operators similar to their adjoints - univ-oran1.dz · the aim of this work is to study the...
TRANSCRIPT
– – – – – – – – – – – – – – – – – – – – – – – – – – – – —
Popular and Democratic Republic of Algeria
University of Oran 1 Ahmed Ben Bella
Faculty of exact and applied Sciences
Department of Mathematics
Operators Similar To Their Adjoints
THESIS SUBMITTED FOR THE DEGREEOF DOCTORATE IN MATHEMATICS
Presented by :
Dehimi Souheyb
Doctorate’s Thesis Committee :
President: Professor C.Bouzar University of Oran 1 Ahmed Ben Bella
Thesis Supervisor: Professor M.H.Mortad University of Oran 1 Ahmed Ben Bella
Examiners: Professor K.Belghaba University of Oran 1 Ahmed Ben Bella
Dr. M.Meftah (M.C.A) University of Oran 1 Ahmed Ben Bella
Professor B.Benahmed National Polytechnic School of Oran (ENPO Ex: ENSET)
Professor M.Tlemçani University of Sciences and Technology of Oran (M.B)
Academic year 2016− 2017
Acknowledgement
I would like to thank first and for most my supervisor, Professor Mohammed Hichem
Mortad, for his many suggestions, helpful discussions, patience and constant support
during this research.
I sincerely thank Professor C.Bouzar for giving me the honor of being president of the
jury.
I thank Professors K.Belghaba, B.Benahmed and M.Tlemçani and Doctor M.Meftah
for their time and effort participating in my thesis committee.
Of course, I am grateful to my parents for their support, encouragement, and patience.
Without them this work would never have come into existence.
Contents
Introduction vi
1 Essential background 11
1.1 Banach algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.2 Basic properties of spectra . . . . . . . . . . . . . . . . . . . . . . . 12
1.2 C∗-algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Bounded operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.1 Definitions and properties . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.2 Approximate point spectrum . . . . . . . . . . . . . . . . . . . . . . 18
1.3.3 Resolutions of the identity . . . . . . . . . . . . . . . . . . . . . . . 19
1.4 Polar decomposition of an operator . . . . . . . . . . . . . . . . . . . . . . 21
1.4.1 Isometry and partial isometry . . . . . . . . . . . . . . . . . . . . . 22
1.4.2 Polar decomposition of an operator . . . . . . . . . . . . . . . . . . 22
1.5 Positive operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.6 Numerical range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2 Non-normal operator classes 29
2.1 Compact operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2 Hyponormal operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.1 Definitions and properties . . . . . . . . . . . . . . . . . . . . . . . 31
i
2.2.2 Some conditions implying normality or self-adjointness . . . . . . . 33
2.2.3 p-Hyponormal operators . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3 Normaloid operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.4 Paranormal operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4.1 Definitions and properties . . . . . . . . . . . . . . . . . . . . . . . 42
2.4.2 k-paranormal operators . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5 Convexoid operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.5.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.6 Class A operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.6.1 Quasi-class A operators . . . . . . . . . . . . . . . . . . . . . . . . 54
3 Similarities involving bounded operators 56
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2 Operators similar to their adjoints . . . . . . . . . . . . . . . . . . . . . . . 58
3.2.1 Conditions implying self-adjointness of operators . . . . . . . . . . . 58
3.2.2 Operators similar to self-adjoint ones . . . . . . . . . . . . . . . . . 63
3.3 Operators with inverses similar to their adjoints . . . . . . . . . . . . . . . 64
3.3.1 Operators similar to unitary ones . . . . . . . . . . . . . . . . . . . 64
3.3.2 Operators with left inverses similar to their adjoints . . . . . . . . . 69
3.4 Similarities involving normal operators . . . . . . . . . . . . . . . . . . . . 73
3.5 Quasi-similarity of operators . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4 Similarities involving unbounded operators 78
4.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.1.1 Adjoint Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.1.2 Self-adjoint Operators . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.2 Quasi-similarity of unbounded operators . . . . . . . . . . . . . . . . . . . 86
4.3 Similarities involving unbounded normal operators . . . . . . . . . . . . . . 90
ii
4.4 Unbounded operators similar to their adjoints . . . . . . . . . . . . . . . . 94
Bibliography 99
iii
Résumé
Le but de ce travail est d’étudier la similarité entre les opérateurs linéaires et leurs
adjoints dans un espace de Hilbert. Le travail présenté est organisé selon le plan suivant:
Dans les deux premiers chapitres, un rappel sur des notions essentiel les sur les opérateurs
linéaires bornés. Dans les deux derniers chapitres nous donnons quelques conséquences de
la similarité et quasi-similarité entre les opérateurs linéaires. L’un des principaux objectifs
de cette thèse est de généraliser le théorème de Sheth.
Abstract
The aim of this work is to study the similarity between linear operators and their
adjoints in a Hilbert space. The work is organized according to the following plan. In
the first two chapters, a reminder on essential notions about bounded linear operators.
In the last two chapters we give some consequences of the similarity and quasi-similarity
between the linear operators. One of the main objectives of this thesis is to generalize
Sheth’s theorem.
Keywords: similarity, bounded and unbounded operators, closed, self-adjoint, normal,
hyponomal operators, unitary cramped operators, Sheth.
iv
Notationsk The field of real or complex numbers.
σ(T ) The spectrum of T.
r(T ) The spectral radius of T.
B(H) The Banach algebra of all bounded linear operators.
ran(T ) The range of a linear operator T.
T ∗ The adjoint operator of T.
σp (T ) The point spectrum of T.
σapp (T ) The approximate point spectrum of T.
PM Initial projection
PN Final projection
W (T ) The numerical range of T.
B∞(X, Y ) The set of all compact linear operators.
co σ (T ) The convex hull of σ (T ) .
σe (T ) The essential spectrum of T.
D (T ) The domain of a linear operator T.
G (T ) The graph of an operator T.
T The closure of the closable operator T.
v
Introduction
The theory of Banach algebras is an abstract mathematical theory which is the synthesis
of many specific cases from different areas of mathematics.
They are named after the Polish mathematician Stefan Banach (1892—1945) who had
introduced the concept of Banach space, but Banach had never studied Banach algebras.
In fact, the first one who have defined them is Mitio Nagumo in 1936 ([Nag36]) under
the name “linear metric rings”. In 1941, I.M.Gelfand (1913—2009) introduced them under
the name “normed rings”([Gel41]). In the classical monograph [BD73], the authors write
that, if they had it their way, they would rather speak of “Gelfand algebras”. But in 1945,
Warren Ambrose (1914—1995) came up with the name “Banach algebras”([Amb45]) .
Banach algebras show up naturally in many areas of analysis: Let X be a Banach
space. then B (X) , the algebra of all bounded linear operators on X, is a Banach algebra,
with respect to the usual operator norm. If we have an analytic object that has a Banach
algebra naturally associated with it, then this algebra can provide us with further insight
into the nature of the underlying object.
A notion which will be of great use in this thesis is that of hyponormal operator. A
bounded hyponormal operator is a bounded operator T on a Hilbert space H such that
T ∗T ≥ TT ∗ . This definition was introduced by Paul Halmos [Halm50] in 1950 and
generalizes the concept of a normal operator (where T ∗T = TT ∗). The important thing
is that there is a prominent example of a hyponormal operator, the unilateral shift. If l2
vi
is the Hilbert space of square summable sequences and T is defined on l2 by
T (x0, x1, . . .) = (0, x0, x1, . . .) .
Then T is called the unilateral shift and is the most basic example of hyponormal
operators. The unilateral shift is a well understood non normal operator; it is arguably
the best understood non normal operator on an infinite dimensional space.
Paul Halmos began a strategic attack on operator theory by extracting two properties
of the shift in [Halm50]. One was the definition of hyponormal operators and the other the
idea of a subnormal operator. A subnormal operator is one that has a normal extension;
every subnormal operator is hyponormal. One of the first important results in the theory
of hyponormal operators, due to C. R. Putnam, is the fact that if T is a pure hyponormal
operator, then its real and imaginary parts, X and Y , must be absolutely continuous self-
adjoint operators [Put63]. That is, the spectral measures for X and Y must be absolutely
continuous with respect to Lebesgue measure on the real line. Thus the Spectral Theorem
for self-adjoint operators can be applied to X, and this operator can be represented as a
multiplication operator on L2 [a, b] for some interval in R. The operator Y also has such
a representation, but on a different L2 space. Can Y be represented on the same space
L2 [a, b] in a way that is intimately connected with the representation of X ? Indeed, a
result of Kato [Kat68], though it is not directly related to hyponormal operators, implies
that this can be done. About the same time, many authors began investigating hyponor-
mal operators from this perspective [Pin68, Rus68, Xia63][Rus68][Xia63]. In addition if
T is hyponormal, then
π ‖T ∗T − TT ∗‖ ≤ Area (σ (T )) .
This was proved by Putnam in [Put70] and is well known as Putnam’s inequality.
An operator T ∈ B (H) is said to be p-hyponormal if
(TT ∗)p ≤ (T ∗T )p
vii
for a positive number p. In fact, semi-hyponormal operators were first introduced by
Professor D. Xia in [Xia80]. He also provides an example of a semi-hyponormal operator
which is not hyponormal. In [Xia80] Xia proved that, if T is p-hyponormal, then
π ‖T ∗T − TT ∗‖ ≤ p
∫∫σ(T )
ρ2p−1dρdθ
for p ≥ 12. Cho and Itoh proved that Putnam’s inequality holds for p−hyponormal oper-
ators in [CI95] .
The notion of a paranormal operator dates back to 1960s and is due to V. Istratescu in
[Ist66] he named them “operators of class N ”. T. Furuta in [Fur67] introduced the term
“paranormal operators”. The class of paranormal operators can be seen as a generalization
of other important classes: hyponormal operators and subnormal and normal operators .
In subsequent years paranormal operators have been the subject of further research. For
example, we know that a paranormal operator T is compact if and only if T n is compact
for some n ∈ N. Moreover, compact paranormal operator is normal . Paranormality
appears also to be an important property when studying various problems in operator
theory.
Let A and B be a bounded operators. We say that A is similar to B iff
SA = BS
for some bounded invertible operator S. In 1956 Beck and C.R.Putnam showed that if
T is a bounded operator which is unitarily equivalent to its adjoint T ∗, via cramped
unitary operator U, necessarily T is self-adjoint. Our main work is to answer the next
question: suppose that T is a bounded operator and S is an invertible operator for which
0 /∈ W (S) and ST = T ∗S, where W (S) is the numerical range, then when does it follow
that necessary T is self-adjoint?.
In 1966, Sheth had proved that, if a bounded hyponormal operator T satisfies the above
relation, then T is self-adjoint. In 1969 J. P. Williams, in his paper Operators Similar to
Their Adjoints, had proved an important theorem which is : If T is a bounded operator
viii
such that S−1TS = T ∗, where 0 /∈ W (S), then the spectrum of T is real. Williams
result is considered a motivation of our thesis. In [Cas83], Castern had given a necessary
and suffi cient conditions for a bounded operator T being similar to unitary or self-adjoint
operator.
Embry showed that, If S and T are commuting normal operators and AS = TA, where
0 is not in the numerical range of A, then S = T . Her result includes a slight improvement
of a result of J. P.Williams. In [Wil69] Williams proved that σ (E) is real if AE = E∗A,
where 0 is not in the closure of W (A). Thus if E is normal, then E is self-adjoint.
If H and K are complex Hilbert spaces, the bounded linear operator X : H −→ K
is said to be quasi-invertible iff it is one-to-one and has dense range. Two operators
A : H −→ H and B : K −→ K are quasi-similar provided there exist quasi-invertible
operators X : H −→ K and Y : K −→ H such that XA = BX and Y B = AY . Quasi-
similarity was first introduced by Sz-Nagy and Foias (see, for example [NF70]) in their
theory of an infinite dimensional analogue of the Jordan for certain classes of operators. It
replaces the familiar notion of similarity. Quasi-similarity is the same thing as similarity in
finite dimensional spaces, but in infinite-dimensional spaces it is a much weaker relation,
so weak that two operators can be quasi-similar and yet have unequal spectra [NF70,
p.262]
For normal operators this cannot happen! It follows from the Fuglede-Putnam commu-
tativity theorem that if two normal operators are quasi-similar, they are actually unitarily
equivalent [Dou69] and therefore have equal spectra. This result does not generalize to
hyponormal operators. Sarason has given an example of two hyponormal operators which
are similar but not unitarily equivalent [Hal67].
Closed linear operators are a class of linear operators on Banach spaces. They are
more general than bounded operators, and therefore not necessarily continuous, but they
still retain nice enough properties that one can define the spectrum and (with certain
assumptions) functional calculus for such operators. Many important linear operators
ix
which fail to be bounded turn out to be closed. Let X, Y be two Banach spaces. A linear
operator T : D(T ) ⊂ X → Y is closed if for every sequence xn in D(T ) converging to
x in X such that Txn → y ∈ Y as n→∞ one has x ∈ D(T ) and Tx = y.
A densely defined operator T is said to be hyponormal if: D (T ) ⊆ D (T ∗) , and
‖T ∗x‖ ≤ ‖Tx‖ for all x ∈ D (T ) . In [OS89], Ôta and Schmüdgen proved that quasi-
similar closed hyponormal operators have equal spectra, and in [Mor10], Mortad gener-
alized Embry’s famous theorem. Mortad’s result considered as a further motivation to
generalize Sheth’s theorem.
x
Chapter 1
Essential background
This chapter is divided into two parts. In the first one, we collect fundamental results on
Banach Algebra, mostly without proof. The second part will be dedicated to the study of
bounded linear operators. Most of the material covered in this chapter is from [Con90].
1.1 Banach algebra
1.1.1 Introduction
A complex algebra is a vector space A over the complex field C in which a multiplication
is defined and satisfies
x (yz) = (xy) z,
(x+ y) z = xz + yz,
and
α (xy) = (αx) y = x (αy)
for all x, y, and z in A and for all scalars α.
if, in addition, A is a Banach space with respect to a norm that satisfies the multiplication
inequality
‖xy‖ ≤ ‖x‖ ‖y‖ (x ∈ A, y ∈ A)
11
and if A has an identity e such that
xe = ex = x (x ∈ A)
and
‖e‖ = 1
then A is called a Banach algebra.
The presence of a unit is very often omitted from the definition of a Banach algebra.
However, when there is a unit it makes sense to talk about inverses, so that the spectrum
of an element of A can be defined in a more natural way than is otherwise possible.
1.1.2 Basic properties of spectra
Definition 1.1.1 Let A be a Banach algebra. The spectrum σ (x) of x ∈ A is the set of
all complex numbers λ such that λe − x is not invertible. The resolvent set ρ(x) of x is
the complement of σ (x) ; it consists of all λ ∈ C for which (λe− x)−1 exists.
The spectral radius of x is the number
r (x) = sup |λ| : λ ∈ σ (x) .
Of course, r (x) makes no sense if σ (x) is empty. But this never happens, as we shall
see.
Theorem 1.1.2 If A is a Banach algebra and x ∈ A , then
1) the spectrum σ (x) of x is compact and nonempty.
2) The spectral radius r (x) of x satisfies
r (x) = limn→∞
‖xn‖1n .
Definition 1.1.3 A subset J of a commutative complex algebra A is said to be an ideal
if :
12
(1) J is a subspace of A (in the vector space sense), and
(2) xy ∈ J whenever x ∈ A and y ∈ J.
If J 6= A, J is a proper ideal. Maximal ideals are proper ideals which are not contained
in any larger proper ideal.
Proposition 1.1.4 Let A be a commutative complex algebra, and let ∆ be the set of all
complex homomorphisms of A
(1) An element x ∈ A is invertible in A if and only if h(x) 6= 0 for every h ∈ ∆.
(2) An element x ∈ A is invertible in A if and only if x lies in no proper ideal of A.
(3) λ ∈ σ(x) if and only if h(x) = λ for some h ∈ ∆.
1.2 C∗-algebra
A C∗-algebra is a particular type of Banach algebra that is intimately connected with
the theory of operators on a Hilbert space. Some of the general theory developed in this
section will be used in the next section to prove the spectral theorem, which reveals the
structure of normal operators.
Definition 1.2.1 If A is a Banach algebra, an involution is a map x 7→ x∗ of A into A
such that the following properties hold for x and y in A and α in C :
(1) (x∗)∗ = x
(2)(xy)∗ = y∗x∗
(3) (αx+ y)∗ = αx∗ + y∗
Note that if A has involution and an identity e, then e∗x = (e∗x)∗∗ = (x∗e)∗ = x,
similarly, xe∗ = x. Since the identity is unique, e∗ = e.
Definition 1.2.2 A C∗-algebra is a Banach algebra A with an involution such that for
every x in A:
‖x∗x‖ = ‖x‖2 .
13
Any x ∈ A for which x = x∗ is called hermitian, or self-adjoint.
Proposition 1.2.3 If A is C∗-algebra and x ∈ A, then
‖x∗‖ = ‖x‖
Proof. Note that ‖x‖2 = ‖x∗x‖ ≤ ‖x∗‖ ‖x‖ , so ‖x‖ ≤ ‖x∗‖ . Since x = x∗∗, then
‖x∗‖ ≤ ‖x‖ , which means ‖x∗‖ = ‖x‖ .
Definition 1.2.4 Let ∆ be the set of all complex homomorphisms of a commutative Ba-
nach algebra A . The formula
x(h) = h(x) (h ∈ ∆)
assigns to each x ∈ A a function x : ∆→ C; we call x the Gelfand transform of x.
Theorem 1.2.5 Suppose A is commutative C∗-algebra, with maximal ideal space ∆. The
Gelfand transform is then an isometric isomorphism of A onto C (∆), which has the
additional property that
h(x∗) = h(x) (x ∈ A, h ∈ ∆)
or, equivalently, that
(x∗) ˆ = x (x ∈ A)
In particular, x is hermitian if and only if x is a real-valued function.
The next theorem is a special case of the previous theorem. We shall state it in a
form that involves the inverse of the Gelfand transform, in order to make contact with
the symbolic calculus.
Theorem 1.2.6 If A is commutative C∗-algebra, which contains an element x such that
the polynomials in x and x∗ are dense in A, then the formula
(Ψf) ˆ = f x
14
defines an isometric isomorphism Ψ of C (σ(x)) onto A, which satisfies
Ψf = (Ψf)∗
for every f ∈ C (σ(x)) . Moreover , if f(λ) = λ on σ(x), then Ψf = x.
Definition 1.2.7 In a Banach algebra with involution, the statement ”x ≥ 0” means that
x = x∗ and that σ(x) ⊂ [0,∞[.
Theorem 1.2.8 Every C∗-algebra A has the following properties:
(1) Hermitian elements have real spectra.
(2) If x ∈ A is normal (xx∗ = x∗x), then r(x) = ‖x‖ .
(3) If x ∈ A , then r(xx∗) = ‖x‖2 .
(4) If x ∈ A, y ∈ A, x ≥ 0, and y ≥ 0, then x+ y ≥ 0.
(5) If x ∈ A , then xx∗ ≥ 0.
Proposition 1.2.9 Suppose A is C∗-algebra,B is a closed subalgebra of A , e ∈ B. Then
σB (x) = σA (x) for every x ∈ B.
