Calc 1

Concepts in Calculus I
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
Beta Version
Miklos Bona 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 2011 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 978-1-61610-155-8
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
Gainesville, FL 32611-2079
http://www.upf.com
Contents
Chapter 1. Functions 1
1. Functions 1
2. Classes of Functions 2
3. Operations on Functions 4
4. Viewing the Graphs of Functions 6
5. Inverse Functions 8
6. The Velocity Problem and the Tangent Problem 11
Chapter 2. Limits and Derivatives 15
7. The Limit of a Function 15
8. Limit Laws 21
9. Continuous Functions 25
10. Limits at Infinity 29
11. Derivatives 34
12. The Derivative as a Function 36
Chapter 3. Rules of Differentiation 41
13. Derivatives of Polynomial and Exponential Functions 41
14. The Product and Quotient Rules 44
15. Derivatives of Trigonometric Functions 47
16. The Chain Rule 49
17. Implicit Differentiation 53
18. Derivatives of Logarithmic Functions 55
19. Applications of Rates of Change 58
20. Related Rates 61
21. Linear Approximations and Differentials 65
Chapter 4. Applications of Differentiation 71
22. Minimum and Maximum Values 71
23. The Mean Value Theorem 76
24. The First and Second Derivative Tests 79
25. Taylor Polynomials and the Local Behavior of a Function 84
26. L’Hospital’s Rule 88
27. Analyzing the Shape of a Graph 92
28. Optimization Problems 97
vi CONTENTS
29. Newton’s Method 101

30. Antiderivatives 106
Chapter 5. Integration 111
31. Areas and Distances 111
32. The Definite Integral 115
33. The Fundamental Theorem of Calculus 120
34. Indefinite Integrals and the Net Change 124
35. The Substitution Rule 128
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.
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, 1 to 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
⎧
⎨ 0.1x if 0 ≤ x ≤ 40,
f (x) = 4 + 0.15(x − 40) if 40 < x ≤ 80,
⎩ 10 + 0.2(x − 80) if x > 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 below
$60,000 is taxed at a rate of 15%, and income above $80,000 is 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,

x if 0 ≤ x,
f (x) = |x| =
−x if x < 0.
1
2 1. FUNCTIONS
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 x to f (x) can only be given by listing
the value of f (x) for each x, 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 x, let f (x) be the
amount of precipitation that x had in 2009. Then f is a function since
it sends each x ∈ 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 coor-
dinate 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 ∈ 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.
2. Classes of Functions
2.1. Power Functions. A power function is a function f given by the
rule f (x) = xa , where a is a fixed real number. Note that x−a = 1/xa ,
so, for instance, x−3 = 1/x3 . The special case of a = −1, that is, the
function f (x) = 1/x, is called the reciprocal function. Note that the
rule g(x) = 1 for all real numbers x also defines a power function, one
2. CLASSES OF FUNCTIONS 3
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) = xa = x1/n = n a
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:
• 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) = .
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 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.
4 1. FUNCTIONS
For example, sin and cos are both periodic with period 2π, and
tan and cot are periodic with period π. 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) = (x + 1)/(x − 1).
3
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. 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 sec and csc periodic functions? If yes, what is their period?
(2) Let f (x) = x2/3 . Is f an algebraic function?
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,
3. OPERATIONS ON FUNCTIONS 5
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).
Similarly, if h is the function given 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 effect of multiplication and division on functions can be de-
scribed similarly. If f is a function and g is the function given by
f (x) = c · g(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 .
At this point, the reader should stop and think about what happens
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
|c|), but each point on the graph will be reflected through the horizontal
axis.
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
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 f g is the intersection
of the domains of f and g. Furthermore, (f + g)(x) = f (x) + g(x),
6 1. FUNCTIONS
(f −g)(x) = f (x)−g(x), and (f g)(x) = f (x)g(x). 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).
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
way sends x to g(f (x)) and is called the composition of f and g. It is
denoted by g ◦ f . Note that in g ◦ f , first f , and then g is applied.
Example 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
(g ◦ f )(x) = g(f (x)) = x2 + 1,
while
(f ◦ g)(x) = f (g(x)) = (x + 1)2 = x2 + 2x + 1.
Note that f ◦ g and g ◦ f are, in general, different functions.
3.3. Exercises.
(1) Sketch the graph of f (x) = x2 , g(x) = (x − 3)2 , and h(x) =
(2x + 5)2 .
(2) Sketch the graph of f (x) = cos 2x, g(x) = sin(x − 2), and
h(x) = 3 tan x.
(3) Show examples for f and g when g ◦ f is defined for all real
numbers, but f ◦ g is not.
(4) Show examples when f ◦ g = g ◦ f .
4. Viewing the Graphs of Functions

The graph of a function f is the set {(x, f (x))|x ∈ D(f )}. It is
a good way of visually describing what a function does. Today, we
have plenty of advanced tools, such as computer 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 (x ), then we say
that f is increasing on I. Visually, this means that the graph of f goes
roughly from the southwest to the northeast while x ∈ I. Similarly,
if, for all real numbers x and x in I, it is true that x < x implies
f (x) > f (x ), then we say that f is decreasing on I. In terms of the
4. VIEWING THE GRAPHS OF FUNCTIONS 7
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 [x1 , x2 ] in which
the value of x can range, we may get an error message, or the com-
puter may simply substitute default values for x1 and x2 . For example,
the software package Maple 13 uses the default values x1 = −10 and
x2 = 10. The interval [x1 , x2 ] is often called the viewing window.
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.
For functions like f (x) = x, g(x) = |x|, 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 (−∞, −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 x = −1/2.
Trigonometric functions, with their periodicity, are particularly
good examples to demonstrate what software package can and cannot
do. The reader is encouraged to plot the graph of the functions sin x,
8 1. FUNCTIONS
cos 2x, tan(x/4), and, finally, sin(1/x) 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.
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 ∈ 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. A function f : A → B is called one-to-one if it sends
different elements into different elements, that is, if x = x implies that
f (x) = f (x ).
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 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 −1 (y) = x if f (x) = y.
Example 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
(y − 7)/2 = x. As x = f −1 (y), it follows that f −1 (y) = (y − 7)/2.
2
The preceding example shows a general strategy for finding the in-
verse of a function. Write the equation f (x) = y, with the appropriate
algebraic expression replacing f (x). Then solve for x. 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
f −1 (y).
Example 3. If A is the set of positive real numbers, B is the set
√ than 1, and f : A → B is given by
of real numbers that are larger
−1
f (x) = x + 1, then f (y) = y − 1.
2
5. INVERSE FUNCTIONS 9
Solution: We have f (x) = x2 + 1 = y. So x2 = y − 1, and because we

know that x is positive
√ and y > 1, we can take the
√ square root of both
−1
sides, leading to x = y − 1. Hence, f (y) = y − 1. 2
Finally, we point out that if f is a one-to-one function with domain
A and range B, then f −1 ◦ f is the identity function of A and f ◦ f −1
is the identity function of B.
For instance, using the functions of Example
3, for all positive
√ real
−1
numbers x, the identity (f ◦ f )(x) = (x + 1) −√1 = x = x
2 2
holds, and for all y > 1, the identity (f ◦ f −1 )(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) = xα ,
where α 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 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 logm .
So if f (x) = xm = 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 (xa ) = a log x.
√
(IV) log b x = logb x .
(V) aloga x = x.
(VI) loga (ax ) = x.
10 1. FUNCTIONS
The last two rules simply express the fact that the functions f (x) =
ax and f −1 (y) = loga (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. For positive real numbers a, b, and x, we have
logb (x)
loga x = .
logb (a)
Proof. Start with the identity
x = aloga x .
Now take the logarithm of base b of both sides to get
logb x = loga logb ax.
Now divide both sides by logb to get the identity of the theorem. 2
Example 4. We can use Theorem 1 to compute log16 (256) from
log2 (256) as follows:
log2 (256) 8
log16 (256) = = = 2.
log2 (16) 4
So if a calculator or computer can provide the logarithm of all
positive real numbers in one base, we 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 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 x.
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 2π or π, 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 [−π/2, π/2].
In that interval, sin is a one-to-one function (since it is increasing),
and its range is the interval [−1, 1]. So its inverse is the function
6. THE VELOCITY PROBLEM AND THE TANGENT PROBLEM 11
sin−1 : [−1, 1] → [−π/2, π/2]. That is, if y ∈ [−1, 1], then sin−1 y
is the (only) x ∈ [−π/2, π/2] for which sin x√= y. For instance,
sin−1 (1/2) = π/6, while sin−1 (0) = 0 and sin−1 ( 2/2) = π/4.
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, π]. So cos−1 is a function with domain
[−1, 1] and range [0, π]. Similarly, tan−1 is the inverse function of the
tan function that is restricted to the interval (−π/2, π/2). Its domain
is the set of all real numbers, and its range is the interval (−π/2, π/2).
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 ?
(2) If we are given loga x, how can we compute log1/a (x)?
(3) For which values of a is loga 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) Is it true that if g is the inverse function of the one-to-one
function f , then g is one-to-one?
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) v= ,
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.
12 1. FUNCTIONS
Then Equation (1.1) yields

100 mi mi
v= = 50 ,
2 hr hr
so the average velocity of the car for the given two-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
10 mi 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
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
6. THE VELOCITY PROBLEM AND THE TANGENT PROBLEM 13
velocity for the time period from t1 to t2 , we simply compute the value
of the fraction
s(t2 ) − s(t1 )
.
t2 − t1
This fraction is precisely the slope of the line that intersects the graph
of the function s at points (t1 , s(t1 )) and (t2 , s(t2 )). If we choose t1
and t2 closer and closer together, then these points will get closer and
closer together as well. Finally, if we set t1 = t2 , then we will not
immediately know the slope of the line that touches the graph of s at
the point (t1 , s(t1 )) 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 t1 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 one hour at a speed of 60 miles per hour, then
two hours at a speed of 45 miles per hour. What is the average
speed of the car during this three-hour period?
(2) Consider the car of the previous exercise. What is its average
speed during the first two hours of its trip?
(3) I drove at 40 miles per hour for two 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 three-hour drive?
(4) 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 P Q is
between 0 and 0.01?
(5) 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 P Q
is between 1 and 1.01?
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 + x). Let us compute the values of f (x) for various real numbers
x that are close to 0. We find that
• f (1) = 1/2,
• f (1/2) = 2/3,
• f (1/3) = 3/4, and, in general,
• f (1/n) = n/(n + 1).
Similarly, for negative values of x, we get
• f (−1/2) = 2,
• f (−1/3) = 3/2,
• f (−1/4) = 4/3, and, in general,
• f (−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 < x < 0, then
1 < f (x) < n/(n − 1). So for instance, if we want f (x) to be closer
1 1
than 1000 to 1, then any choice of x in the interval [0, 999 ) or any choice
1
of x in the interval (− 1001 , 0] will work. That is, any choice of x in the
1
interval (− 1001 1
, 999 ) 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 4. 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 (x) = L.
x→a
15
16 2. LIMITS AND DERIVATIVES
So, if f is the starting example of this section, then limx→0 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 5. Let

1 if 0 ≤ x,
g(x) =
0 if x < 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 and sometimes equal 0 for x ∈ I,
no matter how small I is. There is no number L such that both 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→0 g(x) does not exist.
Second, if limx→0 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→0 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→0 f (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→0 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 1016 to 1, but then it cannot
also be closer than 1016 to 1.0001.
An analogous argument shows that no function can have two dif-
ferent limits at any one point.
7. THE LIMIT OF A FUNCTION 17
Sometimes it can happen that h is not even defined in a, but

limx→0 h(x) still exists. Note that the fact that h(a) is not defined
is not a problem since the definition of limits specifically states that x
should not be equal to a anyway.
Example 6. Let h(x) = (x2 − 9)/(x − 3). Then h is defined for
all real numbers except x = 3. Still, limx→3 h(x) = 6. In particular,
limx→3 h(x) exists.
Solution: If x = 3, then
x2 − 9 (x + 3)(x − 3)
f (x) = = = 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. 2
At this point, the reader should test his or her understanding of the
material by finding limx→−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
sin x
lim = 1.
x→0 x
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 ap-
proaches 0, eventually x and sin x will get so small that the computer
will no longer manipulate them, or their ratio, accurately.
Finally, we point out that in the definition of the limit, the require-
ment that f (x) get close to L and stay close to L is important. Consider
the function f (x) = sin (1/x) around x = 0. As x approaches 0, the
value of 1/x will increase very fast, and so it will equal a multiple of π
many times. All those times, f (0) = 0 will hold, so f (x) will be as
close to 0 as possible. However, limx→0 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 not stay arbitrarily close
to 0, no matter how close x is to 0.
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 5. 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 limx→a f (x) = L,
if, for all > 0, there exists δ > 0 such that if |x − a| < δ, then
|f (x) − L| < .
Example 7. We have limx→0 2x sin x = 0.
Solution: Let be any positive real number. Then let δ = /2. We
know that | sin x| ≤ 1 for all x. So if |x − 0| = |x| < δ = /2, then
|f (x) − 0| = |f (x)| = |2x sin x| ≤ |2x| < 2δ = , as required. 2
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 5. 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 limx→0 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.
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 5, then
lim g(x) = 0.
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.
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.
x→a+
For instance, if g is the function defined in Example 5, then
lim g(x) = 1.
x→0+
7. THE LIMIT OF A FUNCTION 19
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 4) 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 verify-
ing 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
x
h(x) =
|x|
as x approaches 0 and deciding if the limits limx→0 h(x), limx→0− h(x),

and limx→0+ h(x), exist.
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 = ∞, then the values of f have to
get arbitrarily close to ∞; that is, they have to get as large as we want.
This is the content of the following definition.
Definition 8. Let f : R → R be a function. We say that the limit

of f in a is ∞ 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 −∞ 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 ∞ is denoted by
lim f (x) = ∞.
x→a
Example 8. Let f (x) = 1/x2 . Then limx→0 f (x) = ∞.

Solution: If we want f (x) to be larger than an arbitrary positive
realnumber N , all we need to do is to choose x from the interval
2
(− 1/N , 1/N ). Then x < 1/N will hold, implying that f (x) =
1/x2 > N . 2
Similarly, if g(x) = −1/x4 , then limx→0 g(x) = −∞. Note that

if the limit of a function at a given point a is ∞ or −∞, 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.
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 ∞ or to be close to a real
number.
Definition 9. Let f : R → R be a function. We say that the limit
of f in a is ∞ if, for all positive real numbers N , there exists > 0
such that if |x − a| < , then f (x) > N .
Similarly, let g : R → R be a function. We say that the limit of g
in a is −∞ if for all negative real numbers M , there exists > 0 such
that if |x − a| < , then g(x) < M .
7.4.2 One-Sided Infinite Limits. One-sided infinite limits are defined
in an analogous way, as we can see in the following definition.
number. We say that the left-hand limit of f in a is ∞ if the values of
f (x) get arbitrarily large and stay arbitrarily large when x is suitably
close to a and x < a.
Similarly, we say that the right-hand limit of f in a is ∞ if the
values of f (x) get arbitrarily large and stay arbitrarily large when x is
suitably close to a and x > a.
Example 9. Let f (x) = 1/x. Then f is not defined in 0. Further-
more, limx→0− = −∞ and limx→0+ = ∞. As the two one-sided limits
are different, limx→0 does not exist.
Solution: We can make f (x) = 1/x smaller than any given negative
number M by choosing x from the interval (1/M, 0). We can make
x larger than any positive number P by choosing x from the interval
(0, P ). 2
8. LIMIT LAWS 21
7.5. Exercises.
2
(1) Find limx→3 x −4x+3
x−3
.
1
(2) Does limx→0 |x| exist?
(3) Give an example of a function f such that limx→0− f (x) = 0
and limx→0+ f (x) = ∞.

(4) Does limx→0 x13 + x12 exist?
(5) Give an example of a function f such that limx→1− f (x) = ∞,
limx→1+ f (x) = −∞, 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),
x→a x→a x→a
(IV)
lim (c · f )(x) = c · lim f (x),
x→a x→a
where c is a real number, and
(V)

f limx→a f (x)
lim (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 f (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 10. 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 |x| = 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) limx→2 (f − g)(x) = limx→2 f (x) − limx→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) = 3 limx→2 f (x) + 2 limx→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)

lim f (x)
f 2 1
lim (x) = x→2 = = .
x→2 g lim g(x) 4 2
x→2
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,
we get a power of that function. Applying law III each time, we get
that for all positive integers n,
n
n
(2.1) lim (f (x)) = lim f (x) .
x→a x→a
Note that we have essentially applied this rule in the special case
of n = 2 when we computed limx→2 x2 in Example 10.
The reader is invited to verify that the limits of the constant
function f (x) = c and the identity function f (x) = x are given by
limx→a c = c for all a and limx→a x = a. Formal proofs will be given in
the next section.
8. LIMIT LAWS 23
Applying Equation (2.1) to the identity function f (x) = x yields

the equation
(2.2) lim xn = an .
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 n f (x) = n lim f (x).
x→a x→a
(Here f (x) has to be nonnegative if n is even.) So, in particular, if

f (x) = x, then
√ √
lim n x = n a.
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) = 2010. Let g(x) = x + 2 for all real numbers. Then
f (x) = g(x) unless x = 2, and hence limx→a f (x) = limx→a g(x) = 4.
The statement 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 can be sig-
nificantly strengthened. See Exercise 8.4(1) for a possible direction for
that.
Second, Equation (2.2) can be interpreted by saying that the limit
of a power function f (x) = xn 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. 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.
Corollary 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, 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)
2
So far all the relationships that we discussed for limits involved
equations. We will now discuss two rules that, involve inequalities.
Theorem 3. Let f and g be two functions and assume that, for all
real numbers x, the inequality f (x) ≤ 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
g(x) > Lf − = Lg + .
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
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. 2
Note that in Theorem 3, the fact that the inequalities are not strict
is important. See Exercise 8.4(2) for a relevant question.
Corollary 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 limx→a g(x) = L.
9. CONTINUOUS FUNCTIONS 25
Proof. If limx→a g(x) exists, then by applying Theorem 3 to f and

g, it follows that L ≤ limx→a g(x), and by applying Theorem 3 to g
and h, it follows that limx→a g(x) ≤ L. So if limx→a g(x) exists, it is
equal to L. In Exercise 8.4(3) you are asked to prove that this limit
exists. 2
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.
Example 11. Let g(x) = x cos(log x). Then limx→0 g(x) = 0.
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→0 f (x) =
limx→0 h(x) = 0, so we can apply Corollary 2 to prove our claim. 2
We could not have used limit law III to compute limx→0 g(x) since
limx→0 cos(log x) does not exist. You are asked to prove this in
Exercise 8.4(4).
8.4. Exercises.
(1) 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
limx→a f (x) = limx→a g(x) as long as these limits exist? Why
or why not?
(2) 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).
(3) Explain why limx→a g(x) exists if the conditions of Corollary
2 hold.
(4) Prove that limx→0 cos(log√x) does not exist.
(5) Prove that limx→0 |x sin( x)| = 0.
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 11. A function f is called continuous at a if the equality

lim f (x) = f (a)
x→a
holds.
Note that Definition 11 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 11 would not
be satisfied.
If a function f : R → R is continuous at all a ∈ 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 defini-
tion of continuity is very close to that of limits, it is not surprising that
their precise definitions are also similar.
Definition 12. Let f be defined in an open interval containing a.
We say that f is continuous in a if, for all > 0, there exists δ > 0
such that if |x − a| < δ, then |f (x) − f (a)| < .
9.1. Examples of Continuous Functions. Let us consider some of the
most frequently used continuous functions.
Example 12. Polynomial functions are continuous.
Solution: This is a direct consequence of Theorem 2, which we dis-
cussed in the last section. Theorem 2 stated that the limit of a polyno-
mial function at a is equal to the value of the polynomial at a, which
is precisely what the definition of continuity requires. 2
There are many classes of functions that are continuous at every

point where they are defined. If they are not defined somewhere, then,
of course, they cannot be continuous there.
Example 13. The following are examples of functions that are con-
tinuous in every point where they are defined.
9. CONTINUOUS FUNCTIONS 27
(I) Rational functions

