Calculus III (part 4): Vector Calculus (by Evan Dummit, 2021, v. 4.

01)

Contents
4 Vector Calculus 1

4.1 Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

4.2 Surfaces and Surface Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

4.2.1 Parametric Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

4.2.2 Surface Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

4.3 Vector Fields, Work, Circulation, Flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4.3.1 Circulation and Work Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4.3.2 Flux Across a Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4.3.3 Flux Across a Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4.4 Conservative Vector Fields, Path-Independence, and Potential Functions . . . . . . . . . . . . . . . . 17

4.5 Green's Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4.6 Stokes's Theorem and Gauss's Divergence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

4.6.1 Stokes's Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.6.2 Gauss's Divergence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4.7 Applications of Vector Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

4.7.1 Newton's Law of Gravitation and Kepler's Laws . . . . . . . . . . . . . . . . . . . . . . . . . 27

4.7.2 The Heat and Wave Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4.7.3 Numerical Methods in Modeling Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4.7.4 Maxwell's Equations and Electromagnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4 Vector Calculus
Our motivating problem for multivariable integration was to generalize the idea of integration to more complicated
regions in space, or (more succinctly) to integrate a function over a region. We might also ask whether there is a
simple way to integrate a function over an arbitrary curve in the plane or in space, and whether there is a way to
integrate a function over an arbitrary surface in space. The answer (as it always has been to this point) is yes: the
generalization of single-variable integration to arbitrary curves is called a line integral, and the generalization of
double integration to arbitrary surfaces is called a surface integral.

After introducing line and surface integrals, we will then discuss vector elds (which are vector-valued functions in
2-space and 3-space) which provide a useful model for the ow of a uid through space. The principal applications
of line and surface integrals are to the calculation of the work done by a vector eld on a particle traveling through
space, the ux of a vector eld across a curve or through a surface, and the circulation of a vector eld along a
curve.

Finally, we discuss several generalizations of the Fundamental Theorem of Calculus: the Fundamental Theorem
of Calculus for line integrals, Green's Theorem, Gauss's Divergence Theorem, and Stokes's Theorem. Collectively,
these theorems unify all of the dierent notions of integration, as they each relate the integral of a function on a
region to the integral of an antiderivative of the function on the region's boundary.

1
4.1 Line Integrals

• The motivating problem for our discussion of line integrals is: given a parametric curve r(t) = hx(t), y(t)i
and a function f (x, y), if we build a surface along the curve with height given by the function z = f (x, y),
how can we calculate the area of this surface? (This is a natural generalization of our typical single-variable
integration problem, in which we build the surface inside a plane, thus making it the area under a curve.)

3
◦ Here is an example (for visualization), with r(t) = t2 , t cos(2πt) , f (x, y) = t2 + 1, for 0≤t≤ :
2

◦ Another closely related question is: given a parametric curve r(t) = hx(t), y(t), z(t)i and a function
f (x, y, z), how can we calculate the average value of f (x, y, z) on the curve?

◦ A third question: given a thin wire shaped along some curve r(t) = hx(t), y(t)i with variable density
δ(x, y), what is the wire's mass, and what are its moments about the coordinate axes?

• As with all other types of integrals we have examined so far, we use Riemann sums to give the formal denition
of the line integral of a function f (x, y) on a plane curve C. (Also as before, we will use the formal denition
as infrequently as possible!)

◦ The idea is to approximate the curve with straight line segments, sum (over all the segments) the function
value times the length of the segment, and then take the limit as the segment lengths approach zero.

◦ Denition: For a curve C , a partition of Cpinto n pieces is a list of points (x0 , y0 ), ... , (xn , yn ) on C ,
with the nth segment having length ∆si = (∆xi )2 + (∆yi )2 . The norm of the partition P is the largest
number among all of the segment lengths in P .

◦ Denition: For f (x, y) a continuous function and P a partition of the curve C , we dene the Riemann sum
n
X
of f (x, y) on D corresponding to P to be RSP (f ) = f (xk , yk ) ∆sk .
k=1
ˆ
◦ Denition: For a function f (x, y), we dene the line integral of f on the curve C, denoted f (x, y) ds,
C
to be the value of L such that, for every  > 0, there exists a δ>0 (depending on ) such that for every
partition P with norm(P ) < δ , we have |RSP (f ) − L| < .
◦ Remark: It can be proven (with signicant eort) that, if f (x, y) is continuous and the curve C is smooth,
then a value of L satisfying the hypotheses actually does exist.

◦ Remark: The dierential ds in the denition of the line integral is the dierential of arclength, which
we discussed earlier in our study of vector-valued functions.

• In exactly the same way, we can use Riemann sums to give a formal denition of the line integral along a
curve C in 3-space. (We simply add the appropriate z -terms to all the denitions.)

• Like with the other types of integrals, line integrals have a number of formal properties which can be deduced
from the Riemann sum denition. Specically, for D an arbitrary constant and f (x, y) and g(x, y) continuous
functions, the following properties hold:

´
◦ Integral of constant: D ds = D · Arclength(C).
´ C
´
◦ Constant multiple of a function: C D f (x, y) ds = D · C f (x, y) ds.
´ ´ ´
◦ Addition of functions: C f (x, y) ds + C g(x, y) ds = C [f (x, y) + g(x, y)] ds.

2
 ´ ´ ´
◦ Subtraction of functions: f (x, y) ds − C g(x, y) ds = C [f (x, y) − g(x, y)] ds.
C
´
◦ Nonnegativity: if f (x, y) ≥ 0, then C f (x, y) ds ≥ 0.
◦ Union: If C1 and C2 are curves such that
´ C2 ´starts where C1 ´ends, and C is the curve obtained by gluing
the curves end-to-end, then
C1
f (x, y) ds + C2 f (x, y) ds = C f (x, y) ds.
◦ Remark: These same properties also all hold for line integrals of a function f (x, y, z) in 3-space.

• The key observation is that we can reduce calculations of line integrals to traditional single integrals:

• Proposition (Line Integrals in the Plane): If the curve C can be parametrized as x = x(t), y = y(t) for
ˆ ˆ b
ds ds p
a ≤ t ≤ b, then f (x, y) ds = f (x(t), y(t)) dt, where = x0 (t)2 + y 0 (t)2 is the derivative of
C a dt dt
arclength.

• Proposition (Line Integrals in 3-Space): If the curve C can be parametrized as x = x(t), y = y(t), z = z(t)
ˆ ˆ b
ds ds p
for a ≤ t ≤ b, then f (x, y, z) ds = f (x(t), y(t), z(t)) dt, where = x0 (t)2 + y 0 (t)2 + z 0 (t)2 is the
C a dt dt
derivative of arclength.

n
X
◦ The proof of both of these results is simply to observe that the Riemann sum f (xk , yk ) ∆sk for the line
k=1
´ n
X ∆sk ´b ds
integral
C
f (x, y) ds is also a Riemann sum f (xk , yk ) ∆tk for the integral
a
f (x(t), y(t)) dt.
∆tk dt
k=1
◦ Equivalently: we have made a substitution in the integral by changing from s-coordinates to t-coordinates,
ds
where the dierential changes using the rule ds = dt.
dt
´
• Thus, to evaluate the line integral of f on the curve C (i.e., the line integral
C
f (x, y, z) ds), follow these steps:

1. Parametrize the curve C as a function of t, as r(t) = hx(t), y(t), z(t)i for a ≤ t ≤ b.

2. Write the function f t: f (x, y, z) = f (x(t), y(t), z(t)).
in terms of

ds p
3. Write the dierential ds = dt = ||v(t)|| dt = x0 (t)2 + y 0 (t)2 + z 0 (t)2 dt in terms of t.
dt
´b p
4. Evaluate the resulting single-variable integral
a
f (x(t), y(t), z(t)) x0 (t)2 + y 0 (t)2 + z 0 (t)2 dt.

• Example: Integrate the function f (x, y, z) = yz − 6x along the curve r(t) = t3 , 6t, 3t2 from t = 0 to t = 1.
2 3 3
p
◦ √
We have f (x, y, z) = yz − 6x = (6t)(3t ) − 6t = 12t , and we also have ds = (3t2 )2 + (6)2 + (6t)2 =
2
9t4 + 36t2 + 36 = 3t + 6.
´1 ´1
◦ The integral is therefore 0 (12t3 )(3t2 + 6)dt = 0 (36t5 + 72t3 ) dt = 24 .

• Example: Integrate the function f (x, y) = x2 + y along the top half of the unit circle x2 + y 2 = 1, starting at
(1, 0) and ending at (−1, 0).

◦ The unit circle is parametrized by r(t) = hcos t, sin ti:

0 ≤ t ≤ π. the range we want is
2 2
p
◦ We have f (x, y) = x + y = cos t + sin t, and we also have ds = (− sin t) + (cos t)2 = 1.
2

´π ´ π 1 + cos 2t
 
 π
◦ The integral is therefore 0 cos2 t + sin t dt = 0 + sin t dt = +2 .
2 2

• To nd the average value of a function on a curve, we simply integrate the function over the curve, and then
divide by the curve's arclength.

• Example: Find the average value of the function f (x, y, z) = x2 +y 2 +z 2 along the line segment from (1, −1, 0)
to (2, 2, 1).

◦ The direction vector for the line is v = h2, 2, 1i − h1, −1, 0i = h1, 3, 1i. Thus, we can parametrize the line
segment as hx, y, zi = h1, −1, 0i + t h1, 3, 1i for 0 ≤ t ≤ 1.

3
 ◦ So the line segment is parametrized explicitly by x = 1 + t, y = −1 + 3t, z = t for 0 ≤ t ≤ 1.
◦ Now we set up the integral: the function is f (x, y, z) = x + y + z = (1 + t) + (−1 + 3t)2 + (t)2 =
2 2 2 2
2
11t − 4t + 2.
ds √ 2 √
◦ Since x0 (t) = 1, y 0 (t) = 3, and z 0 (t) = 1,
= 1 + 32 + 12 = 11.
we also have
dt
√
´1 2
1
√ √

11 3 2 11 11
◦ The integral of f is therefore 0 11t − 4t + 2 11dt = 11 t − 2t + 2t = .
3 t=0 3
´1 ´1√ √
◦ To compute the average value, we divide by the arclength, which is 0 1 ds = 0 11dt = 11.
11
◦ Thus, the average value is .
3

• We also have formulas for the mass and moments of a wire of variable density:

• Center of Mass and Moment Formulas (Thin Wire): Given a 1-dimensional wire of variable density δ(x, y, z)
along a parametric curve C in 3-space:

´
◦ The total mass M is given by M= δ(x, y, z) ds.
´ C
◦ The x-moment Myz is given by Myz = C x δ(x, y, z) ds.
´
◦ The y -moment Mxz is given by Mxz = C y δ(x, y, z) ds.
´
◦ The z -moment Mxy is given by Mxy = C z δ(x, y, z) ds.
 
Myz Mxz Mxy
◦ The center of mass (x̄, ȳ, z̄) has coordinates , , .
M M M
◦ Note: For a wire in 2-space, the formulas are essentially the same (except without the z -coordinate),
though the x-moment is denoted My and the y -moment is denoted Mx .

• Example: Find the total mass, and the center of mass, of a thin wire in the xy -plane having the shape of the
unit circle with variable density δ(x, y) = 2 + x.
ds p
◦ We can parametrize the unit circle with x = cos t, y = sin t, so = (− sin t)2 + (cos t)2 = 1.
dt
´ ´ 2π
◦ The total mass M is M =
C
δ(x, y) ds = 0 (2 + cos t) dt = 2π .
´ ´ 2π
  2π
1 1
◦ The x-moment My is My =
C
x δ(x, y) ds = 0 cos t(2 + cos t) dt = 2 sin t + t + sin(2t) = π.
2 4 t=0
´ ´ 2π
  2π
1
◦ The y -moment Mx is Mx =
C
y δ(x, y) ds = 0
sin t(2 + cos t) dt = −2 cos t − cos(2t) = 0.
4 t=0
   
M y Mx 1
◦ Therefore, the center of mass is , = ,0 .
M M 2

• ˆ
We will also be interested in computing line integrals involving the dierentials dx, dy , and dz rather than

ds: namely, expressions of the form f dx + g dy + h dz .

C

• We evaluate such line integrals by making the appropriate substitutions: if

ˆ C is parametrized by x = x(t),
y = y(t), z = z(t) for a ≤ t ≤ b, then the line integral f dx + g dy + h dz is given by the single-variable
ˆ b  C
dx dy dz
integral f +g +h dt.
a dt dt dt
´
• Example: Find C y dx + z dy + x2 dz , where C is the curve (x, y, z) = (t, t2 , t3 ) ranging from t = 0 to t = 1.

◦ We have x = t, y = t2 , and z = t3 , so that dx = dt, dy = 2t dt, and dz = 3t2 dt.

´1 2  ´1  73
◦ The integral is
0
t · dt + 3t2 · 2t dt + t2 · 3t2 dt = 0 t2 + 6t3 + 3t4 dt = .
30

4
 ´
• Example: Find
C
x dy − y dx, where C is the upper half of the ellipse x2 /9 + y 2 /16 = 1, starting at (3, 0) and
ending at (−3, 0).

◦ This ellipse is parametrized by r(t) = h3 cos t, 4 sin ti: the range we want is 0 ≤ t ≤ π .
◦ We have x = 3 cos t and y = 4 sin t, so that dx = −3 sin t dt and dy = 4 cos t dt.
´π ´π
[3 cos t · (4 cos t dt) − 4 sin t · (−3 sin t dt)] = 0 12 cos2 t + 12 sin2 t dt = 12π

◦ The desired integral is
0
.

4.2 Surfaces and Surface Integrals

• We would now like to consider the problem of computing the integral of a function on a surface in 3-dimensional
space. In a similar way to how we computed line integrals using (single) integrals, we will be able to compute
surface integrals as double integrals.

• There are essentially two ways to describe a surface in 3-space: either as an implicit surface of the form
f (x, y, z) = c, or as a parametric surface r(s, t) = hx(s, t), y(s, t), z(s, t)i for two parameters s and t.
◦ Note that the explicit surfacez = g(x, y) is simply a special case of the general implicit surface, since
g(x, y) − z = 0 f (x, y, z) = c with f (x, y, z) = g(x, y) − z and c = 0.
has the form

◦ In cases where the functions x, y , and z are suciently simple or nice, it can be possible to eliminate
the variables s and t from the system x = x(s, t), y = y(s, t), z = z(s, t), and obtain an equation for the
surface as an implicit surface f (x, y, z) = c.

