = 0. h

(4.156)

f is dierentiable at c by showing that there exists a number L and a function

satisfying the two conditions of part (3) of Theorem 4.2, p. 84. Dene

k1 (h) , k2 (h) = (c + h) (c) = (x (c + h) x (c) , y (c + h) y (c)) .

Thus, we have that

(4.157)

f (c + h) f (c)

= = = = = = =

f ( (c + h)) f ( (c)) f (x (c + h) , y (c + h)) f (x (c) , y (c)) f (a + k1 (h) , b + k2 (h)) f (a, b) L1 k1 (h) + L2 k2 (h) + f (k1 (h) , k2 (h)) l1 (x (c + h) x (c)) + L2 (y (c + h) y (c)) + f (k1 (h) , k2 (h)) L1 x' (c) h + x (h) + L2 y ' (c) h + y (h) + f (k1 (h) , k2 (h)) L1 x' (c) + L2 y ' (c) h + L1 x (h) + L2 y (h) + f (k1 (h) , k2 (h)) .
(4.158)

We dene

L = L1 x' (c) + L2 y ' (c)

and

(h) = l1 x (h) + L2 y (h) + f (k1 (h) , k2 (h)) .

By these

denitions and the calculation above we have Equation (4.1)

f (c + h) f (c) = Lh + (h) ,
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

(4.159)

115

so that it only remains to verify (4.14) for the function parts of

We have seen above that the rst two

satisfy the desired limit condition, so that it is just the third part of

that requires

some proof. The required argument is analogous to the last part of the proof of the Chain Rule (Theorem 4.7, Chain Rule, p. 89), and we leave it as an exercise.

Exercise 4.11.4
a. Finish the proof to the preceding theorem by showing that

f (k1 (h) , k2 (h)) = 0. h0 h lim

HINT: Review the corresponding part of the proof to Theorem 4.7, Chain Rule, p. 89. b. Suppose

(4.160)

f : S R

is as in the preceding theorem and that

is a real-valued function of

a real variable. Suppose

is dierentiable, as a function of two real variables, at the point

(a, b) ,
c.

the partial derivatives of the real-valued function

c = f (a, b) . Let g = f. Find a formula for g of two real variables. (A generalized Mean Value Theorem) Suppose u is a real-valued function of two real variables, both of whose partial derivatives exist at each point in a disk Br (a, b) . Show that, for any
and that is dierentiable at the point two points joining

(x, y )
to

and

x' , y '

in

Br (a, b) ,

there exists a point

^ ^

x, y

on the line segment

(x, y )

x' , y '

such that

u (x, y ) u x' , y ' =

HINT: Let

tialu tialx

^ ^

x, y

x x' +

tialu tialy

^ ^

x, y

y y' .

(4.161)

: [0, 1] R2

be dened by

(t) = (1 t) x' , y ' + t (x, y ) .

Now use the

preceding theorem. d. Verify that the assignment

f tialf /tialx

is linear; i.e., that

tialf tialg tial (f + g ) = + . tialx tialx tialx

Check that the same is true for partial derivatives with respect to

(4.162)

y.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

116

CHAPTER 4. DIFFERENTIATION, LOCAL BEHAVIOR

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

Chapter 5

Integration, Average Behavior

5.1 Integration, Average Behavior A= r^2
In this chapter we will derive the formula

A = r2

for the area of a circle of radius

r.

As a matter of fact,

we will rst have to settle on exactly what is the denition of the area of a region in the plane, and more subtle than that, we must decide what kinds of regions in the plane have areas. Before we consider the problem of area, we will develop the notion of the integral (or average value) of a function dened on an interval

[a, b] ,

which notion we will use later to compute areas, once they have been dened.

The main results of this chapter include: 1. The denition of function, 2. The 3. The 4. The 6.

integrability

of a function, and the denition of the

integral

of an integrable 131),

Fundamental Theorem of Calculus (Theorem 5.9, Fundamental Theorem of Calculus, p. General Binomial Theorem (Theorem 5.13, General Binomial Theorem, p. area of a geometric set, Integral Test (Theorem 5.17, p.
(Theorem 5.15, p. 139), and 141).

Integral Form of Taylor's Remainder Theorem (Theorem 5.12, Integral Form of Taylor's
134),

Remainder Theorem, p. 134), 5. The denition of the

A = r2

7. The

5.2 Integrals of Step Functions

We begin by dening the integral of certain (but not all) bounded, real-valued functions whose domains are closed bounded intervals. Later, we will extend the denition of integral to certain kinds of unbounded complex-valued functions whose domains are still intervals, but which need not be either closed or bounded. First, we recall from Section 3.1 the following denitions. Let [a, b] be a closed bounded interval of real numbers. By a partition of [a, b] we mean a nite set P = {x0 < x1 < ... < xn } of n + 1 points, where x0 = a and xn = b. The n intervals {[xi1 , xi ]} are called the closed subintervals of the partition P, and the n intervals {(xi1 , xi )} are called the open subintervals or elements of P. We write P for the maximum of the numbers (lengths of the subintervals) {xi xi1 }, and call P the mesh size of the partition P.

Denition 5.1:

1 This 2 This

content is available online at <http://cnx.org/content/m36207/1.2/>. content is available online at <http://cnx.org/content/m36208/1.2/>. Available for free at Connexions <http://cnx.org/content/col11249/1.1>

117

118

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

If a partition then we say that Let

P = {xi } is contained Q is ner than P.

in another partition

Q = {yj },

i.e., each

xi

equals some

yj ,

be a function on an interval

[a, b] ,

and let

P = {x0 < ... < xn }

be a partition of

[a, b] .

Physicists often consider sums of the form

SP,{yi } =
i=1
where

f (yi ) (xi xi1 ) ,

(5.1)

yi

is a point in the subinterval

(xi1 , xi ) .

These sums (called Riemann sums) are approxi-

mations of physical quantities, and the limit of these sums, as the mesh of the partition becomes smaller and smaller, should represent a precise value of the physical quantity. What precisely is meant by the  limit of such sums is already a subtle question, but even having decided on what that denition should be, it is as important and dicult to determine whether or not such a limit exists for many (or even any) functions

f. We approach this question from a slightly dierent point

of view, but we will revisit Riemann sums in the end. Again we recall from Section 3.1 the following.

Denition 5.2:
function
Let

[a, b]

be a closed bounded interval in

if there exists a partition

there exists a number

5.1: REMARK

ai

such

R. A real-valued function h : [a, b] R P = {x0 < x1 < ... < xn } of [a, b] such that for that h (x) = ai for all x (xi1 , xi ) . h

is called a each

step

1in

A step function

is constant on the open subintervals (or elements) of a certain Indeed, if

partition.

Of course, the partition is not unique.

is such a partition, we may add

more points to it, making a larger partition having more subintervals, and the function various distinct partitions.

will still

be constant on these new open subintervals. That is, a given step function can be described using Also, the values of a step function at the partition points themselves is irrelevant. require that it be constant on the open subintervals. We only

Exercise 5.2.1
Let that

h be a step function on [a, b] , and let P = {x0 < x1 < ... < xn } h (x) = ai on the subinterval (xi1 , xi ) determined by P. h
range?

be a partition of

[a, b]

such

a. Prove that the range of b. Prove that of c. d. e. f.

is a nite set. What is an upper bound on the cardinality of this

h is dierentiable at all but a nite number of points in [a, b] . What is the value h' at such a point? ' Let f be a function on [a, b] . Prove that f is a step function if and only if f (x) exists and = 0 for every x (a, b) except possibly for a nite number of points. What can be said about the values of h at the endpoints {xi } of the subintervals of P ? (e) Let h be a step function on [a, b] , and let j be a function on [a, b] for which h (x) = j (x) for all x [a, b] except for one point c. Show that j is also a step function. If k is a function on [a, b] that agrees with a step function h except at a nite number of points c1 , c2 , ..., cN , show that k is also a step function.

Exercise 5.2.2
Let

[a, b]

be a xed closed bounded interval in

R,

and let

H ([a, b])

denote the set of all step

functions on

[a, b] . H ([a, b])

is a vector space of functions;

a. Using Part (c) of Exercise 5.2.1, prove that the set

i.e., it is closed under addition and scalar multiplication.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

119

b. Show that

H ([a, b]) H ([a, b])

is closed under multiplication; i.e., if

h1 , h2 H ([a, b]) ,

then

h1 h2

H ([a, b]) .
c. Show that is closed under taking maximum and minimum and that it contains all the real-valued constant functions. d. We call a function

an indicator function if it equals 1 on an interval (c, d) and is 0 outside [c, d] . To be precise, we will denote this indicator function by (c,d) . Prove that every indicator function is a step function, and show also that every step function h is a linear combination
of indicator functions:

h=
j =1
e. Dene a function

aj (cj ,dj ) .
if

(5.2)

on

[0, 1]

by setting

k ( x) = 0

is an irrational number. Prove that the range of

x is a rational number and k (x) = 1 if k is a nite set, but that k is not a step

function. Our rst theorem in this chapter is a fundamental consistency result about the area under the graph of a step function. Of course, the graph of a step function looks like a collection of horizontal line segments, and the region under this graph is just a collection of rectangles. Actually, in this remark, we are implicitly thinking that the values

{ai }

of the step function are positive. If some of these values are negative, then we

must re-think what we mean by the area under the graph. We rst introduce the following bit of notation.

Denition 5.3:
Let

be a step function on the closed interval dened by

[a, b] .

partition of

[a, b] such that h (x) = ai hrelative to P to be the number SP (h)

on the interval

Suppose P = {x0 < x1 < ... < xn } is a (xi1 , xi ) . Dene the weighted average of

SP (h) =
i=1

ai (xi xi1 ) .

(5.3)

5.2: REMARK Notice the similarity between the formula for a weighted average and the formula for
a Riemann sum. Note also that if the interval is a single point, i.e.,

a = b,

then the only partition

of the interval consists of the single point

x0 = a,

and every weighted average

SP (h) = 0.
It is simply the

The next theorem is not a surprise, although its proof takes some careful thinking. assertion that the weighted averages are independent of the choice of partition.

of

h be a step function on the closed interval [a, b] . Suppose P = {x0 < x1 < ... < xn } is a partition [a, b] such that h (x) = ai on the interval (xi1 , xi ) , and suppose Q = {y0 < y1 < ... < ym } is another partition of [a, b] such that h (x) = bj on the interval (yj 1 , yj ) . Then the weighted average of h relative to P is the same as the weighted average of h relative to Q. That is, SP (h) = SQ (h) .
Let

Theorem 5.1:

Proof:
Then

Suppose rst that the partition

m = n + 1,

and there exists an

Q is obtained from the partition P by adding one additional point. i0 between 1 and n 1 such that

1. for 2. 3.

0 i i0 we have yi = xi . xi0 < yi0 +1 < xi0 +1 . For i0 < i n we have xi = yi+1 . yi0 +1
is the only point of

In other words,

that is not a point of

P,

and

yi0 +1

lies strictly between

x i0

xi0 +1 . Because h is
and

constant on the interval

(xi0 , xi0 +1 ) = (yi0 , yi0 +2 ) ,

it follows that

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

120

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

1. For 2. 3. So,

1 i i0 ,ai = bi . bi0 +1 = bi0 +2 = ai0 +1 . For i0 + 1 i n,ai = bi+1 .

SP (h)

= = = = = = =
i0 i=1 bi i0 i=1

n i=1

ai (xi xi1 ) ai (xi xi1 ) (yi+1 yi )

(5.4)

ai (xi xi1 ) + ai0 +1 (xi0 +1 xi0 ) +

n i=i0 +2

i0 i=1 bi

(yi yi1 ) + ai0 +1 (yi0 +2 yi0 ) + +

n i=i0 +2 bi+1

(yi yi1 ) + ai0 +1 (yi0 +2 yi0 +1 + yi0 +1 yi0 )

n+1 i=i0 +3 bi

(yi yi1 ) (yi yi1 )

i0 i=1 bi

(yi yi1 ) + bi0 +1 (yi0 +1 yi0 ) + bi0 +2 (yi0 +2 yi0 +1 ) +

m i=i0 +3 bi m i=1 bi

(yi yi1 )

SQ (h) , Q is obtained from P P

by adding just one more

which proves the theorem in this special case where point. It follows easily now by induction that if of additional points, then

is obtained from

by adding any nite number

Q, and SQ (h) = SP (h) . Finally, let Q = {y0 < y1 < ... < ym } be an arbitrary partition of [a, b] , for which h is constant on each of the open subintervals (yj 1 , yj ) determined by Q. Dene R to be the partition of [a, b] obtained by taking the union of the partition points {xi } and {yj }. Then R is a partition of [a, b] that is obtained by adding a nite number of points to the partition P, whence SR (h) = SP (h) . Likewise, R is obtained from the partition Q by adding a nite number of points, whence SR (h) = SQ (h) , and this proves that SQ (h) = SP (h) , as desired.
is constant on each of the open subintervals determined by

Denition 5.4:

Let [a, b] be a xed closed bounded interval in R. We dene the integral of a step function h on [a, b] , and denote it by h, as follows: If P = {x0 < x1 < ... < xn } is a partition of [a, b] , for which h (x) = ai for all x (xi1 , xi ) , then

h = SP (h) =
i=1

ai (xi xi1 ) .

(5.5)

5.3: REMARK The integral of a step function h is dened to be the weighted average of h relative
to a partition of

of

[a, b] .

Notice that the preceding theorem is crucial in order that this denition The integral of a step function should

be unambiguously dened.

not

depend on which

partition is used. Theorem 5.1, p. 119 asserts precisely this fact. Note also that if the interval is a single point, i.e.,

a = b, then the integral of every step function

is 0.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

121

We use a variety of notations for the integral of

h:
b

h=
a

h=
a

h (t) dt.

(5.6)

The following exercise provides a very useful way of describing the integral of a step function. Not only does it show that the integral of a step function looks like a Riemann sum, but it provides a description of the integral that makes certain calculations easier. See, for example, the proof of the next theorem.

Exercise 5.2.3
Suppose which

is a step function on

[a, b]

and that

is constant on each subinterval

R = {z0 < z1 < ... < zn } (zi1 , zi ) of R.

n

is a partition of

[a, b]

for

a. Prove that

h = SR (h) =
i=1
where, for each b. Show that

h (wi ) (zi zi1 ) , (zi1 , zi ) . h

(5.7)

1 i n,wi

is any point in

(Note then that the integral of a step

function takes the form of a Riemann sum.)

is independent of the values of

at the points

{zi }

of the partition

R.

Exercise 5.2.4
Let

h1

and

h2

be two step functions on

[a, b] .

a. Suppose that HINT: Let

h1 (x) = h2 (x)

for all is

be a partition of

subintervals, and for which calculate the two integrals. b. Suppose

x [a, b] except for one point c. Prove that h1 = h2 . [a, b] , for which both h1 and h2 are constant on its open one of the points of P. Now use the preceding exercise to c1 , ..., cN [a, b] .
Prove that

h1 =

h1 (x) = h2 (x) h2 .

for all but a nite number of points

We have used the terminology weighted average of a step function relative to a partition

P.

The next

exercise shows how the integral of a step function can be related to an actual average value of the function.

Exercise 5.2.5
Let

be a step function on the closed interval

partition of

[a, b]

for which and

as an interval of time, time between

x i 1

[a, b] , and let P = {x0 < x1 < ... < xn } be a (xi1 , xi ) . Let us think of the interval [a, b] and suppose that the function h assumes the value ai for the interval of xi . Show that the average value A (h) taken on by h throughout the entire h (x) = ai
on the interval

interval ([a, b]) of time is given by

A (h) =

h . ba

(5.8)

Theorem 5.2:
Let assignment 1. 2. 3.

H ([a, b]) denote the vector space of all step functions on the closed h h of H ([a, b]) into R has the following properties:
(Linearity) all

interval

[a, b] .

Then the

c h for If h =

H ([a, b]) is a vector space. Furthermore, (h1 + h2 ) = h1 + h2 , and ch = h1 , h2 , h H ([a, b]) , and for all real numbers c. n i=1 ai (ci ,di ) is a linear combination of indicator functions (See part (d) of Exern cise 5.2.2), then h = i=1 ai (di ci ) . (Positivity) If h (x) 0 for all x [a, b] , then h 0.
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

122

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

4. (Order-preserving) If then

h1

and

h2

are step functions for which

h1 (x) h2 (x)

for all

x [a, b] ,

h1

h2 .

Proof:

That H ([a, b]) is a vector space was proved in part (a) of Exercise 5.2.2. Suppose P = {x0 < x1 < ... < xn } is a partition of [a, b] such that h1 (x) is constant for all x (xi1 , xi ) , and suppose Q = {y0 < y1 < ... < ym } is a partition of [a, b] such that h2 (x) is constant for all x (yj 1 , yj ) . Let R = {z0 < z1 < ... < zr } be the partition of [a, b] obtained by taking the union of the xi 's and the yj 's. Then h1 and h2 are both constant on each open subinterval of R, since each such subinterval is contained in some open subinterval of P and also is contained in some open subinterval of Q. Therefore, h1 + h2 is constant on each open subinterval of R. Now, using Exercise 5.2.3, we have that

(h1 + h2 )

= = =
r k=1

r k=1

((h1 + h2 ) (wk )) (zk zk1 )

r k=1

h1 (wk ) (zk zk1 ) + h1 +

h2 (wk ) (zk zk1 )

(5.9)

h2 .

This proves the rst assertion of part (1). Next, let open subinterval of

P = {x0 < x1 < ... < xn } be a partition of [a, b] such that h (x) is constant on each P. Then ch (x) is constant on each open subinterval of P, and using Exercise 5.2.3
n i=1 ch (wi ) (xi xi1 ) n i=1 h (wi ) (xi xi1 )

again, we have that

(ch)

= = c =

(5.10)

c h,

which completes the proof of the other half of part (1).

(ci ,di ) = di ci , for then part (2) will follow from (ci ,di ) is just a step function determined by the four point partition {a, ci , di , b} and values 0 on (a, ci ) and (di , b) and 1 on (ci , di ) . Therefore, we have that (ci ,di ) = di ci . If h (x) 0 for all x, and P = {x0 < x1 < ... < xn } is as above, then
To see part (2), we need only verify that part (1). But

h=
i=1
and this proves part (3). Finally, suppose

h (wi ) (xi xi1 ) 0,

(5.11)

h3 = h2 h1

is a step function on

h1 (x) h2 (x) for all x [a, b] . By Exercise 5.2.2, we know that the function [a, b] . Also, h3 (x) 0 for all x [a, b] . So, by part (3), h3 0.

Then, by part (1),

0
which implies that

h3 =
as desired.

(h2 h1 ) =

h2

h1 ,

(5.12)

h1

h2 ,

Exercise 5.2.6
a. Let b. Let

c.

h be the constant function c on [a, b] . Show that h = c (b a) . a < c < d < b be real numbers, and let h be the step function on [a, b] that equals r for b c < x < d and 0 otherwise. Prove that a h (t) dt = r (d c) . Let h be a step function on [a, b] . Prove that |h| is a step function, and that | h| |h|. HINT: Note that |h| (x) h (x) |h| (x) . Now use the preceding theorem.
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

123

d. Suppose

h is a step function on [a, b] and x [a, b] . Prove that | h| c (b a) .

that

is a constant for which

|h (x) | c

for all

5.3 Integrable Functions

those functions that are

We now wish to extend the denition of the integral to a wider class of functions. This class will consist of

uniform limits of step functions.

The requirement that these limits be uniform is

crucial. Pointwise limits of step functions doesn't work, as we will see in Exercise 5.3.1 below. The initial step in carrying out this generalization is the following.

Theorem 5.3:

Let

uniformly to a function numbers.

[a, b] be a closed bounded interval, and let {hn } be a sequence of step functions that converges f on [a, b] . Then the sequence { hn } is a convergent sequence of real

Proof:
for any

We will show that

nN

and any

{ hn } is a Cauchy sequence x [a, b] , we have

in

R.

Thus, given an

> 0,

choose an

such that

|f (x) hn (x) | <

Then, for any

. 2 (b a)
we have

(5.13)

and

both

and any

x [a, b] ,

|hn (x) hm (x) | |hn (x) f (x) | + |f (x) hm (x) | <

Therefore,

. ba

(5.14)

|
as desired.

hn

hm | = |

(hn hm ) |

|hn hm |

= , ba

(5.15)

The preceding theorem provides us with a perfectly good idea of how to dene the integral of a function

that is the uniform limit of a sequence of step functions. However, we rst need to establish another kind

of consistency result.

Theorem 5.4:
If

{hn }

and

same function

{kn } are f, then

two sequences of step functions on

[a, b] ,

each converging uniformly to the

lim

hn = lim

kn .

(5.16)

Proof:
Given all

such that

> 0, choose N so that if n N, then |hn (x) f (x) | < / (2 (b a)) for all x [a, b] , and |f (x) kn (x) | < / (2 (b a)) for all x [a, b] . Then, |hn (x) kn (x) | < / (b a) for x [a, b] if n N. So, | hn kn | |hn kn | = ba
(5.17)

3 This

content is available online at <http://cnx.org/content/m36209/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

124

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

if

n N.

Taking limits gives

|lim
Since this is true for arbitrary

hn lim

kn | . lim hn = lim kn ,
as desired.

(5.18)

> 0,

it follows that

Denition 5.5:
Let on

[a, b] be a closed bounded interval of real numbers. A function f : [a, b] R is called integrable [a, b] if it is the uniform limit of a sequence {hn } of step functions. Let I ([a, b]) denote the set of all functions that are integrable on [a, b] . If f I ([a, b]) , dene the integral of f, denoted f, by f = lim
where

hn , f
on

(5.19)

{hn }

is some (any) sequence of step functions that converges uniformly to

[a, b] .

As in the case of step functions, we use the following notations:

f=
a

f=
a

f (t) dt.

(5.20)

5.4: REMARK Note that Theorem 5.4, p.

123 is crucial in order that this denition be unambiguous.

Indeed, we will see below that this critical consistency result is one place where uniform limits of

a a f = 0, because a h = 0 for any step function. In a fact, we will derive almost everything about the integral of a general integrable function from the
also that it follows from this denition that corresponding results about the integral of a step function. mathematical analysis, approximation. No surprise. This is the essence of

step functions works while pointwise limits do not. See parts (c) and (d) of Exercise 5.3.1. Note

Exercise 5.3.1
if

Dene a function

on the closed interval

[0, 1] by f (x) = 1 if x is a rational number and f (x) = 0

is an irrational number.

a. Suppose

h is a step function on [0, 1] . Prove that there must exist an x [0, 1] such that |f (x) h (x) | 1/2. HINT: Let (xi1 , xi ) be an interval on which h is a constant c. Now use
the fact that there are both rationals and irrationals in this interval.

b. Prove that

is not the uniform limit of a sequence of step functions. That is,

is

not an

integrable function. c. Consider the two sequences by

d.

{hn } and {kn } of step functions dened on the interval [0, 1] kn = n(0,1/n) . Show that both sequences {hn } and {kn } converge pointwise to the 0 function on [0, 1] . HINT: All functions are 0 at x = 0. For x > 0, choose N so that 1/N < x. Then, for any n N,hn (x) = kn (x) = 0. Let hn and kn be as in part (c). Show that lim hn = 0, but lim kn = 1. Conclude that the hn = (0,1/n) ,
and consistency result in Theorem 5.4, p. 123 does not hold for pointwise limits of step functions.

Exercise 5.3.2
Dene a function

on the closed interval

[0, 1]

by

f (x) = x.

a. For each positive integer

n, let Pn be the partition of [0, 1] given by the points {0 < 1/n < 2/n < 3/n < ... < (n 1) /n < 1}. Dene a step function hn on [0, 1] by setting hn (x) = i/n i i1 if n < x < n , and hn (i/n) = i/n for all 0 i n. Prove that |f (x) hn (x) | < 1/n for all x [0, 1] , and then conclude that f is the uniform limit of the hn 's whence f I ([0, 1]) .
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

125

b. Show that

hn =
i=1
c. Show that

n (n + 1) i = . 2 n 2 n2

(5.21)

1 f (t) dt = 1/2. The next exercise establishes some additional properties of 0 integrable functions on an interval [a, b] .

Exercise 5.3.3
Let

[a, b]

be a closed and bounded interval, and let

be an element of

I ([a, b]) . |f (x) h (x) | <

a. Show that, for each for all

>0

there exists a step function

on

[a, b]

such that

x [a, b] .

b. For each positive integer

c. d.

n let hn be a step function satisfying the conclusion of part (a) for = 1/n. Dene kn = hn 1/n and ln = hn + 1/n. Show that kn and ln are step functions, that kn (x) < f (x) < ln (x) for all x [a, b] , and that |ln (x) kn (x) | = ln (x) kn (x) = 2/n b 2 (b a) . for all x. Hence, (l kn ) = n a n Conclude from part (b) that, given any > 0, there exist step functions k and l such that k (x) f (x) l (x) for which (l (x) k (x)) < . Prove that there exists a sequence {jn } of step functions on [a, b] , for which jn (x) jn+1 (x) ' f (x) for all x, that converges uniformly to f. Show also that there exists a sequence {jn } of ' ' step functions on [a, b] , for which jn (x) jn+1 (x) f (x) for all x, that converges uniformly to f. That is, if f I ([a, b]) , then f is the uniform limit of a nondecreasing sequence of step
functions and also is the uniform limit of a nonincreasing sequence of step functions. HINT: To construct the

jn 's

and

' 's, jn

use the step functions

kn

and

ln

of part (b), and recall that

the maximum and minimum of step functions is again a step function. e. Show that if HINT: Write

f (x) 0 for all x [a, b] , and g is dened by g (x) = f (x), then g I ([a, b]) . f = limhn where hn (x) 0 for all x and n. Then use part (g) of Exercise 3.28. > 0,
there exists a partition

f. (Riemann sums again.) Show that, given an

such that if

Q = {x0 < x1 < ... < xn } wi (xi1 , xi ) , then |

is any partition ner than

P,

and

{wi }

are any points for which

f (t) dt
a i=1

f (wi ) (xi xi1 ) | < . k

and

(5.22)

HINT: Let and hence

be a partition for which both the step functions

on the open subintervals of

P.

Verify that for any ner partition

l of part (c) are constant Q,l (wi ) f (wi ) k (wi ) ,

(5.23)

l (wi ) (xi xi1 )

i i

f (wi ) (xi xi1 )

i

k (wi ) (xi xi1 ) .

Denition 5.6:
if, given any

f on a closed bounded interval [a, b] is called Riemann-integrable > 0, there exist step functions k and l, on [a, b] for which k (x) f (x) l (x) for all x, such that (l k ) < . We denote the set of all functions on [a, b] that are Riemann-integrable by IR ([a, b]) .
A bounded real-valued function

5.5: REMARK The notion of Riemann-integrability was introduced by Riemann in the mid nineteenth

century and was the rst formal denition of integrability. Since then several other denitions have been given for an integral, culminating in the theory of Lebesgue integration. The denition of integrability that we are using in this book is slightly dierent and less general from that of Riemann, and both of these are very dierent and less general from the denition given by Lebesgue in the

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

126

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

early twentieth century. Part (c) of Exercise 5.3.3 above shows that the functions we are calling integrable are necessarily Riemann-integrable. We will see in Exercise 5.3.4 that there are Riemannintegrable functions that are integrable function

not integrable in our sense.

In both cases, Riemann's and ours, an

f must be trapped between two step functions k and l. In our denition, we must have l (x) k (x) < for all x [a, b] , while in Riemann's denition, we only need that l k < . The distinction is that a small step function must have a small integral, but it isn't
necessary for a step function to be (uniformly) small in order for it to have a small integral. It only has to be small on most of the interval

[a, b] . [a, b]
include among the integrable

On the other hand, all the denitions of integrability on

functions the continuous ones. And, all the dierent denitions of integral give the same value to a continuous function. The dierences then in these denitions shows up at the point of saying exactly which functions are integrable. Perhaps the most enlightening thing to say in this connection is that it is impossible to make a good denition of integrability in such a way that every function is integrable. Subtle points in set theory arise in such attempts, and many fascinating and deep mathematical ideas have come from them. However, we will stick with our denition, since it is simpler than Riemann's and is completely sucient for our purposes.

Theorem 5.5:
Let

[a, b]

be a xed closed and bounded interval, and let

I ([a, b])

denote the set of integrable

functions on

[a, b] .

Then:

1. Every element of 2. 3.

I ([a, b])

is a bounded function. That is, integrable functions are necessarily

bounded functions.

I ([a, b]) I ([a, b]) f

is a vector space of functions. is closed under multiplication; i.e., if

and

g I ([a, b]) ,
is in

then

f g I ([a, b]) .

4. Every step function is in 5. If

I ([a, b]) . [a, b] ,

then

is a continuous real-valued function on

uous real-valued function on

[a, b]

is integrable on

f [a, b] .

I ([a, b]) .

That is, every contin-

Proof:
Let

uniformly to

f I ([a, b]) , and write f = limhn , where {hn } is a sequence of step functions that converges f. Given the positive number = 1, choose N so that |f (x) hN (x) | < 1 for all x [a, b] . Then |f (x) | |hN (x) | + 1 for all x [a, b] . Because hN is a step function, its range is a nite set, so that there exists a number M for which |hN (x) | M for all x [a, b] . Hence, |f (x) | M + 1 for all x [a, b] , and this proves part (1). Next, let f and g be integrable, and write f = limhn and g = limkn , where {hn } and {kn } are sequences of step functions that converge uniformly to f and g respectively. If s and t are real numbers, then the sequence {shn + tkn } converges uniformly to the function sf + tg. See parts (c) and (d) of Exercise 3.10.1. Therefore, sf + tg I ([a, b]) , and I ([a, b]) is a vector space, proving
part (2). Note that part (3) does not follow immediately from Exercise 3.10.1; the product of uniformly

f = limhn and g = limkn be elements of I ([a, b]) . By part (1), both f and g are bounded, and we write Mf and Mg for numbers that satisfy |f (x) | Mf and |g (x) | Mg for all x [a, b] . Because the sequence {kn } converges uniformly to g, there exists an N such that if n N we have |g (x) kn (x) | < 1 for all x [a, b] . This implies that, if n N, then |kn (x) | Mg + 1 for all x [a, b] . Now we show that f g is the uniform limit of the sequence hn kn . For, if n N, then
convergent sequences may not be uniformly convergent. To see it for this case, let

|f (x) g (x) hn (x) kn (x) | = |f (x) g (x) f (x) kn (x) + f (x) kn (x) hn (x) kn (x) |
which implies that

|f (x) ||g (x) kn (x) | + |kn (x) ||f (x) hn (x) | Mf |g (x) kn (x) | + (Mg + 1) |f (x) hn (x) |,

(5.24)

f g = lim (hn kn ) .

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

127

If

is itself a step function, then it is obviously the uniform limit of the constant sequence

{h},

which implies that Finally, if

is integrable.

is continuous on

[a, b] ,

it follows from Theorem 3.21, p. 78 that

is the uniform

limit of a sequence of step functions, whence

f I ([a, b]) .

Exercise 5.3.4
Let

be the function dened on

[0, 1]

by

f (x) = sin (1/x)

if

x=0

and

f (0) = 0.

a. Show that b.

c.

x and discontinuous at 0. HINT: Observe that, sin (1/x) attains both the values 1 and 1. Show that f is not integrable on [0, 1] . HINT: Suppose f = limhn . Choose N so that |f (x) hN (x) | < 1/2 for all x [0, 1] . Let P be a partition for which hN is constant on its open subintervals, and examine the situation for x's in the interval (x0 , x1 ) . Show that f is Riemann-integrable on [0, 1] . Conclude that I ([a, b]) is a proper subset of IR ([a, b]) .
is continuous at every nonzero on any interval

(0, ) ,

the function

Exercise 5.3.5
a. Let all

b.

c.

f be an integrable function on [a, b] . Suppose g is a function for which g (x) = f (x) for x [a, b] except for one point c. Prove that g is integrable and that g = f. HINT: If f = limhn , dene kn (x) = hn (x) for all x = c and kn (c) = g (c) . Then use Exercise 5.2.4. Again, let f be an integrable function on [a, b] . Suppose g is a function for which g (x) = f (x) for all but a nite number of points c1 , ..., cN [a, b] . Prove that g I ([a, b]) , and that g = f. Suppose f is a function on the closed interval [a, b] , that is uniformly continuous on the open interval (a, b) . Prove that f is integrable on [a, b] . HINT: Just reproduce the proof to
Theorem 3.21, p. 78.

5.6: REMARK

In view of part (b) of the preceding exercise, we see that whether a function

is

integrable or not is totally independent of the values of the function at a xed nite set of points. Indeed, the function needn't even be dened at a xed nite set of points, and still it can be integrable. This observation is helpful in many instances, e.g., in parts (d) and (e) of Exercise 5.7.1.

Theorem 5.6:
1. 2. 3. 4. 5. 6.

The assignment

on

I ([a, b])

satises the following properties.

(Linearity) I ([a, b]) is a vector space, and (f + g ) = f + g for all f, g I ([a, b])and , R. (Positivity) If f (x) 0 for all x [a, b] , then f 0. (Order-preserving) If f, g I ([a, b]) and f (x) g (x) for all x [a, b] , then f g. If f I ([a, b]) , then so is |f |, and | f | |f |. If f is the uniform limit of functions fn , each of which is in I ([a, b]) , then f I ([a, b]) and f = lim fn . Let {un } be a sequence of functions in I ([a, b]) . Suppose that for each n there is a number mn , for which |un (x) | mn for all x [a, b] , and such that the innite series mn converges. Then the innite series un converges uniformly to an integrable function, and un = un .

Proof:
That I ([a, b]) is a vector space was proved in part (2) of Theorem 5.5, p. 126. Let f and g I ([a, b]) , and write f = limhn and g = limkn , where the hn 's and the kn 's are step functions. be in Then

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

128

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

f + g = lim (hn + kn ) ,
have so that, by Theorem 5.2, p. 121 and the denition of the integral, we

(f + g )

= = =

lim (hn + kn ) lim hn + f + g, kn hn + lim kn

(5.25)

= lim

which proves part (1). Next, if for all

f I ([a, b]) n.

satises

f ( x) 0

step functions that converges uniformly to

for all x [a, b] , let {ln } be a nonincreasing sequence of f. See part (d) of Exercise 5.3.3. Then ln (x) f (x) 0

and all

So, again by Theorem 5.2, p. 121, we have that

f = lim
This proves part (2).

ln 0 .

(5.26)

Part (3) now follows by combining parts (1) and (2) just as in the proof of Theorem 5.2, p. 121. To see part (4), let

f I ([a, b])

be given. Write

f = limhn .

Then

|f | = lim|hn |.

For (5.27)

||f (x) | |hn (x) || |f (x) hn (x) |.

Therefore,

|f |

is integrable. Also,

|f | = lim
To see part (5), let each

|hn | lim|

hn | = |lim

hn | = |

f |.

(5.28)

{fn }

be a sequence of elements of

n,

let

hn

be a step function on

[a, b]

such that

I ([a, b]) , and suppose that f = limfn . For |fn (x) hn (x) | < 1/n for all x [a, b] . Note ba . n

also that it follows from parts (3) and (4) that

|
Now

fn

hn | <

(5.29)

{hn }

converges uniformly to

f.

For,

|f (x) hn (x) |

<

|f (x) fn (x) | + |fn (x) hn (x) |

1 |f (x) fn (x) | + n ,
Moreover,

(5.30)

showing that

f = limhn .

Therefore,

f I ([a, b]) .

f = lim hn .

Finally,

f =

lim fn ,

for

| f

fn |

hn | + | hn hn | +
b a n .

fn |

| f

(5.31)

This completes the proof of part (5). Part (6) follows directly from part (5) and the Weierstrass M Test (Theorem 3.19, Weierstrass M-Test, p. 77). For, part (1) of that theorem implies that the innite series

un

converges

uniformly, and then

un =

un

follows from part (5) of this theorem.

As a nal extension of our notion of integral, we dene the integral of certain complex-valued functions.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

129

Denition 5.7:
integrable
Let

[a, b]

be a xed bounded and closed interval. A complex-valued function if its real and imaginary parts

f = u + iv

is called

u
b

and

are integrable. In this case, we dene

f=
a a

(u + iv ) =
a

u+i
a

v.

(5.32)

Theorem 5.7:
1. The set of all integrable complex-valued functions on complex numbers, and

[a, b]

is a vector space over the eld of

(f + g ) =
a
for all integrable complex-valued functions 2. If

f +
a a

g
complex numbers so is

(5.33)

and

is an integrable complex-valued function on

g and all [a, b] , then

|f |,

and

b a

and . b f | a |f |.

Proof:
We leave the verication of part (1) to the exercise that follows. To see part (2), suppose that

is integrable, and write

|f | is integrable by Theorem 5.5, p. z

in polar coordinates as

f = u + iv. Then |f | = u2 + v 2 , so that b 126 and part (e) of Exercise 5.3.3. Now write z = f, and write a r = |z | = |
b a

z = rei ,
i
^

where

f |.

(See Exercise 4.6.3 (Polar coordinates).) Then ^

Dene a function

by

g (x) = e

f (x)
^

and notice that

|g | = |f |. r=
^

b a

g = ei

b a

f = r,
^

which So,

is a real number. Writing

g =u +i v, |

we then have that

u +i

v,

implying that

v = 0.

b a

f| = = = = = =
b ^ u a

r
b a

g
b ^ v a
(5.34)

+i

b ^ u a b ^ | a u| b ^ |u| a b |g | a b |f |, a

as desired.

Exercise 5.3.6
Prove part (1) of the preceding theorem. HINT: Break

, ,

f,

and

into real and imaginary parts.

5.4 The Fundamental Theorem of Calculus

later proofs, so it's good to state it precisely now.

We begin this section with a result that is certainly not a surprise, but we will need it at various places in

4 This

content is available online at <http://cnx.org/content/m36210/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

130

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

Theorem 5.8:
Suppose

f I ([a, b]) ,

and suppose

a < c < b.
b

Then

f I ([a, c]) ,f I ([c, b]) ,

b

and

f=

f+
a c

f.

(5.35)

h is a step function on [a, b] , and let P = {x0 < x1 < ... < xn } be a partition h (x) = ai on the subinterval (xi1 , xi ) of P. Of course, we may assume without loss of generality that c is one of the points of P, say c = xk . Clearly h is a step function on both intervals [a, c] and [c, b] . Now, let Q1 = {a = x0 < x1 < ... < c = xk } be the partition of [a, c] obtained by intersecting P with [a, c] , and let Q2 = {c = xk < xk+1 < ... < xn = b} be the partition of [c, b] obtained by intersecting P with [c, b] . We have that
Suppose rst that of

Proof:
[a, b]

such that

b a

h = = = = =
k i=1 n i=1

SP (h) ai (xi xi1 )

n i=k+1

ai (xi xi1 ) +
c a

ai (xi xi1 )

(5.36)

SQ1 (h) + SQ2 (h) h+

b c

h, [a, b] .
Then clearly

which proves the theorem for step functions. Now, write

[a, c] ,

which shows that

f = limhn , where each hn f I ([a, c]) , and

is a step function on

f = limhn

on

f = lim
a
Similarly,

hn .
a
and

(5.37)

f = limhn

on

[c, b] ,

showing that

f I ([c, b]) ,
b

f = lim
c
Finally,

hn .
c

(5.38)

b a

= = = = lim lim

lim
c a c a

b a

hn
b c

hn +
b c

hn
b c

(5.39)

hn + lim f+ f,

hn

c a

as desired.

I's time for the trumpets again!

What we call the Fundamental Theorem of Calculus was discovered Perhaps the main theoretical consequence of It all came down to

by Newton and Leibniz more or less simultaneously in the seventeenth century, and it is without doubt the cornerstone of all we call mathematical analysis today. this theorem is that it provides a procedure for inventing new functions. Polynomials are rather natural functions, power series are a simple generalization of polynomials, and then what? thinking of a function of a variable varying point

as being the area beneath a curve between a xed point

and the

x.

By now, we have polished and massaged these ideas into a careful, detailed development

of the subject, which has substantially obscured the original ingenious insights of Newton and Leibniz. On the other hand, our development and proofs are complete, while theirs were based heavily on their intuition. So, here it is.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

131

Theorem 5.9:
Suppose 1. 3.

Fundamental Theorem of Calculus

is an arbitrary element of

I ([a, b]) . Dene a function F

on

[a, b] by F (x) =

x a

f. Then:

2. If

F (a) = 0. c (a, b) , then F is dierentiable at c and F ' (c) = f (c) . Suppose that f is continuous on [a, b] . If G is any continuous function on [a, b] that is dier' entiable on (a, b) and satises G (x) = f (x) for all x (a, b) , then
is continuous on and

[a, b] ,

is continuous at a point

f (t) dt = G (b) G (a) .

a

(5.40)

REMARK Part (2) of this theorem is the heart of it, the great discovery of Newton and Leibniz,
although most beginning calculus students often think of part (3) as the main statement. Of course it is that third part that enables us to actually compute integrals.

