J.

OPERATOR THEORY © Copyright by T HETA, 2020

83:2(2020), 299–319
doi: 10.7900/jot.2018aug09.2252

JORDAN MAPS AND PSEUDOSPECTRUM IN C ∗ -ALGEBRAS

A. BOURHIM and J. MASHREGHI

Communicated by Florian-Horia Vasilescu

A BSTRACT. We show that a surjective map ϕ between two unital C ∗ -algebras

A and B , with ϕ(0) = 0, satisfies
Λε ( ϕ( x1 ) − ϕ( x2 )) = Λε ( x1 − x2 ), ( x1 , x2 ∈ A),
where Λε denotes the ε-pseudospectrum, if and only if ϕ is a Jordan ∗-isomor-
phism. We also characterize maps ϕ1 and ϕ2 from A onto B that satisfy
Λε ( ϕ1 ( x1 ) ϕ2 ( x2 )) = Λε ( x1 x2 ), ( x1 , x2 ∈ A),
or some other binary operations, in terms of Jordan ∗-isomorphisms. The main
results imply several other characterizations of Jordan ∗-isomorphisms which
are interesting in their own right.

K EYWORDS : Pseudospectrum, spectrum, nonlinear preservers, Jordan homomor-

phism.

MSC (2010): Primary 47B49; Secondary 46B04, 46H05.

INTRODUCTION

Nonlinear preserver problems, in the most general setting, demand the char-
acterization of maps on an algebra A preserving a certain property, a particular
relation, or a subset of various binary operations on A such as the difference, the
product and the triple product. These problems have been treated mainly in ma-
trix theory and in operator theory. Kowalski and Słodkowski were the first ones
to consider these types of preserver problems, showing in [35] that a complex-
valued function f on a Banach algebra A is linear and multiplicative provided
that f (0) = 0 and f ( x ) − f (y) lies in the spectrum of x − y for all x and y in
A. This generalizes the well-known theorem of Gleason–Kahane–Żelazko in the
theory of Banach algebra [22], [33], and shows that linearity is a part of the con-
clusion rather than a hypothesis. Since then, a number of techniques have been
developed to treat nonlinear preserver problems and many results have been ob-
tained notably by Brešar, Molnár, Šemrl and others; see for instance [5], [7], [17],
[29], [40], [41], [42], [45] and the references therein. In [40], Molnár studied maps
300 A. B OURHIM AND J. M ASHREGHI

preserving the spectrum of operator or matrix products, and his result has been
extended in several directions for uniform algebras and semisimple commuta-
tive Banach algebras; see for instance [9], [10], [16], [21], [23], [24], [26], [27], [28],
[31], [39].
The pseudospectra form an important mathematical object that has found
wealth of applications in diverse areas of mathematics and engineering, and have
been widely studied by many authors in recent years; see for instance [2], [15],
[18], [20], [36], [37], [43], [47] and the references therein. In [47], Trefethen and
Embree discussed decades of ongoing research and advocated the use of pseu-
dospectra when studying matrices and operators. In [15], the pseudospectra of
several special classes of matrices are computed, and complete descriptions are
given of all maps on matrices preserving the pseudospectra of T • S of matrix
pairs for different kinds of binary operations “•” such as the difference T − S and
the product TS. In [18], Cui, Li and Poon developed additional tools to study
similar problems on the algebra of all bounded linear operators on an infinite-
dimensional complex Hilbert space. Linear preservers of pseudospectrum be-
tween complex Banach algebras have been also studied in [37] by Kumar and
Kulkarni. Among other things, it is in particular shown that a linear map be-
tween complex Banach algebras preserving the pseudospectrum also preserves
spectrum, and in some special cases such a map is shown to be an isometry.
Let A be a unital complex Banach algebra, and L(H) be the algebra of all
bounded linear operators on a complex Hilbert space H. For any ε > 0, the ε-
pseudospectrum of an element x ∈ A is
n 1o
Λε ( x ) := λ ∈ C : k(λ − x )−1 k > ,
ε

with the convention that k(λ − x )−1 k = ∞ if λ − x is not invertible. It is a compact

subset of C and contains σ ( x ), the spectrum of x. In this paper, we extend the
results of [15], [18], [37] to the general setting of C ∗ -algebras. We show that a
surjective map ϕ : A → B between two C ∗ -algebras A and B satisfies

Λε ( ϕ( x1 ) − ϕ( x2 )) = Λε ( x1 − x2 ), ( x1 , x2 ∈ A),

if and only if ϕ − ϕ(0) is a unital bijective isometry. We also address the question
of describing all surjective maps ϕ : A → B satisfying

Λε ( ϕ1 ( x1 ) ϕ2 ( x2 )) = Λε ( x1 x2 ), ( x1 , x2 ∈ A).

Furthermore, we also study all surjective maps ϕ : A → B satisfying

Λε ( ϕ( x1 ) • ϕ( x2 )) = Λε ( x1 • x2 ), ( x1 , x2 ∈ A),

where x1 • x2 stands either for the triple product x1 x2 x1 or the skew product x1 x2∗
or the skew triple product x1 x2∗ x1 .
 J ORDAN MAPS AND PSEUDOSPECTRUM IN C ∗ - ALGEBRA 301

1. MAIN RESULTS

Throughout this paper, let A and B be unital C ∗ -algebras, and denote by Asa
(respectively Bsa ) the real vector space of all self-adjoint elements of A (respec-
tively of B ). Recall that a linear map ϕ from A into B is said to be a homomorphism
(respectively anti-homomorphism) if ϕ( x1 x2 ) = ϕ( x1 ) ϕ( x2 ) (respectively ϕ( x1 x2 ) =
ϕ( x2 ) ϕ( x1 )) for all x1 , x2 ∈ A. It is called a Jordan homomorphism if ϕ( x2 ) = ϕ( x )2
for all x ∈ A, or equivalently if ϕ( x1 x2 + x2 x1 ) = ϕ( x1 ) ϕ( x2 ) + ϕ( x2 ) ϕ( x1 ) for
all x1 , x2 ∈ A. Note that if ϕ : A → B is a Jordan homomorphism then
(1.1) ϕ ( x1 x2 x1 ) = ϕ ( x1 ) ϕ ( x2 ) ϕ ( x1 ), ( x1 , x2 ∈ A).
A bijective homomorphism (respectively anti-homomorphism) is called an isomor-
phism (respectively anti-isomorphism) and a bijective Jordan homomorphism is
called a Jordan isomorphism. Finally, recall that an isomorphism ϕ from A to B
is called a ∗-isomorphism if it is an isomorphism and ϕ( x ∗ ) = ϕ( x )∗ for all x ∈ A.
In a similar way, ∗-anti-isomorphisms and Jordan ∗-isomorphisms are defined. Kadi-
son, in his celebrated paper [32], proved that a surjective linear map between two
C ∗ -algebras A and B is an isometry if and only if it is a Jordan ∗-isomorphism
multiplied by a unitary element in B . It is also worth mentioning that a surjective
linear map ϕ on L(H) is an isometry if and only if there are two unitary operators
U, V ∈ L(H) such that
either ϕ( T ) = UTV, or ϕ( T ) = UT tr V, ( T ∈ L(H)).
Here, T tr denotes the transpose of T relative to an arbitrary but fixed orthonormal
basis of H.
The following theorem is the first main result of this paper. It describes all
ε-pseudospectrum preservers of the difference of element pairs in C ∗ -algebras. It
generalizes Theorem 3.1 of [18] which states that if ε > 0, then a surjective map
on L(H) satisfies
(1.2) Λε ( ϕ( T ) − ϕ(S)) = Λε ( T − S), (S, T ∈ L(H)),
if and only if ϕ − ϕ(0) takes one of the above forms with V = U ∗ , the adjoint of U.
T HEOREM 1.1. Let ε > 0, and let ϕ be a map from A onto B . Then
(1.3) Λε ( ϕ( x1 ) − ϕ( x2 )) = Λε ( x1 − x2 ), ( x1 , x2 ∈ A),
if and only if ϕ − ϕ(0) is a Jordan ∗-isomorphism.
The proof of Theorem 1.1 is long and several techniques are used. In par-
ticular, we use certain properties of the ε-pseudospectrum together with some
arguments that are influenced by ideas from [18], [36]. Such tools are used to
show that if (1.2) is satisfied then ϕ − ϕ(0) is a linear map preserving self-adjoint
elements in both directions. In the rest of the proof, we use different arguments
and methods than the ones presented in the proof of Theorem 3.1 of [18], which
302 A. B OURHIM AND J. M ASHREGHI

are based on rank one operators, as a matter of fact, that approach cannot be used
in the general case since there are C ∗ -algebras without any rank one element.
In Theorem 4.1 of [18], Cui, Li and Poon proved that if ε > 0, then a surjec-
tive map on L(H) satisfies