◦ We will also remark that parametric descriptions of surfaces are often easier to work with than implicit
descriptions. For example, graphing a parametric surface requires only plugging in values for (s, t) and
plotting the resulting points (x, y, z), whereas graphing an implicit surface requires nding solutions to
the implicit equation, which is typically much harder.

• We will describe how to nd parametrizations of some common surfaces, give the denition of a surface
integral, and then show how to compute surface integrals on both parametric and implicit surfaces.

4.2.1 Parametric Surfaces

• If we graph a vector-valued function of two variables r(s, t) = hx(s, t), y(s, t), z(s, t)i as s and t vary, we will
obtain a surface in space (barring something strange happening).

• Example: The surface r(s, t) = hx0 , y0 , z0 i + t hv1 , v2 , v3 i + s hw1 , w2 , w3 i is the plane passing through the point
(x0 , y0 , z0 ) that contains the two vectors v = hv1 , v2 , v3 i and w = hw1 , w2 , w3 i, provided that v and w are not
parallel.

◦ We could also describe the plane as an implicit surface of the form ax+by +cz = d, where ha, b, ci = v×w
is the normal vector to the plane and d = ax0 + by0 + cz0 .
◦ There are many ways to describe a given plane as a parametric surface. For example, both of the
parametrizations r(s, t) = hs, t, 1 − s − ti and r(s, t) = h−3 + s − 2t, 2 + t + 2s, 2 + t − 3si describe the
same plane x + y + z = 1.
• Example: For two positive radius parameters r and R with r < R, the surface dened parametrically by
r(s, t) = hcos(t)[R + r cos(s)], sin(t)[R + r cos(s)], r sin(s)i, for 0 ≤ t ≤ 2π and 0 ≤ s ≤ 2π is a donut-shaped
surface called a torus.

◦ It is the surface obtained by taking a vertical circle of radius r and moving its center along the circle
x2 + y 2 = R 2 in the xy -plane.
◦ Four tori, with respective parameters (r, R) equal to (1, 5), (2, 5), (3, 5), and (4, 5), are plotted below:

5
• Example: The surface dened parametrically by r(s, t) = hcos(s) + cos(t), s + t, sin(s) + sin(t)i, for 0≤t≤
4π and 0 ≤ s ≤ 4π is a helical ribbon:

• In general, it can be a somewhat involved problem to convert a geometric or verbal description of a surface
into a parametrization: it is really more of an art form than a general procedure.

◦ To parametrize parts of cylinders, cones, and spheres, it is almost always a very good idea to consider
whether cylindrical or spherical coordinates can be of assistance.

◦ Using translations and rescalings, we can also parametrize surfaces like ellipsoids.

• There are many dierent ways to parametrize the same surface, and which description is best will depend on
what the parametrization will be used for.

√ p
◦ For example, x = s, y = t, z = s2 + t2 parametrizes the cone z = x2 + y 2 , but so does the
parametrization x = s cos t, y = s sin t, z = s.

◦ If we want to describe the points lying over a rectangular region in the xy -plane, the rst parametrization
is more useful, but if we want to describe the points on the cone up to a specic height in the z -direction,
the second parametrization is more useful.

• Example: Parametrize the portion of the cylinder x2 + y 2 = 4 lying between the planes z = −2 and z = 2.

◦ In cylindrical coordinates, we know that x = r cos θ, y = r sin θ, and z = z.

◦ Since the given cylinder has equation r = 2 in cylindrical coordinates, we see that a parametrization of
the full cylinder is x = 2 cos t, y = 2 sin t, z = s, where 0 ≤ t ≤ 2π but with no restrictions on s. (Here
we think of t as θ and s as z .)
◦ To obtain just the portion with −2 ≤ z ≤ 2 we just restrict the range for s.
◦ Thus the parametrization of the desired portion of the cylinder is x = 2 cos t, y = 2 sin t, z = s, where
0 ≤ t ≤ 2π and −2 ≤ s ≤ 2.

• Example: Parametrize the portion of the cylinder x2 +y 2 = 4 lying between the planes z = y −2 and z = x+4.

◦ Like in the previous example, we take the parametrization of the full cylinder as x = 2 cos t, y = 2 sin t,
z = s, and then restrict the ranges for s and t appropriately. In this case, we want the portion of the
surface where y − 2 ≤ z ≤ x + 4.
◦ It is straightforward to check that the two planes do not intersect inside the cylinder (since y−2 ≤ 0
inside the cylinder, while x + 4 ≥ 2).
◦ So in this case, we take 0 ≤ t ≤ 2π and 2 sin t ≤ s ≤ 2 cos t + 4.

• Example: Parametrize the sphere x2 + y 2 + z 2 = 9.

◦ In spherical coordinates, we know that x = ρ cos(θ) sin(ϕ), y = ρ sin(θ) sin(ϕ), z = ρ cos(ϕ).

6
 ◦ The sphere has equation ρ = 3, so we can immediately see that x = 3 cos(t) sin(s), y = 3 sin(t) sin(s),
z = 3 cos(s), with 0 ≤ t ≤ 2π and 0 ≤ s ≤ π , will parametrize the sphere. (Here, we are thinking of t as
θ and s as ϕ.)

• Example: Parametrize the sphere (x − 2)2 + (y + 1)2 + (z − 6)2 = 4.

◦ It is not so easy to describe this sphere using spherical coordinates directly. However, if we shift the
coordinates to center the sphere at the origin, we can easily write down the parametrization.

◦ By translating back, we can see that x = 2 + 2 cos(t) sin(s), y = −1 + 2 sin(t) sin(s), z = 6 + 2 cos(s),
with 0 ≤ t ≤ 2π and 0 ≤ s ≤ π, will parametrize the sphere.

x2 y2 z2
• Example: Parametrize the ellipsoid + + = 1.
4 9 16
◦ It is again not so easy to write down the parametrization using any of our coordinate systems directly.
However, if we rescale the coordinates by setting x0 = x/2, y 0 = y/3, and z 0 = z/4, then we see
0 2 0 2 0 2
(x ) + (y ) + (z ) = 1, and we can use spherical coordinates to parametrize the coordinates x0 , y 0 , z 0 .
◦ By rescaling back, we can see that x = 2 cos(t) sin(s), y = 3 sin(t) sin(s), z = 4 cos(s), with 0 ≤ t ≤ 2π
and 0 ≤ s ≤ π, will parametrize this ellipsoid.
p
• Example: Parametrize the portion of the cone z = 3 x2 + y 2 that lies below the plane z = 1 + x + y.

◦ In cylindrical, the equations are z = 3r and z = 2 + r cos θ + r sin θ. They are equal when 3r =
2 √
2 + r cos θ + r sin θ, or r= . (Note that sin θ + cos θ ≤ 2, so the denominator is never
3 − cos θ − sin θ
zero.)

◦ The full surface is parametrized by x = s cos(t), y = s sin(t), z = 3s.

2
◦ The portion under the plane corresponds to 0 ≤ s ≤ , with 0 ≤ t ≤ 2π .
3 − cos t − sin t
• If we have a parametrization of a surface, we can use the parametrization to nd the tangent plane to the
surface at a given point.

◦ The key observation is that if the surface S is parametrized by the vector-valued function r(s, t) =
∂r ∂r
hx(s, t), y(s, t), z(s, t)i, then the two partial derivatives rs = and rt = are both tangent to the
∂s ∂t
surface.
∂r ∂r
◦ Therefore, the cross product × will be perpendicular to the tangent plane, and is thus a normal
∂s ∂t
vector for the tangent plane.

• Example: Find an equation for the tangent plane to the surface r(s, t) = s cos(t), s sin(t), s2 when s=1
and t = π/2.

◦ We compute rs (s, t) = hcos t, sin t, 2si and rt (s, t) = h−s sin t, s cos t, 0i.
◦ Thus, we see rs (1, π/2) = h0, 1, 2i, and rt (1, π/2) = h−1, 0, 0i, and so the normal vector to the tangent
plane is n = h0, 1, 2i × h−1, 0, 0i = h0, −2, 1i.
◦ The tangent plane passes through the point on the surface where s = 1 and t = π/2, which is r(1, π/2) =
h0, 1, 1i.
◦ Thus, an equation for the tangent plane is given by 0(x − 0) − 2(y − 1) + 1(z − 1) = 0 or equivalently

−2y + z = −1 .

• Example: Find an equation for the plane tangent to the surface r(s, t) = s2 , 2st, t3 at the point (4, 4, −1).

◦ First, we need to nd the values of s and t at the point (4, 4, −1). If h4, 4, −1i = s2 , 2st, t3 then we see
3
t = −1 so t = −1, and then 2st = 4 gives s = −2.

7
 ◦ Now, we have rs (s, t) = h2s, 2t, 0i and rt (s, t) = 0, 2s, 3t2 , so rs (−2, −1) = h−4, −2, 0i and rt (−2, −1) =
h0, −4, 3i.
◦ Thus, the normal vector to the tangent plane is n = h−4, −2, 0i × h0, −4, 3i = h−6, 12, 16i.
◦ Thus, an equation for the tangent plane is given by −6(x − 4) + 12(y − 4) + 16(z + 1) = 0 or equiva-

lently −6x + 12y + 16z = 8 .

4.2.2 Surface Integrals

• The motivating problem for our discussion of surface integrals is as follows: given a parametric surface
r(s, t) = hx(s, t), y(s, t), z(s, t)i and a function f (x, y, z), we would like to integrate the function on that
surface. Like with line integrals, we have two natural applications: computing the average value of a function
on the surface, and analyzing the physical properties of a thin surface with variable density.

◦ As with all the other types of integrals, the idea is to approximate the surface with small patches, sum
(over all the patches) the function value times the area of the patch, and then take the limit as the patch
sizes approach zero.

◦ Denition: For a parametric surface S, a partition ofS into n pieces is a list of disjoint subregions inside
S, where the k th subregion corresponds to sk ≤ s ≤ s0k , tk ≤ t ≤ t0k , and has area ∆σk . The norm of the
partition P is the largest number among the areas of the rectangles in P .

◦ Denition: For f (x, y, z) a continuous function and P a partition of the surface S, we dene the
n
X
Riemann sum of f (x, y, z) on R corresponding to P to be RSP (f ) = f (r(sk , tk )) ∆σk .
k=1
¨
◦ Denition: For a function f (x, y, z), we dene the surface integral of f on S, denoted f dσ , to be the

S
value of L such that, for every  > 0, there exists a δ>0 (depending on ) such that for every partition
P with norm(P ) < δ , we have |RSP (f ) − L| < .
◦ Remark: It can be proven (with signicant eort) that, if f (x, y, z) is continuous, then a value of L
satisfying the hypotheses actually does exist.

• As with all of the other types of integrals, surface integrals possess some formal properties. For any continuous
f and g and any constant C, we have the following:

˜
◦ Integral of constant: C dσ = C · Area(S).
S
˜ ˜
◦ Constant multiple of a function: C f dσ = C · S f dσ .
˜ ˜ S
˜
◦ Addition of functions: f dσ + S g dσ = S [f + g] dσ .
S
˜ ˜ ˜
◦ Subtraction of functions: f dσ − S g dσ = S [f − g] dσ .
S
˜
◦ Nonnegativity: if f ≥ 0, then f dσ ≥ 0.
S
˜ ˜ ˜
◦ Union: If S1 and S2 don't overlap and have union S , then
S1
f dσ + S2 f dσ = S f dσ .

• We were able to reduce line integral calculations to standard one-variable integrals. We can similarly reduce
calculations of surface integrals to double integrals:

• Proposition (Parametric Surface Integrals): If f (x, y, z) is continuous on the surface S which is parametrized
as r(s, t) = hx(s, t), y(s, t), z(s, t)i, where S is described by a region R in st-coordinates, then the surface
integral of f on S is

¨ ¨
∂r ∂r
f dσ = f (x(s, t), y(s, t), z(s, t)) × dt ds.
S R ∂s ∂t

◦ The key step is to recognize the Riemann sum for the surface integral as the Riemann sum for a particular
double integral.

8
 ∂r ∂r
◦ Ultimately, the dierential of surface area dσ = × dt ds arises from computing the area of
∂s ∂t
∂r
a small patch in st-coordinates: when s changes slightly, the change in r is given by , and when t
∂s
∂r
changes slightly, the change in r is given by .
∂t
◦ These two vectors form a small parallelogram that closely approximates the surface S , so the dierential
∂r ∂r
of surface area dσ is roughly equal to the area of this parallelogram, which is × , times the
∂s ∂t
dierential dt ds.

• We can also calculate surface integrals over implicit surfaces of the form g(x, y, z) = c:

• Proposition (Implicit Surface Integrals): If f (x, y, z) is continuous on the implicit surface S dened by
g(x, y, z) = c, R is the projection of S into the xy -plane, and ∂g/∂z 6= 0 on R, then the surface integral
of f on S is

¨ ¨
||∇g||
f dσ = f (x, y, z) dy dx
S R |∇g · k|

where ∇g is the gradient of g and k = h0, 0, 1i. (Thus, ∇g · k = ∂g/∂z .)

◦ The statement that ∂g/∂z 6= 0 on R is equivalent to saying that the tangent plane to g(x, y, z) = c is
never vertical above R. In particular this implies that the surface never doubles back on itself over the
region R.
◦ Thus for example, we could not use the method directly to compute a surface integral on the entire unit
sphere, because it has a vertical tangent plane above its projection x2 + y 2 ≤ 1 in the xy -plane.
◦ This formula can be derived from the parametric surface integral formula: after some simplication, it
is what one obtains by using the parametrization r(s, t) = hs, t, z(s, t)i, where z(s, t) is dened implicitly
via the relation f (s, t, z(s, t)) = c.

