M.Sc.

(Mathematics), SEM- I
Paper - II
ANALYSIS – I
PSMT102

Note- There will be some addition to this study material. You should download it again after few
weeks.
 CONTENT

Unit No. Title

1. Differentiation of Functions of
Several Variables
2. Derivatives of Higher Orders
3. Applications of Derivatives
4. Inverse and Implicit Function
Theorems
5. Riemann Integral - I
6. Measure Zero Set

***
 SYLLABUS
Unit I. Euclidean space

Euclidean space : inner product <

and properties, norm

Cauchy-Schwarz inequality,
properties of the norm function .

(Ref. W. Rudin or M. Spivak).

Standard topology on : open subsets of , closed subsets of , interior

and boundary of a subset of .

(ref. M. Spivak)

Operator norm of a linear transformation

( :v & }) and its properties such as:

For all linear maps and

1.

2. ,and

3.

(Ref. C. C. Pugh or A. Browder)

Compactness: Open cover of a subset of , Compact subsets of (A subset K
of is compact if every open cover of K contains a finite subover), Heine-Borel
theorem (statement only), the Cartesian product of two compact subsets of
compact (statement only), every closed and bounded subset of is compact.

Bolzano-Weierstrass theorem: Any bounded sequence in has a converging

subsequence.

Brief review of following three topics:

1. Functions and Continuity Notation: arbitary non-empty set. A

function f : A → and its component functions, continuity of a function( , δ
definition). A function f : A → is continuous if and only if for every open
subset V there is an open subset U of such that

2. Continuity and compactness: Let K be a compact subset and f : K →

be any continuous function. Then f is uniformly continuous, and f(K) is a compact
subset of .

3. Continuity and connectedness: Connected subsets of are intervals. If

is continuous where and E is connected, then f(E) is
connected.

Unit II. Differentiable functions

Differentiable functions on , the total derivative of a differentiable

function at where is open in , uniqueness of total
derivative, differentiability implies continuity.

(ref:[1] C.C.Pugh or[2] A.Browder)

Chain rule. Applications of chain rule such as:

1. Let be a differentiable curve in an open subset of . Let be

a differentiable function and let . Then

2. Computation of total derivatives of real valued functions such as

(a) the determinant function ,

(b) the Euclidean inner product function .

(ref. M. Spivak, W. Rudin )

Results on total derivative:

1. If is a constant function,then .

2. If is a linear map, then .

3. A function : is differentiable at if and only if

each is differentiable at , and .

(ref. M. Spivak).

Partial derivatives, directional derivative of a function at in the

direction of the unit vector, Jacobian matrix, Jacobian determinant. Results such
as :

1. If the total derivative of a map ( open subset of

) exists at , then all the partial derivatives exists at .

2. If all the partial derivatives of a map ( open

subset of ) exist and are continuous on , then is differentiable.

(ref. W. Rudin)

Derivatives of higher order, -functions, -functions.(ref. T. Apostol)

Unit III. Inverse function theorem and Implicit function theorem

Theorem (Mean Value Inequality): Suppose is differentiable on an

open subset of and there is a real number such that
. If the segment is contained in , then .

(ref. C. C. Pugh or A. Browder).

Mean Value Theorem: Let is a differentiable on an open subset of

. Let such that the segment is contained in . Then for every
vector there is a point such that <
>. (ref:T. Apostol)

If is differentiable on a connected open subset of and

, then f is a constant map.

Taylor expansion for a real valued -function defined on an open subset of ,

stationary points(critical points), maxima, minima, saddle points, second
derivative test for extrema at a stationary point of a real valued -function
defined on an open subset of . Lagrange's method of undetermined multipliers.
(ref. T. Apostol)

Contraction mapping theorem. Inverse function theorem, Implicit function

theorem.(ref. A. Browder)

Unit IV. Riemann Integration(15 Lectures)

Riemann Integration over a rectangle in , Riemann Integrable functions,

Continuous functions are Riemann integrable, Measure zero sets, Lebesgues
Theorem(statement only), Fubini’s Theorem and applications.

(Reference for Unit IV: M. Spivak, Calculus on Manifolds)

 5
DIFFERENTIATION OF FUNCTIONS OF SEVERAL
VARIABLES
Unit Structure
5.0 Objectives
5.1 Introduction
5.2 Total Derivative
5.3 Partial Derivatives
5.4 Directional Derivatives

5.5 Summary

5.0 OBJECTIVES
After reading this unit you should be able to

 define a differentiable function of several variables

 define and calculate the partial and directional derivatives (if they exist) of a function
of several variables
 establish the connection between the total, partial and directional derivatives of a
differentiable function at a point

5.1 INTRODUCTION
You have seen how to extend the concepts of limit and continuity to functions between
metric spaces. Another important concept is differentiation. If we try to apply this to
functions between metric spaces, we encounter a problem. We realise that apart from the
distance notion, the domain and codomain also need to have an algebraic structure. So, let us
consider Euclidean spaces like Rn, which have which have both metric and algebraic
structures. Functions between two Euclidean spaces are what we call functions of several
variables.
In this chapter we shall introduce the concept of differentiability of a function of several
variables. The extension of this concept from one to several variables was not easy. Many
different approaches were tried before this final one was accepted. The definition may seem a
little difficult in the beginning, but as you will see, it allows us to extend all our knowledge of
derivatives of functions one variable to the several variables case. You may have studied
these concepts in T. Y. So, here we shall try to go a little deeper into these concepts, and deal
with vector functions of several variables.

5.2 TOTAL DERIVATIVE

To arrive at a suitable definition of differentiability of functions of several variables,
mathematicians had to closely examine the concept of derivative of a function of a single
variable. To decide on the approach to extension of the concept, it was important to know
what was the essence and role of a derivative. So, let us recall the definition of the derivative
of a function f: R R.

f ( a  h)  f ( a )
We say that f is differentiable at a R, if the limit, lim exists.
h 0 h

f ( a  h)  f ( a )
In that case, we say that the derivative of f at a, f1(a) = lim .......(5.1)
h 0 h
So, we take the limit of the ratio of the increment in f(x) to the increment in x. Now, when our
function is defined on Rn, the increment in the independent variable will be a vector. Since
division by a vector is not defined, we cannot write a ratio similar to the one in (5.1). But
(5.1) can be rewritten as

f ( a  h)  f ( a )
lim [ − f1(a) ] = 0, or
h 0 h

lim [ ] = 0, or
h 0

lim = 0, where r(h) = f(a+h) – f(a) – f1(a).h.

h 0

So, we can write f(a+h) = f(a) + f1(a).h + r(h), .........................(5.2)

where the “remainder” r(h) is so small, that tends to zero as h tends to zero.

For a fixed a, f(a), and f1(a) are fixed real numbers. This means, except for the remainder,
r(h), (5.2) expresses f(a + h) as a linear function of h. This also helps us in “linearizing” f. We
say that for points close to a, the graph of the function f can be approximated by a line. Thus,
f1(a) gives rise to a linear function L from R to R.

L: R R, h f1(a).h, which helps us in linearizing the given function f near the given

point a. (5.2) then transforms to

f(a + h) = f(a) + L(h) + r(h) . ...........................(5.3)
It is this idea of linearization that we are now going to extend to a function of several
variables.

Definition 5.1 Suppose E is an open set in Rn, f : E Rm, and a E. We say that f is
differentiable at a, if there exists a linear transformation T : Rn Rm, such that

lim =0 .........................(5.4)
h 0

and we write f1(a) = T.

If f is differentiable at every point in E, we say that f is differentiable in E.
Remark 5.1 i) Bold letters indicate vectors.

ii) Since E is open, , such that B(a, r) E. We choose h, such that h < r, so that

a+h E.

iii) The norm in the numerator of (5.4) is the norm in Rm, whereas the one in the denominator
is the norm in Rn.
iv) The linear transformation T depends on the point a. So, when we have to deal with more
than one point, we use the notation, Ta, Tb, and so on.
We have seen that in the one variable case, the derivative defines a linear function,

h f1(a).h from R to R. Similarly, here the derivative is a linear transformation from Rn to

Rm. With every such transformation, we have an associated m  n matrix. The jth column of
this matrix is T(ej), where ej is a basis vector in the standard basis of Rn.

For a given point a, the linear transformation Ta is called the total derivative of f at a, and is
denoted by f1(a) or Df(a). We can then write

f(a + h) = f(a) + Ta(h) + r(h), where , as h . .............................(5.5)

We now give a few examples.

Example 5.1 : Consider f: Rn Rn, f(x) = a + x, where a is a fixed vector in Rn . Find the
total derivative of f at a point p Rn, if it exists.
Solution : Now, f(p + h) – f(p) = h. So, if we take T to be the identity transformation from
Rn to Rn, then we get

f(p + h) – f(p) – T(h) = 0, and hence

lim 0.
h 0

Comparing this with 5.5, we conclude that the identity transformation is the total derivative
of f at the point p.

Example 5.2 : Find the total derivative, if it exists, for f : R2 R2, f(x, y) = (x2, y2), at a point

a = (a1, a2).

Solution : If f is differentiable, we expect Ta to be a 22 matrix. Let h = (h1, h2). Now,

f(a + h) – f(a) = (

=( )

= (2 )+( )
 = +( )

We take Ta = , and r(h) = ( ), and write

f(a+h) = f(a) + Ta(h) + r(h), where 0, as h .

Thus Ta is the total derivative of f at a.

Now that we have defined the total derivative, let us see how many of the results that we
know about derivatives of functions of a single variable, hold for these total derivatives.

Theorem 5.1: If f : Rn Rm is differentiable at a Rn, then its total derivative is unique.

Proof : Suppose f has two derivatives, T1 and T2 at a, and let T = T1 – T2. Let h Rn,

h 0, and t R, such that t 0.

Then th 0 as t 0.
Since T1 is a total derivative of f at a,

lim lim 0 ......................(5.6)

t 0 t 0

Since T2 is also a total derivative of f at a,

lim lim 0 ......................(5.7)

t 0 t 0

Thus,

Therefore, +

Since T is a linear transformation, T(th) = tT(h). Therefore,

+ .

So, using (5.6) and (5.7) , we get

0 lim lim lim 0

t 0 t 0 t 0

Since is independent of t, this means = 0, which means that = 0.

Now, h was any non-zero vector in Rn. Further, T(0) = 0. Hence we conclude that T(h) = 0
for all h Rn. Thus T = T1 – T2 is the zero linear transformation. Thus, T1 = T2. That is, the
derivative is unique.

In the next example we find the derivatives of some standard functions.

Example 5.3 : i) Find the total derivative f1(a), if f : Rn Rm , f(x) = c, where c is a fixed
vector in Rm and a Rn.

ii) If f : Rn Rm is a linear transformation, show that Df(a) = f for every a Rn .

Solution : i) Since f is a constant function, we expect its derivative to be the zero

transformation.

Here f(a + h) – f(a) = c – c = 0.

If we take T to be the zero transformation,

lim lim 0.
h 0 h 0

Hence f1(a) exists and is equal to 0 for every a Rn.

ii) Since f is a linear transformation, f(a + h) = f(a) + f(h). If we take T = f,

r(h) = f(a + h)  f(a)  f(h) = 0

We have defined the total derivative of a function as a linear transformation. Now we prove a
result about linear transformations which we may use later.

Proposition 5.1 : Every linear transformation T from Rn to Rm is continuous on Rn.

Proof : If T is the zero linear transformation, it is clearly continuous. If T 0, let p Rn,

p = (p1, p2, ..., pn), and . Suppose {e1, e2, ..., en} is the standard basis for Rn. Choose
M, where M = ....... + .

If x = (x1, x2, ..., xn) is such that < , then |xi – pi| < for i = 1, 2, ..., n.

Also, < =

| | +| | + ....... +| |

< ( +.......+ )

Thus, T is continuous at p. Since p was an arbitrary point of Rn, we conclude that T is

continuous on Rn.
In fact, since did not depend on p, we can conclude that T is uniformly continuous on Rn.

For functions of a single variable, we know that differentiability implies continuity. The next
theorem shows that this holds for functions of several variables too.

Theorem 5.2 : If f : Rn Rm is differentiable at p, then f is continuous at p.

Proof : Since f is differentiable at p, there exists a linear transformation Tp such that

lim 0.
h 0

Thus, such that

<

Choose Then

<( )

By Proposition 5.1, Tp is continuous at 0, and Tp(0) = 0. So, there exists such that

Now choose Then

< .

Thus, , and f is continuous at p.

With your knowledge of functions of one variable, you would expect that the converse of
Theorem 5.2 does not hold. That is, continuity does not imply differentiability. The following
example shows that it is indeed so.

Example 5.4 : Consider the function f : R R2, f(x) = (|x|, |x|). We shall show that f is
continuous at 0, but is not differentiable there.

Given choose . Then

|x| < .

Hence, f is continuous at x = 0.

Now suppose f is differentiable at x = 0. Then there exists a linear transformation

T:R R2, such that

lim =0 lim =0
h 0 h 0

Now, (1, 1) and ( −1, −1) are two distinct points in R2, and B((1, 1), 1) B((−1, −1), 1) = .

For = 1, > 0, such that

. .............................(5.8)

Putting h = in (5.8), we get . This means

T(1) B((1, 1), 1).

Similarly, taking h = − , we get that T(1) B((−1, −1), 1). But this contradicts the fact
that B((1, 1), 1) and B(( −1, − 1), 1) are disjoint.

Thus, f is not differentiable at x = 0.

If f : Rn Rm , then, as you know, we can write f = (f1,f2, ...,fm), where each fi : Rn R,

i = 1, 2, ..., m. These fis are called coordinate functions of f. Similarly, a linear transformation

T : Rn Rm can be written as T = (T1,T2, ...,Tm), where each Ti is a linear transformation

from Rn to R.

Theorem 5.3 : Let f = (f1,f2, ...,fm) : Rn Rm, and p Rn. f is differentiable at p, if and only
if each fi, 1 m is differentiable at p.

Proof : f is differentiable at p if and if there exists a linear transformation Tp : Rn Rm, such

that lim = 0, that is, if only if

h 0

lim = 0, where {e1, e2, ..., em} is the standard basis of Rm,
h 0

if and only if, lim = 0, i, 1 m.

h 0

That is, if and only if each fi is differentiable and Dfi = Ti, ,1 m.

Thus, Df(p) = Tp = (Df1(p), Df2(p), ....., Dfm(p)).

Theorem 5.4 : Let f : Rn Rm and g : Rn Rm be two functions differentiable at p Rn. If

k R, then f + g and kf are also differentiable at p. Moreover,

D(f + g)(p) = Df(p) + Dg(p), and D(kf)(p) = kDf(p).

Proof : Let Df(p) = T1, and Dg(p) = T2. Then T1 + T2 is also a linear transformation from Rn
to Rm, and

0 lim
h 0

= lim
h 0

lim + lim = 0.
h 0 h 0

Therefore, f + g is differentiable at p, and D(f + g)(p) = T1 + T2 = Df(p) + Dg(p).

Now, lim |k| lim = 0.

h 0 h 0

Therefore, kf is also differentiable and D(kf)(p) = kT1 = kDf(p).

5.3 PARTIAL DERIVATIVES

We know that the derivative of a function of one variable denotes the rate at which the
function value changes with change in the domain variable. In the case of functions of several
variables, change in the domain vector variable means a change in any or all of its
components. But if we consider change in only one component and study the rate at which
the function value changes, we get what is known as the partial derivative of the function.
Corresponding to each component of the variable, there will be a partial derivative. Here is
the formal definition.

Definition 5.2 Let f : E Rm, where E Rn. Let x = (x1, x2, ..., xn) be an interior point of E.
Then for every i, i = 1, 2, ..., n, the limit

lim , if it exists, is called the ith partial derivative of f

h 0

with respect to xi at x. It is denoted by . We write to indicate the point

at which the partial derivative is calculated.

Remark 5.2 : i) If a function f has partial derivatives at every point of the set E, we say that f
has partial derivatives on E.

ii) It is clear from the definition that a partial derivative can be defined at an interior point of
E, and not on its boundary.

iii) If a function has a partial derivative at a point, its value depends on the values of the
function in a neighbourhood of that point. So, if the function values outside this
neighbourhood are changed, it does not affect the value of the partial derivative.

The following examples will make the concept clear.

Example 5.5 : Find the partial derivative of the function, f(x, y, z) = xyz + x2z.
Solution : This is a real-valued function. You are already familiar with the partial
differentiation of such a function.

= lim = yz + 2xz. Similarly, you can check that fy = xz, and

h 0
2
fz = xy + x .

Let us take a vector-valued function in the next example.

Example 5.6 : Find the partial derivatives of the function, f : R3 R2, f(x, y, z) = (xy, z2), if
they exist.

Solution : lim = lim

h 0 h 0

= ( lim lim = (y, 0).

h 0 h 0

Therefore, = (y, 0).

