Contents

14 Partial Derivatives 2
14.1 Functions of Several Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
14.1.1 Functions of one real variable . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
14.1.2 Functions of two real variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
14.1.3 Neighbourhoods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
14.1.4 Interior and boundary points in XY-plane . . . . . . . . . . . . . . . . . . . 4
14.1.5 Level Curves, and Contours of Functions of Two Variables . . . . . . . . . 5
14.2 Interior and boundary points in XYZ-space . . . . . . . . . . . . . . . . . . . . . . 5
14.3 Limits and Continuity in Higher Dimensions . . . . . . . . . . . . . . . . . . . . . . 6
14.3.1 Limit of a function of one real variable . . . . . . . . . . . . . . . . . . . . . 6
14.3.2 Limit of a function of two real variables . . . . . . . . . . . . . . . . . . . . . 8
14.4 Continuity of a function of one real variable . . . . . . . . . . . . . . . . . . . . . . 11
14.4.1 Continuity of a function of two real variables . . . . . . . . . . . . . . . . . 12
14.5 Partial Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
14.5.1 Differentiability of a function of one real variable . . . . . . . . . . . . . . . 12
14.5.2 Differentiability of a function of two real variables . . . . . . . . . . . . . . 14
14.6 The Chain Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
14.6.1 Implicit Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
14.6.2 Higher Order Partial Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . 19
14.7 Directional Derivatives and Gradient Vectors . . . . . . . . . . . . . . . . . . . . . 21
14.8 Tangent Planes and Differentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
14.8.1 Estimating the change in f in a given direction . . . . . . . . . . . . . . . . 23
14.8.2 Taylor’s formula and linearization of f (x) . . . . . . . . . . . . . . . . . . . . 23
14.8.3 Taylor’s formula and linearization of f (x, y) . . . . . . . . . . . . . . . . . . . 24
14.8.4 Differentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
14.9 Extreme Values and Saddle Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
14.9.1 Maxima and minima of a function of one variable . . . . . . . . . . . . . . 26
14.9.2 First derivative test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
14.9.3 Second derivative test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
14.10Maxima and minima of a function of two variables . . . . . . . . . . . . . . . . . . 28
14.10.1 Test for maxima and minima of f (x, y) . . . . . . . . . . . . . . . . . . . . . . 29
14.11Lagrange Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
14.11.1 Constrained Maxima and Minima . . . . . . . . . . . . . . . . . . . . . . . . . 30
14.11.2 Method of Lagrange Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . 31
14.11.3 Lagrange Multipliers with Two Constraints . . . . . . . . . . . . . . . . . . 32
14.12Some useful tips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Note: Comments/suggestions on these lecture notes are welcome on the e-mail: [email protected] to
Dr. Suresh Kumar.

1
Chapter 14

Partial Derivatives

14.1 Functions of Several Variables

14.1.1 Functions of one real variable
Let A and B be subsets of R, the set of real numbers. Then a rule f that assigns to each member x ∈ A
a unique member y ∈ B, is called a function from A to B, written as f : A → B. The element y is called
image of x under f . It is also called as the value of f at x and is denoted by f (x), that is, y = f (x) while
x is called pre-image of y under f . The set A (the set of all pre-images) is called domain of f (denoted
by D(f )) while the set of images of all the elements of A under f , that is, R(f ) = {f (x) : x ∈ A} is called
range of f . We call the set B as co-domain of f . The set G(f ) = {(x, f (x)) : x ∈ A} is called graph
of f . Notice that the graph of f contains ordered pairs (x, f (x)) of real numbers, which can be shown
geometrically as points (x, f (x)) in the XY-plane. √
Ex. Consider
√ the function f : [−1, 1] → R defined by f (x)√= 1 − x2 . Then, we have D(f ) = [−1, 1],
R(f ) = { 1 − x : x ∈ [−1, 1]} = [0, 1] and G(f ) = {(x, 1 − x2 ) : x ∈ [−1, 1]}, which is displayed
2

geometrically in XY-plane in the left panel of Figure 1. So domain of f in the interval [−1, 1] on the
X-axis, and geometrically its graph is the semicircle.

Z
1.0

0.5
Y
-1

1.0

-1.0 - 0.5
0.0 0 Y
0.8 0.0
0.5
1.0
0.6
1
0.4

X
0.2

-1.0 - 0.5 0.5 1.0

X

√ p
Figure 14.1: Left panel: f (x) = 1 − x2 . Right panel: f (x, y) = 1 − x2 − y 2

2
14.1.2 Functions of two real variables
Let A, B and C be subsets of R, the set of real numbers. Then a rule f that assigns to each member
(x, y) ∈ A × B a unique member z ∈ C, is called a function from A × B to C, written as f : A × B → C.
The element z is called image of (x, y) under f . It is also called as the value of f at (x, y) and is denoted
by f (x, y), that is, z = f (x, y) while (x, y) is called pre-image of z under f . The set A × B (the set of
all pre-images) is called domain of f (denoted by D(f )) while the set of images of all the elements of
A × B under f , that is, R(f ) = {f (x, y) : (x, y) ∈ A × B} is called range of f . We call the set C as
co-domain of f . The set G(f ) = {(x, y, f (x, y)) : (x, y) ∈ A × B} is called graph of f . Notice that the
graph of f contains the triplets (x, y, f (x, y)) of real numbers, which can be shown geometrically as points
(x, y, f (x, y)) in the three dimensional XYZ-space. p
Ex. Consider the function f : {(x, y) : x2 + y 2 ≤ p 1} → R defined by f (x, y) = 1 − x2 − y 2 . Then,
2 2 2 2
we have p D(f ) = {(x, y) : x + y ≤ 1}, R(f ) = { 1 − x − y : (x, y) ∈ D(f )} = [0, 1] and G(f ) =
2 2
{(x, y, 1 − x − y ) : (x, y) ∈ D(f )}, which is displayed geometrically in the XYZ-space in the right
panel of Figure 2. So domain of f is the region within and on the circle x2 + y 2 = 1 in the XY-plane, and
geometrically its graph is the hemispherical surface.
Remark: Likewise, we can define a function of three variables. However, functions of more than two
variables can not be represented geometrically, and therefore are not p of much interest. Consider the
2 2 2
function f : {(x, y, z) : x + y + z ≤ 1} → R defined by f (x) = 1 − x2 − y 2 − z 2 . Then, we have
p
2 2 2
D(f ) = {(x, y, z)p: x + y + z ≤ 1}, R(f ) = { 1 − x2 − y 2 − z 2 : (x, y, z) ∈ D(f )} = [0, 1] and
G(f ) = {(x, y, z, 1 − x2 − y 2 − z 2 ) : (x, y, z) ∈ D(f )}. So domain of f is the region within and on
the sphere x2 + y 2 + z 2 = 1 in the XYZ-space, and of course its graph G(f ) can not be represented
geometrically.
Table 1 shows some two and three variable functions with their respective domains and ranges.

Table 14.1: Some functions with domains and ranges.

Function
p Domain Range
z = y − x2 y≥x 2 [0, ∞)
1
z = xy xy ̸= 0 (−∞, 0) ∪ (0, ∞)
z = sin
p xy Entire plane [−1, 1]
w = x2 + y 2 + z 2 Entire space [0, ∞)
w = x2 +y12 +z 2 (x, y, z) ̸= (0, 0, 0) (0, ∞)
w = xy ln z Half-space z > 0 (−∞, ∞)

14.1.3 Neighbourhoods
Roughly speaking, neighbourhood of a point is any open region (the region carrying only its interior
points and hence without the boundary points) containing the point.
Formally, neighbourhood of a point x0 on X-axis is any open interval containing the point x0 . For
any positive real number δ, the interval (x0 − δ, x0 + δ) is a neighbourhood of x0 . We shall denote it by
Nδ (x0 ) and call it δ-neighbourhood of x0 . In fact, the interval (x0 − δ, x0 + δ) is neighbourhood of its
every point. A neighbourhood of x0 without x0 is called deleted neighbourhood of x0 .
Neighbourhood of a point (x0 , y0 ) in the XY-plane is any open area or region containing the point
(x0 , y0 ). For any positive real number δ, the area {(x, y) : |x − x0 | < δ, |yp − y0 | < δ} is a neighbourhood
of (x0 , y0 ) and is a square neighbourhood of (x0 , y0 ). Similarly, {(x, y) : (x − x0 )2 + (y − y0 )2 < δ} is
circular neighbourhood of (x0 , y0 ). We shall denote any δ-neighbourhood of (x0 , y0 ) by Nδ (x0 , y0 ).
Likewise, neighbourhood of a point (x0 , y0 , z0 ) in the XYZ-space is any open region containing the
point (x0 , y0 , z0 ). For instance, Nδ (x0 , y0 , z0 ) = {(x, y, z) : |x−x0 | < δ, |y −y0 | < δ, |z −z0 | < δ} is cubical

3
 p
δ-neighbourhood of (x0 , y0 , z0 ) while Nδ (x0 , y0 , z0 ) = {(x, y, z) : (x − x0 )2 + (y − y0 )2 + (z − z0 )2 < δ}
is spherical δ-neighbourhood of (x0 , y0 , z0 ).

14.1.4 Interior and boundary points in XY-plane

A point (x0 , y0 ) in a region (set) R in XY-plane is an interior point of R if there exists a nbd of (x0 , y0 )
entirely lying inside R as shown in the left panel of Figure 2. If every nbd of (x0 , y0 ) contains points inside
as well as outside of R, then it is called a boundary point of R (see right panel of Figure 2). The interior
points of a region, as a set, make up the interior of the region. The region’s boundary points make up its
boundary. A region is open if it consists entirely of interior points. A region is closed if it contains all its
boundary points.

Figure 14.2: Left: Interior point (x0 , y0 ) of R. Right: Boundary point (x0 , y0 ) of R.

A region in the plane is bounded if it lies inside a disk of finite

p radius. A region is unbounded if it
is not bounded. For example, domain of the function f (x, y) = y − x2 is given by y ≥ x2 , which is a
closed and unbounded region as shown in Figure 3.

4
 Figure 14.3:

14.1.5 Level Curves, and Contours of Functions of Two Variables

The set of points in the domain of the function z = f (x, y) where it has a constant value f (x, y) = c is
called a level curve of f . A contour curve is a curve f (x, y) = c in space in which the plane z = c cuts
the surface z = f (x, y). For example, consider the function f (x, y) = 100 − x2 − y 2 . The level curves
f (x, y) = 10, f (x, y) = 51, and f (x, y) = 75 in the domain of f are shown in left panel of Figure 4.

Figure 14.4:

14.2 Interior and boundary points in XYZ-space

A point (x0 , y0 , z0 ) in a region (set) R in XYZ-space is an interior point of R if there exists a nbd of
(x0 , y0 , z0 ) entirely lying inside R as shown in the left panel of Figure 5. If every nbd of (x0 , y0 ) contains
points inside as well as outside of R, then it is called a boundary point of R (see right panel of Figure
5). The interior points of a region, as a set, make up the interior of the region. The region’s boundary

5
points make up its boundary. A region is open if it consists entirely of interior points. A region is closed
if it contains all its boundary points.

Figure 14.5: Left: Interior point (x0 , y0 , z0 ) of R. Right: Boundary point (x0 , y0 , z0 ) of R.

The set of points (x, y, z) in space where a function of three independent variables has a constant value
f (x, y, z) = c is called a level surface of f . For example, the level surfaces of the function f (x, y, z) =
p
x2 + y 2 + z 2 are concentric spheres as shown in Figure 6.

p
Figure 14.6: The level surfaces of f (x, y, z) = x2 + y 2 + z 2 are concentric spheres.

