Concepts in Calculus I Second Edition UNIVERSITY PRESS OF FLORIDA Florida A&M University, Tallahassee Florida Atlantic University, Boca Raton Florida Gulf Coast University, Ft. Myers Florida International University, Miami Florida State University, Tallahassee New College of Florida, Sarasota University of Central Florida, Orlando University of Florida, Gainesville University of North Florida, Jacksonville University of South Florida, Tampa University of West Florida, Pensacola Orange Grove Texts Plus   Concepts in Calculus I Second Edition Mikl6s B6na and Sergei Shabanov University of Florida Department of Mathematics UNIVERSITY PRESS OF FLORIDA Gainesville * Tallahassee * Tampa * Boca Raton Pensacola * Orlando * Miami * Jacksonville * Ft. Myers * Sarasota  Copyright 2012 by the University of Florida Board of Trustees on behalf of the University of Florida Department of Mathematics This work is licensed under a modified Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 Unported License. To view a copy of this license, visit http:// creativecommons.org/licenses/by-nc-nd/3.0/. You are free to electronically copy, distribute, and transmit this work if you attribute authorship. However, all printing rights are reserved by the University Press of Florida (http://www.upf.com). Please contact UPF for information about how to obtain copies of the work for print distribution. You must attribute the work in the manner specified by the author or licensor (but not in any way that suggests that they endorse you or your use of the work). For any reuse or distribution, you must make clear to others the license terms of this work. Any of the above conditions can be waived if you get permission from the University Press of Florida. Nothing in this license impairs or restricts the author's moral rights. ISBN 978-1-61610-170-1 Orange Grove Texts Plus is an imprint of the University Press of Florida, which is the scholarly publishing agency for the State University System of Florida, comprising Florida A&M University, Florida Atlantic University, Florida Gulf Coast University, Florida International University, Florida State University, New College of Florida, University of Central Florida, University of Florida, University of North Florida, University of South Florida, and University of West Florida. University Press of Florida 15 Northwest 15th Street FoiaDs nc - Flonida Distance Gainesville, FL 32611-2079 - .:w.mLearning Consortium http://www.upf.com  Contents Chapter 1. Functions 1 1. Functions 1 2. Classes of Functions 4 3. Operations on Functions 8 4. Viewing the Graphs of Functions 12 5. Inverse Functions 15 6. The Velocity Problem and the Tangent Problem 21 Chapter 2. Limits and Derivatives 25 7. The Limit of a Function 25 8. Limit Laws 33 9. Continuous Functions 39 10. Limits at Infinity 43 11. Derivatives 48 12. The Derivative as a Function 51 Chapter 3. Rules of Differentiation 57 13. Derivatives of Polynomial and Exponential Functions 57 14. The Product and Quotient Rules 61 15. Derivatives of Trigonometric Functions 64 16. The Chain Rule 67 17. Implicit Differentiation 71 18. Derivatives of Logarithmic Functions 74 19. Applications of Rates of Change 77 20. Related Rates 82 21. Linear Approximations and Differentials 89 Chapter 4. Applications of Differentiation 99 22. Minimum and Maximum Values 99 23. The Mean Value Theorem 106 24. The First and Second Derivative Tests 116 25. Taylor Polynomials and the Local Behavior of a Function 125 26. L'Hospital's Rule 133 27. Analyzing the Shape of a Graph 140 28. Optimization Problems 146 V  vi CONTENTS 29. Newton's Method 153 30. Antiderivatives 161 Chapter 5. Integration 167 31. Areas and Distances 167 32. The Definite Integral 176 33. The Fundamental Theorem of Calculus 188 34. Indefinite Integrals and the Net Change 194 35. The Substitution Rule 200  CHAPTER 1 Functions 1. Functions A function f is a rule that associates to each element x in a set D a unique element f(x) of another set R. Here the set D is called the domain of f, while the set R is called the range of f. The fact that f associates to each element of D an element of R is represented by the symbol f : D - R. Instead of saying that f associates f(x) to x, we often say that f sends x to f(x), which is shorter. See Figure 1.1 for an illustration. Domain Range x a b FIGURE 1.1. Domain and range. If the sets mentioned in the previous definition are sets of numbers, then it is often easier to describe f by an algebraic expression. Let N be the set of all natural numbers (which are the nonnegative integers). Then the function f: N - N given by the rule f(x) = 2x + 3 is the function that sends each nonnegative integer n to the nonnegative integer 2n+3. For instance, it sends 0 to 3, 1to 5, 17 to 37, and so on. In this case, the algebraic description is simpler than actually saying "f is the function that sends n to 2n + 3." The rule that describes f may be simple or complicated. It could be that a function is defined by cases such as r0.1x if 0< x <40, f(x) = 4+0.15(x-40) if40 80. This example could describe an income tax code. The first $40,000 of income is taxed at a rate of 10%, income above $40,000 but be- low $60,000 is taxed at a rate of 15%, and income above $80,000 is 1  2 1. FUNCTIONS taxed at a rate of 20%. The value of f(x) is the amount of tax to be paid after an income of x thousand dollars for any positive real number x. There are times when the rules that apply in various cases are closely connected to each other. A classic example is the absolute value function, that is, { z if f(x) =Iz|=0< . 3.0 2.5 2.0 1.5 1.0 0.5 -3 -2 -1 0 1 2 3 FIGURE 1.2. Graph of Iz. In this case, f(x) = f(-x) for all x. When that happens, we say that f is an even function. For instance, g(x) = cos x and h(x) = x2 are even functions. There are also functions for which -f(x) = f(-x) holds for all x. Then we say that f is an odd function. Examples of odd functions include g(x) = sin x and h(x) = x3. There are times when a plain English description of a function is simpler than an algebraic one. For instance, "let g be the function that sends each integer that is at least 2 into its largest prime divisor" is simpler than describing that function with algebraic symbols (and symbols of formal logic). If the sets D and R are not sets of numbers, an algebraic description may not even be possible. An example of this is when D and R are both sets of people and f (x) is the biological father of person x. Note that it is not by accident that we said that f (x) is the father (and not the son) of x. Indeed, a function must send x to a unique f(x). While a person has only one biological father, he or she may have several sons. Sometimes the rule that sends cc to f (x) can only be given by listing the value of f(xc) for each cc, as opposed to a general rule. For instance, let D be the set of 200 specific cities in the United States, let R be the set of all nonnegative real numbers, and for a city cc, let f(xc) be the  1. FUNCTIONS 3 amount of precipitation that x had in 2011. Then f is a function since it sends each x E D into an element of R. This function is given by its list of values, not by a rule that would specify how to compute f(x) if given x. Finally, functions can also be represented by their graphs. If f : D - R is a function, then let us consider a two-dimensional co- ordinate system such that the horizontal axis corresponds to elements of D, and the vertical axis corresponds to elements of R. The graph of f is the set of all points with coordinates (x, f(x)) such that x E D. The requirement that f(x) is unique for each x will ensure that no vertical line intersects the graph of f more than once. This is called the vertical line test. 1.1. Exercises. (1) For each person x, let f (x) denote the birthday (day, month, and year) of x. Is f a function? (2) For each person y, let g(y) denote the biological mother of y. Is g a function? If yes, what is the domain of g and what is the range of g? (3) For two people x and y, let us say that f(x) = y if y is a child of x. Is f a function? (4) How many functions are there with domain {A, B, C, D} and range {0, 1}? For the remaining exercises in this section, all functions are defined on some real numbers. (5) Let f(x) = x +|zl. Find the domain and the range of f. (6) Let f (x) (x + 1)/(x - 2). Find the domain and the range of f. (7) Let g(x) = z/4zl. Find the domain and the range of f. (8) Let f(x) = n. Find the domain and the range of f. (9) Let f(x) = sin2 x + cos2 c. Find the domain and the range of f. (10) Let h(x) =+3 + x3. Find the domain and the range of f. (11) Let Fc]be the smallest integer y such that x <; y. What is the domain and the range of the function cc - Fcc]l? (12) Let Fc]be defined as in the previous exercise and let f(xc)= Fcc] - cc. Find the domain and the range of cc. (13) Can the graph of a function intersect a vertical line twice? (14) Can the graph of a function contain a circle?  4 1. FUNCTIONS (15) Can the graph of a function intersect a horizontal line twice? (16) Can the graph of a function intersect a horizontal line infinitely many times? (17) An infinite sequence is an infinite array of numbers ai, a2, .... Explain why infinite sequences are, in fact, functions. What is the domain of these functions? (18) Let f(x) = 3x + 2. Find four points that are on the graph of f. What can be said about the curve determined by those four points? (19) Let f and g be two functions and let us assume that there is exactly one point (x, y) that is on the graph of both f and g. What is the algebraic meaning of that fact? (20) Let f, g, and h be three functions and let us assume that there is no point that is on the graph of all three of them. What is the algebraic meaning of that fact? 2. Classes of Functions 2.1. Power Functions. A power function is a function f given by the rule f(x) = ca, where a is a fixed real number. Note that x-a= 1/za, so, for instance, c-3c i/3. The special case of a = -1, that is, the function f(x) = 1/c, is called the reciprocal function. Note that the rule g(x) = 1 for all real numbers x also defines a power function, one in which a = 0. If a = 1/n, where n is a positive integer, then the power function f given by the rule f(X) =a - / n-c is also called a root function. 2.2. Polynomials. A polynomial function is the sum of a finite number of constant multiples of power functions with nonnegative integer ex- ponents, such as the function f given by the rule f(x) = 3x4 + 2x2 + 7x - 5. The domain of these functions is the set of all real numbers. The largest exponent that is present in a polynomial function is called the degree of the polynomial. So the degree of f in the last example is 4. The real numbers that multiply the power functions in a polynomial are called the coefficients of the polynomial. In the last example, they are 3, 2, 7, and -5. Some subclasses of polynomial functions have their own names as follows:  2. CLASSES OF FUNCTIONS 5 " Polynomials of degree 0, such as f(x) = 6, are called constant functions. " Polynomials of degree 1, such as g(x) = 3x - 2, are called linear functions. " Polynomials of degree 2, such as h(x) = x2 - 4x - 21, are called quadratic functions. " Polynomials of degree 3, such as p(x) = x3 - x2 + 6x - 2, are called cubic functions. 2.3. Rational Functions. A rational function is the ratio of two polyno- mial functions such as 3x2 + 4x - 7 R(x) = x - x3 - 8 The domain of a rational function is the set of all real numbers, except for the numbers that make the polynomial in the denominator 0. In the preceding example, the only such real number is x = 2. 2.4. Trigonometric Functions. Periodicity. The reader has surely en- countered the trigonometric functions sin, cos, tan, cot, sec, and csc in earlier courses. We will discuss these functions, and their inverses, later in the text. For now, we mention one of their interesting properties, their periodicity. A function f is called periodic with period T > 0 if f(x) = f(x + T) for all x and T is the smallest positive real number with this property. For example, sin and cos are both periodic with period 27r, and tan and cot are periodic with period 7. See Figure 1.3 for an illustration. The reader will be asked in Exercise 2.7.1 about the periodicity of sec and csc. 2.5. Algebraic Functions. An algebraic function is a function that con- tains only addition, subtraction, multiplication, division, and taking roots. For instance, power functions with integer exponents are al- gebraic functions, since they only use multiplication, though possibly many times. Therefore, polynomials are algebraic functions as well since they are sums of constant multiples of power functions. This implies that rational functions are also algebraic since they are obtained by dividing a polynomial (also an algebraic function) by another one. The preceding list did not contain all algebraic functions since it did not contain any functions in which roots were involved. So we get additional examples if we include roots, such as the functions given by the rules f (x) = x/ + 3, g(x) = x/, h(x) cv + 1)/(x - 1).  6 1. FUNCTIONS 6 1. FUNCTIONS sin x 1- 0.5 - 2z 2 2 2 -1 tan x 1 Cos x .5- 2- l 2 2 2 -0.5- -1- sec x Cot x 1 77 1 i2 E E E E v" i 2-z - 37 - 2 z z 2- FIGURE 1.3. Trigonometric functions. 2.6. Transcendental Functions. Functions that are not algebraic are called transcendental functions. These include trigonometric functions and their inverses, exponential functions, which are functions that con- tain a variable in the exponent, such as f(x) = 2x, and their inverses, which are called logarithmic functions. See Figure 1.4 for an illustra- tion. We will discuss these functions in later sections of this chapter. There are many additional examples, which do not have their own names. 2.7. Exercises. (1) Are secant and cosecant periodic functions? If yes, what is their period? (2) Can a polynomial be a periodic function? (3) Are f(x) = 3x5 + 7x - 31 and g(x) (2x + 7)/(3x - 1) polynomial functions? (4) Are f(x) = 2x and g(x) = sin2 x power functions?  2. CLASSES OF FUNCTIONS 7 Y 5 , 4 - , 3 2 /, 2 ,/ - log2(x) -5 -4 -3 -2 -1,' 1 2 3 4 5 -1- x -2 -3 , / -4 , , ' -5 FIGURE 1.4. Logarithmic functions. (5) Are 1/(x + 3), g(x) = (x2 + 3x + 9)/(x3 + 1), and h(x) (sin x)/(x + 2) rational functions? (6) Show an example of a rational function that is defined for all real numbers. (7) Show an example of a rational function that is defined for all real numbers except 1, 2, and 3. (8) Let f(x) = x2/3. Is f an algebraic function? (9) Is sin(3x) a periodic function? If yes, what is its period? (10) Is cos(1/x) a periodic function? If yes, what is its period? (11) Is sin(|x|) a periodic function? If yes, what is its period? (12) Show an example of a periodic function that has period 1. (13) Let f(x) = x-2/7. Is f an algebraic function? (14) Is g(x) = (2/3)x an algebraic function? (15) Show an example of a periodic function with period Vw7. (16) Is sin x + tan x a periodic function? If yes, what is its period? (17) Is sin x tan x a periodic function? If yes, what is its period? (18) Is sin2 x a periodic function? If yes, what is its period? (19) Let [x] be equal to the largest integer y such that y < x. Now set f(x) = x - [_]. Is f a periodic function? If yes, what is its period? (20) Let f and g be periodic functions defined for all real numbers. Let f have period 5 and let g have period 7. Is f +g a periodic function? If yes, what is its period?  8 1. FUNCTIONS 3. Operations on Functions 3.1. Transformations of a Function. We have seen the basic mathemat- ical functions and their graphs in the last section. In this section, we will look at their transformations. It is easy to see what happens to the graph of a function if we increase or decrease each value of a function by a constant. Indeed, the graph of the function g given by g(x) = f(x) + 5 for all x is simply the graph of the function f translated by five units to the north. Similarly, the graph of the function h given by h(x) = f(x) - 7 is the graph of f translated by seven units to the south. Horizontal translations are a little bit trickier. The reader is invited to verify that if g is the function given by g(x) = f(x - 2), then the graph of g is the graph of f translated by two units to the east, that is, in the positive direction. Indeed, we must substitute a larger number into g to get the same value as from f. For instance, g(8) = f(6). See Figure 1.5 for an illustration. Similarly, if h is the function given Y 10 9 8- -- f(x) - f(x)+5 -- f(x-2) 3 2 1- x -2 -1 0 1 2 3 4 FIGURE 1.5. Horizontal and vertical translations of f(x).  3. OPERATIONS ON FUNCTIONS 9 by h(x) = f(x + 3) for all x, then the graph of h is the graph of f translated by three units to the west, that is, in the negative direction. The interactive website http://www.math.ufl.edu/~mathguy/ufcalc book/translations.html provides further tools to visualize transforma- tions of functions. The effect of multiplication and division on functions can be de- scribed similarly. If f is a function and g is the function given by g(x) = c . f(x), where c > 1 is a real number, then the graph of g is simply the graph of f "stretched" vertically by a factor of c. That is, each point on the graph of g is c times as far away from the hor- izontal axis as the corresponding point on the graph of f. It goes without saying that dividing by c > 1 has the opposite effect. That is, if h(x) = f(x)/c, then the graph of h is a vertically compressed version of the graph of f. In other words, each point on the graph of h is c times as close to the horizontal axis as the corresponding point on the graph of f. See Figure 1.6 for an illustration. At this point, the reader should stop and think about what hap- pens if c < -1 is a negative constant. As the reader probably figured out, the stretching or compressing effect will not change (it will only depend on cl), but each point on the graph will be reflected through the horizontal axis. Y 3 2 - f(x) 1 - 1 - f(x) 3f(x) x 1 2 3 -f(x) -1 -2 FIGURE 1.6. Effects of multiplying a function by a constant. The reader is encouraged to consult the interactive website http:// www.math.ufl.edu/-mathguy/ufcalcbook/squeeze.html for further illustrations. Horizontal transformations involving multiplication and division are similar to their counterparts involving addition and subtraction in that their effect is the opposite of what one might think at first. If c > 1  10 1. FUNCTIONS and g is the function obtained from f by the rule g(x) = f(cx), then the graph of g is the graph of f compressed horizontally by a factor of c. That is, each point on the graph of g is c times as close to the vertical line as the corresponding point on the graph of f. In other words, if (x, y) is a point on the graph of f, then (x/c, y) is a point on the graph of g. On the other hand, if h is obtained from g by the rule h(x) = f(x/c), then the graph of h is a horizontally stretched version of the graph of f. That is, each point on the graph of h is c times as far from the vertical axis as the corresponding point on the graph of f. So if (x, y) is a point on the graph of f, then (cx, y) is a point on the graph of h. Again, the reader should stop for a minute and think about the graphs of the functions f(cx) and f(x/c) when c < -1 is a negative constant. 3.2. Combining Two Functions. If f and g are two functions, then their sum, difference, and product are defined wherever both f and g are defined. That is, the domain of f + g, f - g, and fg is the intersection of the domains of f and g. Furthermore, (f + g)(x) = f(x) + g(x), (f -g)(x) = f(x)--g(x), and (fg)(x) = f(x)g(x). Figure 1.7 illustrates the sum of two functions. We have to be just a little bit more careful with f/g, since this function is not defined when g(x) = 0, even if x is in the domain of both f and g. So the domain of f/g is the intersection of the domain of f and the domain of g, with the exception of the points x satisfying g(x) = 0. For each point of this domain, (f /g)(x) = f (x)/g(x). Y 3- 2 1 - ix --- f(x)+gWx -1- FIGURE 1.7. Adding two functions together. If the range of f is part of the domain of g, then we can compose f and g by first applying f and then g. The function we obtain in this  3. OPERATIONS ON FUNCTIONS 11 way sends x to g(f(x)) and is called the composition of f and g. It is denoted by f o g. Note that in f o g, first f, and then g is applied. EXAMPLE 1.1. Let R be the set of all real numbers. If f and g are both functions from R to R and f (x) = x2 and g(x) = x + 1, then (f og)() =g(fy(X)) = x2 + 1, while (g o f)(z) = f (g(x)) = (X + 1)2 =cX + 2x + 1. Note that f o g and g o f are, in general, different functions. 3.3. Exercises. (1) Sketch the graph of f(x) I / (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (2x + 5)2, Sketch the graph Sketch the graph Sketch the graph Sketch the graph Sketch the graph Sketch the graph Sketch the graph Sketch the graph Sketch the graph Sketch the graph of of of of of of of of of of f(xc) f(xc) f(xc) f(xc) f(xc) f(xc) f(xc) f(xc) f (xc) f(xc) = x2, g(x) =(x - 3)2, and h(x) (x + 4)2 and g(x) = x2 +4. |x + 5| and g(x) cc Iz+ 5. x+ 1 and g(x) cI + ll. sin(z/2) and g(x) = (sin z)/2. sin c|. cc and g(x) = 1/ cc. ex-3 and g(x) = ex - 3. ln(x2) and g(x) =_(ln z)2. cos(x + r) and g(x) =gr + cos c. v/z + 10 and g(x) = v/+10. = cos(2x), g(x) = sin(x - 2), and (12) Sketch the graph of f(x) h(x) = 3tan z. (13) Sketch the graph of f(x) h(x) = tan(3x). (14) Show examples for f and numbers, but f o g is not. 2 cos x, g(x) 2(sin z) - 2, and g when g o f is defined for all real (15) Show examples when f o g= g o f. (16) Show examples when (f o g) o h = f o (g o h). (17) Show examples when (f o g) o h f f o (g o h). (18) Sketch the graph of g(x) =sin(x - ) (19) Let f (x) = sincx and g(x) = x2. Determine f o g and g o f and sketch their graph. (20) Let f (x) = cos x, let g(x) = ex, and let h(x) = lncc. Determine fo(goh) and (fog)oh.  12 1. FUNCTIONS 4. Viewing the Graphs of Functions 4.1. The Graph of Function. The graph of a function f is the set {(x, f(x))| E D(f)}. The graph of a function is a good way of visually describing what a function does. Today, we have plenty of advanced tools, such as com- puter software packages and graphing calculators, to study the graph of functions. In this section, we point out a few of the common mistakes in using these tools. In order to facilitate the discussion, let us agree on some terminol- ogy. If the domain of f contains an interval I and for all real numbers x and x' in I, it is true that x < x' implies f(x) f(c'), then we say that f is decreasing on I. In terms of the graph of f, this means that the graph goes roughly from the northwest to the southeast. If we simply ask a computer or graphing calculator to plot the graph of a function without specifying the interval [zi, x2] in which the value of x can range, we may get an error message, or the computer may simply substitute default values for x1 and x2. For example, the soft- ware package Maple 13 uses the default values x1 = -10 and x2 = 10. The interval [1, x2] is often called the viewing window. See Figure 1.8 for an illustration. 4000 2000 -10 -5 5 10 -2000 -1.0 - -0.6 -0.4 -0.2 FIGURE 1.8. Viewing g(x) = 4x3 + 9x2 + 6x + 1 with viewing window [-10,10 and [-1,01. We have to be careful, however, since not all viewing windows are appropriate for all functions, and choosing an inappropriate viewing window may cause misleading results.  4. VIEWING THE GRAPHS OF FUNCTIONS 13 For functions like f(x) = x, g(x) = Ix, or h(x) = x2 + 3, the view- ing window [-10, 10] is appropriate as the behavior of these functions outside that window is similar to their behavior inside the window. Now let f(x) = (x + 10)2. In this case, using the viewing window [-10, 10], we get the graph of an increasing function. That is misleading since f is decreasing on the interval (-oc, -10]. So, in this case, a viewing window that starts at a point x1 < -10 is necessary. This problem becomes more difficult if we are dealing with functions that change from increasing to decreasing many times, perhaps in an irregular fashion and perhaps far away from the origin. For this reason, it is worth noting that if f is a polynomial function of degree n, then it cannot change directions more than n - 1 times. If we found all n - 1 direction changes, then we can be sure that we did not miss any of them. We will return to this topic in a later chapter, when we discuss the derivative of a function. The preceding example showed why selecting a viewing window that is too small can be misleading. The next example shows why a viewing window that is too large can also mislead us. Plot the graph of the function g(x) = 4x3 + 9x2 + 6x + 1. Using the default viewing window [-10, 10], or some window containing that one, many software packages will show a graph that increases everywhere and disappears in a small interval to the left of 0. This should raise our suspicion that the program does not properly display the graph of g around 0. Indeed, g is defined for all real numbers, so its graph should not disappear anywhere. Taking a closer look, that is, changing the viewing window to [-1, 1], we see a function that is actually decreasing between x = -1 and c = -1/2. Trigonometric functions, with their periodicity, are particularly good examples to demonstrate what software packages can and can- not do. The reader is encouraged to plot the graph of the functions sin x, cos(2x), tan(z/4), and, finally, sin(1/z) and explain the obtained graphs. In particular, the reader should try to explain why, for sin(1/x), the choice of the viewing window is not important as long as it contains x = 0. Applications of graphical representations of functions include count- ing the solutions of certain equations even when we cannot explicitly solve those equations, and finding asymptotes. (Note that a solution obtained by simply viewing the graphical representation of a function is not mathematically rigorous, but it can provide a first step for a more rigorous solution.) A horizontal asymptote of a function f is a horizontal line y =a so that the values of f(xc) are never equal to a,  14 1. FUNCTIONS but get closer and closer to it as x gets closer and closer to positive infinity or negative infinity. A vertical asymptote of f is a line x= b so that the function f is not defined at x = b, but as x gets closer and closer to b, the values of f(x) get closer and closer to infinity, or neg- ative infinity. For instance, the function f(x) = 1/x has a horizontal asymptote at y = 0, and a vertical asymptote at x = 0. We will make these notions more precise in the next chapter, when we introduce the concept of limits. For now, we can use a graphing software package to find asymptotes, as you will be asked to do in the exercises. 4.2. Exercises. In the following exercises, use a graphing software package with the appropriate viewing window to find the number of solutions (among real numbers) for the given equation. Also find the intervals on which the left-hand side is increasing and on which the left- hand side is decreasing. Approximate the endpoints of these intervals to one decimal. (1) X4 - X + 1 = 0. (2) X4c - 1 =0. (3) X3 - 6x + 1= 0. (4) X3 + X2 - 1 =0. (5) X4 - 4X2 + 1. In the following exercises, use a graphing software package with the appropriate viewing window to find the number of solutions (among real numbers) for the given equation. (6) X3 - X2 - 1 - 0. (7) X2 - x - 7 = X3 _ 1 (8) x = sin z. (9) x2 =sin z (10) x/2 = cos x. (11) x +2 = 2 . In the following exercises, use a graphing software package to decide if the given function has a vertical or horizontal asymptote. (12) f(x) - (x + 3)/(x + 2). (13) f(x) - (x - 7)/(x + 5). (14) g(x) 1/(2 - c). (15) h(x) = x+ (1/c). (16) h(xc) =cx2 - (1/cc2). (17) s(xc) = (cc - 4)/(xc - 3). (18) s(xc) = (cc + 5)/(xc - 7). (19) z(xc) =(cc2 + 1)7(2cc2 - 3). (20) z(xc) =(cc2 + 4cc + 5)/(xc + 3).  5. INVERSE FUNCTIONS 15 5. Inverse Functions The inverse f-1 of a function f A - B "undoes" what f did. That is, if f(x) = y, then f-1(y) = x, so f sends x to y, while f-1 sends y back to x. It goes without saying that this f-1 will only be a function if f-1(y) is unambiguous, that is, when there is only one x E A so that f(x) = y. In that case, and only in that case, it is clear that f -1(y) = x. Let us now formalize these concepts. DEFINITION 1.1. A function f A - B is called one-to-one if it sends different elements into different elements, that is, if x z' implies that f(x) f (z'). One-to-one functions are also called injective functions or injections. Visually, no horizontal line can intersect the graph of a one-to-one function more than once. For instance, if A and B are both the set of real numbers, then f (x) = x and g(x) = x3 are both one-to-one, but h(x) = x2 is not. DEFINITION 1.2. Let f be a one-to-one function with domain A and range B. Then the inverse of f is the function f-1 : B - A given by f (y) - x if f (X) = y. EXAMPLE 1.2. Let A and B both be the set of all real numbers. Let f : A -- B be given by f(x) = 2x+ 7. Then f-1(y) (y - 7)/2. Solution: If f(x) = y, then y = 2x + 7, so y - 7 = 2x and so (y - 7)/2 = x. As x = f-i(y), it follows that f-1(y) = (y - 7)/2. Q The preceding example shows a general strategy for finding the inverse of a function. Write the equation f(x) = y, with the appropriate algebraic expression replacing f(x). Then solve for c. If there is more than one solution, then f is not one-to-one, and so it has no inverse function. If there is one solution, then that expression is the value of EXAMPLE 1.3. If A is the set of positive real numbers, B is the set of real numbers that are larger than 1, and f : A - B is given by f (X) = x2 + 1, then f-1(y) = y - 1. Solution: We have f(x) x2 + 1= y. So X2 = y - 1, and because we know that cc is positive and y > 1, we can take the square root of both sides, leading to cc =y - 1. Hence, fi-(y) = y - 1. D Note that the graphs of f and f-i are reflected images of each other through the line y =cc as illustrated in Figure 1.9.  16 1. FUNCTIONS Finally, we point out that if f is a one-to-one function with domain A and range B, then f o f- is the identity function of A and f-- o f is the identity function of B. Y/ 3 - /' - f(x) / -f (x) // // // // // // /x 0 1 2 3 FIGURE 1.9. f(x) and f 1(x) are symmetric about the identity function x. For instance, using the functions of Example 1.3, for all positive real numbers x, the identity (f o f-1)(x) = V(x2 + 1) - 1 = x2 = x holds, and for all y > 1, the identity (f--1 o f)(y) = ( y - 1)2 + 1 y - 1 + 1 = y holds. 5.1. Logarithmic Functions. If a function contains only additions, sub- tractions, multiplications, and divisions, then its inverse is often easy to compute. Power functions, that is, functions of the form f(x) = x0, where c is a real number, are not much more difficult. However, what is the inverse of an exponential function? Let f(x) = 2x. It is easy to see, by plotting the graph of f or otherwise, that f is a one-to-one function whose domain is the set of all real numbers and whose range is the set of all positive real numbers. So the inverse of f is a function from the set of positive reals to the set of all reals. But what is that inverse function f--1? By the definition of inverse functions in general, this is the function that sends 2x to x for all positive real numbers 2x. In particular, f -1(2) = 1, f -1(4) = 2, f -1(32) = 5, and f -1(1/2) = -1. That is, f -1(y) tells us to what power we have to raise 2 if the result is to be y. This important concept has its own name. DEFINITION 1.3. Let m be a positive real number. Then the inverse of the function f(x) = mx is called the logarithmic function with base m, and is denoted by logi.  5. INVERSE FUNCTIONS 17 So if f (x) = zm = y, then logm(y) = x. For instance, log2(64) = 6, log3(81) = 4, log5(1/25) = -2, and log0. 5(16) = -4. Logarithmic functions satisfy certain rules that are very similar to those satisfied by exponential functions and can, in fact, be deduced from them. These are (I) log(xy) = log x + log y. (II) log(x/y) = log x - log y. (III) log (za) = a logx. (IV) log / x lo.x (V) aloga x = x. (VI) loga(ax) = x. The last two rules simply express the fact that the functions f(x) ax and f-1(y)= loge(y) are inverses of each other, so their composition is an identity function. If we know the logarithm of a number in a base and want to compute it in another base, we can do so using the following theorem. THEOREM 1.1. For positive real numbers a, b, and x, we have _loge X logo x = t. loge a PROOF. Start with the identity Now take the logarithm of base b of both sides to get loge x= logo x logo a. Now divide both sides by logo a to get the identity of the theorem. Q EXAMPLE 1.4. We can use Theorem 1.1 to compute 1og16(256) from log2(256) as follows: lo 256) - log2(256) 8 log2(16) 4 So if a calculator or computer can provide the logarithm of all positive real numbers in one base, it can compute the logarithm of any positive real number in any base. For this reason, many calculators and computers are programmed to work primarily with logarithms of one given base, namely of base e, where e ~i 2.718 is an irrational number that will be formally defined in Chapter 2. The logarithm of base e is so important that it has its own name, natural logarithm, and its own notation, ln. So ln x =loge z.  18 1. FUNCTIONS 5.2. Inverses of Trigonometric Functions. Basic trigonometric functions, such as sin, cos, and tan, are very important in calculus, so it is no sur- prise that their inverse functions are important as well. However, we have to be precise when we define them since trigonometric functions are not one-to-one. In fact, they are periodical, of period 27 or 7, and so they take every value in their range infinitely often. In order to get around this difficulty, we will restrict our trigono- metric functions to just a short interval, in which they are one-to-one, and define their inverses based on that restriction. For instance, consider sin as a function whose domain is [-7/2, 7/2]. In that interval, sin is a one-to-one function (since it is increasing), and its range is the interval [-1, 1]. See Figure 1.10 for an illustration. So the inverse of sin [-7/2, 7/2] - [-1, 1] is the function sin-1 [-1, 1] - [-7/2, 7/2]. That is, if y E [-1, 1], then sin-1 y is the (only) x E [-7/2, 7/2] for which sin x = y. For instance, sin-1(1/2) = 7/6, while sin-1(0) = 0 and sin-1(v/2/2) = /4. Figure 1.11 shows the graph of sin-1 X. The inverses of the other trigonometric functions are defined simi- larly, just the intervals to which we restrict the functions (in order to make them one-to-one) can change. That is, cos-1 is the inverse function of the cos function that is restricted to the interval [0, 7]. So cos-1 is a function with domain [-1, 1] and range [0, 7]. Similarly, tan-1 is the inverse function of the tan function that is restricted to the interval (-7/2, 7/2). Its domain is the set of all real numbers, and its range is the interval (-7/2, 7/2). See Figure 1.12 for illustrations. The inverse functions of cot, sec, and csc, while not used often, can also be defined analogously. 5.3. Exercises. (1) Is there a function f defined on all positive real numbers for which f -1 = f? Y - 2w _ -i _-7 7\ 37T 2>7 FIGURE 1.10. sinx is one-to-one on the interval [-7/2, 7/2].  5. INVERSE FUNCTIONS 19 y x FIGURE 1.11. Graph of sin-1 x. i i i t i i / Il I 1 I1 /I / I FIGURE 1.12. Graphs of cos x and tan x with their inverses.  20 1. FUNCTIONS (2) If we are given logo x, how can we compute log/ax z? (3) For which values of a is logo an increasing function, and for which values of a is it a decreasing function? (4) What is the geometric connection between the graphs of f and f-1? (5) What is the value of tan-1 1? (6) Find all real numbers y such that tan-1 y =cot-1 y. (7) Is it true that if g is the inverse function of the one-to-one function f, then g is one-to-one? (8) Let f : R - R be defined by f (x) =Iz|. Is f a one-to-one function? (9) Let f R - R be defined by f(x) = z5. Is f a one-to-one function? (10) Let f : R+ - R+ be defined by f(x) = x2. Is f a one-to-one function? (11) Is f(x) = log cx a one-to-one function on the set of all positive real numbers? (12) Express x in terms of y if loga(loga z) = y. (13) Let us assume that f : R - R is a strictly increasing function, that is, if x < y, then f (x) < f(y). Can we conclude that f : R - R is a one-to-one function? (14) Let f R -- R be a function that has an inverse function f- : R - R. If f is strictly increasing, can we conclude that f-1 is strictly increasing? (15) The hyperbolic sine function is defined by sinh z = (ex -e-x)/2. Prove that sinh-1 y = ln(y + y2 + 1) for all real numbers y. (16) The hyperbolic cosine function is defined by cosh xc= (ex + e-x)/2. Prove that for all real numbers x, the inequality cosh x ;> 1 holds. (17) Prove that cosh-1 y = ln(y + y2 - 1) for all real numbers y>1. (18) Is cosh : R+ - R+ a one-to-one function? (19) The hyperbolic tangent function is defined by tanh cc= sinh c/cosh cc. Prove that for all real numbers cc, the inequality |tanh cc < 1 holds. (20) Prove that if ly| < 1, then tanh-1 y =jl  6. THE VELOCITY PROBLEM AND THE TANGENT PROBLEM 21 6. The Velocity Problem and the Tangent Problem 6.1. The Velocity Problem. Let us assume that a car was on the road from 3:00 p.m. to 5:00 p.m. on a given afternoon, and it traveled a distance of 100 miles, all due west. From the data, it is easy to compute the average speed of the car by the formula s (1.1) o = - t where t is the time passed, s is the distance covered in time t, and v is the average speed for the given time period. In physics, when the direction in which an object is moving is taken into account, we talk about velocity instead of speed, hence the abbreviation v. In the given example, all travel was in one direction (west), so there is no danger of confusion, and we can use either word. Let us assume that time is measured in hours and distance is measure in miles. Then Equation (1.1) yields 100 mi mi v= 2hr =50h, so the average velocity of the car for the given 2-hour period is 50 miles per hour. The car probably did not cover the entire distance at its average velocity. For various traffic-related or other reasons, it sometimes may have gone faster or slower. If we want to know its average velocity for the time period between 4:00 p.m. and 4:10 p.m., then we need know the distance it covered in that time period. If that distance is 10 miles, then we conclude that in that 10-minute time period, the average velocity of the car was 10mi mi v= 60 . 1/6 hr hr If we want more precise information, like the average velocity of the car between 4:02 p.m. and 4:05 p.m., we can proceed similarly, decreasing the value of both the numerator and the denominator of the fraction s/t. However, what if we want to know the instantaneous velocity of the car in a given moment, such as exactly at 4:02:23 p.m. (and not in the second that passed between 4:02:23 p.m. and 4:02:24 p.m.)? In that case, a direct application of Equation (1.1) is impossible, because the denominator t is equal to 0. The numerator s is also equal to 0, since the car needs time to cover any distance; if it is given no time, it will cover no distance. In this section, we will not give a completely formal answer to the problem of defining instantaneous velocity; we will leave that task to  22 1. FUNCTIONS an upcoming section. However, we will say the following. The instan- taneous velocity of a car in a given moment m can be approximated by choosing smaller and smaller time periods containing m and computing the average speed of the car for those time periods. These averages will approximate the instantaneous velocity. 