Vector Analysis and Classical

Chapter 3
Vector Analysis and Classical

Electromagnetic Theory
3.1 Functions of Several Variables
3.1.1 Functions, Sets, and the Gradient

In the previous chapter, we saw how the tools of vector analysis work in
modelling the mechanics of point masses and rigid bodies. In this chapter
we will see how the tools of vector analysis work in continuum mechanics,
that is in the study of continuous media such as fluids, heat flow, and
electromagnetic fields.
Mathematically, a continuous medium is described using certain func-
tions, called scalar and vector fields. Recall that a function is a corre-
spondence between two sets A and B so that to every element A & K there
corresponds at most one element f(A) G B. Despite its simplicity, it was
only in 1837 that the German mathematician JOHANN P E T E R GUSTAV
LEJEUNE DIRICHLET (1805-1859) introduced this definition of a function.
We write / : A -> B.
For most functions in this chapter, the set A will be either R 2 or
R , and the set B, R, R 2 , or R 3 . When B = R, the collection of pairs
3
{(P, f(P)), P G A} is called a s c a l a r f i e l d ; the function / is also called

a scalar field. When B = A, that is, when f(P) is a vector, the collection of
pairs {(P, f(P)), P € A} is called a v e c t o r f i e l d ; the function / is also
called a vector field. Following our convention from the previous chapters,
we will denote vector fields with bold-face letters.
Let us review the main definitions related to the sets in R 2 and R 3 .
Recall that the distance between two points A, B in R™, n = 2,3, is denoted
by \AB\. A neighborhood of a point A is the set {B : \AB\ < b] for some
b > 0, that is, an open disk in R 2 and an open b a l l in R 3 , with center
121
122 Functions of Several Variables
at A and radius b. A point in a set is called i n t e r i o r if there exists a

neighborhood of the point that lies entirely in the set.
A point P is called a boundary point of the s e t G if every neigh-
borhood of the point contains at least one point that is not in G, and at
least one point that belongs to G and is different from P. The collection of
all boundary points of G is called the boundary of G and is denoted by
dG. FOR EXAMPLE, the boundary points of the set G = {B : \AB\ < 1}
are exactly the points P satisfying \AP\ = 1, so that dG — {P : \AP\ = 1}
is the circle with center at A and radius 1.
A point P in a set is called i s o l a t e d if there exists a neighborhood of
P in which P is the only point belonging to the set.
A set is called
• Bounded, if it lies entirely inside an open ball of sufficiently large radius.

• Closed, if it contains all its boundary points.
• Connected, if every two points in the set can be connected by a con-
tinuous curve lying completely in the set.
• Open, if for every point in the set there exists a neighborhood of this
point that is contained in the set. In other words, all points of an open
set are interior points.
• Domain, if it is open and connected.
• Simply connected, if it has no holes. More precisely, consider a curve
that lies entirely in the set and assume that this curve is simple, closed,
and continuous (see page 25). The set is simply connected if there is
a surface that lies entirely in the set and has this curve as the bound-
ary. For a planar set, this condition means that every simple closed
continuous curve in the set encloses a domain that lies entirely in the
set.
The c l o s u r e of a s e t is the set together with all its boundary points.

The complement of a set is the set of all points that are not in the set.
F O R EXAMPLE, the closure of the open set {P : \OP\ < 1} is {P :
\OP\ < 1}, and the complement of that open set is the closed set {P :
\OP\ > 1}. Every disk or ball is both connected and simply connected,
while the set {P : 0 < \OP\ < 1} is connected, and is simply connected in
M3, but not simply connected in E.2 (make sure you understand why).
EXERCISE 3.I.I.' 4 Give an example of a set in R 3 that is not bounded, is
neither open nor closed and is neither connected nor simply connected.
Intuitively, it is obvious that a simple, closed, continuous curve (also
Functions, Sets, Gradient 123
known as a Jordan curve) divides the plane into two domains, and the
domain enclosed by the curve is simply connected, but the rigorous proof
of this statement is surprisingly hard. The French mathematician M A R I E
ENNEMOND CAMILLE JORDAN (1838-1922) was the first to realize the im-
portance and difficulty of the issue. Even though Jordan's proof later turned
out to be wrong, the result is known as the Jordan curve theorem. It was
the American mathematician OSWALD VEBLEN (1880-1960), who, in 1905,
produced the first correct proof. We will take the Jordan curve theorem
for granted. The analog of this result in space, that a closed continuous
simple (that is, without self-intersections) surface in R 3 separates R 3 into
interior and exterior domains, is even harder to prove, and we take it for
granted as well.
We will now discuss continuity of functions on R n , n = 2,3. A scalar
field / is called continuous a t the p o i n t A if / is defined in some neigh-
borhood of A and
lim \f(A)-f(B)\=0. (3.1.1)
Since the point A in the above definition is fixed, the condition \AB\ —» 0
means that B is approaching A: B —> A. Then (3.1.1) can be written in
the form that is completely analogous to the definition of continuity for
functions of a real variable:
lim f(B) = f(A).
Similarly, a vector field F is called continuous at a point A if F is defined

in some neighborhood of A and
lim ||F(J4)-F(B)||=0. (3.1.2)

|AB|->0
Next, we discuss the question of differentiability for scalar fields. Let

/ : A H R be a scalar field, where A is either R 2 or R 3 . Let u
be a unit vector and, for a real t, define the point Bt so that ABt =
tu. A function / is called d i f f e r e n t i a b l e a t t h e p o i n t A i n t h e
d i r e c t i o n of u if the function g(t) = f(Bt) is differentiable for t = 0.
We also call ^'(0) the ( d i r e c t i o n a l ) d e r i v a t i v e of / a t A i n t h e
d i r e c t i o n of u and denote this directional derivative by Dnf(A).
The function / is called dif f e r e n t i a b l e a t t h e p o i n t A if / is de-

fined in some neighborhood of A and there exists a vector a so that
lim \f(B)-f(A)-a.AB\={
(3.1.3)
\AB\-^0 \AB\
Alternatively, for every unit vector u,
lha\KBt)-f(A)-ta.u\^
(3.1.4)
t-^o t
where ABt = t u. If / is differentiable at the point A, then the correspond-

ing vector a in (3.1.3) is denoted by V / ( J 4 ) or grad/(A) and is called
the g r a d i e n t of / at A. We say that the function / is continuously
d i f f e r e n t i a b l e i n a domain G if the vector field V/(^4) is defined at
all points A in G and the vector field V / is continuous in G. Similarly, a
vector field F is called continuously d i f f e r e n t i a b l e in G if, for every
unit vector n, the scalar field F • n is continuously differentiable in G.
A scalar or vector field is called smooth if it is continuously differentiable
infinitely many times.
EXERCISE 3.1.2. (a)c Verify that if f is differentiable at the point A,
then f is differentiable at A in every direction u, and
Duf{A) = Vf{A)-u. (3.1.5)
(b)A What does it mean for a vector field F to be three times continuously
differentiable in a domain G?
By (3.1.5) and the definition of the dot product (1.2.1), page 9, we have
-\\Vf(A)\\<Dzf(A)<\\Vf(A)\\,
which shows that the most rapid rate of increase of / at the point A is in
the direction of V/(^4), and the most rapid rate of decrease of / at the
point A is in the direction of — Vf(A).
Given a point A, a direction at A can be defined by another point B with
the unit vector UQ = -A^m in the direction of AB. Then (3.1.3) becomes
f(B) = f(A) + (Vf(A) • uB) \AB\ + s(B), (3.1.6)
where lim| A B |_ 0 |e(JB)|/|AB| = 0. The number V / ( A ) • UB is

the ( d i r e c t i o n a l ) d e r i v a t i v e of / at p o i n t A i n t h e d i r e c t i o n
of p o i n t B.
EXERCISE 3.I.3. 5 Show that if V / = 0 in a domain G, then f is constant

in G. Hint: For point P g G and a unit vector u, define the point Bt, t £ R,
so that PBt = tu. Using the definition of open set show that Bt is in G for
sufficiently small \t\. Apply the mean-value theorem to g(t) = f(Bt) to show that
f(Bt) = f(P)- Given another point Pi in G, connect P and Pi with a continuous
curve in G, and apply the above argument several times as you move along the
curve from P to Pi.
Let us now introduce cartesian coordinates in the space A. Then every
point P will have coordinates (x, y) or (x, y, z) and we write f(P) = f(x, y)
or f(P) = f(x,y,z). Recall that the p a r t i a l d e r i v a t i v e fx is computed
by differentiating / with respect to x and treating all other variables as
constants:
fx{x0,yo)= lim - - - (3.1.7)

Ai-»0 X
EXERCISE 3.1.4. C Verify that, in cartesian coordinates, fx = D\f.

An alternative notation for fx is df/dx. The partial derivatives fy, fz
are defined similar to (3.1.7) and coincide with the directional derivatives
in the j and K directions, respectively. The higher-order partial derivatives
are defined as partial derivatives of the corresponding lower-order partial
derivatives:
/** - dx , Jxy - dy , Jvx - Qx , etc.
EXERCISE 3.1.5? For two scalar fields f,g, verify the following properties of
the gradient, generalizing the product and quotient rules from one-variable
calculus:
V ( / 5 ) = fVg + gVf, V(f/g) = (gVf - fVg)/g2.
The next two results should be familiar from a course in multi-variable

calculus:
• If the function f = f(x,y,z) has partial derivatives fx,fy,fz in some
neighborhood of the point A and the derivatives are continuous at A, then
the function is differentiate at A and
V / ( A ) = fx(A) i + fy(A) j + fz(A) «• (3-1-8)

• If the partial derivatives fxy and fyx exist and are continuous in some
neighborhood of the point A, then fxy(A) = fyx(A); the result is known
as C l a i r a u t ' s theorem, after the French mathematician ALEXIS CLAUDE
CLAIRAUT (1713-1765).
EXERCISE 3.1.6. C Let r = r(t) be a vector-valued function, and f, a scalar

field. Define the function g(t) = f(r(t)), where f(r(t)) = f(P) if' r(t) —
OP for some fixed reference point O. Assume that r is differentiable at
the point t = to and the function f is differentiable at the point A, where
0l = r(to). Show that the function g is differentiable at the point t = to
and
g'(t0) = Vf(A)-r'(to). (3.1.9)
Try to argue directly by the definitions without using partial derivatives.

Hint: by (3.1.6), g(t) - g(t0) « Vf(A) • (r(t) - r{t0)).
If r = r(t) is a vector-valued function of one variable and G is a vec-
tor field, then R(t) = G(r(t)) is a vector-valued function of one variable.
How to find the derivative R'(t) in terms of r and G? In cartesian co-
ordinates, G — G\ i + G23 + G3K, and G\,G2,G^ are scalar fields. By
(3.1.9), we conclude that R'{t) = r ' ( t ) - V G i ( r ( t ) ) i + r ' ( t ) - V G 2 ( r ( t ) ) j +
r'(t) • VG3(r(t))/«. To make the construction coordinate-free, we define
the vector (r'(t) • V)G(r(t)) so that, for every fixed unit vector n,
((r'(i) • V)G(r(*))) • n = r'(t) • ( V ( G • n)(r(t))).
With this definition,
~G(r(t)) = (r'(t) • V)G(r(t)). (3.1.10)
More generally, for two vector fields F, G, we define the vector field
(F • V ) G so that, for every fixed unit vector n,
n-((F-V)G) =F-(V(G-n)). (3.1.11)
The expressions of the type (F • V).F often appears in the equations de-
scribing the motion of fluids.
Given a scalar field / and a real number c, the set of points {P : f(P) —
c} is called a l e v e l s e t of / . In two dimensions, that is, for A = M2, the
level sets of / are curves. In three dimensions (A = M3), the level sets of
/ are surfaces. This observation leads to several possible ways to define

curves and surfaces.
Consider the space R 2 with cartesian coordinates (x,y). Then each of
the following defines a curve in R 2 :
• A graph y — g(x) of a function of one variable.
• A level set f(x, y) = c of a function of two variables.
• A vector-valued function r(t) = x(t) i+y(t)j, to <t < t\, of one variable.
EXERCISE 3.1.7. (a)B Give an example of a curve that cannot be repre-

sented as a graph y = g(x). (b)A+ Is there a continuous curve that cannot
be represented as a level set of a function of two variables?
Consider the space R 3 with cartesian coordinates (x,y,z). Then each
of the following defines a surface in R 3 :
• A graph z = f{x,y) of a function of two variables;
• A level set F(x,y,z) = c of a function of three variables;
• A a vector-valued function r(u, v) = x(u, v)i + y(u, v) j+ z(u, v) k. of two
variables; this is known as a parametric r e p r e s e n t a t i o n of the surface.
EXERCISE 3.1.8. (a)B Give an example of a surface that cannot be rep-
resented as a graph z = f(x,y). (b)B Think of at least two different ways
3 A+
to represent a curve in R . (c) Is there a surface that can be repre-
sented parametrically using a continuous function r = r(u,v) but cannot
be represented as a level set of a function of three variables?
EXERCISE 3.1.9. Below, we summarize some geometric properties of the
gradient that are usually studied in a vector calculus class.
(a) Let f(x,y) = c be a curve. Let {xo,yo) be a point on the curve and
assume that the function f is differentiable at (xo,yo)- (i) Show that the
curve is smooth at the point (xo,yo) if and only if V / ( x o , j/o) ¥" 0. (ii)
Use (3.1.9) to show that V/(:ro,2/o) is perpendicular to the unit tangent
vector of the curve at the point (xo,yo)- Hint: in this case, g(t) = c and so
g'(to) = 0. (Hi) Write the equation of the normal l i n e to the curve at the
point (xo,yo). Hint: the line has V/(xo,j/o) as the direction vector.
(b) Let F(x, y,z) = c be a surface. Let Po = (xo, yo, ZQ) be a point on the
surface, and C, a smooth curve on the surface passing through the point Po.
Assume that the function F is differentiable at PQ. (i) Show that 'VF(PQ)
is perpendicular to the unit tangent vector to C at PQ. Hint: same arguments
as in part (a), (ii) Write the equation of the tangent plane to the surface
at the point PQ. Hint: the plane has Vi r (Fo) as the normal vector.
(c) Let r = r(u,v) be a parametric representation of the surface. As-

sume that the partial derivatives ru and rv exist, are continuous, and are
not equal to zero at a point (UQ,VO). Let Po be the point on the sur-
face corresponding to the position vector T{UQ,VQ), andC, a smooth curve
on the surface passing through the point PQ. (i) Verify that the vector
ru{uo,vo) x rv(uo,vo) is perpendicular to the unit tangent vector to C at
Po- Hint: C is represented by a vector function r(u(t),v(t)) for suitable functions
u,v; at the point Po this curve has a tangent vector ruv! + rvv'. (ii) Write the
equation of the tangent plane to the surface at the point Po. Hint: the
plane has ru(uo,vo) x rv(uo,vo) as the normal vector.
The same surface can have different representations. The graph z =

f(x,y) is a level set F(x,y,z) = 0 with F(x,y,z) — f(x,y) — z or
F(x,y,z) = z — f(x,y). The same graph can also be written in the
parametric form r(u,v) — ui + vk + f(u,v)k. Similarly, the vector
r — x(u, v) i + y(u, v)j+ z(u, v) k defines a parametric representation of
the level set F(x, y,z) = c if and only if F(x(u, v), y(u, v), z(u, v)) = c for
all (u,v). In particular, one can write the level set in parametric form by
solving the equation F(x,y,z) = c for one of the variables x,y,z. FOR
2 2 2
EXAMPLE, the hemisphere x + y + z = 4, x > 0, can be written as
x = y/A — y2 — z2. Taking u = y and v = z, we find the equivalent para-
metric representation r(u, v) = %/4 — u2 — v2 i+uj+v K, where u2+v2 < 4.
The most convenient representation of the given surface depends on the
particular problem.
EXERCISE 3.1.10. c Verify that the surface x = ^/4 — y2 — z2 can be rep-

