Discover millions of ebooks, audiobooks, and so much more with a free trial

Only $11.99/month after trial. Cancel anytime.

Topological Vector Spaces and Distributions
Topological Vector Spaces and Distributions
Topological Vector Spaces and Distributions
Ebook808 pages6 hours

Topological Vector Spaces and Distributions

Rating: 3 out of 5 stars

3/5

()

Read preview

About this ebook

"The most readable introduction to the theory of vector spaces available in English and possibly any other language."—J. L. B. Cooper, MathSciNet Review
Mathematically rigorous but user-friendly, this classic treatise discusses major modern contributions to the field of topological vector spaces. The self-contained treatment includes complete proofs for all necessary results from algebra and topology. Suitable for undergraduate mathematics majors with a background in advanced calculus, this volume will also assist professional mathematicians, physicists, and engineers.
The precise exposition of the first three chapters—covering Banach spaces, locally convex spaces, and duality—provides an excellent summary of the modern theory of locally convex spaces. The fourth and final chapter develops the theory of distributions in relation to convolutions, tensor products, and Fourier transforms. Augmented with many examples and exercises, the text includes an extensive bibliography.
LanguageEnglish
Release dateOct 3, 2013
ISBN9780486311036
Topological Vector Spaces and Distributions

Related to Topological Vector Spaces and Distributions

Titles in the series (100)

View More

Related ebooks

Mathematics For You

View More

Related articles

Reviews for Topological Vector Spaces and Distributions

Rating: 3 out of 5 stars
3/5

2 ratings0 reviews

What did you think?

Tap to rate

Review must be at least 10 words

    Book preview

    Topological Vector Spaces and Distributions - John Horvath

    CHAPTER 1

    Banach Spaces

    §1. The definition of Banach spaces

    In the whole book the letter K will always stand either for the field R of real numbers or for the field C of complex numbers. The elements of K are called scalars and will be denoted mostly by small Greek letters.

    A structure of vector space on a set E is defined by two maps:

    (1)a map from E × E into E, called addition,

    (2)a map from K × E into E, called multiplication by a scalar.

    These maps (or algebraic operations) must satisfy the following axioms:

    (VS 1)x + y = y + x (commutativity).

    (VS 2)(x + y) + z = x + (y + z) (associativity).

    (VS 3)There exists an element 0 in E such that x + 0 = x for all x E. This element 0 is called the zero vector or the origin of E.

    (VS 4)For every element x E, there exists an element −x E such that x + (−x) = 0. The element −x is called the opposite of x.

    (VS 5)λ(x + y) = λx + λy.

    (VS 6)(λ + μ)x = λx + μx.

    (VS 7)(λμ)x = λ(μx).

    (VS 8)1 · x = x for all x E.

    If these axioms are satisfied, we say that E is a vector space (or linear space) over the field K; the elements of a vector space are called vectors. The first four axioms express the fact that E is an abelian group under addition. Note that the symbol 0 is used to denote both the zero scalar and the zero vector; this ambiguity does not in general lead to any confusion. Axioms 5 and 6 express the distributivity of multiplication by a scalar with respect to the two kinds of addition. Axiom 7 expresses a kind of associativity.

    A vector space over R will also be called a real vector space and a vector space over C a complex vector space.

    Here are some easy consequences of the above definitions.

    (a)0 · x = 0 for all x E.

    (Here on the left-hand side we have the scalar zero and on the right-hand side the vector zero.)

    Proof.We have x = 1 · x = (1 + 0) · x = 1 · x + 0 · x = x + 0 · x. Adding −x to both sides and using (VS 2), we obtain 0 = 0 · x.

    (b)λ · 0 = 0 for all λ K.

    Proof.We have λ · 0 + λ · 0 = λ(0 + 0) = λ · 0. Adding −(λ · 0) to both sides, we obtain λ · 0 = 0.

    (c)The vector 0 is unique.

    Indeed, suppose we have two vectors 0 and 0′ such that x + 0 = x and x + 0′ = x for all x. Then 0 = 0 + 0′ = 0′.

    (d)The opposite vector is unique.

    Indeed, if x + x′ = 0 and x + x″ = 0, then

    x′ = x′ + (x + x″) = (x′ + x) + x″ = x″.

    (e)(−1) · x = −x for all x E.

    Indeed, x + (−1) · x = 1 · x + (−1) · x = [1 + (−1)] · x = 0 · x = 0 by (a) and the assertion follows from (d).

    (f)Given two vectors a and b, there exists a unique vector x such that a + x = b.

    Proof.x = b + (−a) satisfies the equation and its uniqueness follows from the fact that x must be the opposite of a + (−b).

    The element a + (−b) is usually written as a b and called the difference of the vectors a and b.

    DEFINITION 1.Given a vector space E, a norm on E is a map from E into the set R+ of positive real numbers which satisfies the following axioms:

    (N 1)||x|| = 0 if and only if x = 0.

    (N 2)||λx|| = |λ| · ||x|| for all λ K and x E.

    (N 3) (the triangle inequality).

