APP下载

Coburn Type Operators and Compact Perturbations

2021-11-03TingtingZHOUBinLIANGChaoyueWANG

Tingting ZHOU Bin LIANG Chaoyue WANG

Abstract A bounded linear operator T acting on a Hilbert space is called Coburn operator if ker(T −λ)={0} or ker(T −λ)∗={0} for each λ ∈C. In this paper, the authors define other Coburn type properties for Hilbert space operators and investigate the compact perturbations of operators with Coburn type properties. They characterize those operators for which has arbitrarily small compact perturbation to have some fixed Coburn property.Moreover, they study the stability of these properties under small compact perturbations.

Keywords Coburn type properties, Compact perturbations

1 Introduction

This paper is inspired by[1–5,13–15,18–19],where some special properties of Hilbert space operators under compact perturbations are studied. The purpose of this paper is to investigate the small compact perturbations of Coburn type properties.

Throughout this paper,C denotes the set of complex numbers,H will always denote a complex separable infinite dimensional Hilbert space. Let B(H) denote the algebra of all bounded linear operators on H, and let K(H) denote the ideal of compact operators in B(H).

LetT∈B(H). Denote byσ(T)andσp(T), the spectrum ofTand the point spectrum ofT,respectively. Denote by kerTand ranTthe kernel ofTand the range ofTrespectively.

A celebrate theorem of Coburn asserts that a nonzero Toeplitz operator on the Hardy space of the unit disk is injective or its adjoint operator is injective. This result is contained in the proof of[7,Theorem 4.1](see[8,17]also). Inspired by this fact, we give the following concepts.

Definition 1.1T∈B(H)is called a Coburn operator ifker(T−λ)={0}orker(T−λ)∗={0}for each λ∈C, denoted by T∈(C).

Definition 1.2T∈B(H)is called a generalized Coburn operator if the subspacesker(T−λ)andker(T−λ)∗are orthogonal for each λ∈C, denoted by T∈(gC).

There is a lot of work on Coburn theorem, some of which attempted to extend this result to other spaces such as Dirichlet space or the Hardy space of the bidisk (see [6, 12]). In order to state our main results, we first introduce some notations and terminologies.

LetT∈B(H).Tis called a semi-Fredholm operator, if ranTis closed and either nulTor nulT∗is finite, where nulT≜dim kerTand nulT∗≜dim kerT∗; in this case, indT≜nulT−nulT∗is called the index ofT. In particular, if −∞

and

respectively.ρs−F(T)≜Cσlre(T) is called the semi-Fredholm domain ofT. We denote

and

respectively.

Forλ0∈C andδ >0, we denoteBδ(λ0)={λ∈C:|λ−λ0|<δ}.

LetT∈B(H). Ifσis a clopen subset ofσ(T), then there exists an analytic Cauchy domain Ω such thatσ⊂Ω and [σ(T)σ]∩Ω = ∅. LetE(σ;T) denote the Riesz idempotent ofTcorresponding toσ, that is

where Γ=∂Ω is positively oriented with respect to Ω in the sense of complex variable theory.In this case, we denote H(σ;T)= ranE(σ;T). Obviously, H(σ;T) is an invariant subspace ofT. Ifλ∈isoσ(T), then {λ} is a clopen subset ofσ(T) and we simply write H(λ;T) instead of H({λ};T); if, in addition, dim H(λ;T)<∞, thenλis called a normal eigenvalue ofT. The set of all normal eigenvalues ofTwill be denoted byσ0(T). Obviously,σ0(T) consists of at most countable many points ofσ(T).

The main results of this paper are listed as follows.

Theorem 1.1Let T∈B(H). Then for each ε >0, there exists K∈K(H)with‖K‖< ε such that T+K∈(C)if and only if σ0(T)=∅.

Theorem 1.2Given T∈B(H). Then the following are equivalent:

(1)For each ε>0, there exists K∈K(H)with‖K‖<ε such that T+K∈(gC).

(2)For each λ∈σ0(T), there exist unit vectors eλ∈ker(λ−T)and fλ∈ker(λ−T)∗such that(eλ,fλ)=0.

