Dilations of Irreversible Evolutions in Algebraic Quantum Theory

Dilation of Irreversible Evolutions in
Algebraic Quantum Theory

D.E. Evans & J.T. Lewis
Preface
The evolution of a Hamiltonian system is reversible. The evolution of a real system is
not: it always returns to a state of thermal equilibrium at a temperature determined by its
surroundings. The explanation of this phenomenon is the fundamental problem of statistical
mechanics. Beginning around 1900 with the work of Boltzmann and Gibbs, herculean efforts
have been made to solve this in the context of the classical mechanics of systems with a
finite number of degrees of freedom. The main problem remains open, but some beautiful
theorems have been discovered: a new branch of mathematics, Ergodic Theory, has arisen.1
More recently, there has been intense activity in the context of the quantum mechanics of
systems with an infinite number of degrees of freedom. Again, the harvest, so far, has largely
been mathematical. One line of development can be traced to the seminal paper of Ford,
Kac and Mazur [FKM65]; in particular, this paper was studied in 1970-71 by an Oxford
seminar run by one of us (JTL) in collaboration with E.B. Davies. Both of us owe Brian
Davies a debt of gratitude for what we have learned from him. These notes arose from
a Dublin seminar which in 1975-76 studied one of his papers [Dav72a], and we thank G.
Parravicini, J.H. Ranwnsley and W.G. Sullivan for many stimulating discussions during this
period. We have attempted to present a self-contained account of the mathematical results
which are necessary for work on this field. We do not claim to give a complete catalogue
of results; for reviews of the literature see [GFV` 78] and [Dav77a]. The first draft was
written in Dublin in 1975-76. The second draft was completed in 1976-77 by one of us (DEE)
while in Oslo; he is grateful to Erling Strmer and his colleagues for their warm hospitality
and the stimulating atmosphere of their group.
It is a pleasure to thank Mrs. Eithne Maguire whose patience and skill in typing have
produced the camera-ready copy; and Miss Evelyn Wills, the technical editor of this series,
whose professional expertise we have relied on in preparing the manuscript. Needles to say,
those imperfections which remain are attributed solely to the authors.
D.E. Evans
J.T. Lewis
Dublin 9. 11. 77.
For a description of the present situation in an historical context, see [LP73].

iii
Contents
Preface
iii
Contents
iv
Introduction
vi
Chapter 0. Preliminaries
0.1. Banach spaces and one-parameter semigroups
0.2. Banach *-algebras and C*-algebras
0.3. W*-algebras
0.4. Order
0.5. Tensor product
1
1
2
3
3
4
Chapter 1. Positive-definite kernels
Chapter 2. Positive-definite functions
10
Chapter 3. Dilations of Semigroup of Contractions
14
Chapter 4. C -algebras and Positivity
21
Chapter 5. Conditional Expectations
25
Chapter 6. Fock Space
27
Chapter 7. Representation of the Canonical Commutation Relations
30
Chapter 8. Representation of the Canonical Anti-Commutation Relations
34
Chapter 9. Slawnys Theorem
37
Chapter 10. Completely Positive Maps on the CCR Algebra
40
Chapter 11. Completely Positive Maps on the CAR Algebra
45
Chapter 12. Dilations of Quasi-free Dynamical Semi-groups
47
Chapter 13. Dilation of Completely Positive Maps on C -algebras
49
Chapter 14. Generators of Dynamical Semigroups
50
Chapter 15. Canonical Decomposition of Conditional Completely Positive Maps
55
Chapter 16. Isometric Representation of Quantum Dynamical Semigroups
58
Chapter 17. Unitary Dilations of Dynamical Semigroups
61
iv
CONTENTS
References
66
Bibliography
70
Introduction
The purpose of these notes is to consider the problem of whether irreversible evolutions of
a quantum system can be obtained as restriction of reversible dynamics in some larger system.
In the classical theory of Markov processes, the Fokker-Planck semigroup tTt : t 0u can be
factored as Tt N Ut j, t 0, where j is an embedding, Ut is a group of automorphisms,
and N is a conditional expectation. Is such a factorization of an irreversible evolution possible
in algebraic quantum theory? In particular, we consider the mathematical formulation of
this question in Hilbert space and C*-algebra settings.
Positivity is a central theme in any probability theory; the theory of non-commutative
stochastic processes is no exception. In the first section we give a brief account of the theory
of reproducing kernel Hilbert spaces. This allows us to give a short, unified treatment of
various well-known dilation theorems, such as the Naimark- Sz.-Nagy unitary dilation of
positive-definite functions on groups, the GNS-Stinespring construction for C*-algebras, and
related Schwarz-type inequalities, the construction of Fock space, and the algebras of the
canonical commutation and anticommutation relations.
In our first attempt to construct reversible dynamics from irreversible system we consider,
in Chapter 3, the category whose objects are Hilbert spaces and whose morphisms are contractions. Here we show how one can dilate certain families of morphisms to automorphisms
(unitary operators). As shown by [LT74a, LT75, LT74b], this is the mechanism behind
the construction of the FKM-model [FKM65]. This Hilbert space theory is then lifted to
a C*-algebra setting using the algebras of the canonical commutation and anticommutation
relations. We are thus led naturally to the C*-algebraic setting of quantum theory, where
the bounded observables of the system are represented by the self-adjoint elements of the
algebra, and the states by positive linear functionals.
From this point on, we concern ourselves with the category whose objects are C*-algebras,
and whose morphisms are completely positive contractions. Complete positivity is a property
whose study may be motivated both by mathematical and physical arguments. It is a much
stronger property than positivity. However, for commutative C*-algebras the concept of
complete positivity and positivity coincides; for this reason the distinction does not arise in
classical probability theory. It follows from the Schwarz inequality for completely positive
maps that a morphism which has an inverse which is also a morphism is in fact an algebraic
*-isomorphism (and hence merits the name isomorphism). This is not so if one has mere
positivity. Completely positive maps have an interesting physical interpretation [Kra71,
Lin76c]. They arise physically with the study of operations on systems in interaction.
We adopt the view that reversible behaviour is described by a one-parameter group of *automorphisms on a C*-algebra, and irreversible Markovian behaviour is described by a
semigroup of completely positive maps [Lin76c].
In Chapter 5 we are concerned with the mathematical formulation of the embedding
of a quantum mechanical system in a larger one, and the dual operation of restriction to
vi
INTRODUCTION
vii
a subsystem. We thus require a non-commutative analogue of the conditional expectation

