This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Maps preserving the local spectral subspace of skew-product of operators

R. Parvinianzadeh *** Corresponding author:
Department of Mathematics, University of Yasouj, Yasouj 75918, Iran
E-mail addresses: r.parvinian@yu.ac.ir.

Abstract Let B(H)B(H) be the algebra of all bounded linear operators on an infinite-dimensional complex Hilbert space HH. For TB(H)T\in B(H) and λ\lambda\in\mathbb{C}, let HT({λ})H_{T}(\{\lambda\}) denotes the local spectral subspace of TT associated with {λ}\{\lambda\}. We prove that if φ:B(H)B(H)\varphi:B(H)\rightarrow B(H) be an additive map such that its range contains all operators of rank at most two and satisfies

Hφ(T)φ(S)({λ})=HTS({λ})H_{\varphi(T)\varphi(S)^{\ast}}(\{\lambda\})=H_{TS^{\ast}}(\{\lambda\})

for all T,SB(H)T,S\in B(H) and λ\lambda\in\mathbb{C}, then there exist a unitary operator VV in B(H)B(H) and a nonzero scalar μ\mu such that φ(T)=μTV\varphi(T)=\mu TV^{\ast} for all TB(H)T\in B(H). We also show if φ1\varphi_{1} and φ2\varphi_{2} be additive maps from B(H)B(H) into B(H)B(H) such that their ranges contain all operators of rank at most two and satisfies

Hφ1(T)φ2(S)({λ})=HTS({λ})H_{\varphi_{1}(T)\varphi_{2}(S)^{\ast}}(\{\lambda\})=H_{TS^{\ast}}(\{\lambda\})

for all T,SB(H)T,S\in B(H) and λ\lambda\in\mathbb{C}. Then φ2(I)\varphi_{2}(I)^{\ast} is invertible, and φ1(T)=T(φ2(I))1\varphi_{1}(T)=T(\varphi_{2}(I)^{\ast})^{-1} and φ2(T)=φ2(I)T\varphi_{2}(T)=\varphi_{2}(I)^{\ast}T for all TB(H)T\in B(H).

Mathematics Subject Classification: Primary 47B11: Secondary 47A15, 47B48.

Keywords: Local spectrum, Local spectral subspace, Nonlinear preservers, Rank-one operators.

1 Introduction

Throughout this paper, HH and KK are infinite-dimensional complex Hilbert spaces. As usual B(H,K)B(H,K) denotes the space of all bounded linear operators from HH into KK. When H=KH=K we simply write B(H)B(H) instead of B(H,H)B(H,H), and its unit will be denoted by II. The inner product of HH or KK will be denoted by ,\left\langle,\right\rangle if there is no confusion. For an operator TB(H,K)T\in B(H,K), let TT^{\ast} denote as usual its adjoint. A preserver problem generally deals with characterizing those maps on some specific algebraic structures which preserve a particular subset, property or relation. This subject has a long history and its origins goes back well over a century to the so-called first linear preserver problem, due to Frobenius [11], that determines linear maps preserving the determinant of matrices. As we mentioned earlier, the main of this subject goal is to describe the general form of linear maps between two Banach algebras which preserve a certain property, or a certain class of elements, or a certain relation. One of the most famous related problems is Kaplansky’s problem [17] asking whether every surjective unital invertibility preserving linear map between two semisimple Banach algebras is a Jordan homomorphism. His question was motivated by two classical results, the result of Marcus and Moyls [18] on linear maps preserving eigenvalues of matrices and the Gleason-Kahane-Zelazko theorem [15, 16] stating that every unital invertibility preserving linear functional on a unital complex Banach algebra is necessarily multiplicative. The later this result was obtained independently by Gleason in [15] and Kahane-Zelazko in [16], and was refined by Zelazko in [23]. In the non-commutative case, the best known result so far are due to Sourour [22]. He answered to the Kaplansky’s question in the affirmative for bijective unital linear invertibility preserving maps acting on the algebra of all bounded operators on a Banach space. Note that when the maps are unital, then preserving invertibility is equivalent to preserving spectrum. These results opened the gate for many authors who investigate linear (or additive) maps preserving spectrum; see for instance [1, 12, 13] and the references therein. Along this line, Molnar [19] investigated maps preserving the spectrum of operator products without assuming linearity or additivity.

The local resolvent set, ρT(x)\rho_{T}(x), of an operator TB(H)T\in B(H) at a point xHx\in H is the union of all open subsets UU of the complex plane \mathbb{C} for which there is an analytic function f:UHf:U\longrightarrow H such that (μIT)f(μ)=x(\mu I-T)f(\mu)=x for all μU\mu\in U. The complement of local resolvent set is called the local spectrum of TT at xx, denoted by σT(x)\sigma_{T}(x), and is obviously a closed subset (possibly empty) of σ(T)\sigma(T), the spectrum of TT. We recall that an operator TB(H)T\in B(H) is said to have the single-valued extension property (henceforth abbreviated to SVEP) if, for every open subset UU of \mathbb{C}, there exists no nonzero analytic solution, f:UHf:U\longrightarrow H, of the equation

(μIT)f(μ)=0,μU.(\mu I-T)f(\mu)=0,\quad\forall\leavevmode\nobreak\ \mu\in U.

Every operator TB(H)T\in B(H) for which the interior of its point spectrum, σp(T)\sigma_{p}(T), is empty enjoys this property.

For every subset FF\subseteq\mathbb{C} the local spectral subspace HT(F)H_{T}(F) is defined by

HT(F)={xH:σT(x)F}.H_{T}(F)=\{x\in H:\sigma_{T}(x)\subseteq F\}.

Clearly, if F1F2F_{1}\subseteq F_{2} then HT(F1)HT(F2)H_{T}(F_{1})\subseteq H_{T}(F_{2}). For more information about these notions one may see the books [2, 20].

The study of linear and nonlinear local spectra preserver problems attracted the attention of a number of authors. Bourhim and Ransford were the first ones to consider this type of preserver problem, characterizing in [8] additive maps on B(X)B(X), the algebra of all linear bounded operators on infinite-dimensional complex Banach space XX, that preserves the local spectrum of operators at each vector of XX. Their results motivated several authors to describe maps on matrices or operators that preserve local spectrum, local spectral radius, and local inner spectral radius; see, for instance, the last section of the survey article [5] and the references therein. Based on the results from the theory of linear preservers proved by Jafarian and Sourour [14], Dolinar et al. [9], characterised the form of maps preserving the lattice of sum of operators. They showed that the map (not necessarily linear) φ:B(X)B(X)\varphi:B(X)\rightarrow B(X) satisfies Lat(φ(T)+φ(S))=(\varphi(T)+\varphi(S))=Lat(T+S)(T+S) for all T,SB(X)T,S\in B(X), if and only if there are a non zero scalar α\alpha and a map ϕ:B(X)𝔽\phi:B(X)\rightarrow\mathbb{F} such that φ(T)=αT+ϕ(T)I\varphi(T)=\alpha T+\phi(T)I for all TB(X)T\in B(X) (See [9, Theorem 1]), where 𝔽\mathbb{F} is the complex field \mathbb{C} or the real field \mathbb{R} and Lat(T)(T) is denoted the lattice of TT, that is, the set of all invariant subspaces of TT. They proved also, in the same paper, that a not necessarily linear maps φ:B(X)B(X)\varphi:B(X)\rightarrow B(X) satisfies Lat(φ(T)φ(S))=(\varphi(T)\varphi(S))= Lat(TS)(TS) (resp. Lat(φ(T)φ(S)φ(T))=(\varphi(T)\varphi(S)\varphi(T))= Lat(TST)(TST), resp. Lat(φ(T)φ(S)+φ(S)φ(T))=(\varphi(T)\varphi(S)+\varphi(S)\varphi(T))= Lat(TS+ST)(TS+ST)) for all T,SB(X)T,S\in B(X), if and only if there is a map ϕ:B(X)𝔽\phi:B(X)\rightarrow\mathbb{F} such that φ(T)0\varphi(T)\neq 0 if T0T\neq 0 and φ(T)=ϕ(T)T\varphi(T)=\phi(T)T for all TB(X)T\in B(X) (See [9, Theorem 2]).