(1.4) Λε ( ϕ( T ) ϕ(S)) = Λε ( TS), (S, T ∈ L(H)),

if and only if there exists a unitary operator U ∈ L(H) such that ϕ( T ) = ±UTU ∗
for all T ∈ L(H). Our next result describes all ε-pseudospectrum preservers of
the product of element pairs in C ∗ -algebras but when one of the algebra has an
essential socle. It extends Theorem 4.1 of [18] to the general setting of C ∗ -algebras
with essential socles. Before stating such a result, we recall few definitions and
fix some notations. If E is a unital complex Banach algebra and has a minimal left
(or right) ideal, then its socle, denoted by Soc(E ), is the sum of all minimal left (or
right) ideals of E . If E has no minimal one-sided ideals, the socle is then defined
to be trivial; i.e., Soc(E ) = {0}. Note that Soc(E ) is an ideal of E consisting of all
elements a ∈ E for which σ ( xa) is finite for all x ∈ E , and that all its elements
are algebraic. An ideal P of A is essential if 0 is the only element a of E for which
a · P = 0. We refer the reader to [3], [4] for more details. It should be noted that
a semisimple Banach algebra is finite dimensional if and only if it coincides with
its socle; see for instance Theorem 5.4.2 of [3].
T HEOREM 1.2. Let ε > 0, and let ϕ1 and ϕ2 be two maps from A onto B . Assume
that A has an essential socle. Then

(1.5) Λε ( ϕ1 ( x1 ) ϕ2 ( x2 )) = Λε ( x1 x2 ), ( x1 , x2 ∈ A),

if and only if ϕ2 (1) ϕ1 (1) = 1 and ϕ1 ϕ2 (1) = ϕ1 (1) ϕ2 is a ∗-isomorphism.

By Lemma 2.3 below, it is easy to show that if (1.5) is satisfied then both
ϕ1 and ϕ2 are bijective, and thus ϕ1 −1 and ϕ2 −1 as well satisfy (1.5). Hence, the
conclusion of the previous theorem remains valid if one of the two C ∗ -algebras
A and B has an essential socle. We also would like to point out that the proof of
Theorem 1.2 uses as well properties of the ε-pseudospectrum to show that if ϕ1
and ϕ2 are two maps from A onto B for which (1.5) is satisfied, then

σ ( ϕ1 ( x1 ) ϕ2 ( x2 )) = σ ( x1 x2 ), ( x1 , x2 ∈ A).

Once this is proved, Theorem 2.1 of [10] tells us that if A has an essential socle,
then ϕ2 (1) ϕ1 (1) = 1 and both ϕ1 ϕ2 (1) and ϕ1 (1) ϕ2 coincide and are Jordan
isomorphisms. The last part of the proof is then devoted to the self-adjointness
and multiplicativity of both maps ϕ1 ϕ2 (1) and ϕ1 (1) ϕ2 .
The next result describes the form of all maps between C ∗ -algebras preserv-
ing the ε-pseudospectrum of triple product of element pairs in C ∗ -algebras when
one of the algebra has an essential socle.
 J ORDAN MAPS AND PSEUDOSPECTRUM IN C ∗ - ALGEBRA 303

T HEOREM 1.3. Let ϕ be a map from A onto B , and assume that A has an essential
socle. Then

(1.6) Λε ( ϕ( x1 ) ϕ( x2 ) ϕ( x1 )) = Λε ( x1 x2 x1 ), ( x1 , x2 ∈ A),

if and only if ϕ(1) is a central element of B , for which ϕ(1)3 = 1, and ϕ(1)2 ϕ is a Jordan
∗-isomorphism.
The last result describes the form of all maps between C ∗ -algebras preserv-
ing the spectrum and the ε-pseudospectrum of skew triple product of element
pairs in C ∗ -algebras when one of the algebras has an essential socle.
T HEOREM 1.4. Let ϕ be a map from A onto B , and assume that A has an essential
socle. Then the following assertions are equivalent:
(i) for every x1 , x2 ∈ A, we have

(1.7) Λε ( ϕ( x1 ) ϕ( x2 )∗ ϕ( x1 )) = Λε ( x1 x2 ∗ x1 );

(ii) for every x1 , x2 ∈ A, we have

(1.8) σ ( ϕ( x1 ) ϕ( x2 )∗ ϕ( x1 )) = σ( x1 x2 ∗ x1 );

(iii) ϕ is a Jordan ∗-isomorphism.

The rest of this paper is organized as follows. In Section 2, we prove four
lemmas. The first two provide the basic properties of pseudospectra that will be
used through this paper. The other two lemmas give, in term of pseudospectra,
the necessary and sufficient conditions for two elements in a semisimple Banach
algebra to be the same. In Section 3, we present the proof of Theorem 1.1. The
proofs of Theorems 1.2, 1.3 and 1.4 are given in Section 4, Section 5 and Section 6,
respectively. In the last two sections, we list several consequences of our main
results and provide a number of comments and remarks.

2. PRELIMINARIES

All algebras which we consider throughout this paper are assumed to be

unital and complex. In this section, we provide few more notation and some
lemmas that are needed in the sequel. For any λ ∈ C and ε > 0, let D (λ, ε) be the
closed disc of center λ and radius ε and, as usual, denote by =(z) the imaginary
part of any complex number z ∈ C. Let E be a Banach algebra, and denote, as
usual, the spectral radius of any element x ∈ E by r( x ) := max{|λ| : λ ∈ σ ( x )}.
A nonzero element u ∈ E is said to have rank one if for every x ∈ E , the spectrum
σ (ux ) contains at most one nonzero scalar. The set F1 (E ) of all rank one elements
of E is contained in Soc(E ) and in its turn Soc(E ) is equal to the set of all finite
sums of rank one elements of E . In [4], [13], it is shown that there is a function
304 A. B OURHIM AND J. M ASHREGHI

τ : F1 (E ) → F1 (E ) such that
(
{τ (u)} if dim(E ) = 1,
σ(u) =
{0, τ (u)} if dim(E ) > 1,
for all u ∈ F1 (E ), and
(2.1) τ ( x1 u + x2 u ) = τ ( x1 u ) + τ ( x2 u )
for all x1 , x2 ∈ E and u ∈ F1 (E ).
The pseudospectra of matrices and Hilbert space operators have been widely
studied in the literature and several of their essential properties have been estab-
lished. In [2], [36], a treatment of the pseudospectra in the general context of
Banach algebras is presented. It is shown that some of the basic properties of
pseudospectra of matrices and operators remain valid for the pseudospectra of
elements in an arbitrary Banach algebra. The following lemma summarizes some
of them which will be used in the sequel.
L EMMA 2.1. For an element a in a Banach algebra E and ε > 0, the following
statements hold:
(i) Λε (λa) = λΛε/|λ| ( a) for all nonzero λ ∈ C;
(ii) σ ( a) + D (0, ε) ⊂ Λε ( a);
T
(iii) σ ( a) = Λ δ ( a ).
δ >0
For the proof see Lemma 2.1 of [2] or Theorem 2.3 of [36].
Since the definition of the ε-pseudospectrum depends only on the norm and
the spectrum, a number of known properties of the ε-pseudospectrum carry over
verbatim from the L(H) case to the general C ∗ -algebras via isometric embedding
of any C ∗ -algebra into the algebra of all bounded linear operators on a complex
Hilbert space.
L EMMA 2.2. For an element a in a C ∗ -algebra E and ε > 0, the following state-
ments hold:
(i) if a is normal, then Λε ( a) = σ ( a) + D (0, ε);
(ii) if λ ∈ C, then a = λ1 if and only if Λε ( a) = D (λ, ε);
(iii) if t ∈ R, then eit a is self-adjoint if and only if Λε ( a) ⊂ {z ∈ C : |=(eit z)| 6 ε}.
By Gelfand–Naimark theorem, the proof of this lemma follows from Theo-
rem 2.2 and Corollary 2.3 of [18].
In [1], [14], the question of whether for two given elements a and b in a
Banach algebra E the condition σ ( ax ) = σ(bx ) for all x ∈ E entails a = b was
studied and an affirmative answer was obtained for various special cases and
for some classes of algebras, including C ∗ -algebras. Then Braatvedt and Brits
[11] showed that the answer is always affirmative if E is a semisimple Banach
algebra. In particular, they proved that two elements a and b in such an algebra
E coincide if and only if σ( ax ) = σ(bx ) for all x ∈ E for which r( x − 1) < 1.
 J ORDAN MAPS AND PSEUDOSPECTRUM IN C ∗ - ALGEBRA 305

