# Stokes’ theorem 1

**Chapter 13 Stokes’ theorem
**

In the present chapter we shall discuss R

3

only. We shall use a right-handed coordinate

system and the standard unit coordinate vectors ˆı, ˆ ,

ˆ

k. We shall also name the coordinates

x, y, z in the usual way.

The basic theorem relating the fundamental theorem of calculus to multidimensional in-

tegration will still be that of Green. In this chapter, as well as the next one, we shall see

how to generalize this result in two directions. In this chapter we generalize it to surfaces in

R

3

, whereas in the next chapter we generalize to regions contained in R

n

. But in all of these

procedures it is still Green’s theorem that is fundamental.

A. Orientable surfaces

We shall be dealing with a two-dimensional manifold M ⊂ R

3

. We’ll just use the word

surface to describe M. There are two features of M that we need to discuss ﬁrst.

The ﬁrst is the idea of a normal vector for M. We assume that M is of class C

1

, so that at

each point p ∈ M there is a vector of unit norm which is orthogonal to M, in the sense that

it is orthogonal to the tangent space T

p

M. There are of course two choices of such a normal

vector, and we now need to make a choice.

DEFINITION. The surface M is said to be orientable if there exists a unit normal vector

´

N(p) at each point p ∈ M which is a continuous function of p.

The continuity of

´

N(p) is all-important. For instance, one can construct a M¨obius strip

and obtain a surface which is not orientable:

imagecreated

with Mathematica®

In case a surface is described implicitly by an equation

g(x, y, z) = 0

such that ∇g is never 0 at any point of the surface, and if g is of class C

1

, then ∇g is continuous

and we have two choices for

´

N:

´

N =

∇g

∇g

or

´

N = −

∇g

∇g

.

2 Chapter 13

For example, for the ellipsoid

x

2

a

2

+

y

2

b

2

+

z

2

c

2

= 1

we may take

´

N =

x

a

2

,

y

b

2

,

z

c

2

x

2

a

4

+

y

2

b

4

+

z

2

c

4

.

For the sphere S(0, a) we have in particular either

N

´

N =

(x, y, z)

a

or

´

N = −

(x, y, z)

a

.

REMARK. An orientable surface is also said to be two-sided. The reason for this is that

the continuous normal vector

´

N serves to deﬁne a direction of “up” at points of M. Thus at

points of M there is a deﬁnite sense of two sides of M, an “up” side and a “down” side. A

M¨obius strip for example is one-sided, which may be demonstrated by drawing a curve along

the “equator” of M with a pencil.

EXTENSION. Frequently we shall need to analyze a surface M ⊂ R

3

which is not actually

orientable in the above sense, but is “close enough.” The surface may consist of ﬁnitely many

surfaces with the proper orientability. A few examples should suﬃce for a good explanation.

The surface of a cube. If the cube is [−1, 1] ×[−1, 1] ×[−1, 1], then the surface consists of

the six squares making up the boundary of the solid cube. A typical face is

{1} ×[−1, 1] ×[−1, 1] = {(1, y, z) | −1 ≤ y ≤ 1, −1 ≤ z ≤ 1}.

Now we may choose

´

N at each point except those points on the edges to point (say) outward.

Thus for the above face we have

´

N = (1, 0, 0) = ˆı. On the opposite face {−1}×[−1, 1] ×[−1, 1]

we would have

´

N = −ˆı. We thus regard this surface as orientable.

N=k

N= k -

Stokes’ theorem 3

The boundary of a hemiball. For instance consider the hemiball

x

2

+ y

2

+ z

2

≤ a

2

, z ≥ 0.

Then the surface we have in mind consists of the hemisphere

x

2

+ y

2

+ z

2

= a

2

, z ≥ 0,

together with the disk

x

2

+ y

2

≤ a

2

, z = 0.

If we choose the inward normal vector, then we have

´

N =

(−x, −y, −z)

a

on the hemisphere,

´

N =

ˆ

k on the disk.

A cylindrical can. Consider the surface for which one part is given by

x

2

+ y

2

= a

2

, 0 ≤ z ≤ h,

and the other part by

x

2

+ y

2

≤ a

2

, z = 0.

Then we might choose an “outer” unit normal vector

´

N =

(x,y,0)

a

for 0 < z ≤ h,

´

N = −

ˆ

k for z = 0.

PROBLEM 13–1. The following surface is orientable. It consists of the union of the

cylinder

x

2

+ y

2

= a

2

, 0 ≤ z ≤ h;

and the hemisphere

x

2

+ y

2

+ (z −h)

2

= a

2

, h ≤ z ≤ h + a.

Draw a sketch of it and write down expressions for an “outer” unit normal vector.

4 Chapter 13

PROBLEM 13–2. Give formulas for an “ice cream cone” surface, consisting of a

right circular cone topped oﬀ with a hemisphere. Then give formulas for the ‘outer” unit

normal vector.

All of the surfaces we shall be considering will be connected. Each will be piecewise C

1

and any two points on M can be joined by a piecewise C

1

curve lying in M. Each will be

orientable as well and we shall be faced with just two choices for

´

N. This leads to one more

DEFINITION. An orientable surface M is said to be oriented if a deﬁnite choice has been

made of a continuous unit normal vector

´

N for M.

There is actually a touch of vagueness in this deﬁnition in that

´

N may not be continuous

for a piecewise C

1

surface, and certainly may fail to exist at various points. We have escaped

trouble in our examples thanks to an intuitive concept of “outer” or “inner.” The full resolution

of this ambiguity will be given in Section C and then again in Section F in our discussion of

Stokes’ theorem. But for the moment we are content to live with this ambiguity.

B. The boundary of a surface

This is the second feature of a surface that we need to understand. Consider a surface

M ⊂ R

3

and assume it’s a closed set. We want to deﬁne its boundary.

To do this we cannot revert to the deﬁnition of bdM given in Section 10A. For according

to that deﬁnition bdM = M. The reason is that M has no interior points, since interior points

have to do with open balls in R

3

.

Nevertheless it is clear that we would like some concept of the boundary of M. It’s easy to

make it precise. We say that p ∈ M is an interior point if there is a “disk-like’ neighborhood

of p which lies in M. Otherwise, p is a boundary point of M.

Another way of thinking of this concept is to imagine M as being the “universe,” and

dwellers in this universe have their own two-dimensional idea of interior point, and don’t even

Stokes’ theorem 5

know about the ambient R

3

. In other words, they think of intrinsic interior points of M.

NOTATION. The set of boundary points of M will be denoted

∂M.

Here’s a typical sketch:

M

M

In another typical situation we’ll have a sort of edge in M where

´

N is undeﬁned:

The points in this edge are not in ∂M, as they have a “disk-like” neighborhood in M, even

though the disk is bent.

EXAMPLES from the preceding section:

The surface of a cube. ∂M = ∅.

The boundary of a hemiball. ∂M = ∅.

A cylindrical can. Here ∂M consists of the top circle,

x

2

+ y

2

= a

2

, z = h.

6 Chapter 13

Another typical example is a cylinder “open at both ends”:

M

x

2

+ y

2

= a

2

,

0 ≤ z ≤ h.

Here ∂M consists of the two circles

x

2

+ y

2

= a

2

, z = 0 and x

2

+ y

2

= a

2

, z = h.

DEFINITION. A surface M is closed if ∂M = ∅.

Again, this deﬁnition conﬂicts with our use of the same word in Section 10A. Unfortunately,

this terminology has become standard. So you must be careful if someone utters the phrase

“closed surface.” Be sure you understand what is meant. A better term would be “surface

without boundary.”

Typically, if M is equal to bdD for some set D ⊂ R

3

, then M is a closed surface.

Notice that ∂M may consist of several disjoint arcs.

C. Inherited orientation

The two concepts of orientation and boundary have an important relationship. Namely,

suppose the oriented surface M has a nonempty boundary ∂M. Consider an arc belonging

to ∂M. Then we can assign a direction to ∂M by saying that if we walk along ∂M with our

heads “up,” then we see M at our left sides. Of course “up” refers to the chosen unit normal

vector

´

N.

We describe this by saying that ∂M inherits its orientation from M.