• Using these two results, we can reduce calculations of surface integrals to traditional double integrals: given
a description of the surface S, we can convert it to a double integral using one of two methods:

◦ For a parametric surface given in the form r(s, t) = hx(s, t), y(s, t), z(s, t)i:
∗ Step 1: Find the bounds on s and t that parametrize the desired portion of the surface.
∗ Step 2: Express the function f (x, y, z) to be integrated in terms of (s, t).
∂r ∂r
∗ Step 3: Find the dierential of surface area dσ = × ds dt.
∂s ∂t
˜ ∂r ∂r
∗ Step 4: Write down the integral
S
f (x(s, t), y(s, t), z(s, t)) × ds dt and evaluate.
∂s ∂t
◦ For an implicit surface given in the form g(x, y, z) = c:
∗ Step 1: Sketch the surface, determine the shape of its projection R into the xy -plane, and make sure
that the surface does not cover any part of the projection more than once.
˜ ||∇g||
∗ Step 2: Evaluate the integral
R
f (x, y, z) dy dx, where ∇g is the gradient of g and k =
|∇g · k|
h0, 0, 1i.
∗ Note that the only variables allowed in the integral are x and y, so if the integrand has any z terms
we must use the implicit equation g(x, y, z) = c to get rid of them.

◦ Note that, by swapping z with x or with y , the implicit surface procedure can also be used with a
projection into the xz -plane or the yz -plane.
◦ Also note that for a surface of the form z = f (x, y), we could use either method.

• Example: Integrate the function g(x, y, z) = z over the surface with parametrization r(s, t) = hsin(t), cos(t), s + ti
for 0 ≤ t ≤ 2π and 0 ≤ s ≤ π.

9
 ◦ We have an explicit parametrization of the surface, so we use the parametric formula.

◦ On the surface, we have z =s+t so g(x, y, z) = z = s + t.

i j k
∂r ∂r ∂r ∂r
◦ We have = h0, 0, 1i and = hcos(t), − sin(t), 1i, so × = 0 0 1 = hsin(t), cos(t), 0i.
∂s ∂t ∂s ∂t
cos(t) − sin(t) 1
∂r ∂r
Then × = 1.
∂s ∂t
◦ The integral is therefore given by

ˆ 2π ˆ π ˆ 2π ˆ 2π
s2 π2
     2 
π π π 2π
(s + t) ds dt = + st dt = + πt dt = t + t2 = 3π 3 .
0 0 0 2 s=0 0 2 2 2 t=0

• Example: Integrate the function f (x, y, z) = 8xy over the portion of the plane 2x + y + 2z = 1 with 0 ≤ x ≤ 1,
0 ≤ y ≤ 1.

◦ We use the implicit surface formula, with g(x, y, z) = 2x + y + 2z − 1.

√
◦ We have ∇g = h2, 1, 2i so ||∇g|| = 22
+ 12 + 22 = 3 and |∇g · k| = 2.
´1´1 ´1
◦ The desired integral is therefore
0 0
8xy · (3/2) dy dx = 0 6x dx = 3 .

• Example: Integrate the function f (x, y, z) = xz over the portion of the plane 4x + 2y + z = 1 with 0 ≤ x ≤ 1,
0 ≤ y ≤ 1.

◦ We use the implicit surface formula, with g(x, y, z) = 4x + 2y + z − 1.

√ √
◦ We have ∇g = h4, 2, 1i so ||∇g|| = 42 + 22 + 12 = 21 and |∇g · k| = 1.
◦ Since the function involves z , we must use the implicit relation to eliminate it. In this case, z = 1−4x−2y ,
so f (x, y, z) = xz = x − 4x2 − 2xy .
´1´1 √ ´1 √ 4√
◦ The desired integral is therefore
0 0
(x − 4x2 − 2xy) · 21 dy dx = 0
(−4x2 ) 21 dx = − 21 .
3

• To compute surface area, we can simply integrate the function 1 on the surface, in exactly the same way that
integrating 1 on a plane region gives its area or integrating 1 on a solid region gives its volume.

• Example: Find the area of the portion of the surface z = 2 − x2 − y 2 that lies above the xy -plane.

◦ We can rewrite the equation of the surface implicitly as x2 + y 2 + z − 2 = 0, so we use the implicit
surface formula.
2 2
◦ The projection of the surface into the xy -plane is the region
√ R on which 2 − x − y ≥ 0, which is the
2 2
same as x + y ≤ 2, and this describes the disc of radius 2 centered at the origin. Since this surface
is explicit we do not need to worry about having a vertical tangent plane.
p
◦ ∇g = h2x, 2y, 1i so ||∇g|| = 4x2 + 4y 2 + 1 and |∇g · k| = 1. The desired integral is
We have therefore
˜ p
R
4x2 + 4y 2 + 1 dy dx, since to calculate surface area we simply integrate the function 1.
◦ To evaluate this integral, we change to polar coordinates, since both the region and the function to be
√ √
integrated will become simpler: the region is 0≤r≤ 2, 0 ≤ θ ≤ 2π , and the function is 4r2 + 1.
´ 2π ´ √2 √
◦ Since the area dierential in polar is r dr dθ, we obtain the polar integral
0 0
4r2 + 1 r dr dθ.
◦ To evaluate this new integral, we make (another) substitution u = 4r2 + 1, with du = 8r dr:
ˆ 2π ˆ √
2 ˆ 2π ˆ 9 ˆ 2π ˆ 2π
1√
 
p 1 2 3/2 9 26 13π
4r2 + 1 r dr dθ = u du dθ = u dθ = dθ = .
0 0 0 1 8 0 8 3 u=1 0 12 3

◦ Remark: Alternatively, we could have parametrized this surface using cylindrical coordinates, as
√ x =
s cos(t), y = s sin(t), z = 2 − s2 for 0 ≤ s ≤ 2, 0 ≤ t ≤ 2π . This would have led us directly to the
integral that showed up at the end (with s and t in place of r and θ ).

10
• To nd the average value of a function on a surface, we integrate the function on the surface and then divide
by the surface area.

• Example:
p Find the average value of f (x, y, z) = z on the surface S given by the portion of the cone z =
x2 + y 2 that lies inside the cylinder x2 + y 2 = 4.

◦ By using cylindrical coordinates we see that we can parametrize this portion of the cone as x = s cos(t),
y = s sin(t), z = s, for 0≤s≤2 0 ≤ t ≤ 2π .
and

dr dr
◦ We then have r(s, t) = hs cos(t), s sin(t), si, so = hcos(t), sin(t), 1i and = h−s sin(t), s cos(t), 0i.
ds dt
i j k
∂r ∂r
◦ Then × = cos(t) sin(t) 1 = h−s cos(t), s sin(t), si, so the magnitude is given by
∂s ∂t
−s sin(t) s cos(t) 0
∂r ∂r
q √
× = s2 cos2 (t) + s2 sin2 (t) + s2 = s 2.
∂s ∂t
√
˜ ´ 2π ´ 2 √ ´ 2π 8 √ 16π 2
◦ We also have f (x, y, z) = z = s. So
S
z dσ = 0 0
s · s 2 ds dt = 0 3
2 dt = .
3
˜ ´ 2π ´ 2 √ ´ 2π √ √
◦ Also, the surface area is
S
1 dσ = 0 0 s 2 ds dt = 0 2 2 dt = 4π 2.
√
1 ˜ 16π 2/3 4
◦ Thus, the average value is
S
z dσ = √ = .
Area 4π 2 3

• Like with double, triple, and line integrals, we have mass and moment formulas for surface integrals:

• Center of Mass and Moment Formulas (Thin Surface): Given a surface S of variable density δ(x, y, z) in 3-
space:

˜
◦ The total mass M is given by M= δ(x, y, z) dσ .
˜
S
◦ The x-moment Myz is given by Myz = S x δ(x, y, z) dσ .
˜
◦ The y -moment Mxz is given by Mxz = S y δ(x, y, z) dσ .
˜
◦ The z -moment Mxy is given by Mxy = S z δ(x, y, z) dσ .
 
Myz Mxz Mxy
◦ The center of mass (x̄, ȳ, z̄) has coordinates , , .
M M M

• Example: A hill is shaped like the portion of the paraboloid z = 4 − x2 − y√

2
with z ≥ 0, with all coordinates
measured in meters. Snow accumulates on the hill such that the density is 17 − 4z grams per square meter
at height z. Find the total amount of snow on the hill.

◦ We are given the density of snow and want to compute the total mass, which (per the above) is given by
˜ √
the integral
S
17 − 4z dσ where S is the surface representing the hill.

◦ By using cylindrical coordinates, we can parametrize the hill as r(r, θ) = r cos(θ), r sin(θ), 4 − r2 , so
∂r ∂r
= hcos(θ), sin(θ), −2ri and = h−r sin(θ), r cos(θ), 0i.
∂r ∂θ
i j k √
∂r ∂r ∂r ∂r
◦ Then × = cos(θ) sin(θ) −2r = 2r2 cos(θ), 2r2 sin(θ), r , so × = 4r4 + r2 =
∂r ∂θ ∂r ∂θ
−r sin(θ) r cos(θ) 0
√
2
r 4r + 1.
p √
◦ We also have f (x, y, z) = 17 − 4(4 − r2 ) = 4r2 + 1.
´ 2π ´ 2 √ √ ´ 2π ´ 2 ´ 2π
◦ Hence the integral becomes 0 0 4r2 + 1 · r 4r2 + 1 dr dθ = 0 0 (r + 4r3 ) dr dθ = 0 18 dθ = 36π .
◦ Thus, there are 36π g of snow on the hill.

11
4.3 Vector Fields, Work, Circulation, Flux

• Denition: A vector eld in the plane is a function F(x, y) = hP (x, y), Q(x, y)i that associates a vector to
each point in the plane. A vector eld in 3-space is a function F(x, y, z) = hP (x, y, z), Q(x, y, z), R(x, y, z)i
that associates a vector to each point in 3-space.

◦ One vector eld we have already encountered is the vector eld associated to the gradient of a function
f (x, y) or f (x, y, z): for example, if f (x, y) = x2 + xy , then ∇f (x, y) = h2x + y, xi.

• To represent a vector eld visually, we choose some (nice) collection of points (generally in a grid) and draw the
vectors corresponding to those points as arrows pointing in the appropriate direction and with the appropriate
length.

◦ Example: The three vector elds F(x, y) = hx, yi, G(x, y) = h−y, xi, and H(x, y) = x + y 2 , 2 − 2xy
are plotted below on the region with −2 ≤ x ≤ 2, −2 ≤ y ≤ 2:
2 2 2

1 1 1

0 0 0

-1 -1 -1

-2 -2 -2

-2 -1 0 1 2 -2 -1 0 1 2 -2 -1 0 1 2

◦ We can also produce these plots for 3-dimensional vector elds, but the diagrams tend to be quite
cluttered; here is such a diagram for F(x, y, z) = hx, z − y, x + yi:

• We can think of a vector eld as describing the ow of an incompressible uid through space: the vector
F(x, y) at any point (x, y) gives the direction and velocity of the uid's ow there.

• In this context, if we have a particle that travels along some given path r(t) through the uid, we might like
to know how much work the uid does on the particle, or (essentially equivalently) how much the uid is
pushing the particle along its path. This is the central idea behind work integrals and circulation integrals.

◦ Intuitively, we see that the more the vector eld F aligns with the tangent vector T to the particle's
path, the more work it does.

◦ In the picture, a particle moving counterclockwise around the circle will be pushed along its path by the
vector eld:
2

-1

-2

-2 -1 0 1 2

12
 • Alternatively, if we have a particle traveling along a path, we could also ask: how much is the uid pushing
the particle o of the path? This is the central idea behind a ux integral.

◦ Another way of thinking about this is to imagine the path as being a thin membrane, and asking how
much uid is passing across the membrane.

◦ Here, we see that more uid is owing across the membrane if the vector eld F aligns with the normal
vector N to the particle's path:
2

-1

-2

-2 -1 0 1 2

• We can also formulate these ideas in 3-dimensional space: the ideas of circulation and work remain the same,
but the notion of ux requires a surface for the uid to ow across.

4.3.1 Circulation and Work Integrals

• To compute the circulation of a vector eld along a curve, we want to integrate the quantity measuring how
much the vector eld is aligning with the path of motion along the curve.

• Denition: The (counterclockwise) circulation (or ow) of the vector eld

´ F along the curve C is dened to
be
C
F · T ds, where T is the unit tangent to the curve.

◦ What this says is: the circulation is given by integrating the dot product function f (t) = F(x(t), y(t)) ·
T(t) along the curve C. In order to evaluate the integral as written, we would need to parametrize C,
nd the unit tangent vector T(t) to the curve, and then integrate the dot product F(x(t), y(t)) · T(t)
along the curve.

◦ We would like to see if there is a simpler way, so let us suppose that F(x, y) = hP, Qi, where P and Q
are functions of x y , and say C
and r(t) = hx(t), y(t)i from t = a to t = b.
is parametrized by
   
dx dy dx dy dx dy
, hP, Qi · , P +Q
v(t) dt dt dt dt dt dt .
◦ Then T(t) = = so F · T =
, =
||v(t)|| ||v(t)|| ||v(t)|| ||v(t)||
dx dy
´ ´ b P dt + Q dt ´b
 
dx dy
◦ We can then write C F · T ds = a · ||v(t)|| dt = a P +Q dt.
||v(t)|| dt dt
´b
 
dx dy
◦ Thus, the circulation integral can be written more explicitly as a P +Q dt, where P, Q have
dt ´ dt
been rewritten as functions of t. Note that this expression is also equal to P dx + Q dy .
C
◦ We can also pose essentially the same denition for a curve in 3-space, and we obtain an analogous
´b
 
dx dy dz
formula: if F = hP, Q, Ri, then the circulation can be computed as
a
P +Q +R dt.
dt dt dt
◦ Terminology Note: Some authors reserve the term circulation for closed curves, and use ow to refer
to the general case. This terminology can be somewhat confusing given that there is also a ux integral,
and the words ux and ow (in non-technical settings) are synonyms.

• Example: Find the circulation of the vector eld G(x, y) = h−y, xi around a path that winds once counter-
clockwise around the unit circle.

