M.Sc.

MATHEMATICS
FIRST YEAR -SEMESTER-II

REAL ANALYSIS

https://mkuniversity.ac.in/dde/

PMAT22

All copy right privileges are reserved

 M.Sc. First YearMathematics

REAL ANALYSIS
WELCOME
Dear Students,

We welcome you as a student of the second semester

M.Sc., degree course.

This paper deal with the subject REAL ANALYSIS.

The learning material for this paper will be supplemented

by contact lectures. In this book we give new ideas and

techniques which play a prominent role in current research in

global real analysis.

Learning through the Distance Education mode, as you

are all aware, involves self-learning and self- assessment

and in this regard, you are expected to part in disciplined

and dedicated effort. Asour part, we assure of our

guidance and support.

With best regards.

DEPARTMENT OF MATHEMATICS-DDE, MKU

 SYLLABUS

Unit-I: Uniform convergence and Continuity - Uniform convergence and

Integration – Uniform convergence and Differentiation - Double sequences and
series - Iterated limits – Equi continuous Families of Functions - Arzela – Ascoli
Theorem

Unit- II: The Weierstrauss theorem for algebraic polynomials- The Stone -
Weierstrauss Theorem – Power series - The Exponential and Logarithmic
Functions - The Trigonometric Functions – Fourier Series - The Weierstrauss
theorem for the Trigonometric polynomials - The Gamma Functions.

Unit- III: Functions of Several Variables - Linear Transformation - Differentiation -

The Contraction Principle.

Unit- IV: The inverse function Theorem - The Implicit Function Theorem - The
Rank Theorem – Determinants.

Unit V: Integrations of Differential forms- Primitive mappings – partition of unity-

change of variables – differential forms, Simplexes and chains and Stoke’s
theorem.

Text Book: Walter Rudin, Principles of Mathematical Analysis, McGraw Hill

International Editions (1976)

Reference Books:

1. Patrick M. Fitzpatrick Advanced Calculus, Amer. Math. Soc. Pure and

AppliedUndergraduate Texts, Indian Edition, 2006

2. Apostol, Mathematical Analysis, Narosa Publishing House, Indian edition, 1974.

_________________________________________________________________

Lesson Compiled By:

Dr. T. Asir
Assistant Professor and Head i/c
Department of Mathematics-DDE, Madurai Kamaraj University, Madurai.
 Space for Hints

UNIT – 1

SEQUENCES AND SERIES OF FUNCTIONS

Unit Structure:
Section 1.1: Discussion of the main problem
Section 1.2: Uniform convergence and continuity
Section 1.3: uniform convergence and integration
Section 1.4: uniform convergence and Differentiation
Section 1.5: Double sequences and series, Iterated limits
Section 1.6: Equi continuous families of functions
Section 1.7: Arzela – Ascoli Theorem

Introduction

In this unit we confine our attention to complex valued functions, although

many of the theorems and proofs which follow extend without difficulty to
vector-valued functions, and to mappings into general metric spaces. We
discuss the concept of uniform convergence and continuity,differentiation and
integration. This chapter also include Equi continuous families of functions
and Arzela – Ascoli Theorem.

Definition 1.1:

Suppose 𝑓𝑛 , 𝑛 = 1,2, … , is a sequence of functions defined on a set 𝐸,

and suppose that the sequence of numbers {𝑓𝑛(𝑥)} converges forevery
𝑥 ∈ 𝐸.We can then define afunction 𝑓 by

𝑓 𝑥 = lim𝑛 →∞ 𝑓𝑛 (𝑥)( 𝑥 ∈ 𝐸 ) (1)

{𝑓𝑛} converges to 𝑓 point wise on 𝐸 if(1) holds.

Similarly, if 𝛴𝑓𝑛 (𝑥) converges for every 𝑥 ∈ 𝐸,and if wedefine
∞
𝑓 𝑥 = 𝑛 =1 𝑓𝑛 𝑥 𝑥 ∈ 𝐸 , the function 𝑓 iscalled the sum of the series

𝑓𝑛 .
To say that 𝑓 is continuous at 𝑥meanslim𝑡→𝑥 𝑓 𝑡 = 𝑓(𝑥).

1
Space for Hints

Hence to askwhether the limit of a sequence ofcontinuous functions{𝑓𝑛 (𝑡)} is

continuous is the same astoask
whetherlim𝑡→𝑥 lim𝑛 →∞ 𝑓𝑛 (𝑡) = lim𝑛→∞ lim𝑡→𝑥 𝑓𝑛 𝑡 .
That iswhethertheorderinwhichlimitprocessesarecarried out is immaterial. On
the LHS , we first let𝑛 → ∞then 𝑡 → 𝑥; on the RHS 𝑡 → 𝑥 first,then 𝑛 → ∞.

We shall now show by means of several examples that limit processes cannot
in general be interchanged without affecting the result.

Example 1.2:
𝑚
For 𝑚 = 1,2, . . ., and 𝑛 = 1,2, . . ., let𝑠𝑚 ,𝑛 = 𝑚 +𝑛 .
Then, foreveryfixed 𝑛,
1
lim𝑛→∞ 𝑠𝑚 ,𝑛 = lim 𝑛 = 1.
𝑛 →∞ 1+𝑚
So thatlim𝑛→∞ lim 𝑠𝑚 ,𝑛 = 1
𝑚 →∞
On the other hand, for every fixed𝑚, lim𝑛→∞ 𝑠𝑚 ,𝑛 = 0 so that
lim lim 𝑠𝑚 ,𝑛 = 0.
𝑚 →∞ 𝑛→∞

Example 1.3:

𝑥2
Let 𝑓𝑛 (𝑥) (x real; n=0,1,2,...),and consider
1+𝑥 2 𝑛
∞ 𝑥2
𝑓(𝑥) = 𝑓𝑛 (𝑥) = 𝑛=1 1+𝑥 2 𝑛 (1)

Since 𝑓𝑛 (0) = 0, we have 𝑓(0) = 0. For 𝑥 ≠ 0, the last series in (1) is

convergent with sum.

0 𝑥=0
Hence 𝑓 𝑥 = 2
1+𝑥 𝑥 ≠0
𝑓𝑛 is a convergent series of continuous functions and have a discontinuous
sum.

Example 1.4:

For 𝑚 = 1,2, . . .,put 𝑓𝑚 𝑥 = lim𝑛→∞ (cos 𝑚! 𝜋𝑥)2𝑛

When 𝑚! 𝑥 is an integer , 𝑓𝑚 (𝑥) = 1. For all other values of 𝑥, 𝑓𝑚 (𝑥) = 0.

Now let 𝑓 𝑥 = lim𝑚 →∞ 𝑓𝑚 𝑥 .

For irrational 𝑥, 𝑓𝑚 (𝑥) = 0 for every 𝑚; hence 𝑓(𝑥) = 0.
2
 Space for Hints

For rational 𝑥, say 𝑥 = 𝑝/𝑞, where 𝑝 and 𝑞 are integers, we see that 𝑚! 𝑥
is an integer if 𝑚 ≥ 𝑞, so that 𝑓(𝑥) = 1.
0 𝑥 𝑖𝑟𝑟𝑎𝑡𝑖𝑜𝑛𝑎𝑙
Hence lim𝑚 →∞ lim𝑛→∞ 𝐶𝑜𝑠 𝑚! 𝜋𝑥 2𝑛 =
1 𝑥 𝑟𝑎𝑡𝑖𝑜𝑛𝑎𝑙
That is an everywhere discontinuous limit function, which is not Riemann-
integrable.

Example 1.5:
𝑛𝑥
Let 𝑓𝑛 𝑥 = 𝑠𝑖𝑛 √𝑛 (𝑥 real; 𝑛 = 1,2,3, . . . ), and𝑓 𝑥 = lim𝑛 →∞ 𝑓𝑛 𝑥 .
Then 𝑓‘(𝑥) = 0, and 𝑓𝑛′ 𝑥 = √𝑛𝑐𝑜𝑠 𝑛𝑥, so that {𝑓𝑛‘} does notconverge
to𝑓‘.For instance𝑓𝑛 ′(0) = √𝑛 → +∞ as 𝑛 → ∞ , whereas 𝑓’(0) = 0.

Example 1.6:

Let 𝑓𝑛 𝑥 = 𝑛2 𝑥 1 − 𝑥 2 𝑛
(0 ≤ 𝑥 ≤ 1, 𝑛 = 1,2,3, . . . ) (1)

For 0 ≤ 𝑥 ≤ 1, we have lim𝑛→∞ 𝑓𝑛 𝑥 = 0, (Since,if 𝑝 > 0 and 𝛼 is

𝑛𝛼
real,thenlim𝑛→∞ = 0)
1+𝑝 𝑛

Since 𝑓𝑛 (0) = 0, we see thatlim𝑛→∞ 𝑓𝑛 𝑥 = 0 (0 ≤ 𝑥 ≤ 1). (3)

1 1
Also∫0 𝑥 1 − 𝑥 2 𝑛 𝑑𝑥 = 2𝑛+2
If we replace 𝑛2 by 𝑛 in (2), (3) still holds, but we have
1 𝑛 1 1
lim𝑛→∞ ∫0 𝑓𝑛 𝑥 𝑑𝑥 = lim 2𝑛 +2 = 2whereas ∫0 lim𝑛→∞ 𝑓𝑛 𝑥 𝑑𝑥 = 0∫
𝑛→∞

Therefore the limit of the integral need notbe equal to the integral of the limit,
even if both are finite.

CYP Questions:
1. Give more examples for the limit of the integral need not be equal

1.2 Uniform Convergence and continuity

Note 1

A sequence of functions {𝑓𝑛 }, 𝑛 = 1,2,3, . . . ., is said to converge uniformly on

𝐸 to a function 𝑓 if for every 𝜀 > 0 there is an integer 𝑁 such that 𝑛 ≥ 𝑁
implies |𝑓𝑛 (𝑥) – 𝑓(𝑥)| ≤ 𝜀 for all 𝑥 ∈ 𝐸. (1)

Note 2: Every uniformly convergent sequence is point wise convergent.

3
Space for Hints

Note 3: The difference between uniform convergent and point wise

convergent is : If {𝑓𝑛}converges point wise on 𝐸, then there exists a function
𝑓 such that, for every 𝜀 > 0 and for every 𝑥 ∈ 𝐸, there is aninteger 𝑁,
depending on 𝜀 and on 𝑥, such that (1) holds if 𝑛 ≥ 𝑁. If{𝑓𝑛} converges
uniformly on 𝐸, it is possible, for each 𝜀 > 0, to find one integer 𝑁 which will
do for all 𝑥 ∈ 𝐸.
Note 4: We say that the series 𝛴𝑓𝑛 (𝑥) converges uniformly on 𝐸 if the
sequence 𝑠𝑛 of partial sums definedby 𝑛𝑖=1 𝑓𝑖 (𝑥) = 𝑠𝑛 (𝑥)converges𝑖 = 1
uniformly on 𝐸.

Theorem1.7(Cauchy criterion for uniform convergence):

The sequence of functions {𝑓𝑛 }, defined on 𝐸, converges uniformly on 𝐸 if and

only if for every 𝜀 > 0 there is an integer 𝑁 such that 𝑛 ≥ 𝑁, 𝑚 ≥ 𝑁, 𝑥 ∈
𝐸 implies |𝑓𝑛 (𝑥) – 𝑓𝑚 (𝑥)| ≤ 𝜀.

Proof:

Suppose {𝑓𝑛 } converges uniformly on 𝐸.

Let 𝑓 be the limit function and let 𝜀 > 0.
Then there is an integer 𝑁 such that 𝑛 ≥ 𝑁, 𝑥 ∈ 𝐸 implies
𝑓𝑛 𝑥 − 𝑓𝑚 𝑥 ≤ 𝜀. Therefore 𝑓𝑛 𝑥 − 𝑓𝑚 𝑥 ≤ 𝑓𝑛 𝑥 − 𝑓 𝑥 +
𝑓 𝑥 − 𝑓𝑚 𝑥 < 𝜀 + 𝜀 = 2𝜀 𝑖𝑓 𝑛 ≥ 𝑁, 𝑚 ≥ 𝑁, 𝑥 ∈ 𝐸.
Conversely, suppose the Cauchy condition holds.
That is for every 𝜀 > 0 there is an integer 𝑁 such that 𝑛 ≥ 𝑁, 𝑚 ≥ 𝑁,
𝑥 ∈ 𝐸 implies|𝑓𝑛 (𝑥) − 𝑓𝑚 (𝑥)| ≤ 𝜀 (1)

By a theorem, {𝑓𝑛 (𝑥)} converges, say to 𝑓(𝑥) for every 𝑥 ( since 𝑅 is

complete). Thus the sequence {𝑓𝑛 } converges on 𝐸, to 𝑓.

We have to prove that the convergence is uniformly.Fix 𝑛, and let 𝑚 →

∞in(1).

Since 𝑓𝑚 (𝑥) → 𝑓(𝑥) as 𝑚 → ∞, this gives |𝑓𝑛 (𝑥) – 𝑓(𝑥)| ≤ 𝜀 for every
𝑛 ≥ 𝑁 and every 𝑥 ∈ 𝐸.
Therefore {𝑓𝑛} converges uniformly to 𝑓 on 𝐸.
Theorem 1.8:

Suppose 𝑙𝑖𝑚𝑛→∞ 𝑓𝑛 𝑥 = 𝑓 𝑥 𝑥 ∈ 𝐸 . Put 𝑀𝑛 = sup𝑥∈ 𝐸 |𝑓𝑛 (𝑥) −

𝑓(𝑥)|.Then 𝑓𝑛→ 𝑓 uniformly on 𝐸 if and only if𝑀𝑛→ 0 as 𝑛→∞.

4
 Space for Hints

Proof:

Suppose 𝑓𝑛 → 𝑓 uniformly on 𝐸.
By definition, for every 𝜀 > 0 there is an integer 𝑁 such that 𝑛 ≥ 𝑁 implies
|𝑓𝑛 (𝑥) – 𝑓(𝑥)| ≤ 𝜀 for all 𝑥 ∈ 𝐸.
Therefore sup𝑥∈ 𝐸 |𝑓𝑛 (𝑥) − 𝑓(𝑥) | ≤ 𝜀
That is 𝑀𝑛 ≤ 𝜀 if 𝑛 ≥ 𝑁.That is𝑀𝑛 → 0 as 𝑛 → ∞.
Conversely, suppose 𝑀𝑛 → 0 as 𝑛 → ∞.
Then, given 𝜀 > 0 there is an integer 𝑁 such that 𝑛 ≥ 𝑁 ⇒ 𝑀𝑛 ≤ 𝜀.
That is 𝑛 ≥ 𝑁 sup𝑥 ∈ 𝐸 |𝑓𝑛 (𝑥) − 𝑓(𝑥)| ≤ 𝜀 ⇒ |𝑓𝑛(𝑥) − 𝑓(𝑥)| ≤ 𝜀 for
every 𝑥 ∈ 𝐸.

Therefore 𝑓𝑛 → 𝑓 uniformly on 𝐸.

Weierstrass theorem on uniform convergence.

Theorem 1.9:
Suppose{𝑓𝑛 }isasequenceof functions defined on 𝐸, and suppose
|𝑓𝑛 (𝑥)| ≤ 𝑀𝑛 (𝑥 ∈ 𝐸, 𝑛 = 1,2,3, . ).
Then 𝛴𝑓𝑛 convergence uniformly on 𝐸 if 𝛴 𝑀𝑛 converges.
Proof:
Suppose 𝛴𝑀𝑛 converges.
Then,forarbitrary𝜀 > 0, 𝑚 𝑚
𝑖=𝑛 fi x ≤ 𝑖=𝑛 Mi ≤ ε ( x ∈ E )
provided 𝑚 and 𝑛 are large enough.
That is there is an integer 𝑁 such that 𝑛 ≥ 𝑁, 𝑚 ≥ 𝑁, 𝑥 ∈ 𝐸 implies
|𝑓𝑛 (𝑥)– 𝑓𝑚 (𝑥)| ≤ 𝜀.
By theorem 1.7, 𝛴 𝑓𝑛 convergence uniformly on 𝐸.

Theorem 1.10:

Suppose 𝑓𝑛 → 𝑓 uniformly on a set 𝐸 in a metric space.Let 𝑥 be a limit point

of 𝐸, and suppose that lim𝑡→𝑥 fn(t) = 𝐴𝑛 (𝑛 = 1,2,3, . . . . ).
Then 𝐴𝑛 converges,and lim𝑡→𝑥 𝑓𝑛 (𝑡) = lim𝑛→∞ 𝐴𝑛 . In other words, the
conclusion is thatlim𝑡→𝑥 lim𝑛 →∞ 𝑓𝑛 (𝑡) = lim𝑛→∞ lim𝑡→𝑥 𝑓𝑛 (𝑡) .
Proof:
Let 𝜀 > 0 be given.Since {𝑓𝑛 }converges uniformly on 𝐸, there exists 𝑁 such
that 𝑛 ≥ 𝑁, 𝑚 ≥ 𝑁, 𝑡 ∈ 𝐸|𝑓𝑛 (𝑡) − 𝑓𝑚 (𝑡)| ≤ 𝜀 (1)

By hypothesis,for 𝑛 = 1,2,3, . . . . . . , lim𝑡→𝑥 𝑓𝑛 (𝑡) = 𝐴𝑛 (2)

Letting 𝑡 → 𝑥 in (1) and using (2), we get 𝑛, 𝑚 ≥ 𝑁 implies |𝐴𝑛 – 𝐴𝑚 | ≤ 𝜀
Therefore {𝐴𝑛}is a Cauchy sequence in the set of real numbers 𝑅.
Since 𝑅 is complete, {𝐴𝑛 } converges to some point, say, 𝐴.

5
Space for Hints

|𝑓(𝑡) − 𝐴| ≤ |𝑓(𝑡) − 𝑓𝑛 (𝑡) + 𝑓𝑛 (𝑡) − 𝐴𝑛 + 𝐴𝑛 − 𝐴| ≤ |𝑓(𝑡) −

𝑓𝑛 (𝑡)| + |𝑓𝑛 (𝑡) − 𝐴𝑛 | + 𝐴𝑛 − 𝐴 . (3)
Since 𝑓𝑛 → 𝑓 uniformly on 𝐸, choose a positive integer 𝑛 such that
|𝑓(𝑡) − 𝑓𝑛(𝑡)| ≤ 𝜀 , for all 𝑡 ∈ 𝐸 (4)
Since 𝐴𝑛 → 𝐴,we have 𝑚 ≥ 𝑁 implies |𝐴𝑛 − 𝐴| ≤ 𝜀 (5)
Then, forthis 𝑛,wechoosea neighborhood 𝑉 of 𝑥 such that
|𝑓(𝑡) − 𝐴𝑛 | ≤ 𝜀if 𝑡 ∈ 𝑉 ∩ 𝐸, 𝑡 ≠ 𝑥.Sincelim𝑡→𝑥 𝑓𝑛 (𝑡) = 𝐴𝑛 ) (6)
Using (4),(5) and (6) in (3),we get |𝑓(𝑡) − 𝐴| ≤ 𝜀 + 𝜀 + 𝜀 = 3𝜀 if
𝑡 ∈ 𝑉 ∩ 𝐸, 𝑡 ≠ 𝑥.
That is lim𝑡→𝑥 𝑓(𝑡) = 𝐴 = lim𝑛 →∞ 𝐴𝑛
That islim𝑡→𝑥 lim𝑛 →∞ 𝑓𝑛 (𝑡) = lim𝑛 →∞ lim𝑡→𝑥 𝑓𝑛 𝑡 .

Theorem 1.11:
(Corollary to theorem 1.10) If {𝑓𝑛 }is a sequence of continuous functions on 𝐸,
and if 𝑓𝑛 → 𝑓 uniformly on 𝐸, then 𝑓 is continuous on 𝐸.
Proof:
Since {𝑓𝑛 } is a sequence of continuous functions on 𝐸, for every𝑛, we have
lim𝑡→𝑥 𝑓𝑛 (𝑡) = 𝑓𝑛 (𝑥)By theorem 1.10, we have
lim lim 𝑓𝑛 (𝑡) = lim lim 𝑓𝑛 (𝑡)
𝑡→𝑥 𝑛 →∞ 𝑛 →∞ 𝑡→𝑥
𝑖. 𝑒 lim𝑡→𝑥 𝑓𝑛 (𝑡) = lim 𝑓𝑛 (𝑥) = 𝑓(𝑥) since 𝑓𝑛 → 𝑓 uniformly on 𝐸 . 𝑛 →
∞ 𝑓(𝑡) = 𝑓(𝑥)
By definition of continuous function, 𝑓 is continuous on 𝐸.

Note:Theconverseof theabove theorem is need not be true.

Example 1.12:
𝑓𝑛 (𝑥) = 𝑛2 𝑥 1 – 𝑥 2 𝑛 (0 ≤ 𝑥 ≤ 1, 𝑛 = 1,2,3, . )

Theorem 1.13:

Suppose 𝐾 is compact, and

a. {𝑓𝑛 } is a sequence of continuous functions on 𝐾,
b. {𝑓𝑛 }converges point wisetoa continuous function 𝑓 on 𝐾,
c. 𝑓𝑛 (𝑥) ≥ 𝑓𝑛+1 (𝑥) for all 𝑥 ∈ 𝐾, 𝑛 = 1,2,3, . . . ..
Then 𝑓𝑛 → 𝑓 uniformly on 𝐾.
Proof:
Put 𝑔𝑛 = 𝑓𝑛 – 𝑓.
Since 𝑓𝑛 and 𝑓 are continuous, 𝑔𝑛 is also continuous.
Since 𝑓𝑛 → 𝑓 point wise, 𝑔𝑛 → 0 point wise.
Also , since 𝑓𝑛 (𝑥) ≥ 𝑓𝑛+1 (𝑥) for all 𝑥 ∈ 𝐾, 𝑓𝑛 (𝑥) – 𝑓(𝑥) ≥ 𝑓𝑛+1 (𝑥) – 𝑓(𝑥),
forall 𝑥 ∈ 𝐾 𝑔𝑛 (𝑥) ≥ 𝑔𝑛+1 (𝑥) for all 𝑥 ∈ 𝐾.
6
 Space for Hints

We havetoprovethat𝑓𝑛 → 𝑓 uniformly on 𝐾.
That is to prove that 𝑔𝑛 → 0 uniformly on 𝐾.
Let 𝜀 > 0 be given.
Let 𝐾𝑛 = {𝑥 ∈ 𝐾 \ 𝑔𝑛 (𝑥) ≥ 𝜀. }
That is 𝐾𝑛 = {𝑥 ∈ 𝐾 \ 𝑔𝑛 (𝑥) ∈ [𝜀, ∞)}.
That is 𝐾𝑛 = {𝑥 ∈ 𝐾 \ 𝑥 ∈ 𝑔𝑛 –1 ([𝜀, ∞))}.
That is 𝐾𝑛 = 𝑔𝑛 –1 ([𝜀, ∞)).
Since 𝑔𝑛 is continuous and [𝜀, ∞) is closed,𝐾𝑛 is closed ,and hence 𝐾𝑛 is
compact (since closed subsets are compact).
Let 𝑥 ∈ 𝐾𝑛+1 . Then 𝑔𝑛+1 (𝑥) ≥ 𝜀. Since 𝑔𝑛 (𝑥) ≥ 𝑔𝑛+1 (𝑥) ≥ 𝜀, 𝑥 ∈ 𝐾𝑛.
Then 𝐾𝑛 ⊇ 𝐾𝑛+1 ∀ 𝑛.
Fix 𝑥 ∈ 𝐾. Since 𝑔𝑛 (𝑥) → 0 point wise, we see that 𝑥 ∉ 𝐾𝑛 if 𝑛 is
sufficiently large. Thus 𝑥 ∉∩ 𝐾𝑛 .
In other words, ∩ 𝐾𝑛 is empty.
Therefore 𝐾𝑛 is empty for some 𝑁. It follows that 0 ≤ 𝑔𝑛 (𝑥) < 𝜀, ∀𝑥 ∈
𝐾&∀𝑛 ≥ 𝑁. Therefore | 𝑔𝑛 (𝑥) – 0| < 𝜀 for all 𝑥 ∈ 𝐾 and for all 𝑛 ≥ 𝑁.
That is 𝑔𝑛 → 0 uniformly on 𝐾.That is 𝑓𝑛 → 𝑓 uniformly on 𝐾.

Definition 1.14:
If 𝑋 is a metric space, 𝐶(𝑋) will denote the set of all complex valued,
continuous, bounded functions with domain 𝑋. 𝐶(𝑋)consistsofall complex
Continuousfunctions on𝑋if𝑋 iscompact.
We associate with each 𝑓 ∈ 𝐶(𝑋) its supremum norm
𝑓 = 𝑠𝑢𝑝 𝑓 𝑥 , 𝑥 ∈ 𝐸.
Since 𝑓 is bounded, ||𝑓|| < ∞. Also ||𝑓|| = 0 ⇔ 𝑓(𝑥) = 0 for every
𝑥 ∈ 𝑋.
That is ||𝑓|| = 0 ⇔ 𝑓 = 0.
If 𝑕 = 𝑓 + 𝑔, then |𝑕(𝑥)| = |𝑓(𝑥) + 𝑔(𝑥)| ≤ |𝑓(𝑥)| + |𝑔(𝑥)| ≤ ||𝑓|| +
||𝑔|| for all 𝑥 ∈ 𝑋. Hence ||𝑓 + 𝑔|| = ||𝑕|| ≤ ||𝑓|| + ||𝑔||.
Also 𝐶(𝑋) is a metric space with the metric 𝑑(𝑓, 𝑔) = ||𝑓 – 𝑔||.

Theorem 1.15:

𝐶(𝑋) is a complete metric space.

Proof:

Let {𝑓𝑛 }be a Cauchy sequence in 𝐶(𝑋).

Thereforeto each 𝜀 > 0, there exists apositive integer 𝑁 such that 𝑛, 𝑚 ≥
𝑁 implies ||𝑓𝑛 – 𝑓𝑚 || < 𝜀.

7
Space for Hints

It follows that there is a function 𝑓 with domain 𝑋 to which {𝑓𝑛} converges

uniformly, (by Cauchy criterion for uniform convergence). By theorem 1.11,
𝑓 is continuous.
Since 𝑓𝑛 is bounded and there is an 𝑛 such that |𝑓(𝑥) – 𝑓𝑛 (𝑥)| < 1 for all
𝑥 ∈ 𝑋, f is bounded.Thus 𝑓 ∈ 𝐶(𝑋).
Since 𝑓𝑛 → 𝑓 uniformly on 𝑋, we have||𝑓– 𝑓𝑛 || → 0 as 𝑛 → ∞.

Questions
1. If {𝑓𝑛 } and {𝑔𝑛 } converge uniformly on a set 𝐸, prove that
{𝑓𝑛 + 𝑔𝑛 } converges uniformly on 𝐸.
2. If {𝑓𝑛 } and {𝑔𝑛 } converge uniformly on a set 𝐸 and if {𝑓𝑛 } and {𝑔𝑛 }are
sequences of bounded functions, prove that {𝑓𝑛 𝑔𝑛 } converges
uniformly on 𝐸.

8
 Space for Hints

1.3 Uniform Convergence and integration

Theorem 1.16
Let 𝛼 be monotonically increasing on 𝑎, 𝑏 . Suppose 𝑓𝑛 ∈ ℜ(𝛼) on 𝑎, 𝑏 , and
𝑏 𝑏
∫𝑎 𝑓𝑑𝛼 = lim𝑛→∞ ∫𝑎 𝑓𝑛 𝑑𝛼 .
Proof:
Put 𝜀𝑛 = sup 𝑓𝑛 𝑥 − 𝑓 𝑥 , the supremum being taken over 𝑎 ≤ 𝑥 ≤ 𝑏.
Then 𝑓𝑛 𝑥 − 𝑓 𝑥 ≤ 𝜀𝑛 .
That is −𝜀𝑛 ≤ 𝑓 − 𝑓𝑛 ≤ 𝜀𝑛 . (1)
That is 𝑓𝑛 − 𝜀𝑛 ≤ 𝑓 ≤ 𝑓𝑛 + 𝜀𝑛 .
𝑏 𝑏 𝑏 𝑏
(𝑓𝑛 − 𝜀𝑛 ) 𝑑𝛼 ≤ 𝑓𝑑𝛼 ≤ 𝑓𝑑𝛼 ≤ (𝑓𝑛 + 𝜀𝑛 )𝑑𝛼
𝑎 𝑎 𝑎 𝑎

𝑏 𝑏 𝑏 𝑏
(𝑓𝑛 − 𝜀𝑛 ) 𝑑𝛼 ≤ 𝑓𝑑𝛼 ≤ 𝑓𝑑𝛼 ≤ (𝑓𝑛 + 𝜀𝑛 )𝑑𝛼
𝑎 𝑎 𝑎 𝑎
𝑏 𝑏 𝑏 𝑏
0 ≤ 𝑓𝑑𝛼 − 𝑓𝑑𝛼 ≤ (𝑓𝑛 + 𝜀𝑛 )𝑑𝛼 − (𝑓𝑛 − 𝜀𝑛 )𝑑𝛼
𝑎 𝑎 𝑎 𝑎
𝑏 𝑏
≤ ∫𝑎 𝑓𝑛 + 𝜀𝑛 − 𝑓𝑛 + 𝜀𝑛 𝑑𝛼 = 2 ∫𝑎 𝜀𝑛 𝑑𝛼 = 2𝜀𝑛 [𝛼(𝑏) − 𝛼(𝑎)] (2)
By theorem 1.2.2,𝜀𝑛 → 0 as 𝑛 → ∞. (𝑓𝑛 → 𝑓 uniformly on [𝑎, 𝑏])
𝑏 𝑏
From(1), 0 ≤ ∫𝑎 𝑓𝑑𝛼 − ∫𝑎 𝑓𝑑𝛼 → 0 𝑎𝑠 𝑛 → ∞
𝑏 𝑏
Therefore ∫𝑎 𝑓𝑑𝛼 = ∫𝑎 𝑓𝑑𝛼 (3)
That is 𝑓 ∈ ℛ 𝛼 on [𝑎, 𝑏].
𝑏 𝑏 𝑏 𝑏
Using(3) in (1), we get− ∫𝑎 𝜀𝑛 𝑑𝛼 ≤ ∫𝑎 𝑓𝑑𝛼 − ∫𝑎 𝑓𝑛 𝑑𝛼 ≤ ∫𝑎 𝜀𝑛 𝑑𝛼
𝑏 𝑏
𝜀𝑛 [𝛼(𝑏) − 𝛼(𝑎)] ≤ 𝑓𝑑𝛼 − 𝑓𝑛 𝑑𝛼 ≤ 𝜀𝑛 [𝛼(𝑏) − 𝛼(𝑎)]
𝑎 𝑎
𝑏 𝑏
That is ∫𝑎 𝑓𝑑𝛼 − ∫𝑎 𝑓𝑛 𝑑𝛼 ≤ 𝜀𝑛 𝛼 𝑏 − 𝛼 𝑎 → 0 as 𝑛 → ∞.
𝑏 𝑏
Therefore ∫𝑎 𝑓𝑑𝛼 = lim𝑛→∞ ∫𝑎 𝑓𝑛 𝑑𝛼.

CYP Questions:

∞
1. If𝑓𝑛 ∈ ℛ(𝛼).on 𝑎, 𝑏 andif𝑓(𝑥) = 𝑛=1 𝑓𝑛 (𝑥) (𝑎 ≤ 𝑥 ≤ 𝑏),
𝑏
theseries converging uniformly on [𝑎, 𝑏], then ∫𝑎 𝑓𝑑𝛼 =
∞ 𝑏
𝑛 ∫𝑎 𝑓𝑛 𝑑𝛼

9
Space for Hints

1.4Uniform convergence and differentiation

Theorem 1.17:
Suppose {𝑓𝑛 } is a sequence of functions, differentiable on [𝑎, 𝑏] and such that
{𝑓𝑛 (𝑥0 )} converges for some point 𝑥0 on [𝑎, 𝑏]. If {𝑓𝑛 ′} converges uniformly
on [𝑎, 𝑏], then{𝑓𝑛 } converges uniformly on [𝑎, 𝑏],toafunction𝑓,and
𝑓 ′ 𝑥 = lim 𝑓𝑛 ′(𝑥) (𝑎 ≤ 𝑥 ≤ 𝑏)
𝑛→∞
Proof:
Let 𝜀 > 0 be given.Since {𝑓𝑛 (𝑥0 )}converges for some point 𝑥0 on [𝑎, 𝑏], and
every convergent sequence is Cauchy, choose 𝑁 such that 𝑛 ≥ 𝑁, 𝑚 ≥
𝜀
𝑁, 𝑡 ∈ 𝐸 implies 𝑓𝑛 𝑥0 − 𝑓𝑚 𝑥0 ≤ 2 (1)
Also, since {𝑓𝑛 ′} converges uniformly on [𝑎, 𝑏], say to 𝑓‘, we have
𝜀
|𝑓𝑛 ′(𝑡) − 𝑓𝑚 ′(𝑡)| < 2 𝑏−𝑎 (𝑎 ≤ 𝑡 ≤ 𝑏) (2)

Apply Mean Value theoremtothe function 𝑓𝑛 – 𝑓𝑚 , we get

|(𝑓𝑛 − 𝑓𝑚 )(𝑥) − (𝑓𝑛 − 𝑓𝑚 )(𝑡)| = |𝑓𝑛 (𝑥) − 𝑓𝑛 (𝑡) + 𝑓𝑚 (𝑡)|

= |(𝑓𝑛 (𝑥) − 𝑓𝑛 (𝑡)) + (𝑓𝑚 (𝑥) − 𝑓𝑚 (𝑡))|
≤ |𝑥 − 𝑡||𝑓𝑛 ′(𝑡) − 𝑓𝑚 ′(𝑡)| (by MVT)
𝜀
< |𝑥 − 𝑡| 2 𝑏−𝑎 (by(2)) (3)
𝜀
≤ 2 .for any 𝑥 and 𝑡 on[𝑎, 𝑏], if 𝑛, 𝑚 ≥ 𝑁 (4)
Also 𝑓𝑛 𝑥 − 𝑓𝑚 𝑥 = 𝑓𝑛 𝑥 − 𝑓𝑚 𝑥 − 𝑓𝑛 𝑥0 + 𝑓𝑚 𝑥0 + 𝑓𝑛 𝑥0 −
𝑓𝑚𝑥0≤𝑓𝑛𝑥− 𝑓𝑚𝑥− 𝑓𝑛𝑥0+ 𝑓𝑚𝑥0𝑓𝑛𝑥0− 𝑓𝑚𝑥0<𝜀2 +𝜀2(by(1) and (4))

= 𝜀,for any 𝑥 on [𝑎, 𝑏],if 𝑛, 𝑚 ≥ 𝑁 Therefore {𝑓𝑛} converges uniformly on

𝑎, 𝑏 .
Let 𝑓 𝑥 = lim𝑛 →∞ 𝑓𝑛 𝑥 𝑎 ≤ 𝑥 ≤ 𝑏 .
𝑓 𝑡 −𝑓 𝑥
Let us now fix a point 𝑥 on [𝑎, 𝑏] and define 𝜙𝑛 𝑡 = 𝑛 𝑡−𝑥𝑛 ,
𝑓(𝑡) − 𝑓(𝑥)
𝜙(𝑡) = 𝑓𝑜𝑟 𝑎 ≤ 𝑡 ≤ 𝑏, 𝑡 ≠ 𝑥.
𝑡−𝑥
𝑓 𝑡 − 𝑓𝑛 𝑥
Allowing 𝑛 → ∞ in 𝜙𝑛 (𝑡), we get lim𝑛→∞ 𝜙𝑛 (𝑡) = lim𝑛→∞ 𝑛 =
𝑡−𝑥
𝑓(𝑡) − 𝑓(𝑥)
= 𝜙(𝑡) (5)
𝑡−𝑥
𝑓 𝑡 −𝑓 𝑥
Also𝑙𝑖𝑚𝑡→𝑥 𝜙𝑛 (𝑡) = lim𝑡→𝑥 𝑛 𝑡−𝑥𝑛 = 𝑓𝑛 ′(𝑥) (6)
and
𝑓(𝑡) − 𝑓(𝑥)
lim𝑡−𝑥 𝜙 𝑡 = lim𝑡−𝑥 𝑡−𝑥 = 𝑓 ′(𝑥), for 𝑎 ≤ 𝑡 ≤ 𝑏, 𝑡 ≠ 𝑥 (7)
𝑓𝑛 𝑥 − 𝑓𝑛 𝑡 −𝑓𝑚 𝑥 + 𝑓𝑚 𝑡 𝜀
The inequality (3)can be rewritten as <2
𝑡−𝑥 𝑏−𝑎
10
 Space for Hints

𝑓𝑛 𝑥 − 𝑓𝑛 𝑡 𝑓𝑚 𝑥 − 𝑓𝑚 𝑡 𝜀
That is − <2
𝑡−𝑥 𝑡−𝑥 𝑏−𝑎
𝜀
That is |𝜙𝑛 (𝑡) − 𝜙𝑚 (𝑡)| < 2 𝑏−𝑎
The above equation shows that {𝜙𝑛 } converges uniformly to 𝜙 for 𝑡 ≠ 𝑥.
Apply theorem 1.10 to {𝜙𝑛}, we get
lim lim 𝜙𝑛 (𝑡) = lim lim 𝜙𝑛 (𝑡).
𝑡→𝑥 𝑛→∞ 𝑛 →∞ 𝑡→𝑥
That islim𝑡→𝑥 𝜙𝑛 (𝑡) = lim𝑛→∞ 𝑓𝑛 ′(𝑥)(by(5) and (6)).
Therefore 𝑓 ′ 𝑥 = lim𝑛 →∞ 𝑓𝑛 ′(𝑥)(by(7)).

