Monatshefte für Mathematik

https://doi.org/10.1007/s00605-023-01898-z

Decomposability and local spectral properties of a normal

linear relation

Yosra Barkaoui1 · Maher Mnif1

Received: 18 March 2022 / Accepted: 27 August 2023

© The Author(s), under exclusive licence to Springer-Verlag GmbH Austria, part of Springer Nature 2023

Abstract
Our objective in this paper is to show that, similarly to the case of normal operators,
a normal linear relation on a Hilbert space H satisfies several notions related to the
local spectral theory such as the single valued extension property (SVEP), Bishop and
Dunford properties, and more generally the spectral decomposability. To that end, we
shall start by introducing all those notions for a closed linear relation T on a Banach
space X . Then, we will use an approach allowing us to establish that, as soon as a local
spectral property holds for the orthogonal operator part of T , it will automatically hold
for T . Finally, we will apply that approach on a normal linear relation to reach our
objective.

Keywords Decomposable linear relation · Orthogonal operator part · Local spectral

theory · Componentwise sum

Mathematics Subject Classification 47A06 · 47A10 · 47A11 · 47A25 · 47B15 · 47B40

1 Introduction

Let H be a Hilbert space. Then, a linear relation T on H is called normal if D(T ) =

D(T ∗ ), where T ∗ is the adjoint of T and there is an injective isometry

V : G(T ) → G(T ∗ )
(x, y) → (x, z).

Communicated by Michael Kunzinger.

B Maher Mnif
[email protected]
Yosra Barkaoui
[email protected]

1 Department of Mathematics, Faculty of Sciences of Sfax, B.P. 1171, 3000, Sfax, Tunisia

123
 Y. Barkaoui , M. Mnif

This definition was first given in 1973 by E. A. Coddington [8] in the case of closed
linear relations. Then, it was considered by several mathematicians like A. Sandovici
and Vasilescu.F.H [19] and, in a more important way, by Hassi [12] in 2014. The latter
has successfully established a link between normal operators and multi-valued normal
ones basing on the following decomposition:

 Tmul ,
T = Ts ⊕ (1)

where G(Ts ) := {(x, y) ∈ G(T ) y ∈ D(T ∗ )} and Tmul := {0} × T (0). In fact,
he managed to show that, whenever the relation T verifies the decomposition (1), the
following equivalence holds:

T is normal ⇔ Ts is a densely defined normal linear operator in D(T ∗ ). (2)

In the present paper, we will in our turn rely on the decomposition (1) to study not
only the spectral decomposability of a normal linear relation, but also some of its local
spectral properties such as the Dunford and Bishop properties as well as the single
valued extension property (SVEP). We note that the notion of decomposability was
introduced for the first time in the case of linear relation in [4].
As for the local spectral properties mentioned above, it is worth noticing that they
have already been studied in the framework of bounded linear operators and have been
developed by several mathematicians like Kjeld B. Laursen, Michael M. Neumann [16]
and Pietro Aiena [1] among others.
In order to define all these properties, for defining them in the context of linear
relations, we will first need to present some notions related to the local spectral theory
such as the local spectrum and the local spectral subspaces through the use of the
topology of C∞ . Let us note that these notions have already been extended for linear
relations by M.Mnif and A-A.Ouled-Hmed [17]. However, these authors did not con-
sider the point ∞. Therefore, we will extend their results for the Riemann sphere C∞
and then proceeding to achieve our purpose.
Once the tools of the local spectral theory are all gathered, we then proceed to prove
our objective. Our procedure consists first in proving that any linear relation T written
as follows

 Tmul ,
T = Ts ⊕ (3)

where Ts and Tmul are already defined, is decomposable in the sense of Definition 4.3
whenever its single-valued part Ts is decomposable in the classical sense of bounded
operators. Then, we will establish that, as soon as Ts verifies the Bishop and Dunford
properties as well as the SVEP in the classical sense of bounded linear operators, T
will automatically verify these properties in their new sense of linear relations. Such
procedure yields in creating a strong link between the local spectral theory of operators
and that of linear relations.
This will allow us to conclude that the decomposability as well as the local spectral
properties described above constitute necessary conditions for a relation to be normal
exactly as in the bounded linear operators case.

123
Decomposability and local...

The paper is organized as follows. In Sect. 2, we recall the most important definitions
from the theory of linear relations that can be found in [5] and [10].
In Sect. 3, we adopt the definition of a normal linear relation T from [12] and we
develop some properties related to its single valued part Ts . This will be very useful
in the study of the remaining sections.
Section 4 is devoted to the spectral decomposability of a linear relation T . The main
result of this section is Theorem 4.1, where it is proved that T is spectral decomposable
in the sense of Definition 4.3 whenever its single valued part is spectral decomposable
in the classical sense of bounded linear operators
Some local spectral properties of linear relations are the objective of Sect. 5. We
start this section by introducing the single valued extension property for linear relations
in the Riemann sphere C∞ (see Definition 5.3). Then, we characterize it by means
of extended local spectral subspaces in Corollary 5.1. This will help us to deduce, in
Theorem 5.2, the strong link between the SVEP of bounded linear operators and that
of the continuous linear relations.
To conclude with, we state that every normal linear relation has the SVEP, exactly
as in the case of bounded linear operators. The same procedure is applied to the
Bishop and Dunford properties that are introduced in Definition 5.4 and Definition
5.5, respectively. The most important results regarding these two properties are given
in Theorem 5.3 and Theorem 5.4, where we find the strong connection between the
single-valued linear operators and the multi-valued ones. Another main result is given
in Corollary 5.3 and Corollary 5.4 and consists in establishing that any normal relation
has the Bishop and Dunford properties.

2 Preliminaries

Throughout the paper, X and Y (respectively H ) will denote two Banach spaces
(respectively a Hilbert space) over the field K = R or C, unless otherwise stated.
Taking R.Cross [10] as a reference, a mapping T : X → 2Y \∅ from a subspace
D(T ) of X into the collection of non empty subsets of Y is called a linear relation
if a1 T x + a2 T y = T (a1 x + a2 y) for all x, y ∈ D(T ) and a1 , a2 ∈ C\{0}. This
last is entirely determined by its graph defined as G(T ) = {(x, y) ∈ X × Y x ∈
D(T ) and y ∈ T x}. We denote by L R(X , Y ) the class of all linear relations from X
to Y . The inverse of T is given by G(T −1 ) = {(x, y) ∈ Y × X (y, x) ∈ G(T )}.
The Kernel of T is the subspace K er (T ) := T −1 (0). We say that T is injective
whenever K er (T ) = {0}. The range of a subset X 0 ⊆ X and the range of T is denoted
by T (X 0 ) := x∈D(T )∩X 0 T x and I m(T ) := T (X ), respectively. If I m(T ) = Y ,
then T is said to be surjective. Let us note remark that T is a linear operator if and
only if T (0) = {0} and we write T ∈ L O(X , Y ).
We consider the following quantities related to T :

T = QT T and T x0 := Q T T x0 ,

where x0 ∈ D(T ) and Q T : Y −→ Y /T (0) is the natural quotient map. If T < ∞,

then T said to be continuous.

