0% found this document useful (0 votes)
37 views223 pages

Analysis

The document contains lecture notes for the MTH301A Analysis course at IIT Kanpur, referencing various texts on real analysis and topology. It outlines the syllabus, course structure, assessment methods, and philosophical inquiries related to real numbers and their properties. Key topics include completeness, metric spaces, compactness, convergence, and the completeness axiom of real numbers.

Uploaded by

harnoorsingh2509
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
37 views223 pages

Analysis

The document contains lecture notes for the MTH301A Analysis course at IIT Kanpur, referencing various texts on real analysis and topology. It outlines the syllabus, course structure, assessment methods, and philosophical inquiries related to real numbers and their properties. Key topics include completeness, metric spaces, compactness, convergence, and the completeness axiom of real numbers.

Uploaded by

harnoorsingh2509
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 223

Analysis

MTH301A

Sameer Chavan

Semester I, 2020-21
These are the lecture notes prepared for the course MTH301A1
I mostly referred in parts to the following texts:
• Carothers, N. L. Real analysis. Cambridge University Press, Cambridge, 2000. xiv+401 pp.
• Kumar, Ajit; Kumaresan, S. A basic course in real analysis. CRC Press, Boca Raton, FL, 2014. xx+302 pp.
• Kumaresan, S. Topology of metric spaces. Narosa Publishing House, New Delhi, 2005. xii+152 pp.
• M. Miklavčič, Applied Functional Analysis and Partial Differential Equations, World Scientific, Singapore,
1998.
• Pugh, Charles C. Real mathematical analysis. Undergraduate Texts in Mathematics. Springer, Cham,
2015. xi+478 pp.
• Simon, Barry Real analysis. With a 68 page companion booklet. A Comprehensive Course in Analysis, Part
1. American Mathematical Society, Providence, RI, 2015. xx+789 pp.
• Stein, Elias M.; Shakarchi, Rami Fourier analysis. An introduction. Princeton Lectures in Analysis, 1.
Princeton University Press, Princeton, NJ, 2003. xvi+311 pp
• Stein, Elias M.; Shakarchi, Rami Real analysis. Measure theory, integration, and Hilbert spaces. Princeton
Lectures in Analysis, 3. Princeton University Press, Princeton, NJ, 2005. xx+402 pp

1
The instructor of this course owns the copyright of all the course materials. This lecture material was
distributed only to the students attending the course MTH301A: Analysis-I of IIT Kanpur, and should not be
distributed in print or through electronic media without the consent of the instructor. Students can make their own
copies of the course materials for their use.
Syllabus

• R, Completeness property. Countable and Uncountable


• Metric Spaces and Metric space topology
• Nested set theorem, Baire category theorem
• Compactness, Totally bounded, Characterizations of
compactness, Finite intersection property, Continuous
functions on compact sets, Uniform continuity
• Connectedness and Path connectedness, Continuous functions
• Riemann integration, Fundamental theorem of calculus. Set of
measure zero, Cantor set, Integrable functions
• Convergence of sequence and series of functions: Pointwise
and uniform convergence of functions, Series of functions,
Power series, Dini’s theorem, Ascoli’s theorem
• Nowhere-differentiable Continuous functions, Weierstrass
approximation theorem
Course plan

• This course has been split into L + T + D.


• Lectures (3) and Tutorial (1) have been merged into 3 to 4
videos of time length 40-50 minutes every week. The links of
these videos will be shared after the videos are released.2
• There will be a Discussion hour over zoom every Friday from
12.00-12.50 (afternoon).
• About assessment/Evaluation, we opt for
• In-video questions for assignment submission (20 percent)
• Assignments/homework (40 percent)
• Online oral examination (40 percent)
• There will be mid-semester and end-semester examinations.
These will be held during the prescribed examination period.
• For students with limited or no network access, the Institute
will be making the course materials available as found feasible.
2
The course will managed on https://hello.iitk.ac.in/
Curiosity

Questions. Do you believe in real numbers ? Who are these


creatures ? Are these real or imaginary ?
B Assume the existence (be lazy be happy)
A Prove the existence (hitting against the hard)
Difficulty. If you draw a point on a plane paper, and if see it by a
magnifier, you find it to be disc-like! What/where is the point?
Philosophy Is there is a particle (point) of mass (measure) 0?
B ↓ A Understand R as a set containing all rational numbers, which
satisfies (optimal) axioms. First two building blocks:
Z = {0, ±1, ±2, . . .} and Q = {m/n : m, n ∈ Z, n 6= 0}
• Arithmetic operations of addition + and multiplication ·
• Order structure <
Axiomatic approach
Consider the set R of numbers which contains Q (and hence Z)
with arithmetic operations + and · and order structure < satisfying
• x + y = y + x for all x, y ∈ R
• (x + y ) + z = x + (y + z) for all x, y , z ∈ R
• x + 0 = x for all x ∈ R
• for every x ∈ R, there exists a unique y ∈ R (denoted by −x)
such that x + y = 0
• x · y = y · x for all x, y ∈ R
• (x · y ) · z = x · (y · z) for all x, y , z ∈ R
• x · 1 = x for all x ∈ R
• for every nonzero x ∈ R, there exists a unique y ∈ R (denoted
by x −1 or 1/x) such that x · y = 1
• (x + y ) · z = x · y + x · z for all x, y , z ∈ R
• x < y , x = y or x > y for all x, y ∈ R (exactly one possibility)
In the following, we understand that x < y iff y − x > 0. Further,
we assume that sum and product of positive numbers are positive.
Problem
Use the above axioms to verify the following:
1. If x < y and z ∈ R, then x + z < y + z.
2. If x < y and y < z, then x < z.
3. If x < y and z ∈ R, then

x · z < y · z
 if z > 0,
y ·z <x ·z if z < 0,

x ·z =y ·z if z = 0.

4. If x < y , then −y < −x.


5. Either x 2 > 0 or x = 0.
6. If 0 < x < y , then 0 < 1/y < 1/x.

Hint.
Since z − x = z − y + y − x > 0, 2 follows from 1.
Problem
Use only the last problem to verify the following:
1. Given two real numbers x, z such that x < z, there exists
y ∈ R such that x < y < z.
2. If x 6 y + z for all z > 0, then x 6 y .
3. If 0 < x < y , then for any positive integer n, 0 < x n < y n .

Hint.
x+z
For 1, try y = 2 . For 2, prove by contradiction.

• Q satisfies all the axioms mentioned before the last problem.


Question. How do we differentiate R from Q? Is there a property
(of R) not enjoyed by Q ?
Upper bounds

Definition
Let ∅ =
6 A ⊆ R. We say that A is bounded from above if there is
α ∈ R (an upper bound) such that a 6 α for every a ∈ A.
Problem
Show that any finite nonempty subset A of R is bounded with
upper an bound belonging to it. Give an infinite subset of R,
where this fails.
Hint.
This can be proved by the induction on card(A). If card(A) = 1,
then A = {a} for some a ∈ A, and a is an upper bound. Suppose
that the conclusion holds for all sets A with card(A) = n, and let B
be a subset of R such that card(B) = n + 1. Since B 6= ∅, there
exists b ∈ B. Apply the induction hypothesis to B \ {b} and use
the law of trichotomy. For last part, take A = (0, 1). None of the
upper bounds of A belongs to A.
Least upper bound
Definition
Let ∅ =
6 A ⊆ R. We say that α is a least upper bound (lub) for A if
• α is an upper bound for A, that is, a 6 α for all a ∈ A, and
• β is an upper bound for A, then α 6 β.
Remark. lub is unique. If a set contains its upper bound (for
example, (0, 1]), then maximum of that set is its lub.
Problem
Show that α ∈ R is lub for A ⊆ R iff α is an upper bound for A
and if β < α, then there exists a ∈ A such that β < a 6 α.
Hint.
If β < α, then β is not an upper bound of A.
Example
1 is the lub for A = (0, 1). Indeed, if α < 1 is the lub, then
a := 1+α
2 ∈ (α, 1), so α is not an upper bound for (0, 1).
Unboundedness of integers

It is tempting to conclude that Z is not bounded from above


without a proof.
Difficulty. If Z is bounded above by x ∈ R how to ensure that
there is an integer n > x (this problem does not occur if the upper
bound is an integer) ?
Solution. If you can not prove an assertion, convert it into a
property (with no disrespect for Archimedes)!
Archimedean property. Given x, y ∈ R, with x > 0, there exists
a positive integer n such that nx > y .
Interpretation. Howsoever tiny may be your foot-step x, you can
cover any finite distance y (for example, the distance between
earth and moon) in finitely many steps n.
LUB property or completeness axiom
• Any nonempty subset of R, which is bounded above in R, has
lub in R.
This is the property which differentiates R from Q:
Example
The subset A = {a ∈ Q : a2 < 2} is nonempty and bounded above
(1 ∈ A and 2 is an upper bound for A, but it does not have lub in
Q. To see the latter statement, assume that α is an lub of A. If
α2 < 2, then for any positive integer k,

(α + 1/k)2 = α2 + 2α/k + 1/k 2 6 α2 + 5/k,

so that (α + 1/k)2 < 2 for sufficiently large k (we need here the
fact that Z is not bounded from above). This however implies that
α + 1/k ∈ A, contradicting the assumption that α is lub. Thus
α2 ≮ 2. Similarly, one can see that α2 ≯ 2. Thus α2 = 2 or α ∈ / Q.
LUB property and Archimedean property

Theorem
LUB property =⇒ Archimedean property.

Proof.
Assume the LUB property, Suppose that the Archimedean property
fails, that is, there exist x, y with x > 0 such that nx 6 y for every
positive integer n. Thus n 6 y /x for every (positive) integer n.
Thus Z is bounded above and hence by LUB property, it has lub,
say, α. Now α − 1 is not an upper bound, so for some integer N,
α − 1 < N. However, this implies that α < N + 1 and N + 1 is an
integer, which contradicts the assumption that α is lub.

Problem
Describe ∩n>1 (0, 1/n) and ∩n>1 (n, ∞).
Density of rationals and irrationals in R
Problem (Greatest Integer Function)
Assume the LUB property and let x ∈ R. Show that there exists a
(unique) m ∈ Z such that m 6 x < m + 1 (m is denoted by [x]).
Hint.
Consider the nonempty set A = {k ∈ Z : k 6 x}, which is bounded
above. Let α = lub(A). Since α − 1 is not an upper bound, there
exists m ∈ A such that α − 1 < m 6 x. Verify that x < m + 1.
Theorem
Assume the LUB property and let a, b ∈ R be such that a < b.
Then there exists r ∈ Q (resp. r ∈ R\Q) such that a < r < b.
Proof of Theorem.
By Archimedean property, we find an integer n > 1 such that
n(b − a) > 1. By the last problem, there exists m ∈ Z such that
na < m 6 na + 1 (take m = [na] + 1). Thus
√ na < m √ < nb or
r = m/n. For the rest, apply this to a − 2 and b − 2.
Nested Interval Theorem
Theorem
Assume the LUB property and let Jn = [an , bn ] be an interval in R
such that Jn+1 ⊆ Jn for all integers n > 1. Then ∩n>1 Jn 6= ∅.
Proof.
Consider the set A = {x ∈ R : x = an for some n > 1} and note
that A is nonempty and bounded above. Let α = lub(A). Since
each bn is an upper bound for A (since Jn+1 ⊆ Jn for all integers
n > 1), we have an 6 α 6 bn , that is, α ∈ ∩n>1 Jn .
Problem
Assume the LUB (and GLB) property. Show that any bounded
infinite subset A of R has at least one accumulation point.
Hint.
Suppose A ⊆ [a1 , b1 ] with a1 = glb(A), b1 = lub(A). Let [a2 , b2 ]
be one of the intervals from [a1 , (a1 + b1 )/2] and [(a1 + b1 )/2, b1 ]
which contains infinitely many points from A. Continue this.
Complete ordered field
We may understand now R as the ”complete ordered field” (the
set R satisfying all the axioms mentioned earlier including the
completeness axiom).
Question. Does a complete ordered field exist ? If yes, is it unique
(up to isomorphism) ?
To address the issue of the existence of R, we mention two
approaches; one due to Dedekind (based on ”Dedekind cuts”) and
other due to Cantor (based on the completion of Q). We follow
the second approach and for this, we need to know the
convergence of sequences in R.
Recall the definitions of Cauchy and convergent sequences in R.
• Cauchy sequence is bounded and convergent sequence is a
Cauchy sequence.
• Every Cauchy sequence with convergent subsequence is
convergent.
Cauchy Completeness of R

Theorem
Assume the LUB property. Then R is Cauchy complete, that is,
every Cauchy sequence in R is convergent.

Proof I.
Let A = {x ∈ R : there exists N > 1 such that x < xn for n > N}
and let  > 0. Note that there exists n0 > 1 (dependent on ) such
that xn0 − /2 ∈ A (since {xn }n>1 is a Cauchy sequence). Thus A
is a nonempty set. Further, A is bounded above by xn0 + /2
(verify by contradiction). Let α = lub(A). Since xn0 − /2 ∈ A and
A is bounded above by xn0 + /2, |xn0 − α| 6 /2. Thus for n > n0 ,

|xn − α| 6 |xn − xn0 | + |xn0 − α| < .

This completes the proof.


It is possible to obtain the Cauchy completeness without using the
LUB property.
Proof II.
Let A = {x ∈ R : x = xn for some n > 1}. Since A is bounded,
A ⊆ [a1 , b1 ] for some a1 , b1 ∈ R Let [a2 , b2 ] be one of the intervals
from [a1 , (a1 + b1 )/2] and [(a1 + b1 )/2, b1 ] which contains
infinitely many points from A. Continue this to obtain the intervals
[an , bn ], n > 1, containing infinitely many points of A. By the
nested interval theorem, ∩n>1 [an , bn ] contains x for some x ∈ R.
Now for each n > 1, if we choose xkn ∈ [an , bn ] such that
kn 6 kn+1 , we obtain a subsequence {xkn }n>1 such that
|xkn − x| 6 bn − an = (b1 − a1 )/2n → 0 as n → ∞. Since {xn }n>1
is a Cauchy sequence, it must be convergent now.
The proof above yields the following:
Theorem (Bolzano-Weierstrass Theorem)
Every bounded sequences admits a convergent subsequence.
Cauchy’s Construction of R (Incomplete)

Let R be the set of Cauchy sequences in Q. Let {xn }n>1 , {yn }n>1
belong to R and define the equivalence relation

{xn }n>1 ∼ {yn }n>1 if |xn − yn | → 0.

• Define the set R as R/ ∼ (set of equivalence classes [{xn }n>1 ]).


• Identify the element x in Q with constant sequence {xn = x}n>1
• {xn }n>1 < {yn }n>1 if {yn − xn }n>1 eventually consists of positive
numbers.
• It is a laborious work to verify that R defines an ordered field
which satisfies LUB property. Refer to the article http://www.
math.ucsd.edu/~tkemp/140A/Construction.of.R.pdf) 3 .

3
Cauchy’s Construction of R, Todd Kemp
Theorem (Density of rationals)
If r = {rn }n>1 ∈ R and any small (rational) number , then there
exists a rational number q such that |r − q| < , that is, for some
integer N > 1, |rn − q| <  for all n > N.

Proof.
Since {rn }n>1 is a Cauchy sequence, for some N > 1, |rn − rN | < 
for all n > N. Let q = rN ∈ Q.
Let R = {[{xn }n>1 ] : {xn }n>1 ∈ R} and define a metric on R by

d([{xn }], [{yn }]) = lim |xn − yn | (an equivalence class)


n→∞

(the limit exists since {xn − yn }n>1 is a Cauchy sequence and hence
convergent). Then R is a complete metric space, and if i : Q → R
is given by i({x}) = [{x}] for every x ∈ Q, then i is injective.
4
Countable and Uncountable sets

Definition
A set A is called countably infinite if there is a bijection f : N → A.
If A is finite or countably infinite, we say that A is countable.

Example
• Any subset A of N is countable (either A is finite or
A = {nk }k>1 ; in the latter case, define f (k) = nk )
• Any subset of a countable set is countable
• Z is countable (define g : Z → N by g (n) = 2n if n > 1, and
g (n) = −2n + 1 if n 6 0; now let f = g −1 )
• N × N is countable (define g (m, n) = 2m−1 (2n − 1); let f = g −1 )
• Countable union of countable sets is countable (if
A = {xm,n : m ∈ N, n ∈ N}, then g : N × N → A given by
f (m, n) = am,n is bijective)
4
Cantor is a ‘corrupter of youth’. - L. Kronecker, as quoted by Schoenflies and requoted by Barry Simon
Problem
Verify the following:
1. If g : A → B is bijective, show that A is countable iff so is B.
2. If g : A → B is injective, show that A is countable if so is B.

Hint.
1. If h : N → A is bijective, then so is g ◦ h : N → B.
2. If h : N → g (B) is bijective, then so is g −1 ◦ h : N → A.

Problem
Show that Q is countably infinite.

Hint.
Write Q = Q+ ∪ Q− ∪ {0} (union of positive/negative/zero
rationals). To see that Q+ is countable, define f : Q+ → N × N by
f (m/n) = (m, n), where m and n are coprime to each other. Note
that f is injective and N × N is countable.
Problem
For an index set I , let {Jα }α∈I be a collection of disjoint open
intervals such that R = ∪α∈I Jα . Show that I is countable.

Solution.
We know that each interval Jα contains a rational number rα .
Define g : I → Q by g (α) = rα . If, for some α, β ∈ I ,
g (α) = g (β), then rα = rβ , and hence α = β (if α 6= β, then
Jα ∩ Jβ = ∅). This shows that g is injective. Since Q is countable,
by the problem on the previous slide, I is also countable.

Problem
For an index set I , let {(xα }α∈I be a collection of mutually
orthonormal vectors in an inner-product space. X . If X has a
countable dense subset, then show that I is countable.
Discontinuities of a monotone function
For a function f : (0, 1) → R and c ∈ (0, 1), let f (c− ) and f (c+ )
denote the left and right hand side limits of f at c, respectively.
Theorem
Let f : (a, b) → R be monotone. Then the set D of discontinuities
of f is countable.
Proof.
Without loss of generality, we may assume that f is increasing.
Note that D = {c ∈ (a, b) : Ic = (f (c− ), f (c+ )) 6= ∅}. For
c, d ∈ D, let t ∈ [a, b] be such that c < t < d. Then

f (c+ ) = glb{f (x) : x > c} 6 f (t) 6 f (d− ) = lub{f (y ) : y < d}.

This shows that Ic ∩ Id = ∅. Since each interval Ic contains a


rational and rationals are countable, D is countable.
Problem
Show that a monotone surjection f : [a, b] → [c, d] is continuous.
Cantor’s diagonalization theorem

Theorem
There is no surjection f : N → [0, 1].

Proof.
Consider the decimal expansion (possibly more than one) of f (n):

f (n) = 0 · an1 an2 . . . ann . . . , n>1

(for the existence of decimal expansion, refer to Proposition 1.8 of


[Real Analysis, Carothers, N. L.]). Let x = 0 · a1 a2 . . . an . . . ∈ R
where an ∈ {1, . . . , 8}\{ann }. Note that x 6= f (n) for any n > 1,
and hence x ∈ / f (N).
Corollary (Cantor)
The interval [0, 1] in R is not countable.

Problem
Given two points in R × R \ Q × Q, show that there exists
uncountably many paths (finite union of line segments) in
R × R \ Q × Q, which join these two points.

Hint.
Fix two points A and B in the plane and draw a square with one of
the diagonals being the straight line C joining these two points.
Consider the diagonal D orthogonal to L and note that it is not
countable. Now produce one path corresponding to each point on
D which joins A and B. How many of these touch Q × Q ?
Power set of N

Example
Consider the power set P(N) of all subsets of N. Let f : N → P(N)
be any function. We claim that f is not surjective. Indeed, define
A ∈ P(N) by the property that

k ∈ A ⇐⇒ k ∈
/ f (k).

Note that A 6= f (k) for any k > 1, and hence the claim stands
verified. In particular, P(N) is not countable.

Problem
Let A be a nonempty set. There is no map from A into the power
set P(A) of A, which is surjective.
Cantor set

Example
Let C denote the Cantor set obtained by removing 2n−1 centrally
situated disjoint open subintervals U1,n , · · · , U2n−1 ,n of [0, 1] each
of length 1/3n at the nth stage, where n = 1, 2, · · · . Specifically, if
U1,1 = (1/3, 2/3), U2,1 = (1/9, 2/9), U2,2 = (7/9, 8/9), . . . , Un,1 =
(1/3n , 2/3n ), . . . , Un,2n−1 = (1 − 2/3n , 1 − 1/3n ), then
n−1
where Cn = [0, 1] \ ∪nk=1 ∪2j=1 Uk,j .

C = ∩n>1 Cn ,

Clearly, all end-points 1/3, 2/3, 1/9, 2/9, 7/9, 8/9, . . . of Uj,k
belong to ∆. Are there any other points in C ?
Answer Yes, take 1/4 (since 1/4 ∈
/ Uk,j for all j, k)
Question What is the trick to generate more elements in C ?
Example (continued ...)
• P
Any x ∈ [0, 1] has two distinct representations of the form
∞ n
n=1 an /3 with an ∈ {0, 1, 2} if and only if x belongs to the
set {1/3, 2/3, 1/9, 2/9, 7/9, 8/9, . . .}.
• Any x ∈ C has unique representation ∞ n
P
n=1 an /3 , where
an ∈ {0, 2}. (For instance, 1/3 = 0.1 = 0.0222 . . .).
Let {0, 2}N denote the collection of sequences
P∞ in 0 and 2. Define
the map f : {0, 2} → C by f (an ) = n=1 an /3n . By the above
N

facts, f is a bijection. The following shows that C is not countable.


Problem
There exists a bijection between {0, 2}N and P(N).

Hint.
For a subset A of N, define a sequence {an }n>1 by
(
2 if n ∈ A,
an =
0 if n ∈
/ A.
Metric spaces

Definition
A metric d on a nonempty set X (refer as the metric space) is a
map d : X × X → [0, ∞) that obeys
• (Symmetry) For all x, y ∈ X , d(x, y ) = d(y , x)
• (Triangle inequality) For all x, y , z ∈ X ,
d(x, z) 6 d(x, y ) + d(y , z)
• (Zero property) For all x, y ∈ X , d(x, y ) = 0 iff x = y .