14.3 Limits and Continuity in Higher Dimensions

14.3.1 Limit of a function of one real variable
Meaning of x → x0
If a real variable x ̸= x0 takes infinitely many values in each neighbourhood of the point x0 , then we
say that x approaches or tends to x0 , and we write x → x0 . It implies that each δ-neighbourhood
(x0 − δ, x0 + δ) of x0 contains infinitely many values of the variable x. We may choose positive values
of δ very close to 0. Still the δ-neighbourhood will carry infinitely many values of the variable x. Thus,
x takes values arbitrarily close to x0 , and it makes sense to say that x approaches x0 or x → x0 . If x
approaches x0 by taking values less than x0 , then x0 is called as the left hand limit of x, and we write
x → x− 0 . If x approaches x0 by taking values greater than x0 , then x0 is called as the right hand limit of
x, and we write x → x+ 0.
Ex. Suppose x takes values 0.9, 0.99, 0.999, ........ Then x ̸= 1 and takes infinitely many values in each
neighbourhood of 1. So x approaches 1. Also, all the values of x are less than 1. So x → 1− .

6
Ex. Suppose x takes values 1.1, 1.01, 1.001, ........ Then x ̸= 1 and takes infinitely many values in
each neighbourhood of 1. So x approaches 1. Also, all the values of x are greater than 1. So x → 1+ .
Ex. Suppose x takes values 1, 1/2, 1/3, ........ Then x ̸= 0 and takes infinitely many values in each
neighbourhood of 0. So x approaches 0. Also, all the values of x are greater than 0. So x → 0+ .

Limit of f (x) as x → x0
Let f be a function defined in some neighbourhood of x0 except possibly at x0 . Then a real number
l is said to be limit of f (x), symbolically written as lim f (x) = l, if given any positive real number ϵ
x→x0
(however small), there exists δ > 0 (depending on ϵ) such that
0 < |x − x0 | < δ =⇒ |f (x) − l| < ϵ.
or x ∈ (x0 − δ, x0 + δ) − {x0 } =⇒ f (x) ∈ (l − ϵ, l + ϵ).
Thus, lim f (x) = l if corresponding to each ϵ-neighbourhood of l, there exists a deleted δ-neighbourhood
x→x0
of x0 such that the values of f (x) corresponding to the deleted δ-neighbourhood of x0 lie in the ϵ-
neighbourhood of l. Geometrically, it implies that the portion of the curve f (x) corresponding to the
deleted δ-neighbourhood of x0 lies inside the horizontal strip created by the lines f (x) = l − ϵ and
f (x) = l + ϵ parallel to X-axis.
Further, lim f (x) = l (left hand limit of f (x)) if x ∈ (x0 − δ, x0 ) =⇒ f (x) ∈ (l − ϵ, l + ϵ), and
x→x−
0
lim f (x) = l (right hand limit of f (x)) if x ∈ (x0 , x0 + δ) =⇒ f (x) ∈ (l − ϵ, l + ϵ).
x→x+
0
Obviously, lim f (x) = l if and only if lim f (x) = l = lim f (x).
x→x0 x→x− x→x+
0 0
Ex. Show that lim (2x + 1) = 3.
x→1
Sol. Let ϵ > 0 be given. Then
|2x + 1 − 3| = |2x − 2| = 2|x − 1| < ϵ provided |x − 1| < ϵ/2.
Choosing δ = ϵ/2, we have
|x − 1| < δ =⇒ |2x + 1 − 3| < ϵ.
Thus, lim (2x + 1) = 3.
x→1
For the sake of illustration, we plot f (x) = 2x + 1 in left panel of Figure 7. We choose ϵ = 0.4
so that δ = ϵ/2 = 0.2. The horizontal dotted blue lines are f (x) = l − ϵ = 3 − 0.4 = 2.6 and
f (x) = l + ϵ = 3 + 0.4 = 3.4 while the vertical red lines are x = x0 − δ = 1 − 0.2 = 0.8 and
x = x0 + δ = 1 + 0.2 = 1.2. We can see that the part of the curve in the vertical red strip com-
pletely lies inside the horizontal blue strip, as expected.

Y Y

3 3

2 2

1 1

X X
- 0.5 O 0.5 1.0 1.5 O 0.5 1.0 1.5

Figure 14.7: Left panel: f (x) = 2x + 1. Right panel: f (x) = x for x ≤ 1 and f (x) = x + 1 for x > 1.

x if x ≤ 1
Ex. Show that lim f (x), where f (x) = , does not exist.
x→1 x+1 if x > 1

7
Sol. We shall show that lim f (x) = 1 and lim f (x) = 2.
x→1− x→1+
Let ϵ > 0 be given. Then for x < 1, we have |f (x) − 1| = |x − 1| < ϵ provided x ∈ (1 − ϵ, 1).
Choosing δ = ϵ, we have
x ∈ (1 − δ, 1) =⇒ |f (x) − 1| < ϵ.
Thus, lim f (x) = 1.
x→1−
Next for x > 1, we have
|f (x) − 2| = |x + 1 − 2| = |x − 1| < ϵ provided x ∈ (1, 1 + ϵ).
Choosing δ = ϵ, we have
x ∈ (1, 1 + δ) =⇒ |f (x) − 2| < ϵ.
Thus, lim f (x) = 2.
x→1+
Since lim f (x) ̸= lim f (x), so lim f (x) does not exist.
x→1− x→1+ x→1 
x if x ≤ 1
For the sake of illustration, we plot f (x) = in right panel of Figure 7. We
x+1 if x > 1
choose ϵ = 0.2 so that δ = ϵ = 0.2. The horizontal dotted blue lines are f (x) = l − ϵ = 1 − 0.2 = 0.8
and f (x) = l + ϵ = 1 + 0.2 = 1.2 while the horizontal dotted blue lines are f (x) = l − ϵ = 2 − 0.2 = 1.8
and f (x) = 2 + ϵ = 2 + 0.2 = 2.2. The vertical red lines are x = x0 − δ = 1 − 0.2 = 0.8 and
x = x0 + δ = 1 + 0.2 = 1.2. We can see that the part of the curve corresponding to x ∈ (0.8, 1) in the
vertical Red strip completely lies inside the horizontal blue strip while the part of the curve corresponding
to x ∈ (1, 1.2) in the vertical Red p
strip completely lies inside the horizontal Green strip, as expected.
Ex. Discuss the geometry of lim 1 − x2 = 1.
√ x→0
Sol. We know that 1 − x2 is semicircular curve with the domain [−1, 1] on X-axis (See left panel of
Figure 1). It is easy to see that when x → 0 either from left or right in the neighbourhood √ of 0, the
2
corresponding part of the curve converges to the point (0, 1) p on Y-axis. It implies that 1 − x tends to
1 for both the paths along which x → 0. That is why, lim 1 − x2 = 1.
x→0
Remark: If lim f (x) exists, then the graph of the function f (x), geometrically, converges or strikes at
x→x0
the same point (whose y-coordinate is the limit of f (x)) from left as well as right in the neighbourhood of
x0 . For instance, see left panel of Figure 2. If lim f (x) does not exist but lim f (x) and lim f (x) exist
x→x0 x→x− x→x+
0 0
finitely, then the graph of the function f (x), geometrically, from the left strikes at a point different from
the point where it strikes from the right in the neighbourhood of x0 . For an example, see right panel of
Figure 2.
Remark: The limit of a function at a point may exist even if the function is not defined there at. For
example, consider the function f (x) = 2x + 1, x ̸= 1, which is not defined at x = 1 as per its given
definition. But lim (2x + 1) = 3.
x→1
For, let ϵ > 0 be given. Then
|2x + 1 − 3| = |2x − 2| = 2|x − 1| < ϵ provided 0 < |x − 1| < ϵ/2.
Choosing δ = ϵ/2, we have
0 < |x − 1| < δ =⇒ |2x + 1 − 3| < ϵ.
Thus, lim (2x + 1) = 3.
x→1

14.3.2 Limit of a function of two real variables

Meaning of (x, y) → (x0 , y0 )
If the ordered pair (x, y) ̸= (x0 , y0 ) of real variables x and y takes infinitely many values in each neigh-
bourhood of the point (x0 , y0 ), then we say that (x, y) tends to (x0 , y0 ), and we write (x, y) → (x0 , y0 ). It
implies that each δ-neighbourhood Nδ (x0 , y0 ) (square, circular or whatever sahpe) of (x0 , y0 ) contains in-
finitely many values of (x, y). We may choose positive values of δ very close to 0. Still the δ-neighbourhood

8
will carry infinitely many values of (x, y). Thus, (x, y) takes values arbitrarily close to (x0 , y0 ), and it
makes sense to say that (x, y) approaches (x0 , y0 ) or (x, y) → (x0 , y0 ).
Now, see the critical difference between x → x0 and (x, y) → (x0 , y0 ). In the case, x → x0 , the
variable x can take values only on the X-axis (one dimension), and thus can approach x0 from left or
right directions along X-axis. That is why, we talk about left and right hand limits when x → x0 . On
the other hand, in the case (x, y) → (x0 , y0 ), the ordered pair (x, y) can take values in the XY-plane (two
dimensions), and thus can approach (x0 , y0 ) along infinitely many paths in the XY-plane.
Ex. Suppose (x, y) takes values (0.9, 0.9), (0.99, 0.99), (0.999, 0.999), ........ Then (x, y) ̸= (1, 1) and takes
infinitely many values in each neighbourhood of (1, 1). So (x, y) approaches (1, 1). Also, all the values of
(x, y) lie on the line y = x. So (x, y) → (1, 1) by taking values on the straight line path y = x.
Ex. Suppose (x, y) takes values (1.1, (1.1)2 ), (1.01, (1.01)2 ), (1.001, (1.001)3 ), ........ Then (x, y) ̸= (1, 1)
and takes infinitely many values in each neighbourhood of (1, 1). So (x, y) approaches (1, 1). Also, all
the values of (x, y) lie on the parabola y = x2 . So (x, y) → (1, 1) by taking values on the parabolic path
y = x2 .

Limit of f (x, y) as (x, y) → (x0 , y0 )

Let f be a function defined in some neighbourhood of (x0 , y0 ) except possibly at (x0 , y0 ). Then a real
number l is said to be limit of f (x, y), symbolically written as lim f (x, y) = l, if given any positive
(x,y)→(x0 ,y0 )
real number ϵ (however small), there exists δ > 0 (depending on ϵ) such that
0 < |x − x0 | < δ, 0 < |y − y0 | < δ =⇒ |f (x) − l| < ϵ.
or x ∈ (x0 − δ, x0 + δ) − {x0 }, y ∈ (y0 − δ, y0 + δ) − {y0 } =⇒ f (x) ∈ (l − ϵ, l + ϵ).
Thus, lim f (x, y) = l if corresponding to each ϵ-neighbourhood of l, there exists a deleted
(x,y)→(x0 ,y0 )
δ-neighbourhood of (x0 , y0 ) such that the values of f (x, y) corresponding to the deleted δ-neighbourhood
of (x0 , y0 ) lie in the ϵ-neighbourhood of l. Geometrically, it implies that the portion of the surface f (x, y)
corresponding to the deleted δ-neighbourhood of (x0 , y0 ) lies inside the sandwiched space between the
planes f (x) = l − ϵ and f (x) = l + ϵ parallel to XY-plane.
Further, if lim f (x, y) exists and is equal to l (say), then f (x, y) → l for every possible
(x,y)→(x0 ,y0 )
path along which (x, y) approaches (x0 , y0 ), that is, the limit is independent of path. In case, f (x, y)
approaches to two different values for two different paths along which (x, y) approaches (x0 , y0 ), then
lim f (x, y) does not exist.
(x,y)→(x0 ,y0 )
Ex. Show that lim (2x + 2y + 1) = 5.
(x,y)→(1,1)
Sol. Let ϵ > 0 be given. Then
|2x + 2y + 1 − 5| = |2x + 2y − 4| = |2(x − 1) + 2(y − 1)| ≤ 2|x − 1| + 2|y − 1| < ϵ provided |x − 1| < ϵ/2,
|y − 1| < ϵ/2.
Choosing δ = ϵ/2, we have
|x − 1| < δ, |y − 1| < δ =⇒ |2x + 2y + 1 − 5| < ϵ.
Thus, lim (2x + 2y + 1) = 5.
(x,y)→(1,1)
Geometry: For the sake of illustration, we plot the plane f (x) = 2x + 2y + 1 (Red color plane) in the
left panel of Figure 8 in the domain {(x, y) : 0 ≤ x ≤ 2, 0 ≤ y ≤ 3}. Let us choose ϵ = 0.4 so that
δ = ϵ/2 = 0.2. Then f (x) = l − ϵ = 5 − 0.4 = 4.6 (Blue color plane) and f (x) = l + ϵ = 5 + 0.4 = 5.4
(Green color plane) are planes parallel to the XY-plane. Next, the square-neighbourhood of (1, 1) in the
XY-plane is {(x, y) : |x − 1| < 0.2, |y − 1| < 0.2}, which is not shown in Figure 8. We can imagine that
the part of the Red plane corresponding to the square-neighbourhood {(x, y) : |x − 1| < 0.2, |y − 1| < 0.2}
of (1, 1) lies inside the sandwiched space between the Green (f (x) = 4.6) and Blue (f (x) = 5.4) colored
planes, as expected.

