Intro To Analysis

Irena Swanson Reed College Spring 2013
Table of contents
Preface The briefest overview Chapter 1: How we will do mathematics Section 1.1: Statements and proof methods Section 1.2: Statements with quantiers Section 1.3: More proof methods, and negation Section 1.4: Summation Section 1.5: Proofs by (mathematical) induction Section 1.6: Pascals triangle Chapter 2: Concepts with which we will do mathematics Section 2.1: Sets Section 2.2: Cartesian products Section 2.3: Relations, equivalence relations Section 2.4: Functions Section 2.5: Binary operations on sets Section 2.6: Fields Section 2.7: Order on sets, ordered elds, absolute values Section 2.8: Increasing and decreasing functions Chapter 3: Construction of the basic number systems Section 3.1: Inductive sets and a construction of natural numbers Section 3.2: Arithmetic on N0 Section 3.3: Cancellations in N0 Section 3.4: Order on N0 Section 3.5: Construction of Z, arithmetic, and order on Z Section 3.6: Construction of the eld Q of rational numbers Section 3.7: Construction of the eld R of real numbers Section 3.8: Order on R, the least upper/greatest lower bound theorem Section 3.9: Complex numbers
7 8 9 9 18 22 27 29 33 35 35 42 44 48 55 60 62 68 70 70 74 77 78 81 86 91 97 99
4 Section 3.10: Absolute value in C Section 3.11: Topology of the constructed elds Section 3.12: Closed and bounded sets and open balls Chapter 4: Limits of functions Section 4.1: Limit of a function Section 4.2: When something is not a limit Section 4.3: Limit theorems Section 4.4: Innite limits (for real-valued functions) Section 4.5: Limits at innity Chapter 5: Continuity Section 5.1: Continuous functions Section 5.2: Intermediate value theorem, Extreme value theorem Section 5.3: Radical functions Section 5.4: Uniform continuity Chapter 6: Dierentiation Section 6.1: Denition of derivatives Section 6.2: Basic properties of derivatives Section 6.3: The Mean value theorem Section 6.4: Higher-order derivatives, Taylor polynomials Chapter 7: Integration Section 7.1: Approximating areas Section 7.2: What functions are integrable? Section 7.3: The Fundamental theorem of calculus Section 7.4: Natural logarithm and the exponential functions Section 7.5: Applications of integration Chapter 8: Sequences Section 8.1: Introduction to sequences Section 8.2: Convergence of innite sequences Section 8.3: Divergence of innite sequences and innite limits Section 8.4: Convergence theorems via functions Section 8.5: Convergence theorems via epsilon-N proofs Section 8.6: Cauchy sequences, completeness of R, C Section 8.7: Ratio test for convergence of sequences Section 8.8: Subsequences Section 8.9: Liminf, limsup for real-valued sequences 101 107 113 116 116 125 128 134 138 140 140 143 147 151 154 154 158 165 168 171 171 176 181 187 191 197 197 201 206 210 214 219 222 223 226
5 Chapter 9: Innite series and power series Section 9.1: Innite series Section 9.2: Convergence and divergence theorems for series Section 9.3: Power series Section 9.4: Dierentiation of power series Section 9.5: Numerical evaluations of some series Section 9.6: A special function Section 9.7: A special function, continued Appendix A: Advice on writing mathematics Appendix B: What you should never forget Appendix C: Solutions to selected exercises Index 229 229 232 237 240 243 244 247 251 255 258 261
Preface
These notes were written expressly for Mathematics 112 at Reed College. The title of the course is Introduction to Analysis, and its prerequisite is calculus. We have used several textbooks in the past, most recently Ray Mayers in-house notes Introduction to Analysis (2006, available at http://www.reed.edu/~mayer/math112.html/index.html), and Steven R. Lays Analysis, With an Introduction to Proof (Prentice Hall, Inc., Englewood Clis, NJ, 1986, 4th edition). In Math 112, students learn to write proofs while at the same time learning about binary operations, orders, elds, ordered elds, complete elds, complex numbers, limits, continuity, dierentiation, integration, sequences, and series. These notes reect how I have been teaching the material for several years. My aim for these notes is to constitute a self-contained book that covers the standard topics of a course in introductory analysis, that handles complex-valued functions, sequences, and series, that has enough examples and exercises, that is rigorous, and is accessible to Reed College undergraduates. Chapter 1 is about how we do mathematics: basic logic, proof methods, and Pascals triangle for practicing proofs. Chapter 2 introduces foundational concepts: sets, Cartesian products, relations, functions, binary operations, elds, ordered elds. In Chapters 1 and 2 we assume knowledge of high school mathematics, so that we do not practice abstract concepts and methods in a vacuum. Chapter 3 through Section 3.8 takes a step back: we forget most previously learned mathematics, and we use the newly learned abstract tools to construct natural numbers, integers, rational numbers, real numbers, with all the arithmetic and order. I do not teach these sections in great detail; my aim is to give a sense of the constructions and to practice abstract logical thinking. I spend about an hour and a half sketching the construction of real numbers via Dedekind cuts: I ask the students to not take notes but instead to nod vigorously in agreement. The remaining sections in Chapter 3 are new material for most students: complex numbers, the eld of complex numbers, and some topology. Subsequent chapters cover standard material for introduction to analysis: limits, continuity, dierentiation, integration, sequences, series, k ending with the development of the power series k=0 x k! , the (real) exponential function, and the trigonometric functions. An eort is made throughout to use only what had been proved. In particular, trigonometric functions appear properly only in the last chapter. For this reason, the chapters on dierentiation and integration do not have the usual palette of examples of other books where dierentiability and derivatives of trigonometric functions are assumed. I thank Maddie Brandt, Andrew Erlanger, Munyo Frey-Edwards, Kelsey HoustonEdwards, Molly Maguire, and Simon Swanson for valuable feedback and help in writing these notes. Further comments or corrections should be sent to iswanson@reed.edu.
The briefest overview

What are the meanings of the following: 5+6 79 4/5 2 84
48
1/3 = 0.333 . . .
(a b)2 = a2 b2
1 = 3 1/3 = 0.999 . . .
(a + b) (c + d) = ac + ad + bc + bd (a + b) (a b) = a2 b2 (a + b)2 = a2 + 2ab + b2 a b = ab (for which a, b?) Reconcile: 4 9 = (4)(9) = 36 = 6, 4 9 = 2i 3i = 6
You know all of the above except possibly the complex numbers in the last row, where obviously something went wrong. We will not resolve this last issue until later in the semester, but the point for now is that we do need to reason carefully. The main goal of this class is to learn to reason carefully, rigorously. Since one cannot reason in a vacuum, we will (but of course) be learning a lot of mathematics as well: sets, logic, various number systems, elds, the eld of real numbers, the eld of complex numbers, sequences, series, some calculus, and that eix = cos x + i sin x. We will make it all rigorous, i.e., we will be doing proofs. A proof is a sequence of steps that logically follow from previously accepted knowledge. But no matter what you do, never divide by 0. For more wise advice, turn to Appendix A. [Notational convention: Text between square brackets in this font and in red color should be read as a possible reasoning going on in the background in your head, and not as part of formal writing.] Exercise 1 Exercises with a dagger are invoked later in the text. *Exercise 2 Exercises with a star are more dicult.
Chapter 1: How we will do mathematics
1.1
Statements and proof methods
Denition 1.1.1 A statement is a reasonably grammatical and unambiguous sentence that can be declared either true or false. Why do we specify reasonably grammatical? We do not disqualify a statement just because it is not written in awless grammar, nevertheless, we strive to use correct grammar and to express the meaning clearly. Examples and non-examples 1.1.2 (i) The sum of 1 and 2 equals 3. (This is a statement; it is always true.) (ii) I am ve years old. (This is a statement, even if it may be sometimes true and sometimes false.) (iii) Hello. (This is not a statement.) (iv) It is raining now. (This is a statement.) (v) How did you do that? (This is not a statement.) (vi) 1 = 2. (This is a statement, but not written in plain English. Here, = is a verb.) (vii) For every real number > 0 there exists a real number > 0 such that for all x, if 0 < |x a| < then |f (x) L| < . (This is a statement.) (viii) Every even number greater than 4 can be written as a sum of two odd primes. (This is a statement for which it is currently not known if it is true. This particular statement is known as the Goldbachs conjecture.) I want to emphasize that sentences such as I am good, I am the dictator of the world, All functions are expressible as polynomials, The Riemann hypothesis is true, P = NP are statements. We may not agree on or not know their veracity, but we can start with assuming that one is true or false, and then proceed from there. For example, if I am good, then I get a cookie, but if I am not good, then you get the cookie. If I were the dictator of the world, everybody would be good at logic and common sense. If all functions were expressible as polynomials, then we would not have to deal with trigonometric, exponential and logarithmic functions. If the Riemann hypothesis is true, then prime numbers are distributed fairly regularly among all integers. If P = NP, then many computational problems would become tractable, spurring advances in logistics, biology, etc. If P = NP,
10
Section 1.1: Statements and proof methods
then many common problems would not be solvable eciently, and eorts should be made instead in partial and heuristic solutions. On the other hand, if Hello were to be true or false, I would not be able to make any further deductions about the world or my next action. Denition 1.1.3 A tautology is a statement that is always true. For example 1 = 1 or x = x are tautologies. A useful tool for manipulating statements is a truth table: it is a table in which the rst few columns may set up a situation, and the subsequent columns record truth values of statements applying in those particular situations. Here are two examples of truth tables, where T of course stands for true and F for false: f ( x) x | x| 7 x x>0 x>0 x<0 x<0
2
f is a constant function. F F T y >0 0 >0 0
f is continuous. T T T
f is dierentiable. T F T
y y y y
xy > 0 xy 0 xy < 0 T F F F T F F T T F F F
Note that in the second row of the last table, in the exceptional case y = 0, the statement xy < 0 is false, but in the majority of the cases in that row xy < 0 is true. The one counterexample is enough to declare xy < 0 not true, i.e., false. Statements can be manipulated just like numbers and variables can be manipulated, and rather than adding or multiplying statements, we connect them (by compounding the sentences in grammatical ways) with connectors such as not, and, or, and so on. Statement connecting: (1) Negation of a statement P is a statement whose truth values are exactly opposite from the truth values of P (under any circumstance). The negation of P is denoted P (or sometimes ~P or not P ). Some simple examples: the negation of A = B is A = B ; the negation of A B is A > B ; the negation of I am here is I am not here or It is not the case that I am here.
11
(2) Conjunction of statements P and Q is a statement that is true precisely when both P and Q are true, and it is false otherwise. It is denoted P and Q or P Q. We can record this in a truth table as follows: P T T F F Q T F T F P Q T F F F
(3) Disjunction of statements P and Q is a statement that is false precisely when both P and Q are false, and it is true otherwise. We denote it as P or Q or as P Q. In other words, as long as either P or Q is true, than P Q is true. In plain language, unfortunately, we use or in two dierent ways: You may take cream or sugar says you may take cream or sugar or both, just like in the proper logical way, but Tonight we will go to the movies or to the baseball game implies that we will either go to the movies or to the baseball game but we will not do both. The latter connection of two sentences is in logic called exclusive or, often denoted xor. Even either-or does not disambiguate between or and xor, but either-or else is the same as xor. The truth table for the two disjunctions is: P T T F F Q T F T F P Q T T T F P xor Q F T T F
(4) Implication or a conditional statement is a statement of the form P implies Q, or variants thereof, such as If P then Q, Given P , Q follows, Q whenever P . A symbolic abbreviation of implication is P Q. An implication is true when a true conclusion follows a true assumption, or whenever the assumption is false. In other words, P Q is false exactly when P is true and Q is false. In an implication P Q, P is called the antecedent and Q the consequent. The truth table for implication is as follows: P T T F F Q T F T F P Q T F T T
12
Section 1.1: Statements and proof methods It may be counterintuitive that a false antecedent always makes the implication true. Bertrand Russell once lectured on this and claimed that if 1 = 2 then he (Bertrand Russell) was the pope. An audience member challenged him. So Russell reasoned somewhat like this: If I am the pope, then the consequent is true. If the consequent is false, then I am not the pope. But if I am not the pope, then the pope and I are two dierent people. By assumption 1 = 2, so we two people are one, so I am the pope. Thus no matter what, I am the pope. Furthermore, if 1 = 2, then Bertrand Russell is similarly also not the pope. Namely, if he is not the pope, the consequent is true, but if he is the pope, then the pope and he are one, and since one equals two, then the pope and he are two people, so Russell cannot be the pope. A further discussion about why false antecedent makes the implication true is in the next discussion (5). We list here most possible rephrasings of P implies Q: P implies Q. If P then Q. P is a sucient condition for Q. P only if Q. Q if P . Q provided P . Q given P . Q whenever P . Q is a necessary condition for P . Unfortunately, the implication statement is not used consistently in informal spoken language. For example, your grandmother may say: You may have ice cream if you eat your broccoli when she means You may have ice cream only if you eat your broccoli, i.e., Your eating ice cream can only happen if you rst eat your broccoli. Be nice to your grandmother and eat that broccoli even if she does not express herself precisely because you know precisely what she means. But in mathematics classes you do have to express yourself precisely! (Well, read the next paragraph.) Even in mathematics some shortcuts in precise expressions are acceptable. Here is an example. The statements An object x has P if somethingorother holds and An object x has P if and only if somethingorother holds (see (5) below for if and only if) in general have dierent truth values and the proof of the second is longer. However, the denition of what it means for an object to have property P in terms
13
of somethingorother is usually phrased as An object x has P if somethingorother holds, and it is implicit that we will use the adjective P exactly in cases when somethingorother holds. It would be more precise to write An object x has P if and only if somethingorother holds, or For an object x to have P it means that somethingorother holds, but this is not always done. An example of this shortcut is Denition 2.4.9. (5) Equivalence or the logical biconditional of P and Q stands for the compound statement (P Q) (Q P ). It is abbreviated P Q or P i Q, and is true precisely when P and Q have the same truth values. For example, x y + 1 is equivalent to x 1 y . Another example: 2x = 4x2 is equivalent to x = 2x2 , but it is not equivalent to 1 = 2x. (Say why!) We now backtrack on the truth values of P Q. We can certainly ll in this part without qualms, leaving some unknown truth values x and y : P T T F F Q T F T F P Q T F x y QP y x F y P Q T F F T
Since the last column above is the conjunction of the previous two, the last line forces the value of y to be T . If x equals F , then the truth values of P Q are the same as the truth values of P Q, which would say that the statements P Q and P Q are logically the same. But this cannot be: If r > 0 then r 0 is true whereas If r 0 then r > 0 is false. So the truth values for the third and the fth column have to be distinct, and this is only possible if x is T . Here is the truth table for all the connectives so far: P T T F F Q T F T F P F F T T P Q T F F F P Q T T T F P xor Q F T T F P Q T F T T P Q T F F T
One can form more elaborate truth tables if we start not with two statements P and Q but with three or more. Examples of logically compounding P, Q, and R are: P Q R, (P Q) Q, et cetera. For manipulating three statements, we would ll a total of 8 rows of truth values, for four statements there would be 16 rows, and so on.
14
Section 1.1: Statements and proof methods
(6) Proof of P is a series of steps (in statement form) that establish beyond doubt that P is true under all circumstances, weather conditions, political regimes, time of day... Ends of proofs are usually marked by , , //, or QED (Quod erat demonstrandum, which is Latin for that which was to be proved). The most trivial proofs simply invoke a denition or axiom, such as An even integer is of the form 2 times an integer, or, A positive integer is prime if whenever it can be written as a product of two positive integers, one of the two factors is 1. Another type of proof consists of lling in a truth table. For example, P (P ) is always true, no matter what the truth value of P is, and this can be easily veried with a truth table. Yet another type of proof is based on deductive reasoning. The most common form of deductive reasoning is modus ponens. It says that ((P Q) P ) Q is a tautology. One can verify this tautology with truth tables. The most famous example of modus ponens is the following: Every man is mortal. (If X is a man/human, then X is mortal.) Socrates is human. Therefore, Socrates is mortal. Here is a more mathematical example: Every dierentiable function is continuous. f is dierentiable. Therefore, f is continuous. Another form of deductive reasoning is modus tollens. It says that ((P Q) (Q)) (P ) is a tautology. One can verify also this tautology with truth tables. An example of modus tollens is: If x > 5 then x is positive. x is not positive. Therefore, x 5. Most of the time, naturally, a method of proof by syllogism has more ourishes, in the sense that there has to be more explanation, more intermediate steps, more algebra, et cetera. If n is an even whole number, then n = 2m for some whole number m. 9 can be written as 2 4.5, and 4.5 is not a whole number.
Chapter 1: How we will do mathematics If 9 = 2 m, then 2m = 2 4.5, so m = 4.5. 9 cannot be written as two times a whole number. Therefore, 9 is not an even number.
15
There are further forms on deductive reasoning; one simply has to nd tautologies, say possibly with the help of truth tables. Here are two more examples: ((P Q) (Q R)) (P R),
((P Q) (Q R) P ) R.
If n is an even whole number, then n = 2m for some whole number m. P = n is an even whole number. Q = n = 2m for some whole number m. The square of 2m is 4m2 , which is an even number. (If Q then R.) R = The square is an even whole number. Therefore, the square of an even number is even. (If P then R.) The second example similarly analyzes odd numbers. If n is a whole number that is not even, then n = 2m + 1 for some whole number m, and n is called odd. The square of 2m + 1 is, by a little algebra, equal to 2(2m2 + m) + 1, which is an odd number. Therefore, the square of an odd number is odd. A further example of deductive reasoning is combinining previously established results, such as the last two results: A whole number is even if and only if its square is even. Often we prove a result by invoking a previously known result. This is simply another example of deductive reasoning. For example, the Intermediate value theorem (Theorem 5.2.1) says that if a real-valued function is continuous on [a, b] and k is a number strictly between f (a) and f (b), then there exists c strictly between a and b such that f (c) = 0. (We can think of this theorem as being in the form If P then Q.) We are asked to show that f (x) = x3 x 1 has a root. We somehow know that f is continuous (proved later in Theorem 5.1.3), and we notice that f (0) = 1 < 0 < 3 = f (2). Thus the hypotheses of the Intermediate value theorem are satised. (We established P .) Thus we may conclude that there exists c between 0 and 2 such that f (c) = 0. (So Q follows.) Another common method of proof of a statement P is proof by contradiction: you assume that P is true and by using only correct logical and mathematical
16
Section 1.1: Statements and proof methods steps you derive some nonsense (contradiction). This then says that P is false, so that (P ) = P must be true. Beware: proofs by contradiction are in general not considered elegant, nevertheless, they can be very powerful. 2 is not a rational number. Proof by contradiction: (Due to Pythagoras.) Suppose that 2 is rational. Write 2 = a b for some whole numbers a and b with b non-zero. Let d be the greatest common divisor of a and b. Write a0 a = a0 d and b = b0 d for some integers a0 , b0 = 0. Then 2 = a b = b0 , so 2 2 2 b0 2 = a0 , and so by squaring both sides we get that 2b2 0 = a0 . So a0 = 2b0 is an even number, which means (by a previously established fact) that a0 is 2 2 an even number. Write a0 = 2a1 for some integer a1 . Then 4a2 1 = a0 = 2b0 , 2 so that b2 0 = 2a1 is even, whence again b0 is even. But then 2 divides both a0 and b0 , so that 2d divides both a and b, which contradicts the assumption that d was the greatest common divisor of a and b. Thus it is not the case that 2 is rational, so it must be irrational.
Exercises for Section 1.1 1.1.1 Determine whether the following statements are true or false, and justify: i) 3 is odd or 5 is even. ii) If 3 is even, then 12 is prime. iii) If 6 is prime or 7 is odd, then 23 = 7 implies 7 is prime.
1.1.2 Sometimes statements are not written precisely enough. For example, It is not the case that 3 is prime and 5 is even may be saying (3 is prime and 5 is even), or it may be saying ((3 is prime)) and (5 is even). The rst of these options is true and the second is false. Similarly analyze all the possible interpretations of the following ambiguous sentences: i) If 6 is prime, then 7 is even or 5 is odd. ii) It is not the case that 3 is prime or if 6 is prime, then 7 is even or 5 is odd. General advice: Write precisely; aim to not be misunderstood. 1.1.4 Prove that P Q is equivalent to (Q) (P ). 1.1.3 Suppose that P Q is true and that Q is false. Prove that P is false.
1.1.5 Fill in the following extra columns to the truth table in (5): P Q, (P ) (Q), (P ) (Q), P Q, (P ) Q. Are any of the new columns negations of the columns in (5) or of each other? 1.1.6 Simplify the following statements: i) (P P ) P . ii) P P . iii) (P Q) (P Q).
Chapter 1: How we will do mathematics 1.1.7 Prove, using truth tables, that the following statements are tautologies: i) (P Q) [(P Q) (Q P )]. ii) (P Q) (Q P ). iii) (P Q) [P (P Q)]. iv) [P (Q R)] [(P Q) R]. 1.1.8 Determine the truth values of the following statements: i) If there are clouds above, then it is raining. ii) If if is raining, then there are clouds above. iii) If a function is continuous, then it is dierentiable. iv) If a function is dierentiable, then it is continuous. v) is a very large real number. vi) All prime numbers are even. vii) For all real numbers a, b, (a + b)2 = a2 + b2 . viii) For all real numbers a, b, (a + b)2 = a2 + 2ab + b2 . ix) P or not P . x) (P Q) P .
17
1.1.9 In each of the parts below, determine, with proof, the truth values of the subparts. Explain if there is not enough information to make a conclusion. i) Assume that P Q is true and that R Q is false. ii) Assume that (P Q) R is false. iii) It is true that my shirt is either brown or red, but it is not the case that it is either red or green. iv) If it rained yesterday, then today the grass will be wet and there will be worms on the sidewalk. If the grass was watered this morning, the grass will be wet. I go outside, and notice that there are worms on the sidewalk. 1.1.10 Prove that the sum of two odd integers is an even integer. 1.1.11 Prove that the product of two odd integers is an odd integer. 1.1.12 Suppose that the product of two integers is odd. Prove that the sum of those integers is even. 1.1.13 Suppose that the sum of the squares of two integers is odd. Prove that one of the two integers is even and the other is odd. 1.1.14 Is the following correct or false reasoning: If 1 = 2, then seven times each side gives equality 7 = 14. By subtracting 4 from both sides we get equality 3 = 10. Thus 1 = 2 implies 3 = 10. 1.1.15 Prove that 3 is not a rational number.
18
Section 1.2: Statements with quantiers
1.1.16 (Logic circuits) Logic circuits are simple circuits which take as inputs logical values of true and false (or 1 and 0) and give a single output. Logic circuits are composed of logic gates. Each logic gate stands for a logical connective you are familiar with it could be and, or, or not (more complex logic circuits incorporate more). The shapes for logical and, or, not are as follows:
Given inputs, each of these logic gates outputs values equal to the values in the associated truth table. For instance, an and gate only outputs on if both of the wires leading into it are on. From these three logic gates we can build many others. For example, the following circuit is equivalent to xor.
input output input
A circuit that computes xor. The output on the right is on when exactly one of the inputs on the left is on. Make logic circuits that complete the following tasks. (It may be helpful to make logic tables for each one.) i) xor in a dierent way than the circuit above. ii) p implies Q connective. iii) xor for three inputs. iv) Is a 3-digit binary number greater than 2? v) Is a 4-digit binary string a palindrome?
1.2
Statements with quantiers
The number x equals 1 is true for some x and false for some x. Thus x = 1 is not true or false universally. For determining a statements veracity the statement possibly needs to be further qualied. Two of the ways are with universal quantier for all or for every, and with existential quantier there exists or for some. Our example,
19
which is currently ambiguous, could be formally and precisely stated in one of the following two ways: (1) There exists a real number x such that x = 1. (2) For all real numbers x, x = 1, or For every real number x, x = 1. Certainly the rst statement is true and the second is false. For shorthand and for symbolic statement computations we abbreviate for all with the symbol , and there exists with . Thus for example the statements For all real numbers x, x2 is non-negative. There exists a real number x that is negative. can be written symbolically as real numbers x, x2 0. real number x such that x < 0.
Note that after x, the phrase such that is almost automatic, and is often omitted. The reader is invited to read the following symbolic statement: > 0 > 0 x, 0 < |x a| < |f (x) L| < . This last statement happens to say that the limit of the function f at a is L that we will encounter in Denition 4.1.1. For ease of readability it may be better to use full words rather than symbolic abbreviations. The general form of quantier statements is: x with a certain specication, P (x) holds
and x with a certain specication, P (x) holds where p is some property that can be applied to objects in question. When the specication of x is implicitly understood, we can make this form even shorter and simply write x P (x)
and x P (x)
We read for all x, P of x [holds/is true] and there exists x such that P of x [holds/is true], respectively. The part such that only appears with the existential quantier, and can be replaced with for which, but note that for some specic properties P it can be shortened further. For example: There exists a function f such that for all real numbers x, f (x) = f (x) can be rewritten with equal meaning as There exists a function f that is dened for all real numbers and is even, or even shorter as There exists an even function f dened for all real numbers.
20
Section 1.2: Statements with quantiers
A for all statement is true precisely when all x with the given description have the property P , and a there exists statement is true precisely when one or more x with the given description satises property P . One proves a for all statement by determining that each x with the given description has the property P , and one proves a there exists statement by producing one specimen x with the given description and proving that that specimen has property P . If there are no x with the given specication, then any property holds for those no-things x vacuously. For example, any positive real number that is strictly smaller than 1 is also zero, equal to 15, greater than 20, product of distinct prime integers, et cetera. In common language, there exists appears in various forms: at least once, somebody, something, somewhere, etc. Similarly, for all appears in common language as everything, everybody, everywhere, always, never, nobody, nowhere, etc. For example, Water never ows up can be written with for all as For all times t, it is not the case that water at time t ows up. What is the negation of x P (x)? Lets rst write down a truth table with all the possible situations in the rst column with regards to P , and other columns giving the truth values of various quantier statements: a possible situation there are no x of specied type true for all x false for all x true for some, false for some x xP (x) T vacuously T F F x P (x) T vacuously F T F xP (x) F F T T xP (x) F T F T
Notice that among the the columns with truth values, one and three have oppositive values, and two and four have oppositive values. This proves the following: Proposition 1.2.1 The negation of x P (x) is x P (x). The negation of x P (x) is x P (x). For this reason, x P (x) is false if there is even one tiny tiniest example to the contrary. Every prime number is odd is false because 2 is an even odd number. Every whole number divisible by 3 is divisible by 2 is false because 3 is divisible by 3 and is not divisible by 2. Symbolically, these two false statements can be written as prime numbers x, x is odd, and whole numbers x divisible by 3, x is divisible by 2. Remark 1.2.2 The statement For all whole numbers x between 1/3 and 2/3, x2 is irrational is true vacuously. Another reason why For all whole numbers x between 1/3 and 2/3, x2 is irrational is true is that its negation, There exists a whole number x between 1/3 and 2/3 for which x2 is rational, is false because there is no whole number
21
between 1/3 and 2/3: since the negation is false, we get yet more motivation to declare the original statement true. Note that for all and there exists do not encapsulate often, rarely, and similar ambiguously dened quantiers, unless we decree that often means say at least 22% of the times and rarely means say at most 2% of the times. Exercises for Section 1.2 1.2.1 Show that the following statements are false by providing counterexamples. i) No number is its own square. ii) All numbers divisible by 7 are odd. iii) The square root of all real numbers is greater than 0. iv) For every real number x, x4 > 0. 1.2.2 Rewrite the following statements using quantiers: i) No sum of two positive cubes is itself a cube. ii) 7 is prime. iii) There are innitely many prime numbers. iv) Everybody loves Raymond. v) Sometimes a lunar eclipse happens on a cloudless night. vi) Spring break is always in March. vii) Sam never sleeps in late. 1.2.3 Determine the truth value of the following statements, and justify your answers: i) For all real numbers x < 5, x2 > 16. ii) There exists a real number x < 5 such that x2 < 25. iii) There exists a real number x such that x2 = 4. iv) There exists a real number x such that x3 = 8. v) For all real numbers x there exists a positive integer n such that xn > 0. vi) For all real numbers x and integers n, |x| < xn . vii) For all integers m there exists an integer n such that m + n is even. viii) There exists an integer m such that integers n, m + n is even. ix) For every integer n, n2 n is even. x) Every list of 5 consecutive integers has one element that is a multiple of 5. xi) The element in previous part is a multiple of 3. xii) Every odd number is a multiple of 3. 1.2.4 Explain why the following two statements do not have the same truth values: i) For every x > 0 there exists y > 0 such that xy = 1. ii) There exists y > 0 such that for every x > 0, xy = 1.
22
Section 1.3: More proof methods, and negation
1.2.5 Explain why the following statements have the same truth values: i) There exists x such that there exists y such that xy = 1. ii) There exists y such that there exists x such that xy = 1. iii) There exists a pair (x, y ) such that xy = 1. 1.2.6 Explain why the following statements have the same truth values: i) x > 0 y > 0, xy > 0. ii) y > 0 x > 0, xy > 0. iii) (x, y ), x, y > 0 implies xy > 0.
1.3
More proof methods, and negation
When statements are compound, they can be harder to prove. Fortunately, proofs can be broken down into simpler statements. Here is a small chart of this breaking down: Statement P (via contradiction). P and Q. P or Q. How to prove it Suppose P . Establish some nonsense. Prove P . Prove Q. Suppose that P is false. Then prove Q. Alternatively: Suppose that Q is false and then prove P . (It may even be the case that P is true always. Then simply prove P . Or simply prove Q.) Suppose that P is true. Then prove Q. Contrapositively: Suppose that Q is false. Prove that P is false. Prove P Q. Prove Q P . Let x be arbitrary of the specied type. Prove that property P holds for x. Find/construct an x of the specied type. Prove that property P holds for x. Alternatively, invoke a theorem guaranteeing that such x exists. Suppose that x and x are both of specied type and satisfy property P . Prove that x = x . Alternatively, show that x is the only solution to an equation, or the only element on a list, or ....
If P then Q.
P Q. For all x of a specied type, property P holds for x. There exists x of a specied type such that property P holds for x.
An element x of a specied type with property P is unique.
23
Example 1.3.1 Prove that 2 and 3 are prime integers. (Prove that 2 is a prime integer and that 3 is a prime integer.) Proof. Let m and n be whole numbers strictly greater than 1. If m n = 2, then 1 < m, n 2, so m = n = 2, but 2 2 is not equal to 2. Thus 2 cannot be written as a product of two positive numbers dierent from 1, so 2 is a prime number. If instead m n = 3, then 1 < m, n 3. Then all combinations of products are 2 2, 2 3, 3 2, 3 3, none of which is 3. Thus 3 is a prime number. Example 1.3.2 Prove that a positive prime number is either odd or it equals 2. (One may think of 2 as a prime number as well, that is why positive appears. But often the term prime implicitly assumes positivity.) Proof. Let p be a positive prime number. Suppose that p is not odd. Then p must be even. Thus p = 2 q for some positive whole number q . Since p is a prime, it follows that q = 1, so that p = 2. Example 1.3.3 Prove that if an integer is a multiple of 2 and a multiple of 3, then it is a multiple of 6. (Implicit here is that the factors are integers.) Proof. Let n be an integer that is a multiple of 2 and of 3. Write n = 2 p and n = 3 q for some integers p and q . Then 2 p = 3 q is even, which forces that q must be even. Hence q = 2 r for some integer r , so that n = 3 q = 3 2 r = 6 r . Thus n is a multiple of 6. Example 1.3.4 Prove that for all real numbers x, x2 = (x)2 . Proof. Let x be an arbitrary real number. Note that (1)2 = 1, so that x2 = 1 x2 = (1)2 x2 = ((1) x)2 = (x)2 . Example 1.3.5 Prove that there exists a real number x such that x3 3x = 2. Proof. Observe that 2 is a real number and that 23 3 2 = 2. Thus x = 2 satises the conditions. Example 1.3.6 Prove that there exists a real number x such that x3 x = 1. Proof. Observe that f (x) = x3 x is a continuous function. Since 1 is strictly between f (0) = 0 and f (2) = 4, by the Intermediate value theorem (Theorem 5.2.1 in these notes) [invoking a theorem rather than constructing x, as opposed to in the previous example] there exists a real number x strictly between 0 and 2 such that f (x) = 1. Recall that this font in brackets and in red color indicates the reasoning that should go on in the background in your head; these statements are not part of a proof.
24
Example 1.3.7 (Mixture of methods) For every real number x strictly between 0 and 1 1 1 1 there exists a positive real number y such that x +y = xy . Proof. [We have to prove that for all x as specified some property holds.] Let x be in (0, 1). [For this x we have to find y ...] Set y = 1 x. [Was this a lucky find? No matter how we got inspired to determine this y , we now verify that the stated properties hold for x and y .] Since x is strictly smaller than 1, it follows that y is positive. Thus also xy is positive, and y + x = 1. After dividing 1 1 1 the last equation by the positive number xy we get that x +y = xy . Furthermore, y in the previous example is unique: it has no choice but to be 1 x. Example 1.3.8 (The Fundamental theorem of arithmetic) Any positive integer n > 1 can ak 1 a2 be written as pa 1 p2 pk for some positive prime integers p1 < < pk and some positive integers a1 , . . . , ak . (Another standard part of the Fundamental theorem of arithmetic is that the pi and the ai are unique, but we do not prove that. Once you are comfortable with proofs you can prove that part yourself.) Proof. (You may want to skip this proof for now.) Suppose for contradiction that the conclusion fails for some positive integer n. Then on the list 2, 3, 4, . . . , n let m be the smallest integer for which the conclusion fails. If m is a prime, take k = 1 and p1 = n, a1 = 1, and so the conclusion does not fail. Thus m cannot be a prime number, and so m = m1 m2 for some positive integers m1 , m2 strictly bigger than 1. Necessarily 2 m1 , m2 < m. ak 1 a2 By the choice of m, the conclusion is true for m1 and m2 . Write m1 = pa 1 p2 pk and bl b1 b2 m2 = q1 q2 ql for some positive prime integers p1 < < pk , q1 < < qk and some ak b 1 b 2 bl 1 a2 positive integers a1 , . . . , ak , b1 , . . . , bl . Thus m = pa 1 p2 pk q1 q2 ql is a product of positive prime numbers, and after sorting and merging the pi and qj , the conclusion follows also for m. But we assumed that the conclusion fails for m, which yields the desired contradiction. Hence the conclusion does not fail for any positive integer. Example 1.3.9 Any positive rational number can be written in the form a b , where a and b are positive whole numbers, and in any prime factorizations of a and b as in the previous example, the prime factors for a are distinct from the prime factors for b. Proof. [We have to prove that for all ...] Let x be a(n arbitrary) positive rational number. Thus x = a b for some whole numbers a, b. [Rewriting the meaning of assumptions.] If a is negative, since x is positive necessarily b has to be negative. But a then a, b are positive numbers, and x = . Thus by possibly replacing a, b with b a, b we may assume that a, b are positive. [A rewriting trick.] There may be many dierent pairs of a, b, and we choose a pair for which a is the smallest of all possibilities. [A choosing trick. But does the smallest a exist?] Such a does exist because
25
among a subset of positive integers there is always a smallest one. Suppose that a and b have a (positive) prime factor p in common. Write a = a0 p and b = b0 p for some positive 0 whole numbers a0 , b0 . Then x = a b0 , and since 0 < a0 < a, this contradicts the choice of the pair a, b. Thus a and b couldnt have had a prime factor in common. In order to be able to prove statements eectively, we often have to suppose the negation of a part, say for proving statements with or and for proofs by contradiction. Work and think through the following negations: Statement P P and Q P or Q P Q For all x of a specied type, property P holds for x. There exists x of a specied type such that property P holds for x. Negation P (P Q) = (P ) (Q) (P Q) = (P ) (Q) (P Q) = P (Q) There exists x of the specied type such that P is false for x. For all x of the specied type, P is false for x.
Example 1.3.10 (Due to Euclid.) There are innitely many (positive) prime numbers. Proof by contradiction: Suppose that there are only nitely many prime numbers. Lets enumerate them: p1 , p2 , . . . , pn . Let a = (p1 p2 pn ) + 1. Since we know that 2, 3, 5 are primes, necessarily n 3 and so a > 1. By the Fundamental theorem of arithmetic (Example 1.3.8), a has a prime factor p. Since p1 , p2 , . . . , pn are all the primes, necessarily p = pi for some i. But then p = pi divides a and p1 p2 pn , whence it divides 1 = a (p1 p2 pn ), which is a contradiction. So it is not the case that there are only nitely many prime numbers, so there must be innitely many. Exercises for Section 1.3 1.3.1 Prove that every whole number is either odd or even. 1.3.2 Prove that the successor of any odd integer is even. 1.3.3 Prove that if n is an even integer, then either n is a multiple of 4 or n/2 is odd. 1.3.4 Prove that if 0 < x < 1, then x2 < x < x. 1.3.5 Prove that if 1 < x, then x < x < x2 . 1.3.6 Prove that there exists a real number x such that x2 +
5 12 x
=1 6.
26
1.3.8 Why are f is continuous at all points and f is not continuous at 3 not negations of each other? 1.3.9 Why are Some continuous functions are dierentiable and All dierentiable functions are continuous not negations of each other? 1.3.10 Why is P Q not the negation of P Q? 1.3.11 Negate the following statements: i) The function f is continuous at 5. ii) I am reading a book and listening to radio. iii) She will either stay in Portland or move to Indiana. iv) Sam is taller than Pat. v) If x > y then x > z . vi) For every > 0, there exists > 0 such that f ( ) = . vii) For every > 0, there exists > 0 such that for all x, f (x ) = x. viii) For every > 0, there exists > 0 such that for all x, 0 < |x a| < implies |f (x) L| < . 1.3.12 Find at least three functions f such that for all real numbers x, f (x2 ) = x2 .
1.3.7 Prove that the following pairs of statements are negations of one another: i) P Q. (P Q) (P Q). ii) P (P Q). P Q. iii) (P Q) (R Q). (P Q) (R Q). iv) P (Q R). P (Q R).
1.3.13 Prove that f (x) = x is the unique function that is dened for all real numbers and that has the property that for all x, f (x3 ) = x3 . 1.3.14 Prove that there exists a real number z such that for all real numbers x, xz = z . 1.3.15 Prove that for every real number x there exists a real number y such that x + y = 0. 1.3.16 Prove that the following are false: i) For every real number x there exists a real number y such that xy = 1. ii) 5n + 2 is prime for all non-negative integers n. iii) For every real number x, if x2 > 4 then x > 2. iv) For every real number x, x2 = x. v) The cube of every real number is positive. vi) The square of every real number is positive.
27
1.4
Summation
There are many reasons for not writing out 1 + 2 + 3 + + 100 in full length: it would be too long, it would not be any clearer, we would probably start doubting the intelligence of the writer, it would waste paper and ink... The shorter way is with the summation sign :
100 100
k or
k=1 n=1
n.
The counters k and n above are dummy variables, they vary from 1 to 100. In general, if f is a function dened at m, m + 1, m + 2, . . . , n, we use the summation sign for shortening as follows:
n k =m
f (k ) = f (m) + f (m + 1) + f (m + 2) + + f (n).
This is one example where a good notation saves space (and it even sometimes claries the concept). Typographically, when summation is displayed, the two bounds (m and n) appear below and above the summation sign, but when the summation is in-line, the two n bounds appear to the right of the sign, like so: k=m f (k ). (This prevents lines jamming into each other.) Now is a good time to discuss polynomials. A polynomial function is a function of the form f (x) = a0 + a1 x + + an xn for some non-negative integer n and some numbers a0 , a1 , . . . , an . It is convenient to write this polynomial with the shorthand notation
n
f (x) = a0 + a1 x + + an x =
ak xk .
k=0
Here, of course, x0 stands for 1. When we evaluate f at 0, we get a0 = a0 + a1 0+ + an 0n n = k=0 ak 0k , and we deduce that notationally 00 stands for 1 here. Remark 1.4.1 00 could possibly be thought of also as lim 0x , which is surely equal to 0. But then, is 00 equal to 0 or 1 or to something else entirely? Well, it turns out that 00 is not equal to that zero limit you surely know of other functions f for which lim f (x) exists but the limit is not equal to f (c). (Consult also Exercise 6.3.10.) Similarly we can shorten products with the product sign :
n k =m xc x0+
f (k ) = f (m) f (m + 1) f (m + 2) f (n).
n
In particular, for all non-negative integers n, the product k=1 k is used often and is n abbreviated as n! = k=1 k . See Exercise 1.4.5 for the fact that 0! = 1.
28 Examples 1.4.2
4
Section 1.4: Summation
(1)
k=1 12
k 2 = 12 + 22 + 32 + 42 = 30. cos(k ) = cos(10 ) + cos(11 ) + cos(12 ) = 1 1 + 1 = 1. (4k 3 ) = 4 + 0 + 4 + 4 8 = 32.
(2)
k=10 2
(3)
k = 1 5
(4)
k=1 5
k = 120. 2 = 32.
k=1 0
(5)
We can even deal with empty sums such as k=1 ak : here the index starts at k = 1 and keeps increasing and we stop at k = 0, but there are no such indices k . What could possibly be the meaning of such an empty sum? Note that
4 2 4 1 4 0 4
ak =
k=1 k=1
ak +
k=3
ak =
k=1
ak +
k=2
ak =
k=1
ak +
k=1
ak ,
or explicitly written out: a1 + a2 + a3 + a4 = (a1 + a2 ) + (a3 + a4 ) = (a1 ) + (a2 + a3 + a4 ) = () + (a1 + a2 + a3 + a4 ), from which we deduce that this empty sum must be 0. Similarly, every empty sum equals 0. Exercises for Section 1.4 1.4.1 Compute 1.4.2 Prove that
n 4 k=0 (2k n
+ 1), cf (k ).
4 2 k=0 (k
+ 2).
i) c
k =m n
f (k ) =
k =m n
ii)
k =m n
f (k ) +
k =m n
g (k ) = f (k )
k =m m1
(f (k ) + g (k )). f (k ). g (k )) =
n k =m
iii)
k =m
f (k ) =
1.4.3 Show that ( 1.4.4 Prove that
k=1 k=1 n n f ( k )) ( k =m k =m 0 0(1) . k=1 k = 2
f (k ) g (k ) is false (in general).
1.4.5 Prove that the empty product equals 1. In particular, it follows that we can declare 0! = 1. This turns out to be very helpful notationally.
29
1.5
Proofs by (mathematical) induction
So far we have learned a few proof methods. There is another type of proofs that deserves special mention, and this is proof by (mathematical) induction. This method can be used when one wants to prove that a property P holds for all integers n greater than or equal to an integer n0 . Typically, n0 is either 0 or 1, but it can be any integer, even a negative one. Induction is a two-step procedure: (1) Base case: Prove that P holds for n0 . (2) Inductive step: Let n > n0 . Assume that P holds for all integers n0 , n0 + 1, n0 + 2, . . . , n 1. Prove that P holds for n. Why does induction succeed in proving that P holds for all n n0 ? By the base case we know that P holds for n0 . The inductive step then proves that P also holds for n0 + 1. So then we know that the property holds for n0 and n0 + 1, whence the inductive step implies that it also holds for n0 + 2. So then the property holds for n0 , n0 + 1 and n0 + 2, whence the inductive step implies that it also holds for n0 + 3. This establishes that the property holds for n0 , n0 + 1, n0 + 2, and n0 + 3, so that by inductive step it also holds for n0 + 4. We keep going. For any integer n > n0 , in n n0 step we similarly establish that the inductive step holds for n0 , n0 + 1, n0 + 2, . . . , n0 + (n n0 ) = n. Thus for any integer n n0 , we eventually prove that P holds for it. The same method can be phrased with a slightly dierent two-step process, with the same result, and the same name: 1. Base case: Prove that P holds for n0 . 2. Inductive step: Let n > n0 . Assume that P holds for integer n 1. Prove that P holds for n. Similar reasoning as in the previous case also shows that this induction principle succeeds in proving that P holds for all n n0 . Example 1.5.1 Prove the equality
n k=1
k=
n(n+1) 2
for all n 1.
1
Proof. Base case n = 1: The left side of the equation is k=1 k which equals 1. The right side is 1(1+1) which also equals 1. This veries the base case. 2 Inductive step: Let n > 1 and we assume that the equality holds for n 1. [We want to prove the equality for n. We start with the expression on the left side of the desired and not-yet-proved equation for n (the messier of the two)
30
Section 1.5: Proofs by (mathematical) induction
and manipulate the expression until it resembles the right side.] Then
n
k=
k=1
n1 k=1
+n
= = = = as was to be proved.
(n 1)(n 1 + 1) + n (by induction assumption) 2 n2 n 2n + (by algebra) 2 2 n2 + n 2 n(n + 1) , 2

n
+1) for all n 0. Since we have already We can even prove the equality k=1 k = n(n2 0 proved this equality for all n 1, it remains to prove it for n = 0. The left side k=1 k is an empty sum and hence 0, and the right side is 0(0+1) , which is also 0. 2
Example 1.5.2 (This is a basic derivative problem. We introduce derivatives formally in d (xn ) = nxn1 . [ Or should induction start Section 6.1.) Prove that for all n 1, dx at n = 0? In Section 1.4 we have determined to agree that 00 equals 1, but was that simply a notational convenience rather than truth?] Proof. [If n = 0, then x0 = 1 and its derivative is 0 = 0 x01 . At least when x = 0, the conclusion follows, but at x = 0, it looks like we are dividing by 0. So the equality holds most of the time in this case. What kind of mathematical statement is most of the time? We should not like this!] So we start the induction instead at n = 1, as instructed. By calculus we know that the derivative of x1 is 1 = 1 x0 = 1 x11 , so equality holds in this case. Inductive step: Suppose that equality holds for [0, ]1, 2, . . . , n 1. Then d n d (x ) = (x xn1 ) dx dx d d (x) xn1 + x (xn1 ) (by the product rule of dierentiation) = dx dx n1 =x + (n 1)x xn2 (by induction assumption for 1 and n 1) = xn1 + (n 1)xn1 = nxn1 .
Note that if in the previous example we only proved the base case n = 0, the inductive step there does not prove the case n = 1. The moral of the story is that we have to be careful about how we use induction (or proofs) in general.
Chapter 1: How we will do mathematics The following result will be needed many times, so remember it well. Theorem 1.5.3 For any number x and all integers n 1, (1 x)(1 + x + x2 + x3 + + xn ) = 1 xn+1 . Proof. When n = 1, (1 x)(1 + x + x2 + x3 + + xn ) = (1 x)(1 + x) = 1 x2 = 1 xn+1 ,
31
which proves the base case. Now suppose that equality holds for some integer n 1 1. Then (1 x)(1 + x + x2 + + xn1 + xn ) = (1 x) (1 + x + x2 + + xn1 ) + xn = (1 x)(1 + x + x2 + + xn1 ) + (1 x)xn = 1 xn + xn xn+1 (by induction assumption and algebra)
= 1 xn+1 ,
which proves the inductive step. Exercises for Section 1.5: Prove the following properties for n 1 by induction.
n
1.5.1
k=1 n
k2 = k =
k=1 n 3
n(n + 1)(2n + 1) . 6 n(n + 1) 2