In [36], Krishnan and Kulkarni discussed an analogous question of whether for

given ε > 0 and two elements a, b ∈ E the condition Λε ( ax ) = Λε (bx ) for all
x ∈ E implies that a = b. They obtained a positive answer when a is a particular
element or/and E is a particular algebra. The following lemma shows that the
answer is always positive whenever E is a semisimple Banach algebra and will
be used in the proof of Theorem 1.2. Its proof is a combination of some ideas from
[36] and Theorem 2.6 of [11].
L EMMA 2.3. If ε > 0 and a and b are two elements in a semisimple Banach algebra
E , then the following are equivalent:
(i) a = b;
(ii) Λε ( ax ) = Λε (bx ) for all x ∈ E ;
(iii) Λε ( ax ) = Λε (bx ) for all invertible elements x ∈ E .
Proof. We only need to establish the implication (iii) ⇒ (i). So, assume that
Λε ( ax ) = Λε (bx ) for all invertible elements x ∈ E . Then Lemma 2.1(i) tells us that
tΛε/t ( ax ) = Λε (tax ) = Λε (tbx ) = tΛε/t (bx )
for all invertible elements x ∈ E and positive scalars t. This and Lemma 2.1(iii)
imply that σ ( ax ) = σ (bx ) for all invertible elements x ∈ E . Therefore, by Theo-
rem 2.6 of [11], we conclude that a = b.

An analogous question to the one treated in [1], [11], [14] asks whether for
given two elements a and b in a Banach algebra E the condition σ( xax ) = σ ( xbx )
for all x ∈ E entails a = b. This question has been discussed in [10] where it is
shown that the answer is affirmative.
L EMMA 2.4. Let E be a semisimple Banach algebra and let a, b be two elements in
E . Then a = b if and only if σ( xax ) = σ( xbx ) for all x ∈ E .
For the proof see Theorem 2.6 of [10].
When the algebra E has an essential socle, a stronger result is presented in
[10], which is needed in the proof of Theorem 1.4.
L EMMA 2.5. Let E be a semisimple Banach algebra with an essential socle. Then
two elements a and b in E are equal if and only if τ (uau) = τ (ubu) for all u ∈ F1 (E ).
For the proof see Lemma 3.5 of [10].
The last lemma is a variant of Lemma 2.3 and is needed in the proof of The-
orem 1.3. It addresses the question of whether for given ε > 0 and two elements
a and b in a semisimple Banach algebra E the condition Λε ( xax ) = Λε ( xbx ) for
all x ∈ E entails a = b.
L EMMA 2.6. If ε > 0 and a and b are two elements in a semisimple Banach algebra
E , then the following assertions are equivalent:
(i) a = b;
(ii) Λε ( xax ) = Λε ( xbx ) for all x ∈ E ;
306 A. B OURHIM AND J. M ASHREGHI

(iii) Λε ( xax ) = Λε ( xbx ) for all invertible elements x ∈ E ;

(iv) Λε (ex aex ) = Λε (ex bex ) for all x ∈ E .
Proof. We only need to establish the implication (iv) ⇒ (i). Assume that
Λε (ex aex ) = Λε (ex bex )
for all x ∈ E , and note that Λε (ex+t1 aex+t1 ) = Λε (ex+t1 bex+t1 ) for all x ∈ E and
t ∈ R. This and Lemma 2.1(i) imply that tΛε/t (ex aex ) = tΛε/t (ex bex ) for all x ∈ E
and t > 0. This in its turn together with Lemma 2.1(iii) entail that
(2.2) σ (ex aex ) = σ(ex bex )
for all x ∈ E . Now, fix an element x ∈ E such that r( x − 1) < 1 and note that, by
the holomorphic functional calculus, x = ey for some y ∈ E . By 2.2, we have

σ (ey/2 aey/2 ) = σ (ey/2 bey/2 ).

By Jacobson’s lemma, we then have σ ( ax )\{0} = σ (bx )\{0}. But, since x is in-
vertible, we actually obtain σ ( ax ) = σ(bx ). Finally, since x ∈ E is an arbitrary
element for which r( x − 1) < 1, we deduce from Theorem 2.6 of [11] that a = b.

3. PROOF OF THEOREM 1.1

Checking the “if ” part is straightforward since any Jordan ∗-isomorphism is

a unital bijective isometry preserving invertibility in both directions. Conversely,
assume that (1.3) is satisfied and note that, since ϕ − ϕ(0) satisfies (1.3), we may
and will assume that ϕ(0) = 0. We break the proof into several steps, in which
we use some arguments that are influenced by ideas from [8], [18], [36]. We also
make use, on a number of occasions, of the properties of pseudospectra listed in
Lemma 2.1.
Step 1. ϕ is injective and ϕ(1) = 1.
Assume that ϕ( a1 ) = ϕ( a2 ) for some elements a1 , a2 ∈ A. We have
Λε ( a1 − x ) = Λε ( ϕ( a1 ) − ϕ( x )) = Λε ( ϕ( a2 ) − ϕ( x )) = Λε ( a2 − x ), ( x ∈ A).
In particular, for x = a2 , we have Λε ( a1 − a2 ) = Λε (0) = D (0, ε). By Lemma 2.2(ii),
we see that a1 = a2 and ϕ is injective. Now, since ϕ(0) = 0, we deduce Λε ( ϕ(1)) =
Λε (1) = D (1, ε). Again by Lemma 2.2(ii), we see that ϕ(1) = 1.
Step 2. ϕ is R-linear such that ϕ(Asa )= Bsa and σ ( ϕ(h)) = σ (h) for all h ∈ Asa .
Let x1 and x2 be two elements of A. By Lemma 2.2(iii), we see that x2 − x1
is self-adjoint if and only if ϕ( x2 ) − ϕ( x1 ) is self-adjoint. Thus ϕ : Asa → Bsa is
bijective, and by Lemma 2.2(i), it verifies
σ (h2 − h1 )+ D (0, ε) = Λε (h2 − h1 ) = Λε ( ϕ(h2 )− ϕ(h1 )) = σ ( ϕ(h2 )− ϕ(h1 ))+ D (0, ε)
 J ORDAN MAPS AND PSEUDOSPECTRUM IN C ∗ - ALGEBRA 307

