This action might not be possible to undo. Are you sure you want to continue?
Internet Supplement
for Vector Calculus
Fifth Edition
Version: October, 2003
Jerrold E. Marsden
California Institute of Technology
Anthony Tromba
University of California, Santa Cruz
W.H. Freeman and Co.
New York
Page i
Contents
Preface iii
2 Diﬀerentiation 1
2.7 Some Technical Diﬀerentiation Theorems . . . . . . . . . . . 1
3 HigherOrder Derivatives and Extrema 17
3.4A Second Derivative Test: Constrained Extrema . . . . . . . 17
3.4B Proof of the Implicit Function Theorem . . . . . . . . . . 21
4 Vector Valued Functions 25
4.1A Equilibria in Mechanics . . . . . . . . . . . . . . . . . . . 25
4.1B Rotations and the Sunshine Formula . . . . . . . . . . . . 30
4.1C The Principle of Least Action . . . . . . . . . . . . . . . . 42
4.4 Flows and the Geometry of the Divergence . . . . . . . . . 58
5 Double and Triple Integrals 65
5.2 Alternative Deﬁnition of the Integral . . . . . . . . . . . . 65
5.6 Technical Integration Theorems . . . . . . . . . . . . . . . 71
6 Integrals over Curves and Surfaces 83
6.2A A Challenging Example . . . . . . . . . . . . . . . . . . . 84
6.2A The Gaussian Integral . . . . . . . . . . . . . . . . . . . . 88
8 The Integral Theorems of Vector Analysis 91
8.3 Exact Diﬀerentials . . . . . . . . . . . . . . . . . . . . . . . 91
Page ii
8.5 Green’s Functions . . . . . . . . . . . . . . . . . . . . . . . 94
Selected Answers for the Internet Supplement 107
Practice Final Examination 123
Practice Final Examination Solutions 129
Page iii
Preface
The Structure of this Supplement. This Internet Supplement is in
tended to be used with the 5th Edition of our text Vector Calculus. It
contains supplementary material that gives further information on various
topics in Vector Calculus, including diﬀerent applications and also technical
proofs that were omitted from the main text.
The supplement is intended for students who wish to gain a deeper un
derstanding, usually by self study, of the material—both for the theory as
well as the applications.
Corrections and Website. A list of corrections and suggestions con
cerning the text and instructors guide are available available on the book’s
website:
http://www.whfreeman.com/MarsdenVC5e
Please send any new corrections you may ﬁnd to one of us.
More Websites. There is of course a huge number of websites that con
tain a wealth of information. Here are a few sample sites that are relevant
for the book:
1. For spherical geometry in Figure 8.2.13 of the main text, see
http://torus.math.uiuc.edu/jms/java/dragsphere/
which gives a nice JAVA applet for parallel transport on the sphere.
2. For further information on the Sunshine formula (see ¸4.1C of this
internet supplement), see
http://www.math.niu.edu/~rusin/usesmath/position.sun/
Page iv
3. For the Genesis Orbit shown in Figure 4.1.11, see
http://genesismission.jpl.nasa.gov/
4. For more on Newton, see for instance,
http://scienceworld.wolfram.com/biography/Newton.html
and for Feynman, see
http://www.feynman.com/
5. For surface integrals using Mathematica Notebooks, see
http://www.math.umd.edu/~jmr/241/surfint.htm
Practice Final Examination. At the end of this Internet Supplement,
students will ﬁnd a Practice Final Examination that covers topics in the
whole book, complete with solutions. We recommend, if you wish to prac
tice your skills, that you allow yourself 3 hours to take the exam and then
selfmark it, keeping in mind that there is often more than one way to
approach a problem.
Acknowledgements. As with the main text, the student guide, and the
Instructors Manual, we are very grateful to the readers of earlier editions
of the book for providing valuable advice and pointing out places where
the text can be improved. For this internet supplement, we are especially
grateful to Alan Weinstein for his collaboration in writing the supplement
on the sunshine formula (see the supplement to Chapter 4) and for making
a variety of other interesting and useful remarks. We also thank Brian
Bradie and Dave Rusin for their helpful comments.
We send everyone who uses this supplement and the book our best re
gards and hope that you will enjoy your studies of vector calculus and that
you will beneﬁt (both intellectually and practically) from it.
Jerrold Marsden (marsden@cds.caltech.edu)
Control and Dynamical Systems
Caltech 10781
Pasadena, CA 91125
Anthony Tromba (tromba@math.ucsc.edu)
Department of Mathematics
University of California
Santa Cruz, CA 95064
Page 1
2
Diﬀerentiation
In the ﬁrst edition of Principia Newton admitted that Leib
niz was in possession of a similar method (of tangents) but
in the third edition of 1726, following the bitter quarrel be
tween adherents of the two men concerning the independence
and priority of the discovery of the calculus, Newton deleted
the reference to the calculus of Leibniz. It is now fairly clear
that Newton’s discovery antedated that of Leibniz by about
ten years, but that the discovery by Leibniz was independent
of that of Newton. Moreover, Leibniz is entitled to priority of
publication, for he printed an account of his calculus in 1684
in the Acta Eruditorum, a sort of “scientiﬁc monthly” that had
been established only two years before.
Carl B. Boyer
A History of Mathematics
§2.7 Some Technical Diﬀerentiation Theorems
In this section we examine the mathematical foundations of diﬀerential
calculus in further detail and supply some of the proofs omitted from ¸¸2.2,
2.3, and 2.5.
Limit Theorems. We shall begin by supplying the proofs of the limit
theorems presented in ¸2.2 (the theorem numbering in this section cor
2 2 Diﬀerentiation
responds to that in Chapter 2). We ﬁrst recall the deﬁnition of a limit.
1
Deﬁnition of Limit. Let f : A ⊂ R
n
→ R
m
where A is open. Let
x
0
be in A or be a boundary point of A, and let N be a neighborhood of
b ∈ R
m
. We say f is eventually in N as x approaches x
0
if there
exists a neighborhood U of x
0
such that x ,= x
0
, x ∈ U, and x ∈ A implies
f(x) ∈ N. We say f(x) approaches b as x approaches x
0
, or, in symbols,
lim
x→x0
f(x) = b or f(x) → b as x → x
0
,
when, given any neighborhood N of b, f is eventually in N as x approaches
x
0
. If, as x approaches x
0
, the values f(x) do not get close to any particular
number, we say that lim
x→x0
f(x) does not exist.
Let us ﬁrst establish that this deﬁnition is equivalent to the εδ formu
lation of limits. The following result was stated in ¸2.2.
Theorem 6. Let f : A ⊂ R
n
→ R
m
and let x
0
be in A or be a boundary
point of A. Then lim
x→x0
f(x) = b if and only if for every number ε > 0
there is a δ > 0 such that for any x ∈ A satisfying 0 < x − x
0
 < δ, we
have f(x) −b < ε.
Proof. First let us assume that lim
x→x0
f(x) = b. Let ε > 0 be given,
and consider the ε neighborhood N = D
ε
(b), the ball or disk of radius
ε with center b. By the deﬁnition of a limit, f is eventually in D
ε
(b), as
x approaches x
0
, which means there is a neighborhood U of x
0
such that
f(x) ∈ D
ε
(b) if x ∈ U, x ∈ A, and x ,= x
0
. Now since U is open and x
0
∈ U,
there is a δ > 0 such that D
δ
(x
0
) ⊂ U. Consequently, 0 < x − x
0
 < δ
and x ∈ A implies x ∈ D
δ
(x
0
) ⊂ U. Thus f(x) ∈ D
ε
(b), which means that
f(x) −b < ε. This is the εδ assertion we wanted to prove.
We now prove the converse. Assume that for every ε > 0 there is a
δ > 0 such that 0 < x − x
0
 < δ and x ∈ A implies f(x) − b < ε.
Let N be a neighborhood of b. We have to show that f is eventually
in N as x → x
0
; that is, we must ﬁnd an open set U ⊂ R
n
such that
x ∈ U, x ∈ A, and x ,= x
0
implies f(x) ∈ N. Now since N is open, there is
an ε > 0 such that D
ε
(b) ⊂ N. If we choose U = D
δ
(x) (according to our
assumption), then x ∈ U, x ∈ A and x ,= x
0
means f(x) − b < ε, that
is f(x) ∈ D
ε
(b) ⊂ N.
Properties of Limits. The following result was also stated in ¸2.2. Now
we are in a position to provide the proof.
1
For those interested in a diﬀerent pedagogical approach to limits and the derivative,
we recommend Calculus Unlimited by J. Marsden and A. Weinstein. It is freely available
on the Vector Calculus website given in the Preface.
2.7 Diﬀerentiation Theorems 3
Theorem 2. Uniqueness of Limits. If
lim
x→x0
f(x) = b
1
and lim
x→x0
f(x) = b
2
,
then b
1
= b
2
.
Proof. It is convenient to use the εδ formulation of Theorem 6. Suppose
f(x) → b
1
and f(x) → b
2
as x → x
0
. Given ε > 0, we can, by assumption,
ﬁnd δ
1
> 0 such that if x ∈ A and 0 < x−x
0
 < δ
2
, then f(x)−b
1
 < ε,
and similarly, we can ﬁnd δ
1
> 0 such that 0 < x − x
0
 < δ
2
implies
f(x) − b
2
 < ε. Let δ be the smaller of δ
1
and δ
2
. Choose x such that
0 < x − x
0
 < δ and x ∈ A. Such x’s exist, because x
0
is in A or is a
boundary point of A. Thus, using the triangle inequality,
b
1
−b
2
 = (b
1
−f(x)) + (f(x) −b
2
)
≤ b
1
−f(x) +f(x) −b
2
 < ε +ε = 2ε.
Thus for every ε > 0, b
1
− b
2
 < 2ε. Hence b
1
= b
2
, for if b
1
,= b
2
we
could let ε = b
1
−b
2
/2 > 0 and we would have b
1
−b
2
 < b
1
−b
2
,
an impossibility.
The following result was also stated without proof in ¸2.2.
Theorem 3. Properties of Limits. Let f : A ⊂ R
n
→ R
m
, g : A ⊂
R
n
→R
m
, x
0
be in A or be a boundary point of A, b ∈ R
m
, and c ∈ R; the
following assertions than hold:
(i) If lim
x→x0
f(x) = b, then lim
x→x0
cf(x) = cb, where cf : A → R
m
is deﬁned by x −→ c(f(x)).
(ii) If lim
x→x0
f(x) −b
1
and lim
x→x0
g(x) = b
2
, then
lim
x→x0
(f +g)(x) = b
1
+b
2
,
where (f +g) : A →R
m
is deﬁned by x −→ f(x) +g(x).
(iii) If m = 1, lim
x→x0
f(x) = b
1
, and lim
x→x0
g(x) = b
2
, then
lim
x→x0
(fg)(x) = b
1
b
2
,
where (fg) : A →R is deﬁned by x −→ f(x)g(x).
(iv) If m = 1, lim
x→x0
f(x) = b ,= 0, and f(x) ,= 0 for all x ∈ A, then
lim
x→x0
1
f
=
1
b
,
where 1/f : A →R is deﬁned by x −→ 1/f(x).
4 2 Diﬀerentiation
(v) If f(x) = (f
1
(x), . . . , f
m
(x)), where f
i
: A →R, i = 1, . . . , m, are the
component functions of f, then lim
x→x0
f(x) = b = (b
1
, . . . , b
m
) if
and only if lim
x→x0
f
i
(x) = b
i
for each i = 1, . . . , m.
Proof. We shall illustrate the technique of proof by proving assertions (i)
and (ii). The proofs of the other assertions are only a bit more complicated
and may be supplied by the reader. In each case, the εδ formulation of
Theorem 6 is probably the most convenient approach.
To prove rule (i), let ε > 0 be given; we must produce a number δ > 0
such that the inequality cf(x)−cb < ε holds if 0 < x−x
0
 < δ. If c = 0,
any δ will do, so we can suppose c ,= 0. Let ε
= ε/[c[; from the deﬁnition
of limit, there is a δ with the property that 0 < x − x
0
 < δ implies
f(x) − b < ε
= ε/[c[. Thus 0 < x − x
0
 < δ implies cf(x) − cb =
[c[f(x) −b[ < ε, which proves rule (i).
To prove rule (ii), let ε > 0 be given again. Choose δ
1
> 0 such that
0 < x−x
0
 < δ
1
implies f(x) −b
1
 < ε/2. Similarly, choose δ
2
> 0 such
that 0 < x−x
0
 < δ
2
implies g(x) −b
2
 < ε/2. Let δ be the lesser of δ
1
and δ
2
. Then 0 < x −x
0
 < δ implies
f(x) +g(x) −b
1
−b
2
 ≤ f(x) −b
1
 +g(x) −b
2
 <
ε
2
+
ε
2
= ε.
Thus, we have proved that (f +g)(x) → b
1
+b
2
as x → x
0
.
Example 1. Find the following limit if it exists:
lim
(x,y)→(0,0)
_
x
3
−y
3
x
2
+y
2
_
.
Solution. Since
0 ≤
¸
¸
¸
¸
x
3
−y
3
x
2
+y
2
¸
¸
¸
¸
≤
[x[x
2
+[y[y
2
x
2
+y
2
≤
([x[ +[y[)(x
2
+y
2
)
x
2
+y
2
= [x[ +[y[,
we ﬁnd that
lim
(x,y)→(0,0)
x
3
−y
3
x
2
+y
2
= 0.
Continuity. Now that we have the limit theorems available, we can use
this to study continuity; we start with the basic deﬁnition of continuity of
a function.
Deﬁnition. Let f : A ⊂ R
n
→ R
m
be a given function with domain A.
Let x
0
∈ A. We say f is continuous at x
0
if and only if
lim
x→x0
f(x) = f(x
0
).
If we say that f is continuous, we shall mean that f is continuous at each
point x
0
of A.
2.7 Diﬀerentiation Theorems 5
From Theorem 6, we get the εδ criterion for continuity.
Theorem 7. A mapping f : A ⊂ R
n
→ R
m
is continuous at x
0
∈ A if
and only if for every number ε > 0 there is a number δ > 0 such that
x ∈ A and x −x
0
 < δ implies f(x) −f(x
0
) < ε.
One of the properties of continuous functions stated without proof in
¸2.2 was the following:
Theorem 5. Continuity of Compositions. Let f : A ⊂ R
n
→R
m
and
let g : B ⊂ R
m
→ R
p
. Suppose f(A) ⊂ B so that g ◦ f is deﬁned on A. If
f is continuous at x
0
∈ A and g is continuous at y
0
= f(x
0
), then g ◦ f is
continuous at x
0
.
Proof. We use the εδ criterion for continuity. Thus, given ε > 0, we
must ﬁnd δ > 0 such that for x ∈ A.
x −x
0
 < δ implies (g ◦ f)(x) −(g ◦ f)(x
0
) < ε.
Since g is continuous at f(x
0
) = y
0
∈ B, there is a γ > 0 such that for
y ∈ B,
y −y
0
 < γ implies g(y) −g(f(x
0
)) < ε.
Since f is continuous x
0
∈ A, there is, for this γ, a δ > 0 such that for
x ∈ A,
x −x
0
 < δ implies f(x) −f(x
0
) < γ,
which in turn implies
g(f(x)) −g(f(x
0
)) < ε,
which is the desired conclusion.
Diﬀerentiability. The exposition in ¸2.3 was simpliﬁed by assuming, as
part of the deﬁnition of Df(x
0
), that the partial derivatives of f existed.
Our next objective is to show that this assumption can be omitted. Let us
begin by redeﬁning “diﬀerentiable.” Theorem 15 below will show that the
new deﬁnition is equivalent to the old one.
Deﬁnition. Let U be an open set in R
n
and let f : U ⊂ R
n
→ R
m
be a
given function. We say that f is diﬀerentiable at x
0
∈ U if and only if
there exists an mn matrix T such that
lim
x→x0
f(x) −f(x
0
) −T(x −x
0
)
x −x
0

= 0. (1)
6 2 Diﬀerentiation
We call T the derivative of f at x
0
and denote it by Df(x
0
). In matrix
notation, T(x −x
0
) stands for
_
¸
¸
¸
_
T
11
T
12
. . . T
1n
T
21
T
22
. . . T
2n
.
.
.
.
.
.
.
.
.
T
m1
T
m2
. . . T
mn
_
¸
¸
¸
_
_
¸
_
x
1
−x
01
.
.
.
x
n
−x
0n
_
¸
_.
where x = (x
1
, . . . , x
n
), x
0
= (x
01
, . . . , x
0n
), and where the matrix entries
of T are denoted [T
ij
]. Sometimes we write T(y) as T y or just Ty for
the product of the matrix T with the column vector y.
Condition (1) can be rewritten as
lim
h→0
f(x
0
+h) −f(x
0
) −Th
h
= 0 (2)
as we see by letting h = x−x
0
. Written in terms of εδ notation, equation
(2) says that for every ε > 0 there is a δ > 0 such that 0 < h < δ implies
f(x
0
+h) −f(x
0
) −Th
h
< ε,
or, in other words,
f(x
0
+h) −f(x
0
) −Th < εh.
Notice that because U is open, as long as δ is small enough, h < δ implies
x
0
+h ∈ U.
Our ﬁrst task is to show that the matrix T is necessarily the matrix of
partial derivatives, and hence that this abstract deﬁnition agrees with the
deﬁnition of diﬀerentiability given in ¸2.3.
Theorem 15. Suppose f : U ⊂ R
n
→ R
m
is diﬀerentiable at x
0
∈ R
n
.
Then all the partial derivatives of f exist at the point x
0
and the m n
matrix T has entries given by
[T
ij
] =
_
∂f
i
∂x
j
_
,
that is,
T = Df(x
0
) =
_
¸
¸
¸
¸
_
∂f
1
∂x
1
. . .
∂f
1
∂x
n
.
.
.
.
.
.
∂f
m
∂x
1
. . .
∂f
m
∂x
n
_
¸
¸
¸
¸
_
,
where ∂f
i
/∂x
j
is evaluated at x
0
. In particular, this implies that T is
uniquely determined; that is, there is no other matrix satisfying condition
(1).
2.7 Diﬀerentiation Theorems 7
Proof. By Theorem 3(v), condition (2) is the same as
lim
h→0
[f
i
(x
0
+h) −f
i
(x
0
) −(Th)
i
[
h
= 0, 1 ≤ i ≤ m.
Here (Th
i
) stands for the ith component of the column vector Th. Now
let h = ae
j
= (0, . . . , a, . . . , 0), which has the number a in the jth slot and
zeros elsewhere. We get
lim
a→0
[f
i
(x
0
+ae
j
) −f
i
(x
0
) −a(Te
j
)
i
[
[a[
= 0,
or, in other words,
lim
a→0
¸
¸
¸
¸
f
i
(x
0
+ae
j
) −f
i
(x
0
)
a
−(Te
j
)
i
¸
¸
¸
¸
= 0,
so that
lim
a→0
f
i
(x
0
+ae
j
) −f
i
(x
0
)
a
= (Te
j
)
i
.
But this limit is nothing more than the partial derivative ∂f
i
/∂x
j
evaluated
at the point x
0
. Thus, we have proved that ∂f
i
/∂x
j
exists and equals
(Te
j
)
i
. But (Te
j
)
i
= T
ij
(see ¸1.5 of the main text), and so the theorem
follows.
Diﬀerentiability and Continuity. Our next task is to show that dif
ferentiability implies continuity.
Theorem 8. Let f : U ⊂ R
n
→ R
m
be diﬀerentiable at x
0
. Then f is
continuous at x
0
, and furthermore, f(x) −f(x
0
) < M
1
x−x
0
 for some
constant M
1
and x near x
0
, x ,= x
0
.
Proof. We shall use the result of Exercise 2 at the end of this section,
namely that,
Df(x
0
) h ≤ Mh,
where M is the square root of the sum of the squares of the matrix elements
in Df(x
0
).
Choose ε = 1. Then by the deﬁnition of the derivative (see formula (2))
there is a δ
1
> 0 such that 0 < h < δ
1
implies
f(x
0
+h) −f(x
0
) −Df(x
0
) h < εh = h.
If h < δ
1
, then using the triangle inequality,
f(x
0
+h) −f(x
0
) = f(x
0
+h) −f(x
0
) −Df(x
0
) h +Df(x
0
) h
≤ f(x
0
+h) −f(x
0
) −Df(x
0
) h +Df(x
0
) h
< h +Mh = (1 +M)h.
8 2 Diﬀerentiation
Setting x = x
0
+ h and M
1
= 1 + M, we get the second assertion of the
theorem.
Now let ε
be any positive number, and let δ be the smaller of the two
positive numbers δ
1
and ε
/(1 +M). Then h < δ implies
f(x
0
+h) −f(x
0
) < (1 +M)
ε
1 +M
= ε
,
which proves that (see Exercise 15 at the end of this section)
lim
x→x0
f(x) = lim
h→0
f(x
0
+h) = f(x
0
),
so that f is continuous at x
0
.
Criterion for Diﬀerentiability. We asserted in ¸2.3 that an important
criterion for diﬀerentiability is that the partial derivatives exist and are
continuous. We now are able to prove this.
Theorem 9. Let f : U ⊂ R
n
→ R
m
. Suppose the partial derivatives
∂f
i
/∂x
j
of f all exist and are continuous in some neighborhood of a point
x ∈ U. Then f is diﬀerentiable at x.
Proof. In this proof we are going to use the mean value theorem from one
variable calculus—see ¸2.5 of the main text for the statement. To simplify
the exposition, we shall only consider the case m = 1, that is, f : U ⊂
R
n
→ R, leaving the general case to the reader (this is readily supplied
knowing the techniques from the proof of Theorem 15, above).
According to the deﬁnition of the derivative, our objective is to show
that
lim
h→0
¸
¸
¸
¸
f(x +h) −f(x) −
n
i=1
_
∂f
∂x
i
(x)
_
h
i
¸
¸
¸
¸
h
= 0.
Write
f(x
1
+h
1
, . . . , x
n
+h
n
) −f(x
1
, . . . , x
n
)
= f(x
1
+h
1
, . . . , x
n
+h
n
) −f(x
1
, x
2
+h
2
, . . . , x
n
+h
n
)
+f(x
1
, x
2
+h
2
, . . . , x
n
+h
n
) −f(x
1
, x
2
, x
3
+h
3
, . . . , x
n
+h) +. . .
+f(x
1
, . . . , x
n−1
+h
n−1
, x
n
+h
n
) −f(x
1
, . . . , x
n−1
, x
n
+h
n
)
+f(x
1
, . . . , x
n−1
, x
n
+h
n
) −f(x
1
, . . . , x
n
).
This is called a telescoping sum, since each term cancels with the suc
ceeding or preceding one, except the ﬁrst and the last. By the mean value
theorem, this expression may be written as
f(x +h) −f(x) =
_
∂f
∂x
1
(y
1
)
_
h
1
+
_
∂f
∂x
2
(y
2
)
_
h
2
+. . . +
_
∂f
∂x
n
(y
2
)
_
h
n
,
2.7 Diﬀerentiation Theorems 9
where y
1
= (c
1
, x
2
+ h
2
, . . . , x
n
+ h
n
) with c
1
lying between x
1
and x
1
+
h
1
; y
2
= (x
1
, c
2
, x
3
+h
3
, . . . , x
n
+h
n
) with c
2
lying between x
2
and x
2
+h
2
;
and y
n
= (x
1
, . . . , x
n−1
, c
n
) where c
n
lies between x
n
and x
n
+ h
n
. Thus,
we can write
¸
¸
¸
¸
¸
f(x +h) −f(x) −
n
i=1
_
∂f
∂x
i
(x)
_
h
i
¸
¸
¸
¸
¸
=
¸
¸
¸
¸
_
∂f
∂x
1
(y
1
) −
∂f
∂x
1
(x)
_
h
1
+. . . +
_
∂f
∂x
n
(y
n
) −
∂f
∂x
n
(x)
_
h
n
¸
¸
¸
¸
.
By the triangle inequality, this expression is less than or equal to
¸
¸
¸
¸
∂f
∂x
1
(y
1
) −
∂f
∂x
1
(x)
¸
¸
¸
¸
[h
1
[ +. . . +
¸
¸
¸
¸
∂f
∂x
n
(y
n
) −
∂f
∂x
n
(x)
¸
¸
¸
¸
[h
n
[
≤
_¸
¸
¸
¸
∂f
∂x
1
(y
1
) −
∂f
∂x
1
(x)
¸
¸
¸
¸
+. . . +
¸
¸
¸
¸
∂f
∂x
n
(y
n
) −
∂f
∂x
n
(x)
¸
¸
¸
¸
_
h.
since [h
i
[ ≤ h for all i. Thus, we have proved that
¸
¸
¸
¸
f(x +h) −f(x) −
n
i=1
_
∂f
∂x
i
(x)
_
h
i
¸
¸
¸
¸
h
≤
¸
¸
¸
¸
∂f
∂x
1
(y
1
) −
∂f
∂x
1
(x)
¸
¸
¸
¸
+. . . +
¸
¸
¸
¸
∂f
∂x
n
(y
n
) −
∂f
∂x
n
(x)
¸
¸
¸
¸
.
But since the partial derivatives are continuous by assumption, the right
side approaches 0 as h → 0 so that the left side approaches 0 as well.
Chain Rule. As explained in ¸2.5, the Chain Rule is right up there with
the most important results in diﬀerential calculus. We are now in a position
to give a careful proof.
Theorem 11: Chain Rule. Let U ⊂ R
n
and V ⊂ R
m
be open. Let
g : U ⊂ R
n
→ R
m
and f : V ⊂ R
m
→ R
p
be given functions such that g
maps U into V , so that f ◦ g is deﬁned. Suppose g is diﬀerentiable at x
0
and f is diﬀerentiable at y
0
= g(x
0
). Then f ◦ g is diﬀerentiable at x
0
and
D(f ◦ g)(x
0
) = Df(y
0
)Dg(x
0
).
Proof. According to the deﬁnition of the derivative, we must verify that
lim
x→x0
f(g(x)) −f(g(x
0
)) −Df(y
0
)Dg(x
0
) (x −x
0
)
x −x
0

= 0.
10 2 Diﬀerentiation
First rewrite the numerator and apply the triangle inequality as follows:
f(g(x)) −f(g(x
0
)) −Df(y
0
(g(x) −g(x
0
))
+Df(y
0
) [g(x) −g(x
0
) −Dg(x
0
) (x −x
0
)]
≤ f(g(x)) −f(g(x
0
)) −Df(y
0
) (g(x) −g(x
0
))
+Df(y
0
) [g(x) −g(x
0
) −Dg(x
0
) (x −x
0
]). (3)
As in the proof of Theorem 8, Df(y
0
) h ≤ Mh for some constant M.
Thus the righthand side of inequality (3) is less than or equal to
f(g(x)) −f(g(x
0
)) −Df(y
0
) (g(x) −g(x
0
))
+Mg(x) −g(x
0
) −Dg(x
0
) (x −x
0
). (4)
Since g is diﬀerentiable at x
0
, given ε > 0, there is a δ
1
> 0 such that
0 < x −x
0
 < δ
1
implies
g(x) −g(x
0
) −Dg(x
0
) (x −x
0
)
x −x
0

<
ε
2M
.
This makes the second term in expression (4) less than εx −x
0
/2.
Let us turn to the ﬁrst term in expression (4). By Theorem 8,
g(x) −g(x
0
) < M
1
x −x
0

for a constant M
1
if x is near x
0
, say 0 < x − x
0
 < δ
2
. Now choose δ
3
such that 0 < y −y
0
 < δ
3
implies
f(y) −f(y
0
) −Df(y
0
) (y −y
0
) <
εy −y
0

2M
1
.
Since y = g(x) and y
0
= g(x
0
), y − y
0
 < δ
3
if x − x
0
 < δ
3
/M
1
and
x −x
0
 < δ
2
, and so
f(g(x)) − f(g(x
0
)) −Df(y
0
) (g(x) −g(x
0
))
≤
εg(x) −g(x
0
)
2M
1
<
εx −x
0

2
.
Thus if δ = min(δ
1
, δ
2
, δ
3
/M
1
), expression (4) is less than
εx −x
0

2
+
εx −x
0

2
= εx −x
0
,
and so
f(g(x)) −f(g(x
0
)) −Df(y
0
)Dg(x
0
)(x −x
0
)
x −x
0

< ε
for 0 < x −x
0
 < δ. This proves the theorem.
2.7 Diﬀerentiation Theorems 11
A Crinkled Function. The student has already met with a number of
examples illustrating the above theorems. Let us consider one more of a
more technical nature.
Example 2. Let
f(x, y) =
_
_
_
xy
_
x
2
+y
2
(x, y) ,= (0, 0)
0 (x, y) = (0, 0).
Is f diﬀerentiable at (0, 0)? (See Figure 2.7.1.)
x
y
z
Figure 2.7.1. This function is not diﬀerentiable at (0, 0), because it is “crinkled.”
Solution. We note that
∂f
∂x
(0, 0) = lim
x→0
f(x, 0) −f(0, 0)
x
= lim
x→0
(x 0)/
√
x
2
+ 0 −0
x
= lim
x→0
0 −0
x
= 0
and similarly, (∂f/∂y)(0, 0) = 0. Thus the partial derivatives exist at (0, 0).
Also, if (x, y) ,= (0, 0), then
∂f
∂x
=
y
_
x
2
+y
2
−2x(xy)/2
_
x
2
+y
2
x
2
+y
2
=
y
_
x
2
+y
2
−
x
2
y
(x
2
+y
2
)
3/2
,
which does not have a limit as (x, y) → (0, 0). Diﬀerent limits are obtained
for diﬀerent paths of approach, as can be seen by letting x = My. Thus
the partial derivatives are not continuous at (0, 0), and so we cannot apply
Theorem 9.
12 2 Diﬀerentiation
We might now try to show that f is not diﬀerentiable (f is continuous,
however). If Df(0, 0) existed, then by Theorem 15 it would have to be the
zero matrix, since ∂f/∂x and ∂f/∂y are zero at (0, 0). Thus, by deﬁnition
of diﬀerentiability, for any ε > 0 there would be a δ > 0 such that 0 <
(h
1
, h
2
) < δ implies
[f(h
1
, h
2
) −f(0, 0)[
(h
1
, h
2
)
< ε
that is, [f(h
1
, h
2
)[ < ε(h
1
, h
2
), or [h
1
h
2
[ < ε(h
2
1
+ h
2
2
). But if we choose
h
1
= h
2
, this reads 1/2 < ε, which is untrue if we choose ε ≤ 1/2. Hence,
we conclude that f is not diﬀerentiable at (0, 0).
Exercises
2
1. Let f(x, y, z) = (e
x
, cos y, sin z). Compute Df. In general, when will
Df be a diagonal matrix?
2. (a) Let A : R
n
→R
m
be a linear transformation with matrix ¦A
ij
¦
so that Ax has components y
i
=
j
A
ij
x
j
. Let
M =
_
ij
A
2
ij
_
1/2
.
Use the CauchySchwarz inequality to prove that Ax ≤ Mx.
(b) Use the inequality derived in part (a) to show that a linear
transformation T : R
n
→R
m
with matrix [T
ij
] is continuous.
(c) Let A : R
n
→R
m
be a linear transformation. If
lim
x→0
Ax
x
= 0,
show that A = 0.
3. Let f : A → B and g : B → C be maps between open subsets of
Euclidean space, and let x
0
be in A or be a boundary point of A and
y
0
be in B or be a boundary point of B.
(a) If lim
x→0
f(x) = y
0
and lim
y→y0
g(y) = w, show that, in gen
eral, lim
x→x0
g(f(x)) need not equal w.
(b) If y
0
∈ B, and w = g(y
0
), show that lim
x→x0
g(f(x)) = w.
2
Answers, and hints for oddnumbered exercises are found at the end of this supple
ment, as well as selected complete solutions.
2.7 Diﬀerentiation Theorems 13
4. A function f : A ⊂ R
n
→R
m
is said to be uniformly continuous if
for every ε > 0 there is a δ > 0 such that for all points p and q ∈ A,
the condition p − q < δ implies f(p) − f(q) < ε. (Note that a
uniformly continuous function is continuous; describe explicitly the
extra property that a uniformly continuous function has.)
(a) Prove that a linear map T : R
n
→R
m
is uniformly continuous.
[HINT: Use Exercise 2.]
(b) Prove that x −→ 1/x
2
on (0, 1] is continuous, but not uniformly
continuous.
5. Let A = [A
ij
] be a symmetric n n matrix (that is, A
ij
= A
ji
) and
deﬁne f(x) = x Ax, so f : R
n
→R. Show that ∇f(x) is the vector
2Ax.
6. The following function is graphed in Figure 2.7.2:
f(x, y) =
_
_
_
2xy
2
_
x
2
+y
4
(x, y) ,= (0, 0)
0 (x, y) = (0, 0).
x
y
z
Figure 2.7.2. Graph of z = 2xy
2
/(x
2
+y
4
).
Show that ∂f/∂x and ∂f/∂y exist everywhere; in fact all directional
derivatives exist. But show that f is not continuous at (0, 0). Is f
diﬀerentiable?
7. Let f(x, y) = g(x) +h(y), and suppose g is diﬀerentiable at x
0
and h
is diﬀerentiable at y
0
. Prove from the deﬁnition that f is diﬀerentiable
at (x
0
, y
0
).
14 2 Diﬀerentiation
8. Use the CauchySchwarz inequality to prove the following: for any
vector v ∈ R
n
,
lim
x→x0
v x = v x
0
.
9. Prove that if lim
x→x0
f(x) = b for f : A ⊂ R
n
→R, then
lim
x→x0
[f(x)]
2
= b
2
and lim
x→x0
_
[f(x)[ =
_
[b[.
[You may wish to use Exercise 3(b).]
10. Show that in Theorem 9 with m = 1, it is enough to assume that
n − 1 partial derivatives are continuous and merely that the other
one exists. Does this agree with what you expect when n = 1?
11. Deﬁne f : R
2
→R by
f(x, y) =
_
xy
(x
2
+y
2
)
1/2
(x, y) ,= (0, 0)
0 (x, y) = (0, 0).
Show that f is continuous.
12. (a) Does lim
(x,y)→(0,0)
x
x
2
+y
2
exist?
(b) Does lim
(x,y)→(0,0)
x
3
x
2
+y
2
exist?
13. Find lim
(x,y)→(0,0)
xy
2
_
x
2
+y
2
.
14. Prove that s : R
2
→R, (x, y) −→ x +y is continuous.
15. Using the deﬁnition of continuity, prove that f is continuous at x if
and only if
lim
h→0
f(x +h) = f(x).
16. (a) A sequence x
n
of points in R
m
is said to converge to x, written
x
n
→ x as n → ∞, if for any ε > 0 there is an N such that
n ≥ N implies x−x
0
 < ε. Show that y is a boundary point of
an open set A if and only if y is not in A and there is a sequence
of distinct points of A converging to y.
(b) Let f : A ⊂ R
n
→ R
m
and y be in A or be a boundary point
of A. Prove that lim
x→y
f(x) = b if and only if f(x
n
) → b for
every sequence x
n
of points in A with x
n
→ y.
(c) If U ⊂ R
m
is open, show that f : U → R
p
is continuous if and
only if x
n
→ x ∈ U implies f(x
n
) → f(x).
2.7 Diﬀerentiation Theorems 15
17. If f(x) = g(x) for all x ,= A and if lim
x→A
f(x) = b, then show that
lim
x→A
g(x) = b, as well.
18. Let A ⊂ R
n
and let x
0
be a boundary point of A. Let f : A → R
and g : A → R be functions deﬁned on A such that lim
x→x0
f(x)
and lim
x→x0
g(x) exist, and assume that for all x in some deleted
neighborhood of x
0
, f(x) ≤ g(x). (A deleted neighborhood of x
0
is
any neighborhood of x
0
, less x
0
itself.)
(a) Prove that lim
x→x0
f(x) ≤ lim
x→x0
g(x). [HINT: Consider the
function φ(x) = g(x) −f(x); prove that lim
x→x0
φ(x) ≥ 0, and
then use the fact that the limit of the sum of two functions is
the sum of their limits.]
(b) If f(x) < g(x), do we necessarily have strict inequality of the
limits?
19. Given f : A ⊂ R
n
→ R
m
, we say that “f is o(x) as x → 0” if
lim
x→0
f(x)/x = 0.
(a) If f
1
and f
2
are o(x) as x → 0, prove that f
1
+ f
2
is also o(x)
as x → 0.
(b) Let g : A → R be a function with the property that there is a
number c > 0 such that [g(x) ≤ c for all x in A (the function g
is said to be bounded). If f is o(x) as x → 0, prove that gf is
also o(x) as x → 0 [where (gf)(x) = g(x)f(x)].
(c) Show that f(x) = x
2
is o(x) as x → 0. Is g(x) = x also o(x) as
x → 0?
16 2 Diﬀerentiation
Page 17
3
HigherOrder Derivatives and Extrema
Euler’s Analysis Inﬁnitorum (on analytic geometry) was fol
lowed in 1755 by the Institutiones Calculi Diﬀerentialis, to which
it was intended as an introduction. This is the ﬁrst textbook
on the diﬀerential calculus which has any claim to be regarded
as complete, and it may be said that until recently many mod
ern treatises on the subject are based on it; at the same time it
should be added that the exposition of the principles of the sub
ject is often prolix and obscure, and sometimes not altogether
accurate.
W. W. Rouse Ball
A Short Account of the History of Mathematics
Supplement 3.4A
Second Derivative Test: Constrained Extrema
In this supplement, we prove Theorem 10 in ¸3.4. We begin by recalling
the statement from the main text.
Theorem 10. Let f : U ⊂ R
2
→ R and g : U ⊂ R
2
→ R be smooth (at
least C
2
) functions. Let v
0
∈ U, g(v
0
) = c, and let S be the level curve for
g with value c. Assume that ∇g(v
0
) ,= 0 and that there is a real number λ
such that ∇f(v
0
) = λ∇g(v
0
). Form the auxiliary function h = f −λg and
18 3 Higher Derivatives and Extrema
the bordered Hessian determinant
[
¯
H[ =
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
0 −
∂g
∂x
−
∂g
∂y
−
∂g
∂x
∂
2
h
∂x
2
∂
2
h
∂x∂y
−
∂g
∂y
∂
2
h
∂x∂y
∂
2
h
∂y
2
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
evaluated at v
0
.
(i) If [
¯
H[ > 0, then v
0
is a local maximum point for f[S.
(ii) If [
¯
H[ < 0, then v
0
is a local minimum point for f[S.
(iii) If [
¯
H[ = 0, the test is inconclusive and v
0
may be a minimum, a
maximum, or neither.
The proof proceeds as follows. According to the remarks following the
Lagrange multiplier theorem, the constrained extrema of f are found by
looking at the critical points of the auxiliary function
h(x, y, λ) = f(x, y) −λ(g(x, y) −c).
Suppose (x
0
, y
0
, λ) is such a point and let v
0
= (x
0
, y
0
). That is,
∂f
∂x
¸
¸
¸
¸
v0
= λ
∂g
∂x
¸
¸
¸
¸
v0
,
∂f
∂y
¸
¸
¸
¸
v0
= λ
∂g
∂y
¸
¸
¸
¸
v0
, and g(x
0
, y
0
) = c.
In a sense this is a onevariable problem. If the function g is at all reason
able, then the set S deﬁned by g(x, y) = c is a curve and we are interested
in how f varies as we move along this curve. If we can solve the equation
g(x, y) = c for one variable in terms of the other, then we can make this
explicit and use the onevariable secondderivative test. If ∂g/∂y[
v0
,= 0,
then the curve S is not vertical at v
0
and it is reasonable that we can solve
for y as a function of x in a neighborhood of x
0
. We will, in fact, prove this
in ¸3.5 on the implicit function theorem. (If ∂g/∂x[
v0
,= 0, we can similarly
solve for x as a function of y.)
Suppose S is the graph of y = φ(x). Then f[S can be written as a
function of one variable, f(x, y) = f(x, φ(x)). The chain rule gives
df
dx
=
∂f
∂x
+
∂f
∂y
dφ
dx
and
d
2
f
dx
2
=
∂
2
f
∂x
2
+ 2
∂
2
f
∂x∂y
dφ
dx
+
∂
2
f
∂y
2
_
dφ
dx
_
2
+
∂f
∂y
d
2
φ
dx
2
.
_
¸
¸
¸
¸
_
¸
¸
¸
¸
_
(1)
3.4A Second Derivative Test 19
The relation g(x, φ(x)) = c can be used to ﬁnd dφ/dx and d
2
φ/dx
2
. Diﬀer
entiating both sides of g(x, φ(x)) = c with respect to x gives
∂g
∂x
+
∂g
∂y
dφ
dx
= 0
and
∂
2
g
∂x
2
+ 2
∂
2
g
∂x∂y
dφ
dx
+
∂
2
g
∂y
2
_
dφ
dx
_
2
+
∂g
∂y
d
2
φ
dx
2
= 0,
so that
dφ
dx
= −
∂g/∂x
∂g/∂y
and
d
2
φ
dx
2
= −
1
∂g/∂y
_
∂
2
g
∂x
2
−2
∂
2
g
∂x∂y
∂g/∂x
∂g∂y
+
∂
2
g
∂y
2
_
∂g/∂x
∂g/∂y
_
2
_
.
_
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
_
(2)
Substituting equation (7) into equation (6) gives
df
dx
=
∂f
∂x
−
∂f/∂y
∂g/∂y
∂g
∂x
and
d
2
f
dx
2
=
1
(∂g/∂y)
2
_
_
d
2
f
dx
2
−
∂f/∂y
∂g/∂y
∂
2
g
∂x
2
_ _
∂g
∂y
_
2
−2
_
∂
2
f
∂x∂y
−
∂f/∂y
∂g/∂y
∂
2
g
∂x∂y
_
∂g
∂x
∂g
∂y
+
_
∂
2
f
∂y
2
−
∂f/∂y
∂g/∂y
∂
2
g
∂y
2
_ _
∂g
∂x
_
2
_
.
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
(3)
At v
0
, we know that ∂f/∂y = λ∂g/∂y and ∂f/∂x = λ∂g/∂x, and so
equation (3) becomes
df
dx
¸
¸
¸
¸
x0
=
∂f
∂x
¸
¸
¸
¸
x0
−λ
∂g
∂x
¸
¸
¸
¸
x0
= 0
and
d
2
f
dx
2
¸
¸
¸
¸
x0
=
1
(∂g/∂y)
2
_
∂
2
h
∂x
2
_
∂g
∂y
_
2
−2
∂
2
h
∂x∂y
∂g
∂x
∂g
∂y
+
∂
2
h
∂y
2
_
∂g
∂x
_
2
_
= −
1
(∂g/∂y)
2
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
0 −
∂g
∂x
−
∂g
∂y
−
∂g
∂x
∂
2
h
∂x
2
∂
2
h
∂x∂y
−
∂g
∂y
∂
2
h
∂x∂y
∂
2
h
∂y
2
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
20 3 Higher Derivatives and Extrema
where the quantities are evaluated at x
0
and h is the auxiliary function
introduced above. This 33 determinant is, as in the statement of Theorem
10, called a bordered Hessian, and its sign is opposite that of d
2
f/dx
2
.
Therefore, if it is negative, we must be at a local minimum. If it is positive,
we are at a local maximum; and if it is zero, the test is inconclusive.
Exercises.
1. Take the special case of the theorem in which g(x, y) = y, so that
the level curve g(x, y) = c with c = 0 is the xaxis. Does Theorem 10
reduce to a theorem you know from onevariable Calculus?
2. Show that the bordered Hessian of f(x
1
, . . . , x
n
) subject to the single
constraint g(x
1
, . . . , x
n
) = c is the Hessian of the function
f(x
1
, . . . , x
n
) −λg(x
1
, . . . , x
n
)
of the n + 1 variables λ, x
1
, . . . , x
n
(evaluated at the critical point).
Can you use this observation to give another proof of the constrained
secondderivative test using the unconstrained one? HINT: If λ
0
de
notes the value of λ determined by the Lagrange multiplier theorem,
consider the function
F(x
1
, . . . , x
n
, λ) = f(x
1
, . . . , x
n
) −λg(x
1
, . . . , x
n
) ±(λ −λ
0
)
2
.
3.4B Implicit Function Theorem 21
Supplement 3.4B
Proof of the Implicit Function Theorem
We begin by recalling the statement.
Theorem 11. Special Implicit Function Theorem. Suppose that the
function F : R
n+1
→R has continuous partial derivatives. Denoting points
in R
n+1
by (x, z), where x ∈ R
n
and z ∈ R, assume that (x
0
, z
0
) satisﬁes
F(x
0
, z
0
) = 0 and
∂F
∂z
(x
0
, z
0
) ,= 0.
Then there is a ball U containing x
0
in R
n
and a neighborhood V of z
0
in
R such that there is a unique function z = g(x) deﬁned for x in U and z
in V that satisﬁes F(x, g(x)) = 0. Moreover, if x in U and z in V satisfy
F(x, z) = 0, then z = g(x). Finally, z = g(x) is continuously diﬀerentiable,
with the derivative given by
Dg(x) = −
1
∂F
∂z
(x, z)
D
x
F(x, z)
¸
¸
¸
¸
¸
¸
¸
z=g(x)
where D
x
F denotes the derivative matrix of F with respect to the variable
x, that is,
D
x
F =
_
∂F
∂x
1
, . . . ,
∂F
∂x
n
_
;
in other words,
∂g
∂x
i
= −
∂F/∂x
i
∂F/∂z
, i = 1, . . . , n. (1)
We shall prove the case n = 2, so that F : R
3
→R. The case for general
n is done in a similar manner. We write x = (x, y) and x
0
= (x
0
, y
0
).
Since (∂F/∂z)(x
0
, y
0
, z
0
) ,= 0, it is either positive or negative. Suppose for
deﬁniteness that it is positive. By continuity, we can ﬁnd numbers a > 0 and
b > 0 such that if x−x
0
 < a and [z−z
0
[ < a, then (∂F/∂z)(x, z) > b. We
can also assume that the other partial derivatives are bounded by a number
M in this region, that is, [(∂F/∂x)(x, z)[ ≤ M and [(∂F/∂y)(x, z)[ ≤ M,
which also follows from continuity. Write F(x, z) as follows
F(x, z) = F(x, z) −F(x
0
, z
0
)
= [F(x, z) −F(x
0
, z)] + [F(x
0
, z) −F(x
0
, z
0
)]. (2)
Consider the function
h(t) = F(tx + (1 −t)x
0
, z)
22 3 Higher Derivatives and Extrema
for ﬁxed x and z. By the mean value theorem, there is a number θ between
0 and 1 such that
h(1) −h(0) = h
(θ)(1 −0) = h
(θ),
that is, θ is such that
F(x, z) −F(x
0
, z) = [D
x
F(θ
x
+ (1 −θ)x
0
, z)](x −x
0
).
Substitution of this formula into equation (2) along with a similar formula
for the second term of that equation gives
F(x, z) = [D
x
F(θ
x
+ (1 −θ)x
0
, z)](x −x
0
)
+
_
∂F
∂z
(x
0
, φz + (1 −φ)z
0
)
_
(z −z
0
). (3)
where φ is between 0 and 1. Let a
0
satisfy 0 < a
0
< a and choose δ > 0
such that δ < a
0
and δ < ba
0
/2M. If [x − x
0
[ < δ then both [x − x
0
[ and
[y −y
0
[ are less than δ, so that the absolute value of each of the two terms
in
[D
x
F(θx + (1 −θ)x
0
, z)](x −x
0
)
=
_
∂F
∂x
(θ
x
+ (1 −θ)x
0
, z)
_
(x −x
0
) +
_
∂F
∂y
(θ
x
+ (1 −θ)x
0
, z)
_
(y −y
0
)
is less than Mδ < M(ba
0
/2M) = ba
0
/2. Thus, [x −x
0
[ < δ implies
[[D
x
F(θ
x
+ (1 −θ)x
0
, z)](x −x
0
)[ < ba
0
.
Therefore, from equation (3) and the choice of b, x−x
0
 < δ implies that
F(x, z
0
+a
0
) > 0 and F(x, z
0
−a
0
) < 0.
[The inequalities are reversed if (∂F/∂z)(x
0
, z
0
) < 0.] Thus, by the inter
mediate value theorem applied to F(x, z) as a function of z for each x,
there is a z between z
0
− a
0
and z
0
+ a
0
such that F(x, z) = 0. This z is
unique, since, by elementary calculus, a function with a positive derivative
is strictly increasing and thus can have no more than one zero.
Let U be the open ball of radius δ and center x
0
in R
2
and let V be the
open interval on R from z
0
− a
0
to z
0
+ a
0
. We have proved that if x is
conﬁned to U, there is a unique z in V such that F(x, z) = 0. This deﬁnes
the function z = g(x) = g(x, y) required by the theorem. We leave it to
the reader to prove from this construction that z = g(x, y) is a continuous
function.
It remains to establish the continuous diﬀerentiability of z = g(x). From
equation (3), and since F(x, z) = 0 and z
0
= g(x
0
), we have
g(x) −g(x
0
) = −
[D
x
F(θx + (1 −θ)x
0
, z)](x −x
0
)
∂F
∂z
(x
0
, φz + (1 −φ)z
0
)
.
3.4B Implicit Function Theorem 23
If we let x = (x
0
+h, y
0
) then this equation becomes
g(x
0
+h, y
0
) −g(x
0
, y
0
)
h
= −
∂F
∂x
(θ
x
+ (1 −θ)x
0
, z)
∂F
∂z
(x
0
, φz + (1 −φ)z
0
)
.
As h → 0, it follows that x → x
0
and z → z
0
, and so we get
∂g
∂x
(x
0
, y
0
) = lim
h→0
g(x
0
+h, y
0
) −g(x
0
, y
0
)
h
= −
∂F
∂x
(x
0
, z)
∂F
∂z
(x
0
, z)
.
The formula
∂g
∂y
(x
0
, y
0
) = −
∂F
∂y
(x
0
, z)
∂F
∂z
(x
0
, z)
is proved in the same way. This derivation holds at any point (x, y) in
U by the same argument, and so we have proved formula (1). Since the
righthand side of formula (1) is continuous, we have proved the theorem.
24 3 Higher Derivatives and Extrema
Page 25
4
Vector Valued Functions
I don’t know what I may seem to the world, but, as to myself,
I seem to have been only like a boy playing on the seashore,
and diverting myself in now and then ﬁnding a smoother pebble
or a prettier shell than ordinary, whilst the great ocean of truth
lay all undiscovered before me.
Isaac Newton, shortly before his death in 1727
Supplement 4.1A
Equilibria in Mechanics
Let F denote a force ﬁeld deﬁned on a certain domain U of R
3
. Thus,
F : U →R
3
is a given vector ﬁeld. Let us agree that a particle (with mass
m) is to move along a path c(t) in such a way that Newton’s law holds;
mass acceleration = force; that is, the path c(t) is to satisfy the equation
mc
(t) = F(c(t)). (1)
If F is a potential ﬁeld with potential V , that is, if F = −∇V , then
1
2
mc
(t)
2
+V (c(t)) = constant. (2)
26 4 Vector Valued Functions
(The ﬁrst term is called the kinetic energy.) Indeed, by diﬀerentiating
the left side of (2) using the chain rule
d
dt
_
1
2
mc
(t)
2
+V (c(t))
_
= mc
(t) c
(t) +∇V (c(t)) c
(t)
= [mc
(t) +∇V (c(t))] c
(t) = 0.
since mc
(t) = −∇V (c(t)). This proves formula (2).
Deﬁnition. A point x
0
∈ U is called a position of equilibrium if the
force at that point is zero: F : (x
0
) = 0. A point x
0
that is a position
of equilibrium is said to be stable if for every ρ > 0 and > 0, we can
choose numbers ρ
0
> 0 and
0
> 0 such that a point situated anywhere at
a distance less that ρ
0
from x
0
, after initially receiving kinetic energy in
an amount less than
0
, will forever remain a distance from x
0
less then ρ
and possess kinetic energy less then (see Figure 4.1.1).
X
0
X
ρ
0
ρ
Figure 4.1.1. Motion near a stable point x0.
Thus if we have a position of equilibrium, stability at x
0
means that a
slowly moving particle near x
0
will always remain near x
0
and keep moving
slowly. If we have an unstable equilibrium point x
0
, then c(t) = x
0
solves
the equation mc
(t) = F(c(t)), but nearby solutions may move away from
x
0
as time progresses. For example, a pencil balancing on its tip illustrates
an unstable conﬁguration, whereas a ball hanging on a spring illustrates a
stable equilibrium.
Theorem 1.
(i) Critical points of a potential are the positions of equilibrium.
(ii) In a potential ﬁeld, a point x
0
at which the potential takes a strict local
minimum is a position of stable equilibrium. (Recall that a function f
is said to have a strict local minimum at the point x
0
if there exists
4.1A Equilibria in Mechanics 27
a neighborhood U of x
0
such that f(x) > f(x
0
) for all x in U other
than x
0
.)
Proof. The ﬁrst assertion is quite obvious from the deﬁnition F = −∇V ;
equilibrium points x
0
are exactly critical points of V , at which ∇V (x
0
) = 0.
To prove assertion (ii), we shall make use of the law of conservation of
energy, that is, equation (2). We have
1
2
mc
(t)
2
+V (c(t)) =
1
2
mc
(t)
2
+V (c(0)).
We shall argue slightly informally to amplify and illuminate the central
ideas involved. Let us choose a small neighborhood of x
0
and start our par
ticle with a small kinetic energy. As t increases, the particle moves away
from x
0
on a path c(t) and V (c(t)) increases [since V (x
0
) = V (c(0)) is
a strict minimum], so that the kinetic energy must decrease. If the initial
kinetic energy is suﬃciently small, then in order for the particle to escape
from our neighborhood of x
0
, outside of which V has increased by a deﬁ
nite amount, the kinetic energy would have to become negative (which is
impossible). Thus the particle cannot escape the neighborhood.
Example 1. Find the points that are positions of equilibrium, and de
termine whether or not they are stable, if the force ﬁeld F = F
x
i+F
y
j+F
z
k
is given by F
x
= −k
2
x, F
y
= −k
2
y, F
z
= −k
2
z(k ,= 0).
1
Solution. The force ﬁeld F is a potential ﬁeld with potential given by
the function V =
1
2
k
2
(x
2
+ y
2
+ z
2
). The only critical point of V is at
the origin. The Hessian of V at the origin is
1
2
k
2
(h
2
1
+ h
2
2
+ h
2
3
), which is
positivedeﬁnite. It follows that the origin is a strict minimum of V . Thus,
by (i) and (ii) of Theorem 1, we have shown that the origin is a position of
stable equilibrium.
Let a point in a potential ﬁeld V be constrained to remain on the level
surface S given by the equation φ(x, y, z) = 0, with ∇φ ,= 0. If in formula
(1) we replace F by the component of F parallel to S, we ensure that the
particle will remain on S.
2
By analogy with Theorem 1, we have:
Theorem 2.
(i) If at a point P on the surface S the potential V [S has an extreme
value, then the point P is a position of equilibrium on the surface.
1
The force ﬁeld in this example is that governing the motion of a threedimensional
harmonic oscillator.
2
If φ(x, y, z) = x
2
+y
2
+z
2
−r
2
, the particle is constrained to move on a sphere; for
instance, it may be whirling on a string. The part subtracted from F to make it parallel
to S is normal to S and is called the centripetal force.
28 4 Vector Valued Functions
(ii) If a point P ∈ S is a strict local minimum of the potential V [S, then
the point P is a position of stable equilibrium.
The proof of this theorem will be omitted. It is similar to the proof of
Theorem 1., with the additional fact that the equation of motion uses only
the component of F along the surface.
3
Example 2. Let F be the gravitational ﬁeld near the surface of the earth;
that is, let F = (F
x
, F
y
, F
z
), where F
x
= 0, F
y
= 0, and F
z
= −mg, where
g is the acceleration due to gravity. What are the positions of equilibrium,
if a particle with mass m is constrained to the sphere φ(x, y, z) = x
2
+y
2
+
z
2
−r
2
= 0(r > 0)? Which of these are stable?
Solution. Notice that F is a potential ﬁeld with V = mgz. Using the
method of Lagrange multipliers introduced in ¸3.4 to locate the possible
extrema, we have the equations
∇V = λ∇φ
φ = 0
or, in terms of components,
0 = 2λx
0 = 2λy
mg = 2λz
x
2
+y
2
+z
2
−r
2
= 0.
The solution of these simultaneous equations is x = 0, y = 0, z = ±r, λ =
±mg/2r. By Theorem 2, it follows that the points P
1
(0, 0, −r) and P
2
=
(0, 0, r) are positions of equilibrium. By observation of the potential func
tion V = mgz and by Theorem 2, part (ii), it follows that P
1
is a strict
minimum and hence a stable point, whereas P
2
is not. This conclusion
should be physically obvious.
Exercises.
1. Let a particle move in a potential ﬁeld in R
2
given by V (x, y) =
3x
2
+ 2xy +y
2
+y + 4. Find the stable equilibrium points, if any.
3
These ideas can be applied to quite a number of interesting physical situations, such
as molecular vibrations. The stability of such systems is an important question. For
further information consult the physics literature (e.g. H. Goldstein, Classical Mechanics,
AddisonWesley, Reading, Mass., 1950, Chapter 10) and the mathematics literature (e.g.
M. Hirsch and S. Smale, Diﬀerential Equations, Dynamical Systems and Linear Algebra,
Academic Press, New York 1974).
4.1A Equilibria in Mechanics 29
2. Let a particle move in a potential ﬁeld in R
2
given by V (x, y) =
x
2
− 2xy + y
2
+ y
3
+ y
4
. Is the point (0, 0) a position of a stable
equilibrium?
3. (The solution is given at the end of this Internet Supplement). Let
a particle move in a potential ﬁeld in R
2
given by V (x, y) = x
2
−
4xy −y
2
−8x−6y. Find all the equilibrium points. Which, if any, are
stable?
4. Let a particle be constrained to move on the circle x
2
+ y
2
= 25
subject to the potential V = V
1
+ V
2
, where V
1
is the gravitational
potential in Example 2, and V
2
(x, y) = x
2
+ 24xy + 8y
2
. Find the
stable equilibrium points, if any.
5. Let a particle be constrained to move on the sphere x
2
+y
2
+z
2
= 1,
subject to the potential V = V
1
+ V
2
, where V
1
is the gravitational
potential in Example 2, and V
2
(x, y) = x + y. Find the stable equi
librium points, if any.
6. Attempt to formulate a deﬁnition and a theorem saying that if a
potential has a maximum at x
0
, then x
0
is a position of unstable
equilibrium. Watch out for pitfalls in your argument.
30 4 Vector Valued Functions
Supplement 4.1B
Rotations and the Sunshine Formula
In this supplement we use vector methods to derive formulas for the position
of the sun in the sky as a function of latitude and the day of the year.
4
To motivate this, have a look at the fascinating Figure 4.1.12 below which
gives a plot of the length of day as a function of one’s latitude and the day
of the year. Our goal in this supplement is to use vector methods to derive
this formula and to discuss related issues.
This is a nice application of vector calculus ideas because it does not in
volve any special technical knowledge to understand and is something that
everyone can appreciate. While it is not trivial, it also does not require any
particularly advanced ideas—it mainly requires patience and perseverance.
Even if one does not get all the way through the details, one can learn a
lot about rotations along the way.