123
 Y. Barkaoui , M. Mnif

The closure of T is the closure of its graph and denoted by T . We say that T is
closed (respectively open) if T = T or, equivalently, G(T ) = G(T ) (respectively,
T −1 is continuous). The set of all closed linear relations and that of all closed and
continuous linear relations is denoted by C R(X , Y ) and C Rc (X , Y ), respectively. If
X = Y , then C R(X , Y ) = C R(X ) and C Rc (X , Y ) = C Rc (X ).
The sum and the componentwise sum of two linear relations T , S ∈ L R(X , Y ) are
the linear relations respectively given by
 
G(T + S) := (x, u + v) | (x, u) ∈ G(T ), (x, v) ∈ G(S) ,
 
 S) := (x + u, y + v) | (x, y) ∈ G(T ), (u, v) ∈ G(S) .
G(T +

If T and S are orthogonal in X × Y , then their componentwise sum is denoted by

T⊕ S.
The set of closed and that of all bounded linear operators A : X −→ X are denoted
by C O(X ) and End(X ), respectively. If T −1 ∈ End(X ) then T is said to be invertible.
We note that T is invertible if and only if T is both surjective and injective.
Let T ∈ C R(X ). The resolvent set of T is defined as ρ(T ) := {μ ∈ C T −
μI is invertible}. The resolvent function of T is denoted by R(., T ) and defined by
R(μ, T ) := (T − μI )−1 for all μ ∈ ρ(T ). The set C \ ρ(T ) is called the spectrum of
T and denoted by σ (T ).
It may happen that the spectrum of a relation T is void, for example when T = B −1
and B is a quasinilpotent operator. To keep clear of such complications, we shall bring
back to work with the extended spectrum of T which is a subset of C∞ denoted by

σ (T ) and defined by A.Baskakov [5] as σ (T ) wherever the following conditions are
satisfied:
(1) T (0) = {0};
(2) The resolvent of T can be analytically continued to ∞ and lim|μ|→+∞ R(μ, T ) =
0.
Otherwise,  (T ) := C∞ \ 
σ (T ) = σ (T ) ∪ {∞}. The set ρ σ (T ) is called the extended
resolvent set of T .

3 Normal linear relations

In this section, we collect some results related to a normal relation T . As already said,
we will be focusing on the decomposition of T into a sum of a linear operator Ts and
a multi-valued linear part Tmul , so we shall well study the properties of Ts in order to
study those of T in the remaining parts. Let’s start by recalling the adjoint of T in the
following definition.
Definition 3.1 [8] The adjoint T ∗ of a relation T ∈ L R(H ) is given by

G(T ∗ ) = {(h, k) ∈ H × H ; < h, g >=< k, f >, for all ( f , g) ∈ G(T )}.

If T ∗ = T then T is said to be self-adjoint.

123
Decomposability and local...

Definition 3.2 [13] Let T ∈ L R(H ) and let P be the orthogonal projection from H
onto D(T ∗ ) with kernel T ∗∗ (0). The orthogonal operator part of T is defined by

G(Ts ) := {(x, y) ∈ G(T ) y ∈ D(T ∗ )}

and the regular part of T is the relation Tr eg := P T .

In the remaining work, we shall denote by Tmul and PT the relation {0} × T (0) and
the orthogonal projection onto D(T ∗ ) with kernel T ∗∗ (0), respectively.

Proposition 3.1 Let T ∈ C Rc (H ) be such that D(T ) = D(T ∗ ). Then, the following
statements hold:
(1) T ∗ (0) = T (0) = T ∗∗ (0);
(2) D(T ∗ ) = D(T ). In this case, G(Ts ) := {(x, y) ∈ G(T ) y ∈ D(T )};
(3) Ts = Tr eg := PT T and D(Ts ) = D(T );
(4) T = Ts + Tmul ;
(5) T x = Ts x + T (0) for all x ∈ D(T );
(6) (T − μI )s = Ts − μI for all μ ∈ C.

Proof (1) Since T is closed, it follows that T ∗∗ (0) = T (0). On the other hand, we
have D(T ) = D(T ∗ ), so D(T )⊥ = D(T ∗ )⊥ . Hence, T ∗ (0) = T ∗∗ (0), by [10,
Proposition III.1.4]. Consequently, T ∗ (0) = T (0) = T ∗∗ (0).
(2) Since T ∈ C Rc (H ), we obtain from the closed graph theorem [10, Theorem
III.4.2] that D(T ) is closed, and therefore D(T ∗ ) = D(T ) = D(T ). Consequently,
G(Ts ) := {(x, y) ∈ G(T ) y ∈ D(T )}.
(3) We observe from (1) that T x = PT T x + (I − PT )T x ⊆ PT T x + K er (PT ) =
PT T x + T (0) for all x ∈ D(T ). Hence,

T x = PT T x + T (0) for all x ∈ D(T ).

On the other hand, it follows from the second assertion (2) that I m(PT ) = D(T ),
which leads to the following equivalences:

(x, y) ∈ G(Ts ) ⇔ (x, y) ∈ G(T ) and y ∈ D(T )

⇔ x ∈ D(T ), y ∈ T x and y ∈ D(T )
⇔ x ∈ D(T ), y ∈ PT T x + T (0) and y = PT y
⇔ x ∈ D(T ) and y = PT y = PT T x
⇔ (x, y) = (x, PT T x) and x ∈ D(T )
⇔ (x, y) ∈ G(Tr eg ).

Consequently, Ts = Tr eg . By [13, Theorem 3.10], we then conclude that D(T ) =

D(Ts ).
(4) Since D(T ) = D(Ts ), we infer from [13, Theorem 3.10] that T = Ts +  Tmul .
(5) This immediately follows from the fact that D(T ) = D(Ts ).

123
 Y. Barkaoui , M. Mnif

(6) Let x, y ∈ H and μ ∈ C. Then, we infer from the second assertion (2) the
following equivalences

(x, y) ∈ G((T − μI )s ) ⇔ (x, y) ∈ G(T − μI ) and y ∈ D(T − μI ) = D(T )

⇔ (x, y + μx) ∈ G(T ) and y ∈ D(T )
⇔ (x, y + μx) ∈ G(T ) and y + μx ∈ D(T )
⇔ (x, y + μx) ∈ G(Ts )
⇔ (x, y) ∈ G(Ts − μI ).

Consequently, (T − μI )s = Ts − μI .



Definition 3.3 [8] A relation T ∈ L R(H ) is said to be normal if D(T ) = D(T ∗ ) and
there is an injective isometry V : G(T ) → G(T ∗ ) defined by

V (x, y) := (x, z) ∀(x, y) ∈ G(T ), (x, z) ∈ G(T ∗ ).

It is clear that any self-adjoint linear relation is normal. However, the converse is
not true. This may be illustrated in the following example:

Example 3.1 Let us consider the relation T := λI + T (0) ∈ L R(H ) with the property
that λ ∈ C \ R, T (0) is closed and D(T ) = T (0)⊥ . Then, T is a normal linear relation
verifying T ∗ = T .
Indeed, we infer from [10, Proposition III.1.5] that T ∗ = (λI + (T − T ))∗ =
(λI )∗ + (T − T )∗ = λI + T ∗ − T ∗ = λI + T ∗ (0). But, by [10, Proposition III.1.4],
T ∗ (0) = D(T )⊥ = (T (0)⊥ )⊥ = T (0) = T (0), and hence T ∗ = λI + T (0) = T .
On the other hand, we observe from [3, Lemma 1] that T is closed, so, combining
the fact that D(T ) is closed with [10, Theorem III.4.5], we deduce that D(T ∗ ) is closed.
This implies, by [10, Proposition III.1.4], that D(T ∗ ) = D(T ∗ ) = T (0)⊥ = D(T ).
Now, let us consider the application

V1 : G(T ) → G(T ∗ )
λ
(x, y = λx + α) → (x, z = λx + α),
λ

α
where α ∈ T (0). We then have y = λx + α = |λ| x + and z = |λ| x +
λ
λ α = y . Consequently, V1 is an injective isometry and T is a normal linear relation.
1

A characterisation of normal relations involving normal operators is given in the

following lemma:

Lemma 3.1 [12, Proposition 3.1.] Let T ∈ L R(H ). Then T is normal if and only if
 Tmul .
T is closed, Ts is a densely defined normal operator in D(T ∗ ) and T = Ts ⊕

