MTH 303: REAL ANALYSIS I

NIKITA AGARWAL

Abstract. These are lecture notes for the course taught in Semester 1, 2023-24.

Contents
1. Week 1 1
2. Week 2 4
3. Week 3 11
4. Week 4 16
5. Week 5 24
6. Week 6 31
7. Week 7 34
8. Week 8 38
9. Week 9 44
10. Week 10 50
11. Week 11 56
12. Week 12 64
13. Week 13 70
14. Week 14 76
15. Some solved problems 82

1. Week 1
There were only two lectures this week (Tuesday and Thursday). Friday’s lecture was
canceled due to a schedule conflict.
What is this course about?
• Analyzing real numbers.
• Studying sequences and series of real numbers.
• Analyzing real-valued functions.
• A theoretical foundation for the concepts from calculus.
• You will see familiar objects, will see concepts which might seem intuitive. This
course will be about making concepts precise by proving them.
• Gradually introducing ‘short’ proofs in the beginning to ‘long’ proofs in the end.
What are real numbers?
• Natural numbers: 1, 2, 3, . . .
• Integers: 0, ±1, ±2, ±3 . . .
• Rational numbers: of the form p/q, where p, q are integers and q 6= 0.
Theorem 1.1. For any rational number r, there is a unique integer n such that
n ≤ r < n + 1.
Date: November 16, 2023.
1
2 NIKITA AGARWAL

Proof. First assume r is a positive rational number.

Let r = p/q, where p, q are positive integers.
Then p = kq + s, where k is a non-negative integer and 0 ≤ s < q is an integer
(division algorithm – number theory!).
Therefore r = k + s/q.
Take n = k which is a unique choice. This number k is called the greatest integer
of r, and is denoted by [r]. 
• There are gaps in the number line. Why?
– What is the length of the diagonal of a square with each side 1? It is a
number whose square √ is 2, by Pythagoras theorem.
– Let us denote it by 2.
– It is not a rational number.
√
Theorem 1.2. The number 2 is not rational.
√ p
Proof. Suppose it is. Then 2 = ⇒ 2q 2 = p2 , where p, q are positive integers and are
q
in ‘reduced form’, that is p and q have no common factor other than 1.
Thus both p, q cannot be even.
Since 2q 2 = p2 , p2 is even, hence p is even (thus q is odd).
Let p = 2k, then 2q 2 = 4k 2 which implies q 2 = 2k 2 .
This is impossible since q is odd.
Hence we arrive at a contradiction. √
Therefore our original assumption that 2 is a rational number was wrong. 
√
Where is 2 on the number line?
Theorem 1.3. For every rational number r > 0, there exists a non-negative rational
number x such that
x2 < 2 < (x + r)2 .
Proof. Let r > 0 be a fixed rational number.
Suppose there is no such x.
Then for every non-negative rational number x, if x2 < 2, then (x + r)2 < 2.
Take x = 0, 0 < 2, hence r2 < 2.
This implies (r + r + · · · + r)2 = (nr)2 < 2, for every n ∈ N.
Since nr2 < (nr)2 , we get n < 2/r2 , for every n ∈ N, which contradicts Theorem 1.1. 
Summarize: What are real numbers R?
• Natural numbers N.
• Integers Z.
• Rational numbers Q.
• Irrational numbers R \ Q: Any point on the number line which is not a rational
number.
• The points on the number line are in a one-to-one correspondence with the set
of real numbers R. That is, for every point on the number line, there is one and
only one point in R, and vice-a-versa.
Algebraic properties of R. There are two algebraic operations on the set R of real
numbers: addition ‘+’ and multiplication ‘.’. We now list algebraic properties on R.
(1) Commutativity of addition: x + y = y + x, for all x, y ∈ R.
(2) Associativity of addition: x + (y + z) = (x + y) + z, for all x, y, z ∈ R.
(3) Existence of the additive identity: There exists 0 ∈ R such that x + 0 = 0 + x, for
all x ∈ R.
 LECTURE NOTES ON REAL ANALYSIS I 3

(4) Existence of additive inverse: For every x ∈ R, there exists y ∈ R such that
x + y = 0. The inverse y is denoted by −x.
(5) Commutativity of multiplication: x.y = y.x, for all x, y ∈ R.
(6) Associativity of multiplication: x.(y.z) = (x.y).z, for all x, y, z ∈ R.
(7) Existence of the multiplicative identity: There exists 1 ∈ R such that x.1 = x, for
all x ∈ R.
(8) Existence of multiplicative inverse or reciprocal : For every x ∈ R, x 6= 0, there
exists y ∈ R such that x.y = 1. The inverse y is denoted by 1/x or x−1 .
(9) Distributivity of multiplication over addition: x.(y +z) = x.y +x.z, for all x, y, z ∈
R.
Order relation on R. There is an order relation on R, denoted as ‘<’. It satisfies the
following law:

Law of Trichotomy: For any x, y ∈ R, either x = y, or x < y, or y < x.

Note: y < x is also written as x > y.

Facts about the order relation:

(1) Transitivity: If x < y and y < z, then x < z.
(2) If x < y and z ∈ R, then x + z < y + z.
(3) If x < y and z > 0, then xz < yz.
(4) If x < y, then −y < −x.
(5) For any x ∈ R, x2 ≥ 0.
(6) If x > 0 and y < 0, then xy < 0.
(7) If 0 < x < y, then 0 < 1/y < 1/x.
1.1. Cardinality of sets.
• Finite set: A set which is either empty or is in a one-to-one correspondence with
the set {1, 2, . . . , n}, for some fixed n ∈ N. For example, {1, 2, . . . , n} itself is a
finite set.
• Countable set: A set which is either finite or is in a one-to-one correspondence
with N. If a countable set is not finite, then it is known as a countably infinite
set. For example, the set Q of rational numbers is countable. How?
Other than 0, all the positive numbers in Q are included below (but repeated).
Similarly all negative numbers in Q are countable.

Since a union of two countable sets is countable, Q is countable.

• Uncountable set: A set which is not countable.
• What about R? It contains Q, therefore R is not finite. It is, in fact, uncountable.
We will prove this fact later during the course.
4 NIKITA AGARWAL

2. Week 2
Comparing cardinality of sets. Cardinality of a set A is denoted as |A|. A finite set
with n elements is said to have cardinality n. We do not give the definition of cardinality
of infinite sets here. However, we compare cardinality of two sets as follows: Two sets A
and B are said to have the same cardinality if there is a bujection between the two sets.
If there is an injection from a set A to another set B, but no surjection, then we say that
the set B has more cardinality than the set A, in which case, we say that B is bigger
than A.
By the above discussion, if two sets have same cardinality and one of them is countable,
then the other is also countable. In view of this, once we prove that R is uncountable,
we can have that R has more cardinality than Q (since Q is countable).
The cardinality of Q (or N) is denoted as ℵ0 , and the cardinality of R is denoted as c.
Continuum hypothesis states that c = 2ℵ0 .
There is no biggest set! Let A be a non-empty set. Let P A denotes the power set of
A (set of all subsets of A). There is a one-to-one (injective) map from A to P A which
sends every element a ∈ A to the set {a} ∈ P A .
a 7→ {a}.
Hence the cardinality of P A is at least that of A, that is,
|A| ≤ P A .
Theorem 2.1. The cardinality of P A is strictly bigger than that of A. That is, there is
no bijection (one-one and onto) map from A to P A .
Proof. Consider a map f : A → P A .
Let
S = {a ∈ A : a ∈ / f (a)}.
Claim: there is no b ∈ A such that f (b) = S (this will contradict that f is an onto map.)
Suppose there exists b ∈ A such that f (b) = S. This element b is either in S or not in S.
If b ∈ S, then by the construction of S, b ∈
/ f (b) = S, which is absurd.
If b ∈
/ S and since f (b) = S, b ∈/ f (b). Hence b ∈ S, which is absurd as well (Russell’s
paradox.)
Hence we have produced a set S which is not in the image of f . Therefore f is not
onto. 
Let A be a non-empty set. Then
A
|A| < |P A | < |P P | < . . . .
Note: By convention, if we start with an empty set A, then each set in the above sequence
will be an empty set.
Upper bounds and Lower bounds of a set. Let us begin with a finite subset A =
{a1 , . . . , an } of R. Then A has a maximum element max(A) = M and a minimum element
min(A) = m. That is, m ≤ ai ≤ M , for all i = 1, . . . , n. We will now generalize the
notions of maximum and minimum of a finite set. Let A be a non-empty subset of R.
Definition 2.2. (Upper Bound) We say that u ∈ R is an upper bound of A if x ≤ u, for
all x ∈ A.
Definition 2.3. (Lower Bound) We say that ` ∈ R is an lower bound of A if ` ≤ x, for
all x ∈ A.
 LECTURE NOTES ON REAL ANALYSIS I 5

Note: A set need not have an upper and/or lower bound. For example, the set (0, ∞)
has no upper bound, and the set (−∞, 0) has no lower bound. More examples in the
figure below.

Definition 2.4. We say that A is bounded above if it has an upper bound. We say that
A is bounded below if it has a lower bound.
A set which is both bounded above and below is called a bounded set (or just bounded).
Upper bounds and Lower bounds - observations. Let A be a non-empty subset of
R.
• All points in A lie to the left of an upper bound and lie to the right of a lower
bound.
• Any lower bound is less than any upper bound of a non-empty set.
• Define the set −A = {−x : x ∈ A}. Then −A is a non-empty subset of R and
−(−A) = A. If u is an upper bound of A then −u is a lower bound of −A.
• Upper/Lower bound, if exists, is not unique.
• A real number c ∈ R is NOT an upper bound of A if there exists at least one
x ∈ A such that c < x.
• Similarly a real number d ∈ R is NOT a lower bound of A if there exists at least
one x ∈ A such that x < d.
Examples.
(1) Let A = (0, 1): Any number greater than or equal to 1 is an upper bound and
any number less than or equal to 0 is a lower bound for A. Note 0 ∈ / A is a
lower bound and 1 ∈/ A is an upper bound. Moreover no element of A is either an
upper or a lower bound. For if, 0 < x < 1 is an upper bound, then x < x+1 2
∈ A,
which contradicts the definition of upper bound. Similarly, if 0 < x < 1 is a lower
bound, then x > x2 ∈ A, which contradicts the definition of lower bound.
(2) Let A = {1/n :NewnSection
∈ 2N}:
Page 1 Any number greater than or equal to 1 is an upper

bound and any number less than or equal to 0 is a lower bound for A. Note 1 ∈ A
is an upper bound.
Finite set. Let A is a finite non-empty subset of R. Then
• A has a maximum element M = max(A) and a minimum element m = min(A).
• m is a lower bound of A and M is an upper bound of A. These are bounds that
lie in A.
• All other lower bounds are less than m and all other upper bounds are greater
than M .
6 NIKITA AGARWAL

• Thus m is the greatest lower bound (exists+unique) and M is the least upper
bound (exists+unique).
What happens for A = (0, 1).

Definition 2.5. (Least Upper Bound - LUB) Let A be a non-empty subset of R. We

say that u ∈ R is the least upper bound of A if
i) u is an upper bound of A and
ii) If u0 is another upper bound of A, then u ≤ u0 .
LUB is also known as ‘supremum’, denoted by sup(A).
Definition 2.6. (Greatest Lower Bound - GLB) Let A be a non-empty subset of R. We
say that ` ∈ R is the greatest lower bound of A if
i) ` is a lower bound of A and
ii) If `0 is another lower bound of A, then `0 ≤ `.
GLB is also known as ‘infimum’, denoted by inf(A).
Example 2.7. Let A = (0, 1).
• Any number greater than or equal to 1 is an upper bound.
• Any number less than or equal to 0 is a lower bound.
• As per previous discussion about this set, sup(A) = 1 and inf(A) = 0.
• Both bounds 0 and 1 do not lie in A.
Least upper bound property or axiom. Every non-empty subset A of R which is
bounded above in R has a least upper bound in R. That is, there exists u ∈ R such that
u = sup(A). New Section 3 Page 1

This is the single most important property of R which is NOT satisfied by the set of
2
√ the set A = {x ∈ Q : x < 2} is bounded in Q but
rational numbers Q. For instance,
its least upper bound sup(A) = 2 ∈/ Q.

LUB property can be proved using Dedekind cuts, which we omit here.
An application of LUB property - Archimedian property.
Theorem 2.8. (Archimedian property) The set of natural numbers N is not bounded
above in R.
Proof. Proof by contradiction.
Suppose N is bounded above by u ∈ R.
By the lub property, there exists u ∈ R such that u = lub(N).
Since u − 1 < u, u − 1 is not an upper bound of N.
Hence there exists n ∈ N such that u − 1 < n which implies u < n + 1.
Since n + 1 ∈ N, u is not an upper bound of N, which is a contradiction.
Hence our assumption that N is bounded above is wrong. 
 LECTURE NOTES ON REAL ANALYSIS I 7

An equivalent statement of Archimedian property.

Theorem 2.9. The following statements are equivalent:

i) Archimedian property.
ii) Given x, y ∈ R with x > 0, there exists n ∈ N such that nx > y.

Proof. i) ⇒ ii): Suppose there exists no such n ∈ N. Thus nx ≤ y for all n ∈ N. Hence
n ≤ y/x. This implies that y/x is an upper bound of N which is a contradiction to
statement i). Hence our assumption was wrong and statement ii) holds true.

ii) ⇒ i): Suppose N is bounded above. Let u be an upper bound. Take y = u, x = 1 in

the statement ii). Then there exists n ∈ N such that n > u. This is not possible since u
is an upper bound. 

Density of Q in R.

Theorem 2.10. Given any real numbers a, b ∈ R with a < b, there exists r ∈ Q such
that r ∈ (a, b).

Proof. Since b − a > 0, by the equivalent statement of Archimedian property, there exists
n ∈ N such that n(b − a) > 1.
Hence there exists an integer m ∈ Z such that m ∈ (na, nb) which implies m/n ∈ (a, b).
Take r = m/n. 

Density of R \ Q in R.

Theorem 2.11. Given any real numbers a, b ∈ R with a < b, there exists t ∈
/ Q such
that t ∈ (a, b).
√ √
√ (a − 2, b − 2).
Proof. Apply the previous theorem for√the interval
√ r ∈ Q such that r ∈ (a − 2, b − 2).
There exists
Hence r + 2 ∈ (a, b). √
Show that if r ∈ Q, then r + 2 ∈ / Q. 

How many are there? One would like to know how many rational numbers and irra-
tional numbers are there in the interval (a, b)?

Infinitely many. How? Let us ask the following questions.

• How many real numbers are there in (a, b)?
• How many irrational numbers are there in (a, b)?
• Can there be finitely many rational numbers in (a, b)?
If yes, let these be x1 , x2 , . . . , xk with a < x1 < x2 < · · · < xk < b. This can’t be
all ! There exists a rational number r ∈ (a, x1 ) ⊂ (a, b).
8 NIKITA AGARWAL

An example.
• Last time, we made a remark that Q does not enjoy lub property.
• We can define glb property as well: “A non-empty set of real numbers which is
bounded below has a glb/inf in R”.
An example of a countable bounded set A whose lub and glb both are not in A.
√ √
• Let A = {x ∈ Q : x2 < 2}. Claim: sup(A) = 2 and inf(A) = − 2.
• Note that A is a non-empty bounded subset of Q which does not enjoy lub and
glb. √ √
• First of all √ 2 is an upper bound and − 2 is a lower bound of A.
• Let u ∈√(0, 2) be any other upper bound of A. By density of Q in R, there exists
r ∈ (u, 2). Hence r2 < 2. Thus r ∈ √ A, which contradicts the assumption that u
is an upper bound. Hence sup(A) = 2. √
• Notice that A = −A. Hence inf(A) = −sup(−A) = −sup(A) = − 2.

Nested interval theorem (decreasing sets).

Theorem 2.12. Let In = [an , bn ] be closed intervals in R such that In+1 ⊂ In , for all
n ≥ 1. Then ∩n≥1 In 6= ∅.
Idea of proof
Sunday, August 23, 2020 9:11 PM

Proof. • Let L = {an : n ≥ 1} (collection of left-end points).

• Claim: Each bk is an upper bound for the set L.
• Consider an ∈ L. Let k ≥ 1 be given. If k ≥ n, then since Ik ⊂ In , an ≤ ak ≤
bk ≤ bn . If k < n, then since In ⊂ Ik , ak ≤ an ≤ bn ≤ bk . Thus each bk is an
upper bound of L.
• By the lub property, since L is a non-empty set which is bounded above, it has a
least upper bound. Let c = sup(L).
• Claim: c ∈ ∩n≥1 In .
• This follows immediately from two facts: since c = sup(L), i) an ≤ c, for all n, ii)
since each bn is an upper bound of L, c ≤ bn , for all n. Hence c ∈ In , for all n.


Counter-examples. The nested interval theorem might fail to hold if the intervals are
not closed.
Example 2.13. • Let In = (0, 1/n), for all n ∈ N. Prove that ∩n≥1 In = ∅.
• Let In = (0, 1/n], for all n ∈ N. Prove that ∩n≥1 In = ∅.
• Let In = (n, ∞), for all n ∈ N. Prove that ∩n≥1 In = ∅.
• Let In = (−1/n, 1/n), for all n ∈ N. Prove that ∩n≥1 In = {0}.
All above can be proved using Archimedian property.
 LECTURE NOTES ON REAL ANALYSIS I 9

Theorem 2.14. (Greatest Integer) For any real number x, there is a unique integer n
such that n ≤ x < n + 1.
Proof. We already proved this result when x ∈ Q.
Define the set
A = {m ∈ Z : m ≤ x}.
Then A 6= ∅, otherwise m > x, for all m ∈ Z.
Let n ∈ N and m = −n. Then −n > x, for all n ∈ N.
Hence n ≤ −x, for all n ∈ N. This contradicts the Archimedian property.
Thus A 6= ∅.
Also A is bounded above by x.
By the LUB property, A has a least upper bound, say u.
Since u − 1 is not an upper bound of A, there exists m ∈ A such that u − 1 < m ≤ u.
Also m ∈ A implies m ≤ x.
We prove that m + 1 > x.
For if, m+1 ≤ x, then m+1 ∈ A. Thus m+1 ≤ u which implies m ≤ u−1, contradiction.
Hence x < m + 1. Take n = m.
Uniqueness: Suppose n1 ≤ x < n1 + 1 and n2 ≤ x < n2 + 1. If n1 < n2 , then n1 + 1 ≤ n2 ,
hence n1 + 1 ≤ x, which is a contradiction. 
Remark 2.15. The n obtained in the previous result is known as the greatest integer less
than or equal to x and is denoted as [x]. The number x − [x] is known as the fractional
part of x.
Intervals in R.
Definition 2.16. A subset I of R is known as an interval if it satisfies the following
property: If x, y ∈ I, and x < z < y, then z ∈ I.
For real numbers a ≤ b, there are following types of intervals:
• Closed: [a, b]
• Open: (a, b)
• Half-open: [a, b)
• Half-open: (a, b]
• Closed: [a, ∞)
• Open: (a, ∞)
• Open: (−∞, b)
• Closed: (−∞, b]
• (−∞, ∞) = R.
The first four are finite or bounded, the other five are infinite or unbounded.
Definition 2.17. A subset U of R is said to be open if for each x ∈ U , there exists an
open interval (a, b) such that x ∈ (a, b) ⊆ U .
Example 2.18. • ∅ is open since the definition is vacuously satisfied
• An open interval is an open set. In particular, R is open.
• An arbitrary union of open sets is open.
• A finite intersection of open sets is open. Fails to hold for infinite intersections
since ∩n≥1 (−1/n, 1/n) = {0}, which is not open.
• Half-open intervals are not open sets. For instance, there is no open interval
containing the point b which is contained in the half-open interval (a, b].
Definition 2.19. A subset C of R is said to be closed if its complement R \ C is open.
10 NIKITA AGARWAL

Example 2.20. • ∅ is closed since R is open.

• R is closed since ∅ is open.
• A closed interval is a closed set.
• An arbitrary intersection of closed sets is closed.
• A finite union of closed sets is closed. Fails to hold for infinite unions since
∪n≥1 [0, 1 − 1/n] = [0, 1) (Check!), which is not open.
• Half-open intervals are not closed sets. For instance, the complement of (a, b] is
(−∞, a] ∪ (b, ∞), which is not open. There is a problem at the point a.
Definition 2.21. A subset A of R is said to be compact if it satisfies the following
property:
Let J be any index set. If (Ij )j∈J is a collection of open intervals such that A ⊆ ∪j∈J Ij ,
then there exists finitely many indices j1 , . . . , jk ∈ J such that A ⊆ Ij1 ∪ · · · ∪ Ijk .
Remark 2.22. 1) The collection (Ij )i∈J with A ⊆ ∪j∈J Ij is called an open cover of A.
3) A subcollection Ij1 , . . . , Ijk with A ⊆ Ij1 ∪ · · · ∪ Ijk is called a finite subcover of the
open cover (Ij )j∈J .
Theorem 2.23. (0, 1) is not compact.
Proof. In = (1/n, 1) is an open cover of (0, 1) since (0, 1) = ∪n In . Consider any finite
subcollection In1 , . . . , Ink such that (0, 1) ⊆ ∪j Inj and n1 < · · · < nk . Then ∪j Inj = Ink
since In1 ⊂ In2 . . . Ink . Hence (0, 1) ) ∪j Inj = Ink = (1/nk , 1). 
 LECTURE NOTES ON REAL ANALYSIS I 11

3. Week 3
Theorem 3.1. [a, b] is compact.
Proof. Suppose [a, b] is not compact.
Then there exists a collection (Ij )j∈J , of open intervals satisfying [a, b] ⊆ ∪j∈J Ij , but
having no finite subcollection whose union contains [a, b].
Divide [a, b] into two equal intervals [a, (a + b)/2] and [(a + b/2), b].
Then the union of all the intervals in the collection (Ij )j∈J contains [a, b].
For at least one of these subintervals [a, (a + b)/2], [(a + b/2), b], there is no finite sub-
collection of (Ij )j∈J whose union contains [a, b]. For if, there are finitely many intervals
Ii1 , . . . , Iik whose union contains the subinterval [a, (a + b)/2] and there are finitely many
intervals Ij1 , . . . , Ij` whose union contains the subinterval [(a + b)/2, b], then the union of
Ii1 , . . . , Iik , Ij1 , . . . , Ij` contains [a, b], which contradicts the original assumption (marked
in red).
Without loss of generality, suppose [a, (a + b/2)] is such that there is no finite subcollec-
tion of (Ij )j∈J whose union contains [a, (a + b)/2]. Set K1 = [a, (a + b)/2].
Now repeat the bisection process on K1 (subinterval with no finite subcollection of (Ij )j∈J
whose union contains it) to obtain K2 which also satisfies the same property: there is no
finite subcollection of (Ij )j∈J whose union contains K2 .
This way we get a decreasing sequence of closed intervals K1 ⊃ K2 ⊃ . . . Kn ⊃ Kn+1 ⊃
....
By the nested interval theorem, ∩n Kn is non-empty. Let c ∈ ∩n Kn .
Note that the length of the interval Kn is (b − a)/2n , for each n ≥ 1.
Since c ∈ [a, b] ⊆ ∪j∈J Ij , the point c ∈ Ij for some j ∈ J.
Let Ij = (p, q).
Choose N ∈ N such that (b − a)/2N < min{c − p, q − c} (there exists such N by the
Archimedian property).
Then c ∈ KN ⊆ Ij (in a single picture, draw the interval Ij = (p, q), the point c in it, and
the interval KN whose length is (b − a)/2N , which also contains c).
Since KN was chosen so that no finite subcollection of (Ij )j∈J whose union contains KN ,
and now we have produced a single set from this collection which contains KN . This is
a contradiction.
Hence [a, b] is compact. 

Cantor Set. This set was introduced by Georg Cantor in 1883, now known as the Cantor
set. There are several generalizations of the Cantor set. The most standard Cantor set
is described below, known as the Cantor ternary set.
• Let I = [0, 1].
• Let
A1 = I \ (1/3, 2/3) = [0, 1/3] ∪ [2/3, 1]
(remove the middle-third open interval to get 2 disjoint closed intervals).
• Let
A2 = A1 \ {(1/9, 2/9) ∪ (7/9, 8/9)} = [0, 1/9] ∪ [2/9, 1/3] ∪ [2/3, 7/9] ∪ [8/9, 1]
(remove the middle-third open interval to get 22 = 4 disjoint closed intervals of
length 1/32 = 1/9).
• and so on to obtain An , n ≥ 3. At this stage, we have 2n disjoint closed intervals,
each of length 1/3n .
• We obtain a nested sequence A1 ⊃ A2 ⊃ . . . An ⊃ An+1 . . . .
12 NIKITA AGARWAL

• Let
C = ∩∞
n=1 An .

• Then C 6= ∅. All points of the form k/3n , 1 ≤ k ≤ 3n , n ≥ 1, belong to C.

• We will prove later that C is an uncountable set. This will imply that R is
uncountable. Hence the set of irrational numbers is uncountable.
• An important observation: At the first step, we take away an interval of length
1/3. At the second step, we take away 2 intervals each of length 1/32 . At the
nth -step, we take away 2n−1 intervals each of length 1/3n , for each n ≥ 1. Hence
in the entire process, we have taken away intervals of total length
1 21 2n−1
+ 2 + · · · + n + · · · = 1,
3 3 3
which is the length of the interval [0, 1] we started off with.