1.3 Bounded operators
In conformity with notations used earlier, B (H) will now denote the Banach algebra of
all bounded linear operators T on a Hilbert space H 6= 0, normed by
‖T‖ = sup ‖Tx‖ : x ∈ H, ‖x‖ ≤ 1
We shall see that B (H) has an involution which makes it into a C∗-algebra.
Theorem 1.3.1 If T ∈ B (H) and if (Tx, x) = 0 for every x ∈ H, then T = 0
15
Proof. Since (T (x+ y) , x+ y) = 0, we see that
(Tx, y) + (Ty, x) = 0 (x ∈ H, y ∈ H) . (1.1)
If y is replaced by iy in (1.1) , the result is
−i (Tx, y) + i (Ty, x) = 0 (x ∈ H, y ∈ H) . (1.2)
Multiply (1.2) by i and add to (1) , to obtain
(Tx, y) = 0. (1.3)
With y = Tx, (1.3) gives ‖Tx‖2 = 0. Hence Tx = 0.
Corollary 1.3.2 If S, T ∈ B (H) , such that
(Sx, x) = (Tx, x)
for every x ∈ H, then S = T.
Definition 1.3.3 Let T ∈ B (H) , then the unique operator S ∈ B (H) satisfying
(Tx, y) = (x, Sy) (x ∈ H, y ∈ H)
is called the adjoint of T and is denoted by S = T ∗.
We claim that T → T ∗ is an involution on B (H) , that is, that the following properties
hold :(T + S)∗ = T ∗ + S∗
(αT )∗ = αT ∗
(ST )∗ = T ∗S∗
T ∗∗ = T
Since
‖T ∗T‖ = ‖T‖2
holds for every T ∈ B (H) , then B (H) is a C∗-algebra, relative to the involution T → T ∗.
16
1.3.1 Definitions and properties
We recall thatker (T ) = x ∈ H : Tx = 0
ran (T ) = y ∈ H : y = Tx, x ∈ H .
Theorem 1.3.4 Let T ∈ B (H) , then
ker (T ∗) = ran(T )⊥ and ker (T ) = ran(T ∗)⊥.
Definition 1.3.5 An operator T ∈ B (H) is said to be
(1) normal if T ∗T = TT ∗,
(2) self-adjoint (or hermitian) if T ∗ = T ,
(3) unitary if T ∗T = I = TT ∗, where I is the identity operator on H,
(4) an isometry if T ∗T = I
(5) a projection if T 2 = T.
It is clear that self-adjoint operators and unitary operators are normal.
Theorem 1.3.6 An operator T ∈ B (H) is normal if and only if
‖Tx‖ = ‖T ∗x‖
for every x ∈ H. Normal operators T have the following properties:
(1) ker (T ) = ker (T ∗) .
(2) ran (T ) is dense in H if and only if T is one to-one.
(3) T is invertible if and only if there exist c > 0 such that ‖Tx‖ ≥ c ‖x‖ for every x ∈ H.
Theorem 1.3.7 If U ∈ B (H) , the following three statements are equivalent:
(1) U is unitary .
(2) ran (U) = H and (Ux, Uy) = (x, y) for all x ∈ H, y ∈ H.
(3) ran (U) = H and ‖Ux‖ = ‖x‖ for all x ∈ H.
17
Theorem 1.3.8 Each of the following four properties of a projection P ∈ B (H) implies
the other three:
(1) P is self-adjoint.
(2) P is normal.
(3) ran (P ) = ker (P )⊥ .
(4) (Px, x) = ‖Px‖2 .
Property 3) is usually expressed by saying that P is an orthogonal projection.
Theorem 1.3.9
(1) If U is unitary and λ ∈ σ (U) , then |λ| = 1.
(2) If S is self-adjoint and λ ∈ σ (S) , then λ is a real number.
1.3.2 Approximate point spectrum
Definition 1.3.10 Let T ∈ B (H) , the point spectrum of T , σp (T ) is defined by
σp (T ) = λ ∈ C : ker (T − λI) 6= 0 .
Definition 1.3.11 Let T ∈ B (H) , the approximate point spectrum of T, σap (T ) is de-
fined by
σap (T ) = λ ∈ C : there is a sequence xn in H such that ‖xn‖ = 1 and ‖(T − λI)xn‖ → 0. .
Note that σp (T ) ⊂ σap (T ) .
Proposition 1.3.12 Let T ∈ B (H) , the following statements are equivalent:
(1) λ /∈ σap (T ) .
(2) ker (T − λI) = 0 and ran(T ) is closed.
(3) There is a constant c > 0 such that ‖(T − λI)x‖ ≥ c ‖x‖ for all x ∈ H.
Theorem 1.3.13 The approximate point spectrum σap(T ) is a nonempty closed subset of
C that includes the boundary ∂σ(T ) of the spectrum σ(T ).
18
Theorem 1.3.14 (Fuglede-Putnam-Rosenblum) Assume that M,N, T ∈ B (H) , M
and N are normal, and
MT = TN. (1.4)
Then M∗T = TN∗.
Note that the hypotheses of Theorem (1.3.14) do not imply that MT ∗ = T ∗N, even
when M and N are self-adjoint and T is normal. For instance, if
M =
1 0
0 −1
, N =
0 1
1 0
, T =
1 1
−1 1
,
then MT = TN but MT ∗ 6= T ∗N.
1.3.3 Resolutions of the identity
Definition 1.3.15 Let R be a σ-algebra in a set Ω, and let H be a Hilbert space. In this
setting, a resolution of the identity is a mapping
E : R → B (H)
with the following properties:
(1) E (∅) = 0, E (Ω) = 1.
(2) Each E (w) is a self-adjoint projection for all w ∈ R .
(3) E(w′ ∩ w′′
)= E
(w′)E(w′′)for all w
′, w′′ ∈ R .
(4) If w′ ∩ w = ∅′′ , then E
(w′ ∪ w′′
)= E
(w′)
+ E(w′′)for all w
′, w′′ ∈ R .
(5) For every x ∈ H and y ∈ H, the set functions Ex,y defined by
Ex,y(w) = (E(w)x, y)
is a complex measure.
19
When R is the set of all Borel sets on a compact or locally compact Hausdorf space,
it is customary to add another requirement to 4) : each Ex,y should be a regular borel
measure.
Here are some immediate consequences of these properties.
Since each Ex,y is self-adjoint projection, we have
Ex,x(w) = (E(w)x, x) = ‖E(w)x‖2 (w ∈ R, x ∈ H) .
By (3) , any two of the projections E (w) commute with each other.
Theorem 1.3.16 If A is a closed normal subalgebra of B (H) which contains the identity
operator I and if ∆ is the maximal ideal space of A, then the following assertion are true:
(1) There exists a unique resolution E of the identity on the Borel subsets of ∆ which
satisfies
T =
∫∆
T dE
for every T ∈ A, where T is the Gelfand transform of T.
(2) An operator S ∈ B (H) commutes with every T ∈ A if and only if S commutes with
every projection E(w).
We now specialize this theorem to a single operator.
Theorem 1.3.17 If T ∈ B (H) and T is normal, then there exists a unique resolution of
the identity E on the Borel subsets of σ (T ) which satisfies
T =
∫σ(T )
λdE(λ).
Furthermore, every projection E(w) commutes with every S ∈ B (H) which commutes
with T.
Proof. Let A be the smallest closed subalgebra of B (H) that contains I, T, and T ∗. Since
T is normal , Theorem (1.3.16) applies to A. By Theorem (1.2.6), the maximal ideal
20
space of A can be identified with σ (T ) in such way that T (λ) = λ for every λ ∈ σ (T ) .
The existence of E follow now from Theorem (1.3.16).
If ST = TS, then also ST ∗ = T ∗S, by Theorem (1.3.14) ; hence S commutes with every
member of A. By (2) of Theorem 1.3.16 SE(w) = E(w)S for every Borel set w ∈ σ (T ) .
Remark 1.3.18 If E is the spectral decomposition of normal operator T ∈ B (H), and if
f is a bounded Borel function on σ (T ) , it is customary to denote the operator
Ψ(f) =
∫σ(T )
fdE
by f (T ) .
Using this notation, part of the content of Theorems (1.3.16) and (1.3.17) can be sum-
marized as follows:
The mapping f → f (T ) is a homomorphism of the algebra of all bounded Borel functions
on σ (T ) into B (H) , which carries the function 1 to I, and carries the identity function
on σ (T ) to T.
If S ∈ B (H) and ST = TS, then Sf (T ) = f (T )S for every bounded Borel function f .
Proposition 1.3.19 A normal T ∈ B (H) is
(1) self-adjoint if and only if σ (T ) lies in the real axis.
(2) unitary if and only if σ (T ) lies on the unit circle.
1.4 Polar decomposition of an operator
This section ([Na07]) consists of two subsections. In the first one we introduce the concept
of isometry and partial isometry. In the second subsection, we discuss in some detail the
polar decomposition of an operator.
21
1.4.1 Isometry and partial isometry
What is the appropriate analog of a complex number of length one ? If z is a complex
number such that |z| = 1, then it lies on the unit circle in the complex plane. Since
|zw| = |z| |w| = |w|
for all w, we see that multiplication by z preserves length (it does not stretch). We are
therefore led to consider operators that preserve norms. That is for an arbitrary inner
product space H, we shall consider an operator U in B (H) such that for any x ∈ H we
would have, ‖Ux‖ = ‖x‖.
Definition 1.4.1 A bounded linear operator U on a complex Hilbert space H is said to
be a partial isometry operator if there exists a closed subspace M such that ‖Ux‖ = ‖x‖
for all x ∈M , and Ux = 0 for any x ∈M⊥, where M is called the initial space of U , and
the range of U , ran (U) is called the final space, and the projections onto the initial space
and the final space are said to be initial projection and final projection,respectively.
Theorem 1.4.2 ([Na07]) Let U be a partial isometry operator on a complex Hilbert
space H with initial space M and final space N , PM and PN are initial projection and
final projection ,respectively. Then the following hold :
1) UPM = U and U∗U = PM .
2) U∗ is a partial isometry operator with initial space N and final space M , that is
U∗PN = U∗ and U∗U = PN .
1.4.2 Polar decomposition of an operator
We can write a complex number z = a+ ib in polar form using the formulas :
a = r cos θ and b = r sin θ.
.
22
In other words,
z = r (cos θ + i sin θ) .
Using this analogy, how do we write an arbitrary bounded linear operator acting on a
Hilbert space in ”polar form”?
If we use the complex numbers themselves to do this, then, the motivation for polar
decomposition would be the following equation:
z =z
|z| |z| =(z
|z|
)√zz.
We see that√zz is a positive real number, and z
|z| is a complex number of absolute value
equals one. Since adjoints are the operator analog of complex conjugation, we expect that
an arbitrary operator T ∈ B (H) can be written in the form
T = U√T ∗T
where U is an isometry. Amazingly, this works! Since√T ∗T is well-defined. This is
the basic idea behind the polar decomposition theorem.
Theorem 1.4.3 Let M be a dense subspace of a normed space X . Let T be a bounded
linear operator from M to a Banach space Y. Then there exist T which is the unique
extension of T from X to Y , with ‖Tx‖ =∥∥∥T x∥∥∥ .
Theorem 1.4.4 Let S and T be bounded linear operators on a complex Hilbert space H.
If T ∗T = S∗S, then there exists a partial isometry operator U with initial space
M = ran (T ) and final space N = ran (S), and S = UT.
Theorem 1.4.5 Let T be a bounded linear operator on a complex Hilbert space H. Then
the following hold :
1) There exists a partial isometry operator U such that
T = U |T |
23
where |T | = (T ∗T )12 , M and N are initial and final spaces of U , respectively, can be
expressed as follows: M = ran (|T |) and N = ran (U).
2) ker (U) = ker (|T |) and U∗U |T | = |T |.
3) ran (|T |) = ran (T ∗).
Proof. 1) Follows immediately from Theorem 1.4.3.
2) Let x ∈M⊥, then Ux = 0, so x ∈ ker (U). Hence M⊥ ⊆ ker (U).
Conversely, let x ∈ ker (U) then Ux = 0, but M is closed then x can be written
uniquely as, x = y + z : y ∈M and z ∈M⊥. Thus
0 = ‖Ux‖ = ‖Uy‖ = ‖y‖ .
Hence, y = 0 and x = z ∈M⊥. Therefore, ker (U) ⊆M⊥, hence ker (U) = M⊥. By using
the fact that |T | is self-adjoint and
ran (|T |) = ker (|T |)⊥
we have ker (U) = ker (|T |) .
By using (1) above, and by Theorem1.4.2 we have, U∗T = U∗U |T | = PM |T |. ButM ⊃
ran (|T |), therefore, PM |T | = |T |. Hence U∗U |T | = |T |.
3) Since T = U |T |, then
U∗T = U∗U |T | = |T | .
Then,
ran (|T |) = ran (|T |∗) = ran (T ∗U) ⊆ ran (T ∗) .
Conversely, since T ∗ = |T |U∗, then
ran (T ∗) = ran (|T |U∗) ⊆ ran (|T |) .
Therefore, ran (T ∗) =ran (|T |).
24
Lemma 1.4.6 Let H be a complex Hilbert space, and S ∈ B (H) be a positive operator.
Then
1) 〈Sx, x〉 = 0 holds for some x ∈ H if and only if Sx = 0.
2) ker (Sq) = ker (S) holds for any positive real number q.
Theorem 1.4.7 Let T ∈ B (H), H is a complex Hilbert space, and let T = U |T | be the
polar decomposition of T . Then the following holds :
1) ker (T ) = ker (|T |) .
2) |T ∗|q = U |T |q U∗ for any positive number q.
Theorem 1.4.8 Let ∈ B (H), where H is a complex Hilbert space, and let T = U |T | be
the polar decomposition of T . Then T ∗ = U |T ∗| is the polar decomposition of T ∗.
Corollary 1.4.9 Let ∈ B (H), where H is a complex Hilbert space, and let T = U |T | be
the polar decomposition of T . Then
|T |q = U∗ |T ∗|q U for any positive number q.
1.5 Positive operators
Theorem 1.5.1 Suppose T ∈ B (H) , then
(1) (Tx, x) ≥ 0 for every x ∈ H if and only if
(2) T = T ∗ and σ (T ) ⊂ [0,∞[.
If T ∈ B (H) satisfies (1), we call T a positive operator and write T ≥ 0.
Proof. In general, (Tx, x) and (x, Tx) are complex conjugates of each other. But if (1)
holds, then (Tx, x) is real , so that
(x, T ∗x) = (Tx, x) = (x, Tx)
25
for every x ∈ H. By Corollary (1.3.2), T = T ∗, and thus σ (T ) lien in the real axis. If
λ > 0, (1) implies that
λ ‖x‖2 = (λx, x) ≤ ((T + λI)x, x) ≤ ‖(T + λI)x‖ ‖x‖
so that
‖(T + λI)x‖ ≥ λ ‖x‖ .
By Theorem (1.3.6), T +λI is invertible in B (H) , and −λ is not in σ (T ) . It follows that
(1) implies (2).
Assume now that (2) holds, and let E be the spectral decomposition of T , so that
(Tx, x) =
∫σ(T )
λdEx,x(λ) (x ∈ H) .
Since each Ex,x is a positive measure, and since λ ≥ 0 on σ (T ) , we have (Tx, x) ≥ 0.
Thus (2) implies (1).
Theorem 1.5.2 Every positive T ∈ B (H) has a unique positive square root S ∈ B (H) .
If T is invertible, so is S.
Theorem 1.5.3 If T ∈ B (H) , then the positive square root of T ∗T is the only positive
operator P ∈ B (H) that satisfies ‖Px‖ = ‖Tx‖ for every x ∈ H.
The fact that every complex number λ can be factored in the form λ = α |λ| , where
|α| = 1, suggests the problem of trying to factor T ∈ B (H) in the form T = UP, with U
is unitary and P ≥ 0. When this possible we call UP a polar decomposition of T.
Theorem 1.5.4 Let T ∈ B (H) , then
(1) If T is invertible, then T has a unique polar decomposition T = UP.
(2) If T is normal, then T has a polar decomposition T = UP in wich U and P commute
which each other and with T .
26
In (1), no two of T, U, P need to commute. For example 0 1
2 0
=
0 1
1 0
2 0
0 1
The polar decomposition leads to an interesting result concerning similarity of normal
operators.
Theorem 1.5.5 Suppose M,T,N ∈ B (H) , M and N are normal, T is invertible, and
M = TNT−1. (1.5)
If T = UP is the polar decomposition of T, then
M = UNU−1. (1.6)
Two operators which satisfy (1.5) are usually called similar. If U is unitary and (1.6)
holds, M and N are said to be unitarily equivalent.
1.6 Numerical range
In this section we will study the basic properties of the numerical range of an operator.
As the numerical range and radius of an operator are intimately connected, we will draw
more information about the numerical radius in this section.
We begin with the definition of the numerical range.
Definition 1.6.1 Let T ∈ B (H) , The numerical range of T , denoted W (T ), is the non-
empty set
W (T ) = (Tx, x) for some ‖x‖ = 1
Proposition 1.6.2 Let T, S ∈ B (H) , then
(1) W (T ∗) = W (T )∗ .
27
(2) W (T ) contains all of the eigenvalues of T .
(3) If U ∈ B (H) is unitary then W (UTU∗) = W (T ) .
(4) W (T ) ⊂ R if and only if T is self-adjoint.
(5) If H is finite dimensional, W (T ) is closed and thus compact.
(6) W (T + S) ⊂ W (T ) +W (S) .
Theorem 1.6.3 (Toeplitz-Hausdorff [Hau19][Toe18]) Let T ∈ B (H) , then W (T )
is convex.
Theorem 1.6.4 Let T ∈ B (H) , then σ (T ) ⊂ W (T ).
The numerical radius of an operator T ∈ B (H) on a nonzero complex Hilbert space H
is the nonnegative number
w (T ) = supλ∈W (T )
|λ| = sup‖x‖=1
|(Tx, x)| .
It is ready verified that
w (T ∗) = w (T ) and w (T ∗T ) = ‖T‖2 .
The numerical radius is a norm on B (H). That is, 0 ≤ w (T ) for every T ∈ B (H)
and 0 < w (T ) if T 6= 0, w (αT ) = |α|w (T ) , and w (T + S) ≤ w (T ) + w (S) for every
α ∈ C and T, S ∈ B (H) . However, the numerical radius does not have the operator norm
property in the sense that the inequality w (TS) ≤ w (T )w (S) is not true for all operators
T, S ∈ B (H) . Moreover, the numerical radius is a norm equivalent to the operator norm
of B (H) ,as in the next theorem.
Theorem 1.6.5 Let T ∈ B (H) , then
0 ≤ r (T ) ≤ w (T ) ≤ ‖T‖ ≤ 2w (T ) .
28
Chapter 2
Non-normal operator classes
In this chapter, we will investigate some classes of bounded operators such as hyponormal
operators, normaloid operators, and convexoid operators. We will also be discussing the
relations between them.
2.1 Compact operators
Let X and Y be normed spaces. A linear transformation T : X → Y is compact if its
maps bounded sets into relatively compact subsets of Y. That is, T is compact if T (A)
is compact in Y whenever A is bounded in X. Let B∞ (X, Y ) denote the collection of
all compact linear transformation of a normed space X into a normed space Y so that
B∞ (X, Y ) ⊆ B (X, Y ). Set B∞ (X) = B∞ (X,X) for short, the collection of all compact
operators on a normed space X. B∞ (X) is an ideal of the normed algebra B (X) . That
is, B∞ (X) is a subalgebra of B (X) such that the product of a compact operator with a
bounded operator is again compact.