A Bit About Rotations. Consider two unit vectors l and r in space
with the same base point. If we rotate r about the axis passing through l,
then the tip of r describes a circle (Figure 4.1.2). (Imagine l and r glued
rigidly at their base points and then spun about the axis through l.) Assume
that the rotation is at a uniform rate counterclockwise (when viewed from
the tip of l), making a complete revolution in T units of time. The vector
r now is a vector function of time, so we may write r = c(t). Our ﬁrst
aim is to ﬁnd a convenient formula for c(t) in terms of its starting position
r
0
= c(0).
l
r
Figure 4.1.2. If r rotates about l, its tip describes a circle.
Let λ denote the angle between l and r
0
; we can assume that λ ,= 0 and
λ ,= π, i .e., l and r
0
are not parallel, for otherwise r would not rotate. In
4
This material is adapted from Calculus I, II, III by J. Marsden and A. Weinstein,
SpringerVerlag, New York, which can be referred to for more information and some
interesting historical remarks. See also Dave Rusin’s home page http://www.math.niu.
edu/~rusin/ and in particular his notes on the position of the sun in the sky at http://
www.math.niu.edu/~rusin/usesmath/position.sun/ We thank him for his comments
on this section.
4.1B Rotations and the Sunshine Formula 31
fact, we shall take λ in the open interval (0, π). Construct the unit vector
m
0
as shown in Figure 4.1.3. From this ﬁgure we see that
r
0
= (cos λ)l + (sin λ)m
0
. (1)
l
r
0
m
0
λ
π/2 − λ
Figure 4.1.3. The vector m
0
is in the plane of r
0
and l, is orthogonal to l, and makes
an angle of (π/2) −λ with r
0
.
In fact, formula (1) can be taken as the algebraic deﬁnition of m
0
by
writing m
0
= (1/ sin λ)r
0
− (cos λ/ sin λ)l. We assumed that λ ,= 0, and
λ ,= π, so sin λ ,= 0.
Now add to this ﬁgure the unit vector n
0
= l m
0
. (See Figure 4.1.4.)
The triple (l, m
0
, n
0
) consists of three mutually orthogonal unit vectors,
just like (i, j, k).
l
r
0
m
0
π/2 – λ
λ
n
0
Figure 4.1.4. The triple (l, m
0
, n
0
) is a righthanded orthogonal set of unit vectors.
Example 1. Let l = (1/
√
3)(i +j +k) and r
0
= k. Find m
0
and n
0
.
Solution. The angle between l and r
0
is given by cos λ = l r
0
= 1/
√
3.
This was determined by dotting both sides of formula (1) by l and using
the fact that l is a unit vector. Thus, sin λ =
√
1 −cos
2
λ =
_
2/3, and so
from formula (1) we get
m
0
=
1
sin λ
c(0) −
cos λ
sin λ
l
=
_
3
2
k −
1
√
3
_
3
2
1
√
3
(i +j +k) =
2
√
6
k −
1
√
6
(i +j)
32 4 Vector Valued Functions
and
n
0
= l m
0
=
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
i j k
1
√
3
1
√
3
1
√
3
−
1
√
6
−
1
√
2
−
2
√
6
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
=
1
√
2
i −
2
√
2
j.
Return to Figure 4.1.4 and rotate the whole picture about the axis l. Now
the “rotated” vectors m and n will vary with time as well. Since the angle
λ remains constant, formula (1) applied after time t to r and l gives (see
Figure 4.1.5)
m =
1
sin λ
r −
cos λ
sin λ
l (2)
l
v
m
n
Figure 4.1.5. The three vectors v, m, and n all rotate about l.
On the other hand, since m is perpendicular to l, it rotates in a circle in
the plane of m
0
and n
0
. It goes through an angle 2π in time T, so it goes
through an angle 2πt/T in t units of time, and so
m = cos
_
2πt
T
_
m
0
+ sin
_
2πt
T
_
n
0
.
Inserting this in formula (2) and rearranging gives
r(t) = (cos λ)l + sin λcos
_
2πt
T
_
m
0
+ sin λsin
_
2πt
T
_
n
0
. (3)
This formula expresses explicitly how r changes in time as it is rotated
about l, in terms of the basic vector triple (l, m
0
, n
0
).
Example 2. Express the function c(t) explicitly in terms of l, r
0
, and T.
Solution. We have cos λ = l r
0
and sin λ = l r
0
. Furthermore n
0
is
a unit vector perpendicular to both l and r
0
, so we must have (up to an
4.1B Rotations and the Sunshine Formula 33
orientation)
n
0
=
l r
0
l r
0

.
Thus (sin λ)n
0
= l r
0
. Finally, from formula (1), we obtain (sin λ)m
0
=
r
0
−(cos λ)l = r
0
−(r
0
l)l. Substituting all this into formula (3),
r = (r
0
l)l + cos
_
2πt
T
_
[r
0
−(r
0
l)l] + sin
_
2πt
T
_
(l r
0
).
Example 3. Show by a direct geometric argument that the speed of the
tip of r is (2π/T) sin λ. Verify that equation (3) gives the same formula.
Solution. The tip of r sweeps out a circle of radius sin λ, so it covers
a distance 2π sin λ in time T. Its speed is therefore (2π sin λ)/T (Figure
4.1.6).
l
r
sin λ
λ
Figure 4.1.6. The tip of r sweeps out a circle of radius sin λ.
From formula (3), we ﬁnd the velocity vector to be
dr
dt
= −sin λ
2π
T
sin
_
2πt
T
_
m
0
+ sin λ
2π
T
cos
_
2πt
T
_
n
0
,
and its length is (since m
0
and n
0
are unit orthogonal vectors)
_
_
_
_
dr
dt
_
_
_
_
=
¸
sin
2
λ
_
2π
T
_
2
sin
2
_
2πt
T
_
+ sin
2
λ
_
2π
T
_
2
cos
2
_
2πt
T
_
= sin λ
_
2π
T
_
,
as above.
34 4 Vector Valued Functions
Rotation and Revolution of the Earth. Now we apply our study
of rotations to the motion of the earth about the sun, incorporating the
rotation of the earth about its own axis as well. We will use a simpliﬁed
model of the earthsun system, in which the sun is ﬁxed at the origin of our
coordinate system and the earth moves at uniform speed around a circle
centered at the sun. Let u be a unit vector pointing from the sun to the
center of the earth; we have
u = cos(2πt/T
y
)i + sin(2πt/T
y
)j
where T
y
is the length of a year (t and T
y
measured in the same units).
See Figure 4.1.7. Notice that the unit vector pointing from the earth to the
sun is −u and that we have oriented our axes so that u = i when t = 0.
–u
u
i
j
k
Figure 4.1.7. The unit vector u points from the sun to the earth at time t.
Next we wish to take into account the rotation of the earth. The earth
rotates about an axis which we represent by a unit vector l pointing from
the center of the earth to the North Pole. We will assume that l is ﬁxed
5
with respect to i, j, and k; astronomical measurements show that the incli
nation of l (the angle between l and k) is presently about 23.5
◦
. We will
denote this angle by α. If we measure time so that the ﬁrst day of summer
in the northern hemisphere occurs when t = 0, then the axis l tilts in the
direction −i, and so we must have l = cos αk −sin αi. (See Figure 4.1.8)
Now let r be the unit vector at time t from the center of the earth to a
ﬁxed point P on the earth’s surface. Notice that if r is located with its base
5
Actually, the axis l is known to rotate about k once every 21, 000 years. This phe
nomenon called precession or wobble, is due to the irregular shape of the earth and may
play a role in longterm climatic changes, such as ice ages. See pages. 130134 of The
Weather Machine by Nigel Calder, Viking (1974).
4.1B Rotations and the Sunshine Formula 35
α
i
k
k
N
l
Figure 4.1.8. At t = 0, the earth’s axis tilts toward the sun by the angle α.
point at P, then it represents the local vertical direction. We will assume
that P is chosen so that at t = 0, it is noon at the point P; then r lies in
the plane of l and i and makes an angle of less than 90
◦
with −i. Referring
to Figure 4.1.9, we introduce the unit vector m
0
= −(sin α)k − (cos α)i
orthogonal to l. We then have r
0
= (cos λ)l + (sin λ)m
0
, where λ is the
angle between l and r
0
. Since λ = π/2 − , where is the latitude of the
point P, we obtain the expression r
0
= (sin )l + (cos )m
0
. As in Figure
4.1.4, let n
0
= l m
0
.
Example 4. Prove that n
0
= l m
0
= −j.
Solution. Geometrically, l m
0
is a unit vector orthogonal to l and m
0
pointing in the sense given by the righthand rule. But l and m
0
are both
in the i − k plane, so l m
0
points orthogonal to it in the direction −j.
(See Figure 4.1.9).
Algebraically, l = (cos α)k −(sin α)i and m
0
= −(sin α)k −(cos α)i, so
l m
0
=
_
_
i j k
−sin α 0 cos α
−cos α 0 −sin α
_
_
= −j(sin
2
α + cos
2
α) = −j.
Now we apply formula (3) to get
r = (cos λ)l + sin λcos
_
2πt
T
d
_
m
0
+ sin λsin
_
2πt
T
d
_
n
0
,
36 4 Vector Valued Functions
i
j
k
k
l
m
r
l
N
S
λ
Figure 4.1.9. The vector r is the vector from the center of the earth to a ﬁxed location
P. The latitude of P is l and the colatitude is λ = 90
◦
− . The vector m
0
is a unit
vector in the plane of the equator (orthogonal to l) and in the plane of l and r
0
.
where T
d
is the length of time it takes for the earth to rotate once about
its axis (with respect to the “ﬁxed stars”—i.e., our i, j, k vectors).
6
Sub
stituting the expressions derived above for λ, l, m
0
, and n
0
, we get
r = sin (cos αk −sin αi)
+cos cos
_
2πt
T
d
_
(−sin αk −cos αi) −cos sin
_
2πt
T
d
_
j.
Hence
r =−
_
sin sin α + cos cos αcos
_
2πt
T
d
__
i −cos sin
_
2πt
T
d
_
j
+
_
sin cos α −cos sin αcos
_
2πt
T
d
__
k. (4)
Example 5. What is the speed (in kilometers per hour) of a point on
the equator due to the rotation of the earth? A point at latitude 60
◦
? (The
radius of the earth is 6371 kilometers.)
Solution. The speed is
s = (2πR/T
d
) sin λ = (2πR/T
d
) cos ,
where R is the radius of the earth and is the latitude. (The factor R is
inserted since r is a unit vector, the actual vector from the earth’s center
to a point P on its surface is Rr).
6
T
d
is called the length of the sidereal day. It diﬀers from the ordinary, or solar,
day by about 1 part in 365 because of the rotation of the earth about the sun. In fact,
T
d
≈ 23.93 hours.
4.1B Rotations and the Sunshine Formula 37
Using T
d
= 23.93 hours and R = 6371 kilometers, we get s = 1673 cos
kilometers per hour. At the equator = 0, so the speed is 1673 kilometers
per hour; at = 60
◦
, s = 836.4 kilometers per hour.
With formula (3) at our disposal, we are ready to derive the sunshine
formula. The intensity of light on a portion of the earth’s surface (or at
the top of the atmosphere) is proportional to sinA, where A is the angle
of elevation of the sun above the horizon (see Fig. 4.1.10). (At night sin A
is negative, and the intensity then is of course zero.)
A
2
1
sunlight
Figure 4.1.10. The intensity of sunlight is proportional to sin A. The ratio of area 1
to area 2 is sin A.
Thus we want to compute sin A. From Figure 4.1.11 we see that sin A =
−u r. Substituting u = cos(2πt/T
y
)i + sin(2πt/T
y
)j and formula (3) into
this formula for sin A and taking the dot product gives
sin A =cos
_
2πt
T
y
__
sin sin α + cos cos αcos
_
2πt
T
d
__
+ sin
_
2πt
T
y
__
cos sin
_
2πt
T
d
__
=cos
_
2πt
T
y
_
sin sin α + cos
_
cos
_
2πt
T
y
_
cos αcos
_
2πt
T
d
_
+ sin
_
2πt
T
y
_
sin
_
2πt
T
d
__
. (5)
Example 6. Set t = 0 in formula (4). For what is sin A = 0? Interpret
your result.
Solution. With t = 0 we get
sin A = sin sin α + cos cos α = cos( −α).
This is zero when − α = ±π/2. Now sin A = 0 corresponds to the sun
on the horizon (sunrise or sunset), when A = 0 or π. Thus, at t = 0, this
38 4 Vector Valued Functions
P
r
–u
90°– A
sunlight
l
Figure 4.1.11. The geometry for the formula sin A = cos(90
◦
−A) = −u · r.
occurs when = α ± (π/2). The case α + (π/2) is impossible, since lies
between −π/2 and π/2. The case = α −(π/2) corresponds to a point on
the Antarctic Circle; indeed at t = 0 (corresponding to noon on the ﬁrst
day of northern summer) the sun is just on the horizon at the Antarctic
Circle.
Our next goal is to describe the variation of sinA with time on a par
ticular day. For this purpose, the time variable t is not very convenient;
it will be better to measure time from noon on the day in question. To
simplify our calculations, we will assume that the expressions cos(2πt/T
y
)
and sin(2πt/T
y
) are constant over the course of any particular day; since T
y
is approximately 365 times as large as the change in t, this is a reasonable
approximation. On the nth day (measured from June 21), we may replace
2πt/T
y
by 2πn/365, and formula (4.1.4) gives
sin A = (sin )P + (cos )
_
Qcos
_
2πt
T
d
_
+Rsin
_
2πt
T
d
__
,
where
P = cos(2πn/365) sin α, Q = cos(2πn/365) cos α, and R = sin(2πn/365).
We will write the expression Qcos(2πt/T
d
) + Rsin(2πt/T
d
) in the form
U cos[2π(t − t
n
)/T
d
], where t
n
is the time of noon of the nth day. To ﬁnd
U, we use the addition formula to expand the cosine:
U cos
__
2πt
T
d
_
−
_
2πt
n
T
d
__
= U
_
cos
_
2πt
T
d
_
cos
_
2πt
n
T
d
_
+ sin
_
2πt
T
d
_
sin
_
2πt
n
T
d
__
.
Setting this equal to Qcos(2πt/T
d
) + Rsin(2πt/T
d
) and comparing coeﬃ
cients of cos 2πt/T
d
and sin 2πt/T
d
gives
U cos
2πt
n
T
d
= Q and U sin
2πt
n
T
d
= R.
4.1B Rotations and the Sunshine Formula 39
Squaring the two equations and adding gives
7
U
2
= Q
2
+R
2
or U =
_
Q
2
+R
2
,
while dividing the second equation by the ﬁrst gives tan(2πt
n
/T
d
) = R/Q.
We are interested mainly in the formula for U; substituting for Q and R
gives
U =
¸
cos
2
_
2πn
365
_
cos
2
α + sin
2
2πn
365
=
¸
cos
2
_
2πn
365
_
(1 −sin
2
α) + sin
2
2πn
365
=
¸
1 −cos
2
_
2πn
365
_
sin
2
α.
Letting τ be the time in hours from noon on the nth day so that τ/24 =
(t −t
n
)/T
d
, we substitute into formula (5) to obtain the ﬁnal formula:
sin A =sin cos
_
2πn
365
_
sin α
+ cos
¸
1 −cos
2
_
2πn
365
_
sin
2
αcos
2πτ
24
. (6)
Example 7. How high is the sun in the sky in Edinburgh (latitude 56
◦
)
at 2 p.m. on Feb. 1?
Solution. We plug into formula (6): α = 23.5,
◦
= 90
◦
− 56
◦
= 34
◦
, n
= number of days after June 21 = 225, and τ = 2 hours. We get sin A =
0.5196, so A = 31.3.
◦
Formula (4) also tells us how long days are.
8
At the time S of sunset,
A = 0. That is,
cos
_
2πS
24
_
= −tan
sin αcos(2πT/365)
_
1 −sin
2
αcos
2
(2πT/365)
. (4.1.1)
7
We take the positive square root because sin A should have a local maximum when
t = tn.
8
If π/2 − α <  < π/2 (inside the polar circles), there will be some values of t for
which the righthand side of formula (1) does not lie in the interval [−1, 1]. On the days
corresponding to these values of t, the sun will never set (“midnight sun”). If = ±π/2,
then tan = ∞, and the righthand side does not make sense at all. This reﬂects the fact
that, at the poles, it is either light all day or dark all day, depending upon the season.
40 4 Vector Valued Functions
Solving for S, and remembering that S ≥ 0 since sunset occurs after noon,
we get
S =
12
π
cos
−1
_
_
−tan
sin αcos(2πT/365)
_
1 −sin
2
αcos
2
(2πT/365)
_
_
. (4.1.2)
The graph of S is shown in Figure 4.1.12.
0
100
200
300
0
6
12
18
24
–90
90
0
365
Length
of Day
Day of Year
L
a
t
i
t
u
d
e
i
n
D
e
g
r
e
e
s
Figure 4.1.12. Day length as a function of latitude and day of the year.
Exercises.
1. Let l = (j +k)/
√
2 and r
0
= (i −j)/
√
2.
(a) Find m
0
and n
0
.
(b) Find r = c(t) if T = 24.
(c) Find the equation of the line tangent to c(t) at t = 12 and T = 24.
2. From formula (3), verify that c(T/2) n = 0. Also, show this geomet
rically. For what values of t is c(t) n = 0?
3. If the earth rotated in the opposite direction about the sun, would
T
d
be longer or shorter than 24 hours? (Assume the solar day is ﬁxed
at 24 hours.)
4. Show by a direct geometric construction that
r = c(T
d
/4) = −(sin sin α)i −(cos )j + (sin cos α)k.
Does this formula agree with formula (3)?
4.1B Rotations and the Sunshine Formula 41
5. Derive an “exact” formula for the time of sunset from formula (4).
6. Why does formula (6) for sin A not depend on the radius of the earth?
The distance of the earth from the sun?
7. How high is the sun in the sky in Paris at 3 p.m. on January 15?
(The latitude of Paris is 49
◦
N.)
8. How much solar energy (relative to a summer day at the equator)
does Paris receive on January 15? (The latitude of Paris is 49
◦
N).
9. How would your answer in Exercise 8 change if the earth were to roll
to a tilt of 32
◦
instead of 23.5
◦
?
42 4 Vector Valued Functions
Supplement 4.1C
The Principle of Least Action
By Richard Feynman
9
When I was in high school, my physics teacher—whose name was Mr.
Bader—called me down one day after physics class and said, “You look
bored; I want to tell you something interesting.” Then he told me something
which I found absolutely fascinating, and have, since then, always found
fascinating. Every time the subject comes up, I work on it. In fact, when
I began to prepare this lecture I found myself making more analyses on
the thing. Instead of worrying about the lecture, I got involved in a new
problem. The subject is this—the principle of least action.
Mr. Bader told me the following: Suppose you have a particle (in a grav
itational ﬁeld, for instance) which starts somewhere and moves to some
other point by free motion—you throw it, and it goes up and comes down
(Figure 4.1.13).
here
there
t
1
t
2
actual motion
Figure 4.1.13.
It goes from the original place to the ﬁnal place in a certain amount of
time. Now, you try a diﬀerent motion. Suppose that to get from here to
there, it went like this (Figure 4.1.14)
but got there in just the same amount of time. Then he said this: If you
calculate the kinetic energy at every moment on the path, take away the
potential energy, and integrate it over the time during the whole path,
you’ll ﬁnd that the number you’ll get is bigger then that for the actual
motion.
In other words, the laws of Newton could be stated not in the form F =
ma but in the form: the average kinetic energy less the average potential
energy is as little as possible for the path of an object going from one point
to another.
9
Lecture 19 from The Feynman Lectures on Physics.
4.1C Principle of Least Action 43
here
there
t
1
t
2
imagined
motion
Figure 4.1.14.
Let me illustrate a little bit better what it means. If you take the case of
the gravitational ﬁeld, then if the particle has the path x(t) (let’s just take
one dimension for a moment; we take a trajectory that goes up and down
and not sideways), where x is the height above the ground, the kinetic
energy is
1
2
m(dx/dt)
2
, and the potential energy at any time is mgx. Now I
take the kinetic energy minus the potential energy at every moment along
the path and integrate that with respect to time from the initial time to
the ﬁnal time. Let’s suppose that at the original time t
1
we started at some
height and at the end of the time t
2
we are deﬁnitely ending at some other
place (Figure 4.1.15).
t
1
t
2
t
x
Figure 4.1.15.
Then the integral is
_
t2
t1
_
1
2
m
_
dx
dt
_
2
−mgx
_
dt.
The actual motion is some kind of a curve—it’s a parabola if we plot against
the time—and gives a certain value for the integral. But we could imagine
44 4 Vector Valued Functions
some other motion that went very high and came up and down in some
peculiar way (Figure 4.1.16).
t
1
t
2
t
x
Figure 4.1.16.
We can calculate the kinetic energy minus the potential energy and inte
grate for such a path. . . or for any other path we want. The miracle is that
the true path is the one for which that integral is least.
Let’s try it out. First, suppose we take the case of a free particle for
which there is no potential energy at all. Then the rule says that in going
from one point to another in a given amount of time, the kinetic energy
integral is least, so it must go at a uniform speed. (We know that’s the right
answer—to go at a uniform speed.) Why is that? Because if the particle
were to go any other way, the velocities would be sometimes higher and
sometimes lower then the average. The average velocity is the same for
every case because it has to get from ‘here’ to ‘there’ in a given amount of
time.
As and example, say your job is to start from home and get to school in
a given length of time with the car. You can do it several ways: You can
accelerate like mad at the beginning and slow down with the breaks near
the end, or you can go at a uniform speed, or you can go backwards for a
while and then go forward, and so on. The thing is that the average speed
has got to be, of course, the total distance that you have gone over the
time. But if you do anything but go at a uniform speed, then sometimes
you are going too fast and sometimes you are going too slow. Now the
mean square of something that deviates around an average, as you know,
is always greater than the square of the mean; so the kinetic energy integral
would always be higher if you wobbled your velocity than if you went at a
uniform velocity. So we see that the integral is a minimum if the velocity is
a constant (when there are no forces). The correct path is like this (Figure
4.1.17).
Now, an object thrown up in a gravitational ﬁeld does rise faster ﬁrst
and then slow down. That is because there is also the potential energy, and
we must have the least diﬀerence of kinetic and potential energy on the
4.1C Principle of Least Action 45
t
1
t
2
t
x
here
there
no forces
Figure 4.1.17.
average. Because the potential energy rises as we go up in space, we will
get a lower diﬀerence if we can get as soon as possible up to where there
is a high potential energy. Then we can take the potential away from the
kinetic energy and get a lower average. So it is better to take a path which
goes up and gets a lot of negative stuﬀ from the potential energy (Figure
4.1.18).
t
1
t
2
t
x
here
there
more + KE
more – PE
Figure 4.1.18.
On the other hand, you can’t go up too fast, or too far, because you will
then have too much kinetic energy involved—you have to go very fast to
get way up and come down again in the ﬁxed amount of time available. So
you don’t want to go too far up, but you want to go up some. So i turns
out that the solution is some kind of balance between trying to get more
potential energy with the least amount of extra kinetic energy—trying to
get the diﬀerence, kinetic minus the potential, as small as possible.
46 4 Vector Valued Functions
That is all my teacher told me, because he was a very good teacher and
knew when to stop talking. But I don’t know when to stop talking. So
instead of leaving it as an interesting remark, I am going to horrify and
disgust you with the complexities of life by proving that it is so. The kind
of mathematical problem we will have is a very diﬃcult and a new kind.
We have a certain quantity which is called the action, S. It is the kinetic
energy, minus the potential energy, integrated over time.
Action = S =
_
t2
t1
(KE −PE)dt.
Remember that the PE and KE are both functions of time. For each dif
ferent possible path you get a diﬀerent number for this action. Our math
ematical problem is to ﬁnd out for what curve that number is the least.
You say—Oh, that’s just the ordinary calculus of maxima and minima.
You calculate the action and just diﬀerentiate to ﬁnd the minimum.
But watch out. Ordinarily we just have a function of some variable, and
we have to ﬁnd the value of what variable where the function is least or
most. For instance, we have a rod which has been heated in the middle and
the heat is spread around. For each point on the rod we have a temperature,
and we must ﬁnd the point at which that temperature is largest. But now
for each path in space we have a number—quite a diﬀerent thing—and
we have to ﬁnd the path in space for which the number is the minimum.
That is a completely diﬀerent branch of mathematics. It is not the ordinary
calculus. In fact, it is called the calculus of variations.
There are many problems in this kind of mathematics. For example, the
circle is usually deﬁned as the locus of all points at a constant distance
from a ﬁxed point, but another way of deﬁning a circle is this: a circle
is that curve of given length which encloses the biggest area. Any other
curve encloses less area for a given perimeter than the circle does. So if
we give the problem: ﬁnd that curve which encloses the greatest area for a
given perimeter, we would have a problem of the calculus of variations—a
diﬀerent kind of calculus than you’re used to.
So we make the calculation for the path of the object. Here is the way
we are going to do it. The idea is that we imagine that there is a true path
and that any other curve we draw is a false path, so that if we calculate
the action for the false path we will get a value that is bigger that if we
calculate the action for the true path (Figure 4.1.19).
Problem: Find the true path. Where is it? One way, of course, is to
calculate the action for millions and millions of paths and look at which
one is lowest. When you ﬁnd the lowest one, that’s the true path.
That’s a possible way. But we can do it better than that. When we have
a quantity which has a minimum—for instance, in and ordinary function
like the temperature—one of the properties of the minimum is that if we
go away from the minimum in the ﬁrst order, the deviation of the function
from its minimum value is only second order. At any place else on the
4.1C Principle of Least Action 47
t
x
S
2
> S
1
S
1
S
2
fake path
true path
Figure 4.1.19.
curve, if we move a small distance the value of the function changes also in
the ﬁrst order. But at a minimum, a tiny motion away makes, in the ﬁrst
approximation, no diﬀerence (Figure 4.1.20).
t
x
minimum
T ( x)
2
T x
Figure 4.1.20.
That is what we are going to use to calculate the true path. If we have
the true path, a curve which diﬀers only a little bit from it will, in the ﬁrst
approximation, make no diﬀerence in the action. Any diﬀerence will be in
the second approximation, if we really have a minimum.
That is easy to prove. If there is a change in the ﬁrst order when I
deviate the curve a certain way, there is a change in the action that is
proportional to the deviation. The change presumably makes the action
greater; otherwise we haven’t got a minimum. But then if the change is
proportional to the deviation, reversing the sign of the deviation will make
the action less. We would get the action to increase one way and to decrease
the other way. The only way that it could really be a minimum is that in
the ﬁrst approximation it doesn’t make any change, that the changes are
48 4 Vector Valued Functions
proportional to the square of the deviations from the true path.
So we work it this way: We call x(t) (with an underline) the true path—
the one we are trying to ﬁnd. We take some trial path x(t) that diﬀers
from the true path by a small amount which we will call η(t) (eta of t).
(See Figure 4.1.21).
t
x
x(t)
η(t)
x(t)
Figure 4.1.21.
Now the idea is that if we calculate the action S for the path x(t), then
the diﬀerence between that S and the action that we calculated for the
path x(t)—to simplify the writing we can call it S—the diﬀerence of S and
S must be zero in the ﬁrstorder approximation of small η. It can diﬀer in
the second order, but in the ﬁrst order the diﬀerence must be zero.
And that must be true for any η at all. Well, not quite. The method
doesn’t mean anything unless you consider paths which all begin and end
at the same two points—each path begins at a certain point at t
1
and ends
at a certain other point at t
2
, and those points and times are kept ﬁxed. So
the deviations in our η have to be zero at each end, η(t
1
) = 0 and η(t
2
) = 0.
With that condition, we have speciﬁed our mathematical problem.
If you didn’t know any calculus, you might do the same kind of thing
to ﬁnd the minimum of an ordinary function f(x). You could discuss what
happens if you take f(x) and add a small amount h to x and argue that the
correction to f(x) in the ﬁrst order in h must be zero ant the minimum.
You would substitute x + h for x and expand out to the ﬁrst order in h
. . . just as we are going to do with η.
The idea is then that we substitute x(t) = x(t) +η(t) in the formula for
the action:
S =
_ _
m
2
_
dx
dt
_
2
−V (x)
_
dt,
where I call the potential energy V (t). The derivative dx/dt is, of course,
the derivative of x(t) plus the derivative of η(t), so for the action I get this
4.1C Principle of Least Action 49
expression:
S =
_
t2
t1
_
m
2
_
dx
dt
+
dη
dt
_
2
−V (x +η)
_
dt.
Now I must write this out in more detail. For the squared term I get
_
dx
dt
_
2
+ 2
dx
dt
dη
dt
+
_
dη
dt
_
2
But wait. I’m worrying about higher than the ﬁrst order, so I will take all
the terms which involve η
2
and higher powers and put them in a little box
called ‘second and higher order.’ From this term I get only second order,
but there will be more from something else. So the kinetic energy part is
m
2
_
dx
dt
_
2
+m
dx
dt
dη
dt
+ (second and higher order).
Now we need the potential V at x +η. I consider η small, so I can write
V (x) as a Taylor series. It is approximately V (x); in the next approximation
(from the ordinary nature of derivatives) the correction is η times the rate
of change of V with respect to x, and so on:
V (x +η) = V (x) +ηV
(x) +
η
2
2
V
(x) +. . .
I have written V
for the derivative of V with respect to x in order to
save writing. The term in η
2
and the ones beyond fall into the ‘second and
higher order’ category and we don’t have to worry about them. Putting it
all together,
S =
_
t2
t1
_
m
2
_
dx
dt
_
2
−V (x) +m
dx
dt
dη
dt
−ηV
(x) + (second and higher order)
_
dt.
Now if we look carefully at the thing, we see that the ﬁrst two terms
which I have arranged here correspond to the action S that I would have
calculated with the true path x. The thing I want to concentrate on is the
change in S—the diﬀerence between the S and the S that we would get for
the right path. This diﬀerence we will write as δS, called the variation in
S. Leaving out the ‘second and higher order’ terms, I have for δS
δS =
_
t2
t1
_
m
dx
dt
dη
dt
−ηV
(x)
_
dt.
Now the problem is this: Here is a certain integral. I don’t know what
the x is yet, but I do know that no matter what η is, this integral must
50 4 Vector Valued Functions
be zero. Well, you think, the only way that that can happen is that what
multiplies η must be zero. But what about the ﬁrst term with dη/dt? Well,
after all, if η can be anything at all, its derivative is anything also, so you
conclude that the coeﬃcient of dη/dt must also be zero. That isn’t quite
right. It isn’t quite right because there is a connection between η and its
derivative; they are not absolutely independent, because η(t) must be zero
at both t
1
and t
2
.
The method of solving all problems in the calculus of variations always
uses the same general principle. You make the shift in the thing you want
to vary (as we did by adding η); you look at the ﬁrstorder terms; then you
always arrange things in such a form that you get an integral of the form
‘some kind of stuﬀ times the shift (η),’ but with no other derivatives (no
dη/dt). It must be rearranged so it is always ‘something’ times η. You will
see the great value of that in a minute. (There are formulas that tell you
how to do this in some cases without actually calculating, but they are not
general enough to be worth bothering about; the best way is to calculate
it out this way.)
How can I rearrange the term in dη/dt to make it have an η? I can
do that by integrating by parts. It turns out that the whole trick of the
calculus of variations consists of writing down the variation of S and then
integrating by parts so that the derivatives of η disappear. It is always the
same in every problem in which derivatives appear.
You remember the general principle for integrating by parts. If you have
any function f times dη/dt integrated with respect to t, you write down
the derivative of ηf:
d
dt
(ηf) = η
df
dt
+f
dη
dt
.
The integral you want is over the last term, so
_
f
dη
dt
dt = ηf −
_
η
df
dt
dt.
In our formula for δS, the function f is m times dx/dt; therefore, I have
the following formula for δS.
δS = m
dx
dt
η(t)
¸
¸
¸
¸
t2
t1
−
_
t2
t1
d
dt
_
m
dx
dt
_
η(t)dt −
_
t2
t1
V
(x)η(t)dt.
The ﬁrst term must be evaluated at the two limits t
1
and t
2
. Then I must
have the integral from the rest of the integration by parts. The last term
is brought down without change.
Now comes something which always happens—the integrated part dis
appears. (In fact, if the integrand part does not disappear, you restate the
principle, adding conditions to make sure it does!) We have already said
that η must be zero at both ends of the path, because the principle is that
the action is a minimum provided that the varied curve begins and ends
4.1C Principle of Least Action 51
at the chosen points. The condition is that η(t
1
) = 0 and η(t
2
) = 0. So
the integrated term is zero. We collect the other terms together and obtain
this:
δS =
_
t2
t1
_
−m
d
2
x
dt
2
−V
(x)
_
η(t)dt.
The variation in S is now the way we we wanted it—there is the stuﬀ in
brackets, say F, all multiplied by η(t) and integrated from t
1
to t
2
.
We have that an integral of something or other times η(t) is always zero:
_
F(t)η(t)dt = 0.
I have some function of t; I multiply it by η(t); and I integrate it from one
end to the other. And no matter what the η is, I get zero. That means
that the function F(t) is zero. That’s obvious, but anyway I’ll show you
one kind of proof.
Suppose that for η(t) I took something which was zero for all t except
right near one particular value (See Figure 4.1.22). It stays zero until it
gets to this t, then it blips up for a moment and blips right back down.
t
1
t
2
t
η(t)
Figure 4.1.22.
When we do the integral of this η times any function F, the only place that
you get anything other than zero was where η(t) was blipping, and then
you get the value of F at that place times the integral over the blip. The
integral over the blip alone isn’t zero, but when multiplied by F it has to
be; so the function F has to be zero everywhere.
We see that if our integral is zero for any η, then the coeﬃcient of η must
be zero. The action integral will be a minimum for the path that satisﬁes
this complicated diﬀerential equation:
_
−m
d
2
x
dt
2
−V
(x)
_
= 0.
It’s not really so complicated; you have seen it before. It is just F = ma. The
ﬁrst term is the mass times acceleration, and the second is the derivative
of the potential energy, which is the force.
52 4 Vector Valued Functions
So, for a conservative system at least, we have demonstrated that the
principle of least action gives the right answer; it says that the path that
has the minimum action is the one satisfying Newton’s law.
One remark: I did not prove it was a minimum—maybe it’s a maximum.
In fact, it doesn’t really have to be a minimum. It is quite analogous to
what we found for the ‘principle of least time’ which we discussed in optics.
There also, we said at ﬁrst it was ‘least’ time. It turned out, however, that
there were situations in which it wasn’t the least time. The fundamental
principle was that for any ﬁrstorder variation away from the optical path,
the change in time was zero; it is the same story. What we really mean by
‘least’ is that the ﬁrstorder change in the value of S, when you change the
path, is zero. It is not necessarily a ‘minimum’.
See the “Feynman Lectures on Physics” for the remainder of the lecture,
which involves extending the above ideas to three dimensions and similar
topics.
4.3 Orbits of Planets 53
Supplement to §4.3
The Orbits of Planets
The purpose of this supplement is to demonstrate that the motion of a
body moving under the inﬂuence of Newton’s gravitational law—that is,
the inverse square force law—moves in a conic section. In the text we dealt
with circular orbits only, but some of the results, such as Kepler’s law on
the relation between the period and the size of the orbit can be generalized
to the case of elliptical orbits.
History. At this point it is a good idea to review some of the history
given at the beginning of the text in addition to the relevant sections of
the text (especially ¸4.1). Recall, for example, that one of the important
observational points that was part of the Ptolemaic theory as well as being
properly explained by the fact that planets move (to an excellent degree of
approximation) in ellipses about the sun, is the appearance of retrograde
motion of the planets, as in Figure 4.3.1.
Figure 4.3.1. the apparent motion of the planets. The photograph shows the move
ments of Mercury, Venus, Mars, Jupiter, and Saturn. the diagram shows the paths traced
by the planets as seen from the Earth, which the Ptolemaic theory tried to explain.
54 4 Vector Valued Functions
Methodology. Our approach in this supplement is to use the laws of
conservation of energy and conservation of angular momentum. This means
that we will focus on a combination of techniques using the basic laws of
mechanics together with techniques from diﬀerential equations together
with a couple of tricks. Other approaches, and in fact, Newton’s original
approach were much more geometrical.
10
The Kepler Problem. As in the text, we assume that the Sun (of mass
M) is so massive that it remains ﬁxed at the origin and our planet (with
mass m) revolves about the Sun according to Newton’s second law and
moving in the ﬁeld determined by Newton’s Law of Gravitation. Therefore,
the Kepler problem is the following: Demonstrate that the solution r(t) of
the equation
m¨r =
GmM
r
2
(4.3.3)
is a conic section. The procedure is to suppose that we have a solution,
which we will refer to as an orbit, and to show it is a conic section. By
reversing the argument, one can also show that any conic section, with a
suitable time parametrization, is also a solution.
Energy and Angular Momentum. We saw in the paragraph in ¸4.3
of the text entitled Conservation of Energy and Escaping the Earth’s Grav
itational Field that any solution of equation (4.3.3) obeys the law of con
servation of energy; that is, the quantity
E =
1
2
m˙ r
2
−
GmM
r
(4.3.4)
called the energy of the solution, is constant in time.
The second fact that we will need is conservation of angular momen
tum, which was given in Exercise 20, ¸4.1. This states that the angular
momentum of a solution is constant in time, namely the cross product
J = mr ˙ r (4.3.5)
is time independent for each orbit.
Orbits Lie in Planes. The ﬁrst thing we observe is that any solution
must lie in a plane. This is simply because of conservation of angular mo
mentum: since J is a constant and since J = mr ˙ r, it follows that r lies
in the plane perpendicular to the (constant) vector J.
Introducing Polar Coordinates. From the above consideration, we
can assume that our orbit lies in a plane, which we can take to be the usual
10
The geometric approach, along with a lot of interesting history, is also emphasized
in the little book Feynman’s Lost Lecture: The Motion of Planets Around the Sun by
David L. Goodstein and Judith R. Goodstein, W.W. Norton and Co., New York.
4.3 Orbits of Planets 55
xyplane. Let us introduce polar coordinates (r, θ) in that plane as usual,
so that the components of the position vector r are given by
r = (r cos θ, r sin θ).
Diﬀerentiating in t, we see that the velocity vector has components given
by
˙ r = ( ˙ r cos θ −r
˙
θ sin θ, ˙ r sin θ +r
˙
θ cos θ)
and so one readily computes that
˙ r
2
= ˙ r
2
+r
2
˙
θ
2
. (4.3.6)
Substituting (4.3.6) into conservation of energy gives
E =
1
2
m
_
˙ r
2
+r
2
˙
θ
2
_
−
GmM
r
(4.3.7)
Similarly, the angular momentum is computed by taking the cross product
of r and ˙ r:
J = m
¸
¸
¸
¸
¸
¸
i j k
r cos θ r sin θ 0
˙ r cos θ −r
˙
θ sin θ ˙ r sin θ +r
˙
θ cos θ 0
¸
¸
¸
¸
¸
¸
(4.3.8)
= mr
2
˙
θ k (4.3.9)
Thus,
J = mr
2
˙
θ (4.3.10)
is constant in time.
Kepler’s Second Law. We notice that the expression for J is closely
related to the area element in polar coordinates, namely dA =
1
2
r
2
dθ. In
fact, measuring area from a given reference θ
0
, we get
dA
dt
=
J
2m
(4.3.11)
which is a constant. Thus, we get Kepler’s second law, namely that for an
orbit, equal areas are swept out in equal times. Also note that this works for
any central force motion law as it depends only on conservation of angular
momentum.
Rewriting the Energy Equation. Next notice that the energy equa
tion (4.3.7) can be written, with
˙
θ eliminated using (4.3.10), as
E =
1
2
m˙ r
2
+
J
2
2mr
2
−
GmM
r
(4.3.12)
It will be convenient to seek a description of the orbit in polar form as
r = r(θ).
56 4 Vector Valued Functions
A Trick. Solving diﬀerential equations to get explicit formulas often in
volves a combination of clever guesses, insight and luck; the situation with
the Kepler problem is one of these situations. A trick that works is to intro
duce the new dependent variable u = 1/r (of course one has to watch out
for the possibility that the orbit passes through the origin, in which case
r would be zero.) But watch the nice things that happen with this choice.
First of all, by the chain rule,
du
dθ
= −
1
r
2
dr
dθ
and second, notice that, using the chain rule, the preceding equation and
(4.3.10),
˙ r =
dr
dθ
˙
θ = −r
2
˙
θ
du
dθ
= −
J
m
du
dθ
and so from (4.3.12) we ﬁnd that
E =
J
2
2m
_
du
dθ
_
2
+
J
2
u
2
2m
−GmMu, (4.3.13)
that is,
2Em
J
2
=
_
du
dθ
_
2
+u
2
−
2GMm
2
J
2
u, (4.3.14)
Notice the interesting way that only u and its derivative with respect to
θ appears and that the denominators containing r have been eliminated.
This is the main purpose of doing the change of dependent variable from r
to u.
Completing the Square. Next, we are going to eliminate the term
linear in u in (4.3.14) by completing the square. Let α = GMm
2
/J
2
, so
that (4.3.14) becomes
2Em
J
2
=
_
du
dθ
_
2
+u
2
−2αu, (4.3.15)
Now let
v = u −α
so that equation (4.3.15) becomes
_
dv
dθ
_
2
+v
2
= β
2
, (4.3.16)
where
β
2
=
2Em
J
2
−α
2
.
4.3 Orbits of Planets 57
We make one more change of variables to
w =
1
α
v =
1
α
u −1
so that (4.3.16) becomes
_
dw
dθ
_
2
+w
2
= e
2
, (4.3.17)
where
e
2
=
β
2
α
2
=
2Em
α
2
J
2
−1.
Solving the Equation. The solution of the equation (4.3.17) is readily
veriﬁed to be
w = e cos(θ −θ
0
)
where θ
0
is a constant of integration. In terms of the original variable r,
this means that
r (e cos(θ −θ
0
) + 1) =
1
α
. (4.3.18)
Orbits are Conics. Equation (4.3.18) in fact is the equation of a conic
written in polar coordinates, where e is the eccentricity of the orbit, which
determines its shape. The quantity l = 1/α determines its scale and θ
0
its orientation relative to the xyaxes; it is also the angle of the closest
approach to the origin.
In the case that 0 ≤ e < 1 (and E < 0) one has an ellipse of the form
(x +ae)
2
a
2
+
y
2
b
2
= 1
where
a =
l
1 −e
2
= −
GmM
2E
and
b
2
= al = −
J
2
2mE
.
The case e = 0 gives circular orbits. One also gets hyperbolic orbits if e > 1
and parabolic orbits in the special case when e = 1. Thus, as e ranges from
0 to 1, the ellipse gets a more and more elongated shape.
Kepler’s 3rd Law. From Kepler’s second law and the fact that the area
of an ellipse is πab, one ﬁnds that the period T of an elliptical orbit is
related to the semimajor axis a by
_
T
2π
_
2
=
a
3
GM
which is Kepler’s third law. Note that this reduces to what we saw in the
text for circular orbits in the case of a circle (when a is the radius of the
circle).
58 4 Vector Valued Functions
Supplement to §4.4
Flows and the Geometry of the Divergence
Flows of Vector Fields. It is convenient to give the unique solution
through a given point at time 0 a special notation:
φ(x, t) =
_
the position of the point on the ﬂow line
through the point x after time t has elapsed.
_
.
With x as the initial condition, follow along the ﬂow line for a time period
t until the new position φ(x, t) is reached (see Figure 1). Alternatively,
x
[time t = 0]
the flow line
passing through x
φ(x, t)
[time t]
Figure 4.4.1. The deﬁnition of the ﬂow φ(x, t) of F.
φ(x, t) is deﬁned by:
∂
∂t
φ(x, t) = F(φ(x, t))
φ(x, 0) = x
_
. (1)
We call the mapping φ, which is regarded as the function of the variables
x and t, the ﬂow of F.
Let D
x
denote diﬀerentiation with respect to x, holding t ﬁxed. It is
proved in courses on diﬀerential equations that φ is, in fact, a diﬀerentiable
function of x. By diﬀerentiating equation (1) with respect to x, we get
D
x
∂
∂t
φ(x, t) = D
x
[F(φ(x, t))].
The equality of mixed partial derivatives may be used on the lefthand side
of this equation and the chain rule applied to the righthand side, yielding
∂
∂t
D
x
φ(x, t) = DF(φ(x, t))D
x
φ(x, t), (2)
4.4 Flows and the Geometry of the Divergence 59
where DF(φ(x, t)) denotes the derivative of F evaluated φ(x, t). Equation
(2), a linear diﬀerential equation for D
x
φ(x, t) is called the equation of
the ﬁrst variation. It will be useful in our discussion of divergence and
curl in the next section. Both D
x
F(φ) and D
x
φ are 3 3 matrices since F
and φ take values in R
3
and are diﬀerentiated with respect to x ∈ R
3
; for
vector ﬁelds in the plane, they would be 2 2 matrices.
The Geometry of the Divergence. We now study the geometric mean
ing of the divergence in more detail. This discussion depends on the concept
of the ﬂow φ(x, t) of a vector ﬁeld F given in the preceding paragraph. See
Exercises 3, 4, and 5 below for the corresponding discussion of the curl.
Fix a point x and consider the three standard basis vectors i, j, k em
anating from x. Let ε > 0 be small and consider the basis vectors v
1
=
εi, v
2
= εj, v
3
= εk, also emanating from x. These vectors span a paral
lelepiped P(0). As time increases or decreases, the ﬂow φ(x, t) carries P(0)
into some object. For ﬁxed time, φ is a diﬀerentiable function of x (that
is, φ is a diﬀerentiable map of R
3
to R
3
). When ε is small, the image of
P(0) under φ can be approximated by its image under the derivative of
φ with respect to x. Recall that if v is a vector based at a point P
1
and
ending at P
2
, so v = P
2
−P
1
, then φ(P
2
, t) −φ(P
1
, t) ≈ D
x
φ(x, t) v. Thus
for ﬁxed time and small positive ε, P(0) is approximately carried into a
parallelepiped spanned by the vectors v
1
(t), v
2
(t), v
3
(t) given by
v
1
(t) = D
x
φ(x, t) v
1
v
2
(t) = D
x
φ(x, t) v
2
v
3
(t) = D
x
φ(x, t) v
3
_
_
_
. (3)
Since φ(x, 0) = x for all x, it follows that v
1
(0) = v
1
, v
2
(0) = v
2
, and
v
3
(0) = v
3
. In summary, the vectors v
1
(t), v
2
(t), v
3
(t) span a parallelepiped
P(t) that moves in time (see Figure 4.4.2).
flow line
x
v
1
v
2
v
3
v
1
(t)
v
2
(t)
v
3
(t)
P(0)
P(t)
Figure 4.4.2. The moving basis v
1
(t), v
2
(t), v
3
(t) and the associated parallepiped.
60 4 Vector Valued Functions
Let the volume of P(t) be denoted by 1(t). The main geometric meaning
of divergence is given by the following theorem.
Theorem.
div F(x) =
1
1(0)
d
dt
1(t)
¸
¸
¸
¸
t=0
.
Proof. By equation (2) of the previous paragraph,
d
dt
v
i
(t) = DF(φ(x, t)) (D
x
φ(x, t) v
i
) (4)
for i = 1, 2, 3. Since φ(x, 0) = x, it follows that D
x
φ(x, 0) is the identity
matrix, so that evaluation at t = 0 gives
d
dt
v
i
(t)
¸
¸
¸
¸
t=0
= DF(x) v
i
.
The volume 1(t) is given by the triple product:
1(t) = v
1
(t) [v
2
(t) v
3
(t)].
Using the diﬀerentiation rules of ¸4.1 and the identities
v
1
[v
2
v
3
] = v
2
[v
3
v
1
] = v
3
[v
1
v
2
],
equation (4) gives
d1
dt
=
dv
1
dt
[v
2
(t) v
3
(t)] +v
1
(t)
_
dv
2
dt
v
3
(t)
_
+v
1
(t)
_
v
2
(t)
dv
3
dt
_
=
dv
1
dt
_
v
2
(t) v
3
(t)] +
dv
2
dt
[v
3
(t) v
1
(t)] +
dv
3
dt
[v
1
(t) v
2
(t)
_
.
At t = 0, substitution from formula (3) and the facts that
v
1
v
2
= εv
3
, v
3
v
1
= εv
2
and v
2
v
3
= εv
1
,
gives
d1
dt
¸
¸
¸
¸
t=0
= ε
3
[DF(x)i] i +ε
3
[DF(x)j] j +ε
3
[DF(x)k] k. (5)
Since F = F
1
i + F
2
j + F
3
k, we get [DF(x)i] i = ∂F
1
/∂x. Similarly, the
second and third terms of equation (5) are ε
3
(∂F
2
/∂y) and ε
3
(∂F
3
/∂z).
Substituting these facts into equation (5) and dividing by 1(0) = ε
3
proves
the theorem.
4.4 Flows and the Geometry of the Divergence 61
The reader who is familiar with a little more linear algebra can prove
this generalization of the preceding Theorem:
11
Let v
1
, v
2
and v
3
be any
three noncoplanar (not necessarily orthonormal) vectors emanating from x
that ﬂow according to the formula
v
i
(t) = D
x
φ(x, t) v
i
, i = 1, 2, 3.
The vectors v
1
(t), v
2
(t), v
3
(t) span a parallelepiped P(t) with volume
1(t). Then
1
1(0)
d1
dt
¸
¸
¸
¸
t=0
= div F(x). (6)
In other words, the divergence of F at x is the rate at which the volumes
change, per unit volume. “Rate” refers to the rate of change with respect
to time as the volumes are transported by the ﬂow.
Exercises.
1. If f(x, t) is a realvalued function of x and t, deﬁne the material
derivative of f relative to a vector ﬁeld F as
Df
Dt
=
∂f
∂t
+∇f(x) F.
Show that Df/Dt is the t derivative of f(φ(x, t), t) (i.e., the t deriva
tive of f transported by the ﬂow of F).
2. (a) Assuming uniqueness of the ﬂow lines through a given point at a
given time, prove the following property of the ﬂow φ(x, t) of a vector
ﬁeld F:
φ(x, t +s) = φ(φ(x, s), t).
(b) What is the corresponding property for D
x
φ?
3. Let v and w be two vectors emanating from the origin and let them
be moved by the derivative of the ﬂow:
v(t) = D
x
φ(0, t)v, w(t) = D
x
φ(0, t)w,
so that at time t = 0 and at the origin 0 in R
3
,
dv
dt
¸
¸
¸
¸
t=0
= D
x
F(0) v and
dw
dt
¸
¸
¸
¸
t=0
= D
x
F(0) w.
11
The reader will need to know how to write the matrix of a linear transformation
with respect to a given basis and be familiar with the fact that the trace of a matrix is
independent of the basis.
62 4 Vector Valued Functions
Show that
d
dt
v w
¸
¸
¸
¸
t=0
= [D
x
F(0) v] w+v [D
x
F(0) w]
= [(D
x
F(0) + [D
x
F(0)]
T
)v] w].
4. Any matrix A can be written (uniquely) as the sum of a symmetric
matrix (a matrix S is symmetric if S
T
= S) and an antisymmetric
matrix (W
T
= −W) as follows:
A =
1
2
(A+A
T
) +
1
2
(A−A
T
) = S +W.
In particular, for A = D
x
F(0),
S =
1
2
[D
x
F(0) + [D
x
F(0)]
T
]
and
W =
1
2
[D
x
F(0) + [D
x
F(0)]
T
].
We call S the deformation matrix and W the rotation matrix.
Show that the entries of W are determined by
w
12
= −
1
2
(curl F)
3
, w
23
= −
1
2
(curl F)
1
, and w
31
= −
1
2
(curl F)
2
.
5. Let w =
1
2
(∇ F)(0). Assume that axes are chosen so that w is
parallel to the z axis and points in the direction of k. Let v = wr,
where r = xi+yj+zk, so that v is the velocity ﬁeld of a rotation about
the axis w with angular velocity ω = w and with curl v = 2w.
Since r is a function of (x, y, z), v is also a function of (x, y, z). Show
that the derivative of v at the origin is given by
Dv(0) = W =
_
_
0 −ω 0
ω 0 0
0 0 0
_
_
.
Interpret the result.
6. Let
V(x, y, z) = −yi +xj, W(x, y, z) =
V(x, y, z)
(x
2
+y
2
)
1/2
,
and
Y(x, y, z) =
V(x, y, z)
(x
2
+y
2
)
.
(a) Compute the divergence and curl of V, W, and Y.
4.4 Flows and the Geometry of the Divergence 63
(b) Find the ﬂow lines of V, W, and Y.
(a) How will a small paddle wheel behave in the ﬂow of each of
V, W, and Y.
7. Let φ(x, t) be the ﬂow of a vector ﬁeld F. Let x and t be ﬁxed.
For small vectors v
1
= εi, v
2
= εj and v
3
= εk emanating from
x, let P(0) be the parallelepiped spanned by v
1
, v
2
, v
3
. Argue that
for small positive ε, P(0) is carried by the ﬂow to an approximate
parallelepiped spanned by v
1
(t), v
2
(t), v
3
(t) given by formula (3).
64 4 Vector Valued Functions
Page 65
5
Double and Triple Integrals
Both the derivative and integral can be based on the notion
of a transition point. For example, a function is Riemann inte
grable when there is a transition point between the lower and
upper sums.
Alan Weinstein, 1982
Supplement to §5.2
Alternative Deﬁnition of the Integral
There is another approach to the deﬁnition of the integral based on step
functions that you may want to mention or have some of your better stu
dents exposed to; we present this (optional) deﬁnition ﬁrst.
We say that a function g(x, y) deﬁned in R = [a, b] [c, d] is a step
function provided there are partitions
a = x
0
< x
1
< x
2
< . . . < x
n
= b
of the closed interval [a, b] and
c = y
0
< y
1
< y
2
< . . . < y
m
= d
of the closed interval [c, d] such that, in each of the mn open rectangles
R
ij
= (x
i−1
, x
i
) (y
j−1
, y
j
),
66 5 Double and Triple Integrals
the function g(x, y) has a constant value k
ij
. The graph of a generic step
function is shown in Figure 5.2.1.
x
0
= a
x
1
b = x
2
y
3
= d y
1
y
2
y
c = y
0
x
z
Graph of g
Figure 5.2.1. The function g is a step function since it is constant on each subrectangle.
We will deﬁne the integral
__
R
g(x, y)dxdy
of a step function over a rectangle in such a way that, if g(x, y) ≥ 0 or R,
the integral equals the volume of the region V under the graph. Since the
part of V lying over R
ij
has height k
ij
and base area (x
i
−x
i−1
)(y
i
−y
i−1
),
the volume of this part is k
ij
(x
i
− x
i−1
)(y
i
− y
i−1
), or k
ij
∆x
i
∆y
j
where
∆x
i
= x
i
− x
i−1
and ∆y
i
= y
i
− y
i−1
as in onevariable calculus. The
volume of V is the sum of all the k
ij
∆x
i
∆y
j
as i ranges from 1 to n and j
ranges from 1 to m (making nm terms in all). We denote this sum by
n,m
i=1,j=1
.
Using this geometric guide, we deﬁne
__
R
g(x, y)dxdy =
n,m
i=1,j=1
k
ij
∆x
i
∆y
j
for every step function g, whether or not its values k
ij
are all nonnegative.
Example 1. Let g take values on the rectangles as shown in Figure 5.2.2.
Calculate the integral of g over the rectangle R = [0, 5] [0, 3].
5,2 Alternative Deﬁnition of the Integral 67
x
2 5
1
2
3
4
6
–8
–1
3
2
y
Figure 5.2.2. Find
__
D
g(x, y)dxdy if g take the values shown.
Solution. The integral of g is the sum of the values of g times the areas
of the rectangles:
__
R
g(x, y)dxdy =−8 2 + 2 3 + 6 2
+ 3 3 + 4 2 −1 3 = 16.
Of course, the functions which we want to integrate are usually not step
functions. To deﬁne their integrals, we use a comparison method whose
origins go back as far as Archimedes.
If f
1
(x, y) and f
2
(x, y) are any two functions deﬁned on the rectangle
R, then any reasonable deﬁnition of the double integral of f
2
should be
larger than that of f
1
if f
1
(x, y) ≤ f
2
(x, y) everywhere on R. Thus, if f is
any function which we wish to integrate over R, and if g and h are step
functions on R such that g(x, y) ≤ f(x, y) ≤ h(x, y) for all (x, y) in R, then
the number
__
R
f(x, y)dxdy which we are trying to deﬁne must lie between
the numbers
__
R
g(x, y)dxdy and
__
R
h(x, y)dxdy which have already been
deﬁned. This condition on the integral actually becomes a deﬁnition if we
rephrase it as follows.
Alternative Deﬁnition of the Double Integral. A function f deﬁned
on a rectangle R is said to be integrable on R if for any positive number ,
there exist step functions g(x, y) and h(x, y) with g(x, y) ≤ f(x, y) ≤ h(x, y)
for all (x, y) in R and such that the diﬀerence
__
R
h(x, y)dxdy −
__
R
g(x, y)dxdy
is less than . We shall call such step functions g and h “surrounding func
tions”. When this condition holds, it can be shown that there is just one
68 5 Double and Triple Integrals
number I with the property that
__
R
g(x, y)dxdy ≤ I ≤
__
R
h(x, y)dxdy
for surrounding functions g and h as above. The number I is called the
integral of f on R and is denoted by
__
R
f(x, y)dxdy.
Example 2. Let R be the rectangle 0 ≤ x ≤ 2, 1 ≤ y ≤ 3, and let
f(x, y) = x
2
y. Choose a step function h(x, y) ≥ f(x, y) to show that
__
R
f(x, y)dxdy ≤ 25.
Solution. The constant function h(x, y) = 12 ≥ f(x, y) has integral 12
4 = 48, so we get only the crude estimate
__
R
f(x, y)dxdy ≤ 48. To get a
better one, divide R into four pieces:
R
1
= [0, 1] [1, 2], R
3
= [1, 2] [1, 2],
R
2
= [0, 1] [2, 3], R
4
= [1, 2] [2, 3].
Let h be the function given by taking the maximum value of f on each
subrectangle (evaluated at the upper righthand corner); that is,
h(x, y) = 2 on R
1
, 3 on R
2
, 8 on R
3
, and 12 on R
4
.
Therefore, the integral of h is
__
R
h(x, y)dxdy = 2 1 + 3 1 + 8 1 + 12 1 = 25.
Since h ≥ f, we get
__
R
f(x, y)dxdy ≤ 25.
Alternative Proof of Reduction to Iterated Integrals. The upper
and lower sums approach can also be used to give proof of the reduction
to iterated integrals. We give this (optional) proof here. We ﬁrst treat step
functions. Let g be a step function, with g(x, y) = k
ij
on the rectangle
(t
i−1
, t
i
) (s
j−1
, s
j
), so that
__
D
g(x, y)dxdy =
n,m
i=1,j=1
k
ij
∆t
i
∆s
j
.
If the summands k
ij
∆t
i
∆s
j
are laid out in a rectangular array, they may
be added by ﬁrst adding along rows and then adding up the subtotals, just
as in the text (see Figure 5.2.3):
5,2 Alternative Deﬁnition of the Integral 69
k
11
∆t
1
∆s
1
k
21
∆t
2
∆s
1
. . . k
n1
∆t
n
∆s
2
−→
_
n
i=1
k
i1
∆t
i
_
∆s
1
k
12
∆t
1
∆s
2
k
22
∆t
2
∆s
2
. . . k
n2
∆t
n
∆s
2
−→
_
n
i=1
k
i2
∆t
i
_
∆s
2
.