resented as r(u,v) = 2cosvz + 2 c o s u s i n v j + 2sinusint;K, 0 < u < 2TT,
0<v< IT/2.
EXAMPLE. Consider the function f(x, y) — 2x 2 +3y 2 . This function has
continuous partial derivatives of every order everywhere in K 2 : fx = Ax,
fxx = 4, fy = 6y, fyy = 6, and all other derivatives are equal to zero. Thus,
Vf(x,y) = Axi + 6yj. Consider the point A = (1,2). Then V / ( A ) =
Al + 12j, so that the most rapid increase of / at A is in the direction of
i + 3J, with the rate ||Vf(A)\\ — Ay/16. Similarly, the most rapid decrease
of / at A is in the direction of - i — 3j, with the rate - | | V / ( J 4 ) | | = —4y/I6.
T h e rate of change of / at A in the direction toward the origin is - 2 8 / \ / 5
and is computed by (3.1.6) as follows: B — (0,0), AB = — i — 2j, UB =
-(i+2j)/y/5, Vf(A) • uB = -28/V5.
Next, consider the surface z = 2x 2 + 3y2. Writing 2x2 + 3y2 — z = 0,
we conclude that the vector Ax + 12j — k defines the plane tangent to the
surface at the point PQ = (1,2,14). The equation of the plane at PQ is
4x + 12y — z = 14 : note that x = 1, y = 2, z = 14 must satisfy the
equation, and you get the number 14 on the right by plugging in these
values for x,y,z.
EXERCISE 3.1.11. B Verify that r(u,v) = ui + vj + (2u2 + 3v2)k and

r(u,v) = (v/y/2) cosuz4- (v/y/3) sinuj+v2 k are two possible parametric
2 2
representations of the surface z = 2a: 4- 3y . Verify that, for both repre-
sentations, the vector ru x rv at the point (1,2,14) is parallel to the vector
4i+12j- k.
Recall that P is a c r i t i c a l point of a scalar field / if either V / ( P o ) —
0 or / is defined but not differentiable at PQ. HP is not a critical point of
/ , then — V / ( P ) is the direction of the most rapid decrease of the function
/ at the point P . This observation suggests the following iterative method
for finding the point of local minimum of / : given a reference point O, a
starting point Pi, and a collection of positive numbers ak, k = 2 , 3 , . . . ,
define the points Pj, recursively by
OPk+i =OPk- o fc V/(Pfc). (3.1.12)
Sometimes it is convenient to identify the point Pfc with the position vector
rfc = OPk and to write (3.1.12) in an equivalent form as
T-fc+i = r f c -akVf(rk)-
The new point Pk+\ can be denned as long as V'/(Pfc) ^ 0. If the sequence
of the points Pk, k > 1, converges as k —> oo, then the limit P* must be the
point of local minimum of / ; otherwise V / ( P * ) ^ 0 and the sequence can
be continued. As a result, V/(Pfc) —> 0, and in practice, the computations
are stopped as soon as ||V/(Pfc)|| is sufficiently small. Special choices of the
numbers ak can speed up the convergence. This method is known as the
method of s t e e p e s t descent. It was one of the first numerical methods
for minimizing a function. The method can converge very slowly, but is
easy to implement. There exist more sophisticated numerical methods for
finding critical points.
EXERCISE 3.1.12. A (a) Draw a picture illustrating (3.1.12). (b) Write
a computer program that implements (3.1.12) to find the local minimum
of the function f(x,y) = 2x2 + 3y2. Experiment with various starting
points, various values of ak and different stopping criteria (for example,
m
|| V/(Pfc)|| < 10 for some positive integer m).
3.1.2 Integration and Differentiation

Let us recall the definition of the Riemann i n t e g r a l from the one-variable
calculus:
l f{x)dx = lim V/(4)Aa;fc,

Ja maxAn-40^
where a = xo < • • • < XN = b is a partition of the interval [a,b\, Axk =

Xk — Xk-i, and x*k is a point in the interval [xk-i,Xk\. By definition, the
function / is Riemann-integrable if the limit exists and does not depend on
the particular sequence of partitions or on the choice of the points x^. The
same definition is extendable to real-valued functions / defined on a set G
other than an interval, as long as there exists a function m that measures
the sizes of the subsets of the set G; we will not discuss the deep question
of measurability. Using the measure m, we define
N
r
fdm= lim V/(P*)rn(Gfc), (3.1.13)
JG maxm(Gfc)-»0f^
where Gi,..., Gjv are mutually disjoint (Gk C\ Gm = 0 ) sets such that
their union |J f c = 1 Gfe contains the set G and every set Gk is measurable
(the size m(Gfc) of Gk is defined) and has a non-empty intersection with G
(Gfc f) G ^ 0 ) ; Pk is a point in Gk f] G. Note that definition (3.1.13) is not
tied to any coordinate system.
We will integrate functions defined on intervals, curves, planar regions,
surfaces, and solid regions:
Set Measure m dm
Interval Length dx
Curve Arc length ds
Planar region Area dA
Surface Surface Area da
Solid region Volume dV
The limit in (3.1.13) must be independent of the particular choice of

the sets Gfc and the points Pk, which, in general, puts certain restrictions
on both the function / (integrability) and the set G (measurability). If / is
integrable over a measurable set G, then the average value of / over G
Integration and Differentiation 131
is, by definition,
lG= fdm (3 Li4)

w)L - -
The idea of the Riemann integral was introduce by the German mathemati-
cian G E O R G FRIEDRICH BERNHARD RIEMANN (1826-1866) in 1854 in his
H a b i l i t a t i o n dissertation (Habilitation is one step higher than Ph.D. and
is the terminal degree in several European countries).
We now summarize the main facts about line, area, surface, and volume
integrals.
LINE, OR PATH, INTEGRALS. The set G in (3.1.13) is a curve C defined
by the vector-valued function r = r(t), a < t < b. The set is measurable
and has the line element ds if C is piece-wise smooth; see pages 28-29.
The l i n e i n t e g r a l of the f i r s t kind Jcfds is defined for a continu-
ous scalar field / by the equality
[ fds= [ f(r(t))\\r'(t)\\dt, (3.1.15)

JC Ja
where f{r(t)) is the value of the scalar field / at the point on the curve
with position vector r{t). If f{P) represents the linear density of the curve
at the point P, then Jc f ds is the mass of the curve.
The l i n e i n t e g r a l of the second kind LF-dr is defined for a
continuous vector field F by the equality
f Fdr= f F(r{t))-r'{t)dt, (3.1.16)

JC Ja
where F(r(t)) is the value of the vector field F at the point on the curve
with position vector r(t). For a simple closed curve C, the corresponding
line integral of the second kind is denoted by §CF • dr and is called the
c i r c u l a t i o n of F along C. The line integral of the second kind can be
reduced to the line integral of the first kind:
L'-*-L F-uds, (3.1.17)
where u is the unit tangent vector to the curve. The direction of u defines
the orientation of C; the integral reverses sign if the orientation of C is
reversed. In cartesian coordinates, F = Fi i + F2 j + F3K, and JCF -dr is
often written as j c F\dx + i^dy + F$dz.
We say that F is a p o t e n t i a l vector field with potential / if F = V /

for some scalar field f.UF = V / , then, by (3.1.9),
F(r(t))-r'(t) = ftf(r(t)),
and the value of the integral (3.1.16) depends only on the starting and
ending points r(a), r(b) of the curve and not on the curve itself:
I Vf-dr = f(r(b))-f(r(a)). (3.1.18)
In particular, §c V / • dr = 0 for every simple closed curve C.

The alternative name for a potential vector field is conservative, which
comes from the physical interpretation of the line integral. If F is a force
acting on a point mass moving along C, then Jc F-dr is the work done by the
force along the curve C. By the Second Law of Newton, F(r(t)) = mr "(t),
so that F(r(t)) • r'(t) = (l/2)md\\r'(t)\\2/dt = d£K(t)/dt, where £K is
the kinetic energy; verify this. For a potential force field F, we write
F = - V V , where V is a scalar function, called the p o t e n t i a l energy.
Then
J F-dr = £K(b) - SK(a) = V(r(a)) - V(r(b)),
so that £K{O) + V(r(a)) — £*:(&) + V(r(b)), which means that the total
energy £ # + V is conserved as the point mass moves along C.
In general, we say that a vector field F has the path independence
p r o p e r t y in a domain G if §c F • dr = 0 for every simple closed piece-wise
smooth curve C in G. The following result should be familiar from a course
in multi-variable calculus.
Theorem 3.1.1 A continuous vector field has the path independence

property in a domain G if and only if the field has a potential.
EXERCISE 3.1.13. c Prove the above theorem. Hint: in one direction, use
(3.1.18). In another direction, verify that, as stated, the path independence prop-
erty implies that, for every two points A,B inG the value of the integral fcF-dr
does not depend on the particular choice of the curve C in G connecting the
points A and B. Then fix a point A in G and, for each point P in G define
f{P) = fcF • dr, where C is a simple piece-wise smooth curve starting at A and
ending at P. Path independence makes f a well-defined function of P. Verify
that V / = F.
T H E AREA INTEGRAL. We now take the set G in (3.1.13) to be a subset

of the plane K 2 . The set is measurable, and has area A, if the boundary of G
is a continuous, piece-wise smooth simple closed curve. In area integrals, we
use the differential form of the area element dA. In cartesian coordinates,
the area element is dA — dxdy. In the g e n e r a l i z e d p o l a r c o o r d i n a t e s
x = arcos6, y = brsin.6, the area element is dA = ab rdrd6; the usual
polar coordinates correspond to a = b = 1. The area integral is evaluated
by reduction to a suitable iterated integral. The area integral is also known
as the double integral and is often written as JfG f dA. The integral is
defined if / is a continuous scalar field in G. If / is the area density of G,
then fjG f dA is the mass of G. Computation of double integrals is one of
the key skills that must be acquired in the study of multi-variable calculus.
EXERCISE 3.1.14. (a)B Let G be the parallelogram with vertices

(0,0), (3,0), (4,1), (1,1), and f, a continuous function. Write the limits
in the iterated integrals below:
fffdA= J ffdxdy = f ffdydx = f ffrdr dO = f f fdO rdr.

G
Hint: some integrals must be split into several. (b)c Verify that the area of
the ellipse (x2/a2) + {y2/b2) < 1 with semi-axis a and b is irab. Hint: use
generalized polar coordinates.
T H E SURFACE INTEGRAL. We now take the set G in (3.1.13) to be a
surface S defined by the vector-valued function r = r(u, v) of two variables,
where (u,v) belong to some measurable planar set G\. We say that the
surface is piece-wise smooth if the function r is continuous on G\ and
the set Gi can be split into finitely many non-overlapping pieces so that,
on each piece, the vector field ru x r„, representing the normal vectors to
the surface, is continuous and is not equal to zero. We say that the surface
is o r i e n t a b l e if it is piece-wise smooth and has two sides (the intuitive
idea of a side is enough for our discussion). A piece-wise smooth surface S*
is measurable and its area m(S) is defined as
m(S) = If \\ru x rv\\dA(u,v), (3.1.19)

Gi
where dA(u, v) is the area element of G\. We call
da = \\ru x rv\\ dA(u,v)

the surface area element of S.

EXERCISE 3.1.15. B Justify (3.1.19). Hint. Verify that ||r„ x rv\\AuAv is
the area of the parallelogram in the tangent plane to S at the point r(u,v), with
sides Au, Av parallel to the vectors ru, rv, respectively. For small Au, Av, this
area closely approximates a curved parallelogram-like surface element of S with
vertices atr(u,v), r(u + Au,v), r(u,v + Av), r(u + Au,v + Av). Summing up
over all such surface elements and passing to the limit as Au —• 0, At; —» 0 gives
(3.1.19). For a slightly different derivation of (3.1.19), see Exercise 3.1.35(c) on
page 149 below.
T h e s u r f a c e i n t e g r a l of t h e f i r s t k i n d Jf f da is defined for a
s
continuous scalar field / by the equality
J Jfda = J J f(r(u,v)) \\ru x r „ | | d A ( u , i ; ) , (3.1.20)

S Gi
where dA(u, v) is the area element. If f(P) represents the areal density of
the surface a t point P, then JJ f da is the mass of the surface. T h e s u r f a c e
s
i n t e g r a l of t h e s e c o n d k i n d JJ F-dcr is defined for a continuous vector
s
field F with values in E3 and an orientable surface 5" by the equality
ffFd(T= fJ F(r(u,v)) • ( r u x rv)dA(u,v); (3.1.21)

S Gi
the integral is also called the f l u x of F across S. A surface integral of the

second kind can be reduced to a surface integral of the first kind:
IIFd(T= ffF-nsda, (3.1.22)
where fis is the unit normal vector to the surface. T h e direction of ns

