(Monographs in Mathematics 85) K. R. Parthasarathy (Auth.) - An Introduction To Quantum Stochastic Calculus-Birkhäuser Basel (1992)

Monographs in Mathematics
Vol. 85
Managing Editors:
H.Amann
Universitat Ziirich, Switzerland
K. Grove
University of Maryland, College Park
H. Kraft
Universitat Basel, Switzerland
P.-L. Lions
Universite de Paris-Dauphine, France
Associate Editors:
H. Araki, Kyoto University
J. Ball, Heriot-Watt University, Edinburgh
E Brezzi, Universita di Pavia
K.C. Chang, Peking University
N. Hitchin, University of Warwick
H. Hofer, Universitat Bochum
H. Kniirrer, ETH Ziirich
K. Masuda, University of Tokyo
D. Zagier, Max-Planck-Institut Bonn
K.R. Parthasarathy
An Introduction
toQuantum
Stochastic Calculus
Springer Basel AG
Author's address:
K.R. Parthasarathy
Indian Statistical Institute
7, S.J.S. Sansanwal Marg
New Delhi - 110016
India
Library of Congress Cataloging-in-Publication Data

Parthasarathy, K.R.
An introduction to quantum stochastic calculus 1K.R. Parthasarathy.
(Monographs in mathematics; vol. 85)
Includes bibliographical references.
ISBN 978-3-0348-9711-2 ISBN 978-3-0348-8641-3 (eBook)
DOl 10.1007/978-3-0348-8641-3
1. Stochastic processes. 2. Quantum theory. 3. Mathematical physics. I. Title, II. Series:
Monographs in mathematics; v. 85.
QC174.17.S76P37 1992
519.2-dc20
Deutsche Bibliothek Cataloging-in-Publication Data

Parthasarathy, Kalyanapuram R.:
An introduction to quantum stochastic calculus 1 K.R. Parthasarathy. -
Springer Basel AG, 1992
(Monographs in mathematics; Vol. 85)
ISBN 978-3-0348-9711-2
NE-GT
This work is subject to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specifically those of translation, reprinting, re-use of illustrations, broadcasting,
reproduction by photocopying machine or similar means, and storage in data banks. Under § 54 of
the German Copyright Law, where copies are made for other than private use a fee is payable to
«Verwertungsgesellschaft Wort», Munich.
© 1992 Springer Basel AG

Originally published by Birkhiiuser Verlag Basel in 1992
Softcover reprint of the hardcover 1st edition 1992
Revision and Layout by mathScreen online, CH-4056 Basel
ISBN 978-3-0348-9711-2
To
Brahma (the creator)
Vishnu (the protector)
Shiva (the annihilator)
Contents
Preface .................................................................... IX
Chapter I
Events, Observables and States ........................................... .
1 From classical to quantum probability ................................. .
2 Notational preliminaries ................................................ 2
3 Finite dimensional quantum probability spaces ......................... 6
4 Observables in a simple quantum probability space ..................... 9
5 Variance and covariance .............................................. 13
6 Dynamics in finite dimensional quantum probability spaces ............ 18
7 Observables with infinite number of values and the
Hahn-Hellinger Theorem ............................................. 22
8 Probability distributions on 0P('J£) and Gleason's Theorem ............. 31
9 Trace class operators and Schatten's Theorem ......................... 43
10 Spectral integration and Stone's Theorem on
the unitary representations of IRk ...................................... 53
11 Basic notions of the theory of unbounded operators ................... 60
12 Spectral integration of unbounded functions and
von Neumann's Spectral Theorem .................................... 65
13 Stone generators, characteristic functions and moments ................ 73
14 Wigner's Theorem on the automorphisms of 0P('J£) .................... 82
Chapter II
Observables and States in Tensor Products of Hilbert Spaces ............ 91
15 Positive definite kernels and tensor products of Hilbert Spaces ......... 91
16 Operators in tensor products of Hilbert Spaces ........................ 97
17 Symmetric and anti symmetric tensor products ........................ 105
18 Examples of discrete time quantum stochastic flows .................. 111
19 The Fock Spaces .................................................... 123
20 The Weyl Representation ............................................ 134
21 Weyl Representation and infinitely divisible distributions ............. 152
22 The symplectic group of'J£ and Shale's Theorem .................... 162
23 Creation, conservation and annihilation operators in r a ('J£) ........... 172
VIII Contents
Chapter III
Stochastic Integration and Quantum Ito's Formula ..................... 179
24 Adapted processes ................................................... 179
25 Stochastic integration with respect to creation,
conservation and annihilation processes .............................. 183
26 A class of quantum stochastic differential equations .................. 207
27 Stochastic differential equations with infinite
degrees of freedom .................................................. 221
28 Evans-Hudson Flows ................................................ 233
29 A digression on completely positive linear maps and
Stinespring's Theorem ............................................... 250
30 Generators of quantum dynamical semigroups and
the Gorini, Kossakowski, Sudarshan, Lindblad Theorem .............. 257
References 275
Index .................................................................... 283
Author Index ............................................................ 289

Preface
In the physicist's picture of quantum theory the notion of a real valued observable
has, like Brahma, three visible faces : (i) a spectral measure on the line; (ii) a
selfadjoint operator in a Hilbert space; (iii) a unitary representation of the real line
as an additive group. The equivalence of these three descriptions is a consequence
of von Neumann's spectral theorem for a, not necessarily bounded, selfadjoint
operator and Stone's theorem on the infinitesimal generator of a one parameter
unitary group in a Hilbert space. Real valued random variables on a classical
probability space (n,?Jf, P) when viewed as selfadjoint multiplication operators
in the Hilbert space L2(p} are special examples in the quantum description. This
suggests the possibility of developing a theory of quantum probability within the
framework of operators and group representations in a Hilbert space. The first
chapter of the present work is aimed at an exploration of this idea assuming only
a knowledge of elementary functional analysis and classical probability.
Through the papers of R.F. Streater [132] in 1969, H. Araki [9] in 1970 and
K.R. Parthasarathy and K. Schmidt [100] in 1972 there has emerged a string of
ideas exhibiting the close relationship between the infinitely divisible probability
distributions of Levy and Khinchin on the one hand and the projective unitary Weyl
representation of the Euclidean group of a Hilbert space in the associated symmet-
ric (boson) Fock space on the other. The Weyl representation is a neat summary
of the Weyl commutation relations or, equivalently, the canonical commutation
relations (CCR) and the second quantization homomorphism. In its infinitesimal
form the Weyl representation leads naturally to the fundamental notions of cre-
ation, conservation and annihilation operators which obey the (extended) CCR
and can therefore be interpreted as the source of Heisenberg's uncertainty prin-
ciple. Classical stochastic processes with independent increments are then realised
as distributions in the vacuum state of commuting families of observables which
are, essentially, linear combinations of the creation, conservation and annihilation
operators in the backdrop of a filtration determined by a time observable. Such a
realisation depends very much on the theory of tensor products of Hilbert spaces
and operators on them. The reader may find a fairly self-contained approach to
this theme in the second chapter.
In classical probability theory, infinitely divisible distributions are realised
from stochastic processes with independent increments of which the paradigm
examples are the standard Brownian motion and the Poisson process of a given
intensity. Starting with the pioneering efforts of N. Wiener, K. Ito, P. Levy and
J.L. Doob there has now evolved during the last six decades an extraordinarily
rich stochastic calculus around these and more general local semimartingales en-
compassing stochastic integration, Ito's formula, stochastic differential equations
and a variety of applications which constitute the fruit bearing orchard of modem
stochastic analysis. (See, for example, the book [66] of N. Ikeda and S. Watanabe).
The developments in the second chapter indicate the possibility of developing a
quantum stochastic calculus around the basic creation, conservation and annihi-
x Preface
lation operators arising from the Weyl representation of the Euclidean group of
a Hilbert space equipped with a continuous time observable in order to induce
a filtration. The motivation for such an attempt lies in the intuitive feeling that
the description of any dynamical phenomenon depends on the creation of some
objects, their preservation and motion for a period of time followed by their an-
nihilation and, finally, the repetition of such a process in cycles, all subject to the
laws of chance. A theory of quantum stochastic differential equations based on the
notions of creation, conservation and annihilation operator-valued processes and
time may possibly help in examining the point of view expressed above. Starting
with the preliminary ideas in [57], an attempt to develop a quantum stochastic cal-
culus along the lines mentioned already was made in 1984 by R L. Hudson and
K.R. Parthasarathy [59]. The principal aim of the third chapter is to present such
a calculus leading to quantum Ito's formula and highlight the following essential
features: (i) the classical Ito's formula for Brownian motion and Poisson process
can also be viewed as consequences of CCR and hence the Heisenberg's uncer-
tainty principle; (ii) canonical anticommutation relations (CAR) and CCR can be
derived from each other through stochastic integration; (iii) quantum dynamical
semigroups describing irreversible evolutions result from averaging the solutions
of a Heisenberg equation in the presence of noise, strengthening the belief that
irreversible evolutions are most likely to be shadows of reversible evolutions in an
enlarged universe; (iv) quantum stochastic flows (in the sense of Evans-Hudson)
arise from quantum stochastic differential equations; (v) classical Markov chains
can be realised in the quantum framework through stochastic difference and differ-
ential equations for observables with coefficients depending on appropriate group
actions.
The list of topics omitted and reasons for their omission is too long to be
enumerated and hence the number of friends and colleagues to whom I must
apologise is very high. Primarily, the subject is very young and dynamic and the
growth of literature in recent years is so explosive that I can hardly keep pace with
it. Secondly, notions like conditional expectation, stop time, local time etc., are
in the process of assuming a visible shape but the body of interesting physical as
well as mathematical examples seems to require much more nourishment. There
do exist several other approaches to quantum stochastic integration like the one
based on kernels in Guichardet's symmetric measure space due to H. Maassen [83]
and the other on C* - and W* -algebraic methods due to C. Barnett RF. Streater
and I.F. Wilde [16,17] and L. Accardi and F. Fagnola [5]. The interested reader
may find access to the voluminous literature on all these recent developments in
the Lecture Notes of proceedings of the workshops in quantum probability and its
applications [1,2] edited by L. Accardi, W. von Waldenfels and others as well as
the expositions of P.A. Meyer entitled "Elements de probabilites quantiques" in
the Seminaire de Probabilites [88]-[93].
The present work is a revised version of the mimeographed notes [111]
which owes its existence, primarily, to the influence of RL. Hudson with whom
Preface XI
the author had the pleasure of collaboration during the last nine years. Indeed, this
monograph was first proposed to be written jointly with him but the occurrence of
such a happy event was prevented by the difficulties involved in simultaneously
occupying the states of being the chairman of a department in a British university
during hard times and indulging in the lUxury of organising already discovered
results. To R.L. Hudson and his colleagues D. Applebaum and 1.M. Lindsay at
Nottingham, P.A. Meyer at Strasbourg and L. Accardi at Rome, I express my
heartfelt gratitude for uncountable conversations and exchanges of ideas on the
subject of quantum stochastic calculus as well as the nonmeasurable hospitality
I enjoyed during my visits to their departments and homes; to R.F. Streater who
showed and emphasised passionately that Fock space is a fertile soil for new
developments in probability theory; to my friend and colleague K.B. Sinha for
his enthusiastic collaboration and illuminating suggestions whenever analytical
difficulties obstructed imagination; to my devoted audience of three consisting of
S.K. Muthu, A. Mohari and B.V.R. Bhat during my year-long seminar at the Delhi
Centre of the Indian Statistical Institute during 1988; and to V.S. Varadarajan,
the guru who initiated me to the charms of the geomertry of quantum theory
through his marvellous lectures on G.W. Mackey's approach to the mathematical
foundations of quantum mechanics during the year 1965 in Calcutta prior to his
total exit from the Indian mathematical scene. I am deeply indebted to B.V.R. Bhat,
1.M. Lindsay and S.K. Muthu in checking the mushroom growth of mistakes in
my manuscript as well as suggesting ways for its improvement. Special thanks are
due to V.P. Sharma for his patient and elegant preparation of the manuscript in
TEX. Finally, no amount of thanks would suffice for the cheerful and encouraging
support I have received from Shyama on the domestic front where life in Delhi
in the sweltering summer heat of forty plus, accompanied by aandhi winds is an
everyday battle right from facing an erratic power and water supply to procuring
the daily necessities of life like bread, rice, vegetables and milk.
Chapter I
Events, Observables and States
1 From classical to quantum probability

The source of inspiration for doing quantum probability lies in the ingenious
methods adopted by physicists to compute probabilities of events concerning the
subatomic world of elementary particles where the laws of classical mechanics
break: down and the distinction between a particle and a wave becomes vague.
These methods lead to a generalisation of classical probability which may be
described as a study of observable quantities concerning any system subject to
chance and exhibiting the following basic features: (i) each observable quantity
behaves like a classical real valued random variable on a probability space and
hence has a probability distribution on the real line; (ii) it may be impossible to
measure simultaneously two individually observable quantities in the sense that
measurement of one may interfere with and destroy already existing information
about the value of the other; (iii) bounded observable quantities constitute a real
linear space in which the vectorial sum of two individually observable quantities
interfering with each other is different from the arithmetic sum of their actual
values.
In order to understand this generalisation from a mathematical point of view
we first observe that the expectation rEf of a real valued random variable on a
finite probability space {I, 2, ... , n} with probability distribution (Pl,P2, . .. ,Pn)
for the elementary outcomes 1, 2, ... , n respectively can be expressed in three
different ways:
o o
f~J
f(2) o
(1.1 )
o o
o o
f(2) o
o o
2 Chapter I: Events, Observables and States
where tr denotes the trace of a square matrix, namely the sum of all its diago-
nal elements and fh.fh, ... ,On are arbitrary angles. The first expression in (1.1)
underlines the idea that the space of all real valued random variabes is a real
linear space of dimension n and the probability distribution is a non-negative el-
ement of its dual. This is also the motivation for the famous Riesz representation
theorem [96] that integration with respect to a probability measure in the unit
interval is the only normalised non-negative linear functional on the Banach space
e[O, 1] of all continuous functions in the unit interval. The equivalent expressions
in the second and third lines of (1.1) enable us to view the probability distribution
(Plo]J2, ... ,Pn) as an element of the dual of the n 2 -dimensional real linear space
of all complex Hermitian matrices of order n and write the expectation value in the
language of operators in a Hilbert space using quantities like trace and scalar prod-
uct. The essence of quantum probability lies in taking into account the possibility
of using arbitrary Hermitian matrices or operators instead of only the diagonal
matrices that appear in (1.1) and pushing the basic ideas of classical probability
to their logical end with the tools of the extraordinarily rich theory of operators
in a Hilbert space. The innocuous looking angles OJ in the last line of (1.1) reveal
their influence as soon as the diagonal matrix is replaced by a non-diagonal one .
..;p;ei8j is the so called probability amplitude with the phaselactor e i8j •
Notes
For an extensive discussion of the role of notions like probability, amplitude,
phase-factor and interference see Feynman [41], [42].
2 Notational preliminaries
Throughout these lectures, unless otherwise specifically mentioned, by a Hilbert
space 'fJe we shall always mean a complex separable Hilbert space with scalar or
inner product (-,.) which is conjugate linear in the first and linear in the second
variable. In order to indicate the Hilbert space in question sometimes we denote
the scalar product by (-,. he. The norm (u, u) 1/2 of any element u in 'fJe will
be denoted by lIull. A sequence {un} is said to converge to a limit u in 'fJe if
lim lIun - ull = o. It is said to converge weakly if lim (v, un) = (v, u) for all
n~oo n~oo
V in 'Je. For any set S c 'fJe we write Sl.. = {ulu E 'Je, (u, v) = 0 for all v in S}.
Sl.. is a closed subspace of 'Je and Sl..l.. is the smallest closed subspace containing
S. We say that S generates Sl..l.. or S is total in Sl..l... For any vector space S
we denote its dimension by dim S.
en denotes the n-dimensional complex Hilbert space of all complex n x 1
matrices or column vectors with the standard inner product
2 Notational preliminaries 3
By the canonical basis in en we mean the orthonormal basis {e}, e2, . .. ,en}
where ei is the column vector with 1 in the i-th position and 0 elsewhere. Denote
the corresponding n-dimensional real Hilbert space by ~n and adopt the same
conventions. When n = I drop the superscripts and denote the Hilbert spaces 1 e
e
and ~1 by and ~ respectively. ~+ denotes the half interval [0,00) in ~.
The Hilbert space f? is the linear space of all sequences {an} of complex
scalars satisfying the condition I:n Ian 12 < 00. For any two such sequences
u = {an}, V = {b n } their scalar product is defined by (u, v) = I: n an bn. The
enumeration index may run through {I, 2, ... } or {O, I, 2, ... }. The canonical
orthonormal basis {en} in £2 is the sequence of unit vectors en with 1 in the n-th
position and 0 elsewhere.
If (S, CfF, 1-") is a a-finite measure space we denote by L2 (I-") the space of all 1-"-
equivalence classes of complex valued absolutely square summable (or integrable)
functions on S with scalar product
(f,g) = J fgdl-" = J f(x)g(x)l-"(dx), I,g E L2(1-").
If S is a topological space we shall denote by CfF s the Borel a-algebra generated

by the collection of all open sets. If h is a Hilbert space then L2 (I-", h) will denote
the Hilbert space of all I-"-equivalence classes of h-valued square summable maps:
L2(1-'" h) = {III: S -+ h, I is I-" -measurable and
J11/(x)ll~ I-"(dx) < oo}
J
with scalar product
(f,g) = (f(x),g(X))h I-"(dx).
If {~n} is a finite or countable family of Hilbert spaces we denote their direct sum
~ by EEln~n or ~1 EEl 'Je2 EEl ... and view ~n as a subspace of ~ by identifying
any u E 'Je n as an element of ~ with u as the n-tb summand and remaining
summands equal to O.
By a bounded operator T in ~ we mean an everywhere defined linear map
of ~ into itself satisfying IITII = sup IITul1 < 00. The set of all such bounded
Ilull=l
operators in ~ is denoted by (i}i\(~). The adjoint of a bounded operator T is the
unique operator T* satisfying (T*u, v) = (u, Tv) for all u, v in 'Je. (i}i\(~) is an
involutive Banach algebra with norm 11.11 and involution *. Furthermore, for any
T E (i}i\('Je), IITII = IIT*II = IIT*TI11/2. In other words (i}i\('Je) is a C* algebra. A
sequence {Tn} is said to converge in operator norm in (i}i\('Je) if lim IITn -Til = O.
n->oo
It is said to converge strongly if lim IITnu - Tull
n->oo
= 0 for every u in ~ and we
write s.limTn = T. It is said to converge weakly if lim (u, Tnv) = (u, Tv) for
n--+oo n--+oo
all u, v in 'Je and we write w.limTn = T.
n->oo
If >. is a scalar the same symbol will be frequently used to denote the
operator >.1, I denoting identity, For any T in 'lA('lie), R(T) = {Tulu E 'lie} and
N(T) = {ulTu = O} are called the range and null space of T respectively,
R(T) is a linear manifold and N(T) is a closed subspace. N(T*) = R(T).l.
T is said to be selfadjoint if T* = T. dim R(T) is called rank of T. We write
O('lie) = {TIT E 'lA('X), T = T*} and observe that it is a real linear space. An
operator T is said to be positive if (u, Tu) 2 0 for every u in 'X. A positive
operator is necessarily selfadjoint. If T 1, T2 are in O('lie) we write Tl 2 T2 if
Tl - T2 is a positive operator. 2 is a partial order in O('X). By a projection we
shall always mean an orthogonal projection onto a subspace of 'X. Denote the set
of all projections in 'lie by rzJ'('X). T is a projection if and only if T = T* = T2.
In particular, any projection E is a positive selfadjoint operator and 0 :S E :S 1.
Thus rzJ'('lie) c O('lie) c 'lA('lie). O('lie) is closed under the following non-associative
binary operations:
(X, Y) ~ i[X, Y] = i(XY - Y X)
(X,Y) ~ {X,Y} = XY + YX.
Given a family {Ea} of projections, VaEa and /\aEa denote respectively the
projections on the smallest closed subspace containing UaR(Ea) and the closed
subspace na R( Ea). If {En} is a finite or infinite sequence of projections then we
write VnEn = El V E2 V"" /\nEn = E 1 /\E2/\···. For any family {Eala E r}
of projections and any two projections E, F satisfying the relations E 2 E a ,
F :S Ea for all a, one has E 2 VaEa 2 Ef3; F :S /\aEa :S Ef3 for every f3 E r.
In other words Va and /\a can be interpreted as sup and inf operations in rg>('lie)
with the partial order 2. Writing E.l = 1 - E for any projection we observe that
E.l is a projection and R(E.l) = R(E).l, E.l.l = E and
(VaEa).l = (/\aE;,/\aEa).l = VaE;.
If E 2 F in rzJ'('lie) then E.l :S F.l.
If Ei, i = 1,2,3 are projections then El /\
(~V E 3) is not necessarily equal to (E 1 /\E2) V (El/\E3). Indeed, if Ul = U2 +U3
where U2, U3 are unit vectors satisfying (U2' U3) > 0 and Ei is the projection on
the one dimensional subspace CUi generated by Ui then
El /\ (E2 V E 3) = El i- 0 = (El /\ E 2) V (El /\ E3)'
If E 1, E2, E3 are any three commuting projections then El /\ (~ V E 3) = (El /\
E2)V(El/\E3). If {En} is a sequence of projections satisfying EiEj = 0 for i -:f j
then VnEn = El + E2 + ... where the right hand side is a strongly convergent
sum. Conversely, if VnEn = El + E2 + ... then EiEj = 0 for all i -:f j. For any
sequence {En} of mutually commuting projections /\nEn = s.limE1E 2 ··· En.
n
An element T in 'lA('lie) is called a contraction if IITII :S 1. The set of all such
contractions is denoted by Cf6('X). Cf6('lie) is closed under adjoint, multiplication and
weak convergence. An element T in 'lA('lie) is called an isometry or coisometry
according to whether T*T = 1 or TT* = 1. It is said to be unitary if it is
2 Notational preliminaries 5
both an isometry and a coisometry. If T is an isometry then TT* is a projection.

Denote by OU(71C) the set of unitary operators in 71C. If {Tn} is a sequence in
C(&(~), w.limTn = T and T is an isometry then s.limTn = T. In particular, weak
n----+oo n----+oo
and strong convergence are equivalent in OU(~) and they make OU(71C) a topological
group. OU(~) acts on the space ~(71C) through (U,X) -+ UXU- I. This action
leaves O(71C), C!P(~) and c(&(71C) invariant.
An element T in ~(~) is said to be invertible in ~(~) if there exists Sin
~(71C) satisfying ST = TS = 1. Such an S is unique, denoted by T- I and called
the inverse of T. T is invertible in ~(71C) if and only if N(T) = 0, R(T) = ~. In
such a case liT-III ~ IITII- I. All invertible elements in ~(71C) constitute a group
under multiplication.
A map T : 71C -+ 71C is called antilinear (or conjugate linear) if T (u + v) =
Tu+Tv, Tau = aTu for all u, v in ~ and scalars a. An antilinear map T is called
an antiunitary operator if R(T) = ~ and (Tu, Tv) = (v, u) for all u, v in 71C. It is
called a conjugation if, in addition, T2 = 1. Any antiunitary operator is the product
of a conjugation and a unitary operator. To any conjugation T there corresponds an
orthonormal basis {el' e2, . .. ,} satisfying T(I:iaiei) = I:iaiei for all sequences
{aI, a2, .. . 1I:ilaiI2 < oo}. Any antiunitary operator U is invertible and its inverse
U- I is also antiunitary. Furthermore, the map X -+ UXU-I,X E ~(71C) leaves
C!P(71C), O(71C) and OU(71C) invariant.
An element T in ~(~) is said to have finite rank n if dim(T) = n < 00.
Denote by .1>o(71C) the set of all operators of finite rank. If S, T E .1>o(~), X, Y E
~(~) then as + {3T, XTY and T* belong to .1>o(~), a, {3 denoting arbitrary
scalars. In other words .1>o(71C) is a two-sided ideal in ~(~) closed under adjoint.
An element T in ~ (~) is called a compact operator if for every sequence {un}
of unit vectors in 71C the image sequence {TUn} has a convergent subsequence.
We denote the set of all compact operators in ~ by .1>oc,(71C). Then .1>00(71C) is a
*-closed two-sided ideal in the algebra ~(71C). If {Tn} is a sequence in .1>00(71C)
and lim IITn - Til = 0 then T E .1>00 (71C) . .1>o(71C) is dense in .1>00(~) in the
n->oo
operator norm topology.
For any two elements u,v in 71C we define the operator lu)(vl by
lu)(vlw=(v,w)u for all win~, (2.1)
and summarise its properties in the form of a proposition.
Proposition 2.1: The map (u,v) -+ lu)(vl from 71C x 71C into ~(71C) satisfies the
following:
(i) lu)(vl is linear in u and conjugate linear in v;
(ii) (Iu)(vl)* = Iv)(ul;
(iii) lUI) (vlllu2) (v21· . ·Iun)(vn I = {IIf==-/ (Vi, Ui+1 )}luI) (vnl;
(iv) If u ¥- 0, v¥-O the range of lu)(vl is the one dimensional subspace Cu =
{AulA E C};
(v) II lu)(vl II = lIullllvll;
(vi) For any T in 0A(~), Tlu)(vl = ITu)(vl, lu)(vIT = lu)(T*vl;

(vii) An operator T is a projection with dimR(T) = 1 if and only if T = lu)(ul
for some unit vector u, In such a case R(T) = Cu;
(viii) If P is a projection and {el, e2, ... } is any orthonormal basis for the subspace
R( P) then P = I:i lei) (ei I where the right hand side is a strongly convergent
sum;
(ix) An operator T in ~ is of finite rank n if and only if there exists an orthonor-
mal set {Ul,U2, ... ,un} in ~ such that {Tut.Tu2, ... ,Tun} is a linearly
independent set of vectors and
n
T= L ITui)(Uil·
i=l
Proof: All the properties (i)-(ix) are straightforward from definitions. •
Notes
Regarding the efficiency of Dirac's convention that the scalar product (-,.) in a
Hilbert space is conjugate linear in the first and linear in the second variable and
I,) (-I can be interpreted as an operator satisfying Proposition 2.1, see Dirac [29]
and Vol. III of Feynman [42]. See also the discussion after Proposition 20.13.
3 Finite dimensional quantum probability spaces

We shall now describe the quantum analogue of a classical probability space with
n elementary outcomes or sample points. An event in classical probability is a
subset of the sample space which can equally well be described by the indicator
function of the same subset. The indicator function is a random variable assuming
at most two values 0 and 1. In the context of (1.1) such a random variable is a
diagonal matrix with diagonal entries 0 or 1. In other words it is a projection with
a diagonal matrix representation in the canonical basis of Cn . To introduce the
required quantum analogue we consider an n-dimensional Hilbert space ~ and
call any element of rzp(~) an event. We shall now highlight some of the most
natural properties of events. If E 1, E2 are events and El :S E2 we say that El
implies E 2 . The elements 0 and 1 in rzp(~) are called the null and certain events
respectively. The relation 0 :S E :S 1 for any event E is interpreted as the fact that
the null event implies E and E implies the certain event. If E is an event 1 - E
is called its complement. If Ei, i = 1, 2, ... , k are events ViEi is the event of
occurrence of at least one of the Ei' s whereas /\iEi is the event of simultaneous
occurrence of all the Ei'S. Two events E, F are said to interfere with each other if
[E, F] f::- O. A collection {Ea} of events is said to be mutually non-interfering if
[Ea,E a,] = 0 for any two a,a'. Suppose {Ei11 :S i :S k} is a commuting family
3 Finite dimensional quantum probability spaces 7
of projections, or equivalently, a family of mutually non-interfering events. Then

/\iEi = E 1E2··· Ek and
k
ViEi = El + 2)1 - Ed··· (1 - Ei-dEi.
i=2
If El,~, E3 are three mutually non-interfering events then El /\ (E2 V E 3)
(El/\ ~) V (El/\ E3). In general this need not hold. If E 1, E2 and El ~ are events
then El~ = E 2E 1. If E[,~ are events then El + E2 is an event if and only if
El~ = O. Anyone dimensional projection E in rJP(7Je) is an atom in the sense that
it cannot be expressed as the sum of two non-null projections. If {el' e2, ... ,en} is
an orthonormal basis for'1Je and Ei = lei)(eil is the one dimensional projection on
the subspace Cei then E[, E 2, ... ,En are non-interfering events and L.iEi = 1.
The family {O} U{Ei 1+ Ei2 + ... + Ei r , 1 ::; i 1 < i2 < ... < iT ::; n, r = 1,2, ... }
is closed under /\, V and complementation and hence constitutes a Boolean algebra
of events with n atoms. Given two such Boolean algebras ~[, ~2 of events with
exactly n atoms, there exists a unitary operator U such that ~2 = {U EU* IE E
~d. When the sample space of classical probability has n points or elementary
outcomes the Boolean algebra of all events has 2n elements with exactly n atoms.
In the corresponding quantum description, when the underlying Hilbert space '1Je
has dimension n :::: 2, the collection rJP(7Je) of all events has the cardinality of the
continuum and the set of atoms is a manifold whose real dimension is 2n - 2.
For any operator T on the n-dimensional Hilbert space 7Je and any orthonor-
mal basis {e[,e2, ... ,en} the quantity L.j(ej,Tej) is independent of the basis,
called the trace of T and denoted by tr T. The map T -+ tr T satisfies the following
properties:
(i) tr(aTl + j3T2) = atrTl + j3trT2 for all scalars a,j3;
(ii) trT1T2 = trT2Tl. In particular, if T is invertible then trTl = trTT1T- 1;
(iii) tr T = sum of the eigenvalues of T inclusive of multiplicity;
(iv) trT* = trT;
(v) trT:::: 0 whenever T :::: 0;
(vi) The space ~(7Je) with the scalar product (Tl' T2) = trTtT2 is a Hilbert space
of dimension n 2 ;
(vii) If A : ~(7Je) -+ C is a linear map such that A([X, Y]) = 0 for all X, Yin
~(7Je) and A(J) = n then A(X) = tr X for all X.
A positive operator p of unit trace is called a state. The set of all states in
'1Je is denoted by :/('1Je). For any fixed state p the triple (7Je,rJP('1Je), p) is called
a simple or a finite dimensional quantum probability space. For any E in rJP(7Je)
the quantity tr pE is called the probability of the event E in the state p. Since
trpE = L.i(Ui,PUi) where {Ul,U2, ... } is an orthonormal basis for the range of
E it follows that 0 ::; tr pE ::; I for any event E. If E 1, E2, . .. ,Ek are events
and EiEj = {jij Ej for all i, j then El + ~ + ... + Ek is the event that at least
one of the Ei'S occurs and its probability tr p(EI + E2 + ... + E k ) = ~i tr pEi . In
this sense probability is an additive function on CZP('Je) with values in [0,1]. Later
°
we shall prove a converse of this result known as Gleason's Theorem (Theorem
8.9). In any state the probability of the null event is and the certain event is 1.
If Ph P2 are two distinct states then (U, PI u) =I- (U, P2 u) for some unit vector U
and hence for the event E = Iu) (ul, tr PIE =I- tr P2E. In other words two distinct
states yield two distinct probability distributions on CZP('Je). It follows from the
spectral theorem that every state P can be expressed as P = ~jpjluj)(ujl where
Pj > 0, ~Pj = 1 and Uj, j = 1,2, ... is an orthonormal set of eigenvectors of
P such that pUj = PjUj for each j. This shows that the extreme points of the
convex set 9'('Je) are precisely the one dimensional projections in 'Je. Anyone
dimensional projection is called a pure state. Every state P can be expressed as
a convex combination of at most k pure states where k = rank p. If P is a pure
state then p = u) (u for some unit vector U and tr pE = (u, Eu) for any event
1 1
E. By abuse of language we call any unit vector U in 'Je a pure state but, truly
speaking, it stands for the pure state 1u) (u I. In this context it is worth noting that
in a sample space of n elementary outcomes in classical probability the set of
all probability distributions is a convex set whose extreme points are precisely
the n degenerate distributions. In its quantum analogue the set of pure states is a
manifold of dimension 2n - 2. If U is a unitary operator, p is a state and E is an
event then U pU* is a state, U* EU is an event and tr U pU* E = tr pU* EU. The
unitary group OU('Je) acts on 9'('Je) and CZP('Je) and the action is transitive on the set
of pure states and on the set of atomic events. The probability of an event E under
the "transformed" state U pU* is also the probability of the "transformed" event
U* EU under the state p. This may be fruitfully compared with the fact that in an
n-point sample space the group of permutations acts on the space of probability
distributions and on the algebra of events and the action is transitive on the set
of degenerate distributions and the set of atomic events. It is the richness of the
group OU('Je) acting on events and states that makes quantum probability worth
exploring even in finite dimensions.
n
Exercise 3.1: (a) Let = {1,2, ... ,n},2F k = {EIE C n,#E = k} where #E
is the cardinality of E. Then #2Fk = (~).
(b) Let 'Je be a complex Hilbert space of dimension n. Suppose CZPk('Je) = {EIE E
CZP('Je), dim R(E) = k}. Then the unitary group OU n = OU('Je) of dimension n 2 acts
transitively on CZPk('Je) by the action (U, E) - t U EU* with isotropy subgroup
OUk X OU n - k of dimension k2 + (n - k? In particular, CZPk('Je) is a manifold,
indeed a homogeneous space, of dimension n 2 - k 2 - (n - k? = 2k(n - k).
(c) E E CZPk('Je) if and only if E = EI + E2 + .. , + Ek where each Ej is a one
dimensional projection (or atom).
(d) In (b) if'Je is a real Hilbert space then dim CZP k ('Je) = k( n - k).
Notes
As a general reference on the logic of events in quantum mechanics and further
literature on the subject we refer to the book of Varadarajan [135].
4 Observables in a simple quantum probability space 9
4 Observables in a simple quantum probability space

Let 'JC be a Hilbert space of dimension n < 00. Elements of (f)(~), i.e., Hermitian
operators in ~, are called observables. An observable in quantum probability is
what a random variable is in classical probability. Whereas random variables on an
n-point sample space constitute a real linear space of dimension n, observables on
the n-dimensional complex Hilbert space constitute a real linear space of dimen-
sion n 2 • Any observable X, being a selfadjoint operator, has the spectral resolution
X = 2:i xiEf where XI. X2, ... are its distinct eigenvalues and Ef is the event
that X takes the value Xi. The spectral resolution of X must be compared with the
property that every random variable 1 on an n-point sample space can be expressed
as 1 = L.iXiIEi where {XI, X2, ... } is the range of 1, Ei is the subset of the sample
space where 1 assumes the value Xi, i.e., Ei = 1- 1({ Xi}), and lEi is the indicator
of Ei. If 9 is a real valued function on the real line IR then g(X) = L.ig(Xi)Ef
is also an observable. g(X) assumes the values g(Xt},g(X2), ... but it is possible
that these values need not be distinct. The event that g(X) assumes the value y
is given by 2:i:9(Xi)=Y Ef. The mapping 9 --+ g(X) is a homomorphism from
the algebra of real valued functions on IR into the algebra ~(~). Two observables
X, Y are said to interfere with each other if [X, Y] # O. If Xi, i = 1,2, ... ,k
are mutually non-interfering observables in the sense that XiXj = XjXi for all
i,j then there exists a non-interfering family of events {EI' Eh, . .. ,Em}, m :::; n
such that ErEs = 0 for r # s and each Xi has the form Xi = L.rxirEr. Er is the
simultaneous occurrence of the events "Xi takes the value Xir", i = 1,2, ... , k. If
U E OU('JC) and X is an observable with spectral resolution L.iXiEf then U XU*
is an observable with spectral resolution L.iXiUEfu-l. In other words UXU- I
and X take the same values for any unitary operator U. (The same holds when U
is an antiunitary operator). This may be compared with the fact that, in the classi-
cal sample space with n points, if 1 is a random variable and 7r is a permutation
of the sample points 1 and 107r are random variables with the same range. Events
are observables assuming at most two values 0,1.
Let p be a state and let X be an observable with spectral resolution X =
L.iXiEf. The probability of the event Ef, i.e., X takes the value Xi in the
state p, is equal to tr pEf. In other words, in the state p, the observable X has
the discrete distribution with mass tr pEf at Xi, i = 1,2, .... In particular, X
has expectation L.iXi tr pEf = tr PL.iXiEf = tr pX. The k-th moment of X in
the state p is equal to L.iX~ tr pEf = tr pL.iX~ Ef = tr pXk. More generally,
for any real valued function 9 on IR the expectation of g(X) in the state p is
equal to L.ig( Xi) tr pEf = tr pg( X). The characteristic function of X in the state
p is equal to L.je itXj tr pEl = tr pe itX . It is interesting to note that the map
t --+ Ut = eitX is a continuous homomorphism from the additive group IR into the
multiplicative group OU(~). Conversely, every continuous homomorphism t --+ Ut
from IR into OU(~) can be uniquely expressed by Ut = eitX where X is an
observable. This correspondence between observables and one parameter groups
of unitary operators through the identification Ut == eitX plays a fundamental role
in quantum probability, If u is a unit vector in 'JC then in the pure state u (i.e"
when p = lu)(uj) X has the distribution with mass (u,Efu) at Xi for each i,
expectation (u, Xu), k-th moment (u, Xku) and characteristicfunction (u, eitX u),
If X is a non-negative oQservable or, equivalently, X is a positive operator then
tr pX ~ 0 for any state p, Thus expectation in a state is a non-negative linear map
from O('JC) into ~ with value unity for the observable I,
For any state p, observable X and unitary operator U on 'JC we have
tr pU* XU = trU pU* X, (4,1 )
This may be compared with the property that for any permutation 7f of the elemen-
tary outcomes in an n-point sample space n, a random variable I and a probability
distribution p on n, the expectation of the transformed random variable lo7f with
respect to the distribution p is also the expectation of the random variable I with
respect to the transformed probability distribution P7f-], Thus (4,1) may be inter-
preted as a simple change of the variable formula for expectation in the context
of quantum probability,
Example 4.1: Let'JC = ([2 with its canonical basis eo, e] where eo = (6), e] = (~),
Then O('JC) has the basis
0"0= (01 0)1 ,0"]= (01 01) ,0"2= (0i -i)

0 ,0"3= (10 -10) ' (4,2)
O"j, j = 1,2,3 are the well-known Pauli spin matrices of quantum mechan-
ics, {O"j, 0 :S j :S 3} is an orthonormal basis of O('JC) with the scalar product
(X, Y) = ~ tr XY, The Pauli spin matrices obey the following multiplication and
commutator (or Lie bracket) tables:
O"iO"j [O"i' O"j 1
0"1 0"2 0"3
0 0
0"1 0"2 0"3
0"] 0"0 i0"3 -i0"2 0"] 2i0"3 -2i0"2 (4.3)
0"2
0"3
-i0"3 0"0 iO"l
i0"2 -iO"] 0"0
0"2
0"3
-2i0"3
0
2i0"2 -2iO"l
2iO"]
In the pure state eo the observables 0"1 and 0"2 assume the values 1 and -1
!
with equal probability whereas 0"3 assumes the value 1 with probability 1. An
elementary computation shows that
. (cosr ie-ill sin r)
expz(t]O"l + t20"2) = . ill .
ze sm r cosr
,
where t] + it2 = re ill denotes polar decomposition. In particular,

(eo, ei(t 1a l +t2 a 2)eo) = cos(ti + tD]/2 = ¢(t], t2)'
It is to be noted that ¢(t1' t2) is not the characteristic function of any probability
distribution in the plane but, for fixed t], t 2, ¢( tt], tt 2) = cos t( ti + tD 1/2 is the
characteristic function of the probability distribution assigning equal probability
4 Observables in a simple quantum probability space 11
1 to (tI + tDI/2 and -(tI + t~)1/2. From the second table in (4.3) we have
[O"J, 0"2l = 2i0"3. In other words 0"1 and 0"2 interfere with each other and do not
admit a joint distribution in the pure state eo.
In general, it is not clear under what conditions on a unit vector U and a family
of observables X I, X 2, ... ,Xk in a Hilbert space 'Je, the function 'IjJ( tJ, ... ,tk ) =
(u,(expi~jtjXj)u) is the characteristic function of a probability distribution
in ~k.
Going back to the example under consideration, we observe that 0"1 + 0"2
has eigenvalues ±v'2. Whereas 0"1 and 0"2 take values ±l their vectorial sum in
O('Je) takes values ±v'2. This essentially non-linear feature in the behaviour of the
actual values assumed by the sum of two interfering observables stands in sharp
contrast to the way sums of random variables behave in classical probability. It
is to be noted that for two non-interfering observables X, Y assuming the values
{XI, X2,.·· ,xd and {YI, Y2, .. · ,yt} respectively, the observable X + Y assumes
values from the set {Xi + Yjll ~ i ~ k, 1 ~ j ~ l}.
Example 4.2: Let 'Je = en

with its canonical orthonormal basis {ej, 0 ~ j ~
n - I}. For any set {XO,XI, ... ,xn-d of n distinct real numbers consider the
pure state eo and the observable X = ~jxjlej}(ejl. In the canonical basis X has
the diagonal matrix representation
Xo o o
o
X~:
(
o o
For any unitary operator U = (( Uij)) the observable U* XU takes the values
Xo, XI, . .. ,Xn-I and
n-I
(eo, (U* XU)k eo ) = (Ueo,XkUeo) = L xJlujol2. (4.4)
j=O
In other words, in the pure state eo the k-th moment of the observable U* XU is
given by the right hand side of (4.4) or, equivalently, U* XU has the distribution
with probabilities IUjol2 for the respective values Xj,O ~ j ~ n - 1.
Given an arbitrary probability vector (PO,PI, ... ,Pn-J) one can construct a
unitary matrix U = (( Uij)) by choosing its zero-th column to be Lj~d .jfije iIJj ej
where 00 ,01, .. . ,On-I are arbitrary angles and the remaining columns in such
a manner that all the columns of U together constitute an orthonormal basis for
'Je. Choosing X = ~xjlej}(ejl we note that, in the pure state eo, the observ-
able U* XU has the distribution assigning probabilities Pj to the values Xj,j =
0,1,2, ... ,n - 1 respectively. In other words, any given probability distribution
with probability Pj for the value Xj, 0 ~ j ~ n - 1 is realised in the fixed pure
state eo by a whole manifold of observables U* XU. Indeed, this manifold has
dimension (n - l)(n - 2). In particular, every probability distribution on the real
line supported by a finite set of cardinality not exceeding n can be realised as the
distribution of an observable in en
in the fixed pure state eo. It is worth observing
that this cannot be achieved through random variables on any fixed probability
space of finite cardinality. The abundance of choice for the unitary operator U in
the present discussion indicates the tremendous flexibility in realising a distribu-
tion through the observables of quantum probability in contrast with the random
variables of classical probability.
Exercise 4.3: Let X = (p - q)O'3 - 2..jPQO'I in the notations of Example 4.1,

where 0 :::; p :::; 1, q = 1 - p. In the pure state eo, X assumes the values 1 and -1
with probabilities p and q respectively.
c
Exercise 4.4: Let TJC = C2n ,
0 0 0 0
C ~) y=: ~)
X= 0 0'3 0 a2 0
0 0 0'3 0 an
whereO'3= (1 0) 0
-
1 ,aj= (Xj
Yj
yj) ,Xj,Yj EII\\,Xj+Yj
-Xj
lib 2 2 =1, 1:::;J:::;n,
.
and 0'3 is repeated n-fold in X. Then X and Y are observables taking the values
±1 but X + Y takes the values ±J2(1 + Xj), j = 1,2, ... ,n.
Exercise 4.5: Let '/1f, be an n-dimensional Hilbert space with orthonormal basis
{eo, eJ, . .. ,en-t}. Define the truncated annihilation and creation operators:
n-2 n-l
L = L lej)(ej+ll,L* = L lej)(ej-ll· (4.5)
j=o j=l
Then Leo = 0, Lej = ej-l for 1 :::; j :::; n - 1, L*ej = ej+l for 0:::; j :::; n - 2,
L*en-l = O. The following holds:
k
(L + L*)keo = LCk(r)er, 0:::; k:::; n - 1 where eo(O) = 1,
r=O
Ck(r) = ck-l(r + 1) + ck-l(r - 1) if r ~ 1,
= Ck-l(l) if r = O.
These recurrence relations imply
Ck(O) = 0 if k is odd, Ck(O) = ~2~\ if k = 2m :::; n - 1.
In particular, the first n - 1 moments of the observable X = !(L + L*) in the

pure state eo coincide with the first n - 1 moments of the density function
f(x) = {
1(1 - X2)1/2 if Ixi :::; 1, (4.6)
o
11"
otherwise.
5 Variance and covariance 13
It is to be noted that the first n - 1 moments determine the distribution of X

uniquely. The distribution with density function (4.6) is known as the standard
1
Wigner distribution or semicircle law [141]. In particular, the distribution of (L +
L *) in the pure state eo converges weakly to the standard Wigner distribution as
the dimension of the underlying Hilbert space increases to 00. (See Exercise 13.11,
20.18 (b).)
Notes
Wigner's semicircle law is discussed in Wigner [141] and for a survey on the sub-
ject see the paper by Trotter [134]. For an interesting interpretation of Wigner' s dis-
tribution as a limit law in a non-commutative central limit theorem, see Voiculescu
[136] and Speicher [125].
5 Variance and covariance

Let p be a state in a finite dimensional Hilbert space 'Je. Then p induces a non-
negative definite sesquilinear form Kp on !J3('Je) defined by
Kp(X, Y) = tr pX*Y, X, Y E !J3('Je). (5.1)
The covariance between X, Y E !J3('Je) in the state p is defined to be the quantity
cov p(X, Y) = tr pX*Y - (tr pX* )(tr pY). (5.2)
If tr pX = m(X) then
covp(X, Y) = Kp(X - m(X), Y - m(Y)). (5.3)
This shows that cov p is also a non-negative definite sesquilinear form on !J3('Je).
If Xl, X 2 , .. . ,Xk are elements of !J3(';JC) then the k-th order matrix whose ij-th
element is cov p(Xi' Xj) is called the covariance matrix of Xl, X 2, .. . ,Xk . Even
though observables take only real values, the covariance defined as above between
two observables may tum out to be a complex number thanks to interference
between them. For example, the covariance between the Pauli spin observables a]
and a2 in any state p is equal to itrpa3 - (trpad(trpa2) and trpaj is real for
each j. If X and Yare non-interfering observables then tr(pXY)* = tr Y X P =
tr pY X = tr pXY which implies that the covariance between X and Y in the
state p is real.
Proposition 5.1: Let p E :J'(';JC) , X E O(';JC) and let varp(X) denote the variance
of the distribution of X in the state p. Then
varp(X) = tr pX2 - (tr pX)2 = tr p(X - tr pX)2 = COV p(X, X). (5.4)
The distribution of X in the state p is degenerate at a point x if and only if
Xv = xv for any v E R(p). In particular, the distribution of X in a pure state u
is degenerate at a point x if and only if Xu = xu.
For any pure state u and observables X, Y the following inequality holds:
varu(X) varu(Y) 2': ~I(u, i[X, YjU)12. (5.5)

14 Chapter I: Events, Observab1es and States
Proof: The first part of (5.4) follows from the fact that the first and second mo-
ments of the distribution of X in the state p are tr pX and tr pX 2 respectively. The
second and third part of (5.4) are immediate from the properties of trace and (5.1)-
(5.3). To prove the second part of the proposition choose an orthonormal basis
Ul, U2, ... , Uk of eigenvectors of p for R(p). Then PUi = PiUi, Pi > 0, ~Pi = 1.
If tr pX = m then (5.4) implies
k
varp(X) = 2::Pill(X - m)uil12

i=!
and the right hand side vanishes if and only if XUi = mUi for each i, i.e.,
Xu = mu for each U in R(p). This proves the second part. In the case of a pure
state, the range of p is one dimensional and hence the third part is immediate. To
establish (5.5) we write a = (u, Xu), b = (u, Yu). Then a and b are real scalars
and
(u, i[X, Y]u) = i(u, [X - a, Y - b]u)
= i{ ((X - a)u, (Y - b)u) - ((Y - b)u, (X - a)u)}
= -2Im((X - a)u, (Y - b)u)
where 1m denotes imaginary part. By Schwarz's inequality
~I(u, i[X, Yju)1 :::: II(X - a)ullll(Y - b)ull
= [varu(X) varu(y)p/2.
Proposition 5.1 reveals one of the most characteristic features of observables

•
in quantum probability. As already observed, the extreme points of the convex
set of all probability distributions on an n-point sample space are degenerate
distributions under which every random variable has a degenerate distribution or,
equivalently, zero variance. But in a pure state there will always exist observables
with non-degenerate distributions whenever dim ~ 2:2. (See Exercise 13.10.)
Consider two observables X, Y which do not commute with each other.
Then there exists a unit vector u and a real scalar b -=I- 0 such that i[X, Yju = bu.
Then (5.5) implies
Thus any decrease in the variance of X or Y has to be compensated for by an

increase in the variance of the other so that this inequality holds. This is interpreted
as the impossibility of measuring both X and Y with total precision in this pure
state u. Inequality (5.5) is an abstract mathematical description of the well-known
Heisenberg uncertainty principle in quantum mechanics.
It is to be noted that for observables X, Y whose covariance in a state p is

real one has
1
covp(X, Y) = 2: {varp(X + Y) - varp(X) - varp(Y)}.
One may use the right hand side of this equation as an alternative definition of
covariance. This has the advantage that it can be estimated by repeated individual
measurements on the observables X, Y and X + Y.
Example 5.2: To any point .;r. in ~3 with coordinates XI, X2, X3 associate the
observable iT(.;r.) = XliTl + X2iT2 + X3iT3 in (2 where iTj,j = 1,2,3 are the Pauli
spin matrices. Then by (4.3)
(I b2 1~2)
where Zl = X2Y3 - X3Y2, Z2 = X3Yl - XIY3, Z3 = XIY2 - X2Yl· Indeed, K is the
vector cross product of .;r. and JL In the state p = the observable
iT(.;r.) has binomial distribution with probability 1/2 for each of the values ± l.;r.l
and
cov p (iT(.;r.),iT(1!)) =.;r.' JL = LXjYj.
j
n
Since for any third order positive semidefinite matrix C with real entries of the
form
c~ G~
we can find three unit vectors .;r., y, K in ~3 such that .;r.. Y = 0:, y. K = I' K·.;r. = (3, it
follows that C can be realised asthe covariance matri; of three observables iT(.;r.),
iT(JL), iT(K) each having a binomial distribution with probability 1/2 for the values
± 1 in the state p.
In the context of Example 5.2 the following proposition is of great interest.
Proposition 5.3: (Bell's inequality [18] ) Let ~i,i = 1,2,3 be three random
variables defined on a probability space and taking values in the interval [-1, 1].
Then
Proof: For any three scalars x,Y,z in [-1,1] we have (1 + x)y ::::: 1 + X and
hence 1 - xY ~ Y - x. Interchanging x, Y in this inequality and using both, we
obtain 1 - xY ~ ly - xl ~ lz(y - x)l· Substituting X = 6, Y = 6, z = 6 and
taking expectations, we have
•
Example 5.4: Let a,{j be angles such that cos 2 {j < cos 2 a < Icos{jl. Then the
matrix
cos 13
c= ( cO~{j 1
- cos 13 )
- c~s2a
- cos 13 -cos2a
is positive definite but cannot be the covariance matrix of three ± 1 valued random
variables of expectation O. Indeed, positive definiteness follows from the fact that
the determinant of C is 4 sin2 a( cos 2 a - cos 2 13). If there exist ± 1 valued random
variables of expectation 0 with covariance matrix C then we have
11E66 - 1E661 = 21 cos 131 > 2cos 2 a = I -1E66
which contradicts Proposition 5.3. However, the discussion in Example 5.2 shows
that C can be realised as the covariance matrix of three observables in (:2 assuming
the values ±1 with equal probability in the state p = (1~2 1~2)'
We call any observable assuming only the values ±1 a spin observable.
Example 5.2 leads us to questions concerning the availability of room in a fi-
nite dimensional Hilbert space for the construction of spin observables with a
preassigned covariance structure. For example, given a positive definite matrix
C = (( Cij )) of order n with Cii = I, what is the minimal dimension of the
Hilbert space in which there exist n spin observables Xl. X2, ... ,Xn and a state
p for which tr pXj = 0, tr pXiXj = Cij, I :::; i, j :::; n? Given a state p in a
Hilbert space of dimension n, what is the maximum number k of spin observables
Xl, X 2 , I ... ,Xk such that tr pXj = 0, tr pXiXj = Oij' I :::; i, j ::::; k? Even in
classical probability questions of this kind look interesting. In a 2n-point sample
space where all points occur with equal probability, what is the maximum num-
ber of mutually uncorrelated random variables assuming only the values ± I and
having mean O? If this number is 2n - I we get a Hadamard matrix [52] of order
2n with first row having ones only and the remaining rows having ±I as entries.
Any two rows in such a matrix are orthogonal. We now present a weak partial
answer to the first question.
Proposition 5.5: Let C = (( Cij )) be a positive definite matrix with Cii = 1

for each i, I :::; i, j :::; n. Then there exists a positive integer k :::; n + 1, spin
observables Xl, X 2 , ... ,Xn and a pure state u in a Hilbert space of dimension k
such that (U,XiU) = 0, (U,XiXjU) = Cij, I :::; i,j :::; n.
Proof: Consider the n + I-th order positive semidefinite matrix
c~ (:~~) (1,I/h,LH,I/h)+ (r 14: H· 0)

= ((Cij)).
If rank C = k there exists a Hilbert space '1Je of dimension k and vectors Uj, 0 ::;
j ::; n in '1Je such that Cij = (Ui' Uj ),0 ::; i, j ::; n. In particular each Uj is a
unit vector. Let U = UO,Xj = 2Iuj)(ujl- 1,1 ::; j ::; n. Then Xj is a spin
observable and (u, Xju) = 21(uo, Uj) 12 - 1 = O. Similarly, a routine computation
using Proposition 2.1 shows that (U,XiXjU) = Cij' •
Exercise 5.6: (a) Equality in (5.5) is obtained if and only if one of the following
conditions holds:
(i) U is an eigenvector for X or Y;
(ii) there exist scalars a E R f3 E C such that (X + iaY)u = f3u.
(b) In particular, when X = ai, Y = a2 are the first two Pauli spin matrices,
equality in (5.5) is attained if and only if the pure state U belongs to the set
{ (~~~:0 ) ,( ±~os~: 0 ), 0::; 0 < 1r } .
(See Exercise 13.10.)
Exercise 5.7: If there exist observables Xi, 1 ::; i ::; n and a state p in a Hilbert
space '1Je such that ((trpXiXj)), 1::; i,j::; n has rank k then dim'1Je;:::.../k.
Exercise 5.8: Let p be a state of rank k in a Hilbert space of dimension n.

Suppose X" X 2 , ••• ,Xm is a sequence of spin observables satisfying tr pXi = 0,
tr pXiXj = Oij for alII::; i,j ::; m. Let p = L~=I pjluj)(ujl be the spectral
resolution of p so that Pj > 0 for every j and UI, U2, 1... ,Uk is an orthonormal
!
sequence. Define Po = 1, Pi = (1 - Xd. Then
(1) tr pPiPj = ! if i = j and = ! if i -I- j;
(2) The rank of the matrix Ar = (( (PiUr, PjU r ) ))o~i,j~m does not exceed n;
1 1/2 1/2 .. .
k
( 1/2 1/2 1/4 .. . 1/2)
1/4
(3)
~PTAr = ~~~ ~~: ~~~ ~~: 1/4 :;: has rnnk m +I;
1/2 1/4 ...
(4) m::; nk - 1. If k = 1 it is possible to attain the bound m = n - 1. If n
is odd and p = n -I I then there does not exist a spin observable X with
trpX = O.
Notes
For a physical interpretation of Bell's inequality in the context of the famous
EPR paradox in quantum theory, see Bell [18] and the Notes on Chapter IV of
Varadarajan [135].
When two observables X, Y do not commute with each other and hence
cannot be measured simultaneously, it is natural to wonder what is the covariance
between X and Y. In the context of Gleason's Theorem in Section 8 one may at-
tempt to formulate the notion of variance axiomatically and then define covariance
as in the discussion preceding Example 5.2.
Suppose V : O('Je) ---) IR+ = [0, 00) is a map satisfying the following:
(i) V(X) = 0 if and only if X is a real scalar multiple of the identity;
k
(ii) for any finite number of commuting observables XI ,X2 , ... ,Xk,v( L XiXi)
i=1
is a positive semidefinite quadratic form in (x I, ... , Xk) E IRk.
Then it is not unreasonable to expect the existence of a state P such that
V(X) = tr pX 2 - (tr pX? for all X. The same problem may be posed when
O('Je) is the real linear space of all Hermitian elements of a *-unital subalgebra of
9A('Je).
6 Dynamics in finite dimensional quantum probability spaces

In an n-point sample space n the semigroup of all maps from n into itself has
only nn elements and hence there is no possibility for constructing any continuous
dynamics in n. But the picture in a simple quantum probability space is drastically
different due to the existence of a large unitary group in a Hilbert space. To
illustrate this point consider any observable H in the n-dimensional Hilbert space
'Je. As already observed, the map t ---) e- itH is a continuous homomorphism from
the additive group IR into the multiplicative topological group !JU('Je) of all unitary
operators in 'Je. Conversely, if t ---) Ut is any such homomorphism from IR into
!JU('Je) there exists a unique observable H such that Ut = e- itH for all t. H is
called the (Stone) generator of the one parameter group {Utlt E IR}. First we shall
describe a dynamical picture in which observables remain fixed but states change
in time as follows: if p is the state at time 0, the state Pt at time t is given by
Pt = e- itH pe itH . This may be equivalently described by the differential equation
dPt = -z'[H , Pt,1

dt Po = p. (6.1 )
If X is any observable and the state at time 0 is p then the expectation of X at

time t is tr PtX. If H commutes with P then Pt = P for all t and P is called a
stationary state. If P is a pure state Iu) (u I then it follows from Proposition 2.1
that Pt = IUtu) (Utul is also a pure state. The expectation of any observable X at
time t is then (Ut, X Ut) where Ut = Ut U obeys the differential equation
idut _ H (6.2)
dt - Ut, Uo = u.
This is called the Schrodinger equation for the dynamics of pure states. H com-
mutes with Iu)(ul if and only if Hu = AU for some real scalar A. In such a case
Ut = e-itAu and IUt)(utl = lu)(ul for all t.
6 Dynamics in finite dimensional quantum probability spaces 19
Equation (4.1) enables us to present a dual picture of the dynamics described

by (6.1). Here the state p is fixed but the observable X at time 0 changes to the
new observable Xt = eitH X e -itH at time t. The expectation of X t in the fixed
state p is equal to tr pXt = tr PtX where Pt is given by (6.1). The dynamics of
X t is described by the differential equation.
dX t
dt = Z·[H, X]t, Xo=X. (6.3)
This is called the Heisenberg equation for observables. If H commutes with X

then X t = X for all t. The observables i[H, X] and -[H, [H, X]] are respectively
called the velocity and acceleration of X when the dynamics is described by H.
H is called the energy operator or Hamiltonian for the dynamics.
Like differentiation or the action of a vector field in the commutative algebra
of smooth functions on the line or a manifold, the notion of velocity of an observ-
able with respect to the energy operator H yields the linear map £ : X --+ i[H, X]
in the non-commutative algebra 073('3£) satisfying
£(XY) = £(X)Y + X£(Y), £(X*) = £(X)*. (6.4)
Any linear map £ on 073('3£) satisfying (6.4) is called a *-derivation. It is a theorem

in algebra that every *-derivation £ on 073('3£) has the form £(X) == i[H, X] for
some selfadjoint operator H in 073('3£). H is determined uniquely by £ up to an
additive real scalar multiple of the identity. In other words, energy is unique up
to an additive constant.
If the energy operator H in the Heisenberg equation (6.3) has the spectral
resolution H = L;jAjEj where A, < A2 < ... are the distinct eigenvalues and
E" E 2 , ... are the respective spectral projections then the expectation of X t with
respect to the state p is given by
tr pXt = tr pL;jEjX Ej + 2 Re L tr pEjX Ekeit(Aj -Ak).

j<k
Thus tr pXt is a superposition of trigonometric functions of t and lim tr pXt need
t-+oo
not exist. In particular, if all the differences Aj - Ak are integral multiples of a
single scalar A then tr pXt is a periodic function of t with period 211" A-\. It is
interesting to note that
and compare it with the situation in the case of a finite state continuous time
Markov chain determined by a one parameter semigroup {Pt , t 2: O} of stochastic
matrices where, for every function f on the state space, lim Ptf = P00 f exists
t-+oo
and P~ = P oo is also a stochastic matrix. In (6.5) the map T : X --+ L;jEjXEj
is idempotent and identity and positivity-preserving on 0('3£).
Example 6.1: Let aj,j = 0, 1,2,3 be as in (4.2). Then (4.3) implies

(i) °
[a3, [a3, aill = 4ai if i = 1,2, and otherwise;
(ii) e- itUj = ao cos t - iaj sin t;
(iii) eituj ake-ituj = ak cos 2t + iajak sin 2t, j -::J. k, 1 ::; j, k ::; 3;
(iv) If X t = ei t/2Ujake-i t/2Uj, 1 ::; j, k ::; 3, j -::J. k and p = (1~2 1~2)

then tr pXt = 0, tr pXtXs = cos(t - s).
In other words {Xt } is a "process" of ±1 valued observables for which the
covariance between X t and Xs in the state p is cos(t - s). Each X t assumes the
values ± 1 with probability 1/2. On the other hand if {~t} is a classical stochastic
process of real valued random variables for which lE~t = 0, lE~t~s = cos( t - s) for
all s, t then there exists a complex valued random variable ( = ~ + i'f/ where ~,'f/
are real, IE~ = 1E'f/ = 1E~'f/ = 0, lEe = 1E'f/2 = 1 and ~t = ~((eit + (e- it ) for all t.
As t varies in 1Il,~t assumes all values in [-1(1,1(1].
Example 6.2: : Inspired by (i) in Example 6.1 we call a pair (H, X) of observables
in '/JC a generalized harmonic oscillator of period 21T /c if [H, [H, Xll = c2 X, C > °
being a constant. If we write ad H(X) = [H, Xl then the harmonic property
implies
('t)2n (·t)2n+l
eitHXe-itH = eitadH(X) = L
00
_z_, 2n x + L
00
Z ,2n [H,Xl
n=O 2n. n=O 2n + 1.
= X cos ct + Y sin ct,
where Y = c-li[H, Xl. Our definition of harmonicity is equivalent to the property
that the mean acceleration of X in any state p is proportional to the mean value
of X in the same state p, the constant of proportionality being independent of p.
Suppose dim'/JC = n and {eo, el, ... ,en-I} is an orthonormal basis in '/JC.
Along with the truncated annihilation and creation operators L, L * defined by (4.4)
introduce the number operator N:
n-l
N = Ljlej)(ejl.
j=1
(The nomenclature has been chosen with the following intuitive picture in view.
Consider a system of particles which can be created and destroyed but the total
number of particles cannot exceed n - 1. We say that the unit vector ej indicates
the pure state in which the number of particles is j. Since Lej = ej_1 for j ~ 1
and Leo = 0, the effect of operation by L is interpreted as reduction of population
size by one unit. On the other hand L * increases the size by one unit as long
°
as it is less than n - 1 and annihilates it to when the size reaches n - 1. The
distribution of N in the pure state ej is degenerate at the point j and hence N
6 Dynamics in finite dimensional quantum probability spaces 21
deserves its name). We have the relations:

[N,L] = -L, [N,L*] = L*,
n-l-k
Lk = L !ej)(ej+k! if k:::; n - 1,
j=O
(6.7)
= 0 if k 2:: n.
n-l
N = LL*kLk, L*L = l-!eo)(eo!, LL* = l-!e n -l)(e n -l!

k=l
We shall now express any operator A = I:aij!ei)(ej! in terms of L,L* and
N. Indeed, writing
n-l
A = I:iaii!ei)(ei! + L {I:j aii+k!ei)(ei+k! +L ai+ki!ei+k)(ei!} (6.8)

k=l
and defining the functions f, gk, hk on {O, 1,2, ... } by
f(i) = aii,gk(i) = aii+k,hk(i) = ai+ki

where aij = 0 whenever i or j exceeds n - 1 we can express A as
n-l
A = f(N) + L{gk(N)L k +L*khk(N)}. (6.9)
k=l
Then (6.7) implies
+ L {e-itkgk(N)Lk + eitk L*k hk(N)}.

n-l
eitN Ae- itN = f(N) (6.10)
k=l
Suppose A is Hermitian. Then f is real, 11k = h k. Putting Ao = f(N), Ak

gk(N)L k + L*k hk(N) we see that (6.10) can be expressed as
where Ak is an observable, [N, [N, A k]] = k 2 Ak and for any function ¢ on

{0,1,2, ... }
tr¢(N)A = tr¢(N)Ao,
tr¢(N)AiAj = 0 for all i -I- j.
In other words every observable is a superposition of an observable com-
muting with N and a family of mutually uncorrelated harmonic observables with
respect to N in any state of the form ¢(N).
Exercise 6.3: Let Z = VN + lL + L*VN + 1. Suppose

k
Zk eo = LCk(r)er, 0 ~ k ~ n-l
o
where n = dim 'lie. Then eo(O) = 1 and
ck+l(r) = vr+lck(r + 1) + Vrck(r - 1) if r ~ 1,
= Ck(I) if r = O.
These recurrence relations imply that Ck(O) = 0 if k is odd and Ck(O) = 1.3.5 ...
2m - 1 if k = 2m ~ n - 1. Thus (N, Z) is a generalised harmonic oscillator of
period 211" where the first n - 1 moments of Z in the state eo (i.e., the state of
lowest energy according to N) agree with those of standard normal distribution. In
particular, as n -+ 00 the distribution of Z in the state eo converges weakly to the
normal distribution. (For more examples of this kind see [20)). The eigenvalues of
Z are the zeros of the n-th degree Hermitian polynomials (see Exercise 20.18 (b».
Notes
In the context of equations (6.3), (6.4) and Gleason's Theorem in Section 8, one
may formulate the notion of velocity of an observable as follows. Suppose that
the bounded observables concerning a quantum system constitute the real linear
subspace 0 0 of all Hermitian elements of a *-unital subalgebra 00 0 C 00 ('lie). A
linear map i : 0 0 -+ 0 0 is called a velocity map if i(X)Y + Xi(Y) = i(XY)
whenever X and Y commute with each other. Under what circumstances does
there exist a "Hamiltonian" H E 0 0 such that i(X) == i[H,X],X E Oo? When
00 0 = 00 ('lie), 'lie being a separable Hilbert space, it has been shown by B.V.R.
Bhat that the answer is in the affirmative (see: Characterisation of velocity maps
in quantum probability, lSI preprint 1990, to appear in the Pacific J. Math).
Regarding a more detailed investigation of generalised harmonic oscillators
and probability distributions associated with them, see Bhat and Parthasarathy
[20].
7 Observables with infinite number of values

and the Hahn-Hellinger Theorem
In a Hilbert space 'lie of dimension n < 00 an observable was defined to be a
selfadjoint operator X with the interpretation that the event "X takes the value
x" can occur if and only if x is an eigenvalue of X and in such a case it is the
projection ~({ x}) on the subspace {u : Xu = xu} c 'lie. In such a framework
every observable can take at most n values. If F C ~ is any subset then the event
"X takes a value in F" is the projection
~(F) =L ~({x}).
xEF
7 Observables with infinite number of values and the Hahn-Hellinger Theorem 23
Then for any sequence {Fj } of disjoint subsets of IR
~(UjFj) = L ~(Fj). (7.1 )

j
In other words the correspondence F --* ~(F) is an "event valued measure" on

the a-algebra of all subsets of R It may be recalled that when (Di' '2F i ), i = 1,2
are two measurable spaces and f : Dl --* D2 is a measurable map or, equivalently,
f is an D2 -valued random variable, then the event "f takes a value in F" is the
set f-l(F) E '2F 1 and the correspondence F --* f-l(F) satisfies for any sequence
{Fj} of disjoint sets from '2F 2 ,
If-1(UjFj) = L1f-l(Fj)
j
which is parallel to (7.1). In other words F --* If-l(F) can be viewed as an

event valued measure. In order to admit observables assuming an infinite number
of values, we let 'Je be an arbitrary complex separable Hilbert space and define
for any fixed measurable space (D, '2F) an D-valued observable ~ as a mapping
~ : '2F --* 9J>('Je) satisfying ~(D) = I and
~(UjFj) = L~(Fj) if Fi n Fj = ¢ for i i- j (7.2)

j
where j = 1, 2, ... and the infinite series of projection operators on the right hand
side is interpreted as a strongly convergent sum. Thus an D-valued observable ~
is nothing but a spectral or projection valued measure on the measurable space
(D, '2F). In particular ~(E)~(F) = ~(E n F). Whenever D is a topological space
we shall assume that '2F is its Borel a-algebra generated by the open subsets of
the topology and write '2F = '2Fn.
Suppose ~ : '2F --* 9J>('Je) is an D-valued observable and ~(D\E) = 0 for some
E E '2F. Then we say that ~ takes values in E or ~ has support in E. If D = IRk and
~ has support in a bounded subset of IRk we say that the observable ~ is bounded.
If (Dl' '2FJ) is another measurable space and g : D --* Dl is a measurable map
then ~g-1 : '2F 1 --* 9J>('Je) is an D 1-valued observable called the function g of the
observable ~. If 'Je, 'Je' are Hilbert spaces, U : 'Je --* 'Je' is a unitary isomorphism
and ~ is an D-valued observable with respect to 'Je; then there is an D-valued
observable e with respect to 'Je' defined by e(E) = U(~(E)U-l,E E '2F. ~ and
e are said to be unitarily equivalent and we write ~ '" e. '"
as an equivalence
relation. If ~j is an D-valued observable in 'Jej for each j = 1,2, ... then there
exists an observable ~ = ffij~j, called the direct sum of the observables ~j,j =
1, 2, ... and defined by
~(E)u = ffij~j(E)uj if U = ffijuj E ffij'Je j , E E '2F.

Example 7.1: Let (O,,;!F, JL) be any a-finite measure space where ;!F is countably
generated. In the complex separable Hilbert space L 2 (JL), define e:
;!F --t rJJ>(7IC)
by
IE denoting the indicator of E. Then ell is an observable called the canonical

o'-valued observable on the measure space (O,,;!F, JL).
Let v be another a-finite measure on (O,,;!F) equivalent to JL in the sense that
v(E) = 0 if and only if JL(E) = O. Define the unitary operator U : L 2(JL) --t L2(V)
by
d ) 1/2
(Uu)(w) = ( d~ (w)u(w),
Then Uell (E)U- 1 = eV(E) for all E E ;!F. In other words ell and ev are unitarily
equivalent. Conversely, if ell and ev are two unitarily equivalent canonical 0,-
valued observables then the measures JL and v are equivalent in the sense of
measure theory.
The principal aim of this section is to demonstrate that every o'-valued ob-
servable is unitarily equivalent to a direct sum of canonical observables. To this
end we establish a basic proposition which will be frequently used in the course
of our lectures. A subset S c TIC is called total if the smallest closed subspace
containing S is TIC or, equivalently, S.L = {o}.
Proposition 7.2: Let Si be a total subset of the Hilbert space 7IC i , i = l, 2. Suppose
Uo : SI --t S2 is a scalar product preserving map, so that for all u, v E SI
(UOU, Uov) = (u, v). (7.3)

Then there exists a unique linear isometry U : ~1 --t 7IC2 which extends Uo,
i.e., Uu = Uou for all u E SI. If, in addition, Uo is onto then U is a unitary
isomorphism from ~1 onto 7IC2 •
Proof: For Cti,{3j E C,Ui,Vj E SI, 1 ~ i ~ m.l ~ j ~ n, (7.3) implies
(~=CtiUoUi' L{3jUOVj) =L Ui{3j(Ui,Vj)

i j 0
(7.4)
= (LCtiUi, L{3jVj).
i i
In particular,
(7.5)
Define
U1 L CtiUi = L CtiUOUi'
i i
If Li QiUi = Lj {3jVj then (7.5) implies II Li QiUOUi - Lj {3jUovl12 = 0 so

that UI is well-defined on the linear manifold MI generated by SI. (7.4) implies
that UI is a linear isometry on MI and, in particular, bounded. By continuity, UI
extends uniquely to a linear isometry U on the closure 1\11 = 'fIC 1• If V is another
such extension then U - V is a bounded operator vanishing on the total set SI and
hence U - V = O. The range of the isometry U is a closed subspace containing
the image of SI. If Uo is onto then R(U) = 'fIC 2 and U is unitary. •
A vector U E 'fIC is called cyclic for the O-valued observable ~ : :¥ ~ rg>('fIC)
if the set {~(E)u,E E :¥} is total in 7If. For any two a-finite measures >",p, on
(0, :¥), write>.. « p, if>.. is absolutely continuous with respect to p" i.e., >"(E) = 0
whenever p,(E) = 0; >.. == p, if >.. « p, « >.. ; >"1.p, if >.. and p, are singular with
respect to each other, i.e., there exists E E :¥ such that >"(O\E) = p,(E) = O. If
E E :¥ write >"IE for the measure
>"IE(F) = >"(E n F) for all F E :¥
and call it the restriction of >.. to E.
Proposition 7.3: Let ~ : :¥ ~ rg>('fIC) be an O-valued observable with cyclic
vector U and let p, be the totally finite measure defined by p,(E) = (u, ~(E)u) for
all E E :¥. Then ~ '" ~/L.
Proof: Let 7If1 = 'fIC, 'fIC2 = L2(p,), SI = {~(E)u, E E :¥}, S2 = {IE, E E :¥}.
Then Si is total in 'fICi , i = 1,2. Define Uo : SI ~ S2 by Uo~(E)u = IE. Then
(Uo~(E)u, Uo~(F)u) = (IE,IF )
= p,(E n F) = (u, ~(E n F)u)
= (u, ~(E)~(F)u) = (~(E)u, ~(F)u).

By Proposition 7.2, Uo extends to a unique unitary isomorphism U : 'fIC 1 ~ 'fIC2
and
U~(E)U-t IF = U~(E)~(F)u
= U~(E n F)u = I EnF = IEIF = e(E)IF .

Since S2 is total in 7If2, ~ '" ~/L.
•
Proposition 7.4: Let ~ : :¥ ~ rg>(7If) be an O-valued observable. Then there
exists a finite or countable family {p,j} of probability measures on (O,:¥) such
that ~ '" (f}je j .
Proof: If Ut E 'fIC is any unit vector let 'fIC t be the closed subspace spanned by
{~(E)ut, E E :¥}. Then 'fIC t is left invariant by the projections {~(E)} and ~1;1t.L is
I
a spectral measure. We can now repeat this procedure starting from a unit vector
U2 E 'fICt. Using the standard maximality argument based on Zorn's Lemma,
we may express 7If = (f}'fICj where each 7Ifj is left invariant by the projections
~(E), E E :¥ and ~1;1tj has a unit cyclic vector Uj. If we put p,j(E) = (Uj, ~(E)uj)
and observe that ~ = (f}j~j,~jl;1tj'~j '" e j by Proposition 7.3 then it follows that
~ = (f}j~j '" (f}j~/Lj. •
Proposition 7.5: Let ft], ft2, ... be a-finite measures on (0,2/'). Then there exist
a-finite measures VI » V2 » ... such that ffije'j "" ffij~Vj.
Proof: By repeated use of Lebesgue's Decomposition Theorem construct the

following table:
ftl = ftll
ft2 = ft21 + ft22, ft21 1.. ft 11 , ft22 « ft 11
ft3 = ft31 + ft~l,ft311..ftll + ft2l,ft~l «ftll + ft21
ft~1 = ft32 + ft33, ft321..ft22, ft33 « ft22
ft3 = ft31 + ft32 + ft33
n-l n-l
ftn = ftnl + ft~l,ftnl1.. Lftil,ft~l «Lftil
i=1 i=l
n-l n-l
ftnj
I = ftnj+1 + ftnj+I' ftnj+l .1..
I ""
~ ftij+], ftnj+l«
I ~
" " ftij+l
i=j+l i=j+1
j = 1,2, ... , n - 2
ft~n-l = ftnn
ftn = ftnl + ftn2 + ... + ftnn
This yields a triangular array of measures:
total
ftll ftl
ft2l ft22 ft2
ft3l ft32 ft33 ft3
ftnl ftn2 ftn3 ftnn ftn
Total VI V2 V3
In this array, for every n all the measures in the n-th column are mutually singular
and hence they add up to a a-finite measure vn . Since ft22 « ftll,ftnj+2 «
ft~j+ 1 « L,~:/+ 1 ftij+ I for j = 0, 1,2, ... , n - 2, n = 3,4, ... it follows that
« for n = 2,3, .... We observe that for any finite or infinite sequence
e where A = L.j Aj. Thus
Vn Vn-l
{Aj} of mutually singular a-finite measures, ffije j ""
Hence
•
Theorem 7.6: (Hahn-Hellinger Theorem [53]) Let (O,~) be a measurable space

and let ~ be a O-valued observable in the Hilbert space cae. Then there exists a
sequence AIX" AI, A2, ... of mutually singular 0' -finite measures such that
~ '" {e= ffie= ffi···} ffie ffi {e
1 2 ffie 2 } ffi··· ffi {e
~
i ffi··· ffi e
i } ffi··· (7.6)
j-fold
If A:x" A~, A~, . .. is another sequence of mutually singular O'-finite measures sat-
isfying the same conditions then Aj is equivalent to Aj for each j = 00, 1,2, ... .
Proof: By Proposition 7.4 and 7.5 there exist O'-finite measures VI » V2 » .. .

such that ~ '" ~Vl ffi ~V2 ffi ... Choose and fix a version of the Radon-Nykodym
derivatives ¢n = ddvVnn- l ,n = 2,3, .... Consider the sets
Al =0
A2 = {w : ¢2(W) =I- O}
Then Al :2 A2 :2 '" ,vn(O\An ) = 0 for each n. In view of the discussion in

Example 7.1 we may assume without loss of generality that Vn is the restriction
of VI to the set An, i.e.,
vn(E) = vI(E n An),E E ~,n = 2,3, ...
Then we continue to have ~ '" ffij~Vi. Let Bn = An \An+l' n > 1. We may
represent the Aj's and Bj's diagrammatically as follows:
Al
Bl {
{ ......... .
B'.~~{
B3 {~~{
B~{
~----~~----~------~----------------------~
Each Aj is represented as a column. Since A j ' s decrease they are represented by

columns of decreasing heights. B j is represented by the wavy bracketed portion
of Aj. We write
Then Bj's are disjoint and

Al = n= Boo UBI U B2 U ..... .
Define the measures Aj by
Aj(E) = vI(E n Bj ), j = 00,1,2, ...
Then
Since Bj's are disjoint

~Vj rv e EB ej EB ei+! EB ...
oo
Hence EBj~Vj is equivalent to the right hand side expression in (7.6). This completes
the proof of the first part.
Suppose A:x" A~ , A~, . .. is another sequence of mutually singular measures
satisfying (7.6) with Aj replaced by Ai for each j. By the discussion in Example
7.1 we may replace Aj and Ai by equivalent totally finite measures for each j.
Hence we assume without loss of generality that Aj' Ai are totally finite. Let
'Jeo = EB l::;j::;oo EBj copies L 2 (Aj ),

'Je~ = EBI::;j::;ooEBj copies L 2 (Ai),
~o = EBl::;j::;ooEBj copies e j ,
,
~o = EBl::;j::;ooEBj copies ~
)..'.
J
where 00 indicates the countably infinite cardinal. Let Prs, P;s denote the projec-
tions on the s-th copy of L2(Ar), L2(A~) respectively in 'Jeo, 'Je~. Define ers E 'Jeo
by
Pr's,e rs = 0 (r,s) ¥- (r',s')

if
where s = 1,2, ... ,r if r < 00, s = 1,2, ... if r = 00 and 1 in the first equation
denotes the constant function 1 in L2(Ar)' Let U : 'Jeo -+ 'Je~ be a unitary operator
such that
U~O(E)U-l = ~b(E) for all E E ?:F. (7.7)
Let
f rs = Ue rs , p!.frs
~J
= f'!s . ~J
Choose supports Bj, Cj for Aj' Ai respectively for each j so that the Bj's and
Cj's constitute two partitions of n. Then (7.7) implies
Ai(E n Bi n C j )t5 ss ' = (eiS,IEnBincjeiS')
(7.8)
Putting 8 = 8' we get

Ai IBinGj «: Aj IBinGj
Interchanging the roles of 'Jeo, 'Jeb here and using both the relations we get
Ai IBinGj == Aj IB;nGj' (7.9)
(7.8) and (7.9) imply
L IfjHw)1 2 > 0 a.e. W(AiIB;nCj)' (7.10)

k
L fj~(w)fj~' (w) = 0 a.e. W(AiIBinGj) for 8 -I- 8'. (7.11)

k
Now suppose that AilB;nGj -I- O. Then (7.10) and (7.11) imply the existence of a
Wo E n such that
L IfjH wo)1 2 > 0,
k
L fjHwo)fj~' (wo) = 0, 1::; 8, 8' ::; i if i < 00,

k
1 ::; 8, 8' < 00 if i = 00, S -I- 8'.

