Sei sulla pagina 1di 68

1.

Introduction
This course is designed to develop, grasp and use some essential and important tools to do
scientific calculations.
If we recall our beginning of learning Mathematics, you will go back to counting. But counting
of what? .... The natural objects! Unknowingly, we started learning the very first mathematical
concept of addition. We were also forced to do practice of recognizing certain figures 1, 2, 3,
. . ., 9 in an specific order along with memorizing their names as one, two, three, . . ., nine.
Thanks to this round figure 0 zero, which saved us from forceful recognizing and memorizing
the remaining figures if at all these would be existing. And thus the counting of natural objects
has been represented by
(1.1) 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, . . . .
We called these numbers as natural number and denote this collection by N. We note that
for any x, y N the number x + y is in N itself. In other words N is closed under addition.
The life is not just about taking but also for giving. Sometimes you require to give but you
dont have, for representing the situation of debt we put - minus sign preceding the units of
debt. Thus we established a collection numbers
(1.2) . . . 3, 2, 1, 0, 1, 2, 3, . . .
called as integers and denoted by Z. We also note that for any x, y Z the number x y is
again in Z itself. In other words Z is closed under subtraction.
We also need to do divisions in equal parts. For this fractional numbers were introduced.
As if we divide two breads in three equal parts then one of them is represented by 2/3. The
same amount of bread will be represented by 4/6 if we divide four breads in six equal parts.
We denote all the fractional numbers by Q0 = {m/n : m, n Z, n 6= 0}. But to come out of
ambiguity 2/3 = 4/6 = (8)/(12) = . . . we represent this whole collection by a number of the
form p/q, p Z, q N, and p ,q have no common divisor. Thus
Q = {p/q : p Z, q N, and p, q have no common divisor}
represents fractional numbers which dont have any repetition we call this collection as rational
numbers. Clearly Q is closed under the corresponding subtraction, moreover Q {0} is closed
under division. It is remarkable to note that both Z and Q are totally ordered set.
The set of rational numbers with these operation and order is known as system of rational
numbers.
You know the length of the diagonal of square of unit length is 2. But 2 is not a rational
number (Exercise). It shows that we are already using a larger set of numbers than the set of
rational numbers.
To understand this larger set we draw a straight line and mark the equal length adjacent
segments as . . . , [2, 1], [1, 0], [0, 1], [1, 2], . . . then
all the rational numbers can be represented
by certain points on this line. But observe that 2, which is not a rational number can also be
represented on this line. Thus intuitively we feel that there is a larger set of numbers, say R,
which has a one-one correspondence with points on the straight line. Indeed, one can construct
such a set of numbers from the system of rational numbers. Moreover, on this set we can define
addition and multiplication and an order in such a way that when these operations are restricted
to subset of rational, they coincide with usual operation and order. The numbers in this larger
set R are called real numbers. And the set R with these operations and the total order is called
as system of real numbers.
The real numbers which are not rational numbers are called as irrational numbers. Check
whether the set of irrational numbers is also closed under addition and multiplication?
In a totally ordered number system one can compare any two numbers. And in many calculus
problems we would like to have a solution, which should be located within certain lower and
upper boundary points.
1.1. Bounded set:
2
A subset A of totally ordered system of numbers (R or Q) is said to be bounded above if
there is an element x0 in the system such that for all x A we have x x0 . Such an element
x0 is called an upper bound for A. In a similar way one says that the set A is bounded below if
there is an element y0 A such that x y0 for all x A. And such a y0 is called lower bound
for A.
Example 1.2. The set A = {3, 5, 2, 7, 6} is bounded below by 2. So 2 is the lower bound of A.
At the same time 1 and 0 are also the lower bound for A.
Example 1.3. A finite set is always bounded.(Find out why?)
1
Example 1.4. The lower bound for the set A = {1 + n : n N} is 1.
Example 1.5. The set A = {(1)m + (1)n : m, n N} is bounded below and above by 2
and 2 respectively.
Example 1.6. The set A = {(1)n n2 : n N} is neither bounded below nor bounded above
by any integer. Because for any nonzero N Z, 2|N | N and hence (1)2|N | (2|N |)2 = 4|N |2 ,
which is an element in A, is neither less than N nor equal to N , so N cant be an upper bound
of A. Similarly one can prove that any integer N cant be a lower bound for the set. (Note that
we are within the system of integers.)
Example 1.7. The set A = {n + n1 : n N} is bounded below by 1. And there is no rational
number which can be the upper bound for A. (Prove it.) Hence A is not bounded above. (Note
that we are within the system of rational numbers.)

Example 1.8. The set A = { 2n : n N} is bounded below by 1. But there is no real number
which can be the upper bound for A. To prove this fact we need a property of real number
system, which is not true for system of rational numbers (See the proof for proposition (1.30)).
Example 1.9. The set A = {x : 2 x 3} is called as interval with closed end points or
simply closed interval. We usually denote this set by [2, 3]. Clearly A is bounded below because
2 is a lower bound. But if 2 is a lower bound then any number less then 2 is also a lower bound.
Thus lower bound is not unique. But we can also see that any number greater than 2 can not
be a lower bound, because if y0 > 2, then y0  2. So we conclude that y0 is not a lower bound
because y0 is not dominated by 2, which is the element of the set.
Example 1.10. For any two real numbers a, b R and a < b the set A = {x : a < x < b} is
known as open interval with lower end point a and upper end point b. This set is also denoted
by (a, b). Here b is the upper bound for A, and as in above example no number less than b can
be an upper bound for A. (Though b is not the element of A.)
2
Exercise 1.11. Find the lower and upper bound of the set A = { 3n
n! : n N}.
n
Exercise 1.12. Investigate the boundedness of the set A = { 2n! : n N}.
Question 1.13. How many lower bounds are possible if the set is bounded below?
Proposition 1.14. If x0 is the lower bound for the set A, then any number y, which is less
than x0 i.e. y < x0 , is also a lower bound for A.
Proof. Since x0 is lower bound of the set A, for any x A we have x0 x, which implies
y < x0 x or y < x. Hence y is also a lower bound for A. 
Remark 1.15. As the conclusion of the last proposition if there a set which is bounded below
there are infinitely many lower bounds. And if there is a lower bound, which is within the set
itself, then this lower bound is greater than all other lower bounds of the set. In other words
a lower bound within the set is greatest lower bound of the set. But if a lower bounded set
does not contain any of its lower bounds, then the question arise whether there is a number,
which is greatest among all the lower bounds. We will come back to this existential question in
Observation (1.25) and see that an affirmative answer is not possible in the system of rational
numbers.
3
1.16. Greatest lower bound or inf:
A lower bound x0 of A is said to be g.l.b. or infimum (inf ) of A if whenever u is a lower
bound of A , x0 u.
1.17. Least upper bound or sup:
An upper bound x0 of A is said to be l.u.b. or supremum (sup) of A if whenever u is an
upper bound of A , x0 u.
(1)n
Example 1.18. The infimum of the set A = {n n : n N} is 3/2. (What about
supremum? Think!)
Example 1.19. Let A = {2, 1, 3, 4, 7, 3}, then sup of A is 7 and inf is 3. (Why?)
Remark 1.20. If a non empty set is finite, then both supremum and infimum exist and belong
to the set itself.
Remark 1.21. But in general, infimum or supremum of a set need not belong to the set.
Example 1.22. It is easy to see that 1 is the lower bound of the set A = {2 n1 : n N}. Now
to find inf we first observe that 1, which is the lower bound of A, belongs to the set itself and
if x0 is any lower bound of A then x0 1. Thus 1 is the greatest lower bound of A.
3n
Exercise 1.23. Find the supremum and infimum of the set A = { 2m+n : m, n N}.
Question 1.24. whether the Infimum or Supremum of a bounded subset always exists in any
totally ordered number system.
Observation 1.25. If we consider a subset A = {x Q : 1 x2 2} of rational numbers
Q, then A is bounded above by 3/2. But there is no rational number which is the least upper
bound for set A(Exercise).
On the other hand if we consider the system of rational numbers, that is, B = {x R : 1
x2 2}, then 2 is the least upper bound of B in R.
1.26. Completeness property of real numbers:
Least upper bound property: For every non empty subset of R, which is bounded above,
there exists a least upper bound in the real number system.
Greatest lower bound property: For every non empty subset of R, which is bounded
below, there is a greatest lower bound in the real number system.
By completeness property we mean either least upper bound property or greatest lower bound
property.
Remark 1.27. Note that we did not prove these facts. In fact completeness property comes
as virtue of real number system when it is constructed by rational number system. This con-
struction is bit involved so we dont discuss it here. One can also see that least upper bound
property can be derived by assuming the greatest lower bound property and vise versa.
Exercise 1.28. Show that the least upper bound property of real number system is equivalent
to greatest lower bound property and vise versa.
If A is a bounded above subset of R, and is an upper bound of A iff B = {x R : x A}
is a bounded below subset of R and is a lower bound of B.
Further if we assume G.L.B. property in R and hence there is a infimum of B say x0 , then
x0 is the supremum of A.
1.29. Archimedean property
Proposition 1.30. If x, y are two elements in R and x > 0, then there is a positive integer n0
such that y < n0 x.
4
Proof. If y is negative then n = 1 will work. So consider that y > 0. We will prove this by
contrapositive arguments. Suppose if there is no integer n0 such that n0 x > y. Thus for all
positive integers n we must have nx y (because R is the well ordered set, that is, when ever
a, b R then at least and exactly one of the following holds i)a < b,ii) a = b, iii) a > b). Let
be the subset of R defined by A = {nx : n N}. Then A is bounded above by y. By the least
upper bound property of R, there is a real number x0 , which is the least upper bound for A.
For any n N, (n + 1)x A, and hence (n + 1)x x0 , or nx x0 x. Thus x0 x is also
an upper bound of A. Hence by virtue of x0 being sup, x0 x x0 or 0 x, which gives a
contradiction. 
Problem 1.31. Prove that 0 is the infimum of the set A = { n6 : n N}.
Solution. Since all the elements of A are positive, 0 is the lower bound of A. Now to prove
that 0 is the infimum of A, we need to show that any other lower bound of A is less than 0, or
equivalently, any number greater than 0 cant be a lower bound. (Contrapositive Argument:)
So we assume a contradiction that some number  > 0 is a lower bound and then will try get
a contradiction. (Now observe that if we keep dividing a line segment of six unit length in
to more and more number of equal length line segments, then the length of each piece of the
line segment becomes smaller and smaller and seems to go to 0.) This gives us a feeling that
there should exist some large n0 such that n60 <  and hence  should not be the lower bound (a
contradiction!). But to prove such an existence of n we recall the statement of our Archimedean
property and assume 6 as y and  as x (we can do this as  > 0, which satisfies the requirement
of x being positive) to conclude the existence of n0 such that 6 < n0 .
Example 1.32. Clearly 2 is the upper bound of the set A = {2 n1 : n N}. Then by
completeness property of real number system sup of A exists in R, but to find supremum we
first observe as in the above example that n1 goes near and near to and hence 2 n1 should go
near to 2. So we hope 2 to be the supremum, but to prove it we need to show that any other
upper bound of A is greater than 2, or equivalently, any number less than 2 cant be the
upper bound, or equivalently, we need to show the existence of an element in A, which should
be greater than , that is, the existence of n0 such that 2 n10 > (2 )n0 > 1. but since
2 > 0, the existence of n0 is guaranteed by Archimedean property
2
Exercise 1.33. Show that the infimum of the set A = { 3+7n : n N} is 0 and the supremum
2n 2
of the set B = { 3+7n : n N} is 7 .
1.34. Denseness of rational and irrational numbers in R
We now move towards proving the density of rational as well as irrational numbers in R.
Proposition 1.35. For any two distinct real numbers there is a rational number in between.
Proof. Suppose x, y R, since x 6= y either x > y or y > x. So we assume without loss
of generality that y x > 0. We want two integers m, n such that n 6= 0 and x < m
n < y,
or equivalently nx < m < ny. And this can be guaranteed by first choosing n such that
difference between nx and ny is greater than one and then choosing m = [nx] + 1, because
nx < [nx] + 1 nx + 1 < ny. 
Corollary 1.36. For any two distinct real numbers there is a irrational number in between.
Proof. Suppose x, y R, and x < y. Hence x , y R and x < y , now by the above
3 3 3 3
proposition there is some rational number m
such that x < m
< y and thus x < 3m < y. 
n 3 n 3 n

Proposition 1.37. Suppose A is a nonempty subset of R. Then R is the sup A if and only
if
a for all a A and
For every  > 0, there is some a0 A such that  < a0 .
5
Proof. Since is the greatest lower bound of A, so obliviously the upper bound of A and hence
a for all a A. Further for any  > 0,  < , and being the least upper bound 
cant be the upper bound of A and hence there is some element a0 which is not dominated by
 , that is, a0  , or equivalently, a0 > .
Next if we assume the two statements, the first one is equivalent to say that is the upper
bound of the set A. Now we need to show that if is any other upper bound then .
We will prove this by contrapositive argument. So suppose  , that is, > and if we
denote  = , then  > 0 and according to second assumption there is some a0 A such
that  < a0 or < a0 . This concludes that is not an upper bound of A, which is a
contradiction. Hence our assumption that  is wrong and the only possibility is that
. And we conclude that is the least upper bound. 
Remark 1.38. One can also prove a similar result for infimum. Suppose A is a nonempty
subset of R. Then R is the sup A if and only if i) a for all a A and ii) For every
 > 0, there is some a0 A such that  < a0 .
n o
m
Problem 1.39. Find the supremum and infimum of the set m+n : m, n N .
m
Solution: Since m < m + n, we gave 0 < m+n < 1. Now observe that how close to zero and
1
one we can go. For this if we consider m fixed say m = 1, 1+n seems to go to zero and therefore
we hope that no number greater than zero is a lower bound. To be more precise for any  > 0
1
we need to find n0 such that 1+n 0
<  or 1 < (1 + n0 ), but by Archimedean property there is
some n0 N such that 1 < n0 , which implies 1 < n0  +  and hence 0 is the g.l.b. of A.
m 1
Similarly if we fix some n, say n = 1, then m+1 = 1 m+1 seems to go to 1. But to prove
that 1 is the least upper bound we need to show that any number less than 1 can not be the
upper bound, or for any  > 0 , 1  cant be the upper bound. For this we need to search
some m0 such that 1 m01+1 > 1  or  > m01+1 . Now the existence of m0 can be guaranteed
Archimedean property as above.

2. Sequences
Let us start this section by posing a question.
1 1 1
Question 2.1. Can we find the sum 12 +1
+ 22 +2
+ ... + n2 +n
+ . . .?
Because we dont know how to sum infinitely many terms at a time, what we do, we define
numbers xn as the sum of first n numbers. And we try to find where does xn approaches if n
approaches towards infinity.
We will see that this infinite sum is 1, by showing that xn approaches to 1, when n approaches
towards infinity.
To understand the things rigorously we need to define sequences.
2.2. Sequence
Formally a sequence is a function whose domain is N codomain is R. So any function
f : N R,
is a sequence. But usually we write a sequence by writing the image of this function in the
same order as the order of its domain set, that is, N. So we usually write a sequence as
(f (1), f (2), . . .). Or if we denote f (n) by an , then (a1 , a2 , a3 , . . . , an , . . .) represents a sequence
with the understanding that the real number an is the image of n N and it appears at nth
place from the left in representation of the sequence. Some times if an can be represented in
terms of n like n+5 for all n N simultaneously, then we simply write the sequence as ( n+5 )
n n n=1
or more simply just by its nth term ( n+5
).
n

Example 2.3. {f (1) = 24, f (2) = 3, f (3) = 51, . . .} is sequence if we know all the values
f (n) for all n.
6
Example 2.4. (3n) = (3, 6, 9, . . .).
Example 2.5. ((1)n ) = (1, 1, 1, . . .).
Example 2.6. A sequence is called constant sequence if xn = x1 for all n 2. The sequence
(2) = (2, 2, 2, . . .) is a constant sequence.
Example 2.7. ( n1 ) = (1, 21 , 31 , . . .).

Example 2.8. (n1/n n1 ) = (0, 2 12 , (3)1/3 13 , . . .).
2.9. Convergence of Sequences
We say that a sequence converges if the terms of the sequence approaches to some real
number. In other words if there is a real number such that any neighborhood (an open interval
containing this number and by  neighborhood of x0 we mean (x0 , x0 + e) ) of this number
contains all the terms of the sequence except a finite number of terms.
To be more precise a sequence (xn ) converges if there exists a real number x0 such that for
any  > 0, there exists some N ( N) depending on  such that

xn (x0 , x0 + ), for all n N , or


|xn x0 | < , for all n N .
One can easily see that if such a number x0 exists, then it is unique (Exercise). In this case
we say that the sequence converges to x0 and we call x0 the limit of the sequence (xn ). If x0 is
the limit of the sequence (xn ), we write limxn = x0 or xn x0 .
Example 2.10. The constant sequence (xn = 2) converges to 2, because |xn 2| = |2 2| =
0 <  for any  positive and n 1.
Remark 2.11. The constant sequence always converges to the repeating term of the sequence.
2
Problem 2.12. Investigate the convergence of the sequences ( 3n+1 ).

Solution. The sequence seems to converge to 0. (Why?). But to prove 0 as the limit of the
sequence we need to show that for any given  > 0, all the remaining terms after some finite
number of terms are inside the interval (0 , 0 + e), that is, we need to find some N such that
2 2
3n+1 (, ) for all n N , or equivalently, 3n+1 <  2  < 3n and thus by Archimedean
property with 2  as y and 3 as x we can ensure the existence of n0 such that 2  < 3n0
and hence above inequality is true for all n n0 and thus by considering N as n0 we are done.
Problem 2.13. Show that for p > 0 the sequences ( n1p ) convergence to 0.

Solution. As in above problem for any given  > 0 we need to find N such that n1p <  for all
n N , or equivalently, 1 < np  1 < np and now we can use Archimedean property with 1
as y and p as x to ensure the existence of n0 such that 1 < n0 p , which implies that 1 < np
for all n n0 . Hence by considering N as n0 we are done.
Exercise 2.14. Show that if a sequence xn x0 , then for any real number c the sequence
(cxn ) converges to cx0 .
Theorem 2.15. Show that the sequence (xn )
n=1 converges to x0 if and only if the sequence
(xn+K )
n=1 converges to x0 , where K N.
Proof. Suppose the sequence (xn+K ) n=1 converges to x0 . For a given  > 0, there exists a
number N N such that xn+K (x0 , x0 +) for all n N or equivalently xn (x0 , x0 +)
for all n N + K. And this is equivalent to the convergence of the sequence to x0 . 
2.16. Bounded sequence
7
A sequence (xn ) is said to be bonded if the image set of the sequence function, that is,
{xn : n N } is bounded subset of R. And we say that m and M are the lower and upper bounds
respectively if all the terms of the sequence lies within the interval [m, M ], or equivalently,
m xn M for all n N.
 n

Example 2.17. The sequence 1+3(1) n is a bounded sequence with lower bound as 2 and
upper bound as 74 . Note that the sequence converges to 0.
 nn

Example 2.18. The sequence 1+(1) n is a bounded sequence upper and lower bounds as
3
2 and 23 respectively. But it is not a convergent sequence.
n
Exercise 2.19. Show that the sequence ( 2n2 ) is unbounded.
Question 2.20. Whether an unbounded sequence can converge.
Observation 2.21. No! Because we know that for a convergent sequence except a finite number
of terms, remaining all the terms of the sequence are inside a neighborhood of the limit, and
hence they are bounded.
Theorem 2.22. A convergent sequence is bounded.
Proof. Suppose xn l. As in the observation we will try to find a bounded nbd of the limit
l so that except finitely many terms all the remaining terms of the sequence are within this
bounded neighborhood. By definition, for  = 1 there exists N such that |xn l| < 1 for all
n N and hence |xn | = |xn l + l| |xn l| + |l| < 1 + |l| for all n N . Thus if we choose
M = max{|x1 |, |x2 |, . . . , |xN 1 |, 1 + |l|}, we see the each term of the sequence lies between the
interval [M, M ]. 
Remark 2.23. Thus above theorem provides a test for the non-convergent sequences, that is,
if the sequence is unbounded then it is not convergent.
Question 2.24. Though a bounded sequence need not convergent, but can we make some extra
assumption to a bounded sequence so that it turns out to be a convergent sequence?
2.25. Monotonic sequence
A sequence (xn ) is said to be monotonic increasing if xn xn+1 for all n N. And in this
case symbolically we write (xn ) . Similarly a sequence (xn ) is said to be monotonic decreasing
if xn xn+1 for all n N. To write it symbolically we use the notation (xn ) . A sequence is
called monotonic if it is either monotonic increasing or monotonic decreasing.

Example 2.26. The sequence ( n) is monotonic increasing.
Example 2.27. The sequence ( 2n2n1 ) is monotonic decreasing.
Theorem 2.28. a) If a monotonic increasing sequence is bounded above then it is convergent.
(In fact it converges to its least upper bound).
b) If a monotonic decreasing sequence is bounded below then it converges to its greatest lower
bound.
Proof. a)Suppose (xn ) is a monotonic increasing sequence and the image set A = {xn : n N}
is bounded above. Now if A is a finite set, then what do you hope? The maximum of A will
be the limit of the sequence. (Justify yourself why?) And if A is infinite set, then terms of the
sequence are gradually going up and up, but since A is bounded above, these terms cant cross
any upper bound of A. But A being a subset of R, there is a least upper bound of A, say , and
the terms of the sequence cant cross this LUB, but they are going to densely gather around
. To prove it suppose  is a positive real number and since is the least upper bound, 
cant be an upper bound and hence there will exist some element xn0 in A, which will not be
dominated by , that is, xn0  , or equivalently, xn0 > , or  < xn0 < . But
for any n n0 xn (xn0 , ) ( , ).
b) Analogous to above proof. (Exercise) 
8

Problem 2.29. Suppose x1 = 3 and xn is defined recursively as xn = 6 + xn1 for n > 1.
Show that xn is convergent and find its limit.
Solution. As the problem suggests that we need to prove that the sequence (xn ) is monotonic
increasing and it converges to 3. Here one can use Theorem (2.28) to conclude the convergence
of the sequence (xn ) to 3 either by showing that (xn ) is monotonic increasing and 3 is the LUB
of (xn ) or by showing that (xn ) is monotonic increasing and bounded above and use some extra
argument to show that it converges to 3.
p
First we first try to see that x2 = 6 + x1 = 6 + 3 > 3 = x1 . Now we assume that
xk+1 > xk forall 2 k N and try to prove that xN +1 > xN . Since xN > xN 1 6 + xN >
6 + xN 1 6 + xN > 6 + xN 1 xN +1 > xN . Thus by using mathematical induction
xn+1 > xn for all n N. Now we willagain use Mathematical induction to show that 3 is the
upper bound for (xn ). Clearly x1 = 3 < 3 and we assume that xk 3 for all 2 k < N ,
so that xN 3 xN + 6 9 xN + 6 9 xN +1 3. Thus by induction argument
sequence (xn ) is bounded by 3. So by Monotonic Convergence theorem the sequence converges.
Now instead of proving that 3 is the LUB of the sequence directly, we assume that the sequence
converges to l, then will show that l = 3. Since xn l xn+1 l x2n+1 l2 (xn + 6)
l2 xn l2 6. Since the limit of a convergent sequence is unique we have l2 6 = l. This
implies that either l = 2 or l = 3. But l = 2 is not possible as xn 0 implies the limit l 0
(see theorem (2.42)). So l = 3.
Question 2.30. What if a sequence is unbounded and monotonic?
Remark 2.31. Clearly by Theorem 2.22 we know that the unbounded sequence cant converge.
Okey! But if the sequence is not bounded, terms of the sequence tend to cross each finite
milestone and since it is monotonic, once a milestone is crossed by some term of the sequence,
then all the succeeding one are necessarily going to cross this limit.
Thus one can conclude that the sequence is tending to infinite.
2.32. Tending to infinity
We say that a sequence (xn ) is tending to infinity if for any finite number M > 0 there exists
a natural number N such that xn M for all n N . And in this case we also call that the
sequence is diverging to infinity and write xn .
In a similar way we say that a sequence is diverging to if for any finite number M > 0
there exists a natural number N such that xn M for all n N .
Example 2.33. For p > 0 the sequence (np ) diverges to infinity.
Example 2.34. The sequence ((1)n n + (1)n+1 n2 n2 ) diverges to but it is not mono-
tonic.
Remark 2.35. Thus the divergence of a sequence (xn ) to does not guarantee the mono-
tonicity of the sequence (xn ). But conversely, monotonicity of the sequence (xn ) does guarantee
some approaching behaviour of the sequence.
Theorem 2.36. If a sequence is monotonic, then either sequence is convergent or it diverges
to
Proof. (*) If a monotonic sequence (xn ) is bounded, then by Theorem 2.28 one can guarantee the
convergence of (xn ). So we consider that the sequence is monotonic and unbounded. Without
loss of generality we assume that (xn ) is monotonic decreasing. Since the sequence is unbounded
we feel that the terms of the sequence should go down and down and cross each finite milestone.
Precisely, the image set A = {xn : n N} is bounded above by x1 , the first term of the sequence
and we assumed that the sequence is unbounded, the set A cant be bounded below. And hence
for any number M > 0 the number M cant be a lower bound of A. This implies that there
exists some element in A which cant dominate M . Suppose this element is xN for some
9
N N so that M  xN , that is M > xN but since (xn ) is monotonic decreasing then
xn xN < M for all n N . Hence xn .
Similarly one can assume that (xn ) is monotonic increasing and can prove that xn . 
Remark 2.37. Thus just monotonicity implies that the terms of the sequence are either ap-
proaching to a finite real number or moving towards infinity.
Problem 2.38. Show that for any real number x0 there is increasing/decreasing sequence of
rational/irrationals.
Solution. Recall that between any two real numbers there are infinitely many raionals and
irrationals. So we can always choose a rational number r1 (x0 1, x0 ), then choose other
rational numbers r2 (r1 , x0 ) (x0 21 , x0 ) and so on rn (rn1 , x0 ) (x0 n1 , x0 ). Thus (rn )
is an increasing sequence such that x0 n1 < rn < x0 . So rn x0 by Sandwich theorem.
Similarly we can find other sequences.
Remark 2.39. Using (-) definition to investigate the convergence is not an easy task always.
So it is desirable that some times if we know the the convergence behaviour of certain sequence,
then we should be able to infer the behaviour of other related sequences. The following theorem
is an important tool in this direction.
Theorem 2.40. Suppose xn x0 and yn y0 . Then
(1) xn + yn x0 + y0
(2) xn yn x0 y0
(3) y1n y10 , if y 6= 0 and yn 6= 0 for all n.
(4) And hence xynn xy00 , if y 6= 0 and yn 6= 0 for all n.
Proof. (*) (1) is easy left as an exercise.
(2) For a given  > 0 we need to find N such that whenever n N we have |xn yn x0 y0 | < .
So to control |xn yn x0 y0 | with the help of |xn x0 | and |yn y0 | we proceed as |xn yn x0 y0 | =
|xn yn xn y0 + xn y0 x0 y0 | |xn yn xn y0 | + |xn y0 x0 y0 | = |xn ||yn y0 | + |xn x0 ||y0 |.
Now to dominate the expression by , we will try to dominate each of both the terms of the last
expression separately by /2. Firstly for |xn ||yn y0 |, we recall that since (xn ) is convergent,
there exists some M > 0 such that |xn | < M so that |xn ||yn y0 | < M |yn y0 |. Now since
yn y0 and M M
2 > 0 we can find N1 such that |yn y0 | < 2 M |yn y0 | < 2 whenever


n N1 . Further since xn x0 and |y20 | > 0 we can find N2 such that |xn x0 | < |y20 |
whenever n N2 . Now we can chose N = max{N1 , N2 } so that whenever n N both the
required inequalities hold.
(3) For a given  > 0 we need to find N such that whenever n N we have | y1n y10 | < .
But | y1n y10 | = |y|ynny 0|
y0 | . Thus the things will be in our control if we have control over |yn | ,
1

or equivalently, we want to produce some nonzero lower bound for |yn | which is quite possible
because yn0 s are converging to y0 , which is nonzero. But to make it happen explicitly we take a
neighborhood around y0 (by a neighborhood of a point we mean an open interval containing the
point and by a neighborhood around a point a of radius r we mean the open interval of length
2r and midpoint a, that is, (a r, a + r) ) with radius as half of its distance from origin, that
is |y0 |/2 so that each element of this NBD is at least at |y0 |/2 distance from origin. Now since
|y0 |/2 > 0 and yn y0 , there exists N1 such that yn (y0 |y0 |/2, y0 + |y0 |/2), or equivalently,
|yn y0 | < |y0 |/2 whenever n N1 . But this implies that |y0 | = |y0 yn +yn | |y0 yn |+|yn |
|y0 |/2 + |yn | and hence |y0 |/2 |yn |. Thus | y1n y10 | = |y|ynny 0| |yn y0 |
y0 | |y0 |2 /2 whenever n N1 . Now
since 2/|y0 |2 > 0 and yn y0 , there exists N2 such that |yn y0 | 2/|y0 |2 . Now one can
chose N = max{N1 , N2 } so that both the dominations can happen simultaneously.
(4) It is the implication of (2) and (3). 
Question 2.41. In last theorem we have seen that the convergence behaviour of sequences
respects term wise addition, multiplication and division. Now it natural to ask whether it also
respects term wise comparison.
10
Theorem 2.42. (Comparison Theorem) Let xn yn for all n N. If xn x0 and yn y0 ,
then x0 y0 .
Proof. The statement looks true. Because if x0 and y0 are two different points and eventually
(after n N ) xn s are in (x0 , x0 + ) and yn s are in (y0 , y0 + ), then if  = |x0 y0 |/2
so that these nbds are disjoint such that xN (x0 , x0 + ) and yN (y0 , y0 + ) but
xN < yN x0 +  < y0  x0 < y0 . 
Remark 2.43. If xn < yn for all n N, then it is not necessary that x0 < y0 , for example
xn = n1 and yn = n1 .
Following theorem is a very important tool to verify the convergence of certain sequences,
which are (informally speaking) sandwiched between two convergent sequences approaching to
the same limit.
Theorem 2.44. (Sandwich Theorem) Suppose xn , yn and zn are sequences such that xn
yn zn for all n and that xn l and zn l. Then yn l.
Proof. For a given  > 0, there exists N1 such that whenever n N1
xn (l , l + ).
Similarly for the same  there exists N2 such that whenever n N2
zn (l , l + ).
Thus for n N = max{N1 , N2 }, yn belongs to the closed interval [xn , zn ] and this interval itself
is contained in the interval (l , l + ) (justify why?) and so is yn . 
2
Example 2.45. The convergence of the sequence ( sinnn ) can be assured by Sandwich Theorem
2
and the observation that n1 sinnn n1 , because both ( n1 ) and n1 converges to 0.
Remark 2.46. We know that a geometric sequence (xn ) converges iff |x| < 1 and in this case
limit is 0. Now if there is a sequence (an ) satisfying 0 an xn for all n N and x < 1, then
by sandwich theorem an 0. In the following theorem we will show that if the sequence ( an+1an )
converges to a limit, then eventually the sequence (an ) itself is comparable with a geometric
sequence.
Theorem 2.47. Let (an ) be a sequence of strictly positive real numbers such that
limn an+1
an = l. Then
(1) if l < 1, then an 0,
(2) if l > 1, then an ,
(3) if l = 1, then both the above conclusions can arise.
Proof. First suppose l < 1, so we can find an  = |1l|/2 nbd, i.e., (l , l +) of the point l such
that this nbd will strictly lie on the left of 1. And by convergence of an+1
an 1 eventually all the
terms are within this nbd, more precisely there exists N N such that l |1l|
2 <
an+1
an < l+ |1l|
2
for all n N , but l+ |1l|
2 < 1. So for all n N we have
an+1
an < l+ 2
|1l|
< 1. Now if for simplicity
|1l|
we write l + 2 = r, then an+1 < ran < r.ran1 . . . Thus 0 < an rn arN
< rnN +1 aN . N
for
n aN
all n N , where (r rN ) is a geometric sequence converging to 0. Hence the sequence (an ) is
bounded by a constant sequence (0) and a geometric sequence, which converges to 0. Thus by
sandwich theorem the sequence (an ) itself has to converge to 0.
Similarly if l > 1, then for  = |l 1|/2 > 0, there exists N N such that l |1l| an+1
2 < an <
l + |1l|
2 for all n N . But 1 < l |1l|
2 = r(say), so that as discussed above it implies that
n aN
r rN < an for all n N . So for any given M > 0 by Archimedean property there exists N1
such that M < rn arN N
for all n N1 . Now if N2 = max{N, N1 }, then for all n N2 we have
n aN
M < r rN < an . Thus by definition (an ) diverges to infinity.
11
Now we observe that the sequence ( n1 ) converges to 0, (c + n1 ) converges to c and the sequence
(n) diverges to infinity. But in all the three cases limn an+1
an = 1.