    A vector space on which a norm is defined is called a normed vector space or simply a normed space.

    EXAMPLE 1.The n-dimensional real Euclidean space Rn is the set of all n-tuples x = (x1, …, xn) of real numbers, where addition and multiplication by a scalar λ R are defined by

    The zero vector is (0, …, 0). The eight axioms of a real vector space can be easily verified. The norm on Rn will be denoted by |x| rather than ||x||, and is defined by

    The axioms (N 1) and (N 2) can be readily verified. To prove the triangle inequality, let us first prove the Cauchy-Schwarz inequality:

    The quadratic polynomial in λ,

    is positive for every real λ. Now we know from elementary algebra that if for every real λ, then necessarily . Applying this to the polynomial (2), we obtain

    which is just another form of (1).

    We have, using the Cauchy-Schwarz inequality,

    If we take the positive square roots on both sides, we have also verified (N 3).

    The norm |x| can be thought of as the length of the line segment going from the origin 0 to the point x. The corresponding expression for n = 2 and n = 3 is well known from analytic geometry. The triangle inequality expresses the fact that the length of one side of a triangle is less than or equal to the sum of the lengths of the other two sides. This accounts for its name.

    In a similar fashion the set Cn of all n-tuples z = (z1, …, zn) of complex numbers is a complex normed vector space if we define

    See Example 10.

    EXAMPLE 2.Let I = [a, b] be a finite closed interval of the real line and let or be the set of all continuous functions defined on I and whose values are real numbers. The function f + g is defined by and the function λf (λ R) is defined by . It can be immediately verified that with these operations becomes a vector space over R, where the zero vector is the function which is identically zero for every x I. The norm of is defined by

    It is well known that this maximum exists (Weierstrass’ theorem [2], Theorem 4–20, p. 73). The first two properties of a norm are quite obviously satisfied. To see the third one we observe that for any x I we have

    whence

    If we consider the set of complex-valued functions, we obtain in a similar way a normed vector space over C.

    The norm of a vector space defines a metric in a natural way. Let us recall that a metric on a set X is a map from X × X into the set R+ of positive real numbers which satisfies the following axioms:

    (M 1)δ(x, y) = 0 if and only if x = y.

    (M 2)δ(x, y) = δ(y, x).

    (M 3) (the triangle inequality).

    The value δ(x, y) is called the distance between the points x and y. A set X on which a metric δ is defined is called a metric space, and we say that X is equipped with the metric δ.

    If E is a normed vector space, we set δ(x, y) = ||x y||. The three axioms for a metric are verified:

    (1)

    , where we use (N 1).

    (2)y x = (−1)(x y) and thus by Axiom (N 2)

    ||y x|| = |− 1| · ||x y|| = ||x y||,

    i.e., δ(y, x) = δ(x, y).

    (3)

    by (N 3).

    In the sequel we shall always consider a normed vector space as a metric space equipped with the metric just defined.

    In a metric space X (and thus in a normed vector space) we have the usual notions of topology: closed and open sets, neighborhoods, convergence, etc. (see §2). Let us recall in particular that a sequence (xn)nN of points of X is said to converge to a point x, if for every > 0 there exists an N = N( ) ∈ N such that δ(x, xn) < for every n > N. A sequence is said to be convergent if it converges to some point.

    Let (xn)nN be a convergent sequence. Then it satisfies the following so-called Cauchy condition: For every > 0 there exists N = N( ) ∈ N such that δ(xn, xm) < for n > N and m > N. Indeed we have δ(x, xn) < /2 for n > N, δ(x, xm) < /2 for m > N, and hence by the triangle inequality

    Let us call a sequence which satisfies the Cauchy condition a Cauchy sequence. We have just proved that every convergent sequence is a Cauchy sequence. If, conversely, every Cauchy sequence is convergent, then X is said to be a complete metric space. It is a basic property of the real line R (and also of the complex plane C) that it is a complete metric space. The rational line Q is not a complete metric space.

    DEFINITION 2.A normed vector space E is called a Banach space if it is complete as a metric space.

    EXAMPLE 3.The space Rn is complete. Indeed, let

    (x(m)) = ((x1(m), …, xn(m)))

    be a Cauchy sequence. It follows from the obvious inequality

    that for every i with , the sequence of real numbers is a Cauchy sequence. Thus, by the completeness of R, there exist real numbers xi such that for m > M(i, ). Setting x = (x1, …, xn), we have

    for ; in other words, (x(m)) converges to x.

    It can be shown in an entirely analogous fashion that Cn is a complex Banach space.

    EXAMPLE 4.The space is a Banach space. Indeed, let (fn) be a Cauchy sequence in . For every x I we have

    Thus for every x I the sequence of numbers (fn(x)) converges to some number which we shall denote by f(x). The function defined on I is continuous, and (fn) converges to f in the space . Indeed, for every > 0 there exists an integer M such that ||fn fm|| < for n, m > M, and thus |fn(x) − fm(x)| < for every x I and n, m > M. It follows that

    i.e., the sequence (fn) of continuous functions converges uniformly to the function f; hence f is continuous, i.e., . From (3) we have for n > M, and the proof is complete.