Interpretation
• The distance of x from y is same as the distance of y from x.
• If we consider the triangle 4 with vertices x, y , z and lenghts
of sides being d(x, y ), d(y , z) and d(x, z), then the length of
any side is at most the sum of lengths of remaining two sides.
• The distance of a point from itself is 0, and if the distance
between two points is 0, then these points must coincide.
Real line as a metric space

Example
R is a metric space with metric d(x, y ) = |x − y | for x, y ∈ R. The
symmetry and zero property follow immediately from the facts that
|x| = | − x| and |x| = 0 iff x = 0. To see the triangle inequality,
note that since x 6 |x|, we have
|x − z|2 = (x − y + y − z)2 = (x − y )2 + 2(x − y )(y − z) + (y − z)2
6 (|x − y | + |y − z|)2 . Now take square roots on both sides.

Remark
• One may give an alternate proof of the triangle inequality
which does not involve square root (consider the cases in
which x + y > 0 and x + y < 0, and use the definition of | · |).
• Note that |x + y | = |x| + |y | iff either x and y are both
nonnegative or both nonpositive (Exercise).
Young’s Inequality
Theorem
Let p, q > 1 be conjugate exponents, that is, 1/p + 1/q = 1. For
p q
positive numbers a, b, ab ≤ ap + bq . Equality holds iff ap = b q .
Proof.
Given positive real numbers a ≤ b, consider

D1 = {(x, y ) ∈ R2 : 0 ≤ x ≤ a, 0 ≤ y ≤ x p−1 },

D2 = {(x, y ) ∈ R2 : 0 ≤ y ≤ b, 0 ≤ x ≤ y q−1 }.
We verify the following (see [page 43, Carothers] for a figure):
1. D1 ∩ D2 = {(x, y ) ∈ R2 : 0 ≤ x ≤ a , y = x p−1 }. Indeed,
since (p − 1)(q − 1) = 1, y = x p−1 iff x = y q−1 .
2. R = {(x, y ) ∈ R : 0 ≤ x ≤ a, 0 ≤ y ≤ b} ⊆ D1 ∪ D2 .
Thus Area(R) = ab ≤ Area(D1 ∪ D2 ) 6 Area(D1 ) + Area(D2 ) =
ap /p + b q /q. Moreover, equality holds iff ap = b q .
Hölder’s Inequality
Corollary
Let p, q > 1 be conjugate exponents, that is, p1 + q1 = 1. Let
a = (a1 , . . . , ad ), b = (b1 , . . . , bd ) ∈ Rd = {(x1 , . . . , xd ) :
x1 , . . . , xd ∈ R}. Then
d d
!1/p d
!1/q
X X X
p q
an bn ≤ |an | |bn | .
n=1 n=1 n=1

Proof. P 1/p
d p
Let kakp := n=1 |an | and ãn = |an |/kakp , b̃n = |bn |/kbkq ,
n = 1, . . . , d. By Young’s Inequality, |ãn ||b̃n | ≤ |ãn |p /p + |b̃n |q /q.
d k
1 X X
Thus |an ||bn | = |ãn ||b̃n |
kakp kbkq
d d n=1 n=1
X
p
X
q 1 1
≤ |ãn | /p + |b̃n | /q = + = 1.
p q
n=1 n=1
Example
For a positive integer d > 1, Rd = {(x1 , . . . , xd ) : x1 , . . . , xd ∈ R}
is a metric space with metric
v
u d
uX
d(x, y ) = t |xj − yj |2 , x, y ∈ Rd .
j=1

To see the triangle inequality, it suffices to check that

ka + bk2 6 kak2 + kbk2 , a, b ∈ Rd ,


P 1/2
d 2
where kak2 := n=1 |an | . To see this, note that by Hölder’s
Inequality (with p = 2 = q),
Xd X d d
X
ka + bk22 = |aj |2 + 2 aj b j + |bj |2
j=1 i,j=1 j=1

6 kak22 + 2kak2 kbk2 + kbk22 = (kak2 + kbk2 )2 .


Problem
For a positive integer d > 1 and p > 1, show that Rd is a metric
space with metric dp given by
d
X 1/p
dp (x, y ) = |xj − yj |p , x, y ∈ Rd .
j=1

This result fails for 0 < p < 1.

Hint.
To see that ka + bkp 6 kakp + kbkp , note that
d
X d
X d
X
|aj + bj |p 6 |aj ||aj + bj |p−1 + |bj ||aj + bj |p−1 ,
j=1 j=1 j=1

and apply Hölder’s inequality (two times). To see the second part,
let d = 2, a = (1, 0), b = (0, 1) and p = 1/2.
Problem
For a positive integer d > 1, show that Rd is a metric space with
metric d∞ given by d∞ (x, y ) = maxdj=1 |xj − yj | for x, y ∈ Rd .

Problem
For p > 1 and for a sequence {xn }n>1 of real numbers, define
P 1/p
∞ p
kxkp = j=1 |xj | ∈ [0, ∞]. Verify the following:
1. `p = {{xn }n>1 : kxkp < ∞} is a vector space over R.
2. `p is a metric space with metric dp given by

X 1/p
dp (x, y ) = |xj − yj |p , x, y ∈ `p .
j=1
Proof.
Both parts essentially follow from the triangle inequality. We
already know the conclusion for truncated sequences {an }N k=1 and
{bn }N
k=1 for every integer N > 1. Now let N → ∞.
Let a, b ∈ R be such that a < b.
Lemma
If f : [a, b] → R is continuous, then sup |f (x)| < ∞.
x∈[a,b]

Proof.
Assume that supx∈[a,b] |f (x)| = ∞. Then, for any integer n > 1,
there exists xn ∈ [a, b] such that |f (xn )| > n. By
Bolzano-Weierstrass theorem, {xn }n>1 admits a convergent
subsequence {xnk }k>1 , and hence by (sequential) continuity of f ,
{f (xnk }k>1 is convergent. However, a convergent sequence is
bounded, and on the other hand, |f (xnk )| > nk . Not possible!
Example
Consider the vector space C [a, b] of real-valued continuous
functions defined on [a, b]. Then C [a, b] is a metric space with
metric d∞ given by

d∞ (f , g ) = sup |f (x) − g (x)|. f , g ∈ C [a, b].


x∈[a,b]
Metric topology

Let X be a metric space with metric d (denoted by (X , d)).


Question Given two metric spaces (X , dX ) and (Y , dY ), when can
we say that these are equivalent/isomorphic metric spaces ?
Clue We know that two sets X and Y are equivalent if there is a
bijection between X and X 0 . Similarly, two vector spaces X and Y
are isomorphic if there is a bijective linear map between X and Y .
Guess So, for two metric spaces to be equivalent, we need to
impose some constraint on the given bijection that will take the
metric structures into account (for instance, sequential continuity).
Question What do we mean by metric structures ?
Clue The metric d is the distance function, which measures
distance between two points. One may look for a metric analog of
a disc/ball in the metric space.
Metric balls
Let (X , d) be a metric space. For x0 ∈ X and a real number r > 0,
define the open ball and closed ball around x0 , respectively, by
Br (x0 ) = {x ∈ X : d(x, x0 ) < r }, Br (x0 ) = {x ∈ X : d(x, x0 ) 6 r }.

Example (Discrete Geometry)


Let X be a nonempty set. For every x, y ∈ X , define the
discrete metric d0 by d0 (x, x) = 0 and d0 (x, y ) = 1 if x 6= y . Then
( (
{x0 } if r 6 1, {x0 } if r < 1,
Br (x0 ) = Br (x0 ) =
X otherwise , X otherwise .

The above calculations may be applied to X = R.


Example (One dimensional geometry)
Let R be the real line with d1 (x, y ) = |x − y |, x, y ∈ R. Then

Br (x0 ) = (x0 − r , x0 + r ), Br (x0 ) = [x0 − r , x0 + r ].


Example (Plane geometry)
p
Let R2 be the real plane with d2 (x, y ) = |x1 − y1 |2 + |x2 − y2 |2 ,
x = (x1 , x2 ), y = (y1 , y2 ) ∈ R2 . If x0 = (x01 , x02 ), then

Br (x0 ) = {x ∈ R2 : |x1 − x01 |2 + |x2 − x02 |2 < r 2 } (circular region),

Br (x0 ) = {x ∈ R2 : |x1 − x01 |2 + |x2 − x02 |2 6 r 2 }.


What if d2 is replaced by d∞ (x, y ) = max{|x1 − y1 |, |x2 − y2 |} ?

Br (x0 ) = {x ∈ R2 : |x1 − x01 | < r , |x2 − x02 | < r } (square region),

Br (x0 ) = {x ∈ R2 : |x1 − x01 | 6 r , |x2 − x02 | 6 r }.


What if d2 is replaced by d1 (x, y ) = |x1 − y1 | + |x2 − y2 | ? Why
this metric is known as taxi-cab metric ?

Notation Although the notations Br (x0 ), Br (x0 ) do not indicate


dependence on the underlying metric, the dependence is evident.
Problem
Consider the metric space C [0, 1] with the sup metric
d∞ (f , g ) = supx∈[0,1] |f (x) − g (x)|, f , g ∈ C [0, 1]. Describe the
open and closed balls Br (f ), Br (f ), where f (x) = 0, x and x 2 .

Hint.
Note that
n o
Br (f ) = g ∈ C [0, 1] : sup |g (x) − f (x)| < r
x∈[0,1]

= g ∈ C [0, 1] : |g (x) − f (x)| < r for every x ∈ [0, 1]

= {g ∈ C [0, 1] : g (x) ∈ (f (x) − r , f (x) + r ) for every x ∈ [0, 1]}.

Now draw the graphs of f (x) − r and f (x) + r in the above cases,
and convince yourself that the ”tube” enclosed between these two
graphs is the open ball Br (f ). Draw the diagram.
In the following problem, we need the fact that every continuous
function on [0, 1] is Riemann integrable.
Problem
Consider the
R metric space C [0, 1] with the metric
d(f , g ) = x∈[0,1] |f (x) − g (x)|dx, f , g ∈ C [0, 1].
1. Describe the open ball B1 (0).
2. Give an example of f belonging to B1 (0) and also an example
belonging to C [0, 1] \ B1 (0).
3. Verify that Br (f ) = {f + rg : g ∈ B1 (0)}.
4. Give an example of g belonging to Br (f ) and also an example
belonging to C [0, 1] \ Br (f ).

Hint.
Note that f ∈ B1 (0) if and only if the area under the graph of
|f (x)| is less than 1.
Problem
Let (X , d) be a metric space and let A be a subset of X . Verify the
following:
1. A is a metric with metric dA given by dA (a, b) = d(a, b) for
all a, b ∈ A.
2. Describe the open ball in the metric space (A, dA ).
3. If (X , d) = (R, d1 ) (resp. (R2 , d2 )) and A = Z (resp. Z × Z),
then describe dA and the open ball in the metric space (A, dA ).

Hint.
For the second part, verify that the open ball BrA (a) centred at a
and of radius r in the dA metric equals the intersection of A with
the open ball Br (a). For the last part, draw the lattice Z × Z and
examine its intersection with open balls.
Hausdorff property

Theorem
Let (X , d) be a metric space and let x, y be distinct points in X .
Then there exist positive numbers r , s such that Br (x) ∩ Bs (y ) = ∅.

Proof.
Since x 6= y , d(x, y ) > 0. Let r = d(x, y )/2 = s. We must check
that Br (x) ∩ Bs (y ) = ∅, or equivalently,

Br (x) ⊆ X \ Bs (y ) = {z ∈ X : d(z, y ) > s}.

If z ∈ Br (x), then by the symmetry and the triangle inequality,

d(z, y ) > d(x, y ) − d(x, z) > d(x, y ) − r = s,

since r + s = d(x, y ).
There exist ”topological spaces” without the Hausdorff property!
Open and closed sets

Definition
Let (X , d) be a metric space and let A be a subset of X . We say
that A is open in X if either A is an empty set or if for every
a ∈ A, there exists  > 0 such that B (a) ⊆ A. We say that A is
closed if the complement of A in X is open.
• Both the empty set ∅ and X are open, and hence closed.
• The open ball Br (x0 ) is open. To see this, let a ∈ Br (x0 ).
Thus d(a, x0 ) = r − δ for some δ > 0. We must find an  > 0
such that B (a) ⊆ A = Br (x0 ). For any b ∈ B (a), by the
triangle inequality,

d(b, x0 ) 6 d(b, a) + d(a, x0 ) 6  + r − δ,

which is less than r provided  < δ.


Problem
Show that the closed ball Br (x0 ) in a metric space (X , d) is closed.

Hint.
Note that X \ Br (x0 ) = {x ∈ X : d(x, x0 ) > r }. Let a ∈ X \ Br (x0 )
and write δ = d(x, x0 ) − r > 0. Use the triangle inequality to find
conditions on  > 0 (in terms of δ), so that
B (a) ⊆ A = X \ Br (x0 ).

The closed ball could be an open set in some metric spaces!


Example (Discrete topology)
Consider R enodwed with the discrete metric d0 given by
d0 (x, x) = 0 and d0 (x, y ) = 1 if x 6= y ∈ R. Since single-tons are
open balls ({x0 } = B1 (x0 )), any subset of R is open!

This can not happen in Rd endowed with the metric dp ,


1 6 p 6 ∞ (to be seen later).
Problem
Answer the following:
1. Which of the sets are open/closed in R with metric d1
x, y ∈ R ?
1.1 (0, 1), [0, 1), (0, 1], [0, 1], Z, Q
2. Which of the sets are open/closed in R2 with metric d2 ?
(0, 1) × (0, 1), [0, 1) × [0, 1), (0, 1] × [0, 1), [0, 1] × [0, 1], Z × Z,
Q × Q, {(x, y ) ∈ R2 : x 2 + y 2 = 1}, {(x, y ) ∈ R2 : xy 6= 0}
Justify your answer.

Hint.
The answers in 1 are open, not open and not closed, not open and
not closed, closed, closed, not open and not closed (in order).
The answers in 2 are open, not open and not closed, not open and
not closed, closed, closed, not open and not closed, closed, open
(in order).
Let B[0, 1] denote the metric space of bounded functions
f : [0, 1] → R with the metric d∞ given by

d∞ (f , g ) = sup |f (x) − g (x)|, f , g ∈ B[0, 1].


x∈[0,1]

Problem
Whether the subset C [0, 1] of B[0, 1] is open in B[0, 1] ? Justify
your answer.

Hint.
No. To see this, given any  > 0, find f ∈ B[0, 1] \ C [0, 1] such
that f ∈ B (0) (-neighbourhood of the 0 function), that is, a
bounded function f discontinuous at least at a point in [0, 1] such
that kf k∞ < .

Problem
Let (X , d) be a metric space. For a ∈ X and r > 0, show that
Sr (a) = {x ∈ X : d(x, a) = r } is closed in X .
Let (X , d) be a metric space and let O be the collection of all
open subsets of X .
• O is nonempty: ∅ and X belong to O
• O is closed under arbitrary (finite or infinite) union: If {Uα }
⊆ O, then ∪α Uα is open (if a ∈ ∪α Uα , then a ∈ Uα for some
α, and hence there is r > 0 such that Br (a) ⊆ Uα ⊆ ∪α Uα )
• O need not be closed under countable intersection: Consider
X = R with usual metric d1 (x, y ) = |x − y |, x, y ∈ R. Then
for Un = (−1/n, 1/n) ∈ O, ∩n>1 Un = {0} ∈ /O
• O is closed under finite intersection: If U1 , . . . , Uk ∈ O, then
∩kj=1 Uj ∈ O (if a ∈ ∩kj=1 Uj , then for some rj > 0, Brj (a) ⊆ Uj ,
and hence Br (a) ⊆ ∩kj=1 Uj with r = min{r1 , . . . , rk })

Theorem
For any metric space (X , d), the collection O (known as topology
of X ) of open subsets of X contains empty set and X . Further, O
is closed under arbitrary union and finite intersection.
Open sets in R

Problem
Let U be an open set in R and let x ∈ U. Show that there exists
largest open interval containing x and contained in U.

Hint.
Let ax = −∞ if {a ∈ R : a < x, (a, x) ⊆ U} is not bounded
below, and bx = ∞ if {b ∈ R : b > x, (x, b) ⊆ U} is not bounded
above. Otherwise, set
ax = glb{a ∈ R : a < x, (a, x) ⊆ U},
bx = lub{b ∈ R : b > x, (x, b) ⊆ U},
and consider the open interval Ix = (ax , bx ). Then x ∈ Ix
(otherwise either x 6 ax or x > bx ). We claim that the interval Ix
is contained in U. If Ix 6⊆ U, then there is y ∈ Ix \ U, and hence
either (x, y ) or (y , x) intersects the complement of U, and hence
bx 6 y or ax > y . That’s the contradiction since y ∈ Ix .
Theorem
Any open set U in R can be expressed as the disjoint union of
countably many open intervals.

Proof.
Let Ix denote the largest open interval containing x and contained
in U. Then U = ∪x∈U Ix . If x, y ∈ U, then either Ix ∩ Iy = ∅ or
Ix = Iy . Since rationals are countable and distinct intervals
contains distinct rationals, the collection {Ix }x∈U contains at most
countably many disjoint intervals whose union is U.
The role of an open interval in R may be replaced by open
rectangles (a, b) × (c, d) in R2 .
Question Is it possible to express an open subset of R2 as disjoint
union of countably many open rectangles ?
Open sets in R2

Example
Consider the open unit disc in R2 :

D = {(x, y ) ∈ R2 : x 2 + y 2 < 1}.

We claim that D can not be written as disjoint union of countably


many open rectangles. Indeed, if D = ∪n>1 Rn is union of open
rectangles Rn = (an , bn ) × (cn , dn ) ⊆ D, then at least some point of
the form (an , y ), cn < y < dn will lie in D, and hence (an , y ) must
lie in some open rectangle Rm , m 6= n. In that case, Rn ∩ Rm 6= ∅.

• Major difference between R and R2 is the“natural” order


structure (and LUB property).
Interior of a set

Let (X , d) be a metric space.


• Given the closed ball Br (a) = {x ∈ X : d(x, a) 6 r } in X , we
can distinguish the points lying in Br (a) = {x ∈ X : d(x, a) < r }
from the remaining points {x ∈ X : d(x, a) = r }.
Question Can we do the same for an arbitrary set ?

Definition
Let A be a subset of X and let a ∈ A. We say that a is an
interior point of A if there exists r > 0 such that Br (a) ⊆ A.

Let A◦ = {a ∈ A : a is an interior point of A}.


• Note that A is open in X if and only if A◦ = A.
Let (X , d) be a metric space and let a ∈ X . We know that any
point in Br (a) is an interior point of

Br (a) = {x ∈ X : d(x, a) 6 r }.
Example
In general, the interior of Br (a) need not be Br (a). Indeed, if
x ∈ Br (a) \ Br (a), then x could be an interior point of Br (a). To
see this, let X = R with discrete metric d0 and a = 0, r = 1. Then
B1 (0) = {0} and the interior of B1 (0) = R is R.

This does not happen in X = Rd endowed with any metric dp ,


p > 1, as the following example illustrates.
Example
In (Rd , dp ) the interior of Br (a) is Br (a). We have seen that the
Br (a) is contained in the interior of Br (a). Let b ∈ Br (a) be such
that dp (b, a) = r . We claim that for any s > 0, the ball Bs (b)
intersects Rd \ Br (a).
The line joining a and b is given by a + t(b − a), t > 0. Note that
d
X 1/p
dp (a + t(b − a), b) = |1 − t|p |aj − bj |p ,
j=1

which can be made less than s if |1 − t| is very small. Thus


a + t(b − a) belongs to Bs (b) if |1 − t| is very small. Also, if t > 1,
d
X p
dp (a + t(b − a), a) = t p |aj − bj |p > dp (b, a) = r .
j=1
A proper subset of Q which is open as well as closed

Example
where d1 (x, y ) = |x − y |,
Consider the metric space (Q, d1 ), √
x, y ∈ Q. Let ζ ∈ R \ Q (e.g. ζ = 2) and let
A = {x ∈ Q : x < ζ}. We claim that A is open and closed in Q.
• A is open since A = (−∞, ζ) ∩ Q and (−∞, ζ) is open in R
• A is closed since Q \ A = {x ∈ Q : x > ζ} = {x ∈ Q : x > ζ},
which is again open since it is equal to Q ∩ (ζ, ∞)

The phenomenon above can not occur in R ! This means that


there are no proper open and closed subsets of R (to be seen later).
Problem
Find a proper subset of Q × Q with the metric d2 (x, y ) = kx − y k2
(x, y ∈ Q × Q), which is open and closed.
Limit point of a set

Let (X , d) be a metric space.


Definition
Let A ⊆ X and let a ∈ X . We say that a is a
• limit point5 of A if for every r > 0, Br (a) ∩ A 6= ∅.
• cluster point of A if for every r > 0, Br (a) ∩ (A\{a}) 6= ∅.

• A limit point need not belong to A (0 is a limit point of (0, 1))


• a is not a limit point iff there exists some r > 0 such that
Br (a) ∩ A = ∅ (2 is not a limit point of (0, 1) for B1 (2) ∩ (0, 1) = ∅)
• An interior point is a limit point (this follows from definition)
• A cluster point of A is a limit point of A, but not conversely (e.g.
(R, d0 ), 0 is a limit point of A = {0} but not a cluster point)
5
Some authors define limit point as the point for which Br (a) ∩ (A\{a}) 6= ∅
Example
Let X = R and let d1 (x, y ) = |x − y |, x, y ∈ R. Let A denote the
set of limit points and let cl(A) be the set of cluster points of A:
• If A = [a, b), then A = [a, b] = cl(A)
• If A = (a, b] ∪ {a + b}, then A = [a, b] ∪ {a + b}, cl(A) = [a, b]
• If A = {1/n : n ∈ N}, then A = A ∪ {0}, cl(A) = {0} (apply
the Archimedean property)
• If A = Z, then A = Z, cl(A) = ∅

Problem
Let (X , d) = (R, d1 ). Find all limit points and cluster points of Q.

Hint.
Since rationals are dense in R, any open interval contains infinitely
many rationals. So every real number is a cluster point of R.
Closure of a set

Theorem
Let (X , d) be a metric space and let A ⊆ X . Then the set A of
limit points of A is closed in X .