2
1.5.2
1.5.3 The sum of the rst n odd integers is n2 . 1.5.4 (2k 1) = n2 .
k=1 n
1.5.5 For all real numbers a1 , . . . , an , |a1 + a2 + + an | |a1 | + |a2 | + + |an |. 1.5.6
k=1
(3k 2 i) = n2 (n + 1).
1.5.9 Let An =
1.5.8 Let An = 12 + 22 + 32 + + (2n 1)2 . Find a formula for An and prove it.
1 12
1 1.5.7 1 2 + 2 3 + 3 4 + + n(n + 1) = 3 n(n + 1)(n + 2).
1.5.10 There are exactly n people at a gathering. Everybody shakes everybody elses hands exactly once. How many handshakes are there? 1 xn+1 2 n for x = 1. Exercise 1.5.11 1 + x + x + + x = 1x 1.5.12 7n + 2 is a multiple of 3.
1 23
1 34
++
1 . n(n+1)
Find a formula for An and prove it.
32
Section 1.5: Proofs by (mathematical) induction
1.5.13 3n2 < (n + 1)!. 1 1 1 1 1.5.14 + + + + n. n 1 2 3 1 1 1 1 1 1.5.15 2 + 2 + 2 + + 2 2 . 1 2 3 n n 1.5.16 Let a1 = 2, and for n 2, an = 3an1 . Formulate and prove a theorem giving an in terms of n (no dependence on other ai ). 1.5.17 Every integer greater than 2 is a nite product of primes. 1.5.18 8 divides 5n + 2 3n1 + 1. 1.5.19 1(1!) + 2(2!) + 3(3!) + + n(n!) = (n + 1)! 1. 1.5.20 2n1 n!. 1.5.21 (n 2?)
n n
k=2
1 k2
n+1 . 2n
1.5.22
k=0
2k (k + 1) = 2n+1 n + 1.
Exercise 1.5.23 (This is invoked in Example 9.1.6.) Prove that for all positive integers n, 2n 1 k=1 k n + 2/2. Exercise 1.5.24 (This is invoked in Example 9.1.7.) Prove that for all positive integers 2 n 1 1 n 1 n, k=1 k 2 k=0 2k . Exercise 1.5.25 (This is invoked in the Ratio test for series Theorem 9.2.4.) Suppose that for all positive integers n, an+1 < ran . Prove that for all positive integers n, an+1 < r n a1 . Exercise 1.5.26 (This is invoked in the proof of Theorem 9.4.2.) Prove that for all numbers x, y and all positive integers n, xn y n = (x y )(xn1 + xn2 y + xn3 y 2 + + n1 y n1 = (x y ) k=0 xn1k y k . 1.5.27 (Tower of Hanoi) There are 3 pegs on a board. On one peg, there are n disks, stacked from largest to smallest. The task is to move all of the disks from one peg to a dierent peg, given the following constraints: you may only move one disk at a time, and you may only place a smaller peg on a larger one (never a larger one on a smaller one). i) If n = 1, then what is the least number of moves it takes to complete the task? What if there are 2 disks? Repeat for 3, 4, 5 disks. ii) Let Sn be the least number of moves to complete the task for n disks. Make a recursive formula (dening Sn based on Sn1 ) for this Sn . Then, make a guess for a non-recursive formula for Sn (dening Sn based on n without invoking Sn1 ).
iii) Prove your guess using induction and the recursive formula that you wrote.
33
*1.5.28 Let a1 , . . . , an+1 be positive integers not exceeding 2n. Prove that some element on this list divides another one. 1.5.29 What is wrong with the following proof by induction? I will prove that 5n + 1 is a multiple of 4. Assume that this is true for n 1. Then we can write 5n1 + 1 = 4m for some integer m. Multiply this equation through by 5 to get that 5n + 5 = 20m, whence 5n + 1 = 4(5m 1). As 5m 1 is an integer, this proves that 5n + 1 is a multiple of 4. 1.5.30 What is wrong with the following proof by induction? I will prove that all horses are of the same color. This is the same as saying that for any integer n 1 and any set of n horses, all the horses belonging to the set have the same color. If n = 1, of course this only horse is the same color as itself, so the base case is proved. Now let n > 1. If we remove one horse from this set, the remaining n 1 horses in the set are all of the same color by the induction assumption. Now bring that one horse back into the set and remove another horse. Then again all of these horses are of the same color, so the horse that was removed rst is the same color as all the rest of them.
1.6
Pascals triangle
of Pascals triangle, and the pattern is obvious . . 1 4 10 15 35 56 70 20 35 56 . 1 3 6 10 15 21 28 1 1 2 3 4 5 6 7 8 1 1 1 1 1 1 1
The following is rows 0 through 8 for continuation into further rows: row 0: . . . . . . . row 1: . . . . . . . row 2: . . . . . . . row 3: . . . . . . 1 row 4: . . . . . 1 row 5: . . . . 1 5 row 6: . . . 1 6 row 7: . . 1 7 21 row 8: . 1 8 28
Note that the leftmost and rightmost numbers in each row are all 1, and each of the other numbers is the sum of the two numbers nearest to it in the row above it. We number the slanted columns from left to right starting from 0: the 0th slanted column consists of all 1s, the 1st slanted column consists of consecutive numbers 1, 2, 3, 4, . . ., the 2nd slanted column consists of consecutive numbers 1, 3, 6, 10, . . ., and so on for the subsequent columns. Let the entry in the nth row and k th column be denoted n k . We read this as n choose k. These are loaded words, however, and we will eventually justify these words.
34
Section 1.6: Pascals triangle Pascals triangle is dened so that for all n 0 and all k = 0, 1, . . . , n 1, n n + k+1 k = n+1 . k+1
What would it take to compute 100 ? It seems like we would need to write down rows 5 0 through 100 of Pascals triangle, or actually a little less, only slanted columns 0 through 5 of these 101 rows. That is too much drudgery! We will instead be smart mathematicians and we will prove many properties of Pascals triangle in general, including shortcuts for computing 100 5 . We will accomplish this through exercises, most of which can be proved by mathematical induction. Exercises for Section 1.6 1.6.1 Prove that the sum of the entries in the nth row is 2n .
+1 1.6.2 Prove that n k+1 is the sum of the entries in slanted column k in rows j n. In n +1 k k+1 j other words, prove that n + k+2 ++ n j =k k . k+1 = k + k k k =
1.6.3 Prove that
n k
100 , 8 1.6.4 Compute 4 2 , 2 , 2 , and do cancellations before multiplications.)
n! k!(nk)! . 7 6 5 2 , 2 , 2
100 3
100 4
100 5
. (Hint: be clever about it
1.6.5 Prove that n k is the number of possible k -member teams in a club with exactly n members. For this reason n is read n choose k. k Exercise 1.6.6 (This is invoked in Theorem 2.8.2, Example 8.2.6, Theorem 6.2.3.) Prove that for all non-negative integers n,
n
(a + b) =
k=0
n k nk a b . k
(Since a + b contains two summands, it is called a binomial, and the expansion of (a + b)n is called the binomial expansion, with coecients n being called by yet another name i in this context: binomial coecients.)
n 1.6.7 Prove that for all non-negative integers n and all k = 0, 1, . . . n, n k k! . Fix a positive integer k . Prove that there exists a positive number C such that for all suciently large integers n, Cnk n k .
k
1.6.8 Give reasons why we should have n k = 0 for n < k or if kn < 0. (It is convenient to have the enlargement of the Pascals triangle where we allow all combinations of rows and columns, even negatively indexed ones.)
Chapter 2: Concepts with which we will do mathematics
2.1
Sets
What is a set? Dont we already have an idea of what a set is? The following formal denition is trying to make the intuitive idea precise: Denition 2.1.1 A set is a collection of objects. These objects are called members or elements of that set. If m is a member of a set A, we write m A. If m is not a member of a set A, we write m A. So, while perhaps the denition does not tell us much new about what we think of as sets, it does tell us the notation and vocabulary with which we can write and talk about sets. For example, the set of all polygons contains triangles, squares, rectangles, pentagons, among other polygons. The set of all polygons does not contain circles or disks. The set of all functions contains the trigonometric, logarithmic, exponential, constant functions, and so on. Examples and notation 2.1.2 Study the notation below: (1) Intervals are sets: (0, 1) is the interval from 0 to 1 that does not include 0, 1, (0, 1] is the interval from 0 to 1 that does not include 0 but includes 1, [0, 1) is the interval from 0 to 1 that does not include 1 but includes 0, [0, 1] is the interval from 0 to 1 that includes 0, 1, (5, ) is the interval of all real numbers strictly bigger than 5, [5, ) is the interval of all real numbers bigger than or equal to 5, (, 5) is the interval of all real numbers strictly smaller than 5, (, 5] is the interval of all real numbers smaller than or equal to 5, (, ) is the set of all real numbers. (2) Descriptively we can say for example that a set A is the set of all prime numbers. (3) We can list the elements and surround them with curly braces to dene a set: (i) {1, 2, 3} is the set consisting of precisely 1, 2 and 3. (i) {1, 2, 3, 2} is the set consisting of precisely 1, 2 and 3. Thus {1, 2, 3, 2} = {1, 2, 3}. (iii) {blue, hello, 5} is the set consisting of precisely of the color blue, of the word
36
Section 2.1: Sets hello, and of number 5. (iv) {1, 2, {1, 2}} is the set consisting of precisely of numbers 1 and 2 and of the set {1, 2}. This set has exactly three distinct elements, it is not the same as {1, 2}, and it is not the same as {{1, 2}}.
(5) When the list of elements is not small enough for reasonable explicit listing but the pattern of elements is clear, we can start the list and then add , . . . when the pattern is clear: (i) {1, 2, 3, . . . , 10000} is the set of all positive integers that are at most 10000. (ii) {1, 4, 9, . . . , 169} is the set of the rst 13 squares of integers. (iii) {0, 1, 2, 3, . . .} is the set consisting of all non-negative whole numbers. This set is often denoted by N. WARNING: in some books, N stands for the set of all positive whole numbers. To distinguish between the two, we will write N0 for {0, 1, 2, 3, . . .} and N+ for {1, 2, 3, . . .}.
(4) The set with no elements is called the empty set and is denoted or {}. The set {} is not empty because it contains the empty set.
(6) Warning: {3, 5, 7, . . .} or {3, 5, 7, . . . , 101} could stand for the set of all odd primes (up to 101), or possibly for the set of all odd whole numbers strictly greater than 1 (up to 101). Avoid ambiguities: write more elements, or write an explicit description of the elements instead.
(7) The set of all whole numbers is written Z, the set of all rational numbers is written Q, the set of all real numbers is written R, and the set of all complex numbers is written C. (Comment: at this point I rely on your basic training to understand the examples N, N0 , N+ , Z, Q, R, C; in the next chapter we construct these sets from basic set theory and another axiom.) (8) We can also dene sets propositionally: we write {x : P (x)} or {x A : P (x)} for the set of all x (or x A) for which property P holds. Here are some examples: (i) {x R : x > 0 and x < 1}, and this happens to be the interval (0, 1). (ii) Q = { a b : a, b Z, b = 0}. (iii) {n N : n > 1 and if n = pq for some positive integers p, q then p = 1 or q = 1} is the same as the set of all positive prime numbers. (iv) A = {x : x is a positive integer that equals the sum of its proper factors}. Who are the elements of A? They have to be positive integers, and it is easy to verify that 1, 2, 3, 4, 5 are not in A. But 6 has factors 1, 2, 3, 6, and the sum of the factors 1, 2, 3 other than 6 equals 6. Thus 6 is an element of A. Verify that 28, 496, 8128 are also in A (and if you have a lot of time, program a computer to verify that no other number smaller than 33 million is in A).
37
(9) Proving that a property P holds for all integers n n0 is the same as saying that the set A = {n Z : P holds for n} equals {n0 , n0 + 1, n0 + 2, . . .}. By the principle of mathematical induction, P holds for all integers n n0 is the same as saying that n0 A and that n 1 A implies that n A. Summary of example sets, and their notation = {}: the set with no elements. {a, b, c}, {a, b, . . . , z }. {x : x can be written as a sum of three consecutive integers}. N: the set of all Natural numbers. Depending on the book, this could be the set of all positive integers or it could be the set of all non-negative integers. The symbols below are unambiguous: N0 : the set of all non-negative integers; N+ : the set of all positive integers. Z: the set of all integers (Zahlen in German). Q: the set of all rational numbers (Quotients). R: the set of all Real numbers. C: the set of all Complex numbers. Just like numbers, functions, and logical statements, sets and their elements can also be related and combined in meaningful ways. The list below introduces quite a few new concepts that may be overwhelming at rst, but in a few weeks you will be very comfortable with them. Subsets: A set A is a subset of a set B if every element of A is an element of B . In that case we write A B . For example, N+ N0 Z Q R. The non-subset relation is expressed with the symbol : R N. Every set is a subset of itself, i.e., for every set A, A A. The empty set is a subset of every set, i.e., for every set A, A. If A is a subset of B and A is not equal to B (so B contains at least one element that is not in A), then we say that A is a proper subset of B , and we write A B . For example, N+ N0 Z Q R. Equality: Two sets are equal if they consist of exactly the same elements. In other words, A = B if and only if A B and B A. Intersection: The intersection of sets A and B is the set of all objects that are in A and in B : A B = {x : x A and x B }. When A B = , we say that A and B are disjoint.
38
Section 2.1: Sets
Union: The union of sets A and B is the set of all objects that are either in A or in B : A B = {x : x A or x B }. Intersections and unions of arbitrary families of sets: We have seen intersections and unions of two sets at a time. We can also take intersections and unions of three, four, ve, and even innitely many sets at a time. Verify the equalities below: (A B ) C = A (B C ),
(A B ) (C D) = (A (B C )) D, etc Since parentheses are irrelevant, we simply write the four sets above as A B C , A B C , A B C D, A B C D, respectively. More generally, given sets A1 , A2 , . . . , An , we write
n k=1 n k=1
(A B ) (C D) = A (B C D), etc
(A B ) C = A (B C ),
Ak = A1 A2 An = {a : a Ak for all k = 1, . . . , n}, Ak = A1 A2 An = {a : there exists k = 1, . . . , n such that a Ak },
or even more generally, when Ak are sets as k varies over a possibly innite index set I , then
kI kI
Ak = {a : a Ak for all k I }, Ak = {a : there exists k I such that a Ak }.
When I is the empty index set, one can argue as for empty sums and empty products in Section 1.4 that Ak = the universal set that contains all the Ak ,
k k
Ak = .
We return to this theme in Section 2.5. Complement: The complement of A in B is B \ A = {b B : b A}. Some authors write B A, but that has another meaning as well: B A : {b a : b B, a A}. Always try to use precise and unambiguous notation.
39
We often have an implicit or explicit universal set that contains all elements of our current interest. Perhaps we are talking only about real numbers, or perhaps we are talking about all functions dened on the interval [0, 1] with values being real numbers. In that case, for any subset A of the universal set U , the complement of A is the complement of A in U , thus U \ A, and this is denoted as Ac . Summary notation and vocabulary a A: a is an element of a set A. A B : A is a subset of set B ; every element of A is an element of B . A B : A B and A = B ; A is a proper subset of B . A = B : A B and B A. A B : the set of all elements that are in A and in B . A and B are disjoint: A B = . A B : the set of all elements that are either in A or in B . A \ B : the set of all elements of A that are not in B . Ac : the set of all elements in the universal set that are not in A. Example 2.1.3 For each i N, let Ai = [i, ), Bi = {i, i + 1, i + 2, . . . , }, and Ci = (i, i). Think through the following:
kN kN
Ak = , Ak = [1, ),
kN
Bk = ,
kN
Ck = (1, 1), Ck = R .
kN
Bk = N,
kN
Example 2.1.4 For each real number r , let Ar = {r }, Br = [0, |r |]. Then
r R
Ar = ,
r R
Br = {0},
Ar = R,
r R r R
Br = [0, ).
Set operations can be represented with a Venn diagram, especially in the presence of a universal set U . Here is an example: A B U
On this Venn diagram, sets are represented by the geometric regions: A is the set represented by the left circle, B is represented by the right circle, A B is the part of the
40
Section 2.1: Sets
two circles that is both in A and in B , A B can be represented by the region that is either in A or in B , A \ B is the left crescent after B is chopped out of A, etc. (There is no reason why the regions for sets A and B are drawn as circles, but this is traditional.) Sometimes we draw a few (or all) elements into the diagram. For example, in A y w B t x u U
z v
we read that A = {x, y, t}, A B = {t}, et cetera. Two disjoint sets A and B are represented by a Venn diagram as follows: A B U
Proposition 2.1.5 For all sets A, B, C , we have A (B C ) = (A B ) (A C ). Proof. With the Venn diagram below, B C is the region lled with either horizontal or vertical lines, A is the region lled with the Southeast-Northwest slanted lines, and so A (B C ) is the region that has simultaneously the Southeast-Northwest slanted lines and either horizontal or vertical lines. Also, A C is the region that has horizontal and Southeast-Northwest slanted lines, A B is the region that has vertical and SoutheastNorthwest slanted lines, so that their union (A B ) (A C ) represents the total region of Southeast-Northwest slanted lines that either have horizontal or vertical cross lines as well, which is the same as the region for A (B C ). C A B U
41
Lets prove this also algebraically. We have to prove that A (B C ) (A B ) (A C ) and (A B ) (A C ) A (B C ). Let x be arbitrary in A (B C ). This says that x A and x B C , and the latter says that either x B or x C . But then either x (A B ) or x (A C ), so that x (A B ) (A C ). This proves one inclusion. Now let x be arbitrary in (A B ) (A C ). This says that either x A B or x A C , which in turn says that either x is in A and in B or else that x is in A and in C . In any case, x A, and either x B or x C , so that x A and x B C , which nally says that x A (B C ). This proves the other inclusion. Exercises for Section 2.1 2.1.1 Prove by induction on n that a set with n elements has exactly 2n distinct subsets. 2.1.2 Prove that a set with n elements has
n i
subsets with exactly i elements.
2.1.3 Prove the following: i) {x R : x2 = 3} = { 3, 3}. ii) {x3 : x R} = R. iii) {x2 : x R} = {x R : x 0} = [0, ). iv) {2, 2, 5} = {2, 5} = {5, 2}. v) {x 0 : x is an even prime number} = {2}. vi) is a subset of every set. Elements of are green, smart, sticky, hairy, feathery, prime, whole, negative, positive,... vii) {x : x can be written as a sum of three consecutive integers} = {3n : n Z}. viii) If A B , then A B = A and A B = B . 2.1.4 Let U = {1, 2, 3, 4, 5, 6}, A = {1, 3, 5}, and B = {4, 5, 6}. Find the following sets: i) (A \ B ) (B \ A). ii) U \ (B \ A). iii) U (B \ A). iv) U \ (A B ). v) (U A) (U B ). vi) A \ (A \ B ). vii) B \ (B \ A). viii) {A} {B }.
2.1.5 Assume that A and B are disjoint subsets of U . For each part below, draw a Venn diagram with A and B in U , and shade in the region described by the set: (i) U \ B , (ii) A B , (iii) U \ (U \ A), (iv) (U \ A) (U \ B ), (v) (U B ) (U \ (B A)), (vi) (B U ) (B \ U ), (vii) (U \ (U \ A)) B , (viii) (A B ) (U \ A).
42
Section 2.2: Cartesian products
2.1.6 Let A, B U . i) Prove that there exist at most 16 distinct subsets of U obtained from A, B, U by intersections, unions, and complementation. ii) If A and B are disjoint, prove that there exist at most 8 such distinct subsets. iii) If A = B , prove that there exist at most 4 such distinct subsets. iv) If A = B = U , prove that there exist at most 2 such distinct subsets. v) If A = B = U = , prove that there exists at most 1 such subset. 2.1.7 Let A, B, C U . Prove the following statements: i) (A C ) \ B = (A \ B ) (C \ B ). ii) (A \ B ) (B \ A) = (A B ) \ (A B ). iii) (A B ) (U \ (A B )) = (U \ (A \ B )) \ (B \ A). iv) U \ (A \ B ) = (U \ A) B . v) If U = A B , then A = B = U . 2.1.8 Compute
kN+
(1/k, 1/k ), (1/k, 1/k ),
kN+ kN+
[1/k, 1/k ], [1/k, 1/k ],
kN+ kN+
{1/k, 1/k }. {1/k, 1/k }.
2.1.9 Compute
kN+
2.2
Cartesian products
The set {a, b} is the same as the set {b, a}, as any element of either set is also the element of the other set. Thus, the order of the listing of elements does not matter. But sometimes we want the order to matter. We can then simply make another new notation for ordered pairs, but in general it is not a good idea to be inventing many new notations and concepts; it is much better if we can reuse and recycle old ones. We do this next: Denition 2.2.1 An ordered pair (a, b) is dened as the set {{a}, {a, b}}. So here we dened (a, b) in terms of already known constructions: (a, b) is the set one of whose elements is the set {a} with exactly one element a, and the remaining element sof (a, b) is the set {a, b} that has exactly two elements a, b if a = b and has exactly one element otherwise. Thus for example the familiar ordered pair (2, 3) from high school and calculus classes really stands for {{2}, {2, 3}}, (3, 2) stands for {{3}, {2, 3}}, and (2, 2) stands for {{2}, {2, 2}} = {{2}, {2}} = {{2}}. Proposition 2.2.2 (a, b) = (c, d) if and only if a = c and b = d.
43
Proof. [Recall that P Q is the same as P Q and P Q. Thus the proof will consist of two parts.] Proof of : Suppose that (a, b) = (c, d). Then by the denition of ordered pairs, {{a}, {a, b}} = {{c}, {c, d}}. If a = b, this says that {{a}} = {{c}, {c, d}}, so that {{c}, {c, d}} has only one element, so that {c} = {c, d}, so that c = d. But then {{a}, {a, b}} = {{c}, {c, d}} is saying that {{a}} = {{c}}, so that {a} = {c}, so that a = c. Furthermore, b = a = c = d, which proves the consequent in case a = b. Now suppose that a = b. Then {{a}, {a, b}} = {{c}, {c, d}} has two elements, and so c = d. Note that {a} is an element of {{a}, {a, b}}, hence of {{c}, {c, d}}. Thus necessarily either {a} = {{c} or {a} = {c, d}}. But {a} has only one element and {c, d} has two (since c = d, it follows that {a} = {{c}, so that a = c. But then {a, b} = {c, d}, and since a = c, it follows that b = d. This proves the consequent in the remaining cases. Proof of : If a = c and b = d, then {a} = {c} and {a, b} = {c, d}, so that {{a}, {a, b}} = {{c}, {c, d}}. Note that by our denition an ordered pair is a set of one or two sets. Denition 2.2.3 For any sets A and B , the Cartesian product A B of A and B is the set {(a, b) : a A, b B } of ordered pairs. In general, one can think of A B as the rectangle with A on horizontal side and B on the perpendicular side. Say, if A has 4 elements and B has 3 elements, then A B can be represented by the 12 points in the rectangle with base consisting of elements of A and height consisting of elements of B as follows:
A If instead A and B are intervals as above, then A B is the indicated rectangle. When A and B extend innitely far, then A B is correspondingly a large rectangle: The familiar real plane is R R. (The three-dimensional space is R (R R) or (R R) R. In the former case we write elements in the form (a, (b, c)), and in the latter case we write them in the form ((a, b), c). Those extra parentheses are there only for notation and to slow us down, they serve no better function, so by convention we write elements simply in the form (a, b, c).)
44 Exercises for Section 2.2
Section 2.3: Relations, equivalence relations
2.2.1 Let A have m elements and B have n elements. Prove that A B has mn elements. Prove that A B has 2mn subsets. 2.2.2 How many elements are in R? 2.2.3 Prove that A (B C ) = (A B ) (A C ).
2.2.4 Prove that A (B C ) = (A B ) (A C ).
2.2.5 Prove that (A C ) (B D) = (A B ) (C D).
2.2.6 Prove that (A C ) (B D) = (A B ) (C D).
2.3
Relations, equivalence relations
In this section we introduce relations in a formal way. Most relations that we eventually analyze will be of familiar kind, such as cousin, taller than, , <, et cetera, but we can get much more structure with a formal approach. Denition 2.3.1 A relation on sets A and B is a subset of A B . A relation on A is a subset of A A. We can give relation a name, such as R, and in place of (a, b) R we alternately write aRb. Examples 2.3.2 (1) Some relations on R are , <, =, , >. For example, is a subset of the real plane consisting of all points on or above the line y = x. (2) Some further relations on R: R = {(a, b) : a, b R, a2 < b + 1}. (3) If A = {1, 2} and B = {a, b}, then the following are all the possible relations on A and B : {(1, a), (1, b), (2, a), (2, b)}, {(1, a), (2, a), (2, b)}, {(1, a), (1, b), (2, a)}, {(1, a), (1, b), (2, b)}, {(1, b), (2, a), (2, b)},
{(1, a), (1, b)},
{(1, a), (2, b)}, {(1, b), (2, b)},
{(1, a), (2, a)}, {(1, b), (2, a)},
Chapter 2: Concepts with which we will do mathematics {(2, a), (2, b)}, {(1, b)}, {(1, a)},
45
{(2, b)}, {}.
{(2, a)},
Denition 2.3.3 Let R be relation on A. (1) R is reexive if for all a A, aRa. (2) R is symmetric if for all a, b A, aRb implies bRa. (3) R is transitive if for all a, b, c A, if aRb, bRc, then aRc. (4) R is an equivalence relation if it is reexive, symmetric and transitive. Examples 2.3.4 (1) on R is reexive and transitive but not symmetric. (2) < on R is transitive but not reexive or symmetric. (3) = on any set A is reexive, symmetric, and transitive. (4) Being a cousin is symmetric but not reexive or transitive. (5) Being taller than is ... Denition 2.3.5 Let R be an equivalence relation on a set A. For each a A, the set of all elements b of A such that aRb is called the equivalence class of a. We denote the equivalence class of a with the shorthand [a]. For example, if R is the equality relation, then the equivalence class of a is {a}. If R = A A, then the equivalence class of any a is A. If A = [0, 1] and R is the subset of the square A A consisting of all points on the diagonal or below it, then for all a A, [a] = [0, a], i.e., the equivalence class of a consists of all real numbers on the interval [0, a]. If A is the set of all students in Math 112 this year, and aRb if students a and b are in the same section of Math 112, then [a] is the set of all students that are in the same section as student a. Theorem 2.3.6 Let R be an equivalence relation on a set A. Two equivalence classes are either identical or they have no elements in common. Proof. Let a, b A, and suppose that their equivalence classes have an element in common. Call the element c. We now prove that the equivalence class of a is a subset of the equivalence class of b. Let d be any element in the equivalence class of a. Then aRd, aRc and bRc imply by
46
Section 2.3: Relations, equivalence relations
symmetry that dRa and cRb, so that by transitivity dRc. Then dRc, cRb and transitivity give dRb, so that by symmetry bRd, which says that d is in the equivalence class of b. Thus the equivalence class of a is a subset of the equivalence class of b. A symmetric proof shows that the equivalence class of b is a subset of the equivalence class of a, so that the two equivalence classes are identical. What this says is that whenever R is an equivalence relation on a set A, then every element of A is in a unique equivalence class. Important Example 2.3.7 Let n be a positive integer. Let R be the relation on Z given by aRb if a b is a multiple of n. This relation is called congruence modulo n, and it is an equivalence relation by Exercise 2.3.8. We denote the set of all equivalence classes with Zn . This set consists of [0], [1], [2], . . . , [n 1], [n] = [0], [n + 1] = [1], et cetera, so that Zn has at most n equivalence classes. Since any two numbers among 0, 1, . . . , n 1 have dierence strictly between 0 and n, it follows that this dierence is not an integer multiple of n, so that [0], [1], [2], . . . , [n 1] are distinct. Thus Zn has exactly n equivalence classes. Two natural lists of representatives of equivalence classes are 0, 1, 2, . . . , n 1 and 1, 2, . . . , n. (But there are innitely many other representatives as well.) In everyday life we use congruence modulo 12 (or sometimes 24) for hours, congruence modulo 7 for days of the week, congruence modulo 4 for seasons of the year, congruence modulo 12 for months. In the latter congruence, the equivalence class of 1 is {1, 13, 25, 37, . . .} {11, 23, 35, . . .}. The equivalence class of 12 is all multiples of 12 (including 0). Example 2.3.8 (Construction of Z from N0 .) Consider the Cartesian product N0 N0 . Elements are pairs of the form (a, b), with a, b N0 . If a, b, c, d N0 , we will write (a, b) (c, d) if a + b = c + d. Thus is a relation on N0 N0 . (Certainly you are familiar with Z, in which case you may want to think of this relation simply saying that (a, b) (c, d) if a b = c d. The problem is that in N0 we may not be able to subtract b from a and still get an element from N0 .) (1) is reexive: because for all (a, b) N0 N0 , by commutativity of addition, a + b = b + a, so that by denition of , (a, b) (a, b). (2) is symmetric: if (a, b) (c, d), then by denition a + d = b + c, so that by commutativity of addition, d + a = c + b, and by symmetry of the = relation, c + b = d + a. But by denition this says that (c, d) (a, b). (3) is transitive: if (a, b) (c, d) and (c, d) (e, f ), then by denition a + d = b + c and c + f = d + e. It follows that (a + d) + (c + f ) = (b + c) + (d + e). By associativity and commutativity of addition, (a + f ) + (c + d) = (b + e) + (c + d), and by cancellation then a + f = b + e, which says that (a, b) (e, f ).
47
Now we dene a set Z to be the set of equivalence classes for this relation. Every element (a, b) of N0 N0 is in a unique equivalence class. If a b, then (a, b) (a b, 0), and if a < b, then (a, b) (0, b a). Thus for each (a, b) N0 N0 , there is an element in the equivalence class of (a, b) of the form (0, e) or (e, 0) for some e N0 , and it is left for the reader to verify that this e is unique. With this we can identify the set of equivalence classes with the usual integers as follows: the equivalence class of (e, 0) corresponds to the non-negative integer e and the equivalence class of (0, e) corresponds to the non-positive integer e. This was a construction of Z from N0 . Note that we needed to understand some arithmetic on N0 (commutativity, associativity and cancellation of addition) in the construction of the set Z. Later (see Section 3.5) we will construct also arithmetic on Z. Exercises for Section 2.3 2.3.1 Let A = {a, b} and B = {b, c, d}. i) How many elements are in the set AA? ii) How many elements are in the set AB? iii) How many elements are in the set B B? iv) How many relations are there on set B ? v) How many relations are there on A to B ? On B and A? vi) How many relations are there on the set A B and A B ? 2.3.2 Let A have n elements and B have m elements. How many distinct relations on A and B are there? Prove. 2.3.3 In each part below, nd a relation with the given properties. You may nd such a relation among number or human relations, or you may contrive a relation on a contrived set A. i) Reexive, but not symmetric and not transitive. ii) Reexive and symmetric, but not transitive. iii) Reexive and transitive, but not symmetric. iv) Symmetric, but not reexive and not transitive. v) Transitive, but not symmetric and not reexive. vi) Transitive and symmetric, but not reexive. 2.3.4 Let A be the set of all points in the plane. i) Is the relation is parallel to an equivalence relation? Prove your assertion. ii) Is the relation is perpendicular to an equivalence relation? Prove your assertion. 2.3.5 Let S be a set with 2 elements. How many possible equivalence relations are there on S ? Repeat rst for S with 3 elements, then for S with 4 elements.
48
Section 2.4: Functions
2.3.6 Let A be a set with n elements. Let R be an equivalence relation on A with fewest members. How many members are in R? 2.3.7 Let A be a set. i) Let R be an equivalence relation on A. Prove that the equivalence classes are pairwise disjoint and that their union is A. ii) Conversely, prove that whenever Bi , for i varying over some index set I , are pairwise disjoint subsets of A whose union is A, then R = {(a, b) : i such that a, b Bi } is an equivalence relation on A. iii) Argue that there is a natural correspondence between equivalence relations on A and ways of writing up A as a union of disjoint subsets. Exercise 2.3.8 (Invoked in Example 2.3.7.) Let n be an arbitrary integer. Let R be the relation on Z given by aRb if a b is a multiple of n. Prove that R is reexive, symmetric, and transitive. If aRb for this relation R, we say that a is congruent to b modulo n. Exercise 2.3.9 (Invoked in Section 3.6.) For (a, b), (a, b ) Z Z \ {0}, dene (a, b) (a , b ) if a b = a b. (1) Prove that is an equivalence relation. (Possibly mimic Example 2.3.8.) (2) What are the equivalence class of (0, 1), (1, 1), (2, 3)? (3) Find a natural identication between the equivalence classes and elements of Q.
2.4
Functions
Here is the familiar denition of functions:
Denition 2.4.1 Let A and B be sets. A function from A to B is a rule that assigns to each element of A a unique element of B . We express this with f : A B is a function. The set A is the domain of f and B is the codomain of f . The range of f is Range(f ) = {b B : b = f (a) for some a A}. But in the spirit of introducing few new axioms, lets instead dene functions with the concepts we already know. Convince yourself that the two denitions are the same: Denition 2.4.2 Let A and B be sets. A relation f on A and B is a function if for all a A there exists b B such that (a, b) f and if for all (a, b), (a, c) f , b = c. In this case we say that A is the domain of f , B is the codomain of f , and we write f : A B . The range of f is Range(f ) = {b B : there exists a A such that (a, b) f }. Note that this second formulation is also familiar: it gives us all elements of the graph of the function: b = f (a) if and only if (a, b) is on the graph of f . We will freely change between notations f (a) = b and (a, b) f .
49
One should be aware that if f is a function, then f (x) is an element of the range and is not by itself a function. (But often we speak loosely of f (x) being a function.) Examples 2.4.3 (1) A function can be given with a formula. For example, let f : R R be given by f (x) = sin x. The range is [1, 1]. (2) f (x) = x is NOT a function! (But f (x) = x and f (x) = x are two functions.) (3) Let f : N+ R be given by the description that f (n) equals the nth prime. By Euclids theorem (proved on page 25) there are innitely many primes so that f is indeed dened for all positive integers. We know that f (1) = 2, f (2) = 3, f (3) = 5, and with computers help I get that f (100) = 541, f (500) = 3571. (4) For any set A, the identity function idA : A A takes each x to itself. (5) The function f : A R given by f (x) = 1 for all x is not the identity function (even if A = {1}, because R = {1}). (6) A function can be given by its graph:
From this particular graph we surmise that f (0) = 0, but with the precision of the drawing and our eyesight it might be the case that f (0) = 0.000000000004. Without any further labels on the axes we cannot estimate the numerical values of f at other points. If numerical values are important, the graph should be lled in with more information. (7) A function may be presented by a table: x 1 2 3 f ( x) 1 1 2
Remark 2.4.4 Two functions are the same if they have the same domains, the same codomains, and if to each element of the domain they assign the same element of the codomain. For example, f : R R and g : R [0, ) given by f (x) = x2 = g (x) are not the same! On the other hand, the functions f, g : R R given by f (x) = |x| and g (x) = x2 are the same.
50
Notation 2.4.5 It is common to not specify the domain, in which case the domain is implicitly the largest possible subset of C on which the function is dened. For example, 1 the domain of f (x) = x is the set of all non-zero complex numbers, and the domain of f (x) = x is the set of all non-negative real numbers. Denition 2.4.6 If f : A B and g : B C , then the composition g composed with f is a function g f : A C such that for all a A, (g f )(a) = g (f (a)). If f (x) = (x + 1)2 and g (x) = x3 1, then (g f )(x) = g (f (x) = g ((x + 1)2 ) = ((x + 1)2 )3 1 = (x + 1)6 1, g (x)f (x) = (x3 1)(x + 1)2 , (f g )(x) = f (g (x) = f (x3 1) = (x3 1 + 1)2 = x6 ,
and these three outcomes are distinct. For example, when plugging in x = 1, we get values (g f )(1) = 63, (f g )(1) = 1, g (1)f (1) = 0. Remark 2.4.7 It is common to write f 2 = f f , f 3 = f 2 f = f f f , etc. Some exceptions to this notation are established in trigonometry: sin2 (x) stands for (sin x)2 and not for sin(sin x). Example 2.4.8 Let f (x) = x, and g (x) = x2 . The domain of f is R0 , and the domain of g is C. It is possible to compose g f : R0 C to obtain (g f )(x) = x, but we cannot compose f and g in the other order. If instead h : R R is dened by h(x) = x2 , then g f = h f , and f h : R R is the function (f h)(x) = |x|. We just demonstrated that in composing two functions, the order matters. (Just like rst putting on socks and then shoes gives a dierent outcome from rst putting on shoes and then the socks.) Denition 2.4.9 A function f : A B is injective (or one-to-one) if for all a, a A, f (a) = f (a ) implies a = a . In other words, f is injective if whenever two things map via f to one object, then those two things are actually the same. A function f : A B is surjective (or onto) if for all b B , there exists a A such that f (a) = b. In other words, f is surjective if every element of the codomain is mapped onto via f by some element of the domain. A function f : A B is bijective if it is injective and surjective. For example, the identity function is bijective. The function f : R R given by f (x) = x2 is not injective because f (1) = 1 = f (1). The function f : R0 R given by f (x) = x is injective because f (1) = 1 = f (1), but it is not surjective because 1 is
51
not the square root of any non-negative real number. The function f : R0 R0 given by f (x) = x is both injective and surjective, thus bijective. The following are all the possible functions f : {1, 2} {1, 2} in tabular form: x 1 2 f ( x) 1 1 x 1 2 f ( x) 1 2 x 1 2 f ( x) 2 1 x 1 2 f ( x) 2 2
The rst and the last are neither injective nor surjective, but the middle two are bijective. The following are all the possible functions f : {1, 2, 3} {1, 2}: two constant functions, and seven others given below in tabular form: x 1 2 3 f ( x) 1 1 2 x 1 2 3 f ( x) 1 2 1 x 1 2 3 f ( x) 1 2 2 x 1 2 3 f ( x) 2 1 1 x 1 2 3 f ( x) 2 1 2 x 1 2 3 f ( x) 2 2 1 x 1 2 3 f ( x) 1 2 2
In this case no functions are injective, and all non-constant ones are surjective. Theorem 2.4.10 The composition of two injective functions is injective. The composition of two surjective functions is surjective. The composition of two bijective functions is bijective. Proof. Let f : A B and g : B C . Suppose that f and g are injective. If (g f )(a) = (g f )(a ), then g (f (a)) = g (f (a )). Since g is injective, it follows that f (a) = f (a ), and since f is injective, it follows that a = a . Thus g f is injective. Suppose that f and g are surjective. Let c C . Since g is surjective, there exists b B such that g (b) = c. Since f is surjective, there exists a A such that f (a) = b. Thus (g f )(a) = g (f (a)) = g (b) = c, so that g f is surjective. The last part follows immediately. Denition 2.4.11 For a subset S R, a function f : S R is a polynomial function if there exist a non-negative integer n and c0 , c1 , . . . , cn R such that for all x S , f (x) = c0 + c1 x + c2 x2 + + cn xn . A function f : S R is a rational function if there exist polynomial functions f1 , f2 : S R such that for all x S , f2 (x) = 0 and f (x) = f1 (x)/f2 (x). Similarly there are polynomial and rational functions if all occurrences of R above are replaced by Q or C . Polynomial and rational functions are very special, and are a workhorse of analysis. (The reader has of course encountered trigonometric, exponential, and logarithmic functions, which are not polynomial or rational.) The following are some special properties:
52
Theorem 2.4.12 If a polynomial function is not constant zero, it has only nitely many roots. The domain of a rational function is the complement of a nite subset of C. Proof. Suppose that f (x) = c0 + c1 x + c2 x2 + + cn xn is not constant 0. Thus at least one coecient ci is non-zero. By possibly renaming we may assume that cn = 0. By the non-constant assumption, n 1. If n = 1, then f has only one root, namely c0 /c1 . Suppose that n 2.* Let a be any root of f . We can perform polynomial division f (x) divided by x a to get f (x) = q (x)(x a) + r (x), where q (x) and r (x) are polynomial functions, with r (x) a constant one. But if we plug x = a into both sides we get that r (x) is the constant zero function, so that f (x) = q (x)(x a). Necessarily the degree of q is strictly smaller than the degree of f . By induction on the degree, q has only nitely many roots, say a1 , a2 , . . . , ak . If b is a root of f , then 0 = f (b) = q (b)(b a), so that either q (b) = 0 of b a = 0. Thus b is either a root of q or b = a. Thus the roots of f are a, a1 , . . . , ak . This proves the rst part. A rational function is a quotient of two polynomial functions, and the rational function is dened everywhere except where the denominator is 0. By the rst part this excludes only nitely many numbers. This theorem is useful in that it assures us that the domain of a rational function contains innitely many points, or even better, that the domain is all except nitely many real or complex numbers. (You are aware that the trigonometric functions sine and cosine have innitely many zeroes and that tangent and cotangent are not dened at innitely many real numbers. This fact, together with the previous theorem, is a proof that these four trigonometric functions are not polynomial or rational functions. For similar reasons the logarithmic functions are not polynomial or rational. We have to work harder to prove that the exponential functions are not polynomial or rational.)
1 . There are root If n = 2, then by the quadratic formula the roots of c0 + c1 x + c2 x are 2c2 formulas for cases n = 3, 4, but executing them is very time-consuming, they require familiarity with complex numbers, the solutions involve sums of cube roots with a few square roots thrown in for good measure, and furthermore it can be hard to identify that the ensuing long expression simplies to a nice number such as 2. This, and the existence of computer capabilities, are the reasons that we do not teach such formulas. The formula for solutions of cubic polynomials was discovered by Niccol` o Fontana Tartaglia (15001557) and for quartic ones by Lodovico Ferrari (15221565). Both formulas were popularized in a book by Gerolamo Cardano (15011576). There are no formulas for general polynomials of degree n 5: not only do you and I do not know such a formula, but Niels Henrik Abel (18021829) and Evariste Galois (18111832) proved that no formulas exist using only radicals, sums, products, quotients.
c1
c2 4c0 c2
Chapter 2: Concepts with which we will do mathematics Exercises for Section 2.4
53
2.4.1 Fix a positive integer n. Dene f : Z {0, 1, 2, 3, . . . , n 1} by f (a) is the remainder after a is divided by n. In number theory instead of f (a) = b one says a mod n is b. We can also dene a similar (but not the same!) function g : Z Zn by g (a) = [a]. Graph this function for n = 2 and n = 3. 2.4.2 Dene the oor function : R R to be the function such that for all x R, x is the largest integer that is less than or equal to x. The range is Z. For example, = 3, 1 = 1, 1.5 = 2. Graph this function.
2.4.3 The ceiling function : R R is the function for which x is the smallest integer that is greater than or equal to x. The range is Z. For example, = 4, 1 = 1, 1.5 = 1. Graph this function. 2.4.4 Find a set A and functions f, g : A A such that f g = g f . 2.4.5 Is the composition of functions associative? That is, does f (g h) = (f g ) h? Prove or give a counterexample. 2.4.6 Let f : R2 R and g : R R2 be given by f (x, y ) = x and g (x) = (x, 0). i) Compute f g and g f . ii) Which among f, g, f g, g f are injective, surjective, bijective? 2.4.7 Let A = {1, 2, 3}, B = {2, 4, 6}, and C = {3, 6, 9}. Mark each of the following as a function A B , B C , or C A. i) f = {(2, 6), (4, 3)}. ii) g = {(9, 1), (6, 3), (3, 2)}. iii) h = {(1, 6), (2, 2)}. 2.4.8 For each of the following functions, state the domains, codomains, and how they are compositions of f, g, h from the previous exercise. i) {(2, 3), (4, 2)}. ii) {(9, 6), (3, 2)}. iii) {(2, 6)}. 2.4.9 Let A be a set with 3 elements and B be a set with 2 elements. i) How many possible functions are there from A to B ? ii) How many possible injective functions are there from A to B ? iii) How many possible surjective functions are there from A to B ? From B to A? iv) How many possible bijective functions are there from A to B ? v) How many possible functions from A to B are injective but not surjective? vi) How many possible functions from A to B are surjective but not injective? vii) How many functions from A to B are neither surjective nor injective?
54
2.4.10 The Pigeonhole Principle states that if n items (such as pigeons) are put into m holes with n > m, then at least one hole has more than one item. Let A and B be sets with only nitely many elements. i) Use the Pigeonhole Principle to demonstrate that if A has more elements than B , then f : A B cannot be injective. ii) Use the Pigeonhole Principle to demonstrate that if B has fewer elements than A, then f : A B cannot be surjective. iii) Use the Pigeonhole Principle to demonstrate that if A and B do not have the same number of elements, then f : A B cannot be bijective. 2.4.11 Let A be a set A with m elements and B a set with n elements. i) How many possible functions are there from A to B ? ii) For which combinations of m, n are there injective functions from A to B ? iii) For m, n as in (ii), how many possible injective functions are there from A to B ? iv) For which combinations of m, n are there surjective functions from A to B ? v) For m, n as in (iv), how many possible surjective functions are there from A to B ? vi) For which combinations of m, n are there bijective functions from A to B ? vii) For m, n as in (vi), how many possible bijective functions are there from A to B ? 2.4.12 Find an injective function f : N+ N+ that is not surjective. Compare with parts (ii) and (iii) of the previous exercise and say whether this contradicts the Pigeonhole principle? 2.4.13 In each part below, nd f : R R with the specied condition. i) f is a bijective function. ii) f is neither injective nor surjective. iii) f is injective, but not surjective. iv) f is surjective, but not injective. 2.4.14 Let f : A B and g : B C be functions. i) Suppose that g f is injective. Prove that f is injective. ii) Suppose that g f is surjective. Prove that g is surjective. iii) Give an example of f, g such that g f is injective but g is not injective. iv) Give an example of f, g such that g f is surjective but f is not surjective. 2.4.15 Find functions f, g : R R such that f is surjective but not injective, g is injective but not surjective, and f g is bijective. 2.4.16 Prove that if f and g are both injective functions, then f g is also injective. 2.4.17 Prove that if f and g are both surjective functions, then f g is also surjective. 2.4.18 Prove that if f and g are both bijective functions, then f g is also bijective.
Chapter 2: Concepts with which we will do mathematics 2.4.19 Suppose f : A B and C A. i) Prove f (A) \ f (C ) f (A \ C ). ii) With proof, what condition will make f (A) \ f (C ) = f (A \ C )?
55
2.4.20 For the following polynomial and rational functions, determine the domains (largest sets on which the function is dened): i) f (x) = x2 x31 . x
2 x 1 . ii) f (x) = x x6 +2 1 (iii) * f (x) = x5 17x2 +x4 . (Gottcha! We dont know the roots of this denominator.)
4 3
2.4.21 Prove that f : R R given by f (x) = x3 1 is surjective and injective. Prove is not surjective and not injective. that g : R R given by g (x) = x31 1
2.5
Binary operations on sets
Denition 2.5.1 A binary operation on a set A is a function : A A A. Binary operations are usually denoted by special symbols such as: +, , , /, , , , , or, and. An arbitrary element of A A is an ordered pair (a, b) for a, b A, and when we plug that element into the function , we should perhaps write ((a, b)), but normally one set of parentheses is removed, so we write (a, b). But depending on the exact form of , we often traditionally write a b rather than (a, b) (see examples below). Examples and non-examples 2.5.2 (1) +, on N, Z, Q, R, C (we of course write a + b rather than +(a, b) and a b or even ab rather than (a, b)); (2) on Z, Q, R, C; (3) / on Q \ {0}, R \ {0}, C \ {0}; (4) is not a binary operation on N. (5) Let S be a set, and let A be the set of all functions f : S S . Then composition (of functions) is a binary operation on A. (6) , , \ are binary operations on the set of all subsets of a set S . Denition 2.5.3 Let be a binary operation on a set A. An element e A is an identity element for if for all a A, a e = a = e a. (In this chapter we use e for the identity element; this is a symbol that is unrelated to the base of the exponential function as in Denition 7.4.6.)
56
Section 2.5: Binary operations on sets
Examples 2.5.4 An identity (or is it the identity?) for + on Z, Q, N0 , R is 0. An identity (or is it the identity?) for multiplication on Z, Q, N0 , R is the number 1. An identity (or is it the identity?) for composition of functions is the identity function id (the function with id(x) = x for all x). Subtraction on Z does not have an identity element because if e were an identity element, then 1 e = 1 = e 1, which says that e = 0 and e = 2, which is nonsense. If S is a set and A is the collection of all subsets of S , then S is the identity element for and is the identity element for . The theorem below resolves the problem of an identity versus the identity. Theorem 2.5.5 Let be a binary operation on A. Suppose that e and f are both identities for . Then e = f . In other words, if an identity exists for a binary operation, it is unique. Hence we talk about the identity for . Proof. Since for all a A, e a = a, we get in particular that e f = f . Also, for every a A, a f = a, hence e f = e. Thus e = e f = f . Note: we used symmetry and transitivity of the equality relation. Denition 2.5.6 Let be a binary operation on A and suppose that e is its identity. Let x be an element of A. An inverse of x is an element y A such that x y = e = y x. To emphasize what the operation is, we may also say that y is a -inverse of x (or see specic terms below). Examples and non-examples 2.5.7 (1) Let = + on Z. Then 0 is the identity element and every element has an additive inverse. (2) Let = on Q \ {0}. Then 1 is the identity element and every element has a multiplicative inverse. (3) If S is a set and A is the collection of all subsets of S , then only S has an inverse for , the inverse being S itself, and only has an inverse for , the inverse being . Denition 2.5.8 A binary operation on A is associative if for all a, b, c A, a (b c) = ( a b ) c. Examples and non-examples 2.5.9 (1) + and are associative. (2) , / are not associative. (3) Composition of functions is an associative operation. (4) , are associative. Theorem 2.5.10 Let be an associative binary operation on A with identity e. If x has an inverse, that inverse is unique.
Chapter 2: Concepts with which we will do mathematics Proof. Let y and z be inverses of x. Then y = y e (by property of identity)
57
= y (x z ) (since z is an inverse of x) = (y x) z (since is associative)
= e z (since y is an inverse of x) = z (by property of identity). Thus by the transitivity of equality, y = z .
Denition 2.5.11 We say that x is invertible if x has an inverse. The (abstract) inverse is usually denoted x1 . Be careful! What is the number 51 if equals +? Theorem 2.5.12 If x is invertible, then its inverse is also invertible, and the inverse of the inverse is x. Proof. By denition of inverses of x, x1 x = e = x x1 , which also reads as the inverse of x1 is x. Theorem 2.5.13 Cancellation. Let be an associative binary operation on a set A, let e be the identity and z an invertible element in A. Then for all x, y A, x z = y z x = y, z x = z y x = y. Proof. We prove only the rst implication. If x z = y z , then (x z ) z 1 = (y z ) z 1 , hence by associativity, x (z z 1 ) = y (z z 1 ). Thus by the denition of inverses and identities, x = x e = y e = y . Another proof of the same fact is as follows: x=xe = x ( z z 1 ) = ( y z ) z 1 =ye = y.
= (x z ) z 1 (by associativity) = y (z z 1 ) (by associativity)
58
Section 2.5: Binary operations on sets
If is associative, we will in the future omit parentheses in a b c (or in a b c d et cetera), as the order of the computation does not matter. If is not associative, you need to keep parentheses! For example, in Z, a b c d can have parentheses inserted in many dierent ways, and ve dierent values can be obtained! You may want to nd specic four integers a, b, c, d for which you get 5 distinct values with dierent placements of parentheses. Denition 2.5.14 A binary operation on A is commutative if for all a, b A, a b = b a. Examples and non-examples 2.5.15 (1) +, are commutative on all examples we have seen so far. (If you have seen matrices, you know that matrix multiplication is not commutative.) (2) , are commutative. (3) Function composition is not commutative (cf. Example 2.4.8). Notation 2.5.16 Just like for functions in Remark 2.4.7, also for arbitrary associative binary operation we abbreviate a a with a2 , (a a) a with a3 , et cetera, and in general for all positive integers n we write an = an1 a = a an1 . This notation is familiar also when equals multiplication: then 25 stands for 32 (but when is addition, then the abstract 25 stands for 10, but of course we prefer to not write it this way but write is in the additive notation 2 + 2 + 2 + 2 + 2, or briey, 5 2). The empty product a0 makes sense if the set has the identity, and in that case a0 = e. (See Exercise 1.4.5 for the rst occurrence of an empty product.) If a has a multiplicative inverse, then a1 is that inverse, and in that case if is in addition associative, an = a(n1) a1 . To prove this by induction on n, we multiply (a(n1) a1 ) an = (a(n1) a1 ) (a an1 ) = a(n1) an1 = e, = a(n1) (a1 a) an1
and similarly an (a(n1) a1 ) = e, which proves that an = a(n1) a1 . Notation 2.5.17 More generally, if is associative, we may and do omit parentheses in expressions such as a1 a2 an , as the meaning is unambiguous. If is in addition commutative, the order of the ak may be changed as well. When is commutative multipli-

n
59
cation, we abbreviate this to k=1 ak , and when is commutative addition, we abbreviate n this to k=1 ak . Examples were already worked out in Section 1.4. Example 2.5.18 Let S be a set, and let A be the set of all bijective functions f : S S . Then A contains the identity function e. The obvious binary operation on A is composition : by Theorem 2.4.10, if f, g are bijective, so is g f . By Exercise 2.4.5, composition is an associative operation. By the bijective assumption, every f A has an inverse f 1 , and by Theorem 2.5.10, this inverse is unique. Furthermore, (g f ) f 1 g 1 = g (f f 1 ) g 1 = g e g 1 = e, and similarly f 1 g 1 (g f ) = e, so that (g f )1 = f 1 g 1 . We end this section with an important example. The reader is familiar with manipulations below when n = 12 or n = 24 for hours of the day (we do not say 28 oclock), when n = 7 for days of the week, when n = 3 for standard meals of the day, et cetera. Important Example 2.5.19 Let n be a positive integer. Dene + on Zn as follows: [a] + [b] = [a + b]. Well, rst of all, is this even a function? Namely, we need to verify that whenever [a] = [a ] and [b] = [b ], then [a + b] = [a + b ], which says that any choice of representatives gives the same nal answer. Well, a a and b b are integer multiples of n, hence (a + b) (a + b ) = (a a ) + (b b ) is a sum of two multiples of n and hence also a multiple of n. Thus [a + b] = [a + b ], which says that + is indeed a binary operation on Zn . It is straightforward to verify that + on Zn is commutative and associative, the identity elements is [0], and every element [a] has an additive inverse [a] = [n a]. Exercises for Section 2.5 2.5.1 Let A be the set of all logical statements. i) Verify that , and xor are binary operations on A. ii) Find the identity elements, if they exist, for each of the three binary operations. iii) For each binary operation with identity, which elements of A have inverses? 2.5.2 Let A be the set of all bijective functions f : {1, 2} {1, 2}. i) How many elements are in A? ii) Prove that function composition is a binary operation on A. iii) Write a multiplication table for . iv) What is the identity element? v) Verify that every element of A is its own inverse. vi) Verify that is commutative.
60
Section 2.6: Fields
2.5.4 Find a set A with a binary operation such that for some invertible f, g A, (g f )1 = g 1 f 1 . (This is sometimes called the socks-and-shoes problem. Can you think of why?) 2.5.5 Repeat the previous exercise for the set of all bijective functions f : {1, 2, 3, . . . , n} {1, 2, 3, . . . , n}. 2.5.6 Refer to Example 2.5.19: Write addition tables for Z2 , Z3 , Z4 . 2.5.7 Dene multiplication on Zn in a natural way. Verify that it is well-dened, that it is commutative, associative, and that its identity is [1]. 2.5.8 Write multiplication tables for Z5 , Z6 , Z7 . 2.5.9 Determine all [a] Z12 that have a multiplicative inverse. Determine all [a] Zp that have a multiplicative inverse if p is a prime number. 2.5.10 Here is an opportunity to practice induction. Let f, g : A A be functions, with f invertible. Prove that (f g f 1 )n = f g n f 1 . (Notation is from Remark 2.4.7.)
2.5.3 Let A be the set of all bijective functions f : {1, 2, 3} {1, 2, 3}. i) How many elements are in A? ii) Prove that function composition is a binary operation on A. iii) Find f, g A such that f g = g f .
2.6
Fields
Denition 2.6.1 A non-empty set F is a eld if it has two binary operations on it, typically denoted + and , and special elements 0, 1 F , such that the following identities hold for all m, n, p F : (1) (Additive identity) m + 0 = m = 0 + m. (2) m 0 = 0 = 0 m. (3) (Associativity of addition) m + (n + p) = (m + n) + p. (4) (Commutativity of addition) m + n = n + m. (5) (Multiplicative identity) m 1 = m = 1 m. (6) (Distributivity) m (n + p) = (m n) + (m p). (7) (Associativity of multiplication) m (n p) = (m n) p. (8) (Commutativity of multiplication) m n = n m. (9) 1 = 0. (10) (Existence of additive inverses) There exists r Q such that m + r = r + m = (0, 1) (i.e., m + r = r + m = 0. (11) (Existence of multiplicative inverses) If m = 0, there exists r Q such that m r = r m = 1. 0 is called the additive identity and 1 is called the multiplicative identity.
61
It is standard to omit when no confusion arises. Note that 2 222 + 4 is dierent from 2 222 + 4, but we treat 2 x + 4 the same as 2x + 4. With the usual meaning of +, , the familiar Q, R, C are elds, but N0 , N and Z are not. In the next chapter we construct these elds, and also the non-elds N0 , Z, with all the arithmetic on them. There are many other elds out there, such as the set of all real-valued rational functions with domains subsets of R (each with its own maximal domain). By Section 2.5, we know that the additive and multiplicative identities and inverses are unique in a eld. The additive inverse of m is denoted m, and the multiplicative inverse of a non-zero m is denoted m1 , or also 1/m. The product of n and m1 is also written as n/m. The latter we think of as division of n by m. Note that we cannot divide by 0. Never divide by 0. For one thing, in an abstract eld, dividing by 0 is simply not on the list of allowed operations, and for another, n/m always stands for that unique element of the eld, which, when multiplied by m, yields n. In other words, n = (n/m) m. If m somehow horribly happened to be 0, then we would have n = (n/0) 0, and by part (2) of the denition of elds, n = 0. So, if we were to divide by 0, we could at most divide 0 by 0. That is not much of a division! Furthermore continuing the horrible detour then 0/0 = (0 0)/0 = 0 (0/0) = 0 by (2), but 0/0 should perhaps also be 1? In any case, division by 0 gets us into a muddle, and it is best for mathematics and health if you avoid it. At this stage of your mathematical life, you will of course never write something like 3/0 (oh, my eyes hurt seeing this!), but a common college mistake that is essentially division by 0 is when one is solving an equation such as x2 = 3x, cancels x, and obtains the solution x = 3. However, that cancellation was division by 0 when x was the other possible solution! Please avoid also this hidden division by 0 as you can, just like in this case, fail to nd all solutions. Exercises for Section 2.6 2.6.1 Use the set-up in Example 2.3.7. Prove that Z2 is a eld. Note that in this eld [2] = [0], so [2] does not have a multiplicative inverse. Also note that in this eld, every number has a square and cube root. 2.6.2 Use the set-up in Example 2.3.7. Prove that Z3 is a eld. Note that in this eld [3] = [0], so [3] does not have a multiplicative inverse. Note that in this eld, [2] is not the square of any number. 2.6.3 Use the set-up in Example 2.3.7. Prove that Zn is not a eld if n is not a prime integer. 2.6.4 Use the set-up in Example 2.3.7. Prove that Zp is a eld for any prime integer p. Note that in Z7 , [2] is the square of [3] and [4].
62
Section 2.7: Order on sets, ordered elds, absolute values
2.6.5 Let F be a eld. Prove that for any x F , (x) = x. Prove that for any non-zero x F , 1/(1/x) = x.
2.6.6 Let F be a eld. Prove that for any x, y F , (x) y = (x y ) = x (y ). (Hint: use the denition of additive inverses.) 2.6.7 Let F be a eld. Prove that for any x, y F , (x) (y ) = x y .
2.7
Order on sets, ordered elds, absolute values