We assume that the compact operators act on a complex nonzero Hilbert space H,
although the theory for compact operators equally applies (and is usually developed ) for
operators on Banach spaces.
29
Theorem 2.1.1 Let T : H → H be a linear operator , then the following statements are
equivalent :
1) T ∈ B∞ (H)
2) x 0⇒ Txn → 0.
Theorem 2.1.2 ([Kub12]) If T ∈ B∞ (X, Y ) and λ ∈ C\ 0 , then ran (T − λI) is
closed.
Theorem 2.1.3 If T ∈ B∞ (X, Y ) , λ ∈ C\ 0 , and ker (T − λI) = 0, then
ran (T − λI) = H.
Theorem 2.1.4 (Fredholm Alternative) If T ∈ B∞ (X, Y ) and λ ∈ C\ 0 , then
ran (T − λI) is closed and dim ker (T − λI) = dim ker(T ∗ − λI
)<∞.
Theorem 2.1.5 (Fredholm Alternative) Let T ∈ B∞ (X, Y ) and λ ∈ C\ 0, then
λ ∈ ρ (T ) ∪ σp (T ). Equivalently,
σ (T ) \ 0 = σp (T ) \ 0
Corollary 2.1.6 Let T ∈ B∞ (X, Y ).
(a) 0 is the only possible accumulation point of σ (T ).
(b) If λ ∈ σ (T ) \ 0, then λ is an isolated point of σ (T ).
(c) σ (T ) \ 0 is a discrete subset of C.
(d) σ (T ) is countable.
2.2 Hyponormal operators
In this section we will first examine some general properties of hyponormal operators.
Then we continue with a general discussion of a certain growth condition on the resolvent
set which obtains for hyponormal operators.
30
2.2.1 Definitions and properties
Definition 2.2.1 An operator T ∈ B (H) is hyponormal if T ∗T ≥ TT ∗, which is equiva-
lent to the condition ‖T ∗x‖ ≤ ‖Tx‖ . An operator T ∈ B (H) is cohyponormal if its adjoint
is hyponormal . If it is either hyponormal or cohyponormal, then it is called seminormal.
Proposition 2.2.2 Let T ∈ B (H) , then T is hyponormal operator if and only if T ∗T +
2λTT ∗+ λ2T ∗T > 0, for all λ ∈ R.
Proof. Let λ ∈ R and x ∈ H be given. T is hyponormal operator if and only if
‖T ∗x‖ ≤ ‖Tx‖ ⇔ ‖Tx‖2 + 2λ ‖T ∗x‖2 + λ2 ‖Tx‖2 ≥ 0
⇔ 〈Tx, Tx〉+ 2λ 〈T ∗x, T ∗x〉+ λ2 〈Tx, Tx〉 ≥ 0
⇔ 〈T ∗Tx, x〉+ 2λ 〈TT ∗x, x〉+ λ2 〈T ∗Tx, x〉 ≥ 0
⇔⟨(T ∗T + 2λTT ∗ + λ2T ∗T
)x, x⟩≥ 0
⇔ T ∗T + 2λTT ∗ + λ2T ∗T > 0
Remark 2.2.3 If T ∈ B (H) is hyponormal, then (T − λI) is hyponormal for every
λ ∈ C.
Proposition 2.2.4 Let T ∈ B (H), and λ ∈ C, if T is hyponormal and (T − λI)−1 exists,
then (T − λI)−1 is hyponormal.
Proof. Since hyponormality is preserved under translation, we may assume λ = 0. Thus
T ∗T − TT ∗ ≥ 0 and hence
0 ≤ T−1 (T ∗T − TT ∗)T ∗−1 = T−1T ∗TT ∗−1 − I.
Now since A ≥ I implies A−1 ≤ I we have
I − T ∗T−1T ∗−1T ≥ 0
31
and hence (T ∗−1T−1 − T−1T ∗−1
)= T ∗−1
(I − T ∗T−1T ∗−1T
)T−1 ≥ 0
which completes the proof.
Lemma 2.2.5 Let T ∈ B (H), and λ ∈ C , if ker (T − λI) ⊆ ker(T ∗ − λI
), then
(a) ker (T − λI) ⊥ ker (T − νI) whenever ν 6= λ, and
(b) ker (T − λI) reduces T .
Proof. (a) Let x ∈ ker (T − λI) and y ∈ ker (T − νI). Thus Tx = λx and Ty = νy. if
ker (T − λI) ⊆ ker(T ∗ − λI
), then x ∈ ker
(T ∗ − λI
), and so T ∗x = λx. Then
〈νy, x〉 = 〈Ty, x〉 = 〈y, T ∗x〉 =⟨y, λx
⟩= 〈λy, x〉
and hence
(λ− ν) 〈y, x〉 = 0
which implies that 〈y, x〉 = 0 whenever ν 6= λ.
(b) If x ∈ ker (T − λI)⊆ ker(T ∗ − λI
), then Tx = λx and T ∗x = λx. Thus ker (T − λI)
is T ∗-invariant. But ker (T − λI) is T -invariant. Therefore ker (T − λI) reduce T.
Corollary 2.2.6 if T ∈ B (H) is hyponormal, then
(a) ker (T − λI) ⊥ ker (T − νI) whenever ν 6= λ, and
(b) ker (T − λI) reduce T.
Theorem 2.2.7 If λγγ∈Γ is a (nonempty) family of distinct complex numbers (where
Γ is nonempty index set), and if T ∈ B (H) is hyponormal, then the topological sum
M =
(∑γ∈Γ
ker (T − λγI)
)
reduces T , and the restriction of T to it, T |M ∈ B (M), is normal.
32
Proposition 2.2.8 ([Kub12]) T ∈ B (H) is hyponormal if and only if ‖T ∗x‖ ≤ ‖Tx‖
for every x ∈ H. Moreover, the following assertions are pairwise equivalent
(1) T is normal.
(2) T n is normal for every positive integer n ∈ N.
(3) ‖T ∗nx‖ = ‖T nx‖ for every x ∈ H and every n ∈ N.
2.2.2 Some conditions implying normality or self-adjointness
In this subsection, we will give some important classical results related to hyponormal
operators.
Theorem 2.2.9 ([Sta65]) If T is hyponormal and σ (T ) is an arc, then T is normal.
Corollary 2.2.10 If T is hyponormal and σ (T ) is real, then T is self-adjoint.
Corollary 2.2.11 If T is hyponormal and σ (T ) lies on the unit circle, then T is unitary.
Definition 2.2.12 An operator T is quasi-normal if (T ∗T )T = T (T ∗T ). An operator
T on a Hilbert space H is subnormal if there exists a Hilbert space K, K ⊇ H, and a
normal operator S defined on K with Tx = Sx for x ∈ H.
Remark 2.2.13 One has the following inclusion relation for classes of operators:
Normal ⊂ Quasi-normal ⊂ Subnormal ⊂ Hyponormal.
Theorem 2.2.14 If T is quasi-normal and σ (T ) has no interior, then T is normal.
Theorem 2.2.15 Let T be hyponormal with λ ∈ ρ (T ). Then
∥∥(T − λI)−1∥∥ ≤ 1
d (λ, σ (T ))
or, equivalently, ‖(T − λI)‖ ≥ d (λ, σ (T )) , where d (λ, σ (T )) = min |λ− w| : w ∈ σ (T )
33
Proof. Let λ ∈ ρ (T ) , x ∈ H and, ‖x‖ = 1, then∥∥(T − λI)−1 x∥∥ ≤ ∥∥(T − λI)−1
∥∥ = max|w| : w ∈ σ
((T − λI)−1)
= 1min|w| : w∈σ(T−λI)
= 1min|w−λ| : w∈σ(T )
= 1d(λ,σ(T ))
It will be convenient to refer to the conclusion of the above theorem by stating that T
satisfies condition G1 ; i.e.the resolvent of T has exactly first order rate of growth with
respect to the spectrum of T.
Theorem 2.2.16 ([Nie62]) If T satisfies condition G1 and σ (T ) is real, then T is self-
adjoint.
Theorem 2.2.17 ([Don63]) If T satisfies condition G1 and σ (T ) lies on the unit circle,
then T is unitary
Theorem 2.2.18 ([Sta65]) If T satisfies condition G1 and σ (T ) is a finite set of points,
then T is normal.
However, if T is compact and satisfies condition G1 , T need not be normal. We will
sketch a simple example to illustrate this. The operator
T1 =
0 1
0 0
does not satisfies G1 .
We will now define an operator T2 in such a manner that T = T1 ⊕ T2 does satisfy
condition G1 and moreover is compact. Let fi∞i=1 be an orthogonal basis for H2. We
now set T2fi = aifi where the ai’s are complex numbers placed on circles concentric to
the origin with suffi cient density to ensure that
mini
|λ− ai| ≤ |λ|2 : for each, 0 < |λ| < 1
.
34
. This can clearly be done with zero as the only limit point of the ai’s. The operator
T = T1 ⊕ T2 defined on H1 ⊕H2 is completely continuous and satisfies condition G1 by
construction but it is obviously not normal. This example also illustrates that if T
satisfies condition G1, and M is a reducing subspace of T then T |M may not satisfy
condition G1.
Theorem 2.2.19 ([Halm50]) Let T ∈ B (H)
T is hyponormal ; T 2 is hyponormal .
S. Berberian has asked whether an operator must be subnormal if all its powers are
hyponormal. In fact J. G. Stampfeli ([Sta65]) gives a negative answer to that question.
Let fi+∞i=−∞ be an orthonormal basis for H and define
Tfi =
fi+1, i ≤ 0
2fi+1, i > 0.
Then T kfi = bi,kfi+k where |bi,k| ≤ |bi+1,k| , so T k is hyponormal for k = 1, 2, . . .. Since
‖Tf0 = T ∗f0‖ but ‖T ∗Tf0‖ 6=∥∥T 2f0
∥∥ ,we must conclude that T is not subnormal.
Theorem 2.2.20 ([Sta62], Theorem 5) Let T be a hyponormal operator with T n = B,
where n is a positive integer and B is a normal operator; then T is normal.
2.2.3 p-Hyponormal operators
The semi-hyponormal operator was first introduced by Professor D. Xia. He also provides
an example of a semi-hyponormal operator which is not hyponormal. In this section we
shall study p-Hyponormal operators for p > 0.
35
Definition 2.2.21 An operator T ∈ B (H) is said to be p-hyponormal if
(TT ∗)p ≤ (T ∗T )p
for a positive number p.
Remark 2.2.22 If p = 1, T is hyponormal and if p = 12T is semi-hyponormal.
The following inequality is called Löwner-Heinz’s inequality.
Proposition 2.2.23 ([Löw34], [Hei51]) Let A,B ∈ B (H) satisfy 0 ≤ B ≤ A and
0 < p < 1. Then Bp ≤ Ap.
By Löwner-Heinz’s inequality, every p-hyponormal operator is q-hyponormal if
0 < p < q. There exists a q-hyponormal operator which is not p-hyponormal if 0 < q < p.
Theorem 2.2.24 (Furuta’s inequality [Fur87]) If A ≥ B ≥ 0, then the inequalities
(BrApBr)1q ≥ B
(p+2r)q
and
A(p+2r)
q ≥ (ArBpAr)1q
hold for p, r ≥ 0, q ≥ 1 with (1 + 2r)q ≥ p+ 2r.
Theorem 2.2.25 (Hansen’s inequality [Han80]) If A ≥ 0 and ‖B‖ ≤ 1, then
(B∗AB)p ≥ B∗ApB
for 0 ≤ p ≤ 1.
Theorem 2.2.26 ([AW99]) Let 0 ≤ p ≤ 1. Let T be a p-hyponormal operator. The
inequalities (T n∗T n) pn ≥ (T ∗T )P ≥ (TT ∗)P ≥
(T nT n
∗) pn
hold for all positive integer n.
36
Proof. Let T = U |T | be the polar decomposition of T . For each positive integer n, let
An =(T n∗T n) pn and Bn =
(T nT n
∗) pn We will use induction to establish the inequalities
An ≥ A1 ≥ B1 ≥ Bn. (2.1)
The inequalities (2.1) clearly hold for n = 1. Assume (2.1) hold for n = k. The induction
hypothesis and the assumption that T is p-hyponormal imply
U∗AnU ≥ U∗A1U ≥ A1.
Let Ck =
(U∗A
kp
k U
) pk
. Hansen’s inequality implies
Ck ≥ U∗AkU ≥ A1.
Thus
Ak+1 =(T ∗
k+1T k) pk+1
=(T ∗(T ∗
kT k)T) pk+1
=
(|T |U∗A
kp
k U |T |) p
k+1
=
(A
12p
1 Ckp
k A12p
1
) pk+1
≥ A1
by Furuta’s inequality. On the other hand, the induction hypothesis implies
Bk ≤ B1 ≤ A1.
37
Thus
Bk+1 =(T k+1T ∗
k+1) pk+1
=
(TB
kp
k T∗) p
k+1
=
(U |T |B
kp
k |T |U∗) p
k+1
= U
(|T |B
kp
k |T |) p
k+1
U∗
= U
(A
12p
1 Bkp
k A12p
1
) pk+1
U∗
≤ UA1U∗
= B1
where the inequality follows from Furuta’s inequality. Therefore,
Ak+1 ≥ A1 ≥ B1 ≥ Bk+1
and hence, by induction, inequalities (2.1) holds for n ≥ 1.
Corollary 2.2.27 ([AW99]) Let 0 ≤ p ≤ 1. If the operator T is p-hyponormal, then T n
is(pn
)-hyponormal.
Concrete examples of non-hyponormal p-hyponormal operators are hard to come by.
In [Xia80], Xia gave an example of a singular integral operator which is semi-hyponormal
but not hyponormal. The above Corollary allows us to give another example of a semi-
hyponormal operator which is not hyponormal. Let A be the operator in Halmos’s book,
thus, A is hyponormal but A2 is not hyponormal. By the above Corollary , A2 is semi-
hyponormal. Moreover, A2n is(
12n
)-hyponormal.
Proposition 2.2.28 Let 0 ≤ p ≤ 1. If T is p-hyponormal and T n is normal, then T is
normal.
38
2.3 Normaloid operators
Definition 2.3.1 An operator T ∈ B (H) is normaloid if r (T ) = ‖T‖ , where
r (T ) = sup |λ| : λ ∈ σ (T ) = limn→∞
‖T n‖1n
is the spectral radius of T.
Proposition 2.3.2 r (T ) = ‖T‖ if and only if ‖T n‖ = ‖T‖n .
Theorem 2.3.3 Every hyponormal operator is normaloid.
Proof. Let T ∈ B (H) be a hyponormal operator on a Hilbert space H.
Claim 1 ‖T n‖2 ≤ ‖T n+1‖ ‖T n−1‖ for every positive integer n.
First note that, for any operator T ∈ B (H) ,
‖T nx‖2 = (T nx, T nx) =(T ∗T nx, T n−1x
)≤ ‖T ∗T nx‖
∥∥T n−1x∥∥
for each n ≥ 1 and every x ∈ H. Now if T is hyponormal, then
‖T ∗T nx‖∥∥T n−1x
∥∥ ≤ ∥∥T n+1x∥∥∥∥T n−1x
∥∥ ≤ ∥∥T n+1∥∥∥∥T n−1
∥∥ ‖x‖2
and hence for each n ≥ 1
‖T nx‖2 ≤∥∥T n+1
∥∥∥∥T n−1∥∥ ‖x‖2
which ensures the claimed result, thus completing the proof of Claim 1.
Claim 2 ‖T n‖ = ‖T‖n for every n ≥ 1.
The above result holds trivially if T = 0 and it also holds trivially for n = 1. Let T 6= 0
and suppose the above result holds for some integer n ≥ 1. By Claim 1 we get
‖T‖2n = (‖T‖n)2
= ‖T n‖2 ≤∥∥T n+1
∥∥∥∥T n−1∥∥ ≤ ∥∥T n+1
∥∥ ‖T‖n−1 .
Therefore, as ‖T n‖ ≤ ‖T‖n , and since T 6= 0 ,
‖T‖n+1 = ‖T‖2n (‖T‖n−1)−1 ≤∥∥T n+1
∥∥ ≤ ‖T‖n+1 .
39
Hence ‖T n+1‖ = ‖T‖n+1 . Then the claimed result holds for n + 1 whenever it holds for
n, which concludes the proof of Claim 2 by induction.
Therefore ‖T n‖ = ‖T‖n for every integer n ≥ 1 by Claim 2, and so T is normaloid.
Since ‖T ∗n‖ = ‖T n‖ for each n ≥ 1, it follows that r (T ∗) = r (T ) . Thus an operator T
is normaloid if and only if its adjoint T ∗ is normaloid, and so every seminormal operator
is normaloid.
Proposition 2.3.4 An operator T is normaloid if and only if
‖T‖ = sup‖x‖=1
|〈Tx, x〉| .
Definition 2.3.5 For a compact convex subset X of the plane, a point λ ∈ X is bare if
there is a circle through λ such that no points of X lie outside this circle.
Theorem 2.3.6 ([SY65]) Let T be an operator such that (T − λI) is normaloid for
every complex number λ , then we have
W (T ) = co σ (T ) .
Where co σ (T ) denotes the convex hull of σ (T ) .
To prove the theorem stated above we need the following lemma.
Lemma 2.3.7 Let T be an operator and λ ∈ W (T ) a bare point of W (T ), then there
exists a complex number λ0 satisfying
|λ− λ0| = sup|µ− λ0| : µ ∈ W (T )
.
Lemma 2.3.8 Let C be a nonempty compact convex subset of the plane, and let S be the
collection of all of its bare points. Then C is the closed convex hull of S.
For convenience we state the following known result as a lemma.
40
Lemma 2.3.9 ([Orl63]) For an operator T , and λ ∈ W (T ) , and ‖λ‖ = ‖T‖ imply
λ ∈ σ (T ) .
Proof of Theorem. It is suffi cient to show that each bare point of W (T ) belongs to
σ (T ) ( Lemma 2.3.9 ). Let λ be a bare point of W (T ) , there is a λ0 satisfying
|λ− λ0| = sup|µ− λ0| : µ ∈ W (T )
by lemma 2.3.8 . Thus, by the hypothesis on T and the fact W (T − λ0) = W (T )− λ0,
we have
‖T − λ0I‖ = |λ− λ0| .
Since λ − λ0 ∈ W (T − λ0), λ − λ0 ∈ σ (T − λ0I) by the above Lemma and so we have
λ ∈ σ (T ) . Hence the proof is completed.
In [Ber62], S. K. Berberian conjectured that the closure of the numerical range of
a hyponormal operator coincides with the convex hull of its spectrum. According to
Theorem 2.3.6, we can give an affi rmative answer to his conjecture.
Corollary 2.3.10 For a hyponormal operator T, W (T ) = co σ (T ) .
Corollary 2.3.11 If an operator T ∈ B (H) is compact and normaloid, then σp (T ) 6= ∅
and there exists λ ∈ σp (T ) such that ‖λ‖ = ‖T‖ .
Theorem 2.3.12 ([Kub12]) Every compact hyponormal operator is normal.