2.48. Cauchy Criterion
We have seen that just the boundedness and monotonicity of the sequence implies the con-
vergence. Now our aim is to find another criterion of the convergence of the sequence.
For this we first observe that if a sequence is convergent then eventually all the terms (all the
remaining terms after a finite number of terms) of the sequence are at a desired distance from
the limit of the sequence and hence one can conclude that eventually (after a finite number
of terms) any two terms of the sequence are at a desired distance. This virtue of
a convergent sequence neither depends on the limit of the sequence nor says anything specific
about the convergence. But in fact later we will show that if a sequence possesses this property,
then we can deduce the convergence the sequence. Thus it is another criterion of convergence
and is known as Cauchy criterion.
Precisely, if xn x0 , then for a given  > 0 there exists N N such that

|xp x0 | for all p > N.
2
Hence for all m, n N
|xm xn | = |xm x0 + x0 xn | |xm x0 | + |x0 xn | < .
We say that a sequence (xn ) satisfies Cauchy criterion if for given  > 0 there exists some N N
such that
|xm xn | <  for all m, n N.
Observation 2.49. We first observe that if a sequence satisfies the Cauchy criterion, then ex-
cept a finite number of terms, remaining all the terms of the sequence are inside a neighborhood
of any of the remaining terms. Hence this sequence has to be bounded, which is one property
of a convergent sequence.
Remark 2.50. (Students can skip this remark for the first time reading.) We observed that
the image set of a sequence satisfying the Cauchy criterion is bounded. And in general, we
can also assume it to be infinite set. Thus the image set, which is infinite, is contained in a
bonded interval. Though hypothetically, we know that terms of the sequence are going to be
dense more and more in the neighbourhoods of the later and later terms of the sequence. Or
in more precise words, we can find an interval I1 of unit length, which contains eventually all
the terms of the sequence. Further we can find another interval I2 of length one half, contained
in I1 , and containing all but finitely many terms of the sequence. And this way we can get a
nested sequence of intervals I1 I2 . . . In In+1 . . . such that length of In+1 is half of
the length of In . And except a finite number of terms all the terms of the sequence are inside
each of these interval. And thus we hopeTthat if the sequence converges to a limit, it has to
be contained in each In and hence in the n=1 In . (Note that this intersection can not contain
more than one point because the length of the intervals is decreasing to zero.) And without
knowing the convergence of the sequence, if we know some how that this intersection contains
a point, then the construction of the intervals we know that this point in the intersection has
to be the limit of the sequence. Thus only thing remains to prove is that such an intersection is
nonempty. Note that if there is some point in the intersection it has to be upper bound for the
set of all lower end points of intervals and the lower bound for the set of all upper end points
of the intervals. Thus even if we do not know the existence of such a point in the intersection,
we can try to prove that the supremum of the set of lower end points of the intervals is such a
point. So we get an idea of the following theorem.
Theorem 2.51. (Nested Interval Theorem) For each n, let an , bn R be such that closed
interval [an , bn ] = In (say) satisfies
I1 I2 . . . In In+1 . . .
12
T
and limn (bn an ) = 0. Then n=1 In contains exactly one point.
Proof. (*) It is clear from the assumption that (an ) is a monotonic increasing sequence and
bounded above by b1 and similarly (bn ) is a monotonic decreasing with lower bound as a1 .
Hence by Theorem 2.28 an % a = sup{an : n N} and bn & b = inf{bn : n N}. But
0 = limn (bn T an ) = b a implies a = b. Also an a = b bn implies a In for all n N
and hence a n=1 In . This intersection can not contain any other point than a, because if
it contains some point c, then a, c In for all n and hence length of In is always greater than
|a c| giving a contradiction to the assumption that length of In goes to zero. 
Remark 2.52. We have seen that if we assume monotonicity of a bounded sequence, it turns
out to be convergent. But if we do not assume monotonicity, then what?
We know certain examples of the bounded sequences e.g. (1)n + n1 and sin( n

2 ) which
do not converge. But to prove non convergence of this sequence (directly by - definition)
might not be an easy task. (Think why?...) May be because one need to disqualify each real
number for being the limit of the sequence. Further, to prove that a given number is not a limit
of a sequence we need to show that there is a nbd of the number which cant contain eventually
all the terms of the sequence, that is, there are infinitely many terms of the sequence, which
are not there in the nbd. So we need to deal with a concept of separating out infinite terms of
the sequence. We will define this concept as the next topic.
But in general suppose a bounded sequence has infinite distinct terms, can one hope denseness
of certain terms of this sequence? In other words, if a bounded interval contains infinite image
set of a sequence, can one hope that an infinite subset of this image set is dense somewhere
in the interval? This we feel to be true because if we divide this interval in to finitely many
equal length subintervals, then at least one of these subintervals will contain infinite subset of
image set. We can further subdivide this subinterval and do it repeatedly to find a sequence
of nested intervals whose length goes to zero and each nested subinterval contains smaller and
smaller infinite subset of the image set. Then by Nested interval theorem there is a point in
the intersection such that the infinite subset of the image set is going to be dense around this
point. And if it is possible to find another sequence, whose image set is the infinite subset dense
around the point, then this new sequence seem to converge to the point.
2.53. Subsequence
Let (xn )
n=1 be a sequence of real numbers and let (nk )k=1 be an strict increasing sequence of
natural numbers such that n1 < n2 < . . .. Then the sequence (xnk ) k=1 is called as subsequence
of (xn ).
In other words a subsequence is constructed by deleting some of the terms of the mother
sequence and retaining the remaining once in the same order. For example ( 1+2k 1
)
k=1 is a
1 1 n+1 1
subsequence of ( n ). And ( 2k )k=1 is a subsequence of ((1) n ).
The constant sequence (1, 1, 1, . . .) is the subsequence of ((1)n ).
Remark 2.54. A sequence may have a convergent subsequence though the mother sequence
itself may not converge.
Question 2.55. Does a subsequence of a convergent sequence necessarily converge?
We recall that the terms of the subsequence are nothing but certain terms of the mother
sequence in the same order and since eventually all the terms of a convergent sequence are
within a given neighborhood of the limit, one can infer the same for the subsequence.
Theorem 2.56. If (xn ) x0 , then any subsequence (xnk ) also converges to the same limit x0 .
Proof. Since (xn ) x0 , then for any given  > 0 there exists N N such that
|xn x0 |  for all n N.
But then k N implies nk k N and |xnk x0 | . 
13
Remark 2.57. Last Theorem says that the convergence of a sequence (xn ) implies the conver-
gence of all the subsequences. But what about the converse, that is, if we assume that all the
subsequences of a sequence are convergent, then whether the sequence itself is also convergent.
Here one can observe that the sequence can be considered as a subsequence of itself. And hence
it converges by assumption.
2.58. Criterion for non convergence of a sequence
As we discussed in the last remark that a sequence converges if and only if all possible
subsequences converge. Thus equivalently a sequence does not converge if and only if there
exists a subsequence which does not converge.
Each one of the following four cases implies the existence of a nonconvergent subsequence,
which guarantees the non convergence of the mother sequence.
(1) If there exists a subsequence, which is not bounded hence not convergent.
(2) If there exists a subsequence, which might be bounded but not convergent.
(3) If there exists two subsequences such that they can not converge to single point.
(4) if there exists two subsequence such that they converge to different limits.
Problem 2.59. Show that the sequence (sin n) can not converge.
Solution. Suppose if the sequence is convergent then by the last theorem all subsequences will
converge to same limit. Now we will try to show that there are two subsequences which lie in two
disjoint closed intervals and hence even if these subsequences converge, they can not converge
to the same limit. For this we observe that x [2m + 4 , 2m + 3 1
4 ] sin x [ 2 , 1] and
x [2k 3 1 3
4 , 2k 4 ] sin x [1, 2 ]. But since length of the intervals [2m+ 4 , 2m+ 4 ]
or [2k 3
4 , 2k 4 ] is 2 > 1, each of these subintervals will contain at least one integer say
nm [2m + 4 , 2m + 3 3
4 ] and nk [2k 4 , 2k 4 ] so that the subsequences (xnm ) and
(xnk ) are contained in [ 12 , 1] and [1, 12 ] respectively. Hence these subsequences, in case if
both converge, can not converge to a single point.
Remark 2.60. In general the convergence of one subsequence does not guarantee the conver-
gence of the sequence. But if we impose the condition of monotonicity on the mother sequence,
then the convergence of any one of its subsequence implies the convergence of mother sequence.
Theorem 2.61. Let (an ) be a monotonic sequence. If a subsequence of (an ) is bounded, then
sequence (an ) itself is bounded and therefore it is convergent.
Proof. Without loss of generality suppose (an ) is monotonic increasing sequence. Further sup-
pose (ank ) is bounded subsequence of (an ) with upper bound as M . For N N, N nN
and therefore aN anN M . Thus M is the upper bound for the sequence (an ) and so it is
convergent. 
Remark 2.62. In general, in stead of monotonicity if we assume that each subsequence con-
verges to a fixed limit, then the sequence converges. But to check convergence of all possible
subsequences is not a good idea. On the other hand one can guarantee the convergence of a
sequence by the convergence of certain finite number of subsequences, which exhaust the mother
sequence.
Example 2.63. Let (an ) be a sequence such that the subsequences (a3n ), (a3n1 ) and (a3n2 )
converge to certain limit l. Then for  > 0 there exist N0 , N1 and N2 such that n Ni
|a3ni l| < . Now if N = max{3Ni i}, then n N |an l| < .
Now as discussed in Remark (2.52) in the following theorem we prove the existence of density
in the terms of a bounded sequence.
Theorem 2.64. (Bolzano-Weierstrass Theorem) Every bounded sequence in R has a con-
vergent subsequence.
14
Proof. (*) Suppose (xn ) is a bounded sequence of real numbers with lower and upper bounds
as a and b respectively. Then the set A = {xn : n N } is contained in the interval I1 = [a, b].
If there are only finite number of terms in A, then at least one element of A is repeated infinite
times in (xn ) and hence one can extract a constant subsequence consisting of this repeated
terms of (xn ).
Now suppose there are infinite elements in A. And to construct a convergent subsequence in
this case we need to ensure the denseness of certain elements of A (which we hope to use for
construction of the subsequence) around some real number (for which we hope to be the limit of
the subsequence), and further to find such a real number whose each neighborhood howsoever
small, contains infinite elements of A, it is sufficient to search the nested closed subintervals
such that each of them has infinite elements of A and length of these nested subintervals is
decreasing to 0.
To do all this precisely we first subdivide I1 from its middle point to construct two equal
length subintervals. Then at least one of the subinterval will contain infinite elements of A call
this as I2 . And the length of I2 is half of the length of I1 . Further subdivide I2 from its middle
point in to two equal length intervals and choose the one, which contains infinite number of
elements of A. Call this subinterval of I2 as I3 so that I3 A is infinite and length of I3 is half
of the length of I2 . We continue this process to obtain subintervals In , n N such that
I1 I2 . . . In In+1 . . .
and length of In = 2n1 times length of I1 . Thus length of In is decreasing to zero. T And we
are within the hypothesis of Nested Interval Theorem 2.51 hence can conclude that n=1 In is
a singleton set, say {x0 }.
Now we aim to construct a subsequence whose terms after kth stage are contained in Ik and
hence it is going to be dense around x0 . Precisely first choose any element in A I1 say xn1 and
since there are infinite successors of xn1 , which are contained in I2 , we choose one xn2 I2 A
such that n2 > n1 . And similarly we choose xn3 I3 A such that n3 > n2 and continue this
process to choose xnk Ik A for all k N so that both xnk and x0 are in Ik and hence the
distance between xnk and x0 is less than the length of Ik , that is,
ba
|xnk x0 | k1 .
2
Now for given  > 0, we can find a K N such that 2ba K1 <  and hence conclude that
|xnk x0 |  whenever k K. 
Problem 2.65. A bounded sequence (xn ) is not convergent if and only if there are at least two
different convergent subsequences (xnm ) and (xnk ) converging to different limits.
Solution. (*) Clearly if (xn ) is a bounded sequence, then by Bolzano-Weierstrass Theorem there
is some convergent subsequence say (xnm ). Suppose (xnm ) converges to limit l, by assumption
(xn ) can not converge to l. And hence by definition there exists some  > 0 such that for each
p N there exists some np such that xnp 6 (le, l+). Thus new constructed sequence (xnp ) p=1
is a subsequence of (xn ), so it is bounded. Again by Bolzano-Weierstrass Theorem (xnp ) p=1 has
a convergent subsequence (xnpm )
m=1 . Since (xnp )p=1 lies in the complement of (l e, l + ),

the subsequence (xnpm )m=1 also lies out side (l e, l + ). If this new subsequence converges to
l1 and this l1 (l e, l + ), then there is a 1 > 0 such that (l1 e1 , l1 + 1 ) (l e, l + ),
but eventually (xnpm )m=1 is in (l1 e1 , l1 + 1 ), hence in (l e, l + ), which is a contradiction
because xnp s are outside (l e, l + ). Thus l1 6 (l e, l + ), hence l 6= l1 . One can consider
(xnpm )m=1 as required (xnm ).
Theorem 2.66. A sequence satisfying the cauchy criterion is convergent.
Proof. (*) Suppose if a sequence (xn ) satisfies Cauchy criterion, then firstly we will show that
it is bounded. This fact will guarantee the existence of a convergent subsequence by Bolzano-
Weierstrass Theorem 2.64 and then we will show that the sequence itself converges to the limit
of subsequence.
15
Since (xn ) satisfies Cauchy criterion, for a given  > 0 we can find some N N such that

(2.3) |xm xn | for all n, m N
2
and hence |xn | = |xn xN + xN | |xn xN | + |xN | /2 + |xN | for all n N . Now
M = max{|x1 |, |x2 |, . . . , |xN 1 |, /2 + |xN |} will be a bound of (xn ). So by Bolzano-Weierstrass
Theorem there is a subsequence say (xnk ) k=1 converging to some real number x0 .
Now we aim to show that the sequence (xn ) itself converges to x0 . Since the subsequence
(xnk )
k=1 converges to x0 , we can find K N (depending on the same  as chosen in the beginning
of the proof) such that

(2.4) |xnk x0 | for all k K.
2
Now for the happening of both (2.3) and (2.4) simultaneously for m = nk , we need n N ,
m = nk N and k K nk nK . Thus for n N choose any k K such that nk N
(or equivalently choose any nk max{nK , N }) so that |xn x0 | = |xn xnk + xnk x0 |
|xn xnk | + |xnk x0 | . 
Theorem 2.67. Suppose that 0 < < 1 and that (xn ) is a sequence which satisfies one of the
following conditions
(1) |xn+1 xn | n , for all n N for some N N,
(2) |xn+2 xn+1 | |xn+1 xn |, for all n N for some N N.
Then prove that (xn ) satisfies the Cauchy criterion.
Proof. To show that the sequence satisfies Cauchy criterion, we need to control |xn xm |
eventually. But for n > m N we have
| xn xm | = | xn xn1 + xn1 xn2 + . . . + xm+1 xm |
| xn xn1 | + | xn1 xn2 | + . . . + | xm+1 xm |
n1 + n2 + . . . + m = m (1 + + 2 + . . . + nm1 )
1
m (1 + + 2 + . . . + nm1 + nm + . . .) = m
1
1
Now for  > 0 we can find N1 (by Archimedean property) such that m 1 <  for all m N1 .
Now one can choose N2 = max{N, N1 }, so that whenever n > m N2 , we have | xn xm |< .
Hence (xn ) satisfies Cauchy criterion.
Similarly for the second condition we have |xn+2 xn+1 | |xn+1 xn | so that |xn xn1 |
|xn1 xn2 | 2 |xn2 xn3 | . . . nN 1 |xN +1 xN |. Now for n > m > N2
| xn xm | = | xn xn1 + xn1 xn2 + . . . + xm+1 xm |
| xn xn1 | + | xn1 xn2 | + . . . + | xm+1 xm |
nN 1 |xN +1 xN | + . . . + mN 1 |xN +1 xN |
= |xN +1 xN |mN 1 (1 + + 2 + . . . + nm )
|xN +1 xN |mN 1 (1 + + 2 + . . . + nm + nm+1 + . . .)
1 N 1 |xN +1 xN |
= |xN +1 xN |mN 1 = m
1 1
N 1 |x x |
As above for given  > 0 we can again find some N1 such that m N +1
1
N
< , so that if
N2 = max{N, N1 }, then for n > m > N2 , we have | xn xm |< . Hence (xn ) satisfies Cauchy
criterion. 
Remark 2.68. In above theorem < 1 is a necessary condition. We will see certain examples in
which = 1, but the sequence does not satisfy Cauchy criterion.
If (xn = n), then xn x n1 = 1
but
the sequence diverges to infinity. And if (x
n = n),
then |
x n+2 x n+1 =| n+2
n + 1 |= n+2+1 n+1 n+1+ 1
n
=| n + 1 n |=| n + 2 n + 1 |=| x n+1 x n |. But

( n) diverges to infinity.
16
3. Continuity
Intuitively by continuity of a function the very first thing, which comes in our mind is the
continuity of the graph of the function, that is, there is no jump in the graph or one should be
able to draw it by the continuous motion of the pencil without leaving the paper.
And if the graph of the function is broken at the point (x0 , f (x0 )), or simply there is a jump
in the graph at (x0 , f (x0 )), then we cannot hope the continuity of the function at the point x0 .
By jump of the graph of a function we mean that whenever x approaches to x0 , f (x) does not
approach to f (x0 ), or f (x) takes a vertical jump to reach f (x0 ). If this jump is of  distance,
there will be points near by x0 whose images will be at least at  distance from f (x0 ).
More precisely, we say that the function f : R R is not continuous at a point x0 if there
exists some  > 0 neighborhood around f (x0 ), that is, (f (x0 ) , f (x0 ) + ) such that the image
of any neighborhood around x0 is not contained in (f (x0 ) , f (x0 ) + ), or
for any > 0, f (x0 , x0 + ) * (f (x0 ) , f (x0 ) + ).
The definition of continuity comes as the negation of the above statement. Thus a function
f is said to be continuous at a point x0 if for any given  > 0 there exists a > 0 (depending
on  and point x0 ) such that
f (x0 , x0 + ) (f (x0 ) , f (x0 ) + ), or
f (x) (f (x0 ) , f (x0 ) + ), whenever x (x0 , x0 + ),
x (x0 , x0 + ) f (x) (f (x0 ) , f (x0 ) + ),
or equivalently, |f (x) f (x0 )| < , whenever |x x0 | < ,
|x x0 | < |f (x) f (x0 )| < .
Thus informally, by continuity we mean that whenever x approaches to x0 , f (x) approaches
to f (x0 ). Later, we will prove this criterion of continuity.
Example 3.1. A constant function f : R R is defined by f (x) = c for all x R, where c
is a fixed real number. This function f is continuous at every point of the real line. To check
continuity at some point x0 R, we need to show that for any given  > 0 there is a > 0 (may
be depending on  and x0 ) such that f (x) (f (x0 ) , f (x0 ) + ) whenever x (x0 , x0 + ).
Since f (x) = f (x0 ) = c, we have f (x) (f (x0 ) , f (x0 ) + ) c (c , c + ), which is
always true no matter what x and  we take. Thus we can take any > 0 so that f (x)
(f (x0 ) , f (x0 ) + ) whenever x (x0 , x0 + ). This proves the continuity at the point x0 .
Since to satisfy the continuity criterion any positive number works as , which does not depend
on given  or the point x0 , so one can similarly prove the continuity at any other point y0 R
by assuming any positive number as for a given .
Example 3.2. Now we consider the identity function f (x) = x. To check the continuity at
some point x0 R, foe a given  > 0 we need to produce a > 0(may be depending on  and
x0 ) such that |x x0 | < |f (x) f (x0 )| < . But |f (x) f (x0 )| <  |x x0 | < . So if we
take any 0 < , then |x x0 | < |x x0 | <  |f (x) f (x0 )| < . Thus the function
is continuous at the point x0 . Here the existence of depends on  but does not depend on
the location of the point x0 . Thus we can similarly prove the continuity at any other point by
choosing a =  to satisfy the criterion of continuity.
Exercise 3.3. Use (-) definition to prove that f (x) = |x| is a continuous function on R.
Problem 3.4. Let f : R R be a continuous function and let c R. Show that if x0 R is
such that f (x0 ) > c, then there exists a > 0 such that f (x) > c for all x (x0 , x0 + ).
Solution. Because of continuity of the function at x0 , for any  > 0 there exist > 0 such
that x (x0 , x0 + ) f (x) (f (x0 ) , f (x0 ) + ), so if we can find an  > 0 such that
f (x0 ) > c, then the corresponding will serve our purpose. Thus we can choose  = |f (x02)c| .
3.5. Continuity of a function f on R or on a subset S of R
17
We say that a function f : R R is continuous on R if it is continuous at each point of the
real line.
Let S be a nonempty subset of R and a function f is defined on S, that is,
f : S R,
then we say that f is continuous at a point x0 S, if for a given  > 0 there exists a > 0
such that whenever x S (x0 , x0 + ), or equivalently, x S with |x x0 | < we have
|f (x) f (x0 )| < .
Now if the function is continuous at each point in S, we say that f is continuous on S.
Example 3.6. Suppose f : R R, defined by f (x) = 3x 2. To check the continuity of
the function f at some point x0 in R, we would try to fulfil the definition criterion. So for a
given  > 0 we need to search for a > 0 (may be depending on  and x0 ) such that whenever
|x x0 | < we have |f (x) f (x0 )| < . But in our case |f (x) f (x0 )| <  |3x 3x0 | < 
|x x0 | < 3 . Thus if we choose our = /3, then |x x0 | < implies |f (x) f (x0 )| < . (Note
that no less than /3 will satisfy the continuity criterion.)
We note that here our depends only on . And if we need to check continuity at any other
point y0 R and  as above, then the same as above will work to satisfy the continuity
criterion. And thus we see that the function is continuous at each point in R and hence we say
that f is continuous on R.
Problem 3.7. Let f : R R be such that for every x, y R, |f (x) f (y)| |x y|. Show
that f is continuous.
Solution. Let x0 R. To prove continuity at x0 , we need to show that for any given  > 0,
there exists a > 0 such that |x x0 | < |f (x) f (x0 )| < . But we are given that
|f (x) f (y)| |x y|, so if we choose our as  itself, then |f (x) f (x0 )| |x x0 | < = .
Thus for any given x0 R and a  > 0, we have (= ) such that |xx0 | < |f (x)f (x0 )| < .
So the function is continuous at any given x0 , hence for every x0 R.
Exercise 3.8. Use (-) definition to prove that f (x) = x2 is a continuous function on R.
Example 3.9. Suppose f : (0, 1) R, defined by f (x) = x1 . To check the continuity of
the function f at x0 (0, 1), we start with a given  > 0 and we try to search a (may be
depending on  and x0 ) such that whenever |x x0 | < we have |f (x) f (x0 )| < . Note
that |f (x) f (x0 )| <  |x x0 | |x01||x| < . Now suppose if some > 0 is going to
serve our purpose, then we expect that |x x0 | < should imply that |x01||x| is dominated
by certain number say m so that the product |x01||x| |x x0 | will be dominated by m and
further m should be at max , that is, m can be assumed at max  . Thus if we can find
such that |x x0 | < |x01||x| <  , then this is good enough for our purpose. Now
1 
|x0 ||x| < |x0 | < |x| x0  < x. Thus we want that x0 < x < x0 + should imply
x20 
x0  < x and it can be insured by taking such that x0 = x0  = 1+x0  . Unlike the
previous example, here not only depends on , but the point x0 also.
Problem 3.10. Let f : [0, 1] R be defined as follows: f (x) = 0 if x is 0 or irrational and
f (x) = n1 if x = m
n is rational (where m and n have no common factor). Prove that f (x) is
continuous at every irrational x and discontinuous at every non-zero rational x.
Solution. At any non-zero rational number r function assumes a non-zero value, and at every
irrational it takes the value 0, so if we approach to r through a sequence of irrational numbers
say (in ) in [0, 1], that is, in r, then (f (in )) is a constant sequence (0), hence converging to
0 6= f (r). So the function is discontinuous at r.
Now if b [0, 1] is an irrational number and  > 0, then by Archimedean property, there
exists n0 N such that n10 < . But there are only finite number of rationals with denominator
less than or equal to n0 in the interval [0, 1] say ( 12 , 13 , 23 , 14 , 24 , . . . , n10 , . . . , n0n1
0
). If we name these
18
rational numbers as r1 , r2 , . . . , rk , then only at these points in [0, 1] the functional value can be
greater than or equal to n10 and hence if we choose = min{|ri b| : 1 i k}, then all these ri s
are out side (b, b+). Thus for any rational in (b, b+) functional value is less than n10 and
at irrational it is 0. Thus for all x (b , b + ) we have f (x) [0, n10 ] (f (b) n10 , f (b) + n10 ).
Thus f is continuous at b.
Now to show continuity at x = 0 we can assume = min{|ri 0| : 1 i k}, which is
strictly positive in this case also, so that as above |x 0| < |f (x) f (0)| < n10 < .
3.11. Limit of a function
We say that a function has a limit A at x0 , if whenever x approaches to x0 , f (x) approaches
to A.
For  > 0, there exists > 0 such that
0 < |x x0 | < |f (x) A| < .
If such a number A exists, then it is unique. And we write limxx0 f (x) = A.
There are some more notions of limit. For example we say that limit of a function at x0 is
infinity if for any M > 0, there is a > 0 such that
0 < |x x0 | < M < f (x).
In this case we write limxx0 f (x) = . Similarly one can define limxx0 f (x) = .
We also say that limx f (x) = A, that is, limit of a function at is a real number A, if
for a given  > 0 there exists a positive real number M such that
M < x |f (x) A| < .
In a similar way one defines limx f (x) = . For any given K > 0, there exists some
M > 0 such that
x < M K < f (x).
Remark 3.12. As we expected in the beginning of the section that continuity means that
whenever x approaches to x0 , f (x) also approaches to f (x0 ). Since we approach to a point
through a convergent sequence, and therefore for a continuous function we hope the following
statement to be true:
whenever a sequence (xn ) converges to x0 , the sequence (f (xn )) converges to f (x0 ).
Question 3.13. Now the question arise that whether the above statement is sufficient to
conclude the continuity.
Theorem 3.14. (Sequential Criterion of Continuity ) A real valued function f is contin-
uous at x0 if and only if whenever a sequence (xn ) converges to x0 , then the sequence (f (xn ))
converges to f (x0 ).
Proof. Suppose P is the statement: f is continuous at x0 , and
Q is the statement: whenever a sequence (xn ) converges to x0 , then the sequence (f (xn ))
converges to f (x0 ). Here P if Q means P Q and P only if Q means P Q and thus P if
and only if (iff) Q means P Q.
Firstly, we prove P Q. So we assume that the function f is continuous at x0 and (xn )
converges to x0 and will try to prove that (f (xn )) converges to f (x0 ). So for a given  > 0,
we need to show the existence of some natural number N such that whenever n N , we must
have f (xn ) (f (x0 ) , f (x0 ) + ). But, f (x) (f (x0 ) , f (x0 ) + ) is guaranteed for all
x (x0 , x0 + ), where is some positive number, which exists by virtue of continuity of f
at x0 and it depends on above chosen , in other words, by continuity of f at x0 there exists
> 0 (depending on ) such that
if x (x0 , x0 + ), then f (x) (f (x0 ) , f (x0 ) + ).
In above line x can be chosen as xn for those n0 s such that xn (x0 , x0 + ). But by
convergence of (xn ) to x0 , we know that any given neighborhood of x0 contains all the terms
19
of the sequence except a finite number of terms, thus there exists some finite N N such that
xn (x0 , x0 + ) ( f (xn ) (f (x0 ) , f (x0 ) + )) for all n N .
Next we prove P Q which is equivalent to prove that the negation of P implies the negation
of Q.
So we assume that f is not continuous at x0 and want to conclude the negation of Q. But
the negation of Q is the following statement:
there exists a sequence (xn ) converging to x0 , but the sequence (f (xn )) does not converge to
f (x0 ).
Since f is not continuous at x0 , there exists a positive number  such that for any > 0,
f ((x0 , x0 + )) * (f (x0 ) , f (x0 ) + ), in other words for any > 0 there exists some x
depending on (suffix is just to show the dependence on ) such that x (x0 , x0 + ) and
f (x ) does not belong to (f (x0 ) , f (x0 ) + ). Since can go to zero, we can assume = n1 ,
so that x 1 (x0 n1 , x0 + n1 ), but f (x 1 ) is not in (f (x0 ) , f (x0 ) + ). Thus if we consider
n n
the sequence (yn = x 1 ), then by Sandwich Theorem yn x0 but f (yn ) 6 (f (x0 ) , f (x0 ) + )
n
for all n, hence f (yn ) 6 f (x0 ). 
Remark 3.15. The importance of sequential criterion is to prove the discontinuity of the
function. Because to prove continuity at certain point (through sequential criterion) one needs to
deal with all possible sequences converging to the point. And to characterize all such sequences
is not an easy task.
But the discontinuity at some point x0 can be ensured just by producing one sequence (xn )
converging to x0 such that the corresponding image sequence (f (xn )) does not converge to
f (x0 ).
A similar theorem can be proved about the limit of a function.
Theorem 3.16. Let f be a real valued function, then limxx0 f (x) = A( R) if and only if
whenever a sequence (xn ) converges to x0 , then the sequence (f (xn )) converges to A.
Remark 3.17. Thus to prove the non existence of the limit of a function at some point
x0 , we need to produce a sequence (xn ) converging to x0 , such that (f (xn )) is not convergent.
In other words, either we need to produce a sequence (xn ) such that (f (xn )) is unbounded,
or to produce two sequences (yn ) and (zn ) such that yn x0 and zn x0 , but (f (yn )) and
(f (zn )) do not converge to same limits.
Remark 3.18. One can use limits to define continuity. Following is the limit criterion of
continuity.
A function f is continuous at a point x0 if and only if lim f (x) = f (x0 ).
xx0

Problem 3.19. Let f : R R satisfy f (x + y) = f (x) + f (y) for all x, y R. If f is continuous


at 0, show that f is continuous at every point c R.
Solution. Since f (x + y) = f (x) + f (y) for all x, y R, f (0 + 0) = f (0) + (0) f (0) = 0 and
f (x)+f (x) = f (0) = 0 f (x) = f (x). Thus f (xy) = f (x)+f (y) = f (x)f (y). Now
to prove continuity at some arbitrary point x0 , we need to show that limxx0 f (x) = f (x0 ).
Which is equivalent to show that lim(xx0 )0 f (x)f (x0 ) = 0, or equivalently, lim(xx0 )0 f (x
x0 ) = 0 and this is same as proving limz0 f (z) = 0 if z = (x x0 ). But the last statement is
equivalent to continuity at 0.
Exercise 3.20. Suppose A < f (x) for all x (a, b). If c (a, b) and limxc f (x) = B, then
A B. Show that if f is continuous at c, then A 6= B. Find a function, which is discontinuous
at c and continuous at all other points, such that A = B.
Theorem 3.21. If limxx0 f (x) = A and limxx0 g(x) = B, then
(1) If h(x) = f (x) + g(x), then limxx0 h(x) = A + B
(2) If h(x) = f (x).g(x), then limxx0 h(x) = AB
20
f (x) A
(3) If h(x) = g(x) , then limxx0 h(x) = B , if B 6= 0.