Proof.
We must check that X \ A is open. To see this, let x ∈ X \ A.
Thus, there exists r > 0 such that Br (x) ∩ A = ∅ or Br (x) ⊆ X \ A.
We claim that Br (x) ⊆ X \ A. Otherwise, there exists
y ∈ Br (x) ∩ A, and hence Bs (y ) ∩ A 6= ∅ for every s > 0. However,
for small s > 0, Bs (y ) ⊆ Br (x) (since Br (x) is open), and hence
Br (x) intersects with A, which is a contradiction.
Thus Br (x) ⊆ X \ A and hence X \ A is open or A is closed.

• A is known as the closure of A.


Problem
Let (X , d) be a metric space and let A ⊆ X . Then A is the smallest
closed set containing A, that is, if L ⊆ X is a closed set such that
A ⊆ L ⊆ A, then L = A. Conclude that A is closed iff A = A.

Solution.
Let x ∈ X \ L. Since L is closed, there exists r > 0 such that
Br (x) ∩ X \ L ⊆ X \ A, or Br (x) ∩ A = ∅, so x ∈ X \ A. Thus
X \ L ⊆ X \ A or A ⊆ L, and hence L = A.

Problem
Let (X , d) be a metric space and let A ⊆ X . Then A◦ is the largest
open set contained in A, that is, if O ⊆ X is an open set such that
A◦ ⊆ O ⊆ A, then O = A◦ .

Hint.
Show that O is contained in A◦ .

• For any set A of X , A◦ ⊆ A ⊆ A (strict inclusion may hold).


Convergence in metric spaces

By a sequence {xn }n>1 in a metric space X , we understand a


function from N into X , which maps n to xn .
Definition
Let (X , d) be a metric space and let {xn }n>1 be a sequence in X
We say that {xn }n>1 is a convergent sequence in X if there exists
x ∈ X (limit of {xn }n>1 ) such that

d(xn , x) → 0 as n → ∞.

Remark
• The limit x of a convergent sequence is unique: If d(xn , x) → 0
and d(xn , y ) → 0, then 0 6 d(x, y ) 6 d(x, xn ) + d(xn , y )
= d(xn , x) + d(xn , y ) → 0 as n → ∞.
Theorem
Let (X , d) be a metric space, let A be a subset of X and let
x ∈ X . Then x is a limit point of A if and only if there exists a
sequence {an }n>1 ⊆ A such that d(an , x) → 0 as n → ∞.

Proof.
Let x be a limit point of A.
• If x ∈ A, then take the constant sequence {an = x}n>1 .
• Suppose that x ∈/ A. Then x is a cluster point. Thus for every
r > 0, Br (x) ∩ (A\{x}) 6= ∅. By induction, after letting r = 1/n,
we obtain an ∈ B1/n (x) ∩ (A\{x}), and since d(an , x) < 1/n, we
have d(an , x) → 0 as n → ∞.
Conversely, assume that there exists a sequence {an }n>1 ⊆ A such
that d(an , x) → 0 as n → ∞. Thus for every r > 0, there exists
N > 1 such that d(an , x) < r for every n > N. Thus
aN ∈ Br (x) ∩ A, and hence x is a limit point of A.
Cauchy sequences and complete metric spaces

Definition
Let (X , d) be a metric space and let {xn }n>1 be a sequence in X .
We say that {xn }n>1 is a Cauchy sequence in X if d(xm , xn ) → 0
as m, n → ∞. A metric space is said to be complete if every
Cauchy sequence is convergent.
Remark
• Every convergent sequence is Cauchy: 0 6 d(xm , xn )
6 d(xm , x) + d(x, xn ) = d(xm , x) + d(xn , x) → 0 as m, n → ∞

Example
The metric space (R, d1 ) is complete.

Problem
Which of the following are complete metric spaces: ([a, b], d1 ),
((a, b), d1 ), ([a, b), d1 ), where d1 is the absolute metric.
Theorem
Let (X , d) be a complete metric space. Then every closed subset A
of X is complete when endowed with the (relative) metric

d A (x, y ) = d(x, y ), x, y ∈ A.

Proof.
Let {an }n>1 be a Cauchy sequence in (A, d A ). Clearly, {an }n>1 is a
Cauchy sequence in (X , d). However, X is complete, so that
{an }n>1 converges to some a ∈ X . It follows that a is a limit point
of A. Since A is closed, a ∈ A, and it follows that {an }n>1 is
convergent in A.

Corollary
Any closed subset of (R, d1 ) is complete
Example
Consider the metric space (Q, d1√), where d1 (x, y√) = |x − y |,
x, y ∈ Q. Let x1 ∈ Q such that 2 − 1 < √ x1 < 2 (by density √ of
rationals). Next choose x2 ∈ Q such that 2 − 1/2 < x2 < 2,
and hence by induction,
√ for every positive
√ integer n, there exists
xn ∈ Q such that 2 − 1/n < xn < 2. Thus, as m, n → ∞,
√ √
d1 (xm , xn ) = |xm − xn | 6 |xm − 2| + | 2 − xn | 6 1/m + 1/n → 0.

However, {xn }n>1 is √


not convergent
√ in Q (since the limit being
unique is necessarily 2 and 2 ∈ / Q).

• (Q, d1 ) is not complete.


Problem
Show that (R \ Q, d1 ) is not complete.
Problem
What are the Cauchy sequences in the discrete metric space
(X , d0 ) ?

Solution.
Let {xn }n>1 be a Cauchy sequence in (X , d0 ). Thus for  = 1,
there exists N > 1 such that

d0 (xn , xm ) < 1 for every m, n > N.

However, for the metric d0 , the distance between two points is


either 0 or 1. It follows that xn = xm for every m, n > N.

Problem
Which of the following are complete metric spaces: ([a, b], d0 ),
((a, b), d0 ), ([a, b), d0 ), where d0 is the discrete metric.
Theorem
Let (X , d) be a metric space. Then a Cauchy sequence {xn }n>1 is
convergent if and only if it has a convegent subsequence.

Proof.
If {xn }n>1 is convergent, then clearly any subsequence of {xn }n>1
is convergent (follows from the definition).
Let {xnk }k>1 be a convergent subsequence of {xn }n>1 and let
x ∈ X be its limit. Thus given  > 0, there exists N > 1 such that
d(xnk , x) < /2 for every k > N. Also, for some N 0 > N,
d(xn , xm ) 6 /2 for all m, n > N 0 . Thus for any n > N 0 ,

d(xn , x) 6 d(xn , xnN 0 ) + d(xnN 0 , x) < 

(since nN 0 > N 0 > N).


Bounded sets and diameter
Let (X , d) be a metric space and let ∅ =
6 A ⊆ X . We say that A
bounded if there exists r > 0 and x ∈ X such that A ⊆ Br (x).
• Every Cauchy sequence {xn }n>1 is bounded:
Indeed, for  = 1, there exists N > 1 such that d(xn , xN ) < 1
for all n > N. Thus {xn }n>1 ⊆ Br (x), where x = xN and
r = max{1, d(x1 , xN ), . . . , d(xN−1 , xN )} + 1.
Consider S = {d(x, y ) : x, y ∈ A}. The diameter of A is defined as
(
lub S if S is bounded above,
diam(A) =
∞ otherwise.
Problem
Let (X , d) be a metric space and let ∅ =
6 A ⊆ X . Show that A is
bounded if and only if diam(A) < ∞.
Hint.
diam(Br (x)) = 2r , and if A ⊆ B, then diam(A) 6 diam(B).
Normed linear spaces

Definition
A normed linear space X (denoted by (X , k · k) is a vector space
over R with a function assigning kxk ∈ R to every x ∈ X such that
for every x, y , z ∈ X and α ∈ R,
N1 (Non-negativity) kxk ≥ 0.
N2 (Positive Definiteness) kxk = 0 if and only if x = 0.
N3 (Dilation) kαxk = |α|kxk.
N4 (Triangle Inequality) kx + y k ≤ kxk + ky k.

Remark Every normed linear space is a metric space with metric


d(x, y ) = kx − y k, x, y ∈ X . This metric has the additional
property that d(αx, 0) = |α|d(x, 0) for every α ∈ R and x ∈ X .
Example
We contend that Rd with the norm k · kp is complete, where
d
X 1/p
kxkp = |xj |p , x ∈ Rd .
j=1

(n) (n)
To see this, let {x (n) = (x1 , . . . , xd )}n>1 be a Cauchy sequence
in Rd . Thus for k = 1, . . . , d,
d
X 1/p
(m) (n) (m) (n)
|xk − xk | 6 |xj − xj |p = kx (m) − x (n) kp → 0,
j=1
(n)
that is, {xk }n>1 is a Cauchy sequence in R. Since (R, d1 ) is
(n)
complete, there exists xk ∈ R such that |xk − xk | → 0 as n → ∞.
Clearly, kx (n) − xkp → 0 as n → ∞, where x = (x1 , . . . , xd ) ∈ Rd .
Problem
Show that Rd with the norm k · k∞ is complete.
Problem
(n) (n)
Let {x (n) = {(x1 , x2 , · · · )}n>1 be a Cauchy sequence in l p ,
1 6 p < ∞. For  > 0, verify the following:
(n)
1. {xi }n>1 ⊆ R converges to some xi ∈ R for every i > 1.
2. For k ≥ 1, there exists n0 ≥ 1 (independent of k) such that
k
(n)
X
|xi − xi |p ≤  for all n ≥ n0 .
i=1
Pk p
3. For k ≥ 1, i=1 |xi | ≤ ( + kx (n0 ) kp )p .
4. The normed linear space l p is complete.

Hint.
For part 1, argue as in the last example. For part 2, use the
definition of Cauchy sequence and part 1. For part 3, use the
triangle inequality. By parts 2-3, x ∈ l p and kx (n) − xkp → 0.
Problem
Let p be such that p > 1. For f ∈ C [0, 1], define
Z 1
kf kp = ( |f (x)|p dx)1/p .
0

Show that (C [0, 1], k · kp ) is an incomplete normed linear space.

Hint.
To see triangle inequality, apply Hölder’s equality (see the
argument on Page 35). To see that C [0, 1] is not complete, let
(
1 if x ∈ [0, 1/2),
f (x) =
0 if x ∈ [1/2, 1],

1
 if x ∈ [0, 1/2 − 1/n),
fn (x) = (−n/2)x + (n/2 + 1)/2 if x ∈ [1/2 − 1/n, 1/2 + 1/n],

0 if x ∈ [1/2 + 1/n, 1].

Verify that {fn }n≥3 converges to f ∈/ C [0, 1].
Dense sets and separable metric spaces

Definition
Let (X , d) be a metric space and let A ⊆ X . We say that A is
dense in X if every open ball in X intersects A, that is, for every
x ∈ X and every r > 0, Br (x) ∩ A 6= ∅.

Remark A is dense in X if and only if every x ∈ X is a limit point


of A, or equivalently, if and only if A = X .
Example
• Q and R \ Q are dense in (R, d1 )
• Z is not dense in (Q, d1 )
• Qd is a countable dense in (Rd , dp )

A metric space is separable if it has a countable dense subset.


Theorem
Let 1 ≤ p < ∞. Then l p is separable.
Proof.
For k > 1, let Ak = {(r1 , . . . , rk , 0, . . . , ) : rk ∈ Q}.
• Ak is countable: There is a bijection between Ak and Qk
• A = ∪∞k=1 Ak is countable: Countable union of countable sets
is countable
p p
P∞that Ap is dense in l : Let r > 0, x = (x1 , x2 , . . . , ) ∈ l .
We claim
Then k=1 |xk | < ∞, and hence for some integer N > 2,
( ∞ p 1/p < r /2. For j = 1, . . . , N − 1, let r ∈ Q be such
P
k=N k | )
|x j
r
that |xj − rj | < 2(N−1) 1/p (since Q is dense in R). Then for
y = (r1 , . . . , rN−1 , 0, . . . , ) ∈ A,

N−1
X ∞
X
kx − y kp 6 ( |xj − rj |p )1/p + ( |xk |p )1/p < r /2 + r /2 = r .
j=1 k=N

Thus for every x ∈ lp and r > 0, y ∈ Br (x) ∩ A.


Problem
Show that the set l ∞ of bounded sequences is complete normed
linear space with norm kxk∞ = supj>1 |xj |, x ∈ l ∞ .
Example
We claim that l ∞ is not separable. Consider the uncountable set

{0, 1}N = {x ∈ l ∞ : xn = 0 or 1}.

If x, y ∈ {0, 1}N , then kx − y k∞ = 1 if x 6= y . Thus


[
B1/2 (x) ⊆ l ∞ (disjoint union).
x ∈ {0,1}N

Now if l ∞ has a countable dense set, then each B1/2 (x) would
contain at least one element from this set. However, there are
uncountably disjoint balls of the form B1/2 (x). and hence l ∞ can
not admit a countable dense subset.
Question Whether (C [0, 1], d∞ ) is separable ?
Boundary of a set

Definition
Let (X , d) be a metric space and A ⊆ X . A point x ∈ X is said to
be a boundary point of A in X if for every r > 0,

Br (x) ∩ A 6= ∅ and Br (x) ∩ (X \ A) 6= ∅.

The boundary of A in X (denoted by ∂A) is the set of boundary


points of A in X .
Remark Every boundary point is a limit point, that is, ∂A ⊆ A.
Example
• If (X , d) = (R, d1 ), then for a, b ∈ R such that a < b,

∂[a, b] = ∂(a, b) = ∂[a, b) = ∂(a, b] = {a, b}.

• If (X , d) = (Rd , dp ), then ∂Br (x) = {y ∈ Rd : dp (x, y ) = r }.


Theorem
6 A ⊆ X . Then ∂A = A \ A◦ .
Let (X , d) be a metric space and ∅ =

Proof.
Clearly, x ∈ A \ A◦ if and only if x ∈ A and x ∈
/ A◦ . Note that
• x ∈ A if and only if for every r > 0, Br (x) ∩ A 6= ∅.
/ A◦ if and only if for every r > 0, Br (x) ∩ (X \ A) 6= ∅.
• x∈
Thus x ∈ A \ A◦ if and only if for every r > 0, Br (x) ∩ A 6= ∅ and
Br (x) ∩ (X \ A) 6= ∅, that is, x ∈ ∂A.

Problem
Find the boundary of A = {(x, y ) ∈ R2 : x > 0, y > 0}.

Hint.
Since A is open, A◦ = A. Also, A = {(x, y ) ∈ R2 : x > 0, y > 0}.
It follows that ∂A = {(x, y ) ∈ A : xy = 0}.

• The boundary of Q is equal to R.


Continuity

Definition
Let (X , d) and (Y , ρ) be two metric spaces. Let f : X → Y be a
function and let a ∈ X . We say that f is
C1 continuous
 at a if for every  > 0, there exists
 δ > 0 such that
x ∈ X , d(x, a) < δ =⇒ ρ(f (x), f (a)) <  .
C2 s-continuous at a (or sequential continuous at a) if
 
d(xn , a) → 0 as n → ∞ =⇒ ρ(f (xn ), f (a)) → 0 as n → ∞ .
We say that f is continuous (resp s-continuous) if it is continuous
(resp s-continuous) at every a ∈ X .
Remark
• f is continuous at a if and only if for every  > 0, there exists
δ > 0 such that f (Bδd (a)) ⊆ Bρ (f (a)).
• f is s-continuous at a if and only if f maps convergent
sequences to convergent sequences.
Problem
Let (X , d) and (Y , ρ) be two metric spaces. Let f : X → Y be a
function and let a ∈ X . Then f is continuous at a if and only if f is
s-continuous at a.

Hint.
⇒ follows from the definition: Given  > 0, there exists δ > 0 such
that C1 holds. Now find an integer N > 1 such that d(xn , a) < δ,
so by continuity, ρ(f (xn ), f (a)) < .
For ⇐, argue by contradiction: Suppose that C 2 holds but C 1 fails
to hold. So, for some  > 0, for δ = 1/n, n > 1, find xn ∈ X such
that d(xn , a) < 1/n but ρ(f (xn ), f (a)) > . Then d(xn , a) → 0 but
ρ(f (xn ), f (a)) 6→ 0, a contradiction.

Problem
Let (X , d), Y , ρ) and (Z , η) be metric spaces. If f : X → Y and
g : Y → Z are continuous, then so is g ◦ f : X → Z .
Example
Let p : R → R be a polynomial, that is, for some a0 , . . . , ak ∈ R,
p(x) = a0 + a1 x + · · · + ak x k , x ∈ R.
Notice that its not easy to verify the continuity of p right from its
definition (given , how to find δ ?). However, s-continuity of p
follows immediately from the facts that sum and product of
convergent sequences is convergent:
xn → x, yn → y ⇒ xn + yn → x + y , xnm → x m , m > 1.
Thus if xn → x, then p(xn ) → p(x) as n → ∞. So p is continuous.
Theorem
Let (X , d) be a metric space. Then the set C (X ) of continuous
functions f : X → R forms an algebra, that is, C (X ) is a vector
space over R equipped with the binary operation (f , g ) → fg from
C (X ) × C (X ) to C (X ), which is a binary form.
Proof.
The set of convergent sequences in R forms an algebra.
Let (X , d) be a metric space. Let A be a non-empty subset of X ,
and for x ∈ X , let

d(x, A) = glb Sx , where Sx = {d(x, a) : a ∈ A}

(since d(x, a) > 0, Sx is bounded from below. Also, since A is


nonempty, so is Sx . Hence glb of S exists).
Problem
Show that f : X → [0, ∞) given by f (x) = d(x, A) is continuous.

Solution.
To see this, for x, y ∈ X and a ∈ A, note that

f (x) = glb Sx 6 d(x, a) 6 d(x, y ) + d(y , a)

f (x) − d(x, y ) 6 d(y , a) for every a ∈ A =⇒


f (x) − d(x, y ) 6 glb Sy = f (y ) or f (x) − f (y ) 6 d(x, y ). Changing
roles of x and y , we obtain |f (x) − f (y )| 6 d(x, y ). Let δ = .
Separation by a continuous function

Theorem (Urysohn’s Lemma)


Let (X , d) be a metric space. Given closed non-empty disjoint
subsets A and B of X , there exists a continuous function
f : X → [0, 1] such that f |A = 0 and f |B = 1.

Proof.
For C ⊆ X closed and x ∈ X , let d(x, C ) = inf{d(x, a) : a ∈ C }.
• d(x, C ) is a continuous function of x.
• d(x, C ) = 0 if and only if x ∈ C .
• d(x, A) + d(x, B) > 0 for every x ∈ X .
Define f : X → [0, 1] by
d(x, A)
f (x) = , x ∈ X.
d(x, A) + d(x, B)
Clearly, f is a continuous function. Note that f (a) = 0 and
f (b) = 1 for every a ∈ A and b ∈ B.
A characterization

Theorem
Let (X , d) and (Y , ρ) be two metric spaces. Let f : X → Y be a
function. Then the following statements are equivalent:
(a) f is continuous.
(b) For every open subset O of Y , f −1 (O) = {a ∈ X : f (a) ∈ O}
is an open subset of X .
Proof.
(a) ⇒ (b). Let O ⊆ Y be open and a ∈ f −1 (O). Thus f (a) ∈ O.
Since O is open, there exists  > 0 such that Bρ (f (a)) ⊆ O. By
the continuity, there is δ > 0 such that f (Bδd (a)) ⊆ Bρ (f (a)) ⊆ O.
Thus Bδd (a) ⊆ f −1 (O), which shows that f −1 (O) is open.
(b) ⇒ (a). Let a ∈ X and  > 0. Since O = Bρ (f (a)) is an open
subset of Y , a ∈ f −1 (Bρ (f (a))) is an open subset of X . Hence
there exists δ > 0 such that Bδd (a) ⊆ f −1 (Bρ (f (a))), and hence f
is continuous.
An application to matrices

Problem
Consider the vector space Mn (R) (over R) of n × n matrices
P)16i,j6n with real entries. For A ∈ Mn (R), set
A = (ai,j
kAk = ni,j=1 |aij |. Verify the following:
1. k · k defines a norm on Mn (R).
2. The determinant det : Mn (R) → R is continuous.
3. The set GLn (R) of n × n invertible matrices is open in Mn (R).

Hint.
For 1, the triangle inequality follows from the triangle inequality of
real numbers. For 2, note that determinant is a polynomial in n2
variables, and hence det is s-continuous. For 3, note that
GLn (R) = det−1 (R \ {0}) and R \ {0} is open.
Remark Can you see that GLn (R) has a proper clopen subset ?
Equivalent norms

We say that two norms k · k and k · k0 on a normed linear space X


are equivalent if there exists m, M > 0 such that

mkxk 6 kxk0 6 Mkxk, x ∈ X.

Problem
Verify that the equivalence of norms is an equivalence relation.
Conclude that the all norms k · kp , p > 1, on Rn are equivalent.

Hint.
The first part is a routine verification. To see the second part, by
transitivity of norms, it suffices to check that k · kp is equivalent to
k · k∞ . This followsPfrom kxk∞ ≤ kxkp ≤ n1/p kxk∞ . This in turn
follows from |xj | 6 nj=1 |xj |p 6 n(maxj=1
n |x |)p , j = 1, . . . , d.
j
Problem
Let X be a normed linear space with two norms k · k and k · k0 .
Then the norms k · k and k · k0 are equivalent if and only if the
identity map I from (X , k · k) onto (X , k · k0 ) is continuous at 0
with inverse being continuous at 0.
Hint.
For ⇒, show that the identity map and its inverse are s-continuous.
For ⇐, note that the continuity of I : (X , k · k) → (X , k · k0 ) at 0
implies that for  = 1, there exists δ > 0 such that
kxk < δ ⇒ kxk0 < 1.
For 0 6= x ∈ X , let y = δ x δ
2 kxk . Thus ky k = 2 , and hence ky k0 < 1,
0
that is, 2δ kxk
kxk < 1 or kxk0 ≤ 2δ kxk for every x ∈ X . Similarly, one
may get the other inequality.
Problem
If two norms k · k and k · k0 on X are equivalent, show that
(X , k · k) is complete if and only if (X , k · k0 ) is complete.
Question What are all norms on Rd (up to equivalence) ?
Let us analyze the case of d = 1:
Problem
Describe all norms on R.

Solution.
Let k · k be any norm on R. For any x ∈ R, note that by the
dilation property,
kxk = kx · 1k = |x|k1k.
This means that any norm k · k is of the form α| · |, α > 0.

• Note that all norms on R are equivalent. Indeed, if k · k and k · k0


0
are two norms on R, then for m = k1k
k1k = M, we have

mkxk = kxk0 = Mkxk, x ∈ R.