123
Decomposability and local...

4 Decomposability of a normal linear relation

In this section, we are going to study the notion of extended spectral decomposability
for closed and continuous linear relations. In particular, we will show that closed
continuous normal linear relations are indeed extended spectral decomposable. Before
doing so, we shall start by recalling this notion as well as some of its properties in the
context of bounded linear operators.
Definition 4.1 [15] Let T0 ∈ End(X ). Then,
(1) a T0 -invariant closed linear subspace X 0 ⊆ X is called a spectral maximal space
for T0 if it contains every T0 -invariant linear subspace Y such that σ (T0 |Y ) ⊆
σ (T0 |X 0 ).
(2) T0 ∈ End(X ) is said to be decomposable if for every open covering {U1 , U2 }
of σ (T0 ), there exist X 1 and X 2 two spectral maximal spaces for T0 such that
X = X 1 + X 2 and σ (T0 |X i ) ⊆ Ui or, equivalently, σ (T0 |X i ) ⊆ Ui for i = 1, 2.
Lemma 4.1 Let T0 ∈ End(H ). Then, the following properties hold:
(i) If T0 is normal then T0 is decomposable.
(ii) If X 0 is a spectral maximal subspace for T0 then R(λ, T0 )X 0 ⊆ X 0 for all
λ ∈ ρ(T ).
Proof The proof of (i) easily follows from [18]. To show the assertion (ii), let X 0 be
a spectral maximal subspace for T0 . Then, we infer from [9, Proposition 3.2] that X 0
is invariant under all operators commuting with T0 . As R(λ, T0 )T0 = T0 R(λ, T0 ) for
all λ ∈ ρ(T0 ), we conclude the desired property. 

Moving to the case of linear relations and in order to introduce the extended spectral
decomposability, we take inspiration from Lemma 4.1 to deal with subspaces with
fewer conditions than maximal spaces. Note that those subspaces have been already
studied by A. G. Baskakov and A. S. Zagorskii [7].
Definition 4.2 [7, Definition 2.6.] Let T ∈ C R(X ) be such that ρ(T ) = ∅ and let M
be a closed linear subspace of X . Then, M is said to be T -strongly invariant subspace
if R(λ, T )M ⊆ M for all λ ∈ ρ(T ). The strong restriction of the relation T to a T -
strongly invariant subspace M ⊆ X is the relation T |s M ∈ C R(M) whose resolvent
is R(., T |s M) : ρ(T ) −→ End(M) defined by R(λ, T |s M) := R(λ, T )|M.
Proposition 4.1 Consider T ∈ C R(X ) for which ρ(T ) = ∅ and let Y ⊆ X be a
T -strongly invariant subspace. Then, for all x ∈ D(T |s Y ), we have

T |s Y x ⊆ T x

with equality if T (0) ⊆ Y .

Proof We first observe that, by Remark 2.7 of [6], ρ(T ) ⊆ ρ(T |s Y ). Let x ∈ D(T |s Y ),
y ∈ T |s Y x and let λ ∈ ρ(T ) be arbitrarily given. Then, y − λx ∈ (T |s Y − λI )x, and
therefore

R(λ, T )(y − λx) = R(λ, T )(T |s Y − λI )x = R(λ, T |s Y )(T |s Y − λI )x = x. (4)

123
 Y. Barkaoui , M. Mnif

Applying T − λI on (4), we obtain that y ∈ T x. Hence, T |s Y x ⊆ T x.

Suppose now that T (0) ⊆ Y . We claim that T (0) = T |s Y (0). Indeed, it follows
from the above that T |s Y (0) ⊆ T (0). To prove the other inclusion, consider an arbi-
trary y ∈ T (0) ⊆ Y . Then, for all λ ∈ ρ(T ), we have y ∈ (T − λI )(0), and hence
R(λ, T )y = 0. Since y ∈ Y , it follows that R(λ, T )y = R(λ, T |s Y )y = 0 for
all λ ∈ ρ(T ). Consequently, y ∈ (T |s Y − λI )(0) = T |s Y (0) for all λ ∈ ρ(T ).
We conclude that T (0) ⊆ T |s Y (0), as claimed. Since T |s Y x ⊆ T x, it follows that
T x = T |s Y x + T (0) = T |s Y x + Ts |Y (0) = T |s Y x. 


Let us now present the definition of decomposable linear relations which were
introduced for the first time in our article [4] by means of open subsets of the Riemann
sphere C∞ := C ∪ {∞}.

Definition 4.3 [4, Definition 3.1] Let T ∈ C Rc (X ) be a linear relation such that
ρ(T ) = ∅. We say that T is extended spectral decomposable if every open cover
{U , V } of the extended complex plane by a bounded open subset U of C and an open
subset V of C∞ effects a splitting of the extended spectrum  σ (T ) and of the space X ,
in the sense that there exist T -strongly invariant closed linear subspaces Y and Z of
X with the following properties:
σ (T |s Y ) = σ (T |s Y ) ⊆ U , 
i/  σ (T |s Z ) ⊆ V ;
ii/ T (Z ) ⊆ Z and P(Z  ) ⊆ Z for some bounded linear projection P
 : X −→ X
verifying R( P) = D(T );
iii/ X = Y + Z .

Next, we will create a very strong link between the extended spectral decomposable
linear relations and the decomposable operators.

Theorem 4.1 Let T ∈ C Rc (H ) be such that ρ(T ) = ∅ and D(T ) = D(T ∗ ). If Ts

is a spectral decomposable operator in End(D(T )), then T is an extended spectral
decomposable linear relation.

Proof As D(T ) = D(T ∗ ), we deduce from assertion (2) of Proposition 3.1 that
G(Ts ) := {(x, y) ∈ G(T ) y ∈ D(T )}, D(T ) = D(T ) = D(Ts ) and

T x = Ts x + T (0) for all x ∈ D(T ). (5)

Now, let {U , V } be an open cover of C∞ such that U is an open bounded subset of

C and V is an open subset of C∞ . Then, {U , (V ∩ C)} is an open cover of C. Because
Ts is a decomposable operator in D(T ), there exist two spectral maximal spaces Y
and Z for Ts with the following properties:
(i) D(T ) = Y + Z ;
(ii) σ (Ts |Y ) ⊆ U ;
(iii) σ (Ts |Z ) ⊆ V ∩ C.
Hence Y and Z are Ts -invariant linear subspaces and, by combining the fact that
Ts ∈ End(D(T )) with assertion (ii) of Lemma 4.1, we deduce that both Y and Z are
Ts -strongly invariant linear subspaces.