of classical probability theory; this will be an injection of the states of the system into
those of a larger system (the Schrodinger picture), or the dual operation of averaging or
projection of the observables of a large system onto those of a subsystem (the Heisenberg
picture). Thus we seek a projection N from a unital C*-algebra A onto a unital C*-algebra
B. Since in the dual picture normalized states of B must go into normalized states of A,
we require that N be positive and that it take 1A to 1B . It is shown that such a map is
automatically completely positive. This observation provides us with a second argument for
taking irreversible evolutions to be described by completely positive maps: the restriction
to a subsystem of a reversible evolution is necessarily completely positive.
An abstract dilation theorem for completely positive maps is obtained for C*-algebras
in Chapter 13. For the remainder of this work, we concentrate on W*-algebras and norm
continuous semigroups of completely positive normal maps. A study of generators of such
semigroups in Chapter 14 and Chapter 15 leads to a unitary dilation in Chapter 17, via the
isometric representation of Chapter 16.
We have not given any account of approximate dilations involving a limiting process such
as the weak coupling and the singular coupling limits. We recommend the excellent review
by Gorini et al. [GFV` 78] and Davies [Dav77a].
CHAPTER 0
Preliminaries
We give here a brief summary of the prerequisites for the main text, and establish some
notation. We assume that the reader is familiar with the fundamental elements of functional
analysis on Banach spaces, in particular with the theory of Hilbert spaces and algebras of
operators on Hilbert spaces, such as can be found in Dunford & Schwartz [DS58, DS63],
Reed & Simon [RS72, RS75], Yosida [Yos71], Dixmier [Dix81, Dix77], and Sakai [Sak71].
We work throughout with vector spaces over the complex field, although much of the work
with the CAR and CCR algebras is valid on real spaces.
0.1. Banach spaces and one-parameter semigroups
If X and Y are Banach spaces, BpX, Y q denotes the Banach space of all bounded linear
operators from X into Y . We write X for BpX, Cq and BpXq for BpX, Xq. A contraction
T from X to Y is an element of BpX, Y q s.t. }T } 1; if }T x} }x} for all x P X, then T
is called an isometry.
If X is a Banach space, a one-parameter semigroup tTt : t 0u is a map T : R` BpXq
s.t. T0 1 and Ts Tt Ts`t for all s, t P R` ; the semigroup is said to be strongly continuous
if the maps t Tt pxq are norm continuous for each x P X; or equivalently if t hTt pxq, f i
is continuous at zero for all x P X, and all f P X (see [DS63] p.616, [Yos71] p.233). In
this case, there exists a closed densely defined linear operator L s.t.
Tt x x
Lx lim
t0
t
on the domain DpLq, and DpLq is precisely the set of x P X for which this limit exists in
the norm topology (see [DS63] p.620, [Yos71] p.239,241). The operator L is called the
generator of the semigroup. The domain of L is globally invariant under the semigroup;
moreover,
d
Tt x LTt x Tt Lx
dt
tL
for all x P DpLq (see [DS63] p.619, [Yos71] p.239).
tL Thus wet write the formal symbol e
for Tt . There exist positive numbers M and s.t. e M e for all t 0; for all complex
with Re we then have that lies in pLq, the resolvent set of L, and
8
1
p Lq
etL et dt
0
(see [DS63] p.619,622, [Yos71] p.232,240). Conversely,
n
tL
tL
e lim 1
n8
n
gives the semigroup in terms of the resolvent of the generator (see [HP57], p.352). Moreover,
etL is a contraction semigroup iff the following equivalent conditions hold:
1
0.2. BANACH *-ALGEBRAS AND C*-ALGEBRAS
a) For all x P DpLq, there exists f P X with }f } 1, f pxq }x}, and Rehf, Lxi 0.
b) For all 0 and x P DpLq, we have
}x} }p Lqx} .
(see [DS63] p.626, [Yos71] p.248, [LP61]). The semigroup etL is norm continuous iff L is in
BpXq (see [DS63] p.621); in this case etL can be given by the usual power series expansion
8
ptLqn
tL
e
.
n!
n0
If L is bounded, etL is a contraction semigroup iff
*
"
}1 ` tL} 1
}1 ` tL} 1
: t 0 lim
0
inf
t0
t
t
(see [LP61]).
If L generates a strongly continuous one-parameter semigroup etL , and Z is a bounded
operator on X, then L ` Z generates a strongly continuous one-parameter semigroup etpL`Zq
which satisfies
t
tpL`Zq
tL
e
pxq e pxq ` eptsqL ZespL`Zq pxqds
0
for t 0 and x P X (see [DS63] p.631, [Kat95] p.495). The perturbed semigroup is also
given by the Lie-Trotter product formula
tL tZ n
tpL`Zq
e
pxq lim e n e n
pxq, t 0,
n8
for all x P X (see [Tro59], [Che74]).