What about the case of dimension d > 1 ?


Theorem
Consider (Rd , k · k∞ ). Let S = {x ∈ Rd : kxk∞ = 1} and let
f : S → R be a continuous function. Then there exists a ∈ S such
that f (x) > f (a) for every x ∈ S.

Proof.
Let A = {f (x) : x ∈ S} and let α = inf A. Thus there exists a
sequence {x (n) }n>1 ⊆ S such that f (x (n) ) → α as n → ∞. Now if
(n) (n) (n)
x (n) = (x1 , . . . , xd ), then {x1 }n>1 , being a bounded sequence,
has a convergent subsequence (by the Bolazano-Weierstrass
(n ) (n )
Theorem), say, {x1 j }j>1 . Since {x2 j }j>1 is bounded, it has a
convergent subsequence, and continuing this, we obtain a
(n) (n)
convergent subsequence of x (n) = (x1 , . . . , xd ) converging to
say a in k · k∞ . It follows that f (a) = α and f (x) > inf A = f (a)
for every x ∈ S. Since S is closed, a ∈ S.
Norms on Rd

Theorem
All norms on Rd are equivalent.

Proof.
Let k · k be an arbitrary norm on Rd .
(a) Let e1 , · · · , ed denote the standard basis d
Pd  of R . By the
triangle inequality, kxk ≤ i=1 kei k kxk∞ .

(b) By (a), the function f : (Rd , k · k∞ ) → R given by f (x) = kxk


is (sequential) continuous.
(c) f attains its minimum on S = {x ∈ Rd : kxk∞ = 1}. Thus
there exists a ∈ S such that kxk ≥ kak > 0 for every x ∈ S.
(d) By (a) and (c), the norm k · k is equivalent to k · k∞ .
The desired conclusion now follows from the transitivity of the
equivalence of norms.
One can produce inequivalent norms on infinite-dimensional spaces.
Example
Let R[x]Pdenote the vector space over R of polynomials
p(x) = kn=0 an x n in x. For c := {cn }∞
n=0 , define

k
X
kpkc := |cn ||an |.
n=0

• k · kc defines a norm on R[x] if cn 6= 0 for every n > 0.


This is a routine verification.
• k · kc and k · kd are not equivalent if cn = 1/(n + 1), dn = n + 1.
This is far from being obvious.
Example (Example continued ...)
If possible, then assume that there exists m, M > 0 such that

mkpkc 6 kpkd 6 Mkpkc , p ∈ R[x].

Thus, for any p ∈ R[x],


k k
X X |an | 1
kpkd = (n + 1)|an | < 1 whenever kpkc = < .
n+1 M
n=0 n=0

k
Choose an integer k large enough so that M > 2. Letting a0 = 0
1
and an = 2(n+1)M (n = 1, . . . , k), we get

k k
X |an | 1 X 1 1
= 2
< .
n+1 2M (n + 1) M
n=0 n=1
Pk k
However, n=0 (n + 1)|an | = 2M > 1, which is a contradiction.
Theorem
Let (X , d) and (Y , ρ) be two metric spaces. Let f : X → Y be a
function. Then the following statements are equivalent:
(a) f is continuous.
(b) For every closed subset U of Y , f −1 (U) = {a ∈ X : f (a) ∈ U}
is a closed subset of X .

Proof.
Recall that f −1 (Y \ U) = X \ f −1 (U). Now apply the fact that f is
continuous iff for every open subset O of Y , f −1 (O) is an open
subset of X .

Example
Let X be a metric space endowed with the discrete metric d0 .
Then any f : X → Y is continuous. Indeed, since every subset of
X is closed, by the previous theorem, f is continuous. In particular,
if Z carries the (relative) metric induced from (R, d1 ), then any
function f : Z → Y is continuous.
Problem
Consider the vector space Mn (R) of n × n matrices
Pnwith real entries. Define d : Mn (R) × Mn (R) → R
A = (ai,j )16i,j6n
by d(A, B) = i,j=1 |aij − bij |. Verify the following:
1. For every positive integer k, the map pk : Mn (R) → Mn (R)
given by pk (A) = Ak is continuous.
2. If Nn (R) = {A ∈ Mn (R) : there is k > 1 such that Ak = 0},
then Nn (R) is closed in Mn (R).

Hint.
To see that pk is s-continuous, note that the entries of Ak are
polynomials in the entries of A (please verify this for k = 2 and
n = 2). To see the part 2, note that

Nn (R) = ∪k>1 ker pk = ∪nk=1 ker pk

(justify the second equality). Since kernel of a continuous map is


closed, Nn (R) is closed in Mn (R).
Homeomorphisms

Definition
Let (X , d) and (Y , ρ) be two metric spaces.
• A function f : X → Y is a homeomorphism if f is continuous,
one-one, and onto with a continuous inverse.
• We say that X and Y are homeomorphic if there exists a
homeomorphism f : X → Y . In this case, we say that X ∼
= Y.
Remark Let f : X → Y be a homeomorphism.
1. f −1 is also a homeomorphism.
2. If U is open in X then f (U) is open in Y .
3. If A is a subset of X then A and f (A) are homeomorphic.
4. Composition of homeomorphisms is again a homeomorphism.
In particular, if X is homeomorphic to Y , and Y is
homeomorphic to Z then X is homeomorphic to Z .
∼ is an equivalence relation.
Note that =
Example
Consider the subsets A of the metric space (R, d1 ) as the metric
spaces with relative metric d1A (x, y ) = |x − y |, x, y ∈ A.
• (0, 1)  (0, 1] : If f : (0, 1) → (0, 1] is a homeomorphism, then
f (c) = 1 for some c ∈ (0, 1), and hence it follows that f −1
maps (0, 1) onto (0, c) ∪ (c, 1), which is not possible in view
of the intermediate value property.

Problem
Show that the interval (a, b) ⊆ R is homeomorphic to any other
interval (c, d) ⊆ R.

Hint.
Try α(t − b) + β(t − a) for appropriate scalars α and β.

Problem
Show that e −x is a homeomorphism from (0, ∞) onto (0, 1).
Problem
Which of spaces X and Y are homeomorphic:
(1) X = R and Y = [0, 1)
(2) X = R and Y = [0, 1]
(3) X = [1, ∞) and Y = (0, 1]
(4) X = (−1, 0) and Y = (−∞, −1)
(5) X = R and Y = (0, 1)
(6) X = Q and Y = Z

Hint. For (1), (2), use intermediate value property (No). For (3),
(4), check that 1/x is the desired homeomorphism (Yes). For (5),
write X = (−∞, 1) ∪ [1, ∞), and note that by (3), [1, ∞) ∼= (0, 1].
Also, by (4), (−∞, 1) ∼= (−∞, −1) ∼ = (−1, 0) (No). 6 For (6),
choose an open ball centred at an integer (which contains finitely
many elements), and analyze its image in Q (No).
6
The answer to (5) is Yes. Please try tan (thanks to Sudip and Satyam for catching a careless assertion!)
Theorem
Rn is homeomorphic to R iff n = 1.
Proof.
Suppose that for n > 1. Thus there is a continuous bijection
f : Rn → R, and hence g = f |X is a homeomorphism from
X := Rn \ {0} onto Y := R \ {y0 } for some y0 ∈ R. Choose
y1 , y2 ∈ Y such that y1 < y0 < y2 and let x1 , x2 ∈ X be such that
f (x1 ) = y1 and f (x2 ) = y2 . Let L denote the line segment
connecting x1 and x2 .
• If L does not pass through 0 then let γ(t) = (1 − t)x1 + tx2 .
• If L passes through 0 then choose any point x3 ∈ / L and let
(
(1 − 2t)x1 + 2t x3 if 0 ≤ t ≤ 1/2,
γ(t) =
2(1 − t)x3 + (2t − 1)x2 if 1/2 ≤ t ≤ 1.

Thus γ : [0, 1] → X is a continuous function (since γ(1/2) = x3 )


such that γ(0) = x1 and γ(1) = x2 .
Proof continued ...
We consider the continuous function h : [0, 1] → [y1 , y2 ] by

h(t) = g (γ(t)), t ∈ [0, 1].

By the intermediate value property, there exists t0 ∈ [0, 1] such


that h(t0 ) = y0 , that is, g (γ(t0 )) = y0 . This implies that y0
belongs to the image of g = f |X , which is a contradiction.

Remark It is highly non-trivial fact that Rn is homeomorphic to


Rm iff m = n (beyond the scope of this course).
Problem
Is an open annulus {(x, y ) ∈ R2 : 1 < x 2 + y 2 < 2} homeomorphic
to the unit circle {(x, y ) ∈ R2 : x 2 + y 2 = 1} ? Justify your answer.

Hint.
Argue as in the last theorem.
Objective To find when two metric spaces are homeomorphic, one
needs to look for ”invariants” which are preserved under
homeomorphisms (e.g, existence of a proper clopen set).

Problem
Show that the function g : (0, 1) → R given below is continuous
on irrationals and discontinuous on rationals:
(
1
if x ∈ Q ∩ (0, 1) and x = qp in reduced form
g (x) = q
0 otherwise.

Question Does there exist a function g : (0, 1) → R which is


continuous on rationals and discontinuous on irrationals ?
Answer: No
A solution was first provided by Vito Volterra.
G -delta sets

Definition
A set is a Gδ set if it is countable intersection of open sets.

Example
T
The irrationals R \ Q form a Gδ set for R \ Q = r ∈Q R \ {r }.

Problem
The rationals do not form a Gδ set.

Solution.
Suppose that Q = ∩∞ n=1 Un for open sets Un . Since Q ⊆ Un , each
Un is dense in R. Now note that ∅ = Q ∩ (R \ Q) is countable
intersection of open dense sets Un , n > 1 and R \ {r }, r ∈ Q. This
contradicts BCT (to be proven later).

Theorem (Baire Category Theorem)


A countable intersection of open dense sets in R is dense in R.
A Theorem of Volterra Vito

Theorem
Let U be an open subset of R and let f : U → R be a function
defined on U. Then A = {a ∈ U : f is continuous at a} is a Gδ set.

Proof.
For positive integer n, consider the set An given by

{x0 ∈ U : ∃ δ > 0 such that |f (x)−f (y )| < 1/n, x, y ∈ (x0 −δ, x0 +δ)}.

Note that An is open and A = ∩∞


n=1 An .

Since rationals do not form a Gδ set, we obtain the following:


Corollary (Volterra Vito)
There is no function g : (0, 1) → R which is continuous on
rationals and discontinuous on irrationals.
Baire category theorem

Theorem (Baire Category Theorem)


Let (X , d) be a complete metric space and let {Un }n>1 be a
sequence of open dense subsets of X . Then ∩n>1 Un is dense in X .

Proof.
Let x ∈ X and r > 0. We claim that Br (x) ∩ (∩n>1 Un ) 6= ∅.
• Since U1 is dense in X , there exists x1 ∈ Br (x) ∩ U1 . Since U1
is open, for some 0 < r1 < 1, Br1 (x1 ) ⊆ Br (x) ∩ U1 .
• Since U2 is dense in X , there exists x2 ∈ Br1 (x1 ) ∩ U2 . Since
U2 is open, for some 0 < r2 < 1/2, Br2 (x2 ) ⊆ Br1 (x1 ) ∩ U2 .
By induction on n > 1, there exist {xn }n>1 ⊆ X and 0 < rn < 1/n
such that Brn+1 (xn+1 ) ⊆ Brn (xn ) ∩ Un+1 for every n > 1. Thus
{xn }n>1 is a Cauchy sequence (since d(xn , xm ) < 1/n for every
m > n), and hence converges to some x0 ∈ X (since X is
complete). Note that x0 ∈ Br (x) ∩ (∩n>1 Un ).
Problem
Let X be a complete metric space. If X is a union of closed sets
A1 , A2 , . . . , , then show that there exists N > 1 such that AN has
non-empty interior.

Hint.
Assume that X = ∪n>1 An and that A◦n = ∅ for every integer n > 1.
Then ∅ = ∩n>1 (X \ An ), not possible (since X \ An = X ).

Remark R2 can not be written as the countable union of lines.


Also, R3 can not be written as the countable union of planes.

Problem
Let X be a complete metric space. If f : X → X is surjective, then
show that there exists an integer n > 1 and x0 ∈ X such that
f (Bn (x0 )) has nonempty interior.

Hint.
Apply the last problem to appropriate closed subsets An of X .
An application

Theorem
Let X be a complete normed linear space containing an infinite
linearly independent sequence {xn }n>1 in X . Then the linear span
of {xn }n>1 is a proper subspace of X .

Proof.
For m > 1, let Ym := linspan{x1 , · · · , xm }.
• Ym is a proper closed subspace of X , and hence Ym has
empty interior (if Br (x) ⊆ Ym , then for any nonzero y ∈ X ,
x + 2r kyy k ∈ Ym and hence y ∈ Ym ).
• The complement of Ym is open T and dense. By the Baire Cate-
gory Theorem, the intersection m>1 (X \ Ym ) is dense in X .
T
If linspan {xn }n>1 = X , then any element in m>1 (X \ Ym ) (which
exists) belongs to YN for large N, and hence it belongs to YN and
its complement X \ YN simultaneously, a contradiction.
Problem
If (X , d) is a complete metric space such that every x ∈ X is a
cluster point of X \ {x}, then show that X is not countable.

Solution.
If possible, assume that X is countable. So one can enumerate X
as a sequence {xn }n>1 . Note that for every integer n > 1, X \ {xn }
is open and dense in X . By the Baire Category Theorem,
∩∞ ∞
n=1 (X \ {xn }) is dense in X . However, ∩n=1 (X \ {xn }) = ∅, and
hence X must be uncoutable.
• (Z, d0 ) is a complete metric space even if Z is countable ({n} is
open for every n ∈ Z). Also, the above problem is applicable to
(R, d1 ) providing indirect verification of uncountability of R.
Theorem (Nested Set Theorem)
Let X be a complete metric space and let {Jn }n>1 be a decreasing
sequence of nonempty closed sets in X such that diam(Jn ) → 0 as
n → ∞. Then ∩n>1 Jn contains exactly one point.
Uniform continuity

Definition
Let (X , d) and (Y , ρ) be two metric spaces. Let f : X → Y be a
function. We say that f is uniformly continuous if for every  > 0,
there exists δ > 0 such that
 
x, y ∈ X , d(x, y ) < δ =⇒ ρ(f (x), f (y )) <  .

• Every uniformly continuous function is continuous at every


point in X . The major difference is that unlike the definition
of continuity at a point, single δ works for all points in X .
• f (x) = 1/x, x ∈ (0, 1) is continuous but not uniformly
continuous. Indeed, if m, n > 1 are integers such that
m > n > δ/2, then |1/n − 1/m| < δ & |f (1/n) − f (1/m)| > 1.
• For any subset A of a normed linear space X , the function
f (x) = dist(x, A), x ∈ X is uniformly continuous. Indeed,
since |f (x) − f (y )| 6 kx − y k, x, y ∈ X , one may take δ = .
Problem
Let (R, d1 ) and let f : R → R be uniformly continuous. Then
1. f maps bounded sets to bounded sets.
2. f maps Cauchy sequences to Cauchy sequences.

Hint.
To see 1, let A be a bounded subset of R. Thus A ⊆ [a, b] for some
a, b ∈ R such that a < b. We must check that f ([a, b]) is bounded.
Note that [a, b] can be covered by finitely many open intervals of
length less than δ (the one obtained from the definition of uniform
continuity with  = 1). Part 2 follows from definition.
Problem
Let X , Y be normed linear spaces and let T : X → Y be such that
is, T (x + αy ) = T (x) + αT (y ) for every x, y ∈ X and α ∈ R.
Show that T is continuous at 0 iff T is uniformly continuous.
Hint.
For 0 6= x ∈ X , let x 0 = δ x
2 kxk . Now use the continuity at 0.
Theorem
Let a, b ∈ R be such that a < b. If f : [a, b] → R is continuous,
then f is uniformly continuous.

Proof.
Let  > 0 and let x ∈ [a, b]. Then, by the continuity of f , there
exists δx such that
 
y ∈ [a, b], |x − y | < δx =⇒ |f (x) − f (y )| < /2 .

However, we want single δ, which will work for every x ∈ [a, b].
Note that [a, b] ⊆ ∪x∈[a,b] (x − δx /2, x + δx /2). It turns out that
there are finitely many x1 , . . . , xk ∈ [a, b] such that
[a, b] ⊆ ∪kn=1 (xn − δxn /2, xn + δxn /2)
(to be seen later). Let δ = minkn=1 δxn /2. If |x − y | < δ, then
x ∈ (xn − δxn /2, xn + δxn /2) for some n, and hence |y − xn | < δxn .
It follows that |f (x) − f (y )| < .
Compactness

Definition
Let (X , d) be a metric space. We say that X is compact if for any
collection {Uα : α ∈ I } of open subsets of X such that
X = ∪α∈I Uα , there exists finitely many indices {α1 , . . . , αk } in I
such that X = ∪kj=1 Uαj .
• {Uα : α ∈ I }: open cover of X
• {Uα1 , . . . , Uαk }: open subcover of X

Remark If f : X → Y is a continuous surjection and X is compact,


then Y is also compact. Indeed, if {Vα : α ∈ I } is an open cover
of Y , then {f −1 (Vα ) : α ∈ I } is an open cover of X , and since X
is compact, {f −1 (Vα ) : α ∈ I } admits a finite subcover
{f −1 (Vα1 ), . . . , f −1 (Vαk )} of X . It follows that Y = ∪kj=1 Vαj .
Problem
Show that none of the following metric spaces (X , d) is compact.
1. X = (0, 1], d = relative metric induced by d1
2. X = Z, d = d0
3. X = Rd , d = dp
4. X = normed linear space, d(x, y ) = kx − y k

Solution.
For 1, consider the open cover {(1/n, 1] : n > 2} of (0, 1]. For 2,
consider the open cover {{n} : n ∈ Z}. Part 3 follows from 4, and
for 4, consider the open cover {Bn (0) : n > 1}. Verify that none of
above open covers admits a finite subcover.

Problem
Let X be an arbitrary set endowed with a discrete metric d0 . Then
X is compact if and only if · · · .
Example
We claim that the finite interval [a, b] with relative metric d1 is
compact. If this is false, then there exists an open cover
U = {Uα : α ∈ I } of [a, b], which has no finite subcover.
• Note that U is also an open cover for [a, (a + b)/2] and
[(a + b)/2, b], and U does not admits finite subcover for at
least one of subintervals, say, [a1 , b1 ]. Clearly, b1 − a1 = b−a
2 .
• Note that U is also an open cover for [a1 , (a1 + b1 )/2] and
[(a1 + b1 )/2, b1 ], and U does not admits finite subcover for at
least one of subintervals, say, [a2 , b2 ]. Clearly, b2 − a2 = b−a
4 .
• By induction, for every integer n > 1, U does not admits finite
subcover for [an , bn ] ⊆ [an−1 , bn−1 ] and bn − an = b−a
2n .
• By the nested interval theorem, there exists c ∈ ∩∞ k=1 [ak , bk ].
Since U is an open cover for [a1 , b1 ], c ∈ Uα for some α ∈ I .
• c is an interior point of Uα , ak ↑ c and bk ↓ c, for large k,
[ak , bk ] ⊆ Uα , that is, [ak , bk ] admits a finite subcover. ⇒⇐
Theorem (Heine-Borel Theorem)
Any closed and bounded subset A of (R, d1 ) is compact.

Proof.
We claim that any closed subset A of a compact metric space
(X , d) is compact. Let U = {Uα : α ∈ I } be an open cover of A.
• U ∪ {X \ A} is an open cover of X .
• There exists a finite subcover {Uα1 , . . . , Uαk } ∪ X \ A of X
(since X is compact).
• {Uα1 , . . . , Uαk } is an open subcover of A.
Since A is bounded, A ⊆ [a, b] (finite interval). Since [a, b] is
compact and A is a closed subset of [a, b], A is also compact.

• Every closed subset of a compact metric space is compact.


Question Does there exist a closed and bounded set which is not
compact ?
Compactness in relative metric

Question Let (X , d) be a metric space and Y ⊆ X endowed with


the relative metric d Y (x, y ) = d(x, y ), x, y ∈ X . If K ⊆ Y , then
whether or not

(K , d) compact ⇔ (K , d Y ) compact ?

• If {Uα : α ∈ I } is an open cover of K in X then


{Uα ∩ Y : α ∈ I } is an open cover of K in Y .
• If {Vα : α ∈ I } is an open cover of K in Y , then for every
α ∈ I , Vα = Uα ∩ K for some open set Uα in X . Note that
{Uα : α ∈ I } is an open cover of K in X .

Conclude that (K , d) is compact ⇔ (K , d Y ) is compact.


Theorem
A compact subset K of a metric space (X , d) is closed & bounded.

Proof.
To see that K is closed, it suffices to check that X \ K is open.
• Let x ∈ X \ K and let ry = d(x, y )/2 > 0 for every y ∈ K .
• {Bry (y ) : y ∈ K } is an open cover of K . So it admits a finite
subcover {Bryj (yj ) : j = 1, . . . , N} (since K is compact).
• If r = min{ry1 , . . . , ryN }, then Br (x) ⊆ X \ K . Indeed, if
y ∈ K ∩ Br (x), then y ∈ Bryj (yj ) ∩ Br (x) for some j, and
hence d(y , yj ) < d(x, yj )/2, d(x, y ) < r 6 d(x, yj )/2, ⇒⇐.
Thus x is an interior point of X \ K , and hence K is closed.
To see that K is bounded, let x ∈ X . Consider the open cover
{Br (x) ∩ K : r > 0} of K . Since K is compact, there exists r > 0
such that K ⊆ Br (x), and hence it is bounded.
Example
Consider X = {r ∈ Q : 1 < r 2 < 2} with d being the relative
metric induced by d1 is not compact.
 √Then   √
√ 
X = {r ∈ Q : 1 < |r | < 2} = (1, 2) ∩ Q ∪ (− 2, −1) ∩ Q ,
which is not closed in (Q, d1 ).

Problem
Assume that n > 2. Show that Mn (R) \ GLn (R) is not compact as
a metric space endowed with the (relative)
P metric
d : Mn (R) × Mn (R) → R by d(A, B) = ni,j=1 |aij − bij |.