6.2. The Tangent Problem. The problem of finding the instantaneous velocity of a moving object is simply a special case of a much more general problem, that of finding the slope of a tangent line to a curve at a given point. In the previous problem, the distance the car covered can be viewed as a function of the time that passed since the car started moving. So s(t) is the distance covered from the moment when the car started moving to the moment t hours later. In order to compute the average velocity for the time period from ti to t2, we simply compute the value of the fraction s(t2) - s(ti) t2-t l This fraction is precisely the slope of the line that intersects the graph of the function s at points (t1, s(ti)) and (t2, s(t2)). If we choose ti and t2 closer and closer together, then these points will get closer and closer together as well. Finally, if we set ti1= t2, then we will not immediately know the slope of the line that touches the graph of s at the point (t1, s(ti)) since we will know only one, not two, point of this line. However, and this will be made more precise in the next section, the slope we are looking for will be approximated by the sequence of slopes of the lines that we got when we chose ti and t2 closer and closer together. Finally, we point out that there is nothing magical about the func- tion s(t) here. We could consider any function f: R - R, and ask what the slope of the tangent line to this curve is at the point (x, f(x)). 6.3. Exercises. (1) A car travels 1 hour at a speed of 60 miles per hour, then 2 hours at a speed of 45 miles per hour. What is the average speed of the car during this 3-hour period? (2) Consider the car of the previous exercise. What is its average speed during the first 2 hours of its trip? (3) I drove at 40 miles per hour for 2 hours. How fast do I have to drive in my third hour if I want to reach an average speed of 45 miles per hour for my 3-hour drive?  6. THE VELOCITY PROBLEM AND THE TANGENT PROBLEM 23 (4) A car travels 300 miles on a given day. During the first 100 miles, the car travels at a speed of 40 miles per hour, during the second 100 miles, it travels at a speed of 50 miles per hour, and during the third 100 miles, it travels at a speed of 60 miles per hour. What is the average speed of the car for the entire 300-mile trip? (5) Tim has ridden his bicycle to school, covering a 5-mile distance in half an hour. Can we conclude that there was a segment of his ride for which his average speed was more than 10 miles per hour? (6) Jim has driven his car for 3 days in different conditions. On the first day, he was able to drive 20 miles per gallon of fuel used, for a total of 200 miles. On the second day, he drove 25 miles per gallon of fuel used, for a total of 275 miles. On the third day, he drove 24 miles per gallon of fuel used, for a total of 240 miles. What was his average number of miles driven per gallon of fuel for the entire 3-day trip? (7) A ball is thrown vertically in the air. In t seconds, its height (in meters) is given by the function h(t) = 50t - 20t2. What is the average velocity of the ball during its first 2 seconds of motion? (8) Consider the ball of the previous exercise. What is the average velocity of the ball between t = 1 and t = 2? (9) Consider the ball of the previous exercise. What is the average velocity of the ball between t = 1.4 and t = 1.6? (10) Give a reasonable estimate of the velocity of the ball of the previous exercise in the moment t = 1.5. (11) Consider the function f(x) = x2. Can you find two points P and Q on the graph of f such that the slope of the line PQ is between 0 and 0.01? (12) Let fv(x) =_V/c and let P = (1, 1). Find the slope of the three lines that connect P to the points (4, 2), (2.25, 1.5), and (1.44, 1.2). (13) Let f be as in the previous exercise. Find the slope of the three lines connecting P = (1, 1) to the points (0.25, 0.5), (0.64, 0.8), and (0.81, 0.9). (14) Consider the results of the two preceding exercises. Do you see a trend? (15) Let g~x) =ex and let P =(0, 1). Find the slope of the three lines connecting P to the points (-1, e-1), (1, e), and (ln 2, 2).  24 1. FUNCTIONS (16) Consider the function f(x) = x2. Let P = (1, 1). Can you find a point Q on the graph of g such that the slope of the line PQ is 2? (17) Consider the function g(x) = x3. Let P = (1, 1). Can you find a point Q on the graph of g such that the slope of the line PQ is between 1 and 1.01? (18) Consider the function f(x) = 1/z. Choose two points P and Q of the graph of f such that P / Q and the c coordinates of P and Q are small and positive. What can be said about the slope of the line PQ? (19) Consider the function f(x) = ex. Can you find two points on the graph of f such that the slope of the line PQ is negative? Explain your answer. (20) Consider function f(x) = 1-ln cc. Can you find two points on the graph of f such that the slope of the line PQ is positive? Explain your answer.  CHAPTER 2 Limits and Derivatives 7. The Limit of a Function 7.1. Two-Sided Limits. Consider the function given by the rule f(x) 1/(1 + z). Let us compute the values of f(x) for various real numbers x that are close to 0. We find that Sf (1) = 1/2, Sf (1/2) = 2/3, * f (1/3) = 3/4, and, in general, e f (1/n) = n/(n + 1). Similarly, for negative values of x, we get Sf (-1/2) = 2, Sf (-1/3) = 3/2, " f(-1/4) = 4/3, and, in general, Sf (-1/n) - n/(n - 1). What we see is that if x gets close to 0 (from either side), then f(x) gets close to f(0) = 1. In fact, we can get f (x) to be as close to f(0) = 1 as we want; all we need to do is to choose x sufficiently close to 0. Indeed, looking at the previous examples, we conclude that if 0 < x < 1/n, then n/(n + 1) < f (x) <1, and if -1/n < zx <0, then 1 < f (x) < n/(n - 1). So for instance, if we want f(x) to be closer than a to 1, then any choice of x in the interval [0, 9) or any choice of x in the interval (-o, 0] will work. That is, any choice of x in the interval (-A 9, ) will imply that f(x) - f(0)| < 0.001. This phenomenon, that is, the fact that there exists an interval such that, for each real number in that interval, the value of f(x) is closer to f(0) than a prescribed bound is so important in mathematics that it has its own name. DEFINITION 2.1. Let f : R - R be a function and let a be a real number. We say that the limit of f in a is the real number L if the values of f(x) get arbitrarily close to L and stay arbitrarily close to L when x is suitably close to a without being equal to a. The fact that the limit of f in a is L is expressed by the notation lim f(xc) =L. 25  26 2. LIMITS AND DERIVATIVES So, if f is the starting example of this section, then lim2-o f (x) = 1. Note that the definition of limx-a f(x) requires that f(x) stay close to L when x is close to a, regardless of which of x or a is larger. That is, f(x) has to be close to L if x is a little bit less than a, and f(x) has to be close to L if x is a little bit more than a, though f(x) does not have to be close to L if x = a. Several comments are in order. First, limx-a g(x) does not always exist. EXAMPLE 2.1. Let _ 1 if 0< , Then the limit of g at a = 0 does not exist. Indeed, no matter how small an interval I we take around the point a = 0, that interval I will contain some positive and some negative real numbers. Hence, the values of g(x) will sometimes equal 1 and sometimes equal 0 for x E I, no matter how small I is. There is no number L such that 0 and 1 are arbitrarily close to it-in fact there is no number such that both 0 and 1 are both closer than 0.5 to it. So limx-o g(x) does not exist. Second, if limx-o f(x) exists, it is unique; that is, f cannot have two different limits at any given point a. Let us illustrate this using the introductory example of this section, the function f (x) = 1/(1+x). We have seen that limx-o f(x) = 1. Indeed, we saw that the values of f(x) can get arbitrarily close to 1 if the real numbers x are chosen from a suitably small interval around 0. At this point, one could ask the following question. If 1 satisfies the requirements to be the limx-o fv(x), why does 1.0001 not? After all, what is close to 1 is also close to 1.0001. In order to answer this question, we must have a good understand- ing of the definition of limits. That definition says that if limx-o f (x) L, then the values of f(x) will get arbitrarily close to f(0) if x is chosen from a suitably small interval around 0. The key word in the previous sentence is arbitrarily. While 1.0001 is close to 1, it is not arbitrarily close to 1; it is exactly 0.0001 away. And that is a problem, since we have seen at the beginning of this chapter that, as x approaches 0, the values of f(x) will get arbitrarily close to 1. In particular, if x is close enough to 0, then f(x) will be closer than los to 1, but then it cannot also be closer than 16to 1.0001. An analogous argument shows that no function can have two dif- ferent limits at any one point. Sometimes it can happen that h is not even defined in a, but limo-o h~x) still exists. Note that the fact that h(a) is not defined  7. THE LIMIT OF A FUNCTION 27 8 6 5 -3 -2 -1 0 1 2 3 4 5 FIGURE 2.1. h(x) _ X29 x-3 is not a problem since the definition of limits specifically states that x should not be equal to a anyway. EXAMPLE 2.2. Let h(x) all real numbers except x = lim,-3 h(x) exists. (x2 - 9)/(x - 3) 3. Still, limx,3 h(x) Then h is defined for = 6. In particular, See Figure 2.1 for an illustration. Solution: If x f 3, then 2 f(x) =x (x + 3)(x - 3) x-3 x+3. So if we want f(x) = x + 3 to be closer to 6 than a given distance a, then all we have to do is to choose x such that x - 3 < a. D At this point, the reader should test his or her understanding of the material by finding lime--2((x2 + 3x + 2)/(x + 2)). Sometimes, limits are not easy to determine. Plotting the graph of the function h(x) (sin x)/x, we are led to believe that lim sin x x--0 X See Figure 2.2 for an illustration. 1.  28 However, we have not yet learned the techniques to rigorously prove this. Plotting the graph of the function or producing more numerical data should not be considered as a complete answer, since, as x approaches 0, eventually z and sin x will get so small that the computer will no longer manipulate them, or their ratio, accurately. MAIN RAD AUTu FUNC FIGURE 2.2. Viewing (sinxz)/x on the TI-89 graphing calculator with viewing window [-5, 51 x [-0.5, 1.51. Finally, we point out that in the definition of the limit, the require- ment that f~x) get close to L arnd stay close to L is important. Consider the function f(x) =sin(1/z) around x 0. As x approaches 0, the value of 1/x will increase very fast, and so it will equal a multiple of wr many times. All those times, f(0) =0 will hold, so f~x) will be as close to 0 as possible. However, lim o f~x) does not exist, since f~x) will take all other values in the interval [-1, 1] infinitely often as well as x approaches 0. So the value of f~x) will riot stay arbitrarily close to 0, no matter how close x is to 0. See Figure 2.3 for an illustration. FIGURE 2.3. f(x) =sin(1/x).  7. THE LIMIT OF A FUNCTION 29 7.2. The Precise Definition of Limits. It is time for us to give a precise mathematical definition of limits. The advantage of this formal defini- tion is that we can finally do away with the words arbitrarily close and sufficiently close. The price to pay for that is that we have to use more notation. DEFINITION 2.2. Let f be a function defined on some open interval that contains the real number a, with the possible exception of a itself. Then we say that the limit of f at a is L, denoted by limxa f(x) = L, if, for all c > 0, there exists 6 > 0 such that if x - a < 6, then |f(x) - L < E. See Figure 2.4 for an illustration. 2 3 a 5 -5 FIGURE 2.4. As x approaches a, f(x) approaches L. EXAMPLE 2.3. We have limxso 2x sin x = 0. Solution: know that |f (x) - 0 Let c be any positive sin x < 1 for all x. Sf (x) = 2x sin x < real number. So if x - 0 2x <26=E, Then let 6 = < = as required. : E/2. We c/2, then D 7.3. One-Sided Limits. There are functions that behave in a certain way up to a point a, and then behave very differently after that. We have seen such a function in Example 2.1. The function g of that example satisfied g(x) = 0 for negative values of x, and g(x) = 1 for positive values of x. We have seen that limxso g(x) does not exist, since no real number L is arbitrarily close to both 0 and 1. Nevertheless, there are weaker, one-sided notions of limits that are relevant in this example.  30 2. LIMITS AND DERIVATIVES DEFINITION 2.3. Let f : R -- R be a function and let a be a real number. We say that the left-hand limit of f in a is the real number L if the values of f(x) get arbitrarily close to L and stay arbitrarily close to L when x is suitably close to a and x < a. The fact that L is the left-hand limit of f in a is denoted by lim f (x) = L. x-~a For instance, if g is the function defined in Example 2.1, then lim g(x) = 0. Indeed, if we choose x close to 0 but less than 0, then g(x) = 0, so g(x) is arbitrarily close (in fact, equal) to 0. DEFINITION 2.4. Let f : R -- R be a function and let a be a real number. We say that the right-hand limit of f in a is the real number L if the values of f(x) get arbitrarily close to L and stay arbitrarily close to L when x is suitably close to a and x > a. The fact that L is the right-hand limit of f in a is denoted by lim f (x) = L. For instance, if g is the function defined in Example 2.1, then lim g(x) = 1. Indeed, if we choose x close to 0 but more than 0, then g(x) = 1, so g(x) is arbitrarily close (in fact, equal) to 1. At this point, the reader should compare the definitions of limit, left-hand limit, and right-hand limit. The definition of limit (Defini- tion 2.1) imposes the strongest requirements on the values of f. Indeed, the values of f(x) have to be close to L when x is close to a and x < a and also when x is close to a and x > a. The definitions of the left- hand and right-hand limits impose weaker requirements in that each definition only requires that f(x) be close to L when x is on a given side of a and close to a. It then follows-and the reader should spend a minute verifying it that if limx-a f(x) = L, then limx-a- f(x) =L and limx-a+ f(x) = L. Conversely, if both the left-hand limit and the right-hand limit of f in a is equal to L, then the limit of f in a exists and is equal to L. At this point, the reader should check his or her understanding of the material by considering the function h x) hIz)  7. THE LIMIT OF A FUNCTION 31 as x approaches 0 and deciding if the limits limxso h(x), limx-o- h(x), and limxso+ h(x), exist. It may help to consult Figure 2.5. 2- -2 - FIGURE 2.5. Graph of h(x) = x/lxl. 7.4. Infinite Limits. In our definitions of limits in this section, the limit L was always a real number. In this section, we extend those definitions to the cases of infinite limits. If L = o, then the values of f have to get arbitrarily close to oc; that is, they have to get as large as we want. This is the content of the following definition. DEFINITION 2.5. Let f : R -~ R be a function. We say that the limit of f in a is oc if we can get f(x) arbitrarily large and keep it arbitrarily large if we choose x suitably close to a without being equal to a. Similarly, if g : R -~ R is a function, we say that the limit of g in a is -0o if we can make g(x) a negative number with an arbitrarily large absolute value and keep g(x) that way if we choose x suitably close to a without being equal to a. The fact that the limit of f in a is oc is denoted by lim f (x) = x. x-*+a EXAMPLE 2.4. Let f (x) = 1/x2. Then limxso f (x) = 0. Solution: If we want f(x) to be larger than an arbitrary positive real number N, all we need to do is to choose x from the interval (- 1/N, 1/N). Then x2 < 1/N will hold, implying that f(x) 1/x2 > N. Similarly, if g(x) = -1i/z, then limxso g(x) = -o. Note that if the limit of a function at a given point a is o0 or -o, then, as x approaches a, the graph of the function will approach a vertical line intersecting the horizontal axis at x = a. This phenomenon is referred to by saying that f has a vertical asymptote at a.  32 2. LIMITS AND DERIVATIVES 7.4.1. The Precise Definition of Infinite Limits. The formal definition of infinite limits is similar to that of finite limits. The difference lies in the fact that it is not the same to be close to 00 or to be close to a real number. DEFINITION 2.6. Let f :R - R be a function. We say that the limit of f in a is oo if, for all positive real numbers N, there exists e> 0 such that if x - a| N. Similarly, let g : R -- R be a function. We say that the limit of g in a is -oo if for all negative real numbers M, there exists c6> 0 such that if |x - a| a. EXAMPLE 2.5. Let f (x) = 1/x. Then f is not defined in 0. Fur- thermore, limxo- -oc and lim-,o+ = oo. As the two one-sided limits are different, limxo does not exist. Solution: We can make number M by choosing x x larger than any positive (0, P). f (x) 1 smaller than any given negative from the interval (1/M, 0). We can make number P by choosing x from the interval D- 7.5. Exercises. (1) Find lim-3 x + 7. (2) Find lim-13 x2 - 10x + 7. (3) Find limx-3 x234x+3 (4) Find lim-1 x2+x-2 (5) Does limx-3 3 24x+7 exist? (6) Find lim-o cos x. (7) Find lim-o x. (8) Find limo-1 x. (9) Find limx-2 x_ 8  8. LIMIT LAWS 33 (10) Find lim 2-2 x3+8 (11) Let f(x) =ILx_ be equal to the largest integer that is at most as large as x. So f(3.7) = 3. Note that f is often called the floor function or integer part function. Find the values a for which limx-a f(x) exists. If a is such that f has no two-sided limit at a, decide if f has one-sided limits at a. (12) Let g(x) FxIl be equal to the smallest integer that is at least as large as x. So g(3.7) = 4. Note that g is often called the ceiling function. Find the values a for which limx-,ag(x) exists. If a is such that g has no two-sided limit at a, decide if g has one-sided limits at a. (13) Does limx-72 tan x exist? (14) Does limx-o- cot x exist? (15) Does limx-o+ cot x exist? (16) Does limx-o I exist? (17) Give an example of a function f such that limx-o- f(x) = 0, and limx-o+ f (x) = 00. (18) Does limx-o ( + i) exist? (19) Does limx-o () + ) exist? (20) Give an example of a function f such that limx-1- f(x) = 00, limx-1+ f(x) = -oC, and f (1) is a real number. 8. Limit Laws 8.1. Basic Limit Laws. If f and g are two functions and we know the limit of each of them at a given point a, then we can easily compute the limit at a of their sum, difference, product, constant multiple, and quotient. The rules that provide this limit are given below, and they are very similar to the ways in which the sum, difference, product, constant multiple, and quotient of two functions are defined. Indeed, (I) lim(f + g)(X) = lim f (x) + lim g(x), x-- a x-- a x-- a (II) lim(f - g)(x) = lim f (x) - lim g(x), x-- a x-- a x-- a (III) lim(f - g)(x) =lim f (x) - lim g~x), (IV) lim(c - f)(x) =c - lim f (x), where c is a real number, and  34 2. LIMITS AND DERIVATIVES (V) . f limxa f (x) fm - (x) = x--a (g) limx- a g(x) if limx--a g(x) / 0. It is not difficult to believe that these rules are valid. For instance, if f(x) gets arbitrarily close to L as x approaches a and g(x) gets arbitrarily close to L' as x approaches a, then, as x approaches a, the value of fv(x) +g(x), that is, the value of (f + g) (x), will get arbitrarily close to L + L'. This intuitive argument can be made formal using the precise definition of limits. EXAMPLE 2.6. Let f(x) = |x| and let g(x) = x2. Find the limits of f + g, f - g, f g, 3f + 2g, and f/g at a = 2. Solution: Based on the five limit laws given earlier, it makes sense to first compute the limits of f and g at 2. The reader is invited to verify that lim f (x) = lim Iz| = lim x = 2, x-- 2 x- 2 x- 2 and lim g(x) = lim x2 = lim x -lim x = 2 - 2 = 4, x->2 x-- 2 x-- 2 x->2 where we used the fact that g(x) = x2 = x - x, so law III can be applied to compute the limit of g at 2. Now it is simply a matter of basic algebra to compute the five limits that we have been asked to find. Indeed, applying the five limit laws, we get that (I) limx--2(f + g)(X) = limx--2 f(x) + limx--2 g(x) = 2 + 4 = 6, (II) lime-2(f - g)(z) = limx--2 f(x) - lime-,2 g(x) = 2 - 4 = -2, (III) limx-2(f - g)(x) = limx--2 f(x) - limx--2 g(x) = 2 . 4= 8, (IV) limx--2(3f + 2g)(x) = 3limx--2 f (x) + 2limx-,2 g(x) = 3 - 2 + 2 - 4 = 14 (note that here we applied limit law IV to first f, then to g, and then we applied law I to 3f and 2g), and (V) Slim f (x) 2 1 1-2 g limg(x) 4 2 8.2. Frequently Used Special Cases of Limit Laws. A few special cases of limit laws I-V are used so frequently that it is worth mentioning them separately. First, if we repeatedly multiply a function by itself,  8. LIMIT LAWS 35 we get a power of that function. Applying law III each time, we get that for all positive integers n, (2.1) lim(f(x))" =lim f (x)). Note that we have essentially applied this rule in the special case of n = 2 when we computed limx-2 x2 in Example 2.6. The reader is invited to verify that the limits of the constant func- tion f(x) = c and the identity function f(x) = x are given by limx-a c = c for all a and limx-a zX= a. Formal proofs will be given in the next section. Applying Equation (2.1) to the identity function f(x) = x yields the equation (2.2) lim x" =a". x-- a It turns out (though it is not obvious) that in Equation (2.1) the exponent n can be replaced by 1/n; in other words, powers can be replaced by roots, yielding (2.3) lim ffx) n lim f x). x-- a x- (Here f(x) has to be nonnegative if n is even.) So, in particular, if f (x) = x, then lim /x = na. x-- a 8.3. Other Useful Facts About Limits. In this section, we discuss a few facts about limits that are often used to compute limits, but are slightly different in nature from the limit laws we discussed so far. First, let us recall that the definition of L= limx-a f(x) requires that f(x) get arbitrarily close to L if x is sufficiently close to a but not equal to a. That is, the value of f(a) does not have to satisfy any requirements. In fact, we can change f(a) to anything we want, and L = limx-a f(x) will not change. What matters is what happens at points other than a. Hence, we can conclude that if f(x) = g(x) for all points x / a, then limx-a f (x) = limx-a g(x) as long as these limits exist. For instance, let f(x) =_(x2 - 4)/(x - 2) for all real numbers x / 2 and let f (2) = 2014. Let g(x) = x+2 for all real numbers. Then f (xc) =g~ix) unless x = 2, and hence limx-a f (x) =limx-a g~x) =4. The statement that if f~x) =g~ix) for all points cc / a, then limxsa f(xc) =limxsa g(xc) as long as these limits exist can be sig- nificantly strengthened. See Exercise 8.4.1 for a possible direction for that.  36 2. LIMITS AND DERIVATIVES Second, Equation (2.2) can be interpreted by saying that the limit of a power function f(x) = x" at any point a is simply the value of f (a). Now note that polynomials are nothing else but sums of con- stant multiples of power functions with nonnegative integer exponents. Hence, using limit laws I and IV, we get the following theorem. THEOREM 2.1. Let p be a polynomial function. Then, for any real number a, we have lim p(x) = p(a). x-~a Now recall that a rational function is just the ratio of two polyno- mials. Hence, using limit law V, we get the following statement from Theorem 2.1. COROLLARY 2.1. Let R(x) be a rational function and let a be a real number such that R(a) is defined. Then lim R(x) = R(a). x-- a PROOF. If R(x) = p(x)/q(x), where p and q are polynomials, then by first applying limit law V, and then Theorem 2.1, we get p(x) _ limx-a p(x) p(a) lim R(x) = lim = = = R(a). x-a x-a q(x) limx-a q(x) q(a) So far all the relationships that we discussed for limits involved equations. We will now discuss two rules that, involve inequalities. THEOREM 2.2. Let f and g be two functions and assume that, for all real numbers x, the inequality f (z) g(x) holds. Then (2.4) lim f(x) < lim g(x) x--a x- a for any real number a as long as both limits exist. PROOF. If (2.4) did not hold, then Lf = lim f (x) = D + lim g(x) = D + Lg x-- a x-- a would hold, for some positive real number D. That would lead to a contradiction, since if x is so close to a that |f(x) - Lf| < (D/3), then, in particular, f(x) > Lf - (D/3), so D 2D 3 3 This inequality says that no matter how close x is to a, the distance between g~x) at Lg is more than 2D/3. This contradicts the definition  8. LIMIT LAWS 37 of Lg, since if Lg exists, then the values of g(x) should get arbitrarily close to it, provided that x is sufficiently close to a. D Note that in Theorem 2.2, the fact that the inequalities are not strict is important. See Exercise 8.4.7 for a relevant question. COROLLARY 2.2 (Squeeze Principle). If f, g, and h are functions such that, for all real numbers x, the inequality f(x) < g(x) < h(x) holds and lim f (x) = lim h(x) = L, x-*+a x-*+a then limx-a g(x) exists and limxa g(x) = L. See Figure 2.6 for an illustration of this important principle. PROOF. If limo-a g(x) exists, then by applying Theorem 2.2 to f and g, it follows that L < limsa g(x), and by applying Theorem 2.2 to g and h, it follows that limxoa g(x) < L. So if limxoa g(x) exists, it is equal to L. In Exercise 8.4.3 you are asked to prove that this limit exists. D The squeeze principle is very useful since it allows us to compute the limits of rather complicated functions as long as we can squeeze them between two functions with identical limits. h(x) - g(x) f(x) FIGURE 2.6. Concept of squeeze theorem where f(x) g(x) < h(x). EXAMPLE 2.7. Let g(x) x cos(log x). Then limxso g(x) 0.  38 2. LIMITS AND DERIVATIVES Solution: Indeed, let f(x) = -x and h(x) = x. Then, since cos(log z) is always a real number in the interval [-1, 1], the inequality f(x) < g(x) h(x) holds for all real numbers x. Furthermore, limx-o f(x) lim2-o h(x) = 0, so we can apply Corollary 2.2 to prove our claim. Q We could not have used limit law III to compute limx-o g(x) since limx-o cos(log x) does not exist. You are asked to prove this in Exercise 8.4.4. 8.4. Exercises. (1) Find limx-2 3x2 +44x + 9. (2) Find limx-3 3x2+X-2 (3) Find lim-2 x-42. (4) Find lim 4 2++5 -4x3+1 (5) Find limx-2 - (6) Let f(x) and g(x) be two functions that only differ for a finite number of values of the variable x. Is it true that limxoa f(x) = limx-a g(x) as long as these limits exist? Why or why not? (7) Find an example of two functions f and g such that f(x) < g(x) for all real numbers x, but there exists a real number a such that limx-a f (X) = limx-a g(x). (8) Explain why limx-a g(x) exists if the conditions of Corol- lary 2.2 hold. (9) Prove that limx-o cos(log c) does not exist. (10) Prove that limx-o I sin(cc)| 0. (11) Compute limx-o x3 sin(1/z). (12) Compute lim2-o v/4 + z5 sin(ln x). (13) Compute limx-o o . (14) Compute limx- X.+X (15) Compute lim-1 x+1. (16) Compute limx-io /18 - z. (17) Let a be a positive real number. Prove that lima2a - -- = x-a- (18) Compute lim2-3+ -I (19) Compute limo-4(xc + 4)2.5. Explain which laws you are using. (20) Compute lim-so (cc +-2)0.4. Explain which laws you are using.  9. CONTINUOUS FUNCTIONS 39 9. Continuous Functions Intuitively speaking, a function is called continuous at a point x = a if its graph in a neighborhood of x = a can be drawn without lifting the pencil from the paper, that is, by a "continuous" line. The formal definition of continuity is as follows. DEFINITION 2.8. A function f is called continuous at a if the equality lim f (x) = f (a) x-~a holds. Note that Definition 2.8 really requires three things. The limit of f at a must exist, the function f must be defined in a such that f(a) exists, and the value of f(a) must agree with the limit of f at a. If all these conditions hold, then the behavior of f at a is very similar to the behavior of f around a; in particular, the graph of f can be drawn without lifting the pencil from the paper. If we had to lift the pencil from the paper, that would mean that some kind of "gap" would exist in the graph of f, so the requirements of Definition 2.8 would not be satisfied. If a function f : R -- R is continuous at all a E R, then it is called continuous. If f is continuous at each point of the open interval (c, d), then we say that f is continuous on (c, d). Finally, if you really want a formal definition, the neighborhood of a is a set S that contains an open interval (c, d) containing a. 9.0.1. The Precise Definition of Continuity. As the informal definition of continuity is very close to that of limits, it is not surprising that their precise definitions are also similar. DEFINITION 2.9. Let f be defined in an open interval containing a. We say that f is continuous in a if, for all E > 0, there exists b > 0 such that if |x-a <6, then f(x) - f (a)| 0 and g(x) = 0 if x < 0. As we have seen before, the limit of this function does not exist in a = 0, even if g(0) is defined. So g is not continuous at 0. Finally, it could happen that h is defined in a and the limit of h at a exists, but h(a) is not equal to this limit. That happens, for example, if h(x) = (x + 3)/(x2 - 9) if Iz| / 3 and h(x) = 1 if Iz|j= 3. Let a =-3. Then 1 x- a 6 The interested reader is invited to think about the following example. EXcuRSION 2.1. The following function is not continuous any- where. Let f~x) =1 if xc is rational and let f(xc) =0 if xc is irrational. 9.3. New Continuous Functions from Old. It follows from the limit laws that several transformations preserve the continuous property of func- tions.  9. CONTINUOUS FUNCTIONS 41 THEOREM 2.3. Let f and g be two functions that are continuous at a and let c be a real number. Then all of the following are also continuous functions at a: (I) f + g, (II) f - g, (III) f - g, (IV) cf, and (V) f /g as long as g(a) / 0. EXAMPLE 2.10. It follows from successive applications of the pre- vious theorem that h(I) = ex - sin x + 3 In x - 9/x is continuous at all positive real numbers a. The following important theorem also holds, though it is not a direct consequence of our limit laws. THEOREM 2.4. Let f and g be two functions such that f is contin- uous at a and g is continuous at f(a). Then the composition function f o g is continuous at a. This theorem is important since it enables us to prove the continuity of functions that would otherwise be cumbersome to handle. EXAMPLE 2.11. The function h(x) = 2 + sin x is continuous at all real numbers a. Solution: Let f (x) = 2 + sin x and let g(x) =xV . Then f is contin- uous everywhere, and g is continuous at all positive real numbers. As f (x) is always a positive real number, the statement follows. Q 9.4. One-Sided Continuity. A function may happen to be continuous in only one direction, either from the "left" or from the "right." Formally, this means the following. DEFINITION 2.10. We say that the function f is left-continuous at a if f (a) = limxa- f (x). Similarly, we say that f is right-continuous at a if f (a) = limxa+ f (x). EXAMPLE 2.12. Let g be the function defined by g(x) = 1 if x ;> 0 and gQ(x) = 0 if x < 0. Then lim2-o- g(x) = 0 / 1 =g(0), so g is not left-continuous at 0. On the other hand, lim2-0+ g(Ic) =1 =g(0), so g is right-continuous at 0. The reader is invited to verify that f is continuous at a if and only if f is both left-continuous and right-continuous at a.  42 2. LIMITS AND DERIVATIVES We say that a function is continuous on an interval [a, b] if it is con- tinuous at all points of (a, b), left-continuous at a, and right-continuous at b. 9.5. Intermediate Value Theorem. Perhaps the most important property of continuous functions is that they do not skip any values between two values that they actually take. For instance, if a tree grows from 3 feet to 6 feet, then there is a time in between when the tree is exactly 4.47 feet tall. The intuitive reason for this is that if there were a value in between that is not taken by the function, then there would be a gap in the graph of the function, contradicting the requirement that the function be continuous. This is the content of the next theorem. THEOREM 2.5 (Intermediate Value Theorem). Let f be a function that is continuous on the interval [a, b]. Then, if f(a) = yi and f(b) - Y2 and y is a real number that is between Yi and Y2, then there exists x E [a, b] such that f(x) = y. In other words, f takes all values between Yi and Y2 on the interval [a,b]. EXAMPLE 2.13. There is a real number x in the interval [0, 1] such that x + ex = 2. Solution: Let f (x) = x+ex. Then f is continuous everywhere, f (0) 1, and f(1) = 1 + e > 3.71. So, by the intermediate value theorem, we get that f takes all values between 1 and 1 + e on that interval, including y = 2. D 9.6. Exercises. (1) Is e3x+7 sin x continuous everywhere? (2) Is (x2 + 1) ln(x + 1) continuous everywhere? (3) Is x3+222+3x+4 continuous everywhere? x +4 (4) In what point is V/z right-continuous, but not continuous? (5) Where is lln x continuous? (6) Where is x2+2x+37 continuous? (7) Where is tan x continuous? (8) Where is 1/x not continuous? (9) Is there a point in which x-2 is left-continuous, but not con- tinuous? (10) Where is 1 continuous? (11) Where is 54X"+ continuous? (12) Where is sin(x2) continuous?  10. LIMITS AT INFINITY 43 (13) Let f(x) = x2 + 3 if x / 2. What should f(2) be if f is to be a continuous function? (14) Let f(x) = (x2 - 16)/(x - 4) if x / 4. What should f(4) be if f is to be a continuous function? (15) Let f(x) L=Iz_. Determine the set of points a for which f is continuous at a. What can be said about f at the points where f is not continuous? (16) Let g(x) =FIz|. Determine the set of points a for which g is continuous at a. What can be said about g at the points where g is not continuous? (17) Prove that the equation z5-x-1 = 0 has a root in the interval (-1,2). (18) Prove that the equation x3 - 3x - 1 = 0 has at least two roots in the interval (-1, 2). (19) Prove that the equation x4+ c =_v/2 has at least one solution in the interval (0, 1). (20) Define a function f R -- R that is not continuous in any point a, and f(x) < x holds for all c E R. 10. Limits at Infinity 10.1. Finite Limits at Infinity. In Section 7, we defined what it meant for a function to have a limit L at a real number a. In this section, we extend that definition and define what it means for a function to have a limit L at o0 or at -oo. DEFINITION 2.11. Let f : R - R be a function that is defined on some interval (b, oc). We say that the limit of f at oc is the real number L if the values of f (z) get arbitrarily close to L and stay arbitrarily close to L when x is suitably large. The fact that the limit of f at oc is L is expressed by the notation lim f (x) = L. x-~oo This definition follows the idea of the definition of limits at finite points. Indeed, in order for lim- f(x) = L to hold, we require that the values of f(x) get arbitrarily close to L and stay arbitrarily close to L if x is large enough. Here "c is large enough" means that x is in a suitably selected neighborhood of oc, in other words, in an open interval (c, oo). Recall that this is analogous to what we required in the finite case. There we said that limxa f(cc) =L if f(cc) got arbitrarily close to L and stayed arbitrarily close to L once cc was suitably close to a, that is, when cc was in a suitably selected neighborhood of a.  44 2. LIMITS AND DERIVATIVES EXAMPLE 2.14. Let f (x) = 1/z. Then lim f (x) = 0. Solution: If we want the value of f(x) to be closer than E to 0, all we have to do is to select x such that x > 1/E holds. Once x gets past 1/E, the values of f(x) will stay between 0 and E. D The definition of limits at -oo is what the reader probably expects. DEFINITION 2.12. Let f R -- R be a function defined on some interval (-oc, b). We say that the limit of f at -oo is the real number L if the values of f(x) get arbitrarily close to L and stay arbitrarily close to L when x is a negative number with a suitably large absolute value. The fact that the limit of f at -oo is L is expressed by the notation lim f (x) = L. EXAMPLE 2.15. Let f (x) = 1/c2. Then lim f (x) = 0. Solution: If we want to get f(x) closer than c to 0 and keep it there, it suffices to choose c such that x < -1/ /. Then x2 > 1/E, and hence f (x) = 1/c2 N, then f(x) - L| In M. D The following notation is defined in an analogous way: ()limx_'C f (X) = -0o. IIlimo-_'_ g(x) = 00. (III) limo-_, h(x) = -Co. Each of these definitions refers to a fact that the values of a function get arbitrarily far away from 0 and stay arbitrarily far away from 0 (in the appropriate direction) if x gets sufficiently far away from 0 (in the appropriate direction). The reader should test his or her understand- ing of these concepts by verifying that limx-21 - x = -oo, while limx_ x2 = oc, and lim _,cx3 --_ 10.2.1. The Formal Definition of Infinite Limits at Infinity. By now, the formal definition of infinite limits at infinity probably does not come as a surprise. We are providing a formal definition for one of the four possible scenarios that can occur due to changes in sign. The other three cases are analogous. DEFINITION 2.16. Let f : R -~ R be a function defined on some interval (b, oc). We say that lim-o f~x) =oc if, for all positive real numbers M, there exists a positive real number N such that if x > N; then f(x) > M.  46 2. LIMITS AND DERIVATIVES 10.3. Computing Limits at Infinity. The limit laws that we learned for limits at finite points stay true for limits at infinity as well, provided, of course, that they make sense. Here are a few examples. EXAMPLE 2.17. We have lim X + 3 1. x-o - 4 It would be wrong to argue as follows: "The numerator is the func- tion f(x) = x + 3, and the denominator is the function g(x) = x - 4. At oc, they both have limit oc, so, by the limit law for quotients, the limit of their quotient is 1." The problem with this argument is that oc is not a number. So oc/oc is not defined. It is possible for f and g both to have limit o0 at oc, and for f/g to have limits c at oc, for any given real number c. Indeed, let f (x) = cx and let g(x) = x. Instead, we can solve Example 2.17 as follows. Solution: lim x+3 lim (x-4)+7 x-oo x -4 x-oo x - 4 ( 7 =lim 1+ x-oo x - 4 =1+lim x-o x - 4 =1+0 =1. D We would like to point out other pitfalls when dealing with the application of limit laws and infinite limits. The following expressions are not defined: (I) oc+ (-oc) (II) o - 0 and -o - 0 (III) 1° and 1-- The following theorem is very useful when dealing with limits at oc. THEOREM 2.6. Let r be a positive rational number. Then 1 lim - =0. x-oo If r is an integer, then this statement follows from the fact that lim-o 1/z = 0 by applying limit law III (for products) r times. If r=p/q, where p and q are positive integers, then we can first prove the theorem for x" and then, using the root law, for zP/4 =qW  10. LIMITS AT INFINITY 47 Many limits can be computed with the help of this theorem. EXAMPLE 2.18. We have .ix2 +3x+ 1 him 3 = 0. Solution: We have x2+3x+1 x2 3x 1 33 3 and each of the three summands has limit 0 at oc by the preceding theorem. Hence, by the limit law for sums, so does their sum. Q Note that the limit would not change if we changed the denominator from x3 to x3 + 3x2+ 4x+5. This would have decreased the value of our function, but would have still kept it positive. Hence, by the squeeze principle, we can then conclude that lim x2+3x+1 cco 3 + 3cc2 +4cc+ 5 0. 10.4. Exercises. (1) Find limo-o 2+1 (2) Find limra 3x2+4x+1 00x2 +5 (3) Find lim - 2x2+6 (4) Find limra _ 3x2+4x+1 (5) Compute lim_ 0-x3_14. (6) Let R(x) = p(x)/q(x) be a rational function. Explain how lima-c R(x) depends on p(x) and q(x). (7) Compute lim- i + 2 2x3+1 (8) Compute lim s_ Sin. (9) Does lim-,_s f- exist? 10) Compute lim4 si " 4+cos2 4 11) Compute lim x2+1. 1 12) Compute lim4 42+1.V-21 ( ( ( (13) (14) (15) Compute lim x4+1 . x 02x2+11~ Compute lim 1- 2_x+4 Is there a real number L such that L holds? lim -2x3+1 x-mo° 1000x2 +9x+35  48 2. LIMITS AND DERIVATIVES (16) Compute lim_,,o .