for all h1 , h2 ∈ Asa . Therefore, σ (h2 − h1 ) = σ ( ϕ(h2 ) − ϕ(h1 )) and

k ϕ(h2 ) − ϕ(h1 )k = kh2 − h1 k, (h1 , h2 ∈ Asa ).
By Mazur–Ulam theorem [38], ϕ is R-linear and σ( ϕ(h)) = σ (h) for all h ∈ Asa .
Step 3. ϕ : iAsa → iBsa is bijective R-linear and ϕ(ih) = iϕ(h) for all h ∈ Asa .
Just as in the proof of the previous step, Lemma 2.2(iii) tells us that ϕ maps
iAsa onto iBsa . This and the bijectivity of ϕ ensure that for any h ∈ Asa , there
exists k ∈ Asa such that ϕ(ih) = iϕ(k). Thus, by Lemma 2.2(iii),
{z ∈ C : |=(z)| < ε} ⊃ Λ√2 ε (2h) = Λ√2 ε ((1 + i)(h − ih)) = (1 + i)Λε (h − ih)
= (1 + i)Λε ( ϕ(h) − iϕ(k))
√
= 2 εΛε ((1 + i)( ϕ(h) − iϕ(k))).
Again Lemma 2.2(iii) tells us that (1 + i)( ϕ(h) − iϕ(k)) is self-adjoint. Therefore,
ϕ(h) + iϕ(h) − iϕ(k) + ϕ(k ) = (1 + i)( ϕ(h) − iϕ(k)) = ((1 + i)( ϕ(h) − iϕ(k)))∗
= (1 − i)( ϕ(h) + iϕ(k)) = ϕ(h)− iϕ(h)+ iϕ(k)+ ϕ(k).
This implies that ϕ(h) = ϕ(k) and then h = k, by the injectivity of ϕ. We then
conclude that ϕ(ih) = iϕ(h) and the proof of this step is complete.
Step 4. ϕ : A → B is a bijective C-linear map.
Fix an element x = h + ik ∈ A, and set
ψ ( z ) : = ϕ ( h + z ) − ϕ ( h ), (z ∈ A).
Clearly, ψ satisfies (1.3) and Step 3 tells us that ψ(ik ) = iψ(k). This and the R-
linearity of ϕ when restricted on Asa entail that
ϕ( x ) = ϕ(h + ik ) = ϕ(h) + iϕ(k ).
Now, let λ = t1 + it2 ∈ C and deduce
ϕ(λx ) = ϕ(λ(h + ik)) = ϕ((t1 + it2 )(h + ik )) = ϕ((t1 h − t2 k) + i(t1 k + t2 h))
= ϕ((t1 h − t2 k)) + iϕ((t1 k + t2 h)) = λϕ(h + ik) = λϕ( x ).
Finally, let x1 = h1 + ik1 and x2 = h2 + ik2 , and conclude that
ϕ( x1 + x2 ) = ϕ((h1 + h2 ) + i(k1 + k2 )) = ϕ((h1 + h2 )) + iϕ((k1 + k2 ))
= ϕ(h1 ) + ϕ(h2 ) + iϕ(k1 ) + iϕ(k2 ) = ϕ( x1 ) + ϕ( x2 ).

Step 5. k( ϕ( x ))−1 k = k x −1 k for all invertible elements x ∈ A.

Suppose there exists an invertible element x ∈ A such that k( ϕ( x ))−1 k 6=
k x −1 k .We may and will assume that k( ϕ( x ))−1 k > k x −1 k, as the case when
k( ϕ( x ))−1 k < k x −1 k can be dealt in a similar way. Choose a rational number
t > 0 such that
t
k x −1 k < 6 k( ϕ( x ))−1 k.
ε
308 A. B OURHIM AND J. M ASHREGHI

It then follows that 0 ∈ Λε (t−1 x ) and 0 ∈

/ Λε ( ϕ(t−1 x )), which is a contradiction
− 1 − 1
and thus k( ϕ( x )) k = k x k for all invertible elements x ∈ A.
Step 6. ϕ is a Jordan ∗-isomorphism.
Keep in mind that ϕ preserves self-adjoint elements in both directions. Given
a := h + ik ∈ A where h and k are the real and imaginary parts of z, we have
ϕ( a∗ ) = ϕ(h − ik) = ϕ(h) − iϕ(k) = ( ϕ(h) + iϕ(k))∗ = ϕ( a)∗ .
This implies that ϕ is a self-adjoint map. Since by Step 2 it preserves the spectrum
of self-adjoint elements and positive elements are self-adjoint elements with non-
negative spectrum, we see that ϕ is a positive map. As ϕ(1) = 1, Corollary 1 of
[44] tells us that k ϕk = 1, and then k ϕ(u)k 6 1 for all unitary elements u ∈ A.
Since k( ϕ(u))−1 k = 1 by Step 5, we infer that ϕ(u) is unitary whenever u is a
unitary element of A.
Now, let h be a self-adjoint element in A so that u = eith ∈ A and ϕ(u) are
unitary elements for all reals t ∈ R. We have ϕ(u) = 1 + itϕ(h) − (1/2)t2 ϕ(h2 ) +
· · · and ϕ(u∗ ) = 1 − itϕ(h) − (1/2)t2 ϕ(h2 ) + · · · , and then
 1  1 
1 = ϕ(u) ϕ(u)∗ = 1 + itϕ(h) − t2 ϕ(h2 ) + · · · 1 − itϕ(h) − t2 ϕ(h2 ) + · · · .
2 2
Expand the right side of this identity to conclude that ϕ(h)2 = ϕ(h2 ) for all self-
adjoint elements h ∈ A. Therefore, ϕ is a Jordan ∗-isomorphism.

4. PROOF OF THEOREM 1.2

Assume that ϕ2 (1) ϕ1 (1) = 1 and both maps ϕ1 ϕ2 (1) and ϕ1 (1) ϕ2 coincide
and are ∗-isomorphisms. For every x1 , x2 ∈ A, we have
Λε ( x1 x2 ) = Λε ( ϕ1 (1) ϕ2 ( x1 x2 )) = Λε ({ ϕ1 (1) ϕ2 ( x1 )}{ ϕ1 (1) ϕ2 ( x2 )})
= Λε ({ ϕ1 ( x1 ) ϕ2 (1)}{ ϕ1 (1) ϕ2 ( x2 )}) = Λε ( ϕ1 ( x1 ) ϕ2 ( x2 )),
and (1.5) is satisfied. Conversely, assume that A has an essential socle, and ϕ1
and ϕ2 are two maps from A onto B satisfying (1.5), and let us break down the
proof of Theorem 1.2 into several steps.
Step 1. ϕ1 (1) ϕ2 (1) = ϕ2 (1) ϕ1 (1) = 1.
Since ϕ1 is surjective, there is x0 ∈ A such that ϕ( x0 ) = 1. Then D (0, ε) =
Λε (0) = Λε ( ϕ1 ( x0 ) ϕ2 (0)) = Λε ( ϕ2 (0)), and Lemma 2.2(ii) tells us that ϕ2 (0) = 0.
Now, note that D (1, ε) = Λε (1) = Λε ( ϕ1 (1) ϕ2 (1)), and then Lemma 2.2(ii) tells
us that ϕ1 (1) ϕ2 (1) = 1. So, we only need to show that ϕ2 (1) ϕ1 (1) = 1. As ϕ2 is
surjective, there exists x such that ϕ2 ( x ) = 1 − ϕ2 (1) ϕ1 (1). We have
ϕ1 (1) ϕ2 ( x ) = ϕ1 (1)(1 − ϕ2 (1) ϕ1 (1)) = ϕ1 (1) − ϕ1 (1) ϕ2 (1) ϕ1 (1) = 0.
Then D (0, ε) = Λε ( ϕ1 (1) ϕ2 ( x )) = Λε ( x ), which by Lemma 2.2(ii) gives x = 0.
Therefore, 1 − ϕ2 (1) ϕ1 (1) = ϕ2 ( x ) = ϕ2 (0) = 0 and the result is established.
 J ORDAN MAPS AND PSEUDOSPECTRUM IN C ∗ - ALGEBRA 309

Step 2. The maps ϕ1 and ϕ2 are homogeneous and

(4.1) σ ( ϕ1 ( x1 ) ϕ2 ( x2 )) = σ ( x1 x2 ), ( x1 , x2 ∈ A).