Hint.
Mn (R) \ GLn (R) is closed in Mn (R) but not bounded. Indeed,
there are diagonal matrices B in Mn (R) \ GLn (R) for which
d(0, B) is of arbitrarily large magnitude provided n > 1.
Problem
Consider C2 endowed with the metric

d((z, w ), (z 0 , w 0 )) = (|z −z 0 |2 +|w −w 0 |2 )1/2 , (z, w ), (z 0 , w 0 ) ∈ C2 .

Show that {(z, w ) ∈ C2 : z 2 + w 2 = 1} is not compact.

Hint. √
Find√w ∈ C such that w 2 = 1 + n2 . Thus, for ι = −1, we have
(ιn, 1 + n2 ) ∈ {(z, w ) ∈ C2 : z 2 + w 2 = 1}.

Question What if z 2 + w 2 − 1 is replaced by any nonconstant


polynomial in z, w ?
The answer is yes (this can be deduced from the Fundamental
theorem of algebra).
Theorem
If (X , d) is compact, then every sequence in X has a convergent
subsequence.

Proof.
We claim that an infinite subset A of X has a cluster point in X .
• Suppose A has no cluster point in X , that is, for every x ∈ X ,
there exists rx > 0 such that Brx (x) ∩ A ⊆ {x}.
• {Brx (x) : x ∈ X } is an open cover for X , and hence X has
finite subcover {Brxj (xj ) : j = 1, . . . , N}.
• A = ∪N
j=1 (Brxj (xj ) ∩ A) ⊆ {x1 , . . . , xN } is a finite set.
Applying the above fact to A = {xn }n>1 , we conclude that A has a
cluster point in X . However, for every cluster point, there exists a
sequence in A converging to this cluster point (see slide 62).
Let Y be a proper, closed subspace of a normed linear space X .
• Choose x1 ∈ X \ Y , and note that d(x1 , Y ) > 0.
• There exists x0 ∈ Y such that kx1 − x0 k < 2d(x1 , Y ).
• Note that kx1 − x0 ky + x0 belongs to Y for any y ∈ Y .
−x0
• If x = kxx11 −x0k
(unit vector), then for any y ∈ Y ,
kx1 − x0 + kx1 − x0 ky k 1
kx − y k = > .
kx1 − x0 k 2
Thus there exists a unit vector x ∈ X such that d(x, Y ) > 1/2.
Theorem
If X is infinite dimensional, then unit sphere in X is not compact.
Proof.
• Let x1 ∈ X be a unit vector in X and let X1 := span{x1 }.
• There exists x2 ∈ X such that kx2 k = 1 and d(x2 , X1 ) ≥ 1/2.
• Note that x2 ∈
/ X1 . Let X2 := span{x1 , x2 }.
• There exists x3 ∈ X2 such that kx3 k = 1 and d(x3 , X2 ) ≥ 1/2.
Continuing this, we get {xn }n>1 with no cgt subsequence.
Definition
A metric space X is sequentially compact if every sequence in X
has a subsequence convergent in X .
• A compact metric space is sequentially compact (slide 117).
• If X is sequentially compact and f : X → Y is a continuous
surjection, then Y is sequentially compact (if {yn = f (xn )}n>1
is given, then {xn }n>1 has a convergent subsequence, and
hence {f (xn )}n>1 has a convergent subsequence).
• Every sequentially compact metric space is complete. Indeed,
if {xn }n>1 is a Cauchy sequence, then since {xn }n>1 has a
convergent subsequence, {xn }n>1 is convergent (slide 67).

• A complete metric space is not necessarily sequentially compact.


For example, {n}n∈N in (R, d1 ) has no convergent subsequence.
Question What are all complete metric spaces which are also
sequentially compact ?
Problem
Consider the vector space Mn (R) of n × n matrices
A = (ai,j )16i,j6n with real entries. For X = (x1 , . . . , xn ) ∈ Rn (seen
as a column vector), note that AX ∈ Rn . Define

kAk = sup kAX k2 , A ∈ Mn (R).


X ∈Rn ,kX k2 =1

Verify the following:


1. (Mn (R), k · k) is a complete normed linear space.
2. ISO = {A ∈ Mn (R) : kAX k2 = kX k2 for every X ∈ Rn } is
sequentially compact.
Hint.
Since k · k2 is a norm on Rn , k · k defines a norm on Mn (Rn ). To
2
see that Mn (R) is complete, note that f : Mn (R) → (R n , k · k2 )
given by f (A) = (a11 , . . . , a1n , . . . , an1 , . . . , ann ) is a linear
homeomorphism (since kAm − Ak → 0 if and only if (i, j)th entry
of Am converges to (i, j)th entry of A for every 1 6 i, j 6 n). To
see 2, note that ISO is closed and bounded in Mn (R).
Theorem
Let (X , d) be a metric space and let A ⊆ X be sequentially
compact. Then, for every  > 0, there exist finitely many points
x1 , . . . , xN ∈ X such that A ⊆ ∪N
j=1 B (xj ).

Proof.
Let  > 0 and let x1 ∈ A.
• If A ⊆ B (x1 ), then we are done.
• Otherwise, there exists x2 ∈ A \ B (x1 ). Thus d(x1 , x2 ) > .
• If A ⊆ B (x1 ) ∪ B (x2 ), then we are done.
• Otherwise, there exists x3 ∈ A \ (B (x1 ) ∪ B (x2 )). Thus
d(x1 , x3 ) >  and d(x2 , x3 ) > .
Continuing this, we either have A ⊆ ∪N j=1 B (xj ) for some integer
N > 1 or there exists a sequence {xn }n>1 such that d(xn , xm ) > 
for all positive integers m 6= n. The latter case does not arise since
A is sequentially compact.
Totally bounded spaces

Definition
A metric space X is said to be totally bounded if for every  > 0,
there exist finitely many points x1 , . . . , xN ∈ X such that
X ⊆ ∪N
j=1 B (xj ).

• Unlike compactness and sequential compactness, total


boundedness is not preserved under homeomorphism. To see
this, note that (0, 1) is totally bounded. Indeed, if choose an
integer N > 1 such that N > 1, then

(0, 1) ⊆ (−, ) ∪ (/2, 3/2) ∪ (, 3) ∪ · · · ∪ (N, (N + 2)).

On the other hand, R is homeomorphic to (0, 1), but it is not


totally bounded.
Compact / Sequentially compact

* 
Complete and totally bounded

(see slides 117, 119, 121)


Question Whether Sequentially Compact ⇒ Compact ?
Question Whether Complete and Totally bounded ⇒ Compact ?
• This will give complete the diagram!
Problem
Every totally bounded set is bounded.

Hint.
Note that union of finitely many open balls is contained in a single
open ball of sufficiently large radius.

Problem
Show that a subset of a totally bounded metric space is again
totally bounded.

Solution.
Suppose X is totally bounded and A ⊆ X . Then, for every  > 0,
there exist x1 , . . . , xN ∈ X such that A ⊆ ∪N
j=1 B/2 (xj ). For each j,
choose aj ∈ A ∩ B/2 (xj ), and note that by the triangle inequality,
A ⊆ ∪N j=1 B (aj ).
Problem
Show that union of finitely many totally bounded sets is totally
bounded.

Solution.
This follows from the definition.

Problem
Show that every totally bounded metric space is separable.

Solution.
(k) (k)
For every integer k > 1, there exist x1 , . . . , xNk ∈ X such that
(k) (k) (k)
X ⊆ ∪N j=1 B1/k (xj ). Clearly, Y = ∪k>1 {x1 , . . . , xNk } is
k

countable. Verify that Y is dense in X , that is, every ball of radius


 in X intersects Y non-trivially.
Theorem
Let (X , d) be a metric space. Let {xn }n>1 be a sequence in X and
let A = {x ∈ X : x = xn for some n > 1}. Then
(1) If {xn }n>1 is Cauchy, then A is totally bounded.
(2) If A is totally bounded, then {xn }n>1 has a Cauchy
subsequence.
Proof.
To see (1), let  > 0. Since {xn }n>1 is Cauchy, there exists N > 1
such that d(xn , xm ) <  for all m, n > N. Thus

A = {x1 }∪· · ·∪{xN−1 }∪{xn : n > N} ⊆ B (x1 )∪· · ·∪B (xN−1 )∪B (xN ).
To see (2), we may assume that A is infinite (otherwise, {xn }n>1 is
eventually constant, and hence it has a convergent sequence).
• Cover A by finitely many balls of radius 1, and at least one of
these balls, say, A1 contains infinitely many points in A.
• Cover A1 by finitely many balls of radius 1/2, and at least one
of these balls, say, A2 contains infinitely many points in A.

Proof continued ...
Continuing this, we obtain a decreasing sequnece {An }n>1 of balls
An of radius 1/n, where each An contains infinitely many points of
A. Choose now a sequence {xnk }k>1 such that xnk ∈ Ak and check
that {xnk }k>1 is a Cauchy sequence.
Corollary
Let (X , d) be a metric space. Then X is totally bounded and
complete if and only if it is sequentially compact.
Proof.
We have already seen ⇐ (see slides 119 and 121). To see ⇒, let
{xn }n>1 be given and and let
A = {x ∈ X : x = xn for some n > 1}.
Since X is totally bounded, so is A (see slide 124). By the last
theorem, {xn }n>1 has a Cauchy subsequence, say, {xnk }k>1 .
However, X is complete, so that {xnk }k>1 is convergent in X , and
hence X is sequentially compact.
Hilbert cube
Example
Let H denote the space of all sequences in [0, 1]. Thus

H = {x = {xn }n>1 : 0 6 xn 6 1 for every n > 1}.

One may think of H as the infinite product ∞


Q
n=1 [0, 1]. Define
d : H × H → [0, ∞) by
 |x − y | 
n n
d(x, y ) = sup , x, y ∈ H.
n>1 2n
• d is a metric. Indeed, 0 6 d(x, y ) 6 1/2, d(x, y ) = 0 ⇔
x = y , and d(x, z) 6 d(x, y ) + d(y , z) for every x, y , z ∈ H.
(n)
• d(x (n) , x) → 0 as n → ∞ ⇔ |xk − xk | → 0 as n → ∞ for
every k > 1. The part ⇒ is clear. To see ⇐, choose N > 2
such that 2−N 6 /4, and choose N0 > 1 such that
(n)
|xk − xk | < /2 for all k = 1, . . . , N − 1 and for all n > N0 .
Example (Continued ...)
• Let r > 0 and x ∈ H. Let N > 2 be such that 2N r > 2. Then
 |x − y | 
n n
Br (x) = {y ∈ H : d(x, y ) < r } = {y ∈ H : sup < r}
n>1 2n

⊆ {y ∈ H : |xn − yn | < 2n r for every n > 1}

= {y ∈ H : |xn − yn | < 2n r for n = 1, . . . , N − 1}.

=⇒ Br (x) = {y ∈ H : |xn − yn | < 2n r for n = 1, . . . , N − 1}.

Thus the open ball Br (x) in H is of the form


B2r (x1 ) × · · · × B2N−1 r (xN−1 ) × [0, 1] × · · · .
We claim that H is sequentially compact. It suffices to check that
H is complete and totally bounded.
Example (Continued ...)
• The Hilbert cube H is totally bounded.
Let  > 0. Find an integer N > 1 such that 2N  > 2. Then, by the
discussion on the previous slide, the ball B (x) is of the form

B2 (x1 ) × · · · × B2N−1  (xN−1 ) × [0, 1] × · · · .

Since [0, 1] is totally bounded, for every j = 1, . . . , N − 1, it can be


covered by finitely many balls (intervals) of the form B2j  (xij ) (with
finitely choices of i = 1, . . . , Nj ). If one takes the product

B2 (xi1 1 ) × · · · × B2N−1  (xiN−1 N−1 ) × [0, 1] × · · · ,

then H can be covered by finitely many balls of radius .


Example (Continued ...)
• The Hilbert cube H is complete.
To see that H is complete, let {x (n) }n>1 be Cauchy in H.
• Note that d(x (n) , x (m) ) → 0 as m, n → ∞ if and only if
(n) (m)
|xk − xk | → 0 as n → ∞ for every k > 1.
(n
• Since R is complete, {xk }n>1 converges to some xk in R for
every k > 1.
• Note that x = {xk }k>1 belongs to H. It follows from the
discussion on slide 128 that d(x (n) , x) → 0 as n → ∞.
Since complete and totally bounded metric space is sequentially
compact, the discussion above shows that the Hilbert cube H is
indeed sequentially compact.
Problem
Show that {x = {xn }n>1 : −1 6 xn 6 1 for every n > 1} as a
subset of l ∞ is not compact.

Problem
Give an example of infinite sequentially compact subset of l ∞ .

Solution.
Let K = {x = {xn }n>1 : 0 6 xn 6 1/2n for every n > 1} and note
that f : H → l ∞ given by

f ({x1 , x2 , x3 , . . . , }) = {x1 /2, x2 /4, x3 /8 . . . , }

satisfies kf (x) − f (y )k∞ = d(x, y ). Thus f is continuous, and


since H is sequentially compact, so is f (H). However, f (H) = K is
a subset of l ∞ .

The above example provides an infinite compact subset of l ∞ .


Theorem
A sequentially compact metric space is compact.

Proof.
Let {Uα : α ∈ I } be an open cover of X . For x ∈ X , let rx =
sup{r ∈ R : Br (x) ⊆ Uα for some α} (< ∞ since X is bounded).
• For  = inf{rx : x ∈ X }, there is {xn }n>1 such that rxn → .
• By sequential compactness, {xnk }k>1 converges to x ∈ X .
• x ∈ Uα for some α, and hence Br (x) ⊆ Uα for some r > 0.
• If d(xnk , x) < r /2 for all large k, then rxk > r /2. Thus  > 0.
• Let x1 ∈ X . Either X = B/2 (x1 ) ⊆ Uα for some α or there
exists x2 ∈ X \ B/2 (x1 ).
• Either X = B/2 (x1 ) ∪ B/2 (x2 ) ⊆ Uα ∪ Uβ for some α, β or
there exists x3 ∈ X \ (B/2 (x1 ) ∪ B/2 (x2 )).
One may continue this. However, this process ends in finitely many
steps (otherwise we will get a sequence without a convergent
subsequence contradicting sequential compactness).
Problem
Let (X , d), (Y , ρ) be two metric spaces. For (xj , yj ) ∈ X × Y
define D((x1 , y1 ), (x2 , y2 )) = max{d(x1 , x2 ), ρ(y1 , y2 )}, Verify:
1. D defines a metric on X × Y .
2. If X and Y are compact, then so is X × Y .
Hint.
For 2, it suffices to check that X × Y is sequentially compact.
• Let {(xn , yn )}n>1 be a sequence in X × Y , Then {xn }n>1 ⊆ X
and {yn }n>1 ⊆ Y .
• Since X is sequential compact, {xnk }k>1 is convergent.
• Now {ynk }k>1 ⊆ Y , so by the sequential compactness of Y ,
{ynkl }l>1 is convergent.
Thus {xnkl }l>1 and {ynkl }l>1 are convergent. Then {(xnkl , ynkl )}l>1
is convergent in X × Y . So X × Y is sequentially compact.

Question Can you say that X and Y are compact if so is X × Y ?


One may define sum of two subsets A and B of Rn as
A + B = {a + b : a ∈ A, b ∈ B}.
Problem
Let A be closed, B be compact in Rn . Show that A + B is closed.
Solution.
Let {xn }n>1 be a sequence in A + B such that xn → x in Rn .
• Thus xn = an + bn for some {an }n>1 ⊆ A and {bn }n>1 ⊆ B.
• {bnk }k>1 converges to b ∈ B (B is sequentially compact).
• {xnk }k>1 converges to x, and {ank }k>1 converges to a ∈ A
(difference of cgt sequences is cgt and A is closed).
Thus x = a + b for a ∈ A and b ∈ B. Hence A + B contains all its
limit points showing that it is closed.
Problem
Let A and B be compact in Rn . Show that A + B is compact.
Problem
Give an example of two closed sets whose sum is not closed.
Corollary (Lebesgue covering lemma)
Let (X , d) be a compact metric space. Let {Uα : α ∈ I } be an
open cover of X . Then there is a δ > 0 such that if A ⊆ X with
diameter diam(A) < δ, then there is α ∈ I such that A ⊆ Uα .

Proof.
Suppose there is no δ > 0 with the above property.
• For a positive integer n, letting δ = 1/n, there exists An ⊆ X
with diameter diam(An ) < 1/n, then An 6⊆ Uα for every α ∈ I .
• Choose any xn ∈ An and note that {xn }n>1 has a subsequence
{xnk }k>1 converging to some x ∈ X (by compactness of X ).
• x ∈ Uα for some α ∈ I (since {Uα : α ∈ I }: open cover of X ).
• Since Uα is open, Br (x) ⊆ Uα for some r > 0.
• Find k > 1 such that nk r > 2 and xnk ∈ Br /2 (x).
• Now if a ∈ Ank , then d(x, a) 6 d(x, xnk ) + d(xnk , a) < r .
This shows that Ank is contained in Br (x) ⊆ Uα , ⇒⇐ .
Lebesgue number
Definition
Let U = {Uα : α ∈ I } be an open cover of (X , d). A real number
δ > 0 is said to be Lebesgue number for U if A ⊆ X with diameter
diam(A) < δ, then there is α ∈ I such that A ⊆ Uα .
Example
Consider the metric space [0, 1] with relative metric induced by d1 .
Let δ > 0 be given. We claim that the Lebesgue number for the
open covering U = {[0, δ)} ∪ {(1/n, 1] : n > 1} of [0, 1] equals δ. 7
• If N is a positive integer such that 1/N < δ (which exists by
AP), then {[0, δ), (1/N, 1]} is a finite open subcover of [0, 1].
• Let A be such that diam(A) < δ. Then either A ⊆ [0, δ) or
A ∩ [δ, 1] 6= ∅. Let a ∈ A be such that a > δ.
• If A * (1/n, 1] for any n > 1, (so a > δ) then there is aM ∈ A
such that aM 6 1/M < a − δ < a. Thus diam(A) > δ ⇒⇐
7
I thought for a moment that there is a glitch but this calculation is fine.
Continuous functions on compact metric spaces

Theorem
If f is a continuous function from a compact metric space (X , d)
to any other metric space (Y , ρ), then f (X ) is bounded.

Proof.
Fix y0 ∈ Y and write Y = ∪n>1 Bnρ (y0 ).
• f −1 (Bnρ (y0 )) is open for every n > 1 (since f is continuous).
• X = ∪n>1 f −1 (Bnρ (y0 )).
• Since X is compact, there exist positive integers n1 , . . . , nk
such that X = ∪kj=1 f −1 (Bnρj (y0 )).
Let N = max{n1 , . . . , nk } and note that X = f −1 (BNρ (y0 )). Thus
f (X ) ⊆ BNρ (y0 ) is bounded.
• Let f : X → R be a continuous function. Then supx∈X f (x) and
inf x∈X f (x) exist. Can we find x0 and x1 in X such that
supx∈X f (x) = f (x0 ) and inf x∈X f (x) = f (x1 ) ?
Suppose that there is no x0 ∈ X such that M = f (x0 ), where
M = supx∈X f (x). Thus f (x 0 ) < supx∈X f (x) for every x 0 ∈ X .
• {x 0 ∈ X : f (x 0 ) < M − 1/n} is an open subset of X .
• X = ∪n>1 {x 0 ∈ X : f (x 0 ) < M − 1/n}.
• There exist positive integers n1 , . . . , nk such that

X = ∪kj=1 {x 0 ∈ X : f (x 0 ) < M − 1/nj }.

• X = {x 0 ∈ X : f (x 0 ) < M − 1/N} with N = max{n1 , . . . , nk }.


This would imply that f (x) < M − 1/N for every x ∈ X , and
hence M < M − 1/N ⇒⇐
• Similarly, one can see that there exists x1 ∈ X such that
inf x∈X f (x) = f (x1 ) (Exercise).
Problem
Let X be a compact metric space. Then there exists no continuous
map from X onto (0, 1).
Theorem (Criterion for homeomorphism)
Let X , Y be metric spaces and let f : X → Y be a bijective
continuous map. If X is compact, then f is a homeomorphism.

Proof.
To check that f −1 is continuous, let A be a closed subset of X .
• A is compact (being closed subset of a compact space X ).
• f (A) is compact (continuous image of a compact space).
• f (A) is closed (compact subset of a metric space is closed).
Thus continuous image of a closed set is closed. Hence f −1 is
continuous.

Problem
Let X be a normed linear space with two norms k · k and k · k0 . Let
Y be a compact subset of X . If there exists M > 0 such that
kxk ≤ Mkxk0 for every x ∈ Y , then show that (Y , k · k) and
(Y , k · k0 ) are homeomorphic.
Theorem
Any continuous function f from a compact metric space (X , d) to
any other metric space (Y , ρ) is uniformly continuous.

Proof.
Since f is continuous at x, given  > 0, there exists δx > 0 such
that ρ(f (x), f (x 0 )) < /2 whenever x 0 ∈ X and d(x, x 0 ) < δx .
• {Bδx (x) : x ∈ X } is an open cover of X .
• There is a δ > 0 (Lebesgue number) such that if A ⊆ X with
diameter diam(A) < δ, then there is x ∈ X such that
A ⊆ Bδx (x) (by Lebesgue covering lemma).
• If x1 , x2 ∈ X are such that d(x1 , x2 ) < δ, then apply the above
to A = {x1 , x2 }, we get x ∈ X such that {x1 , x2 } ⊆ Bδx (x).
Thus d(x1 , x) < δx and d(x2 , x) < δx . Hence, by continuity at x,
ρ(f (x1 ), f (x2 )) 6 ρ(f (x1 ), f (x)) + ρ(f (x2 ), f (x)) < .
Example
Let f : (0, 1) → R be a bounded, continuous function, which is
monotonically increasing.
• inf x∈(0,1) f (x) and supx∈[0,1) f (x) exist (since f is bounded).
• Define g : [0, 1] → R by

f (x)
 if x ∈ (0, 1),
g (x) = inf x∈(0,1) f (x) if x = 0,

supx∈[0,1) f (x) if x = 1.

Note that g : [0, 1] → R is a continuous function (since f is


increasing), and hence it is uniformly continuous.

• One can not drop the assumption of boundedness in the above


example (e.g. f (x) = 1/x, x ∈ (0, 1)).
Problem
Given (X , d), (Y , ρ), such that X is compact, verify:
(1) supx∈X ρ(f (x), y ) < ∞ for any y ∈ Y .
(2) Consider C (X , Y ) = {f : X → Y : f is continuous}. Define

D(f , g ) = sup ρ(f (x), g (x)), f , g ∈ C (X , Y ).


x∈X

Then (C (X , Y ), D) is a metric space.