defines the orientation of S; the integral reverses the sign if the orientation
of S is reversed.
If F = pv, where v is the velocity of a fluid flow, and p is the density of
the fluid, then J J F • da is the mass of the fluid passing across the surface
s
S per unit time; verify t h a t
F • fis da = pv • fis da
is the mass of fluid passing across the surface area element da per unit time.
On page 154 we derive the equation of continuity that governs such flows.
T H E VOLUME INTEGRAL. We now take the set G in (3.1.13) to be
a subset of the space K 3 . The set is measurable and has volume V if
the boundary of G is a piece-wise smooth surface. In volume integrals,
we use the differential form of the volume element dV. In cartesian
coordinates, dV = dxdydz. In g e n e r a l i z e d c y l i n d r i c a l c o o r d i n a t e s
x = arcos6, y = brsin9,z — z, dV = ab rdrdOdz; the usual cylindri-
cal coordinates correspond to a = b = 1. In g e n e r a l i z e d s p h e r i c a l
c o o r d i n a t e s x — ar cos9 siny>, y = br sinO sine/?, z = cr costp, dV =
abc r2 sin <p dr d6 d<p; the usual spherical coordinates correspond to a = b =
c = 1. The volume integral is evaluated by reduction to three iterated in-
tegrals; this is why the volume integral is also known as the triple integral
and is often written as JJJ f dV. The integral is defined if / is a continuous
G
scalar field in G. If / is the volume density of G, then JJJ f dV is the mass
a
of G.
EXERCISE 3.1.16. B Find the volume of the ellipsoid (x2/a2) + (y2/b2) +
2
(z /c ) < 1. Hint: use generalized spherical coordinates.
Let us summarize the main facts related to the orientation of domains
and their boundaries. For a domain G in the (?, j) plane, we denote by
dG the boundary of G and assume that this boundary consists of finitely
many simple closed piece-wise smooth curves. The p o s i t i v e o r i e n t a t i o n
of dG means that the domain G stays on your left as you walk around the
boundary in the direction of the orientation of the corresponding curve. An
equivalent mathematical description is as follows. At all but finitely many
points of dG, there exists the unit tangent vector u; the direction of u
defines the orientation. At every point of dG where u exists, consider the
unit vector n that is perpendicular to u and points outside of G. Draw a
picture and convince yourself that the orientation of dG is positive if and
only if fi x u has the same direction as i x j .
For a domain G in R 3 , we denote by dG the boundary of G and assume
that this boundary consists of finitely many piece-wise smooth simple closed
surfaces. The p o s i t i v e o r i e n t a t i o n of dG means that, when it exists,
the normal vector to dG points outside of G.
For a piece-wise smooth surface S that is not closed, we denote by
dS the boundary of the surface and assume that this boundary consists
of finitely many simple closed piece-wise smooth curves. We say that the
o r i e n t a t i o n s of S and dS agree if S stays on your left as you walk
around the boundary in the direction of the orientation of the corresponding
curve with your head in the direction of the normal vector to S. If this is
possible, then S is orientable. An equivalent mathematical description is
as follows. Assume that the orientation of S is fixed and given by the unit
vectors ris that exist at most points of S. Then a small piece of S near dS
looks like a small piece of a domain in the (i, j) plane, with us defining
the direction of i x j . Then the positive orientation of dS means positive
orientation of the boundary of this two-dimensional domain as described
earlier.
A famous example of a smooth non-orientable surface is the Mobius
s t r i p , first discovered by the German mathematician AUGUST FERDINAND
MOBIUS (1790-1868). You can easily make this surface by twisting a long
narrow strip of paper and taping the ends. Verify, for example, by drawing
a line along the strip, that the surface has only one side
We will now use integrals to provide coordinate-free definitions for the
divergence and curl of a vector field. A traditional presentation starts with
the definitions in the cartesian coordinates. Then, after proving Gauss's and
Stokes's Theorems, one can use the results to show that the definitions do
not actually depend on the coordinate system. We take a different approach
and start with the definitions that are not connected to any coordinate
system.
We begin with the DIVERGENCE. Divergence makes a scalar field out
of a vector field and is an example of an operator. This operator arises
naturally in fluid flow models. Let v = v(P) be the velocity at point P in
a moving fluid, and let p = p{P) be the density of the fluid at P. Consider
a small surface of area ACT with unit normal vector u. The rate of flow
per unit time, or flux, of fluid across ACT is pv • u ACT. For a region G
bounded by a smooth simple closed surface dG, the total flux across dG
is 3>(G) = JJ pv • uda. This flux is equal the rate of change in time of
G
the mass of fluid contained in G. The quantity $(G)/m(G), where m(G)
is the volume of G, is the average flux per unit volume in G; in the limit
m(G) —> G, that is, as the region G shrinks to a point, we get an important
local characteristic of the flow. This naturally leads to the definition of
divergence.
Let F be a continuous vector field with values in R3. Then the
d i v e r g e n c e of F at the point P is the number div F(P) so t h a t
div F ( P ) = lim _ —^— /"/" F • d<r, (3.1.23)

m(G)-»o m(G)
9G
where G = G(P) is a measurable solid region t h a t contains the point P,

and m(G) is the volume of G. This definition assumes t h a t the limit exists
and does not depend on the particular choice of the regions G. Interpreting
F as the velocity field of fluid, the right-hand side of expression (3.1.23)
determines the density of sources (div F(P) > 0) or sinks (div F(P) < 0) at
the point P. For an electric field, sources and sinks correspond to positive
and negative charges, respectively. T h e vector field F is called s o l e n o i d a l
in a domain G if div F = 0 everywhere in G. T h e reason for this terminology
is t h a t , in t h e theory of magnetism, there are no analogs of positive and
negative electric charges. Therefore, a magnetic field has no sources or sinks
a n d its divergence is zero everywhere. On the other hand, a magnetic field
is often produced by a s o l e n o i d — a coil of wire connected to an electric
source.
EXERCISE 3.1.17? Let F = F\ 1 + F2J + F3 k and assume that the functions

Fi,F2,Fs have continuous first-order partial derivatives. Show that
divF = Flx + F2y + F3z, (3.1.24)
where, as usual, F\x = dF\jdx, etc. Hint: let P = (xo,yo,zo) and consider a
cube G with center at P, faces parallel to the coordinate planes, and side a. Then
m(G) = a3. The total flux through the two faces that are parallel to the (j, k)
plane is then approximately
(F1(x0 + a/2,y0,zo) - Fi(x0 - a/2,y0,z0))a2; (3.1.25)
draw a picture and keep in mind the orientation of the boundary. Then consider
the faces parallel to the other two coordinate planes. The total flux out of G is the
sum of the resulting three fluxes. After dividing by the volume of G and passing
to the limit, (3.1.25) results in Fix(xo,yo, zo), and Fiv,Fzz come form the fluxes
across the other two pairs of faces.
Next, we discuss the CURL. A S a motivation, we again consider a vector

field v representing the velocity of moving fluid or gas. T h e flow of fluid
or gas can have vortices: think of the juice in a blender or hurricanes,
tornadoes or large-scale circulation of atmospheric winds due to Coriolis
force. Vortices are caused by circular motion of the particles in the flow,
and, for a smooth simple closed planar curveC, the circulation ^ = §v-dr
c
measures the amount of vorticity. If S is the planar region enclosed by C,
then ^/m(S) is the average vorticity in the region, where m ( 5 ) is the area
of S. In the limit m ( 5 ) —> 0 we get another i m p o r t a n t local characteristic
of the flow. This naturally leads to the definition of the curl.
Let F b e a continuous vector field with values in R 3 . Fix a point P a n d
a unit vector n and consider the plane passing through P and having n as
the normal vector. Let C be a piece-wise smooth simple closed curve in the
plane, and S, the p a r t of the plane enclosed by the curve. We assume t h a t
P G S, P £ C, and the orientation of C is counterclockwise as seen from
the top of n . By definition, the c u r l of F at the point P is the vector
curl F(P) so t h a t , for all C and u,
cmlF(P).n= lim -L-JF-dr, (3.1.26)

m(S)—om(i>) Jc
where m(S) is the area of S. This definition assumes t h a t the limit exists
for every unit vector fi, and, for each vector n, the limit does not depend
on the particular choice of the curves C.
E X E R C I S E 3.1.18? Let F = Fii+F2 J+F3K and assume that the functions

Fi,F2, F3 have continuous first-order partial derivatives. Show that c u r l . F
can be computed as the vector given by the symbolic determinant
l JK
curLF A A A (F3y-F2z)i+(Flz-F3x)3+(F2X-Fly)K. (3.1.27)
dx dy dz
F\ F2 F3
Hint: with P — (xo,yo,zo), take the plane through P parallel to one of the coor-
dinate planes and let C be the boundary of the square in that plane with center at
P and side a so that m(S) = a?. Taking the plane so that n = i, we find that
j> F -dr tta(F3{xo,yo + a/2,zo) - F3{x0,yo -a/2,z0))
+ a(F2(xQ,yo,zo -a/2) - F2(xQ,yo,z0 + a/2));
draw the picture and keep track of the orientations. Dividing by a and passing
to the limit a —+ 0 gives the i component of curl F. The other two components
are computed similarly.
Relations (3.1.24) and (3.1.27) suggest the introduction of the vector

differential operator
and the symbolic vector representations of the divergence and curl:
d i v F = V - F , c u r l F = V x F. (3.1.29)
We can also interpret g r a d / as a multiplication of the vector V by the

scalar / . Then it is easy to remember the following identities:
div(curl F) = 0 : V • (V x F) = 0; curl(grad/) = 0 : V x ( V / ) = 0;
V-(/F) = V/-F + /V-F, V x (/F) = V / x F + / V x F ;
V • (F x G) = G • (V x F) - F • (V x G); V • ( V / x Vg) = 0.
(3.1.30)
Not all formulas are nice and easy, though: the expressions for V ( F • G)
and curl(F x G) are rather complicated.
EXERCISE 3.1.19. A Recall that, for two vector fields F, G, we define the
vector field (F • V ) G ; see (3.1.11) on page 126. Show that
{F • V ) G = | ( c u r l ( F x G) + grad(F • G) - G d i v F + F divG
- G x c u r l F - F x curlGV
Use the result to derive the following formulas:

curl(F x G) = F divG - G d i v F + (G • V ) F - ( F • V ) G ,
grad(F -G) = Fx curlG + G x c u r l F + ( F • V ) G + (G • V ) F .
For a scalar field / , div(grad / ) is called the Laplacian of / and denoted

by V 2 / ; an alternative notation A / is also used.
EXERCISE 3.1.20.c Verify that, in cartesian coordinates,
V2f = fxx + fyy + fzz. (3.1.31)
For a vector field F , we define
V 2 F = grad(div F ) - curl(curl F ) . (3.1.32)
EXERCISE 3.1.21. (a)A Verify the equalities in (3.1.30) Hint: choose

cartesian coordinates so that F = F\i. (b) Verify that if F — F\ t+F2 3+F3 k,
then V 2 F = V 2 Fi z + V 2 F 2 j + V 2 F 3 k. (cf Verify that V2(fg) = gV2f+
Given a reference point O, there is a natural vector field, denoted by

r, which maps every point P to the corresponding position vector
OP. The corresponding scalar field r = \\r\\ is the distance from
OtoP.
The following exercise establishes a number of remarkable properties of
the fields r and r.
EXERCISE 3 . 1 . 2 2 . C (a) Verify that if g = g(t) is a differentiate function,
then
Vff(r) = g'(r) r = ^ - r . (3.1.33)
Hint: use cartesian coordinates so that r = x i+yj+z K, and r = \Jx2 + y2 + z2.

(b) Verify that, in M", divr = n, n = 2,3. (c) Verify that, in R 2 , V 2 lnr =
0 for r ^ 0. (d) Verify that, in K 3 , V 2 r _ 1 = 0 for r ^ 0. Hint: for (c)
and (d), use (3.1.33) together with some of the identities in (3.1.30). (e) Let
h = h(t) be a differentiate function. Verify that curl(h(r) r ) = 0. Hint:
the fastest way is to conclude from (3.1.33) that h(r)r = V/7(r) for a suitable
function H; then use (3.1.30). (f) Let b be a constant vector. Verify that
grad(6-r)=6. (3.1.34)
Hint: write the dot product in cartesian coordinates, (g) Let F = cr/r be a
3
central inverse-square force field in R Prove that / F • dr is independent
of the path C joining any two points, as long as C does not pass through the
origin. Hint: show that F is conservative by finding f = f(r) so that F = V / .
Let F = cr/ra = (c/ra+l)r,
EXERCISE 3 . 1 . 2 3 . C where c and a are real
numbers. Verify that the flux of F across the sphere S with center at the
origin and radius R is equal to &-nR2~ac. Hint: Use (3.1.22); Hs = f; on S,
r = R.
EXERCISE 3.1.24? Let BR be a ball with center at the origin and radius R.
Verify that limR^0 fff r~pdV = 0 for all p < 3.
BR
Our motivation of the definition of the curl suggests that curl is asso-
ciated with rotation; in fact, some books denote the curl of F by rotF.
Let us look closer at this connection with rotation. Recall that, for fixed
Curvilinear Coordinate Systems 141
vectors u> and ro, the vector w x r 0 is the velocity of the point with po-
sition vector ro, rotating with constant angular speed u — ||u>|| about a
fixed axis; the vector w is the angular velocity of the rotation, see page
48. The vector field v = u> x r therefore corresponds to the rotation of all
points in space with the same angular velocity CJ. Direct computations in
cartesian coordinates show that curlw = 2u>, establishing the connection
between the curl and rotation: if F = F(P) is interpreted as the velocity
field of a continuum of moving points, then curl F measures the rotational
component of this velocity; the field F is called i r r o t a t i o n a l in a domain
Gif c u r l F = O i n G .
3.1.3 Curvilinear Coordinate Systems

Recall that a vector field (P,F(P)) is a collection of vectors with different
starting points, and, to study the vector field, we need a local coordinate
system, that is, a set of basis vectors, at every point P. So far, we mostly
worked with cartesian coordinates, where this local coordinate system is
the parallel translation of (i, j , k) to the corresponding point P. More
generally, these local coordinate systems are not necessarily parallel trans-
lations of one another. For example, the polar coordinates r, 6 in R 2 give
rise to the unit vectors r,8, see (1.3.22) on page 35 or Figure 3.1.1 below.
At every point, r A. 6, but the corresponding vectors r have, in general,
different directions at different points. For various applications, it is natural
to use cylindrical or spherical coordinates, and then it becomes necessary
to express all the operations of differentiation (gradient, divergence, curl,
Laplacian) in these coordinate systems.
The objective of this section is to study general orthogonal curvilinear
coordinate systems, where the basis vectors at every point are mutually or-
thogonal, but their direction changes from point to point. Our first step
is to establish the general relation between the coordinates and their ba-
sis vectors. For notational convenience, we denote by X — (£1,0:2,0:3) the
cartesian coordinates, with the correspondence x\ = x, X2 = y, X3 = z. For
non-cartesian coordinates Q = {qi,q2,q?), each qk is a given function of
xi,X2,Xz: qk = qfc(a:i,a:2,a;3), k = 1,2,3. The correspondence between the
X and Q coordinates must be invertible, so that each Xk is a correspond-
ing function of q\,q2,qz'- xk = Xk{qi,q2,qz), k = 1,2,3. F O R EXAMPLE,
in c y l i n d r i c a l coordinates, with qi = r, q^ — 6, qz = z, we have
r = \Jx\ + x\, 9 = t a n - 1 (0:2/0:1), z = 0:3, so that 0:1 = rcosO, o:2 =
rsin#, X3 = z. This example also shows that there could be some points
where the correspondence between the X and Q-coordinates is not invert-

ible: the polar angle 6 is not defined when x = y — 0. We call such points
special and allow their existence as long as there are not too many of them;
at this point, we rely on intuition to decide how many is "many" and will
not go into the details.
EXERCISE 3.1.25. B Let 91 = r, 92 = 6, 93 — if be the usual s p h e r i c a l
c o o r d i n a t e s so that Xi = rcosO siny, x-i = rsin# simp, X3 = rcosip.
Express qi,q2,q3 as functions of Xk, k = 1,2,3. Find all special points.
A coordinate curve in system Q is obtained by fixing two of the Q
coordinates, and using the remaining one as a parameter. For every non-
special point P with Q coordinates (cj, 02,03), there are three coordinate
curves passing through P. For example, we can fix 92,93 and vary qi.
The corresponding coordinate curve C\ is given in cartesian coordinates
by xk = x(<7i>c2,C3), k = 1,2,3. The coordinate curves Ci and C3 are
obtained in a similar way. The coordinate system Q is called orthogonal
if the coordinate curves are smooth and are mutually orthogonal at every
non-special point. The unit tangent vectors qk, k = 1,2,3 to these curves
at P define the local orthonormal basis at P ; we choose the orientation and
the ordering of the curves so that the basis is right-handed: q-^ x q2 — q3.
Figure 3.1.1 shows the coordinate curves and unit tangent vectors for the
polar coordinates x = r cos 0,y = r sin 6.
2.
0 = §2 y r = q j
l
o
Fig. 3.1.1 Polar Coordinates
In the Q coordinates, the coordinate curves are also represented as in-

tersections of pairs of surfaces: {(91,92,93) : 9n = cn,qm = c m } , where
m, n = 1,2,3, m < n, and cm, cn are real numbers. If we assume that, at
time t = 0, each curve passes through the point P, then, in the Q coordi-
nates,
• The 91 coordinate curve with 92 = 02, 93 = 03, is the collection of points
(ci +£,c 2 ,c 3 ), teR;
• The 92 coordinate curve with qi= c\, 93 = 03, is the collection of points
(ci,C2 + t,C3), t € R ;
• The qz coordinate curve with q\= c\, q2 = c2, is the collection of points
(ci,c2,c3 + t), i e l .
Since, by assumption, Xk = Xk(qi,q2,q3), the equations of these curves in
cartesian coordinates are
ri(t) = xi(ci + i , c 2 , c 3 ) l + X2(ci + t , c 2 , c 3 ) j + X3(ci +t,c2,c3)k

c
r2{t) = Xi(<=i, 2 + 1 , c3) i + X2(ci ,c2 + t, c3) 3 + Xz{ci ,c2 + t, c3) K
^3(*) = Xl (Cl, C2, C3 + i) Z + X2(Cl, C2, C3 + t) J + X3(Cl, C2, C3 + 0 K
(3.1.35)
We define the functions hk, k = 1,2,3 so that the tangent vectors are
r'fc(0) = ^fc«fc. (3.1.36)
By (3.1.35),
N S(fe). <-->
hk(P) = ||r'fc(0)|| =
where the partial derivatives are evaluated at (c\, c2, c 3 ), the Q coordinates
of the point P. The orthogonality condition means that
7^(0) • r'n(0) = J2^^=0,m^n. (3.1.38)
The functions hk defined in (3.1.36) will play the central role in many
computations to follow. We will assume that, away from the special points,
the functions hk = hk(qi,q2,qs) have continuous partial derivatives of every
order.
EXERCISE 3.1.26? (a) Verify the relations (3.1.37) and (3.1.38). (b) Verify
that cylindrical and spherical coordinates are orthogonal. Hint: a picture is
helpful.
Our next goal is to study ARC LENGTH in Q coordinates. Let C be a
curve represented in cartesian coordinates X by the vector function r(t) —
xi(t)t-\-x2(t)3 + xs(t) ii. We rewrite the expression (1.3.12) on page 29 for
the line element as (ds/dt)2 = J2k=i(dxk/dt)2. Since Xk = Xk{qi,q2,qz),
we have xk(t) = Xk(qi(t),q2(t),q3(t)), where qk(t) = qk(xi(t),x2(t),x3(t)),
k = 1,2,3 By the chain rule, dxk/dt = Tlli=i(dXk/dqm)(dqm/dt), and
then, using (3.1.36) a n d dropping (dt)2, we find t h a t t h e l i n e e l e m e n t in

