Noncommutative Integration and Operator Theory 1st ed. 2023 edition Edition Peter G. Dodds 2024 scribd download
Noncommutative Integration and Operator Theory 1st ed. 2023 edition Edition Peter G. Dodds 2024 scribd download
Noncommutative Integration and Operator Theory 1st ed. 2023 edition Edition Peter G. Dodds 2024 scribd download
com
https://ebookfinal.com/download/noncommutative-integration-
and-operator-theory-1st-ed-2023-edition-edition-peter-g-
dodds/
OR CLICK BUTTON
DOWNLOAD EBOOK
https://ebookfinal.com/download/elliptic-theory-and-noncommutative-
geometry-1st-edition-vladimir-e-nazaykinskiy/
ebookfinal.com
https://ebookfinal.com/download/operator-theory-and-numerical-
methods-1st-edition-hiroshi-fujita/
ebookfinal.com
https://ebookfinal.com/download/introduction-to-operator-space-theory-
gilles-pisier/
ebookfinal.com
https://ebookfinal.com/download/diabetes-and-cardiovascular-
disease-3rd-ed-2023-edition-michael-johnstone/
ebookfinal.com
https://ebookfinal.com/download/the-extended-field-of-operator-
theory-1st-edition-michael-a-dritschel/
ebookfinal.com
https://ebookfinal.com/download/measure-theory-and-integration-second-
edition-gar-de-barra/
ebookfinal.com
Progress in Mathematics
349
Peter G. Dodds
Ben de Pagter
Fedor A. Sukochev
Noncommutative
Integration and
Operator Theory
Progress in Mathematics
Volume 349
Series Editors
Antoine Chambert-Loir , Université Paris-Diderot, Paris, France
Jiang-Hua Lu, The University of Hong Kong, Hong Kong SAR, China
Michael Ruzhansky, Ghent University, Belgium
Queen Mary University of London, London, UK
Yuri Tschinkel, Courant Institute of Mathematical Sciences, New York, USA
Noncommutative Integration
and Operator Theory
Peter G. Dodds Ben de Pagter
College of Science and Engineering Delft Institute of Applied Mathematics
Flinders University at Tonsley Delft University of Technology
Tonsley, SA, Australia Delft, The Netherlands
Fedor A. Sukochev
School of Mathematics & Statistics
University of New South Wales
Sydney, NSW, Australia
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland
AG 2023
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This book is published under the imprint Birkhäuser, www.birkhauser-science.com by the registered
company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
v
vi Preface
the classical theory of rearrangement invariant Banach function spaces now emerge
as special cases. In turn, this unification yields considerable new insights.
These parallels with the classical theory of function spaces suggest very much
the overall directions and even the general results that might be expected. That
said, the lack of commutativity presents substantial technical barriers. At one end
of the spectrum are the (commutative) function spaces in which techniques from
the theory of vector lattices may be employed. On the other hand, the usual triangle
inequality for the absolute value fails in the simplest noncommutative von Neuman
algebra of all two-by-two matrices which is an anti-lattice with respect to the usual
quadratic form ordering. These facts illustrate the intrinsic difficulty in passing from
the commutative to the noncommutative situation.
The idea to write this monograph arose a number of years ago, perhaps more than
the authors are willing to admit, during collaborative visits of the authors to their
respective institutions. However, it is only relatively recently that some sections of
the material could be considered to be in final form. As is often the case, the possible
material that might be included has also expanded rapidly, and so the choice was
made to present in this volume what the authors see as the indispensable core of
the subject, which can then form the framework for the presentation of further,
though no less interesting and important subject areas, in later volume(s). The
monograph itself is no mere synthesis of existing results from the literature; indeed,
many of the main results have been carefully reworked, often leading to newer and
more transparent proofs and yielding a deeper insight through the development of a
coherent structure.
This book cannot exist and could not have been written without the aid of many
others. Too many to name in fact. Of course, there would be no mathematics to
discuss if it were not for the school of research in noncommutative analysis and
Banach space geometry based in Kazan, Tashkent, Delft, Adelaide, and Sydney.
From the bottom of our hearts, we would like to thank the members of these schools
for their many collaborations and contributions to the theory, to which the authors
have dedicated their professional lives.
Of particular and personal standing, we must acknowledge the contributions
of Ovchinnikov (Voronezh); Bikchentaev and Tikhonov (Kazan); Ayupov, Ber,
and Chilin (Tashkent); Lust-Piquard, Pisier, and Xu (France); Junge (Urbana-
Champaign); Hiai and Kosaki (Japan); and Pietsch (Germany).
Special mention must also be made for those who not only paved the way for our
field, but have since unfortunately passed. Their work will always act as a guiding
light and inspiration for the authors and so many other mathematicians. We pay
tribute to those contributions due to Uffe Haagerup, Nigel Kalton, Wim Luxemburg,
Aleksander Pełczyński, Haskell Rosenthal, and Nicole Tomczak-Jaegermann.
It is only through the dedicated and diligent efforts of Eva-Maria Hekkelman,
Jinghao Huang, and Thomas Scheckter in Sydney, as well as Sjoerd Dirksen and
Anna Kaminska, that the book has been able to be edited and reviewed, and is finally
in a publishable state.
The authors, and in particular Peter, would like to thank Theresa Dodds for her
enduring collaboration and support during this project.
vii
Contents
ix
x Contents
Abstract This chapter reviews some of the basic elements of Hilbert space operator
theory and von Neumann algebras which will be used throughout this book. Most of
these results are presented without proofs, which can be readily found in the relevant
standard literature.
In this chapter we collect some notation and terminology concerning Hilbert space
operators and von Neumann algebras which will be used throughout this book. Most
of the results will be stated without proofs. For further details and proofs, we refer
the reader to [13, 23, 35, 70, 71, 74, 115, 125].
Let H be a complex Hilbert space equipped with the inner product .〈·, ·〉 and
corresponding norm .‖·‖H . The inner product is linear in the first and conjugate
linear in the second variable. The elements of H will usually be denoted by
small Greek letters .ξ, η, ζ, . . .. The space of all bounded linear operators in H is
denoted by .B (H ). The elements of .B (H ) will be denoted by small Latin letters
.x, y, u, v, . . .. The identity operator on .H is denoted by .1 = 1H (which is the unit
element in the algebra .B (H )). Equipped with the operator norm .‖·‖B(H ) , given by
.‖x‖B(H ) = sup‖ξ ‖ ≤1 ‖xξ ‖H , the space .B (H ) is a Banach algebra. The closed unit
H
ball in .B (H ) is denoted by .BB(H ) . For any .x ∈ B (H ) its adjoint is denoted by .x ∗ ,
so .〈xξ, η〉 = 〈ξ, x ∗ η〉 for all .ξ, η ∈ H . The mapping .x │−→ x ∗ is a conjugate linear
involution in .B (H ) satisfying .‖x ∗ ‖B(H ) = ‖x‖B(H ) and .‖x‖2B(H ) = ‖x ∗ x‖B(H ) for
all .x ∈ B (H ) (so, .B (H ) is an example of a .C ∗ -algebra, which we will discuss in
Sect. 1.10).
An operator .x ∈ B (H ) satisfying .x ∗ = x is called self-adjoint (or, Hermitian).
The collection of all self-adjoint operators is denoted by .Bh (H ), which is a real
linear subspace of .B (H ). An operator .x ∈ B (H ) is called normal if .xx ∗ = x ∗ x.
Furthermore, if .u ∈ B (H ) satisfies .u∗ u = uu∗ = 1 (equivalently, .u−1 = u∗ ), then
In addition to the norm topology, generated by the operator norm .‖·‖B(H ) , there
are a number of other important topologies on .B (H ). For every .ξ ∈ H , we define
the semi-norm .ρξ on .B (H ) by .ρξ (x) = ‖xξ ‖H , .x ∈ B (H ). The locally
convex
Hausdorff topology on .B (H ) generated by the family of semi-norms . ρξ : ξ ∈ H
is called the strong operator topology (briefly, so-topology). A net .{xα } in .B (H )
so
so-converges to an operator .x ∈ B (H ), denoted by .xα → x, if and only if
.‖xα ξ − xξ ‖H → 0 for all .ξ ∈ H . Clearly, the so-topology is weaker than the
also holds.