-0.1 + -0.9. (17) Compute lim_, x2+ 1X4 (18) Compute lim_, -x+SinX (19) Does lim_,o x sin x exist? (20) Give an example of a function f such that lim_,oc f(x) = 0, but lim_,oc(1/f(x)) does not exist. 11. Derivatives 11.1. Tangent Lines. Let us consider a function, such as f(x) = x2, and its graph. Let us choose a point on the graph, say the point P = (3, 9). Now let us look for the slope of the tangent line to the graph at that point. That is, consider a sequence of points P1, P2, ... that are all on the graph of f and are closer and closer to P. For each of these points, draw the line PP. The slope of these lines will approach a certain slope, and so the lines PP will approach a certain line. That line is called the tangent line of f at P. See Figure 2.7 for an illustration. DEFINITION 2.17. Let f be a function and let P = (a, f(a)) be a point on the graph of f. Then the tangent line to f at P is the line that contains P and has slope (2.5) lim f(x) - f(a) x-a x-a provided that this limit exists. -1 - 11 - 2 V - \ -1 -i- (a,f(a>> (a,f(a)) (x,f(x)) (x, f(x>> FIGURE 2.7. Notice that as x approaches a the secant line approaches the tangent line.  11. DERIVATIVES 49 The interactive website http://www.math.ufl.edu/~mathguy/ufcalc book/derivative_def.html provides further examples of this phenomenon. Note that in the preceding definition, (f(x) -f(a))/(x-a) is simply the slope of the line connecting the points P and (x, f(x)). EXAMPLE 2.19. In our running example, that is, when f(x) = x2 and P = (3, 9), the tangent line is the line that goes through P and has slope f (x) - f (3) X - 9 lim = him = lim(x + 3) = 6. x-3 x-3 ox-3 -3 lim-33 ) 11.2. Velocities. Recall that in Section 6, we mentioned that the av- erage velocity of a moving object, such as a car, can be computed by the rule v= s/t. That is, the average velocity is equal to the distance covered divided by the time needed to cover that distance. However, what can be said about the instantaneous velocity, that is, the velocity in a given moment? We could not answer that question in Section 6 since we did not have the tools to handle the fact that when only a given moment is considered, both the numerator and the denominator of the formula v = s/t are 0. Now that we have learned about limits, we can overcome that difficulty as follows. DEFINITION 2.18. Let f (t) be a function such that f (t) is the dis- tance covered by a moving object in t units of time. Then the instan- taneous velocity of the object a units of time after it starts moving is V (t) = lim , t a t-wa t - a provided that this limit exists. EXAMPLE 2.20. A car starts out by accelerating for 10 seconds so that the distance covered in the first t seconds is obtained (in meters) by the function f (t) = 2t2 if t < 10. What is the instantaneous velocity of the car after 4 seconds? Solution: By the definition of instantaneous velocity, we must compute _- f(t) - f(4) -t2 -16 t + 4 t-4 t-4A t-4 2(t -4) t-4 2 So, at the end of the fourth second (exactly 4 seconds after starting out), the car will move at a rate of 4 meters per second. D  50 2. LIMITS AND DERIVATIVES 11.3. The Derivative of a Function. The fact that the last two concepts, the tangent line and the instantaneous velocity, led to very similar definitions suggests that there is a very general principle at work and we have seen two special cases of that principle. This is indeed the case. DEFINITION 2.19. Let f be a function. The derivative of f at a is the limit f(a) = limf(x)f(a) x-ma x- a if this limit exists and is finite. So, in particular, f'(a) is the slope of the tangent line of f at a (unless that tangent line is vertical). Furthermore, the instantaneous velocity at time a is the derivative of the distance covered (as a function of the time t needed to cover that distance) at t = a. In other words, the derivative is a common generalization of the concepts of tangent line and instantaneous velocity. 11.4. Exercises. (1) Find the slope of the tangent line to the curve f(x) = 3x2 - 7 at the point (2, 5). (2) Find the slope of the tangent line to the curve f(x) = x3 at (3) Find the slope of the tangent line to the curve f(x) = x(1 - z) at x =1/2. (4) Find the slope of the tangent line to the curve f(x) = x2 at three different points. Do you see a pattern? (5) Find the slope of the tangent line to the curve f(x) = x2 + x at three different points. Do you see a pattern? (6) Show an example of a curve that does not have a tangent line at some point a because the limit defined in (2.5) does not exist or is infinite. (7) The distance covered by a car in a certain time period is de- scribed by the function t2(b - mn) f(t) =tm+ 2 2 where b and mn are positive constants. Let us assume that t E [0, 1]. Find the instantaneous velocity of the car at a given moment t =a. (8) A ball is rolling down a hill. The distance it covers in time t is given by the function s(t) =3t + 0.5t2, where t E [0, 51  12. THE DERIVATIVE AS A FUNCTION 51 and time is measured in seconds. What is the instantaneous velocity of the ball at the moment of time t = 3? (9) At the beginning of a daily training session, the distance cov- ered by a runner is described by the function s(t) = 0.5t2, where t E [0, 5], time is measured in seconds, and distance is measured in meters. At what moment will the runner have an instantaneous velocity of 6 m/s? (10) A car is moving at a speed of 20 meters per second when its driver applies the brakes and the car starts slowing down. The car stops 10 seconds later. The distance covered by the car in t seconds, starting at the moment when the driver steps on the brakes, is given by the function f(t) = 20t - t2 for t E [0, 10]. What is the velocity of the car t seconds after the brakes are applied? (11) Prove that, for any constant c, the derivative of the function f(x) = c is 0 at any point a. (12) Find the derivative of the function f (x) = x + 5 at a = 7. (13) Find the derivative of the function f(x) = 3x + 2 at a = 4. What happens if we change the value of a? (14) Find the derivative of the function f(x) = 3x - 11 at a = 4. Compare your result with the result of the previous exercise. (15) Find the derivative of the function f(x) = 2x2 at a = 2. (16) Find the derivative of the functions g(x) = 2x2 + 1 at a = 2 and h(x) =2(x - 1)2 at a = 3. (17) Find the derivative of the function f (x) = x3 at =1. (18) Find the derivative of the function f(x) =fIc at x = 4. (19) Let (2x if 0 0. (4) Compute f'(a) if f (x) = 4x + 1. (5) Let f (x) = x3. Prove that f"(x) = 6 for all real numbers x. (6) Let f(x) = x3. Compute f(4)(x). What can be said about higher-order derivatives of f? (7) Let f(x) = x4. Compute f'(a) at some point a. (8) Let f(x) = 1/x. Compute f'(a) at some point a / 0. (9) Let f(x) = 1/x2. Compute f'(a) at some point a / 0. (10) Let f and g be two functions such that f(x) - g(x) = c for all x, where c is a constant. Is it true that, at every point a where f'(a) exists, g'(a) also exists, and f'(a) =g'(a) holds? (11) Let f be defined on the interval [0, 2] by f(x) =v1 - x2 if the graph of f(x) is the union of two quarters of a unit circle. Prove that, at x =1, the graph of f has a vertical tangent  56 2. LIMITS AND DERIVATIVES line, that is, lim f (X) - f (1) C. xa - 1 (12) Find an example of a function f and a real number a such that f'(a) exists, but f"(a) does not exist. (13) Find an example of a function f and a point a such that f'(a) and f"(a) exist, but f"'(a) does not exist. In the remaining exercises of this section, decide whether the derivative of the given function in the given point exists or not. (14) f (x) = xz at a 0. (15) f (X) c= X2 at a 0. (16) f (x) = I[z_|at a=1. (17) f (x) = Iz] at a=2. (18) f (x) = lnc at a = 0. (19) f (x) = cotxat a = 0. (20) f (x) = 2x2 + x3 at any real number a.  CHAPTER 3 Rules of Differentiation 13. Derivatives of Polynomial and Exponential Functions 13.1. Polynomials. Let us recall that polynomials are sums of power functions with nonnegative integer exponents, such as the function f (x) = 3x2 + 4x + 6. In this section, we will deduce general rules for the derivatives of polynomial functions. We start by their "building blocks," power functions. The simplest of these is the class of constant functions. THEOREM 3.1. Let c be a real number and let f(x) = c for all z. Then f'(a) = 0 for all real numbers a. Before we prove the theorem, we point out that, intuitively, it makes perfect sense. The derivative of a function f describes the rate of change of f, but if f is a constant function, then f never changes (it has zero change). PROOF OF THEOREM 3.1. We have fI(a =limf~cfa f'a x-a z - a = lim C-C x- a c - a =-0. Note that limxma(c - c)/(x - a) = 0 since (c - c)/(x - a) = 0 for all values x / a. D We now turn our attention to a more general class of power func- tions, those of the form f(x) = z", where n is a positive integer. Let us recall the algebraic identity X"n - a"n = (x - al) - (z"-1 + Xzn-2a + -.-. + zaln-2 + a"-1).i T H EOREM 3.2. Let n be a positive integer and let f(c) =c". Then f'(a)=a". 57  58 3. RULES OF DIFFERENTIATION PROOF. We have f' () - f(x) -f(a) f '(a) = limf(x f(a x-ma x- a = lim -a x- a x- a = lim (x - a)-(xz"- + Xn-2a +.---+ xan-2 + a"-1) x- a x-a = lim X-1+ zn-2a . n-2 +n-11 x-- a = na"-1 Note that this agrees with our result from the last section that showed that if f (x) = x3, then f'(x) = 3x2. It turns out that Theorem 3.2 holds even if n is not a positive integer. That is, for all real numbers a, if f(x) = x, then f'(x) ax-1. We will see a formal proof of this fact later. In the exercises, you are asked to prove two special cases of this general result. 13.1.1. Three Simple Rules. Derivatives are limits of certain functions, so it is not surprising that some of the laws governing their computation are very similar to limit laws. That is, if we know the derivative of f and g, then we can easily compute the derivative of f + g, f - g, and cf, where c is a given real number. The rules are as follows. THEOREM 3.3. Let f and g be two functions that are differentiable at a. Then f + g is differentiable at a, and (f + g)'(a) = f'(a) + g'(a). PROOF. We have (f~)'a)- im(f +g)(x)- f g)(a) x-ma x- a = lim f(x) - f(a) g(x) - g(a) x a-a c-a f'(a) + g'(a). The other two rules and their proofs are so similar that they are left as exercises. T HEOREM 3.4. Let f and g be two functions that are differentiable at a. Then f - g is differentiable at a, and (f - g)'(a) =f'(a) - g'(a).  13. DERIVATIVES OF POLYNOMIAL AND EXPONENTIAL FUNCTIONS 59 THEOREM 3.5. Let f be a function that is differentiable at a and let c be a real number. Then cf is differentiable at a and (cf)'(a) = cf'(a). It is very important to point out that the other limit laws do not carry over to derivatives in the same fashion. That is, in general, (fg)' / f'g', and (f/g)' / f'/g'. We will learn some more complicated rules to compute the derivatives of fg and f/g in the next section. Theorems 3.3 to 3.5 enable us to compute the derivative of any polynomial function. EXAMPLE 3.1. Let p(x) = 3x3 + 5x2 - 6x + 8. Find p'(x). Solution: Note that p(x) is just a sum (and difference) of constant multiples of power functions. The derivatives of power functions are computed in Theorem 3.2. Then we can apply Theorems 3.3 to 3.5 to get p'(x) = (3x3)' + (5x2)' - (6x)' + (8)' = 3(x3)' + 5(x2)' - 6(x)' + (8)' =9x2 + 10x - 6. 13.2. Exponential Functions. Let us now compute the derivative of the exponential function f(x) = bx, where b is some positive constant. By the definition of derivatives, we get f/(a) = limf(x)f(a) x-ma x- a = lim x-ma x - a lmba+z - b = lim z->0 Z = balim bz - 1 z--0 z = baf'(0). Several comments are in order. First, note the substitution z =x-a in the third line. Second, note that bo is a constant that does not depend on z; hence, the limit law for constant multiples was used in the fourth line. Third, in the special case when a =0, the definition of the derivative yields f'(0) =limz-0(bz - 1)/z. We used this fact in the last line.  60 3. RULES OF DIFFERENTIATION In other words, (3.1) f'(x) =_f'(O)bx _ f'(O)f(x). That is, the derivative of the function f is a constant multiple of f. The constant in question is f'(O), that is, limz-o(bz - 1)/z. Numerical experimentation suggests that the larger b is, the larger this limit is. Graphical experimentation suggests this as well. Indeed, f'(O) is the slope of the tangent line to the curve of f(x) = bX at the point x = 0, and plotting f for various values of b suggests that the larger b is, the larger this slope is. In particular, it can be proved that there exists a real number e, close to 2.71, such that ez -1 lim z 1. z-0 z This real number e is the basis of the natural logarithm that we denote by ln. The reader may wish to consult the interactive website http://www. math.ufl.edu/~mathguy/ufcalcbook/exponent.html for further illustra- tions. DEFINITION 3.1. Let e be the real number such that ez -1 lim z 1. z-0 z So, in the special case of b = e, Equation (3.1) takes the form (ex)' = ex since ez - 1 f'(0) lim 1. z-0 z That is, the derivative of f(x) = ex is f(x) = ex itself. In Section 16, we will see what that implies for the derivatives of exponential functions with bases different from e. 13.3. Exercises. (1) Let f (x) = x3 + 2x2 + 3x +4. Compute f'(x) and f"(x). (2) Let f(x) = x4 - 3x + 9. Compute f'(x) and f"(x). (3) Let f(cc) = c8 - 2c4 + 1. Compute f'(c), f"(c), and f"'(c). (4) Let f(cc) =-cc - cc3. Compute all derivatives (first, second, third, etc.) of f(cc). (5) Prove that if f is a polynomial function, then f'(cc) is also a polynomial function.  14. THE PRODUCT AND QUOTIENT RULES 61 (6) Prove that if f is a polynomial function of degree d, then f (d+1)(X) = 0 for all real numbers x. (7) Prove that if f is a polynomial function of degree d, then f(d) is a linear function. (8) Let p be a polynomial function of degree d and let k < d be a nonnegative integer. What kind of function is f(k)? (9) Prove that if f(x) =x1/2 and a > 0, then f'(a) = g. (10) Prove that if f(x) = 1/x and a / 0, then f'(a) = - . (11) Prove Theorem 3.4. (12) Prove Theorem 3.5. (13) Let f(x) = 3x3 - 42 + cc - 2 + 4ex. Compute f'(x). (14) Let f(x) = 1// c. Use the remark after the proof of Theorem 3.2 to compute f'(x). (15) Let f (x) = x2 - 2x+7, and let g(x) = ex. Compute (f+g)'(x) and (f - g)'(x). (16) Is there a function f that is not identically zero such that f(k)() = f(x) for all c? (17) Could it happen that f and g are two different functions, but f'(x) = g'(x) for all c? (18) Could it happen that f and g are two different functions, f' and g' are two different functions, but f"(x) = g"(x) for all c? (19) Could it happen that f and g are two polynomial functions of different degree, and f'(x) = g'(x) for all c? (20) Is there a polynomial function f(x) that is not identically zero such that there is a real number x for which f(k) (c) does not depend on k? 14. The Product and Quotient Rules 14.1. The Product Rule. We mentioned in the last section that, in gen- eral, (fg)' / f'g'. For instance, if f(x) = 2x+ 1 and g(x) = x+2, then (fg)(x) = 2x2 + 5x + 2, so (fg)'(x) = (2x2 + 5x + 2)' = 4x + 5, while f'(x) = 2 and g'(x) = 1, so f'(x)g'(x) = 2. It turns out that there is a rule to compute the derivative of a product; it is just a little bit more complicated than the limit law for products. This is the focus of our first theorem in this section. T HEOREM 3.6. Let f and g be two functions that are differentiable at a. Then fg is differentiable at a, and (fg)'(a) = f (a)g'(a) + f'(a)g(a).  62 3. RULES OF DIFFERENTIATION PROOF. By definition, we have (3.2) (fg)'(a) = lim - f(a)g(a) x-ma x - a The crucial idea is to decompose the difference f(x)g(x) - f(a)g(a) as (f(x)g(x) - f(x)g(a)) + (f(x)g(a) - f(a)g(a)) in the numerator of the right-hand side of (3.2). Using this idea, we obtain from Equation (3.2) (fg)'(a) = lim f(x)(x) - f(x)g(a) f(x)g(a) - f(a)g(a) x-ma x-a x-a = lim f(x)g(x) - f(x)g(a) + lim f(x)g(a) - f(a)g(a) x-a x -a x-a x -a g__)__ga f(x) -f(a) = lim f J(x - + lim g(x) - x- a x-a x-a x -a = f (a)g'(a) + g(a)f'(a). EXAMPLE 3.2. The derivative of h(x) = x2ex can be computed as follows. Let f (x) = x2 and g(x) = ex. Then h'(x) = (fg)'(x) = f(x)g'(x)+ f'(x)g(x) = x2(ex)I + (x2)Iex = 2ex + 2xex = ex(X2 + 2x). 14.2. The Quotient Rule. The rule for the derivative of the quotient of two functions is a little bit more complicated than that for the deriva- tive of the product of two functions. Though more complex, both the rule and its proof bear some similarity to the rule given in Theorem 3.6. THEOREM 3.7. Let f and g be two functions that are differentiable at a and let us assume that g(a) / 0. Then f/g is differentiable at a, and we have (a) = g(a)f'(a) - f(a)g'(a) \91 ) (a)2 PROOF. By definition, we have i f(x) _f(a) (3.3) ( ~(a)- lim g4x gCa). ~gj x-a cc- a  14. THE PRODUCT AND QUOTIENT RULES 63 Let us multiply both the numerator and the denominator of the right- hand side by g(x)g(a) to get (f\' (.f(x)g(a) - f(a)g(x) (a)/ lina (x - a)g(x)g(a)~ Now transform the numerator of the right-hand side by subtracting and then adding g(a)f(a) to get (f ' (a) lim f(x)g(a) - g(a)f(a) + g(a)f(a) - f(a)g(x) - x-(=a (x - a)g(x)g(a) g(a) f(x) - f(a) - lim f(a) g(x x-a~x~~a) x - a x-~ag(x)g(a) g(a)f'(a) - f(a)g'(a) g(a)2- ) - g(a) x-a F- Theorem 3.7 now enables us to compute the derivative of rational functions. EXAMPLE 3.3. Let h(x) =_(x + 3)/(x2 + 1). Find h'(x). Solution: Let f(x) = x + 3 and let g(x) = x2 + 1. and g'(x) = 2x. So, by Theorem 3.7, we have h'(x) = g(x)f'(x) - f(c)g'(X) _X2 + 1 - (x + 3)2x g(x)2 4 + 2X2 + 1 Then f'(x) = 1 -2 -26x+1 x4 + 2x2 + 1 14.3. Exercises. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) Let h(x) = exz3. Find h'(x) and h"(x). Let f (x) = (2x + 7)ex. Compute f'(x). Find a rule to compute (f2)'(x). Find a rule to compute (f3)'(x). Find a rule to compute (1/f)'(x). Use the result of the previous exercise to prove a formula for g'(x) if g(x) = x" for a negative integer n. Let g(x) = e-x. Find g'(x). Let h(x) = c/ex. Find h'(x). Let f(x) = . Compute f'(x). Let f(x) = . Compute f'(x). Let f (x) = ex/(x + 2). Compute f'(x). Let g(x) = e2X. Use the product rule to compute g'(x). Let g(x) = e-2x. Compute g'(x). Try to find three different ways to obtain your result.  64 3. RULES OF DIFFERENTIATION (14) Let h(x) = (ex + 1)(ex + 2). Compute h'(x). (15) Let g(x) =(x - 3)/(ex + 1). Compute g'(x). (16) Let f (x) = (2x +3)/(4x + 7). Compute f'(x). Try to find two different ways of getting the same answer. (17) Let f(x) = 1/(1 - x). Find f'(x). (18) Let f(x) = 1/(1 - z)". Find f'(x). (19) Let f(x) = g(x)h(x), where g is a polynomial function of x, and h(x) = ex . Prove that f'(x) and f"(x) are each equal to the product of a polynomial function and the function h(x) em. (20) Prove that if f(x) is a rational function, then f'(x) is also a rational function. 15. Derivatives of Trigonometric Functions In this section, we show how to compute the derivatives of trigono- metric functions. First, we compute (sin x)'. This will be a somewhat lengthy procedure, due to the fact that this is the first trigonometric function we will differentiate and we will have to apply new methods. However, once we know the derivatives of sin x and cos x, it will be much simpler to deduce the derivatives of other trigonometric func- tions, since those functions can be obtained from sin and cos, and then the various differentiation rules can be used. THEOREM 3.8. We have (sincc)'= cos x. PROOF. Recall the identity sin(a + b) = sin a cos b + sin b cos a. We have (sinc)'lim sin(x + h) - sinx h~0 oh = lim sin x cos h + sin h cos x - sin x h~0 oh (C cosh - 1 sin h cos x = lim sincx h + h cosh - 1 sin h = sincc lim + coscxclim . h-0 h h-0O h Note that as, h approaches 0, we certainly have limh~o sin x = sin x and limh~0 cos x = cos x, since these functions do not even depend on h. There remains the task of computing the two nontrivial limits cos h- 1 sin h lim and lim . h-0 h h-0 h We will carry out this task in two lemmas.  15. DERIVATIVES OF TRIGONOMETRIC FUNCTIONS 65 LEMMA 3.1. We have sin h lim = 1. h-0 h PROOF. Let us consider a circle with unit radius and a regular n- gon whose center is at the center 0 of the circle and whose n vertices are all on the unit circle. Then the area of the circle is 7, and the area of the n-gon isn - .-sinca, where a = 2/n is the angle AOB, with A and B being adjacent vertices of our n-gon. Considering just 1/n of both the circle and the n-gon, we see that the area of the triangle AOB is (sin a)/2, and the area of 1/n of the circle bordered by the lines AO, BO, and the arc AB is 7w- a/(27) a/2. So the ratio of the two areas is (sina)/2 sin a a/2 - a On the other hand, as n gets larger and larger, a gets smaller and smaller, while the area of the n-gon gets closer and closer to the area of the circle. Hence, their ratio, sin a/a, will get arbitrarily close to 1 and stay arbitrarily close to 1. D LEMMA 3.2. The equality (3.4) nl ocos h - 1 - (3.4) lim h holds. PROOF. We will manipulate the expression (cos h - 1)/h so that we can use the result of Lemma 3.1. First, we multiply both the numerator and the denominator by cos h + 1 to get cos h - 1 cos2h- 1 -sin2 h h h(1 +cos h) h(1-+-cos h)' Therefore, we have coshh - 1 sin2 h h-0 h h-0h(1 + cos h) -1limsin h sin h he0 hj 1\+cosh = (-1).-0O 0.  66 3. RULES OF DIFFERENTIATION We can now finish the proof of Theorem 3.8. At the end of the first displayed chain of equations in that proof, we saw that cos - 1sin h (sin z)' =sin xlimcosh-i + cos x lims . h-0 h h-O h The previous two lemmas showed that, on the right-hand side, the first limit is 0 and the second limit is 1, so (sin x)' = coscx as claimed. Q The following theorem can be proved by very similar methods. THEOREM 3.9. The equality (cos x)' sin z holds. You are asked to prove this theorem in Exercise 15.1.1. Now that we have the derivatives of sin and cos, the derivatives of other trigonometric functions can be obtained by simply using the quotient rule. The next theorem shows an example of this. THEOREM 3.10. We have (tancc)' = sec2 X. PROOF. Note that tan x = sin c/cos x, so we can apply the quotient rule. This leads to (tancc)' (sin c cos x coscx- (sin z)' - sin x(cos c)' COS2 c cosx + sin2 X COS2 c 1 COS2 c sec2 c. F- The derivatives of the other three trigonometric functions are given in the exercises. 15.1. Exercises. (1) Prove that (2) Prove that (3) Prove that (4) Prove that (5) Let h(x) = (6) Let h(x) = (7) Let h(x) = (8) Let h(x) = (9) Let h(x) = (cos c)' - sin c. (cot c)'= -csc2X. (csccz)'= -csCc cotcc. (sec c)' = seccc tan c. xsincc. Find h'(x). (X2 - 2x + 3) cos c. Find h'(x). tan. Find h'(x). ex cos c. Find h'(x). ex / sinc. Find h'(x).  16. THE CHAIN RULE 67 16. THE CHAIN RULE 67 (1 0 ) L et h (x) (11) Let h(Cx) (12) Let h() (13) Compute (14) Compute ways. (15) Compute (16) Compute (17) Compute (18) Compute ways. (19) Compute cise 13. (20) Compute cise 14. = e2xtancz. Compute h'(x). = e-2x cot z. Compute h'(x). = e +SifX Find h'(x). X2+1 (sin2 c)'. (cos2 x)'. Try to get the same answer in two different (sin x tan z)'. (cot2 X)'. (tan2 x)'. (sec2 x)'. Try to get the same answer in two different (sin3 c)'. You may want to use the result of exer- (cos3 cc'. You may want to use the result of exer- 16. The Chain Rule 16.1. The Derivative of the Composition of Two Functions. In previous sections, we learned how to compute the derivative of the sum, differ- ence, product, and quotient of two functions. We still do not know how to compute the derivative of the composition of functions, such as h(x) = sin(3x), t(x) = /2 + 1, or r(c) = esin. In this section, we will learn a rule, called the chain rule, that applies in these situations. THEOREM 3.11 (Chain Rule). Let h(x) = f(g(x)), where g is dif- ferentiable at x and f is differentiable at g(x). Then h is differentiable at x, and we have h'(x) = f'(g(x))g'(x). In other words, we first differentiate the outside function at a point given by the inside function, then multiply the result by the derivative of the inside function. The proof of the chain rule is somewhat technical, so we will post- pone it until the end of this section. Now we will discuss some examples of the applications of the chain rule. EXAMPLE 3.4. Find the derivative of h(x) = sin(3x). Solution: Let f (x) = sincx and let g(x) = 3c. Then h(x) = f (g()), so, by the chain rule, we have h'(x) = f'(g(x)) - g'(x) =_(cos(3x)) - 3 = 3 cos(3x).  68 3. RULES OF DIFFERENTIATION EXAMPLE 3.5. Let h(x) = v/2+ 1. Find h'(x). Solution: Recall that in Section 13, we mentioned that the identity (z")' =nz"-1 holds for any nonzero real number. Therefore, selecting n = 1/2, we get that ( /c)'= (z1/2)' =1z-1/2 1 Now we can prove the statement of the example. Let f(x) = V, and let g(x) = x2 + 1. Then h(x) = f(g(x)), so, by the chain rule, we have 1 xc h'(x) = f'(g(x))g'(x) -2c 29/z2+1 v/2+1 Sometimes the chain rule is written in the Leibniz notation, that is, as dh dh dg dzc dg dz'~ 16.2. Two Applications of the Chain Rule. 16.2.1. A Simple Way of Obtaining (cos x)'. Recall that in the last sec- tion, it took considerable time and effort to prove that (sin c)' = cos c. Finding (cos c)' with similar methods is just as time-consuming. On the other hand, the chain rule enables us to compute (cos c)' faster. Recall that cos x = sin(x + 2). So we can write cos x as the com- position of two functions, namely cos x = f(g(x)), with f(x) sincx and g(x) = x + 2. So the chain rule applies, and we get (cos c)'= f'(g(x)) - g'(X) = cos (x + 1 7F 7r = cos x cos - - sinx sin- 2 2 =0 - sinx - -sinc. 16.2.2. The Derivatives of Exponential Functions. Recall that we de- fined the number e such that the derivative of the exponential function f (x) = ex was f(x) itself. Now the chain rule enables us to compute the derivatives of exponential functions with any base. T HEOREM 3.12. Let a be a positive real number and let h(cc) =ax. Then we have h'(cc) =a ln a.  16. THE CHAIN RULE 69 PROOF. Note that h(x) =a e") ri" So we have succeeded in writing h as the composition of two functions, namely h(x) = f (g(x)), where f (x) = ex and g(x) = x ln a. Therefore, the chain rule applies, and we get h'(x) =_f'(g(x))- g'(x) eina lna = axlna. 16.3. Proof of the Chain Rule. It is time that we proved the chain rule. PROOF OF THEOREM 3.11. As g is differentiable at x, we know that (3.5) lim g(x+'r) - g(x) - g(X)=0. Set g(x + r) - g(x) ,9() r Note that t depends on r, and as r approaches 0, t approaches 0. Similarly, let y = g(x). As f is differentiable at y, we have (f(y +s) - f(y) , (3.6) lim y f'(Y) = 0. s 0 s Set U = - )f) f /(y). s Again, note that u depends on s and that u approaches 0 as s ap- proaches 0. Now we undertake a series of manipulations of the preceding two equations. Our goal is to express lim f(g(x + r)) - f (g(x)) r-0 r in terms of f'(g(x)) and g'(x). Rearranging the equation that defines the variable t that we just introduced, we get (3.7) g(x + r) g(x) + (g'(x) + t)r. Similarly, rearranging the equation that defines the variable tt, we get (3.8) f (y + s) =f (y) + (f'(y) + tt)s. Now apply the function f to both sides of (3.7) to get (3.9) f(g(x + r)) = f (g(x) + (g'(x) + t)r) .  70 3. RULES OF DIFFERENTIATION Observe that (3.8) holds for all y and s, so, in particular, it holds when y = g(x) and s = (g'(x) + t)r. Making these substitutions in (3.8), Equation (3.9) yields (3.10) f (g(x + r)) f (g(x) + (g'(x) + tjr) (3.11) f(g(x)) + (f'(g(x)) +Qu) - (g'(x) + u)r. We can now express the quotient (f(g(x + r)) - f (g(x)))/r from the equality of the left-hand side of (3.10) and the expression in (3.11) as f (g(x +'r)) - f (g(x)) (f'(g(x)) +u)(g'(x) + t)r r r (f'(g(x)) + U) (g'(x) + t). Finally, we are in a position to compute the derivative we were looking for as the limit of the left-hand side as r approaches 0. We get f (g(x + r)) - f (g(x)) _1('gx)+Q)g()+t lim = g( ) g() lim (f, (g (X)) + U)(g (x) + t) rp r-~0 = (lim f'(g(x)) +lim u). (lim g'(x) +lim t)I = f'(g(x)) - g'(x) since both t and u approach 0 as r approaches 0. D 16.4. Exercises. (1) Let h(x) =(x2 + 1)5. Find h'(x). (2) Let h(x) = cot(2x). Find h'(x). (3) Let h(x) = cos(2x + 8). Find h'(x). (4) Let h(x) = sin(x2). Find h'(x). (5) Let h(x) = sin3 x. Find h'(x). Compare your result to the result of exercise 19 of the previous section. Which other ex- ercises of the previous section can be solved by the chain rule? (6) Let fv(x) = 2x + 3. Find f'(x). (7) Let h(x) = esinx. Find h'(x). (8) Let h(x) = 2cosx. Find h'(x). (9) Let h(x) = ex2sin. Find h'(x). (10) Let h(x) =v/2 + 2x + 7. Find h'(x). (11) Let h(cc) =sin(ex + 5cc + 6). Find h'(cc). (12) Let h(cc) e e+. Find h'(cc). (13) Let h(cc) =esin(x2). Find h'(cc). (14) Let h(cc) =sin(2cc). Find h'(cc). How could you get the same result without using the chain rule?  17. IMPLICIT DIFFERENTIATION 71 (15) Let h(x) = cos(2x). Find h'(x). How could you get the same result without using the chain rule? (16) Let h(x) = 2x . Find h'(x). (17) Let h(x) = 1/(1 - x). Find h'(x). Try to get the same answer in two different ways. (18) Let h(x) = (2 + tan x)3. Find h'(x). (19) Let h(x) 1= - x2. Find h'(x). (20) Let h(x) =v/ + xz + 1. Find h'(x). 17. Implicit Differentiation In the last several sections, we computed the derivatives of many different functions. Although these functions were different, they had one important feature in common. They were explicitly given. That is, they were given by a rule that directly described how f(x) = y is obtained from x. 17.1. Tangent Lines to Implicitly Defined Curves. Sometimes we have to deal with curves that are given by a different kind of rule. Consider the curve given by the equation (3.12) X3 _y3+_J4x. Let us say that we want to compute the slope of the tangent line to this curve at the point (2, 2). If we could express y as a function of x, we could simply take the derivative of that function at x = 2. However, it is not clear how to write y explicitly in terms of x, even if (3.12) implicitly describes this dependence. It is in these situations that we resort to implicit differentiation. Keep in mind that we do not need to explicitly know how y depends on x, that is, we do not need an explicit expression for the function y(x); we only need to know the derivative dy/dz of that function at x= 2. Consider Equation (3.12), and differentiate both sides with respect to the variable x to get d d dz +Y) d(4Y). Now recall that y = y(x) is a function of x. So, when computing (d/dzc)ys on the left-hand side, we need to use the chain rule. On the right-hand side, we need to use the product rule and the chain rule. Using these rules, we get dy dy 3xc2+ 3y2 = 4y + 4x. dzc dzc  72 3. RULES OF DIFFERENTIATION Expressing dy/dz from this equation, we get dy (4y - 3x2) dz (3y2 - 4x) At the point (2, 2), the right-hand side is -4/4 = -1, so the slope of the tangent line at (2, 2) is -1. Note that the fact that the tangent line at (2, 2) has slope -1 makes (intuitively) perfect sense, since the curve in question is symmetric in c and y. That is, if (x, y) is on the curve, then (y, c) is also on the curve. 17.2. Derivatives of Inverse Trigonometric Functions. One place where implicit differentiation is a very powerful tool is in the computation of the derivatives of inverse trigonometric functions. Recall that tan-1 x y is the function that is the inverse of the restriction of the function tan x to the interval (-w/2,wr/2). That is, if tan-1 x = y, then (3.13) x = tany, where y E (-w/2,wr/2). Our goal is to determine d tan-1 x = dy dcc d To that end, let us take the derivative of both sides of (3.13) with respect to c. Recalling that d tanz sec2z and y=y(x), dz we get sec 2 y .dy 1 =sey- . dx Solving for dy/dz and recalling the identity sec2 z = 1 + tan2 z, we obtain dy 1 1 1 dcc sec2y 1 +tan2yl1cc2 In other words, we proved the suprisingly simple formula (tan-1 cc)' =1 .c  17. IMPLICIT DIFFERENTIATION 73 This formula is interesting for two reasons. First, it is surprisingly sim- ple. Second, it does not even contain trigonometric functions. Imagine trying to get this result without implicit differentiation, using just the definition of derivatives. You will be asked to compute the derivatives of the other inverse trigonometric functions in the exercises. 17.3. Exercises. (1) Let C be the circle given by the equation x2+ y2 = 169. Use implicit differentiation to find the slope of the tangent line to C at the point (5, 12). (2) Let E be the ellipse given by the equation x2+ zy + y2 = 108. Use implicit differentiation to find the slope of the tangent line to E at the point (6, 6). (3) Let y be implicitly defined by the equation zi + y5 = 1. Com- pute dy/dz. (4) Let y be implicitly defined by the equation y = sin(xy). Com- pute dy/dz. (5) Let y be implicitly defined by the equation x + y = y. Com- pute dy/dz. (6) Let y be implicitly defined by + fy = 4. Compute dy/dz. (7) Let us assume that h(x) + x cos h(x) = x3 for all real numbers x. Is this sufficient information to determine the value of h'(0)? (8) Let us assume that h(x) + e(x)= 5 for all real numbers x. Prove that then h must be a constant function. (9) Prove that (sin-1 )' = 1 (10) Prove that (cos-1)'= -1 (11) Prove that (cot-1 x)' = -+. (12) Prove that (sec-1 x)' =1 (13) Prove that (csc-1x)'= - 1. (14) Compute (tan-1 c)'. (15) Compute (cos-(2xc+0.1))'. (16) Compute (sin-1(c2)). (17) Compute (sin-1(1/c))'. (18) Compute (tan-1(2xc))'. (19) Compute (csc-l(c/3))'. (20) Compute (sec-1(xc - 0.01))'.  74 3. RULES OF DIFFERENTIATION 18. Derivatives of Logarithmic Functions 18.1. The Formula for (logo x)'. As another powerful application of im- plicit differentiation, we compute the derivative of the function f(x) ln z. THEOREM 3.13. We have (Incz)' PROOF. Set y =Inx. Theneu= respect to x, we get dy dx dy dx However, eC = x by definition, so dy dcc as claimed. 1 cc x. Differentiating both sides with = 1, 1 1 z F- Note that the function ln x is defined for positive values of x. If x < 0, then the function y =ln(-x) is defined. The reader is in- vited to practice the method of implicite differentiation by showing that (ln(-x))' = 1/c for all negative real numbers c. In other words, if y = ln(|x|), then dy/dz= 1/c as long as x is a nonzero real number. It is now a breeze to determine the derivative of logarithmic func- tions of any base. COROLLARY 3.1. Let a -/ 1 be a fixed positive real number. Then (log, c)' 1 x ln a PROOF. Note that X = (el"") "a e(ln a) (og x) So ln x = (ln a) (log x) and ln x f(x) = logc ln= c In a As ln a is a constant, it follows that f'(cc) 1 (lnz)' nlncc 1 c ln a as claimed. F-1  18. DERIVATIVES OF LOGARITHMIC FUNCTIONS 75 18.2. The Chain Rule and In x. An interesting consequence of Theorem 3.13 is the following. COROLLARY 3.2. Let f(x) be a differentiable function that takes positive values only. Then d inf'(x ) dIn f (x) = .x dzc f (c)~ PROOF. By the chain rule, dx d df f'(x) dx df dx f(x) EXAMPLE 3.6. Let f(x) = cos x. Then dln(coscz) -tin x --tanx. dx coscx 18.3. Logarithmic Differentiation. Sometimes we need to compute the derivative of a complicated product. This is sometimes easier by taking the logarithm of the product, which will be a sum, and using implicit differentiation. This procedure, which is called logarithmic differenti- ation, has the inherent advantage that it deals with sums instead of products, and sums are much easier to differentiate than products. EXAMPLE 3.7. Let x3 x -+ 1 Compute dy/dx. Solution: Taking logarithms, we get 1 1 In y = 31n c + -ln(x + 1) - -ln(x -2). 2 2 Now taking derivatives with respect to c and using Corollary 3.2, we have dy1 3 1 1 dc y c 2(x + 1) 2(x - 2) Finally, we can solve this equation for dy/dz to get dy (3 1 2(c1) ccz c- 2 + 2(cc+ 1) -2(cc- 2))  76 3. RULES OF DIFFERENTIATION 18.4. Power Functions Revisited. Recall that in an earlier section, we proved that if n is a fixed positive integer, then (z")' = nz"-1. We stated that this was the case for all nonzero real numbers n, not just positive integers, but we have not proved that claim. Now we have the tools, namely logarithmic differentiation, to prove it. THEOREM 3.14. Let n be any nonzero real number. Then we have d dzc PROOF. Set y =cc". Let us assume for the case of simplicity that x is positive. Taking logarithms, we have lny =nlncc. Differentiating both sides with respect to x, we get dy 1 n dcc y c Solving for dy/dz yields dy ny n cc-nn-1 as claimed. D 18.5. The Number e Revisited. Recall that we have defined the num- ber e, the base of the natural logarithm, as the number for which limh-o (eh - 1)/h = 1. Our new knowledge lets us express e more di- rectly, as a limit. Note that if f(x) = lncx, then f'(x) = 1/c, so f'(1) = 1. By the definition of derivatives, this means that lm ln(1+h) -lIn1 lim = 1. h-0 h Observing that ln 1 = 0 and using the power rule of logarithms, we get lim ln(1 + h)l/h 1 h--0 or, applying the exponential function ez to both sides, we have lim(1 + h)1/h - e. h-0 Equivalently, setting x = 1/h, we get lim I1+ - = e. Either of the last two formulas can help to determine the approximate value 2.712828 of e.  19. APPLICATIONS OF RATES OF CHANGE 77 18.6. Exercises. (1) Let h(x) = 2x + 3x. Compute h'(x). (2) Let h(x) = 4-x. Compute h'(x). (3) Let h(x) = 2X. Compute h'(x). (4) Let h(x) = e2. Compute h'(x). (5) Compute y ln( x+1). (6) Compute (inx)'. (7) Compute (xx)'. (8) Let h(x) = 3x + 1. Compute h'(x). (9) Let h(x) = 4x2 + 2. Computer h'(x). (10) Let h(x) = 52x + 3. Compute h'(x). (11) Compute f'(x) if f(x) X4 .+1 (12) Compute lim-,o (1 + ) . (13) Compute lim-,o (1 + )x. (14) Compute lim x,, (1 - ) . (15) Let h(x) = ln(sin-1 x). Compute h'(x). (16) Let h(x) =sin-1(ln z). Compute h'(x). (17) Let h(x) = in (--). Compute h'(x). (18) Use logarithmic differentiation to find y'(x) xx (19) Use logarithmic differentiation to find y'(x) x. (20) Use logarithmic differentiation to find y'(x) 2 cosx dy/dx if y(x) dy/dx if y(x) dy/dx if y(x) 19. Applications of Rates of Change In this section, we consider a few applications of derivatives in var- ious disciplines. 