For a Banach space XX, it is well-known that XT(F)X_{T}(F), the local spectral subspace of TT associated with a subset FF of \mathbb{C}, is an element of Lat(T)Lat(T), so one can replace the lattice preserving property by the local spectral subspace preserving property. In [10], the authors described additive maps on B(X)B(X) that preserve the local spectral subspace of operators associated with any singleton. More precisely, they proved that the only additive map φ\varphi on B(X)B(X) for which Xφ(T)({λ})=XT({λ})X_{\varphi(T)}(\{\lambda\})=X_{T}(\{\lambda\}) for all TB(X)T\in B(X) and λ\lambda\in\mathbb{C}, is the identity. In [4], Benbouziane et al. characterized the forms of surjective weakly continuous maps φ\varphi from B(X)B(X) into B(X)B(X) which satisfy

Xφ(T)φ(S)({λ})=XTS({λ}),(T,SB(X),λ).X_{\varphi(T)-\varphi(S)}(\{\lambda\})=X_{T-S}(\{\lambda\}),\quad(T,S\in B(X),\lambda\in\mathbb{C}).

Afterwards, in [3], the authors studied surjective maps that preserve the local spectral subspace of the sum of two operators associated with non-fixed singletons. In other words, they characterized surjective maps φ\varphi on B(X)B(X) which satisfy

Xφ(T)+φ(S)({λ})=XT+S({λ}),(T,SB(X),λ).X_{\varphi(T)+\varphi(S)}(\{\lambda\})=X_{T+S}(\{\lambda\}),\quad(T,S\in B(X),\lambda\in\mathbb{C}).

They also gave a characterization of maps on B(X)B(X) that preserve the local spectral subspace of the difference of operators associated with non-fixed singletons. Furthermore, they investigated the product case as well as the triple product case. Namely, they described surjective maps φ\varphi on B(X)B(X) satisfying

Xφ(T)φ(S)({λ})=XTS({λ}),(T,SB(X),λ),X_{\varphi(T)\varphi(S)}(\{\lambda\})=X_{TS}(\{\lambda\}),\quad(T,S\in B(X),\lambda\in\mathbb{C}),

and also surjective maps φ\varphi on B(X)B(X) satisfying

Xφ(T)φ(S)φ(T)({λ})=XTST({λ})(T,SB(X),λ).X_{\varphi(T)\varphi(S)\varphi(T)}(\{\lambda\})=X_{TST}(\{\lambda\})\quad(T,S\in B(X),\lambda\in\mathbb{C}).

Bourhim and Lee [6] investigated the form of all maps φ1\varphi_{1} and φ2\varphi_{2} on B(X)B(X) such that, for every TT and SS in B(X)B(X), the local spectra of TSTS and φ1(T)φ2(S)\varphi_{1}(T)\varphi_{2}(S) are the same at a nonzero fixed vector x0x_{0}. In this paper, We show that if φ:B(H)B(H)\varphi:B(H)\rightarrow B(H) is an additive map such that its range contains all operators of rank at most two and satisfies

Hφ(T)φ(S)({λ})=HTS({λ}),(T,SB(H),λ),H_{\varphi(T)\varphi(S)^{\ast}}(\{\lambda\})=H_{TS^{\ast}}(\{\lambda\}),\leavevmode\nobreak\ \leavevmode\nobreak\ (T,S\in B(H),\lambda\in\mathbb{C}),

then there exist a unitary operator VV in B(H)B(H) and a nonzero scalar μ\mu such that φ(T)=μTV\varphi(T)=\mu TV^{\ast} for all TB(H)T\in B(H). We also investigate the form of all maps φ1\varphi_{1} and φ2\varphi_{2} on B(H)B(H) such that, for every TT and SS in B(H)B(H), the local spectral subspaces of TSTS^{\ast} and φ1(T)φ2(S)\varphi_{1}(T)\varphi_{2}(S)^{\ast}, associated with the singleton {λ}\{\lambda\}, coincide.

2 Preliminaries

The first lemma summarizes some known basic and properties of the local spectrum.

Lemma 2.1.

(See [2, 20].) Let TB(H)T\in B(H). For every x,yHx,y\in H and a scalar α\alpha\in\mathbb{C} the following statements hold.

  • (i)

    σT(αx)=σT(x)\sigma_{T}(\alpha x)=\sigma_{T}(x) if α0\alpha\neq 0, and σαT(x)=ασT(x)\sigma_{\alpha T}(x)=\alpha\sigma_{T}(x).

  • (ii)

    If Tx=λxTx=\lambda x for some λ\lambda\in\mathbb{C}, then σT(x){λ}\sigma_{T}(x)\subseteq\{\lambda\}. In particular, if x0x\neq 0 and TT has SVEP, then σT(x)={λ}.\sigma_{T}(x)=\{\lambda\}.

In the next theorem we collect some of the basic properties of the subspaces HT(F)H_{T}(F).

Lemma 2.2.

(See [2, 20].) Let TB(H)T\in B(H). For FF\subseteq\mathbb{C} the following statements hold.
(i) HT(F)H_{T}(F) is a T-hyperinvariant subspace of HH.
(ii) (TλI)HT(F)=HT(F)(T-\lambda I)H_{T}(F)=H_{T}(F) for every λ\F\lambda\in\mathbb{C}\backslash F.
(iii) If xHx\in H satisfies (TλI)xHT(F)(T-\lambda I)x\in H_{T}(F), then xHT(F)x\in H_{T}(F).
(v) ker(TλI)HT(F)ker(T-\lambda I)\subseteq H_{T}(F).
(iv) HαT(λ)=HT(λα)H_{\alpha T}(\lambda)=H_{T}(\frac{\lambda}{\alpha}) for every λ\lambda\in\mathbb{C} and non-zero scalar α\alpha.

