Multivariable Calculus

George Cain & James Herod

(c)Copyright 1996, 1997 by George Cain & James Herod

All rights reserved
 Table of Contents

Chapter One - Euclidean Three Space

1.1 Introduction
1.2 Coordinates in Three-Space
1.3 Some Geometry
1.4 Some More Geometry--Level Sets

Chapter Two - Vectors--Algebra and Geometry

2.1 Vectors
2.2 Scalar Product
2.3 Vector Product

Chapter Three - Vector Functions

3.1 Relations and Functions
3.2 Vector Functions
3.3 Limits and Continuity

Chapter Four - Derivatives

4.1 Derivatives
4.2 Geometry of Space Curves--Curvature
4.3 Geometry of Space Curves--Torsion
4.4 Motion

Chapter Five - More Dimensions

5.1 The space Rn
5.2 Functions

Chapter Six - Linear Functions and Matrices

6.1 Matrices
6.2 Matrix Algebra

Chapter Seven - Continuity, Derivatives, and All That

7.1 Limits and Continuity
7.2 Derivatives
7.3 The Chain Rule

Chapter Eight - f:Rn-› R

8.1 Introduction
8.2 The Directional Derivative
8.3 Surface Normals
8.4 Maxima and Minima
8.5 Least Squares
8.6 More Maxima and Minima
8.7 Even More Maxima and Minima

Chapter Nine - The Taylor Polynomial

9.1 Introduction
9.2 The Taylor Polynomial
9.3 Error
Supplementary material for Taylor polynomial in several variables.
Chapter Ten - Sequences, Series, and All That
10.1 Introduction
10.2 Sequences
10.3 Series
10.4 More Series
10.5 Even More Series
10.6 A Final Remark

Chapter Eleven - Taylor Series

11.1 Power Series
11.2 Limit of a Power Series
11.3 Taylor Series

Chapter Twelve - Integration

12.1 Introduction
12.2 Two Dimensions

Chapter Thirteen - More Integration

13.1 Some Applications
13.2 Polar Coordinates
13.3 Three Dimensions

Chapter Fourteen - One Dimension Again

14.1 Scalar Line Integrals
14.2 Vector Line Integrals
14.3 Path Independence

Chapter Fifteen - Surfaces Revisited

15.1 Vector Description of Surfaces
15.2 Integration

Chapter Sixteen - Integrating Vector Functions

16.1 Introduction
16.2 Flux

Chapter Seventeen - Gauss and Green

17.1 Gauss's Theorem
17.2 Green's Theorem
17.3 A Pleasing Application

Chapter Eighteen - Stokes

18.1 Stokes's Theorem
18.2 Path Independence Revisited

Chapter Ninteen - Some Physics

19.1 Fluid Mechanics
19.2 Electrostatics

George Cain & James Herod

School of Mathematics
Georgia Institute of Technology
Atlanta, Georgia 30332-0160
 Chapter One

Euclidean Three-Space

1.1 Introduction.

Let us briefly review the way in which we established a correspondence between

the real numbers and the points on a line, and between ordered pairs of real numbers and
the points in a plane. First, the line. We choose a point on a line and call it the origin.
We choose one direction from the origin and call it the positive direction. The opposite
direction, not surprisingly, is called the negative direction. In a picture, we generally
indicate the positive direction with an arrow or a plus sign:

Now we associate with each real number r a point on the line. First choose some
unit of measurement on the line. For r > 0, associate with r the point on the line that is a
distance r units from the origin in the positive direction. For r < 0, associate with r the
point on the line that is a distance r units from the origin in the negative direction. The
number 0 is associated with the origin. A moments reflection should convince you that
this procedure establishes a so-called one-to-one correspondence between the real
numbers and the points on a line. In other words, a real number determines exactly one
point on a line, and, conversely, a point on the line determines exactly one real number.
This line is called a real line.
Next we establish a one-to-one correspondence between ordered pairs of real
numbers and points in a plane. Take a real line, called the first axis, and construct another
real line, called the second axis, perpendicular to it and passing through the origin of the
first axis. Choose this point as the origin for the second axis. Now suppose we have an
ordered pair ( x1 , x2 ) of reals. The point in the plane associated with this ordered pair is

1.1
found by constructing a line parallel to the second axis through the point on the first axis
corresponding to the real number x 1 , and constructing a line parallel to the first axis

through the point on the second axis corresponding to the real number x 2 . The point at

which these two lines intersect is the point associated with the ordered pair ( x1 , x2 ) . A

moments reflection here will convince you that there is exactly one point in the plane thus
associated with an ordered pair (a, b), and each point in the plane is the point associated
with some ordered pair (a, b):

It is traditional to assume the point of view we have taken in this picture, in which
the first axis is horizontal, the second axis is vertical, the positive direction on the first
axis is to the right, and the positive direction on the second axis is up. We thus usually
speak of the horizontal axis and the vertical axis, rather than the first axis and the second
axis. We also frequently abuse the language by speaking of a point ( x1 , x2 ) when, of

course, we actually mean the point associated with the ordered pair ( x1 , x2 ) . The

numbers x 1 and x 2 are called the coordinates of the point- x 1 is the first coordinate and

x 2 is the second coordinate.

Given any collection of ordered pairs( A collection of ordered pairs is called a

relation.), a picture of the collection is obtained by simply looking at the set of points in
the plane corresponding to the pairs in the given collection. Suppose we have an equation

1.2
involving two variables, say x and y. Then this equation defines a collection of ordered
pairs of numbers, namely all ( x , y ) that satisfy the equation. The corresponding picture

in the plane is called the graph of the equation. For example, consider the equation

y 2 = x 4 . Let’s take a look at the graph of this equation. A little algebra (very little,

actually), convinces us that

{( x , y ): y 2 = x 4 } = {( x, y): y = x 2 } ∪ {(x , y ): y = − x 2 } ,

and we remember from the sixth grade that each of the sets on the right hand side of this
equation is a parabola:

What do we do with all this? These constructions are, of course, the bases of
analytic geometry, in which we join the subjects of algebra and geometry, to the benefit of
both. A geometric figure (a subset of the plane ) corresponds to a collection of ordered
pairs of real numbers. Algebraic facts about the collection of ordered pairs of real are
reflected by geometric facts about the subset of the plane, and, conversely, geometric

1.3
facts about the plane subset are reflected by algebraic facts about the collection of pairs of
reals.

Exercises

Draw a picture of the given relation:

1. R = {( x , y ):0 ≤ x ≤ 1, and 1 ≤ y ≤ 4}

2. R = {(x , y ):− 4 ≤ x ≤ 4, and − x 2 ≤ y ≤ x 2 }

3. R = {(x , y ):1 ≤ y ≤ 2} ∩ {( x , y ): y ≥ x 2 }

4. S = {( x , y ): x 2 + y 2 = 1, and x ≥ 0}

5. S = {( x , y ): x 2 + y 2 ≤ 1} ∩ {( x, y): y ≤ x 2 }

6. E = {(r , s):| r| = | s|}

7. T = {(u, v ):| u|+ | v | = 1}

8. R = {( u, v ):|u|+ | v| ≤ 1}

9. T = {( x, y ): x 2 = y 2 }

10. A = {( x, y): x 2 ≤ y 2 }

1.4
11. G = {( s, t ): max {| s|,| t |} = 1}

12. B = {( s, t ): max {| s|,| t |} ≤ 1}

1.2 Coordinates in Three-Space

Now let’s see what’s doing in three dimensions. We shall associate with each
ordered triple of real numbers a point in three space. We continue from where we left off
in the previous section. Start with the plane constructed in the previous section, and
construct a line perpendicular to both the first and second axes, and passing through the
origin. This is the third, axis. Now we must be careful about which direction on this
third axis is chosen as the positive direction; it makes a difference. The positive direction
is chosen to be the direction in which a right-hand threaded bolt would advance if the
positive first axis is rotated to the positive second axis:

We now see how to define a one-to-one correspondence between ordered triples

of real numbers ( x1 , x2 , x3 ) and the points in space. The association is a simple extension

of the way in which we established a correspondence between ordered pair and points in

1.5
a plane. Here’s what we do. Construct a plane perpendicular to the first axis through the
point x 1 , a plane perpendicular to the second axis through x 2 , and a plane perpendicular

to the third axis through x 3 . The point at which these three planes intersect is the point

associated with the ordered triple ( x1 , x2 , x3 ) . Some meditation on this construction

should convince you that this procedure establishes a one-to-one correspondence between
ordered triples of reals and points in space. As in the two dimensional, or plane, case, x 1

is called the first coordinate of the point, x 2 is called the second coordinate of the point,

and x 3 is called the third coordinate of the point. Again, the point corresponding to

(0,00
, ) is called the origin, and we speak of the point ( x1 , x2 , x3 ) , when we actually mean

the point which corresponds to this ordered triple.

The three axes so defined is called a coordinate system for three space, and the
three numbers x, y, and z , where ( x , y , z) is the triple corresponding to the point P, are

called the coordinates of P. The coordinate axes are sometimes given labels-most
commonly, perhaps, the first axis is called the x axis, the second axis is called the y axis,
and the third axis is called the z axis.

1.6
1.3 Some Geometry

Suppose P and Q are two points, and suppose space is endowed with a
coordinate system such that P = ( x , y, z) and Q = ( u, v , w) . How do we find the distance

between P and Q ? This simple enough; look at the picture:

We can see that d 2 = h 2 + (z − w) 2 and h 2 = ( x − u) 2 + ( y − v) 2 . Thus we have

d 2 = ( x − u) 2 + ( y − v ) 2 + ( z − w) 2 , or

d = ( x − u) 2 + ( y − v ) 2 + ( z − w) 2 .

We saw that in the plane an equation in two variables defines in a natural way a collection
of ordered pairs of numbers. The analogous situation obtains in three-space: an equation
in three variables defines a collection of ordered triples. We thus speak of the collection
of triples ( x , y , z) which satisfy the equation

x 2 + y2 + z 2 = 1

1.7
The collection of all such points is the graph of the equation. In this example, it is easy
to see that the graph is precisely the set of all points at a distance of 1 from the origin-a
sphere of radius 1 and center at the origin.
The graph of the equation x = 0 is simply the set of all points with first
coordinate 0, and this is clearly the plane determined by the second axis and the third axis,
or the y axis and the z axis. When the axes are labeled x, y, and z, this is known as the yz
plane. . Similarly, the plane y = 0 is the xz plane, and z = 0 is the xy plane. These

special planes are also called the coordinate planes.

More often than not, it is difficult to see exactly what a graph of equation looks
like, and even more difficult for most of us to draw it. Computers can help, but they
usually draw rather poor pictures whose main application is in stimulating your own
imagination sufficiently to allow you to see the graph in your mind’s eye. An example:

This picture was drawn using Maple.

Let’s look at a more complicated example. What does the graph of

x 2 + y2 − z 2 = 1

look like? We’ll go after a picture of this one by slicing the graph with the coordinate
planes. First, let’s slice through it with the plane z = 0 ; then we see

1.8
 x 2 + y2 = 1 ,

a circle of radius 1 centered at the origin. Next, let’s slice with the plane y = 0 . Here we

see x 2 − z 2 = 1 , a hyperbola:

We, of course, see the same hyperbola when we slice the graph with the plane
x = 0 . What the graph looks like should be fairly clear by now:

1.9
This graph has a name; it is called a hyperboloid.

Exercises

13. Describe the set of points S = {( x , y , z): x ≥ 0, y ≥ 0, and z ≥ 0} .

14. Describe the following sets

a) S = {( x , y , z): z ≥ 0} b) S = {( x , y , z): x ≥ 5}

c) R = {(x , y , z): x 2 + y 2 ≤ 1} d) T = {(r, s, t ): r 2 + s2 + t 2 ≤ 4}

15. Let G be the graph of the equation x 2 + 4 y 2 + 9 z 2 = 36 .

a)Sketch the graphs of the curves sliced from G by the coordinate planes x = 0 ,
y = 0 , and z = 0 .

b)Sketch G. (This graph is called an ellipsoid.)

16. Let G be the graph of the equation x 2 − 3y 2 + 4 z 2 = 12 .

a)Sketch the graphs of the curves sliced from G by the coordinate planes x = 0 ,
y = 0 , and z = 0 .

b)Sketch G. (Does this set look at all familiar to you?.)

1.4 Some More Geometry-Level Sets

The curves that result from slicing the graphs with the coordinate planes are
special cases of what are called level sets of a set. Specifically, if S is a set, the
intersection of S with a plane z = constant is called a level set. In case the level set is a

1.10
curve, it is frequently called a level curve. (The slices by planes x = constant, or y =
constant are also level sets.) A family of level sets can provide a nice stimulant to your
powers of visualization. Everyday examples of the use of level sets to describe a set are
contour maps, in which the contours are, of course, just level curves ; and weather maps,
in which, for instance, the isoclines on a 500mb chart are simply level curves for the
500mb surface. Let’s illustrate with an example.
Let S be the graph of

z 2 − y2 − x2 = 1

Now we look at the level set z = c :

c 2 − y 2 − x 2 = 1 , or

x 2 + y2 = c2 − 1 .

Notice first that we have the same curve for z = c and z = -c. The graph is symmetric
about the plane z = 0. We shall thus look at just that part of the graph that is above the
xy plane.

It is clear that these curves are concentric circles of radius c 2 − 1 centered at the
origin. There are no level sets for | c| < 1, and for c = 1 or -1, the level set is a single point,

the origin.

1.11
 Next, slice with the planes x = 0 and y = 0 to get a better idea of what this thing
looks like. For x = 0, we see

z 2 − y2 = 1 ,

a hyperbola:

The slice by y = 0, of course, is the same. It is rather easy to visualize this graph. Here is
a Maple drawn picture:

1.12
This also is called a hyperboloid. This is a hyperboloid of two sheets, while the
previously described hyperboloid is a hyperboloid of one sheet.

Exercises

17. Let S = {( x 1 , x 2 , x 3 ):| x 1 |+| x 2 |+ | x 3 | = 1} .

a)Sketch the coordinate plane slices of S.

b)Sketch the set S.

18. Let C be the graph of the equation z 2 = 4(x 2 + y 2 ) .

a)Sketch some level sets z = c.

b)Sketch the slices by the planes x = 0 and y = 0.
c)Sketch C. What does the man on the street call this set?

19. Using level sets, coordinate plane slices, and whatever, describe the graph of the

equation z = x 2 + y 2 . (This one has a name also; it is a paraboloid.).

1.13
20. Using level sets, coordinate plane slices, and whatever, describe the graph of the

equation z = x 2 − y 2 .

1.14
 Chapter Two

Vectors-Algebra and Geometry

2.1 Vectors

A directed line segment in space is a line segment together with a direction. Thus
the directed line segment from the point P to the point Q is different from the directed
line segment from Q to P. We frequently denote the direction of a segment by drawing an
arrow head on it pointing in its direction and thus think of a directed segment as a spear.
We say that two segments have the same direction if they are parallel and their directions
are the same:

Here the segments L1 and L2 have the same direction. We define two directed segments L
and M to be equivalent ( L ≅ M ) if they have the same direction and have the same
length. An equivalence class containing a segment L is the set of all directed segments
equivalent with L. Convince yourself every segment in an equivalence class is equivalent
with every other segment in that class, and two different equivalence classes must be
disjoint. These equivalence classes of directed line segments are called vectors. The
members of a vector v are called representatives of v. Given a directed segment u, the
vector which contains u is called the vector determined by u. The length, or magnitude,
of a vector v is defined to be the common length of the representatives of v. It is generally
designated by |v|. The angle between two vectors u and v is simply the angle between the
directions of representatives of u and v.

2.1
 Vectors are just the right mathematical objects to describe certain concepts in
physics. Velocity provides a ready example. Saying the car is traveling 50 miles/hour
doesn’t tell the whole story; you must specify in what direction the car is moving. Thus
velocity is a vector-it has both magnitude and direction. Such physical concepts abound:
force, displacement, acceleration, etc. The real numbers (or sometimes, the complex
numbers) are frequently called scalars in order to distinguish them from vectors.
We now introduce an arithmetic, or algebra, of vectors. First, we define what we
mean by the sum of two vectors u and v. Choose a spear u from u and a spear v from v.
Place the tail of v at the nose of u. The vector which contains the directed segment from
the tail of u to the nose of v is defined to be u + v , the sum of u and v. An easy
consequence of elementary geometry is the fact that | u + v | < | u | + | v |. Look at the
picture and convince yourself that the it does not matter which u spear or v spear you
choose, and that u + v = v + u :

Convince yourself also that addition is associative: u + (v + w) = (u + v) + w.

Since it does not matter where the parentheses occur, it is traditional to omit them and
write simply u + v + w.
Subtraction is defined as the inverse operation of addition. Thus the difference u-v
of two vectors is defined to be the vector you add to v to get u. In pictures, if u is a
representative of u and v is a representative of v, and we put the tails of u and v together,
the directed segment from the nose of v to the nose of u is a representative of u - v:

2.2
 Now, what are we to make of u - u ? We define a special vector with 0 length,
called the zero vector and denoted 0. We may think of 0 as the collection of all degenerate
line segments, or points. Note that the zero vector is special in that it has no direction (If
you are going 0 miles/hour, the direction is not important!). To make our algebra of
vectors nice , we make the zero vector behave as it should:
u - u = 0 and u + 0 = u
for all vectors u.
Next we define the product of a scalar r (i.e., real number) with a vector u. The

product ru is defined to be the vector with length |r||u| and direction the same as the

direction of u if r > 0, and direction opposite the direction of u if r < 0. Convince

yourself that all the following nice properties of this multiplication hold:

(r + s)u = ru + su ,

r ( u + v ) = ru + rv.
0u = 0, and
u + (-1)v = u - v.

It is then perfectly safe to write - u to stand for (-1)u.

Our next move is to define a one-to-one correspondence between vectors and
points in space (This will, of course, also establish a one-to-one correspondence between
vectors and ordered triples of real numbers.). The correspondence is quite easy; simply
take a representative of the vector u and place its tail at the origin. The point at which is

2.3
found the nose of this representative is the point associated with u. We handle the vector
with no representatives by associating the origin with the zero vector. The fact that the
point with coordinates (a, b, c) is associated with the vector u in this manner is
shorthandedly indicated by writing u = (a, b, c). Strictly speaking this equation makes
no sense; an equivalence class of directed line segments cannot possible be the same as a
triple of real numbers, but this shorthand is usually clear and saves a lot of verbiage (The
numbers a, b, and c are called the coordinates, or components, of u.). Thus we frequently
do not distinguish between points and vectors and indiscriminately speak of a vector
(a,b,c) or of a point u.
Suppose u = (a, b, c) and v = (x, y, z). Unleash your vast knowledge of
elementary geometry and convince yourself of the truth of the following statements:

|u| = a 2 + b 2 + c2 ,
u + v = (a + x, b + y, c + d),
u - v = (a - x, b - y, c - d), and
ru = (ra, rb, rc).

Let i be the vector corresponding to the point (1, 0, 0); let j be the vector
corresponding to (0, 1, 0); and let k be the vector corresponding to (0, 0, 1). Any vector
u can now be expressed as a linear combination of these special so-called coordinate
vectors:
u = ( x , y , z) = xi + y j + zk .

Example

Let’s use our new-found knowledge of vectors to find where the medians of a
triangle intersect. Look at the picture:

2.4
We shall find scalars s and t so that
b b−a
a + t ( − a ) = s (a + ).
2 2
Tidying this up gives us
s s t
(1 − t − ) a = ( − ) b .
2 2 2
This means that we must have
s t
− = 0, and
2 2
s
1 − t − = 0.
2

Otherwise, a and b would be nonzero scalar multiples of one another, which would mean
they have the same direction. It follows that
2
s=t= .
3
This is, no doubt, the result you remember from Mrs. Turner’s high school geometry
class.

Exercises

2.5
1. Find the vector such that if its tail is at the point ( x1 , y1 , z1 ) its nose will be at the

point ( x 2 , y 2 , z 2 ) .

2. Find the midpoint of the line segment joining the points (1, 5, 9) and (-3, 2, 3).

3. What is the distance between the points ( x1 , y1 , z1 ) and ( x 2 , y 2 , z 2 ) ?

4. Describe the set of points L = {t i: − ∞ < t < ∞} .

5. Let u = (2,3,8). Describe the set of points L = {t u: - ∞ < t < ∞} .

6. Describe the set of points M = {3k + t i: − ∞ < t < ∞} .

7. Let u = (2,3,8).and v = (1,5,7). Describe the set of points

M = {v + t u: − ∞ < t < ∞ } .

8. Describe the set P = {ti + sj: − ∞ < t < ∞ , and − ∞ < s < ∞ } .

9. Describe the set P = {5k + ti + sj: − ∞ < t < ∞ , and − ∞ < s < ∞} .

10. Let u = (2,-4,1) and v = (1,2,3). Describe the set

P = {tu + sv: − ∞ < t < ∞ , and − ∞ < s < ∞} .

11. Let u = (2,-4,1), v = (1,2,3), and w = (3,6,1). Describe the set

P = {w + tu + sv: − ∞ < t < ∞ , and − ∞ < s < ∞} .

12. Describe the set C = {cost i + sin t j: 0 ≤ t ≤ 2π } .

2.6
13. Describe the set E = {4 cost i + 3 sin t j: 0 ≤ t ≤ 2π } .

14. Describe the set P = {ti + t 2 j : -1 ≤ t ≤ 2} .

15. Let T be the triangle with vertices (2, 5, 7), (-1, 2, 4), and (4, -2, -6). Find the point at
which the medians intersect.

2.2 Scalar Product

You were perhaps puzzled when in grammar school you were first told that the
work done by a force is the product of the force and the displacement since both force
and displacement are, of course, vectors. We now introduce this product. It is a scalar
and hence is called the scalar product. This scalar product u ⋅ v is defined by
u ⋅ v =| u||v |cosθ ,

where θ is the angle between u and v. The scalar product is frequently also called the dot

product. Observe that u ⋅ u =| u|2 , and that u ⋅ v = 0 if and only if u and v are

perpendicular (or orthogonal ), or one or the other of the two is the zero vector. We
avoid having to use the latter weasel words by defining the zero vector to be
perpendicular to every vector; then we can say u ⋅ v = 0 if and only if u and v are
perpendicular.
Study the following picture to see that if |u| = 1, then u ⋅ v is the length of the
projection of v onto u. (More precisely, the length of the projection of a representative of
v onto a representative of u. Generally, where there is no danger of confusion, we omit
mention of this, just as we speak of the length of vectors, the angle between vectors, etc.)

2.7
It is clear that (au) ⋅ (bv ) = (ab )u ⋅ v . Study the following picture until you believe that

u ⋅ ( v + w ) = u ⋅ v + u ⋅ w for any three vectors u, v, and w.

Now let’s get a recipe for the scalar product of u = (a , b, c) and v = ( x , y , z) :

u ⋅ v = (ai + bj + ck ) ⋅ ( xi + y j + zk )
= ax i ⋅ i + ayi ⋅ j + azi ⋅ k + bxj ⋅ i + byj ⋅ j + bzj ⋅ k + cxk ⋅ i + cyk ⋅ j + czk ⋅ k
= ax + by + cz,

since i ⋅ i = j ⋅ j = k ⋅ k = 1 and i ⋅ j = i ⋅ k = j ⋅ k = 0.

We thus see that it is remarkably simple to compute the scalar product of two
vectors when we know their coordinates.

Example

2.8
 Again, let’s see how vectors can make geometry easy by using them to find the
angle between a diagonal of a cube and the diagonal of a face of the cube.
Suppose the cube has edge length s. Introduce a coordinate system so that the
faces are parallel to the coordinate planes, one vertex is the origin and the vertex at the
other end of the diagonal from the origin is (s, s, s). The vector determined by this
diagonal is thus d = si + sj + sk and the vector determined by the diagonal of the face in

the horizontal coordinate plane is f = si + sj . Thus

d ⋅ f =| d || f |cosθ = s 2 + s2 ,

where θ is the angle we seek. This gives us

2 s2 2 s2 2
cosθ = = = .
| d || f | 3s2 2s 2 3

Or,

 2
θ = Cos −1  .
 3

Exercises

16. Find the work done by the force F = 6i − 3 j + 2 k in moving an object from the point

(1, 4, -2) to the point (3, 2, 5).

17. Let L be the line passing through the origin and the point (2, 5), and let M be the line
passing through the points (3, -2) and (5, 3). Find the smaller angle between L and M.

18. Find an angle between the lines 3x + 2 y = 1 and x − 2 y = 3.

2.9
19. Suppose L is the line passing through (1, 2) having slope -2, and suppose M is the

line tangent to the curve y = x 3 at the point (1, 1). ). Find the smaller angle between

L and M.

20. Find an angle between the diagonal and an adjoining edge of a cube.

21. Suppose the lengths of the sides of a triangle are a, b, and c; and suppose γ is the

angle opposite the side having length c. Prove that

c 2 = a 2 + b 2 − 2ab cosγ .

(This is, of course, the celebrated Law of Cosines.)

22. Let v = (1, 2, 5). . What is the graph of the equation v ⋅ ( x , y , z) = 0 ?

2.3 Vector Product

Hark back to grammar school physics once again and recall what you were taught
about the velocity of a point at a distance r from the axis of rotation; you were likely told
that the velocity is rω , where ω is the rate at which the turntable is rotating-the so-

called angular velocity. We now know that these quantities are actually vectors-ω is the

angular velocity, and r is the position vector of the point in question. The grammar school

quantities are the magnitudes of ω (the angular speed) and of r . The velocity of the point

is the so-called vector product of these two vectors. The vector product of vectors u and
v is defined by
u × v =| u|| v||sin θ | n ,

2.10
where θ is the angle between u and v and n is a vector of length 1 (such vectors are called
unit vectors) which is orthogonal to both u and v and which points in the direction a right-
hand threaded bolt would advance if u were rotated into the direction of v.

Note first that this is a somewhat more exciting product than you might be used
to:
the order of the factors makes a difference. Thus u × v = − v × u .
Now let’s find a geometric construction of the vector product u × v . Proceed as
follows. Let P be a plane perpendicular to u. Now project v onto this plane, giving us a
vector v * perpendicular to u and having length | v ||sin θ | . Now rotate this vector v * 90

degrees around u in the “positive direction.” (By the positive direction of rotation about
a vector a, we mean the diction that would cause a right-hand threaded bolt to advance in
the direction of a. ) This gives a vector v ** having the same length as v * and having the
direction of u × v . Thus u × v =| u| v** :

2.11
Now, why did we go to all this trouble to construct u × v in this fashion? Simple. It
makes it much easier to see that for any three vectors u, v, and w, we have
u × (v + w) = u × v + u × w .

(Draw a picture!)
We shall see how to compute this vector product u × v for
u = (a , b, c) = ai + bj + ck and v = ( x , y , z) = xi + y j + zk .

We have
u × v = (ai + bj + ck ) × ( xi + y j + zk )
= ax ( i × i) + ay ( i × j ) + az( i × k ) +
bx ( j × i ) + by ( j × j ) + bz( j × k) +
cx( k × i ) + cy( k × j ) + cz( k × k )

This looks like a terrible mess, until we note that

i× i = j× j = k× k =0,

i × j = −( j × i ) = k ,

j × k = −( k × j ) = i , and

k × i = −( i × k ) = j .

Making these substitutions in the above equation for u × v gives us

2.12
 u × v = (bz − cy )i + (cx − az) j + ( ay − bx ) k .

This is not particularly hard to remember, but there is a nice memory device using
determinants:

i j k
u× v = a b c .
x y z

Example

Let’s find the velocity of a point on the surface of the Earth relative to a
coordinate system whose origin is fixed at its center-we thus shall consider only motion
due to the Earth’s rotation, and neglect its motion about the sun, etc. For our point on the
Earth, choose Room 254, Skiles Classroom Building at Georgia Tech. The latitude of the
room is about 33.75 degrees (North, of course.), and it is about 3960 miles from the center
of the Earth. As we said, the origin of our coordinate system is the center of the Earth.
We choose the third axis to point through the North Pole; In other words, the coordinate
vector k points through the North Pole. The velocity of our room, is of course, not a
constant, but changes as the Earth rotates. We find the velocity at the instant our room is
in the coordinate plane determined by the vectors i and k.
The Earth makes one complete revolution every 24 hours, and so its angular
2π
velocity ω is ω = k ≈ 02618
. k radians/hour. The position vector r of our room is
24

r = 3960(cos(3375
. ) i + sin(3375
. ) k ) ≈ 32926
. i + 22001
. k miles. Our velocity is thus

2.13
 i j k
ω ×r = 0 0 02618
. ≈ 862 j miles/hour.
32926
. 0 22001.

Suppose we want to find the area of a parallelogram, the non-parallel sides of

which are representatives of the vectors a and b:

The area A is clearly A =| a|| b|sin θ = | a × b| .

Example
Find the are of the parallelogram with a vertex (1,4,-2) and the vertices at the other
ends of the sides adjoining this vertex are (4, 7, 8), and (6, 10, 20). This is easy. This is
just as in the above picture with a = (4 − 1) i + (7 − 4) j + (8 − (− 2)) k = 3i + 3 j + 10k and

b = (6 − 1) i + (10 − 4) j + (20 − ( −2 ))k = 5i + 6 j + 22 k . So we have

i j k
a × b = 3 3 10 = 6i − 16 j + 3k ,
5 6 22

and so,

Area =| a × b| = 6 2 + 16 2 + 32 = 301 .

Exercises

2.14
23. Find a a vector perpendicular to the plane containing the points (1,4,6), (-1,2,-7), and
(-3,6,10).

24. Are the points (0,4,7), (2, 6, 8), and (5, 10, 20) collinear? Explain how you know?

25. Find the torque created by the force f = 3i + 2 j − 3k acting at the point

a = i − 2 j − 7k .

