1.

2 Differential Calculus 13

1.2 DIFFERENTIAL CALCULUS

1.2.1 “Ordinary” Derivatives

Suppose we have a function of one variable: f (x). Question: What does the
derivative, d f /d x, do for us? Answer: It tells us how rapidly the function f (x)
varies when we change the argument x by a tiny amount, d x:
 
df
df = d x. (1.33)
dx

In words: If we increment x by an infinitesimal amount d x, then f changes

by an amount d f ; the derivative is the proportionality factor. For example, in
Fig. 1.17(a), the function varies slowly with x, and the derivative is correspond-
ingly small. In Fig. 1.17(b), f increases rapidly with x, and the derivative is large,
as you move away from x = 0.
Geometrical Interpretation: The derivative d f /d x is the slope of the graph of
f versus x.

1.2.2 Gradient
Suppose, now, that we have a function of three variables—say, the temperature
T (x, y, z) in this room. (Start out in one corner, and set up a system of axes; then
for each point (x, y, z) in the room, T gives the temperature at that spot.) We want
to generalize the notion of “derivative” to functions like T , which depend not on
one but on three variables.
A derivative is supposed to tell us how fast the function varies, if we move a
little distance. But this time the situation is more complicated, because it depends
on what direction we move: If we go straight up, then the temperature will prob-
ably increase fairly rapidly, but if we move horizontally, it may not change much
at all. In fact, the question “How fast does T vary?” has an infinite number of
answers, one for each direction we might choose to explore.
Fortunately, the problem is not as bad as it looks. A theorem on partial deriva-
tives states that
     
∂T ∂T ∂T
dT = dx + dy + dz. (1.34)
∂x ∂y ∂z

f f

x x
(a) (b)

FIGURE 1.17
14 Chapter 1 Vector Analysis

This tells us how T changes when we alter all three variables by the infinites-
imal amounts d x, dy, dz. Notice that we do not require an infinite number of
derivatives—three will suffice: the partial derivatives along each of the three co-
ordinate directions.
Equation 1.34 is reminiscent of a dot product:
 
∂T ∂T ∂T
dT = x̂ + ŷ + ẑ · (d x x̂ + dy ŷ + dz ẑ)
∂x ∂y ∂z
= (∇T ) · (dl), (1.35)
where
∂T ∂T ∂T
∇T ≡ x̂ + ŷ + ẑ (1.36)
∂x ∂y ∂z
is the gradient of T . Note that ∇T is a vector quantity, with three components;
it is the generalized derivative we have been looking for. Equation 1.35 is the
three-dimensional version of Eq. 1.33.
Geometrical Interpretation of the Gradient: Like any vector, the gradient has
magnitude and direction. To determine its geometrical meaning, let’s rewrite the
dot product (Eq. 1.35) using Eq. 1.1:
dT = ∇T · dl = |∇T ||dl| cos θ, (1.37)
where θ is the angle between ∇T and dl. Now, if we fix the magnitude |dl| and
search around in various directions (that is, vary θ ), the maximum change in T
evidentally occurs when θ = 0 (for then cos θ = 1). That is, for a fixed distance
|dl|, dT is greatest when I move in the same direction as ∇T . Thus:
The gradient ∇T points in the direction of maximum increase of the
function T .
Moreover:
The magnitude |∇T | gives the slope (rate of increase) along this
maximal direction.
Imagine you are standing on a hillside. Look all around you, and find the di-
rection of steepest ascent. That is the direction of the gradient. Now measure the
slope in that direction (rise over run). That is the magnitude of the gradient. (Here
the function we’re talking about is the height of the hill, and the coordinates it
depends on are positions—latitude and longitude, say. This function depends on
only two variables, not three, but the geometrical meaning of the gradient is easier
to grasp in two dimensions.) Notice from Eq. 1.37 that the direction of maximum
descent is opposite to the direction of maximum ascent, while at right angles
(θ = 90◦ ) the slope is zero (the gradient is perpendicular to the contour lines).
You can conceive of surfaces that do not have these properties, but they always
have “kinks” in them, and correspond to nondifferentiable functions.
What would it mean for the gradient to vanish? If ∇T = 0 at (x, y, z),
then dT = 0 for small displacements about the point (x, y, z). This is, then, a
stationary point of the function T (x, y, z). It could be a maximum (a summit),
 1.2 Differential Calculus 15

a minimum (a valley), a saddle point (a pass), or a “shoulder.” This is analogous

to the situation for functions of one variable, where a vanishing derivative signals
a maximum, a minimum, or an inflection. In particular, if you want to locate the
extrema of a function of three variables, set its gradient equal to zero.


