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

Only $11.99/month after trial. Cancel anytime.

Invariant Subspaces
Invariant Subspaces
Invariant Subspaces
Ebook425 pages4 hours

Invariant Subspaces

Rating: 0 out of 5 stars

()

Read preview

About this ebook

This broad survey spans a wealth of studies on invariant subspaces, focusing on operators on separable Hilbert space. Largely self-contained, it requires only a working knowledge of measure theory, complex analysis, and elementary functional analysis. Subjects include normal operators, analytic functions of operators, shift operators, examples of invariant subspace lattices, compact operators, and the existence of invariant and hyperinvariant subspaces. Additional chapters cover certain results on von Neumann algebras, transitive operator algebras, algebras associated with invariant subspaces, and a selection of unsolved problems. 1973 edition. New appendix on recent developments.
LanguageEnglish
Release dateNov 30, 2011
ISBN9780486153025
Invariant Subspaces

Related to Invariant Subspaces

Titles in the series (100)

View More

Related ebooks

Mathematics For You

View More

Related articles

Reviews for Invariant Subspaces

Rating: 0 out of 5 stars
0 ratings

0 ratings0 reviews

What did you think?

Tap to rate

Review must be at least 10 words

    Book preview

    Invariant Subspaces - Heydar Radjavi

    References

    Chapter 0. Introduction and Preliminaries

    Most of this chapter consists of certain basic definitions and results that will be required later. With the exception of Section 0.4 the results are all very well-known. The reader may wish to use the other sections merely for reference regarding notation.

    0.1 Hilbert Space

    .

    We assume familiarity with the geometry of Hilbert space, as treated in Rudin [1, for all α.

    A linear manifold which is closed under vector addition and under multiplication by complex numbers. A subspace is a linear manifold which is closed in the norm topology; the trivial subspaces then the span are complementary, or are complements . The orthogonal complement .

    An operator . We use the term linear transformation . The identity operator is simply denoted by 1, and the adjoint of A is denoted by Aare complementary subspaces, then the projection on along is the operator P defined by P(x + y) = x ; (continuity of P is called simply the projection on . We assume the basic properties of operators as discussed in Halmos [2], Berberian [1], and Taylor [1]. Convergence of the net {} to A as an operator matrix; (cf. Halmos [3]).

    A unitary operator U is unitary if and only if U is invertible and (Ux, Uy) = (x, y. The operators A and B are unitarily equivalent if there is a unitary operator U such that A = U B U − ¹. We shall generally be interested in unitarily invariant properties of operators, and will often fail to distinguish between unitarily equivalent operators. The operators A and B are similar if there is an invertible operator S such that A = SBS − ¹.

    0.2 Invariant Subspaces

    Definition. is invariant under the operator A invariant under A is denoted Lat A.

    , Lat A is obviously closed under the formation of intersections and spans, and it follows that Lat A . Note that the trivial subspaces are in Lat A for every A.

    , then the representation of A is upper triangular:

    (the restriction of A and where A2 and Arespectively. Thus it is not surprising that there exist various relations between the structure of A and of Lat A.

    is easily seen to be invariant under A. Therefore knowledge of Lat A gives information about the vectors which can be approximated by linear combinations of {Anx}.

    The vector x is cyclic for A is a cyclic subspace for A

    Definition. is hyperinvariant for A for every B which commutes with A.

    Knowledge of the hyperinvariant subspaces of A can give information about the structure of the commutant of A, (the set of all operators B such that AB = BA).

    Much less is known about invariant subspaces than is unknown; the results presented in this book may eventually be fragments of a much more comprehensive theory. The most fundamental unsolved problem, the invariant subspace problem, is: does every operator have a non-trivial invariant subspace? A related question is the hyperinvariant subspace problem: does every operator that is not a complex multiple of 1 have a non-trivial hyperinvariant subspace? Some partial results on the invariant subspace and hyperinvariant subspace problems are obtained in Chapters 5, 6 and 8.

    We begin with some basic facts.

    Theorem 0.1. If and P is any projection onto , then if and only if AP = P AP.

    Proofthen APx is contained in A it follows that P(APx) = APx. Conversely, if AP = PAP , then Px = x and Ax = PAx. Since P(Ax) = Ax

    Theorem 0.2. If and P is the projection on along , then and are both in Lat A if and only if AP = PA.