Proof:

Because dened.

f I ([a, b]) ,

we know that

f I ([a, x])

for every

x [a, b] ,

so that

F ( x)

at least is

Also, we know that Then, if

is bounded; i.e., there exists an

such that

|f (t) | M

for all

t [a, b] .

x, y [a, b]

with

x y,

we have that

|F (x) F (y ) | = = | = =
so that continuous on

|
y a

x a

f |

y a

f| f|
(5.41)

f+

x y f a y x f| y x |f | y x M y

M (x y ) ,

|F (x) F (y ) | M |x y | < if |x y | < = /M. This shows that F is (uniformly) a [a, b] . Obviously, F (a) = a f = 0, and part (1) is proved. Next, suppose that f is continuous at c (a, b) , and write L = f (c) . Let > 0 be given. ' To show that F is dierentiable at c and that F (c) = f (c) , we must nd a > 0 such that if 0 < |h| < then |
Since that

f is continuous at c, choose h > 0 for the moment, we have

F (c + h) F (c) L| < . h > 0 so that |f (t) f (c) | <

that

(5.42) if

|t c| < .

Now, assuming

F (c + h) F (c)

= = =
c a

c+h a

c a

f f
(5.43)

f+

c+h c f a c c+h f, c

and

L=

c+h c

(5.44)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

132

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

So, if

0 < h < ,

then

)F (c) | F (c+hh L| =

R c+ h
c

= = =
where the last inequality follows because for argument holds if

R c+ h f (t) dt L c | h R ch +h (f (t)L) dt c | | R c+ h h | f ( t ) L | dt c h R c+ h |f (t)f (c)| dt c h R c+ h c h

(5.45)

,
we have that

t [c, c + h] ,

|t c| h < .

A similar

h < 0.

(See the following exercise.) This proves part (2).

G is continuous on [a, b] , dierentiable on (a, b) , and that G' (x) = f (x) for all x (a, b) . Then, F G is continuous on [a, b] , dierentiable on (a, b) , and by part (2) ' (F G) (x) = F ' (x)G' (x) = f (x)f (x) = 0 for all x (a, b) . It then follows from Exercise 4.4.2 that F G is a constant function C, whence,
Suppose nally that

G (b) G (a) = F (b) + C F (a) C = F (b) =

a
and the theorem is proved.

f (t) dt,

(5.46)

Exercise 5.4.1
a. Complete the proof of part (2) of the preceding theorem; i.e., take care of the case when

h < 0.

HINT: In this case,

b. Suppose interval

c c+h c f = a f + c+h f. a ' is a continuous function on the closed interval [a, b] , and that f exists and is

a < c + h < c.

Then, write

continuous on the open interval

[a, b] .

Prove

(a, b) . Assume further that f ' is integrable on the closed x ' that f (x) f (a) = f for all x [a, b] . Be careful to understand how a x1
x
we have

this is dierent from the Fundamental Theorem. c. Use the Fundamental Theorem to prove that for

ln (x) = F (x)
1
and for

1 dt, t
1

(5.47)

0<x<1

we have

ln (x) = F (x)
x

1 dt. t x = 1.

(5.48)

HINT: Show that these two functions have the same derivative and agree at

5.5 Consequences of the Fundamental Theorem

The rst two theorems of this section constitute the basic techniques of integration taught in a calculus course. However, the careful formulations of these standard methods of evaluating integrals have some subtle points, i.e., some hypotheses. Calculus students are rarely told about these details.

5 This

content is available online at <http://cnx.org/content/m36211/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

133

Theorem 5.10:
Let by

Integration by Parts Formula

and

be integrable functions on

[a, b] , and as usual let F

x

and

G denote the functions dened

F (x) =
a
Then

f,

and

G (x) =
a

g.

(5.49)

f G = [F (b) G (b) F (a) G (a)]

a
Or, recalling that

F g.
a

(5.50)

f = F'

and

g = G' ,
b

F ' G = [F (b) G (b) F (a) G (a)]

a a

F G' .

(5.51)

Exercise 5.5.1
a. Prove the preceding theorem. HINT: Replace the upper limit

by a variable

x,

and dieren-

tiate both sides. By the way, how do we know that the functions b. Suppose

Fg

and

fG

are integrable?

[a, b] and that both f ' and g ' are continuous on (a, b) and integrable on [a, b] . (Of course f and g ' are not even dened at the endpoints a and b, but they can still be integrable on [a, b] . See the remark following Exercise 5.3.5.) Prove f
and

are integrable functions on

'

that

f g ' = [f (b) g (b) f (a) g (a)]

a a

f ' g.

(5.52)

Theorem 5.11:
Let

Integration by Substitution

[a, b] , and suppose g is a continuous, one-to-one function from [c, d] onto [a, b] such that g is continuously dierentiable on (c, d) , and such that a = g (c) and b = g (d) . Assume nally that g ' is integrable on [c, d] . Then
be a continuous function on

f (t) dt =

f (g (s)) g ' (s) ds.

c

(5.53)

f (g (s)) g ' (s) is continuous on (a, b) and integrable on [c, d] . It also follows from our assumptions that g maps the open interval (c, d) onto the x open interval (a, b) . As usual, let F denote the function on [a, b] dened by F (x) = f (t) dt. a ' Then, by part (2) of the Fundamental Theorem, F is dierentiable on (a, b) , and F = f. Then, by the chain rule, F g is continuous and dierentiable on (c, d) and
It follows from our assumptions that the function

Proof:

(F g ) (s) = F ' (g (s)) g ' (s) = f (g (s)) g ' (s) .

So, by part (3) of the Fundamental Theorem, we have that

'

(5.54)

d c

f (g (s)) g ' (s) ds = = = = =

d c

(F g ) (s) ds
(5.55)

'

(F g ) (d) (F g ) (c) F (g (d)) F (g (c)) F (b) F (a)

b a

f (t) dt,

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

134

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

which nishes the proof.

Exercise 5.5.2
a. Prove the Mean Value Theorem for integrals: If a

is continuous on

[a, b] ,

then there exists

c (a, b)

such that

f (t) dt = f (c) (b a) .
a
b. (Uniform limits of dierentiable functions. a function Compare with Exercise 4.8.1.) Suppose

(5.56)

{fn }

is a sequence of continuous functions on a closed interval

[a, b]

that converges pointwise to

' fn is continuous on the open interval (a, b) , is ' [a, b] , and that the sequence {fn } converges uniformly to ' a function g on (a, b) . Prove that f is dierentiable on (a, b) , and f = g. HINT: Let x be in (a, b) , and let c be in the interval (a, x) . Justify the following equalities, and use them

f.

Suppose that each derivative

integrable on the closed interval

together with the Fundamental Theorem to make the proof.

x
' fn =

f (x) f (c) = lim (fn (x) fn (c)) = lim

c

g.
c

(5.57)

We revisit now the Remainder Theorem of Taylor, which we rst presented in Theorem 4.19, Taylor's Remainder Theorem, p. 107. The point is that there is another form of this theorem, the integral form, and this version is more powerful in some instances than the original one, e.g., in the general Binomial Theorem below.

Theorem 5.12:
Let

Integral Form of Taylor's Remainder Theorem

be a real number, and let

have

f (n+1) I ([c r, c + r]) .

Then for each

n + 1 derivatives c < x < c + r,

x

on

(c r, c + r) ,
n

and suppose that

f (x) T(n f,c) (x) =

c
where

f (n+1) (t) f.

(x t) dt, n!

(5.58)

n Tf

denotes the

Similarly, for

nth Taylor c r < x < c,

polynomial for

f (x) T(n f,c) (x) =

x

f (n+1) (t)

(x t) dt. n!

(5.59)

Exercise 5.5.3
Prove the preceding theorem.

5.7: REMARK We return now to the general Binomial Theorem, rst studied in Theorem 4.21, p.
110. The proof given there used the derivative form of Taylor's remainder Theorem, but we were only able to prove the Binomial Theorem for which

HINT: Argue by induction on

n,

and integrate by parts.

|t| < 1/2.

The theorem below uses the integral form

of Taylor's Remainder Theorem in its proof, and it gives the full binomial theorem, i.e., for all

for

|t| < 1 .
General Binomial Theorem be a xed complex number. Then

Theorem 5.13:
Let

= a + bi

(1 + t) =
k=0

k t k

(5.60)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

135

Proof:
of

for all

t (1, 1) .

For clarity, we repeat some of the proof of Theorem 4.21, p. 110. Given a general consider the function

g : (1, 1) C

dened by

g (t) = (1 + t) .

Observe that the

= a + bi, nth derivative

is given by

g (n) (t) =
Then

( 1) ... ( n + 1) (1 + t)
n

(5.61)

g C ((1, 1)) . k
dene

For each nonnegative integer

ak = g (k) (0) /k ! =
and set

( 1) ... ( k + 1) , = k! k

(5.62)

k k=0 ak t . The radius of convergence for the power series function h is 1, as was shown in Exercise 4.10.1. We wish to show that g (t) = h (t) for all 1 < t < 1. That is, we wish to

h (t) = g

show that

is a Taylor series function around 0. It will suce to show that the sequence

partial sums of the power series function partial sum is just the Now, x a

converges to the function

g.

We note also that the

{Sn } of nth

for g. 1. The argument for t's between 1 and 0 is completely analogous.. Choose an > 0 for which = (1 + ) t < 1. We let C be a numbers such that | n | n C (1 + ) for all nonnegative integers n. See Exercise 4.10.1. We will also need the following Taylor polynomial

nth

n Tg

strictly between 0 and

estimate, which can be easily deduced as a calculus exercise (See part (d) of Exercise 4.4.1.). For all

s between 0 and t, we have (t s) / (1 + s) t. Note also that, for any s (0, t) , we have a a | (1 + s) | = (1 + s) , and this is trapped between 1 and (1 + t) . Hence, there exists a number Mt 1 such that | (1 + s) | Mt for all s (0, t) . We will need this estimate in the calculation that
follows. Then, by the integral form of Taylor's Remainder Theorem, we have:

|g (t)

n k=0

a k tk | = = = = = |

n |g (t) Tg (t) | t (n+1) s)n g (s) (t ds| n! 0 t (n+1) n1 n | 0 (1 + s) (t s) ds| n+1 t 1 ts n | (n + 1) | 1+ | n s | ds +1 || (1 + s) 0 t n | n +1 |Mt (n + 1) t ds 0

(5.63)

C Mt (n + 1)

t 0

(1 + )

n+1 n

t ds

C Mt (n + 1) (1 + )

n+1 n+1

C Mt (n + 1) n+1 , < 1.
This completes the proof for

which tends to 0 as

goes to

because

0 < t < 1.

5.6 Area of Regions in the Plane

number length

It would be desirable to be able to assign to each subset

of the Cartesian plane

R2

a nonnegative real

A (S ) called its area. We would L and W then the number A (S )

insist based on our intuition that (i) if should be

is a rectangle with sides of

LW,

so that this abstract notion of area would generalize

our intuitively fundamental one.

We would also insist that (ii) if

were the union of two disjoint parts,

6 This

content is available online at <http://cnx.org/content/m36212/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

136

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

then

A (S1 ) + A (S2 ) . (We were taught in high school plane geometry that S were the union of an innite number of disjoint parts, S = S with S S = if i = j, we would insist that (iii) A (S ) = i j n=1 n n=1 A (Sn ) . 2 The search for such a denition of area for every subset of R motivated much of modern mathematics.
should be the whole is the sum of its parts.) In fact, even if Whether or not such an assignment exists is intimately related to subtle questions in basic set theory, e.g., the

S = S1 S2 ,

A (S )

Axiom of Choice

and the

Continuum Hypothesis.

Most mathematical analysts assume that the Axiom

of Choice holds, and as a result of that assumption, it has been shown that there can be no assignment

S A (S )

satisfying the above three requirements. Conversely, if one does not assume that the Axiom of

Choice holds, then it has also been shown that it is perfectly consistent to assume as a basic axiom that such an assignment

S A (S )

does exist. We will not pursue these subtle points here, leaving them to a course

in Set Theory or Measure Theory. However, Here's a statement of the Axiom of Choice, and we invite the reader to think about how reasonable it seems.

5.8: AXIOM OF CHOICE Let S

for

be a collection of sets. Then there exists a set

that contains

exactly one element out of each of the sets

in

S.

The diculty mathematicians encountered in trying to dene area turned out to be involved with dening

A (S )

every

subset

S R2 .

To avoid this diculty, we will restrict our attention here to certain 

reasonable subsets geometric sets.

S. Of course, we certainly want these sets to include the rectangles and all other common

Denition 5.8:

2 By a (open) rectangle we will mean a set R = (a, b) (c, d) in R . That is, R = {(x, y ) : a < x < b and c < y < d}. The analogous denition of a closed rectangle [a, b] [c, d] should be clear: [a, b] [c, d] = {(x, y ) : a x b, c y d}. By the area of a (open or closed) rectangle R = (a, b) (c, d) or [a, b] [c, d] we mean the number A (R) = (b a) (d c) . .

The fundamental notion behind our denition of the area of a set S is this. If an open rectangle R = (a, b) (c, d) is a subset of S, then the area A (S ) surely should be greater than or equal to A (R) = (b a) (d c) . And, if S contains the disjoint union of several open rectangles, then the area of S should be greater than or equal to the sum of their areas. We now specify precisely for which sets we will dene the area. interval in Let

and let

and

be two continuous real-valued functions on

[a, b] be a xed closed bounded [a, b] for which l (x) < u (x) for all

x (a, b) .

Denition 5.9:
Given

[a, b] , l, and u as in the above, let S be the set of all pairs (x, y ) R2 , for which a < x < b and l (x) < y < u (x) . Then S is called an open geometric set. If we replace the < signs with signs, i.e., if S is the set of all (x, y ) such that a x b, and l (x) y u (x) , then S is called a closed geometric set. In either case, we say that S is bounded on the left and right by the vertical line segments {(a, y ) : l (a) y u (a)} and {(b, y ) : l (b) y u (b)}, and it is bounded below by the graph of the function l and bounded above by the graph of the function u. We call the union of these four bounding curves the boundary of S, and denote it by CS . If the bounding functions u and l of a geometric set S are smooth or piecewise smooth functions, we will call S a smooth or piecewise smooth geometric set. S
is a closed geometric set, we will indicate the corresponding open geometric set by the symbol

If

S0.

The symbol

S0

we have introduced for the open geometric set corresponding to a closed one is the same

symbol that we have used previously for the interior of a set. Study the exercise that follows to see that the two uses of this notation agree.

Exercise 5.6.1

a. Show that rectangles, triangles, and circles are geometric sets. What in fact is the denition of a circle?

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

137

b. Find some examples of sets that are or a heart on its side. c. Let of

not geometric sets.

Think about a horseshoe on its side, under the graph

be a continuous, nonnegative function on

[a, b] . Show that the region

is a geometric set. Prove that the interior

d. Show that the intersection of two geometric sets is a geometric set. Describe the left, right, upper, and lower boundaries of the intersection. intersection of two geometric sets two interiors. e. Give an example to show that the union of two geometric sets need not be a geometric set. f. Show that every closed geometric set is compact. g. Let

S1

and

S2

coincides with the

(S1 S2 ) of the 0 0 intersection S1 S2 of their

be a closed geometric set. Show that the corresponding open geometric set

S0

coincides

S, i.e., the set of all points in the interior of S. HINT: Suppose a < x < b and l (x) < y < u (x) . Begin by showing that, because both l and u are continuous, there must exist an > 0 and a > 0 such that a < x < x + < b and l (x) < y < y + < u (x) .
with the interior of Now, given a geometric set functions

and

l,

let

S (either open or closed), that is determined by an interval [a, b] and two bounding P = {x0 < x1 < ... < xn } be a partition of [a, b] . For each 1 i n, dene numbers

ci

and

di

as follows:

ci =
Because the functions Of course,

sup
xi1 <x<xi

l ( x) ,

and

di =

inf
xi1 <x<xi

u (x) .

(5.64)

and

are continuous, they are necessarily bounded, so that the supremum and

inmum above are real numbers. For each

1 i n dene Ri to be the open rectangle (xi1 , xi ) (ci , di ) . di may be < ci , in which case the rectangle Ri is the empty set. In any event, we see that the partition P determines a nite set of (possibly empty) rectangles {Ri }, and we denote the union of these n rectangles by the symbol CP . = i=1 (xi1 , xi ) (ci , di ) . The area of the rectangle Ri is (xi xi1 ) (di ci ) if ci < di and 0 otherwise. We may write in general that A (Ri ) = (xi xi1 ) max ((di ci ) , 0) . Dene the number AP by
n

AP =
i=1
Note that that

(xi xi1 ) (di ci ) .

(5.65)

AP is not exactly the sum of the areas of the rectangles determined by P because it may happen di < ci for some i's, so that those terms in the sum would be negative. In any case, it is clear that AP P,
we have

is less than or equal to the sum of the areas of the rectangles, and this notation simplies matters later. For any partition

S CP ,

so that, if

A (S )

is to denote the area of

S,

we want to have

A (S ) = =
n i=1 n i=1

n i=1

A (Ri )
(5.66)

(xi xi1 ) max ((di ci ) , 0) (xi xi1 ) (di ci ) AP . x = a,

Denition 5.10:
Let S be a geometric set x = b, below by the graph (either open or closed), bounded on the left by of

l,

and above by the graph of

u.

Dene the

areaA (S ) of S

on the right by by

A (S ) = supAP =
P

sup
P ={x0 <x1 <...<xn } i=1

(xi xi1 ) (di ci ) , P

of

(5.67)

where the supremum is taken over all partitions as dened above.

[a, b] ,

and where the numbers

ci

and

di

are

Exercise 5.6.2

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

138

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

a. Using the notation of the preceding paragraphs, show that each rectangle the set

Ri

is a subset of

and that

Ri Rj =

if

i = j.

It may help to draw a picture of the set

and the

rectangles b. Suppose Prove that

{Ri }.

Can you draw one so that

di < ci ?
is another geometric set that is contained in

S1

is a geometric set and that

S2

S1 .

A (S2 ) A (S1 ) .

HINT: For each partition

P,

compare the two

AP 's.

Exercise 5.6.3
Let and

T H

be the triangle in the plane with vertices at the three points

that the area

A (T ) ,

as dened above, agrees with the formula

(0, 0) , (0, H ) , and (B, 0) . Show A = (1/2) BH, where B is the base
In fact, this

is the height.

The next theorem gives the connection between area (geometry) and integration (analysis). theorem is what most calculus students think integration is all about.

Theorem 5.14:
Let

be a geometric set, i.e., a subset of

R2
b

that is determined in the above manner by a closed

bounded interval

[a, b]

and two bounding functions

and

u.

Then

A (S ) =

(u (x) l (x)) dx.

a

(5.68)

Proof:
Let

be a partition of [a, b] , and let ci and di be dened as above. Let h di on the open interval (xi1 , xi ) , and let k be a step function that equals ci on the open interval (xi1 , xi ) . Then on each open interval (xi1 , xi ) we have h (x) u (x) and k (x) l (x) . Complete the denitions of h and k by dening them at the partition points so that h (xi ) = k (xi ) for all i. Then we have that h (x) k (x) u (x) l (x) for all x [a, b] . Hence, be a step function that equals

P = {x0 < x1 < ... < xn }

AP =
i=1

(xi xi1 ) (di ci ) =

a

(h k )
a

(u l ) .

(5.69)

Since this is true for every partition

of

[a, b] , it follows by taking the supremum over all partitions

b

that

A (S ) = supAP
P
which proves half of the theorem; i.e., that To see the other inequality, let

(u (x) l (x)) dx,

a

(5.70)

and let

h (x) u (x) for all x, k (x) l (x) for all x. Let P = {x0 < x1 < ... < xn } be a partition of [a, b] for which both h and k are constant on the open subintervals (xi1 , xi ) of P. Let a1 , a2 , ..., an and b1 , b2 , ..., bn be the numbers such that h (x) = ai on (xi1 , xi ) and k (x) = bi on (xi1 , xi ) . It follows, since h (x) u (x) for all x, that ai di . Also, it follows that bi ci . h k
be any step function for which Therefore,

b u l. a be any step function on [a, b] for which

A (S )

(h k ) =
a
Finally, let let

(ai bi ) (xi xi1 )

i=1 i=1

(xi xi1 ) (di ci ) = AP A (S ) .

(5.71)

{km }

be a nonincreasing sequence of step functions that converges uniformly to

{hm } be a nondecreasing sequence of step functions that converges uniformly to u, and l. See part (d)
b b

of Exercise 5.3.3. Then

(u l) = lim
a m a

(hm km ) A (S ) ,

(5.72)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

139

which proves the other half of the theorem.

OK! Trumpet fanfares, please!

Theorem 5.15:
(A

Proof:

= r2 .)

If

is a circle in the plane having radius

r,

then the area

A (S )

of

is

r2 .

Suppose the center of the circle describe the circle with center bounded interval

S is the point (h, k ) . This circle is a geometric set. In fact, we may (h, k ) and radius r as the subset S of R2 determined by the closed

[h r, h + r]

and the functions

u (x) = k +
and

r2 (x h)

(5.73)

l (x) = k
By the preceding theorem, we then have that

r2 (x h) .

(5.74)

h+r

A (S ) =
hr

r2 (x h) dx = r2 .

(5.75)

We leave the verication of the last equality to the following exercise.

Exercise 5.6.4
Evaluate the integral in the above proof:

h+r

2
hr

r2 (x h) dx.

(5.76)

Be careful to explain each step by referring to theorems and exercises in this book. It may seem like an elementary calculus exercise, but we are justifying each step here.

5.9: REMARK There is another formula for the area of a geometric set that is sometimes very useful.
There is really nothing new to this

This formula gives the area in terms of a double integral. as the integral from

formula; it simply makes use of the fact that the number (length)

u (x) l (x)

can be represented

l ( x)

to

u ( x)

of the constant 1. Here's the formula:

u(x)

A (S ) =
a l(x)

1 dy

dx.

(5.77)

The next theorem is a result that justies our denition of area by verifying that the whole is equal to the sum of its parts, something that any good denition of area should satisfy.

Theorem 5.16:
S

Let

be a closed geometric set, and suppose

0 sets for which Si

0 Sj

S = n i=1 Si , where the sets {Si } are closed geometric

n

if

i = j.

Then

A (S ) =
i=1

A (Si ) .

(5.78)

Proof:
Suppose

is determined by the interval

Si

is determined by the interval

[a, b] and the two bounding functions l and u, and suppose [ai , bi ] and the two bounding functions li and ui . Because Si S, it

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

140

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

[ai , bi ] is contained in the interval [a, b] . Initially, the bounding functions li ui are dened and continuous on [ai , bi ] , and we extend their domain to all of [a, b] by dening li (x) = ui (x) = 0 for all x [a, b] that are not in [ai , bi ] . The extended functions li and ui may not be continuous on all of [a, b] , but they are still integrable on [a, b] . (Why?) Notice that we now
must be that the interval and have the formula

bi

A (Si ) =
ai
Next, x an

(ui (x) li (x)) dx =

a

(ui (x) li (x)) dx.

(5.79)

x in the open interval (a, b) . We must have that the vertical intervals (li (x) , ui (x)) (lj (x) , uj (x)) are disjoint if i = j. Otherwise, there would exist a point y in both intervals, 0 0 and this would mean that the point (x, y ) would belong to both Si and Sj , which is impossible by hypothesis. Therefore, for each x (a, b) , the intervals {(li () x) , ui (x)} are pairwise disjoint open intervals, and they are all contained in the interval (l (x) , u (x)) , because the Si 's are subsets of S. Hence, the sum of the lengths of the open intervals {(li (x) , ui (x))} is less than or equal to the length of (l (x) , u (x)) . Also, for any point y in the closed interval [l (x) , u (x)] , the point (x, y ) must belong to one of the Si 's, implying that y is in the closed interval [li (x) , ui (x)] for some i. But this means that the sum of the lengths of the closed intervals [li (x) , ui (x)] is greater than or equal to the length of the interval [l (x) , u (x)] . Since open intervals and closed intervals have the n same length, we then see that (u (x) l (x) = i=1 (ui (x) li (x)) .
and We now have the following calculation:

n i=1

A (Si )

= = = = =

bi n i=1 ai (ui (x) li (x)) dx b n i=1 a (ui (x) li (x)) dx b n i=1 (ui (x) li (x)) dx a b (u (x) l (x)) dx a

(5.80)

A (S ) ,

which completes the proof.

5.7 Extending the Denition of Integrability

Denition 5.11:
Let

We now wish to extend the denition of the integral to a wider class of functions, namely to some that are unbounded and Others whose domains are not closed and bounded intervals. somewhat ad hoc, and these integrals are sometimes called improper integrals. This extended denition is

be a real or complex-valued function on the open interval

is possibly

+ .
'

We say that

is

and bounded subinterval

a' , b' (a, b) ,

improperly-integrable

on

and for each point

(a, b) where a is possibly and (a, b) if it is integrable on each closed c (a, b) we have that the two limits f,
not necessarily dened on all

b c limb' b 0 c f and lima' a+0 a' f exist. More generally, We say that a real or complex-valued function
of the open interval such that

(a, b) ,

is

improperly-integrable
f

on

(a, b)

if there exists a partition

{xi }

of

[a, b]
by

is dened and improperly-integrable on each open interval

(xi1 , xi ) . (a, b)
and

We denote the set of all functions

that are improperly-integrable on an open interval

Ii ((a, b)) .
Analogous denitions are made for a function's being integrable on half-open intervals

[a, b)

(a, b] .

7 This

content is available online at <http://cnx.org/content/m36222/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

141

Note that, in order for

to be improperly-integrable on an open interval, we only require

to be dened

at almost all the points of the interval, i.e., at every point except the endpoints of some partition.

Exercise 5.7.1
a. Let

be

dened

and

improperly-integrable

on

the

open

interval

(a, b) .

Show

that

b. c.

f + limb' b0 c f is the same for all c (a, b) . 1/2 Dene a function f on (0, 1) by f (x) = (1 x) . Show that f is improperly-integrable on (0, 1) and that f is not bounded. (Compare this with part (1) of Theorem 5.5, p. 126.) 1 Dene a function g on (0, 1) by g (x) = (1 x) . Show that g is not improperly-integrable on (0, 1) , and, using part (b), conclude that the product of improperly-integrable functions on (0, 1) need not itself be improperly-integrable. (Compare this with part (3) of Theorem 5.5, lima' a+0
p. 126.)

c a'

b'

d. Dene 126.)

to be the function on

(0, )

given by

h ( x) = 1

for all

x.

Show that

is not

improperly-integrable on

(0, ) .

(Compare this with parts (4) and (5) of Theorem 5.5, p.

Part (a) of the preceding exercise is just the consistency condition we need in order to make a denition of the integral of an improperly-integrable function over an open interval.

Denition 5.12:
Let

be dened and improperly-integrable on an open interval

(a, b) .

We dene the

over the interval

(a, b) ,

and denote it by

b a

integral

of

f,

by

b'

f = lim
a
In general, if of

a' a+0

f + lim
a'

b' b0

f.
c

(5.81)

is improperly-integrable over an open interval, i.e.,

integrable over each subinterval of

f is dened and improperly(a, b) determined by a partition {xi }, then we dene the integral
b n xi

over the interval

(a, b)

by

f=
a i=1 xi1

f.

(5.82)

Theorem 5.17:
Let let 1. 2. 3. 4.

(a, b) be a xed open interval (with a possibly equal to and b possibly Ii ((a, b)) denote the set of improperly-integrable functions on (a, b) . Then: Ii ((a, b))
is a vector space of functions. (Linearity)

equal to

+ ,

and

b a

(f + g ) = f (x) 0
for all

b a

f +

b a

for all

f, g Ii ((a, b))and , C. f 0.
for all

(Positivity) If

x (a, b) ,
and

then

b a

(Order-preserving) If

f, g Ii ((a, b))

f (x) g (x)

x (a, b) ,

then

b a

b a

g.

Exercise 5.7.2
a. Use Theorem 5.5, p. 126, Theorem 5.6, p. 127, Theorem 5.7, p. 129, and properties of limits to prove the preceding theorem. b. Let a

be dened and improperly-integrable on

'

(a, b) .

Show that, given an

'

c. Let

> 0 such that for any a < a < a + and any b < b f be improperly-integrable on an open interval (a, b) . exists a > 0 such that if (c, d) is any open subinterval of (a, b) for which d c < , then d | c f | < . HINT: Let {xi } be a partition of [a, b] such that f is dened and improperlyintegrable on each subinterval (xi1 , xi ) . For each i, choose a i using part (b). Now f is bounded by M on all the intervals [xi1 + i , xi i ] , so = /M should work there.
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

there exists a' b < b we have | a f | + | b' f | < . Show that, given an > 0, there

> 0,

142

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

d. Suppose

is a continuous function on a closed bounded interval

dierentiable on the open interval evaluate

b ' f . HINT: Fix a a to show that the two limits exist.

[a, b] and is continuously (a, b) . Prove that f ' is improperly-integrable on (a, b) , and point c (a, b) , and use the Fundamental Theorem of Calculus

e. (Integration by substitution again.) Let

g : [c, d] [a, b] be continuous on [c, d] and satisfy g (c) = a and g (d) = b. Suppose there exists a partition {x0 < x1 < ... < xn } of the interval [c, d] such that g is continuously dierentiable on each subinterval (xi1 , xi ) . Prove that g ' is improperly-integrable on the open interval (c, d) . Show also that if f is continuous on [a, b] ,
we have that

f (t) dt =
a
HINT: Integrate over the subintervals

f (g (s)) g ' (s) ds.

c
and use part (d).

(5.83)

(xi1 , xi ) ,

5.10: REMARK Note that there are parts of Theorem 5.5, p.

not asserted in Theorem 5.17, p. 141.

126 and Theorem 5.6, p. 127 that are

The point is that these other properties do not hold for

improperly-integrable functions on open intervals. See the following exercise.

Exercise 5.7.3
a. Dene

[1, ) given by f (x) = (1) /n if n 1 x < n. Show that f is improperly-integrable on (1, ) , but that |f | is not improperly-integrable on (1, ) . N (Compare this with part (4) of Theorem 5.6, p. 127.) HINT: Verify that f is a partial 1 N sum of a convergent innite series, and then verify that | f | is a partial sum of a divergent 1 f
to be the function on innite series.

n1

b. Dene the function

c.

f on (1, ) by f (x) = 1/x. For each positive integer n, dene the function fn (x) = 1/x if 1 < x < n and fn (x) = 0 otherwise. Show that each fn is improperly-integrable on (1, ) , that f is the uniform limit of the sequence {fn }, but that f is not improperly-integrable on (1, ) . (Compare this with part (5) of Theorem 5.6.) Suppose f is a nonnegative real-valued function on the half-open interval (a, ) that is in' tegrable on every closed bounded subinterval a, b . For each positive integer n a, dene n yn = a f (x) dx. Prove that f is improperly-integrable on [a, ) if and only if the sequence {yn } is convergent. In that case, show that a f = limyn . fn
on

(1, )

by

We are now able to prove an important result relating integrals over innite intervals and convergence of innite series.

f be a positive function on [1, ) , assume that f is integrable on every closed bounded interval [1, b] , and suppose that f is nonincreasing; i.e., if x < y then f (x) f (y ) . For each positive integer N i, set ai = f (i) , and let SN denote the N th partial sum of the innite series ai :SN = i=1 ai .
Let Then: 1. For each

Theorem 5.18:

N,

we have

SN a1
1
2. For each

f (x) dx SN 1 .

(5.84)

N,

we have that

SN 1
1
i.e., the sequence

f (x) dx a1 aN a1 ;

(5.85)

{SN 1

N 1

f}

is bounded above.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

143

3. The sequence 4. integrable on

N f } is nondecreasing. 1 (Integral Test) The innite series ai converges if and only if the function

{SN 1

is improperly-

(1, ) .

Proof:
For each positive integer N, dene a step function kN on the interval [1, N ] as follows. Let P = {x0 < x1 < ... < xN 1 } be the partition of [1, N ] given by the points {1 < 2 < 3 < ... < N }, i.e., xi = i + 1. Dene kN (x) to be the constant ci = f (i + 1) on the interval [xi1 , xi ) = [i, i + 1) . Complete the denition of kN by setting kN (N ) = f (N ) . Then, because f is nonincreasing, we have that kN (x) f (x) for all x [1, N ] . Also,

N 1

kN

= = = = =

N 1 i=1 ci (xi xi1 ) N 1 i=1 f (i + 1) N i=2 f (i) N i=2 ai

(5.86)

SN a1 ,

which then implies that

SN a1 =
1
This proves half of part (1). For each positive integer above, by setting by setting

kN (x) dx
1

f (x) dx.

(5.87)

N > 1 dene another step function lN , using the same partition P as lN (x) = f (i) if i x < i + 1 for 1 i < N, and complete the denition of lN lN (N ) = f (N ) . Again, because f is nonincreasing, we have that f (x) lN (x) for all
Also

x [1, N ] .

N l 1 N

= = =

N 1 i=1 f (i) N 1 i=1 ai

(5.88)

SN 1 ,

which then implies that

f (x) dx
1
and this proves the other half of part (1). It follows from part (1) that

lN (x) dx = SN 1 ,
1

(5.89)

SN 1
1
and this proves part (2). We see that the sequence

f (x) dx SN 1 SN + a1 = a1 aN ,
N 1

(5.90)

{SN 1
N +1 1

f}

is nondecreasing by observing that

SN

f SN 1 +

N 1

aN

= f (N )

N +1 f N N +1 N f

(5.91)

0,

because

is nonincreasing.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

144

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

Finally, to prove part (4), note that both of the sequences If

{SN }

and

f is improperly-integrable on [1, ) , then limN 1 f exists, and f (x) dx for all 1 N, which implies that ai converges by Theorem 2.14, p. 48. Conversely, if ai converges, then N limSN exists. Since 1 f (x) dx SN 1 , it then follows, again from Theorem 2.14, p. 48, that N limN 1 f (x) dx exists. So, by the preceding exercise, f is improperly-integrable on [1, ) .
We may now resolve a question rst raised in Exercise 2.8.8. That is, for

{ 1 f} SN a1 +

are nondecreasing.

1 < s < 2,

is the innite series

1/ns
number.

convergent or divergent?

We saw in that exercise that this series is convergent if

is a rational

Exercise 5.7.4
a. Let b. Let

s s

be a real number. be a

Use the Integral Test to prove that the innite series Prove that the innite series

1/ns

is

convergent if and only if convergent

s > 1. complex number s = a + bi. if and only if a > 1.

1/ns

is absolutely

Exercise 5.7.5
Let

be the function on

[1, )

dened by

f (x) = 1/x. {

a. Use Theorem 5.18, p. 142 to prove that the sequence number

1.

(This number

N 1 i=1 i lnN } converges to a positive is called Euler's constant.) HINT: Show that this sequence is

bounded above and nondecreasing. b. Prove that

i=1
HINT: Write

(1) i

i+1

= ln2.

(5.92)

S2N

for the

2N th

partial sum of the series. Use the fact that

2N

S2N =
i=1
Now add and subtract

1 1 2 . i 2i i=1

(5.93)

ln (2N )

and use part (a).

5.8 Integration in the Plane

Let

be a closed geometric set in the plane. If

is a real-valued function on

S, we would like to dene what

it means for

to be integrable and then what the integral of

development for integration of functions on a closed interval

f [a, b] .

is. To do this, we will simply mimic our

So, what should be a step function in this context? That is, what should is a partition of partition. Our idea is to replace the subintervals determined by a partition of the interval subsets of the geometric set

be in

this context? Presumably a step function is going to be a function that is constant on the elements of a

[a, b] by geometric

S. S1
and

Denition 5.13:
The

overlap

of two geometric sets and

intersection.

S1

S2

are called

nonoverlapping

S2

is dened to be the interior if this overlap

(S1 S2 )

of their

(S1 S2 )

is the empty set.

8 This

content is available online at <http://cnx.org/content/m36223/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

145

Denition 5.14:
A

partition of a closed geometric set S in R2 is a nite collection {S1 , S2 , ..., Sn } of nonoverlapping

The open subsets A

closed geometric sets for which

step function

0 } {Si

n i=1 Si = S ;

are called the of

elements

i.e., the union of the

Si 's

is all of the geometric set

S.

of the partition.

on the closed geometric set

exists a partition element of the

5.11: REMARK

P = {Si } partition P.

such that

S is a real-valued function h on S for which there 0 h (z ) = ai for all z Si ; i.e., h is constant on each

One example of a partition of a geometric set, though not at all the most general

kind, is the following. Suppose the geometric set bounding functions

S is determined by the interval [a, b] and the two {x0 < x1 < ... < xn } be a partition of the interval [a, b] . We make a partition {Si } of by constructing vertical lines at the points xi from l (xi ) to u (xi ) . Then Si is the geometric set determined by the interval [xi1 , xi ] and the two bounding functions ui and li that are the restrictions of u and l to the interval [xi1 , xi ] . u S
and

l.

Let

A step function is constant on the open geometric sets that form the elements of some partition. We say nothing about the values of

h on the boundaries

of these geometric sets. For a step function

h on an interval

[a, b] ,

we do not worry about the nitely many values of

at the endpoints of the subintervals. However, in

the plane, we are ignoring the values on the boundaries, which are innite sets. As a consequence, a step function on a geometric set may very well have an innite range, and may not even be a bounded function, unlike the case for a step function on an interval. aect the integral (average value) of the function. Before continuing our development of the integral of functions in the plane, we digress to present an analog of Theorem 3.21, p. 78 to functions that are continuous on a closed geometric set. The idea is that the boundaries of geometric sets are negligible sets as far as area is concerned, so that the values of a function on these boundaries shouldn't

Theorem 5.19:
Let

be a continuous real-valued function whose domain is a closed geometric set

S.

Then there

exists a sequence

Proof:

{hn }

of step functions on

that converges uniformly to

f.

As in the proof of Theorem 3.21, p. 78, we use the fact that a continuous function on a compact set is uniformly continuous.

n, let n be a positive number satisfying |f (z ) f (w) | < 1/n if n exists by the uniform continuity of f on S. Because S is compact, it is bounded, and we let R = [a, b] [c, d] be a closed rectangle that contains S. We construct a n n partition {Si } of S as follows. In a checkerboard fashion, we write R as the union Ri of small,
For each positive integer Such a

|z w| < n .

closed rectangles satisfying 1. If 2.

n z and w are in Ri , then |z w| < n . (The rectangles are that small.) n0 n0 Ri Rj = . (The interiors of these small rectangles are disjoint.)

Now dene

n n n0 n0 n n Si = S Ri . Then Si Sj = , and S = Si . Hence, {Si } is a partition of S. n n n n For each i, choose a point zi in Si , and set ai = f (zi ) . We dene a step function hn as follows: n0 n If z belongs to one (and of course only one) of the open geometric sets Si , set hn (z ) = ai . And, if n0 z does not belong to any of the open geometric sets Si , set hn (z ) = f (z ) . It follows immediately that hn is a step function. n0 Now, we claim that |f (z ) hn (z ) | < 1/n for all z S. For any z in one of the Si 's, we have n |f (z ) hn (z ) | = |fz an i | = |f (z ) f (zI ) | < 1/n
(5.94)

n n0 because |z zi | < n . And, for any z not in any of the Si 's, f (z ) hn (z ) = 0. So, we have dened a sequence {hn } of step functions on S, and the sequence {hn } converges uniformly to f by
Exercise 3.10.2.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

146

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

What follows now should be expected. We will dene the integral of a step function geometric set

over a

by

h=
S
We will dene a function

ai A (Si ) .
i=1

(5.95)

on

to be integrable if it is the uniform limit of a sequence

{hn }

of

step functions, and we will then dene the integral of

by

f = lim
S
we had for the denition of the integral on Theorem 5.4, p. 123.

hn .
S
i.e., the analogs of Theorem 5.1, p.

(5.96)

Everything should work out nicely. Of course, we have to check the same two consistency questions

[a, b] ,

119 and

Let S be a closed geometric set, and let h be a step function on S. Suppose P = {S1 , ..., Sn } and 0 Q = {T1 , ..., Tm } are two partitions of S for which h (z ) is the constant ai on Si and h (z ) is the 0 constant bj on Tj . Then

Theorem 5.20:

ai A (Si ) =
i=1 j =1

bj A (Tj ) .

(5.97)

Proof:
We know by part (d) of Exercise 5.6.1 that the intersection of two geometric sets is itself a geometric set. Also, for each xed index

j,

we know that the sets

Theorem 5.16, p. 139, we have that that

A (Tj ) =

n i=1

0 {Tj Si } are pairwise disjoint. Then, by A (Tj Si ) . Similarly, for each xed i, we have

A Si = Tj0 .

m j =1

A (Tj Si ) . zi,j Tj0

Finally, for each pair

and

j,

for which the set because

0 Tj0 Si

is not

empty, choose a point

0 Si , and note that

ai = h (zi,j ) = bj ,

zi,j

belongs to both

0 Si

and

With these observations, we then have that

n i=1

ai A (Si )

= = = = = = =

n m i=1 ai j =1 A (Tj Si ) n m i=1 j =1 ai A (Tj Si ) n m i=1 j =1 h (zi,j ) A (Tj Si ) n m i=1 j =1 bj A (Tj Si ) m n i=1 bj A (Tj Si ) j =1 m n i=1 A (Tj Si ) j =1 bj m j =1 bj A (Tj ) ,

(5.98)

which completes the proof.