0.2. Banach *-algebras and C*-algebras
A Banach algebra A is a complete normed algebra with }xy} }x} }y} for all x, y P A.
If A possesses an identity, written 1A or 1, we require }1} 1; in this case A is said to
be unital. An approximate identity for a Banach algebra A is a net tu : P u in A s.t.
}u } 1 for all , and s.t. for each x P A we have xu x and u x x in the norm
topology as 8. A *-algebra A (also called an algebra with involution) is an algebra
equipped with a conjugate-linear idempotent antiautomorphism x x . An element x in a
*-algebra A is said to be self-adjoin (or hermitian) if x x ; the set of self-adjoint elements
of A is denoted by Ah . Each element x in A has a unique decomposition x x1 ` ix2 with
x1 and x2 in Ah . A linear map T between *-algebras A and B is said to be self-adjoint
if T pAh q Bh , or equivalently if T px q T pxq for all x P A. An element x in a unital
*-algebra is said to be isometric if x x 1, and unitary if both x and x are isometric.
A Banach *-algebra is a Banach algebra with an isometric involution x x ; e.g., if G is
a locally compact group, then L1 pGq with the usual operations is a Banach *-algebra with
approximate identity ([Loo53]).
A C*-algebra A is a Banach *-algebra s.t. }x x} }x}2 for all x P A. If A is a
Banach *-algebra, then the algebra A obtained from A by adjoining an identity is a Banach
algebra containing A as a Banach subalgebra; moreover, if A is a C*-algebra, then so is A
([Sak71] 1.1.7). Every C*-algebra has an approximate identity (see [Dix77] 1.7.2). If T
is a bounded linear map from a C*-algebra A into a Banach space, then }T } supt}T x} :
0.4. ORDER
x unitary in Au, because A is the norm-closed convex hull of its unitaries (see [RD` 66]).
If is a *-homomorphism from a C*-algebra A into another C*-algebra B, then is a
contraction and pAq is norm closed in B; if is faithful it is an isometry (see [Dix77]
1.3.7, [Sak71] 1.2.6, 1.17.4). A norm-closed *-subalgebra of a C*-algebra A is a C*algebra, and is said to be a C*-subalgebra of A. For any Hilbert space H, the algebra BpHq
is a C*-algebra, and its C*-subalgebras are known as C*-algebras on H, or concrete C*algebras. A *-representation of a *-algebra A on a Hilbert space H is a *-homomorphism
from A into BpHq. The Gelfand-Naimark-Segal representation theorem says that every C*algebra has a faithful representation as a concrete C*-algebra on a Hilbert space (see [Dix77]
2.6.1, [Sak71] 1.16.6). If X is a locally compact Hausdorff space, then C0 pXq (the space
of continuous functions which vanish at infinity, equipped with the supremum norm) is a
commutative C*-algebra. Conversely, every commutative C*-algebra is isomorphic to some
C0 pXq (see [Dix77] 1.4.1, [Sak71] 1.2.1, 1.2.2).
0.3. W*-algebras
A W*-algebra A is a C*-algebra which is a dual Banach space (that is, there exists
a Banach space F s.t. A F ). In this case F is uniquely determined up to isometric
isomorphism, and is called the pre-dual of A, written A (see [Sak71] 1.13.3). The weak
*-topology pA, A q is also known as the ultraweak, or -weak (operator), topology. Every
W*-algebra has an identity (see [Sak71] 1.7). If A is a W*-algebra and B is a pA, A qclosed *-subalgebra of A, then B is a W*-algebra with pre-dual A {B0 ; here B0 is the
annihilator of B in A (see [Sak71] 1.1.4). Then B is said to be a W*-subalgebra of A.
The prefix W*- applied, for example, to a homomorphism means a weak *-continuous
homomorphism. Thus a W*-homomorphism from a W*-algebra A into a W*-algebra B is
a weak *-continuous, and in this case pAq is a W*-subalgebra of B (see [Sak71] 1.16.2).
When H is a Hilbert space, BpHq is a W*-algebra; the pre-dual of BpHq can be identified
with the Banach space T pHq of all trace-class operators on H, under the pairing h, xi
trpxq of P T pHq and x P BpHq (see [Sak71] 1.15.3). The W*-subalgebras of BpHq are
also called W*-algebras of H. Consider a W*-algebra A on a Hilbert space H. If A contains
the identity of BpHq, we say that A is a von Neumann algebra on H. In general, the identity
1A of A is merely a projection on H; but A can be viewed also as a von Neumann algebra
on 1A H. If H is a Hilbert space and X a subset of BpHq, then the commutant X 1 of X is
defined as X 1 ty P BpHq : xy yx, @x P Xu. If A is a *-subalgebra of BpHq containing
the identity of BpHq, then A is a von Neumann algebra iff A A2 (see [Dix81] p.42,
[Sak71] 1.20.3). Sakais representation theorem says that every W*-algebra has a faithful
W*-representation as a von Neumann algebra on a Hilbert space (see [Sak71] 1.16.7).
If A is a C*-algebra, then A is a W*-algebra, and can be identified with the von
Neumann algebra generated by A in its universal representation (see [Sak71] 1.17.2). If T
is a bounded linear map from a C*-algebra A into a C*-algebra B, then T can be uniquely
extended to an ultraweakly continuous map from A into B ; if B is in fact a W*-algebra
then T can be uniquely extended to an ultraweakly continuous map from A into B (see
[Sak71] 1.21.13).
0.4. Order
A (partial) ordering of a set is a reflexive, transitive relation, denoted by . If V is
a vector space (over the complex field, as usual), a wedge P in V is a subset satisfying
0.5. TENSOR PRODUCT
P ` P P and R` P P . An ordered vector space is a vector space V equipped with a