    Proof. By if and only if AP = PAP and A(1 − P) = (1 − P) A(1 − P), (since 1 − P ). The second equation is equivalent to A AP = A PA AP + PAP, or 0 = − PA + PAP. The first equation gives 0 = − PA + AP

    Definition. reduces A are both in Lat A.

    reduces A , (reduces A commutes with A. Reducing subspaces are generally easier to treat than arbitrary invariant subspaces, (cf. Chapter 7).

    0.3 Spectra of Operators

    We require a number of facts about spectra of operators. It is convenient to consider the more general situation of spectra of elements of Banach algebras.

    Definition. , then the spectrum σ(x) of x is the set of all complex numbers λ such that x λ ; (x λ is an abbreviation for x λ⋅1). The resolvent set ρ(x) is the complement of σ(x, (the set of complex numbers), and the spectral radius r(x. For each x, σ(x; (cf. Rudin [1], Rickart [1]). The element x is quasinilpotent if σ(xis commutative, the Gelfand theory provides a useful characterization of σ(x).

    Theorem 0.3. If is a complex, commutative Banach algebra with identity and , then

    Proof

    , σ(A) denotes the spectrum of A ; the following theorem makes this possible.

    Theorem 0.4. If is a maximal commutative subalgebra of the complex Banach algebra , if and denote the spectra of x relative to and respectively, then .

    Proof. Clearly if x λ . Suppose that (x λ)y = y(x λ, yz(x λ)y = y(x λ)zy, so that yz = zy. Hence y

    Theorem 0.5 (Spectral Mapping Theorem). If and p is any polynomial, then .

    Proofcontaining x. Then, by Theorems 0.3 and 0.4,

    If A is an operator, there are several possible reasons that A λ may fail to be invertible.

    Definition. The point spectrum of A, denoted Π0(Asuch that A λ , then λ is an eigenvalue of A, and there exist non-zero vectors x (eigenvectors of A) such that Ax = λx.

    Note that an invertible operator A must be bounded below, in the sense that there exists a constant k for all vectors x.

    Definition. The approximate point spectrum of A, denoted Π(Asuch that A λ is not bounded below. The compression spectrum Γ(A) of A is the set of all λ such that the range of A λ is not dense.

    .

    Theorem0.6. If , then .

    Proofimplies, as is easily verified by considering sequential limit points, that the range of A λ , the range of A λ is also dense, and it follows that A λ

    The next result is useful for the study of invariant subspaces. For subsets S we use ∂S to denote the point-set boundary of S.

    Theorem0.7. If .

    Proofwhich converges to λ. We claim that there is a k > 0 and a positive integer N for all x. If this were not the case, then for all positive integers m and N and a vector xm . But

    . Hence there do exist such k and N.

    , then for each n with (A λn)yn = x. Then

    If n .

    , (

    This theorem gives information about the spectra of restrictions to invariant subspaces.

    Definition. , then the full spectrum of A, denoted η(σ(A)), is the union of σ(A) and all the bounded components of ρ(A); i.e., η(σ(A)) is σ(A) together with the holes in σ(A).

    Theorem 0.8. If , then .

    Proofimplies that there exists a sequence {xn} of unit vectors such that ||(A λ)xn. Now, by Theorem 0.7,

    contained points of the unbounded component of ρ(Awould have to meet the unbounded component of ρ(A

    need not be contained in σ(A); e.g., let A (see Chapter 3).

    Definition. The operator A has finite rank is finite.

    Clearly finite-rank operators can be studied using the techniques of finite-dimensional linear algebra, (cf. Proposition 0.5), and therefore we can obtain a great deal of information about them. A wider class of operators to which certain generalizations of finite-dimensional techniques can be applied is the class of compact operators.

    Definition. The operator A is compact, (or, completely continuous), if the closure of A ; equivalently, A , ({xn} converges weakly to x).

    , (Proposition 0.6). We state without proof the well-known spectral properties of compact operators.

    Theorem 0.9. If A is compact, then

    (i) (The Fredholm alternative) .

    (ii) Π0(A) is either finite or consists of a sequence converging to 0.

    (iii) The eigenspace {x: Ax = λx} is finite-dimensional if λ ≠ 0.

    Proof

    We require a result on the relation between the spectrum of an operator and the spectrum of its perturbation by a compact operator.