For a nonzero hHh\in H and TB(H)T\in B(H), we use a useful notation defined by Bourhim and Mashreghi in [7]:

σT(h):={{0} if σT(h)={0},σT(h){0} if σT(h){0}.\sigma^{*}_{T}(h):=\left\{\begin{array}[]{ll}\{0\}&\hbox{ if \> $\sigma_{T}(h)=\{0\}$,}\\ \sigma_{T}(h)\setminus\{0\}&\hbox{ if \> $\sigma_{T}(h)\neq\{0\}$.}\end{array}\right.

For two nonzero vectors xx and yy in HH, let xyx\otimes y stands for the operator of rank at most one defined by

(xy)z=z,yx,zH.(x\otimes y)z=\left\langle z,y\right\rangle x,\qquad\forall\ z\in H.

Note that every rank one operator in B(H)B(H) can be written in this form, and that every finite rank operator TB(H)T\in B(H) can be written as a finite sum of rank one operators; i.e., T=i=1nxiyiT=\sum_{i=1}^{n}x_{i}\otimes y_{i} for some xi,yiHx_{i},y_{i}\in H and i=1,2,,ni=1,2,...,n. By F(H)F(H) and Fn(H)F_{n}(H), we mean the set of all finite rank operators in B(H)B(H). and the set of all operators of rank at most nn, nn is a positive integer, respectively.

The following lemma is an elementary observation which discribes the nonzero local spectrum of any rank one operator.

Lemma 2.3.

(See [7, Lemma 2.2].) Let h0h_{0} be a nonzero vector in HH. For every vectors x,yHx,y\in H, we have

σxy(h0):={{0} if h0,y=0,x,y if h0,y0.\sigma^{*}_{x\otimes y}(h_{0}):=\left\{\begin{array}[]{ll}\{0\}&\hbox{ if \> $\left\langle h_{0},y\right\rangle=0$,}\\ \left\langle x,y\right\rangle&\hbox{ if \> $\left\langle h_{0},y\right\rangle\neq 0$.}\end{array}\right.

The following theorem, which may be of independent interest, gives a spectral characterization of rank one operators in term of local spectrum.

Theorem 2.4.

(See [7, Theorem 4.1].) For a nonzero vector hHh\in H and a nonzero operator RB(H),R\in B(H), the following statements are equivalent.
(a) RR has rank one.
(b) σRT(h)\sigma^{*}_{RT}(h) contains at most one element for all TB(H)T\in B(H).
(c) σRT(h)\sigma^{*}_{RT}(h) contains at most one element for all TF2(H)T\in F_{2}(H).

The following Lemma is a key tool for the proofs in the sequel.

Lemma 2.5.

(See [3, Lemma 1.6].) Let hh be a nonzero vector in HH and T,SB(H)T,S\in B(H). If HT({λ})=HS({λ})H_{T}(\{\lambda\})=H_{S}(\{\lambda\}) for all λ\lambda\in\mathbb{C}. Then, σT(h)={μ}\sigma_{T}(h)=\{\mu\} if and only if σS(h)={μ}\sigma_{S}(h)=\{\mu\} for all μ\mu\in\mathbb{C}.

Moreover, this theorem will be useful in the proofs of our main result.

Theorem 2.6.

(See [3, Theorem 2.1].) Let T,SB(H)T,S\in B(H). The following statements are equivalent.
(1)(1) T=ST=S.
(2)(2) HTR({λ})=HSR({λ})H_{TR}(\{\lambda\})=H_{SR}(\{\lambda\}) for all λ\lambda\in\mathbb{C} and RF1(H)R\in F_{1}(H).

The next theorem describes additive maps on B(H)B(H) that preserve the local spectral subspace of operators associated with any singleton.

Theorem 2.7.

(See [10, Theorem 2.1].) Let φ:B(H)B(H)\varphi:B(H)\rightarrow B(H) be an additive map such that Hφ(T)({λ})=HT({λ})H_{\varphi(T)}(\{\lambda\})=H_{T}(\{\lambda\}) for all TB(H)T\in B(H) and λ\lambda\in\mathbb{C}. Then φ(T)=T\varphi(T)=T for all TB(H)T\in B(H).

The following theorem will be useful in the sequel. We recall that if h:h:\mathbb{C}\rightarrow\mathbb{C} is a ring homomorphism, then an additive map A:HHA:H\rightarrow H satisfying A(αx)=h(α)x,(xH,α)A(\alpha x)=h(\alpha)x,(x\in H,\alpha\in\mathbb{C}) is called an hh-quasilinear operator.

Theorem 2.8.

(See [21, Theorem 3.3].) Let φ:F(H)F(H)\varphi:F(H)\rightarrow F(H) be a bijective additive map preserving rank one operators in both directions. Then there exist a ring automorphism h:h:\mathbb{C}\rightarrow\mathbb{C}, and either there are hh-quasilinear bijective maps A:HHA:H\rightarrow H and B:HHB:H\rightarrow H such that

φ(xy)=AxBy,x,yH,\varphi(x\otimes y)=Ax\otimes By,\leavevmode\nobreak\ \leavevmode\nobreak\ \leavevmode\nobreak\ x,y\in H,

or there are hh-quasilinear bijective maps C:HHC:H\rightarrow H and D:HHD:H\rightarrow H such that

φ(xy)=CyDx,x,yH.\varphi(x\otimes y)=Cy\otimes Dx,\leavevmode\nobreak\ \leavevmode\nobreak\ x,y\in H.

Note that, if in Theorem 2.8 the map φ\varphi is linear, then hh is the identity map on \mathbb{C} and so the maps A,B,CA,B,C and DD are linear.

3 Main Results

The following theorem is the first main result of this paper, which characterizes those maps preserving the local spectral subspace of skew-product operators.

Theorem 3.1.

Let φ:B(H)B(H)\varphi:B(H)\rightarrow B(H) be an additive map such that its range contains F2(H)F_{2}(H). If

Hφ(T)φ(S)({λ})=HTS({λ}),(T,SB(H),λ),H_{\varphi(T)\varphi(S)^{\ast}}(\{\lambda\})=H_{TS^{\ast}}(\{\lambda\}),\leavevmode\nobreak\ \leavevmode\nobreak\ (T,S\in B(H),\lambda\in\mathbb{C}), (1)

then there exist a unitary operator VV in B(H)B(H) and a nonzero scalar μ\mu such that φ(T)=μTV\varphi(T)=\mu TV^{\ast} for all TB(H)T\in B(H).

Proof.

The proof breaks down into several claims.

Claim 1. φ\varphi is injective.

If φ(T)=φ(S)\varphi(T)=\varphi(S) for some T,SB(H)T,S\in B(H), we get that

HTR({λ})=Hφ(T)φ(R)({λ})=Hφ(S)φ(R)({λ})=HSR({λ})H_{TR^{\ast}}(\{\lambda\})=H_{\varphi(T)\varphi(R)^{\ast}}(\{\lambda\})=H_{\varphi(S)\varphi(R^{\ast})}(\{\lambda\})=H_{SR^{\ast}}(\{\lambda\})

for all RF1(X)R\in F_{1}(X) and λ\lambda\in\mathbb{C}. By Theorem 2.6, we see that T=ST=S and hence φ\varphi is injective.

Claim 2. φ\varphi preserves rank one operators in both directions.