Sequences.

Definition 3.2. Let A be a non-empty set. A sequence is a function f : N → A. We

define xn := f (n) for all n ∈ N, it is called the nth term of the sequence. The sequence is
written as
(x1 , x2 , . . . , xn , . . . ),
or simply (xn )n≥1 or (xn ).

Example 3.3. • Constant sequence: (c, c, . . . ), c ∈ R.

• (1, −1, 5, 41, −2, 23, −6, 100, 10100 , . . . ).
• f (1) = f (2) = 1, f (n) = f (n − 1) + f (n − 2), for n ≥ 3. This gives the famous
Fibonacci sequence (1, 1, 2, 3, 5, 8, 13, . . . ).
• (1/2, 2/3, 3/4, 4/5, . . . , n/(n + 1), . . . ).
• Consider the sequence 0.9, 0.99, 0.999, . . . . A few observations about terms:
– increasing as n increases,
– never become 1,
– become closer and closer to 1.
 LECTURE NOTES ON REAL ANALYSIS I 13

Monday, August 24, 2020 9:56 AM

Convergence.
Definition 3.4. Let (xn )n≥1 be a sequence of real numbers. We say that (xn ) converges
to x ∈ R if for any given  > 0, there exists N ∈ N such that |xn − x| < , for all n ≥ N .
Observe that |xn − x| <  ⇐⇒ xn ∈ (x − , x + )
If (xn ) converges, we say that (xn ) is convergent. The number x is called the limit
of the sequence and we write x = limn→∞ xn = limn xn or xn → x.
The sequence (xn ) is said to be divergent if it is not convergent.
Remark 3.5. • N depends on .
• If N works, then any other number larger than N also works.
• If N works for , then it works for any real number larger than .
Question: What does it mean to say that (xn ) does not converge to x? (denoted as
limn xn 6= x or xn 9 x.)
Answer: There exists  > 0 such that for all N ∈ N, |xn − x| ≥  for some n ≥ N .
Example 3.6. Consider the sequence xn = 1/2n , n ≥ 1. We prove that xn → 0.
• Let  > 0 we given.
• To find N ∈ N such that |1/2n − 0| < , for all n ≥ N .
• First note that if 1/2N < , then 1/2n < , for all n ≥ N . (decreasing terms)
• Let us work backwards.
• 1/2N <  if and only if N > log2 (1/). Archimedian property guarantees existence
of such an N ∈ N.
• An important point to note is that N depends on .
Example 3.7. Consider the sequence xn = (−1)n , n ≥ 1. We prove that xn is divergent.
• The sequence is (−1, 1, −1, 1, . . . ). Suppose the sequence converges to x ∈ R.
• Let 0 <  < 1. Since the length of the interval (x − , x + ) is 2 < 2, the interval
can contain either 1 or -1.
• Since all the odd terms equal -1 and all the even terms equal 1, there is no cutoff
N ≥ 1 such that |xn − x| < , for all n ≥ N .
Theorem 3.8. (Uniqueness of limit) Let (xn ) be a sequence of real numbers. If xn → x
and xn → y, then x = y.
|x − y|
Proof. • Suppose x 96=Pagey.1 Let  =
New Section > 0.
2
• Since xn → x, there exists N1 ∈ N such that |xn − x| <  for all n ≥ N1 .
• Since xn → y, there exists N2 ∈ N such that |xn − y| <  for all n ≥ N2 .
• Let N = max{N1 , N2 }. For n ≥ N , |xn − x| <  and |xn − y| < .
• Let n ≥ N . Consider |x − y| ≤ |x − xn | + |xn − y| < 2 = |y − x|, which is absurd.

Remark: Any  smaller than the one chosen here will work for this argument.
14 NIKITA AGARWAL

Example 3.9. Let an = (n + 1)/(2n + 3), for n ≥ 1. Does it converge? If yes, what is
the limit?
It does converge to 1/2.
• Let  > 0 be given. We will find a suitable N ∈ N.
• Let us work backwords.
• |(n + 1)/(2n + 3) − 1/2| = 1/2(2n + 3) < 1/n, since n < 2(2n + 3).
• Choose any natural number N > 1/. Again Archimedian property guarantees
existence of such a N .
• Then for n ≥ N , |(n + 1)/(2n + 3) − 1/2| < 1/n ≤ 1/N < .
• Thus 1/2 is the limit (by uniqueness of limit).
Theorem 3.10. (Change the first few terms) Let (xn ) be a sequence of real numbers
converging to x ∈ R. Fix K ∈ N. Let (yn ) be another sequence of real numbers with
yn = xn , for all n ≥ K. Then yn → x.
Proof. Let  > 0 be given.
Since xn → x, there exists N ≥ 1 such that |xn − x| < , for all n ≥ N .
Let N0 = max{N, K}.
Then for n ≥ N0 , |yn − x| = |xn − x| < . 
Theorem 3.11. Let (xn ) be a sequence of real numbers. If xn → x, then |xn | → |x|.
However the converse is not true.
Proof. Let  > 0 be given.
Since xn → x, there exists N ≥ 1 such that |xn − x| < , for all n ≥ N .
Consider ||xn | − |x|| ≤ |xn − x| < , for all n ≥ N .
Converse: xn = (−1)n diverges but |xn | = 1 converges being a constant sequence. 
Definition 3.12. A New Section 10 Page 1
sequence (xn ) is real numbers is said to be bounded if there exists
M > 0 such that |xn | ≤ M , for all n.
Example 3.13. xn = (−1)n is bounded with M = 1, but is not convergent.
Theorem 3.14. Every convergence sequence of real numbers is bounded. (Converse is
not true!)
Proof. Proof 1: Let (xn ) be a sequence of real numbers converging to x ∈ R. Then
|xn | → |x|. Hence for  = 1, there exists N ≥ 1 such that |xn | < |x| + 1, for all n ≥ N .
Let M = max{1 + |x|, |x1 |, |x2 |, . . . , |xN −1 |} ≥ 0.
Then |xn | ≤ M , for all n.

Proof 2: Let (xn ) be a sequence of real numbers converging to x ∈ R.

Take  = 1 (any number of your choice will work).
 LECTURE NOTES ON REAL ANALYSIS I 15

There exists N ∈ N such that |xn − x| < 1, for all n ≥ N .

Then |xn | ≤ |xn − x| + |x| < 1 + |x|, for all n ≥ N .
Let M = max{1 + |x|, |x1 |, |x2 |, . . . , |xN −1 |} ≥ 0.
Then |xn | ≤ M , for all n. 
Algebra of sequences.
Theorem 3.15. Let xn → x and yn → y and a ∈ R. Then
i) xn + yn → x + y.
ii) axn → ax.
iii) xn yn → xy.
iv) 1/xn → 1/x, if xn , x 6= 0.
Proof. i) We will use |(xn + yn ) − (x + y)| ≤ |xn − x| + |yn − y|, for all n.
Let  > 0 be given.
Since xn → x and yn → y, there exists N1 , N2 ∈ N such that |xn − x| < /2, for all
n ≥ N1 , and |yn − y| < /2, for all n ≥ N2 .
Let N = max{N1 , N2 }. Then for n ≥ N , |(xn + yn ) − (x + y)| ≤ |xn − x| + |yn − y| <
/2 + /2 = .

ii) If a = 0, the result holds trivially. Suppose a 6= 0.

We will use |axn − ax| = |a||xn − x|.
Let  > 0 be given.
Since xn → x, there exists N ∈ N such that |xn − x| < /|a|, for all n ≥ N .
Now for n ≥ N ,
|axn − ax| = |a||xn − x| < |a|./|a| = .

iii) We will use |xn yn − xy| ≤ |xn yn − xyn | + |xyn − xy| = |xn − x||yn | + |x||yn − y|.
Let  > 0 be given. We need three estimates.
a) Since (yn ) is convergent, it is bounded, hence |yn | ≤ M , for all n ∈ N , for some M > 0.
b) Since xn → x, there exists N1 ∈ N such that |xn − x| < /2M , for all n ≥ N1 .
c) Since yn → y, there exists N2 ∈ N such that |yn − y| < /2(|x| + 1), for all n ≥ N2 .
For all n ≥ N = max{N1 , N2 }, combining a), b), and c), we get
|xn yn − xy| ≤ |xn − x||yn | + |x||yn − y| < /2 + |x|/2(|x| + 1) < .

iv) Let  > 0 be given. We need two estimates.

a) Since xn → x, we have |xn | → |x| > 0. Thus for  = |x|/2 > 0, there exists N1 ≥ 1
such that |x|/2 < |xn | < 3|x|/2, for all n ≥ N1 .
b) Since xn → x, for |x|2 /2 > 0, there exists N2 ≥ 1 such that |xn − x| < |x2 /2|, for all
n ≥ N2 .
Hence
|xn − x| |xn − x|
|1/xn − 1/x| = ≤ < ,
|x||xn | |x|||x|/2|
for all n ≥ N = max{N1 , N2 }. 
Remark 3.16. If xn → x and yn → y with yn , y 6= 0, then
xn /yn → x/y.
Since y 6= 0, 1/yn → 1/y. Note xn /yn = xn . (1/yn ) → x. (1/y) = x/y.
16 NIKITA AGARWAL

4. Week 4
Vector space. Let C be the set of all convergent sequences of real numbers. It forms a
vector space under the operations:
(xn ) + (yn ) := (xn + yn )
a.(xn ) := (axn ).
The map T : C → R defined as
T ((xn )) := lim xn ,
n

is a linear transformation (prove this!).

Definition 4.1. A sequence (xn ) of real numbers diverges to plus infinity if for every
real number M > 0, there exists N ∈ N such that xn > M , for all n ≥ N .
Definition 4.2. A sequence (xn ) of real numbers diverges to minus infinity if for every
real number M > 0, there exists N ∈ N such that xn < −M , for all n ≥ N .
Example 4.3. • xn = n. (diverges to infinity)
• x2n = 1, x2n+1 = n. (does not diverge to infinity)
• (xn ) diverges to plus infinity if and only if (−xn ) diverges to minus infinity.
More about divergent sequences. Operations on divergent sequences do not follow
same rules as those for convergent sequences. Let us look at a few situations:
• If (xn ) is a divergent sequence, then (−xn ) is also a divergent sequence. But their
sum is the constant 0 sequence which converges.
• If (xn ) is a divergent sequence, then its multiplication by 0 is the constant 0
sequence which converges.
• If (xn ) is a divergent sequence, then (x2n ) need not be divergent. For example, if
xn = (−1)n , (x2n ) is constant sequence 1 which converges.
Operations on sequences diverging to infinity.
Theorem 4.4. 1) Let (xn ), (yn ) be sequences of real numbers that diverge to infinity, then
(xn + yn ) and (xn yn ) both diverge to infinity.
2) If (xn ) diverges to infinity and (yn ) is bounded, then (xn + yn ) diverges to infinity.
3) If (xn ) diverges to infinity and (yn ) converges, then (xn + yn ) diverges to infinity.
Proofs left as an exercise.
Limit superior. Let (xn ) be a sequence of real numbers that is bounded above. For
each n ∈ N, let
Mn = sup{xk : k ≥ n} = sup{xn , xn+1 , . . . }.
Each Mn exists since the sequence (xn ) is bounded above. (lub property)
Note that Mn+1 ≤ Mn , for all n ∈ N.
Hence (Mn ) is a non-increasing sequence of real numbers.
Theorem 4.5. The sequence (Mn ) either diverges to −∞ or converges to inf{Mn : n ∈
N}.
Proof. Recall Mn = sup{xk : k ≥ n} = sup{xn , xn+1 , . . . }.
If (Mn ) is bounded below and m = inf{Mn : n ∈ N} ∈ R, then given  > 0, there exists
MN such that m ≤ MN < m + .
Since Mn is non-increasing and m is the glb, m ≤ Mn ≤ MN < m + , for all n ≥ N .
Therefore Mn → m.
 LECTURE NOTES ON REAL ANALYSIS I 17

If (Mn ) is not bounded below, given M > 0, there exists MN < −M .

Since (Mn ) is non-increasing, Mn ≤ MN < −M , for all n ≥ N .
Hence (Mn ) diverges to −∞. 
Definition 4.6. Let (xn ) be a sequence of real numbers that is bounded above and let
Mn = sup{xn , xn+1 , . . . }.
a) If (Mn ) converges, we define
lim sup xn = lim Mn = inf{Mn : n ∈ N}.
n→∞ n→∞

b) If (Mn ) diverges to −∞, we define

lim sup xn = −∞.
n→∞

Definition 4.7. Let (xn ) be a sequence of real numbers that is not bounded above, we
define lim supn→∞ xn = ∞.
Examples.
(1) (xn ) = (1, 2, 1, 2, 1, 2, . . . ). Here Mn = sup{xk : k ≥ n} = 2. Hence
lim sup xn = lim Mn = 2.
n→∞ n→∞

(2) xn = (−1)n . Here Mn = sup{xk : k ≥ n} = 1. Hence

lim sup xn = lim Mn = 1.
n→∞ n→∞

(3) (xn ) = (1, 1, 1, 2, 1, 3, . . . , 1, n, . . . ). Since (xn ) is not bounded above,

lim sup xn = ∞.
n→∞

Limit inferior or liminf. Let (xn ) be a sequence of real numbers that is bounded below.
For each n ∈ N, let
mn = inf{xk : k ≥ n} = inf{xn , xn+1 , . . . }.
Each mn exists since the sequence (xn ) is bounded below. (glb property)
Note that mn+1 ≥ mn , for all n ∈ N.
Hence (mn ) is a non-decreasing sequence of real numbers.
Theorem 4.8. The sequence (mn ) either diverges to ∞ or converges to sup{mn : n ∈
N}.
Proof is left as an exercise (similar to the proof of Theorem 4.5).
Definition 4.9. Let (xn ) be a sequence of real numbers that is bounded below and let
mn = inf{xn , xn+1 , . . . }.
a) If (mn ) converges, we define
lim inf xn = lim mn = sup{mn : n ∈ N}.
n→∞ n→∞

b) If (mn ) diverges to ∞, we define

lim inf xn = ∞.
n→∞

Definition 4.10. Let (xn ) be a sequence of real numbers that is not bounded below, we
define lim inf n→∞ xn = −∞.
18 NIKITA AGARWAL

Examples.
(1) (xn ) = (1, 2, 1, 2, 1, 2, . . . ). Here mn = inf{xk : k ≥ n} = 1. Hence
lim inf xn = lim mn = 1.
n→∞ n→∞
n
(2) xn = (−1) . Here mn = inf{xk : k ≥ n} = −1. Hence
lim inf xn = lim mn = −1.
n→∞ n→∞

(3) (xn ) = (1, −1, 1, −2, 1, −3, . . . , 1, −n, . . . ). Since (xn ) is not bounded below,
lim inf xn = −∞.
n→∞

(4) (xn ) = (1, 1, 1, 2, 1, 3, . . . , 1, n, . . . ). Here mn = inf{xk : k ≥ n} = 1. Hence

lim inf xn = lim mn = 1.
n→∞ n→∞

A few observations.
• If (xn ) is a non-increasing sequence of real numbers, then Mn = sup{xk : k ≥
n} = xn , for all n ∈ N. Hence
lim sup xn = lim Mn = lim xn .
n→∞ n→∞ n→∞

• If (xn ) is a non-decreasing sequence of real numbers, then mn = inf{xk : k ≥

n} = xn , for all n ∈ N. Hence
lim inf xn = lim mn = lim xn .
n→∞ n→∞ n→∞

• Changing a few terms of the sequence (xn ) does not change liminf and limsup.
This is because the tail of the sequence (xn ) remains unaltered, hence the tail of
(mn ) and (Mn ) remains unaltered.
Lemma 4.11. Let xn → x and yn → y. Let xn ≤ yn for all n ∈ N. Then x ≤ y.
Proof. On the contrary, suppose x > y.
Let  = (x − y)/2.
Since xn → x, there exists N1 ≥ 1 such that x −  < xn < x + , for all n ≥ N1 .
Since yn → y, there exists N2 ≥ 1 such that y −  < yn < y + , for all n ≥ N2 .
Note y +  = x −  = (x + y)/2.
Let N = max{N1 , N2 }. Then
yN < (x + y)/2 < xN ,
which is a contradiction. Hence x ≤ y. 
Remark 4.12. If xn < yn , for all n ∈ N in the preceding theorem, then x = y is possible.
For instance, take xn = −1/n and yn = 1/n, for all n ∈ N.
Theorem 4.13. Let (xn ) be a convergent sequence in R. Then
lim sup xn = lim inf xn = lim xn .
n→∞ n→∞ n→∞

Proof. Let xn → x. Let  > 0 be given. Then there exists N ∈ N such that |xn −x| < /2,
for all n ≥ N . That is, for all n ≥ N , x − /2 < xn < x + /2.
Thus x+/2 is an upper bound and x−/2 is not an upper bound for the set {xk : k ≥ n}.
Hence for all n ≥ N ,
x −  < x − /2 ≤ mn ≤ Mn ≤ x + /2 < x + .
Hence Mn → x and mn → x. 
 LECTURE NOTES ON REAL ANALYSIS I 19

Theorem 4.14. (Converse) Let (xn ) be a bounded sequence in R with

lim sup xn = lim inf xn = x.
n→∞ n→∞

Then xn → x.
Proof. Let  > 0 be given.
Since lim supn→∞ xn = x, there exists N1 ∈ N such that x −  < Mn < x + , for all
n ≥ N1 .
Similarly since lim inf n→∞ xn = x, there exists N2 ∈ N such that x −  < mn < x + , for
all n ≥ N2 .
Let N = max{N1 , N2 }.
Then for n ≥ N ,
x −  < mN ≤ xn ≤ MN < x + .
Hence xn → x. 
Theorem 4.15. (Relationship between limsup and liminf ) Let (xn ) be a sequence of real
numbers, then
lim inf xn ≤ lim sup xn .
n→∞ n→∞

Proof. a) If (xn ) is bounded: Note that for all n ∈ N, mn ≤ Mn . Taking limits on both
sides, we have the required result (use Lemma 4.11).
b) If (xn ) is bounded above but not below: The result follows immediately since liminf
equals −∞.
c) If (xn ) is bounded below but not above: The result follows immediately since limsup
equals ∞. 
Results discussed without proof in the class.
Theorem 4.16. Let (xn ) be a sequence of real numbers which is bounded above. Then
lim sup xn = − lim inf (−xn ).
n→∞ n→∞

Theorem 4.17. Let (xn ) be a sequence of real numbers with

lim sup xn = lim inf xn = ∞,
n→∞ n→∞

then (xn ) diverges to infinity.

Theorem 4.18. If (xn ), (yn ) are bounded sequences of real numbers and xn ≤ yn , for all
n. Then
lim sup xn ≤ lim sup yn , lim inf xn ≤ lim inf yn .
n→∞ n→∞ n→∞ n→∞

Theorem 4.19. If (xn ), (yn ) are bounded sequences of real numbers and xn ≤ yn , for all
n. Then
lim sup(xn + yn ) ≤ lim sup xn + lim sup yn , lim inf (xn + yn ) ≥ lim inf xn + lim inf yn .
n→∞ n→∞ n→∞ n→∞ n→∞ n→∞

Theorem 4.20. Let (xn ) be a bounded sequence of real numbers.

1) If lim supn→∞ xn = M , then for given  > 0,
a) There exists N ∈ N such that xn < M + , for all n ≥ N .
b) xn > M − , for infinitely many values of n.

2) If lim inf n→∞ xn = m, then for given  > 0,

a) There exists N ∈ N such that xn > m − , for all n ≥ N .
b) xn < m + , for infinitely many values of n.
20 NIKITA AGARWAL

Proof. 1) a) Let  > 0 be given. Then there exists N ≥ 1 such that M − < Mn < M +,
for all n ≥ N . Hence xn ≤ MN < M + , for all n ≥ N .

1 b) Suppose there exists  > 0 such that xn > M −  for only finitely many values
of n.
Then there exists N ∈ N such that xn ≤ M − , for all n ≥ N .
Hence Mn ≤ M − , for all n ≥ N . Using Lemma 4.11, we have M ≤ M − , which is a
contradiction.

2) a) Let  > 0 be given. Then there exists N ≥ 1 such that m −  < mn < m + , for
all n ≥ N . Hence m −  < mN ≤ xn , for all n ≥ N .

2 b) Suppose there exists  > 0 such that xn < m +  for only finitely many values
of n.
Then there exists N ∈ N such that xn ≥ m + , for all n ≥ N .
Hence mn ≥ m + , for all n ≥ N . Using Lemma 4.11, we have m ≥ m + , which is a
contradiction. 
Definition 4.21. Let (xn ) be a sequence of real numbers. Let (nk )k≥1 be an increasing
sequence of natural numbers. That is, n1 < n2 < n3 < . . . . Then (xnk )k≥1 is called a
subsequence of the sequence (xn ).
Note that limk→∞ nk = ∞.
exam: Constant sequences (1) and (−1) are both subsequences of ((−1)n ).
Theorem 4.22. Any bounded sequence of real numbers has a convergent subsequence.
Proof. We will construct a subsequence of (xn ) which converges to lim sup xn . Using sim-
ilar arguments, one can construct a subsequence of (xn ) which converges to lim inf xn .

Construction of a subsequence of (xn ) which converges to lim sup xn : Let (xn )

be a bounded sequence of real numbers with M = lim sup xn .
There are infinitely many values of n for which xn > M − 1. Choose n1 ≥ 1 such that
xn1 > M − 1.
There are infinitely many values of n for which xn > M − 1/2. Choose n2 > n1 such that
xn2 > M − 1/2.
Proceeding in this manner, for each k ≥ 1, there are infinitely many values of n for which
xn > M − 1/k. Choose nk > nk−1 such that xnk > M − 1/k.
Let  > 0 be given. There exists N ∈ N such that xn < M + , for all n ≥ N .
Choose K ∈ N such that 1/K <  and nK > N .
Then M −  < M − 1/K < M − 1/k < xnk < M + , for all k ≥ K.
Hence M −  < xnk < M + , for all k ≥ K.
The subsequence (xnk )k≥1 converges to M .