Notice that this is in complete agreement with our statement of Green’s theorem. There

bdR is given the direction which keeps R on the left, if we suppose a third z direction pointing

“up” from the x −y plane with its usual orientation.

Stokes’ theorem 7

EXAMPLE. Hemisphere.

N

EXAMPLE. Cylinder open at both ends.

This example is extremely typical, and is quite easy, but very important to understand!

It goes without saying that if ∂M = ∅, then we need not worry about an inherited orien-

tation.

Now we can easily explain the orientation of piecewise C

1

surfaces. Each smooth piece

needs to be oriented in such a way that the induced orientations given to any arc in ∂M which

is in the boundary of each piece are opposite. A nice example is a cylinder with a top disk:

N

N

8 Chapter 13

The displayed unit normal vectors

´

N give opposite orientations to the circular arc the two

parts of the surface have in common.

D. The basic calculation

Now we are ready to go! We start with a small piece of an oriented surface, and we actually

assume it’s of class C

2

. Call it M ⊂ R

3

. This surface is to be considered to be parameterized

in the usual way. Let us call the parameters u, v, so we have a parameter mapping Φ from a

region R ⊂ R

2

onto M.

R

M

R

M

We know that the partial derivatives Φ

u

, Φ

v

give a basis for the tangent space to M at

any point, and thus their cross product Φ

u

× Φ

v

is a nonzero vector normal to M. As M is

oriented we are given a unit normal vector

´

N at each point of M. We now want to make

sure that Φ

u

× Φ

v

is a positive scalar multiple of M. This can be achieved by the device of

interchanging u and v if necessary. Thus we have

´

N =

Φ

u

×Φ

v

J

,

where the denominator J is simply the norm

J = Φ

u

×Φ

v

.

From Section 11B we know that area integration on M comes from J as the Jacobian

factor:

darea = Jdudv.

Next, we make sure that we represent the parameter u − v space as a right-handed coor-

dinate system, as shown in the ﬁgure. Then we make the all-important observation that the

Stokes’ theorem 9

positive direction of bdR in the parameter space corresponds to the inherited orientation of

∂M in R

3

. You should check this for yourself:

PROBLEM 13–3. Prove the statement just made about the orientation.

Now we are ready for the computation. The goal we have in mind is to rewrite a general

line integral of the form

∂M

F • dx

as a surface integral of the form

M

(???)darea.

(We don’t yet know what the integrand will be.) In doing this we have to integrate along ∂M

in the direction inherited from

´

N. We need a parameter for describing ∂M, and we’ll just use

a convenient parameter t for describing bdR. That is, bdR may be described by functions

u = u(t), v = v(t), and then ∂M is described by the three coordinates

x = x(u(t), v(t)),

y = y(u(t), v(t)),

z = z(u(t), v(t)).

Here we abuse the notation by thinking of Φ as described by three functions x = x(u, v),

y = y(u, v), z = z(u, v). But we’ll completely suppress t in our calculation. You’ll never need

to see it.

Please notice that we write our surface integrals in the present chapter with two integral

signs, as the manifolds we consider are always two-dimensional ones (and their one-dimensional

boundaries).

Finally, we shall calculate just the particular special case of our line integrals,

∂M

fdx.

In other words, F = (f, 0, 0). We’ll then easily get the other two cases by cycling through the

indices.

10 Chapter 13

Here we go! We have

∂M

fdx

chain rule

=

bdR

f(x

u

du + x

v

dv)

=

bdR

(fx

u

)du + (fx

v

)dv

Green

=

R

[(fx

v

)

u

−(fx

u

)

v

]dudv

=

R

[f

u

x

v

+ fx

vu

−f

v

x

u

−fx

uv

CANCEL

]dudv

=

R

[f

u

x

v

−f

v

x

u

]dudv

chain rule

=

R

[(f

x

x

u

+ f

y

y

u

+ f

z

z

u

)x

v

−(f

x

x

v

CANCEL

+ f

y

y

v

+ f

z

z

v

)x

u

]dudv

=

R

[f

y

(y

u

x

v

−y

v

x

u

) + f

z

(z

u

x

v

−z

v

x

u

)]dudv.

Now we are ready to incorporate the normal vector

´

N into this equation. By deﬁnition of

the cross product we have

´

N = J

−1

Φ

u

×Φ

v

= J

−1

det

¸

ˆı ˆ

ˆ

k

x

u

y

u

z

u

x

v

y

v

z

v

= J

−1

(y

u

z

v

−y

v

z

u

, z

u

x

v

−z

v

x

u

, x

u

y

v

−x

v

y

u

)

= (N

1

, N

2

, N

3

).

Thus we recognize that

∂M

fdx =

R

(−f

y

N

3

+ f

z

N

2

)Jdudv.

Of course, Jdudv = darea. Thus if we write the integrand in the form of a dot product,

we have

∂M

fdx =

M

(0, f

z

, −f

y

) •

´

Ndarea.

Stokes’ theorem 11

This is the end result of our calculation. The parameters have disappeared and everything is

in terms of the surface M.

It is truly wonderful that just knowing the ﬂat Green’s theorem has led to this curved

version by just routine manipulations of the deﬁnitions. Almost no thought was required!

“I cast it into the ﬁre, and there came out this calf ”

−Aaron, Exodus 32

24

E. The basic theorem, curl

We now very quickly extend the result we have just proved by cycling through the coordi-

nates x → y → z → x. Thus we must have

∂M

gdy =

M

(−g

z

, 0, g

x

) •

´

Ndarea

and

∂M

hdz =

M

(h

y

, −h

x

, 0) •

´

Ndarea.

We then add our three basic line integrals to obtain

∂M

fdx + gdy + hdz =

M

(h

y

−g

z

, f

z

−h

x

, g

x

−f

y

) •

´

Ndarea.

We have come to a point where we need to make a fascinating

DEFINITION. Given a vector ﬁeld F = (f, g, h) of class C

1

on R

3

, the curl of F is the

vector ﬁeld

curlF = (h

y

−g

z

, f

z

−h

x

, g

x

−f

y

).

Incidentally, most languages other than English use the word rotation in place of curl,

and write rotF for the vector ﬁeld. On the next page you will ﬁnd examples that have been

photocopied from calculus texts in Russian, German, French, and Italian.

This vector ﬁeld curlF is quite amazing. You should keep in mind how very naturally it

appeared in our calculations. It isn’t something we had to be clever to invent; it simply arose

in the calculations.

There’s a wonderful mnemonic for the curl. Recall the mnemonic for the cross product of

two vectors,

a ×b = det

¸

ˆı ˆ

ˆ

k

a

1

a

2

a

3

b

1

b

2

b

3

,

12 Chapter 13

Stokes’ theorem 13

from Section 7A. If we replace the second row with the “vector”

∇ =

∂

∂x

,

∂

∂y

,

∂

∂z

,

then we obtain

curlF = ∇×F = det

¸

ˆı ˆ

ˆ

k

∂/∂x ∂/∂y ∂/∂z

F

1

F

2

F

3

.

The meaning of this “determinant” is this: expand along the ﬁrst row and regard the entries

of the second row as “operating” on the functions in the third row. For example, the second

component of ∇×F is

−det

∂/∂x ∂/∂z

F

1

F

3

= −(∂/∂x)F

3

+ (∂/∂z)F

1

that is

= −

∂F

3

∂x

+

∂F

1

∂z

.

PROBLEM 13–4. Suppose that the surface M is presented as a graph. Speciﬁcally,

assume M is given as the set

{(x, y, ϕ(x, y)) | (x, y) ∈ R},

where R is a region in the x −y plane and ϕ is a C

2

function. Start from the beginning

and derive Stokes’ theorem for this special case. Just follow the outline given above, using

the “upward” normal

´

N with N

3

> 0. The cases

∂M

fdx and

∂M

gdy

are quite similar, whereas the term

∂M

hdz

requires a slightly diﬀerent treatment.

In terms of curl we can now write Stokes’ theorem in the form

∂M

F • dx =

M

curlF •

´

Ndarea.

14 Chapter 13

This is our basic version of Stokes’ theorem. Now we show how to extend it.

F. Stokes’ theorem

We have proved the result in the preceding section under the restrictive hypothesis that

M is presented in terms of a single parametrization. We now go through the same exercises

we used in Section 12C to extend Green’s theorem.