    For the convenience of the reader let us prove the theorem we have just used, according to which the limit f of a uniformly convergent sequence (fn) of continuous functions is itself continuous. Given > 0 there exists an integer N such that

    if n > N. Let n be a fixed index such that n > N. For every x0 ∈ I there exists an α = α(x0, ) > 0 such that

    if |x0 − x| < α. Thus

    if |x0 − x| < α, i.e., f is continuous at x0.

    Let us also observe that a sequence (fn) converges to f in the space if and only if it converges to f uniformly on I. Indeed, the relation ||f fn|| < is equivalent to "|f(x) − fn(x)| < for every x I."

    EXAMPLE 5.Let be the set of all continuous functions defined on Rn and which vanish at infinity. By this last condition the following is meant: for every and every > 0 there exists a ρ = ρ( , f) > 0 such that |f(x)| < if |x| > ρ. It is clear that is a vector space (real or complex according as we consider real- or complex-valued functions), and if we define again , then the requirements of a norm are satisfied. The maximum exists again; indeed, suppose that f(x) is not identically zero. Then for some x Rn we have |f(x)| = η > 0. Now |f(x)| < η/2 for |x| > ρ. Thus

    and the second maximum exists by Weierstrass’ theorem.

    Finally, is complete. Indeed, it can be shown, exactly as in the previous example, that a Cauchy sequence (fn) tends uniformly to a continuous function f. Thus we have only to prove that , i.e., that f vanishes at infinity. Let > 0. Then |f(x) − fn(x)| < /2 for some n, and |fn(x)| < /2 for |x| > ρ. Thus |f(x)| < for |x| > ρ.

    EXAMPLE 6.Denote by c0 the set of all infinite sequences x = (ξn)nN which tend to zero. Let y = (ηn), and define

    x + y = (ξn + ηn), λx = (λξn).

    Then c0 is clearly a vector space (real or complex according as we consider sequences of real or complex numbers). Define

    Since (ξn) tends to zero, this maximum is attained, and the properties of the norm are trivially verified. We leave it as an exercise to prove that c0 is complete.

    EXAMPLE 7.Let p be a real number, . Let lp be the set of all sequences x = (ξn)nN of elements of K for which the infinite series

    converges. In contrast to the previous examples, it is not at all trivial that lp is a vector space over K. We shall prove this and prove at the same time that

    is a norm on lp (the first two axioms of a norm are obviously satisfied, so that we must only prove the triangle inequality).

    We start with the inequality between the weighted arithmetic and geometric mean: Let α > 0, β > 0, and α + β = 1. Then for every u > 0, v > 0 we have

    To prove this, let us observe that the function is concave downward for t > 0, since its second derivative α(α − 1)− ² is negative because 0 < α < 1. It follows that the curve is below its tangent line at the point t = 1, i.e.,

    Setting t = u/v and multiplying both sides by v, we obtain (4).

    Next we prove the Hölder inequality. Let 1 < p < ∞ and 1/p + 1/q = 1. Then we have

    where (ξi) and (ηi) are two arbitrary finite sequences of real or complex numbers. The case yields the Cauchy-Schwarz inequality (since obviously ). For the proof let us set

    Then it follows from (4) that

    since ∑ ui = ∑ vi = 1. Multiplying by (∑ |ξj|p)¹/p(∑ |ηj|q)¹/q, we obtain (5).

    Finally we prove the Minkowski inequality,

    This is clear for p = 1. For 1 < p < ∞ we have

    Using the Hölder inequality, we have

    since (p − 1)q = p. Adding the similar estimate for ∑ |ξi + ηi|p−1 · | ηi|, we obtain

    Dividing both sides by (∑ |ξi + ηi|p)¹/q, we obtain (6).

    Let us now return to the space lp. If (ξi) ∈ lp and (ηi) ∈ lp, then it follows from (6) that

    for any n ∈ N, and thus

    Consequently, (ξi + ηi) ∈ lp, and the axioms of a vector space can now be immediately verified. The inequality (7) is just the triangle inequality.

    To conclude we prove that lp is complete. Let (x(m)) = ((ξi(m))) be a Cauchy sequence in lp. Then for each i N the numerical sequence (ξi(m))m∈N is a Cauchy sequence and therefore (ξi(m)) converges to some number ξi. There exists a constant μ > 0 such that

    for all m, n, N ∈ N. Letting n → ∞, it follows that

    and hence by the Minkowski inequality

    for all N ∈ N; i.e., the sequence x = (ξi) belongs to lp. Furthermore, for any > 0 there exists n0 ∈ N such that

    for m, n > n0 and for all N ∈ N. Letting n → ∞, it follows that

    for m > n0; and letting N → ∞, we see that ||x x(m)|| ≤ for m > n0. Thus (x(m)) tends to x, and we have proved completely that lp is a Banach space.

