Appendix A
Vector Calculus in Curvilinear
Coordinates
A.I Introduction
In this Appendix I sketch proofs of the three fundamental theorems of vector calculus. My
aim is to convey the essence of the argument, not to track down every epsilon and delta.
A much more elegant, modem, and unified-but necessarily also much longer-treatment
will be found in M. Spivak's book, Calculus on Manifolds (New York: Benjamin, 1965).
For the sake of generality, I shall use arbitrary (orthogonal) curvilinear coordinates
(u, v, w), developing formulas for the gradient, divergence, curl, and Laplacian in any such
system. You can then specialize them to Cartesian, spherical, or cylindrical coordinates, or
any other system you might wish to use. If the generality bothers you on a first reading, and
you'd rather stick to Cartesian coordinates, just read (x, y, z) wherever you see (u, v, w),
and make the associated simplifications as you go along.
A.2 Notation
We identify a point in space by its three coordinates, u, v, and w, (in the Cartesian system,
(x, y, z); in the spherical system, (r, e, ¢); in the cylindrical system, (s, ¢, z)). I shall
assume the system is orthogonal, in the sense that the three unit vectors, D, Y, and w,
pointing in the direction of the increase of the corresponding coordinates, are mutually
perpendicular. Note that the unit vectors are functions of position, since their directions
(except in the Cartesian case) vary from point to point. Any vector can be expressed in
terms of D, Y, and w-in particular, the infinitesimal displacement vector from (u, v, w) to
(u + du, v + dv, w + dw) can be written
dl = f du D+ g dv Y+ h dw w, (A.1)
547
548 APPENDIX A. VECTOR CALCULUS IN CURVILINEAR COORDINATES
where I, g, and h are functions of position characteristic of the particular coordinate
system (in Cartesian coordinates I = g = h = 1; in spherical coordinates 1= 1, g = r,
h = r sine; and in cylindrical coordinates 1= h = 1, g = s). As you'll soon see, these
three functions tell you everything you need to know about a coordinate system.
A.3 Gradient
If you move from point (u, v, w) to point (u + du, v + dv, w + dw), a scalar function
t(u, v, w) changes by an amount
at at at
dt = - du +- dv +- dw; (A2)
au av aw
this is a standard theorem on partial differentiation. 1 We can write it as a dot product,
dt = Vt . dl = (Vt)u I du + (Vt)v g dv + (Vt)w h dw, (A3)
provided we define
1 at I at I at
(Vt)u==jau' (Vt)v == --, (Vt)w == - - .
g av haw
The gradient of t, then, is
I at ~ I at ~ I at ~
Vt == - - u + --v+ --w. (A4)
I au g av haw
If you now pick the appropriate expressions for I, g, and h from Table A.l, you can easily
generate the formulas for V t in Cartesian, spherical, and cylindrical coordinates, as they
appear in the front cover of the book.
System u v w I g h
Cartesian x y z I I I
Spherical r e ¢ I r r sin e
Cylindrical s ¢ z I s I
Table Al
From Eq. A3 it follows that the total change in t, as you go from point a to point b
(Fig. A.1), is
t(b) - t(a) = l b
dt = l\Vt) . dl, (AS)
which is the fundamental theorem for gradients (not much to prove, really, in this case).
Notice that the integral is independent of the path taken from a to b.
1M. Boas, Mathematical Methods in the Physical Sciences, 2nd ed., Chapter 4, Sect. 3 (New York: John Wiley.
1983).
A.4. DIVERGENCE 549
dl
Figure AI
A.4 Divergence
Suppose that we have a vector function,
A(u, v, w) = Au D + A v v + Aww,
and we wish to evaluate the integral fA, da over the surface of the infinitesimal volume
generated by starting at the point (u, v, w) and increasing each of the coordinates in sUCces-
sion by an infinitesimal amount (Fig. A2). Because the coordinates are orthogonal, this is
(at least, in the infinitesimal limit) a rectangular solid, whose sides have lengths diu = f du,
dl v = g dv, and dl w = h dw, and whose volume is therefore
dr = diu dl v dl w = (fgh)dudvdw. (A6)
(The sides are not just du, dv, dw-after all, v might be an angle, in which case dv doesn't
even have the dimensions of length. The correct expressions follow from Eq. AI.)
For the front surface,
da = -(gh) dv dw D,
so that
A . da = -(ghA u ) dv dw.
The back surface is identical (except for the sign), only this time the quantity ghA u is to be
evaluated at (u + du), instead ofu. Since for any (differentiable) function F(u),
dF
F(u + du) - F(u) = - du,
du
(in the limit), the front and back together amount to a contribution
a] I a
u ) dudvdw = --(ghAu)dr.