The fundamental observation is in fact the same we used for Green. If two pieces of M

meet along a seam and Stokes is applied to each, the line integrals along the seam cancel each

other because of our assumption on the orientation of M and the inherited orientation of ∂M.

Here’s a sketch:

M

1

M

2

We then can ﬁnally present our theorem as follows:

STOKES’ THEOREM. Assume M is a piecewise C

2

bounded oriented surface in R

3

whose

boundary ∂M has the inherited orientation. Assume F is a C

1

vector ﬁeld deﬁned on M. Then

∂M

F • dx =

M

curlF •

´

Ndarea.

In summary, Stokes’ theorem may be regarded precisely as a curved version of Green’s

theorem in the plane.

Important special case: In the above theorem, if M is a closed surface (∂M = ∅), then

M

curlF •

´

Ndarea = 0.

Stokes’ theorem 15

EXAMPLE. Let M be the hemisphere x

2

+ y

2

+ z

2

= a

2

, z ≥ 0. Let F = (x, x, y). Then

curlF = (1, 0, 1).

Choose the orientation given by

´

N =

(x,y,z)

a

. Then

M

curlF •

´

Ndarea =

M

x + z

a

darea

symmetry

=

1

a

M

zdarea

sph. coords

=

1

a

2π

0

π/2

0

a cos ϕ a

2

sin ϕdϕdθ

= a

2

· 2π ·

1

2

= πa

2

.

And

∂M

F • dx =

∂M

xdx + xdy + ydz

= 0 +

∂M

xdy + 0

=

2π

0

a cos θ(a cos θdθ) = πa

2

.

Another method that can be used here is to use the larger (closed) surface M

consisting

of the hemisphere M and the disk x

2

+y

2

≤ a

2

, z = 0. The orientation of M

requires that at

points of the disk

´

N = −

ˆ

k. Stokes’ theorem gives

M

curlF •

´

Ndarea = 0.

16 Chapter 13

Thus

M

curlF •

´

Ndarea = −

disk

curlF •

´

Ndarea

= −

disk

(1, 0, 1) • (0, 0, −1)darea

=

disk

darea

= area of disk

= πa

2

.

How nice! We didn’t really have to compute an integral this time.

PROBLEM 13–5. Let ˆ u ∈ R

3

be a unit vector, and let −1 < a < 1 be ﬁxed. Let γ be

the circle deﬁned by

(x, y, z) = 1,

(x, y, z) • ˆ u = a.

Give γ the counterclockwise orientation as seen by a viewer located at the point 10ˆ u.

Compute the line integral

γ

xdy.

PROBLEM 13–6. Let m be a ﬁxed real number, and let γ be the curve of intersection

of the paraboloid z = x

2

+ y

2

and the plane z = mx. Assume the curve has the counter-

clockwise orientation as viewed from (0, 0, r) for large positive r. Compute directly the

line integral

γ

ydz.

Also compute the same line integral using Stokes’ theorem.

(ANSWER: −πm

3

/4.)

PROBLEM 13–7. Repeat the preceding problem but with a plane of the more general

form z = m

1

x + m

2

y.

Stokes’ theorem 17

PROBLEM 13–8. Let M be the surface in R

3

which is the portion of the sphere

x

2

+ y

2

+ z

2

= 1 which lies in the cylinder x

2

+ y

2

≤ y and for which z ≥ 0. Choose the

orientation given by the unit vector (−x, −y, −z). Give ∂M the inherited orientation.

Calculate the six line integrals

∂M

xdy and

∂M

ydx;

∂M

ydz and

∂M

zdy;

∂M

zdx and

∂M

xdz.

(ANSWERS: include the numbers 0, 2/3, π/4.)

PROBLEM 13–9. Let γ be the ellipse which is the intersection of the cylinder

x

2

+ y

2

= 1 and the plane z = ax + by, and give γ the counterclockwise orientation as

viewed from a distant point on the positive z-axis. Calculate the line integral

γ

xyzdz

directly, and also by using Stokes’ theorem.

G. What is curl?

We pause to consider some signiﬁcant examples, and then to give an interpretation of curl

that is completely geometric.

1. CENTRAL SYMMETRIC FIELD. A vector ﬁeld is said to be central if all values of it

point toward a ﬁxed point. We may conveniently take this ﬁxed point to be the origin, so a

vector ﬁeld on R

n

is central if and only if it is given by an expression of the form

F(x) = ϕ(x)x for x ∈ R

n

,

where of course ϕ is a real-valued function. We then say F is also symmetric if ϕ(x) depends

only on x. For the case of R

3

this becomes

F(x, y, z) = g(r)(x, y, z),

18 Chapter 13

where of course r = (x, y, z) is the usual spherical coordinate. Then for example the ﬁrst

component of curlF equals

∂

∂y

(g(r)z) −

∂

∂z

(g(r)y) = g

(r)

yz

r

−g

(r)

zy

r

= 0.

Thus curlF = 0.

2. CONSERVATIVE FIELD. Our deﬁnition from Section 12 states that a vector ﬁeld on R

3

is conservative if there exists a potential function f such that F = ∇f. Assuming that f is of

class C

2

, we conclude that curlF = 0. For instance, the second coordinate of curlF is

∂F

1

∂z

−

∂F

3

∂x

=

∂

∂z

∂f

∂x

−

∂

∂x

∂f

∂z

= 0.

Thus we have the interesting result, that always

curl gradf = 0.

Or in terms of the “del” notation,

∇×∇f = 0.

3. REMARK. Actually, Example 1 is a special case of Example 2, as every central symmetric

ﬁeld is conservative. This is even true for R

n

. For suppose

F(x) = ϕ(r)x

is given on R

n

, and we want to ﬁnd a potential function for F. We would certainly expect

this potential to be spherically symmetric itself, so we look for a function of the form f(r).

We thus want

∇(f(r)) = ϕ(r)x;

that is,

f

(r)

x

r

= ϕ(r)x;

we thus simply need to integrate the equation

f

(r) = rϕ(r)

in order to ﬁnd f.

Stokes’ theorem 19

4. ZERO CURL. We now can explain the strange terminology of Section 12E. There we said

that a vector ﬁeld F on R

n

with

∂F

i

∂x

j

=

∂F

j

∂x

i

has zero curl. In case n = 3 this is exactly the condition that curlF = 0, so by analogy we

say the same for general n. It’s just that for n = 3 we don’t actually have a vector ﬁeld we

call curlF.

5. PLANAR FIELDS. Suppose a vector ﬁeld on R

3

has the special form

F(x, y, z) = (F

1

(x, y), F

2

(x, y), 0),

so that F is parallel to the plane z = 0 and also is independent of z. Then

curlF =

∂F

2

∂x

−

∂F

1

∂y

ˆ

k.

6. EXAMPLES OF PLANAR FIELDS. Our ﬁrst example comes from the famous ∇θ on R

2

,

so that

F =

−y

x

2

+ y

2

,

x

x

2

+ y

2

, 0

.

Then curlF = 0. Two more signiﬁcant examples are

curl(−y, x, 0) = 2

ˆ

k,

curl(y, 0, 0) = −

ˆ

k.

PROBLEM 13–10. Let a ∈ R

3

be an arbitrary ﬁxed vector. Show that

∇×(a ×x) = 2a.

(Here x stands for (x, y, z).)

PROBLEM 13–11. The vector ﬁeld a ×x of the preceding problem is a special case

of a general linear vector ﬁeld. Such a ﬁeld can be written F(x) = Ax, where A is a real

3 ×3 matrix and x is written as a column vector. Show that

∇×F = (a

32

−a

23

, a

13

−a

31

, a

21

−a

12

).

20 Chapter 13

PROBLEM 13–12. Here is a problem from American Mathematical Monthly, Volume 109, Number 7, August-

September, 2002, proposed by Victor Alexandrov, Sobolev Institute of Mathematics, Novosibirsk, Russia:

Let M be a surface contained in the unit sphere in R

3

. Use the outer unit normal

vector

´

N for the sphere. In addition, let ˆ n denote the unit vector at points of ∂M which

is tangent to the unit sphere and is orthogonal to ∂M. (In particular, ˆ n •

´

N = 0.) Prove

that

∂M

ˆ nds + 2

M

´

Ndarea = 0.