    EXAMPLE 8.Our last example is the space m of all bounded sequences x = (ξn). It is clear that if x = (ξn) ∈ m and y = (ηn) ∈ m, then

    x + y = (ξn + ηn) ∈ m

    and

    λx = (λ ξn) ∈ m

    The norm is now defined as

    since a maximum of the values |ξn| does not necessarily exist. The first two properties of a norm are again trivial to verify. To prove the third one, let us observe that

    . Thus . The proof of the completeness of m is left to the reader.

    A very important class of Banach spaces is constituted by the Hilbert spaces, to the definition of which we now turn.

    DEFINITION 3.A vector space E over K is called an inner product space if there is defined a map from E × E into K which has the following properties:

    (SP 1) for every x E.

    (SP 2)(x | x) = 0 if and only if x = 0.

    (SP 3) for every x E, y E.

    (SP 4)(λx + μy | z) = λ(x | z) + μ(y | z) for every λ, μ K, x, y, and z E.

    The value (x | y) is called the inner or scalar product of the vectors x and y.

    Let us observe that in the case when K = R, Axiom (SP 3) simply means that (x | y) = (y | x) for all x E, y E. It follows from (SP 3) and (SP 4) that

    for λ, μ K, x, y, z E, where of course the right-hand side is simply λ(x | y) + μ(x | z) if K = R.

    The scalar product defines a norm on E in a natural way. Indeed, setting

    we have ||x|| ∈ R+ by (SP 1) and ||x|| = 0 if and only if x = 0 by (SP 2). Furthermore,

    . Thus ||λx|| = |λ| · ||x||. To prove the triangle inequality, let us first prove the inequality

    By (SP 1) the expression

    is always positive, and in particular for real λ the quadratic polynomial in λ with real coefficients,

    is always positive. As in the proof of (1), we have

    which is just another form of (9). We now have

    Taking the positive square roots on both sides, we obtain the triangle inequality.

    DEFINITION 4.Let E be an inner product space and ||x|| the norm defined by (8). If E is complete for this norm (i.e., E is a Banach space), then E is said to be a Hilbert space.

    EXAMPLE 9.In the vector space Rn we define a scalar product by setting

    Axioms (SP 1) through (SP 4) are trivial to check and the norm obtained from the scalar product is the same as the norm introduced in Example 1. Thus Rn is a real Hilbert space.

    EXAMPLE 10.In the vector space Cn we define the scalar product of two vectors z = (z1, …, zn) and w = (w1, …, wn) by

    It is again trivial to verify that Axioms (SP 1) through (SP 4) hold, and so Cn is an inner product space over C. But we can prove exactly as in Example 3 that Cn is complete; hence Cn is a complex Hilbert space.

    EXAMPLE 11.For two vectors x = (ξi) and y = (ηi) of the space l² we define the scalar product by

    according as we consider real or complex sequences. The right-hand side series converge by virtue of inequality (5). The norms obtained from these scalar products are the same as the norms defined in Example 7, and thus l² is a real or a complex Hilbert space.

    EXAMPLE 12.The set of all continuous functions on an interval I = [a, b] with the inner product

    is an inner product space. We shall see later (Chapter 5) that it is not complete.

    We conclude this section with the proof of the general Cauchy-Schwarz inequality:

    PROPOSITION 1.Let E be an inner product space. Then

    for all x E, y E.

    Proof.For a real E the statement is equivalent to inequality (9). Let E be a complex inner product space. For x E, y E there exists an eiθ such that . But then |(x | y)| = eiθ(x | y), and by (9)

    EXERCISES

    1.Prove that the spaces c0 and m are complete.

    2.Let c be the set of all sequences x = (ξn) which converge to some finite limit. Define on c a structure of Banach space.

    3.Let I = [a, b] be a finite closed interval of the real line and let m be a strictly positive integer. Let be the set of all functions defined on I which have continuous m-th derivatives. Prove that with the norm

    the set becomes a Banach space.

    §2. Some notions from algebra and topology

    Let E be a vector space over a field K. A nonempty subset F of E is said to be a linear subspace (or simply subspace) of E if the following two conditions are satisfied:

    (SS 1)If x F and y F, then x + y F.

    (SS 2)If x F and λ K, then λx F.

    If these conditions are satisfied, then for x F we have also − x = (−1) · x F and x + (− x) = 0 ∈ F. The operations of addition and multiplication by a scalar are defined on F, because F is a subset of E, and by our conditions they always yield elements of F; we say that the operations on E induce the operations on F. It is trivial to check that under the induced operations F is a vector space over K. There are two trivial subspaces of E (except when E = {0}): the space E itself and the subspace {0}. All other subspaces are called proper.

    A linear subspace of a normed space or of an inner product space is in a natural way a normed space or an inner product space itself.

    If ()ιI is an arbitrary family of linear subspaces of E, then the intersection is also a subspace of E. Given any set A of elements of E, we can therefore speak of the smallest linear subspace containing A, i.e., the intersection of all linear subspaces containing A. This smallest linear subspace F will be called the linear subspace generated by A (or the linear hull of A), and A is a set of generators of F. The subspace F is the set of all linear combinations of elements of A, i.e., the set of all expressions , where λi K and xi A.