19.1. Physics. Recall that if an object moves along a line and the distance it covers in time t is described by the function s(t), then (3.14) ds v(t) = it st lim s(t + h) - s(t) h-o h is the instantaneous velocity of the object at time t. We can take this concept one step further. If the object moves at a changing velocity, then the rate of change of the velocity itself can be important information. For instance, when considering a vehicle's performance, we may be interested in how fast it can reach its top speed, not only what its top speed is.  78 3. RULES OF DIFFERENTIATION The corresponding notion in physics is called acceleration, and is denoted by a(t). That is, keeping the previous notation, we have dv (3.15) a(t) v=o'(t) d s"(t). EXAMPLE 3.8. The position of a particle is described by the equation 13 (3.16) s(t) = -t3 - 3t2 + 5t. Here s is measured in meters and t in seconds. (I) What is the velocity of the particle after 3 seconds? (II) Find the acceleration of the particle after 10 seconds. (III) When does the particle move backward? Additional questions about the movement of this particle will be given in the exercises. Solution: (I) The velocity of the particle is described by the function v(t) s'(t) = t2 - 6t + 5. This yields v(3) = 9 - 18 + 5 = -4. So the velocity of the particle after 3 seconds is -4 m/s, meaning that the particle is moving backward at a speed of 4 meters per second after 3 seconds. (II) The acceleration of the particle is given by the formula a(t) v'(t) = 2t - 6. So, after 10 seconds, the particle is accelerating at 14 m/s2 (III) The particle is moving backward when its velocity v(t) is nega- tive. That happens when v(t) = t2 -6t+5 = (t-1)(t -5) <0, that is, when t E (1, 5). In other words, the particle is moving backward between the first and fifth seconds. D 19.2. Economics. Let us say that a company estimates that it costs C(x) dollars to produce x units of a new product. It is often the case that C(x), which is called the cost function, can be described by a polynomial function, such as C(x)=a+bx+cx2+dx3. The reason for this is as follows. There will be some costs, such as designing the product and obtaining permits, that will be present re- gardless of the number of units produced. These will be represented by the constant term a. Then there will be costs, such as renting a location and buying supplies, that will be more or less in direct proportion to the number of units produced. These will be represented by the linear  19. APPLICATIONS OF RATES OF CHANGE 79 term bx. Then there will be other factors, such as hiring workers, mar- keting the product, and organizing production, that will be in direct proportion to a higher power of x as the differences in size turn into differences in kind. Taxes may factor in at an even higher rate. Because the cost function C(x) is not a linear polynomial, producing the 1001st unit does not cost of the same as producing the first unit or the 5001st unit. The cost of increasing production from n units to n + 1 units, in other words, the cost of producing the (n + 1)th unit, can be computed by the formula M(n) = C(n +1) - C(n). The marginal cost function C'(x) describes how the cost function changes. In that, C'(x) and M(n) are similar. There is one important difference. As we know, the derivative C'(x) is given by (3.17) li C(x+Ax) - C(x) However, it could well be that the smallest meaningful positive value of Ax is 1, in case the products are such that fractional units do not make sense (e.g., automobiles). In that case, Ax - 0 is impossible in its precise mathematical meaning; the closest that Ax can get to 0 is when Ax= 1. In that case, however, the expression after the limit symbol in (3.17) simplifies to C(x + 1) - C(x), justifying the approximation (3.18) M(x) = C(x + 1) - C(x) C'(x). EXAMPLE 3.9. The cost function of a bottle of a new medication is given by C(x) = 106+20x+0.001x2+0.000001x3. Find the approximate cost of producing the 101st and the 1001st bottles. Solution: By the preceding discussion, we need to compute the func- tion C'(x). By the rules of differentiating a polynomial function, we get C'(x) = 0.000003x2 + 0.002x + 20. So the 101st bottle costs 0.0003- 1002 + 0.002 - 100 + 20 = 20.23 dollars to produce, while the 1001st bottle costs 0.000003 - 10002 + 0.002- 1002 + 20 = 43 dollars to produce. D It is important to note that the result of the previous example, that is the fact that it costs more to produce the 1001st bottle than the 101st bottle does not mean that the more bottles are produced, the more expensive it is to produce the average bottle. This is because the cost of producing the first bottle is astronomical, since C(1) > 106. Compared to that, the cost of each of the first thousand, or even, first  80 3. RULES OF DIFFERENTIATION ten thousand bottles is very small, so the production of each of them will bring the cost of producing the average bottle down. (The cost of producing the average bottle if n bottles are produced is of course C(n)/n.) In the exercises, you are asked to compare these results to the results obtained by using the formula C(n + 1) - C(n). 19.3. Exercises. (1) Consider the particle of Example 3.8. After 6 seconds, how far from its starting point is that particle? In what direction? (2) Consider the particle of the previous exercise. Are there any moments when the particle is not moving? (3) The location of an object moving vertically is described by the function s(t) =t _ for t E [0, 5], where time is measured in seconds and distance is measured in meters. When will the object have an instantaneous velocity of 0.2 m/s? (4) Consider the object of the previous exercise. When does it have the greatest speed going up? When does it have the greatest speed going down? (5) Consider the object of Exercise 19.3.3. Will its acceleration ever be 1 m/s2? (6) Consider the object of Exercise 19.3.3. When will its acceler- ation be negative? (7) Use the formula M(n) = C(n + 1) - C(n) to find the cost of producing the 101st and 1001st units in Example 3.9. Com- pare your results with the estimates that we found using the function C'(x). (8) Two race cars speed up from a standing start to 60 m/s so that each car has constant acceleration. The first car reaches one-third of its top velocity in 4 seconds, while the second car reaches one-fourth of its top velocity in 3 seconds. Which car will have covered more distance by the time it reaches its top velocity? (9) A ball is rolling down a slope so that its distance from its starting point is described by the function s(t) = 2t2 + 6t, where 0 < t < 10, the time t is measured in seconds, and the distance s(t) is measured in meters. What will be the velocity of the ball after 3 seconds? (10) Consider the ball of the previous exercise. When will its ve- locity reach 40 m/s?  19. APPLICATIONS OF RATES OF CHANGE 81 (11) Consider the ball of the previous exercise and describe its ac- celeration as a function of the time t passed from the start of the ball's movement. (12) The cost function for a company to produce x bicycles is C(x) = 2400 + 3x + 0.6x2 + 0.002x3. Find the marginal cost function of this product. (13) Consider the cost function C(x) of the previous exercise and use it to determine the actual cost of producing the 1001st bicycle. (14) The cost function for a company to produce c laptops is C(x) = 1500 + 2x + 0.4x2 + 0.01c3. Find the marginal cost function for this product. (15) Consider the cost function of the previous exercise. Explain the meaning of C'(200). (16) Let us say that the function f(t) describes the growth of a certain bacteria population over time. That is, f (0) is the size of the population at the beginning of the observation period, while f(t) is the size of the population t hours after that. Explain why f'(t) describes the growth rate of the population in the moment of time corresponding to t. (17) A certain insect population has been exposed to an insecti- cide, which results in the population changing according to the function f(t) = 10,000 - 1000t - 500t2, where t is mea- sured in hours. Find the growth rate of the insect population after 1 hour and after 5 hours. (18) A certain bacteria population has an initial size of 1000, and it doubles in each hour for the next 10 hours. Describe the growth of this bacteria population by a function, then use that function to determine the growth rate of the population after 3.5 hours. (19) A certain rumor spreads according to the function 1 (3.19) P(t) 1 + 1 +30e-2t Here P(t) is the proportion of the observed population that heard the rumor t days after it started circulating. Explain why P'(t) should be defined as the rate at which the rumor is spreading. Then compute that rate at t =4. (20) Consider the rumor discussed in the previous exercise. If (3.19) remains correct as t goes to infinity, what fraction of the ob- served population will eventually hear the rumor?  82 3. RULES OF DIFFERENTIATION 20. Related Rates 20.1. Preliminaries. An intuitive idea of the notion of related rates comes from a simple fact of everyday life: If there are two related quan- tities that are changing with time, then their rates of change should also be related. For example, the volume V of water in a pool of area 20 m2 is related to the water level h (the pool depth in meters) as V = 20 h. Suppose the water level is low and needs to be increased. A hose is put into the pool that can pump water at a rate of 0.2 m3/h. At what rate does the water level increase? The volume and the water level are both functions of time, V = V(t) and h = h(t). For every instance of time t, their values are related as V(t) = 20h(t) and so must be their derivatives or rates of change: (3.20) V(t) = 20h(t) -- V'(t) = 20h'(t). Now the question is easy to answer. Since V'(t) = 0.2 m3/h, h'(t) V'(t)/20 = 0.01 m/h = 1 cm/h. The water level rises by 1 cm every hour. A somewhat practical estimate! You would know exactly when to come back and turn off the water if you needed an inch or so of the water level increase. Apparently, the same idea of related rates would work for lowering the water level after rain. 20.2. Units. It is important to bring all the quantities to the same system of units. For example, in the above problem the pool area is often given in square feet, for example, 200 ft2, while the pump- ing rate is given in gallons per hour, for example, V' = 60 gal/h. One gallon is 3.785 . 10-3 m3 and therefore V' = 60 - 3.785 - 10-3 0.2271 m3/h. One square foot is 9.29 - 10-2 m2, so the pool area is 200- 9.29- 10-2 = 18.58 m2. Hence, h' = 0.2271/18.58 1.2 cm/h. In 1999, NASA lost a $125 million Mars orbiter because a Lockheed Martin engineering team used English units of measurement while the agency's team used the more conventional metric system for a key spacecraft operation. 20.3. Formal Definition of Related Rates. DEFINITION 3.2 (Related quantities). Two quantities y and x are said to be related if there is a function f such that y = f(x). In the previous example, V =f(h) =20h. Suppose now that the quantities y and x are functions of another variable t (e.g., t is time): x =x(t) and y =y(t). Then the rate of change of x or y with respect to t is nothing but the derivative z'(t) or y'(t). The problem of "related  20. RELATED RATES 83 rates" can now be cast in the proper mathematical terms: What is the relationship between the derivatives z'(t) and y'(t) if the values of x(t) and y(t) are related by y = f(x)? The values of the functions x(t) and y(t) are related as y(t) = f(z(t)) for any t. Taking the derivative of both sides with respect to t by means of the chain rule (Theorem 3.11), we obtain a generalization of (3.20): (3.21) y(t) = f (z(t)) =- y'(t) = f'(z(t))z'(t). Equation (3.21) establishes the sought-after relation between the rates y' and x'. However, it seems somewhat different from (3.20): The rates are still proportional to one another, but the proportionality coefficient f'(x) is no longer a constant, but a function. How do we use it? Take a particular value of t = to. Let the values of x and y at t = to be zo = x(to) and yo = y(to). The number a = f'(zo) can be calculated. Then the equality y'(to) = az'(to) determines the relation between the rates y' and x' at the instance when x has the value zo (or y has the value yo = f (zo)). EXAMPLE 3.10. Let a laser pointer be positioned at a distance D = 1 m from a wall. The pointer can be rotated so that the bright spot created by the laser beam travels horizontally on the wall. (I) At what speed does the bright spot travel along the wall if the pointer revolves at a constant rate w rad/s? (II) At what direction of the laser beam does the bright spot travel at the speed v = 4,rm/s if w = , rad/s? Solution: (i) The analysis of any problem on related rates must begin with defining the quantities whose rates are being studied. In other words, one has to answer the question: How are these quan- tities measured? The orientation of the laser beam can be described by the angle p between the perpendicular to the wall and the laser beam. The position of the bright spot may be set by the distance y traveled by it from the point on the wall when y = 0, that is, when the laser beam is perpendic- ular to the wall. If the pointer rotates, the angle becomes a function of time, p = p(t), and so does the position of the bright spot, y =y(t). Thus, the question is about the relation between the rates y'(t) =v (the speed at which the bright spot travels) and p'(t) =w (the rate at which the pointer rotates). (ii) The next step is to find a function that determines the relation between the quantities of interest, that is, between the distance  84 3. RULES OF DIFFERENTIATION D FIGURE 3.1. A laser pointer is positioned at a distance D from a wall and rotates clockwise. Its beam makes a bright spot that moves with the speed v to the right along the wall. The laser beam direction is determined by the angle o and the position of the bright spot is determined by the distance y. y and the angle p: y = f(). It is clear that D and y are related as the catheti of the right triangle whose hypotenuse is the laser beam: y = D tan y = f(). (iii) Once the relation between the quantities of interest has been established, the relation between their rates can be found. Since (tan p)' = 1/ cos2 o, Equation (3.21) yields D D (3.22) y = Dtan - y' = c ' -- v = 2 O . cos 2 CO co2 The first question is answered. (iv) Note that the rate y' = v is not constant even if the rate o' = w is constant. To answer the second question, one has to find the value of o when v = 47 m/s, D = 1 m, and w = r rad/s. It follows from Equation (3.22) that Dw 1 r cos2 9 _ v9 _ v 4 3'  20. RELATED RATES 85 that is, the bright spot moves at the speed 47rm/s when the laser beam makes 600 with the perpendicular to the wall. Q 20.4. Can Anything Travel Faster Than Light? The solution (3.22) has an interesting feature. When c approaches 900, that is, the laser beam is getting closer to being parallel to the wall, the cosine, cos y, tends to 0 in Equation (3.22), and hence the rate y' = v grows unboundedly. It seems like just with merely a laser pointer, a superluminal object can be created in a lecture hall! Let us investigate this. The speed of light is c 300,000 km/s 186,000 mi/sec. The light can make a trip around the world in merely 0.13 seconds! Example 3.10 is now supplemented by two additional questions: (III) Is it possible that v can exceed the speed of light? If so, at which direction of the laser beam does it happen? (IV) At which position of the bright spot does it happen? The answers read: (III) Setting D = 1 m = 10-3 km (watch the units: all distances are now in kilometers!) and v = c = 3.- 105 km/s, the angle at which the bright spot exceeds the speed of light satisfies the equation cos2 co= Dw/c 1.05.- 10-8, and hence co 89.994140. So the bright spot becomes superluminal if co> 89.994140! (IV) Since y = D tan5, v > c if y > 9772 m. Well, a lecture hall appears to be a "bit" small for this experiment! Take a Dremel miniature grinder (sold in Lowe's stores) for which w r103 rad/s (it can be used to rotate the pointer), and set D = 0.1 m, then v > c if y > 98 m; not yet exactly a lecture hall experiment, but it can be managed on the campus! Einstein's theory of relativity states that no material object can travel faster than light. Has a counterexample to Einstein's theory just been found? The answer is "no." In the motion of the bright spot, no material object actually moves along the wall. Bright spots at y and y + Ay are created by different portions of the laser beam that are emitted by the laser at two distinct moments of time. A lump of light that arrived at y was reflected by the wall (that is why we see the bright spot!), and hence it could not appear at the next position y + Ay (at this position arrived a different lump of light emitted by the laser at a later time). So the rate Ay/At cannot possibly be associated with the motion of any material object along the wall.  86 3. RULES OF DIFFERENTIATION 20.5. Related Problem. The next time you watch a Florida sunset, look at your shadow. Does there exist a position of the Sun above the horizon at which your shadow extends faster than the speed of light? 20.6. More Than Two Related Rates. There are situations when several quantities are related among themselves. If these quantities become functions of a variable t, then their rates are linearly related. A proof of this statement is given in Calculus 3, where functions of several vari- ables are studied. However, the basic idea of finding relations between the rates has not changed: They are obtained by differentiating the relations between the quantities in question with respect to t. The procedure is illustrated in the following example. EXAMPLE 3.11. Consider a rectangle with sides x and y. Sup- pose that x and y change with time. Find their rates of change when x = 3 cm and y = 1 cm if, at that moment, the area of the rectangle decreases at a rate of 2 cm2/s while the perimeter does not change. Solution: (i) There are four quantities involved: the rectangle dimensions x and y, the area S, and the perimeter P. (ii) There are two relations between them: S = zy, P = 2(x + y). (iii) If x = x(t) and y = y(t), then S(t) = x(t)y(t) and P(t) 2(z(t) + y(t)). Using the derivative of the product and the sum of two functions, the linear relations between the rates are obtained S' = z'y + zy', P'= 2(z'+ y'). (iv) Since P' = 0 (the perimeter does not change), x' = -y' and S' = (x - y)y'. Now let S' = -2 cm2/s because S decreases (S' must be negative). With xc= 3 cm and y = 1 cm, one has -2 = (3 - 1)y' and y' = -1 cm/s. It then follows that ' =-y' =1cm/s. 20.7. Exercises. (1) Consider a triangle with vertices A, B, C such that |ABI ACI 2 cm. Let 0 be the angle at the vertex A. If the angle decreases at the rate 0.3 rad/s, what is the rate of change of the length of the side BC at the instance when 0 = g/3.  20. RELATED RATES 87 (2) Consider a triangle with vertices A, B, C such that |ABI = |ACl = 3 cm. Let 0 be the angle at the vertex A. If the angle increases at the rate 0.3 rad/s, what is the rate of change of the area of the triangle at the instance when 0 = r/3. (3) The sides of a rectangle change with time such that the area of the rectangle does not change. If the rate of change of one side is 1 m/s, find the rate of change of the other side at the instance when both sides are of equal length. (4) The sides of a rectangle change with time at the rates 1 m/s and -3 m/s. Find the rate of change of the diagonal of the rectangle at the instance when both sides are of equal length. (5) At what rate does the area of a disk increase at the instance when the radius is R = 10 cm if the radius increases at a constant rate of 2 cm/s? (6) At what rate do the area and the length of the diagonal of a rectangle change at the instance when one side is x = 20 m and the other is y = 15 m if the former is decreasing at a rate of 1 m/s, while the latter is increasing at a rate of 2 m/s? (7) Two ships, A and B, leave a harbor at the same time, one heading north and the other heading east. At what rate is the distance between the ships increasing if the speed of ship A is 30 km/h and the speed of ship B is 40 km/h? (8) The surface area of a ball is increasing at a constant rate of 4 m2/min. At what rate do the radius and volume of the ball change at the instance when the ball has radius 3 m? (9) A ladder 24 ft long leans against a vertical wall. If the lower end is being moved away from the wall at a rate of 3 ft/sec, how fast is the top descending when the lower end is 8 ft from from the wall? When are the lower and upper ends moving at the same rate? (10) A man 6 ft tall walks away from an are light 15 ft high at a rate of 3 miles per hour. How fast is the farther end of his shadow moving? How fast is his shadow lengthening? (11) The volume of a sphere is increasing at a rate of 16 cm3/s. How fast is the radius increasing when it is 6 cm? How fast is the surface area increasing when it is 36 cm2? (12) Sand is being poured on the ground from an elevated pipe and forms a pile that always has the shape of a circular cone whose height is equal to the radius of the base. If the sand falls at a rate of 0.5 m3/min, how fast is the height of the pile increasing when it is 2 m?  88 3. RULES OF DIFFERENTIATION (13) A particle moves along the curve defined by the algebraic equa- tion x2 - 2y3 = 9 so that the coordinate x increases steadily at a rate of 3 units of length per second. Find the rate of change of the coordinate y when the particle is at the point (x, y) = (5, 2). (14) The velocity of a particle moving along a straight line satisfies the condition v2 = c + 2b/s, where a and b are constants and s is the distance traveled by the particle. Show that the acceleration (the rate of change in velocity with respect to time) is a = dv/dt = -b/s2. (15) Consider two lines y+x= 2a and y -x = 0, where a is a num- ber. Suppose that a particle moves along the first line toward the point of intersection of the lines at a constant speed vi, while another particle moves along the second line in the direc- tion away from the point of intersection at a constant speed v2. Find the rate of change of the distance between the particles when the first and second particles are at the distances si and s2 from the point of intersection, respectively. In particular, what is the value of this rate if si = s2 and vi = v2? (16) The blades of a pair of scissors have width 2h. Find the rate at which the point of intersection of the edges of the blades is moving if the angle between the blades decreases at a con- stant rate w. Assume that the blades are attached by a screw through the midpoint of each blade (i.e., through a point that is at distance h from the edges of the blade). If h = 4 mm and w= -2 rad/s, how long should the blades be to see the point of intersection going superluminal? (17) If y2 = 2x and c is decreasing steadily at a rate of 0.25 units per second, find how fast the slope of the graph is changing at the point (x, y) = (8, -4). (18) A pool has a spherical bottom of radius R and the maximal depth h < R at the pool center. A man walks on the bottom of the pool toward the pool center at a constant speed v. Find the rate at which the man is submerging under the water. (19) Consider a rectangle with sides x and y. Suppose that x and y change with time. Find their rates of change when x = 3 cm and y = 1 cm if, at that moment, the area of the rectangle decreases at a rate of 2 cm2/s while the perimeter increases at a rate of 4 cm2/s. (20) Consider a planar region that is a sector of a disk with radius R and angle yp. Suppose that R and y change with time so that the area of the region does not change, while its perimeter  21. LINEAR APPROXIMATIONS AND DIFFERENTIALS 89 increases at a rate of 2 m/s. Find the rate of change of the angle c and the radius R at the moment when co= 300 and R=10 m. 21. Linear Approximations and Differentials 21.1. Tangent Line Approximation. The derivative of a function f(x) at a point x =0zo defines the slope of the line tangent to the graph y = f(x) at the point (zo, f(xo)) (see Equation (2.5)). The equation of the tangent line is y - f(zo) = f'(xo) or y = f(xo) + f'(xo)(x - o). - o DEFINITION 3.3. Suppose f(x) is differentiable at x = xo. The linear function (3.23) L(x) = f (xo) + f'(xo)(x - zo) is called the linearization of f(x) in a neighborhood of zo. Since the values of f and L coincide at x = zo, one might expect that the difference f(x) - L(x) is small, provided x is close enough to zo. More precisely, consider the limit lim f(x) - L(x) - limf - f(xo)-f(x)-f(x) - f'(xo)-0 XaXo 1: - 1:0 XaXo 1 - 1:0 where the definition of the derivative has been used. This shows that the error of the approximation of f(x) by L(x) decreases to zero faster than x - zo as x approaches xo: f (x) = L(x) + (x - xo) (x -xo), wherec (x - xo) - 0 as x - xo So the linear function L(x) may be used to approximate values of f(x) in a small neighborhood of zo. This approximation is called the lin- ear approximation or tangent line approximation. The concept of the tangent line approximation is illustrated in Figure 3.2. EXAMPLE 3.12. Use the linear approximation to estimate the value v3.92. Solution: (i) Consider f(x) / . The closest value of x to 3.92 at which the square root can be evaluated without a calculator is zo = 4: f~zo) =2. Note the two important steps here: the choice of f(x) suitable for the problem and the choice of zo0 near which the linear approximation is to be used.  90 3. RULES OF DIFFERENTIATION y = L(x) y = f(x) 8-xo xo 8+xo FIGURE 3.2. Tangent line approximation. In a neigh- borhood of xo (an interval [xo - 6, zo + 6]), the tangent line y = L(x) stays close to the graph y = f(x). By re- ducing the width of the interval 6, one can make the error E of the tangent line approximation as small as desired, i.e., If (x) - L(x) < for all x E [xo - b, xo + 6]. (ii) Since f'(x) = (x)' = 1/(20) and f'(4) = 1/4, by Equation (3.23) the linearization of x near x = 4 is L(x) = 2 + -(x - 4). 4 (iii) The linear approximation means that the value f(3.92) 3.92 is approximated by the value L(3.92): 1 3.92 r L(3.92) = 2 + -(3.92 - 4) = 1.98. A calculator gives 3.92 r 1.9799. So the approximation error is 3.92 - L(3.92) < 1.02. 10-4. It is easy to see that L(4.08) = 2.02 and 4.08 - L(4.08) < 1.02. 10-4. In notations given in the caption of Figure 3.2, this observation can be summarized by the following inequality: - L(x) < 1.02 . 10-4 = E if x - 4 < 0.08 = 6. In other words, the values of x and its linearization differ by no more than 1.02.10-4 for all 3.92 < x < 4.08. Naturally, a decrease (increase) in the upper bound for the error would lead to a decrease (increase) in the size of a neighborhood of x = 4 where the linear approximation is accurate.  21. LINEAR APPROXIMATIONS AND DIFFERENTIALS 91 21.2. Accuracy of the Linear Approximation. The previous example leads to a problem that is extremely important in applications: Given an up- per bound for the error E of the linear approximation of a function f (x) near zo, find b such that |f(x) - L(xm)| < Eif Ix-x0 l 5c, or, alternatively, given b, that is, the neighborhood zo- b x cc z 1, can always be neglected. So one might think of differentials as infinitesimal variations of variables. From this  94 3. RULES OF DIFFERENTIATION point of view, the definition (3.25) looks rather natural: Infinitesimal variations of two related variables must be related linearly as their higher powers can always be neglected. The concept of the differential becomes rather practical when one has to establish relations between variations of related quantities in situations when these variations may be viewed as infinitesimal. 21.5. Inverse Function and the Differential. The concept of the differ- ential offers a simple way to find the derivative of an inverse function. Suppose that a function f has the inverse g = f-1 and g is differen- tiable (conditions under which g exists and is differentiable are stated later in the inverse function theorem of Section 23). If y = f(x), then the differentials are related as dy = f'(x) dx. On the other hand, x = f-1(y) = g(y) and hence dxz= g'(y) dy. Since the ratio of the differentials is the derivative, it follows that dxz 1 1 1 dy - <- g'(f(x))= <-> g'(y)= . dy d f'(x) gY f'(gy) For example, f (x) = tan x and g(y) =tan-1 y. Then 1 1 1 f'(x)-cos 1+ (f (x))2 _ g'(Y)1+ f'(g(y)) 1 + y2 where the relation f(g(y)) = y has been used. 21.6. Related Errors. Every physical quantity is known only with a certain degree of accuracy. Errors are inherent in the very process of taking measurements. As a point of fact, a value of a physical quan- tity given without its measurement error does not make much sense; neither should one draw any conclusion from data without a proper analysis of the errors. One of the important practical applications of the differential lies in the error analysis of related quantities. Suppose there is a relation between two quantities y and x, y f (x). Let x be measured with an error. This means the following. After n measurements, one gets n values xi, x2, ... , x. The average is x - (xi+x2+- - -+xz)/n is viewed as the actual value. The measured values deviate from the average by amounts Ax1= x1 - x,"... , Az= xn - . If Ax = max{|AxI,..., lAA} (i.e., Ax is the maximal of the absolute errors), then all measured values lie in the interval [x- Ax, x+Az]. The quantity Ax is the maximum uncertainty in the value of x (or an error bound). One writes for the measured value x + Ax to indicate the average measured value and its maximum uncertainty. The number  21. LINEAR APPROXIMATIONS AND DIFFERENTIALS 95 Ax is usually known and determined by the very process of taking measurements. A standard question in error analysis: What is the accuracy of the value y = f(x)? Apparently, x and Ax are independent variables as the error bound Ax depends on the way in which the variable is measured (there are more and less accurate methods which would lead to smaller and higher values of Ax independently of the value of x). Naturally, one might assume that the errors are small; that is, they are infinitesimal variations of measured quantities. Then the errors of the related quantities must be related as their differentials. This is a standard assumption of the error analysis. In other words, if y = f(x) where x is the measured mean value, then the error in the value of y in each measurement is assumed to be Ayi = yi - y f (Xi) - f(x) = f (x + Axi) - f(x) = df(x) = f'(x)Axi which is nothing but dy = df(x) where dy =Ay if dx= Ax. The absolute value of the differential |dyl represents an absolute error of y = f(x). The quantity |dy/yl - 100% is called a relative error. The absolute and relative error bounds are, respectively Ay = I f'(x)|Ax, A 100% f ) Ax 100% y| f (x)| and one writes for the measured value y t Ay to indicate the maximum uncertainty in the value of y. EXAMPLE 3.14. What are the absolute and relative error bounds of the volume of a cube if its side is 10 +0.1 cm? Remark. When measuring the length by a ruler with a grid, the measurement error should not exceed the ruler grid spacing (e.g., a ruler with a millimeter grid). Solution: The volume V and side x are related as V = x3. So dV 3x2 dx. Setting dx= 0.1 cm and x = 10 cm, dV = 30 cm3 and V = 1000 + 30 cm3. The relative error bound is dV/V = 0.03 or 3% (note that dx/x 0.01, i.e., only 1%).D The error analysis for several related quantities is studied in multi- variable calculus courses. It is based on the concept of the differential of functions of several variables.  96 3. RULES OF DIFFERENTIATION 21.7. Exercises. (1) Find the linearizations of each of the following functions at the specified point: (i) cos x, x = 7/4 (ii) tan x, x= 0 (iii) ex2, X - 0 (iv) In x, x= e (v) 1+x, x= 3 (2) Estimate the error of the tangent line approximation of each of the following functions over an interval |x - xo < for the specified point xo and the width S: (i) 1 + z, zo = 3, 6 = 0.1 (ii) In x, x = 1, 5 = 0.2 (iii) tan x, x = 0, b = 7r/4 (3) Find the differentials of each of the following functions: (i) x(1 - X23 (ii) (y - 2)/(y + 1) (iii) 1 + x2/x (iv) sin2 t + cos(t2) (v) ln(x + 2) + sex (vi) a-1 tan-1(x/a), a / 0 (vii)lnx+ x2+a (viii) (2a)-1 In |(x - a)/(x + a) (4) Use differentials (the tangent line approximation) to estimate the following numbers and assess the accuracy of the estimates: (i) 24.6 (ii) e0.08 (iii) V1.02 (iv) sin 290 (v) tan-1 1.05 (vi) log1o 11 (5) Prove the approximation formula x na"+ x ~.a + a , a>0 and use it to calculate approximately the numbers (i) 39 (ii) 850 (iii) #100 (iv) 10 1000  21. LINEAR APPROXIMATIONS AND DIFFERENTIALS 97 (6) Calculate Af (1) and df (1) for the function f (x) = x3 - 2x + 1 and compare them in the following three cases: AOx= 1, Ax= 0.1, and Ax= 0.01. (7) Let u, v, and w be differentiable functions. Find dy if (i) y = uvw (ii) y = =/v2 (iii) y =(u2 + v2)-1/2 (iv) y = In /2 + v2 (v) y =tan-1(u/v) (8) Find dy in terms of x, y, and dx if (i) /+ y = 4 (ii) y3 +x3 = 2xy (iii) cos(x + 3y) = sin(xy) (9) Find an approximate formula for the area of a circular ring of radius r and width dr. What is the exact formula? (10) Find an approximate formula for the volume of a spherical shell of radius r and thickness dr. Assess the accuracy of the approximation by stating the condition on r and dr so that the relative error does not exceed e = 0.01 (i.e., 1%). (11) What is an admissible relative error in measurements of the radius of a ball in order for the relative error of the volume to be less than 1%? (12) A sector of a disk of radius R = 100 cm has an angle 0 = 600. How much is the area of the sector changed if (i) the radius R is increased by 1 cm? (ii) the angle 0 is decreased by 30'? Give the exact and approximate solutions. Compare them. (13) The period of a pendulum is determined by the equation l T = 2r -, where 1 is the length of the pendulum (in cm) and g = 981 cm/s2 is the free-fall acceleration. How much should the length 1 = 20 cm be changed in order to increase the period by 0.05 s? (14) To determine the free-fall acceleration, the period of a pendu- lum is measured so that by the above equation g =47r2/2/T2. How do the measured values of g vary if (i) T is measured with a relative error bound c? (ii) L is measured with a relative error bound c?  98 3. RULES OF DIFFERENTIATION (15) Find the absolute error of log10 x (x > 0) if the relative error of x is E. (16) Use differentials to find the derivatives of the inverse functions sin-1x and cos-1X. (17) Prove that the linearization of a differentiable function f(x) in a neighborhood of xo is unique in the sense that if L(x) b+rm(x - xo) and lim f (x) - L(x) = 0, zazxo x - z then b = f(xo) and m = f'(xo). In other words, the lineariza- tion is the only linear approximation whose error decreases to zero faster than x - x as x approaches xo. (18) Find the tangent line and the normal line (the line perpendic- ular to the tangent line) to the curve y = (x + 1) 3/3 - x at the points (-1, 0), (2, 3), and (3, 0). (19) Find the point(s) of the parabola y = 2+ x - X2 at which the tangent line is (i) parallel to the x axis and (ii) parallel to the line y = x. (20) Prove the relation X X2 a22+X=a+ -R, 0 0 and cc> 0.  CHAPTER 4 Applications of Differentiation 22. Minimum and Maximum Values Some of the most important applications of calculus are optimiza- tion problems. An example of an ancient optimization problem: A man can throw a stone at a speed of vo. At what angle should the stone be thrown in order to get the maximal range? An example of a modern optimization problem: How can one optimize the information flow in the World Wide Web to avoid crashes of servers? Many of these prob- lems can be reduced to finding the maximal and minimal values of a given function. DEFINITION 4.1 (Absolute Maximum and Minimum). A function f has an absolute maximum at c if f (z) < f (c) for all x in the domain D of f. Similarly, the value f(c) is called the maximum value of f. A function f has an absolute minimum at c if f(x) ;> f(c) for all x in the domain D of f. The value f(c) is called the minimum value of f. The maximum and minimum values of f are called the extreme values of f. For example, the function f(x) = cos x attains its maximum value 1 at x = 27n, where n = 0, +1, +2, ..., and its minimum value -1 at x = r+27rn. A function does not always have a maximum or minimum value. For instance, the function f(x) = 1/x defined for all real xz / 0 has neither maximum nor minimum value because, for any real M, one can always find x such that f (x) > M (0 < x < 1/IM). So no real number can be the maximum value of f (x). Similarly, for any real M, f (x) < M if -1/IMI < x < 0; that is, no minimum value exists. The function f (x) = x2 has no maximum value on the real axis, but it does have an absolute minimum at x = 0 because x2 > 0 for all x and f(0) = 0, that is, f(x) > f(0). 22.1. Relative Maxima and Minima. DEFINITION 4.2 (Local Maximum and Minimum). A function f has a local (or relative) maximum at c if f(cc) <; f(c) for all zc in some open interval containing c. Similarly, a function f has a local (or 99  100 4. APPLICATIONS OF DIFFERENTIATION relative) minimum at c if f(x) > f(c) for all x in some open interval containing c. EXAMPLE 4.1. Does the function f(x) = x3 - x - x(x2 - 1) have an absolute maximum (minimum) value and relative maxima (minima) on the real axis? Y I - _ -1L - FIGURE 4.1. Graph of the function f(x) = x3 - x x(x2 - 1). It does not have an absolute maximum or minimum value. However, it does have a relative maxi- mum at x = -1/ / and a relative minimum at x 1//3. Solution: 1. The function has neither an absolute maximum nor an absolute minimum because it grows unboundedly with increasing x and it decreases unboundedly as x attains larger negative values. 2. The function vanishes at three points x = 0, ±1. It can have relative minima and maxima between its zeros because the values of f are bounded from above and below: f(x) x 3 + x 2 for x < 1, that is, -2 < f(x) 2 if -1 < x < 1. 3. Consider the open interval x E (0, 1). The function is strictly negative in it and bounded from below: M < f(x) < 0 for all x E (0, 1) (e.g., M = -3). By increasing M, one can eventually reach the situation when there is 0 < c < 1 such that M = f(c) f(x) for all 0 < x < 1. This happens when the horizontal line y = M touches the graph y = f(x). Thus, f must have a relative minimum in (0, 1). Remark. The actual value c = 1//3. How is it obtained? There is a technique to find c, which will be studied shortly. 4. Similarly, f is strictly positive in (-1, 0) and bounded from above 0 < f(x) < M for some M. By lowering the horizontal line y = M (or decreasing M) to the point when it touches the graph y = f(x), one  22. MINIMUM AND MAXIMUM VALUES 101 can find a point c E (-1, 0) such that f(x) < f(c) for all x E (-1, 0); that is, f has a relative maximum in (-1, 0). Remark. The actual value is c = -1/v/ (see below). Q One of the lessons that can be learned from this example is that one can think of a relative minimum (maximum) as an absolute minimum (maximum) when f is restricted to a sufficiently small subset in its domain. This observation is accurately stated by the following theorem. THEOREM 4.1 (The Extreme Value Theorem). If f is a continuous function on a closed interval [a, b], then f attains its absolute maximum and minimum values in [a, b] ; that is, there exist c1 and c2 in [a, b] such that f(ci) < f(x) f(c2) for all x in [a, b]. ftx)J a c1 bJW FIGURE 4.2. Extreme value theorem. An example of a continuous function with several local minima and max- ima. The minimal value coincides with one of the lo- cal minima, while the maximal value is reached at the endpoint of the interval: f(c1) < f(x) < f(b) for all x E [a, b]. The hypothesis of the closedness of the inter- val is crucial. If the point b is excluded, then f has no maximal value on [a, b). The continuity hypothesis is essential. In fact, the continuity of f(x) = 3- x was implicitly used in Example 4.1 to establish the existence of its relative maximum and minimum! The following exam- ple illustrates the point. Consider the function f~x) =2x if x E [0, 1) and f(x) =1 if x E [1, 2]. So the function is defined on the closed  102 4. APPLICATIONS OF DIFFERENTIATION interval [0, 2] and bounded from above f(x) < M (e.g., M > 2). An attempt to establish the existence of a maximum value of f by lowering M fails! Indeed, the lowest upper bound is M = 2, but there is no c such that f(c) = 2. The values of f approach 2 as x approaches 1 from the left, but f(1) = 1! For any positive c > 0, f(1 - E) < f(x) for x E (1 - E,1) no matter how small e is. Thus, f does not have an abso- lute maximum value because of its discontinuity at x = 1. The absolute minimum exists: f(0) < f(x). Note that the function f(x) = 2x when c E [0, 1] and f(x) = 1 when c E (1, 2] has an absolute maximum and minimum, f(0) f(xc) f(1), despite its discontinuity at x = 1. So the continuity hypothesis is a sufficient condition, but not necessary. The hypothesis of the closedness of the interval is also a sufficient condition, but not necessary. The continuous function f(x) = x does not attain its absolute maximum or minimum value on any open in- terval (a, b). But it does so if the interval becomes closed: f(a) < f(x) < f(b) for any c E [a, b]. On the other hand, the continuous function f(x) = x3 - x in the open interval (-1, 1) attains its absolute maximum and minimum value as one can see in Figure 4.1. 