26. Find the area of the triangle whose vertices are (0,0,0), (1,2,3), and (4,7,12).

27. Find the volume of the parallelepiped

2.15
 Chapter Three

Vector Functions

3.1 Relations and Functions

We begin with a review of the idea of a function. Suppose A and B are sets. The
Cartesian product A × B of these sets is the collection of all ordered pairs (a ,b) such
that a ∈ A and b ∈ B . A relation R is simply a subset of A × B . The domain of R is
the set dom R = {a ∈ A:( a , b) ∈ R} . In case A = B and the domain of R is all of A, we call
R a relation on A. A relation R ⊂ A × B such that (a ,b) ∈ R and (a , c) ∈ R only if b =
c is called a function. In other words, if R is a function, and a ∈ dom R , there is exactly
one ordered pair (a ,b) ∈ R . The second “coordinate” b is thus uniquely determined by a.
It is usually denoted R (a ) . If R ⊂ A × B is a relation, the inverse of R is the relation

R −1 ⊂ B × A defined by R −1 = {(b, a ):( a, b) ∈ R} .

Example
Let A be the set of all people who have ever lived and let S ⊂ A × A be the relation
defined by S = {( a, b): b is the mother of a} . The S is a relation on A, and is, in fact, a

function. The relation S −1 is not a function, and domS −1 ≠ A .

The fact that f ⊂ A × B is a function with dom f = A is frequently indicated by

writing f : A → B , and we say f is a function from A to B. Very often a function f is
defined by specifying the domain, and giving a recipe for finding f(a). Thus we may
define the function f from the interval [0,1] to the real numbers by f ( x) = x 2 . This says

that f is the collection of all ordered pairs ( x , x 2 ) in which x ∈[01

, ].

Exercises

3.1
1. Let A be the set of all Georgia Tech students, and let B be the set of real numbers.
Define the relation W ⊂ A × B by W = {( a, b): b is the weight (in pounds) of a} . Is

W a function? Is W −1 a function? Explain.

2. Let X be set of all states of the U. S., and let Y be the set of all U. S. municipalities.
Define the relation c ⊂ X × Y by c = {(x , y ): y is the capital of x} . Explain why c is
a function, and find c(Nevada), c(Missouri), and c(Kentucky).

3. With X , Y as in Exercise 2, let b be the function b = {( x , y ): y is the largest city in x} .

a)What is b(South Carolina)?
b)What is b(California)?
c)Let f = c ∩ b , where c is the function defined in Exercise 2. Find dom f.

4. Suppose f ⊂ X × Y and g ⊂ X × Y . If f is a function, is it necessarily true that

f ∩ g is a function? Prove your answer.

5. Suppose f ⊂ X × Y and g ⊂ X × Y . If f and g are both functions, is it necessarily

true that f ∪ g is a function? Prove your answer.

6. Suppose f : X → Y is a function and the inverse f −1

is also a function.
−1
a)What is f ( f ( x )) ? Explain.

b)If y ∈ dom f −1
, what is f ( f −1
( y )) ? Explain.

3.2 Vector Functions

Our interest now will be focused on functions f : X → Y in which Y is a set of

vectors. These are called vector functions, or sometimes, vector-valued functions.
Initially, we shall be solely interested in the special case in which X is a “nice” set of real
numbers, such as an interval. As the drama unfolds, we shall see that such functions
provide just the right tool for describing curves in space.
Let’s begin with a simple example. Let X be the entire real line and let the function
f be defined by f (t ) = ti + t 2 j . It should be reasonably clear that if we place the tail of

3.2
f (t ) (actually,. a representative of f (t ) ) at the origin, the nose will lie on the curve

y = x 2 . In fact, as t varies over the reals, the nose traces out this curve. The function f is
called a vector description of the curve. Let’s look at another example. This time, let
g (t ) = cos t i + sin t j for 0 ≤ t ≤ 4π . What is the curve described by this function? First,
note that for all t, we have | g(t )| = 1. The nose of g thus always lies on the circle of radius

one centered at the origin. It’s not difficult to see that, in fact, as t varies from 0 to 2π, the

nose moves around the circle once, and as t varies on from 2π to 4π, the nose traces out the
circle again.
The real usefulness of vector descriptions is most evident when we consider curves
in space. Let f (t ) = cos ti + sin tj + t k , for all t ≥ 0 . Now, what curve is followed by the
nose of f(t)? Notice first that if we look down on this curve from someplace up the
positive third axis (In other words, k is pointing directly at us.), we see the circle described
by cos ti + sin tj . As t increases, we run around this circle and the third component of our
position increases linearly. Convince yourself now that this curve looks like this:

This curve is called a helix, or more precisely, a right circular helix. The picture was
drawn by Maple. Let’s draw another. How about the curve described by the vector
function g (t ) = cos t i + sin t j + sin(2 t ) k ? This one is just a bit more exciting. Here’s a
computer drawn picture:

3.3
(This time we put the axes where they are “supposed to be.”)
Observe that in giving a vector description, we are in effect specifying the three
coordinates of points on the curves as ordinary real valued functions defined on a subset of
the reals. Assuming the axes are labeled x, y, and z, the curve described by the vector
function
r(t ) = f (t ) i + g (t ) j + h( t ) k
is equivalently described by the equations

x = f (t )
y = g (t )
z = h(t )

These are called parametric equations of the curve (The variable t is called the
parameter.).

Exercises

7 . Sketch or otherwise describe the curve given by f (t ) = ti + t 3 k for −1 ≤ t ≤ 3 .

8 . Sketch or otherwise describe the curve given by f (t ) = (2t − 3) i + ( 3t + 1) j .

[Hint: Find an equation in x and y, the graph of which is the given curve.]

3.4
9 . Sketch or otherwise describe the curve given by c( t ) = cos ti + sin t j + 7 k .

10. Sketch or otherwise describe the curve given by

c( t ) = cos(t 2 ) i + sin(t 2 ) j + 7 k .

11. Find an equation in x and y, the graph of which is the curve

g (t ) = 3 cos ti + 4 sin tj .

12. a)Find a vector equation for the graph of y = x 3 + 2 x 2 + x + 5.

b)Find a vector equation for the graph of x = y 3 + 2 y 2 + y + 5.

2 2

13. Find a vector equation for the graph of x 3 + y 3 = 1 .

14. a)Sketch or otherwise describe the curve given by the function r(t ) = a + tb , where
a = 2 i − j + 3k and b = i + 3 j − 5k .
b)Express r(t) in the form r(t ) = f (t ) i + g (t ) j + h( t ) k .

15. Describe the curve given by L (t ) = (3t + 1)i + (1 − t ) j + 2tk .

16. Find a vector function for the straight line passing through the point (1,4,-2) in the
direction of the vector v = i − j + 2k .

17. a)Find a vector function for the straight line passing through the points (1,2,4) and
(3,1,5).
b)Find a vector function for the line segment joining the points (1,2,4) and (3,1,5).

18. Let L be the line through the points (1,5,-2) and (2,2,4); and let M be the line
through the points (2,4,6) and (-3,1,-2). Find a vector description of the line which
passes through the point (1,1,2) and is perpendicular to both L and M.

3.3 Limits and Continuity

3.5
 Recall from grammar school what we mean when we say the limit at t 0 of a real-
valued, or scalar, function f is L. The definition for vector functions is essentially the
same. Specifically, suppose f is a vector valued function, t 0 is a real number, and L is a

vector such that for every real number ε > 0, there is a δ > 0 such that | f ( t ) − L | < ε

whenever 0 < | t − t 0 | < δ and t is in the domain of f. This is traditionally written

lim f (t ) = L .
t → t0

The vector L is called a limit of f at a.

Suppose α( t ) is a scalar function for which lim α (t ) = a , and f is a vector function
t → t0

for which lim f (t ) = L . It is but a modest exercise to show that

t → t0

lim(α (t ) f ( t )) = aL .
t → t0

To see this, we use the “behold!” method. Let ε > 0 be given. Choose δ 1 ,δ 2 ,δ 3 , and
δ 4 so that
ε
| f ( t) − L|< for 0 < | t − t 0 | < δ 1 ;
3(1+| a |)

ε
| f ( t) − L|< for 0 < | t − t 0 | < δ 2 ;
3
ε
| α (t ) − a| < for 0 < | t − t 0 | < δ 3 ; and
3(1+ | L |)

ε
| α (t ) − a| < for 0 < | t − t 0 | < δ 4 .
3

Now let δ = min{δ 1 , δ 2 , δ 3 , δ 4 } suppose t is such that 0 < | t − t 0 | < δ . Then

| α (t ) f (t ) − aL| = | a ( f (t ) − L ) + L (α (t ) − a) + (α (t ) − a )( f (t ) − L )|
≤ | a( f (t ) − L )|+| L (α ( t ) − a )|+ |(α (t ) − a )||( f (t ) − L )|
|a| | L| ε ε ε ε ε
< ε + ε + < + + =ε
3(1+|a|) 3(1+| L|) 3 3 3 3 3

3.6
Or, in other words,
lim(α (t ) f ( t )) = aL ,
t → t0

which is what we set out to show.

Now suppose f (t ) = x ( t ) i + y (t ) j + z(t ) k and L = ai + bj + c k . Then we see
that lim f (t ) = L if and only if
t → t0

lim x( t ) = a ,
t → t0

lim y (t ) = b, and
t → t0

lim z( t ) = c.
t → t0

It is now easy to show that all the usual nice properties of limits are valid for vector
functions:

lim( f (t ) + g( t )) = lim f ( t ) + lim g(t ).

t → t0 t → t0 t → t0

lim( f (t ) ⋅ g(t )) = (lim f (t )) ⋅ (lim g(t )) .

t → t0 t →t 0 t → t0

lim( f (t ) × g(t )) = (lim f (t )) × (lim g(t )) .

t → t0 t →t 0 t → t0

We are now ready to say what we mean by a vector function’s being continuous at
a point of its domain. Suppose t 0 is in the domain of the vector function f. Then we say f
is continuous at t 0 if it is true that lim f ( t ) = f ( t 0 ) . It is easy to see that if
t→ t0

f (t ) = x ( t ) i + y (t ) j + z (t ) k ,
then f is continuous at t 0 if and only if each of the everyday scalar functions x ( t ), y (t ), and
z(t ) is continuous at t 0 . This shows there is nothing particularly mysterious or exotic
about continuity of vector functions.
If f is continuous at each point of its domain, then we say simply that f is
continuous,

Exercises

3.7
19. Is it possible for a function f to have more than one limit at t = t 0 ? Prove your
answer.

20. Suppose m is a continuous real-valued function and f is a continuous vector-valued

function. Is the vector function h defined by h(t ) = m(t ) f (t ) also continuous?
Explain.

21. Let f and g be continuous at t = t 0 . Is the function h defined by

h( t ) = f ( t ) × g( t ) continuous? Explain. How about the function r(t ) = f (t ) ⋅ g( t ) ?

1
22. Let r(t ) = ti + t 2 j + k . Is r a continuous function? Explain.
t

23. Suppose r is a continuous function. Explain how you know that the length
function n(t ) =| r(t )| is continuous.

3.8
 Chapter Four

Derivatives

4.1 Derivatives
Suppose f is a vector function and t 0 is a point in the interior of the domain of f

( t 0 in the interior of a set S of real numbers means there is an interval centered at t 0 that

is a subset of S.). The derivative is defined just as it is for a plain old everyday real
valued function, except, of course, the derivative is a vector. Specifically, we say that f is
differentiable at t 0 if there is a vector v such that

1
lim [ f (t 0 + h) − f (t 0 )] = v .
t → t0 h

The vector v is called the derivative of f at t 0 .

Now, how would we find such a thing? Suppose f (t ) = a( t ) i + b (t ) j + c (t ) k .

Then

1  a (t + h) − a( t 0 )   b(t 0 + h) − b(t 0 )   c(t + h) − c(t 0 ) 

[ f (t 0 + h) − f (t 0 )] =  0 i +   j + 0 k .
h  h   h   h 

It should now be clear that the vector function f is differentiable at t 0 if and only if each

of the coordinate functions a( t ), b(t ), and c(t ) is. Moreover, the vector derivative v is

v = a '( t ) i + b'( t ) j + c'( t ) k .

Now we “know” what the derivative of a vector function is, and we know how to

compute it, but what is it, really? Let’s see. Let f (t ) = ti + t 3 j . This is, of course, a

vector function which describes the graph of the function y = x 3 . Let’s look at the

4.1
derivative of f at t 0 : v = i + 3t 02 j . Convince yourself that the direction of the vector v is

the direction tangent to the graph of y = x 3 at the point (t 0 , t 03 ) . It is not so clear what

we should define to be the tangent to a curve other than a plane curve. Again, vectors
come to our rescue. If f is a vector description of a space curve, the direction of the
derivative f '( t ) vector is the tangent direction at the point f (t ) -the derivative f '( t ) is

said to be tangent to the curve at f (t ) .

If f (t ) specifies the position of a particle at time t, then, of course, the derivative

is the velocity of the particle, and its length | f '( t )| is the speed. Thus the distance the

particle travels from time t = a to time t = b is given by the integral of the speed:

d = ∫ | f '( t )| dt .
a

If the particle behaves nicely, this distance is precisely the length of the arc of the curve
from f (a ) to f (b) . It should be clear what we mean by “behaves nicely”. . For the

distance traveled by the particle to be the same as the length of its path, there must be no
“backtracking”, or reversing direction. This means we must not allow the velocity to be
zero for any t between a and b.

Example
Consider the function r(t ) = cos ti + sin tj . Then the derivative, or velocity, is

r'( t ) = − sin t i + cos tj . This vector is indeed tangent to the curve described by r (which

we already know to be a circle of radius 1 centered at the origin.) at r(t ) . Note that the

scalar product r(t ) ⋅ r'( t ) = − sin t cost + sin t cos t = 0 , and so the tangent vector and the

vector from the center of the circle to the point on the circle are perpendicular-a well-
known fact you learned from Mrs. Turner in 4th grade. Note that the derivative is never

4.2
zero-there is no value of t for which both cost and sint vanish. The length of a piece of the
curve can thus be found by integrating the speed:

2π 2π 2π

p= ∫ | r'( t )| dt = ∫ sin 2 t + cos 2 tdt = ∫ dt = 2π .

0 0 0

No surprise here.

Exercises

1. a)Find a vector tangent to the curve f (t ) = t 2 i + t 3 j + (1 − t )k at the point (1, 1, 0).

b)Find a vector equation for the line tangent to this same curve at the point (1, 1, 0).

2. The position of a particle is given by r(t ) = cos(t 3 ) i + sin(t 3 ) j .

a)Find the velocity of the particle.

b)Find the speed of the particle.
c)Describe the path of the particle, and find its length.

3. Let L be the line tangent to the curve g (t ) = 10 cos ti + 10 sin t j + 16tk at the point

10 10
( , ,4π ) . Find the point at which L intersects the i-j plane.
2 2

4. Let L be the straight line passing through the point (5, 0, 3) in the direction of the
vector a = i + 2 j − k , and let M be the straight line passing through the point (0, 0, 6)

in the direction of b = i − 3 j + 2k .

a)Are L and M parallel? Explain.

b)Do L and M intersect? Explain.

4.3
5. Let L be the straight line passing through the point (1, 1, 3) in the direction of the
vector a = 2 i + j − k , and let M be the straight line passing through the point (0, 1, 5)

in the direction of b = 3i − j + 2k . Find the distance between L and M.

6. Find the length of the arc of the curve R (t ) = 3cos t i + 3 sin t j + 4tk between the

points (3, 0, 0) and (3, 0, 16π).

7. Find an integral the value of which is the length of the curve y = x 2 between the

points (-1, 1) and (1, 1).

4.2 Geometry of Space Curves-Curvature

Let R(t ) be a vector description of a curve. Then the distance s(t ) along the

curve from the point R(t 0 ) to the point R(t ) is, as we have seen, simply
t

s(t ) = ∫ | R '( ξ )| dξ ;
t0

assuming, of course, that R'( t ) ≠ 0 . The speed is

ds
=| R'( t )| .
dt
Now then the vector
R '( t ) R'( t ) dt dR
T= = = R'( t ) =
| R'( t )| ds / dt ds ds

is tangent to R and has length one. It is called the unit tangent vector.

Consider next the derivative

4.4
 d dT dT dT
T ⋅T = T ⋅ + ⋅ T = 2T ⋅ .
ds ds ds ds

dT dT
But we know that T ⋅ T = |T | 2 = 1. Thus T ⋅ = 0, which means that the vector is
ds ds
perpendicular, or orthogonal, or normal, to the tangent vector T. The length of this vector

is called the curvature and is usually denoted by the letter κ. Thus

dT
κ= .
ds

The unit vector

1 dT
N=
κ ds

is called the principal unit normal vector, and its direction is sometimes called the
principal normal direction.

Example
Consider the circle of radius a and center at the origin: R (t ) = a cost i + a sin t j .

ds
Then R '( t ) = − a sin ti + a cos tj , and = | R'( t )| = a 2 sin 2 t + a 2 cos 2 t = a 2 = | a| = a .
dt
Thus

1
T= R' (t ) = − sin ti + cost j .
a

Let’s not stop now.

dT dT dt 1 dT 1
= = = ( − cos t i − sin t j ) .
ds dt ds a dt a

4.5
 dT 1
Thus κ = = , and N = − (cos t i + sin t j ) . So the curvature is the reciprocal of the
ds a

radius and the principal normal vector points back toward the center of the circle.

Another Example

This time let R = (t + 1) i + 2t j + t 2 k . First, R '( t ) = i + 2 j + 2tk , and so

ds
= | R'( t )| = 5 + 4t 2 . The unit tangent is then
dt
1
T= ( i + 2 j + 2tk ) .
5 + 4t 2

It’s a bit of a chore now to find the curvature and the principal normal, so let’s use a
computer algebra system; viz., Maple:
First, let’s enter the unit tangent vector T:

See if we got it right:

T(t);

Fine. Now differentiate:

A(t);

We need to tidy this up:

4.6
 B(t);

dT
This vector is, of course, the normal . We continue and find the curvature κ and the
ds

principal normal N.

kappa:=t->simplify(sqrt(dotprod(B(t),B(t))));
kappa(t);

N(t);

ds
So there we have at last the speed , the unit tangent T, the curvature κ., and the
dt

principal normal N.

Exercises

8. Find a line tangent to the curve R (t ) = (t 2 + t ) i + (t + 1) j − ( t 3 + 5) k and passing

through the point (5, -2, 15), or show there is no such line.

4.7
9. Find the unit tangent T, the principal normal N, and the curvature κ, for the curves:

a) R (t ) = 5 cos(t )i + 5 sin(t ) j + 2 tk

b) R (t ) = (2t + 3) i + (5 − t 2 ) j

c) R (t ) = e t cos t i + e t sin tj + 6k

10. Find the curvature of the curve y = f ( x ) at ( x0 , f ( x0 )) .

11. Find the curvature of R (t ) = ti + t 2 j . At what point on the curve is the curvature the

largest? smallest?

12. Find the curvature of R (t ) = ti + t 3 j . At what point on the curve is the curvature the

largest? smallest?

4.3 Geometry of Space Curves-Torsion

Let R(t ) be a vector description of a curve. If T is the unit tangent and N is the

principal unit normal, the unit vector B = T × N is called the binormal. Note that the
dB
binormal is orthogonal to both T and N. Let’s see about its derivative with respect
ds
dB
to arclength s. First, note that B ⋅ B = | B|2 = 1, and so B ⋅ = 0 , which means that
ds
dB
being orthogonal to B, the derivative is in the plane of T and N. Next, note that B is
ds
dB
perpendicular to the tangent vector T, and so B ⋅ T = 0 . Thus ⋅ T = 0 . So what have
ds

4.8
 dB
we here? The vector is perpendicular to both B and T, and so must have the
ds
direction of N (or, of course, - N). This means

dB
= −τ N .
ds

The scalar τ is called the torsion.

Example

Let’s find the torsion of the helix R (t ) = a cos ti + a sin tj + btk . Here we go!

ds
R '( t ) = − a sin ti + a cos tj + bk . Thus = | R'( t )| = a 2 + b 2 , and we have
dt

1
T= (− a sin t i + a cost j + bk ) .
a + b2
2

Now then
dT dT dt −a
= = 2 (cos ti + sin tj ) .
ds dt ds (a + b 2 )

Therefore,
a
κ= and N = − (cos t i + sin t j ) .
(a + b 2 )
2

Let’s don’t stop now:

4.9
 i j k
1 1
B =T ×N = − a sin t a cos t b = (b sin ti − b cost j + ak ) ;
a2 + b2 a2 + b2
− cos t − sin t 0

and
dB dB dt b −b
= = 2 (cos ti + sin tj ) = 2 N.
ds dt ds ( a + b )
2
(a + b2 )

The torsion, at last:

b
τ = .
a + b2
2

Suppose the curve R(t ) is such that the torsion is zero for all values of t. In other

dB
words, ≡ 0 . Look at
ds
d dR dB
[( R( t ) − R (t 0 )) ⋅ B] = ⋅ B + ( R( t ) − R (t 0 )) ⋅ = 0.
ds ds ds

Thus the scalar product ( R(t ) − R (t 0 )) ⋅ B is constant. It is 0 at t 0 , and hence it is 0 for

all values of t. This means that R (t ) − R( t 0 ) and B are perpendicular for all t, and so

R (t ) − R( t 0 ) lies in a plane perpendicular to B. In other words, the curve described by

R(t ) is a plane curve.

Exercises

13. Find the binormal and torsion for the curve R (t ) = 4 cos ti + 3sin tk .

sin t sin t
14. Find the binormal and torsion for the curve R (t ) = i + cos t j + k.
2 2

4.10
15. Find the curvature and torsion for R (t ) = ti + t 2 j + t 3 k.

1+ t 1− t 2
16. Show that the curve R (t ) = ti + j+ k lies in a plane.
t t

17. What is the vector B × T ? How about N × T ?

4.4 Motion
Suppose t is time and R(t ) is the position vector of a body. Then the curve

dR
described by R(t ) is the path, or trajectory, of the body, v( t ) = is the velocity, and
dt
dv ds
a(t ) = is the acceleration. We know that v( t ) = T , and so the direction of the
dt dt
velocity is the unit tangent T. Let’s see about the direction of the acceleration:

dv d 2 s ds dT
a( t ) = = 2T+
dt dt dt dt
2 ,
 ds 
2
d s
= 2 T +  κN
dt  dt 

dT ds
since = κ N . This tells us that the acceleration is always in the plane of the
dt dt

d 2s
vectors T and N. The derivative of the speed is the tangential component of the
dt 2
2
 ds 
acceleration, and κ   is the normal component of the acceleration.
 dt 

Example

4.11
 Suppose a person who weighs 160 pounds moves around a circle having radius 20
feet at a constant speed of 60 miles/hour. What is the magnitude of the force on this
person at any time?
First, we know the force f is the mass times the acceleration: f (t ) = ma (t ) . Thus

2
d 2s  ds 
f =m T + mκ   N
dt 2
 dt 

also have The speed is a constant 60 miles/hour, or 88 feet/second; in other words,

ds d 2s
= 88 and = 0 . Hence,
dt dt

2 2
 ds   ds
| f | =| mκ   N | = mκ   .
 dt   dt 

160 1
The mass m = = 5 slugs, and the curvature κ = . The magnitude of the force is
32 20
5 ⋅882
thus | f |= = 1936 pounds.
20

Exercises

18. The position of an object at time t is given by r(t ) = ti + (t 3 − 2) j + 2t k . Find the

velocity, the speed, and the tangential and normal components of the acceleration.

19. A force f (t ) = t 2 i + (t − 1) j + k newtons is applied to an object of mass 2 kilograms.

At time t = 0, the object is at the origin. Find its position at time t.

4.12
20. A projectile of weight w is fired from the origin with an initial speed v0 in the

direction of the vector cosθ i + sin θ j , and the only force acting on the projectile is

f = −wj .

a)Find a vector description of the trajectory of the projectile.

b)Find an equation the graph of which is the trajectory.

21. A 16 lb. bowling ball is rolled along a track with a circular vertical loop of radius a
feet. What must the speed of the ball be in order for it not to fall from the track?
What must the speed of an 8 lb. ball be in order for it not to fall?

4.13
 Chapter Five

More Dimensions

5.1 The Space R n

We are now prepared to move on to spaces of dimension greater than three. These
spaces are a straightforward generalization of our Euclidean space of three dimensions. Let
n be a positive integer. The n-dimensional Euclidean space R n is simply the set of
all ordered n-tuples of real numbers x = ( x1 , x 2 ,K , x n ) . Thus R 1 is simply the real

numbers, R 2 is the plane, and R 3 is Euclidean three-space. These ordered n-tuples are
called points, or vectors. This definition does not contradict our previous definition of a
vector in case n =3 in that we identified each vector with an ordered triple ( x1 , x2 , x3 ) and
spoke of the triple as being a vector.
We now define various arithmetic operations on R n in the obvious way. If we
have vectors x = ( x1 , x 2 ,K , x n ) and y = ( y1 , y 2 ,K , y n ) in R n , the sum x + y is defined
by
x + y = ( x 1 + y1 , x 2 + y 2 , K, x n + yn ) ,

and multiplication of the vector x by a scalar a is defined by

ax = (ax 1 , ax2 ,K, ax n ) .

It is easy to verify that a( x + y) = ax + ay and (a + b) x = ax + bx .

Again we see that these definitions are entirely consistent with what we have done
in dimensions 1, 2, and 3-there is nothing to unlearn. Continuing, we define the length,
or norm of a vector x in the obvious manner

| x| = x12 + x 22 +K+ xn2 .

The scalar product of x and y is

5.1
 n
x ⋅ y = x1 y1 + x 2 y 2 +K+ x n y n = ∑ x i y i .
i =1

It is again easy to verify the nice properties:

| x| 2 = x ⋅ x ≥ 0,
| ax| =| a || x|,
x ⋅ y = y⋅x ,
x ⋅ ( y + z) = x ⋅ y + x ⋅ z , and
(ax ) ⋅ y = a ( x ⋅ y ) .

The geometric language of the three dimensional setting is retained in higher

dimensions; thus we speak of the “length” of an n-tuple of numbers. In fact, we also speak
of d( x , y ) = | x − y| as the distance between x and y . We can, of course, no longer rely

on our vast knowledge of Euclidean geometry in our reasoning about R n when n > 3.
Thus for n ≤ 3 , the fact that | x + y | ≤ | x | + | y| for any vectors x and y was a simple
consequence of the fact that the sum of the lengths of two sides of a triangle is at least as
big as the length of the third side. This inequality remains true in higher dimensions, and,
in fact, is called the triangle inequality, but requires an essentially algebraic proof.
Let’s see if we can prove it.
Let x = ( x1 , x 2 ,K , x n ) and y = ( y1 , y 2 ,K , y n ) . Then if a is a scalar, we have

| ax + y|2 = (ax + y ) ⋅ (ax + y) ≥ 0 .

Thus,
(ax + y ) ⋅ ( ax + y ) = a 2 x ⋅ x + 2ax ⋅ y + y ⋅ y ≥ 0.

This is a quadratic function in a and is never negative; it must therefore be true that

4( x ⋅ y ) 2 − 4( x ⋅ x )( y ⋅ y ) ≤ 0 , or

| x ⋅ y | ≤| x|| y| .

This last inequality is the celebrated Cauchy-Schwarz-Buniakowsky inequality. It

is exactly the ingredient we need to prove the triangle inequality.

5.2
 | x + y |2 = ( x + y ) ⋅ ( x + y) = x ⋅ x + 2 x ⋅ y + y ⋅ y .

Applying the C-S-B inequality, we have

| x + y |2 ≤| x| 2 + 2| x || y|+ | y| 2 = (| x|+ | y |) 2 , or

| x + y | ≤ | x | + | y| .

Corresponding to the coordinate vectors i , j , and k, the coordinate vectors

e1 , e 2 ,K , e n are defined in R n by
e1 = (10, ,00, ,K ,0)
e2 = (0,1,0 ,0 ,K ,0)
e3 = ( 0,0 ,1,0 ,K,0) ,
M
en = (0,00
, ,K ,01
,)

Thus each vector x = ( x1 , x 2 ,K , x n ) may be written in terms of these coordinate vectors:

n
x = ∑ x i ei .
i= 1

Exercises

1 . Let x and y be two vectors in R n . Prove that | x + y |2 = | x|2 +| y| 2 if and only if

x ⋅ y = 0. (Adopting more geometric language from three space, we say that x and y
are perpendicular or orthogonal if x ⋅ y = 0.)

2 . Let x and y be two vectors in R n . Prove

a)| x + y |2 − | x − y| 2 = 4 x ⋅ y .

b)| x + y |2 +| x − y |2 = 2[| x|2 + | y| 2 ] .

5.3
3 . Let x and y be two vectors in R n . Prove that | | x| − | y | | ≤ | x |+| y| .

4 . Let P ⊂ R 4 be the set of all vectors x = ( x1 , x 2 , x 3 , x 4 ) such that

3x1 + 5x 2 − 2x 3 + x 4 = 15 .

Find vectors n and a such that P = {x ∈ R 4 : n ⋅ ( x − a) = 0} .

5 . Let n and a be vectors in R n , and let P = {x ∈ R n: n ⋅ ( x − a) = 0 .

a)Find an equation in x 1 , x 2 , K, and x n such that x = ( x1 , x 2 ,K , x n ) ∈ P if and only if
the coordinates of x satisfy the equation.
b)Describe the set P be in case n = 3. Describe it in case n =2.
[The set P is called a hyperplane through a.]

5.2 Functions
We now consider functions F: R n → R p . Note that when n = p = 1, we have the
usual grammar school calculus functions, and when n = 1 and p = 2 or 3, we have the
vector valued functions of the previous chapter. Note also that except for very special
circumstances, graphs of functions will not play a big role in our understanding. The set of
points ( x, F ( x)) resides in R n+ p since x ∈ R n and F ( x) ∈ R p ; this is difficult to “see”
unless n + p ≤ 3 .
We begin with a very special kind of functions, the so-called linear functions. A
function F: R n → R p is said to be a linear function if

i) F ( x + y ) = F ( x) + F ( y ) for all x , y ∈ R n , and

ii) F (ax ) = aF ( x ) for all scalars a and x ∈ R n .