(II) Exponential functions
(III) Trigonometric functions
(IV) Logarithmic functions
(V) Inverse trigonometric functions
The reader is invited to recall the graphs of each of these functions
and verify that they consist of continuous lines as long as they are
defined.
9.2. Functions That Are Not Continuous. It is time to stop for a moment
and think about functions that are not continuous at a given point a.
There can be three reasons for this. First, it could be that f (a) is not
defined, for instance, when f is a rational function whose denominator
becomes 0 when x = a. Or it could be that g is defined at a, but
limx→a g(x) does not exist at a. An example of this is the function
defined by g(x) = 1 if x ≥ 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
|x| = 3 and h(x) = 1 if |x| = 3. Let a = −3. Then
1
h(a) = 1 = lim h(x) = − .
x→a 6
The interested reader is invited to think about the following
example.
Excursion 1. The following function is not continuous anywhere.
Let f (x) = 1 if x is rational and let f (x) = 0 if x is irrational.
9.3. New Continuous Functions from Old. It follows from the limit laws
that several transformations preserve the continuous property of
functions.
Theorem 4. 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 14. It follows from successive

√ applications of the previous
theorem that h(x) = ex · sin x + 3 ln x − 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 5. Let f and g be two functions such that f is continuous
at a and g is continuous at f (a). Then the composition function g ◦ f
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 15. 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) = x. 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. 2
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 13. We say that the function f is left-continuous at a
if f (a) = limx→a− f (x). Similarly, we say that f is right-continuous at
a if f (a) = limx→a+ f (x).
Example 16. Let g be the function defined by g(x) = 1 if x ≥ 0
and g(x) = 0 if x < 0. Then limx→0− g(x) = 0 = 1 = g(0), so g is not
left-continuous at 0. On the other hand, limx→0+ g(x) = 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.
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 prop-
erty 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
10. LIMITS AT INFINITY 29
gap in the graph of the function, contradicting the requirement that

the function be continuous. This is the content of the next theorem.
Theorem 6 (Intermediate Value Theorem). Let f be a function
that is continuous on the interval [a, b]. Then, if f (a) = y1 and f (b) =
y2 and y is a real number that is between y1 and y2 , then there exists
x ∈ [a, b] such that f (x) = y.
In other words, f takes all values between y1 and y2 on the interval
[a, b].
Example 17. 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. 2
9.6. Exercises.
(1) Where is tan x continuous?
(2) Where is 1/x not continuous?
(3) Where is 5x23x+2
−6x+1
continuous?
2
(4) Where is sin(x ) continuous?
(5) Prove that the equation x5 −x−1 = 0 has a root in the interval
(−1, 2).
(6) Prove that the equation x3 − 3x − 1 = 0 has at least two roots
in the interval (−1, 2).
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 ∞ or at −∞.
Definition 14. Let f : R → R be a function that is defined on
some interval (b, ∞). We say that the limit of f at ∞ 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 large.
The fact that the limit of f at ∞ is L is expressed by the notation
lim f (x) = L.
x→∞
This definition follows the idea of the definition of limits at finite
points. Indeed, in order for limx→∞ 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 “x is large enough” means that x is
in a suitably selected neighborhood of ∞, in other words, in an open
interval (c, ∞). Recall that this is analogous to what we required in the
finite case. There we said that limx→a f (x) = L if f (x) got arbitrarily
close to L and stayed arbitrarily close to L once x was suitably close
to a, that is, when x was in a suitably selected neighborhood of a.
Example 18. Let f (x) = 1/x. Then
lim f (x) = 0.
x→∞
Solution: If we want the value of f (x) to be closer than to 0, all we

have to do is to select x such that x > 1/ holds. Once x gets past 1/,
the values of f (x) will stay between 0 and . 2
The definition of limits at −∞ is what the reader probably expects.
Definition 15. Let f : R → R be a function defined on some
interval (−∞, b). We say that the limit of f at −∞ 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 −∞ is L is expressed by the notation
lim f (x) = L.
x→−∞
Example 19. Let f (x) = 1/x2 . Then

lim f (x) = 0.
x→−∞
Solution: If we want to get f (x) closer√

than to 0 and keep it there,
it suffices to choose x such that x < −1/ . Then x2 > 1/, and hence
f (x) = 1/x2 < . 2
10.1.1. The Formal Definition of Limits at Infinity. The formal defini-

tion of limits at infinity is very similar to that of limits at finite points.
The only difference is in the formal description of what it means to be
in a neighborhood of infinity versus what it means to be in a neighbor-
hood of a real number.
interval (b, ∞). We say that limx→∞ f (x) = L if, for all positive real
numbers , there exists a positive real number N such that if x > N ,
then |f (x) − L| < .
The formal definition of limits at negative infinity is analogous. The

only difference is again in the formal description of what it means for x
to be in a neighborhood of −∞. It means to be in an interval (−∞, c).

interval (−∞, b). We say that limx→−∞ f (x) = L if, for all positive real
numbers , there exists a negative real number N such that if x < N ,
then |f (x) − L| < .
10.1.2. The Graphical Meaning of a Finite Limit at Infinity. If a func-

tion f has limit L at ∞ or −∞, then the graph of the function will
approach the horizontal line y = L at that infinity. The graph may or
may not actually touch that line or even become that line. The line
y = L is called a horizontal asymptote of the graph of y = f (x) when
limx→∞ f (x) = L or limx→−∞ f (x) = L holds.
10.2. Infinite Limits at Infinity. It can happen that the limit of a function
at ∞ is not a real number but rather ∞ or −∞.

interval (b, ∞). We say that the limit of f at ∞ is ∞, denoted by
lim f (x) = ∞,
x→∞
if f (x) gets arbitrarily large and stays arbitrarily large if x gets suffi-
ciently large.
Example 20. Let f (x) = ex . Then limx→∞ f (x) = ∞.
Solution: In order to get f (x) to be larger than some given positive

real number M , it suffices to choose x > ln M . 2
The following notation is defined in an analogous way:
(I) limx→∞ f (x) = −∞.
(II) limx→−∞ g(x) = ∞.
(III) limx→−∞ h(x) = −∞.
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→∞ 1 − x = −∞, while
limx→−∞ x2 = ∞, and limx→−∞ x3 = −∞.
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.
interval (b, ∞). We say that limx→∞ f (x) = ∞ if, for all positive real
numbers M , there exists a positive real number N such that if x > N ,
then f (x) > M .
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 21. We have
x+3
lim = 1.
x→∞ x − 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 ∞, they both have limit ∞, so, by the limit law for quotients, the
limit of their quotient is 1.”
The problem with this argument is that ∞ is not a number. So
∞/∞ is not defined. It is possible for f and g both to have limit ∞
at ∞, and for f /g to have limits c at ∞, for any given real number c.
Indeed, let f (x) = cx and let g(x) = x.
Instead, we can solve Example 21 as follows.
Solution:
x+3 (x − 4) + 7
lim = lim
x→∞ x − 4
x−4
x→∞
7
= lim 1 +
x→∞ x−4
7
= 1 + lim
x→∞ x − 4
=1+0
= 1.
2
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) ∞ + (−∞)
(II) ∞ · 0 and −∞ · 0
(III) 1∞ and 1−∞
The following theorem is very useful when dealing with limits at ∞.
Theorem 7. Let r be a positive rational number. Then

1
lim = 0.
x→∞ xr
If r is an integer, then this statement follows from the fact that

limx→∞ 1/x = 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
p p/q q p
the theorem for x , and then, using the root law, for x = x .
Many limits can be computed with the help of this theorem.
Example 22. We have
x2 + 3x + 1
lim = 0.
x→∞ x3
Solution: We have
x2 + 3x + 1 x2 3x 1
3
= 3
+ 3 + 3,
x x x x
and each of the three summands has limit 0 at ∞ by the preceding
theorem. Hence, by the limit law for sums, so does their sum. 2
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
x2 + 3x + 1
lim = 0.
x→∞ x3 + 3x2 + 4x + 5
10.4. Exercises.
(1) Find limx→∞ xx+1
2 +4 .
3x2 +4x+1
(2) Find limx→∞ x2 +4 .
3
(3) Find limx→∞ xx2+2x
+4
.
3x2 +4x+1
(4) Find limx→−∞ x2 +4 .
(5) Let R(x) = p(x)/q(x) be a rational function. Explain how
limx→∞ R(x) depends on p(x) and q(x).
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 Pi P . The slope of these lines will approach a certain
slope, and so the lines Pi P will approach a certain line. That line is
called the tangent line of f at P .
Definition 20. 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
f (x) − f (a)
(2.5) lim ,
x→a x−a
provided that this limit exists.
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 23. 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) x2 − 9
lim = lim = lim (x + 3) = 6.
x→3 x−3 x→3 x − 3 x→3
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 21. Let f (t) be a function such that f (t) is the distance
covered by a moving object in t units of time. Then the instantaneous
velocity of the object a units of time after it starts moving is
f (t) − f (a)
v(t) = lim ,
t→a t−a
provided that this limit exists.
11. DERIVATIVES 35
Example 24. 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) = 12 t2 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
v(4) = lim = lim = lim = 4.
t→4 t−4 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. 2
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 22. Let f be a function. The derivative of f at a is
the limit
f (x) − f (a)
f (a) = lim
x→a 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 tangent line to the curve f (x) = x3 at x = 0.
(3) 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.
(4) The distance covered by a car in a certain time period is de-
scribed by the function
t2 (b − m)
f (t) = tm + ,
2
where b and m are positive constants. Let us assume that

t ∈ [0, 1]. Find the instantaneous velocity of the car at a
given moment t = a.
(5) Find the derivative of the function f (x) = x + 5 at a = 3.
12. The Derivative as a Function

12.1. Rates of Change. In the last section, we saw that the derivative
of a function at a given point was a common generalization of the
concepts of tangent lines and instantaneous velocities. We will now
further elaborate on that, in order to understand how far-reaching the
concept of derivatives is.
If f is a function and f (x) = y, then the quantity denoted by y
depends on the quantity denoted by x. This is sometimes expressed
by saying that x is the independent variable and y is the dependent
variable. If x changes, then the change in y can be described in terms
of the change in x.
In particular, if x changes from x1 to x2 , then y = f (x) changes
from y1 = f (x1 ) to y2 = f (x2 ). The average rate of change for the
interval (x1 , x2 ) is then the ratio
y2 − y1 Δy
= ,
x2 − x1 Δx
where Δx is the change (or increment) of x. We have to use the word
“average” since we only have information about the values of y at the
endpoints of the interval (x1 , x2 ); we do not know how f (x) = y behaves
in the rest of the interval. If we want more precise information, such
as the instantaneous rate of change of f (x) = y at a given point, then
we have to use the notion of limits again, just as we have done twice
in the last section. That is, at a given point x = a, we define the
instantaneous rate of change of f (x) = y as
f (x2 ) − f (a) Δy
lim = lim .
x2 →a x2 − a Δx→0 Δx
12.2. The Derivative of the Function f . Recall that, at a given point a,

the derivative of the function f is defined as the limit
f (x) − f (a)
f (a) = lim .
x→a x−a
Note that this definition associates the real number f (a) to the
real number a. That is, f : R → R is a function. The function f is
called the derivative of f . The operation that takes f into f is called
differentiation. This explains the following definition.
12. THE DERIVATIVE AS A FUNCTION 37
Definition 23. A function f is called differentiable at a if f (a)

exists.
We say that f is differentiable on the interval (a, b) if f is differen-
tiable at d for all d ∈ (a, b).
Example 25. The function f (x) = x3 is differentiable in every real
number a, and f (a) = 3a2 .
Solution: We have
f (x) − f (a) x3 − a3
lim = lim
x→a x−a x→a x − a
(x − a)(x2 + xa + a2 )
= lim
x→a
2 x − a
= lim x + xa + a2
x→a
2
= 3a .
2
The functions we have considered so far had only one independent
variable, usually the variable x. The dependent variable was usually
denoted by y, so y = f (x) held. So it was always clear that the de-
rivative was taken with respect to x. However, there are circumstances
when this is not so clear, usually when f depends on more than one
variable. Therefore, there are additional ways to denote the function
f such as
• dx
dy
,
• dx ,
df
• dx
d
f (x),
• Dx f (x), or
• Df (x).
12.3. Differentiability Versus Continuity. The definitions of differentia-

bility and continuity are similar. Which one imposes stronger require-
ments on a function at a given point? The following theorem shows
that differentiability is the stronger requirement.
Theorem 8. If f is differentiable at a, then f is continuous at a.
Proof. If f is differentiable at a, then
f (x) − f (a)
f (a) = lim ;
x→a x−a
in particular, the limit shown on the right-hand side exists. Multiplying

both sides by the function g(x) = x − a, we get
f (x) − f (a)
f (a)(x − a) = (x − a) lim ;
x→a x−a
Now, taking limits at a on both sides, we obtain
(2.6) f (a) · lim (x − a) = lim (f (x) − f (a)),
x→a x→a
since we can apply the limit law for products on the right-hand side to
get that

f (x) − f (a) f (x) − f (a)

lim (x − a) lim = lim (x − a) ·
x→a x→a x−a x→a x−a
= lim (f (x) − f (a)) .
x→a
Finally, note that the left-hand side of (2.6) is equal to 0 since

f (a)(x − a) is a polynomial that takes value 0 when x = a. Hence, the
right-hand side of (2.6) is equal to 0 as well, that is,
0 = lim (f (x) − f (a)) = (lim f (x)) − f (a).
x→a x→a
Adding f (a) to both the far left and far right sides, we get that
f (a) = lim f (x),
x→a
which means that f is continuous at a. 2

The converse of Theorem 8 is not true. Indeed, the function f (x) =
|x| is continuous at a = 0, but it is not differentiable. The reader is
invited to prove this by showing that
|x| − 0 |x| |x| |x| − 0
lim− = lim− = lim+ = lim+ ,
x→0 x−0 x→0 x x→0 x x→0 x−0
and hence
|x| − 0
f (0) = lim
x→0 x − 0
does not exist.
In general, there are several reasons a continuous function may fail
to be differentiable at a given point. It could be that the graph of the
function has a “corner,” like that of |x| at 0, and hence the slope of
the tangent line cannot be defined because the left-hand limit and the
right-hand limit of the lines approaching the purported tangent line
are not equal. Or, it could be that the function has a vertical tangent
line at the given point. See Exercise 12.5(5) for an example of this.
12. THE DERIVATIVE AS A FUNCTION 39
12.4 Higher-Order Derivatives. In upcoming chapters, it will often be

useful to consider not only the derivative of a function but also the
derivative of the derivative and even the derivative of the derivative
of the derivative. These functions appear so often that they have their
own names.
If f is a differentiable function on an interval (a, b) and its derivative
f is also differentiable on (a, b), then the derivative of f is called

the second derivative of f and is denoted by f . Similarly, if f is

differentiable on (a, b), then its derivative is called the third derivative
of f and is denoted by f . Higher-order derivatives are defined in
an analogous way, but denoted slightly differently. For instance, the
seventh derivative of f is denoted by f (7) , and, in general, the nth
derivative is denoted by f (n) .
Example 26. We have seen in Example 25 that if f (x) = x3 , then

f (x) = 3x2 . Therefore,
f (x) − f (a)
f (a) = lim
x→a x−a
3(x2 − a2 )
= lim
x→a x−a
= lim 3(x + a)
x→a
= 6a.
So f (x) = 6x.
In Exercise 12.5(1), you are asked to prove that f (x) = 6 for
all x, and in Exercise 12.5(2), you are asked to compute higher-order
derivatives of f .
12.5. Exercises.
(1) Let f (x) = x3 . Prove that f (x) = 6 for all real numbers x.
(2) Let f (x) = x3 . Compute f 4 (x). What can be said about
higher-order√derivatives of f ?
(3) Let f (x) = x. Compute f (a) at some point a > 0.
(4) Let f (x) = 1/x. Compute f (a) at some point a = √ 0.
(5) Let f be defined on the interval [0, 2] by f (x) = 1 − x2 if
0 ≤ x ≤ 1, and f (x) = − 1 − (x − 2)2 if 1 < x ≤ 2. So
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
line, that is,

f (x) − f (1)
lim = ∞.
x→1 x−1
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 9. Let c be a real number and let f (x) = c for all x.