Proof. The proof of this theorem is the direct consequence of Theorem 2.40. Let xn x0 , then
by limxx0 f (x) = A and limxx0 g(x) = B we have f (xn ) A and g(xn ) B and then by
application of Theorem 2.40 f (xn ) + g(xn ) A + B, f (xn ).g(xn ) AB. Further if B 6= 0,
then fg(x
(xn )
n)
A
B . 

The next theorem is obvious from the last theorem.


Theorem 3.22. If f (x) and g(x) are two continuous functions at x0 R, then f (x) + g(x) and
f (x).g(x) are also continuous at x0 . Further if g(x) 6= 0 for all x R, then fg(x)
(x)
is continuous
at x0 .
Above theorem helps us to prove continuity of many functions.
Example 3.23. Since f (x) = x is continuous and multiplication of two continuous functions
is again continuous, f (x).f (x) = x.x = x2 is a continuous function on R. One can prove
continuity of x2 just by using (-) definition to realize the importance of this multiplication
rule. Now if g(x) = x2 is again multiplied by f (x) = x, we get a new continuous function
g(x).f (x) = x2 .x = x3 . Proceeding in this way after finitely many multiplication of functions
we can show that any positive integer power of x is a continuous function on real line. Further
since constant function is continuous and addition of finitely many continuous functions is again
continuous we can conclude that polynomials are continuous functions on R.
Now if p(x) and q(x) are two polynomials such that q(x) 6= 0 for all x R, then the quotient
function p(x) 3x+2
q(x) is continuous on R. For example x2 +5 is continuous on R.

Exercise 3.24. Suppose f, g, h : R R be three functions such that f (x) g(x) h(x) for
all x R. If f and h are continuous at x0 and f (x0 ) = g(x0 ) = h(x0 ), prove that g is also
continuous at x0 .
Problem 3.25. If limxx0 |f (x)| = 0, then limxx0 f (x) = 0.
Solution. Suppose xn x0 , then since limxx0 |f (x)| = 0, |f (xn )| 0. Thus for given  > 0,
there is some N N such that for all n N , ||f (xn )| 0| <  |f (xn ) 0| < . Hence
f (xn ) 0. So limxx0 f (x) = 0.
Problem 3.26. Suppose f : [1, 3] R is a continuous function such that there exists a point,
in every neighborhood of 2, at which f takes the value 1. Show that f (2) = 1.
Solution. The statement of the problem suggests that we can get a sequence of points converg-
ing to 2 such that the value of the function at all the these points of the sequence is 1. More
precisely, for any n N there is a point say xn (2 n1 , 2 + n1 ) such that f (xn ) = 1. Now by
continuity since xn 2, we have f (xn ) f (2) but (f (xn )) is a constant sequence with each
term as 1. So by uniqueness of limit f (2) = 1.
Theorem 3.27. Let A and B are to nonempty subsets of R. Let f : A B and g : B R
be two functions such that f is continuous at some point x0 A and g is continuous at the
point f (x0 ) B, then the composite function g f : A R defined by g f (x) = g (f (x)) is
continuous at x0 .
Proof. This proof is easy by sequential criterion. Let (xn ) be a sequence converging to x0 . Then
by continuity of f at x0 , we have that f (xn ) f (x0 ). Now by continuity of g at f (x0 ) B,
we know that g (f (xn )) converges to g (f (x0 )), that is, g f (xn ) g f (x0 ). 
Example 3.28. Let f : R R be a function defined by f (x) = |x| and g : R R be defined by
g(x) = x3 +3. Since both f and g are continuous functions, the composite function g f : R R
given by g f (x) = g(f (x)) = g(|x|) = |x|3 + 3 is a continuous function.
21
Example 3.29. To check the continuity of the function
sin x1 , if x 6= 0
 
f (x) =
0, if x = 0
We need to produce one sequence (xn ) such that f (xn ) is not converging to 0. If we look at the
graph of the function it is touching y = 1 line again and again more frequently before reaching
to Y axis. The points where function is taking the value one is described by sin x1 = 1, or
equivalently, x1 = 2n + /2 x = (4n+1) 2
. Thus if we consider the sequence xn = (4n+1) 2
,
then xn 0, but f (xn ) = 1, hence the sequence (f (xn )) is a constant sequence and converges
to 1 6= f (0). This gives a contradiction to sequential criterion. Hence f is not continuous at 0.
Further to check the existence of the limit at x = 0. We notice that we already obtained
2
a sequence xn = (4n+1) 0, and f (xn ) 1. Now either we might hope that 1 is the limit
at 0 or if it is not the case we can try to find another sequence yn again converging to 0 such
that f (yn ) is not converging to 1. But by the graph it is clear that sin x1 is crossing the X-axis
in a dense way before reaching to Y -axis. And these points can be obtained by sin x1 = 0 or
1 1
x = n. So consider the sequence yn = n , which is converging to 0, but (f (yn )) is a constant
sequence converging to 0. Thus we have two sequences (xn ) and (yn ) both converge to 0, but
the sequences (f (xn )) and (f (yn )) do not converge to the same limit. And hence the given
function does not possess a limit at 0.
Remark 3.30. Though we defined continuity at a point. But in most of these examples we
are proving continuity at all points of the domain except in the last example where we proved
discontinuity at one point and in fact in this example we can not give any value for f (0), so
that modified function becomes continuous at 0. Thus the continuity of a function at a point
does not only depend on the functional value at the point but also at functional values in the
neighbouring points. Thus continuity is a point wise local (depending on some nbd) property
of the function. In the following example we show that the only continuity of the function is at
x = 0.
Problem 3.31. Determine the points of continuity for the function f : [1, 1] [0, 1] defined
by  
|x|, if x is rational
f (x) = .
0, if x is irrational
Solution. Here it is important to note that the both the functions |x| and constant 0 are
continuous, so the function f would be continuous on an open subinterval of [1, 1] where f
takes exactly one form (see Problem (4.10)). But in our case we can not find any subinterval
where f takes the value |x| and the same is true for 0 also and this is because we can not
separate rational and irrational by intervals. Since both graphs matches at x = 0, so we should
not hope continuity at any point except at 0.
Here we will use sequential criterion of continuity to prove discontinuity at any point x0 6= 0.
Recall that for any real number x0 there is a sequence of rationals (rn ) and irrationals (in )
converging to x0 (see Problem (2.38). But then f (rn ) = |rn | |x0 | and f (in ) = 0 0. Now
the function can not be continuous unless both the limits are same, that is, |x0 | = 0.
Now to prove continuity at 0, we consider any  > 0 and try to find > 0 such that
|x 0| < |f (x) f (0)| < . Now for x rational |f (x) f (0)| = |0 0| = 0 and we dont
need to do any extra work, but for irrational x, |f (x) f (0)| = ||x| 0| = |x|, so if we choose
 = , then |x 0| < implies |f (x) f (0)| < .
3.32. Boundedness of a function on set
Let S be a nonempty subset of R. We say that a function f : S R is bounded on S if the
image set of f , that is, f (S) = {f (x) R : x S} is bounded subset of R.
Remark 3.33. In Example (3.9) we observe that even if the function is continuous, the image
set of a bounded domain might be unbounded.
22
Question 3.34. Unlike as the example (3.9), if we need to draw the graph of a continuous
function on closed interval [a, b] (instead of open interval (a, b)), then the graph has to start
with the point (a, f (a)) and howsoever height and depth it can attain but it has to return back
and end at the point (b, f (b)). Thus intuitively, we feel that the function has to be bounded.
The following theorem gives a criterion for boundedness of a continuous function.
Theorem 3.35. Let f : [a, b] R be a continuous function on R. Then f is bounded on [a, b].
Proof. We will prove it by contradiction. Suppose without loss of generality that the image
set of f , that is, f ([a, b]) is unbounded above. So for any positive integer n N, there is a
point in [a, b] such that the image of this point will dominate n (otherwise n will be an upper
bound, a contradiction to assumption). Let us name this point as xn , so that f (xn ) n. This
way we can find a sequence (xn ) in [a, b] such that (f (xn )) is unbounded. Now since (xn ) is
bounded sequence in [a, b]. Then by Bolzano-Weierstrass theorem there is a subsequence (xnk ),
which converges. But as a xnk b, by Sandwich Theorem the limit of this sequence, say xn0
is again in the interval [a, b], so that limk xnk = xn0 , but f (xnk ) nk , that is, (f (xnk )) is
unbounded, and hence (f (xnk )) is not convergent. Giving a contradiction to sequential criterion
of continuity at the point xn0 . 
3.36. Maximum and Minimum for a function
Let f be a function defined on a subset S of R. If x0 S satisfies the property that
f (x0 ) f (x) for all x S, then we say that x0 is the maximum for f on S and in this
case f (x0 ) = max{f (x) : x S}, that is, f (x0 ) is the least upper bound of the image set
{f (x) : x S} of f . Similarly if y0 S satisfies the property that f (y0 ) f (x) for all x S,
then we say that y0 is the minimum for f on S and f (y0 ) is the minimum value of f .
Example 3.37. Let f : (0, 1) R be defined by
 4 1

f (x) = n , if x = n .
1
0, if x =6 n
Then 12 is the maximum for f on (0, 1) with maximum value of f as 2. On the other hand every
number in (0, 1), which is not of the form 1/n is the minimum for f on (0, 1) and minimum
value of f is 0.
Example 3.38. Let f : [3, 2)[0, 1) R be a continuous function defined by f (x) = x2 +1,
then 3 is the maximum for f on [3, 2) [0, 1) and f (3) = 10 is the maximum value of f .
Similarly 0 is the minimum for f on [3, 2) [0, 1) and f (0) = 1 is the minimum value of f .
Example 3.39. Let f : [1, 2) R be defined by f (x) = x2 + 1. Now there is no x0 [1, 2) such
that f (x0 ) f (x) for all x [1, 2). So maximum for f on (1, 2) does not exist. But minimum
of f on [1, 2) is 1 and the minimum value of f is 2.
Remark 3.40. From the above examples it is clear that maximum or minimum does not exist
even for a continuous function on a given set. whether there any criterion for the existence
of maximum or minimum for a continuous function. To answer this in an affirmative way one
has to make sure the boundedness of image set. And this can be insured by Theorem (3.35)
for a continuous function on a closed interval. Now since image of f on [a, b] is bounded, by
completeness property of R there will be sup and inf of f ([a, b]) in R and only thing remains to
see is that these sup and inf belongs to f ([a, b]) itself.
Theorem 3.41. Let f be a continuous function on a closed interval [a, b]. Then there exists
x0 and y0 in [a, b] such that x0 is the maximum for f on [a, b] and y0 is the minimum for f on
[a, b].
Proof. (*) As it is clear from the last Remark, f ([a, b]) is bounded and there is real number
M = sup f ([a, b]). We aim to prove that there is an element x0 [a, b] such that f (x0 ) = M .
Since M is supremum of f ([a, b]) so for any natural number n, M n1 cant be the upper
23
bound of f ([a, b]), hence there will be some element in f ([a, b]), which will dominate M n1 ,
but any element of f ([a, b]) is of the form f (x) for some x [a, b], so there will be some element
f (xn ) f ([a, b]), for some xn [a, b] such that M n1 f (xn ). And thus we can find a sequence
(xn ) in [a, b] such that M n1 < f (xn ) M , that is, the sequence (f (xn )) converges to M . If
this sequence (xn ) would be convergent, then by sequential criterion M would be image of its
limit. But we should not hope this because there may be many points in [a, b] where f can take
value M (you can draw some such figures to justify yourself) and (xn ) can be so chosen that its
two subsequences converge to different points whose image is M . So only thing we hope that
(xn ) would have a convergent subsequence such that image of its limit is M . And to prove this
we observe that (xn ) is a bounded sequence in [a, b], by Bolzano-Weierstrass Theorem there is
convergent subsequence of (xn ), say (xnk ). Now if (xnk ) converges to x0 and since a (xnk ) < b,
we have a x0 b. Now because f is continuous at x0 , and xnk x0 by sequential criterion
f (xnk ) f (x0 ). Further since f (xn ) M , its subsequence f (xnk ) also converges to M . Now
by the uniqueness of limit of a convergent sequence f (x0 ) = M .
In a similar way one can prove that there exists an element y0 [a, b] such that f (y0 ) =
inf f ([a, b]). 

Remark 3.42. Note that the above result just gives the existence of minimum and maximum
for a continuous function on the closed interval. But it does not say any thing specific about
where they are located in the interval. Later we will see certain criterion for possible location
of minimum and maximum for the function.

Observation 3.43. Suppose a function f : [a, b] R is continuous and f (a) is a negative value
whereas f (b) is some positive value. Then if we draw a continuous graph of the function staring
from (a, f (a)) such that it ends at (b, f (b). Then we need to cross the X-axis, that is, there will
be some point x0 [a, b] such that f (x0 ) = 0. And there is nothing special about X-axis, but
the graph has to cross each parallel line to X-axis which lies between the points (a, f (a)) and
(b, f (b). Thus intuitively we feel that whenever s is some intermediate value between f (a) and
f (b), that is, f (a) < s < f (b), there should exist some point xs (a, b) such that f (xs ) = s.

Theorem 3.44. (Intermediate Value Property ) Let f be a continuous function on a closed


interval [a, b]. And let s be some intermediate value between f (a) and f (b). Then there exists
a point x0 [a, b] such that f (x0 ) = s.

Proof. (*)(Students can leave the first paragraph of the proof.) In theorem (3.41) we found a
point in [a, b] whose image is the supremum of the image set. In this case also, we need to find
an element in [a, b] whose image is the a given number s in between f (a) and f (b). Without
loss of generality we can assume that f (a) < s < f (b). Now if we want to run the similar proof
as in Theorem (3.41) and since s is not the suremum or infimum of the image set, we need to
find a subset of the image set, whose supremum is s. Then it is natural to consider the subset
{f (x) : x [a, b] and f (x) s} of the image set f ([a, b]). Clearly s is the upper bound for
the subset and (by continuity of the graph ) we feel that it should be supremum also. But we
dont see any direct argument to prove it so we hope that if we would assume a contradiction,
then some argument would give a contradiction to continuity at a point where we hope the
function to take value s. But there might be many points in the set S = {x [a, b] : f (x) s},
where the function can take value s. So if we draw the figures of these possibilities we observe
that supremum of the set S is the point, where the function takes the value s in each case.
Let x0 = sup S. Clearly x0 [a, b]. Now after this search of x0 we can continue the idea of
assuming a contradiction that f (x0 ) 6= s and we can proceed to prove a contradiction. Since x0
is the supremum of S, there is a sequence (xn ) in S such that xn x0 and hence by continuity
f (xn ) f (x0 ). But xn S f (xn ) s f (x0 ) s. Can we use a similar arguments to
prove that f (x0 ) s. For this we observe that S (x0 , b] = and we can find a sequence (yn )
such that yn (x0 , x0 + bx
n so that yn x0 and f (yn ) s. But by continuity f (yn ) f (x0 )
0

and hence f (x0 ) s. Thus we conclude that f (x0 ) = s a contradiction to assumption that
24
f (x0 ) 6= s. So by contrapositive arguments our assumption f (x0 ) 6= s is wrong and hence
f (x0 ) = s.
In summery, we consider the set S = {x [a, b] : f (x) s}. Suppose x0 = sup S, so
there is a sequence (xn ) in S such that xn x0 and hence by continuity f (xn ) f (x0 ). But
xn S f (xn ) s f (x0 ) s. Further since S (x0 , b] = and we can find a sequence (yn )
such that yn (x0 , x0 + bx
n so that yn x0 and f (yn ) s. But by continuity f (yn ) f (x0 )
0

and hence f (x0 ) s. Thus f (x0 ) = s. 


Problem 3.45. Let f : [0, 1] R be continuous for all x in [0, 1]. If f (x) 6= f (y) when x 6= y,
that is, f is injective, prove that f (x) is either strict increasing or strict decreasing on [0, 1].
Solution. Suppose contradiction, that is, f is neither increasing nor decreasing. Since f is
not an increasing function, there exist x0 < x1 such that f (x0 ) > f (x1 ) and since f is not
a decreasing function, there exist y0 < y1 such that f (y0 ) < f (y1 ). Now there are many
possibilities like x0 y0 < y1 x1 , x0 y0 x1 y1 , x0 < x1 y0 < y1 , y0 < y1
x0 < x1 , y0 x0 y1 < x1 , y0 x0 < x1 y1 . In all these cases we will be able to find
x, y, z {x0 , x1 , y0 , y1 } such that x < y < z and (f (y) f (x))(f (y) f (z)) > 0. Now note
that if |f (y) f (x)| = |f (y) f (z)|, then f (x) = f (z), a contradiction. So assume WLG
|f (y) f (x)| < |f (y) f (z)|, which implies that f (x) is some point in between f (y) and f (z).
Thus by IVP there will exist some point say w in between y and z such that f (w) = f (x) again
a contradiction because x 6= w. Thus by contrapositive arguments f is either strict increasing
or strict decreasing.
Problem 3.46. Let f : R R be a continuous function which takes only irrational values.
Show that f is a constant function.
Solution. Suppose if f is not a constant function, then there are two points x, y R such that
f (x) 6= f (y). But then there will be many rational values in between f (x) and f (y). Suppose
r is such a rational point in between f (x) and f (y), therefore by IVP there is some point z in
between x and y such that f (z) = r, a contradiction to the fact that f assumes only irrational
values. So by contrapositive arguments f has to be a constant function.
Problem 3.47. Suppose g : [a, b] [a, b] is a continuous function (g is not necessarily one-
one and onto). Then prove that there is a point c [a, b] such that g(c) = c. If x satisfies
the condition g(x) = x, we call the point x as the fixed point of g. Hence show that
cos(10x) = x has a real solution in the interval (1, 1). Further show by one example that the
result is not true in general, if interval [a, b] is replaced by (a, b).
Solution. By applying IVT for the function g we can only guarantee that any point in between
g(a) and g(b) is of the form g(x) for some x [a, b], but one can think of many continuous
functions g such that a 6= g(a) = g(b) 6= b. So going back to our requirement that g(c) = c, we
mean that the graph of g intersects the line f (x) = x. This gives us an idea to find the root of
the equation f (x) = g(x) in [a, b]. Or if we define the function h(x) = f (x) g(x) = x g(x)
on [a, b], then to find the root of h in [a, b]. But h is a continuous function on [a, b] such that
h(a) = a f (a) 0 and f (b) = b f (b) 0. Thus by IVT for h there is a point c in [a, b] such
that h(c) = 0, because 0 lies between h(a) and h(b).
Now since g : [1, 1] [1, 1] defined by g(x) = cos(10x) is continuous, so there is a fixed
point say c [1, 1] for g, that is, g(c) = c but g(1) 6= 1 and g(1) 6= 1 so c (1, 1).
Next we consider the continuous function g : (0, 1) (0, 1) defined by g(x) = x2 . Note that
this function can not have any fixed point in the interval (0, 1).
Problem 3.48. Show that the polynomial x4 + 4x3 10 has at least two real roots.
Solution. We observe that P (x) = x4 + 4x3 10 being a polynomial is a continuous function
on R. Now observe that x4 + 4x3 10 = x4 (1 + 4/x 10/x3 ), so that we hope that in the
limit case |x| approaching to infinity the polynomial value approaches to infinity. But at x = 0
it takes the value 10. So it means to approach to infinity on both sides of Y -axis, the graph
25
has to cross both positive real axis and negative real axis at least once. Hence at least two
roots should be there. But if we want to prove that for some given K > 0, there exists some
M > 0 such that |x| > M implies that x4 (1 + 4/x 10/x3 ) > K. We know that it is easy
to see that lim|x| x4 = , so thing will be in our control if we can produce some positive
lower bound for (1 + 4/x 10/x3 ) for large x. For this we observe that for |x| > 16 we
have |4/x| < 1/4 and for |x| > 51/3 we have |10/x3 | < 1/4, then certainly for |x| > 16 we have
|4/x10/x3 | |4/x|+|10/x3 | < 1/2 and hence 1+4/x10/x3 > 1|4/x10/x3 | 1 12 = 21 .
4
Thus x4 + 4x3 10 = x4 (1 + 4/x 10/x3 ) x2 for all |x| > 16.
Note that though we have shown explicitly the lower bound for (1 + 4/x 10/x3 ), but it is
not needed to be that explicit. In fact one can say that since limx x1 = 0 limx x4 =
0, limx x13 = 0 and limx x103 = 0. Thus there exists M1 , M2 > 0 such that x > M1
| x4 | < 14 and x > M2 | x13 | < 41 . Thus for x M = max{M1 , M2 }, we have |4/x| + |10/x3 | <
4
1/2. And as above we can conclude that x4 + 4x3 10 = x4 (1 + 4/x 10/x3 ) x2 M 4 /2 > 0.
So between [0, M ] we can apply IVP to conclude that there exists some point c [0, M ] such
that f (c) = 0 because f (0) = 10 < 0 and f (M ) > 0.
Similarly we can show it on negative real axis.
Remark 3.49. Clearly by IVT the image of the continuous function f on [a, b] contains the
closed interval I whose endpoints are f (a) and f (b). But there might be many points in the
image f ([a, b]), which are not in the interval I. Now how large the image f ([a, b]) can be, surely
if m and M are inf and sup of f ([a, b]), then f ([a, b]) [m, M ], but by Theorem (3.41), there
are points x0 and y0 in [a, b] such that f (x0 ) = m and f (y0 ) = M . So again by IVT, any point
in [m, M ] is the image of some point in between x0 and y0 and hence image of some point in
[a, b]. Thus [m, M ] f ([a, b]), hence f ([a, b]) = [m, M ]. Thus we conclude the following result.
Theorem 3.50. If f : [a, b] R is a continuous function, then range of f is the closed
bounded interval [m, M ], where m, M are respectively infimum and supremum of the range
{f (x) R : x [a, b]}.
3.51. Uniform Continuity
We recall that we defined continuity by understanding discontinuity at some particular point
with the knowledge of functional values in the neighbourhood of this point. Thus it is a local
concept. And to say that the function is continuous on a set we mean that it is continuous at
each point of the set.
If we consider the function f : (0, 1) R defined by f (x) = x1 . Then by our previous
knowledge we know that it is continuous on (0, 1). But suppose if we want to conclude continuity
at x0 by the definition itself and assume our  > 0 as any positive number, then maximum value
 x2
of , which can serve our purpose, is 1+0 00x0 , which goes to zero as x0 0. Thus for any given
 > 0, we can not find any fixed working uniformly for all the points in (0, 1).
We say that a function is uniformly continuous on a set A if for any given  > 0, there exists
a > 0 such that if x, y A and |x y| < , then |f (x) f (y)| < .
Pictorially, if for a given  > 0, there exists a > 0 such that if we put the rectangle of width
2 and hight 2 with center on any point on the graph (x0 , f (x0 )) and sides parallel to axis,
then the graph should not upper and lower sides (sides of length 2).
Example 3.52. A function f , satisfying |f (x) f (y)| < M |x y| for all x, y A, is uniformly
continuous on A. In fact for any  > 0, = /M is such that |x y| < M |x y| < 
|f (x) f (y)| < .
Example 3.53. The function f (x) = x2 is uniformly continuous on [a, b], because |x2 y 2 | =
|x y| |x + y| |x y|(|x| + |y|) |x y|2(|a| + |b|). Thus by above example f is uniformly
continuous. But it is not uniformly continuous on R because |x2 y 2 | = |x y| |x + y| < 
 
|x y| < |x+y| . But |x+y| goes to zero if x, y . This show that we can not find uniform ,
working uniformly for all x, y R.
26
Theorem 3.54. (**) If f is continuous on [a, b], then it is uniformly continuous.
Proof. (**) Suppose contradiction that f is continuous on [a, b], but it is not uniformly contin-
uous on [a, b]. So there exists 0 > 0, such that for any > 0, there exist points x , y such that
|x y | < but |f (x ) f (y )| 0 . Since it is true for all > 0, for n N we can choose
= n1 so that there should exist points xn , yn such that |xn yn | < n1 , but |f (xn ) f (yn )| e0 .
Thus we can get two sequences (xn ), (yn ) in [a, b] such that the difference sequence (xn yn )
goes to zero. But (xn ) being a bounded sequence, it has a convergent subsequence say (xnk ).
Suppose if xnk x0 ( [a, b]), then ynk = xnk (xnk ynk ) x0 0 = x0 . Thus by continuity
at x0 , f (xnk ) f (x0 ) and f (ynk ) f (x0 ). But this implies that for 0 > 0 there exists
K1 , K2 such that k K1 |f (xnk ) f (x0 )| < 0 /2 and k K2 |f (ynk ) f (x0 )| < 0 /2.
Thus k max K1 , K2 implies |f (ynk ) f (xnk )| |f (ynk ) f (x0 )| + |f (xnk ) f (x0 )| 0 , a
contradiction to |f (xn ) f (yn )| e0 . 

4. Differentiability
We know that the derivative of a function f : R R at a point x0 R is the slop of the
tangent line at the point (x0 , f (x0 )) of the graph of the function. But it is not necessary in
general that we can draw a tangent line at any point of the graph. For example at point (0, 0)
of the graph of the function f (x) = |x|, we cant draw a tangent line.
In fact we say that a function is differentiable at the point x0 if the slop of the chords passing
through (x0 , f (x0 )) and (x, f (x)) approaches to a fixed value whenever x approaches to x0 , that
is if
f (x) f (x0 )
lim
xx0 x x0
exists. In this case we say that f has a derivative limxx0 f (x)f xx0
(x0 )
at x0 . If a function is
differentiable at all points of R, we can define the derivative function as
f (x) f (x0 )
f 0 (x0 ) = lim .
xx0 x x0
But in general if our function is defined on some interval I of the form (a, b], then there is
no point of talking about the derivative at a, because the function itself is not defined at a, but
the derivative at b means the left hand derivative at b. Note that
f (x) f (x0 ) f (x0 + h) f (x0 )
lim = lim .
xx0 x x0 h0 h
Thus f (x0 +h)f
h
(x0 )
can be considered as a new function in variable h defined in a neighborhood
of 0 except at the point 0 itself. And we say that the function f is differentiable at the point
x0 if the limit of this quotient function exists whenever h goes to zero.
Remark 4.1. Note that when we say that the function is differentiable at x0 , we consider the
existence of the limit of certain quotient expression, which involves the value of the function in
a nbd of x0 . This implies that there is a nbd of x0 where the function is defined.
Problem 4.2. Prove that f (x) = x3 is differentiable function on R.
Solution. Let x0 R. To show that f is differentiable at x0 , we need to show the ex-
istence of limit limh0 f (x0 +h)f
h
(x0 )
. We note that the function g : R R defined by
g(h) = x0 + h is continuous and f is polynomial it is also continuous on R. Since com-
position of two continuous functions is continuous h f (g(h)) = f (x0 + h) is continuous
again since the addition of continuous functions is again continuous h f (x0 + h) f (x0 )
is continuous on R. Thus again by Theorem (3.21), f (x0 +h)fh
(x0 )
is a continuous function in
3 3
R {0}. Though at {0} this is not defined butlimh0 f (x0 +h)f (x0 )
h = limh0 (x0 +h)h(x0 ) =
3 3 x3 +h3 +3x20 h+3x0 h2 x30
limh0 (x0 +h)h(x0 ) = limh0 0 h limh0 (h2 + 3x20 + 3x0 h) = 3x20 . Thus
= at
f (x0 +h)f (x0 )
x0 limh0 h = 3x20 . So the function f is differentiable at any given x0 R and the
value of the derivative is 3x20 and now one can say that the derivative function is f 0 (x) = 3x2 .
27
Problem 4.3. Let f : R R be such that |f (x) f (y)| |x y|2 for all x and y in R. Prove
that f (x) is a constant for all x.
Solution. Using the hypothesis, we have
|f (x0 + h) f (x0 )| |x0 + h x0 |2
|h|2 f (x0 + h) f (x0 ) |h|2
h2 f (x0 + h) f (x0 ) h2
h2 f (x0 + h) f (x0 ) h2

h h h
f (x0 + h) f (x0 )
h h
h
f (x0 + h) f (x0 )
lim h lim lim h
h0 h0 h h0
Since first and last term goes to 0, by sandwich the middle expression also goes to 0. Thus f (x)
is a constant.
0
Problem 4.4. f (0) = 0 and f (0) = 1. For a positive integer k, show that
1n x x  x o 1 1
lim f (x) + f +f + ... + f = 1 + + ... +
x0 x 2 3 k 2 k
x x
f( ) 1 f( )0
Solution. Observe that for any strict positive real number , limx0 x = limx0 x =

x
f( )f (0) f (0+h)f (0)
lim x 0 x = 1
limh0 h = 1 f 0 (0) = 1 .

