Directional Derivatives

The Question

Suppose that you leave the point (a, b) moving with velocity ~v = hv1 , v2 i. Suppose further
that the temperature at (x, y) is f (x, y). Then what rate of change of temperature do you feel?

The Answers

Let’s set the beginning of time, t = 0, to the time at which you leave (a, b). Then at time t you
are at (a + v1 t, b + v2 t) and feel the temperature f (a + v1 t, b + v2 t). So the change in temperature
between time 0 and time t is f (a + v1 t, b + v2 t) − f (a, b), the average rate of change of temperature,
per unit time, between time 0 and time t is f (a+v1 t,b+v t
2 t)−f (a,b)
and the instantaneous rate of
f (a+v1 t,b+v2 t)−f (a,b)
change of temperature per unit time as you leave (a, b) is limt→0 t . We apply
the approximation

f (a + ∆x, b + ∆y) − f (a, b) ≈ fx (a, b) ∆x + fy (a, b) ∆y

with ∆x = v1 t and ∆y = v2 t. In the limit as t → 0, the approximation becomes exact and we have
f (a+v1 t,b+v2 t)−f (a,b) fx (a,b) v1 t+fy (a,b) v2 t
lim t = lim t
t→0 t→0
= fx (a, b) v1 + fy (a, b) v2
= hfx (a, b), fy (a, b)i · hv1 , v2 i

The vector hfx (a, b), fy (a, b)i is denoted ∇f ~ (a, b) and is called “the gradient of the function
f at the point (a, b)”. It has one component for each variable of f . The j th component is
the partial derivative of f with respect to the j th variable, evaluated at (a, b). The expression
~ (a, b) · ~v is often denoted D~v f (a, b). So we conclude that the rate
hfx (a, b), fy (a, b)i · hv1 , v2 i = ∇f
of change of f per unit time as we leave (a, b) moving with velocity ~v is
~ (a, b) · ~v
D~v f (a, b) = ∇f

We can compute the rate of change of temperature per unit distance in a similar way. The
change in temperature between time 0 and time t is f (a + v1 t, b + v2 t) − f (a, b). Between time 0
and time t you have travelled a distance |~v |t. So the instantaneous rate of change of temperature
per unit distance as you leave (a, b) is
f (a+v1 t,b+v2 t)−f (a,b)
lim v|
t|~
t→0

This is exactly |~v1| times limt→0 f (a+v1 t,b+v

t
2 t)−f (a,b)
which we computed above to be D~v f (a, b). So
the rate of change of f per unit distance as we leave (a, b) moving in direction ~v is
~ (a, b) ·
∇f ~
v
=D ~
v f (a, b)
|~
v| |~
v|

This is called the directional derivative of the function f at the point (a, b) in the direction ~v .

c Joel Feldman. 2004. All rights reserved. 1

Implications

We have just seen that the instantaneous rate of change of f per unit distance as we leave
(a, b) moving in direction ~v is

~ (a, b) ·
∇f ~
v ~ (a, b)| cos θ
= |∇f
|~
v|

where θ is the angle between the gradient vector ∇f ~ (a, b) and the direction vector ~v . Since cos θ
is always between −1 and +1
◦ the direction of maximum rate of increase is that having θ = 0. So to get maximum rate of
increase per unit distance, as you leave (a, b), you should move in the same direction as the
~ (a, b). Then the rate of increase per unit distance is |∇f
gradient ∇f ~ (a, b)|.
◦ The direction of minimum (i.e. most negative) rate of increase is that having θ = 180◦ . To
get minimum rate of increase per unit distance you should move in the direction opposite
~ (a, b). Then the rate of increase per unit distance is −|∇f
∇f ~ (a, b)|.
◦ The directions giving zero rate of increase are those perpendicular to ∇f ~ (a, b). If you move
in a direction perpendicular to ∇f~ (a, b), f (x, y) remains constant as you leave (a, b). That
is, at that instant you are moving along the level curve f (x, y) = f (a, b). So ∇f ~ (a, b) is
perpendicular to the level curve f (x, y) = f (a, b). We have already seen the corresponding
statement in three dimensions. A good way to find a vector normal to the surface F (x, y, z) = 0
~ (a, b, c).
at the point (a, b, c) is to compute the gradient ∇F

An example

Let
f (x, y) = 5 − x2 − 2y 2


(x0 , y0 ) = − 1, −1
2 2
Note that for any fixed z0 < 5, f (x,qy) = z0 is the ellipse x + 2y = 5 − z0 . This ellipse has x–
√
semi–axis 5 − z0 and y–semi–axis 5−z 2
0
. The graph z = f (x, y) consists of a bunch of horizontal
ellipses stacked one on top of each other, starting with a point on the z axis when z0 = 5 and
increasing in size as z0 decreases, as illustrated in the first figure below. Several level curves are
sketched in the second figure below. The gradient vector


∇f (x0 , y0 ) = − 2x, −4y (−1,−1)
= (2, 4) = 2(1, 2)

at (x0 , y0 ) is also illustrated in the second sketch. We have that, at (x0 , y0 )

◦ the direction of maximum rate of increase is √15 (1, 2) and the maximum rate of increase is
√
|(2, 4)| = 2 5.
◦ the direction of minimum rate of increase is − √15 (1, 2) and that minimum rate is −|(2, 4)| =
√
−2 5.
◦ the directions giving zero rate of increase are perpendicular to ∇f (x0 , y0 ), that is ± √15 (2, −1).
These are the directions of the tangent vector at (x0 , y0 ) to the level curve of f through
(x0 , y0 ).

c Joel Feldman. 2004. All rights reserved. 2

 y
z

z=4 z=0

z=3
z=4
z=2 x

z=1
z=0y

(−1, −1)

c Joel Feldman. 2004. All rights reserved. 3