Here ds denotes arc length and the integrals with vector integrands are to be interpreted

in the sense of integrating the components and then combining the results into vectors.

(HINT: apply Stokes to the vector ﬁeld F = a ×x for any ﬁxed vector a.)

Stokes’ theorem 21

PROBLEM 13–13. Let C denote the unit circle

{(x, y, 0) | x

2

+ y

2

= 1}

in R

3

. Deﬁne the “function” g on R

3

−C by

g(x, y, z) = arctan

z

x

2

+ y

2

−1

.

Of course, g is not completely well deﬁned.

a. Prove that the vector ﬁeld F = ∇g is a well deﬁned vector ﬁeld on R

3

− C, and

that it is C

∞

.

b. Prove that F is irrotational.

c. Prove that F is not conservative by showing that there is a loop in R

3

− C along

which the line integral of F is not zero.

d. Calculate F explicitly, and show that on the speciﬁc loop γ

y = 0, z

2

= 2x

2

−x

4

(x ≥ 0),

the line integral of F equals

γ

−2xzdx + (x

2

−1)dz.

Calculate this integral directly and show it equals 2π.

GEOMETRY. We now present a geometric interpretation of curl which provides intuition

for the above examples and others. First, we need to understand line integrals in a certain

geometric way. Suppose that F is a vector ﬁeld on R

n

and γ is a curve in R

n

, say γ = γ(t),

a ≤ t ≤ b. Then the dot product

F(γ(t)) • γ

(t)

represents the component of F in the direction tangent to the curve, multiplied by the speed

of the curve. Thus in a very signiﬁcant sense the integral

b

a

F(γ(t)) • γ

(t)dt

22 Chapter 13

represents the total net component of F tangent to the curve γ. In particular, if γ is a closed

curve, then this integral represents the net increase of the tangential component of F around

γ. We thus say that

γ

F • dx = the net circulation of F around γ.

In case F represents a force, then we might think of the circulation as the net work done

by F in going around γ. In case F represents the velocity of a ﬂuid, then it would be the net

ﬂow of the ﬂuid.

In particular, a vector ﬁeld is conservative ⇐⇒ it has zero circulation around every closed

curve.

PROBLEM 13–14. Let F be a vector ﬁeld in R

2

which is given by

F = r

α

(−y, x). (r =

x

2

+ y

2

)

Find the net circulation of F around the counterclockwise circle x

2

+ y

2

= a

2

. For which

value of α is the result independent of the radius a?

Stokes’ theorem thus asserts that in the case of R

3

the net circulation of F around ∂M

may be measured by calculating the surface integral over M of the normal component of the

curl of F (with proper orientation).

This idea may be presented inﬁnitesimally in such a way as to give an entirely diﬀerent

way of deﬁning curl. To see this, suppose F is a vector ﬁeld on R

2

which is deﬁned in some

neighborhood of a ﬁxed point p

0

. We shall derive a formula for curlF(p

0

) without using any

coordinate system and even without using any partial derivatives of any components of F!

In order to know curlF(p

0

), it suﬃces to know the number curlF(p

0

)• ˆ u for any unit vector

ˆ u. This dot product is of course the component of the vector curlF(p

0

) in the direction ˆ u.

Construct the disk D

with center p

0

, radius , orthogonal to ˆ u.

Use the vector ˆ u as the unit normal vector for D

, thus rendering D

**an oriented surface. The
**

bounding circle ∂D

**is of course supplied with the induced orientation: it travels counterclock-
**

Stokes’ theorem 23

wise as viewed from the tip of ˆ u. Then Stokes’ theorem gives

D

curlF • ˆ udarea =

∂D

F • dx.

If is small, the left side of this equation is very close to curlF(p

0

) • ˆ u times the area of D

,

thanks to the continuity of the integrand. Thus if we divide by π

2

and let → 0, we obtain

curlF(p

0

) • ˆ u = lim

→0

1

π

2

∂D

F • dx.

There is a signiﬁcant way to think about the right side of this equation in terms of the idea

of circulation, and this lead to the sentence

curlF(p

0

) • ˆ u = “the counterclockwise circulation per unit area of F at p

0

with respect to ˆ u.”

MORAL. The right sides of these expressions have a deﬁnite meaning independent of any

choice of coordinate system. Thus the same must be true of curlF(p

0

) • ˆ u. As this is true for

every choice of the unit vector ˆ u, we conclude that

the curl of a vector ﬁeld on R

3

is a geometric property of the ﬁeld, depending only

on the choice of the orientation of R

3

.

Notice how very naturally this description of curl has arisen. Natural as it is, it is nevertheless

stunning in view of our initial deﬁnition in terms of the coordinate system,

curl(F

1

ˆı + F

2

ˆ + F

3

ˆ

k) = det

¸

ˆı ˆ

ˆ

k

∂/∂x ∂/∂y ∂/∂z

F

1

F

2

F

3

.

All three rows of the matrix depend on the coordinate system, but the output does not!

Once again we observe the wonderful interplay between algebra and geometry! Stokes’

theorem provides us with the deep geometric signiﬁcance of curl, while the initial deﬁnition

gives us a handy way of computing the curl of vector ﬁeld.

For instance, suppose {

ˆ

φ

1

,

ˆ

φ

2

,

ˆ

φ

3

} is a right-handed orthonormal frame for R

3

and deﬁne

coordinates t

1

, t

2

, t

3

by the formula

(x, y, z) = t

1

ˆ

φ

1

+ t

2

ˆ

φ

2

+ t

3

ˆ

φ

3

.

24 Chapter 13

Then we conclude immediately that

curlF = det

¸

ˆ

φ

1

ˆ

φ

2

ˆ

φ

3

∂/∂t

1

∂/∂t

2

∂/∂t

3

F •

ˆ

φ

1

F •

ˆ

φ

2

F •

ˆ

φ

3

.

No calculations needed!

PROBLEM 13–15. How does this formula change if {

ˆ

φ

1

,

ˆ

φ

2

,

ˆ

φ

3

} is a left-handed

orthonormal frame?

This is a good time to remember that we stressed a similar geometry/algebra connection

back in Section 2H in our discussion of the gradient of a function. We thus arrive at two

signiﬁcant geometric insights into the “diﬀerential operator” ∇ = (∂/∂x, ∂/∂y, ∂/∂z), one

when it acts on functions to produce conservative vector ﬁelds ∇f, and now one when it acts

on vector ﬁelds to produce new vector ﬁelds ∇×F.

You should also notice why curlF is also called the rotation of F and is sometimes written

rotF, as we mentioned in Section E. It all has to do with the geometric description of curl in

terms of the circulation, or rotation, of the vector ﬁeld.

There are some straightforward calculus results that are frequently quite helpful in manip-

ulating curl. We have seen one already: ∇×∇f = 0. Here is a useful product rule:

PROBLEM 13–16. Show that

∇×(fF) = f∇×F +∇f ×F.

In Chapter 15 we shall greatly extend the formula given above for curlF to allow nonorthog-

onal frames, and in fact to allow curvilinear coordinates.

H. Curlometer

Imagine a ﬂuid ﬂowing in R

3

, and imagine the velocity vector at each point x. This gives

some sort of a vector ﬁeld F(x), which we suppose to be independent of time.

The streamlines of this ﬂuid are obtained by solving the system of ordinary diﬀerential

equations for the curves x(t):

dx

dt

= F(x(t)).

Stokes’ theorem 25

Now imagine that we want to have a mechanical device for measuring the curl of F.

Imagine then a propeller free to spin, attached to a movable stick:

We might then situate the propeller at any point and the stick pointing in any direction. The

speed (and direction) of rotation could then presumably describe the component of ∇×F at

that point and in that direction. This is what Stokes’ theorem guarantees.

We now give three illuminating examples of this idea. Each is a planar ﬁeld, so that

F = (F

1

(x, y), F

2

(x, y), 0). We then know of course that

∇×F =

∂F

2

∂x

−

∂F

1

∂y

ˆ

k.

1. SHEAR FLOW. Here F = (ay, 0, 0). Here’s a picture of the streamlines, supposing a > 0.

But the lines with y large are moving faster than those with y small, so we expect that ∇×F

will be nonzero. In fact,

∇×F = −a

ˆ

k.