123
Decomposability and local...

On the other hand, [10, Proposition III.1.4] and assertion (1) of Proposition 3.1
imply that T (0) = T ∗ (0) = D(T )⊥ , so H = D(T )⊕T (0). Thus, H = Y + Z +T (0),
by (i). Now, let Y1 := Y and Z 1 := Z + T (0).
Claim 1. ρ(T ) ⊆ ρ(Ts ).
To prove the claim, consider λ ∈ ρ(T ) and let x ∈ K er (Ts − λI ). Then,
(Ts −λI )x = 0, and therefore (Ts −λI )x + T (0) = T (0). By (5), we obtain
that (T − λI )x = T (0), which means that x ∈ K er (T − λI ) = {0}. Thus,
Ts − λI is injective.
To see the surjectivity of Ts , let w ∈ D(T ) ⊆ H = I m(T − λI ). Then,
there exists x ∈ D(T ) such that w ∈ (T − λI )x = (Ts − λI )x + T (0). This
implies that w − (Ts − λI )x ∈ T (0) ∩ D(T ) = {0}, so w = (Ts − λI )x ∈
I m(Ts − λI ). Consequently, D(T ) = I m(Ts − λI ), as desired. We then
conclude that λ ∈ ρ(Ts ), and hence ρ(T ) ⊆ ρ(Ts ).
Claim 2. Y1 and Z 1 are two T -strongly invariant linear subspaces which verify H =
Y1 + Z 1 .
We shall first show that Z 1 is closed. To see this, let (xn )n∈N ⊆ Z 1 ⊆ D(T )
be such that xn −→ x ∈ H . As PT is continuous and Z and T (0) are both
n→+∞
closed, we conclude that xn = (I −PT )xn +PT xn −→ (I −PT )x+PT x =
n→+∞
x ∈ Z + T (0) = Z 1 .
Now, for an arbitrary λ ∈ ρ(T ), let y ∈ Y1 ⊆ D(T ) and put x = R(λ, T )y.
Then, the equation (5) entails that y ∈ (T − λI )x = (Ts − λI )x + T (0), and
therefore y − (Ts − λI )x ∈ D(T ) ∩ T (0) = {0}. Hence, y = (Ts − λI )x.
By Claim 1, ρ(T ) ⊆ ρ(Ts ) and R(λ, Ts )y = R(λ, Ts )(Ts − λI )x = x.
Moreover, since Y1 is a Ts -strongly invariant linear subspace, it follows that
x = R(λ, Ts )y ∈ Y1 . This means that R(λ, T )Y1 ⊆ Y1 . In the same way,
we obtain that R(λ, T )Z ⊆ Z for all λ ∈ ρ(T ). Hence, R(λ, T )Z 1 =
R(λ, T )(Z + T (0)) = R(λ, T )Z ⊆ Z ⊆ Z 1 for all λ ∈ ρ(T ). Because Y1
and Z 1 are closed, we then conclude that both Y1 and Z 1 are two T -strongly
invariant linear subspaces with the property that H = Y1 + Z 1 .
σ (T |s Y1 ) = σ (T |s Y1 ) ⊆ U .
Claim 3. 
We first infer from (5) and Proposition 4.1 that

T |s Y1 x − Ts |Y1 x ∈ T (0) ∩ D(T ) = {0} for all x ∈ Y1 . (6)

On the other hand, we have D(T |s Y1 ) = Y1 . In fact, D(T |s Y1 ) ⊆ D(T ) ∩

Y1 = Y1 . To see the other inclusion, let x ∈ Y1 . Then, for all λ ∈ ρ(T ), we
have

x = R(λ, T )(T − λI )x = R(λ, T )(Ts x − λx + T (0))

= R(λ, T )(Ts − λI )x.

But (Ts − λI )x ∈ (Ts − λI )Y1 ⊆ Y1 , so R(λ, T )(Ts − λI )x =

R(λ, T |s Y1 )(Ts − λI )x ∈ I m (R(λ, T |s Y1 )) = D(T |s Y1 ). Hence, Y1 ⊆
D(T |s Y1 ), as desired. This means, by (6), that T |s Y1 = Ts |Y1 ∈

123
 Y. Barkaoui , M. Mnif

End(Y1 ). Using [5, Lemma 2.2] and assertion (ii), we finally conclude
σ (T |s Y1 ) = 
that  σ (Ts |Y1 ) = σ (Ts |Y1 ) ⊆ U .
σ (T |s Z 1 ) ⊆ V .
Claim 4. 
Since Y1 and Z play a symmetrical role, using the same argument of the
previous claim, we deduce from assertion (iii) that σ (T |s Z ) ⊆ V ∩ C.
Hence, it suffices to prove that σ (T |s Z 1 ) ⊆ σ (T |s Z ).
To this end, let λ ∈ ρ(T |s Z ) and let x ∈ K er (T |s Z 1 − λI ). Then, (T |s Z 1 −
λI )x = T |s Z 1 (0). Since T (0) ⊆ Z 1 and Z 1 is a T -strongly invariant linear
subspace, it follows from Proposition 4.1 that

(T − λI )x = (T |s Z 1 − λI )x = T |s Z 1 (0) = T (0). (7)

On the other hand, as above, we obtain that T |s Z ∈ End(Z ), so D(T |s Z ) =

Z . Furthermore, since Z ⊆ D(T ), it follows that D(T ) ∩ (Z + T (0)) ⊆
(D(T ) ∩ Z ) + (D(T ) ∩ T (0)) = Z . Hence,

D(T |s Z 1 ) ⊆ D(T ) ∩ Z 1 = D(T ) ∩ (Z + T (0)) ⊆ Z = D(T |s Z ),

and therefore x ∈ D(T |s Z ). A combination of (7) and Proposition 4.1

implies that T (0) = (T |s Z − λI )x + T (0), so

(T |s Z − λI )x ∈ Z ∩ T (0) ⊆ D(T ) ∩ T (0) = {0}.

Consequently, x ∈ K er (T |s Z − λI ) = {0}, which means that T |s Z 1 − λI

is injective.
To prove the surjectivity of T |s Z 1 − λI , let z 1 ∈ Z 1 . Then, there exist
(z, t) ∈ Z × T (0) such that z 1 = z + t. But Z = I m(T |s Z − λI ), so there
is z  ∈ D(T |s Z ) for which z = (T |s Z − λI )z  . This entails that

z 1 = (T |s Z − λI )z  + t ∈ (T − λI )z  + T (0) = (T − λI )z  .

As T (0) ⊆ Z 1 , we conclude from Proposition 4.1 that z 1 ∈ (T |s Z 1 −λI )z  ∈

I m(T |s Z 1 − λI ). Hence, λ ∈ ρ(T |s Z 1 ), as desired. Finally, we have


σ (T |s Z 1 ) ⊆ σ (T |s Z ) ∪ {∞} ⊆ (V ∩ C) ∪ {∞} = V . (III)

Claim 5. PT (Z 1 ) ⊆ Z 1 and T (Z 1 ) ⊆ Z 1 .
We observe that PT (Z ) = Z since Z ⊆ D(T ), and therefore PT (Z 1 ) =
PT (Z + T (0)) ⊆ Z + PT (T (0)) = Z ⊆ Z 1 . To establish the remaining
inclusion, let x ∈ T (Z 1 ). Then, there exists z 1 ∈ Z 1 ∩ D(T ) ⊆ Z such that
x ∈ T z 1 . Using (5), we deduce that

x ∈ T z 1 = Ts z 1 + T (0) ⊆ Ts (Z ) + T (0) ⊆ Z + T (0) = Z 1 .

Consequently, T (Z 1 ) ⊆ Z 1 , as desired.

123
Decomposability and local...

Combining Claims (2)−(5), we conclude that T is an extended spectral decomposable

linear relation. 


We can now deduce the most important result of this section.

Corollary 4.1 If T ∈ C Rc (H ) is normal and satisfies ρ(T ) = ∅, then T is an extended

spectral decomposable linear relation.

Proof Since T is normal, it follows from Lemma 3.1 that Ts is a normal linear operator.
Moreover, as D(T ) = D(T ∗ ), we infer from assertions (2) and (3) of Proposition 3.1
that Ts ∈ End(D(T )). Consequently, Ts is a bounded normal linear operator in D(T ),
and hence decomposable, by Lemma 4.1. Using Theorem 4.1, we conclude that T is
an extended spectral decomposable linear relation. 