Proceeding similarly, we find that = (x, 0), and = (0, 2z).

You must have observed that the partial derivatives of a vector function are formed by taking
the partial derivatives of its coordinate functions. In fact we have the following theorem,
which establishes the connection between differentiability of a vector-valued function and the
existence of partial derivatives of its coordinate functions

Theorem 5.5 : Let E be an open subset of Rn, and f : E Rm. Suppose f = (f1,f2, ...,fm) is
differentiable at p E. Then the partial derivatives exist for i = 1, 2, ..., m, j = 1, 2, ..., n.

Proof : Since f is differentiable at p, there exists a linear transformation T, such that

lim . Let h = tej, where {e1, e2, ...,en} is the standard basis of Rn.
h 0

Then, h 0 if and only if t 0. Thus,

lim . Therefore, lim T( ).

t 0 t 0

That is,

( lim lim lim )

t 0 t 0 t 0

= T( ).

Hence the limits exist, and (p) exists for all i = 1, 2, ..., m.

Since j was arbitrary, we conclude that (p) exists for all i = 1, 2, ..., m, j = 1, 2, ..., n.
If f : E Rm, where E is an open subset of Rn, and if f is differentiable at p E , then using
Theorem 5.5, the matrix of the linear transformation T can be written as

This m n matrix is called the Jacobian matrix of f at p, and is denoted by [f’(p)] or [Df(p)].

If m = n, the determinant of the Jacobian matrix is called the Jacobian of f at p, and is denoted
by .

Thus, if f is differentiable at p, then the total derivative of f at p, T : Rn Rm is given by the

Jacobian matrix. For x = (x1, x2, ..., xn) Rn,

T(x) = [f’(p)] .

When m = 1, f is a real-valued function, and T(ej) = . Hence, the Jacobian matrix of T

is the row matrix, [ ].

The vector form, ( ) is called the gradient of f at p , and is denoted

by f(p), or gradf(p).

If h = (h1, h2, ..., hn) Rn,

Tp(h) = [ ] .

Thus, T(h) = , or Tp(h) = f(p)  h.

So, we can say that the total derivative Tp of a real-valued function is given by

Tp (h) = f(p) h.

Example 5.7 : Find the Jacobian matrix of i) f(x, y) = (x2y, exy)

ii) f(x, y, z) = (xsinz, yez) at (1, 2, 1).

Solution : i) f1(x, y) = x2y, and f2(x, y) = exy. Therefore, = 2xy, = x2,

= yexy, and = xexy.

Hence, [fi(x, y)] =

= sinyz, and (1, 2, 1) =  sin2

(1, 2,  1) =  cos2, (1, 2,  1) = 2 cos2,

(1, 2, 1) = 0, (1, 2,  1) =  e-1, (1, 2,  1) =  2e-1.

Thus, [fi(1, 2,  1)] =

In the next section we shall consider yet another type of derivative.

5.4 DIRECTIONAL DERIVATIVES

Partial derivatives measure the rate of change of a function in the directions of the standard
basis vectors. Directional derivatives measure the rate of change in any given direction.

Definition 5.3 : Let f : E R, where E is an open subset of Rn. Let u be a unit vector in Rn,
and p E. If lim exists, then it is called the directional derivative of f at p in
t 0

the direction u. It is denoted by or fu(p).

Example 5.8 : Find the directional derivatives of the following functions:

i) f(x, y) = 2xy + 3y2 at p = (1, 1), in the direction of v = (1, 1).
ii) f(x, y) = x2y at p = (3, 4), in the direction of v = (1, 1).

Solution : i) The unit vector u in the given direction is ( ). Hence the required

directional derivative is lim .

t 0

= lim
t 0

= lim = lim =5 .
t 0 t 0

ii) We have the same unit vector u here. Therefore,

 Duf(p) = lim = lim = .
t 0 t 0

Example 5.9 : Find the directional derivatives, if they exist, in the following cases:

i) f(x, y) = , at (0, 0), u = (u1, u2), ||u|| = 1

ii) f(x, y) = at (0,0), u = (1/ , 1/ ).

Solution: i) if u1 0, u2 0, lim = lim , which

t 0 t 0

does not exist. If either u1 or u2 is zero, we get the standard basis vectors, (1, 0) and (0, 1).

If u = (1, 0), lim = = 1.

t 0

Similarly, if u = (0, 1), lim = 1.

t 0

Thus, the directional derivatives in these two directions exist, and are equal to one. In any
other direction, the derivative does not exist. Note that the directional derivative in the
direction (1, 0) is fx, and that in the direction (0, 1) is fy. Thus, this function has both the
partial derivatives at (0, 0).

ii) == == = 1/ .
Thus, Duf(0, 0) = 1/ .
In fact, if we take u = (cos , sin ), then we can show that f has directional derivative at (0, 0)
in the direction of u, whatever be . That is, the directional derivatives of f at (0, 0) exist in
all directions. But you can easily show that this function is not continuous at (0, 0) by using
the two-path test. Recall, that you need to show that the limits of f, at (0, 0) along two
different paths are different. Then by Theorem 5.2 we can conclude that f is not
differentiable at (0, 0).
This example shows that the existence of all directional derivatives at a point does not
guarantee differentiability there. But we have the following theorem:

Theorem 5.7: Let f : E R, where E is an open subset of Rn. If f is differentiable at p Rn,

then the directional derivatives of f at p exist in all directions.
Proof : Since f is differentiable at p, there exists a linear transformation, T: Rn R, such that
lim .
h 0

Let u be any unit vector in Rn, and take h = tu. Then h 0, as t 0. Therefore,
lim . This means,
t 0
 lim . That is,
t 0

lim T(u), or, Duf(p) = T(u). ......................(5.5)

t 0

Since u was an arbitrary unit vector, we conclude that the directional derivatives of f at p
exist in all directions.
Now, if u = (u1, u2, ..., un), T(u) = T( u1e1 + u2e2 + ... + unen), where {e1, e2, ..., en} is the
standard basis of Rn. Therefore, by (5.5),
T(u) = u1T(e1) + u2T(e2) + ... + unT(en)

= u1 f(p) + u2 f(p) + ... + un f(p)

= u1 + u2 + ... + un

= f(p)  u

Thus, Duf(p) = f(p)  u ........................ (5.6)

(5.6 ) gives an easy way to find a directional derivative of a differentiable function, if its
partial derivatives are known. For example, if f(x, y) = x2 + y2, then fx and fy at (1, 2) are 2
and 4, respectively. So, the directional derivative of f at (1, 2) in the direction 2i – 3j is given
by (2i + 4j) = .

This concept of directional derivatives can be extended to vector-valued functions. The

directional derivative of a vector-valued function is a vector formed by the directional
derivatives of its coordinate functions. Thus, to find the directional derivative of
f(x, y) = (x + y, x2), at (1, 2) in the direction of (3, 4) , we first find the directional derivatives
of f1(x, y) = x + y, and f2(x, y) = x2 . You can check that these are 7/5 and 6/5, respectively.
Therefore, the required directional derivative of f is (7/5, 6/5).
We have seen in Theorems 5.6 and 5.7, that differentiability of f at a point guarantees the
existence of partial and directional derivatives there. We have also noted that the converse
statements are not true. Our next theorem gives us a sufficient condition which guarantees the
differentiability of a function at a point.

Theorem 5.8 : Let E be an open subset of Rn, and f : E Rm, f = (f1,f2, ...,fm). If all the
partial derivatives, Djfi(x) of all the coordinate functions of f exist in an open set containing
a, and if each function Djfi is continuous at a, then f is differentiable at a.
Proof : In the light of Theorem 5.3, it is enough to prove this theorem for the case m = 1. So,
we consider a scalar function f from Rn to R, all whose partial derivatives Djf are continuous
at a. Since E is open, for a given > 0, we can find r > 0, such that the open ball,

B(a, r) , and || x – a || < r | Djf(x)  Djf(a) | < /n, for j = 1, 2, ... , n. ..................(5.7)
Now, suppose h = (h1, h2, ... , hn), ||h|| < r. Let v0 = 0, v1 = h1e1, v2 = v1 + h2e2, ... ,

vn = vn – 1 + hnen. Then f(a + h) – f(a) = . ...............(5.8)

Since ||vj|| < r, vj B(a, r), and since B(a, r) is convex, the line segment joining the points,
a+ vj – 1 and a + vj lies in it, for all j = 1, 2, ... , n. Therefore, we can apply the Mean Value
Theorem to the jth term in the sum (5.8), and get

f(a + vj) – f(a + vj − 1) = hjDjf(a + vj – 1 + hjej) , for some (0, 1). Then, using (5.7), we
can write

|f(a + h) – f(a)  (a)| = | (a + vj − 1 + hjej )- (a)|

, for all h, such that ||h|| < r.

This means that

lim 0, where is the linear transformation, whose matrix

h 0

[ ] consists of the row, (D1f(a), D2f(a), ...., Dnf(a)).

Thus, f is differentiable at a.

Definition 5.4 : A function f : E Rm, f = (f1,f2, ...,fm), where E is an open subset of Rn,
is said to be continuously differentiable, or, a C1 function, if Djfi is continuous on E for
all j, j = 1, 2, ..., n, and for all i, i = 1, 2, ..., m.

The continuity of partial derivatives assumed in Theorem 5.8, is only a sufficient

condition, and not a necessary one. That is, there may be functions which are
differentiable at a point, but do not have continuous partial derivatives there. We now
give you an example, and ask you to work out the details (See Exercise 3.)

Example 5.10 : Consider the function f : R2→ R given by

f(x, y) =

This function is differentiable at (0, 0), but neither ,

nor is continuous at (0, 0).

Here are some exercises that you should try.

Exercises:

1) Show that the following function is differentiable at all x in Rn.

f : Rn , f(x) = x  T(x), where T : Rn Rn is a linear transformation.

 2) Let f(x, y) = (x3 + x, x2 – y2, 2x + 3y3), p = (2, 1), v = (4, 5). Compute the partial
derivatives of f, and the directional derivative of f in the direction v, at p.
3) Prove the assertions in Example 5.10. (Hint : To show that f is differentiable, check
that f(h, k) - f(0, 0) – h(hsin ) + k(ksin ) = 0, and so, Df = (hsin , ksin ) ).

5.5 SUMMARY

In this unit we have extended the concept of differentiation from functions of one variable
to functions of several variables. Apart from the total derivatives, we have also defined
partial derivatives, and directional derivatives. We have proved that differentiability
implies the existence of all partial and directional derivatives at a point, but the converse
is not true. As in the case of functions of one variable, we prove that differentiable
functions are continuous, but not vice versa. We have also derived a sufficient condition
for differentiability in terms of the partial derivatives.
 6

DERIVATIVES OF HIGHER ORDER

Unit Structure

6.0 Objectives

6.1 Introduction

6.2 Jacobian Matrix and Chain Rule

6.3 Higher order partial derivatives

6.4 Mean Value Theorem

6.5 Summary

6.0 OBJECTIVES
After reading this chapter, you should be able to

 differentiate a composite of two vector-valued functions

 define and calculate derivatives of higher order
 derive the conditions for the equality of mixed partial derivatives
 state and prove the Mean Value Theorem

6.1 INTRODUCTION
In the last chapter you have seen how functions of several variables are differentiated. Now
we shall start by discussing how a composite function of two differentiable functions can be
differentiated. The Jacobian matrix introduced in the last chapter proves useful in this.

One of the important applications of derivatives is the location of extreme points of a

function. In the next chapter we are going to see how this concept can be extended to scalar
functions of several variables. But we shall do the necessary spade-work in this chapter. So,
we shall introduce higher order derivatives. We shall also study the conditions under which
mixed partial derivatives are equal. You may recall that the Mean Value Theorem was one of
the most important theorems that you studied in Calculus in F. Y. B. Sc. We shall see
whether this theorem can be applied to functions of several variables.

6.2 JACOBIAN MATRIX AND CHAIN RULE

We have seen in Theorem 5.5, that if f: Rn Rm, is differentiable at p, then all partial
derivatives of all coordinate functions of f exist at p. That is, if f = (f1, f2, ... , fm), then Djfi(p)
exists for all i = 1, 2, ..., m and all j = 1, 2, ..., n. We have also seen that if {e1, e2, ..., en} is the
standard basis for Rn, then

(p)(ej) = (Djf1(p), Djf2(p), ..., Djfm(p)).

If h = is a vector in Rn, then

(p)(h) = (p), which is a linear transformation from Rn to Rm, thus has

the matrix,
,

As we have already mentioned in Chapter 5, this m  n matrix, called the Jacobian matrix, is
denoted by [Df(p)]. The kth row of this matrix is the gradient vector, fk(p), and the jth
column is the image of ej under the linear transformation Djf(p).

Thus, the Jacobian matrix of f is formed by all first order partial derivatives of f. This means,
we can write the Jacobian matrix of any function, all of whose partial derivatives exist. As we
have noted earlier, the existence of partial derivatives does not guarantee differentiability. So,
even when a function is not differentiable we would be able to write its Jacobian matrix,
provided all its partial derivatives exist.

If f : Rn R, then its Jacobian matrix, if it exists, will be a 1 x n matrix, or a matrix vector.

If f : Rn Rm is differentiable at p Rn, and if h is any vector in Rn, then

(p)(h) = [Df(p)]h is obtained by multiplying the m  n matrix [Df(p)] with the n  1

column matrix h. Thus,

|| (p)(h)|| = ||  ||  || =  |,
since

|| ej || = 1, 1 n.

Cauchy-Schwartz inequality for inner products says that | u  v | || u || || v||. Using this we
get || (p)(h)|| = || h || .

If we take M = , then

|| (p)(h)|| M || h ||. .........................(6.1)

We have seen in Theorem 5.4 how to get the derivative of the sum of two differentiable
functions, and also that of a scalar multiple of a differentiable function. The next theorem,
which is known as the chain rule, tells us how to get the total derivative of a composite of two
functions.

Theorem 6.1 (Chain Rule) : Let f and g be two differentiable functions, such that the
composite function f  g is defined in a neighbourhood of a point a Rn. Suppose g is
differentiable at a, g(a) = p, and f is differentiable at p. Then f  g is differentiable at a, and

(a) = (p) (a) = [Df(p)] [Dg(a)].

Proof : If h is such that || h || is small, then a + h will belong to the above neighbourhood of
a, in which f  g is defined. Now, since g is differentiable at a,

k = g(a + h) – g(a) = (a)(h) + || h || Ea(h), ............(6.2)

where Ea(h) 0, as h 0.

f is differentiable at p = g(a), and therefore,

f(g(a + h)) – f(g(a)) = f(p + k) – f(p) = (p)(k) + || k || Ep(k), where Ep(k) 0, as k 0.

= (g(a))[ g(a + h) – g(a)] + || k || Ep(k)

= (g(a))[ (a)(h) + || h || Ea(h)] + || k || Ep(k), using (6.2).

= (g(a)) (a)(h) + (g(a)) [|| h || Ea(h)] + || k || Ep(k), since

(g(a)) is a linear transformation. Thus, we can write

f(g(a + h)) – f(g(a)) = (g(a)) (a)(h) + || h ||[ (g(a)) Ea(h) + Ep(k)], if h 0. ..(6.3)

To complete the proof we need to show that the vector in the square brackets in (6.3) tends to
zero, as h tends to zero.

We know that Ea(h) 0, as h 0. ..............(*)

|| k || = || g(a + h) – g(a) || || (a)(h) || + || h || || Ea (h) ||, using (6.2).

If M = , then using (6.1), we can write || (a)(h) || M || h ||. Thus,

|| k || M || h || + || h || || Ea(h) || = || h || (M + || Ea(h) ||). Therefore,

M + || Ea(h) ||. This means that is bounded. Thus,

0, as h 0, since h 0 . ....(**)

Using (*) and (**), we can say that the term in the square brackets in (6.3) tends to zero as
h 0. Therefore,

–
0 as h 0.

This shows that f  g is differentiable at a, and  (a) = (g(a)) (a).

The Chain Rule can be written in terms of Jacobian matrices as follows:

D(f  g) (a) = [D(f(g(a)))] [D(g(a))].

Here the product on the right hand side is matrix multiplication. If y = g(x), and z = f(y),
comparing the entries in the matrices in (6.3), we get

= , where = Dk(f  g)i , = Dj(f )i , and = Dk(g)j .

Example 6.1 : Write the matrices for , and  for the following functions, and
evaluate them at the point (2, 5). f(x, y) = (x + y, x + y , 2x + 3y), g(u, v) = (x, y) = (u2, v3).
2 2

Solution : Here f1(x, y) = x + y, f 2(x, y) = x2 + y2, f3(x, y) = 2x + 3y,

g1(u, v) = u2 and g2(u, v) = v3. This means, D(f) = , and D(g) = .