If < is a relation on a set S , we can dene relations , >, on on S by a > b if b < a. a b if a < b or a = b.
a b if a < b. Conversely, if is a relation on S , then we can dene < on S by a < b if a b and a = b, which by before also denes >, . Similarly, each of >, determines all four relations of this form. Thus one of these relations on a set S implies that we have all four relations naturally derived from the one. These relations impose the usual order on a set. We use these relations in the usual sense on R, but we can also use < in other contexts: Examples 2.7.1 (1) < can be the relation is strictly older than on the set of all people, in which case is is older than or of equal age, and >, have the obvious meanings. (2) < can be the relation is a proper subset of on a set S of all subsets of some universal set U . In this case, means is a subset of, > means properly contains, and means contains. Denition 2.7.2 Let be a relation on a set S . A subset T of S is bounded above (resp. bounded below)* (in S ) if there exists b S such that for all t T , t b (resp. b t). The element b is called an upper bound (resp. lower bound) of T (in S ). A set that is bounded above and below is called bounded. An upper bound c S of T is a least upper bound, or supremum, if for all upper bounds b of T , c b. If c T , then c is also called the maximum of T .
A sentence of the form P is Q (resp. Q ) if R (resp. R ) is a shorthand for two sentences: P is Q if R and P is Q if R .
63
A lower bound c S of T is a greatest lower bound, or inmum, if for all lower bounds b of T , b c. If c T , then c is also called the minimum of T . The obvious standard abbreviations are: lub (T ) = sup (T ), glb (T ) = inf (T ), max (T ), min (T ), possibly without parentheses around T . Examples 2.7.3 (1) The set N in R has minimum 0 and it is not bounded above. (2) The set {1/n : n N} has maximum 1, it is bounded below, the inmum is 0, and there is no minimum. (3) The set {sin(x) : x R} has maximum 1 and minimum 1. (4) The set {ex : x R} has no upper bound, it is bounded below with inmum 0, but there is no minimum. (5) The empty subset obviously has no minima nor maxima. Every element of S is vacuously an upper and a lower bound of the empty set. In the sequel we restrict < to relations that satisfy the trichotomy property: Denition 2.7.4 A relation < on a set S satises trichotomy if for all s, t S , exactly one of the following relations holds: s = t, s < t, or t < s. Theorem 2.7.5 Let < on S satisfy trichotomy. Then a supremum (resp. inmum) of a non-empty subset T of S , if it exists, is unique. Proof. Suppose that c, c are suprema of T in S . Both c and c are upper bounds on T , and since c is a least upper bound, necessarily c c. Similarly c c , so that by assumption c = c . This proves that suprema are unique, and a similar proof shows that inma are unique. Why did we assume that T be non-empty? By denition every element of S is an upper bound for , so in particular if S has no minimum, then has no least upper bound. Denition 2.7.6 A set S with relation is well-ordered if inf (T ) = min (T ) for every non-empty subset T of S that is bounded below. The element min (T ) is called the least element of T . Examples 2.7.7 (1) Any nite set with relation is well-ordered (simply check the nitely many pairings for which element is smaller). (2) Z is not well-ordered as there is no smallest whole number. (3) The set of all positive rational numbers is not well-ordered as it itself does not have a minimum.
64
(4) N0 is well-ordered because for any non-empty subset T of N0 , by the fact that the set is not emtpy there exists an element n T , and after that one has to check if any of the nitely many numbers 0, 1 through n which of them is the smallest one in T . (5) N+ is well-ordered. Denition 2.7.8 Let F be a set with binary operations +, , with additive identity 0 F and multiplicative identity 1 F , and with a relation < satisfying trichotomy. Dene F + = {x F : 0 < x}, and F = {x F : x < 0}. Elements of F + are called positive, and element of F are called negative. Elements of F \ F + are called non-positive and element of F \ F are called non-negative. We dene intervals in F to be sets of the following form, where a, b F with a < b: (a, b] = {x F : a < x b}, (a, b) = {x F : a < x < b},
[a, ) = {x F : a x}, (, b] = {x F : x b}. (, b) = {x F : x < b},
(a, ) = {x F : a < x},
[a, b] = {x F : a x b},
[a, b) = {x F : a x < b},
We say that F is ordered if in addition the following properties hold for all x, y, z F : (1) (Transitivity of <) If x < y and y < z then x < z . (2) (Compatibility of < with addition) If x < y then x + z < y + z . (3) (Compatibility of < with multiplication by positive elements) If x < y and 0 < z then xz < yz . If F is a eld, it is called an ordered eld. Theorem 2.7.9 Let F be an ordered set as in Denition 2.7.8. (1) 1 F + . (2) For x F , x F + if and only if x F , and x F if and only if x F + . Proof. (1) We know that 1 = 0. If 1 F + , then by trichotomy 1 < 0. Then by compatibility of < with addition, 0 = 1 + (1) < 0 + (1) = 1, so that by compatibility of < with multiplication by positive numbers, 0 = 0 (1) < (1) (1) = (1) = 1, which says that 1 < 0 and 0 < 1, which contradicts trichotomy. So 1 F + . (2) x F + if and only if 0 < x, and by compatibility of < with addition this implies that x = 0 x < x x = 0, so that x F . The rest of (2) is equally easy.
Chapter 2: Concepts with which we will do mathematics Denition 2.7.10 Let F be an ordered set. a function dened as 0; | x| = x; x;
65
The absolute value function | | : F F is if x = 0; if x F + ; if x F .
The following theorem lists the familiar properties of absolute values, and the reader may wish to prove it without reading the given proof. Theorem 2.7.11 Let F be an ordered set. (1) For all x F , x 0 if and only if x = |x|; and x 0 if and only if x = |x|. (In particular, |1| = 1.) (2) For all x F , |x| x |x| = | x|. (3) For all x, a F , |x| a if and only if a x a. (4) For all x, a F , |x| < a if and only if a < x < a. (5) For all x, y F , |xy | = |x||y |. Proof. (1) If x 0, then |x| = x, so the conclusion holds, and if x < 0, then x F , so that x F + and 0 < |x| = x = | x|. Thus by transitivity of <, x < |x|, whence x |x|. Thus in particular x | x| = |x|, so that |x| x. This also proves (2). The parenthetical remark in (1) follows from Theorem 2.7.9. (3) Suppose that |x| a. Since 0 |x|, by transitivity also 0 a. If x > a, then also x > 0, so that |x| = x > a, which contradicts the assumption |x| a and the trichotomy. Thus x a. If x < a, then by compatibility of < with addition, a < x, so that x F + and a < x = | x| = |x|, which contradicts trichotomy. So necessarily x a, and from before we have that x a, which gives a x a. The other direction and (4) are proved similarly. (5) If either x or y is 0, then xy = 0, so that |xy | = 0, and either |x| or |y | is 0, so that |x||y | = 0. Thus (5) is true if either x or y is 0. So we may assume that neither is 0. If 0 < x, y , then by property (4) of <, 0 = 0 y < x y , and then |xy | = xy = |x||y |. If x, y < 0, then 0 < x, y , so that |xy | = |(x)(y )| = (x)(y ) = | x|| y | = |x||y |. If x < 0 and 0 < y , then by property (4) of <, xy < 0, so that |xy | = (xy ) = (x)y = | x||y | = |x||y |. The remaining proof in case 0 < x and y < 0 is very similar. This exhausts all the cases and proves (5). Theorem 2.7.12 (Triangle and reverse triangle inequalities) Let F be an ordered set. (1) Triangle inequality: For all x, y F , |x y | |x| + |y |. (2) Reverse triangle inequality: For all x, y F , |x y | ||x| |y || = ||y | |x||. Proof. (1) By the rst part of the previous theorem, (|x| + |y |) x + y |x| + |y |, so that by the second part of the previous theorem, |x + y | |x| + |y |. Thus |x y | = |x + (y )| |x| + | y |, and by the second part again this is equal to |x| + |y |.
66
(2) By (1), |x| = |x y y | |x y | + |y |, so that |x| |y | |x y |. Similarly, |y | |x| |y x|. But by the second part of the previous theorem, |y x| = | (y x)| = |x y | and |y + x| = |x + y |, so that |x y | |x| |y | and |x y | |y | |x| = (|x| |y |). Since ||x| |y || is either |x| |y | or |y | |x|, (2) follows. Theorem 2.7.13 Let F be an ordered eld. Let r F . (1) If r F + , then r 1 F + . (2) If r < for all F + , then r 0. (3) If r > for all F + , then r 0. (4) If |r | < for all F + , then r = 0. Proof. Suppose that r F + . Then r = 0, so that r 1 F . If r 1 = 0, then 1 = r 1 r = 0 r = 0, which contradicts the axioms of a eld. If r 1 F , then by Theorem 2.7.9, r 1 F + , so that then 1 = r 1 r F + by compatibility of < with multiplication by positive numbers, which is impossible by Theorem 2.7.9. Thus r 1 F + . This proves (1). Proof of (2): By Theorem 2.7.9, 1 F + , so that 0 < 1, and by compatibility of < with addition, 1 < 2. Thus by transitivity of <, 0 < 2, so that 2 is positive, and by (1), 21 F + . Since r > 0, by compatibility of < with multiplication by positive numbers, = r/2 is a positive number. By assumption, r < = r/2. Again by compatibility of < with multiplication, by multiplying through by 2r 1 , we get that 2 < 1, which contradicts the trichotomy (since we already established that 1 < 2). Thus r F + , so that r 0. The proof of (3) is similar. For (4), if |r | < for all F + , then < r < . Then by (2) and (3), 0 r 0. Since F + F = by trichotomy, it follows that r = 0. Exercises for Section 2.7 2.7.2 Prove that < (resp. ) is transitive if and only if > (resp. ) is transitive. 2.7.1 Prove that if < is transitive then is transitive.
2.7.3 In computing limits (see Section 4.1), especially for nding , one in general needs to work with minima of nite sets. For example, under what conditions is the minimum of {0.2, /7} equal to 0.2, and when is the minimum /7? Similarly determine min {0.2, /7, 2/4}. 2.7.4 Does the set {2n /n : n N+ } have a lower (resp. upper bound)? Is it well-ordered? Justify. Repeat with {n/2n : n N+ }.
2.7.5 Suppose that a subset T of an ordered eld has a minimum (resp., maximum, inmum, supremum) b. Prove that the set T = {t : t T } has a maximum (resp., minimum, supremum, inmum) b.
67
2.7.6 Suppose that a subset T of positive elements of an ordered eld has a minimum (resp., maximum, inmum, supremum) b. What can you say about the maximum (resp., minimum, supremum, inmum) of the set {1/t : t T }? 2.7.7 For each of the sets below, determine its minimum, maximum, inmum, supremum, if applicable. Justify all anwers. i) {1, 2, , 7}. ii) {(1)n /n : n N}. iii) The set of all positive prime numbers. iv) {x R : 1 < x < 5}. v) {x R : 2 x < 5}. vi) {x R : x2 < 2}. vii) {x Q : x2 < 2}. viii) {x R : x2 + x 1 = 0}. ix) {x Q : x2 + x 1 = 0}.
2.7.10 Let F be an ordered eld and let x be non-zero element of F . Prove that x F + if and only if x1 F + . 2.7.11 Let F be an ordered eld. Prove that 2, 3 are positive (and so not zero). 2.7.12 Let F be an ordered set and x F . Prove that x2 = 0 if and only if x = 0.
2.7.9 Let F be an ordered eld. Prove that for all x, y F , x + y F and x y F + . (Hint: Use that for all x, y F + , x + y, x y F + .)
Exercise 2.7.8 (Invoked in Theorem 7.2.6.) Let S and T be subsets of an ordered eld F . Let S + T = {s + t : s S, t T }. i) If S and T are bounded above, prove that sup (S + T ) sup S + sup T . ii) If S and T are bounded below, prove that inf (S + T ) inf S + inf T .
2.7.13 Let F be an ordered set and x F . Let x, y F be non-negative (resp. nonpositive) such that x + y = 0. Prove that x = y = 0. 2.7.14 Prove that Zp is not an ordered eld for any prime integer p. (Cf. Exercise 2.6.4.) 2.7.15 Let F be an ordered set. Suppose that x y and p q . Prove that x + p y + q . If in addition x < y or p < q , prove that x + p < y + q . 2.7.16 Let F be an ordered set and x, y F . Prove that x < y if and only if 0 < y x. Prove that x y if and only if 0 y x. 2.7.17 Let F be an ordered eld. i) Prove that if 0 < x < y , then 1/y < 1/x. ii) Suppose that x < y and that x, y are non-zero. Does it follow that 1/y < 1/x? Prove or give a counterexample.
68
Section 2.8: Increasing and decreasing functions
Exercise 2.7.18 (In-betweenness in an ordered set) Let F be an ordered set. Let x, y F with x < y . Prove that if 2 is a unit in F , then x < (x + y )/2 < y . 2.7.19 Let F be an ordered set. Prove that any non-empty nite subset S of F has a maximum and a minimum (cf. Denition 2.7.2). These are denoted max (S ) and min (S ), respectively. 2.7.20 Let F be an ordered set and S a nite subset S . Prove that for all s S , min (S ) s max (S ).
2.8
Increasing and decreasing functions
Denition 2.8.1 Let F, G be ordered sets (as in the previous section), and let A F . A function f : F G is increasing (resp. decreasing) on A if for all x, y A, x < y implies that f (x) f (y ) (resp. f (x) f (y )). If furthermore f (x) < f (y ) (resp. f (x) > f (y ) for all x < y , then we say that f is strictly increasing (resp. strictly decreasing) on A. A function is (strictly) monotone if it is (strictly) increasing or decreasing. Theorem 2.8.2 Let n be a positive integer and F an ordered set. Then the function f : F F dened by f (x) = xn is strictly increasing on F + {0}. Proof. Let x, y F + with x < y . Then by Exercise 1.6.6, f (y ) f (x) = y x = (x + (y x)) x =
n n n n n1 k=0
n k x (y x)nk . k
But x, y x are positive, and n k is a non-negative integer, so that f (y ) f (x) is the sum of products of non-negative elements, so that f (y ) f (x) 0. Furthermore, one 0 n0 of the non-negative summands is n = (y x)n , which is positive, so that 0 x ( y x) f (y ) f (x) > 0. Corollary 2.8.3 Let n be a positive integer and F an ordered set. Suppose that x, y F + {0} have the property that xn < y n . Then x < y . (In other words, the radical function is strictly increasing on F + {0}.) Proof. If x = y , then xn = y n , which contradicts the assumption and trichotomy. If x > y , then by Theorem 2.8.2, xn > y n , which also contradicts the assumption. So by trichotomy x < y. Theorem 2.8.4 If F, G are ordered sets and f : F G is strictly increasing (resp. decreasing), then there exists a strictly increasing (resp. decreasing) function g : Range(f ) F such that for all x F , (g f )(x) = x and for all y G, (f g )(y ) = y . In other words, g is the inverse of the function f : F Range f .
69
Proof. Let y Range f . Then y = f (x) for some x F . Since f is injective, x is unique. Thus we dene g : Range f F by g (y ) = x. Then by denition for all x F , g (f (x)) = x and for all y Range f , f (g (y )) = y . It is clear that f is increasing if and only if g is increasing. Exercises for Section 2.8 2.8.1 Let n be an odd positive integer and F an ordered set. Prove that the function f : F F dened by f (x) = xn is strictly increasing.
2.8.3 Let n be an odd positive integer and F an ordered set. Suppose that a, b F and that an < bn . Prove that a < b.
2.8.2 Let n be an even positive integer and F an ordered set. Prove that the function f : F F dened by f (x) = xn is strictly decreasing on F {0}.
2.8.4 Let F be an ordered set, a F + and f : F F dened by f (x) = ax. Prove that f is a strictly increasing function. 2.8.5 Let F be an ordered set, a F and f : F F dened by f (x) = ax. Prove that f is a strictly decreasing function. 2.8.6 Let F be an ordered set. Prove that the absolute value function on F is increasing on F + {0} and decreasing on F {0}.
2.8.7 Prove that the composition of (strictly) increasing functions is (strictly) increasing. Prove that the composition of (strictly) decreasing functions is (strictly) decreasing. 2.8.8 Prove that the composition of a (strictly) increasing function wth a (strictly) decreasing function, in any order, is (strictly) decreasing. 2.8.9 Suppose that f : F B is strictly increasing on a set A. Prove that the function g : A B given by g (x) = f (x) is injective, and that the function h : A Range(g ) is bijective.
Chapter 3: Construction of the basic number systems
In Chapter 1 we laid the basic groundwork for how we do mathematics: how we reason logically, how we prove further facts from established truths, and we learned some notation. In Chapter 2 we introduced sets, and from those derived functions and binary operations that play a big role in mathematics. In this chapter, we use set theory to derive numbers and basic arithmetic: we will do so on N, Z, Q, R, C, in this order. When I teach this course, I go lightly through the rst eight sections of this very long chapter, but I do spend time on Theorems 3.8.3, 3.8.5, 3.8.6. Dedekind cuts give straightforward proofs, but if one wishes to avoid Dedekind cuts, then these theorems can (and should) be simply accepted as facts. Section 3.9 plays a central role in the rest of the book, so this section and Section 3.10 should be studied thoroughly. Sections 3.11 and 3.12 introduce topology, and they contain more information than strictly necessary for the rest of the book.
3.1
Inductive sets and a construction of natural numbers
We start with the most basic set: . We can make the empty set be the unique element of a new set, like so: {}. These two sets and {} are distinct because the latter contains the empty set and the former contains no elements. With these two distinct sets we can form a new set with precisely these two elements: {, {}}. This set certainly diers from the empty set, and it also diers from {} because {} is not an element of {} (but it is a subset think about this). So now we have three distinct sets, and we can form another set from these: {, {}, {, {}}}, giving us distinct sets , {}, {, {}}, {, {}, {, {}}}. We could call these Zeno, Juan, Drew, Tricia, but more familiarly these sets can be called zero, one, two, three, and written 0, 1, 2, 3 for short. At this point, we are simply giving them names, with no assumption on any arithmetic properties. We could say and so on, but that is not very rigorous, is it. Denition 3.1.1 For any set S , its successor S + is dened as S + = S {S }. By our naming convention, 0 is not the successor of any set, 0+ = 1, 1+ = 2, 2+ = 3. We can also write 0++ = (0+ )+ = 1+ = 2 and 0+++ = ((0+ )+ )+ = (1+ )+ = 2+ = 3.
71
At this point I rely on your basic training for the naming conventions of 3+ = 4, 3++ = 5, . . .. The denition of successors of successors of successors... is recursive. We can see informally that the set of all successive successors of 0 will be innite, in fact we hope that this set would be the familiar N0 , but to get to that we would have to take a union of innitely many sets that at this point need not be elements of some universal set. Taking unions of innitely many sets is subject to pitfalls, just like innite sums have pitfalls. Denition 3.1.2 A set J is called inductive if it satises two conditions: (1) J . (2) For any n J , the successor n+ of n is also in J . We accept as a given (we take it as an axiom) that there exists a set that contains and all its successors: Axiom 3.1.3 There exists an inductive set. An inductive set has to contain the familiar 0, 1, 2, 3, . . ., but possibly it can also contain the familiar numbers 1.3 or , which are not part of the familiar (inductive set) N0 . Theorem 3.1.4 Let S be a set that contains inductive subsets. The intersection of any collection of inductive subsets is inductive. Proof. Each inductive set contains , hence their intersection contants as well; for any n in the intersection, that n is in all the inductive sets, hence n+ is also in those same inductive sets, which says that n+ is in the intersection, thus proving that the intersection of inductive sets is inductive. Denition 3.1.5 (Denition of N0 :) Let J be any inductive set. Dene N0 as the intersection of all inductive subsets of J . Elements of N0 are called natural numbers. (But beware, in some books, 0 is not considered a natural number.) We just veried that N0 is an inductive set, and by denition it is a subset of all inductive subsets of J . It appears that N0 depends on the choice of J . However, Exercise 3.1.6 shows that this is not so. Some of the elements of N0 are: , {}, {, {}}, {, {}, {, {}}}, or written alternatively: 0, 1, 2, 3. In fact, we prefer to write elements of N0 numerically, but for proofs we will often need to resort to the notion of successors of sets. Theorem 3.1.6 0 is not the successor of any element of N0 , and every element of N0 other than 0 is the successor of some element of N0 .
72
Section 3.1: Inductive sets and a construction of natural numbers
Proof. 0 stands for the empty set, which by denition cannot be the successor of any set. Let n N0 \ {0}. If n is not the successor of any element in N0 , then N0 \ {n} is an inductive set which is strictly smaller than N0 . But this contradicts the denition of N0 . Thus there is no such n, which proves that every element of N0 \ {0} is the successor of some element of N0 . At least this says that if were to be in N0 , then 1 perhaps would have to be in N0 as well? Recall mathematical induction from Section 1.5. The notion of inductive sets gives us a new look at induction and it will enable us to eventually rigorously dene the (usual) arithmetic. We will use the following new look at induction over and over: Theorem 3.1.7 (Induction Theorem) Let P be a property applicable to (some) elements of N0 . Suppose that (1) P (0) is true; (2) For all n N0 , (P (n) = P (n+ )) is true. Then P (n) is true for all n N0 . Proof. [To use the new concept of inductive sets, we have to define a set that is hopefully inductive.] Let T = {n N0 : P (n) is true}. [We will prove that this T is inductive.] By assumption (1) we have that 0 T . [So the first property of inductive sets holds for T .] If n T , then P (n) is true, and since (P (n) = P (n+ )) is true, it follows that P (n+ ) is true. Thus n+ T . This means that T is an inductive set. But N0 is the smallest inductive set, which means that N0 T , so that P (n) is true for all n N0 . We present one application of this immediately; many more applications are in the exercises and in the subsequent sections analyzing arithmetic and order properties of N0 . Theorem 3.1.8 Every element of N0 is obtained from 0 by applying the function successor of nitely many times. Proof. Let T be the subset of N0 consisting of all elements that are obtained from 0 by applying successor of nitely many times. Certainly 0 is obtained from 0 by applying successor of no times, so that 0 T . Suppose that n T . Then n is obtained from 0 by applying the function successor of nitely many times. It follows that n+ is obtained from 0 by applying the function successor of one more time, whis is still nitely many times. Thus n+ T , which says that T is an inductive set. Hence by Theorem 3.1.7, T = N0 .
Chapter 3: Construction of the basic number systems Exercises for Section 3.1 3.1.2 Let T = {n N0 : n = n+ }. i) Prove that 0 T . ii) Suppose that n T and that n = m+ . Prove that n+ T . iii) Prove that T is an inductive set. iv) Prove that T = N0 . (Hint: Apply Theorem 3.1.7.) 3.1.1 Prove that for every set S , S S + and S S + .
73
3.1.3 Prove that 0 n+ for all n N0 . (Hint: Prove that T = {n N0 : 0 n+ } is inductive.) Exercise 3.1.4 (Invoked in Lemma 3.3.1.) Dene T = {m N0 : whenever n m and s n, then s m}. i) Prove that T (vacuously). ii) Suppose that m T . Prove that m+ T . (Hint: Assume n m+ and s n. Two cases: n m and n = m because m+ = m {m}.) iii) Prove that T = N0 . Conclude that for all m N0 , whenever n m and s n, then s m.
Exercise 3.1.5 (Invoked in Lemma 3.3.1.) Let T be the subset of N0 containing all n for which n n, n+ n, and n+ n. i) Prove that T . ii) Suppose that n T . Prove (by contradiction) that n+ n+ . Prove (by contradiction) that (n+ )+ n+ . Prove (by contradiction) that (n+ )+ n+ . iii) Prove that T is an inductive set. Conclude that T = N0 , i.e., that for every n N0 , n n, n+ n, and n+ n. 3.1.6 (The goal of this exercise is to prove that N0 as dened in Denition 3.1.5 is independent of the ambient inductive set.) Let J and K be inductive sets. Let N be the intersection of all inductive subsets of J , and let M be the intersection of all inductive subsets of K . Let f : N M be a function dened by f (0) = 0 and for any n N , f (n+ ) = (f (n))+. i) Verify that f is indeed a function with domain N . (Hint: Let T be the subset of all n N at which f is dened. Prove that T is inductive and hence equal to N .) ii) Prove that the image of f is an inductive subset of M . iii) Prove that the image of f is M . iv) Prove that via this natural function f that takes 0 to 0, 1 to 1, et cetera, the elements of N can be identied with the corresponding elements of M .
74
Section 3.2: Arithmetic on N0
3.2
Arithmetic on N0
We can apply the Induction Theorem (Theorem 3.1.7) to dene addition and multiplication on N0 . For example, we want to dene the function A5 : N0 N0 as adding 5, as follows. Dene A5 (0) = 5; A5 (n+ ) = A5 (n)+ for n N0 . Or more generally, for any m N0 , dene the function Am : N0 N0 as Am (0) = m; Am (n+ ) = Am (n)+ for n N0 . The induction theorem says that Am (n) is dened for all m and n: Am (0) is in N0 , and if Am (n) N0 , then Am (n)+ is in N0 , hence Am (n+ ) N0 , so that Am is a function from N0 to N0 . Now we dene a binary operation + on N0 as m + n = Am (n). Example 3.2.1 We will prove that 2 + 2 = 4 (it better be true with this system). 2 + 2 = A2 (2) = A2 (1+ ) = A2 (1)+ = A2 (0+ )+ = (A2 (0)+ )+ = (2+ )+ = 3+ = 4. Theorem 3.2.2 For all m, n N0 , m + 0 = m, and m + (n+ ) = (m + n)+ . Proof. By denitions of + and of Am , m + 0 = Am (0) = m, and m + (n+ ) = Am (n+ ) = Am (n)+ = (m + n)+ . We can now dene multiplication: we dene Mm : N0 N0 as Mm (0) = 0, Mm (n+ ) = Mm (n) + m, whence we dene a binary operation on N0 : m n = Mm (n). Check that these denitions work! Also verify that 2 2 = 4. The following is a notational rephrasing of the denition: Theorem 3.2.3 For all m, n N0 , m 0 = 0, and m (n+ ) = (m n) + m.
Chapter 3: Construction of the basic number systems Theorem 3.2.4 The following identities hold for all m, n, p N0 : (1) (Additive identity) m + 0 = m = 0 + m. (2) m 0 = 0 = 0 m. (3) (Associativity of addition) m + (n + p) = (m + n) + p. (4) (Commutativity of addition) m + n = n + m. (4) (Successor is +1) m + 1 = m+ . (5) (Multiplicative identity) m 1 = m = 1 m. (6) (Distributivity) m (n + p) = (m n) + (m p). (Distributivity) (n + p) m = (n m) + (p m). (7) (Associativity of multiplication) m (n p) = (m n) p. (8) (Commutativity of multiplication) m n = n m.
75
Proof. Most parts are proved by induction, more precisely, by applying Theorem 3.1.7. (1) m + 0 = Am (0) = m. Let T = {m N0 : 0 + m = m}. By m + 0 = Am (0) = m for all m, 0 T . If m T , then 0 + (m+ ) = (0 + m)+ (by Theorem 3.2.2) = m+ (as m T ), so that m+ T . It follows that T is an inductive set, so that as N0 is the smallest inductive set, N0 = T . (2) By denition of , m 0 = Mm (0) = 0. Let T = {m N0 : 0 m = 0}. By m 0 = 0, 0 T . If m T , then 0 (m+ ) = (0 m) + 0 (by Denition of ) = 0 + 0 (as m T ) = 0 (by (1)),
so that m+ T . It follows that T is an inductive set, so that as N0 is the smallest inductive set, N0 = T . (3) Fix m, n N0 , and let T = {p N0 : m + (n + p) = (m + n) + p}. Since m + (n + 0) = m + n = (m + n) + 0, it follows that 0 T . If p T , then m + (n + (p+ )) = m + ((n + p)+ ) (by Theorem 3.2.2) = (m + (n + p))+ (by Theorem 3.2.2) = ((m + n) + p)+ (as p T ) = (m + n) + (p+ ) (by Theorem 3.2.2),
so that p+ T . So T is an inductive set, and as N0 is the smallest inductive set, necessarily N0 T , so (3) holds for all m, n, p N0 .
76
Section 3.2: Arithmetic on N0 (4) You prove this. First prove that n + 1 = 1 + n for all n N0 .
(4) Let T = {m N0 : m+ = m + 1}. Since 0 + 1 = A0 (1) = A0 (0+ ) = A0 (0)+ = 0+ = 1, it follows that 0 T . Suppose that m T . Then m+ + 1 = 1 + m+ (by (4)) = A1 (m+ ) (by Denition of +) = A1 (m)+ (by Denition of A1 ) = (1 + m)+ (by Denition of A1 ) = (m + 1)+ (by (4)) = (m+ )+ (since m T ), so that T is an inductive set. Thus T = N0 . (5) By denition of , m 1 = m (0+ ) = m 0 + m. In (2) we already proved that m 0 = 0, and in (1) we proved that 0 + m = m. Thus m 1 = m 0 + m = 0 + m = m. Let T = {m N0 : 1 m = m}. By (2), 0 T . If m T , then 1 (m+ ) = (1 m) + 1 (by Denition of ) = m + 1 (as m T ), = m+ (by the (4)),
so that m+ T . So T is an inductive set, and as N0 is the smallest inductive set, necessarily N0 = T , so (5) holds for all m N0 .
(6) Fix m, n N0 . Let T = {p N0 : m (n + p) = (m n) + (m p)}. By (2), m 0 = 0, and by (1), (m n) + 0 = m n, so that 0 T . If p T , then m (n + (p+ )) = m ((n + p)+ ) (by Theorem 3.2.2) = ((m n) + (m p)) + m (as p T ) = (m n) + ((m p) + m) (by (3)) = (m (n + p)) + m (by Denition of )
= (m n) + (m (p+ )) (by Denition of ),
so that p+ T . So T is an inductive set, and as N0 is the smallest inductive set, necessarily N0 T , so the rst part of (6) holds for all m, n, p N0 . The second part is proved analogously. (7) You prove (7). (8) Fix m N0 . Let T = {n N0 : n m = m n for all m N0 }. By (2), 0 T . If n T , then m (n+ ) = m n + m (by Denition of )
Chapter 3: Construction of the basic number systems = n m + m (since n T )
77
= n m+ m (by (5))
= (n+ ) m (by the second version of (6)),
so that n+ T . Thus T is an inductive set, so T = N0 , which proves (8). This gives us the familiar N0 with the familiar arithmetic properties. In particular, N0 has a binary operation + with identity 0 and a binary operation with identity 1. We already proved abstractly in Theorem 2.5.5 that such identities are unique. Exercises for Section 3.2 3.2.1 Verify with the new denitions that 3 2 = 6 and that 2 3 = 6. Was one verication easier? 3.2.2 Prove (4) and (7) in Theorem 3.2.4. 3.2.3 Prove that for all m, n N0 , m+ + n = n+ + m. 3.2.4 Prove that for all m, n, p N0 , (m+ ) (p + n) = m (p + n) + (p + n).
3.3
Cancellations in N0
The familiar subtraction cannot be applied to arbitrary pairs of elements in N0 : recall that elements of N0 are sets, and we have no way of identifying the familiar negative numbers with sets. However, we can cancel addition of equal terms, as we prove below in Theorem 3.3.2. We need an intermediate lemma: Lemma 3.3.1 Let m, n N0 . Then m = n if and only if m+ = n+ , which holds if and only if m + 1 = n + 1. Proof. By Theorem 3.2.4 (4), m+ = m + 1, so that m+ = n+ if and only if m + 1 = n + 1. If m = n, then certainly the successor m+ of m is the successor n+ of n. Suppose that m+ = n+ . Then m m+ = n+ = n {n}, so that if m = n, necessarily m n. Similarly, n m. Then by Exercise 3.1.4, m m, but this is impossible by Exercise 3.1.5. So necessarily m = n. Theorem 3.3.2 (Cancellation theorem) Let m, n, p N0 . (1) m + p = n + p if and only if m = n. (2) m n = 0 if and only if m = 0 or n = 0. (3) If p = 0 and p m = p n, then m = n.
78
Section 3.4: Order on N0
Proof. (1) Certainly if m = n, then m + p = Am (p) = An (p) = n + p. Let T = {p N0 : whenever m + p = n + p for some m, n N0 , then m = n}. By Theorem 3.2.4 (1), 0 T . Suppose that p T . We will prove that p+ is also in T . For this, suppose that for some m, n N0 , m + (p+ ) = n + (p+ ). By the denition of + and by commutativity and associativity of addition (Theorem 3.2.4 (4) and (3)), m +(p+ ) = m +(p +1) = m +(1+ p) = (m + 1) + p = (m+ ) + p, and similarly n + (p+ ) = (n+ ) + p, so that (m+ ) + p = (n+ ) + p. As p T and as m+ , n+ N0 , it follows that m+ = n+ . Hence by Lemma 3.3.1, m = n. This proves that p+ T , whence T is an inductive set and necessarily equal to N0 . This proves the additive cancellation. (2) In Theorem 3.2.4 (2) we proved that if m = 0 or n = 0, then m n = 0. Now suppose that m n = 0. If n = 0, then n = p+ for some p N0 (by Theorem 3.1.6), hence 0 = m (p+ ) = m p + m. If m = 0, then m is a successor of some element k of N0 . Thus m p + m = m p + k + = m p + (k + 1) = (m p + k ) + 1 = (m p + k )+ is the succesor of some element of N0 as well, which means that m p + m = 0. So necessarily m = 0. We just proved that n = 0 implies m = 0, or in other words, we just proved that either n = 0 or m = 0. This proves (2). (3) Fix p N0 \ {0}. Let T = {m N0 : whenever p m = p n for some n N0 , then m = n}. If p 0 = p n, then by (2), 0 = p n and n = 0, so that 0 T . Suppose that m T and that for some n N0 , p (m+ ) = p n. By (2), n = 0, so by Theorem 3.1.6, n = r + for some r N0 . By Denition of , p (m+ ) = p m + p, and p n = p (r + ) = p r + p, so that p m + p = p r + p. By (1), p m = p r , so that as m T , necessarily m = r . Then m+ = r + = n, which proves that m+ T . Thus T is an inductive set, so T = N0 . Exercises for Section 3.3 3.3.1 Suppose that m, n N0 satisfy m + n = 0. Prove that m = n = 0. 3.3.2 Suppose that m, p N0 satisfy m (p + m) = 0. Prove that m = 0.
3.3.3 Suppose that m, p, n N0 satisfy m + p = m (n+ ). Prove that p = m n.
3.4
Order on N0
Denition 3.4.1 Dene the relation on N0 as follows: for a, b N0 , we write a b if one of the three conditions below is satised: (1) a = 0; (2) a = b; + (3) a = a+ 0 , b = b0 for some a0 , b0 N0 , and a0 b0 .
Chapter 3: Construction of the basic number systems If none of the conditions above are satised, we write a > b. Dene relations , < on N0 as: a b if b a, and a > b if b < a. Clearly , are reexive relations and >, < are not. Lemma 3.4.2 If m n and n m, then m = n.
79
Proof. Let T = {m N0 : for all n N0 , m n, n m implies m = n}. If m = 0, then since 0 is not the successor of any number in N0 , it follows that that condition (3) for n m in Denition 3.4.1 is not satised, hence necessarily n = 0. Thus 0 T . Now let m T . Let n N0 such that m+ n and n m+ . As for m above, + also n = 0. Thus n = n+ 0 for some n0 N0 . If m = n, we have nothing to prove, so we may suppose otherwise, so by denition of , m n0 and n0 m. But m T , so that + m = n0 , whence m+ = n+ 0 = n. Thus m T . It follows that T is inductive, so T = N0 . Theorem 3.4.3 is a transitive relation. Proof. Let T = {n N0 : n m and m p implies m p}. Then T contains 0 because the assumption m 0 forces m = 0 be the lemma, which implies by condition (1) in the denition of that m p. Now let n T . Suppose that m n+ and n+ p. If m = 0 or m = p, then by denition m p. If m = n+ , then by assumption n+ p we get that m p. Similarly, n+ = p implies that m p. So we may suppose that m is none of 0, n+ or p + + and that n+ = p. Then by assumption m n+ , necessarily m = m+ 0 , n = n0 for some m0 , n0 N0 , and m0 n0 . By cancellation (more precisely, Lemma 3.3.1), n = n0 . Since 0 is not the successor of any element of N0 , we have that n+ = 0, so that the assumption n+ p forces that p = p+ 0 for some p0 N0 , and n p0 . But n T , so that m0 n and + n p0 imply that m0 p0 , and then by condition (3), m = m+ 0 p0 = p. This proves that T is an inductive set, so that T = N0 . Lemma 3.4.4 Let m, n N0 . Then m n if and only if m+ n+ , and m < n if and only if m+ < n+ . Proof. If m n, then m+ n+ by condition (3) of the denition Denition 3.4.1 of . Now suppose that m+ n+ . They by denition of , either m+ = n+ or m n. In the latter case we are done, and in the former case, by Lemma 3.3.1, m = n, whence m n. Now m < n if and only if n m is false, which, by the previous paragraph, is if and only if n+ m+ is false, which is if and only if m+ < n+ . Lemma 3.4.5 For m, n N0 , m < n if and only if m n and m = n.
80
Section 3.4: Order on N0
Proof. Let T = {m N0 : n N0 , m < n if and only if m n and m = n}. Since 0 n is true for all n and 0 < n means that n 0 is false, this means in particular by the denition of that n = 0. Thus 0 T . Now suppose that m T . Let n N0 . Then m+ < n means that n m+ is false. This implies that n = 0. Thus n = n+ 0 for some + n0 N0 . But then n m false implies by Lemma 3.4.4 that n0 m is false, so that n0 > m. Since m T , this implies that m n0 and m = n0 , which in turn implies by Lemma 3.4.4 that m+ n and by Lemma 3.3.1 that m+ = n. Thus m+ < n implies that m+ n and m+ = n. Now suppose that m+ n and m+ = n. If n = 0, then by Lemma 3.4.2, since m+ 0 m+ , we have that m+ = 0, which is a contradiction. So + + necessarily n = n+ 0 for some n0 N0 . Then m n0 , so that by Lemma 3.4.4, m n0 . By Lemma 3.3.1, m = n0 , so that since m T , it follows that m < n0 . Hence n0 m is false, and so by Lemma 3.4.4, n m+ is false, hence m+ < n. This proves that m+ T , so that T is inductive. Proposition 3.4.6 If m, n N0 and m > n, there exists r N such that m = n + r . Proof. The property is vacuously true for m = 0. Suppose that the proposition holds for some m and all n. We have to prove that it holds for m+ and for all n. Namely, we assume that m+ > n, and we will prove that there exists r N0 such that m+ = n + r . If n = 0, we may take r = m+ . Now let n = 0. Then n = p+ for some p N0 . Since m+ > p+ , by Lemma 3.4.4, m > p. Hence by assumption on m, there exists r N0 such that m = p + r . Hence m+ = (p + r )+ = (p + r ) + 1 = p + (r + 1) = p + r + , and r + N0 . So the proposition also holds for m+ , so that the set of all m satisfying the proposition is an inductive set, whence the proposition holds for all m N0 . Lemma 3.4.7 Let a, n N0 be such that n a n+ . Then either a = n or a = n+ .
+ + Proof. Suppose n = 0. If a = 0, then a = a+ 0 for some a0 N0 , so a0 0 , so that by Lemma 3.4.4, a0 0, which is only possible if a0 = 0, which then implies that a = a+ 0 = 1. Thus the lemma holds for n = 0. Suppose that the lemma holds for n. We will prove that it holds for n+ . So suppose that n+ a (n+ )+ . Since n+ a, necessarily a = 0. Thus a = a+ 0 for some a0 N0 . + Then n+ a0 (n+ )+ , so that by Lemma 3.4.4, n a0 n+ . By assumption on n, either a0 = n or a0 = n+ , so that either a = n+ or a = (n+ )+ . Thus the lemma also holds for n+ . This proves that T is an inductive set, which proves the lemma.
Proposition 3.4.8 N0 is well-ordered (see Denition 2.7.6). In other words, every nonempty subset S of N0 has a least element, that is, an element r such that for all t S , r t.
Chapter 3: Construction of the basic number systems Proof. We will prove that the following set is inductive: T = {n N0 : If S N0 and n S , then S has a least element}.
81
Note that 0 T because 0 n for all n N0 (and hence 0 n for all n S ). Suppose that n T . We need to prove that n+ T . Let S N0 and n+ T . By assumption that n T , the set S {n} has a least element. Thus there exists r S {n} such that for all t S , r t and r n. If r S , then we just showed that for all t s, r t, so that S has a least element. Now suppose that r S . So necessarily r = n. Then we claim that n+ is the least element of S . Namely, let t S . If t < n+ , then since n t, we have n t < n+ , which is a contradiction. So necessarily t n+ . This shows that every non-empty subset of N0 has a least element. Exercises for Section 3.4 3.4.2 Prove that if m < n and n p, then m < p. (Verify also other standard combinations.) 3.4.3 Prove that > is a transitive relation. 3.4.4 Let m, n, p N0 . Prove that m n if and only if m + p n + p and that m > n if and only if m + p > n + p. 3.4.5 Suppose that m < n. Prove that mp np for all p and that mp < np for all p = 0. 3.4.7 Prove that for all n N0 , n < 2n . (Recall power notation from Notation 2.5.16.) 3.4.6 Suppose that m, p, n N0 and m n. Prove that m + p < n + p+ . 3.4.1 Prove that for all n N0 , n < n+ .
3.4.8 Find, with proof, all n N0 such that n2 < 2n . (Hint: induction, plus a few special cases.) 3.4.10 Let a, b, c N0 . Is it true that ab = c implies that a c? 3.4.9 Let a, b, c N+ . If ab = c, prove that a c and b c.
3.5
Construction of Z, arithmetic, and order on Z
How are we supposed to think of the familiar 5? Well go about it by using the rigorous construction of N0 and all the arithmetic on it. The following is a rehashing of Example 2.3.8. Denition 3.5.1 Consider the Cartesian product N0 N0 . Elements are pairs of the form (m, n), with m, n N0 . If m, n, m, n N0 , we will write (m, n) (m , n ) if m + n = n + m .
82
Section 3.5: Construction of Z, arithmetic, and order on Z
The relation satises the following properties: (1) (Reexivity) For all m, n N0 , (m, n) (m, n). (2) (Symmetry) If (m, n) (m , n ), then (m , n ) (m, n). (3) (Transitivity) If (m, n) (m , n ) and (m , n ) (m , n ), then (m, n) (m , n ). Thus is an equivalence relation on N0 N0 , and we can talk about the equivalence class [(m, n)] of (m, n). Denition 3.5.2 We dene Z to be the set of all equivalence classes of elements of N0 N0 under the equivalence relation . Elements of Z are called whole numbers, or integers. We have a natural inclusion of N0 into Z by identifying n N0 with [(n, 0)] in Z. Note that if n, m N0 are distinct, then so are [(n, 0)] and [(m, 0)], so that N0 is indeed identied with a natural subset of Z. But Z is strictly larger: [(0, 1)] is not equal to [(m, 0)] for all m N0 . More about the familiar integers and Z is in Notation 3.5.5. To dene arithmetic on Z, we rst dene arithmetic on N0 N0 . Start with addition: for m, n, p, q N0 , set (m, n) + (p, q ) = (m + p, n + q ). (There are three + here: the last two are the already well-studied addition in N0 , and the rst one is the one we are dening now on N0 N0 . Yes, we are using the same name for two dierent operations; but thats ok: many families have a child named Pat, and the many Pats are not all equal.) Since + on N0 is binary, so is + on N0 N0 . We show next that + is compatible with : Theorem 3.5.3 Let m, m , n, n, p, p , q, q N0 such that (m, n) (m , n ) and (p, q ) (p , q ). Then (m + p, n + q ) (m + p , n + q ). Proof. By assumption, m + n = m + n and p + q = p + q . Then by associativity and commutativity of + on N0 , (m + p) + (n + q ) = (m + n ) + (p + q ) = (m + n) + (p + q ) = (n + q ) + (m + p ), so that (m + p, n + q ) (m + p , n + q ). This shows that + is well-dened on the equivalence classes of , hence + makes sense on Z: [(m, n)] + [(p, q )] = [(m + p, n + q )]. For example, identifying 5, 6 N0 with [(5, 0)], [(6, 0)] Z0 , we get 5 + 6 = 11 and [(5, 0)] + [(6, 0)] = [(5 + 6, 0)] = [(11, 0)], as expected. Also, [(5, 0)] + [(0, 6)] = [(5, 6)] = [(0, 1)]. Multiplication on Z is a little more complicated: if (m, n) and (p, q ) are in N0 N0 , then we declare (m, n) (p, q ) = (m p + n q, m q + n p). It is left to Exercise 3.5.5 to
83
prove that this multiplication is compatible with , and so that we can dene product on Z as: [(m, n)] [(p, q )] = [(mp + nq, mq + np)].
Consequently, [(5, 0)] [(6, 0)] = [(5 6 + 0 0, 5 0 + 0 6)] = [(30, 0)], as expected. Similarly compute [(5, 0)] [(0, 6)]. We next prove the basic arithmetic laws on Z.
Theorem 3.5.4 The following identities hold for all m, n, p Z: (1) (Additive identity) m + [(0, 0)] = m = [(0, 0)] + m. (2) m [(0, 0)] = [(0, 0)] = [(0, 0)] m. (3) (Associativity of addition) m + (n + p) = (m + n) + p. (4) (Commutativity of addition) m + n = n + m. (5) (Multiplicative identity) m [(1, 0)] = m = [(1, 0)] m. (6) (Distributivity) m (n + p) = (m n) + (m p). (7) (Associativity of multiplication) m (n p) = (m n) p. (8) (Commutativity of multiplication) m n = n m. (9) (Existence of additive inverses) There exists r Z such that m + r = r + m = [(0, 0)]. Proof. Let a, b, c, d, e, f N0 such that m = [(a, b)], n = [(c, d)] and p = [(e, f )]. With this, the proofs follow easily from the denitions of + and . For example, m + [(0, 0)] = [(a, b)] + [(0, 0)] = [(a + 0, b + 0)] = [(a, b)] = m, and similarly [(0, 0)] + m = m. This proves (1). Properties (2)-(4) are equally easy to prove, and (5) follows from m [(1, 0)] = [(a, b)] [(1, 0)] = [(a 1 + b 0, a 0 + b 1)] = [(a + 0, 0 + b)] = [(a, b)] = m, and similarly [(1, 0)] m = m. Lets check (6), and you explain each step: (m (n + p)) = ([(a, b)] ([(c, d)] + [(e, f )])) = [(a (c + e) + b (d + f ), a (d + f ) + b (c + e))] = [(a, b)] [(c + e, d + f )]
= [(a c + a e + b d + b f, a d + a f + b c + b e)]
(why may we omit parentheses in addition?)
= [(a c + b d, a d + b c)] + [(a e + b f, a f + b e)] = m n + m p. = [(a, b)] [(c, d)] + [(a, b)] [(e, f )]
84
Section 3.5: Construction of Z, arithmetic, and order on Z
You verify (7) and (8). (9) Set r = [(b, a)]. Then m + r = [(a, b)] + [(b, a)] = [(a + b, b + a)] = [(a + b, a + b)] = [(0, 0)], and similarly r + m = [(0, 0)]. Notation 3.5.5 It is important to record that the additive inverse of the equivalence class [(a, b)] of (a, b) is the equivalence class [(b, a)] of (b, a). The additive inverse of an element m is always denoted m. Thus for any n N0 , by identifying n with [(n, 0)] in Z, the additive inverse is n = [(n, 0)] = [(0, n)]. Thus, under our identications, any element [(m, n)] in Z is actually equal to [(m, n)] = [(m, 0)] + [(0, n)] = m + (n), and it is standard shorthand to write this as m n. Thus in the future we will write elements [(m, n)] in Z as m n. In particular, if m n, by Proposition 3.4.6, there exists r N0 such that m = n + r , and then [(m, n)] = [(r, 0)] = r . If instead n m, there exists p N0 such that n = m + p, and then [(m, n)] = [(0, p)] = p. Theorem 3.5.6 Let m, n Z. Then (m) n = m (n) = (m n). Proof. Observe that m (n) + m n = m ((n) + n) (by distributivity: Theorem 3.5.4 (6)) = 0 (by Theorem 3.5.4 (2)). = m 0 (by notation for additive inverses)
Thus m (n) is an additive inverse of m n, and by uniqueness of inverses (Theorem 2.5.10), m (n) is the additive inverse of m n, i.e., m (n) = (m n). With that, (m) n = n (m) (by commutativity of multiplication) = (n m) (by the previous paragraph). Theorem 3.5.7 Let m, n Z. Then m n = 0 if and only if m = 0 or n = 0. Proof. We already proved that if m = 0 or n = 0, then m n = 0. Now suppose that m n = 0. Write m = [(a, b)] with a, b N0 . By Notation 3.5.5 we may assume that either a = 0 or b = 0. Write n = [(c, d)] for c, d N0 . If a = 0, then [(0, 0)] = 0 = m n = [(0, b)] [(c, d)] = [(b d, b c)]. This means (by denition of ) that b d = b c. If b = 0, then m = [(0, 0)] = 0, and if b = 0, then by Theorem 3.3.2, d = c, so that n = [(c, d) = [(0, 0)] = 0. Similarly, if b = 0, then either m = 0 or n = 0. But a = 0 or b = 0 covers all the possible cases for elements of Z, so weve proved the theorem.
85
Denition 3.5.8 (Order on Z) Dene a relation on Z as follows: [(a, b)] [(c, d)] if a + d b + c, where the latter is the already known relation on N0 (see Section 3.4). Similarly dene <, , >. It is left to the reader to verify the following: Lemma 3.5.9 (1) on Z is well-dened (this means that if one picks dierent representatives of the equivalence classes, for one pair holds if and only if it holds for the other pair). (2) , are reexive and transitive but not symmetric. (3) <, > are transitive but not reexive and not symmetric. (4) If m Z and 0 m, then m N0 . (5) If m Z and m 0, then m N0 . Theorem 3.5.10 The following properties hold on Z: (1) (Trichotomy) For any m, n Z, exactly one of the following three conditions holds: (i) m = n. (ii) m < n. (iii) n < m. (2) (Transitivity) For all m, n, p Z, m < n and n < p imply m < p. (3) (Compatibility of < with addition) For all m, n, p Z, m < n then m + p < n + p. (4) (Compatibility of < with multiplication by positive numbers) For all m, n, p Z, m < n and 0 < p imply mp < np. Proof. Suppose that m = [(a, b)], n = [(c, d)], and p = [(e, f )] for some a, b, c, d, e, f N0 . By Denition 3.4.1, either a + d b + c or a + d > b + c. By Lemma 3.4.5, the former condition reduces to either a + d < b + c or a + d = b + c. Thus one of the three conditions in the trichotomy holds, and again by Lemma 3.4.5, exactly one of the three conditions. Suppose that m < n and n < p. This means that a+d < b+c and c+f < d+e. Then by associativity and commutativity of addition in N0 , by Exercise 3.4.4, and by transitivity of < on N0 , (a+f )+(c+d) = (a+d)+(c+f ) < (a+d)+(d+e) < (b+c)+(d+e) = (b+e)+(c+d), so that by Exercise 3.4.4, a + f < b + e. This says that m < p, which proves that < is transitive. This proves (2). The other properties follow similarly and are left to the reader. With this, we can use the adjectives positive, negative, non-positive, nonnegative from Denition 2.7.8 for elements of Z, and in general we can use the properties developed in Section 2.7, such as absolute values, triangle inequality, and reverse triangle inequality.
Section 3.6: Construction of the eld Q of rational numbers
3.5.2 Suppose that a, b, c, d N0 such that (a, b) (c, d) and a c. Prove that there exists e N0 such that a = c + e and b = d + e. 3.5.3 Convince yourself that Z as dened in Denition 3.5.2 is the same as the familiar set of all integers. 3.5.4 (Cancellation in Z) Let m, n, p Z. i) If m + p = n + p, prove that m = n. ii) If p = 0 and m p = n p, prove that m = n.
3.5.1 Prove that for all a, b, c N0 , such that (a, b) (a + c, a + c).
3.5.5 Prove that if (m, n) (m , n ) and (p, q ) (p , q ), then (m p + n q, m q + n p) (m p + n q , m q + n p ). Why does this say that is well-dened on Z? 3.5.7 Let a, b, c, d N0 . Suppose that [(b, a)] < [(c, d)]. Prove that [(a, b)] > [(d, c)] and [(a, b)] + [(c, d)] > [(0, 0)]. 3.5.8 Prove (2), (3), (4), (7), (8) in Theorem 3.5.4. 3.5.9 Prove Lemma 3.5.9. 3.5.10 Prove (3) and (4) in Theorem 3.5.10. 3.5.6 Prove that if (a, b) (c, d) then (a, c) (b, d).
3.6
Construction of the eld Q of rational numbers
We next dene the set Q, arithmetic, and order on it, and we show that Q is a eld. Elements of course correspond to the familiar rational numbers. The following is a rehashing of the construction of Q from Z rst given in Exercise 2.3.9. The relation on Z Z \ {0}, given by (m, n) (m , n ) if m n = m n, is an equivalence relation. The reader may wish to go through the details of the proof, to check that only those properties of Z are used that we have proved in the previous section, and that no well-known knowledge crept into the proof. Denition 3.6.1 We dene the set of all equivalence classes of this relation to be Q. The elements of Q are called rational numbers. We have a natural inclusion of Z into Q (and thus of N0 into Q) by identifying m Z with [(m, 1)]. Note that if m, n Z are distinct, so are [(m, 1)] and [(n, 1)], so that Z is
87
indeed identied with a natural subset of Q. But Q is strictly larger: [(1, 2)] is not equal to [(m, 1)] for any m Z. More about the familiar rational numbers is in Notation 3.6.4. We declare the following binary operations + and on elements of Z (Z \ {0}): (m, n) + (p, q ) = (m q + p n, n q ), (m, n) (p, q ) = (m p, n q ). (The + and inside the pair entries are the operations on Z; the other + and are what we are dening here.) We show next that the new + and are compatible with : Theorem 3.6.2 Let m, m , n, n, p, p , q, q N0 such that (m, n)(m, n ) and (p, q )(p, q ). Then ((m, n) + (p, q )) ((m , n ) + (p , q )) , ((m, n) (p, q )) ((m , n ) (p , q )) .
and
Proof. By assumption, m n = m n and p q = p q . Then (you explain each step): (m q + p n) (n q ) = (n q ) (m q + p n) = (n q ) (m q ) + (n q ) (p n)
= (q n ) (m q ) + n (q (p n))
= q (n (m q )) + n ((q p) n)
= q ((n m) q ) + n ((p q ) n)
= q ((m n ) q ) + n ((p q ) n)
= q ((m n) q ) + n (p (q n))
= q (m (n q )) + n (p (n q ))
= (q m ) (n q ) + (n p ) (n q )
= (n q ) (q m ) + (n q ) (n p ) = (n q ) (m q + p n )
= (n q ) (m q ) + (n q ) (p n ) = (m q + p n ) (n q ).
This says that (m q + p n, n q ), = (m q + p n , n q ), which in turn says that ((m, n) + (p, q )) ((m , n ) + (p , q )). It is even easier to prove that ((m, n) (p, q )) ((m , n ) (p , q )) (verify).
The last theorem proves that that the following binary operations + and on Q are well-dened: [(m, n)] + [(p, q )] = [(m, n) + (p, q )] = [(m q + p n, n q )],
88
Section 3.6: Construction of the eld Q of rational numbers [(m, n)] [(p, q )] = [(m, n) (p, q )] = [(m p, n q )].
The + and on the left are what we are dening here; the + and in the middle are binary operations on Z (Z \ {0}), and the + and on the right are binary operations on Z. We now move to arithmetic laws on Q. Theorem 3.6.3 Q is a eld (see Denition 2.6.1), or explicitly, the following identities hold for all m, n, p Q: (1) (Additive identity) m + [(0, 1)] = m = [(0, 1)] + m. (2) m [(0, 1)] = [(0, 1)] = [(0, 1)] m. (3) (Associativity of addition) m + (n + p) = (m + n) + p. (4) (Commutativity of addition) m + n = n + m. (5) (Multiplicative identity) m [(1, 1)] = m = [(1, 1)] m. (6) (Distributivity) m (n + p) = (m n) + (m p). (7) (Associativity of multiplication) m (n p) = (m n) p. (8) (Commutativity of multiplication) m n = n m. (9) [(1, 1)] = [(0, 1)]. (10) (Existence of additive inverses) There exists r Q such that m + r = r + m = [(0, 1)]. (11) (Existence of multiplicative inverses) If m = [(0, 1)], there exists r Q such that m r = r m = [(1, 1)]. Proof. Let a, b, c, d, e, f Z such that m = [(a, b)], n = [(c, d)] and p = [(e, f )]. With this, the proofs follow easily from the denitions of + and . For example, m + [(0, 1)] = [(a, b)] + [(0, 1)] = [(a + 0, b 1)] = [(a, b)] = m, and similarly [(0, 1)] + m = m. This proves (1). Properties (2)-(5) are equally easy to prove. Lets check (6), and you explain each step: m (n + p) = [(a, b)] ([(c, d)] + [(e, f )])
= [(a (c f + e d), b (d f ))]
= [(a, b)] [(c f + e d, d f )]
= m n + m p.
= [(a, b)] [(c, d)] + [(a, b)] [(e, f )]
= [(a (c f ) + a (e d), b (d f ))]
You verify (7) and (8). (9) By the properties of Z, 0 1 = 0 = 0+ = 1 = 1 1, so that (0, 1) (1, 1), so that [(0, 1)] = [(1, 1)] in Q.
89
(10) By denition of m and Q, [(a, b)] Q, and m +[(a, b)] = [(a b +(a) b, b b)] = [(b a + b (a), b b)] = [(b (a + (a)), b b)] = [(b 0, b b)] = [(0, b b)] = [(0, 1)], and similarly [(a, b)] + m = [(0, 1)]. Thus additive inverses exist in Q. (11) By assumption [(a, b)] = [(0, 1)], This means that a 1 = 0 b, so that a = 0. Then [(b, a)] Q, and [(a, b)] [(b, a)] = [(a b, b a)] = [(a b, a b)] = [(1, 1)]. Similarly [(b, a)] [(a, b)] = [(1, 1)]. Notation 3.6.4 It is important to note that the additive inverse of the equivalence class [(a, b)] of (a, b) is [(a, b)], and that the multiplicative inverse of non-zero [(a, b)] is [(b, a)]. For an element m, its additive inverse is always denoted m, and the multiplicative inverse, when it exists, is written m1 . Thus for any n Z, by identifying n with [(n, 1)] in Q, the additive inverse is n = [(n, 1)] = [(n, 1)], and if n = 0, the multiplicative inverse is n1 = [(n, 1)]1 = [(1, n)]. Thus, under our identications, any element [(m, n)] in Q is equal to [(m, n)] = [(m, 1)] [(1, n)] = m n1 , and it is standard shorthand to write this as m/n. Thus in the future we will write elements [(m, n)] in Q as m/n. Furthermore, whenever m, n Z and n = 0, then m/n Q. Theorem 3.6.5 (Cancellation laws in Q) Let m, n, p Q. (1) If m + p = n + p, then m = n. (2) If m p = n p and if p = 0, then m = n. Proof. For part (1), m = m + 0 (by additive identity) = m + (p + (p)) = (n + p) + (p) = (m + p) + (p) (by associativity of addition) = n + (p + (p)) (by associativity of addition) = n + 0 (by additive identity) = n (by additive identity), which proves (1). The proof of part (2) follows the proof above with the following changes: replace all + by , replace p by /, and correspondingly modify all explanations. Theorem 3.6.6 Q is an ordered eld (see Denition 2.7.8) with the order < given by: [(a, b)] < [(c, d)] if ad < bc; if bd > 0; ad > bc; if bd < 0.
90
Section 3.6: Construction of the eld Q of rational numbers
Proof. Theorem 3.6.3 proves that Q is a eld. Now we prove that Q is ordered. First we need to prove that < is well-dened. Let m, n Q. Write m = [(a, b)] = [(a , b )], n = [(c, d)] = [(c , d )], and m < n. Since [(a, b)] = [(a, b)], bd > 0 if and only if (b)d < 0, and ad > bc if and only if (a)d < (b)c, so by possibly replacing (a, b) with (a, b), we may assume that b > 0. Similarly, we may assume that b , d, d > 0. Then m < n says that ad < bc. To conclude that < is well-dened we need to show that a d < b c . By denition of Q, ab = ba and cd = dc . By associativity and commutativity of multiplication in Z, bb dd (b c a d ) = b(b )2 d (dc ) b d(d )2 (ba ) = b(b )2 d (cd ) b d(d )2 (ab ) = (b )2 (d )2 (bc ad), which is a product of all positive numbers in Z, so it is positive by Theorem 3.5.10. Since b, b , d, d are all positive, so is b c a d , so that b c > a d , which proves that < is well-dened on Q. Proof of trichotomy: let m, n Q. Write m = [(a, b)] and n = [(c, d)]. By Theorem 3.5.10, by trichotomy on Z we have trichotomy on Q. We leave the rest of this proof to the exercises. Thus the following terms and facts, developed in Section 2.7, apply to elements of Q: positive, negative, non-positive, non-negative, intervals, absolute value, triangle inequality, reverse triangle inequality. Theorem 3.6.7 (Archimedean property) Let m, n Q. If m > 0, there exists p N such that n < pm. Proof. Write m = a/b and n = c/d for some a, b, c, d Z. Since a/b = (a)/(b), by possibly replacing a, b with a, b, respectively, we may assume that b > 0. Similarly we may assume that d > 0. Since m > 0, necessarily a > 0. If n 0, set p = 1, and clearly n < 0 < pm; and if n > 0, then c > 0 and set p = bc + 1. Then p is a positive integer, pad bc = (bc +1)ad bc = bc(ad 1) + ad > bc(ad 1). Since ad N0 is non-zero, we have that ad 1, so that ad 1 0, hence bc(ad 1) N0 , whence pad bc = bc(ad 1)+ ad > 0. It follows that pm = p(a/c) > b/d = n. Exercises for Section 3.6 3.6.2 Prove that for all (m, n) Z (Z \ {0}) and all non-zero p Z, (m, n) (m p, n p). In other words, m/n = (m p)/(n p). 3.6.1 Compute (5/6) (3/4), (5/6) + (3/4).
3.6.3 Prove (2), (3), (4), (5), (7), (8) in Theorem 3.6.3.
3.6.4 Finish the proof of Theorem 3.6.6, namely prove properties (2), (3), and (4) of ordered elds for Q. 3.6.5 Let m, n, p, q Z with n, q non-zero. Prove that m/n + p/q and m/n p/q are rational numbers.
91
3.6.6 Let m Q. Prove that there exists a positive integer N such that m < 2N . (Hint: If m 0, this is easy. Now suppose that m > 0. Write m = a/b for some positive integers a, b. It suces to prove that there exists N such that a < 2N . Let T = {x N0 : 2N x for all N N0 }. If T = , then in particular a T , so that there exists N as desired. Now suppose that T = . Then by Proposition 3.4.8, T has a least element. Let c be the least element. Then c is positive, and 2c c, which contradicts Exercise 3.4.7.)
3.7
Construction of the eld R of real numbers
In this section we construct real numbers from Q via Dedekind cuts. In fact, a Dedekind cut stands for a real number and vice versa. We establish the familiar arithmetic on R, and in the subsequent section we show that R is an ordered eld with further good properties. To go through these two sections carefully would take many hours. I typically spend 1.5 class hours going through them, asking the students not to take notes but instead to nod vigorously in agreement and awe at all the constructions and claims. I expect the students to get the gist of the construction, a conviction that any arithmetic and order property of real numbers can be veried by them if necessary, but I do not expect the students to have gone through all the details. Denition 3.7.1 A Dedekind cut is a pair (L, R) with the following properties: (1) L, R are non-empty subsets of Q. (2) For all l L and all r R, l < r . (3) For all l L, there exists l L such that l < l . (4) For all r R, there exists r R such that r < r . (5) L R is either Q or Q \ {m}. In case L R = Q \ {m}, then for all l L and all r R, l < m < r . By condition (2) and trichotomy property of < on Q, L R = . We can visualize a Dedekind cut (L, R) as the separation of the rational number line into the left and right parts, and the separation comes either at a rational number m, or possibly at some non-rational number. The letters L and R are mnemonics for left and right. Examples and non-examples 3.7.2 (1) (Q \ {0}, {0}) is not a Dedekind cut: it violates condition (2). It also violates condition (4). (2) ({x Q : x2 < 2}, {x Q : x2 > 2}) is not a Dedekind cut: it violates condition (2).
92
Section 3.7: Construction of the eld R of real numbers
(3) ({x Q : x < 0}, {x Q : x > 0}) is a Dedekind cut. (4) ({x Q : x < 0}, {x Q : x 0}) is not a Dedekind cut: it violates condition (4). Denition 3.7.3 We dene R to be the set of all Dedekind cuts. Elements of R are therefore Dedekind cuts, but more normally, they are called real numbers. The rst goal is to prove that Q is a subset of R, at least under an identication of elements of Q with Dedekind cuts. Proposition 3.7.4 For any m Q, Dm = ({l Q : l < m}, {r Q : r > m}) is a Dedekind cut. Proof. Let L = {l Q : l < m}, R = {r Q : r > m}. By Theorem 3.6.7, there exists p N such that m < p 1 = p. Since p N Z Q, we have p R. Similarly, there exists q Q such that m < q , which says that the rational number q is strictly smaller than m and is thus in L. Thus L and R are not empty. By trichotomy on Q, for every r Q, exactly one of the following holds: r < m, r = m, r > m. So L R = Q \ {m}. By denition, for any l L and any r R, l < m < r , and by transitivity of <, l < r . If l L, then l < m, and by Exercise 2.7.18, l = (l + m)/2 is a rational number strictly larger than l and strictly smaller than m. Thus, l < l L, and condition (3) of Dedekind cuts holds. Similarly, for any r R, r = (r + m)/2 satises condition (4) of Dedekind cuts. This gives a natural inclusion of Q into R, with a rational number m being identied by the Dedekind cut Dm . If m, n are distinct in Q, say m < n, then by Exercise 2.7.18, (m + n)/2 is in the right side of Dm and the left side of Dn , so that Dm and Dn are distinct as well. Thus we can think of Q as a subset of R. A dierent Dedekind cut is ({r Q : r 2 < 2 or r < 0}, ({r Q : r 2 2 and r 0}). In Exercise 3.7.3 the reader is asked to prove that this is a Dedekind cut that is not of the form Dm for any m Q. The reason was proved on page 16: 2 is not the square of a rational number. Thus Q is a proper subset of R. Now we turn to arithmetic on R. Denition 3.7.5 Let A and B be subsets of Q. Dene A + B = {a + b : a A, b B }, A = {a : a A}, A B = {a b : a A, b B } = A + (B ), and A B = {a b : a A, b B }. Lemma 3.7.6 Let (L1 , R1 ) and (L2 , R2 ) be Dedekind cuts. Then (L1 + L2 , R1 + R2 ) is a Dedekind cut.
93
Proof. Since L1 , L2 , R1 , R2 are non-empty subsets of Q, then L1 + L2 , R1 + R2 are nonempty subsets of Q as well. Let l L1 + L2 and r R1 + R2 . Then l = l1 + l2 and r = r1 + r2 for some li Li and ri Ri (i = 1, 2). Then l1 < r1 and l2 < r2 , so that l = l1 + l2 < l1 + r2 < r1 + r2 = r , and by transitivity of < on Q, l < r . Conditions (3) and (4) hold similarly. Suppose that x Q is not in (L1 + L2 ) (R1 + R2 ). Write x = m/n for some m, n Z. Since m/n = (m)/(n), by possibly replacing n with n we may assume that n > 0. Let l L1 . By Theorem 3.6.7, there exists q Z such that q (1/n) > |l|, so that (q )/n < |l| l, and so by Exercise 3.7.4, (q )/n L1 and the set S = {s Z : (s q )/n R1 } is a subset of N. Let r R1 . By Theorem 3.6.7, there exists p Z such that p(1/n) > r . Thus by Exercise 3.7.4, p(1/n) R1 , which proves that p + q S , so that S is not empty. By Proposition 3.4.8, S contains the smallest element, say s. Then s 1 S , and (s q )/n R1 (s 1 q )/n R1 . This proves that for some a Z, a/n L1 and (a + 1)/n R1 . Similarly, for some b Z, b/n L2 and (b + 1)/n R2 . Now let p Z. If p a + b, then p a b, so that by Exercise 3.7.4, (p a)/n L2 , hence p/n = (a/n) + (p a)/n L1 + L2 . If p a + b + 2, then similarly p/n R1 + R2 . So necessarily m = a + b + 1. In other words, if p is an integer and p < m, then p/n L1 + L2 , and if p > m, then p/n R1 + R2 . Let y Q, and write y x = c/d for some c, d Z. Since c/d = (c)/(d), by possibly replacing d with d we may assume that d > 0. By repeating the paragraph above for x = (dm)/(dn), and by noting that y = (y x) + x = (cn)/(dn) + (dm)/(dn) = (cn + dm)/(dn), we get that if cn + dm < dm, then y L1 + L2 , and if cn + dm > dm, then y R1 + R2 . In other words, for all y Q, if y < x, then y L1 + L2 , and if y > x, then y R1 + R2 . This proves that (L1 + L2 , R1 + R2 ) is a Dedekind cut. Thus we can dene addition on R by (L1 , R1 ) + (L2 , R2 ) = (L1 + L2 , R1 + R2 ). There are two dierent meanings to + above: on the right side, + acts on two sets of rational numbers, and on the left, + stands for the new sum. The proof of the following is left as an exercise. Theorem 3.7.7 Addition on R, as dened above, is commutative and associative. The additive identity if D0 = (Q , Q+ ), and the additive inverse of any (L, R) R is (R, L) (recall Denition 3.7.5). Naturally we will write the additive identity of R as 0, and for any x R, we write its additive inverse as x. Multiplication on R is more complicated, but if one keeps in mind the well-known facts about real numbers, the following denition of multiplication on Dedekind cuts is
94
Section 3.7: Construction of the eld R of real numbers alternate formulation): if if if if R and R contain only positive elements of Q; R but not R contains only positive elements; R but not R contains p onlyositive elements; neither R nor R contains only positive elements,
where in each case is the unique subset of Q that makes the result a Dedekind cut. It is left to Exercise 3.7.9 to prove that there exist such Dedekind cuts (contrast to Exercise 3.7.1). Clearly is a commutative operation, and it is easy to show that D1 = ({x Q : x < 1}, {x Q : x > 1}) is a multiplicative identity (and therefore the unique multiplicative identity by Theorem 2.5.5). The following are multiplicative inverses for (L, R) = D0 : (L, R)1 = ( , {1/r : r R}), if R contains only positive elements of Q; ({1/l : l L}, ), if R does not contain only positive elements.
natural (see Exercise 3.7.11 for an ( , R R ), (R L , ), (L, R) (L , R ) = (L R , ), ( , L L ),
If all elements of R are positive, we are certainly not dividing by 0 in the rst case. In the second case, by Exercise 3.7.4, L contains only negative elements, and so again we are not dividing by 0. It is now straightforward that the listed element is indeed the multiplicative inverse. Proofs of associativity of multiplication and of distributivity of multiplication over addition are more involved. With the denition above, we would have to verify eight cases for associativity, and many cases for distributivity, depending on whether the left parts of the various Dedekind cuts contain positive elements. Yes, this can be done, but it is tedious work. We provide a dierent proof in these notes that is not a brute-force attack but is more conceptual and gives side results as a bonus. We rst need a lemma. Lemma 3.7.8 For any Dedekind cuts x, y , (x) y = (x y ) = x (y ). Proof. Write x = (L, R) and y = (L , R ). If L, L contain positive elements, then R, R do not contain positive elements, and by denition of multiplication, (x y ) + ((x) y ) = ((L, R) (L , R )) + ((R, L) (L , R )) = ( , R R ) + ((R) R , ) = (Q \ (R R ), R R ) + (R R ), )
the last equality holding only up to a point, by Exercise 3.7.2. But (Q \ (R R )) + (R R ) = (, 0) because R R consists of positive numbers strictly greater than numbers in Q \ (R R ). Thus (x y ) + ((x) y ) = D0 . By commutativity of addition this also gives ((x) y ) + (x y ) = D0 , ((x) y ) = (x y ). If L contains positive elements but L does not, then (x y ) + ((x) y ) = ((L, R) (L , R )) + ((R, L) (L , R ))
Chapter 3: Construction of the basic number systems = (R L , ) + ( , (R L )),
95
and similarly to the previous case, this is D0 , so again ((x) y ) = (x y ). The remaining cases of ((x) y ) = (x y ) are proved similarly. Then by commutativity of multiplication, x (y ) = (y ) x, and by the established case this equals (y x) = (x y ). Theorem 3.7.9 R is a eld. Proof. The only parts left to prove are that multiplication is associative and that it distributes over addition. Let x = (L, R), y = (L , R ), z = (L , R ) be in R. To prove that x (y z ) = (x y ) z , it suces to prove that the additive inverses of the two products are identical. By Lemma 3.7.8, we may then in each product replace x = (L, R) by x = (R, L), and similarly for y by y and z by z . Note that R does not contain only positive numbers if and only if L contains only positive numbers, so via these replacements we may assume that R, R , R all contain only positive elements. But then x (y z ) = ( , L (L L )), which is by associativity of multiplication on Q the same as ( , (L L ) L ) = (x y ) z . This proves associativity of multiplication. To prove that x (y + z ) = (x y ) + (x z ), we may similarly assume that R and R + R contain only positive numbers. Then either R or R contains only positive elements, and by commutativity of addition we may assume that R contains only positive elements. From assumptions we have x (y + z ) = ( , R (R + R )). If a R, b R and c R , then a (b + c) = a b + a c by distributivity in Q, which proves that R (R + R ) R R + R R . Suppse that in addition R contains only positive numbers. Let r1 , r2 R, r R , r R . Then (r1 r + r2 r )/(r + r ) is a rational number. If (r1 r + r2 r )/(r + r ) < r1 , then by compatibility of multiplication with < on Q, r1 r + r2 r < r1 r + r1 r , whence 0 < (r1 r2 ) r , which forces r1 > r2 . If in addition (r1 r + r2 r )/(r + r ) < r2 , then we similarly get that r2 > r1 , which is a contradiction. So (r1 r + r2 r )/(r + r ) is at least as large as min {r1 , r2 }, and so (r1 r + r2 r )/(r + r ) R by Exercise 3.7.4. This proves that an arbitrary element of R R + R R is in R (R + R ), which proves that R (R + R ) = R R + R R . Thus x (y + z ) = ( , R R + R R ) = ( , R R ) + ( , R R ) x y + x z. Now suppose that R does not contain only positive numbers. Then L contains only positive numbers. Thus by the previous case x (y + z ) + x (z ) = x ((y + z ) + (z )). By the already established associativity of addition, this is x y . But by Lemma 3.7.8 this says that x (y + z ) = x y + x z . The Dedekind cuts as in Proposition 3.7.4 are the familiar rational numbers, and if we establish some good order on R, we should be able to order the other Dedekind cuts among the rational ones as expected, to get that R consists of the familiar real numbers.
Section 3.7: Construction of the eld R of real numbers
3.7.1 Prove that there exists no R Q such that (Q {0}, R) is a Dedekind cut. Similarly, there exists no L Q such that (L, Q+ {0}) is a Dedekind cut.
3.7.2 Let (L, R) be a Dedekind cut. i) Prove that L = {l Q : there exists l Q such that for all r R, l < l < r }, R = {r Q : there exists r Q such that for all l L, l < r < r }. In other words, either L or R uniquely determines the Dedekind cut. ii) If m Q is the supremum of L, prove that R = Q \ (L {m}). If L does not have a supremum in Q, prove that R = Q \ L. iii) If m Q is the inmum of R, prove that L = Q \ (R {m}). If R does not have an inmum in Q, prove that L = Q \ R. 3.7.3 Prove that ({r Q : r 2 < 2 or r < 0}, ({l Q : l2 2 and l 0}) is a Dedekind cut. Prove that it is not of the form Dm = ((, m), (m, )) for any rational number m. 3.7.4 Let (L, R) be a Dedekind cut, l L, r R, and x Q. Prove the following: i) If x l, then x L. ii) If x r , then x R.
3.7.5 If m and n are rational numbers, prove that Dm + Dn = Dm+n . (Hint: write Dm = (Lm , Rm ), Dn = (Ln , Rn ), and prove that Lm + Ln = Lm+n and Rm + Rn = Rm+n .) 3.7.6 Let x = ({x Q : x < 0 and x2 > 2}, {x Q : x2 2 or x > 0}), y = ({x Q : x < 0 or x2 < 3}, {x Q : x2 3 and x > 0}). Compute x + y , x y .
3.7.7 Prove Theorem 3.7.7.
3.7.10 Let (L, R) be a Dedekind cut. Prove that there exists z Q such that z R and z L. Prove in addition that for any a Q, there exists such z with z |a|.
3.7.9 This exercise proves that multiplication on Dedekind cuts is well-dened. Let (L1 , R1 ), (L2 , R2 ) be Dedekind cuts. i) Suppose that L1 and L2 contain positive elements of Q. Prove that there exists L Q such that (L, R1 R2 ) is a Dedekind cut. ii) Suppose that L1 but not L2 contains positive elements of Q. Prove that there exists R Q such that (R1 L2 , R) is a Dedekind cut. iii) Suppose that L1 but not L2 contains positive elements of Q. Prove that there exists R Q such that (L1 R2 , R) is a Dedekind cut. iv) Suppose that neither L1 nor L2 contains positive elements of Q. Prove that there exists L Q such that (L, L1 L2 ) is a Dedekind cut.
3.7.8 Prove that if (L1 , R1 ), . . . , (Ln, Rn ) are nitely many Dedekind cuts, then n (n k=1 Lk , k=1 Rk ) is a Dedekind cut. Furthermore, this Dedekind cut equals (Lk , Rk ) for some k {1, . . . , n}.
97
3.7.11 For any set S Q, dene S+ = S Q+ and S = S Q . Prove that for any Dedekind sets (L, R), (L , R ),
(L, R) (L, R ) = (L R+ + R+ L + L+ L+ + R R , L L + R L+ + L+ R + R+ R+ ),
where it is understood that A + = A.
*3.7.13 Let x, y R such that x3 = y 3 . Prove that x = y .
*3.7.12 Let x, y R such that x2 = y 2 . Prove that either x = y or x = y .
3.8
Order on R, the least upper/greatest lower bound theorem
Denition 3.8.1 We impose the following order on R: (L, R) (L , R ) if L L , and (L, R) < (L , R ) if L L . Theorem 3.8.2 R is an ordered eld. Proof. By Theorem 3.7.9, R is a eld. Let (L, R), (L, R ), (L , R ) R. Sets L, L can be in exactly on of the following relationships: L = L , L L , L L . If L = L , then by Exercise 3.7.2, (L, R) = (L , R ). If L L , then by denition (L, R) < (L , R ). If L L , let l L \ L . Let l L . By Exercise 3.7.4 applied to the Dedekind cut (L, R), l > l and l L. This proves that L L, and since L = L, it follows that L L, whence (L , R ) < (L, R). This proves trichotomy of <. Suppose that (L, R) < (L , R ) and (L , R ) < (L , R ). Then by denition, L L L , so that L L , which says that (L, R) < (L , R ). Thus transitivity holds for <. Suppose that (L, R) < (L , R ). If L L , then certainly L + L L + L . Suppose for contradiction that L + L = L + L . Then L + L + (R ) = L + L + (R ), or in other words, by associativity of addition on Q and by Theorem 3.7.7, L +(, 0) = L +(, 0). Let m L \ L. Then for all k (, 0), m + k L + (, 0), or in other words, m + k is strictly smaller than some element of L. Thus by Exercise 3.7.4, all rational numbers strictly smaller than m are in L. Since m L, it follows that (L, R) = Dm . By condition (3) of Dedekind cuts, there exists n L such that m < n. But then (m + n)/2 = n + (m n)/2 L + (, 0) = L + (, 0), yet at the same time (m + n)/2 is not in L + (, 0) = (, m) as (m + n)/2 = m + (n m)/2 is strictly larger than m. This gives the desired contradiction, and proves that < is compatible with addition. Finally, suppose that (L, R) < (L , R ) and that D0 < (L , R ). Then D0 < (L , R ) (L, R) = (L , R ) + (R, L) = (L R, R L), so that by denition of multiplication, ((L , R ) (L, R)) (L, R ) = ( , (R L) R ), so that D0 < ((L , R ) (L, R)) (L, R ). But then by associativity, D0 < ((L , R ) (L, R )) ((L, R) (L, R )), so that by compatibility of < with addition, (L, R) (L, R ) = D0 + ((L, R) (L, R )), = (L , R ) (L , R ), which proves that < is compatible with multiplication by positive numbers.
98
Section 3.8: Order on R, the least upper/greatest lower bound theorem
Thus the following terms and facts, developed in Section 2.7, apply to elements of R: positive, negative, non-positive, non-negative, intervals, absolute value, triangle inequality, reverse triangle inequality. One of the main properties that distinguishes R from Q is the following theorem: Theorem 3.8.3 (Least upper/greatest lower bound theorem) Let T be a non-empty subset of R that is bounded above (resp. below). Then sup (T ) (resp. inf (T )) exists in R. Proof. First suppose that T is bounded above. Let U = (L,R)T L. In other words, U is the union of all the left parts of all the Dedekind cuts in T . Since T is not empty, neither is U . Now let (U0 , V0 ) R be an upper bound on T . Then by denition L U0 for all (L, R) T . Thus U U0 . This implies that U is a subset of Q that does not contain at least 2 elements of Q. By Exercise 3.7.2, let V Q be such that (U, V ) is a Dedekind cut. By denition, (U, V ) is an upper bound for T , and this upper bound is less than or equal to an arbitrary upper bound (U0 , V0 ). This proves that sup (T ) exists in R. The proof of inmum is left for Exercise 3.8.2. Remark 3.8.4 In contrast, note that the non-empty bounded set {x Q : x2 < 2} has no inmum nor maximum in Q. But this same set, as a subset of R, has inmum ({x Q : x2 > 2}, ) and supremum ( , {x Q+ : x2 > 2}) in R. Theorem 3.8.5 (Archimedean property) Let m, n R. If m > 0, there exists p N such that n < pm. Proof. Write m = (L, R), n = (L , R ). Since L R is either Q or misses one element of Q, it follows that some positive integer x is in L R . Let y R , and z = max {x, y }. Then by Exercise 3.7.4, z is a positive element of R . By assumption (, 0) L, so there exists a non-negative l L. Either l is 0 or it is positive. If l = 0, then by criterion (3) of Dedekind cuts, L contains a positive elements as well. Thus we may assume that l > 0. By Theorem 3.6.7, there exists p N such that r < pl. Then L (, r ) (, pl) pL, so that n = (L , R ) < Dp (L, R) = p m. Theorem 3.8.6 For any two distinct real numbers there is a rational number strictly between them. Proof. Let a, b R with a < b. Write a = (L, R) and b = (L , R ) as Dedekind cuts. Since a < b, it means that L L . Let r L \ L. Then a < Dr < b, which by identication r = Dr proves the theorem. The following is a rephrasing of Theorem 2.7.13 in this context: Suppose that a, b R have the property that |a b| < for all real numbers > 0. Then a = b.
Chapter 3: Construction of the basic number systems Exercises for Section 3.8
99
3.8.1 Prove that the following conditions are equivalent on Dedekind cuts (L, R), (L , R ): i) (L, R) (L , R ). ii) For all l L there exists l L such that l l . iii) For all l L there exists l L such that l < l . (Hint: condition (3) of Dedekind cuts.) 3.8.2 Let T be a non-empty subset of R that is bounded below. Let (U, V ) be the Dedekind cut with U = (L,R)T (R). Prove that (U, V ) is the inmum of T .
3.8.3 Let m R be a positive number. Prove that there exists a positive integer N such that 1/2N < m. (Hint: Exercise 3.6.6.)
3.9
Complex numbers
Note that there is no real number x such that x2 = 1. In this section we build the smallest possible eld containing R with an element i such that i2 = 1. By Exercise 2.6.7 also (i)2 = i2 = 1. Denition 3.9.1 Let C = R R (the Cartesian product). Dene two binary operations on C: (a, b) + (c, d) = (a + c, b + d), (a, b) (c, d) = (ac bd, ad + bc). Elements of C are called complex numbers. The following are all easy to verify, and the reader is encouraged to do so: (1) , + are associative and commutative. (2) distributes over +. (3) For all x C, (0, 0) + x = x. In other words, C has the additive identity (0, 0). (4) For all (a, b) C, (a, b) + (a, b) = (0, 0). In other words, every element has an additive inverse. By Theorem 2.5.10, the additive inverse is unique. (5) For all x C, (1, 0) x = x. In other words, C has a multiplicative identity (1, 0). (6) (1, 0) = (0, 0). a a (7) For all (a, b) = (0, 0), ( a2 + , b ) C and ( a2 + , b ) (a, b) = (1, 0). In other b 2 a2 + b 2 b 2 a2 + b 2 words, every non-zero element has a multiplicative inverse. By Theorem 2.5.10, the multiplicative inverse is unique. (8) C is a eld. As usual for elds, the additive inverse of x is denoted x, the multi1 . plicative inverse as x1 = x
100
Section 3.9: Complex numbers
(9) (0, 1)2 = (1, 0) and (0, 1)2 = (1, 0). (10) If (a, b)2 = (1, 0), then (a, b) = (0, 1) or (a, b) = (0, 1). Notation 3.9.2 We now switch to the more conventional notation for elements of C: (a, b) = a + bi, with (a, 0) = a and (0, b) = bi. This notational convention does not lose any information, but it does save a few writing strokes. In this notation, the additive identity in C is 0, the multiplicative identity is 1, and 2 i = (1)2 = 1. There is a natural injective function : R C given by (r ) = (r, 0) = r , with the property for all r, s R, (r s) = (r ) (s) and (r + s) = (r ) + (s). In total we have found the following natural inclusions that are compatible with addition and multiplication: N0 Z Q R C. These number systems progressively contain more numbers and more solutions of equations. For example, the equation 1 + x = 0 does not have any solutions in N but it does have them in Z; the equation 1 + 2x = 0 does not have any solutions in Z, but it does have them in Q; the equation 2 x2 = 0 does not have any solutions in Q, but it does have them in R; the equation 2 + x2 = 0 does not have any solutions in R, but it does have them in C. The eld C has many excellent properties, some of which we explore in these notes. Since C = R R, we can represent complex numbers in the real plane, but we can label them either with the pair-notation or with the new notation, like so:
(3, 1) = 3 + i (1, 0) = 1
(0, 1) = i (0, 0) = 0 (1, 0) = 1 (0, 1) = i (2, 0.5) = 2 + 0.5i
Note that the real numbers all lie on a horizontal line. This axis is called the real axis. The vertical axis of numbers whose rst entry is 0 is called the imaginary axis. By the Pythagorean theorem, it makes sense to dene the size of a + bi as a2 + b2 . Because of this two-dimensional quality of C it is perhaps intuitively obvious that C is not an ordered eld in the sense of Denition 2.7.8. A rigorous proof is left for Exercise 3.9.6.
Chapter 3: Construction of the basic number systems Exercises for Section 3.9 3.9.1 Write the following elements in the form a + bi with a, b R: i)(73i) i) (4+ (1i)(3+2i) . ii) (2 3i)5 . (Hint: Exercise 1.6.6.) 3.9.2 Let =
3 2
101
i +2 . Draw , 2 , 3 , 4 , 5 , 6 , 7 . What do you observe?
3.9.3 Draw the following sets in C: i) {x : |x| = 3}. ii) {x : the real part of x is 3}. iii) {x : the imaginary part of x is 3}. iv) {x : the product of the real and imaginary parts of x is 3}. 3.9.5 Prove that the only x C with x2 = 0 is x = 0.
3.9.4 Let x C such that x2 = 1. Prove that either x = i or x = i. 3.9.6 Prove that C is not ordered in the sense of Denition 2.7.8. You may not use non-facts such as 0 < 1.
3.10
Absolute value in C
Despite the fact that C is not ordered (Exercise 3.9.6), we can still talk about sizes of complex numbers and bounded subsets of C, because the absolute value function is still dened on C. The motivation for the denition below comes from the Pythagorean theorem: the distance between (0, 0) and (a, b) is a2 + b2 R. Denition 3.10.1 The norm, length, or the absolute value, of a complex number a + bi, is |a + bi| = a2 + b2 R. The real part of a + bi is Re(a + bi) = a, and the imaginary part is Im(a + bi) = b. The complex conjugate of a + bi, is a + bi = a bi C. This gives have a way to partially compare complex numbers, say by their sizes, or by their real components. (And we reiterate Exercise 3.9.6: C is not an ordered eld.) The absolute value of (a, 0) = a or (0, a) = ia is |a|. The absolute values of (1, 1) = 1+ i is 2, the absolute values of (1, 2) = 1+i 2 is 3, the absolute values of (1, 3) = 1+i 3 is 4 = 2, et cetera. (In this we rediscovered what the Pythagoreans already knew: the square root of every positive integer can be constructed from right triangles. Namely, 2 is the hypothenuse of the right triangle with other two sides both of length 1, 3 is the hypothenuse of the right triangle with other two sides having lengths 1 and 2 which we already know how to construct, et cetera.)
102
Section 3.10: Absolute value in C
Proof. Write x = a + bi, y = c + di. Then |x| = a2 + b2 = a2 + (b)2 = |x|, which proves (1). Since R is an ordered eld, by Exercise 2.7.12, a2 , b2 0, so by Exercise 2.7.13, a2 + b2 = 0 if and only if a = b = 0. This proves (2). Certainly x = x if and only if a + bi = a bi, which is the same as saying that (a, b) = (a, b), i.e., that b = 0. Thus x = x if and only if x = a R. Also, x = 0 if and only if (a, b) = (0, 0), which is the same as saying that x = (a, b) = (0, 0). This proves (3). We have that x x = (a + bi)(a bi) = a2 (bi)2 = a2 + b2 , which is the sum of two non-negative real numbers, and is thus non-negative. This proves (4). Since b2 0, it follows that a2 a2 + b2 , so that by Corollary 2.8.3, |a| = a2 a2 + b2 . This proves that | Re x| |a|. Similarly | Im x| |a|. This proves (5). If x = 0, then by (5), |x| is a non-zero (real, complex) number, and by (4), x/|x|2 is the multiplicative inverse of x. This proves (6). Addition in (7) is straightforward, and x y = (a + bi) (c + di)
Theorem 3.10.2 Let x, y C. Then (1) |x| = |x|. (2) x = 0 if and only if |x| = 0. (3) x = x if and only if x R; and x = 0 if and only if x = 0. (4) x x = |x|2 is a non-negative real number. (5) | Im x|, | Re x| |x|. (6) If x = 0, then x1 = x/|x|2 . (7) x + y = x + y . (8) x y = x y . (9) If y = 0, then (x/y ) = x/y . (10) |xy | = |x| |y |. (11) (Triangle inequality) |x y | |x| + |y |.
= ac bd + (ad + bc)i = (a bi) (c di) =xy
= ac bd (ad + bc)i
proves (8). If y = 0, then by (8), x = (x/y )y = x/y y , and so (9) follows. With this we have |xy |2 = (xy )(xy ) (by (4)) = xyxy (by (8))
Chapter 3: Construction of the basic number systems = xxyy (by associativity and commutativity of in C)
103
= |x|2 |y |2 (by (4))
= (|x||y |)2 (by associativity and commutativity of in R),
which proves that |xy | = |x||y |, i.e., it proves (10). Finally, |x y | = (x y )(x y ) = xx xy yx + yy = (x y )(x y )
2
= |x|2 xy xy + |y |2 = |x|2 2 Re xy + |y |2 |x|2 + 2|xy| + |y |2 (comparison of real numbers by (5))
= |x|2 + 2|x||y| + |y |2 (by (10)) = |x|2 + 2|x||y | + |y |2 (by (1)) = ( | x| + | y | ) 2
proves (11). The absolute value allows the denition of bounded sets in C: Denition 3.10.3 A subset A of C is bounded if there exists a positive number M such that for all x A, |x| < M . For example, any set with only nitely many elements is bounded. The subset Z of C is not bounded. The innite set {x C : |x| = 5} is bounded. The set {in : n N+ } is bounded. The set {1/n : n N+ } is bounded. The set {x C : the angle of x counterclockwise from the positive x-axis is /3} is not bounded. (You may wish to draw these sets.) So far we have expressed complex numbers with pairs of real numbers (in so-called Cartesian coordinates, named after Ren e Descartes (15961650)). The absolute value, together with the angle, provide an alternate expression, via so-called polar coordinates. Note that angles are not uniquely determined: for example, 0 and 2 give the same points. Namely, a point x = (a, b) = a + bi C with |x| = r lies lie on the circle centered at 0 = (0, 0) with radius r . Among all points on this circle, the angle counterclockwise from the positive real axis uniquely identies it. We physically identify the angle 2 with 0, 3/2 with /2, et cetera. Note that the angle is /2 precisely when Re x = 0, that it is 0 when x is a positive real number, that it is when x is a negative real number, et
104
cetera. Furthermore, if the angle is not /2, then the tangent of this angle is precisely Im x/ Re x. We draw a few examples:
radius 1, angle /2 radius 1, angle /4 radius 2, angle radius 2, angle 0 radius 1, angle /2
radius 2, angle /3
Theorem 3.10.4 (Fun fact) Let z be a complex number, of length |z | and with angle counterclockwise from the positive real axis. (If z = 0, arbitrarily set = 0.) Let f, g, h : C C be dened by f (x) = zx = multiplication by z, h(x) = rotate x by angle counterclockwise with center at (0, 0). Then f = g h = h g, or in other words, multiplication by z is the same as stretching by |z | followed by or preceded by rotating by the angle counterclockwise. Proof. First of all, by geometry of points on the plane, rst rotating by an angle and then stretching by a factor is the same as rst stretching by the same factor and then rotating by the angle. So it suces to prove one of the equalities. If z = 0, then f (x) = 0 = (h g )(x), so that the theorem holds in this case. So we may assume that z = 0. Similarly we may assume that x = 0. Write z = (c, d) for some c, d R. If x is a positive real number, then f (x) = zx = (xc, xd). The angle of this complex number is the same as the angle of z , and the length of f (x) is |z |x. Thus (hg )(x) = h(|z |x), which is the point in C of length |z |x and at angle counterclockwise from the positive x-axis. But then (h g )(x) is precisely (xc, xd) = f (x), which proves the theorem in case x is positive. If x is a negative real number, then f (x) = zx = (xc, xd). Because x is negative, the angle of this complex number is the same as the angle of z plus , and the length of g (x) = |z |x = stretch x by a factor of |z |,
105
f (x) is |z |x. Thus (h g )(x) = h(|z |x), which is the point in C of length |zx| and at angle + counterclockwise from the positive x-axis. But then (h g )(x) is precisely f (x), which nishes the proof of the theorem in case x is a real number. Observe that f (i) = (d, c). Thus the stretch from i to f (i) is by a factor of |(d, c)| = (d)2 + c2 = c2 + d2 = |(c, d)| = |z |. Draw the graph with points (c, d) and (d, c) to convince yourself that to get from the former to the latter you simply rotate counterclockwise by angle /2. proves the theorem for x = i. [So far we have proved the theorem for all real x and x = i. Good job! But infinitely many more cases to go?! We do not have energy and time for that! Ah, but do not despair, we have a nice shortcut coming.] In general, if a, b R and x = a + bi, then f (x) = za + zbi = za + b(zi) = f (a) + bf (i) = (h g )(a) + b(h g )(i) (by the established cases above) = |z |h(a) + b|z |h(i) (by the denition of g ) = (g h)(a) + b(g h)(i) (by the geometry as asserted earlier) = |z | (h(a) + bh(i)) (by commutativity and distributivity of multiplication in C) = g (h(a + bi)) = (g h)(x). Theorem 3.10.5 For any complex non-zero number x and any integer n, the angle of xn counterclockwise away from the positive x-axis is n times the angle of x. Proof. If n = 1, this is trivially true. Now suppose that the theorem is true for some positive integer n. Then the angle of xn1 counterclockwise away from the positive x-axis is n 1 times the angle of x, and by Theorem 3.10.4, the angle of xn = xxn1 is the sum of the angles of x and xn1 , so that it is n times the angle of x. Thus by induction the theorem is proved for all positive n. Still keep n positive. Since 1 = xn xn has angle 0 and xn has has angle n times the angle of x, by Theorem 3.10.4, xn must have angle n times the angle of x. Thus the theorem holds for all non-zero n. Finally, if n = 0, then then angle of xn = 1 is 0, which is 0 times the angle of x. Thus for example, since
1+i is on the unit circle at angle 4/3 away x-axis, we have that the second power of 2 from the positive x-axis, and the cube power is on the unit circle at angle 2 , i.e., at angle 1+i 3 0, so that ( ) = 1. 2 1+i 2
= |z |h(a + bi) (by the geometry of rotation)
is on the unit circle at angle 2/3 away from the positive
3.10.1 Write the following elements in the form a + bi with a, b R: i) Complex number of size 1 and at angle /4 counterclockwise from the positive real axis. ii) Complex number of size 1 and at angle /4 counterclockwise from the positive real axis. iii) The product and the sum of the numbers from the previous two parts. 3.10.2 Draw the following points in the real plane (= complex line), and think about Theorem 3.10.4: i) 3 2i, i(3 2i), ii) 2 i, i(2 i). iii) 2 3i, (1 + i)(2 3i). iv) 1 i, (1 + i)(1 i). 3.10.3 Draw the following sets in C: i) {x : the angle of x counterclockwise from the positive real axis is /3}. ii) {x : |x| = 3}. iii) {x : |x 2 + i| = 3}. 3.10.4 Prove that for any non-zero z C there exist exactly two elements in C whose square equals z . (Hint: Theorem 3.10.4.) 3.10.5 Square the pentagon drawn below. Namely, estimate the coordinates (real, imaginary or size, angle) of various points on the pentagon, square the point, and draw its image on a dierent real plane. You need to plot the image not only of the ve vertices, but of several representative points from each side. (Hint: you may want to use Theorem 3.10.4.)
107
3.10.7 Let a C, let B be a positive real number, and let A = {x C : |x a| B }. Draw such a set in the complex plane assuming a = 0, and prove that A is a bounded set. 3.10.8 Let a, b C. Suppose that for all real numbers > 0, |a b| < . Prove that a = b. (Hint: Theorem 2.7.13.)
3.10.6 Deduce the reverse triangle inequality from triangle inequality: for all x, y C, ||x| |y || |x y |.
3.11
Topology of the constructed elds
When reading this section, absorb the following main points: open ball, open set, limit point, closed set. Everything else is for extra enjoyment. The main object of this section is to introduce limit points of sets so that we can in subsequent chapters talk about limits of functions, sequences, series, and so on. Limit points are part of a study of a branch of mathematics called topology. The main concept in that branch of mathematics is the notion of a topology on a set, which simply means that we declare some sets open subject to the conditions that the empty set and the whole set have to be open, that arbitrary unions of open sets be open, and that nite intersections of open sets be open. In any topology, the complement of an open set is called closed. A topology can be imposed on any set, not just R or C. In these notes we focus on F = R or F = C, and our starting point is not the abstract topology, but we declare some sets open, and develop the properties of topology in Theorem 3.11.13. This particular topology is the standard, or Euclidean topology. Denition 3.11.1 Let a F and let r be a positive real number. An open ball with center a and radius r is a set of the form B (a, r ) = {x F : |x a| < r }. An open set in F is any set that can be written as a union of open balls. The following are both B (0, 1), but the left one is a ball in R and the right one is a ball in C. Note that they are dierent: by denition the left set is an open subset of R, but if you think of it as a subset of C, it is not open (see Exercise 3.11.1).
108
Section 3.11: Topology of the constructed elds
Examples 3.11.2 (1) B (a, r ) is open. (2) F = aF B (a, 1) is an open set. (3) The empty set is an open set because it is vacuously a union of open sets (see page 38). (4) For real numbers a < b, an interval in R of the form (a, b) is an open set because it is equal to B ((a + b)/2, (b a)/2). The interval (a, ) is open because it equals n=1 B (a + n, 1). (5) The set A = {x C : Re x < 1, Im x < 2} is open in C. Namely, this set is the union aA B (a, min {1 Re x, 2 Im x}. Theorem 3.11.3 B (a, r ) has innitely many points. Proof. For each integer n 2, a + r/n B (a, r ). Since r > 0, these numbers are all distinct. Since N is innite, so is the set of all integers that are at least 2. Example 3.11.4 Thus if A is an open subset of F , then either A is empty or A has innitely many points. In particular, {a} is not open. Denition 3.11.5 Let A be an arbitrary subset of F and a F . If there exists r > 0 such that B (a, r ) A, then a is called an interior point of A. If for every r > 0, B (a, r ) A = and B (a, r ) (F \ A) = , then a is called a boundary point of A. The set of all interior points of A is denoted Int A, and the set of all boundary points of A is denoted Bd A. Note that Int A is automatically a subset of A. Examples 3.11.6 (1) Int F = F , Bd F = . (2) Int = , Bd = . (3) Int {a} = , Bd {a} = {a}. (4) Int Z = , Bd Z = Z. (5) Int Q = , Bd Q = R. This shows that Bd A need not be a subset of A. (6) Int (R \ Q) = , Bd (R \ Q) = R. Theorem 3.11.7 Int B (a, r ) = B (a, r ). Proof. It suces to prove that every point b of B (a, r ) is an interior point. Namely, |b a| < r . Let x B (b, r |b a|). Then by triangle inequality |x a| = |x b + b a| |x b! + |b a| < (r |b a|) + |b a| = r , so that x B (a, r ). We just proved that B (b, r |b a|) B (a, r ), which proves that b Int A.
109
Denition 3.11.8 Let A be an arbitrary subset of F and a F (not necessarily in A). We say that a is a limit point of A if for all real numbers s > 0, B (a, s) contains elements of A dierent from a. Theorem 3.11.9 Any interior point of A is a limit point of A. Proof. Let a Int A. Then there exists r > 0 such that B (a, r ) A. Now let s be an arbitrary positive real number. By Theorem 3.11.3, B (a, min {r, s}) contains innitely many points, and they are all elements of A, so that B (a, min {r, s}) contains at least one point of A dierent from a. But B (a, min {r, s}) B (a, s), so that B (a, s) contains at least one point of of A dierent from a. Thus a is a limit point of A. Theorem 3.11.10 If a is a limit point of A, then a A Bd A. Proof. To prove that a A Bd A, it suces to prove that if a A, then a Bd A. So assume that a A. Since a is a limit point of A, by denition for all positive real numbers s > 0, B (a, s) contains a point of A other than a. Thus B (a, s) A is not empty. But also B (a, s) (F \ A) is not empty because a is in this intersection. Thus a Bd A. Examples 3.11.11 (1) If A = {a}, then Bd A = {a}, but the set of limit points of A is the empty set. (2) If A = Q, then the set of limit points of A is R. In the rest of the section we develop more topology. Most of it is not needed in the rest of the book, but since I started to talk about topology, a section would not be complete without the rest of this section. Theorem 3.11.12 A set A is open if and only if A = Int A. Proof. Suppose that A is open. We need to prove that A Int A. Let a A. Since A is open, it is a union of open balls, so that a is in one of those balls. Namely, a B (b, r ) A for some b A and positive real number r . Then |a b| < r . Set d = r |a b|. Then d is a positive real number. If y B (a, d), then by the triangle inequality, |y b| = |y a + a b| |y a| + |a b| < d + |a b| = r , so that y B (b, r ). Thus B (a, d) B (b, r ) A, which proves that the arbitrary element a of A is in Int A. Now suppose that A = Int A. Then for all a A there exists ra > 0 such that B (a, ra ) A. Then A aA B (a, ra ) A, so that equality holds and A is a union of open balls, so that A is open. Theorem 3.11.13 (Topology on F ) (1) and F are open. (2) An arbitrary union of open sets is open. (3) A nite intersection of open sets is open.
110
Proof. The empty set can be written as an empty union of open balls, so it is open vacuously, and F = aF B (a, 1), so that F is open. Every open set is a union of open balls, and so the union of open sets is a union of open balls, hence open. If A1 , . . . , An are open sets and a A, then for all i = 1, . . . , n, there exists a positive real number ri such that B (a, ri ) Ai . Set r = min {r1 , . . . , rn }. Then r is a positive real number, and by construction B (a, r ) A1 An . Thus a is in the interior of the intersection of these nitely many sets. Since a was arbitrary in the intersection, this proves that the intersection is open. Note that an arbitrary intersection of open sets need not be open: n=1 B (a, 1/n) = {a}, which is not open. Denition 3.11.14 A set A is closed set if Bd A A. Theorem 3.11.15 A is open if and only if F \ A is closed. Proof. Suppose that A is open. Let x Bd (F \ A). We need to prove that x A. By assumption for all r > 0, B (x, r ) (F \ A) is not empty, so that no ball centered at x is contained in A. Hence x Int A, and since A is open, then x A. Now suppose that F \ A is closed. Let x A. We need to prove that x Int A. If x Int A, then for every positive real number r , B (x, r ) is not contained in A. Thus B (x, r ) contains both points of F \ (F \ A) = A (namely x) and points of F \ A. Thus x Bd (F \ A), so by assumption x F \ A, which contradicts the assumption x A. Thus x must be in Int A. The following is now almost immediate from previous results: Theorem 3.11.16 (Topology on F ) (1) , F are closed sets. (2) Arbitrary intersections of closed sets are closed. (3) Finite unions of closed sets are closed. Proof. The last two parts follow from the last two theorems because F\ Ak =
kI kI
(F \ Ak ),
F\
Ak =
kI kI
(F \ Ak ).
Both and F are open and closed, and these turn out to be the only sets that are both open and closed (see Exercise 3.11.2). Some sets are neither open or closed (see Exercise 3.11.1). Denition 3.11.17 For any set A, its closure A is the smallest closed set containing A.
Chapter 3: Construction of the basic number systems Theorem 3.11.18 A = A Bd A.
111
Proof. Certainly A A by denition. Let x Bd A, and let B be a closed set that contains A. If x B , then by the second part in Exercise 3.11.13, x Bd B . But B is closed, so Bd B B , so that the assumption x B brings to the contradiction x B . So the assumption x B is false (and let us very astray), which means that x B . Thus an arbitrary element of Bd A is contained in an arbitrary closed set containing A, so that Bd A A. Thus A Bd A A. Now we prove that A Bd A is closed. It suces to prove that the complement is open. So let x F \ (A Bd A). Since x Bd A, r > 0 such that either B (x, r ) A = or B (x, r ) (F \ A) = . The latter case is impossible because x B (x, r ) (F \ A). So necessarily B (x, r ) A = . Let y B (x, r ). Then by triangle inequality B (y, r |y x|) B (x, r ), so that y Bd A. This proves that B (x, r ) (A Bd A) is empty, so that B (x, r ) F \ (A Bd A). Since x was arbitrary, this proves that F \ (A Bd A) is open, so that A Bd A is closed. Thus we have found a closed seet A Bd A that contains A and is contained in every closed set containing A. Thus A Bd A is the smallest closed set containing A, and thus by the denition A Bd A equals A. Example 3.11.19 The boundary of the open ball B (a, r ) is {x C : |x a| = r }. Thus the closure of B (a, r ) is {x C : |x a| r }. Proof. Suppose that x C satises |x a| = r . Let > 0. Set = min {, 1}. Set y = x + (a x)/(2r ), z = x (a x)/(2r ). Then |y x|, |z x| have lengths | (a x)/(2r )| = | /2| < , |y a| = |x a + (a x)/(2r )| = |(x a)(1 /(2r ))| < |x a| = r , |y a| = |x a + (a x)/(2r )| = |(x a)(1 + /(2r ))| > |x a| = r , so that y B (x, ) B (x, r ), and z B (x, ) (C \ B (x, r )). This proves that x Bd B (a, r ). Now let x C. Suppose that |x a| < r . Then r |x a| > 0. Let y B (x, r |x a|). Then by triangle inequality |y a| = |y x+xa| |y x|+|xa| < (r |xa|)+|xa| = r , so that y B (a, r ). This proves that B (x, r |x a|) B (a, r ), which means that B (x, r |x a|) (C \ B (a, r )) = , so that x Bd B (a, r ). Now suppose that |x a| > r . Then |x a| r > 0. Let y B (x, |x a| r ). Then by reverse triangle inequality |y a| = |y x + x a| |x a| |y x| > r (r |x a|) = |x a| > r , so that y C \ B (a, r ). This proves that B (x, |x a| r ) C \ B (a, r ), which means that B (x, |x a| r ) B (a, r ) = , so that x Bd B (a, r ). We just proved that if x C such that |x a| = r , then x Bd B (a, r ), so that Bd B (a, r ) {x C : |x a| = r }. By the rst paragraph then Bd B (a, r ) = {x C : |x a| = r }.
3.11.1 Let A be B (0, 1) be an open ball in R. Since R is a subset of C, then A is also a subset of C. Prove that A is neither a closed nor an open subset of C. 3.11.2 Let F be either R or C, and let A be a closed and open subset of F . Prove that A = or A = F . 3.11.3 Prove that Q is neither an open nor a closed subset of R or C. 3.11.4 Sketch the following subsets of C. Determine their interior and boundary sets, the set of limit points, and whether the sets are open, closed, or neither: {x C : Im x = 0, 0 < Re x < 1}, {x C : Im x = 0, 0 Re x 1}, {x C : 2 Im x 2, 0 Re x 1}, {x C : 2 Im x 2, 0 Re x 1}, {1/n : n N+ }. 3.11.5 For each of the following intervals as subsets of R, determine the interior and boundary sets, the set of limit points, and whether the sets are open, closed, or neither: (0, 1), [0, 1], [3, 5), [3, ). 3.11.7 Let a be a limit point of a set A. Suppose that a set B contains A. Prove that a is a limit point of B . 3.11.8 Give examples of sets A B C and a C such that a is a limit point of B but not of A.
1 1 : n N+ } is {0} { n : n N+ }. 3.11.10 Prove that the closure of { n
3.11.6 Find the boundary and the set of limit points of {x C : Re x < 0}.
3.11.9 Prove that the closure of B (a, r ) is {x F : |x a| r }.
3.11.11 Prove that the closure of Q in R and in C is R.
3.11.14 Recall the denition of bounded sets Denition 2.7.2. Let A R be bounded below (resp. above). Prove that inf A A (resp. sup A A. *3.11.15 Prove that a set A is closed if and only if A = A. 3.11.16 Let A be a subset of F and a F \ Int A. Prove that a is a limit point of A if and only if a is a boundary point of A \ {a}.
3.11.13 Let A B F . i) Prove that Int A Int B . ii) Prove that (Bd A) \ B Bd B . iii) Find examples of A and B such that Bd A Bd B .
3.11.12 Let B = {a + bi : a, b Q}. Determine the closure of B in C.
3.11.17 Let A be a subset of R and let a complex number a be a limit point of A. Prove that a R.
113
3.12
Closed and bounded sets and open balls
Closed and bounded sets in C and R have many excellent properties we will for example see in Section 5.2 that when a good (say continuous) real-valued function has a closed and bounded domain, then that function achieves a maximum and minimum value, et cetera. and the concept of uniform continuity (introduced in Section 5.4) needs the following fairly technical results. Theorem 3.12.1 Let A be a closed and bounded subset of R, and for each a A let a be a positive number. Then there exist a positive number > 0 and a nite subset S of A such that for all x A there exists a S such that B (x, ) B (a, a ). In particular, A aS B (a, a ). Proof. First suppose that A = [l, u]. Let C be the subset of all c A for which there exist > 0 and a nite subset S of [l, c] such that for each x [l, c] there exists a S such that B (x, ) B (a, a ). In other words, C consists of all those c A for which the conclusion of the theorem holds for [l, c]. Note that l C , so that C is not empty. Since C A, C is bounded. Thus by the Upper bound theorem (Theorem 3.8.3), c = sup C is a real number. Necessarily c u. Since c = sup C , there exists b C such that c b < c /4. By the denition of C there exist > 0 and a nite subset S of [l, b] such that for each x [l, b] there exists a S such that B (x, ) B (a, a ). But then for each x [l, c + c /2] there exists a S {c} such that B (x, ) B (a, a ). Thus [l, min {c + c /2, u}] C , which proves that c C , and that c = sup C = u. Thus C = [l, u], which proves the theorem in case A = [l, u]. Now let A be an arbitrary closed and bounded set in R. Since A is bounded, there exist l, u R such that A [l, u]. For each x [l, u] \ A, since A is closed, there exists x > 0 such that B (x, x ) A = . Thus we now have x for all x [l, u]. By the rst case, there exist > 0 and a nite subset S of [l, u] such that for all x [l, u] there exists a S such that B (x, ) B (a, a ). In particular, if x A, there exists a S such that B (x, ) B (a, a ). But then necessarily a A, for otherwise B (a, a ) A = , which contradicts that x B (x, ) B (a, a ). The next result is a generalization of the previous one: we consider intervals in R as closed and bounded subsets of C, and we even allow vertical translations of such intervals, and we then prove a more general result: Theorem 3.12.2 Let A be a closed and bounded subset of R, let y R, and think of B = A {y } as a subset of C. For each b B let b be a positive number. Then there exist a positive number > 0 and a nite subset S of B such that for all x A [y , y + ] there
114
Section 3.12: Closed and bounded sets and open balls
exists b S such that we have the following containment of (two-dimensional) complex balls: B (x, ) B (b, b ). Proof. To each a A we set a = (a,y) /2. By Theorem 3.12.1, there exist > 0 and a nite subset S of A such that for all x A there exists a S such that we have the following containment of one-dimensional balls: B (x, ) B (a, a ). In particular, a . Set = /2, and set S = {(a, y ) : a S }.
Now let x A, z [y , y + ], and (c, d) B ((x, z ), ). Then |c x| = | Re((c + di) (x + zi))| < , and similarly |d z | < . By assumption B (x, ) B (x, ) B (a, a ) we get that |c a| < a , and by triangle inequality |d y | |d z | + |z y | < 2 . Thus |(c, d) (a, y )| = |c + di a yi| =
2 2 a + 4 = 2 2 a+
|c a|2 + |d y |2
2 2 a + a = a 2 < (a,y) ,
which proves that B ((x, z ), ) B ((a, y ), (a,y)). Theorem 3.12.3 Let A be a closed and bounded subset of C. For each a A let a be a positive number. Then there exist a positive number > 0 and a nite subset S of A such that for all x A there exists a S such that B (x, ) B (a, a ). In particular, A
aS
B (a, a ).
Let C be the subset of all c [N1 , N2 ] for which the conclusion of the theorem holds for [M1 , M2 ] [N1 , c]. By Theorem 3.12.2, C contains N1 , so that C is not empty. Since C is bounded above by N2 , by the Upper bound theorem (Theorem 3.8.3), c = sup C is a real number. Necessarily c [N1 , N2 ]. By Theorem 3.12.2, there exist 1 > 0 and a nite subset S1 [M1 , M2 ] {u}] such that for all x [M1 , M2 ] [u 1 , u + 1 ] there exists a S1 such that B (x, 1 ) B (a, a ). Since c = sup C , there exists b C such that c b < 1 /2. By denition of C , there exist 2 > 0 and a nite subset S2 [M1 , M2 ] [N1 , b] such that for all x [M1 , M2 ] [N1 , c] there exists a S2 such that B (x, 2 ) B (a, a ).
Proof. First assume that A = [M1 , M2 ] [N1 , N2 ].
Set = min {1 , 2 }. We just proved that for all x [M1 , M2 ] [N1 , c + 1 ] there exists a S1 S2 such that B (x, 2 ) B (a, a ). Thus [N1 , min {c + 1 , N2 }] C , so that c C , and since c = sup C , necessarily c = N2 . This proves the theorem in case A is a rectangle.
Now let A be an arbitrary closed and bounded set in C. By boundedness A [M1 , M2 ] [N1 , N2 ] for some numbers M1 , N2 , N1 , N2 R. For each x in this rectangle that is not in A, since A is closed, there exists x > 0 such that B (x, x ) A = . Thus we now have x for all x [M1 , M2 ] [N1 , N2 ]. By the rst case, there exist > 0 and a nite set S [M1 , M2 ] [N1 , N2 ] such that for all x [M1 , M2 ] [N1 , N2 ] there exists a S such that B (x, ) B (a, a ). In particular, if x A, there exists a [M1 , M2 ] [N1 , N2 ]
115
such that B (x, ) B (a, a ). But then necessarily a A, for otherwise B (a, a ) A = , which contradicts that x B (x, ) B (a, a ). Exercises for Section 3.12 3.12.1 Let A = [2, 3]. For each a A let a = 1/a. Does there exist > 0 as in Theorem 3.12.1? If yes, nd it, if no, explain why not. 3.12.2 Let A = {1/n : n N}. For each a A let a = a/3. Does there exist > 0 as in Theorem 3.12.1? If yes, nd it, if no, explain why not. Repeat with A = N+ . 3.12.3 Let A be a closed and bounded subset of R or C. Let I be a set and for each k I let Uk be an open subset of F . Suppose that A kI Uk . Prove that there exists a nite subset K of I such that A kK Uk . (Hint: Prove that for each a A there exists a > 0 such that B (a, a ) is in some Uk . Apply Theorem 3.12.3.)
Chapter 4: Limits of functions
4.1
Limit of a function
Denition 4.1.1 Let A be a subset of C and let f : A C be a function. Suppose that a complex number a is a limit point of A (see Denition 3.11.8). The limit of f (x) as x approaches a is the complex number L if for every real number > 0 there exists a real number > 0 such that for all x A, if 0 < |x a| < , then |f (x) L| < . When this is the case, we write lim f (x) = L. Alternatively, in order to not make lines crowded with subscripts, we can write limxa f (x) = L. It is important to note that we are not asking for f (a). For one thing, a may or may not be in the domain of f , we only know that a is a limit point of the domain of f . We are asking for the behaviour of the function f at points near and nearer a, and not equal to a! A simple geometric picture is as follows: on the graph of y = f (x) we cover the vertical line x = a as below, and with that information, we conclude that limxa f (x) = L.
xa
L a
The function f itself might be any of the following:
L a
L a
L a
117
In an intuitive sense we are hoping that f (x) for x near a can predict a value of f as we get arbitrarily close to a. For example, we may not be able to bring x to 0 Kelvin, but if we can take measurements f (x) for x getting colder and colder, perhaps we can theoretically predict what may happen at 0 Kelvin. But how believable is our prediction? Perhaps for our theory to be satisfactory, we need to run experiments at temperatures x that give us f (x) within = 10 of the predicted value. Or when instruments get better, perhaps gets smaller, say one thousandth. Or a new material is discovered which allows us to get even smaller. But no matter what is determined ahead of time, for the prediction to be believable, we need to determine a range x within a of a but not equal to a, for which the f -values are within the given of the prediction. A graphical way of representing the epsilon-delta denition of limits is as follows: For every positive there exists a positive such that for all x in the domain with 0 < |x a| < (the x = a in the vertical gray band), the value of f (x) is within of L (in the horizontal grey band):
L+ L L a
Clearly if gets smaller, has to get smaller too. (But if gets larger, we may keep the old .) While these pictures can help our intuition, they do not constitute a proof: the definition is an algebraic formulation, and as such it requires algebraic proofs. In the rest of the section we examine many examples algebraically, with the goal of mastering the epsilon-delta proofs. But epsilon-delta proofs are time-consuming, so in the future we will want to replace them with some shortcuts. We will have to prove that those shortcuts are logically correct, and the proofs will require mastering abstract epsilon-delta proofs. Naturally, before we can master abstract epsilon-delta proofs, we need to be comfortable with epsilon-delta proofs on concrete examples. In short, in order to be able to avoid epsilon-delta proofs, we have to master them. (Ha!) Example 4.1.2 lim (4x 5) = 7.
x3
Proof. The function that takes x to 4x 5 is a polynomial function, so it is dened for all complex numbers. So the domain of the function is C and all points of the domain are limit
118
Section 4.1: Limit of a function
points of the domain. Let > 0. [We are proving that for all real numbers > 0 something-or-other holds. Recall that all proofs of this form start with Let be an arbitrary positive real number, or abbreviated as we did. Now we have to prove that the something-or-other holds. But this something-orother claims that there exists a real number > 0 with a certain property. Thus we have to construct such a . In this first example we simply present a that works, but in subsequent examples we will show how to find a working . In general, the proof should contain a specification of and that it works, but how one finds that is in general left for scratch work or inspiration. ] Set = /4. [Again: never mind how this magic /4 appears here; wait until the next example where we present the process of finding .] Then is a positive real number. [Now we have to prove that for all x, if 0 < |x 3| < , then |f (x) L| < . The proof of for all x . . . starts with:] Let x be an arbitrary complex number. [For this x we now have to prove that if 0 < |x 3| < , then |f (x) L| < . The proof of If P then Q starts with Assume P .] Assume that 0 < |x 3| < . [Now we have to prove Q, i.e., we have to prove that |f (x) L| = |(4x 5) 7| < . We do not simply write |(4x 5) 7| < because we do not know that yet. We write the left side of this inequality, and manipulate it algebraically, with triangle inequalities... until we get < . ] Then |(4x 5) 7| = |4x 12| < 4 = 4 [wasnt this a clever choice of ?] 4 = . In the next example we have no magic appearing, but we do the work in deriving a working form for it. It suces to nd any that works. Example 4.1.3
x2
= 4| x 3|
lim (4x2 5x + 2) = 28.
Proof. The function that takes x to 4x2 5x + 2 is a polynomial function, so it is dened for all complex numbers, so that 2 is a limit point of the domain. Let > 0. Set [ to be determined still; the final write-up will have = this filled in, but as we are writing the proof, we dont know yet!]. Then is a positive real number. [Hoping, anyway.] Let x be any complex number. Suppose that 0 < |x + 2| < . Then |(4x2 5x + 2) 28| = |4x2 5x 26| [Goal #1: x a better be a factor!]
Chapter 4: Limits of functions = |(x + 2)(4x 13)|
119
= |4x 13| |x + 2| [Goal #1 accomplished] [Goal #2: we want to bound the coefficient of |x a| by a constant. But
to make a restriction on : lets make sure that is at most 1 (or 2, or 15, it doesnt matter what positive number you pick in this example).] [So on the line write: = min {1, = |4(x + 2) 21| |x + 2| = |4(x + 2 2) 13| |x + 2| (adding a clever zero) }.]
obviously |4x 13| is not bounded above by a constant for all x. So we need
(|4(x + 2)| + 21) |x + 2| (by triangle inequality |a b| |a| + |b|) = 25 |x + 2| [Goal #2 accomplished: constant B times |x a|] < 25 < (4 1 + 21) |x + 2| (since |x + 2| < 1)
[Now go back to and finish with: = min {1, /B }.] 25 /25
= .
The nal version of the proof of lim (4x2 5x + 2) = 28 then looks like this:
x2
The function that takes x to 4x2 5x + 2 is a polynomial function, so 2 is a limit point of the domain. Let > 0. Set = min {1, /25}. Then is a positive real number. Let x be any complex number. Suppose that 0 < |x + 2| < . Then |(4x2 5x + 2) 28| = |4x2 5x 26| = |(x + 2)(4x 13)|
= |4(x + 2) 21| |x + 2|
= |4(x + 2 2) 13| |x + 2| (adding a clever zero) (|4(x + 2)| + 21) |x + 2| (by triangle inequality |a b| |a| + |b|)
= |4x 13| |x + 2|
= 25 |x + 2| < 25 = . 25 /25
< (4 1 + 21) |x + 2| (since |x + 2| < 1)
120
We repeat the same type of discovery work on the next example, with a few fewer explicit comments: Example 4.1.4
x1
lim (4/x2 ) = 4.
Proof. The function that takes x to 4/x2 is dened for all non-zero complex numbers, so 1 is a limit point of the domain. Let > 0. Set = . Then is a positive real number. Let x be any complex number that satises 0 < |x + 1| < . Then 4 4x2 x2 (1 x)(1 + x) =4 x2 1x |1 + x|[Goal #1 accomplished: something times |x a|] =4 x2 [Now we want something to be at most a some constant. Certainly if we |4/x2 4| = allow x to get close to 0, then (1 x)/x2 will have a very large size, so in order to find an upper bound, we need to make sure that x stays away from is strictly smaller than 1. For example, make sure that 0.4. Thus, on
0. Since x is within of 1, in order to avoid 0 we need to make sure that }.] the line write: = min {0.4, 2 (x + 1) |1 + x| (by rewriting 1 x = 2 (x + 1)) =4 | x| 2 2 + | x + 1| 4 |1 + x| (by triangle inequality) x2 2. 4 4 2 |1 + x| (since 0.4) | x| 9. 6 |1 + x| (by rewriting x = x + 1 1) = | x + 1 1| 2 9. 6 |1 + x| (by reverse triangle inequality 0. 62 |x + 1 1| 1 |x + 1| > 1 1 0.4 = 0.6, [On the line now write: = min {0.4, 0.62/9.6}.] 9. 6 < 0. 62 9. 6 0.62 /9.6 0. 62 = . so that 1/|x + 1 1|2 < 1/0.62)
Chapter 4: Limits of functions And here is another example for good measure, with already lled in: Example 4.1.5 lim
4 x3 + x x2i 8xi
3
121
= 2.
x +x Proof. The domain of 48 consists of all complex numbers dierent from i/8, so 2i is a x i limit point of the domain. Let > 0. Set = min {1, /9}. Let x be any complex number dierent from i/8 such that 0 < |x 2i| < . [Here we merged: Let x be any complex number different from i/8. Let x satisfy 0 < |x 2i| < . into one shorter and logically equivalent statement Let x be any complex number different from i/8 such that 0 < |x 2i| < .] Then
4x3 + x 4x3 + x (2) = +2 8x i 8x i 4x3 + x + 16x 2i = 8x i 3 4x + 17x 2i = 8x i 2 (4x + 8ix + 1)(x 2i) = [ (If x 2i were not a factor, 8x i then either our limit or our algebra would have been wrong.] (4x + 8ix + 1) = |x 2i| (Goal #1: x a is a factor.) 8x i |4x2 | + |8ix| + 1 |x 2i| (by triangle inequality) |8x i| 4|(x 2i + 2i)|2 + 8|x 2i + 2i| + 1 |x 2i| |8(x 2i) + 15i| 4(|x 2i| + 2)2 + 8(|x 2i| + 2) + 1 |x 2i| 8|x 2i| + 15 (by triangle and reverse triangle inequality) 2 4(1 + 2) + 8(1 + 2) + 1 |x 2i| (since |x 2i| < 1) 8 + 15 61 |x 2i| = 7 < 9
2
9/9 = . We now give an example where the limit is at a non-specic a.
122 Example 4.1.6 lim (x2 2x) = a2 2a.