Let R=xyR=x\otimes y be a rank one operator where x,yHx,y\in H. Note that, φ(R)0\varphi(R)\neq 0, since φ(0)=0\varphi(0)=0 and φ\varphi is injective. Let TB(H)T\in B(H) be an arbitrary operator. Since RTx=x,TyxRT^{\ast}x=\left\langle x,Ty\right\rangle x and RTRT^{\ast} has the SVEP, then σRT(x)={x,Ty}\sigma_{RT^{\ast}}(x)=\{\left\langle x,Ty\right\rangle\}. We have

xHRT({x,Ty})=Hφ(R)φ(T)({x,Ty}).x\in H_{RT^{\ast}}(\{\left\langle x,Ty\right\rangle\})=H_{\varphi(R)\varphi(T)^{\ast}}(\{\left\langle x,Ty\right\rangle\}).

As the range of φ\varphi contains F2(H)F_{2}(H), using Lemma 2.5, σφ(R)S(x)\sigma^{*}_{\varphi(R)S^{\ast}}(x) contains at most one element for all operators SF2(H)S\in F_{2}(H). By Theorem 2.4, we see that φ(R)\varphi(R) has rank one. The converse holds in a similar way and thus φ\varphi preserves the rank one operators in both directions.

Claim 3. φ\varphi is linear.

We show that φ\varphi is homogeneous. Let RR be an arbitrary rank-one operator. By the previous claim, there exists a rank one operator SS in B(H)B(H) such that φ(S)=R\varphi(S)=R. For every α,λ\alpha,\lambda\in\mathbb{C} with α0\alpha\neq 0 and TB(H)T\in B(H), we have

Hαφ(T)R({λ})\displaystyle H_{\alpha\varphi(T)R^{\ast}}(\{\lambda\}) =Hαφ(T)φ(S)({λ})=Hφ(T)φ(S)({λα})\displaystyle=H_{\alpha\varphi(T)\varphi(S)^{\ast}}(\{\lambda\})=H_{\varphi(T)\varphi(S)^{\ast}}(\{\frac{\lambda}{\alpha}\})
=HTS({λα})=H(αT)S({λ})\displaystyle=H_{TS^{\ast}}(\{\frac{\lambda}{\alpha}\})=H_{(\alpha T)S^{\ast}}(\{\lambda\})
=Hφ(αT)φ(S)({λ})=Hφ(αT)R({λ}).\displaystyle=H_{\varphi(\alpha T)\varphi(S)^{\ast}}(\{\lambda\})=H_{\varphi(\alpha T)R^{\ast}}(\{\lambda\}).

By Theorem 2.6, we see that φ(αT)=αφ(T)\varphi(\alpha T)=\alpha\varphi(T). Since φ\varphi is assumed to be additive, the map φ\varphi is, in fact, linear.

Claim 4. There are bijective linear mappings A:HHA:H\rightarrow H and B:HHB:H\rightarrow H such that φ(xy)=AxBy\varphi(x\otimes y)=Ax\otimes By for all x,yHx,y\in H.

By the previous claim φ\varphi is a bijective linear map from F(H)F(H) onto F(H)F(H) and preserves rank one operators in both directions, thus by Theorem 2.8, either there are bijective linear mappings A:HHA:H\rightarrow H and B:HHB:H\rightarrow H such that

φ(xy)=AxBy,x,yH,\varphi(x\otimes y)=Ax\otimes By,\leavevmode\nobreak\ \leavevmode\nobreak\ \leavevmode\nobreak\ x,y\in H, (2)

or there are bijective linear mappings C:HHC:H\rightarrow H and D:HHD:H\rightarrow H such that

φ(xy)=CyDx,x,yH.\varphi(x\otimes y)=Cy\otimes Dx,\leavevmode\nobreak\ \leavevmode\nobreak\ x,y\in H. (3)

Assume that φ\varphi takes the form (3)(3). Let y1y_{1} be a nonzero vector in HH. Choose a nonzero vector vv such that y1,v=0\left\langle y_{1},v\right\rangle=0. Set x=C1y1x=C^{-1}y_{1}, since xx and vv are nonzero vectors in HH, there exists a yHy\in H such that x,y=1\left\langle x,y\right\rangle=1 and v,y0\left\langle v,y\right\rangle\neq 0, since xyx\otimes y is idempotent, we have

Hxy({λ})\displaystyle H_{x\otimes y}(\{\lambda\}) =H(xy)(xy)({λ})\displaystyle=H_{(x\otimes y)(x\otimes y)}(\{\lambda\})
=H(xy)(yx)({λ})\displaystyle=H_{(x\otimes y)(y\otimes x)^{\ast}}(\{\lambda\})
=H(CyDx)(CxDy)({λ})\displaystyle=H_{(Cy\otimes Dx)(Cx\otimes Dy)^{\ast}}(\{\lambda\})
=H(CyDx)(DyCx)({λ})\displaystyle=H_{(Cy\otimes Dx)(Dy\otimes Cx)}(\{\lambda\})
=HDy,Dx(CyCx)({λ})\displaystyle=H_{\left\langle Dy,Dx\right\rangle(Cy\otimes Cx)}(\{\lambda\})
=HDy,Dx(Cyy1)({λ}).\displaystyle=H_{\left\langle Dy,Dx\right\rangle(Cy\otimes y_{1})}(\{\lambda\}).

On the other hand, since y1,v=0\left\langle y_{1},v\right\rangle=0, we have σCyy1(v)={0}\sigma^{*}_{Cy\otimes y_{1}}(v)=\{0\} and consequently Dy,DxσCyy1(v)={0}\left\langle Dy,Dx\right\rangle\sigma_{Cy\otimes y_{1}}(v)=\{0\}. This implies that

vHDy,DxCyy1({0})=Hxy({0}).v\in H_{\left\langle Dy,Dx\right\rangle Cy\otimes y_{1}}(\{0\})=H_{x\otimes y}(\{0\}).

Using Lemma 2.5, σxy(v)={0}\sigma_{x\otimes y}(v)=\{0\}. But lemma 2.3 implies that

σxy(v){0}.\sigma^{*}_{x\otimes y}(v)\neq\{0\}.

This contradiction shows that φ\varphi only takes the form (2)(2).

Claim 5. AA and BB are bounded unitary operators multiplied by positive scalars α\alpha and β\beta such that αβ=1\alpha\beta=1.

Let x,yx,y be nonzero vectors in HH, since σ(xy)(yx)(x)={y,yx,x}\sigma_{(x\otimes y)(y\otimes x)}(x)=\{\left\langle y,y\right\rangle\left\langle x,x\right\rangle\}, by the previous claim, we have

H(xy)(yx)({y2x2})\displaystyle H_{(x\otimes y)(y\otimes x)}(\{\|y\|^{2}\|x\|^{2}\}) =H(xy)(xy)({y2x2})\displaystyle=H_{(x\otimes y)(x\otimes y)^{\ast}}(\{\|y\|^{2}\|x\|^{2}\})
=H(AxBy)(AxBy)({y2x2})\displaystyle=H_{(Ax\otimes By)(Ax\otimes By)^{\ast}}(\{\|y\|^{2}\|x\|^{2}\})
=H(AxBy)(ByAx)({y2x2})\displaystyle=H_{(Ax\otimes By)(By\otimes Ax)}(\{\|y\|^{2}\|x\|^{2}\})
=HBy,By(AxAx)({y2x2}).\displaystyle=H_{\left\langle By,By\right\rangle(Ax\otimes Ax)}(\{\|y\|^{2}\|x\|^{2}\}).