OK, the rst consistency condition is satised. Moving right along:

Denition 5.15:
Let set

be a step function on a closed geometric set

S.

Dene the

integral

of

over the geometric

by the formula

h=
S S

H (z ) dz =
i=1

ai A (Si ) ,

(5.99)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

147

where

S1 , ..., Sn

is a partition of

for which

is the constant

ai

on the interior

0 Si

of the set

Si .

Just as in the case of integration on an interval, before checking the second consistency result, we need to establish the following properties of the integral of step functions.

Theorem 5.21:
Let

H (S )

denote the vector space of all step functions on the closed geometric set

S.

Then the

assignment 1. 2. 3. 4.

of

H (S )

into

has the following properties:

H (S ) is a vector space, and S (h1 + h2 ) = S h1 + S h2 , and S ch = c S h for h1 , h2 , h H (S ) , and for all real numbers c. n If h = i=1 ci Si is a linear combination of indicator functions of geometric sets that are n subsets of S, then h = i=1 ci A (Si ) . (Positivity) If h (z ) 0 for all z S, then h 0. S (Order-preserving) If h1 and h2 are step functions on S for which h1 (z ) h2 (z ) for all z S, then h S h2 . S 1
(Linearity) all

Proof:

P = {Si } and h2 is constant on the elements Q = {Tj }. Let V be the partition of the geometric set S whose elements are the sets 0 {Uk } = {Si Tj0 }. Then both h1 and h2 are constant on the elements Uk of V, so that h1 + h2 is also constant on these elements. Therefore, h1 + h2 is a step function, and
Suppose

h1

is constant on the elements of a partition

of a partition

(h1 + h2 ) =
k

(ak + bk ) A (Uk ) =
k

ak A (Uk ) +
k

bk A (Uk ) =

h1 +

h2 ,

(5.100)

and this proves the rst assertion of part (1). The proof of the other half of part (1), as well as parts (2), (3), and (4), are totally analogous to the proofs of the corresponding parts of Theorem 5.2, p. 121, and we omit the arguments here.

Now for the other necessary consistency condition:

Theorem 5.22:
let

be a closed geometric set in the plane.

1. If 2.

{hn }

is a sequence of step functions that converges uniformly to a function sequence of real numbers. of step functions on

on

S,

then

{ S hn } is a convergent If {hn } and {kn } are two sequences same function f, then
the sequence

that converge uniformly to the

lim
S

hn = lim
S

kn .

(5.101)

Exercise 5.8.1 Denition 5.16:

If Prove Theorem 5.22, p. 147. Mimic the proofs of Theorem 5.3, p. 123 and Theorem 5.4, p. 123.

is a real-valued function on a closed geometric set

in the plane, then

is

integrable on S

if

it is the uniform limit of a sequence We dene the

integral

{hn }

of step functions on

S.

of an integrable function

on

by

f
S S

f (z ) dz = lim
S

hn , f.

(5.102)

Theorem 5.23:
Let on

where

{hn }

is a sequence of step functions on

that converges uniformly to

S be a S. Then:

closed geometric set in the plane, and let

I (S )

denote the set of integrable functions

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

148

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

1. 3. 4.

2. If

I (S ) is a vector space of functions. f and g I (S ) , and one of them is bounded, then f g I (S ) . Every step function is in I (S ) . If f is a continuous real-valued function on S, then f is in I (S ) . real-valued function on S is integrable on S.

That is, every continuous

Exercise 5.8.2
a. Prove Theorem 5.23, p. 147. Note that this theorem is the analog of Theorem 5.5, p. 126, but that some things are missing. b. Show that integrable functions on to be bounded. c. Show that, if the interior

are not necessarily bounded; not even step functions have

f I (S ) , and g is a function on S for which f (x, y ) = g (x, y ) for all (x, y ) in S 0 of S, then g I (S ) . That is, integrable functions on S can do whatever they

like on the boundary.

Theorem 5.24:
Let

be a closed geometric set. The assignment (Linearity)

on

I (S ) satises the following properties.

S

1. 2. 3. 4. 5. 6.

I (S )

is a vector space, and

, R.
(Positivity) If (Order-preserving) If

(f + g ) =

f +

for all

f, g I (S )and

7.

f (z ) 0 for all z S, then S f 0. f, g I (S ) and f (z ) g (z ) for all z S, then S f S g. If f I (S ) , then so is |f |, and | f | S |f |. S If f is the uniform limit of functions fn , each of which is in I (S ) , then f I (S ) and f = lim S fn . S Let {un } be a sequence of functions in I (S ) , and suppose that for each n there is a number mn , for which |un (z ) | mn for all z S, and such that the innite series mn converges. Then the innite series un converges uniformly to an integrable function, and S un = u . S n If f I (S ) , and {S1 , ..., Sn } is a partition of S, then f I (Si ) for all i, and
n

=
S i=1 Si

f.

(5.103)

Exercise 5.8.3
Prove Theorem 5.24, p. part, let 148. It is mostly the analog to Theorem 5.6, p. 127. To see the last then examine

hi f h i. S

be the step function that is identically 1 on

Si ;

check that

hi f I (Si ) ; S

Of course, we could now extend the notion of integrability over a geometric set functions just as we did for integrability over an interval sets will suce for the purposes of this book.

to include complex-valued

[a, b] .

However, real-valued functions on geometric

We include here, to be used later in Section 7.1, a somewhat technical theorem about constructing partitions of a geometric set.

Theorem 5.25:
Let

S1 , ..., Sn

be closed, nonoverlapping, geometric sets, all contained in a geometric set

S.

Then

there exists a partition the

S 1 , ..., S M

of

such that for

1in

we have

Si = S i .

In other words,

Proof:

si 's

are the rst

elements of a partition of

S. u
and

Suppose

is determined by the interval

[a, b]

and the two bounding functions

l.

We prove

this theorem by induction on

n.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

149

If Set ^

n = 1,

let

S1

be determined by the interval

[a1 , b1 ]
^

and the two bounding functions ^ as follows:

u1

and l1 .

S 1 = S1 , S2 S3 S4
^

and dene four more geometric sets

S 2 , ..., S 5

1. 2. 3.

is determined by the interval is determined by the interval is determined by the interval

[a, a1 ]

and the two bounding functions

and

restricted to

that interval.

[a1 , b1 ] [a1 , b1 ] [b1 , b]

and the two bounding functions

u l

and

u1

restricted

to that interval. and the two bounding functions and l1 restricted to that interval. ^ 4. S 5 is determined by the interval that interval. ^ ^ ^

and the two bounding functions

and

restricted to

Observe that the ve sets theorem in the case Then, given

S 1 , S 2 , ..., S 5

constitute a partition of the geometric set

S,

proving the

n = 1.

n sets satisfying the hypotheses. S1 , ..., Sn+1 as in the hypothesis of the theorem, apply the inductive hypothesis to the n sets S1 , ..., Sn to obtain a partition T1 , ..., Tm of S for which Ti = Si for all 1 i n. For each ' n + 1 i m, consider the geometric set Si = Sn+1 Ti of the geometric set Ti . We may apply ' ' the case n = 1 of this theorem to this geometric set to conclude that Si is the rst element Si,1 of ' ' ' a partition {Si,1 , Si,2 , ..., Si,m } of the geometric set Ti . i
Suppose next that the theorem is true for any collection of Dene a partition

{S k }

of

as follows: For ^

1 k n,

set

S k = Tk .
^

Set

' S n+1 = m i=n+1 Si,1 = ' Si,j

Sn+1 .

And dene the rest of the partition

{S k }

to be made up of the remaining sets

for

n + 1 i m and 2 j mi . It follows directly that this partition {S k } satises the requirements

of the theorem.

Exercise 5.8.4
Sk is determined by the interval [ak , bk ] and Sk is below Sj , equivalently Sj is above Sk , if there exists a point x such that uk (x) < lj (x) . Note that this implies that x [ak , bk ][aj , bj ] .
Let be as in the preceding theorem. Suppose the two bounding functions

S1 , ..., Sn

uk

and lk . We will say that

a. Suppose is, if

b.

c.

(z, yk ) Sk and (z, yj ) Sj . Show that yj > yk . That Sk can be above Sj . Suppose S2 is below S1 and S3 is below S2 . Show that no part of S3 can be above S1 . HINT: By way of contradiction, let x1 [a1 , b1 ] be such that u2 (x1 ) < l1 (x1 ) ; let x2 [a2 , b2 ] be such that u3 (x2 ) < l2 (x2 ) ; and suppose x3 [a3 , b3 ] is such that u1 (x3 ) < l3 (x3 ) . Derive contradictions for all possible arrangements of the three points x1 , x2 , and x3 . Prove that there exists an index k0 such that Sk0 is minimal in the sense that there is no other Sj that is below Sk0 . HINT: Argue by induction on n. Thus, let {Tl } be the collection of all Sk 's that are below S1 , and note that there are at most n 1 elements of {Tl }. By induction, there is one of the Tl 's, i.e., an Sk0 that is minimal for that collection. Now, using part (b), show that this Sk0 must be minimal for the original collection. Sk
is below

Sj ,

and suppose

Sk

is below

Sj ,

then no part of

There is one more concept about integrating over geometric sets that we will need in later chapters. We have only considered sets that are bounded on the left and right by straight vertical lines and along the top and bottom by graphs of continuous functions of continuous functions

y = u ( x)

and

y = l (x) .

We equally well could have discussed sets

that are bounded above and below by straight horizontal lines and bounded on the left and right by graphs

x = l (y )

and

x = r (y ) .

These additional sets do not provide anything particularly

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

150

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

important, so we do not discuss them. However, there are times when it is helpful to work with geometric sets with the roles of horizontal and vertical reversed. We accomplish this with the following denition.

Denition 5.17:
Let

be a subset of

for which the point

R2 . By the symmetric image (y, x) S. [a, b] [c, d]

of

we mean the set

of all points

(x, y ) R2

The symmetric image of a set is just the reection of the set across the the symmetric image of the rectangle is again a rectangle,

y = x line in the plane. Note that [c, d] [a, b] , and therefore the area

of a rectangle is equal to the area of its symmetric image. This has the implication that if the symmetric image of a geometric set is also a geometric set, then they both have the same area. The symmetric image of a geometric set doesn't have to be a geometric set itself. For instance, consider the examples suggested in part (b) of . But clearly rectangles, triangles, and circles have this property, for their symmetric images are again rectangles, triangles, and circles. For a geometric set, whose symmetric image is again a geometric set, there are some additional computational properties of the area of over

as well as the integral of functions

S,

Exercise 5.8.5
Suppose

and we present them in the following exercises.

is a closed geometric set, which is determined by a closed interval

[a, b]

and two

bounding functions

u (x)

and

l (x) . a, b

Suppose the symmetric image and two bounding functions

^ ^

of

is also a closed geometric and ^

set, determined by an interval

^ ^

u ( x)

l ( x) .

a. Make up an example to show that the numbers the numbers functions

and ^

need not have anything to do with

and

b,

and that the functions

and

need not have anything to do with the

and

l. S
^ have the same area. HINT: use the fact that the area of a geometric set

b. Prove that

and

is approximately equal to the sum of the areas of certain rectangles, and then use the fact that the area of the symmetric image of a rectangle is the same as the area of the rectangle. c. Show that for every point have ^ ^ ^

(x, y ) S,
of

we must have

a y b ,
^

and for every such

y,

we must

l (y ) x u (y ) .

HINT: If

d. (d) Prove that the area

A (S )

(x, y ) S, then (y, x) S . S is given by the formula

b u(x)
^

^ ^

u (y )

A (S ) =
a
(See 5.9, p. 139.) e. Let

1 dydx =
l(x)

1 dxdy.

(5.104)

l (y )

be the right triangle having vertices

(a, c) , (b, c) ,

and

(b, d) ,
^

where

d > c.

Describe the

symmetric image of

S;

i.e., nd the corresponding

^ ^ ^

a, b , u,
d

and

l .

Use part (d) to obtain the

following formulas for the area of

S: 1 dsdt = 1 dsdt.
c
td b+ c d (ab)

tb d+ a b (cd)

A (S ) =
a c

(5.105)

Exercise 5.8.6
a. Prove that if

S1

and

S2

are geometric sets whose symmetric images are again geometric sets,

then the symmetric image of the geometric set

S1 S2

is also a geometric set.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

151

b. Suppose

is a closed geometric set that is contained in a closed geometric set

S.

Assume

that both the symmetric images interval ^ ^

and

are also geometric sets. If ^

is determined by an

a, b

and two bounding functions

and

l,

prove that

^ ^

u (s)

A (T ) =
where if

T (t, s) dtds,

(5.106)

l (s)

T is the indicator function of the set T ; i.e., T (t, s) = 1 if (t, s) T, and T (t, s) = 0 (t, s) / T. HINT: See the proof of Theorem 5.16, p. 139, give names to all the intervals and {Si } is a partition of a geometric set S, and suppose the symmetric images of S and Si 's are also geometric sets. Suppose h is a step function that is the constant ai on n 0 0 , and therefore that element Si of the partition {Si }. Prove that h = i=1 ai S Si S
b u(t)
^

bounding functions, and in the end use part (d) of the preceding exercise. c. Suppose all the the

^ ^

u (s)

h=
S
HINT: Use part (b). d. Let

h (t, s) dsdt =
a l(t)

h (t, s) dtds.

(5.107)

l (s)

be a geometric set whose symmetric image

is also a geometric set, and suppose

is

a continuous function on

S.

Show that ^ ^ ^

u(t)

u (s)

f=
S a l(t)

f (t, s) dsdt =

f (t, s) dtds.

(5.108)

l (s)
145

HINT: Make use of the fact that the step functions constructed in Theorem 5.19, p. satisfy the assumptions of part (c). Then take limits. e. Let

be the triangle in part (e) of the preceding exercise. If

is a continuous function on

S,

show that the integral of

over

is given by the formulas

t b d+ a b (cd)

f=
S a c

f (t, s) dsdt =
c
d b+ s cd (ab)

f (t, s) dtds.

(5.109)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

152

CHAPTER 5. INTEGRATION, AVERAGE BEHAVIOR

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

Chapter 6

Integration over Smooth Curves in the Plane

6.1 Integration Over Smooth Curves in the Plane C=2 r
1
In this chapter we will dene what we mean by a smooth curve in the plane and what is meant by its arc length. These denitions are a good bit more tricky than one might imagine. Indeed, it is the subtlety of the denition of arc length that prevented us from dening the trigonometric functions in terms of wrapping the real line around the circle, a denition frequently used in high school trigonometry courses. Having made a proper denition of arc length, we will then be able to establish the formula of a circle of radius

C = 2r

for the circumference

r. R2 C,
and we will on occasion want to carefully distinguish between

By the plane, we will mean

these two notions of the plane, i.e., two real variables represent the same point in the plane. As of it as a vector in

In various instances, for clarity, we will use notations like

x + iy,

it

x and y as opposed to one complex variable z = x + iy. x + iy and (x, y ) , remembering that both of these is a single complex number, while as (x, y ) we may think

R2

having a magnitude and, if nonzero, a direction.

We also will dene in this chapter three dierent kinds of integrals over such curves. The rst kind, called integration with respect to arc length, will be completely analogous to the integral dened in for functions on a closed and bounded interval, and it will only deal with functions whose domain is the set consisting of the points on the curve. The second kind of integral, called a contour integral, is similar to the rst one, but it emphasizes in a critical way that we are integrating a complex-valued function over a curve in the complex plane

and not simply over a subset of

R2 .

The applications of contour integrals is usually to

functions whose domains are open subsets of the plane that contain the curve as a proper subset, i.e., whose domains are larger than just the curve. The third kind of integral over a curve, called a line integral, is conceptually very dierent from the rst two. In fact, we won't be integrating functions at all but rather a new notion that we call dierential forms. This is actually the beginnings of the subject called dierential geometry, whose intricacies and power are much more evident in higher dimensions than 2. The main points of this chapter include: 1. The denition of a 163),

smooth curve, and the denition of its arc length,

C = 2r
for the circumference of a circle of radius

2. the derivation of the formula 3. the denition of the 4. the denition of a 5. 6.

(Theorem 6.5, p.

integral with respect to arc length, contour integral, the denition of a line integral, and Green's Theorem (Theorem 6.15, Green, p. 177).
content is available online at <http://cnx.org/content/m36224/1.2/>. Available for free at Connexions <http://cnx.org/content/col11249/1.1>

1 This

153

154

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

2

6.2 Smooth Curves in the Plane

are good solid analytical arguments, with lots of

Our rst project is to make a satisfactory denition of a smooth curve in the plane, for there is a good bit of subtlety to such a denition. In fact, the material in this chapter is all surprisingly tricky, and the proofs

's

and references to earlier theorems.

Whatever denition we adopt for a curve, we certainly want straight lines, circles, and other natural geometric objects to be covered by our denition. Our intuition is that a curve in the plane should be a 1-dimensional subset, whatever that may mean. At this point, we have no denition of the dimension of a general set, so this is probably not the way to think about curves. On the other hand, from the point of view of a physicist, we might well dene a curve as the trajectory followed by a particle moving in the plane, whatever that may be. As it happens, we do have some notion of how to describe mathematically the trajectory of a moving particle. We suppose that a particle moving in the plane proceeds in a continuous manner relative to time. That is, the position of the particle at time t is given by a continuous function f (t) = x (t) + iy (t) (x (t) , y (t)) , as t ranges from time a to time b. A good rst guess at a denition of a curve joining two points z1 and z2 might well be that it is the range C of a continuous function f that is dened on some closed bounded interval [a, b] . This would be a curve that joins the two points z1 = f (a) and z2 = f (b) in the plane. Unfortunately, this is also not a satisfactory denition of a curve, because of the following surprising and bizarre mathematical example, rst discovered by Guiseppe Peano in 1890.

6.1: THE PEANO CURVE The so-called Peano curve

[0, 1] ,
whose range is the entire unit square

is a continuous function in

dened on the

interval

[0, 1] [0, 1] f

R2 .

Be careful to realize that we're talking about the range of

and not its graph. The graph of a real-

valued function could never be the entire square. This Peano function is a complex-valued function of a real variable. Anyway, whatever denition we settle on for a curve, we do not want the entire unit square to be a curve, so this rst attempt at a denition is obviously not going to work. Let's go back to the particle tracing out a trajectory. The physicist would probably agree that the particle should have a continuously varying velocity at all times, or at nearly all times, i.e., the function the position of the particle, and that's just the derivative

should

be continuously dierentiable. Recall that the velocity of the particle is dened to be the rate of change of

f'

is never at rest as it traces out the curve, i.e., the derivative once during the time interval from Recall that a function

of f. We might also f ' (t) is never 0. As

assume that the particle a nal simplication, we

could suppose that the curve never crosses itself, i.e., the particle is never at the same position more than

t=a

to

t = b.

In fact, these considerations inspire the formal denition

of a curve that we will adopt below. open interval

[a, b] and continuously dierentiable on the [a, b] . And, if there exists a partition {t0 < t1 < ... < tn } of [a, b] such that f is smooth on each subinterval [ti1 , ti ] , then f is called piecewise smooth on [a, b] . Although the derivative of a smooth function is only dened and continuous on the open interval (a, b) , and
that is continuous on a closed interval

(a, b)

is called a smooth function on

hence possibly is unbounded, it follows from part (d) of Exercise 5.7.2 that this derivative is improperlyintegrable on that open interval. We recall also that just because a function is improperly-integrable on an open interval, its absolute value may not be improperly-integrable. Before giving the formal denition of a smooth curve, which apparently will be related to smooth or piecewise smooth functions, it is prudent to present an approximation theorem about smooth functions. Exercise 3.7.5 asserts that every continuous function on a closed bounded interval is the uniform limit of a sequence of step functions. We give next a similar, but stronger, result about smooth functions. It asserts that a smooth function can be approximated almost uniformly by piecewise linear functions.

Theorem 6.1:
Let

be a smooth function on a closed and bounded interval

[a, b] ,

and assume that

|f ' |

is

improperly-integrable on the open interval function

(a, b) .

Given an

> 0,

there exists a piecewise linear

for which

2 This

content is available online at <http://cnx.org/content/m36225/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

155

1. 2.

|f (x) p (x) | < for all x [a, b] . b |f ' (x) p' (x) | dx < . a f
and

That is, the functions

are close everywhere, and their derivatives are close on average in

the sense that the integral of the absolute value of the dierence of the derivatives is small.

Proof:

Because

is continuous on the compact set

such that if Because

[a, b] , it is uniformly continuous. Hence, let > 0 be x, y [a, b] , and |x y | < , then |f (x) f (y ) | < /2. |f ' | is improperly-integrable on the open interval (a, b) , we may use part (b) of Exera+ ' b

' > 0, which may also be chosen to be < , such that a |f ' | + b' |f ' | < /2, ' and we x such a . ' ' ' Now, because f is uniformly continuous on the compact set a + , b , there exists an ' ' ' > 0 such that |f (x) f (y ) | < /4 (b a) if x and y belong to a + , b ' and |x y | < . ' ' Choose a partition {x0 < x1 < ... < xn } of [a, b] such that x0 = a, x1 = a + , xn1 = b , xn = b, and xi xi1 < min (, ) for 2 i n 1. Dene p to be the piecewise linear function on [a, b] whose graph is the polygonal line joining the n + 1 points (a, f (x1 )) ,{(xi , f (xi ))} for 1 i n 1, and (b, f (xn1 )) . That is, p is constant on the outer subintervals [a, x1 ] and [xn1 , b] determined by the partition, and its graph between x1 and xn1 is the polygonal line joining the points {(x1 , f (x1 )) , ..., (xn1 , f (xn1 ))}. For example, for 2 i n 1, the function p has the
cise 5.7.2 to nd a form

p (x) = f (xi1 ) +
on the interval

f (xi ) f (xi1 ) (x xi1 ) xi xi1

(6.1)

[xi1 , xi ] . So, p (x) lies between the numbers f (xi1 ) and f (xi ) for all i. Therefore,
(6.2)

|f (x) p (x) | |f (x) f (xi ) | + |f (xi ) l (x) | |f (x) f (xi ) | + |f (xi ) f (xi1 ) | < .
Since this inequality holds for all Next, for

i, part (1) is proved. 2 i n 1, and for each x (xi1 , xi ) , we have p' (x) = (f (xi ) f (xi1 )) / (xi xi1 ) , which, by the Mean Value Theorem, is equal to f ' (yi ) for some yi (xi1 , xi ) . So, for each such x (xi1 , xi ) , we have |f ' (x) p' (x) | = |f ' (x) f ' (yi ) |, and this is less than /4 (b a) , because |x yi | < . On the two outer intervals, p (x) is a constant, ' so that p (x) = 0. Hence,
b a

|f ' p' |

= = <

xi n ' ' i=1 xi1 |f p | x1 b n1 |f ' | + i=2 |f ' p' | + xn1 |f ' | a xn1 b a+ ' 1 |f ' | + b' |f ' | + 4(b a) x1 a

(6.3)

The proof is now complete.

6.2: REMARK It should be evident that the preceding theorem can easily be generalized to a piecewise
smooth function subinterval integrable

f, i.e., a function that is continuous on [a, b] , continuously dierentiable on each (ti1 , ti ) of a partition {t0 < t1 < ... < tn }, and whose derivative f ' is absolutely on (a, b) . Indeed, just apply the theorem to each of the subintervals (ti1 , ti ) , and then

carefully piece together the piecewise linear functions on those subintervals. Now we are ready to dene what a smooth curve is.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

156

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

Denition 6.1:
By a

z2 in the plane, we mean a set C C that : [a, b] C, where [a, b] is a bounded closed interval in R, ' where z1 = (a) and z2 = (b) , and satisfying (t) = 0 for all t (a, b) . 2 More generally, if : [a, b] R is 1-1 and piecewise smooth on [a, b] , and if {t0 < t1 < ... < tn } ' is a partition of [a, b] such that (t) = 0 for all t (ti1 , ti ) , then the range C of is called a piecewise smooth curve from z1 = (a) to z2 = (b) . In either of these cases, is called a parameterization of the curve C.
from a point

smooth curve

z1

to a dierent point

is the range of a 1-1, smooth, function

Note that we do not assume that made you think we would.

|' | is improperly-integrable,

though the preceding theorem might have

6.3: REMARK Throughout this chapter we will be continually faced with the fact that a given curve
1 : [a, b] C
is a parameterization, and

can have many dierent parameterizations. Indeed, if if

g : [c, d] [a, b]

is a smooth function having a nonzero derivative, then

2 (s) = 1 (g (s))

is another parameterization of

C.

Since our denitions and proofs about curves often involve a

parametrization, we will frequently need to prove that the results we obtain are independent of the parameterization. The next theorem will help; it shows that any two parameterizations of connected exactly as above, i.e., there always is such a function

are

Let 1 : [a, b] C and 2 : [c, d] C be two parameterizations of a piecewise smooth curve C joining z1 to z2 . Then there exists a piecewise smooth function g : [c, d] [a, b] such that 2 (s) = 1 (g (s)) for all s [c, d] . Moreover, the derivative g ' of g is nonzero for all but a nite number of points in [c, d] .

Theorem 6.2:

relating

and

2 .

Proof:

function

1 and 2 are continuous and 1-1, it follows from Theorem 3.10, p. 66 that the 1 g = 1 2 is continuous and 1-1 from [c, d] onto [a, b] . Moreover, from Theorem 3.11, p. 67, it must also be that g is strictly increasing or strictly decreasing. Write 1 (t) = u1 (t)+ iv1 (t) (u1 (t) , v1 (t)) , and 2 (s) = u2 (s) + iv2 (s) (u2 (s) , v2 (s)) . Let {x0 < x1 < ... < xp } be a ' partition of [a, b] for which 1 is continuous and nonzero on the subintervals (xj 1 , xj ) , and let {y0 < y1 < ... < yq } be a partition of [c, d] for which ' 2 is continuous and nonzero on the subintervals (yk1 , yk ) . Then let {s0 < s1 < ... < sn } be the partition of [c, d] determined by the nitely many 1 points {yk } {g (xj )}. We will show that g is continuously dierentiable at each point s in the subintervals (si1 , si ) . 1 Fix an s in one of the intervals (si1 , si ) , and let t = 1 (2 (s)) = g (s) . Of course this means that 1 (t) = 2 (s) , or u1 (t) = u2 (s) and v1 (t) = v2 (s) . Then t is in some one of the intervals ' (xj 1 , xj ) , so that we know that ' 1 (t) = 0. Therefore, we must have that at least one of u1 (t) ' ' ' or v1 (t) is nonzero. Suppose it is v1 (t) that is nonzero. The argument, in case it is u1 (t) that ' ' is nonzero, is completely analogous. Now, because v1 is continuous at t and v1 (t) = 0, it follows that v1 is strictly monotonic in some neighborhood (t , t + ) of t and therefore is 1-1 on that 1 interval. Then v1 is continuous by Theorem 3.10, p. 66, and is dierentiable at the point v1 (t) 1 by the Inverse Function Theorem. We will show that on this small interval g = v1 v2 , and this will prove that g is continuously dierentiable at s. 1 Note rst that if 2 ( ) = x + iy is a point on the curve C, then v2 2 (x + iy ) = y. Then, for any [a, b] , we have
Because both

1 v1 v2 g 1 ( )

= = = =

1 1 v1 v2 2 (1 ( )) 1 1 v2 v1 2 (u1 ( ) + iv1 ( )) 1 v1 (v1 ( ))

(6.4)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

157

showing that subintervals smooth.

1 v1 v2 = g 1 = g. Hence g is continuously dierentiable at every point s in the 1 ' ' (si1 , si ) . Indeed g ' ( ) = v1 (v2 ( )) v2 ( ) for all near s, and hence g is piecewise

' ' ' ' 2 (s) = 1 (g (s)) g (s) . Since 2 (s) = 0 for all but a nite number of points s, it follows that g (s) = 0 for all but a nite number of points,

Obviously,

2 (s) = 1 (g (s))

for all

s,

implying that
'

and the theorem is proved.

Corollary 6.1:
Let

and

on the curve

2 be as in the C, we have

theorem. Then, for all but a nite number of points

z = 1 (t) = 2 (s)

Proof:
From the theorem we have that

' ' 2 (s) 1 (t) = . ' ' |1 (t) | |2 (s) |

(6.5)

' ' ' ' ' 2 (s) = 1 (g (s)) g (s) = 1 (t) g (s)

(6.6)
'

for all but a nite number of points for all points because

s (c, d) .
or

where

is dierentiable. And in fact,

g ' (s)

is either

1 v1 v2 (s)

'

g is strictly increasing, so that g (s) 0 g ' (s) = 0 for all but a nite number of s's, ' 1 u 1 u2 (s) , and these are nonzero except for a nite
Also,

number of points. Now the corollary follows by direct substitution.

6.4: REMARK If we think of

'

(t) = x' (t) , y ' (t)

as a vector in the plane

R2 ,

then the corollary

asserts that the direction of this vector is independent of the parameterization, at least at all but a nite number of points. This direction vector will come up again as the unit tangent of the curve. The adjective smooth is meant to suggest that the curve is bending in some reasonable way, and specically it should mean that the curve has a tangent, or tangential direction, at each point. We give the denition of tangential direction below, but we note that in the context of a moving particle, the tangential direction is that direction in which the particle would continue to move if the force that is keeping it on the curve were totally removed. If the derivative

' (t) = 0,

then this vector is the velocity vector, and its

direction is exactly what we should mean by the tangential direction. The adjective piecewise will allow us to consider curves that have a nite number of points where there is no tangential direction, e.g., where there are corners. We are carefully orienting our curves at the moment. A curve the same curve from

from

z1

to

z2

is being distinguished from

z2

to

z1 ,

even though the set

is the same in both instances. Which way we traverse

a curve will be of great importance at the end of this chapter, when we come to Green's Theorem.

Denition 6.2:
Let

C,

the range of

: [a, b] C,

be a piecewise smooth curve, and let

a point on the curve. parameterization

We say that the curve

has a tangential direction at

z = (x, y ) = (c) be z, relative to the

if the following limit exists:

lim

tc | (t)

(t) z (t) (c) = lim . z | tc | (t) (c) | R2 ,

to

(6.7)

If this limit exists, it is a vector of length 1 in (relative to the parameterization The curve

and this unit vector is called the unit tangent

has a

unit tangent at the point z if there exists a parameterization for which the

exists.

at

z.

unit tangent at

Exercise 6.2.1

relative to

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

158

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

a. Restate the denition of tangential direction and unit tangent using the plane instead of the b. Suppose

R2

version of the

version. That is, restate the denition in terms of pairs

(x, y )

of real

numbers instead of a complex number is a point where

z.

: [a, b] C is a parameterization of a piecewise smooth curve C, and that t (a, b) is dierentiable with ' (t) = 0. Show that the unit tangent (relative to ' ' the parameterization ) to C at z = (t) exists and equals (t) /| (t) |. Conclude that, except possibly for a nite number of points, the unit tangent to C at z is independent of the C C
be the graph of the function

parameterization. c. Let

f (t) = |t|

for

t [1, 1] .
2

Is

a smooth curve? Is it a

piecewise smooth curve? Does d. Let

have a unit tangent at every point?

be the graph of the function

Is it a piecewise smooth e. Consider the set be dened by

that

f (t) = t2/3 = t1/3 for t [1, 1] . Is C a smooth curve? curve? Does C have a unit tangent at every point? is the right half of the unit circle in the plane. Let 1 : [1, 1] C 1 (t) = cos t , sin t 2 2 , sin t3 2 2 ,
(6.8)

and let

2 : [1, 1] C

be dened by

2 (t) = cos t3
Prove that

(6.9)

f.

1 and 2 are both parameterizations of C. Discuss the existence of a unit tangent (1, 0) = 1 (0) = 2 (0) relative to these two parameterizations. Suppose : [a, b] C is a parameterization of a curve C from z1 to z2 . Dene on [a, b] by (t) = (a + b t) . Show that is a parameterization of a curve from z2 to z1 .
at the point

Exercise 6.2.2
a. Suppose

is a smooth, real-valued function dened on the closed interval

[a, b] , and let C R2

b.

f. Show that C is a smooth curve, and nd a natural parameterization : [a, b] C of C. What is the unit tangent to C at the point (t, f (t))? Let z1 and z2 be two distinct points in C, and dene : [0, 1] c by (t) = (1 t) z1 + tz2 . Show that is a parameterization of the straight line from the point z1 to the point z2 .
be the graph of Consequently, a straight line is a smooth curve. (Indeed, what is the denition of a straight

c. Dene a function d. e.

: [r, r] R2 by (t) = t, r2 t2 . Show that the range C of is a smooth curve, and that is a parameterization of C. it Dene on [0, /2) by (t) = e . For what curve is a parametrization? Let z1 , z2 , ..., zn be n distinct points in the plane, and suppose that the polygonal line joing
these points in order never crosses itself. Construct a parameterization of that polygonal line.

line?)

f. Let

be a piecewise smooth geometric set determined by the interval

[a, b] z2 ,

and the two

piecewise smooth bounding functions

u and l. Suppose z1 a
^

and

z2

are two points in the interior

S0

of

S.

Show that there exists a piecewise smooth curve

C
^

joining

z1

to

i.e., a piecewise

smooth function g. Let Let

: a, b C

^ ^

with

= z1

and

= z2 ,

that lies entirely in

S0.

C be a piecewise smooth curve, and suppose : [a, b] C is a parameterization of C. [c, d] be a subinterval of [a, b] . Show that the range of the restriction of to [c, d] is a

smooth curve.

Exercise 6.2.3
Suppose

is a smooth curve, parameterized by

= u + iv : [a, b] C.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

159

a. Suppose that

b.

u' (t) = 0 for all t (a, b) . Prove that there exists a smooth, real-valued function a' , b' such that C coincides with the graph of f. HINT: f should 1 be something like v u . ' What if v (t) = 0 for all t (a, b)? f
on some closed interval

Exercise 6.2.4
Let

be the curve that is the range of the function

: [1, 1] C,

where

(t) = t3 + t6 i.
and

a. Is b. Is c. d.

C a piecewise smooth curve? Is it a smooth curve? What a parameterization of C ? Find a parameterization for C by a function : [3, 4] C. Find the unit tangent to C and the point 0 + 0i.

points

z1

z2

does it join?

Exercise 6.2.5
Let

be the curve parameterized by

: [, ] C

dened by

(t) = eit = cos (t) + isin (t) . [0, 1 ] .

a. What curve does

parameterize?

b. Find another parameterization of this curve, but base on the interval

6.3 Arc Length

Suppose velocity

is a piecewise smooth curve, parameterized by a function This velocity is thought of as a vector in

Continuing to think like a physicist, The particle is moving with

we might guess that the length of this curve could be computed as follows.

' (t) .

R2 ,

and as such it has a direction and a magnitude

or speed. The speed is just the absolute value

|' (t) |
b

of the velocity vector

' (t) .

Now distance is speed

multiplied by time, and so a good guess for the formula for the length

of the curve

would be

L=
a

|' (t) | dt.

(6.10)

Two questions immediately present themselves. First, and of primary interest, is whether the function is improperly-integrable on

| ' |

(a, b)?

We know by Exercise 5.7.2 that

'

itself is improperly-integrable, but we

also know from Exercise 5.7.3 that a function can be improperly-integrable on an open interval and yet its absolute value is not. In fact, the answer to this rst question is no (See Exercise 6.3.1 (A curve of innite length).). We know only that dierent functions

|' |

exists and is continuous on the open subintervals of a partition of

[a, b] .

The second question is more subtle. What if we parameterize a curve in two dierent ways, i.e., with two

and

2 ?

How do we know that the two integral formulas for the length have to agree? That is, we

Of course, maybe most important of all to us, we also must justify the physicist's intuition.

must give a rigorous mathematical denition of the length of a smooth curve and show that Formula ((6.10)) above does in fact give the length of the curve. First we deal with the independence of parameterization question.

Let C be a smooth curve joining (distinct) points z1 to z2 in C, and let 1 : [a, b] C and 2 : [c, d] C be two parameterizations of C. Suppose |' 2 | is improperly-integrable on (c, d) . Then |' 1 |is improperly-integrable on (a, b) , and

Theorem 6.3:

' 1 (t)
a

dt =
c

' 2 (s)

ds.

(6.11)

3 This

content is available online at <http://cnx.org/content/m36226/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

160

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

Proof:
We will use Theorem 6.2, p. 156. Thus, let by part (d) of Exercise 5.7.2, Let

1 g = 1 2 ,

and recall that

is continuous on

[c, d]

and continuously dierentiable on each open subinterval of a certain partition of

[c, d] .

Therefore,

g ' is improperly-integrable on (c, d) . {x0 < x1 < ... < xp } be a partition of [a, b] for which ' 1 is continuous and nonzero on the ' subintervals (xj 1 , xj ) . To show that |1 | is improperly-integrable on (a, b) , it will suce to show ' ' this integrability on each subinterval (xj 1 , xj ) . Thus, x a closed interval a , b (xj 1 , xj ) , and ' ' ' ' ' ' let c , d be the closed subinterval of [c, d] such that g maps c , d 1-1 and onto a , b . Hence,
by part (e) of Exercise 5.7.2, we have

b' a'

| ' 1 (t) | dt

= = = = =

d' ' |' 1 (g (s)) |g (s) ds c' ' d ' |' 1 (g (s)) ||g () s| ds c' ' d ' |' 1 (g (s)) g (s) | ds c' d' ' |(1 g ) (s) | ds c' d' | ' 2 (s) | ds c' d ' |2 (s) | ds, c

(6.12)

which, by taking limits as over

(xj 1 , xj )

for every

a' goes to xj 1 and b' goes to xj , shows that |' 1 | is improperly-integrable j, and hence integrable over all of (a, b) . Using part (e) of Exercise 5.7.2
b d

again, and a calculation similar to the one above, we deduce the equality

| ' 1| =
a
and the theorem is proved.

|' 2 |,
c

(6.13)

Exercise 6.3.1: A curve of innite length

Let

: [0, 1] : R2

be dened by

smooth curve that is the range of a. Graph this curve. b. Show that

(0) = (0, 0) , .

and for

t > 0, (t) = (t, tsin (1/t)) .

Let

be the

|' (t) | = =
c. Show that

1 + sin2 (1/t)
1 t

sin(2/t) t

cos2 (1/t) t2

t2 + t2 sin2 (1/t) tsin (2/t) + cos2 (1/t).

(6.14)

1/

|' (t) | dt =

d. Show that there exists an e. Conclude that

1 t

1 sin2 (t) sin (2t) + + cos2 (t) dt. t2 t2 t n

we have

(6.15)

> 0

so that for each positive integer

cos2 (t)

sin (2t) /t > 1/2 for all t such that |t n | < . |' | is not improperly-integrable on (0, 1) .

Deduce that, if Formula ((6.10)) is

correct for the length of a curve, then this curve has innite length. Next we develop a denition of the length of a parameterized curve from a purely mathematical or geometric point of view. Happily, it will turn out to coincide with the physically intuitive denition discussed above. Let of

C.

Let

C be a piecewise smooth curve joining the points z1 and z2 , and let : [a, b] C be a parameterization P = {a = t0 < t1 < ... < tn = b} be a partition of the interval [a, b] . For each 0 j n write

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

161

zj = (tj ) ,

and think about the polygonal trajectory joining these points

{zj }

in order. The length

L P

of

this polygonal trajectory is given by the formula

L P =
j =1

|zj zj 1 |, C.

(6.16)

and this length is evidently an approximation to the length of the curve

Indeed, since the straight line

joining two points is the shortest curve joining those points, these polygonal trajectories all should have a length smaller than or equal to the length of the curve. These remarks motivate the following denition.

Denition 6.3:
Let

: [a, b] C

be a parameterization of a piecewise smooth curve

C,

relative to the parameterization

we mean the number

C C. By the lengthL of L = supP L P , where the supremum is

taken over all partitions

6.5:

of

[a, b] .

Of course, the supremum in the denition above could well equal innity in some cases. Though it is possible for a curve to have an innite length, the ones we will study here will have nite lengths. This is another subtlety of this subject. After all, every smooth curve is a compact subset of

R2 ,

since it is the continuous image of a closed and bounded interval, and we think of compact sets as being nite in various ways. However, this niteness does not necessarily extend to the length of a curve.

Exercise 6.3.2
Let

: [a, b] R2 partitions of [a, b] .

a. If b. If

be a parameterization of a piecewise smooth curve

C,

and let

and

be two

P is ner than Q, i.e., Q P, show that L Q LP . (t) = u (t) + iv (t) , express LP in terms of the numbers u (tj )

and

v (tj ) .

Of course, we again face the annoying possibility that the denition of length of a curve will depend on the parameterization we are using. However, the next theorem, taken together with Theorem 6.3, p. 159, will show that this is not the case.