    Theorem 0.10 (Weyl’s Theorem). If and K is a compact operator, then .

    Proof. Then A + K λ = (A λ)(1 + (A λ) − ¹K). Since A λ is invertible, (1 + (A λ) − ¹K) is not invertible, and it follows from the Fredholm alternative that − 1 is an eigenvalue of the compact operator (A λ) − ¹K. Hence A + K λ

    We shall have occasion to consider a slightly wider class of operators.

    Definition. The operator A is polynomially compact if there exists a non-zero polynomial p such that p(A) is compact.

    0.4 Linear Operator Equations

    Properties of the operator equation AX XB = Y, where A and B are given operators on possibly different Hilbert spaces, will be used several times in the sequel. We begin with a simple lemma.

    Lemma 0.11. If A and B are commuting operators on the Banach space , then

    for every polynomial p in two variables.

    Proofcontaining A and B. By Theorems 0.3 and 0.4,

    .

    Theorem 0.12 (Rosenblum’s Theorem). If A and B are operators on the Hilbert spaces and respectively, and if the operator on is defined by , then

    Proof, then (A λ)(A λ) − ¹ X = (A λ) − ¹(A λ)X . By

    The notation of the above theorem is retained for the next corollary.

    Corollary 0.13 (Rosenblum’s Corollary). If , then for each operator Y from to there is a unique operator X from to such that AX XB = Y. In particular, AX = XB implies X = 0.

    Proof. This follows immediately from

    We present two applications of Rosenblum’s theorem below; it will also be used in subsequent chapters.

    we use A B defined by (A B)(x y) = Ax By; matricially, A B .

    Corollary 0.14. If and S is an operator on which commutes with A B, then S = C D where C commutes with A and D commutes with B.

    Proof. Then

    yields, by matrix multiplication, AE = EB and BF = FA. Rosenblum’s corollary implies E = 0 and F

    Corollary 0.15. If , then the operator is similar to A B.

    Proof), so it suffices to show that there is an X such that

    This is equivalent to finding an X satisfying the equation C + XB = AX. Such an X

    0.5 Additional Propositions

    Proposition 0.1. For any operator A.

    Proposition 0.2. separable), is a separable metric space in the weak operator topology.

    Proposition 0.3. For any A and Bare complementary invariant subspaces for A.

    Proposition 0.4. If A is compact, then there is a sequence of finite- rank operators which converges uniformly to A.

    Proposition 0.5. If A is a finite-rank operator, then A is unitarily equivalent to an operator of the form B 0, where B is an operator on a finite-dimensional space.

    Proposition 0.6. .

    Proposition 0.7 , then

    Proposition 0.8. If A and B commute and B is quasinilpotent, then σ(A + B) = σ(A).

    Proposition 0.9. If K denote the nullspace of (K λ)nis finite-dimensional.

    Proposition 0.10. The operator A is compact if and only if A* is compact.

    Proposition 0.11. is finite, then A is compact and

    for every orthonormal basis {fn}.

    Proposition 0.12. is a proper ideal.

    0.6 Notes and Remarks

    The evolution of the basic ideas concerning Hilbert space and linear operators is discussed in Dunford-Schwartz ([1], [2]).

    The elementary but powerful results on spectra in commutative Banach algebras are due to Gelfand [1]. More of the elementary theory can be found in Dunford-Schwartz [2], and a deeper account of the general theory of Banach algebras is presented in Rickart [1].

    Theorem 0.8, which generalizes the result of Bram [1] in the case where A is normal, is due to Scroggs [1]; it was independently found by S. Parrott (see Crimmins-Rosenthal [1]). Theorem 0.10 was discovered by Weyl [1]. Theorem 0.12 is due to Rosenblum [1] in the case where A and B . The simple proof presented above was found by Lumer and Rosenblum [1], and the fact that it applied in the case where A and B are operators on different spaces was observed in Radjavi-Rosenthal [5]. Corollaries 0.14 and 0.15 are taken from Rosenthal [2] and Rosenthal [3] respectively; each has undoubtedly been observed by others.

    Chapter 1. Normal Operators

    The spectral theorem exhibits normal operators as integrals with respect to some of their reducing subspaces. This is the most satisfying structure theorem known for operators on infinite-dimensional spaces: the class of normal operators is a large and useful class of operators and the spectral theorem is an appropriate tool for answering many questions about them. It should be noted, however, that the spectral theorem does not answer all questions about normal operators; in particular, there are still a number of unsolved problems about invariant subspaces of normal operators, (see Section 10.1).