◦ We can parametrize the path as x = cos t, y = sin t for 0 ≤ t ≤ 2π .

dx dy
◦ Thus, P = −y = − sin t and Q = x = cos t, and also = − sin t and = cos t.
dt dt

13
 ´b ´ 2π ´ 2π
 
dx dy
◦ So, the circulation is
a
P +Q dt = 0 ((− sin t)(− sin t) + (cos t)(cos t)) dt = 0 1 dt = 2π .
dt dt

• Example: Find the circulation of the vector eld F(x, y, z) = h2xy, xz, yi along the line segment from (0, 1, 0)
to (2, 2, 2).

◦ We can parametrize the path as x = 2t, y = 1 + t, z = 2t for 0 ≤ t ≤ 1.

2 2
◦ P = 2xy = 4t + 4t , Q = xy = 4t , and R = 1 + t.
Thus,
´b ´1 ´1
 
dx dy dz
◦ So, the circulation is a P +Q +R dt = 0 [(4t + 4t2 ) · 2 + 4t2 · 1 + (1 + t) · 2] dt = 0 (2 +
dt dt dt
2
10t + 12t ) dt = 11 .

• We can also pose a similar denition for the work done by a vector eld on a particle:

• ´Denition: ´The work performed on a particle by a vector eld F as the particle travels along a curve C is

C
F · dr = C F · T ds.

◦ Note that the work integral has the same form as the circulation integral.

◦ Notation: The vector dierential dr is dened as dr = hdx, dyi in the plane and as dr = hdx, dy, dzi
in 3-space.
´ ´ ´b
 
dx dy
◦ Then F · dr = P dx + Q dy , F · dr = C P dx + Q dy = a
so the work integral is
C
P +Q dt in
dt dt
´ ´ ´b
 
dx dy dz
the plane, or as
C
F · dr = C
P dx + Q dy + R dz = a
P + Q + R dt in 3-space.
dt dt dt

• Example: Find the work done by the vector eld F(x, y, z) = h2x + z, yz, xyi on a particle traveling along
the path r(t) = t, t2 , 2t from t=0 to t = 1.

dx dy dz
◦ We have P = 2x + z = 3t, Q = yz = 2t3 , and R = xy = t3 . Also, = 1, = 2t, and = 2.
dt dt dt
´b ´1 ´1
 
dx dy dz 
◦ Therefore, the work is a P +Q +R dt = 0 (3t)(1) + (2t3 )(2t) + (t3 )(2) dt = 0 (3t +
dt dt dt
14
4t4 + 2t3 ) = .
5

4.3.2 Flux Across a Curve

• To compute the ux of a vector eld across a curve, we want to integrate the quantity measuring how much
the vector eld is moving in the direction perpendicular to the curve.
´
• Denition: The ux of the vector eld F across the curve C is
C
F · N ds, where N is the unit normal to the
curve.

◦ As with the circulation integral, we would like an easier way to evaluate the ux integral.

◦ If F(x, y) = hP, Qi and C  = hx(t), y(t)i from t = a to t = b, after some algebra

r(t)
is parametrized by

1 dy dx
we can calculate that N(t) = ,− . (At the very least, it is easy to observe that this is a
||v(t)|| dt dt
unit vector that is orthogonal to T.)
 
dy dx dy dx
hP, Qi · ,− P −Q
dt dt dt dt .
◦ Then F · N = =
||v(t)|| ||v(t)||
dy dx
´ ´ b P dt − Q dt ´b
 
dy dx
◦ Plugging this in gives C F · N ds = a · ||v(t)|| dt = a P −Q dt.
||v(t)|| dt dt
´ ´b
 
dy dx
◦ Thus, the ux integral can be written more explicitly as C P dy − Q dx = a P −Q dt.
dt dt

14
 ◦ Note: The ux integral as dened here only makes sense for curves in the plane. In 3-dimensional space,
the corresponding notion requires a surface integral, since a membrane will be a surface, rather than a
curve.

• Example: Find the ux of the vector eld G(x, y) = hx, yi across a path that winds once counterclockwise
around the unit circle.

◦ x = cos t, y = sin t for 0 ≤ t ≤ 2π .

We can parametrize the path as

dx dy
◦ Thus, P = x = cos t and Q = y = sin t, and also = − sin t and = cos t.
dt dt
´b ´ 2π ´ 2π
 
dy dx
◦ So Flux = a P −Q dt = 0 ((cos t)(cos t) − (sin t)(− sin t)) dt = 0 1 dt = 2π .
dt dt

• Example: For the vector eld F(x, y) = h2x + y, 2y − xi, nd the ux across, and circulation along, the
portion of the curve r(t) = t, t2 between (0, 0) and (2, 4).

◦ Here is a plot of the vector eld, along with the curve:

4

0.0 0.5 1.0 1.5 2.0

◦ From the picture, we would expect the circulation and ux to be roughly equal, since the vector eld
makes roughly a 45-degree angle with the path near the end.

◦ The parametrization given says y = t2 , so that P = 2x + y = 2t + t2 and Q = 2y − x = 2t2 − t.

x=t and
Also, the start is t = 0 and the end is t = 2.

´ ´b ´2
 
dx dy

◦ Then the circulation is C F · T ds = a P +Q dt = 0 (2t + t2 ) · 1 + (2t2 − t) · 2t dt =
dt dt
´2
1 2 52
0
4t3 − t2 + 2t dt = (t4 − t3 + t2 ) = .
3 t=0 3
ˆ 2 ˆ 2
´ ´b
 
dy dx
(2t + t2 ) · 2t − (2t2 − t) · 1 dt =



◦ The ux is C F·N ds = a P −Q dt = 2t3 + 2t2 + 2t dt =
dt dt 0 0
1 4 2 3 2 52
2
( t + t +t ) = .
2 3 t=0 3
◦ Indeed, we see that the ux and circulation are roughly (and exactly) equal.

4.3.3 Flux Across a Surface

• In 3-space, the notion of circulation along a curve remains essentially the same as in the plane. However, in
order to make sense of ux in 3-space, we must instead talk about ux through a surface rather than through
a curve. This requires us to use a surface integral to measure how much the vector eld is owing across the
surface:
˜
• Denition: The (normal) ux of the vector eld F across the surface S is
S
F · n dσ , where n is the outward
unit normal to the surface.

˜
◦ Remark: The integral
S
F · n dσ computes the ux through the surface in the direction of the outward
normal. It is also possible to ask about ux in the direction of a particular unit vector
˜ u; the integral in
that case is
S
F · u dσ , instead. In general, when it is not specied what type of ux integral is meant,
the ux in the direction of the outward normal is intended.

15
 ◦ Recall that the normal vector to a surface is orthogonal to the tangent plane (it is in fact the normal
vector to the tangent plane as we dened it earlier). When speaking of a unit normal to a surface we
will use a lowercase n, to keep the notation dierent from the unit normal N to a curve (which is an
uppercase N).
◦ If S is an implicit surfaceg(x, y, z) = c, then a normal vector is given by the gradient ∇g , so we get a
unit normal vector n = ∇g/ ||∇g||.
◦ If S r(s, t) = hx(s, t), y(s, t), z(s,
is parametrized by t)i, then a
normal vector is given by the cross
∂r ∂r ∂r ∂r ∂r ∂r
product ×so we get a unit normal vector n =
, × × .
∂s ∂t ∂s ∂t ∂s ∂t
◦ Important Warning: If we scale the implicit equation by −1, or write the factors of the cross product
in the opposite order, the resulting normal vector n is multiplied by −1. To remedy this ambiguity, we
must always specify which of these two possible orientations of the normal vector we intend. You should
always check to ensure that the normal vector is pointing in the correct direction: typical conventions
are for it to be pointing outward or upward.

• By plugging these expressions into the surface integral formula, we obtain explicit formulas for the outward
normal ux across a surface S:

◦ If S is parametrized by r(s, t) = hx(s, t), y(s, t), z(s, t)i, then the outward normal ux across S is equal
˜
 
∂r ∂r ∂r ∂r
to
S
F· × ds dt, provided that × is the outward-pointing normal vector to the surface.
∂s ∂t ∂s ∂t
(Conveniently, the unpleasant part of the surface-area dierential cancels out the normalization in the
unit normal vector.)

◦ If S f (x, y, z) = c and R is
is dened implicitly by the projection of S in the xy -plane, then the outward
˜ F · ∇g
normal ux across S is equal to
R |∇g · k|
dy dx. Note here that the denominator term ∇g · k is simply
the partial derivative gz .

◦ Depending on the description of the surface, either of these particular approaches (via a parametrization
or as an implicit surface) may be more convenient for computing a ux integral.

• Example: Find the outward ux of the vector eld F = xz 2 , yz 2 , x3 ey through the portion of the cylinder
2 2
x +y =4 that lies between the planes z = −1 and z = 1.

◦ From cylindrical coordinates, we can parametrize the cylinder as r(s, t) = h2 cos t, 2 sin t, si, where the
desired portion corresponds to −1 ≤ s ≤ 1 and 0 ≤ t ≤ 2π .
i j k
∂r ∂r ∂r ∂r
◦ Then = h−2 sin t, 2 cos t, 0i and = h0, 0, 1i, so × = −2 sin t 2 cos t 0 = h2 cos t, 2 sin t, 0i.
∂t ∂s ∂t ∂s
0 0 1
◦ This is indeed an outward-pointing normal vector since it is the vector pointing from (0, 0, s) to the point
r(s, t) = (2 cos t, 2 sin t, s) on the surface.
 
∂r ∂r
◦ Then F · × = 2s2 cos t, 2s2 sin t, (2 cos t)3 e2 sin t · h2 cos t, 2 sin t, 0i = 4s2 cos2 t + 4s2 sin2 t =
∂t ∂s
4s2 .
´ 2π ´ 1 ´ 2π 8 16π
◦ The ux integral is thus 0 −1 4s2 ds dt = 0 dt = .
3 3

• Example: Find the outward ux of the vector eld F = hx − z, y, x + zi through the portion of the sphere
x2 + y 2 + z 2 = 4 that lies above the plane z = 1.

◦ We use the formula for ux across an implicit surface.

◦ On the sphere, z=1 corresponds to x2 + y 2 = 3, and as z increases to 2, the value of x2 + y 2 decreases

2 2
to 0. Thus the projection of the surface into the xy -plane is the region R : x + y ≤ 3.

F · ∇g 2x2 − 2xz + 2y 2 + 2xz + 2z 2 4

◦ We have ∇g = h2x, 2y, 2zi, so = =p .
|∇g · k| 2z 4 − x2 − y 2

16
 ˜ 4
◦ The ux integral is therefore given by
R
p dy dx. We will evaluate this integral using polar
4 − x2 − y 2
coordinates.
√ ´ 2π ´ √3
4
◦ In polar coordinates, the region is 0≤r≤ 3 and 0 ≤ θ ≤ 2π , so the integral is 0 0
√
r dr dθ.
4 − r2
´ 2π ´ √3 4 ´ 2π ´ 0 2
◦ Substituting u = 4 − r2 in the inner integral gives
0 0
√ r dr dθ = 0 1 − √ du dθ =
4−r 2 u
´ 2π
0
4 dθ = 8π .

◦ Alternatively, we could have observed that for a sphere of radius ρ centered at the origin, the outward
1
unit normal vector is n = hx, y, zi.
ρ
˜ 1 ˜ 1 ˜
◦ The desired integral is therefore S hx, y, zi · hx − z, y, x + zi dσ = S (x2 + y 2 + z 2 ) dσ = S 2 dσ .
2 2
◦ This is twice the surface area of S , which we could compute (using a simpler surface integral) to be 4π ,
meaning that the desired ux is again 8π .

• Example: Find the outward ux of the vector eld F = hx, y, zi through the sphere x2 + y 2 + z 2 = 9.

◦ Using spherical coordinates, we can parametrize the sphere as r(s, t) = h3 sin s cos t, 3 sin s sin t, 3 cos si
0 ≤ s ≤ π and 0 ≤ t ≤ 2π .
for
∂r ∂r ∂r ∂r
◦ Then = h−3 sin s sin t, 3 sin s cos t, 0i and = h3 cos s cos t, 3 cos s sin t, −3 sin si, so × =
∂t ∂s ∂t ∂s
i j k
−3 sin s sin t 3 sin s cos t 0 = −9 sin2 s cos t, −9 sin2 s sin t, −9 sin s cos s .
3 cos s cos t 3 cos s sin t −3 sin s
◦ This is not an outward-pointing normal vector, since it is −3 sin s times the position vector r(s, t), so we
must scale it by −1.
 
∂r ∂r
◦ Then F·− × = h3 sin s cos t, 3 sin s sin t, 3 cos si· 9 sin2 s cos t, 9 sin2 s sin t, 9 sin s cos s = 27 sin3 s cos2 t+
∂t ∂s
27 sin3 s sin2 t + 27 sin s cos2 s = 27 sin s.
´ 2π ´ π ´ 2π
◦ The ux integral is thus 0 0 27 sin s ds dt = 0 54 dt = 108π .

4.4 Conservative Vector Fields, Path-Independence, and Potential Functions

• If we have a vector eld F(x, y) and two dierent paths C´1 and C2 between the same two points, we might
´
wonder if there is any relation between the work integrals
C1
F · dr and
C2
F · dr.

• Example: For the elds F(x, y) = hy, xi and G(x, y) = y 2 , x evaluate the work integrals from (0, 0) to
3 2 7 10
(1, 1) along the three dierent paths C1 : (x, y) = (t, t), C2 : (x, y) = (t , t ), and C3 : (x, y) = (t , t ), for
0 ≤ t ≤ 1.
dx dy
◦ Along C1 we have F = ht, ti, G = t2 , t , = 1, and = 1.
dt dt
´ ´1 ´ ´1  5
◦ Then
C1
F · dr = 0
[t · 1 + t · 1] dt = 1 , and
C1
G · dr = 0 t2 · 1 + t · 1 dt = .
6
dx dy
◦ Along C2 we have F = t2 , t3 , G = t4 , t3 , = 3t2 , and = 2t.
dt dt
´ ´1 2  ´ ´1  29
◦ Then
C2
F · dr = 0
t · 3t2 + t3 · 2t dt = 1 , and
C2
G · dr = 0 t4 · 3t2 + t3 · 2t dt = .
35
dx dy
◦ Along C3 we have F = t10 , t7 , G = t20 , t7
= 7t6 , and , = 10t9 .
dt dt
´ ´1  ´ ´1  389
◦ Then
C3
F · dr = 0 t10 · 7t6 + t7 · 10t9 dt = 1 , and C3 G · dr = 0 t30 · 7t6 + t7 · 10t9 dt = .
459

17
 ◦ Observe that for F, all three paths give the same value, while for G, each path gives a dierent value.

