A Friendly Introduction To Analysis
A Friendly Introduction To Analysis
A Friendly Introduction To Analysis
Maung Min-Oo
April 3, 2004
2
Preface
Analysis is the rigorous and more advanced study of the techniques and re-
sults used in calculus. As is well-known, although calculus was “invented”
by Newton and Leibniz in the 17th century a more precise and rigorous foun-
dation was laid only about 200 years later by Cauchy, Weierstrass Bolzano
Dedekind, Cantor and others. The main difficulty was to give a precise def-
inition of the set of real numbers and understand the notions of infinities,
infinitesimals and limits. Of course, such attempts go back to ancient times
as can be seen in the paradoxes of Zeno of Elea, 5th century B.C.
The main difficulty of modern-day undergraduate students in under-
standing rigorous analysis is the lack of basic training in the nature and
structure of logical arguments and mathematical proofs. This is partly due
to the fact that disciplines like classical Euclidean Geometry, which gener-
ations of human beings have studied, are no longer taught in schools. It is
claimed that after the Bible, Euclid’s Elements is the most widely read book
of western civilization. In any case, proofs not only “scare” many students,
but it is also sometimes hard for students to appreciate why mathemati-
cians go through all that trouble to “prove” what seems like obvious facts.
I do admit that the notation and style of most analysis text books is quite
unappealing, especially to students whose only encounter with mathemat-
ics is a “standardized calculus text-book” where proofs are not rigorous,
definitions are not clearly stated and the emphasis is more on a rather su-
perficial (cook-book-style) mastery of certain techniques. It is therefore not
the fault of the students that they are never exposed to some of the finer
points and perhaps the true nature of mathematical reasoning in the first
year at a University. Education can then degenerate into blindly following
a set of technical instructions (given by instructors (sic)) rather than an
awe-inspiring and mind-opening experience that we all strive for in life. I
admit I’m also not so sure that the prescribed material that I will be teach-
ing you in this course is really that mind-boggling either! ( On a personal
note, I should add that in the early 70’s in Germany, where I was a student,
i
there were no credits for courses and also no tuition fees. I never had to
take a written test or examination. All examinations I took were oral and
one-on-one with my professors. It was challenging and a lot of fun!)
These notes are written in an informal and personal style of a “friendly”
mathematician, whose motto is: “It’s better to be approximately right than
exactly wrong”. The notes are by no means a substitute for my live lectures
(which by the way are in general much more entertaining!). They are aimed
to give a somewhat more organized and abbreviated outline of the material
to be discussed in the lectures. The notes (and also the lectures) are not
supposed to be comprehensive. The art of teaching (and learning) consists
in choosing carefully what is relevant and beautiful. My aim is to let you
discover (McMaster motto!) rather than cover the maximum amount of
material. I therefore strongly advise every student to read at least one
other recommended book in Real Analysis. Some are available on reserve in
the Thode library. I apologize for not including too many exercises in the
notes, but I will be making up on that by giving you more exercises and
problems during the term. I also plan to post extra supplementary material
and interesting links related to the course web site. My last (in my opinion,
unforgivable) mistake is the omission of any pictures in these notes. I am a
geometer and I think visually, but I couldn’t bring myself to spend the time
to make the appropriate ps files etc. I will certainly draw some pictures on
the blackboard during the lectures and hopefully, if I am allowed to teach
this course again and if these notes ever become publishable (you will then
have to pay for them!), I will include some pictures, although, Analysis is
still not Geometry!
Yours truly,
Min-Oo
ii
Contents
Preface i
0 Preliminaries 1
0.1 Some elementary propositional logic . . . . . . . . . . . . . . 1
0.2 Some naive set theory . . . . . . . . . . . . . . . . . . . . . . 2
0.3 Natural Numbers and the Principle of Induction . . . . . . . 4
0.4 Rational Numbers . . . . . . . . . . . . . . . . . . . . . . . . 4
1 Real Numbers 7
1.1 Field Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Axioms of Order . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 The Completeness Axiom . . . . . . . . . . . . . . . . . . . . 13
1.4 Remarks on orders of Infinity . . . . . . . . . . . . . . . . . . 16
1.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5.1 Short solutions to some of the exercise problems . . . 19
iii
3 Topology of the real line R 45
3.1 Open sets and closed sets . . . . . . . . . . . . . . . . . . . . 45
3.2 Connected sets . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.3 Compact sets . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4 Elementary topology of Rn . . . . . . . . . . . . . . . . . . . 48
3.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5.1 Solutions to some exercise problems . . . . . . . . . . 50
4 Continuity 51
4.1 Continuous functions . . . . . . . . . . . . . . . . . . . . . . . 51
4.2 Continuity and Connectedness . . . . . . . . . . . . . . . . . 53
4.3 Continuity and Compactness . . . . . . . . . . . . . . . . . . 55
4.3.1 The Contraction Mapping Principle . . . . . . . . . . 57
4.4 Uniform Continuity . . . . . . . . . . . . . . . . . . . . . . . . 57
4.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.5.1 Short solutions to some of the exercise problems . . . 60
5 Differentiation 63
5.1 Definition and basic properties . . . . . . . . . . . . . . . . . 63
5.2 Local properties of the derivative . . . . . . . . . . . . . . . . 66
5.3 Some global properties of the derivative . . . . . . . . . . . . 67
5.3.1 The Logarithm . . . . . . . . . . . . . . . . . . . . . . 70
5.4 Differentiation in Rn . . . . . . . . . . . . . . . . . . . . . . . 70
5.4.1 Inverse Function Theorem . . . . . . . . . . . . . . . . 71
5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.5.1 Hints and short solutions . . . . . . . . . . . . . . . . 75
6 Integration 77
6.1 Definition of the Riemann Integral . . . . . . . . . . . . . . . 77
6.2 Basic properties of the Integral . . . . . . . . . . . . . . . . . 79
6.3 Fundamental Theorem of Calculus . . . . . . . . . . . . . . . 80
6.4 Improper Integrals . . . . . . . . . . . . . . . . . . . . . . . . 81
6.4.1 The Gamma Function . . . . . . . . . . . . . . . . . . 82
6.4.2 Stirling’s Formula . . . . . . . . . . . . . . . . . . . . 83
6.4.3 The Euler-Maclaurin Summation Formula . . . . . . . 85
6.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.5.1 Hints and short solutions . . . . . . . . . . . . . . . . 90
iv
7 Metric spaces and Uniform Convergence 93
7.1 Definitions and examples . . . . . . . . . . . . . . . . . . . . . 93
7.2 Uniform Convergence . . . . . . . . . . . . . . . . . . . . . . 95
7.2.1 Uniform convergence of power series . . . . . . . . . . 98
7.3 Weierstrass approximation theorem . . . . . . . . . . . . . . . 98
7.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
7.4.1 Hints and short solutions to the exercises . . . . . . . 102
v
vi
Chapter 0
Preliminaries
∀m ∈ Z ∃n ∈ Z s.t. n > m
1
which we can read as: For every integer m there is some integer n such that
n > m . This is obviously true. However, the statement:
∃n ∈ Z ∀m ∈ Z s.t. n > m
which we can read as: There is some integer n such that for every integer
m we have n > m is definitely false. So the negation:
∀n ∈ Z ∃m ∈ Z s.t. n ≤ m
All these sets lie inside the set of real numbers R which is what this course
is mainly about. Here are some more standard notations that we will use
for sets:
2
If A, B are two sets, we say that A is a subset of B and write A ⊂ B if
every element of A is also an element of B. For example, A ⊂ A and ∅ ⊂ A
for any set A , where ∅ denotes the empty set which does not contain any
elements.
A ∪ B = {x | (x ∈ A) ∨ (x ∈ B)} and A ∩ B = {x | (x ∈ A) ∧ (x ∈ B)} will
denote the union and intersection of two sets.
A × B denotes the Cartesian product which is the set of all ordered pairs:
{(a, b)| a ∈ A, b ∈ B} .
{(a, f (a)) ∈ A × B a ∈ A}
of the Cartesian product. (In fact, a more rigorous way to define a function
is through its graph).
f is said to be one-one (or injective) if two distinct elements of A never
map to the same element of B. In other words: f : A → B is one-one iff
(∀ x, y ∈ A)(f (x) = f (y) ⇒ x = y). f is said to be onto (or surjective) if
every element of B has some element of A mapped to it. In other words;
f : A → B is onto iff ∀ b ∈ B ∃ a ∈ A s.t. f (a) = b. A map which is both
one-one and onto is called a bijection (or a one-one correspondence).
If f : A → B is a map, the image of a set X ⊂ A is defined by f (X) =
{b ∈ B | ∃ x ∈ X s.t. b = f (x)} and the inverse image of a set Y ⊂ B is
defined by f −1 (Y ) = {a ∈ A | f (a) ∈ Y }.
3
0.3 Natural Numbers and the Principle of Induc-
tion
Natural numbers are used for counting and ordering. The main property
(or axiom, if you wish) about N that we will use is the Induction Principle
N1 P (n0 ) is true
N2 (induction step) ∀ k ≥ n0 , we have P (k) ⇒ P (k + 1)
We also use the inductive property of the natural numbers for recursive
definitions such as in the following definition of the factorial.
0! = 1
∀ k ≥ 0 (k + 1)! = (k + 1)k!
4
√
Proposition 0.4.1 The real number 2 is not a rational number.
√
Proof (by contradiction) Suppose that 2 = a/b with a ∈ Z and b ∈ N.
We may as well assume that a, b have no common factors else we could
cancel them out. Then 2b2 = a2 and so a is even. But then a2 and hence
2b2 is divisible by 4 and so b2 is even. But then b is also even and so a and b
do
√ have a common factor, viz. 2 . Thus we arrive at a contradiction. Hence
2 is not in Q.
QED
The Pythagoreans
√ realized that one could easily construct a line seg-
mentof length 2 by elementary geometry (for example, the diagonal of a
unit square) and so they were forced to the conclusion that rational numbers
were not sufficient to describe their geometric system. Mathematics need
real numbers to describe reality (or at least the shadow of it!).
5
6
Chapter 1
Real Numbers
One normally omits the symbol · for multiplication and simply juxtapose.
We also write x − y for x + (−y) . Our main algebraic axiom about R is
therefore:
7
Axiom 1.1.1 The real numbers form a field under the usual rules of addi-
tion and multiplication.
1. Q is a field, but N and Z do not satisfy all the field axioms. (Can you see
which of the axioms fail for N and for Z ?)
2. However integers modulo a prime number p denoted by Z/pZ form a
field. (The main axiom to check is the last one about the existence of the
multiplicative invers. can you see that this would fail if p is not a prime, say
p = 12)
3. The complex numbers C form a field. (Do you know any other fields?)
The field axioms trivially imply some algebraic operational rules that we
are all familiar with from elementary school. A few selected examples of
such rules are the following:
(i) The commutative and the associative laws can be extended by induction
to any finite number of elements. The same is true for the distributive law.