Theorem 1.18:

There exists a real continuous function on the real line which is nowhere
differentiable.
Proof:
Define 𝜙(𝑥) = |𝑥|( − 1 ≤ 𝑥 ≤ 1) (1)
Extend the definition of 𝜙(𝑥) to all real 𝑥 by requiring that
𝜙 𝑥+2 = 𝜙 𝑥 . (2)
Then,for all 𝑠 and 𝑡,we have |𝜙(𝑠) − 𝜙(𝑡)| = |𝑠| − |𝑡| ≤ |𝑠 − 𝑡|. (3)
In particular, 𝜙 is continuous on 𝑅1 .
∞ 3 𝑛
Define 𝑓 𝑥 = 𝑛=0 4 𝜙 4𝑛 𝑥 .
∞ 3 𝑛 ∞ 3 𝑛
Since 0 ≤ 𝜙 ≤ 1, we have |𝑓(𝑥)| = 𝑛=0 4 𝜙 4𝑛 𝑥 ≤ 𝑛=0 =
4
∞ 3 𝑛
𝑛=0 4
∞ 3 𝑛 3
Since 𝑛=0 4 is a geometric serieswiththecommonratio < 1 and
4
3 𝑛
∞ 1
hence 𝑛=0 4 converges in 𝑅 .

3 𝑛
Bytheorem1.2.3, theseries ∞𝑛=0 4 𝜙 4𝑛 𝑥
Convergesuniformly𝑜𝑛 𝑅1 . By theorem1.11, 𝑓is continuous on 𝑅1 .
1
Fix a real number 𝑥 and a positiveinteger𝑚. Put𝛿𝑚 = ± 2 . 4−𝑚 where the
sign is so chosen that no integer lies between 4𝑚 𝑥 and 4𝑚 (𝑥 + 𝛿𝑚 ). This
can be done , since 4𝑚 𝑥 + 𝛿𝑚 – 4𝑚 𝑥 = 4𝑚 𝛿𝑚 = 4𝑚 𝛿𝑚 =
1 1 1
4𝑚 ± 2 4−𝑚 = 4𝑚 4−𝑚 2 = 2 .

𝜙 (4 𝑛 (𝑥+𝛿 𝑚 )) − 𝜙(4 𝑛 𝑥)
Define 𝛾𝑚 = .
𝛿𝑚
1 1 1
When 𝑛 > 𝑚, 4 𝛿𝑚 = 4𝑛 ± 2 . 4−𝑚 = ± 2 . 4𝑛 −𝑚 = ± 2 . 22(𝑛−𝑚 ) =
𝑛

22 𝑛 −𝑚 −1
= even integer.
11
Space for Hints

Therefore 𝜑(4𝑛 (𝑥 + 𝛿𝑚 ) – 𝜙(4𝑛 𝑥) = 0. That is 𝛾𝑛 = 0 when 𝑛 > 𝑚.

We conclude that
∞ 3 𝑛 3 𝑛
𝑓(𝑥+𝛿 𝑚 ) − 𝑓(𝑥) 𝑛 =0 4 𝜙 4 𝑛 𝑥+𝛿 𝑚 − ∞
𝑛 =0 𝜙 4𝑛 𝑥 3 𝑛
4 ∞
= = 𝑛=0 4 𝛾𝑛 =
𝛿𝑚 𝛿𝑚

𝑚 3 𝑛
𝑛=0 4 𝛾𝑛 (Since 𝛾𝑛 = 0 𝑤𝑕𝑒𝑛 n>m)
3 𝑚 𝑛 3 𝑚 𝑛
𝑚 −1 3 𝑚 −1 3
≥ 𝛾𝑚 − 𝑛=0 4 𝛾𝑛 ≥ 4 4𝑚 − 𝑛 =0 4 𝛾𝑛 (by (5))
4
3 𝑚 𝑚 −1 3
𝑛
≥ 4𝑚 − 𝑛=0 4 4𝑛 (by (4))
4
𝑚 −1
𝑚
3𝑚 − 1 2. 3𝑚 − 3𝑚 − 1 3𝑚 + 1
≥3 − 3𝑛 ≥ 3𝑚 − ≥ =
3−1 2 2
𝑛=0
As 𝑚 → ∞, 𝛿𝑚 → 0
It follows that 𝑓 is not differentiable at 𝑥.

1.5 Double sequences and series, Iterated limits

Definition 1.19

A function 𝑓 whose domain is 𝑍 + × 𝑍 + is called a double sequence.

NOTE. We shall be interested only in real- or complex-valued double

sequences.

Definition1.20
If 𝑎 ∈ ℂ, we write lim𝑝,𝑞→∞ 𝑓 ( 𝑝, 𝑞) = 𝑎 and we say that the double
sequence 𝑓 converges to 𝑎, provided that the following condition is satisfied:
For every 𝜖 > 0, there exists an 𝑁 such that |𝑓(𝑝, 𝑞) − 𝑎| < 𝜀
whenever both 𝑝 > 𝑁and𝑞 > 𝑁.

Definition 1.21(Iterated limit)

To distinguish lim𝑝,𝑞→∞ 𝑓(𝑝, 𝑞)from lim𝑝→∞ ( lim 𝑓(𝑝, 𝑞)), the first is called
𝑞→∞
a double limit, the second an iterated limit.

Theorem 1.22
Assume that lim𝑝,𝑞→∞ 𝑓(𝑝, 𝑞) = 𝑎. For each fixed 𝑝, assume that the limit
lim𝑝,𝑞→∞ 𝑓(𝑝, 𝑞) exists. Then the lim𝑝→∞ ( lim 𝑓(𝑝, 𝑞)) also exists and has
𝑞→∞
the value a.

12
 Space for Hints

Proof

Let 𝐹(𝑝) = lim𝑞→∞ 𝑓(𝑝, 𝑞). Given 𝜖 > 0, choose 𝑁1 so that

𝜖
𝑓 𝑝, 𝑞 − 𝑎 < 2 ,if 𝑝 > 𝑁1 and 𝑞 > 𝑁1 . (20)

For each 𝑝 we can choose 𝑁2 so that

𝜖
𝑓 𝑝, 𝑞 − 𝑎 < 2 ,if 𝑞 > 𝑁2 . (21)

(Note that 𝑁2 depends on pas well as one.) For each 𝑝 > 𝑁1 choose. 𝑁2 , and
then choose a fixed q greater than both 𝑁𝑖 and 𝑁2 . Then both (20) and (21)
hold and hence
|𝐹 𝑝 − 𝑎| < 𝜖,
Therefore, 𝑙𝑖𝑚𝑝→∞ 𝑓(𝑝) = 𝑎.

NOTE. A similar result holds if we interchange the roles of 𝑝 and 𝑞.

Thus the existence of the double limit lim𝑝,𝑞→∞ 𝑓(𝑝, 𝑞) and of
lim𝑞→∞ 𝑓(𝑝, 𝑞) implies the existence of the iterated
limitlim𝑝→∞ ( lim 𝑓(𝑝, 𝑞)) .
𝑞→∞

The following example shows that the converse is not true.

Example 1.23.
Let
𝑝𝑞
𝑓(𝑝, 𝑞) = 2 (𝑝 = 1, 2, . .. , 𝑞 = 1, 2, . .. ).
𝑝 + 𝑞2
Then lim 𝑓(𝑝, 𝑞) = 0 and hence lim lim 𝑓(𝑝, 𝑞) = 0. But 𝑓(𝑝, 𝑞) =
𝑞→∞ 𝑝→∞ 𝑞→∞
½ when 𝑝 = 𝑞 and 𝑓(𝑝, 𝑞) = ¾ when 𝑝 = 2𝑞, and hence it is clear that
the double limit cannot exist in this case.

Double series

Definition 1.24
Let 𝑓 be a double sequence and let s be the double sequence defined by the
equation
𝑝 𝑞

𝑠 𝑝, 𝑞 = 𝑓(𝑚, 𝑛)
𝑚 =1 𝑛=1

13
Space for Hints

The pair (𝑓, 𝑠) is called a double series and is denoted by the symbol
𝑚 ,𝑛 𝑓(𝑚, 𝑛) or, more briefly, by 𝑓(𝑚, 𝑛). The double series is said to
converge to the sum a if
lim 𝑠(𝑝, 𝑞) = 𝑎
𝑝,𝑞→∞
Each number 𝑓(𝑚, 𝑛) is called a term of the double series and each 𝑠(𝑝, 𝑞) is
a partial sum. If 𝑓(𝑚, 𝑛) has only positive terms, it is easy to show that it
con- verges if, and only if, the set of partial sums is bounded. We say
𝑓(𝑚, 𝑛) converges absolutely if 𝑓 𝑚, 𝑛 converges.

1.6 Equicontinuous families of functions

Definition 1.25

Let {𝑓𝑛} be a sequence of functions defined on a set 𝐸. We say that {𝑓𝑛} is

point wise bounded on 𝐸 if the sequence 𝑓𝑛 𝑥 is bounded for every
𝑥 ∈ 𝐸, that is, if there exists a finite-valued function 𝜙 defined on 𝐸 such
that |𝑓𝑛(𝑥)| < 𝜙(𝑥) (𝑥 ∈ 𝐸, 𝑛 = 1,2,3, . )

We say that {𝑓𝑛}is uniformly bounded on 𝐸 if there exists a number 𝑀 such

that |𝑓𝑛(𝑥)| < 𝑀 (𝑥 ∈ 𝐸, 𝑛 = 1,2,3, . )
Note: If {𝑓𝑛}is uniformly bounded sequence of continuous functions on a
compact set 𝐸, there need not exist a subsequence which converges point wise
on 𝐸.

Example 1.26:

Let 𝑓𝑛 (𝑥) = 𝑠𝑖𝑛 𝑛𝑥 (0 ≤ 𝑥 ≤ 2𝜋, 𝑛 = 1,2,3, . )

Supposethere existsasequence 𝑛𝑘 such that {𝑠𝑖𝑛 𝑛𝑘 𝑥} converges,
for every 𝑥 ∈ [0,2𝜋]. In that case, we have

lim𝑘 →∞ (𝑠𝑖𝑛 𝑛𝑘 𝑥 − 𝑠𝑖𝑛 𝑛𝑘+1 𝑥) = 0, (0 ≤ 𝑥 ≤ 2𝜋) and hence

2
lim sin 𝑛𝑘 𝑥 − sin 𝑛𝑘+1 𝑥 = 0, (0 ≤ 𝑥 ≤ 2𝜋)
𝑘 →∞

2𝜋
ByLebesgue'stheorem, lim ∫0 sin 𝑛𝑘 𝑥 − sin 𝑛𝑘+1 𝑥 2 𝑑𝑥 = 0 (1)
𝑘 →∞

But
2𝜋
∫0 sin 𝑛𝑘 𝑥 − sin 𝑛𝑘+1 𝑥 2 𝑑𝑥 = 2𝜋, which contradicts (1)
0

14
 Space for Hints

Note: Every convergent sequence need not contains a uniformly convergent

subsequence.

𝑥2
For example𝑓 𝑥 = 𝑥 2 + (0 ≤ 𝑥 ≤ 1, 𝑛 = 1,2,3, )
1 − 𝑛𝑥 2
Then | 𝑓𝑛(𝑥) | ≤ 1 so that {𝑓𝑛 }is uniformly bounded on [0,1].

Also lim𝑛 →∞ 𝑓𝑛 (𝑥) = 0, (0 ≤ 𝑥 ≤ 1)

But 𝑓𝑛 (1/𝑛) = 1, (𝑛 = 1,2,3, . . . . ), so that no subsequence can converge

uniformly on [0,1].

Definition 1.27:

A family ℱof complex functions 𝑓defined on a set 𝐸 in a metric space 𝑋 is

said to be equicontinuous on 𝐸 if for every 𝜀 > 0 there exists a 𝛿 > 0 such
that |𝑓(𝑥) – 𝑓(𝑦)| < 𝜀 whenever𝑑(𝑥, 𝑦) < 𝛿, 𝑥 ∈ 𝐸, 𝑦 ∈ 𝐸, and 𝑓 ∈ ℱ .Here
𝑑 denotes the metricof 𝑋.

Note: Every member of an equicontinuous family is uniformly continuous.

Theorem 1.28:

If {𝑓𝑛} is a point wise bounded sequence of complex functions on a countable

set 𝐸, then {𝑓𝑛 } has a subsequence {𝑓𝑛𝑘 } such that {𝑓𝑛𝑘 (𝑥)} converges for
every 𝑥 ∈ 𝐸.

Proof:

Let {𝑥𝑖 }, 𝑖 = 1,2,3, . . ., be the points of 𝐸, arranged in a sequence.

Since {𝑓𝑛 (𝑥1 )} is bounded, there exists a subsequence , which we shall
denote {𝑓1,𝑘 }, such that {𝑓1,𝑘 (𝑥1 )} converges as 𝑘 → ∞. Letusnowconsider
sequences 𝑆1 , 𝑆2 , 𝑆3 , . . . . .,which we represent by the array

𝑆1: 𝑓1,1 𝑓1,2 𝑓1,3 𝑓1,4 . . . . . . . .

𝑆2: 𝑓2,1 𝑓2,2 𝑓2,3 𝑓2,4 . . . . . . . .
𝑆3: 𝑓3,1 𝑓3,2 𝑓3,3 𝑓3,4 . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..
and which havethefollowing properties:
a. 𝑆𝑛 is a subsequence of 𝑆𝑛–1 , for 𝑛 = 2,3,4, . . . ...
b. {𝑓𝑛,𝑘 (𝑥𝑛 )} converges as 𝑘 → ∞.

15
Space for Hints

c. The order in which the functions appearisthesame ineach sequence;

That is, if one function precedes another in 𝑆1 , they are in the same
relation in every 𝑆𝑛 , until one or the other is deleted. Hence, when
going from one row in the above array to the next below, functions
may move to the left but never to the right.

We now go down the diagonal of the arrays; that is, we consider the
sequence 𝑆: 𝑓1,1 𝑓2,2 𝑓3,3 𝑓4,4 . . . . . . ..
By (c), the sequence 𝑆 (except possibly its first 𝑛 – 1 terms) is a subsequence
of 𝑆𝑛 , for 𝑛 = 1,2,3, . . . . ..Hence (b) implies that{𝑓𝑛,𝑛 (𝑥𝑖 )} converges as
𝑛 → ∞,forevery 𝑥𝑖 ∈ 𝐸.

Theorem 1.29:

If 𝐾 is a compact metric space, if 𝑓𝑛 ∈ ℭ(𝐾) for 𝑛 = 1,2,3, . . . .,and if

{𝑓𝑛 }converges uniformly on 𝐾, then{𝑓𝑛 } is , equicontinuous on 𝐾.

Proof:
Let 𝜀 > 0 be given.Since {𝑓𝑛} converges uniformly, there is an integer 𝑁
such that 𝑛 > 𝑁 implies ||𝑓𝑛 – 𝑓𝑁 || < 𝜀. Since continuous functions are
uniformly continuous on compact sets, there is a 𝛿 > 0 such that
|𝑓𝑖 (𝑥)– 𝑓𝑖 (𝑦)| < 𝜀 if 1 ≤ 𝑖 ≤ 𝑁 and 𝑑(𝑥, 𝑦) < 𝛿.
If 𝑛 > 𝑁 and 𝑑(𝑥, 𝑦) < 𝛿, it follows that

|𝑓𝑛 (𝑥) − 𝑓𝑛 (𝑦)| ≤ |𝑓𝑛 (𝑥) − 𝑓𝑁 (𝑥) + 𝑓𝑁 (𝑥) − 𝑓𝑁 (𝑦) + 𝑓𝑁 (𝑦) − 𝑓𝑛 (𝑦)|
≤ |𝑓𝑛 (𝑥) − 𝑓𝑁 (𝑥)| + |𝑓𝑁 (𝑥) − 𝑓𝑁 (𝑦)| + |𝑓𝑁 (𝑦) − 𝑓𝑛 (𝑦)|
≤ ‖ 𝑓𝑛 − 𝑓𝑁 ‖ + |𝑓𝑁 (𝑥) − 𝑓𝑁 (𝑦)| + ‖ 𝑓𝑁 − 𝑓𝑛 ‖
< 𝜀 + 𝜀 + 𝜀 = 3𝜀.

Therefore {𝑓𝑛 } is equicontinuous on 𝐾.

Theorem 1.30:

If 𝐾 is a compact, if 𝑓𝑛 ∈ ℭ(𝐾) for 𝑛 = 1,2,3, . . . ., and if {𝑓𝑛 } is point wise

bounded and equicontinuous on 𝐾, then
a. {𝑓𝑛 } is uniformly bounded on 𝐾,
b. 𝑓𝑛 contains auniformly convergent subsequence.
Proof:
a. Let 𝜀 > 0 be given.
Since {𝑓𝑛 } is equicontinuous on𝐾, there exists a 𝛿 > 0 such that𝑥, 𝑦 ∈ 𝐾

16
 Space for Hints

𝑑 𝑥, 𝑦 < 𝛿 ⇒ 𝑓𝑛 𝑥 − 𝑓𝑛 𝑦 < 𝜀,for all 𝑛. (1)

Since 𝐾 is compact, there are finitely many points 𝑝1 , 𝑝2 , . , 𝑝𝑟 in 𝐾 such that to
every 𝑥 ∈ 𝐾 corresponds at least one 𝑝𝑖 with 𝑑(𝑥, 𝑝𝑖 ) < 𝛿.
Since {𝑓𝑛 } is point wise bounded, there exist 𝑀𝑖 < ∞such that
|𝑓𝑛 (𝑝𝑖)| < 𝑀𝑖 for all 𝑛. (2)

If 𝑀 = 𝑚𝑎𝑥(𝑀1 , 𝑀2 , … , 𝑀𝑟 ), then|𝑓𝑛 (𝑥)| = |𝑓𝑛 (𝑥) − 𝑓𝑛 (𝑝𝑖 ) + 𝑓𝑛 (𝑝𝑖 )|

≤ |𝑓𝑛 𝑥 − 𝑓𝑛 (𝑝𝑖 )| + |𝑓𝑛 (𝑝𝑖 )|
< 𝜀 + 𝑀𝑖. (𝑏𝑦(2) & (3))
≤𝜀+𝑀

Therefore, {𝑓𝑛 } is uniformly bounded on 𝐾.

b. Let 𝐸 be a countable dense subset of 𝐾.
By theorem 2.6.1, {𝑓𝑛 } has a subsequence { 𝑓𝑛𝑖 } such that {𝑓𝑛𝑖 (𝑥)} converges
for every 𝑥 ∈ 𝐸 .
Put 𝑓𝑛𝑖 = 𝑔𝑖 .
We shallprovethat{𝑔𝑖 } converges uniformly on 𝐾.
Let 𝜀 > 0 be given.Since {𝑓𝑛 } is equicontinuous on 𝐾, there exists a 𝛿 > 0
such that𝑑(𝑥, 𝑦) < 𝛿 ⇒ |𝑓𝑛 (𝑥) − 𝑓𝑛 (𝑦)| < 𝜀, for all 𝑛. (4)

Let 𝑉(𝑥, 𝛿) = {𝑦 ∈ 𝐸/𝑑(𝑥, 𝑦) < 𝛿}.

Since 𝐸 is dense in 𝐾 and 𝐾 is compact, there are finitely many points
𝑥1 , 𝑥2 , . . . . . , 𝑥𝑚 in 𝐸 suchthat𝐾 ⊂ 𝑉(𝑥1 , 𝛿) ∪ 𝑉(𝑥2 , 𝛿) ∪. . . .∪ 𝑉(𝑥𝑚 , 𝛿).
Since {𝑔𝑖 (𝑥)} converges for every 𝑥 ∈ 𝐸, there is an integer 𝑁 such that
|𝑔𝑖 (𝑥𝑠 ) − 𝑔𝑗 (𝑥𝑠 )| < 𝜀 whenever 𝑖 ≥ 𝑁, 𝑗 ≥ 𝑁, 1 ≤ 𝑠 ≤ 𝑚 (5)

If𝑥 ∈ 𝐾 ⊂ 𝑉(𝑥1 , 𝛿) ∪ 𝑉(𝑥2 , 𝛿) ∪. . . .∪ 𝑉(𝑥𝑚 , 𝛿), then 𝑥 ∈ 𝑉(𝑥𝑠 , 𝛿) for some s

That is 𝑑(𝑥, 𝑥𝑠 ) < 𝛿
⇒ |𝑔𝑖 (𝑥𝑠 ) – 𝑔𝑗 (𝑥𝑠 )| < 𝜀 for every 𝑖.
If 𝑖 ≥ 𝑁, 𝑗 ≥ 𝑁 ,
𝑔𝑖 𝑥 − 𝑔𝑗 𝑥
= |𝑔𝑖 (𝑥) − 𝑔𝑗 (𝑥𝑠 ) + 𝑔𝑖 (𝑥𝑠 ) − 𝑔𝑗 (𝑥𝑠 ) + 𝑔𝑗 (𝑥𝑠 ) − 𝑔𝑗 (𝑥)|
≤ |𝑔𝑖 (𝑥) − 𝑔𝑖 (𝑥𝑠 )| + |𝑔𝑖 (𝑥𝑠 ) − 𝑔𝑗 (𝑥𝑠 )| + |𝑔𝑗 (𝑥𝑠 ) − 𝑔𝑗 (𝑥)|
< 𝜀 + 𝜀 + 𝜀 = 3𝜀 Therefore,{𝑔𝑖 }converges uniformly on 𝐾.
That is {𝑔𝑖 } converges uniformly on 𝐾
That is { 𝑓𝑛𝑖 } converges uniformly on 𝐾.
That is {𝑓𝑛} contains a uniformly convergent subsequence.

1.7 Arzela—Ascoli Theorem

17
Space for Hints

Our setting is a compact metric space 𝑋 which you can, if you wish, take to
be a compact subset of ℝ𝑛 , or even of the complex plane (with the Euclidean
metric, of course). Let 𝐶(𝑋) denote the space of all continuous functions on
𝑋 with values in ℂ (equally well, you can take the values to lie in ℝ). In
𝐶(𝑋) we always regard the distance between functions f and g in 𝐶(𝑋) to
be
𝑑𝑖𝑠𝑡 (𝑓, 𝑔) = 𝑚𝑎𝑥{|𝑓 (𝑥) − 𝑔(𝑥)| ∶ 𝑥 ∈ 𝑋}.
It is easy to check that ―dist‖ is a metric (henceforth: the ―max-metric‖) on
𝐶(𝑋), in which a sequence is convergent iff it converges uniformly on 𝑋.
Similarly, a sequence in 𝐶(𝑋) is Cauchy iff it is Cauchy uniformly on 𝑋. Thus
the max-metric, which from now on we always assume to be part of the
definition of 𝐶(𝑋), makes that space complete. These notes prove the
fundamental theorem about compactness in 𝐶(𝑋):

Example 1.31

Let 𝐾 be the subset of 𝐶( 0,1 , ℝ) Consisting of all continuous functios𝑓 ∶

[0, 1] → ℝ such that 𝑓 𝑥 < 1 for all 𝑥 in [0, 1].
We can verify that 𝐾 is a closed bounded subset of 𝐶([0, 1], ℝ). However, 𝐾
is not a sequentially compact metric space. To establish this assertion, it is
necessary to find a sequence in 𝐾 that has the property that no subsequence
converges unifoimly toa function in 𝐾. For each positive lnteger 𝑘 define the
function 𝑓𝑘 : 0,1 → ℝ by
𝑓𝑘 (𝑥) = 𝑥 𝑘 for 𝑥 in [0, 1].
Then define the function 𝑓 ∶ 0, 1 → ℝ by
0 𝑖𝑓 0 ≤ 𝑥 ≤ 1
𝑓 𝑥 =
1 𝑖𝑓 𝑥 = 1
Since lim𝑘→∞ 𝑥 𝑘 = 0 if 0 ≤ 𝑥 ≤ 1, it follows that the sequence {𝑓𝑘 : 0,1 →
ℝ} convergespointwise to the function 𝑓 ∶ 0,1 → ℝ. Hence every
subsequence of {𝑓𝑘 : 0,1 → ℝ} also converges pointwise to the function
𝑓 ∶ 0, 1 → ℝ. But the function 𝑓 ∶ 0, 1 → ℝ is not continuous and hence is
not in 𝐾. Thus, there is no subsequence of {𝑓𝑘 : 0,1 → ℝ} that converges in
the metric spac 𝐶([0, 1], ℝ) (that is, converges unifoimly) to a function in 𝐾.

There is a theorem, called the Arzela—Ascoli Theorem, that characterizes the

se- quentially compact subspaces of 𝐶([0, 1], ℝ).

The Arzela-Ascoli Theorem 1.32

18
 Space for Hints

If a sequence 𝑓𝑛 1∞ in 𝐶(𝑋) is bounded and equicontinuous then it has a

uniformly convergent subsequence.

In this statement,
(a) “𝐹 ⊂ 𝐶(𝑋) is bounded” means that there exists a positive constant
𝑀 < ∞ such that |𝑓(𝑥)| ≤ 𝑀 for each 𝑥 ∈ 𝑋 and each 𝑓 ∈ 𝐹, and
(b) “𝐹 ⊂ 𝐶(𝑋) is equicontinuous” means that: for every 𝜀 > 0 there exists
𝛿 > 0 (which depends only on 𝜀) such that for 𝑥, 𝑦 ∈ 𝑋:
𝑑(𝑥, 𝑦) < 𝛿 ⇒ |𝑓 (𝑥) − 𝑓 (𝑦)| < 𝜀 ∀𝑓 ∈ 𝐹 ,
where 𝑑 is the metric on 𝑋.

Proof of the Arzela-Ascoli Theorem

.
Step I.
We show that the compact metric space 𝑋 is separable, that is, has a countable
dense subset 𝑆.
Given a positive integer 𝑛 and a point 𝑥 ∈ 𝑋, let
𝐵(𝑥, 1/𝑛) = {𝑦 ∈ 𝑋 ∶ 𝑑(𝑥, 𝑦) < 1/𝑛},
the open ball of radius 1/𝑛, centered at 𝑥. For a given 𝑛, the collection of
all these balls as 𝑥 runs through 𝑋 is an open cover of 𝑥, so (because 𝑋 is
compact) there is a finite subcollection that also covers 𝑋. Let 𝑆𝑛 denote
the collection of centers of the balls in this finite subcollection. Thus 𝑆𝑛 is
a finite subset of 𝑋 that is ―1/𝑛-dense‖ in the sense that every point of 𝑋
lies within 1/𝑛 of a point of 𝑆𝑛 . Clearly the union 𝑆 of all the sets 𝑆𝑛 is
countable, and dense in 𝑋.

Step II. We find a subsequence of {𝑓𝑛 } that converges pointwise on 𝑆.

This is a standard diagonal argument. Let‘s list the (countably many)
elements of 𝑆 as {𝑥1 , 𝑥2 , . . . }. Then the numerical sequence 𝑓𝑛 𝑥1 ∞
𝑛=1
is bounded, so by Bolzano-Weierstrass it has a convergent subsequence,
∞
which we’ll write using double subscripts: 𝑓1,𝑛 𝑥1 𝑛=1 . Now the numer-
∞
ical sequence 𝑓1,𝑛 𝑥2 𝑛=1
isbounded,soithasaconvergentsubsequence
∞ ∞
𝑓2,𝑛 𝑥2 𝑛=1
. Note that the sequence of functions 𝑓2,𝑛 𝑛=1
, since it is a sub-
∞
sequence of 𝑓1,𝑛 𝑛=1 , converges at both 𝑥1 and 𝑥2 . Proceeding in this fashion
we obtain a countable collection of subsequences of our original sequence:
𝑓1,1 𝑓1,2 𝑓1,3 ···
𝑓2,1 𝑓2,2 𝑓2,3 ···
𝑓3,1 𝑓3,2 𝑓3,3 ···

19
Space for Hints

where the sequence in the 𝑛 − 𝑡𝑕 row converges at the points 𝑥1 , . . . , 𝑥𝑛 , and

each row is a subsequence of the one above it.

Thus the diagonal sequence {𝑓𝑛,𝑛 } is a subsequence of the original sequence

{𝑓𝑛 } that converges at each point of 𝑆.

Step III. Completion of the proof.

Let {𝑔𝑛 } be the diagonal subsequence produced in the previous step,

convergent at each point of the dense set 𝑆. Let 𝜀 > 0 be given, and choose
𝛿 > 0 by equicontinuity of the original sequence, so that 𝑑(𝑥, 𝑦) <
𝛿 implies |𝑔𝑛 (𝑥) − 𝑔𝑛 (𝑦)| < 𝜀/3 for each 𝑥, 𝑦 ∈ 𝑋 and each positive
integer 𝑛. Fix 𝑀 > 1/𝛿 so that the finite subset 𝑆𝑀 ⊂ 𝑆 that we produced in
Step I is 𝛿 −dense in 𝑋. Since {𝑔𝑛 } converges at each point of 𝑆𝑀 , there exists
𝑁 > 0 such that
𝑛, 𝑚 > 𝑁 ⇒ |𝑔𝑛 (𝑠) − 𝑔𝑚 (𝑠)| < 𝜀/3∀𝑠 ∈ 𝑆𝑀 . (1)
Fix 𝑥 ∈ 𝑋. Then 𝑥 lies within 𝛿 of some 𝑠 ∈ 𝑆𝑀 , so if 𝑛, 𝑚 > 𝑀 ∶
|𝑔𝑛 (𝑥) − 𝑔𝑚 (𝑥)|
≤ |𝑔𝑛 (𝑥) − 𝑔𝑛 (𝑠)| + |𝑔𝑛 (𝑠) − 𝑔𝑚 (𝑠)| + |𝑔𝑚 (𝑠)
− 𝑔𝑚 (𝑥)|
The first and last terms on the right are< 𝜀/3 by our choice of 𝛿 (which was
possible because of the equicontinuity of the original sequence), and the same
estimate holds for the middle term by our choice of 𝑁 in (1). In summary:
given 𝜀 > 0 we have produced 𝑁 so that for each 𝑥 ∈ 𝑋,
𝑚, 𝑛 > 𝑁 ⇒ |𝑔𝑛 (𝑥) − 𝑔𝑚 (𝑥)| < 𝜀/3 + 𝜀/3 + 𝜀/3 = 𝜀.
Thus on 𝑋 the subsequence {𝑔𝑛 } of {𝑓𝑛 } is uniformly Cauchy, and therefore
uniformly convergent. This completes the proof of the Arzela-Ascoli
Theorem.

CYP QUESTIONS:

1. Suppose {𝑓𝑛 }, {𝑔𝑛 } are defined on 𝐸, and

a. 𝛴𝑓𝑛 has uniformly bounded partial sums;
b. 𝑔𝑛 → 0 uniformly on 𝐸;
c. 𝑔1 (𝑥) ≥ 𝑔2 (𝑥) ≥ 𝑔3 (𝑥) ≥.....for every 𝑥 ∈ 𝐸.
Prove that 𝛴𝑓𝑛 𝑔𝑛 converges uniformly on 𝐸.

UNIT 2
20
 Space for Hints

THE WEIERSTRAUSS THEOREM

Unit Structure:
Section 2.1: The Stone - Weierstrauss Theorem
Section 2.2: Power series
Section 2.3: The exponential and logarithmic functions
Section 2.4: The trigonometric functions
Section 2.5: Fourier Series
Section 2.6: The Gamma Functions

In this unit we shall discuscuss about some properties of functions

which are represented by power series. We also discuss the concept of the
Exponential, Logarithmic,Weierstrauss theorem.

2.1 The Stone - Weierstrauss Theorem

Theorem 2.1:

If 𝑓 is a continuous complex function on [𝑎, 𝑏], there exists a sequence of

polynomials 𝑃𝑛 such that lim𝑥 →∞ 𝑃𝑛 (𝑥) = 𝑓(𝑥) uniformly on [𝑎, 𝑏]. If 𝑓 is
real, the 𝑃𝑛 may be taken real.
Proof:

We may assume, without loss of generality, that [𝑎, 𝑏] = [0,1]. We may also
assume that 𝑓(0) = 𝑓(1) = 0. For if the theorem is proved for this case,
consider 𝑔(𝑥) = 𝑓(𝑥) – 𝑓(0)– 𝑥[𝑓(1) – 𝑓(0)] (0 ≤ 𝑥 ≤ 1). Here
𝑔(0) = 𝑔(1) = 0, and if 𝑔 can be obtained as the limit of a uniformly
convergent sequence of polynomials, it is clear that the same is true for 𝑓,
since 𝑓 – 𝑔 is a polynomial.
Furthermore, we define 𝑓(𝑥) = 0 for 𝑥 outside [0,1]. Then 𝑓 is uniformly
continuous on the whole line.
We put 𝑄𝑛 (𝑥) = 𝑐𝑛 1 − 𝑥2 𝑛 (𝑛 = 1,2,3, . . . . ), (1)
1
where 𝑐𝑛 is chosen so that ∫−1 𝑄𝑛 (𝑥) 𝑑𝑥 = 1(𝑛 = 1,2,3, . . . ) (2)
1
1 1
Now ∫−1 1 − 𝑥 2 𝑛
𝑑𝑥 = 2 ∫0 1 − 𝑥 2 𝑛 √n
𝑑𝑥 ≥ 2 ∫0 1 − 𝑥2 𝑛
𝑑𝑥

≥ 2 ∫0√n 1 − 𝑛𝑥 2 𝑛
𝑑𝑥(by binomial theorem)
21
Space for Hints

1
𝑛𝑥 3 √𝑛
≥ 2 𝑥−
3 0

1 3
1 𝑛 1 1 4 1
√𝑛
=2 − −0 =2 − = ≥
√𝑛 3 √𝑛 3√𝑛 3 √𝑛 √𝑛
1 𝑛
Equation (2) implies that ∫−1 𝑐𝑛 1 – 𝑥 2 𝑑𝑥
1
⇒ 𝑐𝑛 1 − 𝑥2 𝑛
𝑑𝑥 = 1
−1

1
⇒ 1 > 𝑐𝑛 ⇒ 𝑐𝑛 < √𝑛 (3)
√𝑛
For any 𝛿 > 0 (1) and (3) implies that 𝑄𝑛(𝑥) < √𝑛(1 − 𝛿2)𝑛 where
𝛿 ≤ |𝑥| ≤ 1 (4)
∴ 𝑄𝑛 (𝑥) → 0 uniformly in 𝛿 ≤ |𝑥| ≤ 1.
1
Now set 𝑃𝑛 (𝑥) = ∫−1 𝑓 𝑥 + 𝑡 𝑄𝑛 (𝑡) 𝑑𝑡 (5)
−𝑥 1−𝑥
𝑃𝑛 (𝑥) = 𝑓 𝑥 + 𝑡 𝑄𝑛 (𝑡) 𝑑𝑡 + 𝑓 𝑥 + 𝑡 𝑄𝑛 (𝑡) 𝑑𝑡
−1 −𝑥
1
+ 𝑓 𝑥 + 𝑡 𝑄𝑛 (𝑡) 𝑑𝑡
1−𝑥

Put 𝑥 + 𝑡 = 𝑦. Then 𝑑𝑥 = 𝑑𝑡 & 𝑡 =– 1 ⇒ 𝑦 = 𝑥 – 1 , 𝑡 = – 𝑥 ⇒ 𝑦 =

0, 𝑡 = 1– 𝑥 ⇒ 𝑦 = 1, 𝑡 = 1 ⇒ 𝑦 = 𝑥 + 1.
0 1
Therefore 𝑃𝑛 (𝑥) = ∫𝑥−1 𝑓 𝑦 𝑄𝑛 (𝑦 − 𝑥) 𝑑𝑦 + ∫0 𝑓 𝑦 𝑄𝑛 (𝑦 − 𝑥) 𝑑𝑦 +
𝑥+1
∫1 𝑓 𝑦 𝑄𝑛 (𝑦 − 𝑥) 𝑑𝑦
1
=0+ 𝑓 𝑦 𝑄𝑛 (𝑦 − 𝑥) 𝑑𝑦 + 0
0
The RHS integral is clearly a polynomial in 𝑥. Thus {𝑃𝑛 } is a sequence of
polynomials which are real if 𝑓 is real.
.
Therefore, given 𝜀 > 0, we choose 𝛿 > 0 such that |𝑦 – 𝑥| < 𝛿 implies
|𝑓(𝑦) − 𝑓(𝑥)| < 𝜀 . (6)
Let 𝑀 = sup𝑥∈ [0.1] 𝑓 𝑥 (Since 𝑓) is bounded on [0,1] (7)
If 0 ≤ 𝑥 ≤ 1

1 1
𝑃𝑛 𝑥 − 𝑓 𝑥 = ∫−1 𝑓 𝑥 + 𝑡 𝑄𝑛 𝑡 𝑑𝑡 − 𝑓 𝑥 ∫−1 𝑄𝑛 𝑡 𝑑𝑡 (by (2) and (5))

22
 Space for Hints

1 1
≤ 𝑓 𝑥 + 𝑡 − 𝑓 𝑥 𝑄𝑛 𝑡 𝑑𝑡 ≤ | 𝑓 𝑥 + 𝑡 − 𝑓 𝑥 𝑄𝑛 𝑡 𝑑𝑡|
−1 −1
−𝛿 𝛿
≤ 𝑓 𝑥 + 𝑡 − 𝑓 𝑥 𝑄𝑛 𝑡 𝑑𝑡 + 𝑓 𝑥 + 𝑡 − 𝑓 𝑥 𝑄𝑛 𝑡 𝑑𝑡
−1 −𝛿
1
+ 𝑓 𝑥 + 𝑡 − 𝑓 𝑥 𝑄𝑛 𝑡 𝑑𝑡
𝛿
−𝛿
≤ | 𝑓 𝑥 + 𝑡 + 𝑓 𝑥 | 𝑄𝑛 𝑡 𝑑𝑡
−1
𝛿 −𝛿
+ | 𝑓 𝑥+𝑡 − 𝑓 𝑥 𝑀 + 𝑀 𝑄𝑛 (𝑡)𝑑𝑡 𝑄𝑛 (𝑡) 𝑑𝑡
−𝛿 −1
−𝛿 1
< ∫−1 𝑀 + 𝑀 𝑄𝑛 (𝑡) 𝑑𝑡 + ∫𝛿 𝑀 + 𝑀 𝑄𝑛 (𝑡) 𝑑𝑡(by(7))
−𝛿
2
𝜀 𝛿
≤ 2𝑀 √𝑛 1 − 𝛿 𝑑𝑡 + 𝑄 𝑡 𝑑𝑡
1 2 −𝛿 𝑛
1
+ 2𝑀 √𝑛 (1 − 𝛿 2 ) 𝑑𝑡 (𝑏𝑦(4))
𝛿
𝜀
< 2𝑀 𝑛 1 − 𝛿 2 + 2 1 + 2𝑀√𝑛 1 − 𝛿 2 𝑛 1 − 𝛿 (by(2))
𝑛

𝜀
< 4𝑀√𝑛 1 − 𝛿 2 𝑛 + . .
𝜀 𝜀
2
< 2 + 2 , for large enough 𝑛
=𝜀
Therefore lim𝑛→∞ 𝑃𝑛 (𝑥) = 𝑓(𝑥) uniformly on[𝑎, 𝑏].