Proof. Suppose T ∈ B (H) is a compact hyponormal operator on a nonzero complex
Hilbert space H. The above corollary says that σp (T ) 6= ∅. Consider the subspace
M =
∑λ∈σp(T )
ker (T − λI)
of Theorem 2.2.7 with λγγ∈Γ = σp (T ). Observe that
σp (T |M⊥) = ∅.
41
Indeed, if there is a λ ∈ σp (T |M⊥), then there exists
0 6= x ∈M⊥ such that λx = T |M⊥ x = Tx
, and so x ∈ ker (T − λI) ⊆ M, which is a contradiction. Moreover, recall that T |M⊥ is
compact and hyponormal . Thus, ifM⊥ 6= 0, then Corollary 2.3.11 says that
σp (T |M⊥) 6= ∅
which is another contradiction. Therefore,M⊥ = 0 so thatM = H , and hence
T = T |H = T |M
is normal according to Theorem 2.2.7 .
2.4 Paranormal operators
In this section we discuss a class of paranormal operators. In [Ist66] this is named an
operator of class (N) . We show that this class includes hyponormal operators and is
included in the class of normaloid operators, also we will give a generalization of
Theorem 2.3.12.
2.4.1 Definitions and properties
Definition 2.4.1 An operator T ∈ B (H) is paranormal if ‖T 2x‖ ≥ ‖Tx‖2 for every
unit vector x in H.
Proposition 2.4.2 Every hyponormal operator is paranormal.
Proof. In fact,
‖Tx‖2 = 〈Tx, Tx〉 = 〈T ∗Tx, x〉 ≤ ‖T ∗Tx‖ ≤∥∥T 2x
∥∥ .
42
Theorem 2.4.3 Let T ∈ B (H) . If T is paranormal then
1) T is normaloid.
2) T−1 is also paranormal if T is invertible.
Lemma 2.4.4 Let T be a paranormal operator, then∥∥T k+1x∥∥2 ≥
∥∥T kx∥∥2 ∥∥T 2x∥∥ (Pk)
for a positive integer k ≥ 1, and every unit vector x in H.
Proof. For the case k = 1∥∥T 2x∥∥2
=∥∥T 2x
∥∥∥∥T 2x∥∥ ≥ ∥∥T 2x
∥∥ ‖Tx‖2
and (P1) is clear. Now suppose that (Pk) is valid for k and we assume‖Tx‖ 6= 0, then∥∥T k+2x∥∥2
= ‖Tx‖2∥∥∥T k+1 Tx
‖Tx‖
∥∥∥2
≥ ‖Tx‖2∥∥∥T k Tx
‖Tx‖
∥∥∥2 ∥∥∥T 2 Tx‖Tx‖
∥∥∥≥
∥∥T k+1x∥∥2∥∥∥ T 3x‖Tx‖
∥∥∥≥
∥∥T k+1x∥∥2 ‖T 2x‖
by of Lemma (??) and (Pk). So (Pk+1) is valid and the proof is complete by the mathe-
matical induction.
Theorem 2.4.5 ([Fur67]) If T is a paranormal operator, then T n is paranormal for
every integer n ≥ 1.
Proof. It is suffi cient to show that if T and T k is paranormal, then T k+1 is paranormal
too. We may assume ‖T 2x‖ 6= 0 , then∥∥T 2(k+1)x∥∥2
= ‖T 2x‖2∥∥∥T 2k T 2x
‖Tx‖
∥∥∥2
≥ ‖T 2x‖2∥∥∥T k T 2x
‖Tx‖
∥∥∥2
≥ ‖Tk+2x‖2‖T 2x‖
≥ ‖Tk+1x‖2‖T 2x‖‖T 2x‖
=∥∥T k+1x
∥∥2
43
by (Pk+1) of Lemma (2.4.4). So T k+1 is paranormal.
There exists a paranormal operator which is not hyponormal. That is, the class of hy-
ponormal operators is properly included in the class of paranormal operators. In [Halm50]
Halmos gives an example of hyponormal operator T such that T 2 is not hyponormal. By
Theorem 2.4.5, this T is paranormal. Hence we get an example of non-hyponormal,
paranormal operator.
In Theorem 2.3.12 we prove that every compact hyponormal operator is necessarily
normal. The following Theorem is a slight generalization of it.
Theorem 2.4.6 ([ISY66]) Let T be a paranormal operator such that T ∗p1T q1 · · ·T ∗pmT qm
is compact for some non-negative integers p1, q1, . . . pm, qm . Then T is necessarily a nor-
mal operator.
2.4.2 k-paranormal operators
Definition 2.4.7 An operator T is k-paranormal, if T satisfies∥∥T k+1x∥∥ ≥ ‖Tx‖k+1
for any x ∈ H with ‖x‖ = 1.
Proposition 2.4.8 If T is paranormal then T is k-paranormal.
Theorem 2.4.9 ([FHN67]) If a paranormal operator T has a compact power T k, then
T is compact. However, this is not true for normaloid operators in general.
Proof. Let us suppose that
‖xα‖ → 0 ( weakly), ‖xα‖ ≤ 1.
Since T is (k − 1)-paranormal, then
∥∥T kxα∥∥ ≥ ‖Txα‖k‖xα‖≥ ‖Txα‖k ,
44
which tells us that Txα converges strongly to 0, since∥∥T kxα∥∥→ 0 by the compactness of
T . Therefore, T is compact.
To prove the remainder half of the theorem, let us put H = `2 . Define an operator T
by
T =
1 0 0 0 0 0 · · ·
0 0 0 0 0 0 · · ·
0 1 0 0 0 0 · · ·
0 0 0 0 0 0 · · ·
0 0 0 1 0 0 · · ·
0 0 0 0 0 0 · · ·
· · · · · · · · · · · · · · · · · · · · ·
with respect to the orthonormal basis
e1 =
1
0
0
0...
, e2 =
0
1
0
0...
, e3 =
0
0
1
0...
, . . .
Then, wen can easily deduce
Tei =
e1 (i = 1)
ei+1 (i = 2j) j = 1, 2, . . .
0 (i = 2j + 1)
Hence
‖T‖ = 1 and T k = P (k ≥ 2)
where P is the projection belonging to the subspace spanned by the scalar multiples of
e1. Therefore, ∥∥T k∥∥ = ‖T‖k = 1
45
for all k, which shows that T is a normaloid.
Since T k = P for k ≥ 2, T k is compact for k ≥ 2, whereas T is not compact since the
range of T contains an infinite orthonormal set ei ; i = 1, 3, 5, . . . . The second half of
the theorem is now proved.
2.5 Convexoid operators
In this section we study the class of convexoid operators, also we show the relation between
normaloid operators and convexoid operators.
Definition 2.5.1 An operator T is called to be convexoid if
W (T ) = co σ (T )
where co σ (T ) denotes the convex hull of σ (T ) .
Definition 2.5.2 An operator T is said to be spectraloid if
w (T ) = r (T )
or equivalently
w (T n) = w (T )n (n ∈ N∗)
where w(T ) and r(T ) mean the numerical radius and the spectral radius of T respec-
tively as follows :
w (T ) = sup |λ| ; λ ∈ W (T ) .
r (T ) = sup |λ| ; λ ∈ σ (T ) .
The following theorem gives a characterization of convexoids operators
Theorem 2.5.3 ([FN71]) An operator T is convexoid if and only if (T − λI) is spec-
traloid for every complex λ.
46
2.5.1 Examples
It is known that there exist convexoid operators which are not normaloid and vice versa
and the classes of normaloids and convexoids are both contained in the class of spec-
traloids, and every hyponormal operator is convexoid.
Example 2.5.4 A normaloid operator need not be convexoid. Let H = C3 with the
Euclidean norm given by
‖f‖ = ‖(f1, f2, f3)‖ = |f1|2 + |f2|2 + |f3|2 .
Let
T =
0 1 0
0 0 0
0 0 1
.
Then Tf = (f2, 0, f3) and ‖T 2‖ = 1. On the other hand
〈Tf, f〉 = f2f1 + f3f3
and consequently,
w (T ) = sup‖f‖=1
∣∣f2f1 + f3f3
∣∣ = 1
by taking f = (0, 0, 1) . Hence T is normaloid.
It can be verified easily that
σ (T ) = 0 ∪ 1
and W (T ) is the closed convex set spanned by the discλ : |λ| ≤ 1
2
and one point 1.
Hence T is not convexoid.
Example 2.5.5 A convexoid operator need not be normaloid. Let x1, x2, . . . be an
orthonormal base for H = `2. Define zn = x2n+1, n = 0, 1, 2, . . . and z∼n = x2n, n =
0, 1, 2, . . . . Every x in H can be written as
x =
∞∑k=−∞
αkzk.
47
Let now define the operator S on H by
Sx =1
2
∞∑k=−∞
αkzk+1
where x =∞∑
k=−∞αkzk. we can check easily that
W (T ) =
λ ∈ C : |λ| ≤ 1
2
.
Let us define the operator
L =
0 0
0 1
on C2.
The operator defined on H ⊕ C2 by
T (f, g) = (Lf, Sg)
yields
W (T ) =
λ ∈ C : |λ| ≤ 1
2
= co σ (T )
T is not normaloid since
‖T‖ = 1 w (T ) =1
2.
Example 2.5.6 An example of non-convexoid, non-paranormal, normaloid operator.
Let T be an infinite matrix of the form
T =
1 0 0 0 · · ·
0 M 0 0 · · ·
0 0 M 0 · · ·
0 0 0 M · · ·
· · · · · · · · · · · · · · ·
where M =
0 0
1 0
.
48
Then it is clear that T is normaloid , non-paranormal because
T 2 =
1 0 0 0 · · ·
0 0 0 0 · · ·
0 0 0 0 · · ·
0 0 0 0 · · ·
· · · · · · · · · · · · · · ·
and ‖T‖n = ‖T n‖ = 1. However the relation ‖T 2x‖ ≥ ‖Tx‖2 does not hold for the unit
vectors e2(0, 1, 0, 0, · · · ), e4(0, 0, 0, 1, 0, 0, · · · ) etc. T is non-convexoid. In factW (T ) is the
closed convex, set spanned by the discλ : |λ| ≤ 1
2
and one point 1, σ (T ) = 0∪1 , so
the convex hull of σ (T ) is the closed unit interval [0, 1] , and this unit interval is properly
included in W (T ).
Example 2.5.7 An example (T. Ando) of non-hyponormal, paranormal convexoid oper-
ator . T. Ando has given the following concrete example as follows : when H is a complex
Hilbert space, K denotes the infinite direct sum of copies of H, i.e. K =∞⊕k=1
Hk (Hk∼= H) .
Given two bounded positive operators A and B on H, the infinite matrix TA,B,n is
defined on K, which assigns to a vector
x = (x1, x2, · · · ) the vector y = (y1, y2, · · · )
49
such that, y1 = 0, yj = Axj−1 (1 < j ≤ n) and yj = Bxj−1 (n < j) , that is ,
TA,B,n =
0
A 0
A .
A .
. .
. .
A 0
B 0
B .
. .
.
T. Ando shows that this operator TA,B,n is paranormal if and only if
AB2A− 2λA2 + λ2 ≥ 0 (λ > 0)
and that it is hyponormal if and only if B2 ≥ A. He observed the operator
T = TA,B,n with A = C12 , B =
(C−
12DC−
12
) 12
where
C =
1 1
1 2
and D =
1 2
2 8
,
then T is paranormal , but the tensor product T ⊗ T is not paranormal. He shows that
this paranormal operator T is convexoid and non-paranormal T ⊗ T is also convexoid.
Example 2.5.8 A normaloid operator need not be convexoid. In H = C3, consider the
operator
T =
1 0 0
0 0 0
0 1 0
.
50
We have, as in example 2.5.4, that σ (T ) = 0, 1 and W (T ) = co σ (T ) , S , where
S =
λ ∈ C : |λ| ≤ 1
2
.
Example 2.5.9 A slight modification of the above example produces a spectraloid operator
that is not normaloid. In H = C3, let the operator
T =
1 0 0
0 0 0
0 2 0
.
We have ‖T‖ = 2 and w (T ) = r (T ) = 1.
Since a convexoid operator is not always a normaloid by the above examples.
2.6 Class A operators
In this section We shall introduce a new class “class A”given by an operator inequal-
ity which includes the class of log-hyponormal operators and is included in the class of
paranormal operators.
Definition 2.6.1 An operator T ∈ B (H) is called a log-hyponormal operator if T is
invertible and
log (TT ∗) ≤ log (T ∗T ) .
Remark 2.6.2 Since log :]0,∞] → R is operator monotone, for 0 < p < 1, every
invertible p-hyponormal operator T , is log-hyponormal.
Definition 2.6.3 ([FIY98]) An operator T belongs to class A if
∣∣T 2∣∣ ≥ |T |2 .
51
We would like to remark that class "A " is named after the "‘absolute" values of two
operators |T 2| and |T | . We call an operator T class A operator briefly if T belongs to
class A. We obtain the following results on class A operators.
Theorem 2.6.4
1) Every log-hyponormal operator is class A operator.
2) Every class A operator is paranormal operator.
The following theorems and lemma play an important role in the proof of the above
theorem.
Theorem 2.6.5 ([FFK93][Fur92]) Let A and B be positive invertible operators. Then
the following properties are mutually equivalent
1) logA ≥ logB.
2) Ap ≥(A
p2BpA
p2
) 12 for all p ≥ 0.
3) Ar ≥(A
r2BpA
r2
) rr+p for all p ≥ 0 and r ≥ 0.
Theorem 2.6.6 (Hölder-McCarthy inequality [McC67]) Let A be a positive oper-
ator. Then the following inequalities hold for all x in H :
1) 〈Arx, x〉 ≤ 〈Ax, x〉r ‖x‖2(1−r) for 0 < r ≤ 1.
2) 〈Arx, x〉 ≥ 〈Ax, x〉r ‖x‖2(1−r) for r ≥ 1.
Lemma 2.6.7 ([Fur95]) Let A and B be invertible operators. Then
(BAA∗B∗)λ = BA (A∗B∗BA)(λ−1)A∗B∗
holds for any real number λ .
Proof of Theorem 2.6.4. 1) Suppose that T is log-hyponormal. T is log-hyponormal
iff
log |T |2 ≥ log |T ∗|2 . (2.2)
52
By the equivalence between (1) and (2) of Theorem 2.6.5, (2.2) is equivalent to
|T |2p ≥(|T |p |T ∗|2p |T |p
) 12 for all p ≥ 0. (2.3)
Put p = 1 in (2.3), then we have
|T |2 ≥(|T | |T ∗|2 |T |
) 12 (2.4)
By Lemma 2.6.7 and |T ∗|2 = TT ∗,(2.4) holds iff
|T |2 ≥ |T |T(T ∗ |T |2 T
)− 12 T ∗ |T |
iff (T ∗ |T |2 T
) 12 ≥ T ∗T
so that ∣∣T 2∣∣ ≥ |T |2
that is, T is class A.
2) Suppose that T is class A , i.e.,
∣∣T 2∣∣ ≥ |T |2 .
Then for every unit vector x in H ,
‖T 2x‖2=
⟨(T 2)
∗T 2x, x
⟩=
⟨|T 2|2 x, x
⟩≥ 〈|T 2|x, x〉2 by (2) of Theorem 2.6.6
≥⟨|T |2 x, x
⟩2
= ‖Tx‖4
Hence we have ∥∥T 2x∥∥ ≥ ‖Tx‖2 for every unit vector x in H
so T is paranormal. Whence the proof of Theorem 2.6.4 is complete.
53
2.6.1 Quasi-class A operators
In this section we introduce quasi-class A operators, denoted QA, satisfying
T ∗ |T 2|T ≥ T ∗ |T |2 T and we prove basic structural properties of these operators. The
quasi-class A operators were introduced , and their properties were studied in [JK06].
Definition 2.6.8 ([JK06]) An operator T ∈ B (H) is quasi-class A if
T ∗∣∣T 2∣∣T ≥ T ∗ |T |2 T.
We denote the set of quasi-class A operators by QA To be shown in the next example,
the class of quasi-class A operators properly contains classes of class A operators.
Example 2.6.9 ([JK06]) First, we consider finite dimensional Hilbert space operators.
Let H = C2 and let
T =
0 0
1 0
.
Then by simple calculations we see that T is not paranormal with the unit vector (1, 0) and
even not normaloid but quasi-class A. There exists an example that T is not paranormal
but quasi-class A and normaloid; if
T =
1 0 0
0 0 0
0 1 0
then T is not paranormal but quasi-class A and normaloid. Now we consider unilateral
weighted shift operators as an infinite dimensional Hilbert space operator. Recall that
given a bounded sequence of positive numbers α : α0, α1, . . . (called weights), the unilateral
weighted shift Wα associated with α is the operator on H = `2 defined by
Wαen = αnen+1 for all n ≥ 0,
54
where en∞n=0 is the canonical orthonormal basis for `2. We easily see that Wα can be
never normal, and so in general it is used to giving some easy examples of non-normal
operators. It is well known that Wα is hyponormal if and only if α is monotonically in-
creasing. Also, straightforward calculations show that Wα is class A if and only if α is
monotonically increasing. It is meaningless to use this characterization for distinguishing
some gaps between hyponormal operators and class A operators. However, for QA oper-
ators, Wα has a very useful characterization. Indeed, simple calculations show that Wα
belongs to QA if and only if
Wα =
0
α0 0
α1 0
α2 0. . . . . .
where α0 is arbitrary and α1 ≤ α2 ≤ α3 ≤ · · · . So if Wα has weights α0 = 2 and αi = 1
i
(i ≥ 1),then Wα is quasi-class A but not normaloid because ‖Wα‖ = 2 6= 1 = r (Wα).
Theorem 2.6.10 ([JK06]) Let T ∈ QA and T not have a dense range. Then
T =
A B
0 0
on H = ran (T ) + ker (T ∗)⊥ ,
where A = T |ran (T ) is the restriction of T to ran (T ) , and A is class A operator.
Moreover, σ (T ) = σ (A) ∪ 0 .
Theorem 2.6.11 ([JK06]) Let T ∈ QA and M its invariant subspace. Then the re-
striction T |M of T toM is also a QA operator.
Theorem 2.6.12 ([JK06]) Let T ∈ QA and (T − λI)x = 0 for some λ 6= 0,then
(T − λI)∗ x = 0.
55
Chapter 3
Similarities involving bounded
operators
The purpose of this chapter is finding any conditions implying that a bounded operator
T is self-adjoint or unitary.
3.1 Introduction
In this section, we talk about similarity in C∗-algebra A, by giving a classical result of
Berberian.
Definition 3.1.1 A unitary element u of C∗-algebra A is said to be cramped if its spec-
trum is contained in some semicircle of the unit circle
σ (u) ⊂eiθ : θ0 < θ < θ0 + π
.
Beck and C.R.Putnam showed that if T is a bounded operator which is unitarily
equivalent to its adjoint T ∗, via cramped unitary operator U, necessarily T is self-adjoint.
The proof in [BP56] utilizes the spectral resolution of U. The purpose of the next theorem
is to give a proof in which the spectral resolution is replaced by an application of the
Cayley transform.
56
Theorem 3.1.2 ([Ber62]) Let A be a C∗-algebra, if u is a cramped unitary element of
A, and z is an element of A such that
uzu∗ = z∗
then z is self-adjoint.
Corollary 3.1.3 ([BP56]) Let U, T ∈ B (H) , if u is a cramped unitary such that
UTU∗ = T ∗
then T ∗ = T.