(ii) The “neutral” elements 0 and 1 are unique and the same is true for both
“inverses” −x and x−1 .
(iii) 0 x = 0 for all x.
Proof: 0x = (0 + 0)x = 0x + 0x . Now add the inverse −(0x) to both sides
of the equation and use associativity of multiplication: 0 = −(0x) + 0x =
−(0x) + (0x + 0x) = (−(0x) + 0x) + 0x = 0 + 0x = 0x
This seems like splitting hairs but that is what we are learning here!
(iv) Simple rules such as −(−a) = a ; (a−1 )−1 = a for a 6= 0 ; (−a) b =
a.(−b) = −ab ; (−a)(−b) = ab etc.
Sample proof: 0 = 0a = a0 = a(b + (−b)) = ab + a(−b) and so −(ab) =
0 + (−(ab)) = −(ab) + ((ab) + (a(−b)) = 0 + a(−b) = a(−b).
(v) One important fact is the cancellation rule (or in fancier words, the
absence of zero divisors in a field!): ab = 0 if and only if a = 0 ∨ b = 0.
This implies: ac = ab ∧ a 6= 0 ⇒ c = b . (This is, for example, not true in
Z/12Z) .
Proof: If ab = 0 and suppose a 6= 0. Then ∃ a−1 so that a−1 a = 1 , but
then 0 = a−1 0 = a−1 (ab) = (a−1 a)b = 1b = b . QED
(vi) Powers are defined inductively by: x0 = 1 , xn+1 = x.xn for n ∈ N. We
also define x−n = (x−1 )n = (xn )−1
8
A more important result is the binomial formula:
Proposition 1.1.1
n
n
X n
(x + y) = xn−k y k
k
k=0
Then
k
X k
(x + y)k+1 = (x + y) xk−j y j
j
j=0
k
X k
xk−j+1 y j + xk−j y j+1
=
j
j=0
k k
X k X k
= xk−j+1 y j + xk−j+1 y j
j j−1
j=0 j=1
k+1
X k + 1 k+1−j j
= x y
j
j=0
k k k+1
where we use the Pascal Triangle relation: j + j−1 = j .
This relation follows by a simple calculation with factorials from the defini-
n
tion of the binomial coefficients, but if we use the fact that k represents the
number of different ways of choosing k objects from n objects (that’s why
we read “n choose k” for nk ), it can be seen immediately from the following
combinatorial argument: If we have to choose a committee of j members
from k students (that’s you) and 1 professor (that’s me), then there are only
two mutually exclusive possibilities: either I’m on the committee or I’m not.
9
1.2 Axioms of Order
A field with a relation called an ordering satisfying the three properties above
is called an ordered field. The field Q of rationals is an ordered field (with
the usual ordering) but the field C of complex numbers is not an ordered
field (under any ordering). Our next basic axiom about R is therefore:
We will write a > b or b < a for a−b > 0. It follows from the axioms that
the ordering > is transitive, i.e. , if a > b and b > c, then a > c (because
a − c = (a − b) + (b − c)) We also write a ≥ b to mean the same thing as
a > b ∨ a = b. ( a ≤ b is equivalent to −a ≥ −b because b − a = −a − (−b)).
The axioms trivially imply some simple and familiar properties such as:
(i) 1 > 0 .
(ii) x > 0, y < 0 ⇒ xy < 0 ; x < 0, y < 0 ⇒ xy > 0 .
(iii) If a > b ⇒ a + c > b + c. If c > 0 and a > b, then ac > bc (because
ac − bc = (a − b)c, but if c < 0, then a > b ⇒ bc > ac .
(iv) If 0 < a < b, then a−1 > b−1 > 0, but if a < b < 0, then 0 > a−1 > b−1 .
(v) The square x2 of a real number x 6= 0 is always strictly positive: x2 > 0
(because (−x)(−x) = x2 ). Of course 02 = 0.
The following notation will be used for intervals on the real line R: An open
interval: (a, b) = {x ∈ R| a < x < b} does not contain its end points. A
closed interval: [a, b] = {x ∈ R| a ≤ x ≤ b} contains both end points. We
will also use some other types of intervals: [a, b), (a, b], (−∞, b], (a, ∞) , etc.
Their meaning should be clear from the notation, for example: [a, b) = {x ∈
R| a ≤ x < b} (−∞, b] = {x ∈ R| x ≤ b} .
We define the absolute value of a real number to be:
10
Then we have −|x| ≤ x ≤ |x|. (in fact, it is true that either |x| = x or x = |x|.
It can never happen that −|x| < x < |x| . If we add the two inequalities for
x and y , we obtain: x + y ≤ |x| + |y| and also −x − y ≤ |x| + |y| . This
proves the following basic:
|x + y| ≤ |x| + |y|
for all x, y ∈ R.
Of course, it can happen now that the strict inequality holds, e.g., |2 +
(−1)| = 1 is strictly less than |2| + | − 1| = 2 + 1 = 3 .
Other very useful inequalities are the following:
(1 + x)n ≥ 1 + nx
11
n
X n
X n
X
(xi + yi )2 ≥ 0 ⇒ 2 + 2 xi yi ≥ 0 ⇒ − xi yi ≤ 1
i=1 i=1 i=1
Therefore
n n n
X X X
x2i yi2
xi yi ≤ 1 =
i=1 i=1 i=1
Moreover, strict inequality holds unless all the numbers are equal.
Therefore:
n+1 n+1
AMn+1 1 xn+1 xn+1 xn+1
AMn = 1+ n+1 ( AMn − 1) ≥ 1 + ( AM n
− 1) = AMn
12
1.3 The Completeness Axiom
In other words, M is the upper bound of A such that any other upper bound
of A (different from M ) is strictly larger than M . If A is not bounded from
above we often write sup(A) = +∞
In other words, m is the lower bound of A such that any other lower bound
of A (different from m) is strictly smaller than M . If A is not bounded from
below we often write inf (A) = −∞
13
An ordered field that satisfies the least upper bound property is called a
complete ordered field. So to sum up, all we assume about the real numbers R
is that it is a complete ordered (Archimedean) field. (In fact, one can show
that up to “ isomorphism of ordered fields”, that R is the only complete
(Archimedean) ordered field).
Note that the ordered field Q is not complete. For example, the set
A = {q ∈ Q| q 2 < 2} is bounded but does not have a least upper bound in
Q. We will see why in a little while. We first list some easy consequences of
the completeness axiom. First of all by just changing signs (and flipping the
inequalities) it is obvious that any subset A ⊂ R which has a lower bound
has a greatest lower bound. If B = {x ∈ R| − x ∈ A}. Then B is bounded
from above iff A is bounded from below and glb(A) = −lub(B) .
Before we state the next property we need to describe more precisely
how the N, Z and Q are imbedded into the reals. First of all N is identified
with multiples of the unit element 1 in R, i.e. n = 1 + · · · + 1 where the
sum is over n terms. The key fact here is that this is an injective (one to
one) map. (This is for example, not true for finite fields). After that Z and
Q can be imbedded in the obvious fashion, so p/q ∈ Q is identified with
p · q −1 since R is a field. An important property about the real numbers
which follows from our axioms is the following:
Proof: This is equivalent to saying that N is not bounded above. This seems
like a very obvious fact, but we will prove it from the axioms. Suppose N
were bounded above. Then it would have a least upper bound,M say. But
then M − 1 is not an upper bound and so there is an integer n ∈ N with
n > M − 1. But then n + 1 > M contradicting the fact that M is an upper
bound for all of N. QED
The above property of the real numbers is called the Archimedean
property of the Reals and is been attributed to the famous Greek math-
ematician Archimedes (287 to 212 BC) and appears in Book V of The Ele-
ments of Euclid.
We also deduce the following important fact:
14
Proof: Let a, b ∈ R with (say) a < b. Choose n ∈ N so that 1/n < b − a.
Then look at integer multiples of 1/n . Since these are unbounded, we may
choose the first such multiple with m/n > a. We claim that m/n < b. If
not, then since (m − 1)/n < a and m/n > b we would have 1/n > b − a.
QED
A set A with the property that an element of A lies in every interval
(a, b) of R is called dense in R. We have just proved that the rationals Q
are dense in R. The irrationals (= R ∩ Qc ) are also dense in R.
We now prove the result we stated earlier.
√
Proposition 1.3.3 The real number 2 exists.
√
Proof : We will get the existence of 2 as the least upper bound of the set
A = {q ∈ Q| q 2 < 2}. We know that A is bounded above. 2 is an upper
bound for A . (Why?) Let b = lub(A). We now prove that b2 < 2 and b2 > 2
both lead to contradictions and so we must have b2 = 2 (by the trichotomy
axiom for the ordering).
So first suppose that b2 > 2. By the Archimedean property we can
choose an n ∈ N such that n > b22b −2
so that b2 − 2b 1 2
n > 2. Then (b − n ) =
b2 − 2b 1 2 1
n + ( n ) > 2, since a square is always non-negative. Thus b − n is an
upper bound of A, contradicting the assumption that b is the least upper
bound.
2
Similarly, if b2 < 2, then a = b + 2−b 2b+2
b+2 = b+2 is > b and is in A since
2 2
a2 = 4b +8b+4
b2 +4b+4
= 2 − 2(2−b
(b+2)2
)
< 2. This contradicts the fact that b is an upper
bound of A.
QED
15
2 (and also base 16). The terniary system (base 3) is useful in describing the
Cantor set, which we will encounter in Chapter 3. The Cantor set consists
of all points in [0, 1] whose terniary decimal expansion does not contain 2.
For example 0.12 = 5/9 in base 3, so is not in the Cantor set.
16
Proposition 1.4.1 The set of algebraic numbers is countable. Hence there
are uncountably many transcendental numbers.
(i) x ∈ C iff the ternary (base 3) decimal expansion of x does not contain
1.
(ii) C is uncountable. (This follows from (i), so please prove it!)
(iii) C is closed and bounded and hence compact (see Chapter 3 for defini-
tions).
17
1.5 Exercises
Pn
1. Find a formula for k=1 (2k − 1) and prove your formula.
4. Show that 0 < a < b ⇒ a−1 > b−1 > 0 and a < b < 0 ⇒ 0 > a−1 > b−1
for all a, b ∈ F , where F is any ordered field.
6. Show that between any two distinct real numbers there is at least one
(and hence infinitely many) irrational numbers.
7. Show that the number of subsets with exactly k elements of a set consist-
ing of n elements (k ≤ n) is nk . What are the odds of winning the lottery
“6 out of 49” ?