xa
Proof. Any a is a limit point of the domain of the given polynomial function. Let > 0. Set = min {1, }. Let x satisfy 0 < |x a| < . Then |(x2 2x) (a2 2a)| = |(x2 a2 ) (2x 2a)| (by algebra) = |(x + a 2)(x a)| (by algebra) = |x + a 2| |x a| = |(x + a)(x a) 2(x a)| (by algebra)
= |(x a) + 2a 2| |x a| (by adding a clever 0) (1 + |2a 2|) |x a| (since |x a| < 1)
(|x a| + |2a 2|) |x a| (by triangle inequality)
< (1 + |2a 2|) = .
(1 + |2a 2|)/(1 + |2a 2|)
Example 4.1.7 lim
x3
2x 6 = 0.
Proof. The domain here is all x 3. So 3 is a limit point of the domain. Let > 0. Set = 2 /2. Let x > 3 satisfy 0 < |x 3| < . Then 2x 6 = 2 x 3 < 2 = 2 2 /2 = . Note: Often books consider the last example as a case of a one-sided limit (see denition below) since we can only take the x from one side of 3. Denition 4.1.8 Let A R, a R, L C, and f : A C a function. Suppose that a is a limit point of {x A : x > a} (resp. of {x A : x < a}). We say that the rightsided (resp. left-sided) limit of f (x) as x approaches a is L if for every real number > 0 there exists a real number > 0 such that for all x A, if 0 < x a < (resp. if 0 < a x < ) then |f (x) L| < . When this is the case, we write limxa+ f (x) = lim f (x) = L (resp. limxa f (x) = lim f (x) = L).
xa
xa+
With this, Example 4.1.7 can be phrased as lim
follows: The domain A consists of all x 3, and 3 is a limit point of A {x : x > 3} = A.
x3+
2x 6 = 0, and the proof goes as
Chapter 4: Limits of functions Let > 0. Set = 2 /2. Let x satisfy 0 < x 3 < . Then 2x 6 = 2 x 3 < 2 = 2 2 /2 = . Thus, the two proofs are almost identical. Note that lim
123
3 is not the limit point of A {x R : x < 3} = . One-sided limits can also be used in contexts where limxa f (x) does not exist. Below is one example. Example 4.1.9 Let f : R R be given by f (x) =
x1+
x3
2x 6 does not exist because
lim f (x) = 5, lim f (x) = 3.

x1
x2 + 4, x 2,
if x > 1; if x 1.
Then
Proof. Let > 0. Set = min {1, /3}. Let x satisfy 0 < x 1 < . Then f (x) 5 = x2 + 4 5 (since x > 1) = x2 1 = (x + 1)(x 1)
< (x + 1) (since x > 1, so x + 1 is positive) = (x 1 + 2) (by rewriting) 3/3 = . < (1 + 2) (since x 1 < 1)
This proves that lim+ f (x) = 5. Set = . Let x satisfy 0 < 1 x < . Then f (x) (3) = x 2 + 3 (since x < 1) < =x1
x1
= . This proves that lim f (x) = 3.

x1
4.1.1 Rework Example 4.1.3 with choosing to be at most 2 rather than at most 1. 4.1.2 Fill in the blanks of the following proof that limx2 (x2 3x) = 2. Explain why none of the inequalities can be changed into equalities. so . Let > 0. Set = . Let x satisfy that 2 is a 0 < |x 2| < . Then |(x2 3x) (2)| = x2 3x + 2 = | x 1| | x 2| (because (because ) ) ) (because ) )
(| x 2| + 1| ) | x 2| (3 + 1) |x 2| < 4 4 4 = . (because
(because
4.1.3 Determine the following limits and prove them with the epsilon-delta proofs. i) lim (x3 4). ii) lim
x1
iii)
iv) lim v) lim vi)
1 . x2 x x 4 lim 2 . x3 x +2
4.1.4 Prove that for all b C, the limit of f (x) as x approaches 5 (from the left/right) is the same, where x3 4x2 , if x = 5; f ( x) = b, if x = 5. Exercise 4.1.5 (Invoked in Example 5.1.4.) Suppose that a is a limit point of {x A : x > a} and of {x A : x < a}. Prove that limxa f (x) = L if and only if limxa+ f (x) = L and limxa f (x) = L. 4.1.6 Suppose that a is a limit point of {x A : x > a} but not of {x A : x < a}. Prove that limxa f (x) = L if and only if limxa+ f (x) = L. 4.1.7 Prove that limxa (mx + l) = ma + l, where m and l are constants. 4.1.8 Let f : R R be given by f (x) = limx0 f (x) = 1.
x | x| .
x 2 9 . x3 x3 lim x23 . x3 x 9
x4
x + 5.
Prove that limx0+ f (x) = 1 and that
125
4.2
When something is not a limit
Recall that limxa f (x) = L means that a is the limit point of the domain of f , and that for all real numbers > 0 there exists a real number > 0 such that for all x in the domain of f , if 0 < |x a| < then |f (x) L| < . Think of limxa f (x) = L as statement P , a being a limit point of the domain as statement Q, and the epsilon-delta part as statement R. By definition, P is logically the same as Q R.
Thus if limxa f (x) = L, then either a is not the limit point of the domain of f or else it is not true that for all real numbers > 0 there exists a real number > 0 such that for all x in the domain of f , if 0 < |x a| < then |f (x) L| < . This simply says that P is the same as (Q) (R). In particular, lim f (x) = L and a is a limit point of the domain of f
means that it is not true that for all real numbers > 0 there exists a real number > 0 such that for all x in the domain of f , if 0 < |x a| < then |f (x) L| < . This says that (P ) Q is the same as R. You may want to write truth tables for yourself. Negations of compound sentences, such as in the previous paragraph, are typically hard to process and to work with in proofs. By the usual negation rules of compound statements (see chart on page 25), we successively rewrite this last negation into a form that is easier to handle: For all real numbers > 0 there exists a real number > 0 such that for all x in the domain of f , if 0 < |x a| < then |f (x) L| < . Negation of For all z of some kind, property A holds is There is some z of that kind for which A is false. Hence the following rephrasing: = There exists a real number > 0 such that there exists a real number > 0 such that for all x in the domain of f , if 0 < |x a| < then |f (x) L| < . Negation of There exists z of some kind such that property A holds is For all z of that kind, A is false. Hence the following rephrasing: = There exists a real number > 0 such that for all real numbers > 0, for all x in the domain of f , if 0 < |x a| < then |f (x) L| < . Negation of For all z of some kind,
xa
126
Section 4.2: When something is not a limit property A holds is There is some z of that kind for which A is false. Hence the following rephrasing: = There exists a real number > 0 such that for all real numbers > 0, there exists x in the domain of f such that if 0 < |x a| < then |f (x) L| < . Negation of If A then B is A and not B . Hence the following rephrasing: = There exists a real number > 0 such that for all real numbers > 0, there exists x in the domain of f such that 0 < |x a| < and |f (x) L| < . = There exists a real number > 0 such that for all real numbers > 0, there exists x in the domain of f such that 0 < |x a| < and |f (x) L| . In summary, we just proved that:
Theorem 4.2.1 If a is a limit point of the domain of f , then limxa f (x) = L means that there exists a real number > 0 such that for all real numbers > 0, there exists x in the domain of f such that 0 < |x a| < and |f (x) L| .
x Example 4.2.2 The limit of |x | as x approaches 0 does not exist. In other words, for all x complex numbers L, limx0 |x| = L. x The domain of the function that takes x to |x | is the set of all non-zero complex x numbers. For each non-zero x, |x| is a complex number of length 1 and with the same angle as x. Thus the image of this function is the unit circle in C. Note that it is possible to take two non-zero x very close to 0 but at dierent angles so that their images on the unit circle are far apart. This is a geometric reasoning why the limit cannot exist. Next we give an epsilon-delta proof.
Proof. The domain of the function that takes x to |x x| is the set of all non-zero complex numbers, so that 0 is a limit point of the domain. [Thus if the limit is not L, then it must be the epsilon-delta condition that fails.] Set = 1. Let > 0 be an arbitrary positive number. Let x = /2 if Re(L) 0, and let x = /2 otherwise. Then 0 < |x| = |x 0| < . If Re(L) 0, then x /2 Re L = Re L = 1 Re(L) 1, | x| | /2| so that | |x x| L| 1 = , and if Re(L) < 0, then Re x L | x| = Re /2 L |/2| = 1 Re(L) > 1,
L| > 1 = . This proves the claim of the example. so that again | |x x|
127
i Example 4.2.3 For all L C, limx2 x 2 = L. A geometric reason for the non-existence of this limit is that as x gets closer to 2 (but i not equal to 2), the size of x 2 gets larger and larger.
Proof. Set = 1. Let > 0 be an arbitrary positive number. Set = min {, 1/(|L| + 1)}. Let x = 2 /2. Then 0 < |x 2| < , and 2i i L = L x2 2i |L| (by reverse triangle inequality) 2(|L| + 1) |L| (since 1/(|L| + 1)) 1 = . Example 4.2.4 For f : R R given by the graph below, lim f (x) does not exist because of the jump in the function at 2.
x2
2 1
1 . Let Here is an epsilon-delta proof. Say that the limit exists. Call it L. Set = 4 1 3 be an arbitrary positive number. If L 2 , set x = 2 + min 4 , 2 , and if L < 3 2 , set 1 x = 2 min { 4 , 2 }. In either case,
0 < |x 2| = min
1 , 4 2
< . 2
1 1 If L 3 , by our choice x = 2 + min { 4 , 2 }, so that f (x) = 1 + min { 4 , 2} 1 + 1 , 2 4 3 1 1 whence |f (x) L| 4 = . Similarly, if L < 2 , by our choice x = 2 min { 4 , 2 }, so that 1 3 1 f (x) = 2 min { 4 , 2} 2 1 4 = 1 + 4 , whence |f (x) L| 4 = . Thus no L works, so the limit of f (x) as x approaches 2 does not exist.
128 Exercises for Section 4.2 4.2.1 Prove that limx3 (3x 4) = 3. 4.2.3 Prove that limx3 4.2.2 Prove that limx1 (x2 + 4) = 5.
x 3 x 2 9
Section 4.3: Limit theorems
does not exist.
4.2.4 Prove that for all a R, limxa f (x) does not exist, where f : R R is dened by 1, if x is rational; f ( x) = 0, if x is not rational. 4.2.5 Prove that limx0 ( x x) does not exist. (Hint: reason is dierent from the reasons in all other examples.)
4.3
Limit theorems
While epsilon-delta proofs are a reliable method for proving limits, they do not help in deciding what a limit may be. In this section we prove theorems that will for many functions eciently establish the limits, and remove the need for the time-consuming epsilon-delta proofs. Nevertheless, to prove these theorems we will need epsilon-delta methods. Theorem 4.3.1 If a limit exists, it is unique. Proof. Suppose that both L1 and L2 are limits of f (x) as x approaches a. If L1 = L2 , then |L1 L2 | is a positive real number. Thus for each i = 1, 2, there exists i > 0 such that for all x in the domain of f , if 0 < |x a| < i , then |f (x) Li | < |L1 L2 |/2. Set = min {1 , 2 }. Then is a positive real number. Let x in the domain of f satisfy 0 < |x a| < . Then |L1 L2 | = |L1 f (x) + f (x) L2 | (by adding a clever 0) < 2|L1 L2 |/2 (since 1 , 2 ) = |L1 L2 |, |L1 f (x)| + |f (x) L2 | (by triangle inequality)
which says that |L1 L2 | is strictly smaller than itself. But this contradicts trichotomy on R. Thus the assumption L1 = L2 must have been wrong, which proves the theorem. Theorem 4.3.2 Let a be a limit point of the domain of a function f with limxa f (x) = L. Suppose that L = 0. Then there exists > 0 such that for all x in the domain of f , if 0 < |x a| < , then |f (x)| > |L|/2. In particular, there exists > 0 such that for all x in the domain of f , if 0 < |x a| < , then f (x) = 0.
129
Proof. Set |L|/2 > 0, there exists > 0 such that for all x in the domain of f , if 0 < |x a| < , then |f (x) L| < |L|/2. Hence by the reverse triangle inequality, for the same x, |L|/2 > |f (x) L| |L| |f (x)| = |L| |f (x)|, which proves that |f (x)| > |L|/2. In particular f (x) = 0. The following theorem is very important, so study it carefully. Theorem 4.3.3 Let A be the domain of f and g , and let a be a limit point of A, and let c C. Suppose that limxa f (x) and limxa g (x) both exist. Then (1) (Constant rule) lim c = c. (2) (Linear rule) lim x = a. (3) (Sum/dierence rule) lim (f (x) g (x)) = lim f (x) lim g (x). (4) (Scalar rule) lim cf (x) = c lim f (x).
xa xa xa xa xa xa xa xa
(5) (Product rule) lim (f (x) g (x)) = lim f (x) lim g (x).
xa xa
(6) (Quotient rule) If lim g (x) = 0, then lim

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

xa
Proof. Both (1) and (2) were proved in Exercise 4.1.7. Set L = limxa f (x) and K = limxa g (x). (3) Let > 0. Then /(|c| + 1) is a positive number. (Note that we did not divide by 0.) Since L = limxa f (x), there exists > 0 such that for all x A, 0 < |x a| < implies that |f (x) L| < /(|c| + 1). Hence for the same x, |cf (x) cL| = |c| |f (x) L| < |c| /(|c| + 1) < , which proves that limxa cf (x) = cL = c limxa f (x). (4) Let > 0. Since L = limxa f (x), there exists 1 > 0 such that for all x A, if 0 < |x a| < 1 , then |f (x) L| < /2. Similarly, since K = limxa g (x), there exists 2 > 0 such that for all x A, if 0 < |x a| < 2 , then |g (x) L| < /2. Set = min {1 , 2 }. Then is a positive number. Let x A satisfy 0 < |x a| < . Then |(f (x) g (x)) (L K )| = |(f (x) L) (g (x) K )| (by algebra) |f (x) L| + |g (x) K | (by triangle inequality) < /2 + /2 (because 0 < |x a| < 1 , 2 ) = ,
which proves (4). (5) Let > 0. Then min {/(2|L| + 2), 1}, min {/(2|K | + 1), 1} are positive numbers. Since L = limxa f (x), there exists 1 > 0 such that for all x A, if 0 < |x a| < 1 , then |f (x) L| < min {/(2|K | + 1), 1}. Similarly, since K = limxa g (x), there exists
130
2 > 0 such that for all x A, if 0 < |x a| < 2 , then |g (x) L| < /(2|L| + 2). Set = min {1 , 2 }. Then is a positive number. Let x A satisfy 0 < |x a| < . Then by triangle inequality, | f ( x) | = | f ( x) L + L | | f ( x) L | + | L | < 1 + | L | , and so |f (x) g (x) L K | = |f (x) g (x) f (x)K + f (x)K L K | (by adding a clever zero) = |f (x)| |g (x) K | + |f (x) L| |K | (by factoring) + |K | (since 1 , 2 ) < (1 + |L|) 2| L | + 2 2| K | + 1 < /2 + /2 = . (6) Let > 0. Since K = 0, by Theorem 4.3.2, there exists 0 > 0 such that for all x A, if 0 < |x a| < 0 , then |g (x)| > |K |/2. |f (x) g (x) f (x)K | + |f (x)K L K | (by triangle inequality)
The numbers |K |/4, |K |2 /(4|L| + 1) are positive numbers. Since L = limxa f (x), there exists 1 > 0 such that for all x A, if 0 < |x a| < 1 , then |f (x) L| < |K |/4. Similarly, since K = limxa g (x), there exists 2 > 0 such that for all x A, if 0 < |x a| < 2 , then |g (x) L| < |K |2 /(4|L| +1). Set = min {0 , 1 , 2 }. Then is a positive number. Let x A satisfy 0 < |x a| < . Then L f ( x) f (x)K Lg (x) = (by algebra) g ( x) K Kg (x) f (x)K LK + LK Lg (x) (by adding a clever zero) = Kg (x) f (x)K LK LK Lg (x) + (by triangle inequality) Kg (x) Kg (x) | L | K g ( x) f ( x) L + (by factoring) = g ( x) |K | g ( x) |K | 2 |L| |K |2 2 < + (since 0 , 1 , 2 ) 4 | K | | K | 4| L | + 1 | K | < /2 + /2 = . This proves (6) and thus the theorem. Theorem 4.3.4 (Power rule for limits) Let n be a positive integer. If limxa f (x) = L, then limxa f (x)n = Ln .
131
Proof. The case n = 1 is the assumption. We proceed by mathematical induction. Suppose that we know the result for n 1. Then
xa
lim f (x)n = lim f (x)n1 f (x) (by algebra) =L

xa n1
= Ln .
L (by induction assumption and by the product rule)
Theorem 4.3.5 (Polynomial function rule for limits) Let f be a polynomial function. Then for all complex (or real) a, limxa f (x) = f (a). Proof. Because f is a polynomial function, it can be written as f ( x ) = c0 + c1 x + c2 x 2 + c3 x 3 + + cn x n for some non-negative integer n and some constants c0 , c1 , . . . , cn . By the linear rule, limxa x = a. Hence by the power rule, for all i = 1, . . . , n, limxa xi = ai . By the constant rule, limxa ci = ci , so that by the product rule limxa ci xi = ci ai . Hence by repeating the sum rule,
xa
lim f (x) = lim c0 + c1 x + c2 x2 + + cn xn = c0 + c1 a + c2 a2 + + cn an = f (a).

xa
Theorem 4.3.6 (Rational function rule for limits) Let f be a rational function. Then for all complex (or real) a in the domain of f , limxa f (x) = f (a). Proof. Let a be in the domain of f . Write f (x) = g (x)/h(x) for some polynomial functions g, h such that h(a) = 0. By Theorem 2.4.12, the domain of f is the set of all except nitely many numbers, so that in particular a is a limit point of the domain. By the polynomial function rule for limits, limxa g (x) = g (a) and limxa h(x) = h(a) = 0. Thus by the quotient rule, limxa f (x) = g (a)/h(a) = f (a). Theorem 4.3.7 (Absolute value rule for limits) For all a C, limxa |x| = |a|. Proof. This function is dened for all complex numbers, and so every a C is a limit point of the domain. Let > 0. Set = . Then for all x C, by the reverse triangle inequality, |x| |a| |x a| < = . Theorem 4.3.8 (Real and imaginary parts of limits) Let f : A C be a function, let a be a limit point of A, and let L C. Then limxa f (x) = L if and only if limxa Re f (x) = Re L and limxa Im f (x) = Im L. Proof. First suppose that limxa f (x) = L. Let > 0. By assumption there exists > 0 such that for all x A, 0 < |x a| < implies that |f (x) L| < . Then for the same x, | Re f (x) Re L| = | Re(f (x) L)| |f (x) L| < ,
132
and similarly | Im f (x) Im L| < which proves that limxa Re f (x) = Re L and limxa Im f (x) = Im L. Now suppose that limxa Re f (x) = Re L and limxa Im f (x) = Im L. Let > 0. By assumptions there exist 1 , 2 > 0 such that for all x A, 0 < |x a| < 1 implies that | Re f (x) Re L| < /2 and 0 < |x a| < 2 implies that | Im f (x) Im L| < /2. Let = min {1 , 2 }. Then > 0, and if 0 < |x a| < , then |f (x) L| = | Re f (x) + i Im f (x) Re L i Im L| < /2 + /2 = , | Re f (x) Re L| + |i|| Im f (x) Im L|
which proves that limxa f (x) = L. Theorem 4.3.9 (The composite function theorem) Let h be the composition of functions h = g f . Suppose that a is a limit point of the domain of g f , that lim f (x) = L and that lim g (x) = g (L). Then lim h(x) = g (L).
xL xa xa
Proof. Let > 0. Since lim g (x) = g (L), there exists 1 > 0 such that for all x in the such that for all x in the domain of f , if 0 < |x a| < then |f (x) L| < 1 . Thus for the same , if x is in the domain of h and 0 < |xa| < , then |h(x)g (L)| = |g (f (x))g (L)| < because |f (x) L| < 1 . Theorem 4.3.10 Suppose that f, g : A R, that a is a limit point of A, that limxa f (x) and limxa g (x) both exist, and that for all x A, f (x) g (x). Then
xa xa xa
domain of g , if 0 < |x a| < 1 then |g (x) g (L)| < . Since lim f (x) = L, there exists > 0
xa
xa
lim f (x) lim g (x).

xa
Proof. Let L = lim f (x), K = lim g (x). Let > 0. By assumptions there exists > 0 such that for all x A, if 0 < |x a| < , then |f (x) L|, |g (x) K | < /2. Then for the same x, K L = K g ( x) + g ( x) f ( x) + f ( x) L K g ( x) + f ( x) L |K g (x) + f (x) L| |K g (x)| |f (x) L| (by triangle inequality)
> /2 /2 = .
Since this is true for all > 0, by Theorem 2.7.13, K L, as desired. Theorem 4.3.11 (The squeeze theorem) Suppose that f, g, h : A R, that a is a limit point of A, and that for all x A \{a}, f (x) g (x) h(x). If limxa f (x) and limxa h(x)
Chapter 4: Limits of functions both exist and are equal, then limxa g (x) exists as well and
xa
133
lim f (x) = lim g (x) = lim h(x).

xa xa
Proof. [If we knew that limxa g (x) existed, then by the previous theorem, limxa f (x) limxa g (x) limxa h(x) = limxa f (x) would give that the three limits are equal. But we have yet to prove that limxa g (x) exists.] Let L = limxa f (x) = limxa h(x). Let > 0. Since limxa f (x) = L, there exists 1 > 0 such that for all x, if 0 < |x a| < 1 then |f (x) L| < . Similarly, since limxa g (x) = L, there exists 2 > 0 such that for all x, if 0 < |x a| < 2 then |g (x) L| < . Now set = min {1 , 2 }. Let x satisfy 0 < |x a| < . Then < f (x) L g (x) L h(x) L < , where the rst inequality holds because 1 , and the last inequality holds because 2 . Hence < g (x) L < , which says that |g (x) L| < , so that limxa g (x) = L. Exercises for Section 4.3 4.3.1 Determine the following limits by invoking appropriate results: i) lim (x3 4x 27), lim (x2 + 5).
x2 x2
(x3 4x 27)3 ii) lim . x2 (x2 + 5)2 |x3 4x 27| . iii) lim x2 (x2 + 5)3 4.3.2 If lim f (x) = L and lim g (x) = K , prove that lim (3f (x)2 4g (x)) = 3L2 4K . 4.3.3 Prove that lim x = a for any non-zero complex number a and any integer n. 4.3.4 Prove that for any positive real number a and any whole number n, lim xn = an . 4.3.5 Prove that if limxa f (x) = L, then limxa |f (x)| = |L|. Give an example of a function such that limxa |f (x)| = |L| and limxa f (x) does not exist.
xa xa xa n xa n xa
4.3.6 Find functions f, g such that limxa (f (x) + g (x)) exists but limxa f (x) and limxa g (x) do not exist. Does this contradict the sum rule? 4.3.7 Find functions f, g such that limxa (f (x)g (x)) exists but limxa f (x) and limxa g (x) do not exist. Does this contradict the product rule? 4.3.8 Prove that lim 4.3.9 Prove that lim does exist. Does this contradict the sum rule for limits? Justify. this contradict the product rule for limits? Justify.
x x0 |x| x x0 |x|
and lim 1
x0 | x| x0 x
x | x|
do not exist, but that lim

x | x|
x0
x | x|
+ 1
x | x|
and lim
do not exist, but that lim
x0
x| does exist. Does |x
134
Section 4.4: Innite limits (for real-valued functions)
4.3.10 The following information is known about functions f and g : a 0 1 2 3 4 f (a) 1 1 not dened 4 5
xa
lim f (x) 2 0 3 4 1
g (a) 6 5 6 not dened 4
xa
lim g (x) 4 5 6 7
For each part below provide the limit if there is enough information, and justify. (Careful, the answers to the last two parts are dierent.) i) lim (f (x) g (x)). ii) lim (f (x) + g (x)).
x4 f (x ) . x2 g (x) lim g(x) . x4 f (x) x3 x0
iii) lim (f (x) g (x)). iv) lim v)
vii) lim (g f )(x).

x4
vi) lim (g f )(x).

x1
4.3.11 Let A B C, let a be a limit point of A, and let f : A C, g : B C. Suppose that for all x A \ {x}, g (x) = f (x). Prove that if limxa g (x) = L, then limxa f (x) = L. In particular, if a R is a limit point of A = B R and if limxa g (x) = L, then the restriction of g to A has the same limit point at a. 4.3.12 Let A C, let f, g : A C, and let a be a limit point of A. Suppose that for all x A \ {a}, f (x) = g (x). Prove that limxa f (x) exists if and only if limxa g (x) exists, and in they both exist, then the two limits are equal.
4.4
Innite limits (for real-valued functions)
When the codomain of a function is a subset of an ordered eld such as R, the values of a function may grow larger and larger with no upper bound, or more and more negative with no lower bound. In that case we may want to declare limit to be or . Naturally both the denition and how we operate with innite limits requires dierent handling. Denition 4.4.1 Let A C, f : A R a function, and a C. Suppose that a is a limit point of A (and not necessarily in A). We say that the limit of f (x) as x approaches a is if for every real number M > 0 there exists a real number > 0 such that for all x A, if 0 < |x a| < then

xa
135
limit of f (x) as x approaches a is if for every real number M < 0 there exists a real number > 0 such that for all x A, if 0 < |x a| < then f (x) < M . We write this as lim f (x) = limxa f (x) = .
xa
f (x) > M . We write this as lim f (x) = limxa f (x) = . Similarly we say that the
The limit of f (x) as x approaches a from the right is if a is a limit point of {x A : x > a}, and if for every real number M < 0 there exists a real number > 0 such that for all x A, if 0 < x a < then f (x) < M . This is written as lim f (x) = limxa+ f (x) = . It is left to the reader to spell out the denitions of the following:
xa+ xa xa xa+
lim f (x) = , lim f (x) = , lim f (x) = .
Note that we cannot use epsilon-delta proofs: no real numbers are within of innity. So instead we approximate innity with huge numbers. In fact, innity stands for that thing which is larger than any real number. Thus for all M we can nd x near a with f (x) > M is simply saying that we can take f (x)s arbitrarily large, which is more succinctly expressed as saying that f (x) goes to . (As far as many applications are concerned, a real number larger than the number of atoms in the universe is as close to innity as realistically possible, but for proofs, the number of atoms in the universe is not large enough.) Example 4.4.2 limx0
1 | x| 2
= .
Proof. 0 is a limit point of the domain (in the eld of real or complex numbers) of the function that takes x to 1/|x|2 . Let M be a positive real number. Set = 1/ M . Let x satisfy 0 < |x 0| < , i.e., let x satisfy 0 < |x| < . Then 1 1 > 2 (because 0 < |x| < ) 2 | x| = M. = .
Example 4.4.3 limx5+
x+2 x2 25
Proof. Certainly 5 is the limit point of the domain of the given function. Let M > 0. Set = min {1, 117 M }. Let x satisfy 0 < x 5 < . Then 5+2 x+2 > 2 (because x > 5 and x2 25 > 0) 2 x 25 x 25 1 7 (by algebra) = (x + 5) x 5 7 1 > (because 0 < x 5 < 1, so 0 < x + 5 < 11, and x 5 > 0) 11 x 5 7 1 > (because 0 < x 5 < ) 11
136 7 11 = M. 1
7 11M
Section 4.4: Innite limits (for real-valued functions) (because