Hint.
For a positive integer k, define Uk = {x ∈ X : ρ(f (x), y ) < k}.
Verify that {Uk : k > 1} is an open cover of X . Now (1) follows
from the compactness of X .
By (1) and the triangle inequality, D(f , g ) < ∞ for every
f , g ∈ C (X , Y ). Since ρ is a metric, so is D.
Theorem
Let (X , d), (Y , ρ) be metric spaces and X be compact. If Y is
complete, then C (X , Y ) is complete.
Proof.
Let {fn }n>1 be a Cauchy sequence in C (X , Y ), that is, given  > 0,
there exists N > 1 such that D(fn , fm ) < /3 for every m, n > N.
• For every x ∈ X , {fn (x)}n>1 is a Cauchy sequence in Y .
• For every x ∈ X , {fn (x)}n>1 converges to some f (x) ∈ Y .
• ρ(fn (x), fm (x)) < 3 , m, n > N ⇒ ρ(f (x), fm (x)) 6 3 , m > N
To see that f ∈ C (X , Y ), note that for any x, x 0 ∈ X ,

ρ(f (x), f (x 0 )) 6 ρ(f (x), fN (x))+ρ(fN (x), fN (x 0 ))+ρ(fN (x 0 ), f (x 0 )),

which is 6 2/3 + ρ(fN (x), fN (x 0 )). However, fN is continuous and


X is compact, so that for some δ > 0,
ρ(fN (x), fN (x 0 )) < /3 whenever d(x, x 0 ) < δ.
Thus ρ(f (x), f (x 0 )) <  whenever d(x, x 0 ) < δ.
Corollary
Let (X , d) be a compact metric space and consider

C (X ) = {f : X → R : f is continuous}.

If kf k∞ = supx∈X |f (x)|, f ∈ C (X ), then C (X ) is a complete


normed linear space with k · k∞ .

The rest of the notes is devoted to the study of this metric space.
• What are compact subsets of C (X ) ? (Arzela-Ascoli Theorem)
• Whether C (X ) is separable ? (Weierstrass Theorem)
• Is there any proper clopen subset of C (X ) ?
Answer to the last question is No (Why ?)
Totally bounded subsets of C (X )

Theorem
Let (X , d) be a compact metric space and let F be a totally
bounded subset of C (X ). Then for each  > 0, there exists δ > 0
such that supf ∈F |f (x) − f (y )| <  whenever x, y ∈ X and
d(x, y ) < δ.

Proof.
Let  > 0. Then
• there exist f1 , . . . , fN ∈ F such that F ⊆ ∪N
j=1 B/3 (fj ).
• there exists δj > 0 such that |fj (x) − fj (y )| < /3 whenever
d(x, y ) < δj (since each fj is uniformly continuous; slide 141).
If δ = min{δ1 , . . . , δN } and d(x, y ) < δ, then any f ∈ F lies in
some ball B (fj ), and hence
|f (x)−f (y )| 6 |f (x)−fj (x)|+|fj (x)−fj (y )|+|fj (y )−f (y )| < .
Equicontinuity

Definition
Let (X , d) be a compact metric space and let F be a subset of
C (X ). We say that F is equicontinuous if for each  > 0, there
exists δ > 0 such that supf ∈F |f (x) − f (y )| <  whenever x, y ∈ X
and d(x, y ) < δ.

• Every compact subset of C (X ) is equicontinuous (since compact


metric space is totally bounded; see slides 121 and 133).
Question What are all compact subsets of C (X ) ?
• A compact subset of C (X ) is closed and bounded (see slide 114).
Question Are equicontinuous, closed and bounded subsets of
C (X ) all compact subsets of C (X ) ?
Theorem (Generalized Arzela-Ascoli Theorem)
Suppose that (X , d) is a compact metric space and let {fn }n>1 be
a sequence in C (X ) such that
(A) supn>1 |fn (x)| < ∞ for every x ∈ X ,
(B) for each  > 0 and x ∈ X , there exists δx > 0 such that
supn>1 |fn (x) − fn (y )| <  whenever y ∈ X and d(x, y ) < δx .
Then there exist a subsequence {fnk }k>1 of {fn }n>1 and f ∈ C (X )
such that kfnk − f k∞ → 0 as k → ∞.

Remark Let F be a subset of C (X ).


• If F is bounded, then any sequence {fn }n>1 in F satisfies (A).
• Condition (B) is the ”equicontinuity” of {fn }n>1 at x.
• If all sequences in F satisfy (A) and (B), then F is compact.

Corollary (Arzela-Ascoli Theorem)


A subset F of C (X ) is (sequentially) compact if and only if it is
closed, bounded and equicontinuous.
Proof of Generalized Arzela-Ascoli Theorem.
The proof is divided into two steps:
Step 1
• Let {xn }n>1 be a countable dense subset of X (a compact
metric space is separable; see slide 125).
• There exists a convergent subsequence {fnk (x1 )}k>1 of the
bounded sequence {fn (x1 )}n>1 (use (A) and H-B Theorem)
• There exists a convergent subsequence {fnkl (x2 )}l>1 of the
bounded sequence {fnk (x2 )}k>1 (use (A) and H-B Theorem)
• {fnkl (x1 )}l>1 is convergent (subsequence of a convegent
sequence is convergent)
Continuing this, we obtain a subsequence {fnk }k>1 such that
{fnk (xj )}k>1 is convergent for every j > 1 (form the set {nk }k>1
by picking up first element in first subsequence, second element in
second subsequence and so on).
Proof of Generalized Arzela-Ascoli Theorem continued ...
Step 2 Let δx be as given in (B).
• There exist a1 , . . . , am ∈ X such that X = ∪m
k=1 Bδak (ak )
(since X = ∪a∈X Bδa (a) and X is compact)
• Choose yk ∈ Bδak (ak ) ∩ {xn }n>1 and let N > 1 be such that
|fni (yk ) − fnj (yk )| <  for i, j > N and k = 1, . . . , m (Step 1)
• If x ∈ X . then x ∈ Bδak (ak ) for some k = 1, . . . , m.
Note that for i, j > N, (use (B) four times)

|fni (x)−fnj (x)| 6 |fni (x)−fni (ak )|+|fni (ak )−fni (yk )|+|fni (yk )−fnj (yk )|

+|fnj (yk ) − fnj (ak )| + |fnj (ak ) − fnj (x)| 6 5.


Thus the sequence {fnj }j>1 is Cauchy in C (X ). Since C (X ) is
complete, {fnj }j>1 is convergent.
Problem
Let (X , d) be a compact metric space and let F : X × X → R be a
continuous function. For y ∈ X , define fy (x) = F (x, y ). Show that
the family {fy }y ∈X is bounded and equicontinuous.

Solution.
Note that kfy k∞ = supx∈X |F (x, y )| 6 ||F ||∞ < ∞ (since F is
continuous and X × X is compact; see slide 134). Thus
supy ∈X kfy k∞ < ∞, and hence {fy }y ∈X is bounded.
Since F is continuous on a compact metric space, F is uniformly
continuous. Thus, for given  > 0, there exists δ > 0 such that
|fy (x) − fy 0 (x 0 )| <  whenever x, x 0 , y , y 0 ∈ X , d(x, x 0 ) < δ and
d(y , y 0 ) < δ. Let y 0 = y .

• By (Generalized) Arzela-Ascoli Theorem, any sequence in


{fy }y ∈X has a subsequence convergent in C (X ).
Connected metric spaces

Definition
Let (X , d) be a metric space. We say that X is connected if the
only subsets of X , which are both open and closed are ∅ and X .

Example
We check that R is connected. Write R = A t B (disjoint union)
for nonempty, open sets A and B. Let a ∈ A, b ∈ B with a < b.
• [a, b] = A0 t B0 , where A0 = A ∩ [a, b] and B0 = B ∩ [a, b].
• Let c = sup A0 ∈ [a, b]. Thus either c ∈ A0 or c ∈ B0 .
• If c ∈ A0 then either c = a or a < c < b (since A is closed).
• Since A is open, there is r > 0 such that [c, c + r ) ⊆ A.
This contradicts that c = sup A0 . Similarly, prove that c ∈
/ B0 .

Problem
Show that for any a ∈ R, R \ {a} is not connected.
Example
Any normed linear space X over R is connected. To see this,
assume that X = A t B such that A and B are open subsets of X .
We claim that one of the A and B is empty. If not, then for a ∈ A
and b ∈ B, consider f : R → X given by

f (t) = (1 − t)a + tb, t ∈ R.

Then f (0) = a and f (1) = b. Since f is continuous, f −1 (A) and


f −1 (B) are open subsets of R. Since these are proper and disjoint,
we arrive at the conclusion that R is not connected ⇒⇐

Example
The space Q is not connected. In fact, for any irrational x ∈ R,
the set [−x, x] ∩ Q = (−x, x) ∩ Q is both open and closed in Q.
Theorem
Let f : X → Y be continuous. If X is connected, then so is f (X ).

Proof.
Suppose f (X ) has a nonempty open and closed subset, say, U.
Then f −1 (U) is an open and closed subset of X . However, X is
connected, so f −1 (U) = ∅ or f −1 (U) = X . Since f −1 (U) 6= ∅, we
conclude that f −1 (U) = X , and hence U = f (X ).

• A path connecting points x, y is a continuous function


f : [a, b] → X such that f (a) = x and f (b) = y . Since [a, b] is
connected, so is f ([a, b]).

Problem
Show that union of line segments L1 , . . . , Lm in a normed linear
space is connected if L1 ∩ L2 6= ∅, L2 ∩ L3 6= ∅, . . . , Lm ∩ L1 6= ∅.
Example
Let us see that Rn \ {0} is connected. Suppose there exists a
proper set U which is both open and closed in Rn \ {0}.
• Let x ∈ U and y ∈ (Rn \ {0}) \ U.
• x and y can be connected by union L of line segments.
• L contains open and closed proper subset L ∩ U.
However, L being connected, L ∩ U = L or L ∩ U = ∅ ⇒⇐

Example
The unit sphere Sn = {x ∈ Rn : kxk2 = 1} in Rn is connected.
Indeed, Rn \ {0} is connected and g : Rn \ {0} → Sn given by

g (x) = x/kxk2 , x ∈ Rn \ {0}

is a continuous surjection.
Theorem
Let (X , d) be a metric space and let A ⊆ X . If A is connected and
B is a set such that A ⊆ B ⊆ A, then B is also connected.

Proof.
Let B = C t D for open sets C and D.
• A = (A ∩ C ) t (A ∩ D) for open sets A ∩ C and A ∩ D.
• Since A is connected, either A ∩ C = ∅ or A ⊆ C .
• Thus either A ⊆ C or A ⊆ D.
• Since C and D are closed in B, we obtain A ∩ B ⊆ C or
A∩B ⊆D
It follows that either C = B or D = B. However, in this case,
either C = ∅ or D = ∅, which shows that B is connected.

Corollary
The closure of a connected set is connected.
Example (Topologist’s sine curve)
Consider the continuous function s : (0, 1] → [0, 1] given by

s(x) = sin(1/x), x ∈ (0, 1].

Let S denote the graph of s given by

S = {(x, s(x)) ∈ R2 : x ∈ (0, 1]}.

Note that S = f ((0, 1]), where f (x) = (x, s(x)) is a continuous


function. Thus S is connected, and hence S is also connected.
The metric space S (with relative metric induced by d2 ) is
commonly known as the topologist’s sine curve.
Path-connected metric spaces
Definition
Let (X , d) be a metric space. We say that X is path-connected if
for any two points x, y ∈ X , there exists a continuous function
f : [0, 1] → X such that f (0) = x and f (1) = y .
Remark If U is path-connected then for open sets V , W of U such
that U = V t W and for any path f in U, the range of f being the
continuous image of the connected set [0, 1] is connected, and
hence lies entirely either in V or W . This shows that one of V , W
must be empty, that is, U is connected.
Problem
Show that the complement of any countable subset C in R2 is
path-connected.
Hint.
Given p, q ∈ R2 \ C, consider the uncountable set F given by
{f : f is a path connecting p and q}.
Example (Comb space)
Let K denote the set {1/n : n ≥ 1} and consider

E = ([0, 1] × {0}) ∪ (K × [0, 1]) ⊆ R2 .

• E is path-connected (see diagram).


The comb space C is defined to be the space E ∪ ({0} × [0, 1]).
• C is also connected (since E is connected and E = C)
The deleted comb space C0 is defined as E ∪ {(0, 1)}.
• C0 is connected (since E ⊆ C0 ⊆ E )
• C0 is not path-connected (since there is no path which connects
the points p = (0, 1) and q = (1, 0))
Theorem
The deleted comb space is not path-connected.

Proof.
Suppose, contrary to this, that there is a path γ : [0, 1] → C0 such
that γ(0) = p and γ(1) = q.
• γ −1 ({p}) is a closed subset of [0, 1]. and hence it is compact.
Let t0 ∈ [0, 1] be its maximum.
• Consider the projection P1 (x, y ) = x of R2 onto the X -axis.
• Let {tn }n>1 ⊆ (t0 , 1] be a sequence converging to t0 .
If, for every n > 1, there exists t0 < sn < tn such that
γ(sn ) = (xn , 0) for some xn ∈ [0, 1] \ K , then {sn }n>1 converges to
t0 , by the continuity, (xn , 0) = γ(sn ) → γ(t0 ) = p = (0, 1) ⇒⇐
• There exists t1 ∈ (t0 , 1] such that (P1 ◦ γ)(t0 , t1 ) ⊆ K .
• (P1 ◦ γ)(t0 , t1 ) is a connected subset of K containing 1 (wlog)
Thus (P1 ◦ γ)(t0 , t1 ) = {1}. By continuity, (P1 ◦ γ)[t0 , t1 ) = {1},
which is impossible since (P1 ◦ γ)(t0 ) = 0.
Theorem
Let f : X → Y be a continuous surjection. If X is path-connected,
then so is Y .

Proof.
Let y0 , y1 ∈ Y . Let x0 , x1 ∈ X be such that f (x0 ) = y0 and
f (x1 ) = y1 . If γ is a path connecting x0 and x1 , then f ◦ γ is a
path connecting y0 and y1 .

If p is a polynomial in the real variables x1 , · · · , xn and


Z (p) = {x = (x1 , . . . , xn ) ∈ Rd : p(x) = 0}, then Rn \ Z (p) is not
necessarily path-connected.
• If p(x, y ) = x 2 + y 2 − 1, then Z (p) is equal to

{(x, y ) ∈ R2 : x 2 + y 2 = 1}.

Clearly, R2 \ Z (p) is not connected.


Corollary
If p is a polynomial in the complex variables z1 , · · · , zn and
Z (p) = {(z1 , . . . , zn ) ∈ Cn : p(z) = 0}, then Cn \ Z (p) is
path-connected.

Proof.
Let z, w ∈ Cn \ Z (p). Consider the straight-line path

γ(t) = (1 − t)z + tw , t ∈ C.

• Z = {t ∈ C : γ(t) ∈ Z (p)} is the set of zeros of p ◦ γ


• Z is a finite subset of C (p ◦ γ is a polynomial in one variable)
• C \ Z is path-connected (see slide 158)
• γ maps C \ Z continuously into Cn \ Z (p)
In particular, z and w belong to the path-connected subset
γ(C \ Z ) of Cn \ Z (p).
Example
The general linear group GLn (C) of all invertible n × n matrices
with complex entries is path-connected.
2
• Define f : GLn (C) → Cn by

f (A) = (a11 , . . . , a1n , . . . , an1 , . . . , ann ), A ∈ GLn (C).

• f is a (linear) homeomorphism
2
• f maps GLn (C) onto Cn \ Z (det), where det is the complex
polynomial in the variables ai,j , 1 6 i, j 6 n, which sends f (A)
to the determinant of A
2
Thus GLn (C) is path-connected (since so is Cn \ Z (det)).

Problem
Show that GLn (R) is not connected.
Problem
For an open subset U of Rn , show that U is connected if and only
if U is path-connected.

Solution.
We have already seen that path-connected space is connected. To
see the converse, consider for any p ∈ U, the set

S = {x ∈ U : there is a path connecting p and x}.

We claim that S is the whole of U.


• S is nonempty (since p ∈ U)
• S is open (if x ∈ S and Br (x) ⊆ U for some r > 0, then any
y ∈ Br (x) can be connected to x and x can be connected to
p, so y can be connected to p ⇒ Br (x) ⊆ S)
• S is closed (let {xn }n>1 be a sequence in S converging to x,
connect x and xN (for large N) and xN to p)
Since U is connected, S = U.
Pointwise convergence
Example
For n > 1, consider the function fn (x) = x n for x ∈ [0, 1]. Note
that {fn }n>1 converges pointwise to f , where f (x) = 0 for
x ∈ [0, 1) and f (1) = 1. Thus the pointwise limit of a sequence of
continuous functions is not necessarily continuous.
Problem
For m > 1, consider the function fm (x) = limn→∞ (cos(m!πx))n for
x ∈ R. Verify the following:
(1) {fm }m>1 converges pointwise to f , where f (x) = 0 if
x ∈ R \ Q, and f (x) = 1 for x ∈ Q.
(2) f is discontinuous everywhere.
Hint.
If x = p/q ∈ (0, 1), q 6= 0 then limn→∞ (cos(m!πx))n = 1 for every
m > q, and hence f (p/q) = 1. If x ∈ / Q, then cos(m!πx) < 1, and
hence limn→∞ (cos(m!πx))n = 0 and f (x) = 0.
For a metric space (X , d), consider the vector space problem of all
bounded functions f : X → R.
• B(X ) is a normed linear space with norm kf k∞ = supx∈X |f (x)|
Theorem
B(X ) is a complete normed linear space.
Proof.
Let {fn }n>1 be a Cauchy sequence in B(X ).
• supn>1 kfn k∞ < ∞ (every Cauchy sequence is bounded)
• For any x ∈ X , {fn (x)}n>1 is Cauchy. Indeed,

|fm (x) − fn (x)| 6 kfm − fn k∞ , m, n > 1

• Define f (x) = limn→∞ fn (x), x ∈ X


• f ∈ B(X ) (since kf k∞ 6 kfm − fn k∞ + kfn k∞ )
Given  > 0, there exists N > 1 such that kfm − fn k <  for all
m, n > N. Thus for every x ∈ X , |fm (x) − fn (x)| 6  for every
m, n > N. Now let n → ∞ and take supremum over X .
Problem
Let (X , d) be a metric space and fix z ∈ X . For every x ∈ X ,
define fx (y ) = d(x, y ) − d(y , z). Verify the following:
(1) For each x ∈ X , fx ∈ B(X ).
(2) F : X → B(X ) by F (x) = fx satisfies kF (x)k∞ 6 d(x, z).
(3) For every x, y ∈ X , kF (x) − F (y )k∞ = d(x, y ).
Conclude that X is homeomorphic to (F (X ), k · k∞ ).

Solution.
(1) follows from |fx (y )| 6 d(x, z), while (2) follows from (1). To
see (3), note that for any x, y ∈ X , by the triangle inequality,

kF (x) − F (y )k∞ = sup |d(x, w ) − d(y , w )| 6 d(x, y ).


w ∈X

Since |fx (x) − fy (x)| = | − d(x, z) − d(y , x) + d(x, z)| = d(x, y ),


(3) follows.

• The subspace F (X ) of B(X ) is said to be the completion of X .


Uniform convergence

Let (X , d) be a metric space. For n > 1, let fn , f : X → R be such


that fn − f ∈ B(X ). The sequence {fn }n>1 converges uniformly to
f if
kfn − f k∞ = sup |fn (x) − f (x)| → 0 as n → ∞.
x∈X
Theorem
Let (X , d) be a compact metric space. Let {fn }n>1 be a sequence
of continuous functions on X . If {fn }n>1 converges uniformly to f
on X , then f is continuous.

Proof.
This follows from the fact that C (X ) is complete metric space.

• In general, pointwise convergence ; uniform convergence


Theorem
Let {fn }n>1 be a sequence of continuous functions. If {fn }n>1
converges uniformly to f on [a, b] then
Z b Z b
lim fn (x)dx = f (x)dx.
n→∞ a a

Proof. Rb Rb Rb
Note that | a fn (x)dx − a f (x)dx| = | a (fn (x) − f (x))dx| 6
Rb
a |fn (x) − f (x)|dx 6 kfn − f k∞ (b − a).

Problem
For n > 1, consider the function fn (x) = nx(1 − x 2 )n for x ∈ [0, 1].
Verify that {fn }n>1 converges pointwise to f , where f (x) = 0 for
all x ∈ [0, 1], but {fn }n>1 does not converge uniformly to f .

Hint. R1 R1
For second part, use limn→∞ 0 fn (x)dx 6= 0 f (x)dx.
Dini’s Theorem
Sometimes pointwise convergence ⇒ uniform convergence.
Theorem
Let (X , d) be a compact metric space. Let {fn }n>1 be a sequence
in C (X ) converging pointwise to a continuous function f . If
{fn (x)}n>1 is decreasing for all x ∈ X , then {fn }n>1 converges
uniformly to f .
Proof.
Let gn = fn − f > 0. For  > 0, let Kn = {x ∈ X : gn (x) > }.
• Kn is compact and Kn+1 ⊆ Kn (since Kn is closed, gn > gn+1 )
• If each Kn 6= ∅, then so is finitely many sets from {Kn }n>1
• If each Kn 6= ∅, then by the finite intersection property (see
[Assignment 7, Exercise 2]), ∩∞ n=1 Kn 6= ∅
However, if x ∈ X , then since gn (x) → 0, x ∈
/ Kn for sufficiently
large n. Hence KN is empty for some N, that is, 0 6 gn (x) <  for
every x ∈ X and for every n > N.
Example
Define a sequence {pn }n>0 of polynomials by p0 (x) = 0, and

pn+1 (x) = pn (x) + (x 2 − pn (x)2 )/2, n > 0.

A routine calculation shows that

|x| − Pn+1 (x) = (|x| − Pn (x))(1 − (|x| + Pn (x))/2), n > 0.

One may now verify inductively that

0 6 pn (x) 6 pn+1 (x) 6 |x|, x ∈ [−1, 1], n > 0.