9
 -2

-1
Y
0 1 2 3 0.5 0 X

Z 0.0
1
6 - 0.5
Z
2
4

0.0 0.5 1.0 1.5 2.0

X 2

0
Y
-2

xy
Figure 14.8: Left panel: f (x) = 2x + 2y + 1 (Red plane). Right panel: f (x, y) = x2 +y 2
, (x, y) ̸= (0, 0).

xy
Ex. Show that lim f (x, y), where f (x, y) = x2 +y 2
, (x, y) ̸= (0, 0) does not exist.
(x,y)→(0,0)
Sol. We shall show that the given limit is path dependent.
Suppose (x, y) → (0, 0) along the straight line path y = mx. Then
xy x.mx m
lim f (x, y) = lim = lim 2 = .
(x,y)→(0,0) (x,y)→(0,0) x2 + y 2 x→0 x + m2 x2 1 + m2
So the given limit depends on m, the slope of the path line. Hence, it get different values for different
values of m, and consequently it does not exist.
Geometry: The right panel of Figure 8 shows the the surface f (x, y) = x2xy +y 2
, (x, y) ̸= (0, 0) in Green
color. We can see that the part of the surface corresponding to a neighbourhood of the point (0, 0) is not
smooth rather it looks ruptured. Any path in the neighbourhood of (0, 0) in the XY-plane corresponds
to a curve on the Green surface. When (x, y) → (0, 0) along different paths y = mx, the corresponding
curves end up or converge to points near the ruptured part at different heights, which is evident from
the figure. That is why, the limit get different values along different paths, and consequently it does not
exist.
Remark: Note that the two path approach or path dependent test is used when the limit does not exist
uniquely. In case, the limit exists uniquely, you will get same answer for the limit along every chosen
path. However, this way you can not ensure the existence of limit because you can not exhaust infinitely
many paths. So in order to prove the existence of limit, use ϵ − δ approach. Also, in some problems,
changing the variables to polar coordinate system is quite useful for applying ϵ − δ approach.

4x2 y
Ex. Show that lim = 0.
(x,y)→(0,0) x2 + y 2
Sol. Let ϵ > 0 be given. Then, we have
4x2 y 4x2 y
− 0 = .
x2 + y 2 x2 + y 2
Using x = r cos θ and y = r sin θ, we get
4x2 y 2
p
2 + y 2 < ϵ for
p
− 0 = |4r cos θ sin θ| ≤ 4r = 4 x x2 + y 2 < ϵ/4.
x2 + y 2
Choosing ϵ/4 = δ, we have
4x2 y p
− 0 < ϵ for x2 + y 2 < δ.
x2 + y 2
4x2 y
Hence, lim = 0.
(x,y)→(0,0) x2 + y 2
p
Ex. Discuss the geometry of lim 1 − x2 − y 2 = 1.
(x,y)→(0,0)

10
 p
Sol. We know that 1 − x2 − y 2 is hemispherical surface with the circular domain x2 + y 2 ≤ 1 centred
at (0, 0) in the XY-plane (See right panel of Figure 1). Any path in the neighbourhood of (0, 0) inside the
circular domain in the XY-plane corresponds to a curve on the hemispherical surface. It is easy to imag-
ine/visualize that when (x, y) → (0, 0) along any path in the neighbourhood of (0, 0), the corresponding
p
curve on the hemispherical surface converges to the point (0, 0, 1) on Z-axis. Itpimplies that 1 − x2 − y 2
tends to 1 for every path along which (x, y) → (0, 0). That is why, lim 1 − x2 − y 2 = 1.
(x,y)→(0,0)
Remark: If lim f (x, y) exists, then the graph (surface) of the function f (x, y), geometrically,
(x,y)→(x0 ,y0 )
converges or strikes at the same point (whose z-coordinate is the limit of f (x, y)) from all directions in
the neighbourhood of (x0 , y0 ).
Remark: The limit of a function at a pointp may exist even if the function is not defined there at. For
example, consider the function f (x, y) = 1 − x2 − y 2 , (x, y) ̸= (0, 0), which is not defined at x = (0, 0)
as per its given definition. Its graph is the hemispherical surface punctured at the pointp (0, 0, 1) with
2 2
the circular domain x + y ≤ 1 centered at (0, 0) in the XY-plane. We find lim 1 − x2 − y 2 = 1
(x,y)→(0,0)
since when (x, y) → (0, 0) along any path in the neighbourhood of (0, 0), the corresponding curve on the
hemispherical surface converges to the point (0, 0, 1) on Z-axis.
Sandwich Theorem: Suppose the functions g(x, y), f (x, y) and h(x, y) are defined in some neigh-
bourhood of (x0 , y0 ) except possibly at (x0 , y0 ) such that g(x, y) ≤ f (x, y) ≤ h(x, y) for all (x, y), and
lim g(x, y) = L = lim h(x, y). Then lim f (x, y) = L.
(x,y)→(x0 ,y0 ) (x,y)→(x0 ,y0 ) (x,y)→(x0 ,y0 )

Ex. Show that lim y sin(1/x) = 0.

(x,y)→(0,0)
Sol. Since −1 ≤ sin(1/x) ≤ 1, we have −y ≤ y sin(1/x) ≤ y or y ≤ y sin(1/x) ≤ −y according as y > 0
or y < 0. Also, lim y = 0.
(x,y)→(0,0)
So by Sandwich theorem, we get lim y sin(1/x) = 0.
(x,y)→(0,0)

14.4 Continuity of a function of one real variable

Let f be a function defined in some neighbourhood of x0 . Then f is said to be continuous at x0 if
lim f (x) = f (x0 ). Further, f is left continuous at x0 if lim f (x) = f (x0 ), and right continuous if
x→x0 x→x−
0
lim f (x) = f (x0 ). We say that f is continuous on an interval if it is continuous at each point of the
x→x+
0
interval.
Geometry: The existence of limit at x0 ensures that the graph of f (x) strikes at the same point from
both sides. Also, the limit is equal to f (x0 ). So the point of strike is (x0 , f (x0 )). Therefore, continuity of
f at x0 implies that there exists at least one neighbourhood of x0 in which the f (x) has continuous graph,
that is, without any break point. Consequently, continuity over an interval implies that the function has
continuous graph in the interval.
√
Ex. The function f (x) = 1 − x2 is continuous at x = 0 since lim f (x) = 1 = f (0). In fact,
√ x→0
f (x) = √ 1 − x2 is continuous at every point in its domain interval [−1, 1]. That is why, graph of
f (x) = 1 − x2 is the continuous semicircular curve from (−1, 0) to (1, 0) as shown in left panel of the
Figure 1. 
x if x ≤ 1
Ex. The function f (x) = is not continuous at x = 1 since lim f (x) does not exist.
x+1 if x > 1 x→1
The graph of f (x) breaks at the point x = 1 as may be seen in the right panel of Figure 7. Notice that
the function is continuous at every real number except x = 1.

11
14.4.1 Continuity of a function of two real variables
Let f be a function defined in some neighbourhood of (x0 , y0 ). Then f is said to be continuous at (x0 , y0 )
if lim f (x, y) = f (x0 , y0 ). We say that f is continuous in an open region in the XY-plane if it is
(x,y)→(x0 ,y0 )
continuous at each point of the region.
Geometry: The existence of limit at (x0 , y0 ) ensures that the graph (surface) of f (x, y) points towards
the same point from all directions in each neighbourhood of (x0 , y0 ). Also, the limit is equal to f (x0 , y0 ).
So the point of convergence of the surface is (x0 , y0 , f (x0 , y0 )). Therefore, continuity of f at (x0 , y0 ) implies
that there exists at least one neighbourhood of (x0 , y0 ) in which the f (x, y) has continuous surface, that
is, without any break point or point hole. Consequently, continuity over a region implies that the function
has continuous surface in thepregion.
Ex. The function f (x, y) = 1 − x2 − y 2 is continuous at (0, 0) since lim f (x, y) = 1 = f (0, 0). In fact,
p x→0
f (x, y) = 1 − x2 − yp 2 is continuous at every point in its domain circular region x2 + y 2 ≤ 1. That is

why, graph of f (x) = 1 − x2 − y 2 is the continuous hemispherical surface as shown in right panel of the
Figure 1.
Ex. The function f (x, y) = x2xy +y 2
, (x, y) ̸= (0, 0), is not continuous at (0, 0) since the function is not
defined at (0, 0). Even if the function is defined to have some value at (0, 0), it can not be continuous at
(0,0) since lim f (x, y) does not exist. The graph surface of f (x, y) breaks at the point (0, 0) as may
(x,y)→(0,0)
be seen in the right panel of Figure 8. Notice that the function is continuous everywhere in the XY-plane
except (0, 0).

14.5 Partial Derivatives

14.5.1 Differentiability of a function of one real variable
Let f be a function defined in some neighbourhood of x, and ∆f = f (x + ∆x) − f (x) be change in
f corresponding to an infinitesimal change ∆x in x. Then f is said to be differentiable at x if ∆f =
f (x + ∆x) − f (x) can be expressed in the form

∆f = f (x + ∆x) − f (x) = A∆x + ∆x ϕ(∆x), (14.1)

where A is independent of ∆x, and ϕ is a function of ∆x such that ϕ(∆x) → 0 as ∆x → 0.