1.3 The Lattice of Projections 3
topology is stronger than the so- and uwo-topologies and is weaker than the norm
topology. On norm bounded subsets of .B (H ), the uso- and so-topology coincide.
Convergence of a net .{xα } to an element x in .B (H ) with respect to the uso-topology
uso
is denoted by .xα → x.
The uwo-dual and uso-dual spaces of .B (H ) are denoted by .B (H )'uwo and
'
.B (H )uso , respectively. The following result is similar to Theorem 1.2.1.
In the sequel we shall frequently use the following notation concerning partial
ordering. Let .(X, ≤) be a partially ordered set. If D is a non-empty subset of X
for which the least upper bound (or, supremum) exists, then this least upper bound
is denoted by .sup D or . D. Similarly, .inf D or . D denotes the greatest lower
bound (or, infimum) of D whenever it exists. In the case that .D = {x, y}, we also
write .sup D = x ∨ y and .inf D = x ∧ y. A net .{xα }α∈∆ in X is called increasing
4 1 A Review of Relevant Operator Theory
whenever .i /= j in I ). For each finite subset F of I we may define the projection
.pF = i∈F pi . It is clear that .{pF } is an increasing net (with respect to the
1.4 Closed Linear Operators 5
R.V. Kadison. However, for increasing nets we have the following important result
(an analogous statement holds for decreasing nets).
Theorem 1.3.3 If .{aα } is an increasing net in .Bh (H ) which is bounded above, then
so
there exists .a ∈ Bh (H ) such that .aα → a and .aα ↑ a in .Bh (H ).
It follows, in particular, from Proposition 1.3.2 and Theorem 1.3.3 that, if .{pα }
is an increasing net in .P (B (H )) such that .pα ↑ p in .P (B (H )), then .pα ↑ p in
.Bh (H ).
Note that, via complementation, similar statements hold for infima in .Bh (H ) and
P (B (H )).
.
Many of the linear operators we shall encounter will not be bounded and are only
defined on a (dense) subspace of the Hilbert space H . Here we shall introduce the
necessary notions to deal with such operators. A linear operator x in H is a linear
mapping from its domain .D (x), which is a linear subspace of H , into the space
H . Given two such linear operators x and y in H , the operator y is said to be an
extension of x (or, x is a restriction of y), if .D (x) ⊆ D (y) and .xξ = yξ for all
.ξ ∈ D (x). This is denoted as .x ⊆ y. If .x ⊆ y as well as .y ⊆ x, then we say
that .x = y. The range and kernel of a linear operator x are denoted by .Ran (x) and
.Ker (x), respectively.
• .λx by setting .D (λx) = D (x) and .(λx) ξ = λ (xξ ) for all .ξ ∈ D (λx).
• .x + y by setting .D (x + y) = D (x) ∩ D (y) and .(x + y) ξ = xξ + yξ for all
.ξ ∈ D (x + y).
It follows from (a) and (b) above that we can form, without ambiguity, sums and
products of an arbitrary number of linear operators. In particular, polynomials in
linear operators are well defined.
For a linear operator x in H the graph .𝚪 (x) is defined to be the linear subspace
of .H × H given by .𝚪 (x) = {(ξ, xξ ) : ξ ∈ D (x)}. Note that .x ⊆ y is equivalent to
.𝚪 (x) ⊆ 𝚪 (y). A linear operator x is called closed if .𝚪 (x) is a closed subspace of
.H × H (equipped with the product topology). In other words, x is closed if and only
a closed linear operator and if the domain .D (x) is a closed subspace of H , then
it follows from the Closed Graph Theorem that x is bounded on its domain .D (x).
This applies, in particular, if .D (x) = H . Furthermore, if x is a closed injective
linear operator in H , then its inverse .x
−1 is also closed. Consequently, if x is in
−1
addition surjective, then .D x = Ran (x) = H and so, .x −1 ∈ B (H ).
The linear operator x in H is called pre-closed (or, closable) if the closure .𝚪 (x)
of the graph .𝚪 (x) in .H × H is the graph of some linear operator in H . Note that
.𝚪 (x) is the graph of a linear operator if, and only if, .(0, η) ∈ 𝚪 (x) implies that
∞
.η = 0. In other words, x is pre-closed if, and only if, it follows from .{ξn }
n=1 ⊆
D (x), .η ∈ H , .ξn → 0 and .xξn → η as .n → ∞, that .η = 0. If a linear operator x
is pre-closed, then the linear operator whose graph is .𝚪 (x) is denoted by .x̄ and is
called the closure of x. Clearly, .x ⊆ x̄.
Suppose that x is a linear operator in H and that .D is a linear subspace of .D (x).
Then .D is called a core of x if .𝚪 (x) ⊆ 𝚪 x|D , where .x|D denotes the restriction
of x to .D. In other words, .D is a core of x if and only if for every .ξ ∈ D (x) there
exists a sequence .{ξn }∞n=1 ⊆ D such that .ξn → ξ and .xξn → xξ as .n → ∞. It is
easily verified that two pre-closed operators which coincide on a common core have
identical closures.
1.4 Closed Linear Operators 7
operator in H , then .Ran (a)⊥ = Ker (a) and .Ker (a)⊥ = Ran (a).
Definition 1.4.4 Given a closed, densely defined linear operator x in H we
define:
(i) The projection onto .Ker (x) is called the null projection of x, denoted by .n (x).
(ii) The projection onto .Ran (x) is called the range projection of x, denoted by
.r (x).
(iii) The projection .1 − n (x), which is the projection onto .Ran (x ∗ ), is called the
support projection, denoted by .s (x).
Observe that .n (x ∗ ) = 1 − r (x), .r (x ∗ ) = s (x) and .s (x ∗ ) = r (x). It is easy to
see that .x = r (x) x = xs (x). In particular, if a is a self-adjoint operator in H , then
.s (a) = r (a) and .a = s (a) a = as (a). The following theorem presents alternative
n=1
1.5 The Spectral Theorem 9
∞
and .xξ = n=1 xn ξ for all .ξ ∈ D (x). The linear operator .x : D (x) → H is
∞
denoted by .⊕∞
n=1 xn and is called the direct sum of .{xn }n=1 . We list some of its
properties.
Proposition 1.4.7 Given the sequence .{xn }∞
n=1 in .B (H ) as above, the following
statements hold:
(i) The operator .⊕∞ ∞
n=1 xn is closed and densely defined; .⊕n=1 xn is bounded if and
only if .supn ‖xn ‖B(H ) < ∞.
(ii) The adjoint of .⊕∞ ∞ ∗
n=1 xn is equal to .⊕n=1 xn .
(iii) If the operators .xn (.n = 1, 2, . . .) are normal, then .⊕∞n=1 xn is normal.
(iv) If the operators .xn (.n = 1, 2, . . .) are self-adjoint, then .⊕∞
n=1 xn is self-adjoint.
As before we assume that .(H, 〈·, ·〉) is a complex Hilbert space. Suppose that .Ω is a
non-empty set and that .A is a .σ -algebra of subsets of .Ω, so .(Ω, A) is a measurable
space.
Definition 1.5.1 A spectral measure on .(Ω, A) is a mapping .e : A → B (H ) such
that:
(i) .e (δ) is a projection in H for each .δ ∈ A.
(ii) .e (∅) = 0 and .e (Ω) = 1.
(iii) .e (δ1 ∩ δ2 ) = e (δ1 ) e (δ2 ) for all .δ1 , δ2 ∈ A.
(iv) If .δj ∈ A (.j = 1, 2, . . .) are pairwise disjoint, then .e ∪∞ j =1 δj =
∞
j =1 e δj , where the series is strongly convergent in .B (H ).
Observe that condition (iii) in particular implies that .e (δ1 ) and .e (δ2 ) commute
for any two .δ1 , δ2 ∈ A. If condition (iv) is satisfied only for finite disjoint collections
.δ1 , . . . , δn (.n ∈ N) in .A, then e is referred to as a finitely additive spectral measure.
