4 - 03-17-2021 - 11-27-41 - Mathematical Analysis

M. Sc.
(Mathematics) Semester-I
MATHEMATICAL ANALYSIS
Paper Code: 20MAT21C2
DIRECTORATE OF DISTANCE EDUCATION

MAHARSHI DAYANAND UNIVERSITY, ROHTAK
(A State University established under Haryana Act No. XXV of 1975)
NAAC 'A+’ Grade Accredited University
Author
Dr. Anju Panwar
Assistant Professor, Department of Mathematics
Maharshi Dayanand University, Rohtak
Copyright © 2021, Maharshi Dayanand University, ROHTAK

All Rights Reserved. No part of this publication may be reproduced or stored in a retrieval system or transmitted
in any form or by any means; electronic, mechanical, photocopying, recording or otherwise, without the written
permission of the copyright holder.
Maharshi Dayanand University
ROHTAK – 124 001
MASTER OF SCIENCE (MATHEMATICS)
First Semester
Paper code: 20MAT21C2
Mathematical Analysis
M. Marks = 100
Term End Examination = 80
Assignment = 20
Time = 3 hrs
Course Outcomes
Students would be able to:
CO1 Understand Riemann Stieltjes integral, its properties and rectifiable curves.
CO2 Learn about pointwise and uniform convergence of sequence and series of functions and various
tests for uniform convergence.
CO3 Find the stationary points and extreme values of implicit functions.
CO4 Be familiar with the chain rule, partial derivatives and concept of derivation in an open subset of
Rn.
Unit - I
Riemann-Stieltjes integral, its existence and properties, Integration and differentiation, The fundamental
theorem of calculus, Integration of vector-valued functions, Rectifiable curves.
Unit - II
Sequence and series of functions, Pointwise and uniform convergence, Cauchy criterion for uniform
convergence, Weirstrass’s M test, Abel’s and Dirichlet’s tests for uniform convergence, Uniform
convergence and continuity, Uniform convergence and differentiation, Weierstrass approximation
theorem.
Unit - III
Power series, its uniform convergence and uniqueness theorem, Abel’s theorem, Tauber’s
theorem.Functions of several variables, Linear Transformations, Euclidean space Rn, Derivatives in an
open subset of Rn, Chain Rule, Partial derivatives, Continuously Differentiable Mapping, Young’s and
Schwarz’s theorems.
Unit - IV
Taylor’s theorem. Higher order differentials, Explicit and implicit functions. Implicit function theorem,
Inverse function theorem. Change of variables, Extreme values of explicit functions, Stationary values
of implicit functions. Lagrange’s multipliers method. Jacobian and its properties.
Note : The question paper will consist of five sections. Each of the first four sections will contain two
questions from Unit I , II , III , IV respectively and the students shall be asked to attempt one
question from each section. Section five will contain eight to ten short answer type questions
without any internal choice covering the entire syllabus and shall be compulsory.
Books Recommended:
1. T. M. Apostol, Mathematical Analysis, Narosa Publishing House, New Delhi.
2. H.L. Royden, Real Analysis, Macmillan Pub. Co., Inc. 4th Edition, New York, 1993.
3. G. De Barra, Measure Theory and Integration, Wiley Eastern Limited, 1981.
4. R.R. Goldberg, Methods of Real Analysis, Oxford & IBH Pub. Co. Pvt. Ltd.
5. R. G. Bartle, The Elements of Real Analysis, Wiley International Edition.
6. S.C. Malik and Savita Arora, Mathematical Analysis, New Age International Limited, New
Delhi.
CONTENTS
UNIT I : THE RIEMANN-STIELTJES INTEGRAL
1.0 Introduction 1
1.1 Unit Objectives 2
1.2 Riemann-Stieltjes integral 2
1.3 Existence and properties 4
1.4 Integration and Differentiation 23
1.5 Fundamental Theorem of the Integral Calculus 25
1.6 Integration of Vector –Valued Functions 30
1.7 Rectifiable Curves 32
1.8 References 34
UNIT II: SEQUENCE AND SERIES OF FUNCTIONS
2.0 Introduction 35
2.2 Sequence and Series of Functions 35
2.3 Pointwise and Uniform Convergence 36
2.4 Cauchy Criterion for Uniform Convergence 45
2.5 Test for Uniform Convergence 46
2.6 Uniform Convergence and Continuity 58
2.7 Uniform Convergence and Integration 72
2.8 Uniform Convergence and Differentiation 77
2.9 Weierstrass’s Approximation Theorem 81
2.10 References 86
UNIT III : POWER SERIES AND FUNCTION OF SEVERAL VARIABLES
3.0 Introduction 87
3.2 Power Series 88
3.3 Function of several variables 104
3.4 References 129
UNIT IV: TAYLOR THEOREM
4.0 Introduction 130
4.2 Taylor Theorem 131
4.3 Explicit and Implicit Functions 134
4.4 Higher Order Differentials 138
4.5 Change of Variables 141
4.6 Extreme Values of Explicit Functions 148
4.7 Stationary Values of Implicit Functions 151
4.8 Lagrange Multipliers Method 153
4.9 Jacobian and its Properties 158
4.10 References 173
U N IT – I
THE RIEMANN-STIELTJES INTEGRAL
Structure
1.0 Introduction
1.1 Unit Objectives
1.2 Riemann-Stieltjes integral
1.2.1 Definitions and Notations
 Partition P, P* finer than P, Common refinement, Norm (or Mesh)
 Lower and Upper Riemann-Stieltjes Sums and Integrals
 Riemann-Stieltjes integral
1.3 Existence and properties
1.3.1 Characterization of upper and lower Stieltjes sums and upper and lower Stieltjes integrals
1.3.2 Integrability of continuous and monotonic functions along with properties of Riemann-
Stieltjes integrals
1.3.3 Riemann-Stieltjes integral as limit of sums.
1.4 Integration and Differentiation
1.5 Fundamental Theorem of the Integral Calculus
1.5.1 Theorem on Integration by parts or Partial Integration Formula
1.5.2 Mean Value Theorems for Riemann-Stieltjes Integrals.
 First Mean Value Theorem for Riemann-Stieltjes Integral
 Second Mean Value Theorem for Riemann-Stieltjes Integral
1.5.3 Change of variables
1.6 Integration of Vector –Valued Functions
1.6.1 Fundamental theorem of integral calculus for vector valued function
1.7 Rectifiable Curves
1.8 References
1.0 Introduction
In this unit, we will deal with the Riemann-Stieltjes integral and study its existence and properties. The
Riemann-Stieltjes integral is a generalization of Riemann integral named after Bernhard Riemann and
Thomas Joannes Stieltjes. The reason for introducing this concept is to get a more unified approach to
the theory of random variables. Fundamental Theorem of the Integral Calculus is discussed later on.
2 The Riemann-Stieltjes Integral
1.1 Unit Objectives
After going through this unit, one will be able to
 define Riemann-Stieltjes integral and characterize its properties.
 recognize Riemann-Stieltjes integral as a limit of sums.
 know about Fundamental Theorem of the Integral Calculus and Mean Value Theorems .
 understand the concept of Rectifiable Curves
1.2 Riemann-Stieltjes integral
We have already studied the Riemann integrals in our undergraduate level studies in Mathematics. Now
we consider a more general concept than that of Riemann. This concept is known as Riemann-Stieltjes
integral which involve two functions f and  . In what follows, we shall consider only real-valued
functions.
1.2.1 Definitions and Notations
Definition1. Let [ a, b] be a given interval. By a partition (or subdivision) P of [ a, b] , we mean a finite
set of points
P  {x0 , x1 ,........, xn }
such that
a  x0  x1  x2  ........xn 1  xn  b .
Definition 2. A partition P* of [ a, b] is said to be finer than P (or a refinement of P) if P*  P , that is ,

if every point of P is a point of P* i.e. P  P* .
Definition 3. The P1 and P2 be two partitions of an interval [ a, b] . Then a partition P* is called their
common refinement of P1 and P2 if P*  P1  P2 .
Definition 4. The length of the largest subinterval of a partition P  {x0 , x1 ,...., xn } of [ a, b] is called the
Norm (or Mesh) of P. We denote norm of P by P . Thus
P  max xi  max{xi  xi1 : i  1,2,...., n}
We notice that if P*  P , then P*  P . Thus refinement of a partition decreases its norm.
Definition 5. Lower and Upper Riemann-Stieltjes Sums and Integrals

Let f be a bounded real function defined on a closed interval [ a, b] . Corresponding to each
partition P of [ a, b] , we put
M i  lub f ( x ) ( xi 1  x  xi )
mi  glb f ( x ) ( xi 1  x  xi )
.
Mathematical Analysis 3
Let  be monotonically increasing function on [ a, b] . Then  is bounded on [ a, b] since  (a) and  (b)
are finite.
Corresponding to each partition P={x0,x1,…………,xn } of [a, b], we put
 i   ( xi )   ( xi 1 )
.
The monotonicity of  implies that  i  0 .
For any real valued bounded function f on [ a, b] , we take

n
L ( P , f ,  )   mi  i
i 1
n
U ( P , f ,  )   M i  i ,
i 1
where mi and M i are bounds of f defined above. The sums L( P, f , ) and U ( P, f , ) are respectively
called Lower Stieltjes sum and Upper Stieltjes sum corresponding to the partition P. We further define
b
 fd  lub L( P, f ,  )
a
 fd  glbU ( P, f , ) ,
a
b b
where lub and glb are taken over all possible partitions P of [a, b] . Then  fd and  fd are
a a
respectively called Lower integral and Upper integrals of f with respect to  .

b
If the lower and upper integrals are equal, then their common value, denoted by  fd , is called the
a
Riemann-Stieltjes integral of f with respect to  , over [a, b] and in that case we say that f is
integrable with respect to  , in the Riemann sense and we write f ( ) .
The functions f and  are known as the integrand and the integrator respectively.
In the special case, when  ( x)  x , the Riemann-Stieltjes integral reduces to Riemann-integral. In such
b b
a case, we write L(P, f ), U ( P, f ),  f ,  f and f  respectively in place of L( P, f , ), U ( P, f , ),

a a
b b
 fd ,  fd and f ( ) .

a a
Clearly, the numerical value of  fd depends only on f , , a and b and does not depend on the symbol
x. In fact x is a “dummy variable” and may be replaced by any other convenient symbol.
1.3 Existence and properties
1.3.1 In this section, we shall study characterization of upper and lower Stieltjes sums, and upper and
lower Stieltjes integrals.
The next theorem shows that for increasing function  , refinement of the partition increases the lower
sums and decreases the upper sums.
Theorem1. If P  is a refinement of P, f is bounded real valued function on a, b and  is
monotonically increasing function defined on a, b . Then,
 
L P  , f ,   LP, f ,  
U P , f ,    U P, f ,   .

Proof. Let P  {x 0 , x1 ,......... .......... , x n } be a partition of a, b . Further, let P  be a refinement of P

having one more point.
Let x  be such that point in the sub-interval [ xi 1 , xi ] that is
P   { x 0 , x1 ,......... .., x i 1 , x  , x i ,......... ......, x n } . Then, let
mi  g.l.b. of f in x i 1 , xi 
w1  g.l.b. of f in  xi 1 , x* 
w2  g.l.b of f in  x* , xi 
.
Obviously, mi  w1 ; mi  w2 .
Then,
L  P, f ,    m11  m2  2  ............  mi 1 i 1  mi  i  ............  mn  n
   
L P , f ,   m11  m2  2  ............  mi 1 i 1  w1  x    xi 1    w2   xi    x    
mi 1  i 1  .......... .........  m n  n
Thus,
       
L P  , f ,   L P , f ,    w1  x    x i 1   w 2   x i    x   m i   x i     x i 1 
 w  m  x     x   w
1 i

i 1 2   
 mi   xi    x 
Now, w1  mi  0 ; w2  mi  0
Also,  is monotonically increasing function and x i 1  x   x i . So,
 x      xi 1   0
 xi    x    0
 
 L P  , f ,   LP, f ,    0
 LP , f ,    LP, f ,  

Similarly, U P , f ,    U P, f ,   .

If P  contains more points, then similar process holds and so the result follows.
Theorem 2. For any two partitions P1 and P2 of a, b , let f be a bounded real valued function defined
on a, b and  is monotonically increasing function defined on a, b , then
LP1 , f ,    U P2 , f ,   .
Proof. Let P be the common refinement of P1 and P2 , that is, P  P1  P2 . Then, using Theorem 1, we
have
L ( P1 , f ,  )  L ( P, f ,  )  U ( P, f ,  )  U ( P2 , f ,  ) .
Remark 1. If m  f x   M . Then,
m b    a   LP, f ,    U P, f ,    M  b    a  .

Proof . By hypothesis
m  mi  M i  M
 m i  mi  i  M i  i  M  i
n n n n
 m  i  mi   i  M i   i  M   i
i 1 i 1 i 1 i 1
 m   b     a    L  P , f ,    U  P , f ,    M    b     a   .
Theorem 3. If f is bounded real valued function defined on a, b and  is monotonic function defined
on a, b . Then,

b b
 fd   fd
a a
.
Proof . Let Pa, b denotes the set of all partition of a, b . For P1 , P2  Pa, b , we know that
L  P1, f ,   U  P2 , f ,  1
This holds for each P1  Pa, b , keeping P2 fixed, it follows from 1 that U  P2 , f ,  is an upper
bound of the set {L  P1 , f ,   : P1  P  a, b} .
b
But least upper bound of this set is  f x d .
a
b
i.e,  f  x d  l.u.b.{L  P , f ,   : P  P  a, b}
1 1
a
Since, least upper bound  any upper bound

b
 f  x  d  U  P , f ,  
1  2
a
b
This holds for each P2  Pa, b . So, it follows from  2 that  f x d is a lower bound of the set
a
U P2 , f ,   : P2  Pa, b.

b
But greatest lower bound of this set is  fd .
a
b
i.e,  fd  g.l.b.U  P , f ,  : P  P  a, b
a
2 2
Since, any lower bound  greatest lower bound.


b b
So,  fd   fd .
a a
Example 1. Let  ( x)  x and define f on [0,1] by
1, x  Q
f ( x)  
0, x  Q
Then for every partition P of [0,1] , we have
mi  0 , M i  1 , because every subinterval [ xi 1 , xi ] contain both rational and irrational number.

Therefore
n
L ( P, f ,  )   mi xi
i 1
0
n
U ( P , f ,  )   M i xi
i 1
n
  ( xi  xi 1 )  xn  x0  1  0  1
i 1
Hence, in this case




fd   fd .
Theorem 4. Let  is monotonically increasing on a, b then f  ( ) iff for any  0 , there exists a
partition P of a, b such that
U P, f ,    LP, f ,    
Proof. The condition is necessary:
Let f be integrable on a, b i.e, f    on a, b ,

b b b
so that  fd   fd   fd 1
a a a
Let   0 be given.
b
Since  fd  sup{L  P, f ,  : P is a partition of [a,b]}
a
So, by definition of l.u.b., there exists a partition P1 of a, b such that

b
 b

L  P1 , f ,     fd    fd  (by (1))
2 a 2
a
b

 L  P1 , f ,     fd 
a
2
 b
 L  P1 , f ,      fd  2
2 a
Again,

b
Since  fd  inf{U  P, f ,   : P is a partition of [a,b]}.
a
By the definition of g.l.b., there exists a partition P2 of a, b such that


b  b

U  P2 , f ,     fd    fd 
a
2 a
2
b

 U  P2 , f ,     fd   3
a
2
Let P  P1  P2 be the common refinement of P1 and P2 , so that
U  P, f ,   U  P2 , f ,   4
And L  P1, f ,   L  P, f ,   5
Now, we have
b

U  P, f ,    U  P2 , f ,     fd  (by (3))
a
2
 
 LP1 , f ,     (by (2))
2 2
 L  P1, f ,   
 L  P, f ,    (by (5))
 U  P, f ,   L  P, f ,   
or U  P, f ,   L  P, f ,    .
The condition is sufficient:

Let   0 be any number. Let P be a partition of a, b such that
U  P, f ,    L  P, f ,     6
Since lower integral condition exceed the upper integral.

b b
So,  fd   fd .
a a

b b
  fd   fd  0 7
a a
Now, we know that


b b
L  P, f ,     fd   fd  U  P, f ,  
a a
b b
  fd   fd  U  P, f ,    L  P, f ,     8
a a
From (7) and (8) , we have


b b
0   fd   fd  
a a

b b
The non – negative number  fd   fd being less than every positive number  must be zero,
a a

b b
i.e,  fd   fd  0
a a

b b
  fd   fd .
a a
1.3.2 In this section, we shall discuss integrability of continuous and monotonic functions along with
properties of Riemann-Stieltjes integrals.
Theorem 1. If f is continuous on [ a, b] , then
(i) f  ( )
(ii) to every  0 there corresponds a   0 such that

n b
 f (ti )i   fd 

i 1 a
for every partition P of [ a, b] with P   and for all ti  [ xi 1 , xi ] .

Proof. (i) Let  0 and select   0 such that
 [ (b )   ( a )]  (1)
which is possible by monotonicity of  on [ a, b] . Also f is continuous on compact set [ a, b] .
Hence f is uniformly continuous on [ a, b] . Therefore there exists a   0 such that
f ( x)  f (t )   whenever x  t   for all x  [ a , b ] , t  [ a, b] (2).
Choose a partition P with P   . Then (2) implies
M i  mi   (i  1, 2,......, n )
Hence
n n
U ( P , f ,  )  L ( P , f ,  )   M i  i   mi  i
i 1 i 1
n n
  ( M i  mi )  i     i
i 1 i 1
n
   [ i ( xi )   ( xi 1 )]
i 1
  [ (b )   ( a )]

 .  ,

which is necessary and sufficient condition for f  ( ) .
(ii) We have
n
L ( P , f ,  )   f ( ti )   i  U ( P , f ,  )
i 1
and
b
L( P, f ,  )   fd  U ( P, f ,  )
a
Since f  ( ) , for each  0 there exists   0 such that for all partition P with P   , we have
U ( P , f ,  )  L ( P , f ,  ) 
Thus
n b
 f (ti )i   fd  U ( P, f , )  L( P, f , )

i 1 a

n b
Thus for continuous functions f , lim P  0  f (ti )  i exists and is equal to  fd .
i 1 a
Theorem 2. If f is monotonic on [ a, b] and if  is both monotonic and continuous on [ a, b] , then

f  ( ) .
Proof. Let  be a given positive number. For any positive integer n, choose a partition P of [ a, b] such
that
 (b)   (a)
 i  (i  1, 2,......, n) .
n
This is possible since  is continuous and monotonic on [ a, b] and so assumes every value between its
bounds  ( a ) and  (b) . It is sufficient to prove the result for monotonically increasing function f , the
proof for monotonically decreasing function being analogous. The bounds of f in [ xi 1 , xi ] are then
mi  f ( xi 1 ) , M i  f ( xi ) , i  1, 2,......, n .
Hence
n
U ( P, f ,  )  L ( P, f ,  )   ( M i  mi )  i
i 1
 (b )   ( a ) n

n
 (M
i 1
i  mi )
 (b )   ( a ) n

n
[ f (x )  f (x
i 1
i i 1 )]
 (b)   (a)
 [ f (b)  f (a)]
n
 for large n.
Hence f  ( ) .
Example 1. Let f be a function defined by
f ( x* )  1 and f ( x )  0 for x  x* , a  x*  b .
b
Suppose  is increasing on [ a, b] and is continuous at x * . Then f  ( ) over [ a, b] and  fd  0 .
a
Solution. Let P  {x0 , x1 ,......, xn } be a partition of [ a, b] and let x *  xi . Since  is continuous at x * ,
to each  0 there exists   0 such that

 ( x )   ( x* )  whenever x  x*  
2
Again since  is an increasing function,

 ( x )   ( x* )  for 0  x  x*  
2
and

 ( x* )   ( x )  for 0  x  x*  
2
Then for a partition P of [ a, b] ,
 i   ( xi )   ( xi 1 )
  ( xi )   ( x * )   ( x * )   ( xi 1 )
 
   .
2 2
n 0, ti  x*
Therefore  f (ti ) i  
 i , ti  x
*
i 1
that is,
n
 f (t )
i 1
i i  0 
Hence
n b
lim P 0  f (ti ) i   fd  0
i 1 a
and so f  ( ) and  fd  0 .

a
Theorem 3. Let f1  ( ) and f 2 ( ) on [ a, b] , then ( f1  f 2 )  ( ) and

b b b
( f
a
1  f 2 )d   f1d   f 2 d
a a
Proof. Let P  {a  x0 , x1 ,......, xn  b} be any partition of [ a, b] . Suppose further that M i' , mi' , M i" , mi" and
M i , mi are the bounds of f1 , f 2 and f1  f 2 respectively in the subinterval [ xi 1 , xi ] . If 1 ,  2  [ xi 1 , xi ] ,
then
[ f1 (2 )  f2 (2 )]  [ f1 (1 )  f2 (1 )]

 f1 (2 )  f1 (1 )  f2 (2 )  f2 (1 )
 ( M i'  mi' )  ( M i"  mi" )
Therefore, since this hold for all 1 ,  2  [ xi 1 , xi ] , we have
M i  mi  ( M i'  mi' )  ( M i"  mi" ) (1)
Since f1 , f 2  ( ) , there exists a partition P1 and P2 of [ a, b] such that
 
U ( P1 , f1 ,  )  L( P1 , f1 ,  )  2
 (2)
U ( P , f ,  )  L( P , f ,  )  
 2 2 2 2
2
These inequalities hold if P1 and P2 are replaced by their common refinement P.
Thus using (1), we have for f  f1  f 2 ,

n
U ( P , f ,  )  L ( P , f ,  )   ( M i  mi )  i
i 1
n n
  ( M i'  mi' )  i   ( M i"  mi" )  i
i 1 i 1
 
  (using (2))
2 2
 .
Hence f  f1  f 2  ( ) .
Further, we note that

mi'  mi"  mi  M i  M i'  M i"
Multiplying by  i and adding for i  1, 2,......, n, we get
L ( P, f1 ,  )  L ( P, f 2 ,  )  L ( P, f ,  )  U ( P, f ,  )
 U ( P, f1 ,  )  U ( P, f1 ,  )  U ( P, f 2 ,  ) (3)
Also
b

U ( P, f1 ,  )   f1d  (4)
a
2
b

U ( P, f 2 ,  )   f 2 d  (5)
a
2
Combining (3), (4) and (5), we have
b
 fd  U ( P, f , )  U ( P, f , )  U ( P, f ,  )
a
1 2
b b
 
  f1d   f 2 d  
a a
2 2
Since  is arbitrary positive number, we have

b b b

a
fd   f1d   f 2 d
a a
(6)
Proceeding with ( f1 ), ( f 2 ) in place of f1 and f 2 respectively, we have

b b b
 ( f )d   ( f1 )d   ( f2 )d

a a a
or
b b b

a
fd   f1d   f 2 d
a a
(7)
Now (6) and (7) yield

b b b b
 fd   ( f
a a
1  f 2 )d   f1d   f 2 d .
a a
Theorem 4.If f  ( ) and f  (  ) then f  (   ) and

b b b

a
fd (   )   fd   fd  .
a a
Proof. Since f  ( ) and f  (  ) , there exist partitions P1 and P2 such that

U ( P1 , f ,  )  L( P1 , f ,  ) 
2

U ( P2 , f ,  )  L( P2 , f ,  ) 
2
These inequalities hold if P1 and P2 are replaced by their common refinement P.
Also
 ( i   i )  [ ( xi )   ( xi 1 )]  [  ( xi )   ( xi 1 )]
Hence, if M i and mi are bounds of f in [ xi 1 , xi ] ,

n
U ( P, f , (   ))  L ( P, f , (   ))   ( M i  mi )  ( i   i )
i 1
n n
  ( M i  mi )  i   ( M i  mi )  i
i 1 i 1
 
   .
2 2
Hence f  (   ) .
Further
b

U ( P, f ,  )   fd 
a
2
b

U ( P, f ,  )   fd  
a
2
and
U (P, f ,   )   Mi i   Mi i
Also, then
b
 fd (   )  U ( P, f ,    )  U ( P, f , )  U ( P, f ,  )
a
b b
 
  fd    fd  
a
2 a 2
b b
  fd   fd   
a a
Since  is arbitrary positive number, therefore

b b b

a
fd (   )   fd   fd  .
a a
Replacing f by  f , this inequality is reversed and hence

b b b

a
fd (   )   fd   fd  .
a a
Theorem 5. If f  ( ) on [ a, b] , then f  ( ) on [ a, c ] and f  ( ) on [c, b] where c is a point of

[ a, b] and
b c b

a
fd   fd   fd .
a c
Proof. Since f  ( ) , there exists a partition P such that
U ( P , f ,  )  L ( P , f ,  )  ,  0 .
Let P* be a refinement of P such that P*  P  {c} . Then
L ( P, f ,  )  L ( P* , f ,  )  U ( P* , f ,  )  U ( P, f ,  )
which yields
U ( P* , f ,  )  L ( P*, f ,  )  U ( P, f ,  )  L ( P, f ,  ) (1)

Let P1 and P2 denote the sets of points of P* between [ a, c ] , [c, b] respectively. Then P1 and P2 are
partitions of [ a, c ] and [c, b] and P*  P1  P2 . Also
U ( P * , f ,  )  U ( P1 , f ,  )  U ( P2 , f ,  ) (2)
and
L ( P * , f ,  )  L ( P1 , f ,  )  L ( P2 , f ,  ) (3)
Then (1), (2) and (3) imply that

U ( P * , f ,  )  L ( P * , f ,  )  [U ( P1 , f ,  )  L ( P1 , f ,  )]  [U ( P2 , f ,  )  L ( P2 , f ,  )]

Since each of U ( P1 , f ,  )  L ( P1 , f ,  ) and U ( P2 , f ,  )  L ( P2 , f ,  ) is non-negative, it follows that
U ( P1 , f ,  )  L ( P1 , f ,  ) 
and
U ( P2 , f ,  )  L ( P2 , f ,  ) 
Hence f is integrable on [ a, c ] and [c, b] .
Taking inf for all partitions, the relation (2) yields

b c b
 fd   fd   fd

a a c
(4)
But since f is integrable on [ a, c ] and [c, b] , we have

b c b
a
fd   fd   fd
a c
(5)
The relation (3) similarly yields

b c b
 fd   fd   fd

a a c
(6)
Hence (5) and (6) imply that

b c b
 fd   fd   fd .

a a c
Theorem 6. If f  ( ) , then

b b
(i) cf  ( ) and  (cf )d  c  fd , for every constant c,

a a
(ii) If in addition f ( x)  K on [ a, b] , then

b
 fd
a
 K [ (b)   (a)] .
Proof.(i) Let f  ( ) and let g  cf . Then

n n
U ( P , g ,  )   M i'  i   cM i  i
i 1 i 1
n
 c  M i  i
i 1
 cU ( P , f ,  )
Similarly
L ( P , g ,  )  cL ( P , f ,  )
Since f  ( ) ,  a partition P such that for every  0 ,

U ( P , f ,  )  L ( P, f ,  ) 
c
Hence
U ( P , g ,  )  L ( P , g ,  )  c[U ( P, f ,  )  L ( P , f ,  )]

 c. .
c
Hence g  cf  ( ) .
b

Further, since U ( P, f ,  )   fd  ,
a
2
b
 gd  U ( P, g ,  )  cU ( P, f , )
a
b 
 c   fd  
a 2
Since  is arbitrary
b b
 gd  c  fd
a a
Replacing f by  f , we get
b b
 gd  c  fd
a a
b b
Hence  (cf )d  c  fd .

a a
(ii) If M and m are bounds of f  ( ) on [ a, b] , then it follows that

b
m[ (b)   (a)]   fd  M [ (b)   (a )] for b  a (1).
a
In fact, if a  b , then (1) is trivial. If b  a , then for any partition P, we have

n
m[ (b)   ( a )]   mi  i  L ( P, f ,  )
i 1
b
  fd
a
 U ( P, f ,  )   M i  i
 M [ (b )   ( a )]
which yields
b
m[ (b)   (a)]   fd  M [(b)  (a)] (2)
a
Since f ( x)  K for all x  ( a , b ) , we have
 K  f ( x)  K
so if m and M are the bounds of f in ( a , b ) ,

 K  m  f ( x )  M  K for all x  ( a , b ) .
If b  a , then  (b )   ( a )  0 and we have by (2)

b
 K [ (b)   (a)]  m[ (b)   (a)]   fd
a
 M [ (b )   ( a )]  K [ (b )   ( a )]
Hence
b
 fd
a
 K [ (b)   (a)] .
Theorem 7. Suppose f  ( ) on [ a, b] , m  f  M ,  is continuous on [ m , M ] and h ( x )   [ f ( x )] on

[ a, b] . Then h   ( ) on [ a, b] .
Proof. Let  0 . Since  is continuous on closed and bounded interval [ m , M ] , it is uniformly

continuous on [ m , M ] . Therefore there exists   0 such that   and
 (s)   (t )  if s  t   , s, t  [ m, M ] .
Since f  ( ) , there is a partition P  {x0 , x1 ,........, xn } of [ a, b] such that
U ( P , f ,  )  L ( P, f ,  )   2 (1).
Let M i , mi and M i* , mi* be the lub, glb of f ( x ) and  ( x) respectively in [ xi 1 , xi ] . Divide the number
1,2,……,n into two classes:
i  A if M i  mi  
and
i  B if M i  mi   .
For i  A , our choice of  implies that M i*  mi*  . Also, for i  B , M i*  mi*  2 k where k  lub  (t ) ,
t  [ m, M ] . Hence, using (1), we have
   i   (M i  mi )  i   2 (2)
iB iB
so that  
iB
i   . Then we have
U ( P, h,  )  L( P, h,  )   ( M i*  mi* )  i   ( M i*  mi* )  i
i A iB
 [ (b )   ( a )]  2 k
 [[ (b )   ( a )]  2k ] 
Since  was arbitrary,
U ( P, h,  )  L( P, h,  ) * , *  0 .
Hence h  f ( ) .
Theorem 8. If f  ( ) and g  ( ) on [ a, b] , then fg  ( ), f ( ) and

b b

a
fd   f d .
a
Proof. Let  be defined by  (t )  t 2 on [ a, b] . Then h( x )   [ f ( x )]  f 2  ( ) by Theorem7(in section

1.3.2). Also
1
fg  [( f  g ) 2  ( f  g ) 2 ] .
4
Since f , g   ( ) , f  g  ( ) , f  g  ( ) . Then ( f  g ) 2 and ( f  g ) 2  ( ) and so their
1
difference multiplied by also belong to  ( ) proving that fg  ( ) .
4
If we take  ( f )  t , again Theorem 7 implies that f ( ) . We choose c  1 so that
c  fd  0
Then
 fd  c  fd   cfd   f d
because cf  f .
1.3.3. Riemann-Stieltjes integral as limit of sums. In this section, we shall show that Riemann-
Stieltjes integral  fd can be considered as the limit of a sequence of sums in which M , m i i involved in
the definition of  fd are replaced by the values of f .
Definition 1. Let P  {a  x0 , x1 ,........, xn  b} be a partition of [ a, b] and let points t1 , t2 ,......, tn be such
that ti  [ xi 1 , xi ] . Then the sum
n
S ( P , f ,  )   f (t i )   i
i 1
is called a Riemann-Stieltjes sum of f with respect to  .