Theorem 6.4:
If

is a piecewise smooth curve parameterized by

: [a, b] C,

then

L =
a

|' (t) | dt,

(6.17)

Proof:

specically meaning that one of these quantities is innite if and only if the other one is innite. We prove this theorem for the case when

is a smooth curve, leaving the general argument for Hence, assume that of

a piecewise smooth curve to the exercises. We also only treat here the case when leaving the argument for the innite case to the exercises. smooth function on Let

L is nite, = u + iv

also is a

[a, b]

and that

>0

be given. Choose a

L < . partition P = {t0 < t1 < ... < tn }

n

[a, b]

for which

L L P =L j =1
Because such that

| (tj ) (tj 1 ) | < .

that the

(6.18)

is continuous, we may assume by making a ner partition if necessary | (t1 ) (t0 ) | < and | (tn ) (tn1 ) | < . This means that
n1

tj 's

are

L
j =2

| (tj ) (tj 1 ) | < 3.

(6.19)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

162

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

The point of this step (trick) is that we know that

'

is continuous on the open interval

(a, b) ,

but

we will use that it is uniformly continuous on the compact set

[t1 , tn1 ] . Of

course that means that

| ' |

is integrable on that closed interval, and in fact one of the things we need to prove is that

|' |

is improperly-integrable on the open interval Now, because such that

(a, b) .

' is uniformly continuous on the closed interval [t1 , tn1 ] , there exists a > 0 |' (t) ' (s) | < if |t s| < and t and s are in the interval [t1 , tn1 ] . We may assume, again by taking a ner partition if necessary, that the mesh size of P is less than this . Then, using part (f ) of Exercise 5.3.3, we may also assume that the partition P is such that
tn1 n1

|
t1
no matter what points

|' (t) | dt
j =2

|' (sj ) | (tj tj 1 ) | <

(6.20)

sj

in the interval

(tj 1 , tj ) are chosen.

So, we have the following calculation,

in the middle of which we use the Mean Value Theorem on the two functions

and

v.

= = 3 + |

tn1 |' (t) | dt| t1 n 1 |L j =2 | (tj ) (tj 1 ) | t n1 + | j =2 | (tj ) (tj 1 ) | t1n1 |' (t) | dt| g n1 3 + | j =2 | (tj ) (tj 1 ) | t1n1 |' (t) | dt| tn1 n1 |' j =2 |u (tj ) u (tj 1 ) + i v (tj ) v (tj 1 ) | t1

| L

(t) | dt|

3 + |

n1 j =2

(u (tj ) u (tj 1 )) + (v (tj ) v (tj 1 ))

+ = + = <
This implies that

gn1 |' (t) | dt| t1 2 2 n1 3 + | j =2 (u' (sj )) + (v ' (rj )) (tj tj 1 ) t t1n1 |' (t) | dt| 2 2 n1 3 + | j =2 (u' (sj )) + (v ' (sj )) (tj tj 1 ) t t1n1 |' (t) | dt| 2 2 2 2 n1 ' (u (sj )) + (v ' (sj )) | (tj j =2 | (u (sj )) + (v (rj )) t n1 3 + | j =2 |' (sj ) | (tj tj 1 ) t1n1 |' (t) | dt| 2 2 2 2 n1 ' (u (sj )) + (v ' (sj )) | (tj j =2 | (u (sj )) + (v (rj ))

(6.21)

t j 1 ) t j 1 )

4 +

n1 j =2

(tj tj 1 ) (u' (sj ))2 +(v ' (rj ))2 + (u' (sj ))2 +(v ' (sj ))2 ' ' ' ' n1 |v (rj )v (sj )||v (rj )+v (sj )| + j =2 (tj tj 1 ) |v ' (rj )|+|v ' (sj )| n1 ' ' 4 + j =2 |v (rj ) v (sj ) | (tj tj 1 ) n1 4 + j =2 |' (rj ) ' (sj ) | (tj tj 1 ) n1 4 + j =2 (tj tj 1 ) 4 + (tn1 t1 ) (4 + b a) .

|(v ' (rj )) (v ' (sj )) |

tn1

L (4 + b a)
t1

|' | L + (4 + b a) .

(6.22)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

163

If we now let

t1

approach

and

tn1

approach

b,

we get

L (4 + b a)
a
which completes the proof, since

|' | L + (4 + b a) ,

(6.23)

is arbitrary.

Exercise 6.3.3
a. Take care of the piecewise case in the preceding theorem. b. Take care of the case when

is innite in the preceding theorem.

We now have all the ingredients necessary to dene the length of a smooth curve.

Denition 6.4:
Let

be a piecewise smooth curve in the plane. The

length or arc lengthL L (C ) of C

is dened

by the formula

L (C ) = L = supL P,
P
where If

(6.24)

is any parameterization of

C. C,
we will denote by

and

are two points on a piecewise smooth curve

L (z, w)

the arc

length of the portion of the curve between

6.6: REMARK According to Theorem 6.3, p.

L=

and

w.

159 and Theorem 6.4, p. 161, we have the following

formula for the length of a piecewise smooth curve:

|' (t) | dt,

a

(6.25)

where

is any parameterization of

C.

C from z1 to z2 is the same as the length C, but thought of as joining z2 to z1 . Nevertheless, let us make the calculation to verify this. If : [a, b] C is a parameterization of this curve from z1 to z2 , then we have seen in part (f ) of exercise 6.1 that : [a, b] C, dened by (t) = (a + b t) , is a parameterization of C from z2 to z1 . We just need to check that the two integrals giving the lengths are equal. Thus,
It should come as no surprise that the length of a curve of that same curve

| ' (t) | dt =
a a

|' (a + b t) (1) | dt =
a

|' (a + b t) | dt =
a

|' (s) | ds,

(6.26)

where the last equality follows by changing variables, i.e., setting goals. TRUMPETS?

t = a + b s.

We can now derive the formula for the circumference of a circle, which was one of our main

Theorem 6.5:
Let

Proof:

be a circle of radius

in the plane. Then the length of

is

2r.

Let the center of the circle be denoted by the function

on the interval

[0, ]

by

(h, k ) . We can parameterize the top half of the circle by (t) = h + rcos (t) + i (k + rsin (t)) . So, the length of this

half circle is given by

L=
0
the total length is

|' (t) | dt =
0

| rsin (t) + ircos (t) | dt =

0

r dt = r. r,

(6.27) and hence

The same kind of calculation would show that the lower half of the circle has length

2r.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

164

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

The integral formula for the length of a curve is frequently not much help, especially if you really want

to know how long a curve is. The integrals that show up are frequently not easy to work out.

Exercise 6.3.4
a. Let

C be the : [0, 1] C
of this curve.

portion of the graph of the function

y = x2

between

be the parameterization of this curve given by by

x = 0 (t) = t + t2 i.

and

x = 1.

Let

Find the length

b. Dene

: [0, ] C

(t) = acos (t) + ibsin (t) .

What curve does

parameterize, and

can you nd its length?

6.4 Integration with Respect to Arc Length

whose domain is the curve.

We introduce next what would appear to be the best parameterization of a piecewise smooth curve, i.e., a parameterization by arc length. We will then use this parameterization to dene the integral of a function

Theorem 6.6:
Let

be a piecewise smooth curve of nite length

joining two distinct points

z1

to

there exists a parameterization

(u)
Let

Proof:

is equal to

|u t|

for all

: [0, L] C t < u [0, L] .

for which the arc length of the curve joining

z2 . Then (t) to

: [a, b] C

be a parameterization of

C.

Dene a function

F : [a, b] [0, L]

by

F (t) =
a
In other words,

|' (s) | ds. C

that joins the points

(6.28)

F (t)

is the length of the portion of

z1 = (a)

and

(t) .

By the Fundamental Theorem of Calculus, we know that the function

is continuous on

the entire interval partition

[a, b] and is continuously dierentiable on every subinterval (ti1 , ti ) of the P determined by the piecewise smooth parameterization . Moreover, F ' (t) = |' (t) | > 0 for all t (ti1 , ti ) , implying that F is strictly increasing on these subintervals. Therefore, if we write si = F (ti ) , then the si 's form a partition of the interval [0, L] , and the function F : (ti1 , ti ) (si1 , si ) is invertible, and its inverse F 1 is continuously dierentiable. It follows then 1 that = F : [0, L] C is a parameterization of C. The arc length between the points (t) 1 and (u) is the arc length between F (t) and F 1 (u) , and this is given by the formula
F 1 (u) F 1 (t)

|' (s) | ds = = =

F 1 (u) a

|' (s) | ds

F F 1 (u) F F u t,

F 1 (t) |' a 1

(s) | ds
(6.29)

(t)

which completes the proof.

Corollary 6.2:
If

is the parameterization by arc length of the preceding theorem, then, for all

t (si1 , si ) ,

we

have

| ' (s) | = 1.

4 This

content is available online at <http://cnx.org/content/m36228/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

165

Proof:
We just compute

| ' (s) | = = = = =
as desired.

| F 1 (s) | |' F 1 (s) F 1 (s) |

(6.30)
'

'

1 |' F 1 (s) || F ' (F 1 (s)) | 1 |' f 1 (s) | |' (f 1 (s))|

1,

We are now ready to make the rst of our three denitions of integral over a curve. pretty easy. Suppose

This rst one is

is a piecewise smooth curve joining

z1

to

z2

of nite length

Recall that this means that there is a 1-1 function points

from the interval

L, parameterized by arc length. [0, L] onto C that satises the

condidition that the arc length betweenthe two points

(t) and (s) is exactly the distance between the t and s. We can just identify the curve C with the interval [0, L] , and relative distances will correspond perfectly. A partition of the curve C will correspond naturally to a partition of the interval [0, L] . A step function on the dcurve will correspond in an obvious way to a step function on the interval [0, L] , and the
formula for the integral of a step function on the curve is analogous to what it is on the interval. Here are the formal denitions:

L joining distinct points, and let : [0, L] C C by arc length. By a partition of C we mean a set {z0 , z1 , ..., zn } of points on C such that zj = (tj ) for all j, where the points {t0 < t1 < ... < tn } form a partition of the interval [0, L] . The portions of the curve between the points zj 1 and zj , i.e., the set (tj 1 , tj ) ,
Let

Denition 6.5:
C

be a piecewise smooth curve of nite length

be a parameterization of

are called the A

elements of the partition. step fucntion on C is a real-valued

of

function

on

for which there exists a partition

{z0 , z1 , ..., zn } and zj .

such that

h (z )

is a constant

aj

on the portion of the curve between

zj 1

Before dening the integral of a step function on a curve, we need to establish the usual consistency result, encountered in the previous cases of integration on intervals and integration over geometric sets, the proof of which this time we put in an exercise.

Exercise 6.4.1
Suppose

exist two partitions

L, and assume that there {z0 , z1 , ..., zn } and {w0 , w1 , ..., wm } of C such that h (z ) is a constant ak on the portion of the curve between zk1 and zk , and h (z ) is a constant bj on the portion of the curve between wj 1 and wj . Show that
is a function on a piecewise smooth curve of nite length

ak L (zk1 , zk ) =
k=1
HINT: Make use of the fact that

bj L (wj 1 , wj ) .
j =1

(6.31)

is a step function on the interval

[0, L] .

Now we can make the denition of the integral of a step function on a curve.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

166

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

Denition 6.6:
respect to arc length of h over C
Let

be a step function on a piecewise smooth curve is denoted by

C of nite length L. h ( s ) ds, and is dened by C aj L (zj 1 , zj ) ,

The

integral, with

h (s) ds =
C
where

(6.32)

j =1

between

{z0 , z1 , ..., zn } zj 1 and zj .

is a partition of

for which

h (z )

is the constant

aj

on the portion of

Of course, integrable functions on integrable function.

with respect to arc length will be dened to be functions that are

uniform limits of step functions. Again, there is the consistency issue in the denition of the integral of an

Exercise 6.4.2
a. Suppose sequence b.

{hn }

is a sequence of step functions on a piecewise smooth curve

and assume that the sequence

{hn }

converges uniformly to a function

C f.

of nite length, Prove that the

{ C hn (s) ds} is a convergent sequence of real numbers. Suppose {hn } and {kn } are two sequences of step functions on a piecewise smooth curve C of nite length l, and that both sequences converge uniformly to the same function f. Prove
that

lim
C

hn (s) ds = lim
C

kn (s) ds.

(6.33)

Denition 6.7:
integrable with respect to arc length on C if it is the uniform limit of step functions on C. The integral with respect to arc length of an integrable function f on C is again denoted
f (s) ds, C
and is dened by Let

be a piecewise smooth curve of nite length

L.

A function

with domain

is called by

f (s) ds = lim
C
where

hn (s) ds,
C

(6.34)

{hn }

is a sequence of step functions that converges uniformly to

on

C.

In a sense, we are simply identifying the curve function

with the interval

[0, L] by means of the 1-1 parameterizing

. C

Theorem 6.7:
f

The next theorem makes this quite plain.

Let

be a piecewise smooth curve of nite length is an integrable function on

L,
L

and let

be a parameterization of

by arc

length. If

C,

then

f (s) ds =

f ( (t)) dt.
0

(6.35)

Proof:
First, if for every Then,

h j.

is a step function on Note that

the portion of the curve between

is a

C, let {zj } be a partition of C for which h (z ) is a constant aj on zj 1 and zj . Let {tj } be the partition of [0, L] for which zj = (tj ) step function on [0, L] , and that h (t) = aj for all t (tj 1 , tj ) .
N j =1 n j =1

h (s) ds = = = =

aj L (zj 1 , zj )
(6.36)

aj L ( (tj 1 ) , (tj )) n j =1 aj (tj tj 1 ) L h (t) dt, 0

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

167

which proves the theorem for step functions.

Finally, if

on

f = limhn [0, L] , and so

is an integrable function on

C,

then the sequence

{hn }

converges uniformly to

f (s) ds = = =

lim lim
l 0 L 0

h C n

(s) ds
(6.37)

hn ( (t)) dt

f ( (t)) dt,

where the nal equality follows from Theorem 5.6, p. 127. Hence, Theorem 6.7, p. 166 is proved. Although the basic denitions of integrable and integral, with respect to arc length, are made in terms of the particular parameterization

of the curve, for computational purposes we need to know how to evaluate

these integrals using dierent parameterizations. Here is the result:

Theorem 6.8:
C be a C. If f is

Let of

piecewise smooth curve of nite length an integrable function on

L,

and let

: [a, b] C

be a parameterization

C.

Then

f (s) ds =

f ( (t)) |' (t) | dt.

a

(6.38)

p. 166, we write

: [0, L] C for a parameterization of C by arc length. As in g : [a, b] [0, L] for 1 . Just as in that proof, we ' smooth function on the interval [a, b] . Hence, recalling that | (t) | = 1
Write nite number of points, the following calculation is justied:

Proof:

the proof to Theorem 6.7, know that and

g is a piecewise g ' (t) > 0 for all but a

f (s) ds = = = = = = =
L 0

L 0

f ( (t)) dt

f ( (t)) | ' (t) | dt

(6.39)

b f ( (g (u))) | ' (g (u)) |g ' (u) du a b f ( (g (u))) | ' (g (u)) ||g ' (u) | du a b f ( (u)) | ' (g (u)) g ' (u) | du a ' b f ( (u)) | ' gamma g (u) | du a b f ( (u)) |' (u) | du, a

as desired.

Exercise 6.4.3
Let a. b. c.

(0, 1) and (1, 2) . 2 C. Find the arc length parameterization : 0, 2 Let f be the function on this curve given by f (x, y ) = x y. Compute f (s) ds. C Let f be the function on this curve that is dened by f (x, y ) is the distance from (x, y ) the point (0, 3) . Compute f (s) ds. c
be the straight line joining the points

to

The nal theorem of this section sums up the properties of integrals with respect to arc length. There are no surprises here.

Theorem 6.9:
Let

be a piecewise smooth curve of nite length

L,

and write

I (C )

for the set of all functions

that are integrable with respect to arc length on

C.

Then:

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

168

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

1.

I (C )

is a vector space ovr the real numbers, and

(af (s) + bg (s)) ds = a

C
for all

f (s) ds + b
C C

g (s) ds

(6.40)

2. 3. 4. 5.

f, g I (C ) and all a, b R. f (z ) 0 for all z C, then C f (s) ds 0. If f I (C ) , then so is |f |, and | f (s) ds| C |f (s) | ds. C If f is the uniform limit of functions fn , each of which is in I (C ) , then f I (C ) and f (s) ds = lim C fn (s) ds. C Let {un } be a sequence of functions in I (C ) , and suppose that for each n there is a number mn , for which |un (z ) | mn for all z C, and such that the innite series mn converges. Then the innite series un converges uniformly to an integrable function, and C un (s) ds = u (s) ds. C n
(Positivity) If

Exercise 6.4.4
a. Prove the preceding theorem. Everything is easy if we compose all functions on parameterization b. Suppose

with the

obtaining functions on

[0, L] , f

and then use Theorem 5.6, p. 127.

is a piecewise smooth curve of nite length joining over

z1

and

z2 .
to

Show that the

integral with respect to arc length of a function as being a curve from

is the same whether we think of

z1

to

z2

or, the other way around, a curve from

z2

z1 .

6.7: REMARK Because of the result in part (b) of the preceding exercise, we speak of integrating
over

z1

to

C  when we are integrating with respect to arc length. We z2 , since the direction doesn't matter. This is in marked

do not speak of integrating from contrast to the next two kinds of

integrals over curves that we will discuss. here is one nal bit of notation. Often, the curves of interest to us are graphs of real-valued functions. If graph (g ) .

g : [a, b] R

is a piecewise smooth function, then its graph

is a piecewise smooth

curve, and we write

f (s) ds graph(g )

for the integral with respect to arc length of

over

C =

6.5 Contour Integrals

the curve

We discuss next what appears to be a simpler notion of integral over a curve. In this one, we really do regard

as a subset of the complex plane as opposed to two-dimensional real space; we will be integrating

complex-valued functions; and we explicitly think of the parameterizations of the curve as complex-valued functions on an interval reverse, i.e., the same set

Denition 6.8:
Let

[a, b] . Also, in this denition, a curve C from z1 C thought of as a curve from z2 to z1 .

to

z2

will be distinguished from its

z1 to z2 in the plane C, parameterized by a (complex : [a, b] C. If f is a continuous, complex-valued function on C, The contour 2 integral of f from z1 to z2 along C will be denoted by C f ( ) d or more precisely by C z z1 f ( ) d,
be a piecewise smooth curve from valued) function and is dendd by

b C z2 z1 f

( ) d =
a

f ( (t)) ' (t) dt.

(6.41)

5 This

content is available online at <http://cnx.org/content/m36230/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

169

6.8: REMARK There is, as usual, the question about whether this denition depends on the parameterization. Again, it does not. See the next exercise. The denition of a contour integral looks very like a change of variables formula for integrals. See Theorem 5.11, Integration by Substitution, p. 133 and part (e) of Exercise 5.7.2. This is an example of how mathematicians often use a true formula from one context to make a new denition in another context. Notice that the only dierence between the computation of a contour integral and an integral with respect to arc length on the curve is the absence of the absolute value bars around the factor

' (t) .

This will make contour integrals more subtle than integrals with respect to arc length, just

as conditionally convergent innite series are more subtle than absolutely convergent ones. Note also that there is no question about the integrability of cise 5.7.2.

f ( (t)) ' (t) ,

because of Exer-

Exercise 6.5.1

is bounded,

'

is improperly-integrable on

(a, b) ,

and therefore so is their product.

a. State and prove the independence of parameterization result for contour integrals. b. Prove that

z2 z1 f

( ) d =
C

z1 z2 f

( ) d.

(6.42)

Just remember how to parameterize the curve in the opposite direction. c. Establish the following relation between the absolute value of a contour integral and a corresponding integral with respect to arc length.

|
C

f ( ) d |
C

|f (s) | ds.

(6.43)

Not all the usual properties hold for contour integrals, e.g., like those in Theorem 6.9, p. 167 above. The functions here, and the values of their contour integrals, are complex numbers, so all the properties of integrals having to do with positivity and inequalities, except for the one in part (c) of Exercise 6.5.1, no longer make any sense. However, we do have the following results for contour integrals, the verication of which is just as it was for Theorem 6.9, p. 167.

Theorem 6.10:
Let

be a piecewise smooth curve of nite length joining

z1 a

to

z2 . b

Then the contour integrals of

continuous functions on 1. If

have the following properties.

and

are any two continuous functions on

C,

and

and

are any two complex numbers,

then

(af ( ) + bg ( )) d = a
C
2. If 3.

f ( ) d + b
C C

g ( ) d.

(6.44)

C of a sequence {fn } of continuous functions, then C f ( ) d = f ( ) d. C n Let {un } be a sequence of continuous functions on C, and suppose that for each n there is a number mn , for which |un (z ) | mn for all z C, and such that the innite series mn converges. Then the innite series un converges uniformly to a continuous function, and un ( ) d = u ( ) d. C C n f lim
is the uniform limit on

In the next exercise, we give some important contour integrals, which will be referred to several times in the sequel. Make sure you understand them.

Exercise 6.5.2
Let

be a point in the complex plane, and let

parameterized by integer

: [, ] : C dened n n Z, dene fn (z ) = (z c) .

by

r be a positive number. Let C be the curve (t) = c + reit = c + rcos (t) + irsin (t) . For each

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

170

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

1. What two points 2. Compute 3. 4. 5.

z1 and z2 does C join, and what happens to z2 as approaches f ( ) d for all integers n, positive and negative. C n What happens to the integrals computed in part (b) when approaches 0? Set = , and compute f ( ) d for all integers n. C n Again, set = . Evaluate cos ( c) d c
and

0?

sin ( c) d. c

(6.45)

HINT: Make use of the innite series representations of the trigonometric functions.

6.6 Vector Fields, Dierential Forms, and Line Integrals

much a real variable one, so that we think of the plane as

We motivate our third denition of an integral over a curve by returning to physics. This denition is very

R2

instead of

C.

A connection between this real

variable denition and the complex variable denition of a contour integral will emerge later.

Denition 6.9:
By a

vector eld

on an open subset from

of

V (x, y ) (P (x, y ) , Q (x, y )) of the vector eld V .

We will also speak of component functions

into

R2 , R2 .

we mean nothing more than a continuous function The functions

and

are called the

components

smooth vector elds, by which we will mean vector elds V

Q
have continuous partial derivatives

both of whose

and

tialQ tialP tialP tialQ , , and tialx tialy tialx tialy

on

(6.46)

U.

6.9:
(x, y )
with magnitude

The idea from physics is to think of a vector eld as a force eld, i.e., something that exerts a force at the point

| V (x, y ) |

and acting in the direction of the vector

V (x, y ) .

For a particle to move within a force eld, work must be done, that is energy must be provided to move the particle against the force, or energy is given to the particle as it moves under the inuence of the force eld. In either case, the basic denition of work is the product of force and distance traveled. More precisely, if a particle is moving in a direction

within a force eld, then the work

done on the particle is the product of the component of the force eld in the direction of

and

the distance traveled by the particle in that direction. That is, we must compute dot products of the vectors

V (x, y )

and

with respect to arc length by

u (x, y ) . Therefore, if a particle is moving along a curve C, parameterized : [0, L] C, and we write (t) = (x (t) , y (t)) , then the work z1 = (0)

to

W (z1 , z2 )

done on the particle as it moves from

z2 = (L)

within the force eld

V,

should intuitively be given by the formula

W (z1 , z2 )

= =
L 0

L 0

<V ( (t)) | ' (t) > dt

(6.47)

P (x (t) , y (t)) x' (t) + Q (x (t) , y (t)) y ' (t) dt

C

P dx + Q dy,

where the last expression is explicitly dening the shorthand notation we will be using.

6 This

content is available online at <http://cnx.org/content/m36232/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

171

The preceding discussion leads us to a new notion of what kind of object should be integrated over a curve.

R2 is denoted by = P dx + Qdy, and is determined by two continuous real-valued functions P and Q on U. We say that is bounded or uniformly continuous if the functions P and Q are bounded or uniformly continuous functions on U. We say that the dierential form is smooth of order k if the set U is open, and the functions P and Q have continuous mixed partial derivatives of order k. If = P dx + Qdy is a dierential form on a set U, and if C is any piecewise smooth curve of nite length contained in U, then we dene the line integral of over C by C
A of

Denition 6.10:

dierential form on a subset U

=
C
where

P dx + Q dy =
C 0

P ( (t)) x' (t) + Q ( (t)) y ' (t) dt, C

by arc length.

(6.48)

(t) = (x (t) , y (t))

6.10: REMARK
x'
and

is a parameterization of

There is no doubt that the integral in this denition exists, because are integrable. Therefore

and

are

continuous functions on the compact set both

on (0, L) . really should be called dierential 1-forms. For instance, an example of a dierential 2-form would look like R dxdy, and in higher dimensions, we could introduce notions of dierential forms of higher and higher orders, e.g., in 3 dimension things like P dxdy + Q dzdy + R dxdz. Because we will always be dealing with R2 , we will have no need for higher order dierential These dierential forms forms, but the study of such things is wonderful. Take a course in Dierential Geometry! Again, we must see how this quantity As usual, it does not.

y'

C, hence bounded, and ' is integrable, P ( (t)) x' (t) + Q ( (t)) y ' (t) is integrable

implying that

depends, if it does, on dierent parameterizations.

Exercise 6.6.1
Suppose a. Let

= P dx + Qdy C

is a dierential form on a subset

of

R2 . U
that joins

be a piecewise smooth curve of nite length contained in

z1

to

z2 .

Prove

that

=
C C
for any parameterization b. Let

P dx + Q dy =
a

P ( (t)) x' (t) + Q ( (t)) y ' (t) dt x (t)

and

(6.49)

: [a, b] C C
^

having components

y (t) . C
but thought

C C

be as in part (a), and let

denote the reverse of

C,

i.e., the same set

of as a curve joining c. Let

z2

to

z1 .

Show that

= ^
c

. C

be as in part (a). Prove that

|
C
where

P dx + Q dy | (MP + MQ ) L, |P |
and

(6.50)

MP

and

MQ

are bounds for the continuous functions

|Q|

on the compact set

C,

and where

is the length of

C.

Example 6.1
The simplest interesting example of a dierential form is constructed as follows. Suppose open subset of

is an

R2 ,

and let

f :U R = df,

be a dierentiable real-valued function of two real variables; called the

i.e., both of its partial derivatives exist at every point IV.) Dene a dierential form

dierential

(x, y ) U. (See of f, by

the last section of Chapter

df =

tialf tialf dx + dy, tialx tialy

(6.51)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

172

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE forms.

i.e.,

P = tialf /tialx

and

Q = tialf /tialy.

These dierential forms

df

are called

exact dierential

6.11: REMARK Not every dierential form is exact, i.e., of the form df. Indeed, determining which

's are df 's boils down to what may be the simplest possible partial dierential equation problem. If is given by two functions P and Q, then saying that = df amounts to saying that f is a
solution of the pair of simultaneous partial dierential equations

tialf =P tialx

and

tialf = Q. tialy

(6.52)

See part (b) of the exercise below for an example of a nonexact dierential form. Of course if a real-valued function if

has continuous partial derivatives of the second order, then Theo-

rem 4.22, Theorem on mixed partials, p. 112 tells us that the mixed partials

fxy and fyx must be equal. So, = P dx + Qdy = df for some such f, Then P and Q would have to satisfy tialP/tialy = tialQ/tialx. Certainly not every P and Q would satisfy this equation, so it is in fact trivial to nd examples of dierential
forms that are not dierentials of functions. A good bit more subtle is the question of whether every dierential form

P dx + Qdy,

for which

tialP/tialy = tialQ/tialx,

is equal to some

df.

Even this is not true in

general, as part (c) of the exercise below shows. The open subset of open sets.

on which the dierential form is dened

plays a signicant role, and, in fact, dierential forms provide a way of studying topologically dierent kinds In fact, although it may seem as if a dierential form is really nothing more than a pair of functions, the concept of a dierential form is in part a way of organizing our thoughts about partial dierential equation problems into an abstract mathematical context. This abstraction is a good bit more enlightening in higher dimensional spaces, i.e., in connection with functions of more than two variables. Take a course in Multivariable Analysis!

Exercise 6.6.2

a. Solve the pair of simultaneous partial dierential equations

tialf =x+y tialx

and

tialf = x y. tialy

(6.53)

b. Show that it is impossible to solve the pair of simultaneous partial dierential equations

tialf =x+y tialx

Hence, conclude that the dierential form any real-valued function c. Let

and

tialf = y3 . tialy
is not the dierential

(6.54)

= (x + y ) dx + y 3 dy

df

of

f.

U be the open subset of R2 that is the complement of the single point (0, 0) . Let P (x, y ) = y/ x2 + y 2 and Q (x, y ) = x/ x2 + y 2 . Show that tialP/tialy = tialQ/tialx at every point of U, but that = P dx + Qdy is not the dierential df of any smooth function f on U. HINT: If P were fx , then f would have to be of the form f (x, y ) = tan1 (x/y ) + g (y ) , where g is some dierentiable function of y. Show that if Q = fy then g (y ) is a constant c. 1 Hence, f (x, y ) must be tan (x/y ) + c. But this function f is not continuous, let alone dierentiable, at the point (1, 0) . Consider limf (1, 1/n) and limf (1, 1/n) .
C

The next thing we wish to investigate is the continuity of

as a function of the curve

C.

This brings out

a signicant dierence in the concepts of line integrals versis integrals with respect to arc length. For the latter, we typically think of a xed curve and varying functions, whereas with line integrals, we typically think of a xed dierential form and variable curves. mind. This is not universally true, but should be kept in

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

173

Theorem 6.11:
Let and let is contained in in

= P dx + Qdy be a xed, bounded, uniformly continuous dierential form on a set U in R2 , C be a xed piecewise smooth curve of nite length L, parameterized by : [a, b] C, that U.
Then, given an

>0

there exists a

>0

such that, for any curve

contained

U,|

^ C

| <

whenever the following conditions on the curve

hold:

1. 2. 3.

C is a piecewise smooth curve of nite length L contained in U, parameterized by : [a, b] C . | (t) (t) | <
b a
^
'

for all

t [a, b] .

|' (t) (t) | dt < .

Proof:
P and Q are bounded on U, let MP and MQ be upper bounds |Q| respectively. Also, since both P and Q are uniformly continuous ' ' ' ' on U, there exists a > 0 such that if | (c, d) c , d | < , then |P (c, d) P c , d | < /4L ' ' and |Q (c, d) Q c , d | < /4L. We may also choose this to be less than both /4MP and
Let be given. Because both for the functions

>0

|P |

and

/4MQ .

Now, suppose ^

is a curve of nite length and that

L,
^

^ parameterized by
'

: [a, b] C ,

and that and

| (t) (t) | <

^

for all

t [a, b] ,

b a

|' (t) (t) | < .

Writing

(t) = (x (t) , y (t))

(t) = 0 =

x (t) , y (t) ,

we have

| |
b a b a

P dx + Q dy
^ ^
'

^ C

P dx + Q dy |
^ ^
'

P ( (t)) x (t) P
^

'

(t) x (t) + Q ( (t)) y ' (t) Q (t)

^
'

y (t) dt|
^
'

|P ( (t)) x' (t) P

b a

(t) x (t) | dt +
^

b a

|Q ( (t)) y ' (t) Q (t)

b a

y (t) | dt

|P ( (t)) P
b a

(t) ||x' (t) | dt +

^

|P
b a

(t) ||x' (t) x (t) | dt

^ ^
'

'

|Q ( (t)) Q (t) ||y ' (t) | dt +

4L b a

|Q (t) ||y ' (t) y (t) | dt

^
'

(6.55)

|x' (t) | dt + MP
b a

b a

|x' (t) x (t) | dt

b a

+
4L

4L b a

|y ' (t) | dt + MQ
b a

|y ' (t) y (t) | dt

^
'

'

|' (t) | dt + MP
b a

|' (t) (t) | dt

b a

+ < <
as desired.

4L

|' (t) | dt + MQ
4

|' (t) (t) | dt

'

+ MP + MQ ,

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

174

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

Again, we have a special notation when the curve

function, then its graph of the dierential form

C is a graph. C is a piecewise smooth curve, and we write P dx + Qdy over the curve C = graph (g ) .

If

g : [a, b] R is a piecewise smooth P dx + Q dy for the line integral graph(g )

As alluded to earlier, there is a connection between contour integrals and line integrals. It is that a single contour integral can often be expressed in terms of two line integrals. Here is the precise statement.

Theorem 6.12:
C C.

Suppose

is a piecewise curve of nite length, and that Let

function on

: [a, b] C

be a parameterization of

f = u + iv is a complex-valued, continuous C, and write (t) = x (t) + iy (t) . Then (v dx + u dy ) .

C
(6.56)

f ( ) d =

(U dx v dy ) +
C

Proof:
We just compute:

f ( ) d

= = = = =
b a b a b a

b a

f ( (t)) ' (t) dt

(u ( (t)) + iv ( (t))) x' (t) + iy ' (t) dt u ( (t)) x' (t) v ( (t)) y ' (t)
(6.57)

+ i v ( (t)) x' (t) + u ( (t)) y ' (t) dt u ( (t)) x' (t) v ( (t)) y ' (t) dt
b a

+i
C

v ( (t)) x' (t) + u ( (t)) y ' (t) dt

C

u dx v dy + i

v dx + u dy,

as asserted.

6.7 Integration Around Closed Curves, and Green's Theorem

Thus far, we have discussed integration over curves joining two distinct points

z1

and

z2 .

Very important

in analysis is the concept of integrating around a closed curve, i.e., one that starts and ends at the same point. There is nothing really new here; the formulas for all three kinds of integrals we have dened will look the same, in the sense that they all are described interms of some parameterization

A parameterization

: [a, b] C

of a closed curve

is just like the parameterization for a curve joining two points, except that

the two points

(a)

and

(b)

are equal.

Two problems are immediately apparent concerning integrating around a closed curve. First, where do we start on the curve, which point is the initial point? And second, which way to we go around the curve? Recall tha if is a parameterization of

: [a, b] C C

is a parameterization of that is the reverse of

C,

then

: [a, b] C,

dened by

(t) = (a + b t) ,

i.e., it goes around the curve in the other direction. If

we are integrating with respect to arc length, this reverse direction won't make a dierence, but, for contour integrals and line integrals, integrating in the reverse direction will introduce a minus sign. The rst question mentioned above is not so dicult to handle. It doesn't really matter where we start on a closed curve; the parameterization can easily be shifted.

Exercise 6.7.1
Let

[a, b] R2

be a piecewise smooth function that is 1-1 except that ^ ^ by

(a) = (b) .
^ and

For each

0 < c < b a, dene : [a + c, b + c] : R2 for b t b + c.

7 This

(t) = (t)

for

a + c t b,

(t) = (t b + a

content is available online at <http://cnx.org/content/m36233/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

175

^ 1. Show that ^ range of 2. Let

is a piecewise smooth function, and that the range

of

coincides with the

. C.
'

be an integrable (with respect to arc length) function on

Show that

b+c

f ( (t)) |' (t) | dt =

a
That is, the integral 3. Let

f
a+c

(t) | (t) | dt. C

(6.58)

f (s) ds

of

with respect to arc length around the closed curve

is

independent of where we start.

be a continuous complex-valued function on

C. f

Show that ^ ^
'

b+c

f ( (t)) (t) dt =
a
That is, the contour integral where we start. 4. Let

'

(t) (t) dt. C

(6.59)

a+c

f ( ) d

of

around the closed curve Prove that ^ ^

'

is independent of

= P dx + Qdy
b
'

be a dierential form on

C. P

b+c

P ( (t)) x (t) + Q ( (t)) y (t) dt =

a
That is, the line integral

'

(t) x (t) + Q (t)

'

y (t) dt.

(6.60)

a+c

of

around

is independent of where we start.

The question of which way we proceed around a closed curve is one that leads to quite intricate and dicult mathematics, at least when we consider totaly general smooth curves. For our purposes it wil, suce to study a special kind of closed curve, i.e., curves that are the boundaries of piecewise smooth geometric sets. Indeed, the intricate part of the general situation has a lot to do with determining which is the inside of the closed curve and which is the outside, a question that is easily settled in the case of a geometric set. Simple pictures make this general question seem silly, but precise proofs that there is a denite inside and a denite outside are dicult, and eluded mathematicians for centuries, culminating in the famous Jordan Curve Theorem, which asserts exactly what our intuition predicts:

Theorem 6.13:

Jordan Curve Theorem

The complement of a closed curve is the union of two disjoint components, one bounded and one unbounded. We dene the bounded component to be the inside of the curve and the unbounded component to be the outside. We adopt the following convention for how we integrate around the boundary of a piecewise smooth geometric set

S.

That is, the curve

CS

will consist of four parts: the lower boundary (graph of the lower

bounding function

l),

the righthand boundary (a portion of the vertical line such a

(the graph of the upper bounding function

x = a).

By

integrating around

x = b), the upper boundary u), and nally the lefthand side (a portion of the vertical line curve CS , we will always mean proceeding counterclockwise around

the curves.

Specically, we move from left to right along the lower boundary, from bottom to top along

the righthand boundary, from right to left across the upper boundary, and from top to bottom along the lefthand boundary. Of course, as shown in the exercise above, it doesn't matter where we start.

Exercise 6.7.2
Let

be the closed piecewise smooth geometric set that is determined by the interval

the two piecewise smooth bounding functions nite length. Suppose the graph of

and

l.

Assume that the boundary

CS

intersects the lines

x=a

and

x=b

at the points

[a, b] of S (a, c)

and has and

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

176

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

(b, d) ,
and suppose that the graph of parameterization HINT: Try

l intersects those lines at the points (a, e) and (b, f ) . : a' , b' CS of the curve CS . ' ' using the interval [a, b + d f + b a + c e] as the domain a , b of .

Find a

The next theorem, though simple to state and use, contains in its proof a combinatorial idea that is truly central to all that follows in this chapter. In its simplest form, it is just the realization that the line integral in one direction along a curve is the negative of the line integral in the opposite direction.

Theorem 6.14:
S1 , ..., Sn

Let

be a collection of closed geometric sets that constitute a partition of a geometric set

S,

and assume that the boundaries of all the

Si 's,
n

as well as the boundary of

S,

have nite length.

Suppose

is a continuous dierential form on all the boundaries

{CSk }.

Then

=
CS k=1 CSk

(6.61)

Proof:
We give a careful proof for a special case, and then outline the general argument. Suppose then that functions

[a, b] and the two bounding u and l, and assume that the boundary CS has nite length. Suppose m (x) is a piecewise b smooth function on [a, b] , satisfying |m' | < , and assume that l (x) < m (x) < u (x) for all a x (a, b) . Let S1 be the geometric set determined by the interval [a, b] and the two bounding functions m and l, and let S2 be the geometric set determined by the interval [a, b] and the two bounding functions u and m. We note rst that the two geometric sets S1 and S2 comprise a partition of the geometric set S, so that this is indeed a pspecial case of the theorem.
is a piecewise smooth geometric set, determined by the interval Next, consider the following eight line integrals: First, integrate from left to write along the graph of left across the graph of

x = b from (b, m (b)) to (b, u (b)) , third, integrate from right to u, fourth, integrate down the line x = a from (a, u (a)) to (a, m (a)) , fth, continue down the line x = a from (a, m (a)) to (a, l (a)) , sixth, integrate from left to right across the graph of l, seventh, integrate up the line x = b from (b, l (b)) to (b, m (b)) , and nally, integfrate from right to left across the graph of m. The rst four line integrals comprise the line integral around the geometric set S2 , and the last four comprise the line integral around the geometric set S1 . On the other hand, the rst and eighth
second, up the line line integrals here cancel out, for one is just the reverse of the other. Hence, the sum total of these eight line integrals, integrals 27, is just the line integral around the boundary

m,

CS

of

S.

Therefore

=
CS
as desired.

+
CS1 CS2

(6.62)

S1 , ..., Sn of S. Let Sk be determined by the [ak , bk ] and the two bounding functions uk and lk . Observe that, if the boundary CSk of Sk intersects the boundary CSj of Sj in a curve C, then the line integral of along C, when it is computed as part of integrating counterclockwise around Sk , is the negative of the line integral along C, when it is computed as part of the line integral counterclockwise around Sj . Indeed, the rst line integral is the reverse of the second one. (A picture could be helpful.) Consequently, when we compute the sum of the line integrals of around the CSk 's, All terms cancel out except those line integrals that ar computed along parts of the boundaries of the Sk 's that intersect no other Sj . But such parts of the boundaries of the Sk 's must coincide with parts of the boundary of S. Therefore, the sum of the line integrals of around the boundaries of the Sk 's equals the line integral of around the boundary of S, and this is precisely what the theorem asserts.
We give next an outline of the proof for a general partition interval

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

177

Exercise 6.7.3
closed geometric sets that constitute a partition of a geometric set

S1 , ..., Sn be a collection of S, and assume that the boundaries of all the Si 's, as well as the boundary of S, have nite length. Suppose f is a continuous complexvalued function on all the boundaries {CSk } as well as on the boundary CS . Then
Prove the analog of Theorem 6.14, p. 176 for contour integrals: Let

f ( ) d =
CS k=1 CSk

f ( ) d.

(6.63)

We come now to the most remarkable theorem in the subject of integration over curves, Green's Theorem. Another fanfare, please!