.
.
.
.
.
.
.
.
k
1m
∆t
1
∆s
m
k
2m
∆t
2
∆s
m
. . . k
nm
∆t
n
∆s
m
−→
_
n
i=1
k
im
∆t
i
_
∆s
m
m
j=1
_
n
i=1
k
ij
∆t
i
_
∆s
j
Figure 5.2.3. Reduction to iterated integrals.
The coeﬃcient of ∆s
j
in the sum over the jth row,
n
j=1
k
ij
∆t
i
, is equal
to
_
b
a
g(x, y)dx for any y with s
j−1
< y < s
j
, since, for y ﬁxed, g(x, y) is a
step function of x. Thus, the integral
_
b
a
g(x, y)dx is a step function of y,
and its integral with respect to y is the sum:
_
d
c
_
_
b
a
g(x, y)dx
_
dy =
m
j=1
_
n
i=1
k
ij
∆t
i
_
∆s
j
=
_ _
D
g(x, y)dxdy.
Similarly, by summing ﬁrst over columns and then over rows, we obtain
__
D
g(x, y)dxdy =
_
b
a
_
_
d
c
g(x, y)dy
_
dx.
The theorem is therefore true for step functions.
Now let f be integrable on D = [a, b] [c, d] and assume that the iterated
integral
_
b
a
_
_
d
c
f(x, y)dy
_
dx exists. Denoting this integral by S
0
, we will
show that every lower sum for f on D is less than or equal to S
0
, while
every upper sum is greater than or equal to S
0
, so S
0
must be the integral
of f over D.
To carry out our program, let g be any step function such that
g(x, y) ≤ f(x, y) (1)
70 5 Double and Triple Integrals
for all (x, y) in D. Integrating (1) with respect to y and using “monotonic
ity” of the onevariable integral, we obtain
_
d
c
g(x, y)dy ≤
_
d
c
f(x, y)dy (2)
for all x in [a, b]. Integrating (2) with respect to x and applying monotonic
ity once more gives
_
b
a
_
_
d
c
g(x, y)dy
_
dx ≤
_
b
a
_
_
d
c
f(x, y)dy
_
dx. (3)
Since g is a step function, it follows from the ﬁrst part of this proof that
the lefthand side of (3) is equal to the lower sum
__
D
g(x, y)dxdy; the
righthand side of (3) is just S
0
, so we have shown that every lower sum is
less than or equal to S
0
. The proof that every upper sum is greater than
or equal to S
0
is similar, and so we are done.
5.6 Technical Integration Theorems 71
§5.6 Technical Integration Theorems
This section provides the main ideas of the proofs of the existence and
additivity of the integral that were stated in ¸5.2 of the text. These proofs
require more advanced concepts than those needed for the rest of this chap
ter.
Uniform Continuity. The notions of uniform continuity and the com
pleteness of the real numbers, both of which are usually treated more fully
in a juniorlevel course in mathematical analysis or realvariable theory, are
called upon here.
Deﬁnition. Let D ⊂ R
n
and f : D → R. Recall that f is said to be
continuous at x
0
∈ D provided that for all ε > 0 there is a number δ > 0
such that if x ∈ D and x −x
0
 < δ, then f(x) −f(x
0
) < ε. We say f
is continuous on D if it is continuous at each point of D.
The function f is said to be uniformly continuous on D if for every
number ε > 0 there is a δ > 0 such that whenever x, y ∈ D and x−y < δ,
then f(x) −f(y) < ε.
The main diﬀerence between continuity and uniform continuity is that
for continuity, δ can depend on x
0
as well as ε, whereas in uniform conti
nuity δ depends only on ε. Thus any uniformly continuous function is also
continuous. An example of a function that is continuous but not uniformly
continuous is given in Exercise 2 in this supplement. The distinction be
tween the notions of continuity and uniform continuity can be rephrased:
For a function f that is continuous but not uniformly continuous, δ cannot
be chosen independently of the point of the domain (the x
0
in the deﬁni
tion). The deﬁnition of uniform continuity states explicitly that once you
are given an ε > 0, a δ can be found independent of any point of D.
Recall from ¸3.3 that a set D ⊂ R
n
is bounded if there exists a number
M > 0 such that x ≤ M for all x ∈ D. A set is closed if it contains all
its boundary points. Thus a set is bounded if it can be strictly contained
in some (large) ball. The next theorem states that under some conditions
a continuous function is actually uniformly continuous.
Theorem. The Uniform Continuity Principle. Every function that
is continuous on a closed and bounded set D in R
n
is uniformly continuous
on D.
The proof of this theorem will take us too far aﬁeld;
1
however, we can
prove a special case of it, which is, in fact, suﬃcient for many situations
relevant to this text.
1
The proof can be found in texts on mathematical analysis. See, for example, J.
Marsden and M. Hoﬀman, Elementary Classical Analysis, 2nd ed., Freeman, New York,
1993, or W. Rudin, Principles of Mathematical Analysis, 3rd ed., McGrawHill, New
York, 1976.
72 5 Double and Triple Integrals
Proof of a special case. Let us assume that D = [a, b] is a closed interval
on the line, that f : D → R is continuous, that df/dx exists on the open
interval (a, b), and that df/dx is bounded (that is, there is a constant C > 0
such that [df(x)/dx[ ≤ C for all x in (a, b)). To show that these conditions
imply f is uniformly continuous, we use the mean value theorem as follows:
Let ε > 0 be given and let x and y lie on D. Then by the mean value
theorem,
f(x) −f(y) = f
(c)(x −y)
for some c between x and y. By the assumed boundedness of the derivative,
[f(x) −f(y)[ ≤ C[x −y[.
Let δ = ε/C. If [x −y[ < δ, then
[f(x) −f(y)[ < C
ε
C
= ε.
Thus f is uniformly continuous. (Note that δ depends on neither x nor y,
which is a crucial part of the deﬁnition.)
This proof also works for regions in R
n
that are convex; that is, for any
two points x, y in D, the line segment c(t) = tx + (1 − t)y, 0 ≤ t ≤ 1,
joining them also lies in D. We assume f is diﬀerentiable (on an open set
containing D) and that ∇f(x) ≤ C for a constant C. Then the mean
value theorem applied to the function h(t) = f(c(t)) gives a point t
0
such
that
h(1) −h(0) = [h
(t
0
)][1 −0]
or
f(x) −f(y) −h
(t
0
) = ∇f(c(t
0
)) c
(t
0
) = ∇f(c(t
0
)) (x −y)
by the chain rule. Thus by the CauchySchwartz inequality,
[f(x) −f(y)[ ≤ ∇f(c(t
0
))x −y ≤ Cx −y.
Then, as above, given ε > 0, we can let δ = ε/C.
Completeness. We now move on to the notion of a Cauchy sequence of
real numbers. In the deﬁnition of Riemann sums we obtained a sequence of
numbers ¦S
n
¦, n = 1, . . .. It would be nice if we could say that this sequence
of numbers converges to S (or has a limit S), but how can we obtain such a
limit? In the abstract setting, we know no more about S
n
than that it is the
Riemann sum of a (say, continuous) function, and, although this has not
yet been proved, it should be enough information to ensure its convergence.
Thus we must determine a property for sequences that guarantees their
convergence. We shall deﬁne a class of sequences called Cauchy sequences,
and then take as an axiom of the real number system that all such sequences
5.6 Technical Integration Theorems 73
converge to a limit
2
. The determination in the nineteenth century that
such an axiom was necessary for the foundations of calculus was a major
breakthrough in the history of mathematics and paved the way for the
modern rigorous approach to mathematical analysis. We shall say more
this shortly.
Deﬁnition. A sequence of real numbers ¦S
n
¦, n = 1, . . ., is said to satisfy
the Cauchy criterion if for every ε > 0 there exists an N such that for
all m, n ≥ N, we have [S
n
−S
m
[ < ε.
If a sequence S
n
converges to a limit S, then S
n
is a Cauchy sequence.
To see this, we use the deﬁnition: For every ε > 0 there is an N such
that for all n ≥ N, [S
n
− S[ < ε. Given ε > 0, choose N
1
such that for
n ≥ N
1
, [S
n
−S[ < ε/2 (use the deﬁnition with ε/2 in place of ε). Then if
n, m ≥ N
1
,
[S
n
−S
m
[ = [S
n
−S +S −S
m
[ ≤ [S
n
−S[ +[S −S
m
[ <
ε
2
+
ε
2
= ε,
which proves our contention. The completeness axiom asserts that the con
verse is true as well:
Completeness Axiom of the Real Number. Every Cauchy sequence
¦S
n
¦ converges to some limit S.
Historical Note. Augustin Louis Cauchy (1789–1857), one of the great
est mathematicians of all time, deﬁned what we now call Cauchy sequences
in his Cours d’analyse, published in 1821. This book was a basic work on the
foundations of analysis, although it would be considered somewhat loosely
written by the standards of our time. Cauchy knew that a convergent se
quence was “Cauchy” and remarked that a Cauchy sequence converges. He
did not have a proof, nor could he have had one, since such a proof depends
on the rigorous development of the real number system that was achieved
only in 1872 by the German mathematician Georg Cantor (18451918).
It is now clear what we must do to ensure that the Riemann sums ¦S
n
¦
of, say, a continuous function on a rectangle converge to some limit S, which
would prove that continuous functions on rectangles are integrable; we must
show that ¦S
n
¦ is a Cauchy sequence. In demonstrating this, we use the
uniform continuity principle. The integrability of continuous functions will
be a consequence of the following two lemmas.
Lemma 1. Let f be a continuous function on a rectangle R in the plane,
and let ¦S
n
¦ be a sequence of Riemann sums for f. Then ¦S
n
¦ converges
to some number S.
2
Texts on mathematical analysis, such as those mentioned in the preceding footnote,
sometimes use diﬀerent axioms, such as the least upper bound property. In such a setting,
our completeness axiom becomes a theorem.
74 5 Double and Triple Integrals
Proof. Given a rectangle R ⊂ R
2
, R = [a, b] [c, d], we have the regular
partition of R, a = x
0
< x
1
< < x
n
= b, c = y
0
< y
1
< < y
n
= d,
discussed in ¸5.2 of the text. Recall that
∆x = x
j+1
−x
j
=
b −a
n
, ∆y = y
k+1
−y
k
=
d −c
n
,
and
S
n
=
n−1
j,k=0
f(c
jk
)∆x∆y,
where c
jk
is an arbitrarily chosen point in R
jk
= [x
j
, x
j+1
] [y
k
, y
k+1
].
The sequence ¦S
n
¦ is determined only by the selection of the points c
jk
.
For the purpose of the proof we shall introduce a slightly more compli
cated but very precise notation: we set
∆x
n
=
b −a
n
and ∆y
n
=
d −c
n
.
With this notation we have
S
n
=
n−1
j,k
f(c
jk
)∆x
n
∆y
n
. (1)
To show that ¦S
n
¦ satisﬁes the Cauchy criterion, we must show that
given ε > 0 there exists an N such that for all n, m ≥ N, [S
n
− S
m
[ ≤ ε.
By the uniform continuity principle, f is uniformly continuous on R. Thus
given ε > 0 there exists a δ > 0 such that when x, y ∈ R, x−y < δ, then
[f(x) − f(y)[ < ε/[2 area(R)] (the quantity ε/[2 area(R)] is used in place
of ε in the deﬁnition). Let N be so large that for any m ≥ N the diameter
(length of a diagonal) of any subrectangle R
jk
in the mth regular partition
of R is less than δ. Thus if x, y are points in the same subrectangle, we will
have the inequality [f(x −f(y))[ < ε/[2 area(R)].
Fix m, n ≥ N. We will show that [S
n
− S
m
[ < ε. This shows that ¦S
n
¦
is a Cauchy sequence and hence converges. Consider the mnth = (m times
n)th regular partition of R. Then
S
mn
=
r,t
f(˜ c
rt
)∆x
mn
∆y
mn
,
where ˜ c
rt
is a point in the rtth subrectangle. Note that each subrectangle
of the mnth partition is a subrectangle of both the mth and the nth regular
partitions (see Figure 5.6.1).
Let us denote the subrectangles in the mnth subdivision by
˜
R
rt
and
those in the nth subdivision by R
jk
. Thus each
˜
R
rt
⊂ R
jk
for some jk, and
hence we can rewrite formula (1) as
S
n
=
n−1
j,k
_
_
˜
Rrt⊂R
jk
f(c
jk
)∆x
mn
∆y
mn
_
_
. (1
)
5.6 Technical Integration Theorems 75
Figure 5.6.1. The shaded box shows a subrectangle in the mnth partition, and the
darkly outlined box, a subrectangle in the mth partition.
Here we are using the fact that
˜
Rrt⊂R
jk
f(c
jk
)∆x
mn
∆y
mn
= f(c
jk
)∆x
n
∆y
n
,
where the sum is taken over all subrectangles in the mnth subdivision
contained in a ﬁxed rectangle R
jk
in the nth subdivision. We also have the
identity
S
mn
=
mn−1
r,t
f(˜ c
rt
)∆x
mn
∆y
mn
. (2)
This relation can also be rewritten as
S
mn
=
j,k
˜
Rrt⊂R
jk
f(˜ c
rt
)∆x
mn
∆y
mn
. (2
)
where in equation (2
) we are ﬁrst summing over those subrectangles in
the mnth partition contained in a ﬁxed R
jk
and then summing over j, k.
Subtracting equation (2
) from equation (1
), we get
[S
n
−S
mn
[ =
¸
¸
¸
¸
¸
¸
j,k
˜
Rrt⊂R
jk
[f(c
jk
)∆x
mn
∆y
mn
−f(˜ c
rt
)∆x
mn
∆y
mn
]
¸
¸
¸
¸
¸
¸
≤
j,k
˜
Rrt⊂R
jk
[f(c
jk
) −f(˜ c
rt
)[∆x
mn
∆y
mn
.
76 5 Double and Triple Integrals
By our choice of δ and N, [f(c
jk
) − f(˜ c
rt
)[ < ε/[2 area(R)], and conse
quently the above inequality becomes
[S
n
−S
mn
[ ≤
j,k
˜
Rrt⊂R
jk
ε
2 area (R)
∆x
mn
∆y
mn
=
ε
2
.
Thus, [S
n
− S
mn
[ < ε/2 and similarly one shows that [S
n
− S
mn
[ < ε/2.
Since
[S
n
−S
m
[ = [S
n
−S
mn
+S
mn
−S
m
[ ≤ [S
n
−S
mn
[ +[S
mn
−S
m
[ < ε
for m, n ≥ N, we have shown ¦S
n
¦ satisﬁes the Cauchy criterion and thus
has a limit S.
We have already remarked that each Riemann sum depends on the se
lection of a collection of points c
jk
. In order to show that a continuous
function on a rectangle R is integrable, we must further demonstrate that
the limit S we obtained in Lemma 1 is independent of the choices of the
points c
jk
.
Lemma 2. The limit S in Lemma 1 does not depend on the choice of points
c
jk
.
Proof. Suppose we have two sequences of Riemann sums ¦S
n
¦ and ¦S
∗
n
¦
obtained by selecting two diﬀerent sets of points, say c
jk
and c
∗
jk
in each
nth partition. By Lemma 1 we know that ¦S
n
¦ converges to some number
S and ¦S
∗
n
¦ must also converge to some number, say S
∗
. We want to show
that S = S
∗
and shall do this by showing that given any ε > 0, [S−S
∗
[ < ε,
which implies that S must be equal to S
∗
(why?).
To start, we know that f is uniformly continuous on R. Consequently,
given ε > 0, there exits a δ such that [f(x) −f(y)[ < ε/[3 area(R)] when
ever x − y < δ. We choose N so large that whenever n ≥ N the di
ameter of each subrectangle in the nth regular partition is less than δ.
Since lim
n→∞
S
n
= S and lim
n→∞
S
∗
n
= S
∗
, we can assume that N
has been chosen so large that n ≥ N implies that [S
n
− S[ < ε/3 and
[S
∗
n
− S
∗
[ < ε/3. Also, for n ≥ N we know by uniform continuity that if
c
jk
and c
∗
jk
are points in the same subrectangle R
jk
of the nth partition,
then [f(c
jk
) −f(c
∗
jk
)[ < ε/[3 area(R)]. Thus
[S
n
−S
∗
n
[ =
¸
¸
¸
¸
¸
¸
j,k
f(c
jk
)∆x
n
∆y
n
−
j,k
f(c
∗
jk
)∆x
n
∆y
n
¸
¸
¸
¸
¸
¸
≤
j,k
[f(c
jk
) −f(c
∗
jk
)[∆x
n
∆y
n
<
ε
3
.
We now write
[S−S
∗
[ = [S−S
n
+S
n
−S
∗
n
+S
∗
n
−S
∗
[ ≤ [S−S
n
[ +[S
n
−S
∗
n
[ +[S
∗
n
−S[ < ε
5.6 Technical Integration Theorems 77
and so the lemma is proved.
Putting lemmas 1 and 2 together proves Theorem 1 of ¸5.2 of the main
text:
Theorem 1 of §5.2 of the Text. Any continuous function deﬁned on a
rectangle R is integrable.
Historical Note. Cauchy presented the ﬁrst published proof of this the
orem in his r´esum´e of 1823, in which he points out the need to prove the
existence of the integral as a limit of a sum. In this paper he ﬁrst treats
continuous functions (as we are doing now), but on an interval [a, b]. (The
proof is essentially the same.) However, his proof was not rigorous, since
it lacked the notion of uniform continuity, which was not available at that
time.
The notion of a Riemann sum S
n
for a function f certainly predates
Bernhard Riemann (1826–1866). The sums are probably named after him
because he developed a theoretical approach to the study of integration
in a fundamental paper on trigonometric series in 1854. His approach, al
though later generalized by Darboux (1875) and Stieltjes (1894), was to last
more than half a century until it was augmented by the theory Lebesque
presented to the mathematical world in 1902. This latter approach to inte
gration theory is generally studied in graduate courses in mathematics.
The proof of Theorem 2 of ¸5.2 is left to the reader in Exercises 4 to 6
at the end of this section. The main ideas are essentially contained in the
proof of Theorem 1.
Additivity Theorem. Our next goal will be to present a proof of prop
erty (iv) of the integral from ¸5.2, namely, its additivity. However, because
of some technical diﬃculties in establishing this result in its full generality,
we shall prove it only in the case in which f is continuous.
Theorem. Additivity of the Integral. Let R
1
and R
2
be two disjoint
rectangles (rectangles whose intersection contains no rectangle) such that
Q = R
1
∪R
2
is again a rectangle as in Figure 5.6.2. If f is a function that
is continuous over Q and hence over each R
i
, then
__
Q
f =
__
R1
f +
__
R2
f. (3)
Proof. The proof depends on the ideas that have already been presented
in the proof of Theorem 1.
The fact that f is integrable over Q, R
1
, and R
2
follows from Theorem
1. Thus all three integrals in equation (3) exist, and it is necessary only to
establish equality.
78 5 Double and Triple Integrals
x
y
a b
c
d
R
1
R
2
Q
b
1
R
jk
1
R
jk
2
Figure 5.6.2. Elements of a regular partition of R
1
and R
2
.
Without loss of generality we can assume that
R
1
= [a, b
1
] [c, d] and R
2
= [b
1
, b] [c, d]
(see Figure 5.6.2). Again we must develop some notation. Let
∆x
n
1
=
b
1
−a
n
, ∆x
n
2
=
b −b
1
n
, ∆x
n
=
b −a
n
, and ∆y
n
=
d −c
n
.
Let
S
1
n
=
j,k
f(c
1
jk
)∆x
n
1
∆y
n
(4)
S
2
n
=
j,k
f(c
2
jk
)∆x
n
2
∆y
n
(5)
S
n
=
j,k
f(c
jk
)∆x
n
∆y
n
(6)
where c
1
jk
, c
2
jk
, and c
jk
are points in the jkth subrectangle of the nth
regular partition of R
1
, R
2
, and Q, respectively. Let S
i
= lim
n→∞
S
i
n
, where
i = 1, 2, and S = lim
n→∞
S
n
. It must be shown that S = S
1
+ S
2
, which
we will accomplish by showing that for arbitrary ε > 0, [S −S
1
−S
2
[ < ε.
By the uniform continuity of f on Q we know that given ε > 0 there
is a δ > 0 such that whenever x − y < δ, [f(x) − f(y)[ < ε. Let N be
so big that for all n ≥ N, [S
n
− S[ < ε/3, [S
i
n
− S
i
[ < ε/3, i = 1, 2, and if
5.6 Technical Integration Theorems 79
x, y are any two points in any subrectangle of the nth partition of either
R
1
, R
2
, or Q then [f(x) − f(y)[ < ε/[3 area(Q)]. Let us consider the nth
regular partition of R
1
, R
2
, and Q. These form a collection of subrectangles
that we shall denote by R
1
jk
, R
2
jk
, R
jk
, respectively [see Figures 5.6.2 and
5.6.3(a)].
x
a b
c
d
R
jk
(a)
x
y
a b
c
d
b
1
R
jk
2
R
αβ
~
(b)
y
Figure 5.6.3. (a) A regular partition of Q. (b) The vertical and horizontal lines of this
subdivision are obtained by taking the union of the vertical and horizontal of Figures
5.6.2 and 5.6.3(a).
If we superimpose the subdivision of Q on the nth subdivisions of R
1
and R
2
, we get a new collection of rectangles, say
˜
R
αβ
, β = 1, . . . , n and
α = 1, . . . , m, where m > n; see Figure 5.6.3(b).
Each
˜
R
αβ
is contained in some subrectangle R
jk
of Q and in some sub
rectangle of the nth partition of either R
1
or R
2
. Consider equalities (4),
80 5 Double and Triple Integrals
(5), and (6) above. These can be rewritten as
S
i
n
=
j,k
˜
R
αβ
⊂Ri
f(c
i
jk
) area (
˜
R
αβ
) =
α,β
˜
R
αβ
⊂Ri
f(˜ c
αβ
) area (
˜
R
αβ
).
where ˜ c
αβ
= c
i
αβ
if
˜
R
αβ
⊂ R
i
jk
, i = 1, 2, and
S
n
=
j,k
˜
R
αβ
⊂R
jk
f(c
jk
) area (
˜
R
αβ
) =
α,β
f(c
∗
αβ
) area (
˜
R
αβ
).
where c
∗
α,β
= c
jk
if
˜
R
αβ
⊂ R
jk
.
For the reader encountering such index notation for the ﬁrst time, we
point out that
α,β
˜
R
αβ
⊂Ri
means that the summation is taken over those α’s and β’s such that the
corresponding rectangle
˜
R
αβ
is contained in the rectangle R
i
.
Now the sum for S
n
can be split into two parts:
S
n
=
α,β
˜
R
αβ
⊂R1
f(c
∗
αβ
) area (
˜
R
αβ
) +
α,β
˜
R
αβ
⊂R2
f(c
∗
αβ
) area (
˜
R
αβ
).
From these representations and the triangle inequality, it follows that
[S
n
−S
1
n
−S
2
n
[ ≤
¸
¸
¸
¸
¸
¸
¸
¸
¸
α,β
˜
R
αβ
⊂R1
[f(c
∗
αβ
) −f(˜ c
αβ
)] area(
˜
R
αβ
)
¸
¸
¸
¸
¸
¸
¸
¸
¸
+
¸
¸
¸
¸
¸
¸
¸
¸
¸
α,β
˜
R
αβ
⊂R2
[f(c
∗
αβ
) −f(˜ c
αβ
)] area(
˜
R
αβ
)
¸
¸
¸
¸
¸
¸
¸
¸
¸
≤
ε
3 area Q
α,β
˜
R
αβ
⊂R1
area (
˜
R
αβ
)
+
ε
3 area Q
α,β
˜
R
αβ
⊂R2
area (
˜
R
αβ
) <
ε
3
.
In this step we used the uniform continuity of f. Thus [S
n
−S
1
n
−S
2
n
[ < ε/3
for ≥ N. But
[S −S
n
[ <
ε
3
, [S
1
n
−S
1
[ <
ε
3
and [S
2
n
−S
2
[ <
ε
3
.
5.6 Technical Integration Theorems 81
As in Lemma 2, an application of the triangle inequality shows that [S −
S
1
−S
2
[ < ε, which completes the proof.
Example 6. Let C be the graph of a continuous function φ : [a, b] →R.
Let > 0 be any positive number. Show that C can be placed in a ﬁnite
union of boxes B
i
= [a
i
, b
i
][c
i
, d
i
] such that C does not contain a boundary
point of ∪B
i
and such that
area(B
i
) ≤ .
Solution. Let > 0. Since f is uniformly continuous there exists a δ,
with 0 < δ < 1 such that if [x −ω[ < δ, then [f(x) −f(ω)[ < /32. (We see
why the “32” as we proceed) — of course in reality one does a sketch of
the idea ﬁrst and through a trial run one “discovers” that the right number
to put here is indeed 32. Let n > 1/δ and subdivide the interval [a, b] into
2n equal parts with x
0
< x
1
< . . . < x
2n
the corresponding partition. Let
B
i
be the rectangle centered at (x
i
, f(x
i
)) with width 1/n and height /4.
The area of each rectangle is /4n. There are (2n + 1) such rectangles for
a total area of
_
4n
_
(2n + 1) =
2
+
2n
< .
It remains only to check that C does not contain a boundary point of
∪
i
B
i
. If the graph touches the top edge of some Box B
i
, then there is some
(x, f(x))B
i
such that [f(x) − f(x
i
)[ ≥ /8 (/8 is half the height of B
i
).
But by uniform continuity [f(x) −f(x
i
)[ < /32 a contradiction.
Similarly if C intersects a vertical side of B
i
it must do so in a portion
which is in the complement of B
i−1
∪B
i+1
. This again contradicts uniform
continuity.
Exercises.
1. Show that if a and b are two numbers such that for any ε > 0, [a−b[ <
ε, then a = b.
2. (a) Let f be the function on the halfopen interval (0, 1] deﬁned by
f(x) = 1/x. Show that f is continuous at every point of (0, 1]
but not uniformly continuous.
(b) Generalize this example to R
2
.
3. Let R be the rectangle [a, b] [c, d] and f be a bounded function that
is integrable over R.
(a) Show that f is integrable over [(a +b)/2, b] [c, d].
(b) Let N be any positive integer. Show that f is integrable over
[(a +b)/N, b] [c, d].
Exercises 4 to 6 are intended to give a proof of Theorem 2 of ¸5.2.
82 5 Double and Triple Integrals
4. Let C be the graph of a continuous function φ : [a, b] →R. Let ε > 0
be any positive number. Show that C can be placed in a ﬁnite union
of boxes B
i
= [a
i
, b
i
][c
i
, d
i
] such that C does not contain a boundary
point of ∪B
i
and such that Σ area (B
i
) ≤ ε. (Hint: Use the uniform
continuity principle presented in this section.)
5. Let R and B be rectangles and let B ⊂ R. Consider the nth regular
partition of R and let b
n
be the sum of the areas of all rectangles in
the partition that have a nonempty intersection with B. Show that
lim
n→∞
b
n
= area(B).
6. Let R be a rectangle and C ⊂ R the graph of a continuous function
φ. Suppose that f : R → R is bounded and continuous except on C.
Use Exercises 4 and 5 above and the techniques used in the proof of
Theorem 1 of this section to show that f is integrable over R.
7. (a) Use the uniform continuity principle to show that if φ : [a, b] →R
is a continuous function, then φ is bounded.
(b) Generalize part (a) to show that a given continuous function
f : [a, b] [c, d] →R is bounded.
(c) Generalize part (b) still further to show that if f : D → R is a
continuous function on a closed and bounded set D ⊂ R
n
, then
f is bounded.
Page 83
6
Integrals over Curves and Surfaces
To complete his Habilitation, Riemann had to give a lec
ture. He prepared three lectures, two on electricity and one on
geometry. Gauss had to choose one of the three for Riemann to
deliver and, against Riemann’s expectations, Gauss (his advi
sor) chose the lecture on geometry. Riemann’s lecture
¨
Uber die
Hypothesen welche der Geometrie zu Grunde liegen (On the hy
potheses that lie at the foundations of geometry), delivered on
10 June 1854, became a classic of mathematics.
There were two parts to Riemann’s lecture. In the ﬁrst part
he posed the problem of how to deﬁne an ndimensional space
and ended up giving a deﬁnition of what today we call a Rie
mannian space. Freudenthal writes:
It possesses shortest lines, now called geodesics, which re
semble ordinary straight lines. In fact, at ﬁrst approximation
in a geodesic coordinate system such a metric is ﬂat Euclidean,
in the same way that a curved surface up to higherorder terms
looks like its tangent plane. Beings living on the surface may
discover the curvature of their world and compute it at any
point as a consequence of observed deviations from Pythagoras’
theorem.
In fact the main point of this part of Riemann’s lecture was
the deﬁnition of the curvature tensor . The second part of Rie
mann’s lecture posed deep questions about the relationship of
geometry to the world we live in. He asked what the dimension
84 6 Integrals over Curves and Surfaces
of real space was and what geometry described real space. The
lecture was too far ahead of its time to be appreciated by most
scientists of that time. Monastyrsky writes:
Among Riemann’s audience, only Gauss was able to appre
ciate the depth of Riemann’s thoughts. ... The lecture exceeded
all his expectations and greatly surprised him. Returning to the
faculty meeting, he spoke with the greatest praise and rare en
thusiasm to Wilhelm Weber about the depth of the thoughts that
Riemann had presented.
It was not fully understood until sixty years later. Freuden
thal writes:
The general theory of relativity splendidly justiﬁed his work.
In the mathematical apparatus developed from Riemann’s ad
dress, Einstein found the frame to ﬁt his physical ideas, his cos
mology , and cosmogony : and the spirit of Riemann’s address
was just what physics needed: the metric structure determined
by data.
From the Riemann website
http://wwwgap.dcs.stand.ac.uk/~history/
Mathematicians/Riemann.html
Supplement 6.2A
A Challenging Example
Example. Evaluate
__
R
_
x
2
+y
2
dxdy,
where R = [0, 1] [0, 1].
Solution. This double integral is equal to the volume of the region under
the graph of the function
f(x, y) =
_
x
2
+y
2
over the rectangle R = [0, 1] [0, 1]; that is, it equals the volume of the
threedimensional region shown in Figure 6.1.1.
6.2A Challenging Example 85
x
y
z
(0,1)
(1,0)
R
Figure 6.1.1. Volume of the region under z =
_
x
2
+y
2
and over R = [0, 1] ×[0, 1].
As it stands, this integral is diﬃcult to evaluate. Since the integrand
is a simple function of r
2
= x
2
+ y
2
, we might try a change of variables
to polar coordinates. This will result in a simpliﬁcation of the integrand
but, unfortunately, not in the domain of the integration. However, the
simpliﬁcation is suﬃcient to enable us to evaluate the integral. To apply
Theorem 2 of the text with polar coordinates, refer to Figure 6.1.2.
D
1
*
π
2
π
4
r = sec θ
r
θ
D
2
*
π
2
π
4
r = csc θ
r
θ
y
x
T
T
(1,1) (0,1)
(1,0)
T
2
T
1
r
θ
Figure 6.1.2. The polarcoordinate transformation takes D
∗
1
to the triangle T
1
and
D
∗
2
to T
2
.
86 6 Integrals over Curves and Surfaces
The reader can verify that R is the image under T(r, θ) = (r cos θ, r sin θ)
of the region D
∗
= D
∗
1
∪ D
∗
2
where for D
∗
1
we have 0 ≤ θ ≤
1
4
π and
0 ≤ r ≤ sec θ; for D
∗
2
we have
1
4
π ≤ θ ≤
1
2
π and 0 ≤ r ≤ csc θ. The
transformation T sends D
∗
1
onto a triangle T
1
and D
∗
2
onto a triangle T
2
.
The transformation T is onetoone except when r = 0, and so we can
apply Theorem 2. From the symmetry of z =
_
x
2
+y
2
on R, we can see
that __
R
_
x
2
+y
2
dxdy = 2
__
T1
_
x
2
+y
2
dxdy.
Changing to polar coordinates, we obtain
__
T1
_
x
2
+y
2
dxdy =
__
D
∗
1
√
r
2
r drdθ =
__
D
∗
1
r
2
drdθ.
Next we use iterated integration to obtain
__
D
∗
1
r
2
drdθ =
_
π/4
0
__
sec θ
0
r
2
dr
_
dθ =
1
3
_
π/4
0
sec
3
θ dθ.
Consulting a table of integrals (see the back of the book) to ﬁnd
_
sec
3
x dx,
we have
_
π/4
0
sec
3
θ dθ =
_
sec θ tan θ
2
_
π/4
0
+
1
2
_
π/4
0
sec θ dθ =
√
2
2
+
1
2
_
π/4
0
sec
3
θ dθ.
Consulting the table again for
_
sec x dx, we ﬁnd
1
2
_
π/4
0
sec θ dθ =
1
2
[log [ sec θ + tan θ[]
π/4
0
=
1
2
log(1 +
√
2).
Combining these results and recalling the factor
1
3
, we obtain
__
D
∗
1
r
2
drdθ =
1
3
_√
2
2
+
1
2
log(1 +
√
2)
_
=
1
6
[
√
2 + log(1 +
√
2)].
Multiplying by 2, we obtain the answer
__
R
_
x
2
+y
2
dxdy =
1
3
[
√
2 + log(1 +
√
2)].
Instead of using the substitution T, one can alternatively divide the
original square into two triangles T
1
and T
2
as in the text, write the integral
over T
1
as a double integral (ﬁrst with respect to y, then with respect to
x); in the integral over y, substitute y = xv, then use the standard integral
number 43 at the back of the book.
6.2A Challenging Example 87
Here is another example of the use of the change of variables theorem.
Exercise 15 Evaluate
__
B
exp
_
y −x
y +x
_
dxdy
where B is the inside of the triangle with vertices at (0, 0), (0, 1) and (1, 0).
Solution. Using the change of variables u = y − x, v = y + x, we get
[∂(x, y)/∂(u, v)[ = 1/2. Thus,
__
B
exp
_
y −x
y +x
_
dxdy =
__
B
exp
_
u
v
_
1
2
dudv
=
1
2
_
1
0
_
v
−v
exp
_
u
v
_
dudv
=
1
2
_
1
0
_
v exp
_
u
v
_¸
¸
¸
v
−v
_
dv
=
1
2
_
1
0
v
_
e −e
−1
_
dv =
1
4
(e −e
−1
).
88 6 Integrals over Curves and Surfaces
Supplement 6.2B
The Gaussian Integral
The purpose of this supplement is to prove the equality of the following
two limits
lim
a→∞
__
Da
e
−(x
2
+y
2
)
dxdy = lim
a→∞
__
Ra
e
−(x
2
+y
2
)
dxdy,
which was used in the derivation of the Gaussian integral formula. In the
text, we showed that we showed that
lim
a→∞
_ _
Da
e
−(x
2
+y
2
)
dxdy
exists by directly evaluating it. Thus, it suﬃces to show that
lim
a→∞
___
Ra
e
−(x
2
+y
2
)
dxdy −
__
Da
e
−(x
2
+y
2
)
dxdy
_
equals zero. The limit equals
lim
a→∞
_ _
Ca
e
−(x
2
+y
2
)
dxdy,
where C
a
is the region between R
a
and D
a
(see Figure 6.1.3).
y
a
a
D
a
R
a
C
a
x
Figure 6.1.3. The region Ca lies between the square Ra and the circle Da.
In the region C
a
,
_
x
2
+y
2
≥ a (the radius of D
a
), so e
−(x
2
+y
2
)
≤ e
−a
2
.
Thus,
0 ≤
_ _
Ca
e
−(x
2
+y
2
)
dxdy ≤
_ _
Ca
e
−a
2
dxdy
6.2B The Gaussian Integral 89
= e
−a2
area (C
a
) = e
−a
2
(4a
2
−πa
2
) = (4 −π)a
2
e
−a
2
.
Thus it is enough to show that lim
a→∞
a
2
e
−a
2
= 0. But, by l’Hˆ opital’s rule
lim
a→∞
a
2
e
−a
2
= lim
a→∞
_
a
2
e
a
2
_
= lim
a→∞
_
2a
2ae
a
2
_
= lim
a→∞
1
e
a
2
= 0,
as required.
90 6 Integrals over Curves and Surfaces
Page 91
8
The Integral Theorems of Vector
Analysis
George Green (1793–1841), a selftaught English mathe
matician, undertook to treat static electricity and magnetism in
a thoroughly mathematical fashion. In 1828 Green published a
privately printed booklet, An essay on the Application of Math
ematical Analysis to the Theories of Electricity and Magnetism.
This was neglected until Sir William Thomson (Lord Kelvin,
1824–1907) discovered it, recognized its great value, and had
it published in the Journal f¨ ur Mathematik starting in 1850.
Green, who learned much from Poisson’s papers, also carried
over the notion of the potential function to electricity and mag
netism.
Morris Klein
Mathematical Thought from Ancient to Modern Times
Supplement for §8.3
Exact Diﬀerentials
The main theoretical point is that, to solve a diﬀerential equation in the
variables (x, y) of the form
P(x, y) +Q(x, y)
dy
dx
= 0, (8.2.1)
92 8 Integral Theorems of Vector Analysis
we proceed as follows: ﬁrst we test to see if Pdx +Qdy is exact; that is, if
the corresponding vector ﬁeld Pi +Qj is conservative. If it is, then there is
a function f(x, y) such that
P =
∂f
∂x
; Q =
∂f
∂y
.
Recall that the test for exactness is
∂P
∂y
=
∂Q
∂x
.
and that this corresponds to equality of mixed partial derivatives of f.
If the equation is exact and we ﬁnd a corresponding f, then the equa
tion f(x, y) = constant implicitly deﬁnes solutions as level sets of f. This
assertion is readily checked by implicit diﬀerentiation as follows. If x and
y are each functions of t and lie on the surface constant = f(x, y), then
diﬀerentiating both sides and using the chain rule gives
0 =
∂f
∂x
dx
dt
+
∂f
∂y
dy
dt
= P
dx
dt
+Q
dy
dt
.
In fact, this is one possible way to interpret equation (8.2.1). If y is a
function of x and we similarly diﬀerentiate the equation constant = f(x, y)
with respect to x and use the chain rule in a similar way, then we literally
get equation (8.2.1).
Example 1. Solve the following diﬀerential equation satisfying the given
conditions:
cos y sin x + sin y cos x
dy
dx
= 0, y
_
π
4
_
= 0.
Solution. The equation is exact, since
∂P
∂y
=
∂
∂y
(cos y sin x) = −sin y sin x,
and
∂Q
∂x
=
∂
∂x
(sin y cos x) = −sin y cos x.
The function f(x, y) such that P = ∂f/∂x and Q = ∂f/∂y can then be
found by integration:
_
P dx =
_
cos y sin xdx = −cos y cos x +g(y) +C
1
_
Qdy =
_
sin y cos xdy = −cos y cos x +h(x) +C
2
,
8.3 Exact Diﬀerentials 93
where g(y) is a function of y only (which will show up when integrating Q
with respect to y), h(x) is a function of x only (which will show up when
integrating P with respect to x), and C
2
and C
2
are two constants. Compare
those two results, we ﬁnd that f(x, y) = −cos y cos x = C, a constant. Since
y = 0 when x = π/4, we ﬁnd that C = −cos 0 cos(π/4) = −1/
√
2, thus,
the solution is deﬁned by cos y cos x = 1/
√
2.
Sometimes multiplying an equation by an appropriate factor can make
it exact. Such factors are called integrating factors.
Example 2. Solve the equation
x
dy
dx
= xy
2
+y
by using the integrating factor 1/y
2
.
Solution. The integrating factor 1/y
2
makes our equation exact. To check
this, ﬁrst multiply and then check for exactness: The equation is
x
y
2
dy
dx
−x −
1
y
= 0.
Now,
∂
∂y
_
−x −
1
y
_
=
1
y
2
;
and
∂
∂x
_
x
y
2
_
=
1
y
2
.
Thus, the equation is exact. To ﬁnd an antiderivative, write
_ _
−x −
1
y
_
dx = −
x
2
2
−
x
y
+g
1
(y), (8.2.2)
_
x
y
2
dy = −
x
y
+g
2
(x). (8.2.3)
Comparing (8.2.2) and (8.2.3), we see that if g
2
(x) = −x
2
/2 and g
1
(y) = C,
a constant, then the equation is solved, and so the solution is deﬁned by
the level curves of f, i.e.,
f(x, y) =
x
y
+
x
2
2
= C.
94 8 Integral Theorems of Vector Analysis
Supplement for §8.5
Green’s Functions
Next we show how vector analysis can be used to solve diﬀerential equations
by a method called potential theory or the Green’sfunction method.
The presentation will be quite informal; the reader may consult references,
such as
G.F.D. Duﬀ and D. Naylor, Diﬀerential Equations of Applied Mathemat
ics, Wiley, New York, 1966,
R. Courant and D. Hilbert, Methods of Mathematical Physics. Volumes I
and II, John Wiley & Sons Inc., New York, 1989. Wiley Classics Li
brary, Reprint of the 1962 original, A WileyInterscience Publication.
for further information.
Suppose we wish to solve Poisson’s equation
∇
2
u = ρ
for u(x, y, z), where ρ(x, y, z) is a given function. Recall that this equation
arises from Gauss’ Law if E = ∇u and also in the problem of determining
the gravitational potential from a given mass distribution.
A function G(x, y) that has the properties
G(x, y) = G(y, x) and ∇
2
G(x, y) = δ(x −y) (8.5.4)
(in this expression y is held ﬁxed, that is, which solves the diﬀerential equa
tions with ρ replaced by δ, is called Green’s function for this diﬀerential
equation. Here ρ(x−y) represents the Dirac delta function, “deﬁned” by
1
(i) δ(x −y) = 0 for x ,= y
and
(ii)
___
R
3
δ(x −y) dy = 1.
It has the following operational property that formally follows from condi
tions (i) and (ii): For any continuous function f(x),
___
R
3
f(y)δ(x −y) dy = f(x). (8.5.5)
This is sometimes called the sifting property of δ.
1
This is not a precise deﬁnition; nevertheless, it is enough here to assume that δ is
a symbolic expression with the operational property shown in equation (8.5.5). See the
references in the preceding footnote for a more careful deﬁnition of δ.
8.5 Green’s Functions 95
Theorem 1. If G(x, y) satisﬁes the diﬀerential equation ∇
2
u = ρ with
ρ replaced by δ(x −y), then
u(x) =
___
R
3
G(x, y)ρ(y) dy (8.5.6)
is a solution to ∇
2
u = ρ.
Proof. To see this, note that
∇
2
___
R
3
G(x, y)ρ(y) dy =
___
R
3
(∇
2
G(x, y)ρ(y) dy
=
___
R
3
δ(x −y)ρ(y) dy (by 8.5.4)
= ρ(x) (by 8.5.5)
The “function” ρ(x) = δ(x) represents a unit change concentrated at a
single point [see conditions (i) and (ii), above]. Thus G(x, y) represents the
potential at x due to a charge placed at y.
Green’s Function in R
3
. We claim that equation (8.5.4) is satisﬁed if
we choose
G(x, y) = −
1
4πx −y)
.
Clearly G(x, y) = G(y, x). To check the second part of equation (8.5.4),
we must verify that ∇
2
G(x, y) has the following two properties of the δ
function:
(i) ∇
2
G(x, y) = 0 for x ,= y
and
(ii)
___
R
3
∇
2
G(x, y) = 1.
We will explain the meaning of (ii) in the course of the following discus
sion. Property (i) is true because the gradient of G is
∇G(x, y) =
r
4πr
3
, (8.5.7)
where r = x − y is the vector from y to x and r = r (see Exercise 30,
¸4.4), and therefore for r ,= 0, ∇ ∇G(x, y) = 0 (as in the aforementioned
exercise). For property (ii), let B be a ball about x; by property (i),
___
R
3
∇
2
G(x, y) dy =
___
B
∇
2
G(x, y) dy.
96 8 Integral Theorems of Vector Analysis
This, in turn, equals
__
∂B
∇
2
G(x, y) n dS.
by Gauss’ Theorem. Thus, making use of (8.5.7), we get
__
∂B
∇
2
G(x, y) n dS =
__
∂B
r n
4πr
3
dS = 1,
which proves property (ii). Therefore, a solution of ∇
2
u = ρ is
u(x) =
___
R
3
−ρ(y)
4πx −y
dy. (8.5.8)
by Theorem 1.
Green’s Function in Two Dimensions. In the plane rather than in
R
3
, one can similarly show that
G(x, y) =
1
2π
log x −y, (8.5.9)
and so a solution of the equation ∇
2
u = ρ is
u(x) =
1
2π
__
R
2
ρ(y) log x −y dy.
Green’s Identities. We now turn to the problem of using Green’s func
tions to solve Poisson’s equation in a bounded region with given boundary
conditions. To do this, we need Green’s ﬁrst and second identities, which
can be obtained from the divergence theorem (see Exercise 15, ¸8.4). We
start with the identity
___
V
∇ F dV =
__
S
F n dS.
where V is a region in space, S is its boundary, and n is the outward unit
normal vector. Replacing F by f∇g, where f and g are scalar functions,
we obtain
___
V
∇f ∇g dV +
___
V
f∇
2
g dV =
__
S
f
∂g
∂n
dS. (8.5.10)
where ∂g/∂n = ∇g n. This is Green’s ﬁrst identity. If we simply per
mute f and g and subtract the result from equation (8.5.10), we obtain
Green’s second identity,
___
V
(f∇
2
g −g∇
2
f) dV =
__
S
_
f
∂g
∂n
−
∂f
∂n
_
dS, (8.5.11)
which we shall use shortly.
8.5 Green’s Functions 97
Green’s Functions in Bounded Regions. Consider Poisson’s equa
tion ∇
2
u = ρ in some region V , and the corresponding equations for Green’s
function
G(x, y) = G(y, x) and ∇
2
G(x, y) = δ(x −y).
Inserting u and G into Green’s second identity (8.5.11), we obtain
___
V
(u∇
2
G−G∇
2
u) dV =
__
S
_
u
∂G
∂n
−G
∂u
∂n
_
dS.
Choosing our integration variable to be y and using G(x, y) = G(y, x),
this becomes
___
V
[u(y)δ(x −y) −G(x, y)ρ(y)] dy =
__
S
_
u
∂G
∂n
−G
∂u
∂n
_
dS;
and by equation (8.5.5),
u(x) =
___
V
G(x, y)ρ(y) dy +
__
S
_
u
∂G
∂n
−G
∂u
∂n
_
dS. (8.5.12)
Note that for an unbounded region, this becomes identical to our previous
result, equation (8.5.6), for all of space. Equation (8.5.12) enables us to solve
for u in a bounded region where ρ = 0 by incorporating the conditions that
u must obey on S.
If ρ = 0, equation (8.5.12) reduces to
u =
__
S
_
u
∂G
∂n
−G
∂u
∂n
_
dS,
or, fully,
u(x) =
__
S
_
u(y)
∂G
∂n
(x, y) −G(x, y)
∂u
∂n
(y)
_
dS, (8.5.13)
where u appears on both sides of the equation and the integration variable
is y. The crucial point is that evaluation of the integral requires only that
we know the behavior of u on S. Commonly, either u is given on the bound
ary (for a Dirichlet problem) or ∂u/∂n is given on the boundary (for a
Neumann problem. If we know u on the boundary, we want to make
G∂u/∂n vanish on the boundary so we can evaluate the integral. Therefore
if u is given on S we must ﬁnd a G such that G(x, y) vanishes whenever y
lies on S. This is called the Dirichlet Green’s function for the region
V . Conversely, if ∂u/∂n is given on S we must ﬁnd a G such that ∂G/∂n
vanishes on S. This is the Neumann Green’s function.
Thus, a Dirichlet Green’s function G(x, y) is deﬁned for x and y in the
volume V and satisﬁes these three conditions:
98 8 Integral Theorems of Vector Analysis
(a) G(x, y) = G(y, x),
(b) ∇
2
G(x, y) = δ(x −y),
(c) G(x, y) = 0 when y lies on S, the boundary of the region V .
[Note that by condition (a), in conditions (b) and (c) the variables x and
y can be interchanged without changing the condition.]
It is interesting to note that condition (a) is actually a consequence of
conditions (b) and (c), provided (b) and (c) also hold with x and y inter
changed.
To see this, we ﬁx points y and w and use equation (8.5.11) with f(x) =
G(x, y) and g(x) = G(w, x). By condition (b),
∇
2
f(x) = δ(x −y) and ∇
2
g(x) = δ(x −w),
and by condition (c), f and g vanish when x lies on S and so the right
hand side of (8.5.11) is zero. On the left hand side, we substitute f(x) and
g(x) to give
___
V
[G(x, y)δ(x −w) −G(w, x)δ(x −y)] dx = 0
which gives
G(w, y) = G(y, w),
which is the asserted symmetry of G. This means, in eﬀect, that if (b) and
(c) hold with x and y internchanged, then it is not necessary to check con
dition (a). (This result is sometimes called the principle of reciprocity.)
Solving any particular Dirichlet or Neumann problem thus reduces to the
task of ﬁnding Laplace’s equations on all of R
2
or R
3
, namely, equations
(8.5.8) and (8.5.9).
Green’s Function for a Disk. We shall now use the twodimensional
Green’s function method to construct the Dirichlet Green’s function for the
disk of radius R (see Figure 8.5.1). This will enable us to solve ∇
2
u = 0
(or ∇
2
u = ρ) with u given on the boundary circle.
In Figure 8.5.1 we have draw the point x on the circumference because
that is where we want G to vanish. [According to the procedure above,
G(x, y) is supposed to vanish when either x or y is on C. We have chosen
x on C to begin with.] The Green’s function G(x, y) that we shall ﬁnd
will, of course, be valid for all x, y in the disk. The point y
represents
the “reﬂection” of the point y into the region outside the circle, such that
ab = R
2
. Now when x ∈ C, by the similarity of the triangles xOy and
xOy
,
r
R
=
r
b
, that is, r =
r
R
b
=
r
a
R
.
Hence, if we choose our Green’s function to be
G(x, y) =
1
2π
_
log r −log
r
a
R
_
, (8.5.14)
8.5 Green’s Functions 99
a
b
r
R
O
C
r''
x
n
y
y'
y
x
θ
θ'
Figure 8.5.1. Geometry of the construction of Green’s function for a disk.
we see that G is zero if x is on C. Since r
a/R reduces to r when y is on C, G
also vanishes when y is on C. If we can show that G satisﬁes ∇
2
G = δ(x−y)
in the circle, then we will have proved that G is indeed the Dirichlet Green’s
function. From equation (8.5.9) we know that ∇
2
(logr)/2π = δ(x −y), so
that
∇
2
G(x, y) = δ(x −y) −δ(x −y
),
but y
is always outside the circle, and so x can never be equal to y
and
δ(x −y
) is always zero. Hence,
∇
2
G(x, y) = δ(x −y)
and thus G is the Dirichlet Green’s function for the circle.
Now we shall consider the problem of solving
∇
2
u = 0
in this circle if u(R, θ) = f(θ) is the given boundary condition. By equation
(8.5.13) we have a solution
u =
_
C
_
u
∂G
∂n
−G
∂u
∂n
_
ds.
But G = 0 on C, and so we are left with the integral
u =
_
C
u
∂G
∂n
ds,
where we can replace u by f(θ), since the integral is around C. Thus the
task of solving the Dirichlet problem in the circle is reduced to ﬁnding
100 8 Integral Theorems of Vector Analysis
∂G/∂n. From equation (8.5.14) we can write
∂G
∂n
=
1
2π
_
1
r
∂r
∂n
−
1
r
r
∂n
_
.
Now
∂r
∂n
= ∇r n and ∇r =
r
r
,
where r = x −y, and so
∂r
∂n
=
r n
r
=
r cos(nr)
r
= cos(nr),
where (nr) represents the angle between n and r. Likewise,
∂r
∂n
= cos(nr
).
In triangle xyO, we have, by the cosine law, a
2
= r
2
+ R
2
− 2rRcos(nr),
and in triangle xy
0, we get b
2
= (r
)
2
+R
2
−2r
Rcos(nr
), and so
∂r
∂n
= cos(nr) =
R
2
+r
2
−a
2
2rR
and
∂r
∂n
= cos(nr
) =
R
2
+ (r
)
2
−a
2
2r
R
.
Hence
∂G
∂n
=
1
2π
_
R
2
+r
2
−a
2
2rR
−
R
2
+ (r
)
2
−a
2
2r
R
_
.
Using the relationship between r and r
when x is on C, we get
∂G
∂n
¸
¸
¸
¸
x∈C
=
1
2π
_
R
2
−a
2
Rr
2
_
.
Thus the solution can be written as
u =
1
2π
_
C
f(θ)
R
2
−a
2
Rr
2
ds.
Let us write this in a more useful form. Note that in triangle xy0, we can
write
r = [a
2
+R
2
−2aRcos(θ −θ
)]
1/2
,
where θ and θ
are the polar angles in x and y space, respectively. Second,
our solution must be valid for all y in the circle; hence the distance of y
from the origin must now become a variable, which we shall call r
. Finally,
note that ds = R dθ on C, so we can write the solution in polar coordinates
as
u(r
, θ
) =
R
2
−(r
)
2
2π
_
2π
0
f(θ) dθ
(r
)
2
+R
2
−2r
Rcos(θ −θ
)
.
8.5 Green’s Functions 101
This is know as Poisson’s formula in two dimensions.
2
As an exercise,
the reader should use this to write down the solution of ∇
2
u = ρ with u a
given function f(θ) on the boundary.
Exercises
1. (a) With notation as in Figure 8.5.1, show that the Dirichlet prob
lem for the sphere of radius R in three dimensions has Green’s
function
G(x, y) =
1
4π
_
R
ar
−
1
r
_
.
(b) Prove Poisson’s formula in three dimensions:
u(y) =
R(R
2
−a
2
)
4π
_
2π
0
_
π
0
f(θ, φ) sin φ dθdφ
(R
2
−a
2
−2Ra cos γ)
3/2
2. Let H denote the upper half space z ≥ 0. For a point x = (x, y, z)
in H, let R(x) = (x, y, −z), the reﬂection of x in the xy plane. Let
G(x, y) = −1/(4πx −y) be the Green’s function for all of R
3
.
(a) Verify that the function
˜
G deﬁned by
˜
G(x, y) = G(x, y) −G(R(x), y)
is the Green’s function for the Laplacian in H.
(b) Write down a formula for the solution u of the problem
∇
2
u = ρ in H, and u(x, y, 0) = φ(x, y).
Exercises 3 through 9 give some sample applications of vector calculus
to shock waves.
3
3. Consider the equation
u
t
+uu
x
= 0
for a function u(x, t), −∞ < x < ∞, t ≥ 0, where u
t
= ∂u/∂t and
u
x
= ∂u/∂x. Let u(x, 0) = u
0
(x) be the given value of u at t = 0.
The curves (x(s), t(s)) in the xt plane deﬁned by
˙ x = u,
˙
t = 1
2
There are several other ways of deriving this famous formula. For the method of
complex variables, see J. Marsden and M. Hoﬀman, Basic Complex Analysis, 2d ed.,
Freeman, New York, 1987, p. 195. For the method of Fourier series, see J. Marsden,
Elementary Classical Analysis, Freeman, New York, 1974, p. 466.
3
For additional details, consult A. J. Chorin and J. E. Marsden, A Mathematical
Introduction to Fluid Mechanics, 3rd ed., SpringerVerlag, New York, 1992, and P. D.
Lax, “The Formation and Decay of Shock Waves,” Am. Math. Monthly 79 (1972): 227
241. We are grateful to Joel Smoller for suggesting this series of exercises.
102 8 Integral Theorems of Vector Analysis
are called characteristic curves (the overdot denotes the derivative
with respect to s).
(a) Show that u is constant along each characteristic curve by show
ing that ˙ u = 0.
(b) Show that the slopes of the characteristic curves are given by
dt/dx = 1/u, and use it to prove that the characteristic curves
are straight lines determined by the initial data.
(c) Suppose that x
1
< x
2
and u
0
(x
1
) > u
0
(x
2
) > 0. Show that
the two characteristics through the points (x
1
, 0) and (x
2
, 0)
intersect at a point P = (¯ x,
¯
t) with
¯
t > 0. Show that this together
with the result in part (a) implies that the solution cannot be
continuous at P (see Figure 8.5.2).
t
x
x
1
x
2
P
Figure 8.5.2. Characteristics of the equation ut +uux = 0.
(d) Calculate
¯
t.
4. Repeat Exercise 3 for the equation
u
t
+f(u)
x
= 0, (8.5.15)
where f
> 0 and f
(u
0
(x
2
)) > 0. The characteristics are now de
ﬁned by ˙ x = f
(u),
˙
t = 1. We call equation (8.5.15) and equation in
divergence form. (This exercise shows that a continuous solution
is generally impossible—irrespective of the smoothness of f!)
5. (Weak solutions) Since equations of the form in Exercise 4 arise in
many physical applications [gas dynamics, magnetohydrodynamics,
nonlinear optics (lasers)] and because it would be nice for a solution to
exist for all time (t), it is desirable to make sense out of the equation
by reinterpreting it when discontinuities develop. To this end, let
φ = φ(x, t) be a C
1
function. Let D be a rectangle in the xt plane
determined by −a ≤ x ≤ a and 0 ≤ t ≤ T, such that φ(x, t) = 0 for
x = ±a, x = T, and for all (x, t) in the upper half plane outside D.
Let u be a “genuine” solution of equation (8.5.15).
8.5 Green’s Functions 103
(a) Show that
__
t≥0
[uφ
t
+f(u)φ
x
] dxdt +
_
t=0
u
0
(x)φ(x, 0) dx = 0. (8.5.16)
(HINT: Start with
__
D
[u
t
+f(u)
x
]φ dxdx = 0.)
Thus, if u is a smooth solution, then equation (8.5.16) holds
for all φ as above. We call the function u a weak solution of
equation (8.5.15) if equation (8.5.16) holds for all such φ.
(b) Show that if u is a weak solution that is C
1
in an open set Ω in
the upper half of the xt plane, then u is a genuine solution of
equation (8.5.15) in Ω.
6. (The jump condition, which is also known in gas dynamics as the
RankineHugoniot condition.) The deﬁnition of a weak solution
given in Exercise 5 clearly allows discontinuous solutions. However,
the reader shall now determine that not every type of discontinuity is
admissible, for there is a connection between the discontinuity curve
and the values of the solution on both sides of the discontinuity.
Let u be a (weak) solution of equation (8.5.15) and suppose Γ is a
smooth curve in the xt plane such that u “jumps” across a curve Γ;
that is, u is of a class C
1
except for jump discontinuity across Γ. We
call that Γ a shock wave. Choose a point P ∈ Γ and construct, near
P, a “rectangle” D = D
1
∪ D
2
, as shown in Figure 8.5.3. Choose φ
to vanish on D and outside D.
R
t
x
P
Q
D
1
D
2
Γ
Figure 8.5.3. The solution u jumps in value from u
1
to u
2
across Γ.
(a) Show that
__
D
[uφ
t
+f(u)φ
x
] dxdt = 0
104 8 Integral Theorems of Vector Analysis
and
__
D1
[uφ
t
+f(u)φ
x
] dxdt =
__
D1
[(uφ)
t
+ (f(u)φ)
x
] dxdt.