Definition 2. We say that
lim P 0 S (P, f , )  A
If for every  0 , there exists a   0 such that P   implies
S ( P, f , )  A  .
Theorem 1. If lim P 0 S ( P, f , ) exists, then f  ( ) and

b
lim P 0 S ( P, f ,  )   fd .
a
Proof. Suppose lim P 0 S ( P, f , ) exists and is equal to A. Then given  0 there exists a   0 such
that P   implies

S ( P, f ,  )  A 
2
or
 
A  S ( P, f ,  )  A  (1)
2 2
If we choose partition P satisfying P   and if we allow the points ti to range over [ xi 1 , xi ] , taking lub
and glb of the numbers S ( P , f ,  ) obtained in this way, the relation (1) gives
 
A  L ( P, f ,  )  U ( P, f ,  )  A 
2 2
and so
 
U ( P, f ,  )  L ( P, f ,  )   
2 2
Hence f  ( ) . Further
 
A  L( P, f ,  )   fd  U ( P, f ,  )  A 
2 2
which yields
 
A   fd  A 
2 2
or
 fd  A  lim P 0
S ( P, f ,  ) .
Theorem 2. If
(i) f is continuous, then
b
lim P 0 S ( P, f ,  )   fd
a
(ii) f  ( ) and  is continuous on [ a, b] , then

b
lim P 0 S ( P, f ,  )   fd .
a
Proof. Part (i) is already proved in Theorem 1(ii) of section 1.3.2 of this unit.
(ii) Let f  ( ) ,  be continuous and  0 . Then there exists a partition P* such that

U ( P* , f ,  )   fd  (1)
4
Now,  being uniformly continuous, there exists 1  0 such that for any partition P of [ a, b] with
P  1 , we have

 i   ( xi )   ( xi 1 )  for all i
4Mn
where n is the number of intervals into which P* divides [ a, b] . Consider the sum U ( P, f ,  ) . Those
intervals of P which contains a point of P* in their interior contribute no more than:
(n  1)  M 
(n  1) max  i .M   to U(P*,f,  ) (2)
4Mn 4
Then (1) and (2) yield

U ( P, f ,  )   fd  (3)
2
for all P with P  1 .
Similarly, we can show that there exists a  2  0 such that

L( P, f ,  )   fd  (4)
2
for all P with P   2 .
Taking   min{1 ,  2 } , it follows that (2) and (3) hold for every P such that P   .
Since
L( P, f ,  )  S ( P, f ,  )  U ( P, f ,  )
(3) and (4) yield

S ( P, f ,  )   fd 
2
and

S ( P, f ,  ) >  fd 
2
Hence

S ( P, f ,  )   fd 
2
for all P such that P   and so
lim P 0 S ( P, f ,  )   fd
This completes the proof of the theorem.

The Abel’s Transformation (Partial Summation Formula) for sequences reads as follows:
Let  an  and  bn  be sequences and let
An  a0  a1  ........  an ( A1  0) ,
then
q q 1
 anbn   An (bn  bn1 )  Aqbq  Ap 1bp

n p n p
1.4 Integration and Differentiation. In this section, we show that integration and differentiation are
inverse operations.
Definition 1. If f   on [a, b] , then the function F defined by
t
F (t )   f ( x)dx , t  [ a, b]
a
is called the “Integral Function” of the function f.

Theorem 1. If f   on [a, b] , then the integral function F of f is continuous on [a, b] .
Proof. We have
t
F (t )   f ( x)dx
a
Since f   , it is bounded and therefore there exists a number M such that for all x in [a, b] ,
f ( x)  M .
Let  be any positive number and c be any point of [a, b] . Then
c ch
F (c)   f ( x)dx , F (c  h)   f ( x)dx
a a
Therefore
ch c
F ( c  h )  F (c )  
a
f ( x)dx   f ( x)dx
a
ch
 
c
f ( x)dx
M h

 if h  .
M

Thus (c  h)  c    implies F (c  h)  F (c)  . Hence F is continuous at any point c  [ a, b]
M
and is so continuous in the interval [a, b] .
Theorem 2. If f is continuous on [a, b] , then the integral function F is differentiable and
F ( x0 )  f ( x0 ) , x  [ a, b] .
Proof. Let f be continuous at x0 in [a, b] . Then there exists   0 for every  0 such that
f (t )  f ( x0 )  (1)
whenever t  x0   . Let x0    s  x0  t  x0   and a  s  t  b , then

t
F (t )  F ( s ) 1
t  s s
 f ( x0 )  f ( x)dx  f ( x0 )
ts
t t
1 1
 
ts s
f ( x)dx 
t  s s
f ( x0 )dx
t t
1 1
 
ts s
[ f ( x)  f ( x0 )]dx 
t  s s
[ f ( x)  f ( x0 )]dx  ,
(using (1)).
Hence F ( x0 )  f ( x0 ) . This completes the proof of the theorem.
Definition 2. A derivable function F such that F ' is equal to a given function f in [a, b] is called
Primitive of f .
Thus the above theorem asserts that “Every continuous function f possesses a Primitive, viz the integral
t
function  f ( x)dx .”
a
Furthermore, the continuity of a function is not necessary for the existence of primitive. In other words,
the function possessing primitive is not necessary continuous. For example, consider the function f on
[0,1] defined by
 1 1
2 x sin  cos , x  0
f ( x)   x x
0, x  0
It has primitive
 2 1
 x sin , x  0
F ( x)   x
0, x  0
Clearly F ' ( x)  f ( x) but f ( x ) is not continuous at x  0 , i.e., f is not continuous in [0,1] .
1.5 Fundamental Theorem of the Integral Calculus

Theorem 1 (Fundamental Theorem of the Integral Calculus). If f ∈ R on [a, b] and if there is a
differential function F on [a, b] such that F   f , then
b
 f ( x)dx  F (b)  F (a) .

a
Proof. Let P be a partition of [a, b] and choose ti , (i  1, 2,......, n) such that xi 1  ti  xi . Then, by
Lagrange’s Mean Value Theorem, we have
F ( xi )  F ( xi 1 )  ( xi  xi 1 ) F (ti )  ( xi  xi 1 ) f (ti ) (Since F   f ).
Further
n
F (b)  F (a )   [ F ( xi )  F ( xi 1 )]
i 1
n
  f (ti )( xi  xi 1 )
i 1
n
  f (ti ) xi
i 1
b
and the last sum tends to  f ( x)dx as
a
P  0 , by theorem 1 of section 1.3.3, taking  ( x )  x . Hence
b
 f ( x)dx  F (b)  F (a) .

a

The next theorem tells us that the symbols d ( x) can be replaced by  ( x)dx in the Riemann-Stieltjes
b
integral  f ( x)d ( x) . This is the situation in which Riemann-Stieltjes integral reduces to Riemann
a
integral.
Theorem 2. If f   and     on [a, b] , then f  ( ) and
b b
 fd   f ( x) ( x)dx

a a
Proof. Since f   ,     , it follows that their product f     . Let  0 be given. Choose M such
that f  M . Since f     and     , using Theorem 2(ii) of section 1.3.3 for integrator as x, we
have
 f (t ) (t )x   f   
i i i (1).
if P  1 and xi 1  ti  xi and
  (t )x     
i i (2).
if P   2 and xi 1  ti  xi . Letting ti vary in (2), we have
  (s )x     
i i (3).
if P   2 and xi 1  si  xi . From (2) and (3) it follows that
  (t )x            (s )x

i i i i
   (t )x        (s )x    

i i i i
   2 
or
  (t )   (s ) x
i i i  2 (4).
if P   2 and xi 1  ti  xi , xi 1  si  xi .
Now choose a partition P so that P    min{1 ,  2 } and choose ti  [ xi 1 , xi ] . By Mean Value

Theorem,
 i   ( xi )   ( xi 1 )   ( si )( xi  xi 1 )
  ( si )xi
Then, we have
 f (t )   f (t ) (t )x   f (t )[ (s )   (t )]x

i i i i i i i i i (5).
Thus, by (1) and (4), it follows that
 f (t )   f     f (t ) (t )x   f     f (t )[ (s )   (t )]x

i i i i i i i i i
 2  M  (1  2 M )
Hence
b
lim P 0  f (ti )xi   f ( x) ( x)dx
a
or
b b
 fd   f ( x) ( x)dx .

a a
2 2
 x dx  [ x]dx
2 2 2
Example 1. Evaluate (i) , (ii) .
0 0
Solution. We know that

b b
 fd   f ( x) ( x)dx

a a
Therefore
2 2 2
 x dx   x (2 x)dx   2 x dx
2 2 2 3
0 0 0
2
x4
2 8 .
4 0
and
2 2
 [ x]dx   [ x]2 xdx

2
0 0
1 2
  [ x]2 xdx   [ x]2 xdx
0 1
2 2
x2
 0   2 xdx  0  2
1
2 1
=0+3 = 3.
We now establish a connection between the integrand and the integrator in a Riemann-Stieltjes integral.
We shall show that existence of  fd implies the existence of   df .
We recall that Abel’s transformation (Partial Summation Formula) for sequences reads as follows:
“Let  an  and  bn  be two sequences and let An  a0  a1  ......  an ( A1  0) . Then
q q 1
 anbn   An (bn  bn1 )  Aqbq  Ap 1bp

n p n p
(*).”
1.5.1 Theorem (Integration by parts). If f  ( ) on [a, b] , then   ( f ) on [a, b] and
 f ( x)d ( x)  f (b) (b)  f (a) (a)    ( x)df ( x)

(Due to analogy with (*), the above expression is also known as Partial Integration Formula).
Proof. Let P  {a  x0 , x1 ,......, xn  b} be a partition of [a, b] . Choose t1 , t2 ,......, tn such that xi 1  ti  xi
and take t0  a , tn 1  b . Suppose Q is the partition {t1 , t2 ,......, tn 1} of [a, b] . By
partial summation, we have
n n 1
S ( P, f ,  )   f (ti )[ ( xi )   ( xi 1 )]  f (b) (b)  f (a ) ( a )    ( xi 1 )[ f (ti )  f (ti 1 )]
i 1 i 1
 f (b) (b)  f (a ) (a )  S (Q,  , f )
since ti 1  xi 1  ti . If P  0 , Q  0 , then
S ( P, f ,  )   fd and S (Q,  , f )    df .
Hence
 fd  f (b) (b)  f (a) (a)    df .

1.5.2 Mean Value Theorems for Riemann-Stieltjes Integrals. In this section, we establish Mean
Value Theorems which are used to get estimate value of an integral rather than its exact value.
Theorem 1.5.2(a). (First Mean Value Theorem for Riemann-Stieltjes Integral). If f is continuous
and real valued and  be is monotonically increasing on [a, b] , then there exists a point x in [a, b] such
that
b
 fd  f ( x)[ (b)   (a)] .

a
Proof. If  (a )   (b) , the theorem holds trivially, both sides being 0 in that case (  become constant
and so d   0 ). Hence we assume that  ( a )   (b) . Let
M  lub f ( x ) , m  glb f ( x ) . a  x  b
Then
m  f ( x)  M
or
m[ (b)   (a)]   fd M [ (b)   (a)]
Hence there exists some c satisfying m  c  M such that

b
 fd  c[ (b)   (a)]

a
Since f is continuous, there is a point x  [ a, b] such that f ( x )  c and so we have

b
 f ( x)d ( x)  f ( x)[ (b)   (a)]

a

Theorem 1.5.2(b) (Second Mean Value Theorem for Riemann-Stieltjes Integral). Let f be
monotonic and  be real and continuous. Then there is a point x  [ a, b] such that
b
 fd  f (a)[ ( x)   (a)]  f (b)[ (b)   ( x)]

a
Proof. By Partial Integration Formula, we have

b b
 fd  f (b) (b)  f (a) (a)    df

a a
The use of First Mean Value Theorem for Riemann-Stieltjes Integral yields that there is x in [a, b] such
that
b
  df   ( x)[ f (b)  f (a)]

a
Hence, for some x  [ a, b] , we have

b
 fd  f (b) (b)  f (a) (a)   ( x)[ f (b)  f (a)]

a
 f ( a )[ ( x )   ( a )]  f (b)[ (b)   ( x )]

which proves the theorem.
1.5.3 We discuss now change of variable. In this direction we prove the following result.
Theorem 1. Let f and  be continuous on [a, b] . If  is strictly increasing on [ ,  ] , where
a   ( ) , b      , then
b 
 f ( x)dx   f ( ( y))d ( y)
a
(this corresponds to change of variable in  f ( x)dx by taking x   ( y ) ).

a
Proof. Since  is strictly monotonically increasing, it is invertible and so
   1 (a ) ,    1 (b) .
Let P  {a  x0 , x1 ,......, xn  b} be any partition of [a, b] and Q  {  y0 , y1 ,......, yn   } be the
corresponding partition of [ ,  ] , where yi   1 ( xi ) . Then
xi  xi  xi 1
  ( yi )   ( yi 1 )
 i .
Let for any ci  xi , di  yi , where ci   (di ) . Putting g ( y )  f [ ( y )] , we have

n
S ( P, f )   f (ci )xi (1)
i 1
  f ( (di ))i
i
  g (di )i
i
 S (Q , g ,  )
b
Continuity of f implies that S ( P, f )   f ( x)dx as P  0 and continuity of g implies that
a

S (Q, g ,  )   g ( y )d as Q  0 .

Since uniform continuity of  on [a, b] implies that Q  0 as P  0 . Hence letting P  0 in (1),

we have
b  

a
f ( x)dx   g ( y )d   f ( ( y ))d ( y )
 

1.6 Integration of Vector –Valued Functions. Let f1 , f 2 ,......, f k be real valued functions defined on
[a, b] and let f  ( f1 , f 2 ,......, f k ) be the corresponding mapping of [a, b] into R k .
Let  be monotonically increasing function on [a, b] . If fi ( ) for i  1, 2,......, k , we say that
f  ( ) and then the integral of f is defined as
b b b b
 fd  ( f d ,  f d ,........,  f d ) .
a a
1
a
2
a
k
b b
 fd  f d .
k
Thus is the point in R whose ith coordinate is i
a a
It can be shown that if f  ( ) , g  ( ) ,

then
b b b
(i)  ( f  g )d   fd   gd
a a a
b c b
(ii)  fd   fd   fd ,
a a c
a c b.
(iii) if f (1 ) , f ( 2 ) , then f (1   2 )

and
b b b

a
fd (1   2 )   fd1   fd 2
a a
To prove these results, we have to apply earlier results to each coordinate of f . Also, fundamental
theorem of integral calculus holds for vector valued function f . We have
Theorem 1. If f and F map [a, b] into  k , if f  ( ) if F   f , then

b
 f (t )dt  F (b)  F (a)

a
Theorem 2. If f maps [a, b] into R k and if f  R ( ) for some monotonically increasing function  on
[a, b] , then f  R( ) and
b b

a
fd   f d .
a
Proof. Let
f  ( f1 ,......, f k ) .
Then
f  ( f12  ......  f k2 )1/ 2
Since each fi  R( ) , the function f i 2  R ( ) and so their sum f12  ......  f k2  R ( ) . Since x 2 is a
continuous function of x, the square root function is continuous on [0, M ] for every real M. Therefore
f  R( ) .
Now, let y  ( y1 , y2 ,.... yk ) , where yi   fi d , then
y   fd
and
y   yi2   yi  fi d
2
  ( yi fi )d
But, by Schwarz inequality
 y f (t ) 
i i y f (t ) , (a  t  b)
Then

2
y  y f d (1)
If y  0 , then the result follows. If y  0 , then divide (1) by y and get
y   f d
b b
or 
a
fd   f d .
a
1.7 Rectifiable Curves. The aim of this section is to consider application of results studied in this
chapter to geometry.
Definition 1. A continuous mapping  of an interval [a, b] into R k is called a curve in R k .
If  :[a, b]  R k is continuous and one-to-one, then it is called an arc.
If for a curve  :[a, b]  R k ,
 ( a )   (b )
but
 (t1 )   (t2 )
for every other pair of distinct points t1 , t2 in [a, b] , then the curve  is called a simple closed curve.
Definition 2. Let f :[a, b]  R k be a map. If P  {x0 , x1 ,...., xn } is a partition of [a, b] , then

n
V ( f , a, b)  lub  f ( xi )  f ( xi 1 ) ,
i 1
where the lub is taken over all possible partitions of [a, b] , is called total variation of f on [a, b] . The
function f is said to be of bounded variation on [a, b] if V ( f , a, b)   .
Definition 3. A curve  :[a, b]  R k is called rectifiable if  is of bounded variation. The length of a

rectifiable curve  is defined as total variation of  , i.e., V ( , a, b) . Thus length of rectifiable curve
n
  lub   ( xi )   ( xi 1 ) for the partition (a  x0  x1  ....  xn  b) .
i 1
The ith term  ( xi )   ( xi 1 ) in this sum is the distance in R k between the points  ( xi 1 ) and  ( xi ) .
n
Further   (x )   (x
i 1
i i 1 ) is the length of a polygon whose vertices are at the points
 ( x0 ),  ( x1 ),....,  ( xn ) . As the norm of our partition tends to zero, then those polygons approach the
range of  more and more closely.
Theorem 1. Let  be a curve in R k . If   is continuous on [a, b] , then  is rectifiable and has length
b
  (t ) dt .
a
Proof. It is sufficient to show that     V ( , a, b) . So, let {x ,...., x } be a partition of [a, b] .

0 n
Using Fundamental Theorem of Calculus for vector valued function, we have

n n xi
  (x )   (x
i 1
i i 1 )    (t )dt
i 1 xi 1
n xi
   (t ) dt
i 1 xi 1
b
   (t ) dt
a
Thus
V ( , a, b)     (1). .
To prove the reverse inequality, let  be a positive number. Since   is uniformly continuous on [a, b] ,
there exists   0 such that
 ( s)   (t )  , if s  t   .
If mesh (norm) of the partition P is less than  and xi 1  t  xi , then we have
 (t )   ( xi )   ,
so that
xi

xi 1
 (t ) dt   xi   ( xi ) xi
xi
  [ (t )   ( x )   (t )]dt

xi 1
i
xi xi
 
xi 1
 (t )dt   [ ( x )   (t )]dt
xi 1
i
  ( xi )   ( xi 1 )   xi
Adding these inequalities for i  1, 2,...., n , we get
b n
  (t ) dt    ( xi )   ( xi 1 )  2  (b  a)
a i 1
 V ( , a, b)  2  (b  a )
Since  is arbitrary, it follows that
b
  (t ) dt  V ( , a, b)
a
(2).
Combining (1) and (2), we have

b
  (t ) dt  V ( , a, b)
a
Hence the length of  is   (t ) dt .

a
1.8 References
1. Walter Rudin, Principles of Mathematical Analysis (3rd edition) McGraw-Hill, Kogakusha, 1976,
International Student Edition.
2. T. M. Apostol, Mathematical Analysis, Narosa Publishing House, New Delhi, 1974.
5. R.R. Goldberg, Methods of Real Analysis, Oxford & IBH Pub. Co. Pvt. Ltd, 1976.
6. R. G. Bartle, The Elements of Real Analysis, Wiley International Edition, 2011.
7. S.C. Malik and Savita Arora, Mathematical Analysis, New Age International Limited, New Delhi,
2012.
U N IT – II
SEQUENCE AND SERIES OF FUNCTIONS
Structure
2.0 Introduction
2.1 Unit Objectives
2.2 Sequence and Series of Functions
2.3 Pointwise and Uniform Convergence
2.4 Cauchy Criterion for Uniform Convergence
2.5 Test for Uniform Convergence
2.6 Uniform Convergence and Continuity
2.7 Uniform Convergence and Integration
2.8 Uniform Convergence and Differentiation
2.9 Weierstrass’s Approximation Theorem
2.10 References
2.0 Introduction
In this unit, we will consider sequence and series of functions whose terms depend on a variable.
Uniform convergence of sequence or series is a concept of great importance in its domain. With the help
of tests for uniform convergence, we will naturally inquire how we can determine whether the given
sequence or series does or does not converge uniformly in a given interval. The Weierstrass
approximation theorem describes that every continuous function can be “uniformly approximated” by
polynomials to within any degree of accuracy.
2.1 Unit Objectives
 learn about pointwise and uniform convergence of sequence and series of functions
 examine uniform convergence through various tests for uniform convergence.
 study uniform convergence and continuity.
 understand importance of Weierstrass approximation theorem.
2.2 Sequence and Series of Functions
Let fn be a real valued function defined on an interval I (or on a subset D of R ) and for each n  N , then
<f1, f2,……… ,fn,……..> is called a sequence of real valued functions on I. It is denoted by {fn} or <fn>.
36 Sequence & Series of Functions
If <fn> is a sequence of real valued functions on an interval I, then f1+ f2+……… +fn+……..is called a

series of real valued functions defined on I. This series is denoted by f
n 1
n or simply f n . That is, we
shall consider sequences whose terms are functions rather than real numbers. These sequences are useful
in obtaining approximations to a given function.
2.3 Pointwise and Uniform Convergence of Sequences of Functions
We shall study two different notations of convergence for a sequence of functions: Pointwise
convergence and uniform convergence.
Definition 1. Let A  R and suppose that for each n  N there is a function f n : A  R . Then f n  is
called a sequence of functions on A. For each x  A , this sequence gives rise to a sequence of real
numbers, namely the sequence fn  x  .
Definition 2. Let A  R and let f n  be a sequence of functions on A. Let A 0  A and suppose

f : A0  R . Then the sequence f n  is said to converge on A 0 to f if for each x  A 0 , the sequence
fn  x  converges to f(x) in R.
In such a case f is called the limit function on A 0 of the sequence f n  .
When such a function f exists, we say that the sequence f n  is convergent on A 0 or that f n 
converges pointwise on A 0 to f and we write f  x   limn fn  x  , x  A 0 .
Similarly, iff  x  converges for every x  A

n 0 , and if f  x    
f  x  , x  A 0 . The function f is
n 1 n
called the sum of the series  f . n
The question arises: If each function of a sequence fn has certain property, such as continuity,
differentiability or integrability, then to what extent is this property transferred to the limit function? For
example, if each function f n is continuous at a point x 0 , is the limit function f also continuous at x 0 ? In
general, it is not true. Thus, pointwise convergence is not so strong concept which transfers above
mentioned property to the limit function. Therefore some stronger methods of convergence are needed.
One of these methods is the notion of uniform convergence:
We know that f n is continuous at x 0 if limxx0 fn  x   fn  x0  . On the other hand,
f is continuous at x 0 if limxx0 f  x   f  x0  (1)
But (1) can be written as
limxx0 limn fn  x   limn limxx0 fn  x  (2)

Thus our question of continuity reduces to “can we interchange the limit symbols in (2)?” or “Is the
order in which limit processes are carried out immaterial “. The following examples show that the limit
symbols cannot in general be interchanged.
Example 1. A sequence of continuous functions whose limit function is discontinuous:
Let
x 2n
fn  x   , x  R, n = 1, 2, ………
1  x 2n
We note that
0 if | x |  1
1

limn  f n  x   f  x    if | x |  1.
2
1 if | x |  1
Each f n is continuous on R but the limit function f is discontinuous at x = 1 and x = -1.
Example 2. A double sequence in which limit process cannot be interchanged:
For m =1, 2,….,
n = 1, 2,3,..., let us consider the double sequence
m
Smn  .
mn
For every fixed n, we have
limm Smn  1
and so
lim n  lim m Smn  1
On the other hand, for every fixed m, we have
1
lim n  Smn  lim n  0
n
1
m
and so
lim m lim n  Smn  0
Hence lim n  lim m Smn  lim m lim n  Smn

Example 3. A sequence of functions for which limit of the integral is not equal to integral of the
limit: Let
f n  x   n 2 x 1  x  , x  R, n  1, 2,......
n
If 0  x  1, then
f  x   limn fn  x   0
and so
1
 f (x) dx  0.
0
1 1
f  x 1  x 
n
But n (x) dx  n 2
dx
0 0
n2 n2
 
n 1 n  2
n2

(n  1)  n  2 
.
1
and so lim n   f n (x) dx  1
0
1 1
Hence lim n   f n (x) dx   (lim n  f n (x)) dx .
0 0
Example 4. A sequence of differentiable functions { f n } with limit 0 for which { f n ' } diverges.
Let
sin nx
fn  x   if x  R, n = 1, 2,
n
Then limn fn  x   0 for all x.
But f n'  x   n cos nx
and so limn fn'  x  does not exist for any x.

Definition 3. A sequence of functions f n  is said to converge uniformly to a function f on a set E if for

every ε> 0 there exists an integer N (depending only on ε) such that n > N implies
fn  x   f  x    for all x  E (*).
Geometrical Interpretation of uniform convergence:
If each term of the sequence f n  is real-valued, then the expression (*) can be written as
f  x     fn  x   f  x    for all n > N and for all x  E .
This shows that the entire graph of f n lies between a “band” of height 2ε situated symmetrically about
the graph of f.
Definition 4. A series f  x  is said to converge uniformly on E if the sequence S  of partial sums

n n
n
defined by Sn  x    f i  x  converges uniformly on E.
i 1
Theorem 1. Every uniformly convergent sequence is pointwise convergent but not conversely.
Proof. Let { f n } be a sequence of functions which converges uniformly to f on E.
 For given   0 , there exists a positive integer N (depending only on  ) such that
f n ( x )  f ( x )  for all n  N ………….(1)
Since (1) is true for all x  E .
f n x   f x    for all n  N
is true for every x E ,
Hence f n converges pointwise to f on E.
The converse is not true which is shown by following example.

Example 5. Consider the sequence  f n  defined by
1
f n x   ,0  x  1
nx  1
1
Then, f ( x)  lim f n x   lim 0
n  n  nx  1
Hence,  f n  converges pointwise to 0 for all 0  x  1 .
Let   0 be given. Then for convergence, we have
f n  x  f  x   , n  n0
1
or  0   , n  n0 .
nx  1
1
or  .
nx  1
1
or 
nx
1
or nx  .

1
or n .
x
1
If n 0 is taken as integer greater than , then
x
f n x   f x    for all n  n0 .
Since n 0 depends both on  & x in (0,1) , so f n does not converge uniformly on (0,1).
1
Example 6. Consider the sequence Sn  defined by Sn  x   in any interval [a, b], a > 0. Then
xn
1
S  x   lim Sn  x   lim 0
n  n  x  n
For the convergence, we must have
Sn  x   S  x    , n  n0
(1)
1
or 0  , n  n0
xn
1
or 
xn
1
or xn 

1
or n x

1 1
If we select n 0 as integer next higher to , then (1) is satisfied for m(integer) greater than which
 
does not depend on x  [a, b] . Hence the sequence Sn  is uniformly convergent to S(x) in [a, b].
Example 7. Consider the sequence f n  defined by
x
fn  x   , x0
1  nx
Then
x
f  x   lim  0 for all x  0 .
n  1  nx
Then f n  converges pointwise to 0 for all x  0 . Let   0 , then for convergence we must have
f n  x   f  x   , n  n 0
x
or  0  , n  n 0
1  nx
x

1  nx
x    nx
nx  x  
x 
n
x
x 1
n 
x 
1
If n 0 is taken as integer greater than , then

f n  x   f  x   , for all n  n 0 and for all x   0, 
Hence f n  converges uniformly to f on  0, .

fn  x   xn , 0  x  1
Then
0 if 0  x  1
f n  x   lim x n  
n 
1 if x  1
Let   0 be given. Then for convergence, we must have
f n  x   f  x   , n  n 0
or xn  
n
1 1
or   
x 
1
log
or n .
1
log
x
1
log
Thus we should take n 0 to be an integer next higher to  . If we take x = 1, then m does not exist.
1
log
x
Thus the sequence in question is not uniformly convergent to f in the interval which contains 1.
Definition 5 (Point of non–uniform convergence). A point which is such as the sequence is non –
uniformly convergent in any interval containing that point is called a point of non–uniform convergence.
In the following example x = 0 is a point of non–uniform convergence.
nx
Example 9. Consider the sequence f n  defined by f n  x   , 0xa.
1  n2 x2
Then if x = 0, then fn  x   0
and so f  x   limn fn  x   0.
If x  0 , then
nx
f  x   limn  f n  x   limn   0.
1 n2x2
Thus f is continuous at x = 0. For convergence, we must have
f n  x   f  x   , n  n 0 .
nx
or  .
1 n2x2
nx
or 1 n2x2   0.

1 1 1
or nx    4.
2 2  2
Thus we can find an upper bound for n in any interval 0  a  x  b , but the upper bound is infinite if the
interval includes 0. Hence the given sequence is non-uniformly convergent in any interval which
includes the origin. So 0 is the point of non-uniform convergence for this sequence.
f n  x   tan 1 nx, 0  x  a.
Then

 if x  0
f  x   lim f n  x    2 .
n 
0 if x  0
Thus the function is discontinuous at x = 0.