    A family ()ιI of elements of E is said to be algebraically free if a relation ∑ιI λιxι = 0 can hold only if all λι = 0. Equivalently, ()ιI is algebraically free if no belongs to the subspace generated by the with κ ι. A family consisting of a single nonzero element is algebraically free, and if () is an algebraically free family, then for ι κ. A subset L of E is said to be algebraically free if the family defined by the identical bijection of L onto itself is algebraically free. The elements of an algebraically free set are also said to be linearly independent. A family ()ιI which is not algebraically free is said to be linearly dependent; this means that there exists a family (λι)ιI of scalars, a finite number of which are different from zero, such that ∑ιI λιxι = 0.

    An algebraically free family of generators of a subspace F is called an algebraic basis (or Hamel basis) of F. It follows from Zorn’s lemma that every subspace F has an algebraic basis. More precisely, given a set of generators S of F and an algebraically free subset L in F such that S L, there exists an algebraic basis B of F such that S B L. In particular, an algebraic basis of a subspace F of E can always be completed to become an algebraic basis of E. An algebraic basis is a maximal algebraically free subset and also a minimal set of generators. A family ()ιI of elements of F is an algebraic basis of F if and only if every x F has a unique representation of the form x = ∑ιI λιyι (where all except a finite number of λι are equal to zero).

    If a vector space E has an algebraic basis with a finite number of elements, then every other algebraic basis of E is finite and has the same number of elements. It is clearly enough to prove that if an algebraic basis B of E has n elements, then any other algebraic basis C has at most n elements; this is a consequence of the following lemma:

    Let y1, …, yn+1 be n + 1 elements of E which are all linear combinations of n elements x1, …, xn of E. Then the elements y1, …, yn+1 are linearly dependent.

    The proof goes by induction. For n = 1 the lemma is true, since if y1 = λx and y2 = μx, then μy1 − λy2 = 0. Assume therefore that the lemma is true for n − 1. We can write

    and it is no restriction to suppose that λ11 ≠ 0. The n elements

    are linear combinations of x2, …, xn, and therefore by our induction hypothesis we have a linear relation

    where not all μj are zero. From (1) we obtain the relation

    which proves that the yj (j = 1, 2, …, n + 1) are linearly dependent. Q.E.D.

    Coming back to our original statement, suppose that C has more than n elements. Then we can pick n + 1 elements from C, and these n + 1 elements are linear combinations of the n elements of B. Hence, by the lemma, the n + 1 elements are linearly dependent, which contradicts the fact that C is an algebraic basis.

    If E has a finite algebraic basis, then the number n of elements in any algebraic basis of E is called the algebraic dimension of E and denoted by dimK E or dim E. If E does not have a finite algebraic basis, then we say that E is infinite-dimensional. It can be proved that even in the infinite-dimensional case two algebraic bases of E have the same cardinal ([7], §1, no. 12, Corollary 12 of Proposition 23 or [52], §7, 4.(2)).

    EXAMPLE 1.An algebraic basis of the vector space Rn is clearly given by the n elements e1 = (1, 0, …, 0), e2 = (0, 1, …, 0), …, en = (0, 0, …, 1). Thus the dimension of Rn over R is n, which justifies its name. The basis is called the canonical basis of Rn.

    Similarly, it can be seen that dimC Cn = n.

    If E is a vector space x E and A E, then x + A is the set of all vectors x + y with y A and is called the translate of A by x. Similarly, if A E, B E, then A + B is the set of all vectors x + y with x A, y B, and A B is the set of all vectors x y with x A, y B. Finally, if λ K and A E, then λA is the set of all vectors λx with x A.

    Before we link the algebraic concepts just recalled with the norm on the vector space, we must also recall some notions and results from the topology of metric spaces. This will be an excellent occasion to review some topological properties of the space Rn, with which the reader is anyhow expected to be familiar.

    Let X be a metric space with distance δ(x, y) defined on it (cf. §1). A subset A of X is in a natural way a metric space under the metric induced by that of X. Let a be a point of X and ρ a strictly positive real number. The set (a) of all points x X such that is called the closed ball with center a and radius ρ. A set V X is a neighborhood of a if it contains some closed ball with center a. A sequence (xn)nN of points of X converges to a if and only if for every neighborhood V of a there exists an integer N such that xn V for every n > N.

    Let A be a subset of X. A point x X is said to adhere to A if every neighborhood of x contains points of A. The point x adheres to A if and only if there exists a sequence of points of A which converges to x. The set of all points adherent to A is called the adherence or closure of A. Clearly, . If , then the set A is said to be closed. The intersection of any family of closed sets is closed; the union of a finite family of closed sets is closed.

    If A and B are two subsets of X, we denote by δ(A, B) the greatest lower bound of all numbers δ(x, y), where x varies in A, y varies in B, and we call it the distance between the sets A and B. If A = {a} is reduced to one element, we write δ(a, B) instead of δ({a}, B). We have δ(x, A) = 0 if and only if .