In other words we can find i mutually orthogonal j-dimensional vectors. Hence
i ::; j. Reversing the roles of 'Jeo, 'Jeb we get j ::; i. Thus Ai has support in Bi n Ci
and Ai == A~ for 1 ::; i ::; 00. •
When a spectral measure ~ satisfies the equivalence relation (7.6) and Aj -I- 0
for some j then we say that the measure class Aj or spectral type Aj occurs in the
spectrum of ~ with multiplicity j.
Suppose n is a separable metric space and ~n is its Borel a-algebra. Let /-t
be a fixed non-atomic a-finite measure on ;9fn. Then each measure Aj occurring
in (7.6) can be decomposed as
Aj = /-tj + Vj + Pj, 1::; j ::; 00
where /-tj «: /-t, Vj is non-atomic and Vj.l/-t and Pj has its support in a countable
set. Such a decomposition is unique. Define
> -- ""1<"<
'c(1) "" copies'>
W _J_OOWj . eJl.j ,
> -_ fT\l<"<
'c(2) fT\ "eVj
W _J_OOWj copies'> ,
> -- ffil<"<
'c(3) fT\ . ePj
W _J_OOWj copies'> ,
By the Hahn-Hellinger Theorem we can decompose 'Je into the direct sum
~ as 6 EEl 6 EEl 6 where ~i : ~n ---.
'Je l EEl 'Je2 EEl 'Je3 and express the spectral measure
C!P('Jei ) is a spectral measure unitarily equivalent to ~(i), i = 1,2,3. We say that
~J, 6, 6 are respectively the parts of ~ absolutely continuous with respect to /-t,
singular continuous with respect to /-t and pure point part. The pure point part is
independent of /-t. Such a decomposition is unique.
Example 7.7: Let n= ~n,;qf = ;qf[J;ln, For any ~n-valued observable ~ in 'JC and
Q E ~nlet
~S!.(E) = ~(E - Q), E E ;qf[J;ln
where E - Q {.;r. - QI.;r. E E}. Then ~S!. is an ~n-valued observable. The

relation ~S!. rv ~ holds for all Q E ~n if and only if in (7.6) all but one of the
Aj'S vanish and the non-vanishing Aj is equivalent to the Lebesgue measure.
This is a consequence of the fact that in ~n, the Lebesgue measure class is the
only measure class invariant under all translations. If Ajo = A is a non-vanishing
measure equivalent to the Lebesgue measure we say that ~ has Lebesgue spectrum
of multiplicity jo . jo may take any of the values 00,1,2, .... When jo = 00 we
say that ~ has a countable Lebesgue spectrum.
Example 7.8: Let G be a locally compact second countable group and let He G
be any closed subgroup. Choose and fix a G-quasi invariant measure j.t on the Borel
a-algebra of the homogeneous space G/H = {gH,g E G} with the quotient
topology. If ~ is any G / H -valued observable and g E G then
~g(E) = ~(g-IE),E a Borel set in G/H
defines another observable ~g. If ~g rv ~ for every g E G then
~ rv e E!.l e E!.l • • •
"---,.-0"
for some j = 00,1,2, ...
j-fold
Example 7.9: Let T be a bounded nonnal operator in 'JC so that T*T = TT*.
By the spectral theorem for such operators there exists a unique spectral measure
~ on the Borel a-algebra of the complex plane with support in {zl Izl :::; IITII}
Jz~(dz).
such that
T= (7.12)
Thus ~ is a complex valued observable taking values in {zl Izl :::; IITII}. Con-
versely, every bounded complex valued observable ~ detennines uniquely a bound-
ed nonnal operator T through the definition (7.12).
T is unitary if and only if the corresponding ~ in (7.12) has its support in the
torus {z : Iz I = I}. T is selfadjoint if and only if ~ has its support in the inteval
[-IITII, IITlll· T is a positive operator if and only if ~ has support in [0, IITlll. Thus
there is a one-to-one correspondence between real valued bounded observables and
bounded selfadjoint operators on 'JC via the map ~ ~ T~ = J
x~( dx). In view
of this correspondence we may as well call the selfadjoint operator T~ itself a
bounded real valued observable. Thus, whenever a bounded selfadjoint operator
T is viewed as a real valued observable it is to be understood that the event "the
observable takes a value in the Borel set E C ~" is the spectral projection ~(E)
J
occurring in the spectral representation (or resolution) T = x~( dx). In particular,
to every real valued observable taking a finite number of values Xl, X2, ... ,Xk
8 Probability distributions on (l»(~) and Gleason's Theorem 31
there corresponds a direct sum decomposition 'ffC = EBJ=I 'ffC j where 'Je j is the
range of the projection ~({ Xj}) and the corresponding selfadjoint operator T~ is
Lj Xj~( {Xj}).
Two bounded normal operators TI and T2 are said to be unitarily equivalent
if T2 = UTI U- I for some unitary operator U. TI and T2 are unitarily equivalent if
and only if their corresponding complex valued observables 6 and 6 are unitarily
equivalent. Two such observables take the same values or have the same support.
Notes
The proof of the Hahn-Hellinger Theorem is patterned after that of Halmos [53].
For an elegant proof based on the structure of general abelian C* -algebras, see
Arveson [10],
8 Probability distributions on czp('fJe) and

Gleason's Theorem
Let 'Je be a fixed Hilbert space and let (jl('Je) be the set of all events (or projections)
in 'ffC. Suppose that to every event P E (jl('Je) there is a probability j.t(P) for its
occurrence so that 0::; j.t(P)::; 1,j.t(I) = 1. Consider 0 = N = {1,2, ... } with
the a-algebra :?F of all its subsets. Suppose we demand that for every N-valued
observable ~ in 'Je the probability that ~ takes a value in E, namely j.t(~(E)), as
a function of E is a probability distribution on (N, :?F). This is equivalent to the
condition that for every sequence {Pj} in (jl('ffC) for which PiPj = 0 for all i i= j
(i.e., for every seqeunce of projections whose ranges are mutually orthogonal)
j.t(LPj) = Lj.t(Pj )
j j
and j.t( 1) = 1. Such a map j.t : (jl('ffC) ~ [0, 1] is called a probability distribution
on (jl('ffC). The main aim of this section is to establish Gleason's Theorem that
to every probability distribution j.t on (jl('ffC) with dim'ffC 2 3 there corresponds a
unique operator TJ.! on 'Je for which j.t(lu) (ul) = (u, TJ.!u) for every unit vector u
in 'Je.
Suppose j.t is a probability distribution on (jl('ffC) and fJ.!(u) = j.t(lu)(ul)
where II u I = 1. Then f J.! (cu) = f J.! ( u) for any scalar c of modulus unity and
for any orthonormal basis {Uj} in 'ffC we have Lj IUj) (Uj I = 1 and therefore
Lj f J.! (Uj) = 1. We say that a complex valued function f defined on the unit
sphere {ul lIull = I} c 'Je is a frame function of weight W if f( u) = f( cu)
for any unit vector u and scalar c of modulus unity and Lj f( Uj) = W for
every orthonormal basis {Uj} in 'ffC, where the infinite series converges absolutely
whenever dim'Je = 00. Such a frame function is said to be regular if there exists
a bounded operator T on 'ffC such that f (u) = (u, Tu) for all unit vectors u.
Example 8.1: Let 9 be an arbitrary function defined on the interval [0, ~) satis-
fying 0 ~ g( 0) ~ 1 for all O. For any unit vector (cos 0, sin 0) = U in IR define
f(u) by
- {g(O) if ~ 0 < 7r /2,
f(u) - 1 - g(O - 7r/2) if 7r/2 ~ 0 < 7r,
f(u) = f( -u).
Then f is a non-negative frame function of weight unity. In general f need not
be regular.
Proposition 8.2: Let f be a frame function in the real Hilbert space 1R3. If f is
infinitely differentiable on the unit sphere 51 C 1R3 then f is regular.
Proof: Let <Po denote the set of all infinitely differentiable frame functions in 1R3.
Represent any point p on 8 2 by its spherical polar coordinates (0, ¢ ), 0 ~ 0 ~
7r,0 ~ ¢ < 27r. See Figure 8.1.
z
Fig. 8.1
Then any frame function f is a function of (0, ¢) which is periodic in 0 of

period 7r and in ¢ of period 27r. Let 0 3 denote the group of all rotations in 1R3
about the origin O. For any R E 0 3 let R(O, ¢) be the point obtained by applying
R to the point (O,¢). If f E <Po then foR E <Po where foR is the composition of
f with R. Let R~, R~ be the rotations through an angle a about the x and z axis
respectively. Then
R~( 0, ¢) = (0, ¢ + a) mod (7r, 27r),

R~(O, ¢) = (0' , ¢/)
8 Probability distributions on '2J>(iIf) and Gleason's Theorem 33
where (0, </», (0', </>') are related by the equation
1 0 0 ) (Sin 0 cos </> ) sin 0' cos </>' )

( o co~ a sin a sin 0 sin </> = ( sin 0' sin </>' .
o - sm a cos a cos 0 cos 0'
Since <Po is a linear space invariant under 0 3 action we have for any f E <Po,
f(R~(O,</») = f(O,</> + a) E <Po, (8.1)
d (
da f Ro 0, </>
x( )) I0=0 = sm
. </> {){)Of + cot 0 cos </> {)</>
{) f E <Po· (8.2)
Thus
Jor27r f(O, </> + a)e-inada = eincfi Jro27r f(O, a)-inoda
= eincfiFn(O) E <Po
and
+00
f(O,</» = 2~ I>incfiFn(O)
-00
is the Fourier expansion of f (0, </» in the variable </>. We shall now prove that
Fn(O) = 0 for Inl > 2 and Fn has a special fonn when Inl ~ 2. To this end let
g(O,</» = eincfiF(O) E <Po. (8.3)
By (8.2)
eincfi {sin </> F' (0) + in cot 0 cos </> F( O)} E <Po. (8.4)
Changing </> to </> + 1r /2 in (8.4) and using (8.1) we get
eincfi {cos </>F' (0) - in cot 0 sin </> F( O)} E <Po. (8.5)
Multiplying (8.4) by i and adding to and subtracting from (8.5) we obtain
ei(n+l)cfi{F'(O) - nF(O)cotO} E <Po,
(8.6)
ei(n-l)cfi{F'(O) + nF(O) cotO} E <Po.
These two relations imply in particular that in the plane </> = 0, F( 0) and nF( 0)
cot 0 are frame functions for ~2. Thus there exist constants a, b such that
1r
F( 0) + F( 0 + "2 ) = a,
F(O)cotO + F(O + ~ )cot(O + ~) = b if n # O.

Solving for F( 0) we have
F(O) = asin2 0 + bsinOcosO if n # o. (8.7)

Using (8.6)
ei(n+l)tP{(2 - n)asinOcosO + (n - 2)bsin2 0 + (1- n)b} E <Po
whenever (8.3) and (8.7) hold. If Inl ~ 2 then n ± 1 =j:. 0 and (8.7) implies b = O.
If Inl > 2 then
(2 - n)asinOcosOei(n+l)tP E <Po
which implies a = O. Thus
a = b = 0 if Inl > 2,
b=0 if Inl = 2. (8.8)
Let Inl = 1. Using the property of frame functions for the orthonormal basis
(O,¢),(O+f,¢),(f,¢+f) we have
eintP F( 0) + eintP F( 0 + 1f /2) + ein (</>+1r/2) F( 1f /2) = C
where c is a constant. Substituting for F from (8.7) we conclude a = c = O. Thus

a = 0 if Inl = 1. (8.9)
Let now n = O. By the first relation in (8.6) eitP F' (0) E <Po whenever F( 0) E <Po.
In such a case (8.7) and (8.9) imply that F' (0) = d sin 0 cos 0, for some constant
d. Thus F( 0) = Cl + C2 cos 2 0 for some constants Cl, C2. Combining this with
(8.7)-(8.9) we have for any f E <Po,
1 00
f(O,¢) = 21f LeintPFn(O)

-00
= ao + al cos 2 0 + (a2e itP + a3e-itP) sin 0 cos 0
+( a4 e2itP + ase- 2itP ) sin2 0

where aj, 0 :::; j :::; 5 are constants. Writing x = sin 0 cos ¢, y = sin 0 sin ¢,
z = cos 0 we see that f (0, ¢) is a homogeneous quadratic form in x, y, z restricted
to x 2 + y2 + Z2 = 1. In other words f is regular. •
Proposition 8.3: Every continuous frame function in the real Hilbert space ~3 is
regular.
Proof: Let f : S2 ----+ IC be a continuous frame function, S2 being the unit sphere
in ~3. Let S03 be the group of proper rotations in ~3, i.e., all orthogonal linear
transformations of unit determinant. Then S03 acts transitively on S2. Define the
lifted function j on S03 by
j(g) = f(g(O,O)), g E S03
where (0, ¢) = (0,0) is the north pole in S2. Then j is a continuous function on
S03' For any infinitely differentiable function ¢ on S03 consider the convolution
(¢ * j)(g) = J j(h-lg)¢(h)dh
8 Probability distributions on (iJ>('JC) and Gleason's Theorem 35
where dh denotes integration with respect to the normalised Haar measure of the
compact group S03. Then (</J* j)(gk) = (¢ * j)(g) for all rotations k about the z
axis. Furthermore ¢ * j is infinitely differentiable. Hence there exists an infinitely
differentiable function 'Ij; on S2 such that
(¢ * j)(g) = 'Ij;(g(O, 0)) for all g E S03.

Since frame functions constitute a linear space which is closei under pointwise
convergence, it follows that 'Ij; is an infinitely differentiable frame function and
hence, by Proposition 8.3, regular. Choose a sequence {¢n} of non-negative in-
finitely differentiable functions on S03 satisfying (i) J ¢n (g )dg = 1; (ii) for every
neighbourhood N of the identity
lim
n--+oo
r ¢n(g)dg
iN'
= 0,
N' denoting the complement of N. Construct the frame function 'lj;n satisfying
'lj;n(g(O,O)) = (¢n * j)(g) and observe that 'lj;n converges to f pointwise on S2.
Since each 'lj;n is regular f is also regular. •
We shall now establish a few lemmas leading to the result that every non-
negative frame function on S2 is continuous. For any map f : S2 --+ C and any
set V C S2 write
osc(f, V) = L
If(x) - f(y)l·
x,yEV
Lemma 8.4: Let f be a frame function on S2. If for some p E S2 there exists a
neighbourhood Vp of p with osc(S, Vp) < 'TJ then for every q E S2 there exists a
neighbourhood Vq of q such that osc(f, Vq) < 4'TJ.
Proof: Without loss of generality we may assume p to be the north pole and Vp
to be a spherical cap with centre p and arc length e. See Figure 8.2.
Let qo be any point on the equator. Draw the great circle pqo and consider the
point ro at distance e/2 from qo along this great circle. Construct a neighbourhood
VqO of qo so that for any q in VqO the points q' and r' at right angles to qo
and ro respectively along the great circle roq fall in Vp. Then the map q --+
(r', q') is continuous in Vqo' If q[, q2 E Vqo and their images under this map are
(r;, q;), (r~, q~) respectively then
i = 1,2.
Taking differences,
which implies
Fig. 8.2
If S is any point on 8 2 then the equatorial circles for p and

S meet at two points,
one of which may be called qo. Then S lies on the equatorial circle of qo. Hence
we can find a neighbourhood Vs of s such that osc(f, Vs) < 41]. •
To state the next lemma we introduce some notation. Let p denote the north
pole and N the northern open hemisphere on 8 2 • For any p' E N - {p} the great
circles on the equatorial planes of p and p' intersect at two points q, q'. The great
circle through qp' q' is denoted by Cpl. If 0 denotes the centre of 8 2 the angles
p' oq and p' oq' are 7r /2 each.
Lemma 8.5: Let p be the north pole of 8 2 and qo E N - {p}. Then the set
WqO = {qlq E N - {p},qo E Cr for some r E Cq n (N - {p})} (8.10)
has a non-empty interior.
Proof: Without loss of generality assume that qo has cartesian coordinates (sine,
0, cos e) where the origin 0 is the centre of 8 2 and p = (0,0, 1). Let So =
(~o, 1]0, (0) E N - {p} be such that the great circle through qoso is also the circle
C so through So. Suppose C so meets the equator of p at (coso;, sino;, 0). Let
8 Probability distributions on 'iJ'('~) and Gleason's Theorem 37
(C, m, n) be the direction cosines of the normal to the plane of Cso ' Then
Csin () + m.O + n cos () = 0,
+ m",o + nCo = 0,
C~o
Ccosa + msina + n.O = 0,
and by the definition of C so '
~o cos a + "'0 sin a = O.

Eliminating C, m, n, a we have
(~5 + "'5) cos () - ~o(o sin () = O.

Putting 'IjJ(~,,,,,() = (e+ ",2)cos() - ~(sin() we conclude that the great circle
through (~,"',() and (sin(), 0, cos()) is the same as C s where S = (~,"',() if
'IjJ(~,,,,,() = O. Observe that 'IjJ((,,,,, 0) = (e
+ ",2) cos () > O. Let
Va = {(~,"',()Ie + ",2 + (2 = 1, cos() < « 1,
C I (1 - (2)cot() < ~ < (1 _ (2)1/2}.
If S E Va, S = (~, "', () then 'IjJ(s) < O. If C s meets the equator of p at SI

(6,,,,],0) then 'IjJ(sJ) > O. Hence there exists an intermediate point So between s
and Sl where 'IjJ(so) = O. Thus qo E C so ' So E C s n N - {p}. In other words the
set WqO defined by (8.10) includes the open set Vo. •
Proposition 8.6: Every frame function on 8 2 is regular.
Proof: Let f :::: 0 be a frame function of weight W. Since constants are regular
frame functions and f - inf f is non-negative we may assume without loss of
generality that
inf{f(x),x E 8 2 } = O.
Let", > 0 be arbitrary. Choose p E 8 2 such that f(p) < ",. Let a be the
rotation through the angle 1r /2 about the axis op,o being the centre of 8 2 . Put
g(x) = f(x) + f(ax). Then 9 is a frame function of weight 2W. For any point q
on the equator of p,g(q) = W - f(p). Let r E N - {p}, s, t E C r - N - {p}, sl.t.
If q is a point on the intersection of C r and the equator of p then
2W :::: g(s) + g(t) = g(r) + g(q) = g(r) + W - f(p) :::: g(r) +W - ",.
Thus
g(r) :::; W + '" for all r E N - {p}.
In particular
g(r) + W - '" :::; g(s) + g(t) :::; g(s) + W + ",.
Thus
g(r) - g(s) :::; 2", for r E N - {p}, s E Cr. (8.11)
Let
f3 = inf{g(q),q E N - {p}}.
Choose qo EN - {p} such that g(qo) < f3 + 'TJ. Choose any q E WqO defined by
(8.10). Then there exists r satisfying r E Cq , qo E Cr. By (8.11)
g(q) - g(r) ~ 2'TJ, g(r) - g(qo) ~ 2'TJ

Thus
f3 ~ g(q) ~ g(r) + 2'TJ ~ g(qo) + 4'TJ ~ f3 + 5'TJ.
In other words
By Lemma 8.4 it follows that p has a neighbourhood Vp satisfying osc(g, Vp) ~

20'TJ. Hence for any q E Vp
o ~ g(q) ~ g(q) - g(p) + g(p) ~ 20'TJ + g(p) = 20'TJ + 2f(p) ~ 22'TJ.

Thus
o~ f(q) ~ g(q) ~ 22'TJ for all q E Vp.
Once again by Lemma 8.4 every point q has a neighbourhood Vq such that
osc(f, Vq ) ~ 88'TJ. Since 'TJ is arbitrary f is continuous. By Proposition 8.3, f
is regular. •
Lemma 8.7: Let f be a non-negative regular frame function of weight W in a

Hilbert space <tie. Then for any two unit vectors u, v
If(u) - f(v)1 ~ 2Wllu - vii.
Proof: By hypothesis f (u) = (u, Tu) for any unit vector u where T is a bounded
selfadjoint operator. Then
If(u) - f(v)1 = l(u,Tu - Tv) + (u - v,Tv)1 ~ 211Tliliu - vii.

But IITII = sup l(u,Tu)1 ~ W.
Ilull=l
•
Let 71f be a Hilbert space. A closed set S c <tie is called a completely real
subspace iffor all u, v E S, (u, v) is real and au + bv E S for all real scalars a, b.
Proposition 8.8: Let <tie be a complex Hilbert space of dimension 2 and let f
be a non-negative frame function for <tie. If f is regular in every completely real
subspace then it is regular in 71f.
8 Probability distributions on (5J>('~) and Gleason's Theorem 39
Proof: Let M = sup{f( u), lIull = I}. For any u E 'JC define
F(u) = {011U112 f(llull-1u) if u i- 0,

otherwise.
Choose unit vectors Un such that
lim f(u n ) = M, lim Un = UO.
n--+oo n--+oo
Define
Then Vn is a unit vector, lim Vn = Uo and the real linear span Sn of vn and Uo
n-+oo
is completely real. If the weight of f is W it follows from Lemma 8.7 that
Thus
f( uo) = lim f( vn ) = lim f( un) = M.
n--+oo n--+oo
Let Vo be any unit vector orthogonal to uo. Then f is regular in the real linear
span So of Uo and Vo. Since the quadratic form corresponding to f in So attains
its supremum on the unit sphere at Uo it follows that
F(auo + bvo) = a2 M + b2 (W - M) for a, bE !fit
If a, bEe, ab i- 0 then
F(auo + bvo) = lal 2 F(uo + ba-1vo)
= lal 2 F(uo + ba-lv~)
where v~ = ba- 1lab- 1Ivo is a unit vector orthogonal to Uo. Hence the last term in
the above equation is equal to laI 2 (M+lba- 1 2 (W-M)) = laI 2 M+lbI 2 (W-M).
1
If either a or b vanishes the same relation holds trivially. Since dim'JC = 2

it follows that Fis a quadratic form with the matrix (~ W~ M) in the
orthonormal basis uo, Vo. •
Theorem 8.9: (Gleason's Theorem [48]) Let 'JC be a real or complex separable
Hilbert space of dimension ~ 3. Then every non-negative frame function f on '1JC
is regular.
Proof: Let S c '1JC be any two dimensional subspace and So c S a two di-
mensional real subspace. Choose a unit vector uo-LSo. In the real linear subspace
spanned by So and Uo the restriction of f is a non-negative frame function and
hence by Proposition 8.6 regular. Thus f is regular in So and by Proposition 8.8
regular in S. Define
F( u) = { IIul1 2 f(llull-1u) if u i- ?,
o otherwIse.
Then for any two dimensional subspace 8 c ?Je there exists a positive operator
As satisfying
F(u) = (u, Asu), u E 8.
Let u, v E ?Je. Either v = au for some scalar a, or u and v span a unique two
dimensional subspace 8. In the latter case define
A(u,v) = (u,Asv).
In the former case choose some two dimensional subspace 8 containing u (and
therefore v) and define
A(u,v) = (u,Asv).
If 8' is another two dimensional subspace containing u then
(u,Asv) = a(u,Asu) = aF(u) = a(u,Aslu) = (u,As1v).

Thus A( u, v) is a well-defined function of u and v. Furthermore
(i) A(u, v) = A(v, u)
(ii) A( u, av) = aA( u, v) for all scalars a,
(iii) 4ReA(u,v)=F(u+v)-F(u-v),
(iv) F(u + v) + F(u - v) = 2{F(u) + F(v)}
where Re denotes real part. For any u, v, w in ?Je we have
ReA(u,w) +ReA(v,w)
1
= 4{F(u + w) - F(u - w) + F(v + w) - F(v - w)}
I
= 8{F(u + v + 2w) - F(u + v - 2w)}
= ReA(u + v,w).
Changing w to -iw in this relation we obtain
ImA(u,w) + ImA(v,w) = ImA(u + v,w),

1m denoting imaginary part. Thus A( u, v) is a bounded Hermitian sesquilinear
form satisfying 0 :::; F( u) = A( u, u) :::; W whenever lIull = 1. Hence there exists
a unique positive operator T such that f(u) = (u,Tu) for each unit vector u . •
Corollary 8.10: Let f..t be a probability distribution on rzl'(?Je) where dim?Je ~ 3.

Then there exists an orthonormal set {Uj} in ?Je and positive scalars {Pj}, j =
1,2, ... , such that
f..t(P) = LPj(Uj,PUj) for every P E rzl'(?Je) (8.12)

j
8 Probability distributions on (lJ>('Je) and Gleason's Theorem 41
Proof: By Gleason's Theorem there exists a bounded positive operator T such

that J,t(lu)(ul) = (u,Tu) for any unit vector u. The spectral representation of T
has the form
T = r
i[O,IITII]
x~(dx).
Choose any 0 < c < IITII and denote by RE the range of the spectral projection
~([c, IITII]). Let {ej} be any orthonormal basis for 'Je which includes an orthonor-
mal basis for R E • Then
1 = LJ,t(lej)(ejl) ;::: L
j j
1[E,IITII]
x(ej,~(dx)ej)
;::: c dim RE
Thus dim RE < 00 for every c > O. In other words T has the form: T =
LjPjluj)(ujl where Pj > 0 and {Uj} is an orthonormal set. Thus J,t(lu)(ul) =
LPj I(u, Uj W= LPj(Uj, Iu) (uluj) for every unit vector u and since any projec-
tion P is of the form Li IVi)(vil for some finite or infinite orthonormal sequence
we have
J,t(P) = LJ,t(lvi)(Vil)
i
= LPj(Uj, IVi)(viluj)
i,j
= LPj(Uj,PUj).
j
•
Corollary 8.11: The set 'J' of all probability distributions on CZJl('~Je) is convex. If
dim 'lie ;::: 3 an element J,t E 'J' is extremal if and only if there exists a unit vector
U such that
J,t(P) = (u,Pu) for all P E CZJl('Je). (8.13)
Proof: The first part is obvious. To prove the second part, consider a probability
distribution J,t on CZJl('Je) defined by (8.13) through a unit vector u. Suppose
(u,Pu) = PJ,tl(P) + (1 - P)J,t2(P), P E CZJl('lIe)
where 0 0 and {Uji' j = 1,2, ... } is an orthonormal set for
each i. Choose P = 1 -Iu)(ul. Then J,tl(P) = J,t2(P) = O. In other words the
orthonormal set {Uji,j = 1,2, ... } is a singleton {Uli}, i = 1,2, and Uli is a scalar
multiple of U for each i. Thus P,I = P,2. This proves sufficiency. The necessity of
the second part is immediate from Corollary 8.10. •
When dim 'Je ~ 3, any extremal probability distribution on CJP('Je) is just a
pure state described by a unit vector up to a scalar multiple of modulus unity
called phaselactor. An arbitrary probability distribution on CJP('Je) can then be
expressed as a convex combination of at most a countable number of pure states.
If ~ is an O-valued observable where (0,2l') is any measurable space then its
distribution in the pure state determined by a unit vector U is just the probability
measure (u,~(·)u).
Example 8.12: Consider the canonical O-valued observable ~JL of Example 7.1
and any unit vector U in L 2 (p,). Then in the pure state u, the distribution v of the
observable ~JL, is given by
EE2l'
where p, itself is a probability measure and U is the constant function 1; then

v = p,. This shows how one can realise an arbitrary probability measure p, on
(0,2l') as the distribution of an observable in a pure state.
Example 8.13: For any complex separable Hilbert space h and any separable
O"-finite measure space (0, 2l', p,) consider the spectral measure ~ on (0,2l') with
respect to 'Je = L2(p" h) defined by
[~(F)ul(w) = IF(w)u(w), u(-) E 'Je, FE 2l'.

~ is an O-valued observable in 'Je and has distribution v in the pure state U given
LIlu(w)II~p,(dw).
by
v(F) =
If {ej}, j = 1,2, ... is an orthonormal basis in h then the correspondence

u(w) ---; {(ej,u(w))h, j = 1,2, ... } is a unitary isomorphism U from L2(p"h)
onto L2(p,) Ell L2(p,) Ell ... and U~(-)U-I is the observable ~JL Ell ~JL Ell ....
Exercise 8.14: Let (0,2l',p,) be as in Example 8.12 and let 'Je = L 2(p,). Suppose
p is a probability distribution in CJP('Je) given by
pep) = ~Pj(Uj,PUj), P E CJP('Je)

j
where {Uj} is an orthonormal sequence in 'Je and Pj > 0, L: Pj 1. For any

function of the form f(wI, W2) = u(WdV(W2) in L2(p, x p,) define
9 Trace class operators and Schatten' s Theorem 43
Then P extends uniquely to a proje~tion in '3f = L 2 (J.L X J.L), If

'IjJ(WI,W2) = L JPjUj (wduj (W2)
j
where the convergence in the right hand side is in L2 (J.L x J.L), then 'IjJ is a unit
vector in 'fie and
p(P) = ('IjJ, P'IjJ) for all P E rzJ>('3f).
This enables one to look at P as an event in '3f and express the E.robability
distribution p in rzJ>('3f) as a pure state in the "enlarged" Hilbert space '3f.
Notes
Gleason's Theorem was first posed as a problem by G.W. Mackey in his lec-
tures on the mathematical foundations of quantum mechanics and then proved by
Gleason [48]. The proof of Gleason's Theorem here is adapted from Gleason [48]
and Parthasarathy [97]. For a comprehensive survey of recent progress in connec-
tion with Gleason's Theorem for probability distributions on projections in von
Neumann algebras, see Maeda [87].
9 Trace class operators and Schatten's Theorem

In the first six sections where '3f was assumed to be of finite dimension, the notion
of trace of an operator played an important role in describing a state as well as the
expectation of an observable in a state. In order to allow observables assuming an
infinite number of values, the restriction that dim '3f < 00 was removed in Section
7. The aim of this section is to extend the notion of trace when dim '3f is no longer
finite and pave the way for describing the expectation of a real valued observable
in any state.
We begin with some definitions. Suppose 5 1,52 are closed subspaces of a
Hilbert space '3f and U : 51 ---+ 52 is a unitary isomorphism from 51 onto 52.
Then U is called a partial isometry with initial space 51 and range 52. For any
bounded operator T on '3f the positive operator (T*T)I/2 is called the modulus
of T and denoted by ITI. For any set 5 C '3f denote its closure by S. Recall that
R(T) denotes the range of T.
Proposition 9.1: (Polar Decomposition Theorem) Let T E C!A('3f). Then there

exists a unique partial isometry U satisfying the following: (i) U has initial space
R(ITI) and range R(T); (ii) T = UITI.
Proof: By definition (ITlu, ITlv) = (u,ITI 2 v) = (u, T*Tv) = (Tu, Tv) for all
u, v E '3f. Thus the correspondence ITlu ---+ Tu is an isometry from R(ITI) onto
R(T). An application of Proposition 7.2 completes the proof. •
We recall without proof a proposition in basic operator theory concerning
the ideals .1>o('3f) and .1> 00 ('3f) of finite rank and compact operators in C!A('3f)
respectively. (See Section 2 for notations.)
Proposition 9.2:
(i) J>o('Je) c J> 00 ('Je) c 0JI('Je);
(ii) J>o('Je) and J> 00 ('Je) are *-closed two-sided ideals in 0JI('Je);
(iii) J>00('Je) is closed in the norm topology and J>o('Je) is dense in J>00('Je);
(iv) T E J>o('Je) if and only if there exist finite orthonormal sets {Uj}, {Vj} and
positive scalars {Sj}, j = 1,2, ... ,n such that
n
T = L sjluj)(vjl;
j=l
(v) If {Sj} is a non-increasing sequence of positive scalars such that Sj ---+ 0 as

j ---+ 00 and {Uj}, {Vj}, j = 1,2, ... are orthonormal sequences then
T = LSjIUj)(Vjl
j
is a compact operator and the sum on the right hand side converges in
operator norm. The polar decomposition T = UITI is determined by
ITI = LSjlvj)(vjl,
j
UVj=Uj, j=I,2, ... ;
(vi) A bounded selfadjoint operator T is compact if and only if there exists an

orthonormal sequence {Uj} and non-zero scalars {Aj} such that Aj ---+ 0 as
j ---+ 00 whenever the sequence is infinite and T = I: Ajluj)(ujl· In such a
case the infinite series on the right hand side converges in norm.
Proof: Omitted.
•
Example 9.3: Let (fl, '!f, p,) be any seperable a-finite measure space and let
'Je = LZ(p,). Suppose K(Wl,WZ) is a complex valued measurable function on
(fl x fl, '!f x '!f, P, x p,) satisfying
r
inxn
IK(Wl,Wz)IZp,(dwdp,(dwz) < 00.
J
Define
(Tu)(w) = K(w,wdu(wdp,(dwd, U E 'Je
Then Fubini's Theorem and Schwarz's inequality at once imply that T is a bounded
operator on 'Je. Indeed, T is a compact operator. It is called the integral operator
induced by the kernel K.
Proposition 9.4: (Minimax principle) Let )'1 2 A2 2 ... be a complete enumera-

tion, inclusive of multiplicity, of all the positive eigenvalues of a positive compact
operator T. Then
An = min max (u, Tu), n = 1,2, ... (9.1 )
dimS=n-1 lIull=i
uES.L
where S denotes any subspace of 'fIC.
Proof: By (vi) in Proposition 9.2 there exists an orthonormal sequence {Uj}

(which may be finite or infinite) such that T = Lj Ajluj)(ujl. Let S be any
subspace of dimension n - 1. Consider U1, U2, ... ,Un' Since for any orthonormal
basis {e j }, 1 :S j :S n - 1 in S, the rank of the matrix (( (ui, e j ) ) ), 1 :S i :S n,
1 :S j :S n - 1 is at most n - 1, there exist scalars a), a2, ... ,an such that
(Li aiui, ej) = 0 for each j and Li lai 12 = 1. If v = I:aiui then v E S.l,
Ilvll = 1 and
n
(v,Tv) = LAilail2 2 An.
i=1
Thus the right hand side of (9.1) is not less than An. On the other hand choose
So to be the span of U1, U2, ... ,Un -1. Then any vESt can be expressed as
v = Lj::~n {3jUj + w where Tw = 0 and
(v, Tv) = L Aj l{3j12 2 Anllvl12.

j?n
If v = Un then (v, Tv) = An. Thus
sup (v, Tv) = An,
vE s t·llvll=1
and the required minimum in (9.1) is attained at So and equals An.
•
Corollary 9.5: For any compact operator T on 'fIC let 81 (T) 2 82(T) 2 ... be
a complete enumeration, inclusive of multiplicity, of the positive eigenvalues of
ITI. Then
8 n (T) = . min max IITull, n = 1,2, ...
dlmS=n-1 lIull=1
uES-.l
where S denotes a subspace of 'fIC.
Proof: This is immediate from Proposition 9.4 if we observe that IITuI12 =

(u, T*Tu) = (u, ITI 2u) and the eigenvalues of ITI2 are the squares of the eigen-
values of ITI. •
The sequence {8 n (T)} (which is empty when T = 0 and finite or infinite
according to whether the rank of T is finite or infinite) in the Corollary above is
called the sequence of singular values of T. If it is infinite the compactness of T
implies that 8 n (T) decreases to 0 as n ---t 00.
Proposition 9.6: Let T be a compact operator on 7fC with singular values Sl (T) ~
s2(T) ~ ... and let T = UITI be its polar decomposition. Suppose {Uj} and {Vj}
are orthonormal sequences such that ITluj = sj(T)uj, UUj = Vj, j = 1,2, ....
Then
T = L sj(T)lvj)(ujl (9.2)
j
where the right hand side converges in norm. In particular, every compact operator
is the norm limit of a sequence of finite rank operators.
Proof: By (vi) in Proposition 9.2, ITI = L: j sj(T)luj)(ujl where the right hand
side converges in norm. Now left multiplication by the partial isometry U (of
Proposition 9.1) yields the required result. •
Corollary 9.7: Let T be as in Proposition 9.6. Then T* is a compact operator

and
T* = LSj(T)luj)(vjl, sj(T*) = sj(T), j = 1,2, ...
j
where {Uj} and {Vj} are the orthonormal sequences in (9.2).
Proof: Immediate.
•
Proposition 9.8: For any T E .1' 00 (7fC), X E rzA(7Je) the following holds:
(i) sj(XT) :::; IIXlIsj(T); (ii) sj(TX) :::; IIXlIsj(T).
Proof: (i) follows from Corollary 9.5. Now from Corollary 9.7
•
An operator T E .1' 00 (7fC) is said to be of trace class if
IITlh = LSj(T) < 00.

j
Denote by .1'1 (7fC) the set of all trace class operators in 7fC.
Proposition 9.9: For any T E .1' 1(7fC), X E rzA(7Je) the following holds:
(i) IIT*1I1 = IITlh; (ii) IITXll1 :::; IIXIlIITlh; (iii) IIXTll1 :::; IIXIlIITlh·
Proof: Immediate from the definition of I . Ill, Corollary 9.7 and Proposi-
tion 9.8. •
Proposition 9.10: Let T E .1'1 (7fC). For any orthonormal basis {ej} in 7fC the
series L: j (ej, Tej) converges absolutely to a limit independent of the basis.
Proof: Consider the representation
T = I>j(T)IVj)(Ujl
j
of Proposition 9.6. Then
L I(ek, Tek)1 = L I L sj(T)(ek, Vj)(Uj, ek)1

k k j
::::; L Sj(T)I(ek, Vj)(Uj, ek)1

j,k
::::; LSj(T){L l(ek,vj)1 2}1/2{L I(Uj,ek)1 2}i/2
j k k
::::; L Sj(T) = IITIII < 00.

j
In particular the double series L:j,k Sj (T)(ek' Vj) (Uj, ek) converges absolutely.
Summing over k first and then over j we get
L(ek, Tek) = L sj(T)(uj, Vj)

k j
and the right hand side is independent of the basis {ed. •

For any T E j l (';Je) the unique scalar L:k (ek' Tek) defined by an arbitrary
orthonormal basis {ek} is called the trace of T and denoted by T.
Proposition 9.11: j I ('Je) is a linear space.
Proof: We shall prove the proposition by establishing the inequality:

sj+k+l(Tl + T2) ::::; sj+I(TI) + sk+I(T2) (9.3)
for any T 1,T2 E joo('Je), j,k = 0,1,2, .... To this end consider any finite set
E C 'Je and for any compact operator T write
q(E, T) = sup IITull.
Ilull=l,UEE-L
By Corollary 9.5
inf q(E, T) = sn(T),
#E=n-I
# denoting cardinality. q(E, T) is decreasing in E and hence
q(E U F, TI + T2) ::::; q(E U F, TJ) + q(E U F, T2)
::::; q(E, Tl) + q(F, T2)'
Thus
Taking infimum over E,F successively but keeping #E = j, #F = k we obtain

(9.3). Thus for any Tl,T2 E j 1(,(Ie) we have
2:>j(T1 + T2) :::; 2 2:>2k+ 1(T1 + T2)

j k~O
: :; 2{~:::>k+l (Tt) + L
k~O k~O
Sk+l (T2)} < 00. •
Propostion 9.12: The map T --+ tr T on j 1 ('(It:) satisfies the following:

(i) trT* = trT;
(ii) trTX = tr XT for all X E ~(~);
(iii) sup IIXII=l
X E !II (:If)
ItrTXI = IITIl1.
Proof: (i) is immediate from the definition. Using the representation (9.2) of
Proposition 9.6 we have for any orthonormal basis {ek} in '(It:
L(ek, TXek) = LLsj(T)(ek,Vj}(X*Uj,ek}

k k j
(9.4)
= LSj(T)(X*Uj,Vj}.
j
thanks to the absolute convergence of the double series in the middle. Similarly
L(ek,XTek} = LLsj(T)(ek,XVj}(Uj,ek}
j k j
(9.5)
= LSj(T)(Uj,XVj}.
j
Comparing (9.4) and (9.5) we obtain (ii). From (9.5)
trXT:::; LSj(T)IIXIl = IIXIlIITIl1. (9.6)

j
If T = UITI is the polar decomposition and we extend U to ~ by putting Uu = 0

for U E R(ITI).L = N(ITI) we have U*UITI = ITI. Thus for X = U* we have
tr XT = tr ITI = IITIll. IIXII = 1 (9.7)
Now (9.6) and (9.7) imply (iii).

•
Proposition 9.13: j 1('(It:) is a Banach space with norm II . 111.
Proof: That 11·lh is a norm follows from (iii) in Proposition 9.12. Suppose {Tn}
is a Cauchy sequence in the norm 11·111. Since IITn -Tmlll ~ IITn -Tmll it follows
that there exists a compact operator T such that IITn - Til -+ 0 as n -+ 00. By
Corollary 9.5
By Fatou's Lemma
L-t sj(T) ~ limn_H)O 'L-t

'" lim IITnlll < 00.
" sj(Tn) = n~oo
j j
Thus T is of trace class. Since
lim sup ItrTmX - trTnXI = lim IITm - Tnlh = 0

m,n->oo Ilxll=l.xE~(~) m,n-> 00
there exists a linear functional A on ?A('1Je) satisfying
lim sup ItrTnX - A(X)I = O.

n->oo Ilxll=l,XE~(~)
Choosing X = lu)(vl with Ilull = Ilvll = 1 we have
A(lu)(vl) = lim trTnlu)(vl = lim (v, Tnu)

n----;.oo n-400
= (v, Tu) = trTlu)(vl.
Thus A(X) = trTX for every X in ?A('1Je) and IITn - Till -+ 0 as n -+ 00. •
Proposition 9.14: Let T be a positive operator and {ej} an orthonormal basis in

'1Je such that Lj(ej,Tej) <00. Then T E 9 1 ('1Je) and trT = Lj(ej,Tej)'
Proof: Let {Jd be another orthonormal basis in '1Je. Then
L Uk, T fk) = l: IITI/2 fk 112 = L I(T1/2 fk, ej) 12

k k k,j
j j
In other words (u, Tu) is a regular non-negative frame function in the language
of Section 8. The rest of the proof is identical to that of Corollary 8.10. •
Proposition 9.15: Let A be any bounded linear functional on the Banach space
9 00 ('1Je). Then there exists a unique T E 9 1('1Je) such that A(X) = trTX for all
X. Furthermore
sup IA(X)I = IITI11'
XE.9\,o(~),llxll=1
Conversely, for any T E 9 1('1Je) the map X -+ tr T X is a bounded linear functional

on 9 00 ('1Je).
Proof: For any u, v E 'Je, Iu)(vl E .9i 00 ('Je) and B(v, u) = A(lu)(vl) is a sesquilin-
ear form satisfying
IB(v, u)1 :=; IIAllllullllvll·
Hence there exists an operator T E rJ3('Je) such that A(lu) (vi) = (v, TU) =
trTlu)(vl for all u,v in 'Je. Consider the polar decomposition T = UITI and
choose an orthonormal basis in {Uj} in R(ITI) so that Vj = UUj is an orthonormal
basis in R(T). Then for any fixed n, LJ=I IUj)(vjl = Xn is a finite rank operator
of unit norm and
n n n
A(Xn) = '2)Vj, TUj) = '2)U*Vj, ITluj) = L(Uj, ITluj).
j=1 j=1 j=1
In particular
00
L(Uj, ITluj) :=; IIAII·

j=1
Choose an orthonormal basis for N(ITI) = R(ITI)l.., pool it with {Uj} and con-
struct thereby an orthonormal basis {fd in 'Je so that
L(/b ITlfk):=; IIAII·

k
By Proposition 9.14, T E .9i 1 ('Je),
IITIII = L(/k, ITlfk):=; IIAII

k
and A(X) = trTX for all X in .9i 00 ('Je). On the other hand by (iii) in Proposition
9.12, IIAII :=; IITIII' Thus IIAII = IITIII' The converse is immediate. •
Proposition 9.16: Let A be any bounded linear functional on the Banach space
.9i 1('Je) with I . lit. Then there exists a unique operator X E rJ3('Je) such that
A(T) = trTX for all T in .9i 1 ('Je) and IIAII = IIXII. Conversely, for any X in
rJ3('Je) the map T --) trTX is a bounded linear functional on .9i 1 ('Je).
Proof: As in the proof of Proposition 9.15 the map (v,u) --) A(lu)(vl) is a
bounded sesquilinear form and hence there exists X in rJ3('Je) satisfying
A(lu)(vl) = tr Xlu)(vl.
By Proposition 9.6 any T in .9i 1 ('Je) can be expressed as T = Lj sj(T)lvj)(Ujl
where LjSj(T) < 00 and {Uj},{Vj} are orthonormal sets. Hence A(T)
trTX = tr XT. Furthermore
Itr XTI = IL Sj(T)(Uj,Xvj)l:=; IIXIIIITIIt·
j
Thus IIAII :=; IIXII. If T = Iu)(vl where Ilull = Ilvll = 1 then Itr XTI
l(v,Xu)1 and IITIII = 1. Thus
sup Itr XTI = IIXII
IITIII=1
and IIAII = IIXII. This proves the first two parts. The converse is immediate. •
Theorem 9.17: (Schatten's Theorem) Let 'Je be any complex separable Hilbert
space. For any trace class operator T and any bounded operator X in 'Je let
(T, X) = trTX. Then the following holds: (i) Under the norm IITlh = tr ITI
the set :J I('Je) of all trace class operators in 'Je is a Banach space isometrically
isomorphic to the dual of the Banach space :J 00 ('Je) of all compact operators in 'Je
with the operator norm under the correspondence T --+ (T, ·)I.~oo(ilf)' T E :JI('Je);
(ii) The Banach space 'Zl3('Je) of all bounded operators in 'Je with the operator norm
is isometrically isomorphic to the dual of :J I('Je) with the norm I . III under the
correspondence X --+ C,X)IJ1(ilf)' X E 'Zl3('Je).
Proof: This is just a summary of Proposition 9.15, 9.16.

We can now restate Corollary 8.10 and describe any probability distribution
•
on the collection f!P('Je) of all events or orthogonal projections on 'Je in terms of
trace as follows:
Theorem 9.18: Let 'Je be a Hilbert space of dimension ;::: 3. Then a map fJ, :
f!P('Je) --+ [0, 1] is a probability distribution if and only if there exists a positive
operator PJ.L E :J 1 ('Je) such that tr PJ.L = 1 and
fJ,(P) = tr PJ.LP for all P E f!P('Je). (9.8)
The correspondence fJ, --+ PJ.L is one to one.
Proof: Suppose fJ, is a probability distribution on f!P('Je). In the notation of Corol-

lary 8.10, use (8.12) to define the positive operator PJ.L = LjPjluj)(ujl. Thus
tr PJ.L = 1 and (9.8) is obtained. Conversely, if p is a positive operator of unit trace
then for P E f!P('Je), tr pP = tr pp2 = tr P pP ;::: 0 since P pP ;::: O. If {Pj } is a
sequence of projections such that PiPj = 0 for i i= j choose orthonormal bases
{Ujj, Uj2, ... } for R( Pj ) for each j and note that
tr pPj = IJUjk> PUjk).

k
The collection {Uj k, k = 1, 2, ... , j = 1, 2, ... } is an orthonormal basis for

R(LjPj) and
trPLPj = L(Ujk,PUjk) = LtrpPj.
j j,k j
Thus the map P --+ tr pP is a probability distribution on f!P('Je). The last part is
immediate from the identity fJ,(lu)(ul) = (u,PJ.Lu) if lIull = 1. •
A positive operator p in :J 1 ('Je) of unit trace is called a state. By Proposition
9.14 a state p is an operator satisfying the relations (u, pu) ;::: 0 for all U and
r.j(ej,pej) = 1 for some orthonormal basis {ej} in 'Je. Such a p is also called
a density matrix. The convex set of all states is denoted by 9'('Je). Any extreme
point of 9'('Je) is called a pure state for the algebra 'Zl3('Je). A state p is pure if
and only if p = Iu) (ul for some unit vector u. By abuse of language (and for
convenience) any unit vector U in 'Je is called a pure state but, truly speaking, it
stands for the state lu)(ul, When 7Je = L 2 (J.t) where J.t is a O"-finite measure any
W,
unit vector U in 7Je is also called a wave function and Iu( w w E 0 is called a
probability wave function, If E is any measurable set in the measure space where
J.t is defined we recall that JE Iu(wW J.t( dw) is the probability that the canonical
e
observable l assumes a value E in the pure state u, The complex scalar u(w) is
called the (probability) amplitude for ~Jl to assume the value w.
Any state p has the decomposition p = ~j pjluj)(ujl where Pj > 0, ~Pj =
1 and {Uj} is an orthonormal set. If (O,;!F) is any measurable space and ~ is an
O-valued observable in 7Je then the distribution of ~ in the state p is the probability
measure E --+ trp~(E),E E;!F.
Proposition 9.19: Let p be a state and let X be a bounded selfadjoint operator

on 7Je with spectral representation X = J x~(dx). Then the expectation of the
bounded observable ~ (or equivalently X) in the state p is tr pX.
Proof: Let p = ~jpjluj)(ujl, Pj > 0 for each j,~Pj = 1,{uj} being an

orthonormal set. Then the distribution v of the observable ~ in the state p is the
probability measure given by v(E) = ~jpjvj(E) where
vj(E) = (uj,~(E)uj) for each j and E E ;!F[J;l.
Since ~ has support in the bounded interval [-IIXII, IIXII] the required expectation
is equal to
J xV(dx) = ~jPj Jx(uj,~(dx)uj) = ~jPj(Uj,XUj)

= 'E.j(Uj,pXUj) = trpX.
•
In the context of Schatten's Theorem and Proposition 9.19, it is instructive to
compare the situation in quantum probability with the role of Riesz' s representation
theorem in classical probability. Suppose 0 is a compact metric space. Denote
by C(O), M(O), B(O) respectively the space of all complex valued continuous
functions, totally finite measures and bounded measurable functions on O. They
are Banach spaces satisfying C(O)* = M(O), M(O)* ~ B(O), * indicating the
Banach space dual. Correspondingly in the case of .<J 00 (7Je),.<J 1 (7Je) and !J3(7Je) we
have a happier situation: .<J 00 (7Je)* = .<J 1 (7Je), .<J 1(7Je)* = !J3(7Je).
Exercise 9.20: An element X E !J3(7Je) is called a Hilbert-Schmidt operator if for
some orthonormal basis {ej} in 7Je,~jIlXejI12 < 00. Denote by .<J 2 (7Je) the set of
all Hilbert-Schmidt operators in 7Je. The space .<J 2 (7Je) satisfies the following:
(i) For any X E .<J 2 (7Je) and any orthonormal basis {ej} in 7Je, ~j IIXejl12
converges to a finite limit independent of the basis;
(ii) .<J 2 (7Je) is a linear space. If X, Y E .<J 2 (7Je) and {ej} is any orthonormal basis
in 7Je then ~j(Xej, Yej) converges absolutely, X*Y E .<J 1 (7Je) and
tr X*Y = L(Xe, Yej);

j
10 Spectral integration and Stone's Theorem on the unitary representations of IRk 53
(iii) With the scalar product (X, Y) = tr X*Y, :P 2 (,(J£) becomes a Hilbert space;
(iv) :P 2 ('(J£) is a *-closed two-sided ideal in <;JA('(J£).
Exercise 9.21: Let K(-,·) E L2(fJ, X fJ,) where (n,';iF,fJ,) is a a-finite separable
measure space. Then the integral operator TK with kernel K (defined in Example
9.3) is Hilbert-Schmidt. If K 1, K2 E L2(fJ, X fJ,) then
trT~ITK2 = J K 1(Wl,W2)K2(Wl,W2)fJ,(dwd/L(dw2).
Notes
Schatten's Theorem is proved in Schatten [118J. The proofs here are based exten-
sively on the first few sections of Simon [124].
10 Spectral integration and Stone's Theorem

on the unitary representations of [Rk
Let (n, ';iF) be a measurable space and ~ : ';iF -7 r;g>('(J£) be an n-valued observable
in a Hilbert space '(J£. By a function on n we shall mean a complex valued
measurable function. Denote by B(n) the space of all bounded functions on n
and by S(n) c B(n) the subspace of all simple functions, i.e., functions with finite
range. Thus to every f in S(n) there corresponds a finite partition {E j }, E j E ';iF
and scalars {G.j} such that
f(w) = LG.jIEj(w), (10.1 )

j
IE denoting indicator of E. Define the integral of such a simple f with respect to
Jfd~
~ by
= LG.j~(Ej). (10.2)
J
We shall now extend this notation to the space B(n) and use it to express any
unitary representation of ~k as the Fourier integral with respect to an ~k-valued
observable. Such a result facilitates the evaluation of characteristic functions (or
Fourier transforms) of distributions of ~k-valued observables in any state.
A set E E ';iF is called ~-null if ~(E) = O. For any real valued function f on
n write
ess.sup f = inf sup f(w)
~ E:~-null wfl-E
and call it the essential supremum of f with respect to ~. Since
sup I(u, ~(E)v) I:::; Ilullllvll

EE?Ji
for all u, v in 7JC, (u, ~(-)v) is a complex valued measure of finite variation on '2F.
If a function I is integrable with respect to this measure we write
JId(u,~(·)v) JI(w)(u,~(dw)v).
=
Two functions I, 9 on 0 are said to be equal almost everywhere with respect to ~

if {w : I(w) # g(w)} is ~-null and in such a case we write I = 9 a.e. (~).
Proposition 10.1: The map I ~ J Id~ from 8(0) into '!A(7JC) defined by (5.1)
satisfies the following:
(i) I ~ J Id~ is linear in I;
(ii) (j Id~)(j gd~) = J Igd~;
(iii) (j Id~)* = J ld~;
(iv) ((j Id~)u, (j gd~)v) = J 19(w)(u,~(dw)v);
(v) II J Id~11 = ess.sup~ III;
(vi) JId~ = Jgd~ if and only if 1= 9 a.e. (~).
Proof: This is elementary and we leave it to the reader with the hint that for
proving (ii) the property ~(E)~(F) = ~(E n F) may be used. •
Suppose I E B(O). Choose In E 8(0), n = 1,2, ... such that
lim sup Iln(w) - l(w)1 = O.
n----+oo w
By (v) in Proposition 5.1
limm,n---.ooll JImd~ JInd~11

- :S limm,n---.oos~pl/m(w) - In(w)1 = O.
J
Thus lim n ---. oo Ind~ in norm exists in '!A(7JC) and routine argument shows that
the limit is independent of the approximating sequence {In} of f. We call this
unique limiting operator the integral of I with respect to ~ and denote it by J Id~.
Proposition 10.2: The map I ~ J Id~ from B(O) into '!A(7JC) satisfies the
properties (i)-(vi) of Proposition 10.1 with 8(0) replaced by B(O).
Proof: Immediate from the definition of integral and Proposition 10.1. •

If I is essentially bounded in the sense that ess.sup~ III < 00 then for any
9 E B(O) such that 9 = I a.e. (~) the integral Jgd~ is independent of the choice
of 9 and we write J Id~ = Jgd~. Then Proposition 10.2 holds with B(O) replaced
by the space of all ~-essentially bounded functions on O. If I and = I-I are 7
essentially bounded then
(J fdO- Jr-Id~.1 =
In particular, the operator J Id~ is unitary if and only if III = 1 a.e. (~).
10 Spectral integration and Stone's Theorem on the unitary representations of ~k 55
We say that two functions I,g on n are ~-equivalent if 1= g a.e. (~). Then
~-equivalence is, indeed, an equivalence relation. By a function we mean its ~
equivalence class in the present context. The space B~(n) of all (such equivalence
classes) of ~-essentially bounded functions on n is an involutive and commutative
Banach algebra with identity where the involution is defined by the map I ----+ f
and IIIII~ = ess.sup~ III. The map I ----+ J Id~ is an injective and involutive
Banach algebra homomorphism from B~(n) into !J3('fIC). It is to be noted that
J Id~ is always a normal operator.
A sequence {In} of functions on n is said to converge in ~-measure to a
function I if
s.lim ~({wllln(w) - l(w)1

n---7OO
> E}) = 0 for every E > O.
{In} is said to converge almost everywhere (~) to I if
~({wlln(w) -F I(w) as n ----+ oo}) = O.

In such a case we write
lim In = I a.e. (~).
n---7OO
Convergence almost everywhere (~) implies convergence in ~-measure. Conver-

gence in ~-measure implies the almost everywhere convergence of some subse-
quence. Indeed, these are immediate from Proposition 7.4.
Proposition 10.3: Let {In} be a sequence of complex valued functions on (n,:!F)

satisfying the following: (i) supn ess.sup~ Ilnl < 00; (ii) In ----+ I in ~-measure
as n ----+ 00. Then ess.sup~ III < 00 and
s.lim
n---7OO
Jlnd~ = Jld~.
Proof: It is immediate from the preceding discussions that I is ~-essentially

bounded. From Proposition 10.2 and (i) and (iv) in Proposition 10.1 we have
for every u in 'fIC.

•
Example 10.4: Let ~/1 be the canonical observable in L2 (JJ,) where (n,:!F, J.l) is
a separable IT-finite measure space. Then for any J.l-essentially bounded function
Ion n
Proposition 10.5: Let ~n : C!J' -+ <;}('ZJen ) be a sequence of n-valued observables

and let ~ = EI1n ~n be their direct sum in EI1n 'ZJen . If / is ~-essentially bounded
then / is ~n-essentially bounded for each n and
J/d~ = EI1n J/d~n.

Proof: Immediate.
•
Proposition 10.6: Suppose 'ZJe, 'ZJe' are Hilbert spaces, (n, C!J') is a measurable space,
U : 'ZJe -+ 'ZJe' is a unitary isomorphism and ~ is a n-valued observable in 'ZJe. Let
e be the n-valued observable in 'ZJe' defined by f(E) = U~(E)U-l, E E C!J'. For
any ~-essentially bounded function / on n, / is e-essentially bounded and
U Jfd~U-l = J /d(.
Proof: Immediate.
•
Proposition 10.7: Let (n, C!J'), (n', C!J") be measurable spaces and let ¢ : n -+ n'
e
be a measurable map. If ~ is an n-valued observable in 'ZJe, and = ~¢-l and /'
e
is a -essentially bounded function on n' then / = l' o¢ is ~-essentially bounded
and
Proof: Immediate.
•
Theorem 10.8: (Stone's Theorem) Let 'ZJe be a Hilbert space and let ~ be an
~k-valued observable in 'ZJe. Let
(10.3)
where ~'J!.. is the Euclidean scalar product LjXjYj. Then the map ~ -+ Ui!<. is a
strongly continuous homomorphism from the additive group ~k into the unitary
group 0u,('ZJe).
Conversely, if ~ -+ Ux is a strongly continuous homomorphism from ~k
into 0u,('ZJe) then there exists-a unique ~k-valued observable ~ for which (10.3)
holds.
Proof: The first part is immediate from Proposition 10.2 and (10.3). To prove
the converse, first assume that there exists a unit vector U E 0u,('ZJe) such that
{Ui!<.u, ~ E ~k} is total in 'ZJe. We call such a vector u-cyclic for the homomorphism
U. Consider the continuous function
(10.4)
10 Spectral integration and Stone's Theorem on the unitary representations of IRk 57
For any finite set {.;!Cj' 1 ::; j ::; n} C ~k and scalars aj, 1 ::; j ::; n we have
~)lras (P(;Ks - ;!Cr) = IILarU~r uI1 2 ~ O.

r,s r
In other words cP is positive definite in the sense of Bochner and cp(Q) = 1. Hence
by Bochner's Theorem [81] there exists a unique probability measure J.l satisfying
CP(;K) = / e-i~'1L dJ.l(yJ for all;K. (10.5)
Let 8 1 = {UxU,;K E ~k},82 = {e-i~',;!C E ~k} C L 2(J.l). Then 8 1 and 82 are

total respectively in '1Je and L 2(J.l). (1004) and (10.5) imply
(U:f.u, u,Y..,u) = cp(J!.. - ;K) = / ei~'~e-i1L'~ dJ.l(~).

Thus the correspondence Uxu -+ ei~. is an isometry from 8, onto 8 2 . By Propo-
sition 7.2 there exists a unitary isomorphism V: '1Je -+ L 2(J.l) satisfying
(VU:f.u)(J!..) = e-i~'1L for each;!C E ~k.
Let !J. E ~k be fixed. Then by Example lOA

(VU!!U:f.u)(J!..) = e-i!!'1Le-i~'1L
= [/ e-i!!.~ ~11(d~)VU:f.ul(J!..)'
~11 being the canonical observable on (~k,;!FlRk,J.l). The totality of 8, implies
VU!! = Je-iQ:~ e(d~)V for each Q E ~k.
Je-i!!.~ ~(d~)
By Proposition 10.6
u!! =
where ~(E) = V-'~I1(E)V,E E ;!FlRk. This completes the proof when a cyclic
vector exists. Otherwise consider the closed cyclic subspace '1Je, generated by the
vectors {Uxu,;!C E ~k} and observe that the operators Ux leave '1Je, and '1Jet
invariant. The map ;K -+ U-X I:J£.LI yields a homomorphism from ~k into GJ.L('1Jet).
We can repeat the argument with a cyclic subspace generated by a vector in '1Jet.
By a routine maximality argument using Zorn's Lemma we can express '1Je as a
direct sum '1Je = tfJ j '1Je j such that U:f. leaves '1Je j invariant and '1Je j admits a unit
cyclic vector Uj for the family ui!.) = U:f.I:J£j' Thus there exists an ~k-valued
observable ~j in '1Je j such that
ui!.) = / e-i~1L ~j(dJ!).

58 Chapter I: Events. Observables and States
Putting ~ = ffij~j and using Proposition 10.5 we obtain (10.3). If 1] is another

/Rk-valued observable satisfying (10.3) then
Je-i~;:J!.(u,~(dJl)v) J = e-iif:J!.(U, 1] (d1!) v)
for all ;f, u, v. By the uniqueness property of Fourier transforms of totally finite
measures we obtain (u, ~(E)v) == (u, 1](E)v), E E ~lI\lk and u, v E 7Je. Thus ~ = 1] .
•
Corollary 10.9: (von Neumann) Suppose ~,1] are /Rm and /Rn-valued observables
in 7Je satisfying
~(E)1](F) = 1](F)~(E) for all E E ~lI\lm, F E ~lI\ln.
Then there exists a unique /Rm+n-valued observable ( in 7Je satisfying
((E x F) = ~(E)1](F) for all E E ~lI\lm, F E ~lI\ln.
Proof: Let
W;!<.J!. = U;!<VJ!..' ;f E /Rm, J!.. E /R n .

The fact that [~(E), 1](F)] == 0 implies U;!< VJ!.. = VJ!..U;!< and hence
W;!<'J!.. W§..1 = W;!<+§.'J!..+!'

The map (;f, J!..) --t W;!<.y is a continuous homomorphism from /R m +n into OU(7Je).
By Stone's Theorem th~re exists a unique /Rm+n-valued observable ( such that
W(;!<.y) = f e-i(;!<'J!.)·(§.·1)((d~ dt).

- }lI\lmXll\ln
Putting J!.. = Q and ;f = Q successively we obtain
U~ = f e- i;!<'§. ((d~ dt)

}lI\lmxll\ln
Vy = f e-iJ!.·1((d~ dt).
- }lI\lmxll\ln
By the uniqueness part of Stone's Theorem ~(E) = ((Ex /Rn), 1](F) = ((/Rm xF)
and hence ~(E)1](F) = ({(E x /R n ) n (/R m x FH = ((E x F). •
If (!1i' ~i)' i = 1,2, are measurable spaces and ~i is an !1i-valued observ-
able in 7Je for each i we say that 6 and ~2 commute with each other or do not
interfere with each other if 6 (Ft)6(F2) = 6(F2)~1(Ft) for all Fi E ~i, i = 1,2.
According to Corollary 10.9 two non-interfering observables with values in /Rm
and /Rn can always be viewed as a single /Rm+n-valued observable.
11 Basic notions of the theory of unbounded operators 59
Stone's Theorem enables us to view an IRk-valued observable as a continuous

homomorphism from the additve group IRk into the unitary group OU(~). Such a
homomorphism is called a unitary representation of IRk or a k-parameter unitary
group. We can now express the characteristic function of the distribution of an
IRk-valued observable in any state p in terms of the representation (10.3).
Proposition 10.10: Let ~ be an IRk-valued observable and let p be a state in ~.

Then the characteristic function ¢(t) of the probability distribution of ~ in the
state p is given by
¢(t) = tr prj e 1-:f ~(d~)l·

i
If P is the pure state lu)(ul where lIull = 1 then

¢(tJ = (u, J eif·:f ~(d~)u).
Proof: It is exactly along the same lines as the proof of Proposition 9.l9. •
Exercise 10.11: Let {~t, t E T} be a family of mutually non-interfering observ-

abIes in ~ with values in IR. Then there exists a measurable space (0, ':iF), an
O-valued observable ~ in ~ and a family of real valued measurable functions
{it, t E T} on 0 such that
~ e isx ~t( dx) = 10 eisft(w) ~(dw) for all S E IR, t E T.
If P is a fixed state and P is the probability distribution of ~ in the state p then

(0, ':iF, P) is a classical probability space and {ftO, t E T} becomes a classical
stochastic process. (Hint: Use Bochner's Theorem, Kolmogorov's Consistency
Theorem and Zorn's Lemma.)
Exercise 10.12: Let U be any unitary operator in ~. Considering the homo-

morphism n ----7 Un from the additive group Z into OU(~) and using Bochner's
Theorem for positive definite sequences deduce the existence of a unique observ-
able ~ in ~ with values in the torus {z : Iz I = I} C C such that
U= 1 Izl=!
z ~(dz).
Notes
For a detailed treatment of Stone's Theorem on unitary representations of IRk,
see Stone [131], Yosida [142]. Regarding the generalisation of Corollary 10.9 to
quantum logics, see Varadarajan [135].
11 Basic notions of the theory of unbounded operators

In the discussion of Example 7.9 we have seen that a bounded real valued observ-
able in 'ZJe can be described by a bounded selfadjoint operator T. By Proposition
9.19 its expectation in any state p is tr pT. In order to extend this treatment to
cover all real valued observables we have to deal with unbounded operators which
are not defined on the whole Hilbert space but only on a domain which is usually
a dense linear manifold in 'ZJe. For the reader's convenience we include a brief
summary of those aspects of the theory of unbounded operators which are very
useful in quantum probability.
Let 'ZJe be a fixed Hilbert space. An operator T on 'ZJe is a pair (T, D(T))
where D(T) C 'ZJe is a linear manifold called the domain of T and T : D(T) ---t 'ZJe
is a linear map. The graph of such an operator T is the linear manifold G(T) C
'ZJe EEl 'ZJe defined by
G(T) = {(u,Tu)lu E D(T)}
where (u, v) denotes u EEl v in 'ZJe EEl 'ZJe. T is said to be closed if G(T) is closed in
'ZJe EEl 'ZJe. T is said to be closable if there exists an operator T such that G(T) is
the closure G(T) of G(T) in 'ZJe EEl 'ZJe. Such an operator T is unique and called the
closure of T. T is said to be densely defined if D(T) is dense in 'ZJe. The linear
manifolds N(T) = {ulu E D(T), Tu = O} and R(T) = {Tulu E D(T)} are
respectively called the null space and range of T. T is said to be invertible if
N(T) = {O} and its inverse T- 1 is the operator defined by
D(T- 1) = R(T), T-1(Tu) = u, u E D(T).
If Tl,T2 are operators such that D(TJ) C D(T2) and T2U = TIU for all u in
D(Tl) then T2 is called an extension of Tl and Tl a restriction of T2. In such a
case we write Tl C T2. Tl C T2 if and only if G(TJ) C G(T2)' If T is closable
then T C T.
For any two operators Tl, T2 on 'ZJe we write Tl +T2 for the operator defined by
D(TI +T2) = D(TJ)nD(T2) and (Tl +T2)U = T1u+T2u for u in D(TI +T2).T1T2
is the operator with D(T1T2) = {ulu E D(T2),T2U E D(T1)} and (T1T2)U =
T1(T2U) for u in D(TIT2). It is quite possible that D(TI + T2) or D(TIT2) may
just be {O}.
Suppose T is densely defined. Then its adjoint T* is the operator defined as
follows:
D(T*) = {vi sup l(v,Tu)1 < oo}
uED(T),lIull=1
or, equivalently, v E D(T*) if and only if the map u ---t (v, Tu) from D(T) into
C is continuous at O. Since D(T) is dense in 'ZJe an application of Riesz's Theorem
shows that there exists a unique v* E 'ZJe such that (v, Tu) = (v*, u) for all u in
D(T). Write T*v = v*. Then D(T*) is a linear manifold and T* : D(T*) ---t 'ZJe
is a linear map. When T E ~('ZJe), T* coincides with the usual adjoint of T. T is
said to be selfadjoint if T = T*.
Example 11.1: Let (0, cg;, fJ,) be any separable O"-finite measure space, '1fC = L2(fJ,)
and f a complex valued measurable function on '1fC. Let T be the operator defined
by
D(T) = {ul J(1 + IfI 2 )luI 2dfJ, < oo},
Tu = fu, u E D(T).
Then T is closed, densely defined, D(T*) = D(T) and T*u = /u for all u in
D(T).
When 0 = R cg; = cg;~, fJ, = the Lebesgue measure and f(x) = x, we say
T = q. Then q is a selfadjoint operator.
Example 11.2: Let'1fC = L2(1R). A function u on IR is called absolutely continuous

if there exists a function v on IR such that
u(b) - u(a) = lb v(x)dx for all a < b.
In such a case we write v = u'. u' is called the derivative of u. v is determined

uniquely almost everywhere. Define
D(T) = {ulu is absolutely continuous, J(IU I2 + lu'1 2)dx < oo},

Tu=u'.
Then T is a closed densely defined operator and T* = -T. Write p = -iT.
The operators q and p where q is as defined in Example 11.1 satisfy the relation
qp - pq c i.
Proposition 11.3: The mapping (u, v) --+ (v, u) from '1fC ffi '1fC onto itself is a unitary
isomorphism. For any linear manifold S C '1fC ffi '1fC let S' = {( v, u) I(u, v) E S}. If
an operator Ton '1fC is invertible then G(T- 1 ) = G(T)'. In particular, T is closed
if and only if T- 1 is closed.
Proof: Immediate.
•
Proposition 11.4: Let T be a densely defined operator on '1fC. Then G(T*) =
G( _T)'.l. In particular, T* is always closed.
Proof: Let (v, T*v) E G(T*). Then (T*v, u) = (v, Tu) for u in D(T). This
can be expressed in '1fCffi'1fC as ((v,T*v), (-Tu,u)) = 0 for all u in D(T). Since
(u, -Tu) E G( -T) we get (v, T*v) E G( _T)'.l.
Conversely, let ((v,w),(-Tu,u)) = 0 for all u in D(T). Then (w,u) =
(v,Tu) is continuous in u. Thus v E D(T*) and T*v = w. i.e., (v,w) =
(v, T*v) E G(T*). This proves the first part. Since S.l is closed for any set
S C '1fC it follows that G(T*) is closed and hence T* is closed. •
Proposition 11.5: Let T be a closable and densely defined operator on '/Ie. Then
T* is also densely defined and T** is the closure of T.
Proof: Suppose that D(T*) is a proper subspace of '/Ie. Then there exists Uo E
D(T*).l,uo "I- O. Thus ((-T*v,v), (O,uo)) = 0 for all v in D(T*). Since the
operations 1 and 1- commute in the space of linear manifolds in 7Je EB 7Je and
S" = S, S.l.l = S we obtain from Proposition 11.4
(O,uo) E G(-T*)'.l = G(T)'.l'.l = G(T) = G(T). (11.1)
Thus Uo = TO = 0, a contradiction. This proves the first part. By repeated
application of Proposition 11.4 we have as in (11.1)
G(T**) = G( _T*)'.l = G(T)'.l'.l = G(T).
Thus T** = T. •
Proposition 11.6: (Closed Graph Theorem) Let T be a closed operator on 7Je with
D(T) = 7Je. Then T E OO(7Je).
Proof: This is a basic result of functional analysis and we omit its proof. •
Proposition 11.7: Let T be a closed, densely defined and invertible operator on

7Je. If T- 1 E OO('/Ie) then 1'* is invertible and T*-l = T- h E OO(7Je).
Proof: Suppose W E N(T*). Then 0 = (T*w, u) = (w, Tu) for u in D(T).

Thus w E R(T).l = '/Ie.l = {O}. Hence N(T*) = 0 and T* is invertible. By
Proposition 11.3, 11.4
G(T*-l) = G(T*)' = G( _T)'.l' = G( _T).l,
G(T-h) = G( _T- 1 ),.l = G( _T)".l = G( _T).l.

Since T*-l and T- h have the same graph they are equal.
•
Proposition 11.8: Let T be a closed and densely defined operator in '/Ie. If T* is
invertible and T*-l E OO(7Je) then T is invertible, T- 1 E OO('/Ie) and T- h = T*-l.
Proof: Suppose u E N(T). Then 0 = (v, Tu) = (T*v, u) for all v in D(T*).
By hypothesis R(T*) = 7Je and hence u = O. Thus T is invertible. By Proposition
11.3, 11.4 we have
G((T*-l )*) = G( _T*-l ),.l = G( _T*).l
= G(T)'.l.l = G(T)' = G(T- 1 ).
Thus (T*-l)* = T- 1 and T- h = T*-l.
For any closed operator T on 7Je let
•
P(T) = {zlz E C,T - z is invertible and (T - Z)-l E OO(7Je)},
I:(T) = C\P(T).
The sets P(T) and L:(T) are respectively called the resolvent set and spec-
trum of T. For z E P(T) the bounded operator R(z, T) = (T - Z)-I is called
the resolvent of T at z.
Proposition 11.9: Let B E OO(~), IIBII < 1. Then 1- B is invertible, (l-B)-1 E

OO(~) and (1 - B)-I = 1 + B + B2 + ... + B n + ... where the right hand side
converges in norm.
Proof: This is straightforward and we leave it to the reader.

•
Proposition 11.10: For any closed operator T the resolvent set P(T) is open.
Proof: If P(T) = 0 there is nothing to prove. Otherwise let Zo E P(T). Then

T - z = (T - zo) - (z - zo) = {I - (z - zo)R(zo, T)}(T - zo).
Let
D = {zllz - zol < IIR(zo,T)II-I} c c.
By Propostion 11.9, {1 - (z - zo)R(zo,T)}-1 E OO(~) and N(T - z) = 0 for
every zED. Furthermore, for ZED,
(T - Z)-I = R(zo, T){l - (z - zo)R(zo, T)} -I E OO(~).
Thus D C P(T).
•
Proposition 11.11: Let T be a closed operator on ~. For any z, z' E P(T)
TR(z, T) = 1 + zR(z, T),
R(z, T) - R(z', T) = (z - z')R(z, T)R(z', T).
In particular, R(z, T) and R(z', T) commute with each other.
Proof: For any z E P(T) we have

TR(z, T) = (T - z + z)R(z, T) = 1 + zR(z, T).
For z, z' E P(T) we have
R(z,T) - R(z',T) = (T - z)-I{(T - z') - (T - z)}(T - Z')-I
= R(z,T)(z - z')R(z',T).
The second equation in Proposition 11.11 is known as the resolvent identity.

•
Proposition 11.12: Let T be a closed and densely defined operator on ~. Then
z E P(T) if and only if Z E P(T*) and
R(z, T)* = R(z, T*) for all z E P(T).
Proof: Immediate from Proposition 11.7, 11.8.
•
A closable and densely defined operator T on ~ is called symmetric if

T C T*. T is called selfadjoint if T = T*. T is called essentially selfadjoint if its
closure T is selfadjoint.
Let Do C D(T) be a linear manifold and let To be the restriction of T to
Do. If T is selfadjoint and the closure of To is T then Do is called a core for T.
Proposition 11.13: (von Neumann's Theorem) A closed symmetric operator T is
selfadjoint if and only if -i and i belong to the resolvent set P(T) of T. In such
a case ~(T) C IR and (T + i)(T - i)-I is a unitary operator.
Proof: Let ± i E P(T) where ± means + and -. Then (T ± i)-I E ~(~) and
R(T±i) = ~. If u E D(T*) there exists v E D(T) such that (T* +i)u = (T+i)v.
The symmetry of T implies that T C T* and hence (T* + i)u = (T* + i)v. By
Proposition 11.12, ±i E P(T*) and hence u = v E D(T). Thus D(T*) = D(T)
and T* = T.
Conversely, let T* = T. Then
(u, Tu) = (Tu, u) = (u, Tu) for all u E D(T).
Thus
II(T ± i)u11 2 = IITul1 2 + IIul1 2 ~ IluW·
Hence N(T ± i) = {O} and R(T ± i) is closed. Let
(u, (T + i)v) = 0 for all v in D(T).
Then u E D(T*) = D(T) and ((T - i)u, v) = 0 for all v in D(T). Since D(T) is
dense in ~,(T - i)u = 0 or u = 0 and R(T + i) = ~. Similarly R(T - i) = '3e.
T±i being closed operators, the Closed Graph Theorem (Proposition 11.6) implies
that (T ± i)-I E ~(~), i.e. ±i E P(T).
Let T = T*, z = a + ib, b =I- O. Then b- 1 (T - a) is selfadjoint. Since
±i E P(b-I(T - a)) it follows that a ± ib E P(T). i.e., ~(T) C R
To prove the last part observe that
(T+i)(T-i)-1 = (T-i+2i)(T-i)-1 = 1 +2iR(i,T).
By Proposition 11.11, 11.12
{I + 2iR( i, T)}{ 1 + 2iR( i, Tn * = {I + 2i R( i, T)}{ 1 - 2iR( -i, Tn
= 1 + 4 R(i, T)R( -i, T) + 2i{R(i, T) - R( -i, Tn = 1.
Similarly {I +
In view of Exercise
2i R( i, Tn * {I + 2i R( i, Tn = 1. •
10.12 every selfadjoint operator determines uniquely a
torus valued observable ~ through the spectral representation of (T + i)(T - i)-I.
We shall exploit this fact to construct an observable on the real line in the next
section.
Notes
For a comprehensive account of the theory of unbounded operators and its appli-
cations to quantum theory, see Kato [69], Reed and Simon [117].
12 Spectral integration of unbounded functions and von Neumann's Spectral Theorem 65
12 Spectral integration of unbounded functions

and von Neumann's Spectral Theorem
Let '1ff be a Hilbert space, (D,2Ji) a measurable space and let ~ be an D-valued
observable in '1ff. For any complex valued measurable function I (modulo ~ mea-
J
sure zero) we shall now introduce an operator Id~, called the integral of I with
respect to ~, which is an extension of the notion introduced for bounded functions
in Section 10. To this end we write
Df = {ulu E '1ff, JII(wW(u,~(dw)u) < oo} (12.1)
Let f-tu(E) = (u,~(E)u), E E 2Ji. Since f-tu(E) = 11~(E)uI12 and hence

f-tu+v(E) S 2{f-tu(E) + f-tv(E)}
it follows that Df is a linear manifold in '1ff. For any u in '1ff define Un = ~(En)u
where En = {ul II(w)1 s
n}, n = 1,2, .... Then Un E Df and Un -+ u as
n -+ 00. Thus Df is dense in '1ff. If I is bounded then Df = '1ff. For any I,
Df = Df·
Let u E D f. Choose a sequence {f n} of bounded functions such that
lim
n--.oo
J lin - 11 2d f-tu = O. ( 12.2)
If Vn = (J Ind~)u is defined as in Section 10 then by Proposition 10.2 and (iv)

in Proposition 10.1 we have
lim
m,n....----+oo
Ilvm- vn l1 2 = lim
m,n--+oo
J11m - Inl 2d f-tu = O.
Define
( / Id~)u = lim Vn, U E Df
n--.oo
and observe that the limit on the right hand side is independent of the approxi-
mating sequence {In} in (12.2). If u, v E Df we can select {In} so that (12.2)
holds for both u, v and this implies (J Id~)( au + (3v) = a(J Id~)u + (3(J Id~)v.
Denote by T(f) the operator for which D(T(f)) = Df and
T(f)u = (J IdOu.
T(f) is called the integral of I with respect to ~. It depends only on the ~

equivalence class of f.
Proposition 12.1: The map I -+ T(f) satisfies the following:

(i) IIT(f)uI1 2= J II(wW(u,~(dw)u) for u in Df;
(ii) If g is ~-essentially bounded, u E D f then u E D fg , T(g)u E Df and
T(fg)u = T(f)T(g)u = T(g)T(f)u;
(iii) T(IE f)u = ~(E)T(f)u = T(f)~(E)u for all u E D f , E E 2Ji.
Proof: (i) is immediate from definitions and the fact that (i) holds for bounded
f, To prove (ii) observe that (12.1) implies Df C Dfg. Let
fn(w) = {f(w) if If(w)1 ::; n,

o otherwIse.
J
Then
nl~moo Ifn(w) - f(wW(1 + Ig(w)12)(U, e(dw)u) = O. (12.3)
As before, writing Pn(dw) for (u,e(dw)u) we have for v = T(g)u, from Propo-
sition 10.1, 10.2
Pv(E) = (T(g)u,e(E)T(g)u) = (u,T(IElgI2)U) = llgl2dPu.
J J
Thus
Ifl2dpv = Ifgl 2dpu < 00
and T(g)u E Df . By (12.3) it follows that
T(fng)u, T(fn)T(g)u and T(g)T(fn)u
converge respectively to
T(fg)u, T(f)T(g)u and T(g)T(f)u as n ~ 00.
Since for each n, T(fn)T(g) = T(g)T(fn) = T(fng) we obtain (ii). (iii) is

immediate from (ii) if we put 9 = IE and note that T(g) = e(E). •
Theorem 12.2: For every complex valued measurable function f on (n, 'Ji) the
operator T(f) is closed and densely defined. Furthermore T(f)* = T(J). If f is
real then T(f) is selfadjoint and
R(z,T(f))=T((f-Z)-I) ifIm z¥=O.
Proof: Let Un E Df' Un ~ u, T(f)u n ~ v as n ~ 00. Let Ek = {wllf(w)1 ::;

k}, k = 1,2, .... By Proposition 12.1
e(Ek)v = lim e(Ek)T(f)u n
n-->oo
(12.4 )
Thus
J
Letting k ~ 00 we get
Ifl 2dpu = IIvl12 < 00
and u E Df. From (12.4) we have T(f)u = v. Thus T(f) is closed. Since Df is
dense in "JC the first part is proved.
For u, V E D f = D f we have from the boundedness of f I Ek
(u, T(j IEk )u) = (T(fIEk )u, v).
Letting k ----+ 00 we get
(u, T(j)v) = (T(f)u, v) for all u, v in Df.

Thus T(j)* ~ T(f). To prove the opposite inclusion consider u E D(T(j)*).
Then
sup l(u,T(j)v)1 = m < 00.
VEDJoIIvll~1
By (iii) in Propostion 12.1 and the inequality 11~(E)vll ::; Ilvll we have
r Ifl2dJlu = IIT(fIEk)uI12
iEk
sup l(u,T(jIEk )v)1 2
VEDJoIIvll~l
< sup I(u, T(j)v)12 = m2.

VEDJoIIvl19
Letting k ----+ 00 we get J Ifl2dJlu < 00. Thus u E Df. In other words T(j)* =
T(f). The last part is immediate from (ii) in Proposition 12.1 and the fact that
(j - z) -I is a bounded function of w whenever 1m z -I- O. •
Proposition 12.3: Let (D, ;Jf), (D', ;Jf') be measurable spaces and let ¢ : D ----+ D'
be a measurable map. If ~ is an D-valued observable in '1Je and = ~¢-1 is e
the D'-valued observable induced by ¢, then for any complex valued measurable
function f on D',
r f de = r f o¢ d~.
in' in
Proof: Immediate.
•
Theorem 12.4: (von Neumann's Spectral Theorem [137]) Let T be a selfadjoint
operator in '1Je. Then there exists a unique real valued observable ~ : ;Jf~ ----+ fJ'('1Je)
such that T = J~ x~( dx). Conversely, for any real valued observable ~ the operator
J
T = x~( dx) is selfadjoint.
Proof: Let T be a selfadjoint operator. By Proposition 11.13 the operator

(T + i)(T - i)-l = I + 2i R(i, T) is unitary. By Exercise 10.12 there is a
unique spectral measure TJ on the Borel subsets of {z I Iz I = I} such that
(T + i)(T - i)-l = 1Izl=l

ZTJ(dz).
We claim that 'T/( {I}) = O. Indeed, if TJ( {I}) ¥- 0 there exists a unit vector u such
that (T + i)(T - i)-lU = U. If v = (T - i)-lU then (T + i)v = (T - i)v. Thus
v = 0, a contradiction. Hence
(T + i)(T - i)-l = 1..I I=I, z'T/(dz).