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 9. We have

f (x) − f (a)
f (a) = lim
x→a x−a
c−c
= lim
x→a x − a
= 0.
Note that limx→a (c − c)/(x − a) = 0 since (c − c)/(x − a) = 0 for all

values x = a. 2
We now turn our attention to a more general class of power func-

tions, those of the form f (x) = xn , where n is a positive integer. Let
us recall the algebraic identity
xn − an = (x − a) · (xn−1 + xn−2 a + · · · + xan−2 + an−1 ).
Theorem 10. Let n be a positive integer and let f (x) = xn . Then
f (a) = nan−1 .
41
42 3. RULES OF DIFFERENTIATION
Proof. We have
f (x) − f (a)
f (a) = lim
x→a x−a
xn − an
= lim
x→a x − a
(x − a) · (xn−1 + xn−2 a + · · · + xan−2 + an−1 )

= lim
x→a x−a
= lim (x n−1
+ x a + · · · + xan−2 + an−1 )
n−2
x→a
= nan−1
. 2
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 10 holds even if n is not a positive integer.
That is, for all real numbers α, if f (x) = xα , then f (x) = αxα−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 func-
tions, 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 11. 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 + g)(x) − (f + g)(a)
(f + g) (a) = lim
x→a
x−a
f (x) − f (a) g(x) − g(a)

= lim +
x→a x−a x−a

= f (a) + g (a). 2
The other two rules and their proofs are so similar that they are
left as exercises.
at a. Then f − g is differentiable at a, and
(f − g) (a) = f (a) − g (a).
13. DERIVATIVES OF POLYNOMIAL AND EXPONENTIAL FUNCTIONS 43
Theorem 13. 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,
(f g) = f g , and (f /g) = f /g . We will learn some more complicated
rules to compute the derivatives of f g and f /g in the next section.
Theorems 11, 12, and 13 enable us to compute the derivative of any
polynomial function.
Example 27. 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 10. Then we can apply Theorems 11, 12, and
13 to get
p (x) = (3x3 ) + (5x2 ) − (6x) + (8)
= 3(x3 ) + 5(x2 ) − 6(x) + (8)
= 9x2 + 10x − 6. 2
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 (x) − f (a)
f (a) = lim
x→a x−a
b − ba
x
= lim
x→a x − a
ba+z − ba
= lim
z→0 z
b −1
z
= ba lim
z→0 z
= ba f (0).
Several comments are in order. First, note the substitution z = x−a
in the third line. Second, note that ba 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.
In other words,
(3.1) f (x) = f (0)bx = f (0)f (x).
That is, the derivative of the function f is a constant multiple of f .

The constant in question is f (0), that is, limz→0 (bz − 1)/z. Numerical
experimentation suggests that the larger b is, the larger this limit is.
Graphical experimentation suggests this as well. Indeed, f (0) 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 = 1.
z→0 z
Definition 24. Let e be the real number such that
ez − 1
lim = 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 the next
section, we will see what this implies for the derivatives of exponential
functions with bases different from e.
13.3. Exercises.
(1) Prove that if f (x) = x1/2 and a = 0, then f (a) = 2√1 a .
(2) Prove that if f (x) = 1/x and a = 0, then f (a) = − a12 .
(3) Prove Theorem 12.
(4) Prove Theorem 13.
(5) Let f (x) = 3x3 − 4x2 + x − 2 + 4ex . Compute f (x).
14. The Product and Quotient Rules

14.1. The Product Rule. We mentioned in the last section that, in gen-
eral, (f g) = f g . For instance, if f (x) = 2x + 1 and g(x) = x + 2, then
(f g)(x) = 2x2 + 5x + 2, so (f g) (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.
14. THE PRODUCT AND QUOTIENT RULES 45

at a. Then f g is differentiable at a, and
(f g) (a) = f (a)g (a) + f (a)g(a).
Proof. By definition, we have
f (x)g(x) − f (a)g(a)
(3.2) (f g) (a) = lim .
x→a 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)

f (x)g(x) − f (x)g(a) f (x)g(a) − f (a)g(a)

(f g) (a) = lim +
x→a x−a x−a
f (x)g(x) − f (x)g(a) f (x)g(a) − f (a)g(a)
= lim + lim
x→a x−a x→a x−a
g(x) − g(a) f (x) − f (a)
= lim f (x) · + lim g(x) ·
x→a x−a x→a x−a

= f (a)g (a) + g(a)f (a). 2
Example 28. The derivative of h(x) = x2 ex can be computed as
follows. Let f (x) = x2 and g(x) = ex . Then
h (x) = (f g) (x)
= f (x)g (x) + f (x)g(x)
= x2 (ex ) + (x2 ) ex
= x2 ex + 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 14.
at a and let us assume that g(a) = 0. Then f /g is differentiable at a,
and we have

f g(a)f (a) − f (a)g (a)
(a) = .
g g(a)2
Proof. By definition, we have

f
f (x)
g(x)
− f (a)
g(a)
(3.3) (a) = lim .
g x→a x−a
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) = lim .
g x→a (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 f (x)g(a) − g(a)f (a) + g(a)f (a) − f (a)g(x)
(a) = lim
g x→a (x − a)g(x)g(a)
g(a) f (x) − f (a) f (a) g(x) − g(a)
= lim · − lim ·
x→a g(x)g(a) x−a x→a g(x)g(a) x−a

g(a)f (a) − f (a)g (a)
= .
g(a)2 2
Theorem 15 now enables us to compute the derivative of rational

functions.
Example 29. Let h(x) = (x + 3)/(x2 + 1). Find h (x).

Solution: Let f (x) = x + 3 and let g(x) = x2 + 1. Then f (x) = 1
and g (x) = 2x. So, by Theorem 15, we have
g(x)f (x) − f (x)g (x) x2 + 1 − (x + 3)2x −x2 − 6x + 1
h (x) = = = .
g(x)2 x4 + 2x2 + 1 x4 + 2x2 + 1
2
14.3. Exercises.
(1) Let h(x) = ex x3 . Find h (x) and h (x).
(2) Find a rule to compute (f 2 ) (x).
(3) Find a rule to compute (1/f ) (x).
(4) Use the result of the previous exercise to prove a formula for
g (x) if g(x) = xn for a negative integer n.
(5) Let g(x) = e−x . Find g (x).
(6) Let h(x) = x/ex . Find h (x).
15. DERIVATIVES OF TRIGONOMETRIC FUNCTIONS 47
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 16. We have (sin x) = cos x.
Proof. Recall the identity sin(a + b) = sin a cos b + sin b cos a. We
have
sin(x + h) − sin x
(sin x) = lim
h→0 h
sin x cos h + sin h cos x − sin x
= lim
h→0
h
(cos h) − 1 sin h cos x

= lim sin x +
h→0 h h
(cos h) − 1 sin h
= sin x lim + cos x lim .
h→0 h h→0 h
Note that as, h approaches 0, we certainly have limh→0 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.
Lemma 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 O of the circle and whose n vertices
are all on the unit circle. Then the area of the circle is π, and the area
of the n-gon is n · 12 · sin α, where α = 2π/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 α)/2, and the area of 1/n of the
circle bordered by the lines AO, BO, and the arc AB is π · α/(2π) =
α/2. So the ratio of the two areas is
(sin α)/2 sin α
= .
α/2 α
On the other hand, as n gets larger and larger, α 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 α/α, will get arbitrarily close to 1
and stay arbitrarily close to 1. 2
Lemma 2. The equality
(cos h) − 1
(3.4) lim =0
h→0 h
holds.
Proof. We will manipulate the expression ((cos h) − 1)/h so that
we can use the result of Lemma 1. First, we multiply both the numer-
ator and the denominator by cos h + 1 to get
(cos h) − 1 (cos2 h) − 1 − sin2 h
= = .
h h(1 + cos h) h(1 + cos h)
Therefore, we have
(cos h) − 1 sin2 h
lim = − lim
h→0 h h→0 h(1 + cos h)

sin h sin h
= − lim ·
h→0 h 1 + cos h

sin h sin h
= − lim · lim
h→0 h h→0 1 + cos h
= (−1) · 0 = 0. 2
We can now finish the proof of Theorem 16. At the end of the first
displayed chain of equations in that proof, we saw that
(cos h) − 1 sin h
(sin x) = sin x lim + cos x lim .
h→0 h h→0 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) = cos x as claimed. 2
The following theorem can be proved by very similar methods.
Theorem 17. The equality (cos x) = − sin x holds.
16. THE CHAIN RULE 49
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 18. We have (tan x) = sec2 x.
Proof. Note that tan x = sin x/cos x, so we can apply the quotient
rule. This leads to

sin x
(tan x) =
cos x
cos x · (sin x) − sin x(cos x)
=
cos2 x
cosx + sin2 x
=
cos2 x
1
=
cos2 x
= sec2 x. 2
The derivatives of the other three trigonometric functions are given
in the exercises.
15.1. Exercises.
(1) Prove that (cos x) = − sin x.
(2) Prove that (cot x) = − csc2 x.
(3) Prove that (csc x) = − csc x · cot x.
(4) Prove that (sec x) = sec x tan x.
(5) Let h(x) = ex cos x. Find h (x).
(6) Let h(x) = ex / sin x. Find h (x).
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) = x2 + 1, or r(x) = e4x . In this section, we will
learn a rule, called the chain rule, that applies in these situations.
Theorem 19 (Chain Rule). Let h(x) = f (g(x)), where g is differ-
entiable 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).
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 30. Find the derivative of h(x) = sin 3x.
Solution: Let f (x) = sin x and let g(x) = 3x. Then h(x) = f (g(x)),
so, by the chain rule, we have
h (x) = f (g(x)) · g (x) = (cos 3x) · 3 = 3 cos 3x. 2

√
Example 31. Let h(x) = x2 + 1. Find h (x).
√
Solution: Let f (x) = x and let g(x) = x2 + 1. Then h(x) = f (g(x)),
so, by the chain rule, we have
1 x
h (x) = f (g(x))g (x) = √ · 2x = √ . 2
2
2 x +1 2
x +1
Sometimes the chain rule is written in the Leibniz notation, that
is, as
dh dh dg
= · .
dx dg dx
16.2. Two Applications of the Chain Rule.

16.2.1. A Simple Way of Obtaining (cos x) . Recall that in the last
section, it took considerable time and effort to prove that (sin x) =
cos x. Finding (cos x) with similar methods is just as time-consuming.
On the other hand, the chain
rule enables us to compute (cos x) faster.
Recall that cos x = sin x + π2 . So we can write cos x as the com-
position of two functions, namely cos x = f (g(x)), with f (x) = sin x
and g(x) = x + π2 . So the chain rule applies, and we get
(cos x) = f (g(x)) · g (x)

π
= cos x + ·1
2
π π
= cos x cos − sin x sin
2 2
= 0 − sin x
= − sin x.
16. THE CHAIN RULE 51
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.
Theorem 20. Let a be a positive real number and let h(x) = ax .
Then we have
h (x) = ax ln a.
Proof. Note that
h(x) = ax = (eln a )x = ex ln a .
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) = ex ln a · ln a = ax ln a.
2
16.3. Proof of the Chain Rule. It is time that we proved the chain rule.
Proof of Theorem 19. As g is differentiable at x, we know that

g(x + r) − g(x)
(3.5) lim − g (x) = 0.
r→0 r
Set
g(x + r) − g(x)
t= − g (x).
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 − f (y) = 0.
s→0 s
Set
f (y + s) − f (y)
u= − 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
f (g(x + r)) − f (g(x))
f (g(x)) = lim
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 u, we get
(3.8) f (y + s) = f (y) + (f (y) + u)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) .
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) + t)r)
(3.11) = f (g(x)) + (f (g(x)) + u) · (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))
lim = lim (f (g(x)) + u)(g (x) + t)
r→0 r r→0

= lim f (g(x)) + lim u · lim g (x) + lim t
r→0 r→0 r→0 r→0

= f (g(x)) · g (x)
since both t and u approach 0 as r approaches 0. 2
16.4. Exercises.
(1) Let h(x) = (x2 + 1)5 . Find h (x).
(2) Let h(x) = sin (x2 ). Find h (x).
(3) Let h(x) = tan 3x. Find h (x).
(4) Let h(x) = esin x . Find h (x).
2
(5) Let h(x) = ex sin x . Find h (x).
17. IMPLICIT DIFFERENTIATION 53
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 + y 3 = 4xy.
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/dx of that function at
x = 2.
Consider Equation (3.12), and differentiate both sides with respect
to the variable x to get
d 3 d
x + y3 = (4xy).
dx dx
Now recall that y = y(x) is a function of x. So, when computing
(d/dx)y 3 on the left-hand side, we need to use the chain rule. On the
right-hand side, we need to use the product rule. Using these rules,
we get
dy dy
3x2 + 3y 2 = 4y + 4x .
dx dx
Expressing dy/dx from this equation, we get
dy (4y − 3x2 )
= .
dx (3y 2 − 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
x and y. That is, if (x, y) is on the curve, then (y, x) 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 (−π/2, π/2). That is, if
tan−1 (x) = y,
then
(3.13) x = tan y,
where y ∈ (−π/2, π/2).
Our goal is to determine
d dy
tan−1 (x) = .
dx dx
To that end, let us take the derivative of both sides of (3.13) with
respect to x. Recalling that
d
tan z = sec2 z and y = y(x),
dz
we get
dy
1 = sec2 y ·.
dx
Solving for dy/dx and recalling the identity sec2 z = 1 + tan2 z, we
obtain
dy 1 1 1
= 2
= 2
= .
dx sec y 1 + tan y 1 + x2
In other words, we proved the suprisingly simple formula
1
(tan−1 x) = .
1 + x2
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.
18. DERIVATIVES OF LOGARITHMIC FUNCTIONS 55
17.3. Exercises.
(1) Let C be the circle given by the equation x2 + y 2 = 169. Use
implicit differentiation to find the slope of the tangent line to
C at the point (5, 12).
(2) Prove that (sin−1 x) = √1−x
1
2.
−1
(3) Prove that (cos x) = − 1−x2 .
√ 1
(4) Prove that (cot−1 x) = − 1+x1

2.
−1 1
(5) Prove that (sec x) = x√x2 −1 .
(6) Prove that (csc−1 x) = − x√x12 −1 .
18. Derivatives of Logarithmic Functions

18.1. The Formula for (loga x) . As another powerful application of im-
plicit differentiation, we compute the derivative of the function f (x) =
ln x.
Theorem 21. We have
1
(ln x) =.
x
Proof. Set y = ln x. Then ey = x. Differentiating both sides with
respect to x, we get
dy
ey · = 1,
dx
dy 1
= y.
dx e
However, ey = x by definition, so
dy 1
=
dx x
as claimed. 2
It is now a breeze to determine the derivative of logarithmic func-
tions of any base.
Corollary 3. Let a = 1 be a fixed positive real number. Then
1
(loga x) = .
x ln a
Proof. Note that
log x
x = eln a a = e(ln a)(loga x) .
So ln x = (ln a) · (loga x) and

ln x
f (x) = loga x = .
ln a
As ln a is a constant, it follows that
1 1
f (x) = (ln x) =
ln a x ln a
as claimed. 2
18.2. The Chain Rule and ln x. An interesting consequence of Theorem
21 is the following.
Corollary 4. Let f (x) be a differentiable function that takes pos-
itive values only. Then
d f (x)
ln f (x) = .
dx f (x)
Proof. By the chain rule,
d d df f (x)
ln f = · = . 2
dx df dx f (x)
Example 32. Let f (x) = cos x. Then
d − sin x
ln(cos x) = = − tan x.
dx cos x
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 33. Let √
x3 x + 1
y= √ .
x−2
Compute dy/dx.
Solution: Taking logarithms, we get
1 1
ln y = 3 ln x + ln(x + 1) − ln(x − 2).
2 2
Now taking derivatives with respect to x and using Corollary 4, we
have
dy 1 3 1 1
· = + − .
dx y x 2(x + 1) 2(x − 2)
18. DERIVATIVES OF LOGARITHMIC FUNCTIONS 57
Finally, we can solve this equation for dy/dx to get

dy 3 1 1
=y + −
dx x 2(x + 1) 2(x − 2)
√
x3 x + 1 3 1 1
= √ · + − .
x−2 x 2(x + 1) 2(x − 2) 2
18.4. Power Functions Revisited. Recall that in an earlier section, we
proved that if n is a fixed positive integer, then (xn ) = nxn−1 . We
stated that this was the case for all 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 22. Let n be any real number. Then we have
d n
x = nxn−1 .
dx
Proof. Set y = xn . Let us assume for the case of simplicity that
x is positive. Taking logarithms, we have
ln y = n ln x.
Differentiating both sides with respect to x, we get
dy 1 n
· = .
dx y x
Solving for dy/dx yields
dy ny nxn
= = = nxn−1
dx x x
as claimed. 2
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→0 (eh − 1)/h = 1. Our new knowledge lets us express e more di-
rectly, as a limit.
Note that if f (x) = ln x, then f (x) = 1/x, so f (1) = 1. By the
definition of derivatives, this means that
ln(1 + h) − ln 1
lim = 1.
h→0 h
Observing that ln 1 = 0 and using the power rule of logarithms, we get
lim ln(1 + h)1/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

x
1
lim 1 + = e.
x→∞ x
Either of the last two formulas can help to determine the approximate
value 2.712828 of e.
18.6. Exercises.
d
√
(1) Compute dx ln( x + 1).
(2) Compute (ln |x|) .
(3) Compute (xx ) . √
3
(4) Compute f (x) if f (x) = x4 x+1
x+4
.