Example
Let n = p = 1, and define F by F ( x ) = 3x . Then
F ( x + y ) = 3( x + y ) = 3x + 3 y = F ( x ) + F ( y ) and
F (ax ) = 3(ax ) = a3 x = aF ( x ) .
This F is a linear function.

5.4
Another Example
Let F: R → R 3 be defined by F (t ) = t i + 2tj − 7tk = (t ,2t ,−7t ) . Then
F (t + s) = (t + s) i + 2( t + s) j − 7 (t + s)k
= [ti + 2tj − 7tk ] + [ si + 2sj − 7 sk ]
= F (t ) + F ( s )
Also,
F (at ) = ati + 2 atj − 7at k
= a[ ti + 2t j − 7t k ] = aF (t )
We see yet another linear function.

One More Example

Let F: R 3 → R 4 be defined by

F (( x1 , x 2 , x 3 )) = (2 x1 − x 2 + 3 x3 , x1 + 4 x 2 − 5x 3 , − x 1 + 2 x2 + x 3 , x 1 + x 3 ) .

It is easy to verify that F is indeed a linear function.

A translation is a function T:R p → R p such that T( x ) = a + x , where a is a

fixed vector in R n . A function that is the composition of a linear function followed by a

translation is called an affine function. An affine function F thus has the form
F ( x) = a + L( x) , where L is a linear function.

Example
Let F: R → R 3 be defined by F (t ) = (2 + t , 4t − 3, t ) . Then F is affine. Let
a = (2 ,4 ,0) and L(t ) = (t , 4t , t ) . Clearly F (t ) = a + L(t ) .

Exercises
6 . Which of the following functions are linear? Explain your answers.
a) f ( x) = −7 x b) g ( x ) = 2 x − 5
c) F ( x1 , x 2 ) = (2 x 1 + x2 , x 1 − x 2 , 3 x1 , 5 x1 − 2 x 2 , x1 )
d) G( x1 , x 2 , x 3 ) = x 1 x 2 + x3 e) F (t ) = (2t , t , 0, − 2t )
f) h( x1 , x 2 , x 3 , x 4 ) = (1, 0, 0) g) f ( x) = sin x

5.5
7 . a)Describe the graph of a linear function from R to R.
b)Describe the graph of an affine function from R to R.

5.6
 Chapter Six

Linear Functions and Matrices

6.1 Matrices

Suppose f : R n → R p be a linear function. Let e1 , e 2 ,K , e n be the coordinate

vectors for R n . For any x ∈ R n , we have x = x 1e 1 + x 2 e 2 +K+ x n e n . Thus

f ( x) = f ( x1 e 1 + x 2 e2 +K+ x n e n ) = x1 f ( e1 ) + x 2 f ( e2 )+K+ xn f (e n ) .

Meditate on this; it says that a linear function is entirely determined by its values
f ( e1 ), f (e 2 ),K , f (e n ) . Specifically, suppose

f ( e1 ) = (a11 , a21 , K, a p1 ),
f ( e 2 ) = (a12 , a 22 ,K , a p 2 ),
M
f ( e n ) = (a1 n , a 2 n ,K , a pn ).

Then
f ( x) = ( a11 x1 + a12 x 2 +K+ a1n x n , a 21 x1 + a 22 x 2 +K+ a 2 n x n ,K ,
a p 1 x 1 + a p 2 x 2 +K+ a pn x n ).

The numbers aij thus tell us everything about the linear function f. . To avoid labeling

these numbers, we arrange them in a rectangular array, called a matrix:

6.1
  a11 a12 K a1 n 
a 
 21 a 22 K a2 n 
 M M 
 
 a p 1 a p 2 K a pn 

The matrix is said to represent the linear function f.

For example, suppose f : R 2 → R 3 is given by the receipt

f ( x1 , x 2 ) = (2 x1 − x 2 , x1 + 5x 2 , 3x 1 − 2 x2 ) .

Then f ( e1 ) = f (10
, ) = ( 2,1,3) , and f ( e 2 ) = f (01
, ) = ( −15
, ,− 2) . The matrix representing f

is thus

 2 −1
1 5 
 
 3 − 2

Given the matrix of a linear function, we can use the matrix to compute f ( x) for

any x. This calculation is systematized by introducing an arithmetic of matrices. First,

we need some jargon. For the matrix

 a11 a12 K a1n 

a a22 K a 2 n 
A=
21
,
 M M 
 
 a p1 a p2 K a pn 

6.2
  a1 j 
a 
the matrices [ai 1 , a i2 ,K , ain ] are called rows of A, and the matrices   are called
2j

 M 
 
 a pj 

columns of A. Thus A has p rows and n columns; the size of A is said to be p × n . A

vector in R n can be displayed as a matrix in the obvious way, either as a 1× n matrix, in

which case the matrix is called a row vector, or as a n × 1 matrix, called a column vector.

Thus the matrix representation of f is simply the matrix whose columns are the column

vectors f ( e1 ), f (e1 ),K , f (e n ) .

Example

Suppose f : R 3 → R 2 is defined by

f ( x 1 , x 2 , x 3 ) = (2 x1 − 3 x2 + x 3 , − x 1 + 2 x2 − 5x 3 ) .

So f ( e1 ) = f (10
, ,0) = (2,− 1) , f ( e 2 ) = f (0,1,0 ) = ( −32
, ) , and f ( e 3 ) = f ( 0,0 ,1) = (1,− 5) .

The matrix which represents f is thus

 2 −3 1 
 −1 2 − 5
 

Now the recipe for computing f(x) can be systematized by defining the product of

a matrix A and a column vector x. Suppose A is a p × n matrix and x is a n × 1 column

6.3
vector. For each i = 12
, , K, p, let ri denote the i th row of A . We define the product Ax

to be the p × 1column vector given by

 r1 ⋅ x 
r ⋅ x 
Ax =  .
2

 M 
 
 rp ⋅ x 

If we consider all vectors to be represented by column vectors, then a linear function

f : R n → R p is given by f ( x) = Ax , where, of course, A is the matrix representation of

f.

Example

Consider the preceding example:

f ( x 1 , x 2 , x 3 ) = (2 x1 − 3 x2 + x 3 , − x 1 + 2 x2 − 5x 3 ) .

We found the matrix representing f to be

 2 −3 1 
A= .
 −1 2 −5

Then

 x1 
 2 − 3 1     2x 1 − 3x 2 + x 3 
Ax =   x2 =   = f ( x)
 − 1 2 − 5  x   − x1 + 2 x 2 − 5x 3 
 3

Exercises

6.4
1. Find the matrix representation of each of the following linear functions:

a) f ( x1 , x 2 ) = (2 x1 − x 2 , x1 + 4 x 2 , -7x 1 , 3x1 + 5x 2 ) .

b) R (t ) = 4t i − 5tj − 2t k .

c) L( x ) = 6x .

2 −1
 −2 1 
2. Let g be define by g( x ) = Ax , where A =  . Find g(3,−9 ) .
0 −3
 
3 5

3. Let f : R 2 → R 2 be the function in which f(x) is the vector that results from rotating

π
the vector x about the origin in the counterclockwise direction.
4

a)Explain why f is a linear function.

b)Find the matrix representation for f.

d)Find f(4,-9).

4. Let f : R 2 → R 2 be the function in which f(x) is the vector that results from rotating

the vector x about the origin θ in the counterclockwise direction. Find the matrix

representation for f.

5. Suppose g: R 2 → R 2 is a linear function such that g(1,2) = (4,7) and g(-2,1) = (2,2).

6.5
 Find the matrix representation of g.

6. Suppose f : R n → R p and g: R p → R q are linear functions. Prove that the

composition g o f : R n → R q is a linear function.

7. Suppose f : R n → R p and g: R n → R p are linear functions. Prove that the function

f + g: R n → R p defined by ( f + g)( x ) = f ( x ) + g ( x) is a linear function.

6.2 Matrix Algebra

Let us consider the composition h = g o f of two linear functions f : R n → R p

and g: R p → R q . Suppose A is the matrix of f and B is the matrix of g. Let’s see about

the matrix C of h. We know the columns of C are the vectors g ( f (e j )), j = 12

, ,K, n ,

where, of course, the vectors e j are the coordinate vectors for R n . Now the columns of

A are just the vectors f ( e j ), j = 1,2,K , n . Thus the vectors g ( f (e j )) are simply the

products Bf (e j ) . In other words, if we denote the columns of A by ki , i = 12

, ,K, n , so

that A = [k1 , k 2 ,K , k n ] , then the columns of C are Bk1 , Bk2 ,K, Bkn , or in other words,

C = [ Bk1 , Bk2 ,K, Bkn ] .

Example

6.6
 1 0 2
−1 −5 8 
Let the matrix B of g be given by B =   and let the matrix A of f be
 2 7 −3
 
 2 −2 1 

3 1
given by A =  1 2  . Thus f : R 2 → R 3 and g: R 3 → R 4 (Note that for the
 −4 −3

composition h = g o f to be defined, it must be true that the number of columns of B be

 3 1
the same as the number of rows of A.). Now, k1 =  1  and k2 =  2  , and so
 

 − 4  −3

 −5   −5 
 −40   −35
Bk1 =  and Bk =
25  2  25  . The matrix C of the composition is thus
 0   −3 

 −5 −5 
−40 −35 
C= .
25 25 
 0 −3 

These results inspire us to define a product of matrices. Thus, if B is an n × p

matrix, and A is a p × q matrix, the product BA of these matrices is defined to be the

n × q matrix whose columns are the column vectors Bk j , where k j is the j th column of

A. Now we can simply say that the matrix representation of the composition of two

linear functions is the product of the matrices representing the two functions.

6.7
 There are several interesting and important things to note regarding matrix

products. First and foremost is the fact that in general BA ≠ AB , even when both

products are defined (The product BA obviously defined only when the number of

columns of B is the same as the number of rows of A.). Next, note that it follows directly

from the fact that h o ( f o g ) = (h o f ) o g that for C(BA) = (CB)A. Since it does not

matter where we insert the parentheses in a product of three or more matrices, we usually

omit them entirely.

It should be clear that if f and g are both functions from R n to R p , then the

matrix representation for the sum f + g: R n → R p is simply the matrix

 a11 + b11 a12 + b12 K a1 n + b1 n 

a + b a22 + b22 K a 2 n + b2 n 
A+ B = 
21 21
,
 M 
 
 a p 1 + bp 1 a p2 + b p2 K a pn + b pn 

where

 a11 a12 L a1n 

a a22 L a 2 n 
A=
21

 M 
 
a p1 a p2 L a pn 

is the matrix of f, and

6.8
  b11 b12 L b1n 
b b22 L b2 n 
B=  21

 M 
 
b p 1 b p 2 L bpn 

is the matrix of g. Meditating on the properties of linear functions should convince you

that for any three matrices (of the appropriate sizes) A, B, and C, it is true that

A( B + C ) = AB + AC .

Similarly, for appropriately sized matrices, we have ( A + B)C = AC + BC .

Exercises

8. Find the products:

 2 1  −2   2 1 1
a)    b)   
 0 3  1   0 3 3

1 5
 −2 3
 2 1  −2 1
c)    d) [1 −3 2 − 1] 
 0 3  1 3 0 2
 
 −3 4

 1 0 0 a 11 a12 a13   0 0 0 a 11 a12 a13 

9. Find a)  0 1 0 a 21 a 22 a 23  b)  0 0 0 a 21 a 22 a 23 
 0 0 1 a 31 a 32 a 33   0 0 0 a 31 a 32 a 33 

6.9
10. Let A(θ ) be the 2 × 2 matrix for the linear function that rotates the plane θ

counterclockwise. Compute the product A(θ ) A(η ) , and use the result to give

identities for cos(θ + η ) and sin(θ + η ) in terms of cosθ , cosη , sinθ , and sinη .

11. a)Find the matrix for the linear function that rotates R 3 about the coordinate vector j

π
by (In the positive direction, according to the usual “right hand rule” for rotation.).
4

b)Find a vector description for the curve that results from applying the linear

transformation in a) to the curve R (t ) = cos ti + sin tj + t k .

12. Suppose f : R 2 → R 2 is linear. Let C be the circle of radius 1 and center at the origin.

Find a vector description for the curve f(C).

13. Suppose g: R 2 → R n is linear. Suppose moreover that , ) = (2,3)

g(11 and

g( −11
, ) = ( 4,− 5) . Find the matrix of g.

6.10
 Chapter Seven

Continuity, Derivatives, and All That

7.1 Limits and Continuity

Let x 0 ∈ R n and r > 0. The set B (a ; r ) = {x ∈ R n :| x − a| < r} is called the open

ball of radius r centered at x 0 . The closed ball of radius r centered at x 0 is the set

B (a ; r ) = {x ∈ R n :| x − a| ≤ r} . Now suppose D ⊂ R n . A point a ∈ D is called an

interior point of D if there is an open ball B (a; r) ⊂ D . The collection of all interior

points of D is called the interior of D, and is usually denoted int D. A set U is said to be
open if U = int U.

Suppose f : D → R p , where D ⊂ R n and suppose a ∈ R n is such that every

open ball centered at a meets the domain D. If y ∈ R p is such that for every ε > 0, there

is a δ > 0 so that| f ( x ) − y| < ε whenever 0 < | x − a | < δ , then we say that y is the limit of

f at a. This is written

lim f ( x) = y ,
x →a

and y is called the limit of f at a.

Notice that this agrees with our previous definitions in case n = 1 and p =1,2, or 3.
The usual properties of limits are relatively easy to establish:

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

x →a x →a x →a

lim af ( x ) = a lim f ( x ) .
x →a x →a

Now we are ready to say what we mean by a continuous function f : D → R p ,

where D ⊂ R n . Again this definition will not contradict our previous lower dimensional

7.1
definitions. Specifically, we say that f is continuous at a ∈ D if lim f ( x) = f ( a) . If f is
x →a

continuous at each point of its domain D, we say simply that f is continuous.

Example

Every linear function is continuous. To see this, suppose f : R n → R p is linear

ε
and a ∈ R n . Let ε > 0. Now let M = max{| f (e 1 )|,| f ( e2 )|,K,| f (e n )|} and let δ = .
nM

Then for x such that 0 < | x − a | < δ , we have

| f ( x ) − f (a )| = | f ( x1 e1 + x 2 e 2 +K+ xn en ) − f (a1 e 1 + a2 e 2 +K+ a n e n )|

=|( x 1 − a1 ) f ( e1 ) + ( x2 − a2 ) f (e 2 ) +K+ ( x n − an ) f (e n )|
≤ | x1 − a1 || f (e1 )|+ | x 2 − a2 || f ( e 2 )|+K+ | xn − an || f (e n )|
≤ (| x1 − a1 |+ | x2 − a 2 |+ K+ | x n − an |) M
≤ n| x − a | M
<ε

Thus lim f ( x) = f ( a) and so f is continuous.

x →a

Another Example

 x1 x 2
 , for x12 + x 22 ≠ 0
Let f : R → R be defined by
2
f ( x ) = f ( x1 , x2 ) =  x12 + x22 .
 0, otherwise

Let’s see about lim f ( x). Let x = α (11

, ) . Then for all α ≠ 0, we have
x →( 0 ,0 )

α2 1
f ( x) = f (α ,α ) = 2 = .
α +α 2
2

7.2
Now. let x = α (10
, ) = (α ,0) . It follows that all α ≠ 0, f ( x) = 0 . What does this tell

us? It tells us that for any δ > 0 , there are vectors x with 0 < | x − ( 00
, )| < δ such that

1
f ( x) = and such that f ( x) = 0 . This, of course, means that lim f ( x) does not
2 x →( 0 ,0 )

exist.

7.2 Derivatives

Let f : D → R p , where D ⊂ R n , and let x 0 ∈int D . Then f is differentiable at

x 0 if there is a linear function L such that

1
lim [ f ( x0 + h) − f ( x 0 ) − L( h)] = 0 .
h→0 | h|

The linear function L is called the derivative of f at x 0 . It is usual to identify the linear

function L with its matrix representation and think of the derivative at a p × n matrix.

Note that in case n = p = 1, the matrix L is simply the 1 × 1 matrix whose sole entry is the
every day grammar school derivative of f .
Now, how do find the derivative of f ? Suppose f has a derivative at x 0 . First, let

h = te j = ( 00
, ,K ,0, t ,0,K,0) . Then

 f 1 ( x1 , x 2 ,K, x j + t ,K , xn ) 
 f ( x , x ,K , x + t ,K, x ) 
f ( x + h) = f ( x1 , x 2 ,K , x j + t ,K, x n ) = 
2 1 2 j n 
,
 M 
 
 f p ( x1 , x 2 ,K, x j + t ,K , xn )

and

7.3
  0
 m11 m12 L m1n   0  m1 j t 
m m22 L m2 n   M  m t 
Lh =  21
=  2j ,
 M t   M 
    
 mp 1 m p 2 L m pn   M   mpj t 
 0

where x 0 = ( x 1 , x 2 ,K , x n ) , etc.

Now then,

1
[ f (x 0 + h) − f (x0 ) − L(h)]
| h|
 f1(x1, x 2 ,K, xj + t,K,x n ) − f1(x1,x 2,K, x n ) − m1 j t 
1  f2 (x1 , x2,K, x j + t,K, xn ) − f2 (x1 , x 2,K, xn ) − m2 j t 
=  
t M
 f (x , x ,K, x + t,K,x ) − f (x , x ,K, x ) − m t 
 p 1 2 j n p 1 2 n pj 

 f1(x1,x 2,K, x j + t,K, xn ) − f1(x1, x2 ,K,x n ) 

− m1 j
 t 
f (x
 2 1 2 , x ,K, x + t,K, x ) − f (x ,x ,K,, x )
− m2 j 
j n 2 1 2 n
=
 t 
M
 f p (x1 ,x2 ,K, x j + t,K, xn ) − f p (x1 ,x2 ,K,, xn ) 
− m
 t pj 


Meditate on this vector. For each component,

fi (x1 , x2 ,K,x j + t,K, xn ) − fi (x1,x 2,K,, xn )
lim
t→0 t
d
= f (x , x ,K, s,K, xn )
ds i 1 2 s = xj

This derivative has a name. It is called the partial derivative of f i with respect to the j th

variable. There are many different notations for the partial derivatives of a function
g ( x1 , x 2 ,K , xn ) . The two most common are:

7.4
 g , j ( x 1 , x 2 ,K , x n )
∂
g ( x1 , x 2 ,K , xn )
∂x j

1
The requirement that lim [ f ( x0 + h) − f ( x 0 ) − L( h)] = 0 now translates into
h→0 | h|

∂f i
mij = ,
∂x j

and, mirabile dictu, we have found the matrix L !

Example

 3x sin x 2 
Let f : R 2 → R 2 be given by f ( x1 . x2 ) =  3 1 2
. Assume f is differentiable
 x1 + x 1 x 2 
and let’s find the derivative (more precisely, the matrix of the derivative. This matrix will,

m m12 
of course, be 2 × 2 : L =  11 . Now
 m21 m22 

f 1 ( x1 , x 2 ) = 3x1 sin x 2 , and

f 2 ( x1 , x2 ) = x 13 + x1 x22

Compute the partial derivatives:

∂f 1
= 3 sin x 2
∂x 1
,
∂f 2
= 3x 1 + x 2
2 2

∂x 1

7.5
and
∂f 1
= 3x 1 cos x 2
∂x 2
.
∂f 2
= 2 x1 x 2
∂x 2

The derivative is thus

 3 sin x 3x 1 cos x 2 
L =  2 22 .
 3x1 + x 2 2 x 1 x 2 

We now know how to find the derivative of f at x if we know the derivative exists;
but how do we know when there is a derivative? The function f is differentiable at x if the
partial derivatives exist and are continuous. It should be noted that it is not sufficient
just for the partial derivatives to exist.

Exercises

1. Find all partial derivatives of the given functions:

a) f ( x, y) = x 2 y 3 b) f ( x, y, z) = x 2 yz + z cos( xy)

x 3 sin( e x1 )
c) g ( x1 , x 2 , x 3 ) = x1 x 2 x 3 + x 2 d) h( x1 , x 2 , x 3 , x 4 ) =
x2 + x 4

 1 3 2
2. Find the derivative of the linear function whose matrix is  .
 −2 7 0

3. What is the derivative a linear function whose matrix is A ?

7.6
4. Find the derivative of R (t ) = cos ti + sin tj + t k .

5. Find the derivative of f ( x, y) = x 2 y 3 .

6. Find the derivative of

 x1 x3 + e x 2 
x3 log(x1 + x 22 )
f (x1 , x2, x3 ) =  .
x2
 
 x1 x 3 + 5 
2

7.3 The Chain Rule

Recall from elementary one dimensional calculus that if a function is differentiable
at a point, it is also continuous there. The same is true here in the more general setting of

functions f : R n → R p . Let’s see why this is so. Suppose f is differentiable at a with

derivative L. Let h = x − a . Then lim f ( x) = lim f ( a + h) . Now,

x →a h→0

 f (a + h) − f (a ) − L (h) 
f ( a + h) − f (a ) = | h|   − L (h)
 | h| 

Now look at the limit of this as | h| → 0 :

 f ( a + h) − f (a ) − L (h) 
lim   =0
h→0
 | h| 

7.7
because f is differentiable at a, and lim L( h) = L (0) = 0 because the linear function L is
h→0

continuous. Thus lim( f (a + h) − f (a )) = 0 , or lim f (a + h) = f ( a) , which means f is

h→0 h→0

continuous at a.
Next, let’s see what the celebrated chain rule looks like in higher dimensions. Let

f : R n → R p and g: R p → R q . Suppose the derivative of f at a is L and the derivative of

g at f ( a) is M. We go on a quest for the derivative of the composition g o f : R n → R q

at a . Let r = g o f , and look at r(a + h) − r(a ) = g( f (a + h)) − g( f ( a)) . Next, let

k = f (a + h) − f (a ) . Then we may write

r(a + h) − r(a ) − ML (h) = g ( f (a + h)) − g ( f (a )) − ML (h)

= g ( f (a ) + k ) − g ( f (a )) − M ( k ) + M ( k) − ML( h) .
= g ( f (a ) + k ) − g ( f (a )) − M ( k ) + M ( k − L( h))

Thus,
r(a + h) − r(a ) − ML (h) g( f ( a) + k ) − g( f ( a)) − M( k ) k − L (h )
= + M( )
| h| | h| | h|

Now we are ready to see what happens as | h| → 0 . look at the second term first:

k − L (h )  f ( a + h) − f (a ) − L (h)   f (a + h) − f (a ) − L ( h) 
lim M( ) = lim M   = M (lim )
h→0 | h| h→0  | h|  h→0  | h| 
= M (0) = 0

since L is the derivative of f at a and M is linear, and hence continuous.

Now we need to see what happens to the term

 g ( f (a) + k ) − g( f ( a)) − M ( k ) 
lim  .
h→0  | h| 

7.8
This is a bit tricky. Note first that because f is differentiable at a , we know that

| k| | f ( a + h) − f (a)|
=
| h| | h|

behaves nicely as | h| → 0 . Next,

 g ( f (a ) + k ) − g ( f (a )) − M ( k ) | k |
lim  ⋅ 
h→0  | h| | k |
 g ( f (a ) + k ) − g ( f (a )) − M ( k )  | k|
= lim   =0
h→0  | k|  | h|

| k|
since the derivative of g at f ( a) is M, and is well-behaved. Finally at last, we have
| h|

shown that

 r (a + h) − r (a) − ML( h) 
lim  = 0,
h→0  | h| 

which means the derivative of the composition r = g o f is simply the composition, or

matrix product, of the derivatives. What could be more pleasing from an esthetic point of
view!

Example

Let f ( t ) = (t 2 ,1 + t 3 ) and g ( x1 , x 2 ) = ( 2x 1 − x2 ) 3 , and let r = g o f . First, we

shall find the derivative of r at t = 2 using the Chain Rule. The derivative of f is

7.9
  2t 
L =  2,
 3t 

and the derivative of g is

[
M = 6( 2 x1 − x2 ) 2 −3(2 x1 − x 2 ) 2 .]

4
At t = 2 , L =   ; and at g ( f (2 )) = g (4 ,9 ) , M = [6 −3]. Thus the derivative of the
12 

4
composition is ML = [6 − 3]  = [ − 12] = − 12 .
12
Now for fun, let’s find an explicit recipe for r and differentiate:

r(t ) = g( f ( t )) = g( t 2 ,1 + t 3 ) = (2t 2 − 1 − t 3 ) 3 . Thus r'( t ) = 3( 2t 2 − 1 − t 3 ) 2 (4 t − 3t 2 ) ,

and so r'( 2) = 3(1)(8 − 12 ) = − 12. It is, of course, very comforting to get the same answer

as before.
There are several different notations for the matrix of the derivative of

f : R n → R p at x ∈ R n The most usual is simply f '( x ) .

Exercises

7. Let g ( x1 , x 2 , x 3 ) = ( x1 x3 , x 2 x 3 + 1) and f ( x1 , x 2 ) = ( x1 x2 sin x 1 , x 1 + 3x 2 , x 2 − 2 x 12 ) .

Find the derivative of g o f at (2,-4).

8. Let u( x , y , z) = ( x + y 2 ,2 xy, x sin y , x 3 y 2 ) and v( r, s, t , q ) = (r + s − q 3 ,( r − t )e s ) .

a)Which, if either, of the composition functions u o v or v o u is defined? Explain.

b)Find the derivative of your answer to part a).

9. Let f ( x, y) = (e ( x + y ) , e ( x− y ) ) and g ( x , y ) = ( x − y 3 , x 2 + y ) .

7.10
 a)Find the derivative of f o g at the point (1,-2).

b)Find the derivative of g o f at the point (1,-2).

c) Find the derivative of f o f at the point (1,-2).

d) Find the derivative of g o g at the point (1,-2).

∂r ∂r
10. Suppose r = t 2 cos t and t = x 2 − 3y 2 . Find the partial derivatives and .
∂x ∂y

7.4 More Chain Rule Stuff

In the everyday cruel world, we seldom compute the derivative of the
composition of two functions by explicitly multiplying the two derivative matrices.

Suppose, as usual, we have r = g o f : R n → R q . The the derivative is, as we now know,

 ∂ r1 ∂ r1 ∂ r1 
∂ x L
∂ x2 ∂ xn 
 1 
 ∂ r2 ∂ r2
L
∂ r2 
r'( x) = r '( x1 , x 2 ,K , x n ) =  ∂ x1 ∂ x2 ∂ xn  .
 M 
∂ r
 p ∂ rp ∂ rp 
L
 ∂ x1 ∂ x2 ∂ x n 

We can thus find the derivative using the Chain Rule only in the very special case in

which the compsite function is real valued. Specifically, suppose g: R p → R and

f : Rn → R p . Let r = go f . Then r is simply a real-valued function of

x = ( x1 , x 2 ,K , x n ) . Let’s use the Chain Rule to find the partial derivatives.

7.11
  ∂ f1 ∂ f1 ∂ f1 
∂ x L
∂ x2 ∂ xn 
 1 
∂f ∂ f2 ∂ f2 
∂r ∂r ∂r  ∂ g ∂g ∂ g  2 L
r'( x) =  L = L  ∂ x ∂ x2 ∂ xn 
 ∂ x1 ∂ x2 ∂ xn   ∂ y1 ∂ y2 ∂ y p   M 1 
∂ f
 p ∂ fp ∂ f p 
L
 ∂ x1 ∂ x2 ∂ x n 

Thus makes it clear that

∂r ∂ g ∂ f1 ∂ g ∂ f 2 ∂ g ∂ fp
= + + L+ .
∂ x j ∂ y1 ∂ x j ∂ y 2 ∂ x j ∂ yp ∂ xj

Frequently, engineers and other malefactors do not use a different name for the
composition g o f , and simply use the name g to denote both the composition

g o f ( x 1 , x 2 , K, x n ) = g ( f 1 ( x1 , x 2 ,K , xn ), f 2 ( x1 , x 2 ,K , xn ),K, f p ( x1 , x 2 ,K, x n )) and the

function g given by g ( y ) = g( y1 , y 2 ,K , y p ) . Since y j = f j ( x1 , x 2 ,K , xn ) , these same

folks also frequently just use y j to denote the function f j . The Chain Rule given above

then looks even nicer:

∂g ∂ g ∂ y1 ∂ g ∂ y 2 ∂ g ∂ yp
= + +L+ .
∂ x j ∂ y1 ∂ x j ∂ y 2 ∂ x j ∂ yp ∂ xj

Example

Suppose g ( x , y , z) = x 2 y + yez and x = s + t , y = st 3 , and z = s 2 + 3t 2 . Let us

∂g ∂g
find the partial derivatives and . We know that
∂r ∂t

7.12
 ∂ g ∂ g ∂ x ∂ g ∂ y ∂ g ∂z
= + +
∂ s ∂x ∂ s ∂y ∂ s ∂z ∂ s
= 2xy(1) +(x 2 + ez )t3 + ye z (2s)
= 2xy +(x 2 + e z )t 3 + 2syez

Similarly,

∂ g ∂g ∂ x ∂ g ∂y ∂ g ∂ z
= + +
∂ t ∂x ∂ t ∂y ∂ t ∂z ∂ t
= 2xy(1) + (x 2 + ez )3st 2 + yez (6t)
= 2xy + 3(x 2 + ez )st 2 + 6tyez

These notational shortcuts are fine and everyone uses them; you should, however,
be aware that it is a practice sometimes fraught with peril. Suppose, for instance, you

have g ( x , y , z) = x 2 + y 2 + z 2 , and x = t + z , y = t 2 + 2 z , and z = t 3 . Now it is not at all

∂g
clear what is meant by the symbol . Meditate on this.
∂z

Exercises

∂g ∂g
11. Suppose g ( x , y ) = f ( x − y , y − s) . Find + .
∂x ∂y

12. Suppose the temperature T at the point ( x , y , z) in space is given by the function

T( x , y , z) = x 2 + xyz − zy 2 . Find the derivative with respect to t of a particle moving

along the curve described by r(t ) = cos ti + sin tj + 3tk .