Z
ztl
Consider the map g : {zllzl = 1, z ¥- I} -+ Iffi defined by g(z) = i(z+ 1)(Z_1)-l

and put ~ = TJg- l . Then ~ is a real valued observable. Using Theorem 12.2 define
the selfadjoint operator T' = J~ x~(dx). By a change of variables (Proposition
12.3) we have
R(i,T') = j(x - i)-l~(dx) = j(g(Z) - i)-lTJ(dz)
= ~
2z
j(z - I)TJ(dz) = R(i, T)
or T' = T. This proves the first part. The converse is contained in Theo-
rem 12.2.
In view of the one-to-one correspondence between selfadjoint operators and
•
spectral measures on the real line as described by Theorem 12.4, we may define a
real valued observable X as a selfadjoint operator but always keeping in view that
the associated spectral projection ~(E) (in the representation X = Jx~(dx)) is
the event that X takes a value in the Borel set E c R Just as sums of real valued
random variables play a fundamental role in classical probability, so do sums of
real valued observables in quantum probability. But there is a complication unique
to this subject. If a real valued observable is interpreted as a selfadjoint operator
then it may not be possible to add two selfadjoint operators Tl and T2 and obtain
a selfadjoint operator Tl + T2 because the domains of Tl and T2 can be different.
In such cases one is forced to interpret the sum of Tl and T2 as a new observable
depending on the special circumstances that arise. In this context the next result
is somewhat general and useful.
Theorem 12.5: (Rellich's Theorem [69]) Let T be a selfadjoint operator and let S
be a symmetric operator in 'J£ with D(S) :::> D(T). Suppose there exist constants
a ~ 0, 0 ~ b < 1 such that
IISul1 ~ allull + bllTul1 for all u in D(T). (12.5)
Then the operator T +S with domain D(T) is selfadjoint.
Proof: Since T and S are symmetric we have

(u, (T + S)v) = ((T + S)u, v) for all u, v in D(T).
Thus (T + S)* :::> T + S. By Proposition 11.13 it is enough to prove that ±i E
P(T + S), the resolvent set of T + S. Since selfadjointness is preserved by real
scalar multiplication it is enough to show that ±iy E P(T + 8) for some y f:. O.
Putting U = R( iy, T)v in (12.5) we obtain
118R(iy, T)II ::; aIIR(iy, T)II + biiTR(iy, T)II. (12.6)

By von Neumann's Spectral Theorem there exists a real valued observable ~ such
J
that T = x~(dx) and
R(iy,T) = j(X - iy)-I~(dx),

TR(iy,T) = j x(x - iy)-I~(dx).
By Proposition 10.1, 10.2
IIR(iy,T)II::; suplx-iyl-I = Iyl-I,

xEIR
IITR(iy, T)II ::; suplxllx - iyl-I = 1.

xEIR
By (12.6)
118R(iy, T)II ::; alyl-I + b.
+ b < 1. Then
Choose Iyl so large that alyl-I
T + 8 - iy = {I + 8R(iy, T)}(T - iy).
By Proposition 11.9, {I + 8R(iy,Tn- 1 E ~('X) and

(T + 8 - iy)-I = R(iy, T){l + 8R(iy, Tn -I E ~(~).
Thus ±iy E P(T + 8) whenever alyl-I + b < 1.

•
Corollary 12.6: If T is a selfadjoint operator and 8 is a bounded selfadjoint
operator in '1Je then T + 8 is selfadjoint.
Proof: 8 satisfies (12.5) with a = 11811, b = O. •

Exercise 12.7: (i) Let Tj be a selfadjoint operator in 'Xj , j = 1,2, .... In the
Hilbert space 'X = ffi'1Jej define
D(T) = {ulu = ffijUj,Uj E D(Tj ), 2)1 Tj Uj 112 < oo} C 'X,

j
TffijUj = ffijTjUj.
Then T is selfadjoint. T is called the direct sum of {Tj } and denoted by ffijTj. If
Dj is a core for each Tj , j = 1,2, ... then
D = {ffijUj, Uj E Dj, Uj = 0 for all but a finite number of j's } is a core
for T.
(ii) Let T be a selfadjoint operator in a Hilbert space ~ and let U : ~ -+ ~'

be a unitary isomorphism onto another Hilbert space ~'. Define the operator T'
by
D(T') = {Uu, u E D(T)},
T'Uu = UTu, u E D(T).
Then T' is a selfadjoint operator. T' = UTU- I •
If two selfadjoint operators T, T' obey the relations above then they are
said to be unitarily equivalent. Two selfadjoint operators T = xe(dx), T' = J
J x((dx), where e,( are spectral measures on the real line, are unitarily equiv-
alent if and only if e, (
are unitarily equivalent.
(iii) If jl is a a-finite measure on (~, ~IR) the operator TIL defined by
D(TIL) = {uJ J(1 + x2)JU(x)J 2djl(x) < oo} C L2(jl),
(TILu)(x) = xu(x)
is a selfadjoint operator, called the canonical (selfadjoint) multiplication operator
J
with spectral resolution
TIL = xe(dx)
where elL is the canonical real valued observable in L2(jl).

Every selfadjoint operator T is unitarily equivalent to a selfadjoint operator
of the form ffil~j~ooTj where
Tj = rILj ffi ... ffi TILj/
v
j = 1, 2, ...
j-fold
Too = TlLoo ffi TlLoo ffi ... ,
jloo, jll, jl2, ... being a family of mutually singular a-finite measures ~. If 1/00'
1/1, 1/2, ... is another such sequence of mutually singular a-finite measures with
the same property then jlj == I/j for aliI::; j ::; 00. (Hint: Use the Hahn-Hellinger
Theorem.)
Exercise 12.8: Let q,p be the selfadjoint operators in L2(~) defined in Example
11.1, 11.2. Define the Fourier transform IF : L2(~) -+ L2(~) by
(lFu)(x) = £.i.m'N---.00(21r)-~
iyi<N
1
e-iXYu(y)dy
where f.i.m. denotes the limit in the mean square sense with respect to Lebesgue
measure. Then IF is a unitary operator and
IFqlF- I = -po
The selfadjoint operators q, -q,p, -p are unitarily equivalent.
Exercise 12.9: Let T = fIR X~( dx) be a selfadjoint operator in '[JC. Then the linear
manifold Do generated by the set {~( E)u, u E '[JC, E a bounded Borel set } is a
core for T.
In particular if {.An} is a sequence of real numbers, {Un} is an orthonormal
basis in '!JC and Do is the linear manifold generated by {un, n = 1, 2, ... } then
there is a unique essentially selfadjoint operator To with domain Do satisfying
Tou n = .AnU n , n = 1,2, ...
If T is the closure of To then
D(T) = {LCnUn : L(1 + .A~)lcnI2 < oo}.

n n
Example 12.10: Let (O,;;;,Jl) be a separable IT-finite measure space. Suppose

'!JC = L 2(Jl),f is a real valued measurable function on 0 and T is the selfadjoint
operator defined by
D(T) = {ul 1(1 + f2)lul 2dJl < oo},
Tu = fu, U E D(T).
Let w be a complex valued measurable function such that Iw(w)1 > 0 a.e.
(Jl) and v be the IT-finite measure defined by dv = (1 + f2)lwI 2dJl. Suppose
{'l/Jn, n = 1, 2, ... } is an orthonormal basis in L 2 (v). Then the linear manifold Do
generated by {w'l/Jn, n = 1,2, ... } is a core for the selfadjoint operator T. Indeed,
let (u,Tu) E G(T), the graph ofT. Then uw- 1 E L2(V). By Parseval's identity
uw- 1 = L.Cn'l/Jn where Cn = l1PnUW-ldV, n = 1,2, ...
1
Thus
lim
n---+oo
lu - ~Cj'l/JjWI2(1
L..J
+ f2)dJl = O.
1
In other words L.~Cj1/JjW E Do and

n n
lim ( " Cj1/JjW, f " Cj'l/Jjw) = (u, fU) in L 2(Jl) ffi L 2(Jl).
n---+oo ~ ~
1 1
This shows that Do is a core for T.

As a special case consider '!JC = L2(lRk) with Jl =Lebesgue measure. Suppose
Po = Po(x[, X2,' .. ,Xk) is any polynomial with real coefficients and T is the
selfadjoint operator on '[JC with
D(T) = {ul 1(1 + P5)luI2d~ < oo}
Tu = PoU.
If CI, C2, ... ,Ck are scalars such that Re Cj > °for each j then the linear manifold
(12.7)
is a core for T. This is immediate since there is an orthonormal basis of polyno-
mials for L2(V) with dv = (1 + #o)exp{- LjRecj x]}dxI" ·dXk.
In L2(lRk) define the Fourier transform operator IF by
(lFu)(~) = l.i.mN->OO(27r)-~ r e-i~·]lu(JL)dJL (12.8)

J1]lI<N
where l.i.m denotes the limit in the mean square sense with respect to Lebesgue
measure. IF is a unitary operator with the property
In (12.7) put Cj = 1, j = 1,2, ... ,k. Then the corresponding domain Do is left in-
variant by IF. Denote by qj the selfadjoint operator T defined in the example above
when the polynomial Po is given by PO(XI, X2, ... ,Xk) = Xj, j = 1,2, ... ,k.
Then IFqjlF- I is a selfadjoint operator with Do as a core and
au
aXj '
IF qj IF -I u = . D
2- U E o.
We write
Pj = -lFqjlF- I , j = 1,2, ... ,k.
For any polynomial PO(XI,X2, ... ,Xk) with real coefficients the linear partial
differential operator Po( -i 8~1 ' •.. ,-i 8~k) is essentially selfadjoint in Do.
We say that two selfadjoint operators Tj = X~j (dx), j = 1,2 commuteJ
with each other if their spectral projections ~I (E) and 6(F) commute with each
other for all Borel subsets E, F of IR. Then {ql, q2, ... ,qk} is a commuting family
of selfadjoint operators. So is the family {PI,P2,'" ,pd. If i -# j then qi and Pj
commute with each other. But qjPj - Pjqj C i where i denotes times the R
identity. These are expressed rather loosely in the following form;
[qr,qs) = [Pr,Ps) = 0,
[qr,Ps) = iors for alII:=:; r,s:=:; k (12.9)
and called Heisenberg commutation relations or canonical commutation relations

or simply GGR.
Notes
Von Neumann's Spectral Theorem is proved in von Neumann [137]. The proof of
Rellich's Theorem has been taken from Kato [69]. For far-reaching generalisations
of Rellich 's Theorem and their applications to the study of Hamiltonian or energy
operators of concrete physical systems, see Kato [69], Reed and Simon [117].
13 Stone generators, characteristic functions and moments 73
13 Stone generators, characteristic functions and moments

We have already seen that a real valued obsevable can be described in three
different ways, namely, as a spectral measure on the real line, a one parameter
group of unitary operators thanks to Stone's Representation Theorem and, finally,
as a selfadjoint operator thanks to von Neumann's Spectral Theorem. The next
result enables us to express the equivalence of the three approaches.
Theorem 13.1: (Stone's Theorem [142]) Let ~ be a real valued observable in '/if
and let {Ut} be the one parameter unitary group defined by Ut = e-itx~(dx), J
J
t E R Let T = x~( dx) be the selfadjoint operator defined by Theorem 12.4.
Then for any U E '/if,limt->ot-1(Utu - u) exists if and only if U E D(T). If
U E D(T) then UtU E D(T) for all t and
:tUtU=-iTUtU=Ut(-iT)U, UED(T),t EIR (13.1)
where the derivative is in the strong sense.
Proof: By (i) in Proposition 12.1
IIT- 1(Ut u - U + it Tu)11 2 = J I

e-itx - 1 + itx
t 12(U,~(dx)u) (13.2)
for all U in D(T). Since

e- itx - 1 + itx
I t I :S Kltlx2 for all t -I- 0, x E IR
for some positive constant K and
Jx2(u,~(dx)u) < 00 for all U in D(T)
it follows from (13.2) and the dominated convergence theorem that

limC1(Utu - u) = -iTu for all U in D(T). (13.3)
t->o
By Fatou's Lemma and (i) in Proposition 12.1
limhOIICl(Utu - u)11 2 = limt->o J I

eitx 1
t- 12(U,~(dx)u)
2: Jx2(u,~(dx)u).
In particular, if t-1(UtU - u) converges as t -+ 0 the right hand side of the
inequality above is finite and U E D(T). This proves the first part.
Let U E D(T). By (13.3)
dU r UHhu - Utu
dt t U = h~O h
= UI ' UhU-U
t 1m =
U( . )
t -zTu .
h->O h
On the other hand, putting UtU = Vt, we have
dU
dt t U
r
= h~
UhVt - Vt
h
.
= -zTvt· •
The selfadjoint operator T obeying the differential equations (13.1) is called
the Stone generator or simply generator of the one parameter unitary group {Ut}.
A fairly general procedure by which important examples of observables are con-
structed in quantum probability goes as follows. Suppose G is a topological group
and t --+ gW is a continuous homomorphism from ~k into G. Let 9 --+ Ug be a
continuous homomorphism from G into the unitary group OU(~) of a Hilbert space
~. Such a homomorphism is called a unitary representation of G in 'Je. Writing
Vr = U g(1) we obtain a k-parameter group which yields an ~k-valued observ-
able in ~. When k = 1 the Stone generator of the corresponding one parameter
group yields a selfadjoint operator which has a domain. When the homomorphism
t --+ g(t) varies these domains also vary. When G is a connected Lie group we
can say something about the common portion of these domains which is useful in
applications.
Proposition 13.2: Let G be a connected Lie group with Lie algebra 'fl and let
9 --+ Ug be a unitary representation of G. For any X E 'fl let Tx be the Stone
generator of the one parameter unitary group {Ug(t), t E ~} where g(t) = exptX.
Suppose
D = {ulu E 'Je, the map 9 --+ Ugu is infinitely differentiable.}
Then D is a core for Tx for every X in 'fl.
Proof: For any U in D the map t --+ Ug(t)u is differentiable. By Theorem

13.1, U E D(Tx). Thus D C D(Tx) for each X in 'fl. Using the fact that
G has a neighbourhood of the identity e which is diffeomorphic to ~d where
d = dim G, we can construct a sequence {<Pn} of probability densities with respect
to a chosen and fixed left invariant Haar measure L of G satisfying the following:
there exists a decreasing sequence {Nn} of compact neighbourhoods of e such that
nn N n = {e}, <Pn has support in N n and <Pn is infinitely differentiable for each
n. We have L(Eg) = L(E)~(g) for all E E CZFG and 9 in G where 9 --+ ~(g)
is a homomorphism from G into the multiplicative group of positive reals. Thus
there exists a real number A such that ~(g(t)) = eAt and
'ljJn(x) = (27r)-t 1 e- At - tt2 <Pn(g(t)xg(t)-l )dt (13.4)
is an infinitely differentiable probability density function for each n and the corres-
ponding probability measure converges weakly in the sense of probability theory
to the Dirac measure at e as n --+ 00. Let U E D(Tx ), v = Txu. Put
Un = l 'ljJn(x)Uxu dx (13.5)
where integration is in the Bochner sense with respect to L. Since
Ugun = L 'ljJn(x)Ugxudx = L 'ljJn(g-lx)Uxudx
it follows that Un E D for each n and Un ---+ U as n ---+ 00. From (13.4) and (13.5)
we also have
Ug(s)Un = J'ljJn(x)Ug(s)Xg(s)-lUg(s)U dx
= Je),s'IjJn(g(S)-lxg(s))Uxg(s)U dx
= (27r)-! r
JGxlR
e),(s-t)-!t2 ¢n(g(t - s)xg(t - S)-I )Uxg(s)u dx dt
Differentiating with respect to s at s = 0 and using (13.4), once again (13.6)

becomes
-iTx Un = L 'ljJn(X)Ux( -iTxu)dx
-(27r)-~ r
JGXIR
te-),t-~e¢n(g(t)xg(t)-I)Uxu dx dt.
The first term on the right hand side converges to -iTxu as n ---+ 00. The second
term on the right hand side can be expressed as
-(27r)-! ~ te-!t2 {fa ¢n(x)Ug(t)-lxg(t)udx}dt

where the integral within the brackets { } converges to U as n ---+ 00 for each
t and remains bounded in norm by lIull. Thus the second term converges to 0 as
n ---+ 00. In other words (Un, Txun) ---+ (u,Txu) as n ---+ 00. •
Example 13.3: Let G = IRk be the k-dimensional additive group of the vector
space IRk in which {ej, j = 1, 2, ... , k} is the canonical basis. Define the unitary
representations t ---+ Ut and t ---+ Vt of IRk in L2(lRk) by
(Utf)(;r) = e-if.if f(;r),
(l!if)(;r) = f(;r - D, t E IRk,

and put U?) = Utej' v,,(j) = lItej' t E IR, 1 ~ j ~ k. If IF is the Fourier transform
defined by (12.8) then
The Stone generators of uji) and V?) are qi and Pi respectively where qi and
Pi are as defined in (12.10). The operators U§.. and Vi satisfy the relations:
(13.7)
These are known as Weylcommutation relations. This is a rigorous description of
the Heisenberg commutation relations (12.9) in tenns of unitary operators. If ~ is
the canonical spectral measure in L2(~k) then
U§.. = Je-i.H.~(d~)
and (13.7) can be expressed as an imprimitivity relation in the sense of G.W.
Mackey [85]:
Vi~(E)ll-l = ~(E + t.) for all Borel sets E C ~k. (13.8)
Example 13.4: Let 80(3) be the compact group of all third order real orthogonal
matrices of unit detenninant and let G be the Euclidean group in ~3:
G = {(!!,R)I!! E ~3,R E 80(3)}
where the group operation is given by
(!!, R)o(/!., 8) = (!! + R/!., R8).
and the topology is the product of the topologies of ~k and 80(3). Suppose
'Je = L2(~3) and
(V(~,R)f)(~) = f(R-l(~ - !!)), f E 'Je.
Then (!!, R) --+ V(~.R) is a unitary representation of G. The family {V(~,l)'!! E ~3}
is a 3-parameter unitary group which is the same as {Va,!! E ~3} in Example
13.3 when k = 3.
Denoting a point ~ in ~3 as a column vector of three coordinates Xl, X2, X3
consider the rotation about the X3 axis through an angle () and denote it by R~3).
Then
COS () - sin () 0 )
R~3) = ( sin () cos () 0 .
001
The family {V(Q,R~3», () E ~} is a one parameter unitary group. Its Stone generator
L3 is given by
( -iL3f)(~)
= :() f(Xl cos () + X2 sin (), -Xl sin(} + X2 cos (), x3)III=O
=X2
of
--Xl-
of
OXl OX2
for all f in the domain Do defined by (12.7) with Cj = 1 for j = 1,2,3. This is
expressed as
L3 = qlP2 - q2Pl
where qj,Pj are as in Example 12.10 with k = 3. L3 is also essentially selfadjoint
on Do. It is called the orbital angular momentum about the X3 axis of a particle
moving in 3-dimensional space. Similarly the orbital angular momenta Ll and L2
about the Xl and X2 axes can be defined and expressed in vectorial notation:
l = TX p, T= (ql,q2,q3), p= (Pl,P2,P3)
where qj is the j-th coordinate of the position observable and Pj is the j-th coor-
dinate of the momentum observable. In the domain Do we have the commutation
relations:
[L l , L2] = iL3, [L2, L3] = iLl, [L3, LJ] = iL2.
Since R~j) is periodic in () of period 2n, it follows from the uniqueness part of
Stone's Representation Theorem that the observables L j, j = 1, 2, 3 take values
in the set Z of all integers whereas qj and Pj, j = 1, 2, 3 take all values in the
real line.
Example 13.5: Let SU2 denote the group of all second order complex unitary
matrices of unit determinant. Any element 9 in SU2 can be expressed as
9 = (_$ ~) where lal 2 + 1,612 = 1.

SU2 acts as rotations in ~3 as follows: Express ;f E ~3 as
X3 iX2)
;f = (
Xl
+ .
2X2
Xl -
-X3
= X10"1 + X20"2 + X30"3
in terms of the Pauli spin matrices 0"j, j = 1, 2, 3 and put
9(;f) = g( Xl 0"1 + X20"2 + X30"3 )g*. (13.9)

Let'JC = L2(~3)EBL2(~3) where every element 'I/J is expressed as a column vector
(~~),'l/Jj E L2(~3), j = 1,2. Consider the group-
G = {(~,g),~ E ~3,g E SUz}
with multiplication
(~,g)o(Q, h) = (~+ g(Q),gh)
and product topology. Define the unitary operators
(V(Q,g)J£)(;f) = gJ£(9-1(;f - ~))

where gJ£ is left multiplication by the matrix 9 but 9 -1 (;f - ~) is defined through
(13.9). Then {V(Q,l)'~ E ~3} is a 3-parameter unitary group which is a direct
sum of two copies of the unitary group of Example 13.3 when k = 3. If gj(t) =
exp -i ! (Jj then {V(Q,gj(t))} is a one parameter unitary group with Stone generator
Lj +4(Jj where Lj is the direct sum of two copies of the orbital angular momentum
observable in Example 13.4 and (Jj is the matrix multiplication operator: '!!!.. --+ (Jj'!!!...
We shall now conclude this section with an extension of Proposition 9.19
which enables us to compute the expectation of an unbounded real valued observ-
able in a state p. We shall adopt the following conventions. By an observable we
shall mean a real valued observable which is described by a selfadjoint operator
X in a Hilbert space '(Jf,. Suppose {Xl.X2, ... ,Xn} is a commuting family of
selfadjoint operators with spectral representations
Xj = 1 xej(dx), 1 ~ j ~ n.
By Corollary 10.9 there exists an ~n-valued observable esuch that
Xj = f xje(d*.), 1~ j ~ n.
JRn
For any complex valued measurable function f on ~n we write
f(XI,X2, ... ,Xn) = f f(XI, ... ,xn)e(dXI···dxn ) = f za-l(dz)

kn k
so that f(Xl. X 2, . .. , Xn) is a complex valued observable. If f is real valued then
f(X I,X2, ... ,Xn) is selfadjoint and hence a real valued observable. If If I == 1
then f(Xl. X2, . .. , Xn) is a unitary operator. For example,
eWK = ei(tlXl+·+tnXn ) = f ei!-~e(d*-)

JRn
is a unitary operator for every t in ~n. If p is any state (Xl.X2, ... ,Xn) has the
joint probability distribution tr pe(·) with characteristic function ¢a) given by
¢(t) = tr pei!'K = Jei!-~ tr pe( d*-).
If p = Iu)(ul where u is a unit vector then

¢(t) = (u,ei!-Ku) = Jei!-~(u,e(d*-)u).
Proposition 13.6: Let X be an observable and p a state in 'JC such that
p = LPjluj)(ujl, Pj > O,j = 1,2, ... , LPj =1
j j
where {Uj} is an orthonormal sequence (of finite or infinite length). Then the n-th
moment of (xn)p of X in the state p exists if and only if the following holds:
(i) Uj E D(IXln/2) for each j;
(ii) LjPj111Xln/2Uj 112 < 00.
In such a case
(Xn)p = Epj(IXl n/ 2Uj, (sgn xtlxln/2uj). (13.10)
Proof: The probability distribution of X in the state p is the measure J-L given
by J-L(E) = L.jPj(uj,e(E)uj),E being any Borel subset of ~ and the spectral e
measure in the resolution of X. Then the n-th moment (xn) p exists if and only
if
j IxlndJ-L(x) = ~ Pj j Ixln(Uj, e(dx)uj) < 00.
J
By Proposition 12.1, property (i) and the definition of spectral integrals, this is
equivalent to conditions (i) and (ii) of the present proposition. In such a case by
property (ii) of Proposition 12.1 we get
j xndJ-L(x) = L Pj j(sgn x)n Ixln(uj,e(dx)uj)

J
= L pj(IXl n/2uj, (sgn xtl x ln/2uj)

j
which proves (13.10).

•
Corollary 13.7: Let X,p be as in Proposition 13.6. Then the second moment
of X in the state p is finite if and only if Uj E D(X) for each j and L.jPj
IIXujl12 < 00. In such a case its mean and variance are respectively given by
(X)p = L Pj(Uj,XUj) = m,
j
((X - m)2)p = L pjll(X - m)ujI12.

j
Proof: Immediate.
•
Exercise 13.8: (Stone-von Neumann Theorem [135]) Let Uf , Vi,i E ~k be two
strongly continuous k-parameter groups of unitary operators in a Hilbert space 7JC
satisfying the Weyl commutation relations:
ViU~ = ei~'iU~ Vf for all Q, i in ~k.
Suppose eis the spectral measure in ~k satisfying
U~ = j e-i§.·if e(d;K.).
Then
Vie(E)Vi- 1 = e(E + i), E E ;Jf~k, i E ~k.
From this equivalence between the spectral measures eo

and e(· + t) and the
Hahn-Hellinger Theorem it follows that there exists a unitary isomorphism r :
7JC ---+ 7JCo = EBkL2(~k), where the direct sum is over at most a countable number
of copies of L2(~k), satisfying
(re(E)r-1[)(;r) = IE(;r)[(;K.), t E EBL2(~k).
Define unitary operators vt in '11f by
Then vtV~t commutes with ~(E) for every E in ';iF. This implies the existence of
a unitary operator A of the form
in '11fo where A(·) is a Lebesgue measurable unitary matrix valued function and
r
the unitary isomorphism 0 = Ar satisfies the following:
(roU§fo\OC;r) = e- i §.' 5fl.(:J;.),
(rovtro1l.)(:J;.) = l.(:J;. - t).

(It is to be noted that any element f
in '11fo is being viewed as a finite or infinite
column vector valued function on IRk.) Thus up to unitary equivalence any solution
for Weyl commutation relations is a direct sum of copies of the example discussed
in Example 13.3.
Exercise 13.9: The expectation of an observable X in a state p is finite if and

only if IXll/2pl/2 is a Hilbert-Schmidt operator. If the second moment of X in
the state p is finite then X pl/2 is Hilbert-Schmidt and (X) p = tr pl/2 X pl/2.
Exercise 13.10: (Heisenberg's uncertainty principle) Let X, Y be two observables

in '11f and let u be a unit vector in D(XY) n D(Y X) such that
(u,Xu) = a, (u,Yu) = b.
Then in the pure state u
(13.11)
When '11f = L2(1R), X = q, Y = P where q and p are as in Example 11.1, 11.2

then
(13.l2)
Equality in (13.11) is obtained if and only if one of the following conditions holds:
(i) Either Xu = au or Yu = bu;
(ii) (X + iaY)u = (a + iab)u for some real scalar a.
Equality in (13.12) is obtained if and only if
u(x) = A(27ra)-i exp[-~(a + iab - X)2]

2a
where a is a positive scalar and A is a scalar of modulus unity. (See Proposition
5.1, Exercise 5.6.)
l3 Stone generators, characteristic functions and moments 81
Exercise 13.11: (i) Let qj, Pj be as in Example 13.4. The operators qlP2 and
PI q2 are essentially selfadjoint and their closures are observables with Lebesgue
spectrum in (- 00, 00 ). The operator qlP2 - PI q2 is essentially selfadjoint and its
closure is an observable whose spectrum is the discrete set Z of all integers.
(ii) Let ej, j = 0,1,2, ... be an orthonormal basis for a Hilbert space of
infinite dimension. Suppose
00
x = 2)le2j) (e2j+1 I + le2j+1)(e2j I),

j=O
00
Y = 2)le2j+1)(e2j+2 I + le2j+2) (e2j+I!),

j=O
1 00
Z = 2:(X + Y) = 2)lej) (ej+1 I + lej+I)(ejl),

j=O
Z' = ~(X - Y).

Then X and Y are bounded observables with finite spectrum { -1,1 } and { -1,0,1 }
respectively, whereas Z and Z' have Lebesgue spectrum of multiplicity one in
°
the interval [-1, 1]. In the pure state eo, X has binomial distribution with equal
probability for -1 and 1, Y has degenerate distribution at and the observables
Z, Z' have the standard Wigner distribution with density function ~ (1 - X2) I /2 in
the interval [-1,1]. (See Exercise 4.5.)
The "observables" qlP2 and -PI q2 take all values in the real line whereas
qlP2 - PI q2 takes only integral values. The observables Z, Z' take all values in
[-1, 1] but Z + Z' takes only the two values -1 and 1. X assumes the values
-1, 1, Y the values -1,0, 1 but X + Y assumes every value in [-2,2]. Thus, due
to interference between observables, waves can produce quanta and, conversely,
quanta can produce waves.
Exercise 13.12: In Example 13.3 let qj,Pj, 1 :::; j :::; k be the Stone generators
of U?), ~(j) respectively in L2(~k). Consider the dense linear manifold
9' = {e -I.q;l p(~) Ip a polynomial}
in L2(~k). Define the operators {W?), t E ~,1 :::; j :::; k} by
(Wt(j)u)(~) = e-!u(xI,X2,'" ,e-txj,'" ,Xk), U E L2(~k)

Then {Wt(j)} is a one parameter group of unitary operators. Denote by Xj its
Stone generator. Then {Xj, 1 :::; j :::; k} is a commuting family of selfadjoint
operators for each of which 9' is a core. Furthermore
1
Xru = 2: (qrPr + Prqr)u for all U E 9',
i[qrPs,qsPr]u = (Xr - Xs)u for all u E 9'.

Notes
For a detailed treatment of Stone generators, see Stone [131], Yosida [142]. The
interplay between group representations and observables of quantum theory is a
major part of the work of E.P. Wigner [140]. It is the central theme of G.W.
Mackey's lectures. For an extensive treatment of this subject and a rich collection
of references and historical comments see Mackey [85], [86] and Varadarajan
[135].
In the context of Proposition 5.1, Exercise 5.6, 13.10 it is important to
note the possibility of describing the uncertainty principle in terms of quantities
not dependent on the existence of second moments. We refer to the survey by
Bialynicki-Birula [21].
14 Wigner's Theorem on the aotomorphisms of CZP(7Je)

In Section 6 we have already seen the role of the action of the unitary group OU(Z1C)
on events, observables and states in Z1C in the context of dynamics in continuous
time. In such a situation it is natural to ask whether there exists a group larger than
OU(Z1C) acting on the collection rzJ>(Z1C) of all events. Indeed, if U is an antiunitary
operator in Z1C then the map E --+ UEU- 1 induces an "automorphism" of rzJ>(Z1C).
To elaborate, we introduce the following definition. A map T : rzJ>(Z1C) --+ rzJ>(Z1C)
is called an automorphism of rzJ>(Z1C) if it is one to one, onto and the following
conditions hold: (i) T(O) = 0, T(l) = 1; (ii) T(VjEj) = VjT(Ej), T(/\jEj) =
/\jT(Ej) for every sequence {E j } in rzJ>(Z1C); (iii) T(l - E) = 1 - T(E).
All the automorphisms of rzJ>(Z1C) constitute a group under composition. We
denote this group by Aut rzJ>(Z1C). If U is a unitary or antiunitary operator in Z1C
then the map Tu : E --+ U EU-], E E rzJ>(Z1C) is an automorphism of rzJ>(Z1C). The
aim of this section is to prove the converse of this statement when dim Z1C ~ 3.
For any u, v in ':JC say that u is equivalent to v, or u '" v in symbols, if there
exists a scalar A of modulus unity such that v = AU. Denote by ii, the equivalence
class of all vectors equivalent to u. ii, is called a ray of length Iluli. If Ilull = 1,
ii, is called a unit ray.
Proposition 14.1: Let u be a unit vector and let T E Aut rzJ>(':JC). Then there
exists a unit vector v such that T(lu)(ul) = Iv)(vl.
Proof: Let E = T(lu)(ul). If dim R(E) ~ 2 we can find two non-zero projections
E], E2 such that E = El +E2, EIE2 = E2El = O. Then T-l(Ed and T- 1(E2) are
mutually orthogonal non-zero projections satisfying lu)(ul = T-1(Ed +T- 1(E2),
a contradiction. Thus E is a one dimensional projection. •
Let p be any state in ':JC. For any T E Aut rzJ>(':JC) consider the function
/L(E) = tr pT(E), E E rzJ>(':JC). Then /L is a probability distribution on rzJ>(':JC). By
Gleason's Theorem there exists a unique state pT satisfying
tr pT E = tr pT(E) for all E in rzJ>(':JC).
14 Wigner's Theorem on the automorphisms of'lf>('iJf) 83
Extend this correspondence p ---+ pT to the convex set of all positive operators
with finite trace by putting
TT = (trT){(trT)-ITY if T i- 0,
=0 if T = O.
Then trTT = trT. If T = ~cjTj where T ;::: 0, Tj ;::: 0, Cj ;::: 0 and trT < 00
then
TT = LCjTJ.
j
According to Proposition 14.1 the automorphism r induces a one to one onto

map r : u ---+ v on the set of unit rays where Iv)(vl = r(lu)(ul). If {Uj}
is an orthonormal basis in 'lIC and r(luj)(ujl) = IVj)(vjl then {Vj} is also an
orthonormal basis in 'lIC.
Proposition 14.2: If T = 1u) (uland II U II = 1 then there exists a unit vector

v E r-I(u) such that TT = Iv)(vi.
Proof: By definition, for any v E r-I(u), r(lv)(vl) = Iu)(ul. If w is any unit
vector such that (w, v) = 0 then r(lw)(wl) and r(lv)(vl) are mutually orthogonal
projections. We have tr TT = 1 and
(W,TTW) = trTTlw)(wl = trTr(lw)(wl)
= trr(lv)(vl)r(lw)(wl) = O.
Thus TT = Iv)(vl. •
Proposition 14.3: For any positive operator T of finite trace
(T2y = (TT)2.
Proof: Choose Ai ;::: 0 and an orthonormal basis {Ui} so that

T = LAilui)(Uil, T2 = L A;lui)(Uil.
i i
Then by Proposition 14.2
(T2y = LA;(lui)(UilY = LA;lvi)(Vil

i i
where Vi E r-I(ui) and {vil is also an orthonormal basis. Thus
(T2y = (LAilvi)(Vilf = (TTf

•
We observe that for any u, v in 'lIC and any UI E U, VI E v, I(UI, vI)1 =
1(u, v) I. Thus we may as well write
l(u,v)1 = l(u,v)l·
Proposition 14.4: For any two unit rays U, v and any 7 E Aut 9J('JC)
l(u,v)1 = 1(7(U),7(V))I·
Proof: Consider the positive operator T = lu)(ul + Iv)(vl where u,v are unit
vectors. By Proposition 14.2
(Iu)(ulr = lu')(u'l, (Iv)(vlr = Iv')(v'l

where u' E 7- 1(U), v' E 7-1(V). Since
trT2 = 2{1 + l(u,v1l 2}, tr(TTf = 2{1 + l(u',v')1 2}
and trT2 = tr(T 2)7 = tr((TT)2 we have
l(u,v)1 2 = l(u',v')1 2
which is equivalent to I (u, v) 1 =
For any non-zero u E '1JC define
1 (7- 1 (u), 7- 1(v)) I·
•
7(U) = lIuIl7(llull- 1u)}
Then the correspondence U -+ T ( u) is a one to one onto map on the set of all
rays. By Proposition 14.4
l(u,v)1 = I(T(U),7(V))I· (14.1 )
Theorem 14.5: (Wigner's Theorem [135]) Let 'JC be a Hilbert space of dimension
2: 3. Then, to every automorphism 7 of 9J('JC), there corresponds a unitary or
antiunitary operator U satisfying
T(E) = UEU- 1 for all E E 9J('1JC).
If V is another unitary or antiunitary operator in 'JC satisfying the identity T(E) =
V EV- 1 for all E in 9J('JC) then there exists a scalar c of modulus unity such that
V=cU.
Proof: Let T E Aut 9J('JC). Consider the induced map u -+ T( u) on the set of
all rays satisfying (14.1). Choose and fix a unit ray e. Set
Ue = e', e' E 7(e)
where e E e, e' E T( e) are again chosen and fixed. Consider
u=e+z,zEe1-, v=lIzl1- 1z
Since T is an automorphism it is clear that any vector W E T( u) is of the form
ae' + j3v' where v' E 7(V). By (14.1)
I(ae' + j3v', e')1 = lal = I(e + z,e)1 = 1,
I(ae' + j3v', v')1 = 1131 = I(e + z,v)1 = IIzll·
14 Wigner's Theorem on the automorphisms of '2i'(~) 85
Thus e' + a-I (3v' E T( it) and this is the only vector in T( it) with unit coefficient
for e'. Put
Vz = a- l (3v',U(e + z) = e' + a- l (3v' E T(it),}
(14.2)
Uz = Vz, Z E e.l
If A is any scalar it follows from (14.2) that
V(AZ) = X(A,Z)VZ, IX(A,Z)I = IAI (14.3 )
where X(A, z) is a scalar depending on A and z. Furthermore if ZI, Z2 E e.l we
have from (14.2) and (14.3)
I(VZI, VZ2)1 = I(ZI,Z2)1,
(14.4 )
I(U(e + ZI), U(e + z2))1 = I(e + ZI,e + z2)1,
i.e.,
I(e' + VZI,e' + VZ2)1 = I(e + ZI,e + z2)1
or
11 + (VZI, Vz 2 )1 = 11 + (zJ,z2)1· (14.5)
From (14.4) and (14.5) we obtain
Re(VzI, Vz 2 ) = Re(zl,z2), (14.6)
(V ZI, V Z2) = (ZI, Z2) if (ZI, Z2) is real. (14.7)

Putting ZI = AIZ, Z2 = A2Z we get from (14.3) and (14.6)
ReX(AI,z)X(A2,Z) = Re"X I A2. ( 14.8)

Putting Al = i,A2 = 1 we have Rex(i,z) = O. By (14.3) we get X(i,z) = ±i,
i.e., X(i,z) = 'TJ(z)i where 'TJ(z) = ±1. By (14.8)
Rex(i,z)X(A,Z) = Re(-iA) = ImA

= 'TJ(Z) Re(-iX(A, z)) = 'TJ(z)lmx(A,z).
Thus
1m X(A, z) = 'TJ(z) 1m A.
(14.8) also implies
Re X( A, z) = Re A.
Thus X( A, z) == A or "X for all A when Z is fixed. Since e.l is at least two
dimensional we can choose two vectors ZI, Z2 E e.l such that Ilzdl = IIz211 = 1
and (ZI, Z2) = O. Consider aZI + (3Z2. Then
(azi + (3Z2,a-- I ZI) = 1.
From (14.7) we get
But
V(azl + {3Z2) = a'VzI + {3'VZ2
where la'i = lal, 1{3'1 = 1{31 thanks to (14.1). Furthermore (VZI, VZ2) = O. Now
(14.3) implies
a'x(a- I , zJ) = 1. (14.9)
Since (azl, a-I Zl) = 1, (14.7) implies
X(a,zJ)X(a-l,zJ) = 1. (14.10)
(14.9) and (14.10) imply a' = x(a,zJ). Similarly {3' = X({3,Z2). Thus
+ {3Z2) = x(a,zJ)VzI + X({3,Z2)VZ2.
V(azl (14.11)
Whenever IlzI11 = IIz211 = 1, (ZJ,Z2) = 0, ZJ,Z2 E e-L. Putting a = {3 we have
x(a,zl + Z2)V(ZI + Z2) = x(a,zl + Z2)VZI + x(a,zl + Z2)VZ2
= x(a,zJ)VzI + x(a,x2)Vz2.
Since (V Zl, V Z2) = 0 we have X( a, zJ) = X( a, Z2) = X( a, Zl + Z2). Let Z =
aZI + {3Z2. Then for any scalar 'Y
V'Yz = X('Y,z)Vz = X('Y,z)x(a,zJ)VzI + X('Y,z)X({3,Z2)VZ2.
On the other hand
V'Yz = x(a'Y,zJ)VzI + X('Y{3,Z2)VZ2.
Since (V ZJ, V Z2) = 0 we conclude from the last two equations that
x(a'Y,zJ) = X('Y,z)x(a,zJ).
Since X( A, z) == A or == X when z is fixed in e-L we have
x(a'Y, zJ) = x(a, zJ)X('Y, zd·
Thus X('Y, z) = X('Y, zd is independent of z so that we can write X('Y, z) = X('Y).
Furthermore X('Y) == 'Y or X('Y) == 'Y and
Vv = X('Y)Vz for all z E e-L. (14.12)
By (14.11) and (14.12) we have
V(x + y) = Vx + Vy for all x,y E e-L. (14.13)
Changing Zl to iZI in (14.6) and using (14.12) we get from (14.6)
(Vx, Vy) = X((x,y)) for all x,y E e-L. (14.14)
Now for any vector ae + {3z, z E e-L, a # 0 define
U(ae + {3z) = x(a)e' + X({3)Vz.
Then (14.12)-(14.14) are easily verified with U in place of V and x,y,z in '(ff,.
For any x,Ux E T(X). This implies
Ulx)(xIU- 1 = T(lx)(xl) for any unit vector x.
14 Wigner's Theorem on the automorphisms of 'l/'('iJe) 87
Thus UEU- 1 = r(E) for all E in !Jl('lfC). U is unitary if x(,) ==, and antiunitary
if x(f) == 'Y. This proves the first part.
To prove the second part consider two operators U1, U2 each of which is
either unitary or anti unitary and
U1EU1- 1 = U2EU2- 1 for every E E !Jl('lfC).
Using this identity for one dimensional projections we conclude that U1x =
c(x )U2X for every x in 'lfC where c(x) is a scalar of modulus unity. If x, Y are
linearly independent we get
U1(x + y) = U1x + U1Y = C(X)U2X + C(y)U2Y,
U 1(x + y) = c(x + y)U2(x + y) = c(x + y)U2x + c(x + y)U2Y.
Since U2X and U2Y are also linearly independent we have c(x) = c(y) = c(x+y).
Suppose y = ax where a is a non-zero scalar. Choose z linearly independent of x.
Then by the previous argument c(y) = c(ax+z). Since x and ax+z are linearly
independent c(ax+z) = c(x). Thus c(x) = c(y). In other words U1 = CU2 where
c is a constant of modulus unity. •
Wigner's Theorem enables us to describe the group Aut !Jl('lfC) in the follow-
ing manner. All unitary and anti unitary operators on 'lfC constitute a group which
we shall denote by OUs1('lfC). The product of an antiunitary operator and a unitary
operator is antiunitary. OUs1('lfC) with strong topology is a complete and separable
metric group in which OU('lfC) is an open normal subgroup and OUs1('lfC)jOU('lfC) is
a group consisting of two elements. If s is any conjugation in 'lfC then
OUs1('lfC) = OU('lfC) u sOU('lfC)
is a coset decomposition of OUs1('lfC). The subgroup T = {AIIIAI = I} is the centre
of OUs1('lfC). Wigner's Theorem is equivalent to saying that the groups Aut !Jl('lfC)
and OUs1('lfC)jT are isomorphic. Aut !Jl('lfC) with the topology inherited from the
quotient topology of OUs1('lfC)jT is a complete and separable metric group.
By a quantum flow in 'lfC we mean a family {r(s,t),O ::; s ::; t < oo} of
automorphisms of !Jl('lfC) satisfying the following.
(i) r( s, s) = 1 for all 0 ::; s < 00;
(ii) r( s, t )r( t, u) = r( s, u) for all 0 ::; s ::; t ::; u < 00;

(iii) limtlsr(s, t) = id
where id denotes the identity automorphism and convergence is in the topology
described in the preceding paragraph. The flow {r(s,t)} is said to be stationary
or translation invariant if r( s, t) = r( t - s) for all 0 ::; s ::; t < 00 where {r( t)}
is a one parameter family of automorphisms.
For any quantum flow {r( s, t)} it follows from Wigner's Theorem that there
exists a family {U(s, t)} of elements in the group OUs1('lfC) such that
r(s,t)(E) = U(s,t)*EU(s,t)
where
U(t, U)U(S, t) = a(s, t, U)U(S, u) for all s :S t :S u, (14.15)
a(s,t,u) being a scalar of modulus unity. If s > t define U(s,t) as U(t,S)-I.
Then (14.15) holds whenever s :S t :S u or s ~ t ~ u. Define V(s, t) =
U(I, t)U(I, s )-1 for all s, t > O. Then V(s, t) differs from U(s, t) by a scalar and
V(t, u)V(s, t) = V(s,u) for all s:S t:S u,
r(s,t)(E) = V(s,t)*EV(s,t).
When {r( s, t)} is a stationary flow so that r( s, t) = r(t - s) there exists a family
{Ut} of unitary operators such that
r(t)(E) = ut EUt for every E in ~(7fC)
and
UsUt = a(s, t)Us+t for all s, t ~ O.
It is a theorem of V. Bargmann [14 ] that there exists a function a(t) of modulus
unity such that
a(s, t) = a(s)a(t)a(s + t)-I.
Defining Vi = a(t)-IUt we have Vs Vi = Vs+t, yt* EVi = r(t)(E) for every E in
~(7fC). Defining V-t = yt* we obtain a one parameter group of unitary operators
inducing the flow {r(t)}. Using Stone's Theorem we can express Vi = e- itH
where H is a selfadjoint operator. Thus every stationary quantum flow is of the
form
r(t)(E) = eitH Ee- itH for all t E ~
where H is a selfadjoint operator called the Hamiltonian or energy operator of
the flow. This yields the Heisenberg picture of dynamics:
(14.16)
If we write Xo = X, X t = eitH Xe- itH we may express this as a differential
equation:
dt = 2'[H , Xlt·
dX t (14.17)
However, one must keep in view the domain problems concerning unbounded
operators in a Banach space.
If P is a state then the expectation of X t = r(t)(X) in the state p is given
by
tr pr(t)(X) = tre- itH p eitH X.
Putting Pt = e- itH p eitH we may express it as a differential equation in the
Liouville picture:
dpt = -2'[H ,Pt·1
dt (14.18)
14 Wigner's Theorem on the automorphisms of ?J'('iJt) 89
If P =
lu)(ul,u being a unit vector in 'Je, then Pt = IUt)(utl where Ut = e-itHu.
We may express this as a differential equation in the Schrodinger picture:
idut
dt = HUt· (14.19)
Equations (14.17)-(14.19) may be compared with (6.1)-(6.3).
Notes
Wigner's Theorem has its origin in Wigner [140]. The proof here is adapted from
Bargmann [14].
The Heisenberg, Liouville and Schrodinger pictures of quantum dynamics
have their origin in the foundations of quantum theory. Some of the principal ref-
erences are Dirac [29], von Neumann [137], Feynman [42], Mackey [84], Varadara-
jan [135], Wey1 [139].
Chapter II
Observables and States in Tensor
Products of Hilbert Spaces
15 Positive definite kernels and tensor

products of Hilbert spaces
Suppose (Oi, ~i)' 1 :S i :S n are sample spaces describing the elementary out-
comes and events concerning n different statistical systems in classical proba-
bility. To integrate them into a unified picture under the umbrella of a single
sample space one takes their cartesian product (O,~) where 0 = 0 1 X ... x On,
~ = ~I X ... X ~n, the smallest a-algebra containing all rectangles of the form
FI X F2 X ... x Fn , Fj E ~j for each j. Now we wish to search for an analogue of
this description in quantum probability when we have n systems where the events
concerning the j-th system are described by the set P('1Je j ) of all projections in
a Hilbert space '1Je j, j = 1, 2, ... ,n. Such an attempt leads us to consider tensor
products of Hilbert spaces. We shall present a somewhat statistically oriented ap-
proach to the definition of tensor products which is at the same time coordinate
free in character. To this end we introduce the notion of a positive definite kernel.
Let 2e be any set and let K : 2e x 2e --7 C be a map satisfying the following:
LQi(XjK(Xi,Xj) 2: 0
i,j
for all (Xi E C, Xi E 2e, i = 1,2, ... ,n. Such a map K is called a positive definite
kernel or simply a kernel on 2e. We denote by K(2e) the set of all such kernels on
2e.
If 2e = {I, 2, ... ,n}, a kernel on 2e is just a positive (semi) definite matrix.
If'1Je is a Hilbert space the scalar product K(x, y) = (x, y) is a kernel on '1Je. If G
is a group and g --7 Ug is a homomorphism from G into the unitary group OU('1Je)
of a Hilbert space '1Je then K (g, h) = (u, Ug-I h u) is a kernel on G for every u in
'1Je. If P is a state in '1Je then K(X, Y) = tr pX*Y is a kernel on f%('1Je).
Proposition 15.1: Let ((aij)),((b ij )),l :S i,j :S n be two positive definite
matrices. Then (( aij bij )) is positive definite.
Proof: Let A = (( aij ) ). Choose any matrix C of order n such that C C* = A. Let
(Xl, ... ,(Xn,f3 1 , ••• ,!3n be any 2n independent and identically distributed standard
Gaussian (normal) random variables. Write 'Yj = 2-~((Xj +i!3j), ~ = C'Y where 'Y
denotes the column vector with j-th entry 'Yj. Then ~ is a complex valued-Gaussian
random vector satisfying IE{ = Q, IE~C = CC* = A.
Using the procedure described above select a pair of independent complex
Gaussian random vectors {, '!l such that lE~j = lE1]j = 0, lE~i~j = aij, lEr;i1]j = bij
92 Chapter II: Observables and States in Tensor Products of Hilbert Spaces
for 1 ::; i, j ::; n. Let (j = ~jfJj. Then lE(j = 0, lE(i(j = aijbij . Thus (( aijbij))
is the covariance matrix of the complex random vector ( and hence positive
definite. - .
Corollary 15.2: The space K('l£) of all kernels on 'l£ is closed under pointwise
multiplication.
Proof: This follows immediately from Proposition 15.1 and the fact that K is
a kernel on 'l£ if and only if for any finite set {XI,X2, ... ,xn } C 'l£ the matrix
(( aij )) where aij = K (Xi, Xj) is positive definite. •
Corollary 15.3: Let 'l£i, 1 ::; i ::; n be sets and let Ki E K('l£i) for each i. Define
'l£ = 'l£I X ... x 'l£n and
K({f, J!J = rrf=I Ki(Xi, Yi), {f = (Xl, ... ,Xn ), J!.. = (Yl, ... ,Yn)
where xi, Yi E 'l£i. Then K E K('l£).
Proof: Let {f(r) = (Xlr, X2r, ... 'x nr ) E 'l£, 1 < r < m. Putting a~s
Ki(Xir, Xis) we observe that
K({f(r),{f(s)) = rrf=la~S.
Since ((a~s))' 1 ::; r, S ::; m is positive definite for each fixed i the required result
follows from Proposition 15.1. •
Proposition 15.4: Let 'l£ be any set and let K E K('l£). Then there exists a
(not necessarily separable) Hilbert space '7JC and a map). : 'l£ -+ 7Je satisfying the
following: (i) the set {A(X), X E 'l£} is total in '7JC; (ii) K(x,y) = ().(x),).(y)) for
all x, Y in 'l£.
If '7JC' is another Hilbert space and A' : 'l£ -+ '7JC' is another map satisfy-
ing (i) and (ii) with 7Je,). replaced by 7Je', A' respectively then there is a unitary
isomorphism U: '7JC -+ '7JC' such that U)'(x) = A'(x) for all X in 'l£.
Proof: For any finite set F = {x 1, X2, ... ,Xn } C 'l£ it follows from the argument
in the proof of Proposition 15.1 that there exists a complex Gaussian random
vector (({, (!, ... ,(;) such that
IE([ (J = K(Xi' Xj), IE([ = 0, 1::; i, j ::; n.
If G = {Xl, X2, ... ,Xn , Xn+I} =:J F then the marginal distribution of ((p, ... ,(if)
derived from that of ((p, ... , (if, (if+ 1) is the same as the distribution of ((
... , (;). Hence by Kolmogorov's Consistency Theorem there exists a Gaussian
r,
family {(x, X E 'l£} of complex valued random variables on a probability space
(O,?:F, P) such that
lE(x = 0, lE(x(y = K(x,y) for all x,y E 'l£.
If 7Je is the closed linear span of {(x,x E 'l£} in L 2 (P) and ).(x) = (x then the
first part of the proposition holds.
15 Positive definite kernels and tensor products of Hilbert spaces 93
To prove the second part consider 8 = {A (x) Ix E 2'£}, 8' = {A' (x) Ix E 2'£}
and the map U : A(x) -+ A' (x) from 8 onto 8'. Then U is scalar product
preserving and 8 and 8' are total in '7Ji and '7Ji' respectively. By the obvious
generalisation of Proposition 7.2 for not necessarily separable Hilbert spaces, U
extends uniquely to a unitary isomorphism from '7Ji onto '7Ji'. •
The pair ('7Ji, A) determined uniquely up to a unitary isomorphism by the
kernel K on 2'£ is called a Gelfand pair associated with K.
We are now ready to introduce the notion of tensor products of Hilbert
spaces using Proposition 15.4. Let '7Ji i , 1 S; i S; n be Hilbert spaces and let 2'£ =
'7Ji b X ... x '7Ji n be their cartesian product as a set. Then the function Ki (U, v) =
(u, v), U, v E '7Ji i is a kernel on '7Ji i for each i. By Corollary 15.3 the function
where ui, Vi E '7Ji i for each i, is a kernel on 2'£. Consider any Gelfand pair ('7Ji, A)
associated with K and satisfying (i) and (ii) of Proposition 15.4. Then '7Ji is called
a tensor product of '7Ji i , i = 1,2, ... , n. We write
n
'7Ji = '7Ji 1 ® '7Ji 2 ® ... ® '7Ji n = Q9'7Jii , (15.1)
i=1
n
A(~) = U1 ® U2 ® ... ® Un = Q9Ui (15.2)
i=1
and call A(~) the tensor product of the vectors Ui, 1 S; i S; n. If '7Ji i = h for all i
then '7Ji is called the n-fold tensor product of h and denoted by h ® n • If, in addition,
Ui = U for all i in (15.2) then A(~) is denoted by u®n and called the n-th power
of u. (Since ('7Ji, A) is determined uniquely upto a Hilbert space isomorphism we
take the liberty of calling '7Ji the tensor product of '7Ji i , 1 S; i S; n in (15.1».
Proposition 15.5: The map (UI, U2, ... , un) -+ Ul ® U2 ® ... ® Un from '7Ji 1 x
'7Ji2 x··· x '7Ji n into '7Ji 1 ®'7Ji2®·· ·®'7Jin defined by (15.1) and (15.2) is multilinear:
for all scalars ex, f3
(15.3)
Furthermore
n n
(Q9Ui, Q9Vi) = II~=I (Ui, Vi). (15.5)
i=1 i=l
Proof: Only (15.3) remains to be proved. Straightforward computation using

(15.4) and the sesquilinearity of scalar products show that for each fixed i
n
IluI @ ... @ Ui-I @ (aui + {JVi) @ Ui+1 @ .•. @ Un - aQ9Uj
j=1
-{JUI @ ... @ Ui-I @ Vi @ Ui+1 @ ... @ un l1 2
= II#iIIUjI12{llaui + {Jvi11 2 + lal211uil12 + If3I211vil12
-2 Re( (aui + {JVi' aUi) + (aUi + {JVi' {JVi) + (aUi' {JVi))}
=Q •
For any Ui E 'tIC i , 1 ::; i ::; n the product vector Q9~=1 Ui may be interpreted
as the multi-antilinear functional
n
(Q9Ui)(VI,V2, ... ,vn) = IIf=I(Vi,Ui).
i=1
Such muIti-antilinear functionals generate a linear manifold M to which the scalar
product (15.4) can be extended by sesquilinearity to make it a pre-Hilbert space.
M is the usual algebraic tensor product of the vector spaces 'tIC i , 1 ::; i ::; nand
Q9~=1 'tICi is its completion.
Exercise 15.6: (i) Let K be a kernel on 2t'. A bijective map g : 2t' ---+ 2t' is
said to leave K invariant if K(g(x),g(y)) = K(x,y) for all x,y in 2t'. Let G K
denote the group of all such bijective transformations of 2t' leaving K invariant
and let ('tIC, A) be a Gelfand pair associated with K. Then there exists a unique
homomorphism g ---+ Ug from G K into the unitary group OU('tIC) of 'tIC satisfying
the relation
UgA(X) = >.(g(x)) for all x E 2t',g E G K .
If ('tIC', A') is another Gelfand pair associated with K, V: 'tIC ---+ 'tIC' is the unitary
isomorphism satisfying V>.(x) = A'(x) for all x and g ---+ U; is the homomor-
phism from G K into U('tIC') satisfying U~A'(x) = A'(g(x)) for all x E 2t', g E G K
then VUgV-I = U~ for all g.
(ii) In (i) let 2t' be a separable metric space and let K be continuous on 2t' x 2t'.
Suppose Go c G K is a subgroup which is a topological group acting continuously
on 2t' x 2t'. Then in any Gelfand pair ('tIC, A) the map A : 2t' ---+ 'tIC is continuous, 'tIC
is separable and the homomorphism g ---+ Ug restricted to Go is continuous.
(Hint: Use Proposition 7.2 for (i) and examine 11>'(x) - A(y)11 for (ii».
Exercise 15.7: (i) Let 'tIC i , 1 ::; i ::; n be Hilbert spaces and let Si C 'tIC i be a total
subset for each i. Then the set {Q9~=1 Ui lUi E Si for each i} is total in Q9~=1 'tIC i .
'tIC is separable if each 'tICi is separable.
15 Positive definite kernels and tensor products of Hilbert spaces 95
(ii) There exist unitary isomorphisms

U12 ,3 : (7Je 1 Q9 7Je 2) Q9 7Je 3 --t 7Je 1 Q9 7Je2 Q9 7Je 3,
UI,23 : 7Je 1 Q9 (7Je 2 Q9 7Je 3) --t 7Je 1 Q9 7Je 2 Q9 7Je 3,
such that
U12,3(UI Q9 U2) Q9 U3 = UI,23UI Q9 (U2 Q9 U3) = UI Q9 U2 Q9 U3
for all i E 7Jei, i = 1,2,3. (Hint: Use Proposition 7.2.)
Exercise 15.8: Let {eij Ij = 1, 2, ... } be an orthonormal basis in 7Je i , i

1,2, ... ,n respectively. Then the set
{elj! Q9 e2iz Q9 .•. Q9 enjnljl = 1,2, ... ,12 = 1,2, ... ,jn = 1,2, ... }
is an orthonormal basis for Q9~=I7Jei' (Note that when dim 7Je i = mi < 00, ji =
1,2, ... ,mi). In particular,
n n
Q9 Ui = L {IIf=1 (eij" Ui)}Qg eiji
i=1 i=1
where the right hand side is a strongly convergent sum in Q9~=I7Jei' If dim 7Je i =
mi < 00 for every i then
n
dim Q97Je i = ml m2 ... m n .
i=1
Exercise 15.9: Let 7Je i = L2(Oi' ?J'i, J-li), 1 ::; i ::; n where (Oi, ?J'i, J-li) is a a-finite
measure space for each i. If (O,?J', J-l) = IIf=1 (Oi, ?J'i, J-li) is the cartesian product
of these measure spaces there exists a unitary isomorphism U : Q9~I7Jei --t
L2(0, '!F, J-l) such that
(UUI Q9'" Q9 Un)(WI, ... ,wn ) = II~lui(wi) a.e. J-l.
Exercise 15.10: Let 7Je n , n = 1,2, ... be a sequence of Hilbert spaces and let
{<Pn} be a sequence of unit vectors where <Pn E 7Je n for each n. Suppose
M={1t-.I1t-.=(UI,U2,"')' UjE7Je j , un=<Pn foralliargen}.
Define
K(1t-.,'Q) = II~I (Uj, Vj), 1t-.,'Q EM.
Then K is a kernel on M. The Hilbert space 7Je in a Gelfand pair (7Je, oX) associated
with K is called the countable tensor product of the sequence {7Je n } with respect
to the stabilizing sequence {<Pn}. We write
oX(1t-.) = UI Q9 U2 Q9 . . . for 1t-. E M.
Suppose {eno, enl, ... } = Sn is an orthonormal basis in 7Je n such that eno = <Pn
for each n. Then the set P(1t-.)I1t-. E M, Uj E Sj for each j} is an orthonormal
basis in 7Je.
Exercise 15.11: Let 'JC n = L 2(nn,?i'n,f-£n), n = 1,2, ... , 'JC = L 2(n,?i',f-£)

where (nn, ?i'n, f-£n) is a probability space for each n and (n,?i', f-£) = n~=1 (nn,
?i'n, f-£n) is the product probability space. Then '//( is the countable tensor product
of the sequence {'JC n } with respect to the stabilising sequence {<Pn} where <Pn is
the constant function 1 in nn for each n. This may be used to construct examples
where different stabilising sequences may lead to different countable tensor prod-
ucts. (Hint: If A and v are two distinct probability measures such that A == v then
A x A x ... and v x v x ... are singular with respect to each other.)
Exerecise 15.12: Let '//( = L2 ([0, 1], h) be the Hilbert space defined in Section 2
where f-£ is the Lebesgue measure in [0, 1]. Then any continuous map f : [0, 1] ---+ '//(
is an element of 'JC. For any such continuous map f define the element en(J) in
(C Efl h)®n by
Then
lim (en(J), en(g)) '= exp(J,g).
n-->oo
Exercise 15.13: Let '//(1, 'JC2 be Hilbert spaces and let {h, fz, . .. } be an or-
thonormal basis in '//(2. Then there exists a unitary isomorphism U : '//(1 I8l '//(2 ---+
'//(1 Efl '//(1 Efl .•. satisfying Uu I8l v = Efli (Ii. v)u for all u E '//(1, V E 'JC2.
Exercise 15.14: Let L2(f-£, h) be as defined in Section 2. Then there exists a unique
unitary isomorphism U : L2(f-£) I8l h ---+ L2(f-£, h) satisfying (U f I8l u)(w) = f(w)u.
Exercise 15.15: Let'//( be a Hilbert space and let I2('JC) be the Hilbert space of
all Hilbert-Schmidt operators in '//( with scalar product (T1' T2) = tr Tt T2. For any
fixed conjugation J there exists a unique unitary isomorphism U : '//(18l'//( ---+ I2('JC)
satisfying Uu I8l v = lu){Jvl for all u, v in '//(. (See Section 9.)
Example 15.16: [20] Let H be any selfadjoint operator in a Hilbert space 'JC
with pure point spectrum ~(H) = S. Then S is a finite or countable subset of
lit Denote by G the countable additive group generated by S and endowed with
the discrete topology. Let G be its compact character group with the normalised
Haar measure. For any bounded operator X on '//( and A E G define the bounded
operator
X).. = faX(A)X(H)XX(H)d X'
Let f-£, v E S, u, v E '//(, Hu = AU, Hv = vv. Then
(u,X)..v) = {kX(f-£ - v - A)dX}(u,Xv).
Hence
(u,X)..v ) = { (u, Xv) if A = f-£ - v,
° if A#- f-£ - v.
16 Operators in tensor products of Hilbert spaces 97
Thus, for any non-zero bounded operator X there exists a A E S - S

{M - viM, v E S} such that X A i= 0 and on the linear manifold D generated
by all the eigenvectors of H
[H, [H, X A)) = A2 X A,
Xu = Xou + L
>->0
(X-A + XA)u, u E D.
>-ES-S
If X is selfadjoint (XA)* = X-A and X is a "superposition" of bounded "har-

monic" observables X o, {X-A + XAIA E S - S, A > O} with respect to H. (See
Example 6.2.) Whenever X, Y are Hilbert-Schmidt operators
tr X*Y = tr X; Yo + L
>-ES-S
tr X~YA
>->0
and
X = Xo + L
>-ES-S
(XA + X-A)
>->0
converges in Hilbert-Schmidt norm (i.e., the norm in I2('X))' See Example 6.2.
Notes
The presentation here is based on the notes of Parthasarathy and Schmidt [99].
Proposition 15.1 is known as Schur's Lemma. Proposition 15.4 and Exercise 15.6
constitute what may be called a probability theorist's translation of the famous
Gelfand-Neumark-Segal or G.N.S. Theorem. Its origin may be traced back to the
theory of second order stationary stochastic processes developed by A.N. Kol-
mogorov, N. Wiener, A.1. Khinchine and K. Karhunen [31]. Proposition 7.2, 15.4
and 19.4 more or less constitute the basic principles around which the fabric of
our exposition in this volume is woven.
It is an interesting idea of Joume and Meyer [88] that (C EB h )'81 n in Exercise
15.12 may be looked upon as a toy Fock space where en(J) may be imagined as
a toy exponential or coherent vector which in the limit becomes the boson Fock
space r(L2[O, 1) QSI h) where e(J) is the true exponential or coherent vector. See
Exercise 29.12, 29.13, Parthasarathy [107], Lindsay and Parthasarathy [78].
16 Operators in tensor products of Hilbert spaces

We shall now define tensor products of operators. To begin with let 'iJC i be a finite
dimensional Hilbert space of dimension mi, i = 1,2, ... ,n and 'X = ®~=l 'iJCi .
Suppose Ti is a selfadjoint operator in 'iJC i with eigenvalues {Aij, 1 :::; j :::; mi}
and corresponding orthnormal set of eigenvectors {eij, 1 :::; j :::; mi} so that
Tieij = Aijeij, 1 :::; j :::; mi, i = 1,2, ... ,n.
Using Exercise 15.8 define a selfadjoint operator T on 'iJC by putting
n n
TQgeii> = II~l Aiji Qg eiji' 1:::; ji :::; mi
i=l i=l
and extending linearly on X. The operator T has eigenvalues rrf=l Aij; and satisfies
n n
TQ9Ui = Q9TiUi for all Ui E Xi, I ::; i ::; n.
i=l i=l
Furthermore
IITII = max (Irrf=l Aij; 1,1 ::; ji ::; mi)
= rr b1 max(IAijl, 1 ::; j ::; mi) = rrf=lIITill.
In particular, we have the identity
N n N n
I l : (lj a krr f=l (Uij, TiUik) I = I(l:ajQ9 Uij' Tl:ajQ9 Uij) I
l:;j,k:;N j=l i=l j=l i=l
N n
::; rrf=11I Tilllll: a jQ9 UijI12 (16.1)
j=l i=l
for all scalars aj, Uij E Xi, 1 ::; j ::; N, 1 ::; i ::; n, N = 1,2, .... We write
T = ®7=1 Ti = Tl (>9 ••• (>9 Tn and call it the tensor product of operators Ti,
1 ::; i ::; n. The next proposition extends this elementary notion to all bounded
operators on Hilbert spaces.
Proposition 16.1: Let Xi, 1 ::; i ::; n be Hilbert spaces and let Ti be a bounded
operator in Xi for each i. Then there exists a unique bounded operator T in
X = ®7=1 Xi satisfying
n n
TQ9Ui = Q9TiUi for all Ui E 'Je i , 1::; i ::; n. ( 16.2)
i=l i=l
Furthermore IITII = rrill T&
Proof: Let S = {®7=lUilui E 'Je i , 1 ::; i ::; n}. For all scalars aj, 1 ::; j ::; N
and product vectors ®7=1 Uij E S, 1 ::; j ::; N we have
N n
lIl:ajQ9Ti UijW = l : aj a k rr f=l(Uij,Tt T i Uik)
j=l i=l 15j,k:;N
l : aj akrrf=1 (Uij, PiTtTiPiUik) (16.3)

l:;j,k<;N
where Pi is the projection on the finite dimensional subspace Mi spanned by
{Uij, 1 ::; j ::; N} in Xi for each i. Since PiTtTiPi is a positive operator in Mi,
it follows from (16.1) and (16.3) that
N n N n
IIl:ajQ9TiUijI12 ::; rrf=11IPiTtTiPilllll:ajQ9uijI12
j=l i=l j=l i=l
N n
::; rrf=11I Tt Tilllll: a jQ9 UijI12. (16.4)
j=l i=l
Define T on the set S by (16.2). Then (16.4) shows that T extends uniquely as a
linear map on the linear manifold generated by S. Since S is total in 'Je this linear
extension can be closed to define a bounded operator T on 'Je satisfying
IITII ~ IIi=IIITtTilll/2 = IIi=dlTiI!-

Let now 0 < c < 1 be arbitrary. Choose unit vectors Ui E 'Jei such that IITiUil1 ;:::
(1- c)IITill for each i, assuming Ti i- O. Then Q9~=IUi is a unit vector in 'Je and
n n
IITQ9Uill = IIQ9TiUill = IIi=IIITiuill ;::: (1- c)nIIi=IIITiI!-
i=1 i=l
Thus IITII ;::: II~IIIT& •
The operator T determined by Proposition 16.1 is called the tensor product
of the operators Ti, 1 ~ i ~ n. We write T = Q9~=1 Ti = TI Q9 ..• Q9Tn . If 'Jei = h,
Ti = S for all i = 1,2, ... , n we write T = S®n and call it the n-th tensor power
of the operator S.
Proposition 16.2: Let 'Jei , 1 ~ i ~ n be Hilbert spaces and let Si, Ti be bounded
operators in 'Jei for each i. Let S = Q9~=1 Si, T = Q9~=1 T i . Then the following
relations hold:
(i) The mapping (TI' T2, .. . ,Tn) -+ T is multilinear from 8('JeJ) x··· x8('Jen )
into 8('Je 1 Q9 .•. Q9 'Je n );
(ii) ST = Q9~=1 SiTi, T* = Q9~=1 Tt;
(iii) If each Ti has a bounded inverse then T has a bounded inverse and T~I =
T ~l
Q9i i ;
(iv) T is a selfadjoint, unitary, normal or projection operator according to whether

each Ti is a selfadjoint, unitary, normal or projection operator;
(v) T is positive if each Ti is positive;
(vi) If Ti = lUi) (Vi I where Ui, Vi E 'Jei for each i then
T = lUI Q9 U2 Q9 ... Q9 Un)(VI Q9 V2 Q9'" Q9 vnl·
Proof: This is straightforward from definitions and we omit the proof. •
Proposition 16.3: Let Ti be a compact operator in 'Je i , i = 1,2, ... ,n, 'Je =
Q9~=1 'Je i , T = Q9~=1 Ti· If Ti has the canonical decomposition in the sense of
Proposition 9.6
Ti =L Sj(Ti)lvij)(Uijl, i = 1,2, ... ,n (16.5)

j
then T is a compact operator with canonical decomposition
T= L sj\(TJ)",sjJTn )I Vlj\Q9,,·Q9V njJ(ulj\Q9,,·Q9u njnl· (16.6)

If Ti E II (~i) for each i then T E II (~) and
IITlh = IIf=IIITillI. trT = IIf=1 trTi·

In particular, if each Ti is a state so is T.
Proof: Since {Uij, j = 1, 2, ... }, {Vij, j = 1,2, ... } are orthononnal sets in ~i for
each i it follows that {Ulit Q9 . .. Q9 Unjn}' {Vljl Q9 . .. Q9 Vnjn} are orthononnal sets in
~. Equation (16.5), the second part of Proposition 16.1 and (vi) in Proposition 16.2
yield (16.6) as an operator nonn convergent sum over the indices (h,h, ... ,jn)'
If IITill1 = Ejlsj(Ti)1 < 00 for each i then we have
IITlh = L ISjl (TI)··· Sjn (Tn)1 = IIf=IIITilh·

jl.h ... · .jn
If {eij, j = 1, 2, ... } is an orthononnal basis in 7Je i for each i it follows that

{eljl Q9 ... Q9 enjn}
is an orthononnal basis in 7Je and
trT = L (eljl Q9 ... Q9 enjn' TI Q9 ... Q9 Tneljl Q9 ... Q9 enjn)
it.··· ,jn
= L IIf=1 (eiji' Tieiji)
jl,'" ,jn
= IIf=IL(eij.Tieij)
j
= IIf=1 trTi'
due to the absolute convergence of all the sums involved. Finally, if each Ti is
positive so is T. If tr Ti = 1 for each i then tr T = 1. •
If (7Je i , P(~i)' pd is a quantum probability space for each i = 1,2, ... ,n
then putting ~ = ®~=I7Jei' P = ®~=IPi we obtain a new quantum probability
space (~, P(~), p) called the product of the quantum probability spaces
(7Jei, P(~i)' Pi)' i = 1,2, ... ,n.
If Xi is a bounded selfadjoint operator in ~i or, equivalently, a bounded real
valued observable in ~i then X = X I Q9 ... Q9 Xn is a bounded real valued
observable in ~ and the expectation of X in the product state P is equal to
IIf=1 tr PiXi, the product of the expectations of Xi in the state Pi, i = 1,2, ... ,n.
Now the stage is set for achieving our goal of combining the description of
events concerning several statistical experiments into those of a single experiment.
Suppose that the events of the i-th experiment are described by the elements of
P(7Jed where ~i is a Hilbert space, i = 1,2, ... ,n. Let 7Je = Q9i7Jei. If Pi E P(7Je i )
then we view the event Pi as the element
Pi = 1 Q9 ... Q9 1 Q9 Pi Q9 ... Q9 1,
(called the ampliation of Pi in 'M.) in ~('M.) and interpret P = ®~=l Pi = nr=l Pi E

~('M.) (see Proposition 16.2) as the event signifying the simultaneous occurrence
of Pi in the i-th experiment for i = 1, 2, ... ,n. These are the quantum probabilis-
tic equivalents of measurable rectangles in products of sample spaces. The next
proposition indicates how ~('M.) is "generated" by P('M. i ), i = 1,2, ... ,n.
Proposition 16.4: Every element P in ~('M.) can be obtained as a strong limit
of linear combinations of projections of the form ®~=lPi' Pi E P('M. i ).
Proof: Choose an increasing sequence {Q ik, k = 1, 2, ... } of finite dimen-
sional projections so that s.limk-+oo Qik = 1 in 'M. i for each i. Then Qk =
®~=l Qik increases to the identity strongly in 'M. and for any bounded opera-
tor T in ~,s.limK -+00 QkTQk = T. Thus it suffices to show that for fixed k the
operator QkTQk can be expressed as a linear combination of product projections
of the form ®~=l Pi, Pi E P('M.i). Replacing 'M. i by the range of Qik if necessary
we may assume that each 'M. i has dimension mi < 00. Let {eir, 1 ::; r ::; mi} be
an orthonormal basis in 'M. i and let E~s = !eis)(eir!. Then every operator Ton
'M. can be expressed as a linear combination of operators of the form ®~=lEtjsj"
Each Ets can be expressed as a linear combination of selfadjoint operators:
j _- "21 (Ers
Ers j + E rsj*).{
+ Z 2i1 (Ers
j j*)} .
- Ers
Since every selfadjoint operator in 'M. i admits a spectral decomposition in terms
of eigen projections the proof is complete. •
Proposition 16.5: Let (Oi, ~d be measurable spaces and let ~i : ~i --. P(1ti)
be an Oi-valued observable in 'M. i for i = 1,2, ... ,n. Let (O,~) = nr=l (Oi, ~i)
be the product measurable space and 'M. = ®~=l 'M. i . Then there exists a unique
O-valued observable ~ : ~ --. ~('M.) such that
n
~(Fl x··· x Fn) = @ ~i(Fd for all Fi E ~i' i = 1,2, ... ,no (16.7)
i=l
Proof: By Proposition 7.4 there exists a finite or countable family of a-finite
measures {J.Lij, j = 1,2, ... , } and a unitary isomorphism Ui : 'M. i --. EBjL2(J.Lij)
such that
UiME)Ui- 1 = EB eij(E) for all E E ~i, 1::; i::; n.
j
We have
n
@{EB j L 2 (J.Lij)} = EB L 2 (J.Lljl x J.L2jz X ... X J.LnjJ.
i=l
Define U = U1 Q9 ..• Q9 Un and
~(F) = U- 1{ EB e1h X···X/l.njn (F)}U,F E~.
Then (16.7) is obtained.

•
The unique observable ~ satisfying (16.7) in Proposition 16.5 is called the

tensor product of the observables ~i, i = 1,2, ... , n. We write ~ = 6 ® ... ® ~n'
Proposition 16.6: Let 'Je l , 'Je2 be Hilbert spaces and let T be a trace class operator
= 'Je l ® 'Je 2 • Then there exists a unique trace class operator TI in 'Je l such
in 'Je
that
trTIX = trT(X ® 1) for all X in B('Jed. (16.8)
If T is a state in 'Je then TI is a state in 'Je I.
Proof: For any compact operator X in 'Je l define >'(X) = trT(X ® 1). By
Proposition 9.12, (iii) we have I>'(X)I ::; IITIIIIIXII. In particular, >. is a continuous
linear functional on Ioo ('Jed. By Schatten's Theorem (Theorem 9.17) there exists
a TI in II ('Jed such that trTIX = trT(X ® 1) for all X in Ioo('Je I ). Since the
maps X ----+ trTIX and X ----+ trT(X ® 1) are strongly continuous in B('Jed we
obtain (16.8). If T is positive we have
(u,Tlu) = trTllu)(ul = trT(lu)(ul ® 1) ~ 0
for all u in 'Je l . Thus TI ~ O. Putting X = 1 in (16.8) we get trTI = trT. This
proves that TI is a state if T is a state. •
The operator TI in Proposition 16.6 is called the relative trace of Tin 'Je l .
If T is a state then the relative trace of TI is the analogue of marginal distribution
in classical probability.
Exercise 16.7: Suppose 'JeJ, 'Je2 are two real finite dimensional Hilbert spaces
of dimensions ml, m2 respectively and 'Je = 'Je l ® 'Je 2 . Let O('Jed,O('Je2 ) and
o('Je I ® 'J(2 ) be the real linear spaces of all selfadjoint operators in 'Je I, 'Je2 and
'Je l ® 'Je2 respectively. Then we have dim O('Jei ) = !mi( mi + 1), i = 1,2 and
dimO('Je 1 ® 'J( 2 ) = !mlm2(mlm2 + 1). In particular,
dim O('Je 1 ® 'J( 2 ) > dim O('Jed . dim O('Je2 ) if mi > 1, i = 1,2,.
On the other hand, if 'Je l , 'Je2 are complex Hilbert spaces dim O('Je i ) = m; and
dimO('Je 1 ® 'J( 2 ) = dimO('JeddimO('Je2 ).
(In the light of Proposition 16.4 this indicates the advantage of working with
complex Hilbert spaces in dealing with observables concerning several quantum
statistical experiments).
Exercise 16.8: (i) Let Tj be (a not necessarily bounded) selfadjoint operator in

'Jej with spectral representation
Tj= ~X~j(dX)' j=1,2, ... ,n,
~j being a real valued observable in 'Jej for each j. Define the selfadjoint operator
TI ®···®Tn = r XIX2",xn6 ®6®···®~n(dxldx2···dxn)

J~n
by Proposition 12.1 and Theorem 12.2. If Dj = D(Tj) is a core for Tj for each
j then the linear manifold M generated by {Uj 181 ... 181 Un IUj E Dj for each j}
is a core for Tj 181 ... 181 Tn and
for all Uj E D(Tj), 1 ~ j ~ n.

(ii) Let Ti = 1181· .. 1811 Q9Ti 1811· .. 11811 (be the i-th ampliation OfTi) where
Ti is in the i-th position. Then TI +... +Tn is essentially selfadjoint on M with its
closure being the selfadjoint operator JlRn (XI + ... +x n )6 181· .. Q9~n(dXI ... dXn).
(iii) If Pj is a state in 7Jej and Tj has finite expectation in the state Pj for each
j then TI 181 ... 181 Tn has finite expectation in the product state P = PI 181 ... 181 Pn
and
in the notation of Proposition 13.6.
Exercise 16.9: Let T E II(7Je 1 181 7Je2 ) and let TI be its relative trace in 7Je 1. If
{li} is an orthonormal basis in 7Je 2 then
(U, Tlv) = L (u Q9li, Tv Q9li) for all u, v in 7Je 1

j
where the right hand side converges absolutely.
Exercise 16.10: Let 7Jej, 7Je 2 be Hilbert spaces and 7Je = 7Je 1 181 7Je2 . Then the
*-algebra generated by {XI 181 X2 1Xi E B(7Jei ),i = 1,2} is strongly dense in
9A(~).
(ii) For any trace class operator P in 7Je2 there exists a unique linear map
IEp : 9A(~) -+ B(7JeJ) satisfying
(u, IEp(X)v) = tr X(lv)(ul 181 p) for all u, v E 7Jej,X E 9A(7Je)

where
(iii) If P is a state then IEp is called the p-conditional expectation map from
9A(7Je) into B(7Je I ). The P conditional expectation map satisfies the following prop-
erties:
(1) IEpl = 1, IEpX* = (lEpX)*, IllEpXl1 ~ IIXII;
(2) IEp(A Q91)X(B 1811) = A(lEpX)B, for all A,B E B(7JeJ), X E 9A(7Je);
(3) I:1::;i,j::;kYi*(lEpXtXj)}j ~ 0 for all Xi E 9A(7Je), Yi E B(7JeJ). In particu-
lar, IEpX ~ 0 whenever X ~ O.
Exercise 16.11: Let {~n, n ~ O} be a sequence of Hilbert spaces and {¢n, n ~ I}

be unit vectors, ¢n E ~n. Define ;}e[n+l = ~n+l ® ~n+2® .. , with respect to
the stabilising sequence ¢n+l, ¢n+2, ... and ;}en) = ;}eo ® ~1 ® '" ® ~n. In the
Hilbert space
~ = ~o ® ~[1 = ;}en] ® ~[n+J, n = 1,2, ...

consider the increasing sequence of *-algebras
Bn] = {X ® I[n+lIX E B(~n])}' n = 0,1,2 ...

Define Bn = B(~n)) and In) to be the identity in Bn. There exists a unique linear
map [En) : Boo - t Bn) satisfying
(u, [En) (X)v) = (u ® ¢[n+J,Xv ® ¢[n+l) for all u, v E ~n)' X E Boo
where ¢[n+l = ¢n+l ® ¢n+2 ® ... and Boo = B('ie). Indeed,
[En] (X) = [ElcP[n+l)(cP[n+ll(X) ® I[n+l'

The maps {[En) }n:;::o satisfy the following properties:
(i) [En) 1 = 1, [En]X* = ([En)X)*, II[En]XII ~ IIXII;
(ii) [En) AX B = A[En) (X)B whenever A, B E Bn);
(iii) [Em) [En] = [En) [Em] = [Em] whenever m ~ n;
(iv) Ll::;i,j9 Yi *[En) (X; Xj)Yj ~ 0 for all Yi E Bn), Xi E Boo. In particular
[En)X ~ 0 whenever X ~ 0;
(v) s.limn->oo IEn)X = X for all X in Boo.
Exercise 16.12: (i) In the notations of Exercise 16.11 a sequence {Xn} in Boo
is said to be adapted if Xn E Bn] for every n. It is called a martingale if
[En-I]Xn = X n- 1 for all n ~ 1

Suppose A = {An}n:;::1 is any sequence of operators where
An E B(~n), (¢n, An¢n) = 0, n = 1,2, ...