1 2x
(5) Compute limx→∞ 1 + 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
ds s(t + h) − s(t)
(3.14) v(t) = = s (t) = lim
dt h→0 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.
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 (t) =
= s (t).
dt
Example 34. The position of a particle is described by the equation
1
(3.16) s(t) = t3 − 3t2 + 5t.
3
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?
19. APPLICATIONS OF RATES OF CHANGE 59
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 ∈ (1, 5). In other words, the particle is moving
backward between the first and fifth seconds. 2
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 pro-
portion to the number of units produced. These will be represented
by the linear term bx. Then there will be other factors, such as hiring
workers, marketing 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
C(x + Δx) − C(x)
(3.17) lim .
Δx→0 Δx
However, it could well be that the smallest meaningful positive value
of Δx is 1, in case the products are such that fractional units do not
make sense (e.g., automobiles). In that case, Δx → 0 is impossible
in its precise mathematical meaning; the closest that Δx can get to
0 is when Δx = 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 35. 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. 2
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
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 34. 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?
20. RELATED RATES 61
(3) The location of an object moving vertically is described by the

2
function s(t) = t − t5 , for t ∈ [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) Use the formula M (n) = C(n + 1) − C(n) to find the cost of
producing the 101st and 1001st units in Example 35. Com-
pare your results with the estimates that we found using the
function C (x).
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? So 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:
(3.19) 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 pumping 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. The very notion of related

quantities can be stated in proper mathematical terms.
Definition 25. 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 quanti-
ties 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 x (t) or y (t). The problem of “related rates”
can now be cast in the proper mathematical terms: What is the rela-
tionship between the derivatives x (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 (x(t)) for any t. Taking the derivative of both
sides with respect to t by means of the chain rule (Theorem 19), we
obtain a generalization of (3.19):
(3.20) y(t) = f (x(t)) → y (t) = f (x(t))x (t).
Equation (3.20) establishes the sought-after relation between the rates
y and x . However, it seems somewhat different from (3.19): 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 = t0 . Let the values of x and y at t = t0 be x0 =
x(t0 ) and y0 = y(t0 ). The number a = f (x0 ) can be calculated. Then
the equality y (t0 ) = ax (t0 ) determines the relation between the rates
y and x at the instance when x has the value x0 (or y has the value
y0 = f (x0 )).
Example 36. 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 ω = π rad/s?
(II) At what direction of the laser beam does the bright spot travel
at the speed v = 4π m/s?
Solution:
(I) The analysis of any problem on related rates must begin with
defining the quantities whose rates are being studied. In other
20. RELATED RATES 63
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 ϕ 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 ϕ = 0, that is, when the laser beam is perpendic-
ular to the wall. If the pointer rotates, the angle becomes a
function of time, ϕ = ϕ(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 ϕ (t) = ω (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
y and the angle ϕ: y = f (ϕ). 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 ϕ = f (ϕ).
(III) Once the relation between the quantities of interest has been
established, the relation between their rates can be found.
Since (tan ϕ) = 1/ cos2 ϕ, Equation (3.20) yields
D
(3.21) y = D tan ϕ → y = 2
ϕ .
cos ϕ
The first question is answered by setting ϕ = ω = π rad/s
and D = 1 m/s, and y = v is measured in meters per second.
(IV) Note that the rate y = v is not constant even if the rate ϕ = ω
is constant. To answer the second question, one has to find
the value of ϕ when v = 4π m/s, D = 1 m, and ω = π rad/s.
It follows from Equation (3.21) that
Dω 1 π
cos2 ϕ = = → ϕ= ;
v 4 3
that is, the bright spot moves at the speed 4π m/s when the
laser beam makes 60◦ with the perpendicular to the wall. 2
20.4. Can Anything Travel Faster Than Light? The solution (3.21) has
an interesting feature. When ϕ approaches 90◦ , that is, the laser beam
is getting closer to being parallel to the wall, the cosine, cos ϕ, tends to
0 in Equation (3.21), 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 36 is now supplemented by

two additional questions:
(I) Is it possible that v can exceed the speed of light? If so, at
which direction of the laser beam does it happen?
(II) At which position of the bright spot does it happen?
The answers read:
(I) 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 ϕ = Dω/c ≈ 1.05 · 10−8 , and hence ϕ ≈
89.99414◦ . So the bright spot becomes superluminal if ϕ >
89.99414◦ !
(II) Since y = D tan ϕ, 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
ω ≈ π · 103 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 + Δy 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 + Δy (at this position arrived a different lump of light emitted by the
laser at a later time). So the rate Δy/Δt cannot possibly be associated
with the motion of any material object.
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 a more advanced course, where the functions
of several variables are studied. However, the basic idea of finding
relations between the rates has not changed: They are obtained by
21. LINEAR APPROXIMATIONS AND DIFFERENTIALS 65
differentiating the relations between the quantities in question with

respect to t. The procedure is illustrated in the following example.
Example 37. 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
the rate 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 = xy , P = 2(x + y) .
(III) If x = x(t) and y = y(t), then S(t) = x(t)y(t) and P (t) =
2(x(t)+y(t)). Using the derivative of the product and the sum
of two functions, the linear relations between the rates are
S = x y + xy , P = 2(x + 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 x = 3 cm and y = 1 cm, one
has −2 = (3 − 1)y and y = −1 cm/s. It then follows that
x = −y = 1 cm/s. 2
21. Linear Approximations and Differentials

21.1. Tangent Line Approximation. The derivative of a function f (x)
at a point x = x0 defines the slope of the line tangent to the graph
y = f (x) at the point (x0 , f (x0 )) (see Equation (2.5)). The equation
of the tangent line is
y − f (x0 )
= f (x0 ) or y = f (x0 ) + f (x0 )(x − x0 ) .
x − x0
Definition 26. Suppose f (x) is differentiable at x = x0 . The
linear function
(3.22) L(x) = f (x0 ) + f (x0 )(x − x0 )
is called the linearization of f (x) in a neighborhood of x0 .
Since the values of f and L coincide at x = x0 , one might expect
that the difference f (x) − L(x) is small, provided x is close enough to
x0 . So the linear function L(x) may be used to approximate f (x) in a
small neighborhood of x0 , that is, f (x) ≈ L(x). This approximation is

called the linear approximation or tangent line approximation.
√ Example 38. Use the linear approximation to estimate the value

3.92.
Solution:
√
(I) Consider f (x) = x. The closest value of x to 3.92 at which
the square root can be evaluated without a calculator is x0 = 4:
f (x0 ) = 2. Note the two important steps here: the choice of
f (x) suitable for the problem and the choice of x0 near which
the linear approximation
√ is√
to be used.
(II) Since f (x) = ( x) = 1/(2√ x) and f (4) = 1/4, by Equation

(3.22) the linearization of x near x = 4 is

1
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 ≈ L(3.92) = 2 + (3.92 − 4) = 1.98. 2
4
√
√ A calculator gives 3.96 ≈ 1.98997. 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 . This observation can be summarized
by the following inequality:
√
| x − L(x)| < 1.02 · 10−4 if |x − 4| ≤ 0.08.
√
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.
21.2. 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 ε of the linear approximation of a function f (x)
near x0 , find δ such that
|f (x) − L(x)| ≤ ε if |x − x0 | ≤ δ,
or, alternatively, given δ, i.e., the neighborhood x0 − δ ≤ x ≤ x0 + δ,
estimate the error ε of the linear approximation. The following theorem
is useful to answer these questions.
Theorem 23. Suppose a function f (x) is twice differentiable in

(a, b) such that |f (x)| ≤ M for all x ∈ (a, b) and some number M .
Let L(x) be the linearization of f (x) at x0 ∈ (a, b). Then
|f (x) − L(x)| ≤ 12 M (x − x0 )2 , x ∈ (a, b).
This theorem is a simpler version of the Taylor theorem, which is

proved in advanced calculus courses. The following example illustrates
the use of this theorem to assess the accuracy of the linear approxima-
tion.
Example 39. Consider the linearization of sin x at x = 0. Find

an interval |x| ≤ δ in which the error of the linear approximation does
not exceed ε = 0.5 · 10−3 .
Solution:
(I) Since f (x) = (sin x) = cos x, f (0) = 1, and f (0) = 0, the
linearization is L(x) = x.
(II) In Theorem 23, let a = −δ and b = δ. Next, one has to
find M . The simplest way to do this is to take the maximal
value of |f (x)| in the interval |x| ≤ δ. Note that there should
be δ < π/2 because L(π/2) − sin(π/2) = π/2 − 1 exceeds
the given error ε. So sin x is monotonic in |x| ≤ δ, and hence
|(sin x) | = | sin x| ≤ sin δ = M for all |x| ≤ δ. By Theorem 23,
(3.23) | sin x − x| ≤ 12 M x2 ≤ 12 M δ 2 = ε if |x| ≤ δ.
With M = sin δ, the solution of the equation δ 2 sin δ = 2ε =

10−3 determines δ. An analytic solution of this equation is im-
possible. So a value of δ has to be found numerically (actually,
δ ≈ 0.100057).
(III) Otherwise, one can choose a larger M , for example, sin x ≤
1 for any x. So M = 1 is acceptable, too. This simplifies
Equation (3.23): δ 2 = 10−3 and hence δ ≈ 0.0362. This value
of δ appears to be smaller than that in the case M = sin δ. It
follows from Equation (3.23) that a larger value of M leads to
a smaller δ. So this option should not be “abused.” A good
M is not too large and yet is simple enough to solve Equation
(3.23). This requires some skills to achieve.
(IV) A good compromise is to use the inequality sin δ ≤ δ. So
the choice M = δ also fulfills the conditions of Theorem 23.
Equation (3.23) becomes δ 3 = 10−3 and δ = 0.1, which is to

be compared with δ = 0.0362 when M = 1 and δ ≈ 0.100057
when M = sin δ. 2
The converse problem is simpler: Find an upper bound for the error
of the linear approximation of sin x at x = 0 in the interval |x| ≤ 0.2.
One has | sin x − x| ≤ ε = 12 M δ 2 = 0.5 · sin(0.2) · (0.2)2 ≈ 3.9734 · 10−3 .
21.3. Differential. For a real variable x, the differential dx is defined as
an increment of x. It can be given the value of any real number inde-
pendently of the value of x; that is, dx is considered as an independent
variable. So, with every real variable, one can associate another real
variable, called the differential. If two real variables are related, the
following rule postulates the relation between their differentials.
Definition 27. Let two variables y and x be related as y = f (x),
where f is a differentiable function. The differential dy = df (x) is
defined by the linear transformation of dx:
(3.24) dy = df (x) = f (x) dx .
Note that the variables x and dx on the right-hand side are independent
variables. Equation (3.24) states that, if the variables y and x are
related, then the differential dy is no longer an independent variable
and is determined by x and dx; specifically, dy depends linearly on dx.
21.4. Geometrical Significance of the Differential. Put dx = Δx, where
Δx is a real number. Consider an increment of the variable y between
x + Δx and x: Δy = f (x + Δx) − f (x). For any x, the differential
of a function df (x) = dy = f (x) Δx does not generally coincide with
its increment Δy. For example, put f (x) = x2 , x = 1, Δx = 0.2, then
Δy = (1 + 0.2)2 − 1 = 0.44, whereas dy = f (1) Δx = 2 · 0.2 = 0.4.
Since the derivative f (x) determines the slope of the tangent line to
the graph y = f (x), the differential is the increment of y along the
tangent line through the point (x, f (x)) in the interval [x, x + Δx].
Thus, dy = Δy because the tangent line does not generally coincide
with the graph.
An intuitive understanding of the differential stems from its geo-
metrical interpretation. Let Δx tend to 0. The ratio
Δy − dy Δy
= − f (x)
Δx Δx
tends to 0 by the existence of f (x). This means that the difference
Δy − dy must go to 0 faster than Δx. An increment Δx is said to
be infinitesimally small if (Δx)n , n > 1, can always be neglected. So
one might think of differentials as infinitesimal variations of variables.

From this point of view, the definition (3.24) looks rather natural:
Infinitesimal variations of two related variables must be related linearly
as their higher powers can always be neglected. Thus, the concept of
the differential becomes rather practical: to establish relations between
variations of related quantities in situations when these variations may
be viewed as infinitesimal.
21.5. 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: x ± Δx; that is, Δx is known.
Using the measured values of x, the values of y are computed by
y = f (x). What is the accuracy of the values of y? Apparently, x
and Δx are independent variables as the error Δx 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 Δx indepen-
dently of the actual 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! That is, y ± dy, where dy = f (x) dx and dx = Δx.
The differential dy represents an absolute error of y. The quantity dy/y
is called a relative error.
Example 40. What are the absolute and relative errors of the vol-
ume of a cube if its side is 10 ± 0.1 cm?
A typical ruler has a grid spacing of 1 mm, which determines the
length measurement error of 0.1 cm.
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 is dV /V = 0.03 or 3% (note
that dx/x = 0.01, i.e., only 1%). 2
The error analysis for several related quantities is given in a more

advanced course; it is based on the differential of a function of several
variables.
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 v0 . 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 28 (Absolute Maximum and Minimum). A function f
has an absolute maximum at c if f (x) ≤ 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 = 2πn, where n = 0, ±1, ±2, ..., and its minimum value −1
at x = π + 2πn. A function does not always have a maximum or
minimum. For instance, the function f (x) = 1/x defined for all real
x = 0 has neither maximum nor minimum because, for any real M ,
one can always find x such that f (x) > M (0 < x < 1/|M |). So
no real number can be the maximum of f (x). Similarly, for any real
M , f (x) < M if −1/|M | < x < 0; that is, no minimum exists. The
function f (x) = x2 has no maximum 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).
Definition 29 (Local Maximum and Minimum). A function f
has a local (or relative) maximum at c if f (x) ≤ f (c) for all x in
some open interval containing c. Similarly, a function f has a local (or
relative) minimum at c if f (x) ≥ f (c) for all x in some open interval
containing c.
71
72 4. APPLICATIONS OF DIFFERENTIATION
Example 41. Does the function f (x) = x3 − x = x(x2 − 1) have

an absolute maximal (minimal) value and relative maxima (minima)
on the real axis?
Solution:
(I) The function has neither an absolute maximum nor an abso-
lute minimum because it grows unboundedly with increasing
x and it decreases unboundedly as x attains larger negative
values.
(II) 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.
(III) Consider the open interval x ∈ (0, 1). The function is strictly
negative in it and bounded from below: M < f (x) < 0 for
all x ∈ (0, 1) (e.g., M = −3). By increasing M , one can even-
tually 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.
(IV) 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 can find a point c ∈ (−1, 0) such
that f (x) ≤ f (c) for all x ∈ (−1, 0); that is, f has a relative
maximum in (−1, 0). √
Remark. The actual value is c = −1/ 3 (see below). 2
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 24 (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 (c1 ) ≤ f (x) ≤ f (c2 ) for all x in [a, b].
The continuity hypothesis is essential. In fact, the continuity of
f (x) = x3 − x was implicitly used in Example 41 to establish the ex-
istence of its relative maximum and minimum! The following example
22. MINIMUM AND MAXIMUM VALUES 73
illustrates the point. Consider the function f (x) = 2x if x ∈ [0, 1) and

f (x) = 1 if x ∈ [1, 2]. So the function is defined on the closed interval
[0, 2] and bounded from above f (x) < M (e.g., M > 2). An attempt
to establish the existence of a maximal 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 > 0, f (2 − ) < f (x) for x ∈ (2 − , 2)
no matter how small is. Thus, f does not have an absolute 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 x ∈ [0, 1]
and f (x) = 1 when x ∈ (1, 2] has an absolute maximum and mini-
mum, f (0) ≤ f (x) ≤ 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 x ∈ [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 (see Example 41).
The second observation resulting from Example 41 is that at the
point where a continuous function attains its local minimum or max-
imum value there is a horizontal line that touches the graph of this
function. So, if the function 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 maxi-
mum or minimum value.
Theorem 25 (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).
f (c + h) − f (c)
lim = f (c).
h→0 h
Therefore, the right-hand and left-hand limits must coincide with f (c)
(see Section 2):
f (c + h) − f (c) f (c + h) − f (c)
(4.1) lim− = f (c) = lim+ .
h→0 h h→0 h
Let f have a local maximum (the case of a local minimum can be
treated similarly). Then f (c) ≥ f (x) or f (x) − f (c) ≤ 0 in some open
interval a < x < b. In particular, [f (c + h) − f (c)]/h ≤ 0 for any
positive h > 0 such that c < c + h < b. By Theorem 3,

f (c + h) − f (c) f (c + h) − f (c)
(4.2) ≤0 → lim+ ≤ 0.
h h→0 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) f (c + h) − f (c)
(4.3) 0≤ → 0 ≤ lim− .
h h→0 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. 2
This theorem provides a powerful tool to determine the actual po-
sitions of local maxima and minima. Let us go back to Example 41
(f (x) = x√3 − x). The slope f (x) = 3x2 − 1 vanishes at two points
x = ±1/ 3. According to the analysis √ carried out in Example 41,
√ maximum at x = −1/ 3 ∈ (−1, 0) and a local minimum
f has a local
at x = 1/ 3 ∈ (0, 1).
Definition 30. 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) = |x|. 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 29). If f (c) exists, then the
second derivative test can be used, which is discussed later.
22. MINIMUM AND MAXIMUM VALUES 75
(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. 2
Example 42. If a stone is thrown at a speed v0 m/s and an angle θ
with the horizontal line, then its trajectory is a parabola:
g
(4.4) y = x tan θ − x2 2 ,
2v0 cos2 θ
where y is the stone height (vertical position), x 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 v0 ?
Solution:
(I) The range as a function of the angle θ has to be found first.
The stone lands when its height y vanishes. This happens at
x = 0 (naturally, this is where the stone was thrown) and at
x = L(θ), where
2v02 2v 2 v2
L(θ) = tan θ cos2 θ = 0 sin θ cos θ = 0 sin(2θ).
g g g
(II) The range L(θ) is a differentiable function of θ so the values
of θ at which L attains its extreme values may be found from
the equation
v02
L (θ) = 0 → 2 cos(2θ) = 0 → cos(2θ) = 0 .
g
This equation has countably many solutions 2θ = π/2 + πn,
where n is any integer. But in the interval of the physical
values of θ ∈ [0, π/2], it has only one solution θ = π/4. Since
sin(2π/4) = 1 (the absolute maximum of the sine), L attains
its maximal value at θ = π/4; that is, the range is maximal,
Lmax = v02 /g, when a stone is thrown at 45◦ . 2
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, the trajectory would deviate a
bit from the parabola (due friction with the air). So the optimal angle
would deviate a bit from π/4. The deviation would also depend on the
mass and the initial speed. The range optimization problem would be
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 45◦ when the stone is thrown, for
example, from a cliff.
23. The Mean Value Theorem