Theorem 6.15:
Let

Green

be a piecewise smooth, closed, geometric set, let

CS S.

denote the closed curve that is the

boundary of

S,

and assume that

CS

is of nite length. Suppose

= P dx + Qdy

is a continuous

dierential form on

that is smooth on the interior

S0

of

Then

=
CS CS

P dx + Q dy =

tialQ tialP . tialy S tialx

(6.64)

6.12: REMARK The rst thing to notice about this theorem is that it connects an integral around
a (1-dimensional) curve with an integral over a (2-dimensional) set, suggesting a kind of connection between a 1-dimensional process and a 2-dimensional one. does. The second thing to think about is the case when function Such a connection seems to be unexpected, and it should therefore have some important implications, as indeed Green's Theorem

is an exact dierential

df

of a smooth

of two real variables. In that case, Green's Theorem says

tialf tialf dx + dy = tialx tialy CS

which would be equal to 0 if Hence, the integral of

(fyx fxy ) ,
S

(6.65)

f C 2 (S ) , 's,
called

by Theorem 4.22, Theorem on mixed partials, p. 112.

df

around any such curve would be 0. If

may or may not be some other dierential forms It turns out

closed dierential forms, having the property that their

is an open subset of

R2 ,

there

integral around every piecewise smooth curve of nite length in

U is 0, and the study of these closed that are not exact dierential forms df has led to much interesting mathematics. that the structure of the open set U, e.g., how many holes there are in it, is what's

important. Take a course in Algebraic Topology! The proof of Green's Theorem is tough, and we break it into several steps.

Lemma 6.1:
Suppose

Proof:
and

is the rectangle

[a, b] [c, d] .

Then Green's Theorem is true.

CS bounding the rectangle as the union of four straight lines, C1 , C2 , C3 C4 , and we parameterize them as follows: Let : [a, b] C1 be dened by (t) = (t, c) ; let : [b, b + d c] C2 be dened by (t) = (b, t b + c) ; let : [b + d c, b + d c + b a] C3 be dened by (t) = (b + d c + b t, d) ; and let : [b + d c + b a, b + d c + b a + d c] C4 be dened by (t) = (a, b + d c + b a + d t) . One can check directly to see that this parameterizes the boundary of the rectangle S = [a, b] [c, d] .
We think of the closed curve

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

178

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

As usual, we write

(t) = (x (t) , y (t)) .

Now, we just compute, use the Fundamental Theorem

of Calculus in the middle, and use part (d) of Exercise 5.8.5 at the end.

CS

= = = +

C1

+
C3

C2

C4

+
C1

+
C2

P dx + Q dy

P dx + Q dy +
C3
'

P dx + Q dy
C4

+
b a

P dx + Q dy +

P ( (t)) x (t) + Q ( (t)) y ' (t) dt

= = = =
proving the lemma.

b+dc P ( (t)) x' (t) + Q ( (t)) y ' (t) dt b b+dc+ba + b+dc P ( (t)) x' (t) + Q ( (t)) y ' (t) dt b+dc+ba+dc + b+dc+ba P ( (t)) x' (t) + Q ( (t)) y ' (t) dt b b+dc P (t, c) dt + b Q (b, t b + c) dt a b+dc+ba + b+dc P (b + d c + b t, d) (1) dt b+dc+ba+dc + b+dc+ba Q (a, b + d c + b a + d t) (1) b d P (t, c) dt + c Q (b, t) dt a b d a P (t, d) dt c Q (a, t) dt d b (Q (b, t) Q (a, t)) dt a (P (t, d) P (t, c)) dt c d b tialQ (s, t) dsdt c a tialx b d tialP a c tialy (t, s) dsdt tialQ tialP tialx tialy , S

(6.66)

dt

Lemma 6.2:
Suppose

is a right triangle whose vertices are of the form

(a, c) , (b, c)

and

(b, d) .

Then Green's

Theorem is true.

Proof:
for

CS of this triangle as follows: For t [a, b] , set (t) = (t, c) ; t [b, b + d c] , set (t) = (b, t + c b) ; and for t [b + d c, b + d c + b a] , set (t) = (b + d c + b t, b + d c + d t) . Again, one can check that this parameterizes the boundary of the triangle S.
We parameterize the boundary

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

179

Write

(t) = (x (t) , y (t)) .

CS

Again, using the Fundamental Theorem and Exercise 5.8.5, we have

= = +
b a

CS
'

P dx + Q dy

P ( (t)) x (t) + Q ( (t)) y ' (t) dt

= + + =

b+dc P ( (t)) x' (t) + Q ( (t)) y ' (t) dt b b+dc+ba + b+dc P ( (t)) x' (t) + Q ( (t)) y ' (t) dt b b+dc P (t, c) dt + b Q (b, t + c b) dt a b+dc+ba P (b + d c + b t, b + d c + d t) (1) b+dc b+dc+ba Q (b + d c + b t, b + d c + d t) (1) b+dc b d P (t, c) dt + c Q (b, t) dt a b sb a P s, d + a ds b (c d) d d c Q b+ s cd (a b) , s ds d d ds Q (b, s) Q b + s cd (a b) , s c b sb a P s, d + a P (s, c) ds b (c d) d b tialQ d tialx (t, s) dtds c b+ s cd (ab) sb b d+ (cd) tialP a c ab tialy (s, t) dtds tialQ tialP tialx tialy , S

dt dt
(6.67)

which proves Lemma 6.2, p. 178.

Lemma 6.3:
Suppose for all

Proof:

S1 , ..., Sn is a partition of the geometric set S, and that the boundary CSk has nite length 1 k n. If Green's Theorem holds for each geometric set Sk , then it holds for S.

From Theorem 6.14, p. 176 we have

=
CS
and from Theorem 5.24, p. 148 we have

,
k=1 CSk

(6.68)

Qx Py =
S
Since Green's Theorem holds for each

Qx Py .
k=1 Sk

(6.69)

k,

we have that

=
CSk
and therefore

Qx Py ,
Sk

(6.70)

=
CS
as desired.

Qx Py ,
S

(6.71)

Exercise 6.7.4
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

180

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

a. Prove Green's Theorem for a right triangle with vertices of the form b. Prove Green's Theorem for a trapezoid having vertices of the form

(a, c) , (b, c) , and (a, d) . (a, c) , (b, c) ,(b, d) , and

(a, e) ,

where both

and

are greater than

c.

HINT: Represent this trapezoid as the union

of a rectangle and a right triangle that share a border. Then use Lemma 6.3, p. 179. c. Prove Green's Theorem for

any quadrilateral that has two vertical sides.

d. Prove Green's Theorem for any geometric set are piecewise linear functions.

whose upper and lower bounding functions

HINT: Show that

can be thought of as a nite union of

quadrilaterals, like those in part (c), each one sharing a vertical boundary with the next. Then, using induction and the previous exercise nish the argument. We need one nal lemma before we can complete the general proof of Green's Theorem. This one is where the analysis shows up; there are carefully chosen

Lemma 6.4:
S

's

and

's.
is smooth on all of

Suppose is true.

is contained in an open set

and that

U.

Then Green's Theorem

Proof:
Let the piecewise smooth geometric set

be determined by the interval 45, choose an

[a, b] and the two bounding

u Nr (S ) U.
functions a. b. c. d. set

and

l.

Using Theorem 2.11, p.

r > 0

such that the neighborhood

Now let

>0

be given, and choose

delta

to satisfy the following conditions:

< r/2, from which it follows that the open neighborhood N (S ) is a subset of the compact N r/2 (S ) . (See part (f ) of Exercise 2.7.5.) < /4M, where M is a common bound for all four continuous functions |P |, |Q|, |Py |, and |Qx | on the compact set N r/2 (S ) . < /4M (b a) . satises the conditions of Theorem 6.11, p. 173. pu
and

Next, using Theorem 6.1, p. 154, choose two piecewise linear functions 1. 2. 3. 4. ^

pl

so that

|u (x) pu (x) | < /2 for all x [a, b] . |l (x) pl (x) | < /2 for all x [a, b] . b |u' (x) p' u (x) | dx < . a
b a

|l' (x) p' l (x) | dx < . [a, b]

and the two bounding functions ^ ^

Let ^

be the geometric set determined by the interval ^

and

l,

where

u= pu + /2

and

l = pl /2. pl

We know that both

and

are piecewise linear functions.

We have to be a bit careful here, since for some could not simply use (2) that ^

x's

it could be that ^

pu (x) < pl (x) .

Hence, we

pu

and and

themselves as bounding functions for ^

S.

We do know from (1) and

u (x) < u (x)

^

l (x) > l (x) ,

^

which implies that the geometric set

is contained in

the geometric set the compact set

S . Also S N r/2 (S ) .

is a subset of the neighborhood

N (s) ,

which in turn is a subset of ^

Now, by part (d) of the preceding exercise, we know that Green's Theorem holds for

S . That is
(6.72)

=
C
^
S

(Qx Py ) .

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

181

We will show that Green's Theorem holds for (ii)

S by showing two things:

(i)

CS

| < 4, and
^
S

(Qx ) Py

(Qx Py ) | < .

We would then have, by the usual adding and subtracting

business, that

|
CS
and, since

(Qx Py ) | < 5,

(6.73)

is an arbitrary positive number, we would obtain

=
CS S

(Qx Py ) . |u (x) u (x) | <

^ for all

(6.74)

Let us estabish (i) rst. We have from (1) above that from (3) that

x [a, b] ,

and

b a
Hence, by Theorem 6.11, p. 173,

|u' (x) u (x) | dx =

a

'

|u' (x) p' u (x) | dx < .

(6.75)

graph(u) graph

! |

< .

(6.76)

Similarly, using (2) and (4) above, we have that

|
graph(l)

0 1 | ^ graph@ l A

< .

(6.77)

Also, the dierence of the line integrals of than

along the righthand boundaries of

and

is less

Thus ^ b, u (b)

! 1 |

(b,u(b)) C (b,l(b))

@b, l (b)A u (b) u(b)

^

u(b) l(b)

Q (b, t) dt

u (b)

Q (b, t) dt|

l (b)

= <
Of course, a similar calculation shows that

Q (b, t) dt| + |
^

l(b)
^

Q (b, t) dt|
^ (6.78)

l (b)

|l (b) l (b) | + |u (b) u (b) | M ( + ) 2M .

1 ^ @a, l (a)A

|
C

(a,l(a)) (a,u(a))

^ a, u (a)

! |

< .

(6.79)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

182

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

These four line integral inequalities combine to give us that

|
CS
establishing (i). Finally, to see (ii), we just compute

C
^
S

| < 4,

(6.80)

= = < | |
^ b u (t) a ^ l (t)

(Qy Px )

(Qy Px ) | (Qx (t, s) Py (t, s)) dsdt|

Qx ((t, s) Py (t, s)) dsdt (Qx (t, s) Py (t, s)) dsdt + 2M

b a
^

b u(t) a l(t)
^ b u (t) a u(t) ^

b l(t) a ^ l (t)

(Qx (t, s) Py (t, s)) dsdt|

(6.81)

|l (t) l (t) | + | u (t) u (t) | dt 4M (b a) .

This establishes (ii), and the proof is complete.

At last, we can nish the proof of this remarkable result.

Theorem 6.16:

PROOF OF GREEN'S THEOREM

Proof:
S be determined by the interval [a, b] and the two bounding functions u and l. Recall u (x) l (x) > 0 for all x (a, b) . For each natural number n > 2, let Sn be the geometric set that is determined by the interval [a + 1/n, b 1/n] and the two bounding functions un and ln , where un = u (u l) /n restricted to the interval [a + 1/n, b 1/n] , and ln = l + (u l) /n restricted to [a + 1/n, b 1/n] . Then each Sn is a piecewise smooth geometric set, whose boundary 0 has nite length, and each Sn is contained in the open set S where by hypothesis is smooth. Hence, by Lemma 4, Green's Theorem holds for each Sn . Now it should follow directly, by taking limits, that Green's Theorem holds for S. In fact, this is the case, and we leave the details to the
As usual, let that exercise that follows.

Exercise 6.7.5
Let

S, ,

and the

Sn 's

be as in the preceding proof.

a. Using Theorem 6.11, p. 173, show that

= lim
CS
b. Let

.
CSn

(6.82)

be a bounded integrable function on the geometric set

S.

Prove that

f = lim
S Sn

f.

(6.83)

c. Complete the proof to Green's Theorem; i.e., take limits.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

183

6.13: REMARK Green's Theorem is primarily a theoretical result.

Green's Theorem says that the area of the geometric set

It is rarely used to compute a

line integral around a curve or an integral of a function over a geometric set. However, there is one amusing exception to this, and that is when the dierential form

= x dy.

For that kind of

can be computed as follows:

A (S ) =
S

1=

tialQ = S tialx

x dy.
CS

(6.84)

This is certainly a dierent way of computing areas of sets from the methods we developed earlier. Try this way out on circles, ellipses, and the like.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

184

CHAPTER 6. INTEGRATION OVER SMOOTH CURVES IN THE PLANE

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

Chapter 7

The Fundamental Theorem of Algebra, and The Fundamental Theorem of Analysis

7.1 The Fundamental Theorem of Algebra, and the Fundamental Theorem of Analysis
1
In this chapter we will discover the incredible dierence between the analysis of functions of a single complex variable as opposed to functions of a single real variable. Up to this point, in some sense, we have treated them as being quite similar subjects, whereas in fact they are extremely dierent in character. Indeed, if is a dierentiable function of a complex variable on an open set expandable in a Taylor series around every point in guaranteed to have innitely many derivatives on The main points of this chapter are: 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.

U C,

then we will see that

is actually

U.

In particular, a function

f of

a complex variable is

if it merely has the rst one on

U.

This is in marked

contrast with functions of a real variable. See part (3) of Theorem 4.17, p. 104.

The Cauchy-Riemann Equations (Theorem 7.1, Cauchy-Riemann equations, p. 186), Cauchy's Theorem (Theorem 7.3, Cauchy's Theorem, Fundamental Theorem of Analysis, p. 188), Cauchy Integral Formula (Theorem 7.4, Cauchy Integral Formula, p. 189), A complex-valued function that is dierentiable on an open set is expandable in a Taylor series around each point of the set (Theorem 7.5, p. 192), The Identity Theorem (Theorem 7.6, Identity Theorem, p. 194), The Fundamental Theorem of Algebra (Theorem 7.7, Fundamental Theorem of Algebra, p. 195), Liouville's Theorem (Theorem 7.8, Liouville, p. 196), The Maximum Modulus Principle (corollary to Corollary 7.4,
198), Maximum Modulus Principle, p.

The Open Mapping Theorem (Theorem 7.10, Open Mapping Theorem, p. 198), The uniform limit of analytic functions is analytic (Theorem 7.12, p. 201), and The Residue Theorem (Theorem 7.17, Residue Theorem, p. 206).
content is available online at <http://cnx.org/content/m36234/1.2/>.

1 This

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

185

186

7.2 Cauchy's Theorem

Theorem 7.1:
f = u + iv

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS
2

We begin with a simple observation connecting dierentiability of a function of a complex variable to a relation among of partial derivatives of the real and imaginary parts of the function. already visited this point in Exercise 4.3.3. Cauchy-Riemann equations Actually, we have

Let

be a complex-valued function of a complex variable

is dierentiable, as a function of a complex variable, at the point

two partial dierential equations, known as the

Cauchy-Riemann Equations, hold:

z = x + iy (x, y ) , and suppose c = (a, b) . Then the following

tialu tialv (a, b) = (a, b) , tialx tialy

and

(7.1)

tialv tialu (a, b) = (a, b) . tialy tialx

(7.2)

Proof:
We know that

f ' (c) = limh 0

and this limit is taken as the complex number real

f (c + h) f (c) , h h's
approaching 0. For real

(7.3)

approaches 0. We simply examine this limit for

h's

approaching 0 and then for purely imaginary

h's,

we have

f ' (c)

= = = = =

f ' (a + ib)
)f (a+ib) lim f (a+h+ibh h0 )u(a,b)iv (a,b) limh 0 u(a+h,b)+iv(a+h,b h )u(a,b) )v (a,b) lim u(a+h,bh + ilim v(a+h,bh h0 h0 tialu tialx tialv (a, b) + i tialx (a, b) .
we have

(7.4)

For purely imaginary

h's,

which we write as

h = ik,

f ' (c)

= = = = =

f ' (a + ib)
))f (a+ib) lim f (a+i(b+k ik k0 k)u(a,b)iv (a,b) lim u(a,b+k)+iv(a,b+ ik k0 )u(a,b) ilim u(a,b+kk + k 0 v (a,b+k)v (a,b) k

(7.5)

i tialu tialy (a, b) +

tialv tialy

(a, b) . f ' (c)

gives the

Equating the real and imaginary parts of these two equivalent expressions for Cauchy-Riemann equations.

As an immediate corollary of this theorem, together with Green's Theorem (Theorem 6.15, Green, p. 177), we get the following result, which is a special case of what is known as Cauchy's Theorem.

2 This

content is available online at <http://cnx.org/content/m36235/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

187

Corollary 7.1:
Let of

be a piecewise smooth geometric set whose boundary

CS

has nite length. Suppose

is a

complex-valued function that is continuous on

S.

Exercise 7.2.1
b. Suppose

Then the contour integral

S and f ( ) d = 0. CS

dierentiable at each point of the interior

S0

a. Prove the preceding corollary. See Theorem 6.12, p. 174.

f = u + iv

is a dierentiable, complex-valued function on an open disk

Br (c)

in

C,

and assume that the real part c. Suppose

is a constant function. Prove that

is a constant function.

Derive the same result assuming that

is a constant function.

and

are two dierentiable, complex-valued functions on an open disk real part of

Br (c)

in

C. Show that, if the real part of f is equal to the k such that f (z ) = g (z ) + k, for all z Br (c) .

g,

then there exists a constant

For future computational purposes, we give the following implications of the Cauchy-Riemann equations. As with Theorem 7.1, Cauchy-Riemann equations, p. 186, this next theorem mixes the notions of dierentiability of a function of a complex variable and the partial derivatives of its real and imaginary parts.

Theorem 7.2:
Let

f = u + iv

be a complex-valued function of a complex variable, and suppose that

is dier-

entiable at the point

c = (a, b) .

Let

be the

22

matrix (7.6)

A = (ux (a, b) vx (a, b) uy (a, b) vy (a, b)) .

Then: 1.

|f ' (c) | = det (A) . V 1 = (ux (a, b) , uy (a, b))

are linearly independent vectors in

2. The two vectors

and

V 2 = (vx (a, b) , vy (a, b)) f ' (c) = 0.

(7.7)

R2

if and only if

3. The vectors

V 3 = (ux (a, b) , vx (a, b))

are linearly independent vectors in

and

V 4 = (uy (a, b) , vy (a, b)) f ' (c) = 0. A

(7.8)

R2

if and only if

Proof:
Using the Cauchy-Riemann equations, we see that the determinant of the matrix is given by

detA = = = = =
proving part (1).

ux (a, b) vy (a, b) uy (a, b) vx (a, b) (ux (a, b)) + (vx (a, b)) f ' (c) f ' (c) |f ' (c) | ,
2 2 2
(7.9)

(ux (a, b) + ivx (a, b)) (ux (a, b) ivx (a, b))

The vectors

and

V1

V2

are the columns of the matrix

A,

and so, from elementary linear algebra, we see

that they are linearly independent if and only if the determinant of from part (1). Similarly, part (3) is a consequence of part (1). It may come as no surprise that the contour integral of a function set

A f

is nonzero. Hence, part (2) follows around the boundary of a geometric

is not necessarily 0 if the function

is not dierentiable at each point in the interior of

S.

However, it

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

188

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS

is exactly these kinds of contour integrals that will occupy our attention in the rest of this chapter, and we shouldn't jump to any conclusions.

Exercise 7.2.2
Let

be a point in dene

C,

and let

be the geometric set that is a closed disk

parameterization of the boundary

n Z,

fn (z ) = (z c) . f Cr n ( d = 0
for all

Cr

of

given by

(t) = c + reit

for

B r (c) . Let be the t [0, 2 ] . For each integer

a. Show that b. Show that

n = 1. 1 d = 2i. Cr c
(7.10)

f1 ( ) d =
Cr

There is a remarkable result about contour integrals of certain functions that aren't dierentiable everywhere within a geometric set, and it is what has been called the Fundamental Theorem of Analysis, or Cauchy's Theorem. This theorem has many general statements, but we present one here that is quite broad and certainly adequate for our purposes.

Theorem 7.3:
Let

Cauchy's Theorem, Fundamental Theorem of Analysis

be a piecewise smooth geometric set whose boundary

CS

has nite length, and let

be a piecewise smooth geometric set, whose boundary

C^
S

also is of nite length. ^

S S0 Suppose f is f
is

0 continuous on S S , i.e., at every ^ 0 dierentiable on S S , i.e., at every

^

0
and assume that

point point

z z

that is in in

but not in

S ,

S0

but not in

S.

(We think of these sets as being

the points between the boundary curves of these geometric sets.) Then the two contour integrals

CS

f ( ) d

and

f ( ) d
^
S

are equal.

Proof:
and

Let the geometric set

be determined by the interval

[a, b]

and the two bounding functions ^ ^

l,

and let the geometric set ^ ^

be determined by the subinterval ^ ^

a, b

of

[a, b]

and the two ^

bounding functions

and

l .

Because

S S0,

we know that

u (t) < u (t)

and

l (t) < l (t)

for all

t a, b .

^ ^

We dene four geometric sets

S1 , ..., S4
^

as follows:

1.

S1

is determined by the interval

a, a

and the two bounding functions

and

restricted to

that interval. 2.

S2

is determined by the interval

^ ^

a, b

and the two bounding functions

and

restricted

to that interval. 3.

S3

is determined by the interval

^ ^

a, b
^

and the two bounding functions

and

restricted to

that interval. 4.

S4

is determined by the interval

b, b

and the two bounding functions

and

restricted to

that interval. Observe that the ve sets

S , S1 , ..., S4

constitute a partition of the geometric set

S.

The corollary

to Theorem 7.1, Cauchy-Riemann equations, p.

186 applies to each of the four geometric sets

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

189

S1 , ..., S4 .

Hence, the contour integral of

around each of the four boundaries of these geometric

sets is 0. So, by Exercise 6.7.3,

CS

f ( ) d

= =

f ( ) d +
^
S

4 k=1

CSk

f ( ) d
(7.11)

f ( ) d,
^
S

as desired.

Exercise 7.2.3
a. Draw a picture of the ve geometric sets in the proof above and justify the claim that the sum of the four contour integrals around the geometric sets

S1 , ..., S4

is the integral around

CS
b. Let

minus the integral around

C^ .
S

S1 , ..., Sn

be pairwise disjoint, piecewise smooth geometric sets, each having a boundary

of nite length, and each contained in a piecewise smooth geometric set also has nite length. Prove that the

whose boundary

Sk 's

are some of the elements of a partition ^

l } {S

of

S,

each of which is piecewise smooth and has a boundary of nite length.

Show that, by

reindexing, c. Suppose

S1 , ..., Sn

can be chosen to be the rst

elements of the partition

{S l }.

HINT:

Just carefully adjust the proof of Theorem 5.25, p. 148.

is a piecewise smooth geometric set whose boundary has nite length, and let

S1 , ..., Sn

be a partition of

for which each

Sk

is piecewise smooth and has a boundary

of nite length. Suppose the boundary

is continuous on each of the boundaries

CSk

of the

CS

of

S,

and assume that

is continuous on each of the

Sk 's,

CSk Sk 's as well as for 1 k m,

and dierentiable at each point of their interiors. Prove that

f ( ) d =
CS k=m+1 CSk

f ( ) d.

(7.12)

d. Prove the following generalization of the Cauchy Theorem: Let interior of a piecewise smooth geometric set

S1 , ..., Sn

be pairwise disjoint,

piecewise smooth geometric sets whose boundaries have nite length, all contained in the

whose boundary also has nite length. Suppose

is continuous at each point of

that is not in the interior of any of the

dierentiable at each point of

S0

that is not an element of any of the

Sk 's, and that f Sk 's. Prove that

is

f ( ) d =
CS k=1 CSk

f ( ) d.

(7.13)

Perhaps the main application of Theorem 7.3, Cauchy's Theorem, Fundamental Theorem of Analysis, p. 188 is what's called the Cauchy Integral Formula. It may not appear to be useful at rst glance, but we will be able to use it over and over throughout this chapter. In addition to its theoretical uses, it is the basis for a technique for actually evaluating contour integrals, line integrals, as well as ordinary integrals.

Theorem 7.4:
Let

Cauchy Integral Formula

be a piecewise smooth geometric set whose boundary

continuous function on we have

that is dierentiable on the interior

CS has nite length, and let f be a S 0 of S. Then, for any point z S 0 ,

f (z ) =

1 2i

f ( ) d. CS z

(7.14)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

190

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS

REMARK This theorem is an initial glimpse at how dierentiable functions of a complex variable
are remarkably dierent from dierentiable functions of a real variable. Indeed, Cauchy's Integral Formula shows that the values of a dierentiable function geometric set

at all points in the interior of a

are completely determined by the values of that function on the boundary of the

set. The analogous thing for a function of a real variable would be to say that all the values of a dierentiable function at the endpoints

at points in the open interval

(a, b) are completely determined by its values

and

b.

This is patently absurd for functions of a real variable, so there surely is

something marvelous going on for dierentiable functions of a complex variable.

Proof:
Let

r be any positive number such that B r (z ) is contained in the interior S 0 B r (z ) is a piecewise smooth geometric set S
^ contained in

of

S, and note that the

close disk

S 0 . We will write Cr instead of C^ for the boundary of this disk, and we will use as a parameterization of the curve Cr the function : [0, 2 ] Cr given by (t) = z + reit . Now the function g ( ) = f ( ) / ( z ) is continuous on 0 ^ ^ S S and dierentiable on S 0 S , so that Theorem 7.3, Cauchy's Theorem, Fundamental Theorem of Analysis, p. 188 applies to the function g. Hence
1 2i f ( ) CS z S

= = = = =

1 2i CS g ( ) d 1 2i CR g ( ) , d f ( ) 1 2i Cr z d it 2 f (z +re ) 1 it 2i 0 z +reit z ire dt 2 1 it dt. 2 0 f z + re

(7.15)

Since the equality established above is valid, independent of case we would be evaluating

r,

we may take the limit as

goes to

0, and the equality will persist. We can evaluate such a limit by replacing the

by

1/n,

in which

1 n 2 lim
where follows

f
0

z+

1 it e n

dt = lim

1 n 2

fn (t) dt,
0

(7.16)

fn (t) = f z + f rac1neit . Finally, because the function f is continuous at the point z, it that the sequence {fn } converges uniformly to the constant function f (z ) on the interval
So, by Theorem 5.6, we have that

[0, 2 ] .

1 n 2 lim
Therefore,

fn (t) dt =
0

1 2

f (z ) dt = f (z ) .
0

(7.17)

1 2i
and the theorem is proved.

f ( 1 d = lim r 0 z 2 CS

f z + reit dt = f (z ) ,
0

(7.18)

The next exercise gives two simple but strong consequences of the Cauchy Integral Formula, and it would be wise to spend a few minutes deriving other similar results.

Exercise 7.2.4
a. Let

and

be as in the preceding theorem, and assume that

f (z ) = 0

for every point on

the boundary

CS

of

S.

Prove that

f (z ) = 0

for every

z S.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

191

b. Let

dierentiable on that

f and g S 0 , and such that f ( ) = g ( ) for f (z ) = g (z ) for all z S.

be as in part (a), and suppose that

are two continuous functions on every point on the boundary of

S, both S. Prove

The preceding exercise shows that two dierentiable functions of a complex variable are equal everywhere on a piecewise smooth geometric set

if they agree on the boundary of the set. More is true. We will see

below in the Identity Theorem that they are equal everywhere on a piecewise smooth geometric set agree just along a single convergent sequence in the interior of

if they

S.

Combining part (b) of Exercise 7.2.3, Exercise 6.7.3, and Theorem 7.3, Cauchy's Theorem, Fundamental Theorem of Analysis, p. 188, we obtain the following corollary:

Corollary 7.2:
Let

S1 , ..., Sn

be pairwise disjoint, piecewise smooth geometric sets whose boundaries have nite

length, all contained in the interior of a piecewise smooth geometric set nite length. Suppose

whose boundary has Then,

is continuous at each point of

that is not in the interior of any of the

Sk 's, and that f is dierentiable at each point of S 0 that is not an element of any of the Sk 's. 0 for any z S that is not an element of any of the Sk 's, we have f (z ) = 1 2i f ( ) d CS z
n CSk

k=1

f ( ) d z

(7.19)

Proof:
Let be a partition of know that

r > 0 be such that B r (z ) is disjoint from all the Sk 's. By part (b) of Exercise 7.2.3, let T1 , ..., Tm S such that Tk = Sk for 1 k n, and Tn+1 = B r (z ) . By Exercise 6.7.3, we f ( ) d = CS z
m CTk

k=1

f ( ) d. z

(7.20)

From the Cauchy Integral Formula, we know that

f ( ) d = 2if (z ) . CTn+1 z
Also, since

(7.21)

f ( ) / ( z )

is dierentiable at each point of the interior of the sets

Tk

for

k > n + 1,

we have from Theorem 7.2, p. 187 that for all

k >n+1
(7.22)

Ctk
Therefore,

f ( ) d = 0. z

f ( ) d = CS z
which completes the proof.

n CSk

k=1

f ( ) d + 2if (z ) , z

(7.23)

Exercise 7.2.5
Suppose

is a piecewise smooth geometric set whose boundary has nite length, and let

be points in except the

S0.

Suppose

is a complex-valued function that is continuous at every point of

c1 , ..., cn S

Ck 's

and dierentiable at every point of

S0

except the

ck 's.

Let

numbers such that the disks

{B Rk (ck )}

are pairwise disjoint and all contained in

r1 , ..., rn S0.

be positive

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

192

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS
a. Prove that

f ( ) d, =
CS
where

f ( ) d
k=1 Ck

(7.24)

b. For any

Ck denotes the boundary of the disk B rk (ck ) . z S 0 that is not in any of the closed disks B rk (ck ) , f ( ) d = 2if (z ) + CS z
n

show that

k=1

f ( ) d. Ck z

(7.25)

c. (c) Specialize part (b) to the case where except at the central point formula

c.

For each

S = B r (c) , and f is analytic at each point of Br (c) z = c in Br (c) , and any 0 < < |z c|, derive the f ( ) d. C z
(7.26)

f (z ) =

1 2i

f ( ) 1 d z 2 i Cr

7.3 Basic Applications of the Cauchy Integral Formula

a function that is a dierentiable function of a complex variable on an open set a Taylor series around every point in

As a major application of the Cauchy Integral Formula, let us show the much alluded to remarkable fact that

is actually expandable in

U,

Theorem 7.5:
Suppose

i.e., is an analytic function on

U. U C, and let c be r > 0 for which

is a dierentiable function of a complex variable on an open set

an element of

U.

Then

is expandable in a Taylor series around

c.

In fact, for any

B r (c) U,

we have

f (z ) =
n=0

an (z c)

(7.27)

Proof:
that

for all

z Br (c) . r>0
such that the closed disk

Choose an

Note that, for all points

B r (c) U, and write Cr for the boundary of this disk. on the curve Cr , and any xed point z in the open disk Br (c) , we have |z c| < r = | c|, whence |z c|/| c| = |z c|/r < 1. Therefore the geometric series

n=0
Moreover, by the Weierstrass uniformly on the curve

zc c

n
converges to

1 c . 1 z c

(7.28)

M -Test, as functions of the variable , this innite series converges Cr . We will use this in the calculation below. Now, according to Theorem 7.4,

3 This

content is available online at <http://cnx.org/content/m36237/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

193

Cauchy Integral Formula, p. 189, we have that

f (z )

= = = = = = =

f ( ) d Cr z f ( ) 1 2i Cr c+cz d f ( ) 1 2i Cr ( c)(1 zc ) d c n f ( ) 1 z c d n=0 c 2i Cr c n f ( ) 1 n=0 ( c)n+1 (z c) d 2i Cr n f ( ) 1 n=0 Cr ( c)n+1 (z c) d 2i n n=0 an (z c) , 1 2i

(7.29)

where we are able to bring the summation sign outside the integral by part (3) of Theorem 6.10, p. 169, and where

an =
This proves that

1 2i

f ( )
Cr

( c)

n+1

d. c,
as desired.

(7.30)

is expandable in a Taylor series around the point

Using what we know about the relationship between the coecients of a Taylor series and the derivatives of the function, together with the Cauchy Integral Theorem, we obtain the following formulas for the derivatives of a dierentiable function Formulas.

of a complex variable. These are sometimes also called the Cauchy Integral

Corollary 7.3:
Suppose

element of

f is a dierentiable function of a complex variable U. Then f is innitely dierentiable at c, and f (n) (c) = n! 2i f ( )
Cs n+1

on an open set

U,

and let

be an

( c)

d, CS

(7.31) has nite length, and for which

for any piecewise smooth geometric set

SU

whose boundary

Exercise 7.3.1
b. Let

belongs to the interior

S0

of

S.

a. Prove the preceding corollary.

f, U,

and

be as in Theorem 7.5, p. 192. Show that the radius of convergence

of the

Taylor series expansion of

around

is at least as large as the supremum of all

for which

Bs (c) U.
c. Conclude that the radius of convergence of the Taylor series expansion of a dierentiable function of a complex variable is as large as possible. That is, if

is dierentiable on a disk

Br (c) ,

d. Consider the real-valued function of a real variable given by

Br (c) . f (x) = 1/ 1 + x2 . Show that f is dierentiable at each real number x. Show that f is expandable in a Taylor series around 0,
then the Taylor series expansion of around converges on all of but show that the radius of convergence of this Taylor series is equal to 1. Does this contradict part (c)?

e. Let

be the complex-valued function of a complex variable given by

We have just replaced the real variable

of part (d) by a complex

f (z ) = 1/ 1 + z 2 . variable z. Explain the

apparent contradiction that parts (c) and (d) present in connection with this function.

Exercise 7.3.2
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

194

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS
a. Let

be a piecewise smooth geometric set whose boundary

be a continuous function on the curve

CS .

Dene a function

CS has nite F on S 0 by

length, and let

F (z ) =
Prove that

f ( ) d. CS z
point

(7.32)

F (z ) =

F is expandable in a Taylor series around each n an (z c) for all z in a disk Br (c) S 0 , where an = n! 2i f ( )
CS

c S0.

Show in fact that

( c)

n+1

d.

(7.33)

HINT: Mimic the proof of Theorem 7.5, p. 192. b. Let

c.

F be as in part (a). Is F dened on the boundary CS of S ? If z belongs to the CS , and z = limzn , where each zn S 0 , Does the sequence {F (zn )} converge, and, if so, does it converge to f (z )? Let S be the closed unit disk B 1 (0) , and let f be dened on the boundary C1 of this disk by f (z ) = z, i.e., f (x + iy ) = x iy. Work out the function F of part (a), and then re-think
and boundary about part (b).

d. Let

and

be as in part (a). If, in addition,

is continuous on all of

and dierentiable on

S0,

show that

F (z ) = 2if (z )

for all

z S0.

Think about this magic constant

2i.

Review

the proof of the Cauchy Integral Formula to understand where this constant comes from. Theorem 3.15,

THEOREM 3.14159 (Denition of ), p.

Identity Theorem

73 and Exercise 3.9.2 constitute what we called

the identity theorem for functions that are expandable in a Taylor series around a point result than that is actually true for functions of a complex variable.

c. An even stronger

Theorem 7.6:

f be a continuous complex-valued function on a piecewise smooth geometric set S, and assume 0 0 that f is dierentiable on the interior S of S. Suppose {zk } is a sequence of distinct points in S 0 that converges to a point c in S . If f (zk ) = 0 for every K, then f (z ) = 0 for every z S.
Let

Proof:
let let

It follows from Exercise 3.9.2 that there exists an ^ ^

r > 0 such that f (z ) = 0 for all z Br (c) . Now w be another point in S 0 , and let us show that f (w) must equal 0. Using part (f ) of Exercise 6.2.2, : a, b C t a, b
^ ^ be a piecewise smooth curve, joining

c to w, that lies entirely in S 0 . Let A be the

^ We claim rst that

set of all

such that

f ( (s)) = 0

for all

s a, t .

is nonempty. ^

Indeed, because if

is continuous, there exists an

>0

such that

| (s) c| = | (s) a | < r

^

|s a | < .

Therefore ^

f ( (s)) = 0

for all

s a, a + ,

^ ^

whence,

a + A.

Obviously,

A
^

is bounded above by whence, since

b,

and we write ^

t0

for the supremum of

A.

We wish to show that

t0 = b ,

is continuous at ^

B ,f (w) = f

= f ( (t0 )) = 0.

Suppose, by way of

t0 < b , and write z0 = (t0 ) . Now z0 S 0 , and z0 = lim (t0 1/k ) because is continuous at t0 . But f ( (t0 1/k )) = 0 for all k. So, again using Exercise 3.9.2, we know that ' there exists an r > 0 such that f (z ) = 0 for all z Br ' (z0 ) . As before, because is continuous
contradiction, that at

t0 , there exists a > 0 such that t0 + < b and | (s) (t0 ) | < r' if |s t0 | < . Hence, f ( (s)) = 0 for all s (t0 , t0 + ) , which implies that t0 + belongs to A. But then t0 could
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

195

not be the supremum of and therefore

A,

and therefore we have arrived at a contradiction. Consequently,

t0 = b ,

S ,

and since

f (w) = 0 for all w S 0 . Of course, since every point in S is a limit of points from f is continuous on S, we see that f (z ) = 0 for all z S, and the theorem is proved.

The next exercise gives some consequences of the Identity Theorem. Part (b) may appear to be a contrived example, but it will be useful later on.

Exercise 7.3.3
a. Suppose

b.

S {zk } is a sequence of elements of S 0 that 0 converges to a point c S , and assume that f (zk ) = g (zk ) for all k. Prove that f (z ) = g (z ) for all z S. Suppose f is a nonconstant dierentiable function dened on the interior of a piecewise smooth 0 0 geometric set S. If c S and B (c) S , show that there must exist an 0 < r < for which f (c) = f (z ) for all z on the boundary of the disk Br (c) .
and are two functions, both continuous on a piecewise smooth geometric set Suppose and both dierentiable on its interior.

7.4 The Fundamental Theorem of Algebra

fanfare, please!

We can now prove the Fundamental Theorem of Algebra, the last of our primary goals. One nal trumpet

Theorem 7.7:
p (z )
such that

Fundamental Theorem of Algebra That is, every nonconstant polynomial of a complex variable has a root in

Let

be a nonconstant polynomial of a complex variable. Then there exists a complex number

z0

p (z0 ) = 0.

the complex numbers.

Proof:
degree

p is a nonconstant polynomial of f (z ) = 1/p (z ) , and observe that f is dened and dierentiable at every point z C. We will show that f is a constant function, implying that p = 1/f is a constant, and that will give the contradiction. We prove that f is constant by showing
We prove this theorem by contradiction. Thus, suppose that

n 1,

and that

p (z )

is never 0.

Set

that its derivative is identically 0, and we compute its derivative by using the Cauchy Integral Formula for the derivative.

n |cn | 2 |z | |p (z ) |, for all z for which |z | B, and where cn is the (nonzero) leading coecient of the polynomial M p. Hence, |f (z ) | |z |n for all |z | B, where we write M for 2/|cn |. Now, x a point c C. Because
From part (4) of Theorem 3.1, p. 57, we recall that there exists a

B>0

such that

is dierentiable on the open set

Formula, p. 189 to compute the derivative of disks

Br (c) ,

and we choose an

U = C, we can use the corollary to Theorem 7.4, Cauchy Integral f at c by using any of the curves Cr that bound the r large enough so that |c + reit | B for all 0 t 2. Then, | 21 i
f ( ) Cr ( c)2 f (c+reit )

|f ' (c) | = =
4 This

d |

2 1 it 2 | 0 (c+reit c)2 ire dt| 2 1 it | dt 2r 0 |f c + re 2 1 M 2r 0 |c+reit |n dt M rB n .

(7.34)

content is available online at <http://cnx.org/content/m36238/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

196

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS
Hence, by letting

tend to innity, we get that

|f ' (c) | lim

and the proof is complete.

M = 0, r rB n

(7.35)

REMARK 7.1:
The Fundamental Theorem of Algebra settles a question rst raised back in Section 1.1. There, we introduced a number

that was a root of the polynomial

x 2 + 1.

We did this in order to build

a number system in which negative numbers would have square roots. We adjoined the number

to the set of real numbers to form the set of complex numbers, and we then saw that in fact

every complex number

has a square root. However, a fear was that, in order to build a system

in which every number has an

nth

root for every

n,

we would continually need to be adjoining new

elements to our number system. However, the Fundamental Theorem of Algebra shows that this is not necessary. The set of complex numbers is already rich enough to contain all even more. Practically the same argument as in the preceding proof establishes another striking result.

nth

roots and

Theorem 7.8:
Suppose

Liouville

is a bounded, everywhere dierentiable function of a complex variable. Then

must

be a constant function.