Thus our propeller experiences a clockwise rotation (supposing a > 0), just as our intuition

expects.

2. CIRCULAR FLOW. Here F = “∇θ” =

−y

x

2

+y

2

,

x

x

2

+y

2

, 0

**. We know that ∇×F = 0. This
**

was mentioned in Section G6. So our propeller would tend to move in counterclockwise circles

around the origin, but with no tendency to rotate as it moves.

On the other hand, consider F = (−y, x, 0). The streamlines are still counterclockwise

circles, but now

∇×F = 2

ˆ

k,

26 Chapter 13

which means that the propeller will spin counterclockwise as it rotates.

streamlines

3. CENTRAL FLOW. Here we take F = f(θ)(x, y, 0), so that the streamlines are rays through

the origin. However, the velocity may vary with θ.

PROBLEM 13–17. Show that in this example

∇×F = −f

(θ)

ˆ

k.

I. Conservative ﬁelds revisited

In Section 12F we considered vector ﬁelds deﬁned on open subsets D ⊂ R

n

which had

“zero curl.” We noticed particularly the simple example in R

2

− {0} described as F = ∇θ.

This ﬁeld is irrotational but not conservative. In case D is simply connected in R

2

, we saw

that irrotational ﬁelds are indeed conservative.

The same is true for simply connected open subsets of R

3

. For instance, suppose that F is

a vector ﬁeld of class C

1

on R

3

−{0} and suppose that ∇×F = 0. Then we can prove that

F is conservative. For consider a suﬃciently “nice” closed curve γ ⊂ R

3

−{0}. Then we can

construct an oriented surface M also contained in R

3

− {0}, with ∂M = γ. Stokes’ theorem

then gives

∂M

F • dx =

M

∇×F •

´

Ndarea

=

M

0 darea

= 0.

Stokes’ theorem 27

This veriﬁes the validity of the second criterion in the theorem of Section 12E, and thus F is

conservative.

The reason for the diﬀerence between R

2

−{0} and R

3

−{0} is clear: in R

3

−{0} there is

room to maneuver to ﬁll in a loop with a surface missing the origin.

PROBLEM 13–18. Show that the same result holds for the open set

R

3

−R ×[0, ∞) ×{0}.

2 For example, for the ellipsoid

Chapter 13

x2 y 2 z 2 + 2 + 2 = 1 a2 b c N=

x y z , , a2 b2 c2 x2 a4

we may take

+

y2 b4

+

z2 c4

.

For the sphere S(0, a) we have in particular either N= (x, y, z) a or N = − (x, y, z) . a

N

REMARK. An orientable surface is also said to be two-sided. The reason for this is that the continuous normal vector N serves to deﬁne a direction of “up” at points of M . Thus at points of M there is a deﬁnite sense of two sides of M , an “up” side and a “down” side. A M¨bius strip for example is one-sided, which may be demonstrated by drawing a curve along o the “equator” of M with a pencil. EXTENSION. Frequently we shall need to analyze a surface M ⊂ R3 which is not actually orientable in the above sense, but is “close enough.” The surface may consist of ﬁnitely many surfaces with the proper orientability. A few examples should suﬃce for a good explanation. The surface of a cube. If the cube is [−1, 1] × [−1, 1] × [−1, 1], then the surface consists of the six squares making up the boundary of the solid cube. A typical face is {1} × [−1, 1] × [−1, 1] = {(1, y, z) | −1 ≤ y ≤ 1, −1 ≤ z ≤ 1}. Now we may choose N at each point except those points on the edges to point (say) outward. Thus for the above face we have N = (1, 0, 0) = ˆ. On the opposite face {−1}×[−1, 1]×[−1, 1] ı we would have N = −ˆ. We thus regard this surface as orientable. ı

N=k

N = -k

0) for 0 < z ≤ h. 0 ≤ z ≤ h. If we choose the inward normal vector. together with the disk x2 + y 2 ≤ a2 .
PROBLEM 13–1. z ≥ 0. h ≤ z ≤ h + a. Consider the surface for which one part is given by x2 + y 2 = a2 . a ˆ N = k on the disk.Stokes’ theorem
3
The boundary of a hemiball. Draw a sketch of it and write down expressions for an “outer” unit normal vector. −y. N= A cylindrical can.
. Then we might choose an “outer” unit normal vector N N = (x. and the other part by x2 + y 2 ≤ a2 . It consists of the union of the x2 + y 2 = a2 . For instance consider the hemiball x2 + y 2 + z 2 ≤ a2 . 0 ≤ z ≤ h. Then the surface we have in mind consists of the hemisphere x2 + y 2 + z 2 = a2 .
x2 + y 2 + (z − h)2 = a2 . z ≥ 0. a ˆ = −k for z = 0. z = 0.y. cylinder and the hemisphere
The following surface is orientable. −z) on the hemisphere. z = 0. then we have (−x.

The reason is that M has no interior points. Another way of thinking of this concept is to imagine M as being the “universe. Each will be orientable as well and we shall be faced with just two choices for N . Then give formulas for the ‘outer” unit normal vector. The boundary of a surface This is the second feature of a surface that we need to understand. since interior points have to do with open balls in R3 .” and dwellers in this universe have their own two-dimensional idea of interior point. and certainly may fail to exist at various points. We want to deﬁne its boundary. But for the moment we are content to live with this ambiguity. Each will be piecewise C 1 and any two points on M can be joined by a piecewise C 1 curve lying in M . Otherwise. We have escaped trouble in our examples thanks to an intuitive concept of “outer” or “inner. We say that p ∈ M is an interior point if there is a “disk-like’ neighborhood of p which lies in M . For according to that deﬁnition bdM = M . Consider a surface M ⊂ R3 and assume it’s a closed set. B. and don’t even
. This leads to one more DEFINITION. An orientable surface M is said to be oriented if a deﬁnite choice has been made of a continuous unit normal vector N for M . consisting of a right circular cone topped oﬀ with a hemisphere. It’s easy to make it precise. Give formulas for an “ice cream cone” surface.
All of the surfaces we shall be considering will be connected.” The full resolution of this ambiguity will be given in Section C and then again in Section F in our discussion of Stokes’ theorem. There is actually a touch of vagueness in this deﬁnition in that N may not be continuous for a piecewise C 1 surface. Nevertheless it is clear that we would like some concept of the boundary of M .4
Chapter 13
PROBLEM 13–2. p is a boundary point of M . To do this we cannot revert to the deﬁnition of bdM given in Section 10A.

∂M = ∅. ∂M = ∅. they think of intrinsic interior points of M .
. Here’s a typical sketch:
M
M
In another typical situation we’ll have a sort of edge in M where N is undeﬁned:
The points in this edge are not in ∂M . The set of boundary points of M will be denoted ∂M. even though the disk is bent. In other words. EXAMPLES from the preceding section: The surface of a cube. NOTATION. as they have a “disk-like” neighborhood in M . x2 + y 2 = a2 . Here ∂M consists of the top circle. z = h. A cylindrical can. The boundary of a hemiball.Stokes’ theorem
5
know about the ambient R3 .

Inherited orientation The two concepts of orientation and boundary have an important relationship. z = 0 and x2 + y 2 = a2 . 0 ≤ z ≤ h. suppose the oriented surface M has a nonempty boundary ∂M . Then we can assign a direction to ∂M by saying that if we walk along ∂M with our heads “up. if M is equal to bdD for some set D ⊂ R3 . DEFINITION. A surface M is closed if ∂M = ∅. Namely. So you must be careful if someone utters the phrase “closed surface.” then we see M at our left sides. Consider an arc belonging to ∂M . Notice that this is in complete agreement with our statement of Green’s theorem. A better term would be “surface without boundary.
We describe this by saying that ∂M inherits its orientation from M .
Chapter 13
M
Again. Unfortunately.6 Another typical example is a cylinder “open at both ends”: x2 + y 2 = a 2 . Notice that ∂M may consist of several disjoint arcs. this deﬁnition conﬂicts with our use of the same word in Section 10A.
. C. this terminology has become standard. if we suppose a third z direction pointing “up” from the x − y plane with its usual orientation. then M is a closed surface. z = h. Of course “up” refers to the chosen unit normal vector N . Here ∂M consists of the two circles x2 + y 2 = a2 .” Typically. There bdR is given the direction which keeps R on the left.” Be sure you understand what is meant.