Q coordinates is
(ds)2 = ^2h2k(qi,q2,q3)(dqk)2. (3.1.39)

fc=i
5
E X E R C I S E 3.1.27. (a) Verify (3.1.39).
(b) Verify that, for small a, the distance between the two nearby points
with the Q coordinates (ci,02,03) and (ci + 0 , 0 2 , 0 3 ) is approximately h\a.
(c) Verify that in cylindrical coordinates, (ds)2 = (dr)2+r2(d0)2 + (dz)2.
(d) Verify that in spherical coordinates, (ds)2 = (dr)2 + r2 sin (p (d9)2 +
r (dtp)2.
2
Hint for (b) and (c): while direct substitution into (3.1.39) will work,
getting closer to the basics produces an easier argument. For example, in spherical
coordinates, if r and ip are fixed and 6 is changing, the corresponding coordinate
curve is a circle of radius rsimp (draw a picture). Then a change of A0 in 8
produces the change r sin tp A9 in the arc length. Similar arguments work for the
other two coordinate curves, and then it remains to use orthogonality. Note that,
with this approach, you get the values of hk without using (3.1.37).
(e) Verify that if the Q coordinate are not orthogonal, then
(ds)2 = Y^ Ha d
Qi d1ji where H^ = ] P -a^-*^- (3.1.40)
0Qi OQj
i,i=i fc=i
We already used a four-dimensional analog of this relation in the study of

general relativity; see (2.4-23) on page 107.
For future reference, let us summarize t h e values of t h e functions hk for

cylindrical and spherical coordinates:
hi hi h3
Cylindrical 1 r 1
Spherical 1 rs'm<p r
We can now derive t h e expression for t h e GRADIENT in Q coordinates.

T h e value of a function a t a point does not depend on t h e coordinate
system, but the formula for computing that value does: if / = X1X2 +x% in
cartesian coordinates, the same function in spherical coordinates becomes
r 2 cos#sin#sin tp + r cosip. Given an expression /(gi,92,(73) for a function
in Q coordinates, we define the partial derivatives df/dqk, k — 1,2,3 in
the usual way; for example,
9f = n m /(gl+t,g2,g3)-/(gl,g2,g3)
dqi t-»o t
Unlike cartesian coordinates, the partial derivative df/dqk is not equal to
the directional derivative Dqkf- Instead, by (3.1.9), (3.1.35), and (3.1.36),
we conclude that
^ - = V / • r'fc(0) = hkVf-qk = hkD9hf. (3.1.41)

dqk
EXERCISE 3.1.28? (a) Convince yourself that, in general, df/dqk ^ Dqkf.

Hint: if P has Q coordinates (91,92,93), and Pt, the coordinates (91 +£,92,93),
then, in general, OPt ^= OP + tq1, but the equality does hold in cartesian coordi-
nates, (b) Verify (3.1.41).
Since the representation V / = !Cfc=i(^/ ' 9fc)9fc holds in every or-
thonormal coordinate system, equality (3.1.41) implies the following for-
mula for computing the gradient in general orthogonal coordinates:
V / = l | ^ 1 + l | ^ 2+ l | ^ 3 , (3.1-42)
h\ dq\ h2 dq2 h3 dq3
where the functions hi,h2,h3 are denned in (3.1.37).
B
EXERCISE 3.1.29. (a) Verify that
d 1 d d
Vf= -Lr dr + - -L§
r 86+ -lzdz
is the gradient in cylindrical coordinates, (b) Verify that
or r sin <p 06 r o<p

is the gradient in spherical coordinates, and verify that, for f(r, 6, ip) = g(r),
the result coincides with (3.1.33).
Next, we derive the expression for the DIVERGENCE in Q coordinates
using the definition (3.1.23). Recall that, to compute the divergence in the
cartesian coordinates, we consider a family of shrinking cubes with faces
parallel to the coordinate planes. The same approach works in any coordi-
nates by considering rectangular boxes whose sides are parallel to the local
basis vectors qi,q2,q3. Let P be a point with Q coordinates (01,02,03),
and, for sufficiently small a > 0, consider a rectangular box with vertices at
the points with the Q coordinates {c\ ± (a/2),C2 ± (a/2),c3 ± (a/2)). The
volume of this box is approximately a 3 |r' 1 (0) • (r 2 (0) x r 3 (0))| = a3hih2h3;
draw a picture and see equations (3.1.35). By (3.1.36) and (3.1.38), The
basis vector qk is normal to two of the faces so that the area of each of
those faces is approximately a2hmhn, where k ^ m ^ n. Consider a
continuously differentiable vector field F written in the Q coordinates as
Tn
F(qi,Q2,q3) = Fi(qi,q2,q3)q1 + ^2(91,92,93)92 +-^3(91,92,93) <Z3- en
the flux of this vector field through the pair of faces with the normal vector
q1 is approximately
a2(h2(ci + (a/2),C2,c3)h3(ci + (a/2),c2,c3)F1(a + (a/2),c 2 ,c 3 )
-h2(ci - (a/2),C2,c 3 )/i 3 (c 1 - (a/2),c 2 ,c 3 )Fi(ci - (a/2),c 2 ,c 3 )J
3 d(/i 2 /t 3 Fi)
09l
with the approximation getting better as a —> 0. Similar expressions hold
for the fluxes across the other two pairs of faces. Summing up all the fluxes,
dividing by the volume of the box, and passing to the limit a —> 0, we get
the formula for the divergence of F in Q coordinates:
a y * 1 = n h r ( d { h t F l ) + d{htF2) + d{htF*]) • (3-1-43)

hih2h3 V OQi 9q2 093 /
B
EXERCISE 3.1.30. (a) Verify (3.1.43)- (b) Verify that, in cylindrical coor-
dinates,
d i v F = i ( ^ ) + ^ + ^ }
r \ dr 89 dz
(c) Verify that, in spherical coordinates,
1 (d{r2smyFx) d{rF2) , d ( r s i n ^ F 3 )
oiv r = —z—;
2 I 1 — h
r siii(p \ dr 09 dtp
Recall that the Laplacian V 2 / of a scalar field / is defined in every

coordinate system as V 2 / = div(grad/). Let / = 7(91,92,93) be a scalar
field defined in an orthogonal coordinate system Q.
EXERCISE 3.1.31. (a)B Using (3.142) and (3.143) verify that
v 2 / = 1 / d (h2h3 df_ \ + _9_/Mi3 df\ + _d_fhih2 3/\\

hih2h3 \dqi\ hi dqi) dq2\ h2 dq2) dq3\ h3 dq3))'
(b) Verify that, in cylindrical coordinates,
w - ridr^\ V
dr ) i S + gdz. '
r d6 2 2 2 ("•«>
and use the result to provide an alternative proof that V 2 l n r = 0 in R 2 .
(c)c Verify that, in spherical coordinates,
2
\<9r dr J r2 sin2 <pd92 r2sin<p d<p \ dip
and use the result to provide an alternative proof that V 2 r _ 1 = 0 in R 3 .

Next, we derive the formula for the CURL using (3.1.26). Consider a
continuously differentiable vector field F written in the Q coordinates as
F(qi,q2,qs) = Fi(q1,q2,q3)q1 + F2(q1,q2,q3) q2 + F3(q1,q2,q3) q3- Let us
compute c u r l F ( P ) • q1: where the point P has Q-coordinates (c\, c2,c3).
For a sufficiently small a > 0, consider a rectangle spanned by the vectors
ar'k(0), k = 2,3, so that P is at the center of the rectangle. The vertices
of the rectangle have the Q coordinates (ci,C2 ± (a/2),c3 ± (a/2)) and
the area of this rectangle is approximately a2h2h3; draw a picture and see
equations (3.1.35). The line integral of F along the two sides parallel to q2
is approximately
a[h2(ci,c2,c3 - (a/2))F 2 (ci,C2,c 3 - (a/2))
- h2(ci, c 2 , c 3 + (a/2))F2 [cu c 2 , c 3 + (a/2))J

^ n2d(h2F2)
with the quality of approximation improving as a —> 0. The line integral

over the remaining two sides is approximately a2d(h3F3)/dq2. Dividing by
the area of the rectangle and passing to the limit a —> 0, we find that
iir/m - 1 fd(h3F3) d(h2F2)\ u

curlF(P) • 9 l = -^ - ^ - fti
The other two components, curl F(P) • q2 and c u r l F ( P ) • q3 are computed

similarly, and then we get the representation of the curl as a symbolic
determinant similar to (3.1.27):
hiqx h2q2 h3q3
1
curl F (3.1.46)
hih2h3 dqi dq2 dqz
hxFx h2F2 h3F3
EXERCISE 3.1.32. A (a) Verify (3.1.46). (b) Use (3.1.46) in spherical

coordinates to show that curl(g(r) r) = 0 in R 3 .
We conclude this section with a brief discussion of INTEGRATION
in curvilinear coordinates. As before, let X be the cartesian coordi-
nates x\,x2,X3 and Q, some other coordinates qi,q2,qz so that Xk =
Xk(qi,q2,qs), k — 1,2,3; we no longer assume that the Q coordinates are
orthogonal. At a non-special point P, consider the three vectors r'fc(0),
k = 1,2,3, defined in (3.1.35), and assume that
»-i(0)-(r'2(0)x^(0))^0. (3.1.47)
EXERCISE 3.1.33.B Verify that

dx\ dX2 9xz
dqi dqi dqi
dxi dX2 9X3

ri(0)-(r'2(0)xr^(0)) = (3.1.48)
dq2 dq2 9q2
dxi 9X2 9X3

dqz 9qz 9q3
Hint: we have cartesian coordinates in (3.1.35).
The determinant in (3.1.48) is called the Jacobian, after the Ger-
man mathematician CARL GUSTAV JACOB JACOBI (1804-1851), who,
in 1841, published a paper containing a detailed study of such determi-
nants. The Jacobian (3.1.48) is denoted either by J — J(qi,q2,q3) or by
9{xi,x2,xs)/d(qi,q2,q3); similarly, we either call it the Jacobian of the
system of functions Xi>X2iX3 o r the Jacobian of the transformation from
the X coordinates to the Q coordinates. It was one of the results in the
original paper by Jacobi that the change of variables is one-to-one if J ^ 0.
EXERCISE 3.1.34. (a)A We know that, in orthogonal coordinates,
J(liiQ2,Q3) = hihzha. Try to derive this equality directly by expand-

ing the determinant in (3.1.48) and using the orthogonality conditions
(3.1.38). (b)c Verify that J(r,6,z) = r (cylindrical coordinates) and
J(r,6,<p) = r2 simp (spherical coordinates).
Let G be a measurable domain in M3 with cartesian coordinates 3£, and

let / = f{xi,X2,X3) be a continuous scalar field on G. A change of coor-
dinates % = (\k(xi,X2,xs), k = 1,2,3, maps every point (xi,X2,£3) to a
3
point (91,92,93), which is also in M . Under this m a p ,
• T h e domain G transforms into another domain G (of course, there are
certain conditions on the functions c\k to ensure t h a t the image of a domain
is again a domain);
• A small box with sides a,b,c around a point P = (xi,:r2,:E3) becomes a
small curvilinear box around the image of P ; by (3.1.47), t h e volume of this
box is approximately abc\J(qi, 92,93)!, with approximation getting better
as max(a, 6, c) —> 0;
• T h e function / becomes / = / ( 9 i , 9 2 , 9 2 ) so t h a t / ( 9 i , 9 2 , 9 3 ) =
/ ( x i ( 9 i , 92,93), X2(9i, 92,93), X3(9i, 92,93))-
Using the general definition of the Riemann integral (3.1.13), we get the
following change of v a r i a b l e s f o r m u l a f o r t h e volume i n t e g r a l :
f(xux2,x3)dV(X) = ^ y " / ( 9 i , 9 2 , 9 3 ) | J ( 9 i , 92,93) W Q ) -

///
G G
(3.1.49)
Notice t h a t the change of coordinates from X to Q is given by the functions
qfc, k — 1,2,3, whereas, to compute the Jacobian and to find the function
/ , we need the functions Xk defining the change in the opposite direction.
A result similar to (3.1.49) holds for area integrals, if we write 9fc =
(\k(^i,x2), k = 1,2, and 93 = £3.
E X E R C I S E 3.1.35? (a) Verify (3.1.49). (b) Write the analogs of (3.1.47) and
(3.1.49) in two dimensions. Hint: r[(0) x r ^ O ) = J{qi,q2)(t- (c) Interpret
the vector parametric representation r = r(u, v), (u, v) G G, of a surface S
as a map from G to S and derive the familiar formula for the surface area
m(S) = / / \\ru x rv\\dA(u,v).