Theorem 26 (Rolle’s Theorem). Let f be a function that satisfies
the following three hypotheses:
(I) f is continuous on the closed interval [a, b].
(II) f is differentiable on the open interval (a, b).
(III) f (a) = f (b).
Then there is a number c in (a, b) such that f (c) = 0.
This theorem provides a useful method to prove the existence of a
local maximum or minimum of a function f when analytic solutions
of the equation f (x) = 0 are hard to find. In fact, it has already
been used in Example 41: The function f (x) = x3 − x on the inter-
vals [−1, 0], [0, 1], [−1, 1] satisfies the hypotheses of Rolle’s theorem
because f (±1) = f (0) = 0. The proof follows closely the arguments of
Example 41.
Proof of Theorem 26.
(I) If f (x) = f (a) = k is a constant function, then f (x) = 0
everywhere.
(II) Let f (x) > f (a) for some x ∈ (a, b) (cf. Example 41 for
x ∈ [−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 ∈ (a, b). By Fer-
mat’s theorem, f (c) = 0 because f is differentiable in (a, b).
(III) If f (x) < f (a) for some x ∈ (a, b) (cf. Example 41 for x ∈
[−1, 1] or x ∈ [0, 1]), then, by the extreme value theorem, f has
a minimum at c ∈ (a, b), and, by Fermat’s theorem, f (c) = 0.
2
Rolle’s theorem is also useful to analyze the root pattern of a
function.
Example 43. How many real roots does the equation x5 + x3 + x −
1 = 0 have?
23. THE MEAN VALUE THEOREM 77
Solution:
(I) Let f (x) = x5 + x3 − 1. Evidently, f (−1) = −4 < 0 and
f (1) = 2 > 0. By continuity, f has to take all intermediate
values between −4 and 2 (the intermediate value theorem). So
f has at least one root in (−1, 1).
(II) 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. 2
Theorem 27 (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 ∈ (a, b) such that
f (b) − f (a)
(4.5) f (c) = or f (b) − f (a) = f (c)(b − a) .
b−a
The geometrical interpretation of the theorem is simple. 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 (x) 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 27.
(I) Consider the line through the points (a, f (a)) and (b, f (b)). Its
equation is
f (b) − f (a)
(4.6) y = L(x) = f (a) + (x − a) ,
b−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) − (x − a) .
b−a
Its values determine the deviation of the graph y = f (x) from
the secant line y = L(x) on the closed interval [a, b].
(II) 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 (x) and −L(x) (a linear function is con-

tinuous). Second, it is differentiable on (a, b) as the sum of two
differentiable functions:
f (b) − f (a)
(4.8) h (x) = f (x) − , x ∈ (a, b) .
b−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).
(III) By Rolle’s theorem, there is a number c ∈ (a, b) such that
f (b) − f (a)
h (c) = 0 → f (c) = ,
b−a
where Equation (4.8) has been used. 2
Example 44. 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.
Another state trooper approached the driver and issued a speeding ticket,
claiming that the driver exceeded 86 mi/hr. 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 ∈ (0, 0.75) when
s(0.75) − s(0) 65
s (c) = v(c) = = ≈ 86.7 mi/hr.
0.75 − 0 0.75
The speeding ticket is justified. 2
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 45. 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 ∈ (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) .
24. THE FIRST AND SECOND DERIVATIVE TESTS 79
This equation is easy to understand with the help of a mechanical anal-

ogy: How far can a car travel in time b − a if its speed is not lower than
m, but cannot exceed M . 2
The derivative of a constant function vanishes. How about the

converse? The following theorem answers this question.
Theorem 28. 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 (x1 ) − f (x2 ) = f (c)(x1 − x2 ). By hypothesis, f (c) = 0 for any c.
Thus, f (x1 ) − f (x2 ) = 0 or f (x1 ) = f (x2 ) for any x1 and x2 in (a, b);
that is, f is constant. 2
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 (−∞, 0) and (0, ∞) 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 5. If f (x) = g (x) for all x in an interval (a, b), then

f − g is constant, that is, f (x) = 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. 2
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, 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 derivatives 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. Properties of the First Derivative.
Theorem 29 (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 (x1 ) < f (x2 ) and decreasing if
f (x1 ) > 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 (x1 ) = f (c)(x2 − x1 ) .
If f > 0, then it follows from (4.9) that f (x2 ) − f (x1 ) > 0 because, by
assumption, x2 > x1 ; that is, the function is increasing. Similarly, for
f < 0, f (x2 ) − f (x1 ) < 0, and the function is decreasing. 2
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 30 (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 easily be dropped
because all that is really needed to apply the increasing-decreasing test
is the sign of the derivative f (x) for x < c and x > c. For example,
f (x) = |x|. 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/x2
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 number. 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!
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 up-
ward as a function of time t. At the beginning, the stone moves up-
ward so H > 0 (the height is increasing). When the stone comes back
to the ground, it moves downward so H < 0 (the height is decreas-
ing). 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 46 (Example 41 Revisited). Find all local maxima and
minima of f (x) = x3 − x and the intervals on which the function is
increasing or decreasing.
Solution:
(I) Since f is differentiable (it is a polynomial), all its critical
points satisfy the equation
√ √
f (x) = 3x2 − 1 = 3 x − 1/ 3 x + 1/ 3 = 0 .
√ √
Hence, the critical points are c1 = −1/ 3 and c2 = 1/ 3.
(II) For x < c1 , the product (x − c1 )(x − c2 ) is positive (as the
product of two negative numbers), and hence f > 0 (f is
increasing on (−∞, c1 )). For c1 < x < c2 , the product (x −
c1 )(x−c2 ) is negative (as the product of a negative and positive
number), and hence f < 0 (f is decreasing on (c1 , c2 )). For
x > c2 , the product (x − c1 )(x − c2 ) is positive (as the product
of two positive numbers), and hence f > 0 (f is increasing on

(c2 , ∞)).
(III) The derivative changes from positive to negative at c1 . There-
fore, f has a local maximum at c1 . The derivative changes
from negative to positive at c2 . Therefore, f has a local mini-
mum at c2 . 2
24.2. Properties of the Second Derivative: Inflection Points.
Definition 31 (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 (x) > 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:
f (c + h) − f (c)
(4.10) h[f (c + h) − f (c)] > 0 → > 0,
h
where the first inequality has been divided by a positive number h2 .
The latter inequality 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.
Theorem 31 (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.
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) = −x4 . Third, f (c−h) > 0 and f (c+h) < 0, that
is, the concavity changes from upward to downward (e.g., f (x) = −x3 ).
Fourth, f (c − h) < 0 and f (c + h) > 0, that is, the concavity changes
from downward to upward (e.g., f (x) = x3 ).
Definition 32 (Inflection Point). A point P on the graph y =
f (x) is called an inflection point if f is continuous there and the graph
changes from concave upward to concave downward or from concave
downward to concave upward.
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.
Theorem 32 (The Second Derivative Test). Suppose f is contin-
uous 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.
In Example 46, the √function f (x) = x3 − x is shown
√ to have√two

critical points: x = ±1/ 3. Since f (x) = 6x, f (−1/ 3) = −2 3 <
√ √
0 (a local maximum) and f (1/ 3) = 2 3 > 0 (a local minimum).
The function also has an inflection point at x = 0: f (x) = 6x < 0
if x < 0 and f (x) = 6x > 0 if x > 0. Note that an inflection point
may not be related a critical point! In other words, the tangent line
at an inflection point can have any slope. In the previous example,
f (0) = −1.
25. Taylor Polynomials and the Local Behavior of a Function

The tangent line approximation L(x) is the best linear approxima-
tion of f (x) near x = a because L(x) and f (x) have the same rate
of change at a. In the previous section, it was shown that the second
derivative at a provides important information about the behavior of
f (x) near a, namely the concavity. The tangent line L(x) has no con-
cavity as L (x) = 0. The question arises whether there is a systematic
method to improve the accuracy of the tangent line approximation to
capture more essential features of the behavior of f (x) near a (i.e., the
local behavior of f ).
25.1. Taylor Polynomials. The function L(x) is a polynomial of the first

degree. Consider the second-degree polynomial
T2 (x) = f (a) + f (a)(x − a) + c2 (x − a)2 = L(x) + c2 (x − a)2 ,
where c2 is an arbitrary coefficient. This polynomial has the same
features as L(x), that is, T2 (a) = L(a) = f (a) and T2 (a) = L (a) =
f (a) because T2 (x) = f (a) + 2c2 (x − a). So it might provide a better
approximation of f (x) than L(x) near a if the coefficient c2 is chosen
so that T2 (x) has the same concavity as f (x) near a. By the concavity
test, it is then reasonable to assume that T2 (a) = f (a), which yields
2c2 = f (a) or c2 = f (a)/2. The idea can be extended to a polynomial
of degree n:
Tn (x) = c0 + c1 (x − a) + c2 (x − a)2 + · · · + cn (x − a)n ,
where the coefficients are fixed by the conditions
Tn (a) = f (a) , Tn (a) = f (a) , Tn (a) = f (a) , . . . , Tn(n) (a) = cn .
The resulting polynomial is called the nth-degree Taylor polynomial:
f (a) f (n) (a)
Tn (x) = f (a) + f (a)(x − a) + (x − a)2 + · · · + (x − a)n .
2! n!
25. TAYLOR POLYNOMIALS AND THE LOCAL BEHAVIOR 85
25.2. Accuracy of Taylor Polynomials. The accuracy of the tangent line

approximation is assessed in Theorem 23. Let us compare it with
the accuracy of higher-degree Taylor polynomials. Consider Taylor
polynomials of the exponential function ex near x = 0. Since (ex ) = ex
and e0 = 1, the Taylor polynomials are
1 2 1 3 1
f (x) = ex: Tn (x) = 1 + x + x + x + · · · + xn .
2 6 n!
Let us take a few values of x near x = 0 and compare the values of the
Taylor polynomials with the value of the function:
x = 1: f = 2.718 T1 = 2.000 T2 = 2.500 T3 = 2.667

x = −0.5 : f = 0.607 T1 = 0.500 T2 = 0.625 T3 = 0.604
x = 0.25 : f = 1.284 T1 = 1.250 T2 = 1.281 T3 = 1.284
Two observations can be made from this table. First, the accuracy
increases with the degree of the Taylor polynomial (reading the rows
of the table). Second, lower-degree Taylor polynomials become more
accurate as the argument gets closer to the point at which the Taylor
polynomials are constructed (reading the columns of the table). For ex-
ample, the approximation ex ≈ T3 (x) is accurate up to four significant
digits if |x| ≤ 1/4. So the accuracy of the approximation ex ≈ T2 (x)
is determined by the difference T2 − T3 = −x3 /6, that is, by the next
monomial to be added to T2 to get the next Taylor polynomial. This
observation is a characteristic feature of Taylor polynomials:
Theorem 33. Let f be continuously differentiable n+1 times on an

open interval I containing a. Let f (n+1) be bounded on I, |f (n+1) (x)| ≤
M . Then
M
(4.11) |f (x) − Tn (x) ≤ |x − a|n+1 ,
(n + 1)!
where Tn is the Taylor polynomial at a.
Theorem 23 is a particular case of this theorem for n = 1. Inequality

(4.11) is a consequence of the Taylor theorem whose proof goes beyond
the scope of this course. For example, what is the accuracy of the Taylor
polynomial T5 (x) near a = 0 for the exponential ex in the interval
[−1, 1]? To get the upper bound on errors, one should take the maximal
value of the right-hand side of (4.11) for n = 5 in the interval, that is,
(ex )(n) = ex ≤ M = e, and |x| ≤ 1, so errors cannot exceed e/6! ≈
0.0038 (less than 4%).
25.3. Taylor Polynomials near Critical Points. Let a be a critical point

of f . Provided f is enough times differentiable, Taylor polynomials can
be constructed near a. The linear term vanishes because f (a) = 0.
The second derivative test is easy to understand by looking at T2 (x) =
f (a) + f (a)(x − a)2 /2. If f (a) > 0, then f looks like a downward
parabola near a (a local maximum). If f (a) < 0, then f looks like an
upward parabola near a (a local minimum). For example, cos x has a
local maximum at a = 0, and it behaves near a = 0 as cos 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 T3 (x) = f (a) + f (a)(x − a)3 /6. So, if
f (a) = 0, f has an inflection point at a. If f (a) = 0, one should look
at T4 (x) = f (a) + f (4) (a)(x − a)4 /24. A function has a local maximum
(minimum) at a if f (4) (a) > 0 (f (4) (a) < 0). Thus, the local behavior
of f near its critical point is determined by a Taylor polynomial that
has the first nonvanishing correction to f (a).
Example 47. 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 : tan x ≈ T3 (x) = x − x3 /3. Therefore,
f (x) ≈ x − T3 (x) = x3 /3. So there is an inflection point at x = 0. 2
25.4. Asymptotes. Taylor polynomials provide a powerful technique to
investigate the local behavior of a differentiable function. How can one
analyze the behavior of a function near a if it is not differentiable at
a, or not even defined at a, or how does it behave in the asymptotic
regions x → ±∞?
Definition 33 (Vertical Asymptotes). The line x = a is a ver-
tical asymptote of the graph y = f (x) if at least one of the limits
limx→a± f (x) is infinite (∞ or −∞).
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
x(x2 + 3) x(x2 + 3)
(4.12) f (x) = =
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 −∞,
while it tends to ∞ if −1 is approached from the right. Similarly,
f (x) → −∞ as x → 1− and f (x) → ∞ 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?
25. TAYLOR POLYNOMIALS AND THE LOCAL BEHAVIOR 87
Definition 34 (Asymptotic Behavior). The functions f (x) and

g(x) on an open interval x > a (including x > −∞) or x < a (including
x < ∞) are said to have the same asymptotic behavior if
(4.13) lim (f (x) − g(x)) = 0 or lim (f (x) − g(x)) = 0 .
x→a+ x→a−
In particular, if x → ±∞ 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(x) → 0
as x → a± . A practical problem is to find as simple a g as possible
with the property (4.13). In other words, one looks for a simple way
to estimate the values of f (x) near a.
Example 48. Find the asymptotic behavior of the function (4.12).
Solution: The function has to be investigated near x = ±1 and also
when x → ±∞.
(I) 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) ≈ h(−1)/(x +
1) = g(x):
1 x(x2 + 3) 2
f (x) = ≈ = g(x) .
x+1 x−1 x+1
Apparently, limx→−1± (f (x) − g(x)) = 0. The graphs of f (x)
and g(x) = 2/(x + 1) are close near x = −1,
(II) Similarly, near x = 1
1 x(x2 + 3) 2
f (x) = ≈ = g(x) .
x−1 x+1 x−1
(III) When x is a large negative or positive number,
x · x2 (1 + x32 ) x3
f (x) = ≈ = x = g(x) ,
x2 (1 − x12 ) x2
where 1/x3 and 3/x2 are small as compared to 1 for large x2
and can be neglected, that is, limx→±∞ (f (x) − x) = 0. So the
graph of f asymptotically approaches the line y = x. Since
1 + 3/x2 > 1 − 1/x2 , the ratio (1 + 3/x2 )/(1 − 1/x2 ) > 1 for
all x2 > 1, and hence f (x) > x. This means that the graph of
f approaches the slant asymptote y = x from above when x is
a large positive or negative number. 2
The following example illustrates the use of Taylor polynomials to

find the asymptotic behavior.
Example 49. Investigate f (x) = x−8/3 (1 − cos x) near x = 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 T3 =
1 − x2 /2 + x4 /24:
f (x) ≈ x−8/3 (1 − T4 (x)) = x−8/3 ( 12 x2 − 1 4
24
x) = 12 x−2/3 − 1 4/3
24
x .
Therefore, f (x) − 12 x−2/3 ≈ 24 x → 0 as x → 0. Note that the use
1 4/3
of T2 in place of T4 would not be enough to establish the asymptotic

behavior of f . 2
26. L’Hospital’s Rule

If a function f is not defined at a, then its behavior 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.
0 ∞
26.1. Indeterminate Forms 0
and ∞
. Consider the behavior of the fol-
lowing functions:
ex − 1 1 − cos(x) tan(x) − x
(4.14) , 2
, as x → 0 .
x x x3
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) as x → 0+ .
x−1
In general, a limit of the form
f (x)
lim
x→a g(x)
is called an indeterminate form of type 00 if both f (x) → 0 and g(x) → 0

∞
as x → a; it is called an indeterminate form of type ∞ if both f (x) → ∞
(or −∞) and g(x) → 0 (or −∞). The limit itself may or may not
exist. The following theorem provides a powerful method to study the
indeterminate forms of these types.
26. L’HOSPITAL’S RULE 89
Theorem 34 (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) = ±∞ and lim g(x) = ±∞.
x→a x→a
Then
f (x) f (x)
(4.16) lim = lim
x→a g(x) x→a g (x)
if the limit on the right-hand side exists (or is infinite).

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:
f (x)−f (a) f (x)−f (a)
f (x) x−a
limx→a x−a f (a) f (x)
lim = lim = = = lim .
x→a g(x) x→a g(x)−g(a) g(x)−g(a)
limx→a x−a g (a) x→a g (x)
x−a
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 → a± and for the limits at ±∞. The conditions 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) in this
case. 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 un-
derstand via the Taylor polynomials:
f (x) f (a) + f (a)(x − a) + 12 f (a)(x − a)2 + · · ·
≈ .
g(x) g(a) + g (a)(x − a) + 12 g (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 50. Investigate the indeterminate forms (4.14) and (4.15).
Solution:
(I) 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 (ex − 1) ex
lim = lim = lim = 1.
x→0 x x→0 (x) x→0 1
(II) Let f (x) = 1 − cos(x) and g(x) = x2 so that f (0) = g(0) = 0.

Then f (x) = sin(x) and g (x) = 2x. Since f (0) = 0 and
g (0) = 0, l’Hospital’s rule can be applied again:
1 − cos(x) sin(x) (sin(x)) cos(x) 1

lim 2
= lim = lim
= lim = .
x→0 x x→0 2x x→0 (2x) x→0 2 2
(III) 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. Some-
times, simple algebraic transformations 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 sec2 (x) − 1 1 − cos2 (x)

lim = lim = lim 2
x→0 x3 x→0 3x2 x→0 3x cos2 (x)

2
sin2 (x) 1 sin(x) 1
= lim 2
= lim = .
x→0 3x 3 x→0 x 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.
(IV) Let f (x) = ln(x) and g(x) = x−1 so that f (x) → −∞ and
g(x) → ∞ as x → 0+ . So the conditions of l’Hospital’s rule
are fulfilled. Therefore,
ln(x) (ln(x)) x−1
lim+ = lim = lim = − lim+ x = 0.
x→0 x−1 x→0+ (x−1 ) x→0+ −x−2 x→0 2
26. L’HOSPITAL’S RULE 91
26.2. Indeterminate Products 0 · ∞. Suppose that f (x) → ∞ and g(x) →

0 as x → a. How can the indeterminate product f (x)g(x) be inves-
tigated when x → a? It turns out the indeterminate product can be
transformed into one of the indeterminate forms to which l’Hospital’s
rule is applicable:
f ∞ g 0
(4.17) f g = ∞·0→ or f g = ∞·0→ .
1/g ∞ 1/f 0
The function x ln(x) is an indeterminate product of the type 0 · ∞ as
x → 0+ . It can be transformed into an indeterminate form of the type
∞
∞
as in (4.15), which is then resolved by l’Hospital’s rule (see Example
50). Note that, although either of the transformations in (4.17) may be
applied with the subsequent use of l’Hospital’s rule, the technicalities
involved might differ substantially. For instance, if the second option
in (4.17) is applied to x ln(x) = x/(1/ ln(x)), then
x 1
lim+ x ln(x) = lim+ = lim+ = − lim+ x ln2 (x) .
x→0 x→0
1
ln(x)
x→0 − ln21(x) x1 x→0
Although our goal has not been achieved, our effort has not been in
vain. Since the left-hand side vanishes by Example 50, it follows that
x ln2 (x) → 0 as x → 0+ . By repeating this procedure recursively, one
can infer that x lnn (x) → 0 as x → 0+ for any n = 1, 2, ....
26.3. Indeterminate Powers 00 , ∞0 , and 1∞ . Several indeterminate forms

arise from the limits of [f (x)]g(x) as x → a:
00 (f (x) → 0 , g(x) → 0) ; ∞0 (f (x) → ∞ , g(x) → 0);

1∞ (f (x) → 1 , g(x) → 0) .
Note c0 = 1 if c = 0 and c = ∞. Similarly, c∞ = 0 if 0 ≤ c < 1 and

c∞ = ∞ if c > 1. The indeterminate powers can be transformed into
an indeterminate product with the help of the identity y = eln(y) :
g(x) )
lim [f (x)]g(x) = lim eln([f (x)] = lim eg(x) ln(f (x)) = elimx→a g(x) ln(f (x)) .
x→a x→a x→a
The limit of g(x) ln(f (x)) is of type 0 · ∞ and can be treated by the
rule (4.17). The procedure is illustrated with an example of the type
∞0 indeterminate power:
1/x )
lim x1/x = lim eln(x = lim eln(x)/x = elimx→∞ ln(x)/x = e0 = 1 .
x→∞ x→∞ x→∞
26.4. Indeterminate Differences ∞−∞. Suppose f (x) → ∞ and g(x) →