    One result which follows from the spectral theorem is the existence of hyperinvariant subspaces for normal operators—this is Theorem 1.16 (Fuglede’s theorem), which is also useful in other contexts.

    1.1 Preliminaries

    The operator A is normal if it commutes with A*, Hermitian if A = A*, positive for all x, and unitary if A* = A − ¹. One can study normal operators by first proving the spectral theorem for Hermitian operators and then deriving the general case (cf. Halmos [2]) but we shall present a more direct approach.

    Theorem 1.1. If A is normal, then ||A|| = r(A).

    Proof. By the spectral radius formula (page 4) it is sufficient to show that ||An|| = ||A||n for infinitely many n. For this it suffices to show that ||A²|| = ||A||²; induction then gives ||A²k|| = ||A||²k for all k. This follows from the computation

    (we have used the easily verified fact (cf. Halmos [2], p. 40) that ||B* B|| = ||B||² for every operator B

    Theorem 1.2. If A is normal and Ax = λx.

    Prooffor all x

    Theorem 1.3 (Spectral Mapping Theorem for Normal Operators). If A is normal and p(⋅, ⋅) is any polynomial in two variables, then

    Proofwhich contains A and A*; by Theorems 0.4 and 0.3,

    Since {ϕ(A): ϕ } = σ(Afor every ϕ.

    Suppose that ϕ(A) = a + bi, ϕ(A*) = c + di. We first show that b + d = 0. To see this, consider the Hermitian operator B = A + A*. Then ϕ(B) = (a + c) + i(b + d). On the other hand, σ(B) is real; (if H is Hermitian, then (Hx, x) is real for all x, hence Π(Himplies that σ(H) is real). It follows that b + d = 0.

    To complete the proof we need only show that a = c. For this we simply apply the above argument to the operator iA

    1.2 Compact Normal Operators

    It follows immediately from Theorem 1.2 that a normal operator on a finite-dimensional space is unitarily equivalent to a diagonal operator, or, in other words, that given a normal operator on a finite-dimensional space there exists an orthonormal basis for the space consisting of eigenvectors of the operator. The same statement is easily proven for compact normal operators on infinite-dimensional spaces.

    Theorem 1.4 (Spectral Theorem for Compact Normal Operators). If A is a compact normal operator, then there exists an orthonormal basis {} such that each eα is an eigenvector of A.

    Proofconsisting of eigenvectors of A.

    . By Theorem 1.2 the one-dimensional space spanned by reduces A for each αreduces Awere 0, the set {is not empty. Thus A , contradicting the maximality of {

    This theorem can be reformulated as follows.

    Corollary 1.5 (Spectral Theorem for Compact Normal Operators). The operator A is compact normal if and only if there exists an orthonormal basis and a sequence of complex numbers such that for all x.

    Proof. It is obvious that every operator of the given form is compact and normal. The converse follows immediately from Theorem 1.4; ({en} − 0 implies {||Aen||} → 0 for compact operators A

    1.3 Spectral Theorem—First Form

    One way of making examples of normal operators is the following. Let (X, μ. The multiplication operator corresponding to ϕ is the operator defined by (Mϕ f)(x) = ϕ(x)f(x. It is easily verified that |||| = ||ϕ||∞ and that σ() = essential range of ϕ = {λ: μ{x: |ϕ(x) − λ| < ε} > 0 for all ε . Thus ()* = ()* = M|ϕ|², and is normal.

    One version of the spectral theorem states that every normal operator is essentially a multiplication operator.

    Theorem 1.6 (Spectral Theorem—First Form). If A is a normal operator on a separable space, then there exists a finite measure space (X, μ) and a such that A is unitarily equivalent to the operator Mϕ on

    Proof. We first consider the special case where there exists a cyclic vector f . The general case will then be shown to follow easily from this special case.

    , or, in other words, that the closure of {p(A, A*)f: p . In this case we shall show that X can be taken as a subset of the complex plane and ϕ the identity function, (i.e., ϕ(z) = z ). The idea behind the proof is the observation that Ap(A, A*)f = q(A, A*)f, where q(z1,

    Enjoying the preview?
    Page 1 of 1