(b) Suppose that u jumps in value from u
1
to u
2
across Γ so that
when (x, t) approaches a point (x
0
, t
0
) on Γ from ∂D
i
, u(x, t)
approaches the value u
i
(x
0
, t
0
). Show that
0 =
_
∂D1
φ[−u dx +f(u) dt] +
_
∂D2
φ[−u dx +f(u) dt]
and deduce that
0 =
_
Γ
φ([−u] dx + [f(u)] dt)
where [α(u)] = α(u
2
) −α(u
1
) denotes the jump in the quantity
α(u) across Γ.
(c) If the curve Γ deﬁnes x implicitly as a function of t and ∂D
intersects Γ at Q = (x(t
1
), t
1
) and R = (x(t
2
), t
2
), show that
0 =
_
R
Q
φ([−u] dx + [f(u)] dt) =
_
t2
t1
φ
_
[−u]
dx
dt
+ [f(u)]
_
dt.
(d) Show that at the point P on Γ,
[u] s = [f(u)], (8.5.17)
where s = dx/dt at P. The number s is called the speed of the
discontinuity. Equation (8.5.17) is called the jump condition;
it is the relationship that any discontinuous solution will satisfy.
7. (Loss of uniqueness) One drawback of accepting weak solutions is
loss of uniqueness. (In gas dynamics, some mathematical solutions
are extraneous and rejected on physical grounds. For example, dis
continuous solutions of rarefaction shock waves are rejected because
they indicate that entropy decreases across the discontinuity.)
Consider the equation
u
t
+
_
u
2
2
_
x
= 0, with initial data u(x, 0) =
_
−1, x ≥ 0
1, x < 0.
8.5 Green’s Functions 105
Show that for every α ≥ 1, u
α
is a weak solution, where u
α
is deﬁned
by
u
α
(x, t) =
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
1, x ≤
1 −α
2
t
−α,
1 −α
2
t ≤ x ≤ 0
α, 0 ≤ x ≤
α −1
2
t
−1,
α −1
2
t < x.
(It can be shown that if f
> 0, uniqueness can be recovered by
imposing an additional constraint on the solutions. Thus, there is a
unique solution satisfying the “entropy” condition
u(x +a, t) −u(x, t)
a
≤
E
t
for some E > 0 and all a ,= 0. Hence for ﬁxed t, u(x, t) can only
“jump down” as x increases. In our example, this holds only for the
solution with α = 1.)
8. (The solution of equation (8.5.15) depends on the particular diver
gence form used.) The equation u
t
+ uu
x
= 0 can be written in the
two divergence forms
u
t
+ (
1
2
u
2
)
x
= 0. (i)
(
1
2
u
2
)
t
+ (
1
3
u
3
)
x
= 0. (ii)
Show that a weak solution of equation (i) need not be a weak solution
of equation (ii). [HINT: The equations have diﬀerent jump conditions:
In equation (i), s =
1
2
(u
2
+ u
1
), while in equation (ii), s =
2
3
(u
2
2
+
u
1
u
2
+u
2
1
)/(u
2
+u
1
).]
9. ( Noninvariance of weak solutions under nonlinear transformation)
Consider equation (8.5.15) where f
> 0.
(a) Show that the transformation v = f
(u) takes this equation into
v
t
+vv
x
= 0. (8.5.18)
(b) Show that the above transformation does not necessarily map
discontinuous solutions of equation (8.5.15) into discontinuous
solutions of equation (8.5.18). (HINT: Check the jump con
ditions; for equation (8.5.18), s[v] =
1
2
[v
2
] implies s[f
(u)] =
1
2
[f
(u)
2
]; for equation (8.5.15), s[u] = [f(u)].)
106 8 Integral Theorems of Vector Analysis
10. Requires a knowledge of complex numbers. Show that Poisson’s for
mula in two dimensions may be written as
u(r
, θ
) =
1
2π
_
2π
0
u(re
iθ
)
r
2
−[z
[
2
[re
iθ
−z
[
2
dθ,
where z
= r
e
iθ
.
Page 107
Selected Answers and Solutions for the
Internet Supplement
§2.7 Some Technical Diﬀerentiation Theorems.
1. Df(x, y, z) =
_
_
e
x
0 0
0 −sin y 0
0 0 cos z
_
_
;
Df is a diagonal matrix if each component function f
i
depends only
on x
i
.
3. (a) Let A = B = C = R with f(x) = 0 and g(x) = 0 if x ,= 0 and
g(0) = 1. Then w = 0 and g(f(x)) −1 for all x.
(b) If > 0, let δ
1
and δ
2
be small enough that D
δ1
(y
0
) ⊂ B and
g(y) − w < whenever y ∈ B and 0 < y − y
0
 < δ
2
. Since
g(y
0
) = w, the 0 < y −y
0
 restriction may be dropped. Let δ
be small enough that f(x)−y
0
 < min (δ
1
, δ
2
) whenever x ∈ A
and 0 < x − x
0
 < δ. Then for such x, f(x) − y
0
 < δ
1
, so
that f(x) ∈ B and g(f(x)) is deﬁned. Also, f(x) − y
0
 < δ
2
,
and so g(f(x)) −w < .
5. Let x = (x
1
, . . . , x
n
) and ﬁx an index k. Then
f(x) = a
kk
x
2
k
+
n
i=1=k
a
kj
x
k
x
j
+
n
j=1=k
a
kj
x
k
x
j
+ [terms not involving x
k
]
108 Answers for the Internet Supplement
and therefore
∂f
∂x
k
= 2a
kk
x
k
+
i=k
a
ki
x
i
+
i=k
a
ki
x
i
= 2
n
j=1
a
kj
x
j
= (2Ax)
k
.
Since the kth components agree for each k, ∇f(x) = 2Ax
k
.
7. The matrix T of partial derivatives is formed by placing Dg(x
0
) and
Dh(y
0
) next to each other in a matrix so that T(x
0
, y
0
)(x − x
0
, y −
y
0
) = Dg(x
0
)(x−x
0
)+Dh(y
0
)(y−y
0
). Now use the triangle inequality
and the fact that (x−x
0
, y −y
0
) is larger than [x−x
0
[ and [y −y
0
[
to show that f(x, y) − f(x
0
, y
0
) − T(x
0
, y
0
)(x − x
0
, y − y
0
)/(x −
x
0
, y −y
0
) goes to 0.
9. Use the limit theorems and the fact that the function g(x) =
_
[x[ is
continuous. (Prove the last statement.)
11. For continuity at (0, 0), use the fact that
¸
¸
¸
¸
xy
(x
2
+y
2
)
1/2
¸
¸
¸
¸
≤
[xy[
(x
2
)
1/2
= [y[
or that [xy[ ≤ (x
2
+y
2
)/2.
13. 0, see Exercise 11.
15. Let x take the role of x
0
and x +h that of x in the deﬁnition.
17. The vector a takes the place of x
0
in the deﬁnition of a limit or in
Theorem 6. In either case, the limit depends only on values of f(x) for
x near x
0
, not for x = x
0
. Therefor f(x) = g(x) for x ,= a certainly
suﬃces to make the limits equal.
19. (a) lim
x→0
(f
1
+f
2
)(x)/x = lim
x→0
f
1
(x)/x+lim
x→0
f
1
(x)/x =
0.
(b) Let > 0. Since f is o(x), there is a δ > 0 such that f(x)/x <
/c whenever 0 < x < δ. Then (gf)(x)/x ≤ g(x)f(x)/x <
, so lim
x→0
(gf)(x)/x = 0.
(c) lim
x→0
f(x)/[x[ = lim
x→0
[x[ = 0, so that f(x) is o(x). But
lim
x→0
g(x)/[x[ does not exist, since g(x)/[x[ = ±1 (as x is
positive or negative). Therefore g(x) is not o(x).
Answers for the Internet Supplement 109
Solutions to Selected Exercises in §2.7
1. Here, we have f(x, y, z) = (e
x
, cos y, sin z) = (f
1
, f
2
, f
3
), so
Df(x, y, z) =
_
_
∂f
1
/∂x ∂f
1
/∂y ∂f
1
/∂z
∂f
2
/∂x ∂f
2
/∂y ∂f
2
/∂z
∂f
3
/∂x ∂f
3
/∂y ∂f
3
/∂z
_
_
=
_
_
e
x
0 0
0 −sin y 0
0 0 cos z
_
_
Df is a diagonal matrix when f
1
depends only on the ﬁrst variable,
f
2
depends only on the second, and so on. Thus, Df is diagonal if f
n
is a function of the n
th
variable only.
4. A uniformly continuous function is not only continuous at all points
in the domain, but more importantly, for given ε > 0, we can ﬁnd one
δ for all x
0
. This is diﬀerent from continuity in that for continuity, we
may ﬁnd a δ which will work for a particular x
0
. A continuous func
tion does not always have to be uniformly continuous (for examples:
f(x) = 1/x
2
on R or g(x) = 1/x on (0, 1].)
(a) T : R
n
→ R
m
is linear implies that Tx ≤ Mx. Given
ε > 0, x and y ∈ R
n
, we have T(x − y) ≤ Mx − y from
exercise 2(a). Since T is linear, T(x−y) ≤ T(x)−T(y). Let
δ = ε/M; then for 0 ≤ x − y ≤ δ, we have T(x) − T(y) ≤
Mδ = ε. Note that this proof works because δ and ε do not
depend on a particular choice of a point in R
n
.
(b) Given ε > 0, we want a δ > 0 such that 0 < [x−x
0
[ < δ implies
[1/x
2
−1/x
0
2
[ < ε for x and x
0
in (0, 1]. We calculate:
¸
¸
¸
¸
1
x
2
−
1
x
0
2
¸
¸
¸
¸
=
¸
¸
x
0
2
−x
2
¸
¸
x
2
x
0
2
=
[x −x
0
[[x +x
0
[
x
2
x
0
2
= [x −x
0
[
¸
¸
¸
¸
1
xx
0
2
+
1
x
0
x
2
¸
¸
¸
¸
< [x −x
0
[
2
x
0
3
< ε
(if x
0
is the smaller of x and x
0
), i.e.,
¸
¸
¸
¸
1
x
2
−
1
x
0
2
¸
¸
¸
¸
≤ [x −x
0
[
2
¦min(x, x
0
)¦
3
.
Note that min(x, x
0
) > 0. Let δ < (ε/2) ¦min(x, x
0
)¦
3
, then
[x −x
0
[ < δ implies
¸
¸
¸
¸
1
x
2
−
1
x
0
2
¸
¸
¸
¸
< δ
2
x
0
3
<
ε
2
x
0
3
2
x
0
3
= .
110 Answers for the Internet Supplement
We have only shown that f(x) = 1/x
2
is continuous. It is not
uniformly continuous since
¸
¸
¸
¸
1
x
2
−
1
x
0
2
¸
¸
¸
¸
=
¸
¸
x
0
2
−x
2
¸
¸
x
2
x
0
2
=
[x −x
0
[[x +x
0
[
x
2
x
0
2
,
so for ﬁxed ε, any δ approaches ∞ as x
0
goes to 0.
7. We have f(x, y) = g(x) + h(y). What we want to prove is that as
matrices,
Df(x, y) = (Dg(x), Dh(y)).
Since g and h are diﬀerentiable at x
0
and y
0
, respectively, we have
lim
x→x0
[g(x) −g(x
0
) −Dg(x −x
0
)[
[x −x
0
[
= 0
and
lim
y→y0
[h(y) −h(y
0
) −Dh(y −y
0
)[
[y −y
0
[
= 0
Now, use the triangle inequality:
[g(x) −g(x
0
) +h(y) −h(y
0
) −(Dg(x −x
0
) +Dh(y −y
0
))[
≤ [g(x) −g(x
0
) −Dg(x −x
0
)[ +[h(y) −h(y
0
) −Dh(y −y
0
)[.
Since (x, y) −(x
0
, y
0
) ≥ x−x
0
 and (x, y) −(x
0
, y
0
) ≥ y −y
0
,
the sum of the two limits is greater than
lim
(x,y)→(x0,y0)
[g(x) +h(y) −(g(x
0
) +h(y
0
)) −(Dg(x −x
0
) +Dh(y −y
0
))[
(x, y) −(x
0
, y
0
)
.
Hence this limit goes to 0, and this satisﬁes the deﬁnition for diﬀer
entiability of f at (x
0
, y
0
).
11. Given ε, note that
¸
¸
¸
¸
¸
xy
(x
2
+y
2
)
1/2
−0
¸
¸
¸
¸
¸
≤
¸
¸
¸
xy
x
¸
¸
¸ = [y[,
and we also know that [y[ ≤
_
x
2
+y
2
. Let δ = ε; then (x, y) −
(0, 0) =
_
x
2
+y
2
< δ implies that
¸
¸
¸
¸
¸
xy
(x
2
+y
2
)
1/2
−0
¸
¸
¸
¸
¸
< [y[ < ε.
Therefore, lim
(x,y)→(0,0)
f(x, y) = 0 and f is continuous.
Answers for the Internet Supplement 111
16. (a) Suppose y is a boundary point of the open set A ⊂ R
n
. Then
every ball centered at y contains points in A and points not in
A. Therefore, the intersection of A and the ball D
ε
(y) is not
empty. Hence
(i) y is not in A since points of A have neighborhoods contained
in A, and
(ii) y is the limit of the sequence from A; choose ε = 1/n,
n = 1, 2, 3, . . . , then D
(1/n)
(y) ∩ A ,= ∅ for n = 1, 2, 3, . . . .
Pick x
n
to be an element of D
(1/n)
(y) ∩A. Then x
n
is in A and
y−x
n
 ≤ 1/n, which goes to zero as n goes to inﬁnity. Suppose
y is not in A, but there is a sequence ¦x
n
¦ in A converging to
y. Let ε > 0. The y is in the intersection of D
ε
(y) and the set
of all points not in A, so the righthand side is not empty. On
the other hand, there exists an N such that n ≥ N implies that
x
n
−y < , i.e., n ≥ N implies that x
n
is in D
ε
(y). x
N
is in A,
so x
N
is in D
ε
(y) ∩A. Hence, the righthand side is a boundary
point of A.
(b) Suppose lim
x→y
f(x) = b, i.e., for every ε > 0, there is a δ > 0
such that for x in A, x−y < δ implies that f(x)−f(y) < .
Let ¦x
n
¦ be a sequence in A converging to y. Fix ε > 0. Choose
δ > 0 as above. Choose N so that n ≥ N implies that x
n
−
y < δ. Then n ≥ N implies f(x
n
) − b < , so the sequence
[f(x
n
)[ converges to b. (We need y to be on the boundary of A
to guarantee the existence of the sequence ¦x
n
¦.)
To go the other way, suppose it is not the case that lim
x→y
f(x) =
b. Then there exist ε > 0 such that for every δ > 0, there exists
an x in A with x−y < δ, but f(x) −b > ε. Choose x
n
in A
such that x
n
−y < 1/n, but f(x
n
)−b > for n = 1, 2, 3, . . .
(corresponding to δ = 1/n). Then x
n
converges to y (as in part
(a)), but f(x) does not converge to b. (Note: We have proved
the “contrapositive” of the theorem. We negate the statement of
the hypothesis (note the changes in “there exists,” “for every,”
and the inequality signs), and prove the negation of the result
we want to arrive at. The diﬀerence between this method and
the method of proving by contradiction is that we do not negate
the hypothesis.)
(c) Let U be open in R
m
with x in U. The proof follows from part
(b). Speciﬁcally,
lim
xn→x
f(x
n
) = f(x) implies f(x
n
) → f(x)
for any sequence x
n
converging to x in U. By part (b), f is
continuous at x.
112 Answers for the Internet Supplement
§3.4A: Second Derivative Test: Constrained Extrema
1. It is the usual second derivative test in onevariable calculus.
§4.1A: Equilibria in Mechanics
1. (−
1
4
, −
1
4
)
3. (2, 1) is an unstable equilibrium.
5. Stable equilibrium point (2 +m
2
g
2
)
−1/2
(−1, −1, −mg)
Solution to Exercise 3. By Theorem 14, the critical points of the
potential V are the equilibrium points. These points are where the gradient
vanishes. In fact,
∇V (x, y) = (2x + 4y −8)i + (4x −2y −6)j = 0
only if 2x +4y −8 = 0 and 4x −2y −6 = 0. Simplifying we get the system
of equations
x + 2y = 4
2x −y = 3
whose solution is x = 2, y = 1. This critical point (2, 1) will be stable if it is
a strict local minimum for V . Using the second test (Theorem 6), we have
∂
2
V
∂x
2
(2, 1) = 2,
∂
2
V
∂y
2
(2, 1) = −2,
∂
2
V
∂x∂y
(2, 1) = 4.
Therefore the discriminant
D =
_
∂
2
V
∂x
2
__
∂
2
V
∂y
2
_
−
_
∂
2
V
∂x∂y
_
2
= −20 < 0.
The second derivative test thus tells us that the point is a saddle; in par
ticular, we cannot conclude that the point is stable. (In fact, it is unstable,
but this has not been discussed carefully, so it is best to just say the sta
bility test fails).
§4.1B Rotations and the Sunshine Formula
1. (a) m
0
= (1/
√
6)(i +j + 2k),
n
0
= (1/2
√
3)(i +j −k)
(b) r = [cos(πt/12)/2
√
2+sin(πt/12)/4]i+[−1/2
√
2+cos(πt/12)/2
√
2+
sin(πt/12)/4]j + [−1/2
√
2 + cos(πt/12)/
√
2 −sin(πt/12)/4]k
Answers for the Internet Supplement 113
(c) (x, y, z) = (−1/2
√
2)(i + 2j + 3k) + (−π/48)(i +j +k)(t −12)
3. T
d
would be longer.
5. The “exact” formula is −tan l sin α = cos(2πt/T
d
)[tan(2πt/T
y
) tan(2πt/T
d
)−
cos α].
7. A = 9.4
◦
9. The equator would receive approximately six times as much solar
energy as Paris.
§4.4 Flows and the Geometry of the Divergence
1. If x = (x
1
, x
2
, x
3
), φ(x, t) = (φ
1
, φ
2
, φ
3
), and f = f(x
1
, x
2
, x
3
, t), then
by the chain rule,
d
dt
(f(φ(x, t), t)) =
∂f
∂t
(x, t) +
3
i=1
∂f
∂x
i
(φ(x, t), t)
∂φ
i
∂t
(x, t)
=
∂f
∂t
(x, t) + [∇f(φ(x, t), t)] [F(φ(x, t))].
3. By the product rule,
d
dt
v w =
dv
dt
w+v
dw
dt
.
Substitute equation (2) into this. For the last equality, use the identity
A
T
v w = v Aw.
5. By the choice of axes,
w =
1
2
(∇F)(0) = ωk.
From the text,
v = −ωyi +ωxj
and therefore
Dv(0) =
_
_
b0 −ω 0
ω 0 0
0 0 0
_
_
.
114 Answers for the Internet Supplement
On the other hand, by the deﬁnitions of W and ∇F,
W =
_
_
bw
1
1 w
1
2 w
1
3
w
2
1 w
2
2 w
2
3
w
3
1 w
3
2 w
3
3
_
_
=
_
¸
¸
¸
¸
¸
¸
_
0
1
2
_
∂F
1
∂y
−
∂F
2
∂x
_
1
2
_
∂F
1
∂z
−
∂F
3
∂x
_
1
2
_
∂F
2
∂x
−
∂F
1
∂y
_
0
1
2
_
∂F
2
∂z
−
∂F
3
∂y
_
1
2
_
∂F
3
∂x
−
∂F
1
∂y
_
1
2
_
∂F
3
∂y
−
∂F
2
∂z
_
0
_
¸
¸
¸
¸
¸
¸
_
=
1
2
_
_
0 −(∇F)
z
(∇F)
y
(∇F)
z
0 −(∇F)
x
−(∇F)
y
(∇F)
x
0
_
_
Our choice of coordinate axes gives
W =
_
_
0 −ω 0
ω 0 0
0 0 0
_
_
.
To interpret the result, we note that the vector ﬁeld v represents
rotation around a ﬁxed axis w. The ﬂow ψ(x, t) of v rotates points
in this ﬁeld, and, for ﬁxed t, its derivative D
x
ψ(x, t) rotates vectors
as well. Let Y be an arbitrary vector and set Y(t) = D
x
ψ(x, t)Y. As
t increases or decreases, Y(t) rotates around w and
dY
dt
¸
¸
¸
¸
t=0
= D
x
v(0)Y.
This gives the rate of change of Y as it is transported (rotated) by
D
x
ψ. By Exercise 3, the rate of change of any vector x at the origin
under transport by the derivative of the ﬂow φ(x, t) of F is given by
dx
dt
¸
¸
¸
¸
t=0
= D
x
F(0)x = (S +W)x.
Thus this rate of change of x has two components: the deformation
matrix, which aﬀects inner products, and the W matrix. Thus the W
matrix is precisely the rate of change of vectors as they undergo an
inﬁnitesimal rotation around the axis (curl F)(0) = (∇ F)(0) by
the mapping D
x
ψ(x, t).
[The deformation matrix S incorporates all the length and angle
changes caused by the ﬂow. In particular, volume changes are con
tained in S. In fact, the trace of S is the divergence; tr S = div F(x).
Answers for the Internet Supplement 115
(The trace of a matrix is the sum of its diagonal entries). The trace
free part of S,
S
= S −
1
3
(trS
I
)
where I is the 3 3 identity, is called the shear.]
7. The line x + λv is carried to the curve λ → φ(x + λv, t) after time
t, which for λ small, is approximated by its tangent line, namely,
λ → φ(x, t) +D
x
φ(x, t) λv.
§5.6 Some Technical Integration Theorems
1. If a ,= b, let = [a −b[/2.
3. Let e = 2d−c, so that d = (c+e)/2. Consider the vertical “doubling”
of R deﬁned by Q = R∪R
1
, where R
1
= [a, b] [d, e]. If f is extended
to Q by letting f be 0 on the added part, then f is integrable over Q
by additivity. The nth regular partition of [(a+b)/2, b] [c, d] is part
of the 2nth regular partition of Q. For large n, the Riemann sums
for that 2nth partition cannot vary by more than as we change the
points from the subrectangles, in particular if we change only those in
[(a+b)/2, b] [c, d]. These changes correspond to the possible changes
for the Riemann sums for the nth partition of [(a + b)/2, b] [c, d].
The argument for part (b) is similar.
5. Let R = [a, b] [c, d] and B = [e, f] [g, h]. Since the rectangles of
a partition of R intersect only along their edges, their areas can be
added, and b
n
is the area of the union of all subrectangles of the nth
regular partition of R that intersect B. Since B is contained in this
union, area (B) ≤ b
n
. On the other hand, if (x, y) is in the union,
then
e −(b −a)/n ≤ x ≤ f + (b −a)/n
and
g −(d −c)/n ≤ y ≤ h + (d −c)/n.
This leads to
b
n
≤ area(B)+2[(b−a)(h−g)+(d−c)(f −e)]/n+4(a−b)(d−c)/n
2
.
Letting n → ∞ and combining the inequalities proves the assertion.
7. (a) The strategy is to go from point to point within [a, b] by short
steps, adding up the changes as you go. Given > 0, φ is uni
formly continuous and therefore there exists a δ > 0 such that
[φ(x) − φ(y)[ ≤ whenever [x − y[ < δ. Let x ∈ [a, b] and in
troduce intermediate points a = x
0
< x
1
< . . . < x
n−1
< x
n
=
116 Answers for the Internet Supplement
x with x
i+1
− x
i
< δ. This can be done with no more than
[(b −a)/δ] + 1 segments. By the triangle inequality,
[φ(x) −φ(a)[ ≤
n
i=1
[φ(x
i
) −φ(x
i−1
)[ ≤
_
b −a
δ + 1
_
.
Thus [φ(x)[ ≤ [φ(a)[ + [(b −a)/(δ + 1)] for every x in [a, b].
(b) Use an argument like that for part (a), moving by short steps
within the rectangle [a, b] [c, d].
(c) This is trickier, since D may be composed of many disconnected
pieces so that the short steps cannot be taken within D. Never
theless, given , there is a δ such that[f(x) −f(y)[ ≤ whenever
x and y are in D and x−y < δ, by the uniform boundedness
principle. Since D is bounded, we may ﬁnd a large “cube” R with
sides of length L such that D ⊂ R. Partition R into subcubes by
dividing each edge into m parts. The diagonal of each subcube
has length
√
nL/m. If we take m >
√
nL/δ, any two points in the
same subcube are less then δ apart, and there are m
n
subcubes.
If R
1
, . . . , R
n
are those that intersect D, choose x
i
∈ D ∩ R
i
.
For any x ∈ D, we have [f(x) < + map([f(x
1
)[, . . . , [f(x)[).
Solution to Exercise 2(a).
Exercise 2(a). Let f be the function on the half open interval (0, 1]
deﬁned by f(x) = 1/x. Show that f is continuous at every point of (0, 1]
but not uniformly continuous.
Solution. Let x
0
∈ (0, 1]. Given > 0, we must ﬁnd a δ > 0 such that
whenever
0 < [x −x
0
[ < δ, x(0, 1], then
¸
¸
¸
¸
1
x
−
1
x
0
¸
¸
¸
¸
< .
Notice that ¸
¸
¸
¸
1
x
−
1
x
0
¸
¸
¸
¸
=
[x −x
0
[
[xx
0
[
.
If [x −x
0
[ < x
0
/2, then x > x
0
/2 > 0 and
[x −x
0
[
[xx
0
[
<
2[x −x
0
[
x
2
0
.
If [x −x
0
[ is also less than x
2
0
/2, then 2[x −x
0
[/x
2
0
< . Thus by picking
δ < min
_
x
0
/2, (/2) x
2
0
_
one has
¸
¸
¸
¸
1
x
−
1
x
0
¸
¸
¸
¸
< .
Answers for the Internet Supplement 117
Thus, f is continuous at x
0
. To show that f is not uniformly continuous, we
suppose it were and derive a contradiction. Let = 1/2. If f were uniformly
continuous there exists a δ > 0 such that
¸
¸
¸
¸
1
x
1
−
1
x
2
¸
¸
¸
¸
<
1
2
whenever [x
2
−x
1
[ < δ. Let N > 2δ. There for all integers n, m > N
¸
¸
¸
¸
1
n
−
1
m
¸
¸
¸
¸
<
1
N
+
1
N
=
2
N
< δ.
Set x
1
= 1/n and x
2
= 1/(n + 1). Then
¸
¸
¸
¸
1
x
1
−
1
x
2
¸
¸
¸
¸
=
1
1
n+1
−
1
1
n
= 1.
But 1 is not less than 1/2 a contradiction.
§8.5 Green’s Functions
1. (a) If x ∈ S, then r
a/R = r, so G = 0. In general, r = x−y and
r
= x −y, and therefore
G =
1
4π
_
R
a
1
x −y

−
1
x −y
_
and
∇
x
G =
1
4π
_
R
a
y
−x
x −y

3
−
y −x
x −y
3
_
and ∇
2
x
G = 0 when x ,= y, just as in the analysis of equation
(15)(x ,= y
, since x is inside and y
is outside the sphere).
Theorem 10 gives ∇
2
G = δ(x − y) − (R/a)δ(x − y
), but the
second term is always 0, since x is never y
. Therefore ∇
2
G =
δ(x −y) for x and y in the sphere.
(b) If x is on the surface of S, then n = x/R is the outward unit
normal, and
∂G
∂n
=
1
4π
_
R
a
∇
_
1
r
_
n −∇
_
1
r
_
n
_
=
1
4π
_
R
a
y
−x
y
−x
3
−
y −x
x −y
3
_
n.
If γ is the angle between x and y, then x−y
2
= r
2
= R
2
+a
2
−
2aRcos γ and x−y

2
= r
2
= R
2
+b
2
−2bRcos γ = (R
2
/a
2
)r
2
.
Then
∂G
∂n
=
1
4πr
3
_
R
a
y
−x
(R/a)
3
−(y −x)
_
n.
118 Answers for the Internet Supplement
But x n = R and n = x/R, and so this becomes
∂G
∂n
=
1
4π
3
_
a
2
R
3
y
x −
a
2
R
−
y x
R
+R
_
=
1
4π
3
R
_
a
2
R
2
y
Rcos γ −yRcos γ +R
2
−a
2
_
=
R
2
−a
2
4πR
1
r
3
since y
 = R
2
/a.
Integrating over the surface of the sphere,
u(y) =
__
S
f
∂G
∂n
dS
=
_
2π
0
_
π
0
_
f(θ, φ)
R
2
−a
2
4πR
1
r
3
R
2
sin φ
_
dφdθ
=
R(R
2
−a
2
)
4π
_
2π
0
_
π
0
f(θ, φ) sin φ dφdθ
(R
2
+a
2
−2aRcos γ)
3/2
.
2. (a) According to equations (12) of this section, we need to show that
˜
G(x, y) =
˜
G(y, x) (1)
and
∇
2
G = δ(x, y). (2)
Deﬁne r = x −y and r
= R(x) −y. To show (1),
˜
G(y, x) = G(y, x)−G(y, R(x)) = −
1
4
πy−x+
1
4
πy−R(x).
It is obvious that r = −r, so
˜
G(y, x) =
˜
G(x, y). To show (2),
we know ∇G(x, y) = r/4πr
3
. Thus ∇G(R(x), y) = r
/4π(r
)
3
.
Then ∇
2
˜
G = δ(x, y) −δ(R(x), y), and
___
R
3
∇
2
˜
Gdy =
___
R
3
[δ(x, y) −δ(R(x), y)] dy.
If x = y, then R(x) ,= y, and the above integral becomes
___
R
3
∇
2
˜
Gdy =
___
R
3
δ(x, y) dy = 1;
If R(x) = y, then x ,= y. Make a change of variable: Let u = x,
v = y, w = −z, so dy = dxdy dz = −dudv dw. Now the integral
Answers for the Internet Supplement 119
becomes
___
R
3
∇
2
˜
Gdy = −
___
R
3
δ(R(x), y) dy
= −
___
R
3
δ(R(x), u) (−du)
=
___
R
3
δ(R(x), u) du = 1,
where u = (u, v, w).
(b) Simply ”stick” our Green’s function into an integral:
u(x) =
___
R
3
˜
G(x, y)ρ(y) dy
=
___
R
3
G(x, y)ρ(y) −G(R(x), y)ρ(y) dy.
3. (a) Use the Chain rule: u
t
+u
xx
= u
t
˙
t +u
x
˙ x = ˙ u = 0.
(b) Use dt/ds = 1 and dx/ds = u. The slope is equal to
t(s)
x(s)
=
dt
dx
=
1
u
.
From part (a), u is a constant, therefore 1/u is also a constant.
Hence the characteristic curves are straight lines.
(c) Two characteristics through (x
1
, 0) and (x
2
, 0) have equations
t = [1/u
0
(x
1
)](x −x
1
) and t = [1/u
0
(x
2
)](x −x
2
),
respectively. The intersection is
[1/u
0
(x
1
)](x −x
1
) = [1/u
0
(x
2
)](x −x
2
).
Simplify and get
x(1/u
0
(x
1
) −1/u
0
(x
2
)) = x
1
/u
0
(x
1
) −x
2
/u
0
(x
2
).
Solve for x:
x =
x1
u0(x1)
−
x2
u0(x2)
1
u0(x1)
−
1
u0(x2)
=
x
1
u
0
(x
2
) −x
2
u
0
(x
1
)
u
0
(x
2
) −u
0
(x
1
)
> 0.
(d) Plug in:
¯
t =
1
u
0
(x
1
)
_
x
1
u
0
(x
2
) −x
2
u
0
(x
1
)
u
0
(x
2
) −u
0
(x
1
)
−x
1
_
= −
x
2
+x
1
u
0
(x
2
) −u
0
(x
1
)
120 Answers for the Internet Supplement
4. (a) ˙ u = (d/ds)[u(x(s), t(s))] = u
x
˙ x +u
t
˙
t = u
x
f
(u) +u
t
= 0.
(b) If the characteristic curve u(x, t) = c (by part (a)), deﬁne t
implicitly as a function of x; then u
x
+ u
t
(dt/dx) = 0. But also
u
t
+ f(u)
x
= 0; that is, u
t
+ f
(u)u
x
= 0. These two equations
together give dt/dx = 1/f
(u) = 1/f
(c). Therefor the curve is
a straight line with slope 1/f
(c).
(c) If x
1
< x
2
, u
0
(x
1
) > u
0
(x
2
) > 0, and f
(u
0
(x
2
)), then f
(u
0
(x
1
)) >
f
(u
0
(x
2
)) > 0, since f
> 0. The characteristic through (x
1
, 0)
has slope 1/f
(u
0
(x
1
)), which is less than 1/f
(u
0
(x
2
)), (that of
the characteristic through (x
2
, 0). So these lines must cross at a
point P = (¯ x,
¯
t) with
¯
t > 0 and ¯ x > x
2
. The solution must be
discontinuous at P, since these two crossing lines would give it
diﬀerent values there.
(d)
¯
t = (x
2
−x
1
)/[f
(u
0
(x
1
)) −f
(u
0
(x
2
))].
6. (a) Since the “rectangle” D does not touch the x axis and φ = 0 on
∂D and outside D, equation (25) becomes
__
[uφ
t
+f(u)φ
x
] dxdt = 0. (i)
Since (uφ)
t
+[f(u)φ]
x
= [u
t
+f(u)
x
]φ+[uφ
t
+f(u)φ
x
], we have
__
Di
[uφ
t
+f(u)φ
x
] dxdt =
__
Di
[(uφ)
t
+ (f(u)φ)
x
] dxdt
−
__
Di
[u
t
+f(u)
x
]φ dxdt.
But u is C
1
on the interior of D
i
, and so Exercise 4(b) says
u
t
+f(u)
x
= 0 there. Thus
__
Di
[uφ
t
+f(u)φ
x
] dxdt =
__
Di
+[(uφ)
t
+(f(u)φ)
x
] dxdt. (ii)
(b) By Green’s theorem,
__
Di
[(f(u)φ)
x
−(−uφ)
t
] dxdt =
_
∂D
i
[(−uφ) dx +f(u)φ dt],
and so expression (ii) becomes
__
[uφ
t
+f(u)φ
x
] dxdt =
_
∂Di
φ[−u dx +f(u) dt]
Adding the above for i = 1, 2 and using expression (i) gives
0 =
_
∂Di
φ[−u dx +f(u) dt] +
_
∂D2
φ[−u dx +f(u) dt]
Answers for the Internet Supplement 121
The union of these two boundaries traverses ∂D once and that
portion of Γ within D in each direction, once with the values u
1
and once with the values u
2
. Since φ = 0 outside of D and on
∂D
2
this becomes 0 =
_
Γ
φ¦[−u] dx + [f(u)] dt¦.
(c) Since φ = 0 outside D, the ﬁrst integral is the same as that of the
second conclusion of part (b). The second integral results from
parameterizing the portion of Γ by α(t) = (x(t), t), t
1
≤ t ≤ t
2
.
(d) If [−u]s + [f(u)] = c > 0 at P, then we can choose a small
disk B
centered at P contained in D (described above) such
that [−u](dx/dt) +[f(u)] > c/2 on the part of Γ inside B
. Now
take a slightly smaller disk B
γ
⊂ B
centered at P and pick φ
suck that φ ≡ 1 on B
γ
. 0 ≤ φ ≤ 1 on the annulus B
B
γ
and
φ ≡ 0 outside B
. If α(t
0
) = P, then there are t
3
and t
4
with
t
1
< t
3
< t
0
< t
4
< t
2
and α(t) ∈ B
γ
fort
3
< t < t
4
. But then
_
t2
t1
φ
_
[−u]
dx
dt
+ [f(u)]
_
dt >
c
2
(t
4
−t
3
) > 0,
contradicting the result of part (c). A similar argument (revers
ing signs) works if c < 0.
8. Setting P = g(u)φ and Q = −f(u)φ, applying Green’s theorem on
rectangular R, and using the function φ as in Exercise 4 shows that
if u is a solution of g(u)
t
+f(u)
x
= 0, then
__
t≥0
[g(u)φ
t
+f(u)φ
x
] dxdt +
_
t = 0g(u
0
(x))φ(x, 0) dx = 0.
This is the appropriate analogy to equation (25), deﬁning weak solu
tions of g(u)
t
+f(u)
x
= 0. Thus we want u such that
__
t≥0
_
uφ
t
+
1
2
u
2
φ
x
_
dxdt +
_
t=0
u
0
(x)φ(x, 0)dx = 0 (i: weak)
holds for all admissible φ but such that
__
t≥0
_
1
2
u
2
φ
t
+
1
3
u
3
φ
x
_
dxdt +
_
t=0
1
2
u
2
0
(x)φ(x, 0) dx = 0
(ii: weak)
fails for some admissible φ. The method of Exercise 11 produces the
jump condition s[g(u)] = [f(u)]. For (a), this is s(u
2
−u
1
) = (
1
2
u
2
2
−
1
2
u
2
1
) or
s =
1
2
(u
2
+u
1
). (i: jump)
For (b), it is s(
1
2
u
2
2
−
1
2
u
2
1
) = (
1
3
u
3
2
−
1
3
u
3
1
) or
s =
2
3
u
2
2
+u
1
u
2
+u
2
1
u
2
+u
1
. (ii: jump)
122 Answers for the Internet Supplement
If we take for u
0
(x) a (Heaviside) function deﬁned by u
0
(x) = 0
for x < 0 and u
0
(x) = 1 for x > 0, we are led to consider the
function u(x, t) = 0 when t > 2x and u(x, t) = 1 when t ≤ 2x. Thus
u
1
= 1, u
2
= 0, and the discontinuity curve Γ is given by t = 2x. Thus
the jump condition (i: jump)(i.e., dx/dt =
1
2
(u
1
+u
2
) is satisﬁed.
For any particular φ, there are numbers T and a such that φ(x, t) = 0
for x ≥ a and t ≤ T. Letting Ω be the region 0 ≤ x ≤ a and 0 ≤ t ≤ T,
condition (i: weak) becomes
0 =
__
Ω
_
φ
t
+
1
2
φ
x
_
dxdt +
_
a
0
φ(x, 0) dx
=
_
∂Ω
_
−φ dx +
φ
2
dt
_
+
_
a
0
φ(x, 0) dx
=−
_
a
0
φ(x, 0) dx +
_
T/2
0
_
−φ(x, 2x)(−dx) +
1
2
φ(x, 2x)(−2dx)
_
+
_
a
0
φ(x, 0) dx
Thus (i: weak) is satisﬁed for every φ, and u is a weak solution of
equation (i). However, (ii: weak) cannot be satisﬁed for every φ, since
the jump condition (ii: jump) fails. Indeed, if we multiply (ii: weak)
by 2 and insert u, (ii: weak) becomes
0 =
__
Ω
_
φ
t
+
2
3
φ
x
_
dxdt +
_
a
0
φ(x, 0) dx.
The factor
1
2
has changed to
2
3
, and the computation above now
becomes
0 = −
1
3
_
π/2
0
φ(x, 2x) dx,
which is certainly not satisﬁed for every admissible φ.
10. For example, write [re
iθ
−z
[
2
= [re
iθ
−r
e
iθ
[
2
= [re
i(θ−θ
)
−r
[
2
, use
e
iφ
= cos φ +i sin φ, [z[
2
= z¯ z, and multiply out.
Page 123
Practice Final Examination
This is a practice examination, consisting of 10 questions that requires
approximately 3 hours to complete. Full solutions are at the end of the
answer section following this exam, so you can check your solutions and
selfmark the exam. It is always important to make sure you know how to
solve the problem completely if you only got it partly correct. Of course
there might be correct solutions that are slightly diﬀerent from those given.
Consult your instructor if you have special concerns.
Note that there are also practice quizzes and exams in the student guide.
Good luck!
1. (a) Let a, b > 0 and let a point of mass m move along the curve in
R
3
deﬁned by c(t) = (a cos t, b sin t, t
2
). Describe the curve and
ﬁnd the tangent line at t = π.
(b) Find the force acting on the particle at t = π.
(c) Deﬁne the function E(z) by
E(z) =
_
π/2
0
_
1 −z
2
sin
2
t dt.
Find a formula for the circumference of the ellipse
x
2
a
2
+
y
2
b
2
= 1,
where 0 < a < b, in terms of E(z).
124 Practice Final Examination
(d) Let f : R
3
→R be deﬁned by
f(x, y, z) = −z + exp [x
2
−y
2
].
In what directions starting from (1, 1, 1) is f decreasing at 50%
of its maximum rate of change?
(e) If f is a smooth function of (x, y), C is the circle x
2
+ y
2
= 1
and D is the unit disk x
2
+y
2
≤ 1, is it true that
_
C
sin (xy)
∂f
∂x
dx = sin (xy)
∂f
∂y
dy =
__
D
cos (xy)
_
y
∂f
∂y
−x
∂f
∂x
_
dxdy?
2. (a) Find the extrema of the function
f(x, y, z) = exp [z
2
−2y
2
+x
2
]
subject to the two constraints g
1
(x, y, z) = 1 and g
2
(x, y, z) = 0,
where
g
1
(x, y, z) = x
2
+y
2
, g
2
(x, y, z) = z
2
−x
2
−y
2
.
(b) Suppose S is the level surface in R
n
given by
¦x ∈ R
n
[ g(x) = 0¦
for a smooth function g : R
n
→ R. Suppose φ : R → R
n
is a
curve whose image lies in S; that is, for every x, φ(x) ∈ S. Show
that ∇g(φ(0)) is perpendicular to φ
(0).
(c) Find the minimum distance between the circle x
2
+y
2
= 1 and
the line x +y = 4.
3. (a) Evaluate the integral
_
1
0
_
1
y
e
−x
2
/2
dxdy.
(b) Evaluate
___
D
xdxdy dz,
where D is the tetrahedron with vertices
(0, 0, 0), (1, 0, 0), (0, 2, 0),
_
0, 0,
1
2
_
.
4. Determine whether each of the following statements below is true
or false. If true, justify (give a brief explanation or quote a rele
vant theorem from the course), and if false, give an explanation, or a
Counterexample.
Practice Final Examination 125
(a) For smooth functions f and g,
div (∇f ∇g) = 0.
(b) There exist conservative vector ﬁelds F and G, such that
F G = (x, y, z).
(c) By a “symmetry” argument, the following holds
_
1
0
_
2
−2
_
2
1
z
−1
xye
e
x
2
(y
2
+ 1)
−3
2
(y
4
x
6
+x
8
y
7
+5)
dy dxdz
=
_
π
−π
cos
4
(t) sin (t) dt.
(d) The surface integral of the vector ﬁeld
F(x, y, z) = (x +y
2
+z
2
) i + (x
2
+y +z
2
) j + (x
2
+y
2
+z) k
over the unit sphere equals 4π.
(e) If F is a smooth vector ﬁeld on R
3
,
S
1
= ¦(x, y, 0) [ x
2
+y
2
≤ 1¦,
oriented with the upward normal and the upper hemisphere,
S
2
= ¦(x, y, z) [ x
2
+y
2
+z
2
= 1, z ≥ 0¦,
also oriented with the upward normal, then
__
S1
(∇F) dS =
__
S2
(∇F) dS.
5. Let W be the region in R
3
deﬁned by x
2
+y
2
+z
2
≤ 1 and y ≤ x.
(a) Compute
___
W
(1 −x
2
−y
2
−z
2
) dxdy dz.
(b) Find the ﬂux of the vector ﬁeld F = (x
3
− 3x)i + (y
3
+ xy)j +
(z
3
−xz)k out of the region W.
6. (a) Verify the divergence theorem for the vector ﬁeld
F = 4xzi +y
2
j +yzk
over the surface of the cube deﬁned by the set of (x, y, z) satis
fying 0 ≤ x ≤ 1, 0 ≤ y ≤ 1, and 0 ≤ z ≤ 1.
126 Practice Final Examination
(b) Evaluate the surface integral
__
S
F ndS, where
F(x, y, z) = i +j +z (x
2
+y
2
)
2
k,
and S is the surface of the cylinder x
2
+ y
2
≤ 1, 0 ≤ z ≤ 1,
including the sides and both lids.
7. (a) Let D be the parallelogram in the xy plane with vertices
(0, 0), (1, 1), (1, 3), (0, 2).
Evaluate the integral
__
D
xy dxdy.
(b) Evaluate
___
D
(x
2
+y
2
+z
2
)
1/2
exp [(x
2
+y
2
+z
2
)
2
] dxdy dz,
where D is the region deﬁned by
1 ≤ x
2
+y
2
+z
2
≤ 4 and z ≥
_
x
2
+y
2
.
8. For each of the questions below, indicate if the statement is true
or false. If true, justify (give a brief explanation or quote a relevant
theorem from the course), and if false, give an explanation or a coun
terexample.
(a) If P(x, y) = Q(x, y), then the vector ﬁeld F = Pi + Qj is a
gradient.
(b) The ﬂux of any gradient out of a closed surface is zero.
(c) There is a vector ﬁeld F such that ∇F = yj.
(d) If f is a smooth function of (x, y), C is the circle x
2
+ y
2
= 1
and D is the unit disc x
2
+y
2
≤ 1, then
_
C
e
xy
∂f
∂x
dx +e
xy
∂f
∂y
dy =
__
D
e
xy
_
y
∂f
∂y
−x
∂f
∂x
_
dxdy.
(e) For any smooth function f(x, y, z), we have
_
1
0
_
x
0
_
x+y
0
f(x, y, z) dz dy dx =
_
1
0
_
y
0
_
x+y
0
f(x, y, z) dz dxdy.
9. Let W be the threedimensional region deﬁned by
x
2
+y
2
≤ 1, z ≥ 0, and x
2
+y
2
+z
2
≤ 4.
Practice Final Examination 127
(a) Find the volume of W.
(b) Find the ﬂux of the vector ﬁeld
F = (2x −3xy)i −yj + 3yzk
out of the region W.
10. Let f(x, y, z) = xyze
xy
.
(a) Compute the gradient vector ﬁeld F = ∇f.
(b) Let C be the curve obtained by intersecting the sphere x
2
+
y
2
+ z
2
= 1 with the plane x = 1/2 and let S be the portion
of the sphere with x ≥ 1/2. Draw a ﬁgure including possible
orientations for C and S; state Stokes’s theorem for this region.
(c) With F as in (a) and S as in (b), let G = F+(z −y)i +yk, and
evaluate the surface integral
__
S
(∇G) dS.
128 Practice Final Examination
Page 129
Solutions to Practice Final
Examination
1. (a) Let a, b > 0 and let a point of mass m move along the curve in
R
3
deﬁned by c(t) = (a cos t, b sin t, t
2
). Describe the curve and
ﬁnd the tangent line at t = π.
Solution. This is a stretched elliptical helix. The derivative is
c
(t) = (−a sin t, b cos t, 2t).
The tangent line is
l(t) = c(π) +tc
(π)
= (−a, 0, π
2
) +t(0, −b, 2π)
= (−a, −tb, π
2
+ 2tπ).
(b) Find the force acting on the particle at t = π.
Solution. The acceleration is
c
(t) = (−a cos t, −b sin t, 2),
so c
(π) = (a, 0, 2). Thus, the force is F = ma = m(a, 0, 2).
130 Practice Final Examination Solutions
(c) Deﬁne the function E(z) by
E(z) =
_
π/2
0
_
1 −z
2
sin
2
t dt.
Find a formula for the circumference of the ellipse
x
2
a
2
+
y
2
b
2
= 1,
where 0 < a < b, in terms of E(z).
Solution. The circumference of the ellipse is
C = 4
_
π/2
0
_
a
2
sin
2
t +b
2
cos
2
t dt.
Now substitute cos
2
t = 1 − sin
2
t, and pull out a factor of b to
give the ﬁnal desired formula:
C = 4bE(
_
1 −(a
2
/b
2
)).
(d) Let f : R
3
→R be deﬁned by
f(x, y, z) = −z + exp [x
2
−y
2
].
In what directions starting from (1, 1, 1) is f decreasing at 50%
of its maximum rate of change?
Solution. The gradient is
∇f(x, y, z) =
_
∂f
∂x
,
∂f
∂y
,
∂f
∂z
_
= (2xexp (x
2
−y
2
), −2y exp (x
2
−y
2
), −1).
Thus, the direction of fastest decrease at (1, 1, 1) is −∇f(1, 1, 1) =
(−2, 2, 1). Suppose now that n is the direction that f decreases
at 50% of its maximum rate and n = 1. Then, because the
rate of change of f in the direction n at the point (1, 1, 1) is
n ∇f(1, 1, 1), we get
n (−∇f(1, 1, 1)) =
1
2
 −∇f(1, 1, 1).
Practice Final Examination Solutions 131
Let the angle between n and −∇f(1, 1, 1) be denoted θ (taken
to be the acute angle). Then from the preceding discussion,
cos θn  −∇f(1, 1, 1) =
1
2
 −∇f(1, 1, 1).
Thus, cos θ = 1/2, so θ = π/3. Therefore, the directions are
those that make an angle of π/3 with the vector (−2, 2, 1); that
is, a cone centered on the vector (−2, 2, 1) with a base angle of
2π/3.
(e) If f is a smooth function of (x, y), C is the circle x
2
+ y
2
= 1
and D is the unit disc x
2
+y
2
≤ 1, is it true that
_
C
sin (xy)
∂f
∂x
dx+sin (xy)
∂f
∂y
dy =
__
D
cos (xy)
_
y
∂f
∂y
−x
∂f
∂x
_
dxdy?
Solution. Yes, this is true. Let
P = sin (xy)
∂f
∂x
and Q = sin (xy)
∂f
∂y
and apply Green’s theorem:
_
C
P dx +Qdy =
__
D
_
∂Q
∂x
−
∂P
∂y
_
dxdy.
We have
∂Q
∂x
= y cos xy
∂f
∂y
+ sin (xy)
∂
2
f
∂x∂y
and
∂P
∂y
= xcos xy
∂f
∂x
+ sin (xy)
∂
2
f
∂x∂y
.
The terms containing second derivatives of f cancel, so one gets
the desired result.
2. (a) Find the extrema of the function
f(x, y, z) = exp [z
2
−2y
2
+x
2
],
subject to the two constraints g
1
(x, y, z) = 1 and g
2
(x, y, z) = 0,
where
g
1
(x, y, z) = x
2
+y
2
, g
2
(x, y, z) = z
2
−x
2
−y
2
.
132 Practice Final Examination Solutions
Solution. We present two possible solutions.
Method 1. The ﬁrst method uses Lagrange multipliers. Accord
ing to the Lagrange multiplier method, the extrema are among
the solutions of the equations
g
1
(x, y, z) = 1
g
2
(x, y, z) = 0
∇f = λ
1
∇g
1
+λ
2
∇g
2
.
This gives the ﬁve equations
x
2
+y
2
= 1
z
2
−x
2
−y
2
= 0
2xf = 2xλ
1
−2xλ
2
−4yf = 2yλ
1
−2yλ
2
2zf = 2zλ
2
.
Because f > 0 and, from the ﬁrst two equations, (x, y) ,= (0, 0),
so z ,= 0. Thus, from the last equation, f = λ
2
. There are two
cases:
i. y ,= 0: Here λ
1
= −f from the fourth equation. Substituting
into the third equation gives x = 0. Thus, y = ±1 and
z = ±1.
ii. x ,= 0: Here from the third equation, λ
1
= 2f. Substituting
into the fourth equation gives y = 0. Thus, x = ±1 and
z = ±1.
Evaluating f at each of the eightcandidate gives a minimum of
e
−1
at the four points (0, ±1, ±1) and a maximum of e
2
at the
four points (±1, 0, ±1).
Method 2. The constraint set is the collection of two circles
x
2
+y
2
= 1, z = 1 and x
2
+y
2
= 1, z = −1.
Thus, we are extremizing the function h(x, y) = exp [1−2y
2
+x
2
]
over the circle x
2
+y
2
= 1. That is, the function k(y) = exp [2−
3y
2
] for y between −1 and 1. Thus, because the function k(y) is
decreasing in y
2
, the maximum of e
2
occurs at (±1, 0, ±1) and
the minimum of e
−1
occurs at (0, ±1, ±1).
(b) Suppose S is the level surface in R
n
given by
¦x ∈ R
n
[ g(x) = 0¦
for a smooth function g : R
n
→ R. Suppose φ : R → R
n
is a
curve whose image lies in S; that is, for every x, φ(x) ∈ S. Show
that ∇g(φ(0)) is perpendicular to φ
(0).
Practice Final Examination Solutions 133
Solution. The chain rule applied to g(φ(t)) = 0 gives the
equality ∇g(φ(0)) φ
(0) = 0, which proves the assertion.
(c) Find the minimum distance between the circle x
2
+y
2
= 1 and
the line x +y = 4.
Solution. We wish to minimize the square of the distance
function from a point (x, y) on the circle to a point (u, v) on
the line, namely, f(x, y, u, v) = (x − u)
2
+ (y − v)
2
, subject to
the constraints
g
1
(x, y, u, v) = x
2
+y
2
−1 = 0
and
g
2
(x, y, u, v) = u +v −4 = 0.
The Lagrange multiplier method gives
g
1
= 0
g
2
= 0
∇f = λ
1
∇g
1
+λ
2
∇g
2
,
which gives the following six equations
x
2
+y
2
= 1
u +v = 4
2(x −u) = 2λ
1
x
2(y −v) = 2λ
1
y
−2(x −u) = λ
2
−2(y −v) = λ
2
.
First, notice that λ
1
,= 0, for then we would get x = u, y = v,
which cannot happen since the line does not intersect the circle.
Hence
x =
x −u
λ
1
= −
1
2
λ
2
λ
1
=
y −v
λ
1
= y.
Substituting x = y into the last two of the preceding equations
gives u = v as well. From the ﬁrst and second equations, we
ﬁnd that the potential extrema are located at x = y = ±
√
2/2.
Plugging these values into the distance function, we ﬁnd the
values
d
1
=
¸
_
−
√
2
2
−2
_
2
+
_
−
√
2
2
−2
_
2
= 2
√
2 + 1
134 Practice Final Examination Solutions
and
d
2
=
¸
_√
2
2
−2
_
2
+
_√
2
2
−2
_
2
= 2
√
2 −1.
Thus, the minimum distance is 2
√
2 −1.
One can also do this problem geometrically by drawing the line
and the circle and the line orthogonal to them, both to compute
the intersection points and then calculate the distance between
them. However, in this method, one has to be very careful to
justify the method.
3. Evaluate the integral
_
1
0
_
1
y
e
−x
2
/2
dxdy.
Drawing a ﬁgure and switching the order of integration:
_
1
0
_
1
y
e
−x
2
/2
dxdy =
_
1
0
_
x
0
e
−x
2
/2
dy dx =
_
1
0
xe
−x
2
/2
dx = 1−e
−1/2
.
Solution. Evaluate
___
D
xdxdy dz,
where D is the tetrahedron with vertices
(0, 0, 0), (1, 0, 0), (0, 2, 0),
_
0, 0,
1
2
_
.
Solution. First compute or observe that the plane x+
1
2
y +2z = 1
contains the tetrahedral face passing through the points
(1, 0, 0), (0, 2, 0),
_
0, 0,
1
2
_
.
Thus, one gets
___
D
xdxdy dz =
_
1
0
_
2−2x
0
_
1/2−1/2x−1/4y
0
xdz dy dx
=
_
1
0
_
2−2x
0
x
2
−
x
2
2
−
xy
4
dy dx
=
_
1
0
_
x
2
−x
2
+
x
3
2
_
dx
=
1
24
.
Practice Final Examination Solutions 135
4. Determine whether each of the following statements below is True
or False. If true, Justify (give a brief explanation or quote a rele
vant theorem from the course), and if false, give an explanation, or a
Counterexample.
(a) For smooth functions f and g,
div(∇f ∇g) = 0.
Solution. This is one of the vector identities; it is true and
follows by equality of mixed partials. In fact,
∇f ∇g =
¸
¸
¸
¸
¸
¸
¸
¸
¸
i j k
∂f
∂x
∂f
∂y
∂f
∂z
∂g
∂x
∂g
∂y
∂g
∂z
¸
¸
¸
¸
¸
¸
¸
¸
¸
=
_
∂f
∂y
∂g
∂z
−
∂f
∂z
∂g
∂y
_
i −
_
∂f
∂x
∂g
∂z
−
∂f
∂z
∂g
∂x
_
j
+
_
∂f
∂x
∂g
∂y
−
∂f
∂y
∂g
∂x
_
k.
After taking the divergence, we see that all terms cancel (the
student should be prepared to say which terms cancel with which
terms).
(b) There exist conservative vector ﬁelds F and G, such that
F G = (x, y, z).
Solution. This is false. By part (a), the left side has zero di
vergence, but the right side clearly does not (its divergence is
3).
(c) By a “symmetry” argument, the following holds
_
1
0
_
2
−2
_
2
1
z
−1
xye
e
x
2
(y
2
+ 1)
−3
2
(y
4
x
6
+x
8
y
7
+5)
dy dxdz
=
_
π
−π
cos
4
(t) sin(t) dt
Solution. This is true; both sides are zero by symmetry; the
ﬁrst is odd in the xintegral and the second is zero because it is
an odd function of t.
136 Practice Final Examination Solutions
(d) The surface integral of the vector ﬁeld
F(x, y, z) = (x +y
2
+z
2
) i + (x
2
+y +z
2
) j + (x
2
+y
2
+z) k
over the unit sphere equals 4π.
Solution. This is true. By the divergence theorem, the ﬂux is
the integral of the divergence. But the divergence is the constant
3, so the integral over the unit ball is 3 (4π)/3 = 4π.
(e) If F is a smooth vector ﬁeld on R
3
,
S
1
= ¦(x, y, 0) [ x
2
+y
2
≤ 1¦,
oriented with the upward normal and the upper hemisphere,
S
2
= ¦(x, y, z) [ x
2
+y
2
+z
2
= 1, z ≥ 0¦,
also oriented with the upward normal, then
__
S1
(∇F) dS =
__
S2
(∇F) dS.
Solution. This is true. By Stokes’ theorem, both sides equal
the integral of F around the common boundary, the unit circle
in the plane. One can also do this by Gauss’ theorem by writing
the diﬀerence of the two sides as the surface integral over the
boundary of the region enclosed, invoking the divergence the
orem and using the fact that the divergence of a curl is zero.
5. Let W be the region in R
3
deﬁned by x
2
+y
2
+z
2
≤ 1 and y ≤ x.
(a) Compute
___
W
(1 −x
2
−y
2
−z
2
) dxdy dz.
Solution. Consider spherical coordinates in R
3
:
x = r cos θ sin φ
y = r sin θ sin φ
z = r cos φ,
with
r ∈ [0, ∞), φ ∈ [0, π), θ ∈ [−π, π).
Practice Final Examination Solutions 137
We can rewrite the conditions
x
2
+y
2
+z
2
≤ 1 and y ≤ x
as
(r ≤ 1) and (r sin θ sin φ ≤ r cos θ sin φ),
which is equivalent (because sin φ is nonnegative for φ ∈ [0, π))
to:
(r ≤ 1), (sin θ ≤ cos θ) or, equivalently r ∈ [0, 1], θ ∈ [−3π/4, π/4].
Therefore, the region W is deﬁned, in spherical coordinates, by
r ∈ [0, 1], φ ∈ [0, π), θ ∈ [−3π/4, π/4].
One can also see these limits by drawing the ﬁgure containing
the ball x
2
+ y
2
+ z
2
≤ 1 and the halfspace y ≤ x and the
corresponding region in the xy plane consisting of unit disk cut
by the halfplane y ≤ x.