Dividing (14.1) throughout by ∆x, and then taking limit as ∆x → 0, we get
∆f f (x + ∆x) − f (x)
lim = lim = A.
∆x→0 ∆x ∆x→0 ∆x
∆f f (x + ∆x) − f (x)
We define the limit lim or lim to be the derivative of f with respect to x,
∆x→0 ∆x ∆x→0 ∆x
df df
and denote it by or f ′ (x). So = f ′ (x) = A. Obviously, the existence of f ′ (x) is necessary for the
dx dx
function f (x) to be differentiable at x. Now, (14.1) can be rewritten as

∆f = f (x + ∆x) − f (x) = f ′ (x)∆x + ∆x ϕ(∆x), (14.2)

Next, we define the expression f ′ (x)∆x to be the differential of f and denote it by df . So df = f ′ (x)∆x. If
we choose f (x) = x, then dx = 1.∆x = ∆x. Therefore, the differential of f can be written as df = f ′ (x)dx.
Remark1: Notice that by definition of differential, dx = ∆x. Likewise, we have df = ∆f . It means,
differentials can be replaced by the corresponding infinitesimals and vice versa. That is why, the expression
df
is interpreted in two ways. First, simply as the ratio of the differentials df and dx or the infinitesimals
dx
df ∆f
∆f and ∆x. Second, it is the limit of the ratio of the infinitesimals ∆f and ∆x, that is, = lim .
dx ∆x→0 ∆x

12
In this case, it is derivative of f with respect to x. Thus, the differential of f is df while the derivative of
df
f , denoted by f ′ (x) or , is the limit of the ratio of the infinitesimals ∆f and ∆x. In fact, the derivative
dx
measures a rate of change, while the differential measures the change itself. In practical situations,
sometimes we are interested in the rate of change of some quantity and sometimes we want to know the
change itself. Therefore, derivative and differential are not same but both are useful.
Remark2: If y is a differentiable function of x, and x is a differentiable function of t, then y is a
differentiable function of t, and
dy dy dx
= ,
dt dx dt
known as the chain rule of derivatives. In the chain rule, it appears like we have multiplied and divided
by dx. But it is not true because the expressions dy dy dx
dt , dx and dt are derivatives and not the ratios of the
differentials dy, dx and dt. For, let ∆t be an infinitesimal change in t, and ∆x and ∆y be corresponding
changes in x and y, respectively. Then we can write,
∆y ∆y ∆x
=
∆t ∆x ∆t
Here, we have really multiplied and divided by ∆x. In the limit ∆t → 0, we get
  
∆y ∆y ∆x
lim = lim lim .
∆t→0 ∆t ∆x→0 ∆x ∆t→0 ∆t

which implies the chain rule

dy dy dx
= .
dt dx dt
Z
Remark3: Consider the integral x sin(x2 )dx. Here, dx is the differential of x. It has no role at all in
the integration, and is optional to write with the integrand function. But it is useful to write it in the
integral for two reasons. First, it tells us explicitly that the variable of integration is x. Second, it helps
us in using the change of variable conveniently. For instance, to solve the given integral, we use x2 = t.
d
This implies that d(x2 ) = dt or dx (x2 )dx = dt or 2xdx = dt. Consequently,
Z Z
2 1 1 1
x sin(x )dx = sin tdt = − cos t + C = − cos(x2 ) + C,
2 2 2
where C is constant of integration.
f (x + ∆x) − f (x)
Geometry of f ′ (x): The ratio is slope of the secant passing through the points
∆x
P (x, f (x)) and Q(x + ∆x, f (x + ∆x)) of the curve f (x). Since , by definition, tangent to the curve at the
point P (x, f (x)) is the limiting position of secant P Q as Q → P or ∆x → 0, so the slope of the secant P Q
f (x + ∆x) − f (x)
becomes the slope of the tangent at P in the limit ∆x → 0. Therefore, f ′ (x) = lim
∆x→0 ∆x
is the slope of the tangent to the curve f (x) at the point P (x, f (x)).

Ex. Show that the function f (x) = x2 is differentiable.

Sol. We have
f (x + ∆x) − f (x) = (x + ∆x)2 − x2 = 2x∆x + (∆x)2 = A∆x + ∆x ϕ(∆x),
where A = 2x is independent of ∆x, and ϕ(∆x) = ∆x → 0 as ∆x → 0. Therefore, f (x) is differentiable
f (x + ∆x) − f (x)
and f ′ (x) = lim = A = 2x.
∆x→0 ∆x

13
Ex. Show that the function f (x) = |x| is not differentiable at x = 0.
Sol. We have

f (0 + ∆x) − f (0) |∆x|

f ′ (0) = lim = lim .
∆x→0 ∆x ∆x→0 ∆x
|∆x| −∆x |∆x| ∆x
But lim = lim = −1 and lim = lim = 1.
∆x→0− ∆x ∆x→0− ∆x ∆x→0+ ∆x ∆x→0+ ∆x
|∆x|
So lim does not exist and hence f ′ (0) does not exist. Consequently, f (x) = |x| is not differentiable
∆x→0 ∆x
at 0.
Geometry: The plot of f (x) = |x| is shown in the Figure 4. We see that tangent at (0, 0) from the left
|∆x|
is the line f (x) = −x with slope lim = −1 (left hand derivative) while the tangent at (0, 0) from
∆x→0− ∆x
|∆x|
the right is the line f (x) = x with slope lim = 1 (right hand derivative). So we do not have a
∆x→0 + ∆x
′
unique tangent at (0, 0), and consequently f (0) does not exist.
Y

1.2

1.0

0.8

0.6

0.4

0.2

-1.0 - 0.5 0.5

X

Figure 14.9: f (x) = |x|.

Differentiability implies continuity: Let a function f (x) be differentiable at x = x0 . Then f ′ (x0 )

exists and f (x0 + ∆x) − f (x0 ) = f ′ (x0 )∆x + ∆x ϕ(∆x), where ϕ(∆x) → 0 as ∆x → 0. So in the limit
∆x → 0, we get lim f (x0 + ∆x) = f (x0 ). This shows that f (x) is continuous at x = x0 .
∆x→0
However, a continuous function need not be differentiable. For example, f (x) = |x| is continuous at
x = 0 but not differentiable at x = 0.
Why is differentiability considered on open intervals in mean value theorems?
We do have a notion of a function being differentiable on a closed interval.
The reason that mean value theorems, for instance, the Rolle’s theorem talks about differentiability on
the open interval (a, b) is that it is a weaker assumption than requiring differentiability on [a, b].
Normally, theorems might try to make the assumptions as weak as possible, to be more generally appli-
cable. For instance, the function:
f (x) = x sin(1/x), x > 0 and f (0) = 0
is continuous at 0, and differentiable everywhere except at 0.
We can still apply Rolle’s theorem to this function on say the interval (0, 1/π). If the statement of Rolle’s
theorem required the use of the closed interval, then we could not apply it to this function.

14.5.2 Differentiability of a function of two real variables

Let f be a function defined in some neighbourhood of (x, y), and ∆f = f (x + ∆x, y + ∆y) − f (x, y) be
change in f corresponding to an infinitesimal change ∆x in x and an infinitesimal change ∆y in y . Then
f is said to be differentiable at (x, y) if ∆f = f (x + ∆x, y + ∆y) − f (x, y) can be expressed in the form
∆f = f (x + ∆x, y + ∆y) − f (x, y) = A∆x + B∆y + ∆x ϕ(∆x, ∆y) + ∆y ψ(∆x, ∆y), (14.3)

14
where A and B both are independent of ∆x and ∆y; and the functions ϕ(∆x, ∆y) and ψ(∆x, ∆y) both
tend to 0 as (∆x, ∆y) → (0, 0).
Choosing ∆y = 0 and dividing (14.3) throughout by ∆x, and then taking limit as ∆x → 0, we get

∆f f (x + ∆x, y) − f (x, y)
lim = lim = A.
∆x→0 ∆x ∆x→0 ∆x
∆f f (x + ∆x, y) − f (x, y)
We define the limit lim or lim to be the partial derivative of f with
∆x→0 ∆x ∆x→0 ∆x
∂f ∂f
respect to x, and denote it by or fx (x, y). So = fx (x, y) = A.
∂x ∂x
Similarly, choosing ∆x = 0 and dividing (14.3) throughout by ∆y, and then taking limit as ∆y → 0,
we get

∆f f (x, y + ∆y) − f (x, y)

lim = lim = B.
∆y→0 ∆y ∆y→0 ∆y

∆f f (x, y + ∆y) − f (x, y)

We define the limit lim or lim to be the partial derivative of f with
∆y→0 ∆y ∆y→0 ∆y
∂f ∂f
respect to y, and denote it by or fy (x, y). So = fy (x, y) = B.
∂y ∂y
∂f ∂f
We notice that the existence of the partial derivatives and is essential for the function f (x, y)
∂x ∂y
to be differentiable at (x, y). Now, (14.3) can be rewritten as

∂f ∂f
∆f = f (x + ∆x, y + ∆y) − f (x, y) = ∆x + ∆y + ∆x ϕ(∆x, ∆y) + ∆y ψ(∆x, ∆y), (14.4)
∂x ∂y
∂f ∂f
Next, we define the expression ∆x + ∆y to be the total differential of f and denote it by df . So
∂x ∂y
∂f ∂f
df = ∆x + ∆y. If we choose f (x, y) = x, then dx = 1.∆x = ∆x. Similarly, the choice f (x, y) = y
∂x ∂y
leads to dy = 1.∆y = ∆y. Therefore, the total differential of f can be written as

∂f ∂f
df = dx + dy.
∂x ∂y

Ex. Show that the function f (x) = x2 + y 2 is differentiable.

Sol. We have
f (x + ∆x) − f (x) = (x + ∆x)2 + (y + ∆y)2 − x2 − y 2 = 2x∆x + 2y∆y + (∆x)2 + (∆y)2
= A∆x + B∆y + ∆x ϕ(∆x, ∆y) + ∆y ϕ(∆x, ∆y),
where A = 2x and B = 2y both are independent of ∆x and ∆y, and ϕ(∆x, ∆y) = ∆x → 0 and
ψ(∆x, ∆y) = ∆y → 0 as (∆x, ∆y) → (0, 0). Therefore, f (x, y) is differentiable at (x, y).
( 2 2
x +y
|x|+|y| if (x, y) ̸= (0, 0)
Ex. Show that the function f (x) = is continuous at (0, 0) but not differ-
0 if (x, y) = (0, 0)
entiable.

Sol. Let ϵ > 0 be given. Then

x2 + y 2 [|x| + |y|]2
|f (x, y) − f (0, 0)| = ≤ = |x| + |y| < ϵ if |x| < ϵ/2 and |y| < ϵ/2.
|x| + |y| |x| + |y|
Choosing δ = ϵ/2, we have
|x| < δ, |y| < δ =⇒ |f (x, y) − f (0, 0)| < ϵ.

15
 Figure 14.10: Graph of f (x, y) shows that the surface is continuous but with a sharp edge at (0, 0, 0).

This shows that f (x, y) is continuous at (0, 0).

f (0 + ∆x, 0) − f (0, 0) ∆x
Now, fx (0, 0) = lim = lim , which does not exist.
∆x→0 ∆x ∆x→0 |∆x|
f (0, 0 + ∆y) − f (0, 0) ∆y
Similarly, fy (0, 0) = lim = lim does not exist.
∆y→0 ∆y ∆y→0 |∆y|
So f (x, y) is not differentiable at (0, 0) since existence of partial derivatives is necessary for differentiability.
Differentiability implies continuity: Let a function f (x, y) be differentiable at (x0 , y0 ). Then fx (x0 , y0 )
and fy (x0 , y0 ) both exists and
f (x0 + ∆x, y0 + ∆y) − f (x, y) = fx (x0 , y0 )∆x + fy (x0 , y0 )∆y + ∆x ϕ(∆x, ∆y) + ∆y ψ(∆x, ∆y),
where ϕ(∆x, ∆y) → 0 and ψ(∆x, ∆y) → 0 as (∆x, ∆y) → (0, 0). So in the limit (∆x, ∆y) → (0, 0), we
get lim f (x0 + ∆x, y0 + ∆y) = f (x0 , y0 ). This shows that f (x, y) is continuous at (x0 , y0 ).
(∆x,∆y)→(0,0)
However, a continuous function need notxybe differentiable as we have seen in the previous example.
x2 +y 2
if (x, y) ̸= (0, 0)
Ex. Show that the function f (x, y) = is not continuous at (0, 0) but par-
0 if (x, y) = (0, 0)
tial derivatives exist.
Sol. We shall show that the limit lim f (x, y) is path dependent.
(x,y)→(0,0)
Suppose (x, y) → (0, 0) along the straight line path y = mx. Then
xy x.mx m
lim f (x, y) = lim 2 2
= lim 2 2 2
= .
(x,y)→(0,0) (x,y)→(0,0) x + y x→0 x +m x 1 + m2
So the given limit depends on m, the slope of the path line. Hence, it get different values for different
values of m, and consequently it does not exist. So f (x, y) is not continuous at (0, 0).
f (0 + ∆x, 0) − f (0, 0) 0
Now, fx (0, 0) = lim = lim = 0.
∆x→0 ∆x ∆x→0 ∆x
f (0, 0 + ∆y) − f (0, 0) 0
Similarly, fy (0, 0) = lim = lim = 0.
∆y→0 ∆y ∆y→0 ∆y
Note: From the previous two examples, we may notice that continuity of f (x, y) has nothing to do with
its partial derivatives.