By the Lemma 2.5, we see that

{y2x2}\displaystyle\{\|y\|^{2}\|x\|^{2}\} =σ(xy)(yx)(x)\displaystyle=\sigma_{(x\otimes y)(y\otimes x)}(x)
=σ(By,By(AxAx))(x)={By2Ax2}.\displaystyle=\sigma_{(\left\langle By,By\right\rangle(Ax\otimes Ax))}(x)=\{\|By\|^{2}\|Ax\|^{2}\}. (4)

Now, let y0y_{0} be a fixed unit vector in HH and let α=1By0\alpha=\frac{1}{\|By_{0}\|}. By (4)(4), we have

Ax2=α2x2\|Ax\|^{2}=\alpha^{2}\|x\|^{2}

for all xHx\in H. Hence, U=1αAU=\frac{1}{\alpha}A is an isometry and thus it is a unitary operator in B(H)B(H), because AA is bijective. Similarly, fix a unit vector x0Hx_{0}\in H and take β=1Ax0\beta=\frac{1}{\|Ax_{0}\|}, and note that V=1βBV=\frac{1}{\beta}B is a unitary operator in B(H)B(H). Finally, by (4)(4), we see that αβ=1\alpha\beta=1.

Claim 6. AA^{\ast} and II are linearly dependent.

Assume, by the way of contradiction, that there exists a nonzero vector xHx\in H such that AxAx and xx are linearly independent. Let uHu\in H be a vector such that x,u=1\left\langle x,u\right\rangle=1 and Ax,u=0\left\langle A^{\ast}x,u\right\rangle=0. Since σxu(x)={1}\sigma_{x\otimes u}(x)=\{1\}, then

xHxu({1})\displaystyle x\in H_{x\otimes u}(\{1\}) =X(xu)(xu)({1})\displaystyle=X_{(x\otimes u)(x\otimes u)}(\{1\})
=H(xu)(ux)({1})\displaystyle=H_{(x\otimes u)(u\otimes x)^{\ast}}(\{1\})
=Hφ(xu)φ(ux)({1})\displaystyle=H_{\varphi(x\otimes u)\varphi(u\otimes x)^{\ast}}(\{1\})
=H(AxBu)(AuBx)({1})\displaystyle=H_{(Ax\otimes Bu)(Au\otimes Bx)^{\ast}}(\{1\})
=H(AxBu)(BxAu)({1})\displaystyle=H_{(Ax\otimes Bu)(Bx\otimes Au)}(\{1\})
=HBx,Bu(AxAu)({1}),\displaystyle=H_{\left\langle Bx,Bu\right\rangle(Ax\otimes Au)}(\{1\}),

using Lemma 2.5, we have

{1}=σxu(x)=σBx,Bu(AxAu)(x)={0}.\displaystyle\{1\}=\sigma_{x\otimes u}(x)=\sigma_{\left\langle Bx,Bu\right\rangle(Ax\otimes Au)}(x)=\{0\}.

This contradiction shows that there is a nonzero scalar γ\gamma\in\mathbb{C} such that A=γIA^{\ast}=\gamma I.

Claim 7. φ(T)=μTV\varphi(T)=\mu TV^{\ast} for all TB(H)T\in B(H), where VV is unitary operators and μ\mu is a nonzero scalar.

By claim 5 we shall assume that A=UA=U and B=VB=V for some unitary operators U,VB(H)U,V\in B(H). Using the previous claim and (1)(1), for every rank one operator RB(H)R\in B(H) and every operator TB(H)T\in B(H) we have

Hφ(T)φ(R)({λ})\displaystyle H_{\varphi(T)\varphi(R)^{\ast}}(\{\lambda\}) =HTR({λ})\displaystyle=H_{TR^{\ast}}(\{\lambda\})
=HUTRU({λ})\displaystyle=H_{UTR^{\ast}U^{\ast}}(\{\lambda\})
=HUTVVRU({λ})\displaystyle=H_{UTV^{\ast}VR^{\ast}U^{\ast}}(\{\lambda\})
=HUTV(URV)({λ})\displaystyle=H_{UTV^{\ast}(URV^{\ast})^{\ast}}(\{\lambda\})
=HUTVφ(R)({λ}).\displaystyle=H_{UTV^{\ast}\varphi(R)^{\ast}}(\{\lambda\}).

Since φ\varphi preserves rank one operators in both directions, Theorem 2.6 shows that φ(T)=UTV\varphi(T)=UTV^{\ast} for all TB(H)T\in B(H). Claim 6 tells us that for ever TB(H)T\in B(H) we have φ(T)=μTV\varphi(T)=\mu TV^{\ast} for some μ\mu\in\mathbb{C}. ∎

From this result, it is easy to deduce a generalization for the case of two different Hilbert spaces H,KH,K.

Corollary 3.2.

Suppose UB(H,K)U\in B(H,K) be a unitary operator. Let φ\varphi be an additive map from B(H)B(H) into B(K)B(K) such that its range contains F2(K)F_{2}(K). If

Kφ(T)φ(S)({λ})=UHTS({λ}),(T,SB(H),λ).K_{\varphi(T)\varphi(S)^{\ast}}(\{\lambda\})=UH_{TS^{\ast}}(\{\lambda\}),\leavevmode\nobreak\ \leavevmode\nobreak\ (T,S\in B(H),\lambda\in\mathbb{C}).

Then there exist a unitary operator V:HKV:H\rightarrow K and a nonzero scalar μ\mu such that φ(T)=μUTV\varphi(T)=\mu UTV^{\ast} for all TB(H)T\in B(H).

Proof.

We consider the map ψ:B(H)B(H)\psi:B(H)\rightarrow B(H) defined by ψ(T)=Uφ(T)U\psi(T)=U^{\ast}\varphi(T)U for all TB(H)T\in B(H). We have,

Hψ(T)ψ(S)({λ})\displaystyle H_{\psi(T)\psi(S)^{\ast}}(\{\lambda\}) =HUφ(T)UUφ(S)U({λ})\displaystyle=H_{U^{\ast}\varphi(T)UU^{\ast}\varphi(S)^{\ast}U}(\{\lambda\})
=HUφ(T)φ(S)U({λ})\displaystyle=H_{U^{\ast}\varphi(T)\varphi(S)^{\ast}U}(\{\lambda\})
=UKφ(T)φ(S)({λ})\displaystyle=U^{\ast}K_{\varphi(T)\varphi^{\ast}(S)}(\{\lambda\})
=HTS({λ})\displaystyle=H_{TS^{\ast}}(\{\lambda\})

for every T,SB(H)T,S\in B(H) and λ\lambda\in\mathbb{C}. So by Theorem 3.1, there exist a unitary operator P:HHP:H\rightarrow H and μ\mu\in\mathbb{C} such that ψ(T)=μTP\psi(T)=\mu TP^{\ast} for all TB(H)T\in B(H). Therefore φ(T)=μUTV\varphi(T)=\mu UTV^{\ast} for all TB(H)T\in B(H), where V=UPV=UP. ∎

In the next theorem, we investigate the form of all maps φ1\varphi_{1} and φ2\varphi_{2} on B(H)B(H) such that, for every TT and SS in B(H)B(H), the local spectral subspaces of TSTS^{\ast} and φ1(T)φ2(S)\varphi_{1}(T)\varphi_{2}(S)^{\ast}, associated with the singleton {λ}\{\lambda\}, coincide.