Corollary 2.2:

For every interval [– 𝑎, 𝑎] there is a sequence of real polynomials 𝑃𝑛 such

that 𝑃𝑛 (0) = 0 and such thatlim𝑛 →∞ 𝑃𝑛 (𝑥) = |𝑥| uniformly on – 𝑎, 𝑎 .

Proof:
By the above theorem, there exists a sequence {𝑃𝑛∗ } of real polynomials which
converges uniformly to |𝑥| on [– 𝑎, 𝑎]. In particular, 𝑃𝑛∗ (0) → 0 as 𝑛 → ∞.
The polynomials 𝑃𝑛 (𝑥) = 𝑃𝑛∗ (𝑥) − 𝑃𝑛∗ (0) for 𝑛 = 1,2,3,. .have the desired
properties.

Defintion 2.3:
Afamily𝐴 of complexfunctions defined on a set 𝐸 is said to be analgebra if
(i) 𝑓 + 𝑔 ∈ 𝐴 (𝑖𝑖)𝑓𝑔 ∈ 𝐴 (𝑖𝑖𝑖) 𝑐𝑓 ∈ 𝐴
𝐴 for all 𝑓 ∈ 𝐴, 𝑔 ∈ 𝐴 and for allcomplex constants 𝑐, that is, if 𝐴 is closed
under addition, multiplication, and scalar multiplication. For the algebra of
real functions , we have to consider (iii) for all real 𝑐.
23
Space for Hints

If 𝐴 has the property that 𝑓 ∈ 𝐴 whenever 𝑓𝑛 ∈ 𝐴 ( 𝑛 = 1,2,3, . . . . ) and

𝑓𝑛 → 𝑓 uniformly on 𝐸, then 𝐴 is said to be uniformly closed.

Let ℬ be the set of all functions which are limits of uniformly convergent
sequence of members of 𝐴 . Then ℬ is called the uniform closure of 𝐴.

Theorem 2.4:
Let ℬ be the uniform closure of an algebra 𝐴 of boundedfunctions. Then ℬ is a
uniformly closed algebra.
Proof:
If 𝑓 ∈ ℬ and 𝑔 ∈ ℬ,there exist uniformly convergent sequences{𝑓𝑛 }, {𝑔𝑛 }such
that 𝑓𝑛 → 𝑓 , 𝑔𝑛 → 𝑔 and 𝑓𝑛 ∈ ℬ , 𝑔𝑛 ∈ ℬ . Since the functions are bounded,
we have 𝑓𝑛 + 𝑔𝑛 → 𝑓 + 𝑔, 𝑓𝑛 𝑔𝑛 → 𝑓𝑔, 𝑐𝑓𝑛 → 𝑐𝑓, where 𝑐 is anyconstant,
the convergence is uniform in each case.
Hence 𝑓 + 𝑔 ∈ ℬ , 𝑓𝑔 ∈ ℬ and 𝑐𝑓 ∈ ℬ

Therefore ℬ is an algebra.Since𝐵 is the closure of 𝐴 , ℬ is closed (uniformly).

Definition 2.5:
Let 𝐴 be a family of functions on a set 𝐸. Then A is said to separate points on
𝐸 if to everypair of distinct point 𝑥1 , 𝑥2 ∈ 𝐸 therecorresponds a function
𝑓 ∈ 𝐴 , such that 𝑓(𝑥1 ) ≠ 𝑓(𝑥2 ).

If to each 𝑥 ∈ 𝐸 there corresponds a function 𝑔 ∈ 𝐴 , such that 𝑔(𝑥) ≠ 0, we

say that A Vanishes at no point of 𝐸.

Theorem 2.6:

Suppose A , is an algebra of functions on a set ,A , separates points on 𝐸, and

A ,vanishes at no point of 𝐸. Suppose 𝑥1 , 𝑥2 are distinct points of 𝐸, and 𝑐1 , 𝑐2
are constants (real if A, is a real algebra). Then A , contains afunction 𝑓 such
that 𝑓(𝑥1 ) = 𝑐1 , 𝑓(𝑥2 ) = 𝑐2 .

Proof:

Since 𝐴 , separates points on 𝐸 and 𝐴, vanishes at no point of 𝐸, we have

𝑔(𝑥1 ) ≠ 𝑔(𝑥2 ), 𝑕(𝑥1 ) ≠ 0, 𝑘(𝑥2 ) ≠ 0, where 𝑔, 𝑕, 𝑘 ∈ 𝐴 ,.
Put 𝑢 = 𝑔𝑘 – 𝑔(𝑥1)𝑘, 𝑣 = 𝑔𝑕 – 𝑔(𝑥2)𝑕.

Since 𝑔, 𝑕, 𝑘 ∈ 𝐴 , and A ,is an algebra, 𝑢, 𝑣 ∈ 𝐴 ,.

Also 𝑢(𝑥1 ) = 𝑔(𝑥1 )𝑘 – 𝑔(𝑥1 )𝑘 = 0 and 𝑣(𝑥2 ) = 𝑔(𝑥2 )𝑕 – 𝑔(𝑥2 )𝑕 = 0,
𝑢(𝑥2 ) = 𝑔(𝑥2 )𝑘 – 𝑔(𝑥1 )𝑘 ≠ 0 & 𝑣(𝑥1 ) = 𝑔(𝑥1 )𝑕– 𝑔(𝑥2 )𝑕 ≠ 0.

24
 Space for Hints

𝑐 𝑣 𝑐 𝑢
Let 𝑓 = 𝑣 1𝑥 + 𝑢 2𝑥
1 2

𝑐1 𝑣 𝑥1 𝑐2 𝑢 𝑥1
𝑇𝑕𝑒𝑛 𝑓 𝑥1 = +
𝑣 𝑥1 𝑢 𝑥2
𝑐 𝑣 𝑥 𝑐 𝑢 𝑥
= 𝑐1 + 0 = 𝑐1 and𝑓 𝑥2 = 1𝑣 𝑥 2 + 2𝑢 𝑥 2
1 2

= 0 + 𝑐2 = 𝑐2

We now have all the material needed for Stone's generalization of the
Weierstrass theorem.

Theorem 2.7:
Let A be an algebra of real continuous functions on a compact set 𝐾. If A
separates points on 𝐾 and if A vanishes at no point of 𝐾, then the uniform
closure ℬ of A consists of allreal continuous functions on 𝐾.

We shall divide the proof into four steps.

STEP 1: If 𝑓 ∈ ℬ then |𝑓| ∈ ℬ

Proof

Let 𝑎 = 𝑠𝑢𝑝 |𝑓(𝑥)|, ( 𝑥 ∈ 𝐾 ). (1)

Let 𝜀 > 0 be given.

By the corollary to the Stone- Weierstrass theorem, there exist real numbers
𝑐1 , 𝑐2 , . . . . , 𝑐𝑛 such that
𝑛 𝑖
𝑖=1 𝑐𝑖 𝑦 − 𝑦 < 𝜀, 𝑦 ∈ [−𝑎, 𝑎]. (2)

Since ℬ is an algebra and 𝑓 ∈ ℬ ⇒ 𝑐𝑖 𝑓 𝑖 ℬ for 𝑖 = 1,2, . . . , 𝑛.

Hence thefunction𝑔 = 𝑛𝑖=1 𝑐𝑖 𝑓𝑖 is amember of ℬ.
By(1) and(2), we have |𝑔(𝑥) − |𝑓(𝑥)|| < 𝜀 (𝑥 ∈ 𝐾)
Since ℬ is uniformly closed and 𝑔(𝑥) ∈ ℬ , |𝑓| ∈ ℬ .

STEP 2: If 𝑓 ∈ ℬ and 𝑔 ∈ ℬ , then 𝑚𝑎𝑥(𝑓, 𝑔) and 𝑚𝑖𝑛(𝑓, 𝑔) ∈ ℬ .

By 𝑚𝑎𝑥(𝑓, 𝑔) we mean the function 𝑕 defined by
𝑓 𝑥 𝑖𝑓 𝑓(𝑥) ≥ 𝑔(𝑥), 𝑓 𝑥 𝑖𝑓 𝑓 𝑥 < 𝑔(𝑥),
𝑕(𝑥) = and 𝑚𝑖𝑛(𝑓, 𝑔) =
𝑔 𝑥 𝑖𝑓 𝑓(𝑥) < 𝑔(𝑥), 𝑔 𝑥 𝑖𝑓 𝑓 𝑥 ≤ 𝑔(𝑥),

Proof:

𝑓+𝑔 |𝑓−𝑔|
Considertheidentities𝑚𝑎𝑥 𝑓, 𝑔 = +
2 2
25
Space for Hints

𝑓+𝑔 |𝑓−𝑔|
and 𝑚𝑖𝑛(𝑓, 𝑔) = 2 + 2
Since ℬ is an algebra and 𝑓 ∈ ℬ and 𝑔 ∈ ℬ , we have
𝑓 + 𝑔, 𝑓 − 𝑔 ∈ 𝐵. Also 𝑓 + 𝑔, |𝑓 − 𝑔| ∈ 𝐵
Therefore 𝑚𝑎𝑥(𝑓, 𝑔) and 𝑚𝑖𝑛(𝑓, 𝑔) ∈ ℬ

By iteration, the result can be extended to any finite set of functions, That is
if 𝑓1 , 𝑓2 , . . . , 𝑓𝑛 ∈ ℬthen𝑚𝑎𝑥(𝑓1 , 𝑓2 , . . . , 𝑓𝑛 ) ∈ ℬ and𝑚𝑖𝑛(𝑓1 , 𝑓2 , . . . , 𝑓𝑛 ) ∈ ℬ

STEP 3: Given a real function 𝑓, continuous on 𝐾, a point 𝑥 ∈ 𝐾, and𝜀 > 0,

there exists a function 𝑔𝑥 ∈ ℬ such that 𝑔𝑥 (𝑥) = 𝑓(𝑥) and 𝑔𝑥 𝑡 >
𝑓 𝑡 – 𝜀. (𝑡 ∈ 𝐾)

Proof:
By hypothesis ℬ ⊂ 𝐴 , and A satisfies the hypothesis of theorem 2.6, ℬ
also satisfies the hypothesis of theorem 2.6.
Hence, for every 𝑦 ∈ 𝐾, we can find a function 𝑕𝑦 ∈ ℬ such that
𝑕𝑦 𝑥 = 𝑓 𝑥 , 𝑕𝑦 = 𝑓(𝑦) (refer theorem 2.6) (3)

By the continuity of 𝑕𝑦 there exists an open set 𝐽𝑦 , containing 𝑦, such that

𝑕𝑦 (𝑡) > 𝑓(𝑡) − 𝜀. (𝑡 ∈ 𝐽𝑦 ) (4)
Since 𝐾 is compact, there is finite set of points 𝑦1 , 𝑦2 , . . . , 𝑦𝑛 such that
𝐾 ⊂ 𝐽𝑦1 ∪ 𝐽𝑦2 ∪ . . .∪ 𝐽𝑦𝑛 . (5)
Put 𝑔𝑥 = 𝑚𝑎𝑥(𝑕𝑦1 , 𝑕𝑦2 , . . . , 𝑕𝑦𝑛 ). By step 2, 𝑔 ∈ ℬ.
By (3), (4) and (5), we have 𝑔𝑥 (𝑥) = 𝑓(𝑥) and 𝑔𝑥 (𝑡) > 𝑓(𝑡) – 𝜀.

STEP 4: Given a real function 𝑓, continuous on 𝐾, and 𝜀 > 0, there exists a

function 𝑕 ∈ ℬ such that |𝑕(𝑥)– 𝑓(𝑥)| < 𝜀. (𝑥 ∈ 𝐾)
Proof:
Let us consider the functions 𝑔𝑥 , for eaph 𝑥 ∈ 𝐾, constructed in Step 3.
By the continuity of 𝑔𝑥 , there exist open sets 𝑉𝑥 containing 𝑥, such that
𝑔𝑥 (𝑡) < 𝑓(𝑡) + 𝜀. (𝑡 ∈ 𝑉𝑥 ). (6)
Since 𝐾 is compact, there exists a finite set of points 𝑥1 , 𝑥2 , . , 𝑥𝑚 such
That𝐾 ⊂ 𝑉𝑥1 ∪ 𝑉𝑥2 ∪ . . .∪ 𝑉𝑥𝑚 . (7)

By Step 2, 𝑕 ∈ ℬ.
By Step 3, 𝑕(𝑡) > 𝑓(𝑡) – 𝜀. (𝑡 ∈ 𝐾).
(6) and (7) implies that 𝑕(𝑡) < 𝑓(𝑡) + 𝜀. (𝑡 ∈ 𝐾).
That is 𝑓(𝑡) – 𝜀 < 𝑕(𝑡) < 𝑓(𝑡) + 𝜀 (𝑡 ∈ 𝐾).
That is– 𝜀 < 𝑕(𝑡) – 𝑓(𝑡) < 𝜀 (𝑡 ∈ 𝐾).
That is |𝑕(𝑡) – 𝑓(𝑡)| < 𝜀 (𝑡 ∈ 𝐾).

26
 Space for Hints

Theorem 2.8:

Suppose A is a self-ad joint algebra of complex continuous functions on a

compact set 𝐾,A separates points on 𝐾 and A vanishes at no point of𝐾. Then
the uniform closure 𝐵 of A consists of all complex continuous functions on 𝐾.
In other words A is dense in 𝐶(𝑋).
Proof:
Let 𝐴𝑟 be the set of all real functions on 𝐾 which belong to A If 𝑓 ∈ 𝐴 and
𝑓 = 𝑢 + 𝑖𝑣, with 𝑢, 𝑣 real, then 2𝑢 = 𝑓 + 𝑓 , and since 𝐴 is a self-ad joint,
we see that 𝑢 ∈ 𝐴𝑟 .
If 𝑥1 ≠ 𝑥2 , there exists 𝑓 ∈ 𝐴 such that 𝑓(𝑥1 ) = 1, 𝑓(𝑥2 ) = 0.
Hence 0 = 𝑢(𝑥2 ) ≠ 𝑢(𝑥1 ) = 1. Therefore 𝐴𝑟 separates points on 𝐾.
If 𝑥 ∈ 𝐾, then 𝑔(𝑥) ≠ 0 for some 𝑔 ∈ 𝐴 , and there is a complex number
𝜆 such that 𝜆𝑔(𝑥) > 0.
If 𝑓 = 𝜆𝑔, 𝑓 = 𝑢 + 𝑖𝑣, it follows that 𝑢(𝑥) > 0.

Hence 𝐴 𝑅 vanishes at no point of 𝐾.

Therefore 𝐴 𝑅 satisfies the hypothesis of theorem 2.7.
It follows that every real continuous function on 𝐾 lies in the uniform closure
of 𝐴𝑅 , and hence lies in ℬ . If 𝑓 is a complex continuous function on 𝐾, 𝑓 =
𝑢 + 𝑖𝑣, then 𝑢 ∈ ℬ , 𝑣 ∈ ℬ .
Hence 𝑓 ∈ ℬ .

CYP QUESTIONS:
1. Prove that the set 𝐶 (𝑋) of all complex valued, continuous, bounded
functions with domain 𝑋, with 𝑑(𝑓, 𝑔) = ||𝑓 – 𝑔|| is a metric space.
2. Distinguish between uniformly convergent and point wise convergent.

2.2 Power series

∞ 𝑛
The form of the power series is 𝑓(𝑥) = 𝑛=0 𝑐𝑛 𝑥

∞ 𝑛
or more generally𝑓(𝑥) = 𝑛=0 𝑐𝑛 𝑥−𝑎

These are called analytic functions. We shall discuss the power series only
for real values of 𝑥. Instead of circles of convergence we shall encounter
intervals of convergence.
∞ 𝑛
The series 𝑛=0 𝑐𝑛 𝑥 convergesfor all 𝑥 in (– ℝ , ℝ), for some ℝ.

If the series ∞ 𝑛
𝑛 =0 𝑐𝑛 𝑥 − 𝑎 converges for |𝑥 – 𝑎 | < 𝑅, then 𝑓 is said
to be expanded in a power series about the point 𝑥 = 𝑎.

27
Space for Hints

Theorem 2.9:
∞ 𝑛
Supposetheseries 𝑛=0 𝑐𝑛 𝑥 converges for |𝑥| < 𝑅, and define
∞ 𝑛 ∞ 𝑛
𝑓(𝑥) = 𝑛 =0 𝑐𝑛 𝑥 (|𝑥| < 𝑅) . Then theseries 𝑛=0 𝑐𝑛 𝑥

converges uniformlyon [– 𝑅 + 𝜀 , 𝑅 − 𝜀], no matter which 𝜀 > 0 is chosen.

The function 𝑓 is continuous and differentiable in (– 𝑅, 𝑅), and
∞

𝑓 ′(𝑥) = 𝑛𝑐𝑛 𝑥 𝑛 −1 (|𝑥| < 𝑅)

𝑛 =0

Proof:

Let 𝜀 > 0 be given.

𝑛
For |𝑥| ≤ 𝑅 – 𝜀, we have |𝑐𝑛 𝑥 𝑛 | ≤ |𝑐𝑛 𝑅 – 𝜀 |.
∞
Since 𝑛=0 𝑐𝑛 (𝑅 − 𝜖)𝑛 convergesabsolutely (Since every power series

converges absolutely in the interior of its interval of convergence , by the root

test)
∞ 𝑛
Therefore the series 𝑛=0 𝑐𝑛 𝑥 convergesuniformly on [ – 𝑅 + 𝜀 , 𝑅 – 𝜀].(by
theorem 2.2.3)
𝑛
Since lim𝑛 → ∞ √𝑛 = 1
𝑛 𝑛
wehavelimsup𝑛 → ∞ 𝑛|𝐶𝑛 | = limsup𝑛 → ∞ |𝐶𝑛 |

Therefore the series 𝑓(𝑥) = ∞ 𝑛

𝑛 =0 𝑐𝑛 𝑥 and the series 𝑓′ 𝑥 =
∞ 𝑛−1
𝑛=0 𝑛𝑐𝑛 𝑥 havethe same interval of convergence.

Since 𝑓′ 𝑥 = ∞ 𝑛 =1 𝑛𝑐𝑛 𝑥
𝑛 −1
is a powerseries, it converges uniformly in
[– 𝑅 + 𝜀 , 𝑅 – 𝜀], for every 𝜀 > 0 we can apply theorem 2.5.1( for series

instead of sequence), (That is)

∞ 𝑛 −1
𝑓′ 𝑥 = 𝑛=1 𝑛𝑐𝑛 𝑥 holds for |𝑥| < 𝑅holds if |𝑥| ≤ 𝑅 – 𝜀.

But, given any 𝑥 such that |𝑥| < 𝑅, we can find an 𝜀 > 0 such that |𝑥| < 𝑅
∞ 𝑛 −1
– 𝜀.This shows that 𝑓′ 𝑥 = 𝑛=1 𝑛𝑐𝑛 𝑥 holds for |𝑥| < 𝑅.
28
 Space for Hints

Corollary 2.10:

Under the hypothesis of the above theorem, 𝑓 has derivatives of all orders in
(– 𝑅 , 𝑅), which are given by

𝑓 𝑘 (𝑥) = 𝑛 𝑛 − 1 ⋯ (𝑛 − 𝑘 + 1)𝑐𝑛 𝑥 𝑛−𝑘

𝑛=𝑘

In particular, 𝑓 𝑘 (0) = 𝑘! 𝑐𝑘 , (𝑘 = 0,1,2, . ).

Proof:
∞ 𝑛
By the above theorem from 𝑓(𝑥) = 𝑛=0 𝑐𝑛 𝑥 we get
∞

𝑓′ 𝑥 = 𝑛𝑐𝑛 𝑥 𝑛 −1
𝑛 =1

Apply theorem 2.9, to f ‗, we get Successively apply theorem 2.9 to ”, 𝑓 3 ,....,

we get
∞

𝑓 𝑘 (𝑥) = 𝑛 𝑛 − 1 ⋯ (𝑛 − 𝑘 + 1)𝑐𝑛 𝑥 𝑛−𝑘

𝑛=𝑘

Putting x = 0 in (1), we get

𝑓 𝑘 (0) = 𝑘! 𝑐𝑘 , (𝑘 = 0,1,2, . ).

Abel's Theorem

Theorem 2.11:
∞ 𝑛
Suppose𝛴𝑐𝑛 converges. Put𝑓(𝑥) = 𝑛=1 𝑐𝑛 𝑥 ( − 1 < 𝑥 < 𝑅)
∞
Then lim𝑥→1 𝑓 𝑥 = 𝑛=1 𝑐𝑛

Proof:

29
Space for Hints

Let 𝑠𝑛 = 𝑐0 + 𝑐1 + . . . . + 𝑐𝑛 , 𝑠– 1 = 0.
∞ 𝑛 ∞
Then 𝑛 =0 𝑐𝑛 𝑥 = 𝑛=0 𝑠𝑛 − 𝑠𝑛−1 𝑥 𝑛

= 𝑠0 − 𝑠−1 𝑥 0 + 𝑠1 − 𝑠0 𝑥1 + 𝑠2 − 𝑠1 𝑥 2 + . . . + 𝑠𝑚 − 𝑠𝑚 −1 𝑥 𝑚

= 𝑠01 + 𝑠1 𝑥1 − 𝑠0 𝑥1 + 𝑠2 𝑥 2 − 𝑠1 𝑥 2 − 𝑠𝑚 −1 𝑥 𝑚 −1 + 𝑠𝑚 𝑥 𝑚

= 1 − 𝑥 𝑠0 + 1 − 𝑥 𝑠1 𝑥1 + 1 − 𝑥 𝑠2 𝑥 2 1 − 𝑥 𝑠𝑚 −1 𝑥 𝑚 −1
+ 𝑠𝑚 𝑥 𝑚
𝑚 −1

= 1−𝑥 𝑠𝑛 𝑥 𝑛 + 𝑠𝑚 𝑥 𝑚
𝑛=0

∞ 𝑛 𝑚 𝑛
For 𝑥 < 1, let 𝑚 → ∞, we get 𝑛=0 𝑐𝑛 𝑥 = (1 − 𝑥) 𝑛=0 𝑠𝑛 𝑥

∞ 𝑛
That is 𝑓(𝑥) = (1 − 𝑥 𝑛=0 𝑠𝑛 𝑥

Since 𝛴𝑐𝑛 converges to s (say).

Then its partial sum sequence {𝑠𝑛 } converges to s.

That islim𝑛→ ∞ 𝑠𝑛 = 𝑠.

Let 𝜀 > 0 be given.

Choose 𝑁 so that 𝑛 > 𝑁 ⇒ |𝑠 − 𝑠𝑛 | < 𝜀 . (1)

∞ 𝑛
By the geometric series test,we have 𝑛=0 𝑥 = 1 if |𝑥| < 1.
∞ 𝑛
That is (1 − 𝑥) 𝑛 =0 𝑥 = 1if |𝑥| < 1. (2).
∞ 𝑛
Now to prove thatlim𝑥→1 𝑓(𝑥) = 𝑛=0 𝑐 = 𝑠
∞

|𝑓 𝑥 − 𝑠| ≤ 1 − 𝑥 |𝑠𝑛 − 𝑠||𝑥 𝑛 | ( ∵ |𝑥 + 𝑦| ≤ |𝑥| + |𝑦|)

𝑛=0

∞ ∞
𝑛 𝑛
≤ 1 − 𝑥 𝑠𝑛 − 𝑠 𝑥 + 1 − 𝑥 𝑠𝑛 − 𝑠 𝑥
𝑛 𝑛=𝑁+1

30
 Space for Hints

∞ ∞
𝑛
𝜀 𝑛
≤ 1 − 𝑥 𝑠𝑛 − 𝑠 𝑥 + 1 − 𝑥 𝑥 (𝑏𝑦(1))
2
𝑛=0 𝑛 =𝑁+1

𝑁 𝑛
< 1 − 𝑥 𝑛=0 𝑠𝑛 − 𝑠 𝑥 + 𝜀 . (3)

Choose 𝛿 > 0 such that 1 − 𝑥 < 𝛿.Then (3) becomes

𝜀 𝜀
𝑓 𝑥 − 𝑠 < + = 𝜀
2 2

That is|𝑓(𝑥) − 𝑠| < 𝜀 if 𝑥 > 1 − 𝛿.

∞
That islim𝑥→1 𝑓(𝑥) = 𝑠 = 𝑛=0 𝑐𝑛

Note:

𝐼𝑓 𝛴 𝑎𝑛 , 𝛴 𝑏𝑛 , 𝛴𝑐𝑛 , converge to 𝐴, 𝐵, 𝐶, and if 𝑐𝑛 = 𝑎0 𝑏𝑛 + 𝑎1 𝑏𝑛–1

+. . . . + 𝑎𝑛 𝑏0 , then 𝐶 = 𝐴𝐵.

Theorem 2.12:

Given a double sequence 𝑎𝑖𝑗 , 𝑖 = 1,2,3, . . . , 𝑗 = 1,2,3, . . .,

suppose that ∞ 𝑗 =1 |𝑎𝑖𝑗 | = 𝑏𝑖 (𝑖 = 1,2,3, . . . . ) and 𝑏𝑖 converges.Then

∞ ∞ ∞ ∞
𝑖=1 𝑗 =1 𝑎𝑖𝑗 = 𝑗 =1 𝑖=1 𝑎𝑖𝑗 .

Let 𝐸 be a countable set, consisting of the points 𝑥0 , 𝑥1 , 𝑥2 , . . . , 𝑥𝑛 , and suppose

𝑥𝑛 → 𝑥0 as → ∞. Define

Proof:
∞
Let 𝑗 =1 |𝑎𝑖𝑗 | = 𝑏𝑖 (𝑖 = 1,2,3, . . . . ) (1)
∞
𝑓𝑖 (𝑥0 ) = 𝑗 =1 𝑎𝑖𝑗 (𝑖 = 1,2,3, . . . ) (2)
∞
𝑓𝑖 (𝑥𝑛 ) = 𝑗 =1 𝑎𝑖𝑗 (𝑖, 𝑛 = 1,2,3, . . . ) (3)
∞
𝑔(𝑥) = 𝑖=1 𝑎𝑖𝑗 𝑓𝑖 (𝑥) ( 𝑥∈𝐸) (4)

31
Space for Hints

Now, (2) and (3), together with (1), we getn

𝑛

𝑓𝑖 (𝑛) = 𝑎𝑖𝑗
𝑗 =1

∞
𝑗 =1 𝑎𝑖𝑗 = 𝑓𝑖 (𝑥0 ) as 𝑛 → ∞

That is𝑥𝑛 → 𝑥0 ⇒ 𝑓𝑖 (𝑥𝑛 ) → 𝑓𝑖 (𝑥0 ) 𝑎𝑠 𝑛 → ∞.

This shows that each fi is continuous at 𝑥0 .

∞
Now |𝑓𝑖 (𝑥)| = 𝑗 =1 𝑎𝑖𝑗 = 𝑏𝑖 for 𝑥 ∈ 𝐸 & 𝑖 = 1,2,3, . . . .
∞
Therefore 𝑔(𝑥) = 𝑖 𝑓𝑖 (𝑥) converges uniformly.

By theorem 2.3.1, 𝑔 is continuous at 𝑥0 .

∞ ∞ ∞
Therefore 𝑖=1 𝑗 =1 𝑎𝑖𝑗 = 𝑖=1 𝑓𝑖 (𝑥0 ) (by(2))

= 𝑔(𝑥0 ) (by(4))

= lim 𝑔(𝑥𝑛 )
𝑛→ ∞

∞
lim 𝑖=1 𝑓𝑖 (𝑥𝑛 )(by(4))
𝑛→ ∞

∞ ∞

= lim 𝑎𝑖𝑗 (𝑏𝑦(3))

𝑛→ ∞
𝑗 =1 𝑖=1

∞ ∞ ∞ ∞
Therefore 𝑖=1 𝑗 =1 𝑎𝑖𝑗 = 𝑗 =1 𝑖=1 𝑎𝑖𝑗

Taylor's theorem

Theorem 2.13:

32
 Space for Hints

Suppose 𝑓(𝑥) = ∞ 𝑛
𝑛=0 𝑐𝑛 𝑥 , the seriesconverging in |𝑥| < 𝑅. If – 𝑅 <
𝑎 < 𝑅, then 𝑓 can be expanded in a power series about the point 𝑥 = 𝑎
whichconverges in | 𝑥 – 𝑎 | < 𝑅 – |𝑎|, and
∞
𝑓𝑛 𝑎
𝑓(𝑥) = 𝑥 − 𝑎 𝑛 (|𝑥 − 𝑎| < 𝑅 − |𝑎|)
𝑛!
𝑛 =0

Proof:
∞ 𝑛
We have 𝑓(𝑥) = 𝑛=0 𝑐𝑛 𝑥

∞
= 𝑛=0 𝑐𝑛 (x − a + a)𝑛

∞ 𝑛 n
= 𝑛=0 𝑐𝑛 (Σ𝑚=0 an−m (x − a)𝑚
m

Therefore 𝑓(𝑥) can be extended in the form of power series about the point
𝑥=a.

∞ 𝑛 𝑛 n ∞
But 𝑛 =0 𝑚 =0 |𝑐𝑛 (Σ𝑚=0 an−m x − a 𝑚
|= 𝑛=0 𝑐𝑛 𝑥−𝑎 + 𝑎 𝑛
m

which converges if |𝑥 – 𝑎| + |𝑎| < 𝑅.That is if |𝑥 – 𝑎| < 𝑅 – |𝑎|

We know that, by corollary to the theorem 2.9,

𝑓 𝑘 (0) = 𝑘! 𝑐𝑘 . (𝑘 = 0,1,2, … ).

𝑓𝑘 0
That is𝑐𝑘 = (𝑘 = 0,1,2, . . . )
𝑘!

𝑛
∞ 𝑓 𝑎 𝑛
Therefore𝑓 𝑥 = 𝑛=0 𝑛! 𝑥−𝑎 𝑥−𝑎 < 𝑅− 𝑎 .

Theorem 2.14:

33
Space for Hints

Suppose the series 𝛴𝑎𝑛 𝑥 𝑛 and 𝛴 𝑏𝑛 𝑥 𝑛 converge in the segment 𝑆 = (– 𝑅, 𝑅).

Let 𝐸 be the set of all 𝑥 ∈ 𝑆 at which
∞ ∞
𝑛
𝑎𝑛 𝑥 = 𝑏𝑛 𝑥 𝑛
𝑛=0 𝑛=0

If 𝐸 has a limit point in 𝑆, then 𝑎𝑛 = 𝑏𝑛 for 𝑛 = 0,1,2,3, . . ..

∞ 𝑛 ∞
Hence 𝑛=0 𝑎𝑛 𝑥 = 0 𝑏𝑛 𝑥 𝑛 holds for all𝑥 ∈ 𝑆.

Proof:

Put𝑐𝑛 = 𝑎𝑛 – 𝑏𝑛 and
∞

𝑓 𝑥 = 𝑐𝑛 𝑥 𝑛 (𝑥 ∈𝑆)
𝑛 =0

∞
= 𝑛=0 𝑎𝑛 − 𝑏𝑛 𝑥 𝑛 = 0 for 𝑥 ∈ 𝑆. (by hypothesis)

Let 𝐴 be the set of all limit points of 𝐸 in 𝑆. Let 𝐵 be the set of all other points
of 𝑆.

By the definition of limit point, 𝐵 is open.

Let 𝑥0 ∈ 𝐴.

By theabovetheorem,
∞
𝑛
𝑓(𝑥) = 𝑑𝑛 𝑥 − 𝑥0 (|𝑥 − 𝑥0 | < 𝑅 − |𝑥0 |)
𝑛 =0

Now to prove that 𝑑𝑛 = 0 for all n.

Suppose there exist a smallest non- negative integer 𝑘 such that 𝑑𝑘 ≠ 0.

𝑓(𝑥) = 𝑥 − 𝑥0 𝑘 𝑔(𝑥)(|𝑥 − 𝑥0 | < 𝑅 − |𝑥0 |),

∞ 𝑚
where 𝑔 𝑥 = 𝑚 =0 𝑑𝑛+𝑘 𝑥 − 𝑥0

34
 Space for Hints

Since 𝑔 is continuous at 𝑥0 , and 𝑔(𝑥0 ) = 𝑑𝑘 ≠ 0, there exists a 𝛿 > 0 such

that 𝑔(𝑥0 ) = 𝑑𝑘 ≠ 0 if |𝑥 – 𝑥0 | < 𝛿.
𝑘
Therefore 𝑓(𝑥) = 𝑥 – 𝑥0 𝑔(𝑥) ≠ 0 if 0 < |𝑥 – 𝑥0 | < 𝛿, which is a

contradiction to the fact that 𝑥0 is a limit point of 𝐸.

Therefore 𝑑𝑛 = 0 for all 𝑛.

∞ 𝑛
So 𝑓(𝑥) = 𝑛=0 𝑑𝑛 𝑥 − 𝑥0 = 0 if |𝑥 − 𝑥0 | < 𝑅 − |𝑥0 |

That is in a neighborhood of 𝑥0 .

Therefore 𝐴 is open.That is𝐴 and 𝐵 are disjoint open sets. That is𝐴 and 𝐵 are
separated.Since 𝑆 = 𝐴 ∪ 𝐵 and 𝑆 is connected,

one of 𝐴 and 𝐵 must be empty. By hypothesis 𝐴 is not empty.Hence 𝐵 is

empty and 𝐴 = 𝑆.

Since 𝑓 is continuous in 𝑆, 𝐴 ⊂ 𝐸. Thus 𝐸 = 𝑆 and

𝑓𝑛 0
𝑐𝑛 = = 0 (𝑛 = 0,1,2, … )
𝑛!
Therefore𝑎𝑛 = 𝑏𝑛 for 𝑛 = 0,1,2,3, . ..
∞ 𝑛 ∞ 𝑛
Therefore 𝑛=0 𝑎𝑛 𝑥 = 𝑛=0 𝑏𝑛 𝑥 holds for all 𝑥 ∈ 𝑆

3.2 The exponential and logarithmic functions

Definition 2.15:
𝑛
∞ 𝑧
The exponential function, 𝐸(𝑧) = 𝑛=0 𝑛!

Note 1:Bytheratio test,𝐸(𝑧) converges for every complex 𝑧.

Note 2: (Addition formula)

∞ 𝑛
𝑧 𝑘 𝑤 𝑛 −𝑘
𝐸(𝑧)𝐸(𝑤) =
𝑘! (𝑛 − 𝑘)!
𝑛=0 𝑘=0

35
Space for Hints

∞ 𝑛 ∞
1 𝑛 𝑘 𝑛 −𝑘 (𝑧 + 𝑤)𝑛
= 𝑧 𝑤 =
𝑛! 𝑘 𝑛!
𝑛=0 𝑘=0 𝑛 =0

= 𝐸(𝑧 + 𝑤), where 𝑧 and 𝑤 are complex numbers.

Note 3:𝐸(𝑧)𝐸(– 𝑧) = 𝐸(𝑧 – 𝑧) = 𝐸(0) = 1, where,𝑧 is a complex number.

Therefore𝐸(𝑧) ≠ 0 for all 𝑧.

By the definition, 𝐸(𝑥) > 0 if 𝑥 > 0 and 𝐸(𝑥) > 0 for all real 𝑥.

Again by the definition, 𝐸(𝑥) → +∞ as 𝑥 → +∞ and 𝐸(𝑥) → 0 as 𝑥 → – ∞

along the real axis.

Also 0 < 𝑥 < 𝑦 implies 𝐸(𝑥) < 𝐸(𝑦) and 𝐸(– 𝑦) < 𝐸(– 𝑥).

Hence 𝐸 is strictly increasing on the whole real axis.

𝐸 𝑧+𝑕 −𝐸 𝑧 𝐸 𝑕 −1
lim = 𝐸(𝑧) lim = 𝐸(𝑧).1 = 𝐸(𝑧)
𝑕=0 𝑕 𝑕=0 𝑕
By iteration of the addition formula gives

𝐸(𝑧1 + 𝑧2 + . . . + 𝑧𝑛 ) = 𝐸(𝑧1 )𝐸(𝑧2 ). . . 𝐸(𝑧𝑛 ).

If 𝑧1 = 𝑧2 = . . . = 𝑧𝑛 = 1,then 𝐸(𝑛) = 𝑒 𝑛 , 𝑛 = 1,2,3, . . . (∵ 𝐸(1) = 𝑒)

𝑛
If𝑝 = 𝑚 , where𝑛 and𝑚are positive integers, then

𝑚
𝐸 𝑝 = 𝐸(𝑚𝑝) = 𝐸(𝑛) = 𝑒 𝑛 , so that

𝐸(𝑝) = 𝑒 𝑝 and 𝐸(– 𝑝) = 𝑒 –𝑝 (𝑝 > 0, 𝑝 rational).

Theorem 2.16:
𝑛
∞𝑥
Let 𝑒𝑥 be definedon𝑒 𝑥 = 𝐸 𝑥 = 0 𝑛!

a. 𝑒𝑥 is continuous and differentiable for all 𝑥;

𝑥 𝑥
b. (𝑒 ) ’= 𝑒 ;
c. 𝑒 𝑥 is a strictly increasing function of 𝑥, and 𝑒 𝑥 > 0;
d. 𝑒 𝑥+𝑦 = 𝑒 𝑥 𝑒 𝑦 ;
e. 𝑒 𝑥 → +∞ as 𝑥 → +∞ and 𝑒 𝑥 → 0 as 𝑥 → – ∞;
36
 Space for Hints

f. lim𝑥→ + ∞ 𝑥 𝑛 𝑒 −𝑥 = 0 for every 𝑛.

Proof:

For the proof of (a) to (e), refer Note 1,2,3.