Now we can compute
___
W
(1 −x
2
−y
2
−z
2
) dxdy dz
=
_
1
0
_
π
0
_
π/4
−3π/4
(1 −r
2
) r
2
sin φdr dθ dφ
=
_
1
0
(r
2
−r
4
) dr
_
π
0
sin φdφ
_
π/4
−3π/4
dθ =
2
15
2 π =
4π
15
.
(b) Find the ﬂux of the vector ﬁeld F = (x
3
− 3x)i + (y
3
+ xy)j +
(z
3
−xz)k out of the region W.
Solution. Let’s ﬁrst compute
∇ F =
∂
∂x
(x
3
−3x)+
∂
∂y
(y
3
+xy)+
∂
∂z
(z
3
−xz) = 3x
2
+3y
2
+3z
2
−3.
Now, using Gauss’ divergence theorem (and invoking part (a)),
we get that the ﬂux of F out of the region W is
__
∂W
F dS =
___
W
(∇ F) dV
=
___
W
(3x
2
+ 3y
2
+ 3z
2
−3) dxdy dz
= (−3)
___
W
(1 −x
2
−y
2
−z
2
) dxdy dz
= (−3)
4π
15
= −
4π
5
.
138 Practice Final Examination Solutions
6. (a) Verify the divergence theorem for the vector ﬁeld
F = 4xzi +y
2
j +yzk
over the surface of the cube deﬁned by the set of (x, y, z) satis
fying 0 ≤ x ≤ 1, 0 ≤ y ≤ 1, and 0 ≤ z ≤ 1.
Solution. First of all, the divergence theorem states that
__
S
F ndS =
___
W
div FdV,
where S is the surface of the cube and W is the cube itself. We
evaluate each side and check that they are equal. The righthand
side is, in this case,
___
W
div FdV =
_
1
0
_
1
0
_
1
0
(4z + 3y) dxdy dz = 2 +
3
2
=
7
2
.
The lefthand side can be evaluated by writing
S = S
1
∪ S
2
∪ S
3
∪ S
4
∪ S
5
∪ S
6
,
where
S
1
: 0 ≤ y, z ≤ 1, x = 0, n
1
= (−1, 0, 0)
S
2
: 0 ≤ y, z ≤ 1, x = 1, n
2
= (1, 0, 0)
S
3
: 0 ≤ x, z ≤ 1, y = 0, n
3
= (0, −1, 0)
S
4
: 0 ≤ s, z ≤ 1, y = 1, n
4
= (0, 1, 0)
S
5
: 0 ≤ x, y ≤ 1, z = 0, n
5
= (0, 0, −1)
S
6
: 0 ≤ x, y ≤ 1, z = 1, n
6
= (0, 0, 1).
Thus, we get
__
S
F ndS
=
6
i=1
__
Si
F ndS
= 0 +
_
1
0
_
1
0
4z dy dz + 0 +
_
1
0
_
1
0
dxdz + 0 +
_
1
0
_
1
0
y dxdy
= 2 + 1 +
1
2
=
7
2
.
Thus, both sides are equal, so the divergence theorem checks in
this case.
Practice Final Examination Solutions 139
(b) Evaluate the surface integral
__
S
F ndS, where
F(x, y, z) = i +j +z(x
2
+y
2
)
2
k,
and S is the surface of the cylinder x
2
+ y
2
≤ 1, 0 ≤ z ≤ 1,
including the sides and both lids.
Solution. Use Gauss’ divergence theorem in space:
__
S
F ndS =
___
W
(div F) dxdy dz.
Here,
div F =
∂
∂x
(1) +
∂
∂y
(1) +
∂
∂z
z(x
2
+y
2
)
2
= (x
2
+y
2
)
2
.
The region W is a cylinder, so it is the easiest to evaluate the
integral in cylindrical coordinates:
_
1
0
_
2π
0
_
1
0
r (r
2
)
2
dr dθ dz =
2π
6
=
π
3
.
7. (a) Let D be the parallelogram in the xy plane with vertices
(0, 0), (1, 1), (1, 3), (0, 2).
Evaluate the integral
__
D
xy dxdy.
Solution. The region is ysimple with sides given by the lines
x=0 and x=1, and top and bottom by the lines y =x and
y =x+2. Therefore, the integral is
__
D
xy dxdy =
_
1
0
_
x+2
x
xy dy dx
=
_
1
0
x
_
(x + 2)
2
2
−
x
2
2
_
dx
=
_
1
0
2(x
2
+x) dx
=
2
3
x
3
+x
2
¸
¸
¸
¸
1
0
=
5
3
.
140 Practice Final Examination Solutions
(b) Evaluate
___
D
(x
2
+y
2
+z
2
)
1/2
exp [(x
2
+y
2
+z
2
)
2
] dxdy dz,
where D is the region deﬁned by 1 ≤ x
2
+ y
2
+ z
2
≤ 4 and
z ≥
_
x
2
+y
2
.
Solution. One should ﬁrst draw a sketch. The region in ques
tion is that between two spheres and inside a cone centered
around the z axis. By drawing a careful ﬁgure and using a little
trigonometry, one ﬁnds that a side of the cone makes an angle
of π/4 with the z axis.
Denoting the integral required by I, we get
I =
_
2π
0
_
π/4
0
_
2
1
ρ
_
e
ρ
4_
ρ
2
sin ϕdρ dϕdθ
= 2π
_
1 −
1
√
2
__
2
1
exp (ρ
4
)ρ
3
dρ
=
π
2
_
1 −
1
√
2
_
e
ρ
4
¸
¸
¸
¸
2
1
=
π
2
_
1 −
1
√
2
_
(e
16
−e).
8. For each of the questions below, indicate if the statement is true
or false. If true, justify (give a brief explanation or quote a relevant
theorem from the course), and if false, give an explanation or a coun
terexample.
(a) If P(x, y) = Q(x, y), then the vector ﬁeld F = Pi + Qj is a
gradient.
Solution. This is false. For example,
xi +xj
is not a gradient because it fails to satisfy the crossderivative
test.
(b) The ﬂux of any gradient out of a closed surface is zero.
Practice Final Examination Solutions 141
Solution. This is false. For example, the vector ﬁeld
F(r) = r
is the gradient of f(r) = r
2
/2, yet its ﬂux out of the unit
sphere is, either by direct evaluation of the surface integral, or
by the divergence theorem, 4π.
(c) There is a vector ﬁeld F such that ∇F = yj.
Solution. This is false. We know that div curl F = 0 for all
vector ﬁelds F, but div(yj) = 1, and so no such F can exist.
(d) If f is a smooth function of (x, y), C is the circle x
2
+ y
2
= 1
and D is the unit disc x
2
+y
2
≤ 1, then
_
C
e
xy
∂f
∂x
dx +e
xy
∂f
∂y
dy =
__
D
e
xy
_
y
∂f
∂y
−x
∂f
∂x
_
dxdy.
Solution. This is true. To see why, let
P(x, y) = e
xy
∂f
∂x
and Q(x, y) = e
xy
∂f
∂y
,
so that the lefthand side of the expression in the problem is the
line integral of P dx +Qdy. By Green’s theorem we have
_
C
P dx +Qdy =
__
D
_
∂Q
∂x
−
∂p
∂y
_
dxdy.
To work out the righthand side in Green’s theorem, we calculate
the partial derivatives:
∂Q
∂x
= ye
xy
∂f
∂y
+e
xy
∂
2
f
∂x∂y
∂P
∂y
= xe
xy
∂f
∂x
+e
xy
∂
2
f
∂y∂x
.
By the equality of mixed partials for f, we get the desired equality
from Green’s theorem.
(e) For any smooth function f(x, y, z), we have
_
1
0
_
x
0
_
x+y
0
f(x, y, z) dz dy dx =
_
1
0
_
y
0
_
x+y
0
f(x, y, z) dz dxdy.
142 Practice Final Examination Solutions
Solution. The lefthand side represents the volume integral of
f over the region W that is between the xy plane and the plane
z = x + y, and over the region in the xy plane bounded by the
x axis, the line x = 1, and the line x = y. The righthand side
represents the volume integral of f over the region W that is
between the xy plane and the plane z = x + y, and over the
region in the xy plane bounded by the x axis, the line y = 1,
and the line x = y. The student should draw a sketch of these
two regions in the xy plane. One sees that these two regions
are diﬀerent, so the two integrals need not be equal. Thus, the
assertion is false.
9. Let W be the threedimensional region deﬁned by
x
2
+y
2
≤ 1, z ≥ 0, and x
2
+y
2
+z
2
≤ 4.
(a) Find the volume of W.
Solution 1. The student should draw a sketch. Setting up the
volume as a triple integral, one gets
Volume = V =
___
W
dxdy dz
=
__
D
_
√
4−x
2
−y
2
0
dz dy dx
=
__
D
_
4 −x
2
−y
2
dy dx,
where D is the unit disc in the xy plane. Changing variables to
polar coordinates gives
V =
_
1
0
_
2π
0
_
4 −r
2
r dr dθ
= 2π
_
1
0
_
4 −r
2
r dr.
This is now integrated using substitution, and after a little com
putation, one gets the answer
V =
2π
3
(8 −3
√
3).
Solution 2. The student should draw a sketch. Using evident
notation, we have
V (W) = V (cap) +V (cylinder of height h),
Practice Final Examination Solutions 143
where the region is divided into a cylinder with a certain height
h and the portion of the sphere above it. Now
V (cyl.) = A(base) height = π h.
To determine h, note that h is the value of z where the cylinder
x
2
+ y
2
= 1 and the sphere x
2
+ y
2
+ z
2
= 4 intersect. At such
an intersection, 1 + z
2
= 4, and so z
2
= 3, that is, z =
√
3 = h.
Therefore,
V (cyl.) = π
√
3.
To ﬁnd the volume of the cap, we think of it as a capped cone
minus a cone with a ﬂat top. We can ﬁnd the volume of the
capped cone using spherical coordinates, and we know the vol
ume of the cone with the ﬂat top (the base) is
1
3
A(base) height.
First of all, using spherical coordinates,
V (capped cone) =
_
2π
0
_
ϕ0
0
_
2
0
ρ
2
sin ϕdρ dϕdθ.
To ﬁnd ϕ
0
, refer to the ﬁgure that we suggested be drawn for
this problem, and one sees that for the relevant right triangle,
cos ϕ
0
=
Adjacent
Hypotenuse
=
√
3
2
and so ϕ
0
=
π
6
.
Therefore,
V (capped cone) =
_
2π
0
_
π/6
0
_
2
0
ρ
2
sin ϕdρ dϕdθ =
16π
3
_
1−
√
3
2
_
.
However,
V (cone with the ﬂat top) =
1
3
Area of base times height
=
1
3
π
√
3 =
π
√
3
3
.
Therefore,
V (cap) =
16π
3
−
8π
√
3
3
−
π
√
3
3
and ﬁnally,
V (W) =
16π
3
−
8π
√
3
3
−
π
√
3
3
+
3π
√
3
3
=
16π
3
−
6π
√
3
3
=
2π
3
(8 −3
√
3).
144 Practice Final Examination Solutions
(b) Find the ﬂux of the vector ﬁeld F = (2x−3xy)i −yj +3yzk out
of the region W.
Solution. The ﬂux is given by
__
∂W
F dS.
To compute this, we use the divergence theorem, which gives
__
∂W
F dS =
___
W
div FdV
=
___
W
(2 −3y −1 + 3y) dV
= V (W) =
2π
3
(8 −3
√
3).
10. Let f(x, y, z) = xyze
xy
.
(a) Compute the gradient vector ﬁeld F = ∇f.
Solution. Using the deﬁnition of the gradient, one gets
F(x, y, z) = e
xy
[(yz +xy
2
z)i + (xz +x
2
yz)j +xyk].
(b) Let C be the curve obtained by intersecting the sphere x
2
+
y
2
+ z
2
= 1 with the plane x = 1/2 and let S be the portion
of the sphere with x ≥ 1/2. Draw a ﬁgure including possible
orientations for C and S; state Stokes’ theorem for this region.
Solution. The student should draw a ﬁgure at this point. One
has to choose an orientation for the surface and the curve. For
example, if the normal points toward the positive x axis, then the
curve should be marked with an arrowhead indicating a counter
clockwise orientation when the curve is viewed from the positive
xaxis. Stokes’ theorem for this region states that for a smooth
vector ﬁeld K:
_
C
K ds =
__
S
(∇K) dS.
Practice Final Examination Solutions 145
(c) With F as in (a) and S as in (b), let G = F+(z −y)i +yk, and
evaluate the surface integral
__
S
(∇G) dS.
Solution. We can write G = F +H, where
H = (z −y, 0, y).
Thus, we can evaluate the given integral as follows:
__
S
(∇G) dS =
__
S
∇(F +H) dS
=
__
S
∇F dS +
__
S
∇H dS.
Because F is a gradient, ∇F = 0, and so
__
S
∇G dS =
__
S
∇H dS =
_
C
H ds
by Stokes’ theorem. To evaluate the line integral, we shall pa
rameterize C; we do this by letting
y =
√
3
2
cos t x =
1
2
and z =
√
3
2
sin t
for t ∈ [0, 2π]. Thus, the line integral of H is
_
C
H ds =
_
2π
0
H(c(t)) c
(t) dt
=
3
4
_
2π
0
(sin t −cos t, 0, cos t) (0, −sin t, cos t) dt
=
3
4
_
2π
0
cos
2
t dt.
Remembering that the average of cos
2
θ over the interval from
0 to 2π is 1/2, we get 3π/4. Thus, the required line integral of
∇G is 3π/4.
The End
Page i
Contents
Preface 2 Diﬀerentiation 2.7 Some Technical Diﬀerentiation Theorems . . . . . . . . . . . 3 HigherOrder Derivatives and Extrema 3.4A Second Derivative Test: Constrained Extrema . . . . . . . 3.4B Proof of the Implicit Function Theorem . . . . . . . . . . 4 Vector Valued Functions 4.1A Equilibria in Mechanics . . . . . . . . . . 4.1B Rotations and the Sunshine Formula . . . 4.1C The Principle of Least Action . . . . . . . 4.4 Flows and the Geometry of the Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
iii 1 1 17 17 21 25 25 30 42 58 65 65 71 83 84 88 91 91
5 Double and Triple Integrals 5.2 Alternative Deﬁnition of the Integral . . . . . . . . . . . . 5.6 Technical Integration Theorems . . . . . . . . . . . . . . . 6 Integrals over Curves and Surfaces 6.2A A Challenging Example . . . . . . . . . . . . . . . . . . . 6.2A The Gaussian Integral . . . . . . . . . . . . . . . . . . . . 8 The Integral Theorems of Vector Analysis 8.3 Exact Diﬀerentials . . . . . . . . . . . . . . . . . . . . . . .
Page ii 8.5 Green’s Functions . . . . . . . . . . . . . . . . . . . . . . . 94 107 123 129
Selected Answers for the Internet Supplement Practice Final Examination Practice Final Examination Solutions
niu. Here are a few sample sites that are relevant for the book: 1. usually by self study.math.13 of the main text.whfreeman. This Internet Supplement is intended to be used with the 5th Edition of our text Vector Calculus.edu/jms/java/dragsphere/ which gives a nice JAVA applet for parallel transport on the sphere.com/MarsdenVC5e Please send any new corrections you may ﬁnd to one of us. of the material—both for the theory as well as the applications. 2. There is of course a huge number of websites that contain a wealth of information. see http://torus. A list of corrections and suggestions concerning the text and instructors guide are available available on the book’s website: http://www.Page iii Preface The Structure of this Supplement. For spherical geometry in Figure 8.math.1C of this internet supplement).uiuc. For further information on the Sunshine formula (see §4. Corrections and Website. It contains supplementary material that gives further information on various topics in Vector Calculus.2. More Websites.edu/~rusin/usesmath/position. see http://www. including diﬀerent applications and also technical proofs that were omitted from the main text. The supplement is intended for students who wish to gain a deeper understanding.sun/ .
com/biography/Newton.edu/~jmr/241/surfint. see http://www. http://scienceworld. and the Instructors Manual. As with the main text.jpl.11.htm Practice Final Examination.feynman. We also thank Brian Bradie and Dave Rusin for their helpful comments. CA 95064 . For the Genesis Orbit shown in Figure 4.html and for Feynman. Jerrold Marsden (marsden@cds. that you allow yourself 3 hours to take the exam and then selfmark it. For this internet supplement. see http://genesismission. For more on Newton.ucsc. We send everyone who uses this supplement and the book our best regards and hope that you will enjoy your studies of vector calculus and that you will beneﬁt (both intellectually and practically) from it.edu) Control and Dynamical Systems Caltech 10781 Pasadena. the student guide. see for instance.nasa.umd.caltech.gov/ 4. keeping in mind that there is often more than one way to approach a problem. At the end of this Internet Supplement. we are very grateful to the readers of earlier editions of the book for providing valuable advice and pointing out places where the text can be improved. Acknowledgements.Page iv 3.math.wolfram. complete with solutions. if you wish to practice your skills.edu) Department of Mathematics University of California Santa Cruz. CA 91125 Anthony Tromba (tromba@math.com/ 5. We recommend.1. For surface integrals using Mathematica Notebooks. see http://www. students will ﬁnd a Practice Final Examination that covers topics in the whole book. we are especially grateful to Alan Weinstein for his collaboration in writing the supplement on the sunshine formula (see the supplement to Chapter 4) and for making a variety of other interesting and useful remarks.
5.2. a sort of “scientiﬁc monthly” that had been established only two years before. following the bitter quarrel between adherents of the two men concerning the independence and priority of the discovery of the calculus.Page 1 2 Diﬀerentiation In the ﬁrst edition of Principia Newton admitted that Leibniz was in possession of a similar method (of tangents) but in the third edition of 1726. and 2. Boyer A History of Mathematics §2. We shall begin by supplying the proofs of the limit theorems presented in §2. It is now fairly clear that Newton’s discovery antedated that of Leibniz by about ten years.2 (the theorem numbering in this section cor .3. but that the discovery by Leibniz was independent of that of Newton. 2. Moreover. Newton deleted the reference to the calculus of Leibniz. Limit Theorems. for he printed an account of his calculus in 1684 in the Acta Eruditorum.7 Some Technical Diﬀerentiation Theorems In this section we examine the mathematical foundations of diﬀerential calculus in further detail and supply some of the proofs omitted from §§2. Leibniz is entitled to priority of publication. Carl B.
The following result was stated in §2. which means there is a neighborhood U of x0 such that f (x) ∈ Dε (b) if x ∈ U . Proof. . x→x0 lim f (x) = b or f (x) → b as x → x0 . that is f (x) ∈ Dε (b) ⊂ N . then x ∈ U . It is freely available on the Vector Calculus website given in the Preface. as x approaches x0 .2. Marsden and A. the ball or disk of radius ε with center b.2 2 Diﬀerentiation responds to that in Chapter 2). Consequently. as x approaches x0 . f is eventually in N as x approaches x0 . or. By the deﬁnition of a limit. x ∈ U . 1 For those interested in a diﬀerent pedagogical approach to limits and the derivative. Properties of Limits. Now since N is open. there is an ε > 0 such that Dε (b) ⊂ N . when. 0 < x − x0 < δ and x ∈ A implies x ∈ Dδ (x0 ) ⊂ U . We say f (x) approaches b as x approaches x0 . there is a δ > 0 such that Dδ (x0 ) ⊂ U . x ∈ A and x = x0 means f (x) − b < ε. Let f : A ⊂ Rn → Rm and let x0 be in A or be a boundary point of A. Theorem 6. Let ε > 0 be given. and x ∈ A implies f (x) ∈ N . Let us ﬁrst establish that this deﬁnition is equivalent to the εδ formulation of limits. We ﬁrst recall the deﬁnition of a limit. We now prove the converse.2. x ∈ A. and consider the ε neighborhood N = Dε (b). Let N be a neighborhood of b. we recommend Calculus Unlimited by J. and let N be a neighborhood of b ∈ Rm . If we choose U = Dδ (x) (according to our assumption). Now since U is open and x0 ∈ U . which means that f (x) − b < ε. We say f is eventually in N as x approaches x0 if there exists a neighborhood U of x0 such that x = x0 . in symbols. Assume that for every ε > 0 there is a δ > 0 such that 0 < x − x0 < δ and x ∈ A implies f (x) − b < ε. Now we are in a position to provide the proof. Let f : A ⊂ Rn → Rm where A is open. given any neighborhood N of b.1 Deﬁnition of Limit. We have to show that f is eventually in N as x → x0 . f is eventually in Dε (b). Weinstein. The following result was also stated in §2. This is the εδ assertion we wanted to prove. First let us assume that limx→x0 f (x) = b. that is. we have f (x) − b < ε. and x = x0 . the values f (x) do not get close to any particular number. we must ﬁnd an open set U ⊂ Rn such that x ∈ U. we say that limx→x0 f (x) does not exist. Then limx→x0 f (x) = b if and only if for every number ε > 0 there is a δ > 0 such that for any x ∈ A satisfying 0 < x − x0 < δ. Let x0 be in A or be a boundary point of A. and x = x0 implies f (x) ∈ N . If. Thus f (x) ∈ Dε (b). x ∈ A.
and f (x) = 0 for all x ∈ A. where (f + g) : A → Rm is deﬁned by x −→ f (x) + g(x). Choose x such that 0 < x − x0 < δ and x ∈ A. It is convenient to use the εδ formulation of Theorem 6. Suppose f (x) → b1 and f (x) → b2 as x → x0 . we can ﬁnd δ1 > 0 such that 0 < x − x0 < δ2 implies f (x) − b2 < ε. and limx→x0 g(x) = b2 . where (f g) : A → R is deﬁned by x −→ f (x)g(x). The following result was also stated without proof in §2. then x→x0 lim 1 1 = . because x0 is in A or is a boundary point of A.2. g : A ⊂ Rn → Rm . f b where 1/f : A → R is deﬁned by x −→ 1/f (x). b1 − b2 = (b1 − f (x)) + (f (x) − b2 ) ≤ b1 − f (x) + f (x) − b2 < ε + ε = 2ε. limx→x0 f (x) = b = 0. Properties of Limits. an impossibility. and similarly. by assumption. . Thus. then f (x)−b1 < ε. b ∈ Rm . for if b1 = b2 we could let ε = b1 − b2 /2 > 0 and we would have b1 − b2 < b1 − b2 . Given ε > 0.7 Diﬀerentiation Theorems 3 Theorem 2. x0 be in A or be a boundary point of A. the following assertions than hold: (i) If limx→x0 f (x) = b. where cf : A → Rm is deﬁned by x −→ c(f (x)). If x→x0 lim f (x) = b1 and x→x0 lim f (x) = b2 . Such x’s exist. then x→x0 lim (f g)(x) = b1 b2 . ﬁnd δ1 > 0 such that if x ∈ A and 0 < x−x0 < δ2 . (ii) If limx→x0 f (x) − b1 and limx→x0 g(x) = b2 . then limx→x0 cf (x) = cb. then b1 = b2 . Let δ be the smaller of δ1 and δ2 . (iv) If m = 1.2. Thus for every ε > 0. Proof. we can. then x→x0 lim (f + g)(x) = b1 + b2 . (iii) If m = 1. Hence b1 = b2 . limx→x0 f (x) = b1 . and c ∈ R. Theorem 3. using the triangle inequality. b1 − b2 < 2ε. Let f : A ⊂ Rn → Rm . Uniqueness of Limits.
Let f : A ⊂ Rn → Rm be a given function with domain A. . . . where fi : A → R. ≤ x2 + y 2 x2 + y 2 x2 + y 2 x3 − y 3 = 0. so we can suppose c = 0. . To prove rule (i).y)→(0. i = 1. To prove rule (ii). we have proved that (f + g)(x) → b1 + b2 as x → x0 . Let δ be the lesser of δ1 and δ2 . . . we can use this to study continuity. In each case. we start with the basic deﬁnition of continuity of a function. . 0≤ 3 Since (x + y)(x2 + y 2 ) x − y3 xx2 + yy 2 ≤ = x + y. Choose δ1 > 0 such that 0 < x − x0 < δ1 implies f (x) − b1 < ε/2. . are the component functions of f . Deﬁnition. .4 2 Diﬀerentiation (v) If f (x) = (f1 (x). from the deﬁnition of limit. . we must produce a number δ > 0 such that the inequality cf (x)−cb < ε holds if 0 < x−x0 < δ. x2 + y 2 Solution. we shall mean that f is continuous at each point x0 of A. Then 0 < x − x0 < δ implies ε ε f (x) + g(x) − b1 − b2 ≤ f (x) − b1 + g(x) − b2 < + = ε. then limx→x0 f (x) = b = (b1 . . . . Now that we have the limit theorems available. We say f is continuous at x0 if and only if x→x0 lim f (x) = f (x0 ). Find the following limit if it exists: (x. . . Similarly. 2 2 Thus. fm (x)). (x. there is a δ with the property that 0 < x − x0 < δ implies f (x) − b < ε = ε/c.y)→(0. let ε > 0 be given. Example 1. If c = 0. Proof. Thus 0 < x − x0 < δ implies cf (x) − cb = c f (x) − b  < ε. the εδ formulation of Theorem 6 is probably the most convenient approach. any δ will do. let ε > 0 be given again. bm ) if and only if limx→x0 fi (x) = bi for each i = 1. m. The proofs of the other assertions are only a bit more complicated and may be supplied by the reader. Let x0 ∈ A. If we say that f is continuous. Let ε = ε/c.0) lim x3 − y 3 .0) x2 + y 2 lim we ﬁnd that Continuity. choose δ2 > 0 such that 0 < x − x0 < δ2 implies g(x) − b2 < ε/2. . m. . We shall illustrate the technique of proof by proving assertions (i) and (ii). which proves rule (i).
Since f is continuous x0 ∈ A. for this γ. The exposition in §2.3 was simpliﬁed by assuming. Suppose f (A) ⊂ B so that g ◦ f is deﬁned on A. Theorem 7. Let us begin by redeﬁning “diﬀerentiable. y − y0 < γ implies g(y) − g(f (x0 )) < ε.2 was the following: Theorem 5. Let U be an open set in Rn and let f : U ⊂ Rn → Rm be a given function. x − x0 < δ implies (g ◦ f )(x) − (g ◦ f )(x0 ) < ε.7 Diﬀerentiation Theorems 5 From Theorem 6. Thus. Diﬀerentiability. x − x0 < δ implies f (x) − f (x0 ) < γ. which is the desired conclusion. One of the properties of continuous functions stated without proof in §2. Continuity of Compositions. we must ﬁnd δ > 0 such that for x ∈ A. x − x0 (1) . Since g is continuous at f (x0 ) = y0 ∈ B. then g ◦ f is continuous at x0 . If f is continuous at x0 ∈ A and g is continuous at y0 = f (x0 ). We use the εδ criterion for continuity. as part of the deﬁnition of Df (x0 ). which in turn implies g(f (x)) − g(f (x0 )) < ε. Proof. given ε > 0. Our next objective is to show that this assumption can be omitted.” Theorem 15 below will show that the new deﬁnition is equivalent to the old one. Deﬁnition. a δ > 0 such that for x ∈ A.2. that the partial derivatives of f existed. we get the εδ criterion for continuity. Let f : A ⊂ Rn → Rm and let g : B ⊂ Rm → Rp . A mapping f : A ⊂ Rn → Rm is continuous at x0 ∈ A if and only if for every number ε > 0 there is a number δ > 0 such that x ∈ A and x − x0 < δ implies f (x) − f (x0 ) < ε. there is. We say that f is diﬀerentiable at x0 ∈ U if and only if there exists an m × n matrix T such that x→x0 lim f (x) − f (x0 ) − T(x − x0 ) = 0. there is a γ > 0 such that for y ∈ B.
. . . x0 = (x01 . In matrix notation. . h < δ implies x0 + h ∈ U . . Suppose f : U ⊂ Rn → Rm is diﬀerentiable at x0 ∈ Rn . . . . f (x0 + h) − f (x0 ) − Th < ε h . ∂x1 ∂xn where ∂fi /∂xj is evaluated at x0 . Our ﬁrst task is to show that the matrix T is necessarily the matrix of partial derivatives. x0n ). . there is no other matrix satisfying condition (1). that is. this implies that T is uniquely determined.. . T = Df (x0 ) = .. . Condition (1) can be rewritten as h→0 lim f (x0 + h) − f (x0 ) − Th =0 h (2) as we see by letting h = x − x0 . Then all the partial derivatives of f exist at the point x0 and the m × n matrix T has entries given by [Tij ] = that is. . ∂x1 ∂xn . . ∂xj ∂f1 ∂f1 . . and hence that this abstract deﬁnition agrees with the deﬁnition of diﬀerentiability given in §2. xn − x0n Tm1 Tm2 . . . . Sometimes we write T(y) as T · y or just Ty for the product of the matrix T with the column vector y. Written in terms of εδ notation. . Theorem 15. .. ∂f ∂fm m . .3. Notice that because U is open. . in other words.6 2 Diﬀerentiation We call T the derivative of f at x0 and denote it by Df (x0 ). as long as δ is small enough. Tmn where x = (x1 . . T1n T21 T22 . . . . . . h or. . equation (2) says that for every ε > 0 there is a δ > 0 such that 0 < h < δ implies f (x0 + h) − f (x0 ) − Th < ε. ∂fi . T(x − x0 ) stands for T11 T12 . . In particular. . and where the matrix entries of T are denoted [Tij ]. T2n x1 − x01 . xn ). . . .
and so the theorem follows. Then f is continuous at x0 . x = x0 . a. we have proved that ∂fi /∂xj exists and equals (Tej )i . a a→0 fi (x0 + aej ) − fi (x0 ) = (Tej )i . f (x) − f (x0 ) < M1 x − x0 for some constant M1 and x near x0 . . 1 ≤ i ≤ m. . Df (x0 ) · h ≤ M h . Proof. Choose ε = 1. where M is the square root of the sum of the squares of the matrix elements in Df (x0 ).7 Diﬀerentiation Theorems 7 Proof. Then by the deﬁnition of the derivative (see formula (2)) there is a δ1 > 0 such that 0 < h < δ1 implies f (x0 + h) − f (x0 ) − Df (x0 ) · h < ε h = h . .5 of the main text). in other words. condition (2) is the same as lim fi (x0 + h) − fi (x0 ) − (Th)i  = 0. We shall use the result of Exercise 2 at the end of this section. a→0 so that lim Diﬀerentiability and Continuity. namely that. then using the triangle inequality. Our next task is to show that dif Theorem 8. 0). By Theorem 3(v). . But (Tej )i = Tij (see §1. h h→0 Here (Thi ) stands for the ith component of the column vector Th. a But this limit is nothing more than the partial derivative ∂fi /∂xj evaluated at the point x0 .2. Now let h = aej = (0. a a→0 or. . If h < δ1 . Let f : U ⊂ Rn → Rm be diﬀerentiable at x0 . Thus. . and furthermore. . . We get lim fi (x0 + aej ) − fi (x0 ) − a(Tej )i  = 0. . f (x0 + h) − f (x0 ) = f (x0 + h) − f (x0 ) − Df (x0 ) · h + Df (x0 ) · h ≤ f (x0 + h) − f (x0 ) − Df (x0 ) · h + Df (x0 ) · h < h + M h = (1 + M ) h . ferentiability implies continuity. which has the number a in the jth slot and zeros elsewhere. lim fi (x0 + aej ) − fi (x0 ) − (Tej )i = 0.
This is called a telescoping sum. Proof. We asserted in §2. . . since each term cancels with the succeeding or preceding one. we get the second assertion of the theorem. According to the deﬁnition of the derivative. h→0 h Write f (x1 + h1 . Theorem 9. . . . . xn−1 + hn−1 . Suppose the partial derivatives ∂fi /∂xj of f all exist and are continuous in some neighborhood of a point x ∈ U . . x2 + h2 . 1+M which proves that (see Exercise 15 at the end of this section) x→x0 lim f (x) = lim f (x0 + h) = f (x0 ). Then f is diﬀerentiable at x. Then h < δ implies f (x0 + h) − f (x0 ) < (1 + M ) ε =ε. . . . xn−1 . . . . xn ). . . . Let f : U ⊂ Rn → Rm . In this proof we are going to use the mean value theorem from onevariable calculus—see §2. By the mean value theorem. To simplify the exposition. . .5 of the main text for the statement. . xn + h) + . . . +f (x1 .3 that an important criterion for diﬀerentiability is that the partial derivatives exist and are continuous. . xn + hn ) +f (x1 . xn + hn ) − f (x1 . . xn ) = f (x1 + h1 . . . . . ∂x1 ∂x2 ∂xn . x3 + h3 . our objective is to show that n ∂f f (x + h) − f (x) − (x) hi ∂xi i=1 lim = 0. . . Now let ε be any positive number. x2 + h2 .8 2 Diﬀerentiation Setting x = x0 + h and M1 = 1 + M . . + (y2 ) hn . h→0 so that f is continuous at x0 . xn + hn ) − f (x1 . x2 . xn + hn ) − f (x1 . xn + hn ) − f (x1 . Criterion for Diﬀerentiability. . xn−1 . . xn + hn ) − f (x1 . . . . that is. xn + hn ) +f (x1 . this expression may be written as f (x + h) − f (x) = ∂f ∂f ∂f (y1 ) h1 + (y2 ) h2 + . . and let δ be the smaller of the two positive numbers δ1 and ε /(1 + M ). We now are able to prove this. . . we shall only consider the case m = 1. . . . . leaving the general case to the reader (this is readily supplied knowing the techniques from the proof of Theorem 15. above). . . except the ﬁrst and the last. f : U ⊂ Rn → R.
the Chain Rule is right up there with the most important results in diﬀerential calculus. the right side approaches 0 as h → 0 so that the left side approaches 0 as well.2. . . + (yn ) − (x) h . this expression is less than or equal to ∂f ∂f ∂f ∂f (y1 ) − (x) h1  + . ∂x1 ∂x1 ∂xn ∂xn since hi  ≤ h for all i. xn−1 . Suppose g is diﬀerentiable at x0 and f is diﬀerentiable at y0 = g(x0 ). and yn = (x1 . We are now in a position to give a careful proof.7 Diﬀerentiation Theorems 9 where y1 = (c1 . Let g : U ⊂ Rn → Rm and f : V ⊂ Rm → Rp be given functions such that g maps U into V . + ∂x1 ∂x1 By the triangle inequality. .5. Thus. Let U ⊂ Rn and V ⊂ Rm be open. ∂x1 ∂x1 ∂xn ∂xn n i=1 ∂f (x) hi ∂xi But since the partial derivatives are continuous by assumption. As explained in §2. we must verify that lim f (g(x)) − f (g(x0 )) − Df (y0 )Dg(x0 ) · (x − x0 ) = 0. . so that f ◦ g is deﬁned. Then f ◦ g is diﬀerentiable at x0 and D(f ◦ g)(x0 ) = Df (y0 )Dg(x0 ). . Proof. + (yn ) − (x) . ∂xn ∂xn = ∂f ∂f (y1 ) − (x) h1 + . + (yn ) − (x) hn  ∂x1 ∂x1 ∂xn ∂xn ∂f ∂f ∂f ∂f ≤ (y1 ) − (x) + . . c2 . According to the deﬁnition of the derivative. we can write n f (x + h) − f (x) − i=1 ∂f (x) hi ∂xi ∂f ∂f (yn ) − (x) hn . . x2 + h2 . . Thus. . . xn + hn ) with c1 lying between x1 and x1 + h1 . . . . . y2 = (x1 . xn + hn ) with c2 lying between x2 and x2 + h2 . Theorem 11: Chain Rule. . we have proved that f (x + h) − f (x) − h ≤ ∂f ∂f ∂f ∂f (y1 ) − (x) + . . cn ) where cn lies between xn and xn + hn . . . . Chain Rule. . x3 + h3 . x − x0 x→x0 .
δ3 /M1 ). 2 2 and so f (g(x)) − f (g(x0 )) − Df (y0 )Dg(x0 )(x − x0 ) <ε x − x0 for 0 < x − x0 < δ. . x − x0 2M This makes the second term in expression (4) less than ε x − x0 /2. say 0 < x − x0 < δ2 . By Theorem 8. there is a δ1 > 0 such that 0 < x − x0 < δ1 implies g(x) − g(x0 ) − Dg(x0 ) · (x − x0 ) ε < . Df (y0 ) · h ≤ M h for some constant M . expression (4) is less than ε x − x0 ε x − x0 + = ε x − x0 . y − y0 < δ3 if x − x0 < δ3 /M1 and x − x0 < δ2 . and so f (g(x)) − f (g(x0 )) − Df (y0 ) · (g(x) − g(x0 )) ε g(x) − g(x0 ) ε x − x0 ≤ < . 2M1 2 Thus if δ = min(δ1 . δ2 . g(x) − g(x0 ) < M1 x − x0 for a constant M1 if x is near x0 . (4) Since g is diﬀerentiable at x0 . This proves the theorem. Let us turn to the ﬁrst term in expression (4). (3) As in the proof of Theorem 8. Thus the righthand side of inequality (3) is less than or equal to f (g(x)) − f (g(x0 )) − Df (y0 ) · (g(x) − g(x0 )) + M g(x) − g(x0 ) − Dg(x0 ) · (x − x0 ) .10 2 Diﬀerentiation First rewrite the numerator and apply the triangle inequality as follows: f (g(x)) − f (g(x0 )) − Df (y0 · (g(x) − g(x0 )) + Df (y0 ) · [g(x) − g(x0 ) − Dg(x0 ) · (x − x0 )] ≤ f (g(x)) − f (g(x0 )) − Df (y0 ) · (g(x) − g(x0 )) + Df (y0 ) · [g(x) − g(x0 ) − Dg(x0 ) · (x − x0 ]) . 2M1 Since y = g(x) and y0 = g(x0 ). given ε > 0. Now choose δ3 such that 0 < y − y0 < δ3 implies f (y) − f (y0 ) − Df (y0 ) · (y − y0 ) < ε y − y0 .
Is f diﬀerentiable at (0. 0). . This function is not diﬀerentiable at (0. if (x. 0). Let f (x.2. y) → (0. The student has already met with a number of examples illustrating the above theorems. as can be seen by letting x = M y. 0) − f (0. 0). 0) (0. We note that ∂f f (x. Diﬀerent limits are obtained for diﬀerent paths of approach.7 Diﬀerentiation Theorems 11 A Crinkled Function. Also. y) = (0. (∂f /∂y)(0. 0)? (See Figure 2. (x2 + y 2 )3/2 which does not have a limit as (x. 0) = lim x→0 ∂x x √ (x · 0)/ x2 + 0 − 0 0−0 = lim = lim =0 x→0 x→0 x x and similarly. Thus the partial derivatives exist at (0.” Solution.7. 0). 0) = 0.1. Example 2. Thus the partial derivatives are not continuous at (0. 0) (x. 0). y) = (0. because it is “crinkled. Let us consider one more of a more technical nature. and so we cannot apply Theorem 9. y) = (0.7. then ∂f y = ∂x x2 + y 2 − 2x(xy)/2 x2 + y 2 x2 + y 2 = y x2 + y 2 − x2 y . 0).) z y x Figure 2.1. y) = 0 xy x2 + y 2 (x.
for any ε > 0 there would be a δ > 0 such that 0 < (h1 . If x→0 lim Ax = 0. however). (c) Let A : Rn → Rm be a linear transformation. show that. h2 ) . cos y. which is untrue if we choose ε ≤ 1/2. show that limx→x0 g(f (x)) = w. Compute Df . In general. limx→x0 g(f (x)) need not equal w. x show that A = 0. If Df (0. 2 Answers. Use the CauchySchwarz inequality to prove that Ax ≤ M x . z) = (ex . (b) If y0 ∈ B. this reads 1/2 < ε. . and w = g(y0 ). 0) existed. h2 ) < ε (h1 . by deﬁnition of diﬀerentiability. in general. 3.12 2 Diﬀerentiation We might now try to show that f is not diﬀerentiable (f is continuous. h2 ) − f (0. (a) Let A : Rn → Rm be a linear transformation with matrix {Aij } so that Ax has components yi = j Aij xj . f (h1 . Let 1/2 M= ij A2 ij . when will Df be a diagonal matrix? 2. as well as selected complete solutions. y. and let x0 be in A or be a boundary point of A and y0 be in B or be a boundary point of B. then by Theorem 15 it would have to be the zero matrix. Exercises2 1. 0). (b) Use the inequality derived in part (a) to show that a linear transformation T : Rn → Rm with matrix [Tij ] is continuous. 0) <ε (h1 . we conclude that f is not diﬀerentiable at (0. Hence. Let f : A → B and g : B → C be maps between open subsets of Euclidean space. (a) If limx→0 f (x) = y0 and limy→y0 g(y) = w. But if we choose 1 2 h1 = h2 . sin z). Let f (x. 0). and hints for oddnumbered exercises are found at the end of this supplement. h2 ) < δ implies f (h1 . since ∂f /∂x and ∂f /∂y are zero at (0. or h1 h2  < ε(h2 + h2 ). Thus. h2 ) that is.
describe explicitly the extra property that a uniformly continuous function has. Show that ∇f (x) is the vector 2Ax.] (b) Prove that x −→ 1/x2 on (0. [HINT: Use Exercise 2. The following function is graphed in Figure 2.7.2: f (x. 5.2. y0 ). in fact all directional derivatives exist. 0). y) = (0.7.2. y) = 0 2xy 2 x2 + y 4 (x. Show that ∂f /∂x and ∂f /∂y exist everywhere. . 1] is continuous.) (a) Prove that a linear map T : Rn → Rm is uniformly continuous.7 Diﬀerentiation Theorems 13 4. Is f diﬀerentiable? 7. so f : Rn → R. (Note that a uniformly continuous function is continuous. 0). y) = (0. Let A = [Aij ] be a symmetric n × n matrix (that is. But show that f is not continuous at (0. 0) (x. Prove from the deﬁnition that f is diﬀerentiable at (x0 . y) = g(x) + h(y). the condition p − q < δ implies f (p) − f (q) < ε. A function f : A ⊂ Rn → Rm is said to be uniformly continuous if for every ε > 0 there is a δ > 0 such that for all points p and q ∈ A. and suppose g is diﬀerentiable at x0 and h is diﬀerentiable at y0 . Aij = Aji ) and deﬁne f (x) = x · Ax. Let f (x. Graph of z = 2xy 2 /(x2 + y 4 ). but not uniformly continuous. z x y Figure 2. 6.
Show that y is a boundary point of an open set A if and only if y is not in A and there is a sequence of distinct points of A converging to y. then x→x0 lim [f (x)]2 = b2 and x→x0 lim f (x) = b. Prove that limx→y f (x) = b if and only if f (xn ) → b for every sequence xn of points in A with xn → y. (a) Does (b) Does 13. x→x0 9.0) lim 14.y)→(0. . 0). it is enough to assume that n − 1 partial derivatives are continuous and merely that the other one exists. y) = (0. (x. Find (x. Show that in Theorem 9 with m = 1. [You may wish to use Exercise 3(b). Prove that s : R2 → R. (b) Let f : A ⊂ Rn → Rm and y be in A or be a boundary point of A. Using the deﬁnition of continuity. prove that f is continuous at x if and only if lim f (x + h) = f (x). h→0 16.14 2 Diﬀerentiation 8. Deﬁne f : R2 → R by f (x.y)→(0. lim x exist? x2 + y 2 x3 exist? (x. Prove that if limx→x0 f (x) = b for f : A ⊂ Rn → R. Use the CauchySchwarz inequality to prove the following: for any vector v ∈ Rn .0) x2 + y 2 lim xy 2 x2 + y 2 . (x. lim v · x = v · x0 .] 10. Does this agree with what you expect when n = 1? 11. (c) If U ⊂ Rm is open. 15. show that f : U → Rp is continuous if and only if xn → x ∈ U implies f (xn ) → f (x).y)→(0. 0) (x. (a) A sequence xn of points in Rm is said to converge to x. 12. y) = (x2 0 Show that f is continuous. y) = (0. y) −→ x + y is continuous. if for any ε > 0 there is an N such that n ≥ N implies x − x0 < ε.0) xy + y 2 )1/2 (x. written xn → x as n → ∞.
do we necessarily have strict inequality of the limits? 19. If f (x) = g(x) for all x = A and if limx→A f (x) = b. Let A ⊂ Rn and let x0 be a boundary point of A. then show that limx→A g(x) = b. we say that “f is o(x) as x → 0” if limx→0 f (x)/ x = 0. If f is o(x) as x → 0. (A deleted neighborhood of x0 is any neighborhood of x0 . (c) Show that f (x) = x2 is o(x) as x → 0. Is g(x) = x also o(x) as x → 0? . less x0 itself. [HINT: Consider the function φ(x) = g(x) − f (x). prove that limx→x0 φ(x) ≥ 0. and assume that for all x in some deleted neighborhood of x0 . 18. as well. (a) If f1 and f2 are o(x) as x → 0. (b) Let g : A → R be a function with the property that there is a number c > 0 such that g(x) ≤ c for all x in A (the function g is said to be bounded). prove that f1 + f2 is also o(x) as x → 0.] (b) If f (x) < g(x). prove that gf is also o(x) as x → 0 [where (gf )(x) = g(x)f (x)]. f (x) ≤ g(x). Given f : A ⊂ Rn → Rm . and then use the fact that the limit of the sum of two functions is the sum of their limits. Let f : A → R and g : A → R be functions deﬁned on A such that limx→x0 f (x) and limx→x0 g(x) exist.7 Diﬀerentiation Theorems 15 17.2.) (a) Prove that limx→x0 f (x) ≤ limx→x0 g(x).
16 2 Diﬀerentiation .
Let f : U ⊂ R2 → R and g : U ⊂ R2 → R be smooth (at least C 2 ) functions. we prove Theorem 10 in §3. W. Assume that ∇g(v0 ) = 0 and that there is a real number λ such that ∇f (v0 ) = λ∇g(v0 ). We begin by recalling the statement from the main text. W.Page 17 3 HigherOrder Derivatives and Extrema Euler’s Analysis Inﬁnitorum (on analytic geometry) was followed in 1755 by the Institutiones Calculi Diﬀerentialis. and it may be said that until recently many modern treatises on the subject are based on it. Form the auxiliary function h = f − λg and . to which it was intended as an introduction.4A Second Derivative Test: Constrained Extrema In this supplement. This is the ﬁrst textbook on the diﬀerential calculus which has any claim to be regarded as complete. Rouse Ball A Short Account of the History of Mathematics Supplement 3.4. and sometimes not altogether accurate. at the same time it should be added that the exposition of the principles of the subject is often prolix and obscure. Let v0 ∈ U. g(v0 ) = c. Theorem 10. and let S be the level curve for g with value c.
then we can make this explicit and use the onevariable secondderivative test. then the curve S is not vertical at v0 and it is reasonable that we can solve for y as a function of x in a neighborhood of x0 . ¯ (ii) If H < 0. the constrained extrema of f are found by looking at the critical points of the auxiliary function h(x. (If ∂g/∂xv0 = 0. That is. λ) is such a point and let v0 = (x0 . y) = f (x.18 3 Higher Derivatives and Extrema the bordered Hessian determinant 0 ¯ H = − − ∂g ∂x ∂g ∂y − ∂g ∂x − ∂g ∂y evaluated at v0 . λ) = f (x. Then f S can be written as a function of one variable. we can similarly solve for x as a function of y. y0 ). v0 In a sense this is a onevariable problem.5 on the implicit function theorem. y) − c). ∂f ∂x =λ v0 ∂g ∂x . the test is inconclusive and v0 may be a minimum. If the function g is at all reasonable. y) − λ(g(x. The chain rule gives df ∂f ∂f dφ = + dx ∂x ∂y dx and d2 f ∂2f ∂ 2 f dφ ∂ 2 f + 2 = +2 2 2 dx ∂x ∂x∂y dx ∂y dφ dx 2 ∂f d2 φ + . If ∂g/∂yv0 = 0. and g(x0 . prove this in §3. then v0 is a local maximum point for f S. y0 ) = c. The proof proceeds as follows. Suppose (x0 . y0 . φ(x)). y) = c for one variable in terms of the other. y) = c is a curve and we are interested in how f varies as we move along this curve. then the set S deﬁned by g(x. or neither. ∂2h ∂x2 ∂2h ∂x∂y ∂2h ∂x∂y ∂2h ∂y 2 ¯ (i) If H > 0. According to the remarks following the Lagrange multiplier theorem. We will.) Suppose S is the graph of y = φ(x). ¯ (iii) If H = 0. in fact. If we can solve the equation g(x. then v0 is a local minimum point for f S. ∂y dx2 (1) . f (x. y. a maximum. v0 ∂f ∂y =λ v0 ∂g ∂y .
we know that ∂f /∂y = λ∂g/∂y and ∂f /∂x = λ∂g/∂x. − ∂y 2 ∂g/∂y ∂y 2 ∂x 2 (3) At v0 . Diﬀerentiating both sides of g(x.4A Second Derivative Test 19 The relation g(x. φ(x)) = c can be used to ﬁnd dφ/dx and d2 φ/dx2 . (2) Substituting equation (7) into equation (6) gives df ∂f ∂f /∂y ∂g = − dx ∂x ∂g/∂y ∂x and d2 f d2 f ∂g 1 ∂f /∂y ∂ 2 g = − dx2 (∂g/∂y)2 dx2 ∂g/∂y ∂x2 ∂y ∂f /∂y ∂ 2 g ∂g ∂g ∂2f − −2 ∂x∂y ∂g/∂y ∂x∂y ∂x ∂y 2 2 ∂ f ∂g ∂f /∂y ∂ 2 g + . ∂y dx2 ∂g/∂x ∂g/∂y 2 so that dφ ∂g/∂x =− dx ∂g/∂y and d2 φ 1 =− 2 dx ∂g/∂y ∂ 2 g ∂g/∂x ∂ 2 g ∂2g −2 + 2 ∂x2 ∂x∂y ∂g∂y ∂y . φ(x)) = c with respect to x gives ∂g ∂g dφ + =0 ∂x ∂y dx and ∂2g ∂ 2 g dφ ∂ 2 g +2 + 2 ∂x ∂x∂y dx ∂y 2 dφ dx 2 + ∂g d2 φ = 0. and so equation (3) becomes df dx and d2 f dx2 = x0 = x0 ∂f ∂x −λ x0 ∂g ∂x =0 x0 1 (∂g/∂y)2 ∂2h ∂x2 0 ∂g ∂y 2 −2 − ∂g ∂x ∂ 2 h ∂g ∂g ∂2h + 2 ∂x∂y ∂x ∂y ∂y − ∂g ∂y ∂g ∂x 2 =− 1 (∂g/∂y)2 − − ∂g ∂x ∂g ∂y ∂2h ∂x2 ∂2h ∂x∂y ∂2h ∂x∂y ∂2h ∂y 2 .3.
. . x1 . . λ) = f (x1 . . . 1. . xn (evaluated at the critical point). xn ) = c is the Hessian of the function f (x1 . . and if it is zero. . xn . so that the level curve g(x. This 3×3 determinant is. .20 3 Higher Derivatives and Extrema where the quantities are evaluated at x0 and h is the auxiliary function introduced above. . . Show that the bordered Hessian of f (x1 . . . xn ) of the n + 1 variables λ. and its sign is opposite that of d2 f /dx2 . . we are at a local maximum. . If it is positive. . Take the special case of the theorem in which g(x. y) = y. . we must be at a local minimum. Can you use this observation to give another proof of the constrained secondderivative test using the unconstrained one? HINT: If λ0 denotes the value of λ determined by the Lagrange multiplier theorem. called a bordered Hessian. . . . . . Exercises. xn ) subject to the single constraint g(x1 . Therefore. . Does Theorem 10 reduce to a theorem you know from onevariable Calculus? 2. xn ) − λg(x1 . . as in the statement of Theorem 10. . xn ) ± (λ − λ0 )2 . if it is negative. y) = c with c = 0 is the xaxis. . . xn ) − λg(x1 . . . the test is inconclusive. consider the function F (x1 . . . . .
i = 1. . . . Theorem 11. ∂x1 ∂xn in other words. we can ﬁnd numbers a > 0 and b > 0 such that if x−x0 < a and z −z0  < a. ∂xi ∂F/∂z (1) We shall prove the case n = 2. z) as follows F (x. z) = F (x. y) and x0 = (x0 . so that F : R3 → R. Suppose for deﬁniteness that it is positive. Write F (x. z) = 0. z). z0 ) = [F (x. then (∂F/∂z)(x. z) ≤ M . z) ≤ M and (∂F/∂y)(x. where x ∈ Rn and z ∈ R. z) ∂F (x. Consider the function h(t) = F (tx + (1 − t)x0 . Finally. ∂z Then there is a ball U containing x0 in Rn and a neighborhood V of z0 in R such that there is a unique function z = g(x) deﬁned for x in U and z in V that satisﬁes F (x. g(x)) = 0. z)] + [F (x0 . z0 ) = 0.4B Implicit Function Theorem 21 Supplement 3. We can also assume that the other partial derivatives are bounded by a number M in this region. Since (∂F/∂z)(x0 . ∂g ∂F/∂xi =− . z) ∂z z=g(x) where Dx F denotes the derivative matrix of F with respect to the variable x.. it is either positive or negative.. . z0 ) = 0 and ∂F (x0 . that is.. By continuity. We write x = (x. Suppose that the function F : Rn+1 → R has continuous partial derivatives. Special Implicit Function Theorem. if x in U and z in V satisfy F (x. z0 )]. with the derivative given by Dg(x) = − 1 Dx F (x. assume that (x0 . z) (2) .3. The case for general n is done in a similar manner. z0 ) = 0. Denoting points in Rn+1 by (x. (∂F/∂x)(x. z0 ) satisﬁes F (x0 . that is. n. then z = g(x). z) > b. Moreover.. which also follows from continuity. y0 . ∂F ∂F Dx F = . z) − F (x0 . y0 ). z) − F (x0 . z = g(x) is continuously diﬀerentiable.4B Proof of the Implicit Function Theorem We begin by recalling the statement. . z) − F (x0 .
y) required by the theorem. z) as a function of z for each x. ∂F (x0 . z) = 0 and z0 = g(x0 ). that is. x − x0  < δ implies [Dx F (θx + (1 − θ)x0 .] Thus. a function with a positive derivative is strictly increasing and thus can have no more than one zero. We leave it to the reader to prove from this construction that z = g(x. from equation (3) and the choice of b. z) = [Dx F (θx + (1 − θ)x0 . there is a z between z0 − a0 and z0 + a0 such that F (x. there is a unique z in V such that F (x. since. z)](x − x0 ) + ∂F (x0 . z0 + a0 ) > 0 and F (x. This z is unique. z) (y − y0 ) ∂x ∂y is less than M δ < M (ba0 /2M ) = ba0 /2. y) is a continuous function. From equation (3).22 3 Higher Derivatives and Extrema for ﬁxed x and z. It remains to establish the continuous diﬀerentiability of z = g(x). z0 ) < 0. Therefore. We have proved that if x is conﬁned to U . and since F (x. Let U be the open ball of radius δ and center x0 in R2 and let V be the open interval on R from z0 − a0 to z0 + a0 . θ is such that F (x. z) (x − x0 ) + (θx + (1 − θ)x0 . φz + (1 − φ)z0 ) ∂z . By the mean value theorem. φz + (1 − φ)z0 ) (z − z0 ). z)](x − x0 ) < ba0 . This deﬁnes the function z = g(x) = g(x. ∂z (3) where φ is between 0 and 1. we have g(x) − g(x0 ) = − [Dx F (θx + (1 − θ)x0 . z)](x − x0 ) . z) = 0. If x − x0  < δ then both x − x0  and y − y0  are less than δ. Substitution of this formula into equation (2) along with a similar formula for the second term of that equation gives F (x. Let a0 satisfy 0 < a0 < a and choose δ > 0 such that δ < a0 and δ < ba0 /2M . by the intermediate value theorem applied to F (x. so that the absolute value of each of the two terms in [Dx F (θx + (1 − θ)x0 . z) = [Dx F (θx + (1 − θ)x0 . z) = 0. x − x0 < δ implies that F (x. Thus. there is a number θ between 0 and 1 such that h(1) − h(0) = h (θ)(1 − 0) = h (θ). by elementary calculus. z0 − a0 ) < 0. z) − F (x0 . z)](x − x0 ). [The inequalities are reversed if (∂F/∂z)(x0 . z)](x − x0 ) ∂F ∂F = (θx + (1 − θ)x0 .
and so we have proved formula (1). z) g(x0 + h. This derivation holds at any point (x. z) ∂g g(x0 + h. y) in U by the same argument. it follows that x → x0 and z → z0 . y0 ) = − ∂x . z) ∂z ∂F (x0 . y0 ) − g(x0 .4B Implicit Function Theorem 23 If we let x = (x0 + h. Since the righthand side of formula (1) is continuous. z) ∂z is proved in the same way. we have proved the theorem. y0 ) = − ∂F ∂y (x0 . ∂F h (x0 . ∂F h→0 ∂x h (x0 . The formula . φz + (1 − φ)z0 ) ∂z As h → 0. and so we get ∂F (x0 . y0 ) (x0 . z) ∂g ∂y (x0 . y0 ) then this equation becomes ∂F (θx + (1 − θ)x0 . y0 ) − g(x0 .3. y0 ) = lim = − ∂x .
24
3 Higher Derivatives and Extrema
Page 25
4
Vector Valued Functions
I don’t know what I may seem to the world, but, as to myself, I seem to have been only like a boy playing on the seashore, and diverting myself in now and then ﬁnding a smoother pebble or a prettier shell than ordinary, whilst the great ocean of truth lay all undiscovered before me. Isaac Newton, shortly before his death in 1727
Supplement 4.1A Equilibria in Mechanics
Let F denote a force ﬁeld deﬁned on a certain domain U of R3 . Thus, F : U → R3 is a given vector ﬁeld. Let us agree that a particle (with mass m) is to move along a path c(t) in such a way that Newton’s law holds; mass × acceleration = force; that is, the path c(t) is to satisfy the equation mc (t) = F(c(t)). If F is a potential ﬁeld with potential V , that is, if F = −∇V , then 1 m c (t) 2
2
(1)
+ V (c(t)) = constant.
(2)
26
4 Vector Valued Functions
(The ﬁrst term is called the kinetic energy.) Indeed, by diﬀerentiating the left side of (2) using the chain rule d 1 m c (t) dt 2
2
+ V (c(t)) = mc (t) · c (t) + ∇V (c(t)) · c (t) = [mc (t) + ∇V (c(t))] · c (t) = 0.
since mc (t) = −∇V (c(t)). This proves formula (2). Deﬁnition. A point x0 ∈ U is called a position of equilibrium if the force at that point is zero: F : (x0 ) = 0. A point x0 that is a position of equilibrium is said to be stable if for every ρ > 0 and > 0, we can choose numbers ρ0 > 0 and 0 > 0 such that a point situated anywhere at a distance less that ρ0 from x0 , after initially receiving kinetic energy in an amount less than 0 , will forever remain a distance from x0 less then ρ and possess kinetic energy less then (see Figure 4.1.1).
ρ
X0
ρ0
X
Figure 4.1.1. Motion near a stable point x0 .