For convergence, we must have for   0 ,
f n  x   f  x   , n  n 0

or  tan 1 nx  
2
or cot 1 nx  
1
or nx 
tan 
1 1
or n  
tan   x 
Thus no upper bound can be found for the function on the right if 0 is an end point of the interval. Hence
the convergence is non-uniform in any interval which includes 0. So, here 0 is the point of non-uniform
convergence.
Definition 6. A sequence fn  is said to be uniformly bounded on E if there exists a constant
M > 0 such that f n  x   M for all x in E and all n. The number M is called a uniform bound for fn  .
For example, the sequence f n  defined by fn  x   sin nx, x  R is uniformly bounded. Infact,
f n  x   sin nx  1 for all x  R and for all n  N .
If each individual function is bounded and if f n  f uniformly on E, then it can be shown that fn  is
uniformly bounded on E. This result generally helps us to conclude that a sequence is not uniformly
convergent.
2.4 Cauchy Criterion for Uniform Convergence
We now find necessary and sufficient condition for uniform convergence of a sequence of
functions.
Theorem 1 (Cauchy criterion for uniform convergence). The sequence of functions fn  , defined on
E, converges uniformly if and only if for every   0 there exists an integer N such that
m  N, n  N, x  E imply f n  x   f m  x   .
Proof. Suppose first that f n  converges uniformly on E to f. Then to each   0 there exists an integer
N such that n > N implies

f n  x   f  x   , for all x  E
2
Similarly for m > N implies

fm  x   f  x   , for all x  E
2
Hence, for n > N, m > N, we have
fn  x   fm  x   fn  x   f  x   f  x   fm  x 
 fn  x   f  x   fm  x   f  x 
  / 2   / 2   for all x  E
Hence the condition is necessary.
Conversely, suppose that the given condition holds. Therefore {fn(x)} is a Cauchy sequence in R for
each x  E. Since R is complete, it follows that {fn(x)} converges to some value f(x), for each x  E &
{fn} converges to f pointwise. We need only to show that the convergence is uniform. to show this let
 >0 be given, then by hypothesis, n0  N (depending only on  ) such that
f n  x   f m  x   ,
n, m > N and x  E
Let n be fixed and let m   , then we have
fn  x   f  x     x  E
Hence f n  f uniformly on E.
We now find necessary and sufficient condition for uniform convergence of a series of functions.
Theorem 2 (Cauchy criterion for uniform convergence). A series of real functions  f n , each
defined on a set X converges uniformly on X iff for every   0,  n0  N (depending only on  )
such that
f n1  x   f n 2  x   .............  f n m  x    for n  n0 , m  1, x  X .
Proof. Let S n  x   f 1  x   f 2  x   .......... .  f n  x , x  X be a partial sum

n
 f  x   f  x   f  x   ...............  f  x  ,
i 1
i 1 2 n x X

so that S n  x  is a sequence of partial sums of the series f
n 1
n . Now the series f n is uniformly
convergent iff the sequence S n  is uniformly convergent.
i.e., for given   0,  a positive integer m such that n  m
S nm  x   S n  x    , m  1,2,........... [By Cauchy criteria of uniform converge of sequence]
 f n1  x   f n2  x   .............  f nm x    , m  1,2,........
This completes the proof of Cauchy’s Criteria for Series.

2.5 Tests for Uniform Convergence
In this section, we study Mn-test, Weierstrass M-test, Abel’s Test and Dirichlet’s Test for uniform
convergence and some examples which emphasis on the applications of these tests.
Theorem 1. Suppose lim f n  x   f  x  , x  E and let M n  lub f n  x   f  x  . Then f n  f uniformly

n  xE
on E if and only if M n  0 as n   . (This result is known as M n - Test for uniform convergence)
Proof. We have
lub f n  x   f  x   M n  0 as n   .
xE
Hence lim f n  x   f  x   0 for all x  E .

n 
Hence to each   0 , there exists an integer N such that n > N, x  E imply
fn  x   f  x   
Hence f n  f uniformly on E.
Example 1. By using Mn – test, show that the sequences { f n } where
nx
f n x   is not uniformly convergent on any interval containing 0.
1 n2 x2
nx
Solution. Here f  x   lim f n  x   lim
n  n  1  n2 x2
x/n
 lim 0
n  1/ n 2  x 2
Thus the sequence  f n  converges pointwise to the function f identically 0.
Now M n  sup f n  x   f  x 
xa ,b 
nx nx
 sup  0  sup
xa ,b  1  n x xa ,b  1  n x
2 2 2 2
nx
Let us find the maximum value of by second derivative test.
1 n2 x2
nx
Let   x  
1  n2 x2
 ' x  
1  n x n  nx.2n x
2 2 2
1  n x  2 2 2
Put  ' x   0 .
Then we have, 1  n 2 x 2 n  nx 2n 2 x   0
1  n x   x  2 xn 
2 2 2
1  2 x 2 n2  n2 x 2  1  n2 x 2
1 1
 x2  2
x .
n n
1 1
or x  or  .
n n
Also,
1  n x  .n  2n x   n 1  n x  4n x  4n x 
2
2 2 2 2 2 2 4 3
 " x  
1  n x 
4
2 2
    .
2
2n3 x 1  n 2 x 2  n 1  n 2 x 2 4n 2 x  4n 4 x 3

1  n x 
4
2 2
1 1
At x  ,  "  x   0 . Therefore d(x) is maximum when x  .
n n
1 1
Also    
n 2
Thus we take an interval [a,b] containing zero ,then
nx 1
M n  sup f n  x   f  x   sup  ,
x a ,b  1  n x
2 2
x a ,b  2
which does not tend to zero as n   .

Hence by Mn – test the sequence  f n  is not uniformly continuous in any interval containing zero.
Example 2. Show that the sequence  f n , where
x
f n x   converges uniformly on R.
1  nx 2
Solution. Here pointwise limit is
x
f  x   lim  0  x  R.
n 1  nx 2
x
Let   x   f n  x   f  x   .
1  nx 2
For maximum & minimum of  x  , we have
1  nx 2  2nx 2
 ' x   0  0
1  nx  2 2
1  nx 2 1
  0  1  nx 2  0  x   .
1  nx 
2
2 n
Now,  '  x  
1  nx 2


 1  nx 2  2
.
     
2 2 2
1  nx 2 1  nx 2 1  nx 2
1 2
  .
1  nx  1  nx2 2
2
2nx 8nx
 ''  x    ..
1  nx2  1  nx2 
2 3
1
Put x  ,
n
 1  2 n 8 n n n
  ''    2 2  23  2  n   2
 n
 1 
 ' '    ve maximum.
 n
1/ n 1
Hence max.   x    .
1  n 1/ n  2 n
Thus M n  sup f n  x   f  x 
xR
x
 sup  0  sup   x   1
xR 1  nx 2 xR 2 n
1
Also so lim M n  lim 0
n n
2 n
Hence by M n  test, the sequence  f n  x  uniformly converges on R.
Example 3. Show that 0 is a point of non – uniformly convergent of the sequence  f n  x  , where
f n x   nxe  nx ; x  0 .
Solution. Here pointwise limit,
f  x   lim f n  x   lim nxe nx  

n n   form  .
By L’Hospital rule, we get

x
 lim 0
n  xe nx
For maximum & minimum value of  x  , where
  x   f n  x   f  x   nxe  nx
 '  x   nx  ne  nx   ne  nx
 ' x   0  n 2 xenx  ne
 nx
Now 0
 ne  nx 1 1
x 2  nx
 x
n e n n
Now  ' '  x    n 2 x  ne  nx    n 2 e  nx    n 2 e  nx 

 n 3 xe nx  2n 2 e  nx
1 1 1 2n 2 n2
' '   n3. .  
n n e e e
1
 ' '  x  ve i.e., maximum at x 
n
1 1
Hence max. of  x   n. e 1  .
n e
Thus M n  sup f n  x   f  x 
xR
1
 sup nxe nx  0  sup   x  
xR xR e
1 1
So lim M n  lim   0
n  n  e e
Hence by M n  test, the sequence of function is not uniform convergent on R.
Weierstrass contributed a very convenient test for the uniformly convergence of infinite series of
functions.
Theorem 2 (Weierstrass M-test). Let f n  be a sequence of functions defined on E and suppose
f n  x   M n ( x  E , n = 1, 2, 3,……), where M n is independent of x. Then f n converges uniformly
as well as absolutely on E if M n converges.
Proof. Absolute convergence follows immediately from comparison test.

To prove uniform convergence, we note that
m m
Sm  x   Sn  x    fn  x    fn .
i 1 i 1
 f n 1  x   f n  2  x   ....  f m  x 
 M n 1  M n  2  ...  M m .
But since M n is convergent, given   0 , there exists N (independent of x) such that
Mn 1  Mn 2  ...  Mm  , n  N.
Hence
Sm  x   Sn  x   , n  N, x  E
and so f  x  converges uniformly by Cauchy criterion for uniform convergence.

n

cos n
Example 4. Consider the series  . We observe that
n 1 np
cos n 1
 p.
np n

1 cos n
Also, we know that n
n 1
p
is convergent if p > 1. Hence, by Weierstrass M-Test, the series  np

sin n
converges absolutely and uniformly for all real values of θ if p > 1. Similarly, the series 
n 1 np
converges absolutely and uniformly by Weierstrass’s M-Test.
Example 5. Taking M n  r n , 0 < r < 1, it can be shown by Weierstrass’s M-Test that the series
r n
cos n, r n
sin n, r n
cos2 n, r n
sin 2 n converge uniformly and absolutely.

x
Example 6. Consider  , xR .
n 1 n 1  nx 2 
We assume that x is positive, for if x is negative, we can change signs of all the terms. We have
x 1
fn  x   and fn '  x   0 implies nx 2 = 1. Thus maximum value of f n  x  is .
n 1  nx 2  2n3/2
1
Hence fn  x  
2n 3/ 2

1 x
Since  n3/2 is convergent, Weierstrass’s M-Test implies that  n 1  nx is uniformly convergent
n 1  2

for all x  R .

x
Example 7. Consider the series  , x  R . We have
n 1 n  x  2 2
x
fn  x  
n  x  2 2
and so fn '  x  
n  x  2 2
 2x  n  x 2  2x
n  x  2 4
Thus fn '  x   0 gives

x 4  n 2  2nx 2  4nx 2  4x 4  0
n 2  2nx 2  3x 4  0
3x 4  2nx 2  n 2  0
n n
x2  or x  .
3 3
3 3 1
Also f n ''  x  is negative. Hence maximum value of f n  x  is . Since n 3/2
is convergent, it
16n 3/ 2
follows by Weierstrass’s M-Test that the given series is uniformly convergent.
an xn 
a n x 2n
Example 8. The series  and 
n 1 1  x n 1 1  x
2n 2n
converge uniformly for all real values of x and a n is absolutely convergent. The solution follow the
same line as for example 7.
Lemma 1 (Abel’s Lemma). If v1 , v 2 ,...., vn be positive and decreasing, the sum u1 v1  u 2 v 2  ....  u n v n
lies between A v1 and B v1 , where A and B are the greatest and least of the quantities
u1 , u1  u 2 , u1  u 2  u 3 ,....., u1  u 2  ....  u n .
Proof. Write
Sn  u1  u 2  ....  u n .
Therefore
u1  S1 , u 2  S2  S1 ,....., u n  Sn  Sn 1.
Hence
n
u v
i 1
i i  u1v1  u 2 v2  ....  u n vn .
 S1v1  (S2  S1 )v 2  (S3  S2 )v3  .....  (Sn  Sn 1 )v n
 S1 (v1  v2 )  S2  v2  v3   .....  Sn 1  vn 1  vn   Sn vn
 A  v1  v2  v2  v3  ....  vn 1  vn  vn  .
 Av1.
Similarly, we can show that
n
u v
i 1
i i  Bv1.
Hence the result follows.


Theorem 3 (Abel’s Test). The series  u  x v  x  converges uniformly on E if
n 1
n n
(i) v  x  is a positive decreasing sequence for all values of x  E

n
(ii)  u  x  is uniformly convergent

n
(iii) v1  x  is bounded for all x  E , i.e., v1  x   M .
Proof. Consider the series  u  x v  x  , where v  x 

n n n is a positive decreasing sequence for each
x  E . By Abel’s Lemma
u n  x  v n  x   u n 1  x  v n 1  x   ....  u m  x  v m  x   A v n  x  ,
where A is greatest of the magnitudes
u n  x  , u n  x   u n 1  x  ,...., u n  x   u n 1  x   ....  u m  x  .
Clearly A is function of x.
Since  u  x  is uniformly convergent, it follows that

n

u n  x   u n 1  x   ....  u m  x   for all n > N, x  E
M

and so A  for all n > N (independent of x) and for all x  E . Also, since vn  x  is decreasing,
M
vn  x   v1  x   M since v1  x  is bounded for all x  E
Hence
u n  x  v n  x   u n 1  x  v n 1  x   ....  u m  x  v m  x   

for n > N and all x  E and so  u  x v  x  is uniformly convergent.
n 1
n n
Example 9. Consider the series
 1
n

x 2n

n 1 np 1  x 2n
.
 1
n
We note that if p > 1, then

np

is absolutely convergent and is independent of x. Hence, by
Weierstrass’s M-Test, the given series is uniformly convergent for all x  R .
 1
n
If 0  p  1 , the series  np
is convergent but not absolutely. Let
x 2n
vn  x  
1  x 2n
Then  vn  x  is monotonically decreasing sequence for |x| < 1, because
x 2n x 2n  2
v n  x   v n 1  x   
1  x 2n 1  x 2n  2
x 2n 1  x 2 
 (+ve)
(1  x 2n ) 1  x 2n  2 
x2
Also v1  x    1.
1 x2
 1
n

x 2n
Hence, by Abel’s Test, the series 
n 1 np
.
1  x 2n
is uniformly convergent for 0  p  1 and |x| < 1.
xn
Example 10. Consider the series  an . 1  x 2n
, under the condition that a n is convergent. Let
xn
vn  x  
1  x 2n
Then
vn  x  1  x 2n  2

v n 1  x  x(1  x 2n )
and so
vn  x  1  x  1  x 2n 1 
1 
v n 1  x  x(1  x 2n )
which is positive if 0 < x < 1. Hence v n  v n 1 and so  vn  x  is monotonically decreasing and positive.
x xn
Also v1  x   is bounded. Hence, by Abel’s test, the series  a n . 1  x 2n is uniformly convergent
1  x2
in (0, 1) if a n is convergent.
nx n 1 1  x 
Example 11. Consider the series a n
1  xn
under the condition that a n is convergent. We
have
nx n 1 1  x 
vn  x   .
1 xn
Then
vn  x  n 1  x n 1
 . .
v n 1  x  (n  1)x (1  x n )
n
Since  0 as n   , taking n sufficient large
(n  1)
vn  x 

1  x   1 if 0 < x < 1.
n 1
v n 1  x  (1  x n )
Hence <vn(x)> is monotonically decreasing and positive. Hence, by Abel’s Test, the given series
converges uniformly in (0, 1).

Theorem 4. (Dirichlet’s Test for uniform convergence). The series  u  x v  x  converges
n 1
n n
uniformly on E if
(i) v  x 
n is a positive decreasing sequence for all values of x  E , which tends to zero
uniformly on E
(ii) u  x
n oscillates or converges in such a way that the moduli of its limits of oscillation
remains less than a fixed number M for all x  E .

Proof. Consider the series u ( x)v ( x)
n n where vn  x  is a positive decreasing sequence tending to
n 1
zero uniformly on E. By Abel’s Lemma
u n  x  v n  x   u n 1  x  v n 1  x   ....  u m  x  v m  x   Av n  x  ,
where A is greatest of the magnitudes
u n  x  , u n  x   u n 1  x  ,...., u n  x   u n 1  x   ....  u m  x 
and A is a function of x.
s
Since  un  x  converges or oscillates finitely in such a way that  u  x   M for all x  E ,
r
n
therefore A is less than M. Furthermore, since vn  x   0 uniformly as n   , to each   0 there

exists an integer N such that

vn  x   for all n > N and all x  E.
M
Hence

u n  x  v n  x   u n 1  x  v n 1  x   ....  u m  x  vm  x   .M 
M

for all n > N and x  E and so u ( x)v ( x) is uniformly convergent on E.
n 1
n n
Another way of Dirichlet’s Test for uniform convergence with proof.
Statement. If V n x  is a monotonic function of x for each fixed value of x in a, b and V n x 
converges uniformly to zero for a  x  b and if there is a number M > 0 s.t.
n
U  x   M n & x   a, b , then the series V xU x

r 1
r n n is uniformly convergent on [a,b].
Proof. Since V n x  converges uniformly to zero thus for any   0,  an integer N (Independent of x)
s.t. for all x  a, b

Vn  x    n  N ...................... 1 .
4M
n
Let S n   U r  x  n  N & x   a , b 
r 1
so that Sn  x   M  n........................  2  .
n p
Now consider V xU x  V xU x  ...........  V xU x
r  n 1
r r n 1 n 1 n p n p
 Vn 1 x S n 1  S n  Vn 2 x S n 2  S n1  .........  Vn p x   Sn  p  S n p 1 
 Vn1x  Sn  Vn 1 x   Vn 2 x  S n1  ...........  Vn  p1  x   Vn  p x  S n p 1  Vn p  x S n  p
n+p-1
  Vr x   Vr 1 x S r x   Vn 1 x S n x   Vn  p x S n p  x 
r  n1
np n p 1
 
 1
Vr x U r x    Vr x  Vr 1 x  S r  x   Vn 1 x  S n x   Vn p  x  S n p x 
r 
n 1
r n
n p 1
 
 n 1 Vr x   Vr 1  x  M  4M .M  4M .M  n  N
r 
(By (1)&(2))

 M Vn 1  x   Vn  p  x  
2

 M  Vn 1  x   Vn  p  x   
2
    
M     .
 4M 4M  2
n p
Hence by Cauchy Criteria, the series V x U x 
r  n 1
r r converges uniformly on [a,b].
n
Remark 1. The statement  U  x   K  x   a, b 
r 1
r &  n is equivalent to saying that the sequence of
partial sum of series U  x 

n is bounded for each value of x  a, b i.e, for every point xi  a, b ,
n
there is a number ki such that U  x   k
r 1
r i i and there exists a number k such that ki  k  i .
This fact is also stated as the partial sum of the series is uniformly bounded.
This, in turm is equivalent to saying that the series u x either converges uniformly or oscillates
n
finitely.
So Dirichlet’s test can be states also as “If Vn  x  is a monotonic function of n for each fixed value of x
in [a,b] and V n  x  converges uniformly to zero for x  a, b and if Un(x) either uniformly converges
to zero or oscillates finitely in [a,b]. Then the series V xU x is uniformly convergent on [a,b].
n n
Cosn Sinn
Example 12. Prove that the series
np
 and 
n p
converges uniformly for all values of p > 0
in an interval  ,2    for 0     .
Solution. When, p > 1 , By Weierstrass M-test at once prove both the series uniformly converge for all
values of  .
When 0  p  1 , U r  Cosr
1
Take bn  p and U n  Cosn  or Sinn 
n
1
Then by Dirichlet’s test p is positive and monotonic decreasing and uniformly tending to zero with
n
n n
U
r 1
r   Cosn
r 1
 Cos  Cos2  ..............  Cosn
Sinn 
 2 .Cos Ist.angle  Last.angle 
Sin  2
2
Sinn 
 2 .Cos   n 
Sin  2
2
 Co sec  n (  Sin n  1 and Cos   1 ).
2
Cosn Sinn
Thus all the conditions of Dirichlet’s test are fulfilled and the series  np
and  np
converges on  ,2    .
2.6 Uniform Convergence and Continuity

We know that if f and g are continuous functions, then f + g is also continuous and this result holds for
the sum of finite number of functions. The question arises “Is the sum of infinite number of continuous
function a continuous function?”. The answer is not necessary. The aim of this section is to obtain
sufficient condition for the sum function of an infinite series of continuous functions to be continuous.
Theorem 1. Let f n  be a sequence of continuous functions on a set E  R and suppose that f n 
converges uniformly on E to a function f : E  R . Then the limit function f is continuous.
Proof. Let c  E be an arbitrary point. If c is an isolated point of E, then f is automatically continuous at
c. So suppose that c is an accumulation point of E. We shall show that f is continuous at c. Since f n  f
uniformly, for every   0 there is an integer N such that n  N implies

fn  x   f  x   for all x  E .
3
Since f M is continuous at c, there is a neighbourhood S  c  such that x  S  c   E (since c is limit

point) implies

fM  x   fM  c  .
3
By triangle inequality, we have
f  x   f  c   f  x   f M  x   f M  x   f M  c   f M  c   f (c)
 f  x   f M  x   f M  x   f M  c   f M  c   f (c)
  
   
3 3 3
Hence
f  x   f  c   , x  S     E
which proves the continuity of f at arbitrary point c  E .

Remark 1. Uniform convergence of f n  in above theorem is sufficient but not necessary to transmit
continuity from the individual terms to the limit function. For example, let f n :  0,1  R be defined for
n  2 by
 2 1
n x for 0  x 
n

  2 1 2
f n  x   n 2  x   for x .
  n n n
 2
0 for  x 1
 n
Each of the function fn is continuous on [0, 1]. Also f n  x   0 as n   for all x   0,1 . Hence the
limit function f vanishes identically and is continuous. But the convergence f n  f is non-uniform.
The series version of Theorem 1 is the following:
Theorem 2. If the series f  x 

n of continuous functions is uniformly convergent to a function f on
[a, b], then the sum function f is also continuous on [a, b].
n
Proof. Let Sn x    fni x , n  N and let   0 . Since f n converges uniformly to f on [a, b], there
i 1
exists a positive integer N such that


Sn  x   f  x   for all n  N and x  a, b (1) .
3
Let c be any point of [a, b], then (1) implies

Sn  c   f  c   for all n  N (2).
3
Since f n is continuous on [a, b] for each n, the partial sum
Sn  x   f1  x   f 2  x   ....  f n  x 
is also continuous on [a, b] for all n. Hence to each   0 there exists a   0 such that

Sn  x   Sn  c   whenever x  c   (3).
3
Now, by triangle inequality, and using (1), (2) and (3), we have
f  x   f  c   f  x   Sn  x   Sn  x   Sn  c   Sn  c   f (c)
 f  x   Sn  x   Sn  x   Sn  c   Sn  c   f (c)
  
     , whenever x  c  .
3 3 3
Hence f is continuous at c. Since c is arbitrary point in [a, b], f is continuous on [a, b].
However, the converse of Theorem 1 is true with some additional condition on the sequence f n  of
continuous functions. The required result goes as follows:
Theorem 3. (Dini‘s theorem on uniform convergence of subsequences (first form)). Let E be
compact and let f n  be a sequence of functions continuous on E which converges to a continuous
function f on E. If f n  x   f n 1  x  for n = 1, 2, 3, …, and for every x  E , then f n  f uniformly on E.
Proof. Take
gn  x   fn  x   f  x 
.
Being the difference of two continuous functions g n  x  is continuous. Also g n  0 and g n  g n 1 . We
shall show that g n  0 uniformly on E.
Let   0 be given. Since g n  0 , there exists an integer n  N x such that
gn  x   0   / 2
In particular
g Nx  x   0   / 2
i.e. 0  g Nx  x    / 2.
The continuity and monotonicity of the sequence g n  imply that there exists an open set J(x) containing
x such that
0  gn  t   
if t  J(x) and n  N x .
Since E is compact, there exists a finite set of points x1 , x 2 ,...., x m such that
E  J(x1 )  J(x 2 )  ...  J(x m ).
Taking

N  max N x1 , N x 2 ,..., N x m . 
it follows that
0  gn  t   
for all t  E and n  N . Hence g n  0 uniformly on E and so f n  f uniformly on E.
Theorem 4. If a sequence  f n  of real valued function converges uniformly to f in [a,b] and let x 0 be a
point of [a,b] s.t. lim f n  x   an ;  n  1, 2,......... . .
x  x0
Then (i) {a n } converges.
(ii) lim f  x   lim an .

x  x0 n 
i.e, lim lim f n  x   lim lim f n  x  .

x x0 n n x  x0
Proof. (i) The sequence f n  converges uniformly on [a,b] .Therefore for   0 , there exists an
integer m (independent of x ) s.t. for all x  a, b
f n  p  x   f  x     n  m, p  1 (By Cauchy’s Criterion).
Keeping n, p fixed and tending x  x 0 , we get
an  p  an    n  m, p  1
So that a n  is a Cauchy sequence and therefore converges to A.
(ii) Since  f n  converges uniformly to f.
Thus for given   0 , there exists an integer N 1 s.t. for all x  a, b .

f n x   f x    n  N1 1
3
Now the sequence a n  converges to A. So there exists an integer N 2 s.t.

an  A   n  N2 (2)
3
Now take a no. N such that N  max.N1 , N 2 
Since we have ,
lim f n  x   an
x  x0
In particular, lim f N  x   a N
x  x0
 for   0,  a   0 such that
f N x  a N   x  x0    3
3 whenever
Now, f  x   A  f x   f N x   f N  x   a N  a N  A
 f x  f N x  f N x  a N  a N  A

  
     whenever x  x0  
3 3 3
 lim f x  exists and is equal to A.
x x0
Thus lim f x   lim an  A .

x x0 n 
Hence the Proof.


Theorem 5. If a series f
n 1
n converges uniformly to f in [a,b] and x 0 is a point of [a,b] such that
lim f n  x   a n ; n  1,2,...............
x  x0

Then (i) a
n 1
n converges

(ii) lim f  x    a n
x  x0
n 1
Proof. (i) Given that the series f n converges uniformly on [a,b], for given   0 , there exists an
integer m such that for all x  a, b
n  p
 f x    n  m, p  1
r
r  n 1
(By Cauchy’s Criterion)
Keeping n,p fixed and taking the limits x  x 0 , we obtain
n p
 a x   
r  n 1
r
 the series a n converges to A.


(ii) Since the series 
n 1
f n converges uniformly to f, therefore for   0 , there exists an integer N 1
such that x  a, b , we have,

n

 f x f x  3  n  N .....................1
r 1
r 1
Again a n converges to A.
 for   02,  N 2 such that
n

a r A n  N 2 .................2
r 1 3
Also it is given that

lim f n x   an ; n  1,2,..........
x  x0
 for the given   0,  a  i  0 such that for i = 1,2,……..


f n x  a n  x  x0   i
Such that 3N whenever .
If we take   min . 1 ,  2 ,......... ...... N  , then we have

f n x   an  for x  x0  
3N
N N N
 
Thus  f x   a
r 1
r
r 1
r   f r x   ar  N .
r 1
 ............3
3N 3
Now for x  x0   , we have

N N N N
f x   A  f x    f r x    f r x    ar  a r A
r 1 r 1 r 1 r 1
.
Using (1),(2) & (3) , we get
f  x  A  
 lim f x  exists and is equal to A.

x x0
We have seen earlier that if sequence f n  is a sequence of continuous functions which converges
pointwise to the function f, then it is not necessary for f to be continuous. However, the concept of
uniform convergence is of much importance as the property of continuity transfers to the limit function
if the given sequence converges.
Theorem 6. If the sequence of continuous function f n  is uniformly convergent to a function f on
[a,b] then f is continuous on [a,b].
Proof. Let   0 be given.
Now given that sequence  f n  is uniformly convergent to f on [a, b], then there exists a positive integer
m such that

f n  x   f  x   n  m & x   a, b 1
3
Let x 0 be any point of [a, b].
In particular then from (1),


fn x0  f  x0  nm (2)
3
Now f n is continuous at x 0  [ a, b] .So, there exists   0 such that

f n x   f n x0   whenever x  x0    3
3
Hence for x  x0   , we have
f  x   f  x 0   f  x   f n  x   f n  x   f n  x0   f n  x 0   f  x0 
 f  x   f n  x   f n  x   f n  x0   f n  x 0   f  x0 
  
    (from (1), (2)& (3))
3 3 3
We get f x   f x0    whenever x  x0  
x 0  a, b 
Hence f is continuous at
 f is continuous on [a,b].

Theorem 7. If a series f
n 1
n of continuous function is uniformly convergent to a function f on [a, b],
then the sum function f is also continuous on [a, b].
Proof. Since the series f n converges uniformly on [a, b] to f on [a, b].
Thus given   0 , we can choose m such that

n

for all x  [a, b]  f  x   f  x   3 n  m.
r 1
r 1
Let x0 be any point in [a,b], then from (1),we have n = N

N

 f  x   f  x   3 n  m
r 1
r 0 0  2
Now it is given that each fn is continuous on [a,b] and in particular at x0.
Hence   0 , there exists   0 such that

N N

 f r  x    f r  x0  
r 1 r 1 3
whenever x  x0    3
Hence for x  x0   ,
N N N N
f  x   f  x 0   f  x    f r  x    f r  x    f r  x 0    f r  x 0   f  x0 
r 1 r 1 r 1 r 1
N N N N
  f r x   f x  
r 1
 f r  x0   f  x0  
r 1
 f r  x    f r  x0 
r 1 r 1
Thus from (1), (2) & (3) , we get

  
   
3 3 3
 f  x   f  x0   
 f is continuous at x0 on [a,b]. Since x0 was chosen arbitrary.
Hence the proof.
Remark 1. (i) Uniform convergence of the sequence  f n  is sufficient but not a necessary condition for
the limit function to be continuous. This means that a sequence of continuous functions may have a
continuous limit function without uniform convergence.
However the above theorem yields a negative test for uniform convergence of a sequence namely “If the
sequence of continuous functions is discontinuous, the sequence cannot be uniformly convergent.”

(ii) The same argument hold good in the case of infinite series fn 1
n .
The following examples illustrate the same:
(1) The sequence x  of continuous functions has a discontinuous limit function f which is given by
n
0, if 0  x 1
f  x  
1, if x 1 .
Then the sequence cannot uniformly convergent on [0, 1].
 nx 
(2) The sequence  2 2  of continuous functions has a continuous limit function but the given
1  n x 
sequence is not uniformly convergent.

(3) The sum of the functions of the series  1  x x

n 1
n
of the continuous functions.
1, if x  0
f  x  
0, if x  0
which is discontinuous on [0,1]. Therefore the series is not uniformly convergent on [0,1].
Note 1. 1  x  x  1  x 1  x  x  .......... .


n 2
n 1
 1 
 1  x     1.
 1 x 
Some important results

Here we state some results which we shall use in the following theorems & examples:
(1) Every monotonically increasing sequence bounded above converges to the least upper bound
(l.u.b.).
(2) Every monotonically decreasing sequence bounded below converges to greatest lower bound
(g.l.b).
(3) A real no.  is said to be a limit point of a sequence a n  if given any   0 and a +ve integer
m, there exists a +ve integer k > m such that ak     .
(4) Every bounded sequence has a cluster point.
(5) If a seq. a n  converges to L or diverges to  or   then every subsequence of a n  also

converges to L or diverges to  or  .
(6) Consider the geometric series
a  ar  ar 2  ............  ar n1  ...............