• We would like to understand what about F in the example above seems to cause it to do the same amount of
work regardless of the path we chose.

• Denition: A vector eld

´ F is conservative
´ on a region R if, for any two paths C1 and C2 (inside R) from P1
to P2 , it is true that
C1 C2
F · dr = F · dr. In other words, F is conservative if any two paths with the same
endpoints yield the same work integral.

◦ Equivalent to the above denition is the following:

¸ F is conservative on a region R if, for any closed
curve C in R, F · dr = 0. (A closed curve is one whose start and end points are the same.)
C
¸
◦ Notation: For a line integral around a closed curve, we often use the notation
C
, the circle being a
suggestive example of a closed curve.

◦ These two statements are equivalent because, if C1 and C2 are two paths from P1 to P2 , then we can
construct a closed path
´ ´ C by following C1 from P1 to P2 and then following C2 from P2 back to P1 .
´
Then
C
F · dr = C1 F · dr − C2 F · dr, and so the left-hand side is zero if and only if the right-hand side
is zero.

• It turns out that we can give a simple but very useful criterion for when a vector eld is conservative:

• Theorem (Fundamental Theorem of Calculus for Line Integrals): The vector eld F is conservative on a
simply-connected region R if and only if there exists a function U , called a potential function for F, such that
´b
F = ∇U . If such a function U exists, then
a
F · dr = U (b) − U (a) along any path from a to b.
´b
◦ Notice the similarity of the statement
a
F · dr = U (b) − U (a) to the Fundamental Theorem of Calculus,
which relates the integral of a derivative of a function to its values at the endpoints of a path.

◦ Technical Note: The term simply-connected is a technical requirement needed for the proof of the
theorem: intuitively, a simply-connected region consists of a single piece that does not have any holes
in it. More rigorously, it means that the region is connected (contains only one piece) and that if we
take any closed loop in the region, we can shrink it to a point without leaving the region. The disc
x2 + y 2 ≤ 4 is simply-connected, whereas the annulus 1 ≤ x2 + y 2 ≤ 4 is not.

◦ The full proof is not especially enlightening. We will instead show one direction of the proof.
 
∂U ∂U ∂U
◦ Proof (Reverse Direction in 3-Space): Suppose that F = ∇U = , , .
∂x ∂y ∂z
◦ C is the
By the (multivariable) Chain Rule, if path with x = x(t), y = y(t), and z = z(t) for a ≤ t ≤ b,
dU ∂U dx ∂U dy ∂U dz
then = · + · + · .
dt ∂x dt ∂y dt ∂z dt
◦ Now we can write

ˆ ˆ b    
∂U ∂U ∂Udx dy dz
F · dr = , ,· , , dt
C a ∂x ∂y ∂z dt dt dt
ˆ b 
∂U dx ∂U dy ∂U dz
= · + · + · dt
a ∂x dt ∂y dt ∂z dt
ˆ b 
dU
= dt = U (r(b)) − U (r(a))
a dt

where we used the Fundamental Theorem of Calculus for the last step.

◦ Notice that this expression does not depend on C: it only involves the potential function U and the two
endpoints r(b) and r(a). Hence we see that the integral is independent of the path, so F is conservative.

• If we can see that a vector eld is conservative, then it is very easy to compute work integrals: we just need
to nd a potential function for the vector eld.

• Example: Find the work done by the vector eld F(x, y) = h2x + y, xi on a particle traveling along the path
r(t) = −2 cos(πet ), tan−1 (t) from t=0 to t = 1.

18
 ◦ If we try to set up the integral directly using the parametrization, it will be rather unpleasant.

◦ However, this vector eld is conservative: it is not hard to see that for U (x, y) = x2 + xy , we have
∇U = h2x + y, xi = F.
◦ By the Fundamental Theorem of Calculus for line integrals, the work done by the vector eld is then
simply the value of U (r(1)) − U (r(0)).
π
◦ Since r(1) = h2, π/4i and r(0) = h−2, 0i, the work is U (2, π/4) − U (−2, 0) = .
2

• We would like to be able to determine easily whether a given vector eld is conservative. To do this, we
require a preliminary denition:

• Denition: If F = hP, Q, Ri then the curl of F is dened to be the vector eld curl F = ∇ × F =
i j k  
∂R ∂Q ∂P ∂R ∂Q ∂P
∂/∂x ∂/∂y ∂/∂z = − , − , − = hRy − Qz , Pz − Rx , Qx − Py i.
∂y ∂z ∂z ∂x ∂x ∂y
P Q R

◦ Example: If F = 3x2 y, xyz, exy then curl F=∇×F= xexy − xy, −yexy , yz − 3x2 .

◦ If F = hP, Qi is a vector eld in the plane then we dene the curl of

  F to be the curl of the vector eld
∂Q ∂P
hP, Q, 0i: namely, 0, 0, − .
∂x ∂y
◦ Since this vector only has one nonzero component, some authors dene the curl of a vector eld in the
∂Q ∂P
plane to be the scalar quantity − . We will not do this: for us, the curl of a vector eld will
∂x ∂y
always be a new vector eld.

• The curl of a vector eld determines whether or not it is conservative:

• Theorem (Zero Curl Implies Conservative): A vector eld on a simply-connected region in the plane or in
3-space is conservative if and only if its curl is zero. More explicitly, the vector eld F = hP, Qi is conservative
on a simply-connected region R in the plane if and only if Py = Qx , and the vector eld F = hP, Q, Ri is
conservative on a simply-connected region D in 3-space if and only if Py = Qx , Pz = Rx , and Qz = Ry .

◦ It is fairly easy to see why we need the equality of the derivatives of the components: if F = hP, Qi = ∇U
then P = Ux and Q = Uy , so by the equality of mixed partial derivatives, we see that Py = Uxy = Uyx =
Qx .
◦ The three necessary equalities when F = hP, Q, Ri follow in the same way: if F = ∇U then P = Ux ,
Q = Uy , and R = Uz , so Py = Uxy = Uyx = Qx , Pz = Uxz = Uzx = Rx , and Qz = Uyz = Uzy = Ry .
◦ The converse statement (that zero curl implies the eld is conservative) is more dicult, and we omit
the verication.

• The two theorems give us an eective procedure for determining whether a eld is conservative: we rst check
whether its curl is zero, and then (if it is) we can try to nd a potential function by computing antiderivatives.

• Example: Determine whether F(x, y) = x2 + y, x + y 2 is conservative, and if so, nd a potential function.

∂  2  ∂  
◦ For F, x +y =1=
we see x + y 2 , so the eld is conservative .
∂y ∂x
◦ To nd a potential function U with ∇U = F, we need to nd U such that Ux = x2 + y and Uy = x + y 2 .
1
◦ Taking the antiderivative of Ux = x2 + y with respect to x yields U = x3 + xy + f (y), for some function
3
f (y).
1
◦ To nd f (y) we dierentiate: Uy = x + f 0 (y), so we get f 0 (y) = y 2 so f (y) = y 3 . (Plus an arbitrary
3
constant, but we can ignore it.)

1 3 1
◦ Thus we see that a potential function for F is U (x, y) = x + xy + y 3 .
3 3

19
 • Example: Determine whether G(x, y) = x + y 2 , x2 + y is conservative, and if so, nd a potential function.

∂   ∂  2 
◦ For G, we see x + y 2 = 2y 6= 2x = x +y , so the eld is not conservative .
∂y ∂x

• Example: Determine whether H(x, y, z) = hy + 2z, x + 3z, 2x + 3yi is conservative, and if so, nd a potential
function.

∂ ∂ ∂ ∂ ∂
◦ For H, we have [y + 2z] = 1 = [x + 3z], [y + 2z] = 2 = [2x + 3y], and [x + 3z] = 3 =
∂y ∂x ∂z ∂x ∂z
∂
[2x + 3y], so the eld is conservative .
∂y
◦ To nd a potential function U with ∇U = H, we need to nd U such that Ux = y + 2z , Uy = x + 3z ,
and Uz = 2x + 3y .
◦ Taking the antiderivative of Ux = y + 2z with respect to x yields U = xy + 2xz + f (y, z), for some
function f (y, z).
◦ To nd f (y, z) we dierentiate: x+fy = x+3z and 2x+fz = 2x+3y , so fy = 3z and fz = 3y . Repeating
the process yields f = 3yz + g(z), where g 0 (z) = 0.
◦ Thus we see that a potential function for H is U (x, y, z) = xy + 2xz + 3yz .

• If we can nd a potential function for a conservative vector eld, then (as we saw above) we can use it to
compute work integrals.

• Example: If F = x3 + 4x3 sin y sin z + y 2 z, 2xyz + y + x4 cos y sin z, z 3 + x4 sin y cos z + xy 2 , nd the work
√
done by F on a particle that travels along the curve C : r(t) = sin(πt), t t + 3, 2t3 + 2 for 0 ≤ t ≤ 1.

◦ In theory we could compute the work integral using the parametrization of the path, but this seems quite
unpleasant. Instead, we will check whether this vector eld is conservative: then determining the answer
only requires us to nd the potential function of the eld.

◦ We have Py = 4x3 cos y sin z + 2yz and Qx = 2yz + 4x3 cos y sin z so they are equal.
3 2 3 2
◦ We have Pz = 4x sin y cos z + y and Rx = 4x sin y cos z + y so they are also equal.
4 3
◦ Finally we have Qz = 2xy + x cos y cos z and Ry = 4x cos y cos z + 2xy , and these are also equal. Thus,
the eld is conservative.

◦ To nd a potential function U with F = ∇U = hUx , Uy , Uz i:

∗ We know Ux = x + 4x sin y sin z + y 2 z
3 3
so taking the antiderivative with respect to x yields U =
1 4
x + x4 sin y sin z + xy 2 z + C(y, z).
4
∗ We then see Uy = x4 cos y sin z + 2xyz + Cy (y, z) must equal 2xyz + y + x4 cos y sin z so we see
1
Cy = y . Then taking the antiderivative with respect to y yields C(y, z) = y 2 + D(z).
2
1 4 1
∗ We now have x + x4 sin y sin z + xy 2 z + y 2 + D(z). Then Uz = x4 sin y cos z + xy 2 + D0 (z)
U=
4 2
3 4 2 0 3 1 4
must equal z + x sin y cos z + xy so we see D (z) = z so we can take D(z) = z .
4
1 1 1
◦ We conclude that a potential function for F is U (x, y, z) = x4 + x4 sin y sin z + xy 2 z + y 2 + z 4 .
4 2 4
◦ Then the desired work integral is equal to U (0, 2, 4) − U (0, 0, 2) = 62 .

4.5 Green's Theorem

• Green's Theorem is a 2-dimensional version of the Fundamental Theorem of Calculus that relates a line
integral of a function around a closed curve C to the double integral of a related function over the region R
that is enclosed by the curve C.

20
• Theorem (Green's Theorem): If C is a simple closed rectiable curve oriented counterclockwise,
ˆ and R is the re-
¨  
∂Q ∂P
gion it encloses, then for any continuously dierentiable functions P (x, y) and Q(x, y), P dx + Q dy = − d
C R ∂x ∂y
◦ Here is an example of a curve C and its corresponding region R:

◦ Green's Theorem, as noted above, is a generalization of the Fundamental Theorem of Calculus: both
theorems show that the integral of the derivative of a function (in an appropriate sense) on a region can
be computed using only the values of the function on the boundary of the region.

◦ Remark: The hypotheses about the curve (simple closed rectiable, oriented counterclockwise) are to
ensure the curve is nice enough for the theorem to hold. Simple means that the curve does not cross
itself, closed means that its starting point is the same as its ending point (e.g., a circle), rectiable
means piecewise-dierentiable (i.e., dierentiable except at a nite number of points), and oriented
counterclockwise means that C runs around the boundary of R in the counterclockwise direction.

◦ It essentially suces to prove Green's Theorem for rectangular regions, as more complicated regions
can be built by gluing together simpler ones (in much the manner of a Riemann sum); overlapping
boundary pieces on two rectangles sharing a side will have opposite orientations and will therefore cancel
out.
´ ´ ´ ´ ´
◦ Proof (rectangular regions): for a rectangular region a ≤ x ≤ b, c ≤ y ≤ d, we have
C C1 C2 C3
= + + + C4
,
where C1 , C2 , C3 , and C4 are the four sides of the rectangle (with the proper orientation), and the func-
tion to be integrated on each curve is P dx + Q dy .
´ ´ ´b
◦ Setting up parametrizations shows [P dx + Q dy] + C3 [P dx + Q dy] = a [P (x, c) − P (x, d)] dx, and
´ ´ ´d C1

C2
[P dx + Q dy] + C4 [P dx + Q dy] = c [Q(b, y) − Q(a, y)] dy .
˜ ∂P ´ b ´ d ∂P ´b
◦ For the double integral we have R − dy dx = a c − dy dx = a [P (x, c) − P (x, d)] dx, and
∂y ∂y
˜ ∂Q ´ c ´ b ∂Q ´d
R ∂x
dx dy = d a dx dy = c [Q(b, y) − Q(a, y)] dy .
∂x
´ ˜
 
∂Q ∂P
◦ By comparing the expressions, we see that C [P dx + Q dy] = R − dy dx, as desired.
∂x ∂y
• Green's Theorem can be used to convert line integrals into double integrals, which can often be easier to
evaluate if the curve is complicated but the region it encloses is simpler to describe.
¸
• Example: Evaluate the integral
C
3x2 dx + 2xy dy , where C is the counterclockwise boundary of the triangle
having vertices (0, 0), (1, 0), and (1, 2).

◦ We will evaluate the integral both as a line integral and using Green's Theorem.
´ ˜
◦ Green's Theorem says that
¸ P dx + Q dy = R (Qx − Py ) dy dx, so setting P = 3x2
C˜
and Q = 2xy
produces
C
3x2 dx + 2xy dy = R 2y dy dx, where R is the interior of the triangle.
◦ To compute the double integral, we need to describe the region R. A quick sketch shows that R is dened
by the inequalities 0 ≤ x ≤ 1 and 0 ≤ y ≤ 2 − 2x.
´ 1 ´ 2−2x ´1 2−2x ´1 4
◦ Thus, the double integral is
0 0
2y dy dx = 0 (y 2 ) y=0
dx = 0
(2 − 2x)2 dx = .
3
◦ To compute the line integral, we need to parametrize each piece of the boundary. There are three pieces.