8. Show that
n
X 1 − xn+1
xk =
1−x
k=0
for any x ∈ R and for any n ∈ N .
(i) nk=0 nk = 2n
P
18
13. Show from first principles that ∀n ∈ N and ∀ y > 0, ∃ a unique x > 0
such that xn = y .
19
n
= n(n − 1)(n − 2) · · · (n − k + 1) ≤ nk .
9. k! k
10.
(i) Just apply the Binomial formula to (1 + 1)n .
(ii) (x + y)n (x + y)m = (x + y)n+m . Now apply the binomial formula to
both sides:
n m n+m
X n + m
X n n−k k X m m−k k
x y x y = xn+m−l y l
k j l
k=0 j=0 l=0
and collect terms on the left hand side and look at the coefficient of xn+m−l y l .
1 n
1 n(n − 1) 1 n(n − 1) . . . 1 1
1+ = 1+n + 2
+ ··· +
n n 2! n n! nn
1 1
< 1 + 1 + + ··· +
2! n!
1 1
< 1 + 1 + + ··· + n < 1 + 2 = 3
2 2
where we used the formula for the sum of a geometric series:
1
1 1 1 1 − 2n+1
1 + + + ··· + n = <2
2 4 2 1 − 12
20
Chapter 2
2.1 Sequences
Definition 2.1.1 A sequence of real numbers is a map a : N → R.
Examples:
21
5. One may define a sequence inductively by a recursive formula such as:
an+2 = an+1 + an with initial terms given by a1 = a2 = 1. This gives the Fi-
bonacci sequence (1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, 377, 610, 987, 1597, . . . ) ,
first introduced by Fibonacci of Pisa (1170 to 1250).
6. Define an+1 = 12 (an + a2n ) with initial term a1 = 1 . This gives (1, 32 , 17 577
12 , 408 , ...).
This sequence can be written√as a “continued fraction” and gives a very good
rational approximation for 2. It was known (at least the first few terms)
to the ancient Sumerians.
2.2 Convergence
Informally, we say that a sequence of real numbers converges to a limiting
number a if all the terms of the sequence become arbitrarily close to a for
all sufficiently large n . The exact definition now follows:
Definition 2.2.1 We say that a sequence of real numbers (an )n∈N converges
to to a limit a ∈ R (and write: limn→∞ an = a or simply lim an = a) if and
only if the following is true:
∀ > 0, ∃ N ∈ N such that |an − a| < for all n ≥ N .
Examples:
22
1. The sequence an = n1 converges to 0 .
This is because ∀ > 0, | n1 − 0| = n1 < for all n ≥ N > 1 and the existence
of such an N ∈ N is guaranteed by the Archimedean property.
1 bis. More generally (as you know from first year calculus), the same
P (n)
argument proves that the sequence an = Q(n) converges to 0 , where P (n)
and Q(n) are polynomials provided the degree of P is strictly less than the
degree of Q . In case they have the same degree the sequence converges to the
ratio of the coefficients of the terms of highest degree and if deg(P ) > deg(Q)
then the sequence diverges.
Here is a very basic limit, which we will use repeatedly. We prove it from
first principles (i.e., we don’t use logarithms and other stuff that we haven’t
defined yet).
23
The following basic arithmetic rules about computing the limits of sums,
differences, products and quotients of convergent sequences are rather obvi-
ous from the definitions and the trivial proofs will be left to the reader as
an exercise (please do it!), but I will give ample hints below.
Proof (Sketch):
(i): Let > 0 be given. Then ∃ n1 , n2 ∈ N such that |an − a| < 31 for all
n ≥ n1 and |bn − b| < 13 for all n ≥ n2 , by the definition of convergence.
Now choose any n0 ∈ N that is greater than both n1 and n2 (for example
n0 = 10(n1 + n2 ) would do), then n ≥ n0 ⇒ |(an + bn ) − (a + b)| =
|(an − a) + (bn − b)| ≤ |an − a| + |bn − b| < 31 + 31 < by the triangle
inequality.
(ii) First we use the fact that any convergent sequence is bounded, so that
for example: |an | ≤ 1+|a| for all sufficiently large n . The proof then follows
the same pattern as above using the inequality:
|an bn − ab| = |an (bn − b) + (an − a)|b| ≤ (1 + |a|)|bn − b| + |an − a|(1 + |b|)
1 1
(iii) We first prove that bn converges to b and then apply (ii).
Since |b| > 0, ∃ n1 ∈ N such that |bn − b| < |b|
2 for all n ≥ n1 . This implies
|b|
(by the triangle inequality) that |bn | > 2 for all n ≥ n1 .
2
Now let > 0 be given. Then ∃ n2 ∈ N such that |bn − b| < |b|2 for all
n ≥ n2 . Now choose any n0 that is greater than both n1 and n2 . Then
1 1 b − bn 2 1 |b|2
n ≥ n0 ⇒ | − |=| |≤ =
bn b bn b |b| |b| 2
QED
We will sometimes use the following notation:
Definition 2.2.4
limn→∞ an = +∞ iff ∀R ∃N ∈ N s.t. n ≥ N ⇒ an ≥ R .
We write limn→∞ an = −∞ if limn→∞ (−an ) = +∞
24
For example, limn→∞ bn = +∞ for all b > 1 , but limn→∞ bn 6= −∞ for
b < −1 .
25
Proof: Let (an ) be a monotonically increasing sequence which is bounded
from above. Let a = lub(A), where A = {a1 , a2 , ..., an , ...}. We will prove
that the sequence converges to a, its least upper bound. So let a be the
least upper bound of the sequence. Given > 0, we’ll show that all except
possibly a finite number of the terms of the sequence are in the interval
(a − , a] ⊂ (a − , a + ). Since a is an upper bound, we have an ≤ a, but
since a is the least upper bound, a − is not an upper bound of the sequence
and therefore there must be a term an0 which is strictly larger than a − .
Now since the sequence is montonically increasing an ≥ an0 for all n ≥ n0 .
Thus we have shown that ∀ > 0, ∃ n0 ∈ N such that a − < an ≤ a < for
all n ≥ n0 .
The statement about a monotonically decreasing sequences follow from
the above proof since (an ) is monotonically decreasing sequence and bounded
from below if and only if (−an ) is monotonically increasing and bounded
from above.
QED
Of course, there aren convergent sequences which are not monotonic. For
example, an = 1 + (−1) n is not monotonic but converges to 1. Also there are
bounded sequences which are not convergent. For example, the sequence
(−1, 1, −1, 1, ...) defined by an = (−1)n is certainly not convergent even
though it is bounded. We have seen some bounded sequences which do not
converge. We can, however, say something about such sequences.
Definition 2.3.3 A subsequence is an infinite ordered subset of a sequence.
Proposition 2.3.1 Any subsequence of a convergent sequence is convergent
(to the same limit).
The proof of the above proposition is trivial. The next result however is
not totally obvious!
Proposition 2.3.2 Every sequence (convergent or not) contains a mono-
tonic subsequence.
Proof: Let M = {m ∈ N | an < am ∀ n > m } . There are two possibilities:
Case 1. M is infinite. M = {m1 < m2 < . . . } . Then bk = amk is a
monotonically (strictly) decreasing sequence.
Case 2. M is finite. Let n1 be larger than the maximum number in M .
Now since n1 ∈ / M, ∃ n2 > n1 such that an1 ≥ an2 . Furthermore n2 ∈ /M
since it is larger than n1 . Thus we can continue this procedure to obtain a
monotonically increasing sequence an1 ≥ an2 ≥ an3 ≥ . . . .
26
QED
Combining this last proposition with the basic theorem about monotonic
sequences (Theorem 2.3.1), we obtian the famous result attributed to the
Czech mathematician and philosopher Bernard Bolzano (1781 to 1848) and
the German mathematician Karl Weierstrass (1815 to 1897):
Note: A bounded sequence may have many convergent subsequences (for ex-
ample, a sequence consisting of a counting of the rationals has subsequences
converging to every real number) or rather few (for example a convergent
sequence has all its subsequences having the same limit). In fact we define:
In simple language, limsup is the largest and liminf is the smallest accumu-
lation point of a sequence. Moreover a sequence is convergent iff limsup is
equal to liminf, so that there is exactly one (finite) accumulation point.
27
2.4 Cauchy sequences
Informally this means that any two terms (not just for two consecutive
terms!) of a Cauchy sequence can be made arbitrarily close to each other
if we go far enough in the sequence. Note that this definition does not use
the limit of the sequence. Here are some immediate properties of Cauchy
sequences:
Proof: By the proposition above, we only have to show that every Cauchy
sequence is convergent, so let (an ) be a Cauchy sequence. Since a Cauchy se-
quence is bounded it has a convergent subsequence bk = ank by the Bolzano-
Weierstrass theorem. Let b be the limit of the subsequence. Then given
> 0, ∃nK such that all terms of the subsequence bk with k ≥ K lie in
the 31 -neighbourhood of b. Now since (an ) is Cauchy ∃ n0 such that all
terms an with n ≥ n0 are at a distance less than 31 apart. Now choose
28
N = max(n0 , nK ). Then all terms of the sequence an with n ≥ N are at a
distance less than from b and hence lim an = b
Remarks: The fact that Cauchy sequences in R are the same as convergent
sequences is called the Cauchy criterion for convergence. The completeness
axiom to prove the last result is crucial. For example, any sequence of ra-
tional numbers converging to an irrational number is a Cauchy sequence
that is trying to converge but cannot converge in Q. In fact, Cantor (1845
- 1918) used the idea of a Cauchy sequence of rationals to give a construc-
tive definition of the Real numbers. Spaces (not just R) where all Cauchy
sequences converge are called complete. In fact, one can formulate the com-
pleteness axiom for R in terms of Cauchy sequences (provided we assume
the Archimedean property). Here are some equivalent formulations of the
axiom:
C1 Every subset which is bounded from above has a least upper bound.
C2 Every bounded sequence has a convergent subsequence.
C3 Every Cauchy sequence is convergent.
The complex numbers are denoted by C and they form a field. The
multiplicative inverse of a non-zero complex number z is given by I assume
that you are familiar with simple arithmetic and algebraic properties of these
numbers. For a complex number z = x + iy, x = Re(z) is called the real
Im(z) is called the imaginary part. z̄ = x − i y is the conjugate
part, y = √
and |z| = z z̄ the absolute value (or modulus) of z. The complex numbers
are denoted by C and they form a field. The multiplicative inverse of a
29
non-zero complex number z is given by z −1 = |z|z̄ 2 . I assume that you are
familiar with simple arithmetic and algebraic properties of these numbers.