5 Some local spectral properties for normal linear relations

It’s well known that for a relation T ∈ C R(X ) with a non-empty resolvent set and for
an element x ∈ X , the function ϕ1 : ρ(T ) → X given by ϕ1 (μ) := R(μ, T )x for all
μ ∈ ρ(T ) is the only analytic solution on ρ(T ) of the equation

x ∈ (T − λI )ϕ(λ). (8)

However, one can find analytic solutions of the equation (8) on open subsets of C
others than ρ(T ), and here comes the notion of the local resolvent set. Note that the
latter was first introduced in [17, Definition 3.1].
In this section, we shall extend this notion to the extended complex plane C∞ and
also deal with different local spectral properties of continuous and closed normal linear
relations. Before doing so, we start by recalling some complex analysis framework,
which can be found in R. A. Holmgren [14], R. M. Timoney [21] and O. Forster [11],
for the better understanding of the topology of C∞ .
The typical distance used on C∞ is called the chordal distance or chordal metric,
denoted by d∞ and defined by

2|λ1 − λ2 |
d∞ (λ1 , λ2 ) :=  for all λ1 , λ2 ∈ C;
(1 + |λ1 |2 )(1 + |λ2 |2 )
2
d∞ (λ, ∞) =  for all λ ∈ C.
(1 + |λ|2 )

The neighborhoods in C∞ correspond to the standard definitions of neighborhoods

in C and add that neighborhoods of ∞ are defined by U∞ := {z ∈ C |z| > ε } ∪ {∞},
where ε > 0.
The open subsets of C∞ are the usual open subsets of C and those of the form
C\K ∪ {∞}, where K is a compact subset of C.
The space of all analytic functions from an open subset  of C into a complex
topological vector space Z is denoted by H (, Z ). By referring to [16], whenever X

123
 Y. Barkaoui , M. Mnif

is a Banach space, H (, X ) becomes a Frechét space with respect to pointwise vector
space operations and the topology of locally uniform convergence.
For an open set ∞ ⊆ (C∞ , d∞ ), the space of all analytic functions f :∞ → Z
is denoted by H  (∞ , Z ) and the space P(∞ , Z ) is defined as P(∞ , Z ) := { f ∈
H  (∞ , Z ) | f (∞) = 0 if ∞ ∈∞ }.
Topologized by means of uniform convergence on all compact subsets of the open
set ∞ , we note that, by [16], the space H  (∞ , X ) is a Frechét space when X is a
Banach space.
Definition 5.1 Let T ∈ C R(X ) and x0 ∈ X . The local resolvent set (respectively, the
extended local resolvent set) of T at x0 is the set ρT (x0 ) (respectively, ρ
 T (x 0 )) of all
λ0 ∈ C (respectively, λ0 ∈ C∞ ) for which there exist an open neighborhood Uλ0 of λ0
in C (respectively, C∞ ) and an analytic function f x0 ,λ0 : Uλ0 → D(T ) (respectively,
a function f x0 ,λ0 ∈ P(Uλ0 , D(T ))) such that

x0 ∈ (T − μ) f x0 ,λ0 (μ) for all μ ∈ Uλ0 (respectively, Uλ0 ∩ C).

The set C \ ρT (x0 ) (respectively, C∞ \

ρT (x0 )) is called the local spectrum (respec-
tively, the extended local spectrum) of T at the point x0 and denoted by σT (x0 )
(respectively, σ
 T (x 0 )).

There exists a relation between the local spectrum of a linear relation T and the
extended one. This is given in the following remark.
Remark 5.1 Let T ∈ C R(X ) and x ∈ X . Then,

σT (x), if ∞ ∈
/σT (x)
σ
 T (x) =
σT (x) ∪ {∞} otherwise.

We collect some properties of the extended local spectrum of a closed linear relation
in the following proposition:
Proposition 5.1 Let T ∈ C R(X ), U be an open subset of C∞ and x ∈ X . Then, the
following properties hold:
i) σ
 T (x) = ∅ for all x ∈ T (0);
ii) If there exists f ∈ P(U , D(T )) verifying

x ∈ (T − λI ) f (λ) for all λ ∈ U ∩ C, (9)

then U ⊆ ρ
 T ( f (λ)) and σ
 T ( f (λ)) ⊆ σ
 T (x) for all λ ∈ U ∩ C.

Proof i) The proof follows immediately from Definition 5.1.

ii) Let λ ∈ U ∩ C. Define the function g : U → D(T ) such that
⎧
⎪ f (λ) − f (μ)
⎪
⎨ if μ ∈ U ∩ C \ {λ},
λ−μ
g(μ) := f  (λ) if μ = λ,
⎪
⎪
⎩
0 if μ = ∞.

123
Decomposability and local...

Then, g ∈ P(U , D(T )) and, for all μ ∈ U ∩ C\{λ}, we have

(T − μI )g(μ) = (T − λI )g(μ) + (λ − μ)g(μ)

1 
= (T − λI ) f (λ) − (T − λI ) f (μ)
λ−μ

+ (λ − μ) f (λ) − (λ − μ) f (μ)
= f (λ) + T (0).

Hence, f (λ) ∈ (T − μI )g(μ) for all μ ∈ U ∩ C\{λ}. Moreover, it follows from

(9) that, for all μ ∈ U ∩ C, we have x := Q T x = Q T T ( f (μ)) − μQ T ( f (μ)).
Hence, for all μ ∈ U ∩ C, we obtain

    
0 = Q T T ( f (μ)) − Q T ( f (μ)) − μQ T ( f (μ)) = Q T (T − μI ) f (μ) − f (μ) ,

  
in particular Q T (T − λI )( f (λ)) − f (λ) = 0. This implies that f (λ) ∈ (T −
λI )g(λ), and therefore

f (λ) ∈ (T − μI )g(μ) for all μ ∈ U ∩ C.

This shows that μ ∈ ρ  T ( f (λ)) for all μ ∈ U . Consequently, U ⊆ ρ T ( f (λ)).

To prove the remaining inclusion, let v ∈ ρ  T (x). If v ∈ U then, as mentioned
earlier, v ∈ ρ
 T ( f (λ)). If v ∈/ U then there exist an open neighborhood V ⊆ C∞
of v and a function h 1 ∈ P(V , D(T )) such that x ∈ (T − μI )h 1 (μ) for all
μ ∈ V ∩ C. Now, let

g1 : V → ⎧ D(T )
⎨ f (λ) − h 1 (μ)
if μ = ∞
μ → λ−μ
⎩ 0 otherwise.

Then, g1 ∈ P(V , D(T )) and it is easy to see that f (λ) ∈ (T − μI )g1 (μ) for
all μ ∈ V ∩ C. Hence, v ∈ ρ
 T ( f (λ)). This shows that ρ
 T (x) ⊆ ρ
 T ( f (λ)), so
σ
 T ( f (λ)) ⊆ σ
 T (x).



Besides the notion of the extended local spectrum, it is important to introduce that
of the extended local spectral subspaces with the aim of studying the local spectral
properties of a normal linear relation. We recall that this kind of subspaces has been
well studied for linear relations while working with the topology of C (see in [17] for
more details), whereas in our work, we define them using the subsets of C∞ .

Definition 5.2 Let T ∈ C R(X ) and let F0 ⊆ C and F ⊆ C∞ . The local spectral
subspace (respectively, the extended local spectral subspace) of T at F0 (respectively,
F) is defined by X T (F0 ) := {x ∈ X σT (x) ⊆ F0 } (respectively, X T (F) := {x ∈
X σT (x) ⊆ F}).