7.13
13. Suppose the temperature T at the point ( x , y , z) in space is given by the function

T( x , y , z) = x 2 + y 2 + z 2 . A particle moves along the curve described by

r(t ) = sin π ti + cos πt j + (t 2 − 2t + 2) k . Find the coldest point on the trajectory.

14. Let r( x , y ) = f ( x ) g ( y ) , and suppose x = t and y = t . Use the Chain Rule to find

dr
.
dt

7.14
 Chapter Eight

f : Rn → R

8.1 Introduction
We shall now turn our attention to the very important special case of functions that
are real, or scalar, valued. These are sometimes called scalar fields. In the very, but
important, special subcase in which the dimension of the domain space is 2, we can

actually look at the graph of a function. Specifically, suppose f : R n → R . The

collection S = {( x 1 , x 2 , x 3 ) ∈ R 3 : f ( x1 , x 2 ) = x 3 } is called the graph of f. If f is a

reasonably nice function, then S is what we call a surface. We shall see more of this later.

Let us now return to the general case of a function f : R n → R . The derivative of f is just

∂f ∂f ∂f 
a row vector f '( x) =  L
∂ x n 
. It is frequently called the gradient of f
∂ x 1 ∂ x2

and denoted grad f or ∇f .

8.2 The Directional Derivative

In the applications of scalar fields it is of interest to talk of the rate of change of the
function in a specified direction. Suppose, for instance, the function T( x , y , z) gives the

temperature at points ( x , y , z) in space, and we might want to know the rate at which the

temperature changes as we move in a specified direction. Let f : R n → R , let a ∈ R n ,

and let u ∈ R n be a vector such that | u| = 1. Then the directional derivative of f at a in

the direction of the vector u is defined to be

d
Du f ( a) = f (a + tu) t =0 .
dt

8.1
 Now that we are experts on the Chain Rule, we know at once how to compute such a
thing. It is simply

d
D u f (a ) = f ( a + t u) t =0 = ∇f ⋅ u .
dt

Example

The surface of a mountain is the graph of f ( x, y) = 700 − x 2 − 5 y 2 . In other words, at

the point (x, y), the height is f (x, y). The positive y-axis points North, and, of course,
then the positive x-axis points East. You are on the mountain side above the point (2, 4)
and begin to walk Southeast. What is the slope of the path at the starting point? Are you
going uphill or downhill? (Which!?).
The answers to these questions call for the directional derivative. We know we are at
the point a = (2 ,4 ) , but we need a unit vector u in the direction we are walking. This is,

1
of course, just u = (1,− 1) . Next we compute the gradient ∇f ( x , y ) = [ −2 x ,− 10 y] . At
2

the point a this becomes ∇f ( 2,4) = [ −2,− 40] , and at last we have

−2 + 40 38
∇f ⋅ u = = . This gives us the slope of the path; it is positive so we are going
2 2

uphill. Can you tell in which direction the path will be level?

Another Example

The temperature in space is given by T( x , y , z) = x 2 y + yz 3 . From the point (1,1,1), in

which direction does the temperature increase most rapidly?

We clearly need the direction in which the directional derivative is largest. The

directional derivative is simply ∇T ⋅ u =|∇ T|cosθ , where θ is the angle between ∇T and

u. Anyone can see that this will be largest when θ = 0. Thus T in creases most rapidly in

8.2
the direction of the gradient of T. Here that direction is [2 xy, x 2 + z 3 ,3 yz 2 ] . At (1,1,1),

this becomes [2, 2, 3].

Exercises

1. Find the derivative of f ( x, y, z) = x log z + 2 xy at (1, 2, 1) in the direction of the

vector [1, 2 ,2] .

2. Find the derivative of f ( x, y, z) = x cos y + 3z 3 − xz at (1, π, 1) in the direction of the

vector [3, - 2 ,2] .

3. Find the directions in which g ( x , y ) = x 2 y + e xy sin y increases and decreases most

rapidly from the point (1, 0).

4. The surface of a hill is the graph of the equation z = 1000 + x 2 − x 4 − y 2 . You stand

on the hill above the point (5,3) and pour out a glass of water. In which direct will it
begin to run? Explain.

5. The position of a particle at time t is given by r(t ) = 3( t 2 − sin t ) i + t j − cos tk , and

the position of another particle is R (t ) = t 2 i + (t 3 + t ) j + sin t k . At time t = π, what

is the rate of change of the distance between the two particles? Are they getting
closer to one another, or are they getting farther apart? (Which!) Explain.

8.3 Surface Normals

8.3
 Let f : R 3 → R be a function and let c be some constant. Recall that the set

S = {( x , y , z) ∈ R 3 : f ( x , y , z) = c} is called a level set, or level surface, of the function f .

Suppose r(t ) = x (t )i + y( t ) j + z(t )k describes a curve in R 3 that lies on the surface S.

This means, of course, that f ( r( t )) = f ( x (t ), y ( t ), z( t )) = c . Now look at the derivative

with respect to t of this equation:

d
f ( r(t )) = ∇f ⋅ r'( t ) = 0 .
dt

In other words, the gradient of f and the tangent to the curve are perpendicular. Note there
was nothing special about our choice of r(t); it is any curve on the surface. The gradient
∇f is thus perpendicular, or normal to the surface f ( x, y, z) = c .

Example
Suppose we want to find an equation of the plane tangent to the surface

x 2 + 3y 2 + 2z 2 = 12

at the point (1, -1, 2). For an equation of a plane, we need a point a on the plane and a
vector N normal to the plane. Then the equation we seek is simply N ⋅ ( x − a) = 0 ,

where x = ( x , y , z) . In the case at hand, we have a point on the plane: a = (1, -1, 2).

Let’s find a normal vector N. We have just learned that the gradient of

f ( x, y, z) = x 2 + 3 y 2 + 2 z2 does the job.

∇f ( x , y , z) = [2 x ,6 y ,4 z] ,

8.4
and so N = ∇ f (1,− 12
, ) = [2 ,−68
, ] . The tangent plane is thus given by the equation

N ⋅ ( x − a) = 0 , which in this case is

2( x − 1) − 6( y + 1) + 8( z − 2) = 0 .

You should note that the discussion here didn’t depend on the dimension of the

domain. Thus if f : R 2 → R , then the set {( x , y ) ∈ R 2 : f ( x , y ) = c} is a level curve of f,

and the gradient of f is normal to such a curve.

Combining these results with what we know about the directional derivative, we see
that at a point the value of a function increases most rapidly in a direction normal to the
level set passing through that point. On a contour map of a portion of the Earth’s
surface, for example, the steepest path is in the direction normal to the contour lines.

Exercises

6. Find an equation for the plane tangent to the surface z = x 2 + 2 y 2 at the point (1,1,3).

7. Find an equation for the plane tangent to the surface z = log( x + y ) at the point
2 2

(10
, ,0) .

8. Find an equation for the plane tangent to the surface cosπ x − x 2 y + e xz + yz = 4 at

the point (0,1,2).

9. Find an equation of the straight line tangent to the curve of intersection of the surfaces

x 3 + 3x 2 y 2 + y 3 + 4 xy − z 2 = 0 and x 2 + y 2 + z 2 = 11 at the point (1, 1, 3).

8.5
8.4 Maxima and Minima

Let f : R n → R . A point a in the domain of f is called a local minimum if there is an

open ball B (a ; r ) centered at a such that f ( x) − f (a) ≥ 0 for all x ∈ B( a; r ) . If f is a nice

function, then this means the directional derivative Du f ( a) ≥ 0 for all unit vectors u. In

other words, ∇f (a ) ⋅ u ≥ 0 . Then it must be true that both ∇f (a ) ⋅ u ≥ 0 and

−∇f (a) ⋅ u = ∇ f ( a) ⋅ (− u) ≥ 0 . This can be so for every u only of ∇f (a ) = 0 . Thus f has

a local minimum at a point at which it has a derivative only if the derivative is zero there.
You should guess the definition of a local maximum and see why it must be true that
the gradient is zero at such a point. Thus if a is a local minimum or a local maximum of f,
and if f has a derivative at a, then the derivative ∇f (a ) = 0. You should be aware of the

fact that here, just as in Mrs. Turner’s elementary calculus class, the converse is not
necessarily true. We may have ∇f (a ) = 0 without a being either a local minimum or a

local maximum.

Example
Let us find all local maxima and local minima of the function

f ( x , y ) = x 2 + xy + y 2 + 3x − 3y + 4 .

Meditate on just how should proceed. This function clearly has a derivative everywhere,
so at any local maximum or minimum, this derivative, or gradient, must be zero. So let’s
begin by finding all points at which ∇f (a ) = 0 . In other words, we want (x, y) at which

∂f ∂f
= 0 and = 0:
∂x ∂y

8.6
 ∂f
= 2x + y + 3 = 0
∂x

∂f
= x + 2y − 3 = 0
∂y

We are thus faced with the border-line trivial problem of solving the system of equations

2 x + y = −3
.
x + 2y = 3

There is just one solution: ( x , y ) = (−3, 3) . Now let us reflect on what we have here.

What we have actually found is all the points that cannot possibly be local minima or
maxima. These are all points except (-3, 3).. All we know right now is that this point is
the only possible candidate. Let’s find out what we have by the hammer and tongs
method of examining the quantity f (− 3 + x , 3 + y ) − f ( − 3,3 ) :

f ( − 3 + x , 3 + y ) − f ( − 3,3 ) = f ( − 3 + x , 3 + y ) − ( − 5)
= ( − 3 + x ) 2 + ( − 3 + x)( 3 + y) + ( 3 + y) 2 + 3( − 3 + x) − 3( 3 + y) + 9
2
 y 3y 2
= x 2 + xy + y 2 =  x +  +
 2 4

It is therefore clear that f (− 3 + x , 3 + y ) − f ( − 3,3 ) ≥ 0 , which means that (-3, 3) is a local

minimum.

Exercises

In each of the following, find all local maxima and minima:

8.7
10. f ( x , y ) = x 2 + 3xy + 3y 2 − 6 x + 3 y − 6

11. f ( x, y) = x 2 + xy + 3x + 2 y + 5

12. f ( x, y) = 2 xy − 5x 2 − 2 y 2 + 4x − 4

13. f ( x, y) = x 2 + 2 xy

14. f ( x, y) = y − x 2

8.5 Least Squares

We shall next look at some very simple, yet important, applications in which the
location of a minimum value of a function is sought.
Suppose we have a set of n points in the plane, say ( x1 , y1 ),( x 2 , y2 ),K,( x n , y n ) ,

and we seek the straight line that "best" fits this collection of points. We first decide
what we mean by "best". Let's say we mean the line that minimizes the sum of the
squares of the vertical distances from the points to the line. We can describe all
nonvertical lines in the world by means of two variables, traditionally called m and b.
Thus every such line has the form y = mx + b . Our quest is thus for the values of m and

b at which the function

n
f ( m, b) = ∑ (mxi + b − y i ) 2
i =1

has its minimum value. Knowing these values will give us our line.
We simply apply our vast and growing knowledge of calculus and find where the
gradient of f is 0:

8.8
 ∂f ∂f
∇f = ( , ) =0 .
∂m ∂ b

Now,
∂f n n n n
= ∑ 2 x i (mxi + b − yi ) = 2[m∑ x i2 + b∑ x i − ∑ xi y i ], and
∂ m i =1 i =1 i= 1 i =1

∂f n n n
= ∑ 2( mxi + b − yi ) = 2[m∑ xi + nb − ∑ yi ].
∂ b i =1 i =1 i =1

We are thus faced with solving the 2 x 2 linear system

n n n
m∑ x i2 + b ∑ xi = ∑ x i y i
i= 1 i =1 i =1
n n
m∑ x i + bn = ∑ yi
i= 1 i= 1

Meditate sufficiently to convince yourself that there is always exactly one

solution to this system, and continue meditating sufficiently to convince yourself that
there must be an honest-to-goodness minimum of the original function at this solution.
Let's have a go at an example. Suppose we have the following table of values:

x y

0 1

1 2

2 4

3 3.5

4 5

8.9
 5 4

7 7

8 9

9 12

10 18

12 21

15 29

The linear system for m and b is

718m + 76b = 1156.5

76m +12b = 115.5

255 993
Solving this system gives us m = and b = − . In other words, the line that best
142 568
fits the data in the “sense of least squares” is

255 993
y= x−
142 568

Here’s a picture of this line together with the data points:

8.10
Looks pretty good!

Exercises

15. Here is a table of Köchel numbers versus year of composition for the compositions of
W. A. Mozart. Find the "least squares" straight line approximation to this table and
use it to estimate the year in which Mozart's Sinfonia Concertante in E-flat major was
composed.

Köchel Year
Number composed

1 1761

75 1771

155 1772

219 1775

271 1777

351 1780

425 1783

503 1786

575 1789

626 1791

[This problem is taken from Calculus and Analytic Geometry (8th Edition), by
Thomas & Finney.]

8.11
16. Find some data somewhere (The Statistical Abstract of the United States is a good
source of interesting data.), find the least squares linear approximation to the data, and
say something intelligent about your results.

8.6 More Maxima and Minima

In real life, one is most likely interested in finding the places at which the largest
and smallest values of a function f : D → R occur, rather than in simply finding local

maxima and minima. (Here D is a subset of R n .). To begin, let's think a moment about
how we can tell if there is a maximum or minimum value of f on D. First, we suppose
that f is continuous—otherwise, anything can happen! Next, what properties of D will
insure the existence of a biggest and smallest value of f ? The answer is fairly simple.

Certainly D must be a closed subset of R n ; consider, for example the function

, ) → R given simply by f ( x) = x , which has neither a maximum nor a minimum on
f :( 01

D = (01
, ) . Having the domain be closed, however, is not sufficient to guarantee the

existence of a maximum and minimum. Consider, for example f : R → R again with

, ) → R given by f ( x) = x . The domain R is certainly closed, but f has neither a

f :( 01

maximum nor a minimum. We need also to have the domain be bounded. It turns out that
for continuous f , if the domain D is both closed and bounded, then there must necessarily
be a maximum and a minimum value for f on D. Let's think a moment about what the
candidates for such points are. If the biggest or smallest value of f occurs in the interior of
D, then surely the point at which it occurs is a local maximum (or minimum). If f has a
gradient there, then the gradient must be 0 . The points at which the largest or smallest
values occur must therefore be either i)points in the interior of D at which the gradient of f
vanishes, ii)points in the interior at which the gradient of f does not exist, or iii)points in
D but not in the interior of D (that is, points on the boundary of D).
Hark back to Mrs. Turner's third grade calculus class. How did you find the
maximum value of a function f whose domain D is a closed interval [a , b] ⊂ R ? Recall

8.12
found all points in the interior (that is, in the open interval (a,b)) at which the derivative
vanishes. You then simply evaluated f at these points, evaluated f at any points in (a,b)
at which there is no derivative, evaluated f at the two end points of the interval (in this
one dimensional case, the boundary of D is particularly simple.), and then picked out the
biggest and smallest numbers you computed. The situation in higher dimensions is a bit
more complicated, mostly because the boundary of even a nice domain D is not a nice
finite set as in the case of an interval, but is an infinite set. Let's look at an example.

Example

A flat circular plate has the shape of the region {( x , y ) ∈ R 2 : x 2 + y 2 ≤ 1} . The

temperature at the point ( x , y ) on the plate is given by T( x , y ) = x 2 + 2 y 2 − x . Our

assignment is to find the hottest and coldest points on the plate. According to our
previous discussion, candidates for the hottest and coldest points are all points inside the
circular boundary at which the gradient of T is 0 and all points on the boundary. (Note
that T has a gradient at all points inside the circle.) First, let's find where among all points

( x , y ) such that x 2 + y 2 < 1 , the ones at which ∇T = (2 x − 14

, y) = 0 . This is easy; it

1
should be clear there is just one such point: ( ,0) . Now for the more difficult part,
2
finding the candidates on the boundary. Note that the boundary may be described by the
vector equation
r(t ) = cos t i + sin t j , where 0 ≤ t ≤ 2π .

The temperature on this set is then given by

T( t ) = T ( r(t )), 0 ≤ t ≤ 2π

[Here we are abusing the notation, as we have done before, by using the same name for
the function T( x , y ) and the composition T( r(t )) .] We are now faced with the one

dimensional problem of finding the maximum and minimum values of a nice differentiable
function of one variable on a closed interval. First, we know the endpoints of the interval
are candidates: t = 0, and t = 2π . We have at this point added one more point to our list

8.13
of candidates: r(0) = r(2π ) = (10
, ) . Now for candidates inside the interval, we seek

dT
places at which the derivative = 0 . From the Chain Rule, we know
dt
dT
= ∇ T( r(t )) ⋅ r'( t ) = (2 cos t − 14
, sin t ) ⋅ (− sin t ,cos t ) = 2 cos t sin t + sin t .
dt
dT
The equation = 0 now becomes
dt
2 cos t sin t + sin t = 0, or
sin t ( 2 cos t + 1) = 0

1
Thus sint = 0 , or 2 cos t + 1 = 0. We have, in other words, y = 0 , or x = − . When
2

1 3 3
y = 0 , then x = 1 or x = −1; and when x = − , then y = or y = − . Thus our
2 2 2

1 3 1 3
new candidates are (1,0), (− 10
, ), (- , ), and (− ,− ) . These together with the one
2 2 2 2
1
we have already found, ( ,0) , make up our entire list of possibilities for the hottest and
2
coldest points on the plate. All we need do now is to compute the temperature at each of
these points:
1 1 1 1
T( ,0) = − = − .
2 4 2 4
T(10, ) = 1− 1 = 0
T( −10, ) = 1+ 1 = 2
1 3 1 3 1 3 1 9
T( − , ) = T( − ,− ) = + + =
2 2 2 2 4 2 2 4

1
Finally, we have our answer. The coldest point is ( ,0) , and the hottest points are
2

1 3 1 3
(− , ) and (− ,− ).
2 2 2 2

8.14
Exercises

17. Find the maximum and minimum value of f ( x, y) = x 2 − xy + y 2 + 4 on the closed

area in the first quadrant bounded by the triangle formed by the lines x = 0 , y = 4 ,

and y = x .

18. Find the maximum and minimum values of f ( x , y ) = (4 y − y 2 )cos x on the closed

π π
area bounded by the rectangle 1 ≤ y ≤ 3 , − ≤x≤ .
4 4

8.7 Even More Maxima and Minima

It should be clear now that the really troublesome part of finding maxima and
minima is in dealing with the constrained problem; that is, the problem of finding the
maxima and minima of a given function on a set of lower dimension than the domain of the
function. In the problems of the previous section, we were fortunate in that it was easy
to find parametric representations of the these sets; in general, this, of course, could be
quite difficult. Let's see what we might do about this difficulty.
Suppose we are faced with the problem of finding the maximum or minimum value

of the function f : D → R , where D = {( x, y) ∈ R 2 : g ( x , y ) = 0} , where g is a nice

function. (In other words, D is a level curve of g .) Suppose r(t ) is a vector description

of the curve D. Now then, we are seeking a maximum or minimum of the function
dF
F (t ) = f ( r( t )) . At a maximum or minimum, we must have = 0 . (Here g is
dt
sufficiently nice to insure that g ( x , y ) = 0 is a closed curve, and so there are no endpoints

dF
to worry about.) The Chain Rule tells us that = ∇f ⋅ r' = 0 . Thus at a maximum or
dt
minimum, the gradient of f must be perpendicular to the tangent to g ( x , y ) = 0 . But if

∇f is perpendicular to the tangent to the level curve g ( x , y ) = 0 , then it must have the

8.15
same direction as the normal to this curve. This is just what we need to know, for the
gradient of g is normal to this curve. Thus at a maximum or minimum, ∇f and ∇g must

"line up". Thus ∇f = λ ∇ g , and there is no need actually to know a vector representation

r for g ( x , y ) = 0 .

Let's see this idea in action. Suppose we wish to find the largest and smallest

values of f ( x, y) = x 2 + y 2 on the curve x 2 − 2 x + y 2 − 4 y = 0 .

Here, we may take g ( x , y ) = x 2 − 2 x + y 2 − 4 y . Then ∇f = 2x i + 2 y j , and

∇g = (2 x − 2) i + (2 y − 4) j , and our equation ∇f = λ ∇ g becomes

2x = λ (2 x − 2)
2 y = λ ( 2 y − 4)

We obtain a third equation from the requirement that the point ( x , y ) be on the curve

g ( x , y ) = 0 . In other words, we need to find all solutions to the system of equations

2 x = λ (2 x − 2)
2 y = λ (2 y − 4)
x 2 − 2 x + y2 − 4 y = 0

The first two equations become

x ( λ - 1) = λ
y (λ − 1) = 2λ

λ 2λ
Thus x = and y = . (What about the possibility that λ − 1 = 0 ?). The last
λ −1 λ −1

λ2 2λ 4λ 2 8λ
equation then becomes − + − = 0 ; or,
(λ − 1) 2
λ − 1 (λ − 1) 2
λ −1

λ2 − 2λ (λ − 1) = 0,
λ2 − 2λ = 0
We have two solutions: λ = 0 and λ = 2 . What do you make of the solution λ = 0 ?

These values of λ give us two candidates for places at which extrema occur: x = 0 and

y = 0 ; and x = 2 and y = 4 . Now then f ( 00

, ) = 0 , and f ( 2,4) = 4 + 16 = 20 . There

8.16
we have them—the minimum value is 0 and it occurs at (0,0); and the maximum value is
20, and it occurs at (2,4).
This method for finding "constrained" extrema is generally called the method of

Lagrange Multipliers. (The variable λ is called a Lagrange multiplier.)

Exercises

19. Use the method of Lagrange multipliers to find the largest and smallest values of

f ( x, y) = 4 x + 3y on the circle x 2 + y 2 = 1 .

20. Find the points on the ellipse x 2 + 2 y 2 = 1 at which f ( x, y) = xy has its extreme

values.

21. Find the points on the curve x 2 + xy + y 2 = 1 that are nearest to and farthest from the

origin.

8.17
 Chapter Nine

The Taylor Polynomial

9.1 Introduction
Let f be a function and let F be a collection of "nice" functions. The approximation
problem is simply to find a function g ∈ F that is "close" to the given function f . There are
two issues immediately. How is the collection F selected, and what do we mean by
"close"? The answers depend on the problem at hand. Presumably we want to do
something to f that is difficult or impossible (This might be something as simple as finding
f ( x ) for some x.). The collection F would thus consist of functions to which it is easy to
do that which we wish to do to f . Our measure of how close one function is to another
would try to reflect the closeness of the results of our operations. Now, what are we
talking about here. Suppose, for example, we wish to find f ( x ) . Our collection F of
functions should include functions that are easy to evaluate at x , and two function would
be "close" simply if there values are close. We might, for instance, want to evaluate sin x
for all x is some interval I. The collection F could be a collection of second degree
polynomials. The approximation problem is then to find elements of F that make the
"distance" max{|sin x − p (x )|: x ∈ I } as small as possible. Similarly, we might want to find
the integral of some function f over an interval I . Here we would want F to consist of
functions easily integrated and measure the distance between functions by the difference of
their integrals over I . In the previous chapter, we found the "best" straight line
approximation to a set of data points. In that case, the collection F consisted of all
nonvertical straight lines, and we measured the distance between functions by the sum of
the squares of their differences on a specified set of points { x1 , x2 ,K , x n } . You can
imagine many other examples.

9.2 The Taylor Polynomial

We look first at a simple but useful problem: Given a nice function f : D ⊂ R → R , a
point a in the interior of the domain D , and an integer n , find a polynomial p of degree
≤ n such that

9.1
 p (a ) = f (a )
p '( a ) = f '( a )
p ''( a ) = f ' ' ( a )
M
p ( n ) (a ) = f ( n)
(a )

We solve the problem by the Behold Method. Simply verify that

f ''( a ) f '''( a ) f ( n ) (a )
p ( x ) = f ( a ) + f '( a )( x − a ) + ( x − a) 2 + ( x − a ) 3 +K+ (x − a ) n
2! 3! n!

does the job! It is also fairly easy to see that this polynomial is the only polynomial of
degree ≤ n that does the job. Suppose q is also a polynomial with degree g ≤ n such that
p (a ) = f (a )
p '( a ) = f '( a )
p ''( a ) = f ' ' ( a )
M
p ( n ) (a ) = f ( n)
(a )
and consider the function r = p − q . Note that r is also a polynomial of degree ≤ n . But

r ( a ) = r '( a ) = r ''( a ) =K = r ( n ) ( a ) = 0 .
Or, in other words, r has a zero of order n + 1, and the only way this can happen is if
r ( x ) ≡ 0 for all x . That is, p ( x ) ≡ q ( x ) identically.

Example
Let f (x ) = sin x and let a = 0 . Let's find the Taylor polynomial for a few different
values of n. For n = 1, we have simply p1 ( x) = f (a ) + f '( a )( x − a ) = sin 0 + cos 0( x) = x .

Note that for n = 2, we have p 2 ( x) = sin 0 + cos0 (x ) − sin 0( x 2 ) = x , also. Let's take a look

x3
at the next Taylor polynomial. Here p 3 (x ) = x − . Let's draw some pictures; we'll look
6
at the graph of p3 and f . We shall use Maple.

9.2
What we see is that the Taylor polynomial looks like a pretty good approximation as long
as we don't get too far away from a = 0. Let us continue. Convince yourself that p 4 = p 3 ,

x3 x5
and p 5 ( x) = x − + . Another picture:
6 120

9.3
Exercises

1 . Find the Taylor polynomial of degree n for f (x ) = e x at a = 0.

2 . Find the Taylor polynomial of degree n for f (x ) = x 3 at a = 1.

3 . Find the Taylor polynomial of degree 3 for f (x ) = log x at a = 1.

4 . Find the Taylor polynomial of degree n for f (x ) = sin x at a = 0.

5 . Find the Taylor polynomial of degree 3 for f (x ) = x at a = 4.

9.3 Error
Let's see how close the Taylor polynomial is to the function f . To do this, suppose p is
the Taylor polynomial of degree ≤ n for the function f at a , and consider the function

( t − a ) n +1
g ( t ) = f ( t ) − p( t ) − ( f ( x ) − p ( x )) .
( x − a ) n +1
(We assume x ≠ a .) Note that g (a ) = g ( x) = 0 . Now, from the Mean Value Theorem (or
Rolle's Theorem, or whatever.) we know that g' (ξ 1 ) = 0 for some ξ 1 between a and x .

( n + 1)(a − a ) n
But note also that g ' ( a ) = f '( a ) − p '( a ) − ( f ( x ) − p ( x )) = 0 . It thus follows
( x − a ) n +1

from the Mean Value Theorem that the derivative of g' is zero at some ξ 2 between a and

( n + 1) n ( a − a ) n −1
ξ 1 . Also, g ' ' ( a ) = f ' ' (a ) − p' ' ( a ) − ( f ( x ) − p ( x )) = 0 . Once again, from
( x − a ) n +1

the celebrated Mean Value Theorem, we conclude that g'''( ξ 3 ) = 0 for some ξ 3 between a

and ξ 2 . Continuing in this fashion, we are finally able to conclude that g ( n +1) (ξ ) = 0 for
some ξ . Let's see what this looks like.

(n + 1)!
g ( n +1) (t ) = f ( n +1)
(t ) − p ( n +1 ) (t ) − ( f ( x) − p( x))
( x − a ) n +1

9.4
and so g ( n +1) (ξ ) = 0 becomes

(n +1) (n + 1)!
f (ξ ) − ( f ( x) − p (x )) = 0 .
( x − a ) n +1

(Remember, p is a polynomial of degree ≤ n , and so p ( n +1) (t ) ≡ 0 . From this we obtain an

expression for the difference between f and the Taylor polynomial g :

f ( n +1) (ξ )
f (x ) − p (x ) = ( x − a ) n +1 .
( n + 1)!

Example

Remember when in 7th grade physics class, Mr. Crews replaced the sine of a "small"
angle θ by θ itself ? He assured us that for small angles this was just fine. Well, what was
going on here? Let's see if our new-found knowledge of Taylor polynomials will help.
Observe that p(θ ) = θ is simply the Taylor polynomial of degree ≤ 2 for f (θ ) = sinθ at
a = 0 . Using the result just derived, we have that
− sin ξ 3
sinθ − θ = θ .
6
Now, we don't know what ξ is, but we do know that |sin ξ } ≤ 1 ; thus

θ3
|sin θ − θ| ≤ ,
6
and we have a precise estimate of the error incurred by substituting θ for sinθ . Suppose,

10 π
for example, that θ = 10 o ; then what? Well, θ = 2π = . Then the error we get when
360 18
π π
we use instead of sin is estimated by
18 18

π π 1θ 
3
sin − ≤   ≤ 0.008862.
18 18 6  18 

Now we know exactly what "pretty close" means. For 10 degrees, I guess that's "not too
bad."

Exercises

9.5
6 . a)Find the Taylor polynomial of degree ≤ 2 for f (x ) = e x at a =0.

b)Use the result of part a) to find an approximation for e .

c)Find as small an upper bound as you can for the difference between your
approximation found in part b) and e.

7 . Use the Taylor polynomial found in Exercise 3 to approximate log(.

11) and find an
upper bound for the magnitude of the difference between your approximation and
11) .
log(.

x3
8 . For what values of x can you replace sin x by x − with an error of magnitude no
6
greater than 3 × 10 −4 ?

9 . Calculate e with an error of less than 10-6 .