1. The segment from (0, 0) to (1, 0), parametrized by x = t, y = 0 for 0 ≤ t ≤ 1. Then dx = dt and
´1 2
dy = 0, so the integral here is
0
3t dt = 1.

21
 2. The segment from (1, 0) to (1, 2), parametrized by x = 1, y = t for 0 ≤ t ≤ 2. Then dx = 0 and
´2
dy = dt, 2t dt = 4.
so the integral here is
0
3. The segment from (1, 2) to (0, 0), parametrized by x = 1 − t, y = 2 − 2t for 0 ≤ t ≤ 1. Then
´1 
dx = −dt and dy = −2dt, so the integral here is 0 3(1 − t)2 · (−dt) + 2(1 − t)(2 − 2t) · (−2dt) =
´1 11
0
[−11t + 22t − 11t2 ] dt = − .
3
11
◦ Thus, the value of the line integral over the entire boundary is the sum of these three, namely 1+4− =
3
4
.
3
◦ As dictated by Green's theorem, we get the same result either way. However, the double integral was
quite a bit less work!

• We can use Green's Theorem to simplify the calculation of circulation and ux integrals on closed curves.

◦ Specically, we can use the theorem to give expressions for circulation and ux either as line integrals or
as double integrals over a region.

◦ Depending on the shape of the region and its boundary, and the nature of the eld F, either the line
integral or the double integral can be easier.

• Theorem (Green's Theorem, Tangential Form): If C is a simple closed rectiable curve oriented counterclock-
wise and R ˛ it encloses,¨and
is the region F is a continuously dierentiable vector eld, then the circulation

around C is equal to F · T ds = (curl F) · k dA.

C R
 
∂Q ∂P ∂Q ∂P
◦ Recall that if F = hP, Qi, then curl F=∇×F= 0, 0, − and (curl F) · k = − . The
∂x ∂y ∂x ∂y
curl measures how much the vector eld is rotating around a given point.

◦ Thus, if we write everything out in terms of vector eld components, the tangential form of Green's
¸ ˜
 
∂Q ∂P
Theorem reads
C
P dx + Q dy = R
− dy dx, which is just the statement we gave above.
∂x ∂y

• Theorem (Green's Theorem, Normal Form): If C is a simple closed rectiable curve oriented counterclockwise
and R is ˛the region it encloses,
¨ and F is a continuously dierentiable vector eld, then the ux across C is

equal to F · N ds = (div F) dA.

C R

∂P ∂Q
◦ Here, if F = hP, Qi then div F = ∇·F = + . This is called the divergence of F and measures
∂x ∂y
how much the vector eld is pushing inward or outward at the given point.

¸ ˜
 
∂P ∂Q
◦ Explicitly, the normal form of Green's Theorem reads C P dy − Q dx = R + dy dx, which
∂x ∂y
we can recognize as the original statement of Green's Theorem except with −Q in place of P and P in
place of Q.

◦ There is a nice interpretation of the normal form of Green's Theorem: imagine that F is modeling
population movement, and that C is the border of a country taking up the region R. At a city along the
border C, the value F·N measures the immigration (in or out) to that city from across the border. At
a city inside the country, the value div F measures the net immigration (into or out of ) that city.

◦ The normal form of Green's Theorem then says: if we add up the net immigration along the border, this
equals the total population ow inside the country. (These two quantities are denitely equal, since they
both tally the net immigration into the country as a whole.)

• Example: Find the outward ux through, and the (counterclockwise) circulation around, the square with
vertices (0, 0), (2, 0), (2, 2), and (0, 2), for the vector eld F(x, y) = x2 − 2xy, y 3 − x .

22
 ◦ We could parametrize the boundary of this region and evaluate the line integrals to nd the ux and
circulation. However, this would be very tedious, as it requires computing four line integrals each time
(one for each side of the square). We can save a lot of eort by using Green's Theorem, which applies
because the boundary is a closed curve.

¸ ˜
 
∂P ∂Q
◦ For the ux, Green's Theorem says that Flux across C= C
F · N ds = R
+ dy dx.
∂x ∂y
◦ Here, we haveP = x2 − 2xy and Q = y 3 − x, and the region is 0 ≤ x ≤ 2 and 0 ≤ y ≤ 2.
∂P ∂Q
◦ Therefore, since = 2x − 2y and = 3y 2 , the ux is
ˆ 2ˆ 2 ∂x ∂y
´2
2 ´2
2x − 2y + 3y 2 dy dx = 0 2xy − y 2 + y 3


dx = 0 (4x + 4) dx = 16 .
0 0 y=0
¸ ˜
 
∂Q ∂P
◦ Green's Theorem also says that Circulation around C = C F · T ds = R − dy dx.
∂x ∂y
ˆ 2ˆ 2
∂Q ∂P ´2
◦ Since = −1 and = −2x, the circulation is (−1 + 2x) dy dx = 0 (−2 + 4x) dx = 4 .
∂x ∂y 0 0

• Example: For F(x, y) = −x2 y, xy 2 , nd the outward ux through and the (counterclockwise) circulation
2 2
around the circle x + y = 4.

◦ We apply Green's Theorem: in this case, the region R is the region x2 + y 2 ≤ 4.

˜ ˜ ˜
 
∂P ∂Q
◦ The ux is
R
+ dy dx = R (−2xy + 2xy) dA = R 0 dA = 0 .
∂x ∂y
˜ ˜ ´ 2π ´ 2
 
∂Q ∂P
◦ The circulation is
R
− dy dx = R (y 2 + x2 ) dA = 0 0 r2 · r dr dθ = 8π , upon switching
∂x ∂y
to polar coordinates.

• One of the many applications of Green's Theorem is to give various ways to compute the area of a planar
region using a line integral around its boundary. Specically, if C is the counterclockwise boundary curve of
the region R (and C and R satisfy the hypotheses of Green's Theorem), then

˛ ˛ ˛
1
Area of R= x dy = −y dx = (x dy − y dx)
C C C 2
˜
because by Green's Theorem, each of the line integrals is equal to
R
1 dy dx, which is the area of R.

◦ One physical application of this idea is the construction of planimeters: they are devices used for mea-
suring the area of a region that operate by tracing along its boundary.

◦ The basic principle is that the planimeter measures the amount of movement perpendicular to its measur-
ing arm: integrating the resulting dot product around the boundary of the curve, per Green's theorem,
then yields the area.

• Example: Compute the area enclosed by the ellipse x = a cos t, y = b sin t, 0 ≤ t ≤ 2π .

◦ Using the third formula, we compute

¸ 1 ´ 2π 1 ´ 2π ab
A= C 2
(x dy − y dx) = 0 [(a cos t)(b cos t) − (b sin t)(−a sin t)] dt = 0 dt = πab .
2 2

4.6 Stokes's Theorem and Gauss's Divergence Theorem

• We now discuss two generalizations of Green's theorem to 3 dimensions: these are Stokes's Theorem and
Gauss's Divergence Theorem.

◦ As with Green's Theorem, these theorems can be used in either direction, depending on which integral
is easier to set up and evaluate.

23
 ◦ Indeed, taken together, the Fundamental Theorem of Calculus for line integrals, Green's Theorem,
Stokes's Theorem, and Gauss's Divergence Theorem collectively unify all of our notions of integration,
and are all dierent generalizations of the Fundamental Theorem of Calculus.

◦ They all relate the integral of a function on the boundary of a region to the integral of a derivative on
the interior of the region.
ˆ ˆ
◦ Symbolically, their statements all read as dω = ω , where dω represents an appropriate dierential
∂R R
of a function ω and ∂R represents the boundary of the region R.

4.6.1 Stokes's Theorem

• We begin with Stokes's theorem, which is the 3-dimensional version of the tangential form of Green's theorem:

• Theorem (Stokes's Theorem): If C is a simple closed rectiable curve in 3-space that is oriented counterclock-
S , and
˛ the surface¨
wise around F is a continuously dierentiable vector eld, then the circulation around C is
given by F · T ds = (curl F) · n dσ , where T is the unit tangent to the curve and n is the unit normal
C S
to the surface.

◦ Important Note: The curve C must run counterclockwise around S: in other words, when walking along
C, the surface should be on its left-hand side. If one wishes to set up a problem where a curve runs
clockwise around a surface, it is equivalent to traversing the curve in the opposite direction, and so the
integral will be scaled by −1.
◦ The hypotheses about the curve (simple closed rectiable, oriented counterclockwise) are the same as
in Green's Theorem, and they ensure the curve is nice enough for the theorem to hold.

i j k  
∂R ∂Q ∂P ∂R ∂Q ∂P
◦ Recall curl F = ∇×F = ∂/∂x ∂/∂y ∂/∂z = − , − , − if F = hP, Q, Ri.
∂y ∂z ∂z ∂x ∂x ∂y
P Q R
◦ Intuitively, if we think of a vector eld as modeling the ow of a uid, the quantity (curl F) · n at (x, y, z)
measures how much the uid is circulating around the point (x, y, z) along the surface. Stokes's Theorem
then says: we can measure how much the uid circulates around the whole surface by measuring how
much it circles around its boundary.

◦ The proof of Stokes's Theorem (which we omit) is essentially the same as the proof of Green's Theorem:
we can reduce to the case of showing the result for simple patches on the surface. Then, by parametrizing
the patches explicitly, we can show Stokes's Theorem is essentially the same as the tangential form of
Green's Theorem on each patch.

• Stokes's Theorem generalizes the tangential form of Green's Theorem to cover 3-dimensional closed curves
and the surfaces they bound. Note that unlike in Green's Theorem, there are many possible surfaces that any
given curve can bound.

◦ For example, the unit circle x2 + y 2 = 1, z = 0 in the xy -plane bounds the upper portions (i.e.,
p where
2 22 2 2
z ≥ 0) of the sphere x + y + z = 1, the paraboloid z = 2(1 − x − y ), and the cone z = 1 − x2 + y 2 ,
as pictured below:

24
• Typically, we use Stokes's Theorem when the line integral over the boundary is dicult, but there is a nicer
surface available.

• Example: Find the circulation of the eld F(x, y, z) = y 2 z 3 , 2xyz 3 , 3xy 2 z 2 around the ellipse given by the
2 2 2 2 2 2
intersection of the upper half of the ellipsoid x + 2y + 2z = 12 with the cone x + 2y = z .

◦ Here is a picture of the surfaces and the ellipse:

◦ We could write down a parametrization for this ellipse with a little bit of eort: substituting the cone's
2 2 2
equation into the sphere's equation gives 3z
√ = 12 hence z = 2. Then using the fact that x + 2y = 4
is parametrized by
√ x = 2 cos(t) and y = 2 sin(t) gives us a parametrization for the curve as r(t) =
2 cos(t), 2 sin(t), 2 . The resulting circulation integral does not look so wonderful, although it is
possible to evaluate it.

◦ Another way is to try to use Stokes's Theorem. We have two obvious surfaces to choose from (ellipsoid
and cone); since the curve runs counterclockwise around the ellipsoid, we will use that.
¸ ˜
◦ Stokes's Theorem tells us that Circulation around C= C
F · T ds = S
(curl F) · n dσ .
i j k
◦ We have curl F = ∂/∂x ∂/∂y ∂/∂z = 6xyz 2 − 6xyz 2 , 3y 2 z 2 − 3y 2 z 2 , 2yz 3 − 2yz 3 = h0, 0, 0i.
y2 z3 2xyz 3 3xy 2 z 2
◦ So the curl of F is zero. Hence (curl F) · n will also be zero, so we see that the circulation is 0 , without
even having to set up the surface integral.

˜
• Example: Find the ux of the curl
S
curl(F) · n dσ , where F = yzi − xzj + ex+y k, S is the surface which is
2 2 2
the part of the sphere x + y + z = 25 below the plane z = 3, and n is the outward normal.

◦ We will use Stokes's Theorem. In this case, we want S to be the part of the sphere x2 + y 2 + z 2 = 25
which is below the plane z = 3.
◦ The boundary of this surface will be the intersection of the plane and the sphere:
 we see that the
curve is the set of points (x, y, z) : x2 + y 2 = 16, z = 3 , which is a circle that we can parametrize as
r(t) = h4 cos(t), 4 sin(t), 3i for 0 ≤ t ≤ 2π .
◦ However: the surface S lies below the curve C, not above it: so, when viewed from below (which is
required because we are using the the outward normal), the curve runs clockwise around the surface.

◦ In order to apply Stokes's Theorem, we need to reverse the orientation of the curve C , which we can do
by interchanging the limits of integration: thus we start at t = 0. t = 2π and end at
´ ´
◦ From Stokes's Theorem, the ux of the curl is given by the line integral C F · dr = C P dx + Q dy + R dz .
◦ We have P = 12 sin(t), Q = −12 cos(t), and R = e4 cos(t)+4 sin(t) , and also dx = −4 sin(t) dt, dy =
4 cos(t) dt, and dz = 0 dt.
´ ´0  
◦ We get C F · dr = 2π (12 sin(t)) · (−4 sin(t) dt) + (−12 cos(t)) · (4 cos(t) dt)) + e4 cos(t)+4 sin(t) · 0 dt =
´0
2π
−48 dt = 96π .

25
4.6.2 Gauss's Divergence Theorem

• Now we discuss Gauss's Divergence Theorem, which is the 3-dimensional version of the normal form of Green's
theorem:

• Theorem (Gauss's Divergence Theorem): If S is a closed, bounded, piecewise-smooth surface that fully encloses

¨ D, and F is a continuously dierentiable vector eld, then the ux across S is given by
a solid region ˚

F · n dσ = (div F) dV , where n is the outward unit normal to the surface.