7 11M ,
so 1/ 1/( 117 M ))
Example 4.4.4 limx0+
1 x
= .
1 x
Proof. Let M > 0. Set = 1/M . Then for all x with 0 < x 0 < , Example 4.4.5 limx0
1 x
< 1/ = M .
= .
Proof. Let M < 0. Set = 1/M . Then M is a positive number. Then for all x with 0 < 0 x < , 1 > 1/ = M. x
1 Example 4.4.6 We conclude that limx0 x cannot be a real number, and it cannot be 1 either or . Thus limx0 x does not exist. The following theorem is straightforward to prove, and so it is left for the exercises.
Theorem 4.4.7 Let f, g, h : A R, and let a be a limit point of A. Suppose that limxa f (x) = L R, limxa g (x) = , limxa h(x) = . Then (1) (Scalar rule) For any c R,
xa
lim (cg (x)) =
, , 0,
if c > 0; if c < 0; if c = 0;
xa
lim (ch(x)) =
, , 0,
if c > 0; if c < 0; if c = 0.
(2) (Sum/dierence rule)

xa
lim (f (x) g (x)) = , lim (f (x) h(x)) = , lim (g (x) h(x)) = , lim (h(x) g (x)) = , lim (g (x) + h(x)) = not enough information.
xa
xa
xa
xa
(3) (Product rule) , if L > 0; lim (f (x) g (x)) = , if L < 0; xa not enough information, if L = 0, , if L > 0; lim (f (x) h(x)) = , if L < 0; xa not enough information, if L = 0, lim (g (x) h(x)) = .
xa
Chapter 4: Limits of functions (4) (Quotient rule) f ( x) = 0, xa g (x) f ( x) lim = 0, xa h(x) lim g ( x) = lim xa f (x) h(x) = xa f (x) lim lim
137
, , not enough information, , , not enough information,
if if if if if if
L > 0; L < 0; L = 0, L > 0; L < 0; L = 0,
g ( x) = not enough information. xa h(x) Example 4.4.8 Dene g, h1 , h2 , h3 , h4 : R \ {0} R by g (x) = 1/x2 , h1 (x) = 1/x2 , h2 (x) = 17 1/x2 , h3 (x) = 1/x2 1/x4 , h4 (x) = 1/x2 1/x3 . We have seen that limx0 g (x) = , and similarly that limx0 h1 (x) = limx0 h2 (x) = limx0 h3 (x) = limx0 h4 (x) = . However,
x0 x0
lim (g (x) + h1 (x)) = lim 0 = 0,

x0 x0
lim (g (x) + h2 (x)) = lim 17 = 17, lim (g (x) + h3 (x)) = lim (1/x4 ) = ,
x0
x0
but limx0 (g (x) + h4 (x)) = limx0 (1/x3 ) does not exist. This justies the not enough information line in the sum/dierence rule in Theorem 4.4.7. Other not enough information lines are left for the exercises. Distinguish between the scalar and the product rules: when limxa f (x) = 0, if f is a constant function, then limxa (f (x)g (x)) = 0, but if f is not a constant function, then we do not have enough information for limxa (f (x)g (x)) = 0. Exercises for Section 4.4 4.4.1 Write the denitions of i) limxa+ f (x) = . ii) limxa f (x) = . iii) limxa f (x) = . 4.4.2 Prove that limx0+ 4.4.3 Prove that limx0
4.4.4 This is about Theorem 4.4.7. Prove the scalar rule, the sum/dierence rule, the product rule, and the quotient rule.
1 = . x 1 x3 = .
138
Section 4.5: Limits at innity
4.4.5 Let g (x) = 1/x2 , f1 (x) = x3 , f2 (x) = x2 , f3 (x) = 17x2 , f4 (x) = x. It is easy to see that limx0 g (x) = , limx0 f1 (x) = limx0 f2 (x) = limx0 f3 (x) = limx0 f4 (x) = 0. i) Compute the limits limx0 (f1 (x)g (x)), limx0 (f2 (x)g (x)), limx0 (f3 (x)g (x)), limx0 (f4 (x)g (x)). ii) Justify the rst not enough information in the product rule in Theorem 4.4.7. iii) Justify the second not enough information in the product rule in Theorem 4.4.7. 4.4.6 Justify the three not enough information in the quotient rule in Theorem 4.4.7. (Hint: study the previous exercise.)
4.5
Limits at innity
Below we switch to real-valued functions. For f : (a, ) R, we say that limx = if for every M > 0 there exists a real number N > 0 such that for all x > N , f (x) > M . We say that limx = if for every M < 0 there exists a real number N > 0 such that for all x > N , f (x) < M . For f : (, a) R, we say that limx = if for every M > 0 there exists a real number N < 0 such that for all x < N , f (x) > M . We say that limx = if for every M < 0 there exists a real number N < 0 such that for all x < N , f (x) < M . Example 4.5.2 limx (x5 16x4 ) = . Proof. Let M > 0. Set N = max {17, M 1/4}. Then for all x > N , x5 16x4 = x4 (x 16) x4 (because x > N 17) > N 4 (because x > N ) = M. Example 4.5.3 limx
x5 16x4 x5 +4x2
Denition 4.5.1 Let a R and L C. For f : (a, ) C, we say that limx = L if for every > 0 there exists a real number N > 0 such that for all x > N , |f (x) L| < . For f : (, a) C, we say that limx = L if for every > 0 there exists a real number N < 0 such that for all x < N , |f (x) L| < .
(M 1/4 )4 (because x > N M 1/4 ) = 1.
Proof. Let > 0. Set N = max {1, 20/}. Then for all x > N , x5 16x4 x5 4x2 x5 16x4 = 1 x5 + 4x2 x5 + 4x2
Chapter 4: Limits of functions 16x4 4x2 x5 + 4x2 16x4 + 4x2 = x5 + 4x2 16x4 + 4x2 = 5 (because x > N 0) x + 4x2 16x4 + 4x2 (because x5 + 4x2 x5 > 0) x5 16x4 + 4x4 (because x > N 1 so that x2 < x4 ) x5 20x4 = 5 x 20 = x 20 (because x > N ) < N 20 (because N 20/) 20/ = . = Exercises for Section 4.5 4.5.2 Prove that limx (x3 x) = and that limx (x3 x) = . 4.5.1 Prove that limx (x2 x) = and that limx (x2 x) = .
139
4.5.4 Prove the following limits: 2 4x+1 i) limx 3x2x = 3/2. 2 +2 iii) iv) v) ii) limx
3x3 4x+1 = . 2x2 +2 2 4x+1 limx 3x2x = 0. 3 +2 9x2 +4 limx x+2 = 3. x2 +4 = 3. limx 9 x+2
4.5.3 Is there a limit rule of the following form: If limx f (x) = and limx g (x) = , then limx (f (x) g (x)) can be determined?
4.5.5 For which rational functions f does limx f (x) = 0 hold? Justify. 4.5.6 For which rational functions f is limx f (x) a complex number? Justify. 4.5.7 Let f : (a, ) C and L C. Prove that limx f (x) = L if and only if limx Re f (x) = Re L and limx Im f (x) = Im L. 4.5.8 Let f : (, b) C and L C. Prove that limx f (x) = L if and only if limx Re f (x) = Re L and limx Im f (x) = Im L.
Chapter 5: Continuity
Continuous functions from an interval in R to R are the ones that we can graph without any holes or jumps, i.e., without lifting the pencil from the paper, so the range of such functions is an interval in R as well. We make this more formal below, and not just for functions with domains and codomains in R. The formal denition involves limits of functions. All the hard work for that was already done in Chapter 4, so this chapter, after absorbing the denition, is really straightforward. The big new results are the Intermediate value theorem and the Extreme value theorem for real-valued functions (Section 5.2), existence of radical functions (Section 5.3), and the new notion of uniform continuity (Section 5.4).
5.1
Continuous functions
Denition 5.1.1 A function f : A B is continuous at a A if for all real numbers > 0 there exists a real number > 0 such that for all x A, |x a| < implies that |f (x) f (a)| < . We say that f is continuous if f is continuous at all points in its domain. Much of the time for us A is an interval in R, a rectangle in C, or B (a, r ) in R or C, et cetera. In the more general case, A may contain a point a that is not a limit point of A. Theorem 5.1.2 Let f : A B be a function and a A. (1) If a is not a limit point of A, then f is continuous at a. (2) If a is a limit point of A, then f is continuous at a if and only if limxa f (x) = f (a). Proof. In the rst case, there exists > 0 such that B (a, ) A = {a}. Thus by denition, the only x A with |x a| < is x = a, whence |f (x) f (a)| = 0 is strictly smaller than an arbitrary positive . Thus f : A B is continuous at a. Now assume that a is a limit point of A. Suppose that f is continuous at a. We need to prove that limxa f (x) = f (a). Let > 0. Since f is continuous at a, there exists > 0 such that for all x A, if |x a| < , then |f (x) f (a)| < . Hence for all 0 < |x a| < , |f (x) f (a)| < , and since a is a limit point of A, this proves that limxa f (x) = f (a). Finally suppose that limxa f (x) = f (a). We need to prove that f is continuous at a. Let > 0. By assumption there exists > 0 such that for all x A, if 0 < |x a| < then |f (x) f (a)| < . If x = a, then |f (x) f (a)| = 0 < , so that for all x A, if |x a| < , then |f (x) f (a)| < . Thus f is continuous at a.
141
Easy applications of Theorems 4.3.3, 4.3.4, 4.3.5, 4.3.6, 4.3.7, 4.3.8, and 4.3.9, at least when the points in question are limit points of the domain, yield the next theorem easy. When the points in question are not limit points of the domain, the results below need somewhat dierent proofs. Theorem 5.1.3 We have: (1) (Constant rule) Constant functions are continuous (at all real/complex numbers). (2) (Linear rule) The function f (x) = x is continuous (at all real/complex numbers). (3) (Absolute value rule) The function f (x) = |x| is continuous (at all real/complex numbers). (4) (Polynomial function rule) Polynomial function are continuous (at all real/complex numbers). (5) (Rational function rule) Rational functions are continuous (at all points in the domain). (6) (Real and imaginary parts) f is continuous at a if and only if Re f and Im f are continuous at a. (7) (The composite function rule) Suppose that f is continuous at a, g is continuous at f (a), and that a is a limit point of the domain of g f . Then g f is continuous at a. Now suppose that f, g : A C, that a A is a limit point of A, and that f and g are continuous at a. Then (8) (Scalar rule) For any c C, cf is continuous at a. (9) (Sum/dierence rule) f + g is continuous at a. (10) (Product rule) f g is continuous at a. (11) (Quotient rule) If g (a) = 0, then f /g is continuous at a. (12) (Power rule) f n is continuous at a. The theorem covers many continuous functions, but the following function for example has to be veried dierently. Example 5.1.4 Let f : R R be dened by f ( x) =
Then f is continuous at 1, because by the polynomial rules,

x1+
x2 4, if x > 1; 3x3 , if x 1.
lim f (x) = lim+ (x2 4) = 12 4 = 3, lim f (x) = lim 3x3 = 3 13 = 3,

x1 x1
x1
so that by Exercise 4.1.5, limx1+ f (x) = 3 = 3 13 = f (1). It is worth noting that
142
Section 5.1: Continuous functions
precisely because of continuity this function can be expressed in the following ways as well: f ( x) = x2 4, if x 1; = 3x3 , if x < 1 x2 4, if x 1; 3x3 , if x 1. x + 1, if x < 0; x 1, if x > 0.
Example 5.1.5 Let f : (1, 0) (0, 1] R be dened by f (x) = y
This function is continuous at all a < 0 because there f is the polynomial/linear function f (x) = x + 1, and f is function is continuous at all a > 0 because there f is the function f (x) = x 1. Thus f is continuous at all points in its domain, so that f is continuous. Compare the example above to Theorem 5.2.4. Exercises for Section 5.1 5.1.1 Prove Theorem 5.1.3. 5.1.2 Prove that for any c C, the functions f, g : C C given by f (x) = c + x and g (x) = cx are continuous. 5.1.3 Prove that the functions Re, Im : C R are continuous. (Recall that if a, b R, then Re(a + bi) = a and Im(a + bi) = b. 5.1.4 The following information is known about functions f and g : c 0 1 2 3 4 f ( c) 1 1 not dened 4 5
xc
lim f (x) 2 0 4 1
g ( c) 3 5 6 not dened 3
xc
lim g (x) 4 5 6
5.1.6 Dene a function f : C \ {0} R such that for all non-zero x C, f (x) is the angle of x counterclockwise from the positive real axis. Argue that f is not continuous.
5.1.5 Consider the continuous function f in Example 5.1.5. Prove that f has an inverse function f 1 : (1, 1) (1, 0) (0, 1]. Graph f 1 and prove that f 1 is not continuous.
i) At which points is f continuous? ii) At which points is |g | continuous?
143
5.1.7 Assume that the trigonometric functions sine and cosine are continuous everywhere. (We will prove this in Theorem 9.7.5.) Dene a function g : C \ {0} R such that for all non-zero x C, g (x) is the sine (respectively cosine) of the angle of x counterclockwise from the positive real axis. Prove that f is continuous. Compare with the previous exercise. 5.1.8 Assume that the trigonometric function sine is continuous everywhere. (We will sin(1/x), if x = 0; prove this in Theorem 9.7.5.) Dene a function g : R R as g (x) = 0, if x = 0. i) Prove that f is not continuous at 0. ii) Prove that f is continuous at all other points. 5.1.9 Assume that the trigonometric function sine is continuous everywhere. (We will x sin(1/x), if x = 0; prove this in Theorem 9.7.5.) Dene a function g : R R as g (x) = 0, if x = 0. i) Is f continuous at 0? Prove. ii) Prove that f is continuous at all other points. x, if x is rational; 5.1.10 Let f : R R be given by f (x) = 0, otherwise. i) Prove that f is continuous at 0. ii) Prove that f is not continuous anywhere else.
5.2
Intermediate value theorem, Extreme value theorem

In this section, all functions are real-valued.
Theorem 5.2.1 (Intermediate value theorem) Let a, b R with b > a, and let f : [a, b] R be continuous. Let k be a real number strictly between f (a) and f (b). Then there exists c (a, b) such that f (c) = k . Here is a picture that illustrates the Intermediate value theorem: for value k on the y axis between f (a) and f (b) there happen to be two c for which f (c) = k . The Intermediate value theorem guarantees that one such c exists but does not say how many such c exist. y f (b) k a c1 f (a) c2 b x
144
Section 5.2: Intermediate value theorem, Extreme value theorem
Proof of Theorem 5.2.1: We rst construct a1 , a2 , . . . and b1 , b2 , . . ., such that a a1 a2 an bn b2 b1 b, and bn an = (b a)/2n . With this we dene c = sup {a1 , a2 , . . .}, and we prove that f (c) = k . Dene a0 = a, b0 = b. Once we have an1 , bn1 [a, b] with k strictly between f (an1 ) and f (bn1 ), we construct either c (an1 , bn1 ) with f (c) = k or else we construct a new pair an , bn using the following directions. Let d = (an1 + bn1 )/2. So d is the midpoint of the interval [an1 , bn1 r . If f (d) = k , we set c = d and we are done. Instead if f (d) = k , then by trichotomy either f (d) < k or f (d) > k . Let e be that one among an1 , bn1 for which the sign of f (e) k is opposite from the sign of f (d) k . Then k is strictly between f (d) and f (e). Now let the smaller of d, e be an , and the larger of the two be bn . Then 1 |bn an | = 2 |bn1 an1 |. With this procedure, either we nd c in nitely many steps, or else we construct intervals [a, b] = [a0 , b0 ] [a1 , b1 ] [a2 , b2 ] [an1 , bn1 ] [an , bn] [an+1 , bn+1 ] such that for all n, bn an = (b a)/2n and k is strictly between f (an ) and f (bn ). The set {an : n N} is bounded above by b, so that by the Upper bound theorem Theorem 3.8.3, this set has a supremum c in R. By the inclusions of the intervals, a a1 a2 an bn b2 b1 b, so that a a1 a2 an c bn b2 b1 b. Let > 0. Since f is continuous, it is continuous at c, so there exists > 0 such that for all x [a, b], if |x c| < , then |f (x) f (c)| < /3. By Exercise 3.8.3, there exists a positive integer n such that 1/2n < /(b a). Then |an c| |an bn | = (b a)/2n < , so that |f (an ) f (c)| < /3. Similarly, |f (bn) f (c)| < /3. Hence by the triangle inequality, |f (bn) f (an)| < 2/3. But k is between f (an ) and f (bn ), so that |f (an) k |, |f (bn) k | < 2/3. But then |f (c) k | |f (c) f (an )| + |f (an ) k | < , and since this true for all > 0, it follows by Theorem 2.7.13 that f (c) = k . An important application of this theorem is in the next section, introducing the radical functions. (So far we have sporadically used the square roots only, relying on some facts from high school that we have not yet proved. The next section remedies this gap.) Example 5.2.2 There exists a real number c such that c5 4 =
2
c2 2 . c2 +2
2 Proof. Let f : R R be dened by f (x) = x5 4 x x2 +2 . This function is a rational function and dened for all real numbers, so that by Theorem 5.1.3, f is continuous. Note that f (0) = 3 < 0 < f (2), so that by Theorem 5.2.1 there exists c in (0, 2) such that 2 2 f (c) = 0. In other words, c5 4 = c c2 +2 .
145
Theorem 5.2.3 Let I be an interval in R, and let f : I R be continuous. Then the image of f is an interval in R. Proof. For any c and d in the image of f , by the Intermediate value theorem (Theorem 5.2.1), any real number between c and d is in the image of f , which proves the theorem. Compare the theorem below to Exercise 5.1.5. Theorem 5.2.4 Let I be an interval in R, and let f : I R be continuous. If f has an inverse, then f 1 is continuous and f, f 1 are either both strictly increasing or both strictly decreasing. Proof. Let a < b be in I . Since f has an inverse, f (a) = f (b), so that by trichotomy, either f (a) < f (b) or f (a) > f (b). For now we assume that f (a) < f (b). Let x < y < z be in I . If f (y ) is not between f (x) and f (z ), then by invertibility either f (x) or f (z ) is between the other two. In the former case, an application of the Intermediate value theorem gives c (y, z ) (so c = x) such that f (c) = f (x), and the latter case gives c (x, y ) (so c = y ) such that f (c) = f (z ), and both of these contradict f s invertibility. So necessarily f (y ) is between f (x) and f (z ). It follows that for x < y < z < w , either f (x) < f (y ) < f (z ) < f (w ) or f (x) > f (y ) > f (z ) > f (w ). But a < b and f (a) < f (b) then force always f (x) < f (y ) < f (z ) < f (w ), so that f is strictly increasing. By Theorem 5.2.3, the domain of f 1 is also an interval, and clearly f 1 is also strictly increasing. Now we prove that f 1 is a continuous function at an arbitrary y be in the domain of f 1 . Let > 0. Let c = f 1 (y ) (so that y = f (a)). If c is not the right endpoint of I , let z (c, c + ) I . Since f is increasing, y = f (c) < f (z ). Let 1 = f (z ) y . Let x (y, y + 1 ). By the Intermediate value theorem x is in the domain of f 1 (that is, in the range of f ), and since f 1 is increasing, c = f 1 (y ) < f 1 (x) < f 1 (f (z )) = z < c + . Thus |f 1 (x) f 1 (y )| < . If instead c is the right endpoint of I , then since f is strictly increasing, y = f (c) is the maximal element in the domain of f 1 , and so in this case for 1 an arbitrary positive number, and any x (y, y + 1 ) in the domain of f 1 , |f 1 (x) f 1 (y )| < . Similarly, there exists 2 > 0 such that for all x (y 2 , y ) that are in the domain of f 1 , |f 1 (x) f 1 (y )| < . Finally, set = min {1 , 2 }. Then by what we just proved, for all x in the domain of f 1 , if |x y | < , then |f 1 (x) f 1 (y )| < . Thus f 1 is continuous at y , and since y was arbitrary, f 1 is continuous. Now suppose that f (a) > f (b) for some a < b in I . Set g = f . Then g is invertible and continuous, g (a) < g (b), and by the work so far we know that g is a strictly increasing function with a continuous strictly increasing inverse g 1 . Thus f = g is a strictly decreasing function, and f 1 = g 1 is a continuous strictly decreasing function.
146
Section 5.2: Intermediate value theorem, Extreme value theorem
Theorem 5.2.5 (Extreme value theorem) Let A be a closed and bounded subset of C, and let f : A R be a continuous function. Then there exists l, u A such that for all x A, f ( l ) f ( x ) f ( u) . In other words, f achieves its maximum value at u and its minimum value at l. Proof. Since f is continuous, for each a A there exists a > 0 such that for all x A B (a, a ), |f (x) f (a)| < 1. By Theorem 3.12.1 or Theorem 3.12.3 there exists a nite subset S of A such that A aS B (a, a ). Thus {|f (a)| + 1 : a S } is a nite subset of R+ , so it has an upper bound M . We claim that the image of f lies in the bounded interval [M, M ]. Namely, if x A, then x B (a, a ) for some a S , so that by triangle inequality, |f (x)| |f (x) f (a)| + |f (a)| < 1 + |f (a)| M . This proves the claim. Thus the image of f is a bounded subset of R, so that the inmum L and the supremum U of the image are real numbers. Suppose that for all a A, f (a) = L. Then for every a A, f (a) L > 0, so there exists a > 0 such that for all x A B (a, a ), |f (x) f (a)| < (f (a) L)/2. By Theorem 3.12.1 or Theorem 3.12.3 there exist > 0 and a nite subset S of A such that for all x A there exists a S such that B (x, ) B (a, a ). Since S is nite, M = max {(f (a)+ L)/2 : a S is a real number strictly bigger than L. For all x A, there exists a S such that x B (a, a ), so that |f (x) f (a)| < (f (a) L)/2, which means that (f (a) L)/2 < f (x) f (a), so that f (x) > (f (a) + L)/2 M > l. But this contradicts the assumption that l is the inmum of the image of f . Thus there must exist l A such that f (l) = L. Similarly, there exists u A such that f (u) = U . Example 5.2.6 The function f : [2, 2] R given by f (x) = x2 6x + 5 achieves a minimum and maximum. We can rewrite the function as f (x) = (x 3)2 4, from which it is obvious that the minimum of the function is achieved at 3 but wait a minute, this function is not dened at 3 and hence cannot achieve a minimum at 3. Here is a correction: the quadratic function (x 3)2 4 achieves its minimum at 3 and is decreasing on (, 3), so that f achieves its minimum on [2, 2] at 2 and its maximum at 2. Exercises for Section 5.2 5.2.1 Let f (x) = x3 + 4, g (x) = x4 + x2 . Prove that there exists c [4, 4] such that f (c) = g (c). 5.2.2 Assuming that cos is a continuous functions, prove that there exists a real number x such that cos x = x. 5.2.3 Let f : R Q be a continuous function. Prove that f is a constant function. Let g : C Q be a continuous function. Prove that g is a constant function.
147
5.2.5 Find real numbers a < b and a continuous function f : [a, b] [a, b] such that f (c) = c for all c (a, b).
5.2.4 (Fixed point theorem) Let f : [a, b] [a, b] be a continuous function. Prove that there exists c [a, b] such that f (c) = c.
1 . Prove that f is continuous and that f 5.2.6 Let f : (0, 1) R be dened by f (x) = x does not achieve either its minimum or its maximum. Explain why this does not contradict the Extreme value theorem (Theorem 5.2.5).
5.3
Radical functions
Let n be a positive integer. Dene the function f (x) = xn with domain R if n is odd and domain R0 otherwise. By Theorem 2.8.2 and Exercise 2.8.1, f is strictly increasing. Thus by Theorem 2.8.4, f has an inverse function f 1 . By the power rule, f is continuous, so that by Theorem 5.2.4, f 1 is strictly increasing and continuous. Denition 5.3.1 The function f (f (x) = xn ) is called exponentiation by n, and its inverse is the nth radical function, and is written as n . For each a in the domain, n a is called the nth root of a. We just established that n is a strictly increasing continuous function for any positive integer n. By continuity we immediately get the following: Theorem 5.3.2 (Radical rule for limits) For any positive integer n and any a in the domain of n , lim n x = n a.
xa
As a consequence of the Intermediate value theorem, and more specically, by Theo rem 5.2.3, the range of n is an interval in R. We can be more specic about the range: Theorem 5.3.3 The domain of Proof. If a = 0, then n is R if n is odd and is R0 otherwise.
n a = 0. If a > 0, 0n = 0 a < a + 1 (a + 1)n ,
so that since exponentiation by n is continuous, by the Intermediate value theorem (The orem 5.2.1), there exists r (0, a + 1) such that a = r n . Thus n a = r . If a < 0 and n is odd, then (a 1)n a 1 < a < 0 = 0n ,
148
Section 5.3: Radical functions
and the Intermediate value theorem guarantees that there exists r (a 1, 0) such that a = r n . Thus n a = r . Raising a non-negative real number to an integer power mth power and then taking the nth radical of the result yields the same things as rst taking the nth radical of the original non-negative real number and then raising the result to the mth power. We record this function as n/m = ( )m/n , and call it exponentiation by m/n. Theorem 5.3.4 Let r Q. If r 0 let A = R0 and if r < 0 let A = R+ . Let f : A A be dened by f (x) = xr . Then f is continuous. (1) If r > 0, then f is strictly increasing and has an increasing inverse. (2) If r < 0, then f is strictly decreasing and has a decreasing inverse. (3) If r = 0, then f is a constant function. Proof. Write r = m/n, where n and m are integers and n = 0. Since m/n = (m)/(n), by possibly multiplying by 1 we may assume that n > 0. Then f is a composition of exponentiation by m with ( n )m , in either order. Exponentiation by non-negative m is continuous by the constant or power rule, exponentiation by negative m is continuous by the quotient rule, and exponentiation by 1/n is continuous by Theorem 5.3.2. Thus f is continuous by the composite rule. If r > 0, then m, n > 0, and then f is the composition of two strictly increasing functions, hence strictly increasing. If r < 0, then m < 0 and n > 0, so f is the composition of a strictly increasing and strictly decreasing function, hence strictly decreasing by Exercise 2.8.8. Monotonicity of inverses follows from Theorem 5.2.4. The last part is obvious. We prove in Theorem 7.4.4 that more generally exponentiation by arbitrary real numbers (not just by rational numbers) is continuous. Exercises for Section 5.3 5.3.1 Let n, m Q, and suppose that a and b are in the domain of exponentiation by n and m. Prove: i) an bn = (ab)n . ii) (an )m = amn . iii) an am = an+m . iv) If a = 0, then an = 1/an . 5.3.2 Let f (x) = x x. i) Prove that for all a in the domain of f , limxa f (x) is inapplicable. ii) Prove that f is continuous.
149
5.3.3 Determine the following limits, and justify all steps by invoking the relevant theorems/rules: i) lim x2 3x + 4. x2 x 2 ii) lim . x2 x2 + 4 x 2 . iii) lim x2 x2 4 x a iv) If a 0, lim 2 . xa x a2 5.3.4 Recall from Exercise 3.10.4 that for every non-zero complex number a there exist exactly two complex numbers whose squares are a. i) Say that for all a in the rst quadrant you choose f (a) in the rst quadrant. Where are then f (a) for a in the remaining quadrants? ii) Is it possible to extend this square root function to a function f on all of C (the positive and negative real and imaginary axes) in such a way as to make limxa f (x) = f (a) for all a C? (4)(9) = 36 = 6, iii) Explain away the problematic claims 4 9 = 4 9 = 2i 3i = 6 from page 8. 5.3.5 Determine the domain of the function f given by f (x) = x2 . 5.3.6 (The goal of this exercise and the next one is to develop exponential functions without derivatives and integrals. You will see in Section 7.3 that derivatives and integrals give a more elegant approach.) Let c (1, ) and let f : Q R+ be the function f ( x ) = cx . i) Why is f a function? (Is it well-dened, i.e., are we allowed to raise positive real numbers to rational exponents?) ii) Prove that f is strictly increasing. (Hint: Theorem 5.3.4.) iii) Let > 0. Justify the following:
n+1 n+1 n+1
( + 1)
n+1
1=
i=1
( + 1) ( + 1)
i1
=
i=1
( + 1)
i1
= (n + 1).
i=1
Use the Archimedean property of R (Theorem 3.8.5) to prove that the set {(+1)n : n N} is not bounded above. iv) Prove that there exists a positive integer n such that c1/n < + 1. v) Prove that there exists 1 > 0 such that for all x (0, 1 ) Q, |cx 1| < . vi) Prove that there exists 2 > 0 such that for all x (2 , 0) Q, |cx 1| < . vii) Prove that limx0,xQ cx = 1. viii) Prove that for any r R, limxr,xQ cx exists and is a real number.
150
Section 5.3: Radical functions
5.3.7 (Related to the previous exercise.) Let c R+ . i) Prove that for any r R, limxr,xQ cx exists and is a real number. (Hint: Case c = 1 is special; case c > 1 done; related case c < 1 to the case c > 1 and the quotient rule for limits.) ii) We denote the limit in the previous part with cr . Prove that for all real numbers c, c1 , c2 , r, r1 , r2 , with c, c1 , c2 > 0, 1 r r 1 +r 2 = cr 1 cr 2 , cr 1 r 2 = ( cr 1 ) r 2 . , ( c1 c2 ) r = cr 1 c2 , c cr iii) Prove that the function g : R R given by g (x) = cx is continuous. (This is easy.) c r =
5.3.8 Here is an alternate proof of Theorem 5.3.2. Study the proof, and provide any missing commentary. Let A be the domain of n . i) Prove that an element of A is a limit point of A. ii) Suppose that a = 0. Set = n . Let x A satisfy 0 < |x a| < . Since nth root function is an increasing function on A, it follows that n n x n a = n x = n |x| < = . iii) Suppose that a > 0. First let = min {, n a}. So is a positive number. Set n = min {( + n a)n a, a ( n a )n }. Note that ( + n a)n a > ( n a)n a = 0, 0 n a < n a, n a ( n a )n > 0, so that is positive. Let x A satisfy 0 < |x a| < . Then < x a < and n a < x < + a. Since = min {( + n a)n a, a ( n a )n }, it follows that n a a ( n a )n a < x < a + a + ( + n a)n a , or in other words, ( n a )n < x < ( + n a)n . Since the nth radical function is increasing on R+ , it follows that n a n a < n x < + n a, whence < n x n a < , so that | n x n a| < , as desired. iv) Assume that a is negative. Then necessarily n is an odd integer. By what we have proved for positive numbers in the domain, such as for a, there exists > 0 such that for all x R, if 0 < |(x) (a)| < , then n x n a < . But then for x R with 0 < |x a| < , since n is odd, n x n a = (1) n x n a = n x n a < .
151
5.4
Uniform continuity
Denition 5.4.1 A function f is uniformly continuous if for all real numbers > 0 there exists a real number > 0 such that for all x and y in the domain, if |x y | < , then |f (x) f (y )| < . For example, constant functions are uniformly continuous. In uniform continuity, given a real number > 0, there exists > 0 that depends only on that makes some conclusion true, whereas in the denition of continuity at a, given a real number > 0, there exists > 0 that depends on and on a for the same conclusion to be true (with y = a). Thus the following is immediate: Theorem 5.4.2 Every uniformly continuous function is continuous. The converse is false in general: Example 5.4.3 Let f (x) = x2 , with domain C. Since f is a polynomial function, it is continuous. Suppose that f is also uniformly continuous. Then in particular for = 1 there exists > 0 such that for all x, y A, if |xy | < then |f (x)f (y )| < 1. Set d = min {, 1}. Then 1/d > d/4. Set x = 1/d d/4 + 1, y = x + d/2. Then |x y | = d/2 < d , and |f (x) f (y )| = |x y ||x + y | = d 2 2 +1 d =1+ d > 1, 2
which contradicts the assumption that f is uniformly continuous. Thus f is not uniformly continuous. (With your calculus background, uniformly continuity for dierentiable functions essentially says that the derivative function is bounded. Examine the veracity of this on the examples and non-examples so far.) The converse of Theorem 5.4.2 does hold with some extra assumptions. Theorem 5.4.4 Let A be a closed and bounded subset of R or C. Let f : A B be continuous. Then f is uniformly continuous. Proof. Let > 0. Since f is continuous, for each a A there exists a > 0 such that for all x A, |x a| < a implies that |f (x) f (a)| < /2. By applying Theorem 3.12.1 when A R or applying Theorem 3.12.3 otherwise, we get that for a = a , there exists > 0 such that for all x A there exists a A such that B (x, ) B (a, a ). Let x, y A with |x y | < . Since x, y B (x, ) B (a, a ), it follows that |x a|, |y a| < a . Thus |f (x) f (y )| = |f (x) f (a) + f (a) f (y )| < /2 + /2 = . |f (x) f (a)| + |f (a) f (y )| (by the triangle inequality)
152 Example 5.4.5 The continuous function
Section 5.4: Uniform continuity is uniformly continuous.
Proof. Let > 0. By the previous theorem there exists 1 > 0 such that for all a, x [0, 2], if |x a| < 1 then | x a| < . Now let a, x 1 such that |x a| < . Then x+ a xa = | x a| = ( x a) x+ a x+ a |x a| (because x, a 1) 2 < . Finally, set = min {1 , , 1}. Let x, a 0 satisfy |x a| < . Since |x a| < 1, necessarily either x, a 1 or x, a 2. But we have analyzed both cases, and we conclude that | x a| < . Exercises for Section 5.4 5.4.1 Let f : C C be given by f (x) = mx + l for some constants m, l. Prove that f is uniformly continuous. 5.4.2 Which of the following functions are uniformly continuous? Justify your answers. i) f : B (0, 1) C, f (x) = x2 . ii) f : (0, 1] R, f (x) = 1/x. iii) f : R R, f (x) = x21 +1 . iv) f : R \ {0} R, f (x) = |x x| . x , if x = 0, v) f : R R, f (x) = |x| 0, otherwise. 1, if x is rational, vi) f : R R, f (x) = 0, otherwise. vii) Re, Im : C R. viii) The absolute value function from C to R. 5.4.3 Suppose that f, g : A C are uniformly continuous and that c C. i) Prove that cf and f g are uniformly continuous. ii) Is f g uniformly continuous? Prove or give a counterexample.
5.4.4 Let g : [a, b] C be continuous. Prove that the function f : (a, b) C dened by f (x) = g (x) for all x (a, b) is uniformly continuous.
5.4.5 Let f : (a, b) C be uniformly continuous. Prove that there exists a continuous function g : [a, b] C such that f (x) = g (x) for all x (a, b).
5.4.6 If the composition of two uniformly continuous functions uniformly continuous? Prove or give a counterexample.
153
5.4.7 Let f : C C be dened by f (x) = x3 . i) Prove that f is continuous but not uniformly continuous. ii) Find a uniformly continuous function g : C C such that g f is uniformly continuous. iii) Find a uniformly continuous function g : C C such that g f is not uniformly continuous.
Chapter 6: Dierentiation
The geometric motivation comes from lines tangent to a graph of a function f at a point (a, f (a)). For example, on the graph below are two secant lines through (a, f (a)): y f (a)
It appears that the line through (a, f (a)) and (x, f (x)) is closer to the tangent line to the graph of f at (a, f (a)) if x is closer to a. Intuitively, the slope of the tangent line is the limit of the slopes of the secant lines.
6.1
Denition of derivatives
Denition 6.1.1 Let A C, and let a A be a limit point of A. A function f : A C is dierentiable at a if f (x) f (a) lim xa xa
exists. In this case, we call this limit the derivative of f at a, and we use either Newtons (x ) . notation f (a) = (f (x)) x=a or Leibnitzs notation dfdx x= a A function is dierentiable if it is dierentiable at all points in its domain. An alternative way of computing the derivatives is f (a) = df (x) dx = lim
x= a
f (a + h) f (a) , h0 h
as this is simply a matter of writing x as a + h, and a + h = x a if and only if h 0. (See Exercise 6.2.1.) Example 6.1.2 Let m, l be complex numbers, and let f (x) = mx + l. Then f (a) = lim (mx + l) (ma + l) m(x a) f (x) f (a) = lim = lim = m. xa xa xa xa xa xa
Chapter 6: Dierentiation Alternatively, f (a) = lim (m(a + h) + l) (ma + l) mh f (a + h) f (a) = lim = lim = m. h0 h0 h h0 h h
155
Example 6.1.3 Let f (x) = x2 . Then f (a) = lim f (x) f (a) xa xa 2 x a2 = lim h0 x a (x a)(x + a) = lim h0 xa = lim (x + a) = 2a. Alternatively, f (a) = lim f (a + h) f (a) h0 h 2 (a + h) a2 = lim h0 h 2 a + 2ah + h2 a2 = lim h0 h 2 2ah + h = lim h0 h (2a + h)h = lim h0 h = lim (2a + h) = 2a.
h0 h0
From now on, we will mostly use the alternative way of computing derivatives. Example 6.1.4 Let f (x) = 1/x. Then f (a) = lim f (a + h) f (a) h0 h 1 1 = lim a+h a h0 h a a+ h a(a+h) a(a+h) = lim (common denominator in the fractions) h0 h
a a h a(a+h) h0
= lim
156 h h0 a(a + h)h 1 = lim h0 a(a + h) 1 = 2 a by the quotient rule for limits. = lim Example 6.1.5 Let f (x) = f (a) = lim
Section 6.1: Denition of derivatives
x. The domain of f is R0 , and for all a > 0,
f (a + h) f (a) h0 h a+h a = lim h0 h a+h a a+h+ a = lim h0 h a+h+ a (a + h) a = lim (since (x y )(x + y ) = x2 y 2 ) h0 h( a + h + a) h = lim h0 h( a + h + a) 1 = lim h0 a+h+ a 1 = 2 a
by the linear, radical, composite, and quotient rules for limits. It is left to Exercise 6.1.4 to show that f is not dierentiable at 0. Example 6.1.6 Let f (x) = x3/2 . The domain of f is R0 , and for all a 0, f (a) = lim
f (a + h) f (a) h0 h 3 /2 (a + h) a3/2 = lim h0 h 3 /2 (a + h) a3/2 (a + h)3/2 + a3/2 = lim h0 h (a + h)3/2 + a3/2 (a + h)3 a3 (since (x y )(x + y ) = x2 y 2 ) = lim h0 h((a + h)3/2 + a3/2 ) a3 + 3a2 h + 3ah2 + h3 a3 = lim h0 h((a + h)3/2 + a3/2 ) 3a2 h + 3ah2 + h3 = lim h0 h((a + h)3/2 + a3/2 )
Chapter 6: Dierentiation = lim
157
(3a2 + 3ah + h2 )h h0 h((a + h)3/2 + a3/2 ) 3a2 + 3ah + h2 = lim h0 (a + h)3/2 + a3/2 3a2 = 3 /2 a + a3/2 3 = a1/2 2 by the linear, radical, composite, and quotient rules for limits. Note that this f is dierentiable even at 0, whereas the square root function (previous example) is not dierentiable at 0. Note that in all these examples, f is a function from some subset of the domain of f to a subset of C, and we can compute f at a number labelled x rather than a: df (x) f ( z ) f ( x) f (x + h) f (x) = lim = lim . z x h0 dx h zx The h-limit is perhaps preferable to the last limit, where it is z that varies and gets closer and closer to x. f ( x) = Example 6.1.7 The absolute value function is not dierentiable at 0.
0| as h goes to 0. Recall the Proof. Suppose for contradiction that L is the limit of |h|| h negation of limits on page 125. Set = 1. Let > 0. Let h1 = /2 and h2 = /2. Then |h1 0| = |h2 0| < , and by the triangle inequality,
2 = |1 1| =
/2 /2 f (/2) f (/2) = /2 /2 /2 /2 f (h1 + 0) f (0) f (h2 + 0) f (0) = h1 h2 f (h1 + 0) f (0) f (h2 + 0) f (0) = L L h1 h2 f (h1 + 0) f (0) f (h2 + 0) f (0) L + L , h1 h2
f (0) f (0) L or f (h2 +0) L is not strictly smaller than = 1. and so one of f (h1 +0) h1 h2 Thus the limit L does not exist, so that the absolute value function is not dierentiable at 0.
This gives an example of a continuous function that is not dierentiable. (Any continuous function with a jagged graph is not dierentiable.)
Section 6.2: Basic properties of derivatives
6.1.1 Prove that f : C C given by f (x) = x3 is dierentiable everywhere, and compute the derivative function. 6.1.2 Prove that f : C \ {0} C given by f (x) = 1/x2 is dierentiable everywhere, and compute the derivative function. 6.1.3 Prove that f : (0, ) R given by f (x) = 1/ x is dierentiable everywhere, and compute the derivative function. 6.1.4 Prove that the square root function is not dierentiable at 0. 6.1.5 Let f : R R be given by f ( x) = Prove that f is dierentiable at 1. 6.1.6 Let f : R R be given by f ( x) = x2 1, x3 x, if x > 1; if x 1.
Prove that f is continuous but not dierentiable at 1. 6.1.7 Let f : R R be given by f ( x) = x2 , x3 x,
x2 1, if x > 1; 4 x 4x, if x 1.
Prove that f is not continuous and not dierentiable at 1. 6.1.8 Determine if the following function is dierentiable at 1: f (x) = x2 , if x > 1; 2x, if x 1.
if x > 1; if x 1.
6.2
Basic properties of derivatives
Theorem 6.2.1 If f is dierentiable at a, then f is continuous at a. Proof. By denition of dierentiability, a is a limit point of the domain of f and a is in the ) f ( a) exists, by the product rule for limits domain of f . Furthermore, since limh0 f (a+hh f (a+h)f (a) ) exists and equals 0 f (a) = 0. In other words, limh0 (f (a + also limh0 (h h h) f (a)) = 0, so that by the sum rule for limits, limh0 f (a + h) = limh0 (f (a + h) f (a) + f (a)) = 0 + f (a) = f (a). Thus by Exercise 6.2.1, limxa f (x) = f (a), so that f is continuous at a. Theorem 6.2.2 (Basic properties of derivatives) (1) (Constant rule) Constant functions are dierentiable (at all real/complex numbers) and the derivative is 0 everywhere.
159
(2) (Linear rule) The function f (x) = x is dierentiable (at all real/complex numbers) and the derivative is 1 everywhere. Let A be a subset of C and let a A be a limit point of A. Suppose that f, g : A C are dierentiable at a. Then (3) (Scalar rule) For any c C, cf is dierentiable at a and (cf ) (a) = cf (a). (4) (Sum/dierence rule) f g is dierentiable at a and (f g ) (a) = f (a) g (a). (5) (Product rule) f g is dierentiable at a and (f g ) (a) = f (a)g (a) + f (a)g (a). (6) (Quotient rule) If g (a) = 0, then f /g is dierentiable at a and (f /g ) (a) = f ( a) g ( a) f ( a) g ( a) . (g (a))2
Proof. Parts (1) and (2) were already proved in part (1) of Example 6.1.2. (3) follows from cf (a + h) cf (h) f (a + h) f (h) (cf )(a + h) (cf )(h) = lim = c lim = cf (a), h0 h0 h0 h h h lim and (4) follows from the sum rule for limits and from f (a + h) + g (a + h) f (h) g (h) (f + g )(a + h) (f + g )(h) = lim h0 h0 h h f (a + h) f (h) + g (a + h) g (h) = lim h0 h f (a + h) f (h) g (a + h) g (h) + = lim h0 h h = f (a) + g (a). lim The following proves the product rule (5): (f g )(a + h) (f g )(a) f (a + h) g (a + h) f (a) g (a) = lim h0 h0 h h f (a + h) g (a + h) f (a) g (a + h) + f (a) g (a + h) f (a) g (a) = lim (adding 0) h0 h (f (a + h) f (a))g (a + h) + f (a)(g (a + h) g (a)) = lim h0 h g (a + h) g (a)) f (a + h) f (a) g (a + h) + f (a) = lim h0 h h = f (a)g (a) + f (a)g (a), lim where in the last step we used that f and g are dierentiable at a and that g is continuous at a (by Theorem 6.2.1). The proof of the quotient rule is similar, and the reader is invited to provide the commentary on the steps: lim
f g (a
+ h) f g (a) h
h0
= lim
f (a+h) g (a+h)
h0
f ( a) g ( a)
160 = lim
Section 6.2: Basic properties of derivatives f (a + h)g (a) f (a)g (a + h) h0 hg (a + h)g (a) f (a + h)g (a) f (a)g (a) + f (a)g (a) f (a)g (a + h) = lim h0 hg (a + h)g (a) f (a + h) g (a + h) f (a) g (a + h) + f (a) g (a) f (a) g (a + h) = lim h0 hg (a + h)g (a) f (a + h) g (a + h) f (a) g (a + h) f (a) g (a + h) f (a) g (a) = lim h0 hg (a + h)g (a) hg (a + h)g (a) f (a + h) f (a) g (a + h) f (a) g (a + h) g (a) = lim h0 h g (a + h)g (a) g (a + h)g (a) h f (a)g (a) f (a)g (a) = . (g (a))2
Theorem 6.2.3 (Power rule) If n is a positive integer, then (xn ) = nxn1 . Proof #1: Part (1) of Example 6.1.2 with m = 1 and l = 0 proves that (x1 ) = 1, so that the theorem is true when n = 1. Now suppose that the theorem holds for some positive integer n. Then (xn ) = (xn1 x) = (xn1 ) x + xn1 x (by the product rule) = (n 1)xn1 + xn1 = (n 1)xn2 x + xn1 (by induction assumption)
= nxn1 ,
so that the theorem holds also for n, and so by induction also for all positive n. Proof #2: The second proof uses binomial expansions as in Exercise 1.6.6: (x + h)n xn (xn ) = lim h0 h n n k nk x h xn = lim k=0 k h0 h n1 n k nk x h = lim k=0 k h0 h n1 n k nk1 h k=0 k x h = lim h0 h = lim =
n1 k=0 n1 k=0 h0
n k nk1 x h k
n k nk1 x 0 (by the polynomial rule for limits) k
Chapter 6: Dierentiation = n xn1 n1
161
= nxn1 .
Theorem 6.2.4 (Polynomial, rational function rule for derivatives) Polynomial functions are dierentiable at all real/complex numbers and rational functions are dierentiable at all points in the domain. Proof. The proof is an application of the sum, scalar, and power rules from Theorems 6.2.2 and 6.2.3. Theorem 6.2.5 (The composite function rule for derivatives, aka the chain rule) Suppose that f is dierentiable at a, that g is dierentiable at f (a), and that a is a limit point of the domain of g f . Then g f is dierentiable at a, and (g f ) (a) = g (f (a)) f (a). Proof. Let > 0. Since f is dierentiable at a, there exists 1 > 0 such that for all a + h ) f ( a) f (a)| < min {1, /(2|g (f (a))| + 2)}. in the domain of f , if 0 < |h| < 1 then | f (a+hh f (a+h)f (a) | < |f (a)| + 1. By assumption g is For all such h, by triangle inequality, | h dierentiable at f (a), so that there exists 2 > 0 such that for all x in the domain of g , if )g (f (a)) 0 < |x f (a)| < 2 , then | g(xx g (f (a))| < /(2|f (a)| + 2). Since f is dierentiable f ( a) and hence continuous at a, there exists 3 > 0 such that for all x in the domain of f , if |x a| < 3 , then |f (x) f (a)| < 1 . Set = min {1 , 2 }. Let a + h be arbitrary in the domain of g f such that 0 < |h| < . In particular a + h is in the domain of f . If f (a + h) = f (a), then g (f (a + h)) g (f (a)) (g f )(a + h) (g f )(a) g (f (a)) f (a) = g (f (a)) f (a) h h g (f (a + h)) g (f (a)) f (a + h) f (a) = g (f (a)) f (a) f (a + h) f (a) h f (a + h) f (a) g (f (a + h)) g (f (a)) g (f (a)) = f (a + h) f (a) h f (a + h) f (a) + g (f (a)) g (f (a)) f (a) h f (a + h) f (a) g (f (a + h)) g (f (a)) g (f (a)) f (a + h) f (a) h f (a + h) f (a) + |g (f (a))| f (a) h (|f (a)| + 1) + |g (f (a))| 2|f (a)| + 2 2|g (f (a))| + 2 < (|f (a)| + 1) + 2 2 = .
162
Thus if there exists as above but possibly smaller such that f (a + h) = f (a) for all a + h in the domain with 0 < |h| < , the above proves the theorem. Now suppose that for all > 0 there exists h as above such that f (a + h) = f (a). ) f ( a) Then f (a) = limh0 f (a+hh , so that in particular when h varies over those innitely many h getting closer to 0 with f (a + h) f (a), we get that f (a) = 0. Also, for such h, (g f )(a + h) (g f )(a) = g (f (a + h)) g (f (a)) = 0, so that (g f )(a + h) (g f )(a) g (f (a)) f (a) = 0 < . h
This analyzes all the cases and nishes the proof of the theorem. Theorem 6.2.6 (Real and imaginary parts) Let A R, and let a A be a limit point of A. Let f : A C. Then f is dierentiable at a if and only if Re f and Im f are dierentiable at a. Proof. Since all h are necessarily real, Re(f (a + h) f (a)) + i Im(f (a + h) f (a)) f (a + h) f (a) = h h f (a + h) f (a) f (a + h) f (a) = Re + i Im h h
and by the denition of limits of complex functions, the limit of the function on the left exists if and only if the limits of its real and imaginary parts exist. Compare this theorem with Exercise 6.2.5. Theorem 6.2.7 (Derivatives of inverses) Let A, B C, and let f : A B be an invertible dierentiable function whose derivative is never 0. Then for all b B , f 1 is invertible at b, and 1 , ( f 1 ) ( b ) = 1 f (f (b)) or in other words, for all a A, f 1 is invertible at f (a) and (f 1 ) (f (a)) = 1 f (a) .
Proof. For all x B , (f f 1 )(x) = x, so that (f f 1 ) (x) = 1, and by the chain rule, f (f 1 (x)) (f 1 ) (x) = 1. Since f is never 0, the conclusion follows. Example 6.2.8 Let f : [0, ) [0, ) be the function f (x) = x2 . We know that f is dierentiable at all points in the domain and that f (x) = 2x. By Example 6.1.5 and Exercise 6.1.4, the inverse of f , namely the square root function, is dierentiable at all
Chapter 6: Dierentiation positive x, but not at 0. The theorem above applies to positive x (but not to x = 0): ( x) = (f 1 ) (x) = 1 f (f 1 (x)) = 1 2 f 1 ( x ) 1 = . 2 x
163
Theorem 6.2.9 Let n be a positive integer. Then for all non-zero x in the domain of 1 ( n x) = x1/n1 . n
Proof. Let A = R+ if n is even and let A = R \ {0} otherwise. Dene f : A A to be f (x) = xn . We have proved that f is invertible and dierentiable. The derivative is f (x) = nxn1 , which is never 0. Thus by the previous theorem, f 1 = n is dierentiable with 1 1 1 1 1 = x(n1)/n = x1/n1 . ( n x) = = = n n n 1 ( n 1) /n f ( x) n ( x) n n nx Theorem 6.2.10 Let r be an arbitrary rational number and f : R+ R+ be given by f (x) = xr . Then for all x, f (x) = rxr1 . Proof. The power rule and quotient rules prove this in case r an integer, and the previous theorem proves it in case r is one over a positive integer. Now suppose that r = m/n for some integers m, n with n = 0. Since m/n = (m)/(n) is also a quotient of two integers, we may write r = m/n so that m Z and n is a positive integer. Thus f is the composition of exponentiation by m and by 1/n. By the chain rule, 1 1/n1 m (m1)/n+1/n1 x = x = rxm/n1/n+1/n1 = rxr1 . n n This proves the theorem for all rational r . f (x) = m(x1/n)m1 The theorem also holds for all real r . The proofs involving only algebra with limits are very laborious, but with basic calculus the proof is much simpler, so it is postponed to Theorem 7.4.4. Exercises for Section 6.2 6.2.1 Let f : A C be a function, and let a A be a limit point of A. Let g : A B be a continuous invertible function such that its inverse is also continuous. Prove that limxa f (x) = L if and only if limz g(a) f (g 1 (z )) = L. 6.2.2 Provide the commentary for the proof of the quotient rule for derivatives in Theorem 6.2.2. 6.2.3 Prove the general power rule for derivatives: If f is dierentiable at a, then (f (x))n x=a = n(f (a))n1f (a). 6.2.4 Prove the general power rule for derivatives: If f is dierentiable at a, then for every positive integer n, f n is dierentiable at a, and (f n ) (a) = n(f (a))n1f (a).
164
6.2.5 Prove that the absolute value function on R is dierentiable at all non-zero a R. Prove that the absolute value function on C is not dierentiable at any non-zero a C. (Hint: let h = (r 1)a for r near 1.)
6.2.6 The following information is known: c f (c) f (c) g (c) g (c) 0 1 2 6 4 1 1 0 5 3 2 2 3 6 6 3 4 2 3 5 4 0 1 4 7 For each of the following, either provide the derivative or write not enough information. In any case, justify every answer. i) (f g ) (1) = ii) (f g ) (2) = iii)
f g
(3) =
6.2.7 A function f is dierentiable on (2, 5) and f (3) = 4, f (3) = 1. Let g (x) = 3x. For each of the statements below determine whether it is true, false, or if there is not enough information. Explain your reasoning. i) f is constant. ii) The slope of the tangent line to the graph of f at 3 is 4. iii) f is continuous on (2, 5). iv) The derivative of (f g ) at 1 is 3. v) (f + g ) (3) = 2.
x 1 6.2.8 Let f (x) = x 2 . i) Find all a in the domain of f such that the tangent line to the graph of f at a has slope 1. ii) Find all a in the domain of f such that the tangent line to the graph of f at a has slope 2. (Solutions need not be real.)
iv) (g f ) (4) =
6.2.11 Make the requisite assumptions about the trigonometric functions to prove that 1 (tan1 ) (x) = 1+ x2 .
6.2.10 Make the requisite assumptions about the trigonometric functions to prove that 1 . (cos1 ) (x) = 1 x 2
6.2.9 We assume the following familiar properties of the trigonometric functions: sin is dierentiable on R, sin = cos, and when restricted to [, ], sin has an inverse. Use Theorem 6.2.7, the fact that (sin(x))2 + (cos(x))2 = 1, and that on [, ] cosine is non1 . negative to prove that (sin1 ) (x) = 1 x2
165
6.3
The Mean value theorem

In this section the domains and codomains of all functions are subsets of R.
Theorem 6.3.1 (Rolles theorem) Let a, b R with a < b, and let f : [a, b] R be a continuous function such that f is dierentiable on (a, b). If f (a) = f (b), then there exists c (a, b) such that f (c) = 0. Proof. By Extreme value theorem (Theorem 5.2.5) there exist l, u [a, b] such that f achieves its minimum at l and its maximum at u. If f (l) = f (u), then the minimum value of f is the same as the maximum value of f , so that f is a constant function, and so f (c) = 0 for all c (a, b). Thus we may assume that f (l) = f (u). Suppose in addition that f (l) = f (a). Then )f (l) a < l < b. For all x [a, b], f (x) f (l), so that in particular for all x (a, l), f (xx 0 l
)f (l) )f (l) and for all x (l, b), f (xx 0. Since f (l) = limxl f (xx exists, it must be both l l non-negative and non-positive, so necessarily it has to be 0. Now if instead f (l) = f (a), then by the assumption that f (l) = f (u), it follows that f (u) = f (u). Then a < u < b. ) f (u ) 0 and For all x [a, b], f (x) f (u), so that in particular for all x (a, u), f (xx u ) f (u ) ) f (u ) for all x (u, b), f (xx 0. Since f (u) = limxu f (xx exists, it must be both u u non-negative and non-positive, so necessarily it has to be 0. Thus in all cases we found c (a, b) such that f (c) = 0.
Theorem 6.3.2 (Mean value theorem) Let a, b R with a < b, and let f : [a, b] R be a continuous function such that f is dierentiable on (a, b). Then there exists c (a, b) such ) f ( a) . that f (c) = f (bb a
) f ( a) Here is an illustration of this theorem: the slope f (bb of the line from (a, f (a)) to a (b, f (b)) also equals the slope of the tangent line to the graph at some c between a and b:
) f ( a) Proof. Let g : [a, b] R be dened by g (x) = f (x) f (bb (x a). By the sum and a scalar rules for continuity and dierentiability, g is continuous on [a, b] and dierentiable
166
Section 6.3: The Mean value theorem
) f ( a) (b a) = f (b) (f (b) f (a)) on (a, b). Also, g (a) = f (a) and g (b) = f (b) f (bb a = f (a) = g (a). Thus by Rolles theorem, there exists c (a, b) such that g (c) = 0. But
g ( x) = f ( x) so that 0 = g (c) = f (c)

f ( b ) f ( a) , b a
whence f (c) =
f (b) f (a) , ba
The rest of this section consists of various applications of the Mean value theorem. More concrete examples are left for the exercises. Theorem 6.3.3 Let a, b R with a < b, and let f : [a, b] R be a continuous function such that f is dierentiable on (a, b). (1) If f (c) 0 for all c (a, b), then f is non-decreasing on [a, b]. (2) If f (c) > 0 for all c (a, b), then f is strictly increasing on [a, b]. (3) If f (c) 0 for all c (a, b), then f is non-increasing on [a, b]. (4) If f (c) < 0 for all c (a, b), then f is strictly decreasing on [a, b]. (5) If f (c) = 0 for all c (a, b), then f is a constant function. Proof. We only prove part (2). Let x, y [a, b] with x < y . By Theorem 6.3.2 there exists ) f (y ) c (x, y ) such that f (c) = f (xx . Since f (c) > 0 and x < y , necessarily f (x) < f (y ). y Since x and y were arbitrary with x < y , then f is strictly increasing on [a, b]. Example 6.3.4 Assuming that sin (x) 1 for all x, we have that sin(x) x for all x 0. Proof. Let f (x) = x sin(x). Then f is dierentiable on R, and f (x) = 1 sin (x) 0. By the previous theorem, f is non-decreasing, so that for all x 0, x sin(x) = f (x) f (0) = 0, whence x sin(x). Theorem 6.3.5 (Cauchys Mean value theorem) Let a < b be real numbers and let f, g : [a, b] R be continuous functions that are dierentiable on (a, b). Suppose that g is injective and the derivative of g is never 0. Then there exists c (a, b) such that 0 = f (c)(g (b) g (a)) g (c)(f (b) f (a)). If g (c) = 0 and g (b) = g (a), this says that f (b) f (a) f ( c) = . g ( c) g (b) g (a)
f ( b ) f ( a) . b a
Proof. Dene h : [a, b] R by h(x) = f (x)(g (b) g (a)) g (x)(f (b) f (a)). Then h is continuous on [a, b] and dierentiable on (a, b). Note that h(a) = f (a)(g (b) g (a)) g (a)(f (b) f (a)) = f (a)g (b) g (a)f (b) = h(b). Then by the Mean value theorem (Theorem 6.3.2) there exists c (a, b) such that h (c) = 0, i.e., 0 = f (c)(g (b) g (a)) g (c)(f (b) f (a)). Theorem 6.3.6 (LH opitals rule) Let f, g : [a, b) R be dierentiable on (a, b). Suppose (x ) = L, and that for all x (a, b), g (x) = 0 and that f (a) = g (a) = 0, that limxa+ f g (x ) g (x) = 0. Then limxa+
f (x ) g (x )
= L.
167
Similarly, if f, g : (c, a] R are dierentiable on (c, a), f (a) = g (a) = 0, (x ) f (x ) limxa f g (x) = L, and for all x (c, a), g (x) = 0 and g (x) = 0, then limxa g (x) = L. Thus if c < a < b, if f, g : (c, b) are dierentiable on (c, a) (a, b), f (a) = g (a) = 0, (x ) (x ) = L, and for all x = a, g (x) = 0 and g (x) = 0, then lim f = L. limxa f g (x ) g (x )
xa
Proof. We only prove the last part here, and the proofs of the other two parts are similar. (x ) Let > 0. Since lim f g (x) = L, there exists > 0 such that for all x in the domain,
(x ) if 0 < |x a| < then | f g (x) L| < . For the same x, g (x) g (a) = g (x) = 0, so that by xa
Theorem 6.3.5 there exists a number c strictly between x and a such that
f (x ) g (x ) .
f (c) g (c)
But then 0 < |c a| < , so that f ( c) f (x) f (a) L = L < . g (x) g (a) g ( c)
f ( x ) f ( a) g ( x ) g ( a)
Example 6.3.7 Compute lim Proof #1: By Theorem 1.5.3, orem 4.3.6,
2 lim x3 1 x1 x 1
x 2 1 . 3 x1 x 1
x 2 1 x 3 1 lim 2x+1 x1 x +x+1 x1
= =
(x1)(x+1) , (x1)(x2 +x+1) 1+1 2 12 +1+1 = 3 .
so that by Exercise 4.3.11 and The-
Proof #2: By Theorem 4.3.5, lim (x2 1) = 0 = lim (x3 1), and lim (x2 1) = lim 2x = 2,
x1
lim (x 1) = lim 3x = 3, so that by Theorem 6.3.6,

x1
x1
Another form of LH opitals rule is in the exercises.
x1 x 2 1 lim 3 equals x1 x 1
2/3.
x1
Exercises for Section 6.3 6.3.1 (Bernoullis inequality) Prove that for all x R0 and all n N0 , (1 + x)n 1 + nx. (Hint: mimic Example 6.3.4.) 6.3.2 Prove the remaining parts of Theorem 6.3.3. 6.3.3 Compute lim
2 x 64 . 3 x+2 x8
6.3.4 What is wrong with the following: Since limx2 x2 +3 that limx2 2 x3 = 2.
x0 x0
2x 2
= 2, by LH opitals rule we have
6.3.5 Explicitly assume the necessary properties of trigonometric functions to prove that x) x) lim sin( = 1, and that lim 1cos( = 0. x x 6.3.6 Let n be a positive integer. i) Use LH opitals rule to prove that limx1 ously, say how. n an n1 . ii) Repeat for limxa x x a = na
x n 1 x 1
= n. We have proved this previ-
168
Section 6.4: Higher-order derivatives, Taylor polynomials
6.3.7 (Here the goal is to prove another version of LH opitals rule) Let f, g : (a, ) R (x ) be dierentiable. Suppose that lim f (x) = lim g (x) = , and that lim f g (x ) = L .
x x xa
(x ) i) Let > 0. Prove that there exists N > a such that for all x > N , | f g (x) L| < . ii) Prove that there exists N > N such that for all x N , f (x), g (x) > 0. iii) Prove that there exists N > N such that for all x N , f (x) > f (N ) and g (x) > g (N ). (c) iv) Prove that for all x > N there exists c (N , x) such that f g (c) =
v)
6.3.8 Assuming that the derivative of ln(x) is 1/x, prove that lim
x 1/ ln x ln x , 1/x
f (x) 1f (N )/f (x) g (x) 1g (N )/g (x) . (x ) Prove that lim f xa g (x)
= L.
x ln x x
= 0.
x0
6.3.9 Be explicit about all the facts about ln that you need to prove that lim+ x ln x = 0. (Hint: also use one of x ln x =
x0+
or x ln x =
and perhaps not both work.)
6.3.10 Compute lim xx . (Hint: compute lim ln(xx ) and use LH opitals rule.)
x0+
6.4
Higher-order derivatives, Taylor polynomials
Let f be a real-valued function dened on an open set containing a. If f is continuous at a, then near a, f is approximately the constant function f (a). If f is dierentiable near a and the derivative is continuous at a, then near a, f is approximately the linear function f (a)(x a) + f (a), i.e., f is approximated by its tangent line. This game keeps going, but for this we need higher-order derivatives: Denition 6.4.1 Let f be dierentiable. If f is dierentiable, we let the derivative of f be f , or f (2) . If f (n1) is dierentiable, we denote its derivative as f (n) . For completeness, f (1) = f and f (0) = f . For example, if f (x) = xm , then for all n m, f (n) (x) = m(m 1)(m 2) (m n + 1)xmn . Theorem 6.4.2 Let f be a function with derivatives of orders 1, 2, . . . , n existing at a point a in the domain. The Taylor polynomial of f (centered) at a of order n is Tn,f,a (x) = f (a)+ f (a)(x a)+ f (n) (a) f (a) (x a)2 + + (x a)n = 2! n!
n k=0
f (k) (a) (x a)k . k!
Chapter 6: Dierentiation Example 6.4.3 If f (x) = x4 3x3 + 4x2 + 7x 10, then with a = 0, T1,f,0(x) = 10 + 7x, T0,f,0(x) = 10,
169
T2,f,0(x) = 10 + 7x + 4x2 ,
Tn,f,0 (x) = 10 + 7x + 4x2 3x3 + x4 for all n 4, and for a = 1, T0,f,1 (x) = 1,
T3,f,0(x) = 10 + 7x + 4x2 3x3 ,
T2,f,1 (x) = 1 + 10(x 1) + (x 2)2 ,
T1,f,1 (x) = 1 + 10(x 1),
Tn,f,1 (x) = 1 + 10(x 1) + (x 1)2 + (x 1)3 + (x 1)4 for all n 4. Note that for all n 4, Tn,f,0 (x) = Tn,f,1 (x) = f (x). The following is a generalization of this observation: Theorem 6.4.4 If f is a polynomial of degree at most d, then for any a C and any integer n d, the nth-order Taylor polynomial of f centered at a equals f . Proof. Write f (x) = c0 + c1 x + + cd xd for some c0 , c1 , . . . , cd C. By elementary algebra, it is possible to rewrite f in the form f (x) = e0 + e1 (x a) + e2 (x a)2 + + ed (x a)d for some e0 , e1 , . . . , ed C. Now observe that f (a) = e0 , f (a) = e1 , f (a) = 2e2 , f (a) = 6e3 = 3!e3 , f (4) (a) = 24e4 = 4!e4 , . . . f (k) (a) = 0 if k > d. But then for n d,
n
T3,f,1 (x) = 1 + 10(x 1) + (x 1)2 + (x 1)3 ,
f (k) (a) = k !ek if k d,
Tn,f,a (x) =
k=0
f (k) (a) (x a)k = k!
d k=0
k !ek (x a)k = k!
d k=0
ek (x a)k = f (x).
170
Section 6.4: Higher-order derivatives, Taylor polynomials
Theorem 6.4.5 (Taylors remainder theorem) Let f be a real-valued function dened on an open interval containing a with continuous derivatives of orders 1, 2, . . . , n + 1. Then for all x B (a, r ) there exist c, d between a and x such that f (x) = Tn,f,a (x) + f (x) = Tn,f,a (x) + f (n+1) (c) (x a)(x c)n , (n + 1)! f (n+1) (d) (x a)n+1 . (n + 1)!
n f ( k ) (t ) k k=0 k! (x t) . Then g (k ) n n f (k+1) (t) (x t)k k=0 kf k! (t) (x t)k1 k=0 k! (k ) (n+1) n (t ) t)k k=1 f (x t)k1 = f n! (t) (x t)n . (k1)!
is dierentiable on B (a, r ), and g (t) =