 ) (u,v) = (u2 + v3, u4 + v6, 2u2 + 3v3). Hence,

D  )= .

At (u, v) = (2, 5), (x, y) = (4, 125). Therefore,

, D(f)(4, 125) = , D(g)(2, 5) = , and D  )(2, 5) = .

You can now easily verify that D(f  g) (2, 5) = [D(f(4, 125)] [D(g(2, 5))].

6.3 HIGHER ORDER PARTIAL DERIVATIVES

You are familiar with the concept of partial derivatives. In the last chapter we have calculated
the partial derivatives of some functions of n variables. If you take a look at those examples,
you will realise that the partial derivatives are themselves functions of n variables. So, we can
talk about their partial derivatives. These, if they exist, will be the second order partial
derivatives of the original function. If we differentiate these again, we will get the third order
partial derivatives of the original function, and so on. We take a simple example to illustrate.

Example 6.2 : Find partial derivatives of all possible orders for the function,
f(x, y, z) = (x2y2, 3xy3z, xz3).

Solution : Since f is a polynomial function, we do not have to worry about the existence of
partial derivatives. We get

fx = (2xy2, 3y3z, z3), fy = (2x2y, 9xy2z, 0), fz = (0, 3xy3, 3xz2).

Then, fxx = (2y2, 0, 0), fxy = = = (4xy, 9y2z, 0), fxz = (0, 3y3, 3z2).

Differentiating fy, we get fyx = (4xy, 9y2, 0), fyy = (2x2, 18xyz, 0), and fyz = (0, 9xy2, 0).

Then differentiating fz we get fzx = (0, 3y3, 3z2), fzy = (0, 9xy2, 0), and fzz = (0, 0, 6xz).

These are all possible second order derivatives of f. Proceeding in this way, we can also get

fxyz = (0, 9y2, 0), fyxz = (0, 0, 0), fzzz = (0, 0, 6x), and so on. There will be 27 third order
partial derivatives of f. See if you can get the remaining.

You know that fxy and fyx differ in the order in which f is differentiated with respect to the
variables x and y. These two derivatives have come out to be equal in Example 6.2. But you
may have seen examples of scalar functions of several variables, for which the two may not
be the same. Here is an example, to jog your memory.

Example 6.3 : Consider this function f from R2 to R, f(x, y) = for (x, y) (0, 0),
and f(0, 0) = 0. You can easily check that

fx(0, 0) = 0, fy(0, 0) = 0, fx(0, k) = lim =  k,

h 0

fy(h, 0) = lim = h.
k 0

Then, fxy(0, 0) = lim = lim =  1, and similarly, fyx(0, 0) = 1.

k 0 k 0

Thus, the mixed partial derivatives of this function both exist, but are not equal.

Remark 6.1 : If f is a function from Rn to R, the partial derivative of f with respect to the ith
variable, xi, is denoted by Dif, and the partial derivative of Dif with respect to xj , that is,
Dj(Dif) is denoted by Djif.

The following theorem gives a sufficient condition for the two mixed partial derivatives of a
function to be equal. Since the behaviour of a vector-valued function is decided by the
behaviour of its coordinate functions, it is enough to derive this sufficient condition for a
scalar function. Without loss of generality, we state the theorem for a function of two
variables.
Theorem 6.2 : Let f : R2 R, such that the partial derivatives, D1f, D2f, D12f and D21f exist
on an open set S in R2. If (a, b) S, and D12f and D21f are both continuous at (a, b), then
D12f(a, b) = D21f(a, b).

Proof : We choose positive real numbers, h and k, which are small enough so that the
rectangle with vertices (a, b), (a + h, b), (a, b + k), (a + h, b + k) lies within S.

Now we consider a function

(h, k) = [f(a + h, b + k) – f(a + h, b)] – [f(a, b + k) – f(a, b)].

We also define a function G on [a, a + h], G(x) = f(x, b + k) – f(x, b).

Now we can write (h, k) = G(a + h) – G(a). Since G is defined in terms of f, and since f has
all the necessary properties, G is continuous on [a, a + h], and is differentiable in (a, a + h).
So, we apply the Mean Value Theorem for functions of a single variable to G, and get

G(a + h) – G(a) = h (c), for some c (a, a + h). Now (x) = D1f(x, b + k) – D1f(x, b). So,
we write (h, k) = G(a + h) – G(a) = h[D1f(c, b + k) – D1f(c, b)].

Now D1f (c, y) is a differentiable function of one variable with derivative equal to D 21f. So
applying MVT to D1f(c, y) on the interval [b, b + k], we get

(h, k) = h[D1f(c, b + k) – D1f(c, b)] = hkD21f(c, d), ..........................(6.4)

for some d (b, b + k).

We now write (h, k) = [f(a + h, b + k) – f(a, b + k)] – [f(a + h, b) – f(a, b)], and define

H(y) = f(a + h, y) – f(a, y), so that (h, k) = H(b + k) – H(b). Using the same arguments
that we used for G, we apply MVT to H, and then to D2f(x, p), we get

(h, k) = k[D2f(a + h, p) – D2f(a, p)] = khD12f(q, p), ............................(6.5)

for some p (b, b + k), and q (a, a + h).

From (6.4) and (6.5) we get D21f(c, d) = D12f(q, p). Since D12f and D21f are continuous, taking
the limit as (h, k) (0,0), we get D12f(a, b) = D21f(a, b).

As we have mentioned earlier, the conditions of this theorem are sufficient, and not
necessary. In fact, the continuity of just one of the mixed partial derivatives is also sufficient
to guarantee equality. Functions whose partial derivatives are continuous play an important
role in Calculus. We classify these functions as follows:

Definition 6.1 : A function f from Rn to Rm is said to be continuously differentiable, or

belong to class C1, if all its partial derivatives Dif are continuous. It is said to belong to class
C’’, if all its second order partial derivatives are continuous, and so on. If all its partial
derivatives of all orders are continuous, then it is said to belong to class .
We have proved that a function in class C1 is differentiable in Theorem 5.8. In Theorem 6.2
we have seen that the mixed partial derivatives of a function belonging to class C’’ are equal.

In the next chapter we shall see that a Ck function, that is a function, all whose partial
derivatives of order up to k are continuous, can be approximated by means of a polynomial of
order k. We shall also discuss the technique to find the maximum and minimum values of a
function belonging to class C’’.

6.4 MEAN VALUE THEOREM

The Mean Value Theorem (MVT) is an important theorem in Calculus. It is used as a tool to
derive many other results. In the last section we have used it in the proof of Theorem 6.2. In
this section we shall see if it also holds good for functions of several variables. But first, let
us recall the one-variable case.

MVT (single variable): If f : [a, b] R is continuous on [a, b], and differentiable on (a, b),
then there exists c (a, b), such that

f(b) – f(a) = (b - a) (c).

If we write b = a + h, then there exists , such that

f(a + h) – f(a) = h .

Unfortunately, it is not possible to extend this theorem to a function f : Rn Rm, when

m > 1. This will be quite clear from the following example.

Example 6.4 : Consider f : [0, 2 ] R2, f(t) = (cost, sint). This function is continuous on

[0, 2 ] and differentiable on (0, 2 ). Now, f(2 ) – f(0) = (1, 0) – (1, 0) = (0, 0).

(t) = ( − sint, cost). For the extension of MVT to hold, we must have

f(2 ) – f(0) = 2 (c) for some c in (0, 2 ). So, we should have (0, 0) = 2 (  sinc, cosc).
But this is impossible, since sinc and cosc both cannot be zero.

So, the extension of MVT in its stated form does not hold. But there is a way around this
difficulty. A slightly modified version of MVT does hold true for all functions of several
variables. We now state and prove this modified theorem for functions from Rn to Rm. As a
special case of this theorem you will realize that MVT holds for real-valued functions of
several variables.

Theorem 6.3 : (Mean Value Theorem) Let f : S Rm, where S is an open subset of Rn.
Suppose f is differentiable on S. Let x and y be two points in S, such that the line segment
joining x and y, L(x, y) = {tx + (1  t)y | 0 1}, also lies in S. Then for every a Rm,
there is a point z S, such that
a  {f(y) – f(x)} = a  { (z)(y  x)} ................................(6.6)

Before we start the proof, let us understand the geometry involved. Let u = y – x. Then x + tu
gives us a point on the line segment L(x, y), if 0 1. Since S is open, we can find a

> 0, such that S, and S. See Fig. 6.1, in which we show the situation

when n = 2. The point p is on the extension of L(x, y) and is equal to x + (1 + )u. Similarly
the point q is also on the extension of L(x, y), and is equal to x – u for some > 0.

p p
Y

S
X

Figure 6.1

Thus we get a > 0, such that x + tu S for every t . Now we start the formal
proof.

Proof : Let a Rn. We define a function F : R, F(t) = a  f(x + tu). This F is

a differentiable function on , and

(t) = a  , using chain rule.

(Recall, that is a linear transformation.)

Thus, we can apply MVT for functions of a single variable, and get

F(1) – F(0) = ( ), for some . ............................(6.7)

Now, F(1) = a  f(x + u) = a  f(y), F(0) = a  f(x), and

( )=a  =a , where z = L(x, y).

Therefore, from (6.7) we get a  {f(y) – f(x)} = a  for some z S.

Remark 6.2 : i) (6.6) is true for all x, y in S, such that the line segment joining x and y is also
in S. This means, if S is a convex open set in Rn, then (6.6) will be true for all x, y in S.

ii) If f is a real-valued function, then m = 1, and a R. Then for a = 1 we have

1 . {f(y) – f(x)} = 1 . = f(z)  (y- x), for some z S.

So, the MVT for functions of a single variable extends directly to real-valued functions of
several variables. We can also directly prove MVT for scalar functions. The proof runs
exactly similar to that of Theorem 6.3, if we put a = 1.

The MVT has a well-known consequence, which we now state:

Theorem 6.4 : Let f : S Rm, where S is an open connected subset of Rn. Suppose f is
differentiable on S, and = 0 for every p S. Then f is a constant function on S.

Proof : The set S is polygonally connected, since it is open and connected. Let x and y be
two points in S. Then x and y are joined by line segments L1, L2, L3, ... , Lr, lying entirely in
S. Suppose Li is a line segment joining pi and pi+1, 1 r, p1 = x, and pr+1 = y.

Let a Rm. Then using Theorem 6.3, we have

a  {f(pi+1) – f(pi)} = a  , zi Li

= 0, since = 0.

This means,

a  {f(y) – f(x)}= a  {f(pr+1) – f(p1)} = a  {f(pi+1) – f(pi)} = 0. ..................(6.8)

(6.8) is true for every a in Rm. So, in particular, it is true for f(y) – f(x). Thus,

{f(y) – f(x)}  {f(y) – f(x)} = ||f(y) – f(x)||2 = 0.

So, f(y) – f(x) = 0, or f(y) = f(x).

Since x and y were any arbitrary points in S, we have thus proved that f is a constant function
on S.

Try a few exercises now.

Exercises :

1) Find the partial derivatives, D1f, D2f, D12f and D21f at (0, 0) , if they exist, for the
following function f from R2 to R.
f(x, y) = y , if (x, y) (0, 0), and f(0, 0) = 0.
2) If u(x, y) = x +y2, x(t) = 3t2 + 4, and y(t) = sin2t, find (t) and (t).
3) If u(x, y) = x – 2y + 3, x = r + s + t, y = rs + t2, find ur, us and ut at (1, 2, 4).
4) Let f : R2 R2, and g : R3 R2 be two vector functions, defined as:
f(x, y) = (sin(2x + y), cos(x + 2y)),
g(r, s, t) = (2r – s – 3t, r2 – 3st).
i) Write the Jacobian matrices for f and g. If h is the composite function, f  g,
compute the Jacobian matrix of h at the point (1, 0, - 2).
5) If f is a function from R2 to R, and D1f = 0 at all points, show that f is independent of
the first variable. If D1f = D2f = 0 at all points, show that f is a constant function.
 6.5 SUMMARY
In this chapter we have derived the chain rule for differentiation of composite of two
functions. We have also seen that the Jacobian matrix for the composite function is the
product of the Jacobian matrices of the two given functions. We have defined higher order
partial derivatives of functions of several variables. We have seen functions, whose second
order mixed partial derivatives depend on the order of the variables with respect to which the
function is differentiated. On the other hand, we have derived sufficient conditions for such
mixed partial derivatives to be equal. Finally, through an example we have seen that the
Mean Value Theorem cannot be extended to all vector functions. We have proved a restricted
form of the MVT for vector functions. Of course, MVT does extend to scalar-valued
functions of several variables. As a result of MVT we have proved that a function defined on
an open connected set is constant, if its derivative is uniformly zero over its domain.
 7
APPLICATIONS OF DERIVATIVES
Unit Structure

7.0 Objectives

7.1 Introduction

7.2 Taylor’s Theorem

7.3 Maxima and Minima

7.4 Lagrange’s Multipliers

7.5 Summary

7.0 OBJECTIVES
After reading this chapter, you should be able to

 state Taylor’s theorem for real-valued functions of several variables

 obtain Taylor’s expansions for some simple functions
 define, locate and classify extreme points of a function of several variables
 obtain the extreme values of a function of n variables, subject to some constraints

7.1 INTRODUCTION
In the two previous chapters we have discussed differentiation of scalar and vector functions
of several variables. Now we shall tell you about some applications of derivatives. In your
study of functions of one variable you have seen that a major application of the concept of
derivatives is the location of maxima and minima of a function. This knowledge is very
crucial for curve tracing. Here we shall see how the derivatives help us in locating the
extreme values of a real-valued function of several variables. But before we do that, we are
going to discuss Taylor’s theorem and Taylor’s expansions, which help us approximate a
function with the help of polynomials. This knowledge will help us derive some tests for
locating and classifying the extreme points of a function.

7.2 TAYLOR’S THEOREM

It will be useful to recall Taylor’s theorem for functions of one variable, which you have
studied in F. Y. B. Sc. Here we shall also give you the proof of this theorem. Our method of
proof involves the use of Rolle’s theorem. You have studied this theorem too in F. Y. We
now state Rolle’s theorem, and then move on to Taylor’s theorem.

Theorem 7.1 (Rolle’s Theorem): If f: [a, b] R is continuous on [a, b], differentiable on

(a, b), and f(a) = f(b), then there exists c (a, b), such that (c) = 0.

Theorem 7.2 (Taylor’s theorem for real functions of one variable): Let f be a real-valued
function defined on the open interval (p, q). Suppose f has derivatives of all orders up to and
including n +1 in (p, q). Let a be any point in (p, q). Then for any x (p, q),

xa ( x  a) 2 ( x  a) n ( x  a) n 1
f(x) = f(a) + (a) + (a) + ... + (a) + (c),...(7.1)
1! 2! n! (n  1)!

where c (a, b).

Proof: We now define a new function g on [a, x], or [x, a], according as a < x, or x < a, by

( x  y) ( x  y) 2 ( x  y) n
g(y) = f(y) + (y) + (y) + ... + (y) + A, ....(7.2)
1! 2! n!

where A is a constant, chosen so as to satisfy g(x) = g(a). We can easily write the expression
for A by using this condition. We leave this to you as an exercise. See Exercise 1).

Using the properties of f, we can see that g satisfies all the conditions of Rolle’s theorem on
its domain. Thus, we can conclude that there exists a point c (a, x), (or (x, a)) such that
(c) = 0. Now, differentiating (7.2), we see that

( x  y) 2 ( x  y ) ( n 1)
(y) = (y) − (y) + (x − y) (y) − (x − y) (y) + (y) − ... −
2! (n  1)!
( x  y) n
(y) + (y) – (n + 1) A.
n!

f ( n1) ( y )
= [ (n + 1)A].
n!

f ( n1) (c)
Hence, (c) = [ − (n + 1)A] = 0.
n!

f ( n1) (c)
This means that A =
n!

Substituting this value of A in (7.2), we get

f(x) = g(x) =

xa ( x  a) 2 ( x  a) n ( x  a) n 1
g(a) = f(a)+ (a) + (a) + ... + (a) + (c),
1! 2! n! (n  1)!

thus proving the theorem.

Remark 7.1 : If the function in Theorem 7.2 has derivatives of all orders in (p, q), then we
can write a Taylor expansion as in (7.1) for any n N. Further, if all the derivatives of all
orders are bounded by a positive number M, that is, if < M for all n, and at all points in

(p, q), then 0 as n for every x in some interval

{x: |x – a| < R}. Therefore, in this case we can write

xa ( x  a) 2 ( x  a) n ( x  a) n 1
f(x) = f(a) + (a) + (a) + ... + (a) + (c),...(7.3)
1! 2! n! (n  1)!

The infinite series in (7.3) is convergent under the given conditions, and is called the Taylor
series of f about a.

