Operators

and
Matrices
Volume 10, Number 2 (2016), 379–387 doi:10.7153/oam-10-20

ADDITIVE MAPS PRESERVING m–NORMAL EIGENVALUES ON B(H )

W EIJUAN S HI AND G UOXING J I

(Communicated by N.-C. Wong)

Abstract. Let H be an infinite-dimensional complex Hilbert space and B(H ) the algebra of
all bounded linear operators on H . For an operator T ∈ B(H ) and a fixed non-negative integer
m , an m -normal eigenvalue λ of T is the normal eigenvalue satisfying dim N(T − λ I) > m .
In this paper, we prove that, if an additive surjective map ϕ on B(H ) preserves m as well as
m + 1 -normal eigenvalues, then there is an invertible operator A ∈ B(H ) such that ϕ (T ) =
ATA−1 for all T ∈ B(H ) or ϕ (T ) = AT tr A−1 for all T ∈ B(H ) , where T tr denotes the
transpose of T with respect to an arbitrary but fixed orthonormal basis of H .

1. Introduction

Linear or additive preserver problems are to characterize those linear or additive

maps on operator algebras preserving certain properties, subsets or relations. Most
important of all, we need to find certain properties which are isomorphism or anti-
isomorphism invariants. The study of the problem has attracted the attention of many
authors in the last decades [2, 3, 5, 6, 7, 11]. As we know, spectrum is a very fun-
damental and key concept in operator theory. Hence many authors have studied lin-
ear or additive maps preserving the spectrum as well as certain parts of the spectrum
[1, 4, 9, 10]. For example, the author showed that additive maps on standard operator
algebras preserving parts of the spectrum is either an isomorphism or anti-isomorphism
in [4]. It is remarkable that various parts of the spectrum may be regarded as invari-
ants of an automorphism or an anti-automorphism on the algebra of all bounded linear
operators on a Banach (or Hilbert) space. It is known that certain parts of spectrum of
operators are introduced to analyze the structure of operators. For example, the set of
normal eigenvalues of an operator is given (cf. [8]). Note that the set of normal eigen-
values is at most countable and is a very “small” subset of spectrum in general. Thus
how may the normal eigenvalues influence the structure of automorphisms on the alge-
bra of all bounded linear operators on a Banach (or Hilbert) space? In this paper, we
consider parts of the set of the normal eigenvalues as an invariant of an automorphism
or an anti-automorphism on the algebra of all bounded linear operators on a complex
infinite-dimensional Hilbert space.
Let H be a complex infinite-dimensional Hilbert space and B(H ) the algebra
of all bounded linear operators on H . For x, y ∈ H , we denote by x, y and x ⊗ y the
Mathematics subject classification (2010): 47B48, 47A10.
Keywords and phrases: Normal eigenvalues, additive map, additive preserver.

c  , Zagreb 379
Paper OaM-10-20
380 W. S HI AND G. J I

inner product of x and y and the rank one operator defined by (x ⊗ y)z = z, yx , ∀z ∈
H , respectively. The operator x ⊗ y is an idempotent if and only if x, y = 1 . Let T ∈
B(H ), we denote by N(T ) and R(T ) the null space and range of T respectively. For
a subset M of H , {M} denotes the closed subspace spanned by M . Let dim M (resp.
codimM ) is the dimension of M ( resp. M ⊥ , the orthogonal complement of M ) if M
is a closed subspace. Recall that an operator T is called Fredholm if it has closed
range such that dim N(T ) < ∞ and codimR(T ) < ∞. The index of a Fredholm operator
T ∈ B(H ) is given by ind(T ) = dim N(T ) − codimR(T ). The ascent asc(T ) of T is
the least non-negative integer n such that N(T n ) = N(T n+1 ) and the descent des(T )
is the least non-negative integer n such that R(T n ) = R(T n+1 ). An operator T is said
to be Browder if it is Fredholm with finite ascent and descent. It is known that T is a
Browder operator if and only if T is a Fredholm operator of index zero and asc(T ) < ∞.
Let T ∈ B(H ). If σ is a clopen subset of the spectrum σ (T ), then there exists
an analytic Cauchy domain Ω such that σ ⊆ Ω and [σ (T )\ σ ]∩Ω = 0/ . We let E(σ ; T )
denote the Riesz idempotent of T corresponding to σ , that is,