𝑛 𝑥2 𝑥𝑛 𝑥 𝑛 +1
∞𝑥 𝑥
f. By definition 𝑒 𝑥 = 0 𝑛! = 1 + 1! + + . . . . + 𝑛! + + . . . ..
2! 𝑛 +1 !

𝑥 𝑛 +1
> !for x>0
𝑛+1

𝑥𝑛 𝑥
That is𝑒 𝑥 > 𝑛+1 !

𝑛+1 !
⇒ 𝑒 −𝑥 <
𝑥𝑛 𝑥
𝑛+1 !
⇒ 𝑒 −𝑥 𝑥 𝑛 < → 0 𝑎𝑠 𝑥 → +∞;
𝑥

That islim𝑥→ +∞ 𝑥 𝑛 𝑒 −𝑥 = 0 for every 𝑛.

Note:

Since E is strictly increasing and differentiable on ℝ1, it has an inverse

function 𝐿, which is also strictly increasing and differentiable and whose
domain is 𝐸(ℝ1 ), that is, the set of all positive numbers, 𝐿 is defined by
𝐸(𝐿(𝑦)) = 𝑦 (𝑦 > 0) or 𝐿(𝐸(𝑥)) = 𝑥 (𝑥 real).

CYP Questions:

1. Find the following limits.

1
𝑒− 1 + 𝑥 𝑥
i. 𝑙𝑖𝑚𝑥→0 𝑥
ii. 𝑙𝑖𝑚𝑛 → ∞ 𝑛/ log 𝑛 [𝑛1/𝑛 − 1]

2.4 The trigonometric functions

37
Space for Hints

Letusdefine
1 1
𝐶 𝑥 = 2 [𝐸(𝑖𝑥) + 𝐸(−𝑖𝑥)] and 𝑆 𝑥 = 2𝑖 [𝐸(𝑖𝑥) − 𝐸(−𝑖𝑥)]

Since𝐸(𝑧) = 𝐸(𝑧), 𝐶(𝑥) and 𝑆(𝑥) arefor real 𝑥.

Also 𝐸 𝑖𝑥 = 𝐶 𝑥 + 𝑖𝑆 𝑋 .That is𝐶(𝑥) and 𝑆(𝑥) are real and imaginary
parts of 𝐸 𝑖𝑥 , respectively, if 𝑥 is real.

2
𝐸 𝑖𝑥 = 𝐸(𝑖𝑥)𝐸(𝑖𝑥) = 𝐸(𝑖𝑥)𝐸(−𝑖𝑥) = 1, so that

|𝐸(𝑖𝑥)| = 1(x real), so that |𝐸(𝑖𝑥)| = 1 (𝑥 real).

Also 𝐶(0) = 1 and 𝑆(0) = 0 , and 𝐶′(𝑥) = – 𝑆(𝑥), 𝑆′(𝑥) = 𝐶(𝑥).

Now to show that there exist a positive real number 𝑥 such that 𝐶(𝑥) = 0.
Suppose 𝐶(𝑥) ≠ 0 for all 𝑥.Since 𝐶(0) = 1, we have 𝐶(𝑥) > 0 for all
𝑥 > 0.

Since 𝑆(0) = 0 and 𝑆′(𝑥) = 𝐶(𝑥), we have 𝑆′(𝑥) > 0 for all 𝑥 > 0.

Therefore𝑆(𝑥) is strictly increasing function. If 0 < 𝑥 < 𝑦 then 𝑆(𝑥) <

𝑆(𝑦).
𝑦 𝑦 ′ 𝑦
That is𝑆(𝑥)(𝑦 − 𝑥) < ∫𝑥 𝑆 𝑡 𝑑𝑡 < ∫𝑥 − 𝐶 𝑡 𝑑𝑡 = − 𝐶 𝑡 𝑥
=
− [𝐶(𝑦) − 𝐶(𝑥)] = [𝐶(𝑥) – 𝐶(𝑦)]
2
Since 1 = 𝐸 𝑖𝑥 = 𝐶 2 (𝑥) + 𝑆 2 (𝑥), 𝐶 2 (𝑥) ≤ 1.

∴ 𝐶(𝑥) 𝑎𝑛𝑑 – 𝐶(𝑥) ≤ 1.

∴ [𝐶(𝑥) – 𝐶(𝑦)] ≤ 2.

That is𝑆(𝑥) (𝑦 – 𝑥) ≤ 2.

Since 𝑆(𝑥) > 0 this inequality cannot be true for large values of 𝑦 , we get a
contradiction.

∴ ∃ a positive number 𝑥 such that 𝐶(𝑥) = 0.

Let 𝑥0 be the smallest positive integer such that

𝐶(𝑥0 ) = 0.
38
 Space for Hints

Define the number 𝜋 by 𝜋 = 2𝑥0 . Then 𝐶(𝜋/2) = 𝐶(𝑥0 ) = 0

Since 𝐸(𝑖𝑥) = 𝐶(𝑥) + 𝑖𝑆(𝑋) & |𝐸(𝑖𝑥)| = 1, we have 𝑆(𝜋/2) = ±1.

Since 𝐶(𝑥) > 0 in (0, 𝜋/2), 𝑆 is an increasing function in (0, 𝜋/2).

Therefore 𝑆(𝜋/2) = 1. Thus (𝜋𝑖 ) = 𝑖

𝐸(𝜋𝑖) = 𝐸(𝜋𝑖 + 𝜋𝑖 ) = 𝐸(𝜋𝑖 ). 𝐸(𝜋𝑖 ) = 𝑖. 𝑖 = −1

𝐸(2𝜋𝑖) = 𝐸(𝜋𝑖 + 𝜋𝑖) = 𝐸(𝜋𝑖). 𝐸(𝜋𝑖) = ( − 1)( − 1) = 1.

Also 𝐸(𝑧 + 2𝜋𝑖) = 𝐸(𝑧). 𝐸(2𝜋𝑖) = 𝐸(𝑧)

Theorem 2.17:

a. The function 𝐸 is periodic, with period 2𝜋𝑖,

b. The functions 𝐶 and 𝑆 are periodic, with period
2𝜋,
c. If 0 < 𝑡 < 2𝜋, then 𝐸(𝑖𝑡) ≠ 1,
d. If 𝑧 is a complex number with |𝑧| = 1, there is a unique 𝑡 in [0, 2𝜋)
such that 𝐸(𝑖𝑡) = 𝑧.

Proof:

a. Since 𝐸(𝑧 + 2𝜋𝑖) = 𝐸(𝑧), 𝐸is periodic, with period 2𝜋𝑖.

1
b. 𝐶 𝑥 + 2𝜋 = 2 𝐸 𝑖 𝑥 + 2𝜋 + 𝐸 −𝑖 𝑥 + 2𝜋
1
= [𝐸(𝑖𝑥 + 2𝜋𝑖) + 𝐸(−𝑖𝑥 − 𝑖2𝜋))]
2
1
= [𝐸(𝑖𝑥) + 𝐸(−𝑖𝑥)](𝑆𝑖𝑛𝑐𝑒 𝐸(2𝜋𝑖) = 1)
2
= 𝐶(𝑥)

Therefore 𝐶(𝑥) is periodic with period 2𝜋.

Similarly𝑆(𝑥) isperiodicwith period 2𝜋.

39
Space for Hints

c. Suppose 0 < 𝑡 < 𝜋/2 and 𝐸(𝑖𝑡) = 𝑥 + 𝑖𝑦, where 𝑥 & 𝑦 are real.

∴ 0 < 𝑥 < 1 & 0 < 𝑦 < 1.

2
1 = 𝐸 𝑖𝑡 = 𝑥2 + 𝑦2 .
4 4
𝐸(4𝑖𝑡) = 𝐸 𝑖𝑡 = 𝑥 + 𝑖𝑦 = 𝑥4

– 62 𝑦 2 + 𝑦 4 + 4𝑖𝑥𝑦(𝑥 2 – 𝑦 2 )

If 𝐸(4𝑖𝑡) is real, then 𝑥 2 – 𝑦 2 = 0, that is 𝑥 2 = 𝑦 2 .

Since 𝑥 2 + 𝑦 2 = 1,

2𝑥 2 = 1 ⇒ 𝑥 2 = ½ and 𝑦 2 = ½.
2 2 2 2
1 1 1 1
∴ 𝐸(4𝑖𝑡) = –6 + + 4𝑖𝑥𝑦(0)
2 2 2 2

= 1/4 – 6/4 + 1/4 = – 4/4 =– 1.

∴ 𝐸(𝑖𝑡) ≠ 1.

d. Choose 𝑧 so that |𝑧| = 1.

Write 𝑧 = 𝑥 + 𝑖𝑦, with 𝑥 and 𝑦 real.Suppose first that 𝑥 ≥ 0 and 𝑦 ≥ 0.

On [0, 𝜋/2], 𝐶 decreases from 1 to 0.Hence 𝐶(𝑡) = 𝑥 for some 𝑡 ∈ [0, 𝜋/2].

Since 𝐶 2 + 𝑆 2 = 1 and 𝑆 ≥ 0 on [0, 𝜋/2], it follows that 𝑧 = 𝐸(𝑖𝑡). If

𝑥 < 0 and 𝑦 ≥ 0, the precedingconditions are satisfied by – 𝑖𝑧 .Hence
– 𝑖𝑧 = 𝐸(𝑖𝑡) for some 𝑡 ∈ [0, 𝜋/2], and since 𝑖 = 𝐸(𝑖𝜋/2), we obtain
𝑧 = 𝐸(𝑖(𝑡 + 𝜋/2)).Finally if 𝑦 < 0, the preceding two cases show that
– 𝑧 = 𝐸(𝑖𝑡)for some 𝑡 ∈ (0, 𝜋). Hence 𝑧 =– 𝐸(𝑖𝑡) = 𝐸(𝑖(𝑡 + 𝜋)).

–1
Suppose 0 ≤ 𝑡1 < 𝑡2 < 2𝜋, 𝐸(𝑖𝑡2 )[𝐸 𝑖𝑡1 = 𝐸(𝑖(𝑡2 – 𝑡1 )) ≠ 1( by (c))

Therefore there is a unique 𝑡 in [0,2𝜋) such that 𝐸(𝑖𝑡) = 𝑧.

CYP Questions:
𝜋 2 𝑠𝑖𝑛𝑥
1. If 0 < 𝑥 < 2 , prove that 𝑛 < < 1.
𝑥

40
 Space for Hints

2.5 Fourier series

Definition 2.18:

A trigonometric polynomial is a finite sum of the form

𝑓 𝑥 = 𝑎0 + 𝑁 𝑛 =1 𝑎𝑛 𝑐𝑜𝑠 𝑛𝑥 + 𝑏𝑛 𝑠𝑖𝑛 𝑛𝑥 𝑥 𝑟𝑒𝑎𝑙 where

𝑎0 , 𝑎1 , . . . , 𝑎𝑁 , 𝑏0 , 𝑏1 , . . . , 𝑏𝑁 are complex numbers,Since

1 1
𝐶 𝑥 = 𝐸 𝑖𝑥 + 𝐸 −𝑖𝑥 𝑎𝑛𝑑 𝑆 𝑥 = [𝐸(𝑖𝑥) − 𝐸(−𝑖𝑥)], 𝑓(𝑥)
2 2𝑖
can also be written in the form
𝑁

𝑓(𝑥) = 𝑐𝑛 𝑒 𝑖𝑛𝑥 (𝑥 𝑟𝑒𝑎𝑙)

−𝑁

Note 1: Every trigonometric polynomial is periodic with period 2𝜋.

Note 2: If 𝑛 is a nonzero integer,𝑒 𝑖𝑛𝑥 is the derivative of

𝑒 𝑖𝑛𝑥
,which is also has period 2𝜋. Hence
𝑖𝑛

𝜋
1 1 𝑖𝑓 𝑛 = 0
𝑒 𝑖𝑛𝑥 𝑑𝑥 =
2𝜋 −𝜋
0 𝑖𝑓 𝑛 = ±1, ±2, . . .

If we multiply 𝑓(𝑥) = 𝑁−𝑁 𝑐𝑛 𝑒

𝑖𝑛𝑥
by 𝑒 𝑖𝑚𝑥 ,where 𝑚 is an integer, if we
integratethe product,weget
1 𝜋
𝑐𝑚 = 2𝜋 ∫−𝜋 𝑓 𝑥 𝑒 −𝑖𝑚𝑥 𝑑𝑥 for |𝑚| ≤ 𝑁.

If |m| > N, the above integral is zero.

𝑁 𝑖𝑛𝑥
Note 3: The trigonometric polynomial𝑓(𝑥) = −𝑁 𝑐𝑛 𝑒 is real iff
𝑐−𝑛 = 𝑐𝑛 for 𝑛 = 0,1, . . . , 𝑁.

Note 4: We define a trigonometric series to be a series of the form

∞ 𝑖𝑛𝑥
−∞ 𝑐𝑛 𝑒 (x real) the 𝑛𝑡𝑕 partial sumof this series is defined to be
41
Space for Hints

𝑐𝑛 𝑒 𝑖𝑛𝑥
−𝑁

Note 5:

If f is an integrable function on [–π,π], the numbers cm for all integers m are

called the Fouriercoefficients of 𝑓, and the series
∞ 𝑖𝑛𝑥
−∞ 𝑐𝑛 𝑒 formed withthesecoefficientsiscalledtheFourierseries of 𝑓.

Definition 2.19:

Let {𝜙𝑛 } (𝑛 = 1,2,3, . . . ) be asequence ofcomplexfunctions on

[𝑎, 𝑏],suchthat
∞
𝜙𝑛 x ϕm (x) dx = 0 (n ≠ m).
∞

Then {𝜙𝑛} issaidtobean orthogonalsystemoffunctionson[𝑎, 𝑏]. If, in

𝑏 2
addition∫𝑎 𝜙𝑛 𝑥 𝑑𝑥 = 1, for all 𝑛, { 𝜙𝑛} is said to be orthonormal.

Example 2.20:
1
−
The functions 2𝜋 2 𝑒 𝑖𝑛𝑥 formanorthonormal system on [– 𝜋, 𝜋].

𝑏
Note: If {𝜙𝑛 } is orthonormal on [𝑎, 𝑏]andif𝑐𝑛 = ∫𝑎 𝑓(𝑡)𝜙𝑛 (𝑡) 𝑑𝑡 𝑛 =
1,2, . .. we call 𝑐𝑛 the 𝑛𝑡𝑕Fouriercoefficientsof𝑓 relative to {𝜙𝑛 }.
∞
We write 𝑓(𝑥) ∼ 1 𝑐𝑛 𝜙𝑛 (𝑥) and call thisseries the Fourier series of 𝑓.

Theorem2.21:

42
 Space for Hints

𝑛
Let{ 𝜙𝑛 } be orthonormal on [𝑎, 𝑏].Let 𝑠𝑛 (𝑥) = 𝑚 =1 𝑐𝑚 𝜙𝑚 (𝑥) be the
𝑛𝑡𝑕partial sum of the Fourier series of𝑓,
andsuppose𝑡𝑛 (𝑥) = 𝑛𝑚 =1 𝛾𝑚 𝜙𝑚 𝑥 .
𝑏 2 𝑏 2
Then∫𝑎 𝑓 − 𝑠𝑛 𝑑𝑥 ≤ ∫𝑎 𝑓 − 𝑡𝑛 𝑑𝑥 and

Equalityholds iff𝛾𝑚 = 𝑐𝑚 , (𝑚 = 1,2, . . . , 𝑛).

Proof:

Let ∫ denote the integral over [𝑎, 𝑏], 𝛴 the sum from 1 to 𝑛.

Then ∫ 𝑓𝑡𝑛 = ∫ 𝑓 𝛾𝑚 𝜙𝑚 = 𝛾𝑚 ∫ 𝑓𝜙𝑚 = 𝛾𝑚 𝑐𝑚

(by the definition of 𝑐𝑚 ).

𝑛 𝑛
2
𝑁𝑜𝑤 ∫ 𝑡𝑛 = ∫ 𝑡𝑛 𝑡𝑛 = 𝛾𝑚 𝜙𝑚 𝛾𝑚 𝜙𝑚
𝑚 =1 𝑚 =1
𝑛 𝑛
= ∫ 𝑚 =1 𝛾𝑚 𝜙𝑚 𝑚 =1 𝛾𝑚 𝜙𝑚 = 𝛴 𝛾𝑚 2 (since {𝜙𝑛} is orthonormal)
𝑏 𝑏
2
𝑓 − 𝑡𝑛 𝑑𝑥 = (𝑓 − 𝑡𝑛 )(𝑓 − 𝑡𝑛 ) 𝑑𝑥
𝑎 𝑎
𝑏 𝑏 𝑏 𝑏
= 𝑓𝑓 𝑑𝑥 − 𝑓𝑡𝑛 𝑑𝑥 − 𝑡𝑛 𝑓 𝑑𝑥 + 𝑡𝑛 𝑡𝑛 𝑑𝑥
𝑎 𝑎 𝑎 𝑎
𝑏
=∫𝑎 𝑓 2 𝑑𝑥 − 𝑐𝑚 2
+ Σ 𝛾𝑚 − 𝑐𝑚 2
(1)

which is evidently minimized iff 𝛾𝑚 − 𝑐𝑚

𝑏 𝑏
2 2
∴ 𝑓 − 𝑠𝑛 𝑑𝑥 ≤ 𝑓 − 𝑡𝑛 𝑑𝑥
𝑎 𝑎

𝑏 2 𝑏
Put in 𝛾𝑚 = 𝑐𝑚 (1), we get ∫𝑎 𝑓 − 𝑡𝑛 𝑑𝑥 = ∫𝑎 𝑓 2 𝑑𝑥 − 𝑐𝑚 2

𝑏 2 𝑏
Since ∫𝑎 𝑓 − 𝑡𝑛 𝑑𝑥 ≥ 0, ∫𝑎 𝑓 2 𝑑𝑥 − 𝑐𝑚 2
≥ 0

𝑏
∴ 𝑓 2 𝑑𝑥 ≥ 𝑐𝑚 2
𝑎

43
Space for Hints

𝑛
2 2 2
∫ 𝑠𝑛 𝑥 𝑑𝑥 = 𝑐𝑚 ≤ ∫ 𝑓 𝑥 𝑑𝑥
𝑚 =1

Theorem 2.22:

(Bessel's inequality) If { 𝜙𝑛 } is orthonormal on[𝑎, 𝑏] and if 𝑓 𝑥 ∼

∞ ∞ 2
𝑛=1 𝑐𝑛 𝜙 𝑥 , then 𝑛=1 𝑐𝑛 ≤ ∫ 𝑓 𝑥 2 𝑑𝑥

In particular lim𝑛 →∞ 𝑐𝑛 = 0.

Proof:

From the above theorem, we have

𝑛
2 2 2
∫ 𝑠𝑛 𝑥 𝑑𝑥 = 𝑐𝑚 ≤ ∫ 𝑓 𝑥 𝑑𝑥
𝑚 =1

Letting 𝑛 → ∞, we get
𝑛
2 2
𝑐𝑚 ≤ ∫ 𝑓 𝑥 𝑑𝑥
𝑚 =1

Also lim𝑛→∞ 𝑐𝑛 = 0.

Trigonometric series:

We shall consider functions 𝑓 that have period 2𝜋 and that are Riemann-
integrable on [ – 𝜋, 𝜋]. The Fourier series of 𝑓is given by

∞ 𝑖𝑛𝑥 1 𝜋
𝑓(𝑥) = −∞ 𝑐𝑛 𝑒 where 𝑐𝑛 = 2𝜋 ∫−𝜋 𝑓(𝑥) 𝑒 −𝑖𝑛𝑥 𝑑𝑥

, the 𝑁𝑡𝑕 partial sum of the Fourier series of 𝑓 is given by

𝑁

𝑆𝑁 (𝑥) = 𝑆𝑁 (𝑓; 𝑥) = 𝑐𝑛 𝑒 𝑖𝑛𝑥

−𝑁

44
 Space for Hints

𝜋 𝑁 𝜋
1 2 2
1 2
∴ 𝑆𝑁 𝑥 𝑑𝑥 = 𝑐𝑛 ≤ 𝑓 𝑥 𝑑𝑥
2𝜋 𝜋 2𝜋 −𝜋
𝑛=−𝑁

The Dirichlet kernel 𝐷𝑁 (𝑥) is defined by

1
𝑠𝑖𝑛 𝑁 + 𝑥
𝑖𝑛𝑥 2
𝐷𝑁 𝑋 = 𝑒 == 𝑥
𝑛=−𝑁
sin⁡
(2 )

Since
1 𝜋 𝑁
𝑐𝑛 = 2𝜋 ∫−𝜋 𝑓 𝑥 𝑒 −𝑖𝑛𝑥 𝑑𝑥, 𝑆𝑁 𝑓; 𝑥 = −𝑁 𝑐𝑛 𝑒
𝑖𝑛𝑥
=
𝑁 1 𝜋
−𝑁 2𝜋 ∫−𝜋 𝑓 𝑡 𝑒 −𝑖𝑛𝑡 𝑑𝑡 𝑒 𝑖𝑛𝑥

𝑁 𝜋
1
= 𝑓 𝑡 𝑒 𝑖𝑛 (𝑥−𝑡) 𝑑𝑡
2𝜋 −𝜋
−𝑁

𝜋 𝑁
1
𝑓 𝑡 𝑒 𝑖𝑛 (𝑥−𝑡) 𝑑𝑡
2𝜋 𝜋 −𝑁

𝜋
1
= 𝑓(𝑡) 𝐷𝑁 (𝑥 − 𝑡)𝑑𝑡
2𝜋 𝜋

𝜋
1
= 𝑓(𝑥 − 𝑡) 𝐷𝑁 (𝑡)𝑑𝑡
2𝜋 𝜋

Theorem 2.23:

If, for some 𝑥, there are constants 𝛿 > 0 and 𝑀 < ∞such that

|𝑓(𝑥 + 𝑡) − 𝑓(𝑥)| ≤ 𝑀|𝑡| for all 𝑡 ∈ (−𝛿, 𝛿), then lim𝑁 → ∞ 𝑆𝑁 (𝑓; 𝑥) =
𝑓(𝑥)

45
Space for Hints

Proof:
𝑓 𝑥−𝑡 − 𝑓 𝑥
Define 𝑔 𝑡 = 𝑡 for 0 < |𝑡| < 𝜋, and put 𝑔(0) = 0
𝑠𝑖𝑛
2

1 𝜋 1 𝜋 𝑁 𝑖𝑛𝑥
Now ∫ 𝐷 (𝑥) 𝑑𝑥 =
2𝜋 𝜋 𝑁
∫
2𝜋 𝜋 𝑛=−𝑁 𝑒 𝑑𝑥

(by the definition of Dirichlet kernel)

𝑁 𝜋 𝑁
1
= 𝑒 𝑖𝑛𝑥 𝑑𝑥 = 𝑒 𝑖𝑛𝑥
2𝜋 𝜋
−𝑁 𝑛=−𝑁

𝜋
1 1 𝑖𝑓 𝑛 = 0
(∵ 𝑒 𝑖𝑛𝑥 𝑑𝑥 = )
2𝜋 −𝜋
0 𝑖𝑓 𝑛 = ±1, ± 2, . . .

𝜋
1
∴ 𝑆𝑁 𝑓; 𝑥 − 𝑓 𝑥 = 𝑓 𝑥 − 𝑡 𝐷𝑁 𝑡 𝑑𝑡 − 𝑓 𝑥
2𝜋 −𝜋
𝜋
1
= 𝑓 𝑥 − 𝑡 𝐷𝑁 𝑡 𝑑𝑡 − 𝑓 𝑥 . 1
2𝜋 −𝜋

𝜋 𝜋
1 1
= 𝑓 𝑥 − 𝑡 𝐷𝑁 𝑡 𝑑𝑡 − 𝑓 𝑥 . 𝐷𝑁 𝑡 𝑑𝑡
2𝜋 −𝜋 2𝜋 −𝜋

𝜋
1
= (𝑓 𝑥 − 𝑡 − 𝑓 𝑥 ) 𝐷𝑁 𝑡 𝑑𝑡
2𝜋 −𝜋

𝜋
1 𝑡
= 𝑔 𝑡 𝑠𝑖𝑛 𝐷 (𝑡) 𝑑𝑡
2𝜋 −𝜋 2 𝑁
1
1 𝜋
𝑡 𝑠𝑖𝑛 𝑁 + 2 𝑡 1 𝜋
1
= 𝑔 𝑡 𝑠𝑖𝑛 𝑡
𝑑𝑡 = 𝑔 𝑡 𝑠𝑖𝑛 𝑁 + 𝑑𝑡
2𝜋 −𝜋 2 𝑠𝑖𝑛 2 2𝜋 −𝜋 2

𝜋
1 𝑡 𝑡
𝑔 𝑡 [𝑠𝑖𝑛 𝑁𝑡 𝑐𝑜𝑠 + 𝑐𝑜𝑠 𝑁𝑡 𝑠𝑖𝑛 ]𝑑𝑡
2𝜋 −𝜋 2 2
46
 Space for Hints

𝜋 𝜋
1 𝑡 1 𝑡
= [𝑔 𝑡 𝑐𝑜𝑠 ]𝑠𝑖𝑛 𝑁𝑡 𝑑𝑡 + [𝑔 𝑡 𝑠𝑖𝑛 ]𝑐𝑜𝑠 𝑁𝑡 𝑑𝑡
2𝜋 −𝜋 2 2𝜋 −𝜋 2

By |𝑓(𝑥 + 𝑡) – 𝑓(𝑥)| ≤ 𝑀|𝑡| for all 𝑡 ∈ (– 𝛿, 𝛿) and the definition of

𝑔(𝑡), 𝑔(𝑡)𝑐𝑜𝑠(𝑡/2) and 𝑔(𝑡)𝑠𝑖𝑛(𝑡/2) arebounded.

The last two integrals tend to 0 as 𝑁 → ∞ (by theorem 2.22)

∴ lim 𝑆𝑁 (𝑓; 𝑥) = 𝑓(𝑥)

𝑁→∞

Theorem 2.24:

If 𝑓 is continuous (with period 2𝜋) and if 𝜀 > 0, then there is a trigonometric

polynomial P such that |𝑃(𝑥) – 𝑓(𝑥)| < 𝜀 for all real 𝑥.

Proof:

If we identify 𝑥 and 𝑥 + 2𝜋, we may regard the 2𝜋 −periodic functions on

𝑅1 as functions on the unit circle 𝑇,by means of the mapping 𝑥 → 𝑒 𝑖𝑥 . The

trigonometric polynomials , that is, the

𝑁 𝑖𝑛𝑥
functions of the form 𝑓(𝑥) = −𝑁 𝑐𝑛 𝑒 ,

form a self-adjoint algebra 𝐴 , which separates point on 𝑇, and which vanishes

at no point of 𝑇. Since 𝑇 is compact, 𝐴 is dense in 𝐴(𝑇).

2.6 The Gamma function

Definition 2.25
∞
For 0 < 𝑥 < ∞, 𝛤 𝑥 ∫0 𝑡 𝑥 −1 𝑒 −𝑡 𝑑𝑡The integral converges for 𝑥 ∈ 0, ∞ .

Theorem 2.26:

a. The functional equation 𝛤(𝑥 + 1) = 𝑥𝛤(𝑥) holds if 0 < 𝑥 < ∞

b. 𝛤(𝑛 + 1) = 𝑛! for 𝑛 = 1,2, . . ..

47
Space for Hints

c. 𝑙𝑜𝑔 𝛤 is convex on (0, ∞).

Proof:

a. Let 0 < 𝑥 < ∞.

∞
𝛤 𝑥 = 𝑡 𝑥 −1 𝑒 −𝑡 𝑑𝑡
0

∞ ∞
𝑥 −𝑡
𝛤 𝑥+1 = 𝑡 𝑒 𝑑𝑡 = 𝑡 𝑥 𝑑(𝑒 −𝑡 )
0 0

∞ ∞
∞
= − 𝑡 𝑥 𝑒 −𝑡 0 + 𝑥𝑡 𝑥 −1 𝑒 −𝑡 𝑑𝑡 = 0 + 𝑥 𝑡 𝑥 −1 𝑒 −𝑡 𝑑𝑡 = 𝑥𝛤(𝑥)
0 0

By 𝑎 𝛤(𝑛 + 1) = 𝑛𝛤(𝑛) = 𝑛(𝑛 − 1)𝛤(𝑛 − 1)

= 𝑛 𝑛 − 1 𝑛 − 2 ⋯ (1)𝛤(1)
∞ ∞ ∞
b. 𝛤 1 = ∫0 𝑡1−1 𝑒 −𝑡 𝑑𝑡 = ∫0 𝑡 0 𝑒 −𝑡 𝑑𝑡 = ∫0 𝑒 −𝑡 𝑑𝑡 = 1

∴ 𝛤(𝑛 + 1) = 𝑛(𝑛 + 1)(𝑛 − 2)1 = 𝑛!

1 1
Let 1 < 𝑝 < ∞ 𝑎𝑛𝑑 𝑝 + 𝑞 = 1.

𝑥 𝑦 𝑥 𝑦 1 1
𝑥 𝑦 ∞ + −1 ∞ + −( + ) 1 1
Now 𝛤(𝑝 + 𝑞 ) = ∫0 𝑡 𝑝 𝑞 𝑒 −𝑡 𝑑𝑡 = ∫0 𝑡 𝑝 𝑞 𝑝 𝑞 𝑒 −𝑡 (𝑝 + 𝑞 )𝑑𝑡

∞ 𝑥 1 𝑦 1 𝑡 𝑡 ∞ 𝑥 −1 𝑡 𝑦 −1 𝑡
− + − − − − −
= 𝑡 𝑝 𝑝 𝑞 𝑞 𝑒 𝑝 𝑞 𝑑𝑡 = 𝑡 𝑝 𝑒 𝑝 𝑡
𝑞 𝑒 𝑞 𝑑𝑡
0 0

1 1
∞ 𝑝 ∞ 𝑞
≤ 𝑡 𝑥−1 𝑒 −𝑡 𝑑𝑡 }. 𝑡 𝑦 −1 𝑒 −𝑡 𝑑𝑡
0 0
(by Holder's inequality)
1 1
= 𝛤 𝑥 𝑝 𝛤 𝑥 𝑞

c.

48
 Space for Hints

1 1
∞ ∞
𝑥 𝑦 𝑝 𝑞
𝛤( + ) ≤ 𝑡 𝑥 −1 𝑒 −𝑡 𝑑𝑡 }. 𝑡 𝑦 −1 𝑒 −𝑡 𝑑𝑡
𝑝 𝑞 0 0

Taking log on both sides,we get

1 1
∞ ∞
𝑥 𝑦 𝑝 𝑞
log 𝛤( + ) ≤ log 𝑡 𝑥−1 𝑒 −𝑡 𝑑𝑡 }. 𝑡 𝑦 −1 𝑒 −𝑡 𝑑𝑡
𝑝 𝑞 0 0

1 1
= log 𝛤 𝑥 𝑝 + log 𝛤 𝑦 𝑞

1 1
= 𝑙𝑜𝑔𝛤 𝑥 + 𝑙𝑜𝑔𝛤(𝑦)
𝑝 𝑞

1 1 1
𝑃𝑢𝑡 𝜆 = . 𝑇𝑕𝑒𝑛 1 − 𝜆 = 1 − =
𝑝 𝑝 𝑞

∴ 𝑙𝑜𝑔𝛤(𝜆𝑥 + (1 − 𝜆)𝑦) ≤ 𝜆 𝑙𝑜𝑔 𝛤(𝑥) + (1 − 𝜆)𝑙𝑜𝑔 𝛤(𝑦)

∴ 𝑙𝑜𝑔 𝛤 𝑖𝑠 𝑐𝑜𝑛𝑣𝑒𝑥 𝑜𝑛(0, ∞).

49
Space for Hints

UNIT- 3
FUNCTIONS OF SEVERAL VARIABLES

Unit Structure:
Section 3.1: Linear Transformation
Section 3.2: Differentiation
Section 3.3: The Contraction Principle

3.1 Linear transformations

We begin this unit with a discussion of sets of vectors in Euclidean n-space
𝑅 𝑛 . The algebraic facts presented here extend without change to finite-
dimensional vector spaces over any field of scalars. However, for our purposes
it is quite sufficient to stay within the familiar framework provided by the
Euclidean spaces.

Definitions 3.1
(a) A nonempty set X ⊂ 𝑅 𝑛 is a vector space if x + y ∈ X and cx ∈ X for all
x ∈X, y ∈ X, and for all scalars c.
(b) If 𝑥1 , … . 𝑥𝑘 ∈ 𝑅 𝑛 and 𝑐1 , … . . 𝑐𝑘 are scalars, the vector is called a linear
combination of𝑥1 , … . 𝑥𝑘 . If S ⊂ 𝑅 𝑛 and if E is the set of all linear
combinations of elements of S, we say that S spans E, or that E is the span of
S. Observe that every span is a vector space.
(c) A set consisting of vectors 𝑥1 , … . 𝑥𝑘 (we shall use the notation {𝑥1 , … . 𝑥𝑘 }
for such a set) is said to be independent if the relation 𝑐1 𝑥1 + ⋯ + 𝑐𝑘 𝑥𝑘 = 0
implies that 𝑐1 = ⋯ 𝑐𝑘 = 0.
Otherwise { 𝑥1 , … . 𝑥𝑘 } is said to be dependent. Observe that no independent
set contains the null vector.

50
 Space for Hints

(d) If a vector space X contains an independent set of r vectors but contains no

independent set of r + I vectors, we say that X has dimension r, and
write : dim X = r.
The set consisting of 0 alone is a vector space; its dimension is 0.
(e) An independent subset of a vector space X which spans X is called a basis
of X. Observe that if B = { 𝑥1 , … . 𝑥𝑟 } is a basis of X, then every x ∈ X has a
unique representation of the form 𝑥 = 𝑐𝑗 𝑥𝑗 . Such a representation exists
since B spans X. and it is unique since B is independent. The numbers𝑐1 , … 𝑐𝑟
are called the coordinates of x with respect to the basis B. The most familiar
example of a basis is the set {𝑒1 , … 𝑒𝑛 }, where 𝑒𝑗 is the vector in 𝑅 𝑛 whose jth
coordinate is 1 and whose other coordinates are all 0.
lf x ∈ 𝑅 𝑛 , x = ( 𝑥1 , … . 𝑥𝑛 ), then x = 𝑥𝑗 𝑒𝑗 .
We shall call {𝑒1 , … 𝑒𝑛 } the standard basis of 𝑅 𝑛 .

Theorem 3.2
Let r be a positive integer. If a vector space X is spanned by a set of r vectors,
then dim X≤ r.
Proof
If this is false. there is a vector space X which contains an independent set
Q = { 𝑦1 , … . 𝑦𝑟+1 } and which is spanned by a set 𝑆𝑜 consisting of r vectors.
Suppose 0 < i < r. and suppose a set 𝑆𝑖 has been constructed which spans X
and which consists of all 𝑦𝑖 with 1 < j < i plus a certain collection of r – i
members of 𝑆𝑜 , say 𝑦1 , … . 𝑦𝑟 . (In other words, 𝑆𝑖 is obtained from 𝑆𝑜 by
replacing i of its elements by members of Q, without altering the span.) Since
𝑆𝑖 spans X, 𝑦𝑖+1 is in the span of 𝑆𝑖 ; hence there are scalars 𝑎1 ,. . . 𝑎𝑖+1 ,
𝑏𝑖 ,…𝑏𝑟−𝑖 with 𝑎𝑖+1 = 1, such that,
𝑖+1 𝑟−𝑖

𝑎𝑗 𝑦𝑗 + 𝑏𝑘 𝑥𝑘 = 0
𝑗 =1 𝑘=1

51
Space for Hints

If all 𝑏𝑘 ′s were 0, the independence of Q would force all 𝑎𝑗 ′𝑠 to be 0, a

contradiction. It follows that some 𝑥𝑘 ∈ 𝑆𝑖 is a linear combination of the other
members of𝑇𝑖 = 𝑆𝑖 ∪ {𝑦𝑖+1 }. Remove this𝑥𝑘 from 𝑇𝑖 and call the remaining
set 𝑆𝑖+1 .Then 𝑆𝑖+1 spans the same set as 𝑇𝑖 namely X, so that 𝑆𝑖+1 has the
properties postulated for 𝑆𝑖 with i + 1 in place of i. Starting with 𝑆𝑜 , we thus
construct sets 𝑆1 , … . 𝑆𝑟 . The last of these consists of 𝑦1 , … . 𝑦𝑟 , and our
construction shows that it spans X. But Q is independent; hence 𝑦𝑟+1 not in
the span of 𝑆𝑟 . This contradiction establishes the theorem.

Corollary
dim Rn = n.
Proof
Since {𝑒1 , … 𝑒𝑛 } spans 𝑅 𝑛 , the theorem shows that dim 𝑅 𝑛 ≤ n. Since
{ 𝑒1 , … 𝑒𝑛 } is independent, dim 𝑅 𝑛 ≥ n.

Theorem 3.3
Suppose X is a vector space and dim X= n.
(a) A set E of n vectors in X spans X if and only if E is independent.
(b) X has a basis, and every basis consists of n vectors.
(c) If 1 ≤ r ≤ n and { 𝑦1 , … . 𝑦𝑟 } is an independent set in X, then X has a basis
containing { 𝑦1 , … . 𝑦𝑟 }.
Proof
Suppose E = { 𝑥1 , … . 𝑥𝑛 }. Since dim X= n, the set { 𝑥1 , … . 𝑥𝑛 . y} is
dependent, for every y ∈ X. If E is independent, it follows that y is in the span
of E; hence E spans X. Conversely, if E is dependent, one of its members can
be removed without changing the span of E. Hence E cannot span X, by
Theorem 3.2. This proves (a).

52
 Space for Hints

Since dim X= n, X contains an independent set of n vectors, and (a) shows that
every such set is a basis of X: (b) now follows from 3. 1 (d) and 3.2.
To prove (c). let { 𝑥1 , … . 𝑥𝑛 } be a basis of X. The set
S = { 𝑦1 , … . 𝑦𝑟 , 𝑥1 , … . 𝑥𝑛 }
spans X and is dependent, since it contains more than n vectors. The argument
used in the proof of Theorem 3.2 shows that one of the 𝑥𝑖 ′s is a linear
combination of the other members of S. If we remove this 𝑥𝑖 from S, the
remaining set still spans X. This process can be repeated r times and leads to a
basis of X which contains { 𝑦1 , … . 𝑦𝑟 }. by (a).