Using the distance |z − w| between two complex numbers z and w, we can
define a Cauchy sequence of complex numbers in exactly the same way as
for real sequences. Using the Pythagorian identity |x + i y|2 = x2 + y 2 and
the triangle inequality |w + z| ≤ |w| + |z| valid in C, one easily observes that
a sequence of complex numbers (zk ) = (xk + i yk ) is convergent iff the two
real sequences (xk ) and (yk ) are both convergent. So we arrive at one main
property that we need about complex numbers:
Remark: The one property for R that is different from C is that we do not
have an ordering for C. There is no such thing as a positive complex number!
On the other hand, there is an algebraic property that is crucial for C (but
not true in R), namely that C is algebraically closed, i.e., every polynomial
(with complex coefficients) has a zero in C. For example z 2 + 1 = 0 is
solvable in C (but not in R) and the two solutions are ± i .
2.5 Series
Examples:
30
a
1. The geometric series a + a r + a r2 + · · · converges to the sum 1−r if
|r| < 1. It is divergent otherwise (assuming a 6= 0). This follows from the
identity:
n
X 1 − rn+1
rk =
1−r
k=0
Proposition 2.5.1
∞
X 1
kp
k=1
Proof:
1 1 1 1 1 1
s2n −1 = 1 + + + + · · · + + · · · + p + · · · + p
2p 3p 4p 7p
2n−1 2n − 1
1 1 1
≤ 1 + 2 p + 4 p + · · · + 2n−1 p
2 4 2n−1
1 1 1
= 1 + 2 p−1 + p−1 + · · · + p−1
2 4 2n−1
1
<
1 − 21−p
Proposition 2.5.2
∞
X 1
kp
k=1
diverges for p ≤ 1
31
P 1
Proof: We first look at the harmonic series k where p = 1.
1 1 1 1 1
s2n = 1+ + + + · · · + n−1 + ··· + n
2 3 4 2 +1 2
3 1 n−1 1
≥ + 2 + ··· + 2
2 4 2n
n
= 1+
2
Therefore sn is unbounded and the harmonic series
P is divergent. For
< 1, we have k1p > k1 for all k, so nk=1 k1p ≥ nk=1 k1 for all n and
P
0 < pP
1
hence kp diverges for 0 < p < 1.
QED
This follows from the fact that the partial sums of bothP series form mono-
tonically
P increasing sequences and the partial sums for bk dominate those
of ak .
For ease of language, we will say that a statement is true for all suffi-
ciently large k if ∃ K such that the statement is true for all k ≥ K .
P P
Definition 2.6.1 A series ai is said to be absolutely convergent if |ai |
is convergent.
Since |am + · · · + an | < ||am | + · · · + |an ||, it is easily seen, by the Cauchy
criterium, that absolute convergence implies convergence, but the converse
is not true.
32
Proposition 2.6.3 If, for all sufficiently large k , |ak | ≤ C rk for some
positive constants
P C and 0 ≤ r < 1 (C and r are independent of k!), then
the series ak is absolutely convergent. (ak can be complex numbers).
P
Proof: Use the basic comparison P ktest to compare the series |ak | with the
convergent geometric series C r .
QED
1
k
Proof: Choose r = 12 (1 + q). Thew for sufficiently large k , |ak | < r and
so |ak | ≤ rk for all k ≥ K.
QED
Definition
P∞ 2.6.2 kAn alternating series is a series of real numbers of the
form ± k=1 (−1) ak , where all the numbers ak ≥ 0.
Proposition 2.6.4 (Leibniz test for alternating series):
An alternating series ∞ k−1 a satisfying
P
k=1 (−1) k
For your pleasure, the simple proof, which can be found in any elemen-
tary calculus text book, is left as an exercise at the end of this chapter. For
P (−1)k−1
example k is convergent (to log(2) although the harmonic series is
divergent. Series which are convergent but not absolutely convergent are
called conditionally convergent. These series are very sensitive to the order
in which they are summed up. In fact the following rather surprising fact is
true.
33
P
Proposition 2.6.5 Let ak be a conditionally convergent series ofPreal
numbers and
P let S be any real number. Then there is a rearrangement ãn
such that ãn = S .
P P
Proof: LetP p l be the series formed
P by the strictly positive terms of ak
and let qm be the rest.
P SinceP ak is convergent but not absolutely
convergent, both series pl and qm are divergent (to ±∞ respectively).
Let us assume (w.l.o.g.) that S ≥ 0 Now we can always add enough terms of
the positive series to get a smallest (first) partial sum that is strictly larger
than S. After that we add up (or if you prefer subtract!) just enough terms
from the other negative series to make the total strictly less than S. Now
we know that all the terms → 0 (since the original seres is convergent after
all!), so by continuing this process, we arrive at a sequence of partial sums
(with terms rearranged as above) converging to S.
QED
We end this section with a product formula that we need in the next
section for power series.
Proof: Since the two series are absolutely convergent, we can assume w.l.o.g.
that all the terms are non-negative,P so that all the partial sums
P are mono-
tonically increasing. Since sn = nk=0 an → S and tn = nk=0 bn → T ,
the product sequence: sn tn → ST . Therefore, by the Cauchy criterium
s2n t2n − sn tn → 0 as P
n → ∞ . Now sm tm is the sum of all terms ai bj with
i, j ≤ m. Let um = m k=0 cm . Then um is the sum of all terms ai bj with
i + j ≤ m . So sn tn ≤ u2n ≤ s2n t2n and hence, un converges to ST .
34
Given a power series, the principal task is to determine the values of the
variable x, for which the series converges and to study the sum of the power
series as a function of x. Note that a power series always converges at x = c
to the value 0 . The first main result states that a power series converges
(absolutely) for all points inside a circle with center c in the complex plane
and that it diverges outside this circle. (The circle could degenerate to a
point or to the whole plane). We will mainly deal with the case c = 0, since
the general case is just a simple shift.
Examples
a
1. The geometric series a + a x + a x2 + · · · converges to the sum 1−x if
|x| < 1. It is divergent otherwise (assuming a 6= 0).
P∞ 1 k
2. The power series: k=0 k! x is one of the most important power series,
since it defines the exponential function ex for both real and and complex
1 1 k x
values of x. Since (k+1)! xk+1 / k! x = k+1 , this series converges for all values
of x (by the ratio test).
k! xk converges only at x = 0 ( again by the ratio
P
3. The power series:
test).
(−1)k+1 k
4. The series ∞
P
k=1 k x is convergent inside the unit circle |x| < 1. It
arises by integrating a geometric series and defines the logarithm as we see
later in the course).
Proposition 2.7.1
an xn0 is convergent,
P
Proof: We compare with a geometric series. If
an x0 → 0 and so in particular, all the terms are bounded: |an xn0 | ≤ C for
n
some constant C > 0 . Let b = |x|x|0 | < 1 . Then the geometric series C bn
P
is
Pconvergent and hence by the basic comparison theorem, so is the series
an xn , since |an xn | = |an xn0 |bn ≤ Cbn . (ii) follows from a comparison
with a divergent geometric series since |x|x|0 | > 1 .
QED
35
The “largest” value R for such that the power series converges for all
|x| < R and diverges for all |x| > R, is called the radius of convergence.
1
This can be computed by the formula R−1 = limsupn→∞ |an | n (by the root
test) or by the formula R−1 = limn→∞ |a|an+1
n|
|
(ratio test).
Remark: The ratio and the root tests are related by the following relation:
Proposition 2.7.2 For any sequence of positive real numbers (an ) , we have
an+1 √
lim inf ≤ lim inf n an
n→∞ an n→∞
an+1 √
lim sup ≥ lim sup n an
n→∞ an n→∞
I will leave the proof of this proposition as an exercise for you (but
please do it!). We will continue with the more important properties of using
power series to define functions (for example differentiating and integrating
power series) when we deal with uniform convergence in Chapter 7, but
I cannot resist introducing the most important “transcendental” function:
the exponential function, using a power series, in the next section.
for any z ∈ C .
36
Theorem 2.8.1
exp(w + z) = exp(w) exp(z)
for all z, w ∈ C
1 P∞ k 1 2k+1
(ii) sin(x) = 2i (exp(ix) − exp(−ix)) = k=0 (−1) (2k+1)! x
valid for any x ∈ R .
All the other trigonometric functions (tangent, cotangent, secant and
cosecant) can be expressed in terms of cosine and sine.
Trigonometric identities such as cos(α+β) = cos(α) cos(β)−sin(α) sin(β)
sin(α + β) = cos(α) sin(β) + sin(α) cos(β) follow from the fundamental func-
tional equation for the exponential function. The fundamental identity:
(cos θ)2 + (sin θ)2 , which relates the trigonometric functions to the unit cir-
cle x2 + y 2 = 1, follows from the fact that the conjugate of eix is e−ix and
so |eix |2 = 1.
Closely related and very useful functions are the hyperbolic functions
defined by:
37
cosh(x) = 12 (ex + e−x )
sinh(x) = 21 (ex − e−x )
38
2.9 Exercises
√ √ p √ √ p
2.
√ Let ak = k + 10 3− k ; b =
k k + k− k ; and c k = k(1 + 10−3 )−
6
k for k ∈ N . Show that ak > bk > ck for all k < 10 , but ak → 0, bk →
0.5; and ck → +∞ as k → ∞ .
4. Compute: r
√
q
1+ 1+ 1 + ...
√
i.e., the limit of the sequence defined recursively by ak+1 = ak + 1 with
a1 = 1 .
5. Compute:
1
1+ 1
1+ 1+...
1
i.e., the limit of the sequence defined recursively by ak+1 = 1 + ak with
a1 = 1 .
39
7. Cauchy condensation test Suppose that (an ) is a monotonically
P decreasing
sequence of positive real numbers suchPthat the series an is convergent.
Let bn = 2n a2n . Prove that the series bn is also convergent.
10. For each of the the following power series determine all values of x for
which the series (i) converge absolutely; (ii) converges conditionally and (iii)
diverges.
P xn
(i) np with p > 0
P n2 n
(ii) 3n (x + 2)
P (n!)2 n
(iii) (2n)! x
P n n! n
(iv) 2 nn x
12. Suppose that 0 < r < 1 and that a sequence (an ) satisfies the
contraction property: |an+1 − an | ≤ r |an − an−1 | for all n ≥ 10 . Prove that
(an ) is a Cauchy sequence and hence converges.
40
18. Show that e is irrational.
n
19. Show that an = 1+ n1 is a bounded monotonically increasing sequence.
What is the limit?
41
2.9.1 Short solutions to some of the exercises
√ √
2. If k < 106 , then k < 103 and 10−3 k < k .
√ p √ p
So: k + 10−3 k < k + k < (k + 103 ) and ak > bk > ck .
p √ 103
ak = (k + 103 ) − k=√ √ →0
k + 103 + k
√
√ √
q
k 1 1
bk = k+ k− k=p √ √ =q →
k+ k+ k 1
1 + √k + 1 2
√
p √ k 10−3 10−3 k
ck = k(1 + 10−3 ) − k=√ √ =√ → ∞;
k + 10−3 k + k 1 + 10−3
1 1 1 1 1 − (− 21 )k−1 2 1
ak = 1 − + − · · · + (− )k−2 = 1 = (1 − (− )k−1 )
2 4 8 2 1 − (− 2 ) 3 2
Therefore
2
lim ak =
3
since lim(− 12 )k−1 = 0 .