G
Hint: show that a small rectangle with sides a,b in G becomes a curvilinear figure
on S with area approximately equal to ab\\ru x rv\\. Compare this with Exercise
3.1.15 on page 134.
150 The Three Theorems
3.2 T h e T h r e e Integral T h e o r e m s of Vector Analysis
The theorems of Green, Gauss, and Stokes are usually the high point of
a multi-variable calculus class. We will use these theorems to derive the
equations of electromagnetic theory.
3.2.1 Green's Theorem
Theorem 3.2.1 In the space R 2 with cartesian coordinates, consider a

domain G whose boundary dG consists of finitely many simple closed piece-
wise smooth curves. Let P = P(x, y), Q = Q(x, y) be two functions having
continuous first-order partial derivatives in an open domain containing G.
Assume that the orientation of dG is positive (G stays to the left as you
walk around dG). Then
If the boundary of G consists of several pieces, then the integral on the left
is the sum of the corresponding integrals over each piece.
This result is known as Green's theorem, after the English scientist
G E O R G E G R E E N (1793-1841), who published it in 1828 as a part of an
essay on electricity and magnetism. The essay was self-published for pri-
vate distribution and did not receive much attention at the time. Green
did all his major scientific work while managing the family mill and bak-
ing business, with about two years of elementary school as his only formal
education. Later, he quit the family business and, in 1837, got an under-
graduate degree from Cambridge University.
EXERCISE 3.2.1. B Prove Green's theorem. Hint: Consider the vector field
F = Pi + Qj + Ok and verify that V x F = (Qx — Pv)it. Then break G into
small pieces Gk and note that, by (3.1.26), (Qx — Py)m(Gk) ~ fgG Pdx + Qdy,
where m(Gfc) is the area of Gk, and dGk is the boundary of Gk with the suitable
orientation. Now sum over all the small pieces and note that the line integrals
over the interior pieces cancel one another. Finally, pass to the limit m(Gfc) —• 0.
An interesting application of (3.2.1) is computing areas by line integrals:
m(G) = I dA= <p xdy = — d> ydx = - i xdy — ydx. (3.2.2)

2
JJ JdG JdG JdG
Gauss's Theorem 151
E X E R C I S E 3 . 2 . 2 . B (a) Verify all equalities in (3.2.2). (b) Verify that, in

polar coordinates x = r cos 6, y = r sin 6, we have
m(G) = \ i A
r2d9.
JddG
Hint: verify that, in polar coordinates, xdy — ydx = r2dd.
F O R E X A M P L E , let us compute the area under one arc of the c y c l o i d ,

the trajectory of a point on the rim of a wheel rolling without slippage
along a straight line. If a is the radius of the wheel, and the wheel is rolling
to the right so t h a t , at time t = 0, the point on the rim is at the origin,
then the vector parametric equation of the cycloid is
r(t) = a{t - sin t)i + a(l - cos t) j . (3.2.3)
One revolution of the wheel corresponds to t = 2n. T h e region G under

one arc is bounded by the x-axis on the b o t t o m and the arc on the t o p . By
(3.2.2), keeping in mind the orientation of the boundary, the area of the
region is m(G) = - §dG ydx — - fQ* Odx - a2 / 2 7 r ( l - cost)2dt = Sna'
On the arc, x'(t) = a(l — cost), y(t) = a ( l — c o s t ) , and, of the three
possible formulas for the area, the one involving only ydx results in the
easiest expression to integrate.
E X E R C I S E 3.2.3. B (a) Verify that (3.2.3) is indeed a vector parametric

equation of the trajectory of the point on the rim of a rolling wheel. Hint:
consider 0 < t < ir. Denote by A the point on the rim, A', the current point of
contact between the wheel and the ground, and C, the center of the wheel. Then
r(t) = OA'+A'C+CA, \OA'\ = at, and the angle between CA' and CA is t; draw
the picture, (b) Draw the picture of the cycloid and verify the integration.
There is a hardware device, called p l a n i m e t e r , which implements the

formula tn(G) = (1/2) §dG xdy — ydx and measures the area of a planar fig-
ure by traversing the perimeter. T h e Swiss mathematician J A C O B A M S L E R
(1823-1912) is credited with the invention of a mechanical planimeter in
1854. Today, both mechanical and electronic planimeters are in use.
3.2.2 The Divergence Theorem of Gauss
Depending on the source, the following result is known as G a u s s ' s

t h e o r e m , the G a u s s - O s t r o g r a d s k y t h e o r e m , or the d i v e r g e n c e t h e o r e m .
This theorem can be motivated by the following physical argument. Let v
and p be the velocity field and the density of a fluid in a region G. In our
discussion of divergence, we saw that div(pv) is the time rate of change of
the amount fluid in G per unit volume. Therefore, fff div(pv)dV is the
G
time rate of change of the amount of fluid in G, and must be equal to the
total flux across the boundary dG of G:
flJdiv(pv)dV= Ijpvdc
G dG
This is the essential meaning of the divergence theorem, which holds for
every continuously differentiable vector field F as stated below.
T h e o r e m 3.2.2 Consider a bounded domain G in R 3 whose boundary
dG consists of finitely many piece-wise smooth orientable surfaces. Let F
be a vector field so that, for every unit vector u, the scalar field F • u has
a continuous gradient in an open set containing G. Then
ffF-dtr= fffdivFdV, (3.2.4)

dG G
where the normal vector to dG is pointing outside of G; if the boundary of

G consists of several pieces, then the integral on the left is the sum of the
corresponding integrals over all those pieces.
The great German mathematician CARL FRIEDRICH GAUSS (1777-

1855) discovered this result in 1813 while studying the laws of electrostatics.
As with many other his discoveries, he was not in a hurry to publish it; the
exact year is known from the detailed mathematical diary Gauss kept all his
adult life. Based on the entries in that diary, many believe that Gauss could
have advanced 19th century mathematics by another 50 years or more, had
he published all his results promptly. The Russian mathematician MIKHAIL
VASIL'EVICH OSTROGRADSKY (1801-1862) made an independent discovery
of (3.2.4) and published it around 1830.
EXERCISE 3.2.4? (a) Prove Gauss's Theorem. Hint: the argument is the same
as in the proof of Green's Theorem, but using (3.1.23). (b) Let F = r/r3 = r/r2
(see page 140). Let S be a piece-wise smooth closed surface so that the origin
O of the frame is not on the surface. Show that
II F.d<r = C'
I 0,
S end
° S e S thG °rigin'
S does not enclose the origin.
(3.2.5)
Gauss's Theorem 153
Hint: Note that d i v F = V 2 ( l / r ) = 0 ifr ^ 0. If S encloses the origin, then there

exists a small sphere centered at the origin and lying inside the domain enclosed
by S. Show that the flux through S is equal to the flux through the sphere; then see
Exercise 3.1.23 on page HO. (c) Let dG be an orientable piece-wise smooth
surface enclosing a domain G, and let F be a vector field, continuously
differentiable in an open set containing G. Show that
I FxnGda = - fff curl F dV, (3.2.6)

dG G
where fie is the outside normal vector to dG. Hint: the left- and right-hand
sides of (3.2.6) are vectors; denote them by v and w, respectively. The claim is
that v-b = w -b for every constant vector b; in cartesian coordinates, it is enough
to verify this equality for only three vectors, i,J,k instead of every b. By the
properties of the scalar triple product, b • (F x HG) = —(F x b) • HG- By Gauss's
Theorem,
ff{F xb)nada = fff div(F x b) dV,
dG G
and by (3.1.30), div(F x b) = b • c u r l F .
We will now use Gauss's Theorem to derive the basic equation of fluid
flow, called the E Q U A T I O N O F C O N T I N U I T Y . Assume t h a t the space M 3 is
filled with a continuum of point masses (particles). This continuum can
represent, for example, liquid, gas, or electric charges; accordingly, "mass"
represents the appropriate characteristic of t h a t continuum, such as the
mass of m a t t e r or electric charge. Denote by v = v(P, t) the velocity of the
point mass at the point P and time t, and by p = p(P,t), the density of
the point masses at the point P and time t. By definition,
where AV is the volume of a region containing the point P, AM is the mass

of t h a t region, and limit is assumed to exist at every point P. We also allow
the existence of sources and sinks, t h a t is, in some areas of the space, the
particles can be injected into the space (source) or removed from the space
(sink). T h e density of the sources and sinks is described by a continuous
scalar field v = v(P, t) so t h a t the net mass per unit time, injected into and
removed from G due to sources and sinks is fff pvdV. We also assume
G
t h a t the functions p, u, v are continuously differentiable as m a n y time as
necessary. Since the total mass M of the particles in G also changes due t o
the flux across dG, we have

dM
dt
HIpudV - ifpv-da; (3.2.7)
G 8G
recall that the positive direction of the flux is out of G, which reduces
the mass. On the other hand, M = fffp dV. Assuming that we can
G
differentiate under the integral, dM/dt = JJf(dp/dt) dV. If the functions
G
p and v are continuously differentiable, then, by Gauss's Theorem (3.2.4),
Jfpv-dcT = fff div(pv)dV. As a result, (3.2.7) implies
dG G
-£ + div(pv) - pi> ) dV = 0.
/ / / (
Given a point in space, this equality holds for every domain G containing
that point, and we therefore conclude that the equation of c o n t i n u i t y
-£+div(pv)=pu (3.2.8)
holds at every point of the space. Conversely, after integration, (3.2.8)

implies (3.2.7).
EXERCISE 3 . 2 . 5 . C
(a) Convince yourself that (3.2.7) implies (3.2.8). Hint:
assume that the left-hand side of (3.2.8) is bigger than up at one point. By conti-
nuity, the inequality will hold in some neighborhood of that point. Integrate over
that neighborhood and get a contradiction, (b) Verify that (3.2.8) is equivalent
to
-?-+pdivv = up. (3.2.9)

dt
Hint: Consider a trajectory r = r(t) of a moving particle so that v = r ' and
p = p(r(t),t). Then dp/dt = Vp • v + dp/dt. On the other hand, by (3.1.30),
div(pu) = Vp • v + p divv.
A continuum is called incompressible if the density p does not change
in time and space: p(t, x) = const. For an incompressible continuum, equa-
tion (3.2.8) becomes divw = u; the velocity v of an incompressible contin-
uum without sources or sinks satisfies divt; = 0.
We can also state Gauss's Theorem in a two dimensional domain G.
Let the functions P, Q and the domain G satisfy the conditions of Green's
Stokes's Theorem 155
Theorem. Define the vector field F — Pi + Qj + 0k and consider the

outside unit normal vector no to dG; since the unit tangent vector UQ
to dG exists at all but finitely many points of dG, so does no, being, by
definition, a unit vector perpendicular to UG- Then, by Gauss's Theorem,
i F • nG ds = / / d i v F d A . (3.2.10)
JdG JJ
G
EXERCISE 3.2.6. c Derive (3.2.10) from Green's Theorem. Hint: ifr(t) =
x(t) i + y(t) j is the parametrization of dG, then r' = x' t + y' j is tangent to dG;
the vector y' i — x'j is therefore perpendicular to r' and points outside (verify
this). Then F • na ds = -Qdx + Pdy. Note that div J 1 = Px + Qv.
3.2.3 Stokes's Theorem

The following result is known as S t o k e s ' s Theorem, after the English
mathematician G E O R G E GABRIEL STOKES (1819-1903).
T h e o r e m 3.2.3 Let S be a piece-wise smooth orientable surface whose
boundary OS consists of finitely many simple closed piece-wise smooth
curves so that the orientations of S and dS agree (see page 135). Let
F be a vector field, continuously differentiable in a domain containing S.
Then
I F • dr = 17 c u r l F • dtr; (3.2.11)
JdS JJ
s
if the boundary of S consists of several closed curves, then the integral on
the left is the sum of the corresponding integrals over all those curves.
According to some accounts, the first statement of this theorem ap-
peared in a letter to Stokes, written in 1850 by another English scientist,
Sir WILLIAM THOMSON (1824-1907), also known as LORD KELVIN.
EXERCISE 3.2.7. (a)c Verify that Stokes's Theorem implies Green's theo-
B
rem. (b) Prove Stokes's Theorem (the arguments are similar to the proof
of Greens Theorem).
The following result should be familiar from a course in multi-variable
calculus; Stokes's Theorem provides a key component for the proof.
T h e o r e m 3.2.4 Let F be a continuous vector field in a simply connected
domain G, and assume that, for every unit vector n, the scalar field F • ri
has a continuous gradient in G. Then the following three conditions are

equivalent:
(i) The field F has the path independence property: §c F • dr = 0 for every
simple closed piece-wise smooth curve C in G.
(ii) The field F has a potential: there exists a function f so that F = V /
everywhere in G.
(Hi) The field F is irrotational: curl F = 0 everywhere in G.
EXERCISE 3.2.8. (a)c Prove the theorem. Hint: for (i) => (ii), see Theorem
3.1.1. For (ii) =>• (Hi), use (3.1.30). For (Hi) =>• (i) use Stokes's Theorem. (b)B
Verify that the vector field F(x, y, z) = (-y/(x2+y2))i+(x/(x2+y2))j+OK
2 2
is irrotational in the domain {(x, y, z) : x +y > 0} (direct computation) but
does not have the path independence property there (integrate over a unit circle
in the (I, j) plane, centered at the origin). Explain why the result is consistent
with the above theorem (note that the function is not defined anywhere on the
z axis).
Assume that the vector field F satisfies the conditions of Stoke's Theo-
rem everywhere in M3. Then, by (3.2.11), the value of J J c u r l F • dcr does
S
not depend on the particular surface S, as long as the boundary of S is
the given simple closed piece-wise smooth curve C. This conclusion is con-
sistent with Gauss's Theorem. Indeed, let S\ and 5*2 be two orientable
piece-wise smooth surfaces with the same boundary C and no other points
in common, and denote by G the region enclosed by Si and S2 (draw a
picture). The boundary dG of G is the union of S\ and S2. If the vec-
tor field F is sufficiently smooth so that c u r l F satisfies the conditions of
Gauss's Theorem, then, by (3.2.11), ff curl F-dcr = J / / d i v ( c u r l F ) d V = 0,
dG G
where the last equality follows from (3.1.30) on page 139, and therefore
/ / curl F -dcr = JJ curl F • da, if the orientations of both Si and S2 agree
Si S2
with the orientation of C.
By now we have seen enough connections between Green's, Gauss's, and
Stokes's Theorems to suspect that there should be a way to unify the three
into one formula. Such a formula does indeed exist:
/ UJ= I duj, (3.2.12)