Definitions 3.4
A mapping A of a vector space X into a vector space Y is said to be a linear
transformation if
A(𝑥1 + 𝑥2 ) = A𝑥1 + 𝐴𝑥2 , A(cx) = cAx for all x, 𝑥1 , 𝑥2 ∈ X and all scalars c.
Note that one often writes Ax instead of A(x) if A is linear.
Observe that A0 = 0 if A is linear. Observe also that a linear transformation A
of X into Y is completely determined by its action on any basis: If { 𝑥1 , … . 𝑥𝑛 }
is a basis of X, then every x ∈ X has a unique representation of the form and
the linearity of A allows us to compute Ax from the vectors 𝐴𝑥1 , … . 𝐴𝑥𝑛 . and
the coordinates 𝑐1 , … . 𝑐𝑛 by the formula
𝑛
Ax = 𝑖=1 𝑐𝑗 𝐴𝑥𝑖

Linear transformations of X into X are often called linear operators on X. If A

is a linear operator on X which (i) is one-to-one and (ii) maps X onto X, we
say that A is invertible. In this case we can define an operator 𝐴−1 on X by
requiring that 𝐴−1 (Ax) = x for all x ∈ X. It is trivial to verify that we then also
have A(𝐴−1 x) = x, for all x ∈ X, and that 𝐴−1 is linear. An important fact
about linear operators on finite-dimensional vector spaces is that each of the
above conditions (i) and (ii) implies the other:

53
Space for Hints

Theorem 3.5
A linear operator A on a finite-dimensional vector space X is one-to-one if
and only if the range of A is all of X.
Proof
Let {𝑥1 , … . 𝑥𝑛 } be a basis of X. The linearity of A shows that its range ℜ(A) is
the span of the set Q = {𝐴𝑥1 , … . 𝐴𝑥𝑛 }· We therefore infer from Theorem
3.3(a) that ℜ (A) = X if and only if Q is independent. We have to prove that
this happens if and only if A is one-to-one.
Suppose A is one-to-one and 𝑐𝑖 𝐴𝑥𝑖 = 0. Then A( 𝑐𝑖 𝐴𝑥𝑖 ) = 0, hence
𝑐𝑖 𝑥𝑖 = 0, hence 𝑐1 = ⋯ = 𝑐𝑛 = 0, and we conclude that Q is independent.
Conversely, suppose Q is independent and A( 𝑐𝑖 𝑥𝑖 ) = 0. Then 𝑐𝑖 𝐴𝑥𝑖 = 0,
hence 𝑐1 = ⋯ = 𝑐𝑛 = 0, and we conclude: Ax = 0 only if x = 0. If now Ax =
Ay, then A(x - y) = Ax - Ay = 0, so that x - y = 0, and this says that A is one-
to-one.

Definitions 3.6
(a) Let L(X, Y) be the set of all linear transformations of the vector space X
into the vector space Y. Instead of L(X, X), we shall simply write L(X).
If 𝐴1 , 𝐴2 ∈ L(X, Y) and if 𝑐1 , 𝑐2 are scalars, define 𝑐1 𝐴1 + 𝑐2 𝐴2 by
(𝑐1 𝐴1 + 𝑐2 𝐴2 )x = 𝑐1 𝐴1 𝑥 + 𝑐2 𝐴2 𝑥 (x ∈ X). It is then clear that𝑐1 𝐴1 +
𝑐2 𝐴2 ∈ L(X, Y).
(b) If X, Y, Z are vector spaces, and if A ∈ L(X, Y) and B ∈ L(Y, Z), we
define their product BA to be the composition of A and B:
(BA)x = B(Ax) (x ∈ X).
Then BA ∈ L(X, Z).
Note that BA need not be the same as A B, even if X = Y = Z.
(c) For A ∈ L(𝑅 𝑛 , 𝑅 𝑚 ), define the norm ∥ 𝐴 ∥ of A to be the sup of all
numbers |Ax|, where x ranges over all vectors 𝑅 𝑛 in with |x| < 1. Observe that
the inequality
54
 Space for Hints

|Ax| ≤∥ 𝐴 ∥ |𝑥| holds for all x ∈ 𝑅 𝑛 .

Also, if 𝜆 is such that |Ax| ≤ 𝜆 |x| for all x ∈ 𝑅 𝑛 , then ∥ 𝐴 ∥≤ 𝜆.

Theorem 3.7
(a) If A ∈∥ 𝐴 ∥ L(𝑅 𝑛 ,𝑅 𝑚 ,), then ∥ 𝐴 ∥<∞ and A is a uniformly continuous
mapping of 𝑅 𝑛 into 𝑅 𝑚 . (b) If A, B ∈L(𝑅 𝑛 ,𝑅 𝑚 ) and c is a scalar, then
∥ 𝐴 + 𝐵 ∥ ≤ ∥ 𝐴 ∥ +∥ 𝐵 ∥, ∥ 𝑐𝐴 ∥ = |𝑐| ∥ 𝐴 ∥
With the distance between A and B defined as ∥ 𝐴 − 𝐵 ∥, L(𝑅 𝑛 ,𝑅 𝑚 ) is a
metric space.
(c) If A ∈ L(𝑅 𝑛 ,𝑅 𝑚 ), and B ∈ L(𝑅 𝑛 ,𝑅 𝑚 ), then ∥ 𝐵𝐴 ∥ ≤ ∥ 𝐵 ∥∥ 𝐴 ∥
Proof
(a) Let { 𝑒1 , … . 𝑒𝑛 } be the standard basis in 𝑅 𝑛 and suppose x = 𝑐𝑖 𝑒𝑖 , |x|
< 1, so that |𝑐𝑖 | < 1for i = 1, . . . n. Then,
|Ax | = | 𝑐𝑖 𝐴𝑒𝑖 | ≤ |𝑐𝑖 | |𝐴𝑒𝑖 | ≤ |𝐴𝑒𝑖 |
so that
𝑛

∥ 𝐴 ∥≤ |𝐴𝑒𝑖 | ≤ ∞.
𝑖=1

Since | Ax - Ay | <∥ 𝐴 ∥ | x - y | if x, y ∈ 𝑅 𝑛 , we see that A is uniformly

continuous.
(b) The inequality in (b) follows from |(A + B)x| = |Ax + Bx| ≤| Ax |+ |Bx|
≤ (∥A∥ +∥ 𝐵 ∥) |x |.
The second part of (b) is proved in the same manner. If
A, B, C ∈L(𝑅 𝑛 ,𝑅 𝑚 ),
we have the triangle inequality,
∥ 𝐴 − 𝐶 ∥= ∥ 𝐴 − 𝐵 + 𝐵 − 𝐶 ∥ ≤ ∥ 𝐴 − 𝐵 ∥ + ∥ 𝐴 − 𝐶 ∥, and it
is easily verified that ∥ 𝐴 − 𝐵 ∥ has the other properties of a metric (
Definition 2. I 5).
(c) Finally, (c) follows from

55
Space for Hints

|(BA )x| = |B(Ax)| ≤∥ 𝐵 ∥ |Ax| ≤ ∥ 𝐵 ∥ ∥ 𝐴 ∥ |x|

Since we now have metrics in the spaces L(𝑅 𝑛 ,𝑅 𝑚 ) the concepts of open set,
continuity, etc . . make sense for these spaces. Our next theorem utilizes these
concepts.

Theorem 3.8
Let Ω be the set of all invertible linear operators on 𝑅 𝑛 .
(a) If A𝜖 Ω, B 𝜖 L(𝑅 𝑛 ). and ∥ 𝐵 − 𝐴 ∥ ∥ 𝐴−1 ∥ < 1 ' then B 𝜖 Ω.
(b) Ω is a11 open subset of L(𝑅 𝑛 ), and the mapping A ⟶ 𝐴−1 is
continuous on Ω (Th is mapping is also obviously a 1- 1 mapping of Ω
onto Ω, which is its own inverse.)
Proof
1
(a) Put ∥ 𝐴−1 ∥ =𝑥 , put ∥ 𝐵 − 𝐴 ∥ = 𝛽. Then 𝛽<𝛼. For every, x 𝜖𝑅 𝑛 ,

𝛼|x| = |𝐴−1 𝐴𝑥| ≤ 𝛼 ∥ 𝐴−1 ∥ |𝐴𝑥 = 𝐴𝑥 ≤ 𝐴 − 𝐵 𝑥 + |𝐵𝑥| ≤ 𝛽|𝑥| +

|𝐵𝑥|,
so that
(1) 𝛼 − 𝛽 𝑥 ≤ 𝐵𝑥 (𝑥 𝜖 𝑅 𝑛 ).
Since 𝛼 − 𝛽> 0. (1) shows that Bx ≠ 0 if x ≠ 0. Hence B is 1 - 1. By Theorem
3.5, B𝜖 Ω .This holds for all B with ∥ 𝐵 − 𝐴 ∥< 𝛼. Thus, we have (a) and the
fact that Ω is open .
(b) Next replace x by 𝐵 −1 𝑦 in (1). The resulting inequality
𝛼 − 𝛽 𝐵 −1 𝑦 ≤ 𝐵𝐵 −1 𝑦 = 𝑦 (𝑦𝜖𝑅 𝑛 )
shows that ∥ 𝐵 −1 ∥≤ (𝛼 − 𝛽)−1
The identity𝐵 −1 − 𝐴−1 = 𝐵 −1 (𝐴 − 𝐵)𝐴−1 , combined with
Theorem 3. 7(c), implies therefore that,
𝛽
∥ 𝐵 −1 − 𝐴−1 ∥ ≤ ∥ 𝐵 −1 ∥∥ 𝐴 − 𝐵 ∥∥ 𝐴−1 ∥≤
𝛼(𝛼 − 𝛽)
This establishes the continuity assertion made in (b), since
𝛽 ⟶ 0 𝑎𝑠 𝐵 ⟶ 𝐴.
56
 Space for Hints

Definition(Matrices) 3.9
Suppose {𝑥1 , … . 𝑥𝑛 } and {𝑦1 , … . 𝑦𝑚 } are bases of vector spaces X and Y,
respectively. Then every A 𝜖 L( X, Y) determines a set of numbers 𝑎𝑖𝑗 such
that
𝑚
(3 ) 𝐴𝑥𝑗 = 𝑖=1 𝑎𝑖𝑗 𝑦𝑗 (1 < j < n). It is convenient to visualize these
numbers in a rectangular array of m rows and n columns, called an m by n
matrix :
𝑎11 ⋯ 𝑎1𝑛
𝐴 = ⋮ ⋱ ⋮
𝑎𝑚1 ⋯ 𝑎𝑚𝑛
Observe that the coordinates 𝑎𝑖𝑗 of the vector 𝐴𝑥𝑗 (with respect to the
basis{𝑦1 , … . 𝑦𝑚 }) appear in the jth column of [A ]. The vectors𝐴𝑥𝑗 are
therefore sometimes called the column vectors of [A ]. With the terminology
the range of A is spanned by the column vectors of [A ].
If x = 𝑐𝑗 𝑥𝑗 . the linearity of A, combined with (3). shows that
𝑚 𝑛
(4) Ax = 𝑖=1( 𝑗 =1 𝑎𝑖𝑗 𝑐𝑗 )𝑦𝑖

Thus the coordinates of Ax are 𝑗 𝑎𝑖𝑗 𝑐𝑗 .

Note that in (3) the summation ranges over the first subscript of 𝑎𝑖𝑗 .
But that we sum over the second subscript when computing coordinates.
Suppose next that an m by n matrix is given with real entries 𝑎𝑖𝑗 . If A is then
defined by (4), it is clear that A 𝜖 L( X, Y) and that [A ] is the given matrix.
Thus there is a natural 1-1 correspondence between L( X, Y) and the set of all
real m by n matrices. We emphasize though that [A ] depends not only on A
but also on the choice of bases in X and Y. The same A may give rise to many
different matrices if we change bases, and vice versa . We shall not pursue this
observation any further, since we shall usually work with fixed bases. (Some
remarks on this may be found in Sec. 3.37)

57
Space for Hints

If Z is a third vector space with basis {𝑧1 , 𝑧2 , … . . 𝑧𝑝 } if A is given by (3), and

if B𝑦𝑖 = 𝑘 𝑏𝑘𝑖 𝑧𝑘 , 𝐵𝐴 𝑥𝑗 = 𝑘 𝑐𝑘𝑗 𝑧𝑘

then A ∈ L(X, Y), B ∈ L( Y. Z), BA ∈ L( X, Z ), and since

B(𝐴𝑥𝑗 ) = B 𝑖 𝑎𝑖𝑗 𝑦𝑖 = 𝑖 𝑎𝑖𝑗 𝐵𝑦𝑖
= 𝑎𝑖𝑗 𝑘 𝑏𝑘𝑖 𝑧𝑘 the independence of {𝑧1 , 𝑧2 , … . . 𝑧𝑝
} implies that
(5) 𝑐𝑘𝑗 = 𝑖 𝑏𝑘𝑖 𝑎𝑖𝑗 (1 ≤ k ≤p, 1 ≤ j ≤ n).
This shows how to compute the p by n matrix [BA ] from [B] and [A]. If we
define the product [B] [A] to be [BA ], then (5) describes the usual rule of
matrix multiplication.
Finally, suppose {𝑥1 , … . 𝑥𝑛 } and {𝑦1 , … . 𝑦𝑚 }are standard bases of𝑅 𝑛 and𝑅 𝑚 ,
and A is given by (4). The Schwarz inequality shows that

|𝐴𝑥|2 = ( 𝑎𝑖𝑗 𝑐𝑗 )2 ≤ ( 𝑎𝑖𝑗 2 𝑐𝑗 2 ) = 𝑎𝑖𝑗 2 |𝑥|2

𝑖 𝑗 𝑖 𝑗 𝑗

Thus
1
(6) ∥ 𝐴 ∥≤ { 𝑖,𝑗 𝑎𝑖𝑗 2 } 2

If we apply (6) to B - A in place of A, where A, B ∈ L(𝑅 𝑛 ,𝑅 𝑚 ), we see that if

the matrix elements 𝑎𝑖𝑗 are continuous functions of a parameter, then the same
is true of A. More precisely :
If S is a metric space, if 𝑎11 , … . 𝑎𝑚𝑛 are real continuous functions on S, and if,
for each p ∈ S, 𝐴𝑝 is the linear transformation 𝑅 𝑛 into𝑅 𝑚 of whose matrix has
entries 𝑎𝑖𝑗 𝑝 , then the mapping p → 𝐴𝑝 is a continuous mapping of S into
L(𝑅 𝑛 ,𝑅 𝑚 ).

3.2 Differentiation
Definitions 3.10
In order to arrive at a definition of the derivative of a function whose domain
is 𝑅 𝑛 (or an open subset of 𝑅 𝑛 ), let us take another look at the familiar case n =
58
 Space for Hints

1, and let us see how to interpret the derivative in that case in a way which will
naturally extend to n > 1.
If f is a real function with domain (a, b) ⊂ 𝑅1 and if x ∈ (a, b), then 𝑓 ′ (x) is
usually defined to be the real number
𝑓 𝑥+𝑕 −𝑓(𝑥)
(7) lim𝑕→0 𝑕

provided, of course, that this limit exists.

Thus,
(8) f(x + h) -f(x) = f'(x)h + r(h)
where the "remainder" r(h) is small, in the sense that
𝑟(𝑕)
(9) lim𝑕→0 𝑕

Note that (8) expresses the difference f(x + h) - f(x) as the sum of the linear
function that takes h to f'(x)h. plus a small remainder. We can therefore regard
the derivative off at x, not as a real number, but as the linear operator on 𝑅1
that takes h to f'(x)h. [Observe that every real number 𝛼 gives rise to a linear
operator on 𝑅1 ; the operator in question is simply multiplication by 𝛼.
Conversely, every linear function that carries 𝑅1 to 𝑅1 is multiplication by
some real number. It is this natural 1-1 correspondence between 𝑅1 and L(𝑅1 )
which motivates the preceding statements .]
Let us next consider a function f that maps (a, b) ⊂ 𝑅1 into 𝑅 𝑚 . In that case, f
'(x) was defined to be that vector y ∈ 𝑅 𝑚 (if there is one) for which
𝑓 𝑥+𝑕 −𝑓(𝑥)
(10) lim𝑕→0 { − 𝑦} = 0
𝑕

We can again rewrite this in the form

( 11 ) f (x + h) - f(x) = hy + r(h), where r(h)/h →0 as h→ 0. The
main term on the right side of (11) is again a linear function of h. Every y
∈ 𝑅 𝑚 induces a linear transformation of 𝑅1 into 𝑅 𝑚 by associating to each
h ∈ 𝑅1 the vector h𝑦 ∈ 𝑅 𝑚 . This identification of 𝑅 𝑚 with L(𝑅1 , 𝑅 𝑚 ) allows
us to regard f '(x) as a member of L(𝑅1 , 𝑅 𝑚 ). Thus, if f is a differentiable
mapping of (a, b)⊂ 𝑅1 into 𝑅 𝑚 , and if x ∈ (a, b), then f '(x) is the linear
59
Space for Hints

transformation of 𝑅1 into 𝑅 𝑚 that satisfies (12)

𝑓 𝑥+𝑕 −𝑓 𝑥 −𝑓 ′ 𝑥 𝑕
lim𝑕→0 =0
𝑕

or, equivalently,
|𝑓 𝑥+𝑕 −𝑓 𝑥 −𝑓 ′ 𝑥 𝑕|
(13) lim𝑕→0 =0
|𝑕|

We are now ready for the case n > 1.

Definition 3.11
Suppose E is an open set in 𝑅 𝑛 , f maps E into 𝑅 𝑚 , and x ∈ E. If there exists a
linear transformation A of 𝑅 𝑛 into 𝑅 𝑚 such that
|𝑓 𝑥+𝑕 −𝑓 𝑥 −𝐴𝑕|
(14) lim𝑕→0 = 0,
𝑕

then we say that f is differentiable at x, and we write

(1 5) f '(x) = A .
If f is differentiable at every x ∈ E, we say that f is differentiable in E.
It is of course understood in (14) that h𝜖𝑅 𝑛 . If |h| is small enough, then x + h
∈ E, since E is open. Thus f (x + h) is defined, f(x + h) 𝜖𝑅 𝑚 , and since A ∈
L(𝑅 𝑛 , 𝑅 𝑚 ), Ah ∈ 𝑅 𝑚 . Thus
f(x + h) - f (x) – Ah ∈ 𝑅 𝑚
The norm in the numerator of (14) is that of𝑅 𝑚 . In the denominator we have
the 𝑅 𝑛 -norm of h. There is an obvious uniqueness problem which has to be
settled before we go any further.

Theorem 3.12
Suppose E and f are as in Definition 3. 1 1 , x ∈ E and ( 1 4) holds with A =𝐴1
and with A = 𝐴2 . Then 𝐴1 = 𝐴2 ( 1 6)
Proof
60
 Space for Hints

If B =𝐴1 - 𝐴2 , the inequality

|Bh| ≤ | f(x + h) - f(x) -𝐴1 h |+ |f(x + h) - f(x) -𝐴2 h| shows
that |Bh|/ | h| →0 as h →0. For fixed h ≠ 0, it follows that
|𝐵(𝑡𝑕)
→0 as t → 0. I th l The linearity of B shows that the left side of (16) is
|𝑡𝑕|

independent of t. Thus Bh = 0 for every h 𝜖𝑅 𝑛 . Hence B = 0.

Remarks 3.13
(a) The relation (14) can be rewritten in the form
(17) f( x + h) - f(x) = f'(x)h + r(h) where the remainder r(h)
satisfies
|𝑟 𝑕 |
lim =0
𝑕→0 |𝑕|

We may interpret (17) as in Sec. 3. 1 0, by saying that for fixed x and small h,
the left side of (17) is approximately equal to f '(x)h. that is to the value of a
linear transformation applied to h.
(b) Suppose f and E are as in definition 3. 11, and f is differentiable in E. For
every x ∈ E, f '(x) is then a function, namely, a linear transformation of𝑅 𝑛 into
Rm. But f' is also a function : f' maps E into L(𝑅 𝑛 , 𝑅 𝑚 ).
(c) A glance at (17) shows that f is continuous at any point at which f is
differentiable.
(d) The derivative defined by (14) or (17) is often called the differential of f at
x, or the total derivative of f at x, to distinguish it from the partial derivatives
that will occur later.

Example 3.14
We have defined derivatives of functions carrying 𝑅 𝑛 to 𝑅 𝑚 to be linear
transformations of 𝑅 𝑛 into 𝑅 𝑚 . What is the derivative of such a linear
transformationA The answer is very simple. If A ∈ L(𝑅 𝑛 , 𝑅 𝑚 ) and if x∈ 𝑅 𝑛 ,
then
61
Space for Hints

(19) A '(x) = A.
Note that x appears on the left side of (19), but not on the right. Both sides of
(19) are members of L(𝑅 𝑛 ,𝑅 𝑚 ), whereas Ax ∈ 𝑅 𝑚 . The proof of (19) is a
triviality, since
(20) A(x + h) - Ax = Ah, by the linearity of A.
With f (x) = Ax, the numerator in (14) is thus 0 for every h ∈ 𝑅 𝑛 .
In (17), r(h) = 0.
We now extend the chain rule (Theorem 5.5) to the present situation.

Theorem 3.15
Suppose E is an open set in 𝑅 𝑛 , f maps E into ∈ 𝑅 𝑚 , f is differentiable at
𝑥0 ∈ 𝐸, g maps an open set containing f (E) into 𝑅 𝑘 , and g is differentiable at
f(x0). Then the mapping F of E into 𝑅 𝑘 defined by
F(x) = g(f (x)) is differentiable at 𝑥0 , and
(2 1) F'(𝑥0 ) = g'(f(𝑥0 ))f '(𝑥0 ).
On the right side of (21), we have the product of two linear transformations, as
defined in Sec. 3.6.
Proof
Put 𝑦𝑜 = 𝑓(𝑥0 )A = f '(𝑥0 ), B = g'(𝑦𝑜 ), and define
u(h) = f (𝑥𝑜 + h) - f(𝑥𝑜 ) - Ah,
v(k) = g(𝑦𝑜 + k) - g(𝑦𝑜 ) - Bk,
for all h ∈ 𝑅 𝑛 , and k ∈ 𝑅 𝑚 , for which f(𝑥𝑜 + h) and g(𝑦𝑜 + k) are defined.
Then
(22) | u(h) |= 𝜀(h)|h|, |v(k)| = 𝜂(k) |k|,
where 𝜀(h) →0 as h ⟶ 0 and 𝜂(k)⟶ 0 as k⟶0.
Given h, put k = f(𝑥𝑜 + h) - f (𝑥𝑜 ). Then,
(23) |k|= |Ah + u(h)| ≤ [ ∥A∥+𝜀(h)]|h|,
and F(𝑥𝑜 + h) - F(𝑥𝑜 ) - BAh = g(𝑦𝑜 + k) - g(𝑦𝑜 ) - BAh
= B(k - Ah) + v(k) = Bu(h) + v(k).
62
 Space for Hints

Hence (22) and (23) imply, for h ≠ 0, that

|𝐹(𝑥𝑜 + h) − F(𝑥𝑜 ) − BAh |
≤∥ 𝐵 ∥ 𝜀 𝑕 + [∥ 𝐴 ∥ +𝜀(𝑕)]𝜂(𝑘)
|𝑕|
Let h → 0. Then 𝜀(h) → 0. Also, k → 0, by (23), so that 𝜂(k) → 0. It follows
that F'(𝑥0 ) = BA , which is what (21) asserts.

Definition(Partial derivatives) 3.16

We again consider a function f that maps an open set E ⊂ 𝑅 𝑛 into 𝑅 𝑚 . Let
{𝑒1 , 𝑒2 , … … 𝑒𝑛 } and {𝑢1 , 𝑢2 , … … 𝑢𝑚 } be the standard bases of 𝑅 𝑛 and 𝑅 𝑚 .
The components of f are the real functions 𝑓1 , 𝑓2 , … … 𝑓𝑚 defined by
𝑚
(24) 𝑓 𝑥 = 𝑖=1 𝑓𝑖 𝑥 𝑢𝑖 𝑥 ∈ 𝐸
or, equivalently, by 𝑓𝑖 𝑥 = f(x)𝑢𝑖 , 1 ≤ i ≤ m.
For 𝑥 ∈ 𝐸 1 ≤ i ≤ m, 1 ≤ j ≤ n, we define
𝑓 𝑖 𝑥+𝑡𝑒 𝑗 −𝑓 𝑖 (𝑥)
(25) 𝐷𝑗 𝑓𝑖 𝑥 = lim
𝑡→0 𝑡

provided the limit exists. Writing 𝑓𝑖 (𝑥1 , 𝑥2 , … … 𝑥𝑛 ) in place of 𝑓𝑖 (x), we see

that 𝐷𝑗 𝑓𝑖 is the derivative of 𝑓𝑖 with respect to 𝑥𝑗 , keeping the other variables
fixed.
The notation
𝜕𝑓 𝑖
(26) 𝜕𝑥 𝑗

is therefore often used in place of 𝐷𝑗 𝑓𝑖 , and 𝐷𝑗 𝑓𝑖 is called a partial derivative.

In many cases where the existence of a derivative is sufficient when dealing
with functions of one variable, continuity or at least boundedness of the partial
derivatives is needed for functions of several variables.

Theorem 3.17
Suppose f maps an open set E ⊂ 𝑅 𝑛 into 𝑅 𝑚 , and f is differentiable at a point
𝑥 ∈ 𝐸 .Then the partial derivatives 𝐷𝑗 𝑓𝑖 (x) exist, and
𝑚
(27) f '(x)𝑒𝑗 = 𝑖=1 𝐷𝑗 𝑓𝑖 𝑥 𝑢𝑖 (1 ≤ j ≤ n)
63
Space for Hints

Here, as in Sec. 3.1 6, {𝑒1 , 𝑒2 , … … 𝑒𝑛 } and {𝑢1 , 𝑢2 , … … 𝑢𝑚 } are the standard

bases of 𝑅 𝑛 and 𝑅 𝑚
Proof
Fix j. Since f is differentiable at x,
f(𝑥 + 𝑡𝑒𝑗 ) − 𝑓 𝑥 = 𝑓 ′ 𝑥 𝑡𝑒𝑗 + 𝑟(𝑡𝑒𝑗 )
𝑟 𝑡𝑒 𝑗
where → 0 𝑎𝑠 𝑡 → 0. The linearity of f '(x) shows therefore that
𝑡
f(𝑥+𝑡𝑒 𝑗 )−𝑓 𝑥
(28) lim𝑡→0 = 𝑓 ′ 𝑥 𝑒𝑗
𝑡

If we now represent f in terms of its components, as in (24), then (28)

becomes
𝑚 𝑓 𝑖 𝑥+𝑡𝑒 𝑗 −𝑓 𝑖 (𝑥)
(29) lim𝑡→0 𝑖=1 𝑢𝑖 = 𝑓 ′ 𝑥 𝑒𝑗
𝑡

It follows that each quotient in this sum has a limit as t →

0(see Theorem 4. 1 0), so that each 𝐷𝑗 𝑓𝑖 𝑥 exists, and then (27) follows from
(29). Here are some consequences of Theorem 3. 17:
Let [f '(x)] be the matrix that represents f '(x) with respect to our standard
bases, as in Sec. 3.3. Then f '(x)𝑒𝑖 is the jth column vector of [f'(x)], and (27)
shows therefore that the number 𝐷𝑗 𝑓𝑖 𝑥 occupies the spot in the 𝑖 𝑡𝑕 row and
𝑗 𝑡𝑕 column of [f '(x)]. Thus
𝐷1 𝑓1 (𝑥) ⋯ 𝐷𝑛 𝑓1 (𝑥)
[f'(x)] = ⋮ ⋱ ⋮
𝐷1 𝑓𝑚 (𝑥) ⋯ 𝐷𝑛 𝑓𝑚 (𝑥)
If h = 𝑕𝑗 𝑒𝑗 is any vector in 𝑅 𝑛 then (27) implies that
𝑚 𝑛
(30) [f‘(x)] = 𝑖=1{ 𝑗 =1(𝐷𝑗 𝑓𝑖 )(𝑥)𝑕𝑗 }𝑢𝑖

Example 3.18
Let 𝛾 be a differentiable mapping of the segment (a, b) ⊂ 𝑅1 into an open set
E ⊂ 𝑅 𝑛 , in other words, 𝛾 is a differentiable curve in E. Let f be a real-valued
differentiable function with domain E. Thus f is a differentiable mapping of E
into 𝑅1 . Define
64
 Space for Hints

(31) g(t) = f(𝛾( t)) .(a < t < b). The chain rule asserts
then that
(32) g'(t) = f'(𝛾 (t))𝛾 '(t) (a < t < b).
Since 𝛾'(t) ∈ L(𝑅1 , 𝑅 𝑛 ) and f'(𝛾 (t))∈ L(𝑅 𝑛 , 𝑅1 ) (32) defines g'(t) as a linear
operator on 𝑅1 . This agrees with the fact that g maps (a, b) into 𝑅1 However
g'(t) can also be regarded as a real number. (This was discussed in Sec. 3. 1 0.)
This number can be computed in terms of the partial derivatives of f and the
derivatives of the components of 𝛾, as we shall now see. With respect to the
standard basis {𝑒1 , 𝑒2 , … … 𝑒𝑛 } of 𝑅 𝑛 [𝛾 '( t)] is the 1 by n matrix (a "column
matrix‖) which has 𝛾𝑖 ′(t) in the 𝑖 𝑡𝑕 row, where {𝛾1 , 𝛾2 , … . . 𝛾𝑛 } n are the
components of y. For every 𝑥 ∈ E. [f‘(x)] is the 1 by n matrix (a "row matrix")
which has (𝐷𝑗 𝑓)(x) in the jth column. Hence [g'(t)] is the 1 by 1 matrix whose
only entry is the real number.
𝑛
(33) g'(t) = 𝑖=1(𝐷𝑖 𝑓)(𝛾 𝑡 )𝛾 ′ 𝑖 (𝑡)
This is a frequently encountered special case of the chain rule. It can be
rephrased in the following manner. Associate with each x ∈ E a vector, the so-
called "gradient" of f at x, defined by,
𝑛
(34) ∇f x = 𝑖=1 𝐷𝑖 𝑓(𝑥)𝑒𝑖

Since,
𝑛 ′
(35) 𝛾'(t) = 𝑖=1 𝛾 𝑖 (𝑡)𝑒𝑖

Since,
(33) can be written in the form
(36) g '(t) = (∇f)(𝛾(t)). 𝛾′(𝑡)
the scalar product of the vectors (∇f)(𝛾(t)) and 𝛾 ′ 𝑡 .
Let us now fix an x ∈ E. let u ∈ 𝑅 𝑛 be a unit vector (that is |u| = 1) and
specialize 𝛾 so that
(37) 𝛾(t) = x + tu ( −∞< t <∞ ). Then 𝛾 '(t) = u for every t.
Hence (36) shows that

65
Space for Hints

(38) g'(0) = (∇𝑓)(x).u.

On the other hand, (37) shows that
g(t) -g(0) = f(x+tu) - f(x)
Hence (38) gives
𝑓 𝑥 +𝑡𝑢 −𝑓 𝑥
(39) lim𝑡→0 = ∇𝑓 𝑥 . 𝑢
𝑡

The limit in (39) is usually called the directional derivative of f at x, in the

direction of the unit vector u, and may be denoted by (𝐷𝑢 𝑓)(x). If f and x are
fixed, but u varies, then (39) shows that (𝐷𝑢 𝑓)(x) attains its maximum when u
is a positive scalar multiple of (∇𝑓)(x). [The case (∇𝑓)(x) = 0 should be
excluded here. ] If u = 𝑢𝑖 𝑒𝑖 then (39) shows that (𝐷𝑢 𝑓)(x) can be expressed
in terms of the partial derivatives off at x by the formula
𝑛
(40) 𝐷𝑢 𝑓)(x) = 𝑖=1 𝐷𝑖 𝑓 𝑥 𝑢𝑖 .
Some of these ideas will play a role in the following theorem.

Theorem 3.19
Suppose f maps a convex open set E⊂ 𝑅 𝑛 into𝑅 𝑚 , f is differentiable in E, and
there is a real number M such that
∥ 𝑓′(𝑥) ∥≤ 𝑀
for every x ∈ E. Then
|f(b) – f(a)| ≤ 𝑀|𝑏 − 𝑎|
for all a ∈ E, b ∈ E.
Proof
Fix a ∈ E, b ∈ E. Define
𝛾(t) = (1 - t)a + tb for all t ∈ 𝑅1 such that 𝛾 (t) ∈ E. Since E is convex, 𝛾(t) ∈ E
if 0 ≤ t ≤ 1. Put g(t) = f(𝛾(t)).
Then g'(t) = f '(𝛾(t))𝛾 '(t)) = f '(𝛾 (t))(b - a), so that
|g' ( t)| <∥ f ' ( 𝛾 ( t)) ∥b – a| < M | b – a| for all t ∈ [0, 1 ]. By
Theorem 5. 1 9, |g(l) - g(0)| < M |b – a|.

66
 Space for Hints

But g(0) = f(a) and g(l) = f(b). This completes the proof.

Corollary
If, in addition, f '(x) = 0 for all x ∈ 𝐸then f is constant.
Proof
To prove this, note that the hypotheses of the theorem hold now with
M = 0.

Definition 3.20
A differentiable mapping f of an open set E ⊂ 𝑅 𝑛 into 𝑅 𝑚 is said to be
continuously differentiable in E if f' is a continuous mapping of E into L(𝑅 𝑛 ,
𝑅 𝑚 ). More explicitly, it is required that to every x ∈ 𝐸 and to every 𝜀 > 0
corresponds 𝛿 > 0 such that
ll f '(y) - f '(x)ll <𝜀 if y ∈ E and |x - y | <𝛿 If this is so, we also say
that f is a ℭ′ -mapping, or that f ∈ ℭ′(𝐸)

Theorem 3.21
Suppose f maps an open set E ⊂ 𝑅 𝑛 into 𝑅 𝑚 . Then f ∈ ℭ′(𝐸)if and only if the
partial derivatives exist and are continuous on E for 1 ≤ i ≤m, 1 ≤ i ≤n
Proof
Assume first that f ∈ ℭ′ 𝐸 . By (27),
(𝐷𝑗 𝑓𝑖 )(x) = (f '(x)𝑒𝑗 )𝑢𝑖
for all i, j and for all. Hence,
(𝐷𝑗 𝑓𝑖 (y) - (𝐷𝑗 𝑓𝑖 )(x) = {[f '(y) - f '(x)]𝑒𝑗 }𝑢𝑖 ,
and since |𝑢𝑖 | = |𝑒𝑗 | = 1, it follows that
𝜀
|𝐷𝑗 𝑓𝑖 (y) - (𝐷𝑗 𝑓𝑖 )(x)| < 𝑛 (y ∈ 𝑆, 1 ≤ 𝑗 ≤ 𝑛)

Hence, 𝐷𝑗 𝑓𝑖 is continuous.
For the converse, it suffices to consider the case m = 1. Fix x ∈ E and 𝜀> 0.

67
Space for Hints

Since E is open, there is an open ball S ⊂ E, with center at x and radius r, and
the continuity of the functions𝐷𝑗 𝑓shows that r can be chosen so that
(y ∈ S, 1 ≤ j ≤ n).
Suppose h = 𝑕𝑗 𝑒𝑗 . |h| < r, put 𝑣𝑜 = 0 and 𝑣𝑘 =𝑕1 𝑒1 + ⋯ … . . +𝑕𝑘 𝑒𝑘 , for 1 ≤
k ≤ n. Then,
𝑛
f(x + h) -f(x) = 𝑗 =1[𝑓(𝑥 + 𝑣𝑗 ) − 𝑓 𝑥 + 𝑣𝑗 −1 )
Since |𝑣𝑘 |< r for 1 ≤ k ≤ n and since S is convex, the segments with end
points x + 𝑣𝑗 −1 and x +𝑣𝑗 lie in S. Since,𝑣𝑗 = 𝑣𝑗 −1 + 𝑕𝑗 𝑒𝑗 the mean value
theorem (5. 1 0) shows that the 𝑗 𝑡𝑕 summand in (42) is equal to
𝑕𝑗 (𝐷𝑗 𝑓)(𝑥 + 𝑣𝑗 −1 + 𝜃𝑗 𝑕𝑗 𝑒𝑗

for some 𝜃𝑗 ∈ (0, 1 ), and this differs from 𝑕𝑗 (𝐷𝑗 𝑓)(𝑥 )by less than ||𝑕𝑗 | 𝜀 𝑛
using (41). By ( 42 ), it follows that
𝑛 1 𝑛
|f(x+h) -f(x) - 𝑗 =1 𝑕𝑗 (𝐷𝑗 𝑓)(𝑥 )|≤ 𝑛 𝑗 =1 |𝑕𝑗 |𝜀 ≤ |𝑕|𝜀

for all h such that |h| < r.

This says that f is differentiable at x and that f'(x) is the linear
function which assigns the number 𝑕𝑗 (𝐷𝑗 𝑓)(𝑥 ) to the vector h= 𝑕𝑗 𝑒𝑗 . The
matrix [f'(x}] consists of the row (𝐷1 𝑓)(𝑥), … … . . (𝐷𝑛 𝑓)(𝑥): since
𝐷1 𝑓, … … . . 𝐷𝑛 𝑓 are continuous functions on E, the concluding remarks of Sec.
3.9 show that f ∈ ℭ′ 𝐸 .

3.3 The contraction principle

We now interrupt our discussion of differentiation to insert a fixed point
theorem that is valid in arbitrary complete metric spaces. It will be used in the
proof of the inverse function theorem.

Definition 3.22

68
 Space for Hints

Let X be a metric space, with metric d. If 𝜑 maps X into X and if there is a

number c < 1 such that
(43) 𝑑 𝜑 𝑥 ,𝜑 𝑦 ≤ 𝑐 𝑑(𝑥, 𝑦)
for all x, y ∈ X, then 𝜑 is said to be a contraction of X in to X.