∞ as x → a. The limit of f (x) − g(x) as x → a is called an indeter-
minate difference. The following transformations might be helpful to
investigate it:

g 1 − g/f f f /g − 1
f −g =f 1− = or f − 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 ∞/∞ and can
also be investigated by l’Hospital’s rule. Suppose that f (x)/g(x) → k
as x → a, where k can be either a non-negative number or k = ∞.
If k < 1, then f − g = g(f /g − 1) → ∞ · (k − 1) = −∞; that
is, g increases faster than f as x → a. If k > 1 or k = ∞, then
f − g = g(f /g − 1) → ∞ · (k − 1) = ∞; that is, f increases faster
than g as x → a. For example,
1 1 1
lim+ ln(x) + = lim+ 1 + x ln(x) = lim+ (1 + 0) = ∞.
x→0 x x→0 x x→0 x
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,
1 1 sin(x)
lim − cot(x) = lim 1 − cos(x) = lim = 0,
x→0 sin(x) x→0 sin(x) x→0 cos(x)
where l’Hospital’s rule has been used in the second equality. Note that
Taylor polynomials allow us to find the local behavior of this function
near x = 0. Use T2 to approximate cos(x) and T3 for sin(x):
1 − cos(x) x2 /2 x/2 x
≈ = ≈ ,
sin(x) x − x /6
3 1 − x /6
2 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.
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., | sin(x)| ≤ 1), and
periodic with a period of 2π. In addition, sin(x) has zeros at x = πn,
n = 0, ±1, ±2, ..., while cos(x) vanishes at π/2 + πn. The function
sin(x) is odd, while cos(x) is even. Their ratio tan(x) = sin(x)/ cos(x)
27. ANALYZING THE SHAPE OF A GRAPH 93
is not defined at roots of cos(x). How does tan(x) behave, say, near
x = π/2? Since both sin(x) and cos(x) are smooth near x = π/2, the
behavior of tan(x) near π/2 can be understood with the help of Taylor
polynomials. Let us approximate sin(x) by T1 (x) = 1 + (x − π/2) and
cos(x) by T3 (x) = −(x − π/2) + (x − π/2)3 /6. To simplify the notation,
write Δx = x − π/2 (the deviation of x from π/2). Then
1 + Δx 1 1 + Δx 1 1
tan(x) ≈ =− ≈− =− ,
−Δx + (Δx) /6
3 Δx 1 − (Δx) /6
2 Δx x − π/2
where the second ratio in the product has been approximated by 1
because Δx is small. Since tan(x + π) = 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 xn , the exponential func-
tion ex , and the logarithmic function ln(x) as x → ∞. The exponential
function grows faster than the power function. Let f (x) = ex and
g(x) = xn . Let us analyze the ratio f /g as x → ∞. The conditions
of l’Hospital’s rule are satisfied: ex → ∞ and xn → ∞ as x → ∞.
L’Hospital’s rule can successively be applied until the indeterminate
form is resolved:
ex ex ex ex
lim n = lim = lim = · · · = lim = ∞.
x→∞ x x→∞ nxn−1 x→∞ n(n − 1)xn−2 x→∞ n!
The conclusion is true for any real n. For any real n, there exists
a positive integer N such that n < N or xn < 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) (ln(x)) 1
x 1
lim = lim = lim = lim =0
x→∞ xn x→∞ (xn ) x→∞ nxn−1 x→∞ nxn
for any n > 0 (n may be any positive real number here).

27.2. Asymptotes at x → ±∞. The asymptotic behavior of rational
functions is easily determined by the highest powers of the numerator
and denominator, as in Example 48. In general, if limx→∞ f (x) is
infinite, then the limit of f /g can be studied for trial g’s with different
growth, g = mx (for slant asymptotes), g = xn , g = ln(x), and so on.
Suppose g(x) is found such that f (x)/g(x) → 1 as x → ∞. 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 ∞ − ∞ converges
∞
to 0 as x → ∞, 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 → ∞. But
the limit limx→∞ (f (x) − g(x)) = limx→∞ sin(x) does not exist. So, even
if g is found to have the property f (x)/g(x) → 1 as x → ∞, the
indeterminate form f − g of type ∞ − ∞ 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 guide-
lines 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 x in the domain,
then the graph y = f (x) is symmetric about the origin (or the
rotation through 180◦ 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.
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 = 2π; tan(x), p = π; cos(4x), p = 2π/4 = π/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 limx→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
27. ANALYZING THE SHAPE OF A GRAPH 95
the shape of y = f (x) near the vertical asymptotes and the

asymptotic shape of the graph when x → ±∞.
(V) Critical numbers of f .
Critical numbers are solutions of f (x) = 0 or the values of
x where f (x) does not exist. If, for example, f (x) tends to
∞ (or −∞) as x approaches c, then the line tangent to the
graph y = f (x) at x = c is vertical. For example, f (x) = x1/3
and f (x) = 1/(3x2/3 ). So f (x) diverges as x → 0. The graph
y = x1/3 has a vertical tangent line at x = 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 x = 1. So f is
negative on (−∞, −1), positive on (−1, 0), negative on (0, 1),
and positive on (1, ∞).
(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 numbers and vertical asymptotes.
As a consequence of this study, the nature of each critical point
is established by the first derivative test.
(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 51. 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) There is no vertical asymptote. For large values x → ±∞, one
has f (x) = x(1−6x−2/3 ) ≈ x. So the graph has the asymptotic
behavior y = x. Since 1 − 6x−2/3 < 1 (for large |x|), the graph
approaches the line y = x from below.
(IV) The function is not periodic, and it is neither odd nor even.
(V) The derivative reads
2 x−3
f (x) = .
3 x (x − 6)1/3
2/3
It vanishes at x = 3 and does not exist at x = 0 and x = 6.

The critical points are 0, 3, and 6. In particular, f (x) → ∞ as
x → 0 and it tends to ±∞ as x → 6± , respectively. Therefore,
the graph has vertical tangent lines at x = 0 and x = 6. Near
x = 0, the graph looks like y = f (x) ≈ 62/3 x1/3 , while near
x = 6, it has a downward cusp y = f (x) ≈ 61/3 (x − 6)2/3 .
(VI) The graph lies below the x axis on (−∞, 0) (f > 0) and above
it on (0, ∞). There is no sign change 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 − 3, 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 (−∞, 0), f > 0 (f is increasing)
on (0, 3), f < 0 (f is decreasing) on (3, 6), and f > 0 (f is
increasing) on (6, ∞). Also, f has a local maximum at x = 3
and a local minimum at x = 6 by the first derivative test.
(VIII) The second derivative reads
4 x2 − 3x + 18
f (x) = − .
9 x5/3 (x − 6)4/3
The polynomial x2 − 3x + 18 > 0 for all x because it has no
real root. The factor (x − 6)4/3 cannot be negative, either. The
sign of f is determined only by that of x5/3 . Thus, f > 0 on
(−∞, 0) (the graph is upward concave) and f < 0 on (0, 6)
and (6, ∞) (the graph is downward concave). So x = 0 is the
inflection point. Also, near x = 3, the graph looks like the
28. OPTIMIZATION PROBLEMS 97
downward parabola y = T2 (x) = f (3) + f (3)(x − 3)2 /2 =

3 − 49 (x − 3)2 . 2
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 right calculator buttons. But what a calculator can-
not do is to provide details of the local behavior of a function 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 = 3, 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 as-
ymptotic 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.
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 .
(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 35 (First Derivative Test for Absolute Extreme Values).
Suppose c is a critical point of a continuous function f defined 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 42. 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
42, 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 .
Example 52. 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.
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 vari-
ables 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 = 2πrh + πr2 + πr2 = 2πrh + 2πr2 . The
volume is V = πr2 h. Since the volume is fixed, the variables r
28. OPTIMIZATION PROBLEMS 99
and h are related as h = V /(πr2 ). Hence, S can be written as

a function of the radius r only:
V 2V
S(r) = 2πr 2 + 2πr2 = + 2πr2 .
πr r
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 /(πr2 ).
(III) The function S(r) is differentiable for all r > 0. Therefore, all
its critical points are roots of the derivative:
2V 4π 3 V
S (r) = − 2 + 4πr = 2 r − = 0.
r r 2π
So the critical point is

1/3
V
rc = .
2π
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 V 4V 1/3
r= ≈ 3.6 cm, h = 2 = = 2rc ≈ 7.2 cm.
2π πrc π
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! 2
Remark. In the previous example, S has been expressed as a func-

tion of r. The same conclusion could be reached if S is expressed
as a
function of the height h only, that is, when the relation r = V /(πh)
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 optimiza-

tion 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 spheri-
cal 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.
Example 53. A small store sells jeans at a price of $80 per pair. Ev-
ery 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 $10 per each weekly order increase of 5 units. How large a rebate
should the store offer to maximize its profit?
Solution:
(I) What is known about the price function p(x)? First, its value
at a particular number of sold units x = x0 = 60 is p0 =
p(60) = 80. Also, if x increases by an amount of Δx = 20,
the price function decreases by Δp = 10 (the rebate). Thus,
the ratio m = −Δp/Δx = −1/2 is the rate of change of p(x)
29. NEWTON’S METHOD 101
(the minus sign indicates the decrease in p(x)). So the price

function is
p(x) = p0 + m(x − x0 ) = 80 − 12 (x − 60) = 110 − 21 x .
(II) What is known about the cost function C(x)? First, its value
at a particular number of supplied units x = x0 = 60 is C0 =
C(60) = 2500. Also, the cost function decreases by ΔC = 20
if x increases by Δx = 5. So the ratio M = −ΔC/Δx = −4 is
the rate of change of C or the marginal cost. Therefore,
C(x) = C0 + M (x − x0 ) = 2500 − 4(x − 60) = 2740 − 4x .
(III) One has to maximize the profit function:
P (x) = xp(x) − C(x) = 114x − 12 x2 − 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.
(IV) 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) = $2,300 whereas P (114) = $3,758. 2
Remark. In fact, the linear (tangent line) approximation has been

used to get the unknown price and cost functions in the previous ex-
ample. This is a benefit of market surveys: They estimate the deriva-
tives (or trends) of the price functions. Naturally, an increase in sales
leads to a decrease in the demand for that particular item. So, af-
ter a successful rebate campaign, the store would need a new mar-
ket survey to estimate p (114) and get the linear approximation at
x = 114. The price may go up then. Similarly, the cost function is
generally highly nonlinear. Its linearization near a particular x = x0
cannot be valid for all x > x0 . Indeed, in the previous example, it van-
ishes at x = 685 and becomes negative after that, which cannot possibly
be true.
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.
29.1. Newton’s Recurrence Relation for Finding a Root. Suppose f (x)

has a root near x1 . Consider the tangent line approximation of f near
x1 : L(x) = f (x1 ) + f (x1 )(x − x1 ). It is easy to find the root of L(x),
which is denoted by x2 :
f (x1 )
L(x) = 0 −→ x = x2 = x 1 − .
f (x1 )
Note that the root of L(x) exists if f (x1 ) = 0 (otherwise, the tangent
line is horizontal and cannot have any root). Since L(x) is only an
approximation to f (x), the number x2 is closer to the root of f than
x1 , but does not coincide with it. In other words, the value f (x2 ) is
closer to 0 than f (x1 ): 0 < |f (x2 )| < |f (x1 )| (the absolute value is
necessary if the function takes negative values). Therefore, the tangent
line constructed at x = x2 , L(x) = f (x2 ) − f (x2 )(x − x2 ), can be
expected to approximate f (x) even better near its root because x2 is
closer to the root than x1 . The root of the new tangent line is given
by the same expression as before where x1 should be replaced by x2 :
x3 = x2 − f (x2 )/f (x2 ). The procedure may be recursively repeated to
generate a sequence of values xn :
f (xn )
(4.18) xn+1 = xn − , n = 1, 2, ...,
f (xn )
provided f (xn ) = 0.
Theorem 36. If f has a single root r in an open interval and

f (x) = 0 on the interval, then there exists x1 sufficiently close to r
such that the sequence (4.18) converges to the root
lim xn = r .
n→∞
In practical terms, the sequence elements are calculated with a par-

ticular number of significant digits (decimal places). Newton’s recur-
rence is applied until xn+1 and xn agree to all the relevant decimal
places. Then r = xn+1 is correct to the relevant decimal places.
Example 54. Find the root of f (x) = x − e−x that is correct to six
decimal places.
Solution:
(I) Determine the position of the root first. The graphs y = x
and y = e−x intersect between 0 and 1. So the root lies in the
interval (0, 1).
(II) Verify the condition f (x) = 0: f (x) = 1 + e−x > 0 for all x.
(III) Pick an initial value of x1 = 0. Then Newton’s sequence for
six decimal places is:
x1 = 0 , x2 = 0.5 , x3 = 0.566311 , x4 = 0.567143 , x5 = 0.567143 .
So the root r = 0.567143 is correct to six decimal places (in
fact, f (0.567143) = −4.5 × 10−7 ). 2
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.
(I) A bad choice of the initial point x1 can produce the value of x2
that is a worse approximation to the root than x1 . 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.
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 < x1 < 2 is close enough to either x = 0 or x = 2, then
x2 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 (x1 ) is close to 0.
(II) A poor choice of the initial point may lead to a cycle in New-
ton’s sequence. Take f (x) = x3 − 2x + 2 and x1 = 0. Since
f (x) = 3x2 − 2, the next elements are x2 = 0 − 2/(−2) = 1,
x3 = 1 − 1/1 = 0 = x1 . That is, Newton’s sequence is a cyclic
sequence, which never converges. The initial point must be
taken closer to the root.
(III) If f (x) → ±∞ as x approaches a root r (the graph y = f (x)
has a vertical tangent line at the root), 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)ν , ν > 0,

and a is constant. Since f (x)/f (x) = ν −1 (x − r), Newton’s
sequence (4.18) xn+1 = xn − ν −1 (xn − r) can also be written
as xn+1 − r = q(xn − r), where q = 1 − ν −1 . Apparently,
xn → r is equivalent to yn = xn − r → 0. But the sequence
yn+1 = qyn = q 2 yn−1 = · · · = q n+1 y1 would converge if q =
1 − ν −1 < 1 or ν < 1/2 unless y1 = 0 (i.e., if the root is
accurately guessed!). For example, for f (x) = x1/3 (ν = 1/3),
Newton’s sequence diverges: xn+1 = (1 − 3)xn = −2xn for any
choice of the initial point x1 = 0. For f (x) = |x|1/2 (ν = 1/2),
Newton’s sequence oscillates xn+1 = (1 − 2)xn = −xn .
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 + P x − A , F2 = F1 + F1 x − A , ..., Fk = Fk−1 + Fk−1 x − A.
Here F1 is the future value of the loan after one payment, which is the
loan P plus the monthly interest P x 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 F1 x minus the payment A, and so on. After
n payments,
Fn = Fn−1 (1 + x) − A
= Fn−2 (1 + x)2 − A[(1 + x) + 1] = · · ·
= F0 (1 + x)n − A[(1 + x)n−1 + (1 + x)n−2 + · · · + (1 + x) + 1]
(1 + x)n − 1
= P (1 + x)n − A ,
x
where, in the last equality, the geometric sum formula 1 + q + q 2 + · · · +
q n−1 = (q n − 1)/(q − 1) has been used (it can be proved by dividing the
polynomial q n − 1 by the polynomial q − 1). Since Fn = 0, the monthly
payment is
Px
(4.19) A= .
1 − (1 + x)−n
For example, a loan of $200,000 for 10 years at a fixed annual inter-

est rate of 6% implies 120 monthly payments of $2,220.41. Indeed, in
Equation (4.19), substitute x = 0.06/12 = 0.005, n = 120, and P =
200,000, then A ≈ 2220.41004. The total amount paid after 10 years is
120 × 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 (or to pick one among
a few offered at different dealerships), one has to know 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). 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) = P x(1 + x)n − A(1 + x)n + 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 55. 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:
(I) For the first car, one has to find the root of Equation (4.20) if
A = 217.42, P = 10,000, and n = 5 × 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%.
(II) For the second car, one has to find the root of Equation (4.20)
if A = 311.38, P = 15,000, and n = 5 × 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.
2
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!
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 35. 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 (xn+1 ) = (n+1)xn , it follows
that if f (x) = xn , n = −1, the antiderivative is F (x) = xn+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 5 given at the end of Section 23. 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 5, F and G may only differ by a constant: G(x) =
F (x)+C. Recall that Corollary 5 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 37. 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 f (x) =
xn , n = −1, is F (x) = xn+1 /(n + 1) + C, and for f (x) = 1/x, it is
F (x) = ln |x| + C. This 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 verti-
cally upward with a velocity of v0 is v(t) = v0 − 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) = v0 . Then it begins to slow down (v(t) decreases because of grav-
ity). Eventually, at t = v0 /g, the chalk stops and begins to fall back. If
30. ANTIDERIVATIVES 107
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) = v0 t − gt2 /2 (indeed, h (t) = v0 − gt). What is the phys-
ical significance of the general antiderivative h(t) = C + v0 t − 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 the 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 (x0 , y0 ), then C is fixed: y0 = F (x0 )+C
or C = y0 − F (x0 ). 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 kF . These
properties are easily verified. Indeed, (F + G) = F + G = f + g
and (kF ) = kF = kf , where the linearity of the derivative has been
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. For example,
(sin(x)) = cos(x), (− cos(x)) = sin(x), (ex ) = ex ,
1 1
(tan(x)) = (sec(x))2 , (sin−1 (x)) = √ , (tan−1 (x)) = .
1−x 2 1 + x2
In particular, this table says that the general antiderivative of f (x) =
1/(1 + x2 ) is F (x) = tan−1 (x) + C. The table of derivatives of basic
functions combined with the linearity of antidifferentiation is a good

source of antiderivatives of more complicated functions.
Example 56. Find the general antiderivative of f (x) = e−2x +
cos(4x) + x2 /(1 + x2 ).
Solution:
(I) By the linearity of the antiderivative, it is sufficient to find
antiderivatives of e−2x , cos(4x), and x2 /(1 + x2 ). The gen-
eral antiderivative is obtained by adding a general constant to
the sum of the particular antiderivatives of the previous three
functions.
(II) From (e−2x ) = −2e−2x , it follows that (−e−2x /2) = e−2x .
Hence, an antiderivative of e−2x is −e−2x /2.
(III) Similarly, from (sin(4x)) = 4 cos(4x), it follows that an anti-
derivative of cos(4x) is sin(4x)/4.
(IV) 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
2
= 2
=1− .
1+x 1+x 1 + x2
So its antiderivative is x − tan−1 (x). Thus, the general anti-
derivative reads:
1 1 2
F (x) = − e−2x + sin(4x) + x − tan−1 (x) + C .
2 4
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 con-
stant. Hence, F (x) = 3x2 +C1 . Its general antiderivative reads F (x) =
x3 +C1 x+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: (C1 x + C2 ) = 0. Similarly, if F (x) is a particular function
30. ANTIDERIVATIVES 109
that satisfies the condition F (n) (x) = f (x), then the general antideriv-
ative of the nth order is F (x) + C1 xn−1 + C2 xn−2 + · · · + Cn−1 x + Cn ,
where C1 , ..., Cn are arbitrary constants. Indeed, the nth derivative of
a polynomial of degree n − 1 is 0. Note that this analysis applied only
when f was defined in an interval. Why?
The following example illustrates the significance of arbitrary con-
stants in general higher-order antiderivatives.
Example 57. 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:
(I) 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.
(II) The general second antiderivative of the constant function
−9.8 is h(t) = −9.8t2 /2 + C1 t + C2 , where C1 and C2 are
arbitrary constants.
(III) To fix C1 and C2 , the initial conditions of the motion must
be used. The initial velocity is v(0) = 7. Since v(t) = h (t) =
−9.8t+C1 , one can infer that v(0) = C1 = 7. The initial height
is h(0) = 1.5. Hence, h(0) = C1 = 1.5.
(IV) 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 maximal height
reached by the chalk is 4 m. Why? 2
CHAPTER 5
Integration
31. Areas and Distances