Recall that two operatorsA,B∈B(H) are similar (denoted byA~B) if there exists an invertible operatorX∈B(H) such thatAX=XB. GivenT∈B(H), the similarity orbit S(T) ofTis the set {X∈B(H) :X~T}. We denote by S(gC) the set {X∈B(H) :X~Tfor someT∈(gC)}.

It is easy to see that Coburn property is invariant under similarity.

Theorem 1.3Given T∈B(H). Then for each ε>0, there exists K∈K(H)with‖K‖<ε such that T+K∈S(gC)if and only ifdim H(λ;T)≥2for each λ∈σ0(T).

Theorem 1.4Given T∈B(H)and ε>0, there exists K∈K(H)with‖K‖<ε such that T+K /∈S(gC).

Remark 1.1Obviously,T∈(C) implies thatT∈(gC). By Theorem 1.4, for anyε >0,there existsK∈K(H) with ‖K‖<εsuch thatT+K /∈(gC) andT+K /∈(C).

The rest part of this paper is organized as follows. In Section 2,we make some preparations.In Section 3, we give the proofs of Theorems 1.1–1.4. In Section 4, we define other Coburn properties and study their compact perturbations.

2 Preparations

In this part,we list some useful results. The following is the famous decomposition theorem due to Riesz.

Lemma 2.1(see [16, Theorem 2.10])Let T∈B(H). Suppose that σ(T) =σ1∪σ2,where σi(i= 1,2)are clopen subsets of σ(T)and σ1∩σ2= ∅. ThenH(σ1;T)+H(σ2;T) =H,H(σ1;T)∩H(σ2;T)={0}and T admits the following matrix representation

where σ(Ti)=σi(i=1,2).

Lemma 2.2(see [9, Corollary 3.22])Let T∈B(H). Suppose that T admits the following representation

where σ(A)∩σ(B)=∅. Then T~A⊕B.

Using the above lemmas,we can obtain the following result and leave the proof to the reader.

Corollary 2.1Let T∈B(H). Suppose that σ is a clopen subset of σ(T). Then

where σ(A)=σ and σ(B)=σ(T)σ.

Lemma 2.3(see [11, Lemma 3.2.6])Let T∈B(H). Suppose that∅/=Γ ⊂σlre(T). Then,given ε>0, there exists K∈K(H)with‖K‖<ε such thatwhere

(1)N is a diagonal normal operator of uniformly infinite multiplicity,σ(N)=σlre(N)=Γ,

(2)σ(T)=σ(A), σlre(T)=σlre(A)andind(T−λ)=ind(A−λ)for all λ∈ρs−F(T).

Ifσ⊂C andλ∈C, then we denote dist(λ,σ)=inf{|λ−µ|:µ∈σ}.

Lemma 2.4(see [19,Corollary 2.9])Given T∈B(H)and ε>0. There exists K∈K(H)with

such that σp(T+K)=

This result was proved in [19] and plays a key role in this work. To make this paper more self-contained,we outline the sketch of its proof here. Given an operatorT∈B(H),the essential minimum modulusme(T) ofTis defined by

Forγ >0, define Δγ(T)={µ∈C:me(µ−T)≤γ}. Define

By using the property ofme(T)and the continuity ofme(T;λ),one can show thatme(T;λ)≤dist(λ,∂σe(T)), ∀λ∈C. By the following result, Lemma 2.4 is clear.

Lemma 2.5(see[10,Proposition 3.4])Let T∈B(H). Given ε>0,there exists K∈K(H)with

such that σp(T+K)=

The following result characterizes the upper semi-continuity of separate parts of the spectrum.

Lemma 2.6(see [9, Theorem 1.1]Let a and b be two elements of the Banach algebraAwith identity. Assume that the spectrum σ(a)is the disjoint union of two compact subsets σ1and σ2such that σ1/= ∅, and letΩbe a Cauchy domain such that σ1⊂Ωand σ2∩Ω = ∅. If‖a−b‖

Lemma 2.7(see [9, Proposition 1.7]Let a be an element of the Banach algebraAwith identity and let f be an analytic function defined in a neighborhoodΩof σ(a). Given ε >0,there exists δ >0such that f(b)is well-defined for all b inAsatisfying‖b−a‖< δ and,moreover,