Theorem 3.23
If X is a complete metric space, and if 𝜑 is a contraction of X into X, then
there exists one and only one x ∈ X such that 𝜑 𝑥 = 𝑥.
In other words, 𝜑 has a unique fixed point. The uniqueness is a triviality for if
𝜑 𝑥 = 𝑥 and 𝜑 𝑦 = 𝑦, then (43) gives 𝑑 𝑥, 𝑦 ≤ 𝑐 𝑑(𝑥, 𝑦) which can only
happen when d(x, y) = 0.
The existence of a fixed point of 𝜑 is the essential part of the theorem. The
proof actually furnishes a constructive method for locating the fixed point.
Proof
Pick 𝑥0 ∈ 𝑋 arbitrarily, and define {𝑥𝑛 } recursively, by setting
(44) 𝑥𝑛+1 = 𝜑 𝑥𝑛 𝑛 = 0,1,2, … .
Choose 𝑐 < 1 so that (43) holds. For 𝑛 ≥ 1 we then have
𝑑 𝑥𝑛+1 , 𝑥𝑛 = 𝑑 𝜑 𝑥𝑛 , 𝜑 𝑥𝑛−1 ≤ 𝑐 𝑑(𝑥𝑛 , 𝑥𝑛−1 ) .
Hence induction gives
(45) 𝑑 𝑥𝑛+1 , 𝑥𝑛 ≤ 𝑐 𝑛 𝑑 𝑥1 , 𝑥0 𝑛 = 0,1,2, … .

If 𝑛 < 𝑚, it follows that

𝑑 𝑥𝑛 , 𝑥𝑚 ≤ 𝑑(𝑥𝑖 , 𝑥𝑖−1 )
𝑖=𝑛+1
𝑛 𝑛+1
≤ (𝑐 + 𝑐 + ⋯ + 𝑐 𝑚 −1 𝑑(𝑥1 , 𝑥0 )
≤ [(1 − 𝑐)−1 𝑑 𝑥1 , 𝑥0 ]𝑐 𝑛 .

69
Space for Hints

Thus 𝑥𝑛 is a Cauchy sequence. Since 𝑋 is complete, lim 𝑥𝑛 = 𝑥 for some

𝑥∈𝑋.
Since 𝜑 is a contraction, 𝜑 is continuous (in fact, uniformly continuous) on 𝑋.
Hence
𝜑 𝑥 = lim𝑛 →∞ 𝜑 𝑥𝑛 = lim𝑛→∞ 𝑥𝑛+1 = 𝑥 .

CYP QUESTIONS:

1. If 𝑆 is a nonem pty subset of a vector space 𝑋, prove thatthe s pan of 𝑆

is a vector space.
2. Prove that BA is linear if 𝐴 and 𝐵 are linear transformations.Prove
also that 𝐴−1 is linear and invertible .
3. Assume 𝐴 ∈ 𝐿( 𝑋, 𝑌) and 𝐴𝑥 = 0 only when 𝑥 = 0. Prove that 𝐴 is
then 1 - 1 .
4. Prove that null spaces and ranges of linear transformations are vector
spaces .
5. Prove that to every 𝐴 ∈ 𝐿( 𝑅", 𝑅^1 ) corresponds a unique 𝑦 ∈ 𝑅"
such that 𝐴𝑥 = 𝑥 • 𝑦.Prove also that ||𝐴 || = 𝑌 .

70
 Space for Hints

UNIT- 4
THE INVERSE FUNCTION THEOREM

Unit Structure:
Section 4.1: The inverse function Theorem
Section 4.2: The Implicit Function Theorem
Section 4.3: The Rank Theorem
Section 4.4: Determinants

The inverse function theorem states, roughly speaking, that a continuously

differentiable mapping f is invertible in a neighbourhood of any point x at
which the linear transformation f’(x) is invertible:

4.1The inverse function theorem

Theorem 4.1
Suppose f is a 𝒞′-mapping of an open set 𝐸 ⊂ 𝑅 𝑛 into 𝑅 𝑛 . f’(a) is invertible
for some 𝒂 ∈ 𝐸 and 𝒃 = 𝒇(𝒂). Then
(a) there exists open sets U and V in 𝑅 𝑛 such that 𝒂 ∈ 𝑈, 𝒃 ∈ 𝑉. f is one
to one on U, and f(U) = V;
(b) if g is the inverse of f [which exists by (a)], defined in V by
g(f(x))=x (𝒙 ∈ 𝑈).
then 𝒈 ∈ 𝓒′(𝑽)
Writing the equation y=f(x) in component form. We arrive at the following
interpretation of the conclusion of the theorem: The system of n equations
𝑦𝑖 = 𝑓𝑖 𝑥1 , … … . , 𝑥𝑛 1 ≤ 𝑖 ≤ 𝑛
can be solved for 𝑥1 , … … , 𝑥𝑛 in terms of 𝑦1 , … … 𝑦𝑛 if we restrict x and y to
small enough neighbourhoods of a and b; the solutions are unique and
continuously differentiable.
71
Space for Hints

Proof
(a) Put 𝒇′ 𝒂 = 𝑨, and choose 𝜆 so that
(46) 2𝜆‖𝐴−1 ‖ = 1.
Since 𝑓 ′ is continuous at a. there is an open ball 𝑈 ⊂ 𝐸, with centre at a such
that
(47) ‖𝑓 ′ 𝑥 − 𝐴‖ < 𝜆 𝑥∈𝐸 .
We associate to each 𝑦 ∈ 𝑅 𝑛 a function 𝜑, defined by
(48) 𝜑 𝑥 = 𝑥 + 𝐴−1 𝑦 − 𝑓 𝑥 (𝑥 ∈ 𝐸)
Note that f(x) = y if and only if x is a fixed point of 𝜑.
Since 𝜑 ′ 𝑥 = 𝐼 − 𝐴−1 𝑓′(𝑥) = 𝐴−1 (𝐴 − 𝑓 ′ 𝑥 ), (46) and (47) imply that
1
(49) ‖𝜑 ′ 𝑥 ‖ < 𝑥 ∈ 𝑈
2

Hence
1
(50) 𝜑 𝑥1 − 𝜑(𝑥2 ) ≤ 𝑥1 − 𝑥2 𝑥1 , 𝑥2 ∈ 𝑈 .
2

by Theorem 3.13. It follows that 𝜑 has at most one fixed point in U. so that
f(x) = y for at most one 𝑥 ∈ 𝑈.
Thus f is 1-1in U.
Next, put V=f(U), and pick 𝑦0 ∈ 𝑉. Then 𝑦0 = 𝑓(𝑥0 ) for some 𝑥0 ∈ 𝑈. Let B
be an open ball with centre at 𝑥0 and radius r >0, so small that its closure 𝐵
lies in U. We will show that 𝑦 ∈ 𝑉 whenever 𝑦 − 𝑦0 < 𝜆𝑟. This proves, of
course that V is open.
Fix y, 𝑦 − 𝑦0 < 𝜆𝑟. With 𝜑 as in (48).
𝑟
𝜑 𝑥 − 𝑥0 = 𝐴−1 (𝑦 − 𝑦0 ) < ‖𝐴−1 ‖ 𝜆𝑟 =
2
If 𝑥 ∈ 𝐵, it therefore follows from (50) that
𝜑 𝑥 − 𝑥0 ≤ 𝜑 𝑥 − 𝜑 𝑥0 + 𝜑 𝑥0 − 𝑥0
1 𝑟
< 𝑥 − 𝑥0 + ≤ 𝑟 ;
2 2
hence 𝜑 𝑥 ∈ 𝐵. Note that (50) holds if 𝑥1 ∈ 𝐵 , 𝑥2 ∈ 𝐵.
Thus 𝜑 is a contraction of 𝐵 into 𝐵. Being a closed subset of 𝑅 𝑛 .
72
 Space for Hints

𝐵 is complete. Theorem 4.23 implies therefore that 𝜑 has a fixed point 𝑥 ∈ 𝐵.

For this x. f(x) – y. Thus 𝑦 ∈ 𝑓 𝐵 ⊂ 𝑓 𝑈 = 𝑉.
This proves part (a) of the theorem.
(b) Pick 𝑦 ∈ 𝑉, 𝑦 + 𝑘 ∈ 𝑉. Then there exist 𝑥 ∈ 𝑈, 𝑥 + 𝑕 ∈ 𝑈, so that
𝑦 = 𝑓 𝑥 , 𝑦 + 𝑘 = 𝑓 𝑥 + 𝑕 . With 𝜑 as in (48),
𝜑 𝑥 + 𝑕 − 𝜑 𝑥 = 𝑕 + 𝐴−1 𝑓 𝑥 − 𝑓 𝑥 + 𝑕 = 𝑕 − 𝐴−1 𝑘.
1 1
By (50), 𝑕 − 𝐴−1 𝑘 ≤ 𝑕 . Hence 𝐴−1 𝑘 ≥ 𝑕 , and
2 2

(51) 𝑕 ≤ 2‖𝐴−1 ‖ 𝑘 = 𝜆−1 𝑘 .

By (46), (47), and Theorem 3.8. 𝑓′(𝑥) has an inverse, say T. Since
𝑔 𝑦 + 𝑘 − 𝑔 𝑦 − 𝑇𝑘 = 𝑕 − 𝑇𝑘 = −𝑇 𝑓 𝑥 + 𝑕 − 𝑓 𝑥 − 𝑓 ′ 𝑥 𝑕 ,
(51) implies
𝑔 𝑦 + 𝑘 − 𝑔 𝑦 − 𝑇𝑘 ‖𝑇‖ 𝑓 𝑥 + 𝑕 − 𝑓 𝑥 − 𝑓 ′ 𝑥 𝑕
≤ .
𝑘 𝜆 𝑕
As 𝑘 → 0, (51) shows that 𝑕 → 0. The right side of the last inequality thus
tends to 0. Hence the same is true of the left. We have thus proved that
𝑔′ 𝑦 = 𝑇. But T was chosen to be the inverse of 𝑓 ′ 𝑥 = 𝑓 ′ 𝑔 𝑦 . Thus
(52) 𝑔′ 𝑦 = {𝑓′(𝑔 𝑦 }−1 𝑦 ∈ 𝑉 .
Finally, note that g is a continuous mapping of V onto U (since g is
differentiable), that 𝑓 ′ is a continuous mapping of U into the set 𝛺 of all
invertible elements of 𝐿 𝑅 𝑛 , and that inversion is a continuous mapping of 𝛺
onto 𝛺, by Theorem 3.8. if we obtain these facts with (52), we see that
𝑔 ∈ 𝒞′ 𝑉 .
This completes the proof.
Remark.
The full force of the assumption that 𝑓 ∈ 𝒞′(𝐸) was only used in the last
paragraph of the preceding proof. Everything else, down to Eq.(52) was
derived from the existence of 𝑓′(𝑥) for 𝑥 ∈ 𝐸, the invertibility of 𝑓 ′ 𝑎 , and

73
Space for Hints

the continuity of 𝑓′ at just the point a. In this connection, we refer to the article
by A. Nijenhuis in Amer Math Monthly vol 81,1974, pp. 969-980.
The following is an immediate consequence of part (a) of the inverse function.

Theorem 4.2
If f is a 𝒞‘- mapping of an open set 𝐸 ⊂ 𝑅 𝑛 into 𝑅 𝑛 and if 𝑓′(𝑥) is invertible
for every 𝑥 ∈ 𝐸, then f(W) is an open subset of 𝑅 𝑛 foe every open set W⊂ E.
In other words, f is an open mapping of E into 𝑅 𝑛 .
The hypotheses made in this theorem ensure that each point x ∈ E has a
neighbourhood in which f is 1-1. This may be expressed by saying that f is
locally one-to-one in E. But f need not be 1-1 in E under these circumstances.

4.2 The implicit function theorem

If f is a continuously differentiable real function in the plane, then the equation

f(x, y) = 0 can be solved for y in terms of x in a neighbourhood of any point
𝜕𝑓
(a, b) at which f(a, b) = 0 and ≠ 0. Likewise, one can solve for x in terms
𝜕𝑦
𝜕𝑓
of y near (a, b) if 𝜕𝑦 ≠ 0 at (a. b). For a simple example which illustrates the
𝜕𝑓
need for assuming ≠ 0, consider f(x, y) = 𝑥 2 + 𝑦 2 - 1.
𝜕𝑦

The preceding very informal statement is the simplest case (the case m = n = I
of Theorem 4.5) of the so-called "implicit function theorem." Its proof makes
strong use of the fact that continuously differentiable transformations behave
locally very much like their derivatives. Accordingly, we first prove Theorem
4.4, the linear version of Theorem 4.5.

Notation 4.3
If x = (𝑥1 , … … , 𝑥𝑛 ) ∈ 𝑅 𝑛 and y = (𝑦1 , … … 𝑦𝑚 ) ∈ 𝑅 𝑛 , let us write (x, y) for the
point (or vector)
74
 Space for Hints

(𝑥1 , … … , 𝑥𝑛 ,𝑦1 , … … 𝑦𝑚 ) ∈ 𝑅 𝑛+𝑚

In what follows, the first entry in (x, y) or in a similar symbol will always be a
vector in 𝑅 𝑛 , the second will be a vector in 𝑅 𝑚 .
Every 𝐴 ∈ 𝐿 𝑅 𝑛+𝑚 , 𝑅 𝑛 can be split into two linear transformations 𝐴𝑥 and𝐴𝑦 ,
defined by
(53) 𝐴𝑥 𝑕 = 𝐴 𝑕, 0 , 𝐴𝑦 𝑘 = 𝐴 0, 𝑘
for any 𝑕 ∈ 𝑅 𝑛 , 𝑘 ∈ 𝑅 𝑚 . Then 𝐴𝑥 ∈ 𝐿 𝑅 𝑛 , 𝐴𝑦 ∈ 𝐿 𝑅 𝑚 , 𝑅 𝑛 , and
(54) 𝐴 𝑕, 𝑘 = 𝐴𝑥 𝑕 + 𝐴𝑦 𝑘
The linear version of the implicit function theorem is now almost obvious.

Theorem 4.4
If 𝐴 ∈ 𝐿 𝑅 𝑛+𝑚 , 𝑅 𝑛 and if 𝐴𝑥 is invertible, then there corresponds to
every 𝑘 ∈ 𝑅 𝑚 a unique 𝑕 ∈ 𝑅 𝑛 such that 𝐴 𝑕, 𝑘 = 0.
This h can be computed from k by the formula
(55) 𝑕 = −(𝐴𝑥 )−1 𝐴𝑦 𝑘
Proof
By (54). A(h, k) = 0 if and only if
𝐴𝑥 𝑕 + 𝐴𝑦 𝑘 = 0,
which is the same as (55) when 𝐴𝑥 is invertible.
The conclusion of Theorem 4.4 is, in other words, that the equation A(h , k) =
0 can be solved (uniquely) for h if k is given, and that the solution h is a linear
function of k. Those who have some acquaintance with linear algebra will
recognize this as a very familiar statement about systems of linear equations.

Theorem 4.5
Let f be a 𝒞 ′ mapping of an open set 𝐸 ⊂ 𝑅 𝑛+𝑚 𝑖𝑛𝑡𝑜 𝑅 𝑛 , such that f(a. b) = 0
for some point (a, b) 𝜖 𝐸.
Put A = f '(a, b) and assume that 𝐴𝑥 is invertible.

75
Space for Hints

Then there exist open sets 𝑈 ⊂ 𝑅 𝑛+𝑚 𝑎𝑛𝑑 𝑊 ⊂ 𝑅 𝑚 , with (a, b) ∈U and
b ∈W, having the following property:
To every y ∈ W corresponds a unique x such that
(56) (x, y) ∈U and f{x, y) = 0.
If this x is defined to be g(y), then g is a 𝒞‘ mapping of W into 𝑅 𝑛 , g(b) = a,
Type equation here.
(57) f(g(y), y) = 0 (y ∈ W),
and
(58) 𝑔′ 𝑏 = −(𝐴𝑥 )−1 𝐴𝑦
The function g is "implicitly" defined by (57). Hence the name of the theorem.
The equation f(x, y) = 0 can be written as a system of n equations in n + m
variables:
(59) 𝑓1 𝑥1 , … . . 𝑥𝑛 , 𝑦1 , … . . . 𝑦𝑚 = 0
………………………………….
𝑓𝑛 𝑥1 , … . . 𝑥𝑛 , 𝑦1 , … . . . 𝑦𝑚 = 0
The assumption that 𝐴𝑥 is invertible means that the n by n matrix
𝐷1 𝑓1 ⋯ 𝐷𝑛 𝑓1
⋮ ⋱ ⋮
𝐷1 𝑓𝑛 ⋯ 𝐷𝑛 𝑓𝑛
evaluated at (a, b) defines an invertible linear operator in 𝑅𝑛 ; in other words,
its column vectors should be independent, or, equivalently, its determinant
should be ≠ 0. (See Theorem 4.13.) If, furthermore, (59) holds when x =a and
y = b, then the conclusion of the theorem is that (59) can be solved for
𝑥1 , … . . 𝑥𝑛 in terms of 𝑦1 , … . . . 𝑦𝑚 , for every y near b, and that these solutions
are continuously differentiable functions of y.
Proof
Define F by
(60) F(x, y) = (f(x, y), y) ((x, y) e E).

76
 Space for Hints

Then F is a 𝒞‘ mapping of E into 𝑅 𝑛+𝑚 . We claim that F'(a, b) is an invertible

element of L(𝑅 𝑛+𝑚 ) :
Since f(a, b) = 0, we have
f(a + h, b + k) = A(h, k) + r(h, k),
where r is the remainder that occurs in the definition of f'(a, b). Since
F(a + h, b + k) - F{a, b) = (f (a + h, b + k), k)
= (A(h, k), k) + (r(h, k), 0)
it follows that F'(a. b) is the linear operator on 𝑅 𝑛+𝑚 that maps (h, k) to (A(h,
k), k). If this image vector is 0, then A(h, k) = 0 and k = 0, hence A (h, 0) = 0,
and Theorem 4.4 implies that h = 0. It follows that F'(a. b) is 1-1 ; hence it is
invertible (Theorem 4.5).
The inverse function theorem can therefore be applied to F. It shows that there
exist open sets U and V in 𝑅 𝑛+𝑚 , with (a, b) ∈ U. (0. b) ∈ V, such that F is a
1-1 mapping of U onto V.
We let W be the set of all y ∈ 𝑅 𝑚 such that (0, y) ∈ V. Note that b ∈ W.
It is clear that W is open since V is open. If y ∈ W, then (0, y) = F(x. y) for
some (x, y) E U. By (60), f(x, y) = 0 for this x.
Suppose, with the same y, that (x', y' ) ∈ U and f (x' , y) = 0. Then
F(x', y) = (f(x' , y), y) = (f(x. y). y) = F(x. y).
Since F is 1-l in U, it follows that x' = x.
This proves the first part of the theorem.
For the second part, define g(y), for y ∈ W. so that (g(y), y) ∈ U and (57)
holds. Then
(61) F(g(y), y) = (0, y) (y ∈ W).
If G is the mapping of V onto U that inverts F, then G ∈𝒞‘, by the inverse
function theorem, and ( 61) gives
(62) (g(y), y) = G(0, y) (y E W).
Since G ∈ 𝒞′, (62), shows that g ∈ 𝒞′.
Finally, to compute g'(b), put (g(y), y) = 𝜑 (y). Then
77
Space for Hints

(63) 𝜑'(y)k = (g'(y)k, k) (y ∈ W, k ∈ 𝑅 𝑚 ) .

By (57), f(𝜑(y)) = 0 in W. The chain rule shows therefore that
f '(𝜑(y))𝜑'(y) = 0.
When y = b, then 𝜑(y) = (a, b), and f '(𝜑(y)) = A. Thus
(64) A𝜑'(b) = 0.
It now follows from (64), (63), and (54), that
𝐴𝑥 g'(b)k + 𝐴𝑦 k = A (g'(b)k, k) = A𝜑'(b)k = 0
for every k ∈ 𝑅 𝑚 . Thus
(65) 𝐴𝑥 g'(b)k + 𝐴𝑦 = 0
This is equivalent to (58), and completes the proof.
Note. In terms of the components of f and g, (65) becomes
𝑛

(𝐷𝑗 𝑓𝑖 )(𝑎, 𝑏)(𝐷𝑘 𝑔𝑗 )(𝑏) = −(𝐷𝑛+𝑘 𝑓𝑖 )(𝑎, 𝑏)

𝑗 =1

or
𝑛
𝜕𝑓𝑖 𝜕𝑔𝑗 𝜕𝑓𝑖
= −( )
𝜕𝑥𝑗 𝜕𝑦𝑘 𝜕𝑦𝑘
𝑗 =1

where 1 ≤ i ≤ n, 1 ≤ k ≤ m.
𝜕𝑔
For each k, this is a system of n linear equations in which the derivatives𝜕𝑦 𝑗 ,
𝑘

( 1 ≤ j ≤n) are the unknowns.

Example 4.6
Take n = 2, m = 3, and consider the mapping f = (𝑓1 , 𝑓2 ) of 𝑅 5 into 𝑅 2 given by
𝑓1 𝑥1 , 𝑥2 , 𝑦1 , 𝑦2 , 𝑦3 = 2𝑒 𝑥 1 + 𝑥2 𝑦1 − 4𝑦2 + 3
𝑓2 𝑥1 , 𝑥2 , 𝑦1 , 𝑦2 , 𝑦3 = 𝑥2 cos 𝑥1 − 6𝑥1 + 2𝑦1 − 𝑦3
If a = (0, 1) and b = (3, 2, 7), then f(a, b) = 0.
With respect to the standard bases, the matrix of the transformation A = f '(a,
b) is

78
 Space for Hints

2 3 1 −4 0
𝐴 =
−6 1 2 0 −1
Hence
2 3 1 −4 0
𝐴𝑥 = , 𝐴𝑦 = .
−6 1 2 0 −1
We see that the column vectors of 𝐴𝑥 are independent. Hence 𝐴𝑥 is
invertible and the implicit function theorem asserts the existence of a 𝒞' -
mapping g, defined in a neighborhood of (3, 2, 7), such that g(3, 2, 7) = (0, 1)
and f (g(y), y) = 0.
We can use (58) to compute g'(3, 2, 7) : Since
1 1 −3
(𝐴𝑥 )−1 = 𝐴𝑥 −1
=
20 6 2
(58) gives
1 1 −3
1 1 −3 1 −4 0
𝑔′(3,2,7) = − = 4 5 20
20 6 2 2 0 −1 −1 6 1
2 5 10
In terms of partial derivatives, the conclusion is that
1 1 −3
𝐷1 𝑔1 = 𝐷2 𝑔1 = 𝐷3 𝑔1 =
4 5 20
−1 6 1
𝐷1 𝑔2 = 𝐷2 𝑔2 = 𝐷3 𝑔2 =
2 5 10
at the point (3, 2, 7).

4.3 The rank theorem

Although this theorem is not as important as the inverse function theorem or
the implicit function theorem, we include it as another interesting illustration
of the general principle that the local behaviour of a continuously
differentiable mapping F near a point x is similar to that of the linear
transformation F'(x).
Before stating it, we need a few more facts about linear transformations.

79
Space for Hints

Definitions 4.7
Suppose X and Y are vector spaces, and A ∈ L( X, Y), as in Definition 3.6.
The null space of A, Ɲ(A), is the set of all x ∈ X at which Ax = 0. It is clear
that Ɲ(A) is a vector space in X.
Likewise, the range of A, R(A), is a vector space in Y.
The rank of A is defined to be the dimension of R(A).
For example, the invertible elements of L(𝑅 𝑛 ) are precisely those whose rank
is n. This follows from Theorem 3.5.
If A ∈ L(X, Y) and A has rank 0, then Ax = 0 for all x ∈A, hence Ɲ(A) = X.

Definition(Projections)4.8

Le t X be a vector space. An operator P ∈ L( X) is said to be a projection in X

if 𝑃2 = P.
More explicitly, the requirement is that P(Px) = Px for every x ∈ X. In other
words, P fixes every vector in its range 𝑅(P).
Here are some elementary properties of projections:
(a) If P is a projection in X. then every x ∈ X has a unique representation
of the form
x =𝑥1 + 𝑥2
where 𝑥1 ∈ 𝑅 𝑃 , 𝑥2 ∈ Ɲ(𝑃).
To obtain the representation, put 𝑥1 = Px, 𝑥2 = x -𝑥1 . Then
P𝑥2 = Px - P𝑥1 = Px - 𝑃2 x = 0. As regards the uniqueness, apply P to the
equation x =𝑥1 + 𝑥2 . Since 𝑥1 ∈ 𝑅( P). P𝑥1 = 𝑥1 ; since P 𝑥2 = 0, it follows
that 𝑥1 = Px.
(b) If X is a finite-dimensional vector space and if 𝑋1 is a vector space in
X, then there is a projection P in X with R(P) = 𝑋1 .
If 𝑋1 contains only 0, this is trivial : put Px = 0 for all X ∈ X.

80
 Space for Hints

Assume dim 𝑋1 = k > 0. By Theorem 3. 3, X has then a basis {𝑢1 , … . 𝑢𝑛 } such

that {𝑢1 , … . 𝑢𝑘 } is a basis of 𝑋1 . Define
P(𝑐1 𝑢1 + ⋯ + 𝑐𝑛 𝑢𝑛 ) = 𝑐1 𝑢1 + ⋯ + 𝑐𝑛 𝑢𝑛
for arbitrary scalars 𝑐1 , … 𝑐2
Then Px = x for every x∈ 𝑋1 , and 𝑋1 = R(P).
Note that {𝑢𝑘+1 , … 𝑢𝑘 } is a basis of Ɲ(P). Note also that there are infinitely
many projections in X, with range 𝑋1 , if 0 < dim 𝑋1 < dim X.

Theorem 4.9
Suppose m, n, r are nonnegative integers, m ≥ r. n ≥r, F is a 𝒞' -mapping of an
open set E ⊂𝑅 𝑛 into 𝑅 𝑛 , and F'(x) has rank r for every x ∈ 𝐸.
Fix a ∈ 𝐸, put A = F'(a), let 𝑌1 be the range of A. and let P be a projection in
𝑅 𝑚 whose range is 𝑌1 . Let 𝑌2 be the null space of P.
Then there are open sets U and V in 𝑅 𝑛 , ·with a ∈ U, U ⊂ E, and there is a 1-
1 𝒞'-mapping H of V onto U (whose inverse is also of class 𝒞') such that
(66) F(H(x)) = Ax + 𝜑(Ax) (x ∈ V)
where cp is a 𝒞' -mapping of the open set A(V) ⊂𝑌1 into 𝑌2 .
After the proof we shall give a more geometric description of the information
that (66) contains.
Proof
If r = 0, Theorem 3. 19 shows that F(x) is constant in a neighborhood U of a,
and (66) holds trivially, with V = U, H(x) = x, 𝜑(0) = F(a).
From now on we assume r > 0. Since dim 𝑌1 = r, 𝑌1 has a basis {𝑦1 , … 𝑦𝑟 }.
Choose 𝑧𝑖 ∈ 𝑅 𝑛 so that 𝐴𝑧𝑖 = 𝑦𝑖 (1≤ i ≤ r ), and define a linear mapping S of
𝑌1 into 𝑅 𝑛 by setting
(67) S(𝑐1 𝑦1 + ⋯ + 𝑐𝑟 𝑦𝑟 ) = 𝑐1 𝑧1 + ⋯ + 𝑐𝑟 𝑧𝑟
for all scalars 𝑐1 , … 𝑐𝑟 .
Then 𝐴𝑆𝑦𝑖 = 𝐴𝑧𝑖 = 𝑦𝑖 𝑓𝑜𝑟 1 ≤ 𝑖 ≤ 𝑟. Thus
(68) ASy =y (y ∈ 𝑌1 )
81
Space for Hints

Define a mapping G of E into 𝑅 𝑛 by setting

(69) G(x) = x + SP[F(x) - Ax ] (x ∈ E ).
Since F'(a) = A . differentiation of (69) shows that G'(a) = I. the identity
operator on 𝑅 𝑛 . By the inverse function theorem, there are open sets U and V
in𝑅 𝑛 , with a ∈ U, such that G is a 1-1 mapping of U onto V whose inverse
H is also of class 𝒞'. Moreover, by shrinking U and V, if necessary, we can
arrange it so that V is convex and H'(x) is invertible for every x ∈ V.
Note that ASP A = A, since PA = A and (68) holds. Therefore (69) gives
(70) AG(x) = PF(x) (x ∈ 𝐸).
In particular, (70) holds for x ∈ U. If we replace x by H(x), we obtain
(71) PF(H(x)) = Ax (x ∈ V).
Define
(72) 𝜓(x) = F(H(x)) - Ax (x ∈ V).
Since PA = A, (71) implies that Pψ(x) = 0 for all x ∈ V. Thus ψ is a 𝒞' -
mapping of V into 𝑌2 .
Since V is open, it is clear that A(V) is an open subset of its range R(A) = 𝑌1
To complete the proof, that is , to go from (72) to (66), we have to show that
there is a 𝒞'-mapping 𝜑 of A( V) into 𝑌2 which satisfies
(73) 𝜑(Ax) = ψ(x) (x ∈V).
As a step toward (73), we will first prove that
(74) 𝜓(𝑥1 ) = 𝜓(𝑥2 )
If𝑥1 ∈ V, 𝑥2 ∈ V, A𝑥1 = A𝑥2
Put 𝛷(x) = F(H(x)), for x ∈ V. Since H '(x) has rank n for every x ∈ V, and
F'(x) has rank r for every x ∈ U, it follows that
(75) rank Φ'(x) = rank F'(H(x))H'(x) = r (x ∈ V).
Fix x ∈ V. Let M be the range of Φ'(x). Then M c: R m , dim M = r. By (7 1 ),
(76) PΦ'(x) = A.
Thus P maps M onto R(A) = 𝑌1 . Since M and 𝑌1 have the same dimension, it
follows that P (restricted to M) is 1 - 1 .
82
 Space for Hints

Suppose now that Ah = 0. Then PΦ'(x)h = 0, by (76). But Φ'(x)h ∈ M, and P

is 1-1 on M. Hence Φ'(x)h = 0. A look at (72) shows now that we have proved
the following :
lf x ∈ V and Ah = 0, then 𝜓'(x)h = 0.
We can now prove (74). Suppose 𝑥1 ∈ 𝑉, 𝑥2 ∈ 𝑉, 𝐴𝑥1 = 𝐴𝑥2 . Put 𝑕 = 𝑥1 −
𝑥2 and define
(77) 𝑔 𝑡 = 𝜓 𝑥1 + 𝑡𝑕 (0 ≤ 𝑡 ≤ 1)
The convexity of V shows that 𝑥1 + 𝑡𝑕 ∈ V for these t. Hence
(78) g'(t) = ψ' 𝑥1 + 𝑡𝑕 𝑕 = 0 (0 ≤ 𝑡 ≤ 1),
so that g(1) = g(0). But g(l) = ψ(𝑥2 )and g(0) = ψ(𝑥1 ). Thi s proves (74).
By (74), ψ(x) depends only on Ax, for x ∈ V. Hence (73) defines 𝛷
unambiguously in A( V). It only remains to be proved that 𝜑 ∈ 𝒞'.
Fix 𝑦0 ∈ A(V), fix 𝑥0 ∈ V so that A𝑥0 =𝑦0 . Since V is open,𝑦0 has a
neighbourhood W in 𝑌1 such that the vector
(79) x = 𝑥0 + S(y - 𝑦0 )
lies in V for all y ∈ W. By (68),
Ax = A𝑥0 + y - 𝑦0 = y·
Thus (73) and (79) give (80)
𝜑 𝑦 = 𝜓(𝑥0 − 𝑆𝑦0 + 𝑆𝑦) (y ∈W).
This formula shows that 𝜑 ∈ 𝒞' in W, hence in A(V), since 𝑦0 was chosen
arbitrari ly in A( V).
The proof is now complete.
Here is what the theorem tells us about the geometry of the mapping F.
If y ∈ F(U) then y = F(H(x)) for some x ∈V, and (66) shows that Py = Ax.
Therefore
(81) y = Py + 𝜑(Py) (y ∈ F(U)).
This shows that y is determined by its projection Py, and that P, restricted to
F(U), is a 1-1 mapping of F( U) onto A ( V). Thus F(U) is a an "r-dimensional

83
Space for Hints

surface" with precisely one point "over" each point of A( V). We may also
regard F( U) as the graph of 𝜑.
If Φ(x) = F(H(x)), as in the proof, then (66) shows that the level sets of
Φ(these are the sets on which Φ attains a given value) are precisely the level
sets of A in V. These are ― flat‖ since they are intersections with V of
translates of the vector space Ɲ(A). Note that dim Ɲ(A) = n - r.
The level sets of F in U are the images under H of the flat level sets of Φ in V.
They are thus "(n - r)-dimensional surfaces" in U.

4.4 Determinants
Determinants are numbers associated to square matrices, and hence to the
operators represented by such matrices. They are 0 if and only if the
corresponding operator fails to be invertible. They can therefore be used to
decide whether the hypotheses of some of the preceding theorems are
satisfied.

Definition 4.10
If (𝑗1 , … 𝑗𝑛 ) is an ordered n-tuple of integers, define
(82) s(𝑗1 , … 𝑗𝑛 ) = 𝑝<𝑞 𝑠𝑔𝑛 (𝑗𝑞 − 𝑗𝑝 ),
Where sgn x = 1 if x > 0, sgn x = -1 if x<0, sgn x = 0 if x = 0. Then s(𝑗1 , … 𝑗𝑛 )
= 1, - 1, or 0, and it changes sign if any two of the j's are interchanged.
Let [A] be the matrix of a linear operator A on 𝑅 𝑛 . relative to the standard
basis {𝑒1 , … 𝑒𝑛 }, with entries a(i,j) in the ith row and jth column. The
determinant of [A] is defined to be the number
(83) det 𝐴 = 𝑠 𝑗1 , … 𝑗𝑛 𝑎(1, 𝑗1 )𝑎 2, 𝑗2 … 𝑎 𝑛, 𝑗𝑛 .
The sum in (83) extends over all ordered n-tuples of integers (𝑗1 , … 𝑗𝑛 ) with
1 ≤ 𝑗𝑟 ≤ 𝑛.
The column vectors 𝑥 𝑖 of [A] are
𝑛
(84) 𝑥𝑗 = 𝑖=1 𝑎(𝑖, 𝑗)𝑒𝑖 (1 ≤ 𝑗 ≤ 𝑛)
84
 Space for Hints

It will be convenient to think of det [A] as a function of the column vectors of

[A]. If we write
det (𝑥1 , … 𝑥𝑛 ) = det [A],
det is now a real function on the set of all ordered n- tuples of vectors in 𝑅 𝑛 .

Theorem 4.11
(a) If I is the identity operator on 𝑅 𝑛 , then
det [I] = det (𝑒1 , … 𝑒𝑛 ) = 1.
(b) det is a linear function of each of the column vectors 𝑥𝑗 , if the others are
held fixed.
(c) If [A]1 is obtained from [A] by interchanging two columns, then
det [A]1 = - det [A ].
(c) If [A ] has two equal columns, then det [A] = 0.

Proof
If A = I, then a(i, i) = I and a(i,j) = 0 for i ≠j. Hence
det [I] = s( 1, 2, . . . , n) = 1,
which proves (a). By (82), s(𝑗1 , … 𝑗𝑛 ) = 0 if any two of the j's are equal . Each
of the remaining n! products in (83) contains exactly one factor from each
column. This proves (b). Part (c) is an immediate consequence of the fact that
s(𝑗1 , … 𝑗𝑛 ) changes sign if any two of the j's are interchanged, and (d) is a
corollary of (c).

Theorem 4.12
1f [A] and [B] are n by n matrices, then
det ([B][A]) = det [B] det [A ].
Proof
If 𝑥1 , … 𝑥𝑛 are the columns of [A], define
∆𝐵 𝑥1 , … 𝑥𝑛 = ∆𝐵 𝐴 = det 𝐵 𝐴 .
85
Space for Hints

The columns of [B][A] are the vectors 𝐵𝑥1 , … 𝐵𝑥𝑛 . Thus

(86) ∆𝐵 𝑥1 , … 𝑥𝑛 = det( 𝐵𝑥1 , … 𝐵𝑥𝑛 )
By (86) and Theorem 4.11, ∆𝐵 also has properties 4.11 (b) to (d). By (b) and
(84)

∆𝐵 𝐴 = ∆𝐵 ( 𝑎 𝑖, 1 𝑒𝑖 , 𝑥2 , … 𝑥𝑛 ) = 𝑎 𝑖, 1 ∆𝐵 (𝑒𝑖 , 𝑥2 , … 𝑥𝑛 )
𝑖 𝑖

Repeating this process with 𝑥2 , … 𝑥𝑛 , we obtain

(87) ∆𝐵 𝐴 = 𝑎 𝑖1 , 1 𝑎 𝑖2 , 2 … 𝑎 𝑖𝑛 , 𝑛 ∆𝐵 𝑒𝑖1 , … 𝑒𝑖𝑛
the sum being extended over all ordered n-tuples (𝑖1 , … 𝑖𝑛 ) with 1 ≤ 𝑖𝑟 ≤ n.
By (c) and (d),
(88) ∆𝐵 𝑒𝑖1 , … 𝑒𝑖𝑛 = 𝑡(𝑖1 , … 𝑖𝑛 )∆𝐵 (𝑒1 , … 𝑒𝑛 ),
where t = 1, 0, or -1 , and since [B] [I] = [B], (85) shows that
(89) ∆𝐵 (𝑒1 , … 𝑒𝑛 ), = det [B].
Substituting (89) and (88) into (87), we obtain
det ([B][A ]) = { 𝑎 𝑖1 , 1 … 𝑎 𝑖𝑛 , 𝑛 𝑡 𝑖1 , … 𝑖𝑛 } det [B],
for all n by n matrices [A] and [B]. Taking B = I, we see that the above sum in
braces is det[A]. This proves the theorem.