Problem 4.5. Prove that if f : R R is an even function and has a derivative at every point,
0
then the derivative f is an odd function.
f (x+h)f (x)
Solution. Suppose x > 0, and by definition f 0 (x) = limh0 h =

limh0 f (xh)f
h
(x)
= limh0 f (x+
h)f

h
(x)
= limh0

f (x+h)f

h
(x)
= f 0 (x).

Problem 4.6. If f : R R is differentiable at some point c R, show that f 0 (c) = lim(nf (c +


1/n) nf (c)). However, show by example that existence of such a limit does guarantee the
differentiability of the function at the point c.
Solution. If f is differentiable at c, then f 0 (c) = limh0 f (c+h)f
h
(c)
. Since this limit exist it has
to be same for through any sequence (hn ) converging to 0, particularly for (hn ) = ( n1 ). Thus
1
f (c+h)f (c) f (c+hn )f (c) f (c+ n )f (c)
f 0 (c) = limh0 h = limhn 0 hn = lim 1 0 1 = limn (nf (c + n1 )
n n
nf (c)).
Note that lim x af (x) = A if and only if for any sequence (xn ) converging to a we have
lim xn af (xn ) = limn f (xn ) = A. But in general, limn f (xn ) = A for one sequence
xn a does not imply the existence of lim x af (x) = A. So in our case also we should
not hope that limn (nf (c + n1 ) nf (c)) can imply in general that limh0 f (c+h)f
h
(c)
exists.
Suppose  
|x c|, if x is rational
f (x) = .
0, if x is irrational
Then f (c) = 0. WLG assume that c is a rational number, so that c+ n1 is also a rational number.
Thus the limit limn (nf (c + n1 ) nf (c)) = limn n|c + n1 c| n.0 = 1 exists. But for
the differentiability at c we need that limh0 f (c+h)f
h
(c)
should exist, but if (in ) is a sequence
f (c+in )f (c)
of irrationals converging to 0, then limin 0 in = limin 0 00
in = limin 0 0 = 0. Thus
f (c+h)f (c)
limit limh0 h does not exist through sequential criterion.
Following theorems are important to show differentiability of functions.
28
Theorem 4.7. Suppose that f, g : R R are differentiable at some point x0 R. Prove that
(1) (f + g) is differentiable at x0 and (f + g)0 (x0 ) = f 0 (x0 ) + g 0 (x0 ),
(2) (f g) is differentiable at x0 and (f g)0 (x0 ) = f 0 (x0 ) g(x0 ) + f (x0 ) g 0 (x0 ),
(3) if g(x) 6= 0 for all x, then fg(x)
(x)
is also differentiable at x0 and
 0
f g(x0 )f 0 (x0 ) g 0 (x0 )f (x0 )
(x0 ) = .
g g 2 (x0 )
Proof. (**) For proof see some reference book. 
Theorem 4.8. Let f be a function defined on an interval I. If f is differentiable at a point
a I, then f is continuous at c.
f (x)f (c)
Proof. limxc f (x) f (c) = limxc xc limxc (x c) = f 0 (c) 0 = 0. 
Remark 4.9. This shows that if the function is not continuous at some point, then it can not
be differentiable at that point. But converse of this is not true.
Problem 4.10. Use (-) definition to show that the function f : [1, 1] R defined by
 2 
x + 3, if x > 0
f (x) =
3(x + 1), if x 0
is continuous at x = 0 but not differentiable.
Solution. First we show the continuity of f at x = 0. Let  > 0 be given positive number,
and we need to find > 0 such that x (0 , 0 + ) f (x) (f (0) , f (0) +). Thus we
2 2
want that 0 < x < should imply 3  < x + 3 < 3 +   < x <   < x < 
and < x < 0 should imply 3  < 3s + 3 < 3 +  /3 < x < /3. Thus if we choose
min{ , /3}, then x (, ) f (x) (f (0) , f (0) + ). So f is continuous at 0 and it is
continuous at all other point because it is a polynomial there.
Now to check differentiability at 0, we need to see whether the limit limh0 f (0+h)f
h
(0)
exists.
By expression of the function it is clear that on left of 0 the function is a two degree polynomial
and on right it is a line. So on right we hope that the slope of the line should be limit of this
expression, that is, if hn > 0 and hn 0+ , then limhn 0+ 3hnh+33n
= limhn 0+ 3 = 3. Now if
the limit limh0 f (0+h)f
h
(0)
exist, then limh0 f (0+h)fh
(0)
should be equal to 3. Let us try to
f (0+h)f (0) 2
h +33
find this limit limh0 h = limh0 h = limh0 h = 0. Thus for any sequence
f (0+hn )f (0)
hn 0 we have limhn 0

hn
= limhn 0 hn = 0. Thus we have two sequence
(hn ) and (hn ), both converging to 0 but corresponding functional sequences do not converge to
f (0+h)f (0)
unique number. So limit limh0 h does not exist, hence f is not differentiable at 0.
Problem 4.11. If f (x) is differentiable at x0 , prove that |f (x)| is also differentiable at x0 ,
provided f (x0 ) 6= 0.
Solution. Since f is differentiable at x0 , it is continuous at x0 . Now since f (x0 ) 6= 0, we
assume without loss of generality that f (x0 ) > 0. Thus for  = f (x0 )/2 > 0 we can find a > 0
such that for x (x0 , x0 + ) we have f (x) (f (x0 ) f (x0 )/2, f (x0 ) + f (x0 )/2) and hence
f (x) > 0. Now for x (x0 , x0 + ) we have |f (x)| = f (x), which is differentiable at x0 .
Problem 4.12. If both f (x) and g(x) are differentiable at x0 , prove that h(x) =
max{f (x), g(x)} is also differentiable at x0 , provided f (x0 ) 6= g(x0 ).
Solution. WLG we assume that f (x0 ) > g(x0 ) and as in the above problem we can use
continuity of the function F (x) = f (x) g(x) at x0 to find a > 0 such that f (x) g(x) =
F (x) > 0 for all x (x0 , x0 + ). Thus for all x in the nbd (x0 , x0 + ) we have
h(x) = max{f (x), g(x)} = f (x). Therefore the function h is differentiable at x0 because f is
differentiable at x0 .
29
Theorem 4.13. Let A and B are to nonempty subsets of R. Let f : A B and g : B R be
two functions such that f is differentiable at some point x0 A and g is differentiable at the
point f (x0 ) B, then the composite function g f : A R defined by g f (x) = g (f (x)) is
differentiable at x0 . And
(g f )0 (x0 ) = g 0 (f (x0 )) f 0 (x0 ).
Proof. This proof is easy by sequential criterion of limit existence. Let (xn ) be a se-
quence converging to x0 . Then since f is differentiable at x0 , it is continuous also and
we have that f (xn ) f (x0 ). Now by differentiability hence by continuity of g at
f (x0 ) B, we know that g (f (xn )) converges to g (f (x0 )), that is, g f (xn ) g
f (x0 ). Now limxx0 gf (x)gf
xx0
(x0 )
= limxn x0 gf (xxnn)gf
x0
(x0 )
= limxn x0 gff (x n )gf (x0 )
(xn )f (x0 )
f (xn )f (x0 ) g(f (xn ))g(f (x0 )) f (xn )f (x0 )
limxn x0 xn x0 = limxn x0 f (xn )f (x0 ) limf (xn )f (x0 ) xn x0 = g 0 (f (x0 ))
f 0 (x0 ). 
Remark 4.14. A function is not differentiable at x0 if the limit limh0 f (x0 +h)fh
(x0 )
does not
exist.
 By sequentialcriterion of limit existence (see Remark (3.17)), it means that (hn 0) 6
f (x0 +hn )f (x0 )
hn l .
We defined maximum and minimum for a function in the last section. Suppose if x0 (a, b)
is the maximum of a continuous function f defined on [a, b] and f is differentiable at x0 , then
we feel that the tangent line at x0 is parallel to X-axis. (One can draw some figures to visualize
it.)
4.15. Local Maximum and Local Minimum for a function
Let f be a function defined on a subset S of R. If for x0 S there exist a > 0 such that
f (x0 ) f (x) for all x (x0 , x0 + ) S, then we say that x0 is the local maximum for f on
S. Similarly one can define local minimum of f .
Theorem 4.16. Let f : [a, b] R be a function. Suppose that f has either a local maximum
or local minimum at x0 (a, b). If f is differentiable at x0 , then f 0 (x0 ) = 0.
Proof. Suppose f has a local maximum at x0 , then there a > 0 such that x (x0 , x0 +)
f (x) f (x0 ). And hence x (x0 , x0 ) f (x)f
xx0
(x0 )
0 limxx0 f (x)f
xx0
(x0 )
0
f (x)f (x0 ) f (x)f (x0 )
f 0 (x0 ) 0. Similarly x (x0 , x0 + ) xx0 0 limxx0 xx0 0 f 0 (x0 ) 0.
Hence f 0 (x0 ) = 0. 
Remark 4.17. Note that the theorem is not valid if x0 is the end point of the interval. For
example f : [0, 1] R, defined by f (x) = 1 x has both end points as local extremum but f 0
is never zero.
Remark 4.18. Above theorem can be used to prove the following important theorem, known
as Rolles theorem.
Theorem 4.19. (Rolles Theorem) Let f : [a, b] R be a continuous function. If f (a) = f (b)
and f is differentiable in (a, b), then there is point x0 (a, b) such that f 0 (x0 ) = 0.
f (x)f (c)
Proof. If f is a constant function on [a, b], then for x, c in (a, b), xc = 0
limxc f (x)f
xc
(c)
= 0. And if f is not a constant function, then by Theorem (3.41), there
are x0 , y0 [a, b] suchthat f (x0 ) = sup{f (x) : x [a, b]} and f (y0 ) = inf{f (x) : x [a, b]}.
But f being a non constant function f (x0 ) > f (y0 ). Suppose WLG f (x0 ) 6= f (a) = f (b) so
that x0 (a, b), but x0 being the maximum by Theorem (4.16), f 0 (x0 ) = 0. 
Remark 4.20. Geometrically, Rolles theorem implies the existence of a point at which the
tangent line is parallel to a line joining the points (a, f (a)) and (b, f (b)). Now we can ask the
question that if f (a) 6= f (b) in the above theorem, can one still guarantee the existence of
a point in (a, b), where the tangent line is parallel to a line joining the points (a, f (a)) and
(b, f (b)).
30
Theorem 4.21. (Mean Value Theorem) Let f : [a, b] R be a continuous function. If f is
differentiable in (a, b), then there is point x0 (a, b) such that f 0 (x0 ) = f (b)f
ba
(a)
.

Proof. Consider the function g : [a, b] R defined by g(x) = (b a)f (x) (f (b) f (a))x, then
g(a) = bf (a) af (b) = g(b) and g is continuous on [a, b] and differentiable in (a, b), we can
apply Rolles theorem to function g to conclude the existence of a point x0 (a, b) such that
g 0 (x0 ) = 0 f 0 (x0 ) = f (b)f
ba
(a)
. 

Theorem 4.22. Suppose f : [a, b] R is continuous on [a, b] and differentiable on (a, b). Then
the function is constant if and only if f 0 (x) = 0 for all x (a, b).
Proof. It is easy to see that if f is constant function, then derivative of f is 0 at all points in
(a, b). (See the proof of Rolles Theorem.)
Now we assume that derivative of f is 0 at all points in (a, b). Let a < x b. Then by MVT
there is some point cx (a, x) such that f (x) f (a) = (x a)f 0 (cx ) = (x a).0 = 0. Thus f
is constant function. 
Remark 4.23. Note that if we use directly the definition of f 0 (x) = 0 for all the points, then
we dont see any easy way to conclude that the function is constant.
Problem 4.24. Suppose f is continuous on [a, b], differentiable on (a, b) and satisfies f 2 (a)
0
f 2 (b) = a2 b2 . Then show that the equation f (x)f (x) = x has at least one root in (a, b).
Solution. We need to show the existence of a point c (a, b) such that f 0 (c)f (c) = c. If we
want to apply Rolles theorem here we should try to find out some function g such that g 0 (c) = 0
should imply f 0 (c)f (c) = c or 2f (c)f 0 (c) 2c = 0. Now we can assume g(x) = f 2 (x) x2 so
that g(a) = f ( a) a2 = f 2 (b) b2 = g(b). Thus by Rolles theorem there is some c in (a, b)
such that g 0 (c) = 0 or f 0 (c)f (c) = c.
Problem 4.25. Let f and g be functions continuous on [a, b], differentiable on (a, b) and let
0 0
f (a) = f (b) = 0. Prove that there is a point c (a, b) such that g (c)f (c) + f (c) = 0.

Solution. Assume h(x) = f (x)eg(x) so that h is differentiable (by Theorem (4.13) and (4.7))
on (a, b) and h(a) = h(b) = 0. Thus by Rolles theorem there exists some c in (a, b) such that
h0 (c) = 0, that is f 0 (x)eg(x) +f (x)eg(x) g 0 (x) is equal to 0 at x = c. Thus [f 0 (c)+f (c)g 0 (c)]eg(c) = 0
which implies that f 0 (c) + f (c)g 0 (c) = 0.
Problem 4.26. Use the Lagrange mean value theorem to prove the following inequalities:
(1) | sin x sin y || x y | for all x, y R.
(2) x1
x < ln x < x 1, for x > 1.
(3) ex 1 + x, for all x R.
Solution. 1) Consider the function f (x) = sin x and apply MVT in the interval [x, y], so that
sin xsin y
xy = f 0 (c) = cos c. This implies that |sin|xy|
xsin y|
= | cos c| < 1 for all x, y R.
2) Consider the function f (x) = ln x and apply MVT in the interval [1, x], so that ln x1
xln 1
= 1c
where 1 < c < x or x1 < 1c < 1, which implies that x1 < ln xln
x1
1
< 1. And thus x1
x < ln x < x1,
for x > 1.
x e0
3) Consider the function f (x) = ex and apply MVT in the interval [0, x], so that ex0 = ec
x
for some c in [0, x]. But 1 < ec < ex , which implies that 1 < e x1 < ex and hence ex 1 + x,
and ex (x 1) < 1 for all x (0, ). Similarly one can apply MVT in [x, 0] to conclude it in
(, 0).
Problem 4.27. Let f be a differentiable function on [0, 1]. Suppose f (0) = f (1) = 0 and
f ( 14 ) = f ( 43 ) = 1. Show that there are three distinct elements a, b, c (0, 1) such that f 0 (a) +
f 0 (b) + f 0 (c) = 0.
31
Solution. According to the given data we can apply Mean Value Theorem in subintervals
0
[0, 1/4], [1/4, 3/4] and [3/4, 1]. By MVT there exists an a (0, 1/4) such that f (a) =
f (1/4)f (0)
1/40 = 4.
0 f (3/4)f (1/4)
Similarly b (1/4, 3/4) such that f (b) = 3/41/4 = 0. And there exists an c (3/4, 0)
0 f (1)f (3/4)
such that f (c) = 13/4 = 4. Thus there exist three distinct elements a, b, c (0, 1) such
0 0 0
that f (a) + f (b) + f (c) = 0.
Problem 4.28. Let a > 0 and f : [a, a] R be continuous suppose f 0 (x) exists and f 0 (x) 1
for all x (a, a). If f (a) = a and f (a) = a, then show that f (x) = x for every x (a, a).
Solution. Consider g : [a, a] R defined by g(x) = f (x) x so that g is continuous on
0 0
[a, a] and differentiable on (-a,a). Further g (x) = f (x) 1 0 for all x (a, a). i.e.; g is
decreasing and g(a) = 0 = g(a). Assume that g is not constant, then there exists x0 (a, a)
such that g(x0 ) 6= 0. Now if g(x0 ) < 0, one can apply MVT on the [x0 , a] so that there exists
a point c1 in (x0 , a) such that g 0 (c1 ) = g(a)g(x
ax0
0)
> 0, which is a contradiction. Similarly if
g(x0 ) > 0, then one can apply MVT on the [a, x0 ] so that there exists a point c2 in (a, x0 )
such that g 0 (c2 ) = g(xx00 (a)
)g(a)
> 0, which is also a contradiction. So our assumption is wrong.
Thus g(x) is a constant function.
Problem 4.29. Suppose f is continuous on [a, b] and differentiable on (a, b). Assume further
that f (b) f (a) = b a. Prove that for every positive integer n, there exist distinct points
c1 , c2 , ..., cn (a, b) such that f 0 (c1 ) + f 0 (c2 ) + . . . + f 0 (cn ) = n.
Solution. Clearly if n = 1, then the result is true by MVT. Now if n = 2 assume a0 = a, a1 =
(a + b)/2, a2 = b and apply MVT in both subintervals. Applying MVT in [a0 , a1 ] to get a
c1 (a0 , a1 ) such that f 0 (c1 ) = f (aa11)f
a0
(a0 )
= f (a(ba)/2
1 )f (a0 )
. And in [a1 , a2 ] we get c2 (a1 , a2 )
f (a2 )f (a1 ) f (a2 )f (a1 ) f (a1 )f (a0 ) f (a2 )f (a1 )
such that f 0 (c2 ) = a2 a1 = (ba)/2 . Thus f 0 (c1 ) + f 0 (c2 ) = (ba)/2 + (ba)/2 =
f (a2 )f (a0 ) f (b)f (a)
(ba)/2 = (ba)/2 = 2.
Similarly one can divide [a, b] in to n equal subintervals say [ai1 , ai ], each of length ba n .
And if we apply MVT in each of the subinterval [ai1 , ai ], then there exists ci (ai1 , ai ) such
that f 0 (ci ) = f (aai )f (ai1 )
i ai1
= f (a(ba)/n
i )f (ai1 )
. Thus we get
n n
X X f (ai ) f (ai1 ) f (b) f (a)
ci = = = n.
(b a)/n (b a)/n
i=1 i=1

Theorem 4.30. Suppose f : [a, b] R is continuous on [a, b] and differentiable on (a, b).
(1) If f 0 (x) 6= 0 for all x (a, b), then f is one-one,
(2) If f 0 (x) 0 for all s (a, b), then f is increasing on [a, b],
(3) If f 0 (x) > 0 for all s (a, b), then f is strict increasing on [a, b],
(4) If f 0 (x) 0 for all s (a, b), then f is decreasing on [a, b],
(5) If f 0 (x) < 0 for all s (a, b), then f is strict decreasing on [a, b],
Proof. All these results will be proved by contradiction. Otherwise we might be in trouble.
(1)Suppose if f is not one-one, i.e., there are two points x0 , y0 (a, b) such that f (x0 ) = f (y0 ),
then by MVT there is a point c in between x0 and y0 such that 0 = f (x0 ) f (y0 ) = (x0
y0 )f 0 (c) f 0 (c) = 0 giving a contradiction to f (x) 6= 0 for all x (a, b).
(2) Suppose f is not an increasing function on [a, b], then there exists two points x0 , y0 [a, b]
such that x0 < y0 and f (x0 ) 6 f (y0 ) f (x0 ) > f (y0 ). Which implies that f (xx00)f y0
(y0 )
< 0,
then by MVT there exists c (x0 , y0 ) such that f (xx00)f (y0 )
y0 = f 0 (c) and hence f 0 (c) < 0 a
contradiction to assumption.
(3) Similarly if f is not an strict increasing function on [a, b], then there exists two points
x0 , y0 [a, b] such that x0 < y0 and f (x0 ) 6< f (y0 ) f (x0 ) f (y0 ). Which implies that
32
f (x0 )f (y0 ) f (x0 )f (y0 )
x0 y0 0, then by MVT there exists c (x0 , y0 ) such that x0 y0 = f 0 (c) and hence
0
f (c) 0 a contradiction to assumption.
(4) and (5) can be proved in similar lines. 
Remark 4.31. Converse of the result is not true. For example f (x) = x3 is one one and strict
increasing even then f 0 (0) = 0.
Problem 4.32. For x > 1 and x 6= 0, prove that
(1) (1 + x) > 1 + x, whenever < 0, or > 1,
(2) (1 + x) < 1 + x, whenever 0 < < 1.
Solution. Let us assume the function g(x) = (1 + x) 1 x. Then g 0 (x) = (1 + x)1 =
[(1 + x)1 1] and g 00 (x) = ( 1)(1 + x)2 .
First suppose < 0, then for x > 0 we have 0 < (1 + x)1 < 1, which implies that g 0 (x) > 0
so g is strict increasing in the interval (0, ). Since g(0) = 0, g(x) > g(0) = 0, hence the result.
Next if x < 0, then 1 < (1 + x)1 and g 0 (x) < 0. Thus g is decreasing on (1, 0) and hence if
x (1, 0), then g(x) > g(0) = 0, the result.
Secondly suppose 0 < < 1, then for x > 0 we have 0 < (1 + x)1 < 1, which implies that
g (x) < 0, or g strict decreasing, that is, g(x) < g(0). Next if 1 < x < 0, then g 0 (x) > 0 so
0

g(x) increasing and hence g(x) < g(0).


Now we assume that > 1, then for x > 0, we have 1 < (1 + x)1 and g 0 (x) > 0, so we
have g(x) > g(0) = 0. And if 1 < x < 0, then 0 < (1 + x)1 < 1 and g 0 (x) < 0 implies g is
strict decreasing and hence g(x) > g(0) = 0.
Problem 4.33. Prove that the equation x7 + 5x3 6 = 0 has exactly one real root.
Solution. Since p(x) = x7 + 5x3 6 is a polynomial of odd degree, p(x) has at least one
root (justify it). For x < 0, p(x) < 6 so no root in this case. But p0 (x) = 7x6 + 15x2 > 0
for all x > 0, so p is strict increasing there but p(0) = 6 so p(x) has to cross positive real
axis to approach to infinity. One can also say that if p has at least two roots on positive
real axis, then p0 must have at least one root there (Rolles Theorem), a contradiction because
p0 (x) = 7x6 + 15x2 > 0.
Problem 4.34. Show that there is no real number x, which satisfies the equation x2 = x sin x+
cos x 3.
Solution. We need to show that f (x) = x2 x sin x cos x + 3 has no root in R. For this
we first observe that f 0 (x) = x(2 cos x). Thus f 0 (x) > 0 for x > 0 and f 0 (x) < 0 for x < 0,
which implies that f is strict increasing on (0, ) and strict decreasing on (, 0). So the
function f can have at max one root on (0, ) and at max one root on (, 0), but f (0) = 2
so f (x) > f (0) = 2 for all x 6= 0. Thus f can not have any root.
Theorem 4.35. (Cauchy Mean Value Theorem) Let f and g be continuous functions
defined on [a, b] and differentiable on (a, b). If g 0 (x) 6= 0 for all x (a, b), then there exists
c (a, b) such that
f (a) f (b) f 0 (c)
= 0 .
g(a) g(b) g (c)
Proof. Consider the function h(x) = (f (a) f (b)) g(x) (g(a) g(b)) f (x). Then h is contin-
uous on [a, b] and differentiable on (a, b), moreover h(a) = f (a)g(b) f (b)g(a) = h(b). So we
can apply Rolles Theorem to guarantee the root of the function h in (a, b), that is, there exists
c (a, b) such that h0 (c) = 0, which implies (f (a) f (b)) g 0 (c) (g(a) g(b)) f 0 (c) = 0 hence
the result. 
x2
Problem 4.36. Using Cauchy Mean Value theorem prove that 1 2! < cos x and cos x <
2 4
1 x2! + x4! for x 6= 0.
33
2
Solution. Consider f (x) = 1 cos x and g(x) = x2 , so that g 0 (x) = 2x 6= 0 for all x 6= 0.
0 (c)
Therefore by Cauchy Mean Value Theorem, there exists c (0, x) such that fg(x)g(0)
(x)f (0)
= fg0 (c)
1cos x0 sin(c) 1 x2
x2 0
= 2c < 2 1 cos x < 2 . Similarly apply CMVT on [x, 0].
2 4 3
For the next part assume that f (x) = 1 cos x x2 and g(x) = x4! so that g 0 (x) = x3! Now
0 (c)
apply CMVT in [0, x], so that there exists c [0, x] such that fg(x)g(0)
(x)f (0)
= fg0 (c) , which implies
1cos xx2 /2 sin cc
that x4 /4!
= c3 /6
. Here one can again apply MVTC on [0, c] with new functions sin xx
3 sin cc cos d1 cos d1
and x /6 to get a d in [0, c] such that c3 /6
= d2 /2
. Now from the first part d2 /2
< 1.
1cos xx2 /2
Thus x4 /4!
< 1, which implies the result.
Remark 4.37. We found certain necessary conditions for local minimum and local maximum
in Theorem (4.16). Following theorem gives a sufficient condition for local maximum.
Theorem 4.38. Let f : (a, b) R be continuous at c (a, b). If for some > 0, f is increasing
on (c , c) and decreasing on (c, c + ), then f has a local maximum at c.
Proof. To prove that f has maximum at c, we need to show that if x is in some nbd of c, then
f (x) f (c). If x (c , c), then we wish to show that f (x) f (c). But increasing property of
the function in open interval (c , c) cant guarantee any comparison for f (x) and f (c), on the
other hand if y (c, c) such that x < y, then f (x) f (y). But this y can go near to c and we
can use continuity at c. Thus f (x) f (y) for all y (x, c) implies f (x) limyc f (y) = f (c)
(See Exercise (3.20)). Similarly we can prove that if x (c, c + ), then f (x) f (c). 
Remark 4.39. Note that continuity of f at c is a necessary condition. Because if we consider
a function f (x) = |x|, x 6= 0 and f (0) = 1, then 0 is not a local maxima even though f is
increasing on (, 0) and decreasing on (0, ).
Observation 4.40. In Theorem (4.38) if we assume differentiability in (c , c) and (c, c +
), then by theorem (4.30) increasing property of the function in (c , c) is equivalent to
nonnegativity of the derivative of f and similarly decreasing property of the function in (c, c+)
is equivalent to f 0 (x) 0 for all x (c, c + ). Thus following statement is a Corollary to
Theorem (4.38).
Corollary 4.41. Let f : (a, b) R be continuous at c (a, b). If for some > 0, f is
differentiable on (c , c) (c, c + ) and f 0 (x) 0 f 0 (y) for all x (c , c) and y (c, c + ),
then f has a local maximum at c.
Observation 4.42. If we assume the differentiability of the function at c in the above Corollary.
Then by Theorem (4.16) f 0 (c) = 0. Now we observe that the condition of f 0 in the Corollary
can be deduced by assumption that f 0 is decreasing in a nbd of c and f 0 (c) = 0. Note that by
theorem (4.30) decreasing property of f 0 can be the implication of f 00 (x) 0 for all x in a open
nbd of c.
Next Corollary shows that the assumptions of previous corollary can be derived by assuming
the second time differentiability of f at the point c only but with strict inequality f 00 (c) < 0.
And in this case we dont guarantee the increasing or decreasing property of f 0 as discussed
above.
Corollary 4.43. Let c (a, b) and f : (a, b) R be two times differentiable at c such that
f 0 (c) = 0 and f 00 (c) < 0, then f has a local maximum at c.
f 0 (x)f 0 (c)
Proof. Here we assumed that f 00 (c) exists. By definition f 00 (c) = limxc xc . Since
|f 00 (c)| |f 00 (c)|
f 00 (c)
< 0 and > 0 so that
2 f 00 (c) + 2 < 0, there exists a > 0 such that f (x) 0 exists
for all x (c , c) (c, c + ) and
f 0 (x) f 0 (c) |f 00 (c)| 00 |f 00 (c)|
 
f 00 (c) , f (c) + .
xc 2 2
34
00 0
But f 00 (c) + |f 2(c)| < 0 and f 0 (c) = 0, so fxc
(x)
< 0 for all x (c , c) (c, c + ). And hence
0 0
f (x) 0 for all x (c , c) and f (x) 0 for all x (c, c + ). Thus by previous corollary c
is the local maximum. 
Remark 4.44. Similar results are true for minimum also.
Theorem 4.45. Let f : (a, b) R be continuous at c (a, b). If for some > 0, f is decreasing
on (c , c) and increasing on (c, c + ), then f has a local minimum at c.
Corollary 4.46. Let f : (a, b) R be continuous at c (a, b). If for some > 0, f is
differentiable on (c , c) (c, c + ) and f 0 (x) 0 f 0 (y) for all x (c , c) and y (c, c + ),
then f has a local minimum at c.
Remark 4.47. Converse of the above results is not true. Suppose f (x) = x2 sin2 x1 for x 6= 0