This series
(i) converges if r < 1.
(ii) diverges to  if r  1.
(iii) oscillate finitely if r = -1.
(iv) oscillates infinitely if r < -1.

 n 1
(7) Leibnitz’s Rule. The alternative series   1

n 1
an is convergent if
(i) a n 1  a n n
(ii) a n  0 as n  
(8) For every limit point of a sequence we can form a subsequence converging to limit point. Limit
point is also called subsequential limit.
Theorem 8 (Dini’s theorem on uniform convergence of subsequences(2nd form)). If a sequence of
continuous function  f n  defined on [a,b] is monotonically increasing & converges pointwise to a
continuous function f, then the convergence is uniform on [a,b].
Proof. The sequence  f n  is monotonically increasing and converges to f on [a,b].
Therefore, for any   0 and for a point x  a, b there is an integer N s.t.
0  f  x   fn  x     n  N 1
We consider Rn  f  x   f n  x  ; n = 1,2,…………..
Since the sequence  f n  is monotonically increasing. So, the seq. R n  x  is monotonically decreasing.
i.e, R1  x   R2  x   R3  x   ...........  Rn  x   2 .
Also, the sequence R n  x  is bounded below by 0.
Hence the seq. R n  converges pointwise to 0 on [a,b].

We claim that this convergence is uniform.
Suppose if possible for a fixed a 0  0,  no integer N which works for all x  a, b .
Then for each n = 1,2,3,………, there exists x n  a , b  such that
Rn  xn   a0  3
The seq. x n  of points belonging to the interval [a,b] is bounded and thus has atleast one limit say '  '
in [a,b].
Consequently, we can assume that there is a subsequence x n k

 of seq. x  converges to ' '
n
i.e, x n   as k  .
k
Now the function,

Rn  x   f  x   f n  x  is continuous being the difference of two continuous functions and thus for
every fixed m, we have
k 
 
lim Rm x nk  Rm    x nk   as k   .
Now for every m and any sufficiently large k, we have

n k  m, k  m .
Since Rm  is a decreasing sequence, we have

   
R m x nk  R nk x nk  a 0 (from (3))
 R x   a
m
.
nk 0
But this is contradiction to the fact that sequence Rm  converges pointwise to 0 i.e.,
lim Rm    0
n 
Thus the convergence must be uniform and this completes the proof.
Theorem 9 (Dini’s theorem on uniform convergence for series). If the sum function of a series
f n with non negative terms defined on an interval [a, b] is continuous on [a,b], then the series is
uniformly convergent on the interval [a, b].
Proof. Consider the partial sum of the given series
n
S n x    f r x 
r 1
Since all the function f r are non –ve . So, the seq. of partial sum S n  should be increasing.
Therefore, Sn  x   Sn1  x   n
i.e., S n  is an increasing sequence of continuous functions converges pointwise to a continuous function

f. Hence by Theorem 8, the sequence S n  converges uniformly and the given series is also uniformly
convergent.
This completes the proof.
Example 1. Show that the series
x4 x4
x 
4
  ...................
1 x4 1 x4  
2
is not uniformly convergent on [a,b].

Solution. The terms of the given series are quotient of two polynomials and hence continuous (Since
the polynomials are continuous and quotient of two continuous function is continuous).
Now, Let us find the sum function for the given series. Let, f  x  denotes the sum function of the given
series.
1 1
If x  0 then the series is a geometric series with common ratio 4 and
 1x  0,1 .
1 x 1 x4
Hence the sum function is given by
x4
f x    1 x4
1
1
1 x4
1  x 4 if x  0
Thus, f  x  
0, if x  0
which is discontinuous on 0 and hence on [0,1] . So, the series cannot converge uniformly on [0,1].
x
Example 2. Show that the series  nx  1n  1x  1 is uniformly convergent on any interval [a, b],
0 < a < b, but only pointwise on [0, b].
x 1 1
Solution. Let f n  x    
nx  1n  1x  1 n  1x  1 nx  1
Therefore nth partial sum is
n
S n  x    f r  x   f 1  x   f 2  x   ..........  f n  x 
r 1
 1   1 1   1 1 
 1      ..........    
 x  1   x  1 2x  1   n  1x  1 nx  1 
1
 1
nx  1
The sum function f  x   lim

n 
S n x 
 1 
 lim 1  
n
 nx  1 
1 if x  0

0, if x  0
Clearly f is discontinuous at x = 0 and hence discontinuous on [0, b].
This implies that the convergence is not uniform on [0, b] i.e, it is only pointwise.
Now take the interval [a,b] such that 0 < a < b, then the given series is uniformly convergent on [a,b] if
for given   0 .
1
S n x   f x   
nx  1
11 
i.e, if n    1
x  
11 
Now,   1 decreasing with x and its maximum value is
x  
11 
  1   m0 (say).
a 
If we take m  m 0 then for all x  a, b

Sn  x   f  x    n  m.
Hence the series converges uniformly on [a,b] s.t. 0 < a < b.

 1n1
Example 3. Show that the series  n x
n 1
2 is uniformly convergent but not absolutely for all real
values of x.

 1n1
Solution. The given series is  n x
n 1
2 .
1
Let an  .
n  x2
a n 1  a n n and a n  0 as n   .

 1n1
Hence by Leibnitz’s rule, the alternative series  is convergent.
n 1 n  x2
 
We know that a series a

n 1
n is said to be absolutely convergent if the series a
n 1
n is convergent.
 1
n 1
 
1
Now, n 1 n x 2

n 1 n  x
2 which behaves like  1n and hence is divergent.
It remains to prove that the given series is uniformly convergent.
Let S n  x  denotes the partial sum and S  x  denote the sum of the series.
Now, consider
1 1 1 1 1
S2n  x       ............. 
1 x 2
2 x 3 x
2 2
4 x 2
2n  x 2
 1 1   1 1   1 1 
 S2n  x        .................   
 1 x
2   
2  x2   3  x2 4  x2    2n  1  x 2 2n  x 2 
 
Now, note that each bracket in the above expression is positive. Hence S 2 n  x  is positive and increasing
to the sum S  x  .
 S x   S 2 n x   0
1 1 1
Also S  x   S 2 n  x      ......
2n  1  x 2n  2   x 2n  3  x 3
2 2
1  1 1 
      ...........
2 
2n  1  x  2n  2  x 2n  3  x 
2 2
1

2n  1  x 2
1
 .
2n  1
1
So, 0  S  x   S2 n  x   1 .
2n  1
Also, consider
1 1 1
S 2 n 1  x   S  x      .......... ....
2n  2   x 2n  3  x 2n  4   x 2
2 2
 1 1   1 1 
     
2 
   ..............
2 
 2n  2  x 2n  3  x   2n  4  x 2n  5  x 
2 2
1 1
S 2 n 1 x   S  x    .
2n  2  x 2n  2
2
1 1
 0  S2 n 1  x   S  x     2 .
2n  2 2n  1
Inequality (1) & (2) yield that for any   0 ,

we can choose an integer m s.t. for all values of x.
S x  S n x     n  m
 The series converges uniformly for all real values of x.
Example 4. Consider the seq.  f n  where

nx
fn  x   .
1  n2 x2
Show that the sequence of differentiable functions f n  does not converge uniformly in an interval
containing zero.
nx
Solution. Here f n  x  
1 n2 x2
 f  x   lim f n x   0
n 
 f '  x   0 for all real x

' x 
1  n x  n  2nx.n x  n
2 2 2 3  2 
 1/ n  x  2 x 
2 2
Now, f n
1  n x   
 1/ n 2  x 2 2 
2 4
2 2n
 
Now lim f n '  x   0 for x  0 .

n 
Thus lim f n ' x   f ' x 

n 
But at x=0; f n '  x   n and lim f n ' 0  

n
Thus at x  0; f '  x   lim

n 
fn '  x  .
Hence the sequence f n ' does not converges uniformly in an interval that contains zero.
2.7 Uniform Convergence and Integrability.
We know that if f and g are integrable, then  f  g    f   g and this result holds for the sum
of a finite number of functions.
The aim of this section is to find sufficient condition to extend this result to an infinite number of
functions.
Theorem 1. Let α be monotonically increasing on [a, b]. Suppose that each term of the sequence
f n  is a real valued function such that f n  R() on [a, b] for n = 1, 2,.. and suppose f n  f uniformly
on [a, b]. Then f  R() on [a, b] and
b b
 f d  lim  f
a
n 
a
n d ,
b b
that is,  lim f n (x) d(x)  lim  f n (x) d(x)
a
n  n 
a
(Thus limit and integral can be interchanged in this case. This property is generally described by saying
that a uniformly convergent sequence can be integrated term by term).
Proof. Let  be a positive number. Choose   0 such that

   b     a    1
3
This is possible since  is monotonically increasing. Since f n  f uniformly on [a, b], to each   0
there exists an integer n such that
f n  x   f  x    , x   a, b (2)
Since f n  R    , we choose a partition P of [a, b] such that


U  P, f n ,    L  P, f n ,    (3)
3
The expression (2) implies
fn  x     f  x   fn  x   
Now f  x   f n  x    implies, by (1) that

U  P, f ,    U  P, f n ,    (4)
3
Similarly, f  x   f n  x    implies

L  P, f ,    L  P, f n ,    (5)
3
Combining (3), (4) and (5), we get
U  P, f ,    L  P, f ,    
Hence f  R    on [a, b].
Further uniform convergence implies that to each   0 , there exists an integer N such that n  N

fn  x   f  x   , x   a, b 
   b     a  
Then for n > N,

b b b b
 f d   f n d 
a a
  f  f n  d   f  f n
a a
d
b

d  x  dx
   b     a   a

    b     a  
  .
 b   a 
b b
Hence  f d  lim  f
a
n 
a
n d
and the result follows.

The series version of Theorem 1 is
Theorem 2. Let f n  R , n = 1, 2, … If f n converges uniformly to f on [a, b], then f  R and

b  b
 f  x  d    f n  x  d ,i.e., the series

n 1 a
f n is integrable term by term.
a
Proof. Let Sn  denotes the sequence of partial sums of f n . Since f n converges uniformly to f on
[a, b], the sequence Sn  converges uniformly to f. Then Sn being the sum of n integrable functions is
integrable for each n. Therefore, by theorem 1, f is also integrable in Riemann sense and
b b
 f  x  dx  lim  S  x  dx
a
n 
a
n
b b b b
But  Sn  x dx   f 1  x dx   f 2 x  dx  ....   f n x dx

a a a a
n b
   fi  x  dx
i 1 a
b n b
 f  x  dx  lim   f  x  dx
a
n 
i 1 a
i
 b
   f i  x  d
i 1 a
and the proof of the theorem is complete.
Example 1. Consider the sequence f n  for which fn  x   nxe , n  N, x   0,1 . We note that
2
 nx
f  x   lim f n  x 
n 
nx
 lim 2
 0, x   0,1
n  nx n2x4
1   ......
1! 2!
Then
1
 f  x  dx  0
0
1 1
 f n x dx   nxe  nx dx
2
0 0
n
1 t
 
20
e dt, t  nx 2
1
 1  e  n 
2
Therefore
1 1
lim  f n  x  dx  lim 1  e n   .
n  n  2 2
1
If f n  were uniformly convergent, then  f  x  dx should have been equal to lim  f n  x  dx .
n 
0
But it is not the case. Hence the given sequence is not uniformly convergent to f. In fact, x = 0 is the
point of non-uniform convergence.

x
Example 2. Consider the series  . This series is uniformly convergent and so is integrable
 n  x2 
2
n 1
term by term. Thus
1    m 1
x x
0  
n 1  n  x 
2 2
dx  lim
 m
 
n 1 0  n  x 
2 2
 
m 1
 lim   x  n  x 2  dx
2
m 
n 1 0
1
  n  x 2 1 
m
 lim   
m
n 1  2 
 0
m
11 1 
 lim    
m
n 1 2  n n 1
1  1   1 1  1 1 
 lim 1        ...    
m 2
 2   2 3   m m  1 
1 1  1
 lim 1  
m 2
 m 1 2
 
Example 3. Consider the series  
nx 

 n  1 x , a  x  1.
n 1 1  n x 


2 2
1   n  1 x 2
2
  

Let Sn  x  denote the partial sum of the series. Then
nx
Sn  x  
1  n x 
2 2
and so f  x   lim Sn  x   0 for all x   0,1

n 
As we know that 0 is point of non-uniform convergence of the sequence Sn  x  , the given series is not
uniformly convergent on [0, 1]. But
1 1
 f  x  dx   0 dx  0
0 0
1 1
nx
and  S  x  dx   1  n x  dx
0
n
0
2 2
1
1 2n 2 x
2n 0 1  n 2 x 2 
 dx
1 
 
1
 log 1  n 2 2
x 
2n  0
1 
  log 1  n 2   .
2n
Hence
1

lim  Sn  x  dx  lim
1 
log 1  n 2
   form 
n  2n   
n 
0 
n  
 lim 2 
form 
n  1 n   
1
 lim  0.
n  2n
Thus
1 1
 f  x  dx  lim  S  x  dx ,
0
n 
0
n
and so the series is integrable term by term although 0 is a point of non-uniform convergence.
Theorem 3. Let g n  be a sequence of functions of bounded variation on [a, b] such that g n  a   0 ,

and suppose that there is a function g such that
lim V  g  g n   0
n 
and g(a) = 0. Then for every continuous function f on [a, b], we have
b b
lim  f dg n  lim  f dg
n  n 
a a
and g n  g uniformly on [a, b].
Proof. If V denotes the total variation on [a, b], then

V  g   V  gn   V  g  gn 
Since g n is of bounded variation and lim V  g  g n   0 it follows that total variation of g is finite and
n 
so g is of bounded variation on [a, b]. Thus the integrals in the assertion of the theorem exist.
Suppose f  x   M on [a, b]. Then
b b b
 f dg   f dg
a a
n   f d g  g 
a
n
 M V g  gn 
.
Since V  g  g n   0 as n   , it follows that

b b
 f dg  lim  f dg n .
a
n 
a
Furthermore,
g  x   gn  x   V g  gn  , axb
Therefore, as n   , we have
g n  g uniformly.
2.8. Uniform Convergence and Differentiation

If f and g are derivable, then
d d d
f  x   g  x    f x  gx
dx dx dx
and that this can be extended to finite number of derivable functions.
In this section, we shall extend this phenomenon under some suitable condition to infinite number of
functions.
Theorem 1. Suppose f n  is a sequence of functions, differentiable on [a, b] and such that f n  x 0 

converges for some point x 0 on [a, b]. If f n ' converges uniformly on [a, b], then f n  converges
uniformly on [a, b], to a function f, and
f '  x   lim f n '  x   a  x  b  .
n 
Proof. Let   0 be given. Choose N such that n  N, m  N implies

fn  x0   fm  x0   (1)
2
and

fn ' t   fm ' t   a  t  b (2).
2b  a 
Application of mean value theorem to the function f n  f m , (2) yields
x t  
fn  x   fm  x  f n  t  fm  t   (3)
2  b  a 2
for any x and t on [a, b] if n  N, m  N . Since
fn  x   fm  x   fn  x   fm  x   fn  x0   fm  x0   fn  x0   fm  x0  .
the relation (1) and (3) imply for n  N, m  N ,
fn  x   fm  x    / 2   / 2   a  x  b  .
Hence, by Cauchy criterion for uniform convergence, it follows that f n  converges uniformly on [a, b].
Let
f  x   lim f n  x  a  x  b .
n 
For a fixed point x   a, b , let us define
fn  t   fn  x  f t  f x
n  t   , t  (4)
tx tx
for a  t  b, t  x . Then
fn  t   fn  x 
lim  n  t   lim  fn ' x  (n = 1, 2,…) (5)
tx tx tx
Further, (3) implies

n  t   m  t   n  N, m  N .
2(b  a)
Hence n  converges uniformly for t  x . We have proved just now that f n  converges to f uniformly
on [a, b]. Therefore (4) implies that
lim  n  t     t  (6)
n 
uniformly for  a  t  b  , t  x . Therefore using uniform convergence of  n  and (5), we have

lim   t   lim lim  n  t 
tx t  x n 
 lim lim  n  t 
n  t  x
 lim f n '  x  .
n 
But lim   t   f '(x) . Hence

tx
f '  x   lim f n '  x  .

n 
Remark 1. If in addition to the above hypothesis, each f n ' is continuous, then the proof becomes
simpler. Infact, we have then
Theorem 2. Let  f n  be a sequence of functions such that
(i) each f n is differentiable on [a, b].
(ii) each f n ' is continuous on [a, b].
(iii)  f n  converges to f on [a, b].
(iv)  f n ' converges uniformly to g on [a, b], then f is differentiable and f n '  x   g  x  for all
x   a, b .
Proof. Since each f n ' is continuous on [a, b] and  f n '  converges uniformly to g on [a, b], the
application of Theorem 1 of section 2.6 of this unit implies that g is continuous and hence Riemann
integrable. Therefore, Theorem 1 of section 2.7 of this unit implies
t t
 g  x  dx  lim  f n '  x  dx
a
n 
a
But, by Fundamental theorem of integral calculus,

t
 f '  x  dx  f  t   f  a 
a
n n n
Hence
t
 g  x  dx  lim f  t   f  a 

a
n 
n n
Since  f n  converges to f on [a, b], we have
lim f n  t   f  t  and lim f n  a   f  a  .

n  n 
Hence
t
 g  x  dx  f  t   f  a 
a
and so
d 
t
  g  x  dx   f '  t 
dt  a 
or g  t   f '  t  , t   a, b .

The series version of Theorem 2 is
Theorem 3. If a series f n converges to f on [a, b] and
(i) each f n is differentiable on [a, b]
(ii) each f n ' is continuous on [a, b]
(iii)the series f n ' converges uniformly to g on [a, b]
then f is differentiable on [a, b] and f '  x   g  x  for all x   a, b .


Proof. Let Sn  be the sequence of partial sums of the series f
n 1
n . Since f n converges to f on [a, b],
the sequence Sn  converges to f on [a, b]. Further, since f n ' converges uniformly to g on [a, b], the
sequence Sn ' of partial sums converges uniformly to g on [a, b].
Hence, theorem 2 is applicable and we have
f '  x   g  x  for all x   a, b .
 
Example 1. Consider the series  
nx 

 n  1 x .
n 1 1  n x 


2 2
1   n  1 x 2
2
  

For this series, we have
nx
Sn  x   , 0  x 1
1  n x 
2 2
We have seen that 0 is a point of non-uniform convergence for this sequence. We have
nx
f  x   limSn  x   lim
n  n 
1  n x 
2 2
0 for 0  x  1.
Therefore
f '  0  0
Sn  0  h   Sn  0 
Sn '  0   lim
h  h
n
 lim n

h 0 1  n 2 h 2

Hence
limSn '  0  .
n 
Then
f '  0  limSn '  0 .

n 

sin nx
Example 2. Consider the series  , x  R . We have
n 1 n3
sin nx
fn  x  
n3
cos nx
fn '  x   .
n2
Thus
cos nx
 f ' x   n
n 2
cos nx 1 1
Since
n 2
 2 and
n
n 2
is convergent, therefore, by Weierstrass’s M-test the series f '  x 
n is
uniformly as well as absolutely convergent for all x  R and so f n can be differentiated term by
term.
'
   
Hence   n   fn '
f 
 n 1  n 1
'
  sin nx  
cos nx
or   n3   n 2

 n 1  n 1
2.9 Weierstrass’s Approximation Theorem
Weierstrass proved an important result regarding approximation of continuous function which has many
applications in numerical methods and other branches of mathematics.
The following computation shall be required for the proof of Weierstrass’s approximation theroem.
For any p,q  R , we have, by Binomial Theorem
n
n
  k p q k nk
 ( p  q)n , , n I , (1)
k 0  
where
n n!
 
 k  k ! n  k  !
Differentiating with respect to p, we obtain
n
n
  k kp k 1 n  k
q  n( p  q ) n 1 ,
k 0  
which implies
n
k n
 n  k p q k nk
 p( p  q ) n 1 , n  I (2).
k 0  
Differentiating once more, we have
n
k 2  n  k 1 n  k
 n2
  p q  p(n  1)( p  q)  ( p  q)
k 0 n  k 
n 1
and so
n
k 2  n  k nk 1 p
 2  
k 0 n  k 
p q  p 2 (1  )( p  q) n  2  ( p  q) n 1
n n
(3).
Now if x  [0,1] , take p = x and q = 1-x. Then (1), (2) and (3) yield
 n n k
  x 1  x   1
nk
 k 0  k 
 n k  n 
  x 1  x   x
k nk
(4).
 k 0 n  k 
 n k2 n
 2  x k 1  x n  k  x 2  1   
1 x
 k 0 n  k   n n
2
k 
On expanding   x  , it follows from (4) that
n 
x 1  x 
2
n
k  n k
   x   k x 1  x  
nk
0  x 1 (5).
k 0  n    n
For any f [0,1] , we define a sequence of polynomials Bn n 1 as follows:


n
n k
Bn  x     x k 1  x  f   ,
nk
0  x  1, n  I (6).
k 0  k  n
The polynomial B n is called the nth Bernstein Polynomial for f.
We are in a position to state and prove Weierstrass’s Theorem.

Theorem 1 (Weierstrass’s Approximation Theorem). If f is real continuous function defined on [a,b]
then there exists a sequence of real polynomials Pn  which converges uniformly to f(x) on [a,b]
i.e., lim Pn  x   f  x  uniformly on [a,b].

n 
Proof. If a = b, then f(x) = f(a).

Then, the theorem is true by taking Pn  x  to be a constant polynomial defined by
Pn  x   f  a   n
Thus we assume that a < b
xa
f  is continuous mapping of [a,b] onto [0,1].
ba
So, in our discussion W.L.O.G. we take a = 0, b = 1.
n
Now we know that for positive integer n and k where 0  k  n , the binomial coefficients   i.e,
k 
 n n!
nc k is defined as   
 k  k!n  k !
Now, we define the polynomial Bn where
n k
Bn  x      x k 1  x  f    
nk
k  n
The polynomial defined in (*) is called Bernstain polynomial as shown in above equation (6).
We shall prove that certain Bernstain polynomial exists which uniformly converges to f on [0,1].
Now consider the identity
n
  n
  k  x 1  x    x  1  x    1 1
k n k n
k 0  
[This is the binomial exp. of x  1  x  ]

n
Differentiating w.r.t. x, we get

  k kx 1  x   0
n
n
 n  k x k 1  x 
k 1 nk n  k 1
k 0  
n
n
    x k 1 1  x  k  nx   0
n  k 1
k 0  k 
Multiplying by x(1-x) yields

n
n
     x k 1  x   k  nx    0  2
nk
k 0  k 
 
Differentiating again w.r.t. x, we get
  k  nx 1  x k  nx2   0
n
n
 x k 1 1  x 
k nk n  k 1
k 0  
which on applying (1),we get
  k x 1  x k  nx2   n
n
 n k 1 n  k 1
k 0  
Multiplying by x(1-x), we get
n
  n
  k  x 1  x   k  nx   nx 1  x 
k nk 2
k 0  
x 1  x 
2
n
n
nk  k
    x k 1  x   x     3
k 0  k   n n
Since the maximum value of x(1-x) in [0,1] is ¼.
f  x   x 1  x  , f '  x   1  2 x
 f ' x  0  1 2x  0
 x  1/ 2  f 1/ 2   1/ 4.
So, (3) can be written as
2
n
n
nk  k 1
    x k 1  x   x     4
k 0  k   n  4n
Now f is continuous on [0,1]. So, f is bounded and uniformly continuous on [0,1].
  K  0 such that
f  x   K  x   0,1
and by uniform continuity for given   0 , there exists   0 such that for all x  0,1 .
k  k
 f  x   f    whenever x    5.
n 2 n
Now for any fixed but arbitrary x in [0,1], then n- values 0,1,2,………..,n of k can be divided into two
parts as follows:
k
Let A be the set of values of k for which x    and B be the set of remaining values for which
n
k
x  .
n
Now for k  B , we get by (4)

2
n k n k nk  k 1
   x 1  x       x 1  x   x   
nk 2
kB  k  kB  k   n 4n
n 1
   x k 1  x   6
nk
kB  k  4n 2
Now
n
n k
f  x   Bn  x   1. f  x      x k 1  x  f  
nk
k 0  k  n
(By (1))
n
n
  k 
  k  x 1  x   f x  
nk
 k
f  
k 0     n 
n
n k
f  x   Bn x      x k 1  x  f  x   f  
n k
k 0  k  n
We split the summation on R.H.S into two parts accordingly as
k k
x   or x   
n n
Let k  A or k  B .
Thus we have
n k n k
f  x   Bn  x      x k 1  x  f  x   f       x k 1  x  f  x   f  
nk k
k A  k   n  kB  k  n
   n n
   x 1  x   2 K    x k 1  x 
nk nk
 k
2 k
k A   kB  k 
 2K K
    for all values of n  2 .
2 4n 2

Thus B n  x  converges uniformly to f(x) on [0, 1].

Hence the proof.
 x f x dx  0 for n=0,1,2,………. Then show that f(x) = 0

n
Example1. If f is continuous on [0,1] and if
on [0,1].
Solution. Let p x   a 0  a1 x  a 2 x  .......... .....  a n x be a polynomial with real co-efficients defined

2 n
on [0,1], then
1 1
  
0 p  x  f  x dx  0   a n x n  f  x dx
n 0 
 1 
  an  x n f  x dx   an .0  0.
n 0 0 n 0
Thus the integral of product of f with any polynomial is zero.

Now, since f is continuous on [0,1], therefore by Weierstrass’s approximation theorem, there exists a
seq. p n  of real polynomial such that p n  f uniformly on [0,1].
 pn f  f 2
is uniformly on [0,1]
Since f being continuous and bounded on [0,1], therefore
1 1

0
f 2 dx  lim  p n. . f dx  0
n 
0
Therefore, f
2
x   0 on [0,1].
Hence f(x) = 0 on [0,1].
2.10 References
1. Walter Rudin, Principles of Mathematical Analysis (3rd edition) McGraw-Hill, Kogakusha, 1976,
International Student Edition.
Delhi, 2012.
U N IT – III
POWER SERIES AND FUNCTION OF SEVERAL VARIABLES
Structure
3.0 Introduction
3.1 Unit Objectives
3.2 Power Series
3.2.1 Power series
3.2.2 Uniform convergence and uniqueness theorem
3.2.3 Abel theorem
3.2.4 Tauber theorem
3.3 Function of several variables
3.3.1 Linear transformation
 Euclidean space Rn
3.3.2 Derivatives in an open subset E of Rn
 Chain rule
3.3.3 Partial derivatives
 Continuously differentiable mapping
 Young theorem
 Schwarz theorem
3.4 References
3.0 Introduction
In this unit, we study convergence and divergence of a power series and applications of Abel’s theorem.
Tauber showed that the converse of Abel’s theorem can be obtained by imposing additional condition on
coefficients, whenever the converse of Abel’s theorem is false in general. Many of the concepts i.e.,
continuity, differentiability, chain rule, partial derivatives etc are extended to functions of more than one
independent variable.
3.1 Unit Objectives
 understand the concept of power series and radius of convergence.
 identify the notation associated with functions of several variables
 familiar with the chain rule, partial derivatives and concept of derivation in an open subset of Rn.
 know the features of Young and Schwarz’s Theorems.
3.2 Power Series
A very important class of series to study is power series. A power series is a type of series with terms
involving a variable. Evidently, if the variable is x, then all the terms of the series involve powers of x.
So we can say that a power series can be design of as an infinite polynomial. In this section we will give
the definition of the power series as well as the definition of the radius of convergence, uniform
convergence and uniqueness theorem, Abel and Tauber theorems.

Definition 1. A power series is an infinite series of the form a
n 0
n x n where a n ' s are called its
coefficients.
Definition 2 (Convergence of power series). It is clear that for x = 0, every power series is convergent,
independent of the values of the coefficients. Now, we are given three possible cases about the
convergence of a power series.
(a) The series converges for only x = 0 which is trivial point of convergence, then it is called
“nowhere convergent”
e.g.  n! x n
converges only for x = 0 and for x  0 , we have
lim n! x n  .
n
Thus the terms of the series do not converge for x  0 and thus the series converges only for x =
0. Hence it is ‘Nowhere convergent’ series.
(b) The series converges absolutely for all values of x, then it is called “Everywhere convergent”.
e.g. The series converges absolutely for all values of x,
xn x n 1
un  , un 1 
n! n  1!
un x n (n  1)! n  1
lim   n 1   .
n  un 1 n! x x
By D-Ratio test, the series converges for all values of x. So, it is called “Everywhere convergent”
series.
(c) The series converges for some values of x and diverges for others.

e.g. The series x
n 0
n
converges for x < 1 and diverges for x > 1.
The collection of points x for which the series is convergent is called its “Region of
convergence”.
88 Power Series & Function of Several Variables

Definition 3. Let a
n 0
n x n be a power series. Then, applying Cauchy’s root test, we observe that the

power series a
n 0
n x n is convergent if
1
x  ,
L
where
1/n
L  lim a n .
1
The series is divergent if x  .
L
Taking
1
R 1/ n
.
lim a n
We will prove that the power series is absolutely convergent if x  R and divergent if x  R . If
a 0 , a1 ,.... are all real and if x is real, we get an interval R  x  R inside which the series is convergent.

If x is replaced by a complex number z, the power series a
n 0
n z n converges absolutely at all points z
inside the circle z  R and does not converge at any point outside this circle. The circle is known as
circle of convergence and R is called radius of convergence. In case of real power series, the interval
(-R, R) is called interval of convergence.
1/n
If lim a n  0, then R   and the power series converges for all finite values of x. The function
represented by the sum of series is then called an Entire function or an integral function. For example,
e z , sin z and cos z are integral functions.
1/n
If lim a n  , R  0, the power series does not converge for any value of x except x = 0.