Theorem 3.1.4 ([McC64]) Let φ be a linear transformation on a algebra with involu-
tion such that ϕ (x) is self-adjoint whenever x is, and such that −1 is not in the point
spectrum of ϕ. Then ϕ (z) = z∗ implies that z = z∗.
It perhaps not immediately apparent that Theorem 3.1.4 implies Theorem 3.1.2,. To
see this consider a Banach algebra A with unit and with involution *. For r, s in A,
denoted by Lr, Ls the linear operators on A defined by
Lr (z) = rz and Ls (z) = zs
the operators Lr, Ls commute and the spectrum of LrLs is contained in λµ : λ ∈ σ (Lr) , µ ∈ σ (Ls) .
This in turn is contained in λµ : λ ∈ σ (r) , µ ∈ σ (s) . Now let u given and let
φ (z) = LuLu∗z = uzu∗. If z is self-adjoint then
φ (z)∗ = uz∗u∗ = uzu∗ = φ (z) .
If u is unitary, then the spectrum of φ is contained in λµ : λ, µ ∈ σ (u) , and if u is
cramped , this set can not contain −1.
57
3.2 Operators similar to their adjoints
3.2.1 Conditions implying self-adjointness of operators
In this section we try to answer the following question :
Question suppose that T is a bounded operator and S is an invertible operator for
which 0 /∈ W (S) and ST = T ∗S, then when does it follow that necessary T is self-adjoint!?
Definition 3.2.1 Let A and B be a bounded operators. We said that A is similar to B
iff
SA = BS
for some bounded invertible operator S.
Theorem 3.2.2 ([She66]) Let T be a bounded hyponormal operator. if S is any bounded
operator for which 0 /∈ W (S), then
ST = T ∗S ⇒ T = T ∗.
For proving this theorem, we need certain results which we formulate in the form of
lemmas.
Lemma 3.2.3 ([Ber61]) Let T be a bounded hyponormal operator and let λ1 ,λ2 ∈
σapp (T ) , such that λ1 6= λ2. if xn and yn are the sequences of unit vectors of H such
that ‖(T − λ1)xn‖ → 0 and ‖(T − λ2) yn‖ → 0, then 〈xn, yn〉 → 0.
Lemma 3.2.4 ([Ber65]) If T is a bounded hyponormal operator, then σ (T ∗) = σapp (T ∗) .
Lemma 3.2.5 ([Sta65]) If T is a bounded hyponormal operator such that σ (T ) is a set
of real numbers, then T is self adjoint.
Lemma 3.2.6 If an operator A is similar to an operator B, then A is bounded below iff
B is bounded below. In other words if A and B are similar, then σapp (A) = σapp (B) .
58
PROOF OF THE THEOREM. Since 0 /∈ W (S), T is invertible. Hence
T = S−1T ∗S and it follows from Lemmas 3.2.4 and 3.2.6 that
σ (T ) = σ (T ∗) = σapp (T ∗) = σapp (T )
Now, it is suffi cient, by virtue of Lemma 3.2.5, to prove that σ (T ) is real. Suppose that
there exists a λ ∈ σ (T ) such that λ 6= λ. Since λ ∈ σ (T ) = σapp (T ) , there exists a
sequence xn of unit vectors such that∥∥(T ∗ − λ)xn∥∥ ≤ ‖(T − λ)xn‖ → 0.
Since 0 /∈ W (S), the relation∥∥(T ∗ − λ)xn∥∥ =∥∥(STS−1 − λ
)xn∥∥ =
∥∥S (T − λ)S−1xn∥∥→ 0
implies that∥∥(T − λ)S−1xn
∥∥ → 0. Hence 〈xn, S−1xn〉 = 〈SS−1xn, S−1xn〉 . Put yn =
S−1xn‖S−1xn‖ , then ‖yn‖ = 1 and 〈Syn, yn〉 → 0 i.e. 0 ∈ W (S), a contradiction. This complete
the proof of the theorem.
Corollary 3.2.7 Let T be a bounded seminormal operator. If for any bounded operator
S, for which 0 /∈ W (S), then
ST = T ∗S ⇒ T = T ∗.
Theorem 3.2.8 ([Wil69]) If T is any operator such that S−1TS = T ∗, where 0 /∈
W (S), then the spectrum of T is real.
Proof. It is enough to show that the boundary of σ (T ) lies on the real axis. Since this is
a subset of the approximate point spectrum of T , it suffi ces to show that if xnn∈N is a
sequence of unit vectors such that (T − λ)xn, then λ is real. This latter assertion follows
from the inequality∣∣(λ− λ) 〈S−1xn, xn〉∣∣ =
∣∣〈(T ∗ − λ)S−1xn, xn〉 −⟨(T ∗ − λ
)S−1xn, xn
⟩∣∣≤ ‖(T ∗ − λ)S−1xn‖+ ‖S−1‖ ‖(T − λ)xn‖
= ‖S−1 (T − λ)xn‖+ ‖S−1‖ ‖(T − λ)xn‖
≤ 2 ‖S−1‖ ‖(T − λ)xn‖
59
and the fact that 0 /∈ W (S) implies 0 /∈ W (S−1).
To recover Theorem 3.2.2, from Theorem 3.2.8, we need only observe that if T is
hyponormal, then W (T ) is the convex hull of σ (T ) .
Corollary 3.2.9 Let T ∈ B (H) be a convexoid operator, if S−1TS = T ∗, where
0 /∈ W (S), then T = T ∗.
Theorem 3.2.10 If T or T ∗ is p-hyponormal operator and S is an operator for which
p > 0, 0 /∈ W (S) and ST = T ∗S, then T is self-adjoint.
Proof. Suppose that T or T ∗ is p-hyponormal. According to Theorem 3.2.8 we get
σ (T ∗) = σ (T ) ⊂ R.
Thus m2 (σ (T )) = m2 (σ (T ∗)) = 0, for the planer Lebesgue measure m2. Now apply
Putnam’s inequality for p-hyponormal operators to T or T ∗ (depending upon which is
p-hyponormal) to get
‖(T ∗T )p − (TT ∗)p‖ ≤ p
π
∫∫σ(T )
r2p−1drdθ = 0.
It follows that T is normal. Since σ (T ∗) = σ (T ) ⊂ R here, T must be self-adjoint.
Lemma 3.2.11 ([JKTU08]) Let T = U |T | be the polar decomposition of a class A
operator T . Then |T |U |T | is semi-hyponormal and
σ (|T |U |T |) =r2eiθ : reiθ ∈ σ (T )
.
Lemma 3.2.12 ([JKTU08]) For any operator T ∈ B (H)
(T ∗ |T |2 T
) 12 ≥ |T |2 ⇐⇒
(T ∗ |T |2 T
) 12 ≥ |T ∗|2
The next Theorem was mentioned in ([JKTU08]).
60
Theorem 3.2.13 Let T be a class A operator. If S is an arbitrary operator for which
0 /∈ W (S) and ST = T ∗S, then T is self-adjoint.
Proof. Since ST = T ∗S and 0 /∈ W (S), then σ (T ) ⊂ R by Theorem3.2.8. Let T = U |T |
be the polar decomposition of T . Then |T |U |T | is semi-hyponormal and
σ (|T |U |T |) =r2eiθ : reiθ ∈ σ (T )
by Lemma 3.2.11 because T is class A operator. On
the other hand, since T is class A operator, the following inequalities hold by Lemma
3.2.12 : ∣∣T 2∣∣ =
(T ∗ |T |2 T
) 12 ≥ |T |2 ⇐⇒
(T ∗ |T |2 T
) 12 ≥ |T ∗|2
Using the M. Ito and T. Yamazaki’s result, we have
|T |2 ≥(|T | |T ∗|2 |T |
) 12 .
Therefore, we have∥∥|T 2| − |T |2∥∥ ≤ ∥∥∥(T ∗ |T |2 T) 12 − (|T | |T ∗|2 |T |) 12∥∥∥
=∥∥∥((|T |T )∗ (|T |T ))
12 − ((|T |T ) (|T |T )∗)
12
∥∥∥Since σ (T ) is real, σ (|T |T ) is also real. Thus m2σ (|T |T ) = 0, where m2 is the planer
Lebesgue measure. Applying the Putnam’s inequality for semi-hyponormal operators:∥∥∥(T ∗T )12 − (TT ∗)
12
∥∥∥ ≤ 1
2π
∫∫σ(T )
drdθ = 0
we have |T 2| − |T |22 because |T |T = |T |U |T | is semi-hyponormal .
Now let
T =
A B
0 0
on ran (T ) + ker (T ∗)
be a 2 × 2 matrix representation of T , and let P be the orthogonal projection onto
ran (T ). Then since T ∗(T ∗T − TT ∗)T = 0, we have P (T ∗T − TT ∗)P = 0. Therefore,
A∗A− AA∗ = BB∗ and hence A is hyponormal. Let
S =
S1 S2
S3 S4
61
then since ST = T ∗S and 0 /∈ W (S), we have 0 /∈ W (S1) and S1T = T ∗S1. Since A is
hyponormal, Ais self-adjoint and hence B = 0. Hence T =
A B
0 0
is self-adjoint.
The following is an extension of Theorem 3.2.13 and I. H. Kim ([Kim06, Corrolary 5])
to the quasi-class A operators.
Theorem 3.2.14 ([JKTU08]) If T is a quasi-class A operator and S is an arbitrary
operator for which 0 /∈ W (S) and ST = T ∗S, then T is self-adjoint.
Theorem 3.2.15 Let S, T be two bounded operators satisfying:
S−1T ∗S = T, S∗ST = TS∗S and 0 /∈ W (S)
then T is self-adjoint.
The following Lemma plays an important role in the proof of the above Theorem.
Lemma 3.2.16 ([Ber64]) Let T be a bounded operator for which 0 /∈ W (T ). Then T is
invertible and the unitary operator T (T ∗T )−12 is cramped.
Proof of Thoerem 3.2.15. Since 0 /∈ W (S), S is invertible. So, let S = UP be
its polar decomposition. Remember that P = (S∗S)12 is positive and U = S(S∗S)−
12 is
unitary. By Lemma 3.2.16 , U is even cramped.
Since S∗ST = TS∗S, we have
P 2T = TP 2 or PT = TP.
Hence we may write
S−1T ∗S = T
⇐⇒ P−1U∗T ∗UP = T
⇐⇒ U∗T ∗U = PTP−1
⇐⇒ U∗T ∗U = TPP−1
⇐⇒ U∗T ∗U = T
⇐⇒ T ∗ = UTU∗.
62
As U is cramped, Theorem 3.1.2 applies and yields the self-adjointness of T , establishing
the result.
3.2.2 Operators similar to self-adjoint ones
Let T be a bounded linear operator on a Hilbert space. In this section a suffi cient con-
ditions are given in order that S−1TS is self-adjoint for some bounded invertible linear
operator S .
Theorem 3.2.8, fails if the operator S is merely required to be invertible. Even normality
of both S and T does not help, as the following example shows :
Example 3.2.17 Let T be the bilateral shift (Ten = en+1) on the span `2 of the ortho-
normal set en∞−∞ and let S be the self-adjoint unitary defined by Sen = e−n. Then
S−1TS = T−1 = T ∗
but the spectrum of T is not real.
Here is the promised generalization :
Theorem 3.2.18 If S−1TS = T ∗ where 0 /∈ W (S), then T is similar to a self-adjoint
operator.
Proof. Since 0 /∈ W (S) is convex and does not contain 0, we can separate 0 from W (S)
by a half-plane. If necessary, we may replace S by eiθS for suitably chosen θ to insure
that this half-plane is Re z ≥ ε for some ε > 0. Then, if
A =1
2(S + S∗)
the numerical range of A lies on the real axis to the right of ε, hence A is positive and
invertible. Since
TA = AT ∗
63
it follows that A−12TA
12 is self-adjoint.
The proof of Theorem 3.2.18, shows that if TS = ST ∗ where S is positive and invertible,
then T is similar to a self-adjoint operator. Both assumptions on S are essential here.
Example 3.2.19 The condition S−1TS = T ∗ do not imply that T is normal. It suffi ces
to take T = SB where S is positive, B is self-adjoint, and S and B do not commute.
Proposition 3.2.20 Let T ∈ B (H) , if T is similar to a self-adjoint operator, then T is
similar to T ∗ and the similarity can be implemented by an S with 0 /∈ W (S).
Proof. If R−1TR is self-adjoint, then
(RR∗)−1 T (RR∗) = T ∗
and 0 /∈ W (RR∗) because RR∗ is positive and invertible.
3.3 Operators with inverses similar to their adjoints
Let T be a bounded linear operator on a Hilbert space. In this section we give a necessary
and suffi cient conditions are given in order that S−1TS is unitary for some bounded
invertible linear operator S. Also we discuss the case when T is left invertible and similar
to their adjoint.
3.3.1 Operators similar to unitary ones
If P is a positive invertible operator and if TP 2 = P 2T ∗ then P−1TP = PT ∗P−1 is
self-adjoint. Similarly the condition T−1P 2 = P 2T ∗ implies that P−1T−1P = PT ∗P−1 is
unitary. Hence T is similar to a self-adjoint operator (to a unitary operator) if and only
if T and T ∗ are conjugate (T−1 and T ∗ are conjugate) by means of a positive invertible
operator.
64
Theorem 3.3.1 ([SM73]) Let J ∈ B (H), such that J (X∗) = J (X)∗ for all X ∈
B (H) , then J (S) = 0 for some S such that 0 /∈ W (S) if and only if J (A) = 0 for some
positive invertible A.
We have the following important corollaries :
Corollary 3.3.2 ([Wil69]) If S−1TS = T ∗ where 0 /∈ W (S), then T is similar to a
self-adjoint operator.
Corollary 3.3.3 ([SM73]) Let T be an invertible operator, such that S−1T−1S = T ∗
where
0 /∈ W (S), then T is similar to a unitary operator.
For the proof of these corollaries it suffi ces to take
J (X) = i (TX −XT ∗) and J (X) = TXT ∗ − T, respectively.
Definition 3.3.4 Let T ∈ B (H) , we said that T is power bounded if
sup ‖T n‖ : n ∈ N is finite.
Theorem 3.3.5 ([Cas83]) Let T, S ∈ B (H) , such that S is invertible. The following
assertions are equivalent :
(1) T is similar to a unitary operator.
(2) T and S are power bounded.
(3) T is power bounded,(I − λS)−1 exists for |λ| < 1 and
sup
(1− |λ|)∥∥(I − λS)−1
∥∥ : |λ| < 1
is finite.
(4) For each x in H the expressions
sup
(n+ 1)−1
n∑k=0
∥∥T kx∥∥2: n ∈ N
65
and
sup
(n+ 1)−1
n∑k=0
∥∥T ∗kx∥∥2: n ∈ N
are finite, (I − λS)−1 exists for |λ| < 1 and
sup
(1− |λ|)∥∥(I − λS)−1
∥∥ : |λ| < 1
is finite.
(5) For each x in H the expressions
sup
(n+ 1)−1
n∑k=0
∥∥T ∗kx∥∥2: n ∈ N
and
sup
(n+ 1)−1
n∑k=0
∥∥Skx∥∥2: n ∈ N
are finite.
(6) For |λ| < 1 the inverses (I − λT )−1 and (I − λS)−1 exist and for every x and y in H
the expression
sup
(1− r2) +π∫−π
∣∣∣⟨(1− re−iθT)−1 (1− reiθS
)−1x, y⟩∣∣∣ dθ : 0 ≤ r < 1
is finite
Theorem 3.3.6 If T is an operator such that T ∗ = U∗T−1U , where U is a cramped
unitary operator, then T is unitary.
Proof. From T ∗ = U∗T−1U we have
UT ∗ = T−1U (3.1)
Now by taking the inverses, we get UT ∗−1 = TU . Again by taking the adjoints, we have
UT−1 = T ∗U , and hence
T ∗U2 = U2T ∗.
66
It follows by an argument similar to that of W. A. Beck and C. R. Putnam [BP56] that
UT ∗ = T ∗U (3.2)
Hence from (3.1) and (3.2), T ∗U = T−1U which implies that T is unitary.
The following assertion appears as Corollary 3 in [SM73]
(∗) Let S, T ∈ B (H) with T invertible, if
S−1T−1S = T ∗ (3.3)
where 0 /∈ W (S), then
T is normaloid⇒ T is unitary.
Deprima constructs a suitable counterexample to the above assertion. First note that by
Corollary 3.3.3, (3.3) holds iff T is similar to a unitary. Consequently (3.3) implies that
r(T ) = 1 so that T is then normaloid iff ‖T‖ = 1. Thus assume H is separable and let
U0 be the bilateral shift defined by
U0ek = ek+1, k ∈ Z,
where ek∈Z is an orthonormal basis for H. U0 is unitary and
σ (U0) = λ; |λ| = 1 .
Let Q be the positive operator defined by
Qek = αkek+1, k ∈ Z,
with
αk = α−1−k = 2 for k > 0 and α0 = 1.
Set
T0 = Q−1U0Q
67
then
T0ek =
ek+1 for k /∈ 0,−112ek+1 otherwise
Clearly ‖T0‖ = 1, but T0 is not unitary.
Since in the above example σ (T0) = σ (U0) and ‖T0‖ = 1, we see that W (S) =
λ; ‖λ‖ ≤ 1 so that T0 is also convexoid. Hence (∗) is also false when we replace the
assumption T is normaloid by the assumption that T is convexoid.
Theorem 3.3.7 ([DeP74]) Let T, S ∈ B (H) with T invertible. If
(1) S−1T−1S = T ∗ with 0 /∈ W (S),
(2) T is either convexoid or normaloid,
(3) T−1 is either convexoid or normaloid, then T is unitary.
Remark 3.3.8 If one of the conditions (2) and (3) is a normaloid condition, the appli-
cation of Corollary ?? may be replaced by Corollary ?? .
Corollary 3.3.9 Let T, S ∈ B (H) with T invertible. If (1) holds and if T is hyponormal,
then T is unitary.
Corollary 3.3.10 If T, S ∈ B (H) with T invertible. If T−1S = ST ∗ with 0 /∈ W (S),
and T is normal, then T is unitary.
Theorem 3.3.11 Let S, T be two bounded operators satisfying:
S−1T−1S = T ∗, S∗ST = TS∗S and 0 /∈ W (S),
then T is unitary.
Proof. Let S = UP where U is unitary and P is positive (where P = (S∗S)12 ). We then
have
S∗ST = TS∗S =⇒ T−1S∗S = S∗ST−1
68
hence
P 2T−1 = T−1P 2 so that PT−1 = T−1P.
Therefore,
S−1T−1S = T ∗
⇐⇒ P−1U∗T ∗UP = T−1
⇐⇒ U∗T ∗U = PT−1P−1
⇐⇒ U∗T ∗U = T−1PP−1
⇐⇒ U∗T ∗U = T−1
⇐⇒ T ∗ = UT−1U∗.
Since 0 /∈ W (S), U is cramped so that Theorem 3.3.6 applies and gives us T ∗ = T−1,
completing the proof.
3.3.2 Operators with left inverses similar to their adjoints
In this section we discuss the case when T is similar to an isometry.
Theorem 3.3.12 ([Pat73]) If T is a left invertible operator with a left inverse T1 and
if there exist an operator S such that S−1T1S = T ∗ and 0 /∈ W (S), then T is similar to
an isometry.