9.6
 Taylor’s Theorem

1. Introduction. Suppose f is a one-variable function that has n + 1 derivatives on an interval

about the point x = a. Then recall from Ms. Turner’s class the single variable version of Taylor’s
Theorem tells us that there is exactly one polynomial p of degree ² n such that pÝaÞ = fÝaÞ,
p v ÝaÞ = f v ÝaÞ, p vv ÝaÞ = f vv ÝaÞ, u, p ÝnÞ ÝaÞ = f ÝnÞ ÝaÞ. This polynomial is given by
f vv ÝaÞ ÝnÞ
pÝxÞ = fÝaÞ + f v ÝaÞÝx ? aÞ + Ýx ? aÞ 2 + u + f Ýx ? aÞ n
2! n!
We also know the difference between fÝxÞ and pÝxÞ:

f Ýn+1Þ ÝYÞ
fÝxÞ ? pÝxÞ = Ýx ? aÞ n+1 ,
Ýn + 1Þ!

where Y is somewhere between a and x.

The polynomial p is called the Taylor Polynomial of degree ² n for f at a.

Before we worry about what the Taylor polynomial might be in higher dimensions, we need to be
sure we understand what is a polynomial in more than one dimension. In two dimensions, a
polynomial pÝx, yÞ of degree ² n is a function of the form

i+j=n
pÝx, yÞ = > a ij x i y j.
i,j=0

Thus a polynomial of degree ² 2 (perhaps more commonly known as a quadratic) looks like

pÝx, yÞ = a 00 + a 10 x + a 01 y + +a 11 xy + a 20 x 2 + a 02 y 2 .

I hope it easy to guess what one means by a polynomial in three variables, Ýx, y, zÞ, or indeed, in
any number of variables.

Now, how might we extend the idea of the Taylor polynomial of degree ² n for a function f at a
point a ? Simple enough. It’s a polynomial pÝxÞ of degree ² n so that

/ i 1 +u+i q fÝaÞ / i 1 +u+i q pÝaÞ

iq
= i
,
/x 1 /x 2 u/x q
i1 i2
/x i11 /x i22 u/x qq

for all i 1 , i 2 , u, i q such that i 1 + i 2 + u + i q ² n.

This looks pretty ferocious in general, so let’s see what it says for just two variables. In this case,
we have a =Ýa, bÞ and the Taylor polynomial pÝx, yÞ at a becomes the polynomial such that

1
 / i+j fÝaÞ / i+j pÝaÞ
= ,
/ i x/ j y / i x/ j y

for all i + j ² n.

Example

2 y2
Let fÝx, yÞ = cosÝx + yÞ, and let pÝx, yÞ = 1 ? x2 ? xy ? 2
. Let’s verify that p is the Taylor
polynomial of degree ² 2 for f at Ý0, 0Þ. He we go.
fÝ0, 0Þ = 1, and pÝ0, 0Þ = 1;
/f = ? sinÝx + yÞ, and /p = ?x ? y;
/x /x
/f = ? sinÝx + yÞ, and /p = ?x ? y;
/y /y
/ 2 f = ? cosÝx + yÞ, and / p = ?1,
2

/x 2 /x 2
/ 2 f = ? cosÝx + yÞ, and / p = ?1,
2

/y 2 /y 2
/ 2 f = ? cosÝx + yÞ, and / p = ?1.
2

/x/y /x/y
Now it’s easy to see that

fÝ0, 0Þ = 0 = pÝ0, 0Þ;

/f Ý0, 0Þ = 0 = /p Ý0, 0Þ;
/x /x
/f Ý0, 0Þ = 0 = /p Ý0, 0Þ;
/y /y
/ 2 f Ý0, 0Þ = ?1 = / p Ý0, 0Þ;
2

/x 2 /x 2
/ 2 f Ý0, 0Þ = ?1 = / p Ý0, 0Þ; and
2

/y 2 /y 2
/ 2 f Ý0, 0Þ = ?1 = / p Ý0, 0Þ.
2

/x/y /x/y

Exercises

1. Verify that the polynomial in the Example is also the Taylor polynomial for f at (0,0) of degree
² 3.

2. Let fÝx, yÞ = sinÝx + yÞ.Which Which of the following is the Taylor polynomial of degree ² 2 for
f at (0,0)? Explain.
a) pÝx, yÞ = 1 + x 2 + y 2 b) pÝx, yÞ = xy

2
c) pÝx, yÞ = x 2 + xy + 2y d) pÝx, yÞ = x + y

2. Derivatives. Prior to finding a general recipe for the Taylor polynomial, we need look at finding
higher order derivatives of certain composite functions. Let f be a real-valued function defined on a
subset of R q . Suppose that in a neighborhood of the point x, the function f has a lot of continuous
partial derivatives. Define the function g by

gÝtÞ = fÝa + thÞ,

where a = Ýa 1 , a 2 , u, a q Þ and h = Ýh 1 , h 2 , u, h q Þ. We know from the chain rule that g v ÝtÞ is
given by

g v ÝtÞ = 4fÝa + thÞ 6 h

= /f , /f , u, /f 6 Ýh 1 , h 2 , u, h q Þ
/x 1 /x 2 /x q

= h1 / + h2 / + u + hq / f
/x 1 /x 2 /x q Ýa+thÞ

In keeping with our general practice of restricting ourselves to dimensions one, two, or three, let’s
look first at the case q = 2. As usual, we’ll write x =Ýx, yÞ and h = Ýh, kÞ. The expression for g v ÝtÞ
now looks like:

g v ÝtÞ = h / +k / f
/x /y Ýx+thÞ

We are now in business, for we have a nice recipe for higher order derivatives of g :

m
g ÝmÞ ÝtÞ = h / +k / f
/x /y Ýx+thÞ

For example,

2
g vv ÝtÞ = h / +k / f
/x /y
h 2 / 2 + 2hk / + k 2 / 2
2 2 2
= f
/x /x/y /y
= h 2 / 2f + 2hk / f + k 2 / 2f
2 2 2

/x /x/y /y
Example
Suppose fÝx, yÞ = x 2 y 3 + y 2 . Let’s find the second derivative of the function

gÝtÞ = fÝ1 + 3t, ?2 + tÞ

3
First,
2
g vv ÝtÞ = 3 / + / f
/x /y
= 9 / 2f + 6 / f + / 2f
2 2 2

/x /x/y /y
/f /f /2f /2f /2f
Now, /x
= 2xy 3 , and /y
= 3x 2 y 2 + 2y, and so /x 2
= 2y 3 , /y/x
= 6y 2 , and /y 2
= 6x 2 y + 2.
Thus,

g vv ÝtÞ = 18Ý?2 + tÞ 3 + 36Ý?2 + tÞ 2 + 6Ý1 + 3tÞ 2 Ý?2 + tÞ + 2

Exercises
3. Let fÝx, yÞ = xe y . Find the derivative of gÝtÞ = fÝ1 + t, 3 ? 4tÞ.

4. Find the second derivative of the function g defined in Problem 3.

5. Let FÝu, vÞ = u 3 v + v 2 . Find the second derivative of RÝzÞ = FÝz, 3zÞ.

6. Find g vvv ÝtÞ, where g is the function defined in the Example.

3. The Taylor polynomial. To find the Taylor polynomial for a function f of several variables at a
point a, we shall simply apply the one-dimensional results to the function

gÝtÞ = fÝa + thÞ.

Thus,
n ÝmÞ g Ýn+1Þ ÝYÞ n+1
gÝtÞ = > g m!Ý0Þ t m + Ýn + 1Þ!
t ,
m=0

where Y is a number between 0 and t. Next, substitute t = 1 into the above:

n
g ÝmÞ Ý0Þ g Ýn+1Þ ÝYÞ
gÝ1Þ = fÝaÞ = > +
m=0
m! Ýn + 1Þ!

We know the value of g ÝkÞ from Section 2:

n m
fÝa + hÞ = > m!
1 h1 / + h2 / + u + hq / fÝaÞ
m=0
/x 1 /x 2 /x q

4
 n+1
+ 1 h1 / + h2 / + u + hq / fÝcÞ
Ýn + 1! /x 1 /x 2 /x q

The point c lies somewhere on the line segment joining a and a + h.

The polynomial
n m
pÝhÞ =pÝh 1 , h 2 , u, h q Þ = > m!
1 h1 / + h2 / + u + hq / fÝaÞ
m=0
/x 1 /x 2 /x q

is the Taylor polynomial of degree ² n for f at a; the last term is traditionally called the error term
or sometimes, the remainder term. Actually, if we let h = x ? a, then qÝxÞ =pÝx ? aÞ is the thing
we called the Taylor polynomial in the first section.

This is pretty fierce looking. Let’s look at the two variable case:

n m
fÝa 1 + h, a 2 + kÞ = > 1 h / +k / fÝa 1 , a 2 Þ
m=0
m! /x /y
n+1
+ 1 h / +k / fÝc 1 , c 2 Þ
Ýn + 1!Þ /x /y

where Ýc 1 , c 2 Þ is on the line joining Ýa 1 , a 2 Þ and Ýa 1 + h, a 2 + kÞ.

Example

Let fÝx, yÞ = sin x sin y. For n = 2 and a = Ý0, 0Þ, Taylor’s polynomial becomes

pÝh, kÞ = fÝ0, 0Þ + h /f Ý0, 0Þ + k /f Ý0, 0Þ + h / 2f Ý0, 0Þ + hk / f Ý0, 0Þ + k / 2f Ý0, 0Þ

2 2 2 2 2

/x /y 2 /x /x/y 2 /y

We have
/f /f /2f /2f /2f
/x
= cos x sin y; /y
= sin x cos y; /x 2
= ? sin x sin y; /x/y
= cos xcosy; /y 2
= ? sin x sin y.
Thus,
pÝh, kÞ = hk.

Let’s get an estimate for how well this approximates sin x sin y near Ý0, 0Þ. We know that

3
|sin x sin y ? xy| = 1 x / +y / fÝY, WÞ
3! /x /y

where ÝY, WÞ is one the segment joining Ýx, yÞ and the origin. Now,

5
 3
x / +y / f = x 3 / 3f + 3x 2 y /2 f + 3xy 2 / f 2 + y 3 / 3f .
3 3 3 3

/x /y /x /x /y /x/y /x

Next, let’s suppose that |x| ² c and |y| ² c for some constant c. Noting that all the partial
derivatives in the above expression are simply products of sine and cosines, we can estimate

3
x / +y / f ² 8c 3 ,
/x /y
and so, at last,

3
|sin x sin y ? xy| ² 8c = 4 c 3
6 3

Exercises

7. Find the Taylor polynomial of degree ² 1 for fÝx, yÞ = e xy at Ý0, 0Þ.

8. Find the Taylor polynomial of degree ² 2 for fÝx, yÞ = e xy at Ý0, 0Þ.

9. Find the Taylor polynomial of degree ² 3 for fÝx, yÞ = e xy at Ý0, 0Þ.

10. Find the Taylor polynomial of degree ² 1 for fÝx, yÞ = e x cos y at Ý0, 0Þ.

11. Use Taylor’s Theorem to find a quadratic approximation of e x cos y at the origin.

12. Estimate the error in the approximation found in Problem 11 if |x| ² 0.1 and |y| ² 0.1.

6
 Chapter Ten

Sequences, Series, and All That

10.1 Introduction
Suppose we want to compute an approximation of the number e by using the Taylor
polynomial pn for f (x ) = e x at a =0. This polynomial is easily seen to be

x2 x3 xn
p n ( x) = 1 + x + + +K+ .
2 6 n!

We could now use p n (1) as an approximation to e . We know from the previous chapter
that the error is given by
eξ
e − p n (1) = 1 n +1 ,
( n + 1)!
where 0 < ξ < 1 . Assume we know that e <3, and we have the estimate

3
0 ≤ e − pn ( 1) ≤ .
( n + 1)!

Meditate on this error estimate. It tells us that we can make this error as small as we like by
choosing n sufficiently large. This is expressed formally by saying that the limit of
p n (1) as n becomes infinite is e . This is the idea we shall study in this chapter.

1 0 . 2 Sequences
A sequence of real numbers is simply a function from a subset of the nonnegative
integers into the reals. If the domain is infinite, we say the sequence is an infinite
sequence. (Guess what a finite sequence is.) We shall be concerned only with infinite
sequences, and so the modifier will usually be omitted. We shall also almost always
consider sequences in which the domain is either the entire set of nonnegative or positive
integers.
There are several notational conventions involved in writing and talking about
sequences. If f : Z + → R , it is customary to denote f (n ) by f n , and the sequence itself
 1
by ( f n ) . (Here Z + denotes the positive integers.) Thus, for example,   is the sequence
 n

10.1
 1
f defined by f (n) = . The function values f n are called terms of the sequence.
n
Frequently one sees a sequence described by writing something like
, ,9 ,K, n 2 ,K .
14

This is simply another way of describing the sequence (n 2 ).

Let (a n ) be a sequence and suppose there is a number L such that for any ε >0,

there is an integer N such that | a n − L| < ε for all n > N . Then L is said to be a limit of the
sequence, and (a n ) is said to converge to L . This is usually written lim a n = L . Now,
n →∞

what does this really mean? It says simply that as n gets big, the terms of the sequence get
 1
close to L . I hope it is clear that 0 is a limit of the sequence   . From the discussion
 n
in the Introduction to this chapter, it should be reasonably clear that a limit of the sequence
 1 1 1
1 + + +K+  is e .
 2 6 n!
The graph of a sequence is pretty dreary compared with the graph of a function
whose domain is an interval of reals, but nevertheless, a look at some pictures can help
understand some of these definitions. Suppose the sequence (a n ) converges to L . Look at
the graph of (a n ) :

The fact that L is a limit of the sequence means that for any ε >0, there is an N so that to the

right of N , all the spots are in the strip of width 2ε centered at L.

Exercises

10.2
1. Prove that a sequence can have at most one limit (We may thus speak of the limit of
a sequence.).

2. Give an example of a sequence that does not have a limit. Explain.

3. Suppose the sequence (a n ) = a 0 , a 1 , a 2 ,K converges to L. Explain how you know

that the sequence (a n+5 ) = a5 , a 6 , a 7 ,K also converges to L.

 3
4. Find the limit of the sequence   , or explain why it does not converge.
 n2 

 3n 2 + 2n − 7 
5. Find the limit of the sequence   , or explain why it does not converge.
 n2 

 5n 3 − n 2 + 7 n + 2 
6. Find the limit of the sequence   , or explain why it does not
 3n 3 + n 2 − n + 10 
converge.

 logn 
7. Find the limit of the sequence   , or explain why it does not converge.
 n 

10.3 Series
Suppose (a n ) is a sequence. The sequence (a 0 + a1 +K+ a n ) is called a series. It is a

 n

little neater to write if we use the usual summation notation:  ∑ a k  . We have seen an
 k =0 
example of such a thing previously; viz.,

 1  n 1 
∑
1 1
1 + + +K+  =  .
 2 6 n !   k =0 k ! 
n ∞
It is usual to replace lim ∑ a k by ∑a k . Thus, one would, for example, write
n →∞
k=0 k =0

∑k!.
1
e=
k =0

10.3
 ∞
One also frequently sees the limit ∑a k written as a 0 + a 1 +K+a n +K . And one more word
k =0

∞
of warning. Some poor misguided souls also use ∑a k to stand simply for the series
k =0

 n 
∑
 a k  . It is usually clear whether the series or the limit of the series is meant, but it is
 k =0 
nevertheless an offensive practice that should be ruthlessly and brutally suppressed.

Example

 1 
n
1
Let's consider the series  ∑
1 1
 = 1 + + +K+ n  . Let
 k =0 2   2 
k
2 4

1 1 1 1
Sn = 1 + + + +K+ n . Then
2 4 8 2
1 1 1 1 1 1
S n = + + +K+ n + n +1 .
2 2 4 8 2 2
Thus
Sn 1 1
= S n − S n = 1 − n +1 .
2 2 2

Sn
This makes it quite easy to see that lim = 1 , or lim S n = 2 . In other words,
n→∞ 2 n→∞

∑2
1
k
=2.
k =0

 n

Observe that for series  ∑ a k  to converge, it must be true that lim a n = 0 . To see
 k =0  n →∞

∞ n n −1
this, suppose L = ∑ a k , and observe that a n = ∑ a k − ∑ a k . Thus,
k =0 k =0 k =0

 n n −1  n n −1
lim a n = lim a k −
n →∞ n →∞
k=0
∑ ∑a
k =0
k

= lim
n →∞
∑a
k =0
k − lim
n→∞
∑a
k =0
k

= L − L = 0.

10.4
  n

In other words, if lim a n ≠ 0 , then the series  ∑ a k  does not have a limit.
n →∞  k =0 

Another Example
 n
1
Consider the series  ∑  . First, note that lim
1
= 0 . Thus we do not know that
 k =1 k n →∞ k

the series does not converge; that is, we still don't know anything. Look at the following
picture:

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 2 4 6 8 10 12

11

∑k.
1 1
The curve is the graph of y = . Observe that the area under the "stairs" is simply
x k =1

∑k
1 1
Now convince yourself that is larger than the area under the curve y = from x =1
k =1 x
to x = n+1. In other words,

n n +1
∑ k >∫
1 1
dx = log(n + 1) .
k =1
1 x

10.5
We know that log(n + 1) can be made as large as we wish by choosing n sufficiently large.
n

∑k
1
Thus can be made as large as we wish by choosing n sufficiently large. From this it
k =1

 n
1
follows that the series  ∑  does not have a limit. (This series has a name. It is called
 k =1 k
the harmonic series. )
The method we used to show that the harmonic series does not converge can be used
on many other series. We simply consider a picture like the one above. Suppose we have a
 n

series  ∑ a k  such that a k > 0 for all k . Suppose f is a decreasing function such that
 k =1 
R

f ( k ) = a k for all k . Then if the limit lim

R →∞ ∫ f ( x)dx does not exist, the series is divergent.
1

Exercises

 n
1
8. Find the limit of the series  ∑  , or explain why it does not converge.
 k =0 3 
n

 n

Find the limit of the series  ∑
5
9.  , or explain why it does not converge.
 k =0 n + 3

1 1 1
1 0 . Find a value of n that will insure that 1 + + +K+ > 10 6 .
2 3 n

∞
θ 2 k +1
1 1 . Let 0 ≤ θ ≤ 1 . Prove that sinθ = ∑ (−1)k .
k =0 ( 2 k + 1)!
n
θ 2 k +1
[Hint: p 2 n +1 (θ ) = ∑ (−1) k is the Taylor polynomial of degree < 2n+1 for
k =0 ( 2 k + 1)!
the function f (θ ) = sinθ at a = 0.]

10.6
  n

1 2 . Suppose we have a series  ∑ a k  such that a k > 0 for all k , and suppose f is a
 k =1 

decreasing function such that f (k ) = a k for all k . Show that if the limit
R

lim
R →∞ ∫ f ( x)dx exists, then the series is convergent.
1

 n
1 
1 3 . a) Find all p for which the series  ∑  converges.
 k =1 kp
b) Find all p for which the series in a) diverges.

10.4 More Series

 n

Consider a series  ∑ a k  in which a k ≥ 0 for all k . This is called a positive
 k =0 

 n

series. Let  ∑ b k  be another positive series. Suppose that b k ≤ a k for all k > N , where
 k =0 

 n

N is simply some integer. Now suppose further that we know that  ∑ a k  converges.
 k =0 

 n

This tells us all about the series  ∑ b k  . Specifically, it tells us that this series also
 k =0 

 n

converges. Let's see why that is. First note the obvious:  ∑ b k  converges if and only if
 k =0 

 n  n n

∑
 b k  converges. Next, observe that for all n , we have
k= N 
∑ bk ≤ ∑a k , from which it
k=N k=N

n
follows at once that lim ∑ bk exists.
n →∞
k=N

Example

10.7
  n

What about the convergence of the series  ∑
1
 ? Observe first that
 k =1 n + 3n + n + 4 
3 2

 n
1
Then observe that the series  ∑
1 1
< .  converges because
n + 3n + n + 4
3 2
n3  k =1 n 
3

 −1 1  1  n 
∑ 3
R 1 1
lim
R →∞ 1∫ x 3
dx = lim
R→∞

 3R 2
+  =
3 3
. Thus  converges.
 k =1 n + 3n + n + 4 
2

 n
  n

Suppose that, as before,  ∑ a k  and  ∑ b k  are positive series, and b k ≤ a k for all
 k =0   k =0 

 n

k > N , where N is some number. This time, suppose we know that  ∑ b k  is divergent.
 k =0 

 n

Then it should not be too hard for you to convince yourself that  ∑ a k  must be
 k =0 
divergent, also.

Exercises

Which of the following series are convergent and which are divergent? Explain your
answers.

 n
1 
1 4 . ∑ 
 k =0 2 e + k 
k

 n
1 
1 5 . ∑ 
 k =0 2 k + 1

 n
1 
1 6 . ∑ 
 k =2 log k 

 n

1 7 . ∑
1

 k =0 k + k − 1
2

10.8
10.5 Even More Series
We look at one more very nice way to help us determine if a positive series has a
 n

limit. Consider a series  ∑ a k  , and suppose a k > 0 for all k. Next suppose the
 k =0 

 a k +1 
sequence   is convergent, and let
 ak 

a k +1
r = lim .
k →∞ ak

 n

The number r tells us almost everything about the convergence of the series  ∑ a k  . Let's
 k =0 
see about it.
1− r
First, suppose that r < 1. Then the number ρ = r + is positive and less than 1.
2
a k +1
For all sufficiently large k, we know that ≤ ρ . In other words, there is an N so that
ak
a k +1 ≤ a k ρ for all k ≥ N . Thus

a k +1 ≤ a k ρ ≤ a k −1 ρ 2 ≤ a k −2 ρ 3 ≤K≤ a N ρ k +1− N .
Look now at the series
 n 
∑ (
 a N ρ k − N  = a N (1 + ρ + ρ 2 +K ρ n −N ) .
k= N 
)
 n

This one converges because the Geometric series  ∑ ρ k  converges (Recall that
 k =0 

 n 
0 < ρ < 1. ). It now follows from the previous section that our original series  a k  has a
 k =0 
∑
limit.
 n

A similar argument should convince you that if r > 1, then the series  ∑ a k  does
 k =0 
not have a limit.
The "method" of the previous section is usually called the Comparison Test, while
that of this section is usually called the Ratio Test.

10.9
Exercises

Which of the following series are convergent and which are divergent? Explain your
answers.

 10 k 
n
1 8 . ∑ 
 k =0 k ! 

 3 2 k +1 
n
1 9 . ∑ 
 k =0 5 
k

 3 2 k +1 
n
2 0 . ∑ 
 k =0 10 
k

 n
3k 
2 1 . ∑ 
 k =1 5 ( k + k + 1) 
k 4

 3 k ( k 4 + k + 1) 
n
2 2 . ∑ 
 k =1 5k 

10.6 A Final Remark

The "tests" for convergence of series that we have seen so far all depended on the
series having positive terms. We need to say a word about the situations in which this is
 n

not necessarily the case. First, if the terms of a series  ∑ a k  alternate in sign, and if it is
 k =0 

true that | a k +1 | ≤ | a k | for all k , then lim a k = 0 is sufficient to insure convergence of the
k →∞

series. This is not too hard to see—meditate on it for a while.

The second result is a bit harder to see, and we'll just put out the result as the word,
 n

asking that you accept it on faith. It says simply that if the series  ∑ | a k | converges,
 k =0 

 n
  n

then so also does the series  ∑ a k  . Thus, faced with an arbitrary series  ∑ a k  , we
 k =0   k =0 

10.10
  n

may unleash out arsenal of tests on the series  ∑ | a k | . If we find this one to be
 k =0 
convergent, then the original series is also convergent. If, of course, this series turns out
not to be convergent, then we still do not know about the original series.

10.11
 Chapter Eleven

Taylor Series

11.1 Power Series

Now that we are knowledgeable about series, we can return to the problem of
investigating the approximation of functions by Taylor polynomials of higher and higher
degree. We begin with the idea of a so-called power series. A power series is a series of
the form
 n 
∑
 k= 0
ck (x − a) k .


A power series is thus a sequence of special polynomials: each term is obtained from
the previous one by adding a constant multiple of the next higher power of (x − a).

Clearly the question of convergence will depend on x , as will the limit where there is one.
The k th term of the series is ck (x − a)k so the Ratio Test calculation looks like

c k +1 (x − a)k +1 c
r(x) = lim = x − a lim k +1 .
k →∞ ck (x − a) k k →∞ c
k

Recall that our series converges for r(x) < 1 and diverges for r(x) > 1 . Thus this

ck +1
series converges absolutely for all values of x if the number lim = 0 . Otherwise, we
k →∞ ck

ck
have absolute convergence for | x − a | < lim and divergence for
k →∞ c
k +1

ck ck
| x − a | > lim . The number R = lim is called the radius of convergence,
k →∞ c k →∞ ck +1
k +1

and the interval | x − a |< R is called the interval of convergence. There are thus exactly
 n 
three possibilities for the convergence of our power series  ∑ ck (x − a) k :
 k= 0 

(i)The series converges for no value of x except x = a ; or

11.1
 (ii)The series converges for all values of x ; or
(iii)There is a positive number R so that the series converges for | x − a |< R and
diverges for | x − a | > R .

Note that the Ratio Test tells us nothing about the convergence or divergence of the
series at the two points where | x − a |= R .

Example
 n  c k! 1
Consider the series  ∑ k!x k . Then R = lim k = lim = lim = 0.
k=0  k →∞ c
k +1
k →∞ (k +1)! k →∞ k +1

Thus this series converges only when x = 0.

Another Example
 n  c 3k 1 1
Now look at the series  ∑ 3k (x −1)k . Here R = lim k = lim k +1 = lim = .
k=0  k →∞ c
k +1
k →∞ 3 k →∞ 3 3
1 1
Thus, this one converges for | x − 1|< and diverges for | x − 1|> .
3 3

Exercises

Find the interval of convergence for each of the following power series:

 n 
1.  ∑ (x + 5)k
k=0 

 n 1 
2. ∑ (x −1)k
k=0 k 

11.2
  n k 
3. ∑ (x − 4)k
 k = 0 3k + 1 

 n 3k 
4.  ∑ (x +1) k
 k = 0 k! 

 n k! 
5.  ∑ (x − 9) k
 k = 0 7(k +1)
2


11.2 Limit of a Power Series

 n 
If the interval of convergence of the power series  ∑ ck (x − a) k is | x − a |< R , then,
 k= 0 

of course, the limit of the series defines a function f :

∞
f (x) = ∑ c (x − a)
k
k
, for | x − a |< R .
k=0

It is known that this function has a derivative, and this derivative is the limit of the
derivative of the series. Moreover, the differentiated series has the same interval of
convergence as that of the series defining f . Thus for all x in the interval of convergence,
we have
∞
f '( x) = ∑ kc (x − a)
k
k −1
.
k=1

We can now apply this result to the power series for the derivative and conclude that
f has all derivatives, and they are given by
∞
f (p ) (x) = ∑ k(k −1)K(k − p + 1)c (x − a) k
k− p
.
k= p

Example

11.3
 ∞
1
We know that = ∑ x k for | x |< 1 . It follows that
1 − x k=0
∞
1
(1 − x) 2
= ∑ kx k −1 = 1+ 2x + 3x 2 + 4x3 +K
k =1

for | x |< 1 .

It is, miraculously enough, also true that the limit of a power series can be integrated,
and the integral of the limit is the limit of the integral. Once again, the interval of
convergence of the integrated series remains the same as that of the original series:
x ∞
ck
∫ f (t)dt = ∑ k +1(x − a)
k +1
.
a k=0

Example
We may simply integrate the Geometric series to get
∞
xk +1
log(1− x) = − ∑ , for −1 < x < 1, or 0 < 1 − x < 2.
k=0 k + 1

It is also valid to perform all the usual arithmetic operations on power series. Thus if
∞ ∞
f (x) = ∑c x k
k
and g(x) = ∑d x k
k
for | x |< r , then
k=0 k=0

∞
f (x) ± g(x) = ∑ (c k ± dk )x k , for | x |< r .
k =0

Also,
∞
 k 
f (x)g(x) = ∑  ∑ ci dk −i c kx k , for | x |< r .

k= 0 i= 0 

The essence of the story is that power series behave as if they were “infinite degree”
polynomials—the limits of power series are just about the nicest functions in the world.

11.4
Exercises

 n 
6. What is the limit of the series  ∑ x 2k ? What is its interval of convergence?
k=0 

 n 
7. What is the limit of the series  ∑ 2(−1) k kx 2k −1 ? What is its interval of convergence?
 k =1 

8. Find a power series that converges to tan −1 x on some nontrivial interval.

∞
9. Suppose f (x) = ∑ c (x − a)
k
k
. What is f (p ) (a) ?
k=0

11.3 Taylor Series

Our major interest in finding a power series that converges to a given function. The
obvious candidate for such a series is simply the sequence of Taylor polynomials of
increasing degree. Thus if f is a given function, and a is a point in the interior of the
domain of f, the Taylor Series for f at a is the series
 n f (k ) (a) 
∑
 k = 0 k!
(x − a)k .


The Taylor Series is thus an “infinite degree” Taylor Polynomial>

In general, the Taylor series for a function may not converge on any nontrivial interval
to f , but, mercifully, for many sufficiently nice functions it does. In such cases, we are
provided with the nice answer to the question proposed back in Chapter Nine: Can we
approximate the function f as well as we like by a Taylor Polynomial for sufficiently large
degree?

11.5
Example
 n x 2k + 1 
The Taylor series for f (x) = sin x at x = a is simply  ∑ (−1)k . An easy
k=0 (2k +1)!

calculation shows us that the radius of convergence is infinite, or in other words, this
power series converges for all x . But is the limit sin x ? That’s easy to decide. From
Section 9.3, we know that
n
x 2 k+ 1 | x |2 n+ 3
sin x − ∑ (−1)k ≤ ,
k =0 (2k +1)! (2n + 3)!

and we know that

| x |2 n + 3
lim = 0,
n→ ∞ (2n + 3)!

no mater what x is. Thus we have

∞
x 2 k+ 1
sin x = ∑(−1) k
, for all x .
k=0 (2k +1)!