    The interior Å of a subset A of X is the set of those points x X which possess a neighborhood contained in A. Clearly Å A. If Å = A, then the set A is said to be open. A is open if and only if its complementary CA is closed. The union of any family of open sets is open; the intersection of a finite family of open sets is open. The total space X and the empty set ∅ are both open and closed. For any set A X the set is closed and Å is open. If A X, B X, and , then A is dense in B. If then A is everywhere dense.

    Let X and Y be two metric spaces with their respective metrics δ and η. A map f from X into Y is said to be continuous at the point a X, if for every neighborhood W of the point f(a) ∈ Y there exists a neighborhood V of a such that f(x) ∈ W for every x V (i.e., f(V) ⊂ W). This is equivalent to the following: to every > 0, there exists α > 0 such that δ(a, x) < α implies η(f(a), f(x)) < , and also to the following: for every sequence (xn) tending to a the sequence (f(xn)) tends to f(a). A map f: X Y is said to be continuous if it is continuous at every point of X. The map f is continuous if and only if for every closed (resp. open) set A in Y the set f−1(A) is closed (resp. open) in X. More generally, let be a finite family of metric spaces and let f be a map of the cartesian product into Y (i.e., a function of n variables, where xi varies in Xi). The map f is said to be continuous at the point (a1, …, an) if for every neighborhood W of the point f(a1, …, an) there exist neighborhoods Vi of the points such that f(x1, …, xn) ∈ W for (i.e., ). A continuous map from into Y is a map which is continuous at every point of .

    Two metric spaces X and Y are isometric if there exists a bijection f: X Y (called isometry) such that η(f(x), f(y)) = δ(x, y) for every x, y X. Two metric spaces X and Y are homeomorphic if there exists a bijection f: X Y (called homeomorphism) which is continuous together with its inverse f−1. Clearly, an isometry is a homeomorphism, but the converse is not necessarily true. A homeomorphism transforms closed sets into closed sets and open sets into open sets.

    Let E be a normed vector space and a E. The bijection , called translation by a, is an isometry of E onto itself, since

    ||x + a − (y + a)|| = ||x y||.

    If = (0) then (a) = + a; i.e., the ball with center a and radius ρ is obtained through translation by a from the ball with center 0 and radius ρ. If λ is a nonzero scalar, the bijection , called homothecy by λ, is a homeomorphism of E onto itself. Indeed,

    ||λx λy|| = |λ| · ||x y|| <

    if ||x y|| < /|λ|, and the inverse map, given by , is also continuous.

    PROPOSITION 1.Let E be a normed vector space. Then the map from E × E into E, the map from K × E into E, and the map from E into R+ are continuous. If E is an inner product space, then the map from E × E into K is also continuous.

    Proof.We have

    . Thus

    ||a + b − (x + y)|| <

    if

    Next we have the identity

    ξx λa = (ξ λ)(x a) + (ξ λ)a + λ(x a);

    hence ||ξx λa|| < , provided:

    The continuity of the norm follows from the inequality

    To prove this inequality let us observe that

    i.e., . For reasons of symmetry we also have . The last two inequalities imply (2).

    Finally, the continuity of the scalar product follows from the identity

    (x | y) − (a | b) = (x a | y b) + (x a | b) + (a | y b)

    similarly as in the proof of the continuity of multiplication by a scalar.

    An open cover of a set A in a metric space X is a family ()ιI of open sets such that . If J I and ()ιJ is a cover of A, we say that ()ιJ is a subcover of ()ιI. A set A X is compact if any of the following equivalent conditions is satisfied:

    (1)Every sequence (xn)nN of points of A has a subsequence which converges to a point of A.

    (2)Every open cover of A has a finite subcover (Heine-Borel-Lebesgue theorem).

    (3)If ()ι I is any family of closed sets such that is disjoint from A, then there exists a finite subfamily of ()ι I whose intersection is already disjoint from A (Cantor’s theorem).

    Let us make the important remark that these conditions are equivalent only in metric spaces and not in general topological spaces (see Chapter 2, §10). A compact set is always closed. If f is a continuous map of a metric space X into a metric space Y and A is compact in X, then f(A) is compact in Y. A set A is relatively compact if its closure is compact or, equivalently, if every sequence of points of A has a subsequence which converges to a point of X (but not necessarily of A).

    In a metric space X a set A is bounded if it is contained in some ball (a).

    In a normed vector space a bounded set is always contained in a ball (0) = with center at the origin. Indeed, if for every x in A, then

    i.e., A is contained in the ball with center 0 and radius σ = ρ + ||a||.

    A compact set is always bounded. It is a basic property of the spaces Rn and Cn that in them the converse is also true:

    Weierstrass-Bolzano theorem: A closed bounded subset of Rn or Cn is compact. A bounded set of Rn or Cn is relatively compact.

    This theorem is not true for infinite-dimensional Banach spaces; we shall give an example below. In fact we shall see in Chapter 3, §9 that if in a Banach space E every bounded set is relatively compact, then E is necessarily finite dimensional.