Now, (7.1) can be written as f(x) = Pn(x) + Rn(x), where

xa ( x  a) 2 ( x  a) n
Pn(x) = f(a) + (a) + (a) + ... + (a) is called the nth Taylor
1! 2! n!
( x  a) n 1
polynomial of f about a, and Rn(x) = (c), is called the remainder.
(n  1)!

We now state Taylor’s theorem for functions of two variables, and then find Taylor
expansions of some functions.

Theorem 7.3 (Taylor’s theorem for f: R2 R): Let f be a real-valued Cn+1 function on an
open convex set E R2. Let (a, b) E. Then for any (x, y) E,

f(x, y) = f(a, b) + (h )f(a, b) + f(a, b) + ... + f(a, b)

+ f(c, d), ..................................(7.4)

where h = x – a, k = y – b, and (c, d) is some point on the line segment joining (a, b)

and (x, y).

We are not going to prove this theorem. But, note the following points:

1. Recall that f is Cn+1 means f has continuous partial derivatives of all orders n + 1.
This ensures that all the relevant mixed partial derivatives are equal.
2. E is convex. This guarantees that the line segment joining any two points of E, lies in
E, the domain of f.

Pn(x, y) = f(a, b)+ (h )f(a, b) + f(a, b) + ... + f(a, b)

where h = x - a, and k = y – b, is called the nth Taylor polynomial, and

Rn(x, y) = f(c, d) is called the remainder of order n.

Let us use this theorem to get the expansions of some functions.

Example 7.1: Find the Taylor expansions of the following functions about the given points
up to the third order.

i) f(x, y) = x3 + 2xy2 – 3xy + 4x + 5, (a, b) = (1, 2)

ii) f(x, y) = sin(2x + 3y) (a, b) = (0, 0).

Solution: i) Since f(x,y) = x3 + 2xy2 – 3xy + 4x + 5 is a polynomial, it has partial derivatives

of all orders. Further, its partial derivatives of order > 3 are all zero. In fact,

fx = 3x2 + 2y2 – 3y + 4, fy = 4xy – 3x, fxx = 6x, fxy = 4y – 3, fyy = 4x, fxxx = 6, fxxy = 0,

fxyy = 4, fyyy = 0, and all higher partial derivatives are zero. Calculating all these partial
derivatives at (1, 2), we write

f(1 + h, 2 + k) = 12 + 9h + 5k + (6h2 + 10hk + 4k2) + (6h3 + 12hk2) + R3 .

Now, R3 involves all fourth order derivatives, and therefore is zero. Hence,

f(1 + h, 2 + k) = 12 + 9h + 5k + (6h2 + 10hk + 4k2) + (6h3 + 12hk2) .

ii) f(x, y) = sin(2x + 3y) also has derivatives of all orders.

fx = 2cos(2x + 3y) = 2 at (0, 0), fy = 3cos(2x + 3y) = 3 at (0, 0),
fxx =  4sin(2x + 3y), fxy =  6sin(2x + 3y), fyy =  9sin(2x + 3y). These second
order derivatives are all zero at (0, 0).

fxxx =  8cos(2x + 3y), fxxy =  12cos(2x + 3y), fxyy =  18cos(2x + 3y),

fyyy =  27cos(2x + 3y).

These are, respectively,  8,  12,  18, and – 27 at (0, 0). Thus,

f(h, k) = 0 + (2h + 3k) + .0 + ( 8h3 – 3.12h2k – 3.18hk2 – 27h3) + R3, where

R3 = (h )4sin(2c + 3d), where (c, d) is some point on the line segment joining (0, 0)
and (h, k).

We are now going to state Taylor’s theorem for real-valued functions of n variables. For this,
let us first take a close look at the Taylor expansion of a function of two variables.

If we write (x, y) as (a + h, b + k), we get

f(a + h, b + k) = f(a, b) + (h )f(a, b) + f(a, b) + ... +

f(a, b) + f(c, d),

If we take the variables to x1, x2, instead of x and y, take (a, b) to be (a1, a2), and (h, k) to be
 f(a1 + h, a2 + h2) = f(a1, a2) + ( )f(a1, a2) + f(a1, a2) + ...

+ f(a1, a2) + f(c, d),

= f(a1, a2) + Rn(c, d),

= f(a1, a2) ... + Rn(c, d),

where = , and ... , = 1 or 2, and the sum is taken over all

ordered k-tuples ( ... , ). For example,

= D11f(a1, a2)h12 + D12f(a1, a2)h1h2 + D21f(a1, a2)h2h1 + D22f(a1,

a2)h22

= )f(a1, a2) .

Similarly,

= D111f(a1, a2)h13 + D112f(a1, a2)h12 h2 + D121 f(a1, a2)h1 h2 h1 +

D211 f(a1, a2)h2 h12 + D122 f(a1, a2)h1h22 + D212f(a1, a2)h2 h1 h2 + D221f(a1, a2)h22 h1

+ D222f(a1, a2)h23

= )f(a1, a2) .

You must have noticed that we have added the mixed partial derivative terms, for example,
D12f and D21f, or D112f , D121f, and D211f. We could do this, since f ensures that that
these partial derivatives are equal. Now we state Taylor’s theorem for real-valued functions
of several variables.

Theorem 7.4 : Let f : E R, where E is a convex open subset of Rn. Further, let

a = (a1, a2, ..., an) E, h = (h1, h2, ..., hn) Rn, such that a + h D. If f Cm, then

f(a + h) = f(a) ... + Rm-1(c), ......................(7.5)

where ... , take values from the set {1, 2, ..., n}, and the inner summation in (7.5) is
taken over all possible such k-tuples.

Further, the remainder Rm-1(c) = f(c) . This sum is taken over all
possible m-tuples (i1, i2, ..., im), where i1, i2, ..., im take values from {1, 2, ..., n},and c is some
point on the line segment joining a and a + h.
This theorem is used to approximate a given function by a polynomial. In the next section we
shall use it to derive conditions for locating and classifying extreme points of a function.

Exercises: 1) Write the expression for A appearing in Theorem 7.2.

7.3 MAXIMA AND MINIMA

One of the most interesting and well-known applications of Calculus is the location and
classification of extreme points of a function. You have solved many such problems
involving functions of one or two variables. We shall now extend the definitions of maxima
and minima to functions of n variables, and derive suitable tests for their location.

Definition 7.1 : Let f : Rn R. A point a Rn is said to be a local maximum (or relative

maximum) if there exists a neighbourhood N of a, such that f(x) f(a) for every x N.

f(a) is then called the local or relative maximum value.

A local minimum (or relative minimum) is defined in a similar manner. You will agree that

the function f : R5 R, f(x1, x2, x3, x4, x5) = x12 + x22 + x32 + x42 + x52, clearly has a local
minimum at (0, 0, 0, 0, 0). Can you find an example of a function with a local maximum?
Definition 7.2 : A point a Rn is called a saddle point of a function f : Rn R, if every ball
B(a, r), r > 0, contains points x, such that f(x) f(a), and also other points y, such that f(y)
f(a).

In general, it is not easy to spot the local maximum or local minimum merely by observation.
For differentiable functions we can derive tests to locate these values. You know that in the
case of a differentiable function of a single variable, the derivative vanishes at an extreme
point. We have a very similar test for the location of extreme points of a function of n
variables, as you can see in the next theorem.

Theorem 7.5 : If f : Rn R has a local maximum at a Rn, then i = 1, 2, ..., n,

(a), if it exists, is equal to zero.

Proof : Since f has a local maximum at a, r > 0, such that x B(a, r) f(x) f(a).

For i = 1, 2, ..., n, consider a function gi : (ai – r, ai + r) R, such that

gi(x) = f(a1, a2, ..., ai – 1, x, ai+1, ..., an). Since f(a) is the local maximum value of f, gi(ai) is the
maximum value of gi. If (a) exists, then (ai) also exists, and the two are equal. By
applying the first derivative test for functions of one variable to gi, we get

(a) = (ai) = 0.
An exactly similar proof will help us conclude that (a), if it exists, is equal to zero, even
when a is a local minimum of f.

Thus, if f has a local extremum at a, and all the partial derivatives exist at a, then f(a) = 0.

As in the case of functions of one variable, the condition in theorem 7.5 is a necessary one,
and is not sufficient. That is, if all the partial derivatives of a function at a point a are zero,
we cannot say that a is a local maximum or local minimum point. It may be neither.

An example is the function f : R2 R, f(x, y) = 1 – x2 + y2. Here fx = - 2x, and fy = 2y. So,
fx(0, 0) = 0 and fy(0, 0) = 0. But you can see clearly, that f has a maximum in the direction of
the x-axis, and a minimum in the direction of the y-axis at (0, 0). This means, f has neither a
minimum, nor a maximum at (0, 0). In fact (0, 0) is a saddle point for this function.

Definition 7.3 : Let f : Rn R be differentiable, and a Rn. If (a) is equal to zero

for i = 1, 2, ..., n, then a is called a critical point, or a stationary point of f.

Theorem 7.5, tells us to look for extreme points among the critical points of a function. We
shall now see how to classify these points as local maxima, local minima, or saddle points.
This involves second order partial derivatives. This is to be expected, since in one variable
functions too, we have a second derivative test to classify stationary points. The proof of the
test for several variables involves quadratic forms. You have studied them in T. Y. B. A. /B.
Sc. We start with a definition and recall the relevant results.

Definition 7.4 : If A = (aij) is a real symmetric n x n matrix, and x = (x1, x2, ..., xn) Rn,
then Q(x) = is called a quadratic form associated with A.

We can write Q(x) = xAxt. If A is a diagonal matrix, then Q(x) = is called a

diagonal form. Since A is real symmetric, its eigen values are all real. If all the eigen values
of A are positive, then Q(x) 0 for every x, and Q(x) = 0 x = 0. Such a quadratic form is
said to be positive definite. If all the eigen values of A are negative, then Q(x) 0 for every
x, and Q(x) = 0 x = 0. Such a quadratic form is called negative definite.

It may not be very easy to get the eigen values. But we have an easier way to decide.

A principal minor of a square matrix, A, is the determinant of the matrix obtained by taking
the first k rows, and the first k columns of A, 1 n.

If all the principal minors are positive, then the associated quadratic form is positive definite.

If the principal minors are alternately positive and negative, starting with a negative minor for
k = 1, then the associated quadratic form is negative definite.

If a principal minor of order k is negative, when k is an even number, then Q(x) takes both
positive and negative values.
We now use these facts about quadratic forms to derive the second derivative test. A
definition first.

Definition 7.5 : If f is a C2 function from Rn to R, then the symmetric matrix

A = H(x) = is called the Hessian matrix of f at x. Thus,

A = H(x) = .

If a Rn , the first order Taylor formula for f about a gives us the value of f(a + h) for small
values of ||h|| as

f(a + h) = f(a) +  + R1(c).

If a is a critical point, then = 0, and therefore we get

f(a + h) − f(a) = R1(c).

Now, R1(c) = , where 0 < < 1.

= hH(a+ h)ht . We write,

hH(a+ h)ht − hH(a)ht = h[H(a+ h) – H(a)]ht = ||h||2E(a, ) . Thus,

||h||2 |E(a, )| = |

Therefore, |E(a, )| when h 0. ...................(7.6)

Each term in the finite sum on the right hand side tends to zero as h 0, since f C2, and
hence the second order derivatives are continuous. Therefore, E(a, ) 0, as h 0. We write
hH(a+ h)ht = hH(a)ht + ||h||2E(a, ), where E(a, ) 0, as h 0.

Hence, f(a + h) – f(a) = hH(a)ht + ||h||2E(a, ). ..........................(7.7)

Theorem 7.6 : If f is a function from Rn to R, and has continuous second order partial
derivatives in a ball B(a; r) around a stationary point a of f, then

i) f has a relative minimum at a, if H(a) is positive definite

ii) f has a relative maximum at a, if h(a) is negative definite
iii) f has a saddle point at a, if H(a) has both positive and negative eigen values.
Proof : Using the notations that we have used in the discussion just before this theorem, we
can write f(a + h) – f(a) = hH(a)ht + ||h||2E(a, ). Since E(a, ) 0, as h 0, we can
conclude that the sign of f(a + h) – f(a) will depend on that of hH(a)ht .

i) This value will be positive for all h, if H(a) is positive definite. Hence,

f(a + h) – f(a) > 0 for all h, such that 0 < ||h|| < r. This tells us that f(a + h) f(a) for
every h B(a; r), that is , a is a relative minimum point of f.

The argument for proving ii) and iii) are exactly similar, and we are sure you can write those.

.Remark 7.2 : i) If an even principal minor, that is a principal minor of even order is
negative, then the point is a saddle point.

ii) If detH(a) = 0, the test is inconclusive, and a is called a degenerate stationary

point of f.

Go through the following examples carefully, they illustrate our discussion here.

Example 7.2: Locate and classify the stationary points of the functions given by

i) x2 + xy + 2x + 2y + 1, ii) x3 + y3 – 3xy, iii) (x − 1)exy.

Solution : i) Let f(x, y) = x2 + xy + 2x + 2y + 1. Then fx = 2x + y + 2, f y = x + 2.

fx = fy = 0 x + 2 = 0, and 2x + y + 2 = 0 x = − 2 and y = 2. Therefore, f has only one

stationary point, ( − 2, 2). Now, fxx = 2, fyy = 1, and fxy = 0.

Thus, H(( −2, 2)) = , and det (H(( − 2, 2))) = −1.

Therefore, f has a saddle point at ( − 2, 2).

ii) Let f(x, y) = x3 + y3 – 3xy. Then, fx = 3x2 – 3y, fy = 3y2 – 3x.

fx = fy = 0 y = x2, and x = y2 x = y= 0, or x = y = 1. Therefore, the stationary points

are (0, 0) and (1, 1). Now, fxx = 6x, fyy = 6y, and fxy = - 3. Hence,

H((0, 0)) = . det(H(0, 0)) = - 9 < 0, and (0, 0) is a saddle point.

H((1, 1)) = . The principal minors are 6, and 27. Both are positive, and hence f
has a local minimum at (1, 1).

iii) Let f(x, y) = (x - 1)exy. Then fx = exy(xy – y + 1), fy = x(x - 1)exy

fx = 0 xy – y + 1 = 0, and fy = 0 x(x - 1) = 0 x = 0, or x = 1.

x=0 y = 1, and x = 1 contradicts fx = 0. So, (0, 1) is the only stationary point.

fxx = exy(y + xy2 – y2 + y), fxy = exy(x – 1 + x2y – xy + x), fyy = x2(x - 1)exy.

Therefore, H((0, 1)) = . det(H(0, 1)) = - 1 < 0. Hence, (0, 1) is a saddle point.

Example 7.3 : Locate and classify the stationary points of f(x, y, z) = i) xyz ,

ii) x2y + y2z + z2 - 8 x, iii) x2 – xy + yz3 – 6z.

Solution : i) fx = yz − 2x2yz = yz(1 – 2x2)

fy = xz(1 – 2y2) , fz = xy(1 – 2z2). Equating to zero these partial

derivatives, and solving the resultant equations, we get (a, 0, 0), (0, b, 0), (0, 0, c),
( ), where a, b, c are real numbers, as the stationary points.

fxx = − 4xyz − 2xyz(1 – 2x2)

fxy = z(1 – 2x2) − 2y2z(1 – 2x2) ,

fyz = x(1 – 2y2) – 2xz2 (1 – 2y2).

We have indicated the procedure. We are sure now you will be able to get fxz, fyy, and fzz.
Evaluating these second order partial derivatives at the stationary points, we find,

H((a, 0, 0)) = detH((a, 0, 0)) = 0. Therefore, (a, 0, 0) is a degenerate

point of f. Similarly, (0, b, 0) and ( 0, 0, c) are also degenerate points.

H(( )) = . The minors of this matrix are

, 2e- 3, . Therefore, ( ) is a local maximum. Check the

remaining 7 points. You should get local maxima at ( ), ( ), ( ),
and local minima at ( ), ( ), ( ), ( ).

ii) fx = 2xy - 8 , fy = x2 + 2yz, fz = y2 + 2z. Equating these to zero, we get xy = 4 ,

x2 = −2yz , y2 = − 2z. If x, y, and z are non-zero, we get x = 2 , y = 2, and z = − 2. So, the

stationary points are (0, 0, 0) and (2 , 2, − 2).

You will find that (0, 0, 0) is a degenerate stationary point, and (2 , 2, − 2) is a saddle point.
iii) fx = 2x – y, fy = - x + z3, fz = 3yz2 – 6. Equating these to zero, we get (1, 2, 1) as the

stationary point. Check that H((1, 2, 1)) = , and the principal minors are 2, -

1, - 6. Hence, (1, 2, 1) is a saddle point.

See if you can solve these exercises now.

Exercises:

x x
1) Find the stationary points of f(x, y) = i) ii) (x + y)exy.
x  y2  4
2
x  y2  4
2

2) Find the extreme values of f(x, y) = x2 + y3 + 3xy2 – 2x.