Proof. .Since 0 /∈ W (S), 0 can be separated from W (S) by a half plane. If necessary,
we may replace S by Seiθ for suitable θ, so that this half plane lies strictly on the right of
the imaginary axis. Let A = 12
(S + S∗); then A is positive and invertible. Let A12 denote
the positive square root of A. Then A12 is invertible. Now
T1AT∗1 = 1
2(T1ST
∗1 + T1S
∗T ∗1 )
= 12
(ST ∗T ∗1 + TTS∗)
= 12
(S + S∗)
= A
69
Thus
T1AT∗1 = A. (3.4)
Let B = A12T ∗1A
− 12 , then
B∗B =(A
12T ∗1A
− 12
)∗ (A
12T ∗1A
− 12
)= A−
12T1AT
∗1A
12
= I
by using (3.4). Since
B = A12S∗−1TS∗A−
12 =
(S∗A−
12
)−1
T(S∗A−
12
),
T is similar to an isometry.
The preceding Theorem may not be true if S is merely invertible, and it does not
satisfy the condition 0 /∈ W (S). This can be seen by the following example.
Example 3.3.13 Let H be a two-dimensional Hilbert space and T be an operator on H
with the matrix √2 1
1√
2
.
Now if S is the invertible operator with the matrix1 0
0 −1
then T ∗ = S−1T−1S. However, T cannot be similar to an isometry as its spectrum does
not lie in the unit disc.
The following is the converse to Theorem 3.3.12.
Theorem 3.3.14 If T is similar to an isometry, then T has a left inverse T1 satisfying
S−1T1S = T ∗ for some operator S with 0 /∈ W (S).
70
Proof. Our hypothesis implies the existence of an invertible operator R such that R−1TR
is an isometry. If we put S = RR∗, then 0 /∈ W (S) and T1 = ST ∗S−1 is a left inverse of
T .
Our next result shows that a suitable restriction on T in Theorem 3.3.12 guarantees T
to be an isometry.
Theorem 3.3.15 Let T be a left invertible normaloid operator satisfying the hypothesis
of theorem 3.3.12. If T1 is also normaloid then T is an isometry.
Proof. By Theorem 3.3.12, T and T1 are similar to an isometry and coisometry respec-
tively. Consequently, r(T ) = 1 and r(T1) = 1. Since T and T1 are normaloid, ‖T‖ = 1
and ‖T1‖ = 1. Then for any x in H,
‖x‖ = ‖T1Tx‖ ≤ ‖Tx‖ ≤ ‖x‖ .
Thus ‖Tx‖ = ‖x‖ for every x in H, which shows that T is an isometry.
If T1 is not normaloid in theorem 3.3.15, then T may not be an isometry. We mention
the following example to illustrate this point.
Example 3.3.16 Let ei∞i=1 be an orthonormal basis for H. Define T as follows :
Te1 =1
2e2 and Ten = en+1 for n 6= 1.
Then T is a left invertible operator with left inverse T1 defined as follows :
Ten =
Te1 = 0
Te2 = 2e1
Ten = en−1 for all n 6= 1 and n 6= 2
Clearly T is normaloid. Since ‖T1‖ = 2 and r(T1) = 1, T1 is not normaloid. Also it can
be seen that T is not an isometry. However, if S is a diagonal operator with diagonal
−1, 1, 1, . . . , then S−1T1S = T ∗ with 0 /∈ W (S).
71
Theorem 3.3.17 ([Pat73]) Let T be a left invertible operator with the polar decompo-
sition UP , where U is an isometry and P is positive and invertible. Let T1 be the left
inverse of T with the polar decomposition P−1U∗. If, for a cramped unitary operator V ,
T ∗ = V ∗T1V , then T is an isometry.
The next Theorem was proved by S.M.Patel in 1973.
Theorem 3.3.18 Let T be a left invertible operator with a left inverse T1. If there exists
an operator S such that T ∗ = S−1T p1S, 0 /∈ W (S) and p is a nonnegative integer, then
σ (T ) lies in the unit disc.
Although Theorem 3.3.18 is an extension of Theorem 3.3.12, it does not generalize it,
for if p = 1, the conclusion of the theorem cannot say whether T is similar to an isometry.
However, we have
Theorem 3.3.19 Let T be a left invertible operator with a left inverse T1. If there exists
a self-adjoint operator S such that T ∗ = S−1T p1S where 0 /∈ W (S) and p is a nonnegative
integer, then T is similar to an isometry.
Corollary 3.3.20 Let T, S ∈ B (H). If T ∗q = S−1T pS where 0 /∈ W (S) and p, q are
integers, then σ (T p) is real where p 6= q and σ (T ) lies in the unit circle whose p+ q = 0.
Corollary 3.3.21 If T is nilpotent operator such that T p+1 = 0 then T does not satisfy
T ∗ = S−1T pS where 0 /∈ W (S).
The following Theorem is a natural generalization of the result of De Prima in [DeP74].
Theorem 3.3.22 If T and T−1 are spectraloid and T ∗q = S−1T pS where 0 /∈ W (S) and
p 6= q, then T is unitary.
Theorem 3.3.23 If T satisfies the condition T ∗q = S−1T pS for |p| 6= |q| where S is an
invertible self-adjoint operator, then σ (T ) lies in the unit circle.
72
Proof. It is suffi cient to prove the theorem for positive integers p and q. Since
T ∗q = S−1T pS and S−1T ∗qS = T p, we have
S−1T ∗p2
S = T pq and S−1T ∗q2
S = T pq
i.e.
T ∗p2
= T ∗q2
or T p2−q2 = I.
By spectral mapping theorem, λp2−q2 = 1 for each λ ∈ σ (T ) . This gives us that σ (T ) is
a finite set and lies in the unit circle.
3.4 Similarities involving normal operators
The purpose of this section is to discuss Embry’s result.
Theorem 3.4.1 ([Emb70]) If H and K are commuting normal operators and
AH = KA, where 0 is not in the numerical range of A, then H = K.
Proof. Let h and k be the spectral resolutions ofH andK respectively. Since AH = KA,
then
Ah (α) = k (α)
for each complex Borel set α by [Put54]. This last equation together with the fact that
h (α) and k (α) are commuting projections implies that
p (α)Ap (α)∗ = q (α)Aq (α)∗ = 0 (3.5)
for each Borel set α, where
p (α) = (I − h (α))Ah (α)
q (α) = h (α)A (I − h (α))(3.6)
since 0 /∈ W (A), equation (3.5) implies that p (α) = q (α) = 0. Thus by (3.6)
Ah (α) = h (α)A
73
for each Borel set α and consequently, AH = HA. Finally, HA = KA and since 0 /∈
W (A), H = K.
The following two examples show that if H and K are normal and AH = KA, then H
and K may differ if 0 /∈ W (A) or if H and K do not commute, even if A is unitary.
Example 3.4.2 If
K =
1 0
0 2
, A =
0 1
1 0
, and H =
2 0
0 1
then H and K are normal, commute and AH = KA, but H 6= K.
Example 3.4.3 If
K =
1 1
1 2
, A =
1 0
0 i
, and H =
1 i
−i 2
then H and K are normal, AH = KA and 0 /∈ W (A), but H 6= K.
Corollary 3.4.4 ([Emb66]) If AA∗ and A∗A commute and 0 /∈ W (A), then A is nor-
mal.
Proof. Let H = A∗A, K = AA∗ and note that AH = KA, so that Theorem 3.4.1 is
applicable.
Corollary 3.4.5 If 0 /∈ W (A) and there exist real numbers r and s such that r2 + s2 6= 0
and A commutes with rAA∗ + sA∗A, then A is normal.
Several special cases of Corollary 3.4.5 are known. If A is quasi normal and 0 /∈ W (A),
then A is normal [Bro53] . If A commutes with AA∗ − A∗A, then A is normal [Put54].
This last follows from Corollary 3.4.5 by applying the corollary to A−λI,which commutes
with
(A− λI) (A− λI)∗ − (A− λI)∗ (A− λI)
for λ /∈ W (A) .
74
Corollary 3.4.6 ([Put57]) If A2 is normal and 0 /∈ W (A), then A is normal.
Theorem 3.4.7 ([Emb68]) Let A ∈ B (H) . If σ (A) ∩ σ (−A) = ∅, then A and A2
commute with exactly the same operators.
Let D be the set of all bounded operators A for which either 0 /∈ W (A) or σ (A) ∩
σ (−A) = ∅.
Corollary 3.4.8 If A ∈ D and AE = −EA, where either A or E is normal, then E = 0.
Proof. If σ (A) ∩ σ (−A) = ∅, then by Theorem 3.4.7 AE = EA since A2E = EA2.
Therefore E = 0. Assume now that 0 /∈ W (A). If E is normal, we apply Theorem 3.4.1
and have E = −E or E = 0. If A is normal, then A∗E = −EA∗ , and thus
A (E − E∗) = − (E − E∗)A
Since E−E∗ is normal,E = E∗ by Theorem 3.4.1. Consequently,E is normal and a second
application of Theorem 3.4.1 yields E = −E = 0.
Corollary 3.4.9 If A is a normal element of D, then A and A2 commute with exactly
the same operators.
Proof. Assume that A2E = EA2 and let H = AE − EA. Then
AH = −HA
and by Corollary 3.4.8, H = 0.
Corollary 3.4.10 If AE = E∗A and AE∗ = EA, where A ∈ D, then E is self-adjoint.
Corollary 3.4.11 If AE = E∗A, where A ∈ D and either A is unitary or E is normal,
then E is self-adjoint.
75
Proof. If E is normal, then AE∗ = EA, if A is unitary, then
EA∗ = A∗E∗
and consequently,
AE∗ = EA.
Thus in either case Corollary 3.4.10 may be applied.
Corollary 3.4.11 includes a slight improvement of a result of J. P.Williams. In [Wil69]
Williams proved that σ (E) is real if AE = E∗A, where 0 is not in the closure of W (A).
Thus if E is normal, E is self-adjoint. In particular, Williams noted that if E is normal
and AE = E∗A, where A is a cramped unitary operator, then E is self adjoint. More
generally, in [BP56] W. A. Beck and C. R. Putnam and in [Ber62] S. K. Berberian proved
this same result without the hypothesis that A is normal. Finally, in [McC64] C. A.
McCarthy obtained a generalization from which it follows that if AE = E∗A, A unitary
and σ (A)∩σ (−A) = ∅, then E is self-adjoint. All of these results are included in Corollary
3.4.11.
Corollary 3.4.12 Let AH = KA and A∗H = KA∗, where A ∈ D. If A is unitary or H
and K are normal, then H = K.
Corollary 3.4.13 Let H,K ∈ B (H) , such that HH∗ = KK∗ and H∗H = K∗K. If
there exists an element A of D such that AH = KA and A∗H = KA∗, then H = K.
3.5 Quasi-similarity of operators
Two operators A : H −→ H and B : K −→ K are quasi-similar provided there exist
quasi-invertible operators X : H −→ K and Y : K −→ H such that XA = BX and
Y B = AY ..
Theorem 3.5.1 If A : H −→ H is invertible, B : K −→ K is hyponormal, and
X : H −→ K has dense range and satisfies XA = BX, then B is invertible.
76
Theorem 3.5.2 ([Cla75]) Quasi-similar hyponormal operators have equal spectra.
Proof. If A and B are quasi-similar hyponormal operators, then, for any complex number
λ, A− λ and B − λ are also quasi-similar and hyponormal, so by Theorem 3.5.1 they are
both invertible or both non invertible. Thus the spectrum of A is the same as that of B.
(Note that the injectivity of the interwining maps was never used.)
Define the natural map (or the natural quotient map) π : B (H) −→ B (H) /B∞ (H)
by
π (T ) = S ∈ B (H) : S = T +K for some K ∈ B∞ (H) = T + B∞ (H) .
Definition 3.5.3 The essential spectrum (or the Calkin spectrum) σe (T ) of T ∈ B (H)
is the spectrum of π (T ) in unital Banach algebra B (H) /B∞ (H),
σe (T ) = σ (π (T )) .
S. Clary ([Cla75]) proved that quasi-similar hyponormal operators have equal spec-
tra and he asked whether quasi-similar hyponormal operators also have essential spectra.
Later L. R. Williams [Wil80]) showed that two quasi-similar quasi-normal operators and
under certain conditions two quasi-similar hyponormal operators have equal essential spec-
tra, B. C. Gupta ([Gup85]) showed that quasi-similar k-quasihyponormal operators have
equal essential spectra and L. Yang ([Yan93])proved that quasi-similar M -hyponormal
operators have equal essential spectra, and R. Yingbin and Y. Zikun ([YZ99]) showed
that quasi-similar p-hyponormal operators have also equal spectra and essential spectra.
Very recently, I. H. Jeon, J. I. Lee and A. Uchiyama ([JLU03]) showed that quasi-similar
injective p-quasihyponormal operators have equal spectra and essential spectra and A. H.
Kim and I. H. Kim ([KK06]) showed that quasi-similar (p, k)-quasihyponormal operators
have equal spectra and essential spectra.
Theorem 3.5.4 If S, T ∈ B (H) are quasi-similar quasi-class A operators, then
σ (S) = σ (T ) and σe (S) = σe (T ).
77
Chapter 4
Similarities involving unbounded
operators
In this chapter, we develop basic concepts and results about general closed operators on
Hilbert space, also we discuss some conditions implying self-adjointness of unbounded
operators. At the end of the chapter we deal with similarity of unbounded operators.
4.1 Preliminaries
Definition 4.1.1 A linear operator from a Hilbert space H1 into H2 is a linear mapping
T of a linear subspace of H1 , called the domain of T and denoted by D (T ), into H2.
Let S and T be two linear operators from H1 into H2. By definition we have S = T if
and only if D (S) = D (T ) and S(x) = T (x) for all x ∈ D(S) = D(T ). We shall say that T
is an extension of S or that S is a restriction of T and write S ⊆ T , when D (S) ⊆ D (T )
and S(x) = T (x) for all x ∈ D(S).
78
The complex multiple αT for α ∈ C, α 6= 0, and the sum S+T are the linear operators
from H1 into H2 defined by
D (αT ) = D (T ) , (αT )x = αTx for x ∈ D (T ) .
D (S + T ) = D (S) ∩D (T ) , (S + T )x = Sx+ Tx for x ∈ D (S + T ) .
If R is a linear operator from H2 into a Hilbert space H3, then the product RT is the
linear operator from H1 into H3 given by
D (RT ) = x ∈ D (T ) : Tx ∈ D (R) , (RT )x = RTx or x ∈ D (RT ) .
It is easily checked that the sum and product of linear operators are associative and
that the two distributivity laws
(S + T )Q = ST + TQ and R (S + T ) =⊇ RS +RT
hold, where Q is a linear operator from a Hilbert space H0 into H1.
Definition 4.1.2 The graph G (T ) of an operator T from H1 into H2 is the set
G (T ) = (x, Tx) : x ∈ D (T ) .
Obviously, the relation S ⊆ T is equivalent to the inclusion G (S) ⊆ G (T ) .
Proposition 4.1.3 Let T be a linear operator on H such that D(T ) is dense in H. If
〈Tx, x〉 = 0 for all x ∈ D(T ), then Tx = 0 for all x ∈ D(T ).
Definition 4.1.4 An operator T is called closed if its graph G (T ) is a closed subset of
the Hilbert space H × H, and T is called closable (or preclosed) if there exists a closed
linear operator S such that T ⊆ S.
Remark 4.1.5 The operator T is called the closure of the closable operator T .
79
4.1.1 Adjoint Operators
Let (H1, 〈., .〉1) and (H2, 〈., .〉2) be Hilbert spaces. Let T be a linear operator from H1 into
H2 such that the domain D(T ) is dense in H1 . Set
D(T ∗) = y ∈ H2 : There exist u ∈ H1 such that 〈Tx, y〉2 = 〈x, u〉1 for x ∈ D(T ) .
Since D(T ) is dense in H1 , the vector u ∈ H1 satisfying 〈Tx, y〉2 = 〈x, u〉1 for all
x ∈ D(T ) is uniquely determined by y. Therefore, setting T ∗y = u, we obtain a well-
defined mapping T ∗ from H2 into H1. It is easily seen that T ∗ is linear.
Definition 4.1.6 The linear operator T ∗ is called the adjoint operator of T .
Let T be a densely defined linear operator on H. Then T is called symmetric if T ⊆ T ∗.
Further, we say that T is self-adjoint if T = T ∗ and that T is essentially self-adjoint if its
closure T is self-adjoint.
We now begin to develop basic properties of adjoint operators.
Proposition 4.1.7 ([Con90]) Let T : H1 −→ H2 and S : H2 → H3 be linear operators
such that D(ST ) is dense in H1, then
1) If D(S) is dense in H2, then (ST )∗ ⊇ T ∗S∗.
2) If S is bounded and D(S) = H2, then (ST )∗ = T ∗S∗.
3) If T is invertible with inverse T−1 in B(H), then (ST )∗ = T ∗S∗.
Theorem 4.1.8 ([Con90]) Let T be a densely defined linear operator from H1 into H2.
1) T is closable if and only if D(T ∗) is dense in H2.
2) If T is closable, then (T )∗
= T ∗, and setting T ∗∗ := (T ∗)∗, we have
T = T ∗∗.
3) T is closed if and only if T = T ∗∗.
80
4) Suppose that ker (T ) = 0 and ran (T ) is dense in H2, then T ∗ is invertible and
(T ∗)−1 =(T−1
)∗.
5) Suppose that T is closable and ker (T ) = 0. Then the inverse T−1 of T is closable if
and only if ker(T)
= 0. If this holds, then
(T)−1
= (T−1).
6) If T is invertible, then T is closed if and only if T−1 is closed.
Definition 4.1.9 Let T be a closed linear operator on a Hilbert space H. A complex
number λ belongs to the resolvent set ρ (T ) of T if the operator T − λI has a bounded
everywhere on H defined inverse (T − λI)−1 .
The set σ (T ) = C \ρ (T ) is called the spectrum of the operator T.
Definition 4.1.10 σp (T ) := λ ∈ C : ker (T − λI) = 0 is the point spectrum of T.
Remark 4.1.11 Formally, the preceding definition could be also used to define the spec-
trum for a not necessarily closed operator T. But if λ ∈ ρ (T ), then the bounded everywhere
defined operator (T − λI)−1 is closed, so is its inverse T−λI by Theorem 4.1.8, and hence
T . Therefore, if T is not closed, we would always have that ρ (T ) = ∅ and σ (T ) = C
according to Definition 4.1.9, so the notion of spectrum becomes trivial.
By Definition 4.1.9, a complex number λ is in ρ (T ) if and only if there is an operator
B ∈ B(H) such that
B (T − λI)T ⊆ I and TB (T − λI) = I.
Proposition 4.1.12 ρ (T ) is an open subset, and σ (T ) is a closed subset of C.
81
4.1.2 Self-adjoint Operators
Self-adjointness is the most important notion on unbounded operators . The main results
about self-adjoint operators are the spectral theorem and the corresponding functional
calculus based on it.
In this section, T denotes a densely defined linear operator on a Hilbert space H.
Definition 4.1.13 We say that T is normal if D(T ) = D(T ∗) and
‖Tx‖ = ‖T ∗x‖ for all x ∈ D(T ).