Definition 4. Let R be the radius of convergence of the power series a
n 0
n z n , then the open interval
(-R, R) is called the interval of convergence for the given power series.
1
a
1/ n
Theorem 1. Let n x n be a power series such that lim a n  .
n  R
Then the power series is convergent with radius of convergence R.
Proof. Given that
1/ n 1
lim a n 
n  R
1/ n x
So, lim a n x n 
n R
x x
Hence by Cauchy’s Root test, the series a x n
n
is convergence if
R
 1 and divergent if
R
 1 i.e,
convergent if x  R and divergent if x  R . Hence by definition, R is radius of convergence of the

given power series.
Remark 1. (i) From the proof of above theorem, it follows that if for the series a x n
n
,
1/ n 1
lim a n 
n  R
then the series is absolutely convergent.
(ii) In view of the last theorem, we define the power series of convergence in the following way:
Consider the power series  an x n , then the radius of convergence of this series is given by
1 1/ n
R 1/ n
when lim a n 0
lim a n
1/ n
 0 when lim a n 
1/ n
  when lim an  0.
Obviously R   for an “everywhere convergent” and R  0 for a “nowhere convergent” series.
Theorem 2. If a power series a x n

n
converges for x  x 0 then it is absolutely convergent for every
x  x1 where x1  x0 .
Proof. Given that the series a x n

n
is convergent.
Thus a n x 0n  0 as n   .
Hence for   1 / 2 (say), there exists an integer N such that

1
a n x0n  n  N
2
Thus, we have
n
x
a x  an x 1
n
n 1
n
0
x0
n
1 x1
 n  N (*)
2 x0
x1
Now x1  x0   1.
x0
n
x1
Thus  x0
is geometric series with common ratio less than 1. So, it is convergent. By comparison
test, the series a n

x
n 1 converges.
  an x n is absolutely convergent for every x  x1 where x1  x0 .
Theorem 3. If a power series a x n

n
diverges for x  x ' then it diverges for every x  x" , where
x"  x' .
Proof. Given that the series a x n

n
diverges at x  x ' .
Let x" be such that x"  x' .
Let if possible, the series is convergent for x  x" , then by theorem 2, it must be convergent for all x
such that x  x" .
In particular, it must be convergent at x' which is contradiction to the given hypothesis.
Hence the series diverges for every x  x" , where x"  x' .
Definition 5 (Radius of Convergence).
an
For the power series a x n
n
, the radius of convergence is also defined by the relation R  lim
n  a
,
n 1
provided the limit exists.

This definition is commonly used for numerical purpose as illustrated below:
Find the radius of convergence of following:
x2 x3
(1) x    ...............
2! 3!
(2) 1  x  2! x 2  3! x 3  ....................
1 1.3 2 1.3.5 3
(3) x x  x  .................
2 2.5 2.5.8
1 2 1.2 3 1.2.3 4
(4) x  x  3 x  4 x  ...............
22 3 4
1
Solution. (1) Here a n 
n!
an 1
R  lim  lim  n  1!  
n  a n n!
n 1
The series converges for all values of x i.e, everywhere convergent.

(2) Here a n  n!
n!
R  lim 0
n  n  1!
So, the series converges for no value of x other than zero. So, it is nowhere convergent series.
1.3.5...................2n  1
(3) Here a n 
2.5.8.................3n  1
1.3.5................2n  1 2.5.8..............3n  13n  2

R  lim 
n 2.5.8................3n  1 1.3.5................2n  12n  1
3 2/n 3
 lim 
n  2  1/ n 2
3
So series converges for all x where x  .
2
(4) Here a n 
n  1!
nn
 n  1!   n  1
n 1 n 1
 1
R  lim  lim  1    e.
n  nn n! n 
 n
So the series is convergent for all x where x  e .
Definition 6. Let f(x) be a function which can be express in terms of the power series as

f  x    an x n ,
n 0
then f(x) is called sum function of the power series  an x n .

Remark 2. We have defined the uniform convergence of a series in a closed interval always. Thus, if a
power series converges uniformly for x  R , then we must express this fact by saying that the series
converges uniformly in closed interval  R   , R    , where   0 may be arbitrary chosen, however
if a power series converges absolutely for x  R , then we can directly say that the series converges
absolutely in (-R, R).

Theorem 4. Suppose the series a
n 0
n x n converges for x  R and define

f (x)   a n x n  x  R .
n 0
Then

(i) a n 0
n x n converges uniformly on [R  , R ], 0.
(ii) The function f in continuous and differentiable in (-R, R)


(iii) f ' (x)   na n x n 1  x  R
n 1
Proof. (i) Let  be a positive number. If x  R  , we have
a n x n  a n  R  
n
Since every power series converges absolutely in interior of its interval of convergence by Cauchy’s root
test, the series  a n  R   converges absolutely and so, by Weierstrass’s M-test,  a n x n converges
n
uniformly on [R  , R  ].
(ii) Also then the sum f(x) of  a n x n is continuous and differentiable on (-R, R) and a x n
n
is
uniformly convergent on   R   , R    .
Therefore, its sum function is continuous and differentiable on (-R, R).
(iii) Now consider the series  nan x n1 .
Since  n 
1/n
 1 as n  , we have
lim  n | a n |  lim | a n |
1/n 1/n
Hence the series  a n x n and  na n x n 1 have the same interval of convergence. Since  na n x n 1 is a
power series, it converges uniformly in   R   , R    for every  0. Then, by term by term
differentiation yields
 na n x n 1  f '(x) if x  R  .
But, given any x such that x  R we can find an  0 such that x  R. Hence
 na n x n 1  f '(x) if x  R.
Note. It follows from the above theorem 4 that by repeated application of the theorem f can be
differentiable any number of time and series obtained by differentiation at each step has the same radius
of convergence as series a x n
n
.
Theorem 5. Under the hypothesis of Theorem 4, f has derivative of all orders in (-R, R) which are given
by

f (k ) (x)   n(n  1)(n  2)...(n  k  1)a n x n  k .
n k
In particular
f ( k ) (0)  k ak , k  0,1, 2,...........
Proof. Let

f ( x )   nan x n .
n 0
Then by theorem 4,

f ' ( x )   n an x n 1.
n 1
Again applying theorem 4 to f ' (x) , we have


f " ( x)   n (n  1)an x n  2
n2
........................................
........................................
........................................

f ( k ) ( x)   n(n  1)(n  2)......(n  k  1) an x n  k .
nk
Clearly f ( k ) (0)  k ak the other sum vanish at x = 0.

Remark 3. If the coefficients of a power series are known, the values of the derivatives of f at the centre
of the interval of convergence can be found from the relation
f ( k ) (0)  k ak .
Also we can find coefficient from the values at origin of f , f ', f ",...
Theorem 6 (Uniqueness theorem). If  a n x n and  b n x n converge on some interval (-R, R), R>0
to some function f, then
a n  b n for all n  N.
Proof. Under the given condition, the function f have derivatives of all order in (-R, R) given by

f ( k ) ( x )   n( n  1)( n  2)......( n  k  1) an x n  k .
nk
Putting x = 0, this yields

f ( k ) (0)  k !a k and f ( k ) (0)  k !b k .
for all k  N. Hence

a k  b k for all k  N.


Theorem 7 (Abel’s Theorem (First form)). If a power series a
no
n x n converges at the point R of the
interval of convergence (-R, R), then it uniformly converges in the interval [0, R].
Proof. Consider the sum
Sn, p  an1Rn1  an2 Rn2  ..........  an p Rn p ; p  1, 2,................
Then, we have
S n ,1  a n 1 R n 1
S n , 2  a n 1 R n 1  a n  2 R n  2
and so on.
This gives
an 1 R n 1  Sn ,1 

an  2 R n  2  Sn ,2  Sn ,1 

..............................  (1)
............................. 

an  p R n  p  Sn , p  Sn , p 1 

Let є > 0 be given.

Now the series a
n 0
n x n is convergent at x = R.

The series of numbers a R
n 0
n
n
is convergent and hence by Cauchy’s general principle of convergence,
there exists an integer N such that
an 1 R n 1  an  2 R n  2  .......  an  q R n  q    n  N  q  1, 2,..........
 Sn ,q    n  N & q  1, 2,.......... (2)
Now if we take x   0, R i.e, 0  x  R , then we have

n p n  p 1 n 1
x x x
     .............     1.  3
R R R
Now, consider for all n  N ,
a n 1 x n 1  a n  2 x n  2  ...............  a n  p x n  p
n 1 n2 n p
n 1 x n2 x n p x
 an 1 R    an  2 R    ............  an  p R  
R R R
n 1 n2 n p
x x x
 Sn ,1     Sn ,2  Sn ,1     ............   Sn , p  Sn , p 1   
R R R
 x n 1  x n  2   x n  2  x n 3  x
n p
 Sn ,1        Sn ,2        ........  Sn , p  
 R   R    R   R   R
 x n 1  x n  2   x n  2  x n 3  x
n p
 Sn ,1        Sn ,2        .........  Sn , p  
 R   R    R   R   R
 x   x  
n 1 n2 n2 n3 n  p 1 n p n p
x x x x x
              ...........        
 R  R R R R R  R  
x
     .1   (by (3))
R
Thus we have proved that
an 1 x n 1  an  2 x n  2  .........  an  p x n  p    p  1, x   0, R  .

Hence by Cauchy’s criterion of convergence of series, the series a x
n 0
n
n
converges uniformly on [0, R].
Remark 4. (i) In case, a power series with interval of convergence (-R, R) converges at x   R , then
the series is uniformly convergent in [-R, 0].
Similarly, if a series convergent at the end points -R and R, then the series is uniformly convergent on [-
R, R].
(ii) If a power series with interval of convergence (-R, R) diverges at end point x  R , then it cannot be
uniformly convergent on [0, R].
For, if the series is uniformly convergent on [0, R], it will converge at x = R. A contradiction to the
given hypothesis.

Theorem 8 (Abel’s theorem (second form)). Let a

n0
n x n be a power series with finite radius of
convergence R and let f  x    an x ; x  R. If the series n

a n
x n converges at end point x = R then
lim f  x    an R n .
x R
Proof. First we show that there is no loss of generality if we take R  1 .
a xn
n
  an R n y n   bn y n where bn  a n R n
.
Now, this is a power series with radius R’, where
1 1 R
R'     1.
n 1/ n 1/ n
R
lim an R lim an R
So, if any series is given, we can transform it in another power series with unit radius of convergence.
Hence we can take R = 1.

Thus, now it is sufficient to prove that let a x
n 0
n
n
be a power series with unit radius of convergence and

let f x    an x n ; x  1, if the series  an converges then lim f  x    an . .Let us proceed to prove
x 1
n 0
the same.
S n  a 0  a1  .......... ..  a n

S 1  0 and a
n0
n  S.
m m
Then a n x n   S n  S n 1 x n
n 0 n0
m m
  S n x n   S n 1 x n
n 0 n 0
m 1 m
  S n x n  S m x m  x  S n 1 x n 1
n 0 n 0
m 1 m
  S n x n  x  S n 1 x n 1  S m x m
n 0 n 0
m 1 m 1
  Sn x n  x Sn x n  Sm x m  S1  0
n 0 n 0
m 1
 1  x   S n x n  S m x m .
n 0
Now, for x  1; x m  0 as m   and S m  S .
m m 1
 lim  an x n  lim 1  x   Sn x n  lim Sm x m
m  m  m 
n 0 n 0

 f  x    1  x Sn x n 1
n 0
Now, since S n  S , therefore for   0 , there exists integer N such that

Sn  S  n  N (2)
2
Also, we have

1  x   x n  1  3
n 0
Hence for n  N , we have


f x   S  1  x  S n x n  S
n 0
 
 1  x  S n x n  1  x  Sx n (by (3))
n 0 n 0

 1  x    Sn  S  x n
n 0
N
 
 1  x   Sn  S x n  (1  x)  x n
n 0 2 n  N 1
N

 1  x   Sn  S x n 
n 0 2
N
Now for a fixed N, 1  x  S n  S x n is continuous function of x having zero value at x = 1.
n 0
Thus, there exists   0 such that 1    x  1 .
1  x  S n  S x n  
N
n0 2
 
 f x   S     whenever 1    x  1
2 2

Hence, lim f  x   S  an .
x 1
n 0
Remark 5. We state some result related to Cauchy product of two series which will use in following
theorem, which is infact an application of Abel’s theorem.
  
(i) Let  an and
n0
 bn , then the series
n0
c
n 0
n where cn  a0bn  a1bn 1  ..........  an b0 is called
 
Cauchy product of series a
n0
n &  bn .
n 0
 
(ii) Cauchy’s Theorem. Let  an &
n0
b
n 0
n be absolutely convergent series such that
a n  A,  bn  B , then Cauchy’s product series c

n
n is also absolutely convergent and
c
n
n  AB .
  
Theorem 9. If  a n &  bn and
n0 n 0
c
n 0
n converges to sum A, B & C respectively and if c n be
Cauchy product of a n and b n then AB = C.

  
Proof.  c n is the Cauchy product of
n 0
 a n and
n 0
b
n 0
n .
 c n  a 0 bn  a1bn 1  ..........  a n b0
  
Let f x    a n x n , g x    bn x n and h x    c n x n ; 0  x  1 .
n 0 n0 n 0
For x  1, the three series converge absolutely
 c x n
n
 f  x g  x  (By Cauchy’s theorem in Remark 5(ii))
 h  x   f  x  .g  x  ;0  x  1. (1)
Now by Abel’s theorem


lim f  x    an  f  x   A as x  1

x 1
n0
Similarly, g  x   B, h  x   C as x  1 2
Thus from (1) & (2), we have
AB = C.
Example 1. Show that
x3 x5 x7
(i ) tan 1 x  x 
   ..............
3 5 7
 1 1 1
(ii )  1     .........
4 3 5 7
Solution. (i) We know that
1  x 
1
2
 1  x 2  x 4  x 6  ........; x  1 1
The series on the right is a power series with radius of convergence 1, so it is absolutely convergent in (-
1, 1) and uniformly convergent in [-k, k] where │k│< 1.
Now integrating (1), we get
x3 x5 x7
tan 1 x  c  x     ............; x  1
3 5 7
Putting x = 0, we obtain c = 0, so that
x3 x5 x7
tan 1 x  x     ...............; x  1
3 5 7
The series on R.H.S is a power series with radius of convergence equal to 1. However, the series
x3 x5 x7
x    ...........
3 5 7
is convergent at ±1.
Hence by Abel’s theorem, it is uniformly convergent in [-1, 1] and hence

x3 x5 x7
tan 1 x  x     ..................;1  x  1
3 5 7
(ii) At x = 1. By Abel’s theorem (Second form)
tan 1 x  lim tan 1 x
x 1
 1 1 1
 1     ..........
4 3 5 7
Example 2. Show that for -1≤ x ≤ 1,
x2 x3 x4
(i ) log(1  x)  x     .........
2 3 4
1 1 1
(ii ) log 2  1     ...........
2 3 4
Solution. (i) We know that
1  x 1  1  x  x 2  x 3  .......... ......; 1  x  1
On integrating, we get
x2 x3
log 1  x   x    .................;1  x  1
2 3
The power series on R.H.S. converges at x = 1.
So, by Abel’s theorem
x2 x3
log1  x   x    ...............;1  x  1
2 3
(ii) Put x = 1, in above series we get result
1 1 1
log 1  1  1     ............; 1  x  1
2 3 4
1 1 1
 log 2  1     ................; 1  x  1.
2 3 4
Tauber’s Theorem. The converse of Abel’s theorem proved above is false in general. If f is given by

f (x)   a n x n , r  x  r
n 0

the limit f (r ) may exists but yet the series a
n 0
n r n may fail to converge. For example, if

a n   1 , f x     1 x n
n n
n0

   x  , x  1
n
n 0
 1  x  x 2  x 3  .............
Then
1
f ( x)  ,  1  x  1.
1 x
1
 
f 1  lim f  x   lim
x 1 x 1 1 x
.
Put x = 1- h, if x→1-, h→0

1 1
 lim 
h 0 1  1  h 2
 
 f 1 exists.
However,
 
 a    1 is not convergent because this is oscillating between

n
n  1 & 1.
n 0 n 0
Tauber showed that the converse of Abel’s theorem can be obtained by imposing additional condition on
coefficients a n . A large number of such results are known now a days as Tauberian Theorems. We
present here only Tauber’s first theorem.

Theorem 10 (Tauber). Let f (x)   a n x n , for  1  x  1 and suppose that lim n  na n  0. If
n 0

f (x)  S as x  1 , then a
n 0
n converges and has the sum S.

Proof. Let n n   k a k . Then  n  0 as n  . (1)
k 0
1
Also, lim f ( xn )  S where xn  1  . (2)
n  n
 when 
n  , xn  1 , f ( xn )  S .
Therefore to each  0, we can choose an integer N such that n  N implies
  
 n  0  , f ( xn )  S  , nan  0 
3 3 3
i.e.,
  
 n  , f ( xn )  S  , n an  n  N . (3)
3 3 3
n
Let S n   ak . Then for 1  x  1, we have
k 0
n
S n  S   ak  S
k 0
n 
  ak  S  f ( x)   ak x k
k 0 k 0
n n 
 f ( x)  S   ak   ak x  k
ax k
k
k 0 k 0 k  n 1
n 
 f ( x)  S   ak (1  x k )  ax. k
k
k 0 k  n 1
n 
S n  S  f ( x)  S   ak (1  x k )  ax k
k
. (4)
k 0 k  n 1
Let x  (0,1). Then
(1  x k )  (1  x )(1  x  ........  x k 1 )  k (1  x )
for each k. Therefore, if n  N and 0  x  1, we have

n 
S n  S  f ( x)  S   ak (1  x k )  ax k
k
k 0 k  n 1
n 
 f ( x)  S   ak (1  x k ) 
k 0
ax
k  n 1
k
k
n 
 f ( x)  S   ak k (1  x) 
k 0
ax
k  n 1
k
k
n 
 f ( x)  S  (1  x) k ak  ax k
k
k 0 k  n 1
n
  k
 f ( x)  S  (1  x) k ak  x
k 0 3n k  n 1
n

 f ( x)  S  (1  x) k ak  .
k 0 3n(1  x)
1
Putting x  x n  1  , we find that
n
1
 (1  x) 
n
n
k ak 
Sn  S  f ( x)  S   
k 0 n 1
3n.
n
  
    .
3 3 3

a
n0
k converges & has sum S, which completes the proof.
3.3 Functions of Several Variables

This section is devoted to calculus of functions of several variables in which we study derivatives and
partial derivatives of functions of several variables along with their properties. The notation for a
function of two or more variables is similar to that for a function of a single variable. A function of two
variables is a rule that assigns a real number f(x, y) to each pair of real numbers (x, y) in the domain of
the function which can be extended to three and more variables.
3.3.1 Linear transformation
Definition 1. A mapping f of a vector space X into a vector space Y is said to be a linear transformation
if
f (x1  x 2 )  f (x1 )  f (x 2 ),
f (cx)  cf (x)
for all x, x1 , x 2  X and all scalars c.
Clearly, if f is linear transformation, then f(0) = 0.

A linear transformation of a vector space X into X is called linear operator on X.
If a linear operator T on a vector space X is one-to-one and onto, then T is invertible and its inverse is
denoted by T 1. Clearly, T 1 (Tx)  x for all x  X. Also, if T is linear, then T 1 is also linear.
Theorem 1. A linear operator T on a finite dimensional vector space X is one-to-one if and only if the
range of T is equal to X. i.e, T(X) = X.
Proof. Let R(T) denotes range of T. Let x1, x 2 ,..., x n  be basis of X. Since T is linear the set
Tx1,Tx2 ,...,Tx n  spans R(T). The range of T will be whole of X if and only if Tx1,Tx2 ,...,Tx n  is
linearly independent.
So, suppose first that T is one-to-one. We shall prove that Tx1 ,Tx 2 ,...,Tx n  is linearly independent.
Hence, let
c1Tx1  c 2 Tx 2  ...  c n Tx n  0
Since T is linear, this yields

T(c1x1  c 2 x 2  ...  c n x n )  0
and so c1 x1  c 2 x 2  ...  c n x n  0
Since x1 , x 2 ,..., x n  is linearly independent, we have, c1  c 2  ...c n  0.
Thus Tx1 ,Tx 2 ,...,Tx n  is linearly independent and so R(T) = X if T is one-to-one.
Conversely, suppose that Tx1 ,Tx 2 ,...,Tx n  is linearly independent and so
c1Tx1  c 2 Tx 2  ...  c n Tx n  0 (1)

implies c1  c 2  ...c n  0. Since T is linear (1) implies
T(c1x1  c 2 x 2  ...  c n x n )  0
 c1x1  c 2 x 2  ...  c n x n  0
Thus T(x) = 0 only if x = 0. Now

T(x)  T(y)  T(x  y)  0  xy0  xy
and so T is one-to-one. This completes the proof of theorem.
Definition 2. Let L(X, Y) be the set of all linear transformations of the vector space X into the vector
space Y. If T1, T2 ∈ L(X,Y) and if c1,c2 are scalars, then
(c1T1+c2T2)(x)= c1T1x+c2 T2x ; x ∈ X. It can be shown that c1T1+c2T2∈ L(X, Y).
Definition 3. Let X, Y and Z be vector spaces over the same field. If S, T ∈ L(X,Y), then we define their
product ST by
ST(x) = S(T(x)); x ∈ X.
Also, ST  L( X , Y ).
Euclidean space Rn. A point in two dimensional space is an ordered pair of real no. (x1, x2). Similarly, a
point in three dimensional space is an ordered triplet of real no. (x1, x2, x3). It is just as easy to consider
an ordered n-tuple of real no. (x1, x2,……, xn) and refer to this as a point in n-dimensional space.
Definition 4. Let n > 0 be an integer. An ordered set of n real no. (x1, x2,…….., xn) is called an n-
dimensional point or a vector with n-component points. Vector will usually be denoted by single bold
face letter.
e.g. x = (x1, x2,………., xn)
y = (y1, y2,…………, yn)
The number xk is called the kth co-ordinate of point x or kth component of the vector x.
The set of all n-dimensional point is called n-dimensional Euclidean space or n-space and is denoted by Rn.
Algebraic operations in Rn - n-dimensional Euclidean space are as follow:
Let x = (x1,x2,……,xn) and y = (y1,y2,…………,yn) be in Rn.
We define
(a) Equality x = y iff x1 = y1, x2 = y2, …………., xn = yn.
(b) Sum x + y = ((x1 + y1, x2 + y2,………,xn + yn)
(c) Multiplication by real no. (Scalar):

ax = a(x1, x2,……., xn) = (ax1,ax2,……., axn)
(d) Difference x - y = x+( -)y
(e) Zero vector or origin 0 = (0, 0, ……. ,0).
(f) Inner product or dot product

n
xy   xk yk .
k 1
(g) For all x ϵ Rn. Also if λ is such that
Tx   | x |, x  R n , then T  .
(h) Norm or length

If T∈ L(Rn, Rm). Then
lub  Tx : x  R n , x  1
is called Norm of T and is denoted by ||T||. The inequality

Tx  T x
and
1/2
 n 
x    xk2  .
 k 1 
The norm x  y is called the distance between x & y.
(i) Also, Let x and y denote points in Rn, then the following results hold:
x  0 and x  0 iff x  0.
(i)
(ii) ax  a . x for every real a.

(iii) x y  yx
(iv) Cauchy Schwarz Inequality:
2
 x. y   x y.
(v) x y  x  y .
Note 1. Sometimes the triangle inequality is written in the form
xz  x y  yz .
This follows form (v) by replacing x by x-y and y by y-z. We also have
x  y  x y
Definition 5. The unit co-ordinate vector uk in Rn is the vector whose kth component is 1 and remaining
components are zero. Then
u1  1, 0, 0, ......., 0 
u2   0, 1, 0,........, 0 
..........................
..........................
un   0, 0, .............. ,1 .
If x   x1 , x2 ,............, xn  , then
x  x1u1  x2u2  .........  xnun

&
.
x1  xu1 , x2  xu2 ,..........., xn  xun
The vectors u1, u2,………,un are also called basis vectors.

Theorem 2. Let T,S  L(R n , R m ) and c be a scalar. Then
(a ) T   and T is uniformly continuous mappings of R n and R m .

(b) T  S  T  S and cT  c T .
(c) If d (T , S )  T  S , then d is a metric.
n
Proof. (a) Let {e1, e2,….,en} be the standard basis in Rn and let x ϵ Rn. Then x   ci ei .
i 1
Suppose x  1 so that ci  1 for i  1, 2,..., n. Then

Tx   c Te   c . Te
i i i i
  Te i
Taking lub over x  R n , x  1

Tx   Tei  .
Further
Tx  Ty  T ( x  y)  T x  y ; x, y  Rn

So if x  y  , then
T
Tx  Ty ; x, y  Rn .
Hence, T is uniformly continuous.

(b) We have
(T  S)x  Tx  Sy
 Tx  Sx
 T x S x
 T  S  x
Taking lub over x  R n , x  1, we have

T S  T  S .
Similarly, it can be shown that
cT  c T .
(c) We have d(T,S)  T  S  0 and d(T,S)  T  S  0  T  S.

Also d(T,S)  T  S  S  T  d(S, T)
Further, if S, T, U  L(R n , R m ), then
S U  ST  T  U
 ST  T  U
Hence, d is a metric.
Theorem 3. If T ϵ L(Rn, Rm) and S ϵ L(Rn, Rm), then
ST  S T
Proof. We have
(ST)x  S(Tx)  S Tx
 S T x
Taking sup over x, x  1, we have

ST  S T .
In theorem 2, we have seen that the set of linear transformation form a metric space. Hence the concepts
of convergence, continuity, open sets etc. make sense in Rn.
Theorem 4. Let C be the collection of all invertible linear operators on Rn.
1
(a) If T  C, T 1  , S  L(R n , R m ) and S  T     , then S  C.

(b) C is an open subset of L(R n , R m ) and mapping T  T 1 is continuous on C.
Proof. We note that

x  T 1Tx  T 1 Tx
1
 Tx for all x  R n

and so
(   ) x   x   x
 Tx   x
 Tx  ( S  T ) x
 Sx  x  Rn . (1)
Thus kernel of S consists of 0 only. Hence S is one-to-one. Then Theorem 1 implies that S is also onto.
Hence S is invertible and so S ϵ C. But this holds for all S satisfying ||S - T|| < α. Hence every point of C
is an interior point and so C is open.
Replacing x by S-1y in (1), we have
(   ) S1 y  SS1 y  y
y
or S1 y 
 
1
and so S1 
 
sin ce S1  T 1  S1 (T  S)T 1
We have
S 1  T 1  S 1 T  S T 1

 . (2)
 (   )
Thus if f is the mapping which maps T → T-1, then (2) implies
ST
f (S)  f (T) 
 (   ) .
Hence, if ||S –T|| → 0 then f(S) → f(T) and so f is continuous. This completes the proof of the theorem.
3.3.2 Derivatives in an open subset E of Rn
In one-dimensional case, a function f with a derivative at c can be approximated by a linear
polynomial. In fact if fʹ(c) exists, let r(h) denotes the difference
f ( x  h)  f ( x )
r ( h)   f ' ( x) if h  0 (1)
h
and let r(0) = 0. Then we have
f ( x  h)  f ( x )  h f ' ( x )  hr ( h), (2)
an equation which holds also for h = 0. The equation (2) is called the First order Taylor formula for
approximating f(x + h) – f(x) by h fʹ(x). The error committed in this approximation is h r(h). From (1),
we observe that r(h) → 0 as h → 0. The error h r(h) is said to be of smaller order than h as h → 0. We
also note that h fʹ(x) is a linear function of h. Thus, if we write Ah = h fʹ(x), then
A(ah1  bh 2 )  aAh1  bAh 2
Here, the aim is to study total derivative of a function f from Rn to Rm in such a way that the above said
properties of hfʹ(x) and hr(h) are preserved.
Definition 1(Open ball and open sets in Rn). Let ‘a’ be a given point in Rn and let r be a given positive
number, then the set of all points x in Rn such that
xa  r
is called an open n-ball of radius ‘r’ and centre ‘a’.
We denote this set by B(a) or B(a, r) . The B (a, r) consists of all points whose distance from ‘a’ is less
than r.
In R1, this is simply an open interval with centre at a.
In R2, it is a circular disc.
In R3, it is a spherical solid with centre at a and radius r.
Definition 2 (Interior point). Let E be a subset of Rn and assume that a ∈ E, then a is called an interior
point of E if there is an open ball with centre surrounded by an n-ball. i.e.,
B  a   E.
The set of all interior points of E, is called the interior of E and is denoted by int E.
Any set containing a ball with centre ‘a’ is sometime called a neighbourhood of a.
Definition 3 (Open set). A set E in Rn is called open if all points are interior points.
Note 1. A set E is open if and only if E = interior of E.
Every open n-ball is an open set in Rn.
The cartesian product ( a1 , b1 )  a 2 , b2   .......... ....  a n , bn  of n-dimensional open interval
(a1,b1),…………,(an,bn) is an open set is Rn called n-dimensional open interval, we denote it by (a, b)
where
a   a1 , a2 ,............., an 
b   b1 , b2 ,.................., bn  .
Remark 1. (i) Union of any collection of open sets is an open set.