A set .δ ∈ A is called an e-null set if .e (δ) = 0. The collection of all e-null sets,
denoted by .Ne , is a .σ -ideal in .A.
Given a spectral measure e on .(Ω, A) and .ξ, η ∈ H , we define the .σ -additive
measure .eξ,η : A → C by .eξ,η (δ) = 〈e (δ) ξ, η〉 for all .δ ∈ A. The variation .eξ,η
of .eξ,η satisfies
. eξ,η (δ) ≤ ‖e (δ) ξ ‖H ‖e (δ) η‖H
for all .δ ∈ A.In particular, the total variation of .eξ,η , which is denoted by .eξ,η ,
satisfies .eξ,η ≤ ‖ξ ‖H ‖η‖H .
10 1 A Review of Relevant Operator Theory
⎧ ⎫
⎨ n ⎬
.sim (Ω, A) = αj χδj : αj ∈ C, δj ∈ A, j = 1, . . . , n; n ∈ N .
⎩ ⎭
j =1
For .s ∈ sim (Ω, A) we denote .‖s‖∞ = maxω∈Ω |s (ω)|. Clearly, .sim (Ω, A) is an
to the pointwise operations. Given a spectral measure .e : A →
algebra with respect
B (H ) and .s = nj=1 αj χδj in .sim (Ω, A), we define
n
. sde = αj e δj .
Ω j =1
The integration mapping .s │−→ Ω sde is an algebra homomorphism from
.sim (Ω, A) into .B (H ) satisfying in addition for all .s ∈ sim (Ω, A) and all
.ξ, η ∈ H :
∗
(a) .Ω s̄de =
Ω sde .
(b) . Ω sdeB(H ) ≤ ‖s‖∞ .
(c) . Ω sde ξ, η = Ω sdeξ,η .
Let .Bb (Ω, A) be the algebra of all bounded complex valued .A-measurable
functions on .Ω. For .f ∈ Bb (Ω, A) let .‖f ‖∞ = supω∈Ω |f (ω)|. With respect
to the norm .‖·‖∞ , the space .Bb (Ω, A) is a Banach algebra and .sim (Ω, A) is
a subalgebra. Now we shall extend the definition of the integral with respect to
a spectral measure e on .(Ω, A) from .sim (Ω, A) to .Bb (Ω, A). For any .f ∈
Bb (Ω, A) there exists a sequence .{sn }∞ n=1 in .sim (Ω, A) such that .‖f − sn ‖∞ → 0
∞
as .n → ∞. From the estimate given in (b) above it follows that . Ω sn de n=1 is
a Cauchy sequence in .B (H ). Consequently, this sequence is convergent in .B (H )
and it is readily verified that its limit depends only on the function f and not on the
choice of the particular sequence .{sn }∞ n=1 . This justifies the following definition.
Definition 1.5.2 Let .e : A → B (H ) be a spectral measure on the measurable
space .(Ω, A). For .f ∈ Bb (Ω, A) the integral with respect to e is defined
∞
by
.
Ω f de = lim n→∞ Ω ns de (norm convergence in .B (H )), where .{s }
n n=1 is any
sequence in .sim (Ω, A) such that .‖f − sn ‖∞ → 0 as .n → ∞.
In the next theorem we collect the basic properties of the integration mapping.
Theorem 1.5.3 If .e : A → B (H ) is a spectral measure on the measurable space
(Ω, A), then the following statements hold.
.
(i) The integration mapping .f │−→ Ω f de is an algebra homomorphism from
.Bb (Ω, A) into .B (H ).
∗
(ii) . Ω f¯de = Ω f de for all .f ∈ Bb (Ω, A).
1.5 The Spectral Theorem 11
(iii) . Ω f deB(H ) ≤ ‖f ‖∞ for all .f ∈ Bb (Ω, A).
(iv) For every .f ∈ Bb (Ω, A) the operator . Ω f de is normal and if f is real
valued,
then
. Ω f de is self-adjoint.
(v) . Ω f de ξ, η = Ω f deξ,η for all .ξ, η ∈ H and all .f ∈ Bb (Ω, A).
2
(vi) . Ω f de ξ H = Ω |f |2 deξ,ξ for all .ξ ∈ H and all .f ∈ Bb (Ω, A).
(vii) If .f ≥ 0 in .Bb (Ω, A), then . Ω f de ≥ 0 in .B (H ).
(viii) If .f, g ∈ Bb (Ω, A) are such that .|f | ≤ |g|, then . f de ξ H ≤
Ω
Ω gde ξ H for all .ξ ∈ H . In particular, . Ω f de B(H ) ≤ Ω gde B(H ) .
∞
(ix) If .{fn }n=1 is a uniformly bounded sequence in .Bb (Ω, A) and if .f ∈
that .fn (ω) → f (ω) as .n → ∞ for all .ω ∈ Ω, then
Bb (Ω, A) such
. f
Ω n de → Ω f de strongly as .n → ∞.
The integral of a function over a subset of .Ω is defined in the usual manner.
Definition
1.5.4 Given .f ∈ Bb (Ω, A) and .∆ ∈ A, we define . ∆ f de =
Ω f χ∆ de.
From the multiplicativity of the integration mapping it is obvious that
. f de = e (∆) f de = f de e (∆) .
∆ Ω Ω
and .xξ = ∞ n=1 xn ξ for all .ξ ∈ D (x). Then .x : D (x) → H is a normal operator
in H . It is not too difficult to show that the operator x does not depend on the choice
of the admissible sequence. This justifies the following definition.
Definition 1.5.6 Let .e : A → B (H ) be a spectral measure on the measurable
space .(Ω, A). Let .{∆n }∞
n=1 be an admissible sequence for the function .f ∈
B (Ω, A). The integral of f with respect to e is defined by
∞
. f de = f de.
Ω n=1 ∆n
In the next theorem we collect the basic properties of the integration mapping.
For notational convenience we will also write .Фe (f ) = Ω f de for all .f ∈
B (Ω, A). Then .Фe maps .B (Ω, A) into the collection .C (H ) of all closed, densely
defined linear operators on H . If .x, y ∈ C (H ) are such that the operator .x + y is
pre-closed, then .x +̂y denotes the closure of .x + y. Similarly, if xy is pre-closed,
then .xˆ·y denotes the closure of xy.
Theorem 1.5.7 Let .e : A → B (H ) be a spectral measure on the measurable
space .(Ω, A) and define .Фe : B (Ω, A) → C (H ) by .Фe (f ) = Ω f de. For all
.f, g ∈ B (Ω, A) the following statements hold:
(i) .D (Фe (f )) = ξ ∈ H : Ω |f |2 deξ,ξ < ∞ and .‖Фe (f ) ξ ‖2H =
Ω |f | 2
de ξ,ξ for all .ξ ∈ D (Ф e (f ));
.ξ ∈ D (Фe (f )) and .η ∈ H , then .f ∈ L1 eξ,η and .〈Фe (f ) ξ, η〉 =
(ii) if
Ω f de ξ,η ;
(iii) if .{∆n }∞ n=1 is an admissible sequence for f , then the algebraic direct sum
.⊕
∞ e (∆ ) H is a core for the operator .Ф (f );
n
n=1 e
(iv) .Фe f¯ = Фe (f )∗ ;
(v) if f is real valued, then .Фe (f ) is self-adjoint and if .f ≥ 0, then .Фe (f ) ≥ 0;
(vi) .Фe (f + g) = Фe (f ) +̂Фe (g);
(vii) .Фe (fg) = Фe (f ) ˆ·Фe (g). Moreover, .D (Фe (f ) Фe (g)) = D (Фe (g)) ∩
D (Фe (fg));
(viii) if .g ∈ Bb (Ω, A), then .Фe (f g) = Фe (f ) Фe (g);
(ix) .Фe (p (f )) = p (Фe (f )) for all polynomials p. In particular, .p (Фe (f )) is
closed for any polynomial p;
(x) if .∆ ∈ A such that .f χ∆ ∈ Bb (Ω, A), then .e (∆) H ⊆ D (Фe (f )) and
.Фe (f χ∆ ) = Фe (f ) e (∆).