22.2. Derivatives at Local Maxima and Minima. The second observation resulting from Example 4.1 is that at the point where a continuous func- tion attains its local minimum or maximum value there is a horizontal line that touches the graph of this function. So, if, in addition, the func- tion is differentiable, then this horizontal line is a tangent line with the vanishing slope; that is, the derivative of the function vanishes at points where the function attains its local maximum or minimum value. THEOREM 4.2 (Fermat's Theorem). If f has a local maximum or minimum at c, and if f'(c) exists, then f'(c) = 0. PROOF. By the existence of f'(c) lim f(c+h) - f(c) , hi =f'(c). h~0 h Therefore, the right and left limits must coincide with f'(c) (see Section 7.3): ( l \ im. f (c + h) - f(c) f'( f(c + h) - f (c) h~- h h-+ h Let f have a local maximum (the case of a local minimum can be treated similarly). Then f(c) ;> f(cc) or f(cc) - f(c) 0 in some open interval a < cc < b. In particular, [f(c + h) - f(c)]/h 0 for any  22. MINIMUM AND MAXIMUM VALUES 103 positive h > 0 such that c < c + h < b. By Theorem 2.2, (4.2) f (c + h) - f(c) _ f (c + h) - f(c) h 0-o+ h Similarly, for any negative h < 0 such that a < c + h < c, one has [f(c + h) - f(c)] < 0 and [f(c + h) - f(c)]/h> 0. Hence, () f(c +h) - f (c) 0< m f (c +h) - f (c) h n-o- h By inequalities (4.2) and (4.3), it follows from (4.1) that 0 < f'(c) 0, which is only possible if f'(c) = 0. D This theorem provides a powerful tool to determine the actual po- sitions of local maxima and minima. Let us go back to Example 4.1 (f(x) = x3 - x). The slope f'(x) = 3x2 - 1 vanishes at two points x = +1/v3. According to the analysis carried out in Example 4.1, f has a local maximum at x = -1/v/ 3E (-1, 0) and a local minimum at x = 1 /v/53_E C(0, 1).- DEFINITION 4.3. A number c in the domain of a function f is said to be a critical point of f if either f'(c) = 0 or f'(c) does not exist. Does the equation f'(x) = 0 determine all local maxima and minima of f? (I) A function may have a local minimum or maximum at a point where the derivative does not exist. A simple example is the function f(x) =Ic. It has an absolute minimum at x = 0, but f'(x) does not exist at x = 0. So this minimum cannot be found from f'(x) = 0. (II) If f is differentiable everywhere, then, by solving f'(x) = 0, all local minima and maxima can be found. However, not all the solutions generally correspond to either a local maximum or a local minimum. The function f (x) = x3 has no minimum or maximum, but its derivative f'(x) = 3x2 vanishes at x = 0. In other words, the converse of Fermat's theorem is false. (III) If all critical points of a function are found, then their type (local maximum, local minimum, or none of the above) can be analyzed by comparing values f(c + h) with f(c), where c is a critical point (cf. Definition 4.2). If f"(c) exists, then the second derivative test can be used, which is discussed later.  104 4. APPLICATIONS OF DIFFERENTIATION (IV) A function defined on a closed interval [a, b] can have its ab- solute maximum or minimum at the endpoints. When finding the absolute maximum and minimum values, the values of f at the critical points must be compared with f(a) and f(b). The largest (smallest) of them is the absolute maximum (min- imum) value. D EXAMPLE 4.2. If a stone is thrown at a speed vo m/s and an angle 0 with the horizontal line, then its trajectory is a parabola: (4.4) y=xtan08- x2 g 2vo cos2 0' where y is the stone height (vertical position), z is the horizontal posi- tion (all the positions are in meters), and g = 9.8 m/s2 is a constant universal for all objects near the surface of the Earth (the free-fall ac- celeration). This is a consequence of the Newton's second law. At what angle should one throw a stone to reach the maximal range at a given speed vo? Solution: 1. The range as a function of the angle 0 has to be found first. The stone lands when its height y vanishes. The equation y = 0 has two solutions x = 0 (naturally, this is where the stone was thrown) and x = L(O), where L(0) = v°tQanQcost2 8 2v sin 6cosO 6=vsin(20). g g 9 2. The range L(0) is a differentiable function of 0 so the values of 0 at which L attains its extreme values may be found from the equation 2 L'() = 0 v 2 cos(2) = 0 - cos(2) = 0. 9 This equation has countably many solutions 2 w= /2 + ern, where n is any integer. But in the interval of the physical values of 0 E [0, 7/2], it has only one solution w= /4. Since sin(2w/4) = 1 (the absolute maximum of the sine), L attains its maximum value at w= /4. So the range is maximal, Lmax =v2/g, when a stone is thrown at 45 . E Remark. The conclusion in the preceding example is independent of the stone's mass and its initial speed v0. In reality, for larger values of v0, like a projectile shot by a gun, trajectory would deviate from the parabola (due to friction with the air that increases with increasing the speed). So the optimal angle would deviate from w/4. The devia- tion would also depend on the mass and the initial speed. The range  22. MINIMUM AND MAXIMUM VALUES 105 optimization problem becomes more involved and would require the theory of differential equations. It should also be noted that the angle at which the maximal range is attained depends on the initial height at which the stone is thrown. So the angle would be different from 450 when, for example, the stone is thrown from a cliff. 22.3. Exercises. (1) Examine the following functions for maxima and minima. Draw the graph in each case. (i) y= 2 + x -x2 (ii) y = (x - 1)3 (iii) y = (x + 1)4 (iv) y = x2 - 5x + 3 (v) y = 2x3 - 3x2 + 6x - 3 (vi) y x2 + 16/c (vii) y = x2 - 1/c2 (viii) y = 4x/(x2 + 1) (ix) y = sinc + cosx (x) y =ccex (xi) y = z"(1 - ")m, where n and m are positive integers (xii) y = 1/3(1 - X)2/3 (2) Find all critical points of the following functions and determine whether there is a local maximum, a local minimum, or none of the above at each critical point. (i) y =I - 3| (ii)y =I2 _-4| + 2x (iii) y = I -1 (iv) y = | sin(2x) (v)y cc 3ex (vi) y (x - 1)1/3 (vii) y = x(x + 1)2/3 (viii) y = (1 - X2)3/2 (3) Find the extreme values of the following functions on the spec- ified interval or show that such values do not exist. (i) y =X4 - 4x2, -3 , 0 f(a) for some x E (a, b) (cf. Example 4.1 for x E [-1, 0]). Since f is continuous, the extreme value theorem applies, and therefore f has a maximum in [a, b]. Since f(a) = f(b), the maximal value must be attained at c E (a, b). By Fermat's theorem, f'(c) = 0 because f is differentiable in (a, b). 3. If f(x) < f(a) for some x E (a, b) (cf. Example 4.1 for x E [-1, 1] or x E [0, 1]), then, by the extreme value theorem, f has a minimum at c E (a, b), and, by Fermat's theorem, f'(c) = 0. Rolle's theorem is also useful to analyze the root pattern of a function. E XAMPL E 4.3. How many real roots does the equation z5 +z - - 1 =0 have? Solution: 1. Let f(x) =x5 + x3 - 1. Evidently, f(-1) =-4 < 0 and f (1) =2 > 0. By continuity, f has to t ake all intermediate values  108 4. APPLICATIONS OF DIFFERENTIATION between -4 and 2 (the intermediate value theorem). So f has at least one root in (-1, 1). 2. Suppose it has two roots a and b, that is, f(a) = f(b) = 0. Then, by Rolle's theorem, f'(x) has to vanish somewhere in (a, b). But this is not possible because f'(x) = 5x4 + 3x2 + 1 > 0 for any x. Thus, f has the only real root. D THEOREM 4.4 (The Mean Value Theorem). Let f be a function that satisfies the following hypotheses: (I) f is continuous on the closed interval [a, b]. (II) f is differentiable on the open interval (a, b). Then there is a number c E (a, b) such that f(b) - f(a) (4.5) f'(c) (b-a or f (b) - f (a) = f'(c)(b - a). The geometrical interpretation of the theorem is simple (see Figure 4.4). Consider the line through the points (a, f (a)) and (b, f (b)). Its slope is (f(b) - f(a))/(b - a). The theorem asserts the existence of a point where the graph y= f(cx) has a tangent line with the same slope (cf. Equation (4.5)) (as f'(c) is the slope of the tangent line at x = c). Let us turn to a formal proof. PROOF OF THEOREM 4.4. 1. Consider the line through the points (a, f(a)) and (b, f(b)). Its equation is (4.6) y = L(x) = f (a) +-f(b f(a (x - a) , L(a) = f (a) , L(b) = f (b) . Next, consider the function f(b) -f(a) (4.7) h(x) = f(x) - L(x) = f(x) - f(a) - b _a(x - a). Its values determine the deviation of the graph y= f(cx) from the se- cant line y = L(x) on the closed interval [a, b]. 2. The function h(x) satisfies the three hypotheses of Rolle's theorem. First, it is continuous on [a, b] as the sum of two continuous functions f(cc) and -L(c) (a linear function is continuous). Second, it is differ- entiable on (a, b) as the sum of two differentiable functions: /f(b) -f(a) (4.8) h'(c)=f'(c)- b- ccE (a, b) .  23. THE MEAN VALUE THEOREM 109 y (b, f(b)) 'N (a, f(a)) FIGURE 4.4. Mean value theorem. The secant line of the graph of f through the points (a, f(a)) and (b, f(b)) has the slope tan a = [f(b) - f(a)]/(b - a), where a is the angle between the secant line and the horizontal line. If f does not coincide with the secant line, then near x = a the slope of the tangent line does not coincide with tan a. Here the case when this slope is greater than tan a is shown. Then the graph of f lies above the secant line near x = a. But the graph has to return to the secant line again. Near the point where the graph and the secant lines meet again, the tangent line has to have a smaller slope than tan a. So at some point c the tangent line has to be parallel to the secant line, meaning that f'(c) tan a. Finally, by (4.6) and (4.7), h(a) = f(a) - L(a) = 0 and h(b) = f(b) - L(b) = 0, that is, h(a) = h(b). 3. By Rolle's theorem, there is a number c E (a, b) such that f(b) -f(a) h'(c) =0 - f'(c) = b)-fa where Equation (4.8) has been used.D E XAMPLE 4.4. A speeding car was pulled over on an interstate road and a state trooper gave a warning to the driver. Forty five minutes later and passed 65 miles on the road, the car stopped at a rest area.  110 4. APPLICATIONS OF DIFFERENTIATION Another state trooper approached the driver and issued a speeding ticket, claiming that the driver ignored the warning and exceeded 86 miles per hour. Was the trooper's claim correct? Solution: Let s(t) be the distance traveled by the car after it was pulled over the first time. The rate of change s'(t) = v(t) is the speed of the car at any moment of time. The function s(t) is defined between t = 0 and t = 45 min = 0.75 hr so that s(0) = 0 and s(0.75) = 65 mi. It is differentiable as s'(t) is the car speed! By the mean value theorem, there is a time moment t = c E (0, 0.75) when (=s(0.75) - s(0) 65 86.7mi/hr. 0.75 - 0 0.75 The speeding ticket is justified. D For any two moments of time a and b, the ratio (s(b) -s(a))/(b-a) is the average speed on the time interval [a, b]. The mean value theorem simply states that a moving object always attains its average speed at least at one moment of time between a and b. So, if at time moment b the object appears to be traveling slower than its average speed, prior to that it must have been traveling faster than its average speed. EXAMPLE 4.5. Suppose the derivative f' exists and is bounded on (a, b), that is, m < f'(x) M. If f(a) is given, how small and how large can f(b) possibly be? Solution: By the mean value theorem, there is a c E (a, b) such that f(b) = f (a) + f'(c)(b - a). Since m < f'(c) < M, f (a) + m(b - a) f (b) f (a) + M(b - a). This equation is easy to understand with the help of a mechanical analogy: How far can a car travel in time b - a if its speed is not lower than m, but cannot exceed M? D 23.1. Properties of the First Derivative. The derivative of a constant function vanishes. How about the converse? The following theorem answers this question. THEOREM 4.5. If f'x) = 0 for all x in an interval (a, b), then f is constant on (a, b). PROOF. Take any two numbers x1 and x2 between a and b. By the mean value theorem, there is a number c between x1 and x2 such that f~zi) - f(x2) =f'(c)(z1 - xc2). By hypothesis, f'(c) =0 for any c.  23. THE MEAN VALUE THEOREM 111 Thus, f(xi) - f(x2) = 0 or f(xi) = f(x2) for any x1 and x2 in (a, b); that is, f is constant. D The hypothesis that f'(x) = 0 in a single interval is crucial. For example, the sign function f(x) = 1 if x > 0, and f(x) = -1 if x < 0, has zero derivative at any point of its domain, but it is not constant. The key point to note is that the domain is not a single interval, but a union of two disjoint intervals (-oo, 0) and (0, oc). So the mean value theorem is not applicable to any interval containing x = 0. This example is easily extended to the case when the domain is any collection of disjoint intervals and f takes different constant values on different intervals. COROLLARY 4.1. If f'() = g'(x) for all x in an interval (a, b), then f - g is constant, that is, f (z) = g(x)+ k, where k is a constant. PROOF. Let h(x) = f(x) - g(x). Since h'= f' - g'= 0 in (a, b), h is constant, and the conclusion follows. D The sing of the first derivative defines intervals of growth and de- crease of a function. THEOREM 4.6 (Increasing-Decreasing Test). (I) If f' > 0 on an interval, then f is increasing on that interval. (II) If f' <0 on an interval, then f is decreasing on that interval. PROOF. Take any two numbers x1 and x2 in the interval so that x1 < x2. A function is increasing if f(xi) < f(x2) and decreasing if f (Xi) > f(X2). Since f is differentiable, the mean value theorem states that there is a number c between x1 and x2 such that (4.9) f (X2) - f (zi) = f'(c)(X2 - Xi) . If f' > 0, then it follows from (4.9) that f(x2) - f(zi) > 0 because, by assumption, x2 > xi; that is, the function is increasing. Similarly, for f' < 0, f(x2) - f(xi) <0, and the function is decreasing. Q The function f is said to be monotonic in an interval if it is increas- ing or decreasing in this interval. By the increasing-decreasing test, the function is monotonic in an interval if its derivative f' does not vanish in the interval. The increasing-decreasing test is further illustrated on the interac- tive website at http: //www.math.ufl.edu/~mathguy/ufcalcbook/inc_ dec. html.  112 4. APPLICATIONS OF DIFFERENTIATION 23.2. The Inverse Function Theorem. A Baby Version. Given a function f, its inverse function exists if f is one-to-one as explained in Section 5. A simple rule to calculate the derivative of the inverse function was pre- sented in Sections 17 and 21. However, the very question of whether the inverse function is actually differentiable has not been addressed. It ap- pears that if f is differentiable, then the questions about the existence of the inverse function f-1 and its differentiability can be answered by looking at the sign of the derivative f'. THEOREM 4.7. (A Baby Version of the Inverse Function Theorem). Let f be a function on -oc 0 (or f'(x) <0) for all x e (a, b). Then f has the inverse g = f-1 on (c, d) for some -oo 0. The other case is similar. By the increasing- decreasing test, f(xi) < f(x2) for any a < x1 < x2 < b. Therefore, X1 / x2 implies that f(xi) # f(x2) and hence f is one-to-one. So f has the inverse g = f-1. 2. The function f is continuous on (a, b) because it is differentiable on (a, b). By the intermediate value theorem, f takes all intermedi- ate values between f(xi) and f(x2) > f(xi) for any interval [Xi, X2] in (a, b). This shows that the range of f is a single interval (c, d) for some -oo < c < d < c and if f [Xi, x2] -I [f(zi),f(x2)], then g : [f(xi), f (x2)] - [Xi, x2] for any [xi, x2] c (a, b). 3. To show that g is differentiable, fix x E (a, b) and y= f(x) E (c, d). Put Af = f(x + Ax) - f (x). Then Af /Ax - f'(x) as Ax - 0. Any interval I with endpoints x and x+ Ax is mapped by f onto an interval I' with the corresponding endpoints f (x) = y and f(x+ Ax) = y+Ay, where Ay =Af. Put Ag = g(y + Ay) - g(y). Since g maps the in- terval I' onto I, Ag =Ax. The limit Ay - 0 implies that Ax - 0. Therefore, Ag/Ay = Ax/Ay = 1/(Ay/Ax) -- 1/f'(x) as Ax 0, which shows that g'(y) exists, and g'(y) = 1/f'(x), where y = f(x). A graphic illustration is given in Figure 4.5. D For example, the exponential function f(x) =ex has the derivative f'(x) =ex > 0 for all -oo < cc < oc. The image of the interval (-oc, oc) is (0, oc). So the exponential function has the differentiable inverse on (0, 00), which is, of course, the natural logarithm ln cc. The function f(x) =cos x has the derivative f'(x) =- sin cc, which is  23. THE MEAN VALUE THEOREM 113 = f '(x) Of .. . . . . . . . . . . . ............... Ax FIGURE 4.5. Inverse function theorem. An increasing function f, f' > 0, is one-to-one and hence has the in- verse g = f-1. The graphs of f and g are obtained from one another by the reflection about the line y = x. A se- cant line of the graph of f with the slope tan c = Af /Ax is mapped on the secant line of the graph of g with the slope tan, = Af--1/Af by this reflection. The an- gles c and , are related as ca + / = r/2 and hence tan, = 1/tancv. In the limit Ax -- 0, which also implies Af - 0, the secant lines become the tangent lines so that tan cv -- f'(x) and tang -- g'(y), where y = f (x). Hence, g'(f (x)) = 1/f'(x). negative on, for example, (0, r). Since the image of (0, r) is (-1, 1), the inverse cos- x exists and is a differentiable function on (-1, 1). 23.3. Exercises. (1) Verify Rolle's theorem for the function f(x) = (x - 1)(x - 2)(x - 3). (2) The function f(x) = 1 - 37 vanishes at x = a - 1 and x = b = 1; nevertheless, f'(x) $ 0 in the interval (-1, 1). Does this example contradict Rolle's theorem? Explain.  114 4. APPLICATIONS OF DIFFERENTIATION (3) Is the following assertion true? If so, prove it. If a function f has a derivative at each point of an open interval (a, b), if f is continuous at a and at b, and if f (a) = f (b) = 0, then there is a point c, with a < c < b, such that f'(c) = 0. (4) Is the following assertion true? If so, prove it. If a function f has a derivative at each point of an open interval (a, b) and if f is continuous at a and at b, then there is a point c, with a < c < b, such that f(b) = f(a) + (b - a)f'(c). (5) Let f be a function whose domain of definition is the closed interval [a, b] and which is differentiable at each point of this interval. Show that if f'(a) > 0, then f(a) is a relative mini- mum value of f. Remark. The derivative at a is defined as the right limit, f /(a limf (a + h)- f (a) h-0o+ h (6) Let f be a function whose domain of definition is the closed interval [a, b], which is continuous at a and b, which is differ- entiable in the interval (a, b), and for which f'(a) = 0 or does not exist. Show that if there is a point c such that f'(x) > 0 for a < x < c, then f(a) is a relative minimum value for f. Give examples of functions with nonexisting f'(a) to illustrate this result. (7) State and prove results similar to those given in the two previ- ous exercises for relative maxima instead of minima and also for b instead of a. (8) Investigate whether each of the following equations has solu- tions. If so, for each solution, find an interval in which no other solution lies. (i) cos x = X2 (ii) ex = 4 - x4 - x2 (iii) In x= 4 - x2 (iv) 2 tan x =tan-1 x (v) X6 + 3x4 + 3x2 + x - 7 =0 (9) Find the point(s) on the curve y = x3 at which the tangent line is parallel to the secant line through the points (-1, -1) and (2, 8). (10) Does the mean value theorem apply to the function f(cc)= 1/cc, a 1 (iii) | tan-1 a -tan-1 b| < |a - b| (15) Suppose that f'(x) = m = const for -oc < x < oc. Show that the only function that satisfies this condition is a linear function f(x) = b+ mx. Can one claim that the above asser- tion holds if the range of x is restricted to a disjoint interval? Explain. (16) Show that if (i) the function f(x) has the second derivative f"(x) in an interval [a, b] and (ii) f'(a) = f'(b) = 0, then there is a point in the interval (a, b) such that 4 If"(c)| ( )2 If(a) - f (b)|. (b - a)2 (17) Use the result of the previous exercise to show that if a car begins to move from some initial position and finishes the ride in t seconds, passing the distance s meters, then the absolute value of the car's acceleration is no less than 4s/t2 m/s2 at some moment of time. (18) Find intervals in which the given function f is increasing or decreasing. (i) f(x) =2 + cc- x2 (ii) f (x) =3cc - cc3  116 4. APPLICATIONS OF DIFFERENTIATION (iii) f (x) = 2x3 - 9x2 + 12j (iv) f (x) = 2x/(1 +9X2) (v) f (x) V/x/(x + 100), x (vi) f (x) = cos2 x + 1 (vii) f (x) = cos(7r/x) (viii) f (x) = sin x +4/2 (ix) f (x) ex + 3e-x + 2x (x) f (X) - §ve-x, n> 0, c (xi) f (X) =X2 - ln(x2) (xii) f(x) =X22-x (19) Show that the function (1 (-oo, -1) and (0, oo). (20) Is the derivative of a monoto the example f (x) = x + sin (21) Show that the function + x-1)X increases in the intervals )nic function monotonic? Consider f(x) =x+cx22sin(- ifxc# 0 and f(0) = 0 is increasing at x = 0 (f'(0) > 0) but is not increasing in any interval (-a, a), where a > 0 can be arbitrarily small. Graph the function. (22) Suppose that f(x) and g(x) are differentiable so that f(a) g(a) and f'(x) > g'(x) for x > a. Show that f(x) > g(x) for x > a. (23) Use the result of the previous exercise to establish the inequal- ities: (i) ex > 1+ cc, cc / 0 (ii) x - 2x2 < ln(1+ z) 0 (iii)x- 6x3 0 (iv) tancc> cc + 1cz3, 0< cc < Illustrate the above inequalities graphically. (24) Find intervals in which the function f has a differentiable in- verse f-1. (i) f (x) = 2x3 - 9x2 + 12x + 1 (ii) f (X) X2 + 16/c (iii) f (x) = ex + e-x 24. The First and Second Derivative Tests Suppose the critical points of a function f are known. If f is differ- entiable, then all critical points can be found by solving the equation f'(x) = 0. How can one figure out the nature of a critical point, that is,  24. THE FIRST AND SECOND DERIVATIVE TESTS 117 whether it is a local maximum, local minimum, or none of the above? It turns out that this question can be answered by studying the deriva- tives f' and f". In addition, many qualitative features of the graph y = f(x) can be deduced from properties of the derivatives of f. 24.1. The First Derivative Test. By the increasing-decreasing test, f is increasing on interval if its derivative is positive, and f is decreasing on an interval if its derivative is negative. Suppose f' is continuous such that f'(a) = m and f'(b) = M. Then, on the interval [a, b], f' must take all intermediate values between m and M. Suppose m < 0 and M > 0 or m > 0 and M < 0, that is, the derivative changes its sign on the interval [a, b], then f' must vanish between a and b. This means that f has a critical point a < c < b, f'(c) = 0. More to the point, if the derivative f' changes from negative to positive at c, then, according to the increasing-decreasing test, the function f changes from increasing to decreasing at c, that is, f(c - h) < f(c) and f(c) > f(c + h) for some small positive h. We can then conclude that f attains its local maximum at c. Similarly, if the derivative f' changes from negative to positive at c, then f changes from decreasing to increasing at c, f(c - h) > f(c) and f(c) < f(c + h), and hence f attains its local minimum at c. Naturally, there is a possibility that f'(c) = 0 but f'(x) does not change its sign at c. In such a situation, the increasing-decreasing test yields f(c - h) < f(c) < f(c + h) or f(c - h) > f(c) > f(c + h); that is, in either case the function f has neither a local minimum nor a local maximum. The findings are summarized in the following theorem. THEOREM 4.8 (The First Derivative Test). Suppose that c is a crit- ical point of a continuous function f. (I) If f' changes from positive to negative at c, then f has a local maximum at c. (II) If f' changes from negative to positive at c, then f has a local minimum at c. (III) If f' does not changes its sign at c, then f has neither a local maximum nor a local minimum at c. It is important to note that the very existence of f' at c is not required in the first derivative test. Recall the definition of a critical point (f'(c) =0 or f'(c) does not exist). In fact, in the preceding proof of the first derivative test, the condition f'(c) =0 can be dropped because all that is needed to apply the increasing-decreasing test is the sign of the derivative f'(xc) for x < c and x > c. For example,  118 4. APPLICATIONS OF DIFFERENTIATION f(x) =I. Then f'(x) = -1 for x < 0 (the function is decreasing) and f'(x) = 1 for x > 0 (the function is increasing). Hence, f(x) has a minimum at x = 0, even though f' does not exist at x = 0. The continuity hypothesis is also crucial. Consider the function f(x) = 1/z2 for x / 0 and f (0) = 0. Then f'(x) = -2/x3 for x / 0 and f'(0) does not exist. So x = 0 is a critical point. The function is increasing for x < 0 because f' > 0, and it is decreasing for x > 0 because f' < 0. However, f has no maximum at x = 0 because f is discontinuous at x = 0. In fact, it attains its absolute minimum at x = 0. The first derivative test is summarized in the following table. The first derivative test Case sign f', c < c signf', cc> c Conclusion I + + No extremum II + Maximum III + Minimum IV No extremum There are plenty of mechanical analogies of the first derivative test. Let H(t) be the height (relative to the ground) of a stone thrown upward as a function of time t. At the beginning, the stone moves upward so H' > 0 (the height is increasing). When the stone comes back to the ground, it moves downward so H' < 0 (the height is decreasing). Naturally, at some moment of time, the stone has to reach the maximal height. Analyze the motion of a pendulum (or a see-saw) from this point of view! The height would have two maxima and one minimum. EXAMPLE 4.6 (Example 4.1 Revisited). Find all local maxima and minima of f(x) = x3 - x and the intervals on which the function is increasing or decreasing (the function is depicted in Figure 4.1). Solution: 1. Since f is differentiable (it is a polynomial), all its critical points satisfy the equation f'(x)=32 -13(c-1/v/)(c+1/v)=0. Hence, the critical points are ci1 -1/v/5 and c2 =1/v 3. 2. For cc < ci, the product (cc- ci)(cc- c2) is positive (as the product of two negative numbers), and hence f' > 0 (f is increasing on (-oo, ci)). For ci < cc < c2, the product (xc- ci)(xc- c2) is negative (as the product  24. THE FIRST AND SECOND DERIVATIVE TESTS 119 of a negative and positive number), and hence f' <0 (f is decreasing on (ci, c2)). For x > c2, the product (x - ci)(X - c2) is positive (as the product of two positive numbers), and hence f' > 0 (f is increasing on (C2, 0C)). 3. The derivative changes from positive to negative at c1. Therefore, f has a local maximum at c1. The derivative changes from negative to positive at c2. Therefore, f has a local minimum at c2. D 24.2. Properties of the Second Derivative: Inflection Points. DEFINITION 4.4 (Concavity). The graph of a function f is called concave upward on an interval I if it lies above all of its tangent lines on I. The graph is called concave downward on I if it lies below all of its tangent lines on I. Note that the notion of concavity implies that f is differentiable (otherwise, the tangent lines do not exist). If f is twice differentiable, then the concavity is determined by the sign of the second derivative f". Suppose that the graph of f is concave upward on I. Consider the tangent lines at two points c and c + h in I: L1(x) = f (c) + f'(c)(x - c) , L2(X) = f (c + h) + f'(c + h)(x - c - h). The graph of f lies above the lines L1 and L2, that is, f(x) - L1(z) > 0 and f(x) - L2(x) > 0 for all x in I. Putting x = c in the last inequality and x = c + h in the former one, we obtain f (c) - L2(c) = f (c) - f (c + h) + f'(c + h)h > 0, f (c + h) - L1(c + h) = f (c + h) - f (c) - f'(c)h > 0 The sum of the right-hand sides of these inequalities is positive as the sum of two positive numbers: (4.10) h[f'(c + h) - f'(c)]> 0 - f'(c+h)-f(c) 0, h where the first inequality has been divided by a positive number h2. Inequality (4.10) is true for any h. Therefore, by taking the limit h - 0, we can conclude that f"(c) > 0 if the graph is concave upward. Inequality (4.10) shows that f'(c + h) > f'(c) for h > 0 and f'(c) > f'(c+h) for h < 0. In other words, the derivative f', or the slope of the tangent line of the graph of f, increases for the upward concavity, and hence (f')' =f" must be positive by the increasing-decreasing test. Similarly, the downward concavity implies that f" is negative. It turns out that the converse is also true.  120 4. APPLICATIONS OF DIFFERENTIATION THEOREM 4.9 (The Concavity Test). Let f be twice differentiable on an interval I. (I) If f"(x) > 0 for all x in I, then the graph of f is concave upward on I. (II) If f"(x) < 0 for all x in I, then the graph of f is concave downward on I. y y f "(x) > 0 f"(x)< 0 >M-x - >x xo xo FIGURE 4.6. Concavity near a point xo at which f"(zo) = 0. The graph is concave downward if f"(x) > 0 for x > xo and x < xo (left panel). Such a local behavior can be illustrated by f(x) = c + (x - xo)4, where c is a constant, so that f"(x) = 12(x - zo)2 > 0. The graph is concave upward if f"(x) < 0 for x > xo and x < xo (right panel). Such a local behavior can be illustrated by f(x) = c - (x - zo)4 so that f"(x) -12(x - zo)2 <0. How does the graph of f look near a point c where f"(c) = 0? There are four possibilities. First, f"(c + h) > 0 for some small h > 0. This means that the graph is concave upward to the left and right of c. As an example, consider f (x) = x4. Second, f"(c + h) < 0. This implies that the graph is concave downward to the left and right of c. As an example, take f(x) =_-z4. These two cases are depicted in Figure 4.6. Third, f"(c - h) > 0 and f"(c + h) < 0, that is, the concavity changes from upward to downward (e.g., f(x) -cc3). Fourth, f"(c - h) < 0 and f"(c + h) > 0, that is, the concavity changes from downward to upward (e.g., f(cc) =cc3) DEFINITION 4.5 (Inflection Point). A point P on the graph y f(xc) is called an inflection point if f is continuous there and the graph  24. THE FIRST AND SECOND DERIVATIVE TESTS 121 changes from concave upward downward to concave upward. to concave downward or from concave y x xo FIGURE 4.7. Concavity near a point x0 at which f"(xo) = 0 (the case not depicted in Figure 4.6). An in- flection point. The second derivative changes its sign at the inflection point x = x0. The concavity of the graph of f also changes at the inflection point. Such a local behav- ior can be illustrated by f(x) = c + a(x - xo) + (x - z0)3, where c and a are constants, so that f"(x) = 6(x - xo) changes its sign at xo. Note that f'(xo) = a; i.e., a de- fines the slope of the graph at x = x0. Let c be a critical point of f. Suppose f" is continuous near c. What can f"(c) tell us about the nature of the critical number (local minimum or maximum)? There are three possibilities. First, f"(c) > 0. This means that f"(x) > 0 for all x in some neighborhood of c (by the continuity of f"). Hence, f is concave upward near c; that is, its graph lies above the tangent line at c, which is a horizontal line because f'(c) = 0. So f must have a local minimum. Similarly, the condition f"(c) < 0 implies that the concavity is downward near c and f has a local maximum. If f"(c) = 0, then the concavity may or may not change at c as discussed earlier. The function may have a local maximum, a local minimum, or an inflection point; that is, no conclusion about the nature of the critical point can be reached.  122 4. APPLICATIONS OF DIFFERENTIATION THEOREM 4.10 (The Second Derivative Test). Suppose f" is con- tinuous near c. (I) If f'(c) = 0 and f"(c) > 0, then f has a local minimum at c. (II) If f'(c) = 0 and f"(c) <0, then f has a local maximum at c. (III) If f'(c) = 0 and f"(c) = 0, then f may have a local maximum, a local minimum, or an inflection point. y y y f"(c) <0 f"(c) > 0 f"(c)=0 (inflection) ............... C C C FIGURE 4.8. Second derivative test. The graph of f has a horizontal tangent line at a critical point, f'(c) = 0. If f"(c) < 0 and f"(x) is continuous, then f"(x) < 0 is some open interval containing c. Hence, the graph is concave upward near x = c and f has a local maximum at c (left panel). If f"(c) > 0, the graph is concave downward near x = c and f has a local minimum at c (middle panel). If f"(c) = 0, the graph may have an inflection point when f" changes its sign at c (right panel), but f" may not change its sign at c, and hence the behavior depicted in the left and middle panels is also possible in the case f"(c) = 0. The second derivative test is inconclusive. The function may have a local minimum, a local maximum, or an inflection. Examples are given in the captions of Figures 4.6 and 4.7 (with a = 0). In Example 4.6, the function f(x) = x3 - x is shown to have two critical points: x = +1/v/ as depicted in Figure 4.1. Since f"(x) = 6x, f"(-1/v/) -2v/5 < 0 (a local maximum) and f"(1/v) = 2/ > 0 (a local minimum). The function also has an inflection point at x = 0: f"(x) =Ox < 0 if xc < 0 and f"(x) =Ox > 0 if cc > 0. Note that an inflection point may not be a critical point! In other words, the tangent line at an inflection point can have any slope. In the example discussed, f'(0) =-1 (see also Figure 4.7).  24. THE FIRST AND SECOND DERIVATIVE TESTS 123 24.3. Exercises. (1) Find all critical points of the given function in its domain. Use either the first or second derivative test to determine whether there is a local maximum, a local minimum, or an inflection at each critical point. (i) f(x) - (x - 1)2 (ii) f (x) =(X - 1)3 (iii) f (x) =a(x -1)4 (iv) f (x) = (x - a)(x - b)(x - c), a < b < c (v) f (x) (x - a)(x - b)2, a < b (vi) f (x) = x3 - 6x2 + 9x - 4 (vii) f (x) = 2X3 - X4 (viii) f (x) = 8x3 - 9x2 + 1 (ix) f(X) =X(X - 1)2(X - 2)2 (x) f(cc) =X4 - 6x2 + 8x + 2 (xi) f(x)=cx+ (xii) f (X) =X2 - 2/c (xiii) f(x) 2x(1 + X2) (xiv) f(x) (x2 - 3x + 2)/(x2 + 2x + 1) (xv) f (X)X = 1 -X2 (xvi) f (X) = cxv - X2 (xvii) f (x) = 2x - X2 (xviii) f(c) Xcx - 1 (xix) f(x) =V/ccIncc (xx) f (X) =-1 ln(x2) (xxi) f (x) = sin(2x) + x (xxii) f (x) = In c - x3 (xxiii) f(x) =_X2-4 (xxiv) f (x) = cos x + j cos(2x) (xxv) f (x) = 10/(1 + sin2c) (xxvi) f (x) = ex sincc (xxvii) f (x) = (ex - e-x)/(ex + e-x) (xxviii) f(X) =_X2/3(1 _x)1/3 (xxix) f (x) = tan-1 x - 2 ln(1 + x2) (xxx) f (X) = x-1-l (2) Find the absolute extreme values of the function on a given interval: (i) f (x) =2x, [-1, 4] (ii) f (c) =cc2 - 4cc + 6, [-3, 101 (iii) f(xc) ccI2 - 3 + 2|, [-10, 10]  124 4. APPLICATIONS OF DIFFERENTIATION (iv) f (X) =X + , [0.01, 100] (v) f (x) = 5 - 4x, [-1,1] (3) Prove the inequalities: (i) |3x-t3 1|< 2 for Iz <2 (ii) |a sincxt+ b cosx z| I< a2 + b2 (4) Find the greatest lower bound and the smallest upper bound of the function on a given interval: (i) f (X) =e-0.01X, (0, oc) (ii) f (X) (1 + X2)/(1 + X4), (0, oC) (5) Prove the inequality 2 2+ 2 < ___ < 2 -oo 0, then near a the graph of f looks like a parabola concave upward (see the middle panel of Figure 4.8) and has a local minimum. If f"(a) < 0, then near a the graph of f looks like a parabola concave downward as depicted in the left panel of Figure 4.8 (a local maximum). For example, cos x has a local maximum at a = 0, and it behaves near a =0 as cosc x T2(x) =1 - x2/2. The second derivative test is inconclusive if f"(a) = 0. In this case, f (x) behaves near a as f(x) T3(c=) f(a) + 'f"'(a) (cc- )3. So, if f"'(a) / 0, f has an inflection point at a as depicted in the right panel of Figure 4.8. If f"'(a) = 0, one should look at f (x) T4(xc) =f (a) + & f(4)(a)(xc - a)4 . A function has a local maximum (minimum) at a if f(4) (a) < 0 (f(4) (a) > 0) as the concavity does not change at cc= a. This is to be compared with examples given in the caption of Figure 4.6. It is now  128 4. APPLICATIONS OF DIFFERENTIATION clear that the local behavior of f near its critical point is determined by a Taylor polynomial that has the first nonvanishing correction to f(a), provided the function is differentiable sufficiently many times. EXAMPLE 4.7. Investigate f(x) = x - tan x near x = 0. Solution: Find a Taylor polynomial for tan x with two nontrivial terms. In this case, it is T3: tanx T3(x) = x x3/3 (see Section 25.3). Therefore, f (x) x - T3(x) = -3/3. So there is an inflection point at x = 0. D 25.5. Asymptotes. How can the behavior of a function near a be an- alyzed if the function is not differentiable at a, or not even defined at a, or how does it behave in the asymptotic regions x -- too? DEFINITION 4.6 (Vertical Asymptotes). The line xc= a is a ver- tical asymptote of the graph y = f(x) if at least one of the limits limxma+ f(x) is infinite (oc or -oo). In other words, the function f(x) increases (decreases) unboundedly as x approaches a from either the left or the right. For example, the function _ (X2 + 3) X(X2 + 3) (4.12) f (x) = =c~1 -(c-1(c]1 x2 - 1 (x -1)(x +1) has two vertical asymptotes because the denominator vanishes at x = 1 and x = -1. When x approaches -1 from the left, f(x) tends to -oc, while it tends to 0o if -1 is approached from the right. Similarly, f (x) - -oo as x - 1- and f(x) - oc as x - 1+. Suppose f has a vertical asymptote at a. How does it behave near a? How "fast" does it diverge when x gets closer to a? DEFINITION 4.7 (Asymptotic Behavior). The functions f(x) and g(x) on an open interval x > a (including x> -oC) or x < a (including x < oo) are said to have the same asymptotic behavior at x = a if (4.13) lim (f(x) - g(x)) = 0 or lim (f(x) - g(x)) = 0 . -a+ a In particular, if x -- koo and g(x) = mx + b, then f is said to have a slant asymptote, and for m = 0, the slant asymptote is called a horizontal asymptote. For a given f, there are many g that have the same asymptotic be- havior because one can always change g by adding h such that h(xc) -~ 0 as cc - a*. A practical problem is to find as simple a g as possible  25. TAYLOR POLYNOMIALS AND THE LOCAL BEHAVIOR 129 with the property (4.13). In other words, one looks for a simple way to estimate the values of f(x) near a. EXAMPLE 4.8. Find the asymptotic behavior of the function (4.12) at x =±1. Solution: The function has to be investigated near x = ±1 and also when x -+ oo. 1. Near x = -1, the unbounded growth of f(x) is associated with the divergent factor 1/(x + 1) so that f(x) = h(x)/(1 + x), where h(x) is finite near x = -1. Then f(x) r h(-1)/(x + 1) = g(x): 1 x(x2+3) 2 (x) = ~ x+1 x -1 x+1 g(x). Apparently, limx_ 2/(x + 1) are close .l (f(x)-g(x)) near x =-1. 0. The graphs of f(x) and g(x) 5 10 FIGURE 4.9. Graph of f(x) given in (4.12) (the blue solid curve). It has a slant asymptote g(x) = x as x -- ±oo (the dashed line). In these asymptotic regions, f (x) ~ g(x) = x. The function also has two vertical asymptotes x = 1 (the red vertical line) and x = -1 (the blue vertical line). The red solid curve is the graph of g(x) = 2/(x - 1), which shows the asymptotic behav- ior of f(x) near x = 1. In a neighborhood of x = 1, f(x) ~ g(x) near x = 1. The function f exhibits a simi- lar behavior near x = -1 (not depicted here).  130 4. APPLICATIONS OF DIFFERENTIATION 2. Similarly, near x = 1 1 x(x2+3) 2 (X) fz-)1 x+1 - 3. To find an asymptotic behavior when x is large it is convenient to factor out the largest power of x in the numerator and denominator: f z-z= 2(1 + ) 1 +x f)= 1 =c1-1 Since u= 1/cc2 is small in the asymptotic region, the factor (1- )-1 1 + u can be linearized. Therefore the asymptotic behavior of f(x) is f (cc c(+3X2) 2) 2) c x where the terms x-4 have been neglected. This shows that the graph has the slant asymptote y = x. Since f(x) -x 4/cc is positive if x> 0 and is negative if x < 0, the graph approaches the slant asymptote from above as x -- 0o and from below as x - -oo. D 25.6. Asymptotic Behavior and Taylor Polynomials. Taylor polynomials also provide a powerful technique to investigate an asymptotic behavior of a function. This is illustrated by the following example. EXAMPLE 4.9. Investigate f(x) = -8/3(1 - cos c) near c = 0. Solution: The factor x-8/3 diverges as x -- 0, but cos x is smooth near x = 0 and can be approximated by the Taylor polynomial: cosXcT4(x) = T4 = 1 - jz2 +1cz4, f(x) cX-8/3(1 - T4() =cc8/3(1cc2 - 1z4) -1-2/3 - 1cz4/3 Therefore, for a sufficiently small x, f(x) c-2/3 = g(x) because f(X) - g(x) ~24/3 - 0 as x - 0. Note that the use of T2 in place of T4 would not be enough to establish the asymptotic behavior of f. D 25.7. Exercises. (1) Find Taylor polynomials of f (x) = x4 - x3 + 5x2 - 2x + 1 at x = 0. What can be said about Taylor polynomials of a general polynomial function P,(xc) =anz"+ an_1x"- +.