Sufficient condition for differentiability: If the partial derivatives fx and fy are continuous at (x0 , y0 ),
then f (x, y) is differentiable at (x0 , y0 ).
Geometry of partial derivative: Consider the partial derivative fx (x0 , y0 ) of f (x, y) at (x0 , y0 ). The
partial derivative fx (x0 , y0 ), in fact, is the ordinary derivative of f (x, y0 ), which is the curve in XYZ-space
determined by the section of the plane y = y0 with the surface z = f (x, y). Therefore, fx (x0 , y0 ) is the
slope of the tangent to the curve z = f (x, y0 ) at the point (x0 , y0 , f (x0 , y0 )) as shown in Figure 10.

16
 Figure 14.11:
 
Ex. Find fx and fy given that f (x, y) = tan−1 xy .
Sol. We have
y2 1
 
1 ∂ x y
fx = 2 = 2 2
= 2 .
1+ 2 x ∂x y x +y y x + y2
y

y 2 −x
 
1 ∂ x −x
fy = x2
= 2 2 2
= 2 .
1+ ∂y y x +y y x + y2
y2

14.6 The Chain Rule

In the following, we state some chain rules of differentiation without proof.
• If f is differentiable function of x, and x is differentiable function of t, then f is differentiable
function of t, and
df df dx
= .
dt dx dt
Here, to reach from f to t, there is one path, namely, f to x and then x to t.

• If f is differentiable function of x, y, and x, y are differentiable function of t, then f is differentiable

function of t, and
df ∂f dx ∂f dy
= + .
dt ∂x dt ∂y dt
In this formula, two terms are appearing in addition because there are two possible paths to reach
from f to t. One is from f to x and x to t. Second is from f to y and y to t. See the tree diagram
14.12

17
 f

x y

t t

Figure 14.12: There are two paths in the tree diagram: f → x → t, and f → y → t.

• If f is differentiable function of x, y, and x, y are differentiable function of u, v, then f is differentiable

function of u, v and
∂f ∂f ∂x ∂f ∂y
= + ,
∂u ∂x ∂u ∂y ∂u
∂f ∂f ∂x ∂f ∂y
= + .
∂v ∂x ∂v ∂y ∂v
Here, we have two paths from f to u, namely, f → x → u and f → y → u. Similarly, the two paths
from f to v are f → x → v and f → y → v. See the tree diagram 14.13.

x y

u v u v

Figure 14.13: There are four paths in the tree diagram: f → x → u, f → y → u, f → x → v and
f → y → v.

Likewise, we can write the chain rule for a composite function f with any number of variables.

Ex. Use the Chain Rule to find the derivative of f (x, y) = xy with respect to t along the path x = cos t,
y = sin t.
Sol. We have
df ∂f dx ∂f dy
= + = y(− sin t) + x cos t = − sin2 t + cos2 t = cos 2t.
dt ∂x dt ∂y dt
∂g ∂g
Ex. If g(x, y, z) = f (x − y, y − z, z − x), show that ∂x + ∂y + ∂g
∂z = 0.
Sol. Let p = x − y, q = y − z and r = z − x so that f is a function of p, q, r, and p, q, r are functions of
x, y, z. So
∂g ∂f ∂p ∂f ∂q ∂f ∂r ∂f ∂f ∂f ∂f ∂f
= + + = .1 + .0 + (−1) = − .
∂x ∂p ∂x ∂q ∂x ∂r ∂x ∂p ∂q ∂r ∂p ∂r

18
Similarly,
∂g ∂f ∂p ∂f ∂q ∂f ∂r ∂f ∂f
= + + = −
∂y ∂p ∂y ∂q ∂y ∂r ∂y ∂q ∂p
and
∂g ∂f ∂p ∂f ∂q ∂f ∂r ∂f ∂f
= + + = − .
∂z ∂p ∂z ∂q ∂z ∂r ∂z ∂r ∂q
∂g ∂g ∂g
Adding ∂x , ∂y and ∂z , we get the required result.

14.6.1 Implicit Differentiation

Let f (x, y) = 0 be an implicit relation connecting x and y. Then y may be regarded as function of x via
the relation f (x, y) = 0. Thus, f is function of x, y, and x, y both may be regarded as functions of x. So

df ∂f dx ∂f dy ∂f ∂f dy
= + = + .
dx ∂x dx ∂y dx ∂x ∂y dx
df
Also, f (x, y) = 0 implies dx = 0. Thus, we obtain

∂f
dy ∂x fx
= − ∂f =− .
dx fy
∂y

dy
This formula expresses dx in terms of the partial derivatives of f (x, y). It eases the life when we calculate
dy
dx from an implicit relation in x and y.

dy
Ex. Find dx from the relation xy + y x = 1.
Sol. Assuming f (x, y) = xy + y x − 1, we get

dy fx yxy−1 + y x ln y
=− = y .
dx fy x ln x + xy x−1
∂z ∂z
Ex. Find ∂x and ∂y from the relation x3 + z 2 + yexz + z cos y = 0.
Sol. Assuming f (x, y, z) = x3 + z 2 + yexz + z cos y, we get

∂z fx 3x2 + yzexz
=− =− ,
∂x fz 2z + xyexz + cos y
∂z fy exz − z sin y
=− =− .
∂y fz 2z + xyexz + cos y

14.6.2 Higher Order Partial Derivatives

Let f (x, y) be a function of x and y. Then, as we have seen earlier, its first order derivatives are

∂f f (x + h, y) − f (x, y)
fx (x, y) = = lim
∂x h→0 h
and
∂f f (x, y + k) − f (x, y)
fy (x, y) = = lim .
∂y k→0 k

19
The second order partial derivatives of f (x, y) are

∂2f
 
∂ ∂f fx (x + h, y) − fx (x, y)
fxx (x, y) = 2
= = lim ,
∂x ∂x ∂x h→0 h

∂2f
 
∂ ∂f fx (x, y + k) − fx (x, y)
fxy (x, y) = = = lim ,
∂y∂x ∂y ∂x k→0 k
∂2f
 
∂ ∂f fy (x + h, y) − fy (x, y)
fyx (x, y) = = = lim ,
∂x∂y ∂x ∂y h→0 h
∂2f
 
∂ ∂f fy (x, y + k) − fy (x, y)
fyy (x, y) = 2
= = lim .
∂y ∂y ∂y k→0 k
Likewise we can define third order and higher order partial derivatives.

Ex. Find 2nd order partial derivatives of f (x, y) = x3 + x2 y 2 + y 3 .

Sol. We have fx = 3x2 + 2xy 2 , fy = 2x2 y + 3y 2 .

fxx = 6x + 2y 2 , fxy = 4xy, fyx = 4xy, fyy = 2x2 + 6y.

Caution: While dealing with problems in your text book, you will notice that fxy = fyx . In other words,
it does not matter in which order we calculate the partial derivatives. However, you should keep in mind
that the equality fxy = fyx is ensured only when fxy and fyx are continuous functions, otherwise fxy and
fyx need not be equal.
For example,
( consider the function
xy(x2 −y 2 )
x2 +y 2
if (x, y) ̸= (0, 0)
f (x, y) =
0 if (x, y) = (0, 0)
Then it can be shown that fyx (0, 0) = 1 and fxy (0, 0) = −1. So fxy (0, 0) ̸= fyx (0, 0).
For,

fy (h, 0) − fy (0, 0)
fyx (0, 0) = lim .
h→0 h
Next,
hk(h2 −k2 )
f (h, k) − f (h, 0) h2 +k2
−0
fy (h, 0) = lim = lim =h
k→0 k k→0 k
and
f (0, k) − f (0, 0) 0−0
fy (0, 0) = lim = lim = 0.
k→0 k k→0 k
It follows that
fy (h, 0) − fy (0, 0) h−0
fyx (0, 0) = lim = lim = 1.
h→0 h h→0 h
Likewise, we get

fx (0, k) − fx (0, 0) −k − 0
fxy (0, 0) = lim = lim = −1.
k→0 k k→0 k

20
14.7 Directional Derivatives and Gradient Vectors
The directional derivative of a function
 f (x, y, z) at a point (x, y, z) in the direction of a unit vector
df
n̂ = lî + mĵ + nk̂, denoted by dt or Dn̂ f , is defined as the limit
n̂
 
df f (x + lt, y + mt, z + nt) − f (x, y, z)
= lim ,
dt n̂ t→0 t

provided it exists. Note that (x + lt, y + mt, z + nt) is a neighbouring point of (x, y, z) on a line through
(x, y, z) in the direction of n̂ = lî + mĵ + nk̂. In particular, the directional derivative of f (x, y, z) in a
direction parallel to x-axis, where n̂ = î, reads as
 
df f (x + t, y, z) − f (x, y, z)
= lim ,
dt î t→0 t

which is nothing but the partial derivative of f with respect to x, that is, fx . Similarly, directional
derivatives of f (x, y, z) in directions parallel to y-axis and z-axis are fy and fz , respectively.
  So we see
df
that the partial derivatives fx , fy and fz are special cases of the directional derivative dt n̂ of f (x, y, z).
The parametric equations of the line through (x, y, z) in the direction n̂ = lî+mĵ +nk̂ are x(t) = x+lt,
y(t) = y + mt and z(t) = z + nt. So by chain rule, we have
 
df ∂f ∂f ∂f ∂f ∂f ∂f
= l+ m+ n= î + ĵ + k̂ .(lî + mĵ + nk̂).
dt ∂x ∂y ∂z ∂x ∂y ∂z
 
So the directional derivative df
dt may be written as
n̂
 
df
= ∇f.n̂ ,
dt n̂

∂f ∂f ∂f
where ∇f = ∂x î + ∂y ĵ + ∂z k̂ is known as gradient vector of f (x, y, z).
 
df
This vector representation of directional derivative is useful in view of the fact that dt n̂ represents
the rate of change of f (x, y, z) in the direction n̂. If θ is angle between ∇f and n̂, then we have
 
df
= |∇f | cos θ.
dt n̂

Obviously, maximum value of directional derivative is |∇f |, and is obtained in the direction of ∇f .
Similarly, minimum value of directional derivative is −|∇f |, in the direction opposite to ∇f . It is 0 in
the direction perpendicular to ∇f .
Let P (x, y, z) be a point on a level surface f (x, y, z) = c, where c is a constant. Suppose → −r (t) =
x(t)î + y(t)ĵ + z(t)k̂ be any curve on the level surface f (x, y, z) = c passing through the point P (x, y, z)
so thatf (x(t), y(t), z(t)) = c. Therefore, we have

df ∂f dx ∂f dy ∂f dz d→
−r
0= = + + = ∇f. .
dt ∂x dt ∂y dt ∂z dt dt
−
→
Since ddtr is tangent vector to each curve → −r (t) = x(t)î + y(t)ĵ + z(t)k̂ on the level surface f (x, y, z) = c
passing through the point P (x, y, z), so it follows that the gradient vector ∇f is normal to the level surface
at P (x, y, z). Further, it implies that the directional derivative at any point of a level surface is maximum
along the direction of normal to the surface at that point.