Define
An = In-I] ® An ® I[n+J,
0 if n = 0,
MnCA) = { Al + A2 + ... + An if n ~ 1.
Then {Mn (A)}n>o is a martingale. For any two sequences A, B of operators
where An, Bn E B(~n)' (¢n, An¢n) = (¢n, Bn¢n) = 0 for each n
[Em]{Mn(A) - Mm(A)}*{Mn(B) - Mm(B)}
n
= l:= (Aj¢j,Bj¢j) for all n>m~O.
j=m+l
17 Symmetric and antisymmetric tensor products 105
(ii) For any sequence E = (EJ,~, ... ) of operators such that

En E Bn-Ij, n = 1,2, ...
define
if n = 0,
In (A, E) = {~~ En{Mn(A) - Mn-I(A)} if n ;::: 1.
Then {In (A, E)}n>O is a martingale. Furthermore

IE n-1]In (A, E)* In(B, F) = In-I (A, E)* In-I (B, F)
+ (An¢n, Bn¢n)E~Fn
for all n ;::: 1.
Exercise 16.13: For any selfadjoint operator X in the Hilbert space 'JC define
the operator S(X) in (:2 ® 'Je by S(X) = (0'1 ® l)exp[ia2 ® X] where aj,
1 ~ j ~ 3 are the Pauli spin matrices. Then S(X) = S(X)-I = S(X)* and
HS(X) + S( -X)] = 0'1 ® cos X. Thus S(X) and S( -X) are spin observables
with two-point spectrum {-I, I} but their average can have arbitrary spectrum in
the interval [-1,1]. (See also Exercise 4.4,13.11.)
Notes
The role of tensor products of Hilbert spaces and operators in the construction of
observables concerning mUltiple quantum systems is explained in Mackey [84].
For a discussion of conditional expectation in non-commutative probability theory,
see Accardi and Cecchini [4]. Exercise 16.13 arose from discussions with B.V. R.
Bhat.
17 Symmetric and antisymmetric tensor products

There is a special feature of quantum mechanics which necessitates the introduction
of symmetric and anti symmetric tensor products of Hilbert spaces. Suppose that
a physical system consists of n identical particles which are indistinguishable
from one another. A transition may occur in the system resulting in merely the
interchange of particles regarding some physical characteristic (like position for
example) and it may not be possible to detect such a change by any observable
means. Suppose the statistical features of the dynamics of each particle in isolation
are described by states in some Hilbert space 'Je. According to the procedure
outlined in Section 15, 16 the events concerning all the n particles are described
by the elements of <!P('Je 0n ). If Pi E <!P('Je), 1 ~ i ~ n then ®~=I Pi signifies the
event that Pi occurs for each i. If the particles i and j (i < j) are interchanged and
a change cannot be detected then we should not distinguish between the events
PI ®.. '®Pn and PI ® .. ,®Pi-I ®Pj®PHI ®.. '®Pj-I ®Pi®Pj+1 ® .. ·®Pn , where
in the second product the positions of Pi and Pj are interchanged. This suggests
that the Hilbert space 'JC0 n is too large and therefore admits too many projections
and does not qualify for the description of events concerning n identical particles.
In order to take into account this curtailment in the degrees of freedom, the desired
reduction of the Hilbert space may be achieved by restriction to a suitable subspace
of Je0 n which is "invariant under permutations". We shall make this statement
more precise in the sequel.
Let Sn denote the group of all permutations of the set {I, 2, ... ,n}. Thus
any a E Sn is a one-to-one map of {I, 2, ... ,n} onto itself. For each a E Sn let
U(T be defined on the product vectors in Je0 n by
(17.1)
where a-I is the inverse of a. Then U(T is a scalar product preserving the map
of the total set of product vectors in Je0 n onto itself. Hence by Proposition 7.2,
U(T extends uniquely to a unitary operator on Je0 n, which we shall denote by U(T
itself. Clearly
(17.2)
Thus a ----t U(T is a homomorphism from the finite group Sn into OU(Je0 n ). The
closed subspaces
Je®n = {u E Je0 n IU(Tu = U for all a E Sn}, (17.3)
Je@n = {u E Je 0n 1U(Tu = €(a)u for all a E Sn}, (17.4)
where €(a) = ±1 according to whether the permutation a is even or odd are

called respectively the n-fold symmetric and antisymmetric tensor products of Je.
They are left invariant by the unitary operators U(T, a E Sn. There do exist other
such permutation invariant subspaces of Je0 n but it seems that they do not feature
frequently in a significant form in the physical description of n identical particles.
If the statistical features of the dynamics of a single particle are described by states
on rz;(Je) and the dynamics of n such identical particles is described by states in
rz;(Je®n ) for n = 2,3, ... then such a particle is called boson. Instead, if it is
described by states on rz;(Je@n ) for every n then such a particle is called fermion.
(This nomenclature is in honour of the physicists S.N. Bose and E. Fermi who
pioneered the investigation of statistics of such particles).
We shall now present some of the basic properties of Je®n and Je@n in the
next few propositions.
Proposition 17.1: Let E and F be operators in Je0 n defined by
(17.5)
(17.6)
where Uu is the unitary operator satisfying (17.1) and e (a) is the signature of the
pennutation a. Then E and F are projections onto the subspaces 'tIe®n and '/1f,@n
respectively. If K C Sn is any subgroup
1
E = E #K LUu, (17.7)
uEK
1
F = F #K Le(a)Uu (17.8)
uEK
where #K is the cardinality of K. Furthennore, for any Ui, Vi E 'tie, 1 ~ i ~ n
(EUl 181 ••• 181 Un, EVl 181 ••• 181 Vn ) = (Ul 181 •.• 181 Un, EVl 181 ••• 181 Vn )
= --; L IIf=l (Ui' VU(i))' (17.9)

n. uES n
(Full8I"'18I Un,FVll8I"'18I Vn ) = (Ull8I"'18I un,Fvll8I"'18I Vn )
= --;
n. uESn
L e(a)IIf=l(Ui,Vu(i))
1
= ,det(((ui,Vj))). (17.10)
n.
Proof: From (17.2) we have
E * = -1
n!
L Uu* = -1
n!
L UU-l = E,
.
uESn uESn
EUu = UuE = E2, (17.11)
F* = --; L
n. uESn
e(a)U; = --; L
n. uESn
e(a- l )UU-l = F,
FUu = UuF = e(a)F,F2 = F.

Thus E and F are projections. Furthennore (17.3), (17.4) and (17.11) imply that
Eu = U if and only if U E '/1f,®n and Fv = V if and only if v E '/1f,@n . This
proves the first part. Summing up over a E K and dividing by #K in the second
equation of (17.11) we obtain (17.7). Multiplying by e(a), summing over a E K
and dividing by #K in the last equation of (17.11) we obtain (17.8).
The first part of (17.9) is a consequence of the fact that E is a projection.
By definition
(Ul 181 •.• 181 Un, EVl 181 ••• 181 Vn ) =,

1
L
n. uESn
(Ul 181 ••• 181 Un, Vu(l) 181 ••• 181 Vu(n))
=,
I
L IIi(Ui,Vu(i))'
n. uES n
This proves (17.9). (17.10) follows exactly along the same lines.
•
It may be mentioned here that the quantity I:O"ESn IIf=1 (Ui, VO"(i)) occurring
on the right hand side of (17.9) is the permanent of the matrix (((Ui,Vj))) in
contrast to the determinant occurring in (17 .10).
Corollary 17.2: Let En and Fn denote respectively the projections E and F

defined on 'J(0 n by (17.5) and (17.6) for each n = 1, 2, .... Then for U E 'J(0 l ,
v E 'J(0m, W E 'J(0 n
E Hm +n {EHm (U (>9 v) (>9 w} = E Hm +n {U (>9 Em+n ( v (>9 w) }

= EHm+n(u (>9 v (>9 w),
FHm+n{FHm(U (>9 v) (>9 w} = FHm+n{U (>9 Fm+n(v (>9 w)}
= FHm+n(U (>9 v (>9 w).
Proof: This is immediate from (17.7) and (17.8) if we consider the permutation
groups SHm and Sm+n as subgroups of SHm+n so that a E SHm leaves the
last n elements of {I, 2, ... , £ + m + n} fixed whereas a E Sm+n leaves the first
£ elements of {I, 2, ... , £ + m + n} fixed. •
Proposition 17.3: Let {ei, i = 1,2, ... } be an orthonormal basis for 'J(. For any
Ui E 'J(, i = 1,2, ... , k and positive integers Tl, T2, ... , Tk satisfying Tl + ... +
k 0 ' r
Tk = n denote by Q9i=IU i 'the element UI(>9'" (>9UI (>9U2 (>9 ••. (>9U2 (>9 ••• (>9 Uk (>9
where Ui is repeated Ti times for each i. Let E and F be the projections
••• (>9 Uk
defined respectively by (17.5) and (17.6) in 'J(0 n . Then the sets
Tj ~ 1 for each 1 :S j :S k, Tl + T2 + ... + Tk = n, k = 1,2, ... , n},
{(n!)1/2 F ei 1 (>9 ei2 (>9 ••• (>9 ei n Iii < i2 < ... < in}
are orthonormal bases in 'J(®n and 'J(@n respectively. In particular, if dim'J( =
N < 00 then
dim'J(®n = (N+:-l).
dim 'J(@n= {o(~) if n :S N,
otherwise.
Proof: The first part follows from (17.9) and (17.10). The formula for dim 'J(®n
is immediate if we identify it as the number of ways in which n indistinguishable
balls can be thrown in N cells. Similarly the dimension of 'J(@n can be identified
with the number of ways in which n indistinguishable balls can be thrown in N
cells so that no cell has more than one ball. •
Using Proposition 17.3 it is possible to compare the different probability

distributions that arise in the "statistics of occupancy". More precisely, let p be a
state in 'Je, with dim'Je, = N < 00. Let {ej, j = 1,2, ... ,N} be an orthonormal
basis of 'Je such that p = LjPjlej}(ejl. Let lej}(ejl signify the event "particle
occupies cell number j". Now consider n distinguishable particles whose statis-
tics are described by the quantum probability space ('Je,®n, CfP('Je,®n), p®n). The
projection lej] ® ... ® ejn }(ej] ® ... ® ejJ signifies the event "particle i occupies
cell ji for each i = 1,2, ... ,n". Let r = (rj, ... ,rN) where rj is the number of
particles in cell j so that rJ + ... + rN = n. Define
(17.12)
where the summation on the right hand side is over all (j\, ... ,jn) such that the
cardinality of {ilji = j} is rj for j = 1,2, ... ,N. E~ is a projection whose range
has dimension Tl!'~~N! and it signifies the event th;t cell j has rj particles for
each j. Then
(17.13)
In this case we say that the particles obey the Maxwell-Boltzmann statistics and
the probability that there are rj particles in cell j for each j in the state p®n is
given by (17.13).
Suppose that the n particles under consideration are n identical bosons.
Then the Hilbert space 'Je,®n is replaced by 'Je,®n and correspondingly p®n by its
restricton to 'Je,®n . To make this restriction a state we put
and observe that c- J p®n is a state, E being defined by (17.5). The quantum
probability space ('Je,®n ,CfP('Je,®n ), c- J p®n ) describes the statistics of n identical
bosons. Denote by Eib) the projection on the one dimensional subspace generated
by the vector
N
e,- = E QS)e®Tj, r = (rJ,'" ,rN), rl + ... + rN = n.
j=J
Then the probability of finding rj particles in cell j for each j = 1,2, ... ,N is
T] TN
trc-Ip®n E(b) = PI ... PN (17.14)
'- '"
L..Js]+",sN=nPIS] ... PN
SN
In this case we say that the particles obey the Bose-Einstein statistics.
When the particles are n identical fermions 'Je,®n is replaced by 'Je,@n which
is non-trivial if and only if n :::; N, i.e., the number of particles does not exceed
the number of cells. The state p@n is replaced by its restriction to 'Je@n . Once
again to make this restriction a state we have to divide it by
Then the quantum probability space describing the statistics of n identical fermions
is ('Je@n ,Cj>('Je@n ), C,-l p@n ) where
If E£f) denotes the one dimensional projection on the subspace generated by the
vector
where r. = (f]' ... ,fN), fij = 1, j = 1,2, ... ,n and the remaining fi are 0 and
F is defined by (17.6) then the probability that cell i j is occupied by one particle
for each j = 1,2, ... ,n and the remaining cells are unoccupied is given by
(17.15)
In this case we say that the particles obey the Fermi-Dirac statistics.
For a comparison of the three distributions (17.13)-(17.15) consider the case
of two cells and n particles. Let the state p of a single particle be given by
where {e], e2} is an orthonormal basis in 'Je. According to Maxwell-Boltzmann

statistics the number of particles in cell I has a binomial distribution given by
Pr (cellI has kparticles) = (~)2-n, 0:::; k:::; n

whereas Bose-Einstein statistics yield
. 1
Pr (cell I has k partIcles) = - - , 0:::; k :::; n.
n+l
According to the first distribution the probability that all the particles occupy
a particular cell is 2 -n whereas the second distribution assigns the enhanced
probability n~l to the same event.
In the case of fermions it is impossible to have more than two of them when
there are only two cells available and if a cell is occupied by one particle the
second one has to be occupied by the other. Bosons tend to crowd more than
particles obeying Maxwell-Boltzmann statistics and fermions tend to avoid each
other.
18 Examples of discrete time quantum stochastic flows 111
Exercise 17.4: Let 7r be an irreducible unitary representation of Sn and let

X( (T) = tf7r( (T) be its character. Suppose X( 1) = d x denotes the dimension of the
representation 7r. Define
where U~ is the unitary operator satisfying (17.1). Then Ex is a projection for

each X, ExU~ = U~Ex for all (T E Sn, EXl EX2 = 0 if Xl #- X2 and LxEx = 1.
(Hint: Use Schur orthogonality relations).
Example 17.5: [74] (i) The volume of the region

~ = {(XloX2, ... ,xN-dlxj ~ ofor allj,O:::; XI + ... +XN-I:::; I}
in IRN-I is [(N - 1)!]-1.
(ii) Let rl, ... ,rN be non-negative integers such that rl + ... + rN = n.
Then
1 n! r
----PI l
~rl!···rN!
•..
r ()
PN' N - 1 !dp l dp2··· dPN-I = (N + n -
n
1)-1
where PN = (1 - PI - P2 - ... - PN-l).

This identity has the following interpretation. Suppose all the probability
distributions (Plo P2, ... ,PN) for the occupancy of the cells (1,2, ... ,N) by a
particle are equally likely and for any chosen prior distribution (PI, P2, ... ,PN ) the
distribution (17.4) with Pj = *',

particle obeys Maxwell-Boltzmann statistics. Then one obtains the Bose-Einstein
j = 1,2, ... ,N.
Notes
Regarding the role of symmetric and antisymmetric tensor products of Hilbert
spaces in the statistics of indistinguishable particles, see Dirac [29]. For an inter-
esting historical account of indistinguishable particles and Bose-Einstein statistics,
see Bach [11]. Example I7.Slinking Bose-Einstein and Maxwell-Boltzmann statis-
tics in the context of Bayesian inference is from Kunte [74].
18 Examples of discrete time quantum stochastic flows

Using the notion of a countable tensor product of a sequence of Hilbert spaces
with respect to a stabilising sequence of unit vectors and properties of conditional
expectation (see Exercise 16.10, 16.11) we shall now outline an elementary pro-
cedure of constructing a "quantum stochastic flow" in discrete time which is an
analogue of a classical Markov chain induced by a transition probability matrix.
For a Hilbert space 'J£ any subalgebra 0:1 C 0:I('J£) which is closed under the
involution * and weak topology is called a W* algebra or a von Neumann algebra.
If 'J£i, i = 1,2 are Hilbert spaces, 0:l i C 0:I('J£i), i = 1,2 are von Neumann algebras
denote by 0:1 1 tg) 0:1 2 the smallest von Neumann algebra containing {XI tg) X21Xi E
013 i , i1,2} in 013('Xl 0 '( 2 ). If p is a trace class operator in 'X 2 , following

Exercise 16.10, define the operator IEp(Z), Z E 013('Xl 0 '( 2 ) by the relation
(u, IEp(Z)v) = tr Z Iv) (ul 0 p, u, v E 'Xl (18.1 )
Proposition 18.1: IEp( Z) E 013 1 if Z E 013 1 0013 2 .
Proof: If Z = Xl 0 X2 then (18.1) implies that IEp(Z) = (tr pX2)Xl. Thus the
proposition holds for any finite linear combination of product operators in 013 1 0013 2 •
Suppose that p = I;pjlej)(ejl is a state where Pj > O,I;pj = 1 and {ej} is an
orthonormal set and w.limn--+oo Zn = Z in 013('Xl 0 '( 2 ). Then by (18.1)
lim (u,lEp(Zn)v) = lim ""pj(u0ej,ZnV0ej)

n----+oo n----+oo ~
j
= I;pj(u 0 ej, Zv 0 ej)

= (u, IEp(Z)v) for all u, v E 'Xl.
In other words IEp is weakly continuous if p is a state. Since IEp is linear in p
the same property follows for any trace class operator. Now the required result is
immediate from the definition of 013 1 0013 2 . •
Let 'Xo, 'X be Hilbert spaces where dim 'X = d < 00. Let {eo, el, ... ,ed-I}
be a fixed orthonormal basis in 'X and let 013 0 C 013('Xo) be a von Neumann algebra
with identity. Putting 'Xn = 'X, ¢n = eo for all n 2: 1 in Exercise 16.11 construct
the Hilbert spaces 'X nj , 'X[n+l for each n 2: O. Define the von Neumann algebras
®n
013 nj = {X 0 l[n+lIX E 013 0 0013('X )}, n 2: 0,
013 = 013 0 0 013('X[1).
Property (v) in Exercise 16.11 implies that 013 is the smallest von Neumann algebra
containing all the 013 n , n 2: O. {013 nj } is increasing in n. By Proposition 18.1 the
¢[n+l-conditional expectation IEnj of Exercise 16.11 maps 013 onto 013 nj .
Any algebra with identity and an involution * is called a *-unital algebra.
If 013 1 ,013 2 are *-unital algebras and () : 013 1 ---7013 2 is a mapping preserving * and
identity then () is called a *-unital map.
Proposition 18.2: Let () : 013 0 ---7 013 0 0013('X) be a *-unital homomorphism. Define
the linear maps (); : 013 0 ---7 013 0 by
()j(X) = 1E1ej)(e;I(()(X)), O:S i, j :S d - 1, X E 013 0 (18.2)

Then the following holds:
(i) ();(l) = oj, ();(X*) = (){(X)*;
(ii) ();(XY) = ~~:~()k(X)()J(Y) for all X, Y E 013 0 .
Proof: From (18.1) and (18.2) we have

(u,oj(X)v) = (u tg) ei,O(X)V tg) ej).
If X = 1 the right hand side of this equation is (u, v)t5J. Furthermore
(u, oj(X*)v) = (u tg) ei, O(X*)v tg) ej) = (v tg) ej, O(X)u tg) ei)
= (v, O{ (X)u) = (u, 0{ (X)*v).
This proves (i). To prove (ii) choose an orthonormal basis {un} in 'lie o and observe
that
(u,oj(XY)v) = (u tg) ei,O(X)B(Y)v tg) ej)

= (O(X*)u tg) ei, O(Y)v tg) ej)
= L(u tg) ei,O(X)U r tg) ek)(u r tg) ek,O(Y)V tg) ej)

r,k
= L(Ok(X)*u, OJ(Y)v)
k
~.
= (u, Ljk(X)Bj (Y)v).
k
k
•
Proposition 18.3: Let 0, 'lieo, 'lie l , ~o be as in Proposition 18.2. Define the maps
jn : ~o - t ~n]' n = 0,1,2, . .. inductively by
jo(X) = X tg) 1[!,jl(X) = O(X) tg) 1[2,
(18.3)
jn(X) = L jn-I(Ok(X))l n- l] tg) lei)(ekl tg) l[n+I'
O::;i,j::;d-I
Then jn is a *-unital homomorphism for every n. Furthermore
[En-1]jn(X) = jn-l(Og(X)) for all n 2 1, X E ~o,
where [En -I] is the ¢[n -conditional expectation map of Exercise 16.11.
Proof: We prove by induction. For n = 0, 1 it is immediate. Let n 2 2. Then by

(i) in Proposition 18.2 and the induction hypothesis we have
jn(1) = Ljn-I(t5j)l n- l] tg) lei)(ejl tg) l[n+!

i,j
= In-I] tg) Llei)(eil tg) l[n+1 = 1

i
and
jn(X)* = Vn-1(ot(X*))ln-1] Q9lej)(eil Q91[n+l

i,j
= jn(X*).
By (ii) in Proposition 18.2 and induction hypothesis we have
jn(X)jn(Y) = L jn-l(0}(X))jn-l(0:(Y))l n- 1] Q9 OJlei)(etl Q9 l[n+l

i,j,k,l
= Ljn-l LO~(X)O:(Y))ln-l] Q9lei)(etl Q9 l[n+l

i,l k
= Ljn-l(O~(XY))ln-l] Q9lei)(etl Q9 l[n+l'

i.l
This proves the first part. By the definition of lEn] in Exercise 16.11 and the fact
that (eo, lei)(ejleo) = O&oJ, the second part follows from (18.3). •
Corollary 18.4: Let {jn, n ;:::: O} be the *-unital homomorphisms of Proposition

18.3. Write T = og. Then for 0 ::; no < nl < ... < nk < 00, Xi E rA o, 1 ::; i ::; k
IEnoJin! (Xdjn2 (X2) ... jnk (Xk) =

(18.4)
jno(Tn!-no(X,Tn2-n'(X2'" (Xk_lTnk-nk-l(X))···)
Proof: By Exercise 16.11 IEno] = IEno]lEnk_d' Since jnJXd .. · jnk_JXk-d is

an element of rAnk_I] it follows from the same exercise that
IEno]jn! (Xd··· jnk(Xk)

(18.5)
= IEnoljn! (XI) ... jnk_! (Xk-dlEnk_d (jnk (Xk)).
Substituting this in (18.5), using the fact that jnk-l is a homomorphism and re-
peating this argument successively we arrive at (18.4). •
Proposition 18.5: The map T = og from rAo into itself satisfies the following:
(i) T is a *-unitallinear map on rA(~o); (ii) for any Xi, Yi E rA o, 1 ::; i ::; k,
Ll::;i,j 9 Yi*T(Xt Xj)lj ;:::: 0 for every k. In particular, T(X) ;:::: 0 whenever
X;::::O.
Proof: Since () is a *-unital homomorphism from \%0 into \%0 Q9 \%(7Je) and
T(X) = 1E1eo)(eol((}(X)), (i) is immediate from Exercise 16.10. Using the same
exercise once again we have
i,j i,j
= 1E1eo)(eol({L(}(Xi)Yi Q91}*{L (}(Xj)Yj Q91}) 2: O.

i j
Putting k = 1, Yi = 1, XI = X in this relation we get the last part.

We may now compare the situation in Proposition 18.3, Corollary 18.4 and
•
Proposition 18.5 with the one that is obtained in the theory of classical Markov
chains. Consider a Markov chain with state space S = {I, 2, ... ,N} and transition
probability matrix P = ((Pij)), 1 :::; i, j :::; N. Denote by Boo the *-unital
commutative algebra of all bounded complex valued measurable functions on the
space SOO = So X SI x· .. X Sn X ... where Sn = S for every n. Let Bn] C Boo be
the *-subalgebra of all functions which depend only on the first n + 1 coordinates.
Denote by lEn] the conditional expectation map determined by
(IEn]g)(io,iJ, ... ,in) = lE(glXo = io,X I = i l , ... ,Xn = in)
where X o, XI, ... is the Markov chain starting in the state Xo = io with stationary
transition probability matrix P. For any function 1 on S define
jn(1)(i) = l(i n ) where i = (io, i l , ... ,in, ... ) E SOO.
Then jn is a *-unital homomorphism from Bo into Bn] and the Markov property
implies that
lEnD] jn! (11 )jn2 (h) ... jnk (h) =
(18.6)
where
N
(Tf)(i) = LPijlU), 1 E Bo = Bo],

j=1
no < nl < ... < nk and 11, ... ,lk E Bo. T is a *-unital positive linear map on
Bo. Then (18.4) is the non-commutative or quantum probabilistic analogue of the
classical Markov property (18.6) expressed in the language of *-unital commuta-
tive algebras. For this reason we call the family {jn, n 2: O} of homomorphisms in
Proposition 18.3 a quantum stochastic flow induced by the *-unital homomorphism
() : \%0 ---t \%0 Q9 \%(7Je).
Proposition 18.6: Suppose that the von Neumann algebra \%0 in Proposition 18.3
is abelian. Then for any X, Y E \%0, m, n 2: 0
[jm(X),jn(Y)] = O. (18.7)
116 Chapter II: Observab1es and States in Tensor Products of Hilbert Spaces
Proof: Since jn is a homomorphism we have [jn(X),jn(Y)] = jn([X, Y]) = O.

Thus (18.7) is trivial if m = n. Suppose m < n. By induction on (18.3) we have
jn(X) =
L jm([Ok\'" O~:-=-: (X))lmJ ® leil)(ekll ® ... ® lein_m)(ek n_m1 ® l[n+1
os; it ,i2"" ,
kl,k2,'" ,::;<1-1
from which it follows that
[jn(X),jm(Y)] =
L jm([Ok\ ... O~:-=-:JX), Y])lmJ ® leil)(ekll ® ... ® lein_m)(ek n_m1 ® l[n+1

it ,i2,'"
kl,k2 ....
= 0 for all X, Y E ~o.
If ~o is abelian and X I , X 2, . . . ,Xk is any finite set of selfadjoint elements

in ~o then jnl (Xt},jn2(X2),'" ,jnk(Xk ) is a commuting family of observables
and hence possesses a joint distribution in IRk in any state P on the countable
tensor product 'Jeo ® {'Je ® 'Je ® ... } where the Hilbert space within the braces
{ } is with respect to the constant stabilising sequence of unit vectors eo. For
any state Po in 'Jeo, the family {jn(X)IX E ~o, n 2: O} can be interpreted as a
classical Markov flow in the state Po ® 1eo ® eo ® ... ) (eo ® eo ® ... I.
Example 18.7: Let (8, '3', J.L) be any measure space and let <Pj : 8 ----t 8 be
measurable maps satisfying J.L<P j I « J.L, 0 ~ j ~ d - 1. Suppose Pj : 8 ----t
[0, 1], 0 ~ j ~ d - 1 are measurable functions satisfying L,jPj (x) == 1. Let
~o = L 00 (J.L) C ~ (L2 (J.L)) when bounded measurable functions are considered
as bounded multiplication operators. We write 'Jeo = L 2 (J.L), 'Je = Cd and choose
{eo, el, ... ,ed- J} to be the canonical orthonormal basis in 'Je. Define the map
T: ~o ----t ~o by
d-I
Tj = LPjjo<Pj (18.8)
j=O
where 0 denotes composition. The map T can be interpreted as the transition op-
erator of a Markov chain with state space S for which the state changes in one
step from x to one of the states <Po (x ), <PI (x), . .. ,<Pd-I (x) with respective proba-
bilities PO(X),PI(X), ... ,Pd-I(X). However, it is possible that <Pi(X) = <pj(x) for
some i i- j. If S is a finite set of cardinality k it can be shown that every Markov
transition operator is of the form (18.8) with J.L being counting measure. In many
practically interesting models of classical probability d is small.
Consider a unitary d x d matrix valued measurable function U on 8 for
which U = ((Uij)),O ~ i,j ~ d - 1, UOj = y'Pj for all j. For example one may
choose the orthogonal matrix

1/2 1/2 1/2
Po Pl······ Pd-I
1/2
U= -PI
1- Q
1/2
-Pd-I
where Q = ((qij)),qij = (PiPj)I/2(1 + P6/ 2)-I,i,j > 1. Define the *-unital

homomorphism
10 ¢0
B: 1 ~ ((Bj(j))) = U ( 0 10¢1 (18.9)
··10¢d-1
from (lJilo into (lJilo @ (lJil('Je) so that
d-I
Bj(j) = LUirUjr10¢r,
r=O
Bg(j) = T f.
where T is defined by (18.8). Note that in (18.9) the right hand side is to be
interpreted as a matrix multiplication operator in the Hilbert space
L2(fj,) @ Cd = L2(ft) Ell··· Ell L2(ft)
, v "
d-fold
and any element in the right hand side version of the Hilbert space is expressed as
a column vector of elements in L2(ft). By Proposition 18.3 and 18.6 there exists
a quantum stochastic flow {jn, n :::: O} of *-unital homomorphisms from (lJilo into
(lJil induced by the *-unital homomorphism B of (18.9) satisfying
[jm(j),jn(g)] = 0,
IEn-IJln(j) = jn-I(Tj)
for all 1, g E (lJilo = L 00 (ft). If S is a countable or finite set and ft is
the counting measure then for any XES, in the quantum theoretical state
Ox @ eo @ eo @ ... in L2 (ft) @ 'Je @ 'Je @ ... the sequence of observables jo(j) =
1,jl(j), ... ,jn(j),· .. has the same probability distribution as the sequence of
random variables 1(~0(x, w)), 1(6 (x, w)), 1(~n(x, w)), ... where {~n(x, w)} is a
discrete time classical Markov chain with state space S, ~o (x, w) = x and tran-
sition operator T,1 is any element of (lJilo. In other words {jn} can be identified
with a classical Markovian stochastic flow with transition operator T.
It is interesting to note that in the quantized construction of the classical
Markov chain we get the following bonus. By confining ourselves to {jn,O ::S
n ::S N} for any finite time period in the Hilbert space 'Jeo @ 'Je®n and defining
the conditional expectations 1E~1 with respect to an arbitrary unit vector 'IjJ, in 'Je
replacing eo, we get a Markov flow with the property
lE~o]jnl (fl) ... jnk (fk) = jno (T;I-nO

(h T ;2- nl (f2 ... (Ik_IT;k-nk-l (fk)) ... )
for all °:S no < nl < ... < nk :S N where T1/! is the Markov transition operator
d-I
T1/!j = LI(ej,U(·)'IjJ)1 2jo¢j.
j=O
T 1/! describes the chain where the state changes from x to one of the states
¢O(X),¢I(X)"" ,¢d-I(X) with respective probabilities l(ej,U(x)'IjJ)jZ,j = 0,1,
... , d - 1. Thus a quantum probabilistic description enables us to describe a whole
class of Markov chains with transition operators T1/!, 'IjJ E 'Je,II'ljJ11 = 1 within the
framework of a single Hilbert space 'Jeo ® 'Je@N if we confine ourselves to the
finite time period [0, N].
Example 18.8: We examine Example 18.7 when d = 2. Denote the maps ¢o and
¢I on S by ¢ and 'IjJ respectively. Write Po = p, PI = q so that P + q = 1. Then
the *-unital homomorphism () in (18.9) assumes the simple form
()(f) = ((()~(f))) = ( pjo¢ + qjo'IjJ y'pQ(fo'IjJ - jO¢)) (18.10)

J y'pQ(fo'IjJ - jo¢) qjo¢ + pjo'IjJ
when
U= (yip yq).
-yq yip
Define
an = In-I] ® leo)(ell ® 1[n+J,

a~ = In-I] ® lel)(eol ® 1[n+J, n = 1,2, ... ,
so that
ana~ = In-I] ® leo)(eol ® I[n+J,
a~an = In-I] ® lel)(ell ® I[n+J,
and ana~ + a~an = 1 for all n ~ 1. Then the flow {jn, n ~ o} induced by () in
(18.3) assumes the form
jn(f) = jn-I (T f) + jn-I (Lj)(an + a~) + jn-I (K f)a~an (18.11)
for all j E LOO(/L), n ~ 1 where
Tj = pjo¢ + qjo'IjJ,
(18.12)
Lj = y'pQ(fo'IjJ - jo¢), Kj = (p - q){jo'IjJ - jo¢}
If An =
at al
° for n = ° and al + .,. + an for n ~ 1 and An = for n =
+ ... + ahan for n ~ 1 we may express (18.11) in the form
° ° and
jn(f) = jn-l(Tf)+jn-l(Lf)(An-An- 1 +Ah -AL1)+jn-l(Kf)(An -An-d

(18.13)
°
where An, Ah and An, n ~ are three martingales with respect to the conditional
expectations [nj (see Exercise 16.12). In Chapter III we shall treat continuous
time analogues of flows satisfying (18.13) where difference equations will become
differential equations.
Example 18.9: (Hypergeometric model) Consider an urn with a white balls and
b black balls. Draw a ball at random successively without replacement. The state
of the Markov chain at any time is denoted (x, y) where x is the number of white
balls and y is the number of black balls. Then
S = {(x,y)IO::; x::; a,O::; y::; b}
is the state space. Define maps ¢, 'IjJ on S by
(X-l,y) ifx>O
¢(x,y)= { (O,y-l) ~fx=O,y>O,
(0,0) If x = 0, y = 0,
(X,y-l) if Y > 0,
'IjJ(x,y) = { (x - 1,0) if Y = 0, x > 0,
(0,0) if x = Y = 0.
Define
....L if x> 0,
p(x, y) = { 2+Yi if x = 0.
Then
X y
- f ( x -1,y) + -f(x,y - 1) if x> 0, y > 0,
{
x+y x+y
(Tf)(x,y) = f(x -1,0) if x> 0, y = 0,
f(O,y-l) if x = 0, y > 0,
f(O,O) if x = 0, y = 0.
We may call {jn, n ~ O} defined by the *-unital homomorphism () in (18.10)-
(18.12) the hypergeometric flow.
Example 18.10: (Ehrenfest's model) There are two urns with a and b balls so
that a + b = c. One of the c balls is chosen at random and shifted from its urn to
the other. The state of the system is the number of balls in the first urn. Then
S = {O, 1,2, ... ,c}.
Define the maps ¢, 'IjJ on S by
if x> 0,
¢(x) = {~- 1 if x = 0,
'IjJ(x) = {x+l ifx<c,
c-l ifx=c.
{e-1x e,
Define
( x) = if 0 < x 0, b > O. Draw a ball at random, replace it and add e balls
of its colour. The state of the system can be described by (x, y) where x and y
are respectively the number of white and black balls. Thus S = N x N where
N = {I, 2, ... }. Define
¢(x, y) = (x + e, y), 'IjJ(x, y) = (x, y + c)
p(x, y) = x(x + y)-l, q(x, y) = y(x + y)-l
Then
(Tf)(x, y) = (x + y)-l{xf(x + e, y) + yf(x, y + en.
We may call the corresponding {jn, n ~ O} the Polya flow.
Exercise 18.12: (i) Let 'ZIeo = L2 (JL) where JL is the counting measure in S =
{I, 2, ... ,N} and let 'ZIe be any Hilbert space. For any function f on S denote
by the same letter the operator of multiplication by f. Let eJil o be the *-unital
abelian algebra of all complex valued functions viewed as an abelian von Neumann
subalgebra of eJiI('ZIeo). A map () : eJil o -* eJil o (S9 eJiI('ZIe) is a *-unital homomorphism
if and only if there exists a matrix Pe = ((Pij))I50i,j50N of projections in 'ZIe
satisfying the following:
(1) Pi! + ... + PiN = 1,
(2) ()( I{ i}) = 2:){j} (S9 Pji for every i.

j
where I{i} is the indicator function of the singleton {i} in S.

(ii) If 'ZIe i , i = 1,2 are Hilbert spaces, (}i : eJil o -* eJil o (S9 eJiI('ZIeo) are *-unital
homomorphisms and Pei = (( p~~))) are the corresponding matrices of projections
in (i) then
is a matrix of projections in 'ZIe 1 (S9 'ZIe2 satisfying property (1) of part (i). Thus
Pel * Pe2 determines a *-unital homomorphism () : eJil o -* eJil o (S9 eJiI('ZIeJ) (S9 eJiI('ZIe 2 )
where
(}(I{i}) = '2){j} (S9 (S9 PjP p;;)
for every i.
j,T
(iii) Let (), Pe be as in (i). Then the quantum stochastic flow {jn, n ~ O}
induced by () is given by
jn(f) = ~I{jo} (S9 Pjoj, (S9 PM2 (S9 .•. (S9 Pjn-tin (S9 f(jn)l[n+l'
If 0 is the unit vector in 'X with respect to which the conditional expectation maps
lEn], n 2': 0 are defined then
IEn-l]in(f) = in-I(T!) where (Tf)(i) = LPij!U),Pij = (O,PijO).
j
Example 18.13: [115] Let G be a compact second countable group and let
9 --+ Lg denote its left regular representation in the complex Hilbert space L2( G)
of all absolutely square integrable functions on G with respect to its normalised
Haar measure so that (Lgf)(x) = f(g-IX), ! E L2(G). Denote by ~o the von
Neumann algebra generated by {L g Ig E G} and its centre by Zoo Let r( G) denote
the countable set of all characters of irreducible unitary representations of G. For
any X E r( G) let UX be an irreducible unitary representation of G with character
X and dimension d(X). If XloX2 E r(G) the map 9 --+ U;l I8i U;2,g E G
defines a unitary representation which decomposes into a direct sum of irreducible
representations. Denote by m(XI, X2 ; X) the multiplicity with which the type UX
appears in such a decomposition of UXI I8i UX2. Define
x _ m(X,X2; Xdd(X2)
PX1 ,X2 - d(X)d(XI) .
Then
L P~I'X2 = 1 for each X,XI E r(G).
x2H(G)
In other words, for every fixed X E r(G), the matrix px = ((P~I'X2)) is a
stochastic matrix over the state space r( G). In each row of px all but a finite
number of entries are 0 and each entry is rational. Thanks to the Peter-Weyl
Theorem L2( G) admits the Plancherel decomposition
L2(G) = E9 'Xx
xH(G)
where dim 'Xx = d(X?, Lg leaves each 'Xx invariant and Lg I~t)(, 9 EGis a
direct sum of d(X) copies of the representation UX. If 7rx denotes the orthogonal
projection onto the component 'Xx then
7rx = d(X) faX(9)Lgd9'

Fix Xo E r( G). Let uXo act in the Hilbert space 'X. Denote by p the state
d(XO)-1 I in '1JC. Fix a positive integer N and consider in '1JC0 N the increasing
sequence of von Neumann subalgebras
~n] = {X I8i 1[n+1.N] IX E ~('X0n)}, n = 1,2, ... ,N
where ~ N] = ~('X0N) and I [a,b] denotes the identity in 'X 0b - a +1 , the tensor
product of the a-th, a+ 1-th, ... , b-th copies of 'X in 'X0 N. Consider the conditional
expectation maps lEn] : ~N] --+ ~n] defined by
IEn]X = (lEp0N-nX) I8i l[n+I,N], 1~ n ~ N - 1
(see Exercise 16.10) so that
IEnlXI Q9 •.. Q9 X N = (II~n+1 tr pXi)(XI Q9 .•. Q9 Xn) Q9 l[n+l,Nl
for all Xi E 0iI(~). By the Peter-Weyl Theorem there exists a unique *-unital
homomorphism j~O : 011 0 --+ 0iI nl satisfying
(i) j~O(Lg) = (U;o/6i n Q9 l[n+l,Nl

(ii) IEn_l]j~O(X) = jn-I(T(X)) for all X E 011 0 where
T(Lg) = d(XO)-IXO(g)Lg, T(7r)J = L P~~,x7rx"

X'Er(G)
(iii) The centre Zo of 011 0 is generated by {7rx IX E f(G)} and [ji}{'(Z),j~O(X)l =

o for all Z E Zo, X E 011 0 , m :S n. In particular, the family {j~o (Z) 11 :S
n :S N, Z E Zo} is commutative and hence {j~O IZo, 1 :S n :S N} induces
a classical Markov chain with state space f( G) and transition probability
matrix pxo for every Xo E f( G).
(iv) When G = SU(2) and Xn denotes the character of the unique equivalence
class of an irreducible unitary representation of dimension n the Clebsch-
Gordan formula implies that
i-I 'f' . 1
~ 1)=2- ,
X2 - { . 1
px"xJ - ~
2i 'f' - 2. + 1,
I) -
o otherwise.
Exercise 18.14: Let (S,?:;,J,l) be as in Example 18.7. Suppose that

d-I d-I
Ttl = LPdo(/Ji, Td = LqdoWi

i=O i=O
are two Markov transition operators as described in (18.8) and
0
( /010 jO<PI
(h(1)=U ) U-'
0 j0<Pd-1 '
('"'"
0
fOWl
82 (1) = V ) V-,
0 jOWd-1 '
19 The Fock Spaces 123
are the *-unital homomorphisms corresponding to (18.9). Define
w= ( Ucoso: V sino: )
- V* sin 0: V*U*V cos 0: '
focpo o
()(f) = W w- 1
o
where 0: is any fixed angle. Then W is a unitary matrix valued function and () is
a *-unital homomorphism for which ()g = Tl cos 2 0: + T2 sin2 0:. Thus the Markov
transition operator of the quantum stochastic flow induced by () is a superposition
(or convex combination) of the two transition operators Ti, i = 1,2.
Notes
The idea of describing a general quantum stochastic process in terms of a time
indexed family {jn} or {jt} of *-unital homom~rphisms from a *-unital initial or
system algebra '2Ao into a larger *-unital algebra '2A made up of '2Ao and noise or heat
bath elements has its origin in Accardi, Frigerio and Lewis [6] modelled on the
description of classical processes by Nelson (J. Funct. Anal., 12 (1973) 97-112,
211-277) and Guerra, Rosen and Simon (Ann. Math., 101 (1975) 111-259).
For a detailed account of Markov chains (or discrete time flows) in *-algebras
of the form '2Ao Q9 '2A Q9 '2A Q9 ... where '2Ao is the initial or system algebra and '2A is the
noise algebra see Kummerer [71,73]. The account given here is in anticipation of
Evans-Hudson flows which are discussed in Section 28. It is based on Parthasarathy
[112] and inspired by Meyer [91]. Example 18.9-18.11 are based on classical
probability theory as described in Feller [40]. Example 18.12 was suggested to
me by B.V.R. Bhat. Example 18.13 is based on the work of Biane [22], von
Waldenfels [138] and Parthasarathy [115].
19 The Fock Spaces

In Section 16-18 we saw how the notion of tensor products of Hilbert spaces
enables us to combine several quantum probability spaces into one. In this context
there is yet another basic construction leading to the combination of an "indefinite"
number of such systems. This idea is illustrated by first raising the following
question: if the events concerning the dynamics of a single particle are described
by the elements of rzJl('J£) where 'J£ is a separable Hilbert space, how does one
construct the Hilbert space for an indefinite number of such particles in a system
where the indefiniteness is due to the fact that "births" and "deaths" of particles
take place or, equivalently, particles are being "created" and "annihilated" subject
to certain laws of chance? We shall try to achieve this by pooling all the finite
order tensor products into a single direct sum. To this end we collect some of
the elementary properties of direct sums of operators in the form of a proposition
without proof.
Proposition 19.1: Let An E ?J3('Jf n ), n = 1,2, ... where {'Jf n } is a sequence

of Hilbert spaces. Suppose sUPn11An11 < 00. Then there exists a unique operator
A = EBn An on 'Jf = EBn 'Jf n satisfying
(i) A EBn Un = EBn Anu; (ii) IIAII = sUPniIAnll.
If {An}, {Bn} are two sequences of operators such that An, Bn E ?J3('Jf n )
for each nand sUPn(IIAnll + IIBnll) < 00 then their direct sums A = EBn An and
B = EBn Bn satisfy the following:
(a) A + B = EBn (An + B n ), AB = EBnAnBn, A* = EBn A~;
(b) If each An has a bounded inverse and sUPn II A;;: III < 00 then A has a
bounded inverse and A -I = EBn A;;: I;
(c) A is a selfadjoint, normal, unitary, positive or projection operator according
to whether each An has the same property;
(d) If A E g, 00 ('Jf) and therefore has finite trace then each An E g, 00 ('Jf n ),
n = 1,2, ... , and IIAIII = I:nIIAnlit. tr A = I:n tr An;
(e) If A = P is a state in 'Jf then there exist states Pn and scalars Pn ~ 0,
n = 1,2, ... such that I:nPn = 1 and P = EBn PnPn·
Proof: Omitted.
Let 'Jf be a Hilbert space and let 'Jf®n, 'Jf®n and 'Jf@n be the n-fold tensor
•
product, symmetric tensor product and anti symmetric tensor product of 'Jf respec-
tively, where the O-fold product is the one dimensional complex plane and the
I-fold product is 'Jf itself in all the three cases. The Hilbert spaces
00 00 00
r Ir ('Jf) = EB'Jf®n, r s ('Jf) = EB'Jf®n ,r a ('Jf) = EB'Jf@n

n=O n=O n=O
are respectively called the free (or Maxwell-Boltzman), the symmetric (or boson)
and the antisymmetric (or fermion) Fock space over 'Jf. The n-th direct summand
in each case is called the n-particle subspace. When n = 0 it is called the vacuum
subspace. Any element of the n-particle subspace is called an n-particle vector.
The vector 1 EB 0 EB 0 EB ... is called the vacuum vector which we shall denote by
<P. We denote by r~r('Jf),r~('Jf) and r~('Jf) the dense linear manifold generated
by all n-particle vectors, n = 0, 1,2, ... in the corresponding Fock space and call
any element in it a finite particle vector. For any U E 'Jf the element
e(u) = EBn(n!)-1/2 u ®n (19.1)
(where O! = 1, u®o = 1) belongs to r s ('Jf) c r Ir ('Jf) and is called the exponential
(or coherent) vector associated with u. For any u, v E 'Jf
(e(u), e(v)) = exp(u, v). (19.2)
The projections E and F from r Ir (~) onto the subspaces r s (~) and r a(~) can
be expressed as
(19.3)
where En and Fn are the projections from ~0n onto ~®n and ~@n respectively
described by Proposition 17.1. If dim ~ = N < 00 then the direct sum in r a (~)
terminates at the N -th stage and by the second part of Proposition 17.3
dimra(~) = ~(~) =2N.

Let Urs : ~0r 0 ~08 -+ ~0r+8 be the unique unitary isomorphism satisfying the
relations
Urs(UI 0··· 0U r ) 0 (VI 0··· 0v s ) = UI 0··· 0ur 0vI 0··· 0vs
for all Ui, Vj E ~, 1 :::; i :::; r, 1 :::; j :::; s. Such an isomorphism is well-
defined in view of Proposition 7.2 and 15.5. We use this isomorphism to identify
~0r 0 ~08 with ~0r+8. In each of the Fock spaces let Pn denote the projection
on the n-particle subspace for every n. If ?P(~) is interpreted as the collection
of events concerning the dynamics of a single particle then ?p(r Ir(~)) can be
interpreted as the collection of events concerning an indefinite number of identical
but distinguishable particles obeying Maxwell-Boltzmann statistics (see (17.13)).
Similarly ?P (r s (~)) and ?P (r a (~)) may be considered as the collection of events
concerning an indefinite number of identical bosons and fermions respectively.
(See (17.14), (17.15) and the succeeding remarks.) In such a case the projection
Pn signifies the event that the number of particles in the system is n.
Proposition 19.2: Define multiplications (u, v) -+ U 0 v, UV, U 1\ v respectively

in the finite particle Fock spaces r~r (~), r~ (~), r~ (~) by
U 0 v = EBn L Pru 0 Psv, uv = Eu 0 v, U 1\ v = Fu 0 V

r+s=n
where 1 0 U = U 0 1 = U and E, F are defined by (19.3). Then the following
properties hold: (i) r~r (~) is an associative algebra; (ii) r~ (~) is a commutative
and associative algebra; (iii) r~ (~) is an associative algebra in which
U 1\ v = (_1)mnv 1\ U for all U E ~@m ,v E ~@n .
Proof: The associativity of the three multiplications is immediate from Exercise

15.7 and Corollary 17.2. (ii) is immediate from definitions. The last part follows
from the fact that the permutation from (1,2, ... ,m+n) to (m+ 1, m+2, . .. ,m+
n, 1,2, ... ,m) can be achieved by mn successive elementary permutations which
are transpositions. •
If we say that the degree or rank of any n-particle vector is equal to n then
the product of two vectors of degrees m and n leads to a vector of degree m + n in
all the three algebras r~r(~),r~(~) and r~(~). These are examples of "graded
algebras". r~r(~) is the tensor algebra over ~ whereas r~(~) and r~(~) are
the symmetric and exterior algebras over ~. When dim ~ = 00 the products of
m-particle and n-particle vectors generate only a dense linear manifold in the
subspace of m + n-particle vectors. In this sense the definitions here are different
from the corresponding notions in the algebraic appoach to tensor products over
a vector space.
If rjr(~) and rfr(~) denote the subspaces spanned by vectors of even
and odd degrees respectively then any element of rjr (~) is called even and any
element of rfr(~) is called odd. If u, v E r~(~) then U /\ v = v /\ U if either U
or v is even and U /\ v = -v /\ U if both u, v are odd. u /\ v is odd if one of them
is odd and the other is even. u /\ v is even if both are odd or both are even. In
other words r~(~) is a Z2-graded algebra under the multiplication /\.
Proposition 19.3: Suppose {ej, j = 1,2, ... ,} is an orthonormal basis in ~.

Then the three sets
{, ei[ ® ... ® ei n lij = 1,2, ... ; j = 1,2, ... ,n; n = 1,2, ... ,},
{, ( ,n!
rl.··· rk.
,)1/2 e~lle~; ... e?h + ... +rk = n,
k
1 :::; il < i2 < ... , (n!)1/2eil /\ ei 2 /\ ••• /\ ein!1 :::; il < i2 < ... < in, n = 1,2, ... }
are respectively orthonormal bases in the Fock spaces r Ir (~), r s (~) and r a (~),
the multiplications being defined according to Proposition 19.2.
Proof: This is immediate from Exercise 15.8 and Proposition 17.3.

•
Proposition 19.4: The set {e( u) lu E ~} of all exponential vectors is linearly
independent and total in rs(~).
Proof: Let {Uj 11 :::; j :::; n} be a finite subset of ~. Since {ul (u, Uj) =J (u, Uk)},
j =J k are open and dense in ~ there exists a v in ~ such that the scalars
()j = (v,Uj), 1:::; j:::; n are distinct. Suppose aj, 1:::; j:::; n are scalars such that
'E.jaje( Uj) = O. Then for any scalar Z
0= (e(zv),:~::~>je(Uj)) = Laje Z9j •

j j
Hence aj = 0 for all j. This proves linear independence. To prove the second
part consider an orthonormal basis {ed in ~. If U = ZI ei l + ... + Zkeik, Zj being
scalars then the coefficient of Z~I Z~2 ... Z~k, rl + ... + rk = n, in u®n belongs
to ~®n . Suppose that S is the closed subspace spanned by all the exponential
vectors. Then the vacuum vector e(O) E S. By (19.1)

n
u®n+! = (n + 1!)l/21imC(n+ll{e(EU) - EB(r!)-1[T U ®r}.
e~O
T=O
Hence it follows by induction that u®n E S for all nand u in '!if. Thus
S = rs('!if). •
Corollary 19.5: Let S be a dense set in '!if. Then the linear manifold '€,(S)
generated by S = {e(u)lu E S} is dense in rs('!if). 1fT: S --t rs('!if) is any map
there exists a unique linear operator T on r s('!if) with domain'€, (S) satisfying
Te(u) = Te(u) for all u E S.
Proof: For any u, v E '!if
Ile(u) - e(v)11 2 = elluI12 + eilvI12 - 2Ree(u.v).
Since the scalar product is continuous in its arguments this shows that the map
u --t e(u) from '!if into rs('!if) is continuous and, in particular, {e(u)lu E S} =
{ e(u) Iu E '!if}. The second part follows from the linear independence of the set
of all exponential vectors. •
The next two propositions indicate how the correspondences '!if --t r s('!if)
and '!if --t r a ('!if) share a functorial property in the "category of Hilbert spaces".
Proposition 19.6: Let '!if l , '!if2 be Hilbert spaces and '!if = '!if j ED '!if 2 • Then there
exists a unique unitary isomorphism U : r s('!if) --t r s('!if j) Q9 r s('!if 2 ) satisfying
the relation
Ue(u ED v) = e(u) Q9 e(v) for all u E '!if l, v E '!if2. (19.4)
Proof: We may assume without loss of generality that '!if j , '!if2 are mutually orthog-
onal subspaces of '!if. By Proposition 19.4 the sets {e( u) lu E '!if}, {e( u) lu E '!ifd,
i = 1,2 are respectively total in rs('!if),rs('!ifi), i = 1,2. By Proposition 15.6 the
set {e(u) Q9 e(v)lu E '!ifj,v E '!if 2} is total in fs('!ifI) Q9 fs('!if 2 ). By (19.2), for
Ui, Vi in '!if i
= (e(ul), e(vJ))(e(u2)' e(v2)).
In other words the map U defined by (19.4) on the set {e (u + v) Iu E '!if l, v E '!if 2 }
is scalar product preserving. Hence by Proposition 7.2 U extends uniquely to the
required unitary isomorphism. •
Proposition 19.7: Let 'Je 1, 'ilC2 be Hilbert spaces and let 'Je = 'Je 1 EEl 'ilC2 • Then
there exists a unique unitary isomorphism U : r a('ilC 1 EEl 'J( 2 ) -+ r a('ilCt} ® r a('ilC 2 )
satisfying the relations:
U[(m + n)!p/2 u1 1\ ... 1\ Urn 1\ VI 1\ ... 1\ Vn
(19.5)
= {(m!)1/2ul 1\ ... 1\ urn} ® {(n!)1/2vl 1\ ... 1\ vn }
for all Ui E 'Je 1o Vj E 'Je2, 1 :::; i :::; m, 1 :::; j :::; n, m = 1,2, ... , n = 1,2, ... and
U<P = <PI ® <P2,<P,<Pl,<P2 being the vacuum vectors in ra('Je),ra('ilCJ),ra('Je2)
respectively.
Proof: Once again, as in the proof of Proposition 19.6, we assume without loss
of generality that 'Je 1, 'ilC2 are orthogonal subspaces of 'iIC. Let
8 = {<p, [(m + n)!]1/2ul 1\ ... 1\ Urn 1\ VI 1\ ... 1\ VnlUi E 'ilC l , Vj E 'Je2,}
l:::;i:::;m, l:::;j:::;n,m=I,2, ... ; n=1,2, ... },
8 1 = {<P1o (m!)1/2Ull\"'1\ Urn,Ui E'ilC l , 1:::; i:::; m, m = 1,2, ... },
i:::; n, n = 1,2, ... }.

8 2 = {<P2, (n!)1/2Vll\· .. 1\ Vn,Vi E'ilC l , 1:::;
By Proposition 19.3, 8,81,82 are total in r a('Je), r a('Jet), r a('ilC2) respectively.
Furthermore, the set {u ® vlu E 810 V E 82} is total in r a('JeJ) ® r a('Je2). Thus
it suffices to show that the map U defined by (19.5) is scalar product preserving.
Let Ui'U~ E 'ilCj, 1 :::; i:::; m, 1 :::; k:::; m', Vj,V~ E 'ilC2, 1 :::; j :::; n, 1 :::; i:::; n'.
We now consider three different cases:
Case 1: m + n #- m' + n'.

Define Ull\·· . 1\ Urn = <PI, Vll\·· ·I\vn = <P2, Ull\·· ·I\UrnI\Vll\·· ·I\vn = <P
when m = n = O. Then we see that since Ul 1\ ... 1\ Urn 1\ VI 1\ ... 1\ Vn and
u~ 1\ ... 1\ u~, 1\ v; 1\ ... 1\ v; 1\ ... 1\ v~, are m + n and m ' + n' -particle vectors,
they are orthogonal. Furthermore, either m #- m ' or n #- n' and hence
(Ul 1\ ... 1\ Urn, u~ 1\ ... 1\ u~, )(Vl 1\ ... 1\ Vn , v; 1\ ... 1\ v~,) = O. (19.6)
Case 2: m + n = m' + n', m #- m'.

Without loss of generality let m < m', n > n'. Then (19.6) holds. On the
other hand, writing
(Ul, ... ,urn,vj, ... ,vn ) = (WI, ... ,wrn+n),
(
Ul""
I
,urn"vl
I I
"" ,vn'
')
= WI"" ,wrn+n
(' ')
we observe that the matrix (((Wi, wj) ) ), 1 :::; i, j :::; m +n has the partitioned
form
C(rn'~rn)xn'
Brnx(rn'-rn)
o )
o Dn'xn'
and hence has rank ~ m+n' < m+n. Thus its determinant vanishes. By (17.10)
and the definition of 1\ we have
(UI 1\ ... 1\ Urn 1\ VI 1\ ... 1\ Vn , u~ 1\ ... 1\ U~, 1\ v; 1\ ... 1\ V~,) = o.
Case 3: m = m', n = n'.
Define (WI, ... , wrn +n ) and (w;, ... , w~+n) as in case 2 and observe that
the matrix (( (Wi, wj) )) has the partitioned form
( Arnxrn
o
0)
Bnxn
where A = (((ui,u~))),B = (((Vj,vi))). By (17.10)
(UI 1\ ... 1\ Urn 1\ VI 1\ ... 1\ Vn , u~ 1\ ... 1\ u~ 1\ v; 1\ ... 1\ v~)
1 1
-------:-, det(((wi,wj))) = - - , detAdetB
m+n. m+n.
= mIn! (UI 1\ ... 1\ urn ® VI 1\ ... 1\ Vn , UI"
1\ ... 1\ U" ® VI 1\ ..)
. 1\ V .
m+n! rn n
It is interesting to note that the correspondence

•
Uu ® V = EEl n L Pru ® Psv, U E r frelled, V E r frell(2)
r+s=n
can be extended by linearity and closure to an isometry from r fr (~I) ® r fr (~2)
into r fr(~1 EEl ~2) but not a unitary isomorphism. Thus the functorial property
established for the boson and fermion Fock spaces fails for the free Fock space.
The next few examples illustrate the connections between Fock spaces and
probability theory.
Example 19.8: Let J.L be the standard normal (or Gaussian) distribution on the
real line. Consider the Hilbert spaces L 2 (J.L) and the boson Fock space rs(C)(=
r fr(C)). Since C 0n = C for all n we have
rs(C) = C EEl C EEl ... = £2.
For any z E C the associated exponential vector e(z) is the sequence
e(z) = (1,z,(2!)-!z2, ... , (n!)-h n , .. . ).
In L2 (J.L) consider the generating function of the Hermite polynomials
zn
ezx -
1
2z
2
= L ,Hn(x),
n.
00
(19.7)
n=O
Hn being the n-th degree Hermite polynomial. There exists a unique unitary
isomorphism U: rs(C) ---+ L 2 (J.L) satisfying
1 2
[Ue(z)](x) = eZx - 2z for all z E C. (19.8)
Indeed, the set of functions {exp Zx IZ E C} is total in L 2(JL) and
~ eZlX-~Z~ . eZ2x-~z~dJ.L(x) = eZJZ2 = (e(zl),e(z2))'

By Proposition 19.4 and 7.2 the correspondence (19.8) extends uniquely to a
unitary isomorphism. Under this unitary isomorphism we have
U(l,O,O, ... ) = 1,
U(O,O, ... ,0,1,0, ... ) = (n!)-1Hn(X)
where, on the left hand side, 1 appears in the n-th position and n varies from to
00. In particular the sequence {( n!) -1 Hn (x), n = 0, 1,2, ... } is an orthonormal
°
basis in L 2(J.L).
In this example we may replace C by C k and the distribution J.L by its k-fold
cartesian product J.L(gl and replace (19.8) by
1 k
[Ue(~)](~) = exp(~· ~ - 2LZJ) (19.9)
j=l
where ~ . ~ = L,jZjXj. Then U is a unitary isomorphism between L 2(J.L@k)

and rs(Ck). When k is the countably infinite cardinal and correspondingly J.L@k
is the countable cartesian product of standard normal distributions we obtain a
unitary correspondence between the L2-space of an independent and identically
distributed sequence of standard Gaussian random variables and the boson Fock
space r s ([2), [2 denoting the Hilbert space of absolutely square summable se-
quences. This leads us to the connection between Brownian motion and the boson
Fock space.
Example 19.9: Let {w(t)lt ~ O} be the standard Brownian motion stochastic
°
process described by the Wiener probability measure P on the space of continuous
functions in [0,00). Then w(O) = and for any °
< tl < t2 < ... < tk <
°
00, W(tl)' w(h) - W(tl)"" , w(t k ) - w(tk-d are independent Gaussian random
variables with mean and lE(w(t) - w(s)? = t - s. For any complex valued
function I in L2(1R+) where IR+ = [0,00) is equipped with Lebesgue measure, let
1000
Idw denote the stochastic integral (in the sense of Wiener) of I with respect to
the path w of the Brownian motion. Then the argument outlined in Example 19.1
shows that there exists a unique unitary isomorphism U : r s (L2 (IR+)) -+ L2 (P)
1o00 Idw - -1100 l(t)2dt).

satisfying
[Ue(f)](w) = exp( (19.10)
2 0
For any t ~ 0, let It] = II[o,t] which agrees with I in the interval [0, t] and
vanishes in the interval (t, 00). Then
[Ue(ft])](w) = exp(lot Idw - ~ lot I(sfds), t ~° (19.11)

where the right hand side as a stochastic process is the well-known exponential
martingale of classical stochastic calculus.
A multidimensional analogue of this example can be constructed as follows.
Suppose {~(t) = (WI(t),W2(t), ... ,wn(t))lt ~ o} is the n-dimensional standard
Brownian motion process whose probability measure in the space of continuous
sample paths with values in IRn is prg;n. Let {ej 11 ~ j ~ n} be the canonical
basis of column vectors in en. Then there exists a unique unitary isomorphism
U : rs(L2(1R+) I8i en) -+ L2(prg;n) satisfying
n
j=1
.
[Ue(Lft~J) I8i ej)](~) =
n t I t
exp L{l f(j)(s)dwj(s) -
j=1 0
2:1 0
f(j)(S)2ds}
for all f(j) in L2(1R+), j = 1,2, ... , n.
Example 19.10: Let I-" be the Poisson distribution with mean value A in the space
Z+ of non-negative integers. In L2(1-") consider the generating function of the
Charlier-Poisson polynomials defined by
00 n
e-v'Az(1 + ~)X = L ; 7rn (A,X),X E Z+. (19.12)
VA n=on.
Then there exists a unique unitary isomorphism U : r s (C) -+ L 2 (I-") satisfying
[U e(z))(x) = e-v'A z(1 + ]x)X for all z E C. (19.13)
Under this isomorphism

U(I,O,O, . .. ) = 1,
U(O,O, ... ,0,1,0, ... ) = (n!)-!7r n (A,X), n = 1,2, ...
where 1 appears in the n-th position and n = 0,1,2, ... , and in particular,
{(n!)-hn(A, x)ln = 0,1,2, ... } is an orthonormal basis in L 2(1-").
Example 19.11: Let {N(t)lt ~ O} be the Poisson process with stationary inde-
pendent increments, right continuous trajectories and intensity parameter A and let
its distribution be described by the probability measure P. Then
P{N(t) - N(s) = j} = e-(t-s) [A(t ~ s)Ji, j = 0, 1,2, ...

J.
° el
and the conditional distribution of the jump points < < 6 < ... < < t, en
given the fact that the process has n jumps in [0, t), is uniform in the simplex
{§.IO < Sl < ... < Sn < t}. This enables us to construct a natural unitary
isomorphism U between rs(L2(1R+)) and L2(P) by putting for every t ~ °
[Ue(ftj))(N) = e-v'A.C I(S)dSIIo~s~dI + [N(S+);XN(S-)) f(s)} (19.14)
(where N(O-) = 0) for all I in L2(~+) and It] is defined as in (19.11). Indeed,
(Ue(ft]), Ue(gt])) =
t 00
(>..t)n
e-V"X Jof [/(s)+g(S)]dS2: e-At - - x
n!
n=O
n!rnl IIJ=I(1 + I(Sj))(1 + g(Sj))dS 1 ••• ds n

O<s,<",<sn<t J>. J>.
= exp{ -J>.lt [](S) + g(s )]ds - >.t + >.It (1 + 1J;! )(1 + gji )ds}
= exp it ](s)g(s)ds = (e(ft]), e(gt])).
In L 2 (P) the totality of the set of random variables occurring on the right hand
side of (19.14) as I varies in L2(~+) and t in ~+ is an exercise for the reader.
Example 19.12: [51] Let (S, CZF, J-L) be a non-atomic, iT-finite and separable mea-
sure space. For any finite set iT C S denote by #iT its cardinality. Let f(S) be the
space of all finite subsets of Sand fn(S) = {iTliT C S,#iT = n}, fo(S) = {0}
where 0 is the empty subset of S. (Thus the empty subset of S is a point in f(S)!).
For any symmetric measurable function I on sn define the function In on f(S)
by
= {1(SI"" ,sn) if ~iT = n,
In (iT) iT - {SI,'" , sn},
o otherwise.
Let CZF r be the smallest iT -algebra which makes all such functions In measurable
for n = 1,2, .•.• Define the measure J-Lr on CZF r as follows. Let ~n C sn denote
the subset {.§. = (SI, ... ,sn)lsi i- Sj for i i- j}. The non-atomicity of J-L implies
that J-L0n(sn\~n) = O. For any E E CZF r put
Then J-Lr is a iT-finite measure whose only atom is 0 with mass at 0 being unity.
(f( S), CZF r , J-Lr) is called the symmetric measure space over (S, CZF, J-L) in the sense
of A. Guichardet [51]. We write dO" = dJ-Lr(iT) when integration is with respect to
J-Lr·
For any function I on S let
if iT = 0,
(19.16)
otherwise.
Then 'Tr1'Trg = 'TrIg and 1f, = 'Try- If f is integrable with respect to JL then (19.15)
implies
r
Jr(s)
'Tr/(a)da = 1 + f ~n. r
n=1 J sn
rrj=d(sj)JL(dst)·.· JL(ds n )
1
(19.17)
= exp fdJL.
Now consider the Hilbert space L2(JLr). It is an exercise for the reader to show
that {'Tr/lf E L 2(JL)} is total in L 2(JL). Equation (19.17) implies that
('Tr I, 'Trg) = J 1fl'Trg( a)da = J 'Tr7g(a)da = exp 1]g dJL.
This at once enables us to see the unitary isomorphism U : f s(L2(JL)) -----+ L 2(JLr)
satisfying the relations
[Ue(f)](a) = 'Tr/(a) for all f in L 2(JL).
Exercise 19.13: Let'Je = en

or £2 according to whether n is the finite or countably
infinite cardinal. Suppose {~j Ij = 1,2, ... } is an n-length sequence of independent
and identically distributed Bernoulli random variables assuming the values ± 1 with
equal probability. Choose and fix an orthonormal basis {el,e2,"'} in 'Je. Then
there exists a unique unitary isomorphism U : fa ('Je) -----+ L2 (P) satisfying
UiI> = 1,
U(kl)I/2e·
• J1 /\e·J2 /\ .. ·/\e·Jk = <"J1<"J2
c. c .... <"Jk'
c. J'I <J'2 < ... <J'k ,
k < 1 + dim'Je where iI> is the vacuum vector and P is the probability measure
of the sequence {~j}. (Hint: Use Proposition 19.3.)
Exercise 19.14: Let JL be any (not necessarily non-atomic!) a-finite measure

in IR+ = [0,00). Define the measurable space (f(IR+), 9F r ) and the symmetric
measure JLr on 9F r as in Example 19.12. Then there exists a unique unitary
isomorphism V: f a (L 2(JL)) -----+ L 2(JLr) satisfying
ViI> = I{0},
if #a =j:. n,
(ViJ /\ 12 /\ ... /\ fn)(a) = {~n!)-~ det((!i(t j ))) if a = {t I, ... ,tn }
< t2 < ... < tn
and tl
for all n ~ 1 and!J, h ... ,fn in L (JL). (Hint: See Proposition 19.2 and (17.10).)
2
Exercise 19.15: [116] In Exercise 19.14 let JL be Lebesgue measure. Then

00
f Ir(L2(1R+)) = e Ell EBL2(1R~).

n=1
There exists a unique unitary isomorphism V : r !r(L2(~+)) --+ L2(p,r) satisfying

Vel> = I{0},
0 if #cr -::j:. n
(V 1) (cr) = { f (t 1 , t2 - t 1, . . . ,tn - tn - J) if cr = {t 1, t 2, . . . ,tn}
and t 1 < t2 < ... < tn
for all n 2: 1 and f E L2(~~).
Notes
Fock spaces were introduced by Fock [43] in his investigations of quantum electro-
dynamics. For a mathematical account of this topic, see Cook [26]. The connection
between Gaussian processes and Fock space appears in Segal [122]. Example 19.12
is from Guichardet [51] and Maassen [83]. This identification of Fock space is the
key idea on which Maassen's kernel approach to quantum stochastic calculus [83,
76] and Meyer's investigations of chaos [88] are based. For further applications
of this idea of kernels, see Parthasarathy [110], Lindsay and Parthasarathy [79].
20 The Weyl Representation

In Section 13 we had already remarked that the route to construct observables
lies in looking at unitary representations of Lie groups and evaluating the Stone
generators of their restrictions to one parameter subgroups. Any Hilbert space '1Jf,
being a vector space, is an additive group. Thanks to the existence of a scalar
product in '1Jf, we have the group 0fL('1Jf) of all unitary operators in '1Jf. A pair
(u, U), U E '1Jf, U E 0fL('1Jf) acts on any element v as follows:
(u,U)v=Uv+u,
first through a "rotation" by U and then a "translation" by u. The map v --+ (u, U)v
is a homeomorphism of '1Jf with inverse being given by the action of the pair
(-U-1u, U- 1). Successive applications by the pairs (U2' U2) and (uj, UJ) on v
leads to
(Ul, Ul){(U2, U2)V} = (Ul' Ul){U2V + U2}
= U1U2V + U1U2 + Ul
= ((Ul + U1U2),U1U2)V.
This suggests the following composition for the pairs (Uj, Uj ), j = 1, 2.
(Ul,Ul)(U2,U2) = (Ul + U1U2,U1U2). (20.1 )
The cartesian product '1Jf x 0fL('1Jf) as a set becomes a group with multiplication
defined by (20.1), identity element (0,1) and inverse (u, U)-l = (-U-1u, U- 1).
'1Jf inherits a topology from its norm and 0fL('1Jf) the corresponding strong topology.
With the product topology and group operation (20.1), '1Jf x 0fL('1Jf) becomes a
topological group which we denote by E('1Jf) and call the Euclidean group over
20 The Weyl Representation 135
'Je. The separability of 'Je implies that E('Je) is, indeed, a complete and separable
metric group. If d = dim'Je < 00 then E('Je) is a connected Lie group of dimension
n(n + 2). In any case E('Je) has a rich supply of one parameter subgroups. Since
(u, U)(v, l)(u, U)-l = (Uv, 1), 'Je is a normal subgroup of E('Je) if we identify
any element u E 'Je as (u, 1) in E('Je). The quotient group E('Je) j'Je is isomorphic
to OU('Je). Any element U in OU('Je) can be identified with (0, U) in E('Je).
We shall now construct a certain canonical (projective) unitary representation
of E('Je) in the boson Fock space f s ('Je) over 'Je and reap a rich harvest of
observables which constitute the building blocks of quantum stochastic calculus.
Let S = {o:e(v)lo: E lC,v E 'Je}. By Proposition 19.4 S is total in fs('Je).
Consider, for any fixed (u, U) E E('Je), the action induced on S by the map
e( v) --+ e(U v + u). This is not inner product preserving. Indeed, for any VI, V2 in
'Je
This shows that the correspondence

1
e( v) --+ {exp( -2:lluI12 - (u, Uv) )}e(Uv + u)
yields an isometry of S onto itself. Hence by Proposition 7.2 there exists a unique
unitary opeator W (u, U) in f s ('Je) satisfying
1 .
W( u, U)e( v) = {exp( -2:lluI12 - (u, Uv) )}e(Uv + u) for all v m 'Je. (20.2)
W( u, U) is called the Weyl operator associated with the pair (u, U) in E('Je).
Proposition 20.1: The correspondence (u, U) --+ W( u, U) from E('Je) into
OU(f s ('Je)) is strongly continuous. Furthermore
W(Ul, Ut)W(U2, U2) = e-iIm(ul,Ulu2)W((Ul, Ut)(U2, U2))
(20.3)
for all (Uj, Uj ) in E('Je), j = 1,2.
Proof: Since the scalar product in 'Je is continuous in its arguments and the map
u --+ e(u) is also continuous (see the proof of Corollary 19.5) it follows from
(20.2) that (u,U) --+ W(u,U)e(v) is a continuous map from E('Je) into fs('Je)
for every fixed v. The totality of exponential vectors and the unitarity of Weyl
operators imply the first part. The second part of the proposition is immediate from
(20.2) on successive applications of W( U2, U2) and W( Ul, Uj ) on an exponential
vector. •
Proposition 20.1 shows that the correspondence (u, U) --+ W (u, U) is a
homomorphism from E('Je) into OU(f s ('Je)) modulo a phase-factor of modulus
unity. Such a correspondence is called a projective unitary representation [135].
As a special case of (20.3) we obtain the following relations by putting
W(u) = W(u,l), f(U) = W(O,U) (20.4 )
for u in 'Je and U in OU('Je):

W(u)W(v) = e-i1m(u,v)W(u + v), (20.5)
W(u)W(v) = W(v)W(u){exp-2iIm(u,v)}, (20.6)
r(U)r(V) = r(UV), (20.7)
r(U)W(u)r(U)-1 = W(Uu), (20,8)
W(su)W(tu) = W(s + tu), s, t E R (20,9)
It is fruitful to compare (20.6) with the Weyl commutation relations in (13.7) of
Example 13.3. Equation (20.9) shows that every element u in 'Je yields a one
parameter unitary group {W (tu) It E Iffi} and hence an observable p( u) through
its Stone generator so that
W(tu) = e-itp(u),t E Iffi,u E 'Je. (20.10)
If S is any completely real subspace of'Je then (u, v) is real for all u, v in S and
the family {p(u)lu E S} is commutative (or non-interfering) in the sense that all
their spectral projections commute with each other.
The operator r(U) defined by (20.4) is called the second quantization of U.
Equation (20.7) shows that for every one parameter unitary group Ut = e- itH in
'Je there corresponds a one parameter unitary group {r(Ut)lt E Iffi} in rs('Je). We
denote its Stone generator by >"(H) so that
r(e- itH ) = e-it>'(H), t ER (20.11)
The observable >"(H) is called the differential second quantization of H.
Through (20.10) and (20.11) we thus obtain the families {p(u)lu E 'Je},
{>..(H)IH an observable in 'Je} of observables in rs('Je). The quantum stochastic
calculus that we develop in the sequel depends very much on the basic properties
of these observables. We shall now investigate the commutation relations obeyed
by them.
Proposition 20.2: For an arbitrary finite set {VI, V2, ... ,Vn } C 'Je the map ~ --+
e(zlvi + ... + znvn) from Cn into rs('Je) is analytic.
Proof: Since L::=olzin I~ < 00 it follows that the map z --+ e(zv) =
L::=ozn ~ is analytic so that the proposition holds for n = 1. When n >
1 choose an orthonormal basis {UI' U2, ... ,um } for the subspace spanned by
{VI,V2,'" ,vn } and note that ~iZiVi = ~jLj(~)uj where Lj(~) = ~iZi(Uj,Vi)
is linear in ~ for each j. Denote by 'Jej the one dimensional subspace CUj for
1 ~ j ~ m and put 'Jem +l = {Uh . .. ,um}l... Since 'Je = EBJ!=~I'Jej we can, by
Proposition 19.6, identify rs('Je) with the tensor product Q9J!=~lrs('Jej) so that
m
e(I>ivi) = {Qge(Lj(~)uj)} ® e(O)
i j=l
and the required analyticity follows from the case n = 1.
•
Proposition 20.3: For any UI, U2, ... ,um, VI, V2, ... ,vn, V in '!Je the map (§.,t) ---+
W(SIUt)··· W(smUm)e(tlvl + hV2 + ... + tnvn + v) from IRm+n intoTs('!Je)
is analytic.
Proof: By (20.2) and (20,S) we have

n m n
W(SIUt)··· W(SmUm)e(2::)jvj + v) = ¢(§.,t)e(Lsiui + LtjVj + v)
j=1 i=1 j=1
where ¢(§., D is the exponential of a second degree polynomial in the variables

Si, tj' 1 :::; i :::; m, 1 :::; j :::; n. The required result is immediate from Proposition
20.2, •
For any set S c '!Je recall (from Corollary 19.5) that e.g(S) denotes the linear
manifold generated by {e(u)lu E S}. When S = '!Je and there is no confusion we
write e.g = e.g('!Je).
Proposition 20.4: For each u in '!Je let p( u) be the observable defined by (20.10),
Then the following holds:
(i) e.g C D(p( UI )p( U2) ... p( un)) for all nand UI, U2, . .. ,Un E '!Je;
(ii) e.g is a core for p( u) for any u in '!Je;
(iii) [P(u),p(v)]e(w) = {2iIm(u,v)}e(w). (20.12)
Proof: (i) is immediate from Proposition 20.3 by putting tl = t2 = ... = 0,

applying Stone's Theorem (Theorem 13.1) and differentiating successively with
respect to Sm, Sm-I, ... ,SI at the origin. To prove (ii) first observe that for any real
S =F 0, p( su) = sp( u) and hence we may assume without loss of generality that
lIull = 1. Let '!Jeo = Cu, '!Je l = '!Jet. By Proposition 19.6 fs('!Je) = fs('!Jeo) 0
f s('!Jet) and for any v E '!Je
e(v) = e( (u, v)u) 0 e(v - (u, v)u),

W(tu)e(v) = e-~t2-t(u.v)e(v + tu)
= {Wo(tu)e( (u, v)u)} 0 e( v - (u, v)u)
where Wo indicates Weyl operator in fs('!Je o). The totality of exponential vectors
implies that
W(tu) = Wo(tu) 01,
1 denoting the identity operator in f s ('!Je I)' Thus
p(u) = Po(u) 01
where Po(u) is the Stone generator of {Wo(tu)lt E IR} in fs('!Jeo). In other words
it suffices to prove (ii) when dim'!Je = 1 or '!Je = C. Let U denote multipliction by
eiO in C. Then f(U)Wo(u)f(U)-1 = Wo(eiOu) and f(U)e(v) = e(eiOv). Thus
r(U)Po(u)r(u)-1 = po(eiBu). Hence it is enough to prove (ii) when 'fie = C and

u = 1. Write Po = Po(l). We have
rs(C)=CEBCEB'"
where any element 't/J can be expressed as
't/J = (zo,zJ, ... ,zn, ... ), Zj E C, ~)zjI2 < 00. (20.13)

j
Let't/J E D(po) be such that ('t/J,Po't/J) is orthogonal to every vector (e(x), poe(x)),
x E IR in rs(C) EB rs(C). Then
('t/J,e(x)) + (Po't/J,poe(x)) = O.
By (i), e(x) E D(ro) and hence
('t/J, (1 + p~)e(x)) = 0 for all x E IR, (20.14)
where
x2 xn
e(x) = (l,x, MI"'" CO'" .).
v2! vn!
Since by definition
(20.13) and (20.14) yield
(2 - X
2)~ zn n+2~nZn n ~n(n-1)_ X n-2_0
L..J-x L..J-x - L..J Zn -. (20.15)
n=oVn! n=lVn! n=2 Vn!
Putting
(20.15) becomes
(2 - x 2)f(x) + 2x!'(x) - !,,(x) = 0
or
+ {3e- X )e 2X
1 2
f(x) = (ae X
where a,{3 are scalars. This shows that
zn = a + ~)n{3lE(e + l)n
n.
where eis a standard normal random variable. Now (20.13) implies
~ la + (-1)n{312 {1E(e + 1)n}2 < 00. (20.16)
L..J n!
n=O
Let n = 2k. Then

IE(~ + Ifk 2:: lEek = 1.3.5··· 2k - 1.
By Stirling's formula
where c > 0 is a constant and !::::: denotes asymptotic equality. Thus
and (20.16) is possible only if 0: + f3 = O. Now consider n = 2k + 1. Then

IE(~ + 1)2k+1 2:: (2k + l)lEe k = 1.3.5··· (2k + 1)
and
{1E(~+1?k+1}2 >(2k 1)(1.3·5 ... 2k-l?!::::: Jk k-t

2k + 1! - + 2k! c as 00
for some c > O. Thus (20.16) is possible only if 0: - f3 = O. In other words

0: = f3 = 0 and hence f(x) = O. Thus 'IjJ = 0 or G(Po) n {(e(x),poe(x))lx E
~}~ = 0, G(Po) denoting the graph of Po. This proves (ii).
To prove (iii) we observe that for any u, v, Wh W2 in '[JC
(p( u)e( wt),p( v)e( W2)) =

fj2
8sOt (W(su)e(wt), W(tv)e(w2))ls=t=o
82 12 212 2 -- (20.17)
= 8sOt exp{ -2: s lIull - 2:t Ilvll - s(u, WI)
-t(VI' W2) + (WI + su, W2 + tv) }Is=t=o
= {(u,v) + ((U,W2) - (WhU))((WI,t? _(v,w~)}e(Wl'W2).
Interchanging u and v in this equation, using (i) of the proposition and the totality
of exponential vectors we obtain
{p(u)p(v) - p(v)p(u)}e(w2) = ((u,v) - (v,u))e(w2).
•
Corollary 20.5: Let S c 'lIC be any dense set. Then p( u) is essentially selfadjoint
in the domain ~(S).
Proof: For any v, w in '!K define
B(v, w) = Ile(v) - e(w)112 + IIp(u)(e(v) - e(w)112

= Ile(v) - e(w)112
82
+ 8s8t (W(su)(e(v) - e(w)), W(tu)(e(v) - e(w)))ls=t=o
Then B (v, w) is a continuous function of v and w. Thus any element of the form
(e(w),p(u)e(w)) in G(p(u)) can be approximated by a sequence of the form
{(e(vn),p(u)e(v n ))} when Vn E S. The rest is immediate from (ii) in Proposition
20.4. •
Corollary 20.6: The linear manifold of all finite particle vectors in r s(7JC) is a
core for every observable p( u), u E '!K.
Proof: This follows easily from Proposition 20.3, 20.4.

•
Proposition 20.7: For any observable H in '!K let >"(H) denote its differential
second quantization in r s ('!K). Then the following holds;
(i) <&(D(H)) C D(>"(H));
(ii) <&(D(H2)) is a core for >"(H);
(iii) For any two bounded observables HI, H2 in '!K and any v in 7JC
i[>"(Hr),>"(H2)]e(v) = >..(i[HI ,H2])e(v).
Proof: Let u E D(H). Stone's Theorem implies that the map t ~ e- itH u is
differentiable. By Proposition 20.2 the map t ~ e-it>.( H) e(u) = e(e -itH u) is
differentiable. Hence e(u) E D(>"(H)). This proves (i).
If v E D(H2) then t ~ e-itHv is twice differentiable and Proposition 20.2
implies that t ~ e-it>.(H)e(v) is a twice differentiable map and hence e(v) E
D(>..(H)2). Now we proceed as in the proof of (ii) in Proposition 20.4. Let 'IjJ E
D(>..(H)) be such that ('IjJ, >"(H)'IjJ) is orthogonal to (e(v), >"(H)e(v)) for all v E
D(H2). Then
('IjJ, e(v)) + (>'(H)'IjJ, >'(H)e(v)) = O.
Since e(v) E D(>.(H?) we have
('IjJ, {I + >'(H?}e(v)) = 0 for all v E D(H2). (20.18)
r
Let 'if n denote the n-particle subspace in s(7JC) and let En be the projection
on 'if n. Then >"(H) = EB~=o>"n(H) where >'n(H) is the generator of the n-fold
tensor product e- itH Q9 ••. Q9 e- itH restricted to 'if n. Since e( tv) = EB~=otn ~
we obtain by changing v to tv in (20.18)
-
(En'IjJ, (1 + >'n(H))v
2 0 n
) = 0, n = 1,2, ... , v E D(H ).
2
A polarisation argument yields
(20.19)
1:::; j :::; n.
where En denotes the symmetrization projection. Since the linear manifold gen-
erated by {EnvI 161"'161 vnlVj E D(H2), 1 :::; j :::; n} is a core for 1 + >'n(H?,
(20.19) implies
Since 1 +>'n(H? has a bounded inverse (thanks to the spectral theorem) its range
is 'iien . Thus En'IjJ = 0 for all n. In other words
G(>.(H)) n {(e(v), >'(H)e(v))lv E D(H2)}~ = O.