Example 1.3. Find the gradient of r = x 2 + y 2 + z 2 (the magnitude of the
position vector).
Solution
∂r ∂r ∂r
∇r = x̂ + ŷ + ẑ
∂x ∂y ∂z
1 2x 1 2y 1 2z
=  x̂ +  ŷ +  ẑ
2 x 2 + y2 + z2 2 x 2 + y2 + z2 2 x 2 + y2 + z2
x x̂ + y ŷ + z ẑ r
= = = r̂.
x +y +z
2 2 2 r
Does this make sense? Well, it says that the distance from the origin increases
most rapidly in the radial direction, and that its rate of increase in that direction
is 1. . . just what you’d expect.

Problem 1.11 Find the gradients of the following functions:

(a) f (x, y, z) = x 2 + y 3 + z 4 .

(b) f (x, y, z) = x 2 y 3 z 4 .
(c) f (x, y, z) = e x sin(y) ln(z).

Problem 1.12 The height of a certain hill (in feet) is given by

h(x, y) = 10(2x y − 3x 2 − 4y 2 − 18x + 28y + 12),
where y is the distance (in miles) north, x the distance east of South Hadley.

(a) Where is the top of the hill located?

(b) How high is the hill?

(c) How steep is the slope (in feet per mile) at a point 1 mile north and one mile
east of South Hadley? In what direction is the slope steepest, at that point?

• Problem 1.13 Let r be the separation vector from a fixed point (x  , y  , z  ) to the
point (x, y, z), and let r be its length. Show that

(a) ∇(r2 ) = 2r.

(b) ∇(1/r) = −r̂/r2 .

(c) What is the general formula for ∇(rn )?
16 Chapter 1 Vector Analysis

! Problem 1.14 Suppose that f is a function of two variables (y and z) only.

Show that the gradient ∇ f = (∂ f /∂ y)ŷ + (∂ f /∂z)ẑ transforms as a vector un-
der rotations, Eq. 1.29. [Hint: (∂ f /∂ y) = (∂ f /∂ y)(∂ y/∂ y) + (∂ f /∂z)(∂z/∂ y),
and the analogous formula for ∂ f /∂z. We know that y = y cos φ + z sin φ and
z = −y sin φ + z cos φ; “solve” these equations for y and z (as functions of y
and z), and compute the needed derivatives ∂ y/∂ y, ∂z/∂ y, etc.]

1.2.3 The Del Operator

The gradient has the formal appearance of a vector, ∇, “multiplying” a scalar T :
 
∂ ∂ ∂
∇T = x̂ + ŷ + ẑ T. (1.38)
∂x ∂y ∂z
(For once, I write the unit vectors to the left, just so no one will think this means
∂ x̂/∂ x, and so on—which would be zero, since x̂ is constant.) The term in paren-
theses is called del:

∂ ∂ ∂
∇ = x̂ + ŷ + ẑ . (1.39)
∂x ∂y ∂z

Of course, del is not a vector, in the usual sense. Indeed, it doesn’t mean much
until we provide it with a function to act upon. Furthermore, it does not “multiply”
T ; rather, it is an instruction to differentiate what follows. To be precise, then, we
say that ∇ is a vector operator that acts upon T , not a vector that multiplies T .
With this qualification, though, ∇ mimics the behavior of an ordinary vector in
virtually every way; almost anything that can be done with other vectors can also
be done with ∇, if we merely translate “multiply” by “act upon.” So by all means
take the vector appearance of ∇ seriously: it is a marvelous piece of notational
simplification, as you will appreciate if you ever consult Maxwell’s original work
on electromagnetism, written without the benefit of ∇.
Now, an ordinary vector A can multiply in three ways:

1. By a scalar a : Aa;
2. By a vector B, via the dot product: A · B;
3. By a vector B via the cross product: A × B.

Correspondingly, there are three ways the operator ∇ can act:

1. On a scalar function T : ∇T (the gradient);

2. On a vector function v, via the dot product: ∇ · v (the divergence);
3. On a vector function v, via the cross product: ∇ × v (the curl).
 1.2 Differential Calculus 17

We have already discussed the gradient. In the following sections we examine the
other two vector derivatives: divergence and curl.