123
 Y. Barkaoui , M. Mnif

There exists a relation between the local spectral subspace of a linear relation T
and the extended one. This can be seen in the following remark.
Remark 5.2 Let T ∈ C R(X ) and F ⊆ C. Then,

T ({∞} ∪ F) = X T (F).
X

5.1 SVEP of a normal linear relation

Inspired by the work of P. Aiena [1], we extend the definition of the SVEP at a point
λ0 ∈ C of a bounded linear operator to that at a point λ0 ∈ C∞ of a closed linear
relation.
Definition 5.3 Let T ∈ C R(X ). The relation T is said to have the extended relatively
single-valued extension property, abbreviated ER-SVEP, at λ0 ∈ C∞ if for every open
connected neighborhood Uλ0 ⊆ C∞ of λ0 and for all function f ∈ P(Uλ0 , D(T ))
which satisfies

0 ∈ (T − μI ) f (μ) for all μ ∈ Uλ0 ∩ C, (10)

we have f (μ) ∈ T (0) ∩ K er (T ) for all μ ∈ Uλ0 . The relation T is said to have the
ER-SVEP if T has the ER-SVEP at every λ ∈ C∞ .
A necessary and sufficient condition for a linear relation to have the property ER-
SVEP at an arbitrary λ0 ∈ C is obtained in the following theorem:
Theorem 5.1 Let T ∈ C Rc (X ) be a relation such that X = D(T ) ⊕ T (0). Then, the
following statements are equivalent:
i) T has the ER-SVEP at λ0 ∈ C;
T (∅) = {0}.
ii) K er (T − λ0 I ) ∩ X
 such that R( P)
Proof We first observe that there exists a bounded linear projection P  =
 
D(T ) and K er ( P) = T (0). It is clear that P T is a continuous selection of T .
i) ⇒ ii) Suppose that T has the ER-SVEP at λ0 ∈ C, and consider x ∈ K er (T −
T (∅). Then, ρ
λ0 I ) ∩ X  T (x) = C∞ , x ∈ D(T − λ0 I ) = D(T ) and

(T − λ0 I )x = T (0). (11)

Hence, λ0 ∈ ρT (x) and there exist an open neighborhood Uλ0 of λ0 in C∞

and a function 
f x,λ0 ∈ P(Uλ0 , D(T )) verifying

x ∈ (T − μI ) 
f x,λ0 (μ) for all μ ∈ Uλ0 ∩ C. (12)

Applying P  to each sides of (12), we obtain that x = P x = P(T

 −

μI ) f x,λ0 (μ) for all μ ∈ Uλ0 ∩ C. Consequently, for all μ ∈ C ∩ Uλ0 , we
have
 
 − μI ) 
(T − λ0 I )x = (T − λ0 I ) P(T f x,λ0 (μ) . (13)

123
Decomposability and local...

On the other hand, for all μ ∈ Uλ0 ∩ C, we have

   
 − μI ) 
(T − λ0 I ) P(T f x,λ0 (μ) =(T − λ0 I ) P T − μI f x,λ0 (μ)
 
 
=(T − μI ) P T − λ0 I f x,λ0 (μ)
 
+ (μ − λ0 ) P T − λ0 I  f x,λ0 (μ)+
(λ0 − μ)(T − λ0 I ) 
f x,λ0 (μ)
 
=(T − μI ) PT − λ0 I  f x,λ0 (μ).

A combination of (11) and ( 13) implies that

T − λ0 I ) 
0 ∈ (T − μI )( P f x,λ0 (μ) for all μ ∈ Uλ0 ∩ C.

Since P T − λ0 I ) 
T is continuous, it follows that ( P f x,λ0 ∈ P(Uλ0 , D(T ))
and, because T has the ER-SVEP at λ0 , we obtain that

T − λ0 I ) 
(P f x,λ0 (μ) ∈ K er (T ) ∩ T (0) ⊆ D(T ) ∩ T (0) = {0} for all μ ∈ Uλ0 ∩ C.

Combining this with (12), we conclude that

x ∈ (T − λ0 I )  T − λ0 I ) 
f x,λ0 (λ0 ) = ( P f x,λ0 (λ0 ) + T (0) = T (0).

T (∅) ⊆ T (0) ∩ D(T ) = {0}, which

Consequently, K er (T − λ0 I ) ∩ X
proves the assertion ii).
T (∅) = {0}, and let Uλ0 ⊆ C∞ be a
ii) ⇒ i) Suppose that K er (T − λ0 I ) ∩ X
connected open neighborhood of λ0 in C∞ and  f λ0 ∈ P(Uλ0 , D(T )) be
such that

0 ∈ (T − λI ) 
f λ0 (λ) for all λ ∈ Uλ0 ∩ C. (14)

Using Proposition 5.1, this show that, for all λ ∈ Uλ0 ∩ C, we have

σ
 T(
f λ0 (λ)) ⊆ σ
 T (0) = ∅,

and hence  f λ0 (λ) ∈ X T (∅). On the other hand, there exists (an )n∈N ⊆
∞
D(T ) such that  f λ0 (λ) = (λ − λ0 )n an for all λ ∈ BC (λ0 , r ), where
n=0
BC (λ0 , r ) := {α ∈ C λ0 − α < r } ⊆ Uλ0 ∩ C. By (14), we obtain

(T − λ0 I ) 
f λ0 (λ0 ) = T (0),

so a0 =  T (∅) = {0}. Hence, for all λ ∈

f λ0 (λ0 ) ∈ K er (T − λ0 I ) ∩ X
BC (λ0 , r ), we have
∞ 

(T − λI )  f λ0 (λ) = (T − λI ) (λ − λ0 ) an
n

n=1

123
 Y. Barkaoui , M. Mnif

∞

= (λ − λ0 )(T − λI ) (λ − λ0 )n−1 an
n=1
∞

= (λ − λ0 )(T − λI ) (λ − λ0 )n an+1 .
n=0

It then follows from (14) that

∞

0 ∈ (T − λI ) (λ − λ0 )n an+1 for all λ ∈ BC (λ0 , r ) \ {λ0 }. (15)
n=0

Now, let

h : BC (λ0 , r ) −→ D(T )
∞

λ −→ (λ − λ0 )n an+1 .
n=0

Then, h ∈ P(BC (λ0 , r ), D(T )) and, by (15), we conclude that

0 ∈ (T − λI )h(λ) for all λ ∈ BC (λ0 , r ) \ {λ0 }. (16)

This implies that 0 = Q T (T − λI )h(λ) = Q T T h(λ) − λQ T h(λ), and hence

0 = lim (Q T T h(λ) − λQ T h(λ))

λ→λ0
= Q T T h(λ0 ) − λ0 Q T h(λ0 ) = Q T (T − λ0 I ) h(λ0 ).

Consequently, 0 ∈ (T − λ0 I )h(λ0 ). which yields, by (16), that 0 ∈ (T − λI )h(λ)

for all λ ∈ BC (λ0 , r ). Proceeding in the same way as in the first part of the proof, we
obtain h(λ0 ) = a1 = 0. By iteration, we can conclude that ai = 0 for all i ∈ N, so


f λ0 (λ) = 0 for all λ ∈ BC (λ0 , r ).

[20, Corollary 4.9] leads to the equality f λ0 (λ) = 0 for all λ ∈ Uλ0 ∩ C. Moreover,
because f λ0 (∞) = 0, we obtain that  f λ0 (λ) = 0 for all λ ∈ Uλ0 . Thus, T has the
ER-SVEP at λ0 . 

Remark 5.3 If T ∈ C R(X ) has the ER-SVEP at every λ ∈ C, then T has the ER-SVEP
at ∞.
Now, it has become easy to establish a global version on C∞ of Theorem 5.1. In
other words, we give a necessary and sufficient condition for a relation to admit the
ER-SVEP.
Corollary 5.1 Let T ∈ C Rc (X ) be a relation such that X = D(T ) ⊕ T (0). Then, the
following statements are equivalent:

123
Decomposability and local...

i) T has the ER-SVEP;

T (∅) = {0} for every λ0 ∈ C.
ii) K er (T − λ0 I ) ∩ X
In the following theorem, we will show how the property of the ER-SVEP of a closed
linear relation is connected to that of the SVEP, already known for a bounded linear
operator, in a powerful way.
Theorem 5.2 Let T ∈ C Rc (H ) be such that D(T ) = D(T ∗ ). If Ts has the SVEP in
the sense of [2, Definition 2.10], then T has the ER-SVEP.
Proof Since D(T ) = D(T ∗ ), we infer from assertions (2) and (3) of Proposition
3.1 that Ts ∈ End(D(T )). Since Ts has the SVEP, [2, Theorem 2.60] implies that
K er (Ts − λ0 I ) ∩ X Ts (∅) = {0} for all λ0 ∈ C. Moreover, it is easy to see from
assertion (5) of Proposition 3.1 that

K er (T − λ0 I ) ⊆ K er (Ts − λ0 I ) for all λ0 ∈ C. (17)

T (∅)) ⊆ X Ts (∅). To see this, consider an element x ∈ X

We claim that PT ( X T (∅)
and let λ0 ∈ C∞ . Then, there exists an open neighborhood U0 of λ0 in C∞ and a
function f x,λ0 ∈ P(U0 , D(T )) such that x ∈ (T − μI ) f (μ) for all μ ∈ U0 ∩ C.
Consequently,

PT x = PT (T − μI ) 
f x,λ0 (μ) for all μ ∈ U0 ∩ C.