Cylinder open at both ends. Each smooth piece needs to be oriented in such a way that the induced orientations given to any arc in ∂M which is in the boundary of each piece are opposite. Now we can easily explain the orientation of piecewise C 1 surfaces.Stokes’ theorem
7
EXAMPLE. then we need not worry about an inherited orientation. A nice example is a cylinder with a top disk:
N
N
. Hemisphere. This example is extremely typical. and is quite easy.
N
EXAMPLE. but very important to understand!
It goes without saying that if ∂M = ∅.

as shown in the ﬁgure. As M is oriented we are given a unit normal vector N at each point of M . and thus their cross product Φu × Φv is a nonzero vector normal to M . v.
R R M M
We know that the partial derivatives Φu .8
Chapter 13
The displayed unit normal vectors N give opposite orientations to the circular arc the two parts of the surface have in common. Φv give a basis for the tangent space to M at any point. Call it M ⊂ R3 . The basic calculation Now we are ready to go! We start with a small piece of an oriented surface. Then we make the all-important observation that the
. From Section 11B we know that area integration on M comes from J as the Jacobian factor: darea = J dudv. J
where the denominator J is simply the norm J = Φu × Φv . This surface is to be considered to be parameterized in the usual way. Thus we have N= Φu × Φv . This can be achieved by the device of interchanging u and v if necessary. and we actually assume it’s of class C 2 . Next. D. we make sure that we represent the parameter u − v space as a right-handed coordinate system. so we have a parameter mapping Φ from a region R ⊂ R2 onto M . We now want to make sure that Φu × Φv is a positive scalar multiple of M . Let us call the parameters u.

We’ll then easily get the other two cases by cycling through the indices. z = z(u(t). as the manifolds we consider are always two-dimensional ones (and their one-dimensional boundaries). We need a parameter for describing ∂M . bdR may be described by functions u = u(t). 0). That is.
. 0. y = y(u(t).
∂M
In other words. v). But we’ll completely suppress t in our calculation. Here we abuse the notation by thinking of Φ as described by three functions x = x(u.Stokes’ theorem
9
positive direction of bdR in the parameter space corresponds to the inherited orientation of ∂M in R3 . v(t)).
M
(We don’t yet know what the integrand will be. and then ∂M is described by the three coordinates x = x(u(t). v). v). v(t)). Prove the statement just made about the orientation. You’ll never need to see it. z = z(u. v = v(t). Please notice that we write our surface integrals in the present chapter with two integral signs. v(t)). The goal we have in mind is to rewrite a general line integral of the form F • dx
∂M
as a surface integral of the form (???)darea. y = y(u. f dx. we shall calculate just the particular special case of our line integrals.) In doing this we have to integrate along ∂M in the direction inherited from N . F = (f. and we’ll just use a convenient parameter t for describing bdR. Now we are ready for the computation. Finally. You should check this for yourself: PROBLEM 13–3.

10 Here we go! We have f dx
∂M chain rule
Chapter 13
=
f (xu du + xv dv)
bdR
=
bdR Green
(f xu )du + (f xv )dv [(f xv )u − (f xu )v ]dudv
R
=
=
R
[fu xv + f xvu − fv xu − f xuv ]dudv
CANCEL
=
R chain rule
[fu xv − fv xu ]dudv [(fx xu + fy yu + fz zu )xv − (fx xv + fy yv + fz zv )xu ]dudv
R CANCEL
=
=
R
[fy (yu xv − yv xu ) + fz (zu xv − zv xu )]dudv. xu yv − xv yu ) = (N1 .
∂M M
. −fy ) • N darea.
Of course. we have f dx = (0. Thus we recognize that f dx =
∂M R
(−fy N3 + fz N2 )J dudv. J dudv = darea. N3 ).
Now we are ready to incorporate the normal vector N into this equation. By deﬁnition of the cross product we have N = J −1 Φu × Φv ˆ k ı ˆ ˆ = J −1 det xu yu zu xv yv zv = J −1 (yu zv − yv zu . fz . zu xv − zv xu . Thus if we write the integrand in the form of a dot product. N2 .

and there came out this calf ” −Aaron. and Italian. most languages other than English use the word rotation in place of curl. h) of class C 1 on R3 . the curl of F is the vector ﬁeld curlF = (hy − gz . it simply arose in the calculations. There’s a wonderful mnemonic for the curl. French. It is truly wonderful that just knowing the ﬂat Green’s theorem has led to this curved version by just routine manipulations of the deﬁnitions. Thus we must have gdy =
∂M M
(−gz . and write rotF for the vector ﬁeld. Exodus 3224 E. 0. curl We now very quickly extend the result we have just proved by cycling through the coordinates x → y → z → x.
We have come to a point where we need to make a fascinating DEFINITION. fz − hx . On the next page you will ﬁnd examples that have been photocopied from calculus texts in Russian.Stokes’ theorem
11
This is the end result of our calculation. gx ) • N darea
and hdz =
∂M M
(hy . Almost no thought was required! “I cast it into the ﬁre. It isn’t something we had to be clever to invent.
We then add our three basic line integrals to obtain f dx + gdy + hdz =
∂M M
(hy − gz . fz − hx . This vector ﬁeld curlF is quite amazing. Incidentally. ˆ k ı ˆ ˆ a × b = det a1 a2 a3 . g. −hx . gx − fy ) • N darea. The parameters have disappeared and everything is in terms of the surface M . Recall the mnemonic for the cross product of two vectors. You should keep in mind how very naturally it appeared in our calculations. b1 b2 b3
. Given a vector ﬁeld F = (f. German. 0) • N darea. The basic theorem. gx − fy ).

12
Chapter 13
.

y. whereas the term hdz
∂M
requires a slightly diﬀerent treatment. For example. Speciﬁcally.
. assume M is given as the set {(x. the second component of × F is − det ∂/∂x ∂/∂z F1 F3 = −(∂/∂x)F3 + (∂/∂z)F1
that is
=
−
∂F3 ∂F1 + . The cases f dx and
∂M ∂M
gdy
are quite similar.Stokes’ theorem
13
from Section 7A. y) ∈ R}. using the “upward” normal N with N3 > 0. ∂x ∂z
PROBLEM 13–4. where R is a region in the x − y plane and ϕ is a C 2 function. ∂x ∂y ∂z . F1 F2 F3
The meaning of this “determinant” is this: expand along the ﬁrst row and regard the entries of the second row as “operating” on the functions in the third row. In terms of curl we can now write Stokes’ theorem in the form F • dx =
∂M M
curlF • N darea. y)) | (x. ϕ(x. If we replace the second row with the “vector” = then we obtain curlF = ∂ ∂ ∂ . Suppose that the surface M is presented as a graph.
ˆ ˆ ı ˆ k × F = det ∂/∂x ∂/∂y ∂/∂z . . Start from the beginning and derive Stokes’ theorem for this special case. Just follow the outline given above.

If two pieces of M meet along a seam and Stokes is applied to each.14 This is our basic version of Stokes’ theorem. if M is a closed surface (∂M = ∅). then curlF • N darea = 0.
Chapter 13
F. Stokes’ theorem may be regarded precisely as a curved version of Green’s theorem in the plane. Assume F is a C 1 vector ﬁeld deﬁned on M . Then F • dx =
∂M M
curlF • N darea. Now we show how to extend it. Important special case: In the above theorem. the line integrals along the seam cancel each other because of our assumption on the orientation of M and the inherited orientation of ∂M . The fundamental observation is in fact the same we used for Green. Stokes’ theorem We have proved the result in the preceding section under the restrictive hypothesis that M is presented in terms of a single parametrization. Here’s a sketch:
M2 M1
We then can ﬁnally present our theorem as follows: STOKES’ THEOREM.
In summary.
M
. Assume M is a piecewise C 2 bounded oriented surface in R3 whose boundary ∂M has the inherited orientation. We now go through the same exercises we used in Section 12C to extend Green’s theorem.

a
Then x+z darea a
curlF • N darea =
M M symmetry
=
1 a
M
zdarea
2π 0 0 π/2
sph. y).z) . z = 0. z ≥ 0. Stokes’ theorem gives
curlF • N darea = 0.Stokes’ theorem
15
EXAMPLE.y. Choose the orientation given by N =
(x.
Another method that can be used here is to use the larger (closed) surface M consisting of the hemisphere M and the disk x2 + y 2 ≤ a2 . Then curlF = (1. coords
=
1 a
a cos ϕ a2 sin ϕdϕdθ
1 = a2 · 2π · = πa2 . Let M be the hemisphere x2 + y 2 + z 2 = a2 . 2 And
F • dx =
∂M ∂M
xdx + xdy + ydz xdy + 0
∂M 2π
= 0+ =
0
a cos θ(a cos θdθ) = πa2 . Let F = (x.
M
. x. 1). 0. The orientation of M requires that at ˆ points of the disk N = −k.