JdM JM
where dM is the boundary of the set M, and du is a certain derivative

of the expression u> being integrated on the left-hand side. Making a pre-
Laplace's and Poisson's Equations 157
rise meaning of this formula and demonstrating that (3.2.1), (3.2.4), and
(3.2.11) all follow as particular cases requires a higher level of abstraction
and is done in differential geometry, where (3.2.12) is known as Stokes's
Theorem. The details are beyond the scope of this book and can be found
in most books on differential geometry, for example, in Section 8.2 of the
book Differentiable Manifolds: A First Course by L. Conlon, 1993.
The following exercise provides yet another application of Stokes's The-
orem. One of the results, namely, (3.2.14), will be used later in the analysis
of the magnetic dipole.
EXERCISE 3.2.9. c Let S be piece-wise smooth orientable surface and its
boundary OS, a simple closed piece-wise smooth curve. Assume that the
orientations of S and dS agree. Denote by n the field of unit normal
vectors on S, and by u, the field of unit tangent vectors on dS. (a) Show
that, for every continuously differentiable scalar function f,
/ / g r a d / x n d a = - I fuds. (3.2.13)
s ds
Hint: the left- and right-hand sides of (3.2.13) are vectors; denote them by v and
w, respectively. The claim is that v • b = w -b for every constant vector b; in the
cartesian coordinates, it is enough to verify this equality for only three vectors,
i,j,K instead of b. Note that, by Stokes's Theorem, w = f/(curl(/6)) -nda.
' s
By (3.1.30), curl(/6) = grad/ x b. Then use the properties of the scalar triple
product, (b) Taking f = TQ • r in (3.2.13), where TQ is a constant vector
and r is the vector field defined on page HO, and using (3.1.34), show that
(p(ro • r)uds =—ro x nda. (3.2.14)

as s
3.2.4 Laplace's and Poisson's Equations

This section is a very brief introduction to potential theory and harmonic
functions. Gauss's Theorem will be the main tool in the investigation.
A scalar function / is called harmonic in a domain G if / is twice con-
tinuously differentiable in G and V 2 / = 0. Recall that V 2 / = div(grad/),
so that, in cartesian coordinates, V 2 / = fxx + fyy in M2 and V 2 / =
fxx + fyy + fzz in K3. Similar expression exist in every M", n > 3. While
we restrict our discussion to M3, all the results of this section are valid in
every Mn, n > 2.
T h e equation
V 2 / - g, (3.2.15)
where g is a known continuous function, is called P o i s s o n ' s equation,

after t h e French m a t h e m a t i c i a n S I M E O N D E N I S P O I S S O N (1781-1840). T h e
particular case of t h e Poisson equation,
V 2 / = 0, (3.2.16)
is i m p o r t a n t enough to have its own name, a n d is called L a p l a c e ' s

e q u a t i o n , after t h e French mathematician P I E R R E - S I M O N LAPLACE
(1749-1827). P o t e n t i a l t h e o r y is t h e branch of mathematics studying
these two equations; t h e large number of problems in b o t h m a t h e m a t i c s
and physics t h a t are reduced t o either (3.2.15) or (3.2.16) justifies t h e allo-
cation of a whole branch of m a t h e m a t i c s t o the study of just two equations.
This section provides t h e most basic introduction t o potential theory. While
the results we discuss are t r u e in both two a n d three dimensions, our pre-
sentation will be in R 3 , b o t h for the sake of concreteness, a n d because in K 2
complex numbers provide a much more efficient method t o study Laplace's
and Poisson's equations; see Theorem 4.2.6 on page 205 below.
Let / be a continuously differentiable function in a domain G in E 3 ,

and S, a closed orientable piece-wise smooth surface. Denote by ris =
ns{P) t h e outside unit normal vector t o S a t t h e point P. T h e n o r m a l
d e r i v a t i v e of / a t P is, by definition,
^ • ( P ) = V / ( P ) • ns(P); where V / = g r a d / . (3.2.17)

an
E X E R C I S E S^.IO.*^ Assume that f is a harmonic function in a domain G

and S, a closed orientable piece-wise smooth surface in G. Assume that the
domain enclosed by S is a subset of G. (a) Show that
-i-da = 0. (3.2.18)
/ / an
s
Hint: Apply Gauss's Theorem to F = V / . (b) Show that (3.2.18) can fail
if the domain enclosed by S contains points that are not in G. Hint: take
/ = r _ 1 and G, the unit ball around the origin with the center removed.
T h e following result is t h e analog of the integration-by-parts formula

from t h e one-variable calculus.
Theorem 3.2.5 Let G be a domain in R 3 . Assume that the boundary dG

of G consists of finitely many closed piece-wise smooth orientable surfaces.
As usual, we choose the orientation of dG so that every unit normal vector
to dG points outside of G. Let f, g be two scalar fields defined in an open
set G\ containing G.
(a) If f is continuously differentiable in G\ and g is twice continuously
differentiable, then
HI (V2<?) fdV = - HI Vf-VgdV + Hf^ da. (3.2.19)

G G dG
(b) If both f and g are twice continuously differentiable in G\, then
///(/V^-9W)^ =/ / ( / | - 9 i ) *. (3.2,0)
G dG
c
EXERCISE 3.2.11. (a) Prove (3.2.19). Hint: apply Gauss's Theorem to the
vector field fVg. (b) Prove (3.2.20). Hint: in (3.2.19), switch f and g, then
take the difference of the two resulting identities.
Equalities (3.2.19) and (3.2.20) are known as Green's f i r s t and
second formulas (or identities).
EXERCISE 3.2.12? State and prove the two-dimensional versions of (3.2.18),
(3.2.19), and (3.2.20). Hint: you just have to replace triple integrals with
double, and surface integrals with line. For example, (3.2.19) becomes
II W2gdA = - II Vf -VgdA + j f ^da. (3.2.21)

G G dG
Laplace's and Poisson's equations are examples of partial differential

equations. While there exist many different methods for finding explicit so-
lutions of ordinary differential equations, most partial differential equations
are not explicitly solvable. As a result, theorems that ensure existence and
uniqueness of solutions of such equations are very useful. Below, we discuss
uniqueness of solution for the Poisson equation.
Theorem 3.2.6 Let S be a closed orientable piece-wise smooth surface.
Consider two functions fi, fi that are continuously differentiable in an open
set containing S and are harmonic everywhere in the domain Gs enclosed
byS.
(a) If fi = /2 on S, then f\ = fi everywhere in Gs-
(b) If dfi/dn = dfi/dn on S then there exists a real number c so that

/ i — fi = c everywhere in Gs-
EXERCISE 3.2.13. c Prove Theorem 3.2.6. Hint: consider the function f =

/i — fi and apply (3.2.19) with g = f to conclude that V / = 0 everywhere in Gs-
Therefore f must be constant (Exercise 3.1.3 on page 125); in part (a), f = 0 on
S, so the constant must be zero.
Let G be a bounded domain whose boundary dG is a closed orientable
piece-wise smooth surface. Let g be a continuous function in G, and h,
a function continuous in some domain containing dG. The D i r i c h l e t , or
f i r s t , boundary value problem for the Poisson equation (3.2.15) in G is
to find a function / that is twice continuously differentiable in an open
set containing G, satisfies equation (3.2.15) everywhere in G, and / = h
on dG. The Neumann, or second, boundary value problem for the Poisson
equation is to find a function / that is twice continuously differentiable in
an open set containing G, satisfies equation (3.2.15) everywhere in G, and
df/dn — h on dG. With some additional constructions, a formulation of
both problems is possible by considering the function / only in the closure
of G, that is, without using an open set containing G
EXERCISE 3.2.14. c (a) Show that the Dirichlet problem for the Poisson
equation can have at most one solution. Hint: use the first part of Theorem
3.2.6; the difference of two solutions is a harmonic function, (b) Show that
every two solutions of the Neumann problem differ by a constant. Hint: use
the second part of Theorem 3.2.6. (c) Show that the Neumann problem for the
Laplace equation has no solutions unless J J hda = 0. Hint: use (3.2.18).
dG
The difference between the Dirichlet and Neumann problems is the
boundary c o n d i t i o n s imposed on the function / : in the first case, we
prescribe the values on the boundary of G to the function itself, and in the
second case, to the normal derivative of the function. A combination of the
two is possible and is called the Robin boundary value problem, when we
prescribe the values on dG to the expression af + (df/dn), where a is a
known function defined on dG. Finally, there is an oblique d e r i v a t i v e
problem, when the values on dG are prescribed to the expression V / • u,
where ix is a continuous unit vector field on dG. We only mention these
boundary conditions to illustrate the numerous problems that can be stud-
ied for the Poisson and Laplace equations.
While the Dirichlet problem is named after the German mathemati-
cian J.P.G.L. Dirichlet, who did pioneering work in potential theory, the
origins of the other two names, Neumann and Robin, are less clear. Most
probably, the Neumann problem is named after the German mathematician
CARL G O T T F R I E D NEUMANN (1832-1925) for his contributions to poten-
tial theory. His father FRANZ E R N S T NEUMANN (1798-1895) worked in
mathematical physics and could have contributed to the subject as well.
We conclude this section by establishing the Dirichlet principle for the
Poisson equation. As before, let G be a bounded domain whose boundary
dG is a closed orientable piece-wise smooth surface. Let g be a continuous
function on G, and h, a function continuous on some domain containing
dG. We assume that all functions are real-valued.
Denote by U the set of functions so that every function
in U is twice continuously differentiable in an open set
containing G and is equal to h on dG.
For every l i g M , define the number
I{u) = Jjj (±\\Vu\\2+g^dV.
In mathematics, a rule that assigns a number to a function is called a

functional; thus, 7 is a functional defined on the set U.
T h e o r e m 3.2.7 ( D i r i c h l e t p r i n c i p l e ) If f is a function in U and

V 2 / = g in G, then
I{f)= mm I(u). (3.2.22)
Conversely, if the function f in U satisfies (3.2.22), then V 2 / = g in G.
Proof. Assume that / is a function in U and V 2 / = g in G. Then
o = / / / ( v 2 / -g)(f- u)dv = fJJ v 2 / (/ - u)dv - JJfgU ~ u)dv

G G G
(3.2.23)
for every u £U. Being elements of the set U, both / and u are equal to h
on dG. Therefore, / - u = 0 on dG, and we find from (3.2.19) that
HI V2/ (/ - u)dV = - HI || V/||2 dV + JJIV/ • VudV. (3.2.24)

Combining (3.2.23) and (3.2.24) yields
/ / / ("V/"2 + f9) dV =
/ / / ( V /
' VU + 9U)
^ (3 2 25)
- -
G G
By the definition of the dot product, | V / • V u | < | | V / | | ||Vu||. Also, for

every two non-negative numbers a, b, 0 < (a — b)2 = a2 + b2 — 2ab, which
means that ab < (a2 + b2)/2. Therefore, (3.2.25) implies
///(ll V/||2 + fg) dV<\ JJJ \\Vf\\2dV + jJj(\\\Vu\\2 + s«) dV.

G G G
Hence, / ( / ) < I(u), which is equivalent to (3.2.22).

Conversely, assume that / satisfies (3.2.22). Take any function v that
is twice continuously differentiable in G and is equal to zero outside of an
open set contained in G. For this function v define the function F = F(t)
by F(t) = I(f + tv). By assumption, t = 0 is a local minimum of F. Direct
computations show that F = F(t) is differentiable and
F'(0) = [[[{Vf • Vv + gv) dV. (3.2.26)

G
From the one-variable calculus we know that F'(0) must be equal to zero.
Applying (3.2.19) and keeping in mind that, by assumption, v = 0 on dG,
we conclude that
(~Vzf + g)vdV = Q. (3.2.27)

///
G
Since the function v is arbitrary, this implies V 2 / = g in G and completes

the proof of the theorem. •
EXERCISE 3.2.15. 5 (a) Verify (3.2.26). Hint: replace u in the definition of I

with f + tv and expanding the squares; note that, by construction, f + tv €U for
every v and t. (b) Verify that (3.2.27) indeed implies V 2 / — g in G. Hint-
assume that V 2 / > g at some point P in G. Argue that the same inequality must
then hold in some open ball centered at P and lying completely in G. Then take
a function v that is equal to one in that ball — there is still some space between
the boundary of the ball and the boundary of G where v can come down to zero
— and come to a contradiction with (3.2.27).
Theorem 3.2.7 holds also in M2, and, in fact, in every R n , n > 2.
Equations in Vacuum 163
3.3 Maxwell's Equations and Electromagnetic Theory
Electromagnetic theory, or electromagnetism, is the branch of physics that

studies electric and magnetic fields. Numerous experiments have shown
that these two fields are physically related. Maxwell's equations provide a
mathematical model that establishes the precise relation between the elec-
tric and magnetic fields. The original publication in 1864 by the Scottish
mathematician and physicist JAMES CLERK MAXWELL (1831-1879) con-
tained 20 equations in 20 unknowns. Later, in the 1880s, J. W. Gibbs and
O. Heaviside put the equations in a more compact vector form, which we
will present below. On the basic physical level, Maxwell's equations gener-
alize a number of experimental facts; there are some physicists these days
who believe that one cannot derive Maxwell's equations, although Maxwell
himself would probably disagree. On the mathematical level, these equa-
tions admit numerous derivations, for different models and on various levels
of abstraction, and we discuss some of these derivations below.
If an equation describes a physical law, the particular form of the equa-
tion often depends on the units used to measure the corresponding phys-
ical quantities. In our discussion of electromagnetism, we will use the
I n t e r n a t i o n a l System of Units (SI). Some of the units in SI are Am-
pere (A) for electric current, Coulomb (C) for electric charge, meter (m)
for length, Newton (N) for force, second (s) for time.
3.3.1 Maxwell's Equations in Vacuum

In vacuum, the electric and magnetic fields are described mathematically
by the vectors E and B, respectively. A possible interpretation of these
vectors is through the force F exerted by the fields on a point charge q
moving with velocity v.
F = q{E + vxB). (3.3.1)
The charges generating the fields can be stationary or moving; moving

charges constitute an electric current. Denote by p the density per unit
volume of free stationary charges, and by J , the density, per unit area, of
electric currents. In general, E, B, p, and J are time-dependent. Maxwell's
equations establish the connection between E, B, p, and J. We begin by
stating the equation, and then show how to derive them by combining
the basic empirical laws of electricity and magnetism with the theorems of
164 Maxwell's Equations
Gauss and Stokes from vector analysis.

In the International System of Units (SI), Maxwell's equations are
divE = P_1 (3.3.2)

£o
d i v B = 0; (3.3.3)
dB
curl E == - (3.3.4)
dt
dE
cmlB = /j,0J + fJ.oSo-Kr- (3.3.5)
The positive numbers Mo = 47T • 10" 7 N/A 2 , s0 = 8.85 • 1 0 - 1 2 C/(N-m 2 ) are

called the magnetic p e r m e a b i l i t y and e l e c t r i c a l p e r m i t t i v i t y of the
free space, respectively; the relation
c2 = — (3.3.6)
Mo£o
holds, where c is the speed of light in vacuum.
The starting point in the derivation of (3.3.2) is Coulomb's Law, discov-
ered experimentally in 1785 by the French physicist CHARLES AUGUSTIN
DE COULOMB (1736-1806). Originally stated as the inverse-square law for
the force between two charges, the result also provides the electric field E
produced by a point charge q: if q is located at the point O, then, for every
point P ^ O,
E{p) = (3 3 7)
= ^\oPJ ^^- --
Let S be closed piece-wise smooth surface enclosing the point O. Then
E(P) is denned at all points P on S. According to (3.2.5), page 152, / / E •
s
da = q/eo. By linearity, if there are finitely many point charges qi,-.-,qn
inside S, and E is the total electric field produced by these charges, then
f f E • dtr = (l/e0)f29k- (3-3.8)

fc=1
s
Now assume that there is a continuum of charges in the domain G enclosed
by S, and a small region GB around a point B G G has the approximate
charge /j(B)m(Gs), where p is a continuous density function and m(Gg) is
the volume of GB ', the smaller the region GB , the better this approximation
of the charge. Then (3.3.8) becomes
S G
and equation (3.3.2) follows from Gauss's Theorem (see Theorem 3.2.2).
EXERCISE 3.3.l. c Verify that (3.3.9) indeed implies (3.3.2).
Equation (3.3.9) is important in its own right and is known as Gauss's
Law of E l e c t r i c Flux.
Equation (3.3.3) is the analog of (3.3.2) for the magnetic field and re-
flects the experimental fact that there are no single magnetic charges, or
monopoles.
The starting point in the derivation of (3.3.4) is Faraday' s Law, discov-
ered experimentally in 1831 by the English physicist and chemist MICHAEL
FARADAY (1791-1867): a change in time of a magnetic field B induces
an electric field E. This law is the underlying principle of electricity gen-
eration: the electric field E produces an electric current in a conductor.
Mathematically, if S is a piece-wise smooth orientable surface with a closed
piece-wise smooth boundary dS so that the orientations of S and dS agree,
then
<3 3io
/ „ * • * — / / £ •"• - >
s
Equation (3.3.4) now follows from Stake's Theorem (Theorem 3.2.3). Note
that §QS E • dr = V is the voltage drop along dS. For a coil without a
ferromagnetic core, electric current I through the coil produces magnetic
field B whose flux is proportional to / : — JJ B • da = LI, where L is
s
the inductance of the coil. Then (3.3.10) becomes V = Ldl/dt, which is a
fundamental relation in the theory of electrical circuits.
EXERCISE 3.3.2.c Verify that (3.3.10) indeed implies (3.3.4).
The starting point in the derivation of (3.3.5) is an experimental fact
that a steady electric current produces a magnetic field. In 1820, the Danish
physicist HANS CHRISTIAN 0RSTED (1777-1851) observed that a steady
electric current deflects a compass needle. The French scientist ANDRE-
M A R I E A M P E R E (1775-1836) learned about this discovery on September
11, 1820; one week later, he presented a paper to the French Academy with
a detailed explantation and generalizations of this phenomenon. In 1827,