Exercise 7.4.1

Prove Liouville's Theorem.

7.5 The Maximum Modulus Principle

Our next goal is to examine so-called max/min problems for coplex-valued functions of complex variables. Since order makes no sense for complex numbers, we will investigate max/min problems for the absolute value of a complex-valued function. For the corresponding question for real-valued functions of real variables, we have as our basic result the First Derivative Test (Theorem 4.8, First Derivative Test for Extreme Values, p. 92). Indeed, when searching for the poinhts where a dierentiable real-valued function

on an interval

[a, b]

attains its extreme values, we consider rst the poinhts where it attains a local max or min, to which

purpose end Theorem 4.8, First Derivative Test for Extreme Values, p. 92 is useful. Of course, to nd the absolute minimum and maximum, we must also check the values of the function at the endpoints. An analog of Theorem 4.8, First Derivative Test for Extreme Values, p. 92 holds in the complex case, but in fact a much dierent result is really valid. Indeed, it is nearly impossible for the absolute value of a dierentiable function of a complex variable to attain a local maximum or minimum.

Theorem 7.9:
Let

S, and assume that f is S 0 at which the real-valued function |f | attains a local maximum. That is, there exists an > 0 such that |f (c) | |f (z ) | for all z satisfying |z c| < . Then f is a constant function on S ; i.e., f (z ) = f (c) for all z S. In other
be a continuous function on a piecewise smooth geometric set dierentiable on the interior

S0

of

S.

Suppose

is a point in

words, the only dierentiable functions of a complex variable, whose absolute value attains a local maximum on the interior of a geometric set, are constant functions on that set.

Proof:
If

Theorem, p. 194),

f (c) = 0, then f (z ) = 0 for all z B (c) . Hence, by the Identity Theorem (Theorem 7.6, Identity f (z ) would equal 0 for all z S. so, we may as well assume that f (c) = 0. Let r be any positive number for which the closed disk B r (c) is contained in B (c) . We claim rst
content is available online at <http://cnx.org/content/m36239/1.2/>.

5 This

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

197

that there exists a point course,

z on the boundary Cr of the disk B r (c) for which |f (z ) | = |f (c) |. Of |f (z | |f (c) | for all z on this boundary by assumption. By way of contradiction, suppose that |f ( ) | < |f (c) | for all on the boundary Cr of the disk. Write M for the maximum value of the function |f | on the compact set Cr . Then, by our assumption, M < |f (c) |. Now, we use the
Cauchy Integral Formula:

|f (c) |

= = = <
1 2 | 1 2

| 21 i

f ( ) d | Cr c it 2 f (c+re ) ireit dt| reit 0 2 |f c + reit | dt 0 2 1 2 0 M dt

(7.36)

M |f (c) |,

and this is a contradiction. Now for each natural number n for which 1/n < , let zn be a point for which |zn c| = 1/n and |f (zn ) | = |f (c) |. We claim that the derivative f ' (zn ) of f at zn = 0 for all n. What we know is that the real-valued function value at

F (x, y ) = |f (x + iy ) | = u (x, y ) + (v (x, y ))

attains a local maximum

zn = (xn , yn ) .
That is

Hence, by Exercise 4.11.3, both partial derivatives of

must be 0 at

(xn , yn ) .

2u (xn , yn )
and

tialv tialu (xn , yn ) + 2v (xn , yn ) (xn , yn ) = 0 tialx tialx tialu tialv (xn , yn ) + 2v (xn , yn ) (xn , yn ) = 0. tialy tialy

(7.37)

2u (xn , yn )
Hence the two vectors

(7.38)

V1 =
and

tialu tialv (xn , yn ) , (xn , yn ) tialx tialx

(7.39)

V2 =
are both perpendicular to the vector

tialu tialv (xn , yn ) , (xn , yn ) tialy tialy

(7.40)

|f (zn ) | = |f (c) | > 0, = 0, according to Theorem 7.2, p. 187. ' 0 Since c = limzn , and f is analytic on S , it follows from the Identity Theorem that there exists ' an r > 0 such that f (z ) = 0 for all z Br (c) . But this implies that f is a constant f (z ) = f (c) for all z Br (c) . And thenm, again using the Identity Theorem, this implies that f (z ) = f (c) for all z S, which completes the proof.

V 3 = (u (xn , yn ) , v (xn , yn )) . But V 3 = 0, because V 3 ' and hence V 1 and V 2 are linearly dependent. But this implies that f (zn )

7.2: REMARK

Of course, the preceding proof contains in it the verication that if

|f |

attains a 92. But,

maximum at a point

where it is dierentiable, then

f ' (c) = 0.

This is the analog for functions

of a complex variable of Theorem 4.8, First Derivative Test for Extreme Values, p.

Theorem 7.9, p. 196 certainly asserts a lot more than that. In fact, it says that it is impossible for

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

198

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS
the absolute value of a nonconstant dierentiable function of a complex variable to attain a local maximum. Here is the coup d'grs:

Corollary 7.4:
f

Maximum Modulus Principle

Let

be a continuous, nonconstant, complex-valued function on a piecewise smooth geometric set

S,

and suppose that

is dierentiable on the interior

S0

continuous, real-valued function does not belong to the interior

|f |

on

S,

and let

be a point in

S0

of

S;

it belongs to the

M be the maximum value of the S for which |f (z ) | = M. Then, z boundary of S. In other words, |f | attains
of Let

S.

its maximum value only on the boundary of

S.

Exercise 7.5.1
b. Let

a. Prove the preceding corollary. a local minimum; i.e., there exists an

c.

U, and let c U be a point at which |f | achieves > 0 such that |f (c) | |f (z ) | for all z B (c) . Show that, if f (c) = 0, then f is constant on B (c) . Show by example that, if f (c) = 0, then f need not be a constant on B (c) . Prove the Minimum Modulus Principle: Let f be a nonzero, continuous, nonconstant, function on a piecewise smooth geometric set S, and let m be the minimum value of the function |f | on S. If z is a point of S at which this minimum value is atgtained, then z belongs to the boundary CS of S. f
be an analytic function on an open set

7.6 The Open Mapping Theorem and the Inverse Function Theorem
Function Theorem, p. 95, the Inverse Function Theorem.

We turn next to a question about functions of a complex variable that is related to Theorem 4.10, Inverse That result asserts, subject to a couple of hypotheses, that the inverse of a one-to-one dierentiable function of a real variable is also dierentiable. Since a function is only dierentiable at points in the interior of its domain, it is necessary to verify that the point

f (c)

is in the interior of the domain

f (S )

of the inverse function

f 1

before the question of

dierentiability at that point can be addressed. And, the peculiar thing is that it is this point about being in the interior of

f (c)

f (S ) that is the subtle part. f (c)

The fact that the inverse function is dierentiable there,

and has the prescribed form, is then only a careful real variables, the fact that

argument. For continuous real-valued functions of

belongs to the interior of

f (S )

boils down to the fact that intervals get

mapped onto intervals by continuous functions, which is basically a consequence of the Intermediate Value Theorem. However, for complex-valued functions of complex variables, the situation is much deeper. For instance, the continuous image of a disk is just not always another disk, and it may not even be an open set. Well, all is not lost; we just have to work a little harder.

Theorem 7.10:
Let

Open Mapping Theorem

be a piecewise smooth geometric set, and write

for the (open) interior

S0

of

is a nonconstant dierentiable, complex-valued function on the set

U.

Then the range

S. Suppose f (U )of f

is an open subset of

Proof:
Let for all

C.

c be in U. Because f is not a constant function, there must exist an r > 0 such that f (c) = f (z ) z on the boundary Cr of the disk Br (c) . See part (b) of Exercise 7.3.3. Let z0 be a point in the compact set Cr at which the continuous real-valued function |f (z ) f (c) | attains its minimum value s. Since f (z ) = f (c) for any z Cr , we must have that s > 0. We claim that the disk Bs/2 (f (c)) belongs to the range f (U ) of f. This will show that the point f (c) belongs to the ihnterior of the set f (U ) , and that will nish the proof.
6 This
content is available online at <http://cnx.org/content/m36240/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

199

Bs/2 (f (c) is not contained in f (U ) ,, and let w Bs/2 (f (c)) f (U ) . We have that |w f (c) | < s/2, which implies that |w f (z ) | > s/2 for all z Cr . Consider the function g dened on the closed disk B r (c) by g (z ) = 1/ (w f (z )) . Then g is continuous on the closed disk B r (c) and dierentiable on Br (c) . Moreover, g is not a constant function, for if it were, f would also be a constant function on Br (c) and therefore, by the Identity Theorem, constant on all of U, whichg is not the case by hypothesis. Hence, by the Maximum Modulus Principle, the maximum value of |g | only occurs on the boundary Cr of this disk. That is, there exists a point z ' Cr such that |g (z ) | < |g z ' | for all z Br (c) .
By way of contradiction, suppose be a complex number that is not in But then

1 1 1 1 2 = < < , ' s s/2 |w f (c) | |w f (z ) | s

which gives the desired contradiction. Therefore, the entire disk hence the point it follows that

(7.41)

f (c) belongs to the interior f (U ) is open, as desired.

of the set

f (U ) .

Since this holds for any point

Bs/2 (f (c)) belongs to f (U ) , and c U,

Now we can give the version of the Inverse Function Theorem for complex variables.

Theorem 7.11:
Let

be a piecewise smooth geometric set, and suppose

f :SC

is continuously dierentiable

at a point

c = a + bi,

and assume that

f ' (c) = 0.

Then:

1. There exists an 2. 3.

r > 0, such that B r (c) S, for which f is one-to-one on B r (c) . f (c) belongs to the interior of f (S ) . 1 If g denotes the restriction of the function f to Br (c) , then g is one-to-one, g is dierentiable 1 ' ' at the point f (c) , and g f (c) = 1/f (c) .

Proof:
f is not one-to-one on any disk B r (c) . Then, for each ' ' n, there must exist two points zn = xn + iyn and zn = x' n + iyn such that ' ' |zn c| < 1/n,|zn c| < 1/n, and f (zn ) = f zn . If we write f = u + iv, then we would have that ' u (xn , yn ) u x' n , yn = 0 for all n. So, by part (c) of Exercise 4.11.4, there must exist for each n
Arguing by contradiction, suppose that natural number a point ^

xn , y n ,

such that

xn , y n tialu tialx

is on the line segment joining ^ ^

zn
^

and ^

' , zn

and for which

' 0 = u (xn , yn ) u x' n , yn =

xn , y n

xn x' n +

tialu tialy

xn , y n

' yn yn .

(7.42)

Similarly, applying the same kind of reasoning to joining

v, there must exist points ( xn , y n ) on the segment

zn

to

' zn

such that

0=

If we dene vectors

tialv tialv ' ( xn , y n ) xn x' ( xn , y n ) yn yn . n + tialx tialy

(7.43)

Un

and

Vn

by

Un =
and

tialu tialx

xn , y n ,

tialu tialy

xn , y n

(7.44)

Vn =

tialv tialv ( xn , y n ) , ( xn , y n ) , tialx tialy

(7.45)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

200

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS

then we have that both

U n and V n are perpendicular to the nonzero vector

Therefore, U n and V n are linearly dependent, whence

' xn x' n , yn yn

det

tialu tialx
^

xn , y n
^

tialu tialy

xn , y n

tialv tialv ( xn , y n ) ( xn , y n ) tialx tialy

the point

= 0.(7.46)
and the partial

Now, since both derivatives of

{xn + i y n } and {x n + iy n } converge to u and v are continuous at c, we deduce that det

c = a + ib,

tialu tialv tialv tialu (a, b) (a, b) (a, b) (a, b) tialx tialy tialx tialy
187, this would imply that

= 0.

(7.47)

Now, from Theorem 7.2, p. Hence, there must exist an Because (2). Now write

is one-to-one on

Mapping Theorem, the

r > 0 for which f is Br (c) ,f is obviously not a constant function. So, by the Open point f (c) belongs to the interior of the range of f, and this proves part f
to the disk

f ' (c) = 0, and this is a contradiction. one-to-one on B r (c) , and this proves part (1).

for the restriction of

Br (c) .

Then

is one-to-one. According to part

(2) of Theorem 4.2, p. 84, we can prove that

g 1

is dierentiable at

f (c)

by showing that

1 g 1 (z ) g 1 (f (c)) = ' . z f (c) f (c) z f (c) lim

That is, we need to show that, given an then

(7.48) such that if

> 0,

there exists a

>0

0 < |z f (c) | <

g 1 (z ) g 1 (f (c)) 1 ' | < . z f (c) f (c) 1/w

is continuous at the point

(7.49) there exists an

First of all, because the function that if

f ' (c) ,

' > 0

such

|w f (c) | < ,

'

'

then

|
Next, because

1 1 | < . w f ' (c) ' > 0

such that if

(7.50)

is dierentiable at

c,

there exists a

0 < |y c| < '

then

|
Now, by Theorem 3.11, p. 67,

f (y ) f (c) f ' (c) | < ' . yc

is continuous at the point

(7.51)

g 1

f (c) ,

and therefore there exists a

>0

such that if

|z f (c) | <

then

|g 1 (z ) g 1 f (c) | < ' .(7.52)

So, if

|z f (c) | < ,

then

|g 1 (z ) c| = |g 1 (z ) g 1 (f (c)) | < ' .

But then,

(7.53)

f g 1 (z ) f (c) f ' (c) | < ' , g 1 (z ) c

(7.54)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

201

from which it follows that

|
as desired.

g 1 (z ) g 1 (f (c)) 1 ' | < , z f (c) f (c)

(7.55)

7.7 Uniform Convergence of Analytic Functions

Part (c) of Exercise 4.8.1 gives an example showing that the uniform limit of a sequence of dierentiable functions of a real variable need not be dierentiable. Indeed, when thinking about uniform convergence of functions, the fundamental result to remember is that the uniform limit of continuous functions is continuous (Theorem 3.18, The uniform limit of continuous functions is continuous., p. 76). The functions in Exercise 4.8.1 were dierentiable functions of a real variable. The fact is that, for functions of a complex variable, things are as usual much more simple. The following theorem is yet another masterpiece of Weierstrass.

Theorem 7.12:
U

Suppose

is an open subset of

verges uniformly to a function functions on an open set

Proof:

C, and that {fn } is a sequence of analytic functions on U that conf. Then f is analytic on U. That is, the uniform limit of dierentiable U in the complex plane is also dierentiable on U.

Though this theorem sounds impressive and perhaps unexpected, it is really just a combination of Theorem 6.10, p. 169 and the Cauchy Integral Formula. Indeed, let be such that to

B r (c) U.
on

Then the sequence

this closed disk. Moreover, for any

c be a point in U, and let r > 0 {fn } converges uniformly to f on the boundary Cr of z Br (c) , the sequence {fn ( ) / ( z )} converges uniformly

f ( ) / ( z )

Cr .

Hence, by Theorem 6.10, p. 169, we have

f (z )

= = =
n

limfn (z ) lim 21 i
1 2i fn ( ) d Cr z f ( ) d. Cr z
(7.56)

Hence, by part (a) of Theorem 7.7, Fundamental Theorem of Algebra, p. 195, a Taylor series around

is expandable in

c,

i.e.,

is analytic on

U.

7.8 Isolated Singularities, and the Residue Theorem

integrals can hardly be overstated.

The rst result we present in this section is a natural extension of Theorem 7.3, Cauchy's Theorem, Fundamental Theorem of Analysis, p. 188. However, as we shall see, its consequences for computing contour

Theorem 7.13:
S

CS has nite length. Suppose c1 , ..., cn S, and that r1 , ..., rn are positive numbers such that the 0 closed disks {B rk (ck )} are contained in S and pairwise disjoint. Suppose f is continuous on S \ Brk (ck ) , i.e., at each point of S that is not in any of the open disks Brk (ck ) , and that f
Let be a piecewise smooth geometric set whose boundary are distinct points in the interior

S0

of

7 This 8 This

content is available online at <http://cnx.org/content/m36241/1.2/>. content is available online at <http://cnx.org/content/m36242/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

202

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS
is dierentiable on

S 0 \ B rk (ck ) ,

i.e., at each point of

S0

that is not in any of the closed disks

B rk (ck ) .

Write

Ck

for the circle that is the boundary of the closed disk

B rk (ck ) .

Then

f ( ) d =
CS k=1 Ck

f ( ) d.

(7.57)

Proof:
This is just a special case of part (d) of Exercise 7.2.3. Let

be continuous on the punctured disk

is undened at the central point

c.

Such points

B ' r (c) , analytic at each point z c are called isolated singularities

in of

' Br (c) , and suppose f f, and we wish now to

classify these kinds of points. Here is the rst kind: A complex number c is called a removable singularity of an analytic function f if there exists an ' r > 0 such that f is continuous on the punctured disk B ' r (c) , analytic at each point in Br (c) , and limzc f (z ) exists.

Denition 7.1:

Exercise 7.8.1
a. Dene b. For

f (z ) = sinz/z for all z = 0. Show that 0 is a removable singularity of f. z = c, dene f (z ) = (1 cos (z c)) / (z c) . Show that c is a removable singularity
2

of

f.
c. For d.

z = c, dene f (z ) = (1 cos (z c)) /(z c) . Show that c is still a removable singularity of f. Let g be an analytic function on Br (c) , and set f (z ) = (g (z ) g (c)) / (z c) for all z ' (c) . Show that c is a removable singularity of f. Br f
at

The following theorem provides a good explanation for the term removable singularity. The idea is that this is not a true singularity; it's just that for some reason the natural denition of made.

has not yet been

Theorem 7.14:
Let

be continuous on the punctured disk

'

B r (c)

'

and dierentiable at each point of the open singularity of

punctured disk for all 1.

Br (c) , and assume that c is a removable ' (c) = limzc f (z ) . Then (c) , and f z Br
analytic on the entire open disk

f.

Dene

by f (z ) = f (z ) f

is f

Br (c) ,

whence

f (z ) =
k=0
for all
' z Br (c) .

ck (z c)

(7.58)

2. For any piecewise smooth geometric set and for which

S Br (c) ,

whose boundary

CS

has nite length,

c S0, f ( ) d = 0.
CS
(7.59)

Proof:
As in part (a) of Exercise 7.3.2, dene

on

Br (c)

by

F (z ) =
Then, by that exercise,

1 2i

f ( ) d. Cr z
We show next that

(7.60)

is analytic on

Br (c) .

(z ) F (z ) = f

on

Br (c) ,

and

this will complete the proof of part (1).

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

203

Let z be a point in Br (c) that is not equal ( ) f (c) | < < |z c|/2 and such that |f we have that

to if

c, and let > 0 be given. Choose > 0 such that | c| < . Then, using part (c) of Exercise 7.2.5, (z ) f

= f (z ) =

1 2i

f ( ) C r z

1 2i

f ( ) C z

d =
(7.61)

F (z )

1 2i

(c) f ( )f z C

1 2i

(c) f C z

d =

F (z )
disk

1 2i

( )f (c) f z C

d,
on the Fundamental Theorem

where the last equality holds because the function of Analysis, p. 188. So,

(c) / ( z ) is an analytic function of f

B (c) , and hence the integral is 0 by Theorem 7.3, Cauchy's Theorem, (z ) F (z ) | = | 1 |f 2i = =

( )f (c) f d | z C ( )f (c)| |f 1 ds 2 C | z | 1 2 C /2 ds 2

(7.62)

2. (z ) = F (z ) f
for all

Since this holds for arbitrary Finally, since

> 0,

we see that

z=c

in

Br (c) .

(c) = limf (z ) = limz cF (z ) = F (c) , f

z c
the equality of

(7.63)

and

on f

all of

Br (c)

is proved. This nishes the proof of part (1).

Exercise 7.8.2
Prove part (2) of the preceding theorem. Now, for the second kind of isolated singularity:

Denition 7.2:

A complex number

is called a

pole

of a function

if there exists an

continuous on the punctured disk removable singularity of

B ' r (c) ,

analytic at each point of

r > 0 such that f is ' Br (c) , the point c is not a

function function

Exercise 7.8.3
a. Let

f, and there exists a positive integer k such that the analytic k (z c) f (z ) has a removable singularity at c. A pole c of f is said to be of order n, if n is the smallest positive integer for which the (z ) (z c)n f (z ) has a removable singularity at c. f c n of Br (c) .
for all

be a pole of order

a function

f,

and write

(z ) = (z c)n f (z ) . f f.

Show that

is f

analytic on some disk b. Dene

f (z ) = sinz/z 3

z = 0.

Show that 0 is a pole of order 2 of

Theorem 7.15:
Let that

f be continuous on a punctured c is a pole of order n of f. Then

disk

B ' r (c) ,

analytic at each point of

' Br (c) ,

and suppose

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

204

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS
1. For all
' z Br (c) ,

f (z ) =
k=n

ak (z c) . K
of
' Br (c) .

(7.64)

2. The innite series of part (1) converges uniformly on each compact subset 3. For any piecewise smooth geometric set and satisfying

S Br (c) ,

whose boundary

CS

has nite length,

c S0, f ( ) d = 2ia1 ,
CS
(7.65)

where

A1

is the coecient of

(z c)

in the series of part (1).

Proof:
For each

' z Br (c) ,

write

(z ) = (z c)n f (z ) . f

Then, by Theorem 7.14, p. 202,

is f

analytic on

Br (c) ,

whence

f (z )

= = =

1 (z c)n k=n

(z ) f (z c)n k=0 ck (z

c)
k

(7.66)

ak (z c) ,
' Br (c) ,

where

ak = cn+k .

This proves part (1).

We leave the proof of the uniform convergence of the series on each compact subset of i.e., the proof of part (2), to the exercises.

Part (3) follows from Cauchy's Theorem (Theorem 7.3, Cauchy's Theorem, Fundamental Theorem of Analysis, p. 188) and the computations in Exercise 7.2.2. Thus:

CS

f ( ) d

= = = =

Cr k=n

Cr k=n

f ( ) d ak (z c) d ( c) d Cr
k k
(7.67)

ak

a1 2i,

as desired. The summation sign comes out of the integral because of the uniform convergence of the series on the compact circle

Cr .

Exercise 7.8.4
a. Complete the proof to part (2) of the preceding theorem. series That is, show that the innite
'

K of Br (c) . HINT: n Use the fact that the Taylor series c ( z c ) for f converges uniformly on the entire n=0 n disk B r (c) , and that if c is not in a compact subset K of Br (c) , then there exists a > 0
converges uniformly on each compact subset such that b. Let

k=n

ak (z c)

f, c,

and

| z c| > be as f

for all

z K. (n1) (c) f . (n 1)!

' Br (c)

in the preceding proof. Show that

a1 =
c. Suppose

(7.68)

is a function dened on a punctured disk

that is given by the formula

g (z ) =
k=n

ak (z c)

(7.69)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

205

for some positive integer

and for all

an = 0.

Show that

is a pole of order

' z Br (c) . n of g.

Suppose in addition that the coecient

Having dened two kinds of isolated singularities of a function

f,

the removable ones and the polls of nite

order, there remain all the others, which we collect into a third type.

Denition 7.3:
Let

be continuous on a punctured disk

is called an

essential singularity

B ' r (c) ,

and analytic at each point of

' Br (c) .

The point

of

if it is neither a removable singularity nor a poll of any

nite order.

singularities.
For

Singularities that are either poles or essential singularities are called

nonremovable

Exercise 7.8.5
z = 0,

Theorem 7.16:
Let that

dene

f (z ) = e1/z .

Show that 0 is an essential singularity of

f.
' Br (c) ,

f be continuous on a punctured disk B ' r (c) , c is an essential singularity of f. Then

' (c) , z Br

analytic at each point of

and suppose

1. For all

f (z ) =
k=
where the sequence

ak (z c) , N

(7.70)

k<N
is, if to

such that

{ak } ak = 0.

has the property that for any negative integer

there is a That

2. The innite series in part (1) converges uniformly on each compact subset

of

Fn

is dened by

on the

Fn (z ) = compact set K. c S0,

we have

n k=n

ak (z c) , then the sequence {Fn } converges uniformly S Br (c) ,

whose boundary

' Br (c) .

3. For any piecewise smooth geometric set and satisfying

CS

has nite length,

f ( ) d = 2ia1 ,
CS
where

(7.71)

a1

is the coecient of

(z c)

in the series of part (1).

Proof:
Dene numbers

{ak }

as follows.

ak =
Note that for any

1 2i

f ( )
Cr

( c)

k+1

d.

(7.72)

0<<r

we have from Cauchy's Theorem that

ak =
where

1 2i

f ( )
C

( c)

k+1

d,

(7.73)

denotes the boundary of the disk

B (c) .

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

206

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS
Let

z = c

be in

Br (c) ,

and choose

> 0

such that

< |z c|.

Then, using part (c) of

Exercise 7.2.5, and then mimicking the proof of Theorem 7.5, p. 192, we have

f (z )

= = = = = = = =

f ( ) f ( ) d 21 i C z d Cr z f ( ) f ( ) 1 1 2i Cr ( c)(z c) d + 2i C (z c)( c) d f ( ) f ( ) 1 1 1 1 d c d + 2i 2i Cr c 1 z C z c 1 c c z c k j f ( ) f ( ) c z c 1 d + 21 d k=0 c j =0 z c 2i Cr c i C z c k j f ( ) 1 1 j =0 2i C f ( ) ( c) d (z k=0 2i Cr ( c)k+1 d (z c) + k 1 k f ( ) 1 k= 2i C ( c)k+1 d (z c) k=0 ak (z c) + k 1 k f ( ) 1 k=0 ak (z c) + k= 2i Cr ( c)k+1 d (z c) k k= ak (z c) , 1 2i

c)

j 1

(7.74)

which proves part (1). We leave the proofs of parts (2) and (3) to the exercises.

Exercise 7.8.6
a. Justify bringing the summation signs out of the integrals in the calculation in the preceding proof. b. Prove parts (2) and (3) of the preceding theorem. Compare this with Exercise 7.8.4.

7.3: REMARK
orem 7.15, p. singularity the point

The representation of

f (z )

in the punctured disk 205 is called the

203 and Theorem 7.16, p.

Laurent expansion
f,

' (c) Br

given in part (1) of Theof

around the

c.

Of course it diers from a Taylor series representation of

as this one contains neg-

ative powers of

z c.

In fact, which negative powers it contains indicates what kind of singularity

is.

Non removable isolated singularities of a function disk centered at the singularity equals

share the property that the integral of

2ia1 ,

where the number

a1

is the coecient of

f around a 1 (z c) in the

f at c, and denote it by Rf (c) .

Laurent expansion of

around

c.

This number

2ia1

is obviously signicant, and we call it the

residue of

Combining Theorem 7.13, p. 201, Theorem 7.15, p. 203, and Theorem 7.16, p. 205, we obtain:

Theorem 7.17:

Residue Theorem

S be a piecewise smooth geometric set whose boundary has nite length, let c1 , ..., cn be points 0 in S , and suppose f is a complex-valued function that is continuous at every point z in S except 0 the ck 's, and dierentiable at every point z S except at the ck 's. Assume nally that each ck is a nonremovable isolated singularity of f. Then
Let

f ( ) d =
CS
That is, the contour integral around

Rf (ck ) .
k=1

(7.75)

Exercise 7.8.7 Exercise 7.8.8

below.

CS

is just the sum of the residues inside

S.

Prove Theorem 7.17, Residue Theorem, p. 206.

Use the Residue Theorem to compute That is, determine the poles of then evaluate the integrals.

CS inside

f ( ) d for the functions f and geometric sets S given S, their orders, the corresponding residues, and

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

207

a. b. c. d. e. f.

f (z ) = sin (3z ) /z 2 , and S = B 1 (0) . f (z ) = e1/z , and S = B 1 (0) . 2 f (z ) = e1/z , and S = B 1 (0) . f (z ) = (1/z (z 1)) , and S = B 2 (0) . f (z ) = 1 z 2 /z 1 + z 2 (2z + 1)
4 2

and

S = B 2 (0) . S = B r (0)
for any

f (z ) = 1/ 1 + z

= 1/ z i

z +i

and

r > 1.

The Residue Theorem, a result about contour integrals of functions of a complex variable, can often provide a tool for evaluating integrals of functions of a real variable.

Example 7.1
Consider the integral

1 dx. 1 + x4

(7.76)

Let us use the Residue Theorem to compute this integral. Of course what we need to compute is

B B

lim

1 dx. 1 + x4

(7.77)

The rst thing we do is to replace the real variable function

f (z ) = 1/ 1 + z 4

is analytic everywhere except at the four points

See part (f ) of the preceding exercise. These are hence are the poles of the function Next, given a positive number

x by a complex variable Z, and observe that the ei/4 and e3i/4 . the four points whose fourth power is 1, and

f. B, we consider the geometric set (rectangle) SB that is determined by the interval [B, B ] and the two bounding functions l (x) = 0 and u (x) = B. Then, as long 0 i/4 as B > 1, we know that f is analytic everywhere in S except at the two points c1 = e and 3i/4 c2 = e , so that the contour integral of f around the boundary of SB is given by 1 d = Rf (c1 ) + Rf (c2 ) . 1 + 4 CSB
(7.78)

Now, this contour integral consists of four parts, the line integrals along the bottom, the two sides, and the top. The magic here is that the integrals along the sides, and the integral along the top, all tend to 0 as

tends to innity, so that the integral along the bottom, which after all is what

we originally were interested in, is in the limit just the sum of the residues inside the geometric set.

Exercise 7.8.9

Verify the details of the preceding example. a. Show that

B B

lim

1 1 + (B + it) 1 1 + (t + iB )
4 4

dt = 0.

(7.79)

b. Verify that

B B
c. Show that

lim

dt = 0.

(7.80)

1 dx = 2. 1 + x4

(7.81)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

208

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS

Methods similar to that employed in the previous example and exercise often suce to compute integrals of real-valued functions. However, the method may have to be varied. For instance, sometimes the appropriate geometric set is a rectangle below the may shed some light.

x-axis

instead of above it, sometimes it should be a semicircle instead

of a rectangle, etc. Indeed, the choice of contour (geometric set) can be quite subtle. The following exercise

Exercise 7.8.10
a. Compute

eix dx 1 + x4 eix dx. 1 + x4

(7.82)

and

(7.83)

b. Compute

sin (x) dx 1 + x3 sinx dx. 1 + x3

(7.84)

and

(7.85)

Example 7.2

sinx/x dx. The techniques described above don't immediately apply to this function, for, even replacing the x by a z, this function has no poles, so
An historically famous integral in analysis is that the Residue Theorem wouldn't seem to be much help. Though the point 0 is a singularity, it is a removable one, so that this function

sinz/z

is essentially analytic everywhere in the complex

plane. However, even in a case like this we can obtain information about integrals of real-valued

sinx/x dx is the imaginary part of eix /x dx, so that we may as well iz evaluate the integral of this function. Let f be the function dened by f (z ) = e /z, and note that
Notice rst that 0 is a pole of order 1 of a geometric set

functions from theorems about integrals of complex-valued functions.

f, and that the residue Rf (0) = 2i. Now, for each B > 0 and > 0 dene SB, , determined by the interval [B, B ] , as follows: The upper bounding function uB, is given by uB, (x) = B, and the lower bounding function lB, is given by lB, (x) = 0 for B x and x B, and lB, (x) = eix/ for < x < . That is, SB, is just like the rectangle SB in Example 1 above, except that the lower boundary is not a straight line. Rather, the lower boundary is a straight line from B to , a semicircle below the x-axis of radius from to , and a straight line again from to B.
By the Residue Theorem, the contour integral

f ( ) d = Rf (0) = 2i.
CSB,
As in the previous example, the contour integrals along the two sides and across the top of tend to 0 as of

(7.86)

SB,

tends to innity. Finally, according to part (e) of Exercise 6.5.2, the contour integral

along the semicircle in the lower boundary is

independent of the value of

So,

B 0
implying then that

lim lim

ei d = i, graph(lB, )

(7.87)

sinx dx = . x

(7.88)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

209

Exercise 7.8.11
a. Justify the steps in the preceding example. In particular, verify that

B B

lim

0 B

ei(B +it) dt = 0, B + it ei(t+iB ) dt = 0, t + iB

(7.89)

B
and

lim

(7.90)

ei d = i, C C
is the semicircle of radius

(7.91)

where

centered at the origin and lying below the

x-axis.
(7.92)

b. Evaluate

sin2 x dx. x2

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

210

CHAPTER 7. THE FUNDAMENTAL THEOREM OF ALGEBRA, AND THE FUNDAMENTAL THEOREM OF ANALYSIS

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

Appendix: Existence and Uniqueness of a Complete Ordered Field

1

This appendix is devoted to the proofs of Theorem 1.1, p. of Dedekind, which he presented in the late 1800's.

13 and Theorem 1.2, p.

13, which together

assert that there exists a unique complete ordered eld. Our construction of this eld will follow the ideas

Denition 8.1:
By a

Dedekind cut,

or simply a

cut,

we will mean a pair

(A, B )

of nonempty (not necessarily

disjoint) subsets of the set 1.

of rational numbers for which the following two conditions hold.

A B = Q.

That is, every rational number is in one or the other of these two sets.

2. For every element

aA

and every element

b B,A b.

That is, every element of

is less

than or equal to every element of

B.

Recall that when we dene the rational numbers as quotients (ordered pairs) of integers, we faced the problem that two dierent quotients determine the same rational number, e.g., equivalence among Dedekind cuts.

2/3 6/9. There is a similar

Denition 8.2:
b2 B 2 ,

Two Dedekind cuts

Exercise 8.1
b. Let

and

a2 b1

(A1 , b1 ) and (A2 , B2 ) are called equivalent if a1 b2 for all a1 A1 and all for all a2 A2 and all b1 B1 . In such a case, we write (A1 , B1 ) (A2 , B2 ) .

a. Show that every rational number equivalent.

determines three distinct Dedekind cuts that are mutually

be the set of all positive rational numbers

whose square is greater than 2, and let

comprise all the rationals not in

B.

Prove that the pair

(A, B )

is a Dedekind cut. Do you

think this cut is not equivalent to any cut determined by a rational number Can you prove this?

as in part (a)?

c. Prove that the denition of equivalence given above satises the three conditions of an equivalence relation. Namely, show that i. (Reexivity) iii.

ii. (Symmetry) If

(A, B ) is equivalent to itself. (A1 , B1 ) (A2 , B2 ) , then (A2 , B2 ) (A1 , B1 ) . (Transitivity) If (A1 , B1 ) (A2 , B2 ) and (A2 , B2 ) (A3 , B3 ) , then (A1 , B1 ) (A3 , B3 ) .

There are three relatively simple-sounding and believable properties of cuts, and we present them in the next theorem. It may be surprising that the proof seems to be more dicult than might have been expected.

Theorem 8.1:
Let

(A, B )

be a Dedekind cut. Then

1 This

content is available online at <http://cnx.org/content/m36243/1.2/>.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

211

212

APPENDIX
1. If 2. If 3.

a A and a' < a, then a' A. b B and b' > b, then b' B. Let be a positive rational number. b a < .

Then there exists an

a A

and a

b B

such that

Proof:
Suppose

is an element of

A,

and let

a' < a

be given. By way of contradiction suppose that

'

does not belong to

A.

Then, by Condition (1) of the denition of a cut, it must be that

a' a B.

But then, by Condition (2) of the denition of a cut, we must have that contradiction, because

a a' ,

and this is a

a < a.

'

This proves part (1). Part (2) is proved in a similar manner.

To prove part (3), let the rational number > 0 be given, and set r = /2. Choose an element a0 A and an element b0 B. Such elements exist, because A and B are nonempty sets. Choose a natural number N such that a0 + N r > b0 . Such a natural number N must exist. For instance, just choose N to be larger than the rational number (b0 a0 ) /r. Now dene a sequence {ak } of rational numbers by ak = a0 + kr, and let K be the rst natural number for which aK B. Obviously, such a number exists, and in fact K must be less than or equal to N. Now, aK 1 is not in B, so it must be in A. Set a = AK 1 and b = AK . Clearly, a A,b B, and

b a = aK aK 1 = a0 + Kr a0 (K 1) r = r =
and this proves part (3).

< , 2

(8.1)

We will make a complete ordered eld cuts. We will call this eld the

Dedekind eld.

whose elements are the set of equivalence classes of Dedekind To make this construction, we must dene addition and

multiplication of equivalence classes of cuts, and verify the six required eld axioms. Then, we must dene the set

that is to be the positive elements of the Dedekind eld

F, and then verify the required properties of

an ordered eld. Finally, we must prove that this eld is a complete ordered eld; i.e., that every nonempty set that is bounded above has a least upper bound. First things rst. If (A1 , B1 ) and (A2 , B2 ) are Dedekind cuts, dene the sum of (A1 , B1 ) and (A2 , B2 ) to be the cut (A3 , B3 ) described as follows: B3 is the set of all rational numbers b3 that can be written as b1 + b2 for some b1 B1 and b2 B2 , and A3 is the set of all rational numbers r such that r < b3 for all b3 B3 . Several things need to be checked. First of all, the pair from the denition that every element of

Denition 8.3:

A3

is less than or equal to every element of

(A3 , B3 ) is again a Dedekind cut. Indeed, it is clear B3 , so that Condition

r be a rational number, and suppose that it is not in r / A3 , there must exist an element b3 = b1 + b2 B3 for which r > b3 . Otherwise, r would be in A3 . But this means that r b2 > b1 , and so by part (2) of ' ' Theorem 8.1, p. 211, we have that r b2 is an element b1 of B1 . Therefore, r = b1 + b2 , implying that r B3 ,
(2) is satised. To see that Condition (1) holds, let We must show that

A3 .

belongs to

B3 .

Now, since

as desired. We dene the

0 cut

to be the pair

A0 = {r : r 0}

and

B0 = {r : r > 0}.

This cut is one of the three

determined by the rational number 0.

Exercise 8.2

a. Prove that addition of Dedekind cuts is commutative and associative. b. Prove that if c. d.

(A1 , B1 ) (C1 , D1 ) and (A2 , B2 ) (C2 , D2 ) , then (A1 , B1 ) + (A2 , B2 ) (C1 , D1 ) + (C2 , D2 ) . Find an example of a cut (A, B ) such that (A, B ) + 0 = (A, B ) . Prove that (A, B ) + 0 (A, B ) for every cut (A, B ) . F
of all equivalence classes of Dedekind cuts as follows:

We dene addition in the set

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

APPENDIX

213

Denition 8.4:
If is the equivalence class of the cut

x is the equivalence class of a cut (A, b) and y is the equivalence class of a cut (C, D) , then x + y (A, B ) + (C, D) . F
is well-dened, commutative, and associative.

It follows from the previous exercise, that addition in We are on our way. We dene the element 0 of

to be the equivalence class of the 0 cut. The next theorem establishes one

of the important eld axioms for

F,

A' = B, i.e., the set of all the negatives of the elements of B, and let B ' = A, i.e., the set of ' ' all the negatives of the elements of A. It is immediate that the pair A , B is a Dedekind cut. Let ' ' ' ' us show that (A, B ) + A , B is equivalent to the zero cut. Let (C, D ) = (A, B ) + A , B . Then, by the denition of the sum of two cuts, we know that D consists of all the elements of the form d = b + b' = b a, where b B and a A. Since a b for all a A and b B, we see then that the elements of D are all greater than or equal to 0. To see that (C, D ) is equivalent to the 0 cut, it will suce to show that D contains all the positive rational numbers. (Why?) Hence, let > 0 be given, and choose an a A and a b B such that b a < . This can be done by Condition (3) of Theorem 8.1, p. 211. Then, the number b a D, and hence, by part (2) of Theorem 8.1, p. 211, D. It follows then that the cut (C, D ) is equivalent to the zero cut (A0 , B0 ) , as desired.
Let We will write

Proof:

(A, B ) is a Dedekind cut, then there exists a cut A' , B ' such that (A, B )+ A' , B ' is equivalent to the 0 cut. Therefore, if x is an element of F, then there exists an element y of F such that x+y = 0.
If

Theorem 8.2:

namely, the existence of an additive inverse for each element of

F.

(A, B )

Exercise 8.3
a. Suppose b. Prove

for the cut

A' , B '

of the preceding proof.

0 cut. Show that

(A, B ) is a cut, and let (C, D) be a cut for which (A, B ) + (C, D) is equivalent to the (C, D) A' , B ' = (A, B ) . that the additive inverse of an element x of the Dedekind eld F is unique. F,
is a bit more tricky. In fact, we will

The denition of multiplication of cuts, as well as multiplication in rst introduce the notion of positivity among Dedekind cuts.

Denition 8.5:

Exercise 8.4
that

A Dedekind cut

x = (A, B )

is called

positive

if

contains at least one positive rational number.

a. Suppose

(A, B ) and (C, D) (C, D) also is positive.

are equivalent cuts, and assume that

(A, B )

is positive.

Prove

Make the obvious denition of positivity in the set

F.

b. Show that the sum of two positive cuts is positive. Conclude that the sum of two positive c. Let

F, i.e., the sum of two equivalence classes of positive cuts, is positive. (A, B ) be a Dedekind cut. Show that one and only one of the following three properties holds for (A, B ) . (i) (A, B ) is a positive cut, (ii) (A, B ) is a positive cut, or (iii) (A, B ) is
elements of

equivalent to the 0 cut. d. Establish the law of tricotomy for three properties holds for an element

F : That is, show that one and only one of the following x F. (i) x is positive, (ii) x is positive, or (iii) x = 0.