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 x0 = 0, x1 = 1/n, x2 = 2/n,..., xn−1 = (n − 1)/n, and
xn = 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 = S 1 + S 2 + · · · + Sn .
The area Sk cannot exceed the area of a rectangle with base 1/n and
height f (xk ) = (k/n)2 . Let us denote this upper bound by SkU = k 2 /n3 .
The area Sk is greater than the area of a rectangle with base 1/n
and height f (xk−1 ) = (k − 1)2 /n2 . The lower bound is denoted by
SkL = (k − 1)2 /n3 . Thus,
(k − 1)2 k2
SkL = 3
< Sk < 3 = SkU .
n n
Therefore, the area A is bounded above by the sum of SkU and below
by the sum of SkL :
S1L + S2L + · · · + SnL = ALn ≤ A ≤ AUn = S1U + S2U + · · · + SnU
for any number n of partition segments. Let us calculate the difference
2k − 1 2n − 1 2
0 < SkU − SkL = ≤ <
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 AUn −ALn :
2 2
0 < AUn − ALn = (S1U − S1L ) + (S2U − S2L ) + · · · + (SnU − SnL ) < n = .
n2 n
111
112 5. INTEGRATION
Thus, if the limit limn→∞ AUn exists, then lim AUn = lim ALn because
0 < AUn − ALn < 2/n → 0 as n → ∞. On the other hand, ALn ≤ A ≤ AUn
for any n. Taking the limit n → ∞ in this inequality yields
lim ALn = A = lim AUn .
n→∞ n→∞
From a geometrical point of view, when n gets larger, the area AUn
approaches A from above while ALn does so from below. For n large
enough, both AUn and ALn may serve as a good approximation of A.
In fact, the error of either of the approximations does not exceed 2/n
because 0 < AUn − ALn < 2/n and ALn ≤ A ≤ AUn . It appears that the
limit limn→∞ AUn can actually be calculated by means of the formula
for the sum of squares of the first n positive integers:
1 n
(5.1) 12 + 22 + · · · + n2 = n(n + 1)(2n + 1) = (2n2 + 3n + 1).
6 6
Indeed, by making use of this formula, one can infer that
1 2 2n2 + 3n + 1 1 1 1 1
AUn = 3
(1 + 2 2
+ · · · + n 2
) = 2
= + + 2 →
n 6n 3 2n 6n 3
as n → ∞. So the area is A = 3 . 1
Let x∗k be a number in the interval [xk−1 , xk ]. Then the area Sk

can also be approximated by the area Sk∗ of a rectangle with base
1/n and height f (x∗k ) = (x∗k )2 , that is, Sk∗ = f (x∗k )/n. Then the total
area under the graph is approximated by the sum A∗n of all Sk . Since
SkL ≤ Sk∗ ≤ SkU (owing to the monotonicity of the function x2 in each
interval [xk−1 , xk ]), the following inequality holds for any n:
ALn ≤ S1∗ + S2∗ + · · · + Sn∗ = A∗n ≤ AUn .
Taking the limit n → ∞ in this inequality leads to a remarkable result
lim A∗n = A ;
n→∞
that is, the limit of A∗n does not depend on the choice of sample points
x∗k . The area could have been approximated by, for example, A∗n with
the sample points as the midpoints x∗k = (xk + xk−1 )/2, or any other
convenient choice. This analysis can be extended to any continuous
function.
31.1. The Area Under the Graph of a Continuous Function. Let f (x) be
continuous on [a, b]. Consider a partition of [a, b] by n segments of
length Δx = (b − a)/n. The endpoints of the partitions segments are
xk = a + k Δx with k = 0, 1, 2, ..., n, such that x0 = a and xn = b. Let
x∗k be a sample point in the interval [xk−1 , xk ].
31. AREAS AND DISTANCES 113
Definition 36. The area A of the region that lies under the graph
of a continuous function f (x) ≥ 0 on an interval [a, b] is

(5.2) A = lim A∗n = lim f (x∗1 ) Δx + f (x∗2 ) Δx + ··· + f (x∗n ) Δx
n→∞ n→∞
for any choice of sample points x∗k .
Let us assess this definition. Any continuous function attains its

maximal and minimal values on a closed interval. Let Mk and mk be
the maximal and minimal 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 SkL = mk Δx ≤ Sk ≤ SkU = Mk Δx. The area Sk∗ = f (x∗k ) Δx is
a continuous function of x∗k on the interval [xk−1 , xk ]. Therefore, Sk∗
must take all the values between its minimal value SkL and its maximal
value SkU . In particular, Sk∗ = Sk for some x∗k . So, for any n, there
is a choice of sample points such that A∗n = A. Thus, one can always
find a set of sample points for which Equation (5.2) gives the area
under the graph of a continuous function. A nontrivial fact is that
Equation (5.2) holds for any choice of x∗k . Continuing the analogy
with the example of f (x) = x2 , the limit (5.2) is independent of the
choice of sample points if the lower sums ALn = S1L + · · · + SnL and
the upper sums AUn = S1U + · · · + SnU converge to the same number as
n → ∞. Indeed, from the inequality ALn ≤ A ≤ AUn it would then
follow that lim ALn = lim AUn , = A; that is, if ALn and AUn converge
to the same number, then this number must be the area under the
graph. On the other hand, ALn ≤ A∗n ≤ AUn for any choice of x∗k and
for any n. Therefore, limn→∞ A∗n = A independently of the choice of
sample points. It will be shown in the next section that for continuous
functions the limits of ALn and AUn exist and coincide. This fact justifies
the previous definition.
In practice, Equation (5.2) can be used to find the area under the
graph that is correct to any desired number of decimal places. Take a
partition of the interval [a, b], that is, fix some n and Δx. Set sample
points x∗k . Convenient choices might be the left points x∗k = xk−1 , the
right points x∗k = xk , or the midpoints x∗k = (xk−1 + xk )/2. Calculate
the sum A∗n , keeping the desired number of decimal places. Refine
the partition by, for example, doubling the number of segments, and
calculate A∗2n . If A∗n and A∗2n coincide in the desired number of decimal
places, then A = A∗2n is correct to that number of decimal places. If
not, refine the partition further and compute A∗4n and compare it with
A∗2n and so on, until the needed accuracy is reached.
114 5. INTEGRATION
31.2. 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
∗ ∗ ∗ ∗
An = S 1 + S 2 + · · · + S n = Sk∗ ,
k=1

where the index k is called the summation index. The symbol means
∗
adding all Sk , starting with k = 1 up to k = n. For example, the
geometric sum formula can now be written as
n
q n+1 − 1
(5.3) 1 + q + q2 + · · · + qn = qk = .
k=0
q − 1
31.3. The Distance Problem. If an object moves with a constant
velocity v during a time interval a ≤ t ≤ b, then the distance trav-
eled by the object is D = v(b − a). How does one calculate the distance
if the speed is a nonconstant function of time v = v(t) ≥ 0? If D(t)
is the distance as a function of time, then v(t) = D (t). The function
D(t) is an antiderivative of v(t). If it is known, then the distance trav-
eled is D(b) − D(a). It turns out that the problem can also be solved
by finding the area under the graph of v(t)! This implies that there is
a relation between an antiderivative of a function and the area under
its graph.
The time interval [a, b] can be partitioned into n time intervals
Δt = (b − a)/n. Let tk = Δt k, k = 0, 1, ..., n, be the end points of
the partition intervals. The distance ΔDk = 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∗k ) Δt for some t∗k in [tk−1 , tk ].
Recall that the value v(t∗k ) is the average velocity over the time inter-
val [tk−1 , tk ]. Thus, the total distance is D = ΔD1 +· · ·+ΔDn . In a plot
showing the graph of v(t), the distance ΔDk = v(t∗k ) Δt is nothing but
the area under the graph over the interval [tk−1 , tk ]. Recall that one can
always find a particular sample point t∗k such that this area coincides
with the area of a rectangle v(t∗k ) Δt. Therefore, D is the area under the
graph of v(t). This implies, in particular, that in order to calculate D,
any sample points t∗k can be used, not necessarily those at which v coin-
cides with the average velocity in each partition interval. But there is a
price for this generalization. Namely, the limit (5.2) must be computed:
n
D = lim v(t∗k ) Δt ,
n→∞
k=1
where t∗k is any set of sample points.
32. THE DEFINITE INTEGRAL 115
Example 58. 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 Δt = 1/n so that tk = k/n, k = 0, 1, ..., n. Take t∗k =
(k − 1)/n, k = 1, 2, ..., n (the left points of partition intervals). Then
ΔDk = v(t∗k ) Δt = q k−1 /n, where q = e−2/n . The distance traveled is
1 k
n−1
1 qn − 1 1 − e−2
D = lim q = lim = ,
n→∞ n n→∞ n q − 1 limn→∞ n(1 − e−2/n )
k=0
where the sum formula (5.3) has been used. To compute the limit in
the denominator, let x = 1/n, that is, x → 0. The limit becomes the
indeterminate form (1 − e−2x )/x of type 0/0, which can be resolved by
l’Hospital’s rule: (1 − e−2x ) /(x) = 2e−2x /1 → 2 as x → 0. Thus, the
distance traveled is D = (1 − e−2 )/2. 2
Remark. An alternative solution uses the general antiderivative

of e−2t : D(t) = −e−2t /2 + C. The distance traveled is D(1) − D(0) =
(1−e−2 )/2. Note that the result is independent of an arbitrary constant
C. 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
beginning of this section). Its antiderivative is D(t) = t3 /3 + C. So the
distance traveled, or the area under the graph of t2 , is D(1) − D(0) =
1/3. It is so simple, isn’t it? Thus, the concept of the antiderivative
and its relation to the area (5.2) must be further investigated.
32. The Definite Integral

A generalization of the concept of the area under a graph leads to
one of the most fundamental concepts in calculus, the definite integral.
Let f be a function on an interval [a, b]. Consider a partition of [a, b]
by n intervals of length Δx = (b − a)/n so that the endpoints of the
partition intervals are xk = a+k Δx, k = 0, 1, ..., n. Let Ik = [xk−1 , xk ],
k = 1, 2, ..., n, denote the partition intervals. Let Mk = maxIk f (x)
(the maximal value of f (x) in the interval Ik ) and mk = minIk f (x)
(the minimal value of f (x) in the interval Ik ). By the analogy of the
area under a graph, define the lower ALn and upper AUn sums for f by
n n
L U
An = mk Δx , An = Mk Δx
k=1 k=1
for every partition of [a, b].

116 5. INTEGRATION
Definition 37 (The Definite Integral). A function f on an interval

[a, b] is said to be integrable 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
b
f (x) dx = lim ALn = lim AUn ;
a n→∞ n→∞
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 non-negative f on [a, b], the def-
inite integral coincides with the area under the graph of f . The geo-
metrical significance of the definite integral in general will be discussed
later after establishing its basic properties.
Let f be integrable on [a, b] and let x∗k be a sample point in Ik for
each k = 1, 2, ..., n. For any number ∈> 0, there exists an integer N
such that
b
n
∗
f (x) dx − f (x ) Δx <
a k

k=1
for every integer n > N and for every choice of x∗k in Ik . Indeed, since,
for any x∗k , mk ≤ f (x∗k ) ≤ Mk and therefore

n
ALn ≤ f (x∗k ) Δx ≤ AUn .
k=1
By the squeeze principle, the sequence of the sums in the middle of

this inequality must converge to the definite integral of f from a to b
as n → ∞:
n b
∗
(5.4) lim f (xk ) Δx = f (x) dx.
n→∞ a
k=1
Hence, by the definition of the limit, no matter how small is, there is
always a large enough integer N such the deviation of the values of the
sequence elements from the limit does not exceed for all n > N . The
sum in (5.4) is called the Riemann sum after the German mathemati-
cian Bernhard Riemann (1826–1866). It follows from the preceding
analysis that the sequence of Riemann sums for an integrable function
converges to the definite integral. Since the limit is independent of the
choice of sample points, any choice convenient to calculate the limit
(5.4) can be made.
32.1. Continuity and Integrability. The relation (5.4) holds and can be
used to calculate the definite integral, provided the function f is inte-
grable. The question of integrability requires investigating the conver-
gence of the sequences of the upper and lower sums, which might be a
tedious task even for such simple functions as, for example, f (x) = x2 ,
as discussed in the previous section. The following theorem is helpful
when studying the question of integrability.
Theorem 38. If f is continuous on [a, b], or if f has only a finite
number of jump discontinuities, then f is integrable on [a, b]; that is,
b
the definite integral a f (x) dx exists.
This theorem justifies the definition of the area under the graph of
a continuous function introduced in the previous section.
Let f (x) be defined on [0, 1] such that f (x) = 1 if x is a rational
number, and f (x) = 0 otherwise (i.e., if x is irrational). The function is
not continuous anywhere in [0, 1]. 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 δ > 0, the interval
( 12 − δ, 12 + δ) always contains both rational and irrational numbers.
This function gives an example of a nonintegrable function. 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 n for any partition,
ALn = 0, whereas the upper sum is AUn = k=1 Δx = 1, that is,
L U
limn→∞ An = 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∗k to be rational numbers or take x∗k to be irrational numbers;
both options are possible since any partition interval always contains
rational and irrational numbers).
32.2. Properties of the Definite Integral. Suppose f (x) = c, where c is

a constant. In this case, for any partition interval Ik , Mk = mk = c
and AUn = ALn = c nk=1 Δx = cn Δx = c(b − a). In other words, a
constant function is integrable and its integral is c(b − a):
b
(5.5) c dx = c(b − a).
a
For any two integrable functions f (x) and g(x) and constants c1 and
c2 , it follows from the convergence of the Riemann sums (5.4) for f and
118 5. INTEGRATION
g that
b
n
[c1 f (x) + c2 g(x)] dx = lim [c1 f (x∗k ) + c2 g(x∗k )] Δx
a n→∞
k=1

n
n
= c1 lim f (x∗k ) Δx + c2 lim g(x∗k ) Δx
n→∞ n→∞
k=1 k=1
b b
(5.6) = c1 f (x) dx + c2 g(x) dx .
a a
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
Δx changes its sign: Δx = (b − a)/n → (a − b)/n. Therefore,
b a
(5.7) f (x) dx = − f (x) dx
a b
and, in particular,
a
f (x) dx = 0 .
a
It can be proved that

b c b
(5.8) f (x) dx = f (x) dx + f (x) dx
a a c
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 (5.8) 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].
32.3. 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 inte-
gral of g is the area A under
b the graph of g. By the linearity of the
b
integral, a f (x) dx = − a g(x) dx = −A. But the area A coincides
with the area above the graph of f and below the x axis. 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 x axis. Now
let f be continuous 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 (5.8) that
b c b
f (x) dx = f (x) dx + f (x) dx = A1 − A2 ,
a 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].
32.4. Comparison Properties of the Integral. The following additional

properties of the definite integral can be established:
b
(5.9) f (x) dx ≥ 0, if f (x) ≥ 0 in [a, b] ,
a
b b
(5.10) f (x) dx ≥ g(x) dx, if f (x) ≥ g(x) in [a, b] ,
a a
(5.11)
b
m(b − a) ≤ f (x) dx ≤ M (b − a), if m ≤ f (x) ≤ M in [a, b] .
a
The property (5.10) follows directly from the definition: 0 ≤ mk ≤

Mk for any partition if f (x) ≥ 0; that is, the upper and lower sums
are nonnegative and so must be the integral. If f is continuous, the
property (5.9) states the obvious that the area under the graph of f
is nonnegative. The property (5.10) follows from (5.9) for the function
f (x) − g(x) ≥ 0 and the linearity of the integral (5.6). The property
(5.11) is also a consequence of the definition. Indeed, for any partition,
m ≤ mk ≤ Mk ≤ M . Hence, m(b − a) ≤ ALn ≤ AUn ≤ M (b − a) for any
n. In the limit n → ∞, this inequality turns into (5.11).
32.5. Evaluation of the Integral by the Riemann Sum. If the integral

exists (f is integrable), then it can be evaluated as the limit of the
Riemann sum (5.4). The limit is independent of the choice of sample
points. The following choices are often used in practice:
x∗k = xk−1 (the left-point rule),
x∗k = xk (the right-point rule),
x∗k = (xk−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), (5.3), and
n n 2
n(n − 1) 3 n(n − 1)
(5.12) k= , k =
k=1
2 k=1
2
120 5. INTEGRATION
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 59. Find the definite integral of f (x) = e−2x − 2x2 + 4x3
from 0 to 1.
Solution:
(I) The function is continuous on [0, 1] and hence integrable; that
is, Equation (5.4) applies for any choice of x∗k . The left-point
rule will be used.
(II) By the linearity of the integral,
1 1 1 1
−2x
f (x) dx = e dx − 2 2
x dx + 4 x3 dx.
0 0 0 0
−2
The first integral is (1 − e )/2 by Example 58 (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 (5.12). Let Δx = 1/n and xk = (k − 1)/n (the left-point
rule), then the Riemann sum (5.4) becomes
1
1 3
n
3 1 n2 (n − 1)2 1
x dx = lim 4 k = lim 4 = .
0 n→∞ n k=1 n→∞ n 4 4
(III) Thus,
1
1 − e−2 2 1 1 − 6e−2
f (x) dx = − + = .
0 2 3 4 12 2
33. The Fundamental Theorem of Calculus