Forλ∈ρs−F(T), the minimal index ofλ−Tis defined as

Lemma 2.8(see[9,Corollary 1.14])Let T∈B(H). Then the function λ→min·ind(λ−T)is constant on every component of ρs−F(T)except for an at most denumerable subset ρss−F(T)of ρs−F(T)without limit points in ρs−F(T). Furthermore,if µ∈ρss−F(T)and λ is a point of ρs−F(T)in the same component as µ but λ /∈ρss−F(T), thenRemark 2.1(T) is called the set of singular points of the semi-Fredholm domainρs−F(T) ofT.(T)=ρs−F(T)(T) is the set of regular points. Ifλ∈ρs−F(T) and nul(λ−T)<∞(nul(λ−T)∗<∞),thenλ∈(T)if and only if ker(λ−T)⊂ran((λ−T)∗)n). Obviously, we haveσ0(T)⊂(T).

3 Proof of Main Theorems

Now we are going to prove Theorem 1.1.

Proof of Theorem 1.1“⇐=” Supposeσ0(T) = ∅. For anyε >0, by Lemma 2.4, there existsK∈K(H) with ‖K‖<εsuch that

It is easy to check thatσ(T+K) =we have ker(T+K−λ)∗={0}. Forλ∈we have ker(T+K−λ)={0}. Forλ∈σlre(T),we have ker(T+K−λ)=ker(T+K−λ)∗={0}. It follows thatT+K∈(C).

“=⇒” Ifσ0(T) /= ∅, then we arbitrarily fixλ0∈σ0(T). Chooseδ >0 such thatBδ(λ0)∩[σ(T){λ0}]=∅. Let

Thenε0>0. For anyK∈K(H)with‖K‖<ε0,by Lemma 2.6,we haveσ(T+K)∩Bδ(λ0)/=∅andσ(T+K)∩∂Bδ(λ0)=∅. Since

we haveσlre(T)∩Bδ(λ0)=∅. Henceσlre(T+K)∩Bδ(λ0)=∅. It follows thatσ(T+K)∩Bδ(λ0)consists of finite many normal eigenvalues ofT+K. Chooseµ0∈σ(T+K)∩Bδ(λ0), we haveµ0∈σ0(T+K) andµ0∈σ0(T+K)∗. It follows that ker(T+K−µ0) /= {0} and ker(T+K−µ0)∗/={0}. So we haveT+K /∈(C).

Lemma 3.1(see [14, Theorem 3.1])Let T∈B(H)and0 ∈σ0(T).and‖Kn‖ →0, then there exists a subsequenceand xk∈ker(T+Knk−λk)with‖xk‖ = 1for k≥1such that xkconverges to some x∈ker(T)in the norm topology.

Corollary 3.1Let T∈B(H)and0 ∈σ0(T). If⊂K(H)and‖Kn‖ →0, then there exists a subsequenceλk→0,xk∈ker(T+Knk−λk)with‖xk‖=1and yk∈ker(T+Knk−λk)∗with‖yk‖=1for k≥1such that xkconverges to some x∈ker(T)and ykconverges to some y∈ker(T∗)in the norm topology.

Proof of Theorem 1.2“(2)=⇒(1)”. For anyε>0, let

Thenσis a finite clopen subset ofσ(T). By Corollary 2.1,Tcan be written as

where H(σ;T) is a finite dimensional space,σ(A)=σandσ(C)=σ(T)σ. Also, we have

for eachλ∈σ0(C). By Lemma 2.4, there exists a compact operatoron H(σ;T)⊥withsuch that

and

We listσas follows

Also by Corollary 2.1,T|H(σ;T)admits the following representation

whereσ(Ai) = {λi}, Hiis a finite dimensional Hilbert space for each 1 ≤i≤k. In fact,{H1,H2,···,Hk} are mutually orthogonal and defined such that=H({λ1,···,λm};T)for each 1 ≤m≤k. By condition (2), there exist unit vectorsei∈ker(λi−T) andfi∈ker(λi−T)∗such thatei⊥fifor 1 ≤i≤k. It follows thatPHiei/= 0,PHifi/= 0 andPHiei⊥PHififor each 1 ≤i≤k. A direct calculation shows that= 0 and= 0. Applying Gramm-Schimidt process, we can choose an orthornormal basissuch thatAiadmits the following representation

Choose an operatorsuch that

whereµj/=0 for all 1 ≤j≤ni−1.