Given .f ∈ B (Ω, A) and .∆ ∈ A we define . ∆ f de = Ω Note
f χ∆ de.
that it follows from (viii) in the above theorem that . ∆ f de = Ω f de e (∆).
Furthermore, it is easy to see that if the spectral measure
e is supported by a set
.∆ ∈ A (i.e., .Ω \ ∆ is an e-null set), then . f de = f de for all .f ∈ B (Ω, A).
Ω ∆
Therefore, in this situation we may as well consider the integration mapping to be
defined on the space .B (∆, A∆ ), where .A∆ = {δ ∈ A : δ ⊆ ∆}.
1.5 The Spectral Theorem 13
is denoted by .B (C), i.e., .B (C) = B (C, B (C)). Similarly, we write .Bb (C) for
.Bb (C, B (C)).
.f (x) = f (λ) dex (λ)
C
smallest closed subset of .C with .e (supp (e)) = 1. Moreover, if U is any open set in
.C such that .U ∩ supp (e) /= ∅, then .e (U ) /= 0. Note that .
C f de = supp(e) f de for
all .f ∈ B (C). For all .ξ, η ∈ H , the support of the Borel measure .eξ,η is contained in
.supp (e). The support of the spectral measure of a normal operator can be identified
precisely.
Theorem 1.5.13 If .x : D (x) → H is a normal operator, then .supp (ex ) = σ (x).
From this result it follows that .f (x) = σ (x) f (λ) dex (λ) for all .f ∈ B (C) and
so for the Borel functional calculus of x we only need to consider .f ∈ B (σ (x)).
Now we shall discuss some consequences of the change of measure formula (see
Theorem 1.5.8).
Theorem 1.5.14 Given the spectral measure .e : Ω → B (H ) on the measurable
space .(Ω, A) and .f ∈ B (Ω, A), define .x = Ω f de. The spectral measure .ex of
x is given by .ex = f (e), i.e., .ex (δ) = e f −1 (δ) for all .δ ∈ B (C). Moreover, for
any .g ∈ B (C) we have .g (x) = Ω g ◦ f de.
Corollary 1.5.15 Let .x : D (x) → H be a normal operator and .f, g ∈ B (C).
Then .ef (x) = f (ex ) and .g (f (x)) = (g ◦ f ) (x).
The following corollary is a useful consequence of the above result.
Corollary 1.5.16 Assume that x and y are two normal operators in H . Suppose
that .f, g ∈ B (C) are such that .g (f (λ)) = λ for all .λ ∈ σ (x) ∪ σ (y). Then
.f (x) = f (y) implies that .x = y.
1.6 Self-adjoint operators 15
and .x̄ is also symmetric. Recall that a densely defined linear operator a in H is self-
adjoint if .a = a ∗ . Obviously, any self-adjoint operator is closed and symmetric.
However, a closed symmetric operator is not necessarily self-adjoint. The following
theorem gives a useful criterion to verify whether a given symmetric operator is
self-adjoint.
Theorem 1.6.1 For a symmetric operator x in H the following conditions are
equivalent:
(i) x is self-adjoint;
(ii) x is closed and .Ker (x ∗ ± i1) = {0};
(iii) x is closed and .Ran (x ± i1)⊥ = {0};
(iv) .Ran (x ± i1) = H .
This shows that a is not positive and the statement is proved. From these observa-
tions the next proposition follows immediately.
Proposition 1.6.3 Let a be a self-adjoint operator in H and define .m, M ∈
[−∞, ∞] by
m = inf 〈aξ, ξ 〉 : ξ ∈ D (a) , ‖ξ ‖H = 1
.
and
M = sup 〈aξ, ξ 〉 : ξ ∈ D (a) , ‖ξ ‖H = 1 .
.
.σ (b1 ) , σ (b2 ) ⊆ [0, ∞), it is clear that .g (f (λ)) = λ whenever .λ ∈ σ (b1 )∪σ (b2 ).
(i) that .D (a) ⊆ D a 1/2 and, by (iii) of the same theorem, .D (a) is a core of .a 1/2 .
The positive part .a + and negative part .a − of a self-adjoint operator a are defined
by
.a+ = λ+ dea (λ) and a − = λ− dea (λ)
R R
1.6 Self-adjoint operators 17
respectively, where .λ+ = max (λ, 0) and .λ− = max (−λ, 0). By Theorem 1.5.7 (v),
+ − ≥ 0 and it follows from (vi) of the same theorem that .a = a + −̂a − . Actually,
.a , a
using Theorem 1.5.7 (i), we see that .D (a) = D a + ∩ D a − and so, .a = a + −
+
a − . Observe that it follows from Corollary 1.5.15 that .s a + = ea (R \ {0}) =
−
ea (0, ∞) and .s a − = ea (R \ {0}) = ea (−∞, 0). Therefore, .s a + s a − = 0.
The above decomposition in positive and negative part is unique in the following
sense.
Proposition 1.6.5 Suppose that a is a self-adjoint operator in the Hilbert space
H . If b and c are self-adjoint positive operators on H such that .a = b − c and
.s (b) s (c) = 0, then .b = a
+ and .c = a − .
From Theorem 1.5.7 it follows that .|a| ≥ 0, .D (|a|) = D (a) and .a = a + −a − . Note
that . 12 (|a| + a) ⊆ a + and . 12 (|a| − a) ⊆ a − . Moreover, by (ix) of Theorem 1.5.7
we have
. |a|2
= |λ|2
de (λ) =
a
λ2 dea (λ) = a 2 .
R R
1/2
Therefore, the uniqueness of the positive square root implies that .|a| = a 2 .
Two self-adjoint operators a and b in H are called resolvent commuting if
.r (λ, a) r (μ, b) = r (μ, b) r (λ, a) for all .λ, μ ∈ C \ R.
Proposition 1.6.6 If a and b are self-adjoint operators in H , then the following two
statements are equivalent:
(i) a and b are resolvent commuting;
(ii) .ea (δ1 ) eb (δ2 ) = eb (δ2 ) ea (δ1 ) for all Borel sets .δ1 , δ2 ⊆ R.
Proof For the proof that (i) implies (ii), we start with the following observation.
Suppose that .x ∈ B (H ) and .λ ∈ C \ R such that .r (λ, a) x = xr (λ, a). We claim
that .xa ⊆ ax. Indeed, let .ξ ∈ D (a) be given. There exists .η ∈ H such that .ξ =
r (λ, a) η and so, .xξ = xr (λ, a) η = r (λ, a) xη, which shows that .xξ ∈ D (a).
Furthermore,
implies that .r (μ, b) ea (δ1 ) = ea (δ1 ) r (μ, b) for all Borel sets .δ1 in .R. Keeping
the Borel set .δ1 fixed for the moment, it follows from the first part of the proof
that .ea (δ1 ) b ⊆ bea (δ1 ). Using Theorem 1.5.12 once more, we may conclude that
.e (δ1 ) e (δ2 ) = e (δ2 ) e (δ1 ) for all Borel sets .δ2 in .R. Since .δ1 is an arbitrary
a b b a
of H . The projection .p = s (v) onto .Ker (v)⊥ is called the initial projection of v
and the projection .q = r (v) onto .Ran (v) is called the final projection of v. It is easy
to see that .p = v ∗ v and .q = vv ∗ . Note that .p = v ∗ qv, .q = vpv ∗ and .v = vp = qv.
Conversely, if .v ∈ B (H ) is such that .v ∗ v is a projection, then v is a partial
isometry with initial projection .v ∗ v (and final projection .vv ∗ ).
Theorem 1.7.3 (Polar Decomposition) Let .x : D (x) → H be a closed and
densely defined operator. There exists a partial isometry v, with initial projection
.r (|x|) and final projection .r (x), such that .x = v |x|. Moreover, if .x = wa, where a
is positive self-adjoint and w is a partial isometry with initial projection .r (a), then
.a = |x| and .w = v.