(n+1)
Proof. Let g : (a r, a + r ) R be dened by g (t) =

n1
(k+1)
c)n (x a) = f (x) Tn,f,a (x), which proves the rst formulation. By Theorem 6.3.5 applied to functions g (t) and h(t) = (x t)n+1 , there exists d between x and a such that h (d)(g (x) g (a)) = g (d)(h(x) h(a)). In other words, (n + 1)(x d)n (f (x) Tn,f,a (x)) = which proves the second formulation. More on Taylor polynomials and Taylor series is in Section 9.3. Exercises for Section 6.4 6.4.1 Prove that if f is a polynomial function, then for every a R, Tn,f,a = f for all n greater than or equal to the degree of f . 6.4.2 Compute the Taylor polynomial of f (x) = 1 + x of degree 5 centered at a = 0. 6.4.3 Compute the Taylor polynomial of f (x) = 1 x of degree 10 centered at a = 0. f (n+1) (d) (x d)n (0 (x a)n+1 ), n!
= f n! (t) (x t)n + k=0 f k! (t) (x Note that g (x) = f (x), g (a) = Tn,f,a (x), If the domain is a subset of R, then by the Mean ) g ( a) value theorem (Theorem 6.3.2), there exists c (0, 1) such that g (c) = g(xx , i.e., a
f (n+1) (c) (x n!
6.4.4 Let n N. 1 i) Compute the Taylor polynomial of f (x) = 1 x of degree n centered at a = 0. ii) Compute Tn,f,0 (0.5). iii) Use Theorem 6.4.5 to determine n such that |Tn,f,0 (0.5) f (0.5)| < 0.001.
6.4.5 Explicitly assume the necessary properties of trigonometric functions to compute the Taylor polynomials of degree n of sin and cos centered at a = 0. i) Estimate the error term with Theorem 6.4.5. ii) To compute values of trigonometric and also exponential functions, calculators and computers use Taylor polynomials. Compute sin(1) and sin(100) to within 0.01 of true value. What degree Taylor polynomial suces for each? Use Theorem 6.4.5.
Chapter 7: Integration
The basic motivation for integration is computing areas of regions bounded by functions. In this chapter we develop the theory of integration for functions whose domains are subsets of R. The rst two sections handle only codomains in R, and at the end of Section 7.3 we extend integration to functions with codomains in C. We do not extend to domains being subsets of C as that would require multi-variable methods, which is not the subject of this course.
7.1
Approximating areas
In this section, domains and codomains of all functions are subsets of R. Thus we can draw the regions and build the geometric intuition together with the formalism. Let f : [a, b] R. The basic aim is to compute the signed area of the region bounded by the x-axis, the graph of y = f (x), and the lines x = a and x = b. By signed area we mean that we add up the areas of the regions above the x-axis and subtract the areas of the regions below the x-axis. Thus a signed area may be negative or zero, even if the region is not just the x-axis. y = f ( x) y
In the plot above, there are many (eight) regions whose boundaries are some of the listed curves, but only the shaded region (comprising of two of the eight regions in the count) is bounded as a subset of the plane. The simplest case of an area is of course when f is a constant function with constant value c. Then the signed area is c (b a), which is positive if c > 0 and non-positive if c 0.
172
Section 7.1: Approximating areas For a general f , we can try to approximate the area by rectangles, such as the following: y = f ( x) y
It may be hard to decide how close the approximation is to the true value. But we can approximate the region more systematically, by having heights of the rectangles be either the least possible height or the largest possible height, as below:
Then clearly the true area is larger than the sum of the areas of the darker rectangles on the left and smaller than the sum of the areas of the darker rectangles on the right. We establish some notation for all this. Denition 7.1.1 A partition of [a, b] is a nite subset of [a, b] that contains a and b. We typically write a partition in the form P = {x0 , x1 , . . . , xn }, where x0 = a < x1 < x2 < < xn1 < xn = b. (For example, in the gures above, n = 10.) Let f : [a, b] R be a bounded function. For each i = 1, . . . , n, let mi = inf {f (x) : x [xi1 , xi ]}, Mi = sup {f (x) : x [xi1 , xi]}. By the bounded assumption, mi , Mi R. The lower sum of f with respect to P is
n
L(f, P ) =
i=1
mi (xi xi1 ).
Chapter 7: Integration The upper sum of f with respect to P is

n
173
U (f, P ) =
i=1
Mi (xi xi1 ).
We have established that for all partitions P of [a, b], L(f, P ) the signed area U (f, P ). In particular, if U (f, P ) L(f, P ) < , then either U (f, P ) or L(f, P ) serves as an approximation of the true signed area within of its true value. For most functions a numerical approximation is the best we can hope for. Example 7.1.2 Approximate the area under the curve y = f (x) = 2x 3 between x = 1 and x = 4. We rst establish a partition Pn of [1, 4] into n equal subintervals. The length of each subinterval is (4 1)/n, and x0 = 1, so that x1 = x0 + 3/n = 1 + 3/n, x2 = x1 + 3/n = 1 + 2 3/n, and in general, xi = 1 + i 3/n. Note that xn = 1 + n 3/n = 4, as needed. Since f is increasing on [1, 4], necessarily mi = f (xi1 ) and Mi = f (xi ). For 13 example, if n = 1, L(f, P ) = f (1) 3 = 3 4 and U (f, P ) = f (4) 3 = 3 2 . Thus the true 3 and 3 213 . Admittedly, this is not much information. area is some number between 4 A computer program produced the following better numerical approximations: n 10 100 1000 10000 100000 1000000 L(f, P ) 651.524 1436.57 1543.79 1554.81 1555.92 1556.03 U (f, P ) 3109.049 1682.32 1568.36 1557.268 1556.16 1556.05123
2
Notice how the lower sums get larger and the upper sums get smaller as we take ner partitions. We conclude that the true area is between 1556.03 and 1556.05124. This is getting closer but may still be insucient precision. For more precision partitions would have to get even ner, but the calculations slow down too. The observed monotonicity of lower and upper sums is not a coincidence: Theorem 7.1.3 Let P, R be partitions of [a, b] such that P R. (Then R is called a renement of P , and R is said to be a ner partition than P .) Then L(f, P ) L(f, R), U (f, P ) U (f, R).
Proof. Write P = {x0 , x1 , . . . , xn }. Let i {1, . . . , n}. Let R [xi1 , xi ] = {y0 = xi1 , y1 , . . . , ym1 , ym = xi }. By set inclusion, inf {f (x) : x [yj 1 , yj ]} inf {f (x) : x
174 [xi1 , xi ]}, so that L(f, {y0, y1 , . . . , ym1 , ym }) =
Section 7.1: Approximating areas
m j =1 m j =1
inf {f (x) : x [yj 1 , yj ]}(yj yj 1 ) inf {f (x) : x [xi1 , xi ]}(yj yj 1 )

m
= inf {f (x) : x [xi1 , xi ]}
j =1
( y j y j 1 )
= inf {f (x) : x [xi1 , xi ]}(xi xi1 ), and hence that

n n
L(f, R) =
i=1
L(f, R [xi1 , xi ])
i=1
inf {f (x) : x [xi1 , xi]}(xi xi1 ) = L(f, P ).
The proof for upper sums is similar. Example 7.1.4 Let f (x) = 1, if x is rational; Then for any partition P of [2, 4], 0, if x is irrational. n for all i, mi = 0 and Mi = 1, so that L(f, P ) = 0 and U (f, P ) = i=1 1 (xi xi1 ) = xn x0 = 4 (2) = 6. Thus in this case, changing the partition does not produce better approximations.
Theorem 7.1.5 For all partitions P and Q of [a, b], and for any bounded function f : [a, b] R, L(f, P ) U (f, Q). Proof. Let R = P Q. (Then R is a renement of both P and Q.) By Theorem 7.1.3, L(f, P ) L(f, R) and U (f, P ) U (f, R). Since always L(f, R) U (f, R), the conclusion follows by transitivity of It follows that the set of all lower sums of f as P varies over all the partitions of [a, b] is bounded above, so that the set of all lower sums has a real least upper bound. Similarly, the set of all upper sums is bounded below and has a real greatest lower bound. Denition 7.1.6 The lower integral of f over [a, b] is L(f ) = sup {L(f, P ) : as P varies over partitions of [a, b]}, and the upper integral of f over [a, b] is U (f ) = inf {U (f, P ) : as P varies over partitions of [a, b]}. We say that f is integrable over [a, b] when L(f ) = U (f ). We call this common value the integral of f over [a, b], and we write it as
b b b
f=
a a
f (x) dx =
a
f (t) dt.
175
All the work shows that L(f ) U (f ), and at least Example 7.1.4 shows that sometimes strict inequality holds. Clearly all constant functions are integrable. Note that we have not yet proved that the function in Example 7.1.2 is integrable, but in the next section we will prove that every continuous function is integrable over a closed bounded interval. Clearly if f is a constant function f (x) = c for all x, then all lower and all upper sums b are c(b a), so that a f = c(b a). A note about notation: In the denition of integral we sometimes write dx or dt, and sometimes we do not. There is no need to write dx when we are simply integrating a function f : we seek the signed area determined by f over the domain from a to b. For this it does not matter if we like to plug x or t or anything else into f . Recall that when f is a function from a subset of R to R, then f (x) or f (t) do not denote a function but a value of f at x and t respectively. If x and t are non-dependent variables, by the constant rule we then have
b a
f (x) dt = f (x)(b a),

4
and similarly, by geometric reasoning,

4
x dx = 8,
0 0
x dt = 4x.
Thus writing dx versus dt is important, and omitting it can lead to confusion: is the answer the constant 8 or 4x depending on x? Furthermore, if x and t are not independent, we can get further values too. Say if x = 3t, then
4 4
x dt =
0 0
3t dt = 24.
In short, use notation precisely. Exercises for Section 7.1 7.1.1 Let f (x) = x. i) Let P = {x0 , x1 , . . . , xn } be a partition of [0, 5] into n equal parts. Prove that xi = 5i/n. ii) Compute mi , Mi . iii) Compute L(f, P ), U (f, P ). Use Example 1.5.1 to simplify the two sums into a more readily computable expression. iv) Now think of n as a positive real number. Compute lim L(f, P ), lim U (f, P ) (as in Section 4.5). v) Compute the area under the curve y = f (x) between x = 0 and x = 5 rst with high school geometry methods and then with methods of this section. Justify all.
n n
176
Section 7.2: What functions are integrable?
7.1.2 Repeat the previous exercise with f (x) = x2 on [0, 5]. (Invoke Exercise 1.5.1.) 7.1.3 Use geometry to compute the following integrals:
a
i)
a r
f= r 2 x2 dx = (4x 10) dx =
ii)
0 5
iii)
3
5, 5 2, f , where f (x) = iv) 1 3x, 9 x, 7.1.4 Compute the following integrals

r
if x < 7; if 7 x < 1; if 1 x < 3; if 3 x. (t and x do not depend on each other):
i)
0 5
r 2 x2 dt = (4x 10) dt =
ii)
3
5, 5 2, iii) f , where f (t) = 1 3t, 9 x,
if if if if
t < 7; 7 t < 1; 1 t < 3; 3 t.
7.2
What functions are integrable?

In this section, we continue to assume that all functions are real-valued.
Theorem 7.2.1 Every continuous real-valued function on [a, b] is integrable over [a, b], where a, b R with a < b. Proof. Let f : [a, b] R be continuous. We need to prove that L(f ) = U (f ). By Theorem 2.7.13 it suces to prove that for all > 0, U (f ) L(f ) . So let > 0. By Theorem 5.4.2, f is uniformly continuous, so there exists > 0 such that for all x, c [a, b], if |x c| < then |f (x) f (c)| < /(b a). Let P = {x0 , x1 , . . . , xn} be a partition of [a, b] such that for all i = 1, . . . , n, xi xi1 < . (For example, this can be accomplished as follows: by Theorem 3.8.5, there exists n N such that (b a) < n , and then P can be taken to be the partition of [a, b] into n equal parts.) Then
n
U (f, P ) L(f, P ) =
i=1 n i=1
(sup {f (x) : x [xi1 , xi ]} inf {f (x) : x [xi1 , xi]}) (xi xi1 ) (xi xi1 ) (by uniform continuity since xi xi1 < ) (b a)
Chapter 7: Integration = (b a) = .
n i=1
177 (xi xi1 )
But U (f ) U (f, P ) and L(f ) L(f, P ), so that 0 U (f ) L(f ) U (f, P ) L(f, P ) . Thus U (f ) = L(f ) by Theorem 2.7.13, so that f is integrable over [a, b]. Furthermore we get the following formulation of integrals that looks very technical but is fundamental for applications (see Section 7.5). The proof is simply a more careful rephrasing of the previous proof: Theorem 7.2.2 Let f : [a, b] R be continuous. For each real number r > 1 let (r ) (r ) (r ) Pr = {x0 , x1 , . . . , xnr } be a partition of [a, b] such that each subinterval has length at (r ) (r ) (r ) most (b a)/r . For each i = 1, 2, . . . , nr , let ci [xi1 , xi ]. Then
nr r
lim
i=1
f (ci )(xi xi1 ) =
(r )
(r )
(r )
f.
a
The superscripts tell us the dependence on the real number r , and the limit is over such r . Somewhat sloppier but perhaps notationally saner is the following formulation:
n P ={x0 ,x1 ,...,xn }ner and ner,ci [xi1 ,xi ] b
lim
i=1
f (ci )(xi xi1 ) =
f.
a
Theorem 7.2.3 Suppose that f is integrable over [a, b] and [b, c]. Then f is integrable c b c over [a, c], and a f = a f + b f . Proof. Let > 0. Since f is integrable over [a, b] there exists a partition P of [a, b] such that U (f, P ) L(f, P ) < /2. Similarly there exists a partition Q of [b, c] such that U (f, Q) L(f, Q) < /2. Let R = P Q. Then R is a partition of [a, c], and U (f, R) = U (f, P ) + U (f, Q) and L(f, R) = L(f, P ) + L(f, Q). Thus U (f, R) L(f, R) < , and so the upper integral of f over [a, c] minus the lower integral of f over [a, c] is between 0 and . Thus by Theorem 2.7.13, U (f ) = L(f ), so that f is integrable over [a, c]. Finally,
c a
f = sup {L(f, R) : R a partition of [a, c]} = sup {L(f, P ) + L(f, Q) : P a partition of [a, b], Q a partition of [b, c]} (as the partitions of [a, b] and [b, c] are independent of each other) = U (f ) on [a, b]) + U (f ) on [b, c]) = inf {U (f, P ) + U (f, Q) : P a partition of [a, b], Q a partition of [b, c]} sup {L(f, R) : b R a partition of [a, c]}
= L(f ) on [a, b]) + L(f ) on [b, c])
178
Section 7.2: What functions are integrable? = inf {U (f, R) : b R a partition of [a, c]}
c
inf {U (f, R) : R a partition of [a, c]} =

a
f.
a b
Notation 7.2.4 What could possibly be the meaning of
f if a < b? In our denition

b
of integrals, all partitions started from a smaller a to a larger b to get a f . If we did reverse b and a, then the widths of the subintervals in each partition would be negative xi1 xi = (xi xi1 ), so that all partial sums and both the lower and upper integrals would get the negative value. Thus it seems reasonable to declare
a b b
f =
f.
a
In fact, this is exactly what makes Theorem 7.2.3 work without any order assumptions b c b on a, b, c. For example, if a < c < b, by Theorem 7.2.3, a f = a f + c f , whence c b b b c f = a f c f = a f + b f , which is exactly what Theorem 7.2.3 says. a Furthermore, from
b a
f=
a a
f+
b a
f we then deduce that
a a
f = 0.
Theorem 7.2.5 Let f : [a, b] R be piecewise continuous, that is, there exists a partition P = {x0 , x1 , . . . , xn } of [a, b] such that f is integrable on each [xi1 , xi ]. Then f b x n is integrable on [a, b], and a f = i=1 xii f. 1 Proof. By Theorem 7.2.1, each follows.
xi xi1
f is a real number, and by Theorem 7.2.3 the rest
Theorem 7.2.6 Suppose that f and g are integrable over [a, b], and that c R. Then b b b f + cg is integrable over [a, b] and a (f + cg ) = a f + c a g . Proof. We rst prove that L(cg ) = U (cg ) = cL(g ) = cU (g ). If c 0, then L(cg ) = sup {L(cg, P ) : P a partition of [a, b]} = c sup {L(g, P ) : P a partition of [a, b]} = sup {cL(g, P ) : P a partition of [a, b]}
= cL(g )
= cU (g ), = inf {cU (g, P ) : P a partition of [a, b]} = U (cg ), = c inf {U (g, P ) : P a partition of [a, b]}
= inf {U (cg, P ) : P a partition of [a, b]}
Chapter 7: Integration and if c < 0, then L(cg ) = sup {L(cg, P ) : P a partition of [a, b]} = c inf {U (g, P ) : P a partition of [a, b]} = cU (g ) = cL(g ), = c inf {L(g, P ) : P a partition of [a, b]} = inf {U (cg, P ) : P a partition of [a, b]}
b b
179
= sup {cU (g, P ) : P a partition of [a, b]}
= sup {cL(g, P ) : P a partition of [a, b]}
= U (cg ).
This proves that cg is integrable with a (cg ) = c a g . Let > 0. By integrability of f and cg there exist partitions P, Q of [a, b] such that U (f, P ) L(f, P ) < /2 and U (cg, Q) L(cg, Q) < /2. Let R = P Q. Then R is a partition of [a, b], and by Theorem 7.1.3, U (f, R) L(f, R) < /2 and U (cg, R) L(cg, R) < /2. By Exercise 2.7.8, for every partition of [a, b], and in particular for the partition R, L(f + cg, R) L(f, R) + L(cg, R), and U (f + cg, R) U (f, R) + U (cg, R). Then 0 U (f + cg ) L(f + cg ) U (f + cg, R) L(f + cg, R) < .
U (f, R) + U (cg, R) L(f, R) L(cg, R) Thus U (f + cg ) L(f + cg ) = 0 by Theorem 2.7.13, and so f + cg is integrable. The inequalities L(f + cg, R) L(f, R) + L(cg, R), and U (f + cg, R) U (f, R) + U (cg, R) furthermore prove that L(f )+L(cg ) L(f +cg ) U (f +cg ) U (f )+U (cg ) = L(f )+L(cg ), so that nally
b b b b b
(f + cg ) =
a a
f+
a
(cg ) =
a
f +c
a
g.
Theorem 7.2.7 Let a, b R with a < b. Let f, g : [a, b] R be integrable functions such that f (x) g (x) for all x [a, b]. Then
b b a
g.
a
Here is a picture that illustrates this theorem: the values of g are at each point in the domain greater than or equal to the values of f , and the area under the graph of g is larger than the area under the graph of f :
180
Section 7.2: What functions are integrable? y = g ( x) y = f ( x)
Proof. By assumption on every subinterval I of [a, b], inf {f (x) : x I } inf {g (x) : x I }. Thus for all partitions P of [a, b], L(f, P ) L(g, P ). Hence L(f ) L(g ), and since f and b b g are integrable, this says that a f a g . Exercises for Section 7.2 7.2.1 The following is known: 1 3 2 (3f 4g ), 1 5f , 2 3g . 1 7.2.2 Let f (x) =
1 0 3 2 3 0 2 1
f = 5,
f = 6,
f = 15,
g = 3. Compute
x, if x is rational; 0, if x is irrational. i) Prove that f is not integrable over [0, 1]. ii) Does this contradict Theorem 7.2.5? Justify.
7.2.3 (There is no integration in this exercise.) Let A = [a, a] or A = (a, a), and let B be a subset of C. A function f : A B is an odd function (resp. even function) if for all x A, f (x) = f (x) (resp. f (x) = f (x)). Let n be a positive integer, c0 , c1 , . . . , cn complex numbers, and f the polynomial function f (x) = c0 + c1 x + c2 x2 + + cn xn . i) Suppose that for all even k , ck = 0. Prove that f is an odd function. ii) Suppose that for all odd k , ck = 0. Prove that f is an even function. 7.2.5 Let f : [a, a] R be an even function. Prove that 7.2.6 Let f : [a, b] R be piecewise continuous. i) Prove that |f | is integrable over [a, b]. ii) Prove that
b a
7.2.4 Let f : [a, a] R be an odd function. Prove that
a f = 0. a a a f =2 0 a
f.
b a
|f |.
7.2.7 Find a function f : [0, 1] R that is not integrable over [0, 1] but such that |f | is integrable over [0, 1]. 7.2.8 So far we have seen that every dierentiable function is continuous and that every continuous function is integrable. i) Give an example of a continuous function that is not dierentiable. ii) Give an example of an integrable function that is not continuous.
181
7.2.9 (Improper integral, I) Let f : [a, ) R be a piecewise continuous function. i) Discuss how our construction/denition of integrals fails when the domain is not bounded. N ii) Prove that for all N [a, ), a f exists. N iii) If limN a f exists, we call the limit the (improper) integral of f over [a, ). We denote it a f . Observe that this is a limit of limits. iv) For rational numbers p < 1, compute 1 xp dx. (The same is true for real p < 1, but we have not developed enough properties for such functions.) 3x + 1, if x < 10; v) Let f (x) = . Compute 0 f . 0, otherwise. 7.2.10 Similarly to part (iii) in the previous exercise, formulate 7.2.11 Let f (x) =
b
f and
1, if x (the oor of x) is even; 1, otherwise. i) Sketch the graph of this function. N ii) Prove that for every positive real number N , N f = 0. iii) Prove that f is not integrable over [0, ) or over (, ) in the extended sense from the previous two exercises.
f.
7.2.12 (Improper integral, II) Let f : (a, b] R be a piecewise continuous function. i) Discuss how our construction/denition of integrals fails when the domain does not include the boundaries of the domain. b ii) Prove that for all N (a, b), N f exists. iii) If limN a+
b N b b
f exists, we call this limit the (improper) integral of f over [a, b],
and we denote it a f . Similarly formulate a f if the domain of f is [a, b) or (a, b). 1 iv) Compute 0 x1/2 dx. 1 v) For rational numbers p > 1, compute 0 xp dx. (The same is true for real p > 1, but we have not developed enough properties for such functions.)
7.3
The Fundamental theorem of calculus
Despite rst appearances, it turns out that integration and dierentiation are related. For this we have two versions of the Fundamental theorem of calculus. Theorem 7.3.1 (The Fundamental theorem of calculus, I) Let f, g : [a, b] R such that f is continuous and g is dierentiable with g = f . Then
b a
f = g (b) g (a).
182
Section 7.3: The Fundamental theorem of calculus
Proof. Let P = {x0 , x1 , . . . , xn} be a partition of [a, b]. Since g is dierentiable on [a, b], it is continuous on each [xi1 , xi] and dierentiable on each (xi1 , xi). Thus by the Mean value theorem (Theorem 6.3.2), there exists ci (xi1 , xi ) such that f (ci ) = g (ci ) = g (xi )g (xi2 ) . By the denition of lower and upper sums, xi xi1
n
L(f, P ) But
n i=1
i=1
f (ci )(xi xi1 ) U (f, P ).

n
f (ci )(xi xi1 ) = =
i=1 n i=1
g (xi ) g (xi1 ) (xi xi1 ) xi xi1 (g (xi) g (xi1 ))
= g ( xn ) g ( x0 ) = g (b) g (a), so that L(f, P ) g (b) g (a) U (f, P ), whence L(f ) = sup {L(f, P ) : P a partition of [a, b]} g (b) g (a) inf {U (f, P ) : P a partition of [a, b]}
= U (f ).
Since f is continuous on [a, b], it is integrable by Theorem 7.2.1, so that L(f ) = U (f ), and b so necessarily a f = g (b) g (a). The general notation for applying Theorem 7.3.1 is as follows: if g = f , then
b b
f = g ( x)
a a
= g (b) g (a).
For example, assuming that cos (x) = sin(x), we have that

0 0
sin x dx = cos(x)
= cos( ) ( cos(0)) = (1) + 1 = 2.
If we instead had to compute this integral with upper and lower sums, it would take us a lot longer and a lot more eort to come up with the answer 2. In general, upper and lower sums and integrals are time-consuming and we want to avoid them if possible. The Fundamental theorem of calculus that we just proved enables us to do that for many functions: to integrate f over [a, b] one needs to nd g with g = f . Such g is called an antiderivative of f . For example, if r is a rational number dierent r +1 from 1, then by the power rule (Theorem 6.2.10), an antiderivative of xr is x r +1 . By the
r +1
183
scalar rule for derivatives, for any constant C , x r +1 + C is also an antiderivative. It does not matter which antiderivative we choose to compute the integral:
b a
x dx =
br+1 +C r+1
ar+1 +C r+1
br+1 ar+1 = , r+1 r+1
so that the choice of the antiderivative is irrelevant. Denition 7.3.2 (Indenite integral) If g is an antiderivative of f , we write also f (x) dx = g (x) + C, where C stands for an arbitrary constant.
1 2 x + C, For example, 3x2 dx = x3 + C , 3 dx = 3x + C , t dx = tx + C , x dx = 2 and so on. (Study the dierences and similarities of the last three.) So far we have seen 2x for rational exponents x. Exercise 5.3.6 also allows real exponents, and proves that this function of x is continuous. Thus by Theorem 7.2.1 this function is integrable. We do not yet know 2x dx, but in Theorem 7.4.5 we will see that 2 1 x 2 + C . For 2(x ) dx instead, you and I do not know an antiderivative, 2x dx = ln 2 we will not know one by the end of the course, and there actually is no closed-form antiderivative. This fact is due to a theory of Joseph Liouville (18091882). What is the meaning of closed-form? Here is an oblique answer: Exercise 9.6.5 claims that there ex2 ists an innite power series (sum of innitely many terms) that is an antiderivative of 2(x ) . 2 Precisely because of this innite sum nature, the values of any antiderivative of 2(x ) cannot be computed precisely, only approximately. Furthermore, according to Liouvilles theory, that innite sum cannot be expressed in terms of the more familiar standard functions, and neither can any other expression for an antiderivative. It is in this sense that we say 2 that 2(x ) does not have a closed-form antiderivative. (It is a fact that in the ocean of all functions, those for which there is a closed-form antiderivative form only a tiny droplet.) At this point we know very few methods for computing antiderivatives. We will in time build up rigorously a bigger stash of functions: see the next section (Section 7.4) and the chapter on power series (Section 9.4). The simplest method for nding more antiderivatives is to rst nd a dierentiable function and compute its derivative, and voil a, the original function is an antiderivative of its derivative. For example, by the chain and power rules, (x2 + 3x)100 is an antiderivative of 100(x2 + 3x)99 (2x + 3).
Theorem 7.3.3 (The Fundamental theorem of calculus, II) Let f : [a, b] R be continuous. Then for all x [a, b], f is integrable over [a, x], and the function g : [a, b] R given
184 by g (x) =
x a
Section 7.3: The Fundamental theorem of calculus f is dierentiable on (a, x) with d dx

x
f = f ( x) .
a
Proof. f is integrable over [a, x] by Theorem 7.2.1, thus g is a well-dened function. Let c (a, x). Let > 0. Since f is continuous on [a, b], there exists > 0 such that for all x [a, b], if |x c| < then |f (x) f (c)| < . Thus on [c , c + ] [a, b], f (c) < f (x) < f (c) + , so that by Theorem 7.2.7,
max {x,c} min {x,c}
(f (c) )
max {x,c}
min {x,c}
max {x,c}
(f (c) + ).
min {x,c}
But by integrals of constant functions,

max {x,c} min {x,c}
(f (c) ) = (max {x, c} min {x, c})(f (c) ) = |x c|(f (c) ).

max {x,c} min {x,c}
Thus |x c|(f (c) ) f |x c|(f (c) ).
If x = c, dividing by |x c| and rewriting the middle term gives this says that f ( c)
x
whence then
x c
f f (c) , xc
x c
f (c) , and
x a
x c
f (c) < . Then for all x [a, b], if 0 < |xc| < ,
c
g ( x ) g ( c) f ( c) = xc =
< . Thus lim
f f (c) (by Theorem 7.2.3 and Notation 7.2.4) xc
x c
f af f ( c) xc
g ( x ) g ( c) exists and equals f (c), i.e., g (c) = f (c). xc xc It is probably a good idea to review the notation again. The integral written as
x x x
x a
f can also be
f=
a a
f (t) dt =
a
f (x) dx.
This is a function of x because x appears in the bound of the domain of integration. Note x similarly that a f (t) dz are functions of t and x but not of z . Thus by the Fundamental
Chapter 7: Integration theorem of calculus, II, d dx

x a
185
d f = f (x), and dx
f (t) dz = f (t).
a
g = (Re g ) + i(Im g ) , so that it would make sense to dene a g as the integral of (Re g ) plus i times the integral of (Im g ) . Indeed, this is the denition:
So far we have dened integrals of real-valued functions. By Theorem 7.3.1, if g = f , b b then a f = g (b) g (a), i.e., a g = g (b) g (a). If g is complex-valued, we know that
b
Denition 7.3.4 Let f : [a, b] C be a function such that Re f and Im f are integrable over [a, b]. Dene the integral of f over [a, b] to be
b b b
f=
a a
Re f + i
a
Im f.
The following are then immediate generalizations of the two versions of the Fundamental theorem of calculus: Theorem 7.3.5 (The Fundamental theorem of calculus, I, for complex-valued functions) Let f, g : [a, b] C such that f is continuous and g is dierentiable with g = f . Then
b a
f = g (b) g (a).
Theorem 7.3.6 (The Fundamental theorem of calculus, II, for complex-valued functions) Let f : [a, b] C be continuous. Then for all x [a, b], f is integrable over [a, x], and the x function g : [a, b] R given by g (x) = a f is dierentiable on (a, x) with d dx Exercises for Section 7.3 7.3.1 Compute the integrals below. Clever guessing is ne.
1 x
f = f ( x) .
a
i)
0 1
x3 dx = x dx = 16x(x2 + 4)7 dx = ( x + 2)(x3/2 + 2x + 3)7 dx =
ii)
0 1
iii)
0 1
iv)
0
186
1
Section 7.3: The Fundamental theorem of calculus
7.3.2 Compute the integrals below, assuming that t and x do not depend on each other. i)
0 1
x3 dx = x3 dt =
ii)
0 x
iii)
0 t
x3 dt = x3 dt =
0
iv)
7.3.5 Suppose that f : [a, b] R+ is continuous and that f (c) > 0 for some c [a, b]. b Prove that a f > 0. 7.3.6 (Integration by substitution) By the chain rule for dierentiation, (f g ) (x) = f (g (x))g (x). i) Prove that ii) Prove that
3 2 b a
7.3.3 Below t and x do not depend on each other. Compute the following derivatives, possibly using Theorem 7.3.3. x d i) x3 dx = dx 0 x d ii) t3 dt = dx 0 t d x3 dx = iii) dx 0 t d t3 dt = iv) dx 0 x d v) t3 dx = dx 0 t d x3 dt = vi) dx 0 *7.3.4 Let f : [a, b] R be monotone. Prove that f is integrable over [a, b].
f (g (x))g (x) dx = f (g (x)) + C .
f (g (x))g (x) dx = f (g (b)) f (g (a)).
iii) Compute the following integrals: explicitly state f, g in applying this rule: (2x 4)10 dx = 4x + 3 dx = (2x2 + 3x)10 4x + 3 dx = 2x2 + 3x)10 (8x + 6)
1
3
1 3
2x2 + 3x dx =
187
7.3.7 (Integration by parts) By the product rule for dierentiation, (f g ) (x) = f (x)g (x)+ f (x)g (x). b b i) Prove that a f (x)g (x)) dx = f (b)g (b) f (a)g (a) a f (x)g (x) dx. ii) Prove that f (x)g (x)) dx = f (x)g (x) f (x)g (x) dx. iii) Compute the following integrals: explicitly state f, g in applying this rule:
1 1 1 1
(4x + 3)(5x + 1)10 dx = 4x + 3 dx = 2x + 4
7.3.8 Compute the following improper integrals: 1 i) 1 x 2 dx = 1 1 ii) 0 x dx = 7.3.9 Work out Exercises 7.2.4, 7.2.5, and 7.2.6 for complex-valued functions. 7.3.10 (Mean value theorem for integrals) Let f : [a, b] R be continuous. Prove that there exists c (a, b) such that b 1 f ( c) = f. ba a
7.4
Natural logarithm and the exponential functions
The function that takes a non-zero x to 1/x is continuous everywhere on its domain since it is a rational function. Thus by Theorem 7.2.1 and Notation 7.2.4, for all x > 0, x 1 dx is well-dened. This function has a familiar name: 1 x Denition 7.4.1 The natural logarithm is the function
x x 1
ln x =
1
1 dx = x
1 dt t
for all x > 0. We prove below all the familiar properties of this familiar function. Remark 7.4.2 1 1 dx = 0. (1) ln 1 = 1 x x 1 x 1 dx > 0, and for x (0, 1), ln x = 1 x dx = (2) By geometry, for x > 1, ln x = 1 x 1 1 x x dx < 0. (3) By the Fundamental theorem of calculus (Theorem 7.3.3), for all b R+ , ln is dier1 entiable on (0, b), so that ln is dierentiable on R+ . Furthermore, ln (x) = x .
188
Section 7.4: Natural logarithm and the exponential functions
(4) ln is continuous (since it is dierentiable) on R+ . (5) The derivative of ln is always positive. Thus by Theorem 6.3.3, ln is everywhere increasing. (6) Let c R+ , and set g (x) = ln(cx). By the chain rule, g is dierentiable, and 1 1 g (x) = cx c = x = ln (x). Thus the function g ln has constant derivative 0. It follows by Theorem 6.3.3 that g ln is a constant function. Hence for all x R+ , ln(cx) ln(x) = g (x) ln(x) = g (1) ln(1) = ln(c) 0 = ln(c). This proves that for all c, x R+ , ln(cx) = ln(c) + ln(x). (7) By the previous item, for all c, x R+ , c = ln(c) ln(x). ln x (8) For all non-negative integers n and all c R+ , ln(cn ) = n ln(c). Proof. We prove this by mathematical induction. If n = 0, then ln(cn ) = ln(1) = 0 = 0 ln c = n ln c. Now suppose that equality holds for some n 1. Then ln(cn ) = ln(cn1 c) = ln(cn1 ) + ln(c) by what we have already established, so that by the induction assumption ln(cn ) = (n 1) ln(c) + ln(c) = n ln(c). (9) For all rational numbers r and all c R+ , ln(cr ) = r ln(c).
Proof. We have proved this result if r is a non-negative integer. If r is a negative integer, then r is a positive integer, so that by the previous case, ln(cr ) = ln(1/cr ) = ln(1) ln(cr ) = 0 (r ) ln(c) = r ln(c), which proves the claim for all integers. Now write r = m n for some integers m, n with n = 0. Then n ln(cr ) = n ln(cm/n ) = ln(cm ) = m ln(c), so that ln(cr ) = m n ln(c) = r ln(c). (10) The range of ln is R = (, ). Proof. By geometry, ln(0.5) < 0 < ln(2). Let y R+ . By Theorem 3.8.5, there exists n N such that y < n ln(2). Hence ln 1 = 0 < y < n ln(2) = ln(2n ), so that since ln is continuous, by the Intermediate value theorem (Theorem 5.2.1), there exists x (1, 2n) such that ln(x) = y . If y R , then by the just proved we have that y = ln(x) for some x R+ , so that y = ln(x) = ln(x1 ). Finally, 0 = ln(1). Thus every real number is in the range of ln. (11) Thus ln : R+ R is a strictly increasing continuous and surjective function. Thus by Theorem 2.8.4, ln has an inverse ln1 : R R+ . By Theorem 5.2.4, ln1 is increasing and continuous.
Chapter 7: Integration (12) By Theorem 6.2.7, the derivative of ln1 is (ln1 ) (x) = (13) For all x, y R, 1 = ln1 (x). 1 ln (ln (x))
189
ln1 (x) = ln1 ln 1 ln (y )
ln1 (x) ln1 (y )
= ln1 ln ln1 (x) ln ln1 (y )
= ln1 (x y ) .
We have proved that for all c R+ and r Q, ln1 (r ln(c)) = ln1 (ln(cr )) = cr , and we have proved that for all r R, ln1 (r ln(c)) is well-dened. This allows us to dene exponentiation with real (not just rational) exponents: Denition 7.4.3 Let c R+ and r R. Set cr = ln1 (r ln(c)). This denition gives rise to two functions: (1) The generalized power function with exponent r when c varies and r is constant; (2) The exponential function with base c when r varies and c is constant. Theorem 7.4.4 Let r R. The function f : R+ R+ given by f (x) = xr is dierentiable, with f (x) = rxr1 . This function is increasing if r > 0 and decreasing if r < 0. Proof. By denition, f (x) = ln1 (r ln(x)), which is dierentiable by the chain and scalar rules and the fact that ln and its inverse are dierentiable. Furthermore, the derivative is 1 (r ln(x)) r = r ln f (x) = ln1 (r ln(x)) x = r ln1 (r ln(x) ln(x)) = r ln1 ((r 1) ln(x)) = ln1 (ln(x)) rxr1 . The monotone properties then follow from Theorem 6.3.3. Theorem 7.4.5 Let c R+ . The function f : R R+ given by f (x) = cx is dierentiable, and f (x) = (ln(c))cx . This function is increasing if c > 1 and decreasing if c (0, 1). Proof. By denition, f (x) = ln1 (x ln(c)), which is dierentiable by the chain and scalar rules and the fact that ln1 is dierentiable. Furthermore, the derivative is f (x) = ln1 (x ln(c)) ln(c) = f (x) ln(c) = (ln(c))cx . The monotone properties then follow from Theorem 6.3.3. We next give a more concrete form to ln1 . Denition 7.4.6 Let e = ln1 (1) (so that ln(e) = 1). The constant e is called Eulers constant. Since ln(e) = 1 > 0 = ln(1), by the increasing property of ln it follows that e > 1. If P is a partition of [1, 3] into 5 equal parts, then L(f, P ) = 0.976934, if P is a partition
190
Section 7.4: Natural logarithm and the exponential functions
of [1, 3] into 6 equal parts, then L(f, P ) = 0.995635, and if P is a partition of [1, 3] into 7 equal parts, then L(f, P ) = 1.00937. This in particular means that ln(3) > 1, so that e < 3. By geometry ln(2) < 1, so that e > 2. We conclude that e is a number strictly between 2 and 3. By further similar work, one can get that e = 2.71828. A reader may want to run a few computer/calculator simulations of this. Theorem 7.4.7 ln1 (x) = ex . Proof. By denitions, ex = ln1 (x ln(e)) = ln1 (x ln(ln1 (1))) = ln1 (x 1) = ln1 (x). We have already proved in page 188 that the derivative of ln1 is ln1 : Theorem 7.4.8 (ex ) = ex .
Exercises for Section 7.4 7.4.1 Let c R+ . i) Prove that for all x R, cx = eln(c
b x c a
x
= ex ln(c) .
1 ( cb ln(c)
ii) Prove that if c = 1, then
dx =
ca ).
7.4.2 Use integration by substitution (Exercise 7.3.6) to compute
x x2 +4
dx. xc(x ) dx.

2
7.4.3 Let c R+ . Use integration by substitution (Exercise 7.3.6) to compute 7.4.4 Let c R+ . Use integration by parts (Exercise 7.3.7) to compute 7.4.5 Use integration by parts (Exercise 7.3.7) to compute ln(x) dx.
xcx dx.
7.4.6 (Logarithmic dierentiation) Sometimes it is hard or even impossible to compute 3 the derivative of a function. Try for example f (x) = xx , f (x) = (x2 + 2)x +4 , or f (x) = x2 (x1)3 (x+4)3 x2 +1 . There is another way if the range of the function consists of positive 3 x+2(x+7)3 (x2)4 real numbers: apply ln of both sides, take derivatives of both sides, and solve for f (x). For example, if f (x) = xx , then ln(f (x)) = ln(xx ) = x ln(x), so that x f ( x) = (ln(f (x))) = (x ln(x)) = ln(x) + = ln(x) + 1, f ( x) x so that f (x) = xx (ln(x) + 1). i) Compute the derivative of f (x) = (x2 + 2)x ii) Compute the derivative of f (x) =
2 3 3
+4
x (x1) (x+4)3 x2 +1 . 3 x+2(x+7)3 (x2)4
7.4.7 Use LH opitals rule (Theorem 6.3.6) to prove that limx0
e x 1 x
= 1.
191
7.4.8 For any positive integer n, compute the Taylor polynomial Tn,f,0 for ex of degree n centered at 0 (Theorem 6.4.2). Use LH opitals rule (Theorem 6.3.6) to prove that ex Tn,f,0 (x) 1 = . n +1 x0 x (n + 1)! lim
7.5
Applications of integration
The Fundamental theorems of calculus relate integration with dierentiation. In parb ticular, to compute a f , if we know an antiderivative g of f , then the integral is easy. However, as already mentioned after the Fundamental theorem of calculus I, many functions do not have a closed-form antiderivative. One can still compute denite integrals up to a desired precision, however: we take ner and ner partitions of [a, b], and when U (f, P ) and L(f, P ) are within a specied distance from each other, we know that the true integral is somewhere in between, and hence up to the specied precision either L(f, P ) or b U (f, P ) stands for a f . In applications, such as in science and engineering, many integrals have to be and are computed in this way because of the lack of closed-form antiderivatives. In this section we look at many applications that exploit the original denition of integrals via sums over ner and ner partitions. For many concrete examples we can then solve the integral via antiderivatives, but for many we have to make do with numerical approximation. 7.5.1 Length of a curve Let f : [a, b] R be a continuous function. If the graph of f is a line, then by the Pythagorean theorem the length of the curve is (b a)2 + (f (b) f (a))2 . For a general curve it is harder to determine its length from (a, f (a)) to (b, f (b)). But we can do the standard calculus trick: let P = {x0 , x1 , . . . , xn } be a partition of [a, b]; on each subinterval [xi1 , xi ] we approximate the curve with the line (xi1 , f (xi1)) to (xi , f (xi)), compute the length of that line as (xi xi1 )2 + (f (xi) f (xi1 ))2 , and sum up all the lengths:
n i=1
(xi xi1 )2 + (f (xi) f (xi1 ))2 .
Whether this is an approximation of the true length depends on the partition, but geometrically it makes sense that the true length of the curve equals
n
lim
i=1
(xi xi1 )2 + (f (xi) f (xi1 ))2 ,
as the partitions {x0 , x1 , . . . , xn} get ner and ner. But this is not yet in form of Theorem 7.2.2. For that we need to furthermore assume that f is dierentiable on (a, b). Then
192
Section 7.5: Applications of integration
by the Mean value theorem (Theorem 6.3.2) for each i = 1, . . . , n there exists ci (xi1 , xi) such that f (xi) f (xi1 ) = f (ci )(xi xi1 ). Then the true length of the curve equals
n P ={x0 ,x1 ,...,xn }ner and ner,ci [xi1 ,xi ]
lim
i=1 n
(xi xi1 )2 + (f (ci )(xi xi1 ))2 1 + (f (ci ))2 (xi xi1 ),
b a
P ={x0 ,x1 ,...,xn }ner and ner,ci [xi1 ,xi ]
lim
i=1
and by Theorem 7.2.2, if f is continuous, this equals We just proved:
1 + (f (x))2 dx.
Theorem 7.5.1 If f : [a, b] R is continuous and dierentiable such that f is continuous on [a, b], then the length of the curve from (a, f (a)) to (b, f (b)) is
b
1 + (f (x))2 dx.
a
7.5.2 Volume of a surface area of revolution, disk method Let f : [a, b] R be continuous. We rotate the region between x = a and x = b and bounded by the x-axis and the graph of f around the x-axis. If the graph is a horizontal line, then the rotated region is a disk of height b a and radius f (a), so that its volume is (f (a))2(b a). For a general f we let P = {x0 , x1 , . . . , xn} be a partition of [a, b]; on each subinterval [xi1 , xi] we approximate the curve with the horizontal line y = f (ci ) for some ci [xi1 , xi], we compute the volume of the solid of revolution obtained by rotating that approximated line over the interval [xi1 , xi ] around the x-axis, and sum up all the volumes:
n i=1
(f (ci))2 (xi xi1 ).

n
Geometrically it makes sense that the true volume equals

P ={x0 ,x1 ,...,xn }ner and ner,ci [xi1 ,xi ]
lim
i=1
(f (ci))2 (xi xi1 ),
and by Theorem 7.2.2 this equals
b (f (x))2 dx. a
This proves:
Theorem 7.5.2 If f : [a, b] R is continuous, then the volume of the solid of revolution obtained by rotating around the x-axis the region between x = a and x = b and bounded by the x-axis and the graph of f is
b
(f (x))2 dx.
Chapter 7: Integration 7.5.3 Volume of a surface area of revolution, shell method
193
Let 0 a b, and let f, g : [a, b] R be continuous such that for all x [a, b] f (x) g (x). We rotate the region between y = a and y = b and bounded by the graphs of x = f (y ) and x = g (y ) around the x-axis. If f (x) = c and g (x) = d are constant functions, then the solid of revolution is a hollowed disk, with the outer border of height d c and radius b and the hole inside it has the same height but radius a. Thus the volume is (d c)(b2 a2 ). For general f and g we let P = {y0 , y1 , . . . , yn} be a partition of [a, b]; on each subinterval [yi1 , yi ] we approximate the curve f, g with the horizontal line f (ci), g (ci ) for some ci [yi1 , yi ], we compute the volume of the solid of revolution obtained by rotating that approximated region over the interval [yi1 , yi ] around the x-axis, and sum up all the volumes:
n i=1 2 2 (g (ci) f (ci ))(yi yi 1 ) .
Geometrically it makes sense that the true volume equals

n P ={y0 ,y1 ,...,yn }ner and ner,ci [yi1 ,yi ]
lim
i=1
2 2 (g (ci) f (ci ))(yi yi 1 ) n
P ={y0 ,y1 ,...,yn }ner and ner,ci [yi1 ,yi ]
lim
i=1
(g (ci) f (ci ))(yi + yi1 )(yi yi1 ),
and by Theorem 7.2.2 this equals
b (g (y ) a
f (y ))2y dy . This proves:
Theorem 7.5.3 If 0 a b and f, g : [a, b] R are continuous, then the volume of the solid of revolution obtained by rotating around the x-axis the region between y = a and y = b and bounded by the graphs of x = f (y ) and x = g (y ), is
b
2
a
y (g (y ) f (y )) dy.
3 Example 7.5.4 The volume of the sphere of radius r is 4 3 r .
Proof. We rotate the upper-half circle of radius r centered at the origin around the xaxis. The circle of radius r centered at the origin consists of all points (x, y ) such that x2 + y 2 = r 2 , so we have g (y ) = r 2 y 2 and f (y ) = r 2 y 2 . Thus the volume is
r
4
0
r 2 y 2 dy =
4 ( r 2 y 2 ) 3 /2 3
=
0
4 4 (r 2)3/2 = r 3 . 3 3
194
7.5.4 Surface area of the surface area of revolution We rotate the line segment y = mx from x = 0 to x = b > 0 around the x-axis. We assume for now that m = 0.
In this way we obtain a right circular cone of height b and base radius |m|b. The perimeter of that base circle is of course 2 |m|b. If we cut the cone in a straight line from a side to the vertex, we cut along an edge of length b2 + (mb)2 , and we get the wedge as follows:
perimeter 2 |m|b radius b2 + (mb)2 Without the clip in the circle, the perimeter would be 2 b2 + (mb)2 , but the perimeter of our wedge is only 2 |m|b. Thus the angle subtended by the wedge is proportionally 2|m|b 2|m| . The area of the full circle is radius squared times one half of the 2 = 2 2 1+m2
2 b +(mb)
full angle, and so proportionally the surface area of our wedge is radius squared times one 2|m| half of our angle, i.e., the surface area of this surface of revolution is ( b2 + (mb)2 )2 2 1+m2 2 2 = |m|b 1 + m . Note that even if b < 0, the surface area is the absolute value of mb2 1 + m2 . If instead we rotate this line from x = a to x = b and 0 a < b, or from a < b 0, the surface area of the surface of revolution is the absolute value of m(b2 a2 ) 1 + m2 . Now suppose that we rotate the line y = mx + l around the x-axis, with m = 0 and a < b and both on the same side of the intersection of the line with the x-axis. This
195
intersection is at x = l/m. By shifting the graph by l/m to the right, this is the same as rotating the line y = mx from x = a + l/m to x = b + l/m, and by the previous case the surface area of this is the absolute value of m((b + l/m)2 (a + l/m)2 ) 1 + m2 1 + m2 1 + m2
= m b2 a2 + 2(b a)l/m
= m b2 + 2bl/m + l2 /m2 (a2 + 2al/m + l2 /m2 )
= m(b a)(b + a + 2l/m) 1 + m2 = (b a)(m(b + a) + 2l) 1 + m2 . If instead we rotate the line y = l (with m = 0) around the x-axis, we get a ring whose surface area is (b a)2 |l|, which is the absolute value of (b a)(m(b + a) + 2l) 1 + m2 . Thus for all m, the surface area of the surface of revolution obtained by rotating the line y = mx + l from x = a to x = b around the x-axis is the absolute value of (b a)(m(b + a) + 2l) 1 + m2 , with the further restriction in case m = 0 that a and b are both on the same side of the x-intercept. Now let f : [a, b] R0 be dierentiable (and not necessarily linear). Let P = {x0 , x1 , . . . , xn} be a partition of [a, b]; on each subinterval [xi1 , xi ] we approximate the curve with the line from (xi1 , f (xi1)) to (xi , f (xi)). By the assumption that f takes on only non-negative values we have that both xi1 , xi are both on the same side of the i )f (xi1 ) (x xi )+ f (xi ), so that x-intercept of that line. The equation of the line is y = f (xx i xi1 m= value theorem (Theorem 6.3.2) there exists ci (xi1 , xi) such that f (ci ) = Thus m = f (ci ) and l = f (xi) f (ci )xi .
f (xi )f (xi1 ) xi xi1
i )f (xi1 ) and l = f (xx xi + f (xi). Since f is dierentiable, by the Mean i xi1
f (xi )f (xi1 ) . xi xi1
We rotate that line segment around the x-axis, compute its volume of that solid of revolution, and add up the volumes for all the subparts:
n i=1
(xi xi1 ) (f (ci )(xi + xi1 ) + 2 (f (xi) f (ci )xi ))

n
1 + (f (ci ))2
=
i=1
|(f (ci )(xi + xi1 ) + 2f (xi))|
1 + (f (ci ))2 (xi xi1 ).
By Theorem 7.2.2 then we get the following:
Theorem 7.5.5 If f : [a, b] R0 is dierentiable with continuous derivative, then the surface area of the solid of revolution obtained by rotating around the x-axis the curve
196 y = f (x) between x = a and x = b is

b
|f (x)(x + x) + 2f (x)|
1 + (f (x))2 dx = 2
a
f (x) 1 + (f (x))2 dx.
Exercises for Section 7.5 7.5.1 Compute the perimeter of the circle with methods from this section. 7.5.2 Compute the volume of the sphere of radius r . 7.5.3 Compute the volume of the ellipsoid whose boundary satises x +y + z = 1. a2 b2 b2 (We need multi-variable calculus to be able to compute the volume of the ellipsoid whose 2 y2 z2 boundary satises x a2 + b2 + c2 = 1.) 7.5.4 Compute the volume of a conical pyramid with base radius r and height h. 7.5.5 Compute the volume of a doughnut (you specify its dimensions). 7.5.6 Compute the surface area of the sphere of radius r . 7.5.7 A tiny particle of mass m is rotating around a circle of radius r . The moment of inertia of the particle is I = mr 2 . Instead of a tiny particle we have a thin circular plate of radius 5 m and mass 4 kg . Let b be the thickness of the plate and the mass divided by the volume. The goal of this exercise is to nd its moment of inertia about its diameter. i) Draw the circular plate with center at origin. Let the axis of rotation be the y -axis. ii) Let P be a partition of [5, 5]. Prove that the moment of inertia of the sliver of the 2 plate between xi1 and xi is approximately c2 i 2 25 ci b, where ci [xi1 , xi ]. iii) Prove that the moment of inertia of the rotating circular plate is 25 kg m2 . 7.5.8 If a constance force F moves an object by d units, the work done is W = F d. Suppose that F depends on the position as F (x) = kx for some constant k . Find the total work done between x = a and x = b. 7.5.9 In hydrostatics, (constant) force equals (constant) pressure times (constant) area, and (constant) pressure equals the weight density of the water w times (constant) depth h below the surface. But most of the time we do not have tiny particles but large objects where depth, pressure, and surface areas vary. For example, an object is completely submerged under water from depth a to depth b. At depth h, the cross section area of the object is A(h). Compute the total force exerted on the object by the water.
2 2 2
Chapter 8: Sequences
In this chapter, Sections 8.4 and 8.5 contain identical results in identical order, but the proofs are dierent. You may want to learn both perspectives, or you may choose to omit one of the two sections.
8.1
Introduction to sequences
Denition 8.1.1 An innite sequence is a function s : N+ C. Instead of writing s(n), it is common to write sn , and the sequence s is commonly expressed also in all of the following notations (and obviously many more):
s = {s1 , s2 , s3 , . . .} = {sn } n=1 = {sn }n1 = {sn }nN+ = {sn+3 }n=2 = {sn4 }n>4 = {sn }.
The nth element in the ordered list is called the nth term of the sequence. We note that while the notation {s1 , s2 , s3 , . . .} usually stands for the set consisting of the elements s1 , s2 , s3 , . . ., and the order of a listing of elements in a set is irrelevant. Here, however, {s1 , s2 , s3 , . . .} stands for a sequence, and the order matters. When the usage is not clear from the context, we add the word sequence or set as appropriate. The rst term of the sequence {2n 1}n4 is 7, the second term is 9, et cetera. The point is that even though the notating of a sequence can start with an arbitrary integer, the counting of the terms always starts with 1. Note that sn is the nth term of the the sequence, whereas {sn } = {sn }n1 is the sequence in which n plays a dummy variable. Thus s = {s n } = s n . Examples and notation 8.1.2 (1) The terms of a sequence need not be distinct. For any complex number c, {c} = {c, c, c, . . .} is called a constant sequence. (2) The sequence {(1)n} = {1, 1, 1, 1, . . .} has an innite number of terms, and its range is the nite set {1, 1}. The range of the sequence {in } is the set {i, 1, i, 1}. (3) The sequence s with s(n) = n +4 for all n 1 can be written as {n +4} = {n +4}n1 = {5, 6, 7, . . .} = {n}n5 . Note that {n} stands for a dierent sequence, namely for the sequence {1, 2, 3, . . .}.
198
Section 8.1: Introduction to sequences
(4) The sequence {2n} is an abbreviation for the sequence {2, 4, 6, 8, . . .}. This is not the same as the sequence {4, 2, 8, 6, 12, 10, . . .}, namely, if we scramble the order of the terms, we change the sequence. (5) The sequence of all odd positive integers (in the natural order) can be written as {2n 1} = {2n 1}n1 = {2n + 1}n0 (and in many other ways). (6) Terms of a sequence can be dened recursively. For example, let s1 = 1, and for each n 1 let sn+1 = 2sn . Then of course sn = 2n1 has a non-recursive form. If however, s1 = 1 and for each n 1, sn+1 = 2s3 n + sn + 2 + ln(sn ) + 1, then we can certainly compute any one sn without invoking sk for k < n, but we will not get a closed-form for sn as in the previous example. (7) (The Fibonacci sequence) Let s1 = 1, s2 = 1, and for all n 2, let sn+1 = sn + sn1 . This sequence starts with 1, 1, 2, 3, 5, 8, 13, 21, 34, . . .. One can prove by induction that n n 1 1 1+ 5 1 5 sn = . 2 2 5 5 (It may seem amazing that these expressions with square roots of 5 always yield positive integers.) (8) Some sequences, just like functions, do not have an algebraic expression for terms. For example, let s be the sequence whose nth term is the nth prime number. This sequence starts with 2, 3, 5, 7, 11, 13, 17, 19, 23, and we could write many more terms out explicitly, but we do not have a formula for them. (This s is indeed an innite sequence since there are innitely many primes, as proved on page 25.) (9) Note that {n}nZ is NOT a sequence because the list has no rst term. (10) However, one can scramble the set Z of all integers into a sequence. One way of doing this as as follows: {0, 1, 1, 2, 2, 3, 3, 4, 4, . . .}, which algebraically equals if n = 1; 0, if n is even; sn = n/2, (n 1)/2, otherwise.
(11) One can scramble the set Q+ of all positive rational numbers into a sequence via a diagonal construction as follows. First of all, each positive rational number can be written in the form a/b for some positive integers a, b. Rather than plotting the fraction a/b, we will plot the point (a, b) in the plane. We start counting at (1, 1), which stands for 1/1 = 1. We then proceed through all the other integer points in the positive quadrant of the plane via diagonals as in Plot 8.1.2. The given instructions would enumerate positive rational numbers as 1/1, 2/1, 1/2, 1/3, 2/2. Ah, but 2/2 has already been counted as 1/1, so we do not count 2/2. Thus, the proper counting of
Chapter 8: Sequences positive rational numbers in this scheme starts with:
199
1/1, 2/1, 1/2, 1/3, 2/2, 3/1, 4/1, 3/2, 2/3, 1/4, 1/5, 2/4, 3/3, 4/2, 5/1, 6/1, 5/2, 4/3, 3/4, et cetera, where the crossed out numbers are not part of the sequence because they had been counted. For example, the fth term is 3.
1 1 2 3 4 5 6
Plot 8.1.2 Beginning of the enumeration of the positive rational numbers. It is important to note that every positive rational number appears on this list, and because we are skipping any repetitions, it follows that every positive rational number appears on this list exactly once. Thus this gives an enumeration of positive rational numbers. An algebraic formulation of this sequence in the style of enumeration of Z is not known. (12) A trick incorporating the last two sequencing strategies yields also a scrambling of Q into a sequence (see Exercise 8.1.6). Incidentally, it is impossible to scramble R or C into a sequence by a so-called Cantors diagonal argument which we are not presenting here, but an interested reader can consult other sources. (13) Sequences are functions, and if all terms of the sequence are real numbers, we can plot sequences in the usual manner for plotting functions. The following is part of a plot of the sequence {1/n}. sn 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 n
Plot 8.1.2 sn = 1/n (Note that the corresponding function f : {1/n : n N+ } R is f (1/n) = n.)
200
Section 8.1: Introduction to sequences
(14) Another way to plot a sequence is to simply plot and label each sn in the complex plane or on the real line. We plot two examples below. s1 , s3 , s5 , . . . 1 s2 , s4 , s6 , . . . 1 s1 s3 1 i/2 s2 s4 1 0 s3 s2 s1 1
Plot 8.1.2 Plots of the image sets of {(1)n }, {1/n} and {(1)n /n + 0.5i/n}. There is an obvious arithmetic on sequences (just like there is on functions): { s n } { tn } = { s n tn } , { s n } { tn } = { s n tn } , c{sn } = {csn },
{sn }/{tn} = {sn /tn } (if tn = 0 for all n). One has to make sure to add/multiply/divide equally numbered terms of the two sequences, such as in the following: {n}n3 + {n}n2 = {n + 1}n2 + {n}n2 = {2n + 1}n2 . Here are a few further examples of arithmetic on sequences: {2n } + {2n } = {0},
{in }/{(i)n} = {(1)n}. Exercises for Section 8.1
{(1)nn} + {2/n} = {(1)n n + 2/n},
{2n 1} + {1} = {2n},
{ 2 n } { 2 n } = { 1 } ,
2{2n } = {2n+1 },
8.1.1 Express algebraically the ordered sequence of all positive integer multiples of 3. 8.1.2 Think of a sequence whose terms are all between 2 and 3. 8.1.3 Plot the sequences {n2 } and {1/n2 }. Compare the plots.
Chapter 8: Sequences 8.1.4 Plot the sequences {(1)n } and {cos(n )}.
201
8.1.5 Plot the image set of the sequence s = {in }: draw the real and imaginary axes and the unit circle centered at the origin; on this circle, plot i1 , i2 , i3 , i4 , i5 , i6 , and label each correspondingly with s1 , s2 , s3 , s5 , s6 . Label also s20 , s100, s101, s345. 8.1.6 Prove that Q can be scrambled into a sequence. (Hint: use the scrambling of Q+ .) 8.1.7 Prove that the dierence of the ordered sequence of all positive odd integers and the constant sequence {1} is the ordered sequence of all non-negative even integers. 8.1.8 Prove that + is a binary operation on the set of all innite sequences. Prove that {0} is the identity for this operation. Prove that every innite sequence has an inverse for this operation. 8.1.9 Prove that is a binary operation on the set of all innite sequences. Prove that {1} is the identity for this operation. Show that not every innite sequence has an inverse for this operation. What innite sequences to have an inverse for this operation? 8.1.10 Sequences can also be nite. Two examples of nite sequences are: (i) last exam scores in a class, arranged in alphabetic order by the student; and (ii) last exam scores in a class, arranged in ascending order by score. Give two more exaples of nite sequences.
8.2
Convergence of innite sequences
Denition 8.2.1 A sequence s = {sn } converges to L C if for all real numbers > 0 there exists a positive real number N such that for all integers n > N , |sn L| < . If s converges to L, we also say that L is the limit of {sn }. We use the following notations for this: sn L, {sn } L, lim s = L, lim sn = L, lim{sn } = L, lim sn = L, lim {sn } = L,
n n
and a minor variation on the last two: limn sn = L, limn {sn } = L. For example, the constant sequence s = {c} converges to L = c because for all n, |sn L| = |c c| = 0 is strictly smaller than any positive real number . The sequence s = {300, 5, , 4, 0.5, 106, 2, 2, 2, 2, 2, . . .} converges to L = 2 because for all n 7, |sn L| = |2 2| = 0 is strictly smaller than any positive real number . In conceptual terms, a sequence {sn } converges if the tail end of the sequence gets closer and closer to L; you can make all sn with n > N get arbitrarily close to L by simply increasing N a sucient amount. We work out some examples of epsilon-N proofs. These proofs are in spirit similar to the epsilon-delta proofs, and we go through them slowly at rst.
202
Section 8.2: Convergence of innite sequences
Depending on the point of view of your class, the reader may wish to skip the rest of this section for an alternative treatment of limits of sequences in terms of limits of functions in Section 8.4. Be aware that this section is more concrete; the next section assumes greater ease with abstraction. Example 8.2.2 Consider the sequence s = {1/n}. Plot 8.1.2 gives a hunch that lim sn = 0, and now we prove it. [Recall that text between square brackets in this font and in red color is what should approximately be going through your thoughts, but it is not something to write down in a final solution. By the definition of convergence, we have to show that for all > 0 some property holds. All proofs of this form start with:] Let be an arbitrary positive number. [Now we have to show that there exists an N for which some other property holds. Thus we have to construct an N . Usually this is done in retrospect, one cannot simply guess an N , but in the final write-up, readers see simply . that educated guess more about how to guess educatedly later:] Set N = 1 Then N is a positive real number. [Now we have to show that for all integers n > N , |sn 0| < . All proofs of statements of the form for all integers n > N start with:] Let n be an (arbitrary) integer with n > N . [Finally, we have to prove the inequality |sn 0| < . We do that by algebraically manipulating the left side until we get the desired final < :] |sn 0| = |1/n 0| = |1/n| = 1/n (because n is positive)
< 1/N (because n > N > 0) 1 (because N = 1/) [Ah, what a clever guess that was!] = (1/) = . So we conclude that |sn 0| < , which proves that lim sn = 0. Just as in the epsilon-delta proofs where one has to nd a , similarly how does one divine an N ? In the following two examples we indicate this step-by-step, not as a book or your nal homework solution would have it recorded.
1 Example 8.2.3 Let sn = n {(1)n + i(1)n+1 }. If we write out the rst few terms, we nd that {sn } = {1 + i, 1/2 i/2, 1/3 + i/3, . . .}, and we may speculate that lim sn = 0. Here is plot of the image set of this sequence in the complex plane:
203
s1 s3 1
i/2 s4
s2
Lets prove that lim sn = 0. Let > 0. Set N = . [We will eventually fill in what the positive real number N should be, but at this point of the proof simply leave some blank space.] Let n be an integer strictly bigger than N . [We want to make sure eventually that n is positive, which is guaranteed if N is positive, but with blank N , we will simply assume in the algebra below that N is positive.] Then 1 ((1)n + i(1)n+1 ) 0 n 1 |(1)n + i(1)n+1 | (because |ab| = |a||b|) = n 1 = |(1)n + i(1)n+1 | (because n is positive) n 1 |(1)n | + |i(1)n+1 | (by triangle inequality) n 1 = (1 + 1) n 1 < 2 (because n > N ) N [Aside: we want/need 2/N , and 2/N = is a possibility, so set N = 2/. | s n 0| = Now go ahead, write that missing information on N in line 1 of this proof!] 2 = (because N = 2/) 2/ = ,
which proves that for all n > 1/, |sn 0| < . Since is arbitrary, this proves that lim sn = 0. Thus a polished version of this example looks like this:
1 Lets prove that lim( n ((1)n + i(1)n+1 ) 0) = 0. Let > 0. Set N = 2/. Then N is a positive real number. Let n be an integer strictly bigger than N . Then
| s n 0| =
1 ((1)n + i(1)n+1 ) 0 n
204 =
Section 8.2: Convergence of innite sequences 1 |(1)n + i(1)n+1 | (because |ab| = |a||b|) n 1 = |(1)n + i(1)n+1 | (because n is positive) n 1 (|(1)n| + |i(1)n+1 |) (by triangle inequality) n 1 = (1 + 1) n 1 < 2 (because n > N ) N 2 (because N = 2/) = 2/ = ,
which proves that for all n > 1/, |sn 0| < . Since is arbitrary, this proves that lim sn = 0.
2n+3n Example 8.2.4 Claim: lim 3+4 = 3. Proof: Let > 0. Set N = n+n2 an integer strictly bigger than N . Then
2
. Let n be
2n + 3n2 3(3 + 4n + n2 ) 2n + 3n2 = 3 3 + 4n + n2 3 + 4n + n2 3 + 4n + n2 9 10n = 3 + 4n + n2 9 + 10n = (because n > 0) [Assuming that N > 0.] 3 + 4n + n2 n + 10n (because n 9) [Assuming that N 8.] 3 + 4n + n2 11n = 3 + 4n + n2 11n 2 (because 3 + 4n + n2 > n2 , so 1/(3 + 4n + n2 ) < 1/n2 ) n 11 = n 11 (because n > N ) < N 11 (because N 11/ so 1/N 1/(11/)) [Assuming this] 11/ = , which was desired. Now (on scratch paper) we gather all the information we used about N : N > 0, N 8, N 11/, and that is it. Thus on the rst line we ll in the blank part: Set N = max {8, 11/}, which says that N is either 8 or 11/, whichever is greater, so that N 8 and N 11/.
Chapter 8: Sequences The polished version of this proof would go as follows:
205
2n+3n Example 8.2.5 Claim: lim 3+4 n+n2 = 3. Proof: Let > 0. Set N = max {8, 11/}. Then N is a positive real number. Let n be an integer strictly bigger than N . Then
3(3 + 4n + n2 ) 2n + 3n2 2n + 3n2 = 3 3 + 4n + n2 3 + 4n + n2 3 + 4n + n2 9 10n = 3 + 4n + n2 9 + 10n = (because n > 0) 3 + 4n + n2 n + 10n (because n 9) 3 + 4n + n2 11n = 3 + 4n + n2 11n 2 (because 3 + 4n + n2 > n2 , so 1/(3 + 4n + n2 ) < 1/n2 ) n 11 = n 11 (because n > N ) < N 11 (because N 11/ so 1/N 1/(11/)) 11/ = , which proves that for all n > N , |sn 3| < . Since is arbitrary, this proves that the limit of this sequence is 3. Example 8.2.6 lim n1/n = 1. Proof. For all n 1, n = (n1/n )n = (1 + (n1/n 1))n
n
=
k=0
n (n1/n 1)k (by Exercise 1.6.6) k
1 1 + n(n 1)(n1/n 1)2 (by using only k = 0, 2 parts). 2