1) • (0. Let m be a ﬁxed real number. u Compute the line integral xdy. 0. r) for large positive r.)
PROBLEM 13–7. y. and let γ be the curve of intersection of the paraboloid z = x2 + y 2 and the plane z = mx. z) = 1. Compute directly the line integral ydz. 0.
γ
PROBLEM 13–6. (x. −1)darea
disk
= − =
disk
darea
= area of disk = πa2 . How nice! We didn’t really have to compute an integral this time. z) • u = a. y. PROBLEM 13–5.16 Thus curlF • N darea = −
M disk
Chapter 13
curlF • N darea (1. and let −1 < a < 1 be ﬁxed. ˆ Give γ the counterclockwise orientation as seen by a viewer located at the point 10ˆ.
(ANSWER: −πm3 /4. Let u ∈ R3 be a unit vector.
γ
Also compute the same line integral using Stokes’ theorem. Assume the curve has the counterclockwise orientation as viewed from (0. Let γ be ˆ the circle deﬁned by (x. 0.
. Repeat the preceding problem but with a plane of the more general form z = m1 x + m2 y.

y. Calculate the six line integrals xdy
∂M
and
∂M
ydx. Let γ be the ellipse which is the intersection of the cylinder x2 + y 2 = 1 and the plane z = ax + by.
∂M
PROBLEM 13–9.
G. CENTRAL SYMMETRIC FIELD. and give γ the counterclockwise orientation as viewed from a distant point on the positive z-axis. y. Choose the orientation given by the unit vector (−x. z). 2/3. −z). A vector ﬁeld is said to be central if all values of it point toward a ﬁxed point. π/4.
∂M
ydz
∂M
and
zdx and
∂M
(ANSWERS: include the numbers 0. Let M be the surface in R3 which is the portion of the sphere x2 + y 2 + z 2 = 1 which lies in the cylinder x2 + y 2 ≤ y and for which z ≥ 0. For the case of R3 this becomes F (x. so a vector ﬁeld on Rn is central if and only if it is given by an expression of the form F (x) = ϕ(x)x for x ∈ Rn . and then to give an interpretation of curl that is completely geometric. zdy.
. −y. and also by using Stokes’ theorem. Calculate the line integral xyzdz
γ
directly. What is curl? We pause to consider some signiﬁcant examples. Give ∂M the inherited orientation.)
xdz. where of course ϕ is a real-valued function. z) = g(r)(x. 1. We then say F is also symmetric if ϕ(x) depends only on x .Stokes’ theorem
17
PROBLEM 13–8. We may conveniently take this ﬁxed point to be the origin.

× f = 0. the second coordinate of curlF is ∂F1 ∂F3 ∂ ∂f ∂ ∂f − = − ∂z ∂x ∂z ∂x ∂x ∂z = 0. that always curl gradf = 0. Assuming that f is of class C 2 . so we look for a function of the form f (r). x = ϕ(r)x. For suppose F (x) = ϕ(r)x is given on Rn . Example 1 is a special case of Example 2. We thus want (f (r)) = ϕ(r)x. that is. REMARK. as every central symmetric ﬁeld is conservative. r we thus simply need to integrate the equation f (r) f (r) = rϕ(r) in order to ﬁnd f . Thus curlF = 0. 3. For instance. 2. Then for example the ﬁrst component of curlF equals ∂ yz zy ∂ (g(r)z) − (g(r)y) = g (r) − g (r) ∂y ∂z r r = 0. z) is the usual spherical coordinate. Our deﬁnition from Section 12 states that a vector ﬁeld on R3 is conservative if there exists a potential function f such that F = f . and we want to ﬁnd a potential function for F .
. we conclude that curlF = 0. Or in terms of the “del” notation. y. This is even true for Rn .18
Chapter 13
where of course r = (x. We would certainly expect this potential to be spherically symmetric itself. Actually. Thus we have the interesting result. CONSERVATIVE FIELD.

so that F is parallel to the plane z = 0 and also is independent of z. z) = (F1 (x. θ on R2 . Our ﬁrst example comes from the famous so that −y x F = . In case n = 3 this is exactly the condition that curlF = 0. so by analogy we say the same for general n.
6. y. PLANAR FIELDS.)
PROBLEM 13–11. 0) = 2k. Such a ﬁeld can be written F (x) = Ax. Show that × F = (a32 − a23 .0 . 5. The vector ﬁeld a × x of the preceding problem is a special case of a general linear vector ﬁeld. F2 (x. x. 0. y). y. ˆ curl(y. a13 − a31 .Stokes’ theorem
19
4. 2 .
. PROBLEM 13–10. EXAMPLES OF PLANAR FIELDS. Two more signiﬁcant examples are ˆ curl(−y. Then curlF = ∂F2 ∂F1 − ∂x ∂y ˆ k. 0). a21 − a12 ). where A is a real 3 × 3 matrix and x is written as a column vector. z). ZERO CURL. It’s just that for n = 3 we don’t actually have a vector ﬁeld we call curlF . y). x2 + y 2 x + y 2 Then curlF = 0. 0) = −k. Suppose a vector ﬁeld on R3 has the special form F (x. Let a ∈ R3 be an arbitrary ﬁxed vector. There we said that a vector ﬁeld F on Rn with ∂Fi ∂Fj = ∂xj ∂xi has zero curl. (Here x stands for (x. We now can explain the strange terminology of Section 12E. Show that × (a × x) = 2a.

Use the outer unit normal vector N for the sphere. n • N = 0. Volume 109. Sobolev Institute of Mathematics. In addition. Russia: 3
Here ds denotes arc length and the integrals with vector integrands are to be interpreted in the sense of integrating the components and then combining the results into vectors.) Prove ˆ that nds + 2 ˆ N darea = 0. Number 7.
Chapter 13
Here is a problem from American Mathematical Monthly. 2002. (In particular. let n denote the unit vector at points of ∂M which ˆ is tangent to the unit sphere and is orthogonal to ∂M . August-
Let M be a surface contained in the unit sphere in R . proposed by Victor Alexandrov. (HINT: apply Stokes to the vector ﬁeld F = a × x for any ﬁxed vector a.
∂M M
September.)
. Novosibirsk.20 PROBLEM 13–12.

Calculate this integral directly and show it equals 2π. z) = arctan Of course. 0) | x2 + y 2 = 1} in R3 . say γ = γ(t). Let C denote the unit circle {(x. Calculate F explicitly. Suppose that F is a vector ﬁeld on Rn and γ is a curve in Rn . Thus in a very signiﬁcant sense the integral
b
F (γ(t)) • γ (t)dt
a
. y.Stokes’ theorem
21
PROBLEM 13–13. Deﬁne the “function” g on R3 − C by g(x. c. b. g is not completely well deﬁned. First.
γ
x2
z . d. a ≤ t ≤ b. we need to understand line integrals in a certain geometric way. and show that on the speciﬁc loop γ y = 0. a. and
(x ≥ 0). + y2 − 1
g is a well deﬁned vector ﬁeld on R3 − C. z 2 = 2x2 − x4 the line integral of F equals −2xzdx + (x2 − 1)dz. y. Prove that F is not conservative by showing that there is a loop in R3 − C along which the line integral of F is not zero. We now present a geometric interpretation of curl which provides intuition for the above examples and others. Prove that the vector ﬁeld F = that it is C ∞ . multiplied by the speed of the curve. Prove that F is irrotational. GEOMETRY. Then the dot product F (γ(t)) • γ (t) represents the component of F in the direction tangent to the curve.