Fix an invertible element a ∈ A and let ψ1 ( x ) := ϕ1 ( xa) and ψ2 ( x ) :=

ϕ2 ( a−1 x ) for all x ∈ A. Clearly, we have
Λε (ψ1 ( x1 )ψ2 ( x2 )) = Λε ( x1 x2 ), ( x1 , x2 ∈ A),
and thus, by Step 1, ψ1 (1) = {ψ2 (1)}−1 . Therefore, ϕ1 ( a) = { ϕ2 ( a−1 )}−1 for all
invertible elements a ∈ A. Now, we prove that ϕ1 and ϕ2 are homogeneous. For
every invertible element x ∈ A, we have
D (λ, ε) = Λε (λ1) = Λε ( ϕ1 (λx ) ϕ2 ( x −1 )), (λ ∈ C).
Thus ϕ1 (λx ) ϕ2 ( x −1 )
= λ1 and ϕ1 (λx ) = λϕ2 ( x −1 ) −1
= λϕ1 ( x ) for all λ ∈ C. In
a similar way, one shows that ϕ2 (λx ) = λϕ2 ( x ) for all invertible elements x ∈ A
and λ ∈ C. Now, take an invertible element x1 ∈ A and an arbitrary element
x2 ∈ A, and let λ ∈ C. We have
Λε ( ϕ1 ( x1 ) ϕ2 (λx2 )) = Λε (λx1 x2 ) = Λε ( ϕ1 (λx1 ) ϕ2 ( x2 )) = Λε (λϕ1 ( x1 ) ϕ2 ( x2 ))
= Λε ( ϕ1 ( x1 )(λϕ2 ( x2 ))).
By Lemma 2.3, we conclude that ϕ2 (λx2 ) = λϕ2 ( x2 ), and thus ϕ2 is homoge-
neous. In a similar way, we show that ϕ1 is also homogeneous.
Lastly, fix two elements x1 , x2 ∈ A and t > 0, and note that
tΛε/t ( ϕ1 ( x1 ) ϕ2 ( x2 )) = Λε ( ϕ1 (tx1 ) ϕ2 ( x2 )) = Λε (tx1 x2 ) = tΛε/t ( x1 x2 ).
This and Lemma 2.1(iii) imply that σ ( ϕ1 ( x1 ) ϕ2 ( x2 )) = σ ( x1 x2 ) for all x1 , x2 ∈ A.
Step 3. The maps ϕ1 ϕ2 (1) and ϕ1 (1) ϕ2 coincide and are Jordan ∗-isomorphisms.
Since (4.1) holds and A has an essential socle, Theorem 2.1 of [10] tells us
that ϕ1 ϕ2 (1) and ϕ1 (1) ϕ2 coincide and are Jordan isomorphisms. Moreover, we
have
Λε ( ϕ1 ( x ) ϕ2 (1)) = Λε ( x ), ( x ∈ A).
Then Lemma 2.2(iii) implies that ϕ1 ϕ2 (1) preserves the self-adjoint elements in
both directions. Therefore, just as in the proof of Theorem 1.1, one can show that
ϕ1 ( x ∗ ) ϕ2 (1) = ( ϕ1 ( x ) ϕ2 (1))∗ , ( x ∈ A).
Thus, ϕ1 ϕ2 (1) and ϕ1 (1) ϕ2 are Jordan ∗-isomorphism.
Step 4. The maps ϕ1 ϕ2 (1) and ϕ1 (1) ϕ2 are multiplicative.
Set ϕ( x ) := ϕ1 ( x ) ϕ2 (1), x ∈ A, and keep in mind that φ is a Jordan ∗-
isomorphism and that
Λε (φ( x1 )φ( x2 )) = Λε ( x1 x2 ), ( x1 , x2 ∈ A).
For any x1 and x2 in A, Lemma 2.2(ii) and the above identity tell us that
(4.2) x1 x2 = 0 ⇔ φ( x1 )φ( x2 ) = 0.
310 A. B OURHIM AND J. M ASHREGHI

We also note that if x1 and x2 are in A such that x1 x2 = 0, then

(4.3) φ ( x2 x1 ) = φ ( x1 x2 + x2 x1 ) = φ ( x1 ) φ ( x2 ) + φ ( x2 ) φ ( x1 ) = φ ( x2 ) φ ( x1 ).
We shall prove that
(4.4) φ(ux ) = φ(u)φ( x )
for all x ∈ A and all rank one elements u ∈ A.
Fix an element x ∈ A and pick up a nonquasinilpotent rank one element
u ∈ A and keep in mind that τ (u) 6= 0 and u2 = τ (u)u. Then xu(τ (u)1 − u) = 0 and
(4.2) applied to x1 := xu and x2 := τ (u)1 − u gives φ( xu)φ((τ (u)1 − u)) = 0 and
(4.5) τ (u)φ( xu) = φ( xu)φ(u).
We also have x (τ (u)1 − u)u = 0 and then (4.3) applied to x1 := x (τ (u)1 − u)
and x2 := u implies that φ(ux (τ (u)1 − u)) = φ(u)φ( x (τ (u)1 − u)). This together
with (1.1) and (4.5) entail that
τ (u)φ(ux ) − φ(uxu) = φ(ux (τ (u)1 − u)) = φ(u)φ( x (τ (u)1 − u))
= φ(u)(τ (u)φ( x ) − φ( xu)) = τ (u)φ(u)φ( x ) − φ(u)φ( xu)
1
= τ (u)φ(u)φ( x ) − φ(u)φ( xu)φ(u)
τ (u)
1
= τ (u)φ(u)φ( x ) − φ(uxu2 ) = τ (u)φ(u)φ( x ) − φ(uxu).
τ (u)
This shows that (4.4) holds for all x ∈ A and all rank one elements u ∈ A for
which τ (u) 6= 0.
Now, pick up a rank one element u ∈ A such that τ (u) = 0 and note that

τ ((1 − u∗ )u) = −τ (u∗ u) = −kuk2 6= 0.

Thus (4.4) tells us that φ(u∗ ux ) = φ(u∗ u)φ( x ) and φ((1 − u∗ )ux ) = φ((1 −
u∗ )u)φ( x ) for all x ∈ A. Therefore,
φ(ux ) = φ((1 − u∗ )ux + u∗ ux ) = φ((1 − u∗ )ux ) + φ(u∗ ux )
= φ((1 − u∗ )u)φ( x ) + φ(u∗ u)φ( x ) = φ(u)φ( x )
for all x ∈ A. Clearly, (4.4) holds as well for all x ∈ A and all rank one elements
u ∈ A for which τ (u) = 0.
Lastly, let us show that φ is multiplicative. For every x1 , x2 ∈ A and rank
one element u ∈ A, we have
φ(u)φ( x1 x2 ) = φ(ux1 x2 ) = φ(ux1 )φ( x2 ) = φ(u)φ( x1 )φ( x2 ).
Since φ(Soc(A)) = Soc(B) and Soc(B) is essential, this implies that φ( x1 x2 ) =
φ( x1 )φ( x2 ) for all x1 , x2 ∈ A. Clearly, φ is multiplicative and the proof is then
complete.
 J ORDAN MAPS AND PSEUDOSPECTRUM IN C ∗ - ALGEBRA 311

5. PROOF OF THEOREM 1.3

Assume that ϕ(1) is a central invertible element of B for which ϕ(1)3 = 1

and ϕ(1)2 ϕ is a Jordan ∗-isomorphism. Then
(5.1) Λε ( ϕ(1)2 ϕ( x )) = Λε ( x )
for all x ∈ A. This and (1.1) together with the fact that ϕ(1) is a central invertible
element of B for which ϕ(1)3 = 1 imply that
Λε ( ϕ( x1 ) ϕ( x2 ) ϕ( x1 )) = Λε ({ ϕ(1)2 ϕ( x1 )}{ ϕ(1)2 ϕ( x2 )}{ ϕ(1)2 ϕ( x1 )})
= Λε ( ϕ(1)2 ϕ( x1 x2 x1 )) = Λε ( x1 x2 x1 )
for all x1 , x2 ∈ A. Conversely, assume that (1.6) is satisfied. Fix an invertible
element x ∈ A, and note that
Λε ( ϕ( x −1 ) ϕ( x2 ) ϕ( x −1 )) = Λε (1) = D (1, ε)
By Lemma 2.2(ii), we see that
(5.2) ϕ( x2 ) = { ϕ( x −1 )}−2 .
We also have
Λε ( ϕ( x −1 ) ϕ(λx2 ) ϕ( x −1 )) = Λε (λ1) = D (λ, ε)
from which and (5.2) it follows that ϕ(λx2 ) = λ{ ϕ( x −1 )}−2 = λϕ( x2 ), for all
λ ∈ C. In particular, we see that
(5.3) ϕ(λex ) = ϕ(λ(ex/2 )2 ) = λϕ((ex/2 )2 ) = λϕ(ex )
for all x ∈ A and λ ∈ C. Let x, y ∈ A and λ ∈ C, and let u ∈ A such that
ϕ(u) = λϕ(y). Then
√ √
Λε (λex yex ) = Λε ( ϕ(ex ) ϕ(λy) ϕ(ex )) = Λε ( ϕ( λex ) ϕ(y) ϕ( λex ))
= Λε (λϕ(ex ) ϕ(y) ϕ(ex )) = Λε ( ϕ(ex ) ϕ(u) ϕ(ex )) = Λε (ex uex ).
By Lemma 2.6, we see that u = λy, and thus λϕ(y) = ϕ(u) = ϕ(λy). This implies
tΛε/t ( xyx ) = Λε (txyx ) = Λε ( ϕ( x ) ϕ(ty) ϕ( x )) = Λε (tϕ( x ) ϕ(y) ϕ( x ))
= tΛε/t ( ϕ( x ) ϕ(y) ϕ( x ))
for all elements x, y ∈ A and positive scalars t. By Lemma 2.1(iii), we get that
(5.4) σ( ϕ( x ) ϕ(y) ϕ( x )) = σ ( xyx )
for all x, y ∈ A. By Theorem 2.4 of [10], we conclude that ϕ(1) is a central inver-
tible element of B for which ϕ(1)3 = 1 and ϕ(1)2 ϕ is a Jordan isomorphism.
It remains to show that ϕ(1)2 ϕ( x ∗ ) = { ϕ(1)2 ϕ( x )}∗ for all x ∈ A. We have
Λε ( ϕ(1)2 ϕ( x )) = Λε ( ϕ(1) ϕ( x ) ϕ(1)) = Λε ( x )
312 A. B OURHIM AND J. M ASHREGHI

for all x ∈ A. Then Lemma 2.2(iii) implies that ϕ(1)2 ϕ preserves the selfajoint
elements in both directions. Therefore, just as in the proof of Theorem 1.1, one can
show that ϕ(1)2 ϕ( x ∗ ) = { ϕ(1)2 ϕ( x )}∗ , for all x ∈ A. The proof is thus complete.