2 /n n(n 1)(n1/n 1)2 , so that for n 2, n (n1 1)2 , and hence that Thus n1 1 2 2 + 1 n1/n . It follows that for all n 2, 1 n1/n 2 + 1. Now let > 0. n n 2 Set N = 2 / . Then N is a positive real number. Let n > N be an integer. Then 2 2 2 1 /n 1 n < , which proves that |n1/n 1| < , and 0 n < N = , so that 0 n hence proves this example.
206
Section 8.3: Divergence of innite sequences and innite limits
Exercises for Section 8.2 8.2.1 Let sn = 1/n2 . Fill in the blanks of the following proof that lim{sn } = 0. . Then if n > N , Let > 0. Set N = |sn s| = 1 0 n2 1 = 2 (because n 1 < 2 (because N (because = = .
1 8.2.2 Prove that for any real number L, lim{ n + L} = L. n) 8.2.4 Prove that lim sin( = 0. n n+1 8.2.5 Prove that lim 2 n2 4 = 0.
) ) )
8.2.3 Prove that for any positive real number k , lim{1/nk } = 0.
+4 3 8.2.6 Prove that lim 3n 2n = 2 . 8.2.7 Prove that lim n + 1 n = 0.
Exercise 8.2.8 (This is invoked in Example 8.3.9.) Let a be a positive real number. Prove that lim a1/n = 1. (Hint: rst prove it for a 1.) 8.2.9 Suppose that the terms of a sequence are given by sn = k=1 n 1 = 1 n+1 . i) Using induction on n, prove that k=1 k(k1 +1) ii) Use this to nd and prove the limit of {sn }.
n 1 k(k+1) .
8.2.10 Prove the following limits: ln(n) i) lim ln( = 1. (Hint: epsilon-N proof.) n+1)
ln(ln(n)) ii) lim ln(ln( n+1)) = 1.
8.3
Divergence of innite sequences and innite limits
The sequence {(1)n} alternates in value between 1 and 1, and does not seem to converge to a single number. The following denition addresses this situation. Denition 8.3.1 A sequence diverges if it does not converge. In other words, {sn } diverges if for all complex numbers L, lim{sn } = L. By the usual negation of statements (see chart on page 25), lim{sn } = L means:
Chapter 8: Sequences For all real numbers > 0 there exists a positive real number N such that for all integers n > N , |sn L| < . = There exists a real number > 0 such that there exists a positive real number N such that for all integers n > N , |sn L| < . = There exists a real number > 0 such that for all positive real numbers N , for all integers n > N , |sn L| < . = There exists a real number > 0 such that for all positive real numbers N there exists an integer n > N such that |sn L| < . = There exists a real number > 0 such that for all positive real numbers N there exists an integer n > N such that |sn L| .
207
Example 8.3.2 {(1)n} is divergent. Namely, for all complex numbers L, lim sn = L. Proof. Set = 1 (half the distance between the two values of the sequence). Let N be an arbitrary positive number. If Re(L) > 0, let n be an odd integer greater than N , and if Re(L) 0, let n be an even integer greater than N . In either case, |sn L| | Re(sn ) Re(L)| 1 = . The sequence in the previous example has no limit, whereas the sequence in the next example has no nite limit: Example 8.3.3 For all complex numbers L, lim{n} = L. Proof. Set = 53 (any positive number will work). Let N be a positive real number. Let n be any integer that is strictly bigger than N and strictly bigger than |L| + 53 (say strictly bigger than N + |L| + 53). Such an integer exists. Then by reverse triangle inequality, |n L| |n| |L| 53 = . The last two examples are dierent: the rst one has no limit at all since the terms oscillate wildly, but for the second example we have a sense that its limit is innity. We formalize this: Denition 8.3.4 A real-valued sequence {sn } diverges to if for every positive real number M there exists a positive number N such that for all integers n > N , sn > M . We write this as lim sn = . A real-valued sequence {sn } diverges to if for every negative real number M there exists a positive real number N such that for all integers n > N , sn < M . We write this as lim sn = . Example 8.3.5 lim n = .
208
Section 8.3: Divergence of innite sequences and innite limits
Proof. Let M > 0. Set N = M . (As in epsilon-delta or epsilon-N proofs, we must gure out what to set N to. In this case, N = M works). Let n N with n > N . Since N = M , we conclude that n > M , and the proof is complete. Theorem 8.3.6 (Comparison theorem for innite limits) Let {sn }, {tn} be real-valued sequences such that for all suciently large n (say for n N for some xed N ), sn < tn . (1) If lim sn = , then lim tn = . (2) If lim tn = , then lim sn = . Proof. (1) By assumption lim sn = for every positive M there exists a positive N such that for all integers n > N , sn > M . Hence by assumption tn sn for all n N we get that for every positive M and for all integers n > max {N, N }, tn > M . Thus by denition lim tn = . Part (2) has an analogous proof.
+1 Example 8.3.7 lim n n = . +1 1 Proof. Note that for all n N+ , n n = n+ n n, and since we already know that 2 +1 = . lim n = , it follows by the comparison theorem above that lim n n 1 Example 8.3.8 lim n n = . 1 1 Proof. Note that for all integers n > 2, n n = n n n 2 . We already know that n lim n = , and it is straightforward to prove that lim 2 = . Hence by the comparison 2 1 = . theorem, lim n n Or we can give an M N proof. Let M > 0. Set N = max {2, 2M }. Let n be an integer strictly bigger than N . Then
2 2 2 2
n2 1 1 n N =n > M. n n 2 2 Example 8.3.9 lim n n! = . N! . NN
Proof. Let N be any positive integer. Then for all integers n N , n! = n(n 1) (N + 1) N ! N nN N ! = N n
N! n Thus n n! N n N n! N , and since N is arbitrary, the N . By Exercise 8.2.8 then conclusion follows. Theorem 8.3.10 Let {sn } be a sequence of positive numbers. Then lim sn = if and only if lim s1 = 0. n
209
Proof. Suppose that lim sn = . Let > 0. By the denition of innite limits, there exists a positive number N such that for all integers n > N , sn > 1/. Then for the same n, 0 s1 < , so that | s1 | < . This proves that lim s1 = 0. n n n 1 Now suppose that lim sn = 0. Let M be a positive number. By assumption lim s1 = 0, n 1 there exists a positive number N such that for all integers n > N , | sn 0| < 1/M . Since < 1/M , so that sn > M . This proves each sn is positive, it follows that for the same n, s1 n that lim sn = . Exercises for Section 8.3 8.3.1 Prove that the following sequences diverge: i) { n}. ii) {2 n }. iii) {(n + 1)3 n3 }. iv) {(1)n + 3/n}.
8.3.5 Let {sn } be a sequence of negative numbers. Prove that lim sn = if and only if = 0. lim s1 n 8.3.6 Given the following sequences, nd and prove the limits, nite or innite, if they exist. Otherwise, prove divergence: n+5 i) { n 3 5 } .
1) }. iii) { ( n3
n
8.3.4 Give an example of two divergent sequences {sn }, {tn } such that {sn + tn } converges.
8.3.3 Suppose that {sn } diverges and {tn } converges to a non-zero number. Prove that {sn tn } diverges.
8.3.2 Suppose that {sn } diverges and {tn } converges. Prove that {sn tn } diverges.
n n ii) { 2 3n2 5 }.
n
8.3.7 Prove or give a counterexample: i) If {sn } and {tn } both diverge, then {sn + tn } diverges. ii) If {sn } converges and {tn } diverges, then {sn + tn } diverges. iii) If {sn } and {tn } both diverge, then {sn tn } diverges. iv) If {sn } converges and {tn } diverges, then {sn tn } diverges.
n vii) { 1 n }. n viii) { 2 n! }. n! ix) { (n+1)! }.
1) n iv) { ( n+1 }. n }. v) { 4 5n n3 8n vi) { n2 +8n }.

2
210
Section 8.4: Convergence theorems via functions
8.3.8 Find examples of the following: i) A sequence {sn } such that lim{sn /sn+1 } = 0. ii) A sequence {sn } such that lim{sn+1 /sn } = 1. iii) A sequence {sn } such that lim{sn+1 /sn } = . iv) A sequence {sn } such that lim{sn+1 sn } = 0. v) A sequence {sn } such that lim{sn+1 sn } = .
8.4
Convergence theorems via functions
In this section we exploit the connection between sequences and functions to prove theorems about limits of sequences more easily. Because of having to keep in mind these connections as well as theorems about limits of functions to get at theorems about limits of sequences, a reader may nd this section fairly abstract. In Section 8.5 we give epsilon-N proofs of the same results. The reader may omit one of these two sections. (Exercises are almost identical in the two sections.) For any sequence s we can dene a function f : {1/n : n N+ } C with f (1/n) = sn . Conversely, for every function f : {1/n : n N+ } C we can dene a sequence s with sn = f (1/n). The domain of f has exactly one limit point, namely 0. With this we have the usual notion of limx0 f (x) with standard theorems from Section 4.3, Theorem 8.4.1 Let s, f be as above. Then lim sn = L if and only if limx0 f (x) = L. Proof. () Suppose that lim sn = L. We have to prove that limx0 f (x) = L. Let > 0. By assumption lim sn = L, there exists a positive real number N such that for all integers n > N , |sn L| < . Let = 1/N . Then is a positive real number. Let x be in the domain of f such that 0 < |x 0| < . Necessarily x = 1/n for some positive integer n. Thus 0 < |x 0| < simply says that 1/n < = 1/N , so that N < n. But then by assumption |f (1/n) L| = |sn L| < , which proves that limx0 f (x) = L. () Now suppose that limx0 f (x) = L. We have to prove that lim sn = L. Let > 0. By assumption limx0 f (x) = L there exists a positive real number such that for all x in the domain of f , if 0 < |x 0| < then |f (x) L| < . Set N = 1/ . Then N is a positive real number. Let n be an integer greater than N . Then 0 < 1/n < 1/N = , so that by assumption |f (1/n) L| < . Hence |sn L| = |f (1/n) L| < , which proves that lim sn = L. Example 8.4.2 (Compare the reasoning in this example with the epsilon-N proofs of 52/n 5n2 1/n n2 Section 8.2.) Let sn = 5 3n+4 . We note that sn = 3n+4 1/n = 3+4/n . The corresponding
211
2 x function f : {1/n : n N+ } is f (x) = 5 3+4x , and by the scalar, sum, dierence, and 0 quotient rules for limits of functions, limx0 f (x) = 5 3+0 = 5/3, so that by Theorem 8.4.1, lim sn = 5/3.
Example 8.4.3 Suppose sn =
sum, corresponding function f : {1/n : n N+ } is f (x) = dierence, product, and quotient rules for limits of functions, lim sn = limx0 f (x) = 0+0 10 = 0. Theorem 8.4.4 The limit of a converging sequence is unique. Proof. Let {sn } be a convergent sequence. Suppose that {sn } converges to both L and L . Let f : {1/n : n N+ be the function corresponding to s. By Theorem 8.4.1, limx0 f (x) = L and limx0 f (x) = L . By Theorem 4.3.1, L = L . This proves uniqueness of limits for sequences. Theorem 8.4.5 Suppose that lim{sn } = L and that L = 0. Then there exists a positive number N such that for all integers n > N , |sn | > |L|/2. In particular, there exists a positive number N such that for all integers n > N , sn = 0. Proof. Let f be the function corresponding to s. Then limx0 f (x) = L, by Theorem 4.3.2 there exists > 0 such that for all x in the domain of f , |f (x)| > |L|/2. Set N = 1/ . Let n be an integer strictly greater than N . Then 1/n < 1/N = , so |sn | = |f (1/n)| > |L|/2. Theorem 8.4.6 Suppose that lim sn = L and lim tn = K . Then (1) (Constant rule) For any complex number c, lim{c} = c. (2) (Linear rule) lim{1/n} = 0. (3) (Sum/dierence rule) lim(sn tn ) = L K . (4) (Scalar rule) For any complex number c, lim(csn ) = cL. (5) (Product rule) lim(sn tn ) = LK . (6) (Quotient rule) If tn = 0 for all n and t = 0, then lim(sn /tn ) = L/K . Proof. Let f, g : {1/n : n N+ } C be given by f (1/n) = sn , g (1/n) = tn . By Theorem 8.4.1, limx0 f (x) = L and limx0 g (x) = K . Theorem 4.3.3 proves parts (3), (4), (5), (6) for f, g , hence via Theorem 8.4.1 also for s, t. For part (1), declare f (1/n) = c and for part (2), declare f (1/n) = 1/n. Again Theorems 4.3.3 and 8.4.1 easily nish the proofs of (1), (2). The following theorem for sequences follows immediately from the corresponding power, polynomial, and rational rules for functions:
3n+2 n2 3 .
Note that sn =
1/n2 3/n+2/n2 3n+2 n2 3 1/n2 = 13/n2 . 2 3x+2x 13x2 , and by the scalar,
The
212
Section 8.4: Convergence theorems via functions
Theorem 8.4.7 (Power, polynomial, rational rules for sequences) Let f be a polynomial function. Then lim{f (1/n)} = f (0). In particular, for any positive integer m, limn {1/nm } = 0. If f is a rational function that is dened at 0, then lim{f (1/n)} = f (0). Theorem 8.4.8 (The composite rule for sequences) Suppose that lim sn = L. Let g be a function whose domain contains L and all terms sn . Suppose that g is continuous at L. Then lim g (sn ) = g (L). Proof. Let f (1/n) = sn . By Theorem 8.4.1, limx0 f (x) = L, and by assumption limxL g (x) = g (L). Thus by the composite function theorem (Theorem 4.3.9), limx0 g (f (x)) = g (L), so that by Theorem 8.4.1, lim g (sn ) = g (L). In particular, since the absolute value function, the real part, and the imaginary part functions are continuous everywhere, we immediately conclude the following: Theorem 8.4.9 Suppose that lim sn = L. Then (1) lim |sn | = |L|. (2) lim Re sn = Re L. (3) lim Im sn = Im L. Furthermore, since the real and imaginary parts determine a complex number, we have: Theorem 8.4.10 A sequence {sn } of complex numbers converges if and only if the sequences {Re sn } and {Im sn } of real numbers converge. Proof. Let f (1/n) = sn . By Theorem 4.3.8, limx0 f (x) = L if and only if limx0 Re f (x) = Re L and limx0 Im f (x) = Im L, which is simply a restatement of the theorem. Theorem 8.4.11 (Comparison of sequences) Let s and t be convergent sequences of complex numbers. Suppose that |sn | |tn | for all except nitely many n. Then | lim sn | | lim tn |. If in addition for all except nitely many n, sn , tn are real numbers with sn tn , then lim sn lim tn . Proof. Let A be the set of those 1/n for which |sn | |tn | in the rst case and for which sn tn in the second case. Let f, g : A R be the functions f (1/n) = |sn |, g (1/n) = |tn | in the rst case, and f (1/n) = sn , g (1/n) = tn in the second case. By assumption for all x in the domain, f (x) g (x). Since 0 is a limit point of the domain (despite omitting nitely many 1/n) and since by Theorem 8.4.1, limx0 f (x) and limx0 g (x)
213
both exist, by Theorem 4.3.10, limx0 f (x) < limx0 g (x). In the rst case, this translates to | lim sn | | lim tn |, and in the second case it translates to lim sn lim tn . Theorem 8.4.12 (The squeeze theorem for sequences) Suppose that s, t, u are sequences of real numbers and that for all n N+ , sn tn un . If lim s and lim u both exist and are equal, then lim t exists as well and lim s = lim t = lim u. Proof. Let f, g, h : {1/n : n N+ } C be functions dened by f (1/n) = sn , g (1/n) = tn , h(1/n) = un . The assumption is that for all x in the domain of of f, g, h, f (x) g (x) h(x), and by Theorem 8.4.1 that limx0 f (x) = limx0 h(x). Then by the squeeze theorem for functions (Theorem 4.3.11), limx0 f (x) = limx0 g (x) = limx0 h(x). Hence by Theorem 8.4.1, lim s = lim t = lim u. Exercises for Section 8.4 8.4.1 Compute the following limits: . i) lim n21 2
n iii) lim 5n2 2 n1 . n2 iv) lim n 4+4 n2 +2 . 3n +2 v) lim 2n3 +n . 2n +2 vi) lim 3n 2 +n1 +2 .
4 3 2
3n +2000 ii) lim 2 n3 2000 .

2
8.4.3 Prove that lim
8.4.2 Suppose that lim |sn | = 0. Prove that lim sn = 0.

sin n n
= 0.
1 + k} = k. 8.4.5 Prove that for any real number k , lim{ n
Exercise 8.4.4 (Invoked in Theorem 8.7.1.) Let {sn } be a convergent sequence of positive real numbers. Prove that lim sn 0. 8.4.6 Prove that for all integers m, limn 8.4.7 Prove that lim 3 n + 1 3 n = 0.
{(1)n n} n+1 m n
= 1.
8.4.8 Does the sequence
have a limit? Prove your assertion.

2
+n 8.4.9 Compare the following two proofs of the fact that lim n3n = 0? Is one proof better 3 than the other? Does one have a missing step? 2 /n4 1/n2 +1/n3 +n 1 , so that the Proof #1: The nth term of the sequence is n3n 3 1/n4 = 3/n
corresponding function is f (x) =
x2 + x3 , 3x
and the limit of this is 0.

n2 +n 3n3
Proof #2: The nth term of the sequence is corresponding function is f (x) =
x+ x 3
2
, and the limit of this is 0.
1/n3 1/n3
1/n+1/n2 , 3
so that the
214
Section 8.5: Convergence theorems via epsilon-N proofs
8.5
Convergence theorems via epsilon-N proofs
All of the theorems below were proved in the previous section (Section 8.4) with a dierent method; here we use the epsilon-N formulation for proofs without explicitly resorting to functions whose domains have a limit point. The reader may omit one of these two sections. (Exercises are almost identical in the two sections.) Theorem 8.5.1 If a sequence converges, then its limit is unique. Proof. Let {sn } be a convergent sequence. Lets suppose that {sn } converges to both L and L . Then for any > 0, there exists an N such that |sn L| < /2 for all n > N . Likewise, for any > 0, there exists an N such that |sn L | < /2 for all n > N . Then by the triangle inequality, |L L | = |L sn + sn L | |L sn | + |sn L | < /2 + /2 = . Since is arbitrary, by Theorem 2.7.13 it must be the case that |L L | = 0, i.e., that L = L . Theorem 8.5.2 Suppose that lim{sn } = L and that L = 0. Then there exists a positive number N such that for all integers n > N , |sn | > |L|/2. In particular, there exists a positive number N such that for all integers n > N , sn = 0. Proof. Note that p = |L|/2 is a positive real number. Since lim{sn } = L, it follows that there exists a real number N such that for all integers n > N , |sn L| < |L|/2. Then by the reverse triangle inequality (proved in Exercise 3.10.6), | s n | = | s n L + L | = | (s n L ) + L | | L | | s n L | > | L | | L | /2 = | L | /2. Theorem 8.5.3 Suppose that lim sn = L and lim tn = K . Then (1) (Constant rule) For any complex number c, lim{c} = c. (2) (Linear rule) lim{1/n} = 0. (3) (Sum/dierence rule) lim(sn tn ) = L K . (4) (Scalar rule) For any complex number c, lim(csn ) = cL. (5) (Product rule) lim(sn tn ) = LK . (6) (Quotient rule) If tn = 0 for all n and t = 0, then lim(sn /tn ) = L/K . Proof. Part (1) was proved immediately after Denition 8.2.1. Part (2) is left for Exercise 8.5.1. Part (3): Let > 0. Since lim sn = L, there exists a positive real number N1 such that for all integers n > N1 , |sn L| < /2. Since lim tn = K , there exists a positive real number N2 such that for all integers n > N2 , |tn K | < /2. Let N = max {N1 , N2 }.
Chapter 8: Sequences Then for all integers n > N , | ( s n tn ) ( L K ) | = | ( s n L ) ( tn K ) |
215
|sn L| + |tn K | (by triangle inequality) < /2 + /2 (since n > N N1 , N2 ) = .
This proves (3). Part (4): Let > 0. Note that /(|c| + 1) is a positive number. Since lim sn = L, there exists N such that for all integers n > N , |sn L| < /(|c| + 1). Then for the same n, | |csn cL| = |c| |sn L| |c|/(|c| + 1) = |c||c +1 < . Since is any positive number, we conclude that csn converges to cL. Part (5): Let > 0. Since lim{sn } = L, there exists N1 such that for all integers n > N1 , |sn L| < 1. Thus for all such n, |sn | = |sn L + L| |sn L| + |L| < 1+ |L|. There also exists N2 such that for all integers n > N2 , |sn L| < /(2|K | + 1). By assumption lim{tn } = K there exists N3 such that for all integers n > N3 , |tn K | < /(2|L| + 2). Let N = max {N1 , N2 , N3 }. Then for all integers n > N , |sn tn LK | = |sn tn sn K + sn K LK | (by adding a clever 0) |sn tn sn K | + |sn K LK | (by triangle inequality) = | s n ( tn K ) | + | ( s n L ) K | = | s n | | tn K | + | s n L | | K | + |K | < (|L| + 1) 2| L | + 2 2| K | + 1 < + 2 2 = , = |(sn tn sn K ) + (sn K LK )|
which proves (5). Part (6): Let > 0. Since lim{sn } = L, there exists N1 such that for all integers n > N1 , |sn L| < 1. Thus for all such n, |sn | = |sn L + L| |sn L| + |L| < 1 + |L|. | . By assumption There also exists N2 such that for all integers n > N2 , |sn L| < |K 2
|K | lim{tn } = L there exists N3 such that for all integers n > N3 , |tn K | < 4(1+ |L|) . Since K = 0, by Theorem 8.5.2 there exists a positive number N4 such that for all integers n > N4 , |tn | > |K |/2. Let N = max {N1 , N2 , N3 , N4 }. Then for all integers n > N ,
2
sn L K s n L tn = tn K K tn
216 =
Section 8.5: Convergence theorems via epsilon-N proofs K s n tn s n + tn s n L tn (by adding a clever 0) K tn K s n tn s n tn s n L tn + (by triangle inequality) K tn K tn 1 1 1 + |L sn | = |K tn ||sn| | K | | tn | K 2 |K | 1 2 |K | 1 < (1 + |L|) + 4(1 + |L|) |K | |K | 2 |K | = + 2 2 = ,
which proves (6). Example 8.5.4 (Compare with Example 8.4.2.) Suppose sn =
5n2 1/n 3n+4 1/n 52/n 3+4/n . 5n2 3n+4 .
To prove that
= By the linear rule, lim 1/n = 0, so by lim sn = 5/3, we note that sn = the scalar rule, lim 2/n = lim 4/n = 0. Thus by the constant, sum, and dierence rules, lim(5 2/n) = 5 and lim(3 + 4/n) = 3, so that by the quotient rule, lim sn = 5/3. Example 8.5.5 (Compare with Example 8.4.3.) Let sn =
1/n 1/n2
2
= . By the linear rule, lim 1/n = 0, so that by the scalar rule, lim 3/n = 0, and by the product and scalar rules, lim 2/n2 = lim 3/n2 = 0. Thus by the sum and dierence rules, lim(3/n + 2/n2 ) = 0 and lim(1 3/n2 ) = 1. Finally, by the quotient rule, lim sn = 0/1 = 0. Theorem 8.5.6 (Power, polynomial, rational rules for sequences) For any positive integer m, limn {1/nm } = 0. If f is a polynomial function, then lim{f (1/n)} = f (0). If f is a rational function that is dened at 0, then lim{f (1/n)} = f (0). Proof. By the linear rule, limn {1/n} = 0. If limn {1/nm1 } = 0, then by the product rule, limn {1/nm } = limn {1/nm1 1/n} limn {1/nm1 } limn {1/n} = 0. Thus by mathematical induction, for all positive integers m, limn {1/nm } = 0. Now write f (x) = a0 + a1 x + + ak xk for some non-negative integer k and some complex numbers a0 , a1 , . . . , ak . By the just-proved power rule, limn 1/nm = 0 for all positive integers m, so that by the scalar rule, for all m = 1, . . . , k , limn am (1/n)m = 0. Thus by the constant rule and the repeated sum rule,
n
3/n+2/n 13/n2
3n+2 n2 3 .
Note that sn =
3n+2 n2 3
lim f (1/n) = lim a0 + a1 (1/n) + a2 (1/n)2 + + ak (1/n)k = a0 = f (0).

n
This proves the polynomial rule. Finally, let f be a rational function. Write f (x) = g (x)/h(x), where g, h are polynomial functions with h(0) = 0. By the just-proved polynomial rule, limn g (1/n) = g (0),
217
limn h(1/n) = h(0) = 0, so that by the quotient rule, limn f (1/n) = g (0)/h(0) = f (0). Theorem 8.5.7 (The composite rule for sequences) Suppose that lim sn = L. Let g be a function whose domain contains L and all terms sn . Suppose that g is continuous at L. Then lim g (sn ) = g (L). Proof. Let > 0. Since g is continuous at L, there exists a positive number > 0 such that for all x in the domain of g , if |x L| < then |g (x) g (L)| < . Since lim sn = L, there exists a positive number N such that for all integers n > N , |sn L| < . Hence for the same n, |g (sn) g (L)| < . In particular, since the absolute value function, the real part, and the imaginary part functions are continuous everywhere, we immediately conclude the following: Theorem 8.5.8 Suppose that lim sn = L. Then (1) lim |sn | = |L|. (2) lim Re sn = Re L. (3) lim Im sn = Im L. Furthermore, since the real and imaginary parts determine a complex number, we moreover get: Theorem 8.5.9 A sequence {sn } of complex numbers converges if and only if the sequences {Re sn } and {Im sn } of real numbers converge. Proof. By Theorem 8.5.8 it suces to prove that if lim{Re sn } = a and lim{Im sn } = b, then lim{sn } = a + bi. Let > 0. By assumptions there exist positive real numbers N1 , N2 such that for all integers n > N1 | Re sn a| < /2 and such that for all integers n > N2 | Im sn b| < /2. Set N = max {N1 , N2 }. Then for all integers n > N , |sn (a + bi )| = | Re sn + i Im sn a bi| | Re sn a||i|| Im sn b| < /2 + /2 = . Theorem 8.5.10 (Comparison of sequences) Let s and t be convergent sequences of complex numbers. Suppose that |sn | |tn | for all except nitely many n. Then | lim sn | | lim tn |. If in addition for all except nitely many n, sn , tn are real numbers with sn tn , then lim sn lim tn . Proof. Let L = lim sn , K = lim tn . By Theorem 8.5.8, lim |sn | = |L|, lim |tn | = |K |. Suppose that |L| > |K |. Set = (|L| |K |)/2|. By the denition of convergence, there exist N1 , N2 > 0 such that if n is an integer, n > N1 implies that < |sn | |L| < . and
218
Section 8.5: Convergence theorems via epsilon-N proofs
n > N2 implies that < |tn | |K | < . Let N3 be a positive number such that for all integers n > N3 , |sn | |tn |. If we let N = max{N1 , N2 , N3 }, then for integers n > N we have that |tn | < |K | + = (|L| + |K |)/2 = |L| < |sn |. This contradicts the assumption |sn | |tn |, so that necessarily |L| |K |. The proof of the second part is very similar and is left to the exercises. Theorem 8.5.11 (The squeeze theorem for sequences) Suppose that s, t, u are sequences of real numbers and that for all n N+ , sn tn un . If lim s and lim u both exist and are equal, then lim t exists as well and lim s = lim t = lim u. Proof. Set L = lim s = lim u. Let > 0. Since lim s = L, there exists a positive N1 such that for all integers n > N1 , |sn L| < . Since lim u = L, there exists a positive N2 such that for all integers n > N2 , |un L| < . Set N max {N1 , N2 }. Let n be an integer strictly greater than N . Then < sn L tn L un L < , so that |tn L| < . Since was arbitrary, this proves that lim t = L. Exercises for Section 8.5 8.5.2 Compute the following limits: i) lim n21 2 .
n iii) lim 5n2 2 n1 . n2 iv) lim n 4+4 n2 +2 . 3n +2 v) lim . 2n3 +n 2n +2 vi) lim 3n 2 +n1 +2 .
4 3 2
8.5.1 Prove that limn {1/n}n = 0.
3n +2000 ii) lim 2 n3 2000 .

2
8.5.4 Prove that lim
8.5.3 Suppose that lim |sn | = 0. Prove that lim sn = 0.

sin n n
= 0.
1 + k} = k. 8.5.6 Prove that for any real number k , lim{ n 2 8.5.8 Prove that limn (3 n ) = 3. 3 n + 1 3 n = 0. 8.5.9 Prove that lim
Exercise 8.5.5 (Invoked in Theorem 8.7.1.) Let {sn } be a convergent sequence of positive real numbers. Prove that lim sn 0. 8.5.7 Prove that for all integers m, limn
n+1 m n
= 1.
8.5.10 Does the sequence
(1)n n
have a limit? Prove your assertion.
219
8.6
Cauchy sequences, completeness of R, C
Denition 8.6.1 A sequence {sn } is Cauchy if for all > 0 there exists a positive real number N such that for all integers m, n > N , |sn sm | < . A sequence {sn } is bounded if there exists a positive real number M such that for all integers n, |sn | < M . Theorem 8.6.2 Every Cauchy sequence is bounded. Proof. Let {sn } be a Cauchy sequence. Thus for = 1 there exists a positive integer N such that for all integers m, n > N , |sn sm | < 1. Then the set {|s1 |, |s2|, . . . , |sN |, |sN +1| + 1} is a nite and hence a bounded subset of R. Let M an upper bound of this set, and let M = M + 1. It follows that for all n = 1, . . . , N , |sn | < M , and for n > N , |sn | = |sn sN +1 + sN +1 | |sn sN +1 | + |sN +1 | < 1 + M = M . Thus {sn } is bounded. Theorem 8.6.3 Every convergent sequence is Cauchy. Proof. Let {sn } be a convergent sequence. Let L be the limit. Let > 0. Since lim sn = L, there exists a positive real number N such that for all n > N , |sn L| < /2. Thus for all integers m, n > N , |sn sm | = |sn L + L sm | |sn L| + |L sm | < /2 + /2 = . Remark 8.6.4 It follows that every convergent sequence is bounded. But not every bounded sequence is convergent or Cauchy. For example, the bounded sequence {(1)n} is not Cauchy as it diverges (see Example 8.3.2). The converse of Theorem 8.6.3 is not true if the eld in which we are working is Q. Namely, let sn be the decimal approximation of 2 to n digits after the decimal point. Then {sn } is a Cauchy sequence of rational numbers: for every > 0, let N be a positive integer such that 1/10N < . Then for all integers n, m > N , sn and sm dier at most in digits N + 1, N + 2, . . . beyond the decimal point, so that |sn sm | 1/10N < . But {sn } does not have a limit in Q, so that {sn } is a Cauchy but not convergent sequence. But over R and C, all Cauchy sequences are convergent, as we prove next. Theorem 8.6.5 (Completeness of R, C) Every Cauchy sequence in R or C is convergent. Proof. Let {sn } be a Cauchy sequence in R or C. By Theorem 8.4.10 or Theorem 8.5.9, {sn } is convergent if and only if {Re sn } and {Im sn } are convergent. We leave it to Exercise 8.6.1 to prove that {sn } is Cauchy if and only if {Re sn } and {Im sn } are Cauchy. If we can prove that each real-valued Cauchy sequence converges in R, then we will have that {Re sn } and {Im sn } are convergent, whence that {sn } is convergent.
220
Section 8.6: Cauchy sequences, completeness of R, C
Thus we may assume without loss of generality that all sn are real. By Theorem 8.6.2, {sn } is bounded. It follows that all subsets {s1 , s2 , s3 , . . .} are bounded too. By the Least upper bound theorem (Theorem 3.8.3), in particular, un = sup {sn , sn+1 , sn+2 , . . .} is a real number. Any lower bound on {s1 , s2 , s3 , . . .} is also a lower bound on {u1 , u2 , u3 , . . .}. Thus there exists the greatest lower bound u of {u1 , u2 , u3 , . . .} in R (another application of Theorem 3.8.3). We claim that u = lim{sn }. Let > 0. Since {sn } is Cauchy, there exists N1 > 0 such that for all integers m n > N1 , |sn sm | < /2. Thus for all m n > N1 , we have that sm < sn + /2. But un is the least upper bound on these sm , so that sn un < sn + /2, whence |sn un | < /2. Since u = inf {u1 , u2 , u3 , . . .}, there exists an integer N2 such that 0 uN2 u < /2. Set N = max {N1 , N2 }. Let n > N be an integer. Since {sN , sN +1 , sN +2 , . . .} {sn , sn+1 , sn+2 , . . .}, it follows that un uN . But u un , so that 0 un u uN u < /2. Hence |sn u| = |sn un + un u| |sn un | + |un u| < /2 + /2 = . Example 8.6.6 The sequence {1 + converge.
1 2
1 3
1 + + n } is not Cauchy and hence does not
Proof. Let sn be the nth term of the sequence. If m > n, then sm sn = 1 1 1 mn n 1 + + + + =1 . m m1 m2 n+1 m m
Thus if m = 1000n, then sm sn > 1 1/1000 = 0.999. In particular, if = 1/2, then |sm sn | < .
1 1 1 By the next theorem below we conclude that the sequence {1 + 2 +3 ++ n } above is not bounded. Thus the terms of this sequence get larger and larger beyond bound. But this sequence gets large notoriously slowly the 1000th of the sequence is smaller than 8, the 10000th is smaller than 10, the 100000th is barely larger than 12. (One could lose patience in trying to see how this sequence grows without bound.)
Denition 8.6.7 A sequence {sn } of real numbers is called non-decreasing (resp. nonincreasing, strictly increasing, strictly decreasing) if for all n, sn sn+1 (resp. sn sn+1 , sn < sn+1 , sn > sn+1 ). Any such sequence is called monotone. Theorem 8.6.8 (Monotone sequences) Let {sn } be a bounded sequence of real numbers that is eventually non-decreasing (resp. non-increasing), i.e., that for some xed integer N , {sn }nN is non-decreasing (resp. non-increasing). Then lim sn exists, and equals the least upper bound (resp. greatest lower bound) of the set {sN , sN +1 , sN +2 , . . .}.
221
Proof. Suppose that for all n N , sn sn+1 . By the Least upper/greatest lower bound theorem (Theorem 3.8.3), the least upper bound of the set {sN , sN +1 , sN +2 , . . .} exists. Call it L. Let > 0. Since L is the least upper bound, there exists a positive integer N N such that 0 L sN < . Hence for all integers n > N , sN sn , so that 0 L sn L sN < , which proves that for all n > N , |sn L| < . Thus lim sn = L. The proof of the case of sn sn+1 for all n N is similar. Exercises for Section 8.6 Exercise 8.6.1 (Invoked in Theorems 8.8.4 and 8.6.5.) Prove that {sn } is Cauchy if and only if {Re sn } and {Im sn } are Cauchy. 8.6.2 Suppose that {sn } and {tn } are Cauchy sequences. i) Prove that {sn tn } is a Cauchy sequence. ii) Prove that {sn tn } is a Cauchy sequence. iii) Under what conditions is {sn /tn } a Cauchy sequence? iv) Prove that for all c C, {csn } is a Cauchy sequence.
8.6.3 Give examples of non-Cauchy sequences {sn }, {tn } such that {sn + tn } is a Cauchy sequence. Why does this not contradict the previous exercise? Repeat for {sn tn }.
8.6.4 Let r be a real number with a known decimal expansion. Let sn be a rational number whose digits n + 1, n + 2, n + 3, et cetera, beyond the decimal point are all 0, and all other digits agree with the digits of r . Prove that {sn } is a Cauchy sequence. Prove that lim{sn } = r . (Repeat with binary expansions if you know what a binary expansion is.) 8.6.5 Let {sn } be a sequence of real numbers such that for all n 1, sn sn+1 (resp. sn sn+1 ). Prove that {sn } is convergent if and only if it is bounded. *8.6.6 Use previous exercise to determine limits of the following sequences: n i) n+1 . n ii) iii) iv) v) 1+ 1+ 1+ 1+
1 2n . 2n 1 2n . n 1 n . 2n n +1 1 . n
8.6.7 Let s1 be a positive real number. For each n 1, let sn+1 = lim sn = 1. (Hint: prove that lim sn = lim s2 n .)
sn . Prove that
222
Section 8.7: Ratio test for convergence of sequences
8.7
Ratio test for convergence of sequences
Theorem 8.7.1 (Ratio test for sequences (version I)) Let r be a complex number with |r | < 1. Then limn r n = 0. Proof. If r = 0, the sequence {r n } is the constant zero sequence, so that lim r n = 0. So we may suppose that r = 0. For all positive integers n, 0 |r |n = |r |n1 |r | < |r |n1 1 = |r |n1 . Thus by Theorem 8.6.8, lim |r |n exists and equals the greatest lower bound L of the set {|r |, |r |2, |r |3, . . .}. By either Exercise 8.4.4 or Exercise 8.5.5, L 0. Suppose that L > 0. Then L(1|r |)/(2|r |) is a positive number. Since L is the inmum of the set {|r |, |r |2, |r |3 , . . .}, there exists a positive integer N such that 0 |r |N L < L(1 |r |)/(2|r |). Hence |r |N +1 = |r |N |r | < L+ L(1 |r |) 2| r | |r | = L(1 + |r |) L(1 + 1) < = L, 2 2
which contradicts the assumption that L is the greatest lower bound of the set. So necessarily L = 0, which proves that lim |r |n = 0. Then lim r n = 0 either by Exercise 8.4.2 or by Exercise 8.5.3. Theorem 8.7.2 (Ratio test for sequences) Let {sn } be a sequence of non-zero complex +1 numbers such that lim sn sn = L. Suppose that |L| < 1. Then lim sn = 0. Proof. Let r be an element of the interval (|L|, 1). Then r |L| is a positive number. +1 Since lim sn sn = L, there exists a positive number N such that for all integers n > N , +1 | sn sn L| < r |L|. This means that sn+1 sn+1 sn+1 L+L L + |L| r |L| + |L| = r. = sn sn sn Let n0 be the smallest integer strictly greater than N . Claim: For all n > n0 , |sn | r nn0 |sn0 |. If n = n0 + 1, then |sn | = |sn0 +1 | = |sn0 +1 /sn0 | |sn0 | r |sn0 | = r nn0 |sn0 |, which proves the base case. Now suppose that we know the inequality for some n 1. Then |sn | = |sn1+1 | = |sn1+1 /sn1 | |sn1 | r |sn1 | rr n1n0 |sn0 | = r nn0 |sn0 |, which proves claim by induction. By Theorem 8.7.1, lim r n = 0, hence by the scalar rule, lim r nn0 |sn0 | = lim r n r n0 |sn0 | = 0.
223
Since for all except nitely many positive integers n, 0 |sn | r nn0 |sn0 |, by Comparison theorem (either Theorem 8.4.11 or Theorem 8.5.10), lim |sn | = 0. Then by either Exercise 8.4.2 or Exercise 8.5.3 lim sn = 0. Exercises for Section 8.7 8.7.1 For each of the following sequences {sn }, you may wish to compute the the limit of {|sn |} via some comparison, decreasing property, et cetera, to prove the indicated. i) limn 21 n = 0. ii) limn iii)
(1)n 3n (4)n limn 7n k n!
n
= 0. = 0.
iv) limn v) limn vi) limn
n kn n kn
m
= 0 for all k C. = 0 for all non-zero k C with |k | > 1. = 0 for all non-zero k C with |k | > 1 and all integers m.
8.7.2 Give an example of a sequence {sn } of non-zero complex numbers such that lim(sn+1 /sn ) exists and has absolute value strictly smaller than 1. 8.7.3 Give an example of a sequence {sn } of non-zero complex numbers such that lim(sn+1 /sn ) = i/2. 8.7.4 Let r C satisfy |r | > 1. Prove that lim r n diverges.
8.8
Subsequences
Denition 8.8.1 A subsequence of an innite sequence {sn } is an innite sequence {sk1 , sk2 , sk3 , . . .} where 1 k1 < k2 < k3 < are integers. Notations for such a subsequence are: {skn }, {skn }n , {skn } n1 , {skn }n1 , {skn }nN+ .
Examples 8.8.2 Every sequence is a subsequence of itself. Sequences {1/2n}, {1/3n}, {1/(2n + 1)} are subsequences of {1/n}, and {1}, {1} are subsequences of {(1)n}. The constant sequence {1} is not a subsequence of {1/n}, because the latter sequence does not have innitely many terms equal to 1. If {sn } is the sequential enumeration of Q+ on n n } are subsequences, but { n+2 } = {1/3, 1/2, . . .} is not. page 198, then {1/n}, {n}, { n+1 Theorem 8.8.3 A subsequence of a convergent sequence is convergent, with the same limit. A subsequence of a Cauchy sequence is Cauchy.
224
Section 8.8: Subsequences
Proof. Let {sn } be a convergent sequence, with limit L, and let {skn } be a subsequence. Let > 0. By assumption there exists a positive number N such that for all integers n > N , |sn L| < . Since n kn , it follows that |skn L| < . Thus {skn } converges. The proof of the second part is similar. Theorem 8.8.4 Every bounded sequence has a Cauchy subsequence. Proof. First suppose that {sn } is a bounded sequence of real numbers. Let M be a positive real number such that for all n, |sn | M . Let a0 = M and b0 = M . The sequence {sn } has innitely many (all) terms on the interval [a0 , b0 ]. Set l0 = 0. We prove that for all m N+ there exists a subsequence {skn } all of whose terms are in the interval [am , bm ], where bm am = 2m (b0 a0 ) = 2m 2M and [am , bm ] [am1 , bm1 ]. Furthermore, we prove that there exists lm > lm1 such that slm [am , bm ]. Namely, given a subsequence {skn } on [am1 , bm1 ], we dene a +b a +b am1 , m1 2 m1 , if innitely many snk lie in am1 , m1 2 m1 ; [am , bm ] = am1 +bm1 , bm1 , otherwise. 2
This proves that every bounded sequence {sn } of real numbers has a Cauchy subsequence.
By construction {sln }n is a subsequence of {sn }n . We next prove that {sln }n is a Cauchy sequence. Let > 0. By Exercise 3.8.3 there exists a positive integer N such that 1/2N < /(2M ). Let m, n > N be integers. Then slm , sln are in [aN , bN ], so that |slm slm | bN aN = 2N 2M < .
In other words, [am , bm ] is the rst half of the interval [am1 , bm1 ] if that interval contains innitely many terms of the subsequence, and otherwise [am , bm ] is the second half of the interval [am1 , bm1 ]. By the choice of [am , bm ], the subsequence {skn } has an innite subsequence that lies in [am , bm ], and that subsequence is also a subsequence of the original {sn }. The facts bm am = 2m 2M and [am , bm ] [am1 , bm1 ] follow easily from the construction. Furthermore, since [am , bm ] contains innitely many terms of {sn }, there exists lm > lm1 such that slm [am , bm ].
Now suppose that {sn } is a bounded sequence of complex numbers. Then {Re sn } and {Im sn } are bounded sequences of real numbers. Thus by the rst part there exists a subsequence {Re skn } that is Cauchy. But {Im skn } is still bounded, and by the rst part this subsequence has a Cauchy subsequence {Im sln }. Since {Re sln } is a subsequence of the Cauchy sequence {Re skn }, it follows that {Re sln } is Cauchy. But then by Exercise 8.6.1, {sln } is Cauchy. The following is now an immediate consequence of Theorem 8.6.5:
Chapter 8: Sequences Theorem 8.8.5 Every bounded sequence in C has a convergent subsequence.
225
Example 8.8.6 We work out the construction of a subsequence as in the proof on the bounded sequence {(1)n 1}. For example, all terms lie on the interval [4, 4]. On this subinterval the second term equals 0. Innitely many terms lie on [4, 0], in particular the third term 2. Innitely many terms lie on [4, 2], in particular the fth term 2. After this all terms are 2, so that we have built the Cauchy subsequence {0, 2, 2, 2, . . .}. We could have built the Cauchy subsequence {0, 0, 2, 2, . . .}. If we started with the interval [8, 8], we could have built the Cauchy subsequence {0, 0, 0, 2, 2, . . .} or {2, 0, 0, 2, 2, . . .}, and so on. Denition 8.8.7 A subsequential limit of a bounded sequence {sn } is a limit of any convergent subsequence of {sn }. Example 8.8.8 The set of subsequential limits of any convergent sequence consists of the limit only. The set of subsequential limits of the sequence {(1)n 1}n is the set {0, 2}. The set of subsequential limits of the sequence {(1)n + 1/n}n = {0, 3/2, 2/3, 5/4, 4/5, 7/6, 6/7, . . .} is the set {1, 1}. The set of subsequential limits of the sequence {in }n is the set {i, 1, i, 1}. The sequence {n} has no real subsequential limits. Theorem 8.8.9 Every unbounded sequence of real numbers has a subsequence that has limit or . Proof. If {sn } is not bounded, choose k1 N+ such that |sk1 | 1, and once kn1 has been chosen, choose an integer kn > kn1 such that |skn | n. Now {skn }n is a subsequence of {sn }. Either innitely many among the skn are positive or else innitely many among the skn are negative. Choose a subsequence {sln }n of {skn }n such that all terms in {sln } all have the same sign. If they are all positive, then since sln n for all n, it follows that limn sln = , and if they are all negative, then since sln n for all n, it follows that limn sln = . Exercises for Section 8.8 8.8.1 Find the subsequential limit sets of the following sequences: i) {(1/2 + i 3/2)n }, {(1/2 i 3/2)n }. ii) {1/2n }. iii) {(1/2 + i 3/2)n + 1/2n }. iv) {(1/2 + i 3/2)n 1/2n}. v) {(1/2 + i 3/2)n + (1/2 i 3/2)n }. vi) {(1/2 + i 3/2)n (1/2 i 3/2)n}.
226
Section 8.9: Liminf, limsup for real-valued sequences
8.8.2 Consider the sequence enumerating Q+ as on page 198 equals R0 . Prove that the set of its subsequential limits equals R0 . 8.8.3 Prove that the real-valued sequence subsequences that each have limit or .
n2 +(1)n n(n+1) n
is unbounded. Find two
8.8.4 Give examples of sequences with the listed properties, if they exist. If they do not exist, justify. i) A convergent sequence that is not Cauchy. ii) A Cauchy sequence that is not convergent. iii) A bounded sequence that is not Cauchy. iv) An increasing sequence that is not Cauchy. v) A Cauchy sequence that is increasing. vi) A sequence with no convergent subsequences. 8.8.5 Suppose that a Cauchy sequence has a convergent subsequence. Prove that the original sequence is convergent as well. 8.8.6 Prove that a sequence of real numbers contains a monotone subsequence.
8.9
Liminf, limsup for real-valued sequences
Recall that by Theorem 3.8.3, every subset T of R bounded above has a least upper bound sup T = lub T in R, and that every subset T bounded below has a greatest lower bound inf T = inf T in R. We extend this denition by declaring sup T = if T is not bounded above,
inf T = if T is not bounded below.
We extend the same denitions also to sequences (thought of as sets). Thus sup N0 = , inf N0 = 0, sup {1/n}n = 1, inf {1/n}n = 0, sup {n}n = , inf {n}n = 1, sup {(1)n }n = , inf {(1)n }n = , et cetera. Denition 8.9.1 Let {sn } be a real-valued sequence. We dene limit superior lim sup and limit inferior lim inf of {sn } as follows: lim inf sn = sup {inf {sn : n m} : m 1}. lim sup sn = inf {sup {sn : n m} : m 1},
In other words, lim sup is the inmum of the set of all the suprema of all the tailend subsequences of {sn }, and analogously, lim inf is the supremum of the set of all the inma of all the tail-end subsequences of {sn }. In the plot below, the sequence oscillates
227
between positive and negative values but with peaks getting smaller and smaller. The red dots denote the sequence {sup {sm : m n}}n and the blue dots denote the sequence {inf {sm : m n}}n :
Clearly if {sn } is bounded above, then lim sup sn R, and if {sn } is bounded below, then lim inf sn R. Theorem 8.9.2 If {sn } converges to L, then lim inf sn = lim sup sn = L. Proof. Let > 0. Then there exists N > 0 such that for all integers n > N , |sn L| < . Thus for all m N , sup {sn : n m} L + , so that L lim sup sn L + . Since this is true for all > 0, it follows by Theorem 2.7.13 that L = lim sup sn . The other part is left to the reader. Theorem 8.9.3 Let {sn }, {tn} be bounded sequences in R. Then lim sup sn +lim sup tn lim sup(sn + tn ) and lim inf sn + lim inf tn lim inf(sn + tn ). Proof. Let a = lim sup sn , b = lim sup tn , c = lim sup(sn + tn ). By boundedness, a, b, c R. Let > 0. Recall that a = inf {sup {sn : n m} : m 1}. Then there exists a positive integer M such that sup {sn : n m} < a sup {sn : n m}. In particular for all n > m, sn < a. By possibly increasing m, we similarly get that for all n > m in addition tn < b. Thus for all n > m, sn + tn 2 < a + b. Thus c = inf {sup {sn + tn : n m} : m 1} a + b + 2. Since is arbitrary, it follows that c a + b, which proves the rst part. The rest is left as an exercise. Theorem 8.9.4 Let {sn } and {tn } be sequences of non-negative real numbers such that lim sn is a positive real number L. Then lim sup(sn tn ) = L lim sup tn and lim inf(sn tn ) = L lim inf tn .
228
Section 8.9: Liminf, limsup for real-valued sequences
Proof. Let > 0. Set = min {L/2, }. By assumption there exists N > 0 such that for all integers n > N , |sn L| < . Then L < sn < L + . Thus each sn is positive, and in fact |sn | > L/2. It follows that (L )tn sn tn (L + )tn . But then since L 0, (L ) lim sup tn = lim sup(L )tn lim sup sn tn lim sup(L + )tn = (L + ) lim sup tn . The proof of the liminf part is similar. Exercises for Section 8.9 8.9.1 Suppose that {sn } converges to L. Finish the proof of Theorem 8.9.2, namely prove that lim inf sn = L. 8.9.2 Find bounded real-valued sequences {sn }, {tn} such that lim sup sn + lim sup tn > lim sup(sn + tn ). (Compare with Theorem 8.9.3.) 8.9.3 Let {sn }, {tn } be bounded sequences in R. i) Finish the proof of Theorem 8.9.3, namely prove that lim inf sn + lim inf tn lim inf(sn + tn ). ii) Find such {sn }, {tn} so that lim inf sn + lim inf tn < lim inf(sn + tn ).
8.9.4 Suppose that lim sn = . Prove that the set of subsequential limits of {sn } is empty. 8.9.5 Compute lim inf and lim sup for the following sequences. n i) {n sin / 2 }. sin n ii) { n }. iii) {(1)n n!}. iv) {2n }. v) {1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, . . .}. vi) The sequence of all positive prime numbers. vii) The sequence of all inverses of positive prime numbers.
*8.9.6 Let {sn } be a bounded sequence of real numbers. Prove that the supremum of the set of all subsequential limits equals lim sup sn , and that the inmum of the set of all subsequential limits equals lim inf sn . 8.9.7 Prove that every real sequence {sn } has a subsequence {skn }n such that either for all n, skn+1 skn , or for all n, skn+1 skn .
Chapter 9: Innite series and power series
In this section we will handle (some) innite sums, and in particular functions that arise as innite sums of higher and higher powers of a variable x. The culmination of the chapter and the course are the exponential and trigonometric function, with their properties. Warning: Finite sums are possible by the eld axioms, but innite sums need not make any sense at all. For example, 1 + (1) + 1 + (1) + 1 + (1) + 1 + (1) + 1 + (1) + 1 + (1) + 1 + (1) may be taken to be 0 or 1 depending on which consecutive pairs are grouped together in a sum, or it could even be taken to be 3 (or 17 or 4, et cetera) by taking the rst three positive 1s, and then matching each successive 1 in the sum with the next not-yetused +1. In this way each 1 in the expression is used exactly once, so that the sum can indeed be taken to be 3. Innite sums require special handling, but limits prepared the ground for that.
9.1
Innite series
Denition 9.1.1 For an innite sequence {an } of complex numbers, dene the corresponding sequence of partial sums {a1 , a1 + a2 , a1 + a2 + a3 , a1 + a2 + a3 + a4 , . . .}. We denote the nth term of this sequence sn = k=1 ak . The (innite) series corre sponding to the sequence {an } is k=1 ak (whether this innite sum makes sense or not). Example 9.1.2 For the sequence {1}, the sequence of partial sums is {n}. If a = 1, by n n 1 Theorem 1.5.3 the sequence of partial sums of {an } is { k=1 ak }n = {a aa 1 }n . In particn
ular, the sequence of partial sums of {(1)n} is {{ (1) 2
}n = {1, 0, 1, 0, 1, 0, . . .}.
n Denition 9.1.3 The series k=1 ak converges to L C if the sequence { k=1 ak }n converges to L. We say then that L is the sum of the series and we write k=1 ak = L.
If the series does not converge, it diverges.
230
Section 9.1: Innite series
Since convergence of series depends on the convergence of sequences, a series may diverge to or to , or it simply has no limit. The following follows immediately from the corresponding results for sequences: Theorem 9.1.4 Let A = (1) If A, B C, then
k=1
ak , B =
k=1 bk ,
and c C.
(ak + cbk ) = A + cB. A, B R {, }, then if if if if if if if if if
k=1
It is hard to immediately present examples of this because we know so few limits of innite series. There are examples in the exercises. Theorem 9.1.5 If r C satises |r | < 1, then the geometric series 1 r k to 1 k=1 r = 1r . r , so
n
n
(2) If all ak , bk are real numbers and , , , , (ak + cbk ) = , k=1 , , , A,
A = , B R0 {} and c 0; A = , B R0 {} and c 0; A = , B R0 {} and c 0; A = , B R0 {} and c 0; A R, B = and c > 0; A R, B = and c < 0; A R, B = and c < 0; A R, B = and c > 0; c = 0.
n k=1
r k1 converges
r n Proof. By Theorem 1.5.3, k=1 r k1 = 11 r . By Theorem 8.7.1, lim r = 0. Thus by the scalar and sum rules for limits of sequences (Theorem 8.4.6 or Theorem 8.5.3), n n
lim
r k1 = lim
k=1
1 r n+1 = lim n n 1r
1 rn lim 1 r 1 r n
1 . 1r
In particular, the familiar decimal expansion 0.33333 of 1/3 can be thought of as 3 the innite sum k=1 10 k . The sequence of its partial sums is {0.3, 0.33, 0.333, 0.3333, . . .}, and by the theorem above,
k=1
3 3 = k 10 10
k=1
1 10
k 1
3 10 3 1 1 = . 1 = 10 1 10 10 9 3 diverges. (Confer also Example 8.6.6.)