Exercises

10. Find the Taylor Series at a = 0 for f (x) = ex . Find the interval of convergence and

show that the series converges to f on this interval.

11. Find the Taylor Series at a = 0 for f (x) = cos x . Find the interval of convergence and

show that the series converges to f on this interval.

12. Find the derivative of the cosine function by differentiating the Taylor Series you
found in Problem #11.

13. Find the Taylor Series at a = 1 for f (x) = logx . Find the interval of convergence and

show that the series converges to f on this interval.

11.6
14. Let the function f be defined by
 0, for x = 0
f (x) =  −1/ x 2 .
e , for x ≠ 0

Find the Taylor Series at a = 0 for f. Find the interval of convergence and the limit of
the series.

11.7
 Chapter Twelve

Integration

12.1 Introduction

We now turn our attention to the idea of an integral in dimensions higher than one.
Consider a real-valued function f : D → R , where the domain D is a nice closed subset of

Euclidean n-space R n . We shall begin by seeing what we mean by the integral of f over
the set D; then later we shall see just what such an abstract thing might be good for in real
life. Mrs. Turner taught us all about the case n = 1 . As it was in extending the definition
of a derivative to higher dimensions, our definition of the integral in higher dimensions will
include the definition for dimension 1 we learned in grammar school—as always, there
will be nothing to unlearn. Let us again hark back to our youth and review what we know
about the integral of f : D → R in case D is a nice connected piece of the real line R.

First, in this context, the only nice closed pieces of R are the closed intervals; we thus
have D is a set [a , b] , where b > a . Recall that we defined a partition P of the interval to

be simply a finite subset {x 0 , x1 ,K , x n } of [a , b] with a = x 0 < x1 < x 2 <K < x n = b .

The mesh of a partition is max{| x i − xi −1 |:i = 12K

, , n} . We then defined a Riemann sum

S ( P ) for this partition to be a sum

n
S ( P ) = ∑ f ( xi* )∆ xi ,
i =1

where ∆x i = x i − x i−1 is simply the length of the subinterval [ xi −1 , xi ] and xi* is any point

in this subinterval. (Thus there is not just one Riemann sum for a partition P; the sum

obviously also depends on the choices of the points xi* . This is not reflected in the

notation.)
Now, if there is a number L such that we can make all Riemann sums as close as we
like to L by choosing the mesh of the partition sufficiently small, then f is said to be

12.1
integrable over the interval, and the number L is called the integral of f over [a, b]. This
b

number L is almost always denoted ∫ f ( x )dx . More formally, we say that L is the
a

integral of f over [a , b] if for every ε > 0 , there is a δ so that | S ( P ) − L| < ε for every

partition P having mesh < δ. You no doubt remember from your first encounter with this

integral that it initially seemed like an impossible thing to compute in any reasonable
situation, but then some version of the Fundamental Theorem of Calculus came to the
rescue.

12.2 Two Dimensions

Let us begin our study of higher dimensional integrals with the two dimensional
case. As we have seen so often in the past, in extending calculus ideas to higher
dimensions, most of the excitement occurs in taking the step from one dimension to two
dimensions—seldom is the step from 97 to 98 dimensions very interesting. We shall thus
begin by looking at the integral of f : D → R for the case in which D is a nice closed

subset of the plane. Complications appear at once. On the real line, nice closed sets are
simply closed intervals; in the plane, nice closed sets are considerably more interesting:

12.2
A moment's reflection convinces us that the domain D can, even in just two dimensions,
be considerably more complicated than it is in one dimension. First, capture D inside a
rectangle with sides parallel to the coordinate axes; and then divide this rectangle into
subrectangles by partitioning each of its sides:

Now, label the subrectangles that meet D, say with subscripts i = 12

, , K, n . The largest
area of all such rectangles is called the mesh of the subdivision. In each such rectangle,

choose a point ( xi* , y *i ) in D. A Riemann sum S now looks like

n
S = ∑ f ( x i* , yi* ) ∆Ai ,
i= 1

where ∆Ai is the area of the rectangle from which ( xi* , y *i ) is chosen. Now if there is a

number L such that we can get as close to L as we like by choosing the mesh of the
subdivision sufficiently small, then f is said to be integrable over D, and the number L is
the integral of f over D. The number L is usually written with two snake signs:

∫∫ f ( x, y) dA .
D

Such integrals over two dimensional domains are frequently referred to as double
integrals.

12.3
 I hope the definition of the integral in case D is a nice subset of R 3 is evident. We
capture D inside a box, and subdivide the box into boxes, etc. , etc. There will be more of
the higher dimensional stuff later.
Let's look a bit at some geometry. For the purpose of drawing a reasonable picture,
let us suppose that f ( x, y) ≥ 0 everywhere on D.

Each term f ( xi* , y *i ) ∆Ai is the volume of a box with base the rectangle Ai and height

f ( xi* , y *i ) . The top of the box thus meets the surface z = f ( x , y ) . The Riemann sum is

thus the total volume of all such boxes. Convince yourself that as the size of the bases of
the boxes goes to 0, the boxes "fill up" the solid bounded below by the x-y plane, above
by the surface z = f ( x , y ) , and on the sides by the cylinder determined by the region D.

The integral ∫∫ f ( x, y) dA is thus equal to the volume of this solid. If f ( x, y) ≤ 0 , then,

D

of course, we get the negative of the volume bounded below by the surface z = f ( x , y ) ,

above by the x-y plane, etc.

Suppose a and b are constants, and D = E ∪ F , where E and F are nice domains
whose interiors do not meet. The following important properties of the double integral
should be evident:

12.4
 ∫∫ [af ( x , y ) + bg ( x , y )]dA = a ∫∫ f ( x , y )dA + b ∫∫ g ( x , y )dA , and
D D D

∫∫ f ( x, y) dA = ∫∫ f ( x , y )dA + ∫∫ f ( x , y )dA .
D E F

Now, how on Earth do we ever find an integral ∫∫ f ( x, y) dA ? Let's see. Again, we

D

shall look at a picture, and again we shall draw our picture as if f ( x, y) ≥ 0 . It should be

clear what happens if this is not the case.

We assume our domain D has a special form; specifically, we suppose it to be
bounded above by the curve y = h ( x ) , below by y = g ( x ) , on the left by x = a , and on

the right by x = b :

It is convenient for us to think of the integral ∫∫ f ( x, y) dA as the volume of the blob

D

bounded below by D in the x-y plane and above by the surface z = f ( x , y ) . Think of

finding this volume by dividing the blob into slices parallel to the y-axis and adding up the
volumes of the slices. To approximate the volumes of these slices, we use slabs:

12.5
We partition the x interval [a, b ]: a = x0 < x1 <K < x n −1 < x n = b . In each subinterval

[ xi −1 , xi ] choose a point xi* . Our approximating slab has as its base the rectangle of

"width" ∆x i = x i − x i−1 and height h( xi* ) − g( x *i ) ; the roof is z = f ( xi* , y ) . The volume
h ( x*i )
of the slab is the cross section area times the thickness, or [ ∫ f ( x i* , y )dy ]∆x i .
g ( x*i )

The sum of the volumes of the approximating slabs is thus

n
h ( xi* )
S = ∑ [∫ f ( x *i , y) dy ]∆ xi .
g ( xi* )
i= 1

The double integral we seek is just the "limit" of these as we take thinner and thinner
slabs; or finer and finer partitions of the interval [a, b]. But Lo! The above sums are

12.6
Riemann sums for the ordinary one dimensional integral of the function

F ( x) = ∫
h( x )
f ( x , y )dy , and so the double integral is given by
g( x )

∫∫ f ( x, y) dA = ∫ F ( x) dx
D a

b h( x )

= ∫[ ∫ f ( x , y )dy ]dx
a g ( x)

The double integral is thus equal to an integral of an integral, usually called an iterated
integral. It is traditional to omit the brackets and write the iterated integral simply as
b h( x )

∫ ∫ f ( x, y )dydx .
a g( x )

Example

∫∫ [ x + y 2 ]dA , where D is the area enclosed by the

2
Let's find the double integral
D

lines y = x , x = 0 , and x + y = 2 . The first item of business here is to draw a picture of

D (We always need a picture of the domain of integration.):

12.7
It should be clear from the picture that in the language of our discussion, g ( x ) = x ,

h( x ) = 2 − x , a = 0 , and b = 1. So slice parallel to the y axis:

The lower end of the slice is at y = x and the upper end is at y = 2 − x . The "volume" is

thus
2 −x y= 2 −x
y3 (2 − x ) 3 x3 (2 − x ) 3 7 3
∫x + = + = x 2 (2 − x) + − x3 − = 2x 2 + − x ,
2 2 2
[ x y ]dy x y
3 y= x
3 3 3 3

and we have such a slice for all x from x = 0 to x = 1. Thus

( 2 − x) 3 7 3
1

∫∫ [ x + y ]dA = ∫ [2 x + − x ]dx
2 2 2

D 0 3 3
1
2 x 3 (2 − x ) 4 7 x 4
= − −
3 12 12 0

16 4
= =
12 3

Exercises

∫∫ x dA , where D is the domain bounded by the curves y = 4 − x 2 and y = 3 x .

2
1. Find
D

12.8
 ∫∫ (x − y )dA , where D is the area in the first quadrant enclosed by the
2
2. Find
D

coordinate axes and the line 2 x + y = 4 .

3. Use double integration to find the area of the region enclosed by the curves x − y = 2

and y = − x 2 .

4. Find the volume of the solid cut from the first octant by the surface z = 4 − x 2 − y .

5. Sketch the domain of integration and evaluate the iterated integral:

1 1

∫∫ y 2
e xy dydx .
0 x

6. Sketch the domain of integration and evaluate the iterated integral:

log 8log x

∫ ∫e x+ y
dydx .
1 0

7. Find the volume of the wedge cut from the first octant by the cylinder z = 12 − 3y 2

and the plane x + y = 2 .

8. Suppose you have a double integral ∫∫ f ( x, y) dA in which the domain D is bounded

D

on the left by the curve x = g( y ) , on the right by x = h( y ) , below by y = a , and

above by y = b .

12.9
 Give an iterated integral for the double integral in which the first integration is with
respect to x , and explain what's going on.

9. Give a double integral for the area of the region bounded by x = y 2 and x = 2 y − y 2 ,

and evaluate the integral.

12.10
 Chapter Thirteen

More Integration

13.1 Some Applications

Think now for a moment back to elementary school physics. Suppose we have a
system of point masses and forces acting on the masses. Specifically, suppose that for
each i = 12
, , K, n we have a point mass mi whose position in space at time t is given by
the vector ri .. Assume moreover that there is a force f i acting on this mass. Thus

according to Sir Isaac Newton, we have

d 2 ri
f i = mi
dt 2
for each i. Now sum these equations to get
n n
d 2 ri
F = ∑ fi = ∑ mi , or
i =1 i =1 dt 2

 n 
2 ∑ i i 
mr
d  i =1 ,
F=M
dt  n 
 ∑ mi 
 i =1 
n
where M = ∑ mi . Reflect for a moment on this equation. If we define R by
i =1

∑mr i i
d 2R
R= i =1
n
, then the equation becomes F = M . Thus the sum of the external
dt 2
∑m
i =1
i

forces on the system of masses is the total mass times the acceleration of the mystical
point R. This point R is called the center of mass of the system.
In case the total mass is continuously distributed in space, the "sum" in the
equation for R becomes an integral. Let's look at what this means in two dimensions.

13.1
Suppose we have a plate and the mass density of the plate at (x,y) is given by ρ( x , y ) .

To find the center of mass of the plate, we approximate its location by chopping it into a
bunch of small pieces and treating each of these pieces as a point mass.

Now choose a point ri = xi* i + y i* j in each rectangle. The mass of this rectangle will be

approximately ρ( x i* , yi* ) ∆Ai , where ∆Ai is the area of the rectangle. The equation for the

center of mass of this system of rectangles is then

n n

~
∑ m r ∑ ρ( x
i i i
*
, yi* ) ri ∆ Ai
R= i =1
n
= i =1
n

∑m i ∑ρ (x i
*
, y *i ) ∆Ai
i =1 i =1

 n  n  
 ∑ i  i +  ∑ ρ ( xi , y i ) yi ∆ Ai  j 
1
= n
ρ ( x i
*
, yi
*
) x *
i ∆A * * *

∑ ρ( xi* , y *i ) ∆Ai   i =1
i =1
  i =1  

The three sums in the previous line are Riemann sums for two dimensional integrals!
Thus as we take smaller and smaller rectangles, etc., we obtain for R, the location of the
center of mass

13.2
 1      
R=   ∫∫ xρ ( x, y) dA i +  ∫∫ yρ ( x , y) dA j 
∫∫ ρ( x , y ) dA   P  P  
P

In other words, the coordinates ( x , y ) of the center of mass of P are given by

   
 ∫∫ x ρ ( x , y )dA   ∫∫ yρ ( x , y )dA 
x = P  , and y =  P ,
M M

where M = ∫∫ ρ( x , y )dA is the total mass of the plate.

P

Example
Let's find the center of mass of a plate having the shape of the plane region
enclosed by the triangle

and having constant density (In this case, we say the mass is uniformly distributed over
the region. Suppose ρ( x , y ) = k . First,
a b (1− x /a ) a
a2 b
∫∫ x ρ( x , y )dA = k ∫ ∫ xdydx = k ∫ xb(1 − x / a )dx = k 6
, and then
T 0 0 0

a b (1 −x / a )
kb 2 a ab 2
∫∫ yρ( x , y ) dA = k ∫ ∫0 =
2 ∫0
− =
2
ydydx ( 1 x / a ) dx k .
T 0
6

13.3
 ab
Also, M = ∫∫ kdA = k ∫∫ dA = k . Thus,
T T
2

a b
x= , and y = .
3 3

Meditate on the fact that the location of the center of mass does not depend on
the value of the constant k. Note that in general, if the density is constant, then the
constant slips out through the integral signs and cancels top and bottom in the recipe for
the coordinates ( x , y ) . This is what most of our intuitions tell us, I believe. It is,

nevertheless, comforting to see this fact come out in the mathematical wash. In this case
of constant density, the center of mass thus depends only on the geometry of the plate; it
is thus a geometric property of the region. It is called the centroid of the region. One
must never confuse the two concepts; intimately related though they be, they are
different. The center of mass is something a physical body has, while the centroid is an
abstract mathematical something.

Exercises

1. Find the center of mass of a plate of density ρ( x , y ) = y + 1 having the shape of the

area bounded by the line y = 1 and the parabola y = x 2 .

2. Find the center of mass of the smaller of the two regions cut from the elliptical region

x 2 + 4 y 2 = 12 by the parabola x = 4 y 2 if the density ρ( x , y ) = 5x .

3. Find the centroid of the semicircular region {( x , y ) ∈ R 2 : x 2 + y 2 ≤ a 2 , and y ≥ 0} .

13.4
4. Find the centroid of the region bounded by the horizontal axis and one arch of the sine
curve. (That is, the region between x = 0 and x = π bounded above by y = sin x and

below by y = 0.)

5. Find the centroid of the region bounded by the curves y 2 = 2x , x + y = 4 , and

y =0.

2 0 4 x

6. The area of a region A is ∫ ∫ dydx + ∫ ∫ dydx . Draw a picture of the region.

0 x2 −4 0 0

7. Let f :D → R be a function defined on a nice subset D ⊂ R 2 . The average value A

1
area of D ∫∫
of f on D is defined to be A = f ( x , y )dA .
D

a)Find the average depth of a bowl having the shape of the bottom half of the sphere

x 2 + y2 + z 2 = 1 .

b)Find the average depth of a bowl having the shape of the part of the

paraboloid z = x 2 + y 2 − 1 below the x-y plane.

8. Let D be the region inside the circle x 2 + (y − a) 2 = a 2 that lies below the line y = a .

a)Find the centroid of D.

b)Find the point on the semicircular boundar of D that is closest to the centroid.

13.2 Polar Coordinates

Now we shall see what happens when we express a double integral as an iterated
integral in some coordinate system other than the usual rectangular, or Cartesian,

13.5
coordinate system. We shall see more of this later; right now, let's look at what happens
in polar coordinates.

Suppose we have the integral ∫∫ f ( x, y) dA . In polar coordinates, we know that

D

we must substitute
x = r cosθ , and
y = r sin θ .

There is, however, more to it than this. When we divided the plane into regions formed
by the curves x = constant and y = constant, we got rectangles, etc., etc. Now we

divide the plane into regions formed by the curves r = constant and θ = constant ,

where r and θ are the usual polar coordinates. This results in funny shaped regions:

Now, a typical region looks like

13.6
The area of this region is thus something like ∆A ≈ r∆r∆θ , and our iterated integral looks
like

∫∫ f ( x, y)dA = ∫∫ f (r cosθ , r sin θ )rdrdθ

D

together with the appropriate limits of integration. (We may, of course, integrate first

with respect to θ and then with respect to r if this is convenient.) We desperately need

to see an example.

Example
Let's find the centroid of the region enclosed by the curve whose equation in polar
coordinates is r = 1 + cosθ . Here is a picture drawn by Maple:

13.7
 The centroid ( x , y ) is given by

∫∫ xdA ∫∫ ydA
x= D
, and y = D
.
∫∫ dA ∫∫ dA
D D

First. let's find the integral ∫∫ xdA . Now, when we hold θ fixed and integrate first with
D

respect to r, the lower limit is independent of θ and is always r = 0 , while the upper

limit depends, of course on θ and is r = 1 + cosθ . We have a slice for each value of θ

from θ = 0 to θ = 2π , and so our iterated integral looks like

2 π1 +cos θ 2π 1+ cosθ

∫∫ xdA = ∫ ∫ r cosθ rdrdθ = ∫ ∫r cosθ drdθ .

2

D 0 0 0 0

It is downhill all the way now:

2 π1+ cosθ 2π
1
∫ ∫ r cosθ drdθ = ∫ (1 + cosθ ) 3 cosθ dθ
2

0 0
30
1 2π
= ∫ [cosθ + 3 cos2 θ + 3 cos3 θ + cos 4 θ ]dθ
3 0
2π 2π
1 3 1
= [ 0 + ∫ (1 + cos 2θ )dθ + 0 + ∫ (1 + cos 2θ ) 2 dθ ]
3 2 0 40
π 1 2π 2 π π 15π 5
=π + + ∫
6 12 0
cos 2θ dθ = π + +
6 12
=
12
= π
4

Now for the other integrals.

2 π 1+cos θ

∫∫ ydA = ∫ ∫ r sin θ drdθ = 0 . Finally,

2
It should be clear that
D 0 0

13.8
 2 π1 +cos θ
1 2π
∫∫ dA = ∫ ∫0 rdrd θ = 2 ∫0 (1 + cosθ ) dθ
2

D 0

1 2π
4 ∫0
=π + (1 + cos 2θ )dθ

π 3
=π + = π
2 2

We are, at last, done.

5
π 5
x= 4 = , and y = 0.
3
π 6
2

Exercises

9. Find the area of the region enclosed by the curve with polar equation r = sin2θ .

10. Evaluate the integral ∫∫ ( x + y )dA , where D is the region in the first quadrant inside
D

the circle x 2 + y 2 = a 2 and below the line y = x 3 .

11. Find the centroid of the region in the first quadrant inside the circle r = a and between
π
the rays θ = 0 and θ = α , where 0 ≤ α ≤ . What is the limiting position of the
2
centroid as α → 0 ?

∫∫ e x 2 + y2
12. Evaluate dA , where R is the semicircular region bounded above by
R

y = 1 − x 2 and below by the x axis.

13.9
13. Find the area enclosed by one leaf of the rose r = cos3θ .

14. Find the area of the region inside r = 1 + cosθ and outside r = 1.

13. 3 Three Dimensions

We move along to integrals in three dimensions. The idea is quite simple.

Suppose we have a function f : D → R , where D is a nice subset of R 3 . Capture D

inside a big box (i.e., a rectangular parallelepiped). Now subdivide this box by partitioning
each of its sides. The volume of the largest such box is called the mesh of the subdivision.

In each box that meets D, choose a point ( xi* , y *i , z*i ) in D. A Riemann sum S now looks

like
n
S = ∑ f ( xi* , y i* , zi* )∆Vi ,
i =1

where ∆Vi is the volume of the box from which ( xi* , y *i , z*i ) was chosen. (The

summation is over all boxes that meet D.) If there is a number L such that | S − L| can be

made arbitrarily small by choosing a subdivision of sufficiently small mesh, then we say
that f is integrable over D, and the number L is called the integral of f over D. This
integral is usually written with three snake signs:

∫∫∫ f ( x , y , z)dV .
D

Let's see how to evaluate such a thing by considering iterated integrals. Here's
what we do. First, project D onto a coordinate plane. (We choose the x-y plane as an
example.)

13.10
Let A be the region in the x-y plane onto which D projects. Assume that a vertical line
through a point ( x , y ) ∈ A enters D through the surface z = g( x , y ) and exits through the

surface z = h( x , y ) . In other words, the blob D is the solid above the region A between

the surfaces z = g( x , y ) and z = h( x , y ) . Now we simply integrate the integral

h( x , y )

∫ f ( x , y , z)dz over the region A:

g ( x ,y )

 h ( x, y ) 
∫∫∫ f ( x , y , z)dV = ∫∫  ∫ f ( x, y, z)dz dA .
D A  g ( x ,y ) 

Example

Let's find the integral ∫∫∫ ( x + 2 y + z)dV , where D is the tetrahedron with vertices
D

(0,0,0), (1,0,0), (0,2,0), and (0,0,1).

13.11
When we project D onto the x-y plane, the bottom of D is the surface z = 0 and the top
y y
of D is x + + z = 1, or z = 1 − x − . The projection is simply the triangle
2 2

 1− x − y /2 
Our iterated integral is thus simply ∫∫  ∫ ( x + 2 y + z) dz dA . We now write the double
A  0 

integral over A as an iterated integral, and we have

 1 − x − y /2 
∫∫∫ ( x + 2 y + z ) dV = ∫∫A  ∫0
 ( x + 2 y + z ) dz  dA

D

1 2 ( 1− x) 1− x − y/ 2

=∫ ∫ ∫ ( x + 2 y + z)dzdydx .
0 0 0

13.12
Again, it is traditional to omit the parentheses in the iterated integral. All we need do now
is integrate three times. Let's use Maple for the calculations, but look at the intermediate
steps, rather than just use one statement. Here we go.
1 − x − y /2

For the first integration, we want ∫ ( x + 2 y + z )dz :

0

int(x+2*y+z,z=0..(1-x-y/2));

Thus,
1 − x − y /2
1 3 7 1
∫ ( x + 2 y + z)dz = − x 2 − 2 xy + y − y 2 + ,
2 2 8 2
0

and our next integral is

2 (1 −x ) 1− x −y / 2 2 (1 −x )
1 3 7 1
∫ ∫ (1 + 2 y + z)dzdy = ∫ ( − x 2 − 2 xy + y − y 2 + )dy .
2 2 8 2
0 0 0

Maple again:

int(-(x^2)/2-2*x*y+(3/2)*y-(7/8)*y^2+1/2,y=0..2*(1-x));

Thus,
2 (1 −x )
1 3 7 1 2 5
∫ ( − x 2 − 2 xy + y − y 2 + )dy = − 4 x − x 3 + 3x 2 + ,
2 2 8 2 3 3
0

and finally,
int(-4*x-(2/3)*x^3+3*x^2+(5/3),x=0..1);

13.13
At last!
1 2 ( 1− x) 1 −x − y /2
1
∫ ∫ ∫ ( x + 2 y + z) dzdydx = 2 .
0 0 0

We make a few obvious observations. First, if S is a solid, the volume V of the

solid is simply V = ∫∫∫ dV . If the mass density of a blob having the shape of S is
S

ρ( x , y , z) , then the mass M of the blob is M = ∫∫∫ ρ( x , y , z) dV , and the location

S

( x , y , z ) of the center of mass is given by

∫∫∫ xρ ( x, y, z)dV
x= S

∫∫∫ yρ( x , y , z)dV

y= S

∫∫∫ zρ( x , y , z) dV
z= S

Exercises

15. Find the volume of the tetrahedron having vertices (0,0,0), (a,0,0),(0,b,0), and (0,0,c).

16. Find the centroid of the tetrahedron in the previous exercise.

∫∫∫ ( xy + z )dV , where S is the set S = {( x , y , z):0 ≤ z ≤ 1− | x|− | y|} .

2
17. Evaluate
S

13.14
18. Find the volume of the region in the first octant bounded by the coordinate planes

and the surface z = 4 − x 2 − y .

19. Write six different iterated integrals for the volume of the tetrahedron cut from the
first octant by the plane 12 x + 4 y + 3z = 12 .

20. A solid is bounded below by the surface z = 4 y 2 , above by the surface z = 4 , and on

the ends by the surfaces x = 1 and x = −1. Find the centroid.

21. Find the volume of the region common to the interiors of the cylinders x 2 + y 2 = 1

and x 2 + z 2 = 1 .

13.15
 Chapter Fourteen

One Dimension Again

14.1 Scalar Line Integrals

Now we again consider the idea of the integral in one dimension. When we were
introduced to the integral back in elementary school, we considered only functions defined
on nice subsets of the real line. The notion of an integral of a function f : D → R in

which D is a nice one dimensional set, but is not a subset of the reals is our next object of
study. To get some idea of why one might care about such a thing, consider the simple
problem of finding the mass of a piece of wire having the shape of an arc of a space curve
C and having a given density ρ(r) . How might we approach such a problem? Simple

enough! We subdivide, or partition, the curve with a finite set of points, say

{r0 , r1 , K, rn } . On the subarc joining ri −1 to ri , we choose a point, say ri* , and evaluate

the function ρ( ri* ) . Now we multiply this times the length of the line segment joining the

points ri −1 and ri for an approximation to the mass of this arc of our curve. Then sum

these to obtain an approximation for the total mass:

n
S = ∑ ρ( ri* )| ri − ri −1 | .
i= 1

Then we all believe that the "limit" of these sums as we choose finer and finer partitions
of the curve should be the actual, honest-to-goodness mass of the wire.
Let's abstract the essence of the discussion. Suppose f :C → R is a function

whose domain C is a curve (in R 2 or R 3 , or wherever). We subdivide the curve as in the

preceding discussion and choose a point ri* on the subarc joining ri −1 to ri . The sum

14.1
 n
S = ∑ f (ri* )| ri − ri−1 |
i =1

again is called a Riemann sum. If there is a number L such that all Riemann sums are
arbitrarily close to L for sufficiently fine partitions, then we say f is integrable on C, and

the number L is called the integral of f on C and is denoted ∫ f ( r)dr . This integral is
C

also frequently referred to as a line integral.

This is wonderful, but how do find such an integral? It is remarkably simple and
easy. Suppose we have a vector description of the curve C; say r(t ), for a ≤ t ≤ b . We

partition the curve by partitioning the interval [a , b] : If {a = t 0 , t 1 , K, t n = b} is a partition

of the interval, then the points {r(t 0 ), r(t 1 ),K , r( t n )} partition the curve C. We obtain

the point ri* on the subarc joining r(t i −1 ) to r(t i ) by choosing t *i ∈ [t i −1 , t i ] and letting

ri* = r(t i* ) . Our Riemann sum now looks like

n
S = ∑ f ( r(t i* )| r(t i ) − r( ti −1 )| .
i= 1

14.2
 ∆t i
Next, multiply the terms on the right by one, but one disguised as , where, of course,
∆t i

∆t i = t i − t i −1 . Then we see

n
r(t i ) − r( ti −1 )
S = ∑ f ( r(t i* ) ∆t i .
i= 1 ∆ti

r (t i ) − r (t i − 1 ) dr
We know that lim = , and so it is not hard to convince oneself that the
∆t →0 ∆t i dt

"limiting" value of the Riemann sums is

b
dr(t )
∫ f ( r(t )) dt
dt .
a

We have thus turned the problem into one we know how to solve—a plain old everyday
elementary calculus integral. Hence,

b
dr(t )
∫ f ( r)dr = ∫ f ( r(t ))
dt
dt .
C a

Example
Suppose we have a wire in the shape of a quarter circle of radius 2, and the
density of the wire is given by ρ( x , y ) = y . What is the mass of the wire? Well, we

know the mass is simply the integral ∫ ydr , where C is the quarter circle:
C

14.3
 π
A vector description of the curve is r(t ) = 2 cos ti + 2 sin tj , for 0 ≤ t ≤ . Thus we have
4
dr
= |− 2 sin t i + 2 cos t j| = 2 , and the integral becomes simply
dt
π/4

∫ ydr = ∫ 4 sin t dt = 4 .
C 0

Let's see what happens if we use a different vector description of the curve, say

dr t 2
r(t ) = ti + 4 − t 2 j for 0 ≤ t ≤ 2 . We have = i− j= . Hence
dt 4 − t2 4 − t2

2
 2  2

∫ ydr = ∫ 4−t   dt = ∫ 2dt = 4 .

2

C 0  4− t2  0

We get, as we must, the same answer.

Exercises

1. Evaluate the integral ∫ ( x − y + z + 2 )dr , where C is the curve r(t ) = ti + (1 − t ) j + k ,

C

0 ≤ t ≤ 1.

14.4
 ∫ x + y dr , where C is the curve r(t ) = 4 cos ti + 4 sin tj + 3tk ,
2 2
2. Evaluate the integral
C

−2π ≤ t ≤ 2π .