We rst dene multiplication of cuts when one of them is positive.

Denition 8.6:
Let all

(A1 , B1 )

and

(A2 , B2 )

cut. We dene the

product(A3 , B3 ) of (A1 , B1 ) and (A2 , B2 ) as follows:

b1 b2
for some

be two Dedekind cuts, and suppose that one of these cuts is a positive Set

b3

that can be written as

numbers

for which

r < b3

for all

b1 B 1 b3 B3 .

and

b2 B 2 .

Then set

B3 equal to the set of A3 to be all the rational

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

214

APPENDIX
Again, things need to be checked.

Exercise 8.5

a. Show that the pair

(A3 , B3 )

of the preceding denition for the product of positive cuts is in

fact a Dedekind cut. b. Prove that multiplication of Dedekind cuts, when one of them is positive, is commutative. c. Suppose

(A1 , B1 )

is a positive cut. Prove that (8.2)

(A1 , B1 ) ((A2 , B2 ) + (A3 , B3 )) = (A1 , B1 ) (A2 , B2 ) + (A1 , B1 ) (A3 , B3 )

for any cuts positive d. Show that, if

(A2 , B2 ) and (A3 , B3 ) . (A1 , B1 ) (A2 , B2 ) and (C1 , D1 ) (C2 , D2 ) cuts, then (A1 , B1 ) (C1 , D1 ) (A2 , B2 ) (C2 , D2 ) . F.

and

(a1 , B1 )

and

(A2 , B2 )

are

e. Show that the product of two positive cuts is again a positive cut. We are ready to dene multiplication in

Denition 8.7:
Let

be elements of F. x or y is positive, dene the product x y to be the equivalence class of the cut (A, B ) (C, D) , where x is the equivalence class of (A, B ) and y is the equivalence class of (C, D) . If either x or y is 0, dene x y to be 0. If both x and y are negative, i.e., both x and y are positive, dene x y = (x) (y ) . and If either The next exercise is tedious. It amounts to checking a bunch of cases.

Exercise 8.6

a. Prove that multiplication in b. Prove that multiplication in c. Prove that multiplication in

F F F

is commutative. is associative. is distributive over addition.

d. Prove that the product of two positive elements of We dene the element 1 of

is again positive.

to be the equivalence class of the cut

A1 , B 1 ,

where

A1 = {r : r 1}

and

B = {r : r > 1}.

Exercise 8.7
a. Prove that the elements 0 and 1 of b. Prove that

are not equal.

x1=x

for every element

x F. xy = 1
and

c. Use the associative law and part (b) to prove that if

xz = 1,

then

y = z.

Theorem 8.3:
With respect to the operations of addition and multiplication dened above, together with the denition of positive elements,

Proof:

is an ordered eld.

F in the preceding F is a eld. Thus, let x F be a nonzero element. We must show the existence of an element y of F for which x y = 1. Suppose rst that x is a positive element of F. Then x is the equivalence class of a positive cut (A, B ) , and therefore A contains some positive rational numbers. Let a0 be a positive number that is contained in A. It follows then that every element of B is greater than or equal to a0 and hence
The rst ve axioms for a eld, given in Section 1.2, have been established for exercises, so that we need only verify axiom 6 to complete the proof that is positive. Dene dene pair

to be the set of all rational numbers

for which ^

r 1/b for every b B. Then

^ It follows directly that the

to be the set of all rationals is a Dedekind cut.

for which

r b

for every

b B .

A, B

^ ^

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

APPENDIX
Let

215

(C, D) = (A, B ) A, B , F.

^ ^

and note that every element

dD

is of the form

d = b b, A1 , B 1

and that

hence is greater than or equal to 1. We claim that determines the element 1 of

(C, D)

is equivalent to the cut

D contains every rational number r that is greater than 1. Thus, let r > 1 be given, and set = a0 (r 1) . From Condition (3) of ' ' ' ' Theorem 8.1, p. 211, choose an a A and a b B such that b a < . Without loss of generality,
To see this we must verify that we may assume that

a' a0 .

Finally, set

b = 1/a' .

Clearly

b 1/b

for all

b B,

so that

b B .

Also

d = b' b D,

and ^

d = b' b =
implying that

b' a' + b' a' = <1+ ' 1+ = r, ' a a' a a0 (C, D)

is equivalent to the cut

(8.3) implying that

r D.

Therefore,

A1 , B 1 ,

(A, B )

A, B

^ ^

is equivalent to the cut ^ ^

A1 , B 1 . Therefore, if y
as desired.

is the element of

that is the equivalence

class of the cut If

A, B , z

then

x y = 1,

is negative, then

is positive. If we write

for the multiplicative inverse of the positive the the

element

x,

then

is the multiplicative inverse of the element

product of two negative elements of The properties that guarantee

x. Indeed, by the denition of F,x (z ) = (x) z = 1. that F is an ordered eld also have been established in

preceding exercises, so that the proof of this theorem is complete.

So, the Dedekind eld is an ordered eld, but we have left to prove that it is complete. This means we must examine upper bounds of sets, and that requires us to understand when one cut is less than another one. We say that a cut that an element

(A, B )

is

x y

Theorem 8.4:
Let

in the ordered eld

less than or equal to a cut C, D if a d for every a A and d D. We say F is less than or equal to an element y if y x is either positive or 0.
F,
and suppose

and

be elements of

the equivalence class of the cut

Proof:

(C, D) .

Then

x is the equivalence class of the cut (A, B () x y if and only if (A, B ) (C, D) .

and

is

as before,

x y if and only if the element y x = y + x is positive or 0. Writing, A' , B ' for the cut (A, B ) , we have that y x is the equivalence class of the cut (C, d) (A, B ) = (C, D)+ A' , B ' , so we need to determine when the cut (G, H ) = (C, D)+ A' , B ' is a positive cut or the 0 cut; which is the case when the set H only contains nonnegative numbers. ' By denition of addition, the set H contains all numbers of the form h = d + b for some d D ' ' ' and some b B . Since B = A, this means that H consists of all elements of the form h = d a for some d D and a A. Now these numbers h are all greater than or equal to 0 if and only if each a A is less than or equal to each d D, i.e., if and only if (A, B ) (C, D ) . This proves the
We have that theorem.

We are now ready to present the rst of the two main theorems of this appendix, that is Theorem 1.1, p. 13 in Section 1.4.

Theorem 8.5:
There exists a complete ordered eld. Indeed, the Dedekind eld Let

Proof:
S

is a complete ordered eld.

be a nonempty subset of

F,

and suppose that there exists an upper bound for

S;

i.e., an

element class of

M of F (A, B ) .

such that

xM

for all

We must show that

x S. Write (A, B ) for a cut such that M there exists a least upper bound for S.

is the equivalence

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

216

APPENDIX
x S, let (Ax , Bx ) be a Dedekind cut for which x is the equivalence class of (Ax , Bx ) , ax b for all ax Ax and all b B. Let A0 be the union of all the sets Ax for x S. Let B0 be the set of all rational numbers r for which r a0 for every a0 A0 . we claim rst that the pair (A0 , B0 ) is a Dedekind cut. Both sets are nonempty; A0 because it is the union of nonempty sets, and B0 because it contains all the elements of the nonempty set B. Clearly Condition (2) for a cut holds from the very denition of this pair. To see Condition (1), let r be a rational number that is not in B0 . We must show that it is in A0 . Now, since r is not in B0 , there must exist some a0 A0 for which r < a0 . But a0 xS Ax , so that there must exist an x S such that a0 Ax , and hence r is also in Ax . But then r A0 , and this proves that (A0 , B0 ) is a Dedekind cut. Let M0 be the equivalence class determined by the cut (A0 , B0 ) . Since each Ax A0 , we see that ax b0 for every ax Ax and every b0 B0 . Hence, (Ax , Bx ) (A0 , B0 ) for every x S, and therefore, by Theorem A.4, x M0 for all x S. This shows that M0 is an upper bound for S. ' ' ' ' Finally, suppose M is another upper bound for S, and let A , B be a cut for which M is ' ' ' ' ' the equivalence class of A , B . Then ax b for every ax Ax and every b B , implying ' ' ' ' ' that a0 b for every a0 A0 and every b B . Therefore, (A0 , B0 ) A , B , implying that M0 M ' . This shows that M0 is the least upper bound for S, and the theorem is proved.
For each and note that We come now to the second major theorem of this appendix, i.e., Theorem 1.2, p. 13 of Section 1.4. This

one asserts the uniqueness, up to isomorphism, of complete ordered elds.

Theorem 8.6:
F
^

Let

be a complete ordered eld. Then there exists an isomorphism of

onto the Dedekind eld and that satises

F.

That is, there exists a one-to-one function 1. 2. 3.

J :F F

that is onto all of

F,

J (x + y ) = J (x) + J (y ) . J (xy ) = J (x) J (y ) . If x > 0, then J (x) > 0.

Proof:
We know from Section 1.1 that, inside any ordered eld, there is a subset that is isomorphic to the eld

Q x

of rational numbers. We will therefore identify this special subset of is an element of

with

Q.

If

F,

let

Ax = {r Q : r x} ax Ax

and let

that the pair

(Ax , Bx )

is a Dedekind cut. Indeed, from the denition of is less than or equal ^

Condition (2), i.e., that each Condition (1) also holds, let either

Bx = {r Q : r > x}. We claim rst Ax and Bx , we see that to each bx Bx , holds. To see that
^ an ordered eld,

rx

or

r > x,

i.e.,

r be a rational number in F . Then, because F is r Ax or r Bx . Hence, (Ax , Bx ) is a Dedekind cut. F

^ into

We dene a function by the cut

from

by setting

J (x) equal to the equivalence class determined

be elements of

(Ax , Bx ) . J

We must check several things.

First of all,

is one-to-one. Indeed, let

and

that are not equal. Assume

without loss of generality that

x < y.

Then, according to , which is a theorem about complete ^

ordered elds and hence applicable to

F ,, there exist two rational numbers r1 and r2 such that x < r1 < r2 < y, which implies that r1 Bx and r2 Ay . Since r2 > r1 , the cut (Ay , By ) is not equivalent to the cut (Ax , Bx ) , and therefore J (x) = J (y ) . Next, we claim that the function J is onto all of the Dedekind eld F. Indeed, let z be an element of F, and let (A, B ) be a Dedekind cut for which z is the equivalence class determined by (A, B ) .
Think of In fact,

A as a subset of the complete ordered eld F . Then A is nonempty and is bounded above. every element of B is an upper bound of A. Let x = supA. (Here is another place where
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

APPENDIX
we are using the completeness of the eld

217

F .) We claim that the cut (A, B ) is equivalent to the cut (Ax , Bx ) , which will imply that J (x) = z. Thus, if ax Ax , then ax x, and x b for every b B, because x is the least upper bound of A. Similarly, if a A, then a x, and x < bx for every bx Bx . This proves that the cuts (A, B ) and (Ax , Bx ) are equivalent, as desired. x and y are elements of F , and bx Bx and by By , then bx > x and by > y, so that bx + by > x + y, and therefore bx + by Bx+y for every bx Bx and by By . On the other hand, if r Bx+y , then r > x + y. Therefore, r x > y, implying, again by Theorem 1.8, p. 18, that there exists an element by By such that y < by < r x. But then r by > x, which means that r by = bx for some bx Bx . So, r = bx + by , and this shows that Bx+y = bx + By . It follows from this that the cuts (Ax+y , Bx+y ) and (Ax , Bx ) + (Ay , By ) are equal, and therefore
If ^

J (x + y ) = J (x) + J (y ) .
If

A consequence of this is that ^

J (x) = J (x)

for all

x F .

and

are two positive elements of ^

paragraph shows that

F , then an argument just like the one in the preceding J (xy ) = J (x) J (y ) . Then, since J (x) = J (x) , the fact that J (xy ) =
follows. ^

J ( x) J ( y )

for all

x, y F

Finally, if

is a positive element of

F,

then the set

Ax

must contain some positive rationals,

and hence the cut

(Ax , Bx )

is a positive cut, implying that

J (x) > 0. F
^ and

We have veried all the requirements for an isomorphism between the two elds the theorem is proved.

F,

and

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

218

GLOSSARY

Glossary

A bounded real-valued function if, given any

f on a closed bounded interval [a, b] is called Riemann-integrable > 0, there exist step functions k and l, on [a, b] for which k (x) f (x) l (x) for all x, such that (l k ) < . We denote the set of all functions on [a, b] that are Riemann-integrable by IR ([a, b]) . c
is called a

A complex number

pole

of a function

if there exists an

continuous on the punctured disk removable singularity of

B ' r (c) ,

analytic at each point of

r > 0 such that f is ' Br (c) , the point c is not

f,

(z c) f (z ) has a removable singularity at c. A pole c of f is said to be of order n, if n is the smallest (z ) (z c)n f (z ) has a removable singularity at c. f
A complex number

and there exists a positive integer

such that the analytic function

positive integer for which the function

c is called a removable singularity of an analytic function f if there exists an ' r > 0 such that f is continuous on the punctured disk B ' r (c) , analytic at each point in Br (c) , and limz c f (z ) exists. x = (A, B )
is called

A Dedekind cut

positive
R2

if

contains at least one positive rational number.

continuous real-valued functions

= P dx + Qdy, and is determined by two is bounded or uniformly continuous if the functions P and Q are bounded or uniformly continuous functions on U. We say that the dierential form is smooth of order k if the set U is open, and the functions P and Q have continuous mixed partial derivatives of order k. If = P dx + Qdy is a dierential form on a set U, and if C is any piecewise smooth curve of nite length contained in U, then we dene the line integral of over C by C
of is denoted by

dierential form on a subset U

and

on

U.

We say that

=
C
where

P dx + Q dy =
C 0

P ( (t)) x' (t) + Q ( (t)) y ' (t) dt, C

by arc length.

(6.48)

(t) = (x (t) , y (t)) F

is called an

is a parameterization of

A eld

ordered eld

if there exists a subset

P F

that satises the following two

properties:

If If

x, y P, then x + y and xy are in P. x F, then one and only one of the following three statements x P, x P, and x = 0. (This property is known as the law of tricotomy.)

is true.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

GLOSSARY
A

219

eld

is a nonempty set

on which there are dened two binary operations, addition (+) and

multiplication (), such that the following six axioms hold: Both addition and multiplication are commutative and associative. Multiplication is distributive over addition; i.e.,

x (y + z ) = x y + x z
for all

(1.6)

x, y, z F. F,
which we will denote by

There exists an element in

0,

that is an identity for addition; i.e.,

x+0=x

for all

There exists a If If

nonzero element in F, which we will denote by 1, that is an identity for

x1=x
of for all

x F.

multiplication; i.e.,

x F.

x F, xF

then there exists a unique element

called the element

additive inverse
x = 0,

yF

such that

x + y = 0.
such that

This element

is

and is denoted by of

x. yF
and is denoted by

and

then there exists a unique element

is called the

multiplicative inverse
S
equals

x y = 1. x1 .

This

A function

whose domain

its domain. It is called an

odd

S,

is called an

even function if f (z ) = f (z ) for all z

for all

in

function if

f (z ) = f (z )
on

in its domain.

A function

f :SC

is called

uniformly continuous

if for each positive number

exists a positive number

such that

|f (x) f (y ) | <

for all

x, y S

satisfying

, there |x y | < .

A nonzero variable,

polynomial

or

polynomial function is a complex-valued function of a complex

n

p : C C,

that is dened by a formula of the form

p (z ) =
k=0
where the

ak z k = a0 + a1 z + a2 z 2 + ... + an z n ,

(3.11)

polynomial

ak 's are complex numbers and an = 0. The integer n is called the degree of the p and is denoted by deg (p) . The numbers a0 , a1 , ..., an are called the coecients of the polynomial. The domain of a polynomial function is all of C ; i.e., p (z ) is dened for every complex number z.
For technical reasons of consistency, the identically 0 function is called the of its coecients are 0 and its degree is dened to be A

zero polynomial.

All

. r (z ) = p (z ) /q (z ) ,

rational function is a function r

q S
of all

that is given by an equation of the form

where

is a nonzero polynomial and

is a (possibly zero) polynomial. The domain of a rational

function is the set

zC

for which

q (z ) = 0,

i.e., for which

r (z )

is dened.

partition of a closed geometric set S

S.
0 {Si }
are called the

in

R2

is a nite collection

{S1 , S2 , ..., Sn }

of

nonoverlapping closed geometric sets for which geometric set

n i=1 Si = S ;

i.e., the union of the

Si 's

is all of the

The open subsets A

step function on the closed geometric set S

P = {Si } partition P.
of

elements

of the partition.

is a real-valued function for all

exists a partition element of the

such that

h (z ) = ai

0 z Si ;

i.e.,

h on S for which there h is constant on each

A real number called an

irrational number.

that is not a rational number, i.e., is not an element of the subset

of

R,

is

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

220

GLOSSARY sequence

of real or complex numbers is dened to be a function from the set

of natural

numbers into the setR or number

C.

Instead of referring to such a function as an assignment

n f (n) ,

we ordinarily use the notation

{an },{an } 1 ,

or

{a1 , a2 , a3 , ...}.

Here, of course,

an

denotes the

f (n) .

A sequence

{an } of real numbers is called nondecreasing if an an+1 for all n, and it is called nonincreasing if an an+1 for all n. It is called strictly increasing if an < an+1 for all n, and strictly decreasing if an > an+1 for all n. A sequence {an } of real numbers is called eventually nondecreasing if there exists a natural number N such that an an+1 for all n N, and it is called eventually nonincreasing if there
exists a natural number

such that

an an+1

for all

n N.

We make analogous denitions of

eventually strictly increasing and eventually strictly decreasing.

A sequence

{an }

of real or complex numbers is a

Cauchy
then

sequence if for every

> 0,

there exists

a natural number

such that if

nN

and

mN

|an am | < . M
such that

A subset in

of

is called

Bounded

if there exists a real number

|z | M

for every

S. F
is called

An ordered eld

complete

if every nonempty subset

of

that has an upper bound

has a least upper bound.

B
By a (open)

R = {(x, y ) : a < x < b and c < y < d}. The analogous denition of a closed rectangle [a, b] [c, d] should be clear: [a, b] [c, d] = {(x, y ) : a x b, c y d}. By the area of a (open or closed) rectangle R = (a, b) (c, d) or [a, b] [c, d] we mean number A (R) = (b a) (d c) . .

rectangle

we will mean a set

R = (a, b) (c, d)

in

R2 .

That is,

the

By a

Dedekind cut, or simply a cut, we will mean a pair (A, B ) of nonempty (not necessarily
Q
of rational numbers for which the following two conditions hold.

disjoint) subsets of the set

A B = Q. That is, every rational number is in one or the other of these two sets. For every element a A and every element b B,A b. That is, every element of A than or equal to every element of B.
By a

is less

that is the range of a 1-1, smooth, function

z2 in the plane, we mean a set C C : [a, b] C, where [a, b] is a bounded closed ' interval in R, where z1 = (a) and z2 = (b) , and satisfying (t) = 0 for all t (a, b) . 2 More generally, if : [a, b] R is 1-1 and piecewise smooth on [a, b] , and if {t0 < t1 < ... < tn } ' is a partition of [a, b] such that (t) = 0 for all t (ti1 , ti ) , then the range C of is called a piecewise smooth curve from z1 = (a) to z2 = (b) . In either of these cases, is called a parameterization of the curve C.
from a point

smooth curve

z1

to a dierent point

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

GLOSSARY
By a

221

V (x, y ) (P (x, y ) , Q (x, y )) of the vector eld V .

We will also speak of component functions

vector eld

on an open subset from

U of R2 , we mean nothing more than a continuous function U into R2 . The functions P and Q are called the components

both of whose

smooth vector elds, by which we will mean vector elds V

P
and

have continuous partial derivatives

tialP tialP tialQ tialQ , , and tialx tialy tialx tialy

on

(6.46)

U. R
of

By the set

real numbers

we mean the (unique) complete ordered eld.

D
Dene a power series function, denoted by exp, as follows:

exp (z ) =

zn . n! n=0

(3.38)

We will call this function, with 20-20 hindsight, the

exponential function.

Dene two power series functions cosh (hyperbolic cosine) and sinh (hyperbolic sine) by

cosh (z ) =

exp (z ) + exp (z ) 2

and

sinh (z ) =

exp (z ) exp (z ) , 2

(3.39)

and two other power series functions cos (cosine) and sin (sine) by

cos (z ) = cosh (iz ) =

and

exp (iz ) + exp (iz ) 2

(3.40)

sin (z ) = isinh (iz ) =

The ve functions just dened are called the functions are called the basic

exp (iz ) exp (iz ) . 2i

(3.41)

elementary transcendental functions, the sinh and hyperbolic functions, and the sine and cosine functions are called the basic trigonometric or circular functions. The connections between the hyperbolic
cosh
functions and hyperbolic geometry, and the connection between the trigonometric functions and circles and triangles, will only emerge in the next chapter. From the very denitions, however, we can see a connection between the hyperbolic functions and the trigonometric functions. It's something like interchanging the roles of the real and imaginary axes. This is probably worth some more thought.

F
For

a positive real number and

an arbitrary complex number, dene

az

by

az = exp (zln (a)) .

(4.72)

G
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

222

GLOSSARY
Given and

[a, b] , l, and u as in the above, let S be the set of all pairs (x, y ) R2 , for which a < x < b l (x) < y < u (x) . Then S is called an open geometric set. If we replace the < signs with signs, i.e., if S is the set of all (x, y ) such that a x b, and l (x) y u (x) , then S is called a closed geometric set. In either case, we say that S is bounded on the left and right by the vertical line segments {(a, y ) : l (a) y u (a)} and {(b, y ) : l (b) y u (b)}, and it is bounded below by the graph of the function l and bounded above by the graph of the function u. We call the union of these four bounding curves the boundary of S, and denote it by CS . If the bounding functions u and l of a geometric set S are smooth or piecewise smooth functions, we will call S a smooth or piecewise smooth geometric set.

If

(A1 , B1 ) and (A2 , B2 ) are Dedekind cuts, dene the sum of (A1 , B1 ) and (A2 , B2 ) to be the (A3 , B3 ) described as follows: B3 is the set of all rational numbers b3 that can be written as b1 + b2 for some b1 B1 and b2 B2 , and A3 is the set of all rational numbers r such that r < b3 for all b3 B3 .
cut

If

f and g are two complex-valued functions with the same domain S, i.e., f : S C and g : S C, and if c is a complex number, we dene f + g, f g, f /g (if g (x) is never 0), and cf
by the familiar formulas:

(f + g ) (x) = f (x) + g (x) , (f g ) (x) = f (x) g (x) , (f /g ) (x) = f (x) /g (x) ,

and

(3.5)

(3.6)

(3.7)

(cf ) (x) = cf (x) .

If

(3.8)

and

are real-valued functions, we dene functions

max (f, g )

and

min (f, g )

by

[max (f, g )] (x) = max (f (x) , g (x))

(the maximum of the numbers

(3.9)

f (x)

and

g (x)),

and

[min (f, g )] (x) = min (f (x) , g (x)) ,

(the minimum of the two numbers If

(3.10)

f (x)

and

g (x)). S,
then we say that

bounded
If

is either a real-valued or a complex-valued function on a domain if there exists a positive number

is

M r S

such that

|f (x) | M

for all

x S.

convergence of the power series. the disk of convergence.

f

is a power series function, the number The disk

of the preceding theorem is called the of radius

radius of
is called

around 0, denoted by

Br (0) ,

If

is a real-valued function on a closed geometric set

in the plane, then

is

integrable on S

if

it is the uniform limit of a sequence

{hn }

of step functions on

S.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

GLOSSARY
We dene the

223

integral

of an integrable function

on

by

f
S
where

f (z ) dz = lim
S S

hn , f.

(5.102)

{hn }

is a sequence of step functions on

that converges uniformly to

If

is an ordered eld, and

that

xy

if either

We say that

x and y are elements x < y or x = y. x > y if y < x, and x y if y x.

of

F,

we say that

x<y

if

y x P.

We say

If

bounded below
An element for

S is a subset of an ordered eld F, then an element x F is called an upper bound for S if x y for every y S. An element z is called a lower bound for S if z y for every y S. A subset S of an ordered eld F is called bounded above if it has an upper bound; it is called
if it has a lower bound; and it is called

bounded

if it has both an upper bound

and a lower bound.

M is called the least upper bound or supremum of a set S if it is an upper bound M x for every other upper bound x of S. That is, M is less than or equal to any other upper bound of S. Similarly, an element m is called the greatest lower bound or inmum of S if it is a lower bound for S and if z m for every other lower bound z of S. That is, m is greater than or equal to any other lower bound of S. S
and if If

is a positive real number, then the symbol

which

y 2 = x.

will denote the unique positive number

for

Of course,

denotes the number 0.

If

x is an element of a eld F, dene inductively elements n x nx of F by 1 x = x, and, if k x is dened, set (k + 1) x = x + k x. The set S of all natural numbers n for which n x is dened is therefore, by the axiom of mathematical induction, all of N. x is the equivalence class of a cut (A, b) and y is the equivalence x + y is the equivalence class of the cut (A, B ) + (C, D) . z = x + yi
is in class of a cut

If

(C, D) ,

then

If

C,

we dene the

absolute value
|z | =

of

by

x2 + y 2 .
and

(1.40)

We dene the

distance d (z, w) between two complex numbers z

we dene the

by

d (z, w) = |z w|. If c C and r > 0,

open disk of radius r around c, and denote it by Br (c) , by

Br (c) = {z C : |z c| < r}.
(1.41)

The

closed disk

of radius

around

is denoted by

B r (c)

and is dened by

B r (c) = {z C : |z c| r}.

(1.42)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

224

GLOSSARY
We also dene open and closed

punctured

disks

' Br (c)

and

B r (c)

'

around

by

' Br (c) = {z : 0 < |z c| < r}

(1.43)

and

B r (c) = {z : 0 < |z c r}.

These punctured disks are just like the regular disks, except that they do not contain the central point

'

(1.44)

c.

S is any subset of C, we dene the open neighborhood of radius r around S, Nr (S ) , to be the set of all z such that there exists a w S for which |z w| < r. That is, Nr (S ) is the set of all complex numbers that are within a distance of r of the set S. We dene the closed neighborhood of radius r around S, and denote it by N r (S ) , to be the set of all z C for which there exists a w S such that |z w | r.
More generally, if denoted by If

z = x + yi, z = x + yi

we say that the real number

that the real number If

is the

imaginary part

is the of

is a complex number, dene the

x = (z ) . (z ) . complex conjugate z of z by z = x yi.

of and write

real part

and write

z y=

We say

L
Let

(A1 , B1 ) b3

and

(A2 , B2 )

cut. We dene the of all rational numbers

product(A3 , B3 ) of (A1 , B1 ) and (A2 , B2 ) as follows:

for which

be two Dedekind cuts, and suppose that one of these cuts is a positive Set

that can be written as

b1 b2 r < b3

for some for all

b1 B 1 b3 B 3 .

and

b2 B2 .

Then set

B3 equal to the set A3 to be all the

Let

[a, b] be a closed bounded interval in R. A real-valued function h : [a, b] R is called a step function if there exists a partition P = {x0 < x1 < ... < xn } of [a, b] such that for each 1 i n ai
such that

there exists a number

h (x) = ai

for all

x (xi1 , xi ) . f : [a, b] R
If is called

Let

integrable
Let the

[a, b]

be a closed bounded interval of real numbers. A function on

[a, b] f,

if it is the uniform limit of a sequence

{hn }

of step functions.

integral

I ([a, b])

denote the set of all functions that are integrable on of denoted

[a, b] .

f I ([a, b]) ,

dene

f,

by

f = lim
where

hn , f
on

(5.19)

{hn }

is some (any) sequence of step functions that converges uniformly to

[a, b] .

As in the case of step functions, we use the following notations:

f=
a
Let set

f=
a

f (t) dt.

(5.20)

[a, b] be a closed bounded interval of real numbers. By a partition of [a, b] we mean a nite P = {x0 < x1 < ... < xn } of n + 1 points, where x0 = a and xn = b. The n intervals {[xi1 , xi ]} are called the closed subintervals of the partition P, and the n intervals {(xi1 , xi )} are called the open subintervals or elements of P. We write P for the maximum of the numbers (lengths of the subintervals) {xi xi1 }, and call P the mesh size of the partition P.
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

GLOSSARY
If a partition

225

then we say that Let

be a

P = {xi } is contained in another partition Q = {yj }, i.e., each xi equals some yj , Q is ner than P. function on an interval [a, b] , and let P = {x0 < ... < xn } be a partition of [a, b] .
n

Physicists often consider sums of the form

SP,{yi } =
i=1
where

f (yi ) (xi xi1 ) ,

These sums (called Riemann sums) are

(5.1)

yi

is a point in the subinterval

(xi1 , xi ) .

approximations of physical quantities, and the limit of these sums, as the mesh of the partition becomes smaller and smaller, should represent a precise value of the physical quantity. What precisely is meant by the  limit of such sums is already a subtle question, but even having decided on what that denition should be, it is as important and dicult to determine whether or not such a limit exists for many (or even any) functions

f.

We approach this question from a

slightly dierent point of view, but we will revisit Riemann sums in the end.

Let set

[a, b] be a closed bounded interval of real numbers. By a partition of [a, b] we mean a nite P = {x0 < x1 < ... < xn } of n + 1 points, where x0 = a and xn = b. The n intervals {[xi1 , xi ]}, for 1 i n, are called the closed subintervals of the partition P, and the n intervals {(xi1 , xi )} are called the open subintervals of P. We write P for the maximum of the numbers (lengths of the subintervals) {xi xi1 }, and call the number P the mesh size of the partition P. A function h : [a, b] C is called a step function if there exists a partition P = {x0 < x1 < ... < xn } of [a, b] and n numbers {a1 , a2 , ..., an } such that h (x) = ai if xi1 < x < xi . That is, h is a step function if it is a constant function on each of the (open) subintervals (xi1 , xi ) determined by a partition P. Note that the values of a step function at the points {xi } of the partition are not restricted in any way. A function l : [a, b] R is called a polygonal function, or a piecewise linear function, if there exists a partition P = {x0 < x1 < ... < xn } of [a, b] and n + 1 numbers {y0 , y1 , ..., yn } such that for each x [xi1 , xi ] ,l (x) is given by the linear equation l (x) = yi1 + mi (x xi1 ) ,
where each of the closed subintervals (3.12)

mi = (yi yi1 ) / (xi xi1 ) . That is, l is a polygonal function if it is a linear function on [xi1 , xi ] determined by a partition P. Note that the values of a piecewise linear function at the points {xi } of the partition P are the same, whether we think of xi in the interval [xi1 , xi ] or [xi , xi+1 ] . (Check the two formulas for l (xi ) .) The graph of a piecewise linear function is the polygonal line joining the n + 1 points {(xi , yi )}. There is a natural generalization of the notion of a step function that works for any domain S, e.g., a rectangle in the plane C. Thus, if S is a set, we dene a partition of S to be a nite collection {E1 , E2 , ..., En } of subsets of S for which n i=1 Ei = S, and Ei Ej = if i = j.
Then, a

step function on S

would be a function

that is constant on each subset

Ei .

We will

encounter an even more elaborate generalized notion of a step function in Chapter V, but for now we will restrict our attention to step functions dened on intervals

[a, b] .

The set of polynomials and the set of step functions are both closed under addition and multiplication, and the set of rational functions is closed under addition, multiplication, and division.

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

226

GLOSSARY

Let

integrable

[a, b]

be a xed bounded and closed interval. A complex-valued function if its real and imaginary parts

f = u + iv

is called

and

are integrable. In this case, we dene

f=
a
Let

(u + iv ) =
a a

u+i
a

v.

(5.32)

[a, b] be a xed closed bounded interval in R. We dene the integral of a step function h on [a, b] , and denote it by h, as follows: If P = {x0 < x1 < ... < xn } is a partition of [a, b] , for which h (x) = ai for all x (xi1 , xi ) , then
n

h = SP (h) =
i=1
Let

ai (xi xi1 ) .

(5.5)

and

be real numbers for which

a < b.

By the

real numbers numbers

for which

for which

a < x < b, a x b.

and by the

open interval (a,b) we mean the set of all closed interval [a,b] we mean the set of all real

Let

be a natural number. We dene inductively natural numbers

an

as follows:

a1 = a,

and,

whenever

is dened, then

k+1

is dened to be

aa .

Let

(complex-valued) function The by

z1 to z2 in the plane C, parameterized by a : [a, b] C. If f is a continuous, complex-valued function on C, contour integral of f from z1 to z2 along C will be denoted by C f ( ) d or more precisely z2 f ( ) d, and is dendd by C z1
be a piecewise smooth curve from

b C
Let

z2 z1 f

( ) d =
a

f ( (t)) ' (t) dt.

(6.41)

be a piecewise smooth curve in the plane. The

length or arc lengthL L (C ) of C

is

dened by the formula

L (C ) = L = supL P,
P
where If

(6.24)

and

is any parameterization of C. w are two points on a piecewise

smooth curve

C,

we will denote by

L (z, w)

the arc

length of the portion of the curve between

and

w.

Let

C be a piecewise smooth curve of nite length L joining distinct points, and let : [0, L] C be a parameterization of C by arc length. By a partition of C we mean a set {z0 , z1 , ..., zn } of points on C such that zj = (tj ) for all j, where the points {t0 < t1 < ... < tn } form a partition of the interval [0, L] . The portions of the curve between the points zj 1 and zj , i.e., the set (tj 1 , tj ) , are called the elements of the partition. A step fucntion on C is a real-valued function h on C for which there exists a partition {z0 , z1 , ..., zn } of C such that h (z ) is a constant aj on the portion of the curve between zj 1 and zj .

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

GLOSSARY
Let

227

integrable with respect to arc length on C if it is the uniform limit of step functions on C. The integral with respect to arc length of an integrable function f on C is again denoted by
C

be a piecewise smooth curve of nite length

L.

A function

with domain

is called

f (s) ds,

and is dened by

f (s) ds = lim
C
where

hn (s) ds,
C

(6.34)

{hn }

is a sequence of step functions that converges uniformly to

on

C.
a

Let

C,

the range of

: [a, b] C,

be a piecewise smooth curve, and let

point on the curve. We say that the curve parameterization

has a tangential direction at

z = (x, y ) = (c) be z, relative to the

if the following limit exists:

lim

(t) (c) (t) z = lim . tc | (t) (c) | tc | (t) z | R2 , )

to

(6.7)

If this limit exists, it is a vector of length 1 in (relative to the parameterization The curve

and this unit vector is called the unit tangent

has a

unit tangent

at the

z. point z
at

if there exists a parameterization

for which the

unit tangent at

relative to

exists.

Let

be a eld, and let

For each natural number

xk

is dened, set

Dene

x0

to be

xk+1 1. 0
m

x be a nonzero element of F. n, we dene inductively an element xn in F as follows: x1 = x, = x xk . Of course, xn is just the product of nx's. n,
dene

and, if

For each natural number Finally, we dene

xn

to be the multiplicative inverse

to be 0 for every positive integer

m,

and we

(xn ) n leave 0

of the element

xn .

and

undened.

Let

f be a real or complex-valued function on the open interval (a, b) where a is possibly b is possibly +. We say that f is improperly-integrable on (a, b) if it is integrable on each ' ' closed and bounded subinterval a , b (a, b) , and for each point c (a, b) we have that the
and two limits

b' c f and lima' a+0 a' f exist. c More generally, We say that a real or complex-valued function

limb' b 0 (a, b) ,

the open interval such that

is

improperly-integrable
f

f,

not necessarily dened on all of

on

(a, b)

if there exists a partition

{xi }

of

[a, b]
by

is dened and improperly-integrable on each open interval

(xi1 , xi ) . (a, b)

We denote the set of all functions

that are improperly-integrable on an open interval

Ii ((a, b)) .
Let

be continuous on a punctured disk

is called an

essential singularity

B ' r (c) ,

and analytic at each point of

' Br (c) .

The point

of

if it is neither a removable singularity nor a poll of any

nite order. Singularities that are either poles or essential singularities are called

singularities.
Let

nonremovable integral

be dened and improperly-integrable on an open interval

(a, b) .

We dene the

of

over the interval

(a, b) ,

and denote it by

b a

f,

by

b'

f = lim
a

a' a+0

f + lim
a'

b' b0

f.
c

(5.81)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

228

GLOSSARY
In general, if dene the

is improperly-integrable over an open interval, i.e.,

is dened and

improperly-integrable over each subinterval of

integral

(a, b)
by

determined by a partition

{xi },

then we

of

over the interval

(a, b)
b

xi

f=
a i=1 xi1

f.

(5.82)

Let

Taylor polynomial

be in

C n (Br (c))

for

a xed complex number,

r > 0,

and

of degree

for

at

to be the polynomial

n a positive integer. Dene the T n T(n f,c) given by the formula:

(4.115)

T(n f,c) (z ) =
j =0
where

aj (z c) ,

aj = f (j ) (c) /j !.

Let and

F1 and F2 be two ordered elds, and write P1 and P2 for the set of positive elements in F1 F2 respectively. A 1-1 correspondence J between F1 and F2 is called an isomorphism if

J (x + y ) = J (x) + J (y ) for all x, y F1 . J (xy ) = J (x) J (y ) for all x, y F1 . x P1 if and only if J (x) P2 .
Let

f :SC

be a function, where

an element of

S.

We say that

S C, and let c be a limit point of S that f has a limit L as z approaches c, and we write
z c

is not necessarily

limf (z ) = L,

(4.1)

if for every If the

> 0 there exists a > 0 such that if z S and 0 < |z c| < , then |f (z ) L| < . domain S is unbounded, we say that f has a limit L as z approaches , and we write L = lim f (z ) ,
z
(4.2)

if for every > 0 there exists a positive number B such that if z S and |z | B, then |f (z ) L| < . Analogously, if S R, we say limx f (x) = L if for every > 0 there exists a real number B such that if x S and x B, then |f (x) L| < . And we say that limx f (x) = L if for every > 0 there exists a real number B such that if x S and x B, then |f (x) L| < . Finally, for f : (a, b) C a function of a real variable, and for c [a, b] , we dene the one-sided (left and right) limits of f at c. We say that f has a left hand limit of L at c, and we write L = limxc0 f (x) , if for every > 0 there exists a > 0 such that if x (a, b) and 0 < c x < then |f (x) L| < . We say that f has a right hand limit of L at c, and write L = limxc+0 f (x) , if for every > 0 there exists a > 0 such that if x S and 0 < x c < then |f (x) L| < .

Let

f : S R be a function whose domain is a subset S of R2 , and let c = (a, b) be a point in S 0 of S. We say that f is dierentiable, as a function of two real variables, at the point (a, b) if there exists a pair of real numbers L1 and L2 and a function such that
the interior

f (a + h1 , b + h2 ) f (a, b) = L1 h1 + L2 h2 + (h1 , h2 )
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

(4.140)

GLOSSARY
and

229

|(h1 ,h2 )|0 | (h1 , h2 ) |

One should compare this denition with part (3) of Each partial derivative of a function
2

lim

(h1 , h2 )
.

= 0.

(4.141)

is again a real-valued function of two real variables, and

so it can have partial derivatives of its own. We use simplifying notation like to indicate higher order mixed partial derivatives. For instance, partial derivative of to

f,

rst with respect to

x,

second with

fxyxx and fyyyxyy... fxxyx denotes the fourth respect to x again, third with respect

partial derivatives.
Let of

y,

and nally fourth with respect to

x.

These higher order partial derivatives are called

mixed

f : S R be a real-valued function of a real variable, and let c be an element of the interior S. Then f is said to attain a local maximum at c if there exists a > 0 such that (c , c + ) S and f (c) f (x) for all x (c , c + ) . The function f is said to attain a local minimum at c if there exists an interval (c , c + ) S such that f (c) f (x) for all x (c , c + ) . f : S R be dened on a set S C R2 , and let c = (a, b) = + + bi be a point in the interior of S. We dene the partial derivative of f with respect to x at the point c = (a, b) by
formula

Let

the

and the

f (a + h, b) f (a, b) tialf (a, b) = lim , h0 tialx h partial derivative of f with respect to y at c = (a, b) by the

(4.25) formula

whenever these limits exist.

tialf f (a, b + h) f (a, b) (a, b) = lim , h0 tialy h (In both these limits, the variable h is a

(4.26) real variable.)(

It is clear that the partial derivatives of a function arise when we x either the real part of the variable or the imaginary part of the variable to be a constant, and then consider the resulting function of the other (real) variable. We will see in function's being dierentiable at a point partial derivatives at the point
3

that there is a denite dierence between a in the complex plane

(a, b)

in

c = (a + bi) R2 .

versus its having

Let

codomain

f : S T and g : T U be functions. We dene a function g f, with domain S and U, by (g f ) (x) = g (f (x)) . If f : S T,g : T S, and g f (x) = x for all x S, then g is called a left inverse of f. If f g (y ) = y for all y T, then g is called a right inverse for f. If g is both a left inverse and 1 right inverse, then g is called an inverse for f,f is called invertible, and we denote g by f . h S
be a step function on a closed geometric set by the formula

Let set

S.