The factorization .x = v |x| in the above theorem is called the polar decom-
position of x. The last statement in this theorem is usually referred to as the
uniqueness of the polar decomposition. Since .v ∗ v = r (|x|), it is also clear that
∗
.|x| = v x. There are a number of statements implicit in the above theorem. In
particular, .D (|x|) = D (x) and .‖|x| ξ ‖H = ‖xξ ‖H for all .ξ ∈ D (x). Hence,
∗
.Ker (|x|) = Ker (x), .n (|x|) = n (x) and .r (|x|) = r (x ) = s (x). This implies
Since .v |x| v ∗ ≥ 0 and .vv ∗ = r (x) = r (v |x| v ∗ ), it follows from the uniqueness
of the polar decomposition that .v |x| v ∗ = |x ∗ | and that .x ∗ = v ∗ |x ∗ | is the polar
decomposition of .x ∗ .
Another application of the uniqueness of the polar decomposition is given in the
following proposition.
Proposition 1.7.4 Suppose that .x : D (x) → H is a closed, densely defined
operator with polar decomposition .x = v |x| and let .u ∈ B (H ) be unitary. Then,
∗
.uxu is closed and densely defined with polar decomposition given by
.uxu∗ = uvu∗ u |x| u∗ .
As before, H is a complex Hilbert space with inner product .〈·, ·〉. If D is a linear
subspace of H and .q : D × D → C is a map which is linear in the first variable and
conjugate linear in the second, then .q is called a sesquilinear form on H with domain
D. The domain of .q will usually be denoted by .D (q). Given a sesquilinear form .q,
the corresponding quadratic form, also denoted by .q, is defined by .q (ξ ) = q (ξ, ξ )
for all .ξ ∈ D (q). It follows from the polarization identity
1 k
3
.q (ξ, η) = i q ξ + i k η, ξ + i k η , ξ, η ∈ D (q) ,
4
k=0
that a sesquilinear form is uniquely determined by its quadratic form. From now on,
both sesquilinear forms and quadratic forms will be simply called forms.
A form .q is called symmetric if .q (ξ, η) = q (η, ξ ) for all .ξ, η ∈ D (q); note
that this implies that .q (ξ ) ∈ R for all .ξ ∈ D (q). Furthermore, .q is called positive
if .q is symmetric and .q (ξ ) ≥ 0 for all .ξ ∈ D (q). From now on we will only
consider positive forms. Note that any positive form satisfies the Cauchy–Schwartz
inequality, that is, .|q (ξ, η)| ≤ q (ξ )1/2 q (η)1/2 for all .ξ, η ∈ D (q), and the map
.ξ │−→ q (ξ ) is a seminorm on .D (q). This implies in particular that .q (ξn ) →
1/2
q (ξ ) for any sequence .{ξn }∞n=1 in .D (q) and .ξ ∈ D (q) satisfying .q (ξn − ξ ) → 0.
20 1 A Review of Relevant Operator Theory
Definition 1.8.1 A positive form .q on H is called closed if for any sequence .{ξn }∞
n=1
in .D (q) and .ξ ∈ H the conditions
(i) .ξn → ξ in H ,
(ii) .q (ξn − ξm ) → 0 as .n, m → ∞,
imply that .ξ ∈ D (q) and .q (ξn − ξ ) → 0 as .n → ∞.
Definition 1.8.2 Let .q be a positive form on H . A linear subspace D of .D (q) is
called a core of .q if for each .ξ ∈ D (q) there exists a sequence .{ξn }∞
n=1 in D such
that .ξn → ξ in H and .q (ξn − ξ ) → 0.
Suppose that .q is a positive form on H . For .ξ, η ∈ D (q) define
Evidently, .〈·, ·〉q is an inner product in .D (q). The corresponding norm is denoted
by .‖·‖q and satisfies
Note that convergence of a sequence with respect to .‖·‖q always implies conver-
gence in H and .q is continuous on .D (q) with respect to .‖·‖q . The following facts
are easily verified:
(a) A positive form .q is closed if and only if .D (q) is a Hilbert space with respect
to .〈·, ·〉q ;
(b) A linear subspace D of .D (q) is a core of .q if and only if D is dense in .D (q)
with respect to the norm .‖·‖q ;
(c) Let .q1 and .q2 be closed positive forms on H and suppose that D is a subspace
of .D (q1 ) ∩ D (q2 ) which is a core for both .q1 and .q2 . If .q1 (ξ ) = q2 (ξ ) for all
.ξ ∈ D, then .q1 = q2 .
q (ξ ) = sup qα (ξ ) = lim qα (ξ ) , ξ ∈ D.
α α
.ξn → ξ in .D (q) with respect to .‖·‖q . We may conclude that .D (q) is complete with
Let .a : D (a) → H be
a positive self-adjoint operator in H . Define the form
qa by .D (qa ) = D a 1/2 and .qa (ξ, η) = a 1/2 ξ, a 1/2 η for all .ξ, η ∈ D (qa ).
.
22 1 A Review of Relevant Operator Theory
2
Clearly, .qa is a positive densely defined form on H and .qa (ξ ) = a 1/2 ξ H for all
.ξ ∈ D (qa ). Using that .a
1/2 is a closed operator, it follows that .q is also closed.
a
The form .qa is called the quadratic form corresponding to the positive self-
adjoint operator a. It is easily verified that the operator a is uniquely determined
by its quadratic form in the following sense: if .a and b are two positive self-adjoint
operators in H such that .qa = qb , then .a = b. Furthermore, we observe that a linear
subspace D of .D (qa ) is a core of .qa if and only if D is a core of .a 1/2 . In particular,
.D (a) is a core of .D (qa ).
Let us denote the set of all positive self-adjoint operators in H by .H+ . By the
above observation, the map .a − │ → qa from .H+ into .Q+ is injective. The next
proposition shows that this map is also surjective (and hence, it is a bijection).
Proposition 1.8.7 If .q is a closed densely defined positive form on H , then there
exists a unique self-adjoint positive operator a in H such that .D (q) = D a 1/2
2
and .q (ξ ) = a 1/2 ξ for all .ξ ∈ D (q) (in other words, .q = qa ).
H
Proof Since .q is closed, .D (q) is a Hilbert space with respect to .〈·, ·〉q . For .ξ ∈ H ,
define the linear functional .ϕξ on .D (q) by .ϕξ (η) = 〈η, ξ 〉. Since
. ϕξ (η) = |〈η, ξ 〉| ≤ ‖ξ ‖H ‖η‖H ≤ ‖ξ ‖H ‖η‖q
for all .η ∈ D (q), it follows that .ϕξ is bounded with respect to .‖·‖q and .ϕξ ≤
‖ξ ‖H . Consequently, there exists a unique element
.bξ ∈ D (q) such that .ϕξ (η) =
〈η, bξ 〉q for all .η ∈ D (q) and .‖bξ ‖q = ϕξ ≤ ‖ξ ‖H . The map .ξ │−→ bξ is linear
from H into .D (q). Composing this map with the embedding of .D (q) into H , we
obtain a linear operator .b : H → H , which is bounded with .‖b‖B(H ) ≤ 1. From the
definition of .bξ it follows that
and so,
(iii) .Ran (b) is dense in .D (q) with respect to .‖·‖q (and hence, .Ran (b) is dense in
H ). Indeed, if .η ∈ D (q) satisfies .〈η, bξ 〉q = 0 for all .ξ ∈ H , then .〈η, ξ 〉 =
〈η, bξ 〉q = 0 for all .ξ ∈ H and hence, .η = 0.
Consider the inverse .b−1 : D b−1 → H , where .D b−1 = Ran (b). From the
above observations
−1 ∗ it follows that .b−1 is closed, densely defined and self-adjoint
−1
(recall that . b = (b ) = b−1 ). Equation (1.5) can then be written as
∗
q (η, ζ ) = η, b−1 ζ − ζ = η, b−1 − 1 ζ , η ∈ D (q) , ζ ∈ D b−1 .
.
