Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Chapter I

Preliminaries
This chapter prepares the reader for undertaking a study of functional
analysis. Basic notions of set theory are given in Section 1. Linear
spaces and metric spaces are introduced in Section 2 and 3, respectively. Korovkins theorem for positive linear maps is employed to
obtain some important approximation results. A review of the theory
of Lebesgue measure on the real line is given in Section 4. A brief
discussion of Fourier series and integrals concludes this section, which
will provide numerous examples of subsequent results.

1 Relations on a Set
As a deductive science, mathematics rests on the theory of sets as its
foundation. A set is supposed to be made up of its elements. In
other words, certain elements belong to a set while certain others do
not. The notation and 6 are used to denote belongs to and does
not belong to.
It was realized at the turn of this century that not every collection of elements can be allowed to constitute a set without leading
to paradoxes. One such famous paradox is due to Russell (1901).
Suppose we allow as sets, things T where T could belong to itself.
Let S be the set of all things T such that T 6 T . Then it is easily
seen that S S if and only if S 6 S, which is clearly paradoxical.
To understand this phenomenon better, consider the following real
life example. In a town lives a barber who shaves exactly all those
who do not shave themselves. The question is whether the barber
shaves himself. Answer: The barber shaves himself if and only if he

Chapter I Preliminaries

does not shave himself! Here the set S of all those who do not shave
themselves is represented by the barber and the question reduces to
Russells paradox.
In order to avoid such paradoxes, some restrictions have to be put
on what one should call sets. These are formulated in various axiom
systems for the theory of sets. The one that we shall tacitly follow is
due to Zermelo and Frankel (1908). (See, for example, [12]). We do
not wish to develop such a set theory here. The preceding remarks
are made only to put things in a proper perspective.
We shall assume familiarity with the elementary notions in set
theory such as a subset (), union (), intersection (), complementation (E c ), and also with finite and infinite sets. A set is called
denumerable if it is in one to one correspondence with the set of
all natural numbers and it is called countable if it is either finite or
denumerable. We shall denote the empty set by .
A relation on a set X is a subset of the cartesian product
XX = {(x, y) : x, y X}. We consider three important types of
relations on a set. These will be extensively used in the sequel.
A relation R on a set X is called a function if (x, y), (x, z) R
implies that y = z. The set {x X : (x, y) R for some y X}
is called the domain of the function and the set {y X : (x, y)
R for some x X} is called the range of the function. Thus a
function associates to every element in its domain a unique element
in its range. Let now X1 and X2 be two sets and X = X1 X2 . If F
is a function on X such that the domain of F is X1 and the range of
F is contained in X2 , we shall adopt the usual notation F : X1 X2
with F (x1 ) = x2 if and only if (x1 , x2 ) F. If Y X1 , then F|Y will
denote the restriction of F to Y, obtained by requiring x1 to belong
to Y . Also, if Z X2 then F 1 (Z) will denote the inverse image
{x1 X : F (x1 ) Z} of Z under F. If F (x1 ) = F (y1 ) implies that
x1 = y1 , then F is said to be injective or one-to-one. If for every
x2 X2 there is some x1 X1 with F (x1 ) = x2 , then F is said to