wedge V ` ; the elements of V which are in V ` are said to be positive. The wedge V ` of
positive elements induces an ordering in V : for x, y P V , x y if x y P V ` . A linear
map T between ordered vector spaces V and W is said to be positive
if T pV ` q W ` . If
`
A is a *-algebra, we introduce the wedge A of all finite sums a a with a P A; we note
that A` Ah . If A is a C*-algebra, then A` ta a : a P Au and A` is a cone (that is,
A` X pA` q t0u); each element x P Ah has a unique decomposition x x` x with x`
and x P A` and x` x 0 (see [Sak71] 1.4). A linear map T between *-algebras A and
B is positive iff T pa aq 0 for all a P A. Any positive linear map from a Banach *-algebra
with approximate identity into a C*-algebra is automatically continuous ([Sin76] 13.11).
Moreover, if A and B are unital C*-algebras, then a bounded linear map T from A into B,
satisfying T p1A q 1B , is positive iff T is of norm one (see [RD` 66])
If A is a C*-algebra, we use the notation x x to mean that tx : P u is a net of
self-adjoint elements of A, filtering upwards, with least upper bound x. Then a positive map
T between C*-algebras A and B is said to be normal if x x in A implies T x T x in
B. A positive map between W*-algebras is normal iff it is weak *-continuous (see [Sak71]
1.7.4, 1.13.2).
0.5. Tensor product
If A and B are Banach spaces, we denote their algebraic tensor product by A d B. Comq denotes the projective tensor product (see [Gro55]);
pletions are denoted as follows: AbB
if A and B are Hilbert spaces, A b B denotes the Hilbert space tensor product ([RS72]).
If p, q is a measure space, and H is a Hilbert space, we let L2 p; Hq denote the space of
(equivalence classes of) functions f : H satisfying:
a) hf pq, xi is measurable for all x P H,
b) there is a separable subspace H0 in H s.t. f pq lies in H0 for almost every ,
c) }f pq} is in L2 pq.
Then L2 p, q is a Hilbert space when equipped with the inner product
hf, gi hf pq, gpqi dpq.
The map f b x f pqx extends uniquely to a unitary map of L2 pq b H onto L2 p; Hq (see