Theorem 3.3.

Let φ1\varphi_{1} and φ2\varphi_{2} be additive maps from B(H)B(H) into B(H)B(H) which satisfy

Hφ1(T)φ2(S)({λ})=HTS({λ}),(T,SB(H),λ).H_{\varphi_{1}(T)\varphi_{2}(S)^{\ast}}(\{\lambda\})=H_{TS^{\ast}}(\{\lambda\}),\leavevmode\nobreak\ \leavevmode\nobreak\ (T,S\in B(H),\lambda\in\mathbb{C}). (5)

If the range of φ1\varphi_{1} and φ2\varphi_{2} contain F2(H)F_{2}(H), then φ2(I)\varphi_{2}(I)^{\ast} is invertible, and φ1(T)=T(φ2(I))1\varphi_{1}(T)=T(\varphi_{2}(I)^{\ast})^{-1} and φ2(T)=φ2(I)T\varphi_{2}(T)=\varphi_{2}(I)^{\ast}T for all TB(H)T\in B(H).

Proof.

The proof is rather long and we break it into several claims.

Claim 1. φ1\varphi_{1} is a one to one linear map preserving rank one operators in both directions.

Similar to the proof of Theorem 3.1, we can shows that φ1\varphi_{1} is a one to one linear map preserving rank one operators in both directions.

Claim 2. There are bijective linear mappings A:HHA:H\rightarrow H and B:HHB:H\rightarrow H such that φ1(xy)=AxBy\varphi_{1}(x\otimes y)=Ax\otimes By for all x,yHx,y\in H.

By the claim 1 φ1:F(H)F(H)\varphi_{1}:F(H)\rightarrow F(H) is a bijective linear map which preserves rank one operators in both directions. Thus by Theorem 2.8, φ1\varphi_{1} has one of the following forms.
(1)(1) There exist bijective linear maps A:HHA:H\rightarrow H and B:HHB:H\rightarrow H such that

φ1(xy)=AxBy,x,yH.\varphi_{1}(x\otimes y)=Ax\otimes By,\leavevmode\nobreak\ \leavevmode\nobreak\ \leavevmode\nobreak\ x,y\in H. (6)

(2)(2) There exist bijective linear maps C:HHC:H\rightarrow H and D:HHD:H\rightarrow H such that

φ1(xy)=CyDx,x,yH.\varphi_{1}(x\otimes y)=Cy\otimes Dx,\leavevmode\nobreak\ \leavevmode\nobreak\ x,y\in H. (7)

Assume that φ1\varphi_{1} takes the form (7)(7). Let yy be a nonzero vector in HH, choose a nonzero vector vHv\in H such that φ2(I)y,v=0\left\langle\varphi_{2}(I)^{\ast}y,v\right\rangle=0. Set x=D1vx=D^{-1}v, since xx and yy are nonzero vectors in HH, there exists a vector uHu\in H such that y,u0\left\langle y,u\right\rangle\neq 0 and x,u0\left\langle x,u\right\rangle\neq 0. Since σ(Cuv)φ2(I)(y)={0}\sigma^{*}_{(Cu\otimes v)\varphi_{2}(I)^{\ast}}(y)=\{0\}, we have

yH(Cuv)φ2(I)({0})=H(CuDx)φ2(I)({0})=Hxu({0}).y\in H_{(Cu\otimes v)\varphi_{2}(I)^{\ast}}(\{0\})=H_{(Cu\otimes Dx)\varphi_{2}(I)^{\ast}}(\{0\})=H_{x\otimes u}(\{0\}).

Using Lemma 2.5, σxu(y)={0}\sigma_{x\otimes u}(y)=\{0\}. But lemma 2.3 implies that

σxu(y)={x,u}{0}.\sigma^{*}_{x\otimes u}(y)=\{\left\langle x,u\right\rangle\}\neq\{0\}.

This contradiction shows that φ1\varphi_{1} only takes the form (6)(6).

Claim 3. For every x,yHx,y\in H, x,y=Ax,φ2(I)(By)\left\langle x,y\right\rangle=\left\langle Ax,\varphi_{2}(I)(By)\right\rangle.

Assume that xx and yy are arbitrary vectors in HH. We have, σxy(x)={x,y}\sigma_{x\otimes y}(x)=\{\left\langle x,y\right\rangle\}, so the previous claim and (5)(5) imply that

xHxy({x,y})=Hφ1(xy)φ2(I)({x,y})=H(AxBy)φ2(I)({x,y}).x\in H_{x\otimes y}(\{\left\langle x,y\right\rangle\})=H_{\varphi_{1}(x\otimes y)\varphi_{2}(I)^{\ast}}(\{\left\langle x,y\right\rangle\})=H_{(Ax\otimes By)\varphi_{2}(I)^{\ast}}(\{\left\langle x,y\right\rangle\}).

Assume first that x,y0\left\langle x,y\right\rangle\neq 0, using lemma 2.5,

{0}{x,y}=σxy(x)=σφ1(xy)φ2(I)(x)=σ(AxBy)φ2(I)(x),\{0\}\neq\{\left\langle x,y\right\rangle\}=\sigma_{x\otimes y}(x)=\sigma_{\varphi_{1}(x\otimes y)\varphi_{2}(I)^{\ast}}(x)=\sigma_{(Ax\otimes By)\varphi_{2}(I)^{\ast}}(x),

which means that x,φ2(I)(By)0\left\langle x,\varphi_{2}(I)(By)\right\rangle\neq 0. Then Lemma 2.3 implies that

{x,y}=σxy(x)=σ(AxBy)φ2(I)(x)={Ax,φ2(I)(By)}.\{\left\langle x,y\right\rangle\}=\sigma_{x\otimes y}(x)=\sigma_{(Ax\otimes By)\varphi_{2}(I)^{\ast}}(x)=\{\left\langle Ax,\varphi_{2}(I)(By)\right\rangle\}.