and f (0) = 0, then 0 is the local minimum of f . But for any > 0, function is not increasing
in (0, ) and there are points in (0, ), where the derivative is strictly negative.
Corollary 4.48. Let c (a, b) and f : (a, b) R be two times differentiable at c such that
f 0 (c) = 0 and f 00 (c) > 0, then f has a local minimum at c.
Remark 4.49. Let f (x) = x4 . Then f 00 (0) = 0, but 0 is the minimum of f . This shows that
the converse of above corollary is not true.
4.50. Convexity, Concavity and Point of Inflection
Definition: Suppose f : (a, b) R is differentiable on (a, b). We say that f is convex on
(a, b) if f 0 is strictly increasing on (a, b). And f is concave on (a, b) if f 0 is strictly decreasing
on (a, b).
By Theorem (4.30) it is clear that if f is second time differentiable, then f is convex on (a, b)
if f 00 (x) > 0 for all x (a, b). And f is concave on (a, b) if f 00 (x) < 0 for all x (a, b).
Example 4.51. f (x) = sin x is convex on (, 2) and concave on (0, ). And g(x) = x2 is
convex on any interval in R.
Definition: Suppose f : (a, b) R is continuous at a point c (a, b). Then the point c is
said to be point of inflection if there exists a > 0 such that either
f is convex on (c , c) and f is concave on (c, c + ), or
f is concave on (c , c) and f is convex on (c, c + ).
Above conditions can also be written as
f 0 is strictly increasing on (c , c) and f 0 is strictly decreasing on (c, c + ), or
f 0 is strictly decreasing on (c , c) and f 0 is strictly increasing on (c, c + ).
Now suppose f is two times differentiable. Then c is point of inflection if
f (x) > 0 for all x (c , c) and f 00 (x) < 0 for all x (c, c + ), or
00

f 00 (x) > 0 for all x (c , c) and f 00 (x) > 0 for all x (c, c + ).
Example 4.52. x = 0 is the point of inflection for the function f (x) = x3 .
Following Theorem gives a necessary condition for the point of inflection when the double
derivative exists at the point.
Theorem 4.53. Let c (a, b) and f : (a, b) R be such that f 00 (c) exists. If f has a point of
inflection at c, then f 00 (c) = 0.
Proof. Since c is point of inflection, there is some > 0 suppose if f 0 is strictly increasing on
0 0 (c)
(c, c) and f 0 is strictly decreasing on (c, c+), then f 00 (c) = limxc f (x)f
xc 0. Similarly
we can see that f 00 (c) 0. And hence the result. 
Remark 4.54. The converse of the above theorem is not true. That is double derivative can
be 0 at a point but it might not be a point of inflection for example f (x) = x6 . It might also
happen that at point double derivative may not exist even then it is a point of inflection. for
1
example f (x) = x 5 has 0 as the point of inflection.
35
Following theorem gives a sufficient condition for the point of inflection when triple derivative
exists at the point.
Theorem 4.55. Let c (a, b) and f : (a, b) R be such that f 00 (c) = 0 and f 000 (c) 6= 0, then
c is the point of inflection.
000
Proof. (*) Without loss of generality we assume that f 000 (c) > 0, so that f 000 (c) |f 2(c)| > 0.
00 00 (c)
Now by the definition of derivative f 000 (c) = limxc f (x)f
xc . Since the right hand side limit
|f 000 (c)|
exists for 2 > 0, we can find a > 0 such that
f 00 (x) f 00 (c) |f 000 (c)| 000 |f 000 (c)|
 
f 000 (c) , f (c) .
xc 2 2
000 00
Thus 0 < f 000 (c) |f 2(c)| < fxc
(x)
for all x (c , c) (c, c + ). And hence f 00 (x) < 0 for all
x (c , c) and f 00 (x) > 0 for all x (c, c + ). Thus f 0 is strict decreasing on (c , c) and
strict increasing on (c, c + ). Hence by definition c is point of inflection. 
Remark 4.56. Converse of above result is not true. That is even if c is a point of inflection
and triple derivative exists at c, the f 000 (c) might be 0. For example f (x) = x7 has 0 as the
point of inflection but f 000 (0) = 0.
Example 4.57. The function f (x) = (x 1)(x 2)(x 3) is a three degree polynomial so it has
to go to infinity and minus infinity whenever x approaches to infinity and minus infinity. Clearly
it crosses X axis at 1, 2 and 3, and it can not cross at any other point because a polynomial
of degree three can have at most three roots. This is many times differentiable function so
we can apply second derivative test to find local maxima and local minima. Possible points
0 2
of extremum due to Theorem (4.16) are given by f (x) = 0, that00 is, 3x 12x + 11 = 0 or
x = 2 1/ 3. Then by Corollary (4.43) or Corollary (4.48), and f (x) = 6x 12, we conclude
that x1 = 2 1/ 3 is local minimum and x2 = 2 + 1/ 3 is local maximum because f 00 (x1 ) < 0
and f 00 (x2 ) > 0. Since f 0 is positive before x1 , between x1 and x2 it is negative and after x2 it
is again positive. So by increasing decreasing test theorem (4.30), function is increasing before
x1 , decreasing between x1 and x2 and again increasing after x2 . Similarly one can check that
before 2 function is concave and after 2 it is convex. And hence 2 is the point of inflection.
Observation 4.58. We observe that if a function f is differentiable on [a, b], then the tangent
line at a can be moved continuously as a chord on the curve to coincide with the tangent line
at the point b. This gives a feeling that for each real number in between f 0 (a) and f 0 (b), one
can find the position of a chord in, whose slop is . And hence by MVT there is a point in [a, b]
at which the derivative of f is .
Theorem 4.59. (Darbouxs Theorem (**) ) Suppose if f is a real differentiable function
on (a, b), then its derivative function satisfies the Intermediate Value Property, that is, if x, y
(a, b) and f 0 (x) < < f 0 (y), then there is a point c in between x and y such that f 0 (c) = .
Proof. WLG consider x < y. If we consider the function g : [a, b] R, defined by g(t) = f (t)
t, then g is differentiable in [x, y] and g 0 (x) = f 0 (x) < 0 and g 0 (y) = f 0 (y) > 0. We recall
that g 0 (x) = limzx g(z)g(x)
zx , which is strict less than zero means there is neighbourhood around
0 0 0
x, say (x , x + ), such that g(z)g(x)

g 0 (x) <  = |g (x)| 3g (x) < g(z)g(x) < g (x) < 0
zx 2 2 zx 2
for all z (x , x + ). Thus x is a maxima of g [x, x + ). Similarly one can show that y
is a maxima of g in (y , y]. Thus in [x, y] function g is not a constant and hence because
of continuity of g, it attains its minima at some point c in (x, y). But g being differentiable in
(x, y) implies that g 0 (c) = 0. Thus f 0 (c) = . 
Observation 4.60. In all above theorems we have seen the applications of differentiability.
But to insure differentiability of a function at a point we need to show the existence of the limit
of quotient of two functions, both of which independently approaches to 0 in the limit.
36
4.61. LHospital Rule
In the following theorems, which are known as LHospital (pronounced as Lopeetal ) Rule,
we try to find such limits in some special cases.
Theorem 4.62. Let f and g be real valued functions defined on (a, b). If both f and g are
differentiable at x0 (a, b) such that g 0 (x0 ) 6= 0 and f (x0 ) = 0 = g(x0 ), then limxx0 fg(x)
(x)
=
f 0 (x0 )
g 0 (x0 ) .
f (x)f (x0 )
f (x) f (x)f (x0 ) limxx0 xx0 f 0 (x0 )
Proof. limxx0 g(x) = limxx0 g(x)g(x0 ) = g(x)g(x0 ) = g 0 (x0 ) . 
limxx0 xx0

Remark 4.63. In the next theorem we consider the interval [x0 , b) and replace the condition
g 0 (x0 ) 6= 0 by g 0 (x) 6= 0 for all x (x0 , b)
Theorem 4.64. Let f and g be real valued continuous functions defined on [x0 , b) and
differentiable on (x0 , b). If g 0 (x) 6= 0 for all x (x0 , b) and f (x0 ) = 0 = g(x0 ), then
0 (x) 0 (x)
limxx+ fg(x)
(x)
= limxx+ fg0 (x) , provided limxx+ fg0 (x) exists.
0 0 0

f (x) f (x)f (x0 )


Proof. (*) limxx+ = limxx+
g(x) g(x)g(x0 ) .
Since f, g are not differentiable at x0 , we cant
0 0
follow the same proof as in the last theorem. But on the other hand we can apply CMVT on
0 (c )
the [x0 , x], to ensure the existence of a point cx (x0 , x) such that fg(x)g(x
(x)f (x0 )
0)
= fg0 (c x
x)
. But
f (x) f (x)f (x0 ) 0 (c )
x x+ +
0 cx x0 and hence limxx+ = limxx+ = limcx x+ fg0 (c x
. Now if the
0 g(x) 0 g(x)g(x 0 ) 0 x)
f 0 (x) f 0 (cx ) f 0 (x)
limit limxx+ g 0 (x) exists, then limcx x+ g0 (cx ) = limxx+ g0 (x) . This completes the proof. 
0 0 0

Remark 4.65. A similar theorem can be proved for left hand side limits.
Theorem 4.66. Let f and g be real valued continuous functions defined on (a, x0 ] and
differentiable on (a, x0 ). If g 0 (x) 6= 0 for all x (a, x0 ) and f (x0 ) = 0 = g(x0 ), then
0 (x) 0 (x)
limxx fg(x)
(x)
= limxx fg0 (x) , provided limxx fg0 (x) exists.
0 0 0

Proof. Proof is similar to last theorem. 


Remark 4.67. In previous theorems we dealt with an expression of the form 00 . In fact one
can deal with other indeterminate forms like 0 , , 1 , . . . by converting them to 00
or
. And the following generalized version of LHospital Rule is useful to find limits in these
cases.
Theorem 4.68. Let a < b and let f, g be differentiable on (a, b) such that g 0 (x) 6= 0
for all x (a, b). Suppose that exactly one of the following two conditions, that is, either
i) limxa+ f (x) = 0 = limxa+ g(x), or
ii) limxa+ g(x) = is satisfied.
Then following statement holds.
0 (x)
If limxa+ fg0 (x) = L R {, }, then limxa+ fg(x)(x)
= L.
Remark 4.69. A similar theorem is also true if limxa+ is replaced with limxb . We leave
the proof of these theorem but we will use them.
4.70. Applications to root finding problems
We know certain formulas for finding the real roots of the polynomial of degree less or equal to
4. But there is no exact formula for finding the roots of a polynomial of degree greater or equal
to 5. There are many functions for which the analytical methods guarantee the existence of
roots, but we cant find the exact values of these roots. For this we try to find the approximate
value with a desired error. If we rewrite the equation f (x) = 0 as x = g(x) for some function
g. Suppose if g is continuous and there exists an interval [a, b] such that image of [a, b] under g
lands up in [a, b] itself, that is, g([a, b]) [a, b]. Then g has a fixed point in [a, b] (See Problem
(3.47)). Thus problem of finding roots for f is now converted to find fixed point of function g.
37
4.71. Fixed point iteration method
To fond such a fixed point we define a sequence with x1 as some point in [a, b] and then
xn = g(xn1 ) for all n 2. If we can ensure the convergence of (xn ), then by continuity of
function g, the limit of this sequence will satisfy the equation x = g(x) and hence it will be a
fixed point of g. Now we would like to put certain more conditions on g so that the sequence
defined above is convergent.
Theorem 4.72. Let g : [a, b] [a, b] be a differentiable function such that
|g 0 (x)| < 1 for all x [a, b].
Then there is exactly one fixed point for g in [a, b] and the sequence, defined by xn = g(xn1 )
and for any x1 [a, b], converges to the fixed point.
Proof. (*) Clearly g has a fixed point (See Problem (3.47)). Firstly we aim to prove that there is
exactly one fixed point of g in [a, b]. Suppose if x0 and y0 are two fixed points for g in [a, b]. Then
by mean value theorem there is some c in between x0 y0 such that |x0 y0 | = |g(x0 ) g(y0 )| =
|g 0 (c)| |x0 y0 |. Now we use the bound for g 0 and get |x0 y0 | |x0 y0 | < |x0 y0 |, a
contradiction. Hence There is only one fixed point for g. Now we will show that the sequence
(xn ) converges to the fixed point of g. For this we observe that by MVT there is some cn2 in
between xn1 and xn2 such that
|xn xn1 | = |g(xn1 ) g(xn2 )| = |g 0 (cn2 )| |xn1 xn2 | |xn1 xn2 |.
Since inequality |xn xn1 | |xn1 xn2 | can be proved similarly for all n 3, we can use
it repeatedly to get
|xn xn1 | |xn1 xn2 | |xn2 xn3 | < . . . < n2 |x2 x1 |.
Thus sequence satisfies Cauchy criterion (see Q.No. 5 of Problem Sheet 1), hence it is convergent.
If l is the limit of the sequence, that is xn l, then by continuity of g, g(xn ) g(l). And
xn = g(xn1 ) l = g(l). 
Example 4.73. At least one root of f (x) = x3 7x + 1 is lying in [0, 1]. Since the function is
strict decreasing in [0, 1], there can be at most one root. Thus exactly one root lies in [0, 1]. To
find a sequence converging to this root we rewrite the equation x3 7x+1 = 0 as x = 17 [x3 +1]. If
we assume g(x) = 17 [x3 + 1], then g([0, 1]) [0, 1] and g 0 (x) = 3x2 /7 3/7 < 1 for all x [0, 1].
Thus last theorem guarantees the convergence of the sequence xn = g(xn1 ) for any x1 [0, 1].
Now to find approximate value of the root l with error less than 1/100. We observe that
n1 n1
|lxn | = |g(l)g(xn1 )| 73 |lxn1 | 37 |lx1 | 37 |10| is true for all n 2. Thus
3 n1 1 3 1 1
/ log 73 .

we need to find xn such that 7 < 100 (n 1) log 7 < log 100 n 1 > log 100
Now one can find the positive integer satisfying this condition. And then we need to iteratively
find xn .
Problem 4.74. Show that g(x) = (x2 1)/3 has a unique fixed point on the interval [1, 1].
Solution. Since g is continuous and some how we know that g([1, 1]) [1, 1], then g
0
has a fixed point in [1, 1]. We note that g (x) = 2x 3 , which is strict positive for x > 0
and strict negative for x < 0. As g(x) is strictly decreasing on [1, 0], in order to check
g([1, 0]) [1, 1], it is enough to check at end points. i.e.; g(1) = 0 and g(0) = 1 3 ,
1
therefore g([1, 0]) [1, 1]. Similarly since g(x) is strictly increasing on [0, 1], and g(0) = 3 ,
g(0) = 0, we have g([0, 1]) [1, 1]. Thus g([1, 1]) [1, 1] and g has a fixed point in [1, 1].
Now to show that g has a unique fixed point. As g : [1, 1] [1, 1] is a differentiable
0
function such that |g (x)| = | 2x 2
3 | 3 < 1 for all x [1, 1], g has exactly one fixed point in
[1, 1].
Remark 4.75. Some times if we know that the root of the function f (the fixed point of g )
lies in [a, b] and |g 0 (x)| < 1 for all x [a, b] but g([a, b]) 6 [a, b], then the following theorem
guarantees the existence of a proper subset [a1 , b1 ] [a, b] such that g([a1 , b1 ]) [a1 , b1 ].
38
Theorem 4.76. Let l0 be a fixed point of g(x). Suppose  > 0 be such that g is differentiable
on [l0 , l0 + ] and |g 0 (x)| < 1 for all x [l0 , l0 + ]. Then the sequence defined by
xn = g(xn1 ) and x1 [l0 , l0 + ], converges to l0 .
Proof. (*) To prove this theorem we first prove that g([l0 , l0 + ]) [l0 , l0 + ]. If
x [l0 , l0 + ], then |l0 g(x)| = |g(l0 ) g(x)| = |g 0 (c)| |l0 x| |l0 x|  < ,
hence g(x) [l0 , l0 + ]. Now g fulfils the condition of Theorem (4.72) with interval [a, b] as
[l0 , l0 + ] so we conclude the rest. 
Remark 4.77. If g is invertible in a nbd of the fixed point l0 of g, then l0 is a fixed point for
g 1 also and we can apply the iteration method to g 1 instead of g.
4.78. Newtons Method of iteration
Suppose we write the equation f (x) = 0 as x = x ff0(x) f (x)
(x) . In this case g(x) = x f 0 (x) if this
g satisfies the required conditions of theorem, then the sequence defined by
f (xn )
xn+1 = xn
f 0 (xn )
converges to the root of f . This is known as Newtons Method. For example one can find
x2 2
the root of x2 2 = 0, by defining g(x) = x ff0(x) x 1
(x) = x 2x = 2 + x and sequence as
xn+1 = 21 (xn + x2n ) with x1 chosen as follows. Now the convergence of (xn ) is because of the
theorem and observation that g 0 (x) = 21 x12 . Now if we want our = 3/4, that is we want
| 12 x12 | 34 . Thus we need x12 54 25 < x. Now we observe that a fixed point of g lies in
[ 25 , 2] and in this interval g 0 is dominated by 3/4 also g([ 25 , 2]) [ 25 , 2]. So it satisfies the
conditions of the theorem and we can choose x1 [ 25 , 2].

4.79. Taylors Theorem


If we recall the geometric interpretation of Newtons Iterative Method, it shows that we
approximated the function with a line with the only information as value of the function at
a point and value of its derivative at that point. And instead of finding exact root of the
function we found the root of this linear approximation of the function and called that as the
first approximation to the exact root. Now, can we think of some other approximations to
the function just with the knowledge of some more information at a fixed point? For this we
recall that linear approximation was obtained with the information of functional value and its
derivative value at the point x0 . And in fact this line was defined with the requirement that it
should take value f (x0 ) at x0 and the derivative of the line should also match with the derivative
of the function at x0 .
In Taylors Theorem we try to approximate the function with a polynomial of degree n, which
not only matches with the function at x0 but first, second, . . ., up to nth derivative also match
with the corresponding derivatives of the function at x0 . And error of approximation is given
in terms of n + 1th derivative of the function at some point near x0 .
Thus Taylor Theorem relates function with its higher derivatives. And it is a generalization
of Mean Value Theorem, which approximates the function with a constant polynomial and
relates function and its first derivative only.
Theorem 4.80. (Taylors Theorem) Let n N and f : [a, b] R be a function such that its
successive derivatives f 0 , f 00 , . . . , f (n) exist and are continuous on [a, b] and f (n) is differentiable
on (a, b), that is, f (n+1) exists on (a, b). If x0 [a, b], then for any x [a, b], f (x) can be
approximated by an nth degree Taylor polynomial
f 00 (x0 ) f (n) (x0 )
Pn (y) = f (x0 ) + f 0 (x0 )(y x0 ) + (y x0 )2 + . . . + (y x0 )n ,
2! (n)!
39
which matches with the function along with its derivatives up to degree n at x0 , and there exists
c in between x and x0 such that the remainder Rn (x) = f (x) Pn (x) has the expression
f (n+1) (c)
Rn (x) = (x x0 )n+1 .
(n + 1)!
Proof. (*) Firstly, we fix some x0 [a, b] and observe that Taylor polynomial, Pn of degree n
(k)
defined as above, satisfies Pn (x0 ) = f (k) (x0 ) for all 0 k n. Now let us fix any other point
x [a, b]. Our aim is to show that there exists some point c in between x and x0 such that
f (n+1) (c)
f (x) = Pn (x) + (x x0 )n+1 .
(n + 1)!
For this define a new function on [a, b] as
(t) = f (t) Pn (t) K(t x0 )n+1 ,
where K is constant, which is determined by the equation (x) = 0, that is, K = f(xx (x)Pn (x)
0)
n+1 .
0 00 n
Now we observe that (x0 ) = 0 = (x0 ) = (x0 ) = . . . = (x0 ). Now we apply Rolles
Theorem for the function between the interval with end points as x0 and x. Because (x0 ) =
0 = (x) = 0, there exists some point c1 in between x0 and x such that 0 (c1 ) = 0, but we also
have 0 (x0 ) = 0, so again by application of Rolle,s theorem in the interval with end points as
x0 and c1 , there exists some point c2 in between x0 and c1 such that 00 (c2 ) = 0. In fact we can
repeatedly apply Rolles theorem in the smaller and smaller intervals of nested type such that
there exists c in all those nested intervals and hence in the interval with end points x0 and x,
such that (n+1) (c) = 0. But this implies that f (n+1) (c) 0 (n + 1)!K = 0. Thus equating
f (n+1) (c) f (x)Pn (x)
the two values of K we get (n+1)! = (xx0 )n+1
, or equivalently,

f (n+1) (c)
f (x) = Pn (x) + (x x0 )n+1 .
(n + 1)!

Theorem 4.81. Let x0 (a, b) and n 2. Suppose f : [a, b] R be such that f, f 0 , f 00 , . . . , f n
exist and continuous on (a, b) and f 0 (x0 ) = f 00 (x0 ) = . . . = f n1 (x0 ) = 0. Then, if n is even
and f n (x0 ) > 0, then f has local minimum at x0 . Similarly, if n is even and f n (x0 ) < 0, then f
has local maximum at x0 . Further if n is odd and f n (x0 ) 6= 0, then x0 is a point of inflection.
Proof. The conclusion of Taylors Theorem for approximating by a polynomial of degree n 1
(n)
can be written as f (x) = Pn1 (x) + (x x0 )n f n!(c) . Using the assumption f 0 (x0 ) = f 00 (x0 ) =
(n)
. . . = f n1 (x0 ) = 0, we see that Pn1 (x) = f (x0 ), so that f (x) = f (x0 ) + (x x0 )n f n!(c) .
Now since f n (x) is continuous at x0 and f n (x0 ) 6= 0, there exists a > 0 such that for all
x (x0 , x0 + ) the sign of f n (x) is same as of f n (x0 ) (Justify it). And hence the sign of
f n (c) is same as of f n (x0 ). Now if n is even, then (x x0 )n > 0 and further if f n (x0 ) > 0, so
that f n (c) is also positive. Thus for all x (x0 , x0 + ) we have f (x) f (x0 ) > 0, showing
that x0 is local minimum. Similarly one can prove the other part. Now suppose that n is odd
and W.L.G. f n (x0 ) > 0, in a nbd of x0 it is positive. Hence f n1 (x) is strict increasing in the
nbd, so that f n1 (x0 ) = 0 implies that f n1 (x) is strict negative and strict positive on left and
right of x0 resp. But it further implies that f n2 (x) is strict decreasing on left of x0 again since
f n2 (x0 ) = 0 so it is negative there. We can repeatedly use this argument to conclude that
f 0 (x) is strict decreasing on left and strict decreasing. And hence x0 is the point of inflection
in this case. 
Problem 4.82. Suppose f is a thrice differentiable function on [1, 1] such that f 000 is contin-
uous on [1, 1] and f (1) = 0, f (1) = 1 and f 0 (0) = 0. Using Taylors theorem prove that
f 000 (c) = 3 for some c (1, 1).
40
Solution. Since f is given three times differentiable, we apply Taylors formula for n = 2
2 3
around the point 0. Thus f (x) = f (0) + f 0 (0)x + f 00 (0) x2 + f 000 (c) x6 . Thus we find the values at
2 3 2 3
1 and 1 as f (1) = f (0) + f 00 (0) 12 + f 000 (c1 ) 16 and f (1) = f (0) + f 00 (0) 12 + f 000 (c1 ) (1)
6 . Now
3 3
by subtraction, f (1) f (1) = f 000 (c1 ) 16 f 000 (c1 ) (1) 00 000
6 . This implies that 6 = f (c1 ) + f (c2 ).
000
But f is continuous we can consider a continuous function h(x) = f (x) 3 such that 000

h(c1 ) + h(c2 ) = 0. Now if h(c1 ) = 0, then done otherwise h(c1 ) and h(c2 ) have different sign so
by IVP, there exist some c in between c1 and c2 such that h(c) = 0, that is f 00 (c) = 3.
Problem 4.83. Using Taylors theorem, for any k N and for all x > 0, show that
1 1 1 1
x x2 + ... + x2k < log(1 + x) < x x2 + ... + x2k+1 .
2 2k 2 2k + 1
Solution. For all x > 0 and for n N, f (x) = log(1 + x) is a function such that its successive
0 00
derivatives f , f , . . . , f (n) exist and are continuous on [0, x] and f (n) is differentiable on (0, x).
i.e.; f (n+1) exists on (0, x). Therefore by Taylors theorem, there exists c (0, x) such that
log(1 + x) Pn (x) = Rn (x), where Pn (x) is a Taylor polynomial given by
0 x2 00 xn (n)
Pn (x) = f (0) + xf (0) + f (0) + . . . + f (0)
2 n!
x(n+1) (n+1)
and remainder term Rn (x) = (n+1)! f (c) for some point c in (0, x). It is easy to find
that f (n) (x) = (1)n1 (n 1)!(1 + x)n so
that f n (0) = (1)n1 (n 1)! Thus Pn (x) =
Pn n1 xn n n1 xn+1
i=1 (1) n and Rn (x) = (1) (1 + c) n+1 . And if n = 2k, then Rn (x) 0 and if
n = 2k + 1, then Rn (x) 0. Thus log(1 + x) P2k (x) and log(1 + x) P2k+1 (x). Thus one
can conclude the result.

5. Series
Let (an ) be a sequence of real numbers. The expression of the form
a1 + a2 + a3 + . . .
is known as infinite series or series and it is denoted by
P
n=1 an . Sum of fist n terms of the
series, that is, a1 + a2 + a3 + . . . + an is called as nth partial sum and denoted by Sn , that is,
Xn
Sn = ak = a1 + a2 + a3 + . . . + an .
k=1
Thus one can define the sequence of partial sums, that is, (Sn ). We say that the series is
convergent if the sequence of partial sums is convergent, that is there is some real number s R
such that Sn s. In this case number s is called limit of the sequence of partial sums of the
series or sum of the series. And if the sequence of partial sums is not convergent we say that
the series diverges.
Example 5.1. The series 1
P
n=1 n(n+1) converges because the sequence of partial sums (Sn =
1 1 1 1 1 1 1 1 1
2 + 6 + . . . + n(n+1) = 1 2 + 2 3 + . . . + n n+1 = 1 n+1 ) converges.

Example 5.2. 1
P
n=1 n is known as harmonic series. But is not convergent because the sequence
of partial sums (Sn ) is monotonic increasing sequence and it can be shown that the sequence
diverges to infinity by showing that one of its subsequence (S2k ) dominates the sequence (1+ k2 ),
which itself diverges to infinity. Now to show that S2k > 1 + k2 , let us write S2k = S1 +
Pk Pk 1 1 1 Pk 1 1
m=1 (S2m S2m1 ) = 1 + m=1 ( 2m1 +1 + 2m1 +2 + . . . + 2m ) 1 + m=1 ( 2m + 2m + . . . +
1 m1 times) = 1 +
P k 2m1 k
2m )(2 m=1 ( 2m ) = 1 + 2 .
P n
Example 5.3. For 1 < r < 1, n=1 r is a seriesn known as geometric series. And it is
convergent because Sn = 1 + r + r + . . . + rn1 = 1r
2 1
1r 1r .
41
P 1
Example 5.4. n=1 log(1 + n ) diverges because Sn = log(n + 1) diverges to infinity.
P n
Example 5.5. n=1 (1) diverges because (Sn ), which has two convergent subsequences of
even and odd terms, converging to 0 and 1 respectively, is not convergent.
Remark 5.6. As we noticed earlier that for the convergence of a sequence first few terms
does not matter. Similarly for the convergence of a series what actually matters is that how it
behaves eventually. In fact first finite number of terms does not matter for the convergence.
Theorem 5.7. If N N, then the series
P P
n=1 an converges if and only if n=N an converges.
P P
Proof. Series n=N an converges. Series n=1 an+N 1 converges. Sequence of partial
1 1 P 1
sums (Sn ), where Sn = nk=1 ak+N 1 = n+N ak = n+N ak N
P P P
k=N k=1 k=1 ak = Sn+N 1
SN 1 , converges. There exists some s R such that Sn s. Sn+N 1 SN 1 s
Sn+N 1 Ps + SN 1 Sn s + SN 1 P (See Theorem (2.15)). Sequence of partial sums of
the series n=1 an converges. Series
n=1 an converges. 
5.8. Necessary conditions for convergence
As weP know that the behaviour of the sequence (an ) is closely related to the convergence of
series an . Clearly if (an ) is unbounded we can not hope for the convergence of the series.
Now if (an ) is bounded, then there is a convergent subsequence with some limit l and for the
convergence of series, besides many other terms, we tend to add infinitely many approximations
of l also. So we feel that this l should be 0, otherwise sequence of partial sums will be unbounded
or oscillatory like in Example (5.5).
Theorem 5.9. If the series
P
n=1 an converges, then the sequence of terms (an ) converges to
0.
Proof. Since the series
P
n=1 an converges, there is a number s such that the sequence of partial
sums (sn ) converges to s. Thus its subsequence (sn+1 ) also converges to s. And hence the
difference of these sequences an = Sn+1 Sn must converge to the corresponding difference of
limits, that is, s s = 0. 
Remark 5.10. Converse of this theorem is not true. For example n1 0, but 1
P
n=1 n is not
convergent (See Example (5.2)).
Remark 5.11. Above theorem is more useful for proving non-convergence of the series.
Example 5.12. The series
P
n=1 cos n diverges because the sequence (cos n) does not converge
to 0. (See Example 2.59).
Example 5.13. For any x R such that |x| > 1, the series n
P
n=1 x does not converge because
xn 6 0.
P
Problem 5.14. Prove that (an an+1 ) converges if and only if the sequence an converges.
n=1
P
Solution. By definitionPn (an an+1 ) converges if and only if sequence of partial sums (Sn )
converges. But Sn = k=1 (ak ak+1 ) = a1 an+1 converges if and only if an converges.
Example 5.15. Series 4
P
n=1 (4n3)(4n+1) converges because we can write the general term of
 
1
the series as the difference of two consecutive terms of a convergent sequence (4n3) , that is,
4 1 1
(4n3)(4n+1) = (4n3) (4(n+1)3) .
P 2n+1 2n+1 1 1
Example 5.16. Series n=1 (n2 )(n+1)2 converges because (n2 )(n+1)2
= n2
(n+1)2
and ( n12 )
converges.
Remark 5.17. We have seen many criterion for the convergence of the sequences and we know
that the convergence of series is closely related to convergence of the sequences. So the question
arise that how these criterions can be useful in view of series convergence. For example to
guarantee the monotonicity of the sequence of partial sums we require that the signs of all the
terms of the series should be same.
42
P
Theorem 5.18. Let an 0 for all n N. Then n=1 an converges if and only if (Sn ) is
bounded above.
Proof. Clearly since an 0 for all n N, the sequence of partial sums (Sn ) is monotonic in-
creasing. Thus (Sn ) is bounded above (by monotonic criterion)Pif and only if (Sn ) is convergent,
by definition, it is equivalent to the convergence of the series
n=1 an . 
Remark 5.19. Above theorem is also useful for proving the convergence of certain series with
terms of different signs. In fact for checking the convergence of
P
n=1 n , weP
a will check the
convergence of a series with terms replaced by its absolute value, that is, with n=1 |an |. But
here we should not relate these series with the limits of their partial sums, instead we should use
some criterion of sequence convergence, which does not use the limit, may be Cauchy criterion
because |Sn Sm | = |am + am1 + . . . + an+1 | |am | + |am1 | + . . . + |an+1 |.
Theorem 5.20. (Absolute Convergence Test) If
P P
n=1 |an | converges, then n=1 an also
converges.
Proof. Since
P
n=1 |an | converges, the sequence of partial sums (Sn = |a1 | + |a2 | + . . . + |an |)
satisfies the Cauchy criterion. Hence for any given  > 0, there exists N N such that
n, m N implies |Sm Sn | <  or |am | + |am1 | + . . . + |an+1 | < . But then |Sn Sm | =
|am + am1 + . . . + an+1 | |am | + |am1 | + . . . + |an+1 | < P
 for all n, m N , which implies the
convergence of (Sn ). And hence the convergence of series n=1 an . 
P 1
Remark 5.21. Converse is not true because we know that n=1 n does not converge, but later
(1)n
we will show that
P
n=1 n converges.
5.22. Absolute Convergence of Series
We say that the series
P
n=1P an converges absolutely if the corresponding series with absolute

values of the terms, that is, n=1 |an | also converges. Clearly by above theorem if a series
absolutely converges then it converges. But by above example a convergent series might not be
absolutely convergent.
P
Theorem
P 5.23. If the series n=1 an is convergent, then for any constant c R, the series
n=1 can is also convergent.
P
Proof. Since the series a converges, the corresponding sequence of partial sums (Sn )
Pn n=1 n
converges, where Sn = k=1 ak . Further, since the sequence of partial sums (Sn ), corresponding
to the series
P Pn
n=1 can , is described as Sn = k=1 cak = cSn and hence is convergent. 
P P P
Theorem 5.24. If the series n=1 an and n=1 bn are convergent, then the series n=1 (an +
bn ) is also convergent.
Proof. The the general term of the sequence of partial sums of the series
P
Pn Pn Pn n=1 (an + bn ) is given
by Tn = k=1 (ak + bk ) = k=1 ak + k=1 bP k = Sn + S ,
n Pwhere Sn and Sn are general terms

of the sequence of partial terms of the series n=1 an and n=1 bn respectively. But both the
series are convergent, hence the sequences of partial sums (Sn ) and (Sn ) also converge. Thus
(Tn = Sn + Sn ) converges, hence the series
P
n=1 (an + bn ) also converges. 
Question 5.25. As in case of sequences, whether we can compare the convergence of two series,
if we have certain comparison between their terms.
Theorem 5.26. (Comparison test) If 0 an bn for all n k for some k. Then
(1) if P
P P
n=1 bn converges, thenP n=1 an converges,
(2) if n=1 an diverges, then

n=1 bn diverges.