In this section, the relation between the definite integral of a func-
tion and its antiderivative will be established. This relation provides
a powerful method for calculating the definite integral that avoids the
use of the Riemann sum (5.4).
33.1. Integration and Differentiation. Consider the definite integral of

f (t) = t from 0 to x for some x > 0. This integral represents the area
33. THE FUNDAMENTAL THEOREM OF CALCULUS 121
under the graph of f (t) = t in the interval [0, x], which is the area of a
right triangle: x
x2
A(x) = t dt = .
0 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 (x) = x = f (x) .
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
traveled by the object in time T is given by the area under the graph
of v(t), that is,
T
s(T ) = v(t) dt .
0
On the other hand, the speed is the rate of change of s(T ), and therefore
there should be s (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 39. If f is continuous on [a, b], then the function defined
by x
g(x) = f (t) dt , a ≤ x ≤ b,
a
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
g(x + h) − g(x)
(5.13) lim = f (x)
h→0 h
for a < x < b. The ratio in the limit can be transformed as follows:
x+h x
g(x + h) − g(x) 1
= f (t) dt − f (t) dt
h h a a
x x+h
x
1
= f (t) dt + f (t) dt − f (t) dt
h a x a

1 x+h
= f (t) dt ,
h x
where the property (5.8) 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
122 5. INTEGRATION
the interval [x, x + h] is contained in (a, b). By the continuity of f (t)

on the interval [x, x + h], the function f attains its absolute maximal
and minimal values in [x, x + h]. Let M = f (v) and m = f (u) be
the absolute maximal and minimal values, respectively, where v and
u are in [x, x + h]. Suppose that h > 0. Then m ≤ f (t) ≤ M for
x ≤ t ≤ x + h and, by the property (5.11),
x+h
(5.14) mh = f (u)h ≤ f (t) dt ≤ f (v)h = M h .
x
Since h > 0, by dividing this inequality by h, one can infer that

1 x+h
(5.15) f (u) ≤ f (t) dt ≤ f (v)
h x
for some u and v in [x, x + h]. Inequality (5.15) can be established
for h < 0 in a similar manner. Indeed, inequality (5.14) holds for the
x
integral x+h f (t) dt. After dividing it by −h > 0, inequality (5.15)
is obtained but with the minus sign at the integral. By the property
(5.7), the sign is reversed, yielding (5.15). Thus,
g(x + h) − g(x)
f (u) ≤ ≤ f (v) .
h
Since u and v lie in the interval [x, x + h],
lim f (u) = f (x) , lim f (v) = f (x) .
h→0 h→0
Then the relation (5.13) follows from the squeeze principle:

g(x + h) − g(x)
f (x) = lim f (u) ≤ lim ≤ lim f (v) = f (x) .
h→0 h→0 h h→0
2
This theorem basically states that if a continuous function is first
integrated and then differentiated, then it remains unchanged:
x
d
(5.16) f (t) dt = f (x) , a < x < b.
dx a
x
In other words, F (x) = a f (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 x < 1 and f (t) = 1 if
x ≥ 1. Let a = 0. Then xf has a jump discontinuity at t = 1. By the
property (5.5), g(x) = 0 f (t) dt = 0 if x < 1. For x ≥ 1, one has
x 1 x
g(x) = f (t) dt = f (t) dt+ f (t) dt = 0+(x−1) = x−1 , x ≥ 1 .
0 0 1
33. THE FUNDAMENTAL THEOREM OF CALCULUS 123
Therefore, g (x) = 0 if x < 1 and g (x) = 1 if x > 1. But g (1) does

not exist.
b 2
Example 60. Let g(x) = x e−t dt. Find g (x).
2
Solution: The function e−t is a continuous function everywhere as a
composition of two continuous functions, the exponential
x −t2 and power
functions. By the property (5.7), g(x) = − b e dt. Therefore,
2
g (x) = −e−x by (5.16). 2
This example is a particular case of the general property

b x
d d
f (t) dt = − f (t) dt = −f (x)
dx x dx b
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 40. (The Fundamental Theorem of Calculus). If f is
continuous on [a, b], then
b
f (x) dx = F (b) − F (a) ,
a
where F is any antiderivative of f , that is, a function such that F = f .

x
Proof. Let g(x) = a f (t) dt. By (5.16), 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.
b
Also, by the definition of g(x), g(a) = 0 and g(b) = a 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,
b
F (b) − F (a) = g(b) + C − g(a) + C = g(b) = f (t) dt .
a
The proof is complete. 2

The fundamental theorem of calculus provides a powerful analytic
tool to evaluate definite integrals.
124 5. INTEGRATION
1
Example 61. Evaluate 0
(1 + x2 )−1 dx.
Solution: An antiderivative of (1+x2 )−1 is F (x) = tan−1 (x). Therefore,
1
1 π π
2
dx = tan−1 (1) − tan−1 (0) = − 0 = . 2
0 1+x 4 4
4 √
Example 62. Evaluate 1 (1 + x)/ x dx.
Solution: By the linearity of the integral,
4 4 4 4
1+x −1/2 −1/2
√ dx = (x 1/2
+ x ) dx = x dx + x1/2 dx .
1 x 1 1 1
An antiderivative of xn is xn+1 /(n + 1) for any real n = 1. By taking

n = −1/2 and n = 1/2, an antiderivative is obtained: F (x) = 2x1/2 +
2x3/2 /3. Hence,
1
1+x 16 2 20
√ dx = F (4) − F (1) = 4 + − 2+ = . 2
0 x 3 3 3
34. Indefinite Integrals and the Net Change

As has been shown in the previous section, the derivative of the
definite integral of a continuous function f with respect to the upper
limit equals the value of f at the upper limit. So integration and differ-
entiation appear as operations inverse to one another. To further stress
this relation between the integration and differentiation, the notion of
an indefinite integral is introduced.
Definition 38 (Indefinite Integral). The function F is called an
indefinite integral of f and is denoted by

F (x) = f (x) dx if F (x) = f (x) .
It follows from this definition that an indefinite integral is noth-

ing but the general antiderivative of f . The reason for introducing
the integral symbol into the antiderivative notation is the fundamental
theorem of calculus:
b
f (x) dx = F (b) − F (a) ,
a
where F is any antiderivative of f . Since all antiderivatives differ only
by a constant, which is always cancelled out in the difference F (b) −
F (a), the definite integral is the difference in values of the indefinite
integral at the upper and lower limits of the definite integral. The
34. INDEFINITE INTEGRALS AND THE NET CHANGE 125
indefinite integral has the same properties as the antiderivative. It is

linear:

(5.17) c1 f (x) + c2 g(x) dx = c1 f (x) dx + c2 g(x) dx
for any constant c1 and c2 and any functions f and g.

Using the table of antiderivatives of basic functions, one can make
a table of indefinite integrals of basic functions. Let C be an arbitrary
constant. Then it is easy to verify the following relationships:

xn+1 1
n
x dx = + C , n = 1 , dx = ln(x) + C ,
n+1 x

cos(ax) sin(ax)
sin(ax) dx = − +C, cos(ax) dx = +C,
a a

x x ax
e dx = e + C , ax dx = + C, a > 0 ,
ln(a)

1 −1 1
dx = tan (x) + C , √ dx = sin−1 (x) + C ,
1+x 2
1−x 2

tan(ax) cot(ax)
sec2 (ax) dx = +C, csc2 (ax) dx = − +C,
a a

sec(x) tan(x) dx = sec(x) + C ,

csc(x) cot(x) dx = − csc(x) + C .
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 disjoint union of two or more in-
tervals (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 an interval.
Example 63. Find a general indefinite integral for x−3 .
Solution: The function x−3 is not defined at x = 0. So its domain
is the union of two disjoint intervals (−∞, 0) and (0, ∞). By the first
equality in the preceding table (n = −3),

−3 x−2 x−2
x dx = − + C1 , x > 0; x−3 dx = − + C2 , x < 0,
2 2
where C1 and C2 are arbitrary constants. 2
126 5. INTEGRATION
The following notation is used in the fundamental theorem of calculus:

b b

(5.18) f (x) dx = F (x) = F (b) − F (a) ,
a a

where F (x) = f (x) dx.
1
Example 64. Evaluate 0
[3x2 − x + 4(1 + x2 )−1 ] dx.
Solution: By the linearity of the indefinite integral (5.17), an indefinite
integral of the integrand is x3 − x2 /2 + 4 tan−1 (x). An arbitrary con-
stant in the indefinite integral may be omitted here because, as already
noted, it is always cancelled out in the definite integral. Therefore,
1 1 1
4 x2 −1
3x2 − x + 2
dx = x 3
− + 4 tan (x) = − π,
0 1 + x 2 0 2
where tan−1 (1) = π/4 has been used. 2
34.1. The Net Change Theorem. Put f (x) = F (x) in the fundamental
theorem of calculus (5.18). The result obtained is known as the net
change theorem.
Theorem 41. The integral of a continuous rate of change is the
net change:
b
F (x) dx = F (b) − F (a) .
a
Note that 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 x = b so that its net change vanishes.
An analogy with an object moving along a straight line can be
made to illustrate the net change. Let s(t) be a position function of
the object relative to some point on the line. Then s (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
t2
v(t) dt = s(t2 ) − s(t1 ).
t1
34. INDEFINITE INTEGRALS AND THE NET CHANGE 127
Example 65. 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:
(I) The indefinite integral of v(t) is s(t) = t − t2 + C. So the net
change of the object position is
1 1
v(t) dt = s (t) dt = s(1) − s(0) = 0 .
0 0
(II) 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.
1 1/2 1
|1 − 2t| dt = (1 − 2t) − (1 − 2t) dt
0 0 1/2
= [s(1/2) − s(0)] − [s(1) − s(1/2)] = 1/2,
where the definition |v| = v if v > 0 and |v| = −v if v < 0 has

been used. 2
Other examples of the net change includes the volume V (t) of water
in a reservoir between two moments of time
t2
V (t) dt = V (t2 ) − V (t1 ),
t1
where V (t) is the rate of change of the volume; the net change of the
population growth
t2
n (t) dt = n(t2 ) − n(t1 ),
t1
128 5. INTEGRATION
where n (t) is the growth rate; the relation between the cost and
marginal cost functions:
t2
C (t) dt = C(t2 ) − C(t1 );
t1
and similarly for many other quantities.
35. The Substitution Rule

An indefinite integral of the derivative F (x) is the function F (x)
itself. Let u = F (x), where u is a new variable defined as a differen-
tiable function of x. Consider the differential du = F (x) dx. Then the
following equalities hold:

F (x) dx = F (x) + C = u + C = du ,
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 (x) dx = 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,

1 √ √
√ dx = d 2 x + 1 = 2 x + 1 + C ,
x+1
√
where the substitution u = 2 x + 1 has been used. This trick can be
generalized.
Let F (u) be an indefinite integral of a continuous function f (u) on
an interval I. Let u = 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) .

(x). On
In other words, F (g(x))+C is an indefinite integral of f (g(x))g
an interval, the most general indefinite
integral of f (u) is f (u) du =
F (u) + C. Therefore, F (g(x)) and f (u) du can differ at most by an
additive constant. This proves the following theorem.
Theorem 42. (The Substitution Rule). If u = g(x) is a differen-
tiable function whose range is an interval I and f is continuous on I,
then
(5.19) f (g(x))g (x) dx = f (g(x)) dg(x) = f (u) du.
35. THE SUBSTITUTION RULE 129
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 66. Find x sin(x2 + 1) dx.
Solution:

2 2 1 2 1
x sin(x + 1) dx = sin(x + 1) d(x + 1) = sin(u) du
2 2
1 1
= − cos(u) + C = − cos(x2 + 1) + C,
2 2
2
where the substitution u = x + 1 has been used. 2

Example 67. Find tan(x) dx.
Solution:

sin(x) d(cos(x)) du
tan(x) dx = dx = − =−
cos(x) cos(x) u
= − ln |u| + C = − ln | cos(x)| + C = ln | sec(x)| + C ,
where the substitution u = cos(x) and the logarithm property
ln(1/a) = − ln(a) have been used. 2
The substitution rule can be used to evaluate definite integrals by

means of the fundamental theorem of calculus.
2 2
Example 68. Evaluate 0 xex dx.
Solution: First, find an indefinite integral:

x2 1 x2 2 1 1 1 2
F (x) = xe dx = e dx = eu du = eu + C = ex + C .
2 2 2 2
where u = x2 . By the fundamental theorem of calculus,
1
2 1 2
xex dx = F (2) − F (0) = (e4 − 1).
0 2
Note that, when evaluating the integral, the original variable x 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 12 eu + 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.
130 5. INTEGRATION
Theorem 43. (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
b g(b)

(5.20) f (g(x))g (x) dx = f (u) du.
a g(a)
Proof. Let F be an antiderivative of f . Then F (g(x)) is an an-

tiderivative of (F (g(x))) = F (g(x))g (x) = f (g(x))g (x). By the fun-
damental theorem of calculus,
b b

f (g(x))g (x) dx = F (g(x)) = F (g(b)) − F (g(a)) .
a a
On the other hand, since F (u) is an antiderivative of f (u), the funda-

mental theorem of calculus yields
g(b) g(b)

f (u) du = F (u) = F (g(b)) − F (g(a)) .
g(a) g(a)
Since the right-hand sides of these equalities coincide, so must their

left-hand sides, which implies (5.20). 2
e
Example 69. Evaluate 1 ln(x)/x dx.
Solution: The integrand can be transformed as
ln(x)
dx = ln(x) d ln(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,
e 1
ln(x) u2 1 1 2
dx = u du = = .
1 x 0 2 0 2
35.1. Symmetry. The calculation of a definite integral over a symmet-
ric interval can be simplified if the integrand possesses symmetry prop-
erties.
Theorem 44. Suppose f is continuous on a symmetric interval
[−a, a]. Then
a a
(5.21) f (x) dx = 2 f (x) dx if f (−x) = f (x) (f is even),
−a
a 0
(5.22) f (x) dx = 0 if f (−x) = −f (x) (f is odd).

−a
35. THE SUBSTITUTION RULE 131
Proof. The integral can be split into two integrals:

a 0 a
−a a
f (x) dx = + f (x) dx = − f (x) dx + f (x) dx.
−a −a 0 0 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 = −du. Hence,
−a a
− f (x) dx = f (−u) du
0 0
and
a a a
f (x) dx = f (−u) du + f (x) dx.
−a 0 0
Now, if f is even, then f (−u) = f (u) and (5.21) follows. If f is odd,

then f (−u) = −f (u) and (5.22) follows. 2
The geometrical interpretation of this theorem

a is transparent. Sup-
pose f (x) ≥ 0 for 0 ≤ x ≤ a. The integral 0 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
0 that on [0, a] by a reflection
about the y axis. Therefore, the area −a f (x) dx must coincide with
A. If f is odd, then its graph on [−a, 0] is obtained by the mirror
reflection about
0 the origin so that the area A appears beneath the x
axis. Hence, −a f (x) dx = −A.
π
Example 70. Evaluate −π sin(x3 ) dx.
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, [−π, π]. Thus, by property (5.22),
π
sin(x3 ) dx = 0 .
−π
2
132 5. INTEGRATION
Remark. In the previous example, take a partition of [−π, π] by

points xk = k Δx, k = −N, −N + 1, ..., −1, 0, 1, ..., N − 1, N , where
Δx = π/N . Consider the Riemann sum with sample points being the
midpoints. It is then straightforward to show that the Riemann sum
vanishes because sin(x∗3 ∗3
−k ) = − sin(xk ) for k = 1, 2, ..., N .

Calc 1

Uploaded by

Copyright:

Available Formats

Calc 1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Calc 1

Uploaded by

Copyright:

Available Formats

Concepts in Calculus I

UNIVERSITY PRESS OF FLORIDA

Orange Grove Texts Plus

Miklos Bona and Sergei Shabanov

University Press of Florida

This work is licensed under a modiﬁed Creative Commons Attribution-Noncommercial-No

University Press of Florida

29. Newton’s Method 101

in which a = 0. If a = 1/n, where n is a positive integer, then the

2.2. Polynomials. A polynomial function is the sum of a ﬁnite number

2.3. Rational Functions. A rational function is the ratio of two polyno-

2.4. Trigonometric Functions. Periodicity. The reader has surely en-

2.6. Transcendental Functions. Functions that are not algebraic are

in the positive direction. Indeed, we must substitute a larger number

(f −g)(x) = f (x)−g(x), and (f g)(x) = f (x)g(x). We have to be just a

4. Viewing the Graphs of Functions

Solution: We have f (x) = x2 + 1 = y. So x2 = y − 1, and because we

holds, and for all y > 1, the identity (f ◦ f −1 )(y) = ( y − 1)2 + 1 =

6. The Velocity Problem and the Tangent Problem

Then Equation (1.1) yields

6.2. The Tangent Problem. The problem of ﬁnding the instantaneous

Limits and Derivatives

7. The Limit of a Function

So, if f is the starting example of this section, then limx→0 f (x) = 1.

Then the limit of g at a = 0 does not exist. Indeed, no matter how

Second, if limx→0 f (x) exists, it is unique; that is, f cannot have

Sometimes it can happen that h is not even deﬁned in a, but

7.2. The Precise Deﬁnition of Limits. It is time for us to give a precise

Definition 5. Let f be a function deﬁned on some open interval

Indeed, if we choose x close to 0 but less than 0, then g(x) = 0, so g(x)

Indeed, if we choose x close to 0 but more than 0, then g(x) = 1, so

as x approaches 0 and deciding if the limits limx→0 h(x), limx→0− h(x),

Definition 8. Let f : R → R be a function. We say that the limit

The fact that the limit of f in a is ∞ is denoted by

Example 8. Let f (x) = 1/x2 . Then limx→0 f (x) = ∞.

Similarly, if g(x) = −1/x4 , then limx→0 g(x) = −∞. Note that

7.4.1 The Precise Definition of Infinite Limits. The formal definition

Applying Equation (2.1) to the identity function f (x) = x yields

It turns out (though it is not obvious) that in Equation (2.1) the

(Here f (x) has to be nonnegative if n is even.) So, in particular, if

Proof. If R(x) = p(x)/q(x), where p and q are polynomials, then

Proof. If limx→a g(x) exists, then by applying Theorem 3 to f and

The squeeze principle is very useful since it allows us to compute

Example 11. Let g(x) = x cos(log x). Then limx→0 g(x) = 0.

Solution: Indeed, let f (x) = −x and h(x) = x. Then, since cos(log z)

Definition 11. A function f is called continuous at a if the equality

There are many classes of functions that are continuous at every

(I) Rational functions

Example 14. It follows from successive

gap in the graph of the function, contradicting the requirement that

10. Limits at Inﬁnity

Solution: If we want the value of f (x) to be closer than to 0, all we

Example 19. Let f (x) = 1/x2 . Then

Solution: If we want to get f (x) closer√

10.1.1. The Formal Definition of Limits at Infinity. The formal defini-

The formal deﬁnition of limits at negative inﬁnity is analogous. The

Definition 17. Let f : R → R be a function deﬁned on some

10.1.2. The Graphical Meaning of a Finite Limit at Inﬁnity. If a func-

Definition 18. Let f : R → R be a function deﬁned on some