    EXAMPLE 2.Consider in l² (Example 1.7) the set U of all unit vectors e0 = (1, 0, 0, …), e1 = (0, 1, 0, …), …. The set U is bounded, since ||en|| = 1 for all n N. Since for n m, any ball with contains at most one element of U. Therefore U is also closed. But U is not compact, since the sets form an open cover of U, and clearly no proper subfamily of (Vn) covers U.

    A continuous function defined on a compact set is always bounded, and attains its maximum and its minimum (Weierstrass’ theorem; we have used this property in Examples 2 and 5 of §1).

    We have already introduced in the previous section the concept of a complete metric space. A compact metric space is always complete. A complete subset of a metric space is always closed. If a metric space X is not complete, then we can construct a complete metric space , called the completion of X, such that X is isometric with a subset X0 of and X0 is dense in . Most often X will be identified with X0, and hence X is considered to be a subset of . The construction of generalizes the familiar construction of the set R of real numbers from the set Q of rational numbers; we shall recall it briefly. Let S be the set of all Cauchy sequences of X. In S we introduce an equivalence relation R: two Cauchy sequences (xn) and (yn) are equivalent if the sequence (δ(xn, yn)) tends to zero. It is trivial to check that R is indeed an equivalence relation. is now the set of all equivalence classes of S modulo R (i.e., is the quotient set S/R). Let us define a metric on . If and are two elements of , let (xn) and (yn) be elements of S representing these equivalence classes. Then we set

    The following points are now easy to check: (1) the limit on the right-hand side really exists, (2) it is independent of the choice of the representatives (xn) and (yn) in the equivalence classes and , (3) is a metric on . Now let x X, let (xn) be the Cauchy sequence in which xn = x for every n N, and let be the class of (xn) modulo R. The map is clearly an isometry of X onto a subset X0 of . Let , let (xn) be a representative of in S, and let be the image in X0 of xn X under the isometry just defined. Then converges to , since

    This proves that X0 is dense in . Finally, we have to prove that is complete. Let be a Cauchy sequence in . For every n N there exists a such that . Let be the image of yn X. Then (yn) is a Cauchy sequence in X, since

    and thus (yn) defines an element . By what we have said above, converges to ; but then also converges to .

    Let X and Y be two metric spaces with their respective metrics δ and η. A map f from X into Y is said to be uniformly continuous if to every > 0 there exists an α > 0 such that η(f(x), f(y)) < for any two points x, y X satisfying δ(x, y) < α. We have a similar definition for a map from a product space into Y. A uniformly continuous map is clearly continuous; the converse is in general false. On a compact set, however, every continuous map is uniformly continuous. If f: X Y is uniformly continuous and (xn) is a Cauchy sequence in X, then (f(xn)) is a Cauchy sequence in Y.

    PROPOSITION 2.Let f be a map from a normed vector space E into a normed vector space F. The map f is uniformly continuous if for every neighborhood W of the origin in F there exists a neighborhood.V of the origin in E such that x y V implies f(x) − f(y) ∈ W for x, y E.

    Proof.Both conditions express the fact that to every > 0 there exists an α > 0 such that ||f(x) − f(y)|| < whenever ||x y|| < α.

    REMARK 1.Though we shall not need nor use the notion of uniform structure in this book, it may be worth while to note for the interested reader that the uniform structure deduced from the metric space structure of a normed vector space coincides with the uniform structure deduced from the abelian group structure. Proposition 2 is an immediate consequence of this rather obvious fact (see also Chapter 2, §9).

    PROPOSITION 3.Let E be a normed vector space. The map from E × E into E and the map from E into R+ are uniformly continuous. For every λ K the map from E into E is uniformly continuous, and for any a E the map from K into E is uniformly continuous. If E is an inner product space and a E, then the map from E into K is uniformly continuous.

    This follows immediately from the inequalities

    Let X be a metric space, Y a complete metric space and A a subset of X. Then to every uniformly continuous map f from A into Y there corresponds a unique uniformly continuous map from into Y such that for every x A (i.e., such that the restriction of to A coincides with f). Let us recall the construction of . Let x be a point which adheres to A but does not belong to A, and let (xn) be a sequence of points of A which converges to x. Then (xn) is a Cauchy sequence in X and, since f is uniformly continuous, (f(xn)) is a Cauchy sequence in Y which converges to some point y Y because Y is complete. Define . It is easy to check that y is independent of the particular choice of the sequence (xn) tending to x. For x A, we define, of course, . Let us prove that is uniformly continuous on . Let > 0 be arbitrary and let α > 0 be such that δ(x, y) < 3α implies η(f(x), f(y)) < /3 for x, y A. We claim that δ(x, y) < α implies

    Indeed, let be such that δ(x, y) < α. There exists a sequence (xn) of points of A tending to x and a sequence (yn) of points of A tending to y. It follows that there exists an m N such that and for . Hence η(f(xn), f(xm)) < /3, η(f(yn), f(ym)) < /3 for , and thus

    . We also have δ(x, xm) < α, δ(y, ym) < α, and thus

    which implies that η(f(xm), f(ym)) < /3. Finally, we obtain

    It is quite obvious that is unique.