S D

◦ To get an idea of the setup, if S x2 + y 2 + z 2 = 1, then D would be the unit ball

is the unit sphere
x2 + y 2 + z 2 ≤ 1 . If S D would be the interior of the cube.
consists of the 6 faces of the unit cube, then

∂P ∂Q ∂R
◦ Here, if F = hP, Q, Ri then div F=∇·F= + + .
∂x ∂y ∂z
◦ Intuitively, if we think of a vector eld as modeling the ow of a uid, the divergence measures whether
there is a source or a sink at a given point (i.e., whether uid is owing inward toward that point,
or outward from that point). The ux through a surface measures how much uid is owing across the
surface.

◦ The Divergence Theorem then says that we can measure how much uid is owing in or out of a solid
region by measuring how much uid is owing across its boundary.

◦ The proof of the Divergence Theorem (which we omit) is essentially the same as the proof of Green's
Theorem: we reduce to the case of showing the result for rectangular boxes, and then parametrize the
boxes explicitly.

• Typically, we want to use the Divergence Theorem to compute the ux through a closed surface, since it is
usually easier to evaluate the triple integral than the surface integral.

• Example: Find the outward ux of the eld F(x, y, z) = x3 − 3y, 2yz + 1, xyz through the cube bounded
by the planes x = ±1, y = ±1, z = ±1.

◦ We could do this directly by computing the ux across each of the six faces of the cube. This is not the
best idea, because it would require setting up six surface integrals.
˜ ˝
◦ Instead, we use the Divergence Theorem: it says Flux across S= S
F · n dσ = (div F) dV .
V


◦ The solid region V is dened by −1 ≤ x ≤ 1, −1 ≤ y ≤ 1, −1 ≤ z ≤ 1, and div F = 3x2 + (2z) + (xy).
◦ Thus, the ux integral is

ˆ 1 ˆ 1 ˆ 1 ˆ 1 ˆ 1 1
3x2 + 2z + xy dz dy dx 3x2 z + z 2 + xyz



= dy dx
−1 −1 −1 −1 −1 z=−1
ˆ 1 ˆ 1
6x2 + 2xy dy dx


=
−1 −1
ˆ 1 1
ˆ 1
2 2
12x2 dx = 8 .


= 6x y + xy dx =
−1 y=−1 −1
‚
• Example: Compute the ux
S
F · n dσ , where F = (x3 + yz)i + (y 3 + xz)j + (z 3 + xy)k, S is the unit sphere
2 2 2
x + y + z = 1, and n is the outward normal.

◦ We will use the Divergence Theorem. If F = hP, Q, Ri then div(F) = Px + Qy + Rz , so here we have
div(F) = 3x2 + 3y 2 + 3z 2 .
◦ The region enclosed by S is the unit ball x2 + y 2 + z 2 ≤ 1 .
˝

◦ Thus the triple integral is 3x2 + 3y 2 + 3z 2 dz dy dx.
x2 +y 2 +z 2 ≤1
◦ To evaluate this integral we switch to spherical coordinates: the region is bounded by the inequalities
0 ≤ ρ ≤ 1, 0 ≤ φ ≤ π , and 0 ≤ θ ≤ 2π , the function is 3ρ2 , and the dierential is ρ2 sin(φ) dρ dφ dθ.
´ 2π ´ π ´ 1 ´ 2π ´ π 3 ´ 2π 6 12π
◦ So we obtain
0 0 0
3ρ2 · ρ2 sin(φ) dρ dφ dθ = 0 0 5
sin(φ) dφ dθ = 0 dθ = .
5 5

26
4.7 Applications of Vector Calculus

• In this section we discuss a number of moderately related applications of vector calculus. Many of our examples
are drawn from physics and engineering, but we also mention some examples that show up in other elds.

4.7.1 Newton's Law of Gravitation and Kepler's Laws

• We begin by discussing planetary motion using some of our properties of vectors and vector elds. The main
tool we will use for our analysis is Newton's law of gravitation, which says that the gravitational attraction
imparted by an object on a particle is directly proportional to each of their mass and inversely proportional
to the square of the distance between them.

◦ The simplest situation is with an object and a particle. Explicitly, suppose the particle has mass m at
r = hx, y, zi, and the object has mass M and is located at the origin (0, 0, 0).
r
◦ The unit vector from r to the origin is − , and the magnitude of the eld F is equal to a constant times
||r||
1
m times M times . The constant of proportionality here is called G, the universal gravitational
||r||2
3
constant, and its value has been measured to be 6.674m /(kg·s2 ).
GmM  r  GmM
◦ Putting all of this together shows that F = 2
· − =− r, which we can explicitly write
||r|| ||r|| ||r||3
GmM
as a function of x, y, z : F(x, y, z) = − hx, y, zi.
(x + y 2 + z 2 )3/2
2

◦ If the object is located at a dierent point, we can simply translate the underlying coordinate sys-
tem: explicitly, if the object is instead located at (a, b, c), the gravitational eld would be F(x, y, z) =
GmM
− hx − a, y − b, z − ci.
((x − a)2 + (y − b)2 + (z − c)2 )3/2
◦ If we have several objects, we then sum all of their contributions to the graviational eld. For example,
if we had an object of mass M1 at (0, 0, 0) and another object of mass M2 at (2, 0, 1), then the resulting
GmM1 GmM
eld would be F(x, y, z) = − hx, y, zi − hx − 2, y, z − 1i.
(x2 + y 2 + z 2 )3/2 ((x − 2)2 + y 2 + (z − 1)2 )3/2

• Once we have written down the gravitational eld, we can use Newton's second law (F = ma) to write down
equations of motion for objects passing through the gravitational eld.

◦ Explicitly, suppose that the path r describes the motion of a particle (a planet of mass m) through space,
and the only force acting on the planet is the gravity of the sun (mass M ) at the origin.
GmM
◦ Then the gravitational force is F(r) = − r, and so by Newton's second law, we obtain the equation
||r||3
GM GM
a=− 3
r. Since a = r00 , we obtain the equation r00 (t) = − 3 r(t).
||r|| ||r(t)||
◦ This is a second-order dierential equation for r which (in general) is quite dicult to solve explicitly
even in simple cases. We could write it out explicitly in terms of the components of r, which would yield
a system of three second-order dierential equations for the components x, y , and z , but that does not
really make the system any easier to solve.

◦ Rather than attempting to solve the system, we will make a few observations about the solutions.

• First, we claim that the planet's orbit lies in a plane passing through the sun.

GM
◦ To see this, observe that our equation a=− r tells us that a is a scalar multiple of r.
||r||3
◦ Now consider the vector n = r × v. Using the product rule for cross products, we see that the derivative
d
of this vector is [r × v] = v × v + r × a. But both terms in this sum are zero, because the cross product
dt
of any vector with itself is zero, and a is a scalar multiple of r.

27
 d
◦ Thus, [r × v] is zero, and so n=r×v is a constant vector.
dt
◦ But this means that the position and velocity of the particle both lie in the plane passing through its
starting position whose normal vector is n, and by our relation, all of the higher derivatives of the
particle's position will also lie in this plane. This means that the particle's motion will stay in the plane
as time moves forward.

• By extending this sort of analysis, one may derive Kepler's famous laws of planetary motion:

1. The orbit of a planet is a conic section with the sun at one focus. Specically, the conic's eccentricity is
r0 v02
e= GM − 1.
2. The radius vector r from the sun to the planet sweeps out equal areas in equal times.

3. The square of the orbital period T is proportional to the cube of the length of the semimajor axis a.
T2 4π 2
Specically,
a3 = GM .

• The best approach to establishing Kepler's laws is to work in polar coordinates. This approach is feasible
because, as we have just shown, the orbit of a planet lies in a plane passing through the sun.

◦ For notational convenience we will use dots to denote time-derivatives.

◦ Place the sun at the origin and dene r = ||r|| to be the radial parameter, with angle parameter θ,
and also dene the unit vector ur = hcos θ, sin θ, 0i in the direction from the sun to the planet and its
orthogonal unit vector uθ = h− sin θ, cos θ, 0i, which is the unit vector in the direction of increasing θ .

dur duθ dur dθ duθ dθ

◦ We see = uθ and = −ur , so the chain rule gives u̇r = = θ̇uθ and u̇θ = = −θ̇ur .
dθ dθ dθ dt dθ dt
dr dur
◦ Then because r = rur , by the product rule we get v = ṙ = ur + r = ṙur + rθ̇uθ .
dt dt
◦ We can then derive the second law by calculating the vector D n=r× E v from earlier: explicitly, we have
2 2
n = (rur ) × (ṙur + rθ̇uθ ) = rṙ(ur × ur ) + r θ̇(ur × uθ ) = 0, 0, r θ̇ .
◦ Now we can compute the area swept out by the radius vector between time t = t1 and time t = t2 .
Using integration in polar coordinates and then a substitution in the resulting line integral, this is
´ θ2 (t) ´ r(t) ´ θ2 (t) 1 2 ´t 1
θ1 (t) 0
1 · r dr dθ = θ1 (t) 2
r dθ = t12 r(t)2 θ̇(t) dt.
2
◦ However, the integrand in the last integral is exactly the z -component of the constant vector n, so the
integral is simply (t2 − t1 ) times a constant. This means the area depends only on the amount of time
t2 − t1 , which (when phrased more elegantly) is Kepler's second law above.

◦ The other laws are a bit more dicult and require careful manipulation of the dierential equation
r00 (t) = − GmM
||r||3 r.
◦ We will not go through the details of these other than to say that the rst law boils down to computing
(1 + e)r0
the polar equation for r in terms of θ, and verifying it has the form r= where r0 is the radius
1 + e cos θ
at perihelion (i.e., the minimal radius), which we take to occur at time t = 0 and θ = 0, while the third
law boils down to comparing two formulas for the area of an ellipse (one of them is the integral formula
above, and the other is π times the semimajor axis times the semiminor axis, which can be derived using
a change of variables in a double integral).

GmM GmM
• We can also calculate the work done by a gravitational eld F(r) = − r=− 2 hx, y, zi.
||r||3 (x + y 2 + z 2 )3/2

◦ Calculating the work integral as a line integral directly is quite messy because of the square root factor
in the denominator.

◦ But, as our physical intuition would suggest, and as we can check explicitly by computing curl F =
h0, 0, 0i, the gravitational eld F is conservative.

◦ If we search for a potential function, we can eventually see that U = GmM (x2 + y 2 + z 2 )−1/2 has the
property that F = hUx , Uy , Uz i, as the chain rule terms 2x, 2y , 2z exactly give the needed factors of
x, y, z in the three components.

28
 ◦ We can therefore easily compute the work done by the vector eld on a particle that travels from point
a to point b: specically, by the fundamental theorem of line integrals, the work done by F is equal to
GmM GmM
U (b) − U (a) = − .
||b|| ||a||
◦ Notice that the work only depends on the distances of the points from the origin, so the work done by
gravity will be the same regardless of the relative positions of the start and end points on the spheres of
constant radius centered at the origin.

• We can even apply this formula to compute the approximate graviational potential on the surface of the earth.

◦ If the start and end points are both approximately a distance R from the origin, then we can estimate
the change in the potential energy using a linearization (or, equivalently, a directional derivative).

GmM GmM
◦ It is straightforward to compute that the linearization of U (x) = at x=R is L(x) = −
x R
GmM
(x − R).
R2
GmM
◦ Therefore, the approximate value of U (R + ∆h) − U (R)
− ∆h, which equals mg∆h where m is
is
R2
GM
the mass of the particle, h is the change in height, and g = is a constant.
R2
◦ If we evaluate this constant g using the known values G = 6.674m3 /(kg · s2 ), the mass of the Earth
M = 5.972 · 1024 kg, and the radius of the Earth R = 6.371 · 106 m, we do in fact obtain the local
2
gravitational constant g = 9.817m/s .

◦ Of course, this should not be very surprising, because by Newton's second law, the magnitude of the
acceleration due to the gravitational eld will be ||F(r)||/m = −GM/||r||2 .

4.7.2 The Heat and Wave Equations

• We can model a great deal of physical phenomena using fairly simple models, and we can use vector calculus
to study the behaviors of the models quite fruitfully in many cases.

◦ Here, we will discuss two very fundamental models: the heat equation and the wave equation.

• The heat equation in its general form is the partial dierential equation ft = γ∇ · (∇f ).

◦ For a function f (x, y, z, t), the heat equation reads as ft = γ(fxx + fyy + fzz ) in standard notation.

◦ The function f (x, y, z, t) models the temperature of an object at a point (x, y, z) at time t, and γ is a
rate constant.

◦ For shorthand (even though it is technically bad notation) we often write the operator ∇·∇ as ∇2 =
2 2 2
∂ ∂ ∂
h ∂x2 , ∂y 2 , ∂z 2 i. This operator is called the Laplacian and is often also written as ∆.
◦ The heat equation is a rephrasing of the second law of thermodynamics and Newton's law of cooling
(heat ows from hot things to cold ones at a rate proportional to the dierence in temperatures).

• In fact, we can derive the heat equation ft = γ∇2 f from these two principles using the divergence theorem.
˚
◦ Explicitly, let H(t) be the amount of heat contained in a region D. Then H(t) = αf (x, y, z, t) dV
D
since temperature is a measure of heat density.
˚ 
d
◦ Taking the derivative with respect to t of both sides yields Ht (t) = αf (x, y, z, t) dV =
˚ dt D

αft (x, y, z, t) dV , where the second equality follows from a general theorem of Leibniz on inter-
D
changing the order of a derivative and an integral (it is, essentially, a combination of the mixed-partials
theorem and Fubini's theorem, and it allows us to move the t-derivative inside the integral).

◦ Since heat can only ow into the solid region D across its boundary, the heat ow Ht (t) is also given by
computing the ux of the heat owing through the boundary of the surface.

29
 ¨
◦ ∇f , so the ux of this eld is the surface integral
The vector eld modeling the heat ow is β(∇f ) · n dσ .
S
˚ ˚
◦ By the divergence theorem, the surface integral is also equal to β∇ · (∇f ) dV = β∇2 f dV .
D D
˚ ˚
◦ By comparing to the triple-integral expression from earlier, we obtain an equality αft dV = β∇2 f dV
D D
on every solid D. The only way this can happen is if the underlying functions αftand β∇2 f are equal
everywhere.

◦ Moving the constant factors around to solve for ft immediately yields the heat equation ft = γ∇2 f .

• The heat equation ft = γ∇2 f , as an abstract dierential equation, also shows up in various other places.