This proves (ii).
To prove (iii) we first observe by using (i)
(e(u),>'(H)e(v)) = i :t (e(u),e(e-itHv))lt=o
v It=o
d exp (u, e -itH) (20.20)
= z. dt
= (u, Hv)e(u,v)
for any observable H in '3C and v E D(H). If HI, H2 are bounded observables
the map (s, t) -+ e- isH1 e- itH2 v is analytic for every v E '3C and, in particular, by
Proposition 20.2 the map (s, t) -+ e( e- isH1 e- itH2 v) is differentiable and
>'(HI )>'(H2 )e(v) 82

= _ _ e(e- ZS
. Hle- z'tH2v)1 __
8s8t s-t-o·
Thus for any u, v E '3C,
(e(u), >'(HI}>'(H2)e(v)) = ;s~2t exp(eiSH1u, e- itH2 v)ls=t=o

= {(u,HIH2V) + (u,H l v)(u,H2 v )}exp(u,v).
Now the totality of exponential vectors and (20.20) imply (iii).
•
Proposition 20.8: Let H be an observable in '3C and u, v E D(H2). Then
i[P(u), >'(H)]e(v) = -p(iHu)e(v).

Proof: Let U,V,W E D(H2), Ut = e- itH . Then the ~-valued functions
f(Ut)W(su)e(w) = e-!s21I uI1 2- s(u,w)e(Ut(su + w)), (20.21)
W(su)f(Ut-1)e(w) = e-!s21IuI12-S(UtU,W)e(Ut-lw + su), (20.22)

f(Ut)W(su)f(Ut-1)e(w) = W(sUtu)e(w)
= e-!s2I1 uIl 2- s(Ut U,w) e( w + sUtw) (20.23)
are twice differentiable in (s, t). Hence by Stone's Theorem and (20.23) we have
02
{)tas (e( v), f(Ut)W(su)f(Ut- 1)e( w))ls=t=o
(20.24)
= {(v,-iHu) - (-iHu,w)}e(v,w).
On the other hand the left hand side of the above equation can be written as
82
as{)t (f(Ut- 1)e(v), W(su)f(Ut- 1)e( w))ls=t=o
= :s {(iA(H)e(v), W(su)e(w)) + (e(v), W(SU)iA(H)e(w))}ls=o (20.25)
= (e(v), {p(u)A(H) - A(H)p(u)}e(w)).

Furthermore, for any u, v, w in ~
(e(v),p(u)e(w)) = i! (e(v), W(tu)e(w))lt=o
= i! exp{ _~t2I1uI12 - t(u, w) + (v, w + tu) }It=o (20.26)
= i( (v, u) - (u, w)) exp(v, w).

Equating the right hand side expressions in (20.24) and (20.25) and using (20.26)
we obtain
i[P(U),A(H)]e(w) = -p(iHu)e(w). •
Proposition 20.9: Let T be any bounded operator in fs('~e) such that TW(u) =
W (u)T for all u in 'Je. Then T is a scalar multiple of the identity.
Proof: Without loss of generality we assume that ~ = £2. The same proof will
go through when dim ~ < 00. Consider the unitary isomorphism U : f s (~) ---t
L 2 (P) discussed in Example 19.8 where P is the probability measure of an in-
dependent and identically distributed sequence of standard Guassian random vari-
ables {= (e106, ... ). Then
1
[UeC~)]({) = exp 2)Zjej - 2zj ).
j
Elementary computation shows that the Weyl operators obey the following rela-
tions thanks to the totality of exponential vectors in f s('Je) and the set
{exp 2:>j~jIQ E £2} in L 2 (P) :
j
for any square summable real sequence Q
{UW( -iQ)U- 1f}({) = e -i L j aj~j !({) (20.27)
{UW( - ~Q)U-I f}({) = e -~ ILI!I12_~ L j ajEj !({ + Q) (20.28)
for all ! E L 2 (P). Let UTU- 1 = S. Then S commutes with UW( -iQ)U- 1
and UW( _~Q)U-l for all real square summable sequences Q. Equation (20.27)
implies that S commutes with the operator of multiplication of ¢(~) for every
bounded random variable ¢({). In particular, for any indicator random variable
IE we have
SIE = SIE 1 = IESl.
If Sl = 'I/J then we conclude that (Sf)({) = 'I/J({)!({) for every! in L 2 (P). Now
using the commutativity of S with the operators defined by (20.28) we conclude
that
'I/J({ + Q)!({ + Q) = 'I/J({)!({ + Q) a.e. {
for each real square summable sequence Q and ! in L 2 (P). Thus
'I/J({ + Q) = 'I/J({) a.e. {
for each Q of the form (ai, a2, . .. ,an, 0, 0, ... ), aj E R This means that 'I/J(~) is
independent of 6,6, ... ,~n for each n. An application of Kolmogorov's 0-1 law
shows that 'I/J is a constant. •
Suppose {Ta la E J} is a family of bounded operators in a Hilbert space
'Je, which is closed under the adjoint operation *. Such a family is said to be
irreducible if for any bounded operator S the identity STa = TaS for all a E J
implies that S is a scalar multiple of the identity. For such an irreducible family
the only closed subspaces invariant under all the Ta are either {O} or 'Je. Indeed,
if P is an orthogonal projection on such an invariant subspace then PTa = TaP
for all a. In this sense the family of Weyl operators {W (u) lu E 'Je} is irreducible
in fs('Je) according to Proposition 20.9. We now summarise our conclusions in
the form of a theorem.
Theorem 20.10: Let 'Je be any complex separable Hilbert space and let f s('Je) be
the boson Fock space over 'Je. Let W( u, U), (u, U) E E('Je) be the Weyl operator
defined by (20.2). The mapping (u, U) -+ W (u, U) is a strongly continuous,
irreducible and unitary projective representation of the group E('Je). For any u in
'Je and any observable H on 'Je there exist observables p( u) and ).. (H) satisfying
W(tu) = W(tu, 1) = e-itp(u)
f(e- itH ) = W(O,e- itH ) = e-it)"(H)
for all t in R The observables p( u) and >.( H) obey the following commutation
relations:
(i) [P(u),p(v)]e(w) = 2iIm(u,v)e(w) for all u,v,w in 'Je;
(ii) i[P(u), >'(H)]e(v) = -p(iHu)e(v) for all u, v E D(H2);
(iii) for any two bounded observables H 1, H2 in 'Je and v E 'Je
i[>'(Ht},>'(H2)]e(v) = >'(i[H1,H2])e(v).
Proof: This is contained in Proposition 20.4, 20.7-20.9.

We shall now introduce a family of operators in terms of which computations
•
involving the Weyl operators or, equivalently, the operators p( u) and >'(H) become
considerably simplified. We write
q(u) = -p(iu),a(u) = ~(q(u) + ip(u)),at(u) = ~(q(u) - ip(u)) (20.29)

for any u in 'Je. For any bounded operator H in 'Je we write
>'(H) = >.( ~(H + H*)) + i>'(;i (H - H*)), >. t (H) = >'(H*). (20.30)
Proposition 20.11: Let T be any operator of the form T = T1 T2 ... Tn where

Ti = p( Ui) or >.( Hi) for some ui E 'Je or some bounded observable Hi in 'Je. Then
the linear manifold ~ generated by all the exponential vectors in r s ('Je) satisfies
the relation ~ C D(T).
Proof: The proof is analogous to those of (i) and (iii) in Proposition 20.4, 20.7
and we leave it to the reader. •
Proposition 20.12: Let a( u), at (u), >'(H), >. t (H) be defined as in (20.29), (20.30)
for u E 'JC, H E '1A('Je). For any operator of the form T = T1 T2 ... Tn where each
Tj is one of the operators a(uj),at(uj),>.(Hj),uj E 'Je, Hj E '1A('Je), I ~ j ~ n,
n = 1,2, ... , the relation ~ C D(T) holds. Furthermore, for any 7/J, 7/J[, '1P2 E ~
the following relations hold:
(i) a(u)e(v) = (u,v)e(v);
(ii) (a t (u)7/J1,7/J2) = (7/J1,a(u)7/J2);
(iii) (>.t (H)7/J1' 7/J2) = (7/J1, >'(H)7/J2);
(iv) The restrictions of a( u) and at (u) to ~ are respectively antilinear and linear
in the variable u. The restriction of >'(H) to ~ is linear in the variable H;
[a(u),a(v)]7/J = [at(u), at (v)]7/J = 0,
[a(u), at (v)]7/J = (u,v)7/J,
[>'(Ht) , >.(H2)]7/J = >.([Hl, H2)])7/J,
[a(u), >.(H)]7/J = a(H*u)7/J,
[at(u), >.(H)]7/J = -a t (Hu)7/J.

(e(w), a(u)e(v)) = (u, v)(e(w), e(v)) for all w in 'JC.
Since exponential vectors are total in r s ('JC) we obtain (i). For the remaining parts
we may assume without loss of generality that 'I/J, 'I/J" 'l/J2 are exponential vectors.
Since'f!, C D(p(u)) for any u in 'JC and p(u) is selfadjoint we have
(at(u)e(w),e(v)) = (~( -p(iu) - ip(u))e(w),e(v))
= (e(w), ~(-P(iU) + ip(u))e(v))

= (e(w),a(u)e(v)).
This proves (ii). When H = H*, >.. t (H) = >..( H) and (iii) is trivial. Now (iii) is
immediate from (20.30). The antilinearity of a( u) in the variable u follows from
(i). The linearity of at (u) in u follows from (i) and (ii). When H is selfadjoint
(e(u),>"(H)e(v)) = i(e(u), !e(e-itHv)lt=o)
= i :t exp(u, e- itH v)
= (u, Hv) exp(u, v).

Thus the linearity of >"(H) in H follows from (20.30). This proves (iv). (v) follows
from Theorem 20.10 and definitions (20.29), (20.30). •
Proposition 20.13: The operators at(u), u E 'JC and >..(H), HE 0A('JC) obey the
following relations:
(i) (e(v),>"(H)e(w)) = (v,Hw)e(v,w);
(ii) + (u"u2)}e(v,w);
(a t (u,)e(v),a t (u2)e(w)) = {(u"W)(V,U2)
(iii) (at(u)e(v),>"(H)e(w)) = {(u,w)(v,Hw) + (u,Hw)}e(v,w);
(iv) (>"(H,)e(v),>"(H2)e(w)) = {(H,v,w)(v,H2w) + (H,v,H2w)}e(v,w).
Proof: (i) already occurs in the proof of Proposition 20.12. (ii) follows from
(i), (ii) and the second commutation relation in (v) of Proposition 20.12. (iii)
follows from the fourth commutation relation of (v) and (i) in Proposition 20.12
and property (i) of the present proposition. (iv) results from
fj2 .H 'tH
(>"(HJ)e(v),>"(H2)e(w)) = asat exp(e-~S Iv,e-~ w)ls=t=o
2
when H" H2 are selfadjoint and the general case from (20.30).
We now make a notational remark in the context of Dirac's bra and ket
•
symbols. Write
a(u) = a((ul), at(u) = at(lu)).
Then we have the suggestive relations:

a((ul)e(v) = (u,v)e(v),
[a((ul),at(lv))] = (u,v),
[a((ul),A(H)] = a((uIH) = a((H*ul),
[A(H),at(lu))] = at(Hlu)) = at(IHu))
in the domain 'B. The remaining relations in Proposition 20.12 and 20.13 also
acquire a natural significance especially in the context of the quantum stochastic
calculus that will be developed later.
Proposition 20.14: The operators a(u),at(u),u E '3e and A(H),H E gj('3e)
satisfy the following relations:
(i) at(u)e(v) = fhe(v + tu)lt=o;
(ii) the linear manifold of all finite particle vectors is contained in the domains
of a(u),at(u) and A(H). Furthermore
a(u)e(O) = 0,a(u)v0 n = vn(u,V)v 0n - 1ifn 2: 1 (20.31)

n
at (u)v0 n = (n+ 1)-~Lv0r 0u0v 0n - r , (20.32)
r=O
n-l
A(H)v 0n = Lvr 0Hv0v n - r - 1• (20.33)
r=O
Proof: (i) follows from the two relations

(at(u)e(v),e(w)) = (e(v),a(u)e(w)) = (u,w)e(v,w),
d d
(dte(v+tu)lt=o,e(w)) = dte(v+tu,W)lt=o = (u,w)e(v,w)
and the totality of exponential vectors. The first part of (ii) is an easy conse-
quence of Stone's Theorem and Proposition 20.2. By (i) in Proposition 20.12
a(u)e(tv) = t(u,v)e(tv) for all t E C. Identifying coefficients of t n on both
on
sides of this equation we obtain (20.31). Expanding e(sv +tu) as EB:=o (svj;H ,
differentiating at t = 0 and using (i) we obtain (20.32). When H is selfadjoint
e(e-itHv) = e-it>.(H)e(v) = EB:=o(e-i:~~eon. Change v to sv and identify
the coefficients of sntO. Then we get (20.33). When H is not selfadjoint (20.33)
follows from (20.30). •
Proposition 20.15: In the domain 18 the Weyl operator W( u, U) admits the
factorisation:
W(u, U) = e-~lluI12 eat(u)r(U)e-a(U-1u) (20.34)
for all u E '3e, U E OU('3e).
20 The WeyJ Representation 147
Proof: By (i) in Proposition 20.12 we have

e-a(U-1u)e(v) = e-(U-1u,v)e(v) = e-(u,Uv)e(v).
By (20.2) and (20.4)
r(U)e-a(U-1U)e( v) = e-(u,uv) e(Uv).
By (i) in Proposition 20.14 and (20.2) we get
eat(u)r(U)e-a(U-1u)e(v) = e-(u,Uv)e(Uv + u)
= e~lluI12W(u, U)e(v).
•
Proposition 20.14 shows that a((u)e(O) = 0 and a(u) transforms an n-
particle vector into an (n - 1)-particle vector whereas at (u) sends an n-particle
vector into an (n + 1)-particle vector. >"(H) leaves the n-particle subspace invari-
ant. In view of these properties we call a( u) the annihilation operator associated
with u and at (u) the creation operator associated with u. >.. (H) is called the con-
servation operator associated with H. The quantum stochastic calculus that we
shall develop in the sequel will depend heavily on the properties of these operators
described in Proposition 20.12-20.15.
Exercise 20.16: (i) Let G be a connected Lie group with Lie algebra Cfj. Suppose
g ---+ Ug is a unitary representation of G in '3e. For any X E Cfj let 1l'(X) denote
the Stone generator of {UexptX It E ~}. Let
D = {v E '3elthe map g ---+ Ugv from G into '3e is infinitely differentiable}.
Then
i[>..(1l'(X)), >"(1l'(Y))]e(v) = ->"(1l'([X, Y]))e(v)
for all v E D, X, Y E Cfj.
(ii) Let a : Cfj ---+ '3e be a linear map satisfying
1l'(X)a(Y) - 1l'(Y)a(X) = ia([X, Y]) for all X, Y E Cfj. (20.35)

Define
q,(X) = at(a(X)) - a(a(X)) - i>..(1l'(X)).
Then
[q,(X), q,(y)] = q,([X, Y]) - 2i Im(a(X), a(Y))
in the domain cg (D).
(iii) If u E D then a(X) = 1l'(X)u satisfies (20.35). If 1lI(X) = q,(X) -
i(u,1l'(X)u) then
[1lI(X),1lI(Y)] = 1lI([X, Y]) for all X, Y E Cfj.
Exercise 20.17: Let h = ffi~=l hk be a direct sum of Hilbert spaces {hk} and let
D = {ulu E h, u = EBkUb Uk = 0 for all but a finite number of k' s}.
Suppose T is an operator on h with domain D such that Tlhk is bounded for
every k and T( h k ) ~ ffij:::,:nk hj where nk --+ 00 as k --+ 00. Then T is closable.
In particular, a( u) and at (u) restricted to the domain of finite particle vectors are
closable and at(u) C a(u)*.
Exercise 20.18: (a) Let {ej Ij = 1,2, ... ,} be an orthonormal basis in 'iff. Define
on the domain f~('iff) of all finite particle vectors the operators aj and a} as the
restrictions of a( ej) and at (ej) respectively for each j = 1,2, .... Let
qj = T1 (aj + aj),pj = -iT1 (aj - aj).
Then qj,Pj are essentially selfadjoint in f~('iff). Furthermore
[aj,akl = [a},akl = 0, [aj,akl = Ojb

[qj, qkl = [Pj,Pkl = 0, [qj,Pk) = iOjk.
(b) When 'X = C,ef = ell, e~j = j ii, = 1,2, ... then {eIl,fI,h, ... } is an
orthonormal basis for fs(C). If al = a, at = at then the matrices of a and at
are respectively
0 1 0 0 0
0 0 ,ji 0 0
a=
0 0 0 v'3 0
0 0 0 0 0 Vn
0 0 0 0
1 0 0 0
0 ,ji 0 0
at = 0 0 v'3 0
0 0 0 Vn
where Ij is considered as a column vector with 1 in the j-th position and 0
elsewhere and 10 = ell. This may be expressed in Dirac's notation as
00 00
a= L JJlii-l)(fjl, at = L JJlii)(ii-ll·
j=l j=l
The operator ata = N is called the number operator because the j-th par-
ticle vector with 1 in the j -th position and 0 elsewhere is an eigenvector for the
eigenvalue j, fo denoting <P. If L denotes the coisometry 2:j:llfJ-l) (fJ I we can

express
a = IN + IL, at = L*JN + 1
on the domain of finite particle vectors. In the pure state <P the observable ~ (L+ L *)
has standard Wigner distribution whereas the closure of a+a* has standard normal
distribution. (See Exercise 4.5, 6.3.)
Exercise 20.19: Let H be a selfadjoint operator in 'Je with a complete orthonormal
eigenbasis {ej Ij = 1,2, ... } satisfying H ej = Ajej for each j. Then its differential
second quantization A(H) in r s ('Je) has the orthonormal eigenbasis
{[h +, ... + r k )!] 1/2 e~le~2

rl···· rk!
... e~klrj 2: 0, j = 1,2, ... , k, k = 1,2, ... }
(20.36)
(expressed in the notation of Proposition 19.3) with corresponding eigenvalues
{riAl + ... + rkAk}. In the linear manifold generated by the set (20.36) A(H)
coincides with the (formal) sum 2:jAja}aj where aj,a} are as in Exercise 20.18.
Exercise 20.20: (a) There exists a unique unitary isomorphism U : rs(C) --+
L2(1R) satisfying
x2 Z2
{Ue(z)}(x) = (21f)-;; exp -("4 - zx + 2) for all z
1
E IC;
(b) Let De, () > 0 denote the unitary dilation operator L2(1R) defined by
(Def)(x) = ()1/2 f(()x), f E L2(1R)
and let Ve = DeU. Then the image under Ve of the domain of all finite particle
vectors in r s (C) is the linear manifold
1 e2 2
.Me = {P(x)e-;; x IP a polynomial in the real variable x}.
If q = 2-1 (a + at),p = -2-~i(a - at) where a, at are as in Exercise 20.18 (b)
then for any f in .Me the following holds:
(i) (VeqVe- 1 f)(x) = 2- 1()xf(x);
(ii) (VepVe- 1 f)(x) = -2-1i()-1 f'(x);
(iii) for () = V2C
1 1 1
{cVy2c(a t a+ 2:)V~j}(x) = -2:f"(x) + 2:C2X2f(x).
(c) The functions
(21f)-;;(j!)-zHj (v2cx)e- zX ,
1 I ~ 12
'l/Jj(x) = j = 0,1,2, ... ,
where {Hj} is the sequence of Hermitian polynomials as defined by (19.7) satisfy
1 "()
-2:'l/Jj x +2:1 cx
2 2 1
'I/Jj ( x) =c(J. + 2:)'l/Jj ( x), j=0,1,2, ...
(Hint: Use the properties of the number operator at a defined in Exercise
20.18 (b).)
(d) Uf( -i)U- 1 = DI/21F where f( -i) is the second quantization of multiplication
by -i in fs(C) and IF is the unitary Fourier transform defined by (12.8);
(e) IF Do = DO-I IF, DOl D02 = DO I02 and the Stone generator of the unitary group
{Detlt E /R} is the closure of ~U(qp+ pq)U-I,q,p being as in (b) (i) and (ii).
Exercise 20.21: Let the Hilbert space ';Je be a direct sum: ';Je = ';Jel EEl ';Je2. In the
factorisation f s(';Je) = f s(';Jet) (>9 f s(';Je2) determined by Proposition 19.6 the Weyl
operators W( u, U) and the observables p( u), A(H) defined by (20.2), (20.10) and
(20.11) satisfy the following:
(i) W(UI EEl U2,U I EEl U2) = W(u[,Ut} (>9 W(U2,U2) for all Uj E ';Jej, Uj E
OU(';Jej), j = 1,2;
(ii) P(UI EEl u2)e(vl EEl V2) = {p(ut}e(vt}} (>9 e(v2) + e(vI) (>9 {p(u2)e(v2)} for
all Uj,Vj E ';Jej, j = 1,2;
(iii) for any two observables Hj in ';Jej, j = 1,2
A(HI EEl H2)e(vl EEl V2) = {>..(Ht}e(vt}} (>9 e(v2) + e(vt) (>9 {>..(H2)e(v2)}
for all Vj E ';Jej, j = 1,2, ....
Exercise 20.22: (i) Let ~(';Je) C 7J3(';Je) be the set of all contraction operators.
Then ~(';Je) is a *-weakly closed convex set. In particular, ~(';Je) is strongly closed.
Under the strong topology ~(';Je) is a multiplicative topological semigroup. If
T E ~(';1f) and T leaves a subspace ';1fo invariant then TI:lfo is a contraction.
Direct sums and tensor products of contractions are also contractions.
(ii) For any T E ~(';1f) the operator f,r (T) in the free Fock space f,r (';Je)
is defined by
Write
fs(T) = f/r(T)lrs(:If),fa(T) = f/r(T)lra(:If)
where f s(';Je) and fa (';Je) are respectively the boson and fermion Fock spaces over
';Je. Then the maps T ~ f Ir(T), f s(T), f a(T) are strongly continuous *-unital ho-
momorphisms from the topological semigroup ~ (';Je) into ~ (f Ir (';Je)), ~ (f s (';Je)),
~ (f a(';Je)) respectively. These are called second quantization homomorphisms.
Second quantization homomorphisms are positivity-preserving.
(iii) If T is a trace class operator in ';Je such that IITIII < 1 then f Ir(T) is also
a trace class operator and Ilf/r(T)III = (1 -IITIIt)-I, trf Ir(T) = (1 - trT)-I.
(iv) Let T be a positive operator of finite trace with eigenvalues {Aj Ij =
1, 2, ... } inclusive of multiplicity and supj IAj I < 1. Then
trfs(T) = IIj(1 - Aj)-l, trfa(T) = IIj(1 + Aj).

In particular, {trfs(T)}-lfs(T) and {trfa(T)}-lfa(T) are states in fs(';Je) and
f a(';Je) respectively.
20 The Weyl Representation lSI
(v) Let dim'Jf < 00 and let p be any positive operator in 'Jf such
that Ilpll < 1. Then for any observable X in 'Jf,det(1 - pe itX )-I(1 - p) and
det( 1 + pe itX ) (1 + p) -I are characteristic functions of probability distributions.
(Hint: Use Proposition 19.3 for computing traces.)
Exercise 20.23: [83] Let (8,?1', fL) and fLr be as in Example 19.12. Write da for
fLr(da). Then for any f E L2(fLr x fLr x··· x fLr), where the product is n-fold,
the following holds:
r
Jr(s)n
f(al, a2, ... , an)dal ... dan = r
Jr(S)
L
0'1 U0'2U",UO'n=0'
f(alo ... ,an)da
where U signifies disjoint union. (This is known as the sum-integral formula.)

Under the unitary isomorphism U : rs(L2(fl,)) --+ L2(fLr) for any f E UCf,),
u E L2(fL)
sEO'
(Ua(u)U-If)(a) = is u(s)f(aU{s})dfL(S)
and for any real valued ?1' measurable function ¢ on 8
(Ur(ei¢)U- 1f)(a) = f(a) expi L ¢(s) for all f E L 2(fLr)

sEO'
where ei ¢ denotes the unitary operator of multiplication by ei ¢ in L2(fL).
Exercise 20.24: [136] For any u E 'Jf there exists a unique bounded operator
r Ir ('Jf) satisfying the relations:
£( u) in the free Fock space
= 0, £( U )VI 0 ... 0 Vn = (u, VI}V2 0 ... 0 Vn for all n 2': 1, Vj E 'Jf
£( u)
(where V2 0 ... 0 Vn = when n = 1). The adjoint £* (u) of £( u) satisfies the
relations:
£* (u) = u, £* (u )Vl 0 ... 0 vn =U 0 VI 0 ... 0 Vn for all n 2': 1, Vj E 'Jf.
If T is a bounded operator in 'Jf then there exists a unique bounded operator Ao(T)
in r Ir ('Jf) satisfying
Ao(T) = 0, Ao(T)VI0"'0 vn = (TvI) 0 v20···0 Vn for all n 2': 1, Vj E 'Jf.
The operators £(u),£*(u) and Ao(T) satisfy the following:
(i) £(u)£*(v) = (u,v);
(ii) 11£(u)11 = Ill*(u)11 = Ilull;
(iii) £( u) is antilinear in u whereas £* (u) is linear in u;
(iv) £(u)Ao(T) = £(T*u), Ao(T)£*(u) = £*(Tu);
(v) Ao(T)* = Ao(T*), IIAo(T)11 = IITII;
(vi) AO is a *-(non-unital) homomorphism from CJ3('JC) into CJ3(r IT ('JC));

(vii) If.e( u)'I/J = 0 for all u E 'JC then 'I/J is a scalar multiple of the vacuum vector
11>. If B E CJ3('JC) and B commutes with .e( u) for all u then BII> = bll> for
some scalar b. If, in addition, B commutes with .e* (u) for all u then B = b.
In other words the family {.e( u),.e* (u) lu E 'JC} is irreducible.
(It is instructive to compare the properties of .e( u),.e* (u) and Ao(T) in r IT ('JC) with
the properties of a( u), at (u) and A(T) in r s ('JC). The operators .e( u) and.e* (u) are
respectively called the free annihilation and creation operators associated with u.
Ao(T) is called the free conservation operator associated with T. It is appropriate
to call A(T) the boson conservation operator associated with T).
Exercise 20.25: [116] Consider the unitary isomorphism V : r/T(L2(~+)) ---+

L2(j.tr) where j.t is Lebesgue measure in ~+ and j.tr and V are as in Exercise
19.15. For any a E r(~+), s E ~+ define a + s = {t + sit E a} if a -:j:. 0 and
= 0 if a = 0; a - s = {t - sit E a} if a -:j:. 0 and if every element in a exceeds
s. Denote by min a the smallest element in a. Then
(i) (V.e(U)V-l f)(a) = Jooo u(s)f((a + s) U {s} )ds;
(ii) (V.e*(u)V-1f)(a) = u(mina) f(a\{mina} - mina) if a -:j:. 0 and = 0 if
a = 0;
(iii) For any bounded measurable function <p on ~+
(VAo( <p )V- 1f( a) = <p(min a )f( a) if a -:j:. 0,

= 0 if a = 0,
where, on the left hand side, .(<P)).
Notes
For an extensive discussion of second quantization, CCR and CAR, see Cook
[26], Garding and Wightman [45], [46], Segal [120], [121], Berezin [19], Bratteli
and Robinson [23]. Exercise 20.23 is from Maassen [83]. Exercise 20.24 [136] is
the starting point of a free Fock space stochastic calculus developed by Speicher
[126,127]. Exercise 20.25 is from Parthasarathy and Sinha [116].
21 Weyl Representation and infinitely divisible distributions

Using the Weyl operators defined by (20.2) we constructed the observables
{p(u)1 u E 'JC}, P(H)IH an observable in 'JC} in rs('JC) through (20.10) and
(20.11). We shall now analyse their probability distributions in every pure state of
the form
(21.1 )
Any state of the form (21.1) is called a coherent state (associated with v). Such
an analysis together with the factorisability property indicated in Exercise 20.21
21 Weyl Representation and infinitely divisible distributions 153
shows how one can realise every infinitely divisible probability distribution as the
distribution of an observable in f s ('Je) in the vacuum state. It may be recalled
that a probability distribution It on the real line (or IRn) is called infinitely divis-
ible if for every positive integer k there exists a probability distribution Itk such
that It = Iti/, the k-fold convolution of Itk. The investigation of such distribu-
tions in the present context leads us to the construction of stochastic processes
with independent increments as linear combinations of creation, conservation and
annihilation operators in f s ('Je).
Proposition 21.1: Let S c 'Je be a completely real subspace such that 'Je =
S + is = {u + ivlu E S, v E S}. Then the following holds:
(i) {p(u)lu E S} and {q(u) = -p(iu)lu E S} are two commuting families of
observables with eg as a common core in fs('Je);
(ii) [q(u),p(v)]'l/J = 2i(u,v)'l/J for all u,v E S, 'l/J E eg;
(iii) For any v E 'Je, Ul, ... , un E S the joint distribution of p( uJ), ... , p( un)
in the coherent state 'l/J(v) is Gaussian with mean vector -2(Im(v, Ul),
Im(v, U2), ... , Im(v, un)) and covariance matrix (( (Ui' Uj))), 1 ~ i, j ~ n.
In the same state the joint distribution of q( uJ), ... , q( un) is Gaussian with
mean vector 2(Re(u, Ul), ... , Re(v, un)) and the same covariance matrix.
Proof: For any u, v E S, Im(u, v) = 0 and (20.3) and (20.4) imply

W(u)W(v) = W(u + v) = W(v)W(u),
W(iu)W(iv) = W(i(u + v)) = W(iv)W(iu).
Now (20.10) implies that {p(u)lu E S} and {q(u)lu E S} are commuting families
of observables. That eg is a core for each p( u) and q( u) is just (ii) in Proposition
20.4. This proves (i). Now (iii) in Proposition 20.4 implies that for u, v E S
[q(u),p(v)]'l/J = -[P(iu),p(v)]'l/J
= -2i Im(iu, v) = 2i(u, v)'l/J, 'l/J E eg.
This proves (ii).
For any u, v E 'Je we have from (21.1), (20.2) and (20.4)
1
('l/J(v), W(u)'l/J(v)) = exp( -llvl1 2 - 211ul12 - (u, v) + (v, u + v))
= exp(2i Im(v, u) - ~ IluI1 2 ).

Thus for any Uj E S, tj E R I ~ j ~ n and v E 'Je we have
('l/J(v),e-it1P(uJ) ... e-itnp(un)'l/J(v)) = ('l/J(v), W(2)jUj)'l/J(v))
j
(21.2)
= exp{2i2:)j Im(v, Uj) - ~ 2)itj(Ui, Uj)}
j i,j
and
j
(21.3)
= exp{-2i~)jRe(v,uj) - ~2)itj(Ui'Uj)}'
j i,i
Equations (21.2) and (21.3) express the characteristic functions of the joint distri-
butions of p(ut), ... ,p(u n ) and q(ut), ... ,q(u n ) respectively in the pure state
1jJ( v). •
In Proposition 21.1, S is a real Hilbert space and we have p(a:u + j3v)1jJ =
a:p(u)1jJ + j3p(v)1jJ, q(a:u + j3v)1jJ = a:q(u)1jJ + j3q(v)1jJ for 1jJ E cg, a:,j3 E R
u, v E S. In view of Property (iii) in Proposition 21.1, p(.) and q(.) may be
looked upon as classical Gaussian random fields over the real Hilbert space S in
each coherent state.
Example 21.2: Let SeX be a completely real subspace and let {Utlt E ~} be
a one parameter unitary group in X leaving S invariant. Consider X t = p(Utu)
where u E S is fixed. Then {Xtlt E ~} is a commuting family of observables
whose distribution in the vacuum state yields a stationary Gaussian stochastic
process with mean 0 and spectral distribution (u, ~(E)u) = f.l(E), E ~ ~ being
a Borel set where Ut = f e-itx~(dx), t E Il\t As a special case we may consider
X = L2(~), S = the subspace of all real valued functions in ~,(Ut!)(x) =
f(x - t), f E X. In this case the spectral density function is 1'11,12 where '11, is the
Fourier transform of u.
Example 21.3: Let SeX be a completely real subspace and let ~ be a spectral
measure on the real line for which S is invariant. For any fixed S let X t =
p(~(( -00, tju), F(t) = (u, ~(( -00, t])u). Then {Xtlt E ~} is a commuting family
of observables (with common core cg) whose distribution in the vacuum state is
a Gaussian process with independent increments, mean 0 and cov(Xt, Xs) =
F( t 1\ s), t 1\ s being the minimum of t and s.
One may interpret ~ as a time observable in X. This example is meaningful
when ~ is a spectral measure in any interval of the real line. When ~ is a spectral
measure in the unit interval and (u, ~([O, t])u) = t for all 0 ::; t ::; I then the
distribution of {Xt 10 ::; t ::; I} is the same as that of standard Brownian motion.
Proposition 21.4: Let H be an observable in X with spectral resolution H =

fIR x~(dx) and let f.lv(E) = (v, ~(E)v), E ~ ~ being a Borel set. The character-
istic function of the distribution of >.( H) in the coherent state 1jJ( v) is equal to
exp(v, (e itH - l)v) and the corresponding distribution v is given by
v=e- Il v112{ 80 + f"'V

11.
I *2 + ... +-11.
+-11.
2! f"'V
I *n + ... }
n! f"'V
(21.4 )
where 80 is the Dirac measure at the origin and *n denotes n-fold convolution.
Proof: From (21.1), (20.4) and (20.11) we have

('I/J( v), e-it>..(H)'I/J( v)) = ('I/J( v), r( e-itH)'I/J(v))
= exp{-llvI12 + (v,e-itHv)}.
This proves the first part. On the other hand the last expression is equal to
exp J(e- itx - l)J.tv(dx) which implies (21.4). •
Example 21.5: Let (n,~) be any measurable space and let TJ be an n-valued
observable in 'Je. For any E E ~,TJ(E) is a projection in 'Je and X(E) = )"(TJ(E))
is an observable in r s ('Je). {X (E) IE E ~} is a commuting family of observables
and by Proposition 20.7 has a common core ~. For any sequence {Ej} of disjoint
sets from ~
X(UiEj)'I/J = LX(Ej )'I/J, 'I/J E ~.
j
In any coherent state 'I/J(v), for any E E ~,X(E) has Poisson distribution with
mean (v,TJ(E)v) and for any finite sequence Ej, 1 ::; j ::; n of disjoint sets from
~, the joint distribution of X(Ej), 1 ::; j ::; n is the product of its marginal
distributions. In short, we have realised a Poisson point process over (n,~) with
intensity measure (v,TJ(·)v) in terms of observables in the quantum probability
space (r s ('Je), (l}l(r s ('Je)), 'I/J( v)).
Proposition 21.6: For any u E 'Je and any observable H in 'Je let
)..(H, u) = W( -u)"(H)W(u). (21.5)
The distribution of the observable )"(H,u) in the vacuum state is also the distri-
bution of )"(H) in the coherent state 'I/J(u). )"(H,u) is essentially selfadjoint on
the linear manifold generated by {e(v + u)lv E D(H2)}. When H is a bounded
observable, ~ is a core for >'(H,u) for every u and
>'(H, u)l~ = >'(H) + a(Hu) + at(Hu) + (u,Hu)I~.
Proof: The first part is immediate from the identity
(, W(-u)eit>..(H)W(u)<I» = ('IjJ(u),eit>"(H)'I/J(u)).
The second part follows from Proposition 20.7 and also implies that ~ is a core
for >'(H,u) when H is bounded. For any 'I/J in ~ we have from Theorem 20.10
and (20.29)
W( -u)>'(H)W(u)'IjJ = eip(u) >.(H)e-iP(u)'I/J
= >'(H)'I/J - p(iHu)'IjJ + (u,Hu)'I/J
= {>.(H) + a(Hu) + at(Hu) + (u,Hu)}'I/J.
In order to combine observables of the form >'(H,u) for varying H,u the
next proposition is very useful.
Proposition 21.7: Let Xj be an observable in the Hilbert space ?ifj,j = 1,2, ... ,
and let 1>j E ?ifj , II1>jll = 1 for each j. Suppose ?if = ®}:I?ifj is the countable
tensor product with respect to the stabilising sequence {1>j}. Let
Pj(t) = (1)j,e- itXj 1>j), j = 1,2, ....
In order that there may exist a one parameter unitary group {Utlt E ~} in ?if
satisfying the relation
00 n 00
Ut Q9Uj = nl~mooQge-itXjUj Q9 Q9 1>j (21.6)

j=1 j=1 j=n+1
for every sequence {Uj} where Uj E ?ifj for each j and Uj = 1>j for all but a
finite number of j's, it is necessary and sufficient that the infinite product pet) =
II~ IPj (t) is defined as a continuous function of t in some open interval containing
O.
Proof: First we prove sufficiency. Let P c ?if denote the set of all product vectors
of the form ®}:IUj, Uj E ?ifj , Uj = 1>j for all but a finite number of j's. Choose
any 'IjJ = ®}:I Uj in P and let Uj = 1>j for all j > jo. For m, n > jo we have
n 00 m 00
IIQge- itXj Uj Q9 Q9 1>j - Qge- itXj Uj Q9 Q9 1>j112

j=1 j=n+1 j=1 j=m+1 (21.7)
= 2II~~IIIUjI12{1- ReIIJ=m+IPj(t)}
provided n > m. Each Pj (t) is the characteristic function of a probability distri-
bution ILj on the real line and by hypothesis p( t) is the characteristic function of
the weakly convergent product IL = ILl * IL2 *, ... , * denoting convolution. If {(j}
is a sequence of independent random variables where (j has distribution ILj then
Lj(j converges in distribution and hence converges almost surely (see Theorem
5.3.4 [24]). Thus limm,n-tooLJ=m+l(j = 0 almost surely and
lim IIJ=m+IPj(t) = 1 for all t.
m,n--+oo
Thus the right hand side of (21.7) converges to 0 as m, n --+ 00. In other words
the limit on the right hand side of (21.6) exists and Ut defined on P by (21.6)
is an isometry. By Proposition 7.2, Ut extends uniquely to an isometry on ?if.
Denote Ut®}:IUj = ®}:le-itXjuj where ®}:IUj E P.1t is clear that for any
fixed t the range of Ut includes all vectors of the form 'IjJ Q9 ®}:n+le-itXjUj,
'IjJ E ?if I Q9 .•. Q9 ?if n . Choose
'ljJn = UI Q9 ••. Q9 Uk Q91>k+1 Q9 •.• Q91>n, n >k
where Uj E ?ifj . Then
00 00
lim 'ljJn Q9 Q9 e- itXj 1>j = UI Q9 •.• Q9 Uk Q9 Q9 1>j.

n-too
j=n+1 j=k+1
This shows that R(Ut} = 'X and Ut is unitary for each t. By definition {Ut\t E IR}
is a unitary group. The strong continuity of the map t -+ Ut is easily established.
This proves sufficiency.
Necessity is immediate from the relation
•
00 00
(@¢j,ulg)¢j) = II~IPj(t).
j=1 j=1
We shall denote the unitary group {Ut} defined by Proposition 21.7 by
00
Ut = @e- itXj .
j=1
It may be noted that ®i=1 ¢j need not be in the domain of the Stone generator
X of Ut since ¢j need not be in the domain of Xj. It is far from clear from this
analysis how a core for X can be constructed from the domains of the individual
operators Xj.
Proposition 21.8: Let Hj be an observable in 'X j with spectral resolution
Hj = kx~j(dx), j = 1,2, ... ,
and let Uj E '1Je j for each j be such that

00
p,(E) = L(uj,~j(E)uj), E E ?:Fn;!

j=1
is a O"-finite measure satisfying the condition
In;! +x
f 2
1 x zdp,(x) < 00. (21.8)
Let Xj be the observable in r s('1Je j ) defined by

Xj = )..(Hj, Uj) - (Uj, Hj(l + HJ)-I Uj ) (21.9)
where )..(Hj,uj) is given by (21.5). Denote by <Pj the vacuum vector in rs('Xj)
for each j. Then there exists an observable X in the Hilbert space
00 00
r s(EB'1Je j ) = @rs('Xj)
j=1 j=1
defined with respect to the stabilising sequence {IJ=I '
(ii) If <P = ®i=I<Pj denotes the vacuum vector in rs(EBi=I'Xj ) then
"tX f"t
(<p, e- z <p) = exp In;! (e- Z x_I + 1 itx
+ X z )dp,(x);
(iii) the distribution of the observable X in the vacuum state cP is infinitely

divisible and has characteristic function
exp l
IR
"t itx
(e Z x - 1 - -1--2)dJ.t(x)"
+x
Proof: By Proposition 21.4, 21.6 and (21.9) we have
(cpj,e- Z"tXiCPj) = exp J(e-

" tZ x - 1 + 1 itx
+ x 2 )dJ.tj(x)
where J.tj(E) = (uj,~j(E)uj)" Since J.t(E) = LjJ.tj(E) the integrability condi-
tion on J.t implies that
II'?" (cp" e- z"tXicp") = exp J(e-

" tZ x -1 itx
+ --)dJ.t(x)
3=1 3' 3 1 + x2
is a continuous function of t" By Proposition 21.7 the existence of the observable
X satisfying (i) and (ii) follows. (iii) is immediate from (ii). •
Corollary 21.9: Let Xo be the observable in r s('Je o) defined by

Xo = m + p(uo)
for some vector Uo E 'Jeo and real scalar m. If X is the observable defined
by Proposition 21.8 and Y is the Stone generator of e- itXo 0 e- itX then the
distribution of Y in the vacuum state of r s ( EB~o 'Je j) has characteristic function
exp{imt - ~t211uo1l2 + JIR(e itx - 1 - 1~~2 )dJ.t(x)}.
Proof: Immediate.
•
Corollary 21.10: Let 'fIe j , Hj, Uj, j = 1,2, ... ,J.t be as in Proposition 21.8 and
let 'Jeo, Uo, m be as in Corollary 21.9. Define
00
'Jej = 'fIej 0 L 2(1ffi+), 'Je = EB'Jej

j=O
uj (t) = Uj 0 I[o,t] , j = 0, 1, 2, ...

Hj(t) = Hj 0 ~L([O,t]), j = 1,2, ... ,
where ~L denotes the canonical spectral measure in L2(1ffi+) with respect to the
Lebesgue measure. Let
Xj(t) = )"(Hj(t), uj(t)) - (uj(t), Hj(t)[1 + Hj(t?r1uj(t))

for j = 1,2,. " .. Then there exists a commuting family of observables {Y(t), t 2
O} in rs('fIe 0 L2(1ffi+)) = ®~ors('fIej) satisfying the following:
(i) e-iay(t) = e-iamte-iap(u~(t)) 0 ®~le-iaXj(t) for all a E Iffi, t 2 0;
(ii) If ' denotes the vacuum vector in rseJC @ L2(~+)) then

(', IIj=1 e -iaj y(tj) eiaj Y( tj - J) ') =
exp{i)tj - tj_J)[-iajm - ~lluoI12a;

j=l
. ia·x
+ j (e- WjX - 1 + -J-dJ.L(x)]}
1 + x2
for all 0 = to < tl < ... < tn, aI, ... , an E ~, n = 1,2, ...
(iii) In the vacuum state ' the observab1es {Y (t) It ::::: O} form a classical stochas-
tic process with stationary independent increments where Y (t) has charac-
teristic function expt{iam -111uo112a2 + JIl?(e iaX - 1 - /~~2)dJ.L(x)} as a
function of a for each t ::::: O.
Proof: The existence of Y (t) satisfying (i) is immediate from the fact that {Hj (t)}
and {uj(t)} satisfy the conditions of Proposition 21.8. (ii) is proved by straight-
forward calculations and (iii) is immediate from (ii). That Y (s) and Y (t) commute
for all s and t follows from (20.3). •
Exercise 21.11: In Corollary 21.9 put

00 00
'Je = EB'Jej , H = 1 E9 EBHj, Ut = e- itH ,

j=O j=l
00
v(t) = tuo E9 EB(e-itHjuj - Uj),

j=l
lit = W(v(t),e-itH)exP[-ij(Sintx - ~)dJ.L(x)l.

l+x
Then v(t) E 'Je for each t and
v(t + s) = v(t) + Utv(s) for all s, t E ~.
The map t ~ lit is strongly continuous and {lItlt E ~} is a one parameter unitary
group in r s('Je) and
(, vt<I» = exp[-~t21IuoI12 -

2
r(e- itx -
JIl?
1 + itx 2 )dJ.L(x)].
1+x
Exercise 21.12: Let J.L be a a-finite measure in ~n satisfying
Put 'Je = L 2 (J.L),

Ct if;K = Q,
(Utf)(;K) = e-if·:f J(;K), v(t)(x) = { -.
- - e-zf':f - 1 if;K i= Q
for each t E [Rn, where C is any n x n matrix with complex entries. Then
{Utlt E [Rn} is an n-parameter unitary group, vet) E 'Je and
vet + §.) = v(t) + Utv(§.) for all §., t E [Rn.
If
Vi = W( vet), U~) exp[-i ~n (sint . Je - 1 ~. ~12 )dJ.l(Je)]
then {Vilt E [Rn} is an n-parameter unitary group in rs('Je) and
1 r· it . x ]
(cI>, Vicl» = exp [ -"2J.l( {O} )IICtll + J~n (e-~Lf - 1 + 1 +- 1~12 )dJ.l(Je) .
2
(21.10)
r
Thus {Vi} determines an [Rn-valued observable in s('Je) whose distribution in the
vacuum state cI> is the infinitely divisible distribution with characteristic function
given by (21.10).
Exercise 21.13: Let G be any group and let g ---+ Ug be any homomorphism from
G into the unitary group OU('Je) of a Hilbert space 'Je. Suppose g ---+ v(g) is a map
from G into 'Je satisfying
v(gh) = v(g) + Ugv(h) for g, hE G. (21.11 )
Let
Vg = W(v(g), Ug), g E G.
Then
Vg Vh = e-ilm(v(g),UgV(h))Vgh for all g, hE G.
In other words g ---+ Vg is a projective unitary representation of G with multiplier

a(g,h) = exp[-iIm(v(g),Ugv(h))] = expiIm(v(rl),v(h)). (See [135].) The
function a(g, h) = (V(g-l), v(h)) satisfies the identity
a(gl,g2) + a(glg2,g3) = a(gj,g2g3) + a(g2,g3) (21.12)
for all gl,g2,g3 E G. Equations (21.11) and (21.12) are known as first and second
order cocycle identities respectively. If there exists a real valued function (3(g),
g E G such that
Im(v(g-l), v( h)) = (3(gh) - (3(g) - (3( h) for all g, h
then the correspondence g ---+ ei /3(g)Vg is a homomorphism from G into OU(r s('Je)).
In such a case
~
(cI>, ei /3(g) vg cI» = exp [i(3(g) - Ilv(g) 112 ]
is an infinitely divisible positive definite function on the group G in the sense that
exp(it (3(g) - ~tllv(g)112) is positive definite in G for each t 2: O. Conversely,
every infinitely divisible positive definite function has this form. The example in
Exercise 21.12 is a special case when G = [Rn.
22 The symplectic group of:JC and Shale's Theorem 161
Exercise 21.14: [100] Let C§ be any real Lie algebra and let X ---7 lI'(X) be a
representation of C§ in '!J3('fie) so that 11' : C§ ---7 '!J3('fie) is a linear map satisfying
[lI'(X),lI'(Y)] = lI'([X,Y]) for all X,Y in C§, Suppose a,(3: C§ ---7 'fie are linear
maps satisfying
lI'(X)a(Y) -lI'(Y)a(X) = a([X, Y]),
lI'(X)* (3(Y) - lI'(Y)* (3(X) = (3([Y, X])
for all X, Y. In r s ('fie) define the operators
<[leX) = at(a(X)) + ,\(lI'(X)) + a((3(X))
with domain 'i8. Then
[<[leX), <[l(Y)] = <[l([X, Y]) + ((3(X), a(Y)) - ((3(Y), a(X))
for all X, Yin C§.
If lI'(X)* == -lI'(X), a(X) = lI'(X)u, (3(X) = -a(X) and W(X)
<[leX) + (u, lI'(X)u) then iw(X) is essentially selfadjoint on 'i8 and [W(X),
w(Y)] = 'ljJ([X, Y]). Furthermore
(e(O), exp <l>(X)e(O)) = exp(u, (e 1f (X) - l)u),
rv denoting closure. (See also Exercise 20.16.)
Exercise 21.15: Let ~ : 9F~+ ---7 cg>('fie) be an IR+ -valued observable in 'fie satisfying
the condition ~ ({t}) = 0 for every t ::::: O. Define
Xt = r(-~([O,t]) +~((t,oo))), t::::: O.
Then {Xtlt ::::: O} is a commuting family of selfadjoint operators satisfying the
following:
(i) xl = 1 for every t, Xo = 1;
(ii) For any u in 'fie and 0 < tl < t2 < ... < t 2n + 1 < 00
('ljJ(u), X tl X t2 ... Xt2n'ljJ(U)) = IIj~a(tzj+J)a(t2j+2)-1
where aCt) = exp 2(u, ~([O, t])u);
(iii) Kolmogorov's Consistency Theorem implies that, in the coherent state 'ljJ( u),
the distribution of the observables {Xtlt ::::: O} is the same as that of a
Markov chain {x( (t) It : : : O} with initial value x(O) = 1, state space {I, -I}
and transition probability matrix
pes, t) = (HI + a(s)a(t)-I)
!(1- a(s)a(t)-I)
!(1- a(s)a(t)-I)) .
!(1 - a(s)a(t)-I)
Notes
The decisive role of the Weyl representation in the geometric approach to infinitely
divisble distributions can be traced to Streater [132], Araki [9], Parthasarathy and
Schmidt [98, 99]. For recent developments on infinitely divisible completely pos-
itive maps on groups, see Holevo [55], [56], Fannes and Quaegebeur [38], [39].
Exercise 21.14 is adapted from Parthasarathy and Schmidt [100]. For an introduc-
tion to the theory of infinitely divisible distributions in classical probability, see
Gnedenko and Kolmogorov [49].
22 The symplectic group of 'fJC and Shale's Theorem

Let ~ be a complex separable Hilbert space and let S : ~ -+ ~ be any real linear
bijective map satisfying the following conditions: (i) Sand S-I are continuous;
(ii) Im(Su, Sv) = Im(u, v) for all u, v E ~. Then S is called a symplectic auto-
morphism of~. All such symplectic automorphisms constitute a group Y'(~) under
multiplication. Every unitary operator in ~ is also a symplectic automorphism of
~. Thus OU(~) is a subgroup of Y'(~). For any Weyl operator W(u) = W(u, 1)
define
Ws(u) = W(Su). (22.1 )
By (20.5) we have
Ws(u)Ws(v) = e-i1m(u,v)Ws(u + v). (22.2)
In other words the correspondence u -+ Ws (u) is another projective unitary
representation of the additive group ~ with multiplier a( u, v) = exp( -i Im(u, v)).
Proposition 20.9 implies that this representation is even irreducible. The question
arises whether Ws is a new representation or there exists a unitary operator f(S)
in f 8 (~) such that
f(S)W(u)f(S)-1 = Ws(u) for all u E ~. (22.3)
When S E OU(~) its second quantization satisfies (22.3). We shall now investigate
how far we can deviate from a unitary operator within the symplectic group Y'(~)
but preserving (22.3).
Let ~o C ~ be a completely real subspace such that ~ = {u + ivlu, v E
~o} = ~o + i~o. For any S E Y'(~) define operators Sij in the real Hilbert space
~o for 1 ~ i, j ~ 2 by putting
S(u + iv) = SllU + iS21 u + S12V + iS22 v (22.4 )
Express any vector in ~o EEl ~o as a column (~), u, v E ~o and define
So (u) = (Sll S12) (U). (22.5)

V S21 S22 V
We have
= ((uI),
VI
(0 1)
-1 0
(u2))
V2
for all Uj, Vj E ~o, 1 ~ i, j ~ 2. In ~o EEl ~o introduce the operator J by
The fact that S preserves the imaginary part of scalar product in ~ can be expressed
in terms of So in ~o EEl ~o by
(So(~:),JSo(~~)) = (G:).J(~~}
22 The symplectic group of ~ and Shale's Theorem 163
for all Uj, Vj E 'lJeo, j = 1,2, i.e.,

o
8 J80 = J. (22.6)
Conversely, if 8 0 is any bounded operator with a bounded inverse in 'lJeo EB 'lJeo
defined by operators 8ij, 1 :::; i, j :::; 2 in 'Jeo through (22.5) and if (22.6) holds
then the map 8 defined by (22.4) in 'Je is a symplectic automorphism of 'lJe.
Proposition 22.1: Let 'lJeo c 'Je be any completely real subspace of 'Je such
that 'Je = 'lJeo + i'lJeo. Then every symplectic automorphism 8 of 'Je admits a
factorisation 8 = UJ TU2 where UJ, U2 are unitary operators in 'lJe and T is a
symplectic automorphism of the form
T(u+iv) = Au+iA-Jv, u,vE'lJeo,
A being a positive bounded operator with a bounded inverse in the real Hilbert
space 'Jeo.
Proof: Define the operator 80 in 'lJeoEB 'lJeo through (22.4) and (22.5) so that (22.6)
holds. Let 80 = UoHo be the polar decomposition of 80 so that Uo is unitary and
Ho is a bounded positive operator with a bounded inverse in 'Jeo EB'lJeo. Then (22.6)
can be expressed as
JUoHo = UoHr;JJ = UoJJ-JHr;JJ
where JUo and UoJ are unitary and Ho and J-J H o- JJ are positive. Thus the
uniqueness of polar decomposition implies
JUo = UoJ, Ho = J- 1Ho- JJ. (22.7)
In particular, there exist symplectic automorphisms U and H of'Je such that Uo
is unitary and Ho is positive in 'Jeo EB 'Jeo. Furthermore U is a unitary operator in
'lJe. The second relation in (22.7) implies that Ho and H o- J are unitarily equivalent
through J. Let ~ denote the spectral measure of Ho in (0,00) so that Ho =
Jx~(dx). Let R],R2,R3 be the ranges of the spectral projections ~((O, 1)),~( {I})
and ~((1, (0)) respectively. Then RJ EB R2 EB R3 = 'lJeo EB 'lJeo. The second relation
in (22.7) implies that J maps RJ onto R3, R3 onto RJ and R2 onto itself. Since
( ( : ), J ( : )) = 0 for all u, v E 'lJeo
we can and do select an orthonormal basis of the form {(~i), J (~i) Ij = I, 2,

1 1
... ,} for the subspace R 2. Let R 2J and R22 be the closed subspaces spanned by
{(~i) Ij = 1,2, ... } and {J (Uvi) Ij = 1,2, ... } respectively so that R2 = R2J EBR22.
1 1
Write MJ = RJ EB R2J, M2 = R22 EB R3. Then 'Jeo EB 'lJeo = MJ EB M 2, J maps MJ
onto M2 and vice versa. Furthermore Ho leaves MJ and M2 invariant. Choose
orthonormal bases {wJ, W2""} and {eJ, e2, ... } in MJ and 'Jeo respectively.
Define a unitary operator Vo in 'Jeo EB 'lJeo by putting
VoWj = (-;j). VoJWj = (~). j = 1,2, ...

164 Chapter II: Observab1es and States in Tensor Products of Hilbert Spaces
Then Va commutes with J and VoHo Va-I leaves the subspaces 'Jeo EEl {O} and
{O} EEl 'Jeo invariant. Thus VaHo Va-I has the form (~ A~I) where A is a
bounded positive operator with bounded inverse in 'Jeo. Hence
So = UOVO- I (~ A~I) Vo·

If U and V are the unitary operators in 'Je which determine the unitary operators
Uo, Va in 'Jeo EEl 'Jeo respectively and T is the symplectic automorphism for which
To = (~ A~I) then S I I
= U TU2 where U = UV- U2 = V. 1, •
Proposition 22.2: For any u, v, w in 'Je
(e( v) e- !p(U)2 e( w)) = (I + IluI1 2)-! exp{ (v w) + ((v, u) - (u, 2

w)?} (228)
, , 2(1 + Ilu11 ) . .
Proof: The left hand side of (22.8) can be expressed as
_1_
,j2i
r(e(v),e-ixP(u)e(w))e-!x2dx = _1_
JIR ,j2i J
r(e(v), W(xu)e(w))e-!x2dx
IR
=
v27r JIR 2
r
~ exp{ -~(1 + Ilu11 2)X 2+ x( (v, u) - (u, w)) + (u, w)}dx.
Now (22.8) follows by a straightforward evaluation of Gaussian integrals. •
Corollary 22.3: Let Ul, U2, . .. , un be mutually orthogonal vectors in 'Je. Then
(e(v),TIY=le-!P(ude(w)) =
j =I(1 + IIUjl1 2)_l} exp {(v,w) + ~((v,Uj)

(22.9)
n - (Uj,W)?}
{
TI 2 ~ 2(1 + lIujl12) .
Proof: Denote by P the projection on the subspace spanned by {u], ... , un}.
By Proposition 19.6 rs('Je) can be identified with the tensor product
n
{Q9r s (CUj)}Q9r s ({u], ... ,un}~)
j=l
so that
By Exercise 20.21
n
TIY=le-!P(uj )2 = {Qge-!p(u j )2} ® 1.
j=l
22 The symplectic group of'Je and Shale's Theorem 165
Then the left hand side of (22.9) is equal to

e(v.(I-P)w)IIf! (e( (Uj, v) U.) e-!p(Uj)2 e( (Uj, W) U .))
3=1 IIUj 112 3' IIUj 112 3 .
Now the corollary is immediate from Proposition 22.2.

•
Proposition 22.4: Let {Un} be a sequence of mutually orthogonal vectors in 'fJ(',
such that limn-too II Un 112 = A. Then
w.lim IIJ=le-f,;;p(ud = e->..t/2 for all
n-too
t~ o. (22.10)
Proof: For any v, W E 'fJ(', we have from (22.9)
(e( v), IIJ=I e- f,;;p(Uj)2 e( w)) = (e( v), IIJ=I e- !p( v/1u;)e( w))
(22.11)
n
= { IIj=1 (1 +-
til Uj 112 ) _l} exp{(v, w) + -t ~((v,Uj)
~
- (Uj,w)?}
( til 112) .
2 1 + Ii Uj
2
n n j=1
By hypothesis
Since {Uj} are mutually orthogonal limn-too (v, un) = 0 for every v. Hence the
right hand side of (22.11) converges to exp{-1At + (v,w)} as n - 00. Since
exponential vectors are total in r s ('lJe) and exp - 1P( u? is a contraction operator
for every U in 'lJe we obtain (22.10). •
Proposition 22.5: Let 'fJ(',o c 'lJe be a completely real subspace such that 'lJe =
'fJ(',o + i'fJ(',o and let A be a bounded positive operator in 'fJ(',o with a bounded inverse.
Suppose there exists a unitary operator r in r 8 ('fJ(',o) satisfying the relations
rW(u)r-1 = W(Au) for all U E 'fJ(',o. (22.12)
Then every point A "I- 1 in the spectrum of A is an isolated eigenvalue of finite
mUltiplicity.
Proof: By hypothesis the spectrum of A is a closed bounded set contained in

the open interval (0,00). If A "I- 1 is an isolated eigenvalue of infinite multiplic-
ity then there exists an orthonormal sequence {un} such that AUn = AUn and
IIAun l1 2 = A2 for n = 1,2, .... If A "I- 1 is a limit point in the spectrum of A
then there exist sequences {An}, {cn} of positive numbers such that An is in the
spectrum of A, An converges monotonically to A, cn - 0 as n - 00 and the in-
J
tervals (Aj - Cj, Aj + Cj), j = 1,2, ... are disjoint. If A = xe( dx) is the spectral
resolution of A we can choose a unit vector Uj in the range of e( (Aj - Cj, Aj + Cj ))
for each j. This shows that if the proposition is not true we can construct an or-
thonormal sequence {un} such that {A un} is an orthogonal sequence of vectors
and
lim IIAun l1 2 = A2. (22.13)
n-too
From (22.12) we get

r ITJ=I
T)- e-t"p(Uj)2r-1 = ITT)- e-t"p(AUj)2
J=I' (22.14)
By Proposition 22.4 and (22.13) the left hand side of (22.14) converges weakly to
e- tt whereas the right hand side converges weakly to e- Ph for each t > 0 and
>. i= 1. This is a contradiction. •
Proposition 22.6: Let dim 'Je < 00, 'Je = 'Jeo + i'Jeo where 'Jeo is a completely
real subspace and let G denote the general linear group of all linear invertible
transformations of 'Jeo. For any C E G let Se denote the symplectic automorphism
of'Je defined by Sc(u + iv) = Cu + iC*-l v for all u, v in 'Jeo. Then there exists
a unitary representation C ----+ Ve of Gin rs('Je) satisfying
(i) VeW(u)Vcl = W(Seu) for all C E G,u E 'Je; (22.15)
(ii) (cI>, VecI» = Idet ~ (C + C*-l) 1- 1/ 2 , (22.16)
where cI> is the vacuum state.
Proof: Let dim'Je = n. Choose and fix an orthonormal basis {e I, e2, ... , en} in
'Jeo. Consider the unitary isomorphism U: rs('Je) ----+ L2(lRn) satisfying
n
[U e( u)](;!lJ = (271')-4 exp Ln I l
{-4 X] + {ej, U)Xj - 2{ej, U)2} (22.17)
j=]
for all u in 'Je. Then (20.27) and (20.28) in the proof of Proposition 20.9 imply
[UW(u)U-If](~) = f(~ - 2.1f), (22.18)

[UW(iU)U-If](~) = eiy,.!ff(~) (22.19)
for any u E 'Jeo, f E L2(lRn) and .1f = ({el, u), ... , (en, u)). Define the operators
Le in L2(lRn) by
(Lef)(~) = I detCl-t f(C-I~), C E G, f E L2(lRn). (22.20)
Put Ve = U- ILeU. Then C ----+ Ve is a unitary representation of G and (22.17)-
(22.20) imply that for any u in 'Jeo
VeW(u)Vcl = W(Cu), VeW(iu)Vc l = W(C*-lU).

Thus for any u in 'Je we obtain (22.15). From (22.17) and (22.20) we have
(cI>, VecI» = (Ue(O),LeUe(O))
= IdetCl-t(271')-:Y kn {exp-~{~,(c*-lC-I + Ih)}d~

•
22 The symplectic group of'iJC and Shale's Theorem 167
We shall now try to extend Proposition 22.6 to the infinite dimensional

case. Choose 'fie = £2 = IC EB IC EB ... and denote by 'fIe k the subspace of all
sequences whose coordinates vanish from the k + I-th stage. We have f s ('fie) =
f s('fie k) ® f s ('fie t) = f s (IC) ® f s (IC) ® . . . where the countable tensor product is
with respect to the stabilising sequence of vacuum vectors. For any A > 0 let VA
be the unitary operator in f s (IC) defined by Proposition 22.6 so that
VAW(Z)VA- 1 = W(AX + iA -I y ), Z = x + iy, (22.21)
(, VA <I» = (A + A-I)~. (22.22)

2
With these notations we have the following proposition.
Proposition 22.7: Let {An} be a sequence of scalars such that An > 0 for each
nand limn-+ooAn = 1. Let
(22.23)
where 1[n+ 1 is the identity operator in f s ('fIe*). Then there exists a sequence {c n }
of scalars such that cnf n converges strongly to a unitary operator f as n ---7 00 if
and only if Cn ---7 C as n ---7 00, Ici = 1 and L,j(Aj -I? < 00.
Proof: First we prove necessity. In the course of the proof we shall view 1/J( u)
for any u in 'fIe k as an element of fs('fIe k ) as well as fs('fIe). Then
n
f n1/J(u) = [VAl ® ... ® VAk 1/J(u)] ® [ @ VAj] ® ® ® ...
j=k+1
for any u in 'fIekon > k where denotes the vacuum in fs(IC). In particular,
Ilfn1/J(u)11 = 111/J(u)11 = 1. Since cnf n converges to a unitary operator it follows
that ICn I ---7 1 as n ---7 00. From (22.22) and (22.23) we have for u E 'fIe k
( ® ® ... , f1/J( u)) = n-+oo
lim Cn ( ® ® ... , f n1/J( u))
= (
@k .
, VAl ® ... ® VAk1/J(u))nl~moocnIIj=k+1
n
( -I) -~
Aj + Aj
2
(22.24)
The totality of all vectors of the form {1/J(u)lu E 'fIekok = 1,2, ... } in fs('fIe)
implies that the left hand side of (22.24) is not equal to 0 for some u in 'fIe k for
some k. This can happen only if Cn ---7 C as n ---7 00, Ici = 1 and
L {1 - ( 2 -I )1/2} < 00 (22.25)

j Aj + Aj
which is equivalent to L,j (Aj - I? < 00 since An ---7 1 as n ---7 00. This proves
necessity.
Conversely, let (22.25) hold. Then for any n > m > k, we have
II [VAk+1 q, Q9 ..• Q9 VAn q,] Q9 q, Q9 q, Q9 ••.
-[VAk +1 q, Q9 .•• Q9 VAm q,] Q9 q, Q9 q, Q9 ... 11 2
)..+).:-1
= 2{1 - nf.l3=m+1 (3 2
3 )-t}
which converges to 0 as m, n ----T 00. Thus for any fixed k and any u in ':If.,k
00
lim
n---too
r n'I/J( u) = [VAl Q9 VA2 Q9 ••. Q9 VAk 'I/J( u)] Q9 to\
'C:I
VAj q,
j=k+1
exists. Furthermore Ilr'I/J(u)11 = 11'I/J(u)1I = 1. Thus r extends to an isometry on

rs(':If.,). By the same argument, for any u E ':If.,k
n
lim [VA-I Q9'" Q9 VA-I'I/J(U)] Q9 to\ VA-Iq, Q9 q, Q9 q, Q9 ...
n-too I k 'C:I j
= rl'I/J(u)
j=k+1
exists and r l extends to an isometry on rs(':If.,). Clearly r l = r* and r is a unitary

operator. •
Proposition 22.8: Let ':If., = h Q9 k be the tensor product of two Hilbert spaces
h, k. Suppose A is a bounded operator in ':If., satisfying the relation
A(B Q9 1) = (B Q9 I)A for all BE <J3(h),
1 being the identity in k. Then there exists aCE ~(k) such that A = 1 Q9 C
where 1 denotes the identity in h.
Proof: Choose and fix a unit vector Uo in h. For any V in k we have

{(I -Iuo}{uol) Q9 1}Auo Q9 v = A(I -Iuo}{uol)uo Q9 v = O.
Thus Auo Q9 v E R(luo} {uol Q9 1). In other words there exists a VI in k such that
A Uo Q9 v = Uo Q9 VI. The correspondence v ----T VI yields a linear operator C E <J3 (k )
such that AUoQ9v = UoQ9Cv. Let now u be any element in h. Choose an operator
B E ~(h) such that Buo = u. Then
Au Q9 v = A(B Q9 I)uo Q9 v = (B Q9I)uo Q9 Cv = u Q9 Cv.
i.e., A = 1 Q9 C.
•
Proposition 22.9: (von Neumann's Double Commutant Theorem [133]): For any
von Neumann algebra X c <J3(~) let XI = {AlA E <J3(':If.,), AB = BA for every
B E X}. Then XI is also a von Neumann algebra and X" = X.
Proof: We omit the proof and refer to [133].

•
Proposition 22.10: Let ':If., be any Hilbert space. Then the smallest weakly closed
algebra containing {W(u)lu E ':If.,} is ~(r8(':If.,».
22 The symplectic group of'iJt and Shale's Theorem 169
Proof: Since W(u)* = W( -u) it follows that the smallest weakly closed algebra
.N containing {W(u)lu E ~} is closed under * and hence a von Neumann algebra.
By Proposition 20.9 .N' = C and .Nil = ~(r 8 (~)). An application of Proposition
22.9 completes the proof. •
Theorem 22.11: (Shale's Theorem [123]): Let ~ = ~o + i~o where ~o is a

completely real subspace. For any symplectic automorphism S of TJC let So be
the operator in ~o EEl ~o defined by (22.4) and (22.5). Then there exists a unitary
operator r( S) in r s (7JC) such that
r(S)W(u)r(S)-l = W(Su) for all u E ~ (22.26)
if and only if S~ So - I is a Hilbert-Schmidt operator in ~o EEl ~o. In such a case
r(S) is determined uniquely up to a scalar multiple of modulus unity.
Proof: Let S be any symplectic automorphism of ~. By Proposition 22.1 we can

express
(22.27)
where A is a bounded positive operator with positive inverse in ~o and Ul, U2 E
au(~). If r(Uj) is the second quantization of Uj we have
r(Uj)W(u)r(Uj)-l = W(Uju),u E~, j = 1,2.

Thus a unitary operator r (S) satisfying (22.16) exists if and only if a unitary
operator r(T) satisfying
r(T)W(u)r(T)-l = W(Tu), u E ~ (22.28)
exists when T is of the form described in (22.27). Suppose dim ~ < 00. Then
the condition that S~ So - 1 is Hilbert-Schmidt is superfluous. Put r(T) = VA
where VA is determined by Proposition 22.6. Then (22.28) obtains and the proof
is complete in this case.
Let now dim ~ = 00. First we prove sufficiency. The condition that S~ So-l
is Hilbert-Schmidt implies that A2 - 1 and A -2 - 1 are Hilbert-Schmidt in ~o.
Hence there exists an orthonormal basis {ej} in ~o such that Aej = Ajej, Aj > 0
for each j and Lj (Aj - 1? < 00. Now identify ~o with real R? via the basis
{ej}, use Proposition 22.7 and define
r(T) = s.lim r n
n--->oo
where r n is defined by (22.23). Then r(T) satisfies (22.28) and the proof of
sufficiency is complete.
To prove necessity suppose that a unitary r(T) satisfying (22.28) exists.
Then r(T)W(u)r(T)-l = W(Au) for all u in ~o. By Proposition 22.5 there
exists an orthonormal basis {ej} in ~o such that Aej = Ajej, Aj > 0 for each j
and limj--->ooAj = 1. Once again identify ~ with £2 via the basis {ej} and define
the unitary operators r n by (22.23). Then
rnW(u)r;;-l=W(Tu)forall UE~n, n=1,2, ...
This together with (22.28) implies that

r;;-1r(T)W(u) = W(u)r;;-1r(T) for all u E ~n.
By Proposition 22.10, the von Neumann algebra generated by {W(u)lu E ~n}
consists of all operators of the form B ® 1 where B E ~(r 8(~n)), 1 denotes
the identity operator in r 8(~;) and r 8(~) = r 8(~n) ® r s(~; ). By Proposition
22.8 there exists a unitary operator r~ in r 8(~;) such that
r(T) = V>q ® VA2 ® ... ® VAn ® r~

for every n. Let CP~ denote the vacuum vector in r 8(~;) and let
Pn = 1 ® Icp~)(cp~1
where 1 is the identity in r8(~n). Then for any u E ~k and n >k
n
Pnr(T)e(u) = Cn[VAI ® ... ® VAke(u)] ® 0 VAjCP ® cP ® cP ® ...
j=k+1
where
Cn = (cp~,r~cp~).
Since limn->oo Pnr(T)e( u) = r(T)e( u) for every u E ~k' k = 1,2, ... ,cnrn
converges strongly to a unitary operator where r n is defined by (22.23). By Propo-
sition 22.7, Cn ---+ C as n ---+ 00, Ici = 1 and Lj(Aj - I? < 00. This implies that
So So - 1 is Hilbert-Schmidt.
We now turn to the uniqueness of r(S) in (22.26). Suppose there is another
unitary operator r'(S) satisfying (22.26) when r(S) is replaced by r'(S). Then
r'(S)r(S)-1 commutes with all the operators W(u), u E ~. By Proposition 20.9
there exists a scalar A of modulus unity such that r'(S) = Ar(S). •
Let ~o C ~ be a completely real subspace such that ~ = ~o + i~o. Define
[fo(~) = {SIS E [f(~), SoSo - 1 is Hilbert-Schmidt in ~o EB ~o}
where [f(~) is the group of all symplectic automorphisms of ~ and So is defined

by (22.4) and (22.5). For each S E [fo(~) select a unitary operator r(S) according
to Shale's Theorem so that
r(S)W(u)r(s)-1 = W(Su) for all u E ~. (22.29)
Then
(22.30)
where (J(S1, S2) is a scalar of modulus unity depending on S1, S2. It is not clear
whether one can choose r(S) in such a way that (J(Sh S2) = 1. If dim ~ < 00
it is clear from Proposition 22.6 that the answer is in the affirmative.
Define
!J[fo(~) = {(u,S)lu E ~,S E [fo(~)}.
22 The symplectic group of 'iii and Shale's Theorem 171
Under the multiplication

(UhSt}(U2,S2) = (Ul +SlU2,SIS2)
9>!fo(7Je) is a group. If W(u,S) = W(u)r(S) where r(.) satisfies (22.29) and
(22.30) then (u, S) -+ W( u, S) is a projective unitary representation of 9>!fo('?Ie).
It is natural to raise the question of an appropriate topology for 9>!fo('?Ie) and
explore the continuity of the map (u, S) -+ W (u, S).
Exercise 22.12: Let 7Je = en = ~n + mn, G = GL(n,~) the general linear
group of all real non-singular matrices of order n and let el, e2, ... ,en be the
canonical basis of ~n. Write
Pj = p(~ej),qj = p(-iej), 1~ i ~n
where p( u) is defined by (20.10). Suppose C -+ Vc is the unitary representation
of G in r s ('?Ie) described in the proof of Proposition 22.6. For any real n x n
matrix L = (( iij )) define the observable XL as the Stone generator of the one
parameter unitary group {VexptLlt E ~}. Then the following holds:
(i) the family {qj, Pj Ii = 1,2, ... } obeys the canonical commutation relations
in the domain of finite particle vectors in rs(7Je);
(ii) XL is the closure of the essentially selfadjoint operator ~j ijj(qjpj + !
Pjqj) + ~i#/ijPiqj on the domain of all finite particle vectors;
(iii) In the coherent state '¢({! + i(jJ, {!, fJ.. E ~n the distribution of XL has the
characteristic function
<P~'f!.(t) =
I det ~ (e tL + e- tL' ) I-! exp{ a'[(1 + etL )(1 + etL' etL)-l (1 + etL' ) - 2]{!
-{3'(1 - etL )(1 + etL' etL )-I(1 - etL')(i
-2ia'(1 + etL )(1 + etL' etL )-I(1 - etLI)fJ..}

where the prime' denotes transpose. In particular, the characteristic function
of XL in the vacuum state is
<P~Q(t) = I det ~(etL + e-tL')I-!;

(iv) When L is symmetric
<P~,Q(t) = Idetsech tLI 1/ 2 expa'(sech tL-l){!
is the characteristic function of an absolutely continuous, symmetric and
infinitely divisible distribution;
(v) When L is skew symmetric
etL + e- tL etL + e- tL
<P~'f!.(t) = exp{a'( 2 - 1){! + fJ..'( 2 - 1)fJ..
-i a'(e tL - e-tL)fJ..}
is the characteristic function of a discrete infinitely divisible distribution. For

suitable values of Q., fl, L, XL has Poisson distribution.
(vi) When n = 2, L = (~ ~)
is an infinitely divisible characteristic function. (It is not clear, in general,

when </>~'f!. is an infinitely divisible characteristic function.)
Exercise 22.13: Let SI, S2 be bounded operators in a Hilbert space 'Je and let J
be an antiunitary operator in 'Je. Define the unitary operators {W(u)lu E 'Je} in
r s ('Je ffi 'Je) = r s ('Je) ® r s ('Je) by
W(u) = W(SIU ffi JS2u).
If SiSI - SiS2 = 1 then
W( u)W( v) = e-i1m(u,v)W( u + v) for all u, v E 'Je.
and
where cI> is the vacuum vector in r s('Je).

Notes
The proof of Shale's Theorem is adapted from Shale [123].
23 Creation, conservation and annihilation operators in r a (~)

There does not seem to exist a natural analogue of the Weyl representation in the
r
fermion Fock space a ('Je) even though the second quantization homomorphism
U ---+ r a (U) from OU('Je) into OU(r a ('Je)) is well-defined by Exercise 20.22. We
shall drop the suffix a from r a(U) when there is no confusion. The projective
unitary representation u ---+ W (u) of the additive group 'Je in r s ('Je) is described
equivalently through the family {p( u) lu E 'Je} of its Stone generators obeying the
commutation relations [P(u),p(v)] = 2iIm(u,v). These can as well be described
by the creation and annihilation operators at (u), a( u), u E 'Je which obey the com-
mutation relations of Proposition 20.14, at(u) maps 'Je®n into 'Je®n+l , whereas
a( u) maps 'Je®n into 'Je®n-l for each n. We shall now introduce analogues of
at(u) and a(u) by exploiting the multiplication 1\ (defined by Proposition 19.2)
for increasing the rank of anti symmetric tensors.
23 Creation, conservation and annihilation operators in r a(:Je) 173
Proposition 23.1: For any U in 'fJe let Mn(u) : 'fJe@n ....... 'fJe@n+l denote the linear
map defined by
Mn(u)vJ /\ V2 /\ ... /\ Vn = U /\ VJ /\ ... /\ Vn
for all Vj in 'lie, 1 ~ j ~ n. Then the adjoint Mn (u) * of Mn (u) is defined by
where the haC over Vj indicates its omission.
Proof: For any Uj, 1 ~ j ~ n + 1 we have from (17.10) and Proposition 19.2
(uJ /\ ... /\ un+J, U /\ VJ /\ ... /\ vn ) =

(uJ,u) (uJ,vJ)
1 (U2' u) (U2' vJ)
n+l!
(Un+J, u) (un+J, vJ) (un+J, vn)
(uJ, vJ) (uJ, vn )
(Uj-J, vn )
(uHJ,vn )
•
Proposition 23.2: For every U in 'lie define the linear operators a( u) and at (u)
on the domain of all finite particle vectors in r a ('lie) by putting
a(u)cI> = O,a(u)I:Je@n = vnM~_J(u),

at(u)cI> = u,at(u)I:Je@n = v'n+!Mn(u), n = 1,2, ...
where Mn(u) and Mn(u)* are as in Proposition 23.1. Then
{a(u)at(v) + at (v)a(u)}'I/J = (u,v)'I/J
for every finite particle vector 'I/J in r a ('fJe).

Proof: By definitions and Proposition 23.1

at(v)wI/\"'/\ Wn = v'n"+1v /\ WI /\ ... /\ Wn,
a(u)at(v)wI/\"'/\ Wn = (U,V)WI/\"'/\ Wn
(23.1 )
n
+2:) -l)j (u, Wj)V /\ WI /\ ... /\ Wj /\ ... /\ Wn,
j=1
n
a(u)wI/\"'/\ Wn = n-!2:)-l)j-l(u,Wj)WI/\"'/\ Wj /\ ... /\ Wn,
j=1
n
at(v)a(u)wI /\ ... /\ Wn = I) -l)j-1 (u, Wj)V /\ WI /\ ... /\ Wj /\ ... /\ wn .
j=1
(23.2)
Adding (23.1) and (23.2) we obtain the required result.
•
Proposition 23.3: In the fermion Fock space fa ('Jf) there exists a unique family
{a( u), at (u) lu E 'Jf} of bounded operators satisfying the following conditions:
(i) a(u) = O,at(u) = u, being the vacuum and u a I-particle vector;
(ii) a(u)vI/\' ,,/\vn = n-~Lj=I(-l)j-l(u,Vj)vI/\" ·MJj/\·· ./\vn,at(u)vI/\
.. ,/\vn = (n+l)I/2u/\vl/\" ,/\vn forallvj E 'Jf, 1:::; j:::; n, n = 1,2, ... ;
(iii) at(u) = a(u)* for all u in 'Jf.
Proof: First define a( u) and at (u) on the dense linear manifold f~ ('Jf) of all
finite particle vectors in f a('Jf) through Proposition 23.2. By Proposition 23.1,
a(u) and at(u) are adjoint to each other on f~('Jf). From Proposition 23.2 we
have for any 'lfJ E f~ ('Jf)
Ila(u)'lfJ112 + Ila t(u)'lfJ112 = ('lfJ,at(u)a(u) + a(u)at(u)'lfJ)
= IluI1 211'lfJ11 2.
In particular,
Ila(u)'lfJ11 ::; lIullll'lfJll, Ilat(u)'lfJ11 ::; lIullll'lfJll, 'lfJ E f~('Jf).
Hence we can and do extend a( u) and at (u) uniquely to bounded operators on
fa('Jf) and denote them by the same symbols so that Ila(u)11 :::; Ilull, Ilat(u)11 :::;
lIull. The rest is immediate. •
Proposition 23.4: The operators a( u) and at (u) defined by Proposition 25.3

satisfy the following:
(i) a(u)a(v) + a(v)a(u) = 0, at(u)at(v) + at(v)at(u) = 0,
a(u)at(v) + at(v)a(u) = (u,v) for all u, v E 'Jf;
(ii) For any U E OU('Jf) its second quantization feU) in f a('Jf) satisfies
f(U)a(u)f(U)-1 = a(Uu) for all u E 'Jf;
23 Creation, conservation and annihilation operators in r a ('fie) 175
(iii) The set {<P, at (uJ) , , . at (un)<PIUj E "JC, 1 :S j :S n, n = 1,2, ... } is total in
r a("JC);
(iv) If T E ~(ra("JC)), Ta(u) = a(u)T, Tat(u) = at(u)T for all u in "JC then
T is a scalar multiple of the identity.
Proof: By (i) and (ii) in Proposition 23.3 the vectors {at (u)a t (v) +a t (v)a t (u)}<P
and {at (u)a t (v) + at (v)a t (u) }WI 1\ ... 1\ Wn are respectively scalar multiples of
u 1\ v + v 1\ u and (u 1\ v + v 1\ u) 1\ WI 1\ ... 1\ Wn which vanish. The density
of finite particle vectors and the boundedness of the operators at (u) imply the
second relation in (i). The first relation in (i) follows from the second by taking
adjoints. The third relation in (i) follows from Proposition 23.2. Since r(U)<p = <P
and r(U)vI 1\ ... 1\ Vn = UVj 1\ ... 1\ UVn for U in OU("JC) we obtain (ii) from
Proposition 23.3. (iii) is just a restatement of the fact that finite particle vectors
are dense in r a("JC). We now prove (iv). For any UI, ... ,Un in "JC we have
(at(uI)" ·at(un)<p,T<P) = (a t (u2)" .at(un)<p,a(uJ)T<P)

= (a t (u2)" ·at(un)<p,Ta(ut)<P) = O.
In other words T<P is orthogonal to every n-particle vector for n ~ 1. Hence there
exists a scalar A such that T<P = A<P. Now
Tat(ut)·· ·at(un)<P = at(ut)·· . at (un)T<P

= Aat(uI)'" at(un)<P
and (iv) follows from (iii). •

The operators at (u), a( u) defined by Proposition 23.3 are respectively called
the fermion creation and annihilation operators associated with u. The relations (i)
in Proposition 23.4 are called the canonical anticommutation relations or CAR. It
is instructive to compare CCR, CAR and the relations satisfied by the free creation
and annihilation operators described in Exercise 20.24 and explore the connections
between them. (See Example 25.18, Exercise 25.35.)
Exercise 23.5: Suppose b(u),bt(u),u E "JC are bounded operators in a Hilbert

space 'ie such that bt (u) = b( u ) * and the map u --+ bt (u) is linear. Let the
following conditions be fulfilled.
(i) b(u)b(v) + b(v)b(u) = 0,
(ii) There exists a unit vector el> E 'ie such that b( u)el> = 0 for all u and the set
{el>, bt (uJ) ... bt (un)el> IUj E "JC, 1 :S j :S n, n = 1,2, ... } is total in 'ie.
Then there exists a unitary isomorphism V : 'ie --+ r a("JC) such that vel> =
<P, Vbt(ut)··· bt(un)el> = at(ut)··· at(un)<P for all Uj E "JC, 1 :S j :S n,
n = 1,2, ... and Vb(U)V-I = a(u) for all u E "JC. (Hint: Use Proposition 7.2.)
Exercise 23.6: Let p(u) = i{at(u) - a(u)},q(u) = a(u) + at(u) in fa(~).