Thus if we have a position of equilibrium, stability at x0 means that a slowly moving particle near x0 will always remain near x0 and keep moving slowly. If we have an unstable equilibrium point x0 , then c(t) = x0 solves the equation mc (t) = F(c(t)), but nearby solutions may move away from x0 as time progresses. For example, a pencil balancing on its tip illustrates an unstable conﬁguration, whereas a ball hanging on a spring illustrates a stable equilibrium. Theorem 1. (i) Critical points of a potential are the positions of equilibrium. (ii) In a potential ﬁeld, a point x0 at which the potential takes a strict local minimum is a position of stable equilibrium. (Recall that a function f is said to have a strict local minimum at the point x0 if there exists
As t increases.2 By analogy with Theorem 1. the particle is constrained to move on a sphere.4. The only critical point of V is at 2 the origin. It follows that the origin is a strict minimum of V . equation (2). the particle moves away from x0 on a path c(t) and V (c(t)) increases [since V (x0 ) = V (c(0)) is a strict minimum]. (i) If at a point P on the surface S the potential V S has an extreme value. The Hessian of V at the origin is 1 k 2 (h2 + h2 + h2 ). Fz = −k 2 z(k = 0). Let us choose a small neighborhood of x0 and start our particle with a small kinetic energy. we have shown that the origin is a position of stable equilibrium. We have 1 m c (t) 2 2 + V (c(t)) = 1 m c (t) 2 2 + V (c(0)). which is 1 2 3 2 positivedeﬁnite. then the point P is a position of equilibrium on the surface. it may be whirling on a string. so that the kinetic energy must decrease. equilibrium points x0 are exactly critical points of V . by (i) and (ii) of Theorem 1. y. then in order for the particle to escape from our neighborhood of x0 . The force ﬁeld F is a potential ﬁeld with potential given by the function V = 1 k 2 (x2 + y 2 + z 2 ).1A Equilibria in Mechanics 27 a neighborhood U of x0 such that f (x) > f (x0 ) for all x in U other than x0 . we ensure that the particle will remain on S. To prove assertion (ii). we shall make use of the law of conservation of energy. y. Fy = −k 2 y. the kinetic energy would have to become negative (which is impossible). if the force ﬁeld F = Fx i+Fy j+Fz k is given by Fx = −k 2 x. We shall argue slightly informally to amplify and illuminate the central ideas involved. z) = 0.) Proof. and determine whether or not they are stable. we have: Theorem 2. outside of which V has increased by a deﬁnite amount. If the initial kinetic energy is suﬃciently small. Thus the particle cannot escape the neighborhood. at which ∇V (x0 ) = 0. 1 The force ﬁeld in this example is that governing the motion of a threedimensional harmonic oscillator. Thus. with ∇φ = 0. Let a point in a potential ﬁeld V be constrained to remain on the level surface S given by the equation φ(x. Example 1. 1 Solution. Find the points that are positions of equilibrium. that is. 2 If φ(x. . for instance. If in formula (1) we replace F by the component of F parallel to S. The part subtracted from F to make it parallel to S is normal to S and is called the centripetal force. The ﬁrst assertion is quite obvious from the deﬁnition F = −∇V . z) = x2 + y 2 + z 2 − r 2 .
It is similar to the proof of Theorem 1. it follows that the points P1 (0. 3 Example 2. 3 These ideas can be applied to quite a number of interesting physical situations. Mass. This conclusion should be physically obvious. where Fx = 0. . Fz ). Goldstein. that is. it follows that P1 is a strict minimum and hence a stable point. Let a particle move in a potential ﬁeld in R2 given by V (x. H. let F = (Fx . The stability of such systems is an important question.. Smale. 0. AddisonWesley. where g is the acceleration due to gravity. part (ii). Let F be the gravitational ﬁeld near the surface of the earth. then the point P is a position of stable equilibrium. Reading. Dynamical Systems and Linear Algebra. Academic Press. The solution of these simultaneous equations is x = 0. if any. z) = x2 + y 2 + z 2 − r2 = 0(r > 0)? Which of these are stable? Solution.28 4 Vector Valued Functions (ii) If a point P ∈ S is a strict local minimum of the potential V S. New York 1974). Exercises. Diﬀerential Equations. The proof of this theorem will be omitted. λ = ±mg/2r. y) = 3x2 + 2xy + y 2 + y + 4. Notice that F is a potential ﬁeld with V = mgz. 1950. z = ±r. 1. y. Chapter 10) and the mathematics literature (e.. −r) and P2 = (0. 0. r) are positions of equilibrium. By observation of the potential function V = mgz and by Theorem 2. such as molecular vibrations. Classical Mechanics. Fy = 0. and Fz = −mg. Find the stable equilibrium points. M. 0 = 2λx 0 = 2λy mg = 2λz x2 + y 2 + z 2 − r2 = 0. By Theorem 2. if a particle with mass m is constrained to the sphere φ(x. in terms of components. Using the method of Lagrange multipliers introduced in §3.g. with the additional fact that the equation of motion uses only the component of F along the surface. y = 0. we have the equations ∇V = λ∇φ φ=0 or. What are the positions of equilibrium. For further information consult the physics literature (e. whereas P2 is not.g.4 to locate the possible extrema. Hirsch and S. Fy .
Let a particle move in a potential ﬁeld in R2 given by V (x. y) = x2 + 24xy + 8y 2 . Which. and V2 (x. subject to the potential V = V1 + V2 .4. if any.1A Equilibria in Mechanics 29 2. where V1 is the gravitational potential in Example 2. Is the point (0. if any. 0) a position of a stable equilibrium? 3. 5. Watch out for pitfalls in your argument. y) = x + y. where V1 is the gravitational potential in Example 2. Attempt to formulate a deﬁnition and a theorem saying that if a potential has a maximum at x0 . then x0 is a position of unstable equilibrium. Find the stable equilibrium points. y) = x2 − 2xy + y 2 + y 3 + y 4 . and V2 (x. Find the stable equilibrium points. (The solution is given at the end of this Internet Supplement). Let a particle be constrained to move on the sphere x2 + y 2 + z 2 = 1. . Let a particle move in a potential ﬁeld in R2 given by V (x. y) = x2 − 4xy − y 2 − 8x − 6y. Let a particle be constrained to move on the circle x2 + y 2 = 25 subject to the potential V = V1 + V2 . are stable? 4. if any. 6. Find all the equilibrium points.
II. This is a nice application of vector calculus ideas because it does not involve any special technical knowledge to understand and is something that everyone can appreciate. for otherwise r would not rotate. Our ﬁrst aim is to ﬁnd a convenient formula for c(t) in terms of its starting position r0 = c(0). If we rotate r about the axis passing through l. The vector r now is a vector function of time.niu.sun/ We thank him for his comments on this section. Marsden and A.30 4 Vector Valued Functions Supplement 4. we can assume that λ = 0 and λ = π.2. Consider two unit vectors l and r in space with the same base point. New York.4 To motivate this. Let λ denote the angle between l and r0 . (Imagine l and r glued rigidly at their base points and then spun about the axis through l. one can learn a lot about rotations along the way.e. making a complete revolution in T units of time. have a look at the fascinating Figure 4.12 below which gives a plot of the length of day as a function of one’s latitude and the day of the year. SpringerVerlag. which can be referred to for more information and some interesting historical remarks.) Assume that the rotation is at a uniform rate counterclockwise (when viewed from the tip of l). III by J. See also Dave Rusin’s home page http://www.1B Rotations and the Sunshine Formula In this supplement we use vector methods to derive formulas for the position of the sun in the sky as a function of latitude and the day of the year. l and r0 are not parallel. While it is not trivial. edu/~rusin/ and in particular his notes on the position of the sun in the sky at http:// www.math.1. so we may write r = c(t). A Bit About Rotations. Weinstein. Our goal in this supplement is to use vector methods to derive this formula and to discuss related issues. r l Figure 4..1. In 4 This material is adapted from Calculus I.2).math. its tip describes a circle. . If r rotates about l. Even if one does not get all the way through the details.edu/~rusin/usesmath/position. i . then the tip of r describes a circle (Figure 4. it also does not require any particularly advanced ideas—it mainly requires patience and perseverance.1.niu.
k).1.1. n0 π/2 – λ m0 r0 λ l Figure 4. This was determined by dotting both sides of formula (1) by l and using √ the fact that l is a unit vector. is orthogonal to l. just like (i. j. so sin λ = 0. and makes an angle of (π/2) − λ with r0 .4. Example 1. Now add to this ﬁgure the unit vector n0 = l × m0 . r0 π/2 − λ (1) λ l m0 Figure 4. formula (1) can be taken as the algebraic deﬁnition of m0 by writing m0 = (1/ sin λ)r0 − (cos λ/ sin λ)l. In fact. Construct the unit vector m0 as shown in Figure 4. π). Find m0 and n0 . and so from formula (1) we get m0 = cos λ 1 c(0) − l sin λ sin λ 3 3 1 2 1 1 = k− √ · √ (i + j + k) = √ k − √ (i + j) 2 3 2 3 6 6 . n0 ) is a righthanded orthogonal set of unit vectors. √ Solution. and λ = π. (See Figure 4. The triple (l. We assumed that λ = 0.1. The vector m0 is in the plane of r0 and l. we shall take λ in the open interval (0.3. m0 .3. n0 ) consists of three mutually orthogonal unit vectors.1.1B Rotations and the Sunshine Formula 31 fact.4.) The triple (l. m0 . sin λ = 1 − cos2 λ = 2/3. From this ﬁgure we see that r0 = (cos λ)l + (sin λ)m0 .4. √ Let l = (1/ 3)(i + j + k) and r0 = k. Thus. The angle between l and r0 is given by cos λ = l · r0 = 1/ 3.
Express the function c(t) explicitly in terms of l. Solution. n0 ). It goes through an angle 2π in time T.1. We have cos λ = l · r0 and sin λ = l × r0 . so we must have (up to an . Furthermore n0 is a unit vector perpendicular to both l and r0 . Example 2. The three vectors v. Since the angle λ remains constant. in terms of the basic vector triple (l. (3) This formula expresses explicitly how r changes in time as it is rotated about l.5. Inserting this in formula (2) and rearranging gives r(t) = (cos λ)l + sin λ cos 2πt T m0 + sin λ sin 2πt T n0 . On the other hand. r0 . and T . and n all rotate about l. m. m0 . 2 2 Return to Figure 4.1. it rotates in a circle in the plane of m0 and n0 . so it goes through an angle 2πt/T in t units of time.4 and rotate the whole picture about the axis l. since m is perpendicular to l.1. formula (1) applied after time t to r and l gives (see Figure 4.5) 1 cos λ m= r− l (2) sin λ sin λ v l m n Figure 4. and so m = cos 2πt T m0 + sin 2πt T n0 .32 4 Vector Valued Functions and n0 = l × m0 = i 1 √ 3 1 −√ 6 j 1 √ 3 1 −√ 2 k 1 √ 3 2 −√ 6 2 1 = √ i − √ j. Now the “rotated” vectors m and n will vary with time as well.
l × r0 Thus (sin λ)n0 = l × r0 . from formula (1). . Verify that equation (3) gives the same formula. Its speed is therefore (2π sin λ)/T (Figure 4.1. Example 3. we ﬁnd the velocity vector to be dr 2π = − sin λ · sin dt T 2πt T m0 + sin λ · 2π cos T 2πt T n0 . The tip of r sweeps out a circle of radius sin λ. 2 sin2 2πt T + sin2 λ · 2π T 2 cos2 2πt T = sin λ · as above.4.6). and its length is (since m0 and n0 are unit orthogonal vectors) dr = dt sin2 λ · 2π T 2π T . l r sin λ λ Figure 4.6. The tip of r sweeps out a circle of radius sin λ.1. Solution. From formula (3). Finally. so it covers a distance 2π sin λ in time T. we obtain (sin λ)m0 = r0 − (cos λ)l = r0 − (r0 · l)l. r = (r0 · l)l + cos 2πt T [r0 − (r0 · l)l] + sin 2πt T (l × r0 ). Show by a direct geometric argument that the speed of the tip of r is (2π/T ) sin λ.1B Rotations and the Sunshine Formula 33 orientation) n0 = l × r0 . Substituting all this into formula (3).
We will use a simpliﬁed model of the earthsun system.1. then the axis l tilts in the direction −i. j.8) Now let r be the unit vector at time t from the center of the earth to a ﬁxed point P on the earth’s surface. . This phenomenon called precession or wobble. 000 years.5◦ . Now we apply our study of rotations to the motion of the earth about the sun. We will denote this angle by α. Notice that the unit vector pointing from the earth to the sun is −u and that we have oriented our axes so that u = i when t = 0. in which the sun is ﬁxed at the origin of our coordinate system and the earth moves at uniform speed around a circle centered at the sun. See Figure 4.1. The unit vector u points from the sun to the earth at time t. We will assume that l is ﬁxed5 with respect to i.7. incorporating the rotation of the earth about its own axis as well. Notice that if r is located with its base 5 Actually.34 4 Vector Valued Functions Rotation and Revolution of the Earth. is due to the irregular shape of the earth and may play a role in longterm climatic changes. the axis l is known to rotate about k once every 21. (See Figure 4. we have u = cos(2πt/Ty )i + sin(2πt/Ty )j where Ty is the length of a year (t and Ty measured in the same units). The earth rotates about an axis which we represent by a unit vector l pointing from the center of the earth to the North Pole. and so we must have l = cos αk − sin αi. Let u be a unit vector pointing from the sun to the center of the earth. See pages. If we measure time so that the ﬁrst day of summer in the northern hemisphere occurs when t = 0. and k. such as ice ages.1. k j u i –u Figure 4.7. Viking (1974). 130134 of The Weather Machine by Nigel Calder. astronomical measurements show that the inclination of l (the angle between l and k) is presently about 23. Next we wish to take into account the rotation of the earth.
where λ is the angle between l and r0 . then r lies in the plane of l and i and makes an angle of less than 90◦ with −i. point at P.1.4. it is noon at the point P . l = (cos α)k − (sin α)i and m0 = −(sin α)k − (cos α)i. As in Figure 4. we obtain the expression r0 = (sin )l + (cos )m0 . Referring to Figure 4.1. the earth’s axis tilts toward the sun by the angle α.1. then it represents the local vertical direction. Prove that n0 = l × m0 = −j. we introduce the unit vector m0 = −(sin α)k − (cos α)i orthogonal to l. At t = 0.1. We then have r0 = (cos λ)l + (sin λ)m0 . let n0 = l × m0 . − sin α Now we apply formula (3) to get r = (cos λ)l + sin λ cos 2πt Td m0 + sin λ sin 2πt Td n0 .9). Solution. so i l × m0 = − sin α − cos α j 0 0 k cos α = −j(sin2 α + cos2 α) = −j. Algebraically. where is the latitude of the point P. . l × m0 is a unit vector orthogonal to l and m0 pointing in the sense given by the righthand rule.4.1B Rotations and the Sunshine Formula 35 k α N l k i Figure 4.8. so l × m0 points orthogonal to it in the direction −j. Since λ = π/2 − . Example 4. We will assume that P is chosen so that at t = 0. Geometrically. But l and m0 are both in the i − k plane. (See Figure 4.9.
It diﬀers from the ordinary. In fact. . 6 Substituting the expressions derived above for λ. k vectors). or solar. The vector r is the vector from the center of the earth to a ﬁxed location P. we get r = sin (cos α k − sin α i) 2πt + cos cos (− sin α k − cos α i) − cos sin Td Hence r = − sin sin α + cos cos α cos + sin cos α − cos sin α cos 2πt Td 2πt Td i − cos sin k. 6 T is called the length of the sidereal day. and n0 .. our i. where Td is the length of time it takes for the earth to rotate once about its axis (with respect to the “ﬁxed stars”—i. j. Td ≈ 23. 2πt Td j (4) 2πt Td j.e. where R is the radius of the earth and is the latitude.36 4 Vector Valued Functions k l k N r l i j m λ S Figure 4.) Solution. l. Example 5. The vector m0 is a unit vector in the plane of the equator (orthogonal to l) and in the plane of l and r0 . The speed is s = (2πR/Td ) sin λ = (2πR/Td ) cos . The latitude of P is l and the colatitude is λ = 90◦ − . m0 . the actual vector from the earth’s center to a point P on its surface is Rr). What is the speed (in kilometers per hour) of a point on the equator due to the rotation of the earth? A point at latitude 60 ◦ ? (The radius of the earth is 6371 kilometers. d day by about 1 part in 365 because of the rotation of the earth about the sun.9.93 hours. (The factor R is inserted since r is a unit vector.1.
10. we get s = 1673 cos kilometers per hour. Substituting u = cos(2πt/Ty )i + sin(2πt/Ty )j and formula (3) into this formula for sin A and taking the dot product gives sin A = cos 2πt 2πt sin sin α + cos cos α cos Ty Td 2πt 2πt + sin cos sin Ty Td 2πt 2πt = cos sin sin α + cos cos cos α cos Ty Ty 2πt 2πt + sin sin . s = 836. 4. For what 2πt Td (5) Example 6. Solution. Thus we want to compute sin A. With formula (3) at our disposal. The ratio of area 1 to area 2 is sin A.1. Thus.11 we see that sin A = −u · r. The intensity of light on a portion of the earth’s surface (or at the top of the atmosphere) is proportional to sin A.1.1B Rotations and the Sunshine Formula 37 Using Td = 23. At the equator = 0. and the intensity then is of course zero.4. when A = 0 or π. The intensity of sunlight is proportional to sin A. Now sin A = 0 corresponds to the sun on the horizon (sunrise or sunset). this .4 kilometers per hour.1. at t = 0.93 hours and R = 6371 kilometers.) sunlight 1 A 2 Figure 4. is sin A = 0? Interpret With t = 0 we get sin A = sin sin α + cos cos α = cos( − α). From Figure 4. This is zero when − α = ±π/2. Ty Td Set t = 0 in formula (4).10). at = 60◦ . where A is the angle of elevation of the sun above the horizon (see Fig. your result. so the speed is 1673 kilometers per hour. (At night sin A is negative. we are ready to derive the sunshine formula.
occurs when = α ± (π/2). To simplify our calculations. Our next goal is to describe the variation of sin A with time on a particular day. and formula (4. Q = cos(2πn/365) cos α. For this purpose. To ﬁnd U. we will assume that the expressions cos(2πt/Ty ) and sin(2πt/Ty ) are constant over the course of any particular day. The case α + (π/2) is impossible. since Ty is approximately 365 times as large as the change in t. We will write the expression Q cos(2πt/Td ) + R sin(2πt/Td ) in the form U cos[2π(t − tn )/Td ]. the time variable t is not very convenient. Td Td . The geometry for the formula sin A = cos(90◦ − A) = −u · r. The case = α − (π/2) corresponds to a point on the Antarctic Circle. where tn is the time of noon of the nth day. since lies between −π/2 and π/2. indeed at t = 0 (corresponding to noon on the ﬁrst day of northern summer) the sun is just on the horizon at the Antarctic Circle.11.4) gives sin A = (sin )P + (cos ) Q cos where P = cos(2πn/365) sin α. Setting this equal to Q cos(2πt/Td ) + R sin(2πt/Td ) and comparing coeﬃcients of cos 2πt/Td and sin 2πt/Td gives U cos 2πtn 2πtn = Q and U sin = R.1. we may replace 2πt/Ty by 2πn/365. On the nth day (measured from June 21).38 4 Vector Valued Functions l r P 90°– A –u sunlight Figure 4. and R = sin(2πn/365). we use the addition formula to expand the cosine: U cos 2πt 2πtn − Td Td 2πt 2πtn = U cos cos Td Td 2πt Td + R sin 2πt Td .1. it will be better to measure time from noon on the day in question. + sin 2πt Td sin 2πtn Td . this is a reasonable approximation.
◦ Formula (4) also tells us how long days are. We are interested mainly in the formula for U . and τ = 2 hours. we substitute into formula (5) to obtain the ﬁnal formula: sin A = sin cos + cos 2πn 365 sin α 2πn 365 sin2 α cos 2πτ . A = 0. the sun will never set (“midnight sun”). This reﬂects the fact that. it is either light all day or dark all day. then tan = ∞. We get sin A = 0. That is. cos 2πS 24 = − tan sin α cos(2πT /365) 1 − sin2 α cos2 (2πT /365) . while dividing the second equation by the ﬁrst gives tan(2πtn /Td ) = R/Q.5196.1) 7 We take the positive square root because sin A should have a local maximum when t = tn . on Feb. .4.m. 24 (6) 1 − cos2 Example 7. If = ±π/2. (4.8 At the time S of sunset. How high is the sun in the sky in Edinburgh (latitude 56 ◦ ) at 2 p. n = number of days after June 21 = 225.3. there will be some values of t for which the righthand side of formula (1) does not lie in the interval [−1.◦ = 90◦ − 56◦ = 34◦ .1B Rotations and the Sunshine Formula 39 Squaring the two equations and adding gives7 U 2 = Q2 + R2 or U = Q2 + R2 .1. On the days corresponding to these values of t. 1 − cos2 Letting τ be the time in hours from noon on the nth day so that τ /24 = (t − tn )/Td . 1]. and the righthand side does not make sense at all. depending upon the season. so A = 31. substituting for Q and R gives U= = = cos2 cos2 2πn 365 2πn 365 cos2 α + sin2 2πn 365 2πn 365 (1 − sin2 α) + sin2 2πn 365 sin2 α.5. 8 If π/2 − α <   < π/2 (inside the polar circles). at the poles. We plug into formula (6): α = 23. 1? Solution.
Also. Exercises. (b) Find r = c(t) if T = 24. From formula (3). Show by a direct geometric construction that r = c(Td /4) = −(sin sin α)i − (cos )j + (sin cos α)k. (a) Find m0 and n0 . Let l = (j + k)/ 2 and r0 = (i − j)/ 2.40 4 Vector Valued Functions Solving for S. and remembering that S ≥ 0 since sunset occurs after noon. √ √ 1. Day length as a function of latitude and day of the year.12.1. we get 12 sin α cos(2πT /365) . show this geometrically.2) π 2 2 (2πT /365) 1 − sin α cos The graph of S is shown in Figure 4. If the earth rotated in the opposite direction about the sun. Does this formula agree with formula (3)? . 2. would Td be longer or shorter than 24 hours? (Assume the solar day is ﬁxed at 24 hours. For what values of t is c(t) · n = 0? 3. S= cos−1 − tan (4. verify that c(T /2) · n = 0.) 4. (c) Find the equation of the line tangent to c(t) at t = 12 and T = 24.12.1. 24 18 Length 12 of Day 6 0 0 100 200 Day of Y ear 300 –90 365 La 90 De gre in es 0 titu de Figure 4.1.
5◦ ? . How would your answer in Exercise 8 change if the earth were to roll to a tilt of 32◦ instead of 23. on January 15? (The latitude of Paris is 49◦ N. How much solar energy (relative to a summer day at the equator) does Paris receive on January 15? (The latitude of Paris is 49◦ N). Why does formula (6) for sin A not depend on the radius of the earth? The distance of the earth from the sun? 7.1B Rotations and the Sunshine Formula 41 5.) 8. 9.4. 6.m. How high is the sun in the sky in Paris at 3 p. Derive an “exact” formula for the time of sunset from formula (4).
Instead of worrying about the lecture. and have.” Then he told me something which I found absolutely fascinating. for instance) which starts somewhere and moves to some other point by free motion—you throw it.13). when I began to prepare this lecture I found myself making more analyses on the thing. 9 Lecture 19 from The Feynman Lectures on Physics. you try a diﬀerent motion. In other words. It goes from the original place to the ﬁnal place in a certain amount of time. “You look bored. since then. Mr. .42 4 Vector Valued Functions Supplement 4.14) but got there in just the same amount of time. Now. and it goes up and comes down (Figure 4.1.1. Every time the subject comes up. Then he said this: If you calculate the kinetic energy at every moment on the path. I want to tell you something interesting. it went like this (Figure 4. you’ll ﬁnd that the number you’ll get is bigger then that for the actual motion. Bader told me the following: Suppose you have a particle (in a gravitational ﬁeld. I got involved in a new problem. the laws of Newton could be stated not in the form F = ma but in the form: the average kinetic energy less the average potential energy is as little as possible for the path of an object going from one point to another. actual motion there t2 here t1 Figure 4. and integrate it over the time during the whole path.1C The Principle of Least Action By Richard Feynman9 When I was in high school. my physics teacher—whose name was Mr. In fact. take away the potential energy.13.1. Suppose that to get from here to there. always found fascinating. I work on it. The subject is this—the principle of least action. Bader—called me down one day after physics class and said.
15.15).4. then if the particle has the path x(t) (let’s just take one dimension for a moment. But we could imagine . If you take the case of the gravitational ﬁeld. Let me illustrate a little bit better what it means. The actual motion is some kind of a curve—it’s a parabola if we plot against the time—and gives a certain value for the integral.1. and the potential energy at any time is mgx. x t1 t2 t Figure 4. Let’s suppose that at the original time t1 we started at some height and at the end of the time t2 we are deﬁnitely ending at some other place (Figure 4. where x is the height above the ground.1. the kinetic energy is 1 m(dx/dt)2 .14. Then the integral is t2 t1 1 m 2 dx dt 2 − mgx dt. Now I 2 take the kinetic energy minus the potential energy at every moment along the path and integrate that with respect to time from the initial time to the ﬁnal time.1C Principle of Least Action 43 imagined motion there t2 here t1 Figure 4. we take a trajectory that goes up and down and not sideways).1.
That is because there is also the potential energy. The thing is that the average speed has got to be. So we see that the integral is a minimum if the velocity is a constant (when there are no forces). an object thrown up in a gravitational ﬁeld does rise faster ﬁrst and then slow down. the kinetic energy integral is least. or for any other path we want. x t1 t2 t Figure 4.16). so it must go at a uniform speed.) Why is that? Because if the particle were to go any other way. Now the mean square of something that deviates around an average. The miracle is that the true path is the one for which that integral is least. The correct path is like this (Figure 4. as you know. As and example. then sometimes you are going too fast and sometimes you are going too slow. . suppose we take the case of a free particle for which there is no potential energy at all. You can do it several ways: You can accelerate like mad at the beginning and slow down with the breaks near the end.44 4 Vector Valued Functions some other motion that went very high and came up and down in some peculiar way (Figure 4.17). But if you do anything but go at a uniform speed. The average velocity is the same for every case because it has to get from ‘here’ to ‘there’ in a given amount of time. the total distance that you have gone over the time. and so on. is always greater than the square of the mean. so the kinetic energy integral would always be higher if you wobbled your velocity than if you went at a uniform velocity. and we must have the least diﬀerence of kinetic and potential energy on the . We can calculate the kinetic energy minus the potential energy and integrate for such a path. (We know that’s the right answer—to go at a uniform speed. .1. or you can go backwards for a while and then go forward. the velocities would be sometimes higher and sometimes lower then the average. say your job is to start from home and get to school in a given length of time with the car. Then the rule says that in going from one point to another in a given amount of time. First. Now.1. or you can go at a uniform speed. of course.16.1. Let’s try it out.
average.1.1C Principle of Least Action x 45 there no forces here t1 t2 t Figure 4.1. as small as possible. On the other hand. but you want to go up some. you can’t go up too fast. because you will then have too much kinetic energy involved—you have to go very fast to get way up and come down again in the ﬁxed amount of time available. or too far.4. kinetic minus the potential. .18.1.17. we will get a lower diﬀerence if we can get as soon as possible up to where there is a high potential energy. Then we can take the potential away from the kinetic energy and get a lower average. So it is better to take a path which goes up and gets a lot of negative stuﬀ from the potential energy (Figure 4. x more – PE there more + KE here t1 t2 t Figure 4.18). So you don’t want to go too far up. So i turns out that the solution is some kind of balance between trying to get more potential energy with the least amount of extra kinetic energy—trying to get the diﬀerence. Because the potential energy rises as we go up in space.
It is not the ordinary calculus. That is a completely diﬀerent branch of mathematics. Our mathematical problem is to ﬁnd out for what curve that number is the least. But we can do it better than that. and we have to ﬁnd the value of what variable where the function is least or most. Ordinarily we just have a function of some variable. the deviation of the function from its minimum value is only second order. The idea is that we imagine that there is a true path and that any other curve we draw is a false path.1. That’s a possible way. For each point on the rod we have a temperature. that’s just the ordinary calculus of maxima and minima. of course. It is the kinetic energy. S. I am going to horrify and disgust you with the complexities of life by proving that it is so. integrated over time. so that if we calculate the action for the false path we will get a value that is bigger that if we calculate the action for the true path (Figure 4. that’s the true path. is to calculate the action for millions and millions of paths and look at which one is lowest. minus the potential energy. Any other curve encloses less area for a given perimeter than the circle does. So if we give the problem: ﬁnd that curve which encloses the greatest area for a given perimeter. At any place else on the . in and ordinary function like the temperature—one of the properties of the minimum is that if we go away from the minimum in the ﬁrst order. it is called the calculus of variations. but another way of deﬁning a circle is this: a circle is that curve of given length which encloses the biggest area. But I don’t know when to stop talking.19). because he was a very good teacher and knew when to stop talking. Problem: Find the true path. For each different possible path you get a diﬀerent number for this action. we would have a problem of the calculus of variations—a diﬀerent kind of calculus than you’re used to. For example. we have a rod which has been heated in the middle and the heat is spread around. For instance. We have a certain quantity which is called the action. But watch out. the circle is usually deﬁned as the locus of all points at a constant distance from a ﬁxed point. There are many problems in this kind of mathematics. In fact.46 4 Vector Valued Functions That is all my teacher told me. You calculate the action and just diﬀerentiate to ﬁnd the minimum. So instead of leaving it as an interesting remark. Here is the way we are going to do it. t2 Action = S = t1 (KE − PE)dt. You say—Oh. When you ﬁnd the lowest one. But now for each path in space we have a number—quite a diﬀerent thing—and we have to ﬁnd the path in space for which the number is the minimum. So we make the calculation for the path of the object. and we must ﬁnd the point at which that temperature is largest. The kind of mathematical problem we will have is a very diﬃcult and a new kind. When we have a quantity which has a minimum—for instance. Remember that the PE and KE are both functions of time. Where is it? One way.
If there is a change in the ﬁrst order when I deviate the curve a certain way. We would get the action to increase one way and to decrease the other way.20). there is a change in the action that is proportional to the deviation.1C Principle of Least Action x 47 S1 true path S2 fake path S2 > S1 t Figure 4. The change presumably makes the action greater. a curve which diﬀers only a little bit from it will. The only way that it could really be a minimum is that in the ﬁrst approximation it doesn’t make any change. Any diﬀerence will be in the second approximation.1. a tiny motion away makes. x minimum T x T ( x)2 t Figure 4.4. If we have the true path. That is what we are going to use to calculate the true path. in the ﬁrst approximation.1.20.19. make no diﬀerence in the action. That is easy to prove. But then if the change is proportional to the deviation. if we move a small distance the value of the function changes also in the ﬁrst order. in the ﬁrst approximation. reversing the sign of the deviation will make the action less. no diﬀerence (Figure 4. But at a minimum. that the changes are . if we really have a minimum. curve. otherwise we haven’t got a minimum.1.
If you didn’t know any calculus. The derivative dx/dt is.1. of course. η(t1 ) = 0 and η(t2 ) = 0. (See Figure 4. so for the action I get this . And that must be true for any η at all.21. It can diﬀer in the second order. then the diﬀerence between that S and the action that we calculated for the path x(t)—to simplify the writing we can call it S—the diﬀerence of S and S must be zero in the ﬁrstorder approximation of small η. You would substitute x + h for x and expand out to the ﬁrst order in h . We take some trial path x(t) that diﬀers from the true path by a small amount which we will call η(t) (eta of t). x x(t) η(t) x(t) t Figure 4. With that condition.21).1. The idea is then that we substitute x(t) = x(t) + η(t) in the formula for the action: 2 m dx S= − V (x) dt. The method doesn’t mean anything unless you consider paths which all begin and end at the same two points—each path begins at a certain point at t1 and ends at a certain other point at t2 . we have speciﬁed our mathematical problem. Well. You could discuss what happens if you take f (x) and add a small amount h to x and argue that the correction to f (x) in the ﬁrst order in h must be zero ant the minimum. and those points and times are kept ﬁxed. So the deviations in our η have to be zero at each end. . the derivative of x(t) plus the derivative of η(t). 2 dt where I call the potential energy V (t). but in the ﬁrst order the diﬀerence must be zero. . you might do the same kind of thing to ﬁnd the minimum of an ordinary function f (x). not quite.48 4 Vector Valued Functions proportional to the square of the deviations from the true path. just as we are going to do with η. Now the idea is that if we calculate the action S for the path x(t). So we work it this way: We call x(t) (with an underline) the true path— the one we are trying to ﬁnd.
4.1C Principle of Least Action
49
expression:
t2
S=
t1
m 2
dx dη + dt dt
2
− V (x + η) dt.
Now I must write this out in more detail. For the squared term I get dx dt
2
+2
dx dη + dt dt
dη dt
2
But wait. I’m worrying about higher than the ﬁrst order, so I will take all the terms which involve η 2 and higher powers and put them in a little box called ‘second and higher order.’ From this term I get only second order, but there will be more from something else. So the kinetic energy part is m 2 dx dt
2
+m
dx dη + (second and higher order). dt dt
Now we need the potential V at x + η. I consider η small, so I can write V (x) as a Taylor series. It is approximately V (x); in the next approximation (from the ordinary nature of derivatives) the correction is η times the rate of change of V with respect to x, and so on: V (x + η) = V (x) + ηV (x) + η2 V (x) + . . . 2
I have written V for the derivative of V with respect to x in order to save writing. The term in η 2 and the ones beyond fall into the ‘second and higher order’ category and we don’t have to worry about them. Putting it all together,
t2
S=
t1
m 2
dx dt
2
− V (x) + m
dx dη dt dt
− ηV (x) + (second and higher order) dt. Now if we look carefully at the thing, we see that the ﬁrst two terms which I have arranged here correspond to the action S that I would have calculated with the true path x. The thing I want to concentrate on is the change in S—the diﬀerence between the S and the S that we would get for the right path. This diﬀerence we will write as δS, called the variation in S. Leaving out the ‘second and higher order’ terms, I have for δS
t2
δS =
t1
m
dx dη − ηV (x) dt. dt dt
Now the problem is this: Here is a certain integral. I don’t know what the x is yet, but I do know that no matter what η is, this integral must
50
4 Vector Valued Functions
be zero. Well, you think, the only way that that can happen is that what multiplies η must be zero. But what about the ﬁrst term with dη/dt? Well, after all, if η can be anything at all, its derivative is anything also, so you conclude that the coeﬃcient of dη/dt must also be zero. That isn’t quite right. It isn’t quite right because there is a connection between η and its derivative; they are not absolutely independent, because η(t) must be zero at both t1 and t2 . The method of solving all problems in the calculus of variations always uses the same general principle. You make the shift in the thing you want to vary (as we did by adding η); you look at the ﬁrstorder terms; then you always arrange things in such a form that you get an integral of the form ‘some kind of stuﬀ times the shift (η),’ but with no other derivatives (no dη/dt). It must be rearranged so it is always ‘something’ times η. You will see the great value of that in a minute. (There are formulas that tell you how to do this in some cases without actually calculating, but they are not general enough to be worth bothering about; the best way is to calculate it out this way.) How can I rearrange the term in dη/dt to make it have an η? I can do that by integrating by parts. It turns out that the whole trick of the calculus of variations consists of writing down the variation of S and then integrating by parts so that the derivatives of η disappear. It is always the same in every problem in which derivatives appear. You remember the general principle for integrating by parts. If you have any function f times dη/dt integrated with respect to t, you write down the derivative of ηf : d df dη (ηf ) = η + f . dt dt dt The integral you want is over the last term, so f dη dt = ηf − dt η df dt. dt
In our formula for δS, the function f is m times dx/dt; therefore, I have the following formula for δS. δS = m dx η(t) dt
t2
−
t1
t2 t1
d dt
m
dx dt
η(t)dt −
t2
V (x)η(t)dt.
t1
The ﬁrst term must be evaluated at the two limits t1 and t2 . Then I must have the integral from the rest of the integration by parts. The last term is brought down without change. Now comes something which always happens—the integrated part disappears. (In fact, if the integrand part does not disappear, you restate the principle, adding conditions to make sure it does!) We have already said that η must be zero at both ends of the path, because the principle is that the action is a minimum provided that the varied curve begins and ends
4.1C Principle of Least Action
51
at the chosen points. The condition is that η(t1 ) = 0 and η(t2 ) = 0. So the integrated term is zero. We collect the other terms together and obtain this: t2 d2 x δS = −m 2 − V (x) η(t)dt. dt t1 The variation in S is now the way we we wanted it—there is the stuﬀ in brackets, say F , all multiplied by η(t) and integrated from t1 to t2 . We have that an integral of something or other times η(t) is always zero: F (t)η(t)dt = 0. I have some function of t; I multiply it by η(t); and I integrate it from one end to the other. And no matter what the η is, I get zero. That means that the function F (t) is zero. That’s obvious, but anyway I’ll show you one kind of proof. Suppose that for η(t) I took something which was zero for all t except right near one particular value (See Figure 4.1.22). It stays zero until it gets to this t, then it blips up for a moment and blips right back down.
η(t)
t1
t2
t
Figure 4.1.22.
When we do the integral of this η times any function F , the only place that you get anything other than zero was where η(t) was blipping, and then you get the value of F at that place times the integral over the blip. The integral over the blip alone isn’t zero, but when multiplied by F it has to be; so the function F has to be zero everywhere. We see that if our integral is zero for any η, then the coeﬃcient of η must be zero. The action integral will be a minimum for the path that satisﬁes this complicated diﬀerential equation: −m d2 x − V (x) = 0. dt2
It’s not really so complicated; you have seen it before. It is just F = ma. The ﬁrst term is the mass times acceleration, and the second is the derivative of the potential energy, which is the force.
the change in time was zero. In fact. for a conservative system at least. . It is quite analogous to what we found for the ‘principle of least time’ which we discussed in optics. is zero. it is the same story. we have demonstrated that the principle of least action gives the right answer. See the “Feynman Lectures on Physics” for the remainder of the lecture. we said at ﬁrst it was ‘least’ time. it says that the path that has the minimum action is the one satisfying Newton’s law. One remark: I did not prove it was a minimum—maybe it’s a maximum. it doesn’t really have to be a minimum. There also.52 4 Vector Valued Functions So. which involves extending the above ideas to three dimensions and similar topics. The fundamental principle was that for any ﬁrstorder variation away from the optical path. It turned out. What we really mean by ‘least’ is that the ﬁrstorder change in the value of S. It is not necessarily a ‘minimum’. however. that there were situations in which it wasn’t the least time. when you change the path.
History. as in Figure 4. Mars. such as Kepler’s law on the relation between the period and the size of the orbit can be generalized to the case of elliptical orbits.3 Orbits of Planets 53 Supplement to §4.3 The Orbits of Planets The purpose of this supplement is to demonstrate that the motion of a body moving under the inﬂuence of Newton’s gravitational law—that is.1.3.1). the apparent motion of the planets. but some of the results. is the appearance of retrograde motion of the planets. which the Ptolemaic theory tried to explain.4. the diagram shows the paths traced by the planets as seen from the Earth. Recall. that one of the important observational points that was part of the Ptolemaic theory as well as being properly explained by the fact that planets move (to an excellent degree of approximation) in ellipses about the sun. Jupiter. for example. the inverse square force law—moves in a conic section. Venus. At this point it is a good idea to review some of the history given at the beginning of the text in addition to the relevant sections of the text (especially §4.3. and Saturn. Figure 4. In the text we dealt with circular orbits only.1. . The photograph shows the movements of Mercury.
3) r2 is a conic section. the Kepler problem is the following: Demonstrate that the solution r(t) of the equation GmM m¨ = r (4. We saw in the paragraph in §4. it follows that r lies in the plane perpendicular to the (constant) vector J. New York. (4.3.5) . The ﬁrst thing we observe is that any solution must lie in a plane. This states that the angular momentum of a solution is constant in time. Introducing Polar Coordinates.4) called the energy of the solution. we can assume that our orbit lies in a plane. Goodstein. Energy and Angular Momentum. is constant in time. The second fact that we will need is conservation of angular momentum. From the above consideration. one can also show that any conic section. By reversing the argument.54 4 Vector Valued Functions Methodology. which we can take to be the usual 10 The geometric approach. is also emphasized in the little book Feynman’s Lost Lecture: The Motion of Planets Around the Sun by David L.. This is simply because of conservation of angular mo˙ mentum: since J is a constant and since J = mr × r.3. with a suitable time parametrization. Other approaches. is also a solution.3 of the text entitled Conservation of Energy and Escaping the Earth’s Gravitational Field that any solution of equation (4.10 The Kepler Problem. which we will refer to as an orbit. Therefore. and to show it is a conic section. we assume that the Sun (of mass M ) is so massive that it remains ﬁxed at the origin and our planet (with mass m) revolves about the Sun according to Newton’s second law and moving in the ﬁeld determined by Newton’s Law of Gravitation. Orbits Lie in Planes. §4. Goodstein and Judith R.3.3. The procedure is to suppose that we have a solution.W. along with a lot of interesting history. namely the cross product ˙ J = mr × r is time independent for each orbit.1. Norton and Co. Newton’s original approach were much more geometrical. and in fact. which was given in Exercise 20. Our approach in this supplement is to use the laws of conservation of energy and conservation of angular momentum. As in the text. the quantity E= 1 ˙ m r 2 2 − GmM r (4. that is. W.3) obeys the law of conservation of energy. This means that we will focus on a combination of techniques using the basic laws of mechanics together with techniques from diﬀerential equations together with a couple of tricks.
3.11) which is a constant.3.7) can be written. as E= 1 2 J2 GmM − mr + ˙ 2 2mr2 r (4. we see that the velocity vector has components given by ˙ ˙ ˙ r = (r cos θ − rθ sin θ.3. ˙ (4. namely that for an orbit. Next notice that the energy equa˙ tion (4.3.4.3. We notice that the expression for J is closely related to the area element in polar coordinates.10) Thus.6) into conservation of energy gives E= 1 GmM ˙ m r2 + r2 θ2 − ˙ 2 r (4.10). Rewriting the Energy Equation.3.6) Substituting (4. .3. equal areas are swept out in equal times.9) (4. Also note that this works for any central force motion law as it depends only on conservation of angular momentum.12) It will be convenient to seek a description of the orbit in polar form as r = r(θ). namely dA = 1 r2 dθ. we get dA J = dt 2m (4. Diﬀerentiating in t. with θ eliminated using (4. r sin θ).3. the angular momentum is computed by taking the cross product ˙ of r and r: J=m i r cos θ ˙ r cos θ − rθ sin θ ˙ ˙ = mr2 θ k j r sin θ ˙ r sin θ + rθ cos θ ˙ k 0 0 (4. Thus. θ) in that plane as usual.7) Similarly.3 Orbits of Planets 55 xyplane.3. measuring area from a given reference θ0 . is constant in time. ˙ J = mr2 θ Kepler’s Second Law. In 2 fact.3.8) (4. r sin θ + rθ cos θ) ˙ ˙ and so one readily computes that ˙ r 2 ˙ = r2 + r2 θ2 . so that the components of the position vector r are given by r = (r cos θ. Let us introduce polar coordinates (r. we get Kepler’s second law.
This is the main purpose of doing the change of dependent variable from r to u. (4.3. the situation with the Kepler problem is one of these situations. A trick that works is to introduce the new dependent variable u = 1/r (of course one has to watch out for the possibility that the orbit passes through the origin. r= ˙ dr ˙ ˙ du θ = −r2 θ dθ dθ J du =− m dθ 2 and so from (4.3. by the chain rule.12) we ﬁnd that E= that is. First of all.56 4 Vector Valued Functions A Trick.3. the preceding equation and (4. we are going to eliminate the term linear in u in (4.3.16) . J2 (4. using the chain rule.) But watch the nice things that happen with this choice. so that (4. insight and luck. notice that.15) + v2 = β 2 .14) by completing the square. Completing the Square.10).14) Notice the interesting way that only u and its derivative with respect to θ appears and that the denominators containing r have been eliminated. 2Em = J2 du dθ 2 J2 2m du dθ + J 2 u2 − GmM u.3. du 1 dr =− 2 dθ r dθ and second. Let α = GM m2 /J 2 . in which case r would be zero.3. 2Em − α2 . J2 (4.14) becomes 2Em = J2 Now let v =u−α so that equation (4.3.15) becomes dv dθ where β2 = 2 du dθ 2 + u2 − 2αu. Solving diﬀerential equations to get explicit formulas often involves a combination of clever guesses. Next.3.13) + u2 − (4.3. 2m 2GM m2 u.
this means that 1 r (e cos(θ − θ0 ) + 1) = . Note that this reduces to what we saw in the text for circular orbits in the case of a circle (when a is the radius of the circle). where e is the eccentricity of the orbit.3. The quantity l = 1/α determines its scale and θ0 its orientation relative to the xyaxes. veriﬁed to be 2 1 1 v = u−1 α α + w2 = e2 . the ellipse gets a more and more elongated shape.17) β2 2Em = 2 2 − 1. 2mE The case e = 0 gives circular orbits. In terms of the original variable r. one ﬁnds that the period T of an elliptical orbit is related to the semimajor axis a by T 2π 2 = a3 GM which is Kepler’s third law.3. (4.17) is readily w = e cos(θ − θ0 ) where θ0 is a constant of integration. it is also the angle of the closest approach to the origin.18) in fact is the equation of a conic written in polar coordinates.16) becomes dw dθ where e2 = Solving the Equation. From Kepler’s second law and the fact that the area of an ellipse is πab. as e ranges from 0 to 1. In the case that 0 ≤ e < 1 (and E < 0) one has an ellipse of the form (x + ae)2 y2 + 2 =1 2 a b where a= and l GmM =− 2 1−e 2E J2 .3.18) α Orbits are Conics. . One also gets hyperbolic orbits if e > 1 and parabolic orbits in the special case when e = 1. 2 α α J The solution of the equation (4. Equation (4. which determines its shape.3. Thus.4. b2 = al = − Kepler’s 3rd Law. (4.3.3 Orbits of Planets 57 We make one more change of variables to w= so that (4.
t) = Dx [F(φ(x. . t) = DF(φ(x. With x as the initial condition. 0) = x . t) [time t] x [time t = 0] Figure 4.1. φ(x. t) = F(φ(x. It is convenient to give the unique solution through a given point at time 0 a special notation: φ(x. It is proved in courses on diﬀerential equations that φ is. the ﬂow of F. holding t ﬁxed. t)) ∂t φ(x. Let Dx denote diﬀerentiation with respect to x. t) is deﬁned by: ∂ φ(x.58 4 Vector Valued Functions Supplement to §4. we get Dx ∂ φ(x. line the flow x through passing φ(x. t) = the position of the point on the ﬂow line through the point x after time t has elapsed. yielding ∂ Dx φ(x. a diﬀerentiable function of x. in fact. which is regarded as the function of the variables x and t.4 Flows and the Geometry of the Divergence Flows of Vector Fields. t). By diﬀerentiating equation (1) with respect to x. ∂t The equality of mixed partial derivatives may be used on the lefthand side of this equation and the chain rule applied to the righthand side. t))]. t) of F. ∂t (2) . Alternatively.4. follow along the ﬂow line for a time period t until the new position φ(x. t))Dx φ(x. The deﬁnition of the ﬂow φ(x. (1) We call the mapping φ. t) is reached (see Figure 1).
See Exercises 3. t) · v1 v2 (t) = Dx φ(x. t) · v. It will be useful in our discussion of divergence and curl in the next section. v3 (t) = Dx φ(x. the ﬂow φ(x. v3 (t) span a parallelepiped P (t) that moves in time (see Figure 4. t) ≈ Dx φ(x. Recall that if v is a vector based at a point P1 and ending at P2 . The moving basis v1 (t). Both Dx F(φ) and Dx φ are 3 × 3 matrices since F and φ take values in R3 and are diﬀerentiated with respect to x ∈ R3 . they would be 2 × 2 matrices. We now study the geometric meaning of the divergence in more detail. This discussion depends on the concept of the ﬂow φ(x. and 5 below for the corresponding discussion of the curl. t) of a vector ﬁeld F given in the preceding paragraph. t)) denotes the derivative of F evaluated φ(x. t) · v2 .4. Equation (2). 0) = x for all x. φ is a diﬀerentiable map of R3 to R3 ). 4. t) · v3 (3) Since φ(x. When ε is small. t) is called the equation of the ﬁrst variation. a linear diﬀerential equation for Dx φ(x. . In summary. φ is a diﬀerentiable function of x (that is.4 Flows and the Geometry of the Divergence 59 where DF(φ(x. and v3 (0) = v3 . t) carries P (0) into some object. the vectors v1 (t). v2 (0) = v2 .4. it follows that v1 (0) = v1 .4. As time increases or decreases. The Geometry of the Divergence. v3 (t) given by v1 (t) = Dx φ(x. t) − φ(P1 . v2 = εj. v2 (t). These vectors span a parallelepiped P (0).2. v3 (t) and the associated parallepiped. For ﬁxed time. also emanating from x. v3 = εk. k emanating from x. t).2). v2 (t). Thus for ﬁxed time and small positive ε. the image of P (0) under φ can be approximated by its image under the derivative of φ with respect to x. P(t) P(0) v3 v3(t) flow line v2(t) x v1 v2 v1(t) Figure 4. Let ε > 0 be small and consider the basis vectors v1 = εi. Fix a point x and consider the three standard basis vectors i. v2 (t). P (0) is approximately carried into a parallelepiped spanned by the vectors v1 (t). then φ(P2 . so v = P2 − P1 . j. for vector ﬁelds in the plane.
it follows that Dx φ(x. equation (4) gives dV dv2 dv3 dv1 = · [v2 (t) × v3 (t)] + v1 (t) · × v3 (t) + v1 (t) · v2 (t) × dt dt dt dt dv1 dv2 dv3 = · v2 (t) × v3 (t)] + · [v3 (t) × v1 (t)] + · [v1 (t) × v2 (t) . substitution from formula (3) and the facts that v1 × v2 = εv3 . t) · vi ) dt (4) for i = 1.60 4 Vector Valued Functions Let the volume of P (t) be denoted by V(t). d vi (t) = DF(φ(x. v3 × v1 = εv2 gives dV dt = ε3 [DF(x)i] · i + ε3 [DF(x)j] · j + ε3 [DF(x)k] · k. t)) · (Dx φ(x. 3. t=0 and v2 × v3 = εv1 . Substituting these facts into equation (5) and dividing by V(0) = ε3 proves the theorem. . dt dt dt At t = 0. t=0 The volume V(t) is given by the triple product: V(t) = v1 (t) · [v2 (t) × v3 (t)]. the second and third terms of equation (5) are ε3 (∂F2 /∂y) and ε3 (∂F3 /∂z). The main geometric meaning of divergence is given by the following theorem. (5) Since F = F1 i + F2 j + F3 k. 0) = x. Using the diﬀerentiation rules of §4. div F(x) = Proof. we get [DF(x)i] · i = ∂F1 /∂x. 0) is the identity matrix. Similarly.1 and the identities v1 · [v2 × v3 ] = v2 · [v3 × v1 ] = v3 · [v1 × v2 ]. 1 d V(t) V(0) dt . Since φ(x. t=0 By equation (2) of the previous paragraph. Theorem. 2. so that evaluation at t = 0 gives d vi (t) dt = DF(x) · vi .
. v2 and v3 be any three noncoplanar (not necessarily orthonormal) vectors emanating from x that ﬂow according to the formula vi (t) = Dx φ(x. v3 (t) span a parallelepiped P (t) with volume V(t). t).4.e. “Rate” refers to the rate of change with respect to time as the volumes are transported by the ﬂow. (6) V(0) dt t=0 In other words.. If f (x.4 Flows and the Geometry of the Divergence 61 The reader who is familiar with a little more linear algebra can prove this generalization of the preceding Theorem:11 Let v1 . Dt ∂t Show that Df /Dt is the t derivative of f (φ(x. t)w. Exercises. (b) What is the corresponding property for Dx φ? 3. t). The vectors v1 (t). w(t) = Dx φ(0. t) (i. v2 (t). dv dt = Dx F(0) · v and t=0 dw dt = Dx F(0) · w. Then 1 dV = div F(x). s). t)v. so that at time t = 0 and at the origin 0 in R3 . Let v and w be two vectors emanating from the origin and let them be moved by the derivative of the ﬂow: v(t) = Dx φ(0. 2. 1. deﬁne the material derivative of f relative to a vector ﬁeld F as Df ∂f = + ∇f (x) · F. the t derivative of f transported by the ﬂow of F). i = 1. t=0 11 The reader will need to know how to write the matrix of a linear transformation with respect to a given basis and be familiar with the fact that the trace of a matrix is independent of the basis. t) is a realvalued function of x and t. the divergence of F at x is the rate at which the volumes change. t) · vi . prove the following property of the ﬂow φ(x. per unit volume. 3. t) of a vector ﬁeld F: φ(x. t + s) = φ(φ(x. (a) Assuming uniqueness of the ﬂow lines through a given point at a given time. 2.
2 2 2 5. z) = V(x. z) = −yi + xj. where r = xi+yj+zk. Let v = w × r.62 4 Vector Valued Functions Show that d v·w dt = [Dx F(0) · v] · w + v · [Dx F(0) · w] t=0 = [(Dx F(0) + [Dx F(0)]T )v] · w]. z). v is also a function of (x. y. and w31 = − (curl F)2 . (x2 + y 2 ) V(x. z) . Let w = 1 (∇ × F)(0). 2 2 In particular. 4. y. z) = and Y(x. y. Assume that axes are chosen so that w is 2 parallel to the z axis and points in the direction of k. y. so that v is the velocity ﬁeld of a rotation about the axis w with angular velocity ω = w and with curl v = 2w. W. for A = Dx F(0). z). y. S= and 1 [Dx F(0) + [Dx F(0)]T ] 2 1 [Dx F(0) + [Dx F(0)]T ]. 2 We call S the deformation matrix and W the rotation matrix. (x2 + y 2 )1/2 (a) Compute the divergence and curl of V. . Show that the derivative of v at the origin is given by 0 −ω 0 0 0 . Since r is a function of (x. and Y. Let V(x. Show that the entries of W are determined by W = 1 1 1 w12 = − (curl F)3 . W(x. y. Any matrix A can be written (uniquely) as the sum of a symmetric matrix (a matrix S is symmetric if S T = S) and an antisymmetric matrix (W T = −W ) as follows: A= 1 1 (A + AT ) + (A − AT ) = S + W. y. 6. Dv(0) = W = ω 0 0 0 Interpret the result. w23 = − (curl F)1 . z) .
4 Flows and the Geometry of the Divergence 63 (b) Find the ﬂow lines of V. and Y. v3 . v2 (t). and Y.4. W. let P (0) be the parallelepiped spanned by v1 . P (0) is carried by the ﬂow to an approximate parallelepiped spanned by v1 (t). t) be the ﬂow of a vector ﬁeld F. (a) How will a small paddle wheel behave in the ﬂow of each of V. v2 . Let x and t be ﬁxed. Argue that for small positive ε. v2 = εj and v3 = εk emanating from x. v3 (t) given by formula (3). W. 7. For small vectors v1 = εi. Let φ(x. .
64 4 Vector Valued Functions .
. .2 Alternative Deﬁnition of the Integral There is another approach to the deﬁnition of the integral based on step functions that you may want to mention or have some of your better students exposed to. a function is Riemann integrable when there is a transition point between the lower and upper sums. xi ) × (yj−1 . We say that a function g(x. . b] and c = y0 < y1 < y2 < . d] such that. < xn = b of the closed interval [a. y) deﬁned in R = [a. Alan Weinstein. 1982 Supplement to §5. b] × [c. < ym = d of the closed interval [c. in each of the mn open rectangles Rij = (xi−1 . yj ). For example. . . we present this (optional) deﬁnition ﬁrst.Page 65 5 Double and Triple Integrals Both the derivative and integral can be based on the notion of a transition point. d] is a step function provided there are partitions a = x0 < x1 < x2 < .
the volume of this part is kij (xi − xi−1 )(yi − yi−1 ). if g(x. 3].1. whether or not its values kij are all nonnegative. y)dx dy R of a step function over a rectangle in such a way that. the integral equals the volume of the region V under the graph. y) has a constant value kij . y) ≥ 0 or R.2. i=1. We denote this sum by n. The volume of V is the sum of all the kij ∆xi ∆yj as i ranges from 1 to n and j ranges from 1 to m (making nm terms in all).2.2. y)dx dy = R i=1. The graph of a generic step function is shown in Figure 5. Calculate the integral of g over the rectangle R = [0. Let g take values on the rectangles as shown in Figure 5.j=1 Using this geometric guide. Since the part of V lying over Rij has height kij and base area (xi − xi−1 )(yi − yi−1 ). 5] × [0.j=1 kij ∆xi ∆yj for every step function g.66 5 Double and Triple Integrals the function g(x.m . We will deﬁne the integral g(x.2. . The function g is a step function since it is constant on each subrectangle. or kij ∆xi ∆yj where ∆xi = xi − xi−1 and ∆yi = yi − yi−1 as in onevariable calculus. Example 1. we deﬁne n.m g(x. z Graph of g c = y0 y1 y2 y3 = d y x0 = a x1 b = x2 x Figure 5.1.
2 Alternative Deﬁnition of the Integral 67 y 3 2 1 –8 2 Figure 5. Solution. y) for all (x. y) in R. y) ≤ f2 (x. it can be shown that there is just one . If f1 (x. y) ≤ h(x. Alternative Deﬁnition of the Double Integral. When this condition holds. we use a comparison method whose origins go back as far as Archimedes. if f is any function which we wish to integrate over R. and if g and h are step functions on R such that g(x. This condition on the integral actually becomes a deﬁnition if we rephrase it as follows. To deﬁne their integrals. Thus. Find 4 6 –1 3 2 5 x D g(x.2. y) ≤ f (x. y) and f2 (x. y) and h(x. y) for all (x. y)dx dy if g take the values shown.2. The integral of g is the sum of the values of g times the areas of the rectangles: g(x. y) with g(x. y)dx dy is less than . y) in R and such that the diﬀerence h(x. then any reasonable deﬁnition of the double integral of f2 should be larger than that of f1 if f1 (x. y) ≤ h(x. Of course. y)dx dy which we are trying to deﬁne must lie between the numbers R g(x. y) everywhere on R. y) are any two functions deﬁned on the rectangle R. the functions which we want to integrate are usually not step functions. then the number R f (x. y)dx dy = − 8 × 2 + 2 × 3 + 6 × 2 R + 3 × 3 + 4 × 2 − 1 × 3 = 16. y) ≤ f (x. We shall call such step functions g and h “surrounding functions”. y)dx dy and R h(x.5. y)dx dy − R R g(x. A function f deﬁned on a rectangle R is said to be integrable on R if for any positive number . there exist step functions g(x. y)dx dy which have already been deﬁned.