he gave a mathematical formulation, connecting the steady electric current
with the induced magnetic field. In modern terms, the formulation is as
follows Let 5 be a piece-wise smooth orientable surface with a piece-wise
smooth boundary OS so that the orientations of S and dS agree. If J is the
density, per unit area, of the stationary (that is, time-independent) electric
current, then the induced magnetic field B satisfies £„„ B-dr = fio JJ J-dcr;
s
see also Exercise 3.3.11 on page 170 below. By Stokes's Theorem, this
implies Ampere' s Law
curlB = /xoJ; (3.3.11)
Maxwell used the equation of continuity (3.2.8) and his first equation (3.3.2)
to extend (3.3.11) to time-varying currents in the form (3.3.5). Recall that
div (curl F) = 0 for every twice continuously differentiable vector field F.
Then, assuming B is twice continuously differentiable, we get from (3.3.11)
that div J = 0. On the other hand, if p is the density, per unit volume, of
the charges moving with velocity v, then J = pv, and, by the equation of
continuity with no sources and sinks,
^ + d i v J = 0, (3.3.12)
or div J = —dp/dt. For a stationary current, dp/dt = 0, which is consistent

with div(curlJB) = 0. For time-varying currents, we use (3.3.2) to find
dp/dt = e0div(dE/dt) so that div ( J + e0dE/dt) = 0. To ensure the
equality div(curlB) = 0, we therefore replace (3.3.11) with (3.3.5).
EXERCISE 3 . 3 . 3 . B GO
over the above arguments and verify that the diver-
gence of the right-hand side of (3.3.5) is indeed equal to zero.
Equations (3.3.4) and (3.3.5) express the observed interdependence be-
tween B and E, leading to the single theory of electromagnetism. This
is a striking example of the power of mathematical modelling in describ-
ing physical phenomena. A successful model can also predict new physical
phenomena, and we will see later how Maxwell's equations lead to the pre-
diction of electromagnetic waves (see page 348 below).
We conclude this section by establishing the connection between a sta-
tionary electromagnetic field and the Poisson equation. We start with the
stationary electric field. Recall that grad(l/r) — —r/r3. Equation (3.3.7)
therefore implies that the electric field E produced by a single point charge
is potential: E(P) = - W ( P ) , where U(P) = q/(4ne0 \OP\), P ^ O. By

linearity, the electric field at the point P, produced by n point charges is
E(P) = -VU(P), U(P) = J2Aqk H 1

„, (3-3.13)
>
where r^ = OOk, k = l,...,n, are the position vectors of the point charges,
rp = OP is the position vector of P, and we assume P ^ Ok for all k (draw
a picture).
EXERCISE 3.3.4.c Verify (3.3.13).
Now assume that there is a continuum of charges in K 3 with density p
so that the charge of a small region GQ around a point Q € M3 is approxi-
mately P(Q)TTI(GQ), where p is a twice continuously differentiable function
in R 3 and ^.(GQ) is the volume of GQ; the approximation becomes better as
the volume of GQ becomes smaller. We assume that p — 0 outside of some
bounded domain. This continuum of charges produces an electric field E
at every point P e K 3 . Using (3.3.13) and the result of Exercise 3.1.24,
E(P) = - W(P), U(P) = J L Jff jW dV, (3.3.14)

3
R
3
with integration over the points Q £ M where p(Q) is not zero.
EXERCISE S.S.S.*7 Verify (3.3.14). Hint: by assumption on p, the integral in
(3.3.14) is actually over a bounded domain.
Equalities (3.3.13) and (3.3.14) show that the stationary electric field
E, produced either by finitely many point charges or by a continuum of
distributed charges, is a potential vector field: E = —'VU. The solutions
of the vector differential equation r(t) = E(r(t)) for all possible initial
conditions are called l i n e s of force, and the surfaces { P e R 3 : U(P) =
c}, for all possible values of c, are called e q u i p o t e n t i a l s u r f a c e s .
EXERCISE 3.3.6? (a) Show that lines of forces are orthogonal to the equipo-
tential surfaces. Hint: VZ7 gives the normal direction to the surface, and, be-
cause E = — Vf7, also given the tangential direction for the line of force. (b)
Verify that the lines of force for a point charge are straight lines and the
equipotential surfaces are spheres with the charge at the center. Hint: for
a suitable f = f(t) and every unit vector u, the function r(t) = uf{t) satisfies
r(t) = cr(t)/\\r(t)f.
By (3.3.2), div.E = p/e0. The first equality in (3.3.14) then implies
V 2 t/ = - —, (3.3.15)
£o
that is, the potential of the electric field satisfies the Poisson equation
(3.3.15) in K 3 . Our arguments also suggest that the second equality (3.3.14)
gives a solution of (3.3.15); see Problem 4.2 on page 427 for the precise
statement and proof of the corresponding result.
EXERCISE 3.3.7. (a)c Verify that (3.3.15) follows from (3.3.2). (b)B
Taking for granted that the function U defined in (3.3.14) is twice con-
tinuously differentiable in M.3, use (3.3.15) to show that U is a harmonic
function outside of some bounded domain. Hint: recall that we assume that
p = 0 outside of some bounded domain.
The analogs of (3.3.2), (3.3.14), and (3.3.15) exist for every central,
inverse-square field, such as the gravitational field.
EXERCISE 3.3.8. c Let E = E{P), P e M3, be the g r a v i t a t i o n a l f i e l d
i n t e n s i t y vector, so that F = mE(P) is the force acting on a point
mass m placed at the point P. Verify that
E = -VV and V 2 V = 4TTGP, (3.3.16)
where G is the universal gravitational constant (see page 45) and p is the
density, per unit volume, of the mass producing the gravitational field. Hint:
for a point mass M at the origin, E{P) = —MGr~2r = M G V ( l / r ) . Then re-
peat the arguments that lead from (3.3.7) to (3.3.9) and from (3.3.13) to (3.3.15).
Let us now consider the magnetic field. By Maxwell's equation (3.3.4),
the vector field B is solenoidal, that is, has zero divergence. The condition
div B — 0 is satisfied if
B = cm\A (3.3.17)
for some vector field A; if we can find A, then we can find B. This field
A is called a v e c t o r p o t e n t i a l of B. Note that, if it exists, A is not
unique: for every sufficiently smooth scalar field / , A\ = A + grad / is
also a vector potential of B, because curl(grad/) = 0. On the other hand,
div A\ = div A + V 2 / ; with a suitable choice of / we can ensure that the
vector potential satisfies
div^4 = 0. (3.3.18)
This choice of A corresponds to Coulomb's g a u g e ; for more a b o u t gauging

see page 351.
Assume t h a t t h e current density J a n d t h e corresponding electric field
E do not depend on time. Then Maxwell's equation (3.3.5) becomes t h e
original Ampere's Law (3.3.11).
E X E R C I S E 3 . 3 . 9 . B Verify that (3.3.11), (3.3.17), and (3.3.18) imply
V 2 A = -HQJ. (3.3.19)
Hint: see (3.1.32) on page 139.
Let us assume t h a t t h e vector field J is twice continuously differentiable

and is equal t o 0 outside of a bounded domain. By analogy with (3.3.14)
a n d (3.3.15) (see also Problem 4.2 on page 427), we write t h e solution of
the vector Poisson equation (3.3.19) as
R3
3
with integration over t h e points Q g R where J is not zero.
E X E R C I S E 3.3.10. c Assuming that differentiation under the integral sign is

justified, verify that (3.3.17) and (3.3.20) imply
R3
Hint: Note that, for the purpose of differentiating the expression under the integral
in (3.3.20), the point Q is fixed and the point P is variable. Accordingly, place the
origin at the point Q; then you need to compute curl(J/r), where J is a constant
vector. Recall that V ( l / r ) = —r/r3 and use a suitable formula from the collection
(3.1.30) on page 139.
Formula (3.3.21) is t h e three-dimensional version of t h e B i o t - S a v a r t

Law, discovered experimentally in 1820 by t h e French physicists J E A N B A P -
T I S T E B I O T (1774-1862) a n d F E L I X SAVART (1791-1841); in its original
form, t h e law states t h a t t h e magnetic field produced by a constant current
/ in a n infinitesimally thin straight wire of infinite length is
B{P) (3 3 22)
-^^PQf' --
where u is the unit vector along the wire in the direction of the current,
Q is the point on the wire closest to P, and J is the current measured in
amperes. In particular, the magnitude of B is inversely proportional to the
distance from the wire.
EXERCISE 3.3.11. (a)B Draw a picture illustrating (3.3.22) and derive
(3.3.22) from (3.3.21). Hint: consider a uniform cylindrical wire of small ra-
dius. (b)c Let C be a circle of radius R in the plane perpendicular to the
wire and center the wire. Show that (3.3.22) implies § B • dr = fi0I- (c)B
C
Show that the result of part (b) is true for every simple closed piece-wise
smooth curve enclosing the wire.
EXERCISE 3.3.12^ LetC be a simple piece-wise smooth curve representing an
infinitesimally thin conducting wire with current I (measured in amperes);
the value of the current does not depend on time but can be different at
different points of the wire. Assume that orientation ofC is in the direction
of the current. Use (3.3.21) to show that the resulting magnetic field B and
the corresponding vector potential A at a point P not in C are
where Q is a (varying) point on C and u is the unit tangent vector to C.

Hint: assume that the wire has a small radius a and is bent as C. Then ira2J = Iu
andfff(---)dV = f(-.-)Tra2ds.
3.3.2 The Electric and Magnetic Dipoles

As an example illustrating some of the general discussions from the previous
section, we will consider electric and magnetic dipoles. In the following
section, these dipoles will help us to study the electromagnetic field in
material media.
An e l e c t r i c dipole is a pair of point charges q and — q, q > 0, with
equal magnitude and opposite signs, placed at a fixed distance from each
other. Let O be a reference point in R 3 , and let the position vectors of the
charges q, -q be r\ and r 2 , respectively. By (3.3.13), the potential U of
the resulting electric field at the point with position vector r is
l 1
U(r) = -?- ( - ) . (3.3.24)
47TE0 V l k - ^ l l l \\r-rA)
Let ro = {r\ + T2)/2 be the position vector of the midpoint of the line
Dipoles 171
segment connecting q and —q. In what follows, we will find the approximate
value of U at the points that are far away from both q and —q, that is, when
\\r — ro|| is much bigger than the distance ||ri — T*2|| between the point
charges. In Problem 4.3, page 429, we discuss the lines of force for the
dipole, that is, the solutions of the differential equation r(t) = —Vf/(r).
To simplify the computations, define x = r — ro and 1*12 = r\ — TI.
Then
U{r) = (3 3 25)
i ( | | t - ( l / 2 ) r 1 2 | | ~ ||t + ( l / 2 ) r 1 2 | | ) " - -
EXERCISE 3.3.13. C Draw a picture and verify (3.3.25).

For every two vectors u,v, we have ||u — v\\2 = (u — v) • (u — v) =
\\ \\ + \\v\\2 — 2u • v. If ||u|| >•
u 2
||v|| (that is, ||u|| is much larger than
jj«||), then ||« - v\\2 « ||u|| 2 (l - (2u • •u/||w|| 2 )). Applying the linear
approximation to the function f(x) = (1 - x)~1/2 at the point x = 0, we
find that f(x) « 1 + (x/2) and
1 1
\\u — v\\ \\u
+ T-^\ HI»NI- ( 3 - 3 - 26 )
For the exact expansion of l/||tt—v\\ in the powers of ||u||/||u||, see Problem
4.4, page 430. Applying this approximation to (3.3.25) with u = x = r — r0
and v = ± r i 2 / 2 , we find
I/(r)«-g-(r~ro)^12, ||r-r0||»||r12||. (3.3.27)

ATTEQ \\r - ro\\6
EXERCISE 3.3.14. B (a) Verify (3.3.27) and show that the approximation
error is of order (||T"I — T , 2||/||r — T"o||)2. (b) Using the relation E = —VC/,
verify that
E(r) * -Z- (^ r ^ - S ) • IkH » lki a (l, (3-3.28)
(i) by computing the gradient of the approximate value U from (3.3.27);

(ii) by computing the gradient of the exact value of U from (3.3.25) and
then approximating the result.
The vector d — qr 12, pointing from the negative charge to the positive
charge, is called the dipole moment of the electric dipole. Taking the limit
as T"i2 —> 0 and q —• 00 so that the vector d stays constant, we get the
point e l e c t r i c dipole. For the point electric dipole at the origin, we
have the potential and the electric field at every point with position vector
r as follows:
1 dr „, . 1 (Z(d-r)r d
U(r) = ~ ^ r ^ d5-, E(r) = ^ - (' ^ - ^ - —
£ ). (3.3.29)
47T£o r 47T£o
EXERCISE 3.3.15. (a)B Verify that V 2 (d • r/r3) = 0, r ^ 0, and ex-

plain how the result is connected with Maxwell's equation (3.3.2). Hint: use
(3.1.33) on page 140, as well as suitable relations from the collection (3.1.30)
on page 139; note that grad(d • r) = d. (b)A Consider a point electric dipole
with moment d in an external electric field E and assume the electric field
is uniform, that is, has the same magnitude and direction at every point
in space: E(P) = E, where E is a constant vector. Show that the total
force acting on the dipole by the field is zero and the torque is T = d x E.
Thus, the external field will tend to align the dipole with the field. Draw the
picture illustrating the stable orientation of the dipole relative to the field.
MAGNETIC DIPOLE. Since currents produce magnetic fields, we define

a magnetic d i p o l e as an infinitesimally thin closed circular wire in the
shape of a simple smooth closed curve C, carrying a constant current / . By
(3.3.23) on page 170, the vector potential characterizing the magnetic field
of this current is
where u is the unit tangent vector to C in the direction of the current, and
Q is a (varying) point on C. Denote by O the center of the circle C. Then
\PQ\ = \\OP — OQ||, and OQ is perpendicular to the tangent vector at the
point Q. As with the electric dipole, we want to get an approximation of
A at the point P that is far away from C, that is, when |OP| is much larger
than the radius \OQ\ of the circle (draw a picture!) Using (3.3.26), we find
1 1 / OP-OQ\
\PQ\ \OP\ \' 1+ \OP\2 J' '
Then (3.3.30) becomes
A(P)
*4^(£" + |wH (3331)
Equations in Material Media 173
Denote by n the unit vector in the direction of OQ x u, and by a, the