Then p( u) and q( u) are selfadjoint operators satisfying the relations:
p(u)p(v) + p(v)p(u) = 2Re(u, v),
q(u)q(v) + q(v)q(u) = 2Re(u, v),
p(u)q(v) + q(v)p(u) = 2Im(u,v),
f(i)q(u)f(i)-I = p(u) for all u, v E ~.
In particular, p(u? = q(u? = Ilu11 2. In the vacuum state as well as any pure
n-particle state of the form IluI /\ ... /\ Un 11- 1 (UI /\ ... /\ Un) the observables q( u)
and p(u) assume the values ±llull with equal probability.
Exercise 23.7: Let Ut E OU(~),. 0 :f. Ut E ~ for each t E ~ be such that
ul = 1, UsUt = UtUs for all s, t
and
uu = {ut - 2us if s :s; t,
s t -Ut if s > t.
Then (us,Ut - us) = 0 and Ilutll is a non-decreasing function of t. Define
Xt = if(Ut}p(ut)
where p( u) is as in Exercise 23.6. Then {X t I t E ~} is a commuting family of
observables satisfying the relations:
xl = Il u tl1 2,
(4), Xtl xl, ... XtZn 4» = IIUtll1211Ut3112 .. '1I Ut2n_111 2,
(4),Xt1 X tZ '' ·Xtzn + I 4» = 0
for all tl < t2 < ... < t 2n +I , n = 0, 1,2, ...
The distribution of the family {llutll-IXtlt E ~} in the vacuum state 4> is the
same as that of a Markov process {Xtlt E~} with state space {l,-l},P(xt =
1) = P( Xt = -1) = ! for all t and the transition probability matrix
P(s,t) = ~ (~~ ii~:ii ii~:ii=: ~ =ii~:ii ii~:ii=:), s < t.

We can realise the process {Xt I t E ~} as
_ { 1 if Yt is even,
Xt - -1 if Yt is odd
where {Ytlt E ~} is a Poisson process with Yo = 0 or 1 with probability! each
and Yt - Ys has Poisson distribution with mean log Ilutll-log Ilusll for all s < t.
(Compare with Exercise 21.15.)
As a special case one has the following example: Choose ~ : ~~+ -+ g>(~),
an ~+ valued observable satisfying the condition ~({ t}) = 0 for every t. Let u E
~ be such that Ut = ~([O, tlu) :f. 0 for each t. Define Ut = -~([O, t]) + ~(( t, 00)).
Then the required conditions of the exercise are fulfilled.
23 Creation, conservation and annihilation operators in r a (~) 177
Example 23.8: Let X be any set and let K(x, y), x, y E X be a positive
definite kernel with K (x, x) = 1 for all x. Using Proposition 15.4 choose a map
v: X --+ '!JC when '!JC is a Hilbert space and K(x,y) = (v(x),v(y)) for all x,y.
Choose Sx = p( v( x)) where p( v) is as in Exercise 23.6. Since v( x) is a unit vector
Si = 1 and hence Sx is a spin observable, i.e., Sx assumes only the values ± 1. In
the vacuum state of r a('!JC), Sx assumes the values ±1 with equal probability
and the covariance between Sx and Sy is K(x,y). Thus an arbitrary correlation
kernel K (x, y) in any set X can be realised as the correlation kernel of a family
of spin observables. See Section 5, Proposition 5.3, 5.5.
Exercise 23.9: Suppose m( u), m t (u), U E '!JC are bounded operators in a Hilbert
space 'if, such that m t (u) = m( u) * and the following conditions are fulfilled:
(i) The map u --+ m t ( u) is linear;
(ii) m(u)mt(v) = (u,v) for all u,v E '!JC;
(iii) There exists a unit vector E 'if, such that m( u) = 0 for all u and the set
{, mt(uJ) ... mt (un)luj E '!JC,1 5. j 5. n, n = 1,2, ... } is total in 'ie.
Then there exists a unitary isomorphism V : 'if, --+ r Ir('!JC) such that V =
<1>, Vmt(uJ)··· mt(un) = l*(uJ)·· ·l*(un) for all Uj E '!JC, 1 5. j 5. n,
n = 1,2, ... where r Ir ('!JC) and l* (.) are as in Exercise 20.24.
Notes
This is a small part of the theory of second quantization and the literature cited at
the end of Section 20 covers it.
Chapter III
Stochastic Integration and Quantum
Ito's Formula
24 Adapted processes
In Section 21 we have already seen how the classical stochastic processes with
independent increments can be realised as suitable linear combinations of the cre-
ation, conservation and annihilation operators in the boson Fock space r s(7Je) over
a Hilbert space 7Je. This includes, in particular, the Brownian motion and Poisson
process. Since a well-developed theory of stochastic integration with respect to
these classical processes exists, it is natural to wonder whether it would be possible
to introduce quantum probabilistic analogues of stochastic integrals with respect
to processes defined by the creation, conservation and annihilation operators. To
this end one needs the notions of filtration and adapted processes. This, in tum,
depends on the notion of time. We adopt the view that time is an IR+ -valued
observable in a Hilbert space 7Je which, as a spectral measure, has no jumps. The
absence of jumps is assumed more as a matter of mathematical convenience than
a philosophical or conceptual necessity. We shall now examine how such a notion
of time leads to a filtration and the definition of adapted processes.
Let 7Je be a fixed complex separable Hilbert space and let ~ : ~~+ ---- CJ>(7Je)
be a fixed IR+-valued observable with no jump points, i.e., ~({t}) = 0 for every
t ~ O. Denote by 7Je t], 7Je[s,t] and 7Je[t the ranges of the projections ~([O, t]), ~([s, t])
and ~([t, 00)) respectively. Then for any 0 < tJ < t2 < ... < tn < 00 we have
the decomposition
7Je = 7JetJ] EB 7Je[tl,t2] EB··· EB 7Je[tj_l,tj] EB··· EB 7Je[t n_l,t n] EB 7Je[tn'

Let 7Jeo be a fixed complex separable Hilbert space called the initial Hilbert space.
We shall deal with operators in the Hilbert space
'ie = ho ® rs('JC).
It helps our intuition to view ho as the Hilbert space for describing the events
and observables concerning a system and r s(7Je) for describing the same objects
concerning a noise process (or heat bath). Then 'ie can be used to describe events
and observables concerning system plus noise. For any 0 < s < t < 00 write
'ieO] = ho, 'ie t] = ho ® r s(7Jet]),

'ie[t = r s(7Je[t), 'ie[s,t] = r s(7Je[s,t])·
For any u in 7Je let
Ut] = ~([O,t])u,u[t = ~([t,oo))u,u[s,t] = ~([s,t])u.

180 Chapter III: Stochastic Integration and Quantum Ito's Formula
Denote by ~[s,tj, <p[t and <P the vacuu~ in if[s,tj' ;1e[t and rs('~e) respectively. We
s~all view ~tj also as a subspace of ~ by identifying 'I/J in ~tj with 'I/J 0 <p[t in
~. For any 0 < tl < ... < tn < 00 we have
,... .......... ..... ...,
~ = ~td 0 ~[t[,t2j 0 ... 0 ~[tn-l,tnj 0 ~[tn
where the identification between the left and right hand sides is done through the
unitary isomorphism U satisfying
UI 0 e(u) = [10 e(Utd)] 0 e(U[t[,t2j) 0··· 0 e(u[tn_[,tnj) 0 e(u[tn)

for all 1 in ho and U in ~. Under this identification if 'l/Jl E ;1etd' 'l/J2 E ;1e[tl,t2j, ... ,
~n E ;1e[t n_[,t nj, 'l/Jn+l E ;1e[tn we shall denote the element 'l/J10'I/J20" '0'I/Jn+l in
~ by 'l/Jl 'l/J2 ... 'l/Jn+ 1 by dropping the cumbersome 0. We shall also write 10 'I/J =
I'I/J for 1 in ;1eo, 'I/J in rs(~tj) or rs(~).
If (0, '!;, '!;t, P) is a classical probability space with the filtration {'!;t, t 2 O}
of an increasing family of sub (T-algebras of '!;, denote by M t the commutative
*-algebra of all P-essentially bounded complex valued '!;t-measurable functions
with complex conjugation as involution. Then the filtration {'!;t} can be equiva-
lently described by the filtration {Mt } of abelian von Neumann algebras. In the
context of the Hilbert space ;1e we write
~tj = {X 0 l[tlX E ~(;1etj), l[t is identity in ;1e[d. (24.1)

Then {~tj, t 2 O} is an in:reasing family of non-commutative von Neumann
algebras. Any projection in ~ which is an element of ~tj can be interpreted as
an event concerning the outcomes up to time t. Any selfadjoint operator X in ;1etj
has the property that for any bounded real valued Borel measurable function ¢ on
~, ¢(X) 01[t E ~tj. Thus {~tj,t 2 O} may be viewed as the quantum analogue
of a filtration in ;1e induced by the time observable ~ in ~.
The stage is now set for defining the notion of an adapted process. Let
Do ~ho, JItt c ~ be linear manifolds such that ~([s, t])u E JItt whenever u E JItt
for all 0 ~ s < t < 00. Denote by Do0 't: (JItt) the linear manifold generated by
all vectors of the form le(u), 1 E Do, u E JItt. A family X = {Xtlt 2 O} of
operators in if is called an adapted process with respect to the triple (~, Do, JItt) if
(i) D( X t ) ;;2 D o0't: (JItt);
(ii) Xde(utj) E iftj and Xde(u) = {Xde(utj)}0e(u[t) for all t 2 0, u E JItt,
1 E Do. It is said to be regular if, in addition, the map t ---+ Xde(u) from
~+ into ~ is continuous for every 1 E Do, u E JItt.
A map m : t ---+ mt from ~+ into ~ is called a ~-martingale if mt E ?1t'tj for

every t and ~([O, s])mt = ms for all s < t. Suppose m, m' are two ~-martingales.
Then for any a > 0 we have (mt, m~) = (m a , ~([O, t])m~) for all 0 ~ t ~ a.
Thus (mt, m~) as a function of t is of bounded variation in every bounded interval
24 Adapted processes 181
and hence there exists a complex valued measure ((m, m')) in IR+ which has finite
variation in every bounded interval and satisfies
((m, m')) ([0, t]) = (mt, m~) for all t ~ O. (24.2)
If u E 'X then Utj = ~([O, t])u, t 2:: 0 is a ~-martingale. If u, v E 7Je and m is a
~-martingale we denote by ((u, v)), ((u, m)) and ((m, v)) the measures determined
by (24.2) where u and v also denote the ~-martingales t --+ Utj and t --+ Vtj
respectively. We shall now introduce the three basic families of regular adapted
processes with respect to which stochastic integrals will be defined subsequently.
Example 24.1: Choose ho = C so that 'ic = rs('X). Let m be any ~-martingale

in 'X. Using the notion of creation and annihilation operators of section 20, Propo-
sition 20.12 (i), 20.14 (i), define the linear operators Aln(t) and Am(t) by putting
(i) D(Aln(t)) = D(Am(t)) = ~('X);
(ii) Aln(t)e(u) = at(mt)e(u) = {!e(utj + £mt)le=o}e(u[t);
(iii) Am(t)e(u) = a(mt)e(u) = {((m,u))([O,t])e(ut])}e(u[t) for all u in 'X
where at (.) and a(·) are given by (20.29). By definition Aln = {Aln(t)lt 2::
O} and Am = {Am(t)lt 2:: O} are adapted with respect to the triple (~, C, 'X).
Since the role of the initial space is trivial in this example we shall say
that Aln and Am are adapted to the pair (~, 'X) and adopt this convention
hereafter. By Proposition 20.12 we have
IIAln(t)e(u) - Aln(s)e(u)112 = lIat(mt - ms)e(u)112
= {((m, m))([s, t]) + I((m, u))([s, tlW} exp Ilu11 2 ,
IIAm(t)e(u) - Am(s)e(u)112 = 1((m,u))([s,t])1 2exp IIul1 2
for all 0 ~ s < t < 00, u E 7Je. Since ~ has a continuous spectrum it follows
that the maps t --+ Aln(t)e(u), t --+ Am(t)e(u) are continuous. In other
words Aln and Am are regular adapted processes with respect to (~, 'X). Aln
and Am are respectively called the creation and annihilation processes in
r s ('X) associated with the ~-martingale m. Sometimes it is suggestive to
write
Example 24.2: Let HE ?A(7Je) and H~([O, t]) = ~([O, t])H = Ht for all t. Define
the operators AH(t) in rs('X) by putting
(i) D(AH(t)) = ~(7Je);
(ii) AH(t)e(u) = )"(Ht)e(u) = {)"(Ht)e(ut])}e(u[t) for all u in 'X where )..(Ht)
is defined by (20.30).
By (iv) in Proposition 20.13 we have
IIAH(t)e(u) - AH(s)e(u)112 = 11)"(H~([s, t]))e(u)112
= {((Hu, Hu))([s, t]) + I((u, Hu)) ([s, t])12} exp Ilull 2 •
182 Chapter m: Stochastic Integration and Quantum Ito's Formula
Once again it follows from the non-atomicity of ~ that
is a regular adapted process with respect to (~, 'JC). We call AH the conservation
process in r s('JC) associated with H.
Example 24.3: If m,m' are two ~-martingales define the operators A1m)(m'l(t)
in r s ('X) by putting
(i) D(A1m)(m'l(t)) = '&('JC);
(ii) Alm)(m'l(t)e(u) = A(lmt}{mWe(u).
Then
is a regular adapted process with respect to (~, 'JC). Following the preceding ex-
ample we may call A1m)(m'l the conservation process associated with the pair
(m, m') of ~-martingales.
Example 24.4: For any ~-martingale m and selfadjoint operator H in 'JC com-
muting with ~ define H t as the closure of ~([O, t])H and define
where the right hand side is the Weyl operator of Section 20. Since the map
t - t mt is continuous and t - t eiHt is strongly continuous it follows that t - t Wt
is a strongly continuous map from IR+ into OU('JC). Exercise 20.21 implies that
W = {Wtlt ~ O} is a unitary operator valued regular adapted process in rs('JC)
with respect to the pair (~, 'X).
If t - t L t is any strongly continuous map from IR+ into ~(ho) and X is
anyone of the processes Am, AH , Am, W (discussed in Example 24.1-24.4) then
define the operators Yi in 'it = ho 0 r 8 ('X) by putting
(i) D(Yi) = ho0'&('JC);
(ii) Yif e( u) = Ld 0 Xte( u).
Then Y = {Yilt ~ O} is a regular adapted process with respect to (~, ho, 'JC).
Notes
Except for the fact that the filtration in ho 0 r 8 ('JC) is based on a non-atomic
spectral measure ~ in IR+ with respect to 'JC, the rest of the material is based on
Hudson and Parthasarathy [59].
25 Stochastic integration with respect to creation, ... 183
25 Stochastic integration with respect to creation,

conservation and annihilation processes
In the Ito calculus of classical probability, stochastic integrals are defined for
processes adapted to Brownian motion, Poisson process and more generally mar-
tingales and local martingales. Following only the rudimentary ideas of this highly
developed subject to govern our intuition, we shall now develop a rather elemen-
tary theory of stochastic integration of adapted operator-valued processes with
respect to the three fundamental processes of creation, conservation and annihila-
tion described in Example 24.1, 24.2.
Let ho be the initial Hilbert space, '1ff a Hilbert space with a non-atomic time
observable ~ and let
'ie = ho <29 f s('1ff).
If M is anyone of the processes Atn,AH,Am defined in Example 24.1, 24.2
where m is a ~-martingale and H is a bounded operator commuting with ~, we
shall define the stochastic integral of an adapted process L with respect to M by
exploiting the basic factorisability property
(Mt - Ms)e(u) = e(us]){(Mt - Ms)e(u[s,t])}e(u[t) (25.1)
for all 0 ::; s < t < 00, u E '1ff. Here (Mt - Ms)e(u[s,t]) E fs('1ff[s,t]). Property
(25.1) must be compared with the notion of processes with independent increments
in classical probability.
By a partition of lR+ we mean a sequence 0 = to < tl < ... < tn < ...
where tn -+ 00 as n -+ 00. Let L be an adapted process in '1ff with respect to
(~, Do, M). L is said to be simple with respect to the partition {t n } if
Lt = Ltj whenever tj ::; t < tHI , j = 0, 1,2,... (25.2)

If L is simple with respect to a partition then it is also simple with respect to a finer
partition. Suppose L satisfies (25.2) and M is one of the processes Atn, A H , Am.
Using Proposition 19.4 and Corollary 19.5 define the linear operators
Xt = It LdM = It LsdMs
by putting
(i) D(Xt) = Do<29~(M);
(ii) Xtfe(u) = {Lofe(O)}M(t)e(u) if 0 ::; t::; t l ,

= Xtnfe(u) + {Ltnfe(utn])}(M(t) - M(tn))e(u[tJ
if tn < t ::; tn+ I , n = 1, 2, ... (25.3)
where the right hand side is determined inductively in n for each f E Do, u E M.
We make the following elementary observations. The definition of X t is
independent of the partition with respect to which L is simple. For any f E Do, u E
M, the map t -+ Xtfe(u) is continuous and Xtfe(u) = {Xtfe(ut])}e(u[t). In
other words {Xtlt 2: O} is a regular adapted process with respect to (~,Do,M).
Proposition 25.1: (The first fundamental lemma) Let L be a simple adapted

process in ik with respect to (e,Do,M). Then for any f E ho, U E 7Je, 9 E Do,
v E M the following holds:
(fe(u), It LdM ge(v)) = l\fe(U),L(s)ge(V))df./,(S)
where f./, = ((u, m)), ((u, Hv)) or ((m, v)) according to whether M = Atn, AH or
Am.
Proof: Let L be simple with respect to the partition {tn}. Suppose tn ~ t < t n+1 •
From (25.3) and the adaptedness of L we have
(fe(u), It LdMge(v)) =
n
{L (fe( Utj_d)' L(tj-dge(Vtj_d)) (e( U[tj_l)' (M(tj) - M(t j_ 1) )e(V[tj_l))}
j=l
(25.4 )
From (i), (ii) of Proposition 20.12 and (i) of Proposition 20.13 we have
(e(U[a)' (M(b) - M(a))e(v[a)) = f./,([a,b])(e(u[a),e(v[a)) (25.5)

where f./, is the complex valued measure described in the proposition. Substituting
(25.5) in (25.4) and using the adaptedness of L as well as the basic property of
exponential vectors we get
(fe(u), It LdMge(v)) =
n
{L (fe( u), L( t j - 1 )ge( v)) f./,([t j- 1, tj ])} + (fe( u), L(tn)ge( v)) f./,([t n , t])
j=l
= I
t
(fe(U),L(s)ge(V))df./,(S), •
Proposition 25.2: (Second fundamental lemma) Let Lj, j = 1,2 be simple
adapted processes with respect to (e, Dj, M j ), j = 1,2 respectively in ik. Suppose
M j is anyone of the processes Atn j , AHj , Amj for each j = 1,2 where ml, m2 are
e-martingales and HI. H2 are bounded operators in 7Je commuting with e. Define
Then for any f E D I , g E D 2 , U E M I , V E M2
(XI (t)fe( u), X2(t)ge( v)) = it (LI (s )fe( u), X2(S )ge( v))dJ.lI (s)
+ i t (X I(S)f e(U),L2(S)ge(V))dJ.l2(S)
+ lot (LI (s )fe( u), L 2(s)ge( v) )dJ.l12 (s)

where J.lI, J.l2 and J.l12 are given by the following tables:
J.lI ((ml, v))

(25.6)
M2 Atn2
M2
MI Atn2 AH2 Am2
(25.7)
Atnl ((m[, m2)) ((ml,H2v)) 0
AHI ((HIU, m2)) ((HI U, H2V)) 0
Ami 0 0 0
Proof: Let {tj} be an arbitrary partition of IR+ with respect to which both LI
and Lz are simple and 0 = to < tl < ... < tn = t. Then
(XI (t)fe( u), Xz(t)ge( v))

n
= Ci~,LI(tj-dfe(utj_d)(MI(tj) - MI(tj-d)e(u[tj_I)'
j=1
n
LL2 (tj_t)ge(Vtj_d)(M2 (tj) - M2 (tj_t))e(v[tj_I))
j=1
n
= L (LI (tj-dfe( Utj_d)' L 2 (tj_t)ge( Vtj_J)
j=1
n
+L (XI (tj-I)fe(Utj_d), L2(tj_t)ge( Vtj_I))
j=1
n
+2)L I (tj-dje( Utj_d)' X2(tj-dge( Vtj_l))
j=1
x ((MI(tj) - M2(tj - I ))e(u[tj_J,e(V[tj_l))

= 8 1 + 82 + 8 3, say . (25.8)
By Proposition 20.12 and 20.13 we have for all 0 ~ a < b < 00
(e(U[a), (M2(b) - M2(a))e(V[a)) = JL2([a, b])(e(u[a), e(V[a))' (25.9)
((MI(b) - MI(a))e(U[a),e(V[a)) = JLI([a, b])(e(u[a)' e(V[a))' (25.10)

((MI(b) - MI(a))e(u[a), (M2(b) - M2(a))e(v[a))
(25.11)
where JLI, JL2 and JLI2 are given by (25.6), (25.7) as stated in the proposition.
Substituting (25.9) in the term 82 of (25.8) we have
n
82 = L(XI(tj-l)je(u),L2(tj-dge(v))JL2([tj -dj]). (25.12)
j=1
Let for any partition 1r = {tj} with respect to which LI, L2 are simple
07r = sup(tj - tj-d.
j "21
Since the map s ----t XI(s)je(u) is continuous it follows from (25.12) that
lim 82 = rt (XI(s)je(u), L2(s)ge(v))dJL2(S).

6".-+0 io
Similarly it follows by substituting (25.10) in the term 8 3 that
lim 83 = rt (LI(s)je(u),X2(s)ge(v))dJLI(s).
6".-+0 io
Substituting (25.11) in the term 8 1 of (25.8) we have
8 1 = i t (LI(s)je(u),L2(s)ge(v))dJL12(s) + RI.
where
n
RI = L (LI (tj-dje( u), L2(tj- dge( v)) JLI ([tj_l, tj ])JL2([tj-h tj]).
j=1
Thus
Since ~ is non-atomic and hence J.L2 is non-atomic it follows that IRll ~ 0 as

b1f ~ O. Thus 8 1 converges to
•
Corollary 25.3: Let L be a simple adapted process in 'if with respect to (~, Do, .M).
Suppose
X(t) = lot LdM, t 2: 0
where M is anyone of the processes Atn, A H , Am. Then for any f E Do, u E .M
IIX(t)fe(u)112 =
2Re lot (X(s)fe(u), L(s)fe(u))dJ.L(s) + lot IIL(s)fe(u)11 2dv(s)

where J.L = ((u,m)), ((u, Hu)) or ((m,u)) and v = ((m,m)), ((Hu, Hu)) or 0
according to whether M = Atn, AH or Am.
Proof: This is just a restatement of the second fundamental lemma where Ll =

L2 = L, Ml = M2 = M, g = f, v = u. •
Corollary 25.4: Let L, M, X, f, u, J.L, v be as in Corollary 25.3. Then
IIX(t)fe(u)112:::; lot IIX(s)fe(u)1121J.LI(ds) + lot IIL(s)fe(u)112(1J.L1 + v)(ds)

where IJ.LI(E) denotes the variation of J.L in the Borel set E.
Proof: This is immediate from Corollary 25.3 if we observe that
for any two vectors <p, 'l/J in '!Ie and
IRe lot hdJ.L1 :::; lot IhldlJ.Ll.

•
Proposition 25.5: Let F, G, a be non-negative continuous functions in Iffi+ and
let F, G be monotonic non-decreasing. Suppose F(O) = 0 and
a(t) :::; G(t) + lot a(s)dF(s) for all t 2: O. (25.13)
Then
a(t) :::; G(t)expF(t) for all t 2: O. (25.14)
188 Chapter III: Stochastic Integration and Quantum Ito's Fonnula
Proof: From (25.13) we have
a(t) S G(t) + lot {G(t2) + lot2 a(t!)dF(t!)}dF(t2)
S G(t) +
Jo
r <t2<t a(t!)dF(t!)dF(t2)'
t G(t!)dF(t!) + JO<tl
Inductively we obtain
a(t) S G(t) + ~ lo<tl <... <tj<t G(t!)dF(t!) ... dF(tj)

(25.15)
+ r
Jo<t 1 <.. ·<t",+l <t
a(tJ)dF(tJ) ... dF(tn+!).
Let f3(t) = sUPo<s<ta(s). Since G is non-decreasing and the continuity of F

implies - -
r
JO<tl< ... <t",<t
1 dF(t!) ... dF(t n ) = F(t)n
n!
so that, by (25.15),
a(t) S G(t){1 + tF\~)j} + f3(t) F(t)n~J

. J J.
J=
n + 1.
for every n. Letting n -+ 00 we get (25.14).
•
Proposition 25.6: Let L be any simple adapted process with respect to (~, Do, M)
and let M be anyone of the processes Atn, A H , Am. Then for any J E Do, u E M,
t~O
II lot LdMJe(u)112 S ell'l([O,t]) lot IIL(s)Je(u)112(v + Iltl)(ds) (25.16)
where It = ((u,m)),((u,Hu ~ or ((m,u)) and v = ((m,m)),((Hu,Hu)) or 0

according to whether M = Atn, AH or Am.
Proof: This is immediate from Corollary 25.4 and Proposition 25.5. •

Using inequality (25.16) we can now extend the notion of stochastic inte-
grals to adapted processes which are not necessarily simple. We begin with some
definitions. A process L adapted to the triple (~, Do, M) is called measurable if
for every J in Do, u in M the map t -+ L(t)Je( u) is Borel measurable. Two such
measurable processes L, L' are called equivalent if
~({tIL(t)Je(u) =I L'(t)Je(u)}) = 0 for all J E Do, u EM.

A (~, Do, M )-adapted measurable process L is called stochastically integrable if

there exists a sequence {Ln} of simple (~,Do,M)-adapted processes such that
lim
n->oo Jort IILn(s)fe(u) - L(s)fe(u)112(v,~(ds)v) = 0 (25.17)
for all v E 7JC, u E M, f E Do.
Let M be anyone of the fundamental processes AJn, A H, Am and let L be stochas-
tically integrable so that (25.17) holds for a sequence {Ln} of simple processes.
By Proposition 25.6
"(lot LkdM -lot LndM )fe(u)112 ::=;

(25.18)
el~I([O,t]) lot II(Lk(S) - Ln(s))fe(u)112(v + 1J.tI)(ds)
for all f E Do, u E M where, for any fixed a > 0, the measures J.t, v satisfy the
following: for any Borel set E C [0, a]
(rna, ~(E)ma) if M = AJn,

v(E) = { (Hu,~(E)Hu) ~f M = AH ,
o tfM=A m
and
1J.tI(E) ::=; 2{ (u, ~(E)u) + (rna, ~(E)ma)) if M = Atn or Am,
::=; 2{ (u, ~(E)u) + (Hu, ~(E)Hu)} if M = A H •
Now (25.17) and (25.18) imply that
lim (
n->oo Jort LndM)fe(u)
exists for all f in Do and u in M and the limit is independent of the sequence {Ln}
satisfying (25.17). We denote this limit by U;
LdM)fe(u) and extend it linearly
to the linear manifold Do0~(M) by using Proposition 19.4 and Corollary 19.5.
Such a procedure yields a linear operator with domain Do 0 ~ (M). We denote it
by J;LdM and call it the stochastic integral of L with respect to M. If L' is
equivalent to L it is clear that L' is also stochastically integrable and J; L'dM =
J; LdM. By a (~,Do,M)-adapted stochastically integrable process L we always
mean its equivalence class. We denote by [l(~, Do, M) the complex linear space
of all such stochastically interrable processes. By a fundamental process we mean
anyone of the processes Am,AH,Am where m is any ~-martingale and H is
a bounded operator in ~ commuting with ~. If L E [l(~,Do,M) and M is a
fundamental process then J;
LdM is a regular (~, Do, M) -adapted process in the
variable t and the map L - t J; LdM is linear.
Proposition 25.7: Let L be any (~, Do, .M)-adapted process satisfying the follow-
ing: for each f E Do. u E.M
(i) the mapping t --t L(t)fe(u) is left continuous;
(ii) sUPo:o:;s:o:;tIIL(s)fe(u)11 < 00 for each t.
Then L E 1L(~,Do,.M).
Proof: Define
Then Ln is simple and (~, Do,.M )-adapted. Furthermore (i) implies that
lim Ln(t)fe(u) = L(t)fe(u) for each f E Do,u E.M.

n->oo
Now (ii) implies
lim (t IILn(s)fe(u) _ L(s)fe(u)11 2dJL(s) = 0

n->oo Jo
for every totally finite measure JL. In particular L is stochastically integrable. •
Corollary 25.8: Every regular (~, Do,.M )-adapted process is stochastically inte-
grable. If L E 1L(~,Do,.M) and MJ,M2 •... ,Mn are fundamental processes then
the repeated stochastic integral
X(t) = ( L(tI)dM1 (tI)dM2(t2)'" dMn(tn)

JO<tl < ..· <tn <t
is well-defined as a regular (~. Do •.M)-adapted process.
Proof: Immediate.
•
Proposition 25.9: (First fundamental lemma) Let L E IL(~, Do,.M) and let M be
a fundamental process. Then
t
(fe(u).l L(s)dM(s)ge(v)) = It (fe(u). L(s)ge(v))dJL(s)
for all f E ho, u E '(ff,. 9 E Do, v E .M where JL = ((u, m)). ((u, Hv)) or ((m. v))
according to whether M = Am. AH or Am.
Proof: Immediate from the first fundamental lemma (Proposition 25.1) for simple
processes and the definition of stochastic integrals. •
Proposition 25.10: (Second fundamental lemma): Let Lj E IL(~, Dj, M j ), j = 1,2

and let M 1, M2 be any two fundamental processes. Then for any f E D I, g E D2,
U E MI, V E M2
t
(I LldMI/e(u), lt L2dM2ge(v))
= It(LI(S)fe(U), 18
L2dM2ge(v))dJ.tl(s)
+ lt (1 8
LldMlfe(u), L2(s)ge(v))dJ.t2(S)
+ lt (LI (s )fe( u), L 2 ( s )ge( v) )dJ.t 12 ( s)
where J.tl, J.t2 and J.t12 are given by the following tables:
J.t2 ((u, m2)) ((u, H2V)) ((m2' v))
M2
MI Am2 AH2 Am2
Ami ((mj, m2)) ((mj, H2V)) 0

AHI ((Hlu, m2)) ((HI u, H2V)) 0
Ami 0 0 0
Proof: Immediate from the second fundamental lemma (Proposition 25.2) for
simple processes and the definition of stochastic integrals. •
Corollary 25.11: For any L E 1L(~,Do,M) and any fundamental process M the
following holds: for every f E Do, u EM and t 2: 0
(i) lilt LdMfe(u)112 =
2Re I
t
(1 8
LdMfe(u),L(s)fe(u))dJ.t(s)+ lt IIL(s)fe(u)11 2dv(s);
(ii) II J~ LdMfe(u)112 :::; eIJL1([O,t]) J~ IIL(s)fe(u)112(v + 1J.t1)(ds) where J.t =

((u, m)), ((u, Hu)) or ((m, u)) and v = ((m, m)), ((Hu, Hu)) or 0 according to
whether M = Am, AH or Am.
Proof: Immediate from Proposition 25.10 after putting u = v, I = g, Ll = L2 =

L and using Proposition 25.5.
A pair (L, Lt) of (~, Do, .M)-adapted processes is called an adjoint pair if
(L(t)le(u),ge(v)) = (fe(u), Lt(t)ge(v)) for all I,g E Do, u,v E.M, t ~ O.
Proposition 25.12: Let M be any fundamental process. Suppose Mt = Am, A H *

or Atn according to whether M = Atn, AH or Am. Then (M, Mt) is an adjoint
pair of (~, 'X)-adapted processes in rs('X).
Suppose (L, L t) is an adjoint pair of (~, Do,.M )-adapted stochastically inte-
grable processes and
X(t) = lot LdM,Xt(t) = lot LtdMt.

Then (X, X t) is an adjoint pair.
Proof: The first part follows from the fact that the creation and annihilation oper-
ators at (u) and a( u) are adjoint to each other whereas the conservation operators
>'(H) and >'(H*) are adjoint to each other on the domain ~('X).
The second part is immediate from the first fundamental lemma. •
Denote by ?f{ the linear space generated by all complex valued measures of
the form ((m, m')) where m, m' are arbitrary ~-martingales. Any T E ?f{ is totally
finite on every bounded interval of IR+ and absolutely continuous with respect to
~ in the sense that T(E) = 0 whenever ~(E) = O. For any simple (~,Do,.M)
adapted process L which is subordinate to the partition {tj} as in (25.2), define
the operator J; LdT with domain Do Q9 ~ (.M) by
(lot LdT)le(u) = {~T([tj-l' tj])L(tj-dle(u)} + T([tn, t])L(tn)le(u)
iftn ~ t < t n +1, n = 0, 1,2, ...
For any L E 1L(~,Do,.M), choose an approximating sequence {Ln} of simple
(~, Do,.M )-adapted processes satisfying (25.17) and observe that
lim
n->oo iort II(Ln(s) - L(s)le(u)11 2 dITI(s) = 0
for all I E Do, u E .M, T E ?f{, t ~ O. This implies the existence of
lim (
n->oo iort Ln(s)dT(s))le(u)
as well as the independence of the limit from the approximating sequence {Ln}
for each I E Do, u E.M. Denote this limit by U;
LdT)le(u) and extend this def-
inition by linearity to Do0~(M). This yields a regular ("Do,M)-adapted process

of operators U; Ld'Tlt ~ O}. A simple application of Schwarz's inequality yields
11(1t Ld'T)fe(u)W ::; I'TI([O, t]) lt IIL(s)fe(u)11 2dl'Tl(s)

(25.19)
for all f E Do, u EM.
Imitating the proofs of the first and second fundamental lemma we obtain the
following identities:
(Je(u), lt Ld'Tge(v)) = I t
(Je(u),L(s)ge(v))d'T(S)
for all L E IL(" Do,M), 'T E 'if{, f E ho, u E 'JC, 9 E Do, v EM, (25.20)
(1t L\d'Tde(u), lt L2d'T2ge(v))
= lt (L\(s)fe(u), 1 8
L2d'T2ge(v))d'T\(s)
+ lt (1 8
L\d'Tde(u),L2(s)ge(v))d'T2(S) (25.21)
for all Lj E 1L("Dj,Mj ), 'Tj E'if{, f E D\, 9 E D2, U E M\, V E M2,
(1t L\dMfe(u), lt L2d'Tge(v))
= lt (L\(s)fe(u), 1 8
L2d'Tge(v))dj.£(s)
(25.22)
where Lj E 1L("Dj,Mj ), j = 1,2, 'T E 'ife, f E Dj, 9 E D2, U E Mj, v E M2,

M is any fundamental process and j.£ is ((m, v)), {(Hu, v)) or {(u, m)) according
to whether M = Atn, AH or Am.
Suppose Xo is any operator in Cje whose domain includes Do 0 ~ (M) and
Xofe(u) = (CoJ)e(u) for all f E Do, u E M
where Co is an operator in ho whose domain is Do, Mj, I ::; j ::; n are fundamental
processes, 'Tj+n, j = I, 2, ... ,k are measures from 'j{ and Li, I ::; i ::; n + k are
adapted processes in IL(', Do,M). Then
X(t) = Xo +L
n
j=\
1
0
t
LjdMj +
n+k
L
j=n+\
1
0
t
Lj d'Tj , t ~0
194 Chapter ill: Stochastic Integration and Quantum Ito's Formula
is a regular (e, Do, M)-adapted process. Whenever a process has this fonn we
write
n n+k
dX = LLjdMj + L Ljd7j.
j=l j=n+1
If (Lj, L j) is an adjoint pair of processes in IL( e, Do, M) for each j

n n+k
dXt = LLjdMJ + L LjdTj
j=l j=n+1
where bar denotes complex conjugation and XO, XJ are adjoint to each other in
Do®~(M), then (X,xt) is an adjoint pair of regular (e,Do,M)-adapted pro-
cesses.
Theorem 25.13: (Quantum Ito's fonnula [59]) Suppose

n k
dX = LKidMi + LLjd7j,
i=l j=l
n' k'
dX' = LK;,dMI, + LLj,d7}"
i'=l j'=l
where Ki,Lj E lL(e, Do,M), K:"Lj, E lL(e, Db,M'), Mi , MI, are fundamental
processes and 7j, 7}' E ;r~ for all i, i', j, j'. Then the following holds: for all
j E Do, 9 E Db, u E M, v E M'
(X(t)je(u),X'(t)ge(v)) - (X(O)je(u),X'(O)ge(v)) =
~ lot (Ki(s)je(u),X'(s)ge(v))dJ.ti(S)
+L 1°
k
i=l
t
(Li(s)je(u),X'(s)ge(v))d7i(S)
+L 1
n' t
(X(s)je(u),K:,(s)ge(v))dJ.tHs)
°
i'=l
+L 1
k' t
(X(s)je(u),Lj,(s)ge(v))d7}'(S)
°
j'=l
+ L I t
(Ki(s)je(u),K:,(s)ge(v))dJ.tii'(S)
°
l<i<n
1:::;i'5 n'
where J.ti, J.t~, and J.tii', are given by the following tables:
Mi Ak AHi Ami
J.ti ((mi'v)) ((HiU, V)) ((u, mil)

Mf, A~'i' AH'i' Am'i'
J.ti, ((u, mi,)) ((u,HI,v)) ((mi"v))

Mf,
Mi A~" , AH'i' Am'i l
A1n i ((mi' mi,)) ((mi' HI,v)) 0

AHi ((Hi U, mi,)) ((Hi U, HI,v)) 0
Ami 0 0 0
Proof: When X(O) = 0, X'(O) = 0 the theorem is immediate from the second
fundamental lemma (Proposition 25.10) and equations (25.21), (25.22). To treat
the general case we observe that
(X(t)fe(u),X'(t)ge(v)) - (X(O)fe(u),X'(O)ge(v))
= ((X(t) - X(O))fe(u), (X'(t) - X'(O))ge(v))
+((X(t) - X(O))fe(u),X'(O)ge(v))
+(X(O)fe(u), (X'(t) - X'(O))ge(v)).
Applying the theorem to the first term and the first fundamental lemma (Proposition
25.9) together with (25.20) to the second and third term we obtain the required
remk •
It may be noted that when Ki = 0, Lj = 0, X(O) = 1 the quantum Ito's
formula reduces to the first fundamental lemma together with (25.20).
Proposition 25.14: Let
n k
dX = LKidMi + LLjdTj
i=l j=l
where Ki,Lj E 1L(~,Do,.M.), Tj E '?J~ and Mi is a fundamental process for all

i,j. Then for all f E Do, u E .M., 0 :::; tl < h :::; T < <Xl
II(X(t2 ) - X(tJ))fe(u)112
where /-ti = ((u, mi)), ((u, HiU)) or ((mi' u)) and Vi = ((mi' mi)), ((HiU, HiU)) or
o according to whether Mi = Atni' AHi or Ami' 1 ~ i ~ n.
Proof: Define
Y(t) = 0 ift ~ tJ,X(t) - X(tt) if t > t 1•
Then
n k
dY = 2:kidMi + 2:£jdTj
i=l j=l
where, for any K E 1L(~,Do,.M),

k(t) = 0 ift < tl, = K(t) ift 2:: tl
and it is to be noted that k E 1L(~,Do,.M). Thus
X(t2) - X(t 1 ) = Y(t2) = 2: it2 kidMi + 2: it2 £jdTj.

i 0 j 0
Since for any n vectors 1/Jj, 1 ~ j ~ n in any Hilbert space
Ii n1~ 1~
L..,.1/Jjll 2 ~ nL..,.ll1/Jjll 2
j j
we obtain
+2: 11 10
k rt2 £jdTjJe(u)112}.
j=l 0
Now the required result follows from property (ii) in Corolary 25.11 and inequality
(25.19).
Example 25.15: Let ho = C, M(t) = Am(t) where m is a ~-martingale in 'X

and let X(t) be the (~, 'X)-adapted process defined by
D(X(t)) = "&('X),X(t) = eM(t\eM(t)e(u) = e(mt,u)e(u) ..
Denote
</>(t) = (e(u),X(t)e(v))
= exp( (mt, v) + (u, v))
= exp( ((m, v)) ([0, t]) + (u, v)).
Since ((m, v)) is a non-atomic measure in ~+ we have
</>(t) - </>(0) = i t </>(s)((m,v))(ds).

Thus
dX=XdM.
This also shows that X(t) = eAm(t), xt(t) = eAJ,,(t), t ~ 0 is an adjoint pair of
(~, ~)-adapted processes satisfying
dX = XdA m , dXt = XtdAk.
where D(Xt(t)) = ~(~), Xt(t)e(u) = e(u + mt).

Example 25.16: Let H be any bounded operator in ~ commuting with ~ and let
X(t) be the (~, ~)-adapted process in r8(~) given by
D(X(t)) = ~(~),X(t) = eAH(t).
From Example 24.2 we have
X(t)e(v) = e'\(Ht)e(v) = e(eHtv)

where Ht = H~([O, t]). Define
cp(t) = (e(u),X(t)e(v)}.
Then
cp(t) = exp(u, eHtv} = exp( (u, (e H - l)vt) + (u, v})
= exp(((u,(e H -l)v}}([O,t]) + (u,v}).
The non-atomicity of (( u, (e H - l)v)} implies that
cp(t) - cp(O) = it cp(s)((u,(e H -l)v}}(ds).

By the first fundamental lemma we conclude that
dX = XdA(eH_l)'
If xt(t) = eAH * (t) then X, xt is an adjoint pair of (~, ~)-adapted processes.

As an interesting special case consider the operators Ut = eAiH(t) where H
is a bounded selfadjoint operator commuting with ~. Then Ut = expA(iHt ) =
r (exp iHt) which is the second quantization of the unitary operator exp iHt
in ~. Thus as a (~, ~)-adapted process the second quantized unitary operators
r( exp iHt) satisfy
Example 25.17: Let m be any ~-martingale and let U be any unitary operator
commuting with ~ in 'JC, Define the unitary operators
Ut = U~([O, t]) + ~([t, (0))
in 'JC and put
W(t) = W(mt, Ut)
where the right hand side is the Weyl operator defined in Section 20, From (20,2)
we have for any u, v in 'JC
1
(e( u), W(t)e( v)) = exp{ -211mt112 - (mt, Utv) + (u, UtV + mt)
1
= exp [{-2((m, m)) - ((m, Uv)) + ((u, (U - l)v))
+((u, m)) }([O, t]) + (u, v)).

If ¢(t) = (e(u), W(t)e(v)) then
¢(t) - ¢(O) = lot ¢(s){ -~((m, m)) - ((m, Uv)) + ((u, (U -1)v)) + ((u, m)) }(ds).
Hence by the first fundamental lemma

1
dW = W{dAtn + dA u - 1 - dAu-l m - 2d((m,m))} (25.23)
where U-1m denotes the ~-martingale defined by (U-1m)t = U-1mt for all t.
°
Since U commutes with ~ it is clear that U-1m is, indeed, a ~-martingale. When
m = we obtain
df(Ut) = f(Ut)dAu-1(t) (25.24)
where f(Ut ) is the second quantization of Ut.
Example 25.18: [62] In Example 25.17 let U = -1. Put J(t) = f(Ut). Then
J(t) = J(t)* = J(t)-l and (25.24) becomes
dJ = -2JdAl (25.25)
since A_2(t) = -2Al (t). For any ~-martingale m define the adjoint pair (Fm, F~)
of adapted processes by
(25.26)
Using quantum Ito's formula we shall examine some of the basic properties of
this pair and arrive at CAR from CCR. (See Section 23.)
Proposition 25.19: For any u, v in 'JC, t 2: °

(J(t)e(u),Fm(t)e(v)) + (F~(t)e(u),J(t)e(v)) = 0.
Proof: Let
¢(t) = (J(t)e(u), Fm(t)e(v)), 'ljJ(t) = (F~(t)e(u), J(t)e(v)).

By quantum Ito's formula, the unitarity of J(t), (25.25) and (25.26) we have
¢(t) = ((m, v)) ([0, t])e(u,v) - 21t ¢(s)((u, v))(ds),
'ljJ(t) = -((m, v)) ([0, t])e(u,v) - 21t 'ljJ(s) ((u, v))(ds).
Adding these two equations we obtain
t
¢(t) + 'ljJ(t) = -21 (¢(s) + 'ljJ(s)) ((u, v))(ds).
The non-atomicity of ((u, v)) implies that
¢(t) + 'ljJ(t) = {¢(o) + 'ljJ(0)} exp{ -2((u, v))([O, t]} = 0. •
Proposition 25.20: For any u, v in '1JC, t 2:: °and any two ~-martingales m, m'
(Fm(t)e(u), Fm' (t)e(v)) + (F~,(t)e(u),F~(t)e(v))
= ((m', m))([O, t])(e(u), e(v)).
Proof: By (25.26) and quantum Ito's formula we have
(Fm(t)e( u), Fm' (t)e( v)) = l t (J(s )e( u), Fm' (s )e( v)) ((u, m))(ds)
(25.27)
+ l t (Fm(s)e(u), J(s)e(v))((m',v))(ds),
(F~, (t)e(u), F~(t)e(v)) = l t (F~, (s)e(u), J(s)e(v))((u, m))(ds)

(25.28)
+ l t (J(s)e(u),F~(s)e(v))((m', v))(ds)
+(e( u), e( v)) ((m', m)) ([0, t]).
Adding (25.27) and (25.28) and using Proposition 25.19 we obtain the required
rewk •
Proposition 25.21: For any u, v in '1JC, t 2:: °and any two ~-martingales m, m'
(F~(t)e(u),Fml(t)e(v)) + (F~,(t)e(u),Fm(t)e(v)) = 0.
Proof: By (25.26) and quantum Ito's fonnula we have
(Fr1(t)e( u), Fm' (t)e( v)} = lo\J(s)e( u), Fm' (s)e( v))((m, v))(ds)
(25.29)
+ lot (Fr1(s)e( u), J(s)e(v)}((m', v))(ds).
Here we interchange m, m' add the resulting identity to (25.29) and use Proposition
25.19. Then we obtain the required result. •
Proposition 25.22: The operator Fm(t) defined by (25.26) has a unique bounded
extension to r 8 (~).
Proof: Putting m = m' in Proposition 25.20 we have for any vector 't/J in ~(~)
IlFm(t)'t/J112 + 11Fr1(t)'t/J11 2 = ((m, m))([O, t])ll 't/J 112.
Thus
IlFm(t)'t/J112 S ((m, m))([O, tJ) 11't/J112.
Since ~ (~) is dense in r 8 (~) the required result is proved.
•
Proposition 25.23: Let Fm(t) and F,h(t) denote the closures of the corresponding
operators defiend by (25.26). Then
(i) Fm(t) and F,h(t) are bounded. F,h(t) is the adjoint of Fm(t);
°
(ii) For any s ~ 0, t ~ and any two e-martingales m, m'
(a) Fm(s)Fm/(t) + Fm/(t)Fm(s) = 0,
(b) Fm(s)F~/(t) + F~/(t)Fm(s) = ((m,m'))([O,s 1\ tJ).
Proof: (i) is immediate from Proposition 25.22 and the fact that (Fm, F,h) is an
adjoint pair of (e, ~)-adapted processes. When s = t, (a) and (b) in (ii) follow
respectively from Proposition 25.21 and 25.20. Now suppose s < t. Define a new
e-martingale m" by putting
m/(td = m(td if tt S s,
=m(s) if tt>s.
Then (a) and (b) in (ii) can be rewritten as
=
Fmll (t)Fml (t) + Fm' (t)Fm ll (t) °
Fmll(t)F~/(t) + F~/(t)Fmll(t) = ((m",m'))([O,tJ).
•
Proposition 25.24: For any u in ~ let Fu(t) = Fm(t) where m denotes the
e-martingale m(t) = e([O, tJ)u. Then
fe(u) = s.limFu(t),/t(u) = s.limFJ(t)
t->oo ~ t->oo
exist. The family of operators {!{(u),fi(u)lu E~} obey CAR, i.e.
f{(u)!{(v) + f{(v)!{(u) = 0,
f{(u)fi(v) + fi(v)f{(u) = (u,v) for all u,v E~.
Proof: If u, v E ~ then the total variation ofthe measure ((u, v)) does not exceed
Ilullllvll. By Proposition 25.14 we have for any 0 :s: tl < t2 < 00, v E ~
II{Fu(t2) - Fu(tl)}e(v)112 :s: I((u, v))I([t l, t2l) exp(llullllvll + IlvI12).
Thus
and by the proof of Proposition 25.22

IlFu(t)ll:S: lIull for all t 2: o.
Now the required result follows from Proposition 25.23.
Thus, rather remarkably, starting from CCR and using quantum stochastic
•
integration and Ito's formula we have arrived at CAR. It is possible to make a
passage to CCR from CAR. In order to achieve this we need to examine quantum
Ito's formula in greater detail and recast it in a slightly different language. We do
it through the following proposition.
Proposition 25.25: Let M I , M2 be fundamental processes in rs(~). Then MIM2

is a (~, ~)-adapted process satisfying the relation
dMIM2 = MldM2 + M2 dM I + dMldM2 (25.30)
where dM IdM2 is given by the following multiplication table:
dM2
dM I dArn z dAHz dA m2
dArn[ 0 0 0
dAH[ dAk[m2 dAH[Hz 0
dA m[ d((ml, m2)) dA H2*m[ 0
Proof: By quantum Ito's formula
(Mt (t)e( u), M2(t)e( v)) = lot (Mt (s )e( u), e( v))dfJds)
+ lot (e(u), M2(s)e(v))dJLI(s) (25.31)
+ lot(e(u),e(V))dJLdS)
where J.l2 = ((u, m2)), ((u, H2V)) or ((m2' v)) according to whether M2 = Atn2' AH2
or A m2 ,J.l[ = ((u,m[)),((u,H[v)) or ((m[,v)) according to whether M[ =
Atn" AH, or AmI and J.l[2 is given by the following table:
M2
M[ Atn2 AH2 Am2
Atn, 0 0 0
AH, ((u,H[m2)) ((u, H[H2v)) 0
AmI ((mt, m2)) ((Him[, v)) 0
Rewriting the first integral on the right hand side of (25.31) as J~ (e (u),
M[ (s )e( v)) dJ.l2 (s) and using the first fundamental lemma we recognise (25.30)
from (25.31). •
Proposition 25.26: Let (Ei ,El), (Fj,FJ) be adjoint pairs of processes in IL(~,
ho, 'JC). Suppose
dX = L.,EidMi + L.,FjdTj,
i j
dXt = "EtdM!
~ ~ ~
+ "FtdT
~ J
i j
where (Mi' Ml) is an adjoint pair of fundamental processes and Tj E '!f~ for all
i,j. Let Gk,Hl be processes in 1L(~,Do,M) and
dY = L.,GkdMk + L.,HldTl
k l
where Mk is a fundamental process and Tl E '!f ~ for all k, t. Suppose that XY is

defined as a (~,Do,M)-adapted process and XGk,XHl, EiY,FjY and EiGk are
all defined as processes in 1L(~,Do,M). Then
dXY = L.,XGkdMk + L.,XHldTl

k l
+L.,EiYdMi + L.,FjYdTj (25.32)

i j
+ L.,EiGkdMidMk
i,k
where dMidMk is defined by Proposition 25.25.
Proof: Let J E ho, 9 E Do, u E H, v E M. By hypothesis and quantum Ito's

formula we have
(f(e(u),X(t)Y(t)ge(v)) = (Xt(t)Je(u), Y(t)ge(v))

= (xt(O)Je(u), Y(O)ge(v))
+~ j (El (s )Je(u), Y(s)ge( V))dJ.1.i(S)

,
+ ~ j (Fj(s)Je(u), Y(s)ge(V))dTj(S)
J
+L j(xt(S)Je(U),Gk(S)ge(V))dJ.1.~(S)
k
+ L j (xt(s)Je(u), Hl(s)ge(v))dTl(s)
l
+ L j(El(s)Je(u),Gk(s)ge(v))dJ.1.ik(S)
i,k
(25.33)
where the measures J.1.i' J.1.~ and J.1.ik are determined appropriately depending on the
fundamental processes Mit and Mk. By hypothesis
(fe( u), X(O)Y(O)ge( v)) = (xt (O)Je( u), Y(O)ge( v))
and the integrands in the five sums on the right hand side of (25.33) are respec-
tively equal to (fe(u), N(s)ge(v)), N(s) = Ei(S)Y(S), Fj(s)Y(s), X(S)Gk(S),
X(S)Hl(S) and Ei(S)Gk(S). Now an appeal to the first fundamental lemma yields
the relation (25.32). •
If we adopt the notation
rt XdY L iort XGkdMk + L iort XHldTi,

io
=
k l
I t(dX)Y = L It EiYdMi + L It FjYdTj

o i 0 j 0
in Proposition 25.26 then (25.32) can be expressed as

dXY = XdY + (dX)Y + dXdY
where
dXdY = LEiGkdMidMk.
i,k
This is a more specialised version of quantum Ito's formula which can be used as
a convenient mnemonic for computational purposes while applying quantum Ito's
formula. It may also be noted that the first fundamental lemma is a special form
of quantum Ito's formula if one puts Ki = Lj = 0, X(t) = X(O) = 1 in Theorem
25.13.
204 Chapter III: Stochastic Integration and Quantum Ito' s Formula
Exercise 25.27: Let mj, j = 1,2, ... ,n be ~-martingales. Then (A m ,Am2

Am n, Atnn Atn n_, ... Atn,) is an adjoint pair of (~, ~)-adapted processes in
r s (~) satisfying the relations
dAm, ... Am n = LAm, ... Amj ... Am ndAmj ,
j
dA mn
t ... Atm! ="
L....t At
mn
. .. Atmj . .. Atml dA mj'
t
j
where ' over an expression implies its omission.
Exercise 25.28: Let H j , j = I, 2, ... , n be bounded operators in r s (~) which

commute with ~. Then AH, ... AHn is a (~, ~)-adapted process satisfying the
relations
it ,i2"" .ir
l~r:5n
where the summation is over all non-empty subsets i 1 < i2 < ... < ir of
{1,2, ... ,n} and in the summand A Hi " . . . ,A Hir are omitted in the coefficient
of dA Hi , ... Hi r '
Example 25.29: According to the table of Proposition 25.25, for any ~-martingale
m
{d(Am + Atn)}2 = d((m, m)). (25.34 )
From the definitions of Am, Atn we note that the operator Am(t) + Atn(t) is
the restriction of the selfadjoint operator q( md = -p( imt) (See (20.19» to the
core '€,(~). In the vacuum state the commuting family {q(mdlt ~ O} is a
Gaussian stochastic process with independent increments, mean 0 and variance at
time t being (mt, mt) = ((m, m))([O, t]). Thus (25.34) captures the essence of the
classical Ito formula for such Gaussian processes.
If ~ = L2(1~+) with respect to the Lebesgue measure, ~ is the canonical
spectral measure and mt = I[o.t] for all t then {q( mt)lt ~ O} is standard Brownian
motion expressed in the framework of quantum probability and (25.34) is the
equivalent of the well-known relation (dw( t)? = dt for the standard Brownian
motion {w(t)} in classical probability.
Example 25.30: Let

X(t) = Al(t) + Am(t) + AJn(t) + ((mt,mt))
where m is any ~-martingale. Then X(t) is essentially selfadjoint on '€,(H). The
closures of X(t), t ~ 0 constitute a commuting family and in the vacuum state 
describe a Poisson process with intensity measure ((m, m)). Indeed, this follows
from Proposition 21.4, 21.6. By the table of Proposition 25.25 we obtain (dX)2 =
dX which is Ito's formula for a Poisson process in classical probability.
Exercise 25.31: For any ~-martingale m
where J, F m are as in Example 25.18 and Proposition 25.19-25.23. If is the

vacuum vector in r s ('Z1e) and mj, j = 1, 2, ... , n are ~ -martingales then
for n = 1,2, ... , e( (J) being the signature of the permutation (J in n symbols.
Exercise 25.32: For any U in 'Z1e let Fu(t),FJ(t) be the operators determined by
Proposition 25.24. Then each of the subsets
{} U {At (t)··· AtJt)lul,'" , Un E 'Z1e, n = 1,2, ... },

{}U{FJ1(t).·.FJJt)lul, ... , Un E'Z1e, n=I,2, ... },
is total in rs('Z1e tj ) for every t > O. (Hint: Use Exercise 25.31.)
Exercise 25.33: Let f~( u), f1 (u), U E 'Z1e be as in Proposition 25.24. Then the set
{} U {fJ(uJ) ... fJ(un)lul, ... , Un E 'Z1e, n = 1,2, ... }

is total in r s('Z1e). By Proposition 25.24 and Exercise 23.5, there exists a unitary
isomorphism Ufo : r s ('Z1e) ~ r a ('Z1e) such that Ufo is the fermion vacuum vector
in r a ('Z1e) and
where a( u) denotes the fermion annihilation operator associated with u.

If ~, 'TJ are two non-atomic spectral measures in ~+ with values in 1jf('Z1e)
then there exists a unitary operator U(~,'TJ) in rs('Z1e) such that
Such a U(~, 'TJ) is determined uniquely up to a scalar multiple of modulus unity.

In particular, if ~, 'TJ, ( are three non-atomic spectral measures in ~+ with values
in 1jf('Z1e) then
U('TJ,()U(~,'TJ) = c(~,'TJ,()U(~,()
where c(~, 'TJ, () is a scalar of modulus unity.

Exercise 25.34: Let 'M, = L 2(1ffi+), ~ = the canonical spectral measure on

Iffi+, mt = and let ftr be the symmetric measure in (f(Iffi+), 9i'r) (as in
I[o,t]
Example 19.12 and Exercise 20.23). Then m = {mtlt ~ O} is a ~-martingale.
Define the fundamental processes
(25.35)
Let U : f s('M,) ---+ L2(ftr) be the unitary isomorphism defined in Example 19.12.
Suppose
a(u) = s.lim U fO u(s)J(s)dA(s)U- 1

t--+oo io
at(u) = s.limU ft u(s)J(s)dAt(s)U-1
t--+oo io
for each u in 'M,. (See (25.26) and Proposition 25.24.) Let J(t) = UJ(t)U- 1.
Then
(J(t)f)( (J) = ( -1 )#D"n[o.t] f( (J),
(a(u)f)((J) = 1 00
u(s)(-l)#D"n[o,s]f((JU{s})ds,
(at(u)f)((J) = 2) -l)#D"n[o,s]u(s)f((J\{s}).
sED"
Then {a(u), at(u)lu E L2(Iffi+n is a realisation of CAR in L 2(ftr) (Hint: Use

Exercise 20.23.)
Exercise 25.35: [116] Suppose {(hit ~ O} is the one parameter semigroup of

shift isometries in L2(1ffi+) defined by
0 if x < t,
UJtu)(x) = { u(x - t) if x ~ t.
Let A,At be as in (25.35). The families {f(Ot)it ~ O} and {f(O;)it ~ O} are not
(~, 'M,)-adapted. For any partition 7r = {tj} as defined in (25.2) and any u in 'M,
define the (Riemann-Stieltjes) sums
£1r(u) = Lu(tdf(Oti+J{A(tHd - A(td},

i
£t(u) = Lu(ti){At(tHd - At(tinf(OtJ

i
on the domain ~(L2(1ffi+)). Then the limits

£(u) = lim £1r(u)e(v),£t(u) = lim £t(u)e(v), v E 'Z1C
11rI--+o 11rI--+o
exist where 17r1 = sUPi(tHl - ti) is the width of the partition. Furthermore £(u)
and £t(u) extend uniquely and linearly to bounded operators on fs('Z1C). Denote
26 A class of quantum stochastic differential equations 207
the extensions by the same symbols. Then the correspondence u -7 £t (u) is linear
on L2(~+),£t(u) = £(u)*,£(u) = O,£(u)it(v) = (u,v) for all u,v and the set
{} U {£t (uJ) ... £t (un)IUj E 'Je, I ~ j 5: n, n = 1,2, ... } is total in f s('Je). In
other words {£( u), £t (u) lu E 'Je} is a realisation of free annihilation and creation
operators in the boson Fock space fs(L2(1R+)). We may write
£(u) = 1 00
u(s)f(O;)dA(s),£t(u) = 1 00
u(s)dAt(s)f(Os)
but keep in mind that these are integrals of non-adapted processes.
Notes
This section is a leisurely account of quantum stochastic integration in boson Fock
space based on Hudson and Parthasarathy [S9]. Example 2S.18 and Proposition
2S.19-2S.24 are adapted from Hudson and Parthasarathy [62] and Parthasarathy
and Sinha [lOS]. Exercise 2S.3S is from Parthasarathy and Sinha [116]. For an
account of quantum stochastic integration based on kernels in a symmetric measure
space, see Maassen [83], Meyer [88], Lindsay and Maassen [76,77]. For other
approaches to integration based on von Neumann algebras, we refer to Barnett,
Streater and Wilde [16], [17] and Accardi and Fagnola [S]. A brief account of this
section is included in [113].
26 A class of quantum stochastic differential equations

Let ho be the initial Hilbert space and let 'Je be a Hilbert space equipped with a
continuous time observable ~ taking values in ~+ = [0,00). Suppose Ki, (1 ~
i ~ k), Lj, (1 ~ j ~ f), Ki" (1 ~ i' ~ k'), Lj" (1 ~ j' ~ £') are bounded
operators in ho. In 'ie = ho @ f s('Je) introduce the constant adapted processes
Ki(t) = Ki @ 1, Lj(t) = Lj @ 1, Ki, (t) = Ki, @ 1, Lj, (t) = Lj, @ 1
for 1 5: i 5: k, 1 ~ j ~ £, 1 5: i' 5: k', 1 5: j' 5: .e'. We adopt the convention

of denoting by Ki the operator Ki in ho as well as the constant adapted process
{Ki(t)lt ;::: O}. Note that 1 here denotes the identity operator in fs('Je) and
Ki, Lj' Ki"Lj, are all (~,ho,'Je)-adapted. Suppose Mi, (1 ~ i ~ k), Mi" (1 ~
i' ~ k') are fundamental processes with respect to ~ and 7j, (1 ~ j ~ f), 7 1"
(1 ~ j' ~ £) are complex valued measures in the linear space 21;. (See Example
24.1, 24.2 and the definitions preceding Proposition 2S.7 and after Proposition
2S.12.) Examples 2S.1S-2S.17 suggest the study of the following simple equation
in the unknown (~, ho, 'Je)-adapted process X:
k l k' l'
dX = XfL:KidMi + LLjd7j} + LKi,dM;' + LLj,d71"X(0) = Xo
i=1 j=1 i'=1 j'=1
(26.1 )
where Xo is of the fonn Xo ® 1, Xo being a bounded operator in ho. Equation

(26.1) has to be interpreted as the stochastic integral equation:
X(t) = Xo + ~ lot X(S)Ki(S)dMi(S) + ~ lot X(s)Lj(s)dTj(S)

l J
(26.2)
+~ lot K:,(s)dMI,(s) + ~ lot Lj, (S)dTj, (s)
l J
for all t 2': O. In this context we have the following proposition.
Proposition 26.1: There exists a unique regular (~, ho, :1e)-adapted process X =
{X(t)lt 2': O} satisfying (26.1) and the condition
sup sup IIX(t)fe(u)11 < 00 for every u E 7Je, T 2': O. (26.3)
O::;t::;T 11/119
Proof: Define
C(t) = ~ lot K:, (s)dMI, (s) + ~ lot Lj, (S)dTj, (s),

l J
Xo(t) = Xo + C(t),
Xn(t) = Xo + ~
i
it
0
Xn-lKidMi + ~
j
it
0
Xn-lLjdTj + C(t), n = 1,2, ...
(26.4)
At the n-th stage observe that X n - 1 is a regular (~, ho, :1e)-adapted process,
Xn-1Ki and Xn-1Lj are in IL(~, ho, 7Je) and hence Xn is also defined as a regular
(~, ho, 7Je)-adapted process. From (26.4) we have
d(Xn - Xn-d = (Xn- 1 - Xn-2){LK i dMi + LLjdTj}.

i j
Let Au denote the measure obtained by summing up the Vi'S, the variations of the
J-ti'S and the Tj'S that occur in the the statement of Proposition 25.14. Then we
have for any fixed T > 0
II(Xn(t) - X n_ 1(t))fe(u)11 2 :::;

(k + f)eAu([O,t]) lot {~II(Xn-l(S) - Xn_2(s))(Kd)e(u)112
l
+ ~ II(Xn-l(S) - Xn-2(S))(Ljf)e(u)ln Au(ds)

i
for all 0 ~ t ~ T. From this inequality a straightforward induction in n leads us

to
II(Xn(t) - Xn_l(t))je(u)112 ~
{(k + ftenAu([O,T]) L IIXoSilSi2'" SinjI121Ie(u)112}
l:Sij :Sk+l
j=1,2, ... ,n
x r
Jo<t 1 < .. ·<t n <t
Au(dt l )A u(dt 2) .. · Au(dt n )
where Si, i = 1, 2, ... , k + f is an enumeration of the sequence K I, ... , K k>

L 1, •.• , Lt. Using the non-atomicity of the measures Au we conclude that
where
CT = (k + f)2 eAu ([O,T]) max{IIKill, IILjlll1 ~ i ~ k, 1~ j ~ fl·
In particular,
00
L sup II(Xn(t) - Xn_l(t))je(u)11 < 00 (26.6)

n=IO::;t::;T
for every T 2: 0, j E ho, u E '!fe. Define the operators X(t) in the domain
ho 09 '& ('!fe) by putting
00
X(t)je(u) = Xoje(u) + C(t)je(u) + L(Xn(t) - Xn_l(t))je(u)

n=1
= n--->oo
lim Xn(t)je(u)
and extending by linearity with the help of Proposition 19.4 and Corollary 19.5.
Inequality (26.6) ensures the existence of this limit uniformly in every bounded
interval and hence the map t ----+ X(t)je(u) is well-defined and continuous for
each j in ho and u in '!fe. Inequality (26.5) implies that X satisfies (26.3). Thus
X = {X(t)lt 2: O} is defined as a (~, ho, '!fe)-adapted regular process. Now (26.3),
(26.6) and the definition of stochastic integrals imply that X satisfies (26.2). This
proves the existence of a solution of (26.1).
Now suppose that X' = {X'(t)lt 2: O}, X'(O) = X o is another solution of
(26.1) satisfying
sup sup IIX'(t)je(u)11 < 00.
O::;t::;T Iltll::;1
Define Yet) = X(t) - X'(t). Then
Yet) = ~ !at Y(S)KidMi(S) + ~ !at Y(s)LjdTj(s).

t J
Once again by the application of Proposition 25.14 we have
IIY(t)je(u)112 ~ (k + l)eA,,([O,T]) it{I: z

IIY(s)(Kd)e(u)112
(26.7)
+ I: IIY(s)(Ljf)e(u)112}Au(ds)
j
for all 0 ~ t ~ T, j E ho, u E 'Je. Let

a(t) = sup IIY(t)je(u)112,
11/119
C= max{IIKiI12, IILjl1211 ~ i ~ k, 1~ j ~ l}.
Then (26.7) implies
(26.8)
where
bT = C(k + If exp A.u([O, T]).
By induction (26.8) implies
a(t) ~ b?}. r
JO<tl <---<tn <t
a(tt)A.u(dtt)··· A.u(dt n ).
If aT = sUPO::;t::;Ta(t) then we have

bJi.A. u([0, t])n
a(t) <
-
aT
n!
for all n = 1,2,. _.
In other words a(t) = 0 for 0 ~ t ~ T. Since T is arbitrary we have Y(t)je( u) =

o for all jE ho, u E 'Je, t 2 o. •
We shall now investigate a special case of (26.1) which ultimately leads to
a solution X (t) which is a unitary operator for each t. That such a possibility
should exist is already revealed by (25.23) in Example 25.17. To this end consider
the equation
dX = X(L1dAtnl + L2dA H + L3dAm2 + L 4 d7),X(0) = 1 (26.9)

where each Lj is a constant adapted process of the form Lj(t) = Lj (>9 1,Lj E
'lf3( ho), 1 denotes the identity operator in r s ('Je), ml, m2 are ~-martingales, 7 E ?J{
and H is a bounded operator in 'Je commuting with~. We now impose the condition
that the unique solution X (t) provided by Proposition 26.1 is the restriction of
an isometry to ho (>9 ~ for every t and examine the conditions to be fulfilled by
m], m2, 7, H, Lj' 1 ~ j ~ 4. Suppose that
(X(t)je(u),X(t)ge(v)) = (fe(u),ge(v))
for all t ~ 0, f,g E ho, u,v E '1fC. By quantum Ito's formula (Theorem 25.13) we
obtain the following identity in complex valued measures:
(f,L1g)((u,ml)) + (f,L 2g)((u, Hv)) + (f,L3g)((m2,v)) + (f,L4g)T

+(LJ/,g)((ml,v)) + (Lz/,g)((Hu,v)) + (L3f,g)((u,m2))
(26.10)
+(L4f,g)T + (LJ/, L1g) ((mJ, ml)) + (L2f,L 2g)((Hu,Hv))
+(LJ/, L2g)((ml' Hv)) + (Lz/,L1g)((Hu,ml)) = °
for all f, g E ho, u, v E '1fC. A sufficient set of conditions under which (26.1 0) is
obtained is the following:
(i) ml = m2 = m, T = ((m, m));

(ii) H = P is a projection commuting with ~ and Pmt = mt for all t ~ 0;
(iii) L2 = 8 - 1 where 8 is an isometry in ho;
(iv) LI = L, L3 = -L*8
(v) L4 = iHo - ~L* L where Ho is a bounded selfadjoint opeator in ho.
Motivated by these considerations we study the following equation:
dX = X(LdAtn + (8 -1)dAp - L*8dAm + (iH - ~L*L)d((m,m))),

X(O) = 1
(26.11 )
where m is a ~-martingale, P is a projection in '1fC commuting with ~ and satisfying
the condition Pmt = mt for all t, L, 8, H are bounded operators in ho,8 is an
isometry and H is selfadjoint.
Proposition 26.2: There exists a unique isometric operator-valued regular (~, ho,
'1fC)-adapted process X satisfying (26.11).
Proof: By Proposition (26.11) there exists a unique regular (~, ho, ;}'f)-adapted
process X satisfying (26.11) and (26.3). In particular, for each t ~ 0, u, v E '1fC
we have
sup I(X(t)fe(u),X(t)ge(v))1 < 00.
Ilfll~I.llgll~1
Thus there exists a bounded operator B(t) = Bu,v(t) in ho such that
(f, B(t)g) = (X(t)fe(u), X(t)ge(v)) (26.12)
for all t ~ 0, f,g E ho, B(O) = exp(u, v). By Theorem 25.13 we have from
(26.11) for any fixed j, 9 E ho

(J, B(t)g) - (J, B(O)g) =
lot{(J,B(S)L9) - (L*Sj,B(s)g) + ((S -1)j,B(s)Lg)}((u,m))(ds)
+ lot {-(J, B(s)L* Sg) + (Lj, B(s)g) + (Lj, B(s)(S - l)g) }((m, v)) (ds)
+ lot{(J,B(S)(S -1)g) + ((S -1)j,B(s)g)

+((S - l)j,B(s)(S - l)g)}((u,Pv))(ds)
+ lot{(J,B(S)(iH - ~L*L)9) + ((iH - ~L*L)j,B(S)9)

+(Lj, B(s )Lg) }((m, m))(ds).
We can express this as the integral equation
B(t) - B(O) =
lot S*[B(s),L]((u,m))(ds) + lot[L*,B(S))S((m,V))(dS)
+ lot {S* B(s)S - B(s)} ((u, PV)) (ds)
+ lot {i[B(s),H) + L* B(s)L - ~(L* LB(s) + B(s)L* L)}((m,m))(ds).

Putting C(t) = B(t) - B(O) we observe that C(O) = 0 and the integral equation
above can be expressed in differential notation as
dC = S*[C, L)d((u, m)) + [L*,C)Sd((m,v)) + (S*CS - C)d((u,pv))
(26.13)
+{i[C,H) + L*CL - ~(L*LC + CL*L)}d((m,m)).
In the Banach space OO(ho) the maps C - t S*[C,L), [L*,C)S, S*CS - c,
i[C,H] + L*CL - i(L* LC + CL* L) are bounded operators. If we denote by A
the sum of the variations of the measures ((u, m)), ((m, v)), ((u, Pv)) and ((m, m))
then (26.13) implies an inequality of the form
IIB(t)1I ::; a lot IIB(s)lldA(s), t ~0

where a is a constant and A is non-atomic and totally finite in every bounded
interval. As in the proof of Proposition 26.1 we conclude that B(t) = 0 for all t.
Now (26.12) implies that
(X(t)je(u),X(t)ge(v)) == (Je(u),ge(v)).
In other words X(t) extends to an isometry of'ie.
•
Theorem 26.3: [59] Let L, 8, H be bounded operators in ho where 8 is unitary

and H is selfadjoint. Suppose P is a projection in 'JC commuting with e and m is a
e-martingale such that Pmt = mt for all t ~ O. Then there exists a unique unitary
operator-valued (e, ho, 'JC)-adapted regular process U = {U(t)lt ~ O} satisfying
dU = U{LdAtn + (8 -1)dAp -L* 8dAm + (iH - ~L* L)d((m,m))}, U(O) = 1.
Proof: By Proposition 26.2 there exists an isometric operator-valued adapted

process U satisfying the required equation. Then the adjoints U(t)* of the operators
U (t) constitute a process U* satisfying the equation
dU* = {-8* LdAtn + (8* - l)dAp + L*dAm - (iH + ~L* L)d((m, m))}U*
where for any fundamental process M and any measure T E '?J{ we adopt the
convention (dM)U* = U*dM and (dT)U* = U*dT. Once again by Theorem
25.13 (in the form described in Proposition 25.26 and the subsequent discussion)
d(UU*) =
U{LdAtn + (8 - l)dAp - L* 8dAm + (iH - ~L* L)d((m,m))}U*
+U{-8*LdAtn + (8* -1)dAp +L*dAm - (iH + ~L*L)d((m,m))}u*

+U{ -(8 - 1)8* LdAtn + (8 - 1)(8* - l)dAp + L* Ld((m, m))
-L* 8(8* - l)dAm}U*
=0.
Since U(O) = 1 it follows that U(t)U(t)* = 1 for all t. This shows that U(t) is
unitary for all t. •
Corollary 26.4: Let L, 8, H be bounded operators in ho where 8 is unitary and
H is selfadjoint. Suppose P is a projection in '!Je commuting with e and m is a
e-martingale such that Pmt = mt for all t ~ O. Then there exists a unique unitary
operator-valued (e, ho, '!Je)-adapted regular process U satisfying the equation
dU = {LdAtn + (8 -1)dAp -L* 8dAm - (iH + ~L* L)d((m,m))}U, U(O) = 1.

(26.14)
Proof: This is simply the adjoint version of the equation in Theorem 26.3 when
8 is replaced by 8* and -8* L is replaced by L. Thus the required result follows
from Theorem 26.3. •
Suppose the e-martingale m satisfies the additional condition that ((m, m))
is the standard Lebesgue measure in IR+. If 8 = 1, L = 0 then (26.14) becomes
the classical Schrodinger equation dU = -iHU dt with energy operator H. This
suggests the interpretation of (26.14) as a Schrodinger equation in the presence of

noise (or a quantum probabilistic analogue of the Langevin equation [95]). The
next corollary presents the Heisenberg picture of (26.14). For any X in !J3(h o)
define
jt(X) = U(t)* X ~ lU(t), t ~ 0 (26.15)
where U(t) is the solution of (26.14) and 1 is the identity operator in rs(J'C).
Corollary 26.5: For any X in !J3(ho), {jt(X)lt ~ O} is a regular (~, ho, J'C)-
adapted process satisfying the stochastic differential equation
djt(X) = jt(S*[X,L])dAtn + jt(S* XS - X)dAp + jt([L*,X]S)dAm
+jt(i[H,X]- ~{L* LX + XL* L - 2L* XL} )d((m, m))
(26.16)
where jo(X) = X~1. The map X ---+ jt(X) is a *-unital homomorphism from
!J3(ho) into !J3 tJ for every t where !J3t] is defined by (24.1). For every X in !J3(ho)
the map t ---+ jt(X) is strongly continuous.
Proof: That jt is a *-unital homomorphism from !J3(ho) into !J3 tJ is obvious from
(26.15) and the fact that U(t) E !J3 tJ for all t. Since U(t) is strongly continuous in
t it follows that the map t ---+ jt(X) is also strongly continuous. To prove (26.16)
we first observe that
dXU = XLUdAtn + X(S - l)UdAp
-XL* SUdAm - X(iH + ~L* L)Ud((m, m)),

dU* = U* L*dAm + U*(S* - l)dAp
-U* S* LdAtn + U*(iH - ~L* L)d((m, m))

and apply quantum Ito's formula
dU* XU = (dU*)XU + U*d(XU) + (dU*)d(XU)
as described in Proposition 25.26 and the succeeding comments.
In the notations of Section 24 we recall that the Hilbert spaces J'C
•
ho ~ rs(J'C),~tJ = ho ~ rs(J'C tJ ) and ~[t = rs(J'C[t) satisfy the relation ~ =
~tJ ~ ~[t so that
!J3(~) = !J3(~tJ) ~ !J3(~[t) for all t.
Denote by [t the vacuum vector in ~[t and let lEt : !J3(~) ---+ !J3(~tJ) denote the
conditional expectation map with respect to the pure state [t so that
('I/J, (IEtX)'I/J') = ('I/J[t, X ('I/J' [t)) for all 'I/J, 'I/J' E ~tJ' X E !J3(~)
for all t ~ O. Define the conditional expectation map IEtJ : !J3(~) ---+ !J3 tJ by
IEtJX = (IEtX) ~ l[t for all X E !J3(~), t ~0

where l[t is the identity operator in ;}e[t. Then [Et] is a *-unital positive linear map
satisfying the conditions:
(i) 1E8]lEt] = 1E811t] where s 1\ t = min (s, t);
(ii) IEt]AXB = A(lEt]X)B for all A,B E ~t], X E ~(;}e);
(iii) L! <;i,j<;nYi* ([Et]xt Xj)Yj 2: 0 for all Xi E ~(;}e), Yi E ~t], 1 :::; i :::; n,
n = 1,2, ...
(See Exercise 16.10, 16.11.)
Proposition 26.6: Let X, E i , Fj, 1 :::; i :::; k, 1 :::; j :::; .e be bounded operator-
valued (~, h Q, X)-adapted processes satisfying
k l
dX = LEidMi + LFjdTj
i=! j=!
where M i , 1 :::; i :::; k are fundamental processes and Tj, 1 :::; j :::; .e are complex
valued measures from ' j~. Then
IEs]X(t) = X(s) + L it{[ES]Fj (t')}Tj (dt') for all s < t.

J
Proof: Let f, g E hQ, U, v E X and let ~([s, oo))u = ~([s, oo))v = O. Then for
any ~-martingale m and any bounded operator H commuting with ~ we observe
that the measures ((m, v)), ((u, Hv)) and ((u, m)) have zero mass in the interval
[s, (0). Hence by the first fundamental formula
(fe(u), it EidMige(V)) = it (fe(u), Ei(t')ge(v))dJ.li(t') = 0,
J.li being a measure of the form described above. Furthermore
(fe(u), it FjdTjge(v)) = it (fe(u), Fj(t')ge(v))Tj(dt').
By the differential relation satisfied by X we obtain
X(t) = X(s) +L
k
i=!
1
8
t
EidMi +L
l
j=!
18
t
FjdTj.
Hence
(fe(u),X(t)ge(v)) = (fe(u),X(s)ge(v)) +L
l
j=! s
1 t
(fe(u), Fj(t')ge(V))Tj(dt').
Since the set {fe( u) If E hQ, u E X, Ws, oo))u = O} is total in ;}es]} the required
result follows. •
Proposition 26.7 In Corollary 26.5 let m be a ~-martingale such that ((m, m)) is
the standard Lebesgue measure in IR+. Let 8 : C!A(h o) -+ C!A(h o) be the bounded
linear map:
8(X) = i[H,X]- ~{L* LX + XL* L - 2L* XL}
and let Tt = et9 be the one parameter semigroup generated by 8. Then

IEs]it(X) = is(Tt-s(X)) for allX E C!A(ho), 0~s~t < 00. (26.17)
Furthermore, if 0 < tl < t2 < ... < tn < 00 and Xi E C!A(ho), 1 ~ i ~ n then
1E0itl (Xl)'" itn(Xn) = Ttl (XITt2-t l (X2'" (Xn-,Ttn-t n_l (Xn ))·· .).
(26.18)
Proof: From (26.16) and Proposition 26.6 we have
IEs]it+s(X) = is(X) l
+ s
s +t
IEs]id8(X))dt' for all t ~ 0, X E C!A(ho).
Fix s, write (/Jt(X) = IEs]it+s(X) and observe that the integral equation above
can be expressed as
d¢t A-. 8
& = 'Pt O ,
A-.
'PO = Js•
so that ¢t = i soe t9 = isoTt for all t ~ O. This proves (26.17). To prove (26.18)
we proceed exactly as in the proof of Corollary 18.4. We have
lEo]itJX,) ... itn (Xn) = 1E0]lEtn_didXd ... jt n(Xn)
= lEO] (jdXd ... jtn_JXn-d{lEtn_djtn (Xn)))
= 1E0jtl (Xd'" itn_2(Xn- 2)jtn_1 (Xn-,Ttn-t n_l (Xn)).
Now repeating the same argument successively we obtain
1E0]jtl (Xd··· jt n(Xn) = jo{TdX,Tt2-dX2··· (Xn-ITtn-tn_JXn))"'))
which is equivalent to (26.18).
•
Corollary 26.8: Let jt, 8 and Tt be as in Proposition 26.7. Then the following
holds:
(i) Iljt(X)11 ~ IIXII for all t ~ 0, X E C!A(ho);
(ii) {Tt} is a uniformly continuous one parameter semigroup of positive *-unital
linear maps of C!A (ho) satisfying
IITt(X)11 ~ IIXII,
(26.19)
'~i,j~n
1 ~ i ~ n, n = 1,2, ...
Proof: (i) is immediate from (26.15) and the unitarity of U(t). The first part of
(ii) is immediate from the properties of conditional expectation. To prove the last
inequality observe that
ja( L Yi*Tt(Xt Xj)Yj) =
lEa] L (Yi 129 l)*jt(Xt Xj)(Yj 1291)

l'Si.j'Sn
= lEa] {2:)t(Xi )(Yi Q91n*{Ljt(Xi)(Yi 129 In ~ O.

i i
When the ~-martingale m satisfies the condition that ((m, m)) is the Lebesgue
measure in ~+ it follows from Corollary 26.5, Proposition 26.7 and Corollary 26.8
that the family {jtlt ~ O} of *-unital positive contractive and adapted homomor-
phisms from CJ3 (h a) into CJ3 (h a129 r s C?I£)) defined by (26.15) is, indeed, a continuous
time analogue of the discrete time quantum stochastic flow described in Section
18.
Example 26.9: Let ha be a Hilbert space with orthonormal basis fa, fI, ... ,fn, ...
(where the pure state fn may be interpreted as the state in which the num-
ber of particles in a system is n). As in Exercise 20.18 introduce the number
operator N and the free annihilation operator L by N = I:~dlfj)(1i1 and
L = I:~II1i~I)(1i1 respectively. Let Ll = J.l(N + I)L, H = v(N) where J.l
and v are bounded functions on {O, 1, 2, ... }, J.l being complex valued but v real
valued. Let 'M = L2 (~+), m( t) = I[a,t] and let ~ be the canonical spectral mea-
sure in 'M. Then m = {m(t)lt ~ O} is a ~-martingale. Define A(t) = Am(t),
At (t) = Atn (t) and consider the quantum stochastic differential equation
dU = {LldAt - LtdA - (iH + ~LtLddt}U, U(O) = 1.