3. Find the centroid of a semicircle of radius a.

4. Find the mass of a wire having the shape of the curve r(t ) = (t 2 − 1) j + 2 tk , 0 ≤ t ≤ 1

3
if the density is ρ(t ) = t.
2

5. Find the center of mass of a wire having the shape of the curve

2 2 3 /2 t2
r(t ) = ti + t j + k, 0 ≤ t ≤ 2 ,
3 2

1
if the density is ρ(t ) = .
t +1

6. What is ∫ dr ?
C

14.2 Vector Line Integrals

Now we are introduce something perhaps a little different from what we have seen
to now—integrals with vector valued integrands. Specifically, suppose C is a space curve

and f : C → R 3 is a function from C into the Euclidean space R 3 . We are going to define

an integral ∫ f ( r) ⋅ dr . Why should we care about such a thing? Again, let's think about a
C

physical model. You learned in fifth grade physics that the work done by a force F acting
through a distance d is simply the product Fd. The force F and the displacement d are, of
course, really vectors, and we saw earlier in life that the "product" of the two is actually

14.5
the scalar, or dot, product of the two vectors. Now, in general, neither of these quantities
will be constant, and we will have a variable force F(r) acting along a curve C in space.
How do we compute the work done in this situation? Let's see. Once more, we partition
the curve by choosing a sequence of points {r0 , r1 , K, rn } on the curve, with r0 being the

initial point and rn being the final point. Now, of course, there is an orientation, or

direction, specified on the curve. One may think of specifying an orientation by simply
putting an arrow on the curve—it thus makes sense to speak of the initial point and the

terminal point of the curve. Exactly as in the scalar integrand case, we choose a point ri*

on the subarc joining ri −1 to ri , and evaluate F( ri* ) . Now then, the work done in going

from ri −1 to ri is approximately the scalar product F ( ri* ) ⋅ (ri − ri−1 ) . Add all these up for

an approximation to the total work done:

n
S = ∑ F (ri* ) ⋅ ( ri − ri −1 ) .
i= 1

The course should be obvious now; we take finer and finer partitions, and the limiting
value of the sums is the integral

∫ F ( r) ⋅ dr .
C

This integral too is called a line integral. To prevent confusion, we sometimes

speak of scalar line integrals and vector line integrals. How to find such a vector integral
should be clear from the discussion of scalar line integrals. We let r(t ), a ≤ t ≤ b , be a

vector description of C. (Here r(a ) is the initial point and r(b) is the terminal point.)

The discussion proceeds almost exactly as it did in the previous section and we get

14.6
 b
dr
∫ F ( r) ⋅ dr = ∫ F (r( t )) ⋅ dt
dt .
C a

Example

∫ [( xy + z ) i + ( x + z) j + 2 yzk ] ⋅ dr , where C is the straight line from the

2
Find
C

origin to the point (1,2,3). The line C has a vector description r(t ) = ti + 2tj + 3tk . Thus,

dr
= i + 2 j + 3k , and so
dt
1

∫ [( xy + z ) i + ( x + z) j + 2 yzk ] ⋅ dr = ∫ [( 2t 2 + 9 t 2 ) i + (t + 3t ) j + 12t 2 k ] ⋅ ( i + 2 j + 3k )dt

2

C 0

1 1

= ∫ ( 2t + 8t + 36t )dt = ∫ (38t 2 + 8t ) dt

2 2

0 0

1
38 3 50
= t + 4t 2 = .
3 0 3

Nothing to it.

Another Example
Now let's integrate the same function from (0,0,0) t0 (1,2,3), but this time along
the path P in the picture:

14.7
Here the path P is the union of the three nice curves, P1 , P2 , and P3 , so our integral is the

sum of three integrals:

∫ F ( x , y , x ) ⋅ dr = ∫ F ( x , y, x ) ⋅ dr + ∫ F ( x, y, x ) ⋅ dr + ∫ F ( x , y , x ) ⋅ dr ,
P P1 P2 P3

where

F ( x , y , z) = ( xy + z 2 ) i + ( x + z) j + 2 yzk .

A vector description of P1 is simply r(t ) = ti , 0 ≤ t ≤ 1. Thus

1 1

∫ F ( x , y , z) ⋅ dr = ∫ F ( t ,00, ) ⋅ i dt = ∫ tj ⋅ i dt = 0 .
P1 0 0

For P2 , we have r(t ) = i + tj , 0 ≤ t ≤ 2 . This gives us

2 2 2

∫ F ( x , y , z ) ⋅ dr = ∫ F (1, t ,0) ⋅ j dt = ∫ (t i + j ) ⋅ j dt = ∫ dt = 2 .
P2 0 0 0

14.8
Finally, for P3 , there is r(t ) = i + 2 j + tk , 0 ≤ t ≤ 3 ; and so

3 3

∫ F ( x , y , z) ⋅ dr = ∫ F (12, , t ) ⋅ k dt = ∫ [( 2 + t 2 )i + (1 + t ) j + 4t k] ⋅ k dt
P3 0 0

= ∫ 4t dt = 18 .
0

At last, we have then ∫ F ( x , y , z) ⋅ dr = 0 + 2 + 18 = 20 .

P

Exercises

7. Evaluate ∫ [ xyi + x 2 j ] ⋅ dr , where C is the arc of the curve y = x 2 from (0,0) to (1,1).
C

8. Evaluate ∫ (cos x i − y j) ⋅ dr where C the part of the curve y = sin x from (0,0) to
C

(π,0).

9. Evaluate the line integral of F ( x , y , z) = xyzi + ( xy + yz ) j + z 2 k from (0,0,0) to

(-1,1,2) along the line segment joining these two points.

10. Evaluate the line integral of F ( x , y , z) = ( x − z) i + ( y − z) j − ( x + y ) k along the

polygonal path from (0,0,0) to (1,0,0) to (1,1,0) to (1,1,1).

1
11. Integrate F ( x , y ) = (− yi + x j) one time around the circle x 2 + y 2 = a 2 in the
x +y
2 2

counterclockwise direction.

14.9
14.3 Path Independence

Suppose we evaluate the vector line integral ∫ F ( r) ⋅ dr , where C is a curve from

C

the point p to the point q. Let r(t ), a ≤ t ≤ b , be a vector description of C. Then, of

course, we have r(a ) = p and r(b) = q . As we have already seen,

b
dr
∫ F ( r) ⋅ dr = ∫ F (r( t )) ⋅ dt
dt .
C a

Now let us make the very special assumption that there exists a real-valued (or scalar)

function g: R 3 → R such that the derivative, or gradient, of g is the integrand F :

∇g = F .

Next let's use the Chain Rule to compute the derivative of the composition
h(t ) = g ( r(t )) :

dr dr
h'( t ) = ∇g ⋅ = F ( r(t )) ⋅ .
dt dt
This is, mirabile dictu, precisely the integrand in our line integral:
b b
dr
∫C F ( r) ⋅ dr = ∫a F ( r(t )) ⋅ dt dt = ∫a h'( t ) dt = h(b) − h(a ) = g( p) − g (q ) .

This is a very exciting result and calls for some meditation. Note that the curve C
has completely disappeared from the answer. The value of the integral depends only on
the values of the function g at the endpoints; the path from p to q does not affect the
answer. The line integral is path independent. The result is esthetically pleasing and is
clearly the lineal descendant of the fundamental theorem of calculus we learned so many
years ago.

14.10
 A moment's reflection on the examples we have seen should convince us that a lot
of integrals are not path independent, thus many very nice functions F (or vector fields )
are not the gradient of any function. A function F that is the gradient of a function g is
said to be conservative and the function g is said to be a potential function for F.

Let's suppose the domain D of the function F: D → R 3 is open and connected

(Thus any two points in D may be joined by a nice path.) We have just seen that if there
exists a function g: D → R such that F = ∇g , then the integral of F between any two

points of D does not depend on the path between the two points. It turns out, as we
shall see, that the converse of this is true. Specifically, if every integral of F in D is path
independent, then there is a function g such that F = ∇g . Let's see why this is so.

Choose a point p = ( x 0 , y 0 , z0 ) ∈ D . Now define g ( s) = g( x , y , z) to be the

integral from p to s along any curve joining these points. We are assuming path
independence of the integral, so it matters not what curve we choose. Okay, now we
∂g
compute the partial derivative . The domain D is open and hence includes an open
∂x
ball centered at s = ( x, y, z) ∈ D . Choose a point q = ( x1 , y , z) in such an open ball, and

let L be the straight line segment from s to q . Then, of course, L lies in D. Now let's
integrate F from p to s by going along any curve C from p to q and then along L from q to
s:

g ( s) = g( x , y , z) = ∫ F ( r) ⋅ dr + ∫ F ( r) ⋅ dr .
C L

∂
∂ x ∫C
The first integral on the right does not depend on x, and so F ( r) ⋅ dr = 0 . Thus

∂g ∂
∂ x ∂ x ∫L
= F ( r) ⋅ dr .

14.11
We clearly need to find ∫ F ( r) ⋅ dr . This is easy. Suppose
L

F (r ) = f 1 (r ) i + f 2 ( r ) j + f 3 ( r ) k .

dr
A vector description of L is simply r(t ) = ti + yj + zk , x1 ≤ t ≤ x . Thus = i , and our
dt
x

line integral becomes simply ∫ F ( r) ⋅ dr = ∫ f 1 ( t , y , z)dt . We are almost done, for note
L x1

that now

∂ ∂ x
∂ x ∫L ∂ x x∫1 1
F ( r ) ⋅ d r = f (t , y, z)dt = f 1 ( x, y , z) .

Hence
∂g
= f1 .
∂x

∂g ∂g
It should be clear to one and all how to show that = f 2 and = f 3 , thus
∂y ∂z

giving us the desired result: F = ∇g .

Exercises

∂g
12. Prove that = f 2 , where g and f 2 are as in the preceding discussion.
∂y

13. Prove that if F: D → R 3 , where D is open and connected, and every ∫ F ( r) ⋅ dr is

C

path independent, then ∫ F ( r) ⋅ dr = 0 for every closed path in D.( A closed path, or
P

14.12
 curve, is one with no endpoints.) [Physicists and others like to use a snake sign with a

little circle superimposed on it ∫ to indicate that the path of integration is closed.]

14. Prove that if F: D → R 3 , where D is open and connected, and ∫ F ( r) ⋅ dr = 0 for

P

every closed path in D, then every ∫ F ( r) ⋅ dr is path independent .

C

15. a)Find a potential function g for the function F ( r) = yzi + xzj + xyk .

b)Evaluate the line integral ∫ F ( r) ⋅ dr , where C is the curve

C

r(t ) = (e t sin t )i + t 2 e 3 t j + cos3 t k , 0 ≤ t ≤ 1 .

16. a)Find a potential function g for the function F ( r) = e y+ 2 z (i + x j + 2 xk ) .

b)Find another potential function for F in part a).

b)Evaluate the line integral ∫ F ( r) ⋅ dr , where C is the curve

C

r(t ) = t cos 2t 2 i + 4t j + e 2 t k , 0 ≤ t ≤ π .

∫ [( e sin y + 3y ) i + (e x cos y + 2x − 2 y ) j ] ⋅ dr
x
17. Evaluate where E is the ellipse
E

4 x 2 + y 2 = 4 oriented clockwise.

[Really good hint: Find the gradient of g ( x , y , z) = e x sin y + 2xy − y 2 .]

14.13
 Chapter Fifiteen

Surfaces Revisited

15.1 Vector Description of Surfaces

We look now at the very special case of functions r: D → R 3 , where D ⊂ R 2 is a

nice subset of the plane. We suppose r is a nice function. As the point (s, t ) ∈ D moves

around in D, if we place the tail of the vector r( s, t ) at the origin, the nose of this vector

will trace out a surface in three-space. Look, for example at the function r: D → R 3 ,

where r( s, t ) = si + t j + ( s2 + t 2 ) k , and D = {(s, t ) ∈ R 2 :− 1 ≤ s, t ≤ 1} . It shouldn't be

difficult to convince yourself that if the tail of r( s, t ) is at the origin, then the nose will be

on the paraboloid z = x 2 + y 2 , and for all (s, t ) ∈ D , we get the part of the paraboloid

above the square −1 ≤ x , y ≤ 1 . It is sometimes helpful to think of the function r as

providing a map from the region D to the surface.

The vector function r is called a vector description of the surface. This is, of course,
exactly the two dimensional analogue of the vector description of a curve.

15.1
For a curve, r is a function from a nice piece of the real line into three space; and for a
surface, r is a function from a nice piece of the plane into three space.
Let's look at another example. Here, let

r( s, t ) = cos s sin t i + sin s sin t j + cos t k ,

for 0 ≤ t ≤ π and 0 ≤ s ≤ 2π . What have we here? First, notice that

| r(s, t )| 2 = (cos s sin t ) 2 + (sin s sin t ) 2 + (cost ) 2

= sin 2 t (cos2 s + sin 2 s) + cos2 t
= sin 2 t + cos2 t = 1

Thus the nose of r is always on the sphere of radius one and centered at the origin.
Notice next, that the variable, or parameter, s is the longitude of r( s, t ) ; and the variable t

is the latitude of r( s, t ) . (More precisely, t is co-latitude.) A moment's reflection on this

will convince you that as r is a description of the entire sphere. We have a map of the
sphere on the rectangle

15.2
Observe that the entire lower edge of the rectangle (the line from (0,0) to (2π ,0) ) is

mapped by r onto the North Pole, while the upper edge is mapped onto the South Pole.
Let r( s, t ), ( s, t ) ∈ D be a vector description of a surface S, and let p = r( s, t ) be

a point on S. Now, c( s) = r( s, t ) is a curve on the surface that passes through he point p.

dc ∂ r
Thus the vector = ( s , t ) is tangent to this curve at the point p. We see in the same
ds ∂ s
∂r
way that the vector (s , t ) is tangent to the curve r(s , t ) at p.
∂t

15.3
 ∂r ∂r
At the point p = r( s, t ) on the surface S, the vectors and are thus tangent to S.
∂s ∂t
∂r ∂r
Hence the vector × is normal to S.
∂s ∂t

Example
Let's find a vector normal to the surface given by the vector description
∂r
r( s, t ) = si + t j + ( s2 + t 2 ) k at a point. We need to find the partial derivatives and
∂s
∂r
:
∂s
∂r ∂r
= i + 2sk , and = j + 2tk .
∂s ∂t

The normal N is

i j k
∂r ∂r
N= × = 1 0 2 s = − 2 si − 2tj + k .
∂s ∂t
0 1 2t

Meditate on the geometry here and convince yourself that this result is at least
reasonable.

Exercises

1. Give a vector description for the surface z = x + 2 y 2 , x , y ≥ 0.

2. Give a vector description for the ellipsoid 4 x 2 + y 2 + 8z 2 = 16 .

3. Give a vector description for the cylinder x 2 + y 2 = 1 .

15.4
4. Describe the surface given by r( s, t ) = s cos t i + s sin tj + sk , 0 ≤ t ≤ 2π , −1 ≤ s ≤ 1 .

5. Describe the surface given by r( s, t ) = s cos t i + s sin tj + s2 k , 0 ≤ t ≤ 2π , 1 ≤ s ≤ 2 .

6. Give a vector description for the sphere having radius 3 and centered at the point
(1,2,3).

7. Find an equation (I.e., a vector description) of the line normal to the sphere
a a a
x 2 + y 2 + z 2 = a 2 at the point ( , ,− ).
3 3 3

8. Find a scalar equation (I.e., of the form f ( x, y, z) = 0 ) of the plane tangent to the

a a a
sphere x 2 + y 2 + z 2 = a 2 at the point ( , ,− ).
3 3 3

9. Find all points on the surface r( s, t ) = (s 2 + t 2 )i + (s + 3t ) j − stk at which the tangent

plane is parallel to the plane 5x − 6 y + 2z = 7 , or show there are no such points.

10. Find an equation of the plane that contains the point (1,-2,3) and is parallel to the

plane tangent to the surface r( s, t ) = (s + t )i + s2 j − 2t 2 k at the point (1, 4,-18).

15.2 Integration
Suppose we have a nice surface S and a function f : S → R defined on the surface.

We want to define an integral of f on S as the limit of some sort of Riemann sum in the
way in which we have already defined various integrals. Here we have a slight problem in
that we really are not sure at this point exactly what we might mean by the area of a

15.5
small piece of surface. We assume the surface is sufficiently smooth to allow us to
approximate the area of a small piece of it by a small planar region, and then add up these
approximations to get a Riemann sum, etc., etc. Let's be specific.
We subdivide S into a number of small pieces S1 , S 2 ,K , S n each having area ∆Ai ,

select points ri * = ( x i* , yi* , zi* ) ∈ Si , and form the Riemann sum

n
R = ∑ f (ri* )∆Ai .
i =1

Then, of course, we take finer and finer subdivisions, and if the corresponding Riemann

sums have a limit, this limit is the thing we call the integral of f on S: ∫∫ f (r) dS .
S

Now, how do find such a thing. We need a vector description of S , say

r: D → r( D) = S . The surface S is subdivided by subdividing the region D ⊂ R 2 into

rectangles in the usual way:

15.6
The images of the vertical lines, s = constant, form a family of "parallel" curves on the
surface, and the images of the horizontal lines t = constant, also form a family of such
curves:

Let's look closely at one of the subdivisions:

15.7
We paste a parallelogram tangent to the surface at the point r( si , t i ) as shown. The

∂r ∂r
lengths of the sides of this parallelogram are ( si , ti )∆si and ( s , t )∆t i . The area
∂s ∂t i i

∂r  ∂r 
is then  ( si , t i ) ∆si  ×  ( si , t i ) ∆t i  , and we use the approximation
∂s  ∂t 

∂r  ∂ r 
∆Ai ≈  (si , t i ) ×  ( si , ti ) ∆ si ∆ t i
∂s  ∂t 

in the Riemann sums:

∂r  ∂r 
n
R = ∑ f ( r( si , t i ))  (s i , t i ) ×  ( si , t i ) ∆si ∆t i .
i= 1
∂s  ∂t 

These are just the Riemann sums for the usual old time double integral of the function

∂ r  ∂ r 
n
F ( s, t ) = ∑ f ( r( si , t i ))  (s i , t i ) ×  (s i , t i )
i= 1
∂s  ∂t 

over the plane region D. Thus,

∂r ∂r
∫∫ f (r) dS = ∫∫ f ( r( s, t )) ∂ s (s, t ) × ∂ t ( s, t ) dA .
S D

Example
Let's use our new-found knowledge to find the area of a sphere of radius a .

Observe that the area of a surface S is simply the integral ∫∫ dS . In the previous section,
S

we found a vector description of the sphere:

15.8
 r( s, t ) = a cos s sin t i + a sin s sin t j + a cos t k ,

0 ≤ t ≤ π and 0 ≤ s ≤ 2π . Compute the partial derivatives:

∂r
= − a sin s sin t i + a cos s sin t j , and
∂s

∂r
= a cos s cos t i + a sin s cos t j − a sin t k
∂t

Then

i j k
∂r ∂r
× = a − sin s sin t cos s sin t
2
0
∂s ∂t
cos s cos t sin s cost -sint

= a 2 [ − cos s sin 2 t i − sin s sin2 t j − sin t cos t k ]

Next we need to find the length of this vector:

∂r ∂r
× = a 2 [cos 2 s sin 4 t + sin 2 s sin 4 t + sin 2 t cos2 t ]1 /2
∂s ∂t
= a 2 [sin 4 t + sin2 t cos2 t ]1/2 = a 2 [sin 2 t (sin 2 t + cos2 t )]1/2
= a 2 |sin t |

Hence,

∂r ∂r
Area = ∫∫ dS = ∫∫ × dA = ∫∫ a 2 |sin t |dA
S D
∂s ∂t D

15.9
 π 2π

=a ∫ ∫ |sin t | dsdt
2

0 0

= 2π a ∫ sin tdt = 4π a
2 2

Another Example
Let's find the centroid of a hemispherical shell H of radius a. Choose our

coordinate system so that the shell is the surface x 2 + y 2 + z 2 = a 2 , z ≥ 0 . The centroid

( x , y , z ) is given by

∫∫ xdS ∫∫ ydS ∫∫ zds

x= H
y= H
and z = H
.
∫∫ dS ∫∫ dS ∫∫ dS
H H H

First, note from the symmetry of the shell that x = y = 0 . Second, it should be clear

∫∫ dS = 2π a 2
from the precious example that . This leaves us with just integral to
H

evaluate:

∫∫ zdS . Most of the work was done in the example before this one. This hemisphere has
H

the same vector description as the sphere, except for the fact that the domain of r is the
π
rectangle 0 ≤ s ≤ 2π , 0 ≤ t ≤ . Thus
2

15.10
 π / 2 2π
∂r ∂r
∫∫ zdS = a ∫ ∫ a cost ∂ s × ∂ t
2
dsdt
H 0 0

π /2 2 π π /2

=a3 ∫ ∫ cos t sin t dsdt = 2π a 3 ∫ cost sin t dt

0 0 0

π /2
= π a 3 sin 2 t = π a3
0

πa3 a
And so we have z = = . Is this the result you expected?
2π a 2
2

Yet One More Example

Our new definition of a surface integral certainly includes the old one for plane
surfaces. Look at the "surface" described by the vector function

r(θ , r ) = r cosθ i + r sin θ j ,

with r defined on some subset D of the θ − r plane. For what we hope will be obvious

reasons, we are using the letters θ and r instead of s and t . Now consider an integral

∫∫ f ( x, y) dS
S

over the surface S described by r. We know this integral to be given by

∂r ∂r
∫∫ f ( x, y) dS = ∫∫ f (r cosθ , r sin θ ) ∂ θ × ∂ r dA .
S D

Let's find the partial derivatives:

15.11
 ∂r
= − r sin θ i + r cos θ j , and
∂θ
∂r
= cosθ i + sin θ j .
∂r
Thus,

i j k
∂r ∂ r
× = − r sin θ r cosθ 0 = −rk ,
∂ θ ∂r
cosθ sinθ 0

∂r ∂r
and we have × = r . Hence,
∂θ ∂r

∂r ∂r
∫∫ f ( x, y) dS = ∫∫ f (r cosθ , r sin θ ) ∂ θ × ∂ r dA = ∫∫ f (r cosθ , r sinθ )rdA .
S D D

This should look familiar!

Exercises

11. Find the area of that part of the surface z = x 2 + y 2 that lies between the planes z = 1

and z = 2.

12. Find the centroid of the surface given in Problem 11.

13. Find the area of that part of the Earth that lies North of latitude 45°. (Assume the
surface of the Earth is a sphere.)

14. A spherical shell of radius a is centered at the origin. Find the centroid of that part of
it which is in the first octant.

15.12
15. a)Find the centroid of the solid right circular cone having base radius a and altitude h.
b)Find the centroid of the lateral surface of the cone in part a).

16. Find the area of the ellipse cut from the plane z = 2x by the cylinder x 2 + y 2 = 1 .

17. Evaluate ∫∫ ( x + y + z)dS , where S is the surface of the cube cut from the first octant
S

by the planes x = a, y = a , and z = a.

∫∫ x y + 1 dS , where S is the surface cut from the paraboloid

2
18. Evaluate
S

y 2 + 4 z = 16 by the planes x = 0, x = 1, and z = 0.

15.13
 Chapter Sixteen
Integrating Vector Functions

16.1 Introduction
Suppose water (or some other incompressible fluid ) flows at a constant velocity v
in space (through a pipe, for instance), and we wish to know the rate at which the water
flows across a rectangular surface S that is normal to the stream lines:

What is the rate at which the fluid flows through S? Let M (t ) denote the total volume of

fluid that has passed through the surface at time t. The amount of fluid that flows through
during the time between t and t + ∆ t is simply
M (t + ∆t ) − M (t ) = | v| a∆t ,

dM
where a is the area of S. Thus, the rate of flow through S is =| v| a .
dt
The result is slightly more complicated when various exciting changes are made.
Clearly there is nothing special about the surface's being a rectangle. But suppose that S
is placed at an angle to the stream lines instead of being placed normal to the them. Then
dM
we have = v ⋅ n a , where n is a unit normal to the surface S.
dt

16. 1
 Observe that matters which unit normal to the plane surface we choose. If we
choose the other normal (- n ), then our rate will be the negative of this one. We must
thus specify an orientation of the surface. We are computing the rate of flow from one
side of the surface to the other, and so we have to specify the "sides", so to speak.

16.2 Flux
Now, let's look at the general situation. The surface is not restricted to being a
plane surface, and the velocity of the flow is not restricted to being constant in space; it
may vary with position as well as time. Specifically, suppose S is a surface, together
with an orientation—that is, some means of specifying two "sides"—and suppose F ( r)

is a function F: R 3 → R 3 , which is the velocity of the incompressible fluid. How do we

find the rate of flow through the surface S from one side to the other?
First, let's come to grips with the problem of specifying an orientation for S. We

say that an orientation for S is a continuous function n:S → R 3 such that n( r) is normal

to S and | n( r)| = 1 for all r ∈ S . A surface together with an orientation is called an

oriented surface. At first blush this looks simple enough, and the unsophisticated might
guess that every surface has an orientation (or may be oriented, as we sometimes say).
But this is not so! There are many surfaces for which an orientation does not exist. You
may recall from grammar school a simple example of such a surface, the so-called Möbius
band, or strip. Here is my feeble attempt to draw one:

16. 2
 Now we see about finding the rate of flow through the oriented surface S. The
strategy should be old-hat by now. We subdivide S and look at "small" parallelograms
tangent to the surface:

As we have done so often, we suppose the subdivisions are small and approximate the
rate of flow, or flux, through the subdivision by the rate of flow through the tangent
parallelogram.

∆S i = F (ri* ) ⋅ n∆ Ai ,
n
and then add them to obtain yet another type of Riemann sum R = ∑ F (ri* ) ⋅ n∆ Ai . If
i= 1

these sums have a limiting value as the size of the subdivisions go to zero, this is what we
call the integral of F over the oriented surface S:

∫∫ F ( r) ⋅ dS .
S

It should be clear now what we do to evaluate such an integral. As usual, we

consider a vector description of the surface S: r: D → S ∈ R 3 , where D ⊂ R 2 . We

subdivide S by subdividing the region D into rectangles formed by lines s = constant and
t= constant, and looking at the curves r(s, t ) and r(s , t ) on the surface, exactly as we did

in integrating a scalar function over a surface S. Most conveniently now, the vector

16. 3
 ∂r ∂r ∂r ∂r
product × gives us not only a vector such that × ∆s∆t is the area of the
∂s ∂t ∂s ∂t

approximating parallelogram, but also one which is normal to the surface. There is the
∂r ∂r
slight problem of the orientation of S. Thus × may not point in the direction of
∂s ∂t
∂r ∂r
the specified orientation, in which case, of course, we simply replace × by its
∂s ∂t
∂r ∂r
negative, × . (We may think of just reversing the roles of s and t.) We have in the
∂ t ∂s
Riemann sums,

 ∂ r ∂ r
n
R = ∑ F (ri* ) ⋅ ×  ∆ s ∆t ,
i= 1
 ∂ s ∂ t i i

and, as before, we obtain

∂r ∂ r
∫∫ F ( r) ⋅ dS = ∫∫ F ( r( s, t )) ⋅  ∂ s × ∂ t  dA .
S D

The concept we have developed here is purely mathematical and is done

independent of any physical interpretation, such as our fluid flow interpretation. What
we have is just an integral of a vector function F (or field) over an oriented surface S.
This is generally called the flux of F over S. There are many physical interpretations of
this concept; you have perhaps seen some of them in elementary school physics. There is
electric flux, the flux of an electric field; magnetic flux; gravitational flux, etc., etc.

Example
Let S be the sphere of radius a oriented so that the normal points "out" of the
c
sphere, and let F ( r) =
| r| 3
r , where c is a constant. Let's find ∫∫ F ( r) ⋅ dS . Use the
S

vector description of S we used in the first Example of the previous section:

16. 4
 r( s, t ) = a cos s sin ti + a sin s sin tj + a cos tk ,

0 ≤ s ≤ 2π , 0 ≤ t ≤ π . We have already found that

∂r ∂r
× = a 2 sin t [ − cos s sin ti − sin s sin tj − cos tk ] .
∂s ∂t

Modest meditation should convince you that this normal points into the sphere, and is
thus the negative of the one we need for the specified orientation of S.
Next, the integrand is given by
c c
F ( r) = 3
r = 3 a[cos s sin t i + sin s sin t j + cos tk ] ,
| r| a

and our integral becomes

π 2π
c
∫∫ F ( r) ⋅ dS = ∫ ∫a 2
[cos s sin ti + sin s sin tj + cos tk ] ⋅a 2 sin t [cos s sin ti + sin s sin tj + cos tk ]dsdt
S 0 0

π 2π

= c∫ ∫ sin t[cos2 s sin 2 t + sin2 s sin 2 t + cos 2 t ]dsdt

0 0

π 2π π

= c ∫ ∫ sint dsdt =2πc∫ sin t dt = 4π c .

0 0 0

Note that the radius a of the sphere has disappeared—the value of the integral is
independent of the radius of the sphere.

Exercises

16. 5
 ∫∫ [ zi + x k ] ⋅ dS , where S is that part of the surface z = x 2 + y 2 that lies above
2
1. Find
S

the square {( x , y ):− 1 ≤ x ≤ 1, and − 1 ≤ y ≤ 1} , oriented so that the normal points

upward.

2. Find the flux of F ( x , y , z) = x i + zj out of the tetrahedron bounded by the coordinate

planes and the plane x + 2 y + 3z = 6 .

c
3. Find the flux of F ( r) = r out of the surface of the cube − a ≤ x, y , z ≤ a , where c
| r| 3

and a are positive constants.

4. Find the flux of the function F ( x , y , z) = 4x i + 4 y j + 2 k outward through the surface

cut from the bottom of the paraboloid z = x 2 + y 2 by the plane z = 1.

5. Find the flux of the function F ( x , y , z) = z 2 i + xj − 3zk upward throught the surface

cut from the cylinder z = 4 − y 2 by the planes x = 0, x = 1, and z = 0 .