1
E(σ ; T ) = (λ − T )−1 d λ ,
2π i Γ

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 ) = R(E(σ ; T )). If λ ∈ isoσ (T ), the isolate
points of σ (T ), then {λ } is a clopen subset of σ (T ) and we simply write H(λ ; T )
instead of H({λ }; T ). If, in addition, dimH(λ ; T ) < ∞, then λ is called a normal
eigenvalue of T . The set of all normal eigenvalues of T will be denoted by σ0 (T ) (cf.
[8]). Clearly, σ0 (T ) is contained in the point spectrum σ p (T ). From Corollary 1.14 in
[8], we can get

σ0 (T ) = {λ ∈ σ (T ) : T − λ I is Browder}
= {λ ∈ isoσ (T ) : T − λ I is Fredholm}.

Given a non-negative integer m, we call an m-normal eigenvalue λ of T is the normal

eigenvalue satisfying dimN(T − λ I) > m. The set of all m-normal eigenvalues of T
will be denoted by σm (T ).
That is
σm (T ) = {λ ∈ σ0 (T ) : dim N(T − λ I) > m}.

Then we obtain that

... ⊆ σm (T )... ⊆ σ2 (T ) ⊆ σ1 (T ) ⊆ σ0 (T ).

In this paper, we characterize an additive surjective map ϕ on B(H ) preserves m

as well as m + 1 -normal eigenvalues for some fixed non-negative integer m, that is
σm (ϕ (T )) = σm (T ) and σm+1 (ϕ (T )) = σm+1 (T ) for all T ∈ B(H ). And we show
that such a surjective map is an automorphism or an anti-automorphism.
 A DDITIVE MAPS PRESERVING m - NORMAL EIGENVALUES ON B(H ) 381

2. Main results

We firstly need some auxiliary results.

L EMMA 1. Let m  0 and T ∈ B(H ). If 0 ∈ σm+1 (T ), then for every rank one
operator F ∈ B(H ), either 0 ∈ σm (T + F) or 0 ∈ σm (T − F).

Proof. Suppose that 0 ∈ σm+1 (T ). Then both T + F and T − F are Fredholm of

index zero for every rank one operator F . Thus dim N(T + F) = codimR(T + F) and
dim N(T − F) = codimR(T − F). Note that R(T + F) ⊆ R(T ) + R(F) and dim N(T ) =
codimR(T ) > m + 1 . Then codimR(T + F) > m, and dim N(T + F) > m. Similarly,
we have dim N(T − F) > m. Since asc(T ) < ∞, we have for every rank one operator
F , either asc(T + F) < ∞ or asc(T − F) < ∞ by Proposition 2.7 in [12]. This implies
that either T + F or T − F is Browder. Hence we get either 0 ∈ σm (T + F) or 0 ∈
σm (T − F). 

P ROPOSITION 1. Let k, m  0 and T ∈ B(H ). If dim R(T )  2 , then there

exists an operator S satisfying 0 ∈ σk (S) such that 0 ∈
/ σm (S + T ) and 0 ∈
/ σm (S − T ).