[VD78]).
We define the C*- and W*-tensor products for concrete algebras as follows: Let A, B be
C*-algebras on Hilbert spaces H,K; the C*-tensor product A b B is the C*-algebra on H b K
generated by A d B. If A and B are W*-algebras, the W*-tensor product AbB is the W*algebra on H b K generated by A d B. For abstract algebras we take representations, since
the definitions of C*- and W*-algebras which we have given are representation-independent
(see [Sak71] 1.22.9, 1.22.11).
Let p, q be a localizable measure space (that is, direct sum of finite measure spaces);
then L8 pq, the space of all essentially bounded locally -measurable functions, is a commutative W*-algebra, whose pre-dual can be naturally identified with L1 pq. Conversely, every
commutative W*-algebra is *-isomorphic to L8 pq, for some p, q (see [Sak71] 1.18). The
map : L8 pq BpL2 pqq given by
rpf qgspq f pqgpq
0.5. TENSOR PRODUCT
is a faithful *-representation of L8 pq as a maximal abelian von Neumann algebra on L2 pq

(see [Sak71] 2.9.3). let M be a W*-algebra with separable pre-dual. Then L8 p; M q,
the space of all M -valued, essentially bounded, weak *--locally measurable functions on
, is a W*-algebra with pre-dual L1 p; M q, the Banach space of all M -valued Bochner
-integrable functions on . Moreover, the mapping f b a f pqa extends uniquely to a
W*-isomorphism of the W*-algebra L8 pqbM onto L8 p; M q. Under this identification,
q of
the pre-dual L1 p; M q of L8 p; M q is naturally identified with the pre-dual L1 pqbM
8
L pqbM (see [Sak71] 1.22.13).
CHAPTER 1
Positive-definite kernels
Throughout this chapter X denotes a set and H a Hilbert space; a map K : X X
BpHq is called a kernel and the set of such kernels is a vector space denoted by KpX; Hq.
Definition 1.1. A kernel K in KpX; Hq is said to be positive-definite if, for each positive
n and each choice of vectors h1 , . . . , hn in H and elements x1 , . . . , xn in X, the inequality
hKpxi , xj qhj , hi i 0
(1.1)
i,j
holds.
Example 1.2. Let H 1 be a Hilbert space, let V be a map from X into BpH, H 1 q, and
put
Kpx, yq V pxq V pyq;
(1.2)
then
i,j
hKpxi , xj qhj , hi i V pxj qhj 0,

j
so that K is positive-definite.
The principal result of this chapter is that a kernel K is positive-definite iff it can be
expressed in the form (1.2).
Definition 1.3. Let K be a kernel in KpX; Hq. Let HV be a Hilbert space and V :
X BpH, HV q a map s.t. Kpx, yq V pxq

V pyq for all x, y P X. Then V is said to be a
Kolmogorov decomposition of K; if HV tV pxqh : x P X, h P Hu then V is said to be
minimal. Two Kolmogorov decompositions V and V 1 are said to be equivalent if there is a
unitary mapping U : HV HV 1 , s.t. V 1 pxq U V pxq for all x P X.
A minimal Kolmogorov decomposition is universal in the following sense:
Lemma 1.4. Let K be in KpX; Hq and let V be a minimal Kolmogorov decomposition of
K. Then to each Kolmogorov decomposition V 1 of K there corresponds a unique isometry
W : HV HV 1 , s.t. V 1 pxq W V pxq for all x P X. Moreover, if V 1 is minimal then W is
unitary.
Proof. Since V is minimal the set of elements of the form j V pxj qhj is dense in HV .