[-(ghA
au fgh au
By the same token, the right and left sides yield
I a
- - ( f h A )dr
fgh av v ,
550 APPENDIX A. VECTOR CALCULUS IN CURVILINEAR COORDINATES
(u + du, v + dv, W + dw)
Figure A.2
and the top and bottom give
I a
- h -(fgA w ) dr.
fg aw
All told, then,
f A· da =-I [a-(ghA u ) + -(fhA
a a ]
v ) + -(fgA w ) dr. (A.7)
fgh au av aw
The coefficient of dr serves to define the divergence of A in curvilinear coordinates:
V· A == - I [a-(ghA u ) + -(fhA
a a ]
v ) + -(fgA w ) , (A.8)
fgh au av aw
AA. DIVERGENCE 551
and Eq. A 7 becomes
fA, da = (V . A) dr. (A9)
Using Table AI, you can now derive the formulas for the divergence in Cartesian, spherical,
and cylindrical coordinates, which appear in the front cover of the book.
As it stands, Eq. A9 does not prove the divergence theorem, for it pertains only to
infinitesimal volumes, and rather special infinitesimal volumes at that. Of course, a finite
volume can be broken up into infinitesimal pieces, and Eq. A9 can be applied to each one.
The trouble is, when you then add up all the bits, the left-hand side is not just an integral
over the outer surface, but over all those tiny internal surfaces as well. Luckily, however,
these contributions cancel in pairs, for each internal surface occurs as the boundary of two
adjacent infinitesimal volumes, and since da always points outward, A .da has the opposite
sign for the two members of each pair (Fig. A3). Only those surfaces that bound a single
chunk-which is to say, only those at the outer boundary-survive when everything is
added up. For finite regions, then,
fA, da = ! (V . A) dr, (A 10)
and you need only integrate over the external surface. 2 This establishes the divergence
theorem.
Figure A3
2What about regions that cannot be fit perfectly by rectangular solids no matter how tiny they are-such as
planes cut at an angle to the coordinate lines? It's not hard to dispose of this case; try thinking it out for yourself,
or look at H. M. Schey's Div, Grad. Curl and All That (New York: W. W. Norton, 1973), starting with Prob. 11-15.
552 APPENDIX A. VECTOR CALCULUS IN CURVILINEAR COORDINATES
A.5 Curl
To obtain the curl in curvilinear coordinates, we calculate the line integral,
around the infinitesimal loop generated by starting at (u, v, w) and successively increasing
u and v by infinitesimal amounts, holding w constant (Fig. A4). The surface is a rectangle
(at least, in the infinitesimal limit), of length diu = f du, width di v = g dv, and area
da = (fg)dudvw. (All)
Assuming the coordinate system is right-handed, w
points out of the page in Fig. A4.
Having chosen this as the positive direction for da, we are obliged by the right-hand rule
to run the line integral counterclockwise, as shown.
(u + du, v, w)
~!4--'-.J--l,"""""'~ tI
Figure A4
Along the bottom segment,
dl =f duo,
so
A . dl = (f Au) duo
Along the top leg, the sign is reversed, and f Au is evaluated at (v + d v) rather than v.
Taken together, these two edges give
A. 5. CURL 553
Similarly, the right and left sides yield
so the total is
J A· dl = [~(gAv)
j au - ~(f Au)]
av du dv
(A12)
= -1 [a-(gA v ) - -a( f Au) JAw· da.
fg au av
The coefficient of da on the right serves to define the w-component of the curl. Constructing
the u and v components in the same way, we have
v x A == -l[a a JA l[aaw Au) - -(hAw)
a JAv
gh av aw v ) u + -fh --(f
-(hAw) - -(gA
au
(Al3)
and Eq. All generalizes to
fA. dl = (V x A) . da. (A14)
Using Table AI, you can now derive the formulas for the curl in Cartesian, spherical, and
cylindrical coordinates.
Equation A14 does not by itself prove Stokes' theorem, however, because at this point
it pertains only to very special infinitesimal surfaces. Again, we can chop any finite surface
into infinitesimal pieces and apply Eq. A14 to each one (Fig. A5). When we add them up,
though, we obtain (on the left) not only a line integral around the outer boundary, but a lot
of tiny line integrals around the internal loops as well. Fortunately, as before, the internal
Figure A5
554 APPENDIX A. VECTOR CALCULUS IN CURVILINEAR COORDINATES
contributions cancel in pairs, because every internal line is the edge of two adjacent loops
running in opposite directions. Consequently, Eq. Al4 can be extended to finite surfaces,
fA, dl = ! (V x A) . da, (AI5)
and the line integral is to be taken over the external boundary only.3 This establishes Stokes'
theorem.
.6 Laplacian
Since the Laplacian of a scalar is by definition the divergence of the gradient, we can read
off from Eqs. A4 and A8 the general formula
2
V t ==
I [a (ghfa;;at) + iha (fh
fih a;;
at) a (fg at )]
g-ih + a;;; ha;;; . (AI6)
Once again, you are invited to use Table Al to derive the Laplacian in Cartesian, spherical,
and cylindrical coordinates, and thus to confirm the formulas inside the front cover.
3What about surfaces that cannot be fit perfectly by tiny rectangles, no matter how small they are (such as
triangles) or surfaces that do not correspond to holding one coordinate fixed? If such cases trouble you, and you
cannot resolve them for yourself, look at H. M. Schey's Div. Grad, Curl, and All That, Prob. III-2 (New York: W.
W. Norton, 1973).