-.-. + a1xc + ao at cc= 0? (2) Decompose the polynomial P(xc) =1 + 3cc + 5cc2 - 2cc3 into the sum of powers of the monomials (1 + z)", where n is a nonnegative integer.  25. TAYLOR POLYNOMIALS AND THE LOCAL BEHAVIOR 131 (3) Approximate f (x) = (1+ x + x2)/(1 - x +x2) by T2(x) about x = 0. Hint: Approximate first (1+u)1 by T2(u) about u = 0 and then use f(x) (1+ x + x2)T2(u), where tt -x + x2 retaining only z", n < 2, in the product. (4) Find T2(x) about x = 0 forf(x) = mam + x, a > 0. (5) Find T2(x) about x = 0 for f (x) = 1 - 2x + x2. Hint: Use T2(u) for 1 + u about u = 0 and set u = -2x + x2. (6) Find T3(x) for f(x) = e2x-x2 about x = 0. Hint: Use a suitable approximation of e" by a Taylor polynomial where u = 2x -x2. (7) Find T4(x) for f(x) = 1 + x2 - 1 - x2 about x = 0. Hint: Use a suitable Taylor polynomial to approximate 1 + t, where u= x2. (8) Find the nth-degree Taylor polynomials of the given function at a specified point: (i) f(x) sin x, c= 0 (ii) f (x) = cos x, c= 0 (iii) f (x) = In x, c= 1 (iv) f (x) = 1/c, cc= 2 (v) f(cc) (1 + c)P, p> 0, cc= 0 (vi) f (X) ex - e-, cc = 0 (9) Estimate the absolute error of the approximation for a given interval: (i) sincxc.;Xc- x3cc (ii) tanx, Xc+ 3, Icc <0.1 (iii) 1l+ x 1 + jz - x2, 0<5x < Hint: In (i) and (ii), compare T3(x) and T4(x) about x = 0. (10) Find an interval in which the approximation cos x 1 - jzc2 is accurate within the absolute error 0.0001. (11) If f is twice continuously differentiable near x = 0 and f(0) 0, find the local behavior of the function F(x) = f(|x|P) near x = 0, where p > 0. (12) Let f and g be twice differentiable at a and g(a) = 0. Find the second-degree Taylor polynomial for the function F(x) f (g(cc)) near cc= a. Hint: Use f (u) T2(u) =f (0)+ f'(0)u+ f"(0)u2/2, where tt= g(cc) and g(cc) is also approximated by the corresponding T2 near a. (13) Find the third-degree Taylor polynomial for the following func- tions at a specified point by using the results from the previous  132 4. APPLICATIONS OF DIFFERENTIATION exercises (i.e., by using Taylor polynomials of a suitably cho- sen argument): (i) f(X) =sin(x3), Xz= 0 (ii) f (x) = sin(sin z), = 0 (iii) f (x) = tan(1 - cos x), x = 0 (14) Use Taylor polynomials to investigate the local behavior of a given function near a specified critical point (whether it has a local maximum, a local minimum, or an inflection): (i) f(X) =sin(z4), c=c 0 (ii) f(x) = 1 - x2/2 - cos x, c= 0 (iii) f (x) = ln(1 + x) - X + X2, c=c 0 (15) Use Taylor polynomials of successive degrees for f (x) = ln(1+ x) near x = 0 to evaluate ln 2. What degree is required to calculate ln 2 correct within the absolute error 10-4? (16) Use Taylor polynomials to find the number correct within the given absolute error : (i) e, e = 10-4 (ii) sin 1°,6 e =10-4 (iii) cos 9°, e = 10-5 (iv) v/5, c = 10-4 (v) log10 11, c= 10-5 (17) Find vertical and slant asymptotes, if any, of a given function. Investigate the asymptotic behavior of the function near the points where it has vertical asymptotes and in the asymptotic regions x - kmoo. (i) f(x) =x + 4/c (ii) f (X) =x2/(x2 - 1) (iii) f () =-(X3 - 3X2)/(X2 - 2x + 1) (iv) f () =-x2/3 (2 - 1)1/3 (v) f() =(coscx - 1)/2 (vi) f () - (X - sin )/4 (18) Approximate the given function by a power function near a specified point by using Taylor polynomials: (i) f(X) c-2/3 ln(1 + c), cc= 0 (ii) f(X) c-4/3 sin2(2cX2/3), cc 0 (iii) f (X) =z-5/3 (cc- tan (c1/3)), cc 0 (iv) f(x) =[sin(x - 1) - x + 1]/(x - 1)3, cc- 1 (19) Suppose that the functions f and g are such that f(a) g(a) =0, f(k)(a) = 0 for k =1, 2, . .. ,mn, and g(k)(a) = 0 for k =1,2,...,mn, while f(n+1)(a) / 0 and g(mi+1)(a) / 0.  26. L' HOSPITAL'S RULE 133 Investigate the local behavior of the function h(x) = f(x)/g(x) near x = a if n = m, if n > m, and if m < n. 26. L'Hospital's Rule If a function f is not defined at a, then its behavior near a depends on the limit of f as x -- a, whether it is finite, infinite, or does not even exist. So this question is of importance when investigating a function. There is a special technique to answer it. 26.1. Indeterminate Forms ° and . Consider the behavior of the fol- lowing functions: (4.14) ex1 1-cosx tangy - x x(23 as Do they have a vertical asymptote at x = 0? These functions have a common feature. They are ratios f/g of two functions f and g such that f(x) -- 0 and g(x) -- 0 as x -- 0. Similarly, one can make ratios where the limits of the numerator and denominator at a particular point are infinite: ln x (4.15) In x as x - 0+ . In general, a limit of the form l f(x) lim x-a g(x) is called an indeterminate form of type °oif both f(x) -- 0 and g(x) - 0 as x - a; it is called an indeterminate form of type if both f(x) -o0 (or -oc) and g(x) -- 0 (or -oc). The limit itself may or may not exist. The following theorem provides a powerful method to study the indeterminate forms of these types. THEOREM 4.12 (L'Hospital's Rule). Suppose f and g are differen- tiable and g'(x) 0 on an open interval that contains a (except possibly at a). Suppose that lim f (x) = lim g(x) = 0 x-- a x-- a or that lim f(x) =+tmoo and lim g(x) =too. Then (4.16) lim =lim x-a g(xc) x-a g'(xc) if the limit on the right-hand side exists (or is infinite).  134 4. APPLICATIONS OF DIFFERENTIATION For the special case in which f(a) = g(a) = 0, the derivatives f' and g' are continuous, and g'(a) / 0, it is not difficult to see why l'Hospital's rule (4.16) holds: ff(x) f (a) f(x)-f(a) lim f(x)-f(a) = l = lim = x-a g(x) x- a g(x) - g(a) x- a g(x)-9(a) lim g)ga 9 4(a) x-a x ax-a f'(a) f'(x) - lim . g'(a) x- a g'(x) The first equality follows from f(a) = g(a) = 0, the second and third equalities are the consequence of the limit laws and the assumption that g'(a) / 0, and the last equality follows from the continuity of the derivatives. This simplified version of l'Hospital's rule can be under- stood geometrically. The functions f and g can be approximated by their tangent lines at a, f(x) f'(a)(x - a) and g(x) g'(a)(x - a), so that f(x)/g(x) f'(a)/g'(a) near a. It is not so easy to prove the general version of l'Hospital's rule (the proof is omitted here). L'Hospital's rule is also valid for one-sided limits x - a1 and for the limits at too. The hypotheses of l'Hospital's rule must be verified for the corresponding limits. What happens if f'(a) = g'(a) = 0? Apparently, the conditions of l'Hospital's rule are satisfied for the derivatives f'(x) and g'(x) if f and g are twice differentiable. So l'Hospital's rule may be applied again to the ratio f'(x)/g'(x). For functions differentiable many times, l'Hospital's rule is easy to understand via the Taylor polynomials. Sup- pose that functions f and g are continuously differentiable sufficiently many times near a. Then by Theorem 4.11 the following approximation holds f (x) f (a) + f'(a)(x - a) + jf"(a)(x - a)2 + -.. g(x) g(a) +g'(a)(x - a) + 2g"(a)(x - a)2 +... If f(a) = g(a) = 0, then the limit of the ratio is determined by f'(a)/g'(a). If f(a) = g(a) = 0 and f'(a) = g'(a) = 0, then the limit is determined by f"(a)/g"(a) and so on. EXAMPLE 4.10. Investigate the indeterminate forms (4.14) and (4.15). Solution: 1. Let f(x) =ex - 1 and g(x) =x. Then f (0) =g(0) =0 (the conditions of l'Hospital's rule are fulfilled). Hence, eX - 1 lm(ex -1)' e x-o z 2-0 x)' -0 1  26. L' HOSPITAL'S RULE 135 26. L'HOSPITAL'S RULE 135 2. Let f(x) = 1 - cosx xand Then f'(x) = sin x and g'(x) l'Hospital's rule can be applied g(x) = x2 so that f(0) = g(0) = 2x. Since f'(0) = 0 and g'(0) again: 0. 0, 1 - cos x x-o c2 lim sin c x-o 2x lim (sin z)' X-0 (2x)' cos x 1 X-0 2 2 3. Let f(x) = tan x - x and g(x) = x3 so that f(0) = g(0) = 0. The derivatives f'(x) = sec2 x - 1 and g'(x) = 3x2 vanish at x = 0. L'Hospital's rule can be used again to resolve the indeterminate form. For complicated functions, taking higher-order derivatives might be quite an algebraic exercise. Sometimes, simple algebraic transforma- tions of an indeterminate form in combination with basic limit laws may lead to the answer faster than a successive use of l'Hospital's rule: tan x - x lim x-o c33 sec2 x - 1 1 - cos2 x lim lim x-o 332 x-O 3c2COS2 imsin2 ccx 1 ( sin zc2> x-o) 3cc2 3 2c-0 cc/ 3 The third equality follows from cos x -- 1 as x -- 0, and therefore cos2 x in the denominator can be replaced by 1 in accord with the basic limit laws. 4. Let f(x) = ln x and g(x) = x-1 so that f(x) -- -oc and g(x) -- o0 as x -- 0+. So the conditions of l'Hospital's rule are fulfilled. Therefore, lim lnxcc x-~O+ x-1 (ln c)' ll x-+ (c-1)' lim cc1 x-~O+ -c-2 lim x X-o+ 0. F- 26.2. Indeterminate Products 0- oo. Suppose that f(x) - oc and g(x) -- 0 as x -- a. How can the indeterminate product f(x)g(x) be investigated when x -- a? It turns out the indeterminate prod- uct can be transformed into one of the indeterminate forms to which l'Hospital's rule is applicable: f9 (4.17) f g - 1/ (0-" ) oo - 0 0 - g or f g =f ( 0 oo 0 - -. 0 The function xln x is an indeterminate product of the type 0 -oc as x - 0+. It can be transformed into an indeterminate form of the type i as in (4.15), which is then resolved by l'Hospital's rule (see Example 4.10). Note that, although either of the transformations in (4.17) may be applied with the subsequent use of l'Hospital's rule, the  136 4. APPLICATIONS OF DIFFERENTIATION technicalities involved might differ substantially. For instance, if the second option in (4.17) is applied to xln x = z/(1/In zx), then 1 1 lim x In = lim x = lim = lim x In2 x . x-o+ x-o+ In x -_+ - l j x x-o+ Although our goal has not been achieved, our effort has not been in vain. Since the left-hand side vanishes by Example 4.10, it follows that l 1n2 x - 0 as x - 0+. By repeating this procedure recursively, one can infer that lim x(ln z)" = 0, n = 1, 2,.... x-o+ 26.3. Indeterminate Powers 00, 000, and 1°. Several indeterminate forms arise from the limits of [f(x)]g(x) as x - a: 00 (f (x) > 0 , g(x) > 0) ; °O0 (f (x) > 00 , g(x) > 0); 1° (f (x) -1, g(x) 0) . Note c0 = 1 if c / 0 and c 00. Similarly, c° = 0 if 0 < c < 1 and co = 00 if c > 1. The indeterminate powers can be transformed into an indeterminate product with the help of the identity y = e1n g: lim[f (x)]g(x) = lim eln([f(x)]=(x)) lim eg(x)ln(f(x)) = elimxag(x)ln(f(x)) x-a x-a x-a The limit of g(x) ln(f (x)) is of type 0"- 00 and can be treated by the rule (4.17). The procedure is illustrated with an example of the type o0 indeterminate power: lim zl/x = lim en(xl/x) = lim eln(x)/x = elimxo ln(x)/x -0 1. x-oo x-oo x-oo 26.4. Indeterminate Differences 00 - oc. Suppose f(x) 00 and g(x) - 00 as x - a. The limit of f (x) - g(x) as x - a is called an indeterminate difference. The following transformations might be helpful to investigate it: f-g =f(g1= 1-g/f _f g f/g-1 f g f1-- o g g=. f 1/f (g-1) 1/g If f(x)/g(x) - 1, then the indeterminate difference is equivalent to an indeterminate form of type 0/0 and can be investigated by l'Hospital's rule. The limit of f/g is an indeterminate form of type 00/00 and can also be investigated by l'Hospital's rule. Suppose that f(xc)/g(xc) -~ k as cc - a, where k can be either a nonnegative number or k =00. If k < 1, then f - g =g(f/g - 1) -~ 00 - (k - 1) =-oo; that is, g increases faster than f as cc -~ a. If k > 1 or k =00, then  26. L' HOSPITAL'S RULE 137 26. L'HOSPITAL'S RULE 137 f - g = g(f/g - 1) -- c- (k - 1) than g as x -- a. For example, oc; that is, f increases faster lim in x + - oO+ x li- 1I+ x In x) oO+ x 1 lim -(1+0) xO+ X 00. If k = 1, then it is also possible that f -g - c, where c is a number. In this case, f and g increase asymptotically at the same rate: f'-g' - 0. If c = 0, the functions f and g have the same asymptotic behavior. For example, lim . o0 smin cot x) =lim . 1 x-od0 sin x sinxz cos x = lims / X-o cos c 0, where l'Hospital's rule has been used in the second equality. An alternative solution is obtained if the local behavior of the func- tions near x = 0 is approximated by the Taylor polynomials. Use T2 to approximate cos x and T3 for sin x: 1 sin xc 1-cos x x2/2 cot x = _~ sincx z x- z3/6 /2 x 1 - x2/6 2 where x2/6 is small as compared to 1 when x is close enough to 0 and can therefore be neglected in the denominator. This method is often technically easier than the use of l'Hospital's rule. 26.5. Exercises. (1) Find the limits: (i) lim ln(1 + X) xo 0x xz-sinzx (ii) lim -i sin cc3 cs (iii) lim sincx - cosc x-ur/4 cos(2x) sin(az) (iv) lim 2-0 sin(bz) (v) lim 1 - cos(x2) X-0x2 sin(x2) sin-1(az) (vi) lim x- 0 sin(bz)  138 138 4. APPLICATIONS OF DIFFERENTIATION (vii) lim sin-1(2x) - 2 sin-1 x (viii) urn sin x in x (ix) urn [1 - cos x] Inx (x) urn silx -x x-W ex - 1 - X2/2 (x) i tan3 x - x x-W ex - 1 (xii) Ii Mrx2[f x+1- 1_ 2] (xiii) lim (1+ ) (xiv) urn (1 - e-x)ex x-~oo (xvi) urn (x -in x)l/ (xvii) urn (sin(ax))snb), a > 0, b > 0 (xviii) In sin (ax)b> l-o n sin(bx)'>0 (xix) limIncot (x) (xxi) urn ax-cc a> 0 x-- a x -aC (2) Use approximations of basic functions by Taylor polynomials to find the limit: (i) lim cos cc - 2ex2/4 (iii) limcc372( cc+ 1+ c-i 2cc 2\/)  26. L' HOSPITAL'S RULE 139 (iv) lima + 2 a > 0 X---O X2 > (v) lim x - x2 In 1 + -00L X (v i) lim -cv -1) ( -0i x sin xz (vii) lim- (! -cot x (3) Suppose that f(x) has the second derivative f"(x). Show that f"(cv) = rn f (x + h) + f(x - h) - 2f (x) h--0 h2 (4) Let y -- 0 as x -- 0. Find the asymptotic behavior of y, that is, the leading term y Cz", C / 0, if (i) y tanv - x (ii) y = tan(sin x) - sin(tan x) (iii) y = (1 + c)x - 1 (iv) y = 1 - e-1(1 + c)1/x (5) If y = x - (a + b cos c) sincx, find a and b such that y Cz5 as x - 0, C E 0. (6) Find a and b such that the approximation 1 + ax2 cot x z + bz2 is correct when the terms of order z5 and higher can be ne- glected as x -- 0. (7) Consider the function f(x) = e-1/x2 if x / 0 and f(0) = 0. Show first that lim cvel/x2 = 0, n> 0. x-- O Use this fact and the definition of derivatives to show that f(k) (0) = 0 for all k. Can Taylor polynomials be used to investigate the local behavior of the function near x = 0 and establish the nature of the critical point x = 0? (8) Find all critical points of the function e-/ vz> 0, f (cv={0,/ cv= 0, -e/ vz< 0, and investigate the behavior of the function near them.  140 4. APPLICATIONS OF DIFFERENTIATION 27. Analyzing the Shape of a Graph To analyze the shape of a graph y = f(x), it is useful to have a clear idea of how the basic functions behave. For example, sin x and cos x are regular everywhere, bounded (e.g., I sin x < 1), and periodic with a period of 27. In addition, sin x has zeros at x =win, n = 0, +1, +2, ..., while cos x vanishes at w/2 + win. The function sin x is odd, while cos x is even. Their ratio tan x = sin x/cos x is not defined at roots of cos x. How does tan x behave, say, near x =w7/2? Since both sin x and cos x are smooth near x =w7/2, the behavior of tan x near w/2 can be understood with the help of Taylor polynomials. Put Ax= x - w/2 (the deviation of x from w/2). Let us approximate sinx Ti(x) = 1+ (x - ) = 1+ Ax COS X T3 (x) -_-- x - )+(x- 2 3 -=z-x 6(0x3 . Then 1+Ax 1 1+Ax 1 1 tanxze-m- -Ax + (Ax)3/6 Ax 1 - (Ax)2/6 Ax x - w/2' where the second ratio in the product has been approximated by 1 because Ax is small. Since tan(x + w) = tan x, this behavior repeats itself at near every root of cos x. 27.1. Growth of the Power, Exponential, and Logarithmic Functions. Let us compare the growth of the power function x", the exponential func- tion ex, and the logarithmic function ln x as x - oo. The exponential function grows faster than the power function. Let f(x) = ex and g(x) x . Let us analyze the ratio f/g as x - oo. The conditions of l'Hospital's rule are satisfied: ex - oc and x" - oc as x - oo. L'Hospital's rule can successively be applied until the indeterminate form is resolved: ex ex ex ex lim -=lim = lim = - --"= lim - 0o. x-o x x- o nx1 x-o in(n - 1)x-2 x-o in! The conclusion is true for any real n. For any real n, there exists a positive integer N such that n < N or x" < xN, x > 1. But ex grows faster than xN. Similarly, it is straightforward to show that the logarithmic function grows slower than any power function: ln x (lnxz)' 1 1 x-oo x" x-oo (z")' x-oo inx- x-oo inza for any in> 0 (in may be any positive real number here).  27. ANALYZING THE SHAPE OF A GRAPH 141 27.2. Asymptotes at x - too. The asymptotic behavior of rational functions is easily determined by the highest powers of the numerator and denominator, as in Example 4.8. In general, if lim , f(x) is infinite, then the limit of f/g can be studied for trial gs with different growth, g = mx (for slant asymptotes), g = z", g =ln x, and so on. Suppose g(x) is found such that f(x)/g(x) - 1 as x - oc. Does this mean that g and f have the same asymptotic behavior? The answer is "no." If the indeterminate form f(x) - g(x) of type oc - oo converges to 0 as x -- 0o, then the indeterminate form f(x)/g(x) of type * converges to 1. Indeed, it follows from 1/g(x) - 0 and f(x) -g(x) - 0 that (1/g(x))(f(x) - g(x)) = f(x)/g(x) - 1 - 0. The converse is not true. Consider the following simple example: f(x) = x + sin x and g(x) = x. Evidently, f(x)/g(x) = 1 + sin x/x - 1 as x - oc. But the limit limx (f (x) - g(x)) = limo sin x does not exist. So, even if g is found to have the property f(x)/g(x) - 1 as x - 0oc, the indeterminate form f - g of type oc - oc must still be investigated in order to determine whether or not g has the same asymptotic behavior as f. 27.3. Guidelines for Analyzing the Shape of a Graph. The following guidelines are useful for sketching the graph of a function. It should be noted that not all the steps can always be carried out. This depends very much on the complexity of the function in question. So these are really guidelines, not a "must-do" algorithm. Given a function f, find: (I) Domain. The domain consists of all values of x at which f(x) is de- fined. Typically, it is a collection of intervals. If f is defined for x > a or x < a, or both, but not at a, the the local behavior of f near a must be studied (see below). (II) Roots of f and the value f(0). Roots of f(x) define the intercepts of the graph y = f(x) with the x axis. They are not always easy to find. The value f(0) (if x = 0 in the domain of f) defines the intercept of y = f(x) with the y axis. (III) Symmetry and periodicity. If f(-x) = f(x) (an even function) for all x in the domain, then the graph y =f(x) is symmetric about the y axis. If f(-x) =-f(x) (an odd function) for all z in the domain, then the graph y =f(x) is symmetric about the origin (or the rotation through 1800 about the origin). If there is a number p such that f (x+p) =f (x), then f is periodic and p is its period.  142 4. APPLICATIONS OF DIFFERENTIATION The graph y = f(x) repeats itself on intervals of length p, for example [a, a+ p], [a+p, a+2p], and so on for any a. Examples are sin x, p = 27; tan x, p = 7; cos(4x), p = 27/4 = 7/2. (IV) Asymptotes and asymptotic behavior of f. If f is a ratio f = h/g, then vertical asymptotes are x = c, where c solves g(c) = 0 and h(c) / 0. If h(c) = 0, find the limits lim-c+ f(x). If one of the limits or both is infinite, investigate the local behavior of f near c (e.g., with the help of Taylor polynomials if possible). The asymptotic behavior of f (x) near c and for large positive and negative x determines the shape of y = f(x) near the vertical asymptotes and the asymptotic shape of the graph when x -- too. (V) Critical points of f. Critical points are solutions of f'(x) = 0 or the values of x where f'(x) does not exist. If, for example, f'(x) tends to 00 (or -oc) as x approaches c, then the line tangent to the graph y= f(x) at x = c is vertical. For example, f(x) - z1/3 and f'(x) = 1/(3x2/3). So f'(x) diverges as x - 0. The graph y =1/3 has a vertical tangent line at xc= 0. (VI) Intervals of positive and negative values of f. These are the intervals where the graph y= f(x) lies above or below the x axis. Roots of f generally separate the intervals of positive and negative values of f. However, this is not always the case. Let c be a root of f. If f'(c) / 0, then the function f is increasing or decreasing at c and hence must change its sign. If f'(c) = 0 or f' does not exist at c, that is, a root of f coincides with its critical point, then f is negative near c if f has a local maximum at c and f is positive near c if it has a local minimum at c. So the sign of the derivative f' must be investigated near c (the first derivative test). Vertical asymptotes can also separate intervals of positive and negative values of f. For example, the function (4.12) has one root x = 0 and two vertical asymptotes at x = -1 and xc= 1. So f is negative on (-oc, -1), positive on (-1, 0), negative on (0, 1), and positive on (1, oc). The graph is shown in Figure 4.9. (VII) Intervals of increase (f' > 0) and decrease (f' < 0). If f' > 0 (f' < 0) on an interval, then f increases (decreases) on it (the increasing-decreasing test). These intervals are gen- erally separated by critical points and vertical asymptotes. As a consequence of this study, the nature of each critical point is established by the first derivative test.  27. ANALYZING THE SHAPE OF A GRAPH 143 (VIII) Intervals of upward and downward concavity. These intervals are separated by inflection points and vertical asymptotes. The sign of f"(x) must be studied. Yet, the second derivative test and Taylor polynomials can be used to establish the nature of a critical point of f. (IX) Values of f at critical points and inflection points. These values set relative scales of the graph (e.g., they show how much the function increases between two critical points). EXAMPLE 4.11. Sketch the graph of f (x) = x1/3(x - 6)2/3. Solution: Following the preceding guidelines: (I) The domain is the whole real line. (II) The roots of f are x= 0 and x= 6 (the intercepts with the x axis). The intercept with the y axis is f(0) = 0. (III) The function is not periodic, and it is neither odd nor even. (IV) There is no vertical asymptote. To study the asymptotic be- havior as x -- too, it is convenient to factor out the largest power of x: f(x) = x(1 - 6/x)2/3 and approximate the second factor using Taylor polynomials in ut= -6/x. One has 4 4 (1 +u) 2/3 mT2(26) = 1+32.6- 9222 4 f(x) cT2(u) = x - 4 - - This shows that the graph has a slant asymptote of the form y = x - 4. It also follows that f(x) - (x - 4) -4/xc <0 if x is large and positive. Hence, the graph approaches the slant asymptote from below. Similarly, f(x) - (x - 4) > 0 if x is large and negative. Hence, the graph approaches the slant asymptote from above. (V) The derivative reads x - 2 f'(x)= X2/3(x - 6)1/3' It vanishes at x = 2 and does not exist at x = 0 and c = 6. The critical points are 0, 2, and 6. In particular, f'(x) - oc as x -- 0 and it tends to too as x - 6±, respectively. Therefore, the graph has vertical tangent lines at cc= 0 and c 6. Near cc= 0, the graph looks like y =f(cc) 62/3cc1/3, while near cc= 6, it has a downward cusp y =f(cc) 61/3(cc - 6)2/3. (VI) The graph lies below the cc axis on (-oo, 0) as f(cc) < 0 and above it on (0, oo) as f(cc) ;> 0. The function does not change  144 4. APPLICATIONS OF DIFFERENTIATION x 4 2 2 .+ 4 8 10 4 2 .' 6 - 8 FIGURE 4.10. Graph of f(x) =1zi/3(x-6)2/3. The roots of f are x= 0 and x= 6, and they define the intercepts with the x axis. It has the slant asymptote f(x) x - 4 as x -- too. The derivative vanishes at x = 2 (a local maximum). It diverges at x = 0 and x= 6; the graph has vertical tangent lines at these points. The second derivative is negative if x < 0 so that the graph is downward concave. It is positive on (0, 6) and (6, oo). The graph is concave downward. The point x = 0 is an inflection point as the concavity changes at it. its sign at the root x = 6 (f must have a local minimum at 6, which is also verified by the first derivative test below). (VII) The derivative is a product of three factors x - 2, x-2/3, and (x - 6)-1/3. By investigating the signs of these factors on the intervals separated by the critical points, we can conclude that f' > 0 (f is increasing) on (-oc, 0), f' > 0 (f is increasing) on (0, 2), f' < 0 (f is decreasing) on (2, 6), and f' > 0 (f is increasing) on (6, oc). Also, f has a local maximum at x = 2 and a local minimum at x = 6 by the first derivative test.  27. ANALYZING THE SHAPE OF A GRAPH 145 (VIII) The second derivative reads 8 f"() X5z/3 (X - 6)4/3' The factor (x - 6)4/3 cannot be negative. The sign of f" is determined only by that of z5/3. Thus, f" > 0 on (-oc, 0) (the graph is upward concave) and f" < 0 on (0, 6) and (6, oc) (the graph is downward concave). So x = 0 is the inflection point. Also, near x = 2, the graph looks like the downward parabola y = T2(x) = f (2) + f"(2)(x - 2)2/2 = (4 -4(x - 2)2)/32. E In the age of graphing calculators, the preceding guidelines might look rather obsolete because finding the shape of a graph can be done just by hitting the appropriate calculator buttons. But what a calcu- lator cannot do is to provide details of the local behavior of a func- tion near points of interest (e.g., critical points, asymptotes, etc.). In science and engineering, this is often much more important than the overall shape of a graph. In the previous example, a calculator would show that there is a slant asymptote, a cusp at x = 6, and a local maximum at x = 2, but it would not be able to determine the local behavior of the function near the cusp, or at the local maximum, or in the asymptotic region. Here a good working knowledge of calculus becomes indispensable, while a graphing calculator is just a useful tool that greatly facilitates the study of a function. 27.4. Exercises. (1) Sketch the graph of each of the (i) f (x)=3x -X3 (ii) f ( ) 1 + X2 - (iii) f (X) =-x2(x - 1)(2 - z) (iv) f(x) = (x + 1)(x - 2)2 (v) f(cx) =X2/(4 + X2) (vi) f (X) =X4/(1 + X2)2 (vii) f (x) = z/[(1 + z)(1 - x2)] (viii) f(X) =x3/(cx2 - 3x + 2) (ix) f (X) (zI - x2)/(Xc+1) (x) f(X) =[(1 + c)/(1 - z)]4 (xi) f (x) = f(x - a)2+ b2 (xii) f (x) = (x - 2)/v/z-2 + 1 (xiii) f (X) = X3 - X2 - X + 1 (xiv) f(x) = 3c2 - 3 X2 + 1 following functions:  146 4. APPLICATIONS OF DIFFERENTIATION (xv) f (x) (xvi) f (x) (xvii) f (x) (xviii) f(x) (xix) f(x) (xx) f(x) =- (xxi) f(x) (xxii) f(x) (xxiii) f(x) (xxiv) f (x): (xxv) f(x) (xxvi) f(x): (xxvii) f (x) (xxviii) f (x) (xxix) f (x) (xxx) f (x) (xxxi) f (x) (xxxii) f (x) -+ l2/3/, _z-1/3(x - 6) -2/3 = X2- 1| -x - [(x - 1)/(x + 1)]1/3 sin x + cos2 x cos x - j cos(2x) sin z/(2 + cosx) = cos x/ cos(2x) = x sin x = 2x - tanx = sin(nx)/ sin x, n = 2 = x + e-x = x2/3e-x = (ex + e-x) cos x = sin2 x/x2 = lnx/ /x = ln(x + vx2 + 1) , 3, 4 (2) Sketch the graph of the polynomial with k real roots: f (x) = A(x- Xi)1(x-x2)"2 (x - xk)"k, where A > 0 and ni, n2,... ,nk are positive integers. Investi- gate first the case when ni = n2 = = nk= 1, then the case when one of the powers ni, n2, ....,nk is greater than 1 (how does the graph look when this power is odd or when it is even?). Then proceed to the general case. (3) Let f and g be second-degree polynomials such that f" > 0 and g" > 0. Sketch all possible shapes of the graph y f (x)/g(x). 28. Optimization Problems Suppose that a quantity Q depends on some variables. The prob- lem of optimizing Q implies finding the values of the variables at which the quantity Q attains it maximal or minimal value. The simplest op- timization problem arises when Q depends on a single variable x such that Q is a function f(x). Then the optimization problem is reduced to the problem of finding extreme values of f(x). The latter problem has been analyzed in Section 22. To determine extreme values of f, one has to: (I) Find all critical points of f.  28. OPTIMIZATION PROBLEMS 147 (II) Investigate the nature of the critical points (local minima and local maxima). The first or second derivative tests can be used for this purpose. (III) Calculate the values of f at the endpoint of the interval [a, b] (if extreme values are sought only in [a, b]) and compare them with values of f at its local maxima and minima to determine absolute extreme values of f. The following test can also be used to find absolute extreme values of a function. THEOREM 4.13 (First Derivative Test for Absolute Extreme Values). Suppose c is a critical point of a continuous function f de- fined on an interval. (I) If f'(x) > 0 for all x < c and f'(x) < 0 for all x > c, then f (c) is the absolute maximum value of f. (II) If f'(x) < 0 for all x < c and f'(x) > 0 for all x > c, then f(c) is the absolute minimum value of f. The conclusion of the theorem is easy to understand. Consider case (I). Since f'(x) > 0 for all x < c, the function increases for all x < c. Since f'(x) < 0 for all x > c, the function decreases for all x > c. By continuity of f, the number f(c) must be the largest value of f. Case (II) is proved similarly. Recall Example 4.2. This is a typical optimization problem. Its solution is rather straightforward, provided Equation (4.4) is given. Without it, the problem of finding an optimal angle for a projectile becomes far more difficult. Its major part now involves a derivation of Equation (4.4)! This is quite typical for optimization problems. As a rule, they arise in various disciplines, and their formulation as the mathematical problem of extreme values requires a specific knowledge outside mathematics, for example, the laws of physics as in Example 4.2, chemistry, biology, economics, and so on. A typical optimization problem may be split into three basic steps: (I) Identify a variable with respect to which a quantity Q is to be optimized. (II) Use the laws of a specific discipline to express Q as a function f of that variable, Q = f(x). (III) Solve the mathematical problem of extreme values of f. E XAMPLE 4.12. An aluminum can has the shape of a cylinder of radius r and height h. Design an aluminum can of volume V =300 cm3 to minimize the cost (or the amount) of material needed to make the can.  148 4. APPLICATIONS OF DIFFERENTIATION Solution: Following the preceding guidelines: (I) Apparently, the least amount of material is used when the surface area of the can is minimal. So one has to minimize the surface area S, which depends on r and h. But the variables r and h are not independent because the volume is fixed. (II) The surface area is the sum of the areas of the side, top, and bottom of the can: S= 2wrh + r2 + r2 = 2wrh + 27r2. The volume is V = wr2h. Since the volume is fixed, the variables r and h are related as h = V/(wr2). Hence, S can be written as a function of the radius r only: S(r) = 2xr 2 + 27r2 2V + One has to find the value of r > 0 at which S(r) attains its absolute minimum. The corresponding value of h is then found from the relation h = V/(wr2). (III) The function S(r) is differentiable for all r > 0. Therefore, all its critical points are roots of the derivative: 2V S'(r) = + 4xr 4{ r3K 0. So the critical point is / V)1/3 Since S'(r) < 0 for all 0 < r < rc and S'(r) > 0 for all r > rc, the function S(r) attains its absolute minimum at rc by the first derivative test for absolute extreme values. The dimensions of the can with minimal costs of material for a given volume V are V)1/3 r = -27 c (4V)1/3 ~3.6 cm, h 2 2rc ~7.2 cm. The analysis has shown that the height and diameter of a can of a given volume must be equal in order to minimize the cost of material (or the surface area of the can). Check out a local supermarket to see if manufacturers use this fact! Q This example is further illustrated on the interactive website at http://www.math.ufl.edu/~mathguy/ufcalcbook/optimize-cylinder .html.  28. OPTIMIZATION PROBLEMS 149 Remark. In the previous example, S has been expressed as a function of r. The same conclusion could be reached if S is expressed as a func- tion of the height h only, that is, when the relation r = /V/(7h) is substituted into the expression for the surface area to obtain S(h). The critical point of S(h) can be shown to be hc = 2rc. Verify this! A Curious Fact. The preceding problem is essential to reduce waste from plastic, glass, and aluminum containers. It can be stated more generally. What is the shape of a container that has the smallest surface area at a given volume? It can be proved by the calculus of variations that such a container must be a sphere. Even in the example of an aluminum can, the optimal dimensions appear to be as close to those of a sphere as the cylindrical geometry would allow: The height and diameter are the same. Should only spherical containers be used to "go green"? To answer this question, a far more complicated optimization problem must be studied. For example, spheres are not optimal for storage and hence for transportation; rectangular containers are far better. Storage maintenance and transportation require energy (hence carbon emissions). The production waste for containers of different shapes is different. Finally, what about consumers' reaction to spherical Coke cans in a vending machine or spherical aluminum cans in the supermarket? 28.1. Applications to Economics. In Section 19, we introduced the cost function C(x), which is the cost of producing x units of a certain prod- uct. The derivative C'(x) is the marginal cost. It determines the cost of increasing production from x units to x + 1 units. Let p(x) be the price per unit that a company can charge if it sells x units. The function p(x) is also called the price function. Naturally, it is generally expected to be a decreasing function because the price per unit usually goes down when a larger number of units is sold. The total revenue R(x) = xp(x) is called the revenue function. The derivative R'(x) is called the mar- ginal revenue function. It determines the change in the revenue when the number of units sold increases from x to x + 1. Finally, the profit function P(x) = R(x) - C(x) = xp(x) - C(x) determines the total profit if x units are sold. Its derivative P'(x) determines the change in the total profit when the number of units sold increases from x to x + 1. The standard optimization problem here is to minimize costs and maximize revenues and profit.  150 4. APPLICATIONS OF DIFFERENTIATION EXAMPLE 4.13. A small store sells jeans at a price of $80 per pair. Every week 60 units are sold. The cost to the store for 60 units is $2500, including the cost of transportation. A market survey indicates that, for each $10 rebate offered to buyers, the number of units sold will increase by 20 a week. Also, the purchase and transportation costs will go down by $2 per each weekly order increase of 5 units. How large a rebate should the store offer to maximize its profit? Solution: 1. What is known about the price function p(x)? First, its value at a particular number of sold units x = xo = 60 is po = p(60) 80. Also, if x increases by an amount of Ax= 20, the price function decreases by Ap = 10 (the rebate). Thus, the ratio m = -Ap/Ax -1/2 is the rate of change of p(x) (the minus sign indicates the decrease in p(x)). So the price function is p(x) =po + m(x - xo) = 80 - 2(x - 60) = 110 - 2x. 2. What is known about the cost function C(x)? First, its value at a particular number of supplied units x = xo = 60 is Co = C(60) 2500. Also, the cost function decreases by AC = 20 if x increases by Ax = 5. So the ratio M = -AC/Ax= -4 is the rate of change of C or the marginal cost. Therefore, C(x) = Co + M(x - xo) = 2500 - 4(x - 60) = 2740 - 4x . 3. One has to maximize the profit function: P(x) = xp(x) - C(x) = 114x - z2x - 2740. Since P'(x) = 114 - x, the function has one critical point x = 114 at which P(x) attains its absolute maximal value by the first derivative test for absolute extreme values. 4. If x = 114 units can be sold, the price per unit is p(114) = 110-57 = 53; that is, the rebate should be p(60) - p(114) = 80 - 53 = 27. Thus, the store should offer a rebate of $27 to maximize its profit. Note also the increase in the weekly profit: P(60) =$2300 whereas P(114) $3758. Remark. In fact, the linear (tangent line) approximation has been used to get the unknown price and cost functions in the previous example. This is a benefit of market surveys: They estimate the derivatives (or trends) of the price functions. Naturally, an increase in sales leads to a decrease in the demand for that particular item. So, after a successful rebate campaign, the store would need a new market survey to estimate p'(114) and get the linear approximation at x =114. The price may go  28. OPTIMIZATION PROBLEMS 151 up then. Similarly, the cost function is generally highly nonlinear. Its linearization near a particular x =0zo cannot be valid for all x > xo. Indeed, in the previous example, it vanishes at x = 685 and becomes negative after that, which cannot possibly be true. 28.2. Exercises. (1) Among all rectangles of a given area S, find one whose perime- ter is minimal. (2) Two ships move along straight lines with constant speeds u and v, and the angle between the lines is 0. Find the minimal distance between the ships if at some moment of time their distances from the point of intersection of the lines were equal to a and b, respectively. (3) Find all angles of a right-angled triangle of the maximal area if the sum of its cathetus and hypotenuse is constant. (4) A piece of wire 1 m long is cut into two pieces. One is bent into a square and the other into a circle. Where should cuts be made if the sum of the areas of the square and circle is to be an extreme? Which of these extremes are relative maxima and which are relative minima? (5) Show that of all triangles inscribed in a circle the equilateral triangle has the greatest area. (6) A tank has the form of a cylinder with hemispherical ends. If the volume is to be V m3, what are the dimensions for a minimum amount of material? (7) The demand for a certain article varies inversely as the cube of the selling price. If the article costs 20 cents to manufacture, find the selling price that yields the maximum profit. (8) A man is in a boat 1 mile from the nearest point, A, of a straight shore. He wishes to arrive as soon as possible at a point, B, 3 miles along the shore from A. He can row 2 miles per hour and walk 4 miles per hour. Where should he land? (9) The stiffness of a rectangular beam varies as the product of the breadth and the cube of the depth. Find the dimensions of the stiffest beam that can be cut from a cylindrical log whose radius is R. (10) A factory A is located at a (shortest) distance a miles from a railroad that goes from north to south through a town B. If C is the point on the railroad at a distance a from A, then B is b miles to the north of C. The cost of shipping by railroad is p dollars for 1 ton per mile and the shipment by truck costs  152 4. APPLICATIONS OF DIFFERENTIATION q > p dollars for 1 ton per mile. At what angle to the railroad should a truck road from A be constructed to minimize the costs of shipping per ton from A to B? (11) If the cost per hour for fuel required to operate a given steamer varies as the cube of its speed and is $40 per hour for a speed of 10 miles per hour, and if other expenses amount to $200 per hour, find the most economical rate to operate the steamer a distance of 500 miles. (12) A channel of width b meters joins a river of width a meters at a right angle. What is the maximal length of a ship that can enter the channel from the river? (13) A railroad company agreed to run a special train for 50 pas- sengers at a uniform fare of $10 each. In order to secure more passengers, the company agreed to deduct 10 cents from this uniform fare for each passenger in excess of the 50 (i.e., if there were 60 passengers, the fare would be $9 each). What number of passengers would give the company the maximum gross receipt? (14) A sheet of paper for a poster is to contain 16 square feet. The margins at the top and the bottom are to be 6 inches, and those on the sides 4 inches. What are the dimensions if the printed area is to be maximal? (15) A taxi company charges 15 cents a mile and logs 600 passenger- miles a day. Twenty-five fewer passenger-miles a day would be logged for each cent increase in the rate per mile. What rate yields the greatest gross income? (16) Two roads intersect at right angles, and a spring is located in an adjoining field 10 m from one road and 5 m from the other. How should a straight path just passing the spring be laid out from one road to the other so as to cut off the least amount of land? How much land is cut off? (17) Illuminance is a measure of how much the incident light il- luminates the surface. If a source of light of luminosity k is positioned above a plane, the illuminance at a point on the plane is I = k cos /r2, where r is the distance from the source to the point and 0 is the angle between the light ray from the source to the point and the normal to the plane. At what distance above the center of a round dining table of radius R should a light bulb be positioned in order for the table border to have a maximal illuminance?  