Construction of a subsequence of (xn ) which converges to lim inf xn : Let (xn ) be

a bounded sequence of real numbers with m = lim inf xn .
There are infinitely many values of n for which xn < m + 1. Choose n1 ≥ 1 such that
xn1 < m + 1.
There are infinitely many values of n for which xn < m + 1/2. Choose n2 > n1 such that
xn2 < m + 1/2.
Proceeding in this manner, for each k ≥ 1, there are infinitely many values of n for which
xn < m + 1/k. Choose nk > nk−1 such that xnk < m + 1/k.
 LECTURE NOTES ON REAL ANALYSIS I 21

Let  > 0 be given. There exists N ∈ N such that xn > m − , for all n ≥ N .
Choose K ∈ N such that 1/K <  and nK > N .
Then m −  < xnk < m + 1/k ≤ m + 1/K < m + , for all k ≥ K.
Hence m −  < xnk < m + , for all k ≥ K.
The subsequence (xnk )k≥1 converges to m. 
Definition 4.23. A sequence (xn ) in R is said to be a Cauchy sequence if given  > 0,
there exists N ∈ N such that |xn − xm | < , for all n, m ≥ N .
Example 4.24. We will prove that xn = n/(n + 1) is a Cauchy sequence. Consider
n m n−m
|xn − xm | = − = .
n+1 m+1 (n + 1)(m + 1)
Without loss of generality, assume that m ≤ n, then
n−m n−m n 1 1
= ≤ ≤ < .
(n + 1)(m + 1) (n + 1)(m + 1) (n + 1)(m + 1) m+1 m
Let  > 0 be given. Choose N ≥ 1 such that 1/N < . Then for all n ≥ m ≥ N ,
1 1
|xn − xm | < ≤ < .
m N
Theorem 4.25. A Cauchy sequence of real numbers is bounded.
Proof. For  = 1, there exists N ∈ N such that |xn − xm | < 1, for all n, m ≥ N .
In particular, setting m = N , |xn − xN | < 1, for all n ≥ N .
Hence |xn | ≤ |xn − xN | + |xN | < 1 + |xN |, for all n ≥ N .
Let C = max{|x1 |, . . . , |xN −1 |} ≥ 0 and let M = max{C, 1 + |xN |} > 0.
Then |xn | ≤ M , for all n ∈ N. 
Theorem 4.26. A convergent sequence of real numbers is Cauchy.
Proof. Let xn → x.
We will use |xn − xm | ≤ |xn − x| + |xm − x|, for all n, m ∈ N.
Given  > 0, there exists N ∈ N such that |xn − x| < /2.
Hence for n, m ≥ N , |xn − xm | ≤ |xn − x| + |xm − x| < . 
Lemma 4.27. If for any given  > 0, x ≤ y + , then x ≤ y.
Proof. Suppose x > y. Choose  = (x − y)/2 > 0. Then y +  = x −  < x, which is a
contradiction. 
Theorem 4.28. A Cauchy sequence of real numbers is convergent. This property of real
numbers is called Cauchy completeness of R or just completeness.
Proof. Let  > 0 be given. Since (xn ) is Cauchy, it is bounded. Hence it is enough to
prove that M = lim supn→∞ xn = lim inf n→∞ xn = m.
Since m ≤ M (Theorem 4.15), it is enough to show that m ≥ M .
Since (xn ) is Cauchy, there exists N ∈ N such that |xn − xm | < /2, for all n, m ≥ N .
Thus xN + /2 and xN − /2 are ub and lb, respectively, of {xk : k ≥ n}, for all n ≥ N .
Therefore xN − /2 ≤ mn ≤ Mn ≤ xN + /2, for all n ≥ N .
Hence Mn − mn ≤  ⇒ Mn ≤ mn + , for all n ≥ N .
Taking limits on both sides, we get M ≤ m + .
Since  > 0 is arbitrary, we obtain M ≤ m. (Lemma 4.27) 
Example 4.29. Let X = {1/n : n ≥ 1} and sequence (xn ) = (1/n). Since (xn ) is
convergent in R, it is Cauchy in R, and hence Cauchy in X. However it is not convergent
in X since 0 ∈
/ X.
22 NIKITA AGARWAL

Example 4.30. (Cauchy sequence of rational √ numbers) We know that there is a sequence
of rational numbers (xn ) that converges to 2.
Since (xn ) is convergent in R, it is Cauchy in R (also in Q.)
But (xn ) is not convergent in Q.
This gives a definition of the set of real numbers. We will get back to it later.
Theorem 4.31. (Well-ordering principle) Every nonempty subset of natural numbers
has a least element.
Proof. Suppose there is no least element.
Then 1 ∈/ S, otherwise 1 is the least element.
Therefore 2 ∈/ S, otherwise it is the least element.
In a similar fashion, if 1, . . . , n ∈
/ S and n + 1 ∈ S, then n + 1 is the least element.
Hence S is an empty set. Thus we have arrived at a contradiction. 
Theorem 4.32. (Enumeration) An infinite subset S of natural numbers can be written
as {n1 < n2 < · · · < nk < nk+1 < . . . }.
Proof. Since S is a non-empty subset of natural numbers, it has a least element, n1 say
(by the well-ordering principle).
Consider the set S1 = S \ {n1 }, which is a non-empty subset of natural numbers, it has
a least element, n2 say. Note n1 < n2 .
In this fashion, suppose we have chosen n1 , . . . , nk ∈ S such that n1 < n2 < · · · < nk .
The set Sk = S \ {n1 , . . . , nk } is non-empty since S is an infinite set. Hence we can chose
nk+1 , the least element of Sk . Note nk < nk+1 .
Note that there is no element in S which lies between nk and nk+1 , otherwise that is the
least element of Sk .
Thus we have recursively defined an increasing sequence of natural numbers (nk )k such
that no element in S which lies between nk and nk+1 .
Claim 1: k ≤ nk .
1 ≤ n1 < n2 . Hence 2 ≤ n2 . Suppose k ≤ nk . Then since k ≤ nk < nk+1 , we get
k + 1 ≤ nk+1 .
Claim 2: This process exhausts all the elements in S.
Suppose not, and let m be the least element of S \ {nk : k ≥ 1}.
Since m ∈ S and m ≤ nm . Since m 6= nm , we have m < nm , hence m must equal some
nk for 1 ≤ k < m, which is a contradiction. 
Theorem 4.33. Let xn → x, then any subsequence (xnk ) converges to x.
Proof. Let  > 0 be given.
There exists N ∈ N such that |xn − x| < , for all n ≥ N .
Then for k ≥ N , since nk ≥ k, we have nk ≥ N .
Therefore for all k ≥ N , |xnk − x| < . 
Theorem 4.34. Every sequence of real numbers has a monotone subsequence (non-
decreasing or non-increasing).
Proof. Let (xn ) be a sequence of real numbers.
Consider the set
S = {n ∈ N : xm < xn for all m > n}.
There are two cases: S is finite or infinite.

a) S finite: (hence bounded)

There exist N ∈ N such that s ≤ N , for all s ∈ S. Let n1 > N (Archimedian property),
 LECTURE NOTES ON REAL ANALYSIS I 23

then n1 ∈/ S. Hence there exists n2 > n1 such that xn2 ≥ xn1 . Since n2 > n1 > N ,
there exists n3 > n2 such that xn3 ≥ xn2 . This way we have constructed a non-decreasing
subsequence (xnk ).

b) S infinite: S can be written as {n1 < n2 < · · · < nk < nk+1 < . . . }.
By the definition of S, xnk+1 < xnk , for all k ≥ 1.
Hence we have a decreasing subsequence (xnk ). 
Theorem 4.35. A non-decreasing sequence is either convergent or diverges to +∞.
Proof. Let (xn ) be a non-decreasing sequence, that is, xn ≤ xn+1 , for all n ≥ 1.
The sequence is bounded below as well, by x1 .
For each n ≥ 1, mn = inf{xk : k ≥ n} = xn .
Since lim inf xn = lim mn = sup{mn : n ≥ 1} = sup{xn : n ≥ 1}, we have
lim xn = sup{xn : n ≥ 1}.
which is either finite (if (xn ) is bounded), or diverges to +∞. 
Theorem 4.36. A non-increasing sequence is either convergent or diverges to −∞.
Theorem 4.37. (Bolzano-Weierstrass Theorem) A bounded sequence of real numbers has
a convergent subsequence.
We proved this earlier. Now we give another proof.
Proof. Let (xn ) be a bounded sequence.
It has a monotone subsequence (xnk )k≥1 .
Since (xn ) is bounded, (xnk ) is also bounded.
Hence (xnk ) is a bounded monotone sequence.
Therefore it converges. 
24 NIKITA AGARWAL

5. Week 5
Definition 5.1. Let E be a subset of R. Let f : E → R be a function. Let a ∈ E and
` ∈ R. We say that f (x) → ` as x → a if given  > 0, there exists δ > 0 such that
|f (x) − `| < , whenever |x − a| < δ, x ∈ E.
Definition 5.2. Let E be a subset of R. Let f : E → R be a function. We say that f is
continuous at a point a ∈ E if f (x) → f (a) as x → a. That is, given  > 0, there exists
δ > 0 such that
|f (x) − f (a)| < , whenever |x − a| < δ, x ∈ E.
If f is continuous at each point of E, we say that f is continuous on E.
Definition 5.3. Let E be a subset of R. Let f : E → R be a function. We say that f is
uniformly continuous on E if for given  > 0, there exists δ > 0 such that
|f (x) − f (y)| < , whenever |x − y| < δ, x ∈ E.
Uniformly continuous implies continuous.
Example 5.4. (1) Let f : [0, 1] → R be defined as f (x) = x2 , for all x ∈ [0, 1]. Then
|f (x) − f (y)| = |x2 − y 2 | = (x + y)|x − y| ≤ 2|x − y|.
For given  > 0, choose δ = /2 > 0. Thus f is uniformly continuous.
In fact, any continuous function on a closed interval [a, b] is uniformly continuous.
(2) In the previous example, if the domain of f is R, instead of [0, 1], then f is
continuous on R but not uniformly continuous.
a) f continuous: Let a ∈ R. Suppose |x − a| < 1, then
|x + a| ≤ |x − a| + 2|a| < 1 + 2|a|.

Let  > 0 be given. Choose δ < min{1, 1+2|a| }, then
|f (x) − f (a)| = |x2 − a2 | = (x + a)|x − a| ≤ (1 + 2|a|)|x − a| < ,
whenever |x − a| < δ.
b) f not uniformly continuous: For  = 1, there exists δ > 0 such that
|f (x) − f (y)| < 1,
whenever |x − y| < δ. Choose m = 1/δ, x = m − δ/4, y = m + δ/4, then
1 > |x − y||x + y| = 1,
which is a contradiction.
(3) Let f : (0, 1] → R be defined as f (x) = 1/x, for all x ∈ (0, 1]. Let a ∈ (0, 1].
a) f is continuous: Suppose |x − a| < a/2, then x > a/2. For given  > 0, choose
δ = min{a/2, a2 /2}. Then
|f (x) − f (a)| = |1/x − 1/a| = |x − a|/(xa) < 2|x − a|/a2 < ,
whenever |x − a| < δ, x ∈ (0, 1].
b) f is not uniformly continuous: For  = 1, let δ > 0 be such that
|f (x) − f (y)| < 1,
whenever |x − y| < δ, x, y ∈ (0, 1].
Choose n ≥ 1 large enough so that x = 1/n, y = 1/n + δ/(2n) ∈ (0, 1], then
|x − y| = δ/(2n) < δ. Hence
2n nδ
1 > |f (x) − f (y)| = |n − |= .
2+δ 2+δ
 LECTURE NOTES ON REAL ANALYSIS I 25

2+δ
Further choose n ≥ δ
to arrive at a contradiction.
Definition 5.5. Let E be an interval in R. Let f : E → R be a function. We say that
f is differentiable at a point a ∈ E if
f (x) − f (a)
lim
x→a,x∈E x−a
exists. If the limit exists, it is known as the derivative of the function f at a, denoted as
f 0 (a). If f is differentiable at each point of E, we say that f is differentiable on E. In
this case, the function f 0 : E → R is known as the derivative of the function f on E.
Definition 5.6. Let f : [a, b] → R be a function. Let P = {a = x0 < x1 < · · · < xn−1 <
xn = b} be a partition of [a, b]. We define the mesh size of P as
kP k = max (xk − xk−1 ).
k=1,2,...,n

We define the upper sum of f with respect to P as

n
X
U (P, f ) = (xk − xk−1 ) sup f (x),
k=1 x∈[xk−1 ,xk ]

and the lower sum of f with respect to P as

n
X
L(P, f ) = (xk − xk−1 ) inf f (x).
x∈[xk−1 ,xk ]
k=1

We say that f is Riemann-integrable on [a, b] if for given  > 0, there exists δ > 0 such
that for each partition P of [a, b] with kP k < δ, we have
|U (P, f ) − L(P, f )| < .
Remark 5.7. The above definition is equivalent to the following:
We say that f is Riemann-integrable on [a, b] if for given  > 0, there exists a partition
P of [a, b] such that
|U (P, f ) − L(P, f )| < .
Sequence of functions. Let E be a non-empty subset of R. For each n ≥ 1, let
fn : E → R be a function. We will study the sequence (fn )∞
n=1 of functions.
Definition 5.8. (Pointwise convergence) We say that the sequence (fn ) converges to a
function f : E → R pointwise on E if for all x ∈ E,
lim fn (x) = f (x).
n→∞
pw
We will often denote it as fn −→ f .
Remark 5.9. 1) Let x ∈ E. Given  > 0, there exists N ∈ N such that for all n ≥ N ,
|fn (x) − f (x)| < .
The natural number N above depends on both x and .
It is quite possible that there is no common N that works for all x ∈ E.
pw pw
2) fn −→ f iff fn − f −→ 0.
Example 5.10. Consider fn (x) = xn on [0, 1]. Note that
lim fn (1) = lim 1 = 1,
n→∞ n→∞
n
lim fn (x) = lim x = 0, for x ∈ [0, 1).
n→∞ n→∞
Define f (x) = 0, x ∈ [0, 1) and f (1) = 1.
pw
Then fn −→ f .
26 NIKITA AGARWAL

Example 5.11. Consider fn (x) = χ(0,1/n) (x) on (0, 1) (characteristic function).

Note that
lim fn (x) = lim χ(0,1/n) (x).
n→∞ n→∞

Let x ∈ (0, 1). By the Archimedian property, there exists N ∈ N such that 1/N < x.
Hence χ(0,1/n) (x) = 0, for all n ≥ N .
pw
Thus fn −→ 0 (the zero function).

Example 5.12. Consider fn (x) = χ(0,n) (x) on (0, ∞) (characteristic function).

Note that
lim fn (x) = lim χ(0,n) (x).
n→∞ n→∞

Let x ∈ (0, ∞). By the Archimedian property, there exists N ∈ N such that N > x.
Hence χ(0,n) (x) = 1, for all n ≥ N .
pw
Thus fn −→ 1 (the constant function 1).

Definition 5.13. (Uniform convergence) We say that the sequence (fn ) converges to a
function f : E → R uniformly on E if for given  > 0, there exists N ∈ N such that

|fn (x) − f (x)| < ,

for all n ≥ N and for each x ∈ E.

u
We will often denote it as fn →
− f.

Remark 5.14. 1) In the above definition, the natural number N depends only on  and
NOT x (as in pointwise convergence).
u pw
2) If fn →
− f then fn −→ f .
u u
3) fn →− f iff fn − f →
− 0.

Example 5.15. Consider fn (x) = xn on [0, 1].

pw
Recall fn −→ f , where f (x) = 0, x ∈ [0, 1) and f (1) = 1.
u
Suppose fn →− f . Let  = 1/2 be given. Then there exists N ∈ N such that

|fn (x) − f (x)| < 1/2,

for all n ≥ N and for each x ∈ E.

Let x ∈ [0, 1), then xN < 1/2.
Taking limx→1 on both sides, we obtain 1 ≤ 1/2, which is absurd. Hence fn does not
uniformly converge to f .
This provides an example of a sequence of functions which converge pointwise but not
uniformly.
sin nx
Example 5.16. Consider fn (x) = , for x ∈ [0, 1].
n
Note that for all n ≥ 1, and x ∈ [0, 1],
1
|fn (x)| ≤ ,
n
Let  > 0 be given. Choose N ∈ N such that 1/N < .
Then |fn (x)| ≤ 1/n ≤ 1/N < , for all n ≥ N .
u
Hence fn →− 0 (the zero function).
 LECTURE NOTES ON REAL ANALYSIS I 27

Not converging uniformly. Observe that fn does not uniformly converge to f if and
only if there exists  > 0 such that for all N ∈ N,
|fn (x) − f (x)| ≥ ,
for some n ≥ N and for some x ∈ E.
Theorem 5.17. (An alternate definition of uniform convergence) The sequence (fn )
converges uniformly to a function f on E if and only if
lim sup |fn (x) − f (x)| = 0.
n→∞ x∈E

u
Proof. Let fn →
− f . Then for given  > 0, there exist N ∈ N such that
|fn (x) − f (x)| < /2,
for all n ≥ N , for each x ∈ E.
Hence
sup |fn (x) − f (x)| ≤ /2 < , for all n ≥ N.
x∈E
Therefore
lim sup |fn (x) − f (x)| = 0.
n→∞ x∈E

Converse: Let limn→∞ supx∈E |fn (x) − f (x)| = 0. Let  > 0 be given. Then there exists
N ≥ 1 such that for all n ≥ N ,
sup |fn (x) − f (x)| < .
x∈E

Hence
|fn (x) − f (x)| ≤ sup |fn (x) − f (x)| < ,
x∈E
u
for all x ∈ E, for all n ≥ N . Hence fn →
− f. 
Definition 5.18. We say that (fn ) is uniformly Cauchy on E if and only if for given
 > 0, there exists N ∈ N such that
|fn (x) − fm (x)| < ,
for all m, n ≥ N , for each x ∈ E.
Theorem 5.19. The sequence (fn ) converges uniformly to the function f on E if and
only if (fn ) is uniformly Cauchy on E.
u
Proof. (⇒) Let  > 0 be given. Since fn →
− f , there exists N ∈ N such that
|fn (x) − f (x)| < /2,
for all n ≥ N and for each x ∈ E.
Thus for all n, m ≥ N and for each x ∈ E,
|fn (x) − fm (x)| ≤ |fn (x) − f (x)| + |fm (x) − f (x)| < /2 + /2 = .
(⇐) Let  > 0 be given. There exists N ∈ N such that |fn (x) − fm (x)| < /2, for all
m, n ≥ N and for each x ∈ E.
Let x ∈ E.
By the Cauchy completeness of R, for each x ∈ E, since (fn (x))n is a Cauchy sequence,
it converges. Let us denote the limit as `x ∈ R.
Now define f (x) := `x , for all x ∈ E. This is nothing but the pointwise limit of (fn ).
u
We prove that fn → − f on E.
Since |fn (x) − fm (x)| < /2, for all m, n ≥ N and for each x ∈ E, fix x ∈ E, n ≥ N , and
28 NIKITA AGARWAL

let limm→∞ on both sides in |fn (x) − fm (x)| < /2, we obtain |fn (x) − f (x)| ≤ /2 < .
This is true for all n ≥ N and for each x ∈ E.
u
Hence fn →− f on E. 
u
Theorem 5.20. Let fn →
− f on E. For all n ≥ 1, let fn be bounded on E. Then f is
bounded on E.
u
Proof. Since fn →
− f , there exists N ∈ N such that for all n ≥ N and for each x ∈ E,
|fn (x) − f (x)| < 1.
Further since fN is bounded on E, there exists M > 0 such that for all x ∈ E,
|fN (x)| < M.
Hence for each x ∈ E,
|f (x)| ≤ |f (x) − fN (x)| + |fN (x)| < 1 + M.

1
Example 5.21. Let fn (x) = χ[1/n,1] , x ∈ (0, 1]. Since |fn (x)| ≤ n, for each n ≥ 1,
x
x ∈ (0, 1], (fn ) is a sequence of bounded functions.
Let f (x) = 1/x, x ∈ (0, 1]. Note that f is not a bounded function.
Let x ∈ (0, 1], then there exists N ≥ 1 such that 1/N < x, hence 1/n ≤ 1/N < x, for all
n ≥ N.
Therefore fn (x) = 1/x, for all n ≥ N . Thus limn→∞ fn (x) = 1/x, for each x ∈ (0, 1].
pw
Hence fn −→ f on (0, 1].
However the limit f cannot be uniform by the preceding result. Let us also prove this
directly. Suppose it is uniform. Then for  = 1, there exists N ≥ 1 such that |fn (x) −
1/x| < 1, for all n ≥ N , for each x ∈ E. In particular, for n = N and x = 1/(2N ),
1 > |fn (x) − 1/x| = 2N , which is a contradiction.
u
Theorem 5.22. Let fn → − f on E. For all n ≥ 1, let fn be continuous at a point a ∈ E.
Then f is continuous at a.
u
Proof. Let  > 0 be given. Since fn →
− f , there exists N ∈ N such that for all n ≥ N and
for each x ∈ E,
|fn (x) − f (x)| < /3.
Further since fN is continuous at a, there exists δ > 0 such that for all x ∈ E with
|x − a| < δ,
|fN (x) − fN (a)| < /3.
Hence for each x ∈ E with |x − a| < δ,
|f (x) − f (a)| ≤ |f (x) − fN (x)| + |fN (x) − fN (a)| + |fN (a) − f (a)| < .

Example 5.23. Let fn (x) = xn , x ∈ [0, 1].
pw
Then fn −→ f , where f (x) = 0, x ∈ [0, 1), f (1) = 1. But f is not a uniform limit.
Each fn is continuous on [0, 1] but f is not continuous at 1.
u
Theorem 5.24. Let E = [a, b], fn : E → R be Riemann integrable on E. Let fn →
− f on
Rb Rb
E. Then f is Riemann integrable on E and a f (x) dx = limn→∞ a fn (x) dx.
 LECTURE NOTES ON REAL ANALYSIS I 29

u
Proof. Let  > 0 be given. Since fn →
− f , there exists N ∈ N such that for all n ≥ N ,
(Theorem 8, Lecture 11)
  
sup |fn (x) − f (x)| < ⇒ fN (x) − < f (x) < fN (x) + , ∀ x ∈ E.
x∈E 4(b − a) 4(b − a) 4(b − a)
Since fN is R.I., there exists a partition P = {a = x0 < x1 < · · · < xn = b} of E such
that

U (P, fN ) − L(P, fN ) < .
2
Let Mj (g) = supx∈[xj−1 ,xj ] g(x) for any function g and j = 1, . . . , n. Then for each j,
Mj (f ) ≤ Mj (fN ) + /4(b − a).
Multiplying both sides by (xj − xj−1 ) and sum over j, we get U (P, f ) ≤ U (P, fN ) + /4.
Similarly L(P, f ) ≥ L(P, fN ) − /4.

We have U (P, fN )−L(P, fN ) < , U (P, f ) ≤ U (P, fN )+/4 and L(P, f ) ≥ L(P, fN )−/4.
2
Combining these three inequalities we obtain U (P, f )−L(P, f ) < . Hence f is integrable
on E.
Note
Z b Z b Z b
fn (x) dx − f (x) dx ≤ |fn (x) − f (x)| dx ≤ sup |fn (x) − f (x)|(b − a) < ,
a a a x∈E

for all n ≥ N . 
u
Corollary 5.24.1. Let E = [a, b], fn : E → R be continuous. Let fn →
− f on E. Then f
Rb Rb
is continuous hence Riemann integrable on E and a f (x) dx = limn→∞ a fn (x) dx.
Proof follows from Theorem 5.24.
Example 5.25. Let (rn ) be an enumeration of Q ∩ [0, 1].
Let fn (x) = χ{r1 ,...,rn } (x), x ∈ [0, 1].
fn is continuous at all points except r1 , . . . , rn , hence is Riemann integrable on [0, 1].
pw
fn −→ f on [0, 1] with f = χQ (Dirichlet’s function).
The limit function f is not Riemann integrable (it is nowhere continuous on [0, 1]).
Theorem 5.26. (Dini’s Theorem) Let E = [a, b], fn , f : E → R be continuous. Let
pw u
fn −→ f on E. Assume that fn (x) ≤ fn+1 (x), for all n ≥ 1 and x ∈ E. Then fn →
− f on
E.
Proof. Note that f (x) = supn≥1 fn (x), for all x ∈ E.
Define gn = f − fn , then for each x ∈ E,
g1 (x) ≥ g2 (x) ≥ · · · ≥ gn (x) ≥ gn+1 (x) ≥ · · · ≥ 0.
pw pw
Since each fn is continuous and fn −→ f , we have that each gn is continuous and gn −→ 0.
u u u
Since fn →
− f if and only if gn →
− 0, we will show that gn →− 0.
pw
Let  > 0 be given. Since gn −→ 0, for each x ∈ E, there exists Nx ∈ N such that
gNx (x) < /2.
Since gNx is continuous at x, there exists δx > 0 s.t. for all y ∈ (x − δx , x + δx ) = Ix ,
gNx (y) < gNx (x) + /2 < .
From the previous steps, we have for all y ∈ (x − δx , x + δx ) = Ix ,
gNx (y) < .
30 NIKITA AGARWAL

The collection (Ix )x∈E is an open cover of E since E = ∪x∈E {x} ⊂ ∪x∈E Ix .
Since E = [a, b] has property (C), there exists finitely many points x1 , . . . , xk such that
E ⊂ ∪ki=1 Ixi .
Let N = max{Nx1 , . . . , Nxk }.
Let y ∈ E, then y ∈ Ixj for some j = 1, . . . , k. Hence
gNxj (y) < .
Since N ≥ Nxj , gN (y) ≤ gNxj (y) < .
Hence 0 ≤ gN (y) < , for each y ∈ E.
Since (gn ) is a decreasing sequence, we get 0 ≤ gn (y) < , for all n ≥ N and for each
y ∈ E.
u
Therefore gn → − 0. 
Corollary 5.26.1. (Not discussed in the class) Let E = [a, b], fn , f : E → R be contin-
pw
uous. Let gn −→ g on E. Assume that gn (x) ≥ gn+1 (x), for all n ≥ 1 and x ∈ E. Then
u
gn →
− g on E.
pw
Proof. Let fn = −gn , for all n ≥ 1, and let f = −g. Then fn −→ f on E. Also
u
fn (x) ≤ fn+1 (x), for all n ≥ 1 and x ∈ E. Applying Dini’s theorem, we get fn →
− f on
u
E. Hence gn → − g on E. 
Example 5.27. Dini’s theorem does not hold true if one of the hypotheses is dropped.
1) Let E = [0, 1], fn (x) = −xn , for all n ≥ 1 and x ∈ E. Then fn ≤ fn+1 , for all n ≥ 1.
The pointwise limit is not continuous. Also fn does not converge uniformly (we saw this
earlier).
2) Let E = (0, 1] (not closed), fn (x) = −1/(nx + 1), for all n ≥ 1 and x ∈ E. Then
pw
fn ≤ fn+1 , for all n ≥ 1. Here fn −→ 0 on E. Since
1
lim sup |fn (x)| = lim sup = lim 1 = 1.
n→∞ x∈E n→∞ x∈E nx + 1 n→∞

Hence fn does not converge uniformly to the zero function on E.

 LECTURE NOTES ON REAL ANALYSIS I 31

6. Week 6
We first completed the proof of Dini’s theorem from the last lecture of Week 5 and
discussed examples.
Theorem 6.1. Let E be a bounded interval of R. Let fn : E → R be a differentiable
function for all n ≥ 1. Let fn0 converges uniformly to a function g on E. Further assume
there exists a point c ∈ E such that the sequence (fn (c))n of real numbers converges. Then
fn uniformly converges to a continuous function f : E → R on E.
Proof. (Recall Mean value theorem) Let h be a differentiable function on (a, b), then for
any x, y in E,
h(x) − h(y) = h0 (t)(x − y),
for some t between x and y.