◦ In probability theory, the heat equation shows up as a very natural continuous model for random walks.
In physics, this is closely connected with the study of Brownian motion.

◦ In nancial mathematics, the Black-Scholes equation (which is used for computing the proper price of
options) is a minor variation of the heat equation: if V is the price of an option as a function of the asset
S and time t, then it says Vt + 21 σ 2 S 2 VSS + rSVS − rV = 0. Up to the coecients and the rst-order
and constant terms, it is essentially Vt = γVSS , which is a one-dimensional heat equation.
∂
◦ In quantum mechanics, Schrödinger's equation reads as H|ψ(t)i = ih ∂t |ψ(t)i. For a single particle, H
2
is (essentially) the Laplacian operator ∇ , so this is (very roughly!) a heat equation with an imaginary
constant factor.

• Our other fundamental dierential equation is the wave equation: ftt = γ∇2 f .

◦ For a function f (x, y, z, t), the wave equation reads as ftt = γ(fxx + fyy + fzz ) in standard notation.

◦ Note the similarity to the heat equation: the only dierence is that the wave equation has a double
t-derivative rather than the single t-derivative as in the heat equation.

◦ The wave equation, as one would expect from its name, models the intensity of a wave at a point (x, y, z)
in space at time t.

• Pleasantly enough, the one-dimensional wave equation ftt = c2 fxx can actually be solved essentially explicitly.

◦ The idea is to change variables and write a = x − ct and b = x + ct.

◦ Then by two applications of the multivariable chain rule, we can verify that the wave equation is equivalent
to fab = 0, which by antidierentiating twice can be seen to have the simple solution f (a, b) = F (a)+G(b)
for arbitrary functions F and G.
◦ Plugging back in yields a general solution f (x, t) = F (x−ct)+G(x+ct). This is the sum of a left-moving
function F (x − ct) and a right-moving function G(x + ct) as t increases.
◦ Below is plotted a solution to the one-dimensional wave equation for a xed value of t. As t changes, the
positions of the wave peaks will move at constant speed, one to the left and the other to the right:

• Like the heat equation, the wave equation can also be derived from basic physical principles using the diver-
gence theorem.

◦ Specically,
˝ suppose D is any
˝ region. Then the acceleration within D is the second t-derivative of

D
f dV , which is f dV .
D tt

30
 ◦ The vector eld F modeling the force imparted by the wave is
˜ ∇f , and so the total˝force acting on D
through its boundary S is equal to the surface integral
S
(∇f ) · n dσ , which equals D ∇2 f dV by the
divergence theorem.

◦ Applying Newton's second law (F = ma) and equating the two triple integrals on every D, using the
same logic as we used earlier for the heat equation, then gives the wave equation: ftt = γ∇2 f .

4.7.3 Numerical Methods in Modeling Applications

• Although we are able to write down the solutions to the one-dimensional wave equation in a convenient way,
in most cases, the dierential equation (or equations) modeling a physical phenomenon are dicult if not
impossible to solve exactly. As such, we often want to nd methods of generating good approximate solutions.

◦ One approach is to employ a step method and linearization: we take a linearization of the system and
then move a small step forward in time (the idea being that for a small step, the linearization is a good
approximation of the original).

◦ We then iterate this procedure with the new system that has been moved forward: we linearize and then
move a small step forward in time, repeatedly.

◦ Techniques like this one can be used to analyze models for weather and climate, urban planning, epi-
demiology (e.g., during global pandemics), ecology, experimental biology, chemistry, and physics, and
just about everywhere else....

• As part of these numerical methods, one often needs to search for a minimum or maximum value of some
function.

◦ For example, if one wants to model a chemical reaction computationally (which is now possible to do
with modern supercomputers), one needs to compute minimum-energy congurations of molecules.

◦ To perform such simulations, the computer must use step methods to iterate each interaction of particles
in small time intervals, and search for the minimum-energy state.

◦ To nd such a state, one may use a gradient-step method: compute the current energy, and then step
in the opposite direction of the gradient of this energy function.

◦ As we have discussed, the gradient points in the direction of maximum increase of a function, so at each
stage, the search will move in the direction that lowers total energy.

◦ Eventually, a gradient-step algorithm will reach a state in which the gradient is zero, which is a critical
point of the energy function.

◦ To determine whether the energy is actually minimized then requires classifying the resulting critical
point as a local minimum, local maximum, or saddle point.

◦ Of course, in actual practice, the search space is much larger than the 2-dimensional examples we have
discussed (typically it has hundreds or thousands of variables).

◦ But the general principle, that one may classify the type of critical point by using a higher-dimensional
version of the second derivatives test, turns out to be very similar.

• In many other applications, we have a model that we want to t to a given data set.

◦ In statistics there are various methods for making parameter estimates of this type: indeed, a major
component of statistics is about developing methods for making parameter estimates from data seta.

• A computationally convenient technique, frequently used in practice, is to employ a least-squares regression:

we minimize the sum of the squared errors between the predicted and observed values.

◦ The reason to use the sum of squares, rather than something else like the sum of the absolute errors, is
because we can minimize the resulting function using calculus.

◦ The simplest example of this kind is to t a linear function y = ax+b to a data set {(x1 , y1 ), (x2 , y2 ), . . . , (xn , yn )}.
◦ The function to minimize for the linear model above is E(a, b) = (ax1 + b − y1 )2 + (ax2 + b − y2 )2 + · · · +
2
(axn + b − yn ) .

31
 ◦ To minimize this function we set the two partial derivatives ∂E/∂a and ∂E/∂b equal to zero.

◦ We have ∂E/∂a = 2x1 (ax1 + b − y1 ) + · · · + 2xn (axn + b − yn ) and ∂E/∂b = 2(ax1 + b − y1 ) + · · · +

2(axn + b − yn ), so that a x2i + b xi =
P P P P P
xi yi and a xi + nb = yi .
P P P
n P xi yi − Pxi yi
◦ This is a linear system for a and b, with solution a = n x2 −( xi )2 and b = n1 ( yi − a xi ).
P P
i

◦ These two values a and b together give the equation for the famous least-squares regression line to a data
set.

◦ Here, for example, is the plot of a data set {(9, 24), (15, 45), (21, 49), (25, 55), (30, 60)} along with its
least-squares regression line y = 1.599x + 14.615 :

• However, the method of least squares is quite robust and can be used with many dierent kinds of models.

◦ For example, we can use more complicated functions, such as a quadratic function y = ax2 + bx + c.
◦ The procedure is essentially the same as before: we write down the sum of squared errors and then
minimize it using calculus, by setting all of the partial derivatives equal to zero.

◦ Here, the function is E(a, b, c) = (ax21 + bx1 + c − y1 )2 + · · · + (ax2n + bxn + c − yn )2 .

◦ We then calculate ∂E/∂a, ∂E/∂b, and ∂E/∂c and set them equal to zero. The resulting system will be
linear in a, b, c and we can then solve it to compute the predicted coecients a, b, c.
◦ Here, for example, is the plot of a data set {(−2, 19), (−1, 7), (0, 4), (1, 2), (2, 7)} along with the parabola
y = −2.5x2 − 2.9x + 2.8 of best t:

32
4.7.4 Maxwell's Equations and Electromagnetism

• We will now give a brief discussion of Maxwell's equations of electromagnetism.

◦ Here, E is the electric eld, B is the magnetic eld, ρ is electric charge density, and 0 and µ0 are
constants. (We assume no current J here.)

• Here are Maxwell's equations:

Law Integral Form Dierential Form

‹ ˚
1 ρ
Gauss (E) E · n dσ = ρ dV ∇·E=
S 0 D 0
‹
Gauss (M) B · n dσ = 0 ∇·B=0
˛ ¨ S 
d ∂B
Maxwell-Faraday E · T ds = − B · n dσ ∇×E=−
C dt Σ ∂t
˛ ¨ 
d ∂E
Ampère B · T ds = µ0 0 E · n dσ ∇ × B = µ0 0
C dt Σ ∂t

◦ In the two Gauss laws, S is a closed surface enclosing the solid region D, while in the other two laws, Σ
is a surface with counterclockwise boundary curve C.

• Notice that each law comes in two dierent forms: an integral form and a dierential form. We may convert
between these dierent forms using the divergence theorem and Stokes's theorem.

◦ Explicitly, in the two Gauss laws, S is a closed surface enclosing the solid region D, so if we apply the
divergence theorem, we may convert the surface integral into a triple integral.
‚ ˝
◦ For the electric eld law, by the divergence theorem we have
˝ ˝ S
E · n dσ = D
(∇ · E) dV , so the integral
1
form is equivalent to saying
D
(∇ · E) dV = 0 D
ρ dV .
◦ This equality holds on every solid region D, so the integrands ∇·E and ρ/0 are equal: this is the
dierential form. A similar argument works for the magnetic eld law.

◦ Σ is a surface with counterclockwise boundary C , so we can apply Stokes's theorem.

In the other two laws,
¸ ˜
◦ For the Maxwell-Faraday law, by Stokes's theorem the integral C E · T ds equals Σ (∇ × E) · n dσ .
˜ d
˜ ˜
◦ Thus, the integral form is equivalent to Σ (∇ × E) · n dσ = − dt Σ
B · n dσ = − Σ ∂B ∂t · n dσ .
◦ Since this holds on every surface, the two elds ∇×E and − ∂B
∂t must be equal, giving the dierential
form. A similar argument yields the two versions of Ampère's law.

• We can actually derive Gauss's law for both electricity and magnetism as a consequence of more general
properties of inverse-square laws.

◦ Coulomb's law says that the electric force between two particles is proportional to each of their charges
and inversely proportional to the square of the distance between them. (Compare this to Newton's law
of gravitation, which has the same form of inverse-square law.)

◦ Via the same sort of analysis we did earlier in nding the gravitational eld, we can see that, for a single
q r
point charge q at the origin, the electric eld E equals E(r) = .
4π0 ||r||3
◦ We can then compute the surface integral through the sphere of radius a centered at the origin directly.

◦ We can see that the unit normal vector is n = r/a, and so the surface integral in spherical coordinates is
ˆ 2π ˆ π ˆ 2π ˆ π
q r r q q
a2
· dϕdθ = a2
3 a
dϕdθ = , because the dot product r · r = a2 on the
0 0 4π0 a 0 0 4a2 π0 0
sphere of radius a.

◦ This agrees with the triple integral of Gauss's law for the case of a single particle (the triple integral is
simply q/0 ).

33
 ◦ This may seem like a very special case of Gauss's law, but we can actually use it to get the general
version. First, we need to extend Gauss's law to arbitrary surfaces.

◦ So suppose we have an arbitrary closed surface T containing the origin. Choose a sphere S that encloses
it and take D to be the region between the two surfaces.
˝ ‚ ‚
◦ Then, by the divergence theorem, we see that
D
(∇ · E) dV = S
E · n dσ − T
E · n dσ (the minus
sign is because the normal vector for T points inward).

q hx, y, zii
◦ For E(x, y, z) = , ∇ · E = 0 for (x, y, z) 6= (0, 0, 0).
we can compute explicitly that
4π0 (x2 + y 2 + z 2 )3/2 ‚ ‚
Since D does not contain the origin, the triple integral is zero, and so
S
E · n dσ = T E · n dσ .
◦ This means that the Gauss law result holds for a single particle and an arbitrary surface T .
◦ Finally, we can use the fact that Gauss's law holds for a single particle and an arbitrary surface to obtain
the result for arbitrary charge distributions and arbitrary surfaces.

◦ The idea is simply to sum over all of the various charges, and observe that both the surface integral and
the triple integral are consistent with summing over charges.

◦ By taking a limit of nite sums of charges, we obtain the result for arbitrary charge distributions, and
so we obtain Gauss's law for electric elds.
‚
◦ For the Gauss law for magnetic elds
S
B · n dσ = 0, the result is quite a bit simpler: the point is that
there is no magnetic equivalent of charge (this would be a magnetic monopole, of which no experimental
observation has ever been made), and so the resulting triple integral of magnetic charge is simply zero.

• We can make a few additional observations about the constraints imposed by Maxwell's equations.

◦ Both E and B have 3 components. The two Gauss's laws each impose one condition on the components,
while the other two laws each impose three conditions. So we seemingly have 8 conditions on the 6
components.

◦ But in fact, there are two redundant conditions, which are accounted for by the div-curl identity, which
says div(curl(F)) = ∇ · (∇ × F) = 0 for any vector eld F.
◦ So in fact, there are six conditions on the six components, which is (in an appropriate sense) exactly
enough to determine them. (In fact, it is a theorem of electrodynamics that the behavior of electric
and magnetic elds is completely determined by Maxwell's equations, in the sense that there are no
additional hidden constraints.)

• As a nal remark, we will observe a connection between Maxwell's equations and electromagnetic waves.

◦ Suppose that the charge density q is zero everywhere: then Gauss's law for the electric eld says that
div(E) = 0.
◦ For any vector eld F, we have the curl-curl identity curl(curl(F)) = grad(div(F)) − ∇2 · F, which is
not hard to verify just by writing it out.

◦ E yields ∇×(∇×E) = grad(div(E))−∇2 ·E = −∇2 ·E since div(E) = 0.

Applying this to the vector eld
∂B ∂E
◦ By Maxwell's equations we have ∇ × E = − , ∇ × B = µ0 0 , so by the equality of mixed partials
∂t  ∂t
∂B ∂ ∂ ∂E
we see ∇ × (∇ × E) = ∇ × − = − [∇ × B] = − µ0 0 = −µ0 0 Ett
∂t ∂t ∂t ∂t
◦ Putting all of this together with the calculation above yields −∇2 · E = −µ0 0 Ett , which means that E
satises the wave equation! (Likewise for B.)

◦ We see, therefore, that Maxwell's equations lead directly to the phenomenon of electromagnetic waves.
◦ Of course, electromagnetic waves are a quite well-understood concept in the 21st century, so it is quite
important to note that Maxwell published his original papers detailing these equations, and deducing
some of these consequences to unify electricity and magnetism, in 1861.

Well, you're at the end of my handout. Hope it was helpful.

Copyright notice: This material is copyright Evan Dummit, 2012-2021. You may not reproduce or distribute this
material without my express permission.

34