By Theorem 26.3 there exists a unique unitary operator-valued solution U
{U(t)lt ~ O} for this equation. Such a solution induces a quantum stochastic flow
{jtlt ~ O} through
jt(X) = U(t)* X 129 lU(t), t ~ 0, X E CJ3(ha)
for which the Markov property IEs]jt(X) = js(Tt~s(X)) holds for all s <t where
Tt = et () and
O(X) = i[v(N),X]- ~{L*LIJ.l(N)12X +XL*IJ.l(N)12

- 2L*M(N + I)XJ.l(N + I)L}
for all X in CJ3(ha). When X = </J(N) where </J is any bounded function on
{O, 1,2, ... } we obtain O(</J(N)) = (M</J)(N) where
(M</J)(k) = 0 if k = 0,
= 1J.l(k)12{</J(k - 1) - </J(kn if k ~ 1.
We observe that M is the generator of a classical Markov pure death process with
death rates 1j.t(kW, k = 1,2, ...
If Ll is replaced by L2 = L * >.( N) in the discussions above where >. is a
bounded complex valued function on {O, 1,2, ... } then the corresponding 0 map
is given by
O(X) = i[v(N), X]- ~{I>'(N)12 X + XI>.(N)12 - i5.(N)LXL* >'(N)}.
If X = <P(N) then O(<P(N)) = (M<P)(N) where

(M<P)(k) = 1>'(k)12{<p(k + 1) - <P(k)} for all k ~ O.
We observe that M is the generator of a classical Markov pure birth process with
birth rates 1>'(kW, k = 0, 1,2, ...
In short the two flows described above may be called quantizations of
pure death and pure birth processes with corresponding death rate amplitudes
j.t( 1), j.t(2), ... and birth rate amplitudes >'(0), >'(1), ... respectively. (See Exam-
ple 27.11.)
Exercise 26.10: (i) Let _ 1)/, ei </»

00
where W denotes the Weyl operator in rs(L2(~+)). Then V(<P)V("p) = V(<P+"p).

In particular, there exists a commuting family {Z( <P)} of selfadjoint operators
satisfying
V(<P) = eiZ (¢),
(e(O), V(<p)e(O)) = exp 1 00

(e i ¢ - 1)1/1 2dt.
In other words the field {Z (<P)} of commuting observables is the Poisson random
field with intensity measure having density 1/12 with respect to the Lebesgue
measure in ~+.
(ii) Define the adapted processes
V¢(t) = ei 1m lot ei¢lfI2dSW( (e i ¢ _ 1)ft, ei¢t)
where <Pt = <Pi[O,t], It = I I[o,t]· Then

dV¢ = (e i ¢ - IHldAt + dA + fdA + 1/1 2dt}V¢, V¢(O) = 1.
(iii) Consider an initial Hilbert space ho and the unitary operator-valued

adapted process U = {U(t)lt ~ O} in ho ® rs(L2(~+)) satisfying
dU = {LdAt - LtdA - (iH + ~L* L)dt}U, U(O) = 1

where L, H E !J3('Jeo) and H is selfadjoint. Define

M(t, </>; X) = lEoU(t)* XV</>(t)U(t)
where X, V</>(t) are viewed as operators in ho (9 fs(L2(1R+)). Then
:tM(t, </>, X) = M(t, </>; B(X))+(e i </> - I)M(t, </>; (f + L)* X(f + L))
where
B(X) = i[H,X]- ~{L* LX + XL* L - 2L* XL}.
(iv) Let Po be any state in ho. Then there exists a unique trace class operator
Pt in ho for every t > 0 such that
tr poM(t, </>, X) = tr PtX for all X E !J3(ho)
where
d~t = B'(pt) + (e i </> - 1)(f + L)p(f + L)*,
B'(p) = -i[H, p]- ~{LL* P + pLL* - 2LpL*}.

(v) Consider the random field {Z(</>)} in (i) and the unitary process U in
ho (9 fs(L2(1R+)) and define the new random field
Zt(</» = U(t)* Z(</»U(t).
Then
lEe iZt (</» = tr Pt
where Pt is as in (iv).
(v) has the following interpretation. If the field {Z (</>)} is observed in the
vacuum state one gets a Poisson random field for counting particles. If it is coupled
to a system in interaction with a heat bath and the system plus bath evolve under
{U(s)IO:::; s :::; t} then due to interference the Poisson field {Z(</>)} changes to
{Zt( </»} at time t and its characteristic function is given by (v). The next example
is a specialisation to the case ho = ([2.
Example 26.11: Let ho = ([2, a = (~ ~), H = w2at a, L = )..a. (Note that

a is the fermion annihilator in ([2). For any state Po in ([2 let Pt be as in (v) of
Exercise 26.10. Then
trpt = e Jo
f'lfI2(e i 4>-I)ds rt.
{I + 10 2(e~</> -1)e- z
~
x Re)..(tr poa)!(s)e- iW2S ds + tr poatall)..,21t (e i </> - l)e-IAI2sds
+ r 1)..1 2 (e i </>(stl _ 1)(e i </>(s2) _ l)e-~(SI+s2)

10<sl <sz<t
x2 Re f( SI )!( s2)e iW2 (SI -sz) dS 1dS2]}.
Changing ¢ to x¢ and differentiating at x = 0 in (v) of Exercise 26.10 and putting

Po = pa*a + qaa* + za + za* one obtains
[t ~
lEoZt(¢) = Jo ¢(sHI/(s) + Aze-(
2
2 +iW)SI2
+(p _lzI2)IAI2e-IAI2S}ds.
In particular, the expected number of "particles" of the "interfered Poisson field"
counted in the interval [a,b] c [O,t] under the evolution {U(s)IO::; s::; t} is
l a
b
{1/(s) + Aze-(
~
2 +iw
2
)S1 2+ (p -lzI2)IAI 2e- 1A1 S}ds.
2
(26.20)
It is to be noted that Po ~ 0 implies IZl2 ::; pq ::; p, p + q = 1, p, q ~ O.

Exercise 26.12: (i) Let L, H E ~(ho) and let H be selfadjoint. Consider the
unitary operator-valued processes {U(t)lt ~ O}, {W/(t)lt ~ O} obeying the equa-
tions
dU = {LdAt - L*dA - (iH + ~L* L)dt}U, U(O) = 1,
dW/ = {fdAt -fdA - ~1/12dt}W/, W/(O) = 1

(where I E L2(~+)) in ho ® rs(L2(~+)). For any density matrix Po in ho
tr(po ® le(O))(e(O)I)U(t)*W/(t)U(t) = tr Pt
where {Pt} is a family of trace class operators satisfying
-
-dp
& = B'(p) + IpL* -
1 2
ILp - -III
2 p.
(ii) In (i) let ho = C2 with canonical basis, a = (~ ~). H = w2a*a,

L = Aa. Then
tr Pt = e-~ lot 1/1 2 ds
x{I+2iIm ltXtrpoa*j(s)dS-2IA,2trPoa*aRe J j(s2)j(sJ)ds 1ds2}

O<Sl<S2<t
where j(s) = I(s) exp[-( J¥ - iW2)S].
When I is real, W = 0, Po = a*a we obtain
tr Pt = e
_1
2
t0 /
2
ds {I -IAI2(Jo[t I(s)e- ~ dsf}.
2
In this case U(t)*W/(t)U(t) = expiZt(f) where Zt(-) is a random field which

is Gaussian when A = 0 but has an indecomposable distribution for any fixed I
in the state Po ® le(O))(e(O)1 when A:I o. (See [82].)
27 Stochastic differential equations with infinite degrees of freedom 221
Notes
The exposition here is based mainly on Hudson and Parthasarathy [59]. One of
the most interesting problems in this subject is to generalise Theorem 26.3 to the
case when L and H are not necessarily bounded. In this context see Journe [68].
Fagnola [37] has analysed Example 26.9 when /-l and v are unbounded. Exercise
26.10,26.11 are inspired by the work of Davies and Srinivas [128], Srinivas [129]
and Barchielli and Lupieri [12].
27 Stochastic differential equations with

infinite degrees of freedom
The aim of the present section is to generalise Corollary 26.4 when the number
of differentials of fundamental processes (or noise terms) is possibly infinite. To
this end we introduce some notation. Let 'lie = V Q9 L2(1ffi+) where V is a finite
or countably infinite dimensional Hilbert space with orthonormal basis {ej Ij =
1, 2, ... }. Suppose ~ is the time observable 1 Q9 ~o where ~o is the canonical spectral
measure in L2(1ffi+) and 1 is the identity operator in V. Define mj(t) = ej Q9I[o,t]
for each j and note that mj is a ~-martingale. Note that the operator lej)(eil Q9 1
commutes with ~ for every i, j. Introduce the canonical fundamental processes
{A~ 10 ~ i, j < oo} associated with the basis {ej}:
Ag(t) = tI,
A~(t) = Am;(t), A~(t) = A~i (t), i = 1,2, ...
A~(t) = Alej)(eiI0 1 (t), i,j ~ 1
It is clear that (A~, A{) is an adjoint pair of fundamental processes and the quantum
Ito's formula implies
where
8i _ {O. if i = 0 or l = 0,
l - 0i otherwise,
o} being the Kronecker O. We shall view the Hilbert space 'lie as L2(1ffi+, V) so
that any element u in 'lie can be viewed as a V-valued function u(·) on Iffi+. Write
for any u in 'Je,
Ui(t) = (u(t), ei), ui(t) = (ei' u(t)) if i ~ 1,
uo(t) = uO(t) = 1.
.M = {ulu E 'lIe,Ui(t) = 0
for all but a finite number of indices i} (27.1)
222 Chapter Ill: Stochastic Integration and Quantum Ito's Formula
N(u) = max{jluj(-) is a non-zero function in L2(~+)}, (27.2)
vu(t) = lot (1 + lIu(s)1I2)ds. (27.3)
Suppose L~,Mj, i,j ~ 0 are processes in lL(e,ho,.M) see the discussion before
Proposition 25.7) where ho is a fixed initial Hilbert space. Define
Xn(t) = L it L~(s)dA{(s), (27.4)

O<:::,i,j<:::,n 0
Yn(t) = L i t Mj(s)dA{(s). (27.5)

O<:::,i,j<:::,n 0
Then the first and second fundamental lemmas assume the following form:
(Je(u),Xn(t)ge(v)) = i t L ui(s)vj(s)(Je(u),L~(s)ge(v))ds, (27.6)

o O<:::,i,j<:::,n
(Xn(t)fe(u), Yn(t)ge(v)) =
it L (Xn(s)fe(u),Mj(s)ge(V))Ui(S)Vj(s)ds
o O<:::,i,j<:::,n
+ it I: (L~(s)fe(u), Yn(S)ge(v))Uj(s)vi(s)ds (27.7)

o O<:::,i,j<:::,n
[t
I: I:
n
+ in (L~(s)fe(u),Ml(s)ge(v))Uj(s)vl(s)ds
° i=l O<:::,j,l<:::,n
for each n ~ 0, f,g E ho, u, v E.M.
Proposition 27.1: Let {L~li,j ~ O} be processes in lL(e, ho,.M) satisfying
1° LIIL~(s)fe(u)1I2dvu(s) <
t 00
i=O
00 for all t ~ 0, f E ho, u E.M, j ~ O. (27.8)
Then there exists a regular (e, ho, .M)-adapted process X satisfying the following:
(i) limn-too sUPo<::;s<:::,tIlXn(s)fe(u) - X(s)fe(u)1I = 0;
(ii) IIX(t)fe(u)1I2:::; 2eVu(t)~;~~) J; ~:oIlL~(s)fe(u)1I2dvu(s)

for all f E ho, u E .M, t ~ 0 where the notations are as in (27.1)-(27.4).
Proof: Putting Yn = X n , g = j, v = u in (27.7) we have

IIXn(t)je(u)112 =
ito
2Re L
O<::i.j<::N(u)
(Xn(s)je(u),L;(s)je(U))Ui(S)Uj(s)ds
+ It:t L (L;(s)je(u), Lt(s)je(u))Uj(s)uk(s)ds

o i=! O<::j.k<::N(u)
1 N(u)
t N(u)
= {2Re L (ui(s)Xn(s)je(u), L uj(s)L;(s)je(u))
o i=O j=O
n N(u)
+LII L uj (s)Lj(s)je(u)112}ds
i=! j=O
:;1 o
t N(u) n
{IIXn(s)je(u)112 + 2 L LIILj(s)je(u)11 2}dvu(s).
j=O i=O
By Proposition 25.5 we conclude that
IIXn(t)je(u)112 :::; 2e vu (t) 1 tN(u) n

{L LIILj(s)je(u)112}dvu(s).
o j=O i=O
(27.9)
Since
L LjdA{ forn >m
m<max(i,j)<::n
the same argument shows that for all n > m > N (u)
II{Xn(t) - Xm(t)}je(u)112 :::;
N(u) t
2e vu (t) L 1{n
L IIL;(s)je(u)112}dvu(s).
j=O 0 i=m+!
This together with (27.8) and (27.9) implies the required result.
•
Corollary 27.2: Let {L}li,j ~ O},{MJli,j ~ O} be processes in 1L(~,ho,.M,)
satisfying the condition (27.8) and let X and Y be (~, ho,.M,)-adapted processes
determined by
X(t) = X(O) + it L
o i,j?O
LjdAi, Y(t) = Y(O) + it L
0 i,j?O
MJdA{
Then
(fe(u),X(t)ge(v)) - (fe(u),X(O)ge(v)) =
it{
o
L
OS;SN(u)
Ui(S)V j (s)(fe(u), Lj(s)ge(v))}ds, (27.10)
DSJSN(v)
(X(t)fe(u), Y(t)ge(v)) - (X(O)fe(u), Y(O)ge(v)) =
lot L ui(s)vi(s){{X(s)fe(u),Mj(s)ge(v))
° O$i$N(u)
O$j$N(v)
(27.11)
+{L{ (s)fe( u), Y(s)ge(v)) + L{L~(s)fe(u), Mf(s)ge(v))}ds.

k2:!
Proof: Immediate from Proposition 27.1, (27.6) and (27.7).
Our next proposition concerns the "linear independence" of the family
•
{dA~ Ii, j ~ O} of the fundamental differentials.
Proposition 27.3: Let {LHli,j ~ O} be regular (~, ho, ~)-adapted processes

satisfying (27.8) and the following:
(i) The map u - t L~(t)fe(u) is strongly continuous for each t ~ 0, i,j ~ 0,
f E ~o;
(ii) {f~ Ei,i2:oL~ (s )dA{ (s)} fe( u) = 0 for all t ~ 0, f E ho, u E M.
Then
L~(t)fe(u) = 0 for all i,j ~ 0, t ~ 0, f E ho, u E ~. (27.12)
Proof: Fix t > 0 and choose u, v in M so that ui, vi are continuous functions
with support C [0, t) for i, j ~ 1. Then (27.10) implies
(fe(u),L~(t)ge(v)) = 0 for all f,g E ho.
Since vectors of the form fe(u), f E ho, u EM, Ui continuous with support in
[0, t) are total in iet ], the continuity of Lg(t)ge( v) in each of the variables t and v
together with the adaptedness of the process Lg implies (27.12) when i = j = O.
Now choose and fix io ~ 1. Vary u, v in M so that Ui, vi are continuous, Uio (t) #- 0,
SUPPUi C [O,t) for all i ~ 1, i #- io and suppvi C [O,t) for all j ~ 1. Then
(27.10) and condition (ii) imply
(fe(u),L~O(t)ge(v)) = O.
This, in tum, implies (27.12) for i = io, j = O. Interchanging the role of u and
v in this argument we obtain (27.12) when i = O. Now choose and fix io ~ 1,
jo ~ 1. Vary u, v in M so that Ui, vi are continuous, Uio(t) #- 0, vio(t) #- 0 but
supp Ui C [0, t), supp vi C [0, t) whenever i ~ 1, j ~ 1, i #- io, j #- jo. Then
condition (ii) and (27.10) imply
(fe(u),L~~(t)ge(v)) = O.
This, in tum, implies (27.12). •
A family {L~ Ii, j ~ O} of processes in IL(~, ho, M) is said to be stochastically
integrable (with respect to the family {A~ Ii, j ~ O}) if (27.8) holds. If
X(t) = X(O) + lot L L~(s)dA{ (s) for all t ~0

° i,i2:0
we write
dX = LL;(s)dA{(s).
i.j ?,O
Before we proceed to our next proposition we establish an elementary lemma.
Lemma 27.4: Let Lj E t!A(h o), j = 1,2, ... and for some constant C > 0 let
LIILjul1 2 ::; c2 11ul1 2for all U E ho.
j
Then LjLjLj converges strongly to a bounded operator M. If h is another Hilbert

space then
LII(Lj ® l)v112 ::; c211vl12 for all v E ho ® h.
j
Proof: Let Mn = L~ LjLj . Then {Mn} is an increasing sequence of positive

operators. The condition of the lemma and a polarisation argument show that
M = w.limn--->oo Mn exists and IIMII ::; 2. Hence s.limn--->oo(M - M n )1/2 = 0
which implies s.limn--->oo Mn = M and proves the first part. To prove the second
part observe that
j j
Proposition 27.5: Let {L) Ii, j 2: O} c t!A( ho) and suppose that there exist positive
constants Cj, j = 0, 1, 2, ... , such that
LIIL)1112 ::; 0111112 for all 1 E ho· (27.13)

i>O
Then there exists a unique regular (~,ho,.M}adapted process X = {X(t)lt > O}

satisfying
dX = (L L)dA{)X, X(O) = Xo ® 1 (27.14)
i,j?,O
where Xo E t!A(ho).
Proof: First we set up the iterative scheme of adapted processes:
Xo(t)=Xo®1
Xn(t) = Xo ® 1 + it o
LL)Xn-1(s)dA{(s),
i,j?,O
n 2: 1. (27.15)
We shall now prove that each Xn is well-defined, {L}Xnli,j ~ O} is

stochastically integrable and for any fixed T > 0
II(Xn(t) - X n_ l (t))fe(u)11 2 ~ fju(T)V~(t)}n IIXoI121IfI121Ie(u)112,

n. (27.16)
o~ t ~ T
where
N(u)
fju(T) = 2e V ,,(T) L S' f E ho, u E M. (27.17)
j=O
Observe that Xl is well-defined thanks to Proposition 27.1, Lemma 27.4 and
(27.13). For n = 1, (27.16) is immediate from (ii) in Proposition 27.1. By Lemma
27.4 and (27.13)
L IIL~Xl(t)fe(u)1I2 ~ SIIXl(t)fe(u)11 2
i
~ 2c]II{Xl(t) - Xo(t)}fe(u)112 + IIXoI121IfI121Ie(u)112

~ a(j,u,T)IIXoI1 2 1IfI1 2 1Ie(u)11 2
for all 0 ~ t ~ T
where a is a constant depending only on j,u and T. Thus {L}Xlli,j ~ O} is
stochastically integrable. Now suppose that (27.16) is verified for 1 ~ n ~ k. By
Lemma 27.4, Xk+l is well-defined and by (ii) in Proposition 27.1 we get
II{Xk+l(t) - Xk(t)}fe(u)11 2 ~
2e V ,,(T) L
N(u)
j=O
1°t
LIIL}(Xk(S) - X k_ l (s))fe(u)11 2dvu(s))
i2'O
which by (27.13), Lemma 27.4 and the induction hypothesis implies
II{Xk+l(t) - Xk(t)}fe(u)11 2 ~
N(u) t
2e v,,(T) L S 111(Xk(S) - X k_l (s))fe(u)11 2dvu(s)
j=O 0
~ fj~(:);:l vu(t)k+lIIXoI121IfI121Ie(u)112,O ~ t ~ T.
This proves that the iterative scheme (27.15) is well-defined and (27.16) holds for
all n. In particular, X(t)fe(u) = s.limn->oo Xn(t)fe(u) exists for all f E ho,
u EM and defines a regular (~, ho,M)-adapted process. That X satisfies (27.14)
is now immediate.
Let X' be another solution of (27.14) with X'(O) = Xo ® 1. By (ii) in
Proposition 27.1 and Lemma 27.4 we have for all 0 ~ t ~ T, f E ho, u EM
II(X(t) - X'(t))fe(u)11 2 ~ a(f,u, T) lt II(X(s) - X'(s))fe(u)1I 2dvu(s)

where a is a constant depending only on j, u and T. By Proposition 25.5 we

conclude that the left hand side of this inequality is O. •
Proposition 27.6: Let {LJli,j ~ O} c ?J3(ho) be such that for some positive
integer n,LJ = 0 if max(i,j) ~ n. Suppose that there exists a unitary operator-
valued (~, ho,.M )-adapted process X satisfying
dX = (2: LJdA{)X, X(O) = 1. (27.18)

i,j '20
Then there exist bounded operators H,Sj, i,j ~ 1, Li,i > satisfying the
following:
(i) L·Z =0 for i > n Si =
-'J
> n'
o~J for max(i 'J')- '
(ii) Li,j'21Sj C9lei)(ejl is a unitary operator in ho C9'V;
(iii) H is selfadjoint and L8 = - (iH + ~ Li'21 L; L i );
(iv) Lb = Li, L~ = -Lj'21LiSl for all i ~ 1;
(v) LJ = Sj - oj for all i,j ~ 1.
Proof: Since the number of terms under the summation on the right hand side of
(27.18) is finite it is clear that X* = {X(t)*lt ~ O} satisfies
dX* = X*( 2: (Lj)*dAJ)' X*(O) = 1

i,j '20
and quantum Ito's formula can be applied to the differential of the product
X*(t)X(t) : : : : 1. Thus
0= dX* X = 2: X* {(LJ)* + L{ + 2:(LJ)* LnXdAJ.

i,j '20
By Proposition 27.3 and the unitarity of X we conclude that
(LJ)* + L{ + 2:(LJ)* L~ = 0 for all i,j ~ O. (27.19)

k'21
Similarly using X(t)X*(t) : : : : 1 we get
0= dXX* = 2: {(LJ)* + L{ + 2:L{(L~)*}dAj.

i,j '20
By Proposition 27.3 we conclude that
(Lj)* + L{ + 2:L{(L~)* = 0 for all i,j ~ O. (27.20)

k>l
Now define Li = Lb for i ~ 1, Sj = LJ + oJ for i,j ~ 1 and iH = -(L8 +
~Li>lL;Ld. Then properties (i)-(v) of the proposition follow from (27.19) and
(27.20) by straightforward examination. •
Lemma 27.7: Let {A k }, {B k}, k = 1, 2, ... be bounded operators in ho such that

the sums I:k AkAk and I:k B'kBk converge strongly. Then I:k AkBk converges
strongly.
Proof: By hypothesis there exists a constant c > 0 such that

L)IIAkUI12 + IIBkUI12) :s: cllul1 2for all U E ho.
k;::l
Thus for 1 :s: m < n < 00 we have for for all u, v in ho

n n
I(u, L AkBkv)1 2 :s: (L IIAkullllBkvI1)2
k=m k=m
n
:s: (LII AkUI1 2) L IIBkVI12
k21 k=m
n
:s: c211ul1 2 L IIBkVI12.
k=m
Taking supremum over all unit vectors u we have
n n
II L AkBkVI12 :s: c2L IIBkVI12
k=m k=m
and the right hand side tends to 0 as m, n -+ 00.
•
Theorem 27.8: [60], [94] Let H, {Li Ii 2: I}, {By Ii, j 2: I} be bounded operators
in ho satisfying the following conditions:
(i) H is selfadjoint;
(ii) There exists a constant c > 0 such that I:i>11ILiUI12 :s: cllul1 2 for all
u E ho; -
(iii) I:i ,j21 By ® lei)(ejl is a unitary operator in ho ®"V. Define
B~
J
- o~
J
ifi,j 2: 1;
Li if i 2: 1, j = 0;
L} = 2: 1, i = 0;
- I:k21 L kBff if j
-(iH + iI:k21LkLk) if i = j = O.
Then there exists a unique unitary operator-valued (~, ho, .M}adapted process
satisfying
dU = (~L}dA{)U, U(O) = 1. (27.21)
Proof: Condition (ii) and Lemma 27.4 imply that LiLt: Li is strongly convergent.
Condition (iii) is equivalent to
(27.22)
By Lemma 27.7 it follows that Lk

L'kSJ is strongly convergent. Thus the operators
L), i, j 2 0 are well-defined and bounded. Furthermore (i), (ii) and (iii) imply that
(27.23)
for all i,j. Equations (27.23) imply that Li>o(L))* L) is a bounded operator when
j :f. O. When j = 0 its boundedness has already been shown. Thus (27.13) holds
and by Proposition 27.5 there exists a unique solution U for (27.21). By (27.11)
of Corollary 27.2 and (27.22) we have for any j, g E ho, u, v E .M
(U(t)je(u),U(t)ge(v)) - (je(u),ge(v)) =
iorot 2: ui(s)vj(s)(U(s)je(u),{L) + (Li)* + 2:(Ln*LnU(s)ge(v))ds

O~i~N(u) k20
O~j~N(v)
=0.
Thus U is an isometric operator valued process. Now U* obeys the equation
dU* = 2: u*(Li)*dAi, U*(O) = 1.

i.j2o
Once again by (27.11) of Corollary 27.2 we obtain

(U*(t)je(u), U*(t)ge(v)) - (je(u),ge(v)) =
lot 2: Ui(S )v j (s){ (U* (s )je( u), U*(s )(Li)* ge( v))
O~i~N(u)
O~j~N(v)
(27.24)
+(U*(s)(L))* je(u), U*(s)ge(v))
+ 2:)U*(s)(Li)* je(u), U*(s)(L{)*ge(v)) }ds.

k21
Define the bounded operators Mu,v(t) in ho by
(j, Mu,v(t)g) = (U*(t)je(u), U*(t)ge(v)).

Then (27.24) implies that
dMu,v
dt O~i~N(u)
O~j~N(v)
(27.25)
and Mu,v(O) = (e(u), e(v))I for each fixed u,v in M. Due to (27.23),
Mu,v(t) == (e(u), e(v))I is a solution of the ordinary differential equation (27.25).
Hence by the uniqueness theorem of ordinary differential equations, it is the only
solution and therefore
(U*(t)Je(u), U*(t)ge(v)) = (Je(u),ge(v)).

Thus U*(t) is also an isometry. In other words U(t) is unitary.
•
Corollary 27.9: Let {U(t)lt 2:: O} be the unitary operator-valued process satis-
fying (27.21) in Theorem 27.8. Let
jt(X) = U(t)* X 0 lU(t), t 2:: 0, X E ~(ho).

Then {it (X) It 2:: O} satisfies the stochastic differential equation
dit(X) = L,jt(O}(X))dAi(t) for all X E ~(ho) (27.26)

i,j?O
where
{~k?l (S~)* XSJ} - 6}X if i 2:: l,j 2:: 1,
~k?I(S~)*[X,Lkl ifi2::1,j=0,
O}(X) = ~k?I[LZ,XlSJ if i = 0, j 2:: 1,
i[H, Xl - ! ~k?l (LZLkX +
XLZLk - 2LZ XLk) if i = 0, j = O.
Proof: In the notation of Theorem 27.8 a routine computation using Corollary

27.2 and Lemma 27.4 yields (27.26) where
O}(X) = XL} + (Li)* X + L,(L:)* XLJ for all i,j 2:: O.

k?l
Substituting for L} from Theorem 27.8 we obtain the required result. •
Corollary 27.10: Let the conditional expectation maps lEt] and lEt be as in Section
26 and let {jtlt 2:: O} be as in Corollary 27.9. Then
IEs]jt(X) = js(Tt-s(X))
where Tt = etlJg • Furthermore, for all 0 ~ tl < ... < tn < 00
Proof: This is identical with the proof of Proposition 26.7.

•
Example 27.11: This is a generalisation of Example 26.9.

Let h = hi 129' .• 129 hb {lij Ij = 0, 1,2, ... } be an orthonormal basis in hi for each
i = 1,2, ... , k and /ljl 129 •.• 129 ikjk be interpreted as the pure state signifying
that there are ji particles of the i-th kind for each i = 1,2, ... , k in a system. Let
Ni = 1129 1 129'" 129 1 129 2)lfij)(fij I 129 1 129'" 129 1,

j
00
Li = 11291129'" 1291129 Llfij-I)(fijl 1291 129'" 1291

j=1
be the i-th particle number and free annihilation operators for each i. Suppose that
Ai, J.ti, v are bounded functions on the set {O, 1,2, .. . }k where v is real and Ai, J.ti
may be complex valued. Define
J.tj(NI, ... ,Nj + 1, ... ,Nk)Lj if j = 1,2, ... ,k,
g- {
J- Li,>..i(NI, ... ,Nk ) ifj=k+i,i=I,2, ... ,k.
Consider a Hilbert space "If of dimension 2k with orthonormal basis el, e2, ... , e2k.
Let A~ = A j , A~ = Aj and consider the quantum stochastic differential equation:
2k 1 2k
dU = L(EjdAj - EjdAj ) - {iv(N) + 2 LEjEj}dt]U, U(O) = 1. (27.27)
j=1 I j=1
Using Theorem 27.8 (after putting Li = Ei,Sj = oj, 1 ::; i,j ::; 2k,H =
v( N) where N = (NI , ... , N k )) we conclude the existence of a unique unitary
operator-valued (~, ho, :~e)-adapted process U satisfying (27.27). Define jt(X) =
U(t)* XU(t), X E ~(ho) where X and X 129 1 are denoted by the same symbol
X. If the maps {oj 10 ::; i, j ::; 2k} are defined as in Corollary 27.9 then
1 k
og(X) = i[v(N),X]- 2 L{(LiLilJ.ti(NW + IAi(N)12)X
i=1
+X(LiLilJ.ti(N)12 + IAi(N)12)
-2 J.ti(N)LiXLiJ.ti(N) - 2 Ai(N)LiXLiAi(N)}.
If ¢ is any bounded function on {O, 1,2, .. Y then og( ¢(N)) = (M ¢ )(N) where
(M ¢ )(jl, ... ,jk) = ~{IAi('DI2( ¢(jJ, ... , ji + 1, ... , jk) - ¢(jJ)

+ lJ.ti(tW(¢(jI, ... ,ji - 1, ... ,jk) - ¢(t))}
t= (jl, ... ,jk) and ¢(jl, ... ,ji - 1, ... ,jk) = ¢(jl, ... ,jk) if ji = 0. In view
of Corollary 27.10 we may interpret the flow {jtlt ::2: O} as a quantization of a
Markov chain describing the birth and death of k different kinds of particles with
birth and death rate amplitudes Ai(j) and J.ti(j) respectively for the i-th kind when
the system has jT particles of the r-th kind fur each r = 1,2, ... , k.
Example 27.12: [114] Let G be a measurable group acting on a separable (J-

finite measure space (st', '?Jf, J.l) so that J.l is quasi-invariant under the G action, i.e.
J.l(E) = 0 if and only if J.l(gE) = 0 for every E in '?Jf and 9 in G. For every
9 E G define the unitary operators Sg in L 2(J.l) by
(Sgf)(x) = {ddJ.l (g-IX)}I/2 f(g-I X), f E L 2(J.l)

J.lg
where J.lg(E) = J.l(gE). Then the map 9 ---+ Sg is a unitary representation of G
in L 2(J.l). Let .>..(g, x), (g, x) E G x st' be a complex valued bounded measurable
function. For any 1 E Loo(J.l) denote by the same letter 1 the bounded operator
of multiplication by 1 in L 2(J.l) with norm 1111100' Then Loo(J.l) is an abelian von
Neumann sub algebra of \5A(L2(J.l)). Define the bounded operators Lg,g E G in
L 2(J.l) by
For any finite or countably infinite set Fe G, F = {gl,g2, .. .}, set "If = L2(F)
with respect to the counting measure in F and choose the canonical orthonormal
basis in it consisting of the indicators of singletons {gi}, i = 1,2,... Write
W
Si = Sgi' Li = Lgi · Assume that L-iLi Li = L-i I'>"(gi, . converges strongly in
L 2(J.l). In 'ie = L2(J.l) ® fs("If ® L2(1R+)) write
A8(t) = t,A~(t) = Ai(t), A~(t) = A!(t), i ~ 1.
Equation (27.21) assumes the form
dU = ({~)LidA! + (Si - 1)dAi - Li SidAi)} - ~ ~LiLidt)U, U(O) = 1.

i i
(27.28)
By Theorem 27.8, equation (27.28) has a unique unitary solution. Define jt(X) =
U(t)* XU(t), X E \5A(L2(J.l)) where X ® 1 is also abbreviated to X. Then Corol-
lary 27.9 becomes
djt(X) = ~ {jt(Si- 1 [X, Li])dA! (t) + jt(Si- 1X Si - X)dAi(t)

i (27.29)
where
og(X) = -~~(LiLiX + XLiLi - 2LiXLi).

i
If X is the operator of multiplication by 1 where 1 is any complex valued bounded

28 Evans-Hudson Flows 233
measurable function on 2e
Sill¢>, Lil = >'(gi, ·)(¢>ogi - ¢»
Si~ I ¢>Si - ¢> = ¢>Ogi - ¢>
[Li,¢>lSi = >'(gi,·)(¢>ogi - ¢»
88(¢» = ~J>.(gi,·)12{¢>ogi - ¢>}

i
It is interesting to note that 88 restricted to L 00 (J.L) is the infinitesimal generator of a

classical continuous time Markov chain in which, from any state x E 2e, transition
takes place to anyone of the states {gxlg E F} with respective intensities of
probability {1>'(g,x)lZlg E F}. (See Example 28.10, 28.11.)
Notes
Theorem 27.8 and Corollary 27.9, 27.10 occur in Hudson and Parthasarathy [60].
The proof here is based on the elegant notation of Evans [34] and adapted from
Mohari and Sinha [94]. Example 27.12 is from Parthasarathy and Sinha [114] and
inspired by Meyer [91]. For an axiomatic characterisation of unitary processes
satisfying (27.21) and based on *-algebras and coalgebras see Schiirmann [119].
28 Evans-Hudson Flows
Following the description of discrete time quantum stochastic flows in Section 18
and the examples of continuous time flows induced by unitary operator-valued
adapted processes in Proposition 26.7 and Corollary 27.9, 27.10 we introduce the
notion of an Evans-Hudson flow (or EH flow). (See [34, 35, 65].)
Let ~o C ~(ho) be a *-unital subalgebra and let {AJli,j 2 O} be the
family of fundamental processes in ho ® rs('v ® L2(1R+)) as described in the
beginning of Section 27. Denote by ~t the von Neumann algebra generated by
{X ® BIX E ~o, B E ~(rs('v ® L2[0, tD)} and by ~tl = {X ® l[tlX E ~t}, l[t
being the identity operator in r s('V ® L2([t, (0)). A family {jtlt 2 O} of *-unital
homomorphisms from ~o into ~(ho ® rseV ® L2(1R+))) is called an Evans-
Hudson flow (or an EH flow) with initial algebra ~o if the following conditions
are fulfilled:
(i) jo(X) = X Q9 1[0 for all X E ~o;
(ii) jt(X) E ~tl for all X E ~o, t 20;
(iii) There exist maps 8J : ~o ---+ ~o, i,j 20 such that {jt(X)lt 2 O} is a regu-
lar (" ho,.Aft )-adapted process satisfying the quantum stochastic differential
equations:
djt(X) = ~jt(8J(X))dA{(t) for every X E ~o. (28.1)

i,j?O
The family {OJli,j ;::: O} is called the family of structure maps of the EH
flow {jt It ;::: OJ. For every t ;::: 0 and X E !J3 t we shall adopt the convention
of denoting by the same symbol X the operator X @ 1[t E !J3tj'
Proposition 28.1: Let {hit;::: O} be an EH flow with initial algebra !J3 0 and
structure maps {ojli,j ;::: O}. Suppose that for some positive integer n,Oj = 0
whenever max(i,j) ;::: nand (28.1) holds with {jt(X)lt ;::: O} as a (~, ho, 'fJe)-
adapted regular process where 'fJe = r s ('V @ L2 (lR+)). Then the following holds:
for all i, j ;::: 0 and X, Y in !J3 0 (i) OJ is linear on !J3 0 ; (ii) oj (1) = 0; (iii)
oj(X)* = O{(X*); (iv) oj(XY) = oj(X)Y + xoj(Y) + Ek>lOt(X)eJ(Y).
Proof: Since jt is a *-unital homomorphism we have
0= djt(1) = Ljt(Oj(l))dA{(t),
i,f?O
0= d[jt(X*) - jt(X)*] = Ljt(oj(X*) - O{(X)*)dA{(t),

i,j ?o
0= d[jt(oX + Y) - ojt(X) - jt(Y)]
= Ljt(oj(oX + Y) - oOj(X) - OJ(Y))dA{(t),

i,j?O
for all X, Y E !J30 and scalars o. By Proposition 27.3 the first three of the equations
above imply (i), (ii) and (iii) of the proposition. The last equation and quantum
Ito's formula imply
Once again by Proposition 27.3 we conclude (iv). •

Motivated by Proposition 28.1 we introduce the following definition: A
family {oj Ii, j ;::: O} of maps from !J3 0 into itself is said to obey the struc-
ture equations if the following holds: for all i,j ;::: 0 and X, Y E !J3 0 (1)
oj is linear on !J3 0 ; (2) OJ(l) = 0; (3) OJ(X*) = O{(X)*; (4) OJ(XY) =
oj(X)Y + XOj(Y) + Ek>lOt(X)eJ(Y) where the last term on the right hand
side is a strongly convergent sum.
In the following we shall assume that {ojli,j ;::: O} obey the structure equa-
tions and the following Mohari-Sinha regularity condition: There exist constants
OJ > 0, countable index sets Ij' j ;::: 0 and a family {D} Ii E I j , j ;::: O} c !J3( h o)
such that for all 1 E ho, X E !J3 0
LIIDjilf :::; 0]111112, (28.2)

iElj
(28.3)
It is to be noted that (28.2), (28.3) and Lemma 27.4 imply that
2)~(X)OJ(Y) = 2)~(X*)*OJ(Y)
is, indeed, a strongly convergent sum. The central aim of this section is to establish
the existence of an EH flow with structure maps {OJ Ii, j 2: O} whenever the
structure equations and the Mohari-Sinha regularity condition are fulfilled.
If there exists a positive integer n such that OJ = 0 whenever max(i,j) > n
and {Ojli,j 2: O} obey the structure equations then the Mohari-Sinha regularity
conditions are automatically fulfilled.
Let {oj Ii, j 2: O} be a fixed family of maps from ~o into itself obeying the
structure equations as well as (28.2), (28.3). Fix T > 0 and for any f E ho, u E .M
and X E ~o set
Ko(X,f,u) = IIXfI12,
Kn(X,f,u) = [2e Vu (T)r L (28.4)
ikEljk
O~jk~N(u),I~k~n
N(u)
K(T, u) = 2e Vu (T) L 0:]' (28.5)
j=O
By (28.2) and (28.3) we have
2e Vu (T) L Kn(Oj(X),f,u)::; Kn+1(X,f,u), (28.6)

i;?::O,
O~j~N(u)
(28.7)
We write
and note that (28.7) implies
S(X, f, u) = lim Sn(X, f, u) ::; ['" K(T, U)k '" vu(T)k l11X11211f112.
n-+oo L..J Jkf L..J Jkf
(28.9)
Write
Ro(X, f, u) = S(X, f, u)
Rn(X,f,u) = Rn-1(Oj(X),f,u) (28.10)
i,;>O,O'5,j'5, N(u)
Proposition 28.2: For every X E ~o there exists a sequence {iin)(X)lt ~ O}

of (~, ho,.M )-adapted processes satisfying
iiO\x) = X,
iin)(X) =X + it° Li~n-l)(oj(X))dA1(s)

i,j';?O
(28.11)
lIui n) (X) - ii n- 1)(X))je(u)11 2 :::; Kn(X,j,~)vu(t)n Ile(u)112, (28.12)

n.
Ilii n)(X)je(u)1I 2 :::; Sn(X,j,u)lle(u)11 2 (28.13)
for all 0 :::; t :::; T, j E ho, u E .M, where Kn and Sn are as in (28.4) and (28.8).
Proof: The proof is along the lines of Proposition 27.5. By Proposition 27.1
ii1)(X) = X + it° LOj(X)dAi

i,j';?O
is well-defined as a (~, ho,.M )-adapted process and thanks to (28.2), (28.3) we

have
IIU?)(X) - iiO) (X))je(u)112 = lilt LOj(X)dAije(u)112
:::; 2e vu (t) L i t Llloj(X)jI121Ie(u)11 2dvu(s)
O::;j::;N(u) ° i
::; 2e vu (t) L IIXDjjI12vu (t)lIe( u) W

iElj
Furthermore by triangle and Schwarz's inequality we get
Ilii 1)(X)je(u)11 2 :::; (1IXje(u)11 + Ilii1)(X) - iiO)(X))je(u)11)2

:::; (1IXje(u)1I + [Kl(X,j,U)vu(t)]1/21Ie(u)II)2
::; {Ko(X, j, u) + K1(X, j, u)}{l + vu(T)}lIe(u)112
= Sl(X,j,u)lle(u)11 2.
Thus the proposition has been proved for n = 1. Suppose that (28.11)-{28.13)
have been proved for 1 ::; n ::; k. Then by (28.6)-(28.8) we have
L lIii k)(oj(X))je(u)11 2 ::; LSk(Oj(X),j,u)lle(u)W

i i
[f [f
where
a(T, u) = K(T, ~ )n] Vu (T;n]. (28.14)
n=O JnI n=O JnI
Thus by (28.2)-(28.8) {ji k)(oj (X))) is stochastically integrable and jik+ 1) is
well-defined and by Proposition 27.1 and (28.6)
II(jik+l)(X) - jik)(X))fe(u)112 =
lilt ~{j~k)(Oj(X)) - j~k-l)(oj(X))}dA{(s)fe(u)112

~,J
::; 2e v ,,(t)
N(u)
~ 1~11[j~k)(Oj(X))
t
- j~k-l)(oj(X))lfe(u)112dvu(s)
::; 2e v ,,(T) ~Kk(OJ(X), f, u{~~~~11Ie(u)112

~,J
V (t)k+l
::; Kk+l(X,f,u) ~ + 1! Ile(u)112.
This proves (28.12). To complete the proof of (28.13) observe that
Ilji k+1)(X)fe(u)11 2 ::; (1Ij?)(X)fe(u)11

+11J?+l)(X)fe(u) - jik)(X)fe(u)ll?
::; (11J?)(X)fe(u)11
+ [Kk+l(X/~ ~)!vu(t)k+l] 1/2 1Ie (u)II)2,

use (28.13) for n = k, Schwarz's inequality and proceed exactly as in the case
k=Q •
Proposition 28.3: Let {jin)(X)lt 2: O,X E ~o} be as in Proposition 28.2. Then
there exists a family {jt(X)lt 2: a}, X E ~o of (~, ho,.M)-adapted processes
satisfying the following: for each f E ho, u E .M, X E ~o and t ::; T °: ;
(i) jt(X)fe(u) = limn-toojin)(X)fe(u) and the map X --+ jt(X)fe(u) is
linear;
(ii) Iljt(X)fe(u)112::; S(X, f, u)lle(u)W ::; a(T, u)IIXI121IfI121Ie(u)112;
(iii) II(jt(X) - jin)(X))fe(u)11 ::; L%:n+l (Kk(X,f~)V"(T)k) 1/2 1Ie (u)ll;
(iv) jt(X) = X + J~ Li,f20jS(Oj(X))dA{(s);
(v) the map (t,X) --+ jt(X)fe(u) is strongly continuous with respect to the
strong operator topology of ~o <,;;; ~(ho);
(vi) For any 9 E ho, u E .At

(ge(v),jt(X*)je(u)) = Ut(X)ge(v), je(u)).
L
00
II(jin)(x) - jin-1)(X))je(u)11 ::;

n=l
f {K(T, U)v~(t)}n/21IXlllljlllle(u)11
n=l v'nT
< 00,
which implies that the right hand side of (i) exists and determines jt(X) as a
(~, ho,.At )-adapted regular process. Now (28.13), (28.9) and (28.14) imply (ii).
From (28.12) we have
n+m
II(jin+m\X) - jin)(X))je(u)ll::; L II(jik)(X) - jik-1)(X))je(u)11
k=n+l
: ; nf (Kk(X'~'!U)VU(t))1/21Ie(u)ll.
k=n+]
Letting m -+ 00 and using (i) we get (iii). To prove (iv) we observe that
II(jt(X) - X - i t ~js(Oj(X))dA{)je(u)112 ::;

z.J
211(jt(X) - ji n)(X))je(u)11 2 (28.15)
+211 it ~ {js(oj(X)) - j~n-l) (oj (X))}dA{ (s )je( u)112.

Z,J
By Proposition 27.1 and (iii) we conclude that the second term on the right hand
side of this inequality does not exceed
f;
N(u) t
4e vu (t) i ~11{js(Oj(X)) - j~n-l)(oj(X))}je(u)112dvu(s) ::;
4e vu (t)vu (t) L (L[ K 00

k
(Oi (X) j u)v (T)k
j 'k!' u ]1/2)21Ie(u)112.
i2: 0 k=n
°SJSN(u)
By Schwarz's inequality in the summation over k, (28.6) and (28.7) the right hand
side of this inequality does not exceed
These remarks, (28.15) and (i) imply (iv). To prove (v) we first observe that by
Proposition 27.1, (28.6) to (28.9) and (iv)
IIUt2(X) - jtl (X))je(u)112 = IIlt2 L.js(OJ(X))dAije(u)112

t.)
~ 2e vu (T) 1t2 N(u)

L LS(OJ(X),j,u)lle(u)11 2dvu (s)
tl j=O i2: 0
(28.16)
On the other hand, we have

Iljt(X - Y)je(u)112 ~ S(X - Y, j, u)lle(u)112
(T)n
~ Ile(u)112 L 00
Vu , L{ (2ev
00
u
(T))n
,
n=O v'nT n=O v'nT (28.17)
x L II(X - Y)D~: ... D~:jI12}.

ikE1jk,O:O:;jk::;N(u)
l:SkSn
Inequalities (28.16), (28.17) and dominated convergence imply (v). To prove (vi)
we use induction. Note that jiO)(x*) = jiO)(X)* = X*. Assume that
(ge(v),jin~l)(X*)je(u)) = Uin~l)(X)ge(v),je(u)).
By (27.10) and structure equations we have
(ge(v),jjn)(X*)je(u)) = (Xge(v), je(u))
+(ge(v), lt L.j~n~l)(OJ(X*))dAi(s)je(u))
t,)
= (Xge(v),je(u)) + it L Vi(S)uj(s)(ge(v),j~n~l)(OJ(X*))je(u))ds
° iSN(v)
jSN(u)
= (Xge(v),je(u)) + it L Vi(S)uj(s)(j~n~l)(ei(X))ge(v),je(u))ds
° iSN(v)
jSN(u)
= uin)(X)ge(v), je(u)).
•
Proposition 28.4: Let {jtlt ~ O} be as in Proposition 28.3. Then for any f, 9 E

ho, X, Y E (lAo, u, v E .M
(fe(u),jt(XY)ge(v)) = Ut(X*)fe(u),jt(Y)ge(v)).
Proof: From (28.9), (28.10), (28.14), (28.2)-(28.4) we have
Rn(X,f,u) ~ S(OJ: ... OJ: (X), f, u)

ik~O
ikSN(u),lSkSn
~ LVu(Tr L (2eVu(~))k ( L a])n+kIIXI121IfI12 (28.18)

k20 JkT JkT j~N(u))
= a(T, u)( L a]tIIXI121IfI12.
j~N(u)
For any n ~ 0 set
Bin)(x, Y) = Uin)(X*)fe(u),jin)(Y)ge(v)) - (fe(u),jin)(XY)ge(v))

(28.19)
where jjn) are as in Proposition 28.2. We claim that
IB(n)
t ' -L...J
n
(X Y)I < '""' v
( ) n-k±1
T 2
k,f -k
Vu,v (T)k
I'
{'""'[G
L...J
i .1"'i,k (X f u· Y 9 V)1/2
k=! .yn +. JJ"'Jk " , , ,
+G i .1'''i,k (y* 9 V' X* f U)1/2]}

Jl"'jk '" "
(28.20)
where v(T) = max(vu(T), vv(T)),
vu,v(T) = It[(1 + Ilu(s)112)(1 + Ilv(s)112)P/2ds,
il"'i k (X f
Gjl "-jk " u ., Y ,g, v )
(28.21)
and the second summation in (28.20) is over ir's and jT'S subject to the constraint
i, = ii, 1 ~ ir + jT ~ 2, 2 ~ r ~ k.
We prove the claim by induction. When n = 0, BiO\x, Y) =0 and the
claim holds trivially.
By (27.10), (27.11) and Proposition 28.2, 28.3 we have
B~n)(X, Y) = ito
L
O:5 i :5 N (u)
Ui(s)vj(s){B~n-l)(X,O}(Y)) + B~n-l)(O}(X), Y)
O:5j:5 N (v)
+ LB~n-l)(ol(x),oJ(Y))}ds + p~n)(X, Y)
k~l
(28.22)
where
p~n)(X,y) = it o
L
O:5 i :5 N (u)
Ui(S)vj(s){((j~n)(x*)
O:5j:5 N (v)
_j~n-l) (X*) )je( u), j~n-l) (O}(Y) )ge( v))
+(j~n-l)(O}(X*))je(u), (j~n)(y) _ j~n-l\Y))ge(v)) }ds

(28.23)
By (28.12), (28.13) and Schwarz's inequality we have
Ip~n)(X, Y)I ~ {[Kn(X*, j, u)R1(Y,g, V)]1/2 + [Kn(Y,g, V)Rl(X*, j, U)]1/2}
x v(T)n/2vu ,v(T) lIe(u)lllle(v)ll. (28.24)

Vnf
When n = 1, B?)(X, Y) = pP)(X, Y) and (28.24) implies (28.20). Equations
(28.4) and (28.10) and an elementary calculation shows that
it{L
o O:5i:5N(u)
IUi(S)vi (s)IG~~''''',~: (X, j, u; O}(Y),g, V)I/2}ds
O:5j:5 N (v)
<
-
Gi.l, ... ,i.k,O (X " j U', Y " 9 v)v'U,v (t)·,
31, ... ,3k,t
(28.25)
it{L
o O:5i:5N(u)
IUi(S)vi (s)IG~~ :::~: (O}(X), j, u; Y,g, V)I/2}ds
O:5j:5N(v)
<
-
G31,0"
i.1, ... ,i.k,l (X j
,3k,O "
U'
,
Y "9 v)v'U,v (t)·, (28.26)
it{L
o O:5i:5N(u)
IUi(S)vj(s)ILG~~:::~: (O~(X), j, u; OJ(X),g, V)I/2}ds
r>l
o:5j:5 N (v) -
(28.27)
Now assume that (28.20) holds for n - 1. Using triangle inequality in (28.22),
(28.23) and the estimates (28.24)-(28.27) we obtain (28.20) for n. This proves the
claim (28.20) for all n. By (28.5), (28.7), (28.18) and (28.21) we have
i.1"",i,k (X f
IC Jl, ... ,)k "
U'
,
Y "g v)1 <
-
N(v)
ca(T, v)( L a;)L:jr [2e Vu (T)r L:ir K(T, ut- k+ L:ir
j=O
exceed
N(v) N(u)
ca(T, v)( La;)L:jr( L a;)L:ir K(T, ut- k S
j=O j=O
N(u) N(v)
ca(T, v){max( L a;, La;, 1)}2k K(T, u)n-k.
j=O j=O
Thus by (28.20)
where C, a, b are constants,

N(u)
a = [v(T)K(T, U)P/2, b = 3v(T) max( La;, K(T, u), K(T, v), 1).
j=l
Since
n n-kbk (b)n
' """
l1m~ a <1ml' a+ =0,
n->ook=l k!J(n-k+1)! -n->oo v'nT
we conclude that Bin)(X, Y) ~ 0 as n ~ 00 and we obtain the required result

from Proposition 28.3. •
Proposition 28.5: Let {jtlt 2: O} be as in Proposition 28.3. It follows that

for every t 2: 0, X E ~o, jt(X) extends uniquely to a bounded operator on
'ie = ho (9 fseV (9 L2(~+)). Denote this extension by the same symbol jt(X).
Then Iljt(X)11 s
IIXII·
- k
Proof: Choose any element 'IjJ E 'fjf of the form 'IjJ = L:i= 1 fie( Ui) where fi E ho
and Ui E .Ai. By the repeated use of Proposition 28.4 we have
Iljt(X)'ljJ11 2 = ('IjJ,it(X* X)'IjJ)
S 11'ljJlllljt(X* X)'ljJ11
S 11'ljJlll+~+ .. '+ztr Iljt(X* X)2n)'ljJllztr
k
S 11'ljJ111+1+"+ztr {Llljt((X* X)2 n )fie(Ui)ll}ztr.
i=l
Using (ii) in Proposition 28.3 we now get

k
Iljt(X)'l/J11 2 ~ 11'l/J111+~+"+zh-IIX* XII(La(T, ui)llfilllle(ui)IDzh-

i=l
for all t ~ T. Letting n -+ 00 on the right hand side we have
Iljt(X)'l/J11 ~ IIXIIII'l/JII·
Since vectors of the form 'l/J are dense in 'JC the proof is complete. •
Proposition 28.6: Let {jt(X)IX E ~o, t ~ O} be as in Proposition 28.5. Then
jt(X) E ~t] for all t ~ O.
Proof: Let 'ie be as in Proposition 28.5. Suppose B E ~('ie) and B commutes

with every operator of the form X Q9 C, X E ~o, C E ~(r seV Q9 L2(1R+))). By
Proposition 22.8 there exists an operator B E ?A~ such that B = B o Q9 1 where
1 is the identity operator in rsCv Q9 L2(1R+)). We shall now show that jt(X)
commutes with B. Consider the operators ji n ) (X) defined in Proposition 28.2.
When n = 0 we have
jjO) (X)Bo = XBo = BoX = BojjO) (X) for all X E ?A o.
Suppose that for any f E ho, U E .At
jin-1)(X)Bofe(u) = Bojin-1)(X)fe(u).
Clearly
jin)(X)Bofe(u) = XBofe(u) + lot L.j~n-l)(O;(X))dA{(S)Bofe(u)

~,J
= BoXfe(u) + Bo lot L.j~n-l)(OJ(X))dA{(s)fe(u)

~,J
= Bojjn) (X)fe(u).
Using induction, letting n -+ 00 and applying (i) of Proposition 28.3 we have
jt(X)Bofe(u) = Bojt(X)fe(u). By Proposition 28.5, jt(X) commutes with B.
By the double commutant theorem jt(X) E ?A~ Q9 ?A(rseV Q9 L2(1R+))). Since
{jt(X)lt ~ O} is adapted it follows that jt(X) E ?At]. •
Theorem 28.7: [35] [94] Let {OJ Ii, j ~ O} be a family of maps from ?A o into
itself obeying the structure equations and the regularity conditions (28.2), (28.3).
Then there exists a unique EH flow {jtlt > O} with initial algebra ?A o satisfying
the quantum stochastic differential equations:
djt(X) = L,jt(O;(X))dA{(t) for all t ~ 0, X E ~o.

i,j
Proof: Proposition 28.3-28.6 imply the existence of an EH flow satisfying the

required equation in any bounded interval. The existence and uniqueness of the
EH flow in IR+ will follow if uniqueness is established in every bounded interval.
Suppose {jW ~ O} is another EH flow satisfying the same equation in [0. T].
Then
(jt(X) - j;(X))fe(u) = It 4={js((lj(X)) - j~(Oj(X))}dAi(s)fe(u)

z.J
for all f E ho• u E.At, 0 :::; t :::; T. By Proposition 27.1

II(jt(X) - j;(X))fe(u)112 :::;
2e v,,(T) 1 f; ~11{js(Oj(X)) - j~(oj(X))}fe(u)112dvu(s).

t N(u)
By induction
II(jt(X) - j;(X))fe(u)112 :::;
(2e V ,,(T)t ( L: II{jtJoj~ ... oj: (X)) (28.28)
}o<tl<···<tn<t
-j;1 (0;: ... 0;: (X))} fe(u)112 x dvu(t 1 )··· dvu(t n )
where the summation under the integral sign is over each i k ~ 0, 0 :::; jk :::;
N(u) for k = 1,2, ... ,no By the proof of Proposition 28.5 both the *-unital
homomorphisms jt and j; are contractive. Thus (28.28) implies
II(jt(X) - j;(X))fe(u)112 :::;
(28.29)
41IfI121Ie(u)112 {2e V
,,(T) vu (t)}n II '" Oi} ... oin(X)*O~1 ... oin(X)11
n! 6 JI In JI In
where the summation inside the norm on the right hand side is over i k ~ 0,
o :::;
jk :::; N(u) for each 1 :::; k :::; n. By the regularity conditions (28.2), (28.3)
we have
IIL:oj(X)*oj(X)11 :::; alllX* XII·
i2°
Repeated use of this inequality and (28.29) yield
II(jt(X) - j;(X))fe(u)W :::;
(2 v (T) (t)))n N(u)
41IfI121Ie(u)112 e u ~u (~al)nIIXI12 for every n.
J=O
Letting n --> 00 we get jt(X) = j;(X).
•
Theorem 28.8: In Theorem 28.7 suppose that ~o is abelian. Then
[js(X),jt(Y)] = 0 for all s, t ~ 0 and X, Y E ~o.
Proof: Without loss of generality we assume s < t. Since j s is a homomorphism

and ~o is abelian we have
By Theorem 28.7
jt(Y) = js(Y) + jt ~ja(Oj(Y))dAi(a).

t,)
Since {jt(X)lt ?: O} is adapted and js(X) commutes with the increments of Ai

in [s, (0) we obtain
[js(X),jt(Y)] = jt ~[js(X),ja(Oj(Y))]dAi(a). (28.30)

t,)
Fix f, g in ho, u, v in .At and write
K(s, t; X, Y) = (je(u), [js(X),jt(Y)]ge(v)). (28.31 )

Then (28.30) implies
K(s,t;X,Y) = jt L ui(a)vj(a)K(s,a;X,Oj(Y))da (28.32)

s O:Si:SN(u)
0:Sj:SN(v)
thanks to Corollary 27.2. Iterating (28.32) N times we get

K(s,t;X,Y) =
L J rr~=ludtr)vjr(tr)K(s,tl; x,oj:oj~ .. . 0j~(Y))dtl·· ·dt N

s<t 1 <···<t N <t
(28.33)
where the summation is over 0 ~ ir ~ N( u), 0 ~ jr ~ N( v) for each 1 ~ r ~ N.
Restrict u, v to be 'V -valued bounded functions, set
f3(T) = sup max{llu(s)ll, Ilv(s)ll, Ilu(s)llllv(s)ll, I} (28.34)

O::;s::;T
and note that
IK(s, t; X, Y)I ~ 21Ifllllglllle(u)lllle(v)IIIIXIIIIYII. (28.35)
Inserting (28.34) and (28.35) in (28.33) we obtain for 0 ~ s ~ t ~ T

IK(s, t; X, Y)I ~
21Ifllllglllle(u)lllle(v)IIIIXII f3(T)2N (t ~~)N IIYII

(28.36)
x L
i i ,··· ,iN
I10j: II· . '1Ioj~ II
iI,··· ,jN
where 0 ~ iT ~ N(u), 0 ~ iT ~ N(v) for each 1 ~ r ~ N. Inequalities (28.2)

and (28.3) imply that 0;
as an operator on ~o satisfies
IIO}(X)II ~ ajllXl1 for all i,i,X E ~o.

Thus (28.36) implies
{{3( T)2(t - s)(,,~(v)a')(N(u) + 1)}N

IK(s,t;X,Y)I~C. L.J3=;! 3
where C is a constant depending on f, g, u, v, X, Y. Letting N ~ 00 we conclude

that K(s, t; X, Y) = O. Now (28.31) implies the required result. •
Theorem 28.8 has the important interpretation that if we have a "classical
system" of observables, i.e. ~o is a commutative *-subalgebra of ~(ho), then the
corresponding quantum EH flow is, indeed a classical stochastic flow.
Theorem 28.9: Let {itlt ~ O} be the EH flow in Theorem 28.7. Let:Io C ~o

be the centre of ~o. Then
[is(X),it(Y)] = 0 for all s, t ~ 0, X, Y E :Io.
Proof: After noting that [js(X),is(Y)] = 0 for all X, Y E:Io the proof is along
the same lines as those of Theorem 28.8.
It is interesting to compare property (iii) of the special discrete time flow
constructed in Example 18.13 with Theorem 28.9.
Exercise 28.10: The EH flow {itlt ~ O} with initial von Neumann algebra ~o
in Theorem 28.7 satisfies the following:
(i) The map (t,X) ~ it(X) is strongly continuous when ~o is equipped with
the strong topology;
(ii) If {lEtJlt ~ O} is the family of conditional expectation maps described in
Section 26 then the Markov property holds: for any 0 ~ to < t < ... <
tn < 00, Xl, .. ' ,Xn E ~o
IEtoJitl (Xdit2 (X2) ... itn (Xn)
= ito (Ttl-to (X1Tt2-tl (X2 ... (Xn-1Ttn-tn_l (Xn)) ... )
where Tt = exptOg;
(iii) If there exists a bounded linear map Too : ~o ~ ~o such that limt->oo
IITt(X) -Too(X)11 = 0 for all X E ~o then there exists a unique contractive
linear map ioo : ~o ~ ~(ho ® rsC'V' ® L2(~+))) such that
w.lim it(X) = ioo(X) for all X E ~o

t->oo
(iv) If, in addition, joo(X* X) = joo(X*)joo(X) in (iii) then

s.lim jt(X) = joo(X);
t-tOO
(v) If joo (X*Too (X)) = joo(X*)joo(X) then

s.lim rl rt js(X)ds = joo(X);
l-too Jo
(vi) If, in (iii), Too(X) is a scalar multiple of the identity for every X we say
that the (Markov) semigroup {Tt} is ergodic. For such an ergodic semigroup
s.lim rl rt js(X)ds = Too(X).

t---+oo Jo
Example 28.11: We continue with Example 27.12 and for any X E ?A(L2(t1,))
define
Ob(X) = S;l [X,Li], O?(X) = [L;,X]Si
O}(X) = (S;IXSi - X)6J for i,j 2: 1,
og(X) = -~ l)L;LiX + XL;Li - 2L;XLi).

i
Then {oj Ii, j 2: O} obey the structure equations described at the beginning of this
section. Furthermore
l:)Ob(X)fI1 2 :S LII(XLi - Li X )f11 2
:S 2(LIIXLdI12 + IILIIIIXfI12)
i2': 1
where L = LiLi Li. If j 2: 1
LIIO}(X)fI1 2= IIO](X)fI1 2:S 211 X Sjfl1 2 + 211Xf112.
By elementary computations using Schwarz's inequality we have
Ilog(X)fll :S ~(IILIIIIXfll + IIXLfll) + IILII~ (LIIXLdI12)~.

i
Thus the regularity conditions (28.2), (28.3) are fulfilled. The flow {jtlt 2: O}
induced by the unitary operators {U(t)lt 2: O} in Example 27.12 is the same as
the EH flow with initial algebra ?A(L2(fJ,)) given by Theorem 28.7 in this case.
By the computations at the end of Example 27.12 it follows that the maps O} leave
the abelian algebra Loo(t1,) invariant. By Theorem 28.8 {jtILOO(/L),t 2: O} yields
a classical Markov flow with generator
og(¢) = LI,\(gi, ·)12(¢ 0 gi - ¢).
i
Example 28.12: Let ~ be the state space of a Markov chain in continuous time
with countably many states and let Pt(x,y),x,y E ~ be the stationary transition
probabilities such that
i(x, y) 20 if xi- y and LYE'lei(X, y) = O. We shall now realise this as an EH

flow. Put any group structure on ~ so that G = ~, J-l is the counting measure and
G acts on itself by left translation. Set
>'(x, y) = {i(y, xy)I/2 if xi- .e,

o otherwIse,
e being the identity element. Using the elements of G as indices (2 1) introduce

the structure maps on the abelian algebra LOO(J-l) C ~(L2(J-l)):
()g(</J)(y) = >'(x,y){</J(xy) - </J(y)}

()~(</J)(Y) = >,(x,y){</J(xy) - </J(y)}
();(</J)(y) = </J(xy) - </J(y)
();,(</J)(y) = 0 if x i- x'
()g(</J)(y) = LI>'(x, y)1 2(</J(xy) - </J(y)).
xE'le
We now assume that sUPxE'leli(x,x)1 = b < 00. Now observe that
Thus the operator sum LXE'leI>'(x,·W is strongly convergent. It follows from

the discussion at the end of Example 28.11 that the classical Markov flow with
transition probability Pt(x,y),x,y E ~ can be realised as an EH flow {jtlt 2 O}
with initial algebra L 00 (J-l) satisfying
+jt(()g(</J))dt
with jo (</J) = </J. (See (27.29).)

29 A digression on completely positive linear maps and Stinespring's Theorem 249
Exercise 28.13: In the notation of Theorem 28,7 let {jt!t :::: O} be an EH flow
with initial algebra (l}3o and structure maps {OJli,j :::: a}. Suppose {m;li,j :::: o}
are scalars satisfying (i) EilmW < 00 for each j; (ii) m; +m{ + Ek>lm~mJ =
m; + m{ + Ek~l m~m{ = 0 for all i,j. Then there exists a unique unitary
operator-valued process W satisfying
W(O) = 1,dW = (L m;dA{)W.
Define jf(X) = W(t)*jt(X)W(t), t :::: O. Then {j~lt :::: O} is an EH flow with

structure maps {¢; Ii, j :::: O} given by
",i
'Pj = Oij + "'(-kOk
~ m i j + mjkOi)
k + '"
~ -k
m i mjlOk
l'
k~l k,l~l
Here W is an adapted process of ampliations of Weyl operators. By varying

{ m;} the structure maps {OJ} can be reconstructed from the vacuum conditional
expectations of jf(X).
Exercise 28.14 : Let (l}3o be a *-unital subalgebra of (l}3(ho) and let X -+

((rJ(X))), 1 :::; i,j :::; n be a *-unital homomorphism from (l}3o into (l}3o Q9 (l}3(C n )
where ho is a Hilbert space and Cn is equipped with the canonical basis. Suppose
o is a bounded derivation of (l}3o satisfying O(X*) = O(X)* for all X E (l}3o and
Rj' 1 :::; j :::; n are elements of (l}3o. Define the linear maps OJ, 0 :::; i, j :::; n in (l}3o
by
oj(X) = rJ(X) - OJX if i,j :::: 1
06(X) = L(rJ(X)Rj - RjX),

j~l
O?(X) = L(Rjr/(X) - XRj) ifi:::: 1,

j~l
og(X) = O(X) - ~L(RjRjX + XRjRj) + LRirJ(X)Rj.

j?l i,j~l
Then {OJ 10 :::; i, j :::; n} obey the structure equations.
Notes
The idea of quantum stochastic flows based on structure maps goes back to Hudson
[65]. It was developed further in Evans [34]. Theorem 28.7 was proved in Evans
[35] for the case of finite degrees of freedom. The present exposition in the case of
infinite degrees of freedom is reproduced from Mohari and Sinha [94]. Theorem
28.9, Exercise 28.10, Example 28.11, 28.12 are from Mohari and Sinha [94] and
inspired by Parthasarathy and Sinha [114]. For an application of these ideas to
chaos representation for Markov chains see Meyer [91]. For a general theory
of Markov dilations and concrete examples of quantum Markov processes see
Kummerer [73], Kummerer and Maassen [72], Frigerio and Maassen [44].
29 A digression on completely positive linear maps

and Stinespring's Theorem
Sections 26-28 in general and Corollary 26.8, 27.10 and Exercise 28.9 in par-
ticular indicate how a certain class of quantum stochastic differential equations
leads us to one parameter semigroups of positive contraction operators in the
Banach space ~(ho). This immediately reminds us of Markov processes in classi-
cal probability theory which are described by one parameter semigroups of positive
contraction operators in the commutative Banach *-algebra of bounded complex
valued measurable functions in a state space. Before taking up the description
of quantum probabilistic analogues of such semigroups we shall first examine
the simpler problem of constructing a quantum analogue of transition probability.
Suppose (Oi, ~i)' i = 1,2 are measurable spaces and P(·,·) is a map from 0 x ~2
into the unit interval [0, 1] satisfying the following conditions: (i) for each fixed
WI E 0[, P( WI, .) is a probability measure in ~2; (ii) for each E E ~2, P(·, E) is
an ~I-measurable function. Such a map P is called a transition probability. Any
such transition probability induces two maps in the following manner: (i) for any
bounded measurable function 9 on (02, ~2)
(29.1)
(ii) for any probability measure j.t on ~I
(T'j.t)(E) = f P(wI,E)j.t(dwt},E E ~2. (29.2)

in 1
Then T transforms a bounded random variable on O2 to a bounded random vari-

able on 0 1• Indeed, Tl = 1, Tg ~ 0 whenever 9 ~ 0 and sUPw1ITg(wt}1 ::;
sUPw2 Ig(W2)1. On the other hand T' transforms a probability measure j.t on 0 1 to a
probability measure T' j.t on O2. In a similar vein consider two Hilbert spaces '1Je i ,
i = 1,2 and a linear operator T : ~('a'C2) --* ~('1Jet) satisfying the following con-
ditions: (i) Tl = 1; (ii) T(X) ~ 0 whenever X ~ 0; (iii) T(X*) = T(X)*; (iv)
IIT(X)II ::; IIXII; (v) w.limn->oo T(Xn) = T(X) whenever w.limn->oo Xn = X.
For any state p in '1Je 1 the quantity tr pT(X) may be interpreted as the expec-
tation of the observable X in '1Je2 after a transition from the initial state p in
'1Je 1 to a new state in 'a'C2 has been made. If P is a projection in 'a'C2 then by
condition (ii) T(P) ~ 0 and hence tr pT(P) ~ O. Suppose that {Pn } is a se-
quence of projections in 'a'C2 such that L-jPj = 1 in the weak operator topology.
Then by (v) L-j T (Pj) = T ( 1) = 1 in the weak operator topology. Suppose
that p = L-jpjIUj)(Uj I where Pj > O,L-jPj = 1 and {Uj} is an orthonormal
sequence. Then
n n
LtrpT(Pi) = LPj(Uj,T(LPi)Uj) (29.3)
i=1 i=1
and
n
T(LPi ) :::; T(l) = 1.
i=[
Thus
n
0:::; (Uj,T(LPi)Uj) :::; 1 for all j,n. (29.4)
i=[
Since T(L~=[ Pi) converges weakly to the identity operator in '1IC[ as n --+
00 it follows from (29.3), (29.4) and the dominated convergence theorem that
L~[ tr pT(Pi) = 1. In other words the correspondence P --+ tr pT(P) is a prob-
ability distribution in g>('1JCz ). Thus there exists a state T'(p) in '1JCz such that
tr pT(P) = trT'(p)P for every projection P in '1ICz. This implies
trpT(X) = trT'(p)X for all X E ~('1JCz).
In the case of the classical transition probability PC·) on O[ x ?:F z there exists a
joint probability measure v on (O[ x Oz,?:F[ x ?:F z ) such that
where p" T are as in (29.1), (29.2) and f, g are arbitrary bounded measurable
functions on 0[, Oz respectively. In the quantum analogue there may not exist a
state p on '1JC[ 0 '1ICz such that tr pX 0 Y = tr pXT(Y) (or tr pT(Y)X) for all
X E ~('1JCt), Y E ~('1JCz).
Consider the n-dimensional Hilbert space en. If h is any Hilbert space
then h 0 en can be identified with the n-fold direct sum hEEl·· . EEl h and any
operator X in h 0 en can be expressed as a matrix (( Xij )) where Xij is an
operator in h for each 1 :::; i,j :::; n. If T : ~('1JCz) --+ ~('1ICd is a linear operator
satisfying the conditions (i)-(v) mentioned above we define the linear operator
T(n) : ~('1ICz 0 en) --+ ~('1JC[ 0 en) by putting T(n)(((Xij))) = ((T(Xij))),
Xij E ~('1JCz), 1 :::; i,j :::; n. In such a case T(n)(x 0 B) = T(X) 0 B for all
X E ~('1ICz), B E ~(en). It is not necessary that T(n) be positivity-preserving.
If however, T(n) is positivity-preserving we say that T is an n-positive map from
~('1JCz) into ~('1JC[). If Tis n-positive for every n = 1,2, ... we say that T is
completely positive.
Proposition 29.1: Let T : ~('1ICz) --+ ~('1IC[) be a completely positive linear

operator. Then for any X[, ... ,Xn E ~('1JCz), Yi, ... ,Yn E ~('1ICd, n = 1,2 ...
L Yi*T(X; Xj)Yj 2:: O. (29.5)

[<::;i,j<::;n
Proof: For any U E 'Je, we have
(u,~Yi*T(X;Xj)ljU) = ((Y;U) , ((T(X;Xj))) (~U)).

~ ~u ~u
Since ((X;Xj)) ?: 0 in 'Je2 Q9 en and hence ((T(X;Xj))) ?: 0 in 'Je, Q9 en it

follows that the left hand side of the equation above is non-negative. •
Proposition 29.2: Let T : ~('Je2) -+ ~('Jed be a completely positive operator.

Define
K(X" Yj, U, ; X 2, Yz, U2) = (u" Yj*T(Xt X2)YzU2)
for all (Xi, Yi, Ui) E ~('Je2) X ~('Jed X 'Je" i = 1,2. Then K is a positive definite
kernel on ~('Je2) x ~('Jed X 'Je,.
Proof: It is enough to show that for any (Xi, Yi, Ui) E ~('Je2) x ~('Je,) X 'Je"
Ci E e, 1 ~ i ~ n
2)iCjK(Xi, Yi, Ui ; Xj, lj, Uj) ?: O. (29.6)

i,j
To this end choose arbitrary elements 0 :j; U E 'Je, and Zi E ~('Je,), 1 ~ i ~ n
such that ZiU = Ui and put Li = CiYiZi. Then the left hand side of (29.6)
can be expressed as (u, Li,jLiT(X; Xj )Lju). Now (29.6) is immediate from
Proposition 29.1. •
Proposition 29.3: Let T : ~('Je2) -+ ~('Jed be a completely positive operator.