Theorem 4.13
A linear operator A on 𝑅 𝑛 is invertible if and only if det [A]≠ 0.
Proof
If A is invertible, Theorem 4.12 shows that
det [A ] det [𝐴−1 ] ] = det [A ] = det [I] = 1, so that det [A] ≠ 0. If A is not
invertible, the columns 𝑥1 , … . . 𝑥𝑛 of [A] are dependent (Theorem 3.5); hence
there is one, say, 𝑥𝑘 , such that
(90) 𝑥𝑘 + 𝑗 ≠𝑘 𝑐𝑗 𝑥𝑗 = 0 for certain scalars 𝑐𝑗 .
By 4.11 (b) and (d), 𝑥𝑘 can be replaced by 𝑥𝑘 + 𝑐𝑗 𝑥𝑗 without altering the
determinant, if j ≠ k. Repeating, we see that 𝑥𝑘 can be replaced by the left side

86
 Space for Hints

of (90), that is , by 0, without altering the determinant. But a matrix which has
0 for one column has determinant 0. Hence det [A] = 0.
Remark 4.14
Suppose {𝑒1 , … … . 𝑒𝑛 } and {𝑢1 , … … . 𝑢𝑛 } are bases in 𝑅 𝑛 . Every linear
operator A on𝑅 𝑛 determines matrices [A ] and [A ]u , with entries 𝑎𝑖𝑗 and 𝛼𝑖𝑗
, given by

𝐴𝑒𝑗 = 𝑎𝑖𝑗 𝑒𝑖 , 𝐴𝑢𝑗 = 𝛼𝑖𝑗 𝑢𝑖

𝑖 𝑖

If 𝑢𝑗 = 𝐵𝑒𝑗 = 𝑏𝑖𝑗 𝑒𝑖, 𝑡𝑕𝑒𝑛 𝐴𝑢𝑗 is equal to

𝛼𝑘𝑗 𝐵𝑒𝑘 = 𝛼𝑘𝑗 𝑏𝑖𝑘 𝑒𝑖 = ( 𝑏𝑖𝑘 𝛼𝑘𝑗 )𝑒𝑖

𝑘 𝑘 𝑖 𝑖 𝑘

and also to
AB𝑒𝑗 = 𝐴 𝑘 𝛼𝑘𝑗 𝑖 𝑏𝑖𝑘 𝑒𝑖 = 𝑖( 𝑘 𝑏𝑖𝑘 𝛼𝑘𝑗 )𝑒𝑖
Thus 𝑏𝑖𝑘 𝛼𝑘𝑗 = 𝑎𝑖𝑘 𝑏𝑘𝑗 , 𝑜𝑟
(91) [B][A ]u = [A ] [B].
Since B is invertible, det [B] ≠ 0. Hence (91), combined with
Theorem 4.12, shows that
(92) det [A ]u = det [A ].
The determinant of the matrix of a linear operator does therefore not depend
on the basis which is used to construct the matrix. It is thus meaningful to
speak of the determinant of a linear operator, without having any basis in
mind.

Definition(Jacobians )4.15
If f maps an open set E ⊂ 𝑅 𝑛 into 𝑅 𝑛 , and if f is differentiable at a point x∈ 𝐸
the determinant of the linear operator f '(x) is called the Jacobian of f at x.
In symbols,
(93) 𝐽𝑓 𝑥 = 𝑑𝑒𝑡𝑓′(𝑥)
we shall also use the notation
87
Space for Hints

𝜕(𝑦 1 ,……𝑦 𝑛 )
(94) 𝜕(𝑥 1 ,….𝑥 𝑛 )

For 𝐽𝑓 𝑥 if 𝑦1 , … … 𝑦𝑛 = 𝑓(𝑥1 , … . 𝑥𝑛 )
In terms of Jacobians, the crucial hypothesis in the inverse function theorem is
that𝐽𝑓 (𝑎) ≠ 0 (compare Theorem 4. 36). If the implicit function theorem is
stated in terms of the functions (59), the assumption made there on A amounts
to
𝜕(𝑓1 , … … 𝑓𝑛)
≠0
𝜕(𝑥1 , … . 𝑥𝑛 )

Definition 4.16
Suppose f is a real function defined in an open set E⊂ 𝑅 𝑛 , with partial
derivatives 𝐷1 𝑓, … . 𝐷𝑛 𝑓. If the functions 𝐷𝑖 𝑓 are themselves differentiable,
then the second-order partial derivatives of f are defined by
𝐷𝑖𝑗 𝑓 =𝐷𝑖 𝐷𝑗 𝑓 (i, j = 1 . . . , n ) .
If all these functions 𝐷𝑖𝑗 𝑓 are continuous in E, we say that f is of class ℭ in E,
or that f 𝜖 ℭ′′ (E). A mapping f of E into 𝑅 𝑚 is said to be of class ℭ′′ if each
component of f is of class ℭ ".
It can happen that 𝐷𝑖𝑗 𝑓 ≠ 𝐷𝑗𝑖 𝑓at some point, although both derivatives exist.
However, we shall see below that𝐷𝑖𝑗 𝑓 = 𝐷𝑗𝑖 𝑓 whenever these derivatives are
continuous. For simplicity (and without loss of generality) we state our next
two theorems for real functions of two variables. The first one is a mean value
theorem.

Theorem 4.17
Suppose f is defined in an open set E⊂ 𝑅 2 , and 𝐷1 𝑓and 𝐷2 𝑓exist at every
point of E. Suppose Q⊂ 𝐸 is a closed rectangle with sides parallel to the
coordinate axes, having (a, b) and (a + h, b + k) as opposite vertices (h ≠0, k ≠
0). Put

88
 Space for Hints

∆(f, Q) = f(a + h, b + k) - f(a + h, b) - f(a, b + k) + f(a, b). Then

there is a point (x, y) in the interior of Q such that
∆ (f, Q) = hk(𝐷21 𝑓)(𝑥, 𝑦))

Note the analogy between (95) and theorem 5.10; the area of q is hk.
Proof
Put u(t) = f(t, b+k) – f(t, b). two applications of theorem 5.10 show that there
is an x between and that there is a y between b and b+k such that,
∆ (f, Q) = u(a+h) – u(a)
= 𝑕𝑢′ 𝑥
= 𝑕[(𝐷1 𝑓) 𝑥, 𝑏 + 𝑘 − 𝐷(1 𝑓)(𝑥, 𝑏)
= 𝑕𝑘(𝐷21 𝑓)(𝑥, 𝑦)

Theorem 4.18
Suppose f is defined in an open set E ⊂ 𝑅 2 , suppose that 𝐷1 𝑓, 𝐷21 𝑓and
𝐷2 𝑓exist at every point of E, and 𝐷21 𝑓 is continuous at some point (a, b) ∈ 𝐸.
Then, 𝐷12 𝑓 exists at (a, b) and
(96 ) (𝐷12 𝑓) 𝑎, 𝑏 = (𝐷21 𝑓)(𝑎, 𝑏)

Corollary
(𝐷21 𝑓) =(𝐷12 𝑓) if f ∈ ℭ′′ 𝐸
Proof
Put A = (𝐷21 𝑓)(𝑎, 𝑏). Choose 𝜀 >0. If Q is a rectangle as in Theorem 4.17,
and if h and k are sufficiently small, we have
|A – (𝐷21 𝑓) 𝑥, 𝑦 | < 𝜀
For all (x, y) ∈ 𝑄. Thus,
∆ 𝑓, 𝑄
−𝐴 <𝜀
𝑕𝑘
89
Space for Hints

by (95). Fix h, and let k → 0. Since 𝐷2 𝑓 exists in E, the last inequality implies
that
𝐷2 𝑓 𝑎+𝑕, 𝑏 −( 𝐷2 𝑓(𝑎,𝑏)
(97) −𝐴 ≤𝜀
𝑕

Since, 𝜀 was arbitrary, and since (97) holds for all sufficiently small h≠
0, it follows that (𝐷12 𝑓) 𝑎, 𝑏 = 𝐴. This gives (96).

Differentiation of integrals
Suppose 𝜑 is a function of two variables which can be integrated with respect
to other. Under what conditions will the result be the same if these two limit
processes are carried out in the opposite orderA To state the question more
precisely: under what conditions on 𝜑 can one prove that the equation
𝑑 𝑏 𝑏 𝜕𝜑
(98) ∫ 𝜑 𝑥, 𝑡 𝑑𝑥 = ∫𝑎 𝑥, 𝑡 𝑑𝑥
𝑑𝑡 𝑎 𝜕𝑡

is trueA
It will be convenient to use the notation
(99) 𝜑 ′ 𝑥 = 𝜑 𝑥, 𝑡
Thus 𝜑 ′ is for each t, a function of one variable.

Theorem 4.19
Suppose
a) 𝜑 𝑥, 𝑡 𝑖𝑠 𝑑𝑒𝑓𝑖𝑛𝑒𝑑 𝑓𝑜𝑟 𝑎 ≤ 𝑥 ≤ 𝑏, 𝑐 ≤ 𝑡 ≤ 𝑑;
b) 𝛼 𝑖𝑠 𝑎𝑛 𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑖𝑛𝑔 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 𝑜𝑛 𝑎, 𝑏 ;
c) 𝜑 ′ 𝜖 ℜ 𝑥 𝑓𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑡 𝜖 𝑐, 𝑑 ;
d) 𝑐 < 𝑠 < 𝑑, 𝑎𝑛𝑑 𝑡𝑜 𝑒𝑣𝑒𝑟𝑦 𝜀 > 0 𝑐𝑜𝑟𝑟𝑒𝑠𝑝𝑜𝑛𝑑𝑠 𝑎 𝛿 > 0 𝑠𝑢𝑐𝑕 𝑡𝑕𝑎𝑡
|(𝐷2 𝜑 𝑥, 𝑡 − (𝐷2 𝜑(𝑥, 𝑠)| < 𝜀
𝑓𝑜𝑟 𝑎𝑙𝑙 𝑥 𝜖 𝑎, 𝑏 and for all t 𝜖 𝑠 − 𝛿, 𝑠 + 𝛿 .
𝐷𝑒𝑓𝑖𝑛𝑒
𝑏
(100) 𝑓 𝑡 = ∫𝑎 𝜑 𝑥, 𝑡 𝑑𝛼(𝑥) (c≤ 𝑡 ≤ 𝑑)
Then (𝐷2 𝜑)𝑠 𝜖 ℜ 𝛼 , 𝑓 ′ 𝑠 𝑒𝑥𝑖𝑠𝑡𝑠, 𝑎𝑛𝑑
90
 Space for Hints

𝑏
(101) 𝑓 ′ 𝑠 = ∫𝑎 𝐷2 𝜑 𝑥, 𝑠 𝑑𝛼 𝑥 .
Note that (c) simply asserts the existence of the integrals (100) for all t 𝜖 𝑐, 𝑑 .
Note also that (d) certainly holds whenever 𝐷2 𝜑 is continuous on the rectangle
on which 𝜑 is defined.
Proof
Consider the difference quotients
𝜑 𝑥,𝑡 −𝜑(𝑥,𝑠)
(101) 𝜓 𝑥, 𝑡 = 𝑡−𝑠

for 0 <|t-s| <𝛿. By theorem 5.10 there corresponds to each (x, t) a number u
between s and t
𝜓 𝑥, 𝑡 = 𝐷2 𝜑 𝑥, 𝑢
Hence d implies that
(102)
𝜓 𝑥, 𝑡 − (𝐷2 𝜑 𝑥, 𝑠 | < 𝜀 𝑎 ≤ 𝑥 ≤ 𝑏, 0 < 𝑡 − 𝑠 < 𝛿
Note that
𝑓 𝑡 −𝑓(𝑠) 𝑏
(103) = ∫𝑎 𝜓 𝑥, 𝑡 𝑑𝛼(𝑥)
𝑡−𝑠

By (102), 𝜓 𝑡 → (𝐷2 𝜑)𝑠 , 𝑢𝑛𝑖𝑓𝑜𝑟𝑚𝑙𝑦 𝑜𝑛 𝑎, 𝑏 , 𝑎𝑠 𝑡 → 𝑠.

Since each 𝜓 𝑡 ∈ ℜ 𝛼 , desired conclusion follows from (103).

Example 4.20
One can of course prove analogues of theorem 4.19 with (−∞, ∞) in place of
[a,b]. Instead of doing this, let us simply look at an example. Define
∞ 2
(104) 𝑓 𝑡 = ∫−∞ 𝑒 −𝑥 cos 𝑥𝑡 𝑑𝑥

∞ −2
(105) g(t) = − ∫−∞ 𝑥𝑒 −𝑥 sin 𝑥𝑡 𝑑𝑥
for −∞ < 𝑡 < ∞. Both integrals exist (they converge absolutely) since the
absolute values of the integrals are at most exp(-𝑥 2 ) and |x| exp(-𝑥 2 )
respectively

91
Space for Hints

Note that g is obtained from f y differentiating the integrand with respect to t.

we claim that f is differentiable and that
(106) f‘(t) = g(t) (−∞ < 𝑡 < ∞)
To prove this, let us first examine the difference quotients of the cosine: if
𝛽 > 0, then
𝐶𝑜𝑠 𝛼 +𝛽 −𝑐𝑜𝑠𝛼 1 𝛼+𝛽
(107) + 𝑠𝑖𝑛𝛼 = ∫𝛼 (𝑠𝑖𝑛𝛼 − 𝑠𝑖𝑛𝑡)𝑑𝑡
𝛽 𝛽

Since |sin𝛼 − 𝑠𝑖𝑛𝑡| ≤ 𝑡 − 𝛼 , the right side of (107) is at most 𝛽/2 in

absolute value; the case 𝛽 < 0 is handled similarly. Thus
𝐶𝑜𝑠 𝛼 +𝛽 −𝑐𝑜𝑠𝛼
(108) + 𝑠𝑖𝑛𝛼 ≤ 𝛽
𝛽

for all 𝛽 (if the left side is interpreted to be 0 when 𝛽 = 0)

Now fix t, and fix h ≠ 0. Apply (108) with
𝛼 = 𝑥𝑡, 𝛽 = 𝑥𝑕; 𝑖𝑡 𝑓𝑜𝑙𝑙𝑜𝑤𝑠 𝑓𝑟𝑜𝑚 104 𝑎𝑛𝑑 105 𝑡𝑕𝑎𝑡
∞
𝑓 𝑡+𝑕 −𝑓 𝑡 2
−𝑔 𝑡 ≤ |𝑕| 𝑥 2 𝑒 −𝑥 𝑑𝑥
𝑕 −∞

When h → 0, we thus obtain (106).

Let us go a step further: an integration by parts, applied to (104), shows that
∞ 2 sin ⁡
(𝑥𝑡 )
(109) f(t) = 2 ∫−∞ 𝑥𝑒 −𝑥 𝑑𝑥
𝑡

Thus t f(t) = -2 g(t), and (106) implies now that f satisfies the differential
equation
(110) 2f‘(t) + tf(t) = 0
If we solve this differential equation and use the fact that
f(0) = √𝜋 (𝑠𝑒𝑒 sec 8.21), we find that
𝑡2
(111) f(t) = √𝜋 exp − 4

The integral (104) is thus explicitly determined.

CYP QUESTIONS:
1. Suppose 𝑓 is a differentiable mapp ng of ℝ1 into ℝ3 such that
𝑓 𝑡 = 1 for every 𝑡.Prove that 𝑓′(𝑡) · 𝑓(𝑡 ) = 0.

92
 Space for Hints

2. Take n = m = 1 in the implici t function theorem, and interpret the

theorem (aswell as its proof) graphically.

93
Space for Hints

UNIT – 5
INTEGRATION OF DIFFERENTIAL FORMS
Unit Structure:
Section 5.1: Integrations of Differential forms
Section 5.2: Primitive mappings
Section 5.3: partition of unity
Section5.4: change of variables
Section 5.5: differential forms
Section 5.6: Simplexes and chains
Section 5.7: Stoke‘s theorem

Intcgration can be studied on many levels. The present unit is devoted to

those aspects of integration theory that are closely related to thegeometry of
euclidean spaces, such as the change of variables formula , line integrals, and
the machinery of differential forms that is used in the statement and proof of
the n-dimensional analogue of the fundamental theorem of calculus, namely
Stokes' theorem.

5.1 Integrations of Differential forms

Definition 5.1
Suppose 𝐼 𝑘 is a k-cell in ℝ𝑘 , consisting of all
𝑥 = (𝑥1 , … , 𝑥𝑘 )

such that
𝑎𝑖 ≤ 𝑥𝑖 ≤ 𝑏𝑖 (𝑖 = 𝐼, . . . , 𝑘), (1 )

𝐼 𝑗 is the 𝑗 −ce11 in 𝑅𝑖 defined by the first 𝑗 inequalities (I), and 𝑓 is a real

continuous function on 𝐼 𝑘 .
Put𝑓 = 𝑓𝑘 , and define𝑓𝑘−1 on 𝐼 𝑘−1 by

𝑏𝑘
𝑓𝑘−1 𝑥1 , … , 𝑥𝑘−1 = 𝑓𝑘 𝑥1 , … , 𝑥𝑘−1 , 𝑥𝑘 𝑑𝑥𝑘 .
𝑎𝑘
The uniform continuity of 𝑓𝑘 on 𝐼 𝑘 shows that 𝑓𝑘−1 is continuous on
𝐼 𝑘−𝑡 .Hence we can repeat this process and obtain functions𝑓𝑗 , continuous on
𝐼 𝑗 such that 𝑓𝑗 −1 is the integral of with respect to 𝑥𝑗 , over [𝑎𝑗 , 𝑏𝑗 ]. After 𝑘 steps
we arrive at a number 𝑓0 , which we call the integral of𝑓 over 𝐼 𝑘 ;we write it in
the form
94
 Space for Hints

∫𝐼 𝑘 𝑓(𝑥) 𝑑𝑥 or ∫𝐼 𝑘 𝑓. (2)
A priori, this definition of the integral depends on the order in which the 𝑘
integrations are carried out. However, this dependence is only apparent. To
prove this, let us introduce the temporary notation 𝐿(𝑓) for the integral and
𝐿′(𝑓) for the result obtained by carrying out the 𝑘 integrations in some other
order.

Theorem 5.2
For every 𝑓 ∈ 𝒞(𝐼 𝑘 ), 𝐿(𝑓) = 𝐿′(𝑓).
Proof If 𝑕 𝑥 = 𝑕1 𝑥1 … 𝑕𝑘 (𝑥𝑘 ), where 𝑕𝑖 ∈ 𝒞([𝑎𝑗 , 𝑏𝑗 ]), then

𝑘 𝑏𝑖
𝐿(𝑕) = 𝑕𝑖 𝑥𝑖 𝑑𝑥𝑖 = 𝐿′(𝑕).
𝑖=1 𝑎𝑖

If 𝒜 is the set of all finite sums of such functions 𝑕, it follows that 𝐿 𝑔 =

𝐿′(𝑔) for all 𝑔 ∈ 𝒜.Also, 𝒜 is an algebra of functions on 𝐼 𝑘 to which the
Stone-Weierstrass theorem applies.

Put 𝑉 = 𝑘𝐼 𝑏𝑖 − 𝑎𝑖 . If 𝑓 ∈ 𝒞(𝐼 𝑘 ) and 𝜖 > 0, there exists 𝑔 ∈ 𝒜 such

𝜖
that‖𝑓 − 𝑔‖ < 𝑉 , where ‖𝑓‖ is defined as 𝑚𝑎𝑥 |𝑓 𝑥 | (𝑥 ∈ 𝐼 𝑘 ).Then
𝐿 𝑓 − 𝑔 < 𝜖, 𝐿′ 𝑓 − 𝑔 < 𝜖,and since
𝐿(𝑓) − 𝐿′(𝑓) = 𝐿(𝑓 − 𝑔) + 𝐿′(𝑔 − 𝑓),
we conclude that 𝐿 𝑓 − 𝐿′ 𝑓 < 2𝜖.

Definition 5.3
The support of a (real or complex) function 𝑓 on 𝑅 𝑘 is the closure of the set of
all points 𝑥 ∈ 𝑅 𝑘 at which 𝑓 𝑥 ≠ 0. If 𝑓 is a continuous function with
compact support, let 𝐼 𝑘 be any 𝑘 −cell which contains the support of𝑓, and
define

∫𝑅 𝑘 𝑓 = ∫𝐼 𝑘 𝑓. (3)

The integral so defined is evidently independent of the choice of𝐼 𝑘 , provided

only that 𝐼 𝑘 contains the support of𝑓.
It is nwtempting to extend the definition of the integral over ℝ𝑘 to
functions which are li mits (in son1e sense) of continuous functions with
compact support. We do not want to discuss the conditions under which this
can be done : the proper setting for this question is the Lebesgue integral . We
shall merely describe one very simple example which will be used in the
proof of Stokes' theorem.

95
Space for Hints

Example 5.4
Let ℚ𝑘 be the 𝑘 − simp1ex which consists of all points
𝑥 = (𝑥1 , 𝑥2 , … , 𝑥𝑘 )in 𝑅 𝑘 for which 𝑥1 + ⋯ + 𝑥𝑘 ≤ 1 and 𝑥𝑖 ≥
𝑘
0for 𝑖 = 1, … , 𝑘. If𝑘 = 3,for example, ℚ is a tetrahedron, with vertices
at 0, 𝑒1 , 𝑒2 , 𝑒3 . If 𝑓 ∈ 𝒞(ℚ𝑘 ), extend 𝑓 to a function on 𝐼 𝑘 by setting 𝑓(𝑥) =
0 off ℚ𝑘 , and define
∫ℚ𝑘 𝑓 = ∫𝐼 𝑘 𝑓 . (4)
Here 𝐼 𝑘 is the ―unit cube‖ defined by
0 ≤ 𝑥𝑖 ≤ 1 1≤𝑖≤𝑘 .
Since 𝑓 may be discontinuous on 𝐼 𝑘 , the existence of the integral on the left of
(4) needs proof. We also wish to show that this integral is independent of the
order in which the 𝑘 single integrations are carried out.
To do this. suppose0 < 𝛿 < 1 put
1 (𝑡 ≤ 1 − 𝛿))
1−𝑡
𝜙 𝑡 = (1 − 𝛿 < 𝑡 ≤ 1) (5)
𝛿
0 1<𝑡 ,

and define
𝐹 𝑥 = 𝜙 𝑥1 + ⋯ + 𝑥𝑘 𝑓 𝑥 𝑥 ∈ 𝐼 𝑘 . (6)
Then𝐹 ∈ 𝒞 𝐼 𝑘 .
Put 𝑦 = (𝑥1 , . . . , 𝑥𝑘−1 ), 𝑥 = (𝑦, 𝑥𝑘 ). For each ∈ 𝐼 𝑘−1 , , the set of all 𝑥𝑘 such
that 𝐹 𝑦, 𝑥𝑘 ≠ 𝑓(𝑦; 𝑥𝑘 ) is either empty or is a segment whose length
does not exceed 𝛿.Since0 ≤ 𝜙 ≤ 1, it follows that
𝐹𝑘−1 𝑦 − 𝑓𝑘−1 𝑦 ≤ 𝛿‖𝑓‖ 𝑦 ∈ 𝐼 𝑘−1 , (7)

where‖𝑓‖ has the same meaning as in the proof of above Theorem

As𝛿 → 0 (7) exhibits 𝑓𝑘−1 as a uniform limit of a sequence of continuous

functions. Thus 𝑓𝑘−1 ∈ 𝒞(𝐼 𝑘−1 ), and the further integrations present no
problem.

This proves the existence of the integral (4). Moreover, (7) shows that

∫𝐼 𝑘 𝐹(𝑥) 𝑑𝑥 − ∫𝐼 𝑘 𝑓(𝑥) 𝑑𝑥 ≤ 𝛿‖𝑓‖. (8)

Note that (8) is true, regardless of the order in which the 𝑘single integrations
are carried out. Since𝐹 ∈ 𝒞(𝐼 𝑘 ), ∫ 𝐹 is unaffected by any change in this order.
Hence (8) shows that the same is true of∫ 𝑓.
This completes the proof.
96
 Space for Hints

Our next goal is the change of variables formula. To facilitate its proof, we
first discuss so-called primitive mappings, and partitions of unity. Primitive
mappings will enable us to get a clearer picture of the local action of a
𝒞 −mapping with invertible derivative, and partitions of unity are a very
useful device that makes it possible to use local informat ion in a global
setting.

5.2 Primitive mappings

Definition 5.4
If G maps an open set 𝐸 ∈ ℝ𝑛 into 𝑅 𝑛 , and if there is an integer 𝑚 and a real
function 𝑔 with domain 𝐸 such that
𝐺 𝑥 = Σ𝑖≠𝑚 𝑥𝑖 𝑒𝑖 + 𝑔 𝑥 𝑒𝑚 (𝑥 ∈ 𝐸),
then we call𝐺primitive. A primitive mapping is thus one that changes at most
one coordinate. Note that (9) can also be written in the form
𝐺(𝑥) = 𝑥 + 𝑔 𝑥 − 𝑥𝑚 𝑒𝑚 . (10)
If 𝑔 is differentiable at some point 𝑎 ∈ 𝐸, £, so is 𝐺.The matrix [𝛼𝑖𝑗 ] of the
operator 𝐺′(𝑎) has
𝐷1 𝑔 𝑎 , . . . , (𝐷𝑚 𝑔)(𝑎), . . . , ( 𝐷𝑛 𝑔)(𝑎) (11)
as its 𝑚 th row. For 𝑗 ≠ 𝑚, we have 𝛼𝑗𝑗 = 1 and 𝛼𝑖𝑗 = 0 0 if 𝑖 ≠ 𝑗. The
Jacobian of 𝐺 at 𝑎 is thus given by
𝐽𝐺 (𝑎) = 𝑑𝑒𝑡[𝐺′(𝑎)] = (𝐷𝑚 𝑔)(𝑎), (12)
and we know that 𝐺′(𝑎) is invertible ifand only if 𝐷𝑚 𝑔 𝑎 ≠ 0.

Definition 5.5
A linear operator 𝐵 on ℝ𝑛 that interchanges some pair of members of the
standard basis and leaves the others fixed will be called a flip.
For example, the flip 𝐵 on ℝ4 that interchanges 𝑒 2 and 𝑒 4 has the form
𝐵(𝑥1 𝑒1 + 𝑥2 𝑒2 + 𝑥3 𝑒3 + 𝑥4 𝑒4 ) = 𝑥1 𝑒1 + 𝑥2 𝑒4 + 𝑥3 𝑒3 + 𝑥4 𝑒2 (13)
or, equivalently,
𝐵(𝑥1 𝑒1 + 𝑥2 𝑒2 + 𝑥3 𝑒3 + 𝑥4 𝑒4 ) = 𝑥1 𝑒1 + 𝑥4 𝑒2 + 𝑥3 𝑒3 + 𝑥2 𝑒4 . (14)
Hence 𝐵 can also be thought of asinterchanging two of the coordinates, rather
than two basis vectors.
In the proof that follows, we shall use the projections 𝑃0 , 𝑃1 , … , 𝑃𝑛 ,in ℝ𝑛 ,,
defined by 𝑃0 𝑥 = 0 and
𝑃𝑚 𝑥 = 𝑥1 𝑒1 + 𝑥2 𝑒2 + ⋯ + 𝑥𝑚 𝑒𝑚 (1 5)
for 1 ≤ 𝑚 ≤ 𝑛. Thus 𝑃𝑚 is the projection whose range and nullspace are
spanned by {𝑒1 , … , 𝑒𝑚 } and { 𝑒𝑚+1 , … , 𝑒𝑛 }respectively.

97
Space for Hints

Theorem 5.6
Suppose F is a 𝒞-mapping of an open set 𝐸 ⊂ ℝ𝑛 into ℝ𝑛 ,0 ∈ 𝐸,
𝐹 0 = 0, and 𝐹 ′ (0) is in rertible.
Then there is a neighborhood of· 0 in R'' in which a representation

(16)𝐹 𝑥 = 𝐵1 … . 𝐵𝑛−1 𝐺𝑛 … . . 𝐺1 (𝑥)

Is Valid.
In (16), each 𝐺𝑖 is a primitive 𝒞-mapping in some neighborhood of
0: 𝐺𝑖 0 = 0, 𝐺𝑖 0 is invertible, and each 𝐵𝑖 is either a flip or the identity
operator.
Briefly, ( 16) represents F locally as a composition of primitive mappings and
flips.
Proof Put 𝐹 = 𝐹1 . Assume 1 ≤ 𝑚 ≤ 𝑛 − 1 , and make the following
induction hypothesis (which evidently holds for 𝑚 = 1):
𝑉𝑚 is a neighborhood of 0, 𝐹𝑚 ∈ 𝒞 𝑉𝑚 , 𝐹𝑚 0 = 0, 𝐹 ′ 𝑚 0 is invcrtible,
and
(1 7 ) 𝑃𝑚 −1 𝐹𝑚 𝑥 = 𝑃𝑚 −1 𝑥 (𝑥 ∈ 𝑉𝑚 )
By ( 1 7), we have

𝑛
(18)𝐹𝑚 𝑥 = 𝑃𝑚 −1 𝑥 + 𝑖=𝑚 𝛼𝑖 𝑥 𝑒𝑖 ,

where 𝛼𝑚 , … , 𝛼𝑛 are real 𝒞'-functions in 𝑉𝑚 . Hence

𝑛
(19)𝐹 ′ 𝑚 0 𝑒𝑚 = 𝑖=𝑚 (𝐷𝑚 𝛼𝑖 ) 0 𝑒𝑖 .

Since 𝐹′𝑚 (0) is invertible, the left side of (19) is not 0, and therefore there is
a 𝑘 such that 𝑚 ≤ 𝑘 ≤ 𝑛 and 𝐷𝑚 𝛼𝑘 0 ≠ 0.
Let 𝐵𝑚 be the flip that interchanges 𝑚 and this 𝑘 (if 𝑘 = 𝑚, 𝐵𝑚 is the identity)
and define

(20) 𝐺𝑚 𝑥 = 𝑥 + 𝛼𝑘 𝑥 − 𝑥𝑚 𝑒𝑚 𝑥 ∈ 𝑉𝑚 .

Then𝐺𝑚 ∈ 𝒞 ′ (𝑉𝑚 ),𝐺𝑚 is primitive, and 𝐺 ′ 𝑚 (0) is invertible, Since

𝐷𝑚 𝛼𝑘 0 ≠ 0.
The inverse function theorem shows therefore that there is an open
set 𝑈𝑚 , with 0 ∈ 𝑈𝑚 ⊂ 𝑉𝑚 , such that 𝐺𝑚 1 is a 1-1 mapping of 𝑈𝑚 onto a
neighborhood 𝑉𝑚 +1 of 0, in which𝐺𝑚−1 is continuously differentiable.
Define 𝐹𝑚 +1 by
(21 ) 𝐹𝑚 +1 𝑦 = 𝐵𝑚 𝐹𝑚 ° 𝐺𝑚−1 (𝑦) (𝑦 ∈ 𝑉𝑚 +1 )
98
 Space for Hints

Then 𝐹𝑚 +1 ∈ 𝒞 ′ 𝑉𝑚+1 , 𝐹𝑚 +1 0 = 0, and 𝐹 ′ 𝑚 +1 0 is invertible (by the

chain rule) .Also, for 𝑥 ∈ 𝑈𝑚 ,

(22) 𝑃𝑚 𝐹𝑚 +1 𝐺𝑚 𝑥 = 𝑃𝑚 𝐵𝑚 𝐹𝑚 (𝑥)
= 𝑃𝑚 𝑃𝑚 −1 𝑥 + 𝛼𝑘 𝑥 𝑒𝑚 + ⋯
= 𝑃𝑚 −1 𝑥 + 𝛼𝑘 𝑥 𝑒𝑚
= 𝑃𝑚 𝐺𝑚 (𝑥)
so that
(23) 𝑃𝑚 𝐹𝑚 +1 𝑦 = 𝑃𝑚 𝑦 𝑦 ∈ 𝑉𝑚 +1 .
Our induction hypothesis holds therefore with 𝑚 + 1in place of 𝑚.
[In (22), we first used (21 ), then (18) and the definition of 𝐵𝑚 , then the
definition of 𝑃𝑚 and finally (20).]
Since 𝐵𝑚 𝐵𝑚 = I, (21), with 𝑦 = 𝐺𝑚 𝑥 is equivalent to
(24) 𝐹𝑚 𝑥 = 𝐵𝑚 𝐹𝑚 +1 (𝐺𝑚 𝑥 )𝑥 ∈ 𝑈𝑚 .
If we apply this with 𝑚 = 1, … , 𝑛 − 1,we successively obtain
𝐹 = 𝐹1 = 𝐵1 𝐹2 ° 𝐺1
= 𝐵1 𝐵2 𝐹3 ° 𝐺2 ° 𝐺1 = …
= 𝐵1 … 𝐵𝑛−1 𝐹𝑛 ° 𝐺𝑛−1 ° … ° 𝐺1

in some neighborhood of 0. By (17), 𝐹𝑛 is primitive. This completes the

proof.

5.3 Partition of unity

Theorem 5.7

Suppose 𝐾 is a compact subset of 𝑅 𝑛 , and 𝑉𝛼 is an open cove of 𝐾. Then

there exist functions 𝜓1 , . . . , 𝜓𝑠 ∈ 𝒞 𝑅 𝑛 such that

(a) 0 ≤ 𝜓𝑖 ≤ 1 𝑓𝑜𝑟 1 ≤ 𝑖 ≤ 𝑠;

(b) each 𝜓𝑖 has its support in some 𝑉𝛼 , and

(c) 𝜓1 𝑥 + ⋯ + 𝜓𝑠 𝑥 = 1 𝑓𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑥 ∈ 𝐾.

Because of (c), 𝜓𝑖 is called a partition of unity, and (b) is sometimes

expressed by saying that 𝜓𝑖 is subordinate to the cover 𝑉𝛼 .

Corrollary
99
Space for Hints

If 𝑓 ∈ 𝒞 𝑅 𝑛 and the support of 𝑓 lies in 𝐾, then

𝑠
𝑓= 𝑖=1 𝜓𝑖 𝑓 (25)

Each 𝜓𝑖 𝑓 has its support in some 𝑉𝛼 .

The point of (25) is that it furnishs a representation of 𝑓 as a sum of

continuous functions 𝜓𝑖 𝑓 ―small‖ supports.

Proof Associate with each 𝑥 ∈ 𝐾 an index 𝛼 𝑥 so that 𝑥 ∈ 𝑉𝛼(𝑥) . Then there

are open balls 𝐵 𝑥 and 𝑊 𝑥 , centered at 𝑥, with

𝐵 𝑥 ⊂ 𝑊 𝑥 ⊂ 𝑊 𝑥 ⊂ 𝑉𝛼(𝑥) (26)

Since 𝐾 is compact, there are points 𝑥1 , … … , 𝑥𝑠 in 𝐾 such that

𝐾 ⊂ 𝐵 𝑥1 ∪ … ∪ 𝐵 𝑥𝑠 (27)

By (26), there are functions 𝜓1 , . . . , 𝜓𝑠 ∈ 𝒞 𝑅 𝑛 , such that 𝜑𝑖 𝑥 = 1 on

𝐵 𝑥𝑖 , 𝜑𝑖 𝑥 = 0 outside 𝑊 𝑥𝑖 , and 0 ≤ 𝜑𝑖 𝑥 ≤ 1on 𝑅 𝑛 . Define 𝜓1 = 𝜑1
and

𝜓𝑖+1 = 1 − 𝜑1 … 1 − 𝜑𝑖 𝜑𝑖+1 (28)

for 𝑖 = 1, … , 𝑠 − 1.

Properties (a) and (b) are clear. The relation

𝜓1 + ⋯ + 𝜓𝑠 = 1 − 1 − 𝜑1 … 1 − 𝜑𝑖 (29)

is trivial for 𝑖 = 1. If (29) holds for some 𝑖 < 𝑠, addition of (28) and (29)
yields (29) with 𝑖 + 1 in place of 𝑖. It follows that

𝑠 𝑠
𝑖=1 𝜓𝑖 𝑥 = 1− 𝑖=1 1 − 𝜑𝑖 𝑥 (30)

100
 Space for Hints

If 𝑥 ∈ 𝐾, then 𝑥 ∈ 𝐵 𝑥𝑖 for some 𝑖, hence 𝜑𝑖 𝑥 = 1, and the product in (30)

is 0. This proves (c).

5.4 Change of variables

We can now describe the effect of a change of variables on a multiple integral.

For simplicity, we confine ourselves here to continuous functions with
compact support, although this is too restrictive for many applications. This is
illustrated by Exerciese 9 to 13.

Theorem 5.8

Suppose 𝑇 is a 1-1 𝒞 - mapping of an open set 𝐸 ⊂ 𝑅 𝑘 into 𝑅 𝑘 such that

𝐽𝑇 𝑥 ≠ 0 for all 𝑥 ∈ 𝐸. If 𝑓 is a continuous function on 𝑅 𝑘 whose support is
compact and lies in 𝑇 𝐸 , then

∫𝑅 𝑘 𝑓 𝑦 𝑑𝑦 = ∫𝑅 𝑘 𝑓 𝑇 𝑥 𝐽𝑇 𝑥 𝑑𝑥 (31)

We recall that 𝐽𝑇 is the Jacobian of 𝑇. The assumption 𝐽𝑇 𝑥 ≠ 0 implies, by

the inverse function theorem, that 𝑇 −1 is continuous on 𝑇 𝐸 , and this ensures
that the integrand on the right of (31) has compact support in 𝐸 (Theorem
4.14).

∫𝑅 𝑘 𝑓 𝑦 𝑑𝑦 = ∫𝑅 𝑘 𝑓 𝑇 𝑥 𝑇 ′ 𝑥 𝑑𝑥 (32)

by Theorem 6.19 and 6.17, for all continuous 𝑓 with compact support. But if 𝑇
decreases, then 𝑇 ′ 𝑥 < 0; and if 𝑓 is positive in the interior of its support, the
left side of (32) is positive and the right side is negative. A correct equation is
obtained if 𝑇 ′ is replaced by 𝑇 ′ in (32).

The point is that the integrals we are now considering are integrals of
functions over subsets of 𝑅 𝑘 , and we associate no direction or orientation with

101
Space for Hints

these subsets. We shall adopt a different point of view when we come to

integration of differential forms over surface.

Proof It follows form the remarks just made that (31) is true if 𝑇 is a primitive
𝒞- mapping (see Definition 10.5), and (31) is true if 𝑇 is a linear mapping
which merely interchanges two coordinates.