Proof. We will complete this proof by three cases:

Case (i) dim N(T ) = ∞.
Assume that dim R(T )  2 . Then there exist two vectors x0 , y0 such that T x0 , Ty0
are linearly independent. We can choose suitable vectors of N(T ) to perturb x0 , y0 , then
there exist two vectors u0 , v0 such that the vectors u0 , v0 , Tu0 , T v0 are linearly indepen-
dent. Since dim N(T ) = ∞, we have {u0 , v0 , Tu0 , T v0 }⊥ ∩N(T ) is infinite-dimensional.
It follows that there is an orthonormal subset {ui , vi : i  1} of {u0 , v0 , Tu0 , T v0 }⊥ ∩

N(T ) with  an infinite-dimensional orthogonal complement. Let H1 = {ui , vi : i  0}
and H2 = ({Tu0 , T v0 } ∪ {ui , vi : i  0}). We can choose an orthonormal subset

{ξi }ki=0 ⊆ H2⊥ such that H2⊥ = {ξi : i = 0, 1, · · · , k} ⊕ M , where M is an infinite-
dimensional subspace. Take any two unit orthogonal vectors η1 , η2 ∈ H1⊥ such that
  
H1⊥ = {η1 , η2 } ⊕ H2⊥ = {η1 , η2 } ⊕ M ⊕ {ξi : i = 0, 1, · · · , k}.

We define an operator S ∈ B(H ) by:

⎧
⎪
⎪ Su0 = −Tu0 , Sv0 = T v0 ;
⎨
Sui+1 = ui , Svi+1 = vi , ∀ i  0;
⎪
⎪ Sξi = 0, i = 0, 1, · · · , k;
⎩ 
S : {η1 , η2 } ⊕ M → M is a bounded invertible linear operator.

It follows that 0 ∈ σk (S) and (S + T )u0 = (S − T )v0 = 0 , (S + T )i ui = u0 , (S −

T )i v
i = v0 for all i  0. This implies that asc(S + T ) = asc(S − T ) = ∞, and thus 0 ∈ /
σm (S + T ) and 0 ∈/ σm (S − T ).
Case (ii) T = λ I + F for some non-zero complex number λ ∈ C and F is a finite
rank operator.
382 W. S HI AND G. J I

Note that dim N(F) = ∞. Then there exist two separable infinite-dimensional sub-
spaces H1 , H2 ⊆ N(F) such that H1 ⊥ H2 and (H1 ⊕ H2 )⊥ is also infinite-dimensional.

Let {ωi }ki=0 ⊆ (H1 ⊕ H2 )⊥ be an orthonormal subset such that (H1 ⊕ H2 )⊥ = {ωi : i =
0, 1, · · · , k} ⊕ H3 , where H3 is an infinite-dimensional subspace. We define an operator
S ∈ B(H ) by:
⎧
⎪
⎪ S|H = −λ IH1 + λ3 A1 , where A1 is a backward shift operator on H1 ;
⎨ 1
S|H2 = λ IH2 + λ3 A2 , where A2 is a backward shift operator on H2 ;
⎪
⎪ S| = I;
⎩ H3
Sωi = 0, i = 0, 1, 2, · · · , k.

Then 0 ∈ σk (S). Moreover, S + T |H1 = λ3 A1 , S − T |H2 = λ3 A2 . Hence asc(S +

T ) = asc(S − T ) = ∞. We obtain that 0 ∈ / σm (S + T ) and 0 ∈/ σm (S − T ).
Case (iii) dim N(T ) < ∞ and T = λ I + F for any non-zero complex number λ ∈ C
and for any finite rank operator F .
According to this hypothesis, we have for every closed subspace N ⊆ H with
finite codimension, there exists a vector x ∈ N ∩ T −1 N such that the vectors x, T x are
linearly independent. Without loss of generality, let T   12 . Find a unit vector z0
such that z0 and T z0 are linearly independent. Let H0 = {z0 , T z0 }⊥ . Then there
exists a unit vector z1 ∈ H0 ∩ T −1 H0 such that z1 and T z1 are linearly independent.
Let H1 = {z0 , z1 , T z0 , T z1 }⊥ . Then we can choose a unit vector z2 ∈ H1 ∩ T −1 H1
such that z2 and T z2 are linearly independent. Continuing this process, we can get
a sequence of unit vectors {zi }∞ i=0 such that zi and T zi are linearly independent and
{zi+1 , T zi+1 } ⊥ {z j , T z j : j = 0, 1, 2, ..., i} for all i  0 . We can also assume that the