In the general case the limit is
2 a + 2b
a + (b − a) =
3 3
42
To
√ prove an+1 > an for all n , we proceed by induction. For n = 1 : a2 =
2 > 1 = a1 . Assume ak+1 > ak . Then ak+2 = sqrt1 + ak+1 > ak+1 =
sqrt1 + ak . We now show that an ≤ 2 for all n by induction. For n = 1 :
a1 = 1 < 2 . Assume ak < 2 . Then ak+1 = sqrt1 + ak < sqrt1 + 2 < 2 .
Let a = lim an . Then 0 ≤ a ≤ 2 and satisfies a = sqrt1 + a . Therefore a
is equal to the Golden Ratio:
√
5+1
φ=
2
6.
1
(i) limn→∞ b n = 1 .
Proof: Assume first that b > 1 . Then an > 1 and we put an = 1 + hn with
hn > 0 . Then, by the Bernoulli inequality: b = ann = (1 + hn )n ≥ 1 + nhn ,
so 0 < hn ≤ b−1 n and hence hn → 0 . If b < 1 , then since 1b < 1 and so
1
limn→∞ ( 1b ) n = 1 . If b = 1 , there is nothing to prove!
(ii) We set an = 1 + n .
Now n = (1 + n )n > 12 n(n − 1)2 for all n > 2 , by the binomial formula.
1
2
So < sqrt n−1 and hence n → 0 . Therefore limn→∞ n n = 1 .
1
(iii) limn→∞ (b + c) n = max(b, c) .
1 1
Proof: Assume w.l.o.g. that b > c > 0 . (bn + cn ) n = b(1 + q n ) n , where
1 1 1
0 < q = cb < 1 . Now (1+q n ) n < (1+q) n for n > 1 and so Set (1+q n ) n → 0
by (i).
n!
(iv) limn→∞ nn = 0 since
n! nn−1 21 1
n
= ... <
n n n nn n
43
13. By the Cauchy product formula of two absolutely converging series we
have:
∞ ∞ ∞
X 1 X 1 X 1
=
3k 5l 3k 5l
k=0 l=0 k,l=0
1 1 15
which is the given series and therefore its sum is 1− 13 1− 15
= 8 .
Remark: It follows from this that the number of primes is infinite, since the
harmonic series diverges and each integer is a (unique) product of primes!
15.
n n
1 X 1 X i kθ
+ cos(kθ) = (e + e−i kθ )
2 2
k=1 k=0
n n
1X 1 X
= i kθ
e + e−i nθ ei (n−k)θ
2 2
k=0 k=0
1 1− ei (n+1)θ
= (1 + e−i nθ )
2 1 − ei θ
i (n+1) θ2 −i (n+1) θ2 i (n+1) θ2
1 −i n θ i n θ θ e
−i n 2 e − e
= e 2 e 2 + e
θ θ θ
2 ei 2 (e−i 2 − ei 2 )
cos(n 2θ ) sin((n + 1) 2θ )
=
sin( 2θ )
44
Chapter 3
We decree the empty set to be open (this actually follows logically from
the definition). Open intervals on the real line are open and closed intervals
are closed. It is trivial to see that any union of open sets is open and hence
any intersection of closed sets is also closed. However, in general, only finite
intersections of open sets are open and only finite unions of closed sets are
closed. For example, the union of all the closed intervals [ n1 , 1 − n1 ] is the
open interval (0, 1) and the intersection of all open intervals (− n1 , 1 + n1 ) is
the closed interval [0, 1] .
Definition 3.1.2 The interior of A ⊂ R is the union of all open sets con-
tained in A. It is always open and is denoted by int(A) (sometimes also by
Å). The closure of A ⊂ R is the intersection of all closed sets that contain
A. It is obviously closed and is denoted by Ā (sometimes also by clo(A) .
45
3.2 Connected sets
Proof: If ∃ x ∈ (a, b) such that x ∈ / A, then (−∞, x) and (x, +∞) are
two disjoint non-empty open sets such that would disconnect A , i.e , A ⊂
(−∞, x) ∪ (x, +∞). Now let [a, b] ⊂ U ∪ V with U, V two non-empty disjoint
open subsets of R. Then ∃ x ∈ U ∩ [a, b], y ∈ V ∩ [a, b] , where we can assume
that x < y. Let z = lub(U ∩ [x, y]) . If z ∈ U , then z < y (since y ∈ V )
and so there is an open neighbourhood of z contained in U ∩ [x, y], This
contradicts the fact that z is the least upper bound. On the other hand, if
z ∈ V then x < z and so there is an open neighbourhood of z contained in
V ∩ [x, y] contradicting the fact that z is the least upper bound.
QED
46
Let (an ) be sequence contained in a compact set K. Since Kis bounded,
by the Bolzano-Weierstrass theorem, (an ) has a convergent subsequence
whose limit point belongs to K = K, since K is closed. We have thus
proved:
Let us, for the moment call sets with the property that every open cover
has a finite subcover “kompakt” (German for compact!). First of all such a
set is necessarily bounded, because Un = (−n, n), n ∈ N is an open cover for
any subset of R and unless the set is bounded we will never be able to find
a finite subcover. Secondly, a “kompakt” set has to be closed because if p is
a limit point of the set which does not belong to the set, then we can form
the open cover Acn , n ∈ N, where An is the closed interval [p − n1 , p + n1 ] .
This certainly covers A since it covers everything in R except the point p.
In fact we now prove the key result:
47
Since x is in the open set UN for some N , and since xn ∈
/ UN for all n > N ,
we obtain a contradiction.
QED
Any union of open sets is open and hence any intersection of closed sets
is also closed, but in general, only finite unions of open sets are open and
only finite intersections of closed sets are closed. This basic property of open
sets leads to the following definition of an abstract topological space where
every other “superficial” geometric property is stripped away until we are
left with the bare bones of the topology of open subsets.
48
Definition 3.4.5 A subset A in a topological space is said to be compact if
every open cover of A has a finite subcover.
3.5 Exercises
1. Determine for each of the following sets wheter they are (i) open, (ii)
closed, (iii) connected, (iv) bounded and (v) compact:
2. Give an example of an open cover of the set (−1, +1) which does not
admit any finite subcover.
3. Show that a finite union and an arbitrary intersection of compact sets
is again compact
4. Let (Kn )n∈N be a sequence of compact subsets of the real line satisfying
the property: Kk+1 ⊂ Kk , T Such a sequence is called a nested sequence.
Prove that the intersection {Kn |n ∈ N} is non-empty. (For the sake of
simplicity you can assume that each Kn is a closed interval [an , bn ]).
6*. Show that any (non-empty) open subset of R is a countable union of
disjoint open intervals.
49
3.5.1 Solutions to some exercise problems
1.
(i) (−∞, 1] ∪ (0, 5] = (−∞, 5] is not open, closed, connected, not bounded
and not compact.
(ii) (−∞, 1] ∩ [0, 5) = [0, 1] is not open, closed, connected, bounded and
compact.
(iii) {x : |x| ≥ 2} = (−∞, −2] ∪ [2, +∞) is not open, closed, not connected,
not bounded and not compact.
(iv) {z : 0 < |z| ≤ 1} ⊂ C is not open, not closed, connected, bounded and
not compact.
(v) {z : |z| > 2} ⊂ C is open, not closed, connected, not bounded and not
compact.
50
Chapter 4
Continuity
We will begin with the usual , δ-definition for functions on the real line and
show how it can be reformulated purely in terms of open sets.
51
Proposition 4.1.1 A function f : A → R , where A ⊂ R , is continuous
on A if and only if for every open subset U ⊂ R the inverse image f −1 (U ) =
{a ∈ A | f (a) ∈ U } is relatively open in A.
52
Similarly, we may define the difference, product and quotient of functions.
(The quotient is not defined at points where the denominator is zero). The
following arithmetic properties for continuous functions then follow easily
from the corresponding properties for limits
Examples
53
Proof: Let A be a connected set and let f be continuous on A. If f (A) were
disconnected, then by definition, there are two disjoint non-empty open sets
U and V such that f (A) ⊂ U ∪V . Since f is continuous, f −1 (U ) and f −1 (V )
are two non-empty disjoint open sets such that A ⊂ f −1 (U ) ∪ f −1 (V ) which
contradicts the assumption that A is connected.
QED
Using the fact that intervals on the real line are connected we obtain the
“classical” Intermediate Value Theorem of first year calculus.
Since this is one of the key results, let me give you another “more ele-
mentary” proof of the above theorem.
Alternate Proof: By a some simple translation (i.e., a vertical shift of the
graph of the function), we can assume w.l.o.g. that I = [a, b] , f (a) <
0 , f (b) > 0 and that y = 0. We then define two monotonic sequences (one
increasing and the other decreasing) (ak ) and (bk ) such that f (ak ) ≤ 0 and
f (bk ) ≥ 0 for all k ∈ N recursively as follows:
First put: a1 = a, b1 = b . Now suppose ak , bk are defined. Let ck =
1
2 (ak + bk ) be the mid point of [ak , bk ] . If f (ck ) ≤ 0 set ak+1 = ck and
bk+1 = bk . If f (ck ) > 0 set ak+1 = ak and bk+1 = ck .
We then have: a = a1 ≤ · · · ≤ ak ≤ ak+1 < bk+1 ≤ bk ≤ · · · ≤ b1 = b.
Moreover bk+1 − ak+1 = 21 (bk − ak ) , so that by induction bk+1 − ak+1 =
1
2k
(b − a) → 0 as k → ∞ and hence lim ak = lim bk = x , say. Since f is
continuous, lim f (ak ) = f (x) ≤ 0 and lim f (bk ) = f (x) ≥ 0 and therefore
f (x) = 0 .
QED
Remark: The method used in the above proof can be implemented nu-
merically (say on a computer) to find solve an equation of the form f (x) = 0.
It is called the bisection method. The method of false position uses a similar
idea, but this time instead of taking the midpoint of the interval at each in-
teration step, one uses the intersection point of the x−axis with the straight
line joining (ak , f (ak )) and (bk , f (bk )), i.e.
54
.