1 Relations on a Set

be surjective or F is said to map X1 onto X2 . A function F is said


to be bijective if it is injective as well as surjective. If F is bijective
and F (x1 ) = x2 , then F 1 (x2 ) = x1 defines the inverse F 1 of F . If
F : X1 X2 and G : X2 X3 , then the composition of F and G
is the function GF : X1 X3 given by (GF )(x1 ) = G(F (x1 )) for
x1 X 1 .
A relation R on a set X is called an equivalence relation if
(i) R is reflexive, that is, (x, x) R for every x X, (ii) R is
symmetric, that is, (x, y) R implies that (y, x) R and (iii) R
is transitive, that is, (x, y), (y, z) R implies that (x, z) R. An
equivalence relation R on X partitions the set X into a collection of
subsets, each of which is called an equivalence class. Two elements
x and y are in the same equivalence class if and only if (x, y) R. We
shall write x y if (x, y) R. The equivalence classes are mutually
disjoint and their union is X. For example, the set of all straight lines
in the plane can be partitioned into equivalence classes so that two
straight lines are equivalent if and only if they are parallel.
A relation R on a set X is called a partial order if R is (i) reflexive, (ii) antisymmetric, that is, (x, y), (y, x) R implies that x = y
and (iii) transitive. A partial order R thus differs from an equivalence
relation only in symmetry. We shall write x y if (x, y) R. A
partially ordered set is a set X together with a partial order on it.
For Y X and x X, if y x for every y Y, then x is said to be
an upper bound for Y in X. If x y implies that x = y for every
y X, then x is said to be maximal in X. It should be noted that
maximal elements may not exist at all or may exist in plenty. Natural
numbers with the usual order contain no maximal element. On the
other hand, let us consider the set of all branches of a given tree at
a given time, and put a partial order on it by letting b1 b2 if the
branch b2 has grown out of the branch b1 . Then all the most newly
formed branches are maximal. The tree in Figure 1 has ten maximal

Chapter I Preliminaries

branches. Finally, we define a totally ordered set to be a partially


ordered set X in which x, y X implies that x y or y x, that
is, any two elements of X are comparable. The natural numbers are
totally ordered, but the branches of a tree are not.

Figure 1
We now state an additional axiom of set theory.
Zorns lemma
Let X be a nonempty partially ordered set such that every totally
ordered subset of X has an upper bound in X. Then X contains a
maximal element.
Several remarks are in order. First of all, although this statement
is called a lemma, it is an axiom of set theory. In fact, it can be shown
to be equivalent to the following.
Axiom of choice
If I is a nonempty set and Xi is a nonempty set for every i I, then
S
there exists a function F : I {Xi : i I} such that F (i) Xi for
every i I.
While the selection procedure in the axiom of choice seems to
be readily acceptable, it has been proved to be independent of the

2 Linear Spaces and Linear Maps

Zermelo-Frankel axiom system. It was shown by Godel (1940) and


Cohen (1963) that if the Zermelo-Frankel axiom system is consistent,
then it remains consistent together with the axiom of choice and also
together with the negation of the axiom of choice.
We shall find life much easier if we assume that the axiom of
choice holds. We shall do so by assuming Zorns lemma to be valid.
Of course, it must be pointed out that Zorns lemma is an existential
statement. It asserts the existence of a maximal element in a set under
certain conditions, but gives no clue for finding such an element. We
are thus at a disadvantage from the point of view of construction
procedures. A good account of the constructive methods in analysis
(which does not assume Zorns lemma) can be found in [4].

2 Linear Spaces
and Linear Maps
We introduce an algebraic structure on a set X and study functions
on X which are well-behaved with respect to this structure.
From now onwards, K will denote either R, the set of all real
numbers or C, the set of all complex numbers. For k C, Re k and
Im k will denote the real part and the imaginary part of k.
A linear space (or a vector space) over K is a nonempty set
X along with a function + : XX X, called addition, and a
function : KX X, called scalar multiplication, such that for
all x, y, z X and k, ` K, we have
x + y = y + x,
x + (y + z) = (x + y) + z,
there exists 0 X such that x + 0 = x,
there exists x X such that x + (x) = 0,

Chapter I Preliminaries
k (x + y) = k x + k y,
(k + `) x = k x + ` x,
(k`) x = k (` x),
1 x = x.

We shall write kx in place of k x. We shall also adopt the following


notations. For x, y X, k K and subsets E, F of X,
x + F = {x + y : y F } ,
kE = {kx : x E} .

E + F = {x + y : x E, y E},

Let X be a linear space (over K). A subset E of X is said to be


convex if rx + (1 r)y E whenever x, y E and 0 < r < 1. If
E X, then the smallest convex subset of X containing E is
co (E) = {r1 x1 + +rn xn : x1 , . . . , xn E, r1 , . . . , rn 0,

n
X

rj = 1}.

j=1

It is called the convex hull of E.