29. NEWTON'S METHOD 153 (18) Two light sources of luminosity ki and k2 are positioned at points A and B, respectively. Find the point on the straight line segment AB that has the least illuminance if the distance between the sources is a (see the definition of illuminance in the previous exercise). (19) A pointlike source of light is positioned between two nonin- tersecting spheres on the line connecting the centers of the spheres. If the radii of the spheres are R and r (R > r), find the position of the light source such that the sum of illumi- nated areas of the spheres is maximal. (20) A rectangular box with a square base and an open top is to be made. Find the volume of the largest box that can be made from A cm2 of material. (21) A rectangular field containing S m2 is to be fenced off along the bank of a straight river. If no fence is needed along the river, what must be the dimensions requiring the least amount of fencing? (22) If a stone is thrown from a cliff of height h at a speed vo m/s and an angle 0 with the horizontal line, then its trajectory is a parabola: y = h + x tanO8 - x2 9 2v2 cos2 0' where y is the stone height (vertical position), x is the hori- zontal position (all the positions are in meters), and g = 9.8 m/s2 is a constant universal for all objects near the surface of the Earth (the free-fall acceleration). Compare with Equa- tion (4.4). At what angle is a stone to be thrown to reach the maximal range at a given speed vo? 29. Newton's Method Finding roots of a function f (x) is an important problem in various applications. Unfortunately, an analytic solution of the equation f (x) 0 is impossible in many practical cases. For example, consider f(x) = x - e-x. The equation f(x) = 0 is equivalent to x = e-x. The graphs y =x and y =e-x intersect at some x between 0 and 1. So f (x) has a root. But how can it be calculated? Here we present one of the simplest methods, known as Newton's method. It provides a recurrence relation that allows us to compute a root of a differentiable function with any desired accuracy.  154 4. APPLICATIONS OF DIFFERENTIATION 29.1. Newton's Recurrence Relation for Finding a Root. Suppose f(x) has a root near xo. Consider the tangent line approximation of f near xo: L(x) = f(xo) + f'(xo)(x - xo). It is easy to find the root of L(x), which is denoted by xi: L(x) =0 - =x x1 z- Note that the root of L(x) exists if f'(xo) # 0 (otherwise, the tangent line is horizontal and cannot have any root). x2 x1 FIGURE 4.11. Diagram for Newton's method. Pick xo near the root of f. Find the tangent line of the graph of f at xo. Determine the intersection point x1 of the tangent line with the x axis. Find the tangent line to the graph of f at x1 and its intersection x2 with the x axis. By repeating this procedure a sequence of numbers zo, X1, x2, ... is obtained that converges to the root of f, provided xo was chosen close enough to the root. Since L(x) is only an approximation to f(x), the number x1 is closer to the root of f than xo, but does not coincide with it. In other words, the value f(xi) is closer to 0 than f(xo): 0 < If(xi) < If(xo) (the absolute value is necessary if the function takes negative values). Therefore, the tangent line constructed at x = xi, L(x) = f(Xi) - f'(xi)(x - xi), can be expected to approximate f(x) even better near its root because x1 is closer to the root than xo. The root of the new tangent line is given by the same expression as before where xo should be replaced by x1: x2 = x1 - f(xi)/f'(xi). The procedure may be  29. NEWTON'S METHOD 155 recursively repeated to generate a sequence of values zn: f(z12) (4.18) zn+1 = z, - , ,) n = 0, 1, 2, ..., f (zn) provided f'(z) / 0. THEOREM 4.14. If f has a single root r in an open interval and f'(x) / 0 on the interval, then there exists xo sufficiently close to r such that the sequence (4.18) converges to the root lim xn = r . n-oo The convergence of xn to r means that for all large enough n the numbers xn lie in any small interval r - E < xn < r + E for any choice of E > 0. In practice, the root r need to be found only with some accuracy. So the sequence elements need to be calculated with a particular number of significant digits; that is E = 10-rn where m is some positive integer. It is sufficient to apply Newton's recurrence until xz+1 and xn agree to all the relevant decimal places. Then r x= zn1 is correct to the relevant decimal places. EXAMPLE 4.14. Find the root of f (x) = x - e-x that is correct to six decimal places. Solution: 1. Determine the position of the root first. The graphs y = x and y = e-x intersect only once at a point between 0 and 1. So, in any open interval containing the interval (0, 1), f has only one root. 2. Verify the condition f'(x) / 0: f'(x) = 1 + e-x > 0 for all x. 3. Pick an initial value of Newton's sequence as close to the root as possible, e.g. xo = 0. Then Newton's sequence for six decimal places is: xo = 0 , x1 =0.5 , x2 = 0.566311, x3 = 0.567143 , x4 = 0.567143 . So the root r = 0.567143 is correct to six decimal places (in fact, f (0.567143) -4.5 x 10-7). 29.2. Pitfalls in Newton's Method. Unfortunately, there is no unique recipe for choosing an initial point in Newton's sequence. The choice depends very much on the function in question. In practice, it is de- termined by trying different values. A few possible bad behaviors of Newton's sequence are useful to keep in mind.  156 4. APPLICATIONS OF DIFFERENTIATION Choice of the Initial Point in Newton's Method. A poor choice of the initial point xo can produce the value of x1 that is a worse approx- imation to the root than xo. Consider, for example, the function f(x) = x3 - 3x2+2 in the interval [0, 2] and f(x) = 2 when x < 0 and f(x) =--2 when x > 2. This is depicted in Figure 4.12. The function is continuously differentiable because f'(x) = 3x2 -6x approaches 0 as x - 0+ and x - 2-. The function has the root x = 1 and f'(x) <0 in the open interval (0, 2). If 0 < xo < 2 is close enough to either x = 0 or x = 2, then x1 would be outside the interval (0, 2). Note that the actual behavior of f(x) outside the interval [0, 2] is not relevant for the conclusion. The essential point here is that such a situation is likely to occur when f'(xo) is close to 0. 1 o x2 x1-. FIGURE 4.12. A bad choice of the initial point for New- ton's method. The function has the root r = 1, f (1) = 0. The derivative vanishes at x = 0 and x= 2. If xo is set too close to x= 2, then the slope f'(xo) of the tangent line is small; the tangent line is close to a horizontal line. In this case, the next element of Newton's sequence xz+1 may happen to be outside the interval [r - xz, r + xn] so that Newton's sequence does not converge to the root r. Cycles in Newton's Method. A poor choice of the initial point may lead to a cycle in Newton's sequence. Take f(x) = x3 - 2x + 2 and co0 0. Since f'(x) = 3x2 - 2, the next elements are x1 = 0 - 2/(-2) = 1, x2 = 1-1/1 = 0 =cco. That is, Newton's sequence is a cyclic sequence, which never converges. The initial point must be taken closer to the root. Instabilities of Newton's Method. If f'(cc) -~ too as cc approaches a root r (the graph y =f(xc) has a vertical tangent line at the root),  29. NEWTON'S METHOD 157 Xn- -Xn Xn -Xn+2 FIGURE 4.13. A cycle in a Newton's sequence for f(x) x. The function has the root r = 0, f(0) = 0, and it also has the vertical tangent line at the root because the derivative diverges f'(x) -- oc as x - r = 0. A Newton's sequence exhibits a cycle; it oscillates around the root r never converging to it. In general, such a cycle can occur in a Newton's sequence for a function whose behavior near a root r is f(x) ~ a(x - r)2v, where a is a constant and v = 1/4. Furthermore, if 0 < v < 1/4, then a Newton's sequence does not converge. Newton's sequence may oscillate around r, never converging to it, or it may diverge for any initial point. To understand this phenomenon, suppose f(x) behaves near its root r as f(x) ~ a(x - r)2v, v > 0, and a is constant. The derivative f'(x) 2va(x - r)2v-1 diverges as x -- r when v < 1/2. Since f(x)/f'(x) = a(x - r), Newton's sequence (4.18) on+1 = on - v(n - r) = xn(1 - 2) + vr can also be written as n+1- r =xn(1 - ) + r - r =q(xn -r), where q = 1 - v. Apparently, the condition xn - r is equivalent to yn = on - r -- 0. But the sequence yn+1 = qYn = g2yn-1 = ' ' = qn+1, converges to zero only if q = 1 - yv < 1 or v > 1/4 unless yo = 0 (i.e., if the root is accurately guessed!). Recall the asymptotic behavior of the exponential function ax -- oc as x -- oc if a > 0 and ax - 0 if 0 < a < 1. For example, for f(x) = x1/3 (v = 1/6), Newton's sequence diverges: xn+1 = (1 - 3)xn = -2xn for any choice of the  158 4. APPLICATIONS OF DIFFERENTIATION initial point xo / 0 because |q"| |(-2)= 2. For f(x) = |x1/2 (v = 1/4), Newton's sequence oscillates xz+1 =(1 - 2)xz = -xz (see Figure 4.13). 29.3. Understanding Money Loans. Suppose that one takes a loan of P dollars (the principal) for n months with an annual interest rate of I%. What is the monthly payment? It is calculated as follows. The interest rate per month is x = I/12. For example, an annual interest rate of 6% means that I = 0.06 and x= 0.06/12 = 0.005. Each payment includes the payment toward the principal and the interest. Let Fk be the amount yet to be paid after k monthly payments. It is called the future value of the loan. The sequence Fk satisfies the conditions: F0 = P and Fn = 0 (the loan and interest are paid off after n payments). Let A be the monthly payment. Then F1=P+Px-A, F2=F1+Flx-A,...,Fk=Fk_1+Fklx-A. Here F1 is the future value of the loan after one payment, which is the loan P plus the monthly interest P minus the payment A. After one payment, the loan value is F1. So, after one more payment, its value is the value F1 plus interest Fix minus the payment A, and so on. After n payments, Fn = F,1(1+ zx) - A = F-2(1+ cc)2 - A[(1+ cc) + 1] = Fo(1 + z)" - A[(1 + z)"-1 + (1 + z)n-2 + -.-. + (+ z) + 1 P(1+ )< - A (+ where, in the last equality, the geometric sum formula sn=1l+q+q2+---+q" 1= q q-1 has been used. The latter is proved by noting that the sums qs= q+ q2 + - --+ q"-1 +q" and sn have the same terms except 1 in sn and q" in qsn so that qsn - s = q"- 1 or s , =(q" -1)/(q - 1). Since Fn = 0, the monthly payment is (4.19) A = 1 - (1 + c)- For example, a loan of $200,000 for 10 years at a fixed annual inter- est rate of 6% implies 120 monthly payments of $2220.41. Indeed, in Equation (4.19), substitute cc= 0.06/12 =0.005, nm 120, and P  29. NEWTON'S METHOD 159 200,000, then A 2220.41004. The total amount paid after 10 years is 120 x A = $266,449.20. The interest paid is nA - P = $66,449.20. When selling a car, a dealer might offer a monthly payment for a few years if a customer cannot afford to pay the price in full. In this case, the loan amount P is the price of the car; the monthly payment A and its number n are known. To assess the deal, one has to figure out the interest rate before signing up. It might be the case that the loan for a higher-quality car, meaning a higher price and higher monthly payments, might have a lower interest rate, than the loan for a cheaper car (smaller monthly payments). Knowing the interest rate, one can also shop for a loan at a lower rate elsewhere (e.g., banks) to buy a car. If A, P, and n are given, then x can be found by solving Equation (4.19), which can be written in a more convenient form as (4.20) f(x) =Px(1+x)" -A(1+x)" +A=0. In other words, this is the root-finding problem! It can be solved by Newton's method. The number x should be found up to five decimal places, which is sufficient our purposes. EXAMPLE 4.15. A dealer offers a car at a price of $10,000. It can also be sold for payments of $217.42 per month for 5 years. There is another car being offered at a price of $15,000, which can also be sold for payments of $311.38 per month for 5 years. Which loan has a lower interest rate? Solution: 1. For the first car, one has to find the root of Equation (4.20) if A = 217.42, P = 10,000, and n = 5 x 12 = 60. It is convenient to initiate Newton's sequence at x1 = 0.01, which corresponds to an annual interest rate of 12% (i.e., I = 0.12 and x= 0.12/12 = 0.01). Up to five decimal places, Newton's method yields x = 0.00917, which corresponds to I = 12x = 0.11004, or an annual interest rate of 11%. 2. For the second car, one has to find the root of Equation (4.20) if A = 311.38, P = 15,000, and n = 5 x 12 = 60. Newton's method, initiated again at x1 = 0.01, yields the root x = 0.00750 (up to five decimal places). This corresponds to an annual interest rate of 9%. So the second loan has a lower interest rate. D It is interesting to note that the car prices differ by 50%. Similarly, the monthly payments appear in a similar proportion 311.38/217.42 1.43. The offers might look like as nearly the same deal. In fact, they are not!  160 4. APPLICATIONS OF DIFFERENTIATION 29.4. Exercises. (1) Find the number of real roots of the equation and estimate the intervals that contain a single root: (i) x3 - 6X2 + 9x - 10 = 0 (ii) x3 - 3x2 - 9x + h = 0 (iii) 3x4 - 4x3 - 6x2 + 12x - 20 = 0 (iv) c5 - 5x = a (v) Incc= kx (vi)ex =ax2 (2) Under what condition does the equation x3+ pz+ q = 0 have (i) a single real root and (ii) three real roots? Depict the corresponding regions of (p, q) in a plane. (3) Find the roots of the given equation up to five decimal places: (i) cos x = 2x (ii) ex - e-x = 1 - x2 (iii) tan-1 x X3 (iv) tan-1 x = ln (v) ln(1 + X2) = 4 - x (vi) c5 + x - 4 = 0 (vii) (4 -x2)2 -x+4=0 (4) Newton's method is based on the linear approximation of the function at a sample point xc to generate the next point cc+1 of Newton's sequence. This approximation does not take into account the concavity of the function at zc. Generalize New- ton's method by using the Taylor polynomial T2(x) at xc to generate cc+1 as a root of T2. Take any of the above exer- cises and compare the convergence of Newton's method with its generalization (i.e., the numbers of steps needed to obtain the root correct to the same number of decimal places, e.g., 6, 7, or 10, starting with the same initial point zo). (5) Consider a loan of $250,000 at an annual low interest rate of 4% for 15 years. Find the monthly payments. The interest rate was not fixed and is subject to change so that the monthly payments may increase up to 20%. How much may the annual interest rate increase (percentagewise)? (6) A car dealer offers a car at a price of $15,000 for 36 monthly payments of $477. What is the interest rate? (7) The fixed annual interest rate on a mortgage is 7%. For how long should one take a loan if one wants to pay in total inter- est no more than half of the principal? Does the maximum loan period increase or decrease with increasing or decreasing  30. ANTI DERIVATIVES 161 interest rate? Does the answer change if the payments will be made every 2 weeks (i.e., 30 payments per year instead of 12)? (8) Find the root of the equation tan-1x = 1 - x correct up to four decimal places by initiating Newton's sequence at xo = 1. Investigate the dependence of the number of needed iterations to achieve this accuracy on the initial point by taking xo = n, where n = 1, 2, ..., 10. 30. Antiderivatives In many practical problems, a function is to be recovered from its derivative. For example, if the velocity is given as a function of time, v = v(t), one might want to find the position as a function of time, s = s(t), where s'(t) = v(t). What is s(t)? DEFINITION 4.8. A function F is called an antiderivative of f on an interval I if F'(x) = f (x) for all x in I. For many basic functions, it is not difficult to find the corresponding antiderivative. For example, from the rule (xtn+l)' = (n+1)x", it follows that if f (x) = ", n / -1, the antiderivative is F(x) = zn+1/(n+1). It has also been proved that (ln x|)' = 1/x. So the function F(x) =ln x| is the antiderivative of f (x) = 1/x for all x / 0. 30.1. Uniqueness of the Antiderivative. Suppose F'(x) = f (x) for all x in an interval (a, b). Is such an F(x) unique? This question is answered by Corollary 4.1. Indeed, let F(x) and G(x) be antiderivatives of f(x), that is, F'(x) = G'(x) = f(x) on (a, b). By Corollary 4.1, F and G may only differ by a constant: G(x) = F(x) + C. Recall that Corollary 4.1 does not hold for the union of disjoint intervals. Thus, any two antiderivatives of the same function may differ at most by a constant on an interval. THEOREM 4.15. If F is an antiderivative of f on an interval I, then the most general antiderivative of f on I is F(x) +C, where C is an arbitrary constant. For example, the general antiderivative of the power function z" where n is a positive integer, is n2+1  162 4. APPLICATIONS OF DIFFERENTIATION because x" is defined on the single interval (-oc, oc). The function f(x) = 1/x is defined on the union of disjoint intervals (-oc, 0) and (0, oo). So the general antiderivative is 1 Ilnx+C1 x>0 F'(x) = f (x) = - -> F (x)=' x ln(-x) + C2, X < 0 where C1 and C2 are arbitrary constants. In what follows, the domain is always assumed to be a single interval, unless stated otherwise. The nonuniqueness of the antiderivative is not a drawback of the concept but rather a great advantage. This is explained by the following example. The velocity of a piece of chalk thrown vertically upward with a velocity of vo is v(t) = vo - gt, where g = 9.8 m/s2 is the acceleration of a free fall. At t = 0, the chalk has a velocity of v(0) = vo. Then it begins to slow down (v(t) decreases because of gravity). Eventually, at t = vo/g, the chalk stops and begins to fall back. If h(t) is the height of the chalk relative to the floor, then h'(t) = v(t); that is, the height is an antiderivative of v(t). It is easy to find a particular antiderivative of v(t) using the antiderivative of the power function: h(t) = vot - gt2/2 (indeed, h'(t) = vo-gt). What is the physical significance of the general antiderivative h(t) = C + vot - gt2/2? It appears as if the position of the chalk relative to the floor is not uniquely determined. In particular, h(0) = C is the height at the very moment when the chalk was thrown upward. But the chalk could be thrown upward at 1 m above the floor or 2 m above it with the very same initial velocity. So, in both cases, v(t) is the same, while the h(t) are not. In the first case, h(0) = 1, whereas in the second case, h(0) = 2. Thus, the constant C can be fixed by specifying the value of the antiderivative at a particular point. This feature of the general antiderivative can also be visualized by plotting the graphs y = F(x) + C for different values of C. All such graphs are obtained from the graph y = F(x) by rigid translations along the y axis. If one demands that the graph y = F(x) + C should pass through a particular point (zo, yo), then C is fixed: yo = F(zo)+C or C = yo - F(zo). For example, find f (x) if f'(x) = 3x2 and f (2) = 1. The general antiderivative of 3x2 is f (x) = x3 + C. From f (2) = 1, it follows that f (2) = 8 + C = 1 or C = -7. Therefore, f(x) = x3 - 7. 30.2. Linearity of the Antiderivative. Let F and G be antiderivatives of f and g, respectively. Then an antiderivative of f + g is F + G. An antiderivative of kf, where k is an arbitrary constant, is kE. These properties are easily verified. Indeed, (F + G)' =F' + G' =f + g and (kF)' =kF' =kf, where the linearity of the derivative has been  30. ANTI DERIVATIVES 163 used. In other words, antidifferentiation is a linear operation just like differentiation itself. 30.3. Antiderivatives of Basic Functions. An antiderivative of the power function has been found by studying the derivative of the power and logarithmic functions. The idea is useful for other basis functions. Their antiderivatives can be found by reading the table of derivatives of basic functions backward, that is, from the right to left. Table of antiderivatives of basic functions F(x) f (x) = F'(x) F(x) f (x) = F'(x) sin x + C cos x - cosx + C sin x tan x + C (sec x)2 - cot x + C (cscx)2 sin-1 x + C 1-x2 -cos-1 x + C1 tan-1 x+C 1+2 -cot-1 x+ C1X2 j+lc xmn#-1 lnxl+C =0 ex +C ex ; +C ax, a >Oa#1 2In }i± + C 1 2 lnx+ x2+1|+ CX1 As noted earlier, here F(x) is given on a single interval in the domain of f (x). The table of antiderivatives of basic functions combined with the linearity of antidifferentiation is a good source of antiderivatives of more complicated functions. The chain rule (f(g(x))' = f'(g(x))g'(x) can also be used to obtain antiderivatives. EXAMPLE 4.16. Find the general antiderivative of x/(x2 + 1). Solution: By the chain rule 1 2x__ (ln(x2 + 1))' =I - (X 2 + 1)'2= X2+1 X2+1 Therefore F(x=) = ln(X2 + 1) + C. D EXAMPLE 4.17. Find the general antiderivative of f)(x) = e-2x + cos(4xc) + x2/(1 + 92). Solution: 1. By the linearity of the antiderivative, it is sufficient to find antiderivatives of e-2x, cos (4xc), and x2/(1+ cc2). The general an- tiderivative is obtained by adding a general constant to the sum of the  164 4. APPLICATIONS OF DIFFERENTIATION particular antiderivatives of the previous three functions. 2. From (e-2x)' =_-2e-2x, it follows that (-e-2x/2)' = e-2x. Hence, an antiderivative of e-2x is -e-22x/2. 3. Similarly, from (sin(4x))' = 4 cos(4x), it follows that an antideriva- tive of cos(4x) is sin(4x)/4. 4. The table of derivatives does not appear helpful in the case of x2/(1 + x2). However, a simple algebraic manipulation leads to the goal: x2 1+x2-1 1 1+X2 1+x2 - +X2 So its antiderivative is x - tan-1 x. Thus, the general antiderivative reads: F(x) =±e-2x +± -sin(4x) + x -tan-1 x + C. 2 4 because f(x) is defined on the single interval (-oc, oc). Q 30.4. Antiderivatives of Higher Order. What is F(x) if F"(x) = f(x) for a given f(x)? Or, more generally, what is F(x) if F(n)(x) = f(x)? A function F that satisfies the latter condition is called an antideriv- ative of f of the nth order. To find it, one has to antidifferentiate f n times. For example, F"(x) = 6x. Taking the first antiderivative of f (x) = 6x, one gets F'(x) = 3x2. Taking the antiderivative one more time yields F(x) = x3. What about the uniqueness of higher-order antiderivatives? To find the general antiderivative of a higher order, each time antidifferentiation is carried out, the corresponding general antiderivative must be used. In the preceding example, the general antiderivative of f(x) = 6x is 3x2 + C1, where C1 is an arbitrary constant. Hence, F'(x) = 3x2 + C1. Its general antiderivative reads F(x) = x3 + C1x + C2, where C2 is another arbitrary constant. Thus, the general second antiderivative can be obtained from a particular one by adding a general function whose second derivative is 0, which is a general linear function: (Cix + C2)" = 0. Similarly, if F(x) is a par- ticular function that satisfies the condition F(n)(x) = f(x), then the general antiderivative of the nth order is F(x) + C1x-1 + C2xa-2 + ... + Cn-1x + Cn , where C1,... ., Cn are arbitrary constants. Indeed, the nth derivative of a polynomial of degree n- 1 is 0. Note that this analysis is justified only when f was defined in an interval. The reader is instructed to analyze the situation when the domain of the function f consists of disjoint  30. ANTI DERIVATIVES 165 intervals and, in particular, to consider higher order antiderivatives of the inverse power function x-", where n is a positive integer. The following example illustrates the significance of arbitrary con- stants in general higher-order antiderivatives. EXAMPLE 4.18. Any free-falling object near the surface of the Earth has the free-fall acceleration of 9.8 m/s2. A piece of chalk is thrown vertically upward at a speed of 7 m/s and at 1.5 m above the floor. When does the chalk hit the floor? Solution: 1. Let h(t) be the height of the chalk relative to the floor. Then its velocity is v(t) = h'(t), and its acceleration is a(t) - v'(t) = h"(t). Since all free-falling objects have an acceleration of 9.8 m/s2, one has h"(t) - -9.8. The minus sign indicates that the acceleration is directed downward. 2. The general second antiderivative of the constant function -9.8 is t2 h"(t) -9.8 -- h(t) = -9.8- + C1t + C2, 2 where C1 and C2 are arbitrary constants. 3. To fix C1 and C2, the initial conditions of the motion must be used. The initial velocity is v(O) = 7. Since v(t) = h'(t) -9.8t + C1, one infers that v(O) = C1 = 7. The initial height is h(O) = 1.5. Hence, h(0) = C1 = 1.5. 4. The height is h(t) -9.8t2/2+7t+1.5. The chalk hits the floor when its height vanishes, that is, at the time moment t > 0 when h(t) = 0. A positive root of the quadratic equation -9.8t2/2 + 7t + 1.5 = 0 is t 1.62 s. The maximum height reached by the chalk is 4 m. Why? Q 30.5. Exercises. (1) Use the table of antiderivatives of basic functions to find an antiderivative of each of the following functions: (i) f(x) =(3 - X2)2 (ii) f (x) X2(1 - x)2 (iii) fv(x) - (1 - x)(1 - 2x)(1 - 3x) (iv) f (x) = sin(4x) + x (v) f (x) =1/(X2+4) (vi) f (x) =esx _+ e-3x _+ (vii) f (x) =cos2 x (Hint: 2 cos2 x =1 + cos(2x)) (viii) f (c) =sin(ax) cos(bz) (ix) f (c) =cos(ax) cos(bz) (x) f (c) =sin(ax) sin(bz)  166 4. APPLICATIONS OF DIFFERENTIATION Hint: Express the products of trigonometric functions via the sum of trigonometric functions. (xi) f (x) = (x + 1)/9/2 (xii) f(x) = 3x2(1 + v/) (xiii) f (X) =/(x2 + 1) (xiv) f(X) =1(x2 + a2) (xv) f (X) (X 2 + a2)/(X2 + b2) (xvi) f() =(X4 + 1)/(X2 + 1) Hint: z z44 + 2X2 - 2X2 (xvii) f(c) =IC (xviii) f (c) = ex2 (xix) f(x) = I - a| + x - bl, where a < b (xx) f() =sincz| (xxi) f(x) = 1/(x2 - 1), where -1 < c < 1 Hint: 1/[(x - a)(x - b)] = A[1/(x - a) - 1/(x - b)]; find A. (xxii) f(x) = 1/v/c2 - a2, where cc> a > 0 (xxiii) f (x) = 2X/3X (xxiv) f(x) (2x + 3-x)2 (2) Use the table of antiderivatives to find the general second an- tiderivative of each of the following functions: (i) f ( x (x - 1) (ii) f(x) =-z1/3 (iii) f (x) = sin(2x) (iv) f(cc) - IC (v) f(x) = fo = const if c E [a, b] and f(x) = 0 otherwise (vi) f(cc) - I + 1 (vii) f (x) = sinxcos(2x) (viii) f (x) = e3x + e-3x (ix) f (X) =z/(x2 + 1)2 (x) f(x) 2x - 3-x (3) Find the general antiderivative of the function whose domain is not a single interval: (i) f(cc) =1/(x - 1) (ii) f (X) 1(x2 _ 1) (iii) f (X) = x-4/3 (iv) f (X) = (1- )3 /c4/3 (v) f(cc) =sec2 cc (vi) f(cc) c sc2 cc (4) A car that was at rest accelerates at a rate of 1 in/s2 for 1 minute. Then it decelerates at a rate of 0.5 in/s2 until it stops. Find the distance traveled by the car.  CHAPTER 5 Integration 31. Areas and Distances 31.1. Limit of a Numerical Sequence. A numerical sequence a1, a2, a3, ... is a function whose domain is the set of all positive integers. In other words, it is a rule that assigns a unique number an to each integer n. For example, n an =n , n = 1,2,3,.... If f(x) =x/(x + 1), then the sequence an is formed by the values of f(x) at x = n, an = f(n). It follows from Section 10 that x 1 lim f(x) = lim =lim 11= 1. X-o x-ox+1 x-oo1+- This means that for any given E > 0, there exists a real number N such that values of f (x) lies in the interval 1-E < f (x) < 1+E for all x > N. Since E is arbitrarily, one can say that the values of f are arbitrarily close to 1. In particular, for n large enough, the values an = f(n) are arbitrarily close to 1, and one can conclude that lim an = lim 2= 1. nh- oo m-o n + 1 Since numerical sequences are just a special case of a function, a formal definition of the limit of a sequence can be adopted from Section 10. DEFINITION 5.1 (Limit of a numerical sequence). Let an, n 1, 2, ..., be a numerical sequence. The sequence is said to converge to a number a if for any E > 0 there is an integer N such that |an - a| N, and in this case one writes lim an = a n-0 If the sequence has no limit, it is called divergent. This definition also means that a convergent sequence has all but a finite number of its terms in an arbitrarily small interval (a - E, a + E). For example, take an = 1/np where p > 0. Then evidently limn-o an 0. Indeed, given any small number _> 0, it follows that the inequality 1 1 |a -0 =a = - N where N is an integer such that N > 1/el/p. Therefore, only a finite number of terms of the sequence lie outside the interval (-E, e) and, hence, the sequence converges to 0. This sequence is formed by the values of the function f(x) = 1/zp, an = f(n). In Section 10, it has been shown that f(x) -- 0 as x - oc, which also implies that f(n) -- 0 as n - oc. The relation between the limits of sequences and the limits of functions at infinity lim f (x) = a -- lim f (n) = lim an = a x- oo -oo -oo is a useful tool to calculate the limits of sequences. Note that the converse of the above assertion is not true. Why? A final remark is that the basic limit laws and the squeeze principle hold for the limits of sequences. 31.2. Area Under a Graph. Consider the linear function f(x) = x. What is the area below the graph y = f(x) and above the interval 0 < x < 1? This question is easy to answer because the area in question is the area of the right triangle with catheti of unit length: A = 1/2. Let f(x) = x2. What is the area now? To calculate it, consider a partition of the interval [0,1] by n segments on length 1/n. The partition is defined by the set of points zo = 0, x1 =1/In, x2 2/n, ... , xn_1 = (n - 1)/n, and xc =n/n = 1, that is, Xk= k/n, where k = 0, 1, 2, ... , n. The area under the parabola y = x2 over the interval [0, 1] is the sum of the areas Sk under the parabola over the partition interval [Xk_1, Xk] where k = 1, 2, ... , n, A=S1+S2 + -+ Sn. In the interval [Xk_1, xl], the function f(x) = x2 attains its maximum value at x =Xk and its minimum value at x =Xk_1. Therefore the area Sk cannot exceed the area of a rectangle with base 1/n and height f(zk) = (k/n)2. Let us denote this upper bound by S = k2/n3. The area Sk is greater than the area of a rectangle with base 1/n and height f(zk_1) = (k-1)2/n2. The lower bound is denoted by S/j= (k-1)2/n3. Thus, SL (k -1)2 < k2 in3 So, the area A is bounded above by the sum of Sg$ and below by the sum of S/i: for any number in of partition segments.  31. AREAS AND DISTANCES 169 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 1 1 3 1 0 1 1 3 1 4 2 4 4 2 4 FIGURE 5.1. The upper and lower bounds for the area under the graph y = f(x) = x2 for n = 4 partition in- tervals in [0, 1]. The upper bound is obtained by taking the maximum value of f on each partition interval (left panel). The lower bound is obtained by taking the mini- mum value of f on each partition segment (right panel). When n increases, the upper bound decreases, while the lower bound decreases, both approaching the area under the graph as n -- oc. Let us calculate the difference 2k-1 2n-1 2 n3 n3 n2 for any k = 1, 2, ... , n; in the second inequality, the condition k < n has been used. This inequality allows us to estimate the difference An - An : 0 < A - An = (Sl -S+) + (S2 - SS) +.--+ (Sn Sn ) 2 2 0 on an interval [a, b] is (5.2) A = lim A*n = lim [f(z*) Az + f (z*) Az + -.-. + f (z*) Az for any choice of sample points x*. Let us assess this definition. Any continuous function attains its maximum and minimum values on a closed interval. Let Mk and mk be, respectively, the maximum and minimum values of f(x) on the interval [Xk_1, Xk]. If Sk is the area under the graph y = f(x) on the interval [Xk_1,Xk], then S/f = mk Ax Sk < S = MkAx. The area S* = f(x*) Ax of the rectangle with base Ax and height fz(x*) is a continuous function of x on the interval [Xk_1, Xk]. Therefore, S* must take all the values between its minimum and maximum values, SL and S . In particular, S= S for some x* E [Xk-1, Xk]. Thus, for any fixed n, there is a choice of sample points such that A* = A. Continuing the analogy with the example of f(x) = x2, let us show that the limit (5.2) is independent of the choice of sample points, provided the lower sums A = Sl + - - - + SL and the upper sums A 5 = Si+... + S converge to the same number as n - oc. Indeed, for any choice of sample points S S* < Sj and, hence, by tak- ing the sum over the partition in the latter inequality, one infers that An < A* |AU- |< _ -AL (nAL U Thus, if An - An - 0 as n - o0, then A*n - A for any choice of partition. The following theorem holds. THEOREM 5.1. Let f be a continuous function on [a, b]. Suppose that for any partition x0 = a < x1 < x2 < - - - < xn = b, the length of the largest partition interval, An= maxk AXk, Axk = Xk - xk_1, k 1, 2, ...,n, decreases as the number n of partition intervals increases, An > An+1. Then the upper and lower sums converge to the same limit as n - oo. This theorem justifies the definition (5.2). Note also that the parti- tion is not generally required to be equispaced. The above theorem only requires that the length An of the largest partition interval decreases with increasing the number of partition intervals (A, (b- a)/n> Ax for an equispaced partition).  172 5. INTEGRATION 31.4. Approximating the Area Under a Graph. In practice, Equation (5.2) can be used to find the area under the graph that is correct to any de- sired number of decimal places. Take a partition of the interval [a, b], e.g., fix some n so that AOx= (b- a)/n. Choose sample points Xk -1 < x*< X. Convenient choices might be the left points x* =zk_1, the right points x* = Xk, or the midpoints x*= (zi + x)/2. Calculate the sum A*, keeping the desired number of decimal places. Refine the partition by, for example, doubling the number of segments, and cal- culate An. If A*n and A3n coincide in the desired number of decimal places, then A = A3n is correct to that number of decimal places. If not, refine the partition further and compute An and compare it with A;n and so on, until the needed accuracy is reached. For any n, the absolute error of the approximation may estimated by the inequality on the right in (5.3). 31.5. Sigma Notation for Sums. To avoid writing lengthy expressions for sums of an arbitrary number of terms, it is convenient to adopt the following notation: n A*n= S1+ S*2+.-.-.+ S*,= kSg k=1 where the index k is called the summation index. The symbol E means adding all S*, starting with k = 1 up to k = n. For example, the geometric sum formula can now be written as (5.4) 1qq2q"= q 1 k0 q- 31.6. The Distance Problem. If an object moves with a constant veloc- ity v during a time interval a < t < b, then the distance traveled by the object is D = v(b - a). How does one calculate the distance if the speed is a nonconstant continuous function of time v= v(t) > 0? Let D(t) be the distance as a function of time a < t < b. It satisfies the condition D(a) = 0. Since v(t) > 0, the object travels in the same direction all the time, and D(t) increases because D'(t) - v(t) > 0. Thus, D = D(b). To calculate D(b), consider a partition of [a, b] by interval [t1, tk] where tk = a + At k, At = (b - a)/n, k =0, 1, ... , n. The distance ADk D(tk) - D(tk_1) traveled by the object in the time interval [tk_1, tk] can be found by the mean value theorem: D(tk) - D(tk_1) =v(t~e) At for some t*, in [tk_1, tk]. Recall that v(t~e) is the average velocity over the time interval [tk_1, tk]. The total distance is D =AD1 + - - + AD2. On the other hand, points  31. AREAS AND DISTANCES 173 t* represent a particular choice of sample points in the definition (5.2) applied to a continuous function v(t). Therefore, D is the area under the graph of v(t) and, hence, can be calculated with any choice of sample points t% , not necessarily with those at which v coincides with the average velocity in each partition interval: n D = lim v(t ) At , k=1 Furthermore, by the condition D'(t) = v(t) the function D(t) is the antiderivative of v(t) satisfying the initial condition D(a) = 0. If F(t) is a particular antiderivative of v(t), then D(t) and F(t) can differ only by a constant, D(t) = F(t) + C for a < t < b. The constant C is fixed by the condition D(a) = 0 and, hence, C = -F(a). The distance traveled is D = D(b) = F(b) - F(a) This establishes the following relation between the area under the graph of v(t) and an antiderivative F(t) of v(t): (5.5) k=1 EXAMPLE 5.1. A moving object slows down so that its velocity is v(t) = e-2t. What is the distance traveled by the object during the time interval 0 < t < 1? Solution: Let t = (k - 1)/n, Then v(t ) At At qk = 1/n so that tk _ k/n, k = 0, 1, ... , n. Take 1, 2, ... , n (the left points of partition intervals). -1/n, where q = e-2/n. The distance traveled is D= lim !-1qk n-oo n 1 q" - 1 lim- n-oin q - 1 1 - 2 limn-o n(1 - e-2/n) where the sum formula the denominator, let x: indeterminate form (1 - l'Hospital's rule: (5.4) has been used. To compute the limit in = 1/n, that is, x - 0+. The limit becomes the - e-2x)/x of type , which can be resolved by lim n(1 - e-2/n) n-oo 1 -e-2x lim = x-o+ x lm2e-2x lim = 2 x-O+ 1 lim ( )' x-o+ (xc)'  174 5. INTEGRATION Thus, the distance traveled is D = (1 - e-2)/2. D Alternative solution: Using the table of antiderivatives 1-t F'(t) = v(t) =e-21 F(t) = Ie-2t By Equation (5.5), D = F(1) - F(O) = (1 - e-2)/2. Note that any particular antiderivative can be used. When compared to the previous solution, this one looks like cheat- ing! More to the point, take v(t) = t2 (the example discussed at the be- ginning of this section). Its particular antiderivative is F(t) = t3/3. So the distance traveled, or the area under the graph of t2, is F(1)-F(0) 1/3. It turns out that the relation (5.5) between an antiderivative of a function and the area under the graph of the function is not specific for the distance problem. Its generalization will be established with the help of the concept of the definite integral. 