21
Ex. Find ∇f at the point (1, −2, −1), given that f = 3x2 y − y 3 z 2 .
Ans. −12î − 9ĵ − 16k̂.
Ex. Find the directional derivative of f = x2 − y 2 + 2z 2 at the point P (1, 2, 3) in the direction of line P Q,
where Q is the point (5,0,4). Also find the direction and magnitude of maximum directional derivative at
P. √ √
Ans. 4 21/3, 2î − 4ĵ + 12k̂, 2 41.

Ex. If →
−
r = xî + y ĵ + z k̂, then show that ∇rn = nrn−2 →
−
r.

Ex. Find the angle between the normals to the surfaces x2 + y 2 + z 2 = 9 and z = x2 + y 2 − 3 at the
point (2, −1, 2). Also, find the equation of tangent line at (2, −1, 2) to the curve of intersection of the two
surfaces.
Sol. Angle between two surfaces at a point of intersection is equal to the angle between their normals at
that point. Let f = x2 +y 2 +z 2 −9 = 0 and g = x2 +y 2 −z−3 = 0. Then ∇f and  ∇g at(2, −1, 2) are normal
∇f.∇g
√
vectors to the two given surfaces at (2, −1, 2). So required angle is cos−1 |∇f ||∇g| = cos−1 (8 21/63).
Next, the tangent line at (2, −1, 2) is the intersection of the tangent planes to the two surfaces at (2, −1, 2).
So the tangent line at (2, −1, 2) is perpendicular to the plane of the vectors ∇f and ∇g at (2, −1, 2). Thus,
the vector ∇f ×∇g at (2, −1, 2) is parallel to the tangent line at (2, −1, 2). So it gives direction ratios of the
tangent line, and it is easy to write the equation of tangent line at (2, −1, 2) by using x−x l
0
= y−y 0
m = n .
z−z0

Ex. Find the values of constants a and b so that the surfaces ax2 − byz = (a + 2)x and 4x2 y + z 3 = 4
may intersect orthogonally at the point (1, −1, 2).
Ans. a = 2.5, b = 1.
(
0, y = x
Ex. Let f (x, y) = .
1, y ̸= x
Show that the directional derivative exists at (0, 0) along the line y = x though ∇f does not exists at
(0, 0).
So. The unit vector along the line y = x is n̂ = √12 î + √12 ĵ. Also, f (x, y) = 0 along y = x. So the
directional derivative of f at (0, 0) along the line y = x is given by
 
df f (0 + √12 t, 0 + √12 t) − f (0, 0) 0−0
= lim = lim = 0.
dt n̂ t→0 t t→0 t
Now, ∇f at (0, 0) is given by
∇f = fx (0, 0)î + fy (0, 0)ĵ + fz (0, 0)k̂.
It does not exist since
f (h, 0) − f (0, 0) 1−0
fx (0, 0) = lim = lim
h→0 h h→0 h
does not exist.

Also, from ∇f = fx (0, 0)î + fy (0, 0)ĵ + fz (0, 0)k̂, it is clear that existence of ∇f at (0, 0) requires the
existence of all the three directional derivatives fx (0, 0), fy (0, 0) and fz (0, 0) of f at (0, 0) along x-axis,
y-axis and z-axis, respectively.
(
0, xy ̸= 0
Ex. Let f (x, y) = .
1, xy = 0
Show that the directional derivative does not exist at (0, 0) along the line y = x though ∇f exists at
(0, 0).

22
14.8 Tangent Planes and Differentials
The tangent plane at the point P0 (x0 , y0 , z0 ) on the level surface f (x, y, z) = c of a differentiable function
f is the plane through P0 normal to ∇f |P0 . So direction ratios of the normal to the tangent plane at
P0 (x0 , y0 , z0 ) are fx (P0 ), fy (P0 ) and fz (P0 ). So equation of the tangent plane is

fx (P0 )(x − x0 ) + fy (P0 )(y − y0 ) + fz (P0 )(z − z0 ) = 0,

and the equation of the normal line at P0 is given by

x − x0 y − y0 z − z0
= = .
fx (P0 ) fy (P0 ) fz (P0 )

Ex. Find the equation of tangent plane and normal to the sphere x2 + y 2 + z 2 = 1 at the point (0, 0, 1).
Sol. Normal vector to the given sphere is
→
−n = ∇(x2 + y 2 + z 2 ) = 2xî + 2y ĵ + 2z k̂ = 2k̂, at the point (0, 0, 1). So direction ratios of the normal to
the tangent plane at (0, 0, 1) are 0, 0 and 2. So equation of the tangent plane at (0, 0, 1) is

0(x − 0) + 0(y − 0) + 2(z − 1) = 0 or z = 1.

Equation of normal at (0, 0, 1) is given by

x−0 y−0 z−0
= = ,
0 0 2
which is nothing but the equation of z-axis, as expected.

14.8.1 Estimating the change in f in a given direction

To estimate the change in the value of a differentiable function f when we move a small distance ds from
a point P0 in a particular direction n̂, we use the formula

df = (∇f |P0 ).n̂(ds).

Ex. Estimate how much the value of f (x, y, z) = y sin x + 2yz will change if the point P (x, y, z) moves
0.1 unit from P0 (0, 1, 0) straight towards P1 (2, 2, −2).
−−−→
P0 P1
Sol. We find n̂ = −−−→ = 23 î + 13 ĵ − 23 k̂.
|P0 P1 |
Next, ∇f |P0 = y cos xî + (sin x + 2z)ĵ + 2y k̂|P0 (0,1,0) = î + 2k̂.
So with ds = 0.1, the resulting change in f is

df = (∇f |P0 ).n̂(ds) = (−2/3)(0.1) ≈ −0.067

14.8.2 Taylor’s formula and linearization of f (x)

If a function f (x) possesses continuous derivatives up to order n in the neighborhood N (x0 ) of a point
x0 , then for any x ∈ N (x0 ), there exists some θ ∈ (0, 1) such that
n−1
X 1 1
f (x) = (x − x0 )k f (k) (x0 ) + (x − x0 )n f (n) (x0 + (x − x0 )θ).
k! n!
k=0

1 n (n) (x )
It is called Taylor’s formula with the remainder term Rn+1 = n! (x − x0 ) f 0 after the n terms. The
values of x for which Rn+1 → 0 as n → ∞, we have
∞
X 1
f (x) = (x − x0 )k f (k) (x0 ).
k!
k=0

23
This is called the Taylor series expansion of f (x) about x = x0 .

When n = 2, the Taylor’s formula reads

1 ′′
f (x) = f (x0 ) + f ′ (x0 )(x − x0 ) + (x − x0 )2 f (x0 + (x − x0 )θ).
2!
The approximation using the first two terms, that is,

f (x) ≈ f (x0 ) + f ′ (x0 )(x − x0 ),

is called linearization of f (x) about x = x0 . Since f (x) = f (x0 ) + f (x0 )(x − x0 ) is tangent to the curve
f (x) at x = x0 , the linearization is geometrically the tangent line approximation of the curve f (x) in the
close vicinity of (x0 , f (x0 )). Also, the absolute error in the linear approximation or linearization is given
by
1 ′′
E(x) = (x − x0 )2 |f (x0 + (x − x0 )θ)|.
2!
For example, sin x ≈ x is the linear approximation of sin x about x = 0, and the absolute error is
given by
1 2 1
E(x) = x | cos(θx)| ≤ x2 ≤ 0.5,
2! 2!
in case we choose x ∈ [−1, 1]. It means the maximum possible error is 0.5 in the linear approximation
sin x ≈ x for x ∈ [−1, 1].

14.8.3 Taylor’s formula and linearization of f (x, y)

If a function f (x, y) possesses continuous partial derivatives up to order n in the neighborhood N (x0 , y0 )
of a point (x0 , y0 ), then for any (x, y) ∈ N (x0 ), there exists some θ ∈ (0, 1) such that
n−1
∂ k
 
X 1 ∂
f (x, y) = (x − x0 ) + (y − y0 ) f (x0 , y0 ) + Rn+1 ,
k! ∂x ∂y
k=0

where
∂ n
 
1 ∂
Rn+1 = (x − x0 ) + (y − y0 ) f (x0 + (x − x0 )θ, y0 + (y − y0 )θ)
n! ∂x ∂y

is the remainder term after the n terms. The values of (x, y) for which Rn+1 → 0 as n → ∞, we have
∞
∂ k
 
X 1 ∂
f (x, y) = (x − x0 ) + (y − y0 )(x − x0 ) f (x0 , y0 ).
k! ∂x ∂y
k=0

This is called the Taylor series expansion of f (x, y) about (x0 , y0 ).

When n = 2, the Taylor’s formula reads

f (x, y) = f (x0 , y0 ) + [fx (x0 , y0 )(x − x0 ) + fy (x0 , y0 )(y − y0 )] + R3 ,

where
1 
fxx (c, d)(x − x0 )2 + 2fxy (c, d)(x − x0 )(y − y0 ) + fyy (c, d)(y − y0 )2

R3 =
2!
with c = x0 + (x − x0 )θ) and d = y0 + (y − y0 )θ).

24
 The approximation using the first two terms, that is,

f (x, y) ≈ f (x0 , y0 ) + [fx (x0 , y0 )(x − x0 ) + fy (x0 , y0 )(y − y0 )],

is called linearization of f (x, y) about (x0 , y0 ). Since f (x, y) = f (x0 , y0 )+[fx (x0 , y0 )(x−x0 )+fy (x0 , y0 )(y−
y0 ) is tangent plane to the surface z = f (x, y) at (x0 , y0 , f (x0 , y0 )), the linearization is geometrically the
tangent plane approximation of the surface z = f (x, y) in the close vicinity of (x0 , y0 , f (x0 , y0 )). Also,
the absolute error in the linear approximation or linearization is given by E(x, y) = |R3 |.
If f has continuous first and second partial derivatives throughout an open set containing a rectangle
R centered at (x0 , y0 ) and if M is any upper bound for the values of |fxx |, |fxy | and |fyy | on R, then the
error E(x, y) occurred in replacing f (x, y) on R by its linearization

L(x, y) = f (x0 , y0 ) + fx (x0 , y0 )(x − x0 ) + fy (x0 , y0 )(y − y0 )

satisfies the inequality

1
|E(x, y)| ≤ M (|x − x0 | + |y − y0 |)2 .
2
Ex. Find the linearization of f (x, y) = x2 − xy + 12 y 2 + 3 at the point (3, 2). Also find an upper bound
for the error in the approximation over the rectangle R : |x − 3| ≤ 0.1, |y − 2| ≤ 0.1.
Sol. We find

L(x, y) = f (x0 , y0 )+fx (x0 , y0 )(x−x0 )+fy (x0 , y0 )(y−y0 ) = f (3, 2)+fx (3, 2)(x−3)+fy (3, 2)(y−2) = 4x−y−2.

We know that the error satisfies the inequality

1
|E(x, y)| ≤ M (|x − x0 | + |y − y0 |)2 .
2
To find a suitable value for M , we calculate |fxx | = 2, |fxy | = 1 and |fyy | = 1. The largest of these is
2, so we may safely take M to be 2. With (x0 , y0 ) = (3, 2), over the given rectangle R : |x − 3| ≤ 0.1,
|y − 2| ≤ 0.1, we find
1
|E(x, y)| ≤ (2)(|x − 3| + |y − 2|)2 = 0.04.
2

14.8.4 Differentials
Suppose the first order partial derivatives of a function f (x, y) exist at (x0 , y0 ). If x and y change from
x0 and y0 by small amounts dx and dy respectively, then the total differential

df = fx (x0 , y0 )dx + fy (x0 , y0 )dy

gives a good approximation of the resulting change in f .