Applying Proposition 3.1, we obtain then that 

f x,λ0 ∈ P(U0 , D(Ts )), so 
f x,λ0 |U0 ∩C ∈
H (U0 ∩ C, D(Ts )), and

PT x = (Ts − μI ) 
f x,λ0 |U0 ∩C for all μ ∈ U0 ∩ C.

T (∅)) ⊆
Hence, λ0 ∈ ρTs (PT x), and therefore PT x ∈ X Ts (∅). This means that PT ( X
X Ts (∅), as claimed. Combining this with (17), we obtain that

T (∅) = K er (T − λ0 I ) ∩ (D(T ) ∩ X
K er (T − λ0 I ) ∩ X T (∅))
T (∅) ∩ D(T ))
= K er (T − λ0 I ) ∩ PT ( X
⊆ K er (Ts − λI ) ∩ X Ts (∅) = {0}.

On the other hand, we infer from [10, Proposition III.1.4] that D(T )⊥ = D(T ∗ )⊥ =
T ∗∗ (0) = T (0), so H = D(T ) ⊕ T (0). An application of Corollary 5.1 ensures that
T has the ER-SVEP. 

We finally mention the most important result of this subsection.
Corollary 5.2 Every normal linear relation T ∈ C Rc (H ) has the ER-SVEP.
Proof Since T is normal, it follows from assertions (2) and (3) of Proposition 3.1 that
Ts ∈ End(D(T )). Moreover, we obtain from Lemma 3.1 that Ts is a normal linear
operator in D(T ), so, by [16], it has the SVEP in the sense of [2, Definition 2.10].
Using Theorem 5.2, we conclude that T has the ER-SVEP. 


123
 Y. Barkaoui , M. Mnif

5.2 Bishop and Dunford properties of a normal linear relation

As an extension of the Bishop’s property already known in the case of bounded linear
operators (see [16, Definition 1.2.5]), we consider the extended Bishop’s property for
continuous and closed linear relations, not necessarily bounded, in the sense of the
following definition.
Definition 5.4 Let T ∈ C Rc (X ). We say that T has the extended Bishop’s property
(Eβ ), shortly the property (Eβ ), if, for every open set U ⊆ C∞ and every sequences of
analytic functions ( f n )n∈N ⊆ P(U , D(T )), (gn )n∈N ⊆ H  (U , X ) with the property
that gn (λ) ∈ (T − λI ) f n (λ), for all λ ∈ U ∩ C, we have the following implication

gn −→ 0 in H  (U , X ) ⇒ f n −→ 0 in P(U , D(T )).

n→+∞ n→+∞

In the following theorem, we show the link between the Bishop’s property of the
orthogonal operator part of a linear relation T in the sense of [16, Definition 1.2.5]
and the extended Bishop’s property of T in the sense of Definition 5.4.
Theorem 5.3 Let T ∈ C Rc (H ) be such that D(T ) = D(T ∗ ). If Ts has the Bishop’s
property in the sense of [16, Definition 1.2.5], then T has the extended Bishop’s
property (Eβ ).
Proof Let U ⊆ C∞ be an open set, ( f n )n∈N ⊆ P(U , D(T )), and let (gn )n∈N ⊆
H  (U , X ) be a sequence converging to zero in H  (U , X ) with the property that

gn (μ) ∈ (T − μI ) f n (μ) for all μ ∈ U ∩ C. (18)

Applying PT to both sides of (18), we obtain PT gn (μ) = PT (T − μI ) f n (μ) for all

μ ∈ U ∩ C. Since D(T ) = D(T ∗ ), it follows from Proposition 3.1 that PT T = Ts ,
D(T ) = D(Ts ) and

PT gn (μ) = (Ts − μI ) f n (μ) for all μ ∈ U ∩ C. (19)

On the other hand, we have gn −→ 0 in H  (U , X ), so PT (gn |U ∩C ) −→ 0 in

n→+∞ n→+∞
H (U ∩ C, X ). As Ts has the Bishop’s property in the sense of [16, Definition 1.2.5],
we conclude from (19) that f n |U ∩C −→ 0 in H (U ∩ C, D(T )). But ( f n )n∈N ⊆
n→+∞
P(U , D(T )), so f n −→ 0 in P(U , D(T )). Consequently, T has the property
n→+∞
(Eβ ). 

As a consequence of Theorem 5.3, we derive the first important result of this
subsection in the following corollary.
Corollary 5.3 Every normal linear relation T ∈ C Rc (H ) has the property (Eβ ).
Proof Since T is normal, assertions (2) and (4) of Proposition 3.1 yield that Ts ∈
End(D(T )). Moreover, Lemma 3.1 ensures that Ts is a normal linear operator in
D(T ). We then infer from [16] that Ts has Bishop’s property in the sense of [16,
Definition 1.2.5]. Using Theorem 5.3, we conclude that T has the property (Eβ ). 


123
Decomposability and local...

We will close this section by studying another local spectral property of a normal
linear relation, the extended Dunford’s property.

Definition 5.5 A relation T ∈ C Rc (X ) is said to have the Extended Dunford’s property

T (F) is closed for every closed subset F ⊆ C∞ .
(EC ), shortly the property (EC ), if X

As already done for the properties ER-SVEP and (Eβ ), we will illustrate in Theorem 5.4
the link between the extended Bishop’s property of a closed linear relation T and the
Bishop’s property of the orthogonal operator part of T in the sense of [16, Definition
1.2.18]. Before doing so, we shall introduce the relation TV for every T ∈ C R(X )
and for every open set V ⊆ C∞ in the following definition.

Definition 5.6 [4, Definition 3.3] For every T ∈ C R(X ) and for every open set V ⊆
C∞ , we define the relation TV by

TV : P(V , D(T )) −→ H  (V , X )
f −→ TV ( f ) = {g ∈ H  (V , X ) | g(λ) ∈ (T − λI ) f (λ) ∀λ ∈ V ∩ C}.

Remark 5.4 Let T ∈ C Rc (H ) be such that D(T ) = D(T ∗ ). Then, [4, Theorem 3.1]
leads to the following equivalence:

T has the property (Eβ ) ⇔ I m(TV )is closed for every open set V ⊆ C∞ .