Another simple consequence is the following:
Proof: We assume w.l.o.g. that f (a) < f (b). Let x < y ; x, y ∈ [a, b] . Define
g(t) = f (y(t)) − f (x(t)) , where x(t) = a + t(x − a) and y(t) = b − t(b − y)
for t ∈ [0, 1] , so that ∀ t ∈ (0, 1) y(t) − x(t) = (1 − t)(b − a) + t(y − x) > 0 .
g[0, 1] ⇒ R is a difference of composition of continuous functions and hence
is continuous. At t = 0 : g(0) = f (b) − f (a) > 0 . If g(1) = f (y) − f (x) < 0 ,
then ∃ t∗ ∈ (0, 1) such that g(t∗ ) = 0, by the Intermediate Value Theorem
applied to g . This contradicts the fact that f is injective since y(t∗ ) > x(t∗ ) ,
so f (x) < f (y) .
QED
An important consequence is the following:
55
QED
As you can see from the proof given above, the theorem is valid in any
topological space, but for real-valued functions it boils down to the following
“Extreme Value Theorem” of first year calculus:
Proof: Since f (K) is a bounded set of R, both its least upper bound:
sup(f (K)) and its greatest lower bound inf (f (K)) exist and since f (K)
is closed they are both elements of f (K). QED
Combining the two basic theorems, we can now say that the continuous
image of a closed, connected and bounded interval [a, b] is also a closed,
connected and bounded interval [m, M ] where m is the absolute minimum
and M is the absolute maximum value of f that are attained on [a, b].
As an application of the above ideas and theorems let us state and prove
the following beautiful result.
56
4.3.1 The Contraction Mapping Principle
Theorem 4.3.3 : Let f : R → R be a continuous map with the following
contraction property:
∃ C ∈ (0, 1) such that |f (x) − f (y)| ≤ C|x − y| for all x, y ∈ R.
Then, ∃ a unique fixed point p ∈ R such that f (p) = p.
Proof:
Step 0. We first show that f is continuous:
∀ > 0 ∃ δ = C such that |x − y| < δ ⇒ |f (x) − f (y)| ≤ C|x − y| < for all
x, y .
Step 1. Construction of a sequence converging to the fixed point:
Define: x1 = 1 and recursively: xk+1 = f (xk ).
Step 2. Proof that the sequence is Cauchy:
|xk+2 − xk+1 | = |f (xk+1 − f (xk )| ≤ C|xk+1 − xk | and hence by induction:
|xk+2 − xk+1 | ≤ C k |x2 − x1 | for all k ∈ N. Therefore:
m m
X X Ck
|xk+m+1 −xk+1 | ≤ |xk+i+1 −xk+i | < |x2 −x1 | C k+i−1 < |x2 −x1 |
1−C
i=1 i=1
57
The key thing to note here is that the choice of δ should depend only
on , the function f and the set A; it has to be independent of the points x
and y.
Examples:
58
4.5 Exercises
1. Prove that every polynomial of odd degree with real coefficients has at
least one real root.
8*. Show that there does not exist continuous function f : R → R which
takes on each value (that it takes on) exactly twice. (in other words the
pre-image of a point under f is either empty or consists of exactly two
points).
59
4.5.1 Short solutions to some of the exercise problems
1. This follows from the Intermediate Value Theorem.
4. Suppose limx→0 f (x) exists and is equal to some real number a . Then
for = 21 (say), ∃δ > 0 such that |f (x) − a| < 21 for all x ∈ (−δ, +δ). Now
each such neighbourhood contains both an irrational point s and a rational
point r ; but then |f (s) − f (r)| = 1 > 21 . Contradiction!
Since f (x) is either zero or one, |x f (x)| ≤ |x| for all x . Therefore: ∀ > 0 ,
we choose δ = to have |x − 0| < δ ⇒ |xf (x) − 0| ≤ |x| < . Therefore
limx→0 x f (x) = 0 .
60
However the function g(x) = x2 is not uniformly continuous on (0, ∞) .
Consider the two sequences xn = n and yn = n + n1 . |xn − yn | = n1 , but
|(xn )2 − (yn )2 | = 2 + n12 ≥ 2 . So there exists “” = 2 > 0 such that for each
δ > 0 , we can find points xn and yn in (0, ∞) with |xn − yn | = n1 < δ and
|(xn )2 − (yn )2 | ≥ 2
61
62
Chapter 5
Differentiation
r(x)
lim =0
x→a, x6=a x−a
.
It follows that if f is differentiable at a, then f is continuous at a , since
near a , f differs from a simple (linear) function l(x) , which is definitely
continuous, by a term r(x) that goes to 0 (even faster than |x − a| !) as
x → a . This suggests the notation:
Definition 5.1.2 We say that a function φ(x) is o(|x − a|) (read: little-oh)
as x → a if limx→a φ(x)
x−a = 0
We say that φ(x) is O(|x − a|) (read: big-oh) as x → a if the ratio
|φ(x)|
|x−a| = l stays bounded as x → a
63
The following arithmetic properties for derivatives then follow easily from
the corresponding properties for limits
Proof:
1 1 1 1 g(x) − g(x + h)
lim − = lim
h→0 h g(x + h) g(x) h→0 g(x + h)g(x) h
QED
64
Proof:
Let b = g(a) , y = g(x) , c1 = f 0 (a) , and c2 = f 0 (b) .
f (y) − f (b) = c1 (y − b) + r1 (y) , where r1 (y) is o(|y − b|) and y − b =
c2 (x − a) + r2 (x) with r2 (x) in o(|x − a|) .
Since f (y) − f (b) = c1 (c2 (x − a) + r2 (x)) + r1 (y) , the chain rule now follows
because |y − b| is O(|x − a|) .
QED
Examples
1(bis). By the elementary arithmetic for derivatives (the last two proposi-
tions) we see that all polynomials and rational functions (wherever they are
defined) are differentiable and their derivatives can be computed easily and
mechanically (by Maple for example!) by the above rules.
ea+h − ea eh − 1
= ea
h h
P∞ hk
but by the definition of the exponential function: eh − 1 = k=1 k! =
2
P∞ hk−2
h + r(h) where r(h) = h k=2 k! is obviously o(|h|) , by the estimates
we learned in Chapter 2.
65
5.2 Local properties of the derivative
Suppose f is differentiable at a and that c = f 0 (a) > 0. Let l(x) = f (a) +
c(x − a) . By the definition of the derivative ∃ a δ-neighbourhood U of a,
1
where we have |f (x) − l(x)| < 10 c|x − a| . This implies (for every x ∈ U )
1 9
f (x) > l(x) − 10 c(x − a) = f (a) + 10 c(x − a) which is > f (a) if x > a and <
f (a) if x < a . So there exists δ > 0 such that a − δ < x < a ⇒ f (x) < f (a)
and a < x < a + δ ⇒ f (x) > f (a) . It follows that f is strictly increasing in
a neighbourhood where the derivative is strictly positive. We can apply the
same argument to −f to show
Proposition 5.2.1 A differentiable function is strictly increasing in an
open interval where the derivative is strictly positive (everywhere in that
open interval) and is strictly decreasing if the derivative is strictly negative.
As you well remember from first year, a relative or local maximum of a
function f is a point a such that f (x) ≤ f (a) for all x in a neighbourhood of
a. Similarly, a relative or local minimum is a point b where f (x) ≥ f (b) for
all x in a neighbourhood of b. We say a is a strict relative maximum (resp.
minimum) if we have strict inequalities in the above definition (except of
course at x = a !). A relative extremum is then the collective term used
for either a relative maximum or a relative minimum. Since f can neither
be strictly increasing nor strictly decreasing near a relative extremum, we
obtain as a corollary of the above proposition the following familiar fact:
In fact we can say more if we assume that not only f but also it’s
derivative f 0 is differentiable near a relative extremum a
66
next section, but for now, let me derive a simple formula for the derivative
of the inverse function (not the reciprocal!) of a continuously differentiable
function at a point where the derivative is non-zero (so it is either positive
or negative).
67
constant on the whole interval because of our assumption f (a) = f (b). A
constant function has derivative zero everywhere. If f is not constant either
the absolute minimum or the absolute minimum occurs at an interior point
in (a, b), but there the function is differentiable and by the corollary from
the last section the derivative vanishes (i.e. it is zero) at that point.
QED
By applying Rolle’s Theorem to the function
f (b) − f (a)
g(x) = f (x) − (x − a)
b−a
which obviously satisfies g(a) = g(b)(= f (a)) and the other continuity and
differentiability requirements of Rolle’s theorem, provided f does, we easily
derive the following major result of this section:
f (b) − f (a)
f 0 (x) =
b−a
.
68
on the derivative is not just at a point! Applying this kind of argument to
the derivative of f we have the following global result about convexity, but
first the definition:
z−x y−z
If z is the point (1 − t) x + t y , then t = y−x and 1 − t = y−x . So
convexity implies
(i) f (z) − f (x) ≤ t f (y) − f (x)
(ii) f (z) − f (y) ≤ (1 − t) f (x) − f (y)
This proves
Proposition 5.3.2 If f is convex on [a, b] , then for any three points x <
z < y in the interval [a, b] ,
z−x y−z
Proof: Just put t = y−x and 1 − t = y−x in (i) and (ii) above.
69
5.3.1 The Logarithm
5.4 Differentiation in Rn
70
Differentiability obviously implies continuity as in the scalar case n = 1 .
If f is differentiable at each point of an open set U , then we say that f
is differentiable in U . f is said to be C 1 on U , if the map a 7→ f 0 (a) is
continuous in U .
We have the usual arithmetical rules for the derivative as in the case of
n = 1:
71
Proof:
Step 0.
By two translations, we may assume that a = 0 and f (a) = 0 . Furthermore
by composing f with a linear map A−1 = (df0 )−1 we may also assume that
df0 = I , where I is the identity matrix. Now let h(x) = f (x) − x . Then
dh0 = 0 .
Step 1.
Since f and hence h are C 1 , ∃ r > 0 , s.t. kdhx k < 21 , for all x ∈ B3r (0) ,
where we define kAk to be the maximum length of all column vectors of A .
Applying the Mean Value Theorem to the components of h we find that for
every x1 , x2 ∈ B3r (0) , |f (x1 )−x1 −f (x2 )+x2 | = |h(x1 )−h(x2 )| < 12 |x1 −x2 | .
So if f (x1 ) = f (x2 ) , then x1 = x2 . Therefore f is injective on B3r (0)
Step 2.
To prove surjectivity, we pick a point y in Br (0) and define (inductively
and constructively!) a sequence as follows:
x0 = 0 , x1 = y , xk+1 = y − h(xk ) = y − f (xk ) + xk
Step 3.
As in Step 1, we have: |xk+1 − xk | = |h(xk ) − h(xk−1 )| < 21 |xk − xk−1 ,
provided xk , xk+1 ∈ B2r By induction we show that xn ∈ B2r (0) , and
hence that |xn+1 − xn | < 12 |xn − xn−1 | for all n .