In particular, {pn (x)}n>0 converges pointwise to |x|. Now apply


Dini’s Theorem to fn = −pn , n > 0 to conclude that {pn }n>0
converges uniformly to f (x) = |x| on [−1, 1].
Problem
Show that the function g : R → R given by
(
0 if x 6 0,
g (x) =
x if x > 0.

Show that for any α > 0, g can be uniformly approximated by


polynomials on [−α, α].
Hint.
Note that g (x) = 12 (x + |x|).
Problem
Let {pn }n>1 be a sequence of polynomials of degree dn . Suppose
that kpn − f k∞ → 0 as n → ∞ for some f ∈ C ([a, b]). If f is not
a polynomial, then show that dn → ∞ as n → ∞.
Hint.
The subspace of polynomials of degree less than or equal to d is
finite-dimensional, and hence it is closed in C ([a, b]).
Theorem
• P = {f ∈ B(X ) : ∃ {pn }n>1 such that kpn − f k∞ → 0} ⊆ C (X ).
• There are compact metric spaces X with P ( C (X ).
The first part follows from the completeness of C (X ).

Example (Failure of polynomial approximation)


Let X = {z ∈ C : |z| = 1} with d(z, w ) = |z − w |, z, w ∈ C.
• X is a compact metric space (X ⊆ C is closed and bounded)
• Consider the continuous function f (z) = z (C-conjugate)
R 2π
• If γ(t) = e it , 0 6 t 6 2π, then 0 p(γ(t))γ 0 (t)dt = 0 for any
R 2π
polynomial p and 0 f (γ(t))γ 0 (t)dt = 1
• If {pn }n>1 is a sequence of complex polynomials pn in z such
that kpn − f k∞ → 0, then
Z 2π Z 2π
0
pn (γ(t))γ (t)dt → f (γ(t))γ 0 (t)dt as n → ∞ ⇒⇐
0 0

Thus f ∈ C (X ) \ P.
Lemma
Let (X , d) be a compact metric space. The following are true:
(1) P is a complete normed linear space.
(2) Given f ∈ C (X ), if, for every  > 0, there exists g ∈ P such
that kf − g k∞ < , then f ∈ P.

Proof.
(1): Clearly, linear combination of functions approximated
uniformly by polynomials are again in P. Since P is a closed subset
of C (X ) and C (X ) is complete, P is complete.
(2): Every ball B (f ) intersects P, and hence f is a limit point of
P. Thus f ∈ P.

Theorem (Weierstrass’ approximation theorem)


Any f ∈ C ([a, b]) can be approximated uniformly by polynomials.
Lebesgue’s Proof of Weierstrass’ Theorem*8 .
Since f is uniformly continuous on [a, b], there exists an integer
N > 1 such that |f (x) − f (y )| <  whenever |x − y | < 1/N. For
xi := a + (b − a)(i/N) (i = 0, · · · , N), consider the function h(x)
with graph equal to a polygon (see diagram) of vertices at
(a, f (a)), (x1 , f (x1 )), · · · , (xn−1 , f (xn−1 )), (b, f (b)).
• kf − hk∞ < . Indeed, if x ∈ (xi , xi+1 ) then

f (x) − h(x) = f (x) − ((1 − t)f (xi ) + tf (xi+1 ))


= (1 − t)(f (x) − f (xi )) + t(f (x) − f (xi+1 )) ⇒ |f (x) − h(x)| < 
• If h(x) = f (a) + N−1
P
i=0 ci g (x − xi ) (x ∈ [a, b]) for some
scalars c0 , · · · , cN−1 and g (x) = 12 (x + |x|)
Since g can be approximated uniformly by polynomials (see slide
172), h ∈ P, and hence f ∈ P (see the lemma on last slide).

8
J. Burkill, Lectures on Approximation by Polynomials, Lecture Notes, TIFR, Bombay, 1959
Problem
Use Weierstrass’ Theorem to show that C ([a, b]) is separable.

Hint.
Polynomials with rational coefficients are countable and dense.

Problem Rb
Let f ∈ C [a, b] be such that a t n f (t)dt = 0 for all non-negative
integers n. Show that f (t) = 0 for every t ∈ [a, b].

Solution. R
b
Note that a p(t)f (t)dt = 0 for any polynomial p. If {pn } is a
Rb
sequence converging uniformly to f , then a f (t)2 dt = 0, and
hence f = 0.
Problem
Let f : [a, b] → R be a continuous function. Show that there exists
a sequence {pn }n>1 of polynomials such that
Z b
|pn (x) − f (x)|p dx → 0 as n → ∞.
a
Recall that χA is 1 on A and 0 outside A.
Problem
For 0 ≤ x < y ≤ 1, consider χ[x,y ] : [0, 1] → R. Show that there
exists a sequence {pn } of polynomials such that
Z 1
|pn (x) − χ[x,y ] (x)|dx → 0 as n → ∞.
0

Conclude that finite linear combination of indicator functions of


subintervals of [0, 1] can be approximated by polynomials.
Hint.
Approximate χ[x,y ] by continuous functions.
Problem
Let f : [a, b] → R be a continuouly differentiable function. Show
that there exists a sequence {rn }n>1 of polynomials such that

krn − f k∞ → 0 and krn0 − f 0 k∞ → 0 as n → ∞.

Conclude that C 1 [a, b] (the space of continuously differentiable


functions on [a, b]) is a separable normed linear space with norm
kf k := kf k∞ + kf 0 k∞ .
Solution.
Let g (x) = f (x) − f (a) and note that g 0 = f 0 . Find a sequence
{qn }n>1 of 0
R xpolynomials such that0 kqn − g k∞ → 0. Set
pn (x) := a qn (t)dt. Note that pn = qn , and hence
kpn0 − g 0 k∞ → 0. Also,
Z x Z x
|pn (x) − g (x)| = qn (t)dt − g 0 (t)dt ≤ (b − a)kqn − g 0 k∞ .
a a

Let rn (x) := pn (x) + f (a).


Pointwise limit of a sequence of continuous functions

• We have seen that the pointwise limit of sequence of functions


can be discontinuous at every point (see slide 165)
Question Can the pointwise limit of sequence of continuous
functions be discontinuous at every point ?
Answer No
A subset A of a metric space (X , d) is said to be nowhere dense if
the interior of the closure of A is empty, that is, (A)◦ = ∅.

Theorem (Baire-Osgood Theorem)


If f : [a, b] → R is a pointwise limit of a sequence of continuous
functions on [a, b], then the set D(f ) of discontinuities of f is a
countable union of closed nowhere dense sets.

For a proof of this theorem, we need some preliminaries.


Oscillation
Let f : [a, b] → R and let I (c, r ) = (c − r , c + r ) for c ∈ [a, b] and
r > 0. Define the oscillation of f on I (c, r ) by

osc(f , c, r ) = sup |f (x) − f (y )|.


x,y ∈[a,b]∩I (c,r )

• osc(f , c, r ) exists if f is bounded


• osc(f , c, r ) > 0 and osc(f , c, r ) is decreasing in r
Define the oscillation of f at c by
(
limr →0 osc(f , c, r ) if f is bounded near c
osc(f , c) =
∞ otherwise.

Problem
Show that f is continuous at c if and only if osc(f , c) = 0.
Hint.
The delta in the definition of continuity plays here the role of r .
For  > 0, consider the set

A = {c ∈ [a, b] : osc(f , c) > }.

• The set D(f ) of discontinuities of f is equal to ∪>0 A .


Lemma
For every  > 0, A is a compact subset of [a, b].

Proof.
Let {cn }n>1 be a sequence in A converging to c ∈ [a, b].
• If c ∈
/ A , then δ =  − osc(f , c) > 0 and hence
osc(f , c, r ) <  − δ/2 for some r
• If |cn − c| < r /2, then I (cn , r /2) ⊆ I (c, r ), and hence
osc(f , cn , r /2) = sup |f (x) − f (y )| <  (for c ∈
/ A )
x,y ∈[a,b]∩I (cn ,r /2)
Since osc(f , cn ) 6 osc(f , cn , r /2), osc(f , cn ) <  ⇒⇐ (for cn ∈ A )
Thus A is closed and hence it is compact.
Proof of Baire-Osgood Theorem.
Note that D(f ) = ∪n>1 A1/n . Since A is closed for every  > 0, it
suffices to check that A◦ = ∅.
Claim For any closed interval J ⊆ [a, b], J * A
• En = ∩i,j>n {x ∈ [a, b] : |fi (x) − fj (x)| 6 /5} is closed (since
fi − fj is continuous for every i, j > n)
• ∪n>1 En = [a, b] (since limn→∞ fn (x) = f (x), x ∈ [a, b])
• J = ∪∞
n=1 (En ∩ J)
• There exists N > 1 such that EN ∩ J has nonempty interior
(since J is complete, BCT is applicable, see slide 103)
Thus there exists an open interval K contained in EN ∩ J.
SubClaim: K ⊆ [a, b] \ A (⇒ J * A , and Claim follows)

• |fi (x) − fj (x)| 6 /5 for all x ∈ K and i, j > N (since K ⊆ EN )


• |fN (x) − f (x)| 6 /5 for all x ∈ K (let i = N, fj (x) → f (x))
Proof of Baire-Osgood Theorem continued ...
• For x0 ∈ K , there exists δ > 0 such that
|fN (x) − fN (x0 )| < /5 whenever x ∈ K and |x − x0 | < δ (by
the continuity of fN at x0 )
• |f (x) − fN (x0 )| 6 2/5 for every x ∈ K such that |x − x0 | < δ
• |f (x) − f (y )| 6 4/5 for every x, y ∈ K such that |x − y | < δ
This shows that osc(f , x0 , δ) 6 4/5, and hence osc(f , x0 ) < .
Thus the claim K ⊆ [a, b] \ A stands verified.

Corollary
If f : [a, b] → R is a pointwise limit of a sequence of continuous
functions on [a, b], then the set of continuities of f (that is,
[a, b] \ D(f )) is a dense subset of [a, b].

Proof.
Since [a, b] \ D(f ) = ∩n>1 ([a, b] \ A1/n ) and each [a, b] \ A1/n is
dense in [a, b], BCT yields the desired conclusion.
Example
Consider the function f : [a, b] → R such that f = 1 on rationals
and f = 0 on irrationals. Note that D(f ) = [a, b]. It follows from
the Baire-Osgood Theorem that f can not be a pointwise limit of a
sequence of continuous functions.

Problem
If f : [a, b] → R is continuous and f : (a, b) → R is differentiable,
then show that the set of continuities of f 0 is dense in (a, b).

Solution.
For any x ∈ (a, b), note that f 0 (x) = limn→∞ f (x+1/n)−f
1/n
(x)
. Thus
f 0 is a pointwise limit of continuous functions. By Baire-Osgood
Theorem, the set of continuities of f 0 is dense in any closed
interval [c, d] contained in (a, b). Hence it is dense in (a, b).
Convergence of series of functions
Let (X , d) be a metric space. Let {fn }n>1 be a sequence of
bounded functions Pand let f : X → R be a bounded function. We
say that the series ∞ n=1 fn converges uniformly to f if the
Pk
sequence { n=1 fn }k>1 converges uniformly to f , that is,
k
X k
X
k fn − f k∞ = sup | fn (x) − f (x)| → 0 as k → ∞.
n=1 x∈X n=1

P∞
In this case, we write f = n=1 fn .
Theorem
Let (X , d)Pbe compact. Let {fn }n>1 be a sequence of functions in
C (X ). If ∞ n=1 fn converges uniformly to f , then f ∈ C (X ).

Proof.
If each fn is continuous, then so is kn=1 fn . Now use the fact that
P
uniform limit of continuous functions is continuous.
Problem
Let (X , d) be a compact metric Pspace and let {fn }n>1 be a
sequence in C (X ). Assume that n>1 fn (x) converges pointwise to
some f ∈ C (X ) forPevery x ∈ X . If fn (x) > 0 for every x ∈ X and
every n > 1, then n>1 fn is uniformly convergent.

Hint.
Use Dini’s Theorem.

Problem
Let (X , d) be a compact metric space and let A be a closed subset
of X . Assume that X \ A = t∞ n=1 Xn (disjoint union), where each
Let f ∈ C (X ) be such that f (a) = 0 for
Xn is a clopen set in X . P
every a ∈ A. Show that ∞ n=1 f χXn converges uniformly to f .

Hint.
Note that each χXn is continuous. Now use the last problem.
Weierstrass M-test

Theorem
P {fn }n>1 be a sequence of
Let (X , d) be a metric space and let
bounded functions
P fn : X → R. If n>1 kfn k∞ < ∞ (convergence
in R), then n>1 fn is uniformly convergent. Moreover,
X X
k fn k∞ 6 kfn k∞ .
n>1 n>1

Proof.
For n > 1, let gn = nk=1 fn and g0 = 0. For a positive integers
P
m < n, note that by triangle inequality,
X n Xn
kgn − gm k∞ = k fk k ∞ 6 kfk k∞ , (1)
k=m+1 k=m+1
and hence {gn }n>1 is Cauchy in B(X ). Since B(X ) is complete
(see slide 166), {gn }n>1 is uniformly convergent. Let m = 0,
n → ∞ in (1) to get the remaining part.
Power Series

Definition
A power series is an expansion of the form

X
an x n , where an ∈ R.
n=0
P∞ n
P∞ n
n=0 an x converges absolutely if n=0 |an ||x| < ∞.

Definition (Domain of Convergence)


P∞ n
D = {w ∈ R : n=0 |an ||w | < ∞}
Note that
• w0 ∈ D =⇒ ±w0 ∈ D for any θ ∈ R
• w0 ∈ D =⇒ w ∈ D for any w ∈ R with |w | 6 |w0 |
Conclude that D is either R, (−R, R) or [−R, R] for some R > 0.
Definition
The radius of convergence (for short, RoC) of ∞ n
P
n=0 an x is
defined as
X∞
R = sup{|x| : |an ||x|n < ∞}.
n=0
P∞ n
Remark The series n=0 an x converges uniformly on [−r , r ] for
any 0 < r < R.
Example (Geometric series)
P∞ n,
Consider the series n=0 x x ∈ R.
1
• This series converges absolutely to 1−x for any |x| < 1.
Indeed,
n
X 1 − x n+1
xk = , x n+1 → 0 as n → ∞
1−x
k=0

• Thus R > |x| for every |x| < 1, and hence R > 1
• R = 1 (since ∞ n
P
n=0 x diverges at x = 1)
Theorem (Cauchy-Hadamard Formula)
P∞ n
The RoC of n=0 an x is given by

1
R= ,
lim sup |an |1/n

where we use the convention that 1/0 = ∞ and 1/∞ = 0.


Proof.
Assume R < ∞. If r > R, then lim sup |an |1/n > 1/r , and hence
limk→∞ supn>k |an |1/n > 1/r . Thus there exists a subsequence
{n
Pk∞}k>1 such that |ank |1/nk > 1/r , that is, r nk |ank | 9 0, and hence
n 1/n .
n=0 an r is divergent. Thus RoC ≯ R = lim sup |an |

If r < R, then |an |r n < 1 for all integers n > N. Thus, for |x| < r ,
∞ ∞ 
X X |x| n 1
|an ||x|n 6 6 < ∞.
r 1 − |x|/r
n=N n=N

Thus RoC > r for any r < R or RoC > R = lim sup |an |1/n .
Examples

Pk
• n=0 an x n , an = 0 for n > k, R = ∞
P∞ xn 1
• n=0 n! , an = n! , R = ∞
P∞ n
• n=0 x , an = 1, R = 1
P∞ n
• n=0 n!x , an = n!, R = 0

The coefficients of a power series may not be given by a single


formula.
Example
P∞ n2
Consider the power series n=0 x . Then

ak = 1 if k = n2 , and 0 otherwise.

Clearly, lim sup |an |1/n = 1, and hence R = 1.


Sometimes RoC can be computed without knowing the coefficients
explicitly.
Example
Consider the power series ∞ n
P
n=0 an x , where an is number of
divisors of n 1111 . Note that

1 ≤ an ≤ n1111 .

≤ lim sup |an |1/n ≤ lim sup(n1111 )1/n = 1, and hence


Note that 1P
the RoC of ∞ n
n=0 an x equals 1.

Theorem P∞
If the RoC of n
P∞ n=0 an x is R then the RoC of the power series
n−1
n=1 nan x is also R.

Proof.
1 1
Since limn→∞ n1/n = 1, R = lim sup |nan |1/n
= lim sup |an |1/n
.
Power series is infinitely differentiable
Theorem P
If f (x) = ∞ n
n=0 an x is a power series of radius
P R > 0, n−1then f is
infinitely differentiable with f 0 (x) = g (x) = ∞n=1 nan x .
Proof.
Let x0 ∈ (−R, R), h ∈ R, r > 0 with max{|x0 |, |x0 + h|} < r < R.
• Sk (x) = kn=0 an x n , Ek (x) = ∞ n
P P
n=k+1 an x
f (x0 +h)−f (x0 )
• h − g (x0 ) = A + (Sk0 (x0 ) − g (x0 )) + B, where
 S (x + h) − S (x )   E (x + h) − E (x ) 
k 0 k 0 k 0 k 0
A= −Sk0 (x0 ) , B =
h h
∞ ∞
X (x0 + h)n − x0n X
• |B| 6 |an | 6 |an |nr n−1
h
n=k+1 n=k+1
Since f 0 is a power series, f is infinitely differentiable on (−R, R).
A nowhere differentiable continuous function
• Any continuous function on [0, 1] is a uniform limit of infinitely
differentiable functions (Weierstrass’ Theorem).
It is quite striking that uniform limit of infinitely real differentiable
functions could be nowhere differentiable.
• Let φ(x) = |x| for x ∈ [−1, 1], which is extended periodically
(with period 2) to R by setting φ(x + 2) = φ(x), x ∈ R
• φ is a continuous function such that 0 ≤ φ(x) ≤ 1, x ∈ R
• Plot graphs of φ(x), φ(4x) (period 1/2), φ(16x) (period 1/8)
3 n
• Define the function f : R → R by f (x) = ∞ φ(4n x)
P 
n=0 4
Note that f is continuous. Indeed,
k  n ∞  n ∞  n
X 3 n
X 3 n
X 3
f (x) − φ(4 x) ≤ φ(4 x) ≤ → 0.
4 4 4
n=0 n=k+1 n=k+1

Since φ is continuous, so is f .
Theorem
For any x ∈ R, there exists {δm }m>1 converging to 0 such that
|(f (x + δm ) − f (x))/δm | → ∞ as m → ∞.
In particular, f is not differentiable at any point in R.
Proof.
For an integer m > 1, set δm = ± 12 4−m , where the sign is so
chosen that no integer lies between 4m x and 4m (x + δm ). Define
γn = (φ(4n (x + δm )) − φ(4n x))/δm .
• If n > m, then φ(4n (x + δm )) = φ(4n x ± 4n−m /2) = φ(4n x),
and hence γn = 0.
• |γm | = 4m (since 4n (x + δm ) and 4n x are both > 0 or < 0)
• When 0 ≤ n < m,

||4n (x + δm )| − |4n x|| |4n δm |


|γn | = ≤ = 4n
|δm | |δm |

Now we complete the argument.


Proof continued ...
Since γn = 0 for n > m, |γm | = 4m and |γn | 6 4m , 0 ≤ n < m,
∞  n m  n
X 3 X 3
|(f (x + δm ) − f (x))/δm | = γn = γn
4 4
n=0 n=0

m−1 m−1
X 3 n
 X 1
= γn ± 3m ≥ 3m − 3n = (3m + 1),
4 2
n=0 n=0

where we used m−1 n m


P
n=0 3 = (3 − 1)/2.

Thus (f (x + δm ) − f (x))/δm blows up to ∞ as m → ∞.

Geometrically, a continuous nowhere differentiable function has a


continous graph with ”corner” at every point!
• The set of nowhere differentiable continuous functions turns out
to be dense in C [0, 1] (this may be deduced from BCT).
Riemann integral
Let I = [a, b] and f : I → R be a bounded function.
• P = {x0 = a, . . . , xn = b} (partition)
n−1
• Ii = [xi , xi+1 ], [a, b] = ∪i=0 Ii , `(Ii ) (length of Ii )
• kPk = max `(Ii ) (width of P)
06i6n−1
• mi = inf f (x), Mi = sup f (x)
x∈Ii x∈Ii
n−1
X
• L(P, f ) = mi `(Ii ) (lower Riemann sum)
i=0
n−1
X
• U(P, f ) = Mi `(Ii ) (upper Riemann sum)
i=0
R
• –
f (x)dx = sup L(P, f ) (lower Riemann integral)
P
R–
• f (x)dx = inf U(P, f ) (upper Riemann integral)
P
Definition
A bounded function f :RI → R is Riemann
R– R (or
integrable
Darboux integrable) if – f (x)dx = f (x)dx, say I f (x)dx.

Interpretation Area of the region R enclosed by the lines x = a,


x = b, and the curve y = f (x) > 0.
Example
For x0 ∈ I , consider the function f : I → R defined by
(
1 if x 6= x0 ,
f (x) =
0 if x = x0 .
For  > 0, let P be a partition with kPk = max `(Ii ) < .
06i6n−1
• If x0 ∈ Ii0 is an interior point, then U(P, f ) − L(P, f ) = `(Ii0 ) < 
• If x0 ∈ Ii0 isR an end point,
R then U(P, f ) − L(P, f ) < 2
• L(P, f ) 6 – f (x)dx 6 – f (x)dx 6 U(P, f ) (Exercise)
Conclude that f is Riemann integrable.
Theorem
If f : [a, b] → R is a bounded function, then f is Riemann
integrable if and only if for every  > 0, there exists a partition P
of [a, b] such that U(P, f ) − L(P, f ) < .

Proof.
The sufficiency part ⇐ follows from
Z Z –
L(P, f ) 6 f (x)dx 6 f (x)dx 6 U(P, f ).

To see the necessity


R part ⇒, find partitions
R P and Q of [a, b] such
that U(P, f ) − – f (x)dx < /2 and – f (x)dx − L(Q, f ) < /2.
Let R be a partition obtained from P ∪ Q, and note that

L(Q, f ) 6 L(R, f ) 6 U(R, f ) 6 U(P, f ).