    Of course, we have an entirely analogous theorem for a uniformly continuous map from a subset of a product of metric spaces into a complete space.

    If f is an isometry from A onto f(A), then is an isometry from onto . This shows that the completion of a metric space X is unique up to an isometry.

    A subset A of a metric space X is said to be precompact (or totally bounded) if its closure in is compact. A is precompact if and only if for every > 0 there exists a finite family a1, a2, …, an of elements of X such that given x A we can find an aj which verifies δ(x, aj) < . In other words, A is precompact if and only if for every > 0 it can be covered by a finite number of open balls (see Exercise 2) O (a1), …, O (an) with radius . Indeed, the balls O (a), where a runs through A, cover the closure of A in ; and since is compact, it can be covered by a finite number of these balls, which cover A a fortiori. To prove that the condition is also sufficient, it is enough to show that if it is satisfied, then every sequence (xn) of elements of A has a subsequence which is a Cauchy sequence. First, cover A by balls of radius 1. One of these balls contains infinitely many terms of the sequence (xn); pick one of these terms and call it . Next, cover A by balls of radius . The intersection of one of these balls with the first ball contains infinitely many terms of (xn). Pick one whose index is larger than n1 and call it . Continuing this way we obtain a sequence , where

    .

    If X is complete, then of course precompact sets are the same as relatively compact sets.

    DEFINITION 1.Two normed vector spaces E and F over the same field K are isometrically isomorphic if there exists a bijection f: E F such that

    f(x + y) = f(x) + f(y), f(λx) = λf(x), and ||f(x)|| = ||x||

    for all x, y E, λ K. Two inner product spaces E and F over the same field K are isometrically isomorphic if there exists a bijection f: E F such that

    f(x + y) = f(x) + f(y), f(λx) = λf(x), and (f(x) |f(y)) = (x | y)

    for all x, y E, λ K.

    PROPOSITION 4.Let E be a normed vector space. Then there exists a Banach space such that E is isometrically isomorphic to a linear subspace E0 of and that E0 is dense in . The space is unique up to an isometric isomorphism.

    Let E be an inner product space. Then there exists a Hilbert space such that E is isometrically isomorphic to a linear subspace E0 of and that E0 is dense in . The space is unique up to an isometric isomorphism.

    Proof.Let be the completion of the underlying metric space of E, equipped with the metric , and let E0 be the dense subspace of isometric with E. The functions

    can be transported to E0, and since they are uniformly continuous (Proposition 3), they can be extended uniquely to . Similarly, if E is an inner product space, for each a E0 the map is uniformly continuous on E0 and can be extended uniquely to , and then for each the map is uniformly continuous on E0 and can be extended uniquely to . The maps so defined satisfy Axioms (VS 1) through (VS 8), (N 2), (N 3) (resp. (SP 1), (SP 3), (SP 4)) trivially because E0 is dense in . Let us show this for instance for (VS 1). Let . Then there exist two sequences (xn), (yn) in E0 which tend to x and y respectively. By the continuity of addition (xn + yn) tends to x + y and (yn + xn) tends to y + x. But xn + yn = yn + xn, and since a sequence has at most one limit, x + y = y + x. We can see similarly that the distance already defined on coincides with the value ||x y|| of the extended norm (cf. Exercise 3 below). This implies that Axioms (N 1), resp. (SP 2), are also satisfied on .

    REMARK 2. is called the completion of E. We usually identify E with E0 and thus imbed E into .

    EXERCISES

    1.Supply complete proofs of all the statements made without proof.

    2.Prove that in a metric space X a closed ball is a closed set. The set (a) of all points x X such that δ(a, x) < ρ is called the open ball with center a and radius ρ. Prove that an open ball is an open set. Prove that V is a neighborhood of a point a if and only if it contains some open ball with center a.

    3.Prove that if X is a metric space, the mapping from E × E into R+ is uniformly continuous.

    4.Prove that if A is a bounded subset of K and B a bounded subset of the normed vector space E, then the map from A × B into E is uniformly continuous. Similarly prove that if A and B are bounded subsets of the inner product space E, then the map from A × B into K is uniformly continuous.

    5.(a) Prove that the identity

    holds for any real inner product space.

    (b)Prove that the identity

    holds for any complex inner product space.

    (c)Prove that if the underlying normed vector spaces of two inner product spaces are isometrically isomorphic, then the inner product spaces are also isometrically isomorphic.

    6.Let K be a compact subset of Rn and let be the set of all real- (or complex-) valued continuous functions defined on K. Generalizing Examples 2 and 4 of §1, define a structure of Banach space on .

    7.Let Ω be an open subset of Rn and let be the set of all real- (or complex-) valued continuous functions defined on Ω which vanish at the boundary of Ω in the following sense: for every and every > 0 there exists a compact subset K of Ω such that |f(x)| < for . Define a structure of Banach space on (cf. Example 1.5).

    §3. Subspaces

    A closed subspace F of a normed space E is a

    Enjoying the preview?
    Page 1 of 1