The .a = b
−1 − 1 is closed, densely defined and self-adjoint with .D (a) =
operator
−1
D b = Ran (b) and satisfies
and .〈aξ, ξ 〉 = q (ξ, ξ ) = q (ξ ) ≥ 0 for all .ξ ∈ D (a), which shows that a is positive.
2
Finally we show that .D (q) = D a 1/2 and .q (ξ ) = a 1/2 ξ H for all .ξ ∈ D (q).
Let .qa be the closed positive form corresponding to a, that is, .D (qa ) = D a 1/2
2
and .qa (ξ ) = a 1/2 ξ H for all .ξ ∈ D (qa ). Since .D (a) is a core of the operator .a 1/2 ,
it follows that .D (a) is a core for .qa . As has been observed above, .D (a) = Ran (b)
is dense in .D (q) with respect to .‖·‖q , which implies that .D (a) is a core for .q.
Furthermore,
2
q (ξ ) = 〈aξ, ξ 〉 = a 1/2 ξ, a 1/2 ξ = a 1/2 ξ = qa (ξ ) ,
. ξ ∈ D (a) .
H
characterized by:
⎧ ⎫
⎨ 1/2 ⎬
1/2
.D a
1/2
= ξ∈ D aβ : sup aβ ξ < ∞ ,
⎩ β H ⎭
β
1/2 1/2
a ξ = sup aβ ξ , ξ ∈ D a 1/2 .
H β H
Remark 1.8.10 We wish to point out another useful consequence of the above
considerations. Suppose that a is a positive self-adjoint operator in the Hilbert space
H . If .f : σ (a) → [0, ∞) is a Borel function, then .f (a) ∈ H+ and it follows from
Theorem 1.5.7 (i) that the corresponding quadratic form .qf (a) is given by
! "
.D qf (a) = ξ ∈ H : f (λ) deξ,ξ (λ) < ∞
σ (a)
and
qf (a) (ξ ) =
. f (λ) deξ,ξ (λ) , ξ ∈ D qf (a) .
σ (a)
In this and the next section, we discuss some of the aspects of the theory of abstract
operator algebras, which will be convenient to have available. We start with the
general concept of a .∗-algebra.
1.9 Algebras with an Involution 25
Let .A be an algebra over the complex numbers. The mapping .x │−→ x ∗ from .A
into itself is said to be an involution if
(i) .(x + y)∗ = x ∗ + y ∗ ;
∗ ∗
(ii) .(λx) = λ̄x ;
∗ ∗ ∗
(iii) .(xy) = y x ;
∗ ∗
(iv) .(x ) = x,
1 1
Re (x) =
. x + x ∗ , Im (x) = x − x∗ .
2 2i
Clearly, .Re (x) , Im (x) ∈ Ah and .x = Re (x) + iIm (x) for all .x ∈ A. Conversely,
if for a given .x ∈ A we have .x = x1 + ix2 with .x1 , x2 ∈ Ah , then necessarily
.x1 = Re (x) and .x2 = Im (x).
the element y is unique and denoted by .x −1 , the inverse of x. It is easy to see that
∗
.x ∈ A is invertible if and only if .x is invertible and, in this case, . x
−1 ∗ = (x ∗ )−1 .
If, in addition, .A and .B are unital and .φ (1A ) = 1B , then .φ is called a unital
.∗-homomorphism. If .A and .B are .∗-algebras and .φ : A → B is an injective .∗-
1.10 C ∗ -Algebras
An algebra .A equipped with a norm .‖·‖A such that .A is a Banach space and
(i) .‖xy‖A ≤ ‖x‖A ‖y‖A for all .x, y ∈ A,
is called a Banach algebra. If .A has a unit element .1, then we will assume that
‖1‖A = 1. Furthermore, if .A is a .∗-algebra and the norm satisfies in addition
.
as well.
We list a few simple examples of .C ∗ -algebras.
Example 1.10.1
1. Let X be an arbitrary non-empty set. We denote by .𝓁∞ (X) the .∗-algebra of
all bounded complex-valued functions .f : X → C, with pointwise algebraic
operations and involution given by pointwise complex conjugation. Equipped
with the norm given by .‖f ‖∞ = supt∈X |f (t)| for all .f ∈ 𝓁∞ (X), the algebra
∗
.𝓁∞ (X) is a commutative unital .C -algebra.
. ‖f ‖∞ = esssupt∈Ω |f (t)|
Banach subalgebra of the Banach algebra .A and .x ∈ B, then .σA (x) ⊊ σB (x).
However, for .C ∗ -algebras we have the following important observation: if .B is a
∗ ∗
.C -subalgebra of the .C -algebra .A, then .σA (x) = σB (x) for all .x ∈ B.
for this functional calculus, that is, for every .f ∈ C (σ (x)) we have
for all .f ∈ Bb (σ (x)) (and also for .f ∈ B (σ (x)); see Definition 1.5.11) without
any danger of confusion.
Actually, one may prove the existence of the spectral measure of a normal
operator .x ∈ B (H ) via the result of Theorem 1.10.2 (a method which is analogous
to the Riesz representation theorem for positive functionals on a .C (K)-space).
Let .A be a (unital) .C ∗ -algebra. An element .x ∈ A is said to be positive if .x ∗ = x
and .σ (x) ⊆ [0, ∞). This is denoted by .x ≥ 0. The set of all positive elements in .A
is denoted by .A+ . If .x ∈ A+ , then it follows from Theorem 1.10.2 that there exists
an element .y ∈ A+ such that .y 2 = x (indeed, take .y = f (x), where .f (λ) = λ1/2
for .λ ∈ σ (x) ⊆ [0, ∞) and use (1.6) that .y ≥ 0). The element .y ∈ A+ with
√ to see 1/2
.y
2 = x is unique and denoted by . x or .x , the positive square root of x. The
following proposition describes the basic properties of positive elements.
1.10 C ∗ -Algebras 29
Proposition 1.10.4
(i) .A+ is a cone in .A (that is, .λx + μy ∈ A+ whenever
+
.x, y ∈ A and .λ, μ ∈
+ +
R+ ) which is closed and proper (that is, .A ∩ −A = {0});
(ii) .A+ = {x ∗ x : x ∈ A} = a 2 : a ∈ Ah ;
(iii) if .A is a .C ∗ -subalgebra of .B (H ), for some Hilbert space H , and if .x ∈ A,
then .x ∈ A+ if and only if .〈xξ, ξ 〉 ≥ 0 for all .ξ ∈ H (that is, x is a positive
self-adjoint operator on H ).
In the real vector space .Ah we define a partial ordering by setting .a ≤ b
whenever .a, b ∈ Ah satisfy .b − a ∈ A+ . In the next proposition we collect some
of the properties of this partial ordering.
Proposition 1.10.5
(i) .Ah is a partially ordered vector space (that is, .a + c ≤ b + c and .λa ≤ λb
whenever .a ≤ b in .Ah , .c ∈ Ah and .λ ∈ R+ );
(ii) if .a ≤ b in .Ah , then .x ∗ ax ≤ x ∗ bx for all .x ∈ A;
(iii) if .a, b ∈ A+ and .ab = ba, then .ab ∈ A+ ;
(iv) if .0 ≤ a ≤ b in .Ah , then .‖a‖A ≤ ‖b‖A and .a 1/2 ≤ b1/2 ; moreover, if in
addition a is invertible, then b is also invertible and .0 ≤ b−1 ≤ a −1 ;
(v) if .x ∈ A is normal and .f ∈ C (σ (x)) satisfies .f (λ) ≥ 0 for all .λ ∈ σ (x),
then .f (x) ≥ 0 in .Ah .
Define the functions .f0 , f1 and .f2 on .R by .f0 (λ) = |λ|, .f1 (λ) = λ+ and
.f2 (λ) = λ
− respectively, for all .λ ∈ R. For .a ∈ A we set .|a| = f (a), the
h 0
absolute value of a, .a + = f1 (a), the positive part of a, and .a − = f2 (a), the
negative part of a. From the properties of the functional calculus it is clear that
+ − ∈ A+ , .a = a + − a − and .|a| = a + + a − . This shows in particular that
.|a| , a , a
the positive cone .A+ is generating in .Ah , that is, .Ah = A+ − A+ . Note that the
1/2
absolute value of .a ∈ Ah is also given by .|a| = a 2 .