Ex. Suppose that a cylindrical can is designed to have a radius of 1 in. and a height of 5 in., but that
the radius and height are off by the amounts dr = +0.03 and dh = −0.1. Estimate the resulting absolute
change in the volume of the can.
Sol. To estimate the absolute change in V = πr2 h, we use

∆V ≈ dV = Vr (r0 , h0 )dr + Vh (r0 , h0 )dh.

With Vr = 2πrh and Vh = πr2 , we get

dV = 2πr0 h0 dr + πr02 (r0 , h0 )dh = 2π(1)(5)(0.03) + π(1)2 (−0.1) = 0.2π ≈ 0.63 in3 .

25
14.9 Extreme Values and Saddle Points
14.9.1 Maxima and minima of a function of one variable
Let f be a continuous function in some interval I, and x0 ∈ I. Then (x0 , f (x0 )) is said to be
(i) local maximum point of f (x) if f (x) − f (x0 ) remains negative in some deleted nbd of x0 , that is, there
exists some δ > 0 such that f (x) − f (x0 ) < 0 for all x ∈ (x0 − δ, x0 + δ) − {x0 },
(ii) local minimum point of f (x) if f (x) − f (x0 ) remains positive in some deleted nbd of x0 , that is, there
exists some δ > 0 such that f (x) − f (x0 ) > 0 for all x ∈ (x0 − δ, x0 + δ) − {x0 },
(iii) inflection point of f (x) if f (x) − f (x0 ) attains positive as well as negative values in every nbd of x0 ,
(iv) global maximum or absolute maximum point of f (x) if f (x) − f (x0 ) ≤ 0 for all x ∈ I,
(v) global minimum or absolute minimum point of f (x) if f (x) − f (x0 ) ≥ 0 for all x ∈ I.

Critical point
A critical point of a function is a point where its first derivative vanishes. So (x0 , f (x0 )) is a critical point
of f (x) if f ′ (x0 ) = 0.
It is easy to verify that the local maximum and minimum points of f (x) are its critical points.
For, if f (x) has local maximum value at x = x0 , then f (x) < f (x0 ) throughout some deleted nbd of
x0 . So we have
f (x) − f (x0 )
f ′ (x0 ) = lim ≥0 (∵ x − x0 < 0)
x→x−
0
x − x0

and
f (x) − f (x0 )
f ′ (x0 ) = lim ≤ 0. (∵ x − x0 > 0)
x→x+
0
x − x0

It follows that f ′ (x0 ) = 0. Likewise is the case of local minimum.

14.9.2 First derivative test

Suppose x = x0 is a critical point of f (x). If there exists some nbd of x0 in which f ′ (x) changes its sign
from positive to negative, then (x0 , f (x0 )) is a local maximum point of f (x). This happens because as
x goes from left to right in the nbd of x0 , the angle of the tangent to the curve changes from acute to
obtuse, and consequently slope of the tangent, that is, f ′ (x) changes its sign from positive to negative.
Note that f ′ (x0 ) = 0, so the tangent line to the curve at (x0 , f (x0 )) is parallel to x-axis. Likewise, if f ′ (x)
changes its sign from negative to positive in some nbd of x0 , then (x0 , f (x0 )) is a local minimum point of
f (x).

14.9.3 Second derivative test

Suppose (x0 , f (x0 )) is a critical point of f (x) so that f ′ (x0 ) = 0. Then by Taylor’s formula with remainder
term after the two terms, we have
1
f (x) − f (x0 ) = (x − x0 )2 f ′′ (x0 + (x − x0 )θ),
2!
where 0 < θ < 1, and x lies in some nbd of x0 where continuous derivatives of f exist up to second order.
The continuity of f ′′ at x0 implies that there exists some nbd of x0 in which f ′′ (x) remains either negative
or positive according to f ′′ (x0 ) < 0 or f ′′ (x0 ) > 0 (a well known property of continuous functions!). It
implies that f ′′ (x0 + (x − x0 )θ) keeps the sign of f ′′ (x0 ) in some nbd of x0 . Therefore, if f ′′ (x0 ) < 0,
then from the above Taylor’s formula, we obtain f (x) − f (x0 ) < 0 in some deleted nbd of x0 . It implies

26
that (x0 , f (x0 )) is a local maximum point of f (x). Similarly, f ′′ (x0 ) < 0 implies that (x0 , f (x0 )) is a local
minimum point of f (x).
In case, f ′′ (x0 ) = 0, we can use the Taylor’s formula with remainder after three terms to obtain
1
f (x) − f (x0 ) = (x − x0 )3 f ′′′ (x0 + (x − x0 )θ).
3!
If f ′′′ (x0 ) ̸= 0, then continuity of f ′′′ implies that f ′′′ (x0 + (x − x0 )θ) ̸= 0 retains the same sign as of
f ′′′ (x0 ) in some nbd of x0 . It follows that f (x) − f (x0 ) attains positive as well as negative values in every
nbd of x0 . So (x0 , f (x0 )) is an inflection point of f (x).
In case, f ′′′ (x0 ) also vanishes, we can decide local maximum or minimum based on the sign of fourth
derivative of f at x0 as we did above in case of second derivative.
Ex. Show that (0, 0) is local minimum point of f (x) = x2 .
Sol. Here f ′ (x) = 2x. So f ′ (x) = 0 implies x = 0. We see that f ′ (x) = 2x changes its sign from negative
to positive in the nbd (−0.1, 0.1) of x = 0. So by the first derivative test, f (x) has local minimum at
x = 0, and the local minimum value is f (0) = 0.
Also, f ′′ (x) = 2 is positive at x = 0. So the second derivative test implies that f (x) has local minimum
at x = 0.

Ex. Show that f (x) = x3 has point of inflection at x = 0.

Sol. Here f ′ (x) = 3x2 . So f ′ (x) = 0 implies x = 0. Also, f ′′ (x) = 6x = 0 and f ′′′ (x) = 6 ̸= 0 at x = 0.
This shows that (0, f (0)) = (0, 0) is point of inflection of f (x) = x3 . Also notice that in every nbd of
x = 0, we have f (x) < f (0) for x < 0, and f (x) > f (0) for x > 0, as expected.

Ex. Find global maximum/minimum values of f (x) = x2 in the interval [−1, 2].
Sol. In [−1, 2], we find that f (x) has critical point at x = 0. Also, f (−1) = 1, f (0) = 0 and f (2) = 4 So
in [−1, 2], global minimum value of f (x) = x2 is 0, and global maximum value is 4. Global minimum and
maximum points are (0, 0) and (2, 4), respectively.
In case, we are given to find the global maximum/minimum values of f (x) = x2 in the open interval
(−1, 2), then the global minimum value f (0) = 0 is same as in the previous case of closed interval. But
we can not tell the global maximum value of f (x) = x2 in the open interval (−1, 2) because x = 2 does
not belong to this domain interval of f (x) = x2 .
Thus, if domain of a function is a closed interval, we can find its global maximum and minimum values
by considering the values of the function at the end points of the interval and the x-values of critical points
lying inside the interval. In case, of open interval domain, we consider only the x-values of critical points
lying inside the interval for determining global maximum/minimum values of the function.

27
14.10 Maxima and minima of a function of two variables
Let f be a continuous function defined in some region D of xy-plane, and (x0 , y0 ) ∈ D. Then (x0 , y0 , f (x0 , y0 ))
is said to be
(i) local maximum point of f (x, y) if f (x, y) − f (x0 , y0 ) remains negative in some deleted nbd of (x0 , y0 ),
(ii) local minimum point of f (x, y) if f (x, y) − f (x0 , y0 ) remains positive in some deleted nbd of (x0 , y0 ),
(iii) saddle point1 of f (x, y) if f (x, y) − f (x0 , y0 ) attains positive as well as negative values in every nbd
of (x0 , y0 ),
(iv) global maximum or absolute maximum point of f (x, y) if f (x, y) − f (x0 , y0 ) ≤ 0 for all (x, y) ∈ D,
(v) global minimum or absolute minimum point of f (x, y) if f (x, y) − f (x0 , y0 ) ≥ 0 for all (x, y) ∈ D.

Figure 14.14: Left: The person is standing at the saddle point of the hill surface. Is’nt it? Right: How
many saddle points you eat in a day:)

Critical point
A critical point of a function f (x, y) is a point where its first order partial derivatives vanishes. So
(x0 , y0 , f (x0 , y0 )) is a critical point of f (x, y) if fx (x0 , y0 ) = 0 and fx (x0 , y0 ) = 0.
It is easy to verify that the local maximum and minimum points of f (x, y) are its critical points.
1
Consider the point at the center of the surface of a saddle (seat used on a horse) placed on the back of a horse. It is easy
to imagine that the surface points rise as we move on the surface from the center along the horse while the surface points
fall as we move on the surface from the center perpendicular to the horse. Clearly, in every nbd of the center of the this
surface, there are points which are at more height than the center as well as there are points which are at less height than
the center. It implies that center of the surface of saddle is a saddle point. (That is why the name saddle point!)

28
 For, if f has a local extremum (maximum or minimum) at (x0 , y0 ), then the function g(x) = f (x, y0 )
has a local extremum at x = x0 . Therefore, g ′ (x0 ) = 0. Now g ′ (x0 ) = fx (x0 , y0 ), so fx (x0 , y0 ) = 0. A
similar argument with the function h(y) = f (x0 , y) shows that fy (x0 , y0 ) = 0.

14.10.1 Test for maxima and minima of f (x, y)

Suppose (x0 , y0 ) is a critical point of f (x, y) so that fx (x0 , y0 ) = 0 and fy (x0 , y0 ) = 0. Then by Taylor’s
formula with the remainder term after the two terms, we have
1 
fxx (c, d)(x − x0 )2 + 2fxy (c, d)(x − x0 )(y − y0 ) + fyy (c, d)(y − y0 )2

f (x, y) − f (x0 , y0 ) =
2!
1
= [((x − x0 )fxx + (y − y0 )fxy )2 + (y − y0 )2 (fxx fyy − fxy
2
)](c,d)
2fxx
with c = x0 + (x − x0 )θ and d = y0 + (y − y0 )θ. If fxx and fxx fyy − fxy 2 are nonzero at (x , y ), then these
0 0
2
retain the respective signs of fxx (x0 , y0 ) and [fxx fyy − fxy ](x0 ,y0 ) in some nbd of (x0 , y0 ) due to continuity
of the second order partial derivatives at (x0 , y0 ). It implies that fxx (c, d) and [fxx fyy − fxy 2 ]
(c,d) retain
2
the respective signs of fxx (x0 , y0 ) and [fxx fyy − fxy ](x0 ,y0 ) in some nbd of (x0 , y0 ). So we calculate fxx
and fxx fyy − fxy 2 at (x , y ), and have the following cases:
0 0
2 > 0 and f
(i) If fxx fyy − fxy xx < 0, then f (x, y) − f (x0 , y0 ) remains negative in some deleted nbd of
(x0 , y0 ). Therefore, f (x, y) has local maximum at (x0 , y0 ).
2 > 0 and f
(ii) If fxx fyy − fxy xx > 0, then f (x, y) − f (x0 , y0 ) remains positive in some deleted nbd of
(x0 , y0 ). Therefore, f (x, y) has local minimum at (x0 , y0 ).
2 < 0, then f (x, y) − f (x , y ) attains positive as well as negative values in every
(iii) If fxx fyy − fxy 0 0
nbd of (x0 , y0 ). So f (x, y) has saddle point at (x0 , y0 ).
2 = 0, then we can not decide in general and further investigation is required. In this
(iv) If fxx fyy − fxy
case, we directly check the sign of f (x, y)−f (x0 , y0 ) in the nbd of (x0 , y0 ), and take the decision accordingly.