Let

ThenK∈K(H) and ‖K‖<ε. It is easy to check that

and for each 1 ≤i≤k. It follows that ker(T+K−λi)⊥ker(T+K−λi)∗for each 1 ≤i≤k.Obviously,we haveFor fixedλ∈σp(T+K),we shall show that ker(T+K−λ)and ker(T+K−λ)∗are orthogonal. Ifλ=λifor some 1 ≤i≤k,by(3.3)and(3.4),ker(T+K−λ)and ker(T+K−λ)∗are orthogonal.Ifλ∈σp(C+), by (3.1)and (3.2), we haveλ /∈σp(T+K)∗. Hence ker(T+K−λ)∗={0}.So we haveT+K∈(gC).

“(1)=⇒(2)”. We directly assume thatσ0(T) consists of infinitely many points. The finite case is much easier to prove. Letσ0(T)=be an enumeration ofσ0(T). Since(1)holds forT, there existssuch thatT+Kn∈(gC) for eachn≥1.By Corollary 3.1 and diagonal process, we can findasj→∞,with= 1 such thatx(k)jconverges to somexk∈ker(T−λk) andconverges to someyk∈ker(T−λk)∗asj→∞for eachk≥1. Obviously, we have ‖xk‖ = ‖yk‖ = 1 for eachk≥1. SinceT+Knj∈(gC) for eachj≥1, we have (x(k)j,y(k)j)=0 for eachj≥1. It follows that (xk,yk)=0 for eachk≥1.

Proof of Theorem 1.3“=⇒”. We assume that there exists a sequence of compact operatorswith ‖Kn‖ →0 such thatT+Kn∈S(gC). Then we shall show that dim H(λ;T) ≥2 for eachλ∈σ0(T). Otherwise, there exists someλ0∈σ0(T) such that dim H(λ0;T)=1. Then we can choose aδ >0 such thatBδ(λ0)∩[σ(T){λ0}]=∅. By Lemma 2.6,there existsn0such thatσ(T+Kn)∩Bδ(λ0)/=∅andσ(T+Kn)∩∂Bδ(λ0)=∅for eachn≥n0. Applying Lemma 2.7, we can choosen1large enough such thatE(Bδ(λ0);T+Kn1)andE(Bδ(λ0);T) are similar. Notice that ranE(Bδ(λ0);T) = ranE(λ0;T) = H(λ0;T), we have

SinceT+Kn1∈S(gC), there existsT0∈(gC) such thatT0andT+Kn1are similar.

This means thatBδ(λ0)∩σ(T0) consists of only one point. We denote it by {µ}. Also we have dim H(µ;T0)=1. By Corollary 2.1,T0can be written as

whereµ /∈σ(C). Arbitrarily choose a unit vectore∈H(µ;T0). We have H(µ;T0) =direct calculation shows that ker(T0−µ) =SinceT0∈(gC), we have ker(T0−µ)∗⊂H(µ;T0)⊥. Notice thatµ/∈σ(C∗), we have ker(T0−µ)∗= {0}. This is contradict to the fact thatµ∈σ0(T∗0).

“⇐=”. Apply the same argument as “(2) =⇒(1)” in the proof of Theorem 1.2. For eachε>0, we can chooseK∈K(H) with ‖K‖<εsuch thatT+Kcan be written as

where H1,H2,···,Hk, H(σ;T)⊥are mutually orthogonal and dim Hi≥2 for each 1 ≤i≤k.Also, we have

and

By Corollary 2.1 and the Jordan standard theorem,T+Kcan be similar to the following operatorT1,

where

ni≥2,is an orthonormal basis of Hifor each 1 ≤i≤k.

It is easy to see that

and

It suffices to show thatT1∈(gC). For anyλ∈σp(T1), we shall show that ker(T1−λ) and ker(T1−λ)∗are orthogonal. Ifλ=λifor some 1 ≤i≤k, we have ker(T1−λi)=and ker(T1−λi)∗=Then ker(T1−λi)and ker(T1−λi)∗are orthogonal. Ifλ∈σp(C+K0),by (3.5)–(3.6), we haveλ /∈σp(T∗1). Then ker(T1−λ)∗= {0} and hence ker(T1−λ) and ker(T1−λ)∗are orthogonal. It follows thatT1∈(gC).