Theorem 5.4 Let T ∈ C Rc (H ) be such that D(T ) = D(T ∗ ) and I m(TV∞ ) is closed
for every open neighborhood V∞ of ∞ in C∞ . If Ts has the Dunford’s property in the
sense of [16, Definition 1.2.18], then T has the property (EC ).
T (F) be a sequence which
Proof Let F be a closed subset of C∞ and let (xn )n∈N ⊆ X

converges to x ∈ X . We have to show that x ∈ X T (F). To do this, we first observe
from the equality D(T ) = D(T ∗ ) and from Proposition 3.1 that D(T ) = D(Ts ),
K er (PT ) = T (0) and Ts = PT T . Now, we distinguish two cases:
Case I: ∞ ∈ F. In this case, we first observe that for all n ∈ N, we have C∞ \F ⊆
ρ
 T (x n ) ∩ C, so, by Remark 5.1, C∞ \F = C\(F ∩ C) ⊆ ρT (x n ).
This means that (xn )n∈N ⊆ X T (F ∩ C). Now, let λ0 ∈ C∞ \ F.
Then, there exist an open neighborhood U0 of λ0 in C and ( f n )n∈N ⊆
H (U0 , D(T )) such that

xn ∈ (T − μI ) f n (μ) for all μ ∈ U0 , n ∈ N. (20)

Applying PT to each sides of (20), so we then obtain that

PT xn = (PT T − μI ) f n (μ)
= (Ts − μI ) f n (μ) for all μ ∈ U0 , n ∈ N. (21)

Since ( f n )n∈N ⊆ H (U0 , D(Ts )), we deduce from (21) that

(PT xn )n∈N ⊆ X Ts (F ∩ C). Moreover, as PT xn −→ PT x and
n→+∞

123
 Y. Barkaoui , M. Mnif

Ts has the Dunford’s property in the sense of [16, Definition 1.2.18],

we obtain PT x ∈ X Ts (F ∩ C). Hence, there exist an open neigh-
borhood V0 of λ0 in C and f ∈ H (V0 , D(T )) with the property
that

PT x = (Ts − μI ) f (μ) for all μ ∈ V0 .

This yields that x = (I − PT )x + PT x ∈ T (0) + (Ts − μI ) f (μ)

for all μ ∈ V0 , so, by assertion (6) of Proposition 3.1,

x ∈ (T − μI ) f (μ) for all μ ∈ V0 .

Consequently, σT (x) ⊆ F ∩ C or, equivalently, x ∈ X T (F ∩ C).

Using Remark 5.2, we conclude that x ∈ X T (F ∩ C) = X T ((F ∩
C) ∪ {∞}) = X T (F), as desired.
/ F. In this case, we first observe from Remark 5.1 that σT (xn ) ⊆ F for
Case II: ∞ ∈
all n ∈ N and, showing that x ∈ X T (F) is the same as showing
that σT (x) ⊆ F and ∞ ∈ / σT (x). Arguing as above leads up to
σT (x) ⊆ F.
It remains to be shown that ∞ ∈ /σT (x). To see this, we observe that
∞∈ρ  (x
T n ), so there exists an open neighborhood U∞ of ∞ in C∞
and a sequence of functions ( f n ) ⊆ P(U∞ , D(T )) such that

xn ∈ (T − μI ) f n (μ) for all μ ∈ U∞ ∩ C and n ∈ N.

Consider the constant function h := x in H  (U∞ , X ) and the

sequence of analytic functions (h n )n∈N ⊆ H  (U∞ , X ) defined by

h n : U∞ −→ X
.
μ −→ xn , n ∈ N.

We then have h n ∈ I m(TU∞ ) and h n −→ h in H  (U∞ , X ).

n→+∞
As I m(TU∞ ) is closed, it follows that h ∈ I m(TU∞ ), so there is
f ∈ P(U∞ , D(T )) such that h ∈ TU∞ ( f ). Hence,

x = h(μ) ∈ (T − μI ) f (μ) for all μ ∈ U∞ ∩ C.

This means that ∞ ∈ ρ

 T (x) or, equivalently, ∞ ∈
/σT (x), as desired.




We end this section with a necessary condition for a relation to be a normal one,
other than decomposability and the fact that it verifies the local spectral properties
ER-SVEP and (Eβ ).

Corollary 5.4 Every normal linear relation T ∈ C Rc (H ) has the property (EC ).

123
Decomposability and local...

Proof Since T is normal, a combination of Corollary 5.3 and Remark 5.4 implies that
I m(TV∞ ) is closed for every open neighborhood V∞ of ∞ in C∞ . It also follows from
Lemma 3.1 that Ts is a normal linear operator. On the other hand, we have D(T ) =
D(T ∗ ), so assertions (2) and (4) of Proposition 3.1 ensure that Ts ∈ End(D(T )).
Consequently Ts is a bounded normal linear operator in D(T ), and hence has the
Dunford’s property in the sense of [16, Definition 1.2.18]. Applying Theorem 5.4, we
then conclude that T has the property (EC ). 


Data availability statement: Data sharing is not applicable to this article as no new data were created or
analyzed in this study.

References
1. Aiena, P., Triolo, S.: Projections and isolated points of parts of the spectrum. Adv. Oper. Theory 3,
868–880 (2018)
2. Ainea, P.: Fredholm and local spectral theory II: With application to Weyl-type theorems, vol. 2235 of
Lecture notes in mathematics, Cham: Springer, (2018)
3. Álvarez, T., Wilcox, D.: Perturbation theory of multivalued Atkinson operators in normed spaces. Bull.
Austral. Math. Soc. 76, 195–204 (2007)
4. Barkaoui, Y., Mnif, M.: Necessary Conditions For Extended Spectral Decomposable Multi-valued
Linear Operators, to appear in Turkish jounal of Mathematics, (2022)
5. Baskakov, A.G., Chernyshov, K.I.: Spectral analysis of linear relations and degenerate operator semi-
groups. Sb. Math. 193, 1573 (2002)
6. Baskakov, A.G., Krishtal, I.A.: On completeness of spectral subspaces of linear relations and ordered
pairs of linear operators. J. Math. Anal. Appl. 407, 157–178 (2013)
7. Baskakov, A.G., Zagorskii, A.S.: Spectral theory of linear relations on real Banach spaces. Math. Notes
81, 15–27 (2007)
8. Coddington, E.A.: Extension theory of formally normal and symmetric subspaces, vol. 134. American
Mathematical Society (AMS), Providence, RI (1973)
9. Colojoară, I., Foiaş, C.: Theory of generalized spectral operators, 9. CRC Press, (1968)
10. Cross, R.: Multivalued linear operators, vol. 213. CRC Press, New York (1998)
11. Forster, O.: Lectures on Riemann Surfaces, vol. 81. Springer, New York (1981)
12. Hassi, S., de Snoo, H.S.V.: Factorization, majorization, and domination for linear relations. Ann. Univ.
Sci. Budap. Rolando Eötvös 58, 55–72 (2015)
13. Hassi, S., de Snoo, H.S.V., Szafraniec, F.H.: Componentwise and cartesian decompositions of linear
relations. Diss. Math. 465, 59 (2009)
14. Holmgren, R. A.: A first course in discrete dynamical systems, no. 2, Springer Science & Business
Media, (2000)
15. Lange, R.: Equivalent conditions for decomposable operators. Proc. Am. Math. Soc. 82, 401–406
(1981)
16. Laursen, K. B., Neumann, M. M.: An Introduction to Local Spectral Theory, vol. 20, Oxford University
Press, (2000)
17. Mnif, M., Ouled-Hmed, A.-A.: Local spectral theory and surjective spectrum of linear relations. Ukr.
kyi Mat. Zhurnal 73, 222–237 (2021)
18. Radjabalipour, M.: On subnormal operators. Trans. Am. Math. Soc. 211, 377–389 (1975)
19. Sandovici, A., Vasilescu, F.-H.: Normal extensions of subnormal linear relations via quaternionic
Cayley transforms. Monatsh. Math. 170, 437–463 (2013)
20. Stein, E.M., Shakarchi, R.: Complex analysis, no. 2 in Princeton lectures in analysis, Princeton, NJ:
Princeton University Press, (2007)
21. Timoney, R. M.: Course, chapter 6: The metric space M(G) and normal families, Trinty College Dublin,
(2004)

123
 Y. Barkaoui , M. Mnif

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and applicable
law.

123