(ii) The intersection of a finite collection of open sets is open.
(iii) Arbitrary intersection of open sets need not be open.
e.g. Consider the seq. of open interval such that
 1 1
Gn   ,  ; n  N
 n n
Clearly each Gn is open set but G1  G2 .............Gn  {0} which is being a finite set is not open.
,
Definition 4 (The structure of open sets in R’). In R’ the union of countable collection of disjoint
open interval is an open set in R’ can be obtained in this way.
First we introduce the concept of a component interval.
Definition 5 (Component interval). Let E be an open subset in R’ and open interval I (which may be
finite or infinite) is called a component interval of E
If I  E and if there is no interval J  I s.t. I  J  E.
In other words, a component interval of E is not a proper subset of any other open interval contained in E.
Remark 2. (i) Every point of a nonempty open set E belongs to one and only one component interval of
E.
(ii) Representative theorem for open sets on the real line.
Every nonempty open set E in R’ is the union of a countable collection of disjoint intervals of E.
Definition 6(Closed set). A set in Rn is called closed if and only if its complement Rn - E is open.
Remark 3. (i) The union of a finite collection of closed sets is closed and the intersection of an arbitrary
collection of closed set is closed.
(ii) If A is open and B is closed, then A - B is open and B-A is closed.
A  B  A  Bc .
Definition 7 (Adherent point). Let E be a subset of Rn and x is point in Rn, x is not necessary in E.
Then x is said to be adherent to E if every n-ball B(x) contains atleast one point of E.
E.g. (i) If xєE, then x adherenes to E for the trivial reason that every n-ball B(x) contains x.
(ii) If E is a subset of R which is bounded above. Then sup.E is adherent to E.
Some points adheres to E because every ball B(x) contains points of E distinct from x these are called
adherent points.
Definition 8 (Accumulation point/Limit point). Let E be a subset of Rn and x is a point in Rn, then x is
called an accumulation point of E if every n-ball B(x) contains atleast one point of E distinct from x.
In other words, x is an accumulation point of E if and only if x adheres to E-{x}.
If x є E, but x is not an accumulation point of E, then x is called an isolated point of E.
e.g. (i) The set of numbers of the form 1/n (n=1,2,……..) has 0 as an accumulation point.
( ii) The set of rational numbers has every real number as accumulation point.
(iii) Every point of the closed interval [a, b] is an accumulation point of the set of numbers in the open
interval (a, b).
Remark 4. If x is an accumulation point of E, then every n-ball B(x) contains infinitely many points of
E.
Definition 9 (Closure of a set). The set of all adherent points of a set E is called a closure of E
__
and is denoted by E .
Definition 10 (Derived set). The set of all accumulation points of a set E is called the derived set of E
and is denoted by E’.
Remark 5. (i) A set E in Rn is closed if and only if it contains all its adherent points.
__
(ii) A set E is closed iff E  E .
(iii) A set E in Rn is closed iff it contains all its accumulation points.
Definition 11. Suppose E is an open set in Rn and let f : E → Rn be a function defined on a set E in Rn
with values in Rm. Let x ϵ E and h be a point in Rn such that |h| < r and x + h ϵ B(x, r). Then f is said to
be differentiable at x if there exists a linear transformation A of Rn into Rn such that
f(x+h)=f(x)+Ah+r(h) (1)
where the reminder r(h) is small in the sense that
| r(h) |
lim h  0  0.
|h|
We write fʹ(x) = A.
The equation (1) is called a First order Taylor formula.
f ( x  h)  f ( x)  Ah
 0. (2)
h→0 h
The equation (2) thus can be interpreted as “For fixed x and small h, f(x + h) – f(x) is approximately
equal to fʹ(x)h, that is, the value of a linear function applied to h.”
Also (1) shows that f is continuous at any point at which f is differentiable.
The derivatives Ah derived by (1) or (2) is called total derivative of f at x or the differential of f at x.
In particular, let f be a real valued function of three variables x, y, z say. Then f is differentiable at the
point (x, y, z) if it possesses a determinant value in the neighbourhood of this point and if
f  f (x  x, y  y, z  z)  f (x, y, z)  Ax  By  Cz    , where  | x |  | y |  | z |,
 0 as   0 and A, B, C are independent of x, y, z. In this case Ax  By  Cz is called
differential of f at (x, y, z).
Theorem 1 (Uniqueness of derivative of a function). Let E be an open set in Rn and f maps E in Rm
and x ϵ E. Suppose h ϵ Rn is small enough such that x + h ϵ E. Then f has a unique derivative.
Proof. If possible, let there are two derivatives A1 and A2. Therefore
f (x  h)  f (x)  A1 h
lim h 0 0
h
and
f (x  h)  f (x)  A 2 h
lim h 0 0
h
Consider B = A1 - A2. Then

Bh = A1h – A2h
= f(x + h)- f(x) +f(x)- f(x + h) + A1h – A2h
= f(x + h)- f(x) – A2h + f(x)- f(x + h) + A1h
and so
Bh  f ( x  h)  f ( x)  A2 h  f ( x  h)  f ( x)  A1h
which implies
Bh f (x  h)  f (x)  A1h f (x  h)  f (x)  A 2 h
lim h 0  lim h 0 
|h| |h| |h|
0
For fixed h ≠ 0, it follows that

B(th)
 0  as t  0. (1)
th
The linearity of B shows that L.H.S of (1) is independent of t. Thus Bh = 0 for all h ϵ Rn. Hence B = 0,
that is, A1 = A2, which proves uniqueness of the derivative.
The following theorem, known as chain rule, tells us how to compute the total derivatives of the
composition of two functions.
Theorem 2 (Chain rule). Suppose E is an open set in Rn, f maps E into Rm, f is differentiable at x0 with
total derivative fʹ(x0), g maps an open set containing f(E) into Rk and g is differentiable at f(x0) with total
derivative gʹ(f(x0)). Then the composition map F = fog, a mapping E into Rk and defined by F(x) =
g(f(x)) is differentiable at x0 and has the derivative
Fʹ(x0) = gʹ(f(x0)) fʹ(x0).
Proof. Take
y0 = f(x0), A = fʹ(x0), B = gʹ(y0)
and define
r1 (x)  f (x)  f (x 0 )  A(x  x 0 )
r2 (y)  g(y)  g(y0 )  B(y  y 0 )
r(x)  F(x)  F(x 0 )  BA(x  x 0 ).
To prove the theorem, it is sufficient to show that
Fʹ(x0) = BA,
that is,
r ( x)
0 as x  x0
x  x0 (1)
But, in term of definition of F(x), we have

r(x)  g(f (x))  g(y 0 )  B(f (x)  f (x 0 )  A(x  x 0 ))
so that
r(x)  r2 (f (x))  B r1 (x). (2)
If ϵ > 0, it follows from the definitions of A and B that there exists η > 0 and δ > 0 such that
r2 ( y )
 as y  y0
y  y0
or r2 ( y)  y  y0 as y  y0   i.e., f ( x)  f ( x0 )  
and r1  x   x  x0 if x  x0   .
Hence
r2 ( f ( x))   f ( x)  f ( x0 )
 r1 ( x )  A( x  x0 ) (3)
 2 x  x0   A ( x  x0 )
and
Br1  x   B r1  x 
 B x  x0 if x  x0   (4)
Using (3) and (4), the expression (2) yields

Hence
r ( x)  2 x  x0   A  x  x0    B  x  x0 
r ( x)
 2   A   B
x  x0
   A  B  if x  x0  
Hence,
r ( x)
 0 as x  x0
x  x0
which in turn implies

Fʹ(x0) = BA = gʹ(f(x0)) fʹ(x0).
3.3.3 Partial derivatives.
Let {e1 , e2 ,........en } be the standard basis of Rn. Suppose f maps an open set E  Rn into Rm and
let f1 , f 2 ,..., f m be components of f. Define D k f i on E by
fi ( x  tek )  fi ( x)
( Dk fi )( x)  lim
t 0 t (1)
provided the limit exists.

Writing f i (x1 , x 2 ,..., x n ) in place of fi (x) we observe that D k f i is derivative of fi with respect to x k ,
f i
keeping the other variable fixed. That is why, we use frequently in place of D k fi .
x k
Since f  (f1 , f 2 ,..., f n ), we have
D k f (x)  (D k f1 (x), D k f 2 (x),..., D k f n (x))
which is partial derivative of f with respect to xk.

Furthermore, if f is differentiable at x, then the definition of f′(x) shows that
f ( x  thk )  f ( x)
lim  f '( x)hk
t 0 t (2)
If we take h k  e k , taking components of vector in (2), it follows that

“If f is differentiable at x then all partial derivatives D k f i (x) exist”.
In particular, if f is real valued (m=1), then (1) takes the form

f (x  t)  f (x)
(Dk f )(x)  lim t 0
t .
For example, if f is a function of three variables x, y and z, then
f (x  x, y, z)  f (x, y, z)
Df (x)  lim x 0
x
f (x, y  y, z)  f (x, y, z)
Df (y)  lim y 0
y
f (x, y, z  z)  f (x, y, z)
and Df (z)  lim z 0
z
and are known respectively as partial derivatives of f with respect to x, y, z.
The next theorem shows that A h  f   x  (h) is a linear combination of partial derivatives of f.
Theorem 1. Let E  Rn and let f : E → Rn be differentiable at x (interior point of open set E). If
h  c1e1  c 2 e 2  ...  c n e n where e1 ,e2 ,...,en  is a standard basis for Rn, then
n
f   x  (h)   c k D k f ( x).
k 1
Proof. Using the linearity of f   x  , we have

n
f   x  (h)   f (x)(ck ek )
k 1
n
  ck f ( x )ek
k 1
But, by (2),
f   x  e k  (D k f )(x)
n
Hence f   x  (h)   c k D k (f )(x )
k 1
If f is real valued (m = 1), we have

f   x  (h)  (D1f (x), D 2 f (x),..., D n f (x )) h.
Definition 1 (Continuously differentiable mapping).

A differentiable mapping f of an open set E  Rn into Rm is said to be continuously differentiable in E
if f′ is continuous mapping of E into L(Rn, Rm).
Thus to every ϵ > 0 and every x ϵ E there exists a δ > 0 such that
f '( y)  f '( x)  if y  E and y  x   .
In this case we say that f is a C′-mapping in E or that f ϵ C′(E).
Theorem 2. Suppose f maps an open set E  Rn into Rm. Then f is continuously differentiable if and
only if the partial derivatives Djfi exist and are continuous on E for 1  i  m,1  j  n.
Proof. Suppose first that f is continuously differentiable in E. Therefore to each x ϵ E and ϵ > 0, there
exists a δ > 0 such that
f (y)  f (x)  if y  E and y  x   .
We have then
f '( y )e j  f '( x)e j  ( f '( y )  f '( x))e j

 f '( y )  f '( x) e j (1)
 f '( y )  f '( x)  if y  E and y  x   .
Since f is differentiable, partial derivatives Djfi exist. Taking components of vectors in (1), it follows that
(D jf i )(y)  D jf i (x)  if y  E and y  x   .
Hence D jf i are continuous on E for 1  i  m,1  j  n.
Conversely, suppose that D jf i are continuous on E for 1  i  m,1  j  n. It is sufficient to consider one-
dimensional case, i.e., the case m = 1. Fix x ϵ E and ϵ > 0. Since E is open, x is an interior point of E and
so there is an open ball B  E with centre at x and radius R. The continuity of D jf implies that R can be
chosen so that

( D j f )( y )  ( D j f )( x)  if y  B, 1  j  n. (2)
n
Suppose h   h je j ,| h | R, and take v0  0
and vk  h1e1  h 2e2  ....  h k ek for 1  k  n.
Then
n
f ( x  h)  f ( x )   [ f ( x  v j )  f ( x  v j 1 )].
j 1
(3)
n
 f s  v
j 1
j i  h j e j   f ( x  v j 1 )
Mean value theorem implies

n
f  x  h   f  x    h j D j f  x  v j 1   j h j e j  for some    0,1
j 1
n
subtracting h  D f   x 
j 1
j j
and then taking modulus,

n n
f  x  h   f  x   h j  D j f   x    h D f  x  v
j j j 1   j h j e j   h j  D j f   x  
j 1 j 1
n
  h ( D j f )  x  v j 1   j h j e j     D j f   x  
j   
j 1

n

  | hj |
j 1 n
 n 
 
n j 1
| h | .n | h |  | h |
n
Hence f is differentiable at x and f'(x) is the linear function which assigns the number

f '  x  h  h j  D j f   x  when f'(x) is applied on h. Since  D1 f  x  ,  D2 f  x  ...,  Dn f  x  are
j 1
continuous functions on E, it follows that f' is continuous and hence f  C '  E  .
Hence f is differentiable at x and f′(x) is the linear function which assigns the number  h j (D i f )(x) to
the vector h   h je j . The matrix [f′(x)] consists of the row ((D1f )(x), (D 2 f )(x)...., (D n f )(x)) . Since
(D1f )(x), (D 2 f )(x)...., (D n f )(x) are continuous functions on E, it follows that f′ is continuous and hence
f  C(E).
Classical theory for functions of more than one variable
Consider a variable u connected with the three independent variables x, y and z by the functional
relation
u = u(x, y, z)
If arbitrary increment Δx, Δy, Δz are given to the independent variables, the corresponding increment
Δu of the dependent variable of course depends upon three increments assigned to x, y, z.
Definition 2 (Continuous function). Let u : Rn  R be a function. Then u is said to be continuous at a

point x   x1 , x 2 ,........, x n   R n . If given   0 , there exists a   0 such that
u  x1 , x2 ,..........., xn   u  x1  x1 , x2  x2 ,........., xn  xn  < 
whenever   x12  x22  .............  xn2   .

Definition 3 (Differentiable function). A function u = u(x, y, z) is said to be differentiable at point (x,
y, z) if it posses a determinant value in the neighbourhood of this point and if
u  Ax  By  Cz   ,
where   x  y  z , 0 as   0 and A, B, C are independent of x, y, z.
In the above definition ρ may always be replaced by η, where
  x 2  y 2  z 2 .
So, if u : Rn  R be a function, then u is said to be differentiable at a point x   x1 , x 2 ,......... .., x n   R n if
there exist constants A1 , A2 ,......... ....., An such that for given   0
u( x1  x1 , x2  x2 ,......, xn  xn )  u( x1 , x2 ,...., xn )  A1x1  A2 x2  ........  An xn  

n
where    x
i 1
2
i &   0 whenever   0 .
Definition 4 (Partial derivative). If the increment ratio

u(x  x, y, z)  u(x, y, z)
x
tends to a unique limit as x tends to zero, this limit is called the partial derivative of u with respect to
u
x and is written as or u x .
x
u u
Similarly, and can be defined.
y z
So, if u : R n  R be a function, we define a partial derivative as
u u  x1 , x2 ,............, xi  xi ,........, xn   u  x1 , x2 ,............, xn 

 lim ; i  1, 2,......., n. .
xi xi 0 xi
The differential coefficients. If in the relation

u  Ax  By  Cz  
we suppose that y  z  0, then, on the assumption that u is differentiable at the point (x, y, z),
u  u(x  x, y, z)  u(x, y, z)
 A x    x
u
and by the taking limit as x  0, since  0 as x  0, we get  A.
x
u u
Similarly  B and  C.
y z
u u u
Hence, when the function u = u(x, y, z) is differentiable, the partial derivatives , , are
x y z
respectively the differential coefficients A, B, C and so
u u u
u  x  y  z   
x y z
The differential of the dependent variable du is defined to be the principal part of Δu so that the above
expression may be written as
u  du    .
Now as in the case of functions of one variable, the differentials of the independent variables are
identical with the arbitrary increment of these variables. If we write u = x, u = y, u = z respectively, it
follows that
dx  x, dy  y, dz  z
Therefore, expression for du reduces to
u u u
du  dx  dy  dz
x y z .
Proposition 1. Let f : R n  R be a function. If f is differentiable at a point x  x1 , x 2 ,......., x n   R n

then f  x1  x1 , x 2  x 2 ,......... ., x n  x n   f  x1 , x 2 ,......... ..., x n 
f f f
 x1  x 2  .........  x n  
x1 x 2 x n
n
where    x
i 1
i
2
and   0 as   0 .
Proof. Since f is differentiable at a point x   x1 , x 2 ,......... ., x n  , by definition of differentiability, there

exists constants A1 , A2 ,......... , An such that, for given   0
f  x1  x1 , x 2  x 2 ,......., x n  x n   f  x1 , x 2 ,......... ....., x n 
 A1x1  A2 x2  ...........  An xn   *
where    x 2
i and   0 as   0 .
Taking x j  0 for j  i for some fixed i  1,2,....., n  .
Thus, we have
f  x1 , x2 ,............, xi  xi ,.................., xn   f  x1 , x2 ,..........xn 
 Ai  
xi
Taking xi  0
f  x1 , x2 ,.........., xi  xi ,............., xn   f  x1 , x2 ,............, xn     0as  0

lim  Ai  
xi 0 xi orxi  0 
f
  Ai (By definition of partial derivative)
xi
This is true for every i  1,2,........., n 
f f f
  A1 ,  A2 ,......... .......,  An
x1 x 2 x n
Putting these value in equation (*), we get
f  x1  x1 , x 2  x 2 ,......... , x n  x n   f  x1 , x 2 ,......... .., x n 
f f f
 x1  x 2  .......... ...  x n  
x1 x 2 x n
where    xi 
1/ 2
and   0 as   0 .
Remark 1. If the function u  u x1 , x 2 ,......... ...., x n  is differentiable at point x1 , x 2 ,......... , x n  then the
partial derivative of u w.r.t. x1 , x 2 ,......... ...., x n certainly exist and are finite at this point, because by the
above proposition, they are identical to constants A1 , A2 ,......... .., An respectively.
However converse of this is not true, i.e., partial derivatives may exist at a point but the function need
not be differential at that point.
In other words, we can say partial derivatives need not always be differential coefficients.
The distinction between derivatives and differential coefficients
We know that the necessary and sufficient condition that the function y = f(x) should be differentiable at
the point x is that it possesses a finite definite derivative at that point. Thus for functions of one variable,
the existence of derivative f′(x) implies the differentiability of f(x) at any given point.
For functions of more than one variable this is not true. If the function u = u(x, y, z) is differentiable at
the point (x, y, z), the partial derivatives of u with respect to x, y and z certainly exist and are finite at
this point, for then they are identical with differential coefficients A, B and C respectively. The partial
derivatives, however, may exist at a point when the function is not differentiable at that point. In other
words, the partial derivatives need not always be differential coefficients.
x 3  y3
Example 1. Let f be a function defined by f (x, y)  , where x and y are not simultaneously zero,
x 2  y2
f(0, 0) = 0.
If this function is differentiable at the origin, then, by definition,
f (h, k )  f (0, 0)  Ah  Bk   (1)
where   h 2  k 2 and  0 as   0.
Putting h   cos  , k   sin  in (1) and dividing through by  and taking limit as   0 , we get
cos3   sin 3   A cos   Bsin 

which is impossible, since θ is arbitrary.
The function is therefore not differentiable at (0, 0). But the partial derivatives exist however, for
f (h, 0)  f (0, 0) h 0
f x (0, 0)  lim h 0  lim h 0 1
h h
f (0, k)  f (0, 0) 0k
f y (0, 0)  lim k 0  lim k 0  1.
k k
 xy
Example 2.  if x 2  y 2  0
Let f (x, y)   x 2  y 2 .
0 if x  0, y  0

Then f x (0, 0)  0  f y (0, 0)
and so partial derivatives exist. If it is different, then

df  f (h, k)  f (0, 0)  Ah  Bk    , where A  f x (0, 0), B  f y (0, 0).
This yields
hk
 h 2  k 2 ,  h 2  k 2
h k
2 2
or hk  (h 2  k 2 )
Putting k = mh, we get

mh 2  h 2 (1  m 2 )
m
or 
1  m2
m
Hence lim k 0  0, which is impossible. Hence the function is not differentiable at the origin.
1  m2
Remark 2. (i)Thus the information given by the existence of the two first partial derivatives is limited.
The values of f x (x, y) and f y (x, y) depend only on the values of f(x, y) along two lines through the
point (x, y) respectively parallel to the axes of x and y. This information is incomplete and tells us
nothing at all about the behavior of the function f(x, y) as the point (x, y) is approached along a line
which is inclined to the axis of x at any given angle θ which is not equal to 0 or  / 2.
(ii) Partial derivatives are also in general functions of x, y and z which may possess partial derivatives
with respect to each of the three independent variables, we have the definition
  u  u x (x  x, y, z)  u x (x, y, z)
a)    lim x 0
x  x  x
  u  u x (x, y  y, z)  u x (x, y, z)

b)    lim y 0
y  x  y
  u  u x  x, y, z  z   u x  x, y, z 
c)    lim z 0
z  x  z
provided that each of these limits exist. We shall denote the second order partial derivatives by
2u 2u 2u
or u xx , or u yx and or u zx .
x 2 yx zx
u u
Similarly we may define higher order partial derivatives of and .
y z
The following example shows that certain second partial derivatives of a function may exist at a point at
which the function is not continuous.
 x 3  y3
 when (x, y)  (0, 0)
Example 3. Let   x, y    x  y
0 when (x, y)  (0, 0).

This function is discontinuous at the origin. To show this it is sufficient to prove that if the origin is
approached along different paths, ϕ(x, y) does not tend to the same definite limit. For, if ϕ(x, y) were
continuous at (0, 0), ϕ(x, y) would tend to zero (the value of the function at the origin) by whatever path
the origin were approached.
Let the origin be approached along the three curves
(i) y  x  x 2 (ii) y  x  x 3 (iii) y  x  x 4 ;
Then we have
2x 3  0(x 4 )
(i)   x, y    0 as x  0
x2
2x 3  0(x 4 )
(ii)   x, y    2 as x  0
x3
2x 3  0(x 4 )
(iii)   x, y     as x  0
x4
   
Certain partial derivatives, however, exist at (0, 0), for if xx denote   we have, for example
x  x 
  h, 0     0, 0  h2
x  0, 0   lim h0  lim h0 0
h h
  h, 0   x  0, 0  2h
xx  0, 0   lim h0  lim h0  2,
h h
since  (x, 0)  x 2 , x (x, 0)  2x when x  0.
The following example shows that uxy is not always equal to uyx.
 xy(x 2  y 2 )
 when (x, y)  (0, 0)
Example 4. Let f  x, y    x 2  y2
0 when (x, y)  (0, 0).

When the point (x, y) is not the origin, then
f  x2  y 2 4x2 y 2 
 y 2  2 2 2
x  x  y (x  y ) 
2
(1)
f  x2  y 2 4x2 y 2 
 x 2  2 2
y  x  y (x  y ) 
2 2
(2)
while at origin,
f (h, 0)  f (0, 0)
f x (0, 0)  lim 0
h 0 h (3)
and similarly f y (0, 0)  0.
From (1) and (2), we see that

f x (0, y )   y ( y  0) and f y ( x, 0)  x ( x  0)
(4)
Now we have, using (3) and (4)
f y (h, 0)  f y (0, 0) h
f xy (0, 0)  lim h 0  lim h 0  1
h h
f x (0, k)  f x (0, 0) k
f yx (0, 0)  lim k 0  lim h 0  1.
k k .
and so f xy (0, 0)  f yx (0, 0).
Example 5. Prove that the function
f (x, y)  xy
f f
is not differentiable at the point (0, 0), but that and both exist at the origin and have the value
x y
zero.
Hence deduce that these two partial derivatives are continuous except at the origin.
Solution. We have
f f (h, 0)  f (0, 0)
(0, 0)  lim h 0 0
x h
f f (0, k)  f (0, 0)
(0, 0)  lim k 0 0
y k
If f(x, y) is differentiable at (0, 0), then we must have
f (h, k)  0.h  0.k   h 2  k 2
where  0 as h2  k2  0 .
hk
Now 
h2  k2
Putting h   cos  , k   sin  , we get
 cos  sin 
lim  0  cos  sin   cos  sin   0 which is impossible for arbitrary  .
Hence, f is not differentiable.

Now, suppose that (x, y)  (0, 0). Then
f f (x  h, y)  f (x, y)
 lim h 0
x h
(x  h)y  xy xh  x
 lim h 0  lim h 0 | y |

h (x  h)y  xy  h  xh  x 
y 1 y
Now, we can take h so small that x + h and x have the same sign. Hence the limit is or .
2 xy 2 x
f x 1 x
Similarly,  or . Both of these are continuous except at (0,0). We now prove two
y 2 xy 2 y
theorems, the object of which is to set out precisely under what conditions it is allowable to assume that
f xy (a, b)  f yx (a, b).
Theorem 3 (Young). If (i) f x and f y exist in the neighbourhood of the point (a, b) and (ii) f x and f y are
differentiable at (a, b); then
f xy  f yx .
Proof. We shall prove this theorem by taking equal increment h for both x and y and calculating  2 f in
two different ways, where
 2 f  f (a  h, b  h)  f (a  h, b)  f (a, b  h)  f (a, b).
Let
H(x)  f (x, b  h)  f (x, b)
Then
 2 f  H(a  h)  H(a).
Since f x exists in the neighbourhood of (a, b), the function H(x) is derivable in (a, a  h). Applying mean
value theorem to H(x) for 0    1, we obtain
H(a  h)  H(a)  hH(a   h)

Therefore
 2 f  hH ' (a   h)
 h[ f x (a   h, b  h)  f x (a   h, b)] (1)
By hypothesis (ii) of theorem, f x (x, y) is differentiable at (a, b) so that
f x (a   h, b  h)  f x (a, b)   hf xx (a, b)  hf yx (a, b)   h
and
f x (a   h, b)  f x (a, b)   hf xx (a, b)  h,
where  and  tend to zero as h → 0. Thus, we get (on subtracting)
f x (a   h, b  h)  f x (a   h, b)  hf yx (a, b)  (   ) h
Putting this in (1), we obtain

2 f  h2 f yx  1 h2 (2)
where 1 '  '' , so that 1 tends to zero with h.

Similarly, if we take
K(y)  f (a  h, y)  f (a, y)
Then we can show that
2 f  h2 f xy  2 h2 (3)
where 2  0 with h.
From (2) and (3), we have
 2f
 f yx (a, b) 1  f xy (a, b) 2
h2
Taking limit as h  0, we have
 2f
lim h 0  f yx (a, b)  f xy (a, b)
h2
which establishes the theorem.
Theorem 4 (Schwarz). If (i) f x , f y , f yx all exist in the neighbourhood of the point (a, b) and (ii) f yx is
continuous at (a, b); then f xy also exist at (a, b) and f xy  f yx .
Proof. Let (a  h, b  k) be point in neighbourhood of (a, b). Let (as in the above theorem)
 2 f  f (a  h, b  k)  f (a  h, b)  f (a, b  k)  f (a, b).
and
H(x)  f (x, b  k)  f (x, b)
so that we have
 2 f  H(a  h)  H(a).
Since f x exists in the neighbourhood of (a, b), H(x) is derivable in (a, a  h). Applying Mean value
theorem to H(x) for 0    1, we have
H(a  h)  H(a)  hH(a   h)
and therefore
 2 f  hH (a   h)  h[f x (a   h, b  k)  f x (a   h, b)].
Now, since f yx exists in the neighbourhood of (a, b), the function f x is derivable with respect to y in
(b, b  k). Applying mean value theorem, we have
2f  hkf yx (a   h, b   k), 0     1
That is
1  f (a  h, b  k)  f (a  h, b) f (a, b  k)  f (a, b) 
h  k

k   f yx (a   h, b   k )
Taking limit as k tends to zero, we obtain

1
[ f y (a  h, b)  f y (a, b)]  lim f yx (a   h, b   ' k )  f yx (a   h, b) (1).
h k 0
Since f yx is given to be continuous at (a, b), we have
f yx (a   h, b)  f yx (a, b)  ,
where  0 and h  0.
Hence taking the limit h  0 in (1), we have
f y (a  h, b)  f y (a, b)
lim h  0  lim h  0 [f yx (a, b)  ]
h
that is, f xy (a, b)  f yx (a, b)

Remark 3. The conditions of Young or Schwarz’s Theorem are sufficient for f xy  f yx but they are not
necessary. For example, consider the function
 x 2 y2
 , (x, y)  (0, 0)
f  x, y    x 2  y 2
0 , (x, y)  (0, 0).