6. Let S be the surface defined by

y = log x , 1 ≤ x ≤ e , 0 ≤ z ≤ 1 ,

and let n be the orientation of S such that n(r) ⋅ j > 0 for all r ∈ S . Find the flux

∫∫ [2 yj + zk ] ⋅ dS .
S

16. 6
16. 7
 Chapter Seventeen

Gauss and Green

17.1 Gauss's Theorem

Let B be the box, or rectangular parallelepiped, given by
B = {( x , y , z ): x 0 ≤ x ≤ x1 , y 0 ≤ y ≤ y1 , z 0 ≤ z ≤ z1 } ;

and let S be the surface of B with the orientation that points out of B. Let F : B → R 3 be
a nice function, or field. For reasons that will become apparent as the drama unfolds, let's
compute the flux

∫∫ F ( r) ⋅ dS .
S

We shall do this by computing the surface integral over each of the six sides of B
and adding the results. Let S1 be the side in the plane x = x1 ; let S 2 be the side in the
plane x = x 0 ; let S 3 be the side in the plane y = y1 ; let S 4 be the side in the plane
y = y 0 ; and let S5 and S 6 be the obvious things. We begin by computing the integral

∫∫ F ( r) ⋅ dS .
S1

A vector description of S1 is quite easy to come by; it is, of course, simply

r ( s, t ) = x1 i + sj + tk ,
where y 0 ≤ s ≤ y1 and z 0 ≤ t ≤ z1 . (Obviously, s is simply y, and t is z .) Then
∂r ∂r
× = j×k = i.
∂s ∂t
It is clear this is the specified orientation. If F ( r ) = p( x , y , z ) i + q ( x , y , z ) j + r ( x , y , z ) k ,
then
y1 z1

∫∫ F ( r) ⋅ dS = ∫ ∫ F ( x , s, t ) ⋅ i dtds
S1 y 0 z0
1

y1 z1

= ∫ ∫ p( x , s, t ) dtds
y 0 z0
1

17.1
 A vector description for the opposite side, x = x 0 , is just
r ( s , t ) = x 0 i + s j + tk ,
and we have
y1 z1

∫∫ F ( r) ⋅ dS = ∫ ∫ F ( x
S2 y 0 z0
0 , s, t ) ⋅ ( − i ) dtds

y1 z1

= ∫ ∫ − p( x
y 0 z0
0 , s, t ) dtds

The sum of these two is then

y1 z1

∫∫ F ( r) ⋅ dS + ∫∫ F ( r) ⋅ dS = ∫ ∫ [p(x , s, t) - p(x 1 0 ,s, t)] dtds .

S1 S2 y 0 z0

Observe that
∂p
x1

p( x1 , s, t ) − p( x 0 , s, t ) = ∫ ∂ x ( ξ , s , t ) dξ .
x0

Substitution of this into the previous equation gives us

∂p
y1 z1 x1

∫∫ F ( r) ⋅ dS + ∫∫ F ( r ) ⋅ dS = ∫ ∫ ∫ ∂ x (ξ , s, t )dξdtds
S10 S21 y0 z0 x0

∂p
= ∫∫∫ ∂ x dV
B

and we have turned the sum of the two surface integrals into a plain ol' volume integral .
It should be clear how we also obtain

∂q
∫∫ F ( r) ⋅ dS + ∫∫ F ( r) ⋅ dS = ∫∫∫ ∂ y dV , and
S3 S4 B

∂r
∫∫ F ( r) ⋅ dS + ∫∫ F ( r) ⋅ dS = ∫∫∫ ∂ z dV .
S5 S6 B

17.2
 The flux over the entire surface S is thus the sum of these:

∂p ∂q ∂r
∫∫ F ( r) ⋅ dS = ∫∫∫ ∂ xdV + ∫∫∫ ∂ ydV + ∫∫∫ ∂ zdV
S B B B

(z)
∂ p ∂ q ∂ r 
= ∫∫∫  + +
∂ x ∂ y ∂ z 
dV
B 

We have now found the surface integral, or flux, in terms of an ordinary volume integral.
Now, suppose we have an "arbitrary" solid region B bounded by a surface S,
together with a function F ( r ) = p( x , y , z ) i + q ( x , y , z ) j + r ( x , y , z ) k defined on B. Trap
B in a box and subdivide the box into parallelepipeds. Consider those parallelepipeds
{Bi : i = 1,2,K , n} that meet B. The surface that bounds Bi will be called S i , and oriented
so that the normal points out. The union Pn = ∪{Bi }of all the Bi is thus an
approximation to the original solid B.
Apply the equation (z) to each of these and sum the equations:

∂ p ∂q ∂ r
∑ ∫∫ F ( r ) ⋅ dS = ∑ ∫∫∫  ∂ x + ∂ y + ∂ z dV .
i Si i Bi  

The sum on the right hand side is just the integral over Pn :

∂ p ∂q ∂ r
∑ ∫∫ F ( r) ⋅ dS = ∫∫∫  ∂ x + ∂ y + ∂ z dV .
i Si Pni  

Take a closer look at the sum of the surface integrals on the left hand side of this
equation. Suppose parallelepipeds B j and Bk are adjacent, and call the common side T :

17.3
In the sum of surface integrals, the integral over the common side T appears twice, once
from the integral over S j , the surface of B j and once from the integral over S k , the

surface of Bk . These integrals, will, however, have opposite signs because the
orientation of T has one direction as a part of the surface of B j and the opposite direction

as a part of the surface of Bk . These two terms thus sum to zero and cancel each other.
In the sum of all the surface integrals, we are therefore left with only the integrals over
sides that are not adjacent to another box. A moments reflection, and you see that what is
left is precisely the integral over the boundary S n of Pn with the outward pointing
orientation. Mirabile dictu, this is precisely the equation (z):

∂ p ∂q ∂ r
∫∫ F ( r) ⋅ dS = ∫∫∫  ∂ x + ∂ y + ∂ z dV .
Sn Pn

Now, as everyone can see coming, we look at the limit of this equation as we take smaller
and smaller subdivisions. Then Pn → B and S n → S , giving us precisely the same result
for the arbitrary region B:

∂ p ∂q ∂ r
∫∫ F ( r) ⋅ dS = ∫∫∫  ∂ x + ∂ y + ∂ z dV .
S B

17.4
This is really a big deal—such a big deal that it has its own name. This is called Gauss's
Theorem, or the Divergence Theorem.
The integrand in the volume integral also has a name; it is called the divergence of
the function F. It is usually designated either div F , or ∇ ⋅ F . Thus,

∂p ∂q ∂r
div F = ∇ ⋅ F = + + .
∂x ∂y ∂z

With this new definition, Gauss’s Theorem looks like

∫∫ F(r) ⋅ dS = ∫∫∫ ∇ ⋅ F(r) dV

S

Example
c
Let's find the divergence of F ( r ) = r . First we need to see F in the form
| r|3

F ( x , y , z ) = p( x , y , z ) i + q( x , y , z ) j + r ( x , y , z ) k .

That's easy:
c
F= [ xi + yj + zk ] ,
( x + y + z 2 ) 3/ 2
2 2

and so
cx
p= ,
( x + y 2 + z 2 ) 3/ 2
2

cy
q= ,
( x + y 2 + z 2 ) 3/ 2
2

cz
r= .
( x + y 2 + z 2 ) 3/ 2
2

A bit of elementary school calculus (remember Mrs. Turner!), and we have

17.5
 ∂p x 2 + y 2 + z 2 − 3x 2
=c ,
∂x ( x 2 + y 2 + z 2 ) 5/ 2

∂q x 2 + y 2 + z 2 − 3y 2
=c ,
∂y ( x 2 + y 2 + z 2 ) 5/ 2

∂r x 2 + y 2 + z 2 − 3z 2
=c 2 .
∂z ( x + y 2 + z 2 ) 5/ 2

Hence, ∇ ⋅ F = 0 everywhere (except, of course, for r = 0, where F is not defined.).

Gauss's Theorem now tells us that the integral of F over any closed surface that
does not enclose r = 0 must be zero. This might be the ho-hum of the week save for the
fact that the function F is a common one. It is the gravitational field of a point mass fixed
at the origin, or the electric intensity field for a point charge fixed at the origin, or any field
in which the magnitude is inversely proportional to the distance from the origin and which
points in the direction of the origin.

Exercises

1. Find the outward flux of the function F = ( y − x ) i + ( z − y ) j + ( y − x ) k across the

boundary of the cube bounded by the planes x = ±4 , y = ±4 , and z = ±4 .

2. Find ∫∫ [ yi + xyj − zk ] ⋅ dS , where S is the boundary of the solid inside the cylinder
S

x 2 + y 2 ≤ 1 between z = 0 and z = x 2 + y 2 , with the outward pointing orientation.

 2z y
∫∫ [log( x + y 2 ) i +  tan −1  j + z x 2 + y 2 k ] ⋅ dS , where S is the boundary of
2
3. Find
S
 x x

the solid {( x , y , z ):1 ≤ x 2 + y 2 ≤ 2, − 1 ≤ z ≤ 2} .

17.6
4. Let B a region in R 3 , and let f :B → R be a function such that

∂2 f ∂2 f ∂2 f
+ + = 0 in B (Such a function f is said to be harmonic in B.). Let S
∂ x2 ∂ y2 ∂ z2

be the boundary of B. Show that ∫∫ ∇f ⋅ dS = 0 .

S

17.2 Green's Theorem

Let R be the rectangular region in the plane bounded by the rectangle with vertices
( x 0 , y 0 ),( x1 , y 0 ),( x1 , y1 ), and ( x 0 , y1 ) .

( x 0 , y1 ) ( x1 , y1 )

( x0 , y0 ) ( x1 , y0 )

Suppose F : R → R 2 is a vector function given by F ( x , y ) = p( x , y ) i + q ( x , y ) j . Now,

let's compute the vector line integral of F around the rectangular boundary C in the
counterclockwise direction. We shall compute the integral in four parts: the integrals
along each of the straight line segments making up the boundary.
C3
←

C4 ↓ ↑C2

→
C1

Thus,

17.7
 ∫ F ⋅ dr = ∫ F ⋅ dr + ∫ F ⋅ d r + ∫ F ⋅ d r + ∫ F ⋅ d r .
C C1 C2 C3 C4

We shall work out the evaluation of one of these in some painful detail; it should then be
rather obvious how to do the others. Start with a vector description of C1 :
r (t ) = ti + y 0 j , x 0 ≤ t ≤ x1 .
dr
Then, of course, = i , and our line integral becomes
dt

x1 x1

∫ F ⋅ dr = ∫ [ p(t , y0 ) i + q(t , y0 ) j] ⋅ i dt = ∫ p( t , y 0 )dt .

C1 x0 x0

In a similar fashion, we get

x1

∫ F ⋅ dr = ∫ − p(t , y )dt .1
C3 x0

Thus,
x1

∫ F ⋅ dr + ∫ F ⋅ dr = ∫ −[ p(t , y1 ) − p(t , y 0 )]dt

C1 C3 x0

∂p
x1 y1

= ∫ ∫ − ∂ y (t , s)dsdt
x0 y 0

∂p
= ∫∫ − dA
R
∂y

In essentially the same manner, we find that

∂q
∫ F ⋅ dr + ∫ F ⋅ dr = ∫∫ ∂ x dA .
C2 C4 R

17.8
Thus

∫ F ⋅ dr = ∫ F ⋅ dr + ∫ F ⋅ dr + ∫ F ⋅ dr + ∫ F ⋅ dr
C C1 C2 C3 C4

∂ q ∂ p 
= ∫∫R  ∂ x − ∂ y dA

We have turned a one dimensional vector integral into a double integral, similar to
the way in which in the previous section we turned a two dimensional vector integral into
a triple integral.
Now suppose we have a reasonable region R bounded by a reasonable curve C
with a counterclockwise orientation:

Now cover this region with rectangles, and apply the above recipe to each rectangle, and
add all the equations, etc., etc., just as we did with the parallelepipeds in deriving Gauss's
Theorem. When the dust settles, we have the same result:

∂ q ∂ p
∫ F ⋅ dr = ∫∫  ∂ x − ∂ y dA .
C R

This is called Green's Theorem. You should note that the same equation is valid even if
the region R is bounded by more than one closed curve.

17.9
Here the boundary C consists of three curves with the orientation indicated by the arrows
in the fine picture—meditate on the covering by approximating rectangles and you will see
why the orientation of the "inside" curves is clockwise. The line integral on the left side is
simply the sum of the integrals over the pieces of the boundary curve.

Example

Let's evaluate the line integral ∫ [5yi + 3( x + 1) j ] ⋅ dr ,

C
where C is the circle of

radius 2 centered at the origin, oriented counterclockwise. First, note that

∂q ∂p
= 3, and =5 .
∂x ∂y
Thus,

∂ q ∂ p
∫ [5yi + 3( x + 1) j] ⋅ dr = ∫∫  ∂ x − ∂ y dA
C R

= −2 ∫∫ dA = −8π
R

Exercises

5. Evaluate ∫ [(sin x + 3 y 2 ) i + (2 x − e − y ) j ] ⋅ dr , where C is the boundary of the half-disc

2

x 2 + y 2 ≤ 9, y ≥ 0 oriented counterclockwise.

17.10
  y 
∫ (tan
−1
6. Evaluate ) i + log( x 2 + y 2 ) j  ⋅ dr , where C is the boundary of the region
C
x 

1 ≤ r ≤ 2, 0 ≤ θ ≤ π , oriented clockwise. (These are the usual polar coordinates.)

∫ [ ye i + x 3 e y j ] ⋅ dr , where C is the curve given by

x2
7. Evaluate the line integral
C

r (t ) = sin t i + sin 2t j , 0 ≤ t ≤ 2π by using Green's Theorem.

17.3 A Pleasing Application

Here we shall use Green’s Theorem to find the area of a region R bound by a
polygon P with vertices ( x1 , y1 ), ( x 2 , y 2 ),K, ( x n , y n ). How do we do this? We simply

apply Greens’ Theorem to the function

F( x, y ) = p( x, y )i + q( x, y ) j = xj .
Then Green’s Theorem tells us that
 ∂q ∂p 
∫∫  ∂x − ∂x dA = ∫ F( x, y) ⋅ dr ,
R P

which becomes

∫∫ dA = ∫ xj ⋅dr .
R P

We thus find the area by evaluating the line integral on the right side. This is easy. We
simply integrate over each line segment of the polygon and add up the integrals.
Let’s integrate along the line segment Lk from ( x k , y k ) to ( x k +1 , y k +1 ) . A vector

description of this segment is

r (t ) = (1 − t )( x k i + y k j) + t ( x k +1 i + y k +1 j) , 0 ≤ t ≤ 1 .

Thus r ' (t ) = ( x x +1 − x k )i + ( y k +1 − y k ) j , and we have

17.11
 1
∫ xj ⋅ dr = ∫ [(1 − t ) xk + txk +1 ]j ⋅ r' (t )dt
Lk
0

1
= ( y k +1 − y k ) ∫ [(1 − t ) x k + tx k +1 ]dt
0

( y k +1 − y k )( x k +1 + x k )
=
2

n −1
( y k +1 − y k )( x k +1 + x k ) ( y1 − y n )( x1 + x n )
Thus, Area = ∫∫ dA = ∑
R k =1 2
+
2
.

Meditate on this result. It is really a very simple formula for the area enclosed by a
polygon.

Example. We shall find the area of the quadrilateral with vertices (0, 0), (2, 4), (1, 7),
and (-1, 9):

-1 -0.5 0 0.5 1 1.5 2

x

Area =
1
[(4 − 0)(2 + 0) + (7 − 4)(2 + 1) + (9 − 7)(−1 + 1) + (9 − 0)(−1 + 0)] = 4]
2

Exercises
8. Find the area enclosed by the octagon with vertices (0, 0), (1, 0), (2, 3), (0, 5), (-2, 2),
(-1, -1), (-2, -2), (-1, -3).

9. By means of a clever choice of the function F( x, y ) , use Green’s Theorem and derive a

recipe for the integral ∫∫ xdA , where R is the region enclosed by the polygon with
R

vertices ( x1 , y1 ), ( x 2 , y 2 ),K, ( x n , y n ).

17.12
10. By means of a clever choice of the function F( x, y ) , use Green’s Theorem and derive

a recipe for the integral ∫∫ ydA , where R is the region enclosed by the polygon with
R

vertices ( x1 , y1 ), ( x 2 , y 2 ),K, ( x n , y n ).

11. Find the centroid of the region enclosed by the triangle with vertices (1, 1), (2, 8), and
(5, 5).

17.13
 Chapter Eighteen

Stokes

18.1 Stokes's Theorem

Let F : D → R 3 be a nice vector function. If
F ( x , y , z ) = p( x , y , z ) i + q ( x , y , z ) j + r ( x , y , z ) k ,
the curl of F is defined by

 ∂ r ∂ q  ∂ p ∂ r   ∂ q ∂ p
curlF =  − i + −  j + − k .
 ∂ y ∂ z   ∂ z ∂ x  ∂ x ∂ y

∂ ∂ ∂
Here also the so-called del operator ∇ = i +j +k provides a nice
∂x ∂y ∂z
memory device:

i j k
∂ ∂ ∂
curlF = ∇ × F = .
∂x ∂y ∂z
p q r

This definition allows us to look at Green's Theorem from a new perspective by

observing that in case F ( x , y ) = p( x , y ) i + q ( x , y ) j , Green's Theorem becomes

(♥) ∫ F ⋅ dr = ∫∫ curlF ⋅ dS ,
C R

where we are thinking of the region R as an oriented surface with its orientation pointing
in the direction of k.
We want to look at this formula in case the region R is not necessarily in the i-j
plane, in which case, the word "clockwise" doesn't help in deciding on the orientation of
the boundary C. Once again, we orient things according to our familiar "right-hand" rule.

18.1
Here's the way it goes. Suppose now S is any surface bounded by a finite number of
disjoint curves C1 , C2 ,K Cn . We say simply that C = C1 ∪ C2 ∪K∪Cn is the boundary of
S. Now choose an orientation for the surface S. Look at one of these normal vectors
"close" to a curve C j and imagine a little circle around the base of the normal oriented so

that the normal vector points in the right-hand direction with respect to the direction of
the circle. Then the orientation, or direction, of C j that is consistent with the given

orientation of the surface S is the one that "lines up" with the direction on this little circle.
Look at this picture:

The surface and its boundary in this case are said the be consistently oriented.
Now we do what we have done so many times in the past. Look at a surface S in
three space bounded by C. (Here neither S nor C are assumed to lie in a plane.)
Approximate the surface by a bunch of plane regions tangent to S , apply the equation (♥)
to each of these approximating plane regions, and then sum these equations. The sum of
the surface integrals is just the surface integral over the union of the approximating pieces,
and the sum of the line integrals is just the line integral around the boundary of the union
of the pieces—as in the plane case, the line integrals over the boundaries of adjacent
regions cancel. Then, of course, we think of looking at the limit as we take more and more
approximating regions, etc., and we obtain the equation

∫ F ⋅ dr = ∫∫ curlF ⋅ dS ,
C S

18.2
where S and C are oriented consistently. This result is the celebrated Stokes's Theorem.

Example
Let's use Stokes's Theorem to evaluate the line integral

∫ [− y i + x 3 j − z 3 k ] ⋅ dr ,
3

where C is the intersection of the cylinder x 2 + y 2 = 1 and the plane x + y + z = 1

oriented in the clockwise direction when viewed from above (i.e., looking in the direction
of -k.). The curve C bounds the part of the plane x + y + z = 1 that lies above the set of

( x , y ) such that x 2 + y 2 ≤ 1 . A vector description is thus given by

r ( s, t ) = s cos ti + s sin tj + (1 − s cos t − s sin t ) j , 0 ≤ s ≤ 1, 0 ≤ t ≤ 2π .

Hence,

i j k
∂r ∂r
× = cos t sin t − (cos t + sin t )
∂s ∂t
− s sin t s cos t s(sin t − cos t )

= si + sj + sk

I hope this result is no surprise. Notice that this is the opposite of the orientation
consistent with that specified for the curve C, and so we must use

∂r ∂r
× = − s( i + j + k )
∂t ∂s

in our surface integral. The surface integral ∫∫ curlF ⋅ dS looks like

S

18.3
 2π 1
∂ r ∂ r
∫∫ curlF ⋅ dS =
S
∫ ∫ curlF ⋅  ∂ t × ∂ s dsdt .
0 0

We must find curl F:

i j k
∂ ∂ ∂
curlF = ∇ × F = = 3( x 2 + y 2 ) k .
∂x ∂y ∂z
− y3 x3 −z3

Hence,
2π 1
∂ r ∂ r
∫∫ curlF ⋅ dS =
S
∫ ∫ curlF ⋅  ∂ t × ∂ s dsdt
0 0

2π 1
3 3
= ∫ ∫ 3( s )( − s)dsdt = −2π =− π
2

0 0
4 2

Exercises

1. Let S be the surface S = S1 ∪ S 2 , where S1 = {( x , y , z ): x 2 + y 2 = 1, 0 ≤ z ≤ 1} , and

S 2 = {( x , y , z ): x 2 + y 2 + ( z − 1) 2 = 1, z ≥ 1} . Let the function F be given by

F( x, y , z ) = ( x 2 z 3 + y )i + ( xy + z ) j + (5 x z + y 4 )k .
Compute the flux integral

∫∫ ∇ × F ⋅ dS ,
S

where S has the orientation pointing away from the z- axis.

18.4
2. Let S be the hemisphere x 2 + y 2 + z 2 = 1, z ≤ 0 with the orientation pointing toward
the origin.
a)Describe the boundary of S and its orientation that is consistent with the orientation
of S.

b)Evaluate the flux ∫∫ ∇ × F ⋅ dS

S
, where F( x, y , z ) = 2 yi + xj + zk

3. Let S1 and S 2 be two surfaces with a common boundary C. Draw a picture indicating
the orientations these surfaces must have to insure that

∫∫ ∇ × F ⋅ dS = ∫∫ ∇ × F ⋅ dS .
S1 S21

4. Let S be a surface with boundary C . Suppose they are consistently oriented. Suppose
a is a constant vector. Prove that

∫ (a × r ) ⋅ dr = 2∫∫ a ⋅ dS .
C S

[Remember, r = xi + yj + zk .]

5. Suppose S is a surface with boundary C and F is a vector function such that ∇ × F is

tangent to S at each point of S. Prove that ∫ F ⋅ dr = 0 .

C

6. Let r : B → R 3 be a vector description of the surface S with boundary C. Let F be a

vector function such that

18.5
 1  ∂r ∂r 
∇×F =  × .
∂r ∂r  ∂s ∂t 
×
∂s ∂t

Show that ∫ F ⋅ dc
C
= area of S.

7. Suppose the vector function F on a domain D is conservative. Prove that ∇ × F = 0

everywhere in D.

−y x
8. Let F( x, y , z ) = i+ 2 j, x2 + y 2 ≠ 0 .
x +y
2 2
x +y 2

a)Compute ∇ × F .

b)Prove that F is not conservative. [Hint: Evaluate the line integral ∫ F ⋅ dr , where C
C

is the circle x 2 + y 2 = 1 , z = 0 , with the usual counterclockwise orientation.]

18.2 Path Independence Revisited

Problem 7 at the end of the previous section perhaps raised our hopes that an easy
test for a function F to be conservative in a domain D is simply to see if ∇ × F =0. If so,
these hopes were quickly dashed by Problem 8. In this section, we shall see just what we
can do along this line. The concept introduced next provides the key to understanding and
enlightenment.
An open subset D of R 3 is called simply connected if every simple closed curve
in D is the boundary of some surface contained entirely in D. Thus for instance the region
D = {( x, y, z ) : x 2 + y 2 + z 2 < 1}
is simply connected, while the region
R = {( x, y , z ) : x 2 + y 2 > 1}
is not.

18.6
 Now it easy to see that if F has as domain a simply connected region D, then
∇ × F = 0 everywhere in D implies that F is indeed conservative. We show that F is
conservative by showing that the integral of F around any closed curve is 0. This is easy
to do. Let C be any closed curve in D. Then D is simply connected, so there is a surface S
the boundary of which is C. Now unleash Stokes’s Theorem:

∫ F ⋅ dr = ∫∫ ∇ × F ⋅ dS = 0.
C S

How about that!

Exercises
−y x
9. Explain how you know that F( x, y , z ) = i+ 2 j , x > 0. is conservative.
x +y
2 2
x + y2

10. Find a potential function for the vector function F given in Problem 9.

18.7
 Chapter Nineteen

Some Physics

19.1 Fluid Mechanics

Suppose v ( x, y, z , t ) is the velocity at r = ( x, y, z ) = xi + yj + zk of a fluid flowing
smoothly through a region in space, and suppose ρ ( x, y , z , t ) is the density at r at time t. If
S is an oriented surface, it is not hard to convince yourself that the flux integral

∫∫ ρv ⋅ dr
S

is the rate at which mass flows through the surface S. Now, if S is a closed surface, then
the mass in the region B bounded by S is, of course

∫∫∫ ρdV
B
.

The rate at which this mass is changing is simply

∂ ∂ρ
∫∫∫
∂t B
ρdV = ∫∫∫
B
∂t
dV .

This is the same as the rate at which mass is flowing across S into B: − ∫∫ ρv ⋅ dr , where S
S

is given the outward pointing orientation. Thus,

∂ρ
∫∫∫ ∂t dV = −∫∫ ρv ⋅ dr .
B S

We now apply Gauss’s Theorem and get

∂ρ
∫∫∫ ∂t dV = −∫∫ ρv ⋅ dr = ∫∫∫ − ∇ ⋅ (ρv)dV .
B S B

Thus,
 ∂ρ 
∫∫∫  ∂t
B
+ ∇ ⋅ ( ρv ) dV .

Meditate on this result. The region B is any region, and so it must be true that the
integrand itself is everywhere 0:

19.1
 ∂ρ
+ ∇ ⋅ ( ρv) = 0 .
∂t
This is one of the fundamental equations of fluid dynamics. It is called the equation of
continuity.
In case the fluid is incompressible, the continuity equation becomes quite simple.
∂ρ
Incompressible means simply that the density ρ is constant. Thus = 0 and so we have
∂t
∂ρ
+ ∇ ⋅ ( ρv ) = ∇ ⋅ ( ρv) = ρ∇ ⋅ v = 0, or
∂t
∇⋅v = 0.
Exercise
1. Consider a one dimensional flow in which the velocity of the fluid is given by
v = f (x ) , where f ( x ) > 0 . Suppose further that the density ρ of the fluid does not vary
with time t. Show that
k
ρ ( x) = ,
f ( x)
where k is a constant.

19.2 Electrostatics
Suppose there is a point charge q fixed at the point s. Then the electric field
E q (r ) due to q is given by

r−s
E q (r ) = kq .
r−s
3

It is easy to verify, as we have done in a previous chapter, that this field, or function, is
conservative, with a potential function
− kq
Pq (r ) = ;
|r −s |
so that E q = ∇Pq . Physicists do not like to be bothered with the minus sign in Pq , so they

define the electric potential Vq to be - Pq . Thus,

19.2
 kq
Vq (r ) = ,
|r −s|
and
E q (r ) = −∇Vq (r ) .

We have already seen that the flux out of a closed surface S is

 0 if S does not enclose the origin
∫∫ E
S
q ⋅ dS = 
4πkq if S does enclose the origin

Some meditation will convince you there is nothing special here about the origin; that is, if
the point charge is at s, then
 0 if S does not enclose s
∫∫ ES
q ⋅ dS = 
4πkq if S does enclose s

Next, suppose there are a finite number of point charges q1 at s1 , q 2 at s 2 , K , and

q n at s n . Suppose E j is the electric intensity due to q j . Then it should be clear that the

electric field due to these charges is simply the sum

n n r−sj
E(r ) = ∑ E j = k ∑ q j .
j =1 j =1 | r − s j |3

Also,
n qj
V (r ) = k ∑ ; and
j =1 |r−sj |

E(r ) = −∇V (r ) .
Finally,

∫∫ E ⋅ dS = 4πk ∑ q
S
j

where the sum is over those charges enclosed by S.

19.3
 Things become more exciting if instead of point charges, we have a charge
distribution in space with charge density ρ . To find the electric field E(r ) produced by
this distribution of charge in space, we need to integrate:
(r − s )
E(r ) = ∫∫∫ kρ (s) dV s .
U
| r − s |3

But this appears to be a serious breach of decorum. We are integrating over everything,
and at s = r we have the dreaded 0 in the denominator. Thus what we see above is an
improper integral—that is, it is actually a limit of integrals. Specifically, we integrate not
over everything but over everything outside a spherical solid region of radius a centered at
r. We then look at the limit as a → 0 of this integral. With the integral for the electric
field, this limit exists, and so there is no problem with 0 on the bottom of the integrand. In
the same way, we are safe in writing for the potential
ρ (s)
V (r ) = k ∫∫∫ dVs .
U
|r−s|

Everything works nicely so that we also have

E(r ) = −∇V (r ) .
If R is a solid region bounded by a closed surface S, then we can also integrate to get

(*) ∫∫ E ⋅ dS = 4πk ∫∫∫ ρ (s)dV .

S R

The divergence of E is the troublesome item in extending matters to distributed

charge. If we simply try to calculate the divergence by div ∫∫∫ stuff dV = ∫∫∫ div (stuff) dV ,
U U

then things go wrong because the improper integral of the divergence does not exist.
Gauss saves the day. Let R be any region and let S be the closed surface bounding R.
Then

∫∫ E ⋅ dS = ∫∫∫ ∇ ⋅ E dV .
S R

But from equation (*) we have

∫∫ E ⋅ dS = 4πk ∫∫∫ ρ (s)dV = ∫∫∫ ∇ ⋅ E dV .

S R R

This gives us

19.4
 ∫∫∫ 4πkρdV = ∫∫∫ ∇ ⋅ E dV , or
R R

∫∫∫ (∇ ⋅ E − 4πkρ )dV .

R

But R is any region, and so it must be true that

∇ ⋅ E = 4πkρ
for all r.
Finally, remembering that E = −∇V , we get
∇ ⋅ E = −∇ ⋅ (∇V ) = 4πkρ ;

∇ 2V = −4πkρ , or

∂ 2V ∂ 2V ∂ 2V
+ + = −4πkρ .
∂x 2 ∂y 2 ∂z 2

This is the celebrated Poisson’s Equation, a justly famous partial differential equation, the
study of which is beyond the scope of this course.

19.5