orthogonal complement of {zi , T zi : i  0} is infinite-dimensional. Otherwise we
canreplace {zi }∞ ∞
i=0 by {z2i }i=0 . Let {yi }i=0 ∪
k {xi }∞
i=0 be an orthonormal sequence
of {zi , T zi : i  0}⊥ . For all i  0 , let ζi ∈ {zi , T zi } be a unit vector such that
ζi ⊥ zi. Then T zi = αi zi + βi ζi , where αi , βi ∈ C satisfy |αi |  12 , 0 < |βi |  12 . Let
M1 = {T z0 , T z1 , xi , ζi+2 : i  0} , M2 = {zi : i  0} , M3 = {yi : i = 0, 1, · · · , k} and
M4 = (M1 ⊕ M2 ⊕ M3 )⊥ . We define an operator S ∈ B(H ) by:
⎧
⎪
⎪ Sz0 = −T z0 , Sz1 = T z1 ;
⎪
⎪ i
⎪
⎪ Sz i = zi−2 − (−1) T zi , ∀ i  2;
⎨
Sζ0 = x0 , Sζ1 = x1 ;
⎪
⎪ Sζi+2 = ζi+2 , Sxi = xi+2 ∀ i  0;
⎪
⎪
⎪
⎪ Sy i = 0, ∀ 0  i  k;
⎩
S|M4 = I.
We next prove that S|M1 ⊕M2 is invertible. It is known that S|M1 ⊕M2 is injective
and M1 ⊆ R(S|M1 ⊕M2 ). Let P be the projection on M2 and B1 , B2 ∈ B(M2 ) such that
B1 z0 = B1 z1 = 0 , B1 zi+2 = zi and B2 zi = (−1)i+1 αi zi for all i  0 . Then B1  = 1
and B2   12 . Note that B1 is surjective. So is B1 + B2 . However, we now have
PS|M2 = B1 + B2 . It follows that S|M1 ⊕M2 is surjective. Moreover, Syi = 0 for 0  i  k
and S|M4 = I . Then 0 ∈ σk (S). Also, (S + T )z0 = (S − T )z1 = 0 , (S + T )i z2i = z0 ,
(S − T )i z2i+1 = z1 for all i  0. This implies that asc(S + T ) = asc(S − T ) = ∞, and so
0∈/ σm (S + T ) and 0 ∈ / σm (S − T ).
 A DDITIVE MAPS PRESERVING m - NORMAL EIGENVALUES ON B(H ) 383

According to the three cases, we get that there exists an operator S satisfying
0 ∈ σk (S) such that 0 ∈
/ σm (S + T ) and 0 ∈
/ σm (S − T ). 

C OROLLARY 1. Let m  0 and T ∈ B(H ) be a non-zero operator. Then there

exists an operator S ∈ B(H ) such that 0 ∈ σm (S) but 0 ∈
/ σm (S + T ).

Proof. It is sufficient to assume that T = x ⊗ y is a rank one operator by Propo-

sition 1. Without loss of generality, we may assume that x and y are unit vectors.
Note that σm (·) is similarity invariant. We now can assume that x = y or x, y = 0 ,
that is, T is a rank one projection or a rank one nilpotent operator. Let T = x ⊗ x
be a rank one projection. Put H = {x} ⊕ H1 ⊕ H2 , where dim H1 = m. Let S be
the projection

on H2 . Then 0 ∈ σm (S) but 0 ∈ / σm (S + T ). If x, y = 0 , then we let
H = {x, y} ⊕ H1 ⊕ H2 , where dim H1 = m. In this case, let P be the projection on
H2 and S = y ⊗ x + P. Then S is what we require. 
m
L EMMA 2. Let T = ∑ ei ⊗ fi be a rank-(m+1) operator and λ ∈ C − {0} . If
i=0
λ ∈ σm (T ), then ei , f j  = λ δi j for all i, j = 0, 1, 2, · · · , m, where δi j is the Kronecker
number.