We first prove that (fn ) is uniformly Cauchy which implies that (fn ) is uniformly conver-
gent.
Let x ∈ E. Since both fn , fm are differentiable, fn − fm is differentiable.
Using the mean value theorem to the function fn − fm at the points x and c, we get
(fn (x) − fm (x)) − (fn (c) − fm (c)) = (fn0 (d) − fm
0
(d))(x − c),
for some d between x and c. Note that d ∈ E depends on x, c, m, n.
Let  > 0 be given.
Since (fn (c))n converges, it is a Cauchy sequence. Hence there exists N1 ≥ 1 such that
for all n, m ≥ N1 ,

|fn (c) − fm (c)| < .
2
0 u 0
Since fn → − g, (fn ) is uniformly Cauchy. Hence there exists N2 ≥ 1 such that for all
n, m ≥ N2 and for each t ∈ E,

|fn0 (t) − fm0
(t)| < .
2(b − a)
Hence for all n ≥ N = max{N1 , N2 } and for each x ∈ E,
 
|fn (x) − fm (x)| ≤ |fn (c) − fm (c)| + |fn0 (d) − fm 0
(d)||x − c| < + (b − a) = .
2 2(b − a)
Hence (fn ) is uniformly Cauchy. Therefore it is uniformly convergent to a function
f : E → R, say.
u
Since each fn is differentiable on E, it is continuous on E. Further since fn → − f , f is
continuous on E. 
Theorem 6.2. Let E be a bounded interval of R. Let fn : E → R be a differentiable
u
function for all n ≥ 1. Let there exist functions f, g : E → R such that fn →
− f and
u
fn0 →
− g on E. Then f is differentiable and f 0 = g on E.
Proof. Fix a ∈ E. For each n ≥ 1, define gn : E → R as

 fn (x) − fn (a)
, x=6 a
gn (x) = x−c
f 0 (a), x = a.
n

Then gn is continuous on E and gn converge pointwise to the function h : E → R given

by 
 f (x) − f (a)
, x 6= a
h(x) = x−a
g(a), x = a.
32 NIKITA AGARWAL

We now prove that (gn ) is uniformly Cauchy, hence uniformly convergent to a continuous
function k : E → R.
By the mean value theorem, for n, m ≥ 1, x ∈ E, x 6= a, there exists t between x and a
such that
(fn (x) − fm (x)) − (fn (a) − fm (a))
gn (x) − gm (x) = = fn0 (t) − fm
0
(t).
x−a
Since (fn0 )n is uniformly convergent, it is uniformly Cauchy.
Thus (gn ) is uniformly Cauchy on E.
u
Hence there exists a function k : E → R such that gn → − k.
Since the uniform limit is same as the pointwise limit, k = h on E.
Further, since each gn is continuous, the limit function h is continuous.
Hence
f (x) − f (a)
g(a) = lim .
x→a x−a
Thus f is differentiable at a and f 0 (a) = g(a). 
Corollary 6.2.1. Let E be a bounded interval of R. Let fn : E → R be a differentiable
function for all n ≥ 1. Let fn0 converges uniformly to a function g on E. Further assume
there exists a point c ∈ E such that the sequence (fn (c))n of real numbers converges. Then
fn uniformly converges to a differentiable function f : E → R on E. Further f 0 = g on
E.
Proof follows from Theorems 6.1 and 6.2.
e−nx
Example 6.3. Let E = [0, 1). For each n ≥ 1, let fn : E → R be defined as fn (x) = n
,
for all x ∈ E. Let  > 0 be given. Then
e−nx 1
fn (x) = ≤ < ,
n n
u
for all x ∈ E, n ≥ 1 . Let f be the constant zero function. Hence fn → − f on E.
0 −nx 0
Here fn (x) = −e , for all x ∈ E. Therefore fn (x) converges pointwise to the function
g : E → R defined as g(x) = 0, for all x ∈ (0, 1), and g(0) = −1.
(fn0 ) does not converge uniformly to g on E since each fn0 is continuous on E but g is not.
Clearly f 0 6= g.
Series of functions. Let E be a non-empty subset of R. For each nP ≥ 1, let fn : E → R
be a function. We will study the function given by the infinite series ∞ Pfn . This series
n=1
is defined as the limit of the sequence (sn ) of partial sums, where sn = nk=1 fk .
Definition 6.4. We say that the series ∞
P
n=1 fn is pointwise convergent on E if the
sequence of partial sums (sn ) is pointwise convergent on E.

We say that the series ∞

P
n=1 fn is uniformly convergent on E if the sequence of partial
sums (sn ) is uniformly convergent on E.
Pn P∞
Let σn (x) = k=1 |fk (x)|, for each n ≥ 1. We say that the series n=1 fn is abso-
lutely convergent on E if the sequence of partial sums (σn (x)) is pointwise convergent
on E.
Theorem 6.5. If ∞
P
n=1 fn is absolutely convergent on E then it is pointwise convergent
on E.
 LECTURE NOTES ON REAL ANALYSIS I 33

Proof. Let x ∈ E.
For all n > m ≥ 1, observe that
n
X m
X n
X n
X
|sn (x)−sm (x)| = | fk (x)− fk (x)| = | fk (x)| ≤ |fk (x)| = σn (x)−σm (x).
k=1 k=1 k=m+1 k=m+1

Let  >
P0∞be given.
Since n=1 fn is absolutely convergent on E, the sequence (σn ) of functions is convergent
pointwise on E. Hence there exists N ≥ 1 such that
σn (x) − σm (x) < ,
for all n > m ≥ N .
Hence
|sn (x) − sm (x)| ≤ σn (x) − σm (x) < ,
for all n > m ≥ N . Thus (sn (x)) is Cauchy, hence it is convergent. Therefore ∞
P
n=1 fn is
convergent at x ∈ E. This proves the result. 
Example 6.6. Let E = [0, 1] and fn (x) = xn , for all n ≥ 1, x ∈ E. Then the sequence
of partial sums is given by
n n
X X x(1 − xn )
sn (x) = fk (x) = xk = , x ∈ [0, 1),
k=1 k=1
1 − x
n
X
sn (1) = fk (1) = n.
k=1
Clearly (sn (x)) converges to f (x) = x/(1 − x), when x ∈ [0, 1), but (sn (1)) does not
converge.
Hence the series ∞
P
n=1 fn converges pointwise on [0, 1) but not at 1.
Is the convergence on [0, 1) uniform?
Consider
x x(1 − xn ) xn+1
|f (x) − sn (x)| = − = ≥ xn+1 ,
1−x 1−x 1−x
the last inequality is true since 1 − x ≤ 1. Hence
lim sup |f (x) − sn (x)| ≥ lim sup xn+1 = lim 1 = 1.
n→∞ x∈[0,1) n→∞ x∈[0,1) n→∞
P
Hence n fn does not uniformly converge to f by Theorem 5.17.
P
Theorem 6.7. (Cauchy criterion for uniform convergence of series) Let n fn be an
infinite series on a set E. It is uniformly convergent on E if and only if for given  > 0,
there exists N ∈ N such that for all n, m ≥ N (m ≤ n), and for all x ∈ E,
n
X
fk (x) < .
k=m+1

Proof. Recall Cauchy

P criterion for uniform convergence of sequence of functions.
The infinite series n fn is uniformly convergent on E iff the sequence (sn ) of functions
is uniformly convergent (sn = nk=1 fk ) iff for given  > 0, there exists N ∈ N such that
P
for all n, m ≥ N (m ≤ n), and for all x ∈ E,
n
X
|sn (x) − sm (x)| <  ⇐⇒ fk (x) < .
k=m+1


34 NIKITA AGARWAL

7. Week 7
Theorem 7.1. (Weierstrass M-test) Let (fn ) be a sequence of real-valued functions on a
of real numbers such that |fn (x)| ≤ Mn ,
set E. Assume that there exists a sequence (Mn ) P
n ≥ 1, for each x ∈ E. Also assume that n Mn is a convergent series. Then the
for all P
series n fn is absolutely and uniformly convergent on E.
P
Proof. We prove that n |fn | satisfies the Cauchy criterion.
Let  >P0 be given.
Since n Mn is a convergent series, there exists N ∈ N such that nk=m+1 Mk < , for
P
all n ≥ m ≥ N .
Hence for all n ≥ m ≥ N , and for each x ∈ E,
n
X n
X
|fk (x)| ≤ Mk < .
k=m+1 k=m+1
P
Hence n fn is absolutely and uniformly convergent on E by Theorem 6.7. 
sin nx
Example 7.2. (An application of Weierstrass M-test) Let p > 1 and fn (x) = ,
np
x ∈ R.
For all n ≥ 1 and x ∈ R,
1
|fn (x)| ≤ p .
n
p
Take M Pn = 1/n in the Weierstrass
P M-test.
Since n Mn P is convergent, n fn is absolutely and uniformly convergent on R.
Further since nk=1 fk is a continuous function on R and the convergence is uniform, the
limit function
X X sin nx
fn (x) =
n n
np
is a continuous function on R.
cos nx
Similar is the case for fn (x) = , x ∈ R.
np
Theorem 7.3. (Dirichlet’s test) Let (fn ), (gn ) be sequences of real-valued functions on a
set E. Let Fn (x) = nk=1 fk (x), for all n ≥ 1. Assume that
P
i) There exists M > 0 such that |Fn (x)| ≤ M , for all n ≥ 1, and for each x ∈ E
(uniformly bounded),
ii) gn+1 (x) ≤ gn (x), for all n ≥ 1, and for each x ∈ E,
u
iii) gn →
− 0 on E. P
Then the series n fn gn is uniformly convergent on E.
Proof. Let sn (x) = nk=1 fk (x)gk (x). By induction on n, it can be shown that
P

n
X
sn (x) = Fk (x) (gk (x) − gk+1 (x)) + gn+1 (x)Fn (x).
k=1

The above formula is known as Abel’s partial summation formula.

For n > m
n
X
sn (x) − sm (x) = Fk (x) (gk (x) − gk+1 (x)) + gn+1 (x)Fn (x) − gm+1 (x)Fm (x).
k=m+1
 LECTURE NOTES ON REAL ANALYSIS I 35

Hence
n
X
|sn (x) − sm (x)| ≤ |Fk (x)| |gk (x) − gk+1 (x)| + |gn+1 (x)||Fn (x)| + |gm+1 (x)||Fm (x)|
k=m+1
Xn
≤ M (gk (x) − gk+1 (x)) + M gn+1 (x) + M gm+1 (x)
k=m+1
= 2M gm+1 (x).
u
Finally since gn →
− 0, for given  > 0, there exists N ∈ N such that gn (x) < /(2M ), for
all n ≥ N , and for each x ∈ E.
Thus, for n > m > N , we have |sn (x) − smP (x)| < , for each x ∈ E.
Hence (sn ) is uniformly Cauchy, therefore n fn gn is uniformly convergent. 
Example 7.4. (Application of Dirichlet’s test) Let (gn ) be a sequence of real-valued
functions on a set E. Assume that
i) gn+1 (x) ≤ gn (x), for all n ≥ 1, and for each x ∈ E, and
u
ii) gn →
− 0 on E.P
Then the series n (−1)n gn is uniformly convergent on E.

Take fn (x) = (−1)n in the Dirichlet’s test. Then |Fn (x)| ≤ 1.

Theorem 7.5. (Abel’s test) Let (fn ), (gn ) be sequences of real-valued functions on a set
E.P Assume that
i) n fn is uniformly convergent on E,
ii) There exists M > 0 such that |gn (x)| ≤ M , for all n ≥ 1, and for each x ∈ E
(uniformly bounded),
iii) (gn ) is monotone
P on E,
Then the series n fn gn is uniformly convergent on E.
Proof. Left as an exercise. Proof similar to previous one. 
Only the material covered so far will be covered in the Mid-Semester Examination.
36 NIKITA AGARWAL

Quiz 2 will be based on the material from this point onward will the end of Week 12.

The space C[a, b].

• Let C[a, b] be the collection of all continuous real-valued functions on the closed
interval [a, b].
• It forms a vector space with operations defined as (f, g ∈ C[a, b], α ∈ R):
Addition: (f + g)(x) := f (x) + g(x), x ∈ [a, b].
Scalar multiplication: (αf )(x) := αf (x), x ∈ [a, b].
• Product of two continuous functions defined as

(f g)(x) := f (x)g(x), x ∈ [a, b],

is continuous.
• A polynomial function p on [a, b] is given by

p(x) = a0 + a1 x + · · · + an xn , x ∈ [a, b],

where a0 , a1 , . . . , an ∈ R, n ≥ 0 is called the degree of p. Polynomials on [a, b] are

continuous.

Theorem 7.6. Let f : [a, b] → R be a continuous function. Then f is bounded. That is,
there exists M > 0 such that |f (x)| ≤ M , for all x ∈ [a, b].

Proof. Proof by contradiction. Suppose f is not bounded.

For each n ≥ 1, there exists xn ∈ [a, b] such that |f (xn )| > n.
Then (xn )n is a sequence inside [a, b], hence is bounded.
By the Bolzano-Weierstrass theorem, it has a convergent subsequence (xnk )k in [a, b]. Let
x0 ∈ [a, b] be its limit.
Since f is continuous, f (xnk ) → f (x0 ) ∈ R, as k → ∞.
Since (f (xnk ))k is a convergent sequence, it is bounded, which is a contradiction. 

Example 7.7. Theorem fails to hold if the domain of f is not a closed interval. Let
f (x) = 1/x, x ∈ (0, 1].

Norm. Since each f ∈ C[a, b] is bounded, we can define

kf k := sup |f (x)|,
x∈[a,b]

which is finite. It is known as the norm of f .

Properties: For all f, g ∈ C[a, b],

i) kf k ≥ 0.
ii) kf k = 0 if and only if f (x) = 0, for all x ∈ [a, b].
iii) kαf k = |α|kf k, for all α ∈ R.
iv) kf + gk ≤ kf k + kgk.

Proofs of the first three properties is straightforward.

Proof of iv): For each x ∈ [a, b],

|f (x) + g(x)| ≤ |f (x)| + |g(x)| ≤ kf k + kgk.

Hence kf + gk ≤ kf k + kgk.
 LECTURE NOTES ON REAL ANALYSIS I 37

Distance/Metric. For f, g ∈ C[a, b], define

ρ(f, g) := kf − gk.
Properties: For all f, g, h ∈ C[a, b],
i) ρ(f, g) ≥ 0.
ii) ρ(f, g) = 0 if and only if f (x) = g(x), for all x ∈ [a, b].
iii) ρ(f, g) = ρ(g, f ).
iv) ρ(f, g) ≤ ρ(f, h) + ρ(h, g).

Proofs of the first three properties is straightforward.

Proof of iv): For each x ∈ [a, b],
|f (x) − g(x)| ≤ |f (x) − h(x)| + |h(x) − g(x)| ≤ ρ(f, h) + ρ(h, g).
Hence ρ(f, g) ≤ ρ(f, h) + ρ(h, g).
Definition 7.8. Let fn , f ∈ C[a, b], n ≥ 1. We say that fn converges to f in C[a, b] if
lim ρ(fn , f ) = 0.
n→∞

Theorem 7.9. Let fn , f ∈ C[a, b], n ≥ 1. The sequence fn converges to f in C[a, b] if

u
and only if fn →
− f on [a, b].
Proof. Let  > 0 be given.
u
(⇐) Let fn →− f on [a, b].
Then there exists N ∈ N such that for all n ≥ N and x ∈ [a, b], |fn (x) − f (x)| < /2.
Hence sup{x ∈ [a, b]}|fn (x) − f (x)| ≤ /2 < , for all n ≥ N .
Therefore, ρ(fn , f ) < , for all n ≥ N . Thus limn→∞ ρ(fn , f ) = 0.

(⇒) Let limn→∞ ρ(fn , f ) = 0.

Then there exists N ∈ N such that ρ(fn , f ) < , for all n ≥ N .
Hence for all n ≥ N and x ∈ [a, b], |fn (x) − f (x)| < .
u
Thus fn →
− f on [a, b]. 
38 NIKITA AGARWAL

These notes are for lectures after the mid-semester examination week.
8. Week 8

Definition 8.1. Let (fn ) be a sequence in C[a, b]. We say that (fn ) is Cauchy if for given
 > 0, there exists N ≥ 1 such that
ρ(fn , fm ) < ,
for all n, m ≥ N .

Theorem 8.2. Every Cauchy sequence in C[a, b] is convergent. This property is called
completeness of C[a, b] and we say that C[a, b] is complete.
Proof. Let (fn ) be a Cauchy sequence in C[a, b]. Let  > 0 be given.
There exists N ∈ N such that for all n, m ≥ N , ρ(fn , fm ) < .
Hence for all x ∈ [a, b] and for all n, m ≥ N , |fn (x) − fm (x)| < .
Therefore (fn ) is uniformly Cauchy on [a, b].
By Cauchy criterion, it is uniformly convergent to some function f : [a, b] → R.
The limit function f ∈ C[a, b] since the uniform limit of continuous functions is continu-
ous. 
Theorem 8.3. A continuous function f : [a, b] → R is uniformly continuous on [a, b].
(A function is uniformly continuous if for given  > 0, there exists δ > 0 such that for
all x, y ∈ [a, b] with |x − y| < δ, we have |f (x) − f (y)| < .)
Proof. Suppose f is not uniformly continuous on [a, b], then there exists  > 0 such that
for all δ > 0, there exists u, v ∈ [a, b] with |u − v| < δ, |f (u) − f (v)| ≥ .
Hence for each n ≥ 1, there exists xn , yn ∈ [a, b] with |xn − yn | < 1/n such that |f (xn ) −
f (yn )| ≥ .
Since (xn ) is a bounded sequence, it has a convergent subsequence, say (xnk )k .
Since (ynk )k is a bounded sequence, it has a convergent subsequence, say (ynk` )` .
Hence both (xnk` )` (every subsequence of a convergent sequence is convergent) and (ynk` )`
are convergent to x and y, say, respectively.
Since |xnk` − ynk` | < 1/nk` for all ` ≥ 1, x = y taking limit as ` → ∞.
Further since |f (xnk` ) − f (ynk` )| ≥  for all ` ≥ 1, |f (x) − f (y)| ≥  by taking limit as
` → ∞ and using sequential criteria of continuity of f at x and y.
This is a contradiction since f (x) = f (y). 
Weierstrass approximation theorem.
Theorem 8.4. Let f ∈ C[0, 1]. Given  > 0, there exists a polynomial function P such
that ρ(f, P ) < .
Corollary 8.4.1. Let f ∈ C[a, b]. Given  > 0, there exists a polynomial function P
such that ρ(f, P ) < .
Proof. Let f ∈ C[a, b].
Define the function g : [0, 1] → R defined as g(x) = f (a + (b − a)x), x ∈ [0, 1].
Then g(0) = f (a), g(1) = f (b), g ∈ C[0, 1].
Given  > 0, there exists a polynomial function P such that ρ(g, P ) < .
Define polynomial function Q : [a, b] → R as
 
x−a
Q(x) = P , x ∈ [a, b].
b−a
 LECTURE NOTES ON REAL ANALYSIS I 39

Then Q(a) = P (0), Q(b) = P (1).

Since g(x) = f (a + (b − a)x), x ∈ [0, 1], we get
 
x−a
f (x) = g , x ∈ [a, b].
b−a
Since ρ(g, P ) < ,
   
x−a x−a
 > ρ(g, P ) = kg − P k = sup |g(x) − P (x)| = sup g −P
x∈[0,1] x∈[a,b] b−a b−a

= sup |f (x) − Q(x)| = ρ(f, Q).

x∈[a,b]

Hence ρ(f, Q) < . 

Corollary 8.4.2. Let f ∈ C[a, b]. There exists a sequence (Pn ) of polynomials such that
u
Pn →
− f on [a, b].
Proof. By the preceding corollary, there exists a polynomial Pn such that ρ(Pn , f ) < 1/n.
Let  > 0 be given. Choose N ∈ N such that 1/N < .
Hence for all n ≥ N ,
ρ(Pn , f ) < 1/n ≤ 1/N < .
u
Thus Pn →
− f on [a, b]. 

Bernstein polynomials. For any f ∈ C[0, 1], define for x ∈ [0, 1] and n ≥ 1,
n    
X n k n−k k
Bn (f )(x) = x (1 − x) f .
k=0
k n

Bn (f ) is called nth Bernstein polynomial of f .

n!
Note: nk =


.
k!(n − k)!
Note
B1 (f )(x) = (1 − x)f (0) + xf (1),
B2 (f )(x) = (1 − x)2 f (0) + x(1 − x)f (1/2) + x2 f (1), etc.
Theorem 8.5. Let f ∈ C[0, 1]. The sequence of polynomials (Bn (f ))n converges uni-
formly to f on [0, 1].
This proves Theorem 8.4.

Heuristic idea of the proof.

• Suppose there is a biased coin with x being the probability of heads appearing.
• If you toss this coin n times and the head appears exactly k times, you get a
reward of f (k/n) rupees.
• What is your expected reward?
• The probability of heads appearing exactly k times in n tosses is nk xk (1 − x)n−k .


• Hence the expected reward is precisely Bn (f )(x).
• As the number of tosses n increases, the fraction of times heads appears k/n gets
closer and closer to x.
• Hence the expected reward gets closer and closer to f (x).
40 NIKITA AGARWAL

Intermediate results needed for the proof of Theorem 8.5.

Lemma 8.6. i) Bn (f ) = f for all constant functions f ∈ C[0, 1].
ii) Bn (f ) ≥ Bn (g) for all f ≥ g in C[0, 1]. Hence |Bn (f )| ≤ Bn (|f |) since f ≤ |f | and
−f ≤ |f |.
iii) Bn (f ) > 0, for all f > 0 in C[0, 1].
iv) Bn (αf + g) = αBn (f ) + Bn (g), for all f, g ∈ C[0, 1], α ∈ R.
v) Bn (f1 )(x) = x, for  f1 (x)= x on [0, 1].
1 x
vi) Bn (f2 )(x) = x2 1 − + , for f2 (x) = x2 on [0, 1].
n n
2
x−x
vii) Bn (f ) = (x − x0 )2 + , for f (x) = (x − x0 )2 on [0, 1], for fixed x0 ∈ R.
n
Proof. i), ii), iii), iv) follow from the definition of Bn (f ).

For vii), use iv) to get Bn (f ) = Bn (f2 ) + x20 − 2x0 Bn (f1 ), and v) and vi).
For v) and vi), we will use the Binomial theorem. For p, q ∈ R, n ≥ 1,
n  
X n k n−k
p q = (p + q)n .
k=0
k
Differentiating this w.r.t. p, we get
n   n  
X n k−1 n−k n−1
X n k k n−k
kp q = n(p + q) ⇒ p q = p(p + q)n−1 . (1)
k=0
k k=0
k n
Differentiating once more w.r.t. p, we get
n   2
X n k k−1 n−k
p q = (p + q)n−1 + (n − 1)p(p + q)n−2
k=0
k n
n   2
X n k p(p + q)n−1 n − 1 2
⇒ 2
pk q n−k = + p (p + q)n−2 . (2)
k=0
k n n n
Setting p = x and q = 1 − x gives v) and vi), respectively. 
Proof. (Proof of Theorem 8.5 for [a, b] = [0, 1])
Let f ∈ C[0, 1]. We will prove that the sequence of polynomials (Bn (f ))n converges
uniformly to f on [0, 1].
First suppose f is not the zero function. Hence M = kf k > 0.
Since f is continuous on [0, 1], it is uniformly continuous by Theorem 8.3.
Given  > 0, there exists δ > 0 such that for all x, y ∈ [0, 1] with |x−y| < δ, |f (x)−f (y)| <
/2.
Let x0 ∈ [0, 1].
For any x ∈ [0, 1], there are two cases:

a) if |x − x0 | < δ, |f (x) − f (x0 )| < , else
2
b) if |x − x0 | ≥ δ,
 2
x − x0
|f (x) − f (x0 )| ≤ 2M ≤ 2M .
δ
Hence any x ∈ [0, 1], combining a) and b), we get
 2
x − x0 
|f (x) − f (x0 )| < 2M + .
δ 2
 LECTURE NOTES ON REAL ANALYSIS I 41

Hence
|Bn (f )(x) − f (x0 )| =i) |Bn (f − f (x0 ))(x)|
≤ii) Bn (|f (x) − f (x0 )|)(x)
 2 !
x − x0 
≤ii) Bn 2M + (x)
δ 2
2M 
=iv) 2
Bn ((x − x0 )2 )(x) +
δ 2
2M 2 2M 2


=vii) (x − x 0 ) + x − x + .
δ2 nδ 2 2
We have used Lemma 8.6 in the above set of inequalities, as indicated.
Set x = x0 , and use x0 − x20 ≤ 1, we get
2M
|Bn (f )(x0 ) − f (x0 )| ≤ + /2 < ,
nδ 2
4M
for all n ≥ N where N ∈ N is such that N > 2 . The above is true for all x0 ∈ [0, 1].
δ
Hence supx0 ∈[0,1] |Bn (f )(x0 ) − f (x0 )| ≤ , for all n ≥ N ⇒ kBn (f ) − f k ≤ , for all
n ≥ N. 
Example 8.7. Let f : (a, b) → R be a continuous function which can be extended to
a continuous function g : [a, b] → R such that g(x) = f (x), for all x ∈ (a, b). Then
g(a) = limx→a+ f (x) and g(b) = limx→b− f (x). Since g is continuous on [a, b], we can
approximate g by a polynomial P in [a, b]. Hence we can approximate f by a polynomial
P in the open interval (a, b) as well.
Such an extension of f to g is not possible all the time. For example, f (x) = sin(1/x) is
continuous on (0, 1) but can not be extended to a continuous function on [0, 1].
Example 8.8. Let f (x) = 1/x, x ∈ (0, 1). We prove that there is no sequence of
u
polynomials (Pn ) such that Pn →
− f on (0, 1).
• Suppose there is such a sequence (Pn ).
• Then
sup |Pn (x) − 1/x| → 0.
x∈(0,1)

• Hence for  = 1/2, there exists N ≥ 1 such that

sup |Pn (x) − 1/x| < 1/2, for all n ≥ N.
x∈(0,1)

• That is,
1/2 > sup |PN (x) − 1/x| ≥ |xPN (x) − 1|, for all x ∈ (0, 1).
x∈(0,1)

Thus, |xPN (x) − 1| < 1/2, for all x ∈ (0, 1).