3) Is (0, 0) an extreme point of 2cos(x + y) + exy?

4) Locate and classify the stationary points of

i) f(x, y) = (2 - x)(4 - y)(x + y - 3), ii) f(x, y, z) = 4xyz – x4 – y4 – z4,

iii) f(x, y, z) = 64x2y2 – z2 + 16x + 32y + z, iv) f(x, y, z) = xyz(x + y+ z – 1).

7.4 LAGRANGE’S MULTIPLIERS

Look at these situations: i) A rectangular cardboard sheet is given. We have to make a closed
box out of it. What is the maximum volume that is possible?

ii) Temperature varies on a metal surface according to some formula. Where do the
maximum and minimum temperature occur on the surface?

In both these problems we have to maximize or minimize a certain function: volume in the
first case, and temperature in the second. So these are max-min. Problems. But there is a
difference between these and the problems considered in the last section. Here, an additional
constraint or condition is imposed. The given cardboard sheet has a fixed area. The
maximum/minimum temperature points are to be on the given surface.

In this section we shall see how such problems are solved. A very useful method was
developed by Joseph Louis Lagrange. This method gives a necessary condition for the
extreme points of a function. We now state the theorem and then illustrate its use through
some examples.

Theorem 7.7 : Let f : Rn R, and f C1. Suppose g1, g2, . . ., gm (m < n) are functions
belonging to C1, which vanish on an open set E in Rn. If a E is an extreme point of f, and if
(a), (a), . . . , (a) are independent vectors in Rn, then there exist real numbers, ,
, . . . , , such that
Dif(a) + Dig1(a) + Dig2(a) + . . . + Digm(a) = 0, i = 1, 2, . . . , n.

We can also write the vector equation f(a) + (a) = 0.

When we want to find the extreme values of a function f : Rn R, f C1, subject to some
constraints, g1(x1, x2, . . . ,xn) = 0, g2(x1, x2, . . . ,xn) = 0, . . . , gm(x1, x2, . . . ,xn) = 0, where
m < n, we set up the n equations

Dif(a) + Dig1(a) + Dig2(a) + . . . + Digm(a) = 0, i = 1, 2, . . . , n.

These n equations, along with the m equations, g1(x1, x2, . . . ,xn) = 0, g2(x1, x2, . . . ,xn) = 0, .
. . , gm(x1, x2, . . . ,xn) = 0, are then solved to get the values of the n + m unknowns, x1, x2, . . .
,xn, , , . . . , . The solutions x = (x1, x2, . . . ,xn) are the stationary points, and contain the
extreme points of f .

, ,..., are called Lagrange’s Multipliers. We use one multiplier for each
constraint.

To analytically classify these stationary points into local maximum, minimum, or saddle, is a
very complicated process. It is usually easier to look at the physical or geometrical aspect of
the problem to arrive at any conclusion. We now solve a few problems, so that the entire
process is clear to you.

Example 7.4 : Find the dimensions of the box with maximum volume that can be made with
a cardboard sheet of size 12 cm2.

Solution : If the dimensions of the box are x, y, z cms, then its volume V = xyz c. cms. And
surface area is 2(xy + yz + xz) sq. cms. Here we have to maximize V, subject to a constraint
2(xy + yz + xz) = 12, or (xy + yz + xz) = 6. So, f(x, y, z) = xyz, and

g(x, y, z) = xy + yz + xz – 6. Hence,

f(x, y, z) + g(x, y, z) = 0

f x + gx = 0 yz + (y + z) = 0, fy + gy = 0 xz + (x + z) = 0, f z + gz = 0 xy +
(x + y) = 0.

xyz = (xy + xz) = (xy + yz) = (xz + yz). If = 0, then V = 0, which is the minimum
volume. If 0, then xy + xz = xy + yz = xz + yz. That is, x = y = z (unless, of course, x =
y = z = 0).

Therefore, xy + yz + xz = 6 3x2 = 6 x= cms. Thus, V = 2 c. cms. is the

maximum volume.

Example 7.5 : Find the extreme values of the function given by f(x, y, z) = 2x + y + 3z,
subject to x2 + y2 = 2, x +z = 5.
Solution : Let g1(x, y, z) = x2 + y2 – 2 = 0, and g2(x, y, z) = x + z – 5 = 0. Then

=0

fx + g1x + g2x = 0 2+2 x+ =0

fy + g1y + g2y = 0 1+2 =0

fz + g1z + g2z = 0 3+ = 0. Therefore, = − 3, 2 x = 1, and 2 = − 1.

=0 = −2. But = −3. Therefore cannot be zero. Hence, x = , y= .

Substituting these values in x2 + y2 = 2, we get = . This gives, x = 1, y = 1. Hence,

the stationary points are (1, - 1, 4) and ( - 1, 1, 6), and the extreme values are 13 and 17.

Example 7.6 : Find the minimum distance of a point on the intersection of the planes,

x + y – z = 0, and x + 3y + z = 2 from the origin.

Solution : The distance of P(x, y, z) from the origin is . So, we need to

minimize f(x, y, z) = , subject to g1(x, y, z) = x + y – z = 0, and

g2(x, y, z) = x + 3y + z – 2 = 0.

=0

fx + g1x + g2x = 0 2x + + =0

fy + g1y + g2y = 0 2y + +3 =0

fz + g1z + g2z = 0 2x - + = 0. Therefore, x = ,y= ,

z= . Putting these values in x + y – z = 0, we get + = 0. Therefore, x = 0 and

y = z. Using this in x + 3y + z – 2 = 0, we get y = z = ½. Thus, the stationary point is

(0, 1/2, 1/2). The distance of this point from the origin is .

Geometrically, the constraints are equations of two planes. There is no maximum to the
distance of a point on their line of intersection from the origin. So, the stationary point is a
minimum point.

Here are some problems you can try.

x2 y2
1) Find the extreme values of the function f(x, y) = xy on the surface  = 1.
8 2
 x y
2) Find the extreme values of z =  on the unit circle in the xy-plane.
2 3

3) Find the distance of the point (10, 1, − 6) from the intersection of the planes,

x + y + 2z = 5 and 2x – 3y + z = 12.

7.5 SUMMARY

In this chapter we have introduced Taylor’s theorem for functions of several variables. We
have also seen how to get Taylor polynomials of a given order for a given function. Of
course, to be able to do this, the function must have continuous partial derivatives of higher
orders.

We have then discussed the location of maxima and minima of a real-valued function of
several variables. This has tremendous applications in diverse fields of study. In particular,
we have proved that the extreme points of a function are located among the points at which
the gradient vector of the function is zero. That is, the points at which all the first order partial
derivatives are zero. The classification of these points into maxima, minima, or saddle points
depends on the signs of the principal minors of the Hessian matrix.

We pointed out that there are some situations, where we need to find the extreme values
subject to certain constraints. Such problems, and the method of tackling them is also
discussed, and illustrated through some examples.
 8
INVERSE AND IMPLICIT FUNCTION THEOREMS
Unit Structure

8.0 Objectives

8.1 Introduction

8.2 Inverse Function Theorem

8.3 Implicit Function Theorem

8.4 Summary

8.0 OBJECTIVES
After reading this chapter, you should be able to

 state and prove Inverse Function Theorem for functions of several variables
 check if some simple functions are locally invertible
 state and prove Implicit Function Theorem for functions of several variables

8.1 INTRODUCTION
In this chapter we introduce two very important theorems. You have not come across these
theorems even for functions of a single variable. In each case, we shall first discuss the single
variable case, and then extend the concept to functions of several variables. A word of
caution : these theorems are not easy. To help you understand them better, we are going to
prove some smaller results, and then use them in the proof of the theorems. Do study this
chapter carefully and we are sure you would have no difficulty in digesting the concepts.

8.2 INVERSE FUNCTION THEOREM

The inverse function theorem is a very important theorem in Calculus. You may be familiar
with its one dimensional version. Before we introduce the theorem for functions from Rn to
Rn, we shall recall some results about functions of one variable:

1) If f : [a, b] R is continuous, and f(c) > 0 for some c (a, b), then such that

(c ) (a, b), and f(x) > 0 (c ). In other words, we can always find a
neighbourhood of the point c, in which f(x) has the same sign as f(c).

2) If f : [a, b] R is a continuously differentiable function, and for some

c (a, b), then using 1) we can prove that such that f is an injective function on
(c ) (a, b). Further, f-1: f(c ) (c ) is differentiable at f(c) ,

The statement in 2) is the inverse function theorem. Note that we do not know whether the
inverse of f exists on [a, b]. But what this theorem tells us, is that if , then f is
“locally invertible” at c. For example, we know that the function f : [0, 2 ] R, f(x) = sinx
does not have an inverse. But is a continuous function, and .
So, the theorem says that f is locally invertible at . That is, we can find a neighbourhood
N of , such that f restricted to N has an inverse. Check that f is injective when restricted
to N = ( ), and hence has an inverse on N.

We shall now see if this theorem extends to functions of several variables. Let us start with a
definition.

Definition 8.1 : Let f : E Rn, where E Rn. If f C1, f is said to be locally invertible at
a E, if there exists a neighbourhood N1 of a, N1 E, and a neighbourhood N2 of f(a), such
that f(N1) = N2, f is injective on N1, and f-1 : N2 N1 is a C1 function.

We shall soon state and prove the inverse function theorem. In the proof, we are going to use
some minor results. You have already studied some in the earlier chapters of this course.
Next we state and prove one other result, which will be useful to us.

Theorem 8.1 : Let f = (f1 f2, . . . , fn) : E Rn, where E is an open set in Rn. Suppose f C1.
If the Jacobian of f, J(a) 0 for some a E, then f is injective on a neighbourhood of a in E.

Proof : If X1, X2, . . . , Xn E, we consider a point X = (X1, X2, . . . , Xn) , whose first
n coordinates are the coordinates of X1, the next n are the coordinates of X2, and so on. We
define a function, j, such that

j(X) = det[Djfi(Xi)] = det .

Now, the function j, being an n×n determinant, is a polynomial of its n2 entries, and each
entry, is a continuous function, since f C1. Thus, j is a continuous function on its
domain. We write A = (a, a, . . . , a). Then j(A) = det[Djfi(a)] = J(a) 0. Now, since f C1,
all the entries of j(A) are continuous, and hence, j(A) is also continuous. The continuity of
j(A) ensures that there exists a neighbourhood N of A, such that j(X) 0 , if X N.

In other words, there exists a convex neighbourhood Na of a, such that j(X) 0 , if

X = (X1, X2, . . . , Xn) is a point, for which Xi Na for every i = 1, 2, . . . , n. ..........(8.1)

This Na is the required neighbourhood. We have to show that f is injective on Na. For this,
suppose x, y Na , such that f(x) = f(y). Then fi(x) = fi(y) for every i = 1, 2, . . . , n.

Then, using the Mean Value Theorem for scalar fields (See Remark 6.2 ii).), we get

fi(x) − fi(y) = fi(ci) (x − y) fi(ci) (x − y) = 0 for some ci on the line segment joining
x and y. So, if x – y 0, then fi(ci) = 0 for some ci on the line segment joining x and y, that
is, in the neighbourhood Na, since Na is convex. This means, Djfi(ci) = 0 for every j, 1
. Thus, if C = (c1, c2, . . . , cn), then j(C) = det[Djfi(ci)] = 0. But this contradicts
(8.1). So, we conclude that x – y = 0, which proves that f is injective on Na.

Remark 8.1 : i) A function may not be injective on its entire domain. But if its Jacobian is
non-zero at a point, then it is injective on a neighbourhood of that point. In other words, it is
locally injective.

ii) If the Jacobian is non-zero, then the linear transformation Df, which represents the
derivative of f, is non-singular, and hence, is a linear isomorphism.

Example 8.1 : a) Consider the function f(x, y) = (excosy, exsiny). This function is not
injective, since f(x, 0) = f(x, 2 ). But,

J(x, y) = = e2x 0. Thus, f is locally injective at each point in R2.

Here we have a function, which is locally injective at every point of its domain, but is not
injective on the domain.

b) Consider the function f(x, y) = (x3, y3), defined on R2. The Jacobian of this function is
zero at (0, 0). But the function is locally invertible at (0, 0). In fact, it is an invertible
function.

Theorem 8.2 (The Inverse Function Theorem): Let f = (f1, f2, . . . , fn) C1, f: E Rn , where
E is an open set in Rn. Let T = f(E). Suppose J(a) 0 for some a E. Then there exists a
unique function f-1 from Y to X, where X is open in E, Y is open in T, such that

i) a X, f(a) Y, ii) Y = f(X), iii) f is injective on X, iv) f-1: Y X, f-1(Y) = X, v) f-1 C1

on Y.

Proof : Using Theorem 8.1, we can conclude that f is injective on a neighbourhood N of a in

E. So, f : N f(N) is bijective, and hence has an inverse, f-1 : f(N) N. Let r > 0 be such that
N. Since is compact in Rn , we use Theorem 3.4.1 to conclude that
f( ) is also compact in Rn . Now f is continuous and injective on the compact set
. Hence, using Theorem 3.4.2, we can say that f-1 is continuous on f( ).

Now, B(a, r) is an open set in , and therefore,

 (B(a, r)) is open in f( ). That is, f(B(a, r)) is open in f( ).

Also, f(a) f(B(a, r)). Therefore, there exists a > 0, such that B(f(a), ) f(B(a, r)).

Take X = f-1(B(f(a), )), and Y = B(f(a), ). Then X and Y satisfy i), ii), iii) and iv) in the
statement of the theorem.

To prove the last assertion v) in the statement, we have to show that all the partial derivatives
of all the component functions of f-1 are continuous on Y. For this we first define the function
j(X) = det[Djfi(xi)] , as in Theorem 8.1. Here X = (X1, X2, . . . , Xn). Then, as before, there is a
neighbourhood Na of a, such that j(X) 0, whenever each Xi Na. We can assume that the
neighbourhood N Na. This ensures that j(X) 0, whenever each Xi .

1 1
f ( y  tei )  f ( y)
Now we first prove that Dif-1 exists on Y. Let y Y, and consider ,
t

where ei is the ith coordinate vector, and t is a scalar. Let x = f-1(y), and = f-1(y + tei). Then

f( ) – f(x) = tei. Thus, fi( ) – fi(x) = t, and fj( ) – fj(x) = 0, when i j.

By applying Mean Value Theorem (Remark 6.2 ii)), we can write

f m ( x ' )  f m ( x) x'  x
 fm(xm)  , m = 1, 2, . . . , n. Here xm is a point on the line
t t
segment joining x and .

So, we get a system of n equations (for the n values of m). The left hand side of an equation
in this system is 1, if m = i, otherwise it is 0. The right hand side is of the form

x 1' '  x1 x 2'  x 2 x '  xn

D1fm(xm) + D2fm(xm) + . . . + Dnfm(xm) n , m = 1, 2, . . . , n.
t t t

The determinant of this system of linear equations is j(X), which we know is non-zero. Hence
x 'j  x j
we can solve it by Cramer’s rule and get the variables as the quotient of two
t
determinants. Then, as t tends to zero, approaches x, and hence, each xm also approaches x.
The determinant in the denominator, j(X) = det[Djfi(xi)] then approaches J(x), the Jacobian
x 'j  x j
of f at x, which is again non-zero. Thus, as t tends to zero, the limit of exists. That
t
1 1
f ( y  tei )  f ( y)
is, lim exists. Thus, Dif-1(y) exists for all i, and for all y in Y.
t 0 t

We have obtained the partial derivatives of the components of f-1 as quotients of two
determinants. The entries in these determinants are partial derivatives of the components of f,
which are all continuous. Since a determinant is a polynomial of its entries, we conclude that
the partial derivatives of f-1 are continuous on Y.

Example 8.2 : Show that the function f: R2 R2, f(x, y) = (2xy, x2 – y2) is not invertible on
R2, but is locally invertible at every point of E = {(x, y) | x > 0}. Also find the inverse
function at one such point.

Solution : Here f(1, 1) = f( − 1, − 1) = (2, 0). Therefore f is not injective, and hence is not
invertible on R2. On the other hand, if (x, y) E, then

J(x, y) = = − 4(x2 + y2) 0. Hence by the inverse function theorem, f is locally

invertible.

u u2
Suppose f(x, y) = (u, v). If (x, y) E, then y = , and v = x2 . Therefore,
2x 4x 2

4 2 2 v  v2  u2
2 v  v 2  u 2 1/2
4x - 4x v – u = 0. Thus, x = , and x = ( ) ,
2 2

y = u(2v + 2 )−1/2

8.3 IMPLICIT FUNCTION THEOREM

If x2 + y2 = 0, find . You must have done exercises like this in your under-graduate
dy
classes. Here, we take f(x, y) = x2 + y2, and find fx = 2x, and fy = 2y. Then = 2x/2y = x/y.
dx
Of course, y cannot be zero.