Proof of Theorem 1.4Arbitrarily fixε >0. Since∂σ(T)∩σlre(T) /= ∅, we chooseλ0∈∂σ(T)∩σlre(T). By Lemma 2.3, there existsK1∈K(H) with ‖K1‖

whereeis a unit vector in H andσ(A)=σ(T). We can chooseµ0/∈σ(T)such that|µ0−λ0|<.Let

ThenK2is a rank-one operator. and henceK=K1+K2. We haveK∈K(H) and ‖K‖<ε.Moreover,we have

It suffices to show thatT+K /∈S(gC). Otherwise, there existsT1∈(gC) such thatT1andT+Kare similar. Sinceµ0/∈σ(T) andσ(A)=σ(T), µ0is an isolated point inσ(T+K). It follows thatµ0∈σ0(T+K)and henceµ0∈σ0(T1). Obviously,we have H(µ0;T+K)=1 and hence H(µ0;T1)=1. ThenT1can be written as

whereµ0/∈σ(F).

It is easy to see that ker(T1−µ0) = H(µ0;T1). SinceT1∈(gC), we have ker(T1−µ0)⊥ker(T1−µ0)∗and hence ker(T1−µ0)∗⊂H(µ0;T1)⊥. Notice thatµ0/∈σ(F), it follows that ker(T1−µ0)∗={0}. This is contradict to the fact that

4 Other Coburn Type Properties and Compact Perturbations

First, we give the following definitions.

Definition 4.1T∈B(H)is called aGCoperator if the subspacesker(T−λ)∩ker(T−λ)∗={0}for each λ∈C, denoted by T∈(GC).

Obviously,T∈(gC) implies thatT∈(GC).

Definition 4.2T∈B(H)is said to have property(Re),denoted by T∈(Re),ifker(T−λ)⊂ran(T−λ)nfor each λ∈C.

Definition 4.3T∈B(H)is said to have property(P), denoted by T∈(P), ifker(T−λ)⊂ker(T−λ)∗for each λ∈C.

In this part, we get the following results.

Theorem 4.1Given T∈B(H). Then for each ε>0, there exists K∈K(H)with‖K‖<ε such that T+K∈(GC).

Theorem 4.2Given T∈B(H). Then the following are equivalent:

(1)For any ε>0, there exists K∈K(H)with‖K‖<ε such that T+K∈(Re);

(2)σ0(T)=∅.

Theorem 4.3Given T∈B(H). Then the following statements are equivalent:

(1)For any ε>0, there exists K∈K(H)with‖K‖<ε such that T+K∈(P).

(2)andH(λ;T)reduces T, T|H(λ;T)is normal for each λ∈σ0(T).

Theorem 4.4Given T∈B(H). Then the following are equivalent:

(1)For each ε>0, there exists K∈K(H)with‖K‖<ε such that T+K /∈(GC).

(2)There exists a sequence of unit vectorssuch that

as n→+∞.

Theorem 4.5Given T∈B(H)and ε>0, there exists K∈K(H)with‖K‖<ε such that T+K /∈(Re)and T+K /∈(P).

Proof of Theorem 4.1For anyε>0, let

Thenσis a finite clopen subset ofσ(T). By Corollary 2.1,Tcan be written as

where H(σ;T) is a finite dimensional space,σ(A)=σandσ(C)=σ(T)σ. Also, we have

for eachλ∈σ0(C). By Lemma 2.4, there exists a compact operatoron H(σ;T)⊥withsuch that

and

Assume that dim H(σ;T)=N. We divide the proof into two cases.