68 5 Double and Triple Integrals number I with the property that g(x. sj ). and let f (x. Let g be a step function. Let R be the rectangle 0 ≤ x ≤ 2. 3]. h(x. they may be added by ﬁrst adding along rows and then adding up the subtotals. Example 2. y)dx dy for surrounding functions g and h as above. y) has integral 12 × 4 = 48. so we get only the crude estimate R f (x. The constant function h(x. we get f (x. 1] × [2. y) = 2 on R1 . Let h be the function given by taking the maximum value of f on each subrectangle (evaluated at the upper righthand corner). 8 on R3 . The number I is called the integral of f on R and is denoted by R f (x. y)dx dy ≤ 48. 2]. 2].3): . If the summands kij ∆ti ∆sj are laid out in a rectangular array. that is. The upper and lower sums approach can also be used to give proof of the reduction to iterated integrals. y) = 12 ≥ f (x. y)dx dy ≤ 25. y)dx dy ≤ 25. 2] × [2. 1] × [1. divide R into four pieces: R1 = [0. y) = kij on the rectangle (ti−1 .2. y)dx dy = 2 × 1 + 3 × 1 + 8 × 1 + 12 × 1 = 25. 3 on R2 . the integral of h is h(x. To get a better one. R Solution.j=1 kij ∆ti ∆sj . Choose a step function h(x. R Alternative Proof of Reduction to Iterated Integrals. y)dx dy. Therefore. 1 ≤ y ≤ 3. and 12 on R4 . We ﬁrst treat step functions. R3 = [1. y) to show that f (x. y) ≥ f (x. with g(x. We give this (optional) proof here.m g(x. so that n. y) = x2 y. y)dx dy ≤ I ≤ R R h(x. ti ) × (sj−1 . R Since h ≥ f. y)dx dy = D i=1. just as in the text (see Figure 5. R4 = [1. R2 = [0. 2] × [1. 3].
the integral a g(x. . kn1 ∆tn ∆s2 −→ i=1 n ki1 ∆ti ki2 ∆ti i=1 ∆s1 ∆s2 k12 ∆t1 ∆s2 k22 ∆t2 ∆s2 . n k1m ∆t1 ∆sm k2m ∆t2 ∆sm . y)dx dy = D a c g(x. . y) is a b step function of x. Thus.2. To carry out our program. knm ∆tn ∆sm −→ i=1 m kim ∆ti n ∆sm kij ∆ti j=1 i=1 ∆sj Figure 5. The theorem is therefore true for step functions.5. . y) (1) . n The coeﬃcient of ∆sj in the sum over the jth row. y)dx dy. is equal g(x. Now let f be integrable on D = [a. y)dx dy = j=1 i=1 kij ∆ti ∆sj = D g(x. . . j=1 kij ∆ti . we obtain b d g(x. and its integral with respect to y is the sum: d c a b m n to b a g(x. for y ﬁxed. y)dx is a step function of y. Reduction to iterated integrals. y)dx for any y with sj−1 < y < sj . . by summing ﬁrst over columns and then over rows. we will show that every lower sum for f on D is less than or equal to S0 . so S0 must be the integral of f over D. . while every upper sum is greater than or equal to S0 . y)dy dx. . g(x. since. . b]×[c.2 Alternative Deﬁnition of the Integral 69 n k11 ∆t1 ∆s1 k21 ∆t2 ∆s1 . . let g be any step function such that g(x. . Denoting this integral by S0 . kn2 ∆tn ∆s2 −→ . d] and assume that the iterated b d integral a c f (x. .3. y)dy dx exists. y) ≤ f (x. . . Similarly.
The proof that every upper sum is greater than or equal to S0 is similar. it follows from the ﬁrst part of this proof that the lefthand side of (3) is equal to the lower sum D g(x. Integrating (2) with respect to x and applying monotonicity once more gives b a c d b d g(x.70 5 Double and Triple Integrals for all (x. b]. . y)dy dx. (3) Since g is a step function. the righthand side of (3) is just S0 . y)dy dx ≤ a c f (x. y) in D. we obtain d d g(x. y)dy (2) for all x in [a. y)dy ≤ c c f (x. so we have shown that every lower sum is less than or equal to S0 . Integrating (1) with respect to y and using “monotonicity” of the onevariable integral. and so we are done. y)dx dy.
6 Technical Integration Theorems 71 §5. we can prove a special case of it. or W.6 Technical Integration Theorems This section provides the main ideas of the proofs of the existence and additivity of the integral that were stated in §5. then f (x) − f (y) < ε. Deﬁnition. suﬃcient for many situations relevant to this text. which is.3 that a set D ⊂ Rn is bounded if there exists a number M > 0 such that x ≤ M for all x ∈ D. The main diﬀerence between continuity and uniform continuity is that for continuity. 1 The proof can be found in texts on mathematical analysis. Thus a set is bounded if it can be strictly contained in some (large) ball. whereas in uniform continuity δ depends only on ε. δ cannot be chosen independently of the point of the domain (the x0 in the deﬁnition). The Uniform Continuity Principle. for example. 1976. Let D ⊂ Rn and f : D → R. See. y ∈ D and x−y < δ. .2 of the text. McGrawHill. Thus any uniformly continuous function is also continuous. Hoﬀman. The function f is said to be uniformly continuous on D if for every number ε > 0 there is a δ > 0 such that whenever x. 2nd ed. These proofs require more advanced concepts than those needed for the rest of this chapter. The deﬁnition of uniform continuity states explicitly that once you are given an ε > 0. are called upon here.. Recall from §3. 3rd ed. Elementary Classical Analysis. a δ can be found independent of any point of D. The distinction between the notions of continuity and uniform continuity can be rephrased: For a function f that is continuous but not uniformly continuous. Recall that f is said to be continuous at x0 ∈ D provided that for all ε > 0 there is a number δ > 0 such that if x ∈ D and x − x0 < δ. Theorem. Principles of Mathematical Analysis.5. Every function that is continuous on a closed and bounded set D in Rn is uniformly continuous on D. New York.. both of which are usually treated more fully in a juniorlevel course in mathematical analysis or realvariable theory. Uniform Continuity. The proof of this theorem will take us too far aﬁeld. δ can depend on x0 as well as ε. Freeman. A set is closed if it contains all its boundary points. Rudin. in fact. 1993. Marsden and M. then f (x) − f (x0 ) < ε. J. We say f is continuous on D if it is continuous at each point of D. The next theorem states that under some conditions a continuous function is actually uniformly continuous. An example of a function that is continuous but not uniformly continuous is given in Exercise 2 in this supplement.1 however. The notions of uniform continuity and the completeness of the real numbers. New York.
it should be enough information to ensure its convergence. We assume f is diﬀerentiable (on an open set containing D) and that ∇f (x) ≤ C for a constant C. f (x) − f (y) ≤ ∇f (c(t0 )) x − y ≤ C x − y . there is a constant C > 0 such that df (x)/dx ≤ C for all x in (a. and then take as an axiom of the real number system that all such sequences . and that df /dx is bounded (that is. We shall deﬁne a class of sequences called Cauchy sequences. b). Completeness. joining them also lies in D. Then by the mean value theorem. Let us assume that D = [a. that is. . we use the mean value theorem as follows: Let ε > 0 be given and let x and y lie on D. (Note that δ depends on neither x nor y. that df /dx exists on the open interval (a. Then.. we know no more about Sn than that it is the Riemann sum of a (say. Thus by the CauchySchwartz inequality. . and. 0 ≤ t ≤ 1. Then the mean value theorem applied to the function h(t) = f (c(t)) gives a point t0 such that h(1) − h(0) = [h (t0 )][1 − 0] or f (x) − f (y) − h (t0 ) = ∇f (c(t0 )) · c (t0 ) = ∇f (c(t0 )) · (x − y) by the chain rule. . n = 1. In the deﬁnition of Riemann sums we obtained a sequence of numbers {Sn }. for any two points x. By the assumed boundedness of the derivative. that f : D → R is continuous. Thus we must determine a property for sequences that guarantees their convergence.72 5 Double and Triple Integrals Proof of a special case. we can let δ = ε/C. continuous) function. b)). the line segment c(t) = tx + (1 − t)y. as above. but how can we obtain such a limit? In the abstract setting. which is a crucial part of the deﬁnition. y in D. f (x) − f (y) ≤ Cx − y. f (x) − f (y) = f (c)(x − y) for some c between x and y. It would be nice if we could say that this sequence of numbers converges to S (or has a limit S). b] is a closed interval on the line. then f (x) − f (y) < C ε = ε. To show that these conditions imply f is uniformly continuous. although this has not yet been proved. If x − y < δ. We now move on to the notion of a Cauchy sequence of real numbers. Let δ = ε/C. C Thus f is uniformly continuous. given ε > 0.) This proof also works for regions in Rn that are convex.
Deﬁnition. This book was a basic work on the foundations of analysis. If a sequence Sn converges to a limit S. is said to satisfy the Cauchy criterion if for every ε > 0 there exists an N such that for all m. {Sn } converges to some limit S. choose N1 such that for n ≥ N1 . which would prove that continuous functions on rectangles are integrable. although it would be considered somewhat loosely written by the standards of our time. Then {Sn } converges to some number S. The completeness axiom asserts that the converse is true as well: Completeness Axiom of the Real Number. sometimes use diﬀerent axioms. In demonstrating this. Then if n. Sn − Sm  = Sn − S + S − Sm  ≤ Sn − S + S − Sm  < ε ε + = ε. nor could he have had one. m ≥ N1 . He did not have a proof. such as those mentioned in the preceding footnote. Every Cauchy sequence Historical Note. 2 Texts on mathematical analysis. . one of the greatest mathematicians of all time. we must show that {Sn } is a Cauchy sequence. . our completeness axiom becomes a theorem. we use the deﬁnition: For every ε > 0 there is an N such that for all n ≥ N. such as the least upper bound property. published in 1821. To see this. Given ε > 0. Sn − S < ε/2 (use the deﬁnition with ε/2 in place of ε). In such a setting. We shall say more this shortly. since such a proof depends on the rigorous development of the real number system that was achieved only in 1872 by the German mathematician Georg Cantor (18451918). n = 1. . The determination in the nineteenth century that such an axiom was necessary for the foundations of calculus was a major breakthrough in the history of mathematics and paved the way for the modern rigorous approach to mathematical analysis. It is now clear what we must do to ensure that the Riemann sums {Sn } of.5.6 Technical Integration Theorems 73 converge to a limit2 . Sn − S < ε. we have Sn − Sm  < ε. A sequence of real numbers {Sn }. .. Let f be a continuous function on a rectangle R in the plane. The integrability of continuous functions will be a consequence of the following two lemmas. say. 2 2 which proves our contention. Lemma 1. a continuous function on a rectangle converge to some limit S. we use the uniform continuity principle. and let {Sn } be a sequence of Riemann sums for f . n ≥ N . deﬁned what we now call Cauchy sequences in his Cours d’analyse. then Sn is a Cauchy sequence. Augustin Louis Cauchy (1789–1857). Cauchy knew that a convergent sequence was “Cauchy” and remarked that a Cauchy sequence converges.
1). By the uniform continuity principle.k=0 b−a d−c . Note that each subrectangle of the mnth partition is a subrectangle of both the mth and the nth regular partitions (see Figure 5. Fix m. we must show that given ε > 0 there exists an N such that for all n. This shows that {Sn } is a Cauchy sequence and hence converges. Then Smn = r. n ≥ N . then f (x) − f (y) < ε/[2 area(R)] (the quantity ε/[2 area(R)] is used in place of ε in the deﬁnition). d]. x − y < δ. Sn − Sm  ≤ ε. The sequence {Sn } is determined only by the selection of the points cjk . m ≥ N. y are points in the same subrectangle. b] × [c. ˜ Let us denote the subrectangles in the mnth subdivision by Rrt and ˜ rt ⊂ Rjk for some jk. y ∈ R. Given a rectangle R ⊂ R2 . and those in the nth subdivision by Rjk . For the purpose of the proof we shall introduce a slightly more complicated but very precise notation: we set ∆xn = With this notation we have n−1 b−a d−c and ∆y n = . xj+1 ] × [yk . (1 ) .74 5 Double and Triple Integrals Proof.6.k f (cjk )∆xn ∆y n . a = x0 < x1 < · · · < xn = b. Consider the mnth = (m times n)th regular partition of R. (1) To show that {Sn } satisﬁes the Cauchy criterion. Thus given ε > 0 there exists a δ > 0 such that when x. Let N be so large that for any m ≥ N the diameter (length of a diagonal) of any subrectangle Rjk in the mth regular partition of R is less than δ.2 of the text. ∆y = yk+1 − yk = . f is uniformly continuous on R. n n Sn = j. c ˜ where crt is a point in the rtth subrectangle. R = [a.t f (˜rt )∆xmn ∆y mn . Thus if x. we have the regular partition of R. yk+1 ]. c = y0 < y1 < · · · < yn = d. discussed in §5. We will show that Sn − Sm  < ε. n n n−1 f (cjk )∆x∆y. Recall that ∆x = xj+1 − xj = and Sn = j. Thus each R hence we can rewrite formula (1) as n−1 Sn = j. we will have the inequality f (x − f (y)) < ε/[2 area(R)]. where cjk is an arbitrarily chosen point in Rjk = [xj .k ˜ Rrt ⊂Rjk f (cjk )∆xmn ∆y mn .
k Rrt ⊂Rjk [f (cjk )∆xmn ∆y mn − f (˜rt )∆xmn ∆y mn ] c ≤ ˜ j.6 Technical Integration Theorems 75 Figure 5.t f (˜rt )∆xmn ∆y mn .k Rrt ⊂Rjk f (cjk ) − f (˜rt )∆xmn ∆y mn . We also have the identity mn−1 Smn = r. The shaded box shows a subrectangle in the mnth partition. c (2 ) where in equation (2 ) we are ﬁrst summing over those subrectangles in the mnth partition contained in a ﬁxed Rjk and then summing over j.1. and the darkly outlined box. ˜ Rrt ⊂Rjk where the sum is taken over all subrectangles in the mnth subdivision contained in a ﬁxed rectangle Rjk in the nth subdivision. Subtracting equation (2 ) from equation (1 ).k Rrt ⊂Rjk f (˜rt )∆xmn ∆y mn . c (2) This relation can also be rewritten as Smn = ˜ j. we get Sn − Smn  = ˜ j.6. c . Here we are using the fact that f (cjk )∆xmn ∆y mn = f (cjk )∆xn ∆y n .5. k. a subrectangle in the mth partition.
We choose N so large that whenever n ≥ N the diameter of each subrectangle in the nth regular partition is less than δ. Since Sn − Sm  = Sn − Smn + Smn − Sm  ≤ Sn − Smn  + Smn − Sm  < ε for m.k f (cjk )∆xn ∆y n − j. We want to show that S = S ∗ and shall do this by showing that given any ε > 0. jk then f (cjk ) − f (c∗ ) < ε/[3 area(R)]. and consec quently the above inequality becomes ε ε Sn − Smn  ≤ ∆xmn ∆y mn = . 3 ≤ j. say S ∗ . By Lemma 1 we know that {Sn } converges to some number ∗ S and {Sn } must also converge to some number. we can assume that N has been chosen so large that n ≥ N implies that Sn − S < ε/3 and ∗ Sn − S ∗  < ε/3. To start.76 5 Double and Triple Integrals By our choice of δ and N .k f (c∗ )∆xn ∆y n jk < ε .k f (cjk ) − f (c∗ )∆xn ∆y n jk We now write ∗ ∗ ∗ ∗ S − S ∗  = S − Sn + Sn − Sn + Sn − S ∗  ≤ S − Sn  + Sn − Sn  + Sn − S < ε . In order to show that a continuous function on a rectangle R is integrable. given ε > 0. f (cjk ) − f (˜rt ) < ε/[2 area(R)]. Lemma 2. there exits a δ such that f (x) − f (y) < ε/[3 area(R)] whenever x − y < δ. n ≥ N . for n ≥ N we know by uniform continuity that if cjk and c∗ are points in the same subrectangle Rjk of the nth partition. The limit S in Lemma 1 does not depend on the choice of points cjk . Consequently. S −S ∗  < ε. Also. we have shown {Sn } satisﬁes the Cauchy criterion and thus has a limit S. ∗ Since limn→∞ Sn = S and limn→∞ Sn = S ∗ . we must further demonstrate that the limit S we obtained in Lemma 1 is independent of the choices of the points cjk . which implies that S must be equal to S ∗ (why?). we know that f is uniformly continuous on R. 2 area (R) 2 ˜ j.k Rrt ⊂Rjk Thus. Suppose we have two sequences of Riemann sums {Sn } and {Sn } ∗ obtained by selecting two diﬀerent sets of points. We have already remarked that each Riemann sum depends on the selection of a collection of points cjk . Thus jk ∗ Sn − Sn  = j. ∗ Proof. Sn − Smn  < ε/2 and similarly one shows that Sn − Smn  < ε/2. say cjk and cjk in each nth partition.
6.2 of the Text. The proof depends on the ideas that have already been presented in the proof of Theorem 1. Any continuous function deﬁned on a rectangle R is integrable. Cauchy presented the ﬁrst published proof of this theorem in his r´sum´ of 1823. Thus all three integrals in equation (3) exist. its additivity. we shall prove it only in the case in which f is continuous. This latter approach to integration theory is generally studied in graduate courses in mathematics. In this paper he ﬁrst treats continuous functions (as we are doing now). since it lacked the notion of uniform continuity. but on an interval [a. namely. Let R1 and R2 be two disjoint rectangles (rectangles whose intersection contains no rectangle) such that Q = R1 ∪ R2 is again a rectangle as in Figure 5. Historical Note. His approach.2 is left to the reader in Exercises 4 to 6 at the end of this section.2.5. Theorem. However. The sums are probably named after him because he developed a theoretical approach to the study of integration in a fundamental paper on trigonometric series in 1854. If f is a function that is continuous over Q and hence over each Ri . and it is necessary only to establish equality. (The proof is essentially the same. Additivity of the Integral. . his proof was not rigorous.6 Technical Integration Theorems 77 and so the lemma is proved. and R2 follows from Theorem 1. which was not available at that time. The notion of a Riemann sum Sn for a function f certainly predates Bernhard Riemann (1826–1866).2 of the main text: Theorem 1 of §5. was to last more than half a century until it was augmented by the theory Lebesque presented to the mathematical world in 1902.) However. (3) Proof. Putting lemmas 1 and 2 together proves Theorem 1 of §5. The main ideas are essentially contained in the proof of Theorem 1. b]. in which he points out the need to prove the e e existence of the integral as a limit of a sum. although later generalized by Darboux (1875) and Stieltjes (1894). The proof of Theorem 2 of §5. Additivity Theorem. because of some technical diﬃculties in establishing this result in its full generality. then f= Q R1 f+ R2 f. Our next goal will be to present a proof of property (iv) of the integral from §5. R1 . The fact that f is integrable over Q.2.
Elements of a regular partition of R1 and R2 .2). respectively.78 5 Double and Triple Integrals y Q R1 d R2 Rjk 1 2 Rjk c a b1 b x Figure 5. ∆xn = . Let S i = limn→∞ Sn . where 1 2 i = 1. 2.k 2 Sn = j. R2 . d] (see Figure 5. Without loss of generality we can assume that R1 = [a. and Q.6. and S = limn→∞ Sn . Let N be i so big that for all n ≥ N. c2 . and cjk are points in the jkth subrectangle of the nth jk jk i regular partition of R1 . i = 1. Again we must develop some notation. f (x) − f (y) < ε.2. and if . Sn − S < ε/3. Sn − S i  < ε/3.6. It must be shown that S = S + S . and ∆y n = . which we will accomplish by showing that for arbitrary ε > 0. 2. ∆xn = . d] and R2 = [b1 . By the uniform continuity of f on Q we know that given ε > 0 there is a δ > 0 such that whenever x − y < δ. 2 n n n n f (c1 )∆xn ∆y n jk 1 f (c2 )∆xn ∆y n jk 2 f (cjk )∆xn ∆y n j. S − S 1 − S 2  < ε.k b1 − a b − b1 b−a d−c .k (4) (5) (6) Sn = where c1 . b] × [c. b1 ] × [c. Let ∆xn = 1 Let 1 Sn = j.
5. Rjk . y d Rjk c a (a) b x y d ~ R2 Rαβ jk c a b1 (b) b x Figure 5. .6.3. . Let us consider the nth regular partition of R1 . see Figure 5.6.6. β = 1. R2 .6. we get a new collection of rectangles.3(a)]. m.3(b). . respectively [see Figures 5. say Rαβ . . y are any two points in any subrectangle of the nth partition of either R1 .6.2 and 5. If we superimpose the subdivision of Q on the nth subdivisions of R1 ˜ and R2 . These form a collection of subrectangles 1 2 that we shall denote by Rjk . where m > n. .6 Technical Integration Theorems 79 x. n and α = 1.2 and 5. .6. Rjk . . . ˜ Each Rαβ is contained in some subrectangle Rjk of Q and in some subrectangle of the nth partition of either R1 or R2 . or Q then f (x) − f (y) < ε/[3 area(Q)]. (b) The vertical and horizontal lines of this subdivision are obtained by taking the union of the vertical and horizontal of Figures 5. (a) A regular partition of Q.3(a). . and Q. Consider equalities (4). R2 .
β ˜ Rαβ ⊂R1 ˜ [f (c∗ ) − f (˜αβ )] area(Rαβ ) c αβ + α. and (6) above. Thus Sn −Sn −Sn  < ε/3 for ≥ N . α.β ˜ Rαβ ⊂Ri means that the summation is taken over those α’s and β’s such that the ˜ corresponding rectangle Rαβ is contained in the rectangle Ri . c i ˜ ˜ where cαβ = ci if Rαβ ⊂ Rjk . These can be rewritten as i Sn = ˜ j.β ˜ Rαβ ⊂R2 ε ˜ area (Rαβ ) < .β ˜ Rαβ ⊂Ri ˜ f (˜αβ ) area (Rαβ ).β ˜ Rαβ ⊂R2 ˜ [f (c∗ ) − f (˜αβ )] area(Rαβ ) c αβ ≤ ε 3 area Q + ˜ area (Rαβ ) α.β For the reader encountering such index notation for the ﬁrst time. αβ ˜ where c∗ = cjk if Rαβ ⊂ Rjk .β ˜ Rαβ ⊂R2 ˜ f (c∗ ) area (Rαβ ). But ε ε ε 1 2 S − Sn  < . 3 1 2 In this step we used the uniform continuity of f .β ˜ f (c∗ ) area (Rαβ ).β ˜ Rαβ ⊂R1 ε 3 area Q α. 2. 3 3 3 . it follows that 1 2 Sn − Sn − Sn  ≤ α. i = 1. αβ From these representations and the triangle inequality. Sn − S 1  < and Sn − S 2  < .80 5 Double and Triple Integrals (5).β ˜ Rαβ ⊂R1 ˜ f (c∗ ) area (Rαβ ) + αβ α. and αβ Sn = ˜ j.k Rαβ ⊂Rjk ˜ f (cjk ) area (Rαβ ) = α. we point out that α. Now the sum for Sn can be split into two parts: Sn = α.k Rαβ ⊂Ri ˜ f (ci ) area (Rαβ ) = jk α.
d]. . b] → R. Let > 0. b] × [c.6 Technical Integration Theorems 81 As in Lemma 2. d]. (a) Show that f is integrable over [(a + b)/2. Solution. (a) Let f be the function on the halfopen interval (0. Exercises 4 to 6 are intended to give a proof of Theorem 2 of §5. < x2n the corresponding partition. 2. Show that f is integrable over [(a + b)/N. Let > 0 be any positive number. Show that f is continuous at every point of (0.2. Show that C can be placed in a ﬁnite union of boxes Bi = [ai . b] × [c. f (xi )) with width 1/n and height /4. Similarly if C intersects a vertical side of Bi it must do so in a portion which is in the complement of Bi−1 ∪ Bi+1 . d] and f be a bounded function that is integrable over R.5. then f (x) − f (ω) < /32. (b) Generalize this example to R2 . with 0 < δ < 1 such that if x − ω < δ. 1] but not uniformly continuous. (b) Let N be any positive integer. Example 6. (We see why the “32” as we proceed) — of course in reality one does a sketch of the idea ﬁrst and through a trial run one “discovers” that the right number to put here is indeed 32. If the graph touches the top edge of some Box Bi . 4n 2 2n It remains only to check that C does not contain a boundary point of ∪i Bi . then there is some (x. f (x)) Bi such that f (x) − f (xi ) ≥ /8 ( /8 is half the height of Bi ). Since f is uniformly continuous there exists a δ. 1] deﬁned by f (x) = 1/x. Let Bi be the rectangle centered at (xi . Show that if a and b are two numbers such that for any ε > 0. But by uniform continuity f (x) − f (xi ) < /32 a contradiction. Let n > 1/δ and subdivide the interval [a. 1. This again contradicts uniform continuity. 3. Let C be the graph of a continuous function φ : [a. bi ]×[ci . which completes the proof. . Let R be the rectangle [a. b] into 2n equal parts with x0 < x1 < . a−b < ε. Exercises. di ] such that C does not contain a boundary point of ∪Bi and such that area(Bi ) ≤ . an application of the triangle inequality shows that S − S 1 − S 2  < ε. then a = b. There are (2n + 1) such rectangles for a total area of (2n + 1) = + < . b] × [c. The area of each rectangle is /4n. .
Let R and B be rectangles and let B ⊂ R. Suppose that f : R → R is bounded and continuous except on C. Consider the nth regular partition of R and let bn be the sum of the areas of all rectangles in the partition that have a nonempty intersection with B. di ] such that C does not contain a boundary point of ∪Bi and such that Σ area (Bi ) ≤ ε. bi ]×[ci . . Show that C can be placed in a ﬁnite union of boxes Bi = [ai . b] → R. b] → R is a continuous function. Let C be the graph of a continuous function φ : [a.82 5 Double and Triple Integrals 4. 6. then φ is bounded. (a) Use the uniform continuity principle to show that if φ : [a. b] × [c. 7. Show that limn→∞ bn = area(B). (Hint: Use the uniform continuity principle presented in this section. (b) Generalize part (a) to show that a given continuous function f : [a.) 5. Use Exercises 4 and 5 above and the techniques used in the proof of Theorem 1 of this section to show that f is integrable over R. d] → R is bounded. then f is bounded. Let ε > 0 be any positive number. Let R be a rectangle and C ⊂ R the graph of a continuous function φ. (c) Generalize part (b) still further to show that if f : D → R is a continuous function on a closed and bounded set D ⊂ Rn .
He prepared three lectures. at ﬁrst approximation in a geodesic coordinate system such a metric is ﬂat Euclidean. in the same way that a curved surface up to higherorder terms looks like its tangent plane. The second part of Riemann’s lecture posed deep questions about the relationship of geometry to the world we live in. now called geodesics. became a classic of mathematics. Gauss (his advi¨ sor) chose the lecture on geometry. Riemann’s lecture Uber die Hypothesen welche der Geometrie zu Grunde liegen (On the hypotheses that lie at the foundations of geometry). He asked what the dimension . There were two parts to Riemann’s lecture. delivered on 10 June 1854. In fact the main point of this part of Riemann’s lecture was the deﬁnition of the curvature tensor .Page 83 6 Integrals over Curves and Surfaces To complete his Habilitation. Freudenthal writes: It possesses shortest lines. In the ﬁrst part he posed the problem of how to deﬁne an ndimensional space and ended up giving a deﬁnition of what today we call a Riemannian space. Gauss had to choose one of the three for Riemann to deliver and. which resemble ordinary straight lines. Riemann had to give a lecture. In fact. two on electricity and one on geometry. Beings living on the surface may discover the curvature of their world and compute it at any point as a consequence of observed deviations from Pythagoras’ theorem. against Riemann’s expectations.
y) = x2 + y 2 over the rectangle R = [0. 1]. It was not fully understood until sixty years later. From the Riemann website http://wwwgap. R where R = [0. only Gauss was able to appreciate the depth of Riemann’s thoughts. Einstein found the frame to ﬁt his physical ideas. . .1. Returning to the faculty meeting. In the mathematical apparatus developed from Riemann’s address. The lecture exceeded all his expectations and greatly surprised him. Solution.ac. Evaluate x2 + y 2 dxdy... that is. his cosmology . and cosmogony : and the spirit of Riemann’s address was just what physics needed: the metric structure determined by data.1. The lecture was too far ahead of its time to be appreciated by most scientists of that time. Freudenthal writes: The general theory of relativity splendidly justiﬁed his work. 1] × [0. it equals the volume of the threedimensional region shown in Figure 6. This double integral is equal to the volume of the region under the graph of the function f (x.html Supplement 6. he spoke with the greatest praise and rare enthusiasm to Wilhelm Weber about the depth of the thoughts that Riemann had presented.uk/~history/ Mathematicians/Riemann.stand.2A A Challenging Example Example. 1]. 1] × [0.dcs.84 6 Integrals over Curves and Surfaces of real space was and what geometry described real space. Monastyrsky writes: Among Riemann’s audience.
1. However.1.0) (0.1. Since the integrand is a simple function of r2 = x2 + y 2 .1) y x Figure 6.6. unfortunately. To apply Theorem 2 of the text with polar coordinates. This will result in a simpliﬁcation of the integrand but.1) T x ∗ Figure 6. As it stands. refer to Figure 6. The polarcoordinate transformation takes D1 to the triangle T1 and ∗ D2 to T2 .2.1) T2 r θ T1 (1. the simpliﬁcation is suﬃcient to enable us to evaluate the integral.0) (1. not in the domain of the integration.2. Volume of the region under z = x2 + y 2 and over R = [0. . 1]. this integral is diﬃcult to evaluate. 1] × [0. we might try a change of variables to polar coordinates. θ θ π 2 π 4 * D1 π 2 π 4 * D2 r = csc θ r r r = sec θ y T (0.1.2A Challenging Example 85 z R (1.
From the symmetry of z = x2 + y 2 on R. then with respect to x). 0 Consulting a table of integrals (see the back of the book) to ﬁnd sec3 x dx. we can see that x2 + y 2 dxdy = 2 x2 + y 2 dxdy. we ﬁnd sec θ dθ = 0 √ 1 1 π/4 [log  sec θ + tan θ]0 = log(1 + 2). we obtain 3 √ √ √ 2 1 1 1√ r2 drdθ = + log(1 + 2) = [ 2 + log(1 + 2)]. and so we can apply Theorem 2. Next we use iterated integration to obtain π/4 ∗ D1 sec θ r2 drdθ = 0 0 r2 dr dθ = 1 3 π/4 sec3 θ dθ. 2 2 Combining these results and recalling the factor 1 . 3 Instead of using the substitution T . The transformation T is onetoone except when r = 0. + 2 2 0 2 2 0 0 0 Consulting the table again for 1 2 π/4 sec x dx. for D2 we have 4 π ≤ θ ≤ 1 π and 0 ≤ r ≤ csc θ.86 6 Integrals over Curves and Surfaces The reader can verify that R is the image under T (r. ∗ 3 2 2 6 D1 Multiplying by 2. θ) = (r cos θ. . we have √ π/4 π/4 2 1 π/4 sec θ tan θ 1 π/4 sec3 θ dθ = + sec θ dθ = sec3 θ dθ. one can alternatively divide the original square into two triangles T1 and T2 as in the text. r sin θ) ∗ ∗ ∗ of the region D∗ = D1 ∪ D2 where for D1 we have 0 ≤ θ ≤ 1 π and 4 1 ∗ 0 ≤ r ≤ sec θ. then use the standard integral number 43 at the back of the book. in the integral over y. we obtain the answer x2 + y 2 dxdy = R √ 1√ [ 2 + log(1 + 2)]. we obtain x2 + y 2 dxdy = T1 √ ∗ D1 r2 r drdθ = ∗ D1 r2 drdθ. substitute y = xv. write the integral over T1 as a double integral (ﬁrst with respect to y. The 2 ∗ ∗ transformation T sends D1 onto a triangle T1 and D2 onto a triangle T2 . R T1 Changing to polar coordinates.
4 v e − e−1 dv = . we get ∂(x.2A Challenging Example 87 Here is another example of the use of the change of variables theorem. y)/∂(u. Thus. Exercise 15 Evaluate exp B y−x y+x dx dy where B is the inside of the triangle with vertices at (0. 0). 1) and (1. v = y + x. v) = 1/2. (0. 0). exp B y−x y+x dx dy = B exp 1 0 1 v u v exp 1 du dv 2 u du dv v u v v −v 1 = 2 = = 1 2 1 2 −v v exp 0 1 0 dv 1 (e − e−1 ).6. Solution. Using the change of variables u = y − x.
1. In the text.88 6 Integrals over Curves and Surfaces Supplement 6. so e−(x e−(x Ca 2 2 +y 2 ) ≤ e−a .3. Thus. it suﬃces to show that lim e−(x Ra 2 +y 2 ) a→∞ dx dy − Da e−(x 2 +y 2 ) dx dy equals zero. Thus. y Ra Ca Da x a a Figure 6. In the region Ca . The region Ca lies between the square Ra and the circle Da . The limit equals lim e−(x Ca 2 +y 2 ) a→∞ dx dy.1.3). we showed that we showed that lim e−(x Da 2 +y 2 ) a→∞ dx dy exists by directly evaluating it. where Ca is the region between Ra and Da (see Figure 6. which was used in the derivation of the Gaussian integral formula. 2 +y 2 ) dx dy ≤ Ca e−a dx dy 2 . 0≤ x2 + y 2 ≥ a (the radius of Da ).2B The Gaussian Integral The purpose of this supplement is to prove the equality of the following two limits a→∞ lim e−(x Da 2 +y 2 ) dx dy = lim a→∞ e−(x Ra 2 +y 2 ) dx dy.
2 2 Thus it is enough to show that lima→∞ a2 e−a = 0.2B The Gaussian Integral 89 = e−a2 area (Ca ) = e−a (4a2 − πa2 ) = (4 − π)a2 e−a .6. by l’Hˆpital’s rule o 2 a→∞ lim a2 e−a = lim 2 a→∞ a2 ea2 = lim a→∞ 2a 2aea2 = lim a→∞ 1 = 0. ea2 as required. But. .
90 6 Integrals over Curves and Surfaces .
a selftaught English mathematician. 1824–1907) discovered it.1) . y) dy = 0. This was neglected until Sir William Thomson (Lord Kelvin. In 1828 Green published a privately printed booklet.3 Exact Diﬀerentials The main theoretical point is that. who learned much from Poisson’s papers.2. and had it published in the Journal f¨r Mathematik starting in 1850. y) + Q(x. y) of the form P (x. Morris Klein Mathematical Thought from Ancient to Modern Times Supplement for §8. undertook to treat static electricity and magnetism in a thoroughly mathematical fashion. An essay on the Application of Mathematical Analysis to the Theories of Electricity and Magnetism.Page 91 8 The Integral Theorems of Vector Analysis George Green (1793–1841). also carried over the notion of the potential function to electricity and magnetism. recognized its great value. u Green. to solve a diﬀerential equation in the variables (x. dx (8.
If x and y are each functions of t and lie on the surface constant = f (x. ∂y ∂x and that this corresponds to equality of mixed partial derivatives of f .1).92 8 Integral Theorems of Vector Analysis we proceed as follows: ﬁrst we test to see if P dx + Qdy is exact. y). The equation is exact. then there is a function f (x. ∂y ∂y ∂Q ∂ = (sin y cos x) = − sin y cos x. y = 0. y) such that P = ∂f /∂x and Q = ∂f /∂y can then be found by integration: P dx = cos y sin x dx = − cos y cos x + g(y) + C1 sin y cos x dy = − cos y cos x + h(x) + C2 . dx π 4 cos y sin x + sin y cos x Solution. ∂x dt ∂y dt dt dt In fact. since ∂P ∂ = (cos y sin x) = − sin y sin x. if the corresponding vector ﬁeld P i + Qj is conservative. If the equation is exact and we ﬁnd a corresponding f . y) with respect to x and use the chain rule in a similar way. This assertion is readily checked by implicit diﬀerentiation as follows.2. this is one possible way to interpret equation (8. that is. ∂x ∂x The function f (x. ∂y Recall that the test for exactness is ∂P ∂Q = . Example 1. conditions: Solve the following diﬀerential equation satisfying the given dy = 0. and Q dy = . ∂x Q= ∂f . y) such that P = ∂f . then diﬀerentiating both sides and using the chain rule gives 0= ∂f dx ∂f dy dx dy + =P +Q .1). If it is. then the equation f (x. y) = constant implicitly deﬁnes solutions as level sets of f . If y is a function of x and we similarly diﬀerentiate the equation constant = f (x.2. then we literally get equation (8.
Sometimes multiplying an equation by an appropriate factor can make it exact.e. thus.3 Exact Diﬀerentials 93 where g(y) is a function of y only (which will show up when integrating Q with respect to y). Solution. i. Solve the equation x dy = xy 2 + y dx by using the integrating factor 1/y 2 .2) x x dy = − + g2 (x). the equation is exact. ﬁrst multiply and then check for exactness: The equation is x dy 1 − x − = 0. f (x. a constant. 2 y (8.2. write −x − 1 y dx = − x2 x − + g1 (y). √ the solution is deﬁned by cos y cos x = 1/ 2.. To check this.2. y2 ∂ ∂x 1 . we ﬁnd that f (x. y2 y (8.3) Comparing (8.3). ∂ ∂y and −x − x y2 1 y = = 1 . a constant. we ﬁnd that C = − cos 0 · cos(π/4) = −1/ 2.2) and (8. and so the solution is deﬁned by the level curves of f . y) = x x2 + = C. The integrating factor 1/y 2 makes our equation exact.2. y 2 . and C2 and C2 are two constants. then the equation is solved. y) = − cos y cos x = C. Since √ y = 0 when x = π/4. Compare those two results.8. To ﬁnd an antiderivative. Example 2. y2 Thus. we see that if g2 (x) = −x2 /2 and g1 (y) = C. Such factors are called integrating factors. h(x) is a function of x only (which will show up when integrating P with respect to x). 2 dx y y Now.2.
(8. Naylor. where ρ(x.5.5.. y) = δ(x − y) (8. Volumes I and II. Duﬀ and D.5. Diﬀerential Equations of Applied Mathematics. is called Green’s function for this diﬀerential equation. R.94 8 Integral Theorems of Vector Analysis Supplement for §8. The presentation will be quite informal. it is enough here to assume that δ is a symbolic expression with the operational property shown in equation (8. z). y) that has the properties G(x. A WileyInterscience Publication. such as G.4) (in this expression y is held ﬁxed. Suppose we wish to solve Poisson’s equation ∇2 u = ρ for u(x. y. x) and ∇2 G(x.5 Green’s Functions Next we show how vector analysis can be used to solve diﬀerential equations by a method called potential theory or the Green’sfunction method. y. See the references in the preceding footnote for a more careful deﬁnition of δ.D. . y) = G(y.5). 1 This is not a precise deﬁnition. New York. nevertheless. Hilbert. A function G(x.5) R3 This is sometimes called the sifting property of δ. John Wiley & Sons Inc. Here ρ(x − y) represents the Dirac delta function. Wiley. 1989. Courant and D. Wiley Classics Library. z) is a given function. New York. It has the following operational property that formally follows from conditions (i) and (ii): For any continuous function f (x). Reprint of the 1962 original. which solves the diﬀerential equations with ρ replaced by δ. that is. Recall that this equation arises from Gauss’ Law if E = ∇u and also in the problem of determining the gravitational potential from a given mass distribution. 1966. “deﬁned” by1 (i) and (ii) R3 δ(x − y) = 0 for x=y δ(x − y) dy = 1. for further information.F. f (y)δ(x − y) dy = f (x). Methods of Mathematical Physics. the reader may consult references.
4) is satisﬁed if 1 . y) dy = B R3 ∇2 G(x. y) satisﬁes the diﬀerential equation ∇2 u = ρ with ρ replaced by δ(x − y). ∇ · ∇G(x. Proof. y)ρ(y) dy (8. For property (ii). y) = 0 and (ii) R3 for x = y ∇2 G(x. To check the second part of equation (8. y) = − Clearly G(x. y) = G(y. x). and therefore for r = 0. y) = 1.5. Green’s Function in R3 . we must verify that ∇2 G(x. then u(x) = R3 G(x.5. y) = r .6) is a solution to ∇2 u = ρ.8.5 Green’s Functions 95 Theorem 1.5. y)ρ(y) dy = R3 R3 (∇2 G(x. above].5) = R3 = ρ(x) The “function” ρ(x) = δ(x) represents a unit change concentrated at a single point [see conditions (i) and (ii).4). 4π x − y) G(x. 4πr3 (8. y) has the following two properties of the δ function: (i) ∇2 G(x. y) dy. ∇2 To see this. . If G(x. Property (i) is true because the gradient of G is ∇G(x. §4.4) (by 8. note that G(x. y) represents the potential at x due to a charge placed at y. Thus G(x. by property (i). we choose We claim that equation (8.5. let B be a ball about x. ∇2 G(x.7) where r = x − y is the vector from y to x and r = r (see Exercise 30.5. y)ρ(y) dy δ(x − y)ρ(y) dy (by 8. We will explain the meaning of (ii) in the course of the following discussion.4). y) = 0 (as in the aforementioned exercise).5.
R3 . We now turn to the problem of using Green’s functions to solve Poisson’s equation in a bounded region with given boundary conditions. (8. ∂B by Gauss’ Theorem. y) · n dS. 2π (8. §8. This is Green’s ﬁrst identity. y) = In the plane rather than in 1 log x − y . we get ∇2 G(x.5. To do this. We start with the identity ∇ · F dV = V S F · n dS.10) where ∂g/∂n = ∇g · n. where f and g are scalar functions. 4πr3 which proves property (ii).7). If we simply permute f and g and subtract the result from equation (8.10). a solution of ∇2 u = ρ is u(x) = R3 −ρ(y) dy. one can similarly show that G(x. (f ∇2 g − g∇2 f ) dV = V S f ∂g ∂f − ∂n ∂n dS. which can be obtained from the divergence theorem (see Exercise 15.5. equals ∇2 G(x.5. Replacing F by f ∇g. S is its boundary.5. Green’s Function in Two Dimensions. we need Green’s ﬁrst and second identities. making use of (8.11) which we shall use shortly.9) and so a solution of the equation ∇2 u = ρ is u(x) = 1 2π R2 ρ(y) log x − y dy. Green’s Identities.5. where V is a region in space. . in turn. we obtain Green’s second identity. Thus.96 8 Integral Theorems of Vector Analysis This. Therefore. 4π x − y (8.5. we obtain ∇f · ∇g dV + V V f ∇2 g dV = S f ∂g dS. y) · n dS = ∂B ∂B r·n dS = 1. and n is the outward unit normal vector. ∂n (8.4).8) by Theorem 1.
if ∂u/∂n is given on S we must ﬁnd a G such that ∂G/∂n vanishes on S. This is the Neumann Green’s function. Choosing our integration variable to be y and using G(x. y) − G(x.5. a Dirichlet Green’s function G(x. y) = G(y. If ρ = 0.5.5.13) where u appears on both sides of the equation and the integration variable is y. and by equation (8. fully. Therefore if u is given on S we must ﬁnd a G such that G(x.8. equation (8. x). and the corresponding equations for Green’s function G(x.12) Note that for an unbounded region. The crucial point is that evaluation of the integral requires only that we know the behavior of u on S. (8. or. Consider Poisson’s equation ∇2 u = ρ in some region V . y) (y) ∂n ∂n dS. Thus. we want to make G∂u/∂n vanish on the boundary so we can evaluate the integral.5 Green’s Functions 97 Green’s Functions in Bounded Regions. y) = δ(x − y).5). equation (8. y) = G(y.5. this becomes [u(y)δ(x − y) − G(x.5. y)ρ(y)] dy = V S u ∂u ∂G −G ∂n ∂n dS. u(x) = V G(x. u(x) = S u(y) ∂G ∂u (x. either u is given on the boundary (for a Dirichlet problem) or ∂u/∂n is given on the boundary (for a Neumann problem. this becomes identical to our previous result. y)ρ(y) dy + S u ∂G ∂u −G ∂n ∂n dS. Inserting u and G into Green’s second identity (8.5.11). (8. y) vanishes whenever y lies on S.12) reduces to u= S u ∂G ∂u −G ∂n ∂n dS.12) enables us to solve for u in a bounded region where ρ = 0 by incorporating the conditions that u must obey on S. for all of space. y) is deﬁned for x and y in the volume V and satisﬁes these three conditions: . If we know u on the boundary. x) and ∇2 G(x.6). This is called the Dirichlet Green’s function for the region V . we obtain (u∇2 G − G∇2 u) dV = V S u ∂G ∂u −G ∂n ∂n dS. Commonly.5. Equation (8. Conversely.
y) = 0 when y lies on S.] It is interesting to note that condition (a) is actually a consequence of conditions (b) and (c). by the similarity of the triangles xOy and xOy . This means. and by condition (c). [According to the procedure above. We have chosen x on C to begin with. In Figure 8. y) = G(y.] The Green’s function G(x. The point y represents the “reﬂection” of the point y into the region outside the circle. then it is not necessary to check condition (a). that is.5. equations (8. w). in eﬀect. G(x.5. the boundary of the region V .1 we have draw the point x on the circumference because that is where we want G to vanish.5.11) is zero.8) and (8. that if (b) and (c) hold with x and y internchanged. On the left hand side. ∇2 G(x.) Solving any particular Dirichlet or Neumann problem thus reduces to the task of ﬁnding Laplace’s equations on all of R2 or R3 . By condition (b). such that ab = R2 .9). y) = δ(x − y).14) . we substitute f (x) and g(x) to give [G(x. of course. ∇2 f (x) = δ(x − y) and ∇2 g(x) = δ(x − w). in conditions (b) and (c) the variables x and y can be interchanged without changing the condition. which is the asserted symmetry of G. This will enable us to solve ∇2 u = 0 (or ∇2 u = ρ) with u given on the boundary circle. be valid for all x. [Note that by condition (a).5. y in the disk. y) = 1 2π log r − log r a R . To see this. namely. (This result is sometimes called the principle of reciprocity. y) is supposed to vanish when either x or y is on C.11) with f (x) = G(x. Now when x ∈ C. y)δ(x − w) − G(w. (8. Green’s Function for a Disk.5. x). R b b R Hence.5.98 8 Integral Theorems of Vector Analysis (a) (b) (c) G(x. provided (b) and (c) also hold with x and y interchanged. x). r r r R r a = . We shall now use the twodimensional Green’s function method to construct the Dirichlet Green’s function for the disk of radius R (see Figure 8.5. y) and g(x) = G(w. G(x. we ﬁx points y and w and use equation (8.1). if we choose our Green’s function to be G(x. f and g vanish when x lies on S and so the right hand side of (8. x)δ(x − y)] dx = 0 V which gives G(w. r = = . y) that we shall ﬁnd will. y) = G(y.
8.5 Green’s Functions
y n x R r r'' y'
99
θ
y
θ'
b x
O
a
C
Figure 8.5.1. Geometry of the construction of Green’s function for a disk.
we see that G is zero if x is on C. Since r a/R reduces to r when y is on C, G also vanishes when y is on C. If we can show that G satisﬁes ∇2 G = δ(x−y) in the circle, then we will have proved that G is indeed the Dirichlet Green’s function. From equation (8.5.9) we know that ∇2 (logr)/2π = δ(x − y), so that ∇2 G(x, y) = δ(x − y) − δ(x − y ), but y is always outside the circle, and so x can never be equal to y and δ(x − y ) is always zero. Hence, ∇2 G(x, y) = δ(x − y) and thus G is the Dirichlet Green’s function for the circle. Now we shall consider the problem of solving ∇2 u = 0 in this circle if u(R, θ) = f (θ) is the given boundary condition. By equation (8.5.13) we have a solution u=
C
u
∂G ∂u −G ∂n ∂n
ds.
But G = 0 on C, and so we are left with the integral u=
C
u
∂G ds, ∂n
where we can replace u by f (θ), since the integral is around C. Thus the task of solving the Dirichlet problem in the circle is reduced to ﬁnding
100
8 Integral Theorems of Vector Analysis
∂G/∂n. From equation (8.5.14) we can write ∂G 1 = ∂n 2π Now 1 ∂r 1 r − r ∂n r ∂n .
∂r r = ∇r · n and ∇r = , ∂n r where r = x − y, and so ∂r r·n r cos(nr) = = = cos(nr), ∂n r r
where (nr) represents the angle between n and r. Likewise, ∂r = cos(nr ). ∂n In triangle xyO, we have, by the cosine law, a2 = r2 + R2 − 2rR cos(nr), and in triangle xy 0, we get b2 = (r )2 + R2 − 2r R cos(nr ), and so ∂r R2 + r2 − a2 = cos(nr) = ∂n 2rR Hence and R2 + (r )2 − a2 ∂r = cos(nr ) = . ∂n 2r R
∂G 1 R2 + r2 − a2 R2 + (r )2 − a2 = − . ∂n 2π 2rR 2r R
Using the relationship between r and r when x is on C, we get ∂G ∂n =
x∈C
1 2π
R2 − a2 Rr2
.
Thus the solution can be written as u= 1 2π f (θ)
C
R2 − a2 ds. Rr2
Let us write this in a more useful form. Note that in triangle xy0, we can write r = [a2 + R2 − 2aR cos(θ − θ )]1/2 , where θ and θ are the polar angles in x and y space, respectively. Second, our solution must be valid for all y in the circle; hence the distance of y from the origin must now become a variable, which we shall call r . Finally, note that ds = R dθ on C, so we can write the solution in polar coordinates as f (θ) dθ R2 − (r )2 2π u(r , θ ) = . 2π (r )2 + R2 − 2r R cos(θ − θ ) 0
8.5 Green’s Functions
101
This is know as Poisson’s formula in two dimensions.2 As an exercise, the reader should use this to write down the solution of ∇2 u = ρ with u a given function f (θ) on the boundary.
Exercises
1. (a) With notation as in Figure 8.5.1, lem for the sphere of radius R in function 1 G(x, y) = 4π R(R2 − a2 ) 4π
2π 0 0 π
show that the Dirichlet probthree dimensions has Green’s 1 R − ar r .
(b) Prove Poisson’s formula in three dimensions: u(y) = f (θ, φ) sin φ dθdφ (R2 − a2 − 2Ra cos γ)3/2
2. Let H denote the upper half space z ≥ 0. For a point x = (x, y, z) in H, let R(x) = (x, y, −z), the reﬂection of x in the xy plane. Let G(x, y) = −1/(4π x − y ) be the Green’s function for all of R3 . ˜ (a) Verify that the function G deﬁned by ˜ G(x, y) = G(x, y) − G(R(x), y) is the Green’s function for the Laplacian in H. (b) Write down a formula for the solution u of the problem ∇2 u = ρ in H, and u(x, y, 0) = φ(x, y).
Exercises 3 through 9 give some sample applications of vector calculus to shock waves.3 3. Consider the equation ut + uux = 0 for a function u(x, t), −∞ < x < ∞, t ≥ 0, where ut = ∂u/∂t and ux = ∂u/∂x. Let u(x, 0) = u0 (x) be the given value of u at t = 0. The curves (x(s), t(s)) in the xt plane deﬁned by ˙ x = u, t = 1 ˙
are several other ways of deriving this famous formula. For the method of complex variables, see J. Marsden and M. Hoﬀman, Basic Complex Analysis, 2d ed., Freeman, New York, 1987, p. 195. For the method of Fourier series, see J. Marsden, Elementary Classical Analysis, Freeman, New York, 1974, p. 466. 3 For additional details, consult A. J. Chorin and J. E. Marsden, A Mathematical Introduction to Fluid Mechanics, 3rd ed., SpringerVerlag, New York, 1992, and P. D. Lax, “The Formation and Decay of Shock Waves,” Am. Math. Monthly 79 (1972): 227241. We are grateful to Joel Smoller for suggesting this series of exercises.
2 There
˙ (b) Show that the slopes of the characteristic curves are given by dt/dx = 1/u. 0) and (x2 .5. t) = 0 for x = ±a. Let u be a “genuine” solution of equation (8. ¯ (d) Calculate t. t) with t > 0. and for all (x. . Show that this together x ¯ with the result in part (a) implies that the solution cannot be continuous at P (see Figure 8.5. let φ = φ(x. (8. Characteristics of the equation ut + uux = 0. x = T .15) where f > 0 and f (u0 (x2 )) > 0.5. magnetohydrodynamics. We call equation (8. To this end. Show that the two characteristics through the points (x1 . (a) Show that u is constant along each characteristic curve by showing that u = 0. Repeat Exercise 3 for the equation ut + f (u)x = 0. The characteristics are now de˙ ﬁned by x = f (u).15) and equation in ˙ divergence form. t P x1 x2 x Figure 8. (This exercise shows that a continuous solution is generally impossible—irrespective of the smoothness of f !) 5. (Weak solutions) Since equations of the form in Exercise 4 arise in many physical applications [gas dynamics. t = 1. nonlinear optics (lasers)] and because it would be nice for a solution to exist for all time (t). such that φ(x. 0) ¯ intersect at a point P = (¯.2). it is desirable to make sense out of the equation by reinterpreting it when discontinuities develop. (c) Suppose that x1 < x2 and u0 (x1 ) > u0 (x2 ) > 0. 4. t) be a C 1 function.2.102 8 Integral Theorems of Vector Analysis are called characteristic curves (the overdot denotes the derivative with respect to s). Let D be a rectangle in the xt plane determined by −a ≤ x ≤ a and 0 ≤ t ≤ T .5. and use it to prove that the characteristic curves are straight lines determined by the initial data.5.15). t) in the upper half plane outside D.
(b) Show that if u is a weak solution that is C 1 in an open set Ω in the upper half of the xt plane. Let u be a (weak) solution of equation (8.8.) Thus.15) and suppose Γ is a smooth curve in the xt plane such that u “jumps” across a curve Γ. as shown in Figure 8. 0) dx = 0. We call the function u a weak solution of equation (8. Choose φ to vanish on D and outside D. (The jump condition.5. then equation (8. t Γ P Q D1 R D2 x Figure 8. (8.5 Green’s Functions 103 (a) Show that [uφt + f (u)φx ] dxdt + t≥0 t=0 u0 (x)φ(x. that is. then u is a genuine solution of equation (8.15) if equation (8.16) (HINT: Start with D [ut + f (u)x ]φ dxdx = 0.16) holds for all such φ. Choose a point P ∈ Γ and construct. 6.16) holds for all φ as above.5.5.5.5. the reader shall now determine that not every type of discontinuity is admissible. (a) Show that [uφt + f (u)φx ] dxdt = 0 D .5.3.) The deﬁnition of a weak solution given in Exercise 5 clearly allows discontinuous solutions. near P . We call that Γ a shock wave. a “rectangle” D = D1 ∪ D2 .5.5. if u is a smooth solution. The solution u jumps in value from u1 to u2 across Γ. However.15) in Ω.3. which is also known in gas dynamics as the RankineHugoniot condition. for there is a connection between the discontinuity curve and the values of the solution on both sides of the discontinuity. u is of a class C 1 except for jump discontinuity across Γ.
t1 ) and R = (x(t2 ). Show that 0= ∂D1 φ[−u dx + f (u) dt] + ∂D2 φ[−u dx + f (u) dt] and deduce that 0= Γ φ([−u] dx + [f (u)] dt) where [α(u)] = α(u2 ) − α(u1 ) denotes the jump in the quantity α(u) across Γ.17) is called the jump condition. t2 ). (In gas dynamics.5. (Loss of uniqueness) One drawback of accepting weak solutions is loss of uniqueness. [u] · s = [f (u)]. some mathematical solutions are extraneous and rejected on physical grounds. t) approaches the value ui (x0 . dt (d) Show that at the point P on Γ. 7. The number s is called the speed of the discontinuity.17) where s = dx/dt at P . x ≥ 0 1.) Consider the equation ut + u2 2 = 0. t0 ) on Γ from ∂Di . x < 0. . (8. x with initial data u(x. 0) = −1. For example. show that R t2 0= Q φ([−u] dx + [f (u)] dt) = t1 φ [−u] dx + [f (u)] dt. it is the relationship that any discontinuous solution will satisfy. t0 ).5. discontinuous solutions of rarefaction shock waves are rejected because they indicate that entropy decreases across the discontinuity.104 8 Integral Theorems of Vector Analysis and [uφt + f (u)φx ] dxdt = D1 D1 [(uφ)t + (f (u)φ)x ] dxdt. (b) Suppose that u jumps in value from u1 to u2 across Γ so that when (x. t) approaches a point (x0 . Equation (8. (c) If the curve Γ deﬁnes x implicitly as a function of t and ∂D intersects Γ at Q = (x(t1 ). u(x.
u(x. s = 2 (u2 + 2 2 3 u1 u2 + u2 )/(u2 + u1 ). t) = α. 2 (It can be shown that if f > 0.5. (The solution of equation (8.] 1 9. s = 1 (u2 + u1 ). In our example.5.18).5.15) depends on the particular divergence form used. for equation (8.15) where f > 0. (8. 2 1 2 1 3 ( u )t + ( u )x = 0. t) E ≤ a t for some E > 0 and all a = 0. t < x.15). Thus. t) − u(x. t) can only “jump down” as x increases. ( Noninvariance of weak solutions under nonlinear transformation) Consider equation (8. [HINT: The equations have diﬀerent jump conditions: In equation (i).) The equation ut + uux = 0 can be written in the two divergence forms 1 ut + ( u2 )x = 0.8.) . there is a unique solution satisfying the “entropy” condition u(x + a.5 Green’s Functions 105 Show that for every α ≥ 1. Hence for ﬁxed t.18) (b) Show that the above transformation does not necessarily map discontinuous solutions of equation (8. x ≤ 1 − α t 2 1−α −α. while in equation (ii).) 8. s[u] = [f (u)].15) into discontinuous solutions of equation (8. where uα is deﬁned by 1. 2 3 (i) (ii) Show that a weak solution of equation (i) need not be a weak solution of equation (ii).5. t≤x≤0 2 uα (x.18). uα is a weak solution. (HINT: Check the jump conditions. (a) Show that the transformation v = f (u) takes this equation into vt + vvx = 0. s[v] = 1 [v 2 ] implies s[f (u)] = 2 1 2 2 [f (u) ]. for equation (8.5.5. 0 ≤ x ≤ α − 1 t 2 α−1 −1. uniqueness can be recovered by imposing an additional constraint on the solutions. this holds only for the solution with α = 1.5.
Show that Poisson’s formula in two dimensions may be written as u(r .106 8 Integral Theorems of Vector Analysis 10. reiθ − z 2 . θ ) = where z = r eiθ . 1 2π 2π u(reiθ ) 0 r2 − z 2 dθ. Requires a knowledge of complex numbers.
y. . 1. xn ) and ﬁx an index k. let δ1 and δ2 be small enough that Dδ1 (y0 ) ⊂ B and g(y) − w < whenever y ∈ B and 0 < y − y0 < δ2 . 0 0 ex 0 . Then w = 0 and g(f (x)) − 1 for all x. Since g(y0 ) = w.Page 107 Selected Answers and Solutions for the Internet Supplement §2. . Df (x. f (x) − y0 < δ2 . f (x) − y0 < δ1 . Then n f (x) = akk x2 + k i=1=k n akj xk xj akj xk xj + [terms not involving xk ] j=1=k + . so that f (x) ∈ B and g(f (x)) is deﬁned. . z) = 0 − sin y 0 0 cos z Df is a diagonal matrix if each component function fi depends only on xi . Also. the 0 < y − y0 restriction may be dropped. Then for such x.7 Some Technical Diﬀerentiation Theorems. (b) If > 0. 5. and so g(f (x)) − w < . Let x = (x1 . Let δ be small enough that f (x)−y0 < min (δ1 . δ2 ) whenever x ∈ A and 0 < x − x0 < δ. 3. (a) Let A = B = C = R with f (x) = 0 and g(x) = 0 if x = 0 and g(0) = 1. .