1.2.4 The Divergence

From the definition of ∇ we construct the divergence:
 
∂ ∂ ∂
∇ · v = x̂ + ŷ + ẑ · (vx x̂ + v y ŷ + vz ẑ)
∂x ∂y ∂z
∂vx ∂v y ∂vz
= + + . (1.40)
∂x ∂y ∂z

Observe that the divergence of a vector function6 v is itself a scalar ∇ · v.

Geometrical Interpretation: The name divergence is well chosen, for ∇ · v
is a measure of how much the vector v spreads out (diverges) from the point in
question. For example, the vector function in Fig. 1.18a has a large (positive)
divergence (if the arrows pointed in, it would be a negative divergence), the func-
tion in Fig. 1.18b has zero divergence, and the function in Fig. 1.18c again has a
positive divergence. (Please understand that v here is a function—there’s a differ-
ent vector associated with every point in space. In the diagrams, of course, I can
only draw the arrows at a few representative locations.)
Imagine standing at the edge of a pond. Sprinkle some sawdust or pine needles
on the surface. If the material spreads out, then you dropped it at a point of positive
divergence; if it collects together, you dropped it at a point of negative divergence.
(The vector function v in this model is the velocity of the water at the surface—
this is a two-dimensional example, but it helps give one a “feel” for what the
divergence means. A point of positive divergence is a source, or “faucet”; a point
of negative divergence is a sink, or “drain.”)

(a) (b) (c)

FIGURE 1.18

6 A vector function v(x, y, z) = v (x, y, z) x̂ + v (x, y, z) ŷ + v (x, y, z) ẑ is really three functions—

x y z
one for each component. There’s no such thing as the divergence of a scalar.
18 Chapter 1 Vector Analysis

Example 1.4. Suppose the functions in Fig. 1.18 are va = r = x x̂ + y ŷ + z ẑ,

vb = ẑ, and vc = z ẑ. Calculate their divergences.
Solution
∂ ∂ ∂
∇ · va = (x) + (y) + (z) = 1 + 1 + 1 = 3.
∂x ∂y ∂z
As anticipated, this function has a positive divergence.

∂ ∂ ∂
∇ · vb = (0) + (0) + (1) = 0 + 0 + 0 = 0,
∂x ∂y ∂z
as expected.
∂ ∂ ∂
∇ · vc = (0) + (0) + (z) = 0 + 0 + 1 = 1.
∂x ∂y ∂z

Problem 1.15 Calculate the divergence of the following vector functions:

(a) va = x 2 x̂ + 3x z 2 ŷ − 2x z ẑ.

(b) vb = x y x̂ + 2yz ŷ + 3zx ẑ.

(c) vc = y 2 x̂ + (2x y + z 2 ) ŷ + 2yz ẑ.

• Problem 1.16 Sketch the vector function

r̂
v= ,
r2
and compute its divergence. The answer may surprise you. . . can you explain it?

! Problem 1.17 In two dimensions, show that the divergence transforms as a scalar
under rotations. [Hint: Use Eq. 1.29 to determine v y and v z , and the method of
Prob. 1.14 to calculate the derivatives. Your aim is to show that ∂v y /∂ y + ∂v z /∂z =
∂v y /∂ y + ∂vz /∂z.]

1.2.5 The Curl

From the definition of ∇ we construct the curl:
 
 x̂ ŷ ẑ 

∇ × v =  ∂/∂ x ∂/∂ y ∂/∂z 
 vx vy vz 
     
∂vz ∂v y ∂vx ∂vz ∂v y ∂vx
= x̂ − + ŷ − + ẑ − . (1.41)
∂y ∂z ∂z ∂x ∂x ∂y