6. PROOF OF THEOREM 1.4

The implication (i) ⇒ (ii) is similar to the proof of (5.4). Also, (iii) ⇒ (i) is
an immediate consequence of (1.1) and the fact that Jordan ∗-isomorphisms pre-
serve the ε-pseudospectra. So, we only need to establish the implication (ii) ⇒
(iii). Assume that (1.8) is satisfied and let us first prove that ϕ is homogenous. Fix
arbitrary elements x1 , x2 ∈ A and λ ∈ C, and note that

σ ( ϕ( x1 ){λϕ( x2 )}∗ ϕ( x1 )) = λσ ( ϕ( x1 ) ϕ( x2 )∗ ϕ( x1 )) = λσ( x1 x2 ∗ x1 )

= σ( x1 {λx2 }∗ x1 ) = σ( ϕ( x1 ){ ϕ(λx2 )}∗ ϕ( x1 )).

As ϕ is surjective, Lemma 2.4 tell us that λϕ( x2 ) = ϕ(λx2 ) for all x2 ∈ A and
λ ∈ C, and thus ϕ is homogenous.
Second, let us show that ϕ is additive. By (1.8) and Jacobson lemma, we
have
σ ( ϕ( x )∗ ( ϕ(u))2 ) ∪ {0} = σ ( x ∗ u2 ) ∪ {0} ( x, u ∈ A).
As ϕ is surjective, we see that u2 ∈ F1 (A) if and only if ( ϕ(u))2 ∈ F1 (B). More-
over, replacing x by x1 + x2 , we obtain

σ (( ϕ( x1 + x2 ))∗ ( ϕ(u))2 ) ∪ {0} = σ(( x1 + x2 )∗ u2 ) ∪ {0}

for all x1 , x2 ∈ A and u ∈ A such that u2 ∈ F1 (A). Therefore,

τ (v( ϕ( x1 + x2 ))∗ v) = τ (( ϕ( x1 + x2 ))∗ v2 ) = τ (( ϕ( x1 ) + ϕ( x2 ))∗ v2 )

= τ (v( ϕ( x1 ) + ϕ( x2 ))∗ v)

for all x1 , x2 ∈ A and v ∈ F1 (B). Hence, Lemma 2.5 implies that

ϕ ( x1 + x2 ) = ϕ ( x1 ) + ϕ ( x2 ), ( x1 , x2 ∈ A),
i.e., ϕ is additive. Now, consider the linear map φ( x ) := ϕ(1)( ϕ( x ∗ ))∗ ϕ(1), and
note that, by (1.8), it satisfies σ ( x ) = σ (φ( x )), x ∈ A. As {1} = σ ( ϕ(1) ϕ(1)∗ ϕ(1)),
we see that ϕ(1) is invertible and thus φ is a surjective map. By the main result of
[13], we see that φ is a Jordan isomorphism, and consequently ϕ(1) ϕ(1)∗ ϕ(1) = 1.
It then follow that ϕ(1)∗ ϕ(1) = { ϕ(1)}−1 and ϕ(1) is positive. Therefore, ϕ(1)3 =
1 but, since ϕ(1) is positive, we conclude that ϕ(1) = 1 and φ( x ) is a Jordan
isomorphism.
Finally, let us show that ϕ( x ∗ ) = ϕ( x )∗ for all x ∈ A and φ = ϕ . Let
x1 , x2 ∈ A, and note that, since ϕ is surjective, there is x ∈ A such that ϕ( x ) =
 J ORDAN MAPS AND PSEUDOSPECTRUM IN C ∗ - ALGEBRA 313

ϕ( x2 )∗ . Then
σ ( x1 x2 ∗ x1 ) = σ(φ( x1 x2 ∗ x1 )) = σ (φ( x1 )φ( x2 ∗ )φ( x1 )) = σ ( ϕ( x1 ∗ )∗ ϕ( x2 )∗ ϕ( x1 ∗ )∗ )
= σ( ϕ( x1 ∗ ) ϕ( x2 ) ϕ( x1 ∗ )) = σ( ϕ( x1 ∗ ) ϕ( x )∗ ϕ( x1 ∗ )) = σ( x1 ∗ x ∗ x1 ∗ )
= σ( x1 xx1 ).
By Lemma 2.4, we see that x = x2 ∗ and then ϕ( x2 ∗ ) = ϕ( x ) = ϕ( x2 )∗ . Therefore,
φ( x ) = ϕ( x ∗ )∗ = ϕ( x ) for all x ∈ A and ϕ is Jordan ∗-isomorphism.

7. SOME APPLICATIONS

In this section, we provide several applications of the above main results,

which are interesting in their own right. As an application of Theorem 1.1, we
give a complete description of all maps between A and B that preserve the ε-
pseudospectrum of the sum of element pairs in C ∗ -algebras.
C OROLLARY 7.1. Let ε > 0 and let ϕ be a map from A onto B . Then
Λε ( ϕ( x1 ) + ϕ( x2 )) = Λε ( x1 + x2 ), ( x1 , x2 ∈ A),
if and only if ϕ is a Jordan ∗-isomorphism.
Proof. If the identity holds then Λε ( ϕ(0) + ϕ(0)) = Λε (0) = D (0, ε). By
Lemma 2.2(ii), we see that ϕ(0) = 0. Moreover, we have Λε ( ϕ( x ) + ϕ(− x )) =
Λε (0) = D (0, ε), for all x ∈ A. Again Lemma 2.2(ii) tells us that ϕ(− x ) = − ϕ( x )
for all x ∈ A. Therefore, (1.3) is satisfied and ϕ is a Jordan ∗-isomorphism by
Theorem 1.1.

Clearly, every homomorphism and every anti-homomorphism is a Jordan

homomorphism. Conversely, in [30], Jacobson and Rickart proved that a Jor-
dan homomorphism from an arbitrary ring into a domain is either a homomor-
phism or an antihomomorphism. The same conclusion holds for Jordan homo-
morphisms onto prime rings, as shown by Herstein [25] and Smiley [46]. In fact,
the problem of whether Jordan homomorphisms can be expressed through homo-
morphisms and anti-homomorphisms was considered by several authors; see for
instance [6], [12] and the reference therein. The second corollary is an immediate
consequence of Theorem 1.1 and Herstein’s theorem [25].
C OROLLARY 7.2. If ε > 0 and A is prime, then a map ϕ from A onto B satisfies
(1.3) if and only if ϕ − ϕ(0) is either a ∗-isomorphism or a ∗-anti-isomorphism.
The next corollary is just a reformulation of Corollary 7.1 when one of the
algebras A and B is prime.
C OROLLARY 7.3. If ε > 0 and A is prime, then a map ϕ from A onto B satisfies
Λε ( ϕ( x1 ) + ϕ( x2 )) = Λε ( x1 + x2 ), ( x1 , x2 ∈ A),
314 A. B OURHIM AND J. M ASHREGHI

if and only if ϕ is either a ∗-isomorphism or a ∗-anti-isomorphism.

Now, we turn our attention to some consequences of Theorem 1.2. The
fourth corollary is an immediate consequence of Theorem 1.2 when ϕ1 and ϕ2
are the same.
C OROLLARY 7.4. Let ε > 0 and let ϕ be a map from A onto B . Then

Λε ( ϕ( x1 ) ϕ( x2 )) = Λε ( x1 x2 ), ( x1 , x2 ∈ A),
if and only if ϕ(1) is a central element in B , for which ϕ(1)2 = 1, and ϕ(1) ϕ is a
∗-isomorphism.
Proof. If the identity is satisfied, then Theorem 1.2 tells us that ϕ(1) ϕ and
ϕϕ(1) coincide and are ∗-isomorphism. In particular, one has ϕ(1)2 = 1 and
ϕ(1) ϕ( x ) = ϕ( x ) ϕ(1) for all x ∈ A. This shows that ϕ(1) is a central element in
B . The reverse implication is trivial.