• Taking limit x → 0+ , we get 0 on the LHS since xPN (x) → 0 (PN being a
polynomial is bounded on (0, 1).)
• Hence 1 < 1/2, which is a contradiction.
Example 8.9. Let f (x) = ex , x ∈ R. We prove that there is no sequence of polynomials
u
(Pn ) such that Pn →
− f on R.
• Suppose there is such a sequence (Pn ) of polynomials. Then it is uniformly Cauchy.
Hence, for  = 1, there exists N ∈ N such that |Pn (x) − Pm (x)| < 1, for all
n, m ≥ N , x ∈ R.
42 NIKITA AGARWAL

• In particular, (m = N ), for all n ≥ N , x ∈ R, |Pn (x) − PN (x)| < 1.

• Since Pn − PN is a polynomial, it is unbounded on R unless it is a constant (since
for any non-constant polynomial Q, limx→−∞ Q(x) is either of ±∞.)
• Hence for all n ≥ N , Pn − PN = an for some an ∈ R. That is, Pn = PN + an , for
all n ≥ N .
• Since (Pn ) is uniformly convergent on R to ex , the sequence an is convergent and
ex = PN + a, where an → a.
• Hence ex = PN +a is a polynomial which is a contradiction, since limx→−∞ ex = 0.
Taylor’s Theorem - General Form.
Theorem 8.10. Let n, p ∈ N. Let f : [a, b] → R be a function such that f (n−1) is
continuous on [a, b] and f (n) exists on (a, b). Then there exists c ∈ (a, b) such that
n−1
X (b − a)k (b − c)n−p (b − a)p (n)
f (b) = f (a) + f (k) (a) + Rn , where Rn = f (c).
k=1
k! p(n − 1)!
Remarks:
(b − a)n (n)
• Lagrange’s form of the remainder: When p = n, Rn = f (c).
n!
n−1
(b − c) (b − a) (n)
• Cauchy’s form of the remainder: When p = 1, Rn = f (c).
(n − 1)!
c−a
• Since c ∈ (a, b), c = a + θ(b − a), θ ∈ (0, 1) ⇐⇒ θ = , the remainder in
b−a
the statement of the theorem becomes
(b − a)n (1 − θ)n−p (n)
Rn = f (a + θ(b − a)).
(n − 1)!
Proof. The proof is an application of the Rolle’s theorem (recall). Define for x ∈ (a, b),
n−1
X (b − x)k
F (x) = f (b) − f (x) − f (k) (x)
.
k=1
k!
n−1
 p
(b − x) b − x
Then F 0 (x) = −f (n) (x) . Define g(x) = F (x) − F (a).
(n − 1)! b−a
Since g is continuous on [a, b], differentiable on (a, b), and g(a) = g(b) = 0, there exists
−p(b − c)p−1
c ∈ (a, b) such that g 0 (c) = 0. Hence F 0 (c) = F (a). Therefore
(b − a)p
(b − c)n−1 −p(b − c)p−1 (b − c)n−p (b − a)p (n)
−f (n) (c) = F (a) ⇒ F (a) = f (c).
(n − 1)! (b − a)p p(n − 1)!

Taylor’s Theorem - Integral form.
Theorem 8.11. Let f : [a, b] → R be a function such that f (n) is continuous on [a, b].
Then
n−1 Z b
X
(k) (b − a)k (b − t)n−1 (n)
f (b) = f (a) + f (a) + Rn , where Rn = f (t) dt.
k=1
k! a (n − 1)!
Remark : By the mean value theorem for integrals,
(b − c)n−1 (n)
Rn = (b − a) f (c),
(n − 1)!
for some c ∈ (a, b). Hence we have recovered the Cauchy’s form of the remainder.
 LECTURE NOTES ON REAL ANALYSIS I 43

Definition 8.12. The nth order Taylor polynomial about a is given by

n
X (x − a)k
f (x) = f (a) + f (k) (a) .
k=1
k!
44 NIKITA AGARWAL

9. Week 9
Note: A few TYPOS have been fixed and some clarifications have been added in Week
8 notes. Please see matter marked in red on Page 38 and 43.

(Recall) Let f : [a, b] → R be a function such that f (n−1) is continuous on [a, b] and
f (n) exists on (a, b). Then there exists c ∈ (a, b) such that
n−1
X (b − a)k (b − c)n−p (b − a)p (n)
f (b) = f (a) + f (k) (a) + Rn , where Rn = f (c).
k=1
k! p(n − 1)!
Lagrange’s form: p = n, Cauchy’s form: p = 1.

Using this, the Taylor’s form or formula for the function f about a point x0 ∈ (a, b)
is given by
n−1
X (x − x0 )k (x − c)n−p (x − x0 )p (n)
f (x) = f (x0 ) + f (k) (x0 ) + Rn , where Rn = f (c),
k=1
k! p(n − 1)!
for some c between x and x0 . This form is valid for all x in [a, b].

Another formulation: We can also write x = x0 + h in the above formula to get

n−1
X hk (x0 + h − c)n−p hp (n)
f (x0 + h) = f (x0 ) + f (k) (x0 ) + Rn , where Rn = f (c).
k=1
k! p(n − 1)!
Here note c lies in x0 and x0 + h, hence c = x0 + θh, for some θ ∈ (0, 1). Hence
n−1
X hk (1 − θ)n−p hn (n)
f (x0 + h) = f (x0 ) + f (k) (x0 ) + Rn , where Rn = f (x0 + θh).
k=1
k! p(n − 1)!
Example 9.1. Consider f (x) = log(1 + x), for x ∈ (−1, ∞). Compute the nth Taylor’s
polynomial and Taylor’s formula for the function f using Lagrange’s form of remainder
about x = 0.
Note
1 1 (n − 1)!
f (1) (x) = , f (2) (x) = − 2
, . . . , f (n) (x) = (−1)n−1 .
1+x (1 + x) (1 + x)n
The Taylor’s formula is given by
(1) (2) x2 (n−1) xn−1
f (x) = f (0) + f (0)x + f (0) + · · · + f (0) + Rn
2! (n − 1)!
x2 x3 xn−1
= x− + + · · · + (−1)n−2 + Rn ,
2 3 n−1
where
xn (n) xn (n − 1)! xn
Rn = f (c) = (−1)n−1 = (−1) n−1
,
n! n! (1 + c)n n(1 + c)n
where c lies between 0 and x.

The nth order Taylor polynomial of f about x = 0 is given by

x2 xn
= f (0) + f (1) (0)x + f (2) (0) + · · · + f (n) (0)
2! n!
2 3 n
x x x
= x− + + · · · + (−1)n−1 .
2 3 n
 LECTURE NOTES ON REAL ANALYSIS I 45

Power Series.
Definition 9.2. A power series is an expression of the form
X∞
an (x − a)n ,
n=0

where an , a, x ∈ R.
P∞
Note that this is a special case of infinite series n=0 fn (x) with fn (x) = an (x − a)n .

Goal: To find values of x for which the power series converges.

We have looked at several examples of this type, such as

∞ ∞
X X xn
xn , , etc.
n=0 n=0
n!

Convergence.
Theorem 9.3. Let ∞ n
P
n=0 an (x − a) be a power series. There is a unique R ∈ [0, ∞]
(extended real number) such that the following hold:
(1) For all x with |x − a| < R, the series converges absolutely and uniformly to a
function f on (a − r, a + r) for any 0 < r < R.
(2) If R ∈ (0, ∞], then f is continuous and differentiable on (a − R, a + R) with
∞
X
0
f (x) = nan (x − a)n−1 .
n=1

(3) For all a − R < x < y < a + R,

Z y ∞
X Z y
f (t) dt = an (t − a)n dt.
x n=0 x

(4) For all x with |x − a| > R, the series diverges.

Strategy of the proof. The key components of the proof are to
• Find R.
• Show absolute and uniform convergence on (a − r, a + r), for all 0 < r < R, say
to a limit f .
• Prove f is continuous.
• Prove f is differentiable.
• Prove f is integrable.

Note that this result says nothing about convergence on the circle |x − a| = R if
0 < R < ∞. That is when x = a − R and when x = a + R.
Proof. Without loss
P∞ of generality (WLOG), assume a = 0.
n
Let E = {|x| : n=0 a n x converges}.
Note that 0 ∈ E, hence E 6= ∅.
If E is bounded above, let R = sup E, otherwise let R = ∞.
Clearly by the definition of R, the power series diverges for |x| > R. Hence we have
proved 4).

Let R > 0 and let 0 < r < R. Since R is the least upper bound of R, there exists
46 NIKITA AGARWAL

x0 ∈ E with r < |x0 | < R and ∞ n

P
n=0 an x0 converges.
Hence
P∞ there exists M > 0 such that |an xn0 | ≤ M , for all n ≥ 0. This is true since
n n
n=0 an x0 converges implies an x0 → 0 as n → ∞. Hence there exists N ≥ 1 such that
−1
|an xn0 | < 1, for all n ≥ N . Let M = max{|a1 x0 |, |a2 x20 |, . . . , |aN −1 xN
0 |, 1}.
Let x be such that |x| ≤ r, then for each n ≥ 1,
 n  n
n n n n r r
|an x | = |an ||x| ≤ |an |r = |an x0 | ≤M .
|x0 | |x0 |
 n
r P∞ r
Since < 1, there series n=1 M is convergent. Hence by the Weierstrass
|xP0| |x0 |
M-test, ∞ n
n=0 an x is uniformly and absolutely convergent for all x with |x| ≤ r. Let f
be the limit.
Now we prove that f is continuous on (−R, R).

Let x P∈ (−R, R). Choose r such that |x| < r < R.

Then ∞ n
n=0 an x is uniformly convergent on (−r, r).
Since each partial sum nk=0 ak xk is a polynomial, hence continuous, their uniform limit
P
f is also continuous on (−r, r).
In particular f is continuous at x.
Since x ∈ (−R, R) was arbitrary, f is continuous on (−R, R).
Now we prove that f is differentiable on (−R, R).
P∞
For this, we first prove that the series n=1 nan xn−1 is uniformly convergent on (−r, r)
for any 0 < r < R.
Let x be such that |x| < r, then
∞ ∞ ∞  n−1 ∞  n−1 !
X X X r X r
|nan xn−1 | < n|an |rn−1 = n|an xn0 | |x0 |−1 ≤ M/r n .
n=0 n=0 n=0
|x0 | n=0
|x0 |

The series on the right converges by ratio test.

By Weierstrass M-test, ∞ n−1
P
n=1 na nx is uniformly and absolutely convergent for all x
with |x| < r. Let g be the limit. Pn k
Finally since the sequence of partial sums sn (x) = k=0 ak x converges uniformly to
n
f and s0n (x) = k=1 kak xk−1 converges uniformly to g on (−r, r), we deduce that f is
P
differentiable on (−r, r) and f 0 = g (by Theorem 6.2).

Finally,
Pn let −R < x < y < R. Choose r > 0 such that −R < −r < x < y < r < R. Since
k
k=0 ak x converges uniformly to f on (−r, r), by Theorem 5.24,
Z yX n Z y
k
ak t dt → f (t) dt.
x k=0 x

Observe that
Z y n
X n Z
X y n
X Z y
k k
ak t dt = ak t dt = ak tk dt.
x k=0 k=0 x k=0 x

Hence we have prove statement 4). 

Definition 9.4. ThePextended realnnumber R ∈ [0, ∞] is called the radius of convergence
of the power series ∞n=0 an (x − a) . The interval (a − R, a + R) is called the interval of
convergence of the power series.
 LECTURE NOTES ON REAL ANALYSIS I 47

Theorem 9.5. If a power series ∞ n

P
n=0 an (x − a) has a positive radius of convergence
0 < R ≤ ∞, then its sum defines a function f on the interval (a−R, a+R). The function
f is infinitely differentiable on this interval. The Taylor series of f at a is given by the
f (n) (a)
given power series. Also an = .
n!
Proof follows from previous results. Observe the steps given below.
Since
X∞
f (x) = an (x − a)n ,
n=0
we have that
∞
X
(1)
f (x) = nan (x − a)n−1 .
n=1
Differentiating repeatedly (term by term), we get for each k ≥ 1,
∞
X
f (k) (x) = n(n − 1) . . . (n − k + 1)an (x − a)n−k .
n=k

Upon substituting x = a in each of the above, we get

f (a) = a0 , f (1) (a) = a1 , . . . , f (k) (a) = k!ak .
Hence the Taylor’s expansion of f about a is
(1) f (n)
f (a) + f (a)(x − a) + · · · + (x − a)n + . . . ,
n!
which equals
a0 + a1 (x − a) + · · · + an (x − a)n + . . . ,
which is nothing but the original power series.
Theorem 9.6. Let ∞ n
P
n=0 an (x − a) be a power series. Assume that one of the following
hold:
i) limn |an+1 /an | = `, for some ` ∈ [0, ∞].
ii) limn |an |1/n = `, for some ` ∈ [0, ∞].
Then the radius of convergence of the given power series is given by R = 1/`.
Proof. Fix x ∈ R. Note
|an+1 (x − a)n+1 | |an+1 |
n
= |x − a|.
|an (x − a) | |an |
Hence by the ratio test, the series ∞ n
P
n=0 an (x − a) converges if `|x − a| < 1 and diverges
if `|x − a| > 1.

Similarly note
|an (x − a)n |1/n = |an |1/n |x − a|.
Hence by the root test, the series ∞ n
P
n=0 an (x − a) converges if `|x − a| < 1 and diverges
if `|x − a| > 1. 
Example 9.7. Discuss convergence of the power series
∞
X
n n xn .
n=1
48 NIKITA AGARWAL

This power series is about the point 0. Here an = nn . Since |an |1/n = n → ∞, the radius
of convergence of the power series is given by R = 1/∞ = 0. Thus the power series
diverges at all points other than 0.
Example 9.8. Discuss convergence of the power series
∞
X xn
.
n=0
n!

This power series is about the point 0. Here an = 1/n!. Since

an+1 1
= → 0,
an n+1
the radius of convergence of the power series is given by R = 1/0 = ∞. Thus the power
series is convergent on all of R.
Example 9.9. Discuss convergence of the power series
∞
X n3
(x − 1)n .
n=0
3n
n3
This power series is about the point 1. Here an = 3n
. Since
an+1 (n + 1)3 3n 1
= n+1 3
→ ,
an 3 n 3
the radius of convergence of the power series is given by R = 3. Thus the power series
is convergent for all x with |x − 1| < 3 and divergent for all x with |x − 1| > 3. That is,
convergence on (−2, 4) and divergent on (−∞, P−2) ∪ (4,n∞).
Now at the end points x = −2, the series is ∞ 3
n=0 (−1) n , which diverges since its n
th

term does not convergePto 0.

At x = 4, the series is ∞ 3
n=0 n , which diverges since its n
th
term does not converge to 0.
Example 9.10. Discuss convergence of the power series
∞
X xn
.
n=1
n

Here an = 1/n. Since |an+1 /an | → 1, the radius of convergence is 1.

P∞ 1
For x = 1, the series becomes n=1 which diverges.
n
(−1)n
For x = −1, the series becomes ∞
P
n=1 which is convergent to log 2.
n
Example 9.11. Give an explicit expression for the function represented by the power
series
∞
X xn
.
n=0
n+1
Here an = 1/n + 1. Since |an+1 /an | → 1, the radius of convergence is 1. For x = 1,
1
the series becomes ∞
P
n=1 which diverges. For x = −1, the series is convergent by
n+1
alternating series test.
 LECTURE NOTES ON REAL ANALYSIS I 49

Let f (x) be the function represented by the given power series on (−1, 1). Then
∞  n+1 0 ∞
0
X x X 1
(xf (x)) = = xn = .
n=0
n+1 n=0
1−x
By the fundamental theorem of calculus,
Z x
1
xf (x) = dt = − log(1 − x), x ∈ (−1, 1).
0 1−t
log(1 − x)
Hence f (x) = − , for x ∈ (−1, 1) \ {0}, and f (0) = 0 (set x = 0 in the power
x
series).
50 NIKITA AGARWAL

10. Week 10
Recall the space C[a, b] of continuous functions on the closed interval [a, b] and the
metric ρ on it.
Definition 10.1. Let F be a non-empty subset of C[a, b]. The diameter of F is defined
as diam F = sup{ρ(f, g) : f, g ∈ F}.
Example 10.2. Observe that finite subsets of C[a, b] are bounded (Check!).
Below we give an example of an unbounded subset of C[a, b].
For each n ≥ 1, let fn : [a, b] → R be the constant function with value n. Then fn ∈ C[a, b]
for each n ≥ 1. The collection F = {fn : n ≥ 1} is not bounded since ρ(fn , f1 ) = n − 1,
which diverges as n → ∞.
Definition 10.3. A non-empty subset F of C[a, b] is said to be bounded if diam F < ∞.
Definition 10.4. A subset F of C[a, b] is said to be totally bounded if given  > 0, there
exists a finite number of subsets F1 , . . . , Fk of F such that diam Fi < , i = 1, . . . , k, and
F = ∪ki=1 Fi .
Theorem 10.5. A subset of a totally bounded set in C[a, b] is totally bounded.
Proof. Let F 0 be a subset of a totally bounded set F in C[a, b]. Let  > 0 be given.
Then there exists a finite number of subsets F1 , . . . , Fk of F such that diam Fi < ,
i = 1, . . . , k, and F = ∪ki=1 Fi .
Let Fi0 = F 0 ∩ Fi , for each i = 1, . . . , k.
Then diam Fi0 ≤ diam Fi < , for i = 1, . . . , k, and F 0 = ∪ki=1 Fi0 .
Hence F 0 is totally bounded. 
Theorem 10.6. A totally bounded subset F of C[a, b] is bounded.
Proof. Since F is totally bounded, there exists a finite number of subsets F1 , . . . , Fk of F
such that diam Fi < 1, i = 1, . . . , k, and F = ∪ki=1 Fi .
Let f1 , f2 , . . . , fk be some functions in F1 , . . . , Fk , respectively. Let
M = max{ρ(fi , fj ) : i, j ∈ {1, . . . , k}}.
Let f, g ∈ F. Then f ∈ Fi and g ∈ Fj . Hence
ρ(f, g) ≤ ρ(f, fi ) + ρ(fi , fj ) + ρ(fj , g) < 2 + ρ(fi , fj ) ≤ 2 + M.
Hence diam(F) ≤ 2 + M < ∞. 
Remark 10.7. The converse is not true in general. We will need a result to establish
this.
Theorem 10.8. Let F be a totally bounded subset of C[a, b]. Then every sequence of
functions in F has a Cauchy subsequence.
Proof. Let (fn )n≥1 be a sequence in F.
If {fn : n ≥ 1} is a finite set, then the sequence (fn ) has a convergent subsequence,
hence Cauchy.
Assume that {fn : n ≥ 1} is an infinite set.
Since F is a finite union of its subsets each with diameter < 1 ( = 1 in definition of
totally bounded).
Observe that one of these subsets, say S1 contains infinitely many fn .
Let (f1,n )n be infinite subsequence of (fn )n which is contained in S1 .
Now the set S1 is totally bounded (since it is a subset of F). Hence it is a finite union of
its subsets each with diameter less than 1/2 ( = 1/2 in definition of totally bounded).
 LECTURE NOTES ON REAL ANALYSIS I 51

Observe that one of these subsets, say S2 contains infinitely many f1,n (since (f1,n )n is
infinite.)
Let (f2,n )n be infinite subsequence of (f1,n )n which is contained in S2 .
Proceeding in this manner, we obtain a decreasing sequence of sets S1 ⊃ S2 ⊃ . . .
such that diam Si < 1/i. Moreover we obtain subsequences (fi,n )n of (fn )n such that
(fi+1,n )n ⊂ Si+1 is a subsequence of (fi,n )n ⊂ Si .
Consider the (diagonal) subsequence (fn,n )n of (fn )n . Note that fn,n ∈ Sn , for all n.
Finally for any given  > 0, choose N ∈ N such that 1/N < .
Then for all n, m ≥ N , d(fn,n , fm,m ) ≤ 1/N < .
Hence we have obtained a Cauchy subsequence of (fn ). 
Remark 10.9. 1) The converse is also true. It can be proved by contradiction. We skip
the proof.
2) If F is just bounded, then the result fails to hold true. For example, the sequence of
continuous functions
fn (x) = xn
in C[0, 1] converges pointwise to a discontinuous function f defined as
f (x) = 0, 0 ≤ x < 1, f (1) = 1.
Further note that F = {fn : n ≥ 1} is bounded since
ρ(fn , fm ) = sup |xn − xm | ≤ 2, n, m ≥ 1.
x∈[0,1]

Moreover no subsequence of (fn ) is Cauchy, otherwise it will be convergent to some

g ∈ C[0, 1] (since C[0, 1] is complete).
u
Since 0 ≤ fn+1 ≤ fn , for all n, if fnk → g in C[0, 1], then fn → g in C[0, 1] =⇒ fn →
− g,
which is a contradiction.
Hence by Theorem 10.8, F is not totally bounded. We have produced an example of a
subset F of C[0, 1] which is bounded but not totally bounded.
Definition 10.10. Let F be a subset of C[a, b]. The family of functions F is said to
equicontinuous if given  > 0, there exists δ > 0 such that
|f (x) − f (y)| < ,
for all x, y ∈ [a, b] with |x − y| < δ and for all f ∈ F.
Remark 10.11. 1) For a given  > 0, same δ > 0 should work for all points x, y and for
all f ∈ F.
2) If a given family of functions is equicontinuous, then each function in it is continuous.
3) Observe that if F is a finite collection with F = {f1 , . . . , fk }. Since a continuous
function on [a, b] is uniformly continuous, each fi is uniformly continuous. Hence, for
each i = 1, . . . , k, given  > 0, there exists δi > 0 such that
|fi (x) − fi (y)| < ,
for all x, y ∈ [a, b] with |x − y| < δi . Choose δ = min{δ1 , . . . , δk } > 0. Then
|fi (x) − fi (y)| < ,
for all x, y ∈ [a, b] with |x − y| < δ and for all fi ∈ F. Hence F is an equicontinuous
family.
Thus every finite subset of C[a, b] is an equicontinuous family.
Theorem 10.12. Let F be a subset of C[a, b]. Then F is a bounded set if and only if
there exists M > 0 such that kf k ≤ M , for all f ∈ F (that is, F is uniformly bounded).
52 NIKITA AGARWAL

Proof. (⇒) F bounded implies D = diam F < ∞.