It is worth noting, that for every .a ∈ Ah with .‖a‖A ≤ 1, the element .1 − a 2 is
1/2 1/2
positive and the elements .u1 = a + i 1 − a 2 and .u2 = a − i 1 − a 2 are
unitary. Moreover, .a = 12 (u1 + u2 ). This implies, in particular, that .A is the linear
span of its unitary elements.
A linear functional f on a .C ∗ -algebra .A is said to be positive if .f (a) ≥ 0 for
all .a ∈ A+ . If f is a positive functional on .A, then f is necessarily continuous,
∗ ∗
.f (x ) = f (x) and .|f (x)| ≤ ‖f ‖A∗ f (x x) for all .x ∈ A. Moreover, if .A is
2
. xj + yj = xj + yj , xj yj = xj yj , λ xj = λxj ,
∗ ∗
xj = xj , xj A = sup xj A ,
j
j
30 1 A Review of Relevant Operator Theory
S' = {x ∈ B (H ) : xy = yx ∀y ∈ S} ,
.
Theorem 1.11.2 Suppose that . aβ is an increasing net in .Mh .
(i) If .a ∈ Bh (H ) is such that .aβ ↑ a in .Bh (H ), then .a ∈ Mh (and hence, .aβ ↑ a
in .Mh).
(ii) If . aβ is bounded from above, then there exists .a ∈ Mh such that .aβ ↑ a in
.Mh (and .aβ ↑ a in .Bh (H )).
Suppose that .{Mi : i ∈ I } is a collection of von Neumann algebras acting on the
'
Hilbert space H and define .M = i∈I Mi . Since .x ∈ M if and only if .x ∈ M'i
for all .i ∈ I , it is clear that
$ &'
%
M=
. M'i ,
i∈I
.Z (M) = {x ∈ M : xy = yx ∀y ∈ M} .
on .M given by .ωξ,η (x) = 〈xξ, η〉 for all .x ∈ M (see Sect. 1.2). Let .M∼ be the
linear subspace of .M∗ generated by . ωξ,η : ξ, η ∈ H . Evidently, the wo-topology
is equal to .σ (M, M∼ ). The norm closure of .M∼ is denoted by .M∗ . Furthermore,
we denote by .M'wo , .M'so , .M'uwo and .M'uso the dual spaces of .M with respect to the
wo-, so-, uwo- and uso-topology, respectively. In the following theorem we collect
the main features of the duality theory of von Neumann algebras.
Theorem 1.11.5 Let .M be a von Neumann algebra on the Hilbert space H .
(i) .M'so = M'wo and .ϕ ∈ M'wo if and only if .ϕ is given by .ϕ = nj=1 ωξj ,ηj with
.ξ1 , . . . , ξn ∈ H and .η1 , . . . , ηn ∈ H .
' ' '
(ii) .M
∞ uso = Muwo = M∗ and .ϕ ∈ Muwo if and only if is given by .ϕ
.ϕ∞ ∞=
∗
j =1 ω ξj ,ηj , as a norm convergent series in .M , with . ξ j j =1
and . η j j =1
2
in H satisfying . ∞ ξj < ∞ and .∞ ηj 2 < ∞.
j =1 H j =1 H
(iii) If .ϕ is a linear functional on .M, then .ϕ belongs to .M∗ if and only if the
restriction of .ϕ to the unit ball .BM is wo-continuous (equivalently, is so-, uso-
or uwo-continuous).
(iv) The uwo-topology in .M is equal to .σ (M, M∗ ).
(v) Every .x ∈ M defines a bounded linear functional .x̂ ∈ (M∗ )∗ by defining
.x̂ (ϕ) = ϕ (x) for all .ϕ ∈ M∗ . The mapping .x │−→ x̂ is a isometrically
Definition 1.11.6 Let .ϕ be a positive linear functional on the von Neumann algebra
M.
.
(i) .ϕ is said to be normal if .aβ ↑ a in .M+implies
that.ϕ aβ ↑ ϕ (a).
(ii) .ϕ is called completely additive if .ϕ α pα = α ϕ (pα ) for every system
.{pα } of pairwise orthogonal projections in .M.
that is, .xe = (ex)|K . Note that .xe = (exe)e for all .x ∈ B (H ). For any non-empty
subset .D ⊆ B (H ) we denote .De = {xe : x ∈ D}. We consider, in particular, the
following two situations:
1. Let .M ⊆ B (H ) be a von Neumann algebra and .e ∈ P (M). Define
eMe = {exe : x ∈ M} ,
.
34 1 A Review of Relevant Operator Theory
.φe : eMe → Me , defined by .φe (x) = xe for all .x ∈ eMe, is a surjective unital
.∗-isomorphism.
2. Let .M ⊆ B (H ) be a von Neumann algebra and .e ∈ P M' . If .x ∈ M, then
⊥
.xe = ex, that is, .x (K) ⊆ K and .x K ⊆ K ⊥ . Moreover, .xe = x|K for all
.x ∈ M. The set .Me is a unital .∗-subalgebra of .B (K) and the mapping .ψe :
where .z(e) denotes the central support of e; see Definition 1.14.2 below. In
particular, .ψe is an isomorphism if and only if .z (e) = 1.
Theorem 1.11.11 Let .M be a von Neumann algebra on the Hilbert space H .
Suppose that .e ∈ P (M) and put .K = e (H ).
(i) .Me and . M' e are von Neumann algebras on the Hilbert space K and
'
'
.(Me ) = M .
e
(ii) The center of .Me is equal to .(Z e.
(M))
(iii) If .M is a factor, then .Me and . M' e are also factors.
By (i) of the above theorem, there is no danger of confusion to denote . M' e
simply by .M'e whenever .e ∈ P (M). The von Neumann algebra .Me is called the
reduced von Neumann algebra of .M with respect to .e ∈ P (M). The von Neumann
algebra .M'e is called the induced von Neumann algebra by .M' on .K = e (H ).
Frequently, we shall identify the .∗-algebra .eMe with the von Neumann algebra .Me
on K.
Let .M be a von Neumann algebra on a Hilbert space H and recall that .M∗ denotes
the predual of .M, that is, .M∗ ⊆ M∗ is the space of all uwo-continuous linear
functionals on .M. For .ϕ ∈ M∗ and .y ∈ M the linear functionals .ϕy, yϕ ∈ M∗ are
defined by
respectively.
For a discussion of singular functionals on .M it will be convenient to introduce
the universal enveloping von Neumann algebra .Mu of .M. For the definition and
properties of .Mu the reader is referred to [71], Section 10.1 or [125], Section III.2.
1.12 Singular Functionals on von Neumann Algebras 35
In particular, .Mu is a von Neumann algebra on some Hilbert space K and there is
a natural embedding of .M into .Mu . Furthermore, .Mu can be identified with the
bidual space .M∗∗ of .M, such that the embedding of .M into .Mu ∼ = M∗∗ coincides
with the canonical embedding of .M into .M . This implies that .M∗ is the predual
∗∗
of .Mu , that is, .M∗ = (Mu )∗ . It follows that there exists a central projection .z0 in
∗
.Mu such that .M∗ = M z0 , that is,
M∗ = ϕz0 : ϕ ∈ M∗
.
M∗ = M∗ ⊕ M∗s .
. (1.7)
For further properties of this decomposition the reader is referred to, for instance,
Theorem III.2.14 of [125].
Any .ϕ ∈ M∗ can be written as
ϕ = (ϕ1 − ϕ2 ) + i (ϕ3 − ϕ4 ) ,
.
+
where .ϕj ∈ M∗ , for .j = 1, . . . , 4. Since .z0 is a central projection of .Mu , this
implies the following result.
Proposition 1.12.2
(i) Any .ϕ ∈ M∗s can be written as
ϕ = (ϕ1 − ϕ2 ) + i (ϕ3 − ϕ4 ) ,
.