A nonempty subset E of X is said to be a subspace of X if
kx + ly E whenever x, y E and k, l K. If =
6 E X, then the
smallest subspace of X containing E is
span E = {k1 x1 + + kn xn : x1 , . . . , xn E, k1 , . . . , kn K}
It is called the span of E. If span E = X, we say that E spans X.
A subset E of X is said to be linearly independent if for all
x1 , . . . , xn E and k1 , . . . , kn K, the equation k1 x1 + + kn xn = 0
implies that k1 = = kn = 0. It is called linearly dependent if it
is not linearly independent, that is, if there exist x1 , . . . , xn E and
k1 , . . . , kn K such that k1 x1 + + kn xn = 0, where at least one kj
is nonzero.
A subset E of X is called a Hamel basis or simply a basis for X
if span E = X and E is linearly independent. Does every linear space
have a basis? Every nonempty maximal linearly independent subset

2 Linear Spaces and Linear Maps

as well as every nonzero minimal subset which spans X is a basis for


X. But do such subsets of X exist? One may start with a linearly
independent subset of X and progressively enlarge it until it spans
X, or one may start with a subset which spans X and progressively
reduce it until it becomes linearly independent. While it is not at all
obvious whether one can obtain a basis for X in this manner, Zorns
lemma guarantees this whenever X 6= {0}. If, however, X = {0},
then X has no basis, since the only subsets of X are and {0}, of
which does not span X and {0} is not linearly independent.
2.1 Theorem
Let X be a nonzero linear space. Consider a linearly independent
subset L of X and a subset S of X which spans X such that L S.
Then there exists a basis B for X such that L B S.
Proof:
Let E = {E : L E S, E is linearly independent}. Then E is
nonempty since L E. Also, the inclusion is a partial order on E.
S
Further, if F is a totally ordered subset of E, then E = {F : F F }
is an upper bound for F in E, because L E S and E is linearly
independent since whenever x1 , . . . , xn E, there exists (a linearly
independent) F F such that x1 , . . . , xn F as F is totally ordered.
By Zorns lemma, E contains a maximal element B.

Since B E, we see that L B S and B is linearly independent. It remains show that B spans X. First we prove that
B 6= . Since span S = X 6= {0}, S contains a nonzero element
of X. Since B is maximal in E, it follows that B 6= . Next, we
prove that S span B. Suppose for a moment that there is some
x S with x 6 span B. Then B {x} is a linearly independent
subset of X. For let kx + k1 b1 + + kn bn = 0 for some k K and
k1 b1 + + kn bn = y span B. Hence kx = y span B. Since
x 6 span B, we obtain k = 0. But then y = k1 b1 + + kn bn = 0 and
since B is linearly independent, we obtain k1 = = kn = 0. This

Chapter I Preliminaries

shows that B {x} is an element of E which is larger than B, contrary


to the maximality of B in E. Hence S span B. As span S = X, we
have span B = X, as desired.
2
2.2 Corollary
Let X be a nonzero linear space.
(a) There exists a basis for X.
(b) Let Y be a subspace of X and C be a basis for Y . Then there
is a basis B for X such that C B.
Proof:
(a) Let L = and S = X in Theorem 2.1.
(b) Let L = C and S = X in Theorem 2.1.