Case 1,N≥2. There exists a compact operatoracting on H(σ,T) such that

whereµi/=µjfor eachi/=j,µj/∈σ(C) for 1 ≤j≤Nandνj/= 0 for each 1 ≤j≤N−1,is an orthonormal basis of H(σ;T). Let

and

LetK=K0+K1. Then we haveK∈K(H), ‖K‖<εand

For 1 ≤i≤N, by (4.1) and (4.2) we have

It is easy to see that

and

Hence, we have

To get ker(T+K−µi)∩ker(T+K−µi)∗={0},it suffices to show thatei/∈ker(T+K−µi)∩ker(T+K−µi)∗. For 1 ≤i≤N−1, byνi/= 0, it is easy to see thatei/∈ker(T+K−µi)∗.Fori=N, byνN−1/=0, we haveeN/∈ker(T+K−µN). Hence

for each 1 ≤i≤N.

For anyλ∈σp(C+K0), by (4.1) and (4.2), we haveλ /∈σp(T+K)∗. It follows that

Sinceby (4.3) and (4.4), we have

for eachλ∈σp(T+K). HenceT+K∈(GC).

Case 2,N=1. Choose a unit vectorein H(σ;T). ThenTcan be rewritten as

We also let

Then we have

We can choose a rank-one operatorK1withsuch that

Sinceλ /∈σ(C), then ind(C+K0−λ) = 0. By (4.1) and (4.2), we have kerker(C+−λ)∗={0}. It is easy to see that ker(T+K0+K1−λ)=If (4.5) does not hold, then we havee∈ker(T+K0+K1−λ)∗. SinceB/= 0, a direct calculation shows that(T+K0+K1−λ)∗e/=0. This leads a contradiction.

Proof of Theorem 4.2“(1) =⇒(2)”. Assume that there exists a sequence of compact operatorswith ‖Kn‖ →0 such thatT+Kn∈(Re) for eachn≥1. We shall show thatσ0(T)=∅. Otherwise, arbitrarily chooseλ0∈σ0(T). By Lemma 2.6, there existsn0such thatσ0(T+Kn0)/=∅. Chooseµ0∈σ0(T+Kn0). By Lemma 2.1,T+Kn0can be written as

We denotep= dim H(µ0;T+Kn0),then 1 ≤p <∞. It follows that ker(T+Kn0−µ0) /= ∅and ker(T+Kn0−µ0) ⊂H(µ0;T+Kn0). On the other hand, we have (A−µ0)p= 0 andµ0/∈σ(B). We have ran(T+Kn0−µ0)p=H(σ(T+Kn0){µ0};T+Kn0). SoT+Kn0/∈(Re).This leads a contradiction.

“(2) =⇒(1)”. Assume thatσ0(T) = ∅. For eachε >0, we shall show that there existsK∈K(H) with ‖K‖<εsuch thatT+K∈(Re). By Lemma 2.5, there existsK∈K(H) with‖K‖<εsuch that

It suffices to show thatT+K∈(Re). Givenλ∈C, ifλ∈σlre(T), by (4.6), we have ker(T+K−λ)={0}. Ifλ∈ρs−F(T), we divide the proof into three cases. When ind(T+K−λ)=0,by (4.6), we have ker(T+K−λ) = {0}. When ind(T+K−λ)>0, also by (4.6), we have ker(T+K−λ)∗= {0} and hence ker((T+K−λ)n)∗= {0} for eachn∈{0}∪N. It follows that ran(T+K−λ)n= H and henceran(T+K−λ)n= H. When ind(T+K−λ)<0,by (4.6), we have ker(T+K−λ)={0}. In any case, we have

for eachλ∈C. HenceT+K∈(Re).

Proof of Theorem 4.3“(1) =⇒(2)”. Assume (1) holds forT. First, we shall showOtherwise, we can chooseλ0∈For anyK∈K(H), we have ind(T+K−λ0)>0. It follows that dim ker(T+K−λ0)>dim(T+K−λ0)∗. This means thatT+K /∈(P). This leads a contradiction.

Second, we shall show H(λ0;T) reducesTfor eachλ0∈σ0(T). Letλ0∈σ0(T). Then we can choose aδ >0 such thatBδ(λ0)∩[σ(T){λ0}] = ∅. By Lemma 2.6, there existsn0such thatσ(T+Kn)∩Bδ(λ0) /=∅andσ(T+Kn)∩∂Bδ(λ0)=∅for eachn≥n0. By Lemma 2.7,we have

asn→+∞. SinceT+Kn∈(P), it is easy to see that H(Bδ(λ0);T+Kn) reducesT+Knforn≥n0. This means thatE(Bδ(λ0);T+Kn)=E(Bδ(λ0);T+Kn)∗and

Asn→+∞, we have

By (4.7), we also haveE(Bδ(λ0);T)=E(Bδ(λ0);T)∗. SinceBδ(λ0)∩σ(T)={λ0}, so we have got

Hence H(λ0;T) reducesT.