Another consequence of Theorem 1.2 is the following corollary that gives

a complete description of all maps between the C ∗ -algebras A and B preserving
the ε-pseudospectrum of the skew product of element pairs in C ∗ -algebras.
C OROLLARY 7.5. Let ε > 0, and ϕ be a map from A onto B . Then

Λε ( ϕ( x1 ) ϕ( x2 )∗ ) = Λε ( x1 x2∗ ), ( x1 , x2 ∈ A),
if and only if ϕ is a ∗-isomorphism multiplied by a unitary element in B .
For the proof, set ϕ1 ( x ) := ϕ( x ), ϕ2 ( x ) := { ϕ( x ∗ )}∗ , x ∈ A, and apply
Theorem 1.2.
A C ∗ -algebra is primitive if there exists a faithful irreducible representation.
It is fairly easy to see that any primitive C ∗ -algebra must be prime. Conversely,
following a related result of Kaplansky [34], Dixmier [19] showed that every sep-
arable prime C ∗ -algebra is primitive. The question whether, indeed, every prime
C ∗ -algebra is primitive has been answered negatively by Weaver in [48] where
a prime nonseparable C ∗ -algebra that is not primitive is constructed. Since in
primitive algebras every nonzero ideal is essential, the following corollary is an
immediate consequence of Theorem 1.1 and Herstein’s theorem [25].
C OROLLARY 7.6. Assume that ε > 0 and A is primitive with nonzero socle. A
map ϕ from A onto B satisfies (1.6) if and only if ϕ(1) is a central element of B for which
ϕ(1)3 = 1 and ϕ(1)2 ϕ is either a ∗-isomorphism or a ∗-anti-isomorphism.
When we restrict Theorem 1.2 to the case of the algebra of operators on
a complex Hilbert space H, we obtain the following corollary which improves
Theorem 4.1 of [18]. It also includes the complete description of all surjective
maps on L(H) preserving the pseudospectrum of the skew product of operator
pairs.
 J ORDAN MAPS AND PSEUDOSPECTRUM IN C ∗ - ALGEBRA 315

C OROLLARY 7.7. Let ε > 0 and let ϕ1 and ϕ2 be two surjective maps on L(H).
Then they satisfy
Λε ( ϕ1 ( T ) ϕ2 (S)) = Λε ( TS), (S, T ∈ L(H)),
if and only if there exist a unitary operator U ∈ L(H) and an invertible operator A ∈
L(H) such that ϕ1 ( T ) = UTA and ϕ2 ( T ) = A−1 TU ∗ for all T ∈ L(H).
Proof. Checking the “if ” part is straightforward, so we will only deal with
the “only if ” part. So assume that the identity holds, and note that Theorem 1.2
tells us that ϕ2 (1) ϕ1 (1) = 1 and that there is a unitary operator U ∈ L(H) such
that
ϕ1 ( T ) ϕ2 (1) = ϕ1 (1) ϕ2 ( T ) = UTU ∗ , ( T ∈ L(H)).
Set A := U ∗ ϕ2 (1) and note that A−1 = ϕ1 (1)U. Clearly, we see that ϕ1 ( T ) =
UTA and ϕ2 ( T ) = A−1 TU ∗ for all T ∈ L(H).

The last corollary is an immediate consequence of Corollary 7.5 or Corol-

lary 7.7. It characterizes surjective maps on L(H) preserving the pseudospectrum
of the skew product of operator pairs.
C OROLLARY 7.8. Let ε > 0 and let ϕ be a surjective map on L(H). Then it
satisfies
Λε ( ϕ( T ) ϕ(S)∗ ) = Λε ( TS∗ ), (S, T ∈ L(H)),
if and only if there exist two unitary operators U and V in L(H) such that ϕ( T ) = UTV
for all T ∈ L(H).

8. CONCLUDING REMARKS

We end this paper with some comments and remarks.

(i) Assume that ϕ1 and ϕ2 are two maps from A onto B such that (1.3) is
satisfied. Then the proof of Step 1 and Step 2 of Theorem 1.2 tell us (4.1) holds
even when no one of the C ∗ -algebras A and B has an essential socle.
(ii) Similarly, inspecting the proof of Theorem 1.3, one observes that if ϕ is a
map from A onto B satisfying (1.6), then (5.4) holds even when both A and B
have nonessential socle. The same remark applies when ϕ satisfies (1.7).
(iii) Assume that ϕ1 and ϕ2 are two maps on Mn (C) (no one of them is sup-
posed to be surjective) such that
Λε ( ϕ1 ( T ) ϕ2 (S)) = Λε ( TS), (S, T ∈ Mn (C)).
Then Λε ( ϕ1 (1) ϕ2 (1)) = Λε (1) = D (1, ε) and Lemma 2.2(ii) implies that ϕ1 (1) ϕ2 (1)
= 1, and thus ϕ1 (1) = { ϕ2 (1)}−1 . Just as in the proof of Theorem 1.2, one shows
that ϕ1 ( T ) = { ϕ2 ( T −1 )}−1 for all invertible matrices T ∈ Mn (C) and then
σ ( ϕ1 ( T ) ϕ2 (S)) = σ( TS), (S, T ∈ Mn (C)).
316 A. B OURHIM AND J. M ASHREGHI

By Proof of Theorem 2.1 of [16], there are two invertible matrices A and B in
Mn (C) such that either
ϕ1 ( T ) = ATB and ϕ2 ( T ) = B−1 TA−1 , ( T ∈ Mn (C)),
or
ϕ1 ( T ) = AT tr B and ϕ2 ( T ) = B−1 T tr A−1 , ( T ∈ Mn (C)).
Note that, since Λε ( ϕ1 ( T ) ϕ2 (1)) = Λε ( T ) for all T ∈ Mn (C), Lemma 2.2(iii) tells
us that ϕ1 ( T ∗ ) ϕ2 (1) = { ϕ1 ( T ) ϕ2 (1)}∗ for all T ∈ Mn (C). Thus in either case,
A must be a multiple of a unitary operator U ∈ L(H), and thus the above two
representations become
ϕ1 ( T ) = UTB and ϕ2 ( T ) = B−1 TU ∗ , ( T ∈ Mn (C)),
or
ϕ1 ( T ) = UT tr B and ϕ2 ( T ) = B−1 T tr U ∗ , ( T ∈ Mn (C)).
If the last situation occurs, then one has
Λε ((ST )tr ) = Λε (U (ST )tr U ∗ ) = Λε ( ϕ1 ( T ) ϕ2 (S)) = Λε ( TS)
for all S, T ∈ Mn (C). This and Lemma 2.2(ii) tell us that, for any matrices T and
S in Mn (C), we have
TS = 0 ⇔ ST = 0.
This is impossible and the former representation is the only possibility that may
occur. This result improves Theorem 3.3 of [15] and its proof is simpler than the
one presented in [15].
(iv) Just as shown in the previous remark, the surjective condition on ϕ can be
omitted in Corollary 7.8 when H = Cn is a finite-dimensional space.
(v) Theorem 1.3 and Theorem 1.4 can be exemplified in case A = B = L(H),
and the surjective condition on ϕ can be omitted if H = Cn is a finite-dimensional
space.

Acknowledgements. This work was supported by the NSERC Discovery Grant

(Canada).