Now, if x,y=0\left\langle x,y\right\rangle=0, we choose a vector uHu\in H such that x,u0\left\langle x,u\right\rangle\neq 0. By application of what has been shown previously to both uu and x+ux+u, we have x,u=Ax,φ2(I)(By)\left\langle x,u\right\rangle=\left\langle Ax,\varphi_{2}(I)(By)\right\rangle and x,y+u=Ax,φ2(I)(B(u+y))\left\langle x,y+u\right\rangle=\left\langle Ax,\varphi_{2}(I)(B(u+y))\right\rangle. So

x,y+x,u\displaystyle\left\langle x,y\right\rangle+\left\langle x,u\right\rangle =x,y+u\displaystyle=\left\langle x,y+u\right\rangle
=Ax,φ2(I)(B(y+u)\displaystyle=\left\langle Ax,\varphi_{2}(I)(B(y+u)\right\rangle
=Ax,φ2(I)(By)+Ax,φ2(I)(Bu)\displaystyle=\left\langle Ax,\varphi_{2}(I)(By)\right\rangle+\left\langle Ax,\varphi_{2}(I)(Bu)\right\rangle
=Ax,φ2(I)(By)+x,u.\displaystyle=\left\langle Ax,\varphi_{2}(I)(By)\right\rangle+\left\langle x,u\right\rangle.

This shows that x,y=Ax,φ2(I)(By)\left\langle x,y\right\rangle=\left\langle Ax,\varphi_{2}(I)(By)\right\rangle in this case too.

Claim 4. φ2(I)\varphi_{2}(I)^{\ast} is invertible.

It is clear that φ2(I)\varphi_{2}(I)^{\ast} is injective, if not, there is a nonzero vector yHy\in H such that φ2(I)y=0\varphi_{2}(I)^{\ast}y=0. Take x=A1yx=A^{-1}y, and let uHu\in H be a vector such that x,u=1\left\langle x,u\right\rangle=1. By the previous claim, we have 1=x,u=Ax,φ2(I)(Bu)=y,φ2(I)(Bu)=φ2(I)y,Bu=01=\left\langle x,u\right\rangle=\left\langle Ax,\varphi_{2}(I)(Bu)\right\rangle=\left\langle y,\varphi_{2}(I)(Bu)\right\rangle=\left\langle\varphi_{2}(I)^{\ast}y,Bu\right\rangle=0. This contradiction tells us that φ2(I)\varphi_{2}(I)^{\ast} is injective. Now, we show that AA is continuous and (φ2(I))B=(A)1(\varphi_{2}(I))B=(A^{\ast})^{-1}. Assume that (xn)n(x_{n})_{n} is a sequence in HH such that limnxn=xH\lim_{n\longrightarrow\infty}x_{n}=x\in H and limnAxn=yH\lim_{n\longrightarrow\infty}Ax_{n}=y\in H. Then, for every uHu\in H, we have

y,φ2(I)(Bu)\displaystyle\left\langle y,\varphi_{2}(I)(Bu)\right\rangle =limnAxn,φ2(I)(Bu)\displaystyle=\lim_{n\longrightarrow\infty}\left\langle Ax_{n},\varphi_{2}(I)(Bu)\right\rangle
=limnxn,u)=x,u=Ax,φ2(I)(By).\displaystyle=\lim_{n\longrightarrow\infty}\left\langle x_{n},u)\right\rangle=\left\langle x,u\right\rangle=\left\langle Ax,\varphi_{2}(I)(By)\right\rangle.

Since BB is bijective and uHu\in H is an arbitrary vector, the closed graph theorem shows that AA is continuous. Moreover, we have x,y=Ax,φ2(I)(By)=x,Aφ2(I)(By)\left\langle x,y\right\rangle=\left\langle Ax,\varphi_{2}(I)(By)\right\rangle=\left\langle x,A^{\ast}\varphi_{2}(I)(By)\right\rangle for all x,yHx,y\in H, and thus I=Aφ2(I)BI=A^{\ast}\varphi_{2}(I)B. It follows that φ2(I)\varphi_{2}(I)^{\ast} is invertible.

Claim 5. (A)1(A^{\ast})^{-1} and II are linearly dependent.

Assume, by the way of contradiction, that there exists a nonzero vector xHx\in H such that AxAx and xx are linearly independent. Let yy be a vector in HH such that x,y=1\left\langle x,y\right\rangle=1 and (A)1x,y=0\left\langle(A^{\ast})^{-1}x,y\right\rangle=0. We have (xy)x=x(x\otimes y)x=x and A(xy)(A)1x=0A(x\otimes y)(A^{\ast})^{-1}x=0, then σxy(x)={1}\sigma_{x\otimes y}(x)=\{1\} and σA(xy)(A)1(x)={0}\sigma_{A(x\otimes y)(A^{\ast})^{-1}}(x)=\{0\}. Since Hxy({λ})=Hφ1(xy)φ2(I)({λ})H_{x\otimes y}(\{\lambda\})=H_{\varphi_{1}(x\otimes y)\varphi_{2}(I)^{\ast}}(\{\lambda\}), using Lemma 2.5 and claim 2, we have

{1}\displaystyle\{1\} =σxy(x)\displaystyle=\sigma_{x\otimes y}(x)
=σφ1(xy)φ2(I)(x)\displaystyle=\sigma_{\varphi_{1}(x\otimes y)\varphi_{2}(I)^{\ast}}(x)
=σ(AxBy)φ2(I)(x)\displaystyle=\sigma_{(Ax\otimes By)\varphi_{2}(I)^{\ast}}(x)
=σ(Axφ2(I)By)(x)\displaystyle=\sigma_{(Ax\otimes\varphi_{2}(I)By)}(x)
=σA(xy)(A)1(x)={0}.\displaystyle=\sigma_{A(x\otimes y)(A^{\ast})^{-1}}(x)=\{0\}.

This contradiction shows that there is a nonzero scalar α\alpha\in\mathbb{C} such that (A)1=αI(A^{\ast})^{-1}=\alpha I.

Claim 6. φ1\varphi_{1} and φ1\varphi_{1} have the desired forms.

We define the map ψ1:B(H)B(H)\psi_{1}:B(H)\rightarrow B(H) by ψ1(T)=φ1(T)φ2(I)\psi_{1}(T)=\varphi_{1}(T)\varphi_{2}(I)^{\ast} for all TB(H)T\in B(H). We have,

Hψ1(T)({λ})=Hφ1(T)φ2(I)({λ})=HT({λ})H_{\psi_{1}(T)}(\{\lambda\})=H_{\varphi_{1}(T)\varphi_{2}(I)^{\ast}}(\{\lambda\})=H_{T}(\{\lambda\})

for all TB(H)T\in B(H) and λ\lambda\in\mathbb{C}. So by Theorem 2.7, ψ1(T)=T\psi_{1}(T)=T for all TB(H)T\in B(H), therefore φ1(T)=T(φ2(I))1\varphi_{1}(T)=T(\varphi_{2}(I)^{\ast})^{-1} for all TB(H)T\in B(H). Once again, we consider the map ψ2:B(H)B(H)\psi_{2}:B(H)\rightarrow B(H) defined by ψ2(T)=φ1(I)φ2(T)\psi_{2}(T)=\varphi_{1}(I)\varphi_{2}(T^{\ast})^{\ast} for all TB(H)T\in B(H). We see that for all TB(H)T\in B(H) and λ\lambda\in\mathbb{C},

Hψ2(T)({λ})=HT({λ}),H_{\psi_{2}(T)}(\{\lambda\})=H_{T}(\{\lambda\}),

by Theorem 2.7, ψ2(T)=T\psi_{2}(T)=T for all TB(H)T\in B(H). Hence φ2(T)=(φ1(I))1T\varphi_{2}(T)=(\varphi_{1}(I))^{-1}T for all TB(H)T\in B(H). ∎

Theorem 3.3 leads directly to the following corollary.