31.7. Exercises. (1) Find explicit formulas for the upper and lower sums, An and An, for f(x) = 2x + 1 on [0, 2] using an equispaced parti- tion. Find the limits of An and An as n - oo. What is the geometrical significance of this limit? (2) Find explicit formulas for the upper and lower sums, An and An, for f(x) = x3 on [0, 1] using an equispaced partition. Show that AUL - 1/4 as n - oc. What is the geometrical signifi- cance of this limit? Hint: Z k 3 n2(n-)2. k=1 (3) Find the area under the graph of f(x) = e-X2 on [-1, 1] correct up to five decimal places. (4) Find the area under the graph f(x) = 1 - x2, where -1 < X < 1, correct up to three decimal places. Use the geometrical interpretation of this area to find its exact value. (5) Find the area under the graph of each of the following functions on the given interval using the relation (5.2): (i) f(xc) =3 - 3cc 0 < cc < 1 (ii) f(cc) 1+ cc + cc2, 0 < cc < 2 (iii) f(cc) =eso, -1 < cc < 1  31. AREAS AND DISTANCES 175 (6) Use the relation sin-isin(nx) sin[(n - 1)x] Zsin(2kx) to find the following: (i) The upper and lower sums for f(x) = sin x on the interval [0, 7]. Calculate An - An and investigate its behavior as n increases. What is the significance of this number for a fixed n? (ii) The area under the graph of f(x) = sin x, 0 < x < , using (5.2). (iii) The area under the graph of f(x) = cos x, 0 < x < 7/2, using (5.2). (7) An object travels with velocity v(t) =cos2 t. Find the distance passed by the object over the time interval 0 < t < 27. (8) Use the table of antiderivatives to find the area under the graph of each of the following functions. Sketch the graph of the function on a given interval and explain why this method can be used to find the area. (i) f (X) 1(X2 + 1), -1 < x < 1 (ii) f(x) = sin(ax), 0 < x < w/a (iii) f(x)= /, 0<5x 54 (iv) f (x) =1/z, 1 X < 2 (v) f(x) = tancx, 7/4 X 0. At a time t = b, it begins to slow down so that its speed becomes v(t) = ab + c(b - t), c > 0, until its speed vanishes and the car stops. Find the distance traveled by the car. Hint: Sketch the graph of v(t) in the time interval in which the car was moving. (10) An object travels with speed v(t) =v/a2 - t2, starting at t -a and stopping at t =a. Sketch the graph of v(t) and find the distance traveled by the object. (11) Let f(cc) =(cc5 - 1)/(xc - 1) if 0 m may or may not be a lower bound. So one can find the greatest lower bound that is unique for f on [a, b]. Similarly, one can find the least upper bound of f on [a, b]. These bounds have special names. DEFINITION 5.3. (Infimum and Supremum). The number m is called the infimum of a bounded function f on an interval I = [a, b] if m is a lower bound of f but m + e is not a lower bound for any e > 0. This number is denoted as m = inf1 f. The number M is called the supremum of f on [a, b] if M is an upper bound of f but M-e is not an upper bound for any E > 0. This number is denoted as M = sup1 f . Remark. The completeness axiom for a set of real numbers says that if S is a nonempty set of real numbers that has un upper bound M, then S has the least upper bound inf S. If S has a lower bound m, then it also has the greatest lower bound inf S. The completeness axiom is an expression of the fact that there is no gap or hole in the real number line. The numbers sup S and inf S are unique. Indeed, assume that M M2 are the least upper bounds of S. Since they are not equal one of them should be less than the other, e.g., A1 < A2. Since A2 is the least upper bound, the number Al2 - c is not an upper bound for anyce. Takece =l2 -AMi > 0. Then M2 -e =l1 but M1 is also an upper bound, hence, a contradiction. The uniqueness of inf S is established by similar lines of reasoning.  32. THE DEFINITE INTEGRAL 177 Naturally, if the function is continuous, then sup f is nothing but the maximum value of f and inf f is its minimum value. However, if a function has jump discontinuities, then sup f and inf f always ex- ist, while the maximum and minimum values may not exist. This is illustrated in Figure 5.2. 32.2. Definition of the Definite Integral. Let f be a bounded function on an interval [a, b]. Consider a partition of [a, b] by n intervals Ik [zk_1,§zk], k = 1, 2,... ,n, where a =0zo < x1 < --. < xn = b. Let Mk = supra f(x) (the supremum of f(x) on the interval Ik) and mk inflk f(x) (the infimum of f(x) in the interval Ik). The length of Ik is Axk = Xk-Xk_1. The lower An and upper An sums for f are defined by n n A>ZmkAxk, AU =Z M Axx k=1 k=1 for every partition of [a, b]. Put An =maxk Axk which is the length of the largest partition interval for a fixed n. The sequences of lower and upper sum are defined so that An > An+1; that is, the length of the largest partition interval decreases with increasing the number of partition intervals. DEFINITION 5.4 (The Definite Integral). A bounded function f is said to be integrable on an interval [a, b] if the sequences of its lower and upper sums converge to the same number. This number is called the definite integral of f from a to b and is denoted by f (x) dx = lim An = lim An ; the numbers a and b are called the lower and upper integration limits, respectively, and the function f is called the integrand. Apparently, for a continuous and nonnegative f on [a, b], the definite integral coincides with the area under the graph of f. Similarly to the area under the graph of a continuous nonnegative function, an integrable function has the property Ib ALnf(x) f(c)=1/2 b a c sup f= maxf b 1 1 1/2 1 1/21 I i i m I 11 i m a c inff=1/2 0, there exists an integer N such that Ib for every integer n> N and for every choice of x* in Ik. A proof of this theorem is given as an exercise (see Exercise 32.9.5; see also Exercises 32.9.12 and 32.9.13). The theorem asserts that a Riemann sum for a sufficiently large n can approximate the definite integral with any desired accuracy; that is, for any (small) designated absolute error E, R1(f) differs from fa f dc no more than e for a suffi- ciently large n. In other words, (5.6) lim R(f) = f (x) dc, for any choice of sample points x. Equation (5.6) is the analog of Equa- tion (5.2). It can be understood from the inequality AL < R_(f) < A$, which follows from mk f(zce) Mk for any x* (see Figure 5.3). For an integrable function, An and An converge to the same num- ber, which is the value of the definite integral, and, by the squeeze principle, so should R1(f) independently of the choice of sample points. 32.4. Continuity and Integrability. The relation (5.6) can be used to calculate the definite integral, provided the function f is integrable. The question of integrability requires investigating the convergence of the sequences of the upper and lower sums, which might be a tedious task even for such simple functions as, for example, f(xc) =cc2, as  180 5. INTEGRATION 180 5. INTEGRATION a xi xz X3 x4 x5*b FIGURE 5.3. Riemann sum for n = 5 partition inter- vals. Its value always lies between the lower and upper sums, A5 < R5 A , for any choice of sample points x* because mk < f(* ) < Mk. discussed in the previous section. The following theorem is helpful when studying the question of integrability. THEOREM 5.3. If f is continuous on [a, b], or if f has only a finite number of bounded jump discontinuities, then f is integrable on [a, b]; that is, the definite integral fa f(x) dx exists. An Example of a Nonintegrable Function. A bounded function f with infinitely many jump discontinuities may or may not be integrable. So, in general, the area under the graph of such a function cannot be un- ambiguously defined. As an example, consider a bounded nonnegative function f on [0, 1] such that S1 if x is a rational number 0 if x is a irrational number The function is not continuous anywhere in [0, 1] and has infinitely many jump discontinuities. For example, f(1/2) = 1, but when x approaches 1/2, the value f(x) keeps jumping from 0 to 1 and back, no matter how close x is to 1/2 because, for any 6 > 0, the interval  32. THE DEFINITE INTEGRAL 181 (2 - 8, 2 + 8) always contains both rational and irrational numbers. This function is not integrable. Indeed, take a partition Xk = k/n, k = 0, 1, ... , n. Any partition interval [(k - 1)/n, k/n] contains both rational and irrational numbers. Therefore, mk = 0 and Mk = 1. Hence, the lower sum vanishes for any partition, Anf= 0, whereas the upper sum is A = E"_ A x= 1, that is, limn-, Anf= 0 while limn-, An= 1. The function is not integrable. The integral does not exist. Note that the Riemann sum can still be defined, but its limit would depend on the choice of sample points (e.g., take x* to be rational numbers or take x* to be irrational numbers; both options are possible since any partition interval always contains rational and irrational numbers). In fact, with a suitable choice of sample points, the Riemann sums can converge to any value between 0 and 1 (e.g., in all partition segments to left of 0

= A = cE"_1 Ax = cn AOx= c(b - a). In other words, a constant function is integrable and its integral is c(b - a): (5.7) cdx = c(b - a). For any two integrable functions f(x) and g(x) and constants ci and c2, it follows from the convergence of the Riemann sums (5.6) for f and g that fb[cf(x) + c2g(x)] dx lim [cif(x) + c2g(x)] Axk k=1 n n = cl lim 3 f(x) Axk + c2 lim g(xz) Axk k=1 k=1 (5.8) = ci f(x) dx + 2fbg(x) dx. So the integration is a linear operation. In particular, the integral of the sum of two functions is the sum of their integrals. The integral of a function multiplied by a constant is the product of the constant and the integral of the function. If the integration limits are reversed, then all Axk change their signs as xk becomes less than zx_1. Therefore, (5.9) jf (x) dx ()d  182 5. INTEGRATION and, in particular, (5.10) f(x) dx = 0. It can be proved that (5.11) f(x)dz = cf (x)dz+ f (x) d for f integrable on [a, b] and any a < c < b. The proof is rather technical and is omitted. If f is continuous and positive on [a, b], then the property (5.11) is trivial: The area under the graph of f on [a, b] is the sum of the areas under the graph of f on [a, c] and [c, b]. 32.6. Geometrical Significance of the Definite Integral. As already noted, the definite integral of f from a to b coincides with the area under the graph of f for a continuous and positive f. Suppose f is continuous and negative on [a, b]. Consider the function g(x) = -f(x). The integral of g is the area A under the graph of g and, hence, A also coincides with the area above the graph of f and below the c axis. By the linearity of the integral, fa f(x) dcc - f/a g(x) dc = -A. So, for a negative f, the integral of f coincides with the negative area of the region bounded below by the graph of f and above by the c axis. Now let f be con- tinuous on [a, b]. Let it be positive on [a, c] and negative on [c, b], that is, f(c) = 0. Then it follows from the property (5.11) that f(x)d =f(cx)dz+f f(x)d= A1 -A2, a c where A1 is the area under the graph of f on [a, c] and A2 is the area above the graph of f on [c, b]. This property is illustrated in Figure 5.4. 32.7. Comparison Properties of the Integral. The following additional properties of the definite integral can be established: (5.12) f (x) dc > 0, if f(x) > 0 in [a, b] , fbb (5.13) f(x) dcc >I g(x) dx, if f (x) g(x) in [a,b], a a (5.14) m(b -a) ff(c) dc M(b -a), if m5f(c) Min [a, b]. The property (5.12) follows directly from the definition. Since 0 < mft Mk for any partition if f(cc) 0, the upper and lower sums  32. THE DEFINITE INTEGRAL 183 I= A2-A1 b FIGURE 5.4. Geometrical interpretation of the definite integral. If an integrable function f is nonnegative on an interval [c, b], then its integral over [c, b] is the area A2 under the graph of f above the interval [c, b]. If f is non- positive on an interval [a, c], then its integral over [a, c] is -A1, where A1 is the area above the graph of f and below the interval [a, c] on the x axis. By the additivity of the integral, the integral of f over the interval [a, b] being the union of intervals [a, c] and [c, b] is the difference A2 - A1. This shows that the value on the integral can be any real number. are nonnegative and so must be the integral. If f is continuous, the property (5.12) states the obvious that the area under the graph of f is nonnegative. The property (5.13) follows from (5.12) for the function f(x) - g(x) > 0 and the linearity of the integral (5.8). The property (5.14) is also a consequence of the definition. Indeed, for any partition, m < mk < Mk < M. Hence, m(b - a) < An < An < M(b - a) for any n. In the limit n -- o0, this inequality turns into (5.14). THEOREM 5.4 (Integrability of the absolute value). If a bounded function f(x) is integrable over an interval [a, b], then its absolute value |f(x) is also integrable over [a, b] and Ib b f ()d < f (x) dx a a A proof of this theorem under a simplifying hypothesis that f is continuous is given as an exercise (Exercise 32.9.11). The converse of Theorem 5.4 is not true. The integrability of the absolute value |f(x)  184 5. INTEGRATION M ----- a b FIGURE 5.5. Geometrical interpretation of the property (5.14). The graph of a function f lies between two hor- izontal lines y = m and y = M because m < f(x) < M for all x E [a, b]. So the area A under the graph of f lies between the areas of rectangles with the base b - a and heights m and M, i.e., m(b - a) < A < M(b - a). does not imply the integrability of f(x). Consider the function 1 if x is a rational number -1 if x is a irrational number The absolute value |f(x) = 1 is continuous and, hence, is integrable on any bounded interval. However, f(x) is not integrable. The proof of this assertion is left to the reader as an exercise. 32.8. Evaluation of the Integral by the Riemann Sum. If the integral ex- ists (f is integrable), then it can be evaluated as the limit of the Rie- mann sum (5.6). The limit is independent of the choice of sample points. The following choices are often used in practice: x* = xk_1 (the left-point rule), x* = 4 k (the right-point rule), x* = (zk_1 + xk)/2 (the midpoint rule), in combination with the basic properties of the integral. The evaluation of the Riemann sum is rather technical. Formulas like (5.1), (5.4), and n(n - 1) n n(n - 1) 2 2 , k k=1 k=1  32. THE DEFINITE INTEGRAL 185 can be helpful. However, the Riemann sum is mostly used to calculate the integral approximately with some designated accuracy by means of computer simulations, similarly to approximate calculations of the area discussed in the previous section. EXAMPLE 5.2. Find the definite integral of f (x) = e-2x - 2x2+ 4X3 from 0 to 1. Solution: 1. The function is continuous on [0, 1] and hence integrable; that is, Equation (5.6) applies for any choice of x. The left-point rule will be used. 2. By the linearity of the integral, ff(x)dx f e-2x - f2 x2 d f+4 x The first integral is (1 - e-2)/2 by Example 5.1 (where the area under the graph of e-2x in [0, 1] was calculated). The area under the graph x2 in [0, 1] can be found at the beginning of the previous section and is equal to 1/3. The area under the graph of x3 can be found with the help of the second relation in (5.15). Let AOx= 1/n and Xk (k - 1)/n (the left-point rule), then the Riemann sum (5.6) becomes f1x3 dx=lim 1 k3 _lim 1 n2(-n-1)2 - J0 h-~cD124- k n14 4 4 k=1 3. Thus, / 1- e-2 2 1 1-6e-2 f (x) dz 0n )2 3 4 12 ~E 32.9. Exercises. (1) Let f (x) = sin(1/x) if x / 0 and f (0) = fo. Given any number S > 0, find the supremum and infimum of f on [-5, o]. (2) Find the upper and lower sums for the function f(x) = 1 if x > 0 and f (x) = -2 if x < 0 on the interval [-1, 1]. Use them to show that f is integrable. Find the value of the integral as the limit of the lower and upper sums. (3) Find the upper and lower sums for the function f(x) = 1 if x 0 and f(x) =fo 1 if x =0 on an interval [a,b]. Use them to show that f is integrable on any [a, b]. Find the value of the integral as the limit of the lower and upper sums. (4) Suppose that the length of the largest partition interval, As maxk Azk, decreases as the number in of partition intervals  186 5. INTEGRATION increases (i.e., An+1 < An). Show that An An+landAn> An+1. Deduce from this property that Ib for any n and for any integrable f on [a, b]. (5) Use the inequality from the previous exercise to prove Theorem 5.2. (6) Let a < c < b. Put I1 = [a, c] and I2 = [c, b]. Let f be integrable on [a, b]. If M1 = supri f, M2 = sup12 f, mi= infi f, and m2 = infr2 f, prove that Ib mrt(c -a) +m2(b -c) GfdxJV M(c -a) +V 2(b -c). (7) Use Equation (5.14) to estimate the definite integral of each of the following functions from above and below: (i) f (x) =i, -1 < x < 2 (ii) f (x) sinx| - sin x, 0 < x <27r (iii) f (x) = | cos x - 1/2| - cos x, -7r < x < 7r (8) Use the result of exercise 6 to improve the upper and lower estimates in exercise 7 by a suitable choice of c. (9) Use the geometrical properties of the definite integral to find the exact values of (i f (x) dz , f (x) = 0 / 31,0 0 , fx(x))1< = 2aa/2 (iv) a2 - x2dx (v) 1-x2- Ixdx /1/f (vi) a2 -x2 _ C /a2 - (2xCc-a)2]dz (10) Use Riemann sums for equispaced partitions to evaluate each of the following definite integrals correct up to three decimal places:  32. THE DEFINITE INTEGRAL 187 (i) f2 1/3 dc (ii) f2 sin(x2) dz (iii) f1 exp(sin c) dc (iv) fcxncd (11) Let f be continuous on [a, b]. Then g(x) If(x)| is integrable on [a, b]. Why? Show that fbdz < |f(x)|dc. aab Hint: Compare Riemann sums for f(x) and |f(x)|. (12) Let f have a bounded derivative on [a, b], that is, If'(x)| 0. This integral represents the area under the graph of f(t) =tin the interval [0, c], which is the area of a right triangle: A(x)c= tdt cc. /o 2 The area A(x) can be viewed as a function of the variable x, which is the length of the triangle catheti. This function has an interesting property: A'(cc) =cc f (cc). In other words, the derivative of the definite integral with respect to its upper limit equals the value of the integrand at the upper limit. Recall that if v(t) ;> 0 is the speed of a moving object, then the distance  33. THE FUNDAMENTAL THEOREM OF CALCULUS 189 traveled by the object in time T is given by the area under the graph of v(t): D(T) v(t) dt. 0T On the other hand, the speed is the rate of change of D(T), and there- fore there should be D'(T) = v(T); that is, the derivative of the integral with respect to its upper limit is again the value of the integrand at the upper limit. How general is this property? Does it hold for all integrable functions? The following theorem answers these questions. THEOREM 5.5. If f is continuous on [a, b], then the function defined by g (x) = f (t) dt , a < x < b, is continuous on [a, b] and differentiable on (a, b), and g'(x) = f(x). PROOF. By the definition of the derivative, one has to prove that (5.16) lim g(x + h) - g(x)f hO h for a < x < b. The ratio in the limit can be transformed as follows: g(x +h) - g(x) _ fx+h cit -( 1 --i f (t) dt- f(t) dtj h h aa _ = h (Xf(t) dt + f nf(t) dt - If(t) dtl 1x+h =- f f(t)dt, where the property (5.11) has been used. Note that since a < x < b (i.e., x / a and x / b), for a sufficiently small h / 0, both x and x+ h (h can be positive or negative) always lie in the interval (a, b) so that the interval [x, x + h] is contained in (a, b). By the continuity of f(t) on the interval [z, x + h], the function f attains its absolute maximum and minimum values in [x, x + h]. Let M = f(v) and m = f(u) be the absolute maximum and minimum values, respectively, where v and u are in [x,xz + h]. Suppose that h > 0. Then mn f(t) < M for cc 0, by dividing this inequality by h, one can infer that 1 h (5.18) f (u) <;x-h f (t) dt < f (v) hx for some u and v in [x, x + h]. Inequality (5.18) can be established for h < 0 in a similar manner. Indeed, inequality (5.17) holds for the integral fj+h f(t) dt. After dividing it by -h > 0, inequality (5.18) is obtained but with the minus sign at the integral. By the property (5.9), the sign is reversed, yielding (5.18). Thus, g(x + h) - g(x) f (U) < < f (V). h Since u and v lie in the interval [z, x + h], lim f (u) = f(x), lim f (v) = f(x). h->0 h-0 Then the relation (5.16) follows from the squeeze principle: f (x) = lim f (u) < lim g(<+h) -g() 0 h->0 h h- 0 This theorem basically states that if a continuous function is first integrated and then differentiated, then it remains unchanged: (5.19) xff(t) dt= f (x) , a < z < b . In other words, F(x) = fff(t) dt is an antiderivative of f(x) in an open interval (a, b). The continuity of f on [a, b] is essential for this relation to hold. Take, for example, f(t) = 0 if t < 1 and f(t) = 1 if t > 1. Let a = 0. Then f has a jump discontinuity at t = 1; it is integrable on any interval, but not continuous at t = 1. By the property (5.7), g(x) = fe f(t)dt = 0 if xc < 1. For x> 1, one has g(x) Jfl(t)dt = Jf(t)dt +X f(t)dt 0 0 1 =0 + (X -1)=Xz-1I,X ;> 1. Therefore, g'(x) = 0 if x < 1 and g'(x) = 1 if x > 1. But g'(1) does not exist. EXAMPLE 5.3. Let g(xc) =f~ e_2dt. Find g'(xc). Solution: The function e-t2 is a continuous function everywhere as a composition of two continuous functions, the exponential and power  33. THE FUNDAMENTAL THEOREM OF CALCULUS 191 functions. By the property (5.9), g(x) - fj e-2dt. Therefore, g'(x) = -ex2 by (5.19). This example illustrates the general property: f(t)dt = xf(t)dt = -f(x) for a continuous f. 33.2. The Definite Integral and Antiderivative. The following theorem establishes the relation between the definite integral of a function and its antiderivative. THEOREM 5.6 (The Fundamental Theorem of Calculus). If f is continuous on [a,b], then fbfv(x) dzc= F(b) - F(a), where F is any antiderivative of f, that is, a function such that F' = f . PROOF. Let g(x) = fj f(t) dt. By (5.19), the function g(x) is an antiderivative of f(x) in an open interval (a, b). If F is any other antiderivative of f, then F and g may differ only by a constant, F(x) = g(x)+ C, a < x< b. Also, by the definition of g(x), g(a) = 0 and g(b) = fb f(t) dt. The function g(x) is continuous on [a, b] because limx-a+ g(x) = g(a) = 0 and limx-b- g(x) = g(b). Therefore, F(x) is also continuous on [a, b] (as the sum of two continuous functions). Hence, F(b) - F(a) = (g(b) + C) - (g(a) + C)= g(b) f(t) dt. The proof is complete. D The following notations are adopted in the fundamental theorem of calculus: (5.20) f(x)dz = F(x)b= F(b) - F(a) The vertical bar at an antiderivative F~x) means that one has to cal- culate the difference of the values of F at the indicated points. The fundamental theorem of calculus provides a powerful analytic tool to evaluate definite integrals.  192 5. INTEGRATION EXAMPLE 5.4. Evaluate foJ(1 + x2)-1 dx. Solution: An antiderivative of (1+x2)-1 is F(x) =tan-1 x. Therefore, 1 1 dx=tan-1 x = tan-1(1) -tan-1(0) i4o 0 -. o1+x2 o 4 4 0D EXAMPLE 5.5. Evaluate fj(1+ x)/ /x dx. Solution: By the linearity of the integral, 4 1 + dx1/2 + X1/2) dx -fX-1/2 dx + f X 72 dx. An antiderivative of x" is xn+1/(n + 1) for any real n f 1. By taking n = -1/2 and n = 1/2, an antiderivative is obtained: F(x) = 2x1/2 + 2x3/2/3. Hence, 2x3 16 2 20 j 1 i d x =+2x 1/ + 2J 34 332+_ 3 If an object moves along a straight line, its position relative to a fixed point on the line (the origin) may be defined by a single coordinate x which is a function of time. The velocity v(t) x=z'(t) is positive if the particle moves in the direction in which x increases and is negative if it moves in the opposite direction. The acceleration is a(t) = v'(t) - x"(t). A law according to which the acceleration changes with time is usually established by the laws of physics. Then a practical question is to find the position x(t). Since x(t) is a second antiderivative of the acceleration, it is not unique and two (initial) conditions must be imposed to get a unique solution. EXAMPLE 5.6. A particle moves along the x axis with the acceler- ation a(t) = 2 - 6t. Find the position of the particle at the time t = 3 if its position and velocity at t = 1 were x(1) = 1 and v(1) = 2. Solution: Since v'(t) = a(t), the velocity is the antiderivative of the acceleration subject to the condition v(1) =2. Hence, by the property (5.10), v(t) = v(1)I a(s) ds =2±+(2s -3s2) = 2t -3t2+ 3  33. THE FUNDAMENTAL THEOREM OF CALCULUS 193 Since x'(t) = v(t), the position x(t) is the antiderivative of the ve- locity subject to the condition x(1) = 1. By property (5.10) such an antiderivative reads x(t) = x(1) + v(s) ds = 1 + (s2 - S3 + 3s) = t2 - t3 + 3t - 2 1 1 Therefore x(3) =--11. D 33.3. Exercises. (1) Find the derivative of each of the following functions: (i) f (x) = fj(1 + t6)-1 dt (ii) f (X) = fo sin(t2) dt (iii) f (x) = fx cos (et) dt (iv) f (x) = fX e2dt (2) Let f(x) be a piecewise constant function: 0, < 0 and z > 6 1, 00, a 1 sin-1x+C -cos-1 x+ C cot(a) - +C a - cscxc+ C J sec2(ax) dx I sec xtan xdx=sec x+ C I csc x cot xdx zdc I1-X2 1 1+ C -ln +C 2 1-cc dc =ln x+ c2+1 +C X2+1 Recall that the general antiderivative on a given interval is obtained from a particular antiderivative by adding an arbitrary constant. This does not hold for a domain being a union of two or more disjoint intervals (review the properties of antiderivatives). So, in the preceding table, the convention is used that the given expressions for indefinite integrals are valid only in a single interval of continuity of the integrand. EXAMPLE 5.7. Find a general indefinite integral for x-3. Solution: The function c-3 is not defined at x = 0. So its domain is the union of two disjoint intervals (-oc, 0) and (0, oc). By the first equality in the preceding table (n = -3), Ic-3dc -2 +C1, cc>0; Ic3dc 2 x-2 +C2, c<0, 2  196 5. INTEGRATION where C1 and C2 are arbitrary constants. D EXAMPLE 5.8. Evaluate f0 [3x2 - x + 4(1 + x2)-1] dx. Solution: By the linearity of the indefinite integral (5.21), an indefinite integral of the integrand is x3-x2/2 + 4tan-1 x. An arbitrary constant in the indefinite integral may be omitted here because, as already noted, it is always cancelled out in the definite integral. Therefore, 3X2-_X+ + XJdz = (x3 - 2-+4 ta-1 x) 2- -7 where tan-1(1) =w7/4 has been used. D 34.1. The Net Change Theorem. Put f(x) = F'(x) in the fundamental theorem of calculus (5.20). The result obtained is known as the net change theorem. THEOREM 5.7. The integral of a continuous rate of change is the net change: JbF'(x) dz = F (b) - F (a). /b The continuity of F'(x) is crucial for the above relation to hold. A formal application of the theorem without checking the continuity may lead to incorrect results. For example, the following line of calculations is false: dx( )dz -x-1 -1-1= -2 Indeed, the derivative F'(x) = (-1/x)' = 1/z2 is not defined at x = 0. It is not possible to assign any numerical value to F'(0) to make F'(x) continuous at x = 0 because limx o F'(x) = oc which is not a number. Furthermore, F'(x) > 0 is strictly positive in any interval and, by the property (5.12), so should be the definite integral if it exists. Hence, the above result cannot possibly be true. The rate F'(x) may be positive and negative in the interval [a, b] so that the quantity y = F(x) may increase and decrease. The difference F(b) - F(a) represents the net change of y when x changes from a to b. The net change vanishes if F(b) - F(a) 0. This does not mean that the quantity y does not change at all, but rather this might mean, for example, that the quantity y increases from the value F(a), then, at some c in [a, b], it begins to decrease, returning to its initial value when cc= b so that its net change vanishes.  34. INDEFINITE INTEGRALS AND THE NET CHANGE 197 An analogy with an object moving along a straight line can be made to illustrate the net change. Let x(t) be a position function of the object relative to some point on the line. Then x'(t) = v(t) is its velocity (note that the velocity can be negative so that the object can move back and forth). The net change of the position over the time interval [t1,t2] is v(t) dt =x(t2) - x(t1). t2 EXAMPLE 5.9. Suppose an object travels along a straight line with a velocity of v(t) = 1 - 2t. Find the net change of its position over the time interval [0, 1] and the total distance traveled by the object over the same time interval. Solution: 1. The indefinite integral of v(t) is x(t) the net change of the object position is = t-t2+C. So fv(t)dt 1 I z'(t) dt = x(1) - x(0) = 0. 0 2. Note that the velocity changes its sign at t = 1/2. So, in the interval [0, 1/2], it is positive (i.e., the object moves to the right from its initial position), then the velocity becomes negative in [1/2, 1] (i.e., the object goes back to the initial point). To find the distance traveled by the object, the absolute value |v(t)| must be integrated over the interval [0, 1]. Think of |v(t)| as the speed shown on the speedometer of your car; it is always non negative regardless of the direction in which the car is moving. 1l - 2t| dt 0 r1/2 1 (1 -2t )- (1- 2t ) dt 0 (1/2 ( [x(1/2) - x(0)] - [z(1) - x(1/2)] = 1/2, where the definition |vl used. vif v > 0 and |vl -v if v < 0 has been Other examples of the net change includes the volume V(t) of water in a reservoir between two moments of time t2 V'(t)ct V(t2) -  198 5. INTEGRATION where V'(t) is the rate of change of the volume; the net change of the population growth 1t2mn'(t) dt = n(t2)- n(ti), t2 where n'(t) is the growth rate; the relation between the cost and mar- ginal cost functions: J C'(t) dt =C(t2) - C(t1); ti and similarly for many other quantities. 34.2. Exercises. (1) Find the indefinite integrals. Assume that x lies in a single interval of continuity of the integrand. (i) f6x5 dz (ii) fcc3 dz (iii) f(x2/3 - X-2/3) dz (iv) f(x2 - 4)2 dc (v) f(1+ c)3 dz (vi) f ( + c a)2/ ccdc (vii) fcc/ c2 + 1 d (viii) f (1 +cc-1) cccdcc (ix) f (1 + coscx + cos(2x)) dc (x) f(x + sin(4x)) dc (xi) f (1 + 2x)/ 1 - x2 dc (xii) f (1 - x + c3)/x2 d (xiii) fcx2/(a2+cx2) dc (xiv) f(x2 + 3)/(x2 - 1) dc (xv)fcc2c4 + c-4 +2 d (xvi) f [ 1 -+x2 + l -x2]/ 1 -x4 d (xvii) f[ c2+1- c2-1]/ c4-1d (xviii) f (2x + 3x)2d (xix) f(2x+l - 5x-1)/1Ox dc (xx) f (esx + -)/e +1 dcc (xxi) f 21- sin(2c)dc O 5cc5 w Hint: Use the fundamental trigonometric identity. (xxii) f cot2cdc (xxiii) f t an2 ccdcc  34. INDEFINITE INTEGRALS AND THE NET CHANGE 199 (2) Prove that the existence of the indefinite integral of f(x) im- plies the existence of the indefinite integral of f(ac+ b), a f 0, and ff(x)dzc=F(x) + C - f(azc+ b)dz= F(azc+ b) + C a (3) Use the result of the previous exercise to find (i) f (x +a)-idz (ii) f(2xc-3)10 dz (iii) f 3/1 - 3x dz (iv) f(5xc-2)5/2 dz (v) f(2+3x2)-1dz (vi) f (3 - 2x2)-1 dz (vii) f (4 - 3x2)-1/2 dz (viii) f (2X2 - 5)-1/2 dc (ix) f csc2(2x + 7/4) dz (x) f (1 + coscx)-1 dc (xi) fJ(1 - sin c)-1 dc (xii) f (1 + sin c)-1 dc Hint: Use sin x = 2 sin(c/2) cos(c/2) and the fundamental trigonometric identity. (4) Find a general indefinite integral: (i) fcx-1 dz (ii) fcc1/P d, p > 1 (iii) f sec2 x d (iv) f (1 - X2)-1 dc (v) f(x2 - 1)-id (5) Explain why a formal application of the fundamental theorem of calculus with a given antiderivative F(x) leads to incorrect results if (i) J1d F(x) = ln jx| (ii) xt an- (xdz , F(x) = tan-1x (6) Evaluate ic; c1+2 dc and _ 1+__2 dc/  200 5. INTEGRATION (7) A particle travels with velocity v(t) = sin(t/2). Find the net displacement of the particle over the time interval [0, 27] and the distance traveled by the particle. (8) A bacteria population grows at an exponential rate n'(t) noyery, where no is the initial population and y is a constant. If in the time T the population has doubled, find the constant y. What is the population at t =lOT as compared to the initial population? (9) The decay rate of a radioactive element is proportional to the total amount of the element at each moment of time. Find the law of decay of the radium isotope Ra-226 if at an initial moment of time there were no grams of Ra-226 and in 1600 years its quantity had decreased by two times. If 1 gram of Ra-226 is deposited into a radioactive storage, how much of it will remain in 800 years? 35. The Substitution Rule 35.1. Indefinite Integrals. An indefinite integral of the derivative F'(x) is the function F(x) itself, provided F'(x) is continuous. Let u = F(x), where u is a new variable defined as a differentiable function of x. Consider the differential du = F'(x) dx. Then the following equalities hold: f F'(x)dx =F(x) +C =u+C =fdu, where C is an arbitrary constant and the last equality follows from the fact that an indefinite integral of f(u) = 1 is u. So we can conclude that f F'(x) dxz= f du, provided the variables u and x are related as u = F(x). This also shows that it is permissible to operate with dx and du after the integral sign as if they were differentials. This obser- vation leads to a neat technical trick to calculate indefinite integrals. For example, +Id=d (2/x + 1) = 2/ x+1+ C, where the substitution u = 2/x + 1 has been used. This trick can be generalized. Let F(Li) be an indefinite integral of a continuous function f(Li) on an interval I. Let Li= g(x), where g is differentiable and its range is the interval I. By the chain rule, (F(g(x)))= F'(g(x))g'(x) =f (g(x))g'(x) .  35. THE SUBSTITUTION RULE 201 In other words, F(g(x))+C is an indefinite integral of f(g(x))g'(x). On an interval, the most general indefinite integral of f(u) is f f(u) du F(u) + C. Therefore, F(g(x)) and f f(u) du can differ at most by an additive constant. This proves the following theorem. THEOREM 5.8. (The Substitution Rule). If u = g(x) is a dif- ferentiable function whose range is an interval I and f is continuous on I, then (5.22) f (g(x))g'(x) dx J f(g(x)) dg(x) J f(u) du. The substitution rule is often referred to as a change of the integra- tion variable. It is a powerful method to calculate indefinite integrals. EXAMPLE 5.10. Find f xcsin(x2 + 1) dx. Solution: fx sin(x2 + 1) dz=fsin(x2 + 1)2d(x2 + 1) = sin u du 1 1 = cos L + C - cos(x2 + 1) + C, 2 2 where the substitution u = x2 + 1 has been used. D EXAMPLE 5.11. Find f tan x dz. Solution: sin x f d(cos zc) fdu f tanccdcc f Csxd= oszc I coszc Ju -lnlul +C=-ln coszl +C=ln secccl+C, where the substitution uL= cos x and the logarithm property ln(1/a) - ln a have been used. Here it should be noted that the calculations are valid only in a single interval of continuity of tan x, e.g., -7/2 < z < 7/2. D A general idea of the substitution is to transform the integral in question to one of the basic integrals given in the table. Sometimes the task can only be accomplished with several substitutions. EXAMPLE 5.12. Find f (ex - 1)-1/2dz. Solution: It is suggestive to transform the integrand to a power func- tion with the help of the substitution: u = eX -1 --> zc=ln(Li+ 1) --> dcc =~ Li+ 1  202 5. INTEGRATION Therefore dz 1 du d dv -2 -2 I -1 u+1 u+1 Jv2+1 =2tan-1v+C=2tan-1( L)+C = 2tan-1( ex - 1) + C where the second substitution v =V, 2dv= du/ /, has been made. 35.2. Definite Integrals. The substitution rule can be used to evaluate definite integrals by means of the fundamental theorem of calculus. EXAMPLE 5.13. Evaluate j zcex2 dxc. Solution: First, find an indefinite integral: f 2 1 2 1 laX12 F(x)= ze dc= ed2 I ed e+C= e +C. J 2 2 2 where L = x2. By the fundamental theorem of calculus, fe dx= F(2) - F(0) =-(e - 1). Q 02D Note that, when evaluating the integral, the original variable c has been restored in the indefinite integral in order to apply the fundamen- tal theorem of calculus. The fundamental theorem of calculus can also be applied directly in the new variable u, provided the range of u is properly changed. Indeed, in the previous example, the answer could have been recovered from the indefinite integral je" + C if u = x2 ranges from 0 = 02 to 4 = 22 as x ranges from 0 to 2. This is especially useful when a calculation of a definite integral requires several changes of the integration variable. THEOREM 5.9. (The Substitution Rule for Definite Integrals). If g' is continuous on [a, b] and f is continuous on the range of u = g(x), then pb g(b) (5.23)] f(g(x))g'(x) d = f((u) du. PROOF. Let F be an antiderivative of f. Then F(g(cc)) is an an- tiderivative of (F(g(cc)))' =F'(g(c))g'(cc) =f(g(cc))g'(cc). By the fun- damental theorem of calculus, f (g(cc))g'(cc) dcc= F(g(cc)) =F(g(b)) - F(g(a)) . a a  35. THE SUBSTITUTION RULE 203 On the other hand, since F(u) is an antiderivative of f(u), the funda- mental theorem of calculus yields g(b) g(b) I (u) du = F(u) = F(g(b)) - F(g(a)). g(a) 9(a) Since the right-hand sides of these equalities coincide, so must their left-hand sides, which implies (5.23). D EXAMPLE 5.14. Evaluate fe ln x/x dx. Solution: The integrand can be transformed as lnx dx= ln x dln x. x So the substitution u = ln x can be made. The range of the new integration variable u is determined by the range of the old one: u = 0 when x = 1 and u = 1 when x = e. Thus, felnx dx~ f id.20 2 35.3. Symmetry. The calculation of a definite integral over a symmet- ric interval can be simplified if the integrand possesses symmetry prop- erties. THEOREM 5.10. Suppose f is continuous on a symmetric interval [-a,a]. Then (5.24) f (x) dx = 2 f (x) dx if f (-x) = f (x) (f is even), (5.25) f (x) dxz= 0 if f (-x) = -f (x) (f is odd). PROOF. The integral can be split into two integrals: /a d f0 + aa f (x) dz + f (x) dx f (x) dx + f (x) dx. -a - a 00 0 In the first integral on the very right-hand side, the substitution u = -x is made so that u = 0 when x = 0 and u = a when x = -a and dx -dii. Hence, - f x) =f (-u) du 0 0  204 5. INTEGRATION I=0 A -a a -A FIGURE 5.6. Illustration of the property (5.25). A func- tion is odd if f(-x) = -f(x). Its integral over a sym- metric interval [-a, a] vanishes. The area A under the graph of f and above the interval [0, a] is the same as the area above the graph of f and below the interval [-a, 0] because of the skew symmetry of the function and the symmetry of the interval [-a, a] relative to the reflection x -- -x. By the property depicted in Figure 5.4, the integral of f over [-a, a] is A + (-A) = 0. and /aa a f (x) dx f (-u) du + f (x) dx. a 0 0 Now, if f is even, then f(-u) = f(u) and (5.24) follows. If f is odd, then f(-u) = -f(u) and (5.25) follows. Q The geometrical interpretation of this theorem is transparent (see Fig 5.6). Suppose f(x) > 0 for 0 < x < a. The integral f" f(x) dx A is the area under the graph of f on [0, a]. If f is even, then, by symmetry, the graph of f on [-a, 0] is obtained from that on [0, a] by a reflection about the y axis. Therefore, the area fa f(x) dx must coincide with A. If f is odd, then its graph on [-a, 0] is obtained by the mirror reflection about the origin so that the area A appears beneath the x axis. Hence, f_" f(x) d= -A. EXAMPLE 5.15. Evaluate f_, sin(x3) dx.  35. THE SUBSTITUTION RULE 205 Solution: Unfortunately, an antiderivative of sin(x3) cannot be ex- pressed in elementary functions, and the fundamental theorem of cal- culus cannot be used. One can always evaluate the integral by taking the limit of the sequence of Riemann sums. An alternative solution is due to a simple symmetry argument. Note that sin(x3) is an odd function, sin((-x)3) =sin(-x3) = - sin(x3). The integration interval is also symmetric, [-7, w]. Thus, by property (5.25), f sin(x3)dx 0. Remark. In the previous example, take a partition of [-7, w] by points xk = k Ax, k = -n, -n + 1, ... , -1,0,1,... , n - 1, n, where Ax = 7/n. Consider the Riemann sum with sample points being the midpoints. They have the property that x*_k = -4. It is then straight- forward to show that the Riemann sum vanishes because sin(x*_3) sin((-z )3) - sin(z43) for k = 1, 2,... , n (the terms corresponding to negative x cancel out the terms corresponding to positive x in the Riemann sum). 35.4. Exercises. (1) Use the suggested substitution to find the indefinite integrals: (i) f x3(x4 + 1)1/3 dxa,= x4 + 1 (ii)fsin( x)/ xdx, u = (iii) f sin xecosx dx, u = cos x (iv) fJx2 1 - x2 dx, u =sin x (v) f(lnx)3/xdx u= ln x (vi) f(tanx)"sec2xdx, n -1, u=tanx (vii) f(cot x)" csc2 x dx, n -1, u = cot x (viii) f(sin-1 x)2(1 - x2)-1/2 dx, u= sin-1 x (ix) f ex (e2x + 1)-1 dx, u = e (x) f e2x(1 + em)-1 dx, u = 1 + ex (xi) f sin(2x)(1 + cos2 x)P dx, u = cos2 x (xii)f ex+1dx, u = ex + 1 Hint: See Example 5.12 to proceed. (xiii) f 1l - ex dz, u = 1 - ex Hint: See Example 5.12 to proceed. (2) Use a substitution to find the indefinite integrals: (i)fx z1 +2x dx (ii) fx2/ 2 -3x dx  206 5. INTEGRATION (iv) f e 7- /c/c dc (v) f z/(z4 + 2x2 + 2) d Hint: Complete the squares in the denominator. (vi) fcx tan-1 x(1 + x2)-2 dc (vii) f cos-1cc/v1 - x2 d (viii) f 1 - x2 sin-1 x dc (ix) fJ(1 + cos2 x + cos4 c) sincx dc (x) f sin3 c/(1 + cos2 cc) d (xi) f[/1+ ex+ /1 - ex]-ld Hint: Transform the integrand to make the difference of per- fect squares in the denominator: a2 - b2 = (a + b)(a - b). (3) Use a change of variables and/or symmetry to evaluate the definite integrals: (i)f xv/2+x2d (ii) f1 tan-1 c/(1 + x2) dc (iii) f zxcos(x2) dc (iv) f cc3e X4 dc (v) f22 x(ex2 - e2) dc (vi) f22(ex3 - e-x3) dc (vii) faa g(x) dx, g(x) =-ff cos(t2) dt (viii) f(2x + 1)v/2 + x + 3 d (ix) f/2 sin(2x)31 + cos2 x d (x) f/4(tanc)Psec2 x d, p > 0 (xi) f_/> tan3(3x) sin5(2x) dc (xii) fgv1 - e-xdc, a > 0 (xiii) f33 c3/(1+ + c6) dc