We have
f  h, 0   f  0, 0 
f x  0, 0   lim h0 0
h
f  0, k   f  0, 0 
f y  0, 0   lim k 0 0
k
Also for (x, y)  (0, 0), we have
(x 2  y 2 )2xy 2  x 2 y 2 .2x 2xy 4

f x (x, y)  
(x 2  y 2 ) 2 (x 2  y 2 ) 2
2x 4 y
f y (x, y) 
(x 2  y 2 ) 2
Again
f x (0, k)  f x (0, 0)
f yx (0, 0)  lim k 0  0 and f xy (0, 0)  0
k
So that f xy (0, 0)  f yx (0, 0).
For (x, y)  (0, 0), we have
8xy3 (x 2  y2 )2  2xy4 4y(x 2  y2 ) 8x 3 y3

f yx (x, y)  
(x 2  y2 )4 (x 2  y2 )3
Putting y = mx, we can show that

lim ( x ,y )  ( 0,0 ) f yx (x, y)  0  f yx (0, 0)
so that f xy is not continuous at (0, 0). Thus the condition of Schwarz’s theorem is not satisfied.
To see that conditions of Young’s theorem are also not satisfied, we notice that
f x (h, 0)  f x (0, 0)
f xx (0, 0)  limh 0  0.
h
If f x is differentiable at (0, 0), we should have
f x (h, k)  f x (0, 0)  hf xx (0, 0)  kf yx (0, 0) 

2hk 4
 ,
(h 2  k 2 ) 2
where   h 2  k 2 and  0 as   0.
Put h   cos  , k   sin  , then   h 2  k 2  
so we have
2  cos  . 4 sin 4 
 
4
2 cos  .sin 4  
Taking limit as ρ → 0, we have
2 cos  .sin 4   0
which is impossible for arbitrary θ.
3.4 References
1. Walter Rudin, Principles of Mathematical Analysis (3rd edition) McGraw-Hill, Kogakusha,
1976, International Student Edition.
Delhi, 2012.
U N IT – IV
TAYLOR THEOREM
Structure
4.0 Introduction
4.1 Unit Objectives
4.2 Taylor Theorem
4.3 Explicit and Implicit Functions
4.3.1 Implicit function theorem
4.3.2 Inverse function theorem
4.4 Higher Order Differentials
4.4.1 Choice of independent variables
4.4.2 Higher order derivatives of implicit functions
4.5 Change of Variables
4.6 Extreme Values of Explicit Functions
4.7 Stationary Values of Implicit Functions
4.8 Lagrange Multipliers Method
4.9 Jacobian and its Properties
4.9.1 Jacobian
4.9.2 Properties of Jacobian
4.10 References
4.0 Introduction
In this unit, we study most important mathematical tool of analysis i.e. Taylor theorem. As we know,
Taylor series is an expression of a function as an infinite series whose terms are expressed in term of the
values of the function’s derivatives at a single point. Also we shall be mainly concerned with the
applications of differential calculus to functions of more than one variable such as how to find stationary
points and extreme values of implicit functions, implicit function theorem, Jacobian and its properties etc.
4.1 Unit Objectives
 solve Taylor series expansions.
 find the stationary points and extreme values of implicit functions.
 understand Jacobian and its properties.
 know about the local character of Implicit function i.e. the implicit function is a unique solution
of a function f(x, y)=0 in a certain neighbourhood.
130 Taylor Theorem
4.2 Taylor Theorem
In view of Taylor’s theorem for functions of one variable, it is not unnatural to expect the possibility of
expanding a function of more than one variable f (x  h, y  k, z  m), in a series of ascending powers of
h, k, m. To fix the ideas, consider a function of two variables only; the reasoning in general case is
precisely the same.
Theorem 1 (Taylor’s theorem). If f(x, y) and all its partial derivatives of order n are finite and
continuous for all point (x, y) in domain a  x  a  h, b  y  b  k, then
1 2 1
f (a  h, b  k)  f (a, b)  df (a, b)  d f (a, b)  ...  d n 1f (a, b)  R n
2! (n  1)!
1 n
where Rn  d f (a   h, b   k), 0    1.
n!
Proof. Consider a circular domain of centre (a, b) and radius large enough for the point (a  h, b  k) to
be also with in domain. Suppose that f(x, y) is a function such that all the partial derivatives of order n of
f(x, y) are continuous in the domain. Write
x  a  ht, y  b  kt,
so that, as t ranges from 0 to 1, the point (x, y) moves along the line joining the point (a, b) to the point
(a  h, b  k) ; then
f (x, y)  f (a  ht, b  kt)   (t).
f dx f dy f f
Now,  (t)  .  . h k  df
x dt y dt x y
and
  f f  dx   f f  dy
 "(t )  h  k   h  k 
x  x y  dt y  x y  dt
 2 f dx  2 f dx  2 f dy  2 f dy
h  k  h  k
x 2 dt xy dt yx dt y 2 dt
 2 2 f 2 f 2 f 2  f 
2
 h  hk  hk k 
 x xy yx y 2 
2
 2 2 2 
  h2 2  2hk  k2 2  f (by Schwarz ' s theorem)
 x xy y 
2
   
  h  k  f (a  ht , b  kt )
 x y 
and hence, similarly we get
 (t)  d 2 f ,...,  (n ) (t)  d n f
Also, ϕ(t) and its n derivatives are continuous functions of t in the interval 0  t  1, and so, by
Maclaurin’s theorem
t2 t n (n )
 (t)   (0)  t (0)   (0)  ...   ( t) (1)
2! n!
where 0    1. Now put t = 1 and observe that
 (1)  f (a  h, b  k),
 (0)  f (a, b),
 (0)  df (a, b),
 (0)  d 2 f (a, b),
.....
 (n ) ( t)  d n f (a   h, b   k).
It follows immediately from (1) that
1 2 1
f (a  h, b  k)  f (a, b)  df (a, b)  d f (a, b)  ...  d n 1f (a, b)  R n (2)
2! (n  1)!
1 n
where Rn  d f (a   h, b   k), 0    1.
n!
Here, we assumed that all the partial derivatives of order n are continuous in the domain. Taylor
expansion does not necessarily hold if these derivatives are not continuous.
Remark 1. If we put a  b  0, h  x, k  y, from the equation (2), we get
1 2 1
f (x, y)  f (0, 0)  df (0, 0)  d f (0, 0)  ...  d n 1f (0, 0)  R n
2! (n  1)!
1 n
where Rn  d f ( x, y), 0    1.
n!
This is known as Maclaurin’s theorem.
2. If we put a  h  x, b  k  y, we get
  
f ( x, y)  f (a, b)  ( x  a)  ( y  b)  f (a, b)  ........
 x y 
n 1
1   
  ( x  a )  ( y  b)  f (a, b)  Rn ,
(n  1)!  x y 
n
1  
where Rn  ( x  a)  ( y  b)  f (a  ( x  a) , b  ( y  b) ) .
n!  x y 
This is called Taylor’s expansion of f ( x, y ) about the point (a, b) in power of ( x  a) and ( y  b) .
132 Taylor Theorem
Example 1. If f (x, y)  xy , prove that Taylor’s expansion about the point (x, x) is not valid in any
domain which includes the origin.
Solution. Given that f ( x, y)  xy

.
f (h, 0)  f (0, 0)
We find f x (0, 0)  lim 0
h 0 h
f (0, k )  f (0, 0)
f y (0, 0)  lim 0
k 0 k
1 y
 ,x 0
2 x
Now, f x ( x, y )  
 1 y
 2 x , x  0

1 x
 ,y0
2 y
Also f y ( x, y )  
 1 x
 2 y , y  0

1
 2 , x  0
Thus, f x ( x , x )  f y ( x, x )  
 1 , x  0
 2
Now, Taylor’s expansion about ( x, x) for n  1 is
f ( x  h, x  h)  f ( x, x)  h  f x ( x   h, x   h)  f y ( x   h, x   h)
 x  h, x   h  0

x  h   x  h, x   h  0 (1)

 x , x   h  0.
If the domain (( x, x), ( x  h, x  h)) contains origin then x and x  h must be of opposite sign i.e.
xh  xh, x  x
or x  h  ( x  h) , x x
under these conditions none of the equality in (1) holds.

Hence the expansion is not possible because partial derivatives f x and f y are not continuous in any
domain which contains origin.
(Partial derivatives f x , f y are not continuous at origin and therefore Taylor’s theorem is not necessary
valid).
Example 2. Expand x 2 y  3 y  2 in power of ( x  1), ( y  2) .
Solution. Let us use Taylor’s expansion with a  1 , b  2 .

Then, f ( x, y )  x 2 y  3 y  2 , f (1, 2)  10
f x ( x, y )  2 xy , f x (1, 2)  4
f y ( x, y)  x2  3 , f y (1,  2)  4
f xx ( x, y )  2 y , f xx (1, 2)  4
f xy ( x , y )  2 x , f xy (1,  2)  2
f yy ( x , y )  0 , f yy (1,  2)  0
f xxx ( x, y )  0 , f xxx (1, 2)  0
f yyy ( x , y )  0 , f yyy (1,  2)  0
f yxx ( x , y )  2 , f yxx (1,  2)  2
f xxy ( x , y )  2 , f xxy (1,  2)  2 .
All higher derivatives are zero. Thus, we have

x 2 y  3 y  2  10  4( x  1)  4( y  2)  2( x  1) 2  2( x  1)( y  2)  ( x  1) 2 ( y  2) .
4.3 Explicit and Implicit Functions

The explicit function is one which is given in the independent variable. On the other hand,
implicit functions are usually given in terms of both dependent and independent variables. Here we read
in details:
Explicit function
If we consider set of n independent variables x1 , x2 , x3 ......, xn and one dependent variable u , the
equation
u  f  x1 , x2 , x3......, xn  (*)
denotes the functional relation. In this case if y1 , y2 , y3 ......, yn are the n arbitrarily assigned values of the
independent variables, the corresponding values of the dependent variable u are determined by the
functional relation.
134 Taylor Theorem
The function represented by equation * is an Explicit function but where several variables are
involved, then it is difficult to express one variable explicitly in terms of the others. Thus most of the
functions of more than one variable are implicit function, that is to say we are given a functional relation
  x1 , x2 , x3 ......, xn   0
connecting the n variables x1 , x2 , x3 ......, xn and is not in general possible to solve this equation to find an
explicit function which expresses one of these variables say x1 , in terms of the other n  1 variables.
Implicit function
Let F(x1 , x 2 ,..., x n , u)  0 (1)
be a functional relation between the n  1 variables x1 , x 2 ,..., x n , u and let x1  a1 , x 2  a 2 ,..., x n  a n be

a set of values such that the equation.
F(a1 , a 2 ,..., a n , u)  0 (2)
is satisfied for at least one value of u, that is equation (2) in u has at least one root. We may consider u as
a function of the x 's : u   (x1 , x 2 ,..., x n ) defined in a certain domain, where  (x1 , x 2 ,..., x n ) has
assigned to it at any point (x1 , x 2 ,..., x n ) the roots u of the equation (1) at this point. We say that u is the
implicit function defined by (1). It is, in general, a many valued function.
More generally, consider the set of equations
Fp (x 1 , x 2 ,..., x n , u 1 ,..., u m )  0 (p  1, 2,..., m) (3)
between the n  m variables x1 ,..., x n , u1 ,..., u m and suppose that the set of equations (3) are such that
there are points (x1 , x 2 ,..., x n ) for which these m equations are satisfied for at least one set of values
u1 , u 2 ,..., u m We may consider the u’s as function of x’s.
u p   p (x 1 , x 2 ,..., x n ) (p  1, 2,..., m)
where the function ϕ have assigned to them at the point (x1 , x 2 ,..., x n ) the values of the roots
u1 , u 2 ,..., u m at this point. We say that u1 , u 2 ,..., u m constitute a system of implicit functions defined by
the set of equation (3). These functions are in general many valued.
Definition 1 (Implicit function of two variables). Let f ( x, y ) be a function of two variables and
y   ( x) be a function of x such that for every value of x for which  ( x) is defined, f ( x,  ( x))
vanishes identically i.e., y   ( x) is a root of the functional equation f ( x, y)  0 . Then, y   ( x) is an
implicit function defined by the functional equation f ( x, y)  0 .
4.3.1 Implicit function theorem.

This theorem tells us that whenever we can solve the approximating linear equation for y as a function
of x, then the original equation defines y implicitly as a function of x. This theorem also known as
Existence theorem.
Theorem 1 (Implicit function theorem). Let F(u, x, y) be a continuous function of variables u, x, y.
Suppose that
(i) F (u0 , a, b)  0;
(ii) F(u, a, b) is differentiable at (u0 , a, b);
F
(iii) The partial derivative (u0 , a, b)  0.
. u
Then there exists at least one function u = u(x, y) reducing to u 0 at the point (a, b) and which, in the
neighbourhood of this point, satisfies the equation F(u, x, y) = 0 identically.
Also, every function u which possesses these two properties is continuous and differentiable at the point
(a, b).
F
Proof. Since F(u 0 , a, b)  0 and(u 0 , a, b)  0, the function F is either an increasing or decreasing
u
function of u when u  u 0 . Thus there exists a positive number δ such that F(u 0   , a, b) and
F(u 0   , a, b) have opposite signs. Since F is given to be continuous, a positive number η can be found
so that the functions
F(u 0   , x, y) and F(u 0   , x, y)
the values of which may be as near as we please to

F(u 0   , a, b) and F(u 0   , a, b)
will also have opposite signs so long as x  a   and y  b  .
Let x, y be any two values satisfying the above conditions. Then F(u, x, y) is a continuous function of u
which changes sign between u 0   and u 0   and so vanishes somewhere in this interval. Thus for
these x and y there is a u in [u 0   , u 0   ] for which F(u, x, y) = 0. Thus u is a function of x and y, say
u(x, y) which reduces to u 0 at the point (a, b).
Suppose that u, x, y are the increments of such function u and of the variables x and y measured
from the point (a, b). Since F is differentiable at (u 0 , a, b) we have
F  [ Fu (u0 , a, b) ]u  [ Fx (u0 , a, b) ' ]x  [ Fy (u0 , a, b)  "]y  0.
Since F  0 because of F = 0. The numbers ,' ," tend to zero with u, x & y and can be made
as small as we please with  &  . Let  and  be so small that the numbers ,' ," are all less than
1
Fu  u0 , a, b  , which is not zero by our hypothesis. The above equation then shows that u  0 as
2
x  0 and y  0 which means that the function u  u  x, y  is continuous at (a, b).
136 Taylor Theorem
Moreover, we have
 Fx  u0 , a, b    ' x   Fy  u0 , a, b    " y
u   
Fu  u0 , a, b   
Fx u 0 , a, b  Fy u 0 , a, b 
 x  y  1 x 2 y ,
Fu u 0 , a, b  Fu u 0 , a, b 
1 and 2 tending to zero as x and y tend to zero.

Hence u is differentiable at (a, b).
F
Corollary 1. If exists and is not zero in the neighbourhood of the point u 0 , a, b  , the solution u of
u
the equation F = 0 is unique. Suppose that there are two solutions u1 and u 2 . Then we should have, by
mean value theorem, for u1  u '  u 2
0  F  u1 , x, y   F  u2 , x, y    u1  u2  Fu  u ', x, y  ,
and so Fu u, x, y  would vanish at some point in the neighbourhood of u 0 , a, b  which is contrary to
our hypothesis.
Corollary 2. If F u, x, y  is differentiable in the neighbourhood of u 0 , a, b  , the function u  u  x, y 
is differentiable in the neighbourhood of the point (a, b).
This is immediate, because the preceding proof is then application at every point u , x, y  in that
neighbourhood.
4.3.2 Inverse function theorem.
Corollary 1 is of great importance, for a function of two variables only, F u, x   0 and taking
F u, x   f u   x , we can express the fundamental theorem on inverse functions as follows:
Theorem 1 (Inverse function theorem). If, in the neighbourhood of u  u 0 , the function f u  is a

continuous function of u and if
(i) f u 0   a
(ii) f ' u   0
in the neighbourhood of the point u  u 0 , then there exists a unique continuous function u    x  ,
which is equal to u 0 when x = a, and which satisfies identically the equation
f u   x  0 ,
in the neighbourhood at the point x = a.
The function u    x  thus defined is called the inverse function of x  f u  .
4.4 Higher Order Differentials
Let z  f ( x, y ) be a function of two independent variables x and y defined in a certain domain
and let it be differentiable at the point ( x, y) of the domain. The first differential coefficient of z at the
point ( x, y) is defined as
z z
dz  dx  dy (1)
x y
z z
and if and are differentiable at the point ( x, y) , then the differential coefficient of dz is called
x y
second differential coefficient of z and is denoted by d 2 z and is given by
 z z 
d 2 z  d  dx  dy 
 x y 
 z   z 
 d   dx  d   dy (2)
 x   y 
 z   z 2 z
2
Now, d    2 dx  dy
 x  x yx
 z   2 z 2 z
and d  dx  2 dy.
 y  xy y .
Putting these values in (2), we get
2 z 2 z 2 z
  2 
dy 
2 2
d 2z  dx  2 dxdy 
x 2
yx y
2
  
Thus, d z   dx  dy  z
2
 x y 
3
  
Similarly, d z   dx  dy  z
3
 x y 
Proceeding in this manner, we define the successive differential coefficients d 4 z , d 5 z ,.......... .
Thus, the differential coefficient of nth order d z exists if d n 1 z is differentiable i.e. if all the partial
n
derivatives of (n-1)th order are differentiable. Thus, by mathematical induction, we have
n z n z n( n  1)  n z n z
    n 
dy 
n n 1 n2 2 n
dnz  dx  n n 1
dx dy  n2
dx dy  ........ 
x n
x y 2! x y 2
y
138 Taylor Theorem
n
  
  dx  dy  z .
 x y 
4.4.1. Choice of independent variables
Let F ( x, y, z )  0 (1)
Differentiate (1), we get
F F F
dx  dy  dz  0 (2)
x y z
Now, if z is dependent on the two independent variables x and y in such a way that the equation
F ( x, y, z)  0 is satisfied by z  z ( x, y ) , then
z z
dz  dx  dy (3)
x y
Now, equation (2) can be written as
Fx F
dz   dx  y dy (4)
Fz Fz
Comparing (3) and (4), we get
z F z F
 x ,  y
x Fz y Fz
Similarly, if x is dependent on y and z then
x Fy x F
 ,  z
y Fx z Fx
Similarly, if y is dependent on z and x , then
y F y F
 x ,  z .
x Fy z Fy
4.4.2 Higher order derivatives of implicit functions

Let f ( x, y, z)  0 be a functional relation where z is dependent variable such that z  z( x, y) .
z z  2 z  2 z  2 z
We denote the partial derivatives , , , , by p, q, r, s, t respectively.
x y x 2 xy y 2
Now, we suppose that x is dependent variable so that x  x( y, z) . Then, we will show that how to
express partial derivatives of first and second order w.r.t. y and z in terms of p, q, r, s and t.
Since z  z( x, y)
z z
 dz  dx  dy. (1)
x y
Now, we differentiate (1), taking x as dependent variable, dy and dz as constant so that
 z  z  z 
 0  d   dx  d 2 x  d   dy
 x  x  y 
2 z 2 z z 2 2 z 2 z
  2 
dy 
2 2
 dx  dydx  d x  dxdy 
x 2
xy x xy y
 r  dx   2 sdx.dy  t  dy   pd 2 x.
2 2
(2)
Now, from (1)

dz  pdx  qdy (3)
1
 dx   dz  qdy  (4)
p
Now, putting the value of dx in (2), we get

2
1  1 
0  r   dz  qdy    2s   dz  qdy   dy  t  dy   pd 2 x
2
p  p 
  pd 2 x  r.
1 
p 2 
2

1
p
 2 


dz   q 2  dy   2qdz.dy   2s  dzdy  q  dy    t  dy 
2 2
r  rq 2 2sq   2s 2qr 
2    2   t   dy     2  dzdy
2 2
 dz 
p  p p   p p 
r
 2  dz  
 rq 2  2spq  tp 2   2sp  2qr  dzdy
 dy  
2 2
2
p p p2
r
d x   3  dz  
 2 pqs  rq 2  tp 2   2qr  2sp  dzdy.
 
2 2

2
dy  (5)
p p3 p3
From (4), we have
x Coefficient of 1
 dz in (4)  ;
z p
x q
 Coefficient of dy in (4)   .
y p
From (5), we have

140 Taylor Theorem
2 x r
 Coefficient of  dz  in (5)   3
2
z 2
p
2 x 2 pqs  rq 2  p 2t
 
2
 Coefficient of dy in (5) 
y 2 p3
2 x 1 1  2qr  2 sp 
 Coefficient of dydz in (5) 
yz 2 2 p3
qr  sp
 .
p3
4.5 Change of Variables

In problems involving change of variables it is frequently required to transform a particular
expression involving a combination of derivatives with respect to a set of variables, in term of
derivatives with respect to another set of variables.
2w 2w
Example1. Let w be a function of two variables x and y , then transform the expression  by
x 2 y 2
the formula of polar transformation x  u cos v , y  u sin v .
Solution. Here, x  x(u, v)

x x
dx  du  dv
u v
 cos v.du  u sin v.dv (1)
Since y  y(u, v)
y y
dy  du  dv
u v
 sin v.du  u cos v.dv (2)
Multiplying (1) by cos v and (2) by sin v and adding, we get
du  cos v(dx)  sin v(dy) (3)
Multiplying (1) by sin v and (2) by cos v and subtracting, we get

sin v cos v
dv   (dx)   dy  (4)
u u
From (3) and (4), we get
u u
 cos v ,  sin v
x y
v sin v v cos v
 , 
x u y u
w w u w v
Now,  .  .
x u x v x
w  sin v  w
 cos v  
u  u  v
  sin v  
  cos v  w (5)
 u u v 
w   cos v  
Similarly   sin v  w (6)
y  u u v 
 2 w   w 
Now   
x 2 x  x 
  sin v    w sin v w 
  cos v    cos v  
 u u v   u u v 
 2 w sin v cos v  2 w sin v cos v w sin v cos v  2 w

 cos 2 v   
u 2 u uv u2 v u uv
sin 2 v w sin v cos v w sin 2 v  2 w
   2 (7)
u u u2 v u v 2
 2 w 2sin v cos v  2 w sin 2 v  2 w sin 2 v w 2sin v cos w
 cos 2 v 2   2  
u u uv u uv u u u2 u
2w  2 w sin v cos v  2 w sin v cos v w cos 2 v w
Similarly  sin v 2 
2
 
y 2 u u uv u2 v u u
sin v cos v  2 w cos v sin v w cos 2 v  2 w
   (8)
u u v u2 v u 2 v 2
Adding (7) and (8), we get
 2 w  2 w  2 w 1 w 1  2 w
    .
x 2 y 2 u 2 u u u 2 v 2
Example 2. Transform the expression
 z z 
2

2  z 
2
 z  
2
 x     a  x  y       
2 2
 x y   x   y  
by the substitution x  r cos  , y  r sin .
Solution. We wish to express z as a function of x and y where x and y are the functions of r and
 i.e. z  z( x, y) and given x  x(r, ) , y  y(r, ) .
142 Taylor Theorem
x x
dx  dr  d
r 
 cos dr  ( sin  .r)d
 cos  dr  sin  .rd (1)
y y
Similarly dy  dr  d
r 
 sin  dr  r cos  d (2)
Multiplying (1) by cos  and (2) by sin  and adding, we get
dr  cos dx  sin  dy (3)
Multiplying (1) by sin  and (2) by cos  and subtracting, we get

cos  sin 
d  dy  dx (4)
r r
r r
 cos  ,  sin 
x y
 cos   sin 
 , 
y r x r
z z r z 
Now,  
x r x  x
z sin  z
 cos   (5)
r r 
2 2
 z   z sin  z 
    cos   
 x   r r  
2 2
 z  sin   z  2sin  cos   z
2 2
 cos     2   
2
(6)
 r  r    r r
z z r z 
Similarly  
y r y  y
z cos  z
 sin   (7)
r r 
2 2 2
 z   z  cos   z  2sin  cos   z
2 2
   sin 2
       (8)
 y   r  r 2    r r
Adding (6) and (8), we get
2 2 2 2
 z   z   z  1  z 
       2  
 x   y   r  r   
Multiplying (a 2  r 2 ) on both sides,
 z   z  
2
 z  1  z  
2 2 2
( a  r )         ( a 2  r 2 )    2 
2 2
  (9)
 x   y    r  r    
Multiplying (5) by x and (7) by y and adding,
z z z 1 z
x y   x cos   y sin     y cos   x sin  
x y r r 
z r z
  r cos 2   r sin 2     sin  cos   cos  sin  
r r 
z
r
r
Squaring on both sides,
2 2
 z z  2  z 
x  y  r   (10)
 x y   r 
Adding (9) and (10), we get required result
 z z 
2

2  z 
2
 z  
2

2  z 
2
1  z    z 
2 2
 x  y   ( a  r )        a     2   
2

 x y   x   y    r  r       
 r 2
 x2  y2 
Example 3. If x  r cos  , y  r sin then prove that
  2u  2u   2u  2u u  2u
x 2
 y 2   2  2   4 xy  r2 2  r  2
r 
 x y  xy r
where u is any twice differentiable function of x and y .
Solution. Here, x  x(r, )

x x
dx  dr  d
r 
 cos  dr  r sin  d (1)
Since y  y(r, )
144 Taylor Theorem
y y
dy  dr  d
r 
 sin  dr  r cos  d (2)

sin  cos 
d   dx  dy (4)
r r
r r
 cos  ,  sin 
x y
 sin   cos 
 , 
x r y r
u u r u 
Now,  
x r x  x
u u sin 
 cos  
r  r
  sin   
  cos   u (5)
 r r  
u   cos   
Similarly   sin   u (6)
y  r r  
 2u   u 
Now,   
x 2 x  x 
  sin     u sin  u 
  cos     cos   
 r r    r r  
 2u sin  cos   2u sin  cos  u sin  cos   2u

 cos 2    
r 2 r r  r2  r r 
sin 2  u sin  cos  u sin 2   2u

   (7)
r r r2  r 2  2
 2u  2u sin  cos   2u sin  cos  u cos 2  u

Similarly  sin 2
   
y 2 r 2 r r  r2  r r
sin  cos   2u cos  sin  u cos 2   2u

   (8)
r r  r2  r 2  2
Subtracting (8) from (7), we get
  2u  2u   2u sin 2  2u sin 2 u sin 2  2u
 2  2   cos 2 2   2 
 x y  r r r r  r r
cos 2 u sin 2 u cos 2  2u

  2 
r r r  r 2  2
and we have
x 2
 y 2   r 2  cos 2   sin 2    r 2 cos 2
  2 u  2u  2    2u  2u  2sin 2  2u 2sin 2 u cos 2 u 

 x  y   x2  y 2   r cos 2 cos 2  r 2   2   r r  r 2   r r 
2 2
     
(9)
 2u   u 
Now,   
xy x  y 
  sin     u cos  u 
  cos     sin   
 r r    r r  
 2u cos 2   2u cos 2  u sin  cos  u

 cos  sin    
r 2 r r  r 2  r r
sin 2   2u sin  cos   2u sin 2  u
   2
r r r2  2 r 
 2u   2u 1  2u cos2  u
4 xy  2r 2 sin 2  sin  cos  2  cos 2  2
xy  r r r r 
sin  cos  u sin  cos   2u sin 2  u 

   2  (10)
r r r2  2 r  
Adding (9) and (10), we get required result.
Example 4. If x  r cos  , y  r sin , then show that
 2
 r 2 cos 2 .
xy
Solution. Here, x  x(r, )
x x
dx  dr  d
r 
146 Taylor Theorem
 cos  dr  r sin  d (1)
Since y  y(r, )
y y
dy  dr  d
r 
 sin  dr  r cos  d (2)
sin  cos 
d   dx  dy (4)
r r
r r
 cos  ,  sin 
x y
 sin   cos 
 , 
x r y r
From (4), rd   sin  dx  cos dy.
Now differentiating, we get
drd  rd 2   cos ddx  sin ddy
 cosdx  sin dy d
rd 2  cos dx  sin dy d  drd 
 cos dx  sin dy d  cos dx  sin dy d
 2cos dx  sin dy d
 sin  cos  
 2cos dx  sin dy   dx  dy 
 r r 
2
d 2   cosdx  sin dy  sin dx  cosdy 
r2
2

r 2 
 sin  cos  dx 2  cos 2 dxdy  sin  cos  dy 2 .
 2 2  2  2 2
As d 2  dx  2 dxdy  dy , so we get
x2 xy y 2
 2
 r 2 cos 2 .
xy
4.6 Extreme Values of Explicit Functions
We now investigate the theory of extreme values for explicit functions of more than one variable.
Definition 1. Let u  f ( x, y) be the equation which defines u as a function of two independent variables
x and y . Then, the function u  f ( x, y) has an extreme value at the point (a, b) if the increment
f  f (a  h, b  k )  f (a, b) preserves the same sign for all values of h and k such that h   , k  
where  is a sufficiently small positive number. If f is negative then the value is maximum and if f
is positive then the value is minimum.
Necessary condition for extreme value
The necessary condition that f (a, b) should be an extreme value is that both f x ( a , b ) and f y ( a , b ) are
zero. Values of ( x, y) at which df  0 are called stationary values.
Or A necessary condition for f ( x, y) to have an extreme value at (a, b) is that f x (a, b)  0 , f y ( a , b )  0

provided that these partial derivatives exist.
If f (a, b) is an extreme value of the function f ( x, y) of two variables then it must also be an extreme
value of both the functions f ( x, b) and f (a, y) of one variable.
But the necessary condition that these have extreme values at x  a and y  b respectively is
f x (a, b)  0 and f y ( a , b )  0 .
Sufficient condition for extreme value
 
2
The value f (a, b) is an extreme value of f ( x, y) if f x (a, b)  0 , f y ( a , b )  0 and also f xx . f yy  f xy
and the value is maximum or minimum according as f xx or f yy is negative or positive respectively.
Here, A  f xx , C  f yy , B  f xy
(i) If AC  B  0 , then f (a, b) is a maximum value if A  0 and a minimum value if A  0 .

2
(ii) If AC  B  0 , then f (a, b) is not an extreme value.

2
(iii) If AC  B  0 , this is doubtful case, in which the sign of f (a  h, b  k )  f (a, b) depends on h

2
and k and requires further investigation.

Example 1. Find the extreme value of the function f ( x, y )  x 2  xy  y 2  3 x  2 y  1 .
Solution. Here, f ( x, y )  x 2  xy  y 2  3 x  2 y  1
 fx  2x  y  3 , f xx  2
148 Taylor Theorem
fy  x  2y  2 , f yy  2 , f xy   1 .
For extreme values, f x  0 , fy  0
 2x  y  3 and x  2 y  2
1 4
y  , x  
3 3
 4 1
Thus, the extreme point is   ,  .
 3 3
 4 1
At   ,  , A  f xx  2 , B  f xy   1 , C  f xy  2
 3 3
Now, AC  B2  4 1  3  0 and A  2  0
 4 1  4 1
   ,  is a point of minimum and minimum value  f   , 
 3 3  3 3
16 4 1 2 4
    4  1   .
9 9 9 3 3
Example 2. Show that f ( x, y )  2 x 4  3 x 2 y  y 2 has neither maximum nor minimum at (0, 0).
Solution. Here, f ( x, y )  2 x 4  3 x 2 y  y 2
 f x  8 x 3  6 xy , f xx  24 x 2  6 y
f y  3x2  2 y , f yy  2 , f xy   6 x
For extreme values, f x  0 , fy  0
8 x 3  6 xy  0 and 3 x 2  2 y  0
3x 2
2 x  4 x 2  3 y   0 and y 
2
4x2
 x  0 or y 
3
If x  0  y  0
4x2 3x 2 4 x 2 3x 2
If y  and y    , which is not possible.
3 2 3 2
So, stationary point is (0,0).
Now, A  f xx (0, 0)  0 , B  f xy (0, 0)  0 , C  f yy (0, 0)  2
 AC  B2  0  0  0
So, doubtful case and further investigation is required.
Now, f  f (0  h,0  k )  f (0,0)
 f (h, k )  f (0,0)
 2h 4  3h 2 k  k 2  (2h 2  k )( h 2  k )
k
If h  k  0 and 2h  k  0 i.e. h  k and h2  then f  0 .
2 2 2
2
k
If h  k  0 and 2h  k  0 i.e. h  k and h2  then f  0 .
2 2 2
2
So, for different values of h and k , f does not have the same sign. Hence, f has neither maximum
nor minimum at (0, 0).
Example 3. Find the extreme value of x 3  3axy  y 3 ; a  0. .
Solution. Here, f ( x, y )  x 3  3axy  y 3
 f x  3 x 2  3ay , f xx  6 x
f y  3 y 2  3ax , f yy  6 y , f xy   3a
For extreme value, we put f x  0 , f y  0
3 x 2  3ay  0 and 3 y 2  3ax  0
x2 y2
y and x 
a a
After solving, the stationary points are (0, 0) and (a, a) .
Now, A  f xx (0, 0)  0 , B  f xy (0, 0)   3a , C  f yy (0, 0)  0
AC  B2  9a2  0 at (0,0).
So, given function has no extreme value at (0,0).
Now, A  f xx ( a , a )  6 a , B  f xy ( a , a )   3a , C  f yy ( a , a )  6 a
AC  B2  36a2  9a2  27a2  0 at (a, a) .
& A  6a  0
Hence, the given function has minimum value at (a, a) .