While doing this exercise, actually we have used a theorem, the implicit function theorem. To
recall, in this setting, a function which can be written as y = g(x), is called an explicit
function, and one which can be expressed only as f(x, y) = 0, is called an implicit function.
The implicit function tells us that under certain conditions, we can express an implicit
dy
function as an explicit one, and then we can use this expression to find .
dx

In this section we are going to discuss this implicit function theorem for functions of several
variables. Before we state and prove the general case, we first prove the case for functions
involving only two variables, x and y.

Theorem 8.3 : Let f be a real-valued C1 function, defined on the product , where and
are two intervals in R. Let (a, b) , and f(a, b) = 0, but fy(a, b) 0. Then there
exists an interval I in R, containing a, and a C1 function g : I R, such that g(a) = b, and

f(x, g(x)) = 0 for all x I.

Proof : We consider a function, h: R2, given by h(x, y) = (x, f(x, y)). If we write

h=( ), the Jacobian matrix of h is

Jh(x, y) = = . The determinant of this matrix, is not zero at (a, b).

Thus, h is a C1 function, with a non-zero Jacobian at (a, b). Therefore, by the inverse function
theorem, Theorem 8.2 , we can conclude that h is locally invertible at (a, b). Let u = ( )
be the local inverse of h. You will agree that (x, y) = x for all x and y in R. That is,

u(x, y) = (x, (x, y)) for all x and y in R. We now define g as, g(x) = (x, 0), and show that
it has all the required properties.

Now, since h(a, b) = (a, 0), u(a, 0) = (a, b). This means, (a, 0) = b. Thus, g(a) = b.

Also, (x, 0) = h(u(x, 0)) = h(x, (x, 0)) = h(x, g(x)) = (x, f(x, g(x))). This implies that

f(x, g(x)) = 0.

Since u is a C1 function, g is also C1. Differentiating f(x, g(x)) = 0 with respect to x using
chain rule, we get D1f(x, g(x)) + D2f(x, g(x)) (x) = 0, and thus,

 D1 ( f ( x,g ( x))
(x) = , since D2f(x, g(x)) 0.
D2 f ( x, g ( x))

Basically, this theorem tells us that under certain conditions, the relation f(x, y) = 0, between
x and y can be explicitly written as y = g(x).

Remark 8.2 : If instead of fy(a, b) 0, we take the condition fx(a, b) 0, then we can
express x as an explicit function of y.

Example 8.3 : Can f(x, y) = x3 + y3 – 2xy be expressed by an explicit function y = g(x) in a

neighbourhood of the point (1, 1)?

Solution : Note that f(1,1) = 0, and fy = 3y2 – 2x = 1 at (1, 1). Further, f is a C1 function on R2.
Therefore, we can apply Theorem 8.3, and conclude that there exists a unique function g,
3x 2  2 y
defined on a neighbourhood of 1, such that g(1) = 1. Also, (x) = in this
3 y 2 2 x
neighbourhood.

Example 8.4 : Check whether Theorem 8.3 can be applied at all points, where

f(x, y) = x2 – y2 = 0.
Solution : x2 – y2 = 0 is true at points (0, 0), (1, 1),(1, −1), ( −1, 1), and ( −1, −1). fy = −2y,
and fx = 2x. At the point (0, 0), fx and fy are both zero, and hence we cannot apply the
theorem. At all the remaining points, the function satisfies all the conditions of Theorem 8.3,
and hence it can be applied. You will agree that at each of these points, we will get either

g(x) = x, or g(x) = − x.

We now go a step further, and consider a real-valued function of several variables.

Theorem 8.4 : Let f be a real-valued C1 function, defined on an open set, U, in Rn. Let

a = (a1, a2, ... , an-1) Rn-1, such that (a, b) U, f(a, b) = 0, and Dnf(a, b) 0. Then there
exists a unique C1 function g, defined on a neighbourhood N of a, such that g(a) = b, and

f(x, g(x)) = 0 for all x N.

Proof : We consider a function h : U Rn−1 R, defined by h(x, y) = (x, f(x, y)). If we write
h = (h1, h2, ... , hn), then hi(x, y) = xi, for 1 i n – 1, and hn(x, y) = f(x, y). Therefore, the
Jacobian matrix of h is given by

Jh = .

The determinant of this matrix is Dnf, which is non-zero. Therefore, we can apply the inverse
function theorem (Theorem 8.2), and conclude that h is locally invertible at (a, b). If u is the
local inverse of h, and we write u = (u1, u2), then you will see that u1(x, y) = x for all (x, y).
Thus, u(x, y) = (x, u2(x, y)) for all (x, y). We now define g(x) = u2(x, 0), and show that this
has the required properties.

Now, u(a, 0) = (a, b). This gives g(a) = u2(a, 0) = b.

Also, (x, 0) = h(u(x, 0)) = h(x, (x, 0)) = h(x, g(x)) = (x, f(x, g(x))). This implies that

f(x, g(x)) = 0.

Example 8.5 : Examine whether the function f(x, y, z) = x2 + y2 – 4 can be expressed as a

function y = g(x, z) in a neighbourhood of the point (0, -2, 0).

Solution : We note that f(0, −2, 0) = 0, and D2f = 2y = − 4 at (0, −2, 0). So, applying the
implicit function theorem, there exists the required neighbourhood of (0, −2, 0). In fact, you
can check that in the neighbourhood, N = B((0, − 2, 0), 1), we can express the function as

y = − (4 – x2)1/2 .
Here are some exercises that you should try :

1) Determine whether the following functions are locally invertible at the given points :

i) f(x, y) = (x3y + 3, y2) at (1, 3)

ii) f(x, y, z) = (excosy, exsinz, z) at (1, 1, 1).

2) For each of the following functions, show that the equation f(x, y, z) = 0 defines a
continuously differentiable function z = g(x, y), in a neighbourhood of the given point:

i) f(x, y, z) = x3 + y3+ z3 – xyz – 2 , (1, 1, 1)

ii) f(x, y, z) = x2 + y3 – xysinz , (1, - 1, 0).

That brings us to the end of this chapter. We hope you have studied the concepts carefully,
and have understood them.

8.4 LET US SUM UP

In this chapter we have discussed two very important theorems: the inverse function theorem,
and the implicit function theorem. The proofs of these theorems are a little complicated. So
we have tried to go step by step from functions of one variable to functions of many
variables.

The Inverse Function Theorem: gives the conditions under which a function, even though not
invertible on its domain, is seen to be locally invertible. The Jacobian of the function being
non-zero at a point ensures the local invertibility of the function in a neighbourhood of that
point.

The Implicit Function Theorem: gives the conditions, under which an implicit relationship
between variables can be expressed in an explicit manner. Here, again, the Jacobian plays an
important role.
 1

1
RIEMANN INTEGRAL - I
Unit Structure :

1.1 Introduction
1.2 Partition
1.3 Riemann Criterion
1.4 Properties of Riemann Integral
1.5 Review
1.6 Unit End Exercise

1.1 INTRODUCTION

The Riemann integral dealt with in calculus courses, is well

suited for computations but less suited for dealing with limit
processes.

Bernhard Riemann in 1868 introduced Riemann integral. He

need to prove some new result about Fourier and trigonometric
series. Riemann integral is based on idea of dividing. The domain of
function into small units over each such unit or sub-interval we erect
an approximation rectangle. The sum of the area of these rectangles
approximates the area under the curve.

As the partition of the interval becomes thinner, the number

of sub-interval becomes greater. The approximating rectangles
become narrower and more precise. Hence area under the curve is
more accurate. As limits of sub-interval tends to zero, the values of
the sum of the areas of the rectangles tends to the value of an
integral. Hence the area under curve to be equal to the value of the
integral.

Before going for exact definition of Riemann explained the

following definitions.

1.2 PARTITION

A closed rectangle in  n is a subset A of  n of the forms.

A   a1 , b1    a2 , b2   ....   an , bn  where ai  bi   . Note that
 x1 , x2 ,...., xn   A iff ai  xi  bi i .
 2

The points x1 , x2 ,...., xn are called the partition points.

The closed interval I1   x0 , x1  , I 2   x1 , x2  ,......, I n   xn 1 , xn  are

called the component internal of  a, b  .

Norm : The norm of a portion P is the length of the largest sub-

internal of P and is denoted by P .

For example : Suppose that P1  t0 , t1 ,....tk is a partition of  a1 , b1  and

P2  S0 ,...., Sr is a partition of  a2 , b2  . Then the partition P   P1.P2  of
 a1 , b1    a2 , b2  divides the closed rectangle  a1 , b1    a2 , b2  into Kr-
gub rectangles.

In general if Pi divides  ai , bi  into ki sub-interval then

P   P1 ,....Pn   a1 , b1   ....   an , bn 
into K  k1k2 .....kn sub-rectangle.
These sub-rectangles are called sub-rectangles of the partition p.

Refinement :
Definition : Let A be a rectangle in  n and f : A   be a bounded
function and P be partition of A for each sub-rectangles of the
partition.

ms  f   inf  f  x  : x  S 
 g .l.b.of  f on xs 1 , xs 

Ms  f   sup  f  x  : x  S 
 l.u.b.of  f on xs 1 , xs 
where S  1, 2,...., n

The lower and upper sums of f for ‘p’ are defined by

L  f , p    ms  f   s  and U  f , p    M s  f   s 
s s

Since ms  M s we have L  f , p   U  f , p 

Refinement of a partition : Let P   P1 , P2 ,..., Pn  and P*   P1* ,..., Pn* 

be partition of a rectangle A in  n . We say that a partition P* is a
refinement of P if P  P* .
 3

If P1 and P2 are two partition of A then P  P1  P2 is also a

partition of A is called the common refinement of P1 and P2 .

A function f : A   is called integrable on the rectangle A

in  if ' f ' is bounded  g.l.b of the set of all upper sum of ' f ' and
n

l.u.b of the set of all lower sum of ' f ' exist.

Let U  f   inf U  f , p 
L  f   sup L  f , p 

If U  f   L  f  is called ' f ' is R-integrable over A.

 if can be written as U  f   L  f    f .
A

Theorem :
Let P and P be partitions of a rectangle A in  n . If P
refines P then show that L  f , p   L  f , P  and U  f , P   U  f , p  .

Proof :
Let a function f : A   is bounded on A P & P* are two
partition of A and P is retinement to P.

Any subrectangle S of P is union of some subrectangles

s1 , s2 ,...., sk of P and V  S   V  s1   V  s2   .....  V  sk  .

Now ms  f   inf  f  x  ; x  s  inf  f  x  ; x  si 

 ms  f   ms i  f i  1,...., k

L  f , p    ms  f  V  s 
s p

 ms  f V  s   ms  f  V  s1   ....  V  sk  
 ms1  f V  s1   .....  ms  f V  sk 
k

The sum of LHS for all subrectangle si of P will get

L  f , P  .

 L  f , p   L  f , p1 
Now, M s  f   sup  f  x  ; x  S 
 sup  f  x  ; x  Si 
Ms  f   Ms  f i  1,..., K
i
 4

U  f , p    ms  f  V  s 
s p

Now, Msi  f V  S   Ms  f  V  S1   V  S 2   ....  V  Sk  

 Ms  f  V  s1   .....  M s  f V  s2   ....  M s  f V  sk 

Taking the of L.H.S. for all subrectangle Si of P will get

U  f , P  U  f , P   U  f , P  .

Theorem :
Let P1 & P2 be partitions of rectangle A & f : A   be
bounded function. Show that L  f , P2   U  f , P1  &
L  f , P1     f , P2  .

Proof :
Let a function f : A   be a bounded find P1 & P2 are any
two partition of A.

Let P  P1  P2
 P is a refinement of both P1 & P2
U  f , P   U  f , P1  ……….. (I)
U  f , P   U  f , P2  ……….. (II)
L  f , P   L  f , P1  ……….. (III)
L  f , P   L  f , P2  ……….. (IV)

 We get U  f , P1   U  f , P   L  f , P   L  f , P2  .

Hence U  f , P1   L  f , P2 

Similarly, U  f 2 , P2   U  f , P   L  f , P   L  f , P1  .

Hence, U  f , P2   L  f , P1 

Theorem :
Let a function f : A   be bounded on A then for any
 0,  a partition P on A such that U  f , P   U  f    and
L  f , P  L  f   
 5

Proof :
Let a function f :A be bounded on A
U  f   inf U  f , P  and L  f   sup L  f , P  for any  0, 
partitions P1 & P2 of A such that U  f , P1   U  f    &
L  f , P2   L  f    .

Let P  P1  P2 the common refinement of P1 and P2 .

U  f , P   U  f , P1   U  f   
L  f , P   L  f , P2   L  f   
 U  f , P  U  f   
L  f , P  L  f   

1.3 RIEMANN CRITERION

Let A be a rectangle in  n A bounded function f : A   is

integrable iff for every  0 , there is a partition P of A such that
U  f , P   L  f , P   .

Proof :
Let a function f : A   is bounded.
U  f   inf U  f , P 
L  f   sup L  f , P 

Let f be integrable of A
U  f   L  f 
for any  0, a partition P on A such that U  f , p   U  f   2
and L  f , p   L  f   2 .

U  f , p   U  f   2 &  L  f , p    L  f   2 .
U  f , p   L  f , P   U  f   2  L  f   2 .

U  f , p   L  f  

Conversely,
Let for any  0, a partition P on A such that
U  f , p   L  f , P   .

U  P, f   U  f    U  f   L  f     L  f   L  f , P   
 6

Since U  f , P   U  f   o,
U  f   L f   o
and L  f   L  f , P   o
 we have, o  U  f   L  f  

Since  is arbitrary, U  f   L  f 
 f is integrable over A.

Example 1
Let A be a rectangle in  n and f : A   be a constant
function. Show that f is integrable and  f  C.V  A for some C   .
A

Solution :
f  x   Cx  A
 f is bounded on A

Let P be a partition of A
ms  f   inf  f  x  ; x  s  C
M s  f   sup  f  x  ; x  s  C

 L  f , P    ms  f V  S   C  V  S   CV  A 
S S

U  f , P    M s  f V  S   C  V  S   CV  A 
S S

U  f   L  f   CV  A 
 f is integrable over A.
 by Reimann criterion,  0 s.t.
 f  C.V  A for some C   .
A

Example 2 :
Let F :  0,1 X  0,1  
 oif xisrational
f  x, y   
1if  xisirrational

Show that ‘f’ is not integrable.

Solution :
Let P be a partition of  0,1   0,1 into S subport of P.
 7

Take any point   x1 , y1   S such that x is rational.

 f  x, y   o and   x1 , y1   S such that x1 , is irrational

 f  x1 , y1   1
 ms  f   inf  f  x  ; x  S   0
M s  f   sup  f  x  ; x  S   1

L  f , P    ms  f  V  S   0
S

U  f , P    M s  f V  S   1
S

U  f   1, L  f   0
U  f   L  f 
 f is not integrable  0,1   0,1

1.4 PROPERTIES OF RIEMANN INTEGRAL

1) Let f : A   be integrable and g  f except at finitely many

points show that g is integrable and  f   g .
A A

Proof :
Since f is integrable over A.
 by Riemann Criterion,  a partition P of A.
Such that U  f , P   L  f , P   ……… (I)

Let P be a refinement of P, such that

1) x  A with f  x  g  x  , it belongs to 2n subrectangles of P

2) V  S  
2 n 1
d u   

Where d = numbers of points in A at which f  g

u  sup  g  x   inf  f  x 
xA xA

  inf  g  x   sup  f  x 
x A x A

 P is refines P, we have
L  f , P   L  f , P    U  f , P   U  f , P 
U  f , P   L  f , P   U  f , P   L  f , P  
 8

Now
U  g , P   U  f , P 

   Ms  g   Msij  f  V  sij  
d
 ij
i 1

 On other rectangle, f  g and so Msij  g   Msij  f  .

 Msij  g   sup  g  x  & Msij  f   inf  f  x   Msij  f   inf  f  x 
x A x A xA

Msij  g   Msij  f   u
2 
n

U  g , P   U  f , P      u  V  Sij 
d

i 1  j 1 

 
2n
Let V  sup V  Sij   U  g , P   U  f , P    uV  d 2 nu.v
d
1 1
…….
i 1 j 1

(II)

Now similarly we get L  g , P1   L  f , P1   d 2n V ……... (III)

by (II) & (III) we get.