1
The map W
V
px
qh
j
j
j
j V pxj qhj is well-defined and isometric since
hV pyqk, V pxqhi hKpx, yqk, hi hV 1 pyqk, V 1 pxqhi ,
and hence it extends by continuity to an isometry W : HV HV 1 . The rest is routine.
6
1. POSITIVE-DEFINITE KERNELS
We have yet to show the existence of a Kolmogorov decomposition for an arbitrary

positive-definite kernel; we remedy this by constructing a decomposition canonically associated with the kernel. We employ a Hilbert space of H-valued functions spanned by those of
the form x Kpx, yqh, using the positivity of K to get an inner product. For this purpose
it is convenient to reformulate Definition 1.1. But first we need another definition:
Definition 1.5. Let F0 F0 pX; Hq denote the vector space of H-valued functions on
X having finite support; let F F pX; Hq denote the vector space of all H-valued functions
on X. We identify F with a subspace of the algebraic anti-dual F01 of F0 by defining the
pairing p, f pp, f q of F and F0 by
pp, f q
hppxq, f pxqi .
xPX
(Since f has finite support only finite number of terms in the sum are non-zero.) Given K
in KpX; Hq we define the associated convolution operator K : F0 pX; Hq F pX; Hq by
pKf qpxq
Kpx, yqf pyq.
yPX
Then Definition 1.1 may be reformulated as:

Definition 1.6. The kernel K in KpX; Hq is positive-definite iff the associated convolution operator K : F0 pX; Hq F pX; Hq is positive:
pKf, f q 0 @f P F0 pX; Hq.
Next we need a vector space result:
Lemma 1.7. Let V be a complex vector space, and let V 1 be its algebraic anti-dual, with
the pairing V 1 V C written v 1 , v pv 1 , vq. Let A : V V 1 be a linear mapping s.t.
pAv, vq 0 for all v P V . Then there is a well-defined inner-product on the image-space AV
given by
hAv1 , Av2 i pAv1 , v2 q.
Proof. The sesquilinear form v1 , v2 apv1 , v2 q pAv1 , v2 q is positive, so that the
Schwarz inequality holds:
|apv1 , v2 q|2 apv1 , v1 qapv2 , v2 q.
It follows that the set VA tv P V : apv, vq 0u coincides with the subspace ker A, and
so the natural projection : V V { ker A carries the form ap, q into an inner-product
h, iA on V { ker A given by hpv1 q, pv2 qiA apv1 , v2 q. The vector space isomorphism A1 :
V { ker A AV given by A1 A carries the inner-product h, iA into an inner-product
h, i on AV , given by
hAv1 , Av2 i hA1 pv1 q, A1 pv2 qi hpv1 q, pv2 qiA
apv1 , v2 q pAv1 , v2 q.

Theorem 1.8. For each positive-definite K in KpX; Hq there exists a unique Hilbert
space RpKq of H-valued functions on X s.t.
(a) RpKq tKp, xqh : x P X, h P Hu,

(b) hf pxq, hi hf, Kp, xqhi for all f P RpKq, x P X and h P H.
Proof. Since the kernel K is positive-definite the associated convolution operator K

of F0 F0 pX; Hq into F 1 defined in 1.5 satisfies the hypothesis of Lemma 1.7. Let KF0
be the abstract completion of KF with respect to the norm got from the inner-product
hKf1 , Kf2 i pKf1 , f2 q, and identify KF0 with a dense subset of KF0 . For each x P X and
h P H define the function hx in F0 by putting hx pyq h if y x and hx pyq 0 otherwise;
then pKhx qpyq Kpy, xqh. Define Kx on H by Kx h Khx for all x P X and h P H;
1
then }Kx h} }Kpx, xq} 2 }h}, so that Kx is a bounded linear map from H into KF0 . A
straightforward calculation shows that on KF0 we have Kx f f pxq. The mapping of KF0
into the space of all H-