Proof. Suppose that λ ∈ σm (T ). Then we have dim N(T − λ I) > m. It is known

that N(T − λ I) ⊆ R(T ) and dim

R(T ) = m + 1 . Hence, dim N(T − λ I) = m + 1 ,
that is N(T − λ I) = R(T ) = {ei : i = 0, 1, 2, · · · , m} . Then Te j = λ e j for every
m m
j = 0, 1, 2, · · · , m. Now Te j = ∑ e j , fi ei . So ∑ e j , fi ei = λ e j , this implies that
i=0 i=0
ei , f j  = 0 for i = j and ei , fi  = λ , where 0  i, j  m. 

L EMMA 3. Let m  0 and A, B ∈ B(H ). If σm (A + F) = σm (B + F) for all

operator F ∈ B(H ) with rank not greater than m+1, then A = B.

Proof. Let x ∈ H , fix a scalar α ∈ C such that |α | > A + B . We define an
operator
x−2(A − α I)x ⊗ x, if x = 0;
Fx =
0, if x = 0.
Then we have Fx x = Ax − α x . If x = 0 , then α ∈ σ p (A − Fx ) ⊆ σ (A − Fx ). It follows
that α ∈ σ0 (A − Fx ) from the fact that A − Fx   |α | > A  Ae = A − Fx e ,
where Ae is the essential norm of A.
In the following, we will prove that α ∈ σ0 (B − Fx ) if x = 0 . There are two cases:
Case (1) dim N(A − Fx − α I) > m.
Now, we have α ∈ σm (A − Fx), so α ∈ σm (B − Fx ) ⊆ σ0 (B − Fx ).
Case (2) dim N(A − Fx − α I)  m.
Assume that α ∈ / σ0 (B− Fx ). Note that |α | > B  Be = B− Fx e . We obtain
that B − Fx − α I is invertible. Choose m + 1 orthogonal vectors x0 = x, x1 , x2 , · · · , xm
and let Fm = Fx0 + Fx1 + · · · + Fxm . Then (A − Fm )xi = α xi for all 0  i  m. It implies
384 W. S HI AND G. J I

that dim N(A − Fm − α I) > m. Now, A − Fm   |α | > A  Ae = A − Fm e .

Hence α ∈ σm (A − Fm) and hence α ∈ σm (B − Fm). We know that

B − Fm − α I = (B − Fx0 − α I) − (Fx1 + · · · + Fxm ).

Since B − Fx0 − α I is invertible and Fx1 + · · · + Fxm is a rank-m operator, we have

dim N(B − Fm − α I)  m. This is a contradiction. Therefore, we get that α ∈ σ0 (B −
Fx ).
It follows that there exists a non-zero vector yx ∈ H such that (B − Fx )yx = α yx
for any non-zero vector x ∈ H . We claim that if there exist two vectors y1 , y2 ∈
H such that (B − Fx )y1 = α y1 and (B − Fx )y2 = α y2 , then y1 and y2 are linearly
dependent. According to the assumption, we have (B − α I)y1 = Fx y1 and (B − α I)y2 =
Fx y2 . Then Fx y1 and Fx y2 are linearly dependent since Fx is rank one. We may assume
that Fx y1 = μ Fx y2 for some constant μ ∈ C. Then (B − α I)y1 = μ (B − α I)y2 , that
is, (B − α I)(y1 − μ y2 ) = 0 . We know that y1 and y2 are linearly independent. Then
α ∈ σ p (B). But B − α I is invertible since |α | > B . This is a contradiction. Thus y1
and y2 are linearly dependent.
Note that (A − α I)x = Fx x and (B − α I)yx = Fx yx for any non-zero vector x ∈ H .
Then there is an unique non-zero vector yx such that (A − α I)x = Fx x = (B − α I)yx =
Fx yx for any nonzero x ∈ H . We define yx = 0 if x = 0 . Thus, we can define a map
T on H such that T x = yx . Moreover, we have that (A − α I)x = Fx x = (B − α I)T x =
Fx T x for all x ∈ H . This implies that T = (B − α I)−1(A − α I).
If x = 0 , then Fx = x−2 (A − α I)x ⊗ x and Fx T x = Fx x , thus

x−2T x, x(A − α I)x = (A − α I)x.