Proof.P
WLG we can assume k = 1 because first few terms does not effect the convergence.
Since
n=1 bn converges, the sequence of partial sums (Sn ) is convergentP
hence bounded. But

0 an bn implies that if (Sn ) is the sequence of partial terms of n=1 an , then Sn =
43
P P
m=1 nam m=1 nbm = SnPand hence (Sn ) is also monotonic increasing and bounded. Thus
(Sn ) is convergent and series n=1 an converges.
P
Further if n=1 an diverges, (Sn ) is monotonic increasing and unbounded. But Sn Sn
implies (Sn ) is also unbounded hence
P
n=1 bn diverges. 
Example 5.27. Since
P 1 1 1 P 1
P 1 n=1 n diverges and 0 n np for 0 p < 1, then n=1 np diverges.
In particular n=1 n diverges.
P 1 1 1 P 1
Example 5.28. n=1 (n+1)2 converges because (n+1)2 n(n+1) . This also implies n=1 n2
1 1 P 1
converges. And since np n2 for p > 2, n=1 np converges.
1
n12 for n 4, 1
P
Example 5.29. Since n! n=1 n! converges.

Question 5.30. As in case of sequences, whether we have any sandwich kind of result. Whether
we can conclude the convergence of a series if it is sandwiched between two convergent series,
in the sense if we have sandwich kind of comparison between their terms.
P P
P 5.31. Let an bn cn for all n N. If the series n=1 an and n=1 cn converge
Theorem
then n=1 bn also converges.
Proof. Since the series
P P P
n=1 an and n=1 cn converge, the seriesP n=1 (cn an ) also converges.

Further, since 0 (bn an ) (cn aP n ), by comparison test n=1 (bn an ) also
Pconverges

and hence its addition with the series n=1 an must converge. Thus the series n=1 bn also
converges. 
P P an
Problem 5.32. If an 0 and p > 0, then n=1 an converges if and only if n=1 p+an
converges.
apn , hence whenever
an P P an
Solution. Since p+a n n=1 an converges, n=1 p+an also converges by
comparison test.
an
Now to do other way comparison we try to find some constant c > 0 such that can p+a n
.
Note that we can find such a c if we can find some upper bound for p + an or simply for an .
But such an upper bound can be guaranteed by the convergence of sequence (an ). Now let us
ask ourselves that if we want to show the convergence of (an ), what should P be the limit for
this sequence? Note that our ultimate aim is to show the convergencePof n=1 an , and it can
happen only when an 0. Now we will prove that an 0. Since an
n=1 p+an converges so
an an 1
that p+a n
0. But this implies p+a n
1 1 p+a n
p1 p + an p an 0. Thus
(an ) is bonded, that is, there exists M > 0 such that an M for all n N . But this implies
P
that p + an < p + M an < anp+a (p+M )
n
. Thus by comparison test n=1 an converges because
P an
n=1 p+an converges.

Problem 5.33. Suppose an , bn > 0 and an+1 bn+1


an bn for all n N for some N N. Show that
P P
if n1 bn converges then n1 an also converges.
Solution. Note that if we can prove that an c bn eventually, then the result will follow
through comparison test. But by given conditions abnn abn1
n1
< . . . < an+1 aN aN
an bN . Now bN can
be considered as a constant c for comparison test.
1 n2
nn2 (n+1)n1 en (n)! (1+ n )
Example 5.34. For the series an+1
P
n=1 en n! the ratio an = en+1 (n+1)! (n)n2 = e =
1 n
(1+ n )
1 2
e(1+ n )
< e(1+e 1 )2 = (n+1)
1 1
2 / (n)2 . Thus we can consider bn =
1
(n)2
and can apply above result
P n
because bn converges.
Remark 5.35. In above examples we could see the power of Comparison test. To apply it
we need to compare eventually all the terms of both series. But this comparison can also be
insured if the sequence, formed by ratio of the terms, converges.
44
Theorem 5.36. (Limit Comparison Test) Suppose an , bn > 0 for all n k for some k and
lim abnn L.
L > 0, then
P P
(1) If L R and P n=1 an converges iff n=1 bn converges.
P
(2) If L = 0 and P n=1 bn converges, then P n=1 bn converges.
(3) If L = and n=1 bn diverges, then

n=1 bn diverges.

Proof. 1) We can choose  > 0 such that L  > 0 and then there existsPN N such that
whenever n N we have L  < abnn < L +  and then by comparison test n=1 an converges
iff
P
n=1 nb converges.
2) It is given that L = 0 and an
P
n=1 bn converges. Since lim 0, for a given  > 0 there exist
Pbn
some N N such that bn <  so that by comparison test n=1 an converges because
an P
n=1 bn
converges.
3) We have lim abnn so by definition if we choose M = 1 there exists some N such that
whenever n N we have 1 abnn and now since
P P
n=1 bn diverges, then by comparison n=1 an
diverges. 
P 1 P 1
Problem 5.37. If an > 0 and p > 0, then n=1 an converges if and only if n=1 p+an
converges.
1 1 P 1 P 1
Solution. Since p+a n
an
, hence whenever n=1 an
converges, n=1 p+an also converges by
comparison test.P
Now suppose 1 1 1
n=1 an converges. This implies that an 0. But an 0 an + p p
1 1 1
p+an 1 p+an P 1
an p p an 1. Thus by Limit Comparison Test n=1 p+an converges.
P 1 1
Example 5.38.
P 1 n=1 ( n sin n ) converges by comparing the limits (LCT) with the series
1 1 1 xsin x
n=1 n3 because ( n sin n )/( n3 ) converges to a limit, which is equal to limx0 x3
, if at all
it exists. But by LHospital Rule limx0 x3 = limx0 3x2 = limx0 6x = limx0 cos6 x =
xsin x 1cos x sin x
1
6.

1 cos n1 if we consider
P P P 1
Example 5.39. To check the behaviour of bn = np and want
1
1cos
to apply Limit Comparison test, we need to find the limit limn 1/np n or it suffices to find
the limit limx0 1cos x
xp . Now if weP can find certain p such that limx0 1cos xp
x
is a non zero
1
real number, then the behaviour of np will decide the behaviour of the given sequence. Now
limx0 1cos
xp
x
= lim sin x
x0 pxp1 = lim cos x
x0 p(p1)xp2 . Thus to satisfy all the steps and to get the
1 cos n1 converges
P
last limit as non-zero limit we can consider p = 2. And thus by LCT
P 1
because n2
converges.
P 1 1
Example 5.40. n=1 n log(1 + n ) converges by comparing the limits (LCT) with the se-
P 1
ries n=1 n2 because n log(1 + n1 ) converges to a limit, which is equal to limx0 log(1+x) x =
1
limx0 1+x = 1.
5.41. Cauchy Condensation Test
This test is used to examine the convergence of a series whose terms are positive and decreas-
ing. Since we have ordering between the terms and if we break the series in to infinitely many
parts, each part is bounded above and bounded below by first and last term of the part respec-
tively. And we can choose a right representative term from each of the part and multiply this
representative term with number of terms in that part to give a quantity, which is comparable
to sum of the terms of the part. This way we can get a new sequence of these representative
quantities of each part. And we can find the corresponding series, which is rather dense but
comparable to the original one. Note that in this decomposition if each part have same number
of terms, then it is not effective in its full strength. But if number of terms in each next part
are increased in a certain percentage of the number of terms in the previous part, then the time
taken by computer to justify the convergence is effectively less.
45
Theorem
P P k Test) If an 0 and an+1 an for all n N, then
5.42. (Cauchy Condensation
a
n=1 n converges if and only if k=1 2 a2k converges.

Proof. Terms of both the series are positive, so to examine the P convergence we just need to
examine the boundedness of sequence of partial sums. Let (Sn = nm=1 am ) be the sequence of
partial sums for the series
P Pk m
P k n=1 an and (Tk = m=1 2 a2m ) be the sequence of partial sums
for the series n=1 2 a2P k.

First we assume that n=1 an is convergent, which means there is some number M such that
Sn M for all n N. Now
Tk = 2a2 + 4a4 + 8a8 + . . . + 2k a2k = 2[a2 + 2a4 + 4a8 + . . . + 2k1 a2k ]
2[a2 + (a3 + a4 ) + (a5 + a6 + a7 + a8 ) + . . . + (a2k1 +1 + a2k1 +2 + . . . + a2k )]
= 2[S2k a1 ]
2S2k
2M.
Thus sequence of partial sums (Tk ),P which is already monotonic, is bounded also, therefore it
is convergent. And hence the series k
k=1 2 aP 2k converges.
Now suppose we are given that the series k
k=1 2 a2k is convergent, therefore the sequence
of partial sums (Tk ) is convergent, but (Tk ) is monotonic increasing hence it is bounded above.
If M1 is the upper bound for (Tk ), then for any given n N, there exists an unique k such that
2k n < 2k+1 so that
n
X 2k+1
X1
Sn = am am
m=1 m=1
a1 + (a2 + a3 ) + (a4 + a5 + a6 + a7 ) + . . . + (a2k + a2k +1 + . . . + a2k+1 1 )
a1 + 2a2 + 4a4 + . . . + 2k a2k
k
X
= a1 + 2m a2m = a1 + Tk
m=1
a1 + M1 .
Thus sequence (Sn ) isPconvergent because it is monotonically increasing and bounded above.
And hence the series n=1 an converges. 
P 1
Corollary 5.43. n=1 np converges if p > 1 and diverges if p 1.

Proof. By Cauchy condensation Test converges if and only if k


P P
P 1 n=1 an P k=1 2 a2k converges.
k 1 P k(1p)
Therefore n=1 np converges if and only if k=1 2 (2k )p = k=1 2 converges. But
P k(1p)
k=1 2 is a geometric series, which converges if and only if ratio of the terms is strictly
less than 1, that is, 2(1p) < 1 1 p < 0 1 < p. Thus if 1 6< p, then it diverges. 
P 1
Corollary 5.44. n=1 n(log n)p converges if p > 1 and diverges if p 1.

Proof. By Cauchy condensation Test


P 1 P k 1
n=1 n(log n)p converges if and only if k=1 2 2k (log 2k )p =
P 1 P 1 P 1
k=1 kp (log 2)p converges. But kp (log 2)p converges if and only if k=1 kp converges. And
P k=1 1
now by previous corollary k=1 kp converges if and only if 1 < p. Thus if 1 p, then it
diverges. 
Observation 5.45. We know that if a series converges,P then the sequence of its terms converges
to 0. Now suppose the terms of aP convergent series an are decreasing to 0. Keeping in mind
1
that the terms of divergent series n are also decreasing to 0. We feel that sequence an should
be eventually dominated by sequence ( n1 ).
Theorem 5.46. Let an 0 and an an+1 for all n N. If
P
n=1 an converges, then nan 0.
46
Proof. By Cauchy condensation test
P P k
n=1 an converges if and only if k=1 2 a2k converges.
So 2k a2k 0 as k . But for each n N, there exists a unique k such that 2k n < 2k+1
and hence 0 < nan na2k < 2k+1 a2k , so that 0 < limn nan limk 2k+1 a2k = 0. 
P n
Observation 5.47. We know that P if a geometric series x converges, then it converges
absolutely. Now if there is a series an whose terms eventually satisfy |x|n an |x|n , then
|an | |x|n and by comparison test it converges absolutely.
P
Theorem 5.48. (Ratio Test) Let an be a series of eventually non-zero terms.
(1) If | an | < 1 for all n N for some N N, then
an+1 P
a converges.
an+1 P n=1 n
(2) If | an | 1 for all n N for some N N, then n=1 an diverges.
P
Proof. We will try to deduce the convergence of the series P an by showing its absolute con-
vergence. Clearly |an | |an 1| . . . nN |aN | but n=N P nN |aN | is a geometric series

and it converges because < 1. And hence by comparison test P n=N |an | converges.
Similarly for all n N , |an | |aN | =
6 0 and hence an 6 0 and an diverges. 

Observation 5.49. This eventual comparison can be ensured by the limits of fractions |a|an +1|
n|
,
if at all it exists.
Corollary 5.50. Let an 6= 0 for all n N, and |a|an +1|
P
n | L. If L < 1, then an converges, and
P
if L > 1, then an diverges. But we cant make any conclusion if L = 1.
Proof. If L < 1, then choose  = |1 L|/2 so that L +  < 1 and we can find N N such that
n N |a|an +1|
n|
(L , L + ). Thus |a|an +1|
n|
< (L + ) < 1 and we can apply Ratio Test.
Similarly one can prove that if L > 1 and  = |L 1|/2, then |a|an +1|
n|
> (L ) > 1 for all
n N for some N N. And so an 6 0 showing theP divergence. But L = 1 is possible both for
1
converging as well as diverging series. For example np for p R. 
P 1 |an +1| 1
Example 5.51. n! converges because |an | = n+1 0, which is less than 1.
P nn |an +1|
Example 5.52. n! diverges because |an | = (1 + n1 )n e > 1.
P n log n |an +1| (n+1) log(n+1) 2n
Example 5.53. 2n converges because limn |an | = limn 2n+1 n log n =
1 log(x+1) log(x+1)
2 limx 1+x
x log x = 1 1+x
2 limx x limx log x = 1 1 1
2 limx 1 limx (x+1) / x
1
=
1
2 limx (x+1) = 12 <
x
1.
Following Theorem is again due to comparison with geometric series. Here we want to
compare the root of the terms with the ratio of a geometric series.
n 1/n
Theorem 5.54. (Root Test) Suppose P 0 an x 0 an x for all n N for some
N N and for some 0 < x < 1. Then an converges.
Proof. Proof is just the application of Comparison test. 
Now the following corollary is easy to prove.
5.55. If |an |1/n L, then
P P
Corollary P |an | converges if L < 1 and an diverges if L > 1.
And series an can both converge as well as diverge if L = 1.
P 1 1/n = 1 0, which is less than 1.
Example 5.56. n=2 (log n)n converges because |an | log n
P n n 1
Example 5.57. n=1 ( n+1 ) converges because |an |
1/n =
(1+ 1 )n
1e < 1.
n

Example 5.58. (1 + n1 )n(n+1) diverges because |an |1/n = (1 + n1 )(n+1) = (1 + n1 )(n) (1 + n1 )


P
e.1 = e > 1.
47
P n(1)n n 1
Example 5.59. n=1 2 converges because limn |2n(1) | n =
(1)n (1)n
21 limn 2 n = 1
2 < 1 as 2x is a continuous function and the sequence n 0.
P 1
Example 5.60. n=1 np can converge and diverse as well but ( n1p )1/n 1.
Theorem P5.61. (Lebniz Test) If (an ) is a decreasing sequence of real numbers converging to
0, then n=1 (1) n+1 an converges.
P
Proof. Let (Sn ) be the sequence of partial sums of an . To conclude the result we will show
that (Sn ) is convergent. Though the sequence (Sn ) is not monotonic, but if we observe carefully
we find that there is a monotonicity between odd and even terms separately. So if it is possible
to show the convergence of subsequences (S2n ) and (S2n+1 ) (see example (2.63)), we may hope
for the convergence of (Sn ) itself. S2n = a1 a2 + a3 a4 + a5 . . . + a2n1 a2n = (a1
a2 ) + (a3 a4 ) + . . . + (a2n1 a2n ), this shows that S2(n+1) S2n = a2n+1 a2n+ 0 and
(S2n ) is increasing sequence. But S2n = a1 (a2 a3 ) (a4 a5 ) . . . (a2n2 a2n1 ) a2n ,
so that S2n is bounded above by a1 , hence it converges to certain limit say l, that is, S2n l.
Now S2n+1 = S2n + a2n+1 l + 0 = l. Now for given  > 0, there exists Ne and No such that
n Ne |S2n l| <  and n No |S2n+1 l| < . Thus if n N = max{2Ne , 2No + 1},
then |Sn l| < . Hence (Sn ) is convergent. 
n+1
(1)n n1 , (1)
P P
Example 5.62. n=2 log n are convergent series.
Problem 5.63. Let (an ) be a decreasing
P sequence of real numbers converging to 0 and bn =
a1 +a2 +...+an
n for n N. Show that n=1 (1)n bn converges.
Solution. Since (an ) is decreasing and an 0, so we hope that the sequence of averages (bn )
should also be decreasing and go to zero. First we note that bn+1 bn = a1 +a2 +...+a n+1
n +an+1

a1 +a2 +...+an an+1 a1 +a2 +...+an 1
n = n+1 n(n+1) = n(n+1) [(a1 an+1 ) + (a2 a n+1 ) + . . . + (a n a n+1 )] 0.
Thus (bn ) is a decreasing sequence. Now since an 0, after finite number of terms an s are 0

desirably close to 0, that is for given 1 > 0, there exits N1 N such that |an | < 1 for all
a +a(N1 +2) +...+an
n N1 , so that average of these terms, that is, (N1 +1) nN 1
< 1 . Now observe that
for each N N and n N we have
a1 + a2 + . . . + aN aN +1 + aN +2 + . . . + an
bn = + .
n n
Thus we can control the later expression by 1 by choosing N = N1 . But to control first
expression we observe that a1 + a2 + . . . + aN1 is a fixed number so for any 2 > 0, there exists
a +a +...+a
N2 N such that 1 2 n N +1 < 2 for all n N2 . Now we choose N = max{N1 , N2 } so that
bn < 1 + 2 for all n N . Now since 1 and 2 was our choice in the beginning, for a given
 > 0 we could have chosen our 1 = 2 = 2 . So that bn <  for all n N . Thus bn 0.
5.64. Power Series
P n
We have seen that the geometric series x is constructed using powers of some x R and
converges if |x| < 1 and diverges P if |x| 1. In general for some sequence (an ) of real numbers
and x R one can define a series n
n=1 an x and can ask the question that for what value(s) of
x the
P series converges. More generally for some fixed c R and some x R the series defined
by a
n=1 n (x c) n is known as power series around the point c.

Example 5.65. Let 1 n


P
n=1 n! x be a power series around 0. Let us apply Ratio test to check
1 1 n 1
the convergence of the series at some x R. | (n+1)! xn+1 / n! x | = x n+1 0 < 1. So the series
converges absolutely for all x R.
Example 5.66. Let n
P
n=1 (x c) be a power series around some point c R. If we again
n+1
apply Ratio test, we see that | (xc)
(xc)n | = |(x c)|. Thus whenever x is such that |x c| < 1,
we can find some = |x c| + 1|xc|
2 , depending on x, such that |x c| < < 1. Thus by
ratio test the series converges absolutely if x satisfies |x c| < 1 and diverges if |x c| 1.
48
P n n
Example 5.67. Series n x does not converge for any x 6= 0 because by Ratio test
(n+1)n+1 xn+1
| nn xn | = |x| (n + 1) (1 + n1 )n > 1 eventually for all n if x 6= 0. So the series di-
verges for x 6= 0. And it converges for x = 0 because there is only one non zero term in the
series.
P n
Question
P 5.68. Suppose a power series n=1 an x converges at some point x0 R, that
n
P n=1 ann x0 converges.n For any other point x R
is the general term of the power series
n . But to compare two series, terms of
n=1 a n x , that is, a n x looks comparable with a n x 0
both should be positive! Okey, suppose if an > 0 and x0 > 0, and we only consider x > 0, then
an xn an xn0 x < x0 and the series can converge for all 0 < x < x0 . Following theorem
shows that even if the terms are not positive always, we can prove the absolute convergence of
the series for all x R such that |x| < |x0 |.
Theorem 5.69. If the series n
P
n=1 an (x c) converges for some x0 and diverges for some x1 ,
then it converges absolutely for all x R such that |x c| < |x0 c| and diverges for all x R
such that |x c| > |x1 c|.
Proof. (*) Suppose x0 6= c and n
P
n=1 an (x0 c) converges. Let x R such that |xc| < |x0 c|.
Since the series converges at x0 , the sequence an (x0 c)n 0. So the sequence is bounded and
there is some M > 0 such that |an (x0 c)n | < M for all n N. Therefore, for all n N
n (x c) n

n n |(x c) |
|an (x c) | = |an (x0 c) | M .
|(x0 c)n | (x0 c)

(xc)
P (xc) n
Now since (x 0 c)
< 1, the series M (x0 c) converges and hence by comparison test
n an (x c)n converges absolutely.
P P
M |an (x c) | also converges. And hence
Next suppose x R and |x c| > |x1 c|, now if series converges at x, then by above result
it will converge absolutely for x1 also giving a contradiction to assumption. So the series has to
diverge at all x such that |x c| > |x1 c|. 
Remark 5.70. Above theorem suggests that either a power series can converge at only at c, or
can converge for all x R or there is a unique r > 0 such that the series converges absolutely
for all |x c| < r and diverges
Pfor all |x c|n> r. This r is called the radius of convergence.
if we define S = {x R : n=1 an (x0 c) is convergent.}, then the possibilities for S are
{c}, R, (c r, c + r), [c r, c + r), (c r, c + r], [c r, c + r] for some r > 0.
|an+1 (xc)n+1 |
Example 5.71. By Ratio test the series 1 n
P
n=1 n (x c) is convergent when an (xc)n =
|n(xc)n+1 | n P 1
(n+1)(xc)n = n+1 |x| |x| < 1. And divergent when |x| > 1. Further we know that n
P (1)n
diverges and n converges. Thus the set of all points where the series converges is S =
[1, 1).
Example 5.72. Consider the series 1 n
P
n=1 nn x , for convergence at some x R let us use root
test. So we try to find the limit of |an xn |1/n = | n1n xn |1/n = | nx | 0 < 1. Thus the series
converges for every x in R. Thus S = R and radius of convergence is infinity.
P (3)n n
Example 5.73. Consider the series n=1 n+2 (x + 1) . Let us apply ratio test so that
|an+1 (x+1)n+1 | n+1
= |(3)
|an (x+1)n | n+3
| n+2
|(3)n | |(x + 1)| 3|x + 1|. Thus by Ratio test the series converges
if |x + 1| < 13
and diverges if |x + 1| > 31 . At x + 1 = 31 , the series converges because of Lebnizs
test and x + 1 = 13 diverges because the sequence is comparable to
P1
n . Thus radius of
1 4 2
convergence at x = 1 is 3 and the set S = ( 3 , 3 ].
Remark 5.74. Thus to find the convergence of the series we usually apply ratio test or root
test, which ever is more suitable. But it is difficult to find the sum of the series.
5.75. Taylor Series
49
We recall Taylors theorem for (n + 1) times differentiable functions
f 00 (c) f (n) (c) f (n+1) (c1 )
f (x) = f (c) + f 0 (c)(x c) + (x c)2 + . . . + (x c)n + (x c)n+1
2! (n)! (n + 1)!
where c1 is some point between c and x. Now suppose f is infinitely many times differentiable,
f (n) (c)
for example polynomial, ex , sin x etc, then the series of the form n
P
n=0 (n)! (x c) is known
f (n) (0)
as Taylor series of the function around the point c. And if c = 0 then the series n
P
n=0 (n)! (x)
is known also as Macluarin series.
We recall that Taylor polynomial Pn (x) is the approximation of the function with error as
(n+1)
Rn (x) = f (n+1)!
(c1 )
(x c)n+1 . In general we might hope that the remainder term should go to
zero as n goes to infinity and hence Taylor series should converge to the function. But there
are functions whose Taylor series does not converge and even if converges it might not converge
to the function. We say that Taylor series converges to the function at x if the remainder term
goes to zero. An infinitely differentiable function f is said to be analytic if for each x0 R
there exists a nbd of x0 such that the Taylor series around x0 converges to the function for all
the points in the nbd.
1
Example 5.76. Define a function f : R{1} R such that f (x) = (1x) , then Macluarin series
P n (n+1) (c )
is n=0 x , which converges in (1, 1). Now the remainder term Rn (x) = f (n+1)! 1
(x c)n+1 =
1 |x| 1
(1c1 )n+2
xn+1 goes to zero if |1c1 | < 1 which is possible when 1 < x < 2 because c is in
between 0 and x. Thus we can say that the Taylor series converges to the function in (1, 21 ).
2
Example 5.77. Consider a function f (x) = e1/x for x 6= 0 and f (0) = 0. Using LHospital
rule one can find that all the derivatives of f at 0 as f k (0) = 0 for all k, so that Macluarin series
is around 0 is 0. Thus it can cot converge to the function for any x 6= 0.
P 1 n
Example 5.78. The Taylor series of the function ex is n! x . Forcx > 0 the remainder term
ec e ex
is Rn (x) = (n+1)! xn+1 . This converges to zero because Rn (x) = (n+1)! xn+1 < (n+1)! xn+1 0
because an+1
an = x/(n + 2) 0 < 1.
(1)n 2n+1
Example 5.79. Macluarin series of f (x) = sin x can be written as
P
n=0 (2n+1)! x . And
(1)n+1 2n+3 n+1 1 2n+3 .
the remainder term as Rn (x) = (2n+3)! x f (c) so that |Rn (x)| < (2n+3)! x And it
an+1 x2n+5 (2n+3)! x2
goes to zero as =
an (2n+5)! x2n+3 = (2n+4)(2n+5) 0 < 1. Thus the Taylor series for sin x
converges to the function itself for all x R.

6. Riemann Integration
As it is explained in the lectures that we try to find area under a curve by approximating it
with the sum of areas of rectangles of finer and finer width from above and under the curve.
Now onward for better understanding of the defined quantities and the concepts, students
can draw the figures for themselves.
To find the area between the graph of bounded function f : [a, b] raR and the X-axis, we first
divide [a, b] in to smaller subintervals. Or in other words we consider a partition of the interval
[a, b]. In general by a Partition P of the interval [a, b], we mean a finite number of points say
a = x0 < x1 < . . . < xn = b in [a, b], and we write P = {a = x0 , x1 , . . . , xn = b}. We represent
the length of ith sub interval [xi1 , xi ], by i = xi xi1 for all 1 i < n. And since we
already assumed that the function is bounded, we can very well assume
Mi = sup{f (x) : xi1 x xi } and mi = inf{f (x) : xi1 x xi }.
And we define upper and lower Riemann sums for the partition P as
n
X Xn
U (P, f ) = Mi i and L(P, f ) = mi i respectively.
i=1 i=1
50
If m and M are lower and upper bound for the function on [a, b], that is, m f (x) M , for
all x [a, b]. Then for each 1 i < n, we have

m m i Mi M
mi mi i Mi i M i
n
X n
X n
X Xn
mi m i i Mi i M i .
i=1 i=1 i=1 i=1
Pn
Note that i=1 i = xn x1 = b a. Thus for any partition P ,

(6.5) m(b a) L(P, f ) U (P, f ) M (b a).

Now we define lower Riemann integral of the function f on the interval [a, b] as the supremum
of lower Riemann sums over all possible partitions of [a, b] and upper Riemann integral as the
infimum of U (P, f )s over all possible partitions P of [a, b]. We denote them as
Z b Z b
f dx = inf U (P, f ) and f dx = sup L(P, f ).
a P a P

We say that the function f is Riemann integrable or integrable on [a, b], if the upper and lower
Rb Rb Rb Rb
Riemann integrals are equal. In this case we denote a f dx = a f dx = a f dx = a f (x)dx and
Rb
call a f dx as the Riemann integral of the function f on [a, b].

Example 6.1. Consider a function f : [1, 3] R defined by f (x) = 1, if x 6= 2 and f (2) = 2.