2.3 Theorem
Let a linear space X have a basis consisting of n elements, 1 n < .
Then every basis for X has n elements. Further, if a subset of n
elements in X either spans X or is linearly independent, then it is, in
fact, a basis for X.
Proof:
Let B = {b1 , . . . , bn } be a basis for X consisting of n elements. First,
we show that any linearly independent subset of X has at most n
elements. Let L be a linearly independent subset of X. If L B,
then there is nothing to prove. If, on the other hand, there is some
a1 L with a1 6 B, then since the set {a1 } is linearly independent
and since the set B {a1 } spans X, there is a basis B1 for X such
that {a1 } B1 B {a1 } by 2.2(b). As a1 X = span B and as
B1 is linearly independent, it follows that B1 6= B {a1 }. Hence at
least one element of B does not belong to B1 , so that B1 has at most
n elements. If L B1 , we are through. If, on the other hand, there
is some a2 L with a2 6 B1 , then again by 2.2(b), there is a basis

2 Linear Spaces and Linear Maps

B2 for X such that {a1 , a2 } B2 B1 {a2 }, where at least one


element of B1 does not belong to B2 , so that B2 also has at most n
elements. If L B2 , we are through. Otherwise, we may repeat this
process. We claim that L Bk for some k n, where Bk has at most
n elements. For otherwise, there will be n + 1 elements a1 , . . . , an+1
in L such that {a1 , . . . , an } Bn and an+1 6 Bn . But since Bn has
at most n elements, we see that Bn = {a1 , . . . , an } and since Bn is a
basis for X, an+1 will be a linear combination of a1 , . . . , an , contrary
to the linear independence of L. Thus L has at most n elements.
Let C be any basis for X. Since C is linearly independent, it
has m elements with m n. Interchanging the roles of B and C, we
see that n m. Hence m = n, that is, C has the same number of
elements as B.
Let now L and S be subsets of X having n elements each such
that L is linearly independent and S spans X. By 2.2(b), there are
bases B1 and B2 of X such that L B1 and B2 S. Since B1 and
B2 must have n elements each, we see that L = B1 and S = B2 . Thus
L and S are both bases for X.
2
If a linear space X has a basis consisting of a finite number of
elements, then X is said to be finite dimensional and the number
of elements in a basis for X is called the dimension of X. It follows
from Theorem 2.3 that every basis for a finite dimensional linear space
has the same (finite) number of elements and hence the dimension is
well-defined. The space {0} is said to have zero dimension. Note
that it has no basis !
If a linear space contains an infinite linearly independent subset, then it is said to be infinite dimensional. It can be shown by
using elaborate cardinality arguments that any two bases for an infinite dimensional linear space are in one to one correspondence. (See
[34], Theorem 3.12.) We then define the dimension of X to be the
cardinality of a basis for X. It is denoted by dim X.

10

Chapter I Preliminaries

We shall now give procedures for obtaining new linear spaces from
the given ones. A subspace Y of a linear space X is said to be proper
if Y 6= X. A proper subspace Y of X is a linear space along with the
induced addition and scalar multiplication. Further, for x1 , x2 X, if
we let x1 x2 whenever x1 x2 Y, then is an equivalence relation
on X. Let us denote the equivalence class of x X by x + Y and let
X/Y = {x + Y : x X}.
For x1 + Y, x2 + Y in X/Y and k K, define
(x1 + Y ) + (x2 + Y ) = (x1 + x2 ) + Y

and k(x1 + Y ) = kx1 + Y.

Then it can be readily seen that X/Y along with these functions is a
linear space over K. It is called the quotient space of X by Y.
Next, let X1 , . . . , Xn be linear spaces over K and consider
X1 Xn = {(x1 , . . . , xn ) : xj Xj , j = 1, . . . , n}.
For (x1 , . . . , xn ) and (y1 , . . . , yn ) in X1 Xn and k K, define
(x1 , . . . , xn ) + (y1 , . . . , yn ) = (x1 + y1 , . . . , xn + yn ),
and
k(x1 , . . . , xn ) = (kx1 , . . . , kxn ).
Then X1 Xn along with these functions is a linear space over
K. It is called the product space of X1 , . . . , Xn . The most common
example of a product space is Kn = K K (n times).

Linear Maps
Let X and Y be linear spaces over K. A linear map from X to Y is
a function F : X Y such that
F (k1 x1 + k2 x2 ) = k1 F (x1 ) + k2 F (x2 )

You might also like