For n = 0: x0 = 0, x1 = y ∈ Br (0) ⊂ B2r (0) and since kdhk < 12 in B3r (0),
|x2 − x1 | = |h(x1 ) − h(x0 )| < 12 |x1 − x0 | .
Assume xk ∈ B2r (0) , for all k ≤ n . Then |xk+1 − xk | < 21 |xk − xk−1 | <
· · · < 2−k |x1 − x0 | for all k ≤ n and hence P by summing up the geometric
series we obtain: |xn+1 | = |xn+1 − x0 | ≤ nk=0 |xk+1 − xk | < 2|x1 − x0 | < 2r .
Step 4.
We claim now that xk is a Cauchy sequence and hence converges (This
was actually proved in Problem # 1 of your Assignment #2 !). By the
same argument as in Step 3, |xn+k − xn | ≤ 2−n |x1 − x0 | = |y| , so xn is
a Cauchy sequence and it converges to a limit x , which is in B2r , since
|xn − x0 | < 2|x1 − x0 | = 2|y| for all n.
Step 5.
We have thus proved that f restricted to the inverse image of Br (0) is
a bijective map. The differentiability of the inverse function follows in a
similar fashion as in the case of a single variable and the chain rule then
gives the formula for the derivative of the inverse function.
72
Corollary 5.4.1 (Implicit Function Theorem) Suppose f = (f1 , . . . , fm ) :
Rn × Rm → Rm is continuously differentiable in an open neighbourhood of
∂
(a, b) and let f (a, b) = 0 . Let A be the m×m matrix ∂xn+k fk (a, b) k=1,...,m .
If det(A) 6= 0 , then there exists an open neighbourhood U of a and an open
neighbourhood V of b such that ∀ x ∈ A , there exists a unique y = g(x) ∈ B
satisfying the implicit equation: f (x, g(x)) = 0 . Moreover g is differentiable
in A .
73
5.5 Exercises
6.
Let f : R → R be a twice continuously differentiable function with
f (0) = 0, f (1) = 1 , and with f 0 (0) = f 0 (1) = 0 . Prove that ∃ x ∈ [0, 1]
with |f 00 (x)| ≥ 4| .
In more physical terms; a particle which travels a unit distance in unit time
and starts and ends with zero velocity (like landing on Mars!) has at least
at some time an acceleration ≥ 4 (in absolute value).
74
8. Let f : R → R be a strictly increasing and convex function which is
three times differentiable. Assume that: f (0) = 0 . Starting with an initial
value x1 > 0 , we now define inductively a sequence by the formula (Newton’s
method ):
f (xn
xn+1 = xn − 0
f (xn )
Prove that limn→∞ xn = 0 .
75
76
Chapter 6
Integration
These are all (finite) real numbers for any bounded function defined on
a closed and bounded rectangle. U S stands for upper sum, LS for lower
sum, U I for upper integral and LI for lower integral. It is obvious from the
definitions that U S ≥ LS for any partition and that LS(f, R) ≤ LS(f, S)
and U S(f, R) ≥ U S(f, S) if S is a refinement of R, since inf (A) ≤ inf (B)
and sup(A) ≥ sup(B) if B ⊂ A and volume is an additive function on
77
rectangles. Given two partitions R and S, we can define their join (or
common refinement) R ∨ S to be the partition consisting of all intersections
Ri ∩ Sj . R ∨ S is a refinement of both R and S. By looking at the upper
and lower sums of this common refinement we arrive at the following:
Definition 6.1.1
We say that f is Riemann-integrable on R if U I(f ) = LI(f ) and the com-
mon value is defined to be the integral of f on R:
Z
f = U I(f ) = LI(f )
R
78
A by simply setting f (x) = 0 for all points x ∈ / A. We then say that f is
integrable on A if this exetended function is integrable on R and define the
integral to be the integral over R of the extended function. The integral, if
it exists, does not depend on the extension. However, there are complicated
bounded sets A, for which even constant functions are not integrable.
We will leave the proof of these simple properties as an exercise for you.
Monotonicity is an important property and we state some immediate
consequences:
R R
M1 If f (x) ≥ g(x) for all x ∈ A, then A f ≥ A g.
R
M2 If m ≤ f (x) ≤ M for all x ∈ A, then m|A| ≤ A f ≤ M |A|.
R R
M3 f (x) ≤ |f | provided |f | is integrable.
79
Proof: Let m = f (p) be the absolute minimum value of f on [a, b] and
M = f (q) be its absolute maximum value. Then by monotonicity: m =
1
Rb
f (p) ≤ b−a a f (t) dt ≤ M = f (q) . Now since f is continuous, by the
Intermediate Value Theorem, there exists a point c (between p and q ) such
1
Rb
that f (c) = b−a a f (t) dt
QED
We state and prove now the fundamental theorem of calculus, which relates
integration and differentation for functions of a single variable. The for-
mulation of the corresponding theorems in higher dimensions (for example,
Stokes’ theorem that you learned last term in in Math 2A03) are a lot more
elaborate although fundamentally the proofs are based on the following one
dimensional fundamental theorem of calculus.
1
limh→0 (F (x + h) − F (x)) = f (x)
h
It is easy to see from the estimate above that not only is F continuous on
[a, b], but in fact, it is Lipschitz continuous.
QED
80
Rx
The function F (x) = a f (t) dt is then what is known as an “antideriva-
tive” in first year calculus, since it satisfies F 0 (x) = f (x) . It is also some-
times called the indefinite integral. Two antiderivatives differ by a constant
on any connected interval. This follows from the mean value theorem for
derivatives, since two antiderivatives have the same derivative! Perhaps a
more familiar form of the fundamental theorem that you remember from
first year calculus is the following:
Improper integrals are integrals for unbounded sets and/or unbounded func-
tions. We will restrict ourselves in these notes to the real line.
Let (a, b) be an open interval in R. To deal with unbounded intervals,
we will allow a to be −∞ and b to be ∞. Let f : (a, b) → R be a function
not necessarily bounded or continuous. Then:
If the limit exists we say that the improper integral is convergent. The
order of the limits in the definition actually does not matter.
Here are some familiar examples from first year calculus:
R1
1. 0 x−p dx is convergent iff p < 1.
R∞
2. 1 x−p dx is convergent iff p > 1.
81
R∞ −x
3. 0 e dx converges to 1.
R ∞ −x2 √
4. −∞ e dx converges to π.
Definition 6.4.2
Z ∞
Γ(x) = tx−1 e−t dt (x > 0)
0
82
Proposition 6.4.2 The gamma function Γ(x) is positive and satisfies the
functional equation:
Γ(x + 1) = x Γ(x)
If an and bn are two sequences of real numbers we will use the notation
an ∼ bn to mean limn→∞ abnn = 1.
Theorem 6.4.1
√ 1
n! ∼ 2π nn n 2 e−n
Proof: Let
1
dn = log(n!) + n − (n + ) log(n)
2
Then
1 3
dn − dn+1 = − log(n + 1) − 1 − (n + ) log(n) + (n + ) log(n + 1)
2 2
1 n+1
= (n + ) log( )−1
2 n
1 1+x
= log( )−1
2x 1−x
83
1 1+x 1 x3 x5
log( )−1 = (x + + + ···) − 1
2x 1−x x 3 5
x2 x4
= + + ···
3 5
x2
< (1 + x2 + x4 + · · · )
3
x2
=
3(1 − x2 )
1 1 1 1
= ( − ) since x =
12 n n + 1 2n + 1
Therefore
1 1 1
0 < dn − dn+1 < −
12 n n + 1
This simultaneously shows two important properties of the sequence dn :
It follows from (ii) that dn is bounded from below and so by (i) is a conver-
gent sequence with limit lim dn = d .
√
We claim that ed = 2π .
The proof relies on the famous product formula of John Wallis (1616-1703):
2.2.4.4.6.6. . . . (2n)(2n) π
→
1.3.3.5.5.7. . . . (2n − 1)(2n + 1) 2
as n → ∞
(The derivation of this formula is one of the exercise problems at the end of
this chapter)
Taking the square root we get
r
2.4.6. . . . 2n 1 π
√ →
3.5.7. . . . (2n − 1) 2n + 1 2
84
Taking logarithms we find
1 √
2 log(n!) − log((2n)!) + 2n log(2) − log(n) → log( π)
2
On the other hand, by the definition of dn :
1
e2dn −d2n = 2 log(n!) − log((2n)!) + 2n log(2) − (log(n) + log(2))
2
√
Therefore d = lim(2dn − d2n ) = log( 2π)
+ nk=1 (−1)k P (n−k) (1)f (k) (x) − P (n−k) (0)f (k) (a) (x − a)k
P
R1
= (−1)n (x − a)n+1 0 P (t)f (n+1) (a + t(x − a)) dt
85
Functions for wchich the Taylor-Maclaurin formula remains true (in a
neighbourhood of a) when n → ∞ are called analytic.
To obtain the Euler-Maclaurin summation formula we first introduce a
very important sequence of rational numbers named after Bernoulli. The
Bernoulli numbers B2j , are defined as coefficients in the power series ex-
pansion:
∞
x x X x2j
x
= 1 − + (−1)j−1 B2j
e −1 2 (2j)!
j=1
βn (t + 1) − βn (t) = ntn−1
This implies:
(n−2j) n!
β (n−2j−1) (0) = 0 , βn (0) = (−1)j−1 (2j)! B2j ,
86
(n−1) (n)
βn (0) = − 21 n! , βn (0) = n! .
Moreover by differentiating the recurrence relation βn (t + 1) = βn (t) , we
(n−k) (n−k)
have: βn (1) = βn (0) for all k . Putting these all together for β2n in
the Darboux formula we obtain:
f (x) − f (a) − 12 f 0 (x) + f 0 (a) (x − a)
Pn−1 (−1)( j−1)
+ j=1 (2j)! B2j f (2j) (x) − f (2j) (a) (x − a)2j
1
R1
= Rn+1 = (2n)! (x − a)2n+1 0 β2n (t)f (2n+1) (a + t(x − a)) dt
n n ∞
(−1)j
Z
X 1 X
B2j f (2j−1) (m)−f (2j−1) (n)
f (k)− f (x) dx = (f (m)+f (n)) −
m 2 (2j)!
k=m j=1
87
6.5 Exercises
1. Suppose f : [0, 1] → R is integrable. Show that
R R
1 1
(i) exp 0 f (x) dx ≤ 0 exp(f (x)) dx
R 1
1 p p R1
(ii) 0 |f (x)| dx ≤ 0 |f (x)| dx for 0 < p < 1
R 1
1 p p R1
(iii) 0 |f (x)| dx ≤ 0 |f (x)| dx for p > 1
n
q using the sequence of partitions 0 < a < · · · < a q = b , where
3. By
q = n ab , show that
rb Z b
1 b
n
lim n −1 = dx = log( )
n→∞ a a x a
4.