Ex. Show that (0, 0, 0) is the local minimum point of z = x2 + y 2 .

Sol. Here f (x, y) = x2 + y 2 . So fx = 2x = 0 and fy = 2y = 0 imply (x, y) = (0, 0). Next, fxx = 2,
fxy = 0, and fyy = 2. So at (0, 0), we get fxx = 2 > 0 and fxx fyy − fxy2 = 4 > 0, which in turn imply that
2 2
= (0, 0.f (0, 0)) = (0, 0, 0) is the local minimum point of z = x + y .

Ex. Show that (0, 0, 0) is the saddle point of the hyperbolic paraboloid z = x2 − y 2 .
Sol. Please try yourself.

Figure 14.15: (0, 0, 0) is the saddle point of z = x2 − y 2 .

p
Ex. Find the local maximum/minimum values of the function f (x, y) = 1 − x2 − y 2 .
Sol. Let g(x, y) = f 2 (x, y) = 1 − x2 − y 2 . Then maxima and minima of g(x, y) are same as of f (x, y).

29
Now, gx = −2x = 0 and gy = −2y = 0 imply (x, y) = (0, 0). Next at (0, 0), we have gxx = −2, gxy = 0
2 = 4 > 0 and g
and gyy = −2. Therefore, gxx gyy − gxy xx = −2 < 0. It implies that g(x, y), and hence
f (x, y) has local maximum value at (0, 0), and the maximum value reads as f (0, 0) = 1.

Note: In case, the domain of the function is a closed and bounded region R, its absolute maxi-
mum/minimum values are the maximum/minimum of the values of the function at the critical points
within R and at the boundary points of R as illustrated in the following example.

Ex. Find the absolute maximum and minimum values of

f (x, y) = 2 + 2x + 2y − x2 − y 2

on the triangular region in the first quadrant bounded by the lines x = 0, y = 0, y = 9 − x.

Sol. The given triangular region is shown in Figure 14.

Figure 14.16:

We find fx = 2 − 2x and fy = 2 − 2y. So fx = 0 and fy = 0 yield the only critical point (1, 1) in the
interior of R, and f (1, 1) = 4.
Now we consider the three sides of the triangular region OAB one by one.
Along OA, y = 0. So we have f (x, 0) = 2 + 2x − x2 , which may be treated a function of single variable
x on the interval [0, 9]. Its extreme values may occur at the end points x = 0, x = 9 and at the interior
points, where f ′ (x, 0) = 2 − 2x = 0 or x = 1. The values are f (0, 0) = 2, f (9, 0) = −61 and f (1, 0) = 3.
Along OB, x = 0. So we have f (0, y) = 2 + 2y − y 2 , which may be treated a function of single variable
y on the interval [0, 9]. Its extreme values may occur at the end points y = 0, y = 9 and at the interior
points, where f ′ (0, y) = 2 − 2y = 0 or y = 1. The values are f (0, 0) = 2, f (0, 9) = −61 and f (0, 1) = 3.
Along AB, y = 9 − x. So we have f (x, y) = −61 + 18x − 2x2 , which may be treated a function of
single variable x on the interval [0, 9]. Its extreme values may occur at the end points x = 0, x = 9 and
at the interior points, where f ′ (x, 9 − x) = 18 − 4x = 0 or x = 9/2, and therefore y = 9 − 9/2 = 9/2. The
values are f (0, 0) = 2, f (0, 9) = −61 and f (9/2, 9/2) = −41/2.
Finally we list all the values: 4, 2, −61, 3, −(41/2). So the absolute maximum value is 4, which f
assumes at (1, 1). The absolute minimum value is −61, which f assumes at (0, 9) and (9, 0).

14.11 Lagrange Multipliers

14.11.1 Constrained Maxima and Minima
We first consider a problem where a constrained maximum and minimum can be found by eliminating a
variable.

30
Ex. Find the dimensions of a box open at the top with volume 32 cubic units requiring the least material
for its construction.
Sol. Let x, y, z be length, breadth and height of the box so that its volume is xyz = 32 and surface area
is xy + 2yz + 2zx. So we need to minimize xy + 2yz + 2zx subject to the constraint xyz = 32. Using
32
z = xy into the surface area function, we get the two variable function

64 64
f (x, y) = xy + + .
x y
Applying the maxima/minima test (please try yourself), we find minimum value of f (x, y) for x = 4,
y = 4. Thus, the required dimensions of the box are 4, 4, 2.

14.11.2 Method of Lagrange Multipliers

Let f (x, y, z) be differentiable in an open region containing a smooth curve C given by →
−
r (t) = x(t)î +
y(t)ĵ + z(t)k̂. Then on the curve C, we have

df df dx df dy df dz d→
−r
= + + = ∇f. .
dt dx dt dy dt dz dt dt
−
→
At a point P (x(t), y(t), z(t)) of local maxima or minima of f on the curve C, df dr
dt = 0. So ∇f. dt = 0,
which in turn implies that ∇f is orthogonal to C at the point of maxima or minima.
Now suppose a differentiable function f (x, y, z) takes a local maximum or minimum value at some
point P0 of a surface g(x, y, z) = 0, where g(x, y, z) is differentiable function and ∇g ̸= 0. Then f takes a
local maximum or minimum value at point P0 relative to its values on every differentiable curve passing
through P0 and lying on the surface g(x, y, z) = 0. So ∇f is orthogonal to every differentiable curve C
at the point P0 . It implies that ∇f is normal to the surface g(x, y, z) = 0 at the point P0 . Also, ∇g is
normal to the surface g(x, y, z) = 0 at every point. It implies that the normal vectors ∇f and ∇g at P0
are parallel vectors, and therefore ∇f = λ∇g for some non-zero constant λ, known as Lagrange multiplier.
Thus, to find the local maximum and minimum values of f subject to the constraint g(x, y, z) = 0, we
find the values of x, y, z and λ that simultaneously satisfy the equations

∇f = λ∇g, g(x, y, z) = 0.

Note: It must be noted that Lagrange multiplier method gives the points related to the extremum
(maximum or minimum) values of the function. It does not make distinction between points of maxima
and minima. In other words, it does not tells us the exact nature of the point. We have to judge the
nature of the point from the nature of the problem under consideration. It may be treated as a drawback
of the Lagrange multiplier method.
Ex. Find the dimensions of a box open at the top with volume 32 cubic units requiring the least material
for its construction.
Sol. Let x, y, z be length, breadth and height of the box so that its volume is xyz = 32 and surface area
is xy + 2yz + 2zx. So we need to minimize xy + 2yz + 2zx subject to the constraint xyz = 32. We model
the given problem as a Lagrange multiplier problem with

f (x, y, z) = xy + 2yz + 2zx, g(x, y, z) = xyz − 32

and look for the values of x, y, z and λ that satisfy the equations

∇f = λ∇g, g(x, y, z) = 0.

These are, in fact, four equations given by

y + 2z − λyz = 0

31
x + 2z − λxz = 0
2x + 2y − λxy = 0
xyz − 32 = 0
Multiplying the first three equations by x, y and z, respectively, we get
xy + 2xz − λxyz = 0
xy + 2yz − λxyz = 0
2xz + 2yz − λxyz = 0
Now subtracting the first two and last two of the above equations, we get x = y and y = 2z, respectively.
Using these in xyz − 32 = 0, we get x = 4, y = 4 and z = 2.
2
x2 2
Ex. Find the maximum volume of the cuboid that can be inscribed inside the ellipsoid a2
+ yb2 + zc2 = 1.
2
x2 2
Sol. Let 2x, 2y and 2z be dimensions of the cuboid that can be inscribed inside the ellipsoid a2
+ yb2 + zc2 =
1. So volume of the cuboid is 8xyz. Thus, we need to maximize

f (x, y, z) = 8xyz (14.5)

subject to the constraint

x2 y 2 z 2
g(x, y, z) = + 2 + 2 − 1 = 0. (14.6)
a2 b c
So as per the method of Lagrange multipliers, in order to find the local maximum and minimum values of
f subject to the constraint g(x, y, z) = 0, we find the values of x, y, z and λ that simultaneously satisfy
the equations

∇f = λ∇g, g(x, y, z) = 0.

Now, ∇f = λ∇g leads to the following equations:

x y z
4yz + λ 2
= 0, 4zx + λ 2 = 0, 4xy + λ 2 = 0.
a b c
Multiplying these equations respectively by x, y, z, and adding the resulting equations in view of (14.6),
we get

12xyz + λ = 0.

Solving this equation with 4yz + λ ax2 = 0, we get x = √a .

3
Likewise, we get y = √b
3
and z = √c .
3
So in
8abc
view of (14.5), maximum volume of the cuboid is √ .
3 3

14.11.3 Lagrange Multipliers with Two Constraints

Suppose f assumes a local maximum or minimum value at a point P0 lying on two surfaces g1 (x, y, z) = 0
and g2 (x, y, z) = 0. Then P0 lies on the curve, say C of the intersection of the two surfaces. It follows that
∇f , ∇g1 and ∇g2 all are normal to C at P0 , and hence lie in the same plane so that ∇f = λ∇g1 + µ∇g2
for some non-zero constants λ and µ, known as Lagrange multipliers. Thus, to find the local maximum
and minimum values of f subject to two constraints g1 (x, y, z) = 0 and g2 (x, y, z) = 0, we find the values
of x, y, z, λ and µ that simultaneously satisfy the equations

∇f = λ∇g1 + µ∇g2 , g1 (x, y, z) = 0, g2 (x, y, z) = 0.

Ex. The plane x + y + z = 1 cuts the cylinder x2 + y 2 = 1 in an ellipse (Figure 16). Find the points on
the ellipse that lie closest to and farthest from the origin.
Sol. We find the extreme values of

32
 Figure 14.17:

Figure 14.18:

f (x, y, z) = x2 + y 2 + z 2

(the square of the distance from (x, y, z) to the origin) subject to the constraints

g1 (x, y, z) = x2 + y 2 − 1 = 0,

g2 (x, y, z) = x + y + z − 1 = 0.
The gradient equation ∇f = λ∇g1 + µ∇g2 leads to the following equations:
2x = 2λx + µ,
2y = 2λy + µ,
2z = µ.
Eliminating µ, we get the following equations:
(1 − λ)x = z,
(1 − λ)y = z.
Obviously, these two equations are satisfied simultaneously provided either λ = 1 and z = 0 or λ ̸= 1 and
x = y = z/(1 − λ).

33
 In case z = 0, the solution of the equations x2 + y 2 − 1 = 0 and x + y + z − 1 = 0 yields the two points
(1, 0, 0) and (0, 1, 0).
In case x = y, the solution of the equations x2 + y 2 − 1 = 0 and x + y + z − 1 = 0 gives the two points
√ √ ! √ √ !
2 2 √ − 2 2 √
P1 = , , 1 − 2 , P2 = ,− ,1 + 2 .
2 2 2 2

The distance of the points√(1, 0, 0) and √ (0, 1, 0) from the origin is 1 unit while the distances of P1 and P2
from the origin are 4 − 2 2 and 4 + 2 2 units respectively. So the points on the ellipse closest to the
origin are (1, 0, 0) and (0, 1, 0), and the point farthest from the origin is P2 .

14.12 Some useful tips

• To show that lim f (x, y) exists, either satisfy the ϵ − δ definition of limit or use the Sandwich
(x,y)→(x0 ,y0 )
theorem depending on the problem.

• To show that lim f (x, y) does not exist, prove that it is path dependent.
(x,y)→(x0 ,y0 )

• To prove the differentiability of f (x, y), either satisfy the definition of differentiability or show that
fx and fy are continuous.

• To prove the non-differentiability of f (x, y), show any of the following: (i) fx does not exist (ii) fy
does not exist (iii) f is not continuous.

• .....

34