radius of the circle.
E X E R C I S E 3 . 3 . 1 6 . B (a) Verify that the direction of n is the same for every

point Q of the circle. How should a small bar magnet be placed to produce
the same magnetic field as the magnetic dipole? Hint: draw a picture; ifC is
a circle in the (i, j) plane and is oriented counterclockwise, then n = k. (b)
Verify that (3.3.31) can be written as
Hint: to show that fcuds = 0, use cartesian coordinates so that C is a circle in

the (i, j) plane and is oriented counterclockwise. Then u = — s i n t i + cost j and
ds = adt, 0 < t < 2-7T. For the second integral, use (3.2.14) on page 157 with
ro = OP, r = OQ, and S, the disk enclosed by C.
Similar to the electric dipole, we define the d i p o l e moment of the mag-

netic dipole as the vector m = -Ka2In. In the limit a —> 0 and I —> oo
so t h a t the vector m stays the same, we get the p o i n t m a g n e t i c d i p o l e .
If a point magnetic dipole is placed at the origin, then the corresponding
vector potential is
EXERCISE 3.3.17? Find the magnetic field B of the point magnetic dipole.
3.3.3 Maxwell's Equations in Material Media
In conductors, such as metals and electrolytes, there are charged particles

t h a t can move when an external electric field E is applied. In metals, these
particles are free electrons, and in electrolytes, ions. By O h m ' s Law, the
resulting current density J is proportional to the external field E:
J = aE, (3.3.34)
where a is the conductivity of the material. Similar to other laws of elec-

tricity and magnetism, relation (3.3.34) is a mathematical expression of
empirical observations and was originally published in 1827 by the German
physicist G E O R G S I M O N O H M (1789-1854). Today, (3.3.34) is known as
the microscopic form of Ohm's Law, as opposite to the more familiar macro-
scopic, or averaged-out, version V = IR, where V is the voltage measured

in volts, / is the current in amperes, and R is the resistance in ohms.
If an insulated conductor is placed in an external electric field, then the
induced electric current redistributes the charged particles on the surface
of the conductor so that, in equilibrium, the total electric field inside the
conductor is zero: if there were non-zero field inside the conductor, then,
by Ohm's Law, there would be a non-zero current, which is impossible
for an isolated conductor in equilibrium. The force lines of the electric
field are therefore normal to the surface of the conductor; otherwise, the
tangential component would produce a surface current. There is no charge
accumulations inside the conductor either, as such accumulations would
produce electric current. Thus, a closed conducting surface isolates the
interior from the exterior electric field. For that reason, delicate micro-
chips and electronic equipment are often stored and shipped in aluminum
foil or other conducting cover.
In dielectrics, such as glass, plastics, and rubber, there is little or no

free motion of charged particles when an external electric field is applied.
Instead, the external field penetrates the material, causing polarization of
molecules in the substance and producing numerous point electric dipoles.
In biological materials, cell membranes are good insulators against ion flux,
and ions form dipoles near the surface of cell membranes. Protein molecules
form dipoles by ionization of the amino-acid chains.
Denote by P the resulting density of dipole moments per unit volume
in the dielectric material. Recall that a point dipole has zero net charge;
direct computations show that the flux of the corresponding electric field E
(see (3.3.29) through every closed piece-wise smooth surface is zero, as long
as the dipole is not on the surface. Accordingly, instead of looking at the
flux of E, we will look at the flux of P. Note that P has the dimension of
charge per unit area, and so the divergence div P of P has the dimension
of charge per unit volume. Recall that the direction of the dipole moment
is from the negative charge to the positive. It is then natural to define the
scalar field pb so that pb = — div P. With this definition, if G is any part
of the dielectric, then / / / ( d i v P — pb) dV = 0, which is consistent with
G
the idea that there is no movement of charges inside the dielectric. The
scalar field pb is called the d e n s i t y of bound charges. It is also possible
that some free charges are present inside the dielectric, with density per
unit volume pf. By Gauss's Law of Electric Flux (3.3.9), a region in free
space with density pb + p/ of charges per unit volume produces the electric
field E so that div E = (pb + pf)/so- In the dielectric material, we have

Pb = — d i v P so that div(eo E + P) = pf. It is then natural to define the
vector D = eo E + P, called the e l e c t r i c displacement. We get the
following modification of (3.3.2):
divD = p/. (3.3.35)
Dielectric material is called linear if P is proportional to the external

field E:
P = eoXeE, D = e0E + e0XeE = eE, (3.3.36)
where Xe is called the s u s c e p t i b i l i t y and e = £o(l -I- Xe), the

p e r m i t t i v i t y of the dielectric material; K = 1+Xe is called the d i e l e c t r i c
constant or r e l a t i v e p e r m i t t i v i t y . In general, e (as well as Xe and K)
is a t e n s o r f i e l d and the value of e depends on both the location and
the direction. For time-dependent E, the value of e can also depend on
the frequency of E. In homogeneous isotropic materials, e is a constant
number so that (3.3.35) becomes div.E = Pf/£-
A M P E R E ' S LAW FOR MAGNETIZED MATERIALS. In material media,
there exists an internal magnetic field created by the motion of electrons
around the atoms. At distances much larger than the size of an atom,
the magnetic field produced by an electron moving around an atom is well
approximated by the field of a point magnetic dipole. These magnetic
dipoles affect the overall magnetic field in the material, just as the bound
charges in dielectrics affect the electric field.
Denote by M the density, per unit volume, of the point magnetic dipoles
in the material; this density is called magnetization. According to (3.3.33),
the overall vector potential A of the resulting magnetic field is
where G is the region in space occupied by the material, Q is a point in Q

and P is the point in M3 at which the magnetic field is computed. We now
treat P as a fixed point and Q, as variable. Then
(3 3 38)
^(wod-mr --
and (3.3.37) becomes
A(P) = g fjj M(Q) x grad (j^j dV. (3.3.39)

G
EXERCISE 3.3.18.5 Verify (3.3.38). Hint: use (3.1.33) on page UO, taking P
as the origin of the frame.
By (3.1.30) on page 139,
M(Q) x grad ( ^ ) = ^ curl A*(Q) - curl ( ^ M ) , (3.3.40)
and then we use relation (3.2.6) on page 153 to write (3.3.39) as
A{P) =
£ \IIIwQ\cmlMiQ)dV +
IIwQ\M{Q) X HGda
) '
\ G dG J
(3.3.41)
where dG is the boundary of G, which we assume to be piece-wise smooth,
closed, and orientable, and n g is the outside unit normal to dG. We
call Jb = curliW the bound current, and Kb = M x no, the surface
d e n s i t y of t h e bound current. Recall that a magnetic dipole is mod-
elled by a circular movement of electric charges, and the vector J& can be
interpreted as the total current produced by the charge movement in all
the point magnetic dipoles in G.
EXERCISE 3.3.19. (a)c Verify that (3.3.41) indeed follows from (3.3.39),
(3.3.40), and (3.2.6). (b)B Find the magnetic field corresponding to the
potential (3.3.41). Hint: recall that the magnetic field is curl A; in the resulting
computations, you now treat P as a variable and Q as fixed.
Suppose that the material has non-zero conductivity and there is a
steady current Je flowing in the material due to an external source of
electric field, for example, two electrodes embedded in the material and
connected to an external battery. The total current density in the material
is then J = J\, + Je = curliVT + Je. We assume the the system is in
equilibrium so that both J\, and Je do not depend on time. By the original
Ampere's Law (3.3.11) on page 166, the resulting magnetic field B satisfies
curlB = /x 0 (curlM + Je) or curl (B//xo — M ) = J e . We therefore define
the net magnetic field H, also called the magnetic f i e l d s t r e n g t h , as
H = B/fio - M, (3.3.42)
so that Ampere's Law in material media becomes
curlH = Je. (3.3.43)

B
EXERCISE 3.3.20. Verify that, in material media, Maxwell's equation
(3.3.5) becomes
cm\H
= Je + ^ - . (3.3.44)
at
Hint: use the same arguments as in the derivation of (3.3.5).
In paramagnetic materials, the magnetic dipoles are aligned so that the
induced magnetic field B is stronger than in vacuum; the vector M has, on
average, the same direction as H: Jff(M • H) dV > 0. In diamagnetic
G
materials, the magnetic dipoles are aligned so that the induced magnetic
field B is weaker than in vacuum; the directions of M and H are, on the
average, opposite: JJf(M • H) dV < 0.
G
A material is called linear if magnetization M depends linearly on H,
that is, M = XmH, where Xm is called the magnetic s u s c e p t i b i l i t y of
the material. This linear relation holds for most materials when \\H\\ is
sufficiently small. By (3.3.42),
B = ii H, where /x = fi0(l + Xm)

is called the p e r m e a b i l i t y of the material. In vacuum, Xm — 0 and fi =
I/O- Similar to (electric) permittivity, permeability is, in general, a tensor
field and the value of /i depends on both the location and the direction.
For time-varying magnetic fields, the value of \i can also depend on the
frequency of the field. The value of it usually depends also on temperature
and density. In linear homogeneous isotropic materials, /x is a number and
(3.3.43) becomes
V x B = iiJe, (3.3.45)
which is the same as (3.3.11), but with \i instead of \i§.

EXERCISE 3.3.21. c Assuming the frequencies of B and E are constant,
verify that, in a linear homogeneous isotropic material, equation (3.3-44)
becomes
dE
cmlB = nJe + /j,e~. (3.3.46)
Hint: use (3.3.36).

For most linear materials, \xm\ ls close to zero. In linear homogeneous

isotropic paramagnetic materials, Xm > 0 and fj, > IIQ. For example, alu-
minum has Xm « 2 x 1 0 - 5 . In linear homogeneous isotropic diamagnetic
materials, Xm < 0 and fi < HQ. For example, both copper and water have
Xm ^ —10~5 (we write « rather than = because the precise values are
sensitive to the temperature of the materials and the frequency of the field;
the above values correspond to room temperature and a time-homogeneous
field).
The main examples of nonlinear materials are ferromagnetics and
superconductors. In ferromagnetics, the induced magnetic field is much
stronger than in vacuum. In superconductors the induced magnetic field is
essentially zero.
Let us summarize Maxwell's equations in material media:
d\vD =: P/; (3.3.47)
divB = 0; (3.3.48)
dB
curlJ3 = ) (3.3.49)
at
dD
curl i f := Je + (3.3.50)
dt'
where pf is the density, per unit volume, of the free electric charges, and Je
is the density, per unit area, of the externally produced electric currents.
These equations describe the electromagnetic field at distances much larger
than atomic size (atomic size is usually taken as 1 0 - 1 0 meters.) In Section
6.3.4, we further investigate Maxwell's equations for some simple models,
both in vacuum and in material media.
At sub-atomic distances, of the order of e 2 /(4nem e c 2 ) « 2.8 •
10~ 17 m, Maxwell's equations are no longer applicable, and quantum
electrodynamics takes over. A quantum-theoretic analog of Maxwell's
equations was suggested in 1954 by two American physicists, C H E N - N I N G
YANG (b. 1922) and ROBERT L. MILLS (1927-1999). Mathematical anal-
ysis of these Yang-Mills equations is mostly an open problem and is
outside the scope of our discussion. In fact, this analysis is literally a
million-dollar question, being one of the seven Millennium Problems an-
nounced by the Clay Mathematics Institute in 2000; for details, see the
book The Millennium Problems by K. Devlin, 2002.

Vector Analysis and Classical

Uploaded by

Document Informationclick to expand document informationrevision of mathematical physics

Document Informationclick to expand document information

Copyright:

Available Formats

Vector Analysis and Classical

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Vector Analysis and Classical

Uploaded by

Copyright:

Available Formats

Chapter 3

Vector Analysis and Classical

3.1 Functions of Several Variables

3.1.1 Functions, Sets, and the Gradient

{(P, f(P)), P G A} is called a s c a l a r f i e l d ; the function / is also called

at A and radius b. A point in a set is called i n t e r i o r if there exists a

• Bounded, if it lies entirely inside an open ball of sufficiently large radius.

The c l o s u r e of a s e t is the set together with all its boundary points.

lim \f(A)-f(B)\=0. (3.1.1)

lim f(B) = f(A).

Similarly, a vector field F is called continuous at a point A if F is defined

lim ||F(J4)-F(B)||=0. (3.1.2)

Next, we discuss the question of differentiability for scalar fields. Let

The function / is called dif f e r e n t i a b l e a t t h e p o i n t A if / is de-

where ABt = t u. If / is differentiable at the point A, then the correspond-

Duf{A) = Vf{A)-u. (3.1.5)

f(B) = f(A) + (Vf(A) • uB) \AB\ + s(B), (3.1.6)

where lim| A B |_ 0 |e(JB)|/|AB| = 0. The number V / ( A ) • UB is

EXERCISE 3.I.3. 5 Show that if V / = 0 in a domain G, then f is constant

fx{x0,yo)= lim - - - (3.1.7)

EXERCISE 3.1.4. C Verify that, in cartesian coordinates, fx = D\f.

/** - dx , Jxy - dy , Jvx - Qx , etc.

V ( / 5 ) = fVg + gVf, V(f/g) = (gVf - fVg)/g2.

The next two results should be familiar from a course in multi-variable

V / ( A ) = fx(A) i + fy(A) j + fz(A) «• (3-1-8)

EXERCISE 3.1.6. C Let r = r(t) be a vector-valued function, and f, a scalar

g'(t0) = Vf(A)-r'(to). (3.1.9)

Try to argue directly by the definitions without using partial derivatives.

((r'(i) • V)G(r(*))) • n = r'(t) • ( V ( G • n)(r(t))).

With this definition,

~G(r(t)) = (r'(t) • V)G(r(t)). (3.1.10)

n-((F-V)G) =F-(V(G-n)). (3.1.11)

/ are surfaces. This observation leads to several possible ways to define

EXERCISE 3.1.7. (a)B Give an example of a curve that cannot be repre-

(c) Let r = r(u,v) be a parametric representation of the surface. As-

The same surface can have different representations. The graph z =

EXERCISE 3.1.10. c Verify that the surface x = ^/4 — y2 — z2 can be rep-

EXERCISE 3.1.11. B Verify that r(u,v) = ui + vj + (2u2 + 3v2)k and

OPk+i =OPk- o fc V/(Pfc). (3.1.12)

3.1.2 Integration and Differentiation

l f{x)dx = lim V/(4)Aa;fc,

where a = xo < • • • < XN = b is a partition of the interval [a,b\, Axk =

The limit in (3.1.13) must be independent of the particular choice of

lG= fdm (3 Li4)

[ fds= [ f(r(t))\\r'(t)\\dt, (3.1.15)

f Fdr= f F(r{t))-r'{t)dt, (3.1.16)

L'-*-L F-uds, (3.1.17)

We say that F is a p o t e n t i a l vector field with potential / if F = V /

I Vf-dr = f(r(b))-f(r(a)). (3.1.18)

In particular, §c V / • dr = 0 for every simple closed curve C.

J F-dr = £K(b) - SK(a) = V(r(a)) - V(r(b)),

Theorem 3.1.1 A continuous vector field has the path independence

T H E AREA INTEGRAL. We now take the set G in (3.1.13) to be a subset

EXERCISE 3.1.14. (a)B Let G be the parallelogram with vertices

fffdA= J ffdxdy = f ffdydx = f ffrdr dO = f f fdO rdr.

m(S) = If \\ru x rv\\dA(u,v), (3.1.19)

where dA(u, v) is the area element of G\. We call

da = \\ru x rv\\ dA(u,v)