150 Taylor Theorem
a3 a3
Example 4. Let u  xy   ,
x y
u a3
 y 2
x x
u a3
 x 2 .
y y
Putting x= a, y = a
 2u 2a3  2u  2u 2a3
Hence 2  3  2,  1,  3  2.
x x xy y 2 y
Therefore r and t are positive when x = a = y and rt- = 2.2-1= 3 (positive). Therefore, there is a
minimum value of u viz. u = 3 .
Example 5. Let
, .
It can be verified that
0, 0 0, 0, 0 0
0, 0 0, 0, 0 2
0, 0 0.
So at the origin, we have
However, on writing
It is clear that f(x, y) has a minimum value at the origin, since
Δ , 0, 0
is greater than zero for all values of h and k.
4.7 Stationary Values of Implicit Functions

To find the stationary values of the function
f  x1 , x 2 ,......... , x n , u1 , u 2 ,......... ....., u m  (1)
of n  m variables which are connected by m differentiable equations

r  x1 , x2 ,............., xn , u1 , u2 ,.................., um   0; r  1, 2,................, m (2)
If the m variables u1 , u 2 ,......... ...., u m are determinate as functions of x1 , x 2 ,......... .., x n from the system of
m equations of (2), then f can be regarded as a function of n independent variables x1 , x 2 ,........, x n .
At a stationary point of f, df  0 .
Hence at a stationary point, 0  df  f x dx 1  f x dx 2  .........  f x dx n  f u du 1  .......... ....  f u du m

1 2 n 1 m
(3)
Again differentiating the equation (2), we get
1    
dx1  .........  1 dx n  1 du1  ............  1 du m  0
x1 x n u1 u m

 2  2  2  2 
dx1  .........  dx n  du1  ..........  du m  0 
x1 x n u1 u m 
.....................................................................................  (4)
.................................................................................... 

 m  m  m  m 
dx1  ........  dx n  du1  ..........  du m  0 
x1 x n u1 u m


From these m equations of (4), the differentials du1 , du 2 ,......... ......, du m of the m dependent variables
may be found in terms of the n differentials dx1 , dx 2 ,........., dx n and are substituted in (3). This way df
has been expressed in terms of the differentials of the independent variables, and since the differentials
of the independent variables are arbitrary and df  0 , the coefficients of each of these n differentials may
be equated to zero. These n equations together with the m equations of (2) constitute a system of n  m 
equations to determine the n  m  coordinates of the stationary points of f.
Example 1. F  x, y, z  is a function subject to the constraint condition G  x, y, z   0 . Show that at a

stationary point.
Fx G y  Fy G x  0.
Solution. We may consider z as a function of the independent variables x, y.

At a stationary point, dF  0
 0  dF  Fx dx  Fy dy  Fz dz. (1)
Differentiating the relation G x, y, z   0 ,we get
G x dx  G y dy  G z dz  0. (2)
Putting the values of dz from (2) into (1), or what is same thing, eliminating dz from (1) and (2), we get
152 Taylor Theorem
F x G z  G x F z dx  F y G z  G y F z dy  0
Since dx, dy (being differentials of independent variables) are arbitrary, therefore
FxGz  Gx Fz  0
Fy Gz  Gy Fz  0
which gives Fx G y  G x Fy  0.
4.8 Lagrange Multipliers Method

In this method, we discuss the determination of stationary points from a modified point of view. This
process consists in the introduction of undetermined multipliers, a method due to Lagrange. After his
name, this method also called Lagrange’s method of undetermined multipliers.
Let ϕ x , x , … , x be a function of n variables which are connected by m equations
x ,x ,…,x 0, x ,x ,…,x 0, … , x ,x ,…,x 0,
So that only n-m variables are independent.
When u is maximum or minimum
x x x ⋯ x 0
Also x x x ⋯ x 0
x x x ⋯ x 0
…………………………………………………………………….
……………………………………………………………………..
x x x ⋯ x 0
Multiplying all these respectively by 1, , ,…, and adding, we get a result which may be written
x x x ⋯ x 0,
Where ⋯
The m quantities , ,…, are of our choice. Let us choose them so as to satisfy the m linear
equations
P1 = P2 = ………….= Pm.
The above equation is now reduced to
⋯ 0
It is indifferent which n-m of the n variables are regarded as independent. Let them be
, , … , . Then since n-m quantities , ,… are all independent, their
coefficients must be separately zero. Thus we obtain the additional n-m equations
Pm+1 = Pm+2 = ………….= Pn = 0.
Thus the m+n equations f1 = f2 = ………….= fm=0 and P1 = P2 = ………….= Pn=0 determine the m
multipliers , , … , and values of n variables x , x , … , x for which maximum and minimum
values of u are possible.
Example 1. Find the length of the axes of the section of the ellipsoid 1 by the plane
0.
Solution. We have to find the extreme values of the function where , subject to
the equations of the condition
1 0,

0.
Then 0 (1)
0, (2)

0 (3)
Multiplying these equations by 1, , and adding we get
x
x  1  2l  0 (4)
a2
y
y  1  2 m  0 (5)
b2
z
z  1  2 n  0 (6)
c2
Multiplying (4), (5) and (6) by x, y, z and adding we get
or 0 ⟹ .
From (4), (5) and (6), we have
, ,
154 Taylor Theorem
 l 2a2 m 2b 2 n2c 2 
But 0 2      0 and since 2  0 the equation giving the
r a r 2  b2 r 2  c2 
2 2
values of which are the squares of the length of semi-axes required (quadratic in r2) is
 l 2a2 m 2b 2 n2c 2 
 2    0.
r a r 2  b2 r 2  c2 
2
Example 2. Investigate the maximum and minimum radii vector of the sector of “surface of elasticity”
(x2+y2+z2)2=a2x2+y2b2+z2c2 made by the plane lx + my + nz = 0.
Solution. We have
xdx  ydy  zdz  0 (1)
a 2 xdx  b 2 ydy  c 2 zdz  0 (2)
ldx  mdy  ndz  0 (3)
Multiplying these equations by 1, 12 and adding we get
x  a 2 x1  l 2  0 (4)
y  b 2 y1  m2  0 (5)
z  c 2 z1  n2  0 (6)
Multiplying(4), (5) and (6) by x, y, z respectively and adding we get
x 2
 y 2  z 2   1  a 2 x 2  y 2b 2  z 2 c 2   2  lx  my  nz   0
1
 r 2  1r 4  0  1  
r2
2lr 2 2 mr 2 2 nr 2
x , y , z .
a2  r 2 b2  r 2 c2  r 2
2l 2 r 2 2 m 2 r 2 2 n 2 r 2
Then lx + my + nz = 0     0.
a2  r 2 b2  r 2 c2  r 2
l2 m2 n2
 2  
r  a 2 r 2  b2 r 2  c 2
It is quadratic in r2 and give its required values.
Example 3. Prove that the volume of the greatest rectangular parallelepiped that can be inscribed in the
1 1 1
  ,   
ellipsoid          .
  ,  ,  
    
Solution. Volume of the parallelepiped . Its maximum value is to find under the condition that it
2 2 2
x y z
is inscribed in the ellipsoid 2  2  2  1 , we have
a b c
Therefore
(1)
2x 2y 2z
df1  2
dx  2 dy  2 dz  0 (2)
a b c
Multiplying (1) by 1 and (2) by and adding we get
x
yz   0 (3)
a2
y
zx  2   0 (4)
b
z
xy   0 (5)
c2
From (3), (4) and (5), we get
a 2 yz b2 zx c 2 xy
  
x y z
a 2 yz b 2 zx c 2 xy
and so  
x y z
Dividing throughout by xyz we get
a 2 b2 c 2  x2 y 2 z 2 
   2  2  2  1 .
x2 y 2 z 2  a b c 
3x 2 a b c
Hence 2
 1 or x  .Similarly y  , z
x 3 3 3
8abc
It follows therefore that u  8 xyz 
3 3
Example 4. Find the point of the circle x2 +y2 + z2 = 1, lx  my  nz  0 at which the function
u = ax2 + by2 + cz2 + 2fyz + 2gzx + 2hxy attains its greatest and least value.
Solution. We have
u = ax2 + by2 + cz2 + 2fyz + 2gzx + 2hxy
f1  lx  my  nz
156 Taylor Theorem
f2  x2  y 2  z 2  1
Then
axdx  bydy  czdz  fydz  fzdy  ...  gzdx  gxdz  hxdy  hydx
Multiplying these equations by 1, 1 , 2 and adding, we get

ax  hy  gz  1l  2 x  0
.
Multiplying by x, y, z and adding we get
= -u.
Putting all the values in the above equation we have
hx  y (b  u )  fz  m1  0
.
Eliminating x, y, z and ,we get
= 0.
Example 5. If a, b, c are positive and
Show that a stationary value of u is given by
where is the +ve root of the cubic
Solution. We have
(1)
(2)
Differentiating (1) we get
1  b2 c 2 
 x3  z 2  y 2  dx  0
 
which on multiplication with yields
1
 x b 2

y 2  c 2 z 2 dx  0 (3)
Differentiating (2) we have
(4)
Using Lagrange’s multiplier, we obtain
i.e. (5)
(6)
(7)
Then (6) + (7) - (5) yields
= -
=  (1  2ax 2 ) (By (2))
Therefore
Similarly and .
Substituting these values of in (2), we obtain
which equals to
(8)
Since a, b, c are positive, any one of (5), (6), (7) shows that must be positive. Hence is a positive
root of (8).
158 Taylor Theorem
4.9 Jacobian and its Properties
In this section, we give definition of Jacobian and discuss its properties.
4.9.1 Jacobian
If be n differentiable functions of the n variables then the determinant
 u1 u1 u1 

 , ,..., 
 x1 x2 xn 
 u u2 u2 
 2 , , , 
 x1 x2 xn 
 u un un 
 n , , , 
 x1 x2 xn 
is called the Jacobian of u1 , u2 , , un with regard to x1 , x2 , , xn . The determinant is often denoted by
 (u1 , u2 ,....., un ) (u , u ,....., un )

or J 1 2
 ( x1 , x2 ,....., xn ) ( x1 , x2 ,....., xn )
or shortly J, when there can be no doubt as to the variables referred to.

Theorem 1. If u1 , u2 , , un be n differentiable functions of the n independent variables x1 , x2 , , xn and
there exists an identical differentiable functional relation which does not involve
the x’s explicitly, then the Jacobian
vanishes identically provided that as a function of the u’s has no stationary values in the domain
considered.
Proof. Since
We have
(1)
But
(2)
On substituting these values in (1) we get an equation of the form

(3)
And since are arbitrary differentials of independent variables, it follows that
In other words
(4)
And since by the hypothesis, we cannot have
On eliminating the partial derivatives of from the set of equation (4) we get
which establishes the theorem.

Theorem 2. If be n functions of n variables say
and if then if all differential coefficients
concerned are continuous, there exists a functional relation connecting some or all of the variables
which is independent of .
Proof. First we prove the theorem when n=2. We have and
If v does not depend on y, then and so either or else In the former case u and v are
the functions of x only, and the functional relation sought is obtained from
By regarding x as a function of v and substituting in In the latter case v is constant, and the
functional relation is v = a. If v does depend on y, since the equation v = g(x, y) defines y as a
function of x and v, say
And on substituting in the other equation we get an equation of the form

.
160 Taylor Theorem
(The function F[x, g(x, y)] is the same function of x and y as f(x, y))
Then
(obtained on multiplying the second row by and subtracting from the first ) and so, either
which is contrary to hypothesis or else so that F is a function of v only; hence the functional
relation is
u = F(v)
Now assume that the theorem holds for n  1.
Now u n must involve one of the variables at least, for if not there is a functional relation u n  a. Let one
u n
such variable be called x n since  0 we can solve the equation
x n
u n  f n (x1 , x 2 ,..., x n )
for x n in terms of x1 , x 2 ,..., x n 1 and u n , and on substituting this value in each of the other equations we
get n 1 equations of the form
u r  g r (x1 , x 2 ,..., x n 1 , u n ), (r  1, 2,..., n  1) (1)
If now we substitute f n (x1 , x 2 ,..., x n ) for u n the functions g r (x1 , x 2 ,..., x n 1 , u n ) become
f r (x1 , x 2 ,..., x n 1 , x n ), (r  1, 2,..., n  1)
Then
f1 f1 f
,..., 1
x1 x 2 x n
f 2 f 2 f
,..., 2
0  x1 x 2 x n
...........................
f n f n f
,..., n
x1 x 2 x n
g1 g1 u n g1 g u g u

 . ,...,  1. n , 1. n
x1 u n x1 x n 1 u n x n 1 u n x n
g 2 g 2 u n g 2 g 2 u n g 2 u n
 . ,...,  . , .
 x1 u n x1 x n 1 u n x n 1 u n x n
............................................................................
u n u n u n
, ..., ,
x1 x n 1 x n
g1 g1
,..., , 0
x1 x n 1
g 2 g 2
,..., , 0
 x1 x n 1
...........................
,..., ,
x1 x n 1 x n
by subtracting the elements of the last row multiplied by
g1 g 2 g
, ..., n
from each of the others. Hence
u n  (g1 , g 2 ,..., g n 1 )
.  0.
x n  (x1 , x 2 ,..., x n 1 )
u n  (g1 , g 2 ,..., g n 1 )
Since  0 we must have  0, and so by hypothesis there is a functional relation
x n  (x1 , x 2 ,..., x n 1 )
between g1 , g 2 ,..., g n 1 , that is between u1 , u 2 ,..., u n 1 into which u n may enter, because u n may occur in
set of equation (1) as an auxiliary variable. We have therefore proved by induction that there is a relation
between u1 , u 2 ,..., u n .
4.9.2 Properties of Jacobian
Lemma 1. If U and V are functions of u and v, where u and v are themselves functions of x and y, we
have
 (U, V)  (U, V)  (u, v)
 .
 (x, y)  (u, v)  (x, y)
Proof. Let U = f(u, v), V = F(u, v)

u   (x, y), v   (x, y)
U U u U v
Then  .  .
x u x v x
162 Taylor Theorem
U U u U v
 .  .
y u y v y
V V u V v
 .  .
x u x v x
V V u V v
 .  .
y u y v y
and
U U u u
 (U, V)  (u, v) u v x y
.  
 (u, v)  (x, y) V V v v
u u x y
U u U v U u U v
.  . .  .
u x v x u y v y

V u V v V u V v
.  . .  .
u x v x u y v y
U U
x y  (U, V)
 
V V  (x, y)
x y
The same method of proof applies if there are several functions and the same number of variables.
Lemma 2. If J is the Jacobian of system u, v with regard to x, y and J′ the Jacobian of x, y with regard to
u, v, then J J′ = 1.
Proof. Let u = f(x, y) and v = F(x, y), and suppose that these are solved for x and y giving
x   (u, v) and y   (u, v),
we then have differentiating u = f(x, y) w.r.t u and v; v = F(x, y) w.r.t u and v
u x u y 
1 .  .
x u y u 
 obtained from u = f(x, y)
u x u y 
0 .  .
x v y v 
v x v y 
0 .  .
x u y u 
 obtained from v = F(x, y).
v x v y 
1 .  .
x v y v 
u u x x
x y u v
Also J J  
v v y y
x y u v
u x u y u x u y
.  . .  .
x u y u x v y v

v x v y v x v y
.  . .  .
x u y u x v y v
1 0
 1
0 1
Example 1. If u  x  2y  z, v  x  2y  3z
w  2xy  xz  4yz  2z 2 ,
 (u, v, w)
prove that  0, and find a relation between u, v, w.
 (x, y, z)
Solution. We have
u u u
x y z
 (u, v, w) v v v

 (x, y, z) x y z
w w w
x y z
1 2 1
1 2 3
2y  z 2x  4z  x  4y  4z
1 0 0
1 4 2
2y  z 2x  6z  4y  x  2y  3z
Performing c 2  c 2  2c1 and c3  c3  c1
4 2 0 2
 
2x  6z  4y  x  2y  3z 0  x  2y  3z
 0.
Performing c1  c1  2c 2
Hence a relation between u, v and w exists.
164 Taylor Theorem
Now,
u  v  2x  4z
u  v  4y  2z
w  x(2y  z)  2z(2y  z)
 (x  2z)(2y  z)
 4w  (u  v)(u  v)
 4w  u 2  v2
which is the required relation.

Example 2. Find the condition that the expression px  qy  rz, px  qy  rz are connected with the
expression ax 2  by 2  cz 2  2fyz  2gzx  2hxy, by a functional relation.
Solution. Let
u  px  qy  rz
v  px  q y  rz
w  ax 2  by 2  cz 2  2fyz  2gzx  2hxy
We know that the required condition is

 (u, v, w)
 0.
 (x, y, z)
Therefore
u u u
x y z
v v v
 0.
x y z
w w w
x y z
But
u u u
 p,  q, r
x y z
v v v
 p,  q,  r .
x y z
w
 2ax  2hy  2gz
x
w
 2hx  2by  2fz
y
w
 2gx  2fy  2cz
z
Therefore
p q r
p' q' r' =0
2ax  2hy  2gz 2hx  2by  2fz 2gx  2fy  2cz
p q r p q r p q r
 p' q' r '  0, p ' q' r'  0 , p' q' r'  0
a h g h b f g f c
which is the required condition.

1
Example 3. Prove that if f(0) = 0, f '  x   , then
1  x2
 xy 
f (x)  f (y)  f  .
 1  xy 
Solution. Suppose that

u = f(x) +f(y)
xy
v
1  xy
u u
x y
Now J  u, v  
v v
x y
1 1
1  x2 1  y2
 0
1  y2 1  x2
(1  xy) 2 (1  xy) 2
Therefore u and v are connected by a functional relation

Let u  (v) , that is,
 xy
f (x)  f (y)    
 1  xy 
Putting y = 0, we get
f (x)  f (0)  (x)
 f (x)  0  (x) because f(0) = 0

166 Taylor Theorem
 xy 
Hence f (x)  f (y)  f  .
 1  xy 
Example 4. The roots of the equation in 
  x     y     z   0
3 3 3
  u, v, w  (y  z)(z  x)(x  y)
are u, v, w. Prove that  2 .
  x, y,z  (v  w)(w  u)(u  v)
Solution. Here u, v, w are the roots of the equation

1
 3  (x  y  z) 2  (x 2  y 2  z 2 )  (x 3  y3  z3 )  0
3
1 3
Let x  y  z  , x 2  y 2  z 2   , (x  y3  z 3 )   (1)
3
and then u  v  w  , vw  wu  uv  , uvw   (2)
Then from (1),

1 1 1
 (, ,  )
 2x 2y 2z  2(y  z)(z  x)(x  y) (3)
 (x, y, z)
x2 y2 z 2
Again, from (2), we have

1 1 1
 (, ,  )
 vw wu u  v  (v  w)(w  u)(u  v) (4)
 (u, v, w)
vw wu uv
Then from (3) and (4)

 (u, v, w)  (u, v, w)  (, ,  ) (y  z)(z  x)(x  y)
 .  2
 (x, y, z)  (, ,  )  (x, y, z) (v  w)(w  u)(u  v)
x y z
Example 5. If  , ,  are the roots of the equation    1 in k,
ak bk ck
then
 (x, y, z) (  )(   )(   )
 .
 (, ,  ) (a  b)(b  c)(c  a)
Solution. The equation in k is

k 3  k 2 (a  b  c  x  y  z)  k[ab  bc  ca  x(b  c)  y(c  a)  z(a  b)] (1)
abc  bcx  cay  abz  0.
Now  , ,  are the roots of this equation. Therefore
      (a  b  c)  x  y  z
      ab  bc  ca  x(b  c)  y(c  a)  z(a  b)
and
  abc  bcx  cay  abz
Then, we have
x y z
1  
  
x y z
1  
  
x y z
1  
  
x y z
    (b  c)  (c  a)  (a  b)
  
x y z
    (b  c)  (c  a)  (a  b)
  
x y z
    (b  c)  (c  a)  (a  b)
  
x y z
  bc  ca  ab
  
x y z
  bc  ca  ab
  
x y z
  bc  ca  ab
  
x y z
   1 1 1
x y z
Now, (b  c)  (c  a)  (a  b)
  
bc ca ab
x y z
  
1 1 1
   
  
Hence
 (x, y, z)
(b  c)(c  a)(a  b)  (  )(   )(   )
 (, ,  )
 (x, y, z) (  )(   )(   )
 
 (, ,  ) (b  c)(c  a)(a  b)
168 Taylor Theorem
Second Method. After the equation (1),
let a  b  c  (x  y  z)  
ab  bc  ca  x(b  c)  y(c  a)  z(a  b)  
abc  bcx  cay  abz   (2)

      ,       ,   . (3)
Then
1 1 1
 (, ,  )
 (b  c)  (c  a)  (a  b)  (a  b)(b  c)(c  a) .
 (x, y, z)
bc  ca  ab
and
1 1 1
  ,   
       
  ,  ,  
   
  (  )(   )(   )
 (x, y, z)  (x, y, z)  (, ,  ) (  )(   )(   )

Therefore  .  .
 (, ,  )  (, ,  )  (, ,  ) (a  b)(b  c)(c  a)
Example 6. Prove that the three functions U, V, W are connected by an identical functional relation if
U = x + y – z, V = x – y + z, W = x2 +y2+z2-2yz
and find the functional relation.
Solution. Here
U U U
x y z
 (U, V, W) V V V

 (x, y, z) x y z
W W W
x y z
1 1 1
 1 1 1
2x 2(y  z) 2(z  y)
Performing c3  c3  c 2
1 1 0
 1 1 0 0
2x 2(y  z) 0
Hence there exists some functional relation between U, V and W.

Moreover,
U + V =2x
U – V = 2(y – z)
(U  V)2  (U  V)2  4(x 2  y 2  z 2  2yz)
= 4W
which is the required functional relation.
Example 7. Let V be a function of the two variables x and y. Transform the expression
2 V 2 V

x 2 y 2
by the formulae of plane polar transformation

x  r cos , y  r sin  .
Solution. We are given a function V which is function of x and y and therefore it is a function of r and θ.
From x  r cos , y  r sin  , we have
r  x 2  y 2 ,   tan 1 y / x.
Now
V V r V 
 .  .
x r x  x
V sin  V  r  sin  
 cos    x  cos , x   r 
r r   
V V r V 
And  .  .
y r y  y
V cos  V  r  cos  
 sin     sin ,  
r r   y y r 
   sin   
Therefore   cos   
x  r r  
   cos   
  sin   
y  r r  
 2V   sin     V sin  V 
Hence   cos     cos   
x 2
 r r    r r  
  V sin  V  sin    V sin  V 
 cos   cos     cos   
r  r r   r   r r  
170 Taylor Theorem
  2V sin  V sin   2V 
 cos   cos  2  2  
 r r  r r 
sin    2V V cos  V sin   2V 

  cos   sin    
r  r r r  r  2 
 2V sin  cos   2V sin 2   2V

 cos 2   2 
r 2 r r  r 2  2
sin 2  V 2 sin  cos  V
  (1)
r r r2 
 2V   cos     V cos  V 
and   sin     sin   
y 2
 r r    r r  
  V cos  V  cos    V cos  V 
 sin   sin     sin   
r  r r   r   r r  
  2V cos  V cos   2V 
 sin   sin  2  2  
 r r  r r 
cos    2V V sin  V cos   2V 

  sin   cos    `
r  r r r  r  2 
 2V sin  cos   2V cos  sin  V

 sin 2   
r 2 r r  r2 
cos  sin   2V cos 2   2V cos 2  V
  
r  r r 2  2 r r
sin  cos  V
 (2)
r2 
Adding (1) and (2), we obtain
 2V  2V  2V 1 V 1  2V
   . 
x 2 y 2 r 2 r r r 2  2
which is the required result.
Example 8. Transform the expression
 Z Z 
2

2  Z 
2
 Z  
2
x y   ( a  x  y ) 
2 2
   
 x y   x   y  
by the substitution x  r cos  , y  r sin  .
Solution. If V is a function of x, y, then
V V  x V  y x V y V
   
r x r  y  r r x r y
V V V    
 r x y   x  y V
r x y  x y 
  
 r x y
r x y
  
Similarly x y
 y x
Z Z r Z  Z sin  Z
Now  .  .  cos   (1)
x r x  x r r 
Z Z cos  Z
 sin   (2)
y r r 
2 2 2 2
 Z   Z   Z  1  Z 
Therefore        2 
 x   y   r  r   
and the given expression is equal to
 Z 
2
 Z 2 1  Z 2 
  ( a  r )    2   
2 2
r
 r   r  r    
2 2
 Z   a   Z 
2
 a2    2  1 
  .
 r   r    
Example 9. If x  r cos  , y  r sin  , prove that
  2u  2 u   2u  2u u  2u
( x 2  y 2 )  2  2   4 xy  r2 2  r  2
 x y  xy r r 
where u is any twice differentiable function of x and y.
Solution. We have
u u x  u y
 .  .
r x r y r
u  u x u y  u
 cos   sin   
x  y r x r  y
u u u
 r x y (1)
r x y
  u      u u 
Therefore r  r    x  y  x  y 
r  r   x y  x y 
172 Taylor Theorem
  u u    u u 
x x  y  y x  y 
x  x y  y  x y 
 2u  2u  2u 2  u
2
u u
 x2  xy  xy  y x y
x 2
xy yx y 2
x y
Therefore
 2u u 2  u
2
 2u 2  u
2
u u
 r2  r  x  2 xy  y x y (2)
r 2
r x 2
xy y 2
x y
 2u 2  u
2
 2u 2  u
2
r2
x  2 xy y (using (1))
r 2 x 2 xy y 2
 u u  x  u  y
Again,  .  .
  x   y  
u u
x y
y x
 2u     u u 
Therefore   x  y  x  y 
 2
 y x  y x 
  u u  u  u u 
x x  y  y x  y 
y  y x  x  y x 
 2u  2u 2  u
2
u u
 x2  2 xy  y x y (3)
y 2
yx y 2
x y
From (1), (2) and (3), we get the required result.
4.10 References
1. Walter Rudin, Principles of Mathematical Analysis (3rd edition) McGraw-Hill, Kogakusha,
1976, International Student Edition.
Delhi, 2012.

4 - 03-17-2021 - 11-27-41 - Mathematical Analysis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

4 - 03-17-2021 - 11-27-41 - Mathematical Analysis

Uploaded by

Copyright:

Available Formats

M. Sc.

Paper Code: 20MAT21C2

DIRECTORATE OF DISTANCE EDUCATION

Copyright © 2021, Maharshi Dayanand University, ROHTAK

Definition 2. A partition P* of [ a, b] is said to be finer than P (or a refinement of P) if P*  P , that is ,

P  max xi  max{xi  xi1 : i  1,2,...., n}

We notice that if P*  P , then P*  P . Thus refinement of a partition decreases its norm.

Definition 5. Lower and Upper Riemann-Stieltjes Sums and Integrals

For any real valued bounded function f on [ a, b] , we take

respectively called Lower integral and Upper integrals of f with respect to  .

a case, we write L(P, f ), U ( P, f ),  f ,  f and f  respectively in place of L( P, f , ), U ( P, f , ),

 fd ,  fd and f ( ) .

Proof. Let P  {x 0 , x1 ,......... .......... , x n } be a partition of a, b . Further, let P  be a refinement of P

Also,  is monotonically increasing function and x i 1  x   x i . So,

m b    a   LP, f ,    U P, f ,    M  b    a  .

Since, least upper bound  any upper bound

Since, any lower bound  greatest lower bound.

Example 1. Let  ( x)  x and define f on [0,1] by

mi  0 , M i  1 , because every subinterval [ xi 1 , xi ] contain both rational and irrational number.

Hence, in this case

So, by definition of l.u.b., there exists a partition P1 of a, b such that

By the definition of g.l.b., there exists a partition P2 of a, b such that

Let P  P1  P2 be the common refinement of P1 and P2 , so that

The condition is sufficient:

Now, we know that

From (7) and (8) , we have

(ii) to every  0 there corresponds a   0 such that

 f (ti )i   fd 

for every partition P of [ a, b] with P   and for all ti  [ xi 1 , xi ] .

which is possible by monotonicity of  on [ a, b] . Also f is continuous on compact set [ a, b] .

Hence f is uniformly continuous on [ a, b] . Therefore there exists a   0 such that

f ( x)  f (t )   whenever x  t   for all x  [ a , b ] , t  [ a, b] (2).

Choose a partition P with P   . Then (2) implies

 f (ti )i   fd  U ( P, f , )  L( P, f , )

Theorem 2. If f is monotonic on [ a, b] and if  is both monotonic and continuous on [ a, b] , then

Example 1. Let f be a function defined by

and so f  ( ) and  fd  0 .

Theorem 3. Let f1  ( ) and f 2 ( ) on [ a, b] , then ( f1  f 2 )  ( ) and

[ f1 (2 )  f2 (2 )]  [ f1 (1 )  f2 (1 )]

 f1 (2 )  f1 (1 )  f2 (2 )  f2 (1 )

 ( M i'  mi' )  ( M i"  mi" )

Therefore, since this hold for all 1 ,  2  [ xi 1 , xi ] , we have

M i  mi  ( M i'  mi' )  ( M i"  mi" ) (1)

Since f1 , f 2  ( ) , there exists a partition P1 and P2 of [ a, b] such that

Thus using (1), we have for f  f1  f 2 ,

Further, we note that

Multiplying by  i and adding for i  1, 2,......, n, we get

Since  is arbitrary positive number, we have

Proceeding with ( f1 ), ( f 2 ) in place of f1 and f 2 respectively, we have

 ( f )d   ( f1 )d   ( f2 )d

Now (6) and (7) yield

Theorem 4.If f  ( ) and f  (  ) then f  (   ) and

Hence, if M i and mi are bounds of f in [ xi 1 , xi ] ,

Since  is arbitrary positive number, therefore

Replacing f by  f , this inequality is reversed and hence

Theorem 5. If f  ( ) on [ a, b] , then f  ( ) on [ a, c ] and f  ( ) on [c, b] where c is a point of

Let P* be a refinement of P such that P*  P  {c} . Then

Then (1), (2) and (3) imply that

Hence f is integrable on [ a, c ] and [c, b] .

Taking inf for all partitions, the relation (2) yields

 fd   fd   fd

But since f is integrable on [ a, c ] and [c, b] , we have

The relation (3) similarly yields