Corollary 3.4.

Suppose UB(H,K)U\in B(H,K) be a unitary operator. Let φ1\varphi_{1} and φ2\varphi_{2} be two additive map from B(H)B(H) into B(K)B(K) which satisfy

Kφ1(T)φ2(S)({λ})=UHTS({λ}),(T,SB(H),λ).K_{\varphi_{1}(T)\varphi_{2}(S)^{\ast}}(\{\lambda\})=UH_{TS^{\ast}}(\{\lambda\}),\leavevmode\nobreak\ \leavevmode\nobreak\ (T,S\in B(H),\lambda\in\mathbb{C}).

If the range of φ1\varphi_{1} and φ2\varphi_{2} contain F2(K)F_{2}(K), then there exists a bijective linear map V:KHV:K\rightarrow H such that φ1(T)=UTV\varphi_{1}(T)=UTV and φ2(T)=V1TU\varphi_{2}(T)=V^{-1}TU^{\ast} for all TB(H)T\in B(H).

Proof.

We consider the maps ψ1:B(H)B(H)\psi_{1}:B(H)\rightarrow B(H) defined by ψ1(T)=Uφ1(T)U\psi_{1}(T)=U^{\ast}\varphi_{1}(T)U and ψ2:B(H)B(H)\psi_{2}:B(H)\rightarrow B(H) defined by ψ2(T)=Uφ2(T)U\psi_{2}(T)=U^{\ast}\varphi_{2}(T)U for all TB(H)T\in B(H). We have,

Hψ1(T)ψ2(S)({λ})\displaystyle H_{\psi_{1}(T)\psi_{2}(S)^{\ast}}(\{\lambda\}) =HUφ1(T)UUφ2(S)U({λ})\displaystyle=H_{U^{\ast}\varphi_{1}(T)UU^{\ast}\varphi_{2}(S)^{\ast}U}(\{\lambda\})
=HUφ1(T)φ2(S)U({λ})\displaystyle=H_{U^{\ast}\varphi_{1}(T)\varphi_{2}(S)^{\ast}U}(\{\lambda\})
=U1Kφ1(T)φ2(S)({λ})=HTS({λ})\displaystyle=U^{-1}K_{\varphi_{1}(T)\varphi_{2}(S)^{\ast}}(\{\lambda\})=H_{TS^{\ast}}(\{\lambda\})

for every T,SB(H)T,S\in B(H) and λ\lambda\in\mathbb{C}. So by Theorem 3.3, ψ1(T)=TP1\psi_{1}(T)=TP^{-1} and ψ2(T)=PT\psi_{2}(T)=PT for all TB(H)T\in B(H), where P=ψ2(I)P=\psi_{2}(I)^{\ast}. Therefore φ1(T)=UTV\varphi_{1}(T)=UTV and φ2(T)=V1TU1\varphi_{2}(T)=V^{-1}TU^{-1} for all TB(H)T\in B(H), where V=(UP)1V=(UP)^{-1}. ∎

References

  • [1] B. Aupetit, Spectrum-preserving linear mappings between Banach algebras or Jordan–Banach algebras, J. London Math. Soc. 62 (2000) 917–924.
  • [2] P. Aiena, Fredholm and local spectral theory, with applications to multipliers, Kluwer, Dordrecht, 2004.
  • [3] H. Benbouziane, M.E. Kettani, I. Herrou, Local spectral subspace preservers. Rend. Circ. Mat. Palermo, II. Ser. 68 (2019) 293-303.
  • [4] H. Benbouziane, M.E. Kettani, I. Herrou, Nonlinear maps preserving the local spectral subspace, Linear Multilinear Algebra. 7 (2019) 29-38.
  • [5] A. Bourhim, J. Mashreghi, A survey on preservers of spectra and local spectra, Contemp Math. 45 (2015) 45-98.
  • [6] A. Bourhim, J.E. Lee, Multiplicatively local spectrum-preserving maps, Linear Algebra Appl. 549 (2018) 291–308.
  • [7] A. Bourhim, J. Mashreghi, Maps preserving the local spectrum of product of operators, Glasgow Math. J. 57 (2015) 709-718.
  • [8] Bourhim A, Ransford T. Additive maps preserving local spectrum. Integral Equations Operator Theory. 5 (2006) 377-385.
  • [9] G. Dolinar, S.P. Du, J.C. Hou, P. Legisa, General preservers of invariant subspace lattices, Linear Algebra Appl. 429 (2008) 100-109.
  • [10] M. Elhodaibi, A. Jaatit, On additive maps preserving the local spectral subspace, Int. J. Math. Anal. 6 (21) (2012) 1045-1051.
  • [11] G. Frobenius, Ueber die Darstellung der endlichen Gruppen durch lineare Substitutionen, Berl Ber. Appl. 203 (1897) 994-1015.
  • [12] A.A. Jafarian, A survey of invertibility and Spectrum-preserving linear maps, Bulletin of the Iranian Mathematical Society. 35 (2) (2009) 1-10.
  • [13] A.A. Jafarian, A.R. Sourour, Spectrum-preserving linear maps, J. Funct. Anal. 66 (1986) 255-261.
  • [14] A.A. Jafarian, A.R. Sourour, Linear maps that preserve the commutant, double commutant or the lattice of invariant subspaces, Linear and Multilinear Algebra. 38 (1994) 117-129.
  • [15] A.M. Gleason, A characterization of maximal ideals. J. Analyse Math. 19 (1967) 171-172.
  • [16] J.P. Kahane, W. Zelazko, A characterization of maximal ideals in commutative Banach algebras. Studia Math. 29 (1968) 339-343.
  • [17] I. Kaplansky, Algebraic and analytic aspects of operator algebras, Conference Board of the Mathematical Sciences Regional Conference Series in Mathematics, No. 1. Providence (RI): American Mathematical Society; 1970.
  • [18] M. Marcus, B.N. Moyls, Linear transformations on algebras of matrices, Canad. J. Math. 11 (1959) 61-66.
  • [19] L. Molnar, Some characterizations of the automorphisms of B(H)B(H) and C(X)C(X). Proc Amer Math Soc. 130 (1) (2002) 111–120.
  • [20] K.B. Laursen, M.M. Neumann, An introduction to local spectral theory, London mathematical society monographs, new series. vol. 20. New York (NY): The Clarendon Press, Oxford University Press; 2000
  • [21] M. Omladic, P. Semrl, Additive mappings preserving operators of rank one, Linear Algebra Appl. 182 (1993) 239-256.
  • [22] A.R. Sourour, Invertibility preserving linear maps on L(X), Trans. Amer. Math. Soc. 348 (1996) 13-30.
  • [23] W. Zelazko, A characterization of multiplicative linear functionals in complex Banach algebras. Studia Math. 130 (1968) 83-85.