Then there exists a Hilbert space 'Je and a map A : ~('Je2) x 'Je, -+ 'Je such that
(i) {A(X,u)IX E ~('Je2)'U E 'Jed is total in 'Je;
(ii) A is linear in each of the variables X, u;
(iii) (A(X" ud, A(X2, U2)) = (u" T(Xt X2)U2);
(iv) In particular the map Va : U -+ A(I, u) is an isometry from 'Je, into 'Je.
If ('Je', A') is another such pair satisfying (i) and (iii) then there exists a
unitary isomorphism Uo : 'Je -+ 'Je' such that UOA(X, u) = A'(X, u) for all
X E ~('Je2)' U E 'Je,.
Proof: By Proposition 29.2 and 15.4 there exists a Gelfand pair ('Je, AO) where
AO : ~('Je2) x ~('Je,) X 'Je, -+ 'Je is a map satisfying
(a) (AO(X" Yj, uJ), AO(X2' Y2, U2)) = (uJ, YtT(Xt X2)YzU2)
for all (Xi, Yi, Ui) E ~('Je2) x ~('Jed x 'Je" i = 1,2;
(b) {Ao(X, Y, u)1 (X, Y, u) E ~('Je2) X ~('JeJ) x 'Jed is total in 'Je.
Put A(X, u) = AO(X, 1, u). Then (a) implies
(AO(XJ, YJ, uJ), AO(X, Y, u) - A(X, Yu)) = O.
Now (b) implies A(X, Yu) = AO(X, Y, u) and hence A satisfies (i). Putting Yj =
Y2 = 1 in (a) we obtain (iii). Putting X j = Xl = 1 in (iii) and using T(l) = 1 we
get (iv). (ii) follows from the fact that A( aXj + {3X2' u) - aA(Xj, u) - {3A(X2' u)
and A(X, au + {3v) - aA(X, u) - {3A(X, v) are orthogonal AO(X', Y', u') for all
scalars a,{3 and X,X',X j,X2 E (JA('Je 2 ), Y' E (JA('Je j), u,v,u' E 'Je j. The last
part is immediate from Proposition 7.2.
Proposition 29.4: Let T, A, Vo be as in Proposition 29.3. Then there exists a

*-unital homomorphism 1f : (JA('Je2 ) --+ (JA('Je) such that A(X, u) = 1f(X)Vou for
all X E (JA('Je l ), u E 'Je j .
Proof: For any unitary operator U in 'Je l we have from (iii) in Proposition 29.3
(A(UXj,ud,A(UX2,U2)) = (u\,T(X:U*UX2)U2)
= (u\,T(X:X2)U2) = (A(Xj,Uj),A(X2 ,U2)).
Hence by (i) in Proposition 29.3 and Proposition 7.2 there exists a unitary operator
1f(U) in 'Je such that
1f(U)A(X,U) = A(UX,U) for all X E (JA('Je2 ), u E 'Je j.
This also shows that 1f(Uj )1f(U2) = 1f(UjU2 ) and 1f(1) = 1. For any finite
linear combination '£ ajUj of unitary operators in 'Je 2 define 1f('£jajUj)
'£jaj1f(Uj). If '£jajUj = 0 observe that
112: aj1f(Uj )A(X, u)11 2 = (u, T(X*(2: ajUj )*(2: ajUj )X)u) = O.
Thus 1f is well-defined on the *-algebra generated by the unitary operators in 'Je2
as a *-unital homomorphism. If X is any selfadjoint operator in 'Je 2 such that
IIXII 'S 1 then
X = ~(exp[icos-j Xl + exp[-icos- j Xl)
is a linear combination of two unitary operators. Thus every X in ~('Je2) is a

linear combination of four unitary operators. Hence there exists a 1f(X) in ~('Je)
such that 1f(X)A(X',U) = A(XX',U) for all X,X' in ~('Je2) and u in 'Je j
and the correspondence X --+ 1f(X) is a *-unital homomorphism. By definition
A(X,U) = 1f(X)A(1,U) = 1f(X)Vou. •
Proposition 29.5: Let 1f : (JA('Je2 ) --+ (JA('Je) be any *-unital homomorphism. There
exists a Hilbert space k and a unitary isomorphism r : 'Je --+ 'Je2 (>9 k such that
1f( X) = r- j X (>9 1 r for all X E (JA ('Je2) where 1 is the identity operator in k.
Proof: Let dim 'Je2 = 00. Without loss of generality assume that 'Je 2 = £2(1R).
Choose one parameter unitary groups Ut = e- itq , Vt = e- itp where q,p is
the canonical SchrOdinger pair satisfying [q, p] = i. Let Q, P be the selfadjoint
opeators defined by 1f(Ut) = e- itQ , 1f(Vt) = e- itP . Then 1f(Ut) and 1f(Vt)
obey the Weyl commutation relations, Hence by the Stone-von Neumann Theorem
(Exercise 13,8) on the uniqueness of CCR, there exists a Hilbert space k and a
unitary isomorphism r : 'Je ----+ 'Je 2 0 k such that Q = r-tq 0 lr, P = r-tp 0 lr
or, equivalently, 7l'(Uslit) = r- t (Us lit 01)r for all s, t. Since linear combinations
of Us lit, s, t E ~ are weakly dense in ~('Je2) (See Proposition 20.9) it follows
that 7l'(X) = r- t X 0 lr.
Now suppose dim 'Je 2 = n < 00. Then up to unitary equivalence there exists
a unique irreducible pair U, V of unitary operators in 'Je2 such that un = vn = 1
and UV = VU exp 2~ i . By an analogue of the Stone-von Neumann Theorem for
the group {O, 1,2, ... ,n - I} with addition modulo n (See Exercise 13.8) there
exists a Hilbert space k such that r-tu 01r = 7l'(U), r-tv 01r = 7l'(V). The
algebra generated by U, V is ~('Je2) and by the same argument as in the case
n = 00 we have 7l'(X) = r- t X 0 lr. •
Theorem 29.6: (Stinespring's Theorem [130]) Let 'Jet, 'Je 2 be Hilbert spaces and
let T : ~('Je2) ----+ ~('Jet) be a linear operator satisfying the following conditions:
(i) T(I) = 1, T(X*) = T(X)*; (ii) IIT(X)II : : ; IIXII; (iii) if Xn ----+ X weakly
in 'Je2 then T(Xn) ----+ T(X) weakly in 'Jet; (iv) for every n = 1,2, ... the
correspondence ((Xij )) ----+ ((T(Xij ))), 1 ::::; i, j ::::; n from ~('Je2 0 en) into
~('Jet 0 en) preserves positivity. Then there exists a Hilbert space k, an isometry
V from 'Jet into 'Je2 0 k such that (a) T(X) = V* X 01 V for all X E ~('Je2); (b)
{(X 0 1)V ulX E ~('Je2)' U E 'Jet} is total in 'Je2 0 k; (c) If k' is another Hilbert
space and V' is an isometry from 'Jet into 'Je2 0 k' such that (a) and (b) hold
with k, V replaced respectively by k', V' then there exists a unitary isomorphism
W: 'Je2 0 k ----+ 'Je2 0 k' such that W(X 0 1)V = (X 0 1)V'.
Conversely, if V : 'Jet ----+ 'Je2 0 k is an isometry where k is any Hilbert space
then the map T : ~('Je2) ----+ ~('Jet) defined by T(X) = V* X 0 1 V satisfies the
conditions (i)-(iv).
Proof: To prove the first part consider the homomorphism 71' from ~('Je2) into
~('Je) and the isometry Vo : 'Jet ----+ 'Je given by Proposition 29.4. Write V = rvo
where r is defined by Proposition 29.5. Then
(Ut, V*(X 0 l)Vu2) = (Ut, vo*r-t(X 0 l)rVou2)
= (VoUt,7l'(X)VoU2) = (A(I, ut), A(X, U2))
= (Ulo T(X)U2) for all Ut, U2 E 'Jet.
This proves (a). (b) follows from the fact that {A(X,u)IX E ~('Je2)'U E 'Jet} is
total in 'Je. (c) is immediate from Proposition 7.2. The converse is immediate. •
Proposition 29.7: In Theorem 29.6 if dim 'Jej = nj < 00, j = 1,2 then dim k ::::;
nt n 2·
Proof: Since dim ~('Je2) = n~ condition (b) implies that n2 dim k ::::; n~nt. •
Proposition 29.8: An operator T : ~(~2) ---+ ~(~d satisfies the conditions

(i)-(iv) in Theorem 29.6 if and only if there exist operators Lj : ~l ---+ ~2'
j = 1,2, ... such that LjLjLj = 1 is a strongly convergent sum and
T(X) = LLiXLj for all X E ~(~2)'

j
If dim ~j = nj < 00, j = 1,2 then the number of Lj's can be restricted to be
~ nln2·
Proof: Choose an orthonormal basis {ej} for k in Theorem 29.6 and identify
~2 ® k with ~2 EB ~2 EB ... through the unitary isomorphism U defined by
Uv ® w = E9(ej,w)v, v E ~2'W E k.
j
Then the isometry V : ~l ---+ ~2 ® k can be equivalently described by a sequence

{Lj} of operators from ~l into ~2 such that
Vu=L 1uEBL 2UEB"', UE~l
where 1 = V*V = LjLiLj is a strongly convergent sum. By Theorem 29.6
T(X) = V*(X ® 1)V = LLiXLj.

j
The second part is immediate from Proposition 29.7.
Exercise 29.9: Let T, {Lj} be as in Proposition 29.8. Let T' be the map from the
set of states in ~l into the set of states in ~2 defined by tr pT(X) = trT'(p)X for
all X E '2Ji\(~2)' Then T'(p) = LjLjpLj where the right hand side is convergent
in the norm of the Banach space of trace class operators. (See Section 9.)
Exercise 29.10: Let ~ be an n-dimensional Hilbert space with orthonormal basis

{eJ, ... ,en} and let {Ul,' .. ,un} be a sequence of unit vectors in ~. Suppose
Li = IUi)(eil, 1 ~ i ~ n. Then
n n
T(X) = LLiXLi = L(Ui,Xui)lei)(eil, X E ~(~)
i=l i=l
n
T(X) = LYilei)(eil
i=l
where Yi = L7=lPijXj,Pij = l(ui,ejW, 1 ~ i,j ~ n· ((Pij)) is a stochastic
matrix and (Ui' ej) can be interpreted as the transition amplitude from the classical
Markov state i to the state j.
256 Chapter III: Stochastic Integration and Quantum Ito' s Formula
Exercise 29.11: If Tl : <J3('Jf 2 ) --t <J3('Jfd and T2 : <J3('Jf 3 ) --t <J3('Jf2 ) are
completely positive operators then their composition Tl T2 : <J3('Jf 3 ) --t <J3('Jfd is
also completely positive and (T1T2)' = T~T{ is a mapping from the set of states in
'Jf 1 into the set of states in 'Jf3 , T' being defined as in Exercise 29.9. In particular,
for any completely positive operator T : <J3('Jf) --t <J3('Jf), Tn is a completely
positive operator on <J3('Jf) for every n = 1,2, ...
In the case of Exercise 29.10, for the diagonal matrix X with diagonal entries
Xl,X2, ... ,xn,Tk(X) is the diagonal matrix with diagonal entries Yl,Y2,··· ,Yn
where Yi = L-jPij(k)xj, ((pij(k))) = ((Pij))k, 1 :::; i,j :::; n.
Exercise 29.12: (a) Let ho, 'Jf be Hilbert spaces. Suppose L : ho --t ho 0 'Jf
is an operator such that IILII :::; 1. From the identity L(L*L)n = (LL*)nL for
n = 0, 1, 2, ... it follows that
( (1 - L* L)I/2 -L* )
L (1 - LL*)1/2
is a unitary operator in ho ttl (h o 0 'Jf) = ho 0 (C ttl 'Jf).
(b) Let L : ho --t ho 0 'Jf be any bounded operator. Suppose H is any
bounded selfadjoint operator in ho. For every n > IILI12 the matrix
_ ((I-n- IL*L)I/2
Vn - n-~L
-n-~L*
(1 _ n-ILL*)1/2
) (e- in - 1H
°
e- in - 1H 01
° )
(29.7)
is a unitary operator in ho 0 (C ttl 'Jf).
Exercise 29.13: We introduce the following definition: if 'Jfj, 1 :::; j :::; n are
Hilbert spaces, T is a bounded operator in 'Jf i1 0· .. 0 'Jf ik ' 1 :::; il < ... < i k :::; n
and U is the unitary isomorphism
k
U : 'Jf l 0 ... 0 'Jf n --t Q9'Jf ij 0 {
j=1 l<i<n
ili'{i;,,~ ,i k }
induced by the permutation of indices then the operator U- 1T 0 lU is called the

ampliation of T from ®7=1 'Jf ij to ®~=1 'Jfi .
Let ho, 'Jf, L, H be as in Exercise 29.12. Denote by Vjn the ampliation of
Vn (defined by (29.7» from the tensor product of ho and the j-th copy of C ttl 'Jf
in (C ttl 'Jf)®n to
- ®n
'Jf n = ho 0 (C ttl 'Jf) .
For any °: :; t :::; 1 define the unitary operators
Un(t) = Vjn Vj-In'" Vin, j = [ntl·
Let en be the vector (1 ttl O)®n in (C ttl 'Jf)®n and let I n : ho --t 'Jf n be the
isometry Jnu = U 0 en. Then
30 Generators of quantum dynamical semigroups and the Gorini, ... Theorem 257
defines a completely positive operator in ~(ho). Furthermore Tn(t) = Ohnt ] where

On(X) = ein - 1H(1 - n- 1L * L )1/2 X(1 _ n- 1L * L )1/2 e -in- 1 H
+n-lein-IH L* X Q91Le- in - 1H ,
1 in X Q9 1 being the identity in 'fIC. In particular
lim Tn(t)(X) = etfl(X),X E ~(ho), 0:; t:; 1
n---+oo
where
O(X) = i[H,X]- ~{L* LX + XL* L - 2L* X Q9 lL}
and convergence holds in operator norm.
If {h} is an orthonormal basis in 'fIC and hoQ9'fIC is identified with hoffihoffi' ..
through the unitary isomorphism Wu Q9 v = ffi (fj, v)u then L can be expressed
as
Lu = EBLju, u E ho
j
where LjLjLj is a strongly convergent sum. Then 0 assumes the form
O(X) = i[H,X]- ~2)LiLjX + XLiLj - 2LiXLj).

j
This may be compared with the expression for og in Corollary 27.9, 27.10. Such
a comparison indicates the possibility of "convergence of quantum random walks
to quantum stochastic flows".
Exercise 29.14: The maps IEp in Exercise 16.10, lEn] in Proposition 18.3 and lEt]
in Propostion 26.7 are completely positive.
Notes
Theorem 29.6 appears in Stinespring [130]. Exercise 29.12, 29.13 are inspired by
the exposition of Meyer [88] on toy Fock spaces. For an actual convergence of
random walks in toy Fock spaces to flows in boson Fock spaces, see Parthasarathy
[107], Lindsay and Parthasarathy [78], Accardi and Bach [3].
30 Generators of quantum dynamical semigroups and the

Gorini, Kossakowski, Sudarshan, Lindblad Theorem
In Section 29 we have seen how a completely positive operator T from ~('fIC2)
into ~('fICI) may be interpreted as a quantum probabilistic analogue of a transition
probability when a one step transition is made from states in 'fIC 1 into states in 'fIC 2.
This suggests that the quantum analogue of a homogeneous Markov semigroup
of transition probabilities is a one parameter semigroup of completely positive
operators in ~('fIC). We introduce the following definition accordingly: A quantum
258 Chapter ill: Stochastic Integration and Quantum Ito's Fonnula
dynamical semigroup (in the Heisenberg picture) in a Hilbert space 'ZJe is a one
parameter family {Ttlt ~ O} of operators in the Banach space ~('ZJe) satisfying
the following conditions:
(i) To(X) =X for all X E ~('ZJe), TtTs = Tt+s for all s, t ~ 0;
(ii) limt---+oIITt(X) - XII = 0 for all X E ~('ZJe);
(iii) Tt (1) = 1, IITt(X)11 ::; IIXII,
w.lim Tt(Xn ) = Tt(X) whenever w.lim Xn = X;

n----+oo n---too
(iv) for any Xi, Yi E 27i1('ZJe), 1 ::; i ::; n, n = 1,2, ...
Z)i*Tt(XtXj)Yj ~ O.
i,j
A quantum dynamical semigroup {Ttlt ~ O} is said to be uniformly con-

tinuous if limt---+o sUPllxll9 IITt(X) - XII = O. For such a uniformly continuous
semigroup there exists a bounded operator () in ~(~) such that
(}(X) = lim r 1 (Tt(X) - X) for all X E 27i1(~) (30.1)

t---+o
where the limit is in the operator norm. Then Tt(X) = et9 (X) for all t ~ O. Our
goal in this section is to find the form of () and link it with the quantum stochastic
differential equations of Section 26-29. The operator () in (30.1) is called the
generator of {Tt }. We begin with some examples.
Example 30.1: Let {w(t)lt ~ O} be the standard Brownian motion process and
let L be a selfadjoint operator in the Hilbert space 'ZJe. Define
Tt(X) = lEeiw(t)L Xe-iw(t)L
where IE denotes expectation with respect to the probability measure of the process
w in the strong Bochner sense that for every fixed U E 'ZJe, the ~-valued function
f(x) = eixL Xe-ixLu is integrable in norm with respect to the distribution of
w(t). (See Section 5, Chapter V, [142].) Since the standard Brownian motion has
stationary independent increments
Tt+s(X) = lEei(w(t+s)-w(s))Leiw(S)L Xe-iw(S)Le-i(w(t+s)-W(S))L
=Tt(Ts(X)).
Since eiw(t)L is unitary in ~ it follows that {Tt } is a quantum dynamical semi-
group. If L is bounded then {Ttl is uniformly continuous. In such a case its gener-
ator () can be evaluated as follows. Define ad L in ~(~) by (ad L)(X) = [L,X].
Since w(t) has normal distribution with mean 0 and variance t
Tt(X) = lEeiw(t)ad L(X) = e-!(ad L)\X).

Thus the generator 0 of Tt is given by
O(X) = -~(ad L)2(X)

= _~(L2 X + XL 2 - 2LXL).
2
Example 30.2: A slightly more general case than the preceding example is ob-
tained as follows: let (WI (t), . .. ,Wd(t)) be the standard d-dimensional Brownian
motion and let Lj, 1 ~ j ~ d be bounded selfadjoint operators in 'fIC. Consider
the unitary operator-valued stochastic process U(t) = {U(t,~)It ~ O} satisfying
the stochastic differential equation
dU = {-i'LLjdwj - ~('L Lj)dt}U, U(O) = 1

i j
and put Tt(X) = IEU(t)* XU(t), X E ~('fIC) where IE denotes expectation with
respect to the probability measure of the d-dimensional Brownian motion. It fol-
lows from the classical Ito formula for Brownian motion that {Tt} has generator
o given by
Example 30.3: Let U be any unitary operator in 'Jf, and let {N(t)lt ~ O} be the
Poisson stochastic process with intensity A. Define
Tt(X) = IEU*N(t) XUN(t)
where expectation is taken with respect to the probability measure of the Poisson
process {N(t)}. Then the generator 0 of {Tt} is given by
O(X) = A(U* XU - X) = -~(L* LX + XL* L - 2L* XL)
where L = /).,U.
Example 30.4: Let {Uj 11 ~ j ~ n} be unitary operators in 'Jf, and let {Nj 11 ~
j ~ n} be n independent Poisson processes with respective intensities {Aj 11 ~
j ~ n}. Consider the unitary operator-valued stochastic process {W(t)lt ~ O}
satisfying the stochastic differential equation
dW = {'L(Uj - l)dNj}W, W(O) = 1.

j
Define
Tt(X) = IEW(t)* XW(t), t ~ 0, X E ~('Jf,).
Then {Tt} is a quantum dynamical semigroup with generator 0 given by

n
O(X) = L).j(U; XUj - X)
j=1
n
= -21 '""( *
L.. LjLjX + XLjLj
* - 2Lj* )
XLj
j=1
where Lj = fij Uj. 1 ::; j ::; n.
Example 30.5: Suppose T is a completely positive operator in ~(~) defined
by T(X) = LjLiXLj where LjLiLj = 1 is a strongly convergent sum. Then
Tt = exp ).t(T -1). t ~ 0 defines a quantum dynamical semigroup with generator
ogiven by
O(X) = -~).L(LiLjX + XLiLj - 2LiXLj).
j
This example may be viewed as a quantum probabilistic analogue of an embedded
Markov chain in classical probability theory.
We shall now investigate the properties of the generator of an arbitrary
uniformly continuous quantum dynamical semigroup (q.d.s.).
Proposition 30.1: Let 0 be the generator of a uniformly continuous quantum
dynamical semigroup {Tt} in ~(~). Then the following holds: (i) 0(1) = 0; (ii)
If Xi. Yi E ~(~). 1 ::; i ::; n and L~=IXiYi = 0 then Li,jYi*O(Xt Xj)lj ~ 0;
(iii) for any Xi. Yi E ~(~). 1 ::; i ::; n
L Yi*{O(xt Xj) - O(XnXj - xtO(Xj)}lj ~ O.
l:S;i,j:S;n
Proof: From property (iii) in the definition of a q.d.s.

0(1) = lim r 1 (Tt(1) - I) = O.
t.....,o
From property (iv) in the definition of q.d.s. and the equation LiXiYi = 0 we
get
L Yi*r 1{Tt(X; Xj) - X; Xj}lj = t- 1LYi*Tt(X; Xj)lj ~ o.
l:S;i,j:S;n i,j
Letting t -+ 0 we obtain (ii). To prove (iii) consider arbitrary elements Xi. Yi E
~(~). 1 ::; i ::; n and put Xo = I. Yo = - L~=IXiYi. Then L~=oXiYi = o.
Now applying (ii) to Xi. Yi. 0 ::; i ::; n we get
n n n
(- LXjlj)*O(I)(- LXjlj) + (- LXjlj)*LO(Xj)lj
j=1 j=1 j=1
n n
+ LYi*O(X;)(- LXjlj) + L Yi*O(X; Xj)lj ~ o.
i=1 i=1 l:S;i,j:S;n
Since O( I) = 0 we obtain (iii).
•
Proposition 30.2: Let 0 be as in Proposition 30.1. Then there exists a Hilbert

space 'J{ and for every X E ~('i1e) a bounded linear operator 1I"(X) : 'i1e ~ 'J{
satisfying the following:
(i) 1I"(X)*1I"(Y) = O(X*y) - O(X*)Y - X*O(Y);
(ii) The correspondence X ~ 1I"(X) is linear;
(iii) The set {1I"(X)uIX E ~('i1e), u E ~{} is total in ~{;
(iv) If (~{', 11"') is another pair satisfying (i)-(iii) with (~{, 11") repalced by (~{', 11"')
then there exists a unitary isomorphism r : 'J{ ~ ~(' such that r1l"(X)u =
1I"'(X)u for all X E ~('i1e), u E 'i1e.
Proof: From (iii) in Proposition 30.1 it follows that the function

K((X, u), (Y, v)) = (u, {O(X*Y) - O(X*)Y - X*O(Y)}v)
is a positive definite kernel on the space ~('i1e) x 'i1e. Hence by Proposition 15.4
there exists a Gelfand pair ('J{, A) where A : ~('i1e) x 'i1e ~ 'J{ is a map such that
(A (X, u), A(Y, v)) = (u, {O(X*Y) - O(X*)Y - X*O(Y)}v) (30.2)
for all u, v in 'i1e, X, Y in ~('i1e) and the set {A(X, u)IX E ~('i1e), u E 'i1e} is total
in ~L Since the right hand side of (30.2) is antilinear in each of the variables X, u
and linear in each of the variables Y, v it follows that A(X, u) is linear in each
of the variables X,u. Thus there exists an operator 1I"(X) : 'i1e ~ 'J{ satisfying
1I"(X)u = A(X, u). Further (30.2) implies
111I"(X)uI1 2 = (u, {O(X* X) - O(X*)X - X*O(X)}u).
Since 0 is bounded it follows that 1I"(X) is a bounded operator from 'i1e into 'J{. Now
(30.2) implies that (i) and (ii) hold. (iii) follows from the totality of {A(X,u)}.
Property (iv) is immediate from the last part of Proposition 15.4. •
Proposition 30.3: Let 0,11" be as in Proposition 30.2. Then there exists a *-unital
homomorphism p : ~('i1e) ~ ~('?J{) satisfying the following:
1I"(XY) = 1I"(X)Y + p(X)1I"(Y) for all X, Y E ~('i1e).
Proof: Let U E !JU('i1e). By Proposition 30.1, 30.2 we have

(1I"(UX)u -1I"(U)XU,1I"(UY)v - 1I"(U)Yv) =
(u, {O(X*Y) - O(X*U*)UY - X*U*O(UY)}v)
-(u,O{(X*) - O(X*U*)U - X*U*O(U)}Yv)
-(u,X*{O(Y) - U*O(UY) - O(U*)UY}v)
+(u, X* {-O(U*)U - U*O(U)}Yv)
= (u, {O(X*Y) - O(X*)Y - X*O(Y)}v)
= (1I"(X)u,1I"(X)v).
Thus the correspondence p(U) : 7f(X)u ---+ 7f(UX)u - 7f(U)Xu extends to an

isometry of 'Jf into itself. For any two unitary operators Uj , U2 in ~, p(Uj U2) =
p(Uj )P(U2) and p(l) = 1. Thus p(U) is unitary for every U in UU(~) and
p(U)7f(X) = 7f(UX) - 7f(U)X for all X in 0Ji\(~). For any X = L,CjUj , a
finite linear combination of unitary operators in ~ define p(X) = L,Cjp(Uj). If
L, CjUj = 0 then
I>jp(Uj)7f(Y) = 7f(I>jUjY) - 7f(L:>jUj )Y = o.
Thus p(X) is well-defined whenever X is a finite linear combination of unitary
operators. As in the proof of Proposition 29.4, p(X) is well-defined for all X in
0Ji\(~) and p(X)7f(Y) = 7f(XY) - 7f(X)Y for all Y in 0Ji\(~). This also implies
that the map X ---+ p(X) is a *-unital homomorphism from 0Ji\(~) into 0Ji\('Jf). •
Proposition 30.4: Let p : 0Ji\(~) ---+ 0Ji\('J£) be the *-unital homorphism defined
in Proposition 30.3. If w.limn->oo Xn = X in 0Ji\(~) then w.limn->oo p(Xn) =
p(X) in 0Ji\('J£).
Proof: Let w.limn->oo Xn = X in 0Ji\(~). By the definition of a q.d.s. we have

w.lim Tt(X n) = Tt(X) for all t 2: O.
n-+oo
Thus for any u, v E ~, t 2: 0 we have
tk tk
L k! (u, (}k(X)V).
00 00
~:!!~L k! (u, (}k(Xn)v) = (30.3)

k=O k=O
Putting
tk - 2
L T!(u, (}k(Xn)v),
00
<Pn(t) =
k=2
<p(t) = f t:~2
k=2
(u, (}k(X)V)
we conclude from (30.3) that

lim (u, {(}(Xn) - (}(X)}v)
n-+oo
+ t<Pn(t) - t<p(t) = 0 for all t > O. (30.4)
Since (} is bounded and sUPn11Xn11 < 00 (by uniform boundedness principle) there
exists a > 0 such that
max(l<pn(t)l, 1<p(t)l) ::; eta for all n.
If E > 0 is arbitrary choose to> 0, such that toe ato < i. Then (30.4) implies
limn-+ool(u, {(}(Xn) - (}(X)}v) I ::; limn-+ool(u, {(}(Xn) - (}(X)}v)
+to<Pn(tO) - to<p(to) I + 2toetao

::; E.
Since t is arbitrary
lim (u, B(Xn)v) = (u, B(X)v) for all u, v E 'JC. (30.5)
n--+oo
For any Yi E ~('JC), Ui E 'JC, i = 1,2 we have from the definition of p and 1f in
Proposition 30.3 and (30.5)
lim n --+ oo (1f(Yj )Uj, p(Xn )1f(Yi)U2)
= lim (Uj, {B(Yt XnYi) - B(Yt)XnY2 - Yj*B(XnYi)

n--+oo
-[B(yt Xn) - B(Yt)Xn - YtB(Xn )]Y2}U2)
= (Uj, {B(yt XY2) - B(Yj*)XYi - Yj*B(XY2)
-[B(yt X) - B(Yt)X - YtB(X)]Y2}U2)
= (1f(YI)Uj,p(X)1f(Yi)U2). (30.6)
If IIYII ~ 1 then
and hence
Y + Y* UI + U2 Y - Y* VI + Vi
2 2 2i 2
where Ui, Vi, i = 1,2 are unitary. Thus IIp(Y)11 ~ 2 whenever IIYII ~ 1. Hence
sUPnllp(Xn)11 < 00. Now the totality of the set {1f(Y)uIY E ~('JC),u E 'JC} in 'J{
and (30.6) implies that w.limn->oo p(Xn) = p(X). •
Proposition 30.5: Let B be the generator of a uniformly continuous quantum

dynamical semigroup {Td in ~('JC). Then there exists a Hilbert space k and a
linear map 1f : ~('JC) ---> ~('JC 0 k) such that
(i) {1f(X)uIX E ~('JC), U E 'JC} is total in 'JC 0 k;
(ii) 1f(X)*1f(Y) = B(X*Y) - B(X*)Y - X*B(Y);
(iii) 1f(XY) = 1f(X)Y + (X 0 1 )1f(Y) for all X, Y E ~('JC).
Proof: Let B, 1f, P be as in Proposition 30.3. By Proposition 29.5 and 30.4 there
.
exists a Hilbert space k and a unitary isomorphism r : 'J{ ---> 'JC 0 k such that
p(X) = r- j X 01r for all X in ~('JC). Define 1fo(X) = r1f(X). Then properties
(i)-(iii) with 1f replaced by 1fo follow from Proposition 30.2 and the definition
~~
Proposition 30.6: Let B be the generator of a uniformly continuous quantum

dynamical semigroup {Tt } in ~('JC). Then there exists a (finite or infinite) sequence
{Lj} in ~('JC) satisfying the following
(i) sup LII[Lj,X]uI1 2 < 00; (30.7)

lI u ll:':1 .
IIXII:':' J
(ii) O(X*Y) - O(X*)Y - X*O(Y) = 2)Lj ,Xl*[Lj' Yl

j
for all X, Y in ~('ZJe) where the right hand side is a strongly convergent sum.
Proof: In Proposition 30.5 choose an orthonormal basis {ej} in k and consider the
unitary isomorphism fa : 'ZJe 0 k --+ 'ZJe EB 'ZJe EB ... satisfying fau 0 v = EBj (ej, v)u
for all u E 'ZJe, v E k. Then for any X E ~('ZJe) there exist operators 7fj(X),
j = 1,2, ... such that fa7f(X)u = EBj7fj(X)u for all u E 'ZJe. Then by property
(iii) in Proposition 30.5 we have 7fj(XY) = 7fj(X)Y + X7fj(Y) for each j =
1,2, ... Each 7fj is a bounded derivation of ~('ZJe). Hence there exists Lj E ~('ZJe)
such that 7fj(X) = [Lj,Xl for every j (See [30].) Property (i) follows from the
boundedness of 7f in Proposition 20.5. Property (ii) follows from property (ii) in
Proposition 30.5. •
Proposition 30.7: Let {Lj} be a sequence in ~('ZJe) satisfying (30.7). Then
(30.8)
Proof: In the sum on the left hand side of (30.7) put X = Iu) (u I and ob-
serve that II[Lj,XluI1 2 = IILjul12 - l(u,Lju)12. Thus (30.8) is immediate
~(3~. •
Proposition 30.8: Let {L j } be a sequence in ~('ZJe) satisfying (30.7). Then for

any projection P in 'ZJe
sup 2:11[PLjP,PXPluW < 00.

IIXII<S1 ,
lIull=I,Pu=u J
Proof: When Pu = u we have [PLjP,PXPlu = P[Lj,PXPlu and hence

II [PLjP, P XPlu11 ::::; II[Lj, PXPlull. Now the required result follows from (30.7).
Proposition 30.9: Let dim'ZJe = n < 00. Denote by dU the differential of the
normalised Haar measure of the compact group OU('ZJe). Then there exist positive
constants a, [3 such that a + n[3 = n -I and
(i) J IILUul1 2 dU = n- I tr L* L;
(ii) J I(u, U* LUu)1 2dU = atr L* L + [31 tr LI2
for all L in ~('ZJe) and unit vectors u in 'ZJe.

Proof: By the two-sided invariance of the Haar measure of au(~) the left hand
side of (i) is independent of the unit vector u. Thus for any orthonormal basis
el, ... ,en in ~ we have
j IILUul1 2dU = n- I L j IILUejl12dU = n- I tr L* L

J
which proves (i). To prove (ii) consider the sesquilinear form
B(L, M) =j (u, U* LUu) (u, U* MUu)dU, Ilull = 1.

in the vector space ~(~). Then
B(U* LU, U* MU) = B(L, M) for all L, M E ~(~), U E au(~) (30.9)
and by the argument in the proof of (i)
B(L,I)=n-ItrL. (30.10)
For any two one dimensional projections P, Q such that PQ = 0 put
a = B(P, P), b = B(P, Q).
By (30.9) a and b are positive constants independent of P, Q. By (30.10) a + (n-
1)b = n- I . By Schwarz's inequality
b = j(U,U*Puu)(U,U*QUU)dU
(30.11)
::; {j(u,U*PUU)2dU j(U,U*QUU)2dU}I/2 = a.
For any selfadjoint L with spectral resolution L = I: AjPj where each Pj is a

one dimensional projection we have
B(L,L) = a L AJ + b LAjAk
j j#
= a tr L2 + b{(tr L)2 - tr L2}.

For any two Hermitian L,M we have B(L,M) = HB(L+M,L+M) -B(L-
M, L - M)} and for arbitrary L, M, L = LI + iL2, M = MI + iM2 where
L I, L2, M I , M2 are Hermitian. Thus for any L, M
B(L, M) = (a - b) tr L* M + b tr Ltr M.
Putting a = a - b,{3 = b, L = M and using the relation a + (n - 1)b = n- I we
obtain (ii) where a, {3 are positive and a + n{3 = n -I. •
Proposition 30.10: Let dim ~ = n < 00 and let {L j } be a sequence in ~(~)

satisfying (30.7). Then
L tr(L j - n- I tr L j )*(Lj - n- I tr L j ) < 00.
j
Proof: From Proposition 30.7, 30.9 we have for any unit vector U in 7Je
00 > 2;= j(IILjUUII 2 -1(u,U*LjUu)12)dU

J
= n{3 ~::>r(Lj - n- 1 tr Lj )*(Lj - n- 1 tr L j ).

j
•
Proposition 30.11: Let 7Je be any Hilbert space and let {Lj} be any sequence
in CJ3(7Je) satisfying (30.7). Then there exsits a sequence {Cj} of scalars such that
L,j(Lj - cj)*(Lj - Cj) is strongly convergent.
Proof: Let u, v E 7Je be two linearly independent unit vectors and let aj =
(u, Lju), bj = (v, Ljv). Denote by P the projection on the subspace spanned by
!
u and v. Let mj = tr PLjP. From Proposition 30.8, 30.10 we have
L)II(PLjP - mj)ul1 2 + II(PLjP - mj)vI12} < 00. (30.12)

j
On the other hand from Proposition 30.7 we have
j j
j j
(30.13)
Similarly
LII(PLjP - bj)vl1 2 < 00. (30.14)
j
Inequalities (30.12)-(30.14) imply
LI(u,Lju) - (v,Ljv)1 2 < 00

j
for any two linearly independent unit vectors u, v. Choose and fix any unit vector
Uo and put Cj = (uo,Ljuo). Then for any unit vector W such that wand Uo are
linearly independent we have
j j
+2LI(w,Ljw) - (uo,Ljuo)12 < 00.

j
When W = cUo for some scalar C the left hand side of this inequality is already
finite. •
Theorem 30.12: (Gorini, Kossakowski, Sudershan [50], Lindblad [75].) Let '?J£
be a Hilbert space. An operator () in the Banach space 0l3('?J£) is the generator of
a uniformly continuous quantum dynamical semigroup if and only if there exists
a Hilbert space k, a bounded linear operator L : '?J£ -+ '?J£ 181 k and a bounded
selfadjoint operator H in '?J£ satisfying the following:
(i) O(X) = i[H, X]- HL* LX + XL* L - 2L* X 181 lL} for all X E 0l3('?J£);
(ii) The set {(LX - (X 181 I)L )uIX E 0l3('?J£), u E '?J£} is total in '?J£ 181 k.
Proof: First we prove necessity. Let 0 be the generator of a q.d.s. {Tt} in

0l3('?J£). Using Proposition 30.2 and 30.5 construct a Hilbert space k and a linear
map 1r : 0l3('?J£) -+ 0l3('?J£ 181 k) so that properties (i)-(iii) of Proposition 30.5 hold
and the set {1r(X)uIX E 0l3('?J£),u E '?J£} is total in '?J£ 181 k. As in the proof of
Proposition 30.6 construct the unitary isomorphism fo : '?J£ 181 k -+ '?J£ E!:l '?J£ E!:l ••.
so that fo1r(X)u = EBj[Lj,X]u for all X E 0l3('?J£),u E '?J£ and (30.7) holds. By
Proposition 30.11 we may choose {Cj} such that L.jLiLj is strongly convergent.
Define
Lu = foj{E9Lju}.
j
Then
1r(X) = LX - X 181 1 L, X E 0l3('?J£),
1 being the identity operator in k. Define
OJ(X) = -~{L* LX + XL* L - 2L* X 181 lL}.
By straightforward algebra and property (ii) of Proposition 30.5 we get
OJ(X*Y) - OJ(X*)Y - X*Oj(Y) = 1r(X)*1r(Y)

= O(X*Y) - O(X*)Y - X*O(Y) for all X, Y E 0l3('Je).
Thus () - OJ is a bounded derivation of 0l3('?J£). Furthermore we have O(X*) =

O(X)* and OJ (X*) = OJ (X)*. Hence there exists a bounded selfadjoint operator
H in '?J£ such that O(X) - OJ (X) = i[H, X] for all X. Thus (i) and (ii) hold and
the proof of necessity is complete.
To prove sufficiency consider the unitary isomorphism fo : '?J£ 181 k -+ 'Je E!:l '?J£ E!:l' •.
satisfying fou 181 v = EBj(ej, v)u where ej, e2, ... is an orthonormal basis in k.
Then L is given by foLu = EBjLju for all u in '?J£ where {Lj} is a sequence of
bounded operators in '?J£, L.jLiLj is strongly convergent and IILI12 = L.jLiLj.
Property (i) can be expressed as
()(X) = i[H,X]- ~ ~(LjLjX + XLjLj - 2LjXLj ).

J
Putting SJ = 6J, Ab = Ai, A~ = Ai in Theorem 27.8, consider the unitary solution

{U(t)lt ~ O} of (27.21):
dU = {~)LjdAj - LjdAj ) - (iH + ~2::LjLj)dt}U, U(O) = 1.

j j
By Corollary 27.10
lEoU(t)* X Q9 1 U(t) = etll(X) for all X E 9A('Je).

The left hand side expression shows that Tt = etll is a q.d.s. with generator (). •
Corollary 30.13: A bounded operator () in 9A('Je) is the generator of a uniformly

continuous quantum dynamical semigroup if and only if there exists a bounded
selfadjoint operator H and a sequence {Lj} of bounded operators in 'Je such that
LjLjLj is strongly convergent and
()(X) = i[H,X]- ~ '2;(LjLj X + XLjLj - 2LjXLj)

J
for all X in 9A('Je). In such a case the sequence {Lj} may be so chosen that the
set {EBj[Lj,X]uIX E 9A('Je),u E 'Je} is total in EBj'Je. It is possible that the
sequence {Lj} may be vacuous, finite or infinite. In the vacuous case EBj'Je is
interpreted as {o}.
Proof: This is already contained in the proof of Theorem 30.12.

•
Proposition 30.14: Let () be the generator of a uniformly continuous quantum
dynamical semigroup in 9A('Je). Suppose k and k' are Hilbert spaces, L : 'Je ---+
'Je Q9 k, L' : 'Je ---+ 'Je Q9 k' are bounded linear operators and H, H' are bounded
selfadjoint operators in 'Je satisfying the following:
(i) ()(X) = i[H,X]- ~(L* LX + XL* L - 2L* X Q9 lkL)
= i[H',X]- ~(L'* L'X + XL'* L' - 2L'* X Q9 lklL') (30.15)

for all X E 9A('Je);
(ii) The sets {(LX - X Q9 lkL )ulX E 9A('Je), u E 'Je}, {(L' X - X Q9 lk ' L )ulX E
9A('Je), u E 'Je} are total in 'Je Q9 k, 'Je Q9 k' respectively.
Then there exists a unitary isomorphism V : k ---+ k' a vector v' E k' and a real
constant a such that
L' = (1~ Q9 V)L + EV 1 (30.16)
H' = H + a + ;i(E;'l~ Q9 VL - L*I~ Q9 V*Evl) (30.17)

where EV1 : "J£ ~ "J£ 0 k' is the operator defined by

EV'U = U 0 v' for all U E "J£. (30.18)
Conversely, if k, k' are Hilbert spaces and the pairs (H', L'), (H, L) are related
by (30.16)-(30.18) where V is a unitary isomorphism from k to k', a is a real
scalar and v' E k' then (30.15) holds.
Proof: Define
7r(X) = LX - X 01kL,7r'(X) = L'X - X 01k,L'.
Then (30.15) implies
O(X*Y) - O(X*)Y - X*O(Y) = 7r(X)*7r(Y) = 7r'(X)*7r'(Y).
for all X, Y in '!f3("J£). Thus
(7r(X)u,7r(Y)v) = (7r' (X)u, 7r' (Y)v)
for all X, Y in '!f3("J£) and u, v in "J£. By condition (ii) and Proposition 7.2 there
exists a unitary isomorphism f : "J£ 0 k ~ "J£ 0 k' satisfying
f7r(X) = 7r'(X) for all X. (30.19)
Thus
f7r(XY) = f(7r(X)Y + X 0 lk7r(Y))
= 7r'(XY) = 7r'(X)Y + X 0 1kl7r'(Y)
= f7r(X)Y +X 0 1k,f7r(Y).
By condition (ii) we conclude that
fX 0 lk = X 0 lk,f for all X E '!f3("J£).
In other words there is a unitary isomorphism V : k ~ k' such that f = 1;/£ 0 V.
By (30.19)
1;/£ 0 V(LX - X 01kL) = L'X - X 0 l k,L'
which can be expressed as
{1;/£ 0 VL - L'}X = X 0 1kl {l;/£ 0 VL - L'}. (30.20)
Let
B = 1;/£ 0 V L - L'. (30.21 )
Choose an orthonormal basis {ej} in k' and write
Bu = LBjU0ej
j
where B j E '!f3("J£). Then (30.19) becomes
LBjXU 0 ej = LXBjU 0 ej for all X E '!f3("J£), U E "J£.

j j
Thus each Bj is a scalar -bj and B assumes the fonn
Bu = -u ® L bjej = -u ® v'.
By the definition in (30.18), B = -Ev 1 • Now (30.21) becomes L' = he®VL+Ev'

which is (30.16). Routine algebra shows that
L'* L' + XL'* L' - 2L'* X ® lklL'

= L* LX + XL* L - 2L* X ® lkL + [E;/1:Je ® VL - L*I:Je ® V* EVI,Xl.
Now (30.15) implies
[i(H'-H),Xl = ~[E;/l:Je®VL-L*I:Je®V*Evl,Xl
for all X E 9A(~). This implies (30.17) for some real scalar a. This proves the
first part. The converse is a consequence of routine algebra. •
Proposition 30.15: Let {Lj},j = 1,2, ... be a finite or infinite sequence of

bounded operators in ~ such that L-jLiLj is a strongly convergent sum. Then
the set
S = {$[Lj,XluIX E 9A(~),u E 71f} (30.22)
j
is not total in ~ Ee ~ Ee ... if and only if there exist scalars Co, CJ, • •. such that
0< L-lcjl2 < 00 and
(30.23)
is a strongly convergent sum.
Proof: Define the operator L : ~ -+ ~ Ee ~ Ee •.• by Lu = EBjLju, U E 71f.

Then L is a bounded operator with IILII = IIL-jLiLjII1/2. Furthennore L* is a
bounded operator satisfying
L*$Uj = I)iUj,
j j
II LLjUjl1 ~ IIL*II(LIIUjI12)1/2. (30.24)

j
Let X = X Ee X Ee •.• in ~ Ee 71f Ee .•. , X E 9A(~). Then
(LX - XL)u = $[Lj,Xlu,

j
(30.25)
j j
The set S defined by (30.22) is not total in 'Je EEl 'Je EEl ... if and only if there exists
a vector EBjVj -lOin 'Je EEl 'Je EEl ... such that
(EBVj, EEl [Lj, X]u) = 0 for all X E rzJ3('Je), u E 'Je.

j
By (30.25) this can be expressed as
2)Lj,X]Vj = 0 for all X E rzJ3('Je). (30.26)

j
Without loss of generality assume that LI -10, Ilvlll = 1. Putting PI = IVI)(vd,

•Cj = (Vj, VI) for j :::: 1 and substituting X PI for X in (30.26) we get
2:)jLjXvl = (Vlo LLjVj)Xvl (30.27)

j~l j~l
where L: j lcjl2 ~ L: j llvjl12 < 00 and L:/3jLj and L:jLjvj are well-defined
thanks to (30.24). Putting Co = -(L:jLjvj,vI), (30.27) can be written as (Co +
L:j~lcjLj)XVl = 0 for all X E rzJ3('Je) where CI = 1. This implies Co +
L:j~l cjLj = 0 and proves the only if part. To prove the converse we have
to only observe that (30.23) implies that S is contained in the proper subspace
{EBjWjlL:j~lCjWj = O}.
As a corollary to Proposition 30.14, 30.15 we obtain the following sharpened
version of Theorem 30.12.
Theorem 30.16: Let () be the generator of a uniformly continuous quantum

dynamical semigroup in rzJ3('Je) and let p be any state in 'Je. Then there exists a
bounded selfadjoint operator H and a finite or infinite sequence {Lj},j = 1,2, ...
of elements in rzJ3('Je) satisfying the following:
(i) tr pLj = 0 for each j;
(ii) L:jLjLj is a strongly convergent sum;
(iii) If L: j ICj 12 < 00 and Co + L:jcjLj = 0 for scalars {Cj} then Cj = 0 for each
j;
(iv) ()(X) = i[H,X]- !L:j(LjLjX + XLjLj - 2LjXLj) for all X in rzJ3('Je).
If H', {Lj} is another family of elements in rzJ3('Je) such that H' is selfadjoint
and the sequence {Lj} is finite or infinite then conditions (i)-(iv) are fulfilled
with H, {Lj} replaced respectively by H', {Lj} if and only if the lengths of the
sequences {Lj}, {Lj} are equal and for some scalar a and a unitary matrix (P.'ij))
H' = H + a, L~ = LAijLj for each i.

j
Proof: By Theorem 30.12 and Corollary 30.13 there exist {Lj} satisfying proper-
ties (ii) and (iv) and the property that the set S defined by (30.22) is total in ~(~).
By Proposition 30.15 it follows that (iii) holds for {Lj}. Now consider a state p
and define mj = trpLj and Mj = Lj - mj' Suppose that p = LjPjluj)(ujl
where Pj > 0 for each j and {Uj} is an orthonormal sequence. Then
Imjl2 = ILPi(Ui,LjUi)12
i
~ LPil(Ui,Lj Ui)1 2
~ LPi(Ui, LjLjUi).
i
Thus
j j
Hence {Mj} satisfies properties (i}-(iv) with {Lj } replaced by {Mj}. This proves
the first part.
To prove the second part assume that (H, {Lj}) and (H', {Lj)} satisfy the
properties (i}-(iv). By Proposition 30.15 and (30.14) there exists a unitary matrix
{>.ij} and a sequence {Vj} such that Lj = LkAjkLk + Vj for every j. Since
tr pLj = tr pLj = 0 for all j it follows that Vj = 0 for all j. Using (30.16) and
(30.17) we conclude that v' = O,H' = H + a and V is determined by ((Ajk)).
This completes the proof. •
In reversible quantum dynamics the Hamiltonian H is determined up to
an arbitrary real additive scalar. In uniformly continuous but not necessarily re-
versible quantum dynamics in ~(~) the concerned generator () is determined
by the "parameters" (H, L\, L2, ... ) where H is a bounded selfadjoint oper-
ator and L 1 , L 2 , . •• is an empty, finite or infinite sequence of bounded oper-
ators satisfying the following conditions: (i) LjLjLj is strongly convergent;
(ii) If Lj2:olCjl2 <+ Lj2:1CjLj = 0 then Cj = 0 for every j;
OO,CQ (iii)
()(X) = i[H,X]- ~Lj(LjLjX + XLjLj - 2LjXLj) for all X in ~(~).
If (H',L~,L~, ... ) is another set of parameters satisfying (i}-(iii) when
H, L\, L2, ... are replaced respectively by H', L~, L~, . .. then the lengths of the
sequences {Lj},{Lj} are equal and there exists a unitary matrix ((Aij)) whose
order is equal to the length (= k,) of the sequence {Lj}, a complex scalar valued
sequence {aj} of length k satisfying Lj laj 12 < 00 and a real scalar f3 such that
Li = LAijLj + ai,
j
H' = H + f3 + ;i (S - S*)
where S = Li,jaiAijLj. Thus the arbitrariness in the choice of the parameters

is characterised by the group {(g, A,,6) Ig E k, A E OU( k),,6 E Iffi} with the group
operation
(g, A, ,6) (g', A', ,6') = (g + Ag', AA',,6 + ,6' + Im(g, Ag')),
k being a Hilbert space of dimension k. This is a central extension of the Euclidean
group of k. Each element of this group can be interpreted as a "phase-factor" of
the generator O. When the algebra CJ3('lIC) is replaced by a von Neumann subalgebra
CJ3 0 of CJ3(;¥t') and the dynamical semigroup {Tt } is determined by Tt = et9g in
Theorem 28.6 and Exercise 28.9 it may be interesting to describe the form of 08
more precisely up to a "phase-factor group".
Notes
The notion of quantum dynamical semigroups appears in Kraus [70]. Theorem
30.12 was established in the finite dimensional case by Gorini, Kossakowski and
Sudarshan [50] and in the general case by Lindblad [75]. For a detailed account
of completely positive maps and uniformly continuous quantum dynamical semi-
groups, see Davies [28]. For more recent developments in the case of C*-algebras
and literature on the subject see the survey article by Evans [32].
References
[1] Accardi, L., Frigerio, A, Gorini, V. (Eds): Quantum Probability and Ap-
plications to the Quantum Theory of Irreversible Processes, Springer LNM
1055 (1984) Berlin.
[2] Accardi, L., von Waldenfels, W.: Quantum Probability and Applications
II, Springer LNM 1136 (1985), III 1303 (1988), IV 1396 (1989), V 1442
(1990).
[3] Accardi, L. Bach, A.: The harmonic oscillator as quantum centrallimit of
Bernoulli processes, Probab. Th. ReI. Fields, to appear.
[4] Accardi, L., Cecchini, c.: Conditional expectations in von Neumann al-
gebras and a theorem of Takesaki, J. Funct. Anal. 45, 245-273 (1982).
[5] Accardi, L., Fagnola, F.: Stochastic integration, Springer LNM 1303,6-19
(1988).
[6] Accardi, L., Frigerio, A, Lewis, J.T.: Quantum stochastic processes, Publ.
RIMS, Kyoto Univ. 18,97-133 (1982).
[7] Accardi, L. Journe, J.L., Lindsay, J.M.: On multidimensional Markovian
cocycles, Springer LNM 1396, 59-67 (1989).
[8] Applebaum, D.B., Hudson, R.L.: Fermion Ito's formula and stochastic
evolutions, Commun. Math. Phys. 96, 473-496 (1984).
[9] Araki, H.: Factorizable representations of current algebra, Publ. RIMS,
Kyoto Univ. Ser. A 5,361-422 (1970).
[10] Arveson, W.: An Invitation to C*-algebras, Springer-Verlag Graduate
Texts in Mathematics 34 (1976) Berlin.
[11] Bach, A: Indistinguishable Classical Particles, Mimeographed Notes
(1989) Haus Stapel, D-4409 Havixbeck, Germany.
[12] Barchielli, A, Lupieri, G.: Quantum stochastic calculus, operation-valued
stochastic processes and continual measurements in quantum mechanics.
J. Math. Phys. 26, 2222-2230 (1985).
[13] Barchielli, A.: Input and output channels in quantum systems and quantum
stochastic differential equations, Springer LNM 1303, 37-51 (1988).
[14] Bargmann, V.: On unitary ray representations of continuous groups, Ann.
Math. 59, 1-46 (1954).
[15] Bargmann, V.: Note on Wigner's theorem on symmetry operations, 1.
Math. Phys. 5, 862-868 (1964).
[16] Barnett, c., Streater, R.F., Wilde, I.F.: The Ito-Clifford integral, J. Funct.
Anal. 48, 172-212 (1982).
[17] Barnett, c., Streater, R.F., Wilde, I.F.: The Ito-Clifford integral II, Stochas-
tic differential equations, 1. London Math. Soc. 27, 373-384 (1983).
[18] Bell, J.S.: On the Einstein-Podolsky-Rosen paradox, Physics 1, 195-200
(1964).
276 References
[19] Berezin, F.A: The Method of Second Quantization, Academic Press

(1966) New York.
[20] Bhat, B.V.R., Parthasarathy, K.R.: Generalized harmonic oscillators in
quantum probability, in: Seminaire de Probabilites XXV 1990, 39-51,
Springer LNM 1485 (Ed: J. Azema, P.A Meyer and M. Yor).
[21] Bialynicki-Birula, I.: Entropic uncertainty relations in quantum mechanics,
Springer LNM 1136,90--103 (1985).
[22] Biane, Ph.: Marches de Bernoulli quantiques, in: Seminaire de Probabilites
XXIV 1988/89 Springer LNM 1426 (Ed: J. Azema, P.A. Meyer, M. Yor).
[23] Bratteli, 0., Robinson, D.W.: Operator Algebras and Quantum Statistical
Mechanics I (1979), II (1981) Springer, Berlin.
[24] Chung, K.L.: A Course in Probability Theory, Academic Press (1974)
New York (2nd edition).
[25] Cockroft, AM., Hudson, R.L.: Quantum Mechanical Wiener Process, J.
Multivariate Anal. 7, 107-124 (1977).
[26] Cook, J.M.: The mathematics of second quantization, Trans. Amer. Math.
Soc. 74, 222-245 (1953).
[27] Cushen, C.D., Hudson, R.L.: A quantum mechanical central limit theorem,
J. Appi. Probab. 8,454-469 (1971).
[28] Davies, E.B.: Quantum Theory of Open Systems, Academic Press (1976)
New York.
[29] Dirac, P.A.M.: The Principles of Quantum Mechanics, Oxford University
Press (1958) London.
[30] Dixmier, J.: Von Neumann Algebras, North Holland (1981) Amsterdam
(translated from French).
[31] Doob, J.L.: Stochastic Processes, John Wiley & Sons (1953) New York.
[32] Evans, D.E.: Quantum dynamical semigroups, symmetry groups and lo-
cality, Acta Appi. Mathematicae 2, 333-352 (1984).
[33] Evans, M.P., Hudson, R.L.: Multidimensional diffusions, Springer LNM
1303, 69-88 (1988).
[34] Evans, M.P.: Ph.D. Thesis, University of Nottingham (1988).
[35] Evans, M.P.: Existence of quantum diffusions, Probab. Th. ReI. Fields 81,
473-483 (1989).
[36] Fagnola, F.: A martingale characterization of quantum Poisson Processes,
Probab. Th. ReI. Fields 84, 323-333 (1990).
[37] Fagnola, F.: Pure birth and pure death processes as quantum flows in Fock
space, Universita di Trento preprint (1990), to appear in Sankhya A
[38] Fannes, M., Quaegebeur, J.: Infinite divisibility and central limit theorems
for completely positive mappings, Springer LNM 1136, 177-201 (1985).
[39] Fannes, M., Quaegebeur, J.: Infinitely divisible completely positive map-
pings, Probab. Th. ReI. Fields, 79, 369-403 (1988).
References 277
[40] Feller, W.: An Introduction to Probability Theory and its Applications,

Vol. I, John Wiley & Sons (1977) New York.
[41] Feynman, R.P.: Concept of probability in quantum mechanics, Proc. 2nd
Berkeley Symposium on Mathematical Statistics and Probability, 533-541
(1951 ) Berkeley.
[42] Feynman, R.P., Leighton, R.B., Sands, M.: The Feynman Lectures on
Physics, Vol. III, Addison-Wesley (1965) Reading, Mass.
[43] Fock, V.A.: Konfigurationsraum und zweite Quantelung, Z. Phys. 75, 622-
647 (1932).
[44] Frigerio, A., Maassen, H.: Quantum Poisson processes and dilations of
dynamical semigroups, Universita di Udine preprint (1988).
[45] Garding, L., Wightman, A.S.: Representations of the anticommutation re-
lations, Proc. Nat. Acad. Sci. U.S.A. 40, 617-621 (1954).
[46] Garding, L., Wightman, A.S.: Representations of the commutation rela-
tions, Proc. Nat. Acad. Sci. U.S.A. 40, 622-626 (1954).
[47] Giri, N., Waldenfels, W.v.: An algebraic version of the central limit the-
orem, Z. Wahr. theorie. 42, 129-134 (1978).
[48] Gleason, A.M.: Measures on the closed subspaces of a Hilbert space, J.
Math. and Mech. 6, 885-894 (1957).
[49] Gnedenko, B.V., Kolmogorov, A.N.: Limit Distributions for Sums of In-
dependent Random Variables, Addison-Wesley (1954) Cambridge, Mass.
(translated from Russian).
[50] Gorini, V., Kossakowski, A., Sudarshan, E.C.G.: Completely positive
dynamical semigroups of n-Ievel systems, J. Math. Phys. 17, 821-825
(1976).
[51] Guichardet, A.: Symmetric Hilbert spaces and Related Topics, Springer
LNM 261 (1972) Berlin.
[52] Hall, M.: Combinatorial Theory, John Wiley & Sons (1967) New York.
[53] Halmos, P.R.: Introduction to the Theory of Hilbert Spaces and Spectral
Multiplicity, Chelsea (1957) New York.
[54] Hegerfeldt, G.C.: Noncommutative analogs of probabilistic notions and
results, J. Funct. Anal. 64, 436--456 (1985).
[55] Holevo, A.S.: Conditionally positive definite functions in quantum prob-
ability, Proc. Int. Congo Math. 1011-1020 (1986) Berkeley.
[56] Holevo, A.S.: Limit theorems for repeated measurements and continuous
measurement processes, Springer LNM 1396, 229-255 (1989).
[57] Hudson, R.L., Karandikar, R.L., Parthasarathy, K.R.: Towards a theory
of noncommutative semimartingales adapted to Brownian motion and a
quantum Ito's formula, Springer Lecture Notes in Control and Information
Sciences 49, 96-110 (1983) (Ed: G. Kallianpur).
278 References
[58] Hudson, RL., Parthasarathy, K.R: Quantum diffusions, Ibid., 111-121

(1983).
[59] Hudson, RL., Parthasarathy, K.R: Quantum Ito's formula and stochastic
evolutions, Commun. Math. Phys. 93, 301-323 (1984).
[60] Hudson, RL., Parthasarathy, K.R: Stochastic dilations of uniformly con-
tinuous completely positive semigroups, Acta Appl. Mathematicae 2, 353-
398 (1984).
[61] Hudson, RL., Parthasarathy, K.R: Construction of quantum diffusions,
Springer LNM 1055, 173-198 (1984) Berlin.
[62] Hudson, RL., Parthasarathy, K.R: Unification of fermion and boson
stochastic calculus, Commun. Math. Phys. 104,457-470 (1986).
[63] Hudson, RL., Lindsay, J.M., Parthasarathy, K.R.: Stochastic integral rep-
resentation of some quantum martingales, in: From Local Times to Global
Geometry, Control and Physics, Proc. Warwick Symposium 1984/85, 121-
131 (1986) Pitman RNM. (Ed: K.D. Elworthy).
[64] Hudson, RL., Lindsay, J.M.: On characterising quantum stochastic evo-
lutions, Math. Proc. Camb. Phil. Soc. 102, 363-369 (1987).
[65] Hudson, RL.: Quantum diffusions and cohomology of algebras, in: Proc.
1st World Congress of Bernoulli Soc. 479-485 (1987) VNU Science Press,
Utrecht (Ed: Prohorov, Y. Sazonov, V.V.).
[66] Ikeda, N., Watanabe, S.: Stochastic Differential Equations and Diffusion
Processes, North Holland (1981) Amsterdam.
[67] Joume, J.L., Meyer, P.A.: Une martingale d'o¢rateurs bomes non repre-
sentable en integrale stochastique, Seminaire de Probabilites XX 1984/85,
Springer LNM 1204, 313-316 (1986). (Eds: J. Azema, M. Yor) Berlin.
[68] Joume, J.L.: Structure des cocycles Markoviens sur l'espace de Fock,
Probab. Th. ReI. Fields 75, 291-316 (1987).
[69] Kato, T.: Perturbation Theory for Linear Operators, 2nd edition, Springer
(1976) Berlin.
[70] Kraus, K.: General state changes in quantum theory, Ann. Phys. 64, 311-
335 (1971).
[71] Kiimmerer, B.: Markov dilations on W*-algebras, J. Funct. Anal. 63,
139-177 (1985).
[72] Kiimmerer, B., Maassen, H.: The essentially commutative dilations of
dynamical semigroups on M n , Commun. Math. Phys. 109, 1-22 (1987).
[73] Kiimmerer, B.: Survey on a theory of noncommutative stationary Markov
processes, Springer LNM 1303, 154-182 (1988) Berlin.
[74] Kunte, S.: The multinomial distribution, Dirichlet integrals and Bose-
Einstein Statistics, Sankhya, Ser. A 39, 305-308 (1977).
References 279
[75] Lindblad, G.: On the generators of quantum dynamical semigroups, Com-

mun. Math. Phys. 48, 119-130 (1976).
[76] Lindsay, I.M., Maassen, H.: An integral kernel approach to noise, Springer
LNM 1303, 192-208 (1988) Berlin.
[77] Lindsay, I. M., Maassen, H.: The stochastic calculus of Bose noise, King's
College preprint (1988).
[78] Lindsay, I.M., Parthasarathy, K.R.: The passage from random walk to
diffusion in quantum probability II, Sankhya, Ser A 50, 151-170 (1988).
[79] Lindsay, J.M., Parthasarathy, K.R.: Cohomology of power sets with ap-
plications in quantum probability, Commun. Math. Phys. 124, 337-364
(1989).
[80] Lindsay, I.M., Parthasarathy, K.R.: Rigidity of the Poisson convolution, in
Quantum Probability and Applications V, Springer LNM, 1442, 251-262
(1990).
[81] Lukacs, E.: Characteristic Functions, Griffin (1960) London.
[82] Lukacs, E.: Developments in Characteristic Function Theory, Griffin
(1983) London.
[83] Maassen, H.: Quantum Markov processes on Fock space described by
integral kernels, Springer LNM 1136, 361-374 (1985) Berlin.
[84] Mackey, G.W.: The Mathematical Foundations of Quantum Mechanics,
W.A. Benjamin (1963) New York.
[85] Mackey, G.W.: The Theory of Unitary Group Representations in Physics,
Probability and Number Theory, Benjamin-Cummings (1978) Reading,
Mass.
[86] Mackey, G.W.: Harmonic analysis as the exploitation of symmetry - a
historical survey, Bull. Amer. Math. Soc. (New Series) 3, 543-698 (1980).
[87] Maeda, S.: Probability measures on projections in von Neumann algebras,
Reviews in Math. Phys. 1, 235-290 (1990).
[88] Meyer, P.A.: Elements de probabilites quantiques, Seminaire de Proba-
bilites, Springer LNM 1204, 186-312 (1986); 1247,34-80 (1987); 1321,
101-128 (1988); 1372, 175-185 (1989).
[89] Meyer, P.A.: New Examples of chaotic representation, Strasbourg preprint
(1988).
[90] Meyer, P.A.: Equation de structure des martingales et probabilites quan-
tiques, Springer LNM 1372, 139-141 (1989).
[91] Meyer, P.A.: Chaines de Markov finies et representations chaotique, Stras-
bourg preprint (1989).
[92] Meyer, P.A.: Diffusions quantiques, I,I1,III, in: Seminaire de Probabilites
XXIV 1988/89, 370-396, Springer LNM 1426 (Eds.: I. Azema, P.A.
Meyer, M. Yor).
280 References
[93] Meyer, P.A.: Fock Spaces in Classical and Noncommutative Probability,

Chapters I-IV, Publication de l'IRMA, Strasbourg (1989).
[94] Mohari, A., Sinha, K.B.: Quantum stochastic flows with infinite degrees
of freedom and countable state Markov processes, Sankhya, Ser. A, Part
I, 43-57 (1990).
[95] Nelson, E.: Dynamical Theories of Brownian Motion, Princeton Univer-
sity Press (1967).
[96] Parthasarathy, K.R.: Probability Measures on Metric Spaces, Academic
Press (1967) New York.
[97] Parthasarathy, K.R.: Probability theory on the closed subspaces of a
Hilbert space, in: Les Probabilites sur les Structures Algebriques, CNRS
Colloquium No. 186,265-292 (1970).
[98] Parthasarathy, K.R., Schmidt, K.: Factorizable representations of current
groups and the Araki-Woods imbedding theorem, Acta Mathematica 128,
53-71 (1972).
[99] Parthasarathy, K.R., Schmidt, K.: Positive Definite Kernels, Continuous
Tensor Products and Central Limit Theorems of Probability Theory,
Springer LNM 272 (1972) Berlin.
[100] Parthasarathy, K.R., Schmidt, K.: A new method for constructing factor-
izable representations for current groups and current algebras, Commun.
Math. Phys. 50, 167-175 (1976).
[101] Parthasarathy, K.R.: Boson stochastic calculus, Pramana - J. Phys. 25,
457-465 (1985).
[102] Parthasarathy, K.R.: Some remarks on the integration of SchrOdinger equa-
tion using the quantum stochastic calculus, Springer LNM 1136, 409-419
(1985).
[103] Parthasarathy, K.R.: A remark on the paper "Une martingale d'operateurs
bornes non representable en integrale stochastique" by J.L. Journe and
P.A. Meyer, 317-320; Seminaire de Probabilites XX 1984/85, Springer
LNM 1204 (1986). Some additional remarks on Fock space stochastic
calculus, Ibid. 331-333 (1986).
[104] Parthasarathy, K.R., Sinha, K.B.: Stochastic integral representations of
bounded quantum martingales in Fock space, J. Funct. Anal. 67, 126-151
(1986).
[105] Parthasarathy, K.R., Sinha, K.B.: Boson-fermion relations in several di-
mensions, Pramana - J. Phys. 27, 105-116 (1986).
[106] Parthasarathy, K.R., Sinha, K.B.: Stop times in Fock space stochastic
calculus, Probab. Th. ReI. Fields, 75, 317-349 (1987).
[107] Parthasarathy, K.R.: The passage from random walk to diffusion in quan-
tum probability, in: Celebration Volume in Applied Probability, 231-245
(1988) Applied Probability Trust, Sheffield (Ed: J. Gani).
References 281
[108] Parthasarathy, K.R.: Local measures in Fock space stochastic calculus and
a generalized Ito-Tanaka formula, Springer LNM 1303,213-231 (1988).
[109] Parthasarathy, K.R.: A unified approach to classical, bosonic and fermionic
Brownian motions, Colloque Paul Levy sur les processus stochastiques,
Asterisque 157-158,303-320 (1988).
[110] Parthasarathy, K.R.: Azema martingales and quantum stochastic calculus,
Proc. R.C. Bose symposium, 551-569 (1990), Wiley Eastern, New Delhi
(Ed: R.R. Bahadur), to appear.
[111] Parthasarathy, K.R.: I.S.1. Lectures on Quantum Stochastic Calculus,
Mimeographed Lecture Notes (1988) New Delhi.
[112] Parthasarathy, K.R.: Discrete time quantum stochastic flows, Markov
chains and chaos expansions, Proc. Vth International Vilnius Conference
on Probability and Statistics (1989).
[113] Parthasarathy, K.R.: Quantum Ito's formula, Reviews in Math. Phys. 1,
89-112 (1989).
[114] Parthasarathy, K.R., Sinha, K.B.: Markov chains as Evans-Hudson diffu-
sions in Fock space, in: Seminaire de Probabilites XXIV 1988/89, 362-
369, Springer LNM 1426 (Eds: J. Azema, P.A. Meyer, M. Yor).
[115] Parthasarathy, K.R.: A generalized Biane Process, in: Seminaire de Prob-
abilites XXIV 1988/89 345-348, Springer LNM 1426 (Eds: J. Azema,
P.A. Meyer, M.Yor)
[116] Parthasarathy, K.R., Sinha, K.B.: Unification of quantum noise processes
in Fock spaces, Proc. Trento conference on Quantum Probability and Ap-
plications (1989), to appear.
[117] Reed, M., Simon, B.: Methods of Mathematical Physics Vol. I-IV Aca-
demic Press (1972) New York.
[118] Schatten, R.: Norm Ideals of Completely Continuous Operators, Springer
(1970) Berlin.
[119] Schiirmann, M.: Noncommutative stochastic processes with independent
and stationary increments satisfy stochastic differential equations, Probab.
Th. ReI. Fields, 84, 473-490 (1990).
[120] Segal, I.E.: Distributions in Hilbert space and canonical systems of oper-
ators, Trans. Amer. Math. Soc. 88, 12-41 (1954).
[121] Segal, I.E.: Tensor algebras over Hilbert spaces I, Trans. Amer. Math.
Soc. 81, 106-134 (1956).
[122] Segal, I.E.: Les problemes matMmatiques de la theorie quantiques des
champs, CNRS Paris, 57-103 (1959).
[123] Shale, D.: Linear symmetries of boson fields, Trans. Amer. Math. Soc.
103, 149-167 (1962).
[124] Simon, B.: Trace Ideals and their Applications, London Math. Soc. Lecture
Notes Series 35 (1979) Cambridge University Press.
282 References
[125] Speicher, R.: A new example of "independence" and "white noise", Prob-
abo Th. ReI. Fields 84, 141-159 (1990).
[126] Speicher, R.: Survey on stochastic integration on the full Fock space,
Heidelberg preprint (1990).
[127] Speicher, R.: Stochastic integration on the full Fock space with the help
of a kernel calculus, Heidelberg preprint (1990).
[128] Srinivas, M.D., Davies, E.B.: Optica Acta 28, 981-996 (1981).
[129] Srinivas, M.D.: Quantum theory of continuous measurements, Springer
LNM 1055, 356-363 (1984).
[130] Stinespring, W.F.: Positive functions on C*-algebras, Proc. Amer. Math.
Soc. 6, 211-216 (1955).
[131] Stone, M.H.: Linear Transformations in Hilbert space and their Applica-
tions to Analysis, Amer. Math. Soc. Colloq. Publications Vol. 15 (1951).
[132] Streater, R.F.: Current commutation relations, continuous tensor products
and infinitely divisible group representations, Rendiconti di sc. Inst. di
Fisica E. Fermi. Vol. XI, 247-263 (1969).
[133] Sunder, V.S.: An Invitation to von Neumann Algebras, Springer (1987)
Berlin.
[134] Trotter, H.F.: Eigenvalue distributions of large hermitian matrices,
Wigner's semicircle law and a theorem of Kac, Murdock and Szego,
Advances in Math. 54, 67-82 (1984).
[135] Varadarajan, V.S.: Geometry of Quantum Theory, Second Edition,
Springer (1985) Berlin.
[136] Voiculescu, D.: Symmetries of some reduced free product C* -algebras, in:
operator Algebras and their Connections with Topology and Ergodic The-
ory, Busteni, Romania 1983, Springer LNM 1132, (1985) (Ed: H. Araki).
[137] von Neumann, J.: Mathematical Foundations of Quantum Mechanics,
Princeton University Press (1955) Princeton (translated from German).
[138] von Waldenfels, W.: The Markov process of total spins, in: Seminaire de
Probabilites XXIV 1988/89, 357-361, Springer LNM 1426 (Eds:
J. Azema, P.A. Meyer, M.Yor).
[139] Weyl, H.: The Theory of Groups and Quantum Mechanics, Dover (1931)
New York (translated from German).
[140] Wigner, E.P.: Group Theory and its Applications to the Quantum Theory
of Atomic Spectra, Academic Press (1959) New York.
[141] Wigner, E.P.: Random matrices in Physics, SIAM Review 9, 1-23 (1967).
[142] Yosida, K.: Functional Analysis (4th edition) Springer (1974) Berlin.
Index
absolutely continuous 61 conjugation 5
acceleration of an observable 19 conservation operator 147
adapted 104 conservation process 182
- process 180 convergence 2
adjoint pair of processes 192 - almost everywhere ~ 55
algebra - in operator norm 3
- exterior 126 - in ~-measure 55
- graded 126 - strong 3
- symmetric 126 - weak 2,3
- tensor 126 core of a selfadjoint operator 64
- Z2-graded 126 countable Lebesgue spectrum 30
ampliation 101, 103, 256 countable tensor product 95
annihilation covariance 13
- operator 147 - matrix 13
- process 181 creation operator 147
antisymmetric Fock space 124 creation process 181
atom in rJP('Je) 7 cyclic vector 25
automorphism of rJP('Je) 82 density matrix 51
Bell's inequality 15, 17 direct sum of selfadjoint
Bose-Einstein statistics 109 operators 69
boson 106 double commutant theorem 68
boson conservation operator 152 Ehrenfest flow 120
boson Fock space 124 Ehrenfest's model 119
canonical anticommutation relations embedded Markov chain 260
(CAR) 175 energy operator 19, 88
canonical basis 3 equivalent processes 188
canonical commutation ergodic semi group 247
relations CCR 72 essential supremum with
canonical orthonormal basis 3 respect to ~ 53
canonical selfadjoint multiplication essentially bounded with
operator 70 respect to ~ 55
characteristic function 9, 78 Euclidean group 134
closed graph theorem 62 Evans-Hudson flow (EH flow) 233
cocycle identities 160 even element 126
- of first order 160 event 6
- of second order 160 - certain 6
coherent state 152 - complement of 6
coherent vector 124 - null 6
coisometry 4 events
completely positive map 251 - mutually noninterfering 6
completely real subspace 38 - simultaneous occur-
conditional expectation map 214 rence of 6, 101
284 Index
expectation 9 Heisenberg's uncertainty

exponential vector 124 principle 14, 80
Fermi-Dirac statistics 110 hypergeometric flow 119
fermion 106 hypergeometric model 119
- annihilation operator 175 imprimitivity relation 76
- creation operator 175 infinitely divisible
- Fock space 124 distribution 153, 161
finite particle vector 124 initial algebra 233
Fock Space 123, 124, 125 initial Hilbert space 179
- anti symmetric 124 integral with respect to ~ 54, 65
- symmetric 124 interfered Poisson field 220
Fourier irreducible family of operators 143
- transform 70 isometry 4
- transform operator 72 kernel 44,91
frame function
Lebesgue spectrum 30
- of weight W 31
Liouville's picture 88, 89
- regular 31
free martingale 104
- annihilation operator 152 - ~- 180
- conservation operator 152 Maxwell-Boltzmann
- creation operator 152 - Fock space 124
- Fock space 124 - statistics 109
fundamental process 183, 189 mean 79
Gelfand-Neumark- measurable process 188
Segal theorem 97 minimax principle 45
Gelfand pair 93 Mohari-Sinha regularity
generalized harmonic oscillator 20 condition 234
generates 2 moment 9
generator of 258 momentum observable 77
- one parameter unitary group 74 multiplier 160
- quantum dynamical n-particle
semigroup 258, 260 - vector, degree of 125
Gleason's theorem 39 - vector, rank of 125
Gorini-Kossakowski-Sudershan- - subspace 124
Lindblad theorem 257 - vector 124
Hadamard matrix 16 n-positive map 251
Hahn-Hellinger theorem 27 number operator 20,148
Hamiltonian 19, 88 observable
harmonic observable 21 - bounded 23
Heisenberg - canonical D-valued 23, 24
- commutation relations 72 - expectation of 9
- equation 19 - D-valued 23, 24
- picture 89, 214 - momentum 77
Index 285
- position 77 partial isometry 43

- support of 23 - initial space of 43
observables 9 Pauli spin matrices 10
- commuting 58 permanent 108
- direct sum of 23 phase-factor 2, 42
- interfering 9 Poisson random field 218
- mutually noninterfering 9 Polar decomposition theorem 43
- noninterfering 58 Polya flow 120
- unitarily equivalent 23 Polya's urn scheme 120
odd element 126 position observable 77
operator 60 positive definite kernel 91
- adjoint of 3, 60 probability
- antilinear 5 - amplitude 2, 52
- antiunitary 5 - distribution on '2J>('JC) 31
- bounded 3 - of an event in a state 7
- closable 60 - wave function 52
- closed 60 product of the quantum probability
- closure of 60 spaces 100
- compact 5 projection valued measure 23
- contraction 4 projective unitary
- densely defined 60 representation 135
- domain of 60 pure state 42, 51
- essentially selfadjoint 64 quantization 218
- extension of 60 - of pure birth process 218
- finite rank 5 - of pure death process 218
- graph of 60 - second 136
- Hilbert-Schmidt 52 quantum
- integral 44 - dynamical semigroup
- inverse of 5 (q.d.s.) 258, 260
- invertible 5, 60 - dynamical semigroup, uniformly
- modulus of 43 continuous 258
- normal 30 - flow 87
- null space of 4, 60 - Ito's formula 194
- positive 4 - probability space, finite
- range of 4, 60 dimensional 7
- rank of 4 - probability space, simple
- restriction of 25, 60 dimensional 7
- selfadjoint 4, 60, 64 - stationary 87
- singular values of 45 - stochastic differential
- symmetric 64 equation 207
- trace class 7, 46 - stochastic differential
- unitary 4 equation, with infinite
orbital angular momentum 77 degrees of freedom 221
286 Index
- stochastic flow 111, 115 - product 100

- translation invariant 87 - pure 8,42,51
regular frame function 31 - stationary 18
regular process 180 Stinespring's theorem 254
relative trace 102 stochastic integral 189
Rellich's theorem 68 stochastically integrable
resolvent process 189, 224
- identity 63 Stone generator 18, 74
- of an operator 63 Stone's theorem 56, 73
- set 63 - on the generator of a one parameter
p-conditional expectation 103 unitary 74
Stone-von Neumann theorem 79
Schatten 's theorem 51
structure equations 234
SchrOdinger
structure maps 234
- equation 18
sum-integral formula 151
- equation in the presence
symmetric measure space 132
of noise 214
symmetric Fock space 124
- picture 89
symplectic automorphism 162
second quantization 136
symplectic group 162
- differential 136
- homomorphism 150 tensor power, n-th 99
semicircle law 13 tensor product 93
Shale's theorem 169 - antisymmetric 106
simple adapted process 183 - countable 95
simple function with respect to ~ 53 - n-fold 93
singular values of an operator 45 - of Hilbert spaces 91ff.
spectral measure 23 - of operators 98, 99
- absolutely continuous 29 - of the observables 102
- pure point part 29 - of vectors 93
- singular continuous 29 - of n-th power 93
- support of 29 - symmetric 106
spectral total set 2, 24
- multiplicity 29 toy exponential vector 97
- theorem 8, 30 toy Fock space 97
- type 29 trace of an operator 7, 47
spectrum of an operator 63 transition amplitude 255
spin observable 16 transition probability 250
stabilizing sequence 95 truncated annihilation operator 12
*-derivation 19 truncated creation operator 12
*-unital unit ray 82
- algebra 112 unitarily equivalent 23, 70
- homomorphism 112 - normal operators 31
- map 112 - observables 23
state 7,51 - selfadjoint operators 70
Index 287
unitary group 8 - theorem 64

- k-parameter 59
unitary representation 59, 74 wave function 52
vacuum subspace 124 Wey1
vacuum vector 124 - commutation relations 76
variance 79 - operator 135
velocity map 22 - representation 161
velocity of an observable 19 Wigner distribution 13
von Neumann algebra 111 Wigner's theorem on Aut rzI'('lJC) 84
von Neumann's W* algebra 111
- spectral theorem 67
- theorem on double ~-martinga1e 180
commutant W 168 ~-null 53
Author Index
Accardi X, XI, 105, 123,207,257 Ito IX

Applebaum XI Journe 97,221
Araki IX, 161
Arveson 31 Karhunen 97
Kato 64,72
Bach 111, 257
Barchielli 221 Khinchine 97
Bargmann 88, 89 Kolmogorov 97, 161
Barnett X, 207 Kossakowski 257, 267, 273
Bell 15,17 Kraus 273
Berezin 152 Kummerer 123,249
Bhat 105, 123 Kunte 111
Bialynicki-Birula 82 Levy IX
Biane 123 Lewis 123
Bochner 57
Lindblad 257, 267, 273
Bose 106
Lindsay XI, 97, 134, 207, 257
Bratteli 152
Lupieri 221
Cecchini 105
Cook 134, 152 Maassen 134, 152, 207, 249
Mackey XI, 43, 82, 89, 105
Davies 221, 273
Maeda 43
Dirac 6, 89, 111
Meyer X, XI, 97, 123, 134, 207,
Doob IX
233, 249, 257
Ehrenfest 119
Mohari 233, 249
Evans 233, 249, 273
Fagnola X, 207, 221 Nelson 123
Fannes 161 Parthasarathy X, 43, 97, 123, 134,
Fermi 106 152, 161, 182,207,221,233,249,
Feynman 2, 6, 89 257
Fock 134 Polya 120
Frigerio 123, 249
Quaegebeur 161
Garding 152
Gleason 3943 Reed 64,72
Gnedenko 161 Rellich 72
Gorini 257,267,273 Robinson 152
Guerra 123 Rosen 123
Guichardet 132, 134 Schatten 51, 53
Halmos 31 Schmidt IX, 97, 161
Holevo 161 Schiirmann 233
Hudson X, XI, 182,207,221,233, Segal 134, 152
249 Shale 169
Ikeda IX Simon 53, 64, 72, 123
290 Author Index
Sinha XI, 152, 207, 233, 249 von Neumann 58, 64, 67, 72, 79,
Speicher 13, 152 89, 168
Srinivas 221 von Waldenfels X, 123
Stinespring 254, 257 Watanabe IX
Stone 56, 59, 82 Weyl 89
Streater IX-XI, 161, 207 Wiener IX, 97
Sudarshan 257, 267, 273 Wightman 152
Trotter 13 Wigner 13, 82, 84, 89
Varadarajan XI, 8, 17,59, 82, 89 Wilde X,207
Voiculescu 13 Yosida 59,82
Monographs in Mathematics
The foundations of this outstanding book series were laid in 1944. Until the end of the 1970s, a total of 77
volumes appeared, including works of such distinguished mathematicians as Caratheodory, Nevanlinna,
and Shafarevich, to name a few. The series came to its present name and appearance in the 1980s.
According to its well-established tradition, only monographs of excellent quality will be published in this
collection. Comprehensive, in-depth treatments of areas of current interest are pre',ented to a readership
ranging from graduate students to professional mathematicians. Concrete examples and applications both
within and beyond the immediate domain of mathematics illustrate the import and consequences of the
theory under discussion.
We encourage preparation of manuscripts in TeX for delivery in camera-ready copy which leads to rapid
publication, or in electronic form for interfacing with laser printers or typesetters. Proposals should be sent
directly to the editors or to: Birkhauser Verlag, P.O. Box 133, CH-4010 Basel, Switzerland.
Published in the series Monographs in Mathematics since 1983:
Volume 78: Hans Triebel, Theory of Function Spaces.

1983,284 pages, hardcover, ISBN 3-7643-1381-1.
Volume 79: Gennadi M. HenkinlJiirgen Leiterer, Theory of Functions on

Complex Manifolds.
1984,228 pages, hardcover, ISBN 3-7643-1477-X.
Volume 80: Enrico Giusti, Minimal Surfaces and Functions of Bounded

Variation.
Volume 81: Robert J. Zimmer, Ergodic Theory.

Volume 82: V. I. Arnold/S. M. Gusein-ZadelA. N. Varchenko, Singularities of

Differentiable Maps - Volume I.
Volume 83: V. I. Arnold/S. M. Gusein-ZadelA. N. Varchenko, Singularities of

Differentiable Maps - Volume II.
Volume 84: Hans Triebel, Theory of Function Spaces II.

Volume 85: K.R. Parthasarathy, An Introduction to Quantum Stochastic

Calculus.
Basler Lehrbiicher: A Series of Advanced Textbooks in Mathematics
Managing Editors:
H. Amann (Universitat Ziirich)

H. Kraft (Universitiit Basel)
This series presents, at an advanced level, introductions to some of the fields of current interest in
mathematics. Starting with basic concepts, fundamental results and techniques are covered, and important
applications and new developments discussed. The textbooks are suitable as an introduction for students
and non-specialists, and they can also be used as background material for advanced courses and seminars.
We encourage preparation of manuscripts in TeX for delivery in camera-ready copy which leads to rapid
publication, or in electronic form for interfacing with laser printers or typesetters. Proposals should be sent
directly to the editors or to: Birkhauser Verlag, P.O. Box 133, CH-4010 Basel, Switzerland.
Published in the series Basler Lehrbiicher:
Volume I: Markus Brodmann, Algebraische Geometrie (german).

Volume 2: Ernest B. Vinberg, Linear Representations of Groups.

1989, 146 pages, hardcover, ISBN 3-7643-2288-8.
Volume 3: Konrad Jacobs, Discrete Stochastics.

(Monographs in Mathematics 85) K. R. Parthasarathy (Auth.) - An Introduction To Quantum Stochastic Calculus-Birkhäuser Basel (1992)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Monographs in Mathematics 85) K. R. Parthasarathy (Auth.) - An Introduction To Quantum Stochastic Calculus-Birkhäuser Basel (1992)

Uploaded by

Copyright:

Available Formats

Monographs in Mathematics

Library of Congress Cataloging-in-Publication Data

Deutsche Bibliothek Cataloging-in-Publication Data

© 1992 Springer Basel AG

Index .................................................................... 283

Author Index ............................................................ 289

1 From classical to quantum probability

(f,g) = J fgdl-" = J f(x)g(x)l-"(dx), I,g E L2(1-").

If S is a topological space we shall denote by CfF s the Borel a-algebra generated

J11/(x)ll~ I-"(dx) < oo}

both an isometry and a coisometry. If T is an isometry then TT* is a projection.

(vi) For any T in 0A(~), Tlu)(vl = ITu)(vl, lu)(vIT = lu)(T*vl;

Proof: All the properties (i)-(ix) are straightforward from definitions. •

3 Finite dimensional quantum probability spaces

of projections, or equivalently, a family of mutually non-interfering events. Then

4 Observables in a simple quantum probability space

0"0= (01 0)1 ,0"]= (01 01) ,0"2= (0i -i)

where t] + it2 = re ill denotes polar decomposition. In particular,

Example 4.2: Let 'Je = en

Exercise 4.3: Let X = (p - q)O'3 - 2..jPQO'I in the notations of Example 4.1,

Ck(O) = 0 if k is odd, Ck(O) = ~2~\ if k = 2m :::; n - 1.

In particular, the first n - 1 moments of the observable X = !(L + L*) in the

It is to be noted that the first n - 1 moments determine the distribution of X

5 Variance and covariance

varu(X) varu(Y) 2': ~I(u, i[X, YjU)12. (5.5)

varp(X) = 2::Pill(X - m)uil12

where 1m denotes imaginary part. By Schwarz's inequality

~I(u, i[X, Yju)1 :::: II(X - a)ullll(Y - b)ull

Proposition 5.1 reveals one of the most characteristic features of observables

Thus any decrease in the variance of X or Y has to be compensated for by an

It is to be noted that for observables X, Y whose covariance in a state p is

Proposition 5.5: Let C = (( Cij )) be a positive definite matrix with Cii = 1

c~ (:~~) (1,I/h,LH,I/h)+ (r 14: H· 0)

{ (~~~:0 ) ,( ±~os~: 0 ), 0::; 0 < 1r } .

(See Exercise 13.10.)

Exercise 5.8: Let p be a state of rank k in a Hilbert space of dimension n.

6 Dynamics in finite dimensional quantum probability spaces

dPt = -z'[H , Pt,1

If X is any observable and the state at time 0 is p then the expectation of X at

Equation (4.1) enables us to present a dual picture of the dynamics described

This is called the Heisenberg equation for observables. If H commutes with X

£(XY) = £(X)Y + X£(Y), £(X*) = £(X)*. (6.4)

Any linear map £ on 073('3£) satisfying (6.4) is called a *-derivation. It is a theorem

tr pXt = tr pL;jEjX Ej + 2 Re L tr pEjX Ekeit(Aj -Ak).

Example 6.1: Let aj,j = 0, 1,2,3 be as in (4.2). Then (4.3) implies

(iv) If X t = ei t/2Ujake-i t/2Uj, 1 ::; j, k ::; 3, j -::J. k and p = (1~2 1~2)

deserves its name). We have the relations:

N = LL*kLk, L*L = l-!eo)(eo!, LL* = l-!e n -l)(e n -l!

A = I:iaii!ei)(ei! + L {I:j aii+k!ei)(ei+k! +L ai+ki!ei+k)(ei!} (6.8)

and defining the functions f, gk, hk on {O, 1,2, ... } by

f(i) = aii,gk(i) = aii+k,hk(i) = ai+ki

+ L {e-itkgk(N)Lk + eitk L*k hk(N)}.

Suppose A is Hermitian. Then f is real, 11k = h k. Putting Ao = f(N), Ak

where Ak is an observable, [N, [N, A k]] = k 2 Ak and for any function ¢ on

Exercise 6.3: Let Z = VN + lL + L*VN + 1. Suppose

7 Observables with infinite number of values

Then for any sequence {Fj } of disjoint subsets of IR

~(UjFj) = L ~(Fj). (7.1 )

In other words the correspondence F --* ~(F) is an "event valued measure" on

which is parallel to (7.1). In other words F --* If-l(F) can be viewed as an

~(UjFj) = L~(Fj) if Fi n Fj = ¢ for i i- j (7.2)

~(E)u = ffij~j(E)uj if U = ffijuj E ffij'Je j , E E '2F.

IE denoting the indicator of E. Then ell is an observable called the canonical

£(XY) = £(X)Y + X£(Y), £(X) = £(X). (6.4)

N = LLkLk, LL = l-!eo)(eo!, LL* = l-!e n -l)(e n -l!