(i) Show that
Z 1
log(1 + x) dx = 2 log 2 − 1
0
(ii) Deduce that
1 (2n)! 4
log n → log
n n n! e
as n → ∞
88
R π n
6. (i) Compute Sn = 2
0 sin 2 (x) dx . (Hint: n Sn = (n − 1) Sn−2 ).
(ii) Deduce Wallis’ formula:
2.2.4.4.6.6. . . . (2n)(2n) π
→
1.3.3.5.5.7. . . . (2n − 1)(2n + 1) 2
9. Find the first four non-vanishing terms of the Taylor-Maclaurin series for
the following functions about the point x = 0 :
Rx
(i) f (x) = 0 log(1 + t) dt
(ii) f (x) = exp(sin(x))
(iii) f (x) = log( sin(x)
x ) f (0) = 0
(iv) f (x) = x cot(x) f (0) = 1
89
6.5.1 Hints and short solutions
d
3(i) (1 + x) log(1 + x) − x = log(1 + x) . Therefore by
dx the fundamental
R1
theorem of calculus: 0 log(1 + x) dx = 2 log 2 − 1 = log 4e , since 1 log 1 −
0 = 0.
(ii) Choose the partition 0 < n1 < . . . < n−1
n < 1 and form the Riemann
R1
sum for the integral 0 log(1 + x) dx :
Q
1 Pn n (2n)!
k 1 k+n
, but nk=1 k+n
Q
n k=1 log(1+ n ) = n log k=1 n n = n! nn and the Rie-
1
mann sum converges to the integral as n → ∞ . Therefore log n(2n)!
n
n n! →
log 4e as n → ∞ .
90
Since 0 ≤ sin x ≤ 1 , for x ∈ [0, π2 ] we have 0 < Sn ≤ Sn+1 ≤ Sn+2
and because n Sn = (n − 1) Sn−2 . limn→∞ SS2n+1 2n
= 1 . This proves Wallis’
formula:
2.2.4.4.6.6 . . . 2n.2n π
lim =
n→∞ 1.3.3.5.5.7 . . . (2n − 1).(2n + 1) 2
91
92
Chapter 7
Exactly as for Rn , we define an open ball with centre a and radius r > 0
in a metric space to be the set Br (a) = { x ∈ X | d(x, a) < r } and we
call U ⊂ X open if ∀ a ∈ U ∃ r > 0 such that Br (a) ⊂ U . This defines a
93
topology for X in the sense of Chapter 3 , so metric spaces are special cases
of topological spaces and all the general definitions and theorems, using open
sets as the fundamental concept in Chapters 3 and 4 about connectedness,
compactness and continuity hold for metric spaces (in particular!).
As was the case for real numbers, it is easy to see that every convergent
sequence is a Cauchy sequence, but the converse need not be true in a general
metric space.
Examples
0. The Euclidean vector space Rn with the usual metric d(x, y) = ||x − y|| is
the mother of all metric spaces. (Warning: the four dimensional Minkowski
space-time R1,3 of special relativity where the distance is defined by an inner
product that is not positive-definite is not a metric space.) Similarly, Cn is
usually given the same metric as R2n . All these spaces are complete metric
spaces.
1. The vector space Rn can also be equipped with other non-Euclidean
metrics.p Important
Pn examples are defined by the Lp norms : ||x||p defined by
||x||p = i=1 |xi |p (for p > 0). The distance is then given by: d(x, y) =
||x − y||p . (You will check all the axioms as an exercise at the end of this
chapter. The first two properties of the metric are easy to check. The
triangle inequality is a bit harder.) One can extend this definition to infinite
sequences
P of real numbers provided we restrict to those sequences whose Lp -
norm ∞ |x
i=1 i |p is finite. All these spaces are also complete.
94
2. (The most important metric for the purposes of this section!).
Let X = C 0 [a, b] be the set of all real-valued continuous functions on the
compact interval [a, b]. Since all continuous functions on a compact set are
bounded we can define a metric on this space by:
but I will leave the study of these spaces to your next course in analysis.
Remarks
95
This is in contrast to the following much weaker notion of convergence
we could have used:
Remarks
Proof:
The classic proof of this (which thousands of mathematicians had to
learn in the last 100 years) is actually very simple and is based on the fact
that: |f (x)−f (y)| ≤ |f (x)−fk (x)|+|fk (x)−fk (y)|+|fk (y)−f (y)| (triangle
inequality). All three terms on the right-hand-side are made to go to 0 by
using the assumptions of the theorem, so let’s just do what others have done
before us !
Let x ∈ [a, b] and > 0 be given. Then because we have uniform
convergence we can find a K such that the first term |f (x) − fK (x)| and
1
the last term: |fK (y)−f (y)| are simultaneously less than (say) 10 . Now the
1
second term : |fK (x) − fK (y)| can also be made less than 10 by choosing y
1
sufficiently close to x , (i.e. ∃ δ > 0 s.t. |x − y| < δ ⇒ |fK (x) − fK (y)| < 10 ),
since fK is a continuous at x .
3
Adding the three terms together finishes the proof, since 10 <1!
QED
,
so uniform continuity implies commutativity of two different limits.
96
Corollary 7.2.1 C 0 [a, b] is a complete metric space with respect to the met-
ric defined by the supremum norm.
Proof:
(i) Let > 0 be given. Then ∃ K Rsuch that k ≥RK ⇒
b b
supx∈A |fk (x) − f (x)| < b−a ⇒ a (fk − f ) ≤ a |fk − f | <
0 Rx 0 Rx
(ii) By (i): fk → g uniformly on A ⇒ a fk → a g for any x ∈ A. By the
Rx 0
fundamental theorem
Rx of calculus: fk (x) − f (a) = a fk and since fk (x) →
f (x) we have a g = f (x) − f (a). Therefore again by the fundamental
0
theorem of calculus g(x) = f (x)
QED
Remarks
97
7.2.1 Uniform convergence of power series
The main message of this section is that on any compact set which is strictly
inside its radius of convergence a power series converges uniformly and there-
fore we can integrate and differentiate power series term by term on any
compact set inside its radius of convergence. The standard way to establish
this is to prove a general, but very useful comparison theorem, called the
Weierstrass M-Test for a series of functions (not necessarily a power series)
P
Proposition 7.2.4 (Weierstrass M-test) Suppose Mn is convergent se-
ries of strictly positive numbers. If a sequence of continuous functions (fn )
defined on a compact P set A satisfies |fn (x)| ≤ Mn for all x ∈ A , then the
series of functions fn (x) is uniformly and absolutely convergent on A .
Proof: The partial sums Sn (x) = nk=1 fk (x) form a Cauchy sequence
P
in the complete metric space C 0 (A) , since for any n > m :
n
X
|Sn (x) − Sm (x)| ≤ Mk
k=m
98
n
X k n k
BPn (f, x) = f x (1 − x)n−k
n k
k=0
This implies:
n n
X k
BPn (f, x) − f (x) = f − f (x) xk (1 − x)n−k
n k
k=0
99
7.4 Exercises
(i)
n
X
d(x, y) = |xk − yk |
k=1
(ii)
d(x, y) = max |xk − yk |
1≤k≤n
d(x, y)
ρ(x, y) =
1 + d(x, y)
(i)
nx
fn (x) =
1 + |n x|
(ii)
fn (x) = n x (x − 1)n
(iii)
x
fn (x) =
1 + n x2
# 4. Let [[x]] denote the distance from x to the nearest integer (sketch the
graph !). Show that the function:
∞
X [[3k x]]
f (x) =
3n
k=1
100
# 5. Give an example of a sequence of continuous functions fn defined on
[0, 1] that converges pointwise to zero but such that
Z 1
lim fn (x) dx = 1
n→∞ 0
# 6. Recall that the Fibonacci numbers are defined by: an+1 = an + an−1
with a0 = a1 = 1 .
101
7.4.1 Hints and short solutions to the exercises
2. Except for the triangle inequality, the other axioms are trivially satisfied,
since d satisfies them! The triangle inequality follows from the following fact:
If a, b, c are positive real numbers then
a b c
a≤b+c⇒ ≤ +
1+a 1+b 1+c
which I leave as an easy exercise for you. All sets are bounded in the new
metric, since ρ(x, y) ≤ 1 for all x, y .
3.
nx
(i) fn (x) = 1+|n x|
This sequence converges pointwise to the function f (x) = +1 for 0 <
x ≤ 1 and f (0) = 0 . The convergence is not uniform since the limit function
is not continuous.
(ii)fn (x) = n x (x − 1)n This sequence converges pointwise to the constant
function f (x) = 0 for every x ∈ [0, 1] . Although the limit function is
continuous, the convergence is not uniform since the maximumvalue of
1 n 1 n
fn (x) on the interval [0, 1] (occurring at x = n+1 ) is n+1 1 − n+1 which
approaches e > 0 as n → ∞ .
(iii)fn (x) = 1+nx x2 This sequence converges pointwise to the constant func-
tion f (x) = 0 for every x ∈ [0, 1] . The convergence is uniform since and
the maximum value of fn (x) on the interval [0, 1] which occurs at x = √1n
is 2√1 n which approaches zero as n → ∞ . The limit function has to be
continuous (as it is obviously!). Note that the derivative at zero of all the
fn ’s is f 0 (0) = 1 and so they do not converge to f 0 (0) = 0 !
P∞ n
6. Let f (x) = n=0 an x . Then
∞
X ∞
X ∞
X
2 n n+1
(1 − x − x )f (x) = an x − an x − an xn+2
n=0 n=0 n=0
X∞ X∞ X∞
= an xn − an−1 xn − an−2 xn
n=0 n=1 n=2
X∞
an − an−1 − an−2 xn
= 1+x−x+
n=2
= 1
102
√
Convergence is guaranteed for all |x| < 1
α , where α = lim | an+1
an | =
1+ 5
2
is the golden ratio and α < 2 .
√
−1− 5
(ii) Since the roots of the polynomial x2 + x − 1 are −α = 2 and
√
1 −1+ 5
α = 2 , we have a partial fraction decomposition:
∞
X −1 n 1 1 1
an x = 2 = √ − 1
x +x−1 5 x + α x− α
n=0
and comparing coefficients we get the formula for the Fibonacci numbers:
√ √ n+1 √ n+1
5 an = αn+1 − (−1)n+1 α−n−1 = 1+2 5 − 1−2 5 .
7. Prove that: Z 1 ∞
X
x−x dx = n−n
0 n=1
P∞ (−1)k k k
x−x = e−x log(x) = k=0 k! x log(x) , so all we have to prove is that
1
(−1)k
Z
k
xk log(x) dx = (k + 1)−(k+1)
k! 0
103