U  g , P1   L  g , P1   U  f , P1   d 2 n u  L  f , P1   d 2 n 

  d 2n  u   V
2
 d 2  u    
n

   
2 d 2n 1  u    2 2
U  g , P1   L  g , P1  

By Reimann Criterion G is integrable by equation (II)

U  g , P1  U  f , P1   d 2 n uv
 U  g , P1   U  f , P1   d 2 n u

Note that  g  U  g , P1   U  f , P1   d 2n u
A

 L  f , P1    d 2n u
2
d 2n u 
 L  f , P1     n 1
2 d 2 u   
 9

 L  f , P1     
2 2
 L f , P 
1

  f 
A

This is true for any  0

 g   f ………………….. (IV)
A A

Now  g  L  g , P   L  f , P    2
A

 U  f , P 
  f  f  
2
A A

  f  inf U  f , P 
A

  g   f  
2
A A

 This is true for any  0

  g   f ……… (V)
A A

 from (IV) & (V) we get

g  f
A A

2) Let f : A   be integrable, for any partition P of A and sub-

rectangle S, show that

i) ms  f   ms  g   ms  f  g  and
ii) M s  f   M s  g   M s  f  g 

Deduce that
L  f , P   L  g , P   L  f  g , P  and
U  f  g, P  U  f , P  U  g, P 

Solution :
Let P be a partition of A and S be a Subrectangle
 ms  f   inf  f  x  ; x  S 
 ms  f   f  x  x  S
 10

Similarly ms  g   g  x  x  S
 ms  f   ms  g   f  x   g  x  x  S
 ms  f   ms  g  is lower bound of
 f  x   g  x  ; x  S    f  g  x  ; x  S 
 ms  f   ms  g  is lower bound of
 f  x   g  x  ; x  S    f  g  x  ; x  S 
 m  f   m  g   inf  f  g  x  ; x  S 
s s

 ms  f  g 
 ms  f   ms  g   ms  f  g 

ii) Ms  f   sub  f  x  ; x  s
 Ms  f   f  x x  s

Similarly Ms  g   g  x  x  S
 Ms  f   Ms  g   f  x   g  x  x  S
 Ms  f   Ms  g  is upper bound of
 f  x   g  x  ; x  S    f  g  x  ; x  S 
 Ms  f   Ms  g   sup  f  g  x  ; x  S   Ms  f  g 

 Ms  f   Ms  g   Ms  f  g 

Hence,
L  f , P   L  g , P     Ms  f   Ms  g  V  S 
s p

   Ms  f  g  V  S 
s p

 L  f  g , P 

 L  f , P   L  g, P   L  f  g, P 
U  f , P   U  g , P     Ms  f   Ms  g  V  S 
s

   Ms  f  g   V  S 
s

 U  f  g , P 
U  f , P   U  g , P   U  f  g , P  Proved.
 11

3) Let f : A   be integrable, & g : A   integrable than show

that f  g is integrable and   f  g    f   g .
A A A

Proof :
Let P be any partition of A then
U  f  g , P   L  f  g , P   U  f , P   U  g , P    L  f , P   L  g , P  
 U  f , P   U  g , P   L  f , P   L  g , P  …………………….. (I)
 f is integrable.

By Rieman interion for given  0, a partition P, of A such

that U  f , P1   L  f , P1    2 ……………………………….… (II)

Similarly  g is integrable for  0, a partition P2 of A such that

U  g , P2   L  f , P2    ……………………………………… (III)
2

Then P*  P1  P2 is a refinement of both P1 & P2 .

 L  f , P1   L  f , P*  ; U  f , P1   U  f , P*  & L  g , P2   L  f , P*  ;
U  g , P2   U  g , P*  ………………………………………….. (IV)

 2  U  f , P1   L  f , P1   U  f , P*   L  f , P* 
 2  U  g , P2   L  g , P2   U  g , P*   L  g , P*  ……………….. (V)

The equation I is true for any partition P of A.

In general, it is true for partition P* of A

U  f  g , P*   L  f  g , P* 
 U  f , P*   L  f , P*   U  g , P*   L  g , P* 
  2  2 
U  f  g , P*   L  f  g , P*  
By Riemann Criterian f  g is integrable.

Let  0 since f  sup  f , P so a partition P such that

A

 f   f , P   2 .
A
1
 12

Similarly a partition P2 , P3 ,....Pn of A S

 g  L  g , P   2
A
2

U  f , P3    f  
2
A

U  g , P4    g  
2
A

Let P  P1  P2  P3  P4 .
Then  f   f , P   2  L  f , P   2
A
1

Similarly  g  L  g , P    2
A

U  f , P   f  and U  g , P    g   2
2
A A

 f   g   L  f , P   L  g , P   L  f  g , P    f  g
A A A

 U  f  g, P 
 U  f , P U  g, P 
  f    g 
2 2
A A

  f   g 
A A

 f   g   f  g   f   g 
A A A A A

This is true for any  0

 f   g   f  g   f   g   f  g   f   g
A A A A A A A A

4) Let f : A   be integrable for any constant C, show that

  Cf   C  f .
A A

Proof :
Let C  
Case 1
Let  0 and suppose C  0 .
Let P be a partition of A and S be a subrectangle of P.
 13

M s  Cf   sup  Cf  x  ; x  S 
 sup Cf  x  ; x  S 
 C sup  f  x  ; x  S 
 CMs  f 

Similarly,
ms  Cf   Cms  f 
U  Cf , P    Ms  Cf  v  S   C  Ms  f  v  S 
S S

 CU  f , P 
Similarly L  Cf , P   CL  f , P 
 f is integrable for above  0,  a partition P of A such that
U  f , P  L f , P   C
U  Cf , P   L  Cf , P   CU  f , P   CL  f , P 
 C U  f , P   L  f , P  
 C    C
C
By Riemann Criteria.
 Cf  is integrable
for  0, a partition P of A such that
 
C  f   C   f     CL  f , P   L  Cf , P 
C
A A 
  Cf  U  Cf , P 
A

 
 CU  f , P   C   f   
C
A 
   
   f      Cf  C   f     C  f  
C C
A  A A  A

This is true for any  0

C  f    Cf   C  f
A A A

  Cf  C  f
A A

Case II
Now suppose C  0
Let P be a partition of A and S be any subrectangle in P.
 Ms  Cf   CMs  f  and
 14

ms  Cf   CMs  f 
 L  Cf , P   CU  f , P  and
U  Cf , P   CL  f , P 
 f is integrable for above  0, a partition P of A such that
U  f , P  L  f , P  
 C 
U  Cf , P   L  Cf , P   CL  f , P   CU  f , P 
 C U  f , P   L  f , P  
 C 
C

By Riemann Criteria  Cf  is integrable.
for  0, a partition P of A such that C  f    Cf  C  f   .
A A A

This is true for every  0

C  f   Cf  C  f
A A A

  Cf  C  f
A A

Example 3:
Let f , g : A  R be integrable & suppose f  g show that
 f g .
A A

Solution :
By definition f  inf U  f , P  and  g  inf U  g , P  .
A A

Let P be any partition of A & S be any subrectangle in P

as f  g
ms  f   ms  g 
U  f , P   U  g , P 
inf U  f , P   inf U  g , P 

This is true for any partition

 f   g
A A
 15

Example 4:
If f : A   is integrable show that if is integrable and

 f 
A A
f .

Solution :
 Suppose f is integrable first we have to show that f is integrable.

Let P be a partition of A & S be subrectangle of P then


Ms  f   sup f  x  ; x  S 
 sup  f  x  ; x  S 
 sup  f  x  ; x  S 
 Ms  f 
Similarly
Ms  f
  Ms  f 
U  f , P    M  f  V  S    M  f  V  S 
s s
S S

L  f , P    ms  f  V  S 
S

  
  M s  f   ms  f  V  S    M s  f   ms  f  V  S 
P P

 U  f , P   L  f , P 

f isintegrable, for  0, a partition P such that

U  f , P   L  f , P   .

U  f , P   L  f , P   U  f , P   L  f , P  
 By Riemann criteria
f is integrable over  .

Now F  inf U  f , P 
P
A

 inf  M s  f  V  S 
P
S P

 inf  M s  f  V  S 
P

 inf  M s f V  S 
P
P
 16

 inf  M s f V  S 
P


 inf U  f , P  
 f  f
A A

Example 5:
Let f : A   and P be a partition of A show that f is
integrable iff for each sub-rectangle S the function f s which consist

 f   s .
of f restricted to S is integrable and that in this case f
A S S

 Suppose f : A   is integrable.
Let P be a partition of A & S be a sub-rectangle in P.

Now to show that f s ; S   is integrable.

Let  0, a partition P of A such that U  f , P   L  f , P   ( f
is integrable)
Let P  P  P then P1 is refinement of both P & P .
U  f , P   U  f , P1  & L  f , P   L  f , P1 
U  f , P1   L  f , P1   U  f , P   L  f , P  ………………… (I)
 P1 is refinement of P
 S is union of some subrectangle of P1 say S  U si .
i 1

 U  f , P1   L  f , P1     M s  f   ms  f  V  S  for all rectangle.

S P1

 
k
  Msi  f   msi  f  V  S 
i 1

U f  S  
,P L f
S
,P 
 By Riemann Criterion
f is integrable.
S

Conversely, Suppose f S is integrable for each S  P .

To show that f is integrable.

Let  0, partition PS of S such that
 17

U f  s  
, PS  L f
s 
, PS   k ………………………………. (II)

f is integrable for each S  P where K is number of rectangle in

S
P.

Let P1 be the partition of A obtained by taking all the

subrectangle defined in the partition PS .

There is a refinement PS1 of PS containing subrectangles in

P1 .

U  f s , PS1   L  f s , PS1    k …………………………… (III)

U  f , P1   L  f , P1     M  f   m  f  V  S 
S1 S1
1

S P
1 1

 
 
    Ms1  f   ms1  f  V  S 1  
 
S P  S1PS1 

  U  f s , PS1   L  f s , PS1 
S P

  k
S P

 k , k 

 By Riemann Criterian f is integrable.

Let  0
 
  f
S P  S
S  k    L  f S , PS 
 SP
 
    m1s  f  V  S   

S P  S 1PS1



Let P1 be a partition of A, obtained by taking allthe subrectangle

defined in PS .
 18

 
    f S  
k    ms1  f  V  S 1  
S P  S  S1P1
 L  f , P1    f  U  f , P1 
A

  M  f V  S 
s1
1

S1P1

 
    M s1  f V  S 1  
S P  S 1P1 

 
  U  f S , PS       f S  C 
S P  S
k
S P 
  f S C f    f S  
S P A S P

This is true for all  0

   f S  f   f S
S P SP S

 f    f S
A S P S

Example 6:
Let f : A   be a continues function show that f is
integrable on A.
Solution :
Let f : A   be a continuous function to show that f is
integrable.

Let  0 , since A is closed rectangle it is closed and bounded

in  . n

 A is compact.

f is continuous function on compact set  f is uniformly

continuously on  .
 for the above  0,  0 such that x, g  A,
x  y    f  x   f  y    V  A .

Let P be a partition of A such that side length of each

subrectangle is less than  n .

If x, y  S for some subrectangles S then

x y   x1  y1   ....   xn  yn 
2 2
 19
2

 n  S  

 n
f  x   f  y    V  A

 S is compact
 f is continuous
 f attains its bound in S.

Let S1 , S 2 ,....., S k be the subrectangle in A. Then for

1  i  k , xi , yi  Si such that Msi  f   f  xi  msi  f   f  yi  .

 
k
U  f , P   L  f , P    Msi  f   msi  f  V  Si 
i 1
k
   f  xi   f  yi   V  Si 
i 1
k
  k
 V  Si    V  Si 
i 1 V  A  V  A  V  A

 V  A  
V  A

 By Riemann Criterion f is integrable.

1.5 REVIEW

After reading this chapter you would be knowing.

 Defining R-integral over a rectangle in  n
 Properties of R-integrals
 R-integrabal functions
 Continuity of functions using  -intervals.

1.6 UNIT END EXERCISE

I) Let f ;  0,1   0,1   be defined by

f  x, y   0if 0  y  1
3
 3if  1  y  1
3
show that f is integrable.

II) Let Q be rectangle in  n & f ; Q   be any bounded

function.
 20

a) Show that for any partition P of Q L  f , P   U  f , P 

b) Show that upper integral of function f exit.

III) Let f be a continuous non-negative function on  0,1 and

suppose there exist x0   a, b  such that f  x0   0 show that
 f  x dx  a .
0

IV) Let f be integrable on  a, b and F :  a, b   and

F 1  x   f  x  then prove that  f  x dx  F  b   F  a 
a

V) Which of the following functions are Riemann integrable

over  0,1 . Justify your answer.
a) The characteristic function of the set of rational number in
0,1 .
b) f  x   x sin yx for 0  x  1
f  0  3

VI) Prove that if f is  -integrable then f is also R-integrable is

the converse true? Justify your answer.

VII) Show that a monotone function defined on an interval  a, b  is

R-inegrable.

1 1 1
VIII) A function f ;  0,1   is defined as f  x   n 1
 n  x  n 1
3 3 3
where n 
f  0  0
1
show that f is R-integrable on  0,1 & calculate   f  x dx .
0

IX) f  x   x  x  x  1,3 where  x  denotes the greatest integer

not greater than x show that f is R-integrable on 1,3 .

X) A function f ;  a, b    is continuous on  a, b  f  x  0
b
x   a, b  and  f  x dx  0 show that f  x   0 x   a, b  .
a


 21

2
MEASURE ZERO SET
Unit Structure :

2.1 Introduction
2.2 Measure zero set
2.3 Definition
2.4 Lebesgue Theorem (only statement)
2.5 Characteristic function
2.6 FUBIN’s Theorem
2.7 Reviews
2.8 Unit End Exercises

2.1 INTRODUCTION

As we have seen, we cannot tell if a function is Riemann

integrable or not merely by counting its discontinuities one possible
alternative is to look at how much space the discontinuities take up.
Our question then becomes : (i) How can one tell rigorously, how
much space a set takes up. Is there a useful definition that will
concide with our intuitive understanding of volume or area?

At the same time we will develop a general measure theory

which serves as the basis of contemporary analysis.

In this introductory chapter we set for the some basic

concepts of measure theory.

2.2 MEASURE ZERO SET

Definition :
A subset ‘A’ of  n said to have measure ‘O’ if for every
 0 there is a cover U1 ,U 2 .... of A by closed rectangles such that

the total volume  v Ui   .
i 1

Theorem :
A function ‘f’ is Riemann integrable iff ‘f’ is discontinuous
on a set of Measure zero.
 22

A function is said to have a property of Continuous almost

everywhere if the set on which the property does not hold has
measure zero. Thus, the statement of the theorem is that ‘f’ is
Riemann integrable if and only if it is continuous atmost
everywhere.

Recall positive measure : A measure function u : M   0,   such

 
 
that V   ui    V  ui  .
 i 1  i 1

Example 1:
1) “Counting Measure” : Let X be any set and M  P  X  the set of
all subsets : If E  X is finite, then   E     E  if E  X is
infinite, then   E   
2) “Unit mass to x0 - Dirac delta function” : Let X be any set and
M  P  X  choose x0  X set.
  E   1if  x0  E
 0if  x0  E

Example 2:
Show that A has measure zero if and only if there is countable
collection of open rectangle V1 ,V2 ,.... such that A  Vi and
V  vi   .
Solution :
Suppose A has measure zero.
For  0, countable collection of closed rectangle V1 ,V2 ,....
 

such that A  Vi and
i 1
 V V   2 .
i 1
i

For each i , choose a rectangle ui such that ui  vi and

V  ui   2V  vi  .

    
Then A   vi   ui
i 1 i 1
and V  ui   V  ui    2V  vi 
i 1 i 1 i 1


 2 v  ui   2 
i 1 2

Note that : ui are open rectangles in  n conversely,

 23

Suppose for  0,  countable collection of open rectangles

 
u1 , u2 ,.... such that A   ui and
i 1
V  u  .
i 1
i

For each i, consider Vi  ui then Vi is a closed rectangle and

V  vi   V  ui  .

   
Then A   ui   vi and
i 1 i 1
V  vi   V  ui   .
i 1 i 1

A has measure zero.

Note : Therefore we can replace closed rectangle with open

rectangles in definition of measure zero sets.
Example 3:
Show that a set with finitely many points has measure zero.

Solution :
Let A  a1 ,...., am  be finite subset of  n .
Let  0, ai   ai1 , ai 2 ,....., ain  and
 1   n
1
1   n
1

Vi   ai1   i 1  , ai1   i 1    ...

 22  22  

 1   n
1
1   n
1

...   ain   i 1  , ain   i 1  

 22  22  


1
  n
n

Then V Vi     i 1   i 1
i 1  2  2
Clearly ai  Vi for 1  i  m
m m m
 
1 1
 A  Vi and
i 1
V Vi   
i 1 i 1 2 i 1
i 1 2
i 1
 
2

 By definition of measure of zero

 A has measure of zero.

Example 4:
If A  A1  A2  A3  .... and each Ai has measure zero, then
show that A has measure zero.