Fix f0 ∈ F. Let f ∈ F. Then
kf k ≤ kf − f0 k + kf0 k ≤ D + kf0 k.
Take M = D + kf0 k.
(⇐) Let f, g ∈ F. Then
kf − gk ≤ kf k + kgk ≤ 2M.
Hence diam F ≤ 2M . 
Theorem 10.13. Let F be a bounded equicontinuous subset of C[a, b]. Then F is totally
bounded.
Proof. Since F is bounded, there exists M > 0 such that kf k ≤ M , for all f ∈ F, by
Theorem 10.12.
This implies that |f (x)| ≤ M , for all x ∈ [a, b], f ∈ F.
Let  > 0 be given. Since F is equicontinuous, there exists δ > 0 such that |f (x)−f (x0 )| <
/15, for all x, x0 ∈ [a, b] with |x − x0 | < δ and for all f ∈ F.
Let {a = x0 < x1 < · · · < xk = b} be a partition of [a, b] with |xi+1 − xi | < δ, for all
i = 0, . . . , k − 1.
Let {−M = y0 < y1 < · · · < y` = M } be a partition of [−M, M ] with |yj+1 − yj | < /15,
for all j = 0, . . . , ` − 1.
Thus we have created a rectangular grid in the rectangle [a, b] × [−M, M ], whose each
rectangle has height less than /15 and width less than δ.
Let f ∈ F.
Observe that for each xi , we have f (xi ) ∈ [−M, M ], hence f (xi ) ∈ [yj(i) , yj(i+1) ] for some
j(i). Then |f (xi ) − yj(i) | < /15.
Define a function g as follows:
For each xi , let g(xi ) = yj(i) .
Now extend g on the entire interval [a, b] so that it is linear on each of the intervals
[xi , xi+1 ] (graph of g can be plotted by first plotting the value at each xi , and then join
yj(i) and yj(i+1) by a straight line).
For each i = 0, . . . , k − 1,
|g(xi+1 ) − g(xi )| = |yj(i+1) − yj(i) | ≤ |yj(i+1) − f (xi+1 )| + |f (xi+1 ) − f (xi )| + |f (xi ) − yj(i) |
< /15 + /15 + /15 = /5.
Moreover since g is linear on [xi , xi+1 ],
|g(x) − g(xi )| ≤ |g(xi+1 ) − g(xi )| < /5,
for all x ∈ [xi , xi+1 ].
Let x ∈ [a, b]. Let i be such that x ∈ [xi , xi+1 ]. Then
|g(x) − f (x)| ≤ |g(x) − g(xi )| + |g(xi ) − f (xi )| + |f (xi ) − f (x)| < /5 + /15 + /15 = /3.
Hence kf − gk ≤ /3.
By construction, there are at most (` + 1)k+1 many possible functions g (as we vary f
over the collection F). This is because, for each xi , 0 ≤ i ≤ k, there are ` + 1 possible
values of j(i) in the set {0, 1, . . . , `}.
Let G be the collection of these functions.
For any g, let Ag = {h ∈ F : kg − hk ≤ /3}. Then each f ∈ F lies in one of Ag . Hence
F = ∪g∈G Ag .
Moreover, if h1 , h2 ∈ Ag , then kh1 − h2 k ≤ kh1 − gk + kg − h2 k < 2/3.
Hence diam Ag ≤ 2/3 < . 
 LECTURE NOTES ON REAL ANALYSIS I 53

Theorem 10.14. (Arzela-Ascoli’s Theorem) Let (fn ) be a sequence of functions in C[a, b].
Let {fn : n ≥ 1} be an equicontinuous and bounded subset of C[a, b]. Then (fn ) has a
subsequence which uniformly converges to some function in C[a, b].
Proof. Let F = {fn : n ∈ N}.
By Theorem 10.13, F is totally bounded.
Hence by Theorem 10.8, (fn ) has a Cauchy subsequence (fnk )k .
Since C[a, b] is complete, (fnk ) converges to a function f ∈ C[a, b]. That is ρ(fnk , f ) → 0
as k → ∞.
u
Hence fnk →− f. 
Theorem 10.15. (Converse of Arzela-Ascoli’s Theorem) Let F be a subset of C[a, b].
If every sequence of functions in F has a uniformly convergent subsequence, then F is
equicontinuous and bounded.
We do not discuss proof of the converse.
An application of Arzela-Ascoli’s theorem - bounded derivatives. Let (fn ) be a
uniformly bounded sequence of functions on [0, 1] such that |fn0 (x)| ≤ M , for all x ∈ (0, 1),
n ≥ 1, for some M > 0. Prove that (fn ) has a uniformly convergent subsequence.
• First of all, (fn ) ⊂ C[0, 1] since each fn is differentiable.
• Let  > 0 be given.
Let δ = /M . Then for all n ≥ 1, x, y ∈ [0, 1] with |x − y| < δ,
|fn (x) − fn (y)| = |fn0 (t)||x − y| < M.δ = .
Hence {fn } is an equicontinuous family.
• Now we can use Arzela-Ascoli’s theorem. Since {fn : n ≥ 1} is equicontinu-
ous and bounded subset of C[a, b], the sequence (fn ) has a uniformly convergent
subsequence.
An application of Arzela-Ascoli’s theorem - bounded norm. Let (fn ) be a se-
quence of functions on [a, b] such that kfn k ≤ M ,Rfor all n ≥ 1, for some M > 0. Then
x
the sequence of functions (Fn ) defined as Fn (x) = a fn (t) dt, x ∈ [a, b] lies in C[a, b] and
has a uniformly convergent subsequence.
• Let  > 0 be given.
Let δ = /M (b − a). Then for all n ≥ 1, x, y ∈ [a, b] with |x − y| < δ,
Z x Z x
|Fn (x) − Fn (y)| = | fn (t)| dt ≤ | |fn (t)| dt| ≤ M |x − y| < M (b − a).δ = .
y y

Hence {Fn } is an equicontinuous family, thus each Fn ∈ C[a, b].

• Let x ∈ [a, b] and n ≥ 1, then
Z x Z x
|Fn (x)| = | fn (t) dt| ≤ |fn (t)| dt ≤ M (x − a) ≤ M (b − a).
a a
Since x, y, n are arbitrary, kFn k ≤ M (b − a), for all n ≥ 1.
Hence {Fn } is a bounded subset of C[0, 1].
• Now we can use Arzela-Ascoli’s theorem. Since {Fn : n ≥ 1} is equicontinuous
and bounded subset of C[a, b], the sequence (Fn ) has a uniformly convergent
subsequence.
Example 10.16. We prove that the sequence of functions (fn ), where fn (x) = sin nx,
for each x ∈ [0, 2π] and n ≥ 1 is not an equicontinuous family in C[0, 2π]. We prove this
directly using the definition of equicontinuous family of functions.
54 NIKITA AGARWAL

• Let fn (x) = sin nx.

• For all N ≥ 1, |fN (0) − fN (1/N )| = sin 1.
sin 1
• Choose  = .
2
• Let δ > 0. Choose N ∈ N such that 1/N < δ.
• Then for x = 0 and y = 1/N in [0, π], we have |x − y| = 1/N < δ, but |fN (x) −
fN (y)| = sin 1 ≥ .
• Hence no δ > 0 works for this .
Without using the converse of Arzela-Ascoli’s Theorem, we prove that there is no subse-
quence of (fn ) which converges uniformly. We prove this by contradiction.
Let (fnk )k≥1 be a subsequence of (fn ) which converges uniformly. Then for  = 1, there
exists K ≥ 1 such that |fnk (x) − fn` (x)| < 1, for all k, ` ≥ K, x ∈ [0, 2π].
Since fnK (1/nK ) = sin 1 > 0, there exists an interval (a, b) such that 1/nK ∈ (a, b) ⊂
[0, 2π] and fnk (x) > 0, for all x ∈ (a, b). Choose ` ≥ K such that n` > 2π/(b − a). Then
fn` takes all the values from −1 to 1 in the interval (a, b). Hence there exists x0 ∈ (a, b)
where fn` (x0 ) = −1. Since fnk (x0 ) > 0, we have
1 > |fnk (x0 ) − fn` (x0 )| ≥ fnk (x0 ) + 1 > 1,
which is a contradiction.
Example 10.17. The sequence of continuous functions
fn (x) = xn
on [0, 1] converges pointwise to a discontinuous function f defined as
f (x) = 0, 0 ≤ x < 1, f (1) = 1.
Moreover no subsequence of (fn ) converges uniformly to f since f is discontinuous (since
uniform limit of continuous functions is continuous.)
Further note that F = {fn : n ≥ 1} is bounded since kfn k ≤ 1, for all n ≥ 1.
Hence by Arzela-Ascoli’s theorem, F is not equicontinuous.
Example 10.18. Consider the sequence of functions fn : R → R defined as

 |x|
, |x| ≤ n
fn (x) = n
1, |x| > n.
We will prove that the sequence (fn ) is equicontinuous and uniformly bounded, but it
does not have a uniformly convergent subsequence. This gives a counter-example for the
Arzela-Ascoli’s theorem since the domain of the sequence of function is not of type [a, b].
• Clearly 0 ≤ fn (x) ≤ 1, for all x ∈ R and for all n ≥ 1. Hence the sequence (fn )
is uniformly bounded on R.
• Moreover, fn+1 (x) ≤ fn (x), for all x ∈ R and for all n ≥ 1. Observe that:
a) For |x| ≤ n and |y| ≤ n,
|x − y|
|fn (x) − fn (y)| = ||x|/n − |y|/n| = ≤ |x − y|.
n
b) For |x| ≥ n and |y| ≥ n,
|fn (x) − fn (y)| = 0.
c) For |x| < n and |y| ≥ n,
n − |x| |y| − |x| |y − x|
|fn (x) − fn (y)| = |1 − |x|/n| ≤ ≤ ≤ ≤ |y − x|.
n n n
 LECTURE NOTES ON REAL ANALYSIS I 55

Hence for given  > 0, take δ = , then |fn (x) − fn (y)| < , for all |x − y| < δ,
x, y ∈ R, n ≥ 1. Therefore the sequence (fn ) is equicontinuous on R.
p.w.
• In fact, 0 ≤ fn (x) ≤ |x|/n, for all x ∈ R and for all n ≥ 1. Hence fn −−→ 0 on R.
p.w.
• Finally, let (fnk )k is any subsequence of (fn ). Then fnk −−→ 0. Since supx∈R |fnk (x)| =
1, for all k ≥ 1, we get limk→∞ supx∈R |fnk (x)| = 1 6= 0. Thus fnk does not con-
verge to 0 uniformly.
56 NIKITA AGARWAL

11. Week 11
The goal for the next set of lectures will be to cover the following topics:
• Monotonic functions, functions of bounded variation, rectifiable curves
• Riemann-Stieltjes integral, properties of Riemann-Stieltjes integral, differentiation
of the integral, fundamental theorem of calculus, integration by parts, Gamma
function
• The main text for these topics will be: Chapters 6 and 7, T. M. Apostol,
Mathematical Analysis (2nd Edn.), Narosa Publishing, 1985
Definition 11.1. Let E be a non-empty subset of R.
• A function f : E → R is said to be increasing (non-decreasing) on E if for any
pair of points x, y ∈ E,
x < y ⇒ f (x) ≤ f (y).
The function f is strictly increasing on E if x < y ⇒ f (x) < f (y).
• Let E be a non-empty subset of R. A function f : E → R is said to be decreasing
(non-increasing) on E if for any pair of points x, y ∈ E,
x < y ⇒ f (x) ≥ f (y).
The function f is strictly decreasing on E if x < y ⇒ f (x) > f (y).
• A function is called monotonic on E if it is increasing or decreasing on E.
Note that f is decreasing on E if and only if −f is increasing on E. Because of this,
in many situations relating to monotonic functions, it is enough to consider increasing
functions.

We will use this shorthand for left hand and right hand limits of a function f at a
point c throughout the remainder of the lectures:
f (c−) = lim− f (x), f (c+) = lim+ f (x).
x→c x→c

Theorem 11.2. Let f : [a, b] → R be an increasing function and c ∈ (a, b). Then f (c−)
and f (c+) exist and
f (c−) ≤ f (c) ≤ f (c+).
At the end points c = a or c = b,
f (a) ≤ f (a+), f (b−) ≤ f (b).
For a decreasing function f , all the inequalities will be reversed in the statement of the
theorem.
The proof is immediate by observing the graph of f .
Proof. • Let A = {f (x) : x ∈ (a, c)}. This set is non-empty and bounded above
by f (c) (since f is increasing.)
• Let α = sup A. Then α ≤ f (c).
• We will prove that f (c−) = α. That is, for given  > 0, there exists δ > 0 such
that
c − δ < x < c ⇒ |f (x) − α| = α − f (x) < .
• Let  > 0 be given. Since α = sup A, there exists x1 ∈ (a, c) such that α −  <
f (x1 ) ≤ α.
• Since f is increasing, α −  < f (x1 ) ≤ f (x) ≤ α, for all x ∈ (x1 , c).
• That is, α − f (x) < , for all x1 < x < c.
• Take δ = c − x1 > 0 to obtain the statement in red.
 LECTURE NOTES ON REAL ANALYSIS I 57

3.5

3.0

2.5

2.0

1.5

1.0

0.5

0.2 0.4 0.6 0.8 1.0

Figure 1. An increasing function with jump discontinuities at 0.4 and 0.5

• Thus f (c−) ≤ f (c).

• The proofs for f (c) ≤ f (c+), f (a) ≤ f (a+) and f (b−) ≤ f (b) are similar.

Theorem 11.3. Let f : [a, b] → R be a strictly increasing and continuous function. Then
f −1 exists, continuous and strictly increasing on [f (a), f (b)].
Note: Since f is strictly increasing on [a, b], it is immediate that f : [a, b] → [f (a), f (b)].
Proof. • Since f is strictly increasing on f , it is one-one, hence f −1 exists.
• Let y1 , y2 ∈ [f (a), f (b)], y1 < y2 . Then y1 = f (x1 ), y2 = f (x2 ). Hence x1 =
f −1 (y1 ), x2 = f −1 (y2 ). Then x1 < x2 . For if, x1 ≥ x2 , then f (x1 ) ≥ f (x2 ), which
is a contradiction. Hence f −1 is strictly increasing.
• Let d = f (c), c ∈ (a, b). Let  > 0 be given. Let d2 = f (c + ) and d1 = f (c − ).
Choose δ = min{d2 − d, d − d1 } > 0. Then for d − δ < y < d + δ, since d1 ≤
d − δ < y < d + δ ≤ d2 , since f −1 is strictly increasing, we get
f −1 (d1 ) < f −1 (y) < f −1 (d2 ) ⇐⇒ c −  < f −1 (y) < c + .
Thus f −1 is continuous at d.

Theorem 11.4. Let f : [a, b] → R be an increasing function. Let a = x0 < x1 < · · · <
xn = b. Then
n−1
X
(f (xk +) − f (xk −)) ≤ f (b) − f (a).
k=1

Note that f (xk +) − f (xk −) is the jump of f at xk .

Proof. • For 0 ≤ k ≤ n − 1, let yk ∈ (xk , xk+1 ).
• For 1 ≤ k ≤ n − 1, f (yk−1 ) ≤ f (xk −) and f (xk +) ≤ f (yk ) (by the definition of
left hand and right hand limits).
• Hence f (xk +) − f (xk −) ≤ f (yk ) − f (yk−1 ).
• Take sum over 1 ≤ k ≤ n − 1, to get
n−1
X n−1
X
(f (xk +) − f (xk −)) ≤ f (yk ) − f (yk−1 ) = f (yn−1 ) − f (y0 ) ≤ f (b) − f (a).
k=1 k=1


58 NIKITA AGARWAL

Theorem 11.5. Let f : [a, b] → R be a monotonic function. Then the set of discontinu-
ities of f is countable.
Proof. • Assume that f is increasing.
• Let Em = {x ∈ (a, b) : f (x+) − f (x−) > 1/m}, for m ∈ N.
• If Em 6= ∅, let x1 , . . . , xn ∈ Em , then f (xk +) − f (xk −) > 1/m, hence
n−1
n−1 X
≤ (f (xk +) − f (xk −)) ≤ f (b) − f (a).
m k=1

• Hence Em is a finite set for all m ∈ N.

• Note that each point of discontinuity of f lies in one of the Em .
• Therefore, the set of discontinuities of f is a subset of ∪m Em , which is countable
(countable union of finite sets is countable).

Definition 11.6. Let [a, b] be a closed interval. A set of points P = {x0 , x1 , . . . , xn }
satisfying a = x0 < x1 < · · · < xn = b is called a partition of [a, b].
is called the k th subinterval of P .
Each sub-interval [xk−1 , xk ]P
Let ∆k = xk − xk−1 . Note nk=1 ∆k = b − a.
The collection of all possible partitions of [a, b] will be denoted by P[a, b].
Definition 11.7. Let f : [a, b] → R. For a partition P = {x0 , x1 , . . . , xn } of [a, b], let
∆fk = f (xk ) − f (xk−1 ). If there exists M > 0 such that
X n
|∆fk | ≤ M,
k=1

for all partitions of [a, b], then f is said to be of bounded variation on [a, b].
We write BV [a, b] for the collection of all the functions of bounded variation on [a, b].
Theorem 11.8. If f is monotonic on [a, b], then it is of bounded variation on [a, b].
Proof. • Let f be an increasing function.
• For any partition of [a, b], ∆fk = f (xk ) − f (xk−1 ) ≥ 0.
• Hence
X n Xn
|∆fk | = ∆fk = f (b) − f (a).
k=1 k=1

Theorem 11.9. If f is continuous on [a, b], f 0 exists and is bounded: |f 0 (x)| ≤ A, for
all x ∈ (a, b), then f is of bounded variation on [a, b].
Proof. • By mean-value theorem, ∆fk = f (xk ) − f (xk−1 ) = f 0 (tk )(xk − xk−1 ), for
some tk ∈ (xk−1 , xk ).
• Hence |∆fk | = |f (xk ) − f (xk−1 )| = |f 0 (tk )|(xk − xk−1 ) ≤ A(xk − xk−1 ).
• Hence
Xn X n
|∆fk | ≤ A (xk − xk−1 ) = A(b − a).
k=1 k=1

Theorem 11.10. If f is of bounded variation on [a, b], then it is bounded.
• Since f is of bounded variation, nk=1 |∆fk | ≤ M , for all partitions.
P
Proof.
• Let x ∈ (a, b).
 LECTURE NOTES ON REAL ANALYSIS I 59

• Consider the partition P = {a, x, b}.

• Then |f (x) − f (a)| + |f (b) − f (x)| ≤ M .
• Hence |f (x) − f (a)| ≤ M ⇒ |f (x)| ≤ |f (x) − f (a)| + |f (a)| ≤ M + |f (a)|.
• Therefore f is bounded on [a, b] with M + |f (a)| as an upper bound.

Example 11.11. Let f (x) = x cos π/(2x), x 6= 0, and f (0) = 0, on [0, 1].
• f is continuous on [0, 1].
1 1 1
• Consider partition Pn = {0, , , . . . , , 1}.
2n 2n − 1 2
• Then
2n
X 1 1 1 1
|∆fk | = |f ( ) − f (0)| + |f ( ) − f ( )| + · · · + |f (1) − f ( )|
k=1
2n 2n − 1 2n 2
1 1 1 1 1 1
= + + + + ··· + +
2n 2n 2n − 2 2n − 2 2 2
1 1
= 1 + + ··· + .
2 n
P∞
• Since n=1 1/n = ∞, the sums 2n
P
k=1 |∆fk | are not bounded above for all n ≥ 1.
• Hence f is not of bounded variation on [0, 1].
π
• Aside: Observe that for x ∈ (0, 1], f 0 (x) = cos π/(2x) + sin π/(2x), which is
2x
not uniformly bounded above.
Example 11.12. Let f (x) = x2 cos π/(2x), x 6= 0, and f (0) = 0, on [0, 1].
• f is continuous on [0, 1].
π
• f 0 (x) = 2x cos π/(2x) + sin π/(2x), for all x ∈ (0, 1], and f 0 (0) = 0.
2
• Hence |f 0 (x)| ≤ 2 + π/2, for all x ∈ [0, 1].
• Using Theorem 11.9, f is of bounded variation on [0, 1].

0.2

0.2 0.4 0.6 0.8 1.0

-0.2

-0.4

Figure 2. The functions x cos π/(2x) (blue) and x2 cos π/(2x) (orange) on [0, 1].

Definition 11.13. Let f : [a, b] → R be a function of bounded variation. For a partition

P = {a = x0 < x1 < · · · < xn = b} of [a, b], let
X n
S(P ) := |∆fk |,
k=1
60 NIKITA AGARWAL

where ∆fk = f (xk ) − f (xk−1 ). The number

Vf (a, b) = sup{S(P ) : P ∈ P[a, b]},
is called the total variation of f on the interval [a, b].
Remark 11.14. • If the interval of bounded variation is clearly specified, we will
write Vf instead of Vf (a, b).
• 0 ≤ Vf (a, b) < ∞.
• Moreover Vf (a, b) = 0 if and only if f is constant on [a, b].
Proof : Clearly when f is constant on [a, b], then Vf (a, b) = 0. Conversely, let
x ∈ (a, b) and consider the partition P = {a, x, b} of [a, b]. Then S(P ) = |f (x) −
f (a)| + |f (b) − f (x)| ≤ Vf (a, b) = 0. Hence |f (x) − f (a)| = |f (b) − f (x)| = 0,
which implies f (x) = f (a) = f (b). Since x ∈ (a, b) is an arbitrary point, we have
proved that f is constant on [a, b].
• Vf (a, b) = f (b) − f (a) if f is an increasing function on [a, b].
• Vf (a, b) = f (a) − f (b) if f is a decreasing function on [a, b].
Theorem 11.15. Let f, g : [a, b] → R be function of bounded variation on [a, b]. Then
f ± g, f g are also functions of bounded variation on [a, b], and
Vf ±g ≤ Vf + Vg , Vf g ≤ kf kVg + kgkVf .
Proof. • Let h = f ± g. Then
|∆hk | = |h(xk ) − h(xk−1 )| ≤ |f (xk ) − f (xk−1 )| + |g(xk ) − g(xk−1 )| = |∆fk | + |∆gk |.
• Let h = f g. Then
|∆hk | = |`(xk ) − `(xk−1 )|
≤ |f (xk )g(xk ) − f (xk−1 )g(xk )| + |f (xk−1 )g(xk ) − f (xk−1 )g(xk−1 )|
= |g(xk )||f (xk ) − f (xk−1 )| + |f (xk−1 )||g(xk ) − g(xk−1 )|
≤ kgk|∆fk | + kf k|∆gk |.
P
Take k on both sides, then RHS is bounded above by Vf + Vg . Then take sup over
all partitions on LHS. 
Theorem 11.16. Let f : [a, b] → R be function of bounded variation on [a, b]. Let
0 < m ≤ f (x), for all x ∈ [a, b]. Then 1/f is a function of bounded variation on [a, b],
and
Vf
V1/f ≤ 2 .
m
Proof. Let g = 1/f . Then
1 1 |∆fk | |∆fk |
|∆gk | = |g(xk ) − g(xk−1 )| = | − |= ≤ .
f (xk ) f (xk−1 ) |f (xk )||f (xk−1 )| m2
Take k on both sides, then RHS is bounded above by Vf /m2 . Then take sup over all
P
partitions on LHS. 
Theorem 11.17. Let f : [a, b] → R be function of bounded variation on [a, b]. Let
c ∈ (a, b). Then f is of bounded variation on [a, c] and [c, b]. Moreover
Vf (a, b) = Vf (a, c) + Vf (c, b).
Proof. • Let P1 be a partition of [a, c] and P2 be a partition of [c, b]. Then P = P1 ∪P2
is a partition of [a, b].
• Hence
S(P1 ) + S(P2 ) = S(P ) ≤ Vf (a, b).
 LECTURE NOTES ON REAL ANALYSIS I 61

• Thus S(P1 ), S(P2 ) are bounded above by Vf (a, b) for all partitions P1 of [a, c] and
all partitions P2 of [c, b].
• Therefore f is of bounded variation on both [a, c] and [c, a].
• Also Vf (a, c) + Vf (c, b) ≤ Vf (a, b), by taking sup over P1 and P2 .
Let us now prove the reverse inequality Vf (a, b) ≤ Vf (a, c) + Vf (c, b).
• Let P = {a = x0 < x1 < · · · < xn = b} be a partition of [a, b]. Let P 0 = P ∪ {c}.
That is P 0 is a new partition of [a, b] obtained from P by inserting point c.
• If c ∈ P , then S(P ) = S(P 0 ) clearly.
• Else if c ∈/ P , then c ∈ (xk−1 , xk ) for some k. Then P 0 = {a = x0 < x1 < · · · <
xk−1 < c < xk < xn = b} is a partition of [a, b].
• Then P1 = {a = x0 < x1 < · · · < xk−1 < c} and P2 = {c < xk < · · · < xn = b}
are partitions of [a, c] and [c, b], respectively.
• Moreover since P 0 = P1 ∪ P2 , we have S(P 0 ) = S(P1 ) + S(P2 ).
• Since
|f (xk ) − f (xk−1 )| ≤ |f (xk ) − f (c)| + |f (c) − f (xk−1 )|,
we get S(P ) ≤ S(P 0 ).
• Hence S(P ) ≤ S(P 0 ) = S(P1 ) + S(P2 ) ≤ Vf (a, c) + Vf (c, b).
• Taking sup over partitions P of [a, b] on the LHS, we obtain
Vf (a, b) ≤ Vf (a, c) + Vf (c, b).

The above result is quite useful for computing Vf . One can divide the
interval [a, b] into sub-intervals in which it is easier to compute the variation.
For example, sub-intervals of monotonicity.
Theorem 11.18. (Total variation as a function) Let f : [a, b] → R be of bounded varia-
tion on [a, b]. Let V : [a, b] → R be defined as
V (a) := 0, V (x) := Vf (a, x), x ∈ (a, b].
Then both V and V − f are increasing functions on [a, b].
Proof. • Let a < x < y < b. Then Vf (a, y) = Vf (a, x) + Vf (x, y). Hence V (y) −
V (x) = Vf (x, y) ≥ 0, which implies V (x) ≤ V (y). Hence V is an increasing
function.
• Let W = V − f . Then
W (y) − W (x) = (V (y) − V (x)) − (f (y) − f (x)) ≥ 0,
since f (y) − f (x) ≤ |f (y) − f (x)| ≤ Vf (x, y) = V (y) − V (x). Hence W (x) ≤ W (y).
Thus W is an increasing function.

Theorem 11.19. (Function of bounded variation as a difference of increasing functions)
Let f : [a, b] → R. Then f is of bounded variation on [a, b] iff f can be expressed as a
difference of two increasing functions on [a, b].
Proof. (⇒) Note that f = V − (V − f ), where both V and V − f are increasing.
(⇐) Increasing functions on [a, b] are of bounded variation. Moreover, the difference of
two functions of bounded variation is a function of bounded variation. 
Remark 11.20. • The expression as a difference of two increasing functions is not
unique since V and V − f can be replaced by V + g and V − f + g, respectively,
for any increasing function g.
62 NIKITA AGARWAL

• In the above statement, increasing can be replaced by decreasing with a simple

observation: f = [−(V − f )] − [−V ].
• In the above statement, increasing can be replaced by strictly increasing with a
simple observation: Replace both V and V −f by V +g and V −f +g, respectively,
for any strictly increasing function g.
Example 11.21. Express f (x) = sin 2πx on [0, 1] as a difference of two increasing func-
tions. See Figure 3.

