MA 105 : Calculus

D1 - Lecture 15

Sandip Singh

Department of Mathematics

Autumn 2023, IIT Bombay, Mumbai

Directional Derivatives

The partial derivatives are special cases of the directional

derivative. Let v = (v1 , v2 ) be a unit vector. Then v specifies a
direction in R2 .
Definition: The directional derivative of f in the direction v at a
point x = (x1 , x2 ) is defined as
f (x + tv ) − f (x) f ((x1 + tv1 , x2 + tv2 )) − f ((x1 , x2 ))
∇v f (x) = lim = lim .
t→0 t t→0 t
It measures the rate of change of the function f at x along the
path x + tv .
Observe that if we take v = (1, 0) in the above definition, we
obtain ∂f /∂x1 , while v = (0, 1) yields ∂f /∂x2 .
Consider the function
(
1 if x1 = 0 or if x2 = 0
f (x1 , x2 ) =
0 otherwise.

It should be clear to you that since this function is constant along

the two axes,
∂f ∂f
(0, 0) = 0 and (0, 0) = 0.
∂x1 ∂x2
On the other hand, f (x1 , x2 ) is not continuous at the origin.
(why?)
Thus, a function may have both partial derivatives (and, in fact,
any directional derivative - see the next slide) but still not be
continuous.
This suggests that for a function of two variables, just requiring
that both partial derivatives exist is not a good or useful definition
of “differentiability” (why?).
In the section on iterated limits, we studied the following function
from Exercise 5.5:

x 2y 2
f (x, y ) = for (x, y ) 6= (0, 0).
x 2 y 2 + (x − y )2
Let us further set f (0, 0) = 0. You can check that every directional
derivative of f at (0, 0) exists and is equalto 0, except
 along
1 √1
y = x (that is, along the unit vector v = √
2
, 2 ) when the
directional derivative is not defined.
However, we have already seen that the function is not continuous
at the origin since we have shown that lim(x,y )→(0,0) f (x, y ) does
not exist. For an example with directional derivatives in all
directions see Exercise 5.3(i).
Conclusion: All directional derivatives may exist at a point even if
the function is discontinuous at that point.
Let us go back and examine the notion of differentiability for a
function f (x) of one variable.
Suppose f is differentiable at the point x0 . When is a line passing
through the point (x0 , f (x0 )) on the curve y = f (x) is the tangent
line to the curve?
Recall that the equation of a line passing through the point
(x0 , f (x0 )) and having slope m is
y = g (x) = f (x0 ) + m(x − x0 ).
If we consider the difference
f (x) − g (x) = f (x) − f (x0 ) − m(x − x0 ) and write h = (x − x0 ),
we see that the difference can be rewritten as
f (x0 + h) − f (x0 ) − m · h.
The above line is the tangent line to the curve y = f (x) at the
point (x0 , f (x0 )) on the curve if m = f 0 (x0 ), which is equivalent of
saying that the above difference
f (x0 + h) − f (x0 ) − m · h = o(h) = ε1 (h)h
where ε1 (h) is a function of h that goes to 0 as h goes to 0, and in
this case the function o(h) = ε1 (h)h is a function of h “that goes
to zero faster than h” (that is, limh→0 o(h)
h = 0).
The preceding idea generalises to two (or more) dimensions.
Let f (x, y ) be a function which has both partial derivatives. In the
two variable case we will consider the difference of z = f (x, y )
(defining a surface in R3 ) and z = g (x, y ) (defining a plane in R3 ).
Let us first recall how to find the equation of a plane passing
through the point P = (x0 , y0 , z0 ).
It is the graph of the function

z = g (x, y ) = z0 + a(x − x0 ) + b(y − y0 ).

Now, we determine the values of a and b, for which, the above

plane z = g (x, y ), passing through the point P = (x0 , y0 , z0 ) on
the surface z = f (x, y ), is the tangent plane to the surface.
If we fix the y variable as y = y0 and treat f (x, y ) only as a
function of x, we get a curve on the surface z = f (x, y ).
Similarly, if we treat g (x, y ) as function only of x (by fixing
y = y0 ), we obtain a line on the plane z = g (x, y ).
The tangent plane in a picture

[Link]
4-4-tangent-planes-and-linear-approximations
The tangent to the curve passing through (x0 , y0 , z0 ) must be the
same as the line passing through (x0 , y0 , z0 ), and, in any event,
their slopes (which are given by the derivatives of the curve
z = f (x, y0 ) at x = x0 and the line z = g (x, y0 ) at x = x0 , resp.)
must be the same.
Since the above derivatives with respect to x are same as the
partial derivatives with respect to x, we get
dz ∂f ∂g
(x0 ) = (x0 , y0 ) = (x0 , y0 ) = a.
dx ∂x ∂x
Arguing in exactly the same way but fixing the x variable as x = x0
and varying the y variable, we obtain
∂f ∂g
(x0 , y0 ) = (x0 , y0 ) = b.
∂y ∂y
Hence, the equation of the tangent plane to the surface
z = f (x, y ) at the point (x0 , y0 ) is (remember that z0 = f (x0 , y0 ))
∂f ∂f
z = f (x0 , y0 ) + (x0 , y0 )(x − x0 ) + (x0 , y0 )(y − y0 ).
∂x ∂y
Differentiability for functions of two variables
We now define differentiability for functions of two variables by
imitating the one variable definition, but using the “o(h)” version.
We let (x, y ) = (x0 , y0 ) + (h, k) = (x0 + h, y0 + k).
Definition A function f : U → R is said to be differentiable at a
∂f ∂f
point (x0 , y0 ) if ∂x (x0 , y0 ) and ∂y (x0 , y0 ) exist and
∂f ∂f
|f (x0 + h, y0 + k) − f (x0 , y0 ) − ∂x (x0 , y0 )h − ∂y (x0 , y0 )k|
lim = 0.
(h,k)→(0,0) k(h, k)k
We could rewrite this as
∂f ∂f
f ((x0 , y0 ) + (h, k)) − f (x0 , y0 ) − (x0 , y0 )h − (x0 , y0 )k
∂x ∂y
= ε(h, k)k(h, k)k
where ε(h, k) is a function that goes to 0 as (h, k) → (0, 0).
This form of differentiability now looks exactly like the one variable
version case (put o(h, k) = ε(h, k)k(h, k)k). Can you guess the
derivative f 0 (x0 , y0 ) of the function f (x, y ) at (x0 , y0 )?