We will show that f is Riemann integrable by showing that both upper and lower Riemann
integral is equal to 2.
Note that 1 f (x) 2 for all x [1, 3] and if we take any partition P = {1 = x0P , x1 , . . . , xn =
n
3},
Pn then for each subinterval [x ,
i1 ix ] infimum of f is m i = 1 and hence L(P, f ) = i=1 mi i =
i=1 1.(xi x i1 ) = x n x 0 = 31 = 2. Now since for each partition P the lower Riemann sum
Rb
is 2, lower Riemann integral a f dx = supP L(P, f ) = supP 2 = 2 and 2 = L(P, f ) U (P, f ).
This implies that 2 inf P U (P, f ). Thus to show integrability of the function we will try to
show that 2 = inf P U (P, f ). And for this we only need to show that for a given  > 0, there is
a partition P of [1, 3] such that U (P, f ) 2 . Note that on each subinterval of the partition,
except the subinterval which contains 2, supremum and infimum are same (mi = Mi = 1). So
we consider a partition P = {1 = x0 , x1 = 2 , x2 = 2 + , x3 = 3} for some 0 < < 1/2 so
that L(P, f ) = 1(2 1) + 1(2 + 2 + ) + 1(3 2 ) = 2 and

U (P, f ) = 1(2 1) + 2(2 + 2 + ) + 1(3 2 ) = 2(1 ) + 4 = 2 + 2.

Thus if we consider our to be less than /2, then U (P, f ) L(P, f ) < 2 < .

Question 6.2. Whether every bounded function on a closed interval is Riemann integrable.

Remark 6.3. Since there is no condition of continuity, we can think of constructing a bounded
function such that on each subinterval the difference of supremum and infimum of the function
is always greater than some positive constant and this can be done by defining a function
separately by rational and irrational points by two different continuous functions, which never
coincide.

Problem 6.4. Show that the function defined by



1, if a x b and x is rational
f (x) =
2, if a x b and x is irrational
is not Riemann integrable.
51
Solution. If we take any partition P = {x0 = a, x1 , . . . , xn = b}, then for any subinterval
[xi1 , xi ] we have mi = inf{f (x) : xi1 z xi } = 1 and Mi = sup{f (x) : xi1 z xi } = 2.
And hence
n
X X n
L(P, f ) = mi (xi xi1 ) = 1(xi xi1 ) = b a, and
i=1 i=1
n
X n
X
U (P, f ) = Mi (xi xi1 ) = 2(xi xi1 ) = 2(b a).
i=1 i=1
Rb Rb
Thus = inf P {U (P, f ) : P is a partition of [a, b]} = inf P {2(ba)} = 2(ba) and a f dx =
a f dx
Rb Rb
supP {L(P, f )} = inf P {(b a)} = (b a) so that a f dx a f dx = (b a) > 0. And hence f is
not Riemann integrable.
Observation 6.5. In general to check integrability of the function firstly we need to find upper
and lower Riemann integral. Recall that the lower Riemann integral is defined as the infimum
of L(P, f )0 s for all possible partitions and it is not easy to find such an infimum directly. But
as we observed in figures that if we increase a point in a given partition, then the corresponding
lower Riemann sum can not decrease in fact it increases in most of the cases. Precisely if
P = {x0 = a, x1 , . . . , xk1 , xk , . . . , xn = b} is a partition of [a, b], and we increase a point x in
the sub interval [xk1 , xk ] to form a new partition P = {x0 = a, x1 , . . . , xk1 , x , xk , . . . , xn =
b}. Further let mk = inf{f (x) : xk1 x xk }, mk = inf{f (x) : xk1 x x } and
m
k = inf{f (x) : x x xk }, then clearly mk mk and mk mk , so that

mk (xk xk1 ) = mk (xk x ) + mk (x xk1 ) mk (xk x ) + m


k (x xk1 ).

This implies that


n
X
L(P, f ) = mi (xi xi1 )
i=1
Xn
= mi (xi xi1 ) + mk (xk xk1 )
i=1,i6=k
Xn
mi (xi xi1 ) + mk (xk x ) + m
k (x xk1 )
i=1,i6=k
(6.6) L(P, f ) L(P , f ).
Similarly one can see that U (P, f ) U (P , f ). Now if P and Q are two partitions such that
P Q, then Q contains finite number of extra node points than P and we call Q as the
refinement of P , and one can use above idea to show that U (P, f ) U (Q, f ). Thus every
time when we are refining the partition (increasing the node points in the partition), upper
Riemann sum is more closer to upper Riemann integral and lower Riemann sum is more closer
to lower Riemann integral. Moreover, for any two given partitions P1 and P2 , if P is a common
refinement of both, that is, P1 P2 P , then
L(P1 , f ) L(P, f ) U (P, f ) U (P2 , f ).
This implies that for any two partitions P1 and P2 we have
L(P1 , f ) U (P2 , f ).
Now for any given partition P2 we can vary P1 over all possible partitions in above inequality
and get
Z b
sup L(P1 , f ) U (P2 , f ) f dx U (P2 , f ).
P1 a
52
And now we can vary P2 over all possible partitions to get
Z b Z b Z b
f dx inf U (P2 , f ) f dx f dx.
a P2 a a
Further, since for each partition P upper Riemann sum U (P, f ) is bounded by M (b a),
Rb
and a f dx = supP {U (P, f ) : P is a partition of [a, b]}, this supremum exists and we can find
R 
b
partition P such that U (P, f ) is arbitrarily close to upper Riemann integral a f dx . (Similarly
for lower Riemann integral we can find partition P such that L(P, f ) is as close to lower Riemann
integral as we want.) And in fact we can get a sequence of partitions say (Pk ) such that
Rb
limk U (Pk , f ) = a f dx. But there might be many other sequences of partitions such that
the corresponding sequence of upper Riemann sums converges to upper Riemann integral.
Question 6.6. Now the question arise that whether we can find certain property of sequence
of partitions so that if a sequence satisfies such a property, then the corresponding sequence of
U (P, f )0 s converges to upper Riemann integral.
6.7. Norm of a partition
To answer above question we first define norm of partition as
||P || = max (xi xi1 ) = max i .
1<n 1<n
One can prove the following result.
Theorem 6.8. (Darbouxs Theorem ) For any given
 > 0 there exists a > 0 such that if
Rb
a partition P satisfies kP k < , then U (P, f ) a f dx < .

We will not prove this result. Above result can be used to show that
Z b
(6.7) lim kPn k = 0 lim U (Pn , f ) f dx.
n n a
Similar result is also true for Lower Riemann integral.
Z b
(6.8) lim kPn k = 0 lim L(Pn , f ) f dx.
n n a
We will not discuss the proof of above results but we will use them for our purpose later.
Similar result is also true for lower Riemann integral.
In general it is difficult to show the Riemann integrability of a function. So it is desirable to
have an equivalent criterion for Riemann integrability of a bounded function.
6.9. Criterion for Riemann integrability
Theorem 6.10. Let f be a bounded function on [a, b]. Then f is Riemann integrable on [a, b]
if and only if for given  > 0, there exists a partition P such that
U (P, f ) L(P, f ) < .
Rb Rb
Proof. Let us first consider that the function f is Riemann integrable, that is, a f dx = a f dx.
Rb Rb
Since a f dx = supP {L(P, f )0 s}, there is a partition say P1 such that a f dx L(P1 , f ) < /2,
Rb Rb
further since a f dx = inf P {U (P, f )0 s}, there is a partition say P2 such that U (P2 , f ) a f dx <
/2. Let P = P1 P2 be the common refinement of both the partitions. Then
Z b Z b
f dx L(P, f ) < f dx L(P1 , f ) < /2, and
a a
Z b Z b
U (P, f ) f dx < U (P2 , f ) f dx < /2.
a a
53
Rb Rb
And thus we have a f dx L(P, f ) + U (P, f ) a f dx < /2 + /2, or U (P, f ) L(P, f ) < .
Next suppose if for every given  > 0 there is a partition P such that U (P, f ) L(P, f ) < ,
Rb Rb Rb Rb
then a f dx a f dx U (P, f ) L(P, f ) <  for any given  > 0. Hence a f dx = a f dx 

Corollary 6.11. Let f be a bounded function on [a, b]. Then f is Riemann inte-
grable on [a, b] if and only if there exists a sequence of partitions say (Pk ) such that
limk [U (Pk , f ) L(Pk , f )] = 0.

Proof. Suppose f is Riemann integrable, we can use the above criterion for each  = k1 , k
N to ensure the existence of partition Pk such that U (Pk , f ) L(Pk , f ) < k1 . And hence
limk [U (Pk , f ) L(Pk , f )] = 0.
Further if there exists a sequence of partitions say (Pk ) such that
limk [U (Pk , f ) L(Pk , f )] = 0, then for any given  > 0 there is a k0 N such that
U (Pk , f ) L(Pk , f ) <  for all k k0 . Hence by the above criterion of integrability the function
f is Riemann integrable. 

Problem 6.12. Let f : [0, 1] R be defined as f ( n1 ) = 1


n for all n N and 0 otherwise. Show
R1
that f is R-integrable. Find 0 f (x)dx.

Solution. Our function is constant except discontinuities at 1/n0 s. We already have dealt
with this kind of discontinuity in example (6.1), but there it was only at one point. The ideas
was to shrink the subinterval, which contains the point of discontinuity. So we can use this
idea even if there are finitely many points of discontinuity of this kind. But in our case there
are infinite number of points of this kind. But it is important to observe that if we take any
partition P = {0 = x0 , x1 , . . . , xn = 1} of [0, 1], then except for finite number of discontinuities
all the discontinuities are within the fist subinterval, i.e. [x0 , x1 ] and in this subinterval m1 = 0
and M1 x1 . Thus for given  > 0 we can control the contribution (M1 m1 )(x1 x0 ) to
U (P, f ) L(P, f ) from the subinterval [x0 , x1 ] by /2 > 0 if we consider x1 such that x21 < /2.
Thus if n0 is the smallest integer such that 1/n0 < /2 we can choose x1 as the middle point of
1/(n0 + 1) and 1/n0 .
Further we aim to find the other node points such that each 1/n for 1 < n n0 is contained
in a subinterval of length 2 as the middle point of the subinterval and 1 is contained in the
subinterval [1 , 1] so that the total contribution from these subintervals to U (P, f ) L(P, f ),
that is, (1 0) + (1/2 0)2 + . . . + (1/n0 0)2 is less than /2. And this can be ensured if
we assume 2n0 < /2. Note that the contribution from any other subinterval, which does not
contain 1/n0 s is zero because the difference of supremum and infimum value of the function on
these subintervals is zero. Now by choosing as above and enough small so that we can have
a partition
   
1 1 1 1 1 1 1 1
P = 0, + , , + , , . . . , , + , 1 , 1
2 n0 + 1 n0 n0 n0 n0 1 2 2

such that U (P, f ) L(P, f ) < . Hence f is Riemann integrable.

Now we will use Riemann criterion of integrability to show the integrability of a big class of
functions.

Theorem 6.13. If f is continuous function on [a, b], then f is Riemann integrable.

Proof. For any given partition we have U (P, f ) L(P, f ) = ni=1 (Mi mi )(xi xi1 ). Note
P
that if for a given  > 0 we can find a P such that Mi mi is dominated by /(b a) for all
54
1 i n, then
n
X
U (P, f ) L(P, f ) = (Mi mi )(xi xi1 )
i=1
n
X 
(xi xi1 )
(b a)
i=1
n
 X
(6.9) = (xi xi1 ) = .
(b a)
i=1
Since function is continuous, mi , Mi are attained at some points yi and zi respectively in the
closed interval [xi , xi1 ], that is, Mi mi = f (zi ) f (yi ). Further since f is continuous on
closed interval [a, b], by theorem (3.54) f is uniformly continuous and now for given  > 0 we
can find > 0 such that |f (x) f (y)| < /(b a) whenever |x y| < and x, y [a, b].
Thus we partition [a, b] such that each subinterval is of length less than or equal to so that
|zi yi | |xi xi1 | and hence |Mi mi | = |f (zi ) f (yi )| < /(b a) for all i. Now from
(6.9), U (P, f ) L(P, f ) <  and by Theorem (6.10) f is R-integrable. 
Theorem 6.14. If f : [a, b] R is monotonic and bounded, then f is integrable on [a, b].
Proof. Suppose WLG, f is monotonic increasing. We partition [a, b] in to n number of equal
length subintervals, that is, P = {x0 , x1 , . . . , xn } such that xi xi1 = (ba)
n . Thus
n
X
U (P, f ) L(P, f ) = (Mi mi )(xi xi1 )
i=1
n
X (b a)
(f (xi ) f (xi1 ))
n
i=1
n
(b a) X (b a)(f (b) f (a))
(6.10) = (f (xi ) f (xi1 )) = .
n n
i=1

Now for any given  > 0, we can find n0 large enough such that (ba)(fn(b)f 0
(a))
< . And
hence by partitioning [a, b] in to n0 number of equal length subintervals we can conclude from
(6.10) that U (P, f ) L(P, f ) < , which implies the integrability of f by Theorem (6.10). 
Theorem 6.15. If f and g are R-integrable on [a, b], then
Rb Rb
(1) for any real number 6= 0, f is also R-integrable and a f (x) dx = a f (x) dx,
Rb Rb Rb
(2) (f + g) is also R-integrable and a (f + g)(x) dx = a f (x) dx + a g(x) dx,
Rb Rb
(3) if f (x) g(x) for all x [a, b], then a f (x) dx a g(x) dx,
(4) for any point c (a, b), f is also integrable on intervals [a, c] and [c, b], moreover
Z b Z c Z b
f (x) dx = f (x) dx + f (x) dx.
a a c
We will not prove these results, but we will use them. As an immediate corollary we have
Z b Z b


f (t) dt |f (x)| dt.
a a
Observation 6.16. (**) We developed Riemann integration with the requirement of finding
the area of a region under the graph of a bounded positive function and above the X-axis. Now if
we can find such an area for some function, then we can hope that this area should continuously
increase as we increase the domain of the function continuously because intuitively we hope
that this area can not have a sudden jump, or in other words, if we fix one end point of a closed
interval and shrink the interval continuously by moving the other end point to the fixed end
point, then the area of the region decreases and approaches to zero in the limiting case. And
55
infact one can show that if the function is continuous, then the corresponding area function is
smooth.
In the following theorem we show that in some sense, differentiation and integration are
inverse operations of each other.
Theorem 6.17. (First fundamental R xtheorem of calculus) Let f be an integrable function
on [a, b]. For a x b, let F (x) = a f (t)dt. Then F is continuous on [a, b], further if f is
continuous at x0 , then F is differentiable at x0 and F 0 (x0 ) = f (x0 ).
Rx
Proof. (*) Let a x b. We define the area function as F (x) = a f (t)dt. To show the
continuity of F at some point x0 , we need to show that for given  > 0 there exists a > 0
such that |F (x) RF (x0 )| < Rfor all x whenever R x |x x0R| x< . WLG suppose x0 < x, then
x x
|F (x) F (x0 )| = | a f (t)dt a 0 f (t)dt| = | x0 f (t)dt| x0 |f (t)|dt. Now if M = sup{|f (x)| :
Rx Rx Rx
a x b}, then |F (x) F (x0 )| x0 |f (t)|dt x0 M dt = M x0 dt = M (x x0 ). And
we can assume = /M so that for all x such that |x x0 | < we have |F (x) F (x0 )| <
M |x x0 | < M = . This shows the continuity of F at x0 . In fact it also shows that F is
uniformly continuous.
Now suppose f is continuous at x0 and we need to show that F is differentiable
at x0 and the value of the derivative at x0 is f (x0 ), that is, F 0 (x0 ) = f (x0 ), or
equivalently, limxx0 F (x)F xx
(x0 )
= f (x0 ), or limxx0 F (x)F xx0
(x0 )
limxx0 f (x0 ) = 0, or
0
F (x)F (x0 )
limxx0 xx0 f (x0 ) = 0. But

Rx R x0
F (x) F (x0 ) a f (t)dt a f (t)dt
f (x0 ) = f (x0 )
x x0 x x0
Rx Z x
f (t)dt f (x0 )
= x0 dt
x x0 (x x0 ) x0
Z x
1
= (f (t) f (x0 )) dt.
x x 0 x0

1 Rx Rx
Thus F (x)F (x0 )
f (x ) = (f (t) f (x )) dt |xx1
|f (t) f (x0 )| dt. Now one

xx0 0 xx0 x0 0 0 | x0
can use continuity of f at x0 to show that the limit of the above expression approaches to zero
as x approaches to x0 . For a given  > 0 there is a > 0 such that |t x0 | < implies that
|f (t) f (x0 )| < , which in turn implies that
Z x Z x
F (x) F (x0 ) 1 1
f (x0 ) |f (t) f (x0 )| dt <  dt = .
x x0 |x x0 | x0 |x x0 | x0
Hence F 0 (x0 ) = f (x0 ). 
Following theorem shows that if our area function is smooth, then one can calculate integral
just by knowing the point evaluation of the area function at the end point of the interval.
Theorem 6.18. (Second fundamental theorem of calculus) Let f be an integrable func-
tion on [a, b]. If f has an antiderivative, that is, there is a function F on [a, b] such that
Rb
F 0 (t) = f (t) for all t [a, b], then a f (x) dx = F (b) F (a).
Proof. (*) We first observe that for any partition P = {x0 , x1 , . . . , xn } of [a, b], there is a set of
points {c1 , c2 , . . . , cn } such that
F (xi ) F (xi1 ) = F 0 (ci )(xi xi1 ) = f (ci )(xi xi1 ) and ci (xi1 , xi ).
But mi f (ci ) Mi , hence mi (xi1 , xi ) f (ci )(xi1 , xi ) Mi (xi1 , xi ), which implies
mi (xi1 , xi ) F (xi ) F (xi1 ) Mi (xi1 , xi ).
Pn
And thus by summing the above inequality from i = 1, . . . , n, we get L(P, f ) i=1 F (xi )
F (xi1 ) U (P, f ), or
L(P, f ) F (b) F (a) U (P, f ).
56
Now we use the integrability of f so that for a given  > 0 we can find a partition P such that
U (P, f ) L(P, f ) < , which implies that
Z b Z b
f (x) dx (F (b) F (a)) f (x) dx L(P, f ) U (P, f ) L(P, f ) <  and
a a
Z b Z b
(F (b) F (a)) f (x) dx U (P, f ) f (x) dx U (P, f ) L(P, f ) < .
a a
Rb Rb
Thus |(F (b) F (a)) a f (x) dx| <  for any given  > 0. Hence F (b) F (a) = a f (x) dx. 
If we make the strong assumption in the statement of the above theorem that f is continuous,
then it can be proved as an application of first fundamental theorem.
One can prove integration by parts formula as an application of the above theorem.
Theorem 6.19. (Integration by parts) Suppose F and G are differentiable functions on
[a, b], F 0 = f and G0 = g are integrable functions on [a, b], then
Z b Z b
(6.11) F (x)g(x) dx = F (b)G(b) F (a)G(a) f (x)G(x) dx.
a a

Proof. Let h(x) = F (x)g(x) + f (x)G(x) and H(x) = F (x)G(x). Thus H 0 (x) = h(x) so that
Rb
applying the second fundamental theorem of calculus we get a h(x)dx = H(b) H(a), which
Rb
implies that a [F (x)g(x) + f (x)G(x)] dx = F (b)G(b) F (a)G(a). 
( n1
ny
Problem 6.20. Suppose that gn (y) = 1+y , if 0 y < 1 . Then prove that
0, if y = 1
Z 1 Z 1
1
lim gn (y)dy = where as lim gn (y)dy = 0.
n 0 2 0 n
n1 n1 R 1 n1
Solution. If y (0, 1), then ny1+y 0 (Check it), so that limn ny1+y = 0 and 0 ny1+y dy =
0. Further to check that gn (y) is integrable on [0, 1] we can have a partition with last subinterval

as [1 2n , 1] so that the contribution to U (P, gn ) L(P, gn ) from this interval is less than

(n 0)(1 1 + 2n ) = 2 .

Further since gn is continuous on [0, 1 2n ], it is integrable there and hence there is a
partition P of [0, 1 2n 
] such that U (P , gn ) L(P , gn ) < /2. Using this one can define
a partition P of [0, 1] consisting of all the node points of P and the last node as 1. Then
U (P, gn ) L(P, gn ) U (P , gn ) L(P , gn ) + /2 <  and hence gn is integrable on [0, 1]. And
R 1 n1 yn 1 R 1 yn R 1 yn
one can use integration by parts to show that 0 ny1+y dy = 1+y |0 + 0 (1+y)2 dy = 12 + 0 (1+y)2 dy.
R 1 yn R1 n 1
Now since 0 0 (1+y)2 dy 0 y dy = n+1 , by applying the limit n in above inequality
R 1 yn R1 1
and using sandwich theorem we get limn 0 (1+y) 2 dy = 0 and limn 0 gn (y)dy = 2 +
R 1 yn
limn 0 (1+y)2 dy = 21 .
Remark 6.21. Above example shows that in general we can not interchange of limits and
integral.
Following results are useful for calculating integrals. We will not prove these results but we
will use them.
Theorem 6.22. (Direct Substitution) Let g : [c, d] R be a differentiable function and
its derivative function, that is, g 0 is R-integrable on [c, d]. Let f : g([c, d]) R be continuous.
Then
Z d Z g(d)
0
(6.12) f (g(x)) g (x) dx = f (t) dt
c g(c)
57
Theorem 6.23. (Inverse Substitution) Suppose g : [c, d] R is a differentiable function
and its derivative function g 0 is continuous with g 0 (x) 6= 0 for all x [c, d]. Let g[c, d] = [m, M ]
and f : [a, b] R be continuous. Then
Z M Z g1 (M )
(6.13) f (t) dt = f (g(x)) g 0 (x) dx.
m g 1 (m)

Problem 6.24. Let p R and f : R R be a continuous function. Show that if f is a


R x+p with period p that is, f (x + p) = f (x) for all x R, then g : R R defined
periodic function
as g(x) = x f (t) dt is a constant function on R.
Rx
Solution. WLG assume that p, x > 0 and consider the function F (x) = 0 f (t) dt. Since f
R x+p Rx
is continuous so F is differentiable (by First F.T.C.) and g(x) = 0 f (t) dt 0 f (t) dt =
F (x + p) F (x) and hence the derivative of g is g 0 (x) = F 0 (x + p) F 0 (x) = f (x + p) f (x) = 0
for all x. Thus g is a constant function.
R /2
Problem 6.25. Let f be a continuous function on [0, /2] and 0 f (t) dt = 0. Show that
there exists a pint c (0, /2) such that f (c) = 2 cos 2c.
Rx
Solution. First we assume the function F on [0, /2] defined as F (x) = 0 f (t) dt. Clearly
F (0) = 0 and F (/2) = 0 also by first fundamental theorem of calculus F is differentiable. We
need to show that f (c) = 2 cos 2c or equivalently the derivative of the function F (x) sin 2x is 0
at some point c in (0, /2). So we consider the new function G(x) = F (x) sin 2x. Clearly G is
differentiable and G(0) = 0 = G(/2). So by Rolles theorem there exists a point c(0, /2) such
that G0 (c) = 0 or F 0 (c) 2 cos 2c = 0. But by First F.T.C. F 0 (c) = f (c), hence f (c) = 2 cos 2c.
R x 4t3
Problem 6.26. Show that limx0 x14 0 2+3t 1
6 dt = 2 .

4t3
R x 4t3
Solution. Since 2+3t 6 is continuous function so F (x) = 0 2+3t6 dt is differentiable. Also
R x 4t3
1
limx0 F (x) = 0. So limx0 x4 0 2+3t6 dt is an indeterminate form 00 . We can apply Lhospital
R x 4t3 F (x) 0
rule to get limx0 x14 0 2+3t 6 dt = limx0 x4 = limx0 F4x(x)3 if the later limit exist. But by
4x3 F 0 (x) 4x3
First F.T.C. F 0 (x) = 2+3x6
and hence limx0 4x3
== limx0 1
2+3x6 4x3
= 12 .
6.27. Riemann Sum
Let P = {x0 , x1 , . . . , xn } be a partition of [a, b] and let ci [xi1 , xi ]. Then the Riemann
sum of the function
Pn f : [a, b] R corresponding to the partition P and the points c0i s is defined
as S(P, f ) = i=1 f (ci )(xi xi1 ). It is easy to see that for any partition P we have
L(P, f ) S(P, f ) U (P, f ).
Now if f is R-integrable, then by (6.8),(6.7) we conclude that if there is a sequence of
Rb
partitions say (Pn ) satisfying limn kPn k = 0, then both limn L(Pn , f ) = a f dx and
Rb
limn U (Pn , f ) = a f dx. But for each n N, L(Pn , f ) S(Pn , f ) U (Pn , f ), so by
Rb
sandwich theorem limn S(Pn , f ) = a f dx. Thus if f is integrable, then
Z b
(6.14) lim kPn k = 0 lim S(Pn , f ) f dx.
n n a
 
1 1 1
Problem 6.28. Let an = n2 n3 +13
+ n3 +23
+ ... + n3 +n3
. Convert an into a Riemann sum
and find limn an .
Solution. Note that to write an as the Riemann sum we try to consider the partition with n
1 1
subintervals of equal length
P i = n for all 1 i n and rewrite an as the multiple of n with
n
the expression of the form i=1 f (i/n). If this f turns out to be a continuous function, then an
can be considered
 as the Riemann sum for thePfunction f on [0, 1] with ci = xi . Let us rewrite
1 1 1 1 n i 1 1
an = n 1+(1/n)3 + 1+(2/n)3 + . . . + 1+1 = i=1 f ( n )i , where i = n and f (x) = 1+x3 .
58
Thus an = S(Pn , f ), where Pn = {0, n1 , n2 , . . . , nn = 1} is a partition of [0, 1]. Since the function
R1
f is continuous, hence R-integrable on [0, 1]. Now by (6.14) limn S(Pn , f ) = 0 f (t) dt. This
R1 1
implies that limn an = limn S(Pn , f ) = 0 1+t 3 dt =??

1 1
Problem 6.29. Let an = n+1 +. . .+ n+n . Convert an into a Riemann sum and find limn an .
  P
Solution. Let us rewrite an = 1
n
1
1+(1/n)
1
+ 1+(2/n) 1
+ . . . + 1+1 = ni=1 f ( ni )i , where i =
1
n and f (x) = x1 . Thus an = S(Pn , f ), where Pn = {1, 1 + n1 , 1 + n2 , . . . , 1 + nn = 1} is a partition
of [1, 2] and ci = xi . Since the function f is continuous, hence R-integrable on [1, 2]. Now
R2
by (6.14) limn S(Pn , f ) = 1 f (t) dt. This implies that limn an = limn S(Pn , f ) =
R2 1
1 t dt = log 2.
1
Problem 6.30. Let an = log (n!)
n
n for all n N . Convert an into a Riemann sum and find
limn an .
1
1
Solution. an = log (n!) = log ( nn!n ) n = n1 log ( nn!n ) = n1 ni=1 log ni . As above problems we can
n P
n
consider the partition Pn = {0, n1 , n2 , . . . , nn = 1} of [0, 1] and an = S(Pn , f ) with f (t) = log t,
R1 R1
ci = xi and i = n1 . Then limn an = limn S(Pn , f ) = 0 f (t) dt = 0 log t dt.
Problem 6.31. Suppose that f is a continuous function on [a, b] such that f (x) 0, for all
Rb
x [a, b]. Show that a f (x) dx = 0 if and only if f (x) = 0 x [a, b].
Solution. Clearly if f (x) = 0, then for any partition P both L(P, f ) and U (P, f ) are 0.
Rb Rb Rb
Hence a 0 dx = 0. Since 0 f (x) 0 = a 0dx a f (x) dx. And we need to show that
Rb
a f (x) dx = 0 implies that f (x) = 0. Suppose contradiction, that is, f (c) 6= 0 for some c [a, b].
WLG one can assume that c (a, b). But we already have that f (x) 0 so that f (c) > 0.
Then by continuity there exist a > 0 such that x (c , c + ) implies f (x) ( f (c) 3f (c)
2 , 2 ).
R c+ R c+ f (c) R c
Hence c f (x) dx c 2 dx = f (c) > 0. But 0 f (x) implies a f (x) dx 0 and
Rb Rb R c R c+ Rb
c+ f (x) dx 0 and hence a f dx = a f (x) dx+ c f (x) dx+ c+ f (x) dx 0+f (c)+0 >
Rb
0, which gives a contradiction to a f dx = 0.
Theorem 6.32. (Mean Value Theorem for Integrals) If f and g are continuous functions
on [a, b] and if g(x) 0 for a x b, then show that there exists c [a, b] such that
Rb Rb
a f (x)g(x)dx = f (c) a g(x)dx.
Proof. Since f is continuous on [a, b], it attains its minima and maxima. Let m = f (y) and
M = f (z), where y, z [a, b]. WLG assume that y < z. Further since
Z z Z z Z z
m f (x) M mg(x) f (x)g(x) M g(x) mg(x)dx f (x)g(x)dx M g(x)dx
y y y
Z z Z z Z z
m g(x) dx f (x)g(x) dx M g(x) dx.
y y y
Note that if g = 0 in [y, z], then the conclusion is
R zobvious. But if g is non-zero at some point
of the interval [y, z], then by previous problem y g(x) dx > 0. Then by intermediate Value
Rz
property for the continuous function h(x) = y g(t) dt f (x) on [y, z] and the observation that
Rz Rz Rz
h(y) = m y g(t) dt, h(z) = M y g(t) dt and h(y) y f (x)g(x) dx h(z), there exists a point
Rz Rz Rz
c in [y, z] such that h(c) = y f (x)g(x) dx, or y g(t) dt f (c) = y f (x)g(x) dx. 
6.33. Improper Integrals
We defined Riemann integration for a bounded function defined on a closed interval. But in
many cases it is desirable to find the integral on an unbounded interval.
6.34. Improper Integral of First Kind
59
R
Let f : [a, ) R be integrable on [a, x]R xfor all x > 0. Then we say that a f (x) dx is an
improper integral Rof first kind. If limx a f (t) dt exists and equal M R, we say that the

improper integral a f (x) dx is convergent and converges to M .
Rb
Similarly f (x) dx is an improper integral of second kind if f is integrable in [x, b] for all
Rb
x < b, and we say that it converges to M R if limx x f (t) dt = M .
R
Example 6.35. 1 t12 dt is an improper integral of second kind because f (t) = 1t2 is a contin-
Ruous
x 2
function on [1, x] for all x > 1, and hence
1 x 1
R xintegrable
2
there. Moreover, by Second F.T.C.
1
1 1t dt = t |1 = 1 xR , so that limx 1 1t dt = limx (1 x ) = 1. So the second
1
kind of improper integral 1 t2 dt converges to 1.
R Rx
Example 6.36. The integral 0 cos t dt diverges, because, by second F.T.C. 0 cos t dt =
sin x and limx ( sin x) does not exist.
Many times we also need to deal with the integration of an unbounded function on a closed
interval. To deal these cases we try to find out points in the interval, where the function
approaches to infinity. We further find a subintervals such that the function approaches to
infinity only at one point in the subinterval. One can subdivide this further into two subintervals
such that the function approaches to infinity only at one end point of these subintervals.
6.37. Improper Integral of Second Kind
Let f : [a, b] R be an unbounded function, which is bounded and R-integrable on [a+, b] for
Rb
all  > 0 and limxa f (x) = . Then a f (x)dx is called as improper integral of second kind.
Rb Rb
And if lim0 a+ f (x)dx exists, then we say that the improper integral a f (x)dx converges
to that limit. Rt
Similarly if the a f (x) dx exists for all a x < b and the function could be unbounded in
Rb
[a, b). In this case we also say that the integral a f (x) dx is an improper integral of second
Rt
kind. Moreover, if limtb a f (x) dx converges to some M R, then we say that the improper
Rb
integral of second kind a f (x) dx is convergent and converges to M .
If an improper integral does not converge, we say that it diverges.
R1 x
Example 6.38. 0 log dx is an improper integral of second kind because for every  > 0
x
x = and since f (x) = log
because lim0 logx
x is continuous on [0 + , 1], f is bounded
x
and integrable on [, 1]. Moreover, to evaluate this integral we can use integration by part
R1 x 1/2 R 1 1 x1/2 R 1 1/2
formula as follows  log dx = log x x
x
1 log 
1/2 |  x 1/2 dx = 2 1/2 2  x dx. Now
1/
by LHospital rule lim0 1/2
log 
= lim0 (1/2)3/2 = lim0 
1/2 = 0 and by Second F.T.C.
R 1 1/2 1/2 1 R1 x R1
 x dx = x1/2 |1 = 22 2 . Thus lim0  log
dx = 2 lim0 log
x