Dene the

integral

of

over the geometric

h=
S S

H (z ) dz =
i=1

ai A (Si ) ,

(5.99)

2 http://cnx.org/content/m36206/latest/ 3 http://cnx.org/content/m36186/latest/

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

230

GLOSSARY
where

S1 , ..., Sn

is a partition of

for which

is the constant

ai

on the interior

0 Si

of the set

Si .

Let

respect to arc length of h over C

be a step function on a piecewise smooth curve is denoted by

C of nite length L. The integral, with h ( s ) ds, and is dened by C aj L (zj 1 , zj ) ,

(6.32)

h (s) ds =
C
where

j =1

between

{z0 , z1 , ..., zn } zj 1 and zj .

is a partition of

for which

h (z )

is the constant

aj

on the portion of

Let

[a, b] . Suppose P = {x0 < x1 < ... < xn } is a [a, b] such that h (x) = ai on the interval (xi1 , xi ) . Dene the weighted average hrelative to P to be the number SP (h) dened by h
be a step function on the closed interval partition of

of

SP (h) =
i=1
Let

ai (xi xi1 ) .

(5.3)

denote an object whose square

i2 = 1.

Let

be the set of all objects that can be

represented in the form Dene two operations

z = x + yi, where both x + and on C as follows:

and

are real numbers.

(x + yi) + x' + y ' i = x + x' + y + y ' i,

and

(1.34)

(x + iy ) x' + iy ' = xx' + xiy ' + iyx' + iyiy ' = xx' yy ' + xy ' + yx' i.

(1.35)

Let

be a natural number. As earlier in this chapter, we dene

n!

as follows:

n! = n (n 1) (n 2) ... 2 1.
For later notational convenience, we also dene If

(1.19)

0!

to be 1.

is any integer for which

0 k n,

we dene the

binomial coecient

n k

by

n (n 1) (n 2) ... (n k + 1) n n! = . = k ! (n k )! k! k
Let

(1.20)

|f (x) f (c) | < . The function f is called continuous on S if it is continuous at every c of S. If the domain S of f consists of real numbers, then the function f is called right continuous at c if for every > 0 there exists a > 0 such that |f (x) f (c) | < whenever x S and 0 x c < , and is called left continuous at c if for every > 0 there exists a > 0 such that |f (x) f (c) | < whenever x S and 0 x c > .
then point

pointc of S

and

be sets of complex numbers, and let if for every positive

there exists a positive

f : S T. Then f is said to be continuous at a such that if x S satises |x c| < ,

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

GLOSSARY
Let

231

and

be sets. A

function from S

into

(notation

each element

in

a unique element denoted by

f (x)

in

f : S T) T. T.

is a rule that assigns to

It is useful to think of a function as a mechanism or black box. We use the elements of inputs to the function, and the outputs are elements of the set If

as

f : S T is a function, then S is called the domain of f, and the set T is called the codomain f. The range or image of f is the set of all elements y in the codomain T for which there exists an x in the domain S such that y = f (x) . We denote the range by f (S ) . The codomain
of is the set of all potential outputs, while the range is the set of actual outputs.

S into a set T. If A S, we write f (A) for the subset of T t T for which there exists an s A such that t = f (s) . We call f (A) the image of A under f. Similarly, if B T, we write f 1 (B ) for the subset of S 1 containing all the elements s S such that f (s) B, and we call the set f (B ) the inverse 1 image or preimage of B. The symbol f (B ) is a little confusing, since it could be 1 misinterpreted as the image of the set B under a function called f . We will discuss inverse functions later on, but this notation is not meant to imply that the function f has an inverse. If f : S T, then the graph of f is the subset G of the Cartesian product S T consisting of all the pairs of the form (x, f (x)) . If f : S R is a function, then we call f a real-valued function, and if f : S C, then we call f a complex-valued function. If f : S C is a complex-valued function, then for each x S the complex number f (x) can be written as u (x) + iv (x) , where u (x) and v (x) are the real and imaginary parts of the complex number f (x) . The two real-valued functions u : S R and v : S R are called respectively the real and imaginary parts of the complex-valued function f. If f : S T and S R, then f is called a function of a real variable, and if S C, then f is
Suppose

is a function from a set

containing all the elements

called a function of a If the range of The function

complex variable.

f equals f :ST

the codomain, then

is called

onto.
implies that

is called one-to-one if

f (x1 ) = f (x2 )

x1 = x2 .
the right by of

Let

S be a geometric set (either open or closed), bounded on the left by x = a, on x = b, below by the graph of l, and above by the graph of u. Dene the areaA (S )
n

by

A (S ) = supAP =
P

sup
P ={x0 <x1 <...<xn } i=1

(xi xi1 ) (di ci ) , P

of

(5.67)

where the supremum is taken over all partitions are as dened above.

[a, b] ,

and where the numbers

ci

and

di

Let

be a subset of

(respectively

R.

By an

open cover

of

we mean a sequence

subsets of

(respectively

R)

such that

S Un ;

i.e., for every

x S there

exists an

{Un } of open n such that

x Un .

property, if every open cover of S

then there exists an integer

A subset

of

(respectively

R)

is called

compact, or is said to satisfy the Heine-Borel

{Un }
is an open cover of

has a nite subcover. That is, if

S,

such that

S N n=1 Un .

In other words, only a nite number of

the open sets are necessary to cover

S.
be a complex-valued function, and let

Let

be a subset of

C,

Then

is said to be

expandable in a Taylor series around c

let

f :SC

be a point of

S.

with radius of convergence

if there

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

232

GLOSSARY
exists an

r>0

such that

Br (c) S,

and

f (z )

is given by the formula

f (z ) =
n=0
for all Let

an (z c)

(3.51)

z Br (c) .
is said to be

be a subset of

Then

expandable in a Taylor series around c

(c r, c + r) S,
n

R,

let

f :SR

be a real-valued function on and

S,

and let

be a point of

S.

with radius of convergence

if there

exists an

r>0

such that the interval

f ( x)

is given by the formula

f (x) =
n=0

an (x c)

(3.52)

f :SC c of S. Suppose S is an open subset of R. A function f : S C expandable in a Taylor series around every point c of S.
of

x (c r, c + r) . Suppose S is an open subset

for all

C.

A function

is called

analytic on S

if it is

expandable in a Taylor series around every point

is called

real analytic on S

if it is

Let

S be a subset of C. A complex number x is called a limit point {xn } of elements of S such that x = limxn . A set S C is called closed if every limit point of S belongs to S. S
be a subset of

of

if there exists a sequence

Let

that the open disk points of A subset

S = S0.
an

C. A point x S is called an interior point of S if there exists an > 0 such B (x) of radius around x is entirely contained in S. The set of all interior S is denoted by S 0 and we call S 0 the interior of S. S of C is called an open subset of C if every point of S is an interior point of S ; i.e., if S
be a subset of

Analogously, let

x S is called an interior point of S if there exists (x , x + ) is entirely contained in S. Again, we denote 0 0 the set of all interior points of S by S and call S the interior of S. A subset S of R is called an open subset of R if every point of S is an interior point of S ; i.e., if S = S0. R.
A point

>0

such that the open interval

Let

R (or C ), and Let f : S C be a function of a real (or complex) variable. f is continuously dierentiable on S 0 if f is dierentiable at each point x of S 0 and ' 0 1 the function f is continuous on S . We say that f C (S ) if f is continuous on S and 0 0 continuously dierentiable on S . We say that f is 2-times continuously dierentiable on S if ' 0 the rst derivative f is itself continuously dierentiable on S . And, inductively, we say that f 0 is k-times continuously dierentiable on S if the k 1st derivative of f is itself continuously 0 (k) k dierentiable on S . We write f for the k th derivative of f, and we write f C (S ) if f is 0 k continuous on S and is k times continuously dierentiable on S . Of course, if f C (S ) , then (j ) 0 all the derivatives f , for j k, exist nd are continuous on S . (Why?) (0) For completeness, we dene f to be f itself, and we say that f C (S ) if f is continuous on 0 S and has innitely many continuous derivatives on S ; i.e., all of its derivatives exist and are 0 continuous on S . 4 As in , we say that f is real-analytic (or complex-analytic ) on S if it is expandable in a Taylor 0 series around each point c S S
be a subset of We say that

4 http://cnx.org/content/m36192/latest/

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

GLOSSARY

233

Let

be a subset of

R,

let

be an element of the interior of

f : S C be a complex-valued function (of a S. We say that f is dierentiable at c if lim f (c + h) f (c) h

real variable), and let

h0
exists. (Here, the number Analogously, let

(4.8)

is a real number.)

variable), and let

S be a subset of C, let f : S C be a complex-valued function (of a complex c be an element of the interior of S. We say that f is dierentiable at c if f (c + h) f (c) h0 h lim
(4.9)

exists. (Here, the number If

is a complex number.)

f : S C is a function either of a real variable or a complex variable, and if S ' denotes the ' ' subset of S consisting of the points c where f is dierentiable, we dene a function f : S C
by

f ' (x) = lim

The function

h0

f (x + h) f (x) . h

(4.10)

A continuous function

derivative of f. f : [a, b] C that is dierentiable at each point x (a, b) , and whose ' derivative f is continuous on (a, b) , is called a smooth function on [a, b] . If there exists a partition {a = x0 < x1 < ... < xn = b} of [a, b] such that f is smooth on each subinterval [xi1 , xi ] , then f is called piecewise smooth on [a, b] .
f'
is called the Higher order derivatives are dened inductively. That is,

the

f'

'

is the derivative of

f ',

and so on.

We use the symbol

f (n)

for the

nth

derivative of

f.
^

Let

S be a subset of R2 . By the symmetric image (x, y ) R2 for which the point (y, x) S. x
and

of

we mean the set

of all points

Let

If either

y be elements of F. x or y is positive, dene the product x y to be the equivalence class of the cut (A, B ) (C, D) , where x is the equivalence class of (A, B ) and y is the equivalence class of (C, D) . If either x or y is 0, dene x y to be 0. If both x and y are negative, i.e., both x and y are positive, dene x y = (x) (y ) . {an }
be a sequence of real numbers and let to

Let

converge

L,

or that

is the

limit

be a real number. The sequence

{an }

is said to

of

{an },

if the following condition is satised. For every

positive number

there exists a natural number

such that if

n N,

then

|an L| < .

In symbols, we say

L = liman

or

L = lim an .
n
We also may write If a sequence sequence

(2.1)

an L.
numbers converges to a number

{an } of real or complex {an } is convergent.

L,

we say that the

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

234

GLOSSARY
We say that a sequence

{an }

of real numbers

diverges

to

there exists a natural number such a sequence is convergent.

such that if

n N,

then

+ if for every positive number M, an M. Note that we do not say that

{an } of real numbers diverges to if for every real number N such that if n N, then an M. The denition of convergence for a sequence {zn } of complex numbers is exactly the same as for a sequence of real numbers. Thus, let {zn } be a sequence of complex numbers and let L be a complex number. The sequence {zn } is said to converge to L, or that L is the limit of {zn }, if the following condition is satised. For every positive number , there exists a natural number N such that if n N, then |zn L| < .
Similarly, we say that a sequence

M,

there exists a natural number

Let

S denote its cluster set. If S is nonempty and bounded above, we dene lim supan to be the supremum supS of S. If S is nonempty and bounded below, we dene lim inf an to be the inmum inf S of S. If the sequence {an } of real numbers is not bounded above, we dene lim supan to be , and {an } is not bounded below, we dene lim inf an to be . If {an } diverges to , then we dene lim supan and lim inf an both to be . And, if {an } diverges to , we dene lim supan and lim inf an both to be . We call lim supan the limit superior of the sequence {an }, and lim inf an the limit inferior of {an }.
be a sequence of real numbers and let

{an }

if

Let

{an }

be a sequence of real or complex numbers. A number

is called a

cluster point

of the

sequence

{an }

if there exists a subsequence

{bk }

cluster points of a sequence

{an }

is called the

cluster set

of

{an }

such that

x = limbk .

The set of all

of the sequence.

Let

subsequence of {an } is a sequence {bk } {an } together with a strictly increasing sequence {nk } of natural numbers. The sequence {bk } is dened by bk = ank . That is, the k th term of the sequence {bk } is the nk th term of the original sequence {an }.
{an }
be a sequence of real or complex numbers. A that is determined by the sequence real or complex numbers. By the

Let

{an } 0 be a sequence of sequence {SN } dened by

innite series

an

we mean the

SN =
n=0
The sequence

an .

(2.31)

{SN }

is called the

innite series is said to be partial sums converges to An innite series series If

S.The

sequence of partial sums of the innite series an , and the summable to a number S, or to be convergent, if the sequence {SN } of absolutely summable
or

sum of an innite series is the limit of its partial sums.

absolutely convergent
if the innite

an

is called

|an |

is convergent.

an

is not convergent, it is called

convergent, it is called

divergent. If it is convergent but not absolutely conditionally convergent.

an 's
and the

A few simple formulas relating the

SN 's

are useful:

SN = a0 + a1 + a2 + ... + aN ,

(2.32)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

GLOSSARY
SN +1 = SN + aN +1 ,
and

235

(2.33)

SM SK =
n=K +1
for

an = aK +1 + aK +2 + ... + AM ,

(2.34)

M > K.

Let

{an } 0 be a sequence of real or complex numbers. By the power series n functionf (z ) = n=0 an z we mean the function f : S C where the domain S is the set of all z C for which the innite series an z n converges, and where f is the rule that assigns to such a z S the sum of the series. The numbers {an } dening a power series function are called the coecients of the function. n We associate to a power series function f (z ) = n=0 an z its sequence {SN } of partial sums.
We write

SN (z ) =
n=0

an z n .

(3.31)

Notice that polynomial functions are very special cases of power series functions. They are the power series functions for which the coecients that each partial sum

{an } are all 0 beyond some point. Note also SN for any power series function is itself a polynomial function of degree less than or equal to N. Moreover, if f is a power series function, then for each z in its domain we have f (z ) = limN SN (z ) . Evidently, every power series function is a limit of a sequence of
polynomials. Obviously, the domain

S Sf

of a power series function

depends on the coecients

{an }

determining the function. Our rst goal is to describe this domain.

Let

coecient

be a complex number, and let

be a nonnegative integer. We dene the general

binomial

by

( 1) ... ( k + 1) = . (4.126) k k! If is itself a positive integer and k , then k agrees with the earlier denition of the binomial coecient, and = 0 when k > . However, if is not an integer, but just an k arbitrary complex number, then every k = 0.
Let

: [a, b] C

be a parameterization of a piecewise smooth curve

C,

relative to the parameterization

is taken over all partitions

of

, [a, b] .

we mean the number

C C. By the lengthL of L = supP L P , where the supremum

S
Suppose

is a subset of

R2 ,

and that If all

is a continuous real-valued function on

S.

If both partial

derivatives of

exist at each point of the interior

S0

of

S,

and both are continuous on

S0,

then

is said to belong to

C (S ) .

and all of them are continuous on to

k th order mixed S 0 , then f is said

partial derivatives exist at each point of to belong to

S0,

partial derivatives, of arbitrary orders, exist and are continuous on

C (S ) . Finally, if all mixed S 0 , then f is said to belong

C (S ) .
Available for free at Connexions <http://cnx.org/content/col11249/1.1>

236

GLOSSARY

T
The absolute value of a real number

is denoted by

|x|

and is dened as follows:

The

|0| = 0. x > 0 then |x| = x. (iii) If x < 0 (x > 0) then |x| = x.

(i) (ii) If

overlap

of two geometric sets

S1

intersection.

S1

and

S2

are called

nonoverlapping

and

S2

is dened to be the interior if this overlap

(S1 S2 ) of their 0 (S1 S2 ) is the empty set.

and all

Two Dedekind cuts

(A1 , b1 ) and (A2 , B2 ) are called equivalent if a1 b2 for all a1 A1 b2 B2 , and a2 b1 for all a2 A2 and all b1 B1 . In such a case, we write (A1 , B1 ) (A2 , B2 ) .

W
We call the inverse

logarithm function, and we denote this function by ln.

exp1

of the restriction of the exponential function to

the (natural)

We say that the sequence every

{fn }converges or converges pointwise to a function f : S C if for x S and every > 0 there exists a natural number N, depending on x and , such that for every n N,|fn (x) f (x) | < . That is, equivalently, {fn } converges pointwise to f if for every x S the sequence {fn (x)} of numbers converges to the number f (x) . We say that the sequence {fn }converges uniformly to a function f if for every > 0, there exists an N, depending only on , such that for every n N and every x S,|fn (x) f (x) | < . If {un } is a sequence of functions dened on S, we say that the innite series un converges uniformly if the sequence {SN = N u } of partial sums converges uniformly. n=0 n

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

INDEX

237

Index of Keywords and Terms

Keywords are listed by the section with that keyword (page numbers are in parentheses).
apples, 1.1 (1) Keywords do not necessarily appear in the text of the page. They are merely associated with that section.

Terms are referenced by the page they appear on. Ex.

Ex.

apples, 1

absolutely convergent, 2.8(48) addition formulas, 4.6(99) additive inverse, 1.3(7) algebraic identities, 1.7(19) alternating harmonic series, 2.8(48) alternating series test, 2.8(48) analytic functions, 7.7(201) approximation, 1.6(16) arc length, 6.3(159), 6.4(164), 6.6(170) associative, 8(211) axiom of choice, 5.6(135) behavior at innity, 3.3(57) binomial coecient, 4.10(108) Binomial theorem, 1.7(19), 2.4(34), 4.10(108) Bolzano-Weierstrass, 2.6(39) Bolzano-Weierstrass Theorem, 2.7(44) bounded, 3.2(53) catchy mean value theorem, 4.9(106) Cauchy Criterion, 2.6(39) Cauchy Integral Formula, 7.3(192), 7.5(196) cauchy mean value theorem, 4.7(101) cauchy's theorem, 7.2(186), 7.8(201) cauchy-hadamard, 3.7(67) cauchy-riemann, 7.2(186) chain rule, 4.1(81), 4.3(83), 4.11(111) closed bounded interval, 5.3(123) closed dierential forms, 6.7(174) closed geometric set, 5.8(144) closed neighborhood, 1.8(23) closed sets, 3.5(63) closed subintervals, 3.2(53), 5.2(117) codomain, 3.2(53) commutative, 8(211) comparison test, 2.8(48) complement, 2.7(44) complex number, 4.2(81) complex numbers, 1.1(5), 1.8(23) complex plane, 6.5(168)

complex-analytic, 4.8(104) complex-valued function, 3.2(53), 3.4(60), 3.5(63), 3.6(64), 4.2(81), 6.7(174), 7.3(192) complex-valued project, 4.3(83) composition, 3.3(57) composition of continuous functions, 3.4(60) conditionally convergent, 2.8(48) continuity, 3.4(60), 3.5(63), 4.2(81) continuous 1-1 function, 3.6(64) continuous function, 5.4(129) continuously dierentiable, 4.8(104) continuum hypothesis, 5.6(135) contour integrals, 6.5(168) convergence, 2.2(29), 2.3(32), 3.10(76), 5.3(123), 5.7(140) convergent, 2.8(48) convergent sequences, 2.5(36) converges pointwise, 3.10(76) curve of innite length, 6.3(159) Dedekind cut, 8(211) denominator, 1.3(7) denseness, 1.6(16) derivative, 4.2(81), 4.3(83), 5.4(129) dierential, 6.6(170) distributive over addition, 8(211) divergent, 2.8(48) domain, 3.2(53) double integral, 5.6(135) e, 2.4(34) equivalent, 8(211) essential singularity, 7.8(201) Euler's constant, 5.7(140) expandable, 3.9(75) exponential function, 3.8(72) eld, 1.3(7) rst derivative test, 4.4(92), 4.11(111), 7.5(196) function, 3.1(53)

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

238

INDEX
fundamental theorem of algebra, (1), 7.4(195), 7.7(201) fundamental theorem of analysis, 7.2(186), 7.7(201) fundamental theorem of calculus, (1), 5.4(129), 5.5(132) general binomial theorem, 5.5(132) geometric progression, 1.7(19) geometric set, 7.2(186) greatest lower bound, 1.5(13) Greeks, 2.1(29) Green's theorem, 6.7(174) Heine-Borel Theorem, 2.7(44) hyperbolic functions, 3.8(72) identity theorem, 3.7(67), 3.9(75), 4.6(99), 7.3(192), 7.5(196) image, 3.2(53) improperly-integrable, 5.7(140) innite geometric series, 2.8(48), 3.9(75) innite series, 6.5(168) inection point, 4.9(106) integer, 1.2(5) integral, 5.4(129), 5.6(135), 5.8(144), 6.1(153), 6.4(164), 6.7(174) integral test, 5.7(140) integration, 5.1(117) integration by parts, 5.5(132) integration by substitution, 5.5(132) interior point, 2.7(44) Intermediate value theorem, 3.6(64) intervals, 1.6(16) introduction, 1.1(5), 2.1(29), 3.1(53), 4.1(81), 5.1(117), 6.1(153) inverse function theorem, 4.4(92), 7.6(198) inverse image, 3.2(53) invertible, 3.3(57) irrational number, 1.5(13) isolated singularities, 7.8(201) isomorphism, 8(211) jordan curve theorem, 6.7(174) l'hopital's rule, 4.7(101) laurent expansion, 7.8(201) law of exponents, 4.5(96) leibniz, 5.4(129) less than or equal, 8(211) limit, 3.7(67) limit theorems, 2.5(36) limits, 2.2(29), 4.2(81) line integrals, 6.6(170) linearity, 5.7(140) linearly dependent, 7.6(198) local maximum, 4.4(92), 4.11(111), 7.5(196) local minimum, 4.4(92), 7.5(196) logarithms, 4.5(96) lower bound, 1.4(9) mathematical induction, 1.2(5) mathematical truth, (1) mean value theorem, 4.1(81), 4.4(92), 4.5(96), 4.9(106), 4.11(111), 5.5(132), 6.2(154) Minimum modulus principle, 7.5(196) mixed partials, 4.11(111) multiplicative inverse, 1.3(7) negative numbers, 1.2(5) nonconstant dierentiable function, 7.3(192) nondecreasing, 2.3(32) nonincreasing, 2.3(32) nonoverlapping, 5.8(144) nonremovable singularities, 7.8(201) numerator, 1.3(7) open cover, 2.7(44) open geometric set, 5.6(135) open mapping theorem, 7.6(198) open neighborhood, 1.8(23) open sets, 3.5(63) open subintervals, 5.2(117) order-preserving, 5.7(140) ordered eld, 1.4(9), 1.8(23) parametrization, 6.3(159), 6.5(168), 6.6(170) partial derivatives, 4.11(111) partition, 5.2(117), 5.8(144) peano curve, 6.2(154) pi, 3.8(72) piecewise linear function, 3.2(53) piecewise smooth curve, 6.2(154), 6.3(159), 6.4(164) piecewise smooth function, 6.6(170), 6.7(174) piecewise smooth geometric set, 5.6(135), 7.6(198), 7.8(201) plane, 5.8(144) polar coordinates, 4.6(99) polygonal function, 3.2(53) polynomial function, 3.2(53), 3.7(67) positivity, 5.7(140)

H I

J L

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

INDEX
positivity properties, 1.4(9) power series, 3.7(67) proof, 1.5(13) punctured disks, 1.8(23) Quotient formula, 4.5(96) radius, 5.1(117) range, 3.2(53) ratio test, 2.8(48) rational function, 3.2(53) rational number, 1.6(16), 5.3(123) real number, 1.5(13) real numbers, 1.1(5) real-analytic, 4.8(104) real-valued function, 3.2(53) remainder theorem of taylor, 5.5(132) removable singularity, 7.8(201) residue theorem, 7.8(201) Riemann sums, 5.2(117) Riemann-integrable, 5.3(123) sequence, 2.4(34), 2.7(44) sequence of partial sums, 2.8(48) sequences, 2.2(29) smooth approximation, 4.8(104) smooth curves, 6.1(153) smooth geometric set, 5.6(135) squeeze theorem, 2.5(36) step function, 3.2(53), 5.8(144) strictly decreasing, 2.3(32) strictly increasing, 2.3(32) summable, 2.8(48) symmetric image, 5.8(144) taylor polynomial, 4.9(106), 4.10(108)

239

Q R

Taylor series, 3.9(75), 4.8(104), 7.3(192) taylor's remainder theorem, 4.9(106), 4.10(108) techniques of integration, 5.5(132) test for irrationality, 2.8(48) theorem, 1.5(13) theory of complex variables, 4.8(104) topology, 3.5(63) trigonometric functions, 3.8(72), 4.6(99) uniform convergence, 7.7(201) uniform limit, 3.10(76) uniform limits, 5.3(123) uniformly continuous, 6.6(170) uniqueness of coecients, 3.3(57) uniqueness of limits, 2.6(39) unit tangent, 6.2(154) weierstrass m-test, 3.10(76), 7.3(192) weighted average, 5.2(117) zero, 1.2(5)

W Z

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

240

ATTRIBUTIONS

Attributions
Collection:

Analysis of Functions of a Single Variable

Edited by: Lawrence Baggett URL: http://cnx.org/content/col11249/1.1/ License: http://creativecommons.org/licenses/by/3.0/ Module: "Preface to Analysis of Functions of a Single Variable: A Detailed Development" By: Lawrence Baggett URL: http://cnx.org/content/m36084/1.3/ Pages: 1-4 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Real and Complex Numbers: Denition of the Numbers 1, i, and the square root of 2" Used here as: "Denition of the Numbers 1, i, and the square root of 2" By: Lawrence Baggett URL: http://cnx.org/content/m36082/1.3/ Page: 5 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Real and Complex Numbers: The Natural Numbers and the Integers" Used here as: "The Natural Numbers and the Integers" By: Lawrence Baggett URL: http://cnx.org/content/m36075/1.2/ Pages: 5-7 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Real and Complex Numbers: The Rational Numbers" Used here as: "The Rational Numbers" By: Lawrence Baggett URL: http://cnx.org/content/m36061/1.2/ Pages: 7-9 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Real and Complex Numbers: The Real Numbers" Used here as: "The Real Numbers" By: Lawrence Baggett URL: http://cnx.org/content/m36069/1.2/ Pages: 9-13 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Real and Complex Numbers: Properties of the Real Numbers" Used here as: "Properties of the Real Numbers" By: Lawrence Baggett URL: http://cnx.org/content/m36085/1.2/ Pages: 13-16 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

ATTRIBUTIONS
Module: "The Real and Complex Numbers: Intervals and Approximation" Used here as: "Intervals and Approximation" By: Lawrence Baggett URL: http://cnx.org/content/m36094/1.2/ Pages: 16-18 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Real and Complex Numbers: The Geometric Progression and the Binomial Theorem" Used here as: "The Geometric Progression and the Binomial Theorem" By: Lawrence Baggett URL: http://cnx.org/content/m36104/1.2/ Pages: 19-23 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Real and Complex Numbers: The Complex Numbers" Used here as: "The Complex Numbers" By: Lawrence Baggett URL: http://cnx.org/content/m36113/1.2/ Pages: 23-28 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Introduction to the Limit of a Sequence of Numbers: Denition of the Number e:" Used here as: "Denition of the Number e" By: Lawrence Baggett URL: http://cnx.org/content/m36117/1.2/ Page: 29 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Limit of a Sequence of Numbers: Sequences and Limits" Used here as: "Sequences and Limits" By: Lawrence Baggett URL: http://cnx.org/content/m36118/1.2/ Pages: 29-32 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Limit of a Sequence of Numbers: Existence of Certain Fundamental Limits" Used here as: "Existence of Certain Fundamental Limits" By: Lawrence Baggett URL: http://cnx.org/content/m36120/1.2/ Pages: 32-34 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

241

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

242 Module: "The Limit of a Sequence of Numbers: Denition of e" Used here as: "Denition of e" By: Lawrence Baggett URL: http://cnx.org/content/m36124/1.2/ Pages: 34-36 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Limit of a Sequence of Numbers: Properties of Convergent Sequences" Used here as: "Properties of Convergent Sequences" By: Lawrence Baggett URL: http://cnx.org/content/m36126/1.2/ Pages: 36-39 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Limit of a Sequence of Numbers: Subsequences and Cluster Points" Used here as: "Subsequences and Cluster Points" By: Lawrence Baggett URL: http://cnx.org/content/m36129/1.2/ Pages: 39-44 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Limit of a Sequence of Numbers: A Little Topology" Used here as: "A Little Topology" By: Lawrence Baggett URL: http://cnx.org/content/m36157/1.2/ Pages: 44-47 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Limit of a Sequence of Numbers: Innite Series" Used here as: "Innite Series" By: Lawrence Baggett URL: http://cnx.org/content/m36135/1.2/ Pages: 48-52 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Functions and Continuity Denition of the Number By: Lawrence Baggett URL: http://cnx.org/content/m36131/1.2/ Page: 53 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Functions and Continuity: Functions" Used here as: "Functions" By: Lawrence Baggett URL: http://cnx.org/content/m36141/1.2/ Pages: 53-57 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

ATTRIBUTIONS

"

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

ATTRIBUTIONS
Module: "Functions and Continuity: Polynomial Functions" Used here as: "Polynomial Functions" By: Lawrence Baggett URL: http://cnx.org/content/m36147/1.2/ Pages: 57-60 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Functions and Continuity: Continuity" Used here as: "Continuity" By: Lawrence Baggett URL: http://cnx.org/content/m36150/1.2/ Pages: 60-63 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Functions and Continuity: Continuity and Topology" Used here as: "Continuity and Topology" By: Lawrence Baggett URL: http://cnx.org/content/m36152/1.2/ Pages: 63-64 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Functions and Continuity: Deeper Analytic Properties of Continuous Functions" Used here as: "Deeper Analytic Properties of Continuous Functions" By: Lawrence Baggett URL: http://cnx.org/content/m36167/1.2/ Pages: 64-67 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Functions and Continuity: Power Series Functions" Used here as: "Power Series Functions" By: Lawrence Baggett URL: http://cnx.org/content/m36165/1.2/ Pages: 67-71 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Functions and Continuity: The Elementary Transcendental Functions" Used here as: "The Elementary Transcendental Functions" By: Lawrence Baggett URL: http://cnx.org/content/m36160/1.2/ Pages: 72-74 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

243

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

244 Module: "Functions and Continuity: Analytic Functions and Taylor Series" Used here as: "Analytic Functions and Taylor Series" By: Lawrence Baggett URL: http://cnx.org/content/m36168/1.2/ Page: 75 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Functions and Continuity: Uniform Convergence" Used here as: "Uniform Convergence" By: Lawrence Baggett URL: http://cnx.org/content/m36178/1.2/ Pages: 76-80 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Dierentiation, Local Behavior E^i = -1." By: Lawrence Baggett URL: http://cnx.org/content/m36173/1.2/ Page: 81 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Dierentiation, Local Behavior: The Limit of a Function" Used here as: "The Limit of a Function" By: Lawrence Baggett URL: http://cnx.org/content/m36185/1.2/ Pages: 81-83 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Dierentiation, Local Behavior: The Derivative of a Function" Used here as: "The Derivative of a Function" By: Lawrence Baggett URL: http://cnx.org/content/m36186/1.2/ Pages: 83-91 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

ATTRIBUTIONS

Module: "Dierentiation, Local Behavior: Consequences of Dierentiability, the Mean Value Theorem" Used here as: "Consequences of Dierentiability, the Mean Value Theorem" By: Lawrence Baggett URL: http://cnx.org/content/m36203/1.2/ Pages: 92-96 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Dierentiation, Local Behavior: The Exponential and Logarithm Functions" Used here as: "The Exponential and Logarithm Functions" By: Lawrence Baggett URL: http://cnx.org/content/m36199/1.2/ Pages: 96-99 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

ATTRIBUTIONS
Module: "Dierentiation, Local Behavior: The Trigonometric and Hyperbolic Functions" Used here as: "The Trigonometric and Hyperbolic Functions" By: Lawrence Baggett URL: http://cnx.org/content/m36196/1.2/ Pages: 99-101 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Dierentiation, Local Behavior: L'Hopital's Rule" Used here as: "L'Hopital's Rule" By: Lawrence Baggett URL: http://cnx.org/content/m36201/1.2/ Pages: 101-104 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Dierentiation, Local Behavior: Higher Order Derivatives" Used here as: "Higher Order Derivatives" By: Lawrence Baggett URL: http://cnx.org/content/m36192/1.2/ Pages: 104-106 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Dierentiation, Local Behavior: Taylor Polynomials and Taylor's Remainder Theorem" Used here as: "Taylor Polynomials and Taylor's Remainder Theorem" By: Lawrence Baggett URL: http://cnx.org/content/m36204/1.2/ Pages: 106-108 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Dierentiation, Local Behavior: The General Binomial Theorem" Used here as: "The General Binomial Theorem" By: Lawrence Baggett URL: http://cnx.org/content/m36205/1.2/ Pages: 108-111 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Dierentiation, Local Behavior: More on Partial Derivatives" Used here as: "More on Partial Derivatives" By: Lawrence Baggett URL: http://cnx.org/content/m36206/1.2/ Pages: 111-115 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration, Average Behavior A= r^2" By: Lawrence Baggett URL: http://cnx.org/content/m36207/1.2/ Page: 117 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

245

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

246 Module: "Integration, Average Behavior: Integrals of Step Functions" Used here as: "Integrals of Step Functions" By: Lawrence Baggett URL: http://cnx.org/content/m36208/1.2/ Pages: 117-123 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration, Average Behavior: Integrable Functions" Used here as: "Integrable Functions" By: Lawrence Baggett URL: http://cnx.org/content/m36209/1.2/ Pages: 123-129 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration, Average Behavior: The Fundamental Theorem of Calculus" Used here as: "The Fundamental Theorem of Calculus" By: Lawrence Baggett URL: http://cnx.org/content/m36210/1.2/ Pages: 129-132 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration, Average Behavior: Consequences of the Fundamental Theorem" Used here as: "Consequences of the Fundamental Theorem" By: Lawrence Baggett URL: http://cnx.org/content/m36211/1.2/ Pages: 132-135 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration, Average Behavior: Area of Regions in the Plane" Used here as: "Area of Regions in the Plane" By: Lawrence Baggett URL: http://cnx.org/content/m36212/1.2/ Pages: 135-140 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration, Average Behavior: Extending the Denition of Integrability" Used here as: "Extending the Denition of Integrability" By: Lawrence Baggett URL: http://cnx.org/content/m36222/1.2/ Pages: 140-144 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

ATTRIBUTIONS

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

ATTRIBUTIONS
Module: "Integration, Average Behavior: Integration in the Plane" Used here as: "Integration in the Plane" By: Lawrence Baggett URL: http://cnx.org/content/m36223/1.2/ Pages: 144-151 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration Over Smooth Curves in the Plane C=2 r" By: Lawrence Baggett URL: http://cnx.org/content/m36224/1.2/ Pages: 153-154 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration Over Smooth Curves in the Plane: Smooth Curves in the Plane" Used here as: "Smooth Curves in the Plane" By: Lawrence Baggett URL: http://cnx.org/content/m36225/1.2/ Pages: 154-159 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration over Smooth Curves in the Plane: Arc Length" Used here as: "Arc Length" By: Lawrence Baggett URL: http://cnx.org/content/m36226/1.2/ Pages: 159-164 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration over Smooth Curves in the Plane: Integration with Respect to Arc Length" Used here as: "Integration with Respect to Arc Length" By: Lawrence Baggett URL: http://cnx.org/content/m36228/1.2/ Pages: 164-168 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration over Smooth Curves in the Plane: Contour Integrals" Used here as: "Contour Integrals" By: Lawrence Baggett URL: http://cnx.org/content/m36230/1.2/ Pages: 168-170 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

247

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

248

ATTRIBUTIONS

Module: "Integration Over Smooth Curves in the Plane: Vector Fields, Dierential Forms, and Line Integrals" Used here as: "Vector Fields, Dierential Forms, and Line Integrals" By: Lawrence Baggett URL: http://cnx.org/content/m36232/1.2/ Pages: 170-174 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Integration Over Smooth Curves in the Plane: Integration Around Closed Curves, and Green's Theorem" Used here as: "Integration Around Closed Curves, and Green's Theorem" By: Lawrence Baggett URL: http://cnx.org/content/m36233/1.2/ Pages: 174-183 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Fundamental Theorem of Algebra, and the Fundamental Theorem of Analysis" By: Lawrence Baggett URL: http://cnx.org/content/m36234/1.2/ Page: 185 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Fundamental Theorem of Algebra, Analysis: Cauchy's Theorem" Used here as: "Cauchy's Theorem" By: Lawrence Baggett URL: http://cnx.org/content/m36235/1.2/ Pages: 186-192 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Fundamental Theorem of Algebra, Analysis: Basic Applications of the Cauchy Integral Formula" Used here as: "Basic Applications of the Cauchy Integral Formula" By: Lawrence Baggett URL: http://cnx.org/content/m36237/1.2/ Pages: 192-195 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Fundamental Theorem of Algebra, Analysis: The Fundamental Theorem of Algebra" Used here as: "The Fundamental Theorem of Algebra" By: Lawrence Baggett URL: http://cnx.org/content/m36238/1.2/ Pages: 195-196 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

ATTRIBUTIONS
Module: "Fundamental Theorem of Algebra, Analysis: The Maximum Modulus Principle" Used here as: "The Maximum Modulus Principle" By: Lawrence Baggett URL: http://cnx.org/content/m36239/1.2/ Pages: 196-198 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

249

Module: "Fundamental Theorem of Algebra, Analysis: The Open Mapping Theorem and the Inverse Function Theorem" Used here as: "The Open Mapping Theorem and the Inverse Function Theorem" By: Lawrence Baggett URL: http://cnx.org/content/m36240/1.2/ Pages: 198-201 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "The Fundamental Theorem of Algebra, Analysis: Uniform Convergence of Analytic Functions" Used here as: "Uniform Convergence of Analytic Functions" By: Lawrence Baggett URL: http://cnx.org/content/m36241/1.2/ Page: 201 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Fundamental Theorem of Algebra, Analysis: Isolated Singularities, and the Residue Theorem" Used here as: "Isolated Singularities, and the Residue Theorem" By: Lawrence Baggett URL: http://cnx.org/content/m36242/1.2/ Pages: 201-209 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/ Module: "Appendix: Existence and Uniqueness of a Complete Ordered Field" By: Lawrence Baggett URL: http://cnx.org/content/m36243/1.2/ Pages: 211-217 Copyright: Lawrence Baggett License: http://creativecommons.org/licenses/by/3.0/

Available for free at Connexions <http://cnx.org/content/col11249/1.1>

Analysis of Functions of a Single Variable

I have written this book primarily for serious and talented mathematics scholars, seniors or rst-year graduate students, who by the time they nish their schooling should have had the opportunity to study in some detail the great discoveries of our subject. What did we know and how and when did we know it? I hope this book is useful toward that goal, especially when it comes to the great achievements of that part of mathematics known as analysis. I have tried to write a complete and thorough account of the elementary theories of functions of a single real variable and functions of a single complex variable. Separating these two subjects does not at all jive with their development historically, and to me it seems unnecessary and potentially confusing to do so. On the other hand, functions of several variables seems to me to be a very dierent kettle of sh, so I have decided to limit this book by concentrating on one variable at a time.

About Connexions
Since 1999, Connexions has been pioneering a global system where anyone can create course materials and make them fully accessible and easily reusable free of charge. We are a Web-based authoring, teaching and learning environment open to anyone interested in education, including students, teachers, professors and lifelong learners. We connect ideas and facilitate educational communities. Connexions's modular, interactive courses are in use worldwide by universities, community colleges, K-12 schools, distance learners, and lifelong learners. Connexions materials are in many languages, including English, Spanish, Chinese, Japanese, Italian, Vietnamese, French, Portuguese, and Thai. Connexions is part of an exciting new information distribution system that allows for

Print on Demand Books.

Connexions

has partnered with innovative on-demand publisher QOOP to accelerate the delivery of printed course materials and textbooks into classrooms worldwide at lower prices than traditional academic publishers.

This book was distributed courtesy of:

For your own Unlimited Reading and FREE eBooks today, visit: http://www.Free-eBooks.net

Share this eBook with anyone and everyone automatically by selecting any of the options below:

To show your appreciation to the author and help others have wonderful reading experiences and find helpful information too, we'd be very grateful if you'd kindly post your comments for this book here.

COPYRIGHT INFORMATION
Free-eBooks.net respects the intellectual property of others. When a book's copyright owner submits their work to Free-eBooks.net, they are granting us permission to distribute such material. Unless otherwise stated in this book, this permission is not passed onto others. As such, redistributing this book without the copyright owner's permission can constitute copyright infringement. If you believe that your work has been used in a manner that constitutes copyright infringement, please follow our Notice and Procedure for Making Claims of Copyright Infringement as seen in our Terms of Service here:

http://www.free-ebooks.net/tos.html