If the theorem is true for transformations 𝑃, 𝑄 and if 𝑆(𝑥0 = 𝑃 𝑄 𝑥 ,

then

∫ 𝑓 𝑧 𝑑𝑧 = ∫ 𝑓 𝑃 𝑦 𝐽𝑃 𝑦 𝑑𝑦

= ∫𝑓 𝑃 𝑄 𝑥 𝐽𝑃 𝑄 𝑥 𝐽𝑄 𝑥 𝑑𝑥

= ∫𝑓 𝑆 𝑦 𝐽𝑆 𝑥 𝑑𝑥

𝐽𝑄 𝑄 𝑋 𝐽𝑄 𝑥 = det 𝑃′ (𝑄 𝑥 ) det 𝑄 ′ 𝑥

= det 𝑃′ 𝑄 𝑥 𝑄′ 𝑥 = det 𝑆 ′ 𝑥 = 𝐽𝑆 𝑥

by the multiplication theorem for determinants and the chain rule. Thus the
theorem is also ture for 𝑆.

𝑇 𝑥 = 𝑇 𝑎 + 𝐵1 … 𝐵𝑘−1 𝐺𝑘 , 𝐺𝑘−1 … , 𝐺1 𝑥 − 𝑎 (33)

where 𝐺𝑖 and 𝐵𝑖 are as in Theorem 10.7. Setting 𝑉 = 𝑇 𝑈 , it follows that

(31) holds if the support of 𝑓 lies in 𝑉. Thus:

Each point 𝑦 ∈ 𝑇 𝐸 lies in an open set 𝑉𝑦 ⊂ 𝑇 𝐸 such that (31) holds

for a ll continuous functions whose support lies in 𝑉𝑦 .

Now let 𝑓 be continuous function with compact support 𝐾 ⊂ 𝑇 𝐸 .

Since 𝑉𝑦 covers 𝐾 , the Corollary to Theorem 5.7shows that 𝑓 = 𝜓𝑖 𝑓 ,

102
 Space for Hints

where each 𝜓𝑖 is continuous, and each 𝜓𝑖 has its support in some 𝑉𝑦 . Thus
(31) holds for each 𝜓𝑖 𝑓 and hence also for their sum 𝑓.

5.5 Differential forms

We shall now develop some of the machinery that is needed for the 𝑛 -
dimensional version of the fundamental theorem of calculus which is usually
called Stoke‘s theorem. The original form Stokes theorem arose in application
of vector analysis to electromagnetism and was stated in terms of the curl of a
vector field. Green‘s theorem and the divergence theorem are other special
cases.

It is a curious feature of Stokes theorem that the only thing that is

difficult about it is the elaborate structure of definitions that are needed for its
statement. These definitions concern differential forms, their derivatives,
boundaries and orientation. Once these concepts are understood, the statement
of the theorem is very brief and succinct, and its proof presents little
difficulty.

Up to now we have considered derivatives of functions of several

variables only for functions defined in open sets. This was done to avoid
diffuclties that can occur at aboundary points. It will now be convenient,
however, to discuss differentiable functions on compact sets. We therefore
adopt the following convention:

To say that 𝑓 is a 𝒞 ′ - mapping (or a 𝒞 ′′ - mapping) of a compact set

𝐷 ⊂ 𝑅 𝑘 into 𝑅 𝑛 means that there is a 𝒞 ′ - mapping (or a 𝒞 ′′ - mapping)
𝑔 𝑥 = 𝑓 𝑥 for all 𝑥 ∈ 𝐷.

Definition 5.9

103
Space for Hints

Suppose 𝐸 is an open set in 𝑅 𝑛 . A 𝑘-surface in 𝐸 is 𝒞 ′ - mapping Φ

from a compact set 𝐷 ⊂ 𝑅 𝑘 into 𝐸.

𝐷 is called the parameter domain of Φ. Points of 𝐷 will be denoted by

𝑢 = 𝑢1 , … , 𝑢𝑘 .

We shall confine ourselves to the simple situation in which 𝐷 is either

a 𝑘-cell or the 𝑘-simplec 𝑄 𝑘 described in Example 10.4. The reason for this is
that we shall have to integrate over 𝐷 , and we have not yet discussed
integration over more complicated subsets of 𝑅 𝑘 . It will be seen that this
restriction on 𝐷 (which will be tacitly made form now on) entails no
significant loss of generality in the resulting theory of differential forms.

We stress that 𝑘-surfaces in 𝐸 are defined to mappings into 𝐸 , not

subsets of 𝐸 . This agrees with our earlier definition of curves (Definition
6.26). In fact, 1-surfaces are precisely the same as continuously differentiable
curves.

5.10 Definition Suppose 𝐸 is open set in 𝑅 𝑛 . A differential form of

order 𝑘 ≥ 1 in 𝐸 (briefly, a 𝑘 -form in 𝐸 ) is a functions 𝜔 , symbolically
represented by the sum

𝜔= 𝑎𝑖1 ,…𝑖𝑘 𝑥 𝑑𝑥𝑖 ⋀ … ⋀𝑑𝑥𝑖𝑘 (34)

(the indices 𝑖1 , … 𝑖𝑘 range independently from 1 to 𝑛), which assigns to each

𝑘-surface Φ in 𝐸 a number 𝜔 Φ = ∫Φ 𝜔, according to the rule

𝜕 𝑥 𝑖 1 ,…,𝑥 𝑖 𝑘
∫Φ 𝜔 = ∫D 𝑎𝑖1 ,…𝑖𝑘 Φ 𝑢 𝜕 𝑢 1 ,…,𝑢 𝑘
𝑑𝑢 (35)

where 𝐷 is the parameter domain of Φ.

104
 Space for Hints

The functions 𝑎𝑖1 ,…𝑖𝑘 are assumed to bea real contionuous 𝐸 . If

𝜙1 , … , 𝜙𝑛 are the components of Φ, the Jacobian in (35) is the one determined
by the mapping

𝑢1 , … , 𝑢𝑘 → 𝜙𝑖1 𝑢 , … , 𝜙𝑖𝑘 𝑢

Note that the right side of (35) is an integral over 𝐷, as defined in

Definition 10.1 (or Example 10.4) and that (35) is the definition of the symbol

∫Φ 𝜔.

A 𝑘-form 𝜔 is said to be class 𝒞 ′ or 𝒞 ′′ if the function 𝑎𝑖1 ,…𝑖𝑘 in (34)

are all of class 𝒞 ′ or 𝒞 ′′ .

A 0-form in 𝐸 is defined to be continuous function in 𝐸.

5.11 Examples

(a) Let 𝛾 be a 1-surface (a curve of class 𝒞 ′ ) in 𝑅 3 , with parameter

domain 0,1 .

Write 𝑥, 𝑦, 𝑧 in place of 𝑥1 , 𝑥2 , 𝑥3 , and put

𝜔 = 𝑥𝑑𝑦 + 𝑦𝑑𝑥.

Then

𝜔= 𝛾1 𝑡 𝛾2′ 𝑡 + 𝛾1 𝑡 𝛾1′ 𝑡 𝑑𝑡 = 𝛾1 1 𝛾2 1 − 𝛾1 0 𝛾2 0
𝛾 0

Note that in this example ∫𝛾 𝜔 depends only on the initial point 𝛾 0

and on the end point 𝛾 1 of 𝛾. In particular ∫𝑟 𝜔 = 0 for every closed curve

𝛾. (As we shall see later, this is true for every 1-form 𝜔 which is exact.)

105
Space for Hints

Integrals of 1-forms are often called line integrals.

(b) Fix 𝑎 > 0, 𝑏 > 0 and define

𝛾 𝑡 = 𝑎 cos 𝑡 , 𝑏 sin 𝑡 0 ≤ 1 ≤ 2𝜋

so that 𝛾 is a closed curve in 𝑅 2 . (Its range is an ellipse). Then

2𝑥

𝑥𝑑𝑦 = 𝑎𝑏 cos2 𝑡 𝑑𝑡 = 𝜋𝑎𝑏

𝛾 0

whereas

2𝑥

𝑦𝑑𝑥 = − 𝑎𝑏 sin2 𝑡 𝑑𝑡 = −𝜋𝑎𝑏

𝛾 0

Note that ∫𝛾 𝑥𝑑𝑦 is the area of the region bounded by 𝛾. This is a

special case of Green‘s theorem.

(c) Let 𝐷 be the 3-cell defined by

0 ≤ 𝑟 ≤ 1, 0 ≤ 𝜃 ≤ 𝜋, 0 ≤ 𝜑 ≤ 2𝜋

Define Φ 𝑟, 𝜃, 𝜑 = 𝑥, 𝑦, 𝑧 , where

𝑥 = 𝑟 sin 𝜃 cos 𝜑

𝑦 = 𝑟 sin 𝜃 sin 𝜑

𝑧 = 𝑟 cos 𝜃

Then

𝜕 𝑥, 𝑦, 𝑧
𝐽Φ 𝑟, 𝜃, 𝜑 = = 𝑟 2 sin 𝜃
𝜕 𝑟, 𝜃, 𝜑

106
 Space for Hints

Hence

4𝜋
∫Φ 𝑑𝑥⋀𝑑𝑦⋀𝑑𝑧 = ∫D 𝐽Φ = 3
(36)

Noete that Φ maps 𝐷 onto the closed unit ball of 𝑅 3 , that the mapping
is 1-1 in the interior of 𝐷 (but certain boundary points are identified by Φ),
and that the integral (36) is equal to the volue of Φ D .

Proof Let 𝐷 be the parameter domain of Φ (hence also of 𝑇Φ) and define ∆.

𝜔= 𝜔TΦ = 𝜔T Φ = 𝜔T
TΦ ∆ ∆ Φ

5.6 Simplexes and chains

Definition(Affine simplexes)5.12

A mapping 𝑓 that carries a vector space 𝑋 into a vector space 𝑌 is said

to be affine if 𝑓 − 𝑓 0 is linear. In other words, the requirement is that

𝑓 𝑥 = 𝑓 0 + 𝐴𝑥 (73)

for some 𝐴 ∈ 𝐿 𝑋, 𝑌 .

An affine mapping of 𝑅 𝑘 into 𝑅 𝑛 is thus determined if we know

𝑓 0 and 𝑓 𝑒𝑖 for 1 ≤ 𝑖 ≤ 𝑘; as usual, 𝑒1 , … , 𝑒𝑘 is the standard basis of 𝑅 𝑘 .

We define the standard simplex 𝑄 𝑘 to be the set of all 𝑢 ∈ 𝑅 𝑘

of the form

𝑘
𝑢= 𝑖=1 𝛼𝑖 𝑒𝑖 (74)

107
Space for Hints

such that 𝛼𝑖 ≥ 0 for 𝑖 = 1, … , 𝑘 and 𝛼𝑖 ≤ 1.

Assume now that 𝑃0 , 𝑃1 , … , 𝑃𝑘 are points of 𝑅 𝑛 . The oriented

affine 𝑘-simplex

𝜎 = 𝑃0 , 𝑃1 , … , 𝑃𝑘 (75)

is defined to be the 𝑘-surface in 𝑅 𝑛 with parameter domain 𝑄 𝑘 which

is give by the affine mapping

𝑘
𝜎 𝛼1 𝑒1 + ⋯ + 𝛼𝑘 𝑒𝑘 = 𝑃0 + 𝑖=1 𝛼𝑖 𝑃𝑖 − 𝑃0 (76)

Note that 𝜎 is characterized by

𝜎 0 = 𝑃0 𝜎 𝑒𝑖 = 𝑃𝑖 (for 1 ≤ 𝑖 ≤ 𝑘) (77)

and that

𝜎 𝑢 = 𝑃0 + 𝐴𝑢 (𝑢 ∈ 𝑄 𝑘 ) (78)

where 𝐴 ∈ 𝐿 𝑅 𝑘 , 𝑅 𝑛 and 𝐴𝑒𝑖 = 𝑃𝑖 − 𝑃0 for 1 ≤ 𝑖 ≤ 𝑘.

We call 𝜎 oriented to emphasize that the ordering of the

vertices 𝑃0 , … , 𝑃𝑘 is taken into account. If

𝜎 = 𝑃𝑖0 , 𝑃𝑖1 , … , 𝑃𝑖𝑘 (79)

where 𝑖0 , 𝑖1 , … , 𝑖𝑘 is a permutation of the ordered set 0,1, … , 𝑘 , we

adopt the notation

𝜎 = 𝑠 𝑖0 , 𝑖1 , … , 𝑖𝑘 𝜎 (80)

where 𝑠 is the function defined in Definition 4.11 Thus 𝜎 = ±𝜎 ,

depending on whether 𝑠 = 1 or 𝑠 = −1 . Strictly speaking, having adopted
(75) and (76) as the definition of 𝜎 , we should not write 𝜎 = 𝜎 unless
𝑖0 = 0, … , 𝑖𝑘 = 𝑘 , even if 𝑠 𝑖0 , 𝑖1 , … , 𝑖𝑘 = 1 ; what we have here is an
108
 Space for Hints

equivalence relation, not an equality. However, for our purposes the notation
is justified by Theorem 5.13.

If 𝜎 = 𝜀𝜎 (using the above convention) and if 𝜀 = 1, we say

that 𝜎 and 𝜎 have the same orientation; if 𝜀 = −1 , 𝜎 and 𝜎 said to have
opposite orientation. Note that we have not defined what we mean by the
―orientation of a simplex‖. What we have defined is a relation between pairs
of simplexes having the same set of vertices, the relation being that of ―having
the same orientation.‖

There is, however, one situation where the orientation of a

simplex can be defined in a natural way. This happens when 𝑛 = 𝑘 and when
the vectors 𝑃𝑖 − 𝑃0 1 ≤ 𝑖 ≤ 𝑘 are independent. In that case, the linear
transformation 𝐴 that appears in (78) is invertible, and its determinant (which
is the same as the Jacobian of 𝜎) is not 0. Then 𝜎 is said to be positively (or
negatively) oriented if det 𝐴 is positive (or negative). In particular, the simplex
0, 𝑒1 , … , 𝑒𝑘 in 𝑅 𝑘 , given by the identity mapping, has positive orientation.

So far we have assume that 𝐾 ≥ 1. An oriented 0-simplex is

defined to be a point with a sign attached. We write 𝜎 = +𝑃0 or 𝜎 = −𝑃0 . If
𝜎 = 𝜀𝑃0 𝜀 = ±1 and if 𝑓 is a 0-form (i.e., a real function), we define

𝑓 = 𝜀𝑓 𝑃0
𝜎

5.13 Theorem

If 𝜎 is an oriented rectilinear 𝑘-simplex in an open set 𝐸 ⊂ 𝑅 𝑛 and if

𝜎 = 𝜀𝜎 then

∫𝜎 𝜔 = 𝜀 ∫𝜎 𝜔
(81)
109
Space for Hints

for every 𝑘-form 𝜔 in 𝐸.

proof For 𝑘 = 0, (81) follows from the preceding definition. So we assume

𝐾 ≥ 1 and assume that 𝜎 is given by (75).

Suppose 1 ≤ 𝑗 ≤ 𝑘 , and suppose 𝜎 is obtained from 𝜎 by

interchanging 𝑃0 and 𝑃𝑗 . The 𝜀 = −1, and

𝜎 𝑢 = 𝑃𝑗 + 𝐵𝑢 𝑢 ∈ 𝑄𝑘

where 𝐵 is the linear mapping of 𝑅 𝑘 into 𝑅 𝑛 defined 𝐵𝑒𝑗 = 𝑃0 − 𝑃𝑗 ,

𝐵𝑒𝑖 = 𝑃𝑖 − 𝑃𝑗 if 𝑖 ≠ 𝑗. If write 𝐴𝑒𝑖 = 𝑥𝑖 1 ≤ 𝑖 ≤ 𝑘 , where 𝐴 is given by (78),
the column vectors of 𝐵 (that is, the vectors 𝐵𝑒𝑖 ) are

𝑥1 − 𝑥𝑗 , … , 𝑥𝑗 −1 − 𝑥𝑗 , 𝑥𝑗 +1 , … , 𝑥𝑘 − 𝑥𝑗

If we subtract the 𝑗th column from each of the others, none of the determinants
in (35) are affected, and we obtain columns 𝑥1 , … , 𝑥𝑗 −1 − 𝑥𝑗 , 𝑥𝑗 +1 , … , 𝑥𝑘 .
These differ from those of 𝐴 only in the sign of the 𝑗th column. Hence (81)
holds for this case.

Suppose next that 0 ≤ 𝑖 ≤ 𝑗 ≤ 𝑘 and that 𝜎 is obtained from 𝜎 by

interchanging 𝑃𝑖 and 𝑃𝑗 . Then 𝜎 𝑢 = 𝑃0 + 𝐶𝑢 , where 𝐶 has the asme
columns as 𝐴, except that 𝑖th and 𝑗th columns have been interchanged. This
again implies that (81) holds, since 𝜀 = −1.

The general case follows, since every permutation of 0,1, … , 𝑘 is a

composition of the special cases we have just dealt with.

Definition(Affine chains) 5.14

110
 Space for Hints

An affine 𝑘-chain Γ in an open set 𝐸 ⊂ 𝑅 𝑛 is a collection of finitely many

oriented affine 𝑘 -simplexes 𝜎1 , … , 𝜎𝑟 in 𝐸 . These need not be distinct; a
simplex may thus occur in Γ with a certain multiplicity.

If Γ is above, and 𝜔 is a 𝑘-form in 𝐸, we define

𝑟
∫Γ 𝜔 = 𝑖=1 ∫𝜎 𝑖 𝜔 (82)

We may view a 𝑘-surface Φ in 𝐸 as a function whose domain is

the collection of all 𝑘-foms in 𝐸 and which assigns the number ∫Φ 𝜔 to 𝜔.

Since real-valued functions can be added (as in Definition 4.3), this suggests
the use of the notation

Γ = 𝜎1 + ⋯ + 𝜎𝑟 (83)

or, more compactly,

𝑟
Γ= 𝑖=1 𝜎𝑖 (84)

to state the fact that (82) holds for every 𝑘-form 𝜔 in 𝐸.

To avoid misunderstanding, we point out explicitly that the

notations introduced by (83) and (80) have to be handled with care. The point
is that every oriented affine 𝑘-simplex 𝜎 in 𝑅 𝑛 is a function in two ways, with
different domains and different ranges, and that therefore two entirely
different operations of addition are possible. Originally, 𝜎 was defined as an
𝑅 𝑛 -valued function with domain 𝑄 𝑘 ; accordingly, 𝜎1 + 𝜎2 could be interpreted
to be the function 𝜎 that assigns the vector 𝜎1 𝑢 + 𝜎2 𝑢 is the null vector of
𝑅𝑛 .

Definition(Boundaries) 5.15

For 𝑘 ≥ 1, the boundary of the oriented affine 𝑘-simplex

111
Space for Hints

𝜎 = 𝑃0 , 𝑃1 , … , 𝑃𝑘

is defined to be the affine 𝑘 − 1 -chain

𝑘 𝑗
𝜕𝜎 = 𝑗 =0 −1 𝑃0 , … , 𝑃𝑗 −1 , 𝑃𝑗 +1 , … , 𝑃𝑘 (85)

For example, if 𝜎 = 𝑃0 , 𝑃1 , 𝑃2 , then

𝜕𝜎 = 𝑃1 , 𝑃2 − 𝑃0 , 𝑃2 + 𝑃0 , 𝑃1 = 𝑃0 , 𝑃1 + 𝑃1 , 𝑃2 + 𝑃2 , 𝑃0

which coincides with the usual notion of the oriented boundary of a

triangle.

For 1≤𝑗≤𝑘 , observe that the simples

𝜎𝑗 = 𝑃0 , … , 𝑃𝑗 −1 , 𝑃𝑗 +1 , … , 𝑃𝑘 which occurs in (85) has 𝑄 𝑘−1 as its parameter
domain and that it defined by

𝜎𝑗 𝑢 = 𝑃0 + 𝐵𝑢 𝑢 ∈ 𝑄 𝑘−1 (86)

where 𝐵 is the linear mapping from 𝑅 𝑘−1 to 𝑅 𝑛 determined by

𝐵𝑒𝑖 = 𝑃𝑖 − 𝑃0 𝑖𝑓 1 ≤ 𝑖 ≤ 𝑗 − 1

𝐵𝑒𝑖 = 𝑃𝑖+1 − 𝑃0 𝑖𝑓 𝑗 ≤ 𝑖 ≤ 𝑘 − 1

The simples

𝜎0 = 𝑃0 , 𝑃1 , … , 𝑃𝑘

which also occurs in (85), is given by the mapping

𝜎0 = 𝑃1 + 𝐵𝑢

where 𝐵𝑒𝑖 = 𝑃𝑖+1 − 𝑃1 for 1 ≤ 𝑖 ≤ 𝑘 − 1.

Definition(Differentiable simplexes and chains) 5.16

112
 Space for Hints

Let 𝑇𝒞 ′′ -mapping of an open 𝐸 ⊂ 𝑅 𝑛 into an open set 𝑉 ⊂ 𝑅 𝑚 ; 𝑇 need not be

one-to-one. If 𝜎 is an oriented affine 𝑘 -simplex in 𝐸 , then the composite
mapping Φ = T ∘ 𝜎 (which we shall sometimes write in the simpler form 𝑇𝜎)
is a 𝑘-surface in 𝑉 , with parameter domain 𝑄 𝑘 . We call Φ an oriented 𝑘 -
simplex of class 𝒞 ′′ .

A finite collection Ψ of oriented 𝑘-simplexes Φ1 , … , Φ𝑟 of class 𝒞 ′′ in

𝑉 is called a 𝑘-chain of class 𝒞 ′′ in 𝑉. If 𝜔 is a 𝑘-form in 𝑉, define

𝑟
∫Ψ 𝜔 = 𝑖=1 ∫Φ 𝑖 𝜔 (87)

and use the corresponding notation Ψ = Φ𝑖

If Γ = 𝜎𝑖 is an affine chain and if Φ𝑖 = 𝑇 ∘ 𝜎𝑖 , we also write Ψ =

TΓ, or

𝑇 𝜎𝑖 = T𝜎𝑖 (88)

The boundary 𝜕Φ of the oriented 𝑘-simplex Φ = T ∘ 𝜎 is defined to be

the 𝑘 − 1 chain

𝜕Φ = T 𝜕𝜎 (89)

In justification of (89), observe that if 𝑇 is affine, then Φ = T ∘ 𝜎 is an

oriented affine 𝑘-simplex, in which case (89) is not a matter of definition, but
is seen to be consequence of (85). Thus (89) generalizes this special case.

It is immediate that 𝜕Φ is of class 𝒞 ′′ if this it true of Φ.

Finally, we define the boundary 𝜕Ψ of the 𝑘-chain Ψ = Φ𝑖 to be the

𝑘 − 1 chain

𝜕Ψ = 𝜕Φ𝑖 (90)

113
Space for Hints

Definition(Positively oriented boundaries) 5.17

So far we have associated boundaries to chains, not to subsets of 𝑅 𝑛 .

This notion of boundary is exactly the one that is most suitable for the
statement and proof of Stokes theorem. However, in applications, especially in
𝑅 2 or 𝑅 3 , it is customary and convenient to talk about ―oriented boundaries‖
of certain sets as well. We shall now describe this briefly.

Let 𝑄 𝑛 be the standard simplex in 𝑅 𝑛 , let 𝜎0 be the identity mapping

with domain 𝑄 𝑛 . Here, 𝜎0 may be regarded as a positively oriented 𝑛-simplex
in 𝑅 𝑛 . Its boundary 𝜕𝜎0 is an affine 𝑛 − 1 -chain. This chain is called the
positively oriented boundary of the set 𝑄 𝑛 .

For example, the positively oriented boundary of 𝑄 3 is

𝑒1 , 𝑒2 , 𝑒3 − 0, 𝑒2 , 𝑒3 + 0, 𝑒1 , 𝑒3 − 0, 𝑒1 , 𝑒2

Now let 𝑇 be 1-1 mapping of 𝑄 𝑛 into 𝑅 𝑛 , of class 𝒞 ′′ , whose Jacobian

is positive (at least in the interior of 𝑄 𝑛 ). Let 𝐸 = 𝑇 𝑄 𝑛 . By the inverse
function theorem, 𝐸 is the closure of an open subset of 𝑅 𝑛 . We define the
positively oriented boundary of the set 𝐸 to be the 𝑛 − 1 -chain

𝜕Φ = T 𝜕𝜎0

and we may denote this 𝑛 − 1 -chain by 𝜕𝐸.

An obvious question occurs here: If 𝐸 = 𝑇1 𝑄 𝑛 = 𝑇2 𝑄 𝑛 , and if both

𝑇1 and 𝑇2 have positive Jacobians, is it true that 𝜕𝑇1 = 𝜕𝑇2 ? That is to say,
does the equality

𝜔= 𝜔
𝜕𝑇1 𝜕𝑇2

114
 Space for Hints

hold for every 𝑛 − 1 -form 𝜔? The answer is yes, but we shall omit the
proof.

One can go further. Let

Ω = 𝐸1 ∪ … ∪ 𝐸𝑟

where 𝐸𝑖 = 𝑇𝑖 𝑄 𝑛 , each 𝑇𝑖 has the properties that 𝑇 had above, and the
interiors of the sets 𝐸𝑖 are pairwise disjoint. Then the 𝑛 − 1 -chain

𝜕𝑇1 + ⋯ + 𝜕𝑇𝑟 = 𝜕Ω

is called the positively oriented boundary of Ω.

For example, the unit square 𝐼 2 in 𝑅 2 is the union of 𝜎1 𝑄 2 and 𝜎2 𝑄 2 ,

where

𝜎1 𝑢 = 𝑢, 𝜎1 𝑢 = 𝑒1 + 𝑒2 − 𝑢

Both 𝜎1 and 𝜎2 have Jacobian 1 > 0. Since

𝜎1 = 0, 𝑒1 , 𝑒2 , 𝜎2 = 𝑒1 + 𝑒2 , 𝑒2 , 𝑒1

we have

𝜕𝜎1 = 𝑒1 , 𝑒2 − 0, 𝑒2 + 0, 𝑒1

𝜕𝜎2 = 𝑒2 , 𝑒1 − 𝑒1 + 𝑒2 , 𝑒1 + 𝑒1 + 𝑒2 , 𝑒2

The sum of these two boundaries is

𝜕𝐼 2 = 0, 𝑒1 − 𝑒1 , 𝑒1 + 𝑒2 + 𝑒1 + 𝑒2 , 𝑒2 + 𝑒2 , 0

the positively oriented boundary of 𝐼 2 . Note that 𝑒1 , 𝑒2 canceled 𝑒2 , 𝑒1 ,

If Φ is a 2-surface in 𝑅 𝑚 , with parameter domain 𝐼 2 , then Φ (regarded

as function on 2-forms) is the same as the 2-chain
115
Space for Hints

Φ ∘ 𝜎1 + Φ ∘ 𝜎2

Thus

𝜕Φ = 𝜕 Φ ∘ 𝜎1 + 𝜕 Φ ∘ 𝜎2

= Φ 𝜕𝜎1 + Φ 𝜕𝜎2 = Φ 𝜕𝐼 2

In other words, if the parameter domain of Φ is the square 𝐼 2 , we need

not refer back to simplex 𝑄 2 , but can obtain 𝜕Φ directly from 𝜕I2 .

Example 5.18

For 0 ≤ 𝑢 ≤ 𝜋, 0 ≤ 𝑣 ≤ 2𝜋, define

𝑢, 𝑣 = sin 𝑢 cos 𝑣 , sin 𝑢 sin 𝑣 , cos 𝑢

Then is a 2-surface in 𝑅 3 , whose parameter domain rectangle 𝐷 ⊂ 𝑅 2 , and

whose range is the unit sphere in 𝑅 3 . Its boundary is

𝜕 = 𝜕𝐷 = 𝛾1 + 𝛾2 + 𝛾3 + 𝛾4

where

𝛾1 𝑢 = 𝑢, 0 = sin 𝑢 , 0, cos 𝑢

𝛾2 𝑢 = 𝜋, 𝑣 = 0, 0, −1

𝛾3 𝑢 = 𝜋 − 𝑢, 2𝜋 = sin 𝑢 , 0, − cos 𝑢

𝛾4 𝑢 = 0,2𝜋 − 𝑣 = 0, 0, 1

with 0, 𝜋 and 0,2𝜋 as parameter intervals for 𝑢 and 𝑣, respectively.

Since 𝛾2 and 𝛾4 are constant, their derivatives are 0, hence the integral
of any 1-form over 𝛾2 or 𝛾4 is 0. [See Example 1.12 (a).]

116
 Space for Hints

Since 𝛾3 𝑢 = 𝛾1 𝜋 − 𝑢 , direct application of (35) shows that

𝜔=− 𝜔
𝛾3 𝛾1

for every 1-form 𝜔. Thus ∫𝜕 𝜔 = 0, and we conclude that 𝜕 = 0.

(In geographic terminology, 𝜕 starts at the north pole 𝑁, runs to the

south pole 𝑆 along a meridian, pauses at 𝑆 , returns to 𝑁 along the same
meridian, and finally pauses at 𝑁. The two passages along the meridian are in
opposite directions. The corresponding two line integrals therefore cancel each
other.)

5.7 Stokes’ theorem

Theorem 5.19

If Ψ is a 𝑘 -chain of class 𝒞 ′′ in an open set 𝑉 ⊂ 𝑅 𝑚 and if 𝜔 is a

𝑘 − 1 -form of class 𝒞 ′ in 𝑉, then

∫Ψ 𝑑𝜔 = ∫𝜕Ψ 𝜔 (91)

The case 𝑘 = 𝑚 = 1 is nothing but the fundamental theorem of

calculus (with an additional differentiability assumption). The 𝑘 = 𝑚 = 2 is
Green‘s theorem, and 𝑘 = 𝑚 = 3 gives the so-called ―divergence theorem‖ of
Gauss. The case 𝑘 = 2, 𝑚 = 3 is the one originally discovered by Stokes.
(Spivak‘s book describes some of the historical background).

Proof It is enough to prove that

∫Φ 𝑑𝜔 = ∫𝜕Φ 𝜔 (92)

117
Space for Hints

for every oriented 𝑘-simplex Φ of class 𝒞 ′′ in 𝑉. For if (92) is proved and if

Ψ = Φ𝑖 , then (87) and (89) imply (91).

Fix such a Φ and put

𝜎 = 𝑒0 , 𝑒1 , … , 𝑒𝑘

Thus 𝜎 is the oriented affine 𝑘-simplex with parameter domain 𝑄 𝑘 which is

defined by the identity mapping. Since Φ is also defined on 𝑄 𝑘 (see Definition
5.16) and Φ ∈ 𝒞 ′′ , there is an open set 𝐸 ⊂ 𝑅 𝑘 which contains 𝑄 𝑘 , and there is
a 𝒞 ′′ -mapping 𝑇 of 𝐸 into 𝑉 such that Φ = T ∘ 𝜎. The left side of (92) is equal
to

𝑑𝜔 = 𝑑𝜔 𝑇 = 𝑑 𝜔 𝑇
T𝜎 𝜎 𝜎

By (89), that the right side of (92) is

𝑑𝜔 = 𝑑𝜔 𝑇 = 𝑑 𝜔 𝑇
𝜕 T𝜎 𝑇 𝜕𝜎 𝜕𝜎

Since 𝜔 𝑇 is a 𝑘 − 1 -form in 𝐸, we see that in order to prove (92) we

merely have to show that

∫𝑇𝜎 𝜆 = ∫𝜕𝜎 𝜆 (94)

for the special simplex (93) and for every 𝑘 − 1 -form 𝜆 of class 𝒞 ′ in 𝐸.

If 𝑘 = 1, the definition of an 0-simplex shows that (94) merely asserts

that

1
∫0 𝑓 ′ 𝑢 = 𝑓 1 − 𝑓 0 (95)

118
 Space for Hints

for every continuously differentiable function 𝑓 on 0,1 , which is true by the

fundamental theorem of calculus.

From now onwe assume that 𝑘 > 1, fix an integer 𝑟 1 ≤ 𝑟 ≤ 𝑘 , and

choose 𝑓 ∈ 𝒞 ′ 𝐸 . It is then enough to prove (94) for the case

𝜆 = 𝑓 𝑥 𝑑𝑥1 ⋀ … ⋀𝑑𝑥𝑟−1 ⋀𝑑𝑥𝑟+1 ⋀ … ⋀𝑑𝑥𝑘

since every 𝑘 − 1 -form is a sum of these special ones, for 𝑟 = 1, … , 𝑘.

By (85), the boundary of the simplex (93) is

𝜎 = 𝑒0 , 𝑒1 , … , 𝑒𝑘 + −1 𝑖 𝜏𝑖
𝑖=1

where

𝜏𝑖 = 0, 𝑒1 , … , 𝑒𝑖−1 , 𝑒𝑖+1 , … 𝑒𝑘
for 1 = 1, … , 𝑘. put

𝜏0 = 𝑒𝑟 , 𝑒1 , … , 𝑒𝑟−1 , 𝑒𝑟+1 , … 𝑒𝑘

Note that 𝜏0 is obtained from 𝑒1 , … , 𝑒𝑘 by 𝑟 − 1 successive interchanges of

𝑒𝑟 and its left neighbors. Thus

𝑟−1 𝑘
𝜕𝜎 = −1 𝜏0 + 𝑖=1 −1 𝑖 𝜏𝑖 (97)

Each 𝜏𝑖 has 𝑄 𝑘−1 as parameter domain.

If 1 ≤ 𝑖 ≤ 𝑘, 𝑢 ∈ 𝑄 𝑘−1 , then

𝑢𝑗 1 ≤ 𝑗 ≤ 𝑟 ,
𝑥𝑗 = 1 − 𝑢1 + ⋯ + 𝑢𝑘−1 𝑗 = 𝑟 , (98)
𝑢𝑗 −1 𝑖 < 𝑗 ≤ 𝑘 .

If 1 ≤ 𝑖 ≤ 𝑘, 𝑢 ∈ 𝑄 𝑘−1 and 𝑥 = 𝜏𝑖 𝑢 , then

119
Space for Hints

𝑢𝑗 1 ≤ 𝑗 ≤ 𝑟 ,
𝑥𝑗 = 0 𝑗 = 𝑟 , (99)
𝑢𝑗 −1 𝑖 < 𝑗 ≤ 𝑘 .

For 1 ≤ 𝑖 ≤ 𝑘, let 𝐽𝑖 be the Jacobian of the mapping

𝑢1 , … , 𝑢𝑘−1 → 𝑥1 , … , 𝑥𝑟−1 , 𝑥𝑟+1 , … , 𝑥𝑘 (100)

induced by 𝜏𝑖 . When 𝑖 = 0 and when = 𝑟 , (98) and (99) show that (100) is the
identity mapping. Thus 𝐽0 = 1, 𝐽𝑟 = 1. For other 𝑖, the fact that 𝑥𝑖 = 0 in (99)
shows that 𝐽𝑖 has a row of zeros, 𝐽𝑖 = 0. Thus

∫𝜏 𝜆 = 0 𝑖 ≠ 0, 𝑖 ≠ 𝑟 (101)
𝑖

by (35) and (96). Consequently, (97) gives

𝑟−1 𝑟−1
∫𝜕𝜎 𝜆 = −1 ∫𝜏 𝜆 + −1 ∫𝜏 𝜆 (102)
0 𝑟

𝑟−1
= −1 ∫ 𝑓 𝜏0 𝑢 − 𝑓 𝜏𝑟 𝑢 𝑑𝑢

On the other hand,

𝑑𝜆 = 𝐷𝑟 𝑓 𝑥 𝑑𝑥𝑟 ⋀𝑑𝑥1 ⋀ … ⋀𝑑𝑥𝑟−1 ⋀𝑑𝑥𝑟+1 ⋀ … ⋀𝑑𝑥𝑘

𝑟−1
= −1 𝐷𝑟 𝑓 𝑥 𝑑𝑥1 ⋀ … ⋀𝑑𝑥𝑘

so that

𝑟−1
∫𝜎 𝑑𝜆 = −1 ∫𝑄 𝑘 𝐷𝑟 𝑓 𝑥 𝑑𝑥 (103)

We evaluate (103) by first integrating with respect to 𝑥𝑟 , over the interval

0,1 − 𝑥1 + ⋯ + 𝑥𝑟−1 + 𝑥𝑟+1 + ⋯ + 𝑥𝑘

120
 Space for Hints

put 𝑥1 + ⋯ + 𝑥𝑟−1 + 𝑥𝑟+1 + ⋯ + 𝑥𝑘 = 𝑢1 , … , 𝑢𝑘−1 , and see with the aid

of (98) that the integral over 𝑄 𝑘 in (103) is equal to the integral over 𝑄 𝑘−1 in
(102). Thus (94) holds, and the proof is complete.

Closed forms and exact forms

Definition 5.20

Let 𝜔 be a 𝑘-form in an open set 𝐸 ⊂ 𝑅 𝑛 . If there is a 𝑘 − 1 -form 𝜆

in 𝐸 such that 𝜔 = 𝑑𝜆, the 𝜔 is said to be exact in 𝐸.

If 𝜔 is of class 𝒞 ′ and 𝑑𝜔 = 0, then 𝜔 is said to be closed.

Remarks 5.21

(a) Whether a given 𝑘-form 𝜔 is or is not closed can be verified by

simply differentiating the coefficients in the standard presentation of
𝜔.

For example, a 1-form

𝑛
𝜔= 𝑖=1 𝑓𝑖 𝑥 𝑑𝑥𝑖 (104)

with 𝑓𝑖 𝜖𝒞 ′ 𝐸 for some open set 𝐸 ⊂ 𝑅 𝑛 , is closed if and only if the

equations

𝐷𝑖 𝑓𝑖 𝑥 = 𝐷𝑖 𝑓𝑖 𝑥 (105)

hold for all 𝑖, 𝑗 in 1, … , 𝑛 and for all 𝑥 ∈ 𝐸.

CYP QUESTIONS:
1. Show that the simplex 𝑄 𝑘 is the sn1al lest convex subset of 𝑅 𝑘 that
contains0, 𝑒1 , … , 𝑒𝑘 .

121
Space for Hints

2. Show that affine mappings take convex sets to convex sets .

122