2n 1 k=1 k
Example 9.1.6 The harmonic series
1 k=1 k
Proof. By Exercise 1.5.23, for any positive integer n, 2n 1 and Theorem 8.3.6, { k=1 k }n has limit .
n+2 2 .
By Example 8.3.5
Chapter 9: Innite series and power series Example 9.1.7 The series
1 k=1 k2
231
converges.
Proof. By Exercise 1.5.24 and Theorem 9.1.5, 0

2 1
n
2 n 1 k=1
1 k2
n
n+1 k=1
1 2 k 1
k=1
1 2 k 1
1 = 2, 1 1/2
1 1 so that { k=1 k 2 }n and { k=1 k2 }n are bounded increasing sequences of real numbers. 1 By Theorem 8.6.8, the sequence has a limit that is at most 2, so that k=1 k 2 converges.
It turns out that Theorem 9.1.8 If
1 k=1 k2 k=1
2 , 6
but this is much harder to prove.
ak converges, then lim an = 0.
Proof. Let L = k=1 ak . Let > 0. By assumption there exists a positive number N such n that for all integers n > N , |L k=1 ak | < /2. In particular, for integers n > N + 1, |an | = an + an +
n n1 k=1 n1 k=1
ak L
n1 k=1
ak + L
n1 k=1
ak L +
n1 k=1
ak + L
=
k=1
ak L +
ak + L
< /2 + /2 = . The converse of this theorem is of course false; see Example 9.1.6. Theorem 9.1.9 Let {an } be a sequence of complex numbers, and let m be a positive integer. Then k=1 an converges if and only if k=m an converges. Furthermore in this case, if c = a1 + a2 + + am1 , k=1 an = c + k=m an . Proof. Let sn = a1 + a2 + + an , and tn = am + am+1 + + an . By the constant and sum rules for sequences (Theorem 8.4.6 or Theorem 8.5.3), the sequence {sn } = {c} + {tn} converges if and only if the sequence {tn } converges. Exercises for Section 9.1 9.1.1 Let r C satisfy |r | < 1. Prove that 9.1.2 Let r C satisfy |r | 1. Prove 9.1.3 Compute the following sums:
r k k=1 r = 1r . that k=1 r k diverges. 1 1 1 k=1 2k , k=5 2k , k=6 3k ,
2 k=8 5k .
232
Section 9.2: Convergence and divergence theorems for series
1 9.1.4 Prove that k=1 k(k +1) converges, and nd the sum. (Hint: do some initial experimentation with partial sums, nd a pattern for partial sums, and prove the pattern with mathematical induction.) 2k+1 9.1.5 Prove that k=1 k2 (k+1)2 converges, and nd the sum. (Hint: do some initial experimentation with partial sums, nd a pattern for partial sums, and prove the pattern with mathematical induction.)
9.1.6 For each k N+ let xk be an integer between 0 and 9. Prove that converges. i) What does this say about decimals? ii) Find the limit L if xk = 4 for all k . iii) Find the limit L if {xn } = {1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, . . .}. iv) Prove that whenever the sequence {xn } is eventually periodic, then k=1 rational number. 9.1.7 Prove that k=1 ak converges if and only if k=1 Re ak and Furthermore, k=1 ak = k=1 Re ak + i k=1 Im ak . 9.1.8 Suppose that lim an = 0. Prove that
k=1 ak k=1
xk k=1 10k
xk 10k
is a
Im ak converge.
diverges.
9.1.9 Determine, with proof, which series converge. i) k=1 k1k . 1 ii) k=1 k3 + ik . 1 . iii) k=1 k ! 9.1.10 Let {an } and {bn } be complex sequences, and let m N such that for all n 1, an = bn+m . Prove that k=1 ak converges if and only if k=1 bk converges.
9.2
Convergence and divergence theorems for series
Theorem 9.2.1 (Cauchys criterion for series) The innite series ak converges if and only if for all real numbers > 0 there exists a real number N > 0 such that for all integers n m > N, |am+1 + am+2 + + an | < . Proof. Suppose that ak converges. This means that the sequence {sn } of partial sums converges, and by Theorem 8.6.3 this means that {sn } is a Cauchy sequence. Thus for all > 0 there exists N > 0 such that for all integers m, n > N , |sn sm | < . In particular for n m > N , |am+1 + am+2 + + an | = |sn sm | < . The converse is left for Exercise 9.2.2. Theorem 9.2.2 (Comparison test (for series)) Let {an } be a real and {bn } a complex sequence. If ak converges and if for all n, an |bn |, then bk converges.
233
Proof. Let sn = a1 + a2 + + an , tn = b1 + b2 + + bn . Then for all integers n m, |tn tm | = |bm+1 + bm+2 + + bn | am+1 + am+2 + + an = sn sm = |sn sm |. |bm+1 | + |bm+2 | + + |bn |
Since {sn } is convergent, it is Cauchy, and the inequality above shows that {tn } is Cauchy in C. By Theorem 8.6.5, {tn } has a limit L C, so that by denition of convergence of series, bk = L. Theorem 9.2.3 (Comparison test (for series)) Let {an } be a complex and {bn } a real sequence. If ak diverges and if for all n, bn |an |, then bk diverges. Proof. Let sn = a1 + a2 + + an , tn = b1 + b2 + + bn . Then for all integers n m, |tn tm | = bm+1 + bm+2 + + bn |am+1 | + |am+2 | + + |an | |am+1 + am+2 + + an | = |sn sm |.
Since {sn } is divergent, by Theorem 8.6.5 it is not Cauchy, so that by the inequality above {tn } is not Cauchy. Hence bk diverges. Theorem 9.2.4 (Ratio test) Let {an } be a sequence of non-zero complex numbers. (1) If lim sup
an+1 an
< 1, then
(2) If lim inf
an+1 an
|ak | and
ak converge.
> 1, then
|ak | and
ak diverge.
n+1 | interval (1, L). Since L = sup {inf { |a |an | : n m} : m 1} and r < L, it follows that
n+1 | follows that there exists m 1 such that r > sup { |a |an | : n m}. Thus for all n m, |an+1 | < r |an |. Thus by Exercise 1.5.25, |am+n | < r n |am |. The geometric series k k k r converges by Theorem 9.1.5, and by Theorem 9.1.4, k am r converges. Thus by Theorem 9.2.2, k=1 |am+k | and k=1 am+k converge. Hence by Theorem 9.1.9, |ak | and ak converge. This proves (1). n+1 |, and suppose that L > 1. Let r be a real number in the open Now let L = lim inf | aa n
n+1 | open interval (L, 1). Since L = inf {sup { |a : n m} : m 1} and r > L, it | an |
n+1 Proof. Let L = lim sup | aa |. Suppose that L < 1. Let r be a real number in the n
n+1 | there exists m 1 such that r < inf { |a |an | : n m}. Thus for all n m, |an+1 | > r |an |. Thus by a straightforward modication of Exercise 1.5.25, |am+n | > r n |am |. The geometric series k r k diverges by Exercise 9.1.2, and by Theorem 9.1.4, k am r k diverges. Thus
234

k=1
by Theorem 9.2.2, k=1 |am+k | and and ak diverge. This proves (2).
am+k diverge. Hence by Theorem 9.1.9,
|ak |
n+1 | = 1 or This ratio test for convergence of series does not apply when lim sup | aa n an+1 lim inf | an | = 1. The reason is that under these assumptions the series k |ak | and k ak sometimes converge and sometimes diverge. For example, if an = 1/n for all n, n+1 n+1 lim sup | aa | = lim inf | aa | = 1, and the two series diverge; whereas if an = 1/n2 for all n n n+1 n+1 | = lim inf | aa | = 1, and the two series converge. n, then lim sup | aa n n
Theorem 9.2.5 (Root test for series) Let {an } be a sequence of complex numbers. Let L = lim sup |an |1/n . (1) If L < 1, then k |ak |, k ak converge. (2) If L > 1, then k |ak |, k ak diverge. Proof. If L < 1, choose r (L, 1). Since L = inf {sup {|an |1/n : n m} : m 1} and r > L, there exists m 1 such that r > sup {|an |1/n : n m}. Thus for all n m, r n |an |. As in the proof of of Theorem 9.2.4, the conclusion in (1) follows. The proof of (2) is similar, and is omitted here. Theorem 9.2.6 (Alternating series test) If {an } is a sequence of positive real numbers such that lim an = 0 and a1 a2 a3 , then k=1 (1)k ak converges. Proof. Let m, n be positive integers. Then 0 (an an+1 ) + (an+2 an+3 ) + + (an+2m an+2m+1 ) = an an+1 + an+2 an+3 + + an+2m an+2m+1 = an (an+1 an+2 ) (an+3 ) an+4 ) (an+2m1 an+2m ) an+2m+1
an ,
and similarly 0 an an+1 + an+2 an+3 + + an+2m an . Thus by Cauchys criterion Theorem 9.2.1,
k (1) k
ak converges.
Example 9.2.7 Recall from Example 9.1.6 that the harmonic series k 1/k diverges. k But the alternating series k (1) 1/k converges by this theorem. (In fact, k (1)k 1/k converges to ln 2, but this is harder to prove.) Theorem 9.2.8 (The p-series convergence test) Let p be a real number. The series converges if p < 1 and diverges if p 1.
k
kp
235
Proof. If p = 1, then the series is the harmonic series and hence diverges. If p 1, then np n1 for all n by Theorem 7.4.5. Thus by the comparison test (Theorem 9.2.3), p k k diverges. Now suppose that p < 1. Since p is negative, exponentiation by p is a decreasing function by Theorem 7.4.4, so that for all positive integers n, np+1 (p + 1)np = (n p 1)np < (n 1)(n 1)p = (n 1)p+1 . Division by negative p + 1 yields np < Claim: : For all integers n 1, 1 np+1 (n 1)p+1 . p+1 1 p np+1 + . p+1 p+1
k=1
kp
1 +p p When n = 1, 1p = 1 = 1+ p+1 = p+1 , which proves the base case. Now suppose that the displayed equation holds for some n 1 1. Then n
p+1
kp =
k=1
n1 k=1
k p + np
p 1 1 np+1 (n 1)p+1 (n 1)p+1 + + p+1 p+1 p+1 (by induction assumption and the inequality before the claim) 1 p np+1 + . p+1 p+1
This proves the claim. n 1 In particular, since p+1 np+1 < 0, the sequence { k=1 k p }n of partial sums is an increasing sequence bounded above by p/(p + 1). Thus by Theorem 8.6.8, the sequence n { k=1 k p }n converges. Thus by the denition of series, k=1 k p converges. Theorem 9.2.9 (Integral test for series convergence) Let f : [0, ) [0, ) be a den creasing function. Suppose that for all n, 1 f exists. Then k=1 f (k ) converges if and n only if limn 1 f exists and is a real number. Proof. Since f is decreasing, for all x [n, n + 1], f (n) f (x) f (n + 1). Thus
n+1 n+1 n+1
f (n + 1) =
n
f (n + 1)dx
f (x)dx
f (n)dx = f (n).
n
By denition, k f (k ) converges if and only if limn (f (1) + f (2) + + f (n)) exists. n+1 n+1 But by the displayed formula, 1 f f (1) + f (2) + + f (n), so that { 1 f }n is n+1 a bounded increasing sequence of real numbers, so that limn 1 f exists, and hence n that limn 1 f exists. The proof of the other direction is similar.
9.2.1 For each of the following series, determine, with proof, whether they converge or diverge. You may need to use Examples 8.2.6 and 8.3.9. 3i . i) 2 + k4 k k=1 ii) iii) iv) v) vi) 1 . k k=1
k=1 k=1 k=1 k=1
1 . k3 2k . k! 2k . k3
. k 2
9.2.2 Prove the converse of the Cauchys criterion for series (Theorem 9.2.1). 9.2.3 Find a convergent series lim sup |bn |1/n = 1.
k
ak and a divergent series
k bk
with lim sup |an |1/n =
9.2.4 Make a list of all encountered criteria of convergence for series. 9.2.5 The goal of this exercise is to show that if the ratio test (Theorem 9.2.4) determines the convergence/divergence of a series, then the root test (Theorem 9.2.5) determines it as well. Let {an } be a sequence of non-zero complex numbers. n+1 i) Suppose that lim sup | aa | < 1. Prove that lim sup |an |1/n < 1. n n+1 ii) Suppose that lim inf | aa | > 1. Prove that lim sup |an |1/n > 1. n 9.2.6 Apply the ratio test (Theorem 9.2.4) and the root test (Theorem 9.2.5) to Was one test easier? Repeat for k=1 k1k . 9.2.7 Prove that
(1) k=1 k
k
5k k=1 k! .
= mation and dierence rules?
1 k=1 2k 1 k=1 2k
and
1 k=1 2k+1 1 k=1 2k+1 . Why
diverge. Refer to Example 9.2.7. Argue that does this not contradict the expected sum
9.2.8 Let {an } be a complex sequence, and let c C. Is it true that k=1 ak converges if and only if k=1 cak converges? If true, prove; if false, give a counterexample. 9.2.9 Prove that if
k=1
ak converges then
9.2.10 Let an = (1)n /n, bn = 2(1)n. Then k and k=1 a bk diverges.
ak k=1 k converges. k=1 ak converges,
and for all n, |bn | = 2,
237
9.2.11 (Compare with Exercise 9.2.10.) Suppose that k=1 |ak | converges. Let {bn } be a k sequence of complex numbers such that for all n, |bn | > 1. Prove that k=1 a bk converges.
9.3
Power series
So far all innite series were sums of terms of a sequence: k=1 ak stands for the limit of the sequence {a1 , a1 + a2 , a1 + a2 + a3 , . . .} of partial sums. Thus to obtain the sum, we add the rst term to the second term, the obtained sum to the third term, the new sum to the fourth term, that sum to the fth term, and so on ad innitum. But the meaning is clear when the innite sum shifts the index by a nite number, such as in the following:
k=5 k = 5
ak =
k=1
ak+4 ,
5
ak =
k = 5
ak +
k=1
ak .
In this section we will be dealing with sums where the index varies through N0 , and furthermore, the terms of the sequence will be special functions rather than constants:
k Denition 9.3.1 A power series is an innite series of the form k=0 ak x , where a0 , a1 , a2 , . . . are complex numbers and x is a variable that can be replaced by any complex number. By convention as on page 27, 00 = 1.
The most important question that we address in this section is: for which x does such an innite sum make sense. This is the same as asking for the x-domain of the function n k k k=0 ak x . Certainly when x = 0, the partial sums are k=0 ak 0 = a0 for all n 0, so that k=0 ak xk converges to a0 when x = 0. Thus 0 is always in the domain of a power series function.
k Example 9.3.2 Let f (x) = k=0 x . By Theorem 9.1.5, the domain of f contains all complex numbers with absolute value strictly smaller than 1, and by Exercise 9.1.2, the domain of f contains no other numbers, so that the domain equals {x C : |x| < 1}. For 1 all x in the domain of f , by Exercise 9.1.1, k=0 xk = 1 x . But note that the domain of 1 1x is strictly larger than the domain of f . 3 1 1 1 = 2 , k0 0.1 = 11 In particular, k0 21 k = 1 1 = 2, k 0 3 k = 1 1 0.6 = 2.5, et 6k 2 3 cetera.
Theorem 9.3.3 (Root test for convergence of power series) Let
ak xk be a power series,
238 and let = lim sup |an |1/n. Dene R by 1/, R = 0, ,
Section 9.3: Power series
if 0 < < ; if = ; if = 0. ak xk converge in C, and for all x C
Then for all x C with |x| < R, |ak ||x|k and with |x| > R, |ak ||x|k and ak xk diverge.
Proof. By denition of limits, is a non-negative real number of . We apply the Root test for series (Theorem 9.2.5): lim sup |an xn |1/n = |x| lim sup |an |1/n = |x|. If |x| < 1, then both of the series converge, and if |x| > 1, then the two series diverge. If = 0, then |x| < 1 is true for all x C, so R = has the stated property. If = 0, then |x| < 1 is true only for x = 0, so R = 0 has the stated property. If 0 < < , then |x| < 1 is true only for all x C with |x| < 1/ = R. Denition 9.3.4 The R from Theorem 9.3.3 is called the radius of convergence of the series ak xk . This is really a radius of convergence because inside the circle B (0, R) the series converges and outside of the circle the series diverges. Note that the series ak xk , |ak |xk , and |ak ||x|k have the same radius of convergence. Theorem 9.3.5 (Ratio test for convergence of power series) Suppose that all an are nonzero complex numbers. n , then |ak ||x|k and ak xk converge. (1) If |x| < lim inf aa n+1 an , then |ak ||x|k and ak xk diverge. (2) If |x| > lim sup an +1 n exists, it equals the radius of convergence of ak xk . Thus if lim aa n+1 Warning: Compare with the Ratio test for convergence of series (Theorem 9.2.4) where fractions are dierent. Proof. The two series converge in case x = 0, so that we may assume that x = 0. We may then apply the Ratio test for convergence of series (Theorem 9.2.4): lim sup an+1 xn+1 an+1 = |x| lim sup n an x an an+1 :nm = |x| inf {sup an = |x| inf = sup inf { 1 inf {
an an+1
: m 1} :m1
an an+1
| x|
: n m}
: n m} : m 1
Chapter 9: Innite series and power series = | x| . n lim inf aa n+1
239
If this is strictly smaller than 1, then the two series converge, which proves (1). Similarly, lim inf an+1 xn+1 | x| = , n an xn lim sup aa n+1
and if this is strictly larger than 1, then the two series diverge. The last part is then immediate by the denition of radius of convergence. Examples 9.3.6 (1) xk , kxk ,
k 2 xk ,
k (k 1)xk ,
x2k+1 all have radius of convergence 1.
Proof of the last part: The root test gives lim sup |an |1/n = lim sup{1, 0, 1, 0, 1, 0, 1, 0, . . .} = 1, so that the radius of convergence is 1/1 = 1. The ratio test for power series is inapplicable here. But note that x2k+1 = x (x2 )k , and by the ratio test for series (not power series), this series converges for non-zero x if lim sup |(x2 )k+1 /(x2 )k | < 1, i.e., if |x2 | < 1, i.e., if |x| < 1, and it diverges if |x| > 1. (2) The radius of convergence of = lim sup
1 1/n nn k=1 x n n
= lim sup
1 n
= 0.
is . For this we apply the root test:

xn k=1 n!
(3) By the ratio test, the radius of convergence of
is lim
1 n! 1 (n+1)!
= = lim (n+1)! n!
lim(n + 1) = . The root test gives = lim sup |1/n!|1/n| = lim sup 1/n!1/n, and by Example 8.3.9 this is 0. Thus the radius of convergence if also by the root test. Denition 9.3.7 Let f have derivatives of all orders. If a is in the domain, the Taylor (k ) series of f (centered) at a is the series k=0 f k!(a) (x a)k . Exercises for Section 9.3 9.3.1 Compute the radius of convergence for the following series: i) xn . ii) 3xn . iii) (3x)n . iv) nxn . v) 3nxn . vi) n(3x)n. xn vii) n3 . 3 xn viii) n3 . (3x)n ix) n3 .
240
Section 9.4: Dierentiation of power series
9.3.2 Compute the radius of convergence for the following series: i) ii) iii)
xn . nn n x . n2n xn (2n)n .
9.3.3 What would be a sensible denition for generalized power series k=0 ak (z a)k ? What would be a sensible denition of the radius of convergence of k=0 ak (z a)k ? 9.3.4 Compute the Taylor series of f (x) = convergence.
1 1 x
centered at 0. Determine the radius of
9.3.5 Compute the Taylor series of f (x) = ex centered at 0. Determine the radius of convergence. Use Theorem 6.4.5 to determine for each x in the domain of f the dierence between f (x) and the Taylor series at x. 9.3.6 Compute the Taylor series of f (x) = ln(x + 1) centered at 0. Determine the radius of convergence.
9.4
Dierentiation of power series
Power series are functions, and perhaps they are dierentiable in their domains. Recall ) f (x ) that for any dierentiable function f , f (x) = limh0 f (x+hh , and any power series n k k k k=0 ak x is actually the limit of a sequence: k=0 ak x = lim{ k=0 ak x }n . Thus (
ak xk ) = lim
k=0
h0
lim{
n k=0
ak (x + h)k }n lim{ h
n
n k=0
ak xk }n
.
n k k=0 ak x }n
Certainly by the sum rule for convergent series, lim{ k=0 ak (x+h)k }n lim{ n lim{ k=0 ak ((x + h)k xk )}n , and by the constant rule we get that (
n
ak x ) = lim lim
k=0
h0 n
ak
k=0
(x + h)k xk . h
If we could change the order of limits, then we would get by the polynomial rule for derivatives that (
k=0 n
ak xk ) = lim lim
n h0
kak xk1 .
k=0
In fact, it turns out that this is the correct derivative, but our reasoning was based on an unproven and generally false switch of the two limits. We give a correct proof of derivatives in the rest of the section, and we prove that the radii of convergence of a series and its derivative series are the same.
Chapter 9: Innite series and power series Theorem 9.4.1 The series same radius of convergence. Proof.
k k=0 ak x
241
k 1 k=0 kak x
and
k=1
kak xk1 have the
k 1 k Certainly the radii of convergence of , and = k=1 kak x k=1 kak x k 1/n = 1, so that by Theorem 8.9.4, k=0 kak x are the same. By Example 8.2.6, lim n 1/n 1/n 1/n 1/n lim sup |nan | = lim sup n |an | = lim sup |an | . Thus by the root test for series (Theorem 9.2.5), the radii of convergence of k=1 kak xk1 and k=0 ak xk are the same, so that the radii of convergence of all the listed series are the same.
The following theorem is not necessarily interesting in its own right, but it is a stepping stone in the proof of derivatives of power series. Typically in mathematics literature such stepping stones are called lemmas rather than theorems. Theorem 9.4.2 Let the function g (x) = and is continuous at c.
k k ak x have radius convergence R. Let c C satisfy |c| k 1 + cxk2 + c2 xk3 + + ck1 ) is dened k=1 ak (x
< R. Then on B (0, R)
Proof. There is nothing to prove if R = 0, so we may assume that R > 0.

k 1 First of all, converges on B (0, R) by Theorem 9.4.1. In particular, k=1 kak x k 1 k 1 = k=1 ak (c + cck2 + c2 ck3 + + ck1 ) is well-dened, but this is k=1 kak c simply g (c), so that c is in the domain of g .
Let > 0 and let d R satisfy |c| < d < R. Then again by Theorem 9.4.1, d is in the domain of k ak xk , k kak xk1 , and also k k (k 1)ak xk2 and k k (k 1)|ak ||x|k2. 1 k 2 , and = min {d |c|, D+1 }. Then is positive. Let c C Set D = 2 k k (k 1)|ak |d with 0 < |x c| < . Then by the triangle inequality, |x| = |x c + c| |x c| + |c| < + |c| d |c| + |c| = d < R, so that x is in the domain of the power series k ak xk . But is x in the domain of g ? Since the radius of convergence of k ak xk is R, by Theorem 9.4.1 also the radius of convergence of k kak xk1 is R, so that k kak dk1 converges. Then from the Comparison theorem (Theorem 9.2.2) and from
n
k=1
ak (xk1 + cxk2 + c2 xk3 + + ck1 )|

n
k=1 n k=1 n
|ak |(|x|k1 + |c||x|k2 + |c|2 |x|k3 + + |c|k1 ) (by triangle inequality) |ak |(dk1 + ddk2 + d2 dk3 + + dk1 ) k |ak |dk1
k=1
242
Section 9.4: Dierentiation of power series
we deduce that x is in the domain of g . Furthermore, | g ( x ) g ( c) | = | =| =| =| =| ak (xk1 + cxk2 + c2 xk3 + + ck1 ) kak ck1 |
ak (xk1 + cxk2 + c2 xk3 + + ck1 kck1 )| ak ((xk1 ck1 ) + (cxk2 ck1 ) + (c2 xk3 ck1 ) + + (ck1 ck1 )| ak ((xk1 ck1 ) + c(xk2 ck2 ) + c2 (xk3 ck3 ) + + ck2 (x c))| + c ( x k 3 + x k 4 c + x k 5 c 2 + + c k 3 ) ak (x c)((xk2 + xk3 c + xk4 c2 + + ck2 )
|ak ||x c|((|x|k2 + |x|k3 |c| + |x|k4 |c|2 + + |c|k2 ) + | c | ( | x | k 3 + | x | k 4 | c | + | x | k 5 | c | 2 + + | c | k 3 ) (by triangle inequality)
+ c2 (xk4 + xk5 c + xk6 c2 + + ck4 ) + + ck2 )| (by Exercise 1.5.26)
+ | c | 2 ( | x | k 4 + | x | k 5 | c | + | x | k 6 | c | 2 + + | c | k 4 ) + + | c | k 2 ) |ak |((dk2 + dk3 d + dk4 d2 + + dk2 )
| x c|
+ d ( d k 3 + d k 4 d + d k 5 d 2 + + d k 3 )
= | x c|
+ d 2 ( d k 4 + d k 5 d + d k 6 d 2 + + d k 4 ) + + d k 2 ) |ak | k (k 1) k2 d (by Example 1.5.1) 2
|ak |((k 1)dk2 + (k 2)dk2 + (k 3)dk2 + + dk2 )
= | x c| = | x c| D < D .
This proves that g is continuous at c. Theorem 9.4.3 Let f (x) = k=0 ak xk have radius of convergence R. Then f is dierenk 1 . The derivative function tiable on B (0, R), and for all c B (0, R), f (c) = k=1 kak c k 1 also has radius of convergence R. f (x) = k=1 kak x Proof. Let c, x B (0, R). Then f ( x ) f ( c) = xc = =
k
ak xk k ak ck xc k k k ak (x c ) xc k 1 + x k 2 c + x k 3 c 2 + + c k 1 ) k ak (x c)(x (by Exercise 1.5.26) xc
Chapter 9: Innite series and power series =

k
243
ak (xk1 + xk2 c + xk3 c2 + + ck1 ),
which is the function g from the previous theorem. In that theorem we proved that g is continuous at c, so that f ( x ) f ( c) = lim g (x) = g (c) = f (c) = lim xc xc xc
k=1
kak ck1 .
Then the theorem follows from Theorem 9.4.1. Exercises for Section 9.4
1 and its corresponding geometric series on B (0, 1). 9.4.1 Consider the function f (x) = 1 x i) Compute a derivative of f as a rational function and as a power series. ii) Compute an antiderivative of f as a rational function and as a power series.
9.4.2 Let f (x) = k=0 (2x k+1)k! . i) Find the radius of convergence of the series. ii) Compute the derivative of f . xk x2 iii) Suppose that we know that ex = k=0 k! . Find the series for e . 2 iv) Find an antiderivative of ex . v) Address the no-closed-form discussion on page 183.
2k+1
9.5
Numerical evaluations of some series
1 For all complex numbers x B (0, 1) the geometric series k=0 xk converges to 1 x. 1 Certainly it is much easier to compute 1x than the innite sum. We next exploit geometric series and derivatives of power series to compute a few other innite sums.
Example 9.5.1
k k=1 2k1
= 4.
k Proof. Let f (x) = k=0 x . This is the geometric series with radius of convergence 1 that 1 k 1 k 1 (Example 9.3.2). By Theorem 9.4.3, f (x) = = , converges to 1 k=0 kx k=1 kx x 1 and by Theorem 9.4.1, the radius of convergence of f is also 1. Thus 2 is in the domain of f . From 1 1 f ( x) = = 1x (1 x)2
we deduce that Example 9.5.2
k k=0 2k1 k2 k=0 2k
= f (1/2) = = 6, and
1 (11/2)2
= 4. = 12.
k2 k=0 2k1
244
Section 9.6: A special function
k Proof. As in the previous example we start with the geometric series f (x) = k=0 x 1 k 1 that converges on B (0, 1). Its derivative f (x) = = (1 k=0 kx x)2 also converges k 2 k 1 on B (0, 1). Then xf (x) = also k=0 kx and its derivative (xf (x)) = k=0 k x converge on B (0, 1). From
(xf (x)) = we deduce that to verify.

k2 k=0 2k
x (1 x)2 +
2x 1 + 2 (1 x) (1 x)3
1 (11/2)2
1 (11/2)3
= 4 + 8 = 12. The other part is also easy
Exercises for Section 9.5 9.5.1 Compute 9.5.2 Compute

1 k=1 k2k . 1 k=1 k3k . k k=1 k+23k .
9.5.3 Consider According to my computer the partial sum of the rst 1000 terms is a huge fraction that takes several screen pages. So I decide to compute curtailed decimal expansions instead. Then, according to my computer, the partial sum of the rst 10 terms is (about) 0.719474635555091, the partial sum of the rst 100 terms is (about) 0.719487336054311, the partial sum of the rst 1000 terms is (about) 0.719487336054311. What can you suspect? How would you go about proving it?
9.6
A special function
Dene the power series E ( x) =
k=0
xk . k!
By Examples 9.3.6 (3), the domain of E is C. Thus by Section 9.4, E is dierentiable everywhere. Remarks 9.6.1 k 1 k 1 xk (1) E (x) = k=1 kxk! = k=1 (x k=0 k! = E (x). k1)! = (2) E (0) = 1. (3) Dene f : C C be given by f (x) = E (x) E (x). Then f is a product of two dierentiable functions, hence dierentiable, and f (x) = E (x) E (x) + E (x) E (x) (1) = E (x) E (x) E (x) E (x) = 0, so that f is a constant function. But this constant has to equal E (0) E (0) = 1. 1 We conclude that E (x) is never 0, and that E (x) = E ( x) .
245
(4) Let a C. Dene g : C C be given by g (x) = E (x + a) E (x). Then g is a product of two dierentiable functions, hence dierentiable, and g (x) = E (x + a) E (x) E (x + a) E (x) = 0, so that g is a constant function. But this constant has to equal E (a) E (0) = E (a). 1 We conclude that E (x + a) = E (a) E ( x) = E (a)E (x). Thus for all a, b C, E (a + b) = E (a)E (b). By induction on n and by the previous part, it follows that for all positive integers n and all a C, E (na) = (E (a))n . By parts (3) and (6), E (na) = (E (a))n for all integers n and all a C. Let a, b R. By part (5), E (a + bi) = E (a) E (bi). Thus to understand the function E : C C, it suces to understand E restricted to real numbers and E restricted to i times real numbers.
(5) (6) (7) (8)
For the rest of this section we restrict the domain of E to R, and we analyze the case E restricted to i times real numbers in the next section. Since E : C C is dierentiable, so is the restriction of E to any subset, and hence E : R R is continuous. By part (3) above, E never takes on value 0, so that by the Intermediate value theorem, E of real numbers is either always positive or always negative. Since E (0) = 1, it follows that E takes R to R+ . Let m, n Z with n = 0. Then by part (6) above, m n m = E (m) = E (m 1) = (E (1)) . E n Since E
m n
> 0, it follows that E
m = (E (1))m/n. n Now let r R. Let {sn } be a sequence of rational numbers that converges to r . Then by continuity of E , E (r ) = lim E (sn ) = lim E (1)sn . By Theorem 7.4.5, the exponential function with a xed positive base is dierentiable and hence continuous, so that E (r ) = E (1)r . Thus E restricted to R is simply exponentiating E (1). 1 What is this special number E (1)? The sequence { n k=0 k! }n of partial sums for E (1) 2.66667, 65 2.71667, 1957 is convergent, and starts with 1, 2, 2.5, 8 2.70833, 163 3 = 24 = 60 = 720 = 19601 98641 685 2.71806, 252 = 2.71825, 40320 = 2.71828, 36288 = 2.71828. The last decimal approximation turns out to be correct to 5 decimal places and looks a lot like Eulers constant e from Denition 7.4.6.
246
Section 9.7: A special function, continued
Theorem 9.6.2 E (1) equals e = ln1 (1), and furthermore, for all x R, E (x) = ln1 (x) = ex . Proof. Dene f : R R by f (x) =
E (x ) . ln1 (x)
Then f is dierentiable and
E (x) ln1 (x) E (x)(ln1 ) (x) E (x) ln1 (x) E (x) ln1 (x) f ( x) = = = 0. (ln1 (x))2 (ln1 (x))2 Thus f is a constant function, so that for all x R, f (x) = f (0) =
1 x E (0) 1 = 1 ln1 (0) 1
= 1. Thus
E (x) = ln
(x), and by Theorem 7.4.7, this is e . In particular, E (1) = e = e.
The following is now immediate: Theorem 9.6.3 The function E : R R+ has the following properties: (1) The range is all of R+ . (2) By Theorem 7.4.7, E 1 (x) = ln(x).
Exercises for Section 9.6 9.6.1 Use the denition of ln in this section to prove that for all x, y R+ , ln(xy ) = ln x + ln y and that for all integers n, ln(xn ) = n ln x.
1 . 9.6.2 Use the denition of ln in this section to prove that (ln) (x) = x
9.6.4 Determine a power series whose derivative is e(x ) . (It turns out that there is no simpler, nite-term antiderivative of this function.).
2
9.6.3 Prove that {f : R R dierentiable | f = f } = {cE (restricted to R) : c R}. Similarly, prove that {f : C C dierentiable | f = f } = {cE : c C}.
2
9.6.5 Determine a power series whose derivative is 2(x ) . (It turns out that there is no simpler, nite-term antiderivative of this function refer to the no-closed-form discussion on page 183). 9.6.6 Express 9.6.7 Express
(1)k k=0 4k k! in 2k k=0 3k (k+1)!
terms of e. in terms of e.
9.6.8 (Unusual.) We have seen two ways of computing digits of e: as k=0 x k! and as that number at which ln has value 1. Play with the two methods (on a computer) and see which one is faster. 9.6.9 Find a Taylor polynomial expansion of the function E centered at 0. (Do as little work as possible.) 9.6.10 Numerically evaluate e2 from the power series expansion to 5 signicant digits. Prove that you have achieved desired precision. (Hint: Theorem 6.4.5.)
247
9.7
A special function, continued
In this section restrict E from the previous section to the imaginary axis. Throughout, all x are real, and we thus analyze the function E (ix). Note that E (ix) =
k=0
(ix)k k!
ix (ix)2 (ix)3 (ix)4 (ix)5 (ix)6 (ix)7 (ix)8 (ix)9 + + + + + + + + + 1! 2! 3! 4! 5! 6! 7! 8! 9! x x2 x3 x4 x5 x6 x7 x8 x9 =1+i i + +i i + + i + . 1! 2! 3! 4! 5! 6! 7! 8! 9! We dene two new functions (their names may be purely coincidental, but pronounce them for the time being as cause and sin): =1+ x4 x6 x8 x2 + + + , 2! 4! 6! 8! x x3 x5 x7 x9 + + + . SIN (x) = Im E (ix) = + 1! 3! 5! 7! 9! Since E (ix) converges, so do its real and imaginary parts, so that the domains of COS and SIN are both all of R. COS (x) = Re E (ix) = 1 Remarks 9.7.1 (1) E (ix) = COS (x) + i SIN (x). (2) COS (0) = Re E (i 0) = Re 1 = 1, SIN (0) = Im E (i 0) = Im 1 = 0. COS (x) = COS (x), (4) For all x R, SIN (x) = SIN (x). (3) By the powers appearing in the power series for the two functions, for all x R, Thus E (ix) = COS (x) i SIN (x), which is the complex conjugate of E (ix). (COS (x))2 + (SIN (x))2 = (COS (x) + i SIN (x)) (COS (x) i SIN (x)) = E (ix ix) = E (ix)E (ix)
= E (0) = 1. (5) We conclude that for all x R, Thus |E (ix)| = 1 for all x R.
1 COS (x), SIN (x) 1.
248 (6) Since E is dierentiable,
(E (ix)) = E (x)i = E (x)i = iCOS (x) SIN (x). It follows that (COS (x)) = (Re(E (ix))) = Re((E (ix))) = SIN (x), and (SIN (x)) = (Im(E (ix))) = Im((E (ix)) ) = COS (x). Theorem 9.7.2 There exists a unique real number s (0, 3) such that E (is) = i, i.e., COS (s) = 0 and SIN (s) = 1. Proof. The function t SIN (t) is dierentiable and its derivative is 1 COS (t), which is always non-negative. Thus t SIN (t) is non-decreasing for all real t, so that t SIN (t) 2 0 SIN (0) = 0. Hence the function t2 + COS (t) has a non-negative derivative on [0, ), 2 2 so that t2 + COS (t) is non-decreasing on [0, ). Thus for all t 0, t2 + COS (t) t3 02 2 + COS (0) = 1. It follows that the function 6 t + SIN (t) is non-decreasing on [0, ). 3 [How long will we keep going like this???] Thus for all t 0, t6 t + SIN (t) 3 t4 t2 0 6 0 + SIN (0) = 0. Thus 24 2 COS (t) is non-decreasing on [0, ), so that for all 2 2 04 t4 t2 COS (t) 24 0 t 0, 24 2 COS (0) = 1. We conclude that for all t 0,
2 4 3 3 2 + 1 = 1 and In particular, COS ( 3) 24 8 < 0. Since COS is continuous COS (0) = 1 > 0, by the Intermediate value theorem there exists s (0, 3) such that COS (s) = 1. 3 3 t We proved that for all t 0, t6 t + SIN (t) 0, so that SIN (t) t t6 = 6 (6 t2 ). t (6 t2 ) > 0. But then COS is decreasing on (0, 3]), so If t (0, 3], then SIN (t) 6 that s with the specied properties is unique.
COS (t)
t4 t2 + 1. 24 2
Denition 9.7.3 Any real number x uniquely determines a point in C on the unit circle at angle x radians counterclockwise away from the positive x-axis. This point has coordinates (cos(x), sin(x)). This is our denition of cos and sin. Theorem 9.7.4 COS and SIN are the functions cos and sin, and s = /2. Proof. We know that E (ix) is a point on the unit circle with coordinates (COS (x), SIN (x)). What we do not yet know is whether the angle of this point counterclockwise from the positive real axis equals x radians. Let s be as in Theorem 9.7.2. The angle (in radians, counterclockwise from the positive real axis) of E (is) = 0 + i 1 = i is /2. The proof of Theorem 9.7.2 showed for t [0, 1],
249
the imaginary part SIN (st) of E (ist) is positive, and so by the derivative properties, the real part COS (st) of E (ist) is decreasing. Thus on [0, 1], COS (st) decreases from 1 to 0, and SIN (st) increases from 0 to 1. Thus E (ist) is in the rst quadrant. It follows that for all positive integers n, E (is/n) is a point on the unit circle in the rst quadrant and i = E (is) = (E (is/n))n, so that by Theorem 3.10.5, the angle of E (is/n) is /(2n). Similarly, the angle of E (is/n) is the angle of 1/E (is/n), which is /(2n). Thus again by Theorem 3.10.5, for all rational numbers t, the angle of E (ist) is t/2. Since E is a continuous function, it follows that for all real t, the angle of E (ist) is t/2. Thus COS (st) + i SIN (st) = E (ist) = cos(st) + i sin(st), so that by uniqueness of real and imaginary parts, COS (st) = cos(st) and SIN (st) = sin(st). Since s = 0, it follows that for all real x, COS (x) = cos(x) and SIN (x) = sin(x). In particular, s = /2. Theorem 9.7.5 sin, cos are dierentiable and continuous functions. Proof. By Remarks 9.6.1, E is dierentiable. Hence its real and imaginary parts are dierentiable and continuous, i.e., cos(x) = COS (x) = Re E (ix) and sin(x) = SIN (x) = Im E (ix) are dierentiable and continuous. Theorem 9.7.6 Every complex number x can be written in the form rE (i ), where r = |x| is the length and the angle of x counterclockwise from the positive real axis. Proof. Let r = |x|. Then x lies on the circle centered at 0 and of radius x. If r = 0, then x = 0, and the angle is irrelevant. If instead r is non-zero, it is necessarily positive, and x/r a complex number of length |x|/r = 1 and by Theorem 3.10.4, x/r and x have the same angle. Let be that angle. Then x/r = E (i ), so that x = rE (i ). Notation 9.7.7 It is common to write E (x) = ex for any complex number x. We have seen that equality does hold if x is real, but we adopt this notation also for other numbers. With this, if x, y R, then ex+iy = ex eiy , and ex is the length and y is the angle of ex+iy counterclockwise from the positive x axis.
9.7.1 Prove that ei + 1 = 0. (This has been called one of the most beautiful equalities.) 9.7.2 Prove that cos(x + y ) = cos(x) cos(y ) sin(x) sin(y ), and that sin(x + y ) = sin(x) cos(y ) + cos(x) sin(y ). (You might have seen trigonometric proofs of this in high school. You should provide a proof with methods from this chapter.) 9.7.3 Prove de Moivres formula: for all n Z, (cos(x) + i sin(x))n = cos(nx) + i sin(nx). 9.7.4 Prove that for all x R, (sin(x))2 =
1 cos(2x) , 2
(cos(x))2 =
1 + cos(2x) . 2
9.7.5 For the moment pretend that we do not know that COS = cos and SIN = sin. i) Justify that = 2
1
1+
0
2x 2 1 x2
dx.
(Hint: lengths of curves, and the Greeks dened to be the perimeter of half the circle of radius 1.) ii) Compute the improper integral via substitution u = COS (x) and further other steps. Your answer should be s from Theorem 9.7.2. 9.7.6 Express the following complex numbers in the form ex+iy : i, i, 1, 1, e, 2 + 2i. Can 0 be expressed in this way? Justify. 9.7.7 Compute the derivatives of the standard trigonometric functions: tan(x) = cot(x) =
cos(x) sin(x) , sin(x) cos(x) ,
sec(x) =
1 cos(x) ,
csc(x) =
1 sin(x) .
9.7.8 Use integration by substitution (Exercise 7.3.6) to compute antiderivatives of tan(x) and cot(x). 9.7.9 Use integration by substitution (Exercise 7.3.6) to compute an antiderivative of sec(x)+tan(x) 1 sec(x) = cos( x) = sec(x)(sec(x)+tan(x)) . With a similar rewriting trick compute an antiderivative of csc(x). 9.7.10 Determine a power series whose derivative is sin(x2 ). Repeat for cos(x2 ). (It turns out that there are no simpler, nite-term antiderivatives of these functions.)
Appendix A.
Advice on writing mathematics
Make your arguments succinct and straightforward to read Write preliminary arguments for yourself on scratch paper: your rst attempt may yield some dead ends which denitely should not be on the nal write-up. In the nal write-up, write succinctly and clearly; write what you mean and mean what you write; write with the goal of not being misunderstood. Use good English grammar, punctuate properly. And above all, use correct logical reasoning. Do not allow yourself to turn in work that is half-thought out or that is produced in a hurry. Use your best writing, in correct logical order, with good spatial organization on paper, with only occasional crossing out of words or sections, on neat paper. Represent your reasoning and yourself well. Take pride in your good work. Process is important Perhaps the nal answer to the question is 42. It is not sucient to simply write 42, The answer is 42, or similar, without the process that led to that answer. While it is extremely benecial to have the intuition, the smarts, the mental calculating and reasoning capacity, the inspiration, or what-not, to conclude 42, a huge part of learning and understanding is to be able to explain clearly the reasoning that lead to your answer. I encourage you to discuss the homework with others before, during or after completing it: the explanations back-and-forth will make you a better thinker and expositor. Write your solutions in your own words on your own, and for full disclosure write the names of all of your collaborators on the work that you turn in for credit. I do not take points o, but you should practice full honesty. Sometimes you may want to consult a book or the internet. Again, on the work that you turn in disclose the help that you got from outside sources. Keep in mind that the more you have to consult outside sources, the more fragile your stand-alone knowledge is, the less well understand the material, and the less likely you are to be able to do satisfactory work on closed-book or limited-time projects. Do not divide by 0 Never write 1/0, 0/0, 02 /0, /0. (Erase from your mind that you ever saw this in print! It cannot exist.) Sometimes division by zero creeps in in subtler ways. For example, to nd solutions 2 to x = 3x, it is wrong to simply cancel xs on both sides to get only one solution x = 3. Yes, x = 3 is one of the solutions, but x = 0 is another one. Cancellation of x in x2 = 3x amounts to dividing by 0 in case x is the solution x = 0.
252
Appendix A: Advice on writing mathematics
Never plug numbers into a function that are not in the domain of the function By design, the only numbers you can plug into a function are those that are in the domain of the function. What else is there to say? 1 (see I will say more, by way of examples. Never plug 0 into the function f (x) = x x previous admonition). Even never plug 0 into the function f (x) = x : the latter function is undened at x = 0 and is constant 1 at all other x. Never plug 1 into or into ln. 3x 4, if x > 3; Do not plug x = 0 or x = 1.12 into f that is dened by f (x) = 2x + 1, if x < 1. Order of writing is important The meanings of Everybody is loved by somebody and Somebody loves everybody are very dierent. Another way of phrasing these two statements is as follows: For every x there exists y such that y loves x, and There exists y such that for every x, y loves x. In crisp symbols, without distracting such that, for, and commas, these are written as x y, y loves x and y x, y loves x. Conclusion: order of quantiers matters. Certainly also the order matters in implications; simply consider the truth values of If x > 2 then x > 0 and If x > 0 then x > 2. The statement if A then B can be written also as B if A, but in general it is better to avoid the latter usage. In particular, when writing a long proof, do not write out a very long B and only at the very end add that you were assuming A throughout; you could have lost the doubting-your-statements reader before the disclaimer. Here is a fairly short example where B if A form is not as elegant. With the statement 2 x is ... if x = 0 you might have gone over the abyss of dividing by 0 and no ifs can make you whole again. It is thus better to rst obtain proper assurances, and write if x = 0 2 is ... . then x does not equal 3.14159 If the answer to a problem is 17/59, leave it at that. This is an exact number from which one can get an approximation to arbitrary precision, but from 0.21954437870195 one cannot recover further digits. Never write 17/59 = .21954437870195, but it is ne to write 17/59 = .21954437870195. Usually it is not necessary to write out the numerical approximation, but sometimes it is helpful to write one down to get a sense of the size of the answer, and to check that with any intuition about the problem. The answer to what how far a person can run in one minute certainly should not exceed a kilometer or a mile.
Appendix A: Advice on writing mathematics To prove an (in)equality, manipulate one side in steps to get to the other If you have to prove that
n k=1 1
253
k2 =
n(n+1)(2n+1) 6
for n = 1, do NOT do the following: DO NOT DO THIS!
k2 =
k=1
1(1 + 1)(2 1 + 1) 6
12 =
123 6 1=1
The reasoning above is wrong-headed because in the rst line you are asserting the equality that you are expected to prove, and in subsequent lines you are simply repeating your assumptions more succinctly. If you add question marks over the three equal sums and a check mark on the last line, then you are at least acknowledging that you are not yet sure of the equality. However, even writing with question marks over equal signs is inelegant and long-winded. That kind of writing is what we do on scratch paper to get our bearings on how to tackle the problem. But a cleaned-up version of the proof would be better as follows: 1 1(1 + 1)(2 1 + 1) 123 = . k 2 = 12 = 1 = 6 6
k=1
Another reason why the three-line reasoning above is bad is because it can lead to the following nonsense: ? 1=0 ? 2=1 ? 0=0
add 1 to both sides: multiply both sides by 0:
But we certainly cannot conclude that the rst line 1 = 0 is correct. Write parentheses is not a recognized binary operator. Do not write 5 2; instead write 5 (2).
x1
lim 4 3x = 4 3x, whereas lim (4 3x) = 7.

x1
4 3x dx is terrible grammar; instead write (4 3x) dx. Do not assume what has to be proved If you have to prove that a is positive, do not assume that a is positive. (For a proof by contradiction you may want to suppose that a 0, hopefully you get a contradiction, and hence a must have been positive.)
254 Limit versus function value
Appendix A: Advice on writing mathematics
Asking for limxa f (x) is in general dierent from asking for f (a). (If the latter is always meant, would there be a point to developing the theory of limits?) Do not start a sentence with a mathematical symbol This admonition is in a sense an arbitrary stylistic point, but it helps avoid certain confusions, such as in Let > 0. x can be taken to be negative. Possibly one could read or confuse 0. x as 0 x = 0, but then would have to be both positive and negative. Do not force a reader to have to do a double-take: write unambiguous and correct sentences.
Appendix B.
Logic
What you should never forget
You should remember the basic truth tables, correct usage of or and of implications, how to justify/prove a statement, and how to negate a statement. If A implies B and if A is true, we may conclude B . If A implies B and if B is false, we may conclude that A is false. If A implies B and if A is false, we may not conclude anything. If A implies B and if B is true, we may not conclude anything.
Truth table: P T T F F Q T F T F P F F T T P Q T F F F P Q T T T F P xor Q F T T F P Q T F T T P Q T F F T
Statement P (via contradiction). P and Q. P or Q.
How to prove it Suppose P . Establish some nonsense. Prove P . Prove Q. Suppose that P is false. Then prove Q. Alternatively: Suppose that Q is false and then prove P . (It may even be the case that P is true always. Then simply prove P . Or simply prove Q.) Suppose that P is true. Then prove Q. Contrapositively: Suppose that Q is false. Prove that P is false. Prove P Q. Prove Q P . Let x be arbitrary of the specied type. Prove that property P holds for x. Find/construct an x of the specied type. Prove that property P holds for x. Alternatively, invoke a theorem guaranteeing that such x exists.
If P then Q.
P Q. For all x of a specied type, property P holds for x. There exists x of a specied type such that property P holds for x.
256 An element x of a specied type with property P is unique.
Appendix B: What you should never forget Suppose that x and x are both of specied type and satisfy property P . Prove that x = x . Alternatively, show that x is the only solution to an equation, or the only element on a list, or ....
Statement P P and Q P or Q P Q For all x of a specied type, property P holds for x. There exists x of a specied type such that property P holds for x.
Negation P (P Q) = (P ) (Q) (P Q) = (P ) (Q) (P Q) = P (Q) There exists x of the specied type such that P is false for x. For all x of the specied type, P is false for x.
Geometric series
k=1
For all r C \ {1},
r k diverges if |r | 1 and converges to

n k k=0 r
n+1
1 r 1 .
r 1 r
if |r | < 1.
Mathematical induction The goal is to prove a property for all integers n n0 . First prove the base case, namely that the property holds for n0 . For the inductive step, assume that for some n 1 n0 , the property holds for n 1 (alternatively, for n0 , n0 + 1, . . . , n 1), and then prove the property for n. The limit denition of derivative f (a + h) f (a) f (x) f (a) = lim . xa h0 h xa
f (a) = lim
Appendix B: What you should never forget The limit-partition denition of integrals
257
A partition of [a, b] is a nite set P = {x0 , x1 , . . . , xn } such that x0 = a < x1 < x2 < < xn1 < xn = b. Let f : [a, b] R be a bounded function. For each i = 1, . . . , n, let mi = inf {f (x) : x [xi1 , xi ]}, Mi = sup {f (x) : x [xi1 , xi]}. The lower sum of f with respect to P is
n
L(f, P ) =
i=1
mi (xi xi1 ).
The upper sum of f with respect to P is

n
U (f, P ) =
i=1
Mi (xi xi1 ).
The lower integral of f over [a, b] is L(f ) = sup {L(f, P ) : as P varies over partitions of [a, b]}, and the upper integral of f over [a, b] is U (f ) = inf {U (f, P ) : as P varies over partitions of [a, b]}. We say that f is integrable over [a, b] when L(f ) = U (f ). We call this common value the integral of f over [a, b], and we write it as
b b b
f=
a a
f (x) dx =
a
f (t) dt.
The Fundamental theorems of calculus I: Let f, g : [a, b] R such that f is continuous and g is dierentiable with g = f . Then
b a
f = g (b) g (a).
II: Let f : [a, b] R be continuous. Then for all x [a, b], f is integrable over [a, x], and x the function g : [a, b] R given by g (x) = a f is dierentiable on (a, x) with d dx Never divide by 0 It bears repeating. Similarly do not plug 0 or negative numbers into ln, do not plug negative numbers into the square root function, do not ascribe a function (or a person) a task that makes no sense.
x
f = f ( x) .
a
Appendix C.
Solutions to selected exercises
1.1.9: (1) R is true, Q is false, P is true (2) P is true, Q is true, R is false (3) Brown is true, red is false, green is false (4) Worms is true, wet grass is true, rain is true, grass watered is ambiguous. 1.2.2: (1) For all numbers b, c > 0, there is no number a such that a3 = b3 + c3 . (2) There do not exist whole numbers a, b > 1 such that 7 = a b. (3) For every real number a there exists a whole number b greater than a such that there do not exist whole numbers c, d > 1 such that c d = b. 1.3.2: Let n be an odd integer. Then n = 2m + 1 for some integer m. The successor of n is n + 1 = (2m + 1) + 1 = 2m + 2 = 2(m + 1), and since m + 1 is an integer, it follows that n + 1 is even. 1.5.27: (1) Sn = 1, 3, 7, 15, 31, . . .. (2) We can see that a recursive denition would be Sn = 2(Sn1 ) + 1, where S1 = 1, because moving the nth stack requires moving the (n 1)th stack, then moving the nth piece one peg, and then moving the entire (n 1)th stack on top of that piece. Using intuition and pattern recognition (by observing the numbers) we can see that Sn = 2n 1. (3) Let Pn = 2n 1, and Sn = 2(Sn1 ) + 1. Prove that Pn = Sn for all n Z+ Base case: Let n = 1. S1 = 1, and 21 1 = 1. Induction Step: Assume Pn = Sn for some n. Prove, based on this assumption, that Pn+1 = Sn+1 : Sn+1 = 2Sn + 1 (by the recursive formula we wrote) = 2(2n 1) + 1 (by inductive assumption) = 2 2n 2 + 1 = 2n+1 1
1.5.28: Inductive step only, for which n?: Without loss of generality 1 < a1 < < an+1 . If a1 , . . . , an 2(n 1), then by induction (for which n?) one of a1 , . . . , an divides another one. So we may assume that an > 2(n 1), so that an = 2n 1, an+1 = 2n. If a1 = 2, then a1 divides an+1 . So we may assume that a1 > 2, and similarly that n is not one of the ai . For each ai < n, let ki be a positive integer such that 2ki ai {n + 1, . . . , 2n}. Argue that for all distinct i, i , 2ki ai = 2ki ai . Thus each ai {3, . . . , n 1} eliminates a unique possibility of some 2ki ai being aj for some j . Now ... 2.3.9: (Reexivity) For all m, n N0 , (m, n) (m, n) because m n = m n. (Symmetry) If (m, n) (m , n ), then by denition of , m n = m n. By symmetry of equality, m n = m n , which by denition of says that (m , n ) (m, n). (Transitivity) Suppose that (m, n) (m , n ) and (m , n ) (m , n ). This means that m n = m n and that
Appendix C: Solutions to selected exercises m n = m n . Then (m n ) n = m (n n ) = (m n ) n = (m n) n = m (n n )
259
= (m n ) n = m (n n) = m (n n )
= m (n n)
= m (n n )
= (m n ) n
= (m n) n . Then by Theorem 3.3.2, m n = m n, so that (m, n) (m , n ). 3.1.4: Clearly T , as the dening condition is satised vacuously. Suppose that m T . Let n m+ and s n. As m+ = m {m}, either n = m or n m. In the former case, s m, and in the latter case, since m T , then s m. In either case, since m m+ , it follows that s m+ . Thus m+ T . This proves that T is an inductive set, and as N0 is the smallest inductive set, necessarily N0 = T . 3.1.5: Clearly n = 0, i.e., n = , is in T , since the empty set contains no elements. Now assume that n T . Suppose for contradiction that n+ n+ . Then n+ n {n}, so that either n+ n or n+ = n. Both contradict the assumption that n T , as n+ is neither an element nor a subset of n. Thus n+ n+ . Now suppose for contradiction that (n+ )+ n+ . Then n+ {n+ } n+ , which implies that n+ n+ , which contradicts the previous paragraph. Now suppose for contradiction that (n+ )+ n+ . In other words, (n+ )+ n {n}. Hence (n+ )+ n or (n+ )+ = n. In either case, since n n+ (n+ )+ , by Exercise 3.1.4, n n, which contradicts the assumption n T . Thus (n+ )+ n+ . We just proved that n+ T . Thus T is an inductive subset of N0 . Since N0 is the smallest inductive subset of itself, necessarily T = N0 , which proves the theorem. 3.7.10: Since L and R are non-empty, there exist x L and y R. Both x and y are in Q. By assumption x < y . If y 0, then x < 0 and necessarily also x = |x| > 0 y , so that x R, so z = x works. Similarly, if x > 0, then z = y works. So we may assume that y 0 and that x 0. Let z be y if y |x|, and otherwise let z = |x|. We can also make
260
Appendix 4: Solutions to selected exercises
sure that z is greater than or equal to |a|. Then z y, |a|, so that z R, and z x, so that z L. 3.7.11: This is indeed the same multiplication: (1) When L and L contain positive elements of Q, the new expression has the right side L L + R L+ + L+ R + R+ R+ = L L + R+ R+ = (, 0) (, 0) + R R = (0, ) + R R = R R , as desired. (2) When L but not L contains positive elements, then the new expression has the left side equal to L R+ + R+ L + L+ L+ + R R = (, 0) (0, ) + R+ L = (, 0) + R+ L = R+ L , again as desired. (3) When L but not L contains positive elements, the proof is symmetric to the previous case. (4) When neither L nor L contains positive elements, then the new expression has the right side equal to L L + R L+ + L+ R + R+ R+ = L L + R+ R+ = L L + (0, ) (0, ) = L L , again as desired. 3.11.13: Let a Int A. By denition there exists r > 0 such that B (a, r ) A. But then B (a, r ) B , so that a Int B . Now let a (Bd A) \ B . Since a Bd A, for all r > 0 the sets B (a, r ) A and B (a, r ) (F \ A) are not empty. Since A B , it follows that for all r > 0, B (a, r ) B is non-empty, and since a B , it follows that and B (a, r ) (F \ B ) is not empty. Thus a Bd B . Let A = {0} and B = F . Then Bd A = A and Bd B = . 4.1.8: (1) Let > 0. [Here does not matter, so we do not need to invoke it explicitly, we will simply implicitly use that = 34.] Let x satisfy 0 < x 0 < 34. Then x is positive, so | f ( x) 1| = x 1 = |1 1| = 0 < . | x|
(2) Let > 0. Let x satisfy 0 < 0 x < 34. Then x is negative, and x (1) = | 1 + 1| = 0 < . | x|
4.3.11: Let > 0. Since limxa g (x) = L, there exists > 0 such that for all x B , if 0 < |x a| < , then |g (x) L| < . But then for the same , for all x A with 0 < |x a| < , |f (x) L| = |g (x) L| < . 4.3.12: If limxa f (x) exists and equals L, then by an application of Exercise 4.3.11 with A = B , we have that limxa g (x) exists and equals L. By symmetry, if limxa g (x) exists and equals L, then limxa f (x) exists and equals L.
Index
SYMBOLS abstract binary operation 55 Bd boundary of a set 108 n k , n choose k 33 Ac complement of A 39 \ complement of sets 38 composition of functions 50 cos 248 e 189 {} empty set 36 empty set 36 there exists 19 for all 19 f n function composed with itself 50 sinn trigonometric function (not composition) 50 > (strictly) greater than 62 greater than or equal to 62 e (identity element) 55 id (identity function) 49 if and only if 13 implies 11 in a set 35 inf : inmum 63 Int interior of a set 108 intersection 37 < (strictly) smaller than 62, 63, 64, 79 > (strictly) greater than 79 greater than or equal to 79 less than or equal to 62, 78 lim of sequence 201 liminflim inf 226 limsuplim sup 226 lnnatural logarithm 187 logical not 10 logical or 11 logical and 11 L(f, P ): lower sum 172 Mi : maximum of a function on a subinterval 172 mi : minimum of a function on a subinterval 172
n n k=1 , k=1
product 27, 59
// end of proof 14 QED end of proof 14 end of proof 14 end of proof 14 Range range of a function 48 n 147, 148 sin 248 subset of a set 37
n n k=1 ,
27, 59
k=1
sup : supremum 62, 226 union 38 U (f, P ): upper sum 173 00 27 A absolute value 65 advice 251 alternating test for series 234 and (logical) 11 antecedent 11 antiderivative 182, 183 area 171 associative 56 addition 60 multiplication 60 axiom 71
262 B base (of exponential function) 189 base case (of induction) 29 Bernoullis inequality 167 bijective 50 composition 51 binary operation 55 associative 56 commutative 58 binomial coecient 34 binomial 34 boundary (Bd ) 108 bounded set 62 above 62 and open balls 113 below 62 C C 36, 99 absolute value 101 arithmetic 99 Cartesian coordinates 103 length 101 norm 101 polar coordinates 103 properties 99 reverse triangle inequality 107 triangle inequality 102 cancellation 77 Cartesian coordinates 103 Cartesian product 43 Cauchys criterion for series 232 Cauchys Mean value theorem 166 ceiling function 53 chain rule 161 chart how to prove 22, 255 negation 25, 256
Appendix : Index choose: n choose k, n k 33 closed and boundedcontained in open sets 113, 114, 115 closed set 110 closed-form antiderivative 183 closure (of a set) 110 codomain (of a function) 48 commutative 58 addition 60 multiplication 60 comparison for sequences 217 for series 232, 233 completeness of R, C 219 complex conjugate 101 complex number 99 composition of functions 51 conditional statement 11 antecedent, consequent 11 congruent modulo n 48 congruent modulo n 46 conjunction 11 consequent 11 construction of N0 71 construction of Q from Z 48, 86 construction of R 92 construction of Z from N0 46 continuous 140 and inverse function 145 and monotonocity 145 exponential function 149, 150 exponentiation 148 hence integrable 177 image of interval 145 piecewise 178 properties 141 uniformly continuous 151 cos: trigonometric function 248, 249
Appendix : Index D Dedekind cut 91 denition: beware 13 derivative 154 and monotonicity 166 complex-valued functions 162 inverse 162 Leibnitzs notation 154 Newtons notation 154 power rule for rational exponents 163 properties 158, 160, 161, 162, 163 dierentiable 154 hence continuous 158 disjunction 11 distributivity 60 division 61 do not divide by zero 61 domain (of a function) 48 implicit 50 E element 35 empty product 28, 58 empty set 36 empty sum 28 enumeration of Q+ 198 epsilon-delta denition 116, 122 epsilon-N denition 138 equivalence (logical) 13 equivalence class 45, 82 equivalence relation 45, 82 Euclid 25 Eulers constant 189, 246 even function, its integral 180 exclusive or (xor) 11 existential quantier 18, 19 exponential function 189, 245 continuity 149, 150 exponentiation 147, 148 by real exponents 189 continuity 148 exponent 189 Extreme value theorem 146
263
F Fibonacci sequence 198 eld 60, 88, 95, 99 axioms 60 Fixed point theorem 147 oor function 53 function 48 bijective 50 ceiling 53 codomain 48 composition 50 continuous 140 decreasing 68 inverse 68 domain 48 exponential 149, 150, 189, 245 oor 53 graph 48 identity 49 increasing 68 inverse 68 injective 50 one-to-one 50 onto 50 polynomial 51 range 48 rational 51 surjective 50 Fundamental theorem of arithmetic 24 Fundamental theorem of calculus 181, 183 complex-valued function 185
264 G generalized power function 189 geometric series 230 glb , inf : greatest lower bound 63 Goldbachs conjecture 9 Greatest lower bound theorem 98 greatest lower bound: glb, inmum, inf 63 H harmonic series 230 higher-order derivatives 168 how-to-prove chart 22, 255 I identity element 55 additive 60 multiplicative 60 identity function 49 if and only if, i 11 imaginary axis 100 imaginary part of a complex number 101 implication 11 improper integrals 187 improper integral 181 Indenite integral 183 induction 29, 72 base case 29 inductive step 29 inductive 71 inmum (inf,glb) 63 226 innite sequence: see sequence 197 innite series 229 injective 50 composition 51 integer 82 integrable 174, 176, 178 integral test for series 235
Appendix : Index integral 174 addition 178 applications 191 complex-valued function 185 monotonicity 179 notation 174, 175, 182, 184 integration by parts 187 integration by substitution 186 interior 108 Intermediate value theorem 143 interval 35, 64 inverse 56, 68 additive 56, 60 derivative 162 multiplicative 56, 60 of a strictly monotone function 68 invertible 57 L LH opitals rule 166, 168, 190, 191 least element 63, 80 Least upper bound theorem 98 least upper bound: lub, supremum, sup 62 length of curve 191 limit inferior, liminf 226 limit point (of set) 109 limit superior, limsup 226 limit absolute value 131 at innity 138 composite function theorem 132 epsilon-delta denition 116, 122 epsilon-N denition 138 innite 135 M-delta denition 135 M-N denition 138 of a function 116, 122, 135, 138 of sequence 201
Appendix : Index polynomial rule 131 power rule 130 rational rule 131 real/imaginary part 131 right/left-handed 122, 135 squeeze theorem 132 subsequence 225 theorems for innite limits 136 theorems 129, 130, 131, 132 uniqueness 128 ln, natural logarithm 187, 246 inverse 246 logic circuits 18 logical biconditional 13 lower bound of set 62 lower integral 174 lower sum (integrals) 172 lub , sup : least upper bound 62 M Mean value theorem for integrals 187 Mean value theorem 165 member (of a set) 35 modulo n 48 modus ponens 14 modus tollens 14 monotone function 68 and derivative 166 monotone sequence 220 N N 36 N0 71 properties 75 natural logarithm, ln 187 natural number (in N0 ) 71 negation 10, 25 chart 25, 256 negative 64, 85, 90, 98 non-negative 64, 85, 90, 98 non-positive 64, 85, 90, 98 notation integral 182, 184 set vs. sequence 197 O odd function, its integral 180 one-to-one 50 onto 50 open ball 107 open set 107 or (logical) 11 ordered pair 42 ordered set 64 eld 64 order 62 N0 78 Q 89 R 97 Z 85
265
P p-test for series 234 partial sum (of a sequence) 229 partition (of an interval) 172 ner 173 renement 173 Pascals triangle 33 piecewise continuous 178 pigeonhole principle 54 polynomial function 27, 51, 52, 131, 168, 169, 212, 216 Taylor polynomial 168 positive 64, 85, 90, 98 power function 68, 69, 130, 147, 212, 216 generalized 148, 189
266 power notation 27, 50, 58 00 27 for functions 50, 60 on sets with binary operation 58 trigonometry special 50 power series 237, 240 derivative 242 ratio test 238 root test 237 prime 9, 14, 23, 24, 25 n product k=1 27, 59 proof 14, 22 by contradiction 15 by induction 29 deductive reasoning 14 how-to-prove chart 22, 255 pigeonhole principle 54 2 is not rational 16 proper subset 37 Pythagoras 16 Q Q 36, 86 Archimedean property 90 cancellation 89 eld 88 ordered 89 qed (quod erat demonstrandum) 14 also QED 14 quadratic formula 52 quantier 18, 19 R R 36, 92 Archimedean property 98 eld 95 ordered 97 properties 95, 98
Appendix : Index radical function 147 radius of convergence (of power series) 238 range (of a function) 48 ratio test for sequences 222 for series 233 power series 238 rational number (in Q) 86 rational numbers between real numbers 98 real axis 100 real number 92 real part of a complex number 101 recursive 198 renement partition 173 relation on sets 44 equivalence 45 reexive 45, 79, 82 symmetric 45, 82 transitive 45, 82 reverse triangle inequality 65, 98, 107 over C 107 Rolles theorem 165 root of a number 147, 148 root test for power series 237 root test for series 234 S sequence vs. set notation 197 sequence 197 arithmetic of sequences 200 bounded 219 Cauchy 219 comparison for innite limits 208 comparison 212, 217 completeness of R, C 219 composite rule 212, 217 constant 197
Appendix : Index convergence via functions 210 convergence 201 properties 211, 212, 214, 216, 217 divergence to 207 divergence 206 Fibonacci 198 nite 201 limit 201 unique 211, 214 monotone 220 of partial sums 229 ratio test 222 squeeze theorem 213, 218 subsequence 223 term 197 series alternating test 234 comparison 232, 233 converges 229 diverges 229 geometric 230 harmonic 230 integral test 235 numerical evaluation 243 p-series test 234 power 237, 240 ratio test 233, 238 root test 234, 237 set vs. sequence notation 197 set 35 complement 38 disjoint 37 empty set 36 equality 37 intersection 37, 38 member 35 notation 35 propositionally 36
267 subset 37 sum of 67 union 38 universal 39 signed area 171, 173 sin: trigonometric function 248, 249 socks-and-shoes 60 statement 9 conditional statement 11 conjunction 11 disjunction 11 equivalence 13 if and only if 11 implication 11 logical biconditional 13 negation 10 xor 11 subsequence 223 subsequential limit 225 subset 37 successor 70 sum of sets 67 summation n k=1 27, 59 supremum (sup,lub) 62 226 surface area of a surface of revolution 194 surjective 50 composition 51 T tautology 10 Taylor polynomial 168, 191 Taylor series 239 Taylors remainder theorem 170 term of a sequence 197 topology 107, 109, 110 boundary 108 closed set 110
268 closure 110 interior 108 limit point 109 open ball 107 open set 107 Tower of Hanoi 32 triangle inequality 65 over C 102 reverse 65, 98, 107 trichotomy (of <) 63, 85 truth table 10 U uniform continuity 151 universal quantier 18, 19 universal set 39 upper bound of set 62 upper integral 174 upper sum (integrals) 173 V vacuously true 20 Venn diagram 39 volume of a surface of revolution 192, 193 W well-ordered 63, 80 whole number 82 X xor 11 Z Z 36, 82 properties 83, 85 zero do not divide by it 8, 61, 251 to zeroth power 27
Appendix : Index

Intro To Analysis

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Intro To Analysis

Uploaded by

Copyright:

Available Formats

Irena Swanson Reed College Spring 2013

The briefest overview

(a + b) (c + d) = ac + ad + bc + bd (a + b) (a b) = a2 b2 (a + b)2 = a2 + 2ab + b2 a b = ab (for which a, b?) Reconcile: 4 9 = (4)(9) = 36 = 6, 4 9 = 2i 3i = 6

Chapter 1: How we will do mathematics

Statements and proof methods

Section 1.1: Statements and proof methods

f is a constant function. F F T y >0 0 >0 0

Chapter 1: How we will do mathematics

Chapter 1: How we will do mathematics

Section 1.1: Statements and proof methods

Section 1.2: Statements with quantiers

input output input

Statements with quantiers

Chapter 1: How we will do mathematics

Section 1.2: Statements with quantiers

Chapter 1: How we will do mathematics

Section 1.3: More proof methods, and negation

More proof methods, and negation

An element x of a specied type with property P is unique.

Chapter 1: How we will do mathematics

Section 1.3: More proof methods, and negation

Chapter 1: How we will do mathematics

Section 1.3: More proof methods, and negation

Chapter 1: How we will do mathematics

Section 1.4: Summation

k 2 = 12 + 22 + 32 + 42 = 30. cos(k ) = cos(10 ) + cos(11 ) + cos(12 ) = 1 1 + 1 = 1. (4k 3 ) = 4 + 0 + 4 + 4 8 = 32.

1.4.3 Show that ( 1.4.4 Prove that

k=1 k=1 n n f ( k )) ( k =m k =m 0 0(1) . k=1 k = 2

f (k ) g (k ) is false (in general).

Chapter 1: How we will do mathematics

Proofs by (mathematical) induction

Section 1.5: Proofs by (mathematical) induction

(n 1)(n 1 + 1) + n (by induction assumption) 2 n2 n 2n + (by algebra) 2 2 n2 + n 2 n(n + 1) , 2

n(n + 1)(2n + 1) . 6 n(n + 1) 2

1.5.3 The sum of the rst n odd integers is n2 . 1.5.4 (2k 1) = n2 .

1 1.5.7 1 2 + 2 3 + 3 4 + + n(n + 1) = 3 n(n + 1)(n + 2).

Find a formula for An and prove it.

Section 1.5: Proofs by (mathematical) induction

Chapter 1: How we will do mathematics

1.6.3 Prove that

100 , 8 1.6.4 Compute 4 2 , 2 , 2 , and do cancellations before multiplications.)

. (Hint: be clever about it

Chapter 2: Concepts with which we will do mathematics

Chapter 2: Concepts with which we will do mathematics

Section 2.1: Sets

Ak = A1 A2 An = {a : a Ak for all k = 1, . . . , n}, Ak = A1 A2 An = {a : there exists k = 1, . . . , n such that a Ak },

Ak = {a : a Ak for all k I }, Ak = {a : there exists k I such that a Ak }.

Chapter 2: Concepts with which we will do mathematics

Section 2.1: Sets

Chapter 2: Concepts with which we will do mathematics

subsets with exactly i elements.

Section 2.2: Cartesian products

(1/k, 1/k ), (1/k, 1/k ),

[1/k, 1/k ], [1/k, 1/k ],

{1/k, 1/k }. {1/k, 1/k }.

Chapter 2: Concepts with which we will do mathematics

44 Exercises for Section 2.2

Section 2.3: Relations, equivalence relations

2.2.4 Prove that A (B C ) = (A B ) (A C ).

2.2.5 Prove that (A C ) (B D) = (A B ) (C D).

2.2.6 Prove that (A C ) (B D) = (A B ) (C D).

Relations, equivalence relations