Then U(R,Rf ) − L(R, f )R6 U(P, f ) − L(Q, f ) =


U(P, f ) − – f (x)dx + – f (x)dx − L(Q, f ) < .
Corollary
Every continuous function f : I → R is Riemann integrable.
Proof.
Given  > 0, there exists δ > 0 such that |f (x) − f (x 0 )| < 
whenever x ∈ I , |x − x 0 | < δ. Thus for partition P with kPk < δ,
m−1
X
U(P, f ) − L(P, f ) 6 |Mi − mi |`(Ii ).
i=0

However, Mi = f (ai ) and mi = f (bi ) for some ai , bi ∈ Ii . Thus


m−1
X
U(P, f ) − L(P, f ) = |f (ai ) − f (bi )|`(Ii ) 6 `(I ). Hence f is
i=0
Riemann integrable.

Problem
Show that the indicator function χ[0,1/2] : [0, 1] → R (which is 1 on
[0, 1/2] and 0 outside) is Riemann integrable.
Problem
Show that a bounded function, which is continuous except at a
finite subset F of [a, b], is Riemann integrable.

Hint.
Consider first the case in which F = {x0 }. Assuming f 6= 0, let
I = (x0 − 4kfk∞ , x0 + 4kfk∞ ). Apply the last corollary to
I = [a, b] \ I to find a partition P of I such that
U(P, f ) − L(P, f ) < /2. Let P̃ be the partition formed by taking
union of P and the interval I  . Check that U(P, f ) − L(P, f ) < .
Extend this argument to any finite set F .

Question Does there exist a Riemann integrable function, which is


discontinuous at countably infinite points ?
Answer There are plenty of such functions!
Theorem
Every bounded monotone function f : [a, b] → R is integrable.
Proof.
Let f be increasing, P be a partition with `(Ii ) = b−am for all i.
Pm−1 Pm−1
• U(P, f ) = P i=0 Mi `(Ii ) =P i=0 f (xi+1 ) `(Ii ) and
L(P, f ) = m−1
i=0 mi `(Ii ) =
m−1
i=0 f (xi ) `(Ii )
f (b)−f (a) 1
U(P, f ) − L(P, f ) = (f (xm ) − f (x0 )) b−a
m = b−a m → 0.
Problem
Consider the Zeno’s staircase function Z : [0, 1] → R given by

1

2
 if 0 ≤ x < 12 ,
3
if 12 ≤ x < 34 ,



4

7
Z (x) = 8 if 34 ≤ x < 78 ,

 ..



 .
 2k −1 if 2k−1 −1 ≤ x < 2k −1 , k > 1.

2k 2k−1 2k
Then Z is integrable with countably infinite discontinuites.
Sets of measure 0
Question Does there exist a Riemann integrable function with
uncountably many discontinuities ?
Question Does there exist a Riemann integrable function with
discontinuities containing an open interval ?
Definition
Let E be a subset of R. We say that E is a set of measure 0 if for
every  > 0, there exists a countablePfamily of open intervals

{Ik }k>1 such that E ⊆ ∪∞ k=1 Ik and k=1 `(Ik ) < .
• In case E is compact, there exists a finite family of open intervals
{Ik }N
k=1 satisfying the conditions above.
• If A is of measure 0 and B ⊆ A, then B is of measure 0.
Problem
Show that the Cantor set is of measure 0.
Hint.
C = ∩n>1 Cn (see slide 28) and `(Cn ) → 0 as n → ∞.
Problem
Show that a countable union of sets of measure 0 is of measure 0.
Solution.
Let E1 , E2 , . . . , be countably many sets of measure 0. Let  > 0.
(1)
• Let {Ik }k>1 be a sequence of open intervals such that
(1) (1)
E1 ⊆ ∪∞ and ∞
P
k=1 Ik k=1 `(Ik ) < /2.
(2)
• Let {Ik }k>1 be a sequence of open intervals such that
(2) (2)
E2 ⊆ ∪∞ and ∞
P
k=1 Ik k=1 `(Ik ) < /4.
(j)
Continue like this to get for every j > 1, {Ik }k>1 such that
(j) P∞ (j) (j)
Ej ⊆ ∪∞
k=1 Ik and
j ∞ ∞
k=1 `(Ik ) < /2 . Thus ∪j=1 Ej ⊆ ∪j,k=1 Ik
(j)
(countable union) and ∞
P P∞ j
j,k=1 `(Ik ) < j=1 /2 = .

Problem
Show that any countable subset of R is of measure 0.
Proof.
Any finite set (and hence single-ton) is of measure 0.
Let f : [a, b] → R be bounded with M = supx∈[a,b] |f (x)| < ∞.
• A = {c ∈ [a, b] : osc(f , c) > } is compact (see slide 181)
Lemma
If A is a set of measure 0, then there exists a partition P of [a, b]
such that U(P, f ) − L(P, f ) < (2M + b − a).
Proof.
Since A is a compact set of measureP0, there exists open intervals
N
{Ik }N N
k=1 such that A ⊆ ∪k=1 Ik and k=1 `(Ik ) < .
• K = [a, b] \ (∪N
k=1 Ik ) is a compact subset of [a, b]
• For any c ∈ K , there exists an interval, open in Jc ⊆ [a, b] \ A
such that c ∈ Jc (⇒ supx,y ∈Jc |f (x) − f (y )| 6 )
0
• There exists J1 , . . . , JN 0 such that K ⊆ ∪N
j=1 Jj (K is compact)
If P : partition formed by end-points of I1 , . . . , IN , J1 , . . . , JN 0 , then
XN
U(P, f ) − L(P, f ) 6 2M `(Ij ) + (b − a) < (2M + b − a).
j=1
• A is a subset of D(f ) (the set of discontinuities of f )
• If D(f ) is of measure 0, then so is A for every  > 0
• If D(f ) is of measure 0, then f ∈ R[a, b] (apply Lemma)
Theorem (Lebesgue’s criterion for Riemann integrability)
If f : [a, b] → R is bounded, then f ∈ R[a, b] if and only if D(f ) is
of measure 0.
Proof.
To see (D(f ) is of measure 0 ⇐ f ∈ R[a, b]), let f ∈ R[a, b].
• Countable union of sets of measure zero is of measure 0 (see
slide 204) and D(f ) = ∪n>1 A1/n Claim Each A1/n is of measure 0
• Let  > 0 and choose a partition P = {x0 , . . . , xN } such that
U(P, f ) − L(P, f ) < /n (since f is Riemann integrable)
• Ij = (xj−1 , xj ) and Ij1 , . . . , IjN 0 be sets intersecting with A1/n
• sup f (x) − inf f (x) > sup f (x) − inf f (x) > 1/n
x∈Ijk x∈Ijk x∈Ijk ∩A1/n x∈Ijk ∩A1/n
P 0
Finally, n1 N  0
k=1 `(Ijk ) 6 U(P, f ) − L(P, f ) < n . Choose Ijk ⊇ Ijk so
P 0 N0 0
that N k=1 `(Ijk ) < 2 and A1/n ⊆ ∪k=1 Ijk .
Example
Consider the function g : [0, 1] → R given by
(
1
if x ∈ Q ∩ (0, 1) and x = qp in reduced form
g (x) = q
0 otherwise.
Then g ∈ R[0, 1] since D(g ) = Q ∩ (0, 1] is of measure 0.
Problem
Show by an example that composition of Riemann integrable
functions need not be Riemann integrable.
Hint.
Take f = χ(0,1] and g as above.
Problem
Let f : [a, b] → [c, d] be Riemann integrable. If φ : [c, d] → R is
continuous, then φ ◦ f is Riemann integrable.
Hint.
D(φ ◦ f ) ⊆ D(f ) is of measure 0.
Problem
Show that sum and product of Riemann integrable functions is
Riemann integrable.
Hint.
D(f + g ) ⊆ D(f ) ∪ D(g ) and D(fg ) ⊆ D(f ) ∪ D(g ).
Theorem
The set R[a, b] of Riemann integrable functions f : [a, b] → R is a
vector space over R. Morever, for every f , g ∈ R[a, b] and α ∈ R,
Rb Rb Rb
a (f (x) + αg (x))dx = a f (x)dx + α a g (x)dx.
Rb
For f ∈ R[a, b], define kf k1 = a |f (x)|dx.
Question Is k · k1 a norm on R[a, b] ?
• Clearly, kf k1 > 0, kαf k1 = |α|kf k1 and kf + g k1 6 kf k1 + kg k1
for every f , g ∈ R[a, b] and α ∈ R
• If kf k1 = 0 for f ∈ R[a, b], then f need not be 0
• χ[0,1/2] − χ[0,1/2) is Riemann integrable, kχ[0,1/2] − χ[0,1/2) k1 = 0
but χ[0,1/2] 6= χ[0,1/2)
Discontinuities of positive measure

Example
Let Ĉ denote the Cantor-like set obtained by removing 2k−1
centrally situated open subintervals I1k , · · · , I2k−1 k of I = [0, 1]
each of length 1/4k at the kth stage, where k = 1, 2, · · · .
2k−1
• `(Ĉ ) = 1 − ∞
P P∞ 1
k=1 4k = 1 − k=1 2k+1 = 1/2
• Let Fk : I → I be a continuous function such that Fk = 1 on
k−1
I \ ∪2i=1 Iik and Fk = 0 at the mid-points of I1k , · · · , I2k−1 k
• If fn = ni=1 Fi , then fn+1 (x) 6 fn (x) for every x ∈ I , n > 1
Q

• Let f : I → I be the pointwise limit of {fn }n>1


• For x ∈ Ĉ , there exists a sequence {xn }n>1 converging to x
such that f (xn ) = 0
• f is discontinuous on Ĉ (since f (x) = 1 for every x ∈ Ĉ )
Conclude that f is not Riemann integrable.
• It turns out that kfn − f k1 → 0 as n → ∞!
Problem
For f , g ∈ R[a, b], define f ∼ g if f = g outside a set of measure
0. Verify the following:
(1) ∼ defines an equivalence relation.
(2) If [f ] denotes the equivalence relation containing f and
Rb
k[f ]k1 = a |f (x)|dx, then k[f ]k1 = 0 if and only if f ∼ 0.
(3) R = {[f ] : f ∈ R[a, b]} is a normed linear space endowed with
the norm k · k1 .
(4) (R, k · k1 ) is incomplete.
Problem
Consider the intervals I1 = [0, 1], I2 = [0, 1/2], I3 = [1/2, 1], I4 =
[0, 1/4], I5 = [1/4, 1/2], I6 = [1/2, 3/4], I7 = [3/4, 1] and so on. For
fn = χIn and f = 0, show that fn (x) 9 f (x) for any x ∈ [0, 1].

Hint.
Any x ∈ I1 lies in infinitely many In & infinitely many [0, 1] \ In .
R
• [0,1] |fn (x) − f (x)|dx = `(In ) → 0 as n → ∞.
Problem (Integrable function with uncountable discontinuities)
Consider the indicator function χC of the Cantor set, that is,
(
1 if x ∈ C ,
χC (x) =
0 if x ∈ [0, 1] \ C .

Show that χC is discontinuous precisely at every x ∈ C . Conclude


that χC ∈ R[0, 1].

Hint.
Note that C is closed and nowhere dense, and hence [0, 1] \ C is
dense. Thus for every x ∈ C , there exists a sequence of points
xn ∈ [0, 1] \ C such that xn → x. Clearly, χC (xn ) = 0 does not
converge to χC (x) = 1, and hence C ⊆ D(χC ).
Let x ∈ [0, 1] \ C . Then, some neighborhood of x does not
intersect C (otherwise, x is a limit point of C ), and hence χC is
sequentially continuous at x. Thus D(χC ) = C . Finally, since C is
of measure 0, by Lebesgue’s criterion, χC ∈ R[0, 1].
Theorem
The uniform limit f of a sequence of Riemann
Rb integrable
R b functions
fn is Riemann integrable and limn→∞ a fn (x)dx = a f (x)dx.
Proof.
Let {fn }n>1 ⊆ R[a, b], f : [a, b] → R be such that kfn − f k∞ → 0.
• f is bounded (since kf k∞ 6 kfn − f k∞ + supn>1 kfn k∞ )
• Given  > 0, there exists N > 1 such that for every n > N and
for every x ∈ [a, b], fn (x) −  < f (x) < fn (x) + 
• For
R partitionsR P, Q of [a, b], L(fn − , P) 6 L(f , P) 6


f (x)dx 6 f (x)dx 6 U(Q, f ) 6 U(fn + , Q)
Now take supremum over LHS and infinimum over RHS to get
Z b Z Z – Z b
(fn (x) − )dx 6 f (x)dx 6 f (x)dx 6 (fn (x) + )dx.
a – a
R R
Thus 0 6 – f (x)dx − – f (x)dx 6 2, or f ∈ R[a, b] and
Rb Rb
a fn (x)dx − a f (x)dx 6 (b − a) for every n > N.
• A power series can be integrated termwise
Problem
Let f (x) = ∞ n
P
n=0 an x be a power series of radius of convergence
R > 0. Show that for any −R < a < b < R,
b ∞  b n+1 an+1 
Z X
f (x)dx = an − .
a n+1 n+1
n=0

Hint.
The partial sum of power series is continuous and it converges to
f (x) uniformly on [a, b]. Now apply last problem.
Rb
• One may use the last problem to compute a x1 dx for every
1 < a < b < 2. Indeed, x1 = ∞ n
P
n=0 (1 − x) converges uniformly on
|1 − x| 6 1 −  for every  > 0, and hence on [a, b]. Thus
Z b ∞
X  (1 − b)n+1 (1 − a)n+1 
1
dx = − = log b − log a.
a x n+1 n+1
n=0
Let f : [a, b] → R be Riemann integrable and let x0 ∈ [a, b] \ D(f ).
• Given  > 0, there exists δ > 0 such that |f (x) − f (x0 )| < 
whenever |x − x0 | < δ
• Let J be an interval of length `(J) < δ/2 that contains x0
1 1
R R
• f (x0 ) − `(J) J f (x)dx = `(J) J (f (x0 ) − f (x))dx
1 1 9
R R
• f (x0 ) − `(J) J f (x)dx 6 `(J) J |f (x0 ) − f (x)|dx <  (see )
This proves the following:
Theorem (Fundamental Theorem of Calculus-I)
For every f ∈ R[a, b] and every x0 ∈ [a, b] \ D(f ),
Z
1
lim f (x)dx = f (x0 ).
`(J)→0 `(J) J
x0 ∈J

1
R
Remark The quantity `(J) J f (x)dx is the “average of f over J”.
9
Rb Rb
We used | a
g (x)dx| 6 a
|g (x)|dx, g ∈ R[a, b] (Exercise)
Corollary
R x0
Let f ∈ R[a, b]. Define F : [a, b] → R by F (x0 ) = a f (x)dx,
x0 ∈ [a, b]. Then
(1) F is continuous at every point in [a, b],
(2) F is differentiable at every x0 ∈ [a, b]\D(f ) & F 0 (x0 ) = f (x0 ).

Proof.
The first part follows from the estimate
Z x0  
|F (x)−F (x0 )| = f (t)dt 6 sup |f (t)| |x−x0 |, x ∈ [a, b].
x t∈[a,b]

To see the second part, let J = [x0 , x0 + h] in FTC-I to obtain


x0 +h
F (x0 + h) − F (x0 )
Z
1
= f (x)dx → f (x0 ) as h → 0.
h h x0

This also shows that F 0 (x0 ) = f (x0 ).


Corollary
Rx
Let f ∈ R[a, b]. Define F : [a, b] → R by F (x0 ) = a 0 f (x)dx,
x0 ∈ [a, b]. If D(f ) denotes the set of points of discontinuities of f
and Diff(F ) denotes the set of points at which F is differentiable,
then [a, b]\D(f ) ⊆ Diff(F ).

Problem
Give an example of a continuous function on [0, 1], which is
differentiable at irrationals in [0, 1] (we are not asking for a
function differentiable precisely at irrationals in [0, 1]).

Hint.
Use the last corollary.

• There exist functions differentiable precisely at rationals 10

10
A Continuous Function That Is Differentiable Only at the Rationals, Mark
Lynch, Mathematics Magazine , Vol. 86, No. 2 (April 2013), pp. 132-135
Definition
Given a partition P = {a = x0 < x1 · · · < xn = b} of [a, b] and
g : [a, b] → R, the variation of g over P is given by
n
X
V (g , P) = |g (xj ) − g (xj−1 )|.
j=1

• g is of bounded variation on [a, b] if the total variation


Vab g = supP V (g , P) of g over [a, b] is finite.

Theorem R x0
Let f ∈ R[a, b]. Define F : [a, b] → R by F (x0 ) = a f (x)dx,
x0 ∈ [a, b]. Then F is of bounded variation.

Proof. Pn Pn R xj
Since V (F , P) = j=1 |F (xj ) − F (xj−1 )| 6 j=1 xj−1 |f (x)|dx,
Vab F is at most (b − a) supx∈[a,b] |f (x)|.
Theorem (Fundamental Theorem of Calculus-II)
Let f ∈ R[a, b]. If there is a continuous function F : [a, b] → R
differentiable on (a, b) such that F 0 (x) = f (x), x ∈ (a, b), then
Rb
a f (x)dx = F (b) − F (a).

Proof.
Let  > 0 and P = {a = x0 < x1 · · · < xn = b} be a partition of
[a, b] such that U(P, f ) − L(P, f ) < .
• For i = 1, . . . , n, there exists ci ∈ [xi , xi−1 ] such that
F (xi ) − F (xi−1 ) = f (ci )(xi − xi−1 ) (Mean Value Theorem)
Pn Pn
• i=1 f (ci )(xi −xi−1 ) = i=1 (F (xi )−F (xi−1 )) = F (b)−F (a)
Pn
• L(P, f ) 6 i=1 f (ci )(xi − xi−1 ) 6 U(P, f ) (mi 6 f (ci ) 6 Mi )
Rb
• L(P, f ) 6 a f (x)dx 6 U(P, f )
Rb
Thus a f (x)dx and F (b) − F (a) lie in interval [L(U, f ), U(P, f )]
Rb
of length less than . Hence |F (b) − F (a) − a f (x)dx| < .
The Fundamental Theorem of Calculus-II can be used to compute
integrals
R x nprovided derivatives are known.
x n+1 n+1
• 0 t dt = n+1 (since the derivative of F (x) = xn+1 equals x n )
Rx 1
• 1 t dt = log x (since (log x)0 = 1/x)
Definition
We say that a function G : [a, b] → R is an anti-derivative of
g ∈ R[a, b] if G is differentiable on [a, b] and G 0 (x) = g (x) at
every x ∈ [a, b].
Example
The jump function g = χ(0,1] does not have an anti-derivative G .
Indeed, if there exists a differentiable function G such that G 0 = g ,
then by the Fundamental Theorem of Calculus-II,
Z 1
1−x = g (x)dx = G (1) − G (x), 0 < x < 1.
x
This shows that G (x) = G (1) − 1 + x for every x ∈ (0, 1], and
hence by continuity, G (x) = G (1) − 1 + x for every x ∈ [0, 1].
However, G 0 (0) = 1 6= g (0) ⇒⇐
Corollary (Integration by parts)
Suppose that f , g : [a, b] → R are continuous on [a, b] and
differentiable in [a, b], and f 0 , g 0 are integrable on [a, b]. Then
Z b Z b
0
f (x)g (x)dx = f (b)g (b) − f (a)g (a) − f 0 (x)g (x)dx.
a a

Proof.
Define F (x) = f (x)g (x), x ∈ [0, 1]. Then F is continuous on [a, b],
F is differentiable and
F 0 (x) = f 0 (x)g (x) + f (x)g 0 (x), x ∈ (a, b).
By the Fundamental Theorem of Calculus-II,
Z b  
f 0 (x)g (x)+f (x)g 0 (x) dx = F (b)−F (a) = f (b)g (b)−f (a)g (a).
a

This completes the proof.


• There exists f ∈ R[a, b] such that F 0 = f outside a set of
Rb
measure 0, yet, a f (x)dx 6= F (b) − F (a).
Example (Cantor Function)
Let C denote the Cantor set obtained by removing 2n−1 centrally
situated disjoint open subintervals U1,n , · · · , U2n−1 ,n of [0, 1] each
of length 1/3n at the nth stage, where n = 1, 2, · · · . Thus
n−1
C = ∩n>1 Cn , where Cn = [0, 1] \ ∪nk=1 ∪2j=1 Uk,j .


• F1 : [0, 1] → R a continuous increasing function so that


F1 (0) = 0, F1 = 1/2 on [1/3, 2/3], F1 (1) = 1, F1 linear on C1
• F2 : [0, 1] → R a continuous increasing function so that
F2 (0) = 0, F2 = 1/4 on [1/9, 2/9], F2 = 1/2 on [1/3, 2/3],
F2 = 3/4 on [7/9, 8/9], F1 (1) = 1, F2 linear on C2
Continuing this, we obtain a sequence of continuous increasing
functions {Fn }n>1 such that |Fn+1 (x) − Fn (x)| 6 2−n−1 .
Example (Example continued ...)
Check that {Fn }n>1 is Cauchy in C [0, 1]. Indeed, for m > n,
m
X m
X
|Fm (x) − Fn (x)| 6 |Fj (x) − Fj−1 (x)| 6 2−j ,
j=n+1 j=n+1

and hence {Fn }n>1 converges uniformly to F ∈ C [0, 1].


• We refer to F as Cantor function or devil’s staircase function.
• F is increasing (since so is Fn for every n > 1)
• F 0 = 0 on [0, 1] \ C (since F is constant on each interval of
the complement of C )
• F 0 = 0 outside a set of measure 0 (since C is of measure 0)
R1
This shows that 0 F 0 (x)dx = 0 (why?) and F (1) − F (0) = 1, so
Z 1
F 0 (x)dx 6= F (1) − F (0).
0
• F is non-constant, yet, F 0 = 0 outside a set of measure 0!
Let us summarize the discussion above:
Rx
• If f ∈ R[a, b], then F (x) = a f (x)dx is differentiable outside
the set D(f ) of measure 0.
Rb
• In general, a F 0 (x)dx 6= F (b) − F (a).
This raises the following question:
Question What conditions on a function F on [a, b] guarantee that
F 0 (x) exists (outside a set of measure 0), that this function is
Rb
integrable, and that moreover a F 0 (x)dx = F (b) − F (a) ?
This problem is closely related to the ”averaging problem”:
Question What are conditions on a function f on [a, b] for which
Z
1
lim f (x)dx = f (x0 )
`(J)→0 `(J) J
x0 ∈J

holds for x0 ∈ [a, b] outside a set of measure 0 ?


To answer these questions, one needs to venture into the theory of
Lebesgue integration!

You might also like