We obtain that T x, x = x2 = x, x. Note that if x = 0 , then T x = 0 . So we get that

T x, x = x, x, ∀x ∈ H .

Therefore, T = I . That is, we have (B − α I)−1(A − α I) = I , and so A = B. 

T HEOREM 1. Let ϕ be a surjective additive map on B(H ) and m  0 . If

σm (ϕ (T )) = σm (T ) and σm+1 (ϕ (T )) = σm+1 (T ) for all T ∈ B(H ), then there is
an invertible operator A ∈ B(H ) such that ϕ (T ) = ATA−1 for all T ∈ B(H ) or
ϕ (T ) = AT tr A−1 for all T ∈ B(H ), where T tr denotes the transpose of T with re-
spect to an arbitrary but fixed orthonormal basis of H .

Proof. We first show ϕ is injective. Let ϕ (T ) = 0 . If T = 0 , then by Corollary

1, there exists an operator S such that 0 ∈ σm (S) but 0 ∈
/ σm (S + T ). Note that

σm (S + T ) = σm (ϕ (S + T )) = σm (ϕ (S)) = σm (S).

It is a contradiction. Thus, T = 0 .
Let T ∈ B(H ) with dim R(T )  2 . By Proposition 1, there exists an opera-
tor S satisfying 0 ∈ σm+1 (S) such that 0 ∈
/ σm (S + T ) and 0 ∈
/ σm (S − T ). Then
 A DDITIVE MAPS PRESERVING m - NORMAL EIGENVALUES ON B(H ) 385

0∈/ σm (ϕ (S) + ϕ (T )) and 0 ∈ / σm (ϕ (S) − ϕ (T )). We know that 0 ∈ σm+1 (S) =

σm+1 (ϕ (S)) by the assumption. Then by Lemma 1, we have that dim R(ϕ (T ))  2 .
Since ϕ is bijective and ϕ −1 has the same property as ϕ , it follows that ϕ preserves
the set of operators of rank one in both directions.
We claim that ϕ preserves idempotents of rank one and their linear spans in both
directions. That is ϕ (CP) ⊆ Cϕ (P) for every idempotent of rank one P. Let e0 ⊗ f0
m
be a rank one idempotent and let P = ∑ ei ⊗ fi be a rank-(m + 1) idempotent. For any
i=0
m
non-zero λ ∈ C, then ϕ (λ P) = ∑ ϕ (λ ei ⊗ fi ) is also a rank-(m + 1) operator. Note
i=0
that λ ∈ σm (λ P). Then λ ∈ σm (ϕ (λ P)). It follows from Lemma 2 that ϕ (λ e0 ⊗ f0 ) =
λ yλ ⊗ gλ , where yλ ⊗ gλ is a rank one idempotent. In particular, let λ = 1 and let
ϕ (e0 ⊗ f0 ) = y ⊗ g , then we can get y ⊗ g is a rank one idempotent. Thus ϕ preserves
idempotents of rank one in both directions. We claim that ϕ (Ce0 ⊗ f0 ) ⊆ Cϕ (e0 ⊗ f0 ).
Since ϕ (e0 ⊗ f0 ) = y ⊗ g and y, g = 1 , then we can find two vectors z, h ∈ H
such that z, g = 0 , y, h = 0 and z, h = 1 . For y ⊗ h and z ⊗ g , there exist two rank
one operators u ⊗ k and v ⊗ p such that ϕ (u ⊗ k) = y ⊗ h and ϕ (v ⊗ p) = z ⊗ g as ϕ is
surjective. We know that

ϕ (e0 ⊗ f0 + u ⊗ k) = y ⊗ g + y ⊗ h, ϕ (e0 ⊗ f0 + v ⊗ p) = y ⊗ g + z ⊗ g.