+
where .ϕj ∈ M∗s , for .j = 1, . . . , 4.
(ii) If .ϕ ∈ M∗s , then also .ϕy, yϕ ∈ M∗s for all .y ∈ M.
Another relevant observation is the following proposition (see [125], Theorem
3.8). Here .P (M) denotes the collection of all (orthogonal) projections belonging to
.M.
+
Proposition 1.12.3 Let .ϕ ∈ M∗ . The following two statements are equiva-
lent.
(i) .ϕ ∈ M∗s .
(ii) For every .0 /= p ∈ P (M) there exists .q ∈ P (M) with .0 < q ≤ p satisfying
.ϕ (q) = 0.
36 1 A Review of Relevant Operator Theory
dense in .M if for every .0 < x ∈ M+ there exists .y ∈ J + such that .0 < y ≤ x. The
following characterization of order denseness is readily verified.
Lemma 1.12.4 If .J ⊆ M is an order ideal, then the following three statements are
equivalent:
(i) J is order dense in .M;
(ii) for every .0 < x ∈ M+ there exists and upwards directed system .{xα } in .J +
such that .0 ≤ xα ↑α x;
(iii) for every .0 < p ∈ P (M) there exists .q ∈ P (M) ∩ J such that .0 < q ≤ p.
+
For any .φ ∈ M∗ we denote by .N (φ) its absolute kernel that is,
N (φ) = x ∈ M : φ (|x|) = φ x ∗ = 0 ,
.
φ = φ1 − φ2 + i (φ3 − φ4 ) .
.
As observed in part (i), each .N φj is an order dense order ideal in .M and so,
4
.J = j =1 N φj is an order dense order ideal on which .φ vanishes.
Conversely, suppose that .φ ∈ M∗ vanishes on some order ideal J which
is order dense in .M. Writing .φ = φ1 + φ2 , with .φ1 ∈ M∗ and .φ2 ∈ M∗s , it
follows from what just has been proved that .φ2 vanishes on some order dense
order ideal .J2 ⊆ M. Hence, .φ1 vanishes on the order dense order ideal .J ∩ J2 .
If .0 ≤ x ∈ M, then it follows from Lemma 1.12.4 (ii) that there exists an
1.13 Direct Products of von Neumann Algebras 37
⨆
⨅
∗ +
Recall from Definition 1.11.6 that a positive functional .φ ∈ M is normal
whenever .xα ↓α 0 in .M implies that .φ (xα ) ↓α 0. Furthermore, it is clear that
any .φ ∈ M∗ has the property that .φ (xα ) → 0 whenever .xα ↓α 0 in .M (as
.xα → 0 with respect to the uwo-topology). This motivates the following extension
of Definition 1.11.6.
Definition 1.12.6 A linear functional .φ ∈ M∗ is called normal whenever .xα ↓α 0
in .M implies that .φ (xα ) → 0.
The collection of all normal functionals on .M is denoted by .M∗n . By what just
has been observed, it follows that .M∗ ⊆ M∗n . Next it will be shown that actually
∗
.M∗ = Mn , extending Theorem 1.11.7.
.φ1 (xα ) → φ1 (x). Hence, .φ (x) = φ1 (x), that is, .φ2 (x) = 0. This holds for all
+
.x ∈ M , which implies that .φ2 = 0, that is, .φ = φ1 ∈ M∗ . This suffices for the
proof. ⨆
⨅
In view of this observation, decomposition (1.7) can also be written as
singular parts.
it follows that . j ∈J Hj is a Hilbert space. Note that the norm of . ξj ∈ j ∈J Hj
1/2
is given by . 2 . The Hilbert space . j ∈J Hj , also denoted by . j Hj ,
j ξ j Hj
is called the direct sum of the Hilbert spaces . Hj j ∈J . If .J = {1, . . . , n}, then
'
.
written as .H1 ⊕ · · · ⊕ Hn . Note that the algebraic direct sum . j Hj
j Hj is also
is dense in . j Hj . For any .k ∈ J , the mapping .Uk , which assigns to every .ξ ∈ Hk
the vector . ξj ∈ j Hj given by .ξk = ξ and .ξj = 0 whenever .j /= k, is an
isometric isomorphism from .Hk onto the closed subspace .H̃k of . j Hj . Usually,
we will identify
.Hk with .H̃k and so, we consider .Hk as a closed subspace of the
direct sum . j Hj . Note that the subspaces . Hj j ∈J are pairwise orthogonal and
that . j Hj = j Hj . For .k ∈ J we denote by .pk the projection in . j Hj onto
the subspace .Hk .
For the sake of convenience, we denote the direct sum Hilbert space . j Hj
for the moment simply by H . Suppose that operators .xj ∈ B Hj , .j ∈ J , are
given such that .supj xj B (H ) < ∞. Then, we may define the linear operator
j
.⊕j xj on H by . ⊕j xj
ξj = xjξj for all . ξj ∈ H . The operator .⊕j xj is
.
Mj = ⊕j xj : xj ∈ Mj ∀j ∈ J, sup xj B (H ) < ∞ ,
j
j j
which is a .C ∗ -subalgebra
where .H = j Hj .
of . j B Hj and hence, of .B (H ),
Note that .pk ∈ j Mj for all .k ∈ J . The commutant of . j Mj is given by
1.14 Comparison of Projections 39
' ''
'
. j M j = j M j and so, . j M j = j Mj , which shows that . j Mj
is a von Neumann algebra on H . The algebra . j Mj is called the product von
Neumann algebra of the family . Mj j ∈J . The projections .pk , .k ∈ J , belong to the
center of . j Mj , satisfy . k pk = 1 and . j M j = Mk . If the index set J is
pk
finite, say .J = {1, . . . , n}, then . j Mj is also denoted by .M1 ⊕ · · · ⊕ Mn .
As a converse to the above construction, we observe the following. Suppose that
.M is a von Neumann algebra on a Hilbert space H and assume that .{pα } is a
( )
P (M) = p ∈ M : p 2 = p, p∗ = p .
.
"Sen koiran tulee saada ruoskaa niin että hän sen muistaa kaikkina
elinpäivinänsä", kirkui Shestokow ja tahtoi vielä lisätä uhkauksia, kun
linnan pihaan ajavien vaunujen rätinä herätti hänen huomionsa.
"Te saatte kyllä tietää, mitä sanomia tämä vieras minulle tuopi,
rakas parooni", virkkoi Shestokow. "Kun minä en pyydä teitä
jäämään tänne, niin tapahtuu se ainoastaan nadsiratelin vuoksi, joka
teidän läsnäollessanne ei rohkenisi puhua suutansa puhtaaksi."
Hän astui niin nopeaan, että vouti tuskin jaksoi häntä seurata eikä
hiljentänyt askeleitaan, ennenkuin hän saapui sisarensa pojan
kartanolle.
"Vielä kerran", ähkyi hän, "anna se koira ulos tai minä en lepää
ennenkuin saat lähteä Siperiaan."
Aurinkokin armahainen
Yöhön yhtyy, vaipuu vainen.
Myrsky se öinen
Raivosi töineen,
Se ulvoen lensi, se riehui ja kiehui
Ja pauhinallaan
Se valtasi maan!
Kun vilu, joka oli pannut tutisemaan luita ja ytimiä myöten, oli
voitettu, virkkoi Jerupkow suojatilleen: "Nyt tuonne ukkoraukan
mökille."
Iivana tunsi että hänen isänsä oli lähtenyt ja hän vaipui itkien alas
kurjalle leposijalle. Hänen nyyhkytyksiinsä sekoittui pauhaavan
myrskyn hurjat säveleet, jotka nyt äkkiä täydellä voimalla
tunkeutuivat majaan, sillä ovi kiskaistiin äkkiä auki. Monen lyhdyn
räikeä hohde valaisi puolipimeää huonetta, valaisi myös samalla
Shestokowin ilosta ilkkuvia ja rumia kasvoja, hän kun nyt seisoi
keskellä tuota pientä huonetta.
Ilian kultavesi.