It is clear that T is normal if and only if T ∗ is normal. Moreover, we have
ker (T ) = ker (T ∗) for any normal operator T .
Proposition 4.1.14 T is normal if and only if T is closed and T ∗T = TT ∗.
Definition 4.1.15 The operator T is called symmetric if
〈Tx, y〉 = 〈x, Ty〉 for all x, y ∈ D(T ).
Proposition 4.1.16 ([Con90]) T is symmetric if and only if 〈Tx, x〉 is real for all x ∈
D(T ).
Definition 4.1.17 A densely defined symmetric operator T on a Hilbert spaceH is called
self-adjoint if T = T ∗ and essentially self-adjoint, briefly e.s.a., if T is self-adjoint, or
equivalently, if T = T ∗.
A self-adjoint operator T is symmetric and closed, since T ∗ is always closed.
Proposition 4.1.18 If T is a symmetric operator on H such that ran(T ) = H, then T
is self-adjoint, and its inverse T−1 is a bounded self-adjoint operator on H.
Corollary 4.1.19 A closed symmetric linear operator T on H is self-adjoint if and only
if σ (T ) ⊆ R.
82
Theorem 4.1.20 (Fuglede-Putnam) Let S ∈ B(H). Let T1 and T2 be an unbounded
normal operators and
ST1 ⊆ T2S
then
ST ∗1 ⊆ T ∗2S.
Corollary 4.1.21 If S is bounded and T1 and T2 are unbounded and normal, then
ST1 = T2S =⇒ ST ∗1 = T ∗2S.
Theorem 4.1.22 (Fuglede-Putnam-Mortad) Let S be a closed operator with domain
D(S). Let T1 and T2 be two unbounded normal operators with domains D(T1) and D(T2)
respectively. If D(T2) ⊂ D(ST2), then
ST1 ⊂ T2S =⇒ ST ∗1 ⊂ T ∗2S.
Theorem 4.1.23 Let T and S be two normal operators. Assume that S is bounded. If
ST = TS, then ST (and so TS) is normal.
Proof. Since ST = TS, by Corollary 4.1.21 we have ST ∗ = T ∗S. Then we have
(ST )∗ ST = T ∗S∗ST = T ∗S∗TS ⊂ T ∗TS∗S
and
ST (ST )∗ = STT ∗S∗ = TST ∗S∗ = TT ∗SS∗ = T ∗TS∗S.
Whence
(ST )∗ ST ⊂ ST (ST )∗ .
But ST is closed for it equals TS which is closed since is closed and S is bounded.
Therefore, ST (ST )∗ and (ST )∗ ST are both self-adjoint and hence ST is normal.
83
Theorem 4.1.24 Let T be a bounded and invertible operator. Let S be unbounded and
closed. Assume further that D(S) ⊂ D(STS). Then ST and TS are normal iff STT ∗ =
T ∗TS and S∗ST ⊂ TSS∗.
Consider next the following example:
Example 4.1.25 Let T and S be the two operators defined by
Tf(x) = eixf(x) and Sf(x) = ex2−ixf(x)
on their respective domains
D (T ) = L2 (R) and D (S) =f ∈ L2 (R) : ex
2
f(x) ∈ L2 (R).
Then T is unitary (so STT ∗ = T ∗TS is verified) and S is normal. Moreover, we can
easily check that:
D (S∗ST ) =f ∈ L2 (R) : e2x2f(x) ∈ L2 (R)
and
D (TSS∗) = D (SS∗) =f ∈ L2 (R) : e2x2f(x) ∈ L2 (R)
too. Since
S∗STf(x) = TSS∗f(x), ∀f ∈ D (S∗ST ) = D (TSS∗) ,
we have S∗ST = TSS∗. We also see that both TS and ST are normal on their equal
domains
D (TS) = D (ST ) =f ∈ L2 (R) : ex
2
f(x) ∈ L2 (R)
since they are the multiplication operator by the function ex2. Nonetheless we have
D (STS) =f ∈ L2 (R) : e2x2f(x) ∈ L2 (R)
and so D (S) " D (STS) as, for instance, e−
32x2 ∈ D (S) but e−
32x2 /∈ D (STS) .
This example suggests that replacing "bounded and invertible" by "unitary" might allow
us to drop the condition D (S) ⊂ D (STS) there. This is in fact the case and we have
84
Theorem 4.1.26 Let T be a unitary operator. Let S be unbounded and closed. Then ST
and TS are normal iff S∗ST ⊂ TSS∗.
Proof. The proof of suffi ciency is as before. Note that with T assumed unitary, the first
condition of Theorem 4.1.24 is automatically satisfied.
Let us suppose that ST and TS are normal and let us check that S∗ST ⊂ TSS∗. In
fact, since TS is normal, we have
(TS)∗ TS = S∗T ∗TS = S∗S = TS (TS)∗ = TSS∗T ∗.
Hence SS∗T ∗ = T ∗S∗S. Accordingly by taking adjoints,
TSS∗ = S∗ST
establishing the result.
We now turn to the case of two unbounded normal operators. We have
Theorem 4.1.27 Let T be an unbounded invertible normal operator. Let S be an un-
bounded normal operator. If ST = TS, T ∗S ⊂ ST ∗ and S∗T ⊂ TS∗, then ST is normal.
The same method of proof yields
Theorem 4.1.28 Let T be an unbounded invertible normal operator. Let S be an un-
bounded normal operator. If ST ⊂ TS, T ∗S ⊂ ST ∗ and S∗T ⊂ TS∗, then ST is normal.
Corollary 4.1.29 Let T and S be two unbounded invertible normal operators with do-
mains D(T ) and D(S) respectively. If ST = TS and D(T ), D(S) ⊂ D(ST ), then ST
(and TS) is normal.
Proof. Note first that the closedness of ST is clear. Now we have
ST ⊂ TS =⇒ ST ∗ ⊂ T ∗S =⇒ S∗T ⊂ TS∗.
Similarly we have
TS ⊂ ST =⇒ TS∗ ⊂ S∗T =⇒ T ∗S ⊂ ST ∗
So we came back to the setting of Theorem 4.1.24.
85
4.2 Quasi-similarity of unbounded operators
In this section, we generalized some notions of bounded operators to unbounded operators
on Hilbert space, and we extend some familiar results on quasi-similar bounded operators
to unbounded operators.
Definition 4.2.1 (Quasi-similarity) Let T and S be densely defined linear operators
in Hilbert spaces H1 and H2, respectively. If there exist quasi-invertible operators XTS
from H1 into H2 and XST from H2 into H1 such that
XTST ⊂ SXTS and XSTS ⊂ TXST
then we say that T is quasi-similar to S and is denoted by T ∼ S.
Remark 4.2.2 if T ∼ S, then S∗ ∼ T ∗.
Definition 4.2.3 A densely defined operator T is said to be hyponormal if:
1) D (T ) ⊆ D (T ∗) .
2) ‖T ∗x‖ ≤ ‖Tx‖ for all x ∈ D (T ) .
a densely defined linear operator T in a Hilbert space H is said to be subnormal if
there exist a Hilbert space K containing H as a closed subspace and a normal operator
N in K such that
D (T ) ⊆ D (N) and Tx = Nx for all x ∈ D (T ) .
Proposition 4.2.4 A subnormal operator is hyponormal.
Proof. Suppose T is subnormal operator in a Hilbert space H with normal extension N
in K, and let us take x ∈ D (T ) , for y ∈ D (T ) , one has
〈Tx, y〉H = 〈Nx, y〉K = 〈x,N∗y〉K
86
hence y ∈ D (T ∗) .Moreover if P is the orthogonal projection of K onto H, then the above
equalities implies that
T ∗y = PN∗y for all y ∈ D (T ∗) .
It follows that‖Tx‖ = ‖Nx‖
= ‖N∗x‖
≥ ‖PN∗x‖
= ‖T ∗x‖
for all x ∈ D (T ) , thus T is hyponormal.
The following Theorem is an unbounded version of Clary’s result [Cla75] on unbounded
operators.
Theorem 4.2.5 ([OS89]) Let T be a closed hyponormal operator in a Hilbert space H
and let S be a closed densely defined operator in Hilbert space K. If there exists a linear
bounded transformation X with dense range from K to H such that XS ⊂ TX, then the
spectrum of T is contained in that of S.
Corollary 4.2.6 Quasi-similar closed hyponormal operators have equal spectra.
The next Theorem is a generalization of Douglas’ result in [Dou69] to unbounded
operators.
Theorem 4.2.7 ([OS89]) Let T and S be normal operators. If T and S are quasi-
similar, then T and S are unitarily equivalent.
Proof. Let T1(resp. S1) be the closure of T+T ∗
2( resp. S+S∗
2); the real part of T ( resp
. S) and Let T2(resp. S2) be the closure of T−T∗
2i( resp. S−S∗
2i); the imaginary part of T
( resp. S). It then follows from the spectral theory for a normal operator (ex. Theorem
7.32 in [Wei80]) that all Ti, Si (i = 1, 2) are self-adjoint operators satisfying
T1T2 = T2T1 and S1S2 = S2S1
87
and
T = T1 + iT2, S = S1 + iS2.
Let X = XTS be a quasi-invertible transformation with XT ⊂ SX. It then follows from
the Fuglede-Putnam theorem that XT ∗ ⊂ S∗X, and so
XT1 ⊂ S1X, XT2 ⊂ S2X.
Since both of (Ti − iI) and (Si − iI) are (every where defined) boundedly invertible and
X (Ti − iI) ⊂ (Si − iI)X
for i = 1, 2, it follows that
X (Ti − iI)−1 = (Si − iI)−1X
for i = 1, 2.
LetX = UP be the polar decomposition ofX. Clearly U is unitary. Applying the Proof
of Lemma 4.1 in [Dou69], for bounded normal operators (Ti − iI)−1 and (Si − iI)−1 , it
easily seen that they are unitarily equivalent with the common intertwinig operator U,
namely
U (Ti − iI)−1 = (Si − iI)−1 U
for i = 1, 2. Hence, by noticing that U is unitary, one has
UT = U (T1 + iT2)
= S1U + iT2U
= SU.
This completes the proof of the Theorem.
Theorem 4.2.8 ([OS89]) Let T be a closed densely defined symmetric operator, if T is
quasi-similar to its adjoint T ∗, then T is self-adjoint.
88
Proof. Let X be a quasi-invertible operator such that
XT ∗ ⊂ TX.
For y ∈ D (T ∗) with T ∗y = iy, one has X (y) ∈ D (T ) and
iX (y) = XT ∗ (y) = TX (y) .
Since the point spectrum of a symmetric operator is real, it follows that X (y) = 0, by
the injectiveness of X, y = 0. Analogously,
(ran (T − iI))⊥ = ker (T ∗ + iI) = 0 .
hence the deficiency indices of T are (0, 0) . This means that T is self-adjoint.
Remark 4.2.9 In general the above Theorem fails if the closed operator T is merely
normal, and so T is unitarily equivalent to T ∗.
Corollary 4.2.10 Let T be a densely defined, closed symmetric operator. If T is quasi-
similar to a self-adjoint operator S, then T is also self-adjoint, and moreover T and S
are unitarily equivalent.
Proof. Since T ∼ S, then
S ∼ T ∗ and T ∼ T ∗
according to Theorem 4.2.8, T is self-adjoint. By Theorem 4.2.7 T and S are unitarily
equivalent.
Theorem 4.2.11 ([OS89]) Let T be a closed subnormal operator and S be a self-adjoint
operator in a Hilbert space H. If there is a positive, quasi-invertible operator X in H such
that
XS ⊂ TX
then, T = S.
89
Theorem 4.2.12 Let T be a closed subnormal operator in a Hilbert space H, If T is
quasi-similar to a self-adjoint operator S, then T is also self-adjoint, and moreover T and
S are unitarily equivalent.
Proof. Let X be a quasi-invertible operator such that
XS ⊂ TX.
Let X = UP be the polar decomposition of X. Then U is unitary and P is positive. Since
U∗TUP = U∗TX ⊃ U∗XS = PS
and U∗TU is subnormal, It follows by Theorem 4.2.11, that U∗TU = S.
4.3 Similarities involving unbounded normal opera-
tors
In this section we prove and disprove some generalizations of a result about some simi-
larities involving normal operators due to M. R.Embry in 1970. Some interesting conse-
quences are also given.
If N and M are unbounded normal operators, and AN ⊂MA, then
APBR (N) = PBS (M)A
where PBR (N) and PBS (M) are the spectral projections of N and M respectively and
where
BR (N) = z ∈ C : |z| ≤ R and BS (M) = z ∈ C : |z| ≤ S
are two closed balls in C where R and S are two positive numbers.
The numerical range of an operator A, defined on a Hilbert space H, is denoted by
W (A) and is defined as
W (A) = 〈Ax, x〉 : x ∈ D (A) , ‖x‖ = 1 .
90
We adopt the following definition of commutativity of two unbounded normal operators:
Two normal operators are said to commute if their associated spectral projections do.
As introduced by Devinatz-Nussbaum in [DN57], we say that the unbounded operators
N,H and K have the property P if they are normal and if
N = HK = KH.
Devinatz-Nussbaum proved (in the same paper) the following result
Theorem 4.3.1 If N,H and K have the property P , then H and K commute.
Proposition 4.3.2 Assume that H and K are two bounded self-adjoint operators such
that 0 /∈ W (K). If HK is normal, then it is self-adjoint.
Proof. Let N = HK, Since N is normal and
KN = KHK = N∗K
then HK = (HK)∗ .
Now we give the generalization of Embry Theorem to unbounded H and K.
Theorem 4.3.3 ([Mor10]) Assume N,H and K are unbounded operators having the
property P . Also assume that D (H) ⊂ D (K). Assume further that A is a bounded
operator for which 0 /∈ W (A) and such that AH ⊂ KA. Then H = K.
Proof. Let PBR (H) and PBS (K) be the spectral projections of H and K by respectively.
Then PBR (H) and PBS (K) are two bounded normal operators. The property P (more
precisely Theorem 4.3.1) then guarantees that they are commuting operators.
Now since AH ⊂ KA and since ran (PBR (H)) ⊂ D (H) (by the spectral theorem), we
immediately see that
AHPBR (H) = KAPBR (H) .
91
Whence (by the remark below the Fuglede-Putnam theorem in the introduction)
AHPBR (H) = KAPBR (H) = KPBS (K)A.
Therefore we are in a bounded setting and Embry’sTheorem then applies and implies that
HPBR (H)x = KPBS (K)x for all x ∈ D (H) .
Sending both R and S to infinity (in the strong operator topology) gives us
Hx = Kx for all x ∈ D (H) (⊂ D (K)) .
Whence H ⊂ K. Now since normal operators are maximally normal, H = K. The proof
is complete.
An interesting application of the previous theorem is the following.
Corollary 4.3.4 Assume that A is a bounded operator such that 0 /∈ W (A). If H is an
unbounded normal operator such that AH ⊂ H∗A, then H is self-adjoint.
Proof. Obvious since HH∗ = H∗H as H is normal and also since HH∗ is self-adjoint.
Remark 4.3.5 The necessity of 0 /∈ W (A) was justified in [Emb70] by a counterexample.
Now we give an example which shows that Property P cannot be completely eliminated.
Take A = I (the identity operator on the whole Hilbert space). Now take any non-
closed symmetric operator H and hence it is neither self-adjoint nor normal (e.g., take
H such that Hf (x) = −if ′ (x) on D (H) = C∞0 (R)). Then Property P is not fulfilled,
AH ⊂ H∗A, 0 /∈ W (A) but H is not self-adjoint.
Now we give an analog of Proposition 4.3.2 for unbounded H (this is also akin to a
result obtained in [Mor03]). We have
Proposition 4.3.6 Assume that H and K are two self-adjoint operators such that H is
unbounded and 0 /∈ W (K). Assume further that K is bounded. If HK is normal, then it
is self-adjoint.
92
The case where all operators are unbounded fails to be true in general even if A
is assumed to be self-adjoint and even if "⊂ "is replaced by "= "in the assumption
AH = KA. We have
Theorem 4.3.7 Let A,H and K be unbounded operators. Assume that N,H and K have
the property P . Also assume that A is self-adjoint. Then AH = KA and 0 /∈ W (A) do
not necessarily imply that H = K.
Proof. We give a counterexample. Consider the following operators A and H defined by
Af (x) = (1 + |x|) f (x) and Hf (x) = −i (1 + |x|) f ′ (x)
on their respective domains
D (A) =f ∈ L2 (R) : (1 + |x|) f ∈ L2 (R)
and
D (H) =f ∈ L2 (R) : (1 + |x|) f ′ ∈ L2 (R)
.
In order to find H∗ , the adjoint of H, some technical work is required. One has to do it
first for f ∈ C∞0 (R∗) , the space of smooth functions with compact support away from
the origin. Then one has to mimic the arguments used in [Mor03] for slightly different
operators. One finds the following
H∗f (x) = ±f (x)− i (1 + |x|) f ′ (x)
on
D (H∗) =f ∈ L2 (R) : (1 + |x|) f ′ ∈ L2 (R)
.
Now simple calculations yield
AH∗f (x) = HAf (x) = ±i (1 + |x|) f (x)− i (1 + |x|)2 f ′ (x)
for every f in
D (AH∗) = D (HA) =f ∈ L2 (R) : (1 + |x|) f, (1 + |x|)2 f ′ ∈ L2 (R)
.
93
This shows that AH∗ = HA. Now since H is normal (see [Mor09]), then so is H∗
and besides, HH∗ = H∗H (and HH∗ is self-adjoint) and hence Property P is verified.
Obviously A is self-adjoint on D (A) and 0 /∈ W (A).
As one can see, all these assumptions are not suffi cient to make H = H∗ .
4.4 Unbounded operators similar to their adjoints
Theorem 4.4.1 ([DM17]) Let S be a bounded operator on a C-Hilbert space H such
that 0 /∈ W (S). Let T be an unbounded and closed hyponormal operator with domain
D(T ) ⊂ H. If ST ∗ ⊂ TS, then T is self-adjoint.
To prove the theorem stated above we need the following results.
Lemma 4.4.2 Let T be a densely defined and closed operator in a Hilbert space H, with
domain D(T ) ⊂ H. Assume that for some k > 0,
‖Tx‖ ≥ k ‖x‖ for all x ∈ D(T ).
Then ran(T ) is closed.
Proposition 4.4.3 Let T be an unbounded, closed and hyponormal operator in some
Hilbert space H. Then W (T ) ⊂ co σ(T ), where co σ(T ) denotes the the convex hull of
σ(T ).
Proof. The proof is divided into three claims:
Claim 1: σa (T ∗) = σ (T ∗) . By definition, σa (T ∗) ⊆ σ (T ∗) . To To show the reverse
inclusion, let λ /∈ σa (T ∗). Then there exists a positive number k
‖T ∗x− λx‖ ≥ k ‖x‖ for all x ∈ D(T ∗).
Hence T ∗ − λI is clearly injective. Besides, and by Lemma 4.4.2, ran(T ∗ − λI) is closed
as T ∗ − λI is closed for T ∗ is so. Now, since T is hyponormal, so is T − λI which means
94
that such that
∥∥Tx− λx∥∥ ≥ ‖T ∗x− λx‖ ≥ k ‖x‖ for all x ∈ D(T ) ⊂ D(T ∗).
Whence T − λI is also one-to-one so that
ran (T ∗ − λI)⊥ = ker(T − λI
)= 0 or ran (T ∗ − λI) = H.