Then both e0 ⊗ f0 + u ⊗ k and e0 ⊗ f0 + v ⊗ p are rank one operators. It implies that

λ e0 ⊗ f0 + u ⊗ k and λ e0 ⊗ f0 + v ⊗ p are also rank one operator for any non-zero
λ ∈ C. Fix a non-zero complex number λ , we have ϕ (λ e0 ⊗ f0 ) = λ yλ ⊗ gλ , where
yλ ⊗ gλ is a rank one idempotent. Then ϕ (λ e0 ⊗ f0 + u ⊗ k) = λ yλ ⊗ gλ + y ⊗ h is also
rank one. We obtain that yλ and y are linearly dependent or the same is true for gλ
and h .
We assert that yλ and y are linearly dependent. Otherwise, we have gλ and h
are linearly dependent. Then there exists some non-zero αλ ∈ C such that gλ = αλ h .
Thus
ϕ (λ e0 ⊗ f0 + v ⊗ p) = αλ λ yλ ⊗ h + z ⊗ g.
Since h and g are linearly independent, there is some non-zero βλ ∈ C such that
yλ = βλ z, and so ϕ (λ e0 ⊗ f0 ) = αλ βλ λ z ⊗ h , where z ⊗ h is a rank one idempotent.
We also know that ϕ (λ e0 ⊗ f0 ) = λ yλ ⊗ gλ , where yλ ⊗ gλ is a rank one idempotent.
It follows that αλ βλ λ = λ . Therefore,

ϕ (λ e0 ⊗ f0 ) = λ z ⊗ h.

As ϕ is surjective, we can find two vectors w, l ∈ H such that ϕ (w ⊗ l) = z ⊗ h and

w, l = 1 . It is clear that y ⊗ g + z ⊗ h is a projection of rank two. Then e0 ⊗ f0 + w ⊗ l
is a rank two operator, and then the operator λ e0 ⊗ f0 + w ⊗ l is also rank two. Thus

ϕ (λ e0 ⊗ f0 + w ⊗ l) = (λ + 1)z ⊗ h.

This is a contradiction since ϕ preserves the set of operators of rank one in both direc-
tions. Therefore, we get that yλ and y are linearly dependent. Then there exists some
386 W. S HI AND G. J I

non-zero γλ ∈ C such that yλ = γλ y. Note that ϕ (λ e0 ⊗ f0 + v ⊗ p) = γλ λ y ⊗ gλ + z ⊗

g , which is also rank one. Thus we can find some non-zero μλ ∈ C such that gλ = μλ g
because y and z are linearly independent. So ϕ (λ e0 ⊗ f0 ) = γλ μλ λ y ⊗ g , where y ⊗ g
is a rank one idempotent. We also know that ϕ (λ e0 ⊗ f0 ) = λ yλ ⊗ gλ , where yλ ⊗ gλ
is also a rank-1 idempotent. Thus γλ μλ λ = λ . Therefore,

ϕ (λ e0 ⊗ f0 ) = λ y ⊗ g = λ ϕ (e0 ⊗ f0 ).

According to the above, we have that ϕ preserves idempotents of rank one and their
linear spans in both directions. It follows from Theorem 4.4 in [13] that there is a
bounded invertible linear or conjugate-linear operator A on H such that one of the
following assertions holds.
(1) ϕ (F) = AFA−1 for all finite rank operators F ∈ B(H );
(2) ϕ (F) = AF tr A−1 for all finite rank operators F ∈ B(H ), where F tr is the
transpose of F with respect to an arbitrary but fixed orthonormal basis of H . If
A is conjugate-linear, then ϕ (iP) = A(iP)A−1 = −iϕ (P) or ϕ (iP) = A(iP)tr A−1 =
−iϕ (P) for any rank-(m + 1) idempotent P, which means that σm (iP) = {i} while
σm (ϕ (iP)) = {−i} . This is a contradiction. Thus A must be linear.
Assume that (1) holds. Let T ∈ B(H ) and for any finite rank operator F , we
have

σm (T + F) = σm (ϕ (T ) + ϕ (F))
= σm (ϕ (T ) + AFA−1)
= σm (A(A−1 ϕ (T )A + F)A−1 )
= σm (A−1 ϕ (T )A + F).