Third, we shall show thatT|H(λ0;T)is normal for eachλ∈σ0(T). Assume that there existswith ‖Kn‖ →0 such thatT+Kn∈(P) for eachn≥1. Fixλ0∈σ0(T).Chooseδ >0 such thatBδ(λ0)∩[σ(T){λ0}] =∅. By Lemma 2.6, there existsn0such thatσ(T+Kn)∩Bδ(λ0)/=∅andσ(T+Kn)∩∂Bδ(λ0)=∅for eachn≥n0. We denote

By Lemma 2.7, we have ‖f(T+Kn)−f(T)‖ →0 asn→∞. In fact, we have thatf(T+Kn)|H(Bδ(λ0);T+Kn)=(T+Kn)|H(Bδ(λ0);T+Kn)and its spectrum consists of finite many normal eigenvalues ofT+Kn. We denoteσ((T+Kn)|H(Bδ(λ0);T+Kn)) = {µ1,µ2,···,µkn}. SinceT+Kn∈(P), we have ker(T+Kn−µi)⊂ker(T+Kn−µi)∗for 1 ≤i≤kn. It follows thatf(T+Kn) is normal and hencef(T) is normal. HenceT|H(λ0;T)=f(T)|H(λ0;T)is normal.

“(2)=⇒(1)”. For anyε>0, let

Thenσis a finite clopen subset ofσ(T).By Corollary 2.1 and condition (2),Tcan be written as

whereAis a normal operator acting on the finite dimensional space H(σ;T),σ(A) =σandσ(C)=σ(T)σ. Then it follows that

for eachλ∈σ0(C). By Lemma 2.4, there exists a compact operatoron H(σ;T)⊥withsuch thatLet

ThenK∈K(H) and ‖K‖<ε. Also, we have

So we haveσp(T+K)=σ. SinceAis normal on finite dimensional space,we have ker(A−λ)=ker(A−λ)∗for eachλ∈σ. It follows that

for eachλ∈C.This finishes the proof.

Proof of Theorem 4.4“(1) =⇒(2)”. For anyε >0, we shall choose a unit vectoresuch that ‖T e‖2−|(T e,e)|2+‖T∗e‖2−|(T∗e,e)|2< ε. By (1), there existsK∈K(H) withsuch thatT+K /∈(GC). Hence there existsλ0∈C such that ker(T+K−λ0) ∩ker(T+K−λ0)∗/={0}. We can choose a unit vectore∈ker(T+K−λ0)∩ker(T+K−λ0)∗.Then we have

We rewriteKas follows

It follows that

Since ‖T e‖2−|(T e,e)|2= ‖K21e‖2≤‖K‖2and ‖T∗e‖2−|(T∗e,e)|2=we have

“(2)=⇒(1)”. Assume (2) holds, for anyε>0, we can choosen0∈N such that

ThenTcan be written as

Let

Obviously,K∈K(H). By (4.7), it is easy to see that ‖K‖<ε. Then we have

It follows thaten0∈ker(T−K−λ)∩ker(T−K−λ)∗and henceT−K /∈(GC).

Proof of Theorem 4.5Similarly to the proof of Theorem 1.4, for eachε >0, we can choose a compact operatorKwith ‖K‖<εsuch that

whereµ0is a normal eigenvalue ofT+KandB/=0. By Lemma 2.8 and Remark 2.1,it is easy to see thatµ0is a singular point inρs−F(T+K). Hence“ker(T+K−µ0)⊂ran((T+K−µ0)n)”does not hold. Also, it is easy to see thate∈ker(T+K−µ0)ande /∈ker(T+K−µ0)∗. HenceT+K /∈(Re) andT+K /∈(P).

AcknowledgementThe authors would like to thank the referee for his/her valuable comments. The authors also wish to thank Professor Chunguang Li for valuable suggestions concerning this paper.


登录APP查看全文