(Prove the last statement. For continuity at (0. ∇f (x) = 2Axk . y0 )(x − x0 . Since the kth components agree for each k. y − y0 ) / (x − x0 . < x is . 0. since g(x)/x = ±1 (as x is positive or negative). y − y0 ) = Dg(x0 )(x−x0 )+Dh(y0 )(y−y0 ). so that f (x) is o(x). 17. (c) limx→0 f (x)/x = limx→0 x = 0. y0 )(x − x0 . see Exercise 11. Since f is o(x). Therefore g(x) is not o(x). y − y0 ) goes to 0. 0). so limx→0 (gf )(x)/ x = 0. y) − f (x0 . not for x = x0 . 13. (a) limx→0 (f1 +f2 )(x)/ x = limx→0 f1 (x)/ x +limx→0 f1 (x)/ x = 0. Let x take the role of x0 and x + h that of x in the deﬁnition. Then (gf )(x)/ x ≤ g(x) f (x)/ x . there is a δ > 0 such that f (x)/ x < /c whenever 0 < x < δ. Now use the triangle inequality and the fact that (x − x0 . 9. use the fact that xy xy ≤ 2 1/2 = y (x2 + y 2 )1/2 (x ) or that xy ≤ (x2 + y 2 )/2. The vector a takes the place of x0 in the deﬁnition of a limit or in Theorem 6. Therefor f (x) = g(x) for x = a certainly suﬃces to make the limits equal. 7. the limit depends only on values of f (x) for x near x0 . y − y0 ) is larger than x − x0  and y − y0  to show that f (x.) 11.108 Answers for the Internet Supplement and therefore ∂f = 2akk xk + ∂xk n aki xi + i=k i=k aki xi = 2 j=1 akj xj = (2Ax)k . (b) Let > 0. In either case. 19. The matrix T of partial derivatives is formed by placing Dg(x0 ) and Dh(y0 ) next to each other in a matrix so that T(x0 . But limx→0 g(x)/x does not exist. 15. Use the limit theorems and the fact that the function g(x) = continuous. y0 ) − T(x0 .
we want a δ > 0 such that 0 < x − x0  < δ implies 1/x2 − 1/x0 2  < ε for x and x0 in (0. y. x0 )}3 . y. x and y ∈ Rn . x0 ) > 0. Note that this proof works because δ and ε do not depend on a particular choice of a point in Rn . Since T is linear. we have T(x) − T(y) ≤ M δ = ε.Answers for the Internet Supplement 109 Solutions to Selected Exercises in §2. Given ε > 0. f3 ). we may ﬁnd a δ which will work for a particular x0 . This is diﬀerent from continuity in that for continuity. sin z) = (f1 . 2 x x0 {min(x. 4. i. we can ﬁnd one δ for all x0 . f2 depends only on the second. T(x−y) ≤ T(x)−T(y) . but more importantly. A uniformly continuous function is not only continuous at all points in the domain. Df is diagonal if fn is a function of the nth variable only. then x − x0  < δ implies 1 2 1 ε 2 − 2 < δ · 3 < x0 3 · 3 = . Let δ = ε/M . Let δ < (ε/2) {min(x. (b) Given ε > 0. and so on. 1]. then for 0 ≤ x − y ≤ δ.. z) = ∂f2 /∂x ∂f2 /∂y ∂f2 /∂z ∂f3 /∂x ∂f3 /∂y ∂f3 /∂z x e 0 0 0 = 0 − sin y 0 0 cos z Df is a diagonal matrix when f1 depends only on the ﬁrst variable. for given ε > 0. z) = (ex . x2 x0 x0 2 x0 .e. x0 )}3 Note that min(x. f2 . Here. we have T(x − y) ≤ M x − y from exercise 2(a).7 1. 1 1 2 − 2 ≤ x − x0  . cos y. Thus. 1]. we have f (x. A continuous function does not always have to be uniformly continuous (for examples: f (x) = 1/x2 on R or g(x) = 1/x on (0. so ∂f1 /∂x ∂f1 /∂y ∂f1 /∂z Df (x.) (a) T : Rn → Rm is linear implies that Tx ≤ M x . We calculate: x0 2 − x2 1 1 x − x0 x + x0  − 2 = = 2 2x 2 x x0 x 0 x2 x0 2 1 1 2 < x − x0  3 < ε = x − x0  · + 2 2 xx0 x0 x x0 (if x0 is the smaller of x and x0 ).
y) − (x0 . y0 ). y) = (Dg(x).y)→(x0 . Since (x. Since g and h are diﬀerentiable at x0 and y0 . Given ε. and this satisﬁes the deﬁnition for diﬀerentiability of f at (x0 .110 Answers for the Internet Supplement We have only shown that f (x) = 1/x2 is continuous. What we want to prove is that as matrices. lim(x. . y) − (x0 . y0 ) (x. (x.y0 ) Hence this limit goes to 0. Dh(y)). then (x. any δ approaches ∞ as x0 goes to 0. 7. y0 ) ≥ y − y0 . y0 ) ≥ x − x0 and (x. x2 x0 x2 x0 2 x2 x0 2 so for ﬁxed ε.y)→(0. the sum of the two limits is greater than lim g(x) + h(y) − (g(x0 ) + h(y0 )) − (Dg(x − x0 ) + Dh(y − y0 )) .0) f (x. y) − (0. respectively. Therefore. y) = g(x) + h(y). We have f (x. note that xy (x2 + 1/2 y2 ) −0 ≤ xy = y. y) − (x0 . we have lim g(x) − g(x0 ) − Dg(x − x0 ) =0 x − x0  h(y) − h(y0 ) − Dh(y − y0 ) =0 y − y0  x→x0 and y→y0 lim Now. x and we also know that y ≤ x2 + y 2 . 11. use the triangle inequality: g(x) − g(x0 ) + h(y) − h(y0 ) − (Dg(x − x0 ) + Dh(y − y0 )) ≤ g(x) − g(x0 ) − Dg(x − x0 ) + h(y) − h(y0 ) − Dh(y − y0 ). Df (x. Let δ = ε. y) = 0 and f is continuous. 0) = x2 + y 2 < δ implies that xy (x2 + y2 ) 1/2 − 0 < y < ε. It is not uniformly continuous since x0 2 − x2 1 1 x − x0 x + x0  − 2 = = .
.) To go the other way. (a) Suppose y is a boundary point of the open set A ⊂ Rn . (corresponding to δ = 1/n). Suppose y is not in A. then D(1/n) (y) ∩ A = ∅ for n = 1. the righthand side is a boundary point of A. i. Choose N so that n ≥ N implies that xn − y < δ. .e. . for every ε > 0. which goes to zero as n goes to inﬁnity. . Then xn converges to y (as in part (a)). Choose δ > 0 as above.” and the inequality signs). . choose ε = 1/n. . . Hence (i) y is not in A since points of A have neighborhoods contained in A. but there is a sequence {xn } in A converging to y. Speciﬁcally. Let ε > 0. Choose xn in A such that xn −y < 1/n. Pick xn to be an element of D(1/n) (y) ∩ A. The diﬀerence between this method and the method of proving by contradiction is that we do not negate the hypothesis. . but f (x) does not converge to b. n ≥ N implies that xn is in Dε (y). and prove the negation of the result we want to arrive at. 3. 3. Then xn is in A and y−xn ≤ 1/n. Then every ball centered at y contains points in A and points not in A. 2. Fix ε > 0.) (c) Let U be open in Rm with x in U . On the other hand. so the sequence f (xn ) converges to b. so xN is in Dε (y) ∩ A. (b) Suppose limx→y f (x) = b. (We need y to be on the boundary of A to guarantee the existence of the sequence {xn }..” “for every. so the righthand side is not empty. Hence. 2. but f (x) − b > ε. . x−y < δ implies that f (x)−f (y) < . The y is in the intersection of Dε (y) and the set of all points not in A. We negate the statement of the hypothesis (note the changes in “there exists.e. The proof follows from part (b). the intersection of A and the ball Dε (y) is not empty. . f is continuous at x. i. Then n ≥ N implies f (xn ) − b < . and (ii) y is the limit of the sequence from A. suppose it is not the case that limx→y f (x) = b. there is a δ > 0 such that for x in A. Therefore..Answers for the Internet Supplement 111 16. Then there exist ε > 0 such that for every δ > 0. . xN is in A. . 3. there exists an x in A with x − y < δ. 2. (Note: We have proved the “contrapositive” of the theorem. Let {xn } be a sequence in A converging to y. xn →x lim f (xn ) = f (x) implies f (xn ) → f (x) for any sequence xn converging to x in U . n = 1. By part (b). but f (xn )−b > for n = 1. there exists an N such that n ≥ N implies that xn −y < .
so it is best to just say the stability test fails). y) = (2x + 4y − 8)i + (4x − 2y − 6)j = 0 only if 2x + 4y − 8 = 0 and 4x − 2y − 6 = 0. §4. 1) = 4. The second derivative test thus tells us that the point is a saddle. y = 1. These points are where the gradient vanishes. 1) = −2. ∇V (x. (In fact. ∂y 2 ∂2V (2. the critical points of the potential V are the equilibrium points.1A: Equilibria in Mechanics 1. In fact. in particular. we cannot conclude that the point is stable. Simplifying we get the system of equations x + 2y = 4 2x − y = 3 whose solution is x = 2. 1) = 2. 5. n0 = (1/2 3)(i + j − k) √ √ √ (b) r = [cos(πt/12)/2 2+sin(πt/12)/4]i+[−1/2 2+cos(πt/12)/2 2+ √ √ sin(πt/12)/4]j + [−1/2 2 + cos(πt/12)/ 2 − sin(πt/12)/4]k . Stable equilibrium point (2 + m2 g 2 )−1/2 (−1. This critical point (2. (a) m0 = (1/ √6)(i + j + 2k). §4. −mg) Solution to Exercise 3. (2.112 Answers for the Internet Supplement §3. − 1 ) 4 4 3. we have ∂2V (2. (− 1 . ∂x2 Therefore the discriminant D= ∂2V ∂x2 ∂2V ∂y 2 − ∂2V ∂x∂y 2 ∂2V (2.1B Rotations and the Sunshine Formula √ 1. 1) is an unstable equilibrium. −1. It is the usual second derivative test in onevariable calculus. 1) will be stable if it is a strict local minimum for V . ∂x∂y = −20 < 0. it is unstable. Using the second test (Theorem 6).4A: Second Derivative Test: Constrained Extrema 1. By Theorem 14. but this has not been discussed carefully.
φ(x. t) + (x. v = −ωyi + ωxj and therefore b0 Dv(0) = ω 0 −ω 0 0 0 0 . §4. Td would be longer. t). t). t). x2 . x3 ). A = 9. t)) = (x. then by the chain rule. use the identity AT v · w = v · Aw. t). x2 . d ∂f ∂φi ∂f (φ(x. 7.4◦ 9.4 Flows and the Geometry of the Divergence 1. The “exact” formula is − tan l sin α = cos(2πt/Td )[tan(2πt/Ty ) tan(2πt/Td )− cos α]. and f = f (x1 . t) + [∇f (φ(x. t)] · [F(φ(x. d dv dw v·w = ·w+v· . 5. The equator would receive approximately six times as much solar energy as Paris. For the last equality. φ2 .Answers for the Internet Supplement 113 √ (c) (x. t) dt ∂t ∂xi ∂t i=1 = 3. 0 1 (∇ × F)(0) = ωk. By the product rule. By the choice of axes. 5. x3 . If x = (x1 . 2 ∂f (x. y. ∂t 3 . w= From the text. t) = (φ1 . t) (f (φ(x. φ3 ). z) = (−1/2 2)(i + 2j + 3k) + (−π/48)(i + j + k)(t − 12) 3. t))]. dt dt dt Substitute equation (2) into this.
the rate of change of any vector x at the origin under transport by the derivative of the ﬂow φ(x. t) rotates vectors as well. [The deformation matrix S incorporates all the length and angle changes caused by the ﬂow. t) of F is given by dx dt = Dx F(0)x = (S + W )x. which aﬀects inner products. Thus the W matrix is precisely the rate of change of vectors as they undergo an inﬁnitesimal rotation around the axis (curl F)(0) = (∇ × F)(0) by the mapping Dx ψ(x. Y(t) rotates around w and dY dt = Dx v(0)Y. t=0 This gives the rate of change of Y as it is transported (rotated) by Dx ψ. . t) of v rotates points in this ﬁeld. its derivative Dx ψ(x. In particular. volume changes are contained in S. tr S = div F(x). 0 ∂F3 ∂x ∂F3 ∂y To interpret the result. for ﬁxed t. we note that the vector ﬁeld v represents rotation around a ﬁxed axis w. by the deﬁnitions of W and ∇ × F. As t increases or decreases. and the W matrix. bw1 1 w1 2 w1 3 W = w2 1 w2 2 w2 3 w3 1 w3 2 w3 3 1 ∂F1 ∂F2 1 ∂F1 0 − − 2 ∂y ∂x 2 ∂z 1 ∂F2 ∂F1 1 ∂F2 = 0 − 2 ∂x − ∂y 2 ∂z 1 ∂F3 ∂F1 1 ∂F3 ∂F2 − − 0 2 ∂x ∂y 2 ∂y ∂z 0 −(∇ × F)z (∇ × F)y 1 0 −(∇ × F)x = (∇ × F)z 2 −(∇ × F)y (∇ × F)x 0 Our choice of coordinate axes gives 0 −ω 0 W = ω 0 0 0 0 . t)Y. the trace of S is the divergence. By Exercise 3. The ﬂow ψ(x. Let Y be an arbitrary vector and set Y(t) = Dx ψ(x.114 Answers for the Internet Supplement On the other hand. In fact. t). t=0 Thus this rate of change of x has two components: the deformation matrix. and.
d]. is approximated by its tangent line. Since B is contained in this union. so that d = (c + e)/2. f ] × [g. d] and B = [e. e]. d]. let 3. then f is integrable over Q by additivity. The line x + λv is carried to the curve λ → φ(x + λv. is called the shear. their areas can be added. b] × [c. area (B) ≤ bn .6 Some Technical Integration Theorems = a − b/2. y) is in the union.] 7. where R1 = [a. namely. b] by short steps. The tracefree part of S. Consider the vertical “doubling” of R deﬁned by Q = R ∪ R1 . φ is uniformly continuous and therefore there exists a δ > 0 such that φ(x) − φ(y) ≤ whenever x − y < δ. then e − (b − a)/n ≤ x ≤ f + (b − a)/n and g − (d − c)/n ≤ y ≤ h + (d − c)/n. if (x. This leads to bn ≤ area(B) + 2[(b − a)(h − g) + (d − c)(f − e)]/n + 4(a − b)(d − c)/n2 . and bn is the area of the union of all subrectangles of the nth regular partition of R that intersect B. . t) + Dx φ(x. b] × [c. t) after time t. t) · λv.Answers for the Internet Supplement 115 (The trace of a matrix is the sum of its diagonal entries). d] is part of the 2nth regular partition of Q. 7. < xn−1 < xn = . 1 S = S − (trS I ) 3 where I is the 3 × 3 identity. adding up the changes as you go. The argument for part (b) is similar. which for λ small. . Since the rectangles of a partition of R intersect only along their edges. If a = b. 1. the Riemann sums for that 2nth partition cannot vary by more than as we change the points from the subrectangles. Let R = [a. For large n. Letting n → ∞ and combining the inequalities proves the assertion. in particular if we change only those in [(a+b)/2. On the other hand. Let x ∈ [a. b] × [d. b] × [c. Given > 0. §5. b]×[c. 5. The nth regular partition of [(a + b)/2. If f is extended to Q by letting f be 0 on the added part. Let e = 2d − c. h]. λ → φ(x. (a) The strategy is to go from point to point within [a. These changes correspond to the possible changes for the Riemann sums for the nth partition of [(a + b)/2. b] and introduce intermediate points a = x0 < x1 < .
If we take m > nL/δ. b] × [c. choose xi ∈ D ∩ Ri . then x > x0 /2 > 0 and If x − x0  is also less than · x2 /2. (c) This is trickier. This can be done with no more than [(b − a)/δ] + 1 segments. then 2x − x0 /x2 < . Solution to Exercise 2(a). − x x0 x (0. Rn are those that intersect D. The diagonal of each subcube √ √ has length nL/m. and there are mn subcubes. (b) Use an argument like that for part (a). Partition R into subcubes by dividing each edge into m parts. . Thus φ(x) ≤ φ(a) + [(b − a)/(δ + 1)] for every x in [a. b]. whenever Let x0 ∈ (0. . Notice that > 0. d]. n φ(x) − φ(a) ≤ i=1 φ(xi ) − φ(xi−1 ) ≤ b−a δ+1 . For any x ∈ D. Since D is bounded. by the uniform boundedness principle. Thus by picking 0 0 δ < min x0 /2. xx0  x2 0 If x − x0  < x0 /2. we must ﬁnd a δ > 0 such that 1 1 < . 1] but not uniformly continuous. . x x0 xx0  x − x0  2x − x0  < . Given 0 < x − x0  < δ. 1]. Exercise 2(a). . If R1 . given . Show that f is continuous at every point of (0. there is a δ such thatf (x) − f (y) ≤ whenever x and y are in D and x − y < δ. . . . Nevertheless. ( /2) · x2 one has 0 1 1 < . moving by short steps within the rectangle [a. we have f (x) < + map(f (x1 ). f (x)). Solution. then 1 x − x0  1 = − . any two points in the same subcube are less then δ apart. − x x0 .116 Answers for the Internet Supplement x with xi+1 − xi < δ. we may ﬁnd a large “cube” R with sides of length L such that D ⊂ R. since D may be composed of many disconnected pieces so that the short steps cannot be taken within D. By the triangle inequality. 1] deﬁned by f (x) = 1/x. . 1]. Let f be the function on the half open interval (0.
To show that f is not uniformly continuous. we suppose it were and derive a contradiction. then n = x/R is the outward unit normal.5 Green’s Functions 1. §8. r = x − y and r = x − y . Let = 1/2. then r a/R = r. so G = 0. n m N N N Set x1 = 1/n and x2 = 1/(n + 1). But 1 is not less than 1/2 a contradiction. If f were uniformly continuous there exists a δ > 0 such that 1 1 1 < − x1 x2 2 whenever x2 − x1  < δ. Then 1 1 = − x1 x2 1 1 n+1 − 1 1 n = 1. Then ∂G R y −x 1 = − (y − x) · n. since x is never y . In general. but the second term is always 0. (b) If x is on the surface of S. and ∂G 1 1 R 1 ·n−∇ = ∇ ·n ∂n 4π a r r 1 R y −x y−x = · n. and therefore 1 4π 1 ∇x G = 4π G= R 1 1 − a x−y x−y R y −x y−x − a x−y 3 x−y 3 and and ∇2 G = 0 when x = y. − 4π a y − x 3 x−y 3 If γ is the angle between x and y. (a) If x ∈ S. m > N 1 1 1 1 2 − < + = < δ. just as in the analysis of equation x (15)(x = y . Theorem 10 gives ∇2 G = δ(x − y) − (R/a)δ(x − y ). since x is inside and y is outside the sphere). f is continuous at x0 . There for all integers n. Let N > 2δ. then x−y 2 = r2 = R2 +a2 − 2aR cos γ and x−y 2 = r 2 = R2 +b2 −2bR cos γ = (R2 /a2 )r2 . ∂n 4πr3 a (R/a)3 . Therefore ∇2 G = δ(x − y) for x and y in the sphere.Answers for the Internet Supplement 117 Thus.
To show (2). ˜ Then ∇2 G = δ(x. 1 1 ˜ G(y. y) = r /4π(r ) . and the above integral becomes ˜ ∇2 G dy = δ(x. x) = G(x. 4 4 ˜ ˜ It is obvious that r = −r . y). y) = G(y. w = −z. x) = G(y. Thus ∇G(R(x). Now the integral .118 Answers for the Internet Supplement But x · n = R and n = x/R. u(y) = S 2π f ∂G dS ∂n π = 0 0 f (θ. y) − δ(R(x). φ) sin φ dφdθ . 3 3 we know ∇G(x. y) dy = 1. y). and ˜ ∇2 G dy = [δ(x. so dy = dx dy dz = −du dv dw. y)] dy. (a) According to equations (12) of this section. 4πR r3 Integrating over the surface of the sphere. y). y) − δ(R(x). x) − G(y. and so this becomes ∂G a2 1 y·x a2 y ·x− = − +R 3 3 ∂n 4π R R R 1 a2 = y R cos γ − y R cos γ + R2 − a2 4π 3 R R2 R2 − a2 1 = since y = R2 /a. φ) R(R2 − a2 ) 4π 2π 0 R2 − a2 1 2 R sin φ 4πR r3 π 0 dφdθ = f (θ. Make a change of variable: Let u = x. so G(y. (2) (1) R3 R3 If x = y. Deﬁne r = x − y and r = R(x) − y. R3 R3 If R(x) = y. R(x)) = − π y − x + π y − R(x) . then x = y. y) = r/4πr . v = y. then R(x) = y. x) and ∇2 G = δ(x. To show (1). we need to show that ˜ ˜ G(x. (R2 + a2 − 2aR cos γ)3/2 2.
Answers for the Internet Supplement 119 becomes R3 ˜ ∇2 G dy = − =− = R3 δ(R(x). y) dy R3 δ(R(x). = R3 ˙ 3. y)ρ(y) − G(R(x). w). 0) and (x2 . Simplify and get x(1/u0 (x1 ) − 1/u0 (x2 )) = x1 /u0 (x1 ) − x2 /u0 (x2 ). where u = (u. (c) Two characteristics through (x1 . The intersection is [1/u0 (x1 )](x − x1 ) = [1/u0 (x2 )](x − x2 ). u0 (x2 ) − u0 (x1 ) . Solve for x: x= (d) Plug in: ¯ t= 1 u0 (x1 ) x1 u0 (x2 ) − x2 u0 (x1 ) − x1 u0 (x2 ) − u0 (x1 ) =− x2 + x1 u0 (x2 ) − u0 (x1 ) x1 u0 (x1 ) 1 u0 (x1 ) − − x2 u0 (x2 ) 1 u0 (x2 ) = x1 u0 (x2 ) − x2 u0 (x1 ) > 0. v. respectively. u is a constant. u) (−du) R3 δ(R(x). therefore 1/u is also a constant. y)ρ(y) dy. x(s) dx u From part (a). The slope is equal to t(s) dt 1 = = . u) du = 1. ˙ ˙ (b) Use dt/ds = 1 and dx/ds = u. (b) Simply ”stick” our Green’s function into an integral: u(x) = R3 ˜ G(x. (a) Use the Chain rule: ut + uxx = ut t + ux x = u = 0. 0) have equations t = [1/u0 (x1 )](x − x1 ) and t = [1/u0 (x2 )](x − x2 ). y)ρ(y) dy G(x. Hence the characteristic curves are straight lines.
then ux + ut (dt/dx) = 0. t) with t > 0 and x > x2 . (a) Since the “rectangle” D does not touch the x axis and φ = 0 on ∂D and outside D. and f (u0 (x2 )). Thus [uφt + f (u)φx ] dxdt = Di Di +[(uφ)t + (f (u)φ)x ] dxdt. deﬁne t implicitly as a function of x. u0 (x1 ) > u0 (x2 ) > 0. ˙ ˙ (b) If the characteristic curve u(x. The solution must be x ¯ ¯ discontinuous at P . (c) If x1 < x2 . which is less than 1/f (u0 (x2 )). (i) Since (uφ)t + [f (u)φ]x = [ut + f (u)x ]φ + [uφt + f (u)φx ]. ut + f (u)ux = 0. These two equations together give dt/dx = 1/f (u) = 1/f (c). then f (u0 (x1 )) > f (u0 (x2 )) > 0. equation (25) becomes [uφt + f (u)φx ] dxdt = 0. 0) has slope 1/f (u0 (x1 )). since f > 0. and so expression (ii) becomes [uφt + f (u)φx ] dxdt = ∂Di φ[−u dx + f (u) dt] Adding the above for i = 1. (a) u = (d/ds)[u(x(s). The characteristic through (x1 . 2 and using expression (i) gives 0= ∂Di φ[−u dx + f (u) dt] + ∂D2 φ[−u dx + f (u) dt] . we have [uφt + f (u)φx ] dxdt = Di Di [(uφ)t + (f (u)φ)x ] dxdt − Di [ut + f (u)x ]φ dxdt. So these lines must cross at a ¯ point P = (¯. since these two crossing lines would give it diﬀerent values there. Therefor the curve is a straight line with slope 1/f (c). that is. t(s))] = ux x + ut t = ux f (u) + ut = 0. ¯ (d) t = (x2 − x1 )/[f (u0 (x1 )) − f (u0 (x2 ))]. (ii) (b) By Green’s theorem. and so Exercise 4(b) says ut + f (u)x = 0 there. But also ut + f (u)x = 0. t) = c (by part (a)). (that of the characteristic through (x2 . But u is C 1 on the interior of Di . [(f (u)φ)x − (−uφ)t ] dxdt = Di ∂Di [(−uφ) dx + f (u)φ dt].120 Answers for the Internet Supplement ˙ 4. 0). 6.
then we can choose a small disk B centered at P contained in D (described above) such that [−u](dx/dt) + [f (u)] > c/2 on the part of Γ inside B . A similar argument (reversing signs) works if c < 0. 2 For (b). 0) dx = 0 t≥0 t=0 2 (ii: weak) fails for some admissible φ. once with the values u1 and once with the values u2 . deﬁning weak solutions of g(u)t + f (u)x = 0. it is s( 1 u2 − 1 u2 ) = ( 1 u3 − 1 u3 ) or 2 2 2 1 3 2 3 1 1 2 1 u φt + u3 φx 2 3 dxdt + s= 2 u2 + u1 u2 + u2 2 1 . (c) Since φ = 0 outside D. applying Green’s theorem on rectangular R. 0)dx = 0 t=0 (i: weak) t≥0 holds for all admissible φ but such that 1 2 u0 (x)φ(x.Answers for the Internet Supplement 121 The union of these two boundaries traverses ∂D once and that portion of Γ within D in each direction. and using the function φ as in Exercise 4 shows that if u is a solution of g(u)t + f (u)x = 0. then [g(u)φt + f (u)φx ] dxdt + t≥0 t = 0g(u0 (x))φ(x. This is the appropriate analogy to equation (25). Now take a slightly smaller disk Bγ ⊂ B centered at P and pick φ suck that φ ≡ 1 on Bγ . But then t2 φ [−u] t1 dx + [f (u)] dt dt > c (t4 − t3 ) > 0. 0 ≤ φ ≤ 1 on the annulus B Bγ and φ ≡ 0 outside B . t1 ≤ t ≤ t2 . Thus we want u such that 1 uφt + u2 φx dxdt + 2 u0 (x)φ(x. For (a). the ﬁrst integral is the same as that of the second conclusion of part (b). (d) If [−u]s + [f (u)] = c > 0 at P . 3 u2 + u1 (ii: jump) . Since φ = 0 outside of D and on ∂D2 this becomes 0 = Γ φ{[−u] dx + [f (u)] dt}. 2 contradicting the result of part (c). Setting P = g(u)φ and Q = −f (u)φ. 8. t). The method of Exercise 11 produces the jump condition s[g(u)] = [f (u)]. this is s(u2 − u1 ) = ( 1 u2 − 2 2 1 2 2 u1 ) or 1 (i: jump) s = (u2 + u1 ). If α(t0 ) = P . 0) dx = 0. The second integral results from parameterizing the portion of Γ by α(t) = (x(t). then there are t3 and t4 with t1 < t3 < t0 < t4 < t2 and α(t) ∈ Bγ f ort3 < t < t4 .
z2 = z z . 10. 0) dx 2 0 a φ −φ dx + dt + φ(x. write reiθ − z 2 = reiθ − r eiθ 2 = rei(θ−θ ) − r 2 . For example. if we multiply (ii: weak) by 2 and insert u. we are led to consider the function u(x. t) = 1 when t ≤ 2x. 2 For any particular φ. a dxdt + 0 φ(x. 0) dx Thus (i: weak) is satisﬁed for every φ. since the jump condition (ii: jump) fails. 2x)(−2dx) 2 + 0 φ(x. use eiφ = cos φ + i sin φ. However. and the discontinuity curve Γ is given by t = 2x. ¯ . t) = 0 for x ≥ a and t ≤ T . Indeed. The factor becomes 1 2 has changed to 0=− 1 3 and the computation above now φ(x. t) = 0 when t > 2x and u(x. condition (i: weak) becomes 0 = Ω a 1 φt + φx dxdt + φ(x. (ii: weak) cannot be satisﬁed for every φ. Letting Ω be the region 0 ≤ x ≤ a and 0 ≤ t ≤ T . there are numbers T and a such that φ(x. π/2 0 which is certainly not satisﬁed for every admissible φ. and u is a weak solution of equation (i).122 Answers for the Internet Supplement If we take for u0 (x) a (Heaviside) function deﬁned by u0 (x) = 0 for x < 0 and u0 (x) = 1 for x > 0.e. u2 = 0.. 0) dx + 0 1 −φ(x. 0) dx 2 0 a T /2 = ∂Ω =− 0 a φ(x. (ii: weak) becomes 0= Ω 2 φt + φx 3 2 3. 0) dx. 2x) dx. dx/dt = 1 (u1 + u2 ) is satisﬁed. 2x)(−dx) + φ(x. Thus u1 = 1. Thus the jump condition (i: jump)(i. and multiply out.
(a) Let a. (b) Find the force acting on the particle at t = π. a2 b where 0 < a < b. . consisting of 10 questions that requires approximately 3 hours to complete. in terms of E(z). Consult your instructor if you have special concerns.Page 123 Practice Final Examination This is a practice examination. b sin t. Full solutions are at the end of the answer section following this exam. It is always important to make sure you know how to solve the problem completely if you only got it partly correct. (c) Deﬁne the function E(z) by π/2 E(z) = 0 1 − z 2 sin2 t dt. Of course there might be correct solutions that are slightly diﬀerent from those given. Good luck! 1. Describe the curve and ﬁnd the tangent line at t = π. Find a formula for the circumference of the ellipse x2 y2 + 2 = 1. so you can check your solutions and selfmark the exam. b > 0 and let a point of mass m move along the curve in R3 deﬁned by c(t) = (a cos t. Note that there are also practice quizzes and exams in the student guide. t2 ).
(a) Evaluate the integral 1 0 y 1 e−x 2 /2 dx dy. 1. and if false. 0). justify (give a brief explanation or quote a relevant theorem from the course). Suppose φ : R → Rn is a curve whose image lies in S. give an explanation. 0). y. 4. y. If true. 3. z) = z 2 − x2 − y 2 . (a) Find the extrema of the function f (x. z) = x2 + y 2 .124 Practice Final Examination (d) Let f : R3 → R be deﬁned by f (x. y. . 0. Show that ∇g(φ(0)) is perpendicular to φ (0). for every x. that is. (b) Suppose S is the level surface in Rn given by {x ∈ Rn  g(x) = 0} for a smooth function g : Rn → R. or a Counterexample. z) = exp [z 2 − 2y 2 + x2 ] subject to the two constraints g1 (x. y). C is the circle x2 + y 2 = 1 and D is the unit disk x2 + y 2 ≤ 1. 0). 0. Determine whether each of the following statements below is true or false. 2. z) = 0. 1 2 . φ(x) ∈ S. z) = −z + exp [x2 − y 2 ]. 1) is f decreasing at 50% of its maximum rate of change? (e) If f is a smooth function of (x. y. y. (1. g2 (x. (0. z) = 1 and g2 (x. where g1 (x. is it true that sin (xy) C ∂f ∂f dx = sin (xy) dy = ∂x ∂y cos (xy) y D ∂f ∂f −x dx dy? ∂y ∂x 2. (b) Evaluate x dx dy dz. D where D is the tetrahedron with vertices (0. 0. (c) Find the minimum distance between the circle x2 + y 2 = 1 and the line x + y = 4. In what directions starting from (1. y. 0.
z). y. S1 = {(x. z)  x2 + y 2 + z 2 = 1. (b) There exist conservative vector ﬁelds F and G. the following holds 1 0 2 −2 π −π 1 2 z −1 xyee (y 2 + 1)−3 2(y x2 4 x6 +x8 y 7 +5) dy dx dz = cos4 (t) sin (t) dt. oriented with the upward normal and the upper hemisphere. (e) If F is a smooth vector ﬁeld on R3 . div (∇f × ∇g) = 0. y. also oriented with the upward normal. then (∇ × F) · dS = S1 S2 (∇ × F) · dS. S2 = {(x. (d) The surface integral of the vector ﬁeld F (x. z ≥ 0}. Let W be the region in R3 deﬁned by x2 + y 2 + z 2 ≤ 1 and y ≤ x. . y. (c) By a “symmetry” argument. y. such that F × G = (x. 6. 0)  x2 + y 2 ≤ 1}. 5. and 0 ≤ z ≤ 1. 0 ≤ y ≤ 1. y. W (b) Find the ﬂux of the vector ﬁeld F = (x3 − 3x)i + (y 3 + xy)j + (z 3 − xz)k out of the region W . (a) Verify the divergence theorem for the vector ﬁeld F = 4xzi + y 2 j + yzk over the surface of the cube deﬁned by the set of (x. (a) Compute (1 − x2 − y 2 − z 2 ) dx dy dz. z) = (x + y 2 + z 2 ) i + (x2 + y + z 2 ) j + (x2 + y 2 + z) k over the unit sphere equals 4π. z) satisfying 0 ≤ x ≤ 1.Practice Final Examination 125 (a) For smooth functions f and g.
For each of the questions below. (0. (b) The ﬂux of any gradient out of a closed surface is zero. z). 8. 0). (a) If P (x. y. D where D is the region deﬁned by 1 ≤ x2 + y 2 + z 2 ≤ 4 and z ≥ x2 + y 2 . z) = i + j + z (x2 + y 2 )2 k. y. where F(x. (1. give an explanation or a counterexample. If true. . we have 1 0 0 x 0 x+y 1 y 0 0 x+y f (x. 3). 2). Let W be the threedimensional region deﬁned by x2 + y 2 ≤ 1. and x2 + y 2 + z 2 ≤ 4. y). 9. (d) If f is a smooth function of (x. and if false. z) dz dy dx = 0 f (x. and S is the surface of the cylinder x2 + y 2 ≤ 1. z ≥ 0. (a) Let D be the parallelogram in the xy plane with vertices (0. 7. y. justify (give a brief explanation or quote a relevant theorem from the course). 1). y) = Q(x. D (b) Evaluate (x2 + y 2 + z 2 )1/2 exp [(x2 + y 2 + z 2 )2 ] dx dy dz. y). including the sides and both lids. ∂y ∂x (e) For any smooth function f (x. then exy C ∂f ∂f dx + exy dy = ∂x ∂y exy y D ∂f ∂f −x dx dy. 0 ≤ z ≤ 1. then the vector ﬁeld F = P i + Qj is a gradient. (1. C is the circle x2 + y 2 = 1 and D is the unit disc x2 + y 2 ≤ 1. y. Evaluate the integral xy dx dy.126 Practice Final Examination (b) Evaluate the surface integral S F · n dS. (c) There is a vector ﬁeld F such that ∇ × F = yj. indicate if the statement is true or false. z) dz dx dy.
S . 10.Practice Final Examination 127 (a) Find the volume of W . y. (a) Compute the gradient vector ﬁeld F = ∇f . (c) With F as in (a) and S as in (b). (b) Let C be the curve obtained by intersecting the sphere x2 + y 2 + z 2 = 1 with the plane x = 1/2 and let S be the portion of the sphere with x ≥ 1/2. Draw a ﬁgure including possible orientations for C and S. Let f (x. (b) Find the ﬂux of the vector ﬁeld F = (2x − 3xy)i − yj + 3yzk out of the region W . let G = F + (z − y)i + yk. state Stokes’s theorem for this region. z) = xyzexy . and evaluate the surface integral (∇ × G) · dS.
128 Practice Final Examination .
2). 0. so c (π) = (a. b cos t. 0. 2π) = (−a. π 2 ) + t(0. 2). 2). b sin t. The acceleration is c (t) = (−a cos t. b > 0 and let a point of mass m move along the curve in R3 deﬁned by c(t) = (a cos t. −b. The derivative is c (t) = (−a sin t. The tangent line is l(t) = c(π) + tc (π) = (−a. . 2t). Solution. (a) Let a. t2 ). −tb. Thus.Page 129 Solutions to Practice Final Examination 1. (b) Find the force acting on the particle at t = π. 0. π 2 + 2tπ). This is a stretched elliptical helix. −b sin t. Describe the curve and ﬁnd the tangent line at t = π. Solution. the force is F = ma = m(a.
we get n · (−∇f (1. 1. 2. 1). y. the direction of fastest decrease at (1. . Find a formula for the circumference of the ellipse x2 y2 + 2 = 1. z) = −z + exp [x2 − y 2 ]. 1. 1. Thus. The circumference of the ellipse is π/2 C=4 0 a2 sin2 t + b2 cos2 t dt. 1. −2y exp (x2 − y 2 ). In what directions starting from (1. 1) is f decreasing at 50% of its maximum rate of change? Solution. 1. Now substitute cos2 t = 1 − sin2 t. 1) . Solution. Then. 1. (d) Let f : R3 → R be deﬁned by f (x. y. 1) is n · ∇f (1. 1). in terms of E(z). a2 b where 0 < a < b. 1)) = 1 − ∇f (1. Suppose now that n is the direction that f decreases at 50% of its maximum rate and n = 1. −1). and pull out a factor of b to give the ﬁnal desired formula: C = 4bE( 1 − (a2 /b2 )).130 Practice Final Examination Solutions (c) Deﬁne the function E(z) by π/2 E(z) = 0 1 − z 2 sin2 t dt. 1) = (−2. ∂x ∂y ∂z ∇f (x. The gradient is ∂f ∂f ∂f . 1. because the rate of change of f in the direction n at the point (1. 1) is −∇f (1. z) = = (2x exp (x2 − y 2 ). 2 .
C is the circle x2 + y 2 = 1 and D is the unit disc x2 + y 2 ≤ 1. subject to the two constraints g1 (x. ∂y ∂x ∂x∂y and The terms containing second derivatives of f cancel. z) = 1 and g2 (x. 1. Then from the preceding discussion. (a) Find the extrema of the function f (x. . z) = x2 + y 2 . the directions are those that make an angle of π/3 with the vector (−2. this is true. z) = exp [z 2 − 2y 2 + x2 ]. 2. We have ∂Q ∂f ∂2f = y cos xy + sin (xy) ∂x ∂y ∂x∂y ∂P ∂f ∂2f = x cos xy + sin (xy) . so θ = π/3. cos θ n · − ∇f (1. is it true that sin (xy) C ∂f ∂f dx+sin (xy) dy = ∂x ∂y Yes. 2. y. 1. 2. (e) If f is a smooth function of (x. cos θ = 1/2. y. y). 2 Thus. Therefore. a cone centered on the vector (−2. Let ∂f ∂x and cos (xy) y D ∂f ∂f −x dx dy? ∂y ∂x Solution. 1). y. 1) with a base angle of 2π/3. y. that is.Practice Final Examination Solutions 131 Let the angle between n and −∇f (1. 1. y. 1) . where g1 (x. 1) = 1 − ∇f (1. z) = 0. z) = z 2 − x2 − y 2 . g2 (x. 1) be denoted θ (taken to be the acute angle). P = sin (xy) Q = sin (xy) ∂f ∂y and apply Green’s theorem: P dx + Q dy = C D ∂Q ∂P − ∂x ∂y dx dy. so one gets the desired result.
x = ±1 and z = ±1. Method 2. The ﬁrst method uses Lagrange multipliers. There are two cases: i. That is. y. We present two possible solutions. 0. The constraint set is the collection of two circles x2 + y 2 = 1. so z = 0. Evaluating f at each of the eightcandidate gives a minimum of e−1 at the four points (0. x = 0: Here from the third equation. f = λ2 . Thus. This gives the ﬁve equations x2 + y 2 = 1 z 2 − x2 − y 2 = 0 2xf = 2xλ1 − 2xλ2 −4yf = 2yλ1 − 2yλ2 2zf = 2zλ2 . y = 0: Here λ1 = −f from the fourth equation. ±1). ±1) and the minimum of e−1 occurs at (0. ±1) and a maximum of e2 at the four points (±1. Method 1. y) = (0. z) = 0 ∇f = λ1 ∇g1 + λ2 ∇g2 . Thus. y = ±1 and z = ±1. z = 1 and x2 + y 2 = 1. 0). from the ﬁrst two equations. we are extremizing the function h(x. According to the Lagrange multiplier method. z = −1. (x. y. z) = 1 g2 (x. the extrema are among the solutions of the equations g1 (x. ±1). y) = exp [1−2y 2 +x2 ] over the circle x2 + y 2 = 1. the maximum of e2 occurs at (±1. for every x.132 Practice Final Examination Solutions Solution. because the function k(y) is decreasing in y 2 . from the last equation. 0. Thus. Thus. φ(x) ∈ S. λ1 = 2f . ±1. ii. Substituting into the third equation gives x = 0. ±1. . that is. Because f > 0 and. (b) Suppose S is the level surface in Rn given by {x ∈ Rn  g(x) = 0} for a smooth function g : Rn → R. the function k(y) = exp [2 − 3y 2 ] for y between −1 and 1. Show that ∇g(φ(0)) is perpendicular to φ (0). Suppose φ : R → Rn is a curve whose image lies in S. Substituting into the fourth equation gives y = 0. Thus.
y = v.Practice Final Examination Solutions 133 Solution. We wish to minimize the square of the distance function from a point (x. y. First. notice that λ1 = 0. v) = (x − u)2 + (y − v)2 . Solution. λ1 2 λ1 λ1 Substituting x = y into the last two of the preceding equations gives u = v as well. namely. Plugging these values into the distance function. The chain rule applied to g(φ(t)) = 0 gives the equality ∇g(φ(0)) · φ (0) = 0. which cannot happen since the line does not intersect the circle. y. (c) Find the minimum distance between the circle x2 + y 2 = 1 and the line x + y = 4. which proves the assertion. v) on the line. u. y. for then we would get x = u. u. u. we √ ﬁnd that the potential extrema are located at x = y = ± 2/2. which gives the following six equations x2 + y 2 = 1 u+v=4 2(x − u) = 2λ1 x 2(y − v) = 2λ1 y −2(x − u) = λ2 −2(y − v) = λ2 . v) = x2 + y 2 − 1 = 0 and g2 (x. Hence x−u 1 λ2 y−v x= =− = = y. v) = u + v − 4 = 0. f (x. From the ﬁrst and second equations. y) on the circle to a point (u. we ﬁnd the values d1 = 2 − −2 2 √ 2 √ + − 2 −2 2 2 √ =2 2+1 . subject to the constraints g1 (x. The Lagrange multiplier method gives g1 = 0 g2 = 0 ∇f = λ1 ∇g1 + λ2 ∇g2 .
0). One can also do this problem geometrically by drawing the line and the circle and the line orthogonal to them. Evaluate the integral 1 0 y 1 e−x 2 /2 dx dy. 0. Solution. 0. 0). 2 Solution. one gets 1 2−2x 0 1 2−2x 0 1 0 1/2−1/2x−1/4y 1 . Thus. First compute or observe that the plane x + 1 y + 2z = 1 2 contains the tetrahedral face passing through the points (1. 24 . (0. the minimum distance is 2 2 − 1. one has to be very careful to justify the method. 2.134 Practice Final Examination Solutions and d2 = 2 −2 2 √ Thus. 0. 0. 1 . (0. D where D is the tetrahedron with vertices (0. However. 0. 0). 0). 2 x dx dy dz = D 0 x dz dy dx xy x x2 − − dy dx 2 2 4 dx = 0 = 0 x x3 − x2 + 2 2 = 1 . both to compute the intersection points and then calculate the distance between them. in this method. 0. + √ 2 −2 2 2 √ 2 √ = 2 2 − 1. 3. (1. Drawing a ﬁgure and switching the order of integration: 1 0 y 1 e−x 2 1 /2 x 0 dx dy = 0 e−x 2 1 /2 dy dx = 0 xe−x 2 /2 dx = 1−e−1/2 . 2. Evaluate x dx dy dz. 0. 0).
but the right side clearly does not (its divergence is 3). y. Determine whether each of the following statements below is True or False. In fact. Justify (give a brief explanation or quote a relevant theorem from the course). If true. This is true.Practice Final Examination Solutions 135 4. (b) There exist conservative vector ﬁelds F and G. the following holds 1 0 2 −2 π −π 1 2 z −1 xyee (y 2 + 1)−3 2(y x2 4 x6 +x8 y 7 +5) dy dx dz = cos4 (t) sin(t) dt Solution. the left side has zero divergence. both sides are zero by symmetry. it is true and follows by equality of mixed partials. Solution. div(∇f × ∇g) = 0. such that F × G = (x. or a Counterexample. (a) For smooth functions f and g. Solution. . and if false. ∂x ∂y ∂y ∂x After taking the divergence. By part (a). (c) By a “symmetry” argument. i ∂f ∇f × ∇g = ∂x ∂g ∂x ∂f = ∂y j k ∂f ∂f ∂y ∂z ∂g ∂g ∂y ∂z ∂g ∂f ∂g ∂f ∂g ∂f ∂g − i− − j ∂z ∂z ∂y ∂x ∂z ∂z ∂x ∂f ∂g ∂f ∂g + − k. This is one of the vector identities. give an explanation. we see that all terms cancel (the student should be prepared to say which terms cancel with which terms). z). the ﬁrst is odd in the xintegral and the second is zero because it is an odd function of t. This is false.
Consider spherical coordinates in R3 : x = r cos θ sin φ y = r sin θ sin φ z = r cos φ. (e) If F is a smooth vector ﬁeld on R3 . 0)  x2 + y 2 ≤ 1}. so the integral over the unit ball is 3 × (4π)/3 = 4π. (a) Compute (1 − x2 − y 2 − z 2 ) dx dy dz. π). θ ∈ [−π. the unit circle in the plane.136 Practice Final Examination Solutions (d) The surface integral of the vector ﬁeld F (x. Let W be the region in R3 deﬁned by x2 + y 2 + z 2 ≤ 1 and y ≤ x. z ≥ 0}. y. One can also do this by Gauss’ theorem by writing the diﬀerence of the two sides as the surface integral over the boundary of the region enclosed. ∞). But the divergence is the constant 3. φ ∈ [0. with r ∈ [0. z)  x2 + y 2 + z 2 = 1. then (∇ × F) · dS = S1 S2 (∇ × F) · dS. This is true. invoking the divergence theorem and using the fact that the divergence of a curl is zero. Solution. . This is true. both sides equal the integral of F around the common boundary. S1 = {(x. S2 = {(x. y. W Solution. By Stokes’ theorem. y. 5. Solution. π). the ﬂux is the integral of the divergence. z) = (x + y 2 + z 2 ) i + (x2 + y + z 2 ) j + (x2 + y 2 + z) k over the unit sphere equals 4π. also oriented with the upward normal. oriented with the upward normal and the upper hemisphere. By the divergence theorem.
θ ∈ [−3π/4. π/4].Practice Final Examination Solutions 137 We can rewrite the conditions x2 + y 2 + z 2 ≤ 1 as (r ≤ 1) and (r sin θ sin φ ≤ r cos θ sin φ). Now we can compute (1 − x2 − y 2 − z 2 ) dx dy dz W 1 π 0 1 π/4 −3π/4 = 0 (1 − r2 ) r2 sin φ dr dθ dφ π π/4 = 0 (r2 − r4 ) dr 0 sin φ dφ −3π/4 dθ = 2 4π ·2·π = . 1]. equivalently r ∈ [0. π)) to: (r ≤ 1). 15 15 (b) Find the ﬂux of the vector ﬁeld F = (x3 − 3x)i + (y 3 + xy)j + (z 3 − xz)k out of the region W. θ ∈ [−3π/4. (sin θ ≤ cos θ) or. using Gauss’ divergence theorem (and invoking part (a)). in spherical coordinates. by r ∈ [0. π). the region W is deﬁned. 1]. Solution. π/4]. and y≤x One can also see these limits by drawing the ﬁgure containing the ball x2 + y 2 + z 2 ≤ 1 and the halfspace y ≤ x and the corresponding region in the xy plane consisting of unit disk cut by the halfplane y ≤ x. φ ∈ [0. we get that the ﬂux of F out of the region W is F · dS = ∂W W (∇ · F) dV (3x2 + 3y 2 + 3z 2 − 3) dx dy dz W = = (−3) (1 − x2 − y 2 − z 2 ) dx dy dz W = (−3) 4π 4π =− . ∇·F = Let’s ﬁrst compute ∂ 3 ∂ ∂ (x −3x)+ (y 3 +xy)+ (z 3 −xz) = 3x2 +3y 2 +3z 2 −3. Therefore. 15 5 . ∂x ∂y ∂z Now. which is equivalent (because sin φ is nonnegative for φ ∈ [0.
n6 = (0. z ≤ 1. . −1. y = 1. both sides are equal. 0) S4 : 0 ≤ s. z ≤ 1. 1 1 0 0 1 div F dV = W 0 (4z + 3y) dx dy dz = 2 + 3 7 = . 1. z = 1. −1) S6 : 0 ≤ x.138 Practice Final Examination Solutions 6. 0) S2 : 0 ≤ y. 2 2 The lefthand side can be evaluated by writing S = S 1 ∪ S 2 ∪ S 3 ∪ S 4 ∪ S5 ∪ S 6 . where S is the surface of the cube and W is the cube itself. 0. 0. 0. We evaluate each side and check that they are equal. y ≤ 1. (a) Verify the divergence theorem for the vector ﬁeld F = 4xzi + y 2 j + yzk over the surface of the cube deﬁned by the set of (x. n4 = (0. y. we get F · n dS S 6 = i=1 Si 1 F · n dS 1 1 1 1 1 =0+ 0 0 4z dy dz + 0 + 0 0 dx dz + 0 + 0 0 y dx dy =2+1+ 1 7 = . y ≤ 1. and 0 ≤ z ≤ 1. Thus. z = 0. y = 0. n5 = (0. z) satisfying 0 ≤ x ≤ 1. in this case. 0) S3 : 0 ≤ x. 0 ≤ y ≤ 1. 0) S5 : 0 ≤ x. z ≤ 1. x = 1. Solution. 1). so the divergence theorem checks in this case. the divergence theorem states that F · n dS = S W div F dV. n2 = (1. x = 0. z ≤ 1. where S1 : 0 ≤ y. n1 = (−1. 0. First of all. n3 = (0. The righthand side is. 2 2 Thus.
so it is the easiest to evaluate the integral in cylindrical coordinates: 1 0 0 2π 0 1 r · (r2 )2 dr dθ dz = π 2π = . the integral is 1 x+2 xy dx dy = D 0 1 x xy dy dx x 0 1 = = 0 x2 (x + 2)2 − 2 2 dx 2(x2 + x) dx 2 3 x + x2 3 1 = = 0 5 . Use Gauss’ divergence theorem in space: F · n dS = S W (div F) dx dy dz. The region is ysimple with sides given by the lines x = 0 and x = 1. 0). 1). Therefore. including the sides and both lids. D Solution. 0 ≤ z ≤ 1. Evaluate the integral xy dx dy. 3 . 6 3 7. and S is the surface of the cylinder x2 + y 2 ≤ 1. y. z) = i + j + z(x2 + y 2 )2 k. Here. 2). ∂x ∂y ∂z The region W is a cylinder. (1. (1. where F(x. (a) Let D be the parallelogram in the xy plane with vertices (0.Practice Final Examination Solutions 139 (b) Evaluate the surface integral S F · n dS. and top and bottom by the lines y = x and y = x + 2. (0. 3). Solution. div F = ∂ ∂ ∂ (1) + (1) + z(x2 + y 2 )2 = (x2 + y 2 )2 .
One should ﬁrst draw a sketch. If true.140 Practice Final Examination Solutions (b) Evaluate (x2 + y 2 + z 2 )1/2 exp [(x2 + y 2 + z 2 )2 ] dx dy dz. For each of the questions below. The region in question is that between two spheres and inside a cone centered around the z axis. This is false. then the vector ﬁeld F = P i + Qj is a gradient. and if false. y). give an explanation or a counterexample. one ﬁnds that a side of the cone makes an angle of π/4 with the z axis. y) = Q(x. Solution. justify (give a brief explanation or quote a relevant theorem from the course). indicate if the statement is true or false. (b) The ﬂux of any gradient out of a closed surface is zero. By drawing a careful ﬁgure and using a little trigonometry. Denoting the integral required by I. D where D is the region deﬁned by 1 ≤ x2 + y 2 + z 2 ≤ 4 and z ≥ x2 + y 2 . For example. (a) If P (x. 2 8. we get 2π π/4 0 1 2 2 I= 0 ρ eρ ρ2 sin ϕ dρ dϕ dθ exp (ρ4 )ρ3 dρ 1 2 4 1 = 2π 1 − √ 2 π 1− 2 π = 1− 2 = 4 1 √ eρ 2 1 1 √ (e16 − e). . Solution. xi + xj is not a gradient because it fails to satisfy the crossderivative test.
Practice Final Examination Solutions
141
Solution.
This is false. For example, the vector ﬁeld F(r) = r
is the gradient of f (r) = r 2 /2, yet its ﬂux out of the unit sphere is, either by direct evaluation of the surface integral, or by the divergence theorem, 4π. (c) There is a vector ﬁeld F such that ∇ × F = yj. Solution. This is false. We know that div curl F = 0 for all vector ﬁelds F, but div(yj) = 1, and so no such F can exist. (d) If f is a smooth function of (x, y), C is the circle x2 + y 2 = 1 and D is the unit disc x2 + y 2 ≤ 1, then exy
C
∂f ∂f dx + exy dy = ∂x ∂y
exy y
D
∂f ∂f −x dx dy. ∂y ∂x
Solution.
This is true. To see why, let ∂f ∂x and Q(x, y) = exy ∂f , ∂y
P (x, y) = exy
so that the lefthand side of the expression in the problem is the line integral of P dx + Q dy. By Green’s theorem we have P dx + Q dy =
C D
∂Q ∂p − ∂x ∂y
dx dy.
To work out the righthand side in Green’s theorem, we calculate the partial derivatives: ∂Q ∂f ∂2f = yexy + exy ∂x ∂y ∂x∂y ∂P ∂f ∂2f = xexy + exy . ∂y ∂x ∂y∂x By the equality of mixed partials for f, we get the desired equality from Green’s theorem. (e) For any smooth function f (x, y, z), we have
1 0 0 x 0 x+y 1 y 0 0 x+y
f (x, y, z) dz dy dx =
0
f (x, y, z) dz dx dy.
142
Practice Final Examination Solutions
Solution. The lefthand side represents the volume integral of f over the region W that is between the xy plane and the plane z = x + y, and over the region in the xy plane bounded by the x axis, the line x = 1, and the line x = y. The righthand side represents the volume integral of f over the region W that is between the xy plane and the plane z = x + y, and over the region in the xy plane bounded by the x axis, the line y = 1, and the line x = y. The student should draw a sketch of these two regions in the xy plane. One sees that these two regions are diﬀerent, so the two integrals need not be equal. Thus, the assertion is false. 9. Let W be the threedimensional region deﬁned by x2 + y 2 ≤ 1, z ≥ 0, and x2 + y 2 + z 2 ≤ 4.
(a) Find the volume of W. Solution 1. The student should draw a sketch. Setting up the volume as a triple integral, one gets Volume = V =
W
dx dy dz √ 2
0
4−x −y 2
=
D
dz dy dx 4 − x2 − y 2 dy dx,
D
=
where D is the unit disc in the xy plane. Changing variables to polar coordinates gives
1 2π
V =
0 0 1
4 − r2 r dr dθ 4 − r2 r dr.
0
= 2π
This is now integrated using substitution, and after a little computation, one gets the answer V = √ 2π (8 − 3 3). 3
Solution 2. The student should draw a sketch. Using evident notation, we have V (W ) = V (cap) + V (cylinder of height h),
Practice Final Examination Solutions
143
where the region is divided into a cylinder with a certain height h and the portion of the sphere above it. Now V (cyl.) = A(base) · height = π · h. To determine h, note that h is the value of z where the cylinder x2 + y 2 = 1 and the sphere x2 + y 2 + z 2 = 4 intersect. √ such At an intersection, 1 + z 2 = 4, and so z 2 = 3, that is, z = 3 = h. Therefore, √ V (cyl.) = π 3. To ﬁnd the volume of the cap, we think of it as a capped cone minus a cone with a ﬂat top. We can ﬁnd the volume of the capped cone using spherical coordinates, and we know the volume of the cone with the ﬂat top (the base) is 1 A(base) · height. 3 First of all, using spherical coordinates,
2π ϕ0 0 0 2
V (capped cone) =
0
ρ2 sin ϕ dρ dϕ dθ.
To ﬁnd ϕ0 , refer to the ﬁgure that we suggested be drawn for this problem, and one sees that for the relevant right triangle, √ 3 Adjacent π cos ϕ0 = = and so ϕ0 = . Hypotenuse 2 6 Therefore,
2π π/6 0 0 2
V (capped cone) =
0
√ 3 16π ρ sin ϕ dρ dϕ dθ = 1− . 3 2
2
However, 1 V (cone with the ﬂat top) = Area of base times height 3 √ π 3 1 √ = π 3= . 3 3 Therefore, √ √ 16π 8π 3 π 3 V (cap) = − − 3 3 3 √ √ √ 16π 8π 3 π 3 3π 3 V (W ) = − − + 3 3 3 3 √ √ 16π 6π 3 2π = − = (8 − 3 3). 3 3 3
and ﬁnally,
The ﬂux is given by F · dS. For example. One has to choose an orientation for the surface and the curve. (b) Let C be the curve obtained by intersecting the sphere x2 + y 2 + z 2 = 1 with the plane x = 1/2 and let S be the portion of the sphere with x ≥ 1/2. we use the divergence theorem. y. then the curve should be marked with an arrowhead indicating a counterclockwise orientation when the curve is viewed from the positive xaxis. Using the deﬁnition of the gradient. Solution. Solution. 3 10. . which gives F · dS = ∂W W div F dV (2 − 3y − 1 + 3y) dV W = = V (W ) = √ 2π (8 − 3 3). one gets F(x. Draw a ﬁgure including possible orientations for C and S. (a) Compute the gradient vector ﬁeld F = ∇f . The student should draw a ﬁgure at this point. y. state Stokes’ theorem for this region.144 Practice Final Examination Solutions (b) Find the ﬂux of the vector ﬁeld F = (2x − 3xy)i − yj + 3yzk out of the region W. z) = xyzexy . ∂W To compute this. z) = exy [(yz + xy 2 z)i + (xz + x2 yz)j + xyk]. Solution. Let f (x. if the normal points toward the positive x axis. Stokes’ theorem for this region states that for a smooth vector ﬁeld K: K · ds = C S (∇ × K) · dS.
S Solution. we do this by letting √ √ 3 3 1 y= cos t x= and z= sin t 2 2 2 for t ∈ [0. let G = F + (z − y)i + yk. we shall parameterize C. 0. The End . the line integral of H is 2π H · ds = C 0 H(c(t)) · c (t) dt 3 4 3 4 2π = = (sin t − cos t.Practice Final Examination Solutions 145 (c) With F as in (a) and S as in (b). To evaluate the line integral. and so ∇ × G · dS = S S ∇ × H · dS = C H · ds by Stokes’ theorem. We can write G = F + H. cos t) dt 0 2π cos2 t dt. the required line integral of ∇ × G is 3π/4. 0. and evaluate the surface integral (∇ × G) · dS. ∇ × F = 0. where H = (z − y. 2π]. we get 3π/4. cos t) · (0. Thus. − sin t. 0 Remembering that the average of cos2 θ over the interval from 0 to 2π is 1/2. Because F is a gradient. Thus. y). we can evaluate the given integral as follows: (∇ × G) · dS = S S ∇ × (F + H) · dS ∇ × F · dS + S S = ∇ × H · dS. Thus.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.