1/2
2 lim0  x1/2 dx =
1
2(0) 2 lim0 (2 2 2 ) = 4.
Similar to series convergence there are certain tests to check the convergence of an improper
integral.
First such test is for nonnegative functions.
Theorem 6.39. Let f be integrable on [a, x], for all x > 0 and f (x) R0 for all x > a. Then
Rx
there exists M > 0 such that a f (t) dt M for all x a if and only if a f (t) dt converges.
Rx
Proof. (*) First suppose that there exists M > 0 such that R a f (t) dt M for all x a. Note
x
that by First F.T.C. F : [a,R) R, defined by F (x) = a f (t) dt, is a continuous and for
y
a < x < y, F (y) F (x) = x f (t) dt 0 so that it is monotonic increasing also. Since F is
bounded on [a, ) by M , there exists = sup{F (x) : a < x}. For any given  > 0, 
can not be an upper bound of {F (x) : a < x}, so there exists x0 such that F (x0 ) > .
But monotonicity of F implies R xthat  < F (x0 ) < F (x) R for all x x0 . And hence
limx F (x) = , or limx a f (t) dt = . Hence the integral a f (t) dt is convergent.
60
R
Next if a f (t) dt converges, then limx F (x) exists. Suppose limx F (x) = R xM . Since
F is increasing limx F (x) = sup{F (x) : a < x} (check it). Hence F (x) M , or a f (t) dt
M. 
A similar test can be proved for improper integral of second kind.
Theorem 6.40. Let f be integrable on [x, b], for all a < x b and f (x) 0 for all a x b.
Rb Rb
If there exists M > 0 such that x f (t) dt M for all a < x b, then a f (t) dt converges.
Theorem 6.41. (Comparison Test for First kind of Improper Integral) Suppose R f and
g both are integrable
R on [a, x], for all x > a and 0 f (t) g(t), for all t > a. If a g(t) dt
converges, then a f (t) dt converges.
R
Proof. Proof of this theorem
Rx is a consequence of Theorem (6.39).
Rx If a g(t)
R x dt converges, then
there is M such that a g(t) dt MR for all x a. But a f (t) dt a g(t) dt, and hence
Rx
a f (t) dt M for all x a, so that a f (t) dt converges. 
Corollary 6.42. Suppose f and g both areR integrable on [a, x], for all x > a and 0 f (t) g(t),
R
for all t > a. If a f (t) dt diverges, then a f (t) dt diverges.
R 2
Example 6.43. To find the behaviour of improper integral 0 ex dx we first observe that
2
ex is a continuous function [0, ] and it is bounded by e0 = 1. So this is an improper integral
of secondR kind. To check its convergence using definition we need to evaluate the Riemann
y 2
integral 0 ex dx for all y > 0 and for this we need to know the antiderivative of the function
2 2
ex , that is, a function whose derivative is ex . Since we dont know such function, we will
2
use comparison test. We know that ex is comparable with Rex , because 1 < x x <
2 M
x2 x2 < x1 ex < ex . And since Riemann integral 1 ex dx = eM + e, hence
2
lim M eM + e = e. Thus 1 ex dx converges, hence 1 ex dx converges. Moreover,
R R
R 1 x2 2
0 e dx, is a proper Riemann integral of a continuous function, so ex is R-integrable on
R1 2 R 2 R 2
[0, 1]. Thus 0 ex dx + 1 ex dx exists, hence 0 ex dx converges.
R R
Example 6.44. 1 2+cos x
x
dx diverges because 2+cos
x
x
> x1 > 0 for all x > 1 and 1 x1 dx(=
RM
limM 1 x1 dx = limM log M = ) diverges.
R 2 2 R
Example 6.45. 1 sinx2 x dx converges because sinx2 x > x12 > 0 for all x > 1 and 1 x12 dx
converges.
The Comparison Theorem for second kind improper integral can be stated as follows.
Theorem 6.46. (Comparison Test for Second kind Improper Integral) Suppose f and
Rb
g both are integrable on [a + , b], for all  > 0 and 0 f (t) g(t), for all a < t b. If a g(t) dt
Rb Rb Rb
converges, then a f (t) dt converges. And if a f (t) dt diverges, then a g(t) dt diverges.
Proof. The proof is similar to the comparison test of first kind integral. 
R1
Example 6.47. 0 (1cos1t+x) dx is an improper integral of second kind because 1
(1cos t+ x)
goes to as x 0 . Note that 0 < (1cos1t+x) 1x for all 0 < x 1 and the improper
R1
integral of second kind 0 (1x) dx is convergent. So by comparison theorem the improper integral
R1 1
0 (1cos t+ x) dx converges.
As a consequence of Comparison test one can prove Limit comparison test.
Theorem 6.48. (Limit Comparison Test for Second kind Improper Integral) Suppose
f and g both are integrable on [x, b], for all x > a and 0 f (t), 0 < g(t), for all a < t b.
Rb Rb
If limxa fg(t)
(t)
= l and if l > 0, then a g(t) dt converges if and only if a f (t) dt converges.
Rb
And if l = 0, then the convergence of the improper integral a g(t) dt implies the convergence
Rb Rb Rb
of a f (t) dt, or equivalently, if a f (t) dt diverges, then a g(t) dt diverges.
61
f (t)
Proof. (*) If limxa g(t) = l 6= 0, then there is an  > 0 such that l  > 0 and there exists
f (t)
> 0 such that l  < < l +  for all a < x < a + . Thus f and g are comparable in
g(t)
R a+ R a+
(a, a + , by comparison test a f (t)dt converges if and only if a g(t)dt converges. But by
Rb
hypothesis both f and g are integrable in [a + , b], therefore a f (t)dt converges if and only if
Rb
a g(t)dt converges.
Next if l = 0, then for any  > 0, there exists a > 0 such that 0 fg(t) (t)
< l +  for
R a+
all a < x < a + . Thus by comparison test the convergence of improper integral a g(t)dt
R a+
implies the convergence of a g(t)dt. But by hypothesis both f and g are integrable in [a+, b],
Rb Rb
therefore a f (t)dt converges if a g(t)dt converges. 
R1 1
Example 6.49. The integral 0 log (1+ x)
dx is an improper integral of second kind because

1
log (1+ x)
goes to as x 0+ . But if g(t) = 1x , then limx0+ fg(t)
(t)
= limx0+ log (1+xx) =

limx0+ [(1/2)x1/2 ]/[(1 + x)1 (1/2)x1/2 ] = limx0+ (1 + x) = 1 6= 0. Hence by L.C.T.
R1 1 R1 1
converging behaviour of 0 x dx implies the convergence of 0 log (1+ x)
dx.
R1
Example 6.50. 0 xlog1(1+x) dx is an improper integral of second kind because xlog1(1+x) goes
to as x 0+ . Observe that lim x 0+ xlogx2(1+x) = 1/2 6= 0, so we can consider the function
R1 R1
g(x) = 1/x2 . And since 0 x12 dx diverges, by L.C.T. 0 xlog1(1+x) dx diverges.
Similarly one can prove the Limit comparison test for first kind improper integral.
Theorem 6.51. (Limit Comparison Test for First kind Improper Integral) Suppose
f and g both are integrable on [a, x], for all x a and 0 f (t), 0 < g(t), for all a t. If
R R
limx fg(t) (t)
= l and if l > 0, then a g(t) dt converges if and only if a f (t) dt converges. And
R
if
R l = 0, then the convergenceR of the improper integral R a g(t) dt implies the convergence of

a f (t) dt, or equivalently, if a f (t) dt diverges, then a g(t) dt diverges.
R sin 1t sin 1t 1
Example 6.52. Using L.C.T. the integral 1 t dt converges, because lim t t /( t2 ) =

limt cos( t )(t2 )/(t2 ) = limt cos( t ) = 1 6= 0 and the integral 1 t2 dt converges.
1 1 1
R

R s 1+x R 0 1+x
Problem 6.53. Show that 0 1+x
R s 1+x 2 dx and s 1+x2 dx do not approach a limit as s .
However lims s 1+x2 dx exists.
1+x 1 (1+x)2 R 1+x
Solution. Since lims 1+x 2 / 1+x = lims 1+x2 = 1 6= 0 so by L.C.T. 0 1+x2
dx diverges
R 1
because 0 1+x dx diverges.
1
(Note that if instead of 1+x we use x1 for Limit comparison with 1+x 1+x
2 , then in this case
R
also lims 1+x2 / x = 1. But there is a little ambiguity to say that since 0 x1 dx diverges,
1+x 1
R 1+x R 1 R1 1
0 1+x 2 dx diverges because 0 x dx diverges due to two reasons one that 0 x dx diverges and
R 1 R 1 R 1+x
the integral 1 x dx diverges. On the other hand to say that 1 x dx diverges, so 1 1+x2 dx
diverges is all right.)
R 0 1+x
Now since s 1+x 2 dx is R-integrable for all s > 0 we can use substitution x = y to
R 0 1y R s 1y R s y1 1
observe s 1+y2 (1)dy = 0 1+y 2 dy = 0 1+y 2 dy so that by L.C.T. with 1+y we can say that
R y1 R 1
0 1+y 2 dy diverges because R 0 1+y dy diverges.
s 1+x 1+x
Next to show that lims s 1+x 2 dx exists, we fist observe that 1+x2 is R-integrable in [s, s]
R s 1+x R 0 1+x R s 1+x R s 1y R s 1+x R s 1x
for s > 0 so s 1+x2 dx = s 1+x2 dx + 0 1+x2 dx = 0 1+y2 dy + 0 1+x 2 dx = 0 1+x2 dx +
R s 1+x Rs 1 R 1 1
0 1+x2 dx = 2 0 R1+x2 dx. But 1 1+x2 dx converges because of L.C.T. with the function x2

and the fact that 1 x12 dx converges.
62
R
Problem 6.54. Find the range of p such that 1 et tp dt converges.
Solution. We know that exponential function ex can be represented as Taylor series around
the point 0. And it contains all positive integer powers of x. Keeping this in mind we hope that
n
limt tet should be zero for any n N, in fact one can use LHospital rule repeatedly n times
n
to show that limt tet = limt n!et = 0.
t p
Now coming back to problem, for any p we can show that limt et2t is equal to zero
t p
because if p + 2 < 0, then tp+2 < 1 for all t > 1 and 0 < et2t < et hence by sandwich theorem
t p
for limits limt et2t = 0, further if p + 2 > 0, then the repeated application of LHospital
p+2 p+2
rule for [p + 2] + 1 times to limt t et , we get limt t et = limt Ktp+2[p+2]1 et = 0,
where K is some constant R depending on p and p + 2 [p + R2] 1 < 0. Hence by L.C.T. the
convergence of integral 1 t12 dt implies the convergence of 1 et tp dt.
R x log x
Problem 6.55. Prove that 0 (1+x 2 )2 dx = 0.

R x log x R 1 x log x
Solution. We will try to show that 1 (1+x 2 )2 dx converges and equals to 0 (1+x2 )2 dx.
x log x x log x 1
For this we first observe 0 (1+x 2 )2 for all x > 1 and limx (1+x2 )2 / x2 = 0, so that by
R R x log x
L.C.T. the convergence of 1 x12 dx implies the convergence of 1 (1+x 2 )2 dx. Now since for
x log x
any M > 0, (1+x2 )2
is R-integrable on [0, M ], so that we can use substitution x = 1/y to
R M x log x R 1/M y(1) log y R1
observe that 1 (1+x2 )2 dx = 1 (y 2 +1)2
(1)dy = 1/M (yy2log y
+1)2
dy. Thus it implies that
R M x log x R1 x log x R x log x R 1 x log x
limM 1 (1+x2 )2 dx = limM 1/M (1+x2 )2 dx, or 1 (1+x2 )2 dx = 0 (1+x 2 )2 dx.

R 1ex
Problem 6.56. Find the range of p for which the integral 0 xp (1+x) dx converges.
Solution. The given integral is not only a first type improper integral, but it might be of second
type also because x1p will approach to infinity for some range of p. So we split this integral in
R1 x R 1ex 1ex
two parts 0 x1ep (1+x) dx and
1
1 xp (1+x) dx. Now limx0+ xp (1+x) / xp1 = 1, so that by L.C.T.
R 1 1ex R1 1
0 xp (1+x) dx converges if and only if 0 xp1 dx converges, equivalently p 1 < 1.
x R x
And since limx x1e 1
p (1+x) / xp+1 = 1, by L.C.T. 1 x1e p (1+x) dx converges if and only if
R 1
1 xp+1 dx converges, equivalently p + 1 > 1. Thus the common range of p in which both
parts converges is 0 < p < 2.
Question 6.57. We proved certain results for non-negative functions. Can we use these results
some how to find the behaviour of improper integrals of any real valued functions?
6.58. Absolute Convergence
R R
We say that an improper integral a f (t)dt converges absolutely, if integral a |f (t)|dt con-
verges.
In the following theorem we show that the absolute convergence of an improper integral
implies the convergence of the integral.
Theorem 6.59. (Absolute Convergence Test for Improper R Integrals) If f and |f | both
are
R integrable on [a, x], for all x a and improper integral a |f (t)|dt converges, then integral
a f (t)dt converges.
R
Proof. Note that 0 f (t) + |f (t)| 2|f (t)| for all t a. Now if a |f (t)|dt R converges, by
comparison test for the functions f (t) + |f (t)| and 2|f (t)|, the integral a (f (t) + |f (t)|)dt
converges.
R x (Since weR xare applying Comparison test we need R x the existence of theR x integrals of the
form a f (t)dt and R a |f (t)|dt for all x > a.) So lim x a f (t)dt = lim x a (f (t) + |f (t)|
x Rx
|f (t)|)dt = limx a (f (t) + |f (t)|)dt limx a |f (t)|dt = converges. 
R sin t
Remark 6.60. Converse of the above result is not true. In fact later we will show that 1 t dt
R sin t
converges but 1
t
dt diverges.
63
Remark 6.61. One can also see easily that the condition, f and |f | both are integrable on
[a, x], for all x a, is necessary because if one can consider the function f (x) = x12 for all
R R
rational x in (1,R) and f (x) = x12 for all irrational x in (1, ), then 1 |f (t)|dt = 1 t12 dt
x
converges. But 1 f (t)dt does not exist for any x > 1, hence it can not converge.
A similar result is true for second kind improper integrals.
Rb R
Theorem 6.62. If a second kind improper integral a |f (t)|dt converges, then a f (t)dt con-
verges.
R1
Example 6.63. To show the convergence of the integral 0 sin( 1t )dt we will show that it con-
R1
verges in absolute sense, that is, 0 sin( 1t ) dt is convergent. And the absolute convergence can

R1
be shown using comparison test and observation 0 < sin( 1t ) < 1 for all 0 < t 1 and 0 1dt

is convergent.
Theorem 6.64. (Dirichlet Test ) If f, g : [a, ) R satisfy
(1) f (t) is a decreasing function and limt f (t) = 0, and
Rx
(2) g is continuous and there is some M > 0 such that | a g(t)dt| < M ,
R
then the improper integral a f (t)g(t)dt converges.
We will not prove this result but we will use this for solving problems.
R
Example 6.65. The integral 1 sint t dt converges because f (t) = 1t decreases to zero as x
and g(t) = sin t is continuous and for any x there is a unique k Z such that 1 + 2k < x
Rx R 1+2k Rx Rx R
1 + 2(k + 1) so that 1 sin tdt = 1 sin tdt + 1+2k sin tdt = 0 + 1+2k sin tdt < 0 sin tdt,
Rx R
which implies | 1 sin tdt| < 2. Thus Dirichlet test is applicable to integral 1 sint t dt to conclude
the convergence of the integral.
6.66. Conditionally Convergent
An improper integral is called conditionally convergent if the integral converges but does not
converge absolutely.
R t
Problem 6.67. Show that the integral 1 sin tp dt converges conditionally for 0 < p 1.
Rx
Solution. Since t1p decreases to zero as t for all 0 < p 1 and | 1 sin tdt| < 2 for all
R sin t
x > 1, the integral 1 tp dt converges due to Dirichlet test. But to show that it does not
2
converge in absolute sense we first observe that | sin t|2 < |sint| so that | sintpt| | sin tp
t|
and
R | sin t|2 R sin2 t
try to show that the integral 1 tp dt = 1 tp Rdt diverges. For any M > 0 we know by
M 2
Riemann integrability of continuous functions that 1 sintp t dt is R-integrable so that we can
RM 2 RM RM RM
use integration by parts to get 1 sintp t dt = 1 1cos tp
2t
dt = 1 t1p dt 1 costp2t dt. In limit
RM RM
M , 1 costp2t dt converges by Dirichlet test (check it) and 1 t1p dt diverges for 0 < p 1.
R
So limM 1 sintp t dt diverges and hence 1 | sin
RM 2 t|
tp dt diverges by comparison test.
R 2 R
Problem 6.68. Show that the integrals 0 sinx2 x dx and 0 sinx x dx converge. Further, prove
R 2 R
that 0 sinx2 x dx = 0 sinx x dx.
Solution. Since x1 appears in both the integrals and it approaches to infinity as x 0+ , we
R1 2 R1
can not say directly that the integrals 0 sinx2 x dx and 0 sinx x dx are Riemann integrable. But
R1 2 R1
since limx0+ sinx x /1 = limx0+ cos1 x = 1 6= 0, by L.C.T. 0 sinx2 x dx converges because 0 1dx
R1
converges. And 0 sinx x dx converges absolutely because of L.C.T. with the constant function 1
and the fact that the limit limx0+ sinx x /1 = limx0+ sinx x = 1 6= 0.

R Now
sin2 x
we only Rneed to show the convergence of the improper integral of first
sin x sin2 x 1
R kind,
sin2 x
that is,
1 x2
dx and 1 x dx. Since 0 < x2 < x2 by direct comparison test 1 x2 dx con-
R R
verges because 1 x12 dx converges. Moreover 1 sinx x dx converges by Dirichlet test (see above
example).
64
6.69. Applications of Riemann Integration
We will briefly discuss the formulas for finding the area of a planer region bounded by graphs
of certain known functions.
6.70. Area in Cartesian Coordinates
If f (x) 0, then the area of the region between the graph of the function and the X-axis is
defined as
Z b
(6.15) Area = f (x)dx.
a
Next if f (x) g(x) for all a < x < b, then the area of the region between the graphs for f and
g and x = a and x = b line is defined to be
Z b
(6.16) Area = (f (x) g(x)) dx.
a
6.71. Polar Coordinates
Suppose if O is the origin and OA is the initial line, then if the point P is at a distance r
from origin, that is, OP = r and line segment OP makes the angle with the initial ray OA,
then we say that the polar coordinates of the point P are (r, ). Mostly in XY -plane we assume
the X-axis as initial line and relate the cartesian coordinates (x, y) of the point P with polar
coordinates as
p y
x = r cos , y = r sin , and r = x2 + y 2 , = tan1 ( ).
x
In general to draw the graphs of the function defined by r = f (), following points are helpful.

1-Periodicity:
If there is no change in the equation of the function by replacing by + , then the function
is periodic and we only need to draw the graph between = 0 to = . To draw the graph
in < < 2 and 2 < < 3 and so on we rotate the graph of 0 < region by angles
, 2 and so on.
For example r = 1 + sin can be drown between 0 < 2 and then for other 0 s,
2 < 4 the same graph can be rotated by 2 angle and so there is no extra figures.
The function r = 2 cos 4 is periodic with period /2, so we only need to draw it between
= 0 to = /2 and we can draw the same graphs between /2 < < , < < 23 and
3 3
2 < < 2 by considering the initial ray as = /2, = and = 2 respectively, or
simply by copying and pasting the graph between 0 < 2 three times first after rotation of
angle /2, and then 23 .

2-Symmetry:
Since the reflection of any point P = (r, ) about the origin is a point Q, whose polar
coordinates are (r, ), or (r, + ), we say that the graph of the function represented by
r = f () is symmetric about the origin if there is no change in the equation of the graph either
by replacing r by r or by + . In general, if one coordinates representations satisfies the
equation of the function, then it is not necessary that the other coordinates representation also
satisfies the equation. For example ( 12 , 16 ) satisfies the equation r2 = cos() but ( 12 , 76 )
does not satisfy.
Moreover a function has a symmetry about the ray = , if there is no change in the equation
by replacing by 2. For example r = 1 + sin() is symmetric about the ray = /2.
If the initial ray is X-axis, then the no change in the equation of the function by simultaneously
replacing r to r and to also implies the symmetry about Y -axis and the no change in the
equation by simultaneously replacing r to r and to implies the symmetry about X-axis.

65
3-Valid range of r or :
The above equation also suggests that there is a particular range of the angle , say /2 <
< /2, only for which we can get a real value of r, so we do not need to draw the graph in
the complement of this range, that is /2 < < 3/2.
And since 1 cos() 1, the equation r2 = cos() implies that 1 r2 1. This show
that the complete graph of the function will lie within or on the boundary of the circle |r| = 1.
The graph of the function r = f () = 3 + 2 cos(2) lies in the annulus 1 |r| 5.

4-Analytical Properties:
Certain analytical properties of the function are very helpful for drawing the graph. Suppose
r = f () is differentiable with respect to (), then the derivative d dr
= f 0 () determines the
increasing and decreasing behaviour and local extremum of the function r = f () whenever
particle move continuously on the curve with increasing values of .
Moreover one can show that if the tangent line at some point P = (r, ) of the curve, makes
dr
an angle with the radius vector at the point P , then tan = r/( d ) = f 0r() and the slope of
the tangent vector from Xaxis is given by
f 0 () sin f () cos
tan( + ) = .
f 0 () cos f () sin
Above equation also shows that if the curve passes through the origin at an angle 0 , that is,
0 = f (0 ), then tangent vector makes the angle 0 from Xaxis.
6.72. Area in Polar Coordinates
Here we first note that the elemental area of region bounded by rays = 0 , = 0 + and
the curve r = f (), is given by r2
2 so that the Riemann sum for finding the area between the
rays = , = for some partition P = { = 0 , 1 , . . . , n = } of with ck = k is
given as
n n
X f 2 (i ) X f 2 (i )
S(f, P ) = (i i1 ) = i .
2 2
i=1 i=1
Continuity of f implies the Riemann integrability of f and thus the actual area is given by
Z
1
f 2 ()d.
2
The area between two graphs r = f1 () and r = f2 () such that f1 () f2 () for all < <
is given by
Z
1
[f 2 () f12 ()]d.
2 2
6.73. Volume of a solid by slicing
If a solid is bounded between two parallel planes and the area of the cross section of the
solid by an other parallel plane, which is at a distance t from one fixed plane t = 0 parallel to
boundary planes t = a and t = b, is A(t) (which is continuous in t), then we partition [a, b] as
P = {a = t0 , t1 , t2 , . . . , tn = b} to define the Riemann sum for finding the volume of the solid as
n
X
S(P, A) = A(ti )(ti ti1 ).
i=1
And by theory of integration the actual volume can be defined as Riemann integral
Z b
Volume of the solid = A(t)dt.
a
One can also use this formula for finding the volume of a annular solid.
6.74. Volume of a solid of revolution (Washer Method)
66
Volume of a solid formed by revolving a planer region about a line, which does not intersect
the interior of the region, can also be found using the above method.
If f, g : [a, b] R are two continuous functions such that 0 f (x) g(x) for all t [a, b], then
volume of the solid, formed by revolving the region bounded by the graphs of the functions and
the straight lines x = a, x = b about the X-axis, can be obtained by slicing such a solid by planes
perpendicular to the axis of revolution. Note that in this case we find the elementary volume
as the volume of a solid looking like a washer. And the width of this washer is i = xi xi1
and area is A(xi ) = (g 2 (xi ) f 2 (xi )), so that the Riemann sum for the volume of the solid is
n
X n
X
S(P, A) = A(xi )(xi xi1 ) = [g 2 (xi ) f 2 (xi )]i .
i=1 i=1
And using Riemann integration the volume of solid is given by
Z b
Volume of a solid of revolution = [g 2 (x) f 2 (x)]dx.
a
Next if the the function f is uniformly zero, then the solid formed by revolution is not an
annuls and the volume in this case is
Z b
Volume of a solid of revolution = g 2 (x)dx.
a

6.75. Volume of a solid of revolution (Shell Method)


If f, g : [a, b] R are two continuous functions such that f (x) g(x) for all 0 a x b],
then the volume of the solid, formed by revolving the region bounded by the graphs of the
functions and the straight lines x = a, x = b about the Y -axis (which does not intersect the
interior of the region) can be obtained using shell method. In this case our elementary object is a
shell, which can be defined as the part of the solid, which lies between two cylinders, whose axis
coincide with the axis of revolution and radii are x and x + x and hight is g(x) f (x). Thus
the Riemann sum for finding volume of the solid for a partition P = {a = x0 , x1 , . . . , xn = b} is
given as
n
X n
X
S(P ) = 2xi [g(xi ) f (xi )]i = 2 xi [g(xi ) f (xi )](xi xi1 ).
i=1 i=1
And the volume of such a solid is
Z b Z b
Volume = 2x[g(x) f (x)]dx = 2 (radius of the shell) (height of the shell) dx.
a a
6.76. Length of the plane curve
Let f : [a, b] R be a differentiable function such that f 0 is continuous. To find the
length of the curve obtained as the graph of the function between x = a, x = b, we first find the
elementary length estimate of some small portion of the curvep joining (x, f (x)) and (x+x, f (x+
x )) as length of line segment joining these points, that is, 2x + (f (x + x ) f (x))2 . Thus
the Riemann sum for the total length of the curve with partition P = {a = x0 , x1 , . . . , xn = b}
of [a, b] is given as
n p
X
S(P ) = (xi xi1 )2 + (f (xi ) f (xi1 ))2
i=1
s
n  2
X f (xi ) f (xi1 )
= 1+ |(xi xi1 )|
(xi xi1 )
i=1
n q
X
= 1 + (f 0 (ci ))2 |i |.
i=1
67
The total length of the arc is can be obtained as the limit of Riemann sum S(P ) as kP k 0,
that is,
Z bq
Length of curve = 1 + (f 0 (x))2 dx.
a
In a similar way length of a parametric curve (x(t), y(t)) for a t b is given as
Z bq
Length of a parametric curve = (x0 (t))2 + (y 0 (t))2 dt.
a
Using above equation one can find the length of a curve r = f (), given in polar coordinates,
by defining it as a parametric curve x() = f () cos , y() = f () sin and as the parameter
ranging between as follows
s  2
Z q Z
2 0 2 2
dr
Length of a curve in polar coordinates = (f ()) + (f ()) d = r + d.
d
Note that in above cases the dx, dt and d are considered to be positive.
6.77. Area of a surface of revolution
Suppose f : [a, b] R is a non-negative continuous function so that the graph of the function
defies a curve between a x b. The area of the surface generated by revolving the curve
around the X-axis can be given as
Z b q
2f (x) 1 + (f 0 (x))2 dx.
a
If the curve is given in parametric form say {(x(t), y(t)) : a t b} and the derivatives x0
and y 0 are continuous, then the area of the surface generated by revolving the curve around
an axis, whose perpendicular distance from the point (x(t), y(t)) is (t), (which is a continuous
function of t and depend upon the axis of revolution), is given as
Z b q
2(t) (x0 (t))2 + (y 0 (t))2 dt.
a
In case of polar form representation of the curve the surface area is given by
s  2
Z b
2
dr
2() r + d,
a d
where () is the distance of the point (x(), y()) from the axis of revolution. For example if
the curve is revolved around the X-axis, then
s  2
Z b
2
dr
area of the surface generated by revolution = 2 r() sin r + d.
a d
6.78. Centroid
The center of mass of a curve or a region is known as centroid, if the density function is
constant. Thus the coordinates of the centroid for a parametric curve can be given as
Rb q
(x0 (t))2 + (y 0 (t))2 dt
Rb
x(t)ds a x(t)
x = aR b = R q , and
ds b 0 2 0 2
a a (x (t)) + (y (t)) dt
Rb q
Rb 0 2 0 2
a y(t)ds a y(t) (x (t)) + (y (t)) dt
y = Rb = .
ds L
a
6.79. Pappus Theorem
68
Theorem 6.80. (Pappus Theorem for volume) If a plane region is revolved once about a
line in the plane that does not cut through the regions interior, then the volume of the solid it
generates is equal to the regions area times the distance traveled by the regions centroid during
the revolution. If is the distance from the axis of revolution to the centroid and A is the area
of the region, then
V = 2A.
Theorem 6.81. (Pappus Theorem for area) If an arc of a smooth plane curve is revolved
once about a line in the plane that does not cut through the arcs interior, then the area of the
surface generated by the arc equals the length of the arc times the distance traveled by the arcs
centroid during the revolution. If is the distance from the axis of revolution to the centroid,
and L is the area of the curve, then
S = 2L.
Use the reference book Calculus and Analytic Geometry by Thomas and Finney
for examples and problems of this section.

69

Potrebbero piacerti anche