REFERENCES

[1] J. A LAMINOS , M. B REŠAR , J. E XTREMERA , Š. Š PENKO , A.R. V ILLENA, Determining

elements in C ∗ -algebras through spectral properties, J. Math. Anal. Appl. 405(2013),
214–219.
[2] J. A LAMINOS , J. E XTREMERA , A.R. V ILLENA, Approximately spectrum-preserving
maps, J. Funct. Anal. 261(2011), 233–266.
[3] B. A UPETIT, A Primer on Spectral Theory, Springer-Verlag, New York 1991.
[4] B. A UPETIT, H. DU M OUTON, Spectrum preserving linear mappings in Banach alge-
bras, Studia Math. 109(1994), 91–100.
 J ORDAN MAPS AND PSEUDOSPECTRUM IN C ∗ - ALGEBRA 317

[5] L. B ARIBEAU , T. R ANSFORD, Non-linear spectrum-preserving maps, Bull. London

Math. Soc. 32(2000), 8–14.
[6] W.E. B AXTER , W.S. M ARTINDALE III, Jordan homomorphisms of semiprime rings, J.
Algebra 56(1979), 457–471.
[7] R. B HATIA , P. Š EMRL , A. S OUROUR, Maps on matrices that preserve the spectral
radius distance, Studia Math. 134(1999), 99–110.
[8] A. B OURHIM , M. B URGOS , V.S. S HULMAN, Linear maps preserving the minimum
and reduced minimum moduli, J. Funct. Anal. 258(2010), 50–66.
[9] A. B OURHIM , J. M ASHREGHI, A survey on preservers of spectra and local spectra, in
Invariant Subspaces of the Shift Operator, Contemp. Math., vol. 638, Amer. Math. Soc.,
Providence, RI 2015, pp. 45–98.
[10] A. B OURHIM , J. M ASHREGHI , A. S TEPANYAN, Maps between Banach algebras pre-
serving the spectrum, Arch. Math. 107(2016), 609–621.
[11] G. B RAATVEDT, R. B RITS, Uniqueness and spectral variation in Banach algebras,
Quaest. Math. 36(2013), 155–165.
[12] M. B REŠAR, Jordan homomorphisms revisited, Math. Proc. Cambridge Philos. Soc.
144(2008), 317–328.
[13] M. B REŠAR , A. F OŠNER , P. Š EMRL, A note on invertibility preservers on Banach al-
gebras, Proc. Amer. Math. Soc. 131(2003), 3833–3837.
[14] M. B REŠAR , Š. Š PENKO, Determining elements in Banach algebras through spectral
properties, J. Math. Anal. Appl. 393(2012), 144–150.
[15] J.L. C UI , V. F ORSTALL , C.-K. L I , V. YANNELLO, Properties and preservers of the
pseudospectrum, Linear Algebra Appl. 436(2012), 316–325.
[16] J.T. C HAN , C.K. L I , N.S. S ZE, Mappings preserving spectra of products of matrices,
Proc. Amer. Math. Soc. 135(2007), 977–986.
[17] J.L. C UI , C.K. L I, Maps preserving peripheral spectrum of Jordan products of oper-
ators, Oper. Matrices 6(2012), 129–146.
[18] J.L. C UI , C.-K. L I , Y.-T. P OON, Pseudospectra of special operators and pseudospec-
trum preservers, J. Math. Anal. Appl. 419(2014), 1261–1273.
[19] J. D IXMIER, Sur les C ∗ –algèbres, Bull. Soc. Math. France 88(1960), 95–112.
[20] M. F ORTIER B OURQUE , T. R ANSFORD, Super-identical pseudospectra, J. London Math.
Soc. 79(2009), 511–528.
[21] H.L. G AU , C.K. L I, C ∗ -isomorphisms, Jordan isomorphisms, and numerical range
preserving maps, Proc. Amer. Math. Soc. 135(2007), 2907–2914.
[22] A. G LEASON, A characterization of maximal ideals, J. Analyse Math. 19(1967), 171–
172.
[23] O. H ATORI , T. M IURA , H. TAKAGI, Characterizations of isometric isomorphisms be-
tween uniform algebras via nonlinear range-preserving property, Proc. Amer. Math.
Soc. 134(2006), 2923–2930.
318 A. B OURHIM AND J. M ASHREGHI

[24] O. H ATORI , T. M IURA , H. TAKAGI, Unital and multiplicatively spectrum-preserving

surjections between semisimple commutative Banach algebras are linear and multi-
plicative, J. Math. Anal. Appl. 326(2007), 281–296.
[25] I.N. H ERSTEIN, Jordan homomorphisms, Trans. Amer. Math. Soc. 81(1956), 331–341.
[26] J.C. H OU , Q. D I, Maps preserving numerical ranges of operator products, Proc. Amer.
Math. Soc. 134(2006), 1435–1446.
[27] M. H OSSEINI , F. S ADY, Multiplicatively range-preserving maps between Banach
function algebras, J. Math. Anal. Appl. 357(2009), 314–322.
[28] J.C. H OU , C.K. L I , N.C. W ONG, Jordan isomorphisms and maps preserving spectra
of certain operator products, Studia Math. 184(2008), 31–47.
[29] J.C. H OU , C.K. L I , N.C. W ONG, Maps preserving the spectrum of generalized Jordan
product of operators, Linear Algebra Appl. 432(2010), 1049–1069.
[30] N. J ACOBSON , C. R ICKART, Jordan homomorphisms of rings, Trans. Amer. Math. Soc.
69(1950), 479–502.
[31] A. J IMÉNEZ -VARGAS , A. L UTTMAN , M. V ILLEGAS -VALLECILLOS, Weakly peripher-
ally multiplicative surjections of pointed Lipschitz algebras, Rocky Mountain J. Math.
40(2010), 1903–1922.
[32] R.V. K ADISON, Isometries of operator algebras, Ann. Math. 54(1951), 325–338.
[33] J.P. K AHANE , W. Ż ELAZKO, A characterization of maximal ideals in commutative
Banach algebras, Studia Math. 29(1968), 339–343.
[34] I. K APLANSKY, The structure of certain operator algebras, Trans. Amer. Math. Soc.
70(1951), 219–255.
[35] S. K OWALSKI , Z. S ŁODKOWSKI, A characterization of multiplicative linear function-
als in Banach algebras, Studia Math. 67(1980), 215–223.
[36] A. K RISHNAN , S.H. K ULKARNI, Pseudospectrum of an element of a Banach algebra,
Operators and Matrices 11(2017), 263–287.
[37] G.K. K UMAR , S.H. K ULKARNI, Linear maps preserving pseudospectrum and condi-
tion spectrum, Banach J. Math. Anal. 6(2012), 45–60.
[38] S. M AZUR , S. U LAM, Sur les transformations isométriques d’espaces vectoriels nor-
més, C. R. Acad. Sci. Paris 194(1932), 946–948.
[39] T. M IURA , D. H ONMA, A generalization of peripherally-multiplicative surjections
between standard operator algebras, Cent. Eur. J. Math. 7(2009), 479–486.
[40] L. M OLNÁR, Some characterizations of the automorphisms of B( H ) and C ( X ), Proc.
Amer. Math. Soc. 130(2001), 111–120.
[41] L. M OLNÁR , P. Š EMRL, Transformations of the unitary group on a Hilbert space, J.
Math. Anal. Appl. 388(2012), 1205–1217.
[42] C.K. L I , P. Š EMRL , N.S. S ZE, Maps preserving the nilpotency of products of opera-
tors, Linear Algebra Appl. 424(2007), 222–239.
[43] T. R ANSFORD , S. R AOUAFI, Pseudospectra and holomorphic functions of matrices,
Bull. London Math. Soc. 45(2013), 693–699.
 J ORDAN MAPS AND PSEUDOSPECTRUM IN C ∗ - ALGEBRA 319

[44] B. R USSO , H.A. D YE, A note on unitary operators in C ∗ -algebras, Duke Math. J.
33(1966), 413–416.
[45] P. Š EMRL, Non-linear commutativity preserving maps, Acta Sci. Math. (Szeged)
71(2005), 781–819.
[46] M.F. S MILEY, Jordan homomorphisms onto prime rings, Trans. Amer. Math. Soc.
84(1957), 426–429.
[47] L.N. T REFETHEN , M. E MBREE, Spectra and Pseudospectra. The Behavior of Nonnormal
Matrices and Operators, Princeton Univ. Press, Princeton, NJ 2005.
[48] N. W EAVER, A prime C ∗ -algebra that is not primitive, J. Funct. Anal. 203(2003), 356–
361.

A. BOURHIM, S YRACUSE U NIVERSITY, D EPARTMENT OF M ATHEMATICS ,

215 C ARNEGIE B UILDING , S YRACUSE , NY 13244, U.S.A.
E-mail address: abourhim@syr.edu

J. MASHREGHI, D ÉPARTEMENT DE MATHÉMATIQUES ET DE STATISTIQUE , U NI -

VERSITÉ L AVAL , Q UÉBEC , QC, C ANADA , G1K 0A6
E-mail address: javad.mashreghi@mat.ulaval.ca

Received August 9, 2018; revised September 18, 2018.