Solution :
Let  0 and A  A1  A2  .... with each Ai has measure zero.
 24

 Each Ai has measure zero for i  1, 2,....  a cover

ui1 ,U i 2 ,....,U in  of Ai



By closed rectangle such that V  u   2 ,i  1, 2,....
i 1
ii i

Then the collection of U ii is cover A

 

  V Vi    
i 1 i 1 2i

Thus A  A1  A2  An .... has measure zero.

Example 5:
Let A  n be a Rectangle show that A does not have
measure zero. But A has measure zero.

Proof :
Suppose A has measure zero.
 A is a rectangle in  n
V  A  0

Choose  0 such that  V  A  …………………….. (I)

 A has measure zero


 countable collection of open rectangle ui  such that A   ui
i 1

and V  u  .
i

 A is compact

This open cover has a finite subcover after renaming. We may

assume that u1 , u2 ,....uk  is subcover of the cover ui  .


 A   ui .
i 1

Let P be partition of A that contains all the vertices all ui ' si  1 to
k. Let S1 , S 2 ,...., S n denote the subrectangle of partitions.


V  A    V  S j    V  ui    V  ui  
n k

j 1 i 1 i 1
 25

which is a contradiction to (I)

 A does not have measure zero.

Note that A is a finite union of set of the form

B   a1 , b1    ai , bi   .....   an , bn  , . B can be covered by are closed
rectangle. B   a1 , b1   .....   ai , ai    .....   an , bn  .

Then V  B  depend on  and V  B   0 as   0 .

 B has measure zero
 Boundary of A  A  is finite union of measure zero.
A has measur5e zero.

Example 6:
Let A  n with A   . Show that A does not measure zero.

Solution :
Let A  n , with A  
Let x  A
 r  0 , such that B  x, r   A, But
B  x, r    y  A; y  x  r
 n

  y  A;  yi  xi  r 
 i 1 

 A does not have measure zero.

Example 7:
Show that the closed interval  a, b  does not have measure
zero.

Solution :
Suppose ui i 1 be a cover of  a, b  by open intervals.
  a, b  is compact this open cover has a finite subcover.

After renaming, we may assume u1 , u2 ,...., un  is the subcover of ui 

of  a, b  .
 26

We may assume each ui intersect  a, b  (otherwise replace ui with

ui   a, b  )

n
Let u   ui
i 1

If u is not connected then  a, b  is contained in one of connected

component of u.

  a, b   ui for some i
 a, b   u j   for i  j
Which is not possible
 u is connected
 u is an open interval say u   c, d  Then as  a, b   u   c, d 
 V  ui   d  c  b  a

In particular we cannot find an open cover of  a, b  with total length

ba
of the cover  .
2
 a, b  does not have measure zero.

Example 8:
If A   0,1 is the union of all open intervals  ai , bi  such that
each rational number in (0,1) is contained in some  ai , bi  . If

T    bi  ai   1 then show that the boundary of A does not have
i 1

measure zero.

Solution :
We first show that A   0,1 \ A
Note that A  A \ A
 A is open  A  A
Also Q   0,1  A
 Q   0,1  A
 0,1  A
But A  0,1  A   0,1
 A   0,1
A   0,1 \ A
 27

Let  1  T  0

If A has measure zero then since  0,  a cover of A with open

intervals such that sum of length of intervals  1  T
 A is closed and bounded
 A is compact
  finite subcover ui i 1 for A
n

    ui   1  T

 
  i  n;  ai , bi i 1 cover  0,1 and sum of lengths
Note that ui ;1


of these open intervals is less than 1  T  T  1 which is not possible

 
as  0,1   ui ;1  i  n; ai , bi i 1 A does not have measure zero.


2.3 DEFINITION

A subset ‘A’ of  n has content ‘O’ if for every  0 , there is

a finite cover u1 , u2 ,....., un  of A by closed rectangles such that
n

V  u  
i 1
i

Remark :
1) If A has content O, then A clearly has measure O.
2) Open rectangles can be used instead of closed rectangles in the
definition.

Example 9:
If A is compact and has measure zero then show that A has
content zero.

Solution :
Let A be a compact set in  n
Suppose that A has measure zero

 a cover u1 , u2 ,.... of A such that V  u   for every  0 .
i 1
i

 A is compact, a finite number u1 , u2 ,....., un of ui also covers A and

n 

V  u   V  u  
i 1
i
i 1
i

 A has content zero.

 28

Example 10 :
Give one example that a set A has measure zero but A does
not have content zero.

Solution :
Let A   0,1  Q
Then A is countable
 A has measure zero
Now to show that A does not have content zero.
Let  ai , bi  ;1  i  n be cover of A
 A   ai , bi   ....   an , bn 
 A   a1 , b1   ....   an , bn 
But A   0,1
n
    ai , bi    1
i 1

In particular, we cannot find a finite cover for A such that

n

   a , b   12
i 1
i i

 A does not have content zero.

Example 11:
Show that an unbounded set cannot have content zero.

Solution :
Let A  n be an unbounded set.
To show that A does not have content zero
Suppose A has content zero for  0,  finite cover of closed
k k
rectangles ui i 1 of A such that A   ui and V  u  .
k
i
i 1 i 1

Let ui   ai1 , bi1   ....   ain , bin 

Let ai  min a1i , a2i ,.....aki 
bi  max b1i , b2i ,.....bki 
then ui   a1 , b1   ....   an , bn 
 A   a1 , b1   ....   an , bn 
 A is bounded
Which is contradiction
 A does not have content zero.
 29

Example 12:
f : A   is non-negative and  f 0
A
where A is rectangle,

then show that  x  A; f  x   0 has measure zero.

Solution :

For n   , An  x  A; f  x   1 n 
Note that  x  A, f  x   0   x  A; F  x   0
 f is non-negative}
 n 
 
  x  A; f  x   1  A n
n 1 n 1

We have to show that An has measure zero

  f  0 and f  inf U  f , P   0 for  0,  a partition P such that
P
A A

U  f , P   n

Let S be a subrectangle in P
if S  An    M s  f   1 n
clearly S  P; S  An   covers An and
1  1
 n V  S    M  f V  S   M  f   n 
S P S P
s s

   f , P    n
  V  S  
S  An  
s  p

By definition An has content zero

 An has measure zero
 x  A, f  x   0 is countable union of measure zero set.
 x  A; f  x   0 has measure zero.

* Oscillation o  f , a  of ‘f’ at a
 for   0 , Let M  a, f ,    sup  f  x  ; x  A& x  a   
m  a, f ,    inf  f  x  ; x  A& x  a   
The oscillation o f , a of f at a defined by
o  f , a   lim  M  a, f ,    m  a, f ,   
 o
 30

This limit always exist since M  a, f ,    m  a, f ,   decreases as 

decreases.

Theorem :
Let A be a closed rectangle and let f : A   be a bounded
function such that O  f , x   for all x  A show that there is a
partition P of A with U  f , P   L  f , P  V  A  .

Proof :
Let x  A  U  f , x   lim  M  x, f ,    m  x, f ,    
 O

 a closed rectangle u x containing x in its interior such that

M u x  M u x  by definition of oscillation.
u x ; x  A is a cover of A
 A is compact
 This cover has a finite subcover say u x1 , u x 2 ,...., uxk 
k
 A   ux .
i 1
i

Let P be a partition for A such that there each subrectangle ‘S’ of P

is contained in some u x then M s  f   ms  f   for each
i
subrectangle ‘S’ in f
U  f , P   L  f , P     M s  f   ms  f  V  S 
S P

  V  S 
S P

V  A

2.4 LEBESGUE THEOREM (ONLY STATEMENT)

Let A be a closed rectangle and f : A   is bounded

function. Let B   x ; f is not continuous at x}. Then f is integrable
iff B is a set of measure zero

2.5 CHARACTERISTIC FUNCTION

Let C   n . The characteristics function  c of C is defined by

 c  x   1if x  C
 0if  x  C
 31

If C A where A is a closed rectangle and f : A   is bounded

then  f is defined as  f  c provided  f   c is integrable [i.e. if f
C C

and  c are integrable]

Theorem :
Let A be a closed rectangle and C A . Show that the
function  c : A   is integrable if and only if C has measure zero.

Proof :
To show that C : A   is integrable iff C has measure
zero.

By Lebesgue theorem, it is enough to show that C   x  A :  c is

discontinuous}

Let a  C    an open rectangle ‘u’ containing a such that u  C

  c  n   1n  U
  c is continuous at a.

Let a  Ext  c   Exterior of C

[By definition union of all open sets disjoints from C]
Ext (C) is an open set
 an open rectangle u containing such that U  Ext  c 
  c  n   0n  u
  c is continuous at a
If a  c then  c is continous at a ……………………. (I)

Let a  c  for any open rectangle U with a in its interior contains

a point y  C  & a point z   n c
  c  y   1&  c  z   0
  c is not continuous at a
c   x  A :  c is discontinuous at x }
 By Lebesgue Theorem.
 c is interrable if and only if c has measure zero.

Theorem :
Let A be a closed rectangle and C A
 32

If C is bounded set of measure zero and 

A
c exist then show that


A
c 0.

Proof :
C  A be a bounded set with measure zero.

Suppose   c exist   c is integral

A

To show that   c  0
A

Let P be a partition of A and S be a subrectangle in P.

 S does not have measure zero

 S C
 x  S but x  C
 c  x   0
 ms   c   0

This is true for any subrectangle S in P

 L   c , P    ms   c  V  C   0
This is true for any partition P
   c  sup  L   c , P  ; P is partition of}
A


A
c O

2.6 FUBINI’S THEOREM

Fubini’s Theorem reduces the computation of integrals over

closed rectangles in  n , n  1 to the computation of integrals over
closed intervals in  . Fubini’s Theorem is critically important as it
gives us a method to evaluate double integrals over rectangles
without having to use the definition of a double integral directly.

If f : A  R is a bounded function on a closed rectangle then

the least upper bound of all lower sum and the greatest lower bound
of all upper sums exist. They are called the lower integral and upper
integral of f and is denoted by L  F and U  F respectively.
A A
 33

Fubini’s Theorem
Statement : Let A   n and B   n be closed rectangles and let
f : A  B   be integrable for x  A , Let g x : B   be defined by
g x  y   F  x, y  and let
  x   L  g x  L  f  x, y dy
B B

u  x   U  g x  U  f  x, y dy
B B

 
Then  and  are integable on A and  f   L    L  f  x dy  dx
A B A A B 
 
 f   u  x dx    U  f  x, y dy  dx
A B A A B 

Proof :
Let PA be a partition of A and PB be a partition of B. Then
P   PA , PB  is a partition of A  B
Let S A be a subrectangle in PA and S B be a subrectangle in PB
Then by definition,
S  S A  S B is a subrectangle in P
L  f1 P    ms  f V  S 
S P

 m
S B PB
s A  sB  f V  S A  S B 
 
 m s A  sB  f V  S B  V  S A  …………………. (I)
S A PA  SB PB 

For x  S A , ms A  sB
 f   M s  gx  B

 For x  S A,
 m
S B PB
s A  sB V  S A   V  S B    msB  g x  V  S B 

 L  g x , PB   L  g x  L  x 
B

This is true for any x  A

 
L  f , P  m s A  sB  f V  S B  V  S A 
 SB PB
S A PA 
  ms A  L  x   V  S A 
S A PA

 L    x  , PA  ……………………………………… (II)
 34

 From (I) & (II)

L  f , P    L  x  , PA  ………………………………………… (III)

Now U  f , P    M S  f V  s 
S P

 
S A  PA
M S ASB  f V  S A  S B 
S B PB

 
    M  f V  S B V  S A  …………….. (IV)

S AS B
S A  PA S B PB

For x  S A , M S S  f   M S  g x 
A B B

 For x  S A ,
M
S B PB
S AS B  f V  S B   M
S B PB
SB  g x V S B 

 u  g x , PB   u  g x    x 
B

This is true for any x  A .

 

 S
S A PA   P
M S AS B  f V  S B V  S A 




B B

  M u  xV  S
S A PA
SA A 

 u  x  , PA  ……………………………………….. (V)

from (IV) & (V)

U  f , P   U u  x  , PA  ……………………………. (VI)
 By (III) & (VI)

L  f , P   L   x  , PA   u  L  x  , PA 
 u   x  , PA   U  f , P  ………………………… (VII)

Also
L  f , P   L   x  , PA   L   x  , PA   u   x , PA  …………. (VIII)

 f is integrable
sup  L  f , P   inf U  f , P    f
P P
AB

 sup L   x  , PA   inf u   x  , PA    f
PA PB
AB

   x is integrable
 35

 
 f     x     L  f  x, y dx ………………………. (IX)
 B 
AB A A 

Also by (VIII) & (IX)

sup  L  L  x  , PA   inf U u  x  , PA    f
PA PA
AB

 u  x is integrable.
 
  f   u  x dx   U  f  x, y dx
 B 
AB A A 

Hence Proved

Remark :
The Fubini’s theorem is a result which gives conditions under
which it is possible to compute a double integral using interated
integrals, As a consequence if allows the under integration to be
changed in iterated integrals.
 
 f    L  f  x, y dxdy
 B 
AB B 

 
  U  f  x, y dxdy
 A 
B 

These integrals are called iterated integrals.

Example 13:
Using Fubini’s theorem show that D12 f  D21 f if D12  f  and
D21  f  are continuous.

Solution :
 Let A  R and f : A   continuous
T.P.T D12 f  D21 f
Suppose D12 f  D21 f

 x0 , y0 in domain of f such that

 D12 f a D21 f a   0

without loss of generality,  D12 f a   D21 f a   0 or

 D12 f  D21 f a   0 ………………………………….. (I)
 36

   D12 f  D21 f  x, g   0
A

Let A   a, b c, d 
 By Fubini’s Theorem

d b

D 21 f  x, y    D 21 f  x, y dxdy
A c a
d

   D2 f b, y   D2 f  g , y dy

c

 f b, d   f b, c   f a, d   f a, c 

Similarly,
D 12 f  x, y   f b, d   f b, c   f a, d   f a, c 
A

  D21 f  x, y    D12 f  x, y 
A A

   D21 f  D12 f  x, y   0
A

Which is contradiction to (I)

D12 f  D21 f proved

Example 14:
Use Fubini’s Theorem to compute the following integrals.

1 1 x 2

I  
dy.dx
1)
0 0
1 x2  y 2

Solution :
1 1 x 2

I  
dy.dx
0 0
1 x2  y 2
1 1 x 2

  dx 
dy
0 0
1 x2  y2

 1 y 
1 1 x 2

  dx  tan 1 
 1  x 1  x 2  0
2
0


1


1
dx. .
0 1 x2 4
 37


1

4 0
dx

1 x2

 2
1
 log x  1  x 
4 0

 log  x  1
4  

1 1
  x 2 
I   dy  sin  dx
 2 
ii)
0 y

Solution :
C   x, y ; y  x  1, 0  y  1
By Fubini’s Theorem
1 1
  x 2 
I   sin  dxdy
 2 
0 y
1 x
  x 2 
   sin  dxdy
 2 
0 0
1
  x 2  x
  sin   y  dx
 2  0
0
1
  x 2 
  xsin   dx
 2 
0

 x2 x  1
Put  t,
2 t 0 
2

2 x
dx  dt
2
dt
xdx 

 
2 2
I  sin t   sin tdt  cos t 0 2
dt 1 1 

0
  0 

 0  1 
1 1
 
 38

2.7 REVIEWS

After reading this chapter you would be knowing.

 Definition of Measure zero set and content zero set.
 Oscillation O  f , a 
 Find set contain measure zero on content zero
 Statement of Lebesgue Theorem
 Definition of characteristic function & its properties.
 Fubini’s Theorem & its examples.

2.8 UNIT END EXERCISES

1. If B  A and A has measure zero then show that & has measure
zero.
2. Show that countable set has measure zero.
3. If A is non-empty open set, then show that A is not of measure
zero.
4. Give an example of a bounded set C if measure zero but C does
not have measure zero.
5. Show by an example that a set A has measure zero but A does
not have content zero.
6. Prove that  a1 , b1 .... an , bn  does not have content zero if ai  bi
for each i .
7. If C is a set of content zero show that the boundary of C has
content zero.
8. Give an example of a set A and a bounded subset C of A measure
zero such that  c does not exist.
A

9. If f & g are integrable, then show that f g is integrable.

10. Let U  0,1 be the union of all open intervals ai , bi  such that
each rational number in 0,1 is contained in some ai , bi  . Show
that if f  c except on a set of measure zero, then f is not
integrable on 0,1 .
11. If f :  a, b  a, b    is continuous; then show that
b b b b

 f  x, y  dxdy    f  x, y dydx
a x a x
 
2 2

 dy 
sin x
12. Use Fubini’s theorem, to compute dx
0 0
x y
 39

13. Let A  1,10,  2 and f : A  defined by

f  x, y   x sin y  ye x compute  f
A

14. Let f  x, y, z   z sin  x  y  and A  0,    ,   0,1

 2 2
computer  f.
A