0.2 0.4 0.6 0.8 1.0

-1

Figure 3. Plots of V (blue), V − f (orange), and f (green)

• f is a continuous function on [0, 1]. Moreover, for each x ∈ [0, 1],

|f 0 (x)| = 2π| cos 2πx| ≤ 2π.
Hence f is of bounded variation on [0, 1] since it is continuous and has bounded
derivative.
• Let us compute the function V given by V (x) = Vf (0, x).
• Note that f is increasing on [0, 1/4] and [3/4, 1], and decreasing on [1/4, 3/4].
Hence for 0 ≤ x ≤ 1/4, Vf (0, x) = f (x) − f (0) = sin 2πx.
For 1/4 ≤ x ≤ 3/4, Vf (0, x) = Vf (0, 1/4) + Vf (1/4, x) = 1 − (f (x) − f (1/4)) =
2 − sin 2πx.
For 3/4 ≤ x ≤ 1, Vf (0, x) = Vf (0, 3/4) + Vf (3/4, x) = 3 + (f (x) − f (3/4)) =
2(1 − sin 2πx).
Thus 
f (x) − f (0),
 0 ≤ x ≤ 1/4
V (x) = V (1/4) − (f (x) − f (1/4)), 1/4 ≤ x ≤ 3/4

V (3/4) + (f (x) − f (3/4)), 3/4 ≤ x ≤ 1.
• That is, 
sin 2πx,
 0 ≤ x ≤ 1/4
V (x) = 2 − sin 2πx, 1/4 ≤ x ≤ 3/4 ,

4 + sin 2πx, 3/4 ≤ x ≤ 1.

0,
 0 ≤ x ≤ 1/4
(V − f )(x) = 2(1 − sin 2πx), 1/4 ≤ x ≤ 3/4

4, 3/4 ≤ x ≤ 1.
• Now f = V − (V − f ), both V and V − f are increasing on [0, 1].
• Similarly, f = −(V − f ) − (−V ), both −V and −(V − f ) are decreasing on [0, 1].
 LECTURE NOTES ON REAL ANALYSIS I 63

• Let g(x) = x on [0, 1] be a strictly increasing function of our choice. Then f =

(V + g) − (V − f + g), both V + g and V − f + g are strictly increasing on [0, 1].
Example 11.22. Express f (x) = x2 on [−1, 1] as a difference of two increasing functions.
• f is a continuous function on [−1, 1]. For each x ∈ [−1, 1],
|f 0 (x)| = |2x| ≤ 2.
Hence f is of bounded variation on [−1, 1] since it is continuous and has bounded
derivative.
• Let us compute the function V given by V (x) = Vf (−1, x).
• Note that f is decreasing on [−1, 0] and increasing on [0, 1].
Hence for −1 ≤ x ≤ 0, Vf (−1, x) = f (−1) − f (x) = 1 − x2 .
For 0 ≤ x ≤ 1, Vf (−1, x) = Vf (−1, 0) + Vf (0, x) = 1 + (f (x) − f (0)) = 1 + x2 .
Thus (
1 − x2 , −1 ≤ x ≤ 0
V (x) = ,
1 + x2 , 0 ≤ x ≤ 1.
(
1, −1 ≤ x ≤ 0
(V − f )(x) =
1 + 2x2 , 0 ≤ x ≤ 1.
• Now f = V − (V − f ), both V and V − f are increasing on [−1, 1].
• Similarly, f = −(V −f )−(−V ), both −V and −(V −f ) are decreasing on [−1, 1].
• Let g(x) = x on [−1, 1] be a strictly increasing function of our choice. Then
f = (V + g) − (V − f + g), both V + g and V − f + g are strictly increasing on
[−1, 1].
Example 11.23. Let f : [0, 1] → R be defined as f (x) = 1, if x ∈ Q and f (x) = 0,
otherwise. We show that f ∈ / BV [0, 1].
• Let n ≥ 1 be any natural number.
• By density of rationals and irrationals in [0, 1], pick a sequence 0 < x1 < x2 <
· · · < x2n−1 < 1 such that xk ∈ Q for k even and xk ∈ R \ Q for k odd.
• Consider the partition Pn = {0 = x0 < x1 < x2 < · · · < x2n−1 < 1 = x2n } of [0, 1].
• Then
S(Pn ) = |f (x1 ) − f (0)| + |f (x2 ) − f (x1 )| + · · · + |f (1) − f (x2n−1 )| = 2n,
since f (xk ) = 1 for k even and f (xk ) = 0 for k odd.
• Hence Vf (0, 1) ≥ S(Pn ) = 2n.
• Since n ≥ 1 is arbitrary, Vf (0, 1) is infinite.
64 NIKITA AGARWAL

12. Week 12
Theorem 12.1. Let f : [a, b] → R be of bounded variation on [a, b]. Let V : [a, b] → R
be defined as
V (a) := 0, V (x) := Vf (a, x), x ∈ (a, b].
Then points of continuity of V and f are the same.
Recall : Both V and V − f are increasing functions. Moreover, f = V − (V − f ).

Curves and Paths.

Definition 12.2. Let f : [a, b] → Rn be defined as
f (x) = (f1 (x), . . . , fn (x)),
where f1 , . . . , fn : [a, b] → R are functions on [a, b]. Then f is called a vector-valued
function on [a, b].
Example 12.3. The function f1 : [0, 1] → R2 defined as f1 (t) = (cos 2πt, sin 2πt) is a
vector-valued function.
Definition 12.4. Let V be a vector space over field F. We know that V is isomorphic
to Rn . A norm on V is a function k.k : V → R satisfying the following properties:
1) kvk ≥ 0, for all v ∈ V and kvk = 0 if and only if v = 0.
2) kαvk = |α|kvk, for all α ∈ F and v ∈ V .
3) kv + wk ≤ kvk + kwk, for all v, w ∈ V .
Definition 12.5. We say that two norms k.k1 and k.k2 on a vector space V are equivalent
if there exist two constants a, b > 0 such that for all v ∈ V ,
akvk1 ≤ kvk2 ≤ bkvk1 .
Theorem 12.6. All norms on a finite-dimensional vector space are equivalent. In par-
ticular, on norms on Rn are equivalent.
Proof will be included in the next week’s notes.

In view of this result, we will fix the following norm on Rn : for X = (x1 , . . . , xn ) ∈ Rn ,
kXk = |x1 | + · · · + |xn |.
Check that this is a norm on Rn .
Theorem 12.7. Let k.k be any norm on Rn . The vector-valued function f is continuous
on [a, b] if and only if each of the functions f1 , . . . , fn are continuous on [a, b].
Proof. Let f be continuous on [a, b]. Let t0 ∈ [a, b]. Let  > 0 be given. There exists
δ > 0 such that for all t ∈ [a, b] with |t − t0 | < δ, we have kf (t) − f (t0 )k < . Thus
n
X
 > kf (t) − f (t0 )k = |(fi (t) − fi (t0 ))|.
i=1

Hence for each i = 1, . . . , n,

|(fi (t) − fi (t0 ))| < .
Thus each fi is continuous at t0 .
Conversely suppose each fi is continuous on [a, b]. Let t0 ∈ [a, b]. Let  > 0 be given.
There exists δi > 0 such that for all t ∈ [a, b] with |t − t0 | < δi , we have kfi (t) − fi (t0 )k <
 each trace out the unit circle x2 + y2 = 1, but the points are visited in opposite
directions. The same circle is traced out five times by the function h(t) = e"'11,
0

LECTURE NOTES ON REAL ANALYSIS I 65

6.10 RECTIFIABLE PATHS AND ARC LENGTH
/n. Next we introduce the concept of arc length of a curve. The idea is to approximate
Let δ = min{δ 1 , . .by
the curve n } > 0. polygons,
. , δinscribed Then fora all t ∈ [a,learned
technique |t −ancient
b] withfrom t0 | < geometers.
δ, we haveOur
intuition tells us that the length of any X n inscribed polygon should not exceed that
of the curve (since a straight line
kf (t) − f (t0 )k =is the shortest path
|(fi (t) − between
fi (t0 ))| < .
two points), so the
length of a curve should be an upper bound
i=1
to the lengths of all inscribed polygons.
Therefore, it seems natural to define the length of a curve to be the least upper
Hence f isbound
continuous at t0of
of the lengths . all possible inscribed polygons. 
For most curves that arise in practice, this gives a useful definition of arc
Definition 12.8. Let f : [a, b] → Rn be a continuous vector-valued function.
length. However, as we will see presently, there are curves for which there is no
• Theuppergraph
boundof to
f the
or the curve
lengths described
of the inscribedby f is the set
polygons. of points
Therefore, f (t) in Rn as t
it becomes
varies fromtoa classify
necessary to b. curves into two categories: those which have a length, and
those which do not. The former
• Since f is the continuous are called
image rectifiable,
of a closed the latter
interval, nonrectifiable.
it is compact and connected
nWe now turn to a formal description of these ideas.
in R .
Let f : [a, fb]is-+called
• The function R" be aa path
[Link] R". For any partition of [a, b] given by
• An interpretation: t ∈ [a, b] P can be thought
= {t0, of as the time. The vector f (t) as the
t1, ... , tm},
n
position of a moving particle in R . The function f will be called a motion, with
this interpretation. ... ,
the points f(to), f(t1), are the vertices of an inscribed polygon. (An
example is shown in Fig. 6.1.) The length of this polygon is denoted by Af(P) and
is defined
Rectifiable pathtoand
be the sumlength.
arc
M
• Let f : [a, b] → Rn be a Af(P) path in RnIlf(tk)
= E, . - f(tk-1)If.
• Let P = {a = t0 < t1 < · · · < tmk=1 = b} be a partition of [a, b].
• Consider
Definitionan6.16.
inscribed
If the setpolygon with
of numbers vertices
Af(P) f (t0 ),for
is bounded . . all
. , fpartitions
(tm ) (points onb],the curve).
P of [a,
then the path f is said to be rectifiable and its arc length, denoted by Af(a, b), is

a = tp t1 t2 t3 t4 t5 t$ = b

Figure 4. from Apostol book

• The length of the polygon is defined as

m
X
Λf (P ) = kf (tk ) − f (tk−1 )k.
k=1

Definition 12.9. If the set {Λf (P ) | P ∈ P[a, b]} is bounded above, then we say that
the path f is rectifiable and its arc length is defined as
Λf (a, b) = sup{Λf (P ) | P ∈ P[a, b]}.
If the set {Λf (P ) | P ∈ P[a, b]} is unbounded, we say that f is a nonrectifiable path.
Theorem 12.10. Let f = (f1 , . . . , fn ) : [a, b] → Rn be a path. Then f is rectifiable if
and only if each of the functions f1 , . . . , fn are of bounded variation on [a, b]. Moreover,
if f is rectifiable, then
Vi (a, b) ≤ Λf (a, b) ≤ V1 (a, b) + · · · + Vn (a, b),
for all i = 1, . . . , n, where Vi (a, b) denotes the total variation of fi .
66 NIKITA AGARWAL

Proof. Let P = {a = t0 < t1 < · · · < tm = b} be a partition of [a, b].

m
X m X
X n n X
X m
Λf (P ) = kf (tk ) − f (tk−1 )k = |fi (tk ) − fi (tk−1 )| = |fi (tk ) − fi (tk−1 )|.
k=1 k=1 i=1 i=1 k=1

Hence m
P
k=1 |fi (tk ) − fi (tk−1 )| ≤ Λf (P ) ≤ V1 (a, b) + · · · + Vn (a, b).
Now take sup over all partitions P on the LHS. 
Theorem 12.11. Let f : [a, b] → Rn be a rectifiable path. Let c ∈ (a, b). Then f is
rectifiable on both [a, c] and [c, b]. Moreover
Λf (a, b) = Λf (a, c) + Λf (c, b).
Proof. For any partition P of [a, b], if we include the point c, we obtain partitions P1 of
[a, c] and P2 of [c, b].
By triangle inequality, Λf (P ) ≤ Λf (P1 ) + Λf (P2 ) ≤ Λf (a, c) + Λf (c, b).
Taking sup over partitions P of [a, b] on LHS, we get Λf (a, b) ≤ Λf (a, c) + Λf (c, b).
For the reverse inequality, let P1 be a partition of [a, c] and P2 of [c, b]. Let P = P1 ∪ P2 ,
partition of [a, b].
Then Λf (P1 ) + Λf (P2 ) = Λf (P ) ≤ Λf (a, b).
Taking sup over partitions P1 , P2 of [a, c], [c, b], respectively on LHS, we get Λf (a, c) +
Λf (c, b) ≤ Λf (a, b). 
Theorem 12.12. Let f : [a, b] → Rn be a rectifiable path. Define a function s : [a, b] → R
as
s(a) = 0, s(x) = Λf (a, x), x ∈ (a, b].
Then
i) s is increasing and continuous on [a, b].
ii) If there is no sub-interval of [a, b] on which f is constant, then s is strictly increasing
on [a, b].
Proof. Let a ≤ x < y ≤ b. Then s(y) − s(x) = Λf (a, y) − Λf (a, x) = Λf (x, y) ≥ 0. Hence
s is an increasing function.
Clearly Λf (x, y) = 0 implies Vk (x, y) = 0, for all k = 1, . . . , n which further implies that
f is constant on [x, y].
Since Λf (x, y) ≤ V1 (x, y) + · · · + Vn (x, y), and as y → x+ , each Vi (x, y) → 0. Hence
Λf (x, y) → 0. Therefore, s(x+) − s(x) = 0. Similarly s(x) − s(x−) = 0. Hence s is
continuous on [a, b]. 
Definition 12.13. Let f : [a, b] → Rn and g : [c, d] → Rn be two paths. These paths
are called equivalent if there exists a continuous, onto, and strictly monotone function
φ : [a, b] → [c, d] such that
g ◦ φ(t) = f (t), t ∈ [a, b].
Remark 12.14. A few remarks about equivalent paths.
• The equivalent paths are said to give two parametric representations of the same
curve in Rn .
• The function φ defines a change of parameter.
• Let C be the curve described by f or g.
• If φ is strictly increasing, then φ(a) = c, φ(b) = d, both f and g trace out C in
the same direction. In this case, φ is called orientation-preserving.
• If φ is strictly decreasing, then φ(a) = d, φ(b) = c, both f and g trace out C in
the reverse direction. In this case, φ is called orientation-reversing.
 LECTURE NOTES ON REAL ANALYSIS I 67

Example 12.15. Let f : [0, 1] → R2 , g : [0, 4] → R2 be given as

f (t) = (cos 8πt, sin 8πt),
g(t) = (cos 2πt, sin 2πt),
Example 12.16. Define f : [0, 1] → R2 as
f (t) = (t, 1 + t2 ), t ∈ [0, 1].
Define g : [0, π/4] → R2 as
g(t) = (tan t, sec2 t), t ∈ [0, π/4].
Define φ : [0, π/4] → [0, 1] as φ(t) = tan t.
Then φ is a continuous, strictly increasing and onto function.
Moreover f (φ(t)) = f (tan t) = (tan t, 1 + tan2 t) = (tan t, sec2 t) = g(t).
Hence f and g are equivalent.
Example 12.17. Let f : [0, 1] → R2 , g : [0, 2] → R2 be given as
f (t) = (cos 4πt, sin 4πt),
g(t) = (cos 2πt, − sin 2πt),
Define φ : [0, 1] → [0, 2] as φ(t) = 2(1 − t).
φ is strictly decreasing, continuous and onto.
Moreover
g ◦ φ(t) = g(2 − 2t) = (cos 4π(1 − t), − sin 4π(1 − t)) = (cos 4πt, sin 4πt) = f (t),
for all t ∈ [0, 1].
Hence f and g are equivalent which trace out the unit circle twice, φ is orientation-
reversing.
Example 12.18. Let f : [0, 1] → R2 , g : [0, 1] → R2 be given as
f (t) = (cos 2πt, sin 2πt),
g(t) = (cos 4πt, sin 4πt),
The paths are not one-one since f (0) = f (1) = (1, 0) and g(0) = g(1/2) = g(1) = (1, 0).
Define φ : [0, 1] → [0, 4] as φ(t) = 4t. Then
g ◦ φ(t) = g(4t) = f (t), t ∈ [0, 1].
Hence f and g are equivalent paths which trace out the unit circle four times, φ is
orientation-preserving. Both f and g have the same graph which is the unit circle. But
they are not equivalent. Explicitly let us understand why?
• Suppose there is a change of parameter map φ : [0, 1] → [0, 1] (which is continu-
ous, strictly monotone and onto, by definition.)

• Then f (φ(t)) = g(t), for all t ∈ [0, 1].

• Since g(0) = g(1/2), f (φ(0)) = f (φ(1/2)).

• Since φ is strictly monotone, φ(0) 6= φ(1/2) and φ(1/2) ∈ (0, 1).

• Moreover since f traces the unit circle only once, f (φ(0)) = (1, 0) (since φ(0) is
either 0 or 1) and f (φ(1/2)) 6= (1, 0).
68 NIKITA AGARWAL

Riemann-Stieltjes Integral. Let us first discuss some useful notations and definitions.
• Let f, α : [a, b] → R be two functions.
• For any partition P = {a = x0 < x1 < · · · < xn = b}, we define
∆αk = α(xk ) − α(xk−1 ), k = 1, . . . , n.
• A partition P 0 of [a, b] with P ⊂ P 0 is called a refinement of P .
• The length of the largest subinterval in the partition P is called the norm of P ,
denoted as kP k. That is,
kP k = max{∆xk = xk − xk−1 | k = 1, . . . , n}.
• Note that if P 0 is a refinement of P , then kP 0 k ≤ kP k.
• For any choice points tk ∈ [xk−1 , xk ], k = 1, . . . , n, the Riemann-Stieltjes sum is
given by
X n Xn
S(P, f, α) := f (tk )∆αk = f (tk )(α(xk ) − α(xk−1 )).
k=1 k=1

• If α(x) = x, for all x ∈ [a, b], then the Riemann-Stieltjes sum are nothing but
Riemann sums.
Definition 12.19. A function f is said to be Riemann-Stieltjes integrable with respect to
α if there exists a real number I such that for given  > 0, there exists a partition P of
[a, b] such that for all refinements P of P and for all choices tk ∈ [xk−1 , xk ], k = 1, . . . , n,
|S(P, f, α) − I| < .
Remark 12.20. A few remarks about the definition.
• The number I, if it exists, is unique. ItRis called theR b Riemann-Stieltjes integral of
b
f with respect to α, and is denoted by a f dα or a f (x) dα(x).
• The function f is known as integrand and the function α is known as integrator.
• If f is Riemann-Stieltjes integrable with respect to α, we will say f ∈ R(α).
Rb
• If α or f is the zero function on [a, b], then a f dα = 0.
• If f (x) = c, for all x ∈ [a, b], then
Xn X n
S(P, f, α) = f (tk )∆αk = c (α(xk ) − α(xk−1 )) = c(α(b) − α(a)).
k=1 k=1
Rb
Hence a
f dα = c(α(b) − α(a)).
Theorem 12.21. (Linearity of integrand) Let c1 , c2 ∈ R. If f, g ∈ R(α) on [a, b], then
c1 f + c2 g ∈ R(α) on [a, b]. Moreover
Z b Z b Z b
(c1 f + c2 g) dα = c1 f dα + c2 g dα.
a a a

Proof. Let h = c1 f + c2 g. Let P = {a = x0 < x1 < · · · < xn = b} be a partition of [a, b].

For all choices of tk ∈ [xk−1 , xk ], k = 1, . . . , n,
Xn X n n
X
S(P, h, α) = h(tk )∆αk = c1 f (tk )∆αk + c2 g(tk )∆αk .
k=1 k=1 k=1

Let  > 0 be given. Since f, g ∈ R(α), there exist partitions P0

and P00 of [a, b] such that
0 0 00 00
for all refinements P of P and P of P ,
Z b Z b
0 00
|S(P , f, α) − f dα| < , and |S(P , g, α) − g dα| < .
a a
 LECTURE NOTES ON REAL ANALYSIS I 69

Let P = P0 ∪ P00 , then any refinement P of P is also a refinement of both P0 and P00 .
Let P be a refinement of P . Then
Z b Z b
|S(P, h, α) − c1 f dα − c2 g dα| ≤
a a
Z b Z b
|c1 S(P, f, α) − c1 f dα dα| + |c2 S(P, g, α) − c2 g dα| ≤ (|c1 | + |c2 |).
a a

Theorem 12.22. (Linearity of integrator) Let c1 , c2 ∈ R. If f ∈ R(α) and f ∈ R(β) on
[a, b], then f ∈ R(c1 α + c2 β) on [a, b]. Moreover
Z b Z b Z b
f d(c1 α + c2 β) = c1 f dα + c2 f dβ.
a a a

Proof left as an exercise. It is similar to that for the previous theorem.

Theorem 12.23. (Additivity) Let c ∈ (a, b). If two of the three integrals exist then the
third one also exists, and
Z b Z c Z b
f dα = f dα + f dα.
a a c

Corollary 12.23.1. If a < c1 < · · · < ck < b. If all but one integrals exist then the
remaining one also exists and
Z b Z c1 Z c2 Z ck Z b
f dα = f dα + f dα + · · · + f dα + f dα.
a a c1 ck−1 ck

Proof of corollary by induction on k.

Theorem 12.24. (Reduction to Riemann integral) Let f ∈ R(α) on [a, b]. Let α be
differentiable on [a, b] and α0 be Riemann integrable on [a, b]. Then the function f α0 is
Riemann integrable on [a, b] and
Z b Z b
f (x) dα(x) = f (x)α0 (x) dx.
a a
0
Note: If α has a continuous derivative α , then the above result applies.
70 NIKITA AGARWAL

13. Week 13
Example 13.1. Consider the function f (x) = [x] on the interval [0, 2]. We know that f
is a increasing function, with value 0 in [0, 1), 1 in [1, 2), and 2 at the point 2. Thus f
has discontinuity at the points 1 and 2.
To compute the variation function V on [0, 2], we have to be particularly careful at the
points of discontinuity of f .
Recall that V (x) = Vf (0, x) is the supremum of sums S(P, f ) over all partitions P of
[0, x]. If f is constant on some subinterval [a, b] I of [0, x], we can ignore all the partition
points in I since the corresponding terms in S(P, f ) will contribute to just f (b) − f (a).
Note that since f is constant on [0, 1), V (x) = Vf (0, x) = f (x) − f (0) = 0.
Now, V (1) = Vf (0, 1) = sup{x∈[0,1)} {|f (x)−f (0)|+|f (1)−f (x)|} = sup{x∈[0,1)} {f (1)} = 1
(independent of x).
For x ∈ (1, 2), V (x) = Vf (0, x) = Vf (0, 1) + Vf (1, x) = V (1) + (f (1) − f (x)) = 1.
Finally let x ∈ (1, 2), then V (2) = Vf (0, x)+Vf (x, 2) = V (x)+sup{y∈[1,2)} {|f (2)−f (y)|} =
1 + sup{y∈[1,2)} {2 − 1} = 2 (independent of y).
Thus the total variation of f on [0, 2] is V (2) = Vf (0, 2) = 2.

Suppose you were just asked to compute total variation of f on [0, 2]. Consider any
partition P = {0 = x0 < x1 < · · · < xn } of [0, 2]. WLOG assume that 1 ∈ P and xk = 1.
Let P 0 = {0 < xk < 2} = {0 < 1 < 2}. Then S(P, f ) = S(P 0 , f ) since f is constant on
[0, 1) and [1, 2). Hence Vf (0, 2) = S(P 0 , f ) = (f (1) − f (0)) + (f (2) − f (1)) = 2.
Rb
Definition. If a < b and a f dα exists, we define
Z a Z b
f dα := − f dα.
b a
Moreover we define Z a
f dα = 0.
a

Theorem 13.2. (Integration by parts) If f ∈ R(α) on [a, b], then α ∈ R(f ) on [a, b].
Moreover Z b Z b
f dα + α df = f (b)α(b) − f (a)α(a).
a a

Remark 13.3. We know that if f ∈ R(α) on [a, b] and α has a continuous derivative on
[a, b], then
Z b Z b
f dα = f (x)α0 (x) dx.
a a
Thus if f ∈ R(α) on [a, b] and α and f have continuous derivatives on [a, b], then
Z b Z b Z b
0 0
f (x)α (x) dx + α(x)f (x) dx = d(αf ).
a a a
Note: Z b
d(αf ) = f (b)α(b) − f (a)α(a).
a

Theorem 13.4. (Change of variable) Let f ∈ R(α) on [a, b] and let g : [c, d] → R be
a strictly monotonic continuous function. Assume that g(c) = a and g(d) = b. For all
x ∈ [c, d],
h(x) = f (g(x)), and β(x) = α(g(x)).
 LECTURE NOTES ON REAL ANALYSIS I 71

Then h ∈ R(β) on [c, d]. Moreover

Z b Z d
f dα = h dβ.
a c

That is,
Z g(d) Z d
f (y) dα(y) = f (g(x)) d {α(g(x))} .
g(c) c

In fact, if g is strictly increasing, then c < d. Else if g is strictly decreasing, then

c > d. In this case, replace [c, d] by [d, c] in the above statement.
Remark 13.5. Reduction of the change of variable formula to Riemann integral:
Let α(x) = x, for all x ∈ [a, b].
Then Z g(d) Z d
f (y) dy = f (g(x)) d {g(x)} .
g(c) c

Definition 13.6. A function α : [a, b] → R is known as a step function if there exists

a partition P = {a = x0 < x1 < · · · < xn = b} of [a, b] such that α is constant on each
subinterval (xk−1 , xk ), k = 1, . . . , n.
The value α(xk +) − α(xk −) is called the jump at xk , for k = 1, . . . , n − 1.
The jump at x0 = a is α(a+) − α(a).
The jump at xn = b is α(b) − α(b−).