Thus T ∗ − λI is onto since we already observed that its range was closed. Therefore,
λ /∈ σ (T ∗).
Claim 2: σ(T ) ⊂ R. Let λ ∈ σ (T ∗) = σa (T ∗). Then for some xn ∈ D(T ∗) such that
‖xn‖ = 1 we have ‖T ∗xn − λxn‖ −→ 0 as n −→ ∞. Since ST ∗ ⊂ TS and xn ∈ D(T ∗),
we have ST ∗xn = TSxn so that we may write the following
0 ≤∣∣(λ− λ) 〈Sxn, xn〉∣∣ =
∣∣⟨(ST ∗S−1 − λ+ λ− T)Sxn, xn
⟩∣∣≤ |〈S (T ∗ − λ)xn, xn〉|+
∣∣⟨(λ− T)xn, xn⟩∣∣≤ ‖S‖ ‖T ∗xn − λxn‖+ |〈xn, (λ− T ∗)xn〉|
≤ ‖S‖ ‖T ∗xn − λxn‖+ ‖S‖ ‖T ∗xn − λxn‖
= 2 ‖S‖ ‖T ∗xn − λxn‖ −→ 0.
(where in the second inequality we used the fact that xn ∈ D(T ∗) and Sxn ∈ D(T ) both
coming from ST ∗ ⊂ TS). However, the condition 0 /∈ W (S) forces us to have λ = λ
Accordingly,σ(T ∗) ⊂ R or just σ(T ) ⊂ R (remember that σ(T ∗) =λ : λ ∈ σ(T )
).
Claim 3: T = T ∗. Since σ(T ) ⊂ R, Proposition 4.4.3 implies that W (T ) ⊂ R, which
clearly implies that 〈Tx, x〉 ∈ R for all x ∈ D(T ), which, in its turn, means that T is
symmetric. Hence T is quasi-similar to T ∗ via S and I, so that Theorem 4.2.8 applies
and gives the self-adjointness of T . This completes the proof.
The condition ST ∗ ⊂ TS in the foregoing theorem is not purely conventional, i.e. we
may not obtain the desired result by merely assuming instead that ST ⊂ T ∗S, even with
a slightly stronger condition (i.e. symmetricity in lieu of hyponormality). This is seen in
the following proposition
95
Proposition 4.4.4 There exist a bounded operator S such that 0 /∈ W (S), and an un-
bounded and closed hyponormal T such that ST ⊂ T ∗S whereas T 6= T ∗.
Proof. Consider any unbounded, densely defined, closed and symmetric operator T which
is not self-adjoint. Let S = I, i.e. the identity operator on the Hilbert space. Then S is
bounded and 0 /∈ W (S). Also, T is closed and hyponormal. Finally, it is plain that
T = ST ⊂ T ∗ = T ∗S.
Remark 4.4.5 We have not insisted on the explicitness of the example T in the previous
proof. This was not too important. Besides, there are many of them. For instance, the
interested reader may just consult Exercise 4 on page 316 of [Con90].
Theorem 4.4.6 Let S be a bounded operator such that 0 /∈ W (S). Let T be an unbounded
hyponormal and invertible operator. If ST ⊂ T ∗S, then T is self-adjoint.
Proof. Since T is invertible with an everywhere defined bounded inverse, we have
ST ⊂ T ∗S =⇒ ST−1 =(T−1
)∗S.
Since T is hyponormal, the bounded T−1 too is hyponormal (see [Jan94]). Hence by
[She66], T−1 is self-adjoint. Hence T is self-adjoint.
The condition of invertibility in the foregoing theorem may not simply be dispensed
with. This is illustrated by the following example:
Example 4.4.7 Let A be an unbounded operator defined on a Hilbert space H, with do-
main D(A) H. Set T = A − A, then T is not closed and hence it is surely not
self-adjoint. Finally, let S = I the identity operator on H. Now we verify that the re-
maining conditions (except for invertibility) of the theorem are fulfilled.
1) T is hyponormal for T ∗ = 0 with D(T ∗) = H ⊃ D(A) = D(T ) so that
‖Tx‖ = ‖T ∗x‖ for all x ∈ D(T ).
96
2) Since S = I, obviously 0 /∈ W (S). Moreover,
T = 0D(A) ⊂ T ∗ = 0H so that ST ⊂ T ∗S
Last but not least, we have a very nice and important result which generalizes Berberian
Theorem to unbounded operators.
Theorem 4.4.8 ([DM17]) Let U be a cramped unitary operator. Let T be a closed
operator such that UT = T ∗U . Then T is self-adjoint.
Proof. First we prove that U2T = TU2. Since U is bounded and invertible, we have
(UT )∗ = T ∗U∗ and (T ∗U)∗ = U∗T
(by Proposition 4.1.7). Hence T ∗U∗ = U∗T . We may then write
U2TU∗2 = U (UTU∗)U∗
= UT ∗U∗
= UU∗T ∗
= T ∗,
giving U2T = TU2 or TU∗2 = U∗2T 2 or U2T ∗ = T ∗U2.
Next, we prove that TU = UT ∗. We have
TU = U∗T ∗UU
= U∗UUT ∗
= UT ∗.
Hence also U∗T ∗ = TU∗.
The penultimate step in the proof is to prove that T is normal. To this end, set
S = 12
(U + U∗). Following [Wil69], S > 0.
Then we show that STT ∗ ⊂ T ∗TS. We have
UTT ∗ = T ∗UT ∗ = T ∗TU
97
and
U∗TT ∗ = T ∗U∗T ∗ = T ∗TU∗.
HenceSTT ∗ = 1
2(U + U∗)TT ∗
= 12UTT ∗ + 1
2U∗TT ∗
= 12T ∗TU + 1
2T ∗TU∗
⊂ TT ∗S
So according to Corollary 5.1 in [Sto01], TT ∗ = TT ∗, and remembering that T is taken
to be closed, we immediately conclude that T is normal. Accordingly, and by Corollary
4.3.4 ,
UT = T ∗U
as 0 /∈ W (U), establishing the result.
Remark 4.4.9 Evidently, a hypothesis like UT ⊂ T ∗U would not yield the desired result.
For example, take T to be any symmetric and closed unbounded operator T which is not
self-adjoint. Let U = I be the identity operator on the given Hilbert space. Then clearly
UT ⊂ T ∗U while T 6= T ∗.
Remark 4.4.10 Going back to the previous proof, we observe that this proof may well be
applied to bounded operators. Hence we have just given a new proof of Berberian Theorem
which bypasses the Cayley transform.
Corollary 4.4.11 Let S be a bounded operator and T be an unbounded closed operator
satisfying: S−1T ∗S = T, S∗ST = TS∗S and 0 /∈ W (S). Then T is self-adjoint.
98
Bibliography
[Amb45] W. AMBROSE, Structure theorems for a special class of Banach algebras. Trans.
Amer. Math. Soc. 57 (1945), 364—386.
[And72] T.ANDO, Operators with a norm condition, Acta Sci. Math. Szeged, 33 (1972),
169-178.
[AW99] A. ALUTHGE and D. WANG, Powers of p -hyponormal operators, J. of Inequal.
Appl., 3 (1999 ), 279—284.
[BD73] F. F. BONSALL and J. DUNCAN, Complete Normed Algebras. Ergebnisse der Math-
ematik und ihrer Grenzgebiete 80, Springer Verlag, 1973.
[Ber61] S. K. BERBERIAN, Introduction to Hilbert space, Oxford Univ. Press, New York,
1961.
[Ber62] – – – – – – —, A note on operators unitarily equivalent to their adjoints,
J.London Math. Soc. 37 (1962), 403-404.
[Ber64] – – – – – – —, The numerical range of a normal operator, Duke Math. J., 31,
(1964) 479-483.
[Ber65] – – – – – —, A note on hyponormal operators, Pacific J. Math. 12 (1962), 1871-
1875.
[BP56] W. A. BECK and C. R. PUTNAM, A note on normal operators and their adjoints,
J.London Math. Soc. 31 (1956), 213-216.
99
[Bro53] A. BROWN, On a class of operators, Proc. Amer. Math. Soc. 4 (1953), 723-728.
[Cas83] J.A. VAN CASTEREN, Operators similar to unitary or selfadjoint ones, Pacific
Math. J. 104(1983), 241—255.
[CI95] M. CHO and M. ITOH, Putnam’s inequality for p-hyponormal operators, Proc.
Amer. Math. Soc., 123 (1995) 2435—2440.
[Cla75] S. CLARY, Equality of spectra of quasisimilar hyponormal operators Proc. Amer.
Math. Soc. 53 (1975), 88—90.
[Con90] J. B. CONWAY, A Course in Functional Analysis, Springer, 1990 (2nd edition).
[DeP74] C. R. DEPRIMA, Remarks on "Operators with inverses similar to their adjoints,
Proc. Amer. Math. Soc., 43 (1974) 478-480.
[DM17] S. DEHIMI, M. H. MORTAD, Bounded and unbounded operators similar to their
adjoints, Bull. Korean Math. So c. 54 (2017), No. 1, pp. 215—223
[DN57] A. DEVINATZ, A. E. NUSSBAUM, On the Permutability of Normal Operators, Ann.
of Math. (2), 65 (1957), 144—152.
[Don63] W. F. DONOGHUE, On a problem of Nieminen, Inst. Hautes Études Sei. Publ.
Math, no. 16 (1963),31-33. MR 27 #2864.
[Dou69] R.G. DOUGLAS , On the operator equations S∗XT = X and related topics , Acta
Sci . Math . , 30(1969),19- 2.
[Emb66] M. R. EMBRY, Conditions implying normality in Hilbert space, Pacific. J.
Math.18 (1966), 457-460.
[Emb68] – – – – – , Conditions implying normality in Hilbert space, Pacific. J. Math.
18 (1966), 457-460.
100
[Emb70] – – – – – , Similarities Involving Normal Operators on Hilbert Space, Pacific
J. Math., 35 (1970), 331—336.
[FFK93] M.FUJII, T.FURUTA and E.KAMEI, Furuta’s inequality and its application to
Ando’s theorem, Linear Algebra Appl., 179 (1993), 161-169.
[FHN67] T. FURUTA, M. HORIE and R. NAKAMOTO : A remark on a class of operators.
Proc. Japan Acad., 43, 607-609 (1967).
[FIY98] T. FURUTA, M. ITO and T. YAMAZAKI, A subclass of paranormal operators in-
cluding class of log-hyponormal and several related classes, Sci. Math., 1 (1998),
389—403.
[FN71] T. FURUTA, R. NAKAMOTO, On the numerical range of an operator, Proc. Japan
Acad. 47 (1971), 279—284.
[Fur67] T. FURUTA, On the class of paranormal operators, Proc. Japan Acad. 43 (1967)
594—598.
[Fur87] – – – – , A > B > 0 assures (BrApBr)1q ≥ B
(p+2r)q for r ≥ 0, p ≥ 0, q ≥ 1
with (1 + 2r) q ≥ p+ 2r, Proc. Amer. Math. Soc., 101 (1987), 85-88.
[Fur92] – – – – , Applications of order preserving operator inequalities, Oper. Theory
Adv. Appl., 59 (1992), 180-190.
[Fur95] – – – – , Extension of the Furuta inequality and Ando-Hiai log-majorization,
Linear Algebra Appl., 219 (1995), 139-155.
[Gup85] B. C. GUPTA, Quasisimilarity and k-quasihyponormal operators Math. Today 3
(1985), 49—54.
[Gel41] I. M. GELFAND, Normierte Ringe. Rec. Math. (Mat. Sbornik) 51 (1941), 3—24.
101
[Halm50] P. HALMOS, Normal dilations and extensions of operators, Summa Brasil. Math.
2 (1950), 125-134.
[Hal67] – – – – – – , A Hilbert space problem book, Van Nostrand, Princeton, N.J.,
1967.
[Han80] F. HANSEN, An operator inequality, Math. Ann., 246 (1980), 249-250.
[Hau19] F. HAUSDORFF, Der Wertvorrat einer Bilinearform, Math. Z. 3 (1919), 314—316.
[Hei51] E. HEINZ, Beiträge zur Störungstheorie der Spektralzerlegung, Math. Ann., 123
(1951), 415-438.
[Ist66] V. ISTRATESCU, On some hyponormal operators, Pacific J. Math. 22 (3) (1967)
413—417.
[ISY66] V. ISTRATESCU, T. SAITS, and T. YOSHINO: On a class of operators. TShoku. Math.
Journ., 18, 410-413 (1966).
[Jan94] J. JANAS, On Unbounded Hyponormal Operators, Ark. Mat., 112/1 (1994) 75-82.
[JK06] I. H. JEON, I. H. KIM, On operators satisfying T ∗ |T 2|T ≥ T ∗ |T |2 T.. Linear Alg.
Appl. 418 (2006), 854—862.
[JKTU08] I. H. JEON, I. H. KIM, K. TANAHASHI and A. UCHIYAMA,Conditions implying self—
adjointness of operators, Integral Equations Operator Theory, 61(2008), 549—
557.
[JLU03] I. H. JEON, J. I. LEE and A. UCHIYAMA, On p-quasihyponormal operators and
quasisimilarity Mathematical Inequalities and Applications 6 (2003), 309—315.
[Kat68] T. KATO, Smooth operators and commutators, Studia Math. 31 (1968), 535-546.
[Kim06] I. H. KIM, The Fuglede-Putnam theorem for (p,k)-quasihyponormal operators J.
Inequalities and Applications 9 (2006), 397—403.
102
[KK06] A. H. KIM and I. H. KIM, Essential spectra of quasisimilar (p,k)-quasihyponormal
operators J. Inequalities and Applications 9 (2006), 1—7.
[Kub12] C. S. KUBRUSLY, Spectral theory of operators on Hilbert spaces. New York, NY,
USA: Birkh¨ auser, 2012.
[Löw34] K. LÖWNER, Über monotone Matrixfunktionen, Math. Z., 38 (1934), 177-216.
[McC64] C. A. MCCARTHY, On a theorem of Beck and Putnam, J. London Math. Soc. 39
(1964), 288-290.
[McC67] – – – – – – —, cp, Israel J. Math., 5 (1967), 249-271.
[Mor03] M. H. MORTAD, An Application of the Putnam-Fuglede Theorem to Normal Prod-
ucts of Selfadjoint Operators, Proc. Amer. Math. Soc., 131/10 (2003), 3135—
3141.
[Mor09] – – – – – , On Some Product of Two Unbounded Self-adjoint Operators, Inte-
gral Equations Operator Theory, 64/3 (2009), 399—408.
[Mor10] – – – – – , On a Beck-Putnam-Rehder Theorem, Bull. Belg. Math. Soc. Simon
Stevin, 17/4, (2010) 737-740.
[Na07] N.A.SOMIRY, On Powers of p-hyponormal Operators, The Islamic University of
Gaza, 2007.
[Nag36] M. NAGUMO, Einige analytische Untersuchungen in linearen metrischen Ringen.
Japan J. Math. 13 (1936), 61—80.
[Nie62] T. NIEMINEN, A condition for the self-adjointness of a linear operator, Ann. Acad.
Sei.Fenn. Ser. A I No. 316 (1962),3-5.
[NF70] B. Sz.-NAGY and C. FOIAS, Analyse harmonique des opérateurs de l” espace
de Hilbert, Masson, Paris; Akad. Kiado’, Budapest, 1967; English rev. transl.,
103
NorthHolland, Amsterdam; American Elsevier, New York; Akad. Kiado, Bu-
dapest, 1970.
[Orl63] G. H. ORLAND, On a class of operators, Proc. Amer. Math. Soc, 15(1964), 75-79.
[OS89] S. ÔTA, K. SCHMÜDGEN,On Some Classes of Unbounded Operators, Integral Equa-
tions Operator Theory, 12/2 (1989) 211-226.
[Pat73] S. M. PATEL, Operators with left inverses similar to their adjoints, Proc. Amer.
Math. Sot. 41(1):127-131 (1973).
[Pet08] S. PETRAKIS, Introduction to Banach algebras and the Gelfand-Naimark Theo-
rems. University of thessalonki.
[Pin68] J. D. PINCUS, Commutators and systems of singular integral equations, I,Acta
Math. 121 (1968), 219-249.
[Put54] C. R. PUTNAM, On the spectra of commutators, Proc. Amer. Math. Soc. 5 (1954),
929-931.
[Put57] – – – – – —,On square roots of normal operators, Proc. Amer. Math. Soc. 8
(1957)768-769.
[Put63] – – – – – —, On the structure of semi-normal operators, Bull. Amer. Math. Soc.
(N.S.) 69 (1963), 818-819.
[Put70] – – – – – —, An inequality for the area of hyponormal spectra, Math Z. 116
(1970), 323-330.
[Rus68] M. ROSENBLUM, A spectral theorem for self adjoint singular operators, Amer. J.
Math. 88(1966), 314-328.
[Sch61] M. SCHREIBER, On the spectrum of a contraction, Proc. Amer. Math. Soc.,
12(1961), 709-713.
104
[She66] I. H. SHETH, On hyponormal operators, Proc. Amer. Math. Soc., 17 (1966) 998-
1000.
[SM73] U. N. SINGH and K. MANGLA, Operators with inverses similar to their adjoints,
Proc. Amer. Math. Soc. 38 (1973),258-260.
[Sta62] J.G. STAMPFLI, Hyponormal operators, Pacific J. Math., 12 (1962), 1453-1458.
[Sta65] – – – – – —, Hyponormal operators and spectral density, Trans. Amer. Math.
Soc. 117(1965), 469—476. MR 33:4686.
[Sta67] – – – – – —, Minimal range theorems for operators with their spectra, Pacific J.
Math. 23 (1967), 601-612. MR 37 #4655.
[Sto01] J. STOCHEL, An asymmetric Putnam-Fuglede theorem for unbounded operators,
Proc. Amer. Math. Soc., 129/8 (2001) 2261-2271.
[SY65] T. SAITO and T. YOSHINO, On a conjecture of Berberian. Tohoku. Math. Journ.,
17, 147-149(1965).
[Toe18] O. TOEPLITZ, Das algebraische Analogon zu einem Satz von Fejer, Math. Z. 2
(1918), 187—197.
[Wei80] J. WEIDMANN , Linear operators in Hilbert spaces, Springer-Verlag, Berlin
Heidelberg-New York (1980).
[Wil69] J. P. WILLIAMS, Operators Similar to Their Adjoints, Proc. Amer. Math. Soc.,
20 , (1969) 121-123.
[Wil70] – – – – – —, Finite operators. Proc. Amer. Math. Soc., 21, 129-136 (1970).
[Wil80] L. R. WILLIAMS, Equality of essential spectra of certain quasisimilar seminormal
operators Proc. Amer. Math. Soc. 78(2) (1980), 203—209.
105
[Xia63] D. XIa, On non-normal operators, I, Chinese J. Math. 3 (1963), 232-246.
[Xia80] – – —, On the non-normal operators-semihyponormal operators, Sci. Sininca, 23
(1980), 700-713.
[Yan93] L. YANG, Quasi-similarity of hyponormal and subdecomposable operators J. Func-
tional Analysis 112 (1993), 204—217.
[YZ99] R. YINGBIN and Y. ZIKUN, Spectral structure and subdecomposability of p-
hyponormal operators Proc. Amer. Math. Soc. 128 (1999), 2069—2074.
106