Then we get that T = A−1 ϕ (T )A by Lemma 3. Therefore, ϕ (T ) = ATA−1 for all

T ∈ B(H ).
If (2) holds, then we similarly have that ϕ (T ) = AT tr A−1 for all T ∈ B(H ). 

Acknowledgement. The authors would like to thank the referees for many com-
ments to improve the original draft. And the authors also would like to thank Professor
Xiaohong Cao for her valuable suggestions.
This research was supported by National Natural Science Foundation of China
(Grant No. 11371233, Grant No. 11471200) and by the Fundamental Research Funds
for the Central Universities (Grant No. GK201601004).

REFERENCES

[1] B. AUPETIT, Spectrum-preserving linear mapping between Banach algebra or Jordan Banach alge-
bras, J. London Math. Soc. 62, (2000), 917–924.
[2] M. B ENDAOUD , A. B OURHIM , M. B URGOS AND M. S ARIH, Linear maps preservig Fredholm and
Atkinson elements of C∗ -algebra, Linear Multilinear Alg. 57, (2009), 823–838.
[3] X. C AO , S. C HEN, Linear maps between operator algebras preserving certain spectral functions,
Banach J. Math. Anal. 8, (2014), 39–46.
[4] J. C UI AND J. H OU, Additive maps on standard operator algebras preserving parts of the spectrum,
J. Math. Anal. Appl. 282, (2003), 266–278.
 A DDITIVE MAPS PRESERVING m - NORMAL EIGENVALUES ON B(H ) 387

[5] J. D IEUDONN É, Sur une généralisation du groupe orthogonal à quatre variables, Arch. Math. (Basel)
1, (1949), 282–287.
[6] G. J I , Y. G AO, Maps preserving operator pairs whose products are projections, Linear Algebra Appl.
433, (2010), 1348–1364.
[7] L. L IU , G. J I , Maps preserving product X ∗Y + Y X ∗ on factor von Neumann algebras, Linear and
Multilinear Algebra. 59, (2011), 950–955.
[8] D. A. H ERRERO, Approximation of Hilbert space operators, Vol. 1, 2nd ed., Pitman Res. Notes Math.
Ser., Vol. 224, Longman Scientific and Technical, Harlow, 1989.
[9] J. H OU, Spectrum-preserving elementary operators on B(X) , Chinese Annals of Mathematics. 19,
(1998), 511–516.
[10] A. A. JAFARIAN AND A. R. S OUROUR, Spectrum-preserving linear maps, J. Funct. Anal. 66, (1986),
255–261.
[11] M. M ARCUS AND R. P URVES , Linear transformations on algebras of matrices: The invariance of the
elementary symmetric functions, Canad. J. Math. 11, (1959), 383–396.
[12] M. M BEKHTA , V. M ÜLLER , M. O UDGHIRI , Additive preservers of the ascent, descent and related
subsets, to appear in journal of Operator Theory, 2012.
[13] M. O MLADI , P. Š EMRL, Additive mappings preserving operators of rank one, Linear Alg. Appl. 182,
(1993), 239–256.

(Received March 10, 2015) Weijuan Shi

School of Mathematics and Information Science
Shaanxi Normal University
Xi’an 710062, China
e-mail: [email protected]
Guoxing Ji
School of Mathematics and Information Science
Shaanxi Normal University
Xi’an 710062, China
e-mail: [email protected]

Operators and Matrices

www.ele-math.com
[email protected]