22
Chapter 13
represents the total net component of F tangent to the curve γ. In particular. (r = x2 + y 2 )
Find the net circulation of F around the counterclockwise circle x2 + y 2 = a2 . a vector ﬁeld is conservative ⇐⇒ it has zero circulation around every closed curve. Let F be a vector ﬁeld in R2 which is given by F = rα (−y. To see this. then it would be the net ﬂow of the ﬂuid. ˆ ˆ Construct the disk D with center p0 . In case F represents the velocity of a ﬂuid. We shall derive a formula for curlF (p0 ) without using any coordinate system and even without using any partial derivatives of any components of F ! In order to know curlF (p0 ). if γ is a closed curve. PROBLEM 13–14. it suﬃces to know the number curlF (p0 ) • u for any unit vector ˆ u. thus rendering D an oriented surface. suppose F is a vector ﬁeld on R2 which is deﬁned in some neighborhood of a ﬁxed point p0 . The ˆ bounding circle ∂D is of course supplied with the induced orientation: it travels counterclock-
. then we might think of the circulation as the net work done by F in going around γ.
γ
In case F represents a force. ˆ
Use the vector u as the unit normal vector for D . orthogonal to u. radius . then this integral represents the net increase of the tangential component of F around γ. We thus say that F • dx = the net circulation of F around γ. For which value of α is the result independent of the radius a? Stokes’ theorem thus asserts that in the case of R3 the net circulation of F around ∂M may be measured by calculating the surface integral over M of the normal component of the curl of F (with proper orientation). This dot product is of course the component of the vector curlF (p0 ) in the direction u. x). In particular. This idea may be presented inﬁnitesimally in such a way as to give an entirely diﬀerent way of deﬁning curl.

φ2 . ˆ ˆ ˆ For instance.
. t3 by the formula ˆ ˆ ˆ (x. it is nevertheless stunning in view of our initial deﬁnition in terms of the coordinate system.
If is small. As this is true for ˆ every choice of the unit vector u. The right sides of these expressions have a deﬁnite meaning independent of any choice of coordinate system. Natural as it is. t2 . Thus the same must be true of curlF (p0 ) • u. Thus if we divide by π 2 and let → 0. Notice how very naturally this description of curl has arisen. we conclude that ˆ the curl of a vector ﬁeld on R3 is a geometric property of the ﬁeld. the left side of this equation is very close to curlF (p0 ) • u times the area of D . but the output does not! Once again we observe the wonderful interplay between algebra and geometry! Stokes’ theorem provides us with the deep geometric signiﬁcance of curl. while the initial deﬁnition gives us a handy way of computing the curl of vector ﬁeld. Then Stokes’ theorem gives ˆ curlF • udarea = ˆ
D ∂D
F • dx. suppose {φ1 . φ3 } is a right-handed orthonormal frame for R3 and deﬁne coordinates t1 . we obtain curlF (p0 ) • u = lim ˆ 1 →0 π 2
∂D
F • dx. depending only on the choice of the orientation of R3 .” ˆ ˆ MORAL. ı ˆ F1 F2 F3 All three rows of the matrix depend on the coordinate system. ˆ thanks to the continuity of the integrand. and this lead to the sentence curlF (p0 ) • u = “the counterclockwise circulation per unit area of F at p0 with respect to u.
There is a signiﬁcant way to think about the right side of this equation in terms of the idea of circulation. ˆ ˆ ı ˆ k ˆ curl(F1ˆ + F2 + F3 k) = det ∂/∂x ∂/∂y ∂/∂z . z) = t1 φ1 + t2 φ2 + t3 φ3 . y.Stokes’ theorem
23
wise as viewed from the tip of u.

There are some straightforward calculus results that are frequently quite helpful in manipulating curl. ∂/∂y. φ2 .24 Then we conclude immediately that ˆ ˆ ˆ φ1 φ2 φ3 curlF = det ∂/∂t1 ∂/∂t2 ∂/∂t3 . and in fact to allow curvilinear coordinates.
H. ˆ ˆ ˆ F • φ1 F • φ2 F • φ3 No calculations needed! PROBLEM 13–15. Curlometer Imagine a ﬂuid ﬂowing in R3 . φ3 } is a left-handed
This is a good time to remember that we stressed a similar geometry/algebra connection back in Section 2H in our discussion of the gradient of a function.
In Chapter 15 we shall greatly extend the formula given above for curlF to allow nonorthogonal frames. It all has to do with the geometric description of curl in terms of the circulation. and imagine the velocity vector at each point x. orthonormal frame?
Chapter 13
ˆ ˆ ˆ How does this formula change if {φ1 . one when it acts on functions to produce conservative vector ﬁelds f . The streamlines of this ﬂuid are obtained by solving the system of ordinary diﬀerential equations for the curves x(t): dx = F (x(t)). We thus arrive at two signiﬁcant geometric insights into the “diﬀerential operator” = (∂/∂x. Show that × (f F ) = f ×F + f × F. of the vector ﬁeld. You should also notice why curlF is also called the rotation of F and is sometimes written rotF . and now one when it acts on vector ﬁelds to produce new vector ﬁelds × F . as we mentioned in Section E. ∂/∂z). or rotation. dt
. We have seen one already: × f = 0. which we suppose to be independent of time. This gives some sort of a vector ﬁeld F (x). Here is a useful product rule: PROBLEM 13–16.

CIRCULAR FLOW.
1. 0 . so we expect that will be nonzero. F2 (x. The speed (and direction) of rotation could then presumably describe the component of × F at that point and in that direction. Here F = “ θ” = x2−y 2 . just as our intuition expects. y). ˆ × F = −ak.
.
x 2. attached to a movable stick:
We might then situate the propeller at any point and the stick pointing in any direction. This is what Stokes’ theorem guarantees. In fact. x. but now ˆ × F = 2k.Stokes’ theorem
25
Now imagine that we want to have a mechanical device for measuring the curl of F . We then know of course that ×F = ∂F2 ∂F1 − ∂x ∂y ˆ k. Here’s a picture of the streamlines. supposing a > 0. Each is a planar ﬁeld. The streamlines are still counterclockwise circles. 0. 0). so that F = (F1 (x.
×F
Thus our propeller experiences a clockwise rotation (supposing a > 0). SHEAR FLOW. We know that × F = 0. 0). but with no tendency to rotate as it moves. y). Imagine then a propeller free to spin. On the other hand. So our propeller would tend to move in counterclockwise circles around the origin. We now give three illuminating examples of this idea. 0). Here F = (ay. This +y was mentioned in Section G6.
But the lines with y large are moving faster than those with y small. consider F = (−y. x2 +y2 .

Then we can prove that F is conservative. However. Then we can construct an oriented surface M also contained in R3 − {0}. suppose that F is a vector ﬁeld of class C 1 on R3 − {0} and suppose that × F = 0.” We noticed particularly the simple example in R2 − {0} described as F = θ. 0). Stokes’ theorem then gives F • dx =
∂M M
× F • N darea 0 darea
M
= = 0. Here we take F = f (θ)(x. For consider a suﬃciently “nice” closed curve γ ⊂ R3 − {0}. CENTRAL FLOW.
I. with ∂M = γ. the velocity may vary with θ. In case D is simply connected in R2 .
. For instance. PROBLEM 13–17.26 which means that the propeller will spin counterclockwise as it rotates. y. we saw that irrotational ﬁelds are indeed conservative. Show that in this example ˆ × F = −f (θ)k. The same is true for simply connected open subsets of R3 . Conservative ﬁelds revisited In Section 12F we considered vector ﬁelds deﬁned on open subsets D ⊂ Rn which had “zero curl. This ﬁeld is irrotational but not conservative. so that the streamlines are rays through the origin.
Chapter 13
streamlines
3.

. ∞) × {0}. and thus F is conservative. The reason for the diﬀerence between R2 − {0} and R3 − {0} is clear: in R3 − {0} there is room to maneuver to ﬁll in a loop with a surface missing the origin.Stokes’ theorem
27
This veriﬁes the validity of the second criterion in the theorem of Section 12E. Show that the same result holds for the open set R3 − R × [0. PROBLEM 13–18.