Theorem 13.7. (Step functions as integrators) Let a < c < b. Let α : [a, b] → R be
such that α(a), α(c), α(b) are arbitrary. Further
α(x) = α(a), a ≤ x < c, α(x) = α(b), c < x ≤ b.
Let f : [a, b] → R be such that at least one of the functions f, α are continuous from the
left at c, and at least one of the functions f, α are continuous from the right at c. Then
f ∈ R(α) and
Z b
f dα = f (c)[α(b) − α(a)] = f (c)[α(c+) − α(c−)].
a

The above expression holds even if c = a or c = b.

Proof. Let  > 0 be given. Refer to Definition 13.6, we need to find a partition P . We
will find P which contains the point c.
Let P = {a = x0 < x1 < · · · < xn = b} be a partition of [a, b] containing c and that
xk = c. Then in the Riemann sum, only two terms will survive. Hence
S(P, f, α) = f (tk−1 )[α(c) − α(a)] + f (tk )[α(b) − α(c)],
72 NIKITA AGARWAL

where tk−1 ∈ [xk−1 , xk ] and tk ∈ [xk , xk+1 ].

Let ∆ = S(P, f, α) − f (c)[α(b) − α(a)]. Then
∆ = [f (tk−1 ) − f (c)][α(c) − α(a)] + [f (tk ) − f (c)][α(b) − α(c)].
Thus
|∆| ≤ |f (tk−1 ) − f (c)||α(c) − α(a)| + |f (tk ) − f (c)||α(b) − α(c)|.
Case 1: If f is continuous at c (behavior of α is inconsequential), then for given  > 0,
there exists δ > 0 such that if x ∈ [a, b] with |x − c| < δ, then |f (x) − f (c)| < . Hence if
kP k < δ, then
|f (tk−1 ) − f (c)| < , |f (tk ) − f (c)| < .
Thus
|∆| ≤ (|α(c) − α(a)| + |α(b) − α(c)|).
Case 2: If f is discontinuous at c from both the directions, then by hypothesis, α is
continuous at c. This implies α(a) = α(b) = α(c). Hence ∆ = 0.
Case 3: If f is continuous from the right at c but not from the left, then by hypothesis,
α is continuous from the left at c. Thus for given  > 0, there exists δ > 0 such that if
kP k < δ, then
α(a) = α(c), |f (tk ) − f (c)| < .
Hence
|∆| ≤ (|α(b) − α(c)|).
Case 4 (similar as Case 3): If f is continuous from the left at c but not from the right,
then by hypothesis, α is continuous from the right at c. Thus for given  > 0, there exists
δ > 0 such that if kP k < δ, then
|f (tk−1 ) − f (c)| < , α(b) = α(c).
Hence
|∆| ≤ (|α(c) − α(a)|).
To summarize, any P containing c with kP k < δ would work. 
Corollary 13.7.1. Let α : [a, b] → R be a step function (as above) with jump αk at xk .
Let f : [a, b] → R be such that for each k = 0, . . . , n, at least one of the functions f, α are
continuous from the left at xk , and at least one of the functions f, α are continuous from
the right at xk . Then f ∈ R(α) and
Z b Xn
f dα = f (xk )αk .
a k=0

Example 13.8. The simplest step function is the greatest integer function.
• Let α(x) = [x], 0 ≤ x ≤ n.
• Then α is continuous from the right at each point, but not from the left at
1, 2, . . . , n − 1.
• Let f : [0, n] → R be a function which is continuous from the left at 1, 2, . . . , n−1.
• Then by the corollary on the previous slide, f ∈ R(α) and
Z b Xn Xn
f dα = f (k)αk = f (k),
a k=0 k=1

since α0 = 0, α1 = · · · = αn = 1.
 LECTURE NOTES ON REAL ANALYSIS I 73

Example 13.9. This example shows that by changing the value of f at c in Theorem 13.7,
the integral might not even exist.
Let α(x) = 0, x 6= 0, and α(0) = −1. (jump only at 0)
1) If f is a constant function on [−1, 1], then by Theorem 13.7,
Z 1
f dα = f (0)(α(0+) − α(0−)) = 0.
−1

2) Let f (x) = 1, for x 6= 0, and f (0) = 2. Since α and f are both discontinuous at 0 from
the left as well as from the right, we cannot use Theorem 13.7.
We directly work out the calculations. For any partition of P = {−1 = x0 < x1 <
. . . xk . . . , xn = 1} containing 0 such that xk = 0,
S(P, f, α) = f (tk−1 )[α(0) − α(xk−1 )] + f (tk )[α(xk+1 ) − α(0)] = f (tk ) − f (tk−1 ),
where xk−1 ≤ tk−1 ≤ 0 ≤ tk ≤ xk .
If tk−1 < 0 and tk > 0, then S(P, f, α) = 0.
On the other hand, if tk−1 = 0 and tk > 0, then S(P, f, α) = −1.
Thus the Riemann sum takes different values for any partition P containing 0, depending
upon theR 1 choice of points tk−1 and tk .
Hence −1 f dα does not exist.
Example 13.10. We prove that
n Z n
X 1 1 [x]
= +s dx.
k=1
ks ns−1 1 xs+1
1
• Let f (x) = [x] and α(x) = on [0.5, n].
xs Rb Rb
• Since α has a derivative which is Riemann integrable, a f dα = a f (x)α0 (x) dx.
Hence Z n Z n Z n
0 [x]
f dα = f (x)α (x) dx = −s s+1
dx.
0.5 0.5 1 x
Z n n
X 1
α df = s
.
0.5 k=1
k
n 1
f (n)α(n) − f (0.5)α(0.5) = s
= s−1 .
n n
• Use integration by parts formula.
• If you use [1, n] instead of [0.5, n], everything will work fine. In this,
Z n n n
!
X 1 X 1
α df = s
= s
− 1.
1 k=2
k k=1
k

Theorem 13.11. (Euler’s summation formula) Let f : [a, b] → R has a continuous

derivative on [a, b]. Let α(x) = ((x)) = x − [x], x ∈ [a, b] (floor function). Then
X Z b Z b
f (n) = f (x) dx + f 0 (x)α(x) dx + f (a)α(a) − f (b)α(b).
a<n≤b a a

When a, b ∈ Z, then the above formula becomes

Xb Z b Z b
f (n) = f (x) dx + f 0 (x)α(x) dx + f (a).
n=a a a
74 NIKITA AGARWAL

which is equivalent to
b Z b Z b  
X
0 1 f (a) + f (b)
f (n) = f (x) dx + f (x) α(x) − dx + .
n=a a a 2 2
Proof. Let us use integration by parts formula
Z b Z b
f dα + α df = f (b)α(b) − f (a)α(a).
a a
Since f has a continuous derivative on [a, b], this becomes
Z b Z b
f dα + α(x)f 0 (x) dx = f (b)α(b) − f (a)α(a). (1)
a a
From Example 13.8,
Z b X
f d[x] = f (n). (2)
a a<n≤b

Note α(x) = x − [x]. Hence by linearity of R-S integral w.r.t. the integrator,
Z b Z b Z b
f dα + f d[x] = f (x) dx.
a a a
Adding equations (1) and (2) gives the required identity.

Now suppose a, b ∈ Z. Since α(a) = α(b) = 0, and

b
X b
X b
X
f (n) = f (n) = f (n) − f (a),
n=a n=a+1 n=a

we get the required identity.

Further since
b Z b
1 b 0
Z   Z
0 1 0
f (x) α(x) − dx = f (x)α(x) dx − f (x) dx
a 2 a 2 a
Z b
f (b) − f (a)
= f 0 (x)α(x) dx − ,
a 2
we get the required identity. 
Remark 13.12. • This formula illustrates the use of Riemann-Stieltjes integral to
obtain a formula relation the integral of a function f on an interval [a, b] and its
values at integer points in [a, b].
• The formula is used to approximate integrals by sums, or conversely, sums by
integrals.
A solved exercise not discussed in the class. Prove that
n Z n
X 1 x − [x]
= log n − dx + 1.
k=1
k 1 x2
Solution: Let f (x) = 1/x and α(x) = [x] on [1, n]. Then
Z n n n
!
X 1 X 1
f dα = = − 1.
1 k=2
k k=1
k
 LECTURE NOTES ON REAL ANALYSIS I 75

Moreover
Z n n n Z n Z n
x − [x]
Z Z
0 [x] x
α df = α(x)f (x) dx = − 2
dx = 2
dx − 2
dx
1 1 1 x 1 x 1 x
Z n Z n Z n
x − [x] 1 x − [x]
= 2
dx − dx = dx − log n.
1 x 1 x 1 x2
Using integration by parts formula
Z n Z n
f dα + f dα = f (n)α(n) − f (1)α(1) = 0,
1 1
we get the required identity.
76 NIKITA AGARWAL

14. Week 14
Monotonic Functions as integrators.
Definition 14.1. Let f, α : [a, b] → R, P = {a = x0 < x1 < · · · < xn = b} be a partition
of [a, b], and
Mk (f ) = sup{f (x) | x ∈ [xk−1 , xk ]}, mk (f ) = inf{f (x) | x ∈ [xk−1 , xk ]}.
The numbers
n
X n
X
U (P, f, α) = Mk (f )∆αk , L(P, f, α) = mk (f )∆αk ,
k=1 k=1

are called the upper and lower Riemann Stieltjes sum, respectively, of f with respect to
α for the partition P .
Remark 14.2. 1) mk (f ) ≤ Mk (f ).
2) If α is increasing, then ∆αk ≥ 0. For any tk ∈ [xk−1 , xk ], mk (f ) ≤ f (tk ) ≤ Mk (f ).
Hence L(P, f, α) ≤ S(P, f, α) ≤ U (P, f, α), for all Riemann sums.
Henceforth, we will focus on integrators α which is a monotonically increasing function.
Observe that if α is monotonically decreasing, then β = −α is monotonically increasing,
and Z b Z b
f dα = − f dβ.
a a

Example 14.3. Let f (x) = 1 − x2 , α(x) = x2 on [−1, 2].

• Consider the partition P = {x0 = −1, x1 = −1/2, x2 = 1/2, x3 = 1, x4 = 2}.
• Then M1 (f ) = f (−1/2) = 3/4 (sup on [−1, −1/2]), m1 (f ) = f (−1) = 0 (inf on
[−1, −1/2]).
• M2 (f ) = f (0) = 1 (sup on [−1/2, 1/2]), m2 (f ) = f (−1/2) = 3/4 (inf on
[−1/2, 1/2]).
• Then M3 (f ) = f (1/2) = 3/4 (sup on [1/2, 1]), m3 (f ) = f (1) = 0 (inf on [1/2, 1]).
• Then M4 (f ) = f (1) = 0 (sup on [1, 2]), m4 (f ) = f (2) = −3 (inf on [1, 2]).
• ∆α1 = α(−1/2) − α(−1) = −3/4, ∆α2 = α(1/2) − α(−1/2) = 0, ∆α3 = α(1) −
α(1/2) = 3/4, ∆α4 = α(2) − α(1) = 3.
• Hence
X4 X 4
U (P, f, α) = Mk (f )∆αk = 0, L(P, f, α) = mk (f )∆αk = −9.
k=1 k=1

Theorem 14.4. Let α be an increasing function on [a, b] (denoted by α %). Then

(1) If P 0 is finer than P , then
U (P 0 , f, α) ≤ U (P, f, α), L(P 0 , f, α) ≥ L(P, f, α).
(2) For any two partitions P1 and P2 ,
L(P1 , f, α) ≤ U (P2 , f, α).
Proof. For (1), note the following: If xk−1 < y < xk , then
• ∆αk = α(xk ) − α(xk−1 ) = (α(y) − α(xk−1 )) + (αk − α(y)).
• Mk (f ) ≥ sup{f (x) | x ∈ [xk−1 , y]}, and also Mk (f ) ≥ sup{f (x) | x ∈ [y, xk ]}.
• mk (f ) ≤ inf{f (x) | x ∈ [xk−1 , y]}, and also mk (f ) ≤ inf{f (x) | x ∈ [y, xk ]}.
For (2), note that P1 ∪ P2 is a refinement of both P1 and P2 . Now use (1). 
 LECTURE NOTES ON REAL ANALYSIS I 77

Definition 14.5. Let α % on [a, b].

The upper Stieltjes integral of f with respect to α is defined as
Z b
f dα = inf{U (P, f, α) P ∈ P[a, b]}.
a
The lower Stieltjes integral of f with respect to α is defined as
Z b
f dα = sup{L(P, f, α) P ∈ P[a, b]}.
a

If α(x) = x, the upper and lower Stieltjes integral are called upper and lower Riemann
integral, respectively.
Theorem 14.6. Let α % on [a, b]. Then
Z b Z b
f dα ≤ f dα.
a a

Definition 14.7. (Riemann’s condition) A function f : [a, b] → R is said to satisfy

Riemann’s condition on [a, b] with respect to α, if for given  > 0, there exists a partition
P such that for all refinements P of P , we have
0 ≤ U (P, f, α) − L(P, f, α) < .
Theorem 14.8. Let α % on [a, b]. The the following statements are equivalent:
(1) f ∈ R(α) on [a, b].
(2) f satisfies Riemann’s condition on [a, b] with respect to α.
Rb Rb
(3) a f dα = a f dα.
Theorem 14.9. Let α : [a, b] → R be a function of bounded variation. Let V be the total
variation function of α on [a, b] with V (a) = 0. Let f : [a, b] → R be a bounded function.
If f ∈ R(α) on [a, b], then f ∈ R(V ) on [a, b].
Remark : This result enables us to reduce the theory of R-S integration w.r.t. functions
of bounded variation to increasing functions as integrators.
Theorem 14.10. Let α : [a, b] → R be a function of bounded variation and let f ∈ R(α)
on [a, b]. Then f ∈ R(α) on every subinterval [c, d] of [a, b].
Remark : Compare this with Corollary 12.23.1.
Theorem 14.11. (Necessary conditions for R-S integrability) Let α % on [a, b] and let
a < c < b. Assume that both f and α are discontinuous from the right at c. Then f ∈
/ R(α)
on [a, b]. The same conclusion holds true if both f and α are discontinuous from the left
at c.
Proof. Since both f and α are discontinuous from the right at c, there exists  > 0 such
that for all δ > 0, there exist x, y (depending on δ) in (c, c + δ) such that
|f (x) − f (c)| ≥ , |α(y) − α(c)| ≥ .
• Let P be a partition of [a, b] containing c = xk−1 and with xk − xk−1 = xk − c < δ.
Then
U (P, f, α) − L(P, f, α) ≥ (Mk (f ) − mk (f ))∆αk ≥ ∆αk .
• We can assume that xk is chosen such that |α(xk ) − α(c)| ≥  (if needed take a
refinement of P ).
78 NIKITA AGARWAL

• Then U (P, f, α) − L(P, f, α) ≥ 2 .

• Now suppose Riemann’s condition is satisfied. Then for this , there exists a
partition P of [a, b] such that for all refinements P 0 of P , we have U (P 0 , f, α) −
L(P 0 , f, α) < .
• Consider P 0 = P ∪ P which is a refinement of P . Hence we have arrived at a
contradiction and f ∈ / R(α) on [a, b].
• Similar argument works when both f and α are discontinuous from the left at c.

Theorem 14.12. (First Mean-value theorem for R-S integration) Let α % and let f ∈
R(α) on [a, b]. Let M and m denote the sup and inf of the set {f (x) | x ∈ [a, b]}. Then
Rb Rb
there exists a real number d ∈ [m, M ] such that a f dα = d a dα = d(α(b) − α(a)). In
particular, if f is continuous on [a, b], then d = f (c), for some c ∈ [a, b].
Proof. Note that, for all partitions P of [a, b],
Z b
m(α(b) − α(a)) ≤ L(P, f, α) ≤ f dα ≤ U (P, f, α) ≤ M (α(b) − α(a)).
a

If α(a) = α(b), then the result holds trivially.

Rb Rb
a
f dα a
f dα
Assume α(a) < α(b). Then ∈ [m, M ]. Thus = d, for some
α(b) − α(a) α(b) − α(a)
d ∈ [m, M ], and the result follows.
If f is continuous, by the Intermediate Value Theorem for continuous functions, d = f (c),
for some c ∈ [a, b]. 
Theorem 14.13. Let α be a function Rof bounded variation on [a, b] and let f ∈ R(α) on
x
[a, b]. Define F : [a, b] → R as F (x) = a f dα. Then
(1) F is of bounded variation on [a, b].
(2) Every point of continuity of α is a point of continuity of F .
(3) If α % on [a, b], then the derivative F 0 exists at each point where α0 exists and
where f is continuous. At such points x, F 0 (x) = f (x)α0 (x).
Proof. Note: This proof is slightly different from the one discussed in the class. In the
class, I used mean value theorem, but here I use ideas from mean value theorem but not
the actual theorem.

First of all, F is defined since f ∈ R(α) on [a, b] implies f ∈ R(α) on [a, x].
Further it suffices to assume
Ry that α % on [a, b].
Since F (y) − F (x) = x f dα and α is increasing, we have
m[x,y] (α(y) − α(x)) ≤ F (y) − F (x) ≤ M[x,y] (α(y) − α(x)),
where m[x,y] = inf{f (t) : t ∈ [x, y]} and M[x,y] = sup{f (t) : t ∈ [x, y]}.
Clearly this implies (1) and (2).
Let x ∈ [a, b] be a point at which α0 exists and where f is continuous. We have
α(y) − α(x) F (y) − F (x) α(y) − α(x)
m[x,y] ≤ ≤ M[x,y] ,
y−x y−x y−x
Taking limit y → x on all sides, we get
lim m[x,y] α0 (x) ≤ lim F (y) − F (x) ≤ lim M[x,y] α0 (x),
y→x y→x y→x
 LECTURE NOTES ON REAL ANALYSIS I 79

We now prove that limy→x m[x,y] = limy→x M[x,y] = f (x). Let  > 0 be given. Since f is
continuous at x, there exists δ > 0 such that for all y ∈ (x − δ, x + δ), f (x) −  < f (y) <
f (x) + . Hence, for all y ∈ (x − δ, x + δ),
f (x) −  ≤ m[x,y] ≤ M[x,y] ≤ f (x) + .
Therefore for all y ∈ (x − δ, x + δ),
− ≤ m[x,y] − f (x) ≤ M[x,y] − f (x) ≤ ,
Thus
− ≤ lim m[x,y] − f (x) ≤ lim M[x,y] − f (x) ≤ ,
y→x y→x
Since  > 0 is arbitrary, we get
lim m[x,y] = lim M[x,y] = f (x).
y→x y→x


Lebesgue’s criterion for Riemann integrability.
Definition 14.14. Let S be a subset of real numbers. The set S is said to have measure
P∞  > 0, there exists countably many intervals I1 , I2 , . . . such that S ⊂
zero if for given
∞
∪i=1 Ii and i=1 length(Ii ) < .
Example 14.15. A few examples.
• A countable
 set is of measure zero. If S is a countably infinite set {a1 , a2 , . . . }.
 
Let Ii = ai − i+2 , ai + i+2 , i ≥ 1. Then
2 2
∞
X
∞
S ⊂ ∪i=1 Ii , length(Ii ) < .
i=1
Hence the set Q is rational numbers is measure zero.
The above argument shows that a finite set has measure zero.
• A subset of a measure zero set also has measure zero.
• An interval (a, b) with a < b is not measure zero. For if, (In )n≥1 cover (a, b), then
X∞
b−a≤ length(In ).
n=1

Theorem 14.16. Let I1 , I2 , . . . be countably many sets of measure zero. Then I = ∪∞

i=1 Ii
also has measure zero.
Proof. • Let  > 0 be given.
• For each i, there exists countably many intervals Ii1 , Ii2 , . . . such that
∞
∞
X 
Ii ⊂ ∪j=1 Iij , length(Iij ) < i+1 .
j=1
2
• Then {Iij }∞
i,j=1 is a collection of countably many intervals such that
∞ X
∞ ∞
X X 
I⊂ ∪∞
i=1 ∪∞
j=1 Iij , length(Iij ) ≤ i+1
< .
i=1 j=1 i=1
2

Theorem 14.17. (Lebesgue’s criterion for Riemann integrability) Let f : [a, b] → R be
a bounded function. Let D be the set of discontinuities of f in [a, b]. Then f is Riemann
integrable on [a, b] if and only if D has measure zero.
80 NIKITA AGARWAL

The set of real numbers R is uncountable.

Decimal Expansion. Every real number x can be expressed as

d1 d2 dn
x = d0 + + 2 + · · · + n + · · · = d0 .d1 d2 . . . dn . . . ,
10 10 10
where d0 ∈ Z and dn ∈ {0, 1, . . . , 9}, for all n ∈ N.
• Note that for all k ≥ 1,
 
1 1 1
9 k
+ k+1 + . . . = k−1 .
10 10 10
Hence if dk−1 6= 9, then
dk−1 9 9 dk−1 + 1
k−1
+ k + k+1 + · · · = .
10 10 10 10k−1
• All real numbers except those of the form d/10k for some d ∈ Z and k ≥ 1 have a
unique decimal expansion. These points have two decimal expansions, one ending
with infinitely many 0’s and other ending with infinitely many 9’s.
• Therefore, in representing a number in decimal expansion, we choose the con-
vention that if a number terminates in that way, we will pick the representation
which terminates in infinitely many 0’s, and not 9’s.
• For example, 1/2 can be expressed as
0.49 = 0.4999 . . . , and 0.5 = 0.5000 . . .
We choose 0.5000 . . . .
• Given a real number x, we have
x = [x] + x,
where [x] is the greatest integer less than equal to x, and x = x − [x] ∈ [0, 1) is
the fractional part of x.
Theorem 14.18. The interval [0, 1) is uncountable. Consequently R is uncountable.
Proof. Suppose (0, 1) is countable and (0, 1) = {x1 , x2 , x3 , . . . }.
Let the decimal expansions of numbers in (0, 1) is given by
x1 = 0.d11 d12 . . . d1n . . .
x2 = 0.d21 d22 . . . d2n . . .
..
.
xn = 0.dn1 dn2 . . . dnn . . .
..
.
Consider the point x ∈ (0, 1) with decimal expansion 0.b1 b2 . . . bn . . ., where bn 6= dnn and
bn ∈ {0, 1, . . . , 8}. Then x ∈ [0, 1) since
 
b1 b2 bn 1 1 1
0≤ + + ··· + n + ··· ≤ 8 + + · · · + n + . . . < 1.
10 102 10 10 102 10
This method is known as Cantor’s diagonal method. Then x 6= xn , for all n ≥ 1. Hence
we arrive at a contradiction. Thus (0, 1) is uncountable. 
Remark : Since Q is countable, the above theorem proves that R \ Q is uncountable.
 LECTURE NOTES ON REAL ANALYSIS I 81

e is irrational. We now prove that e is an irrational number. Observe that

ec
  
1 1 1
= n! e − 1 + + + · · · + = n!e − k,
n+1 1! 2! n!
where k ∈ N. Suppose e = p/q, a rational number, where p, q ∈ N, then
 c
p 1 p
= n! − k.
q n+1 q
If n ≥ q, then the RHS is a natural number. Hence LHS is a natural number for all
n ≥ q. However the LHS tends to 0 as n → ∞. This is a contradiction. Therefore e is
an irrational number.
82 NIKITA AGARWAL

15. A solved problem

(1) a) Let f : [a, b] → R be a function satisfying a uniformly Lipschitz condition of
order α > 0 on [a, b]. That is, there exists a constant L > 0 such that
|f (x) − f (y)| ≤ L|x − y|α ,
for all x, y ∈ [a, b]. If α > 1, prove that f is a constant function, whereas if α = 1,
f is of bounded variation on [a, b].
b) Give an example of a function f satisfying a uniformly Lipschitz condition of
order α < 1 on [a, b] such that f is not of bounded variation on [a, b].
c) Give an example of a function f which satisfies no uniformly Lipschitz condi-
tion on [a, b] but is of bounded variation on [a, b].

Solution: Construction of b) is skipped here. We solve a) and c).

a) Suppose α ≥ 1. Then
|f (x) − f (y)|
≤ L|x − y|α−1 ,
|x − y|
for all x, y ∈ [a, b]. Hence
(
f (x) − f (y) 0, α > 1
lim ≤ lim L|x − y|α−1 =
y→x x−y y→x L, α = 1.
Hence if α > 1, f is differentiable at x and f 0 (x) = 0, for all x ∈ [a, b]. Thus f is
a constant function. Further if α = 1, f is differentiable at x and |f 0 (x)| ≤ L, for
all x ∈ [a, b]. Hence f is a function of bounded variation on [a, b].

c) Observe that if f satisfies uniformly Lipschitz condition of order α > 0 on

[a, b], then f is (uniformly) continuous on [a, b]. For this, let  > 0 be given. Let
δ = (/L)1/α > 0. The for |x − y| < δ, x, y ∈ [a, b], we have
|f (x) − f (y)| ≤ L|x − y|α < .
Now any discontinuous function of bounded variation will work as an example for
c). For example, let c ∈ (a, b). Consider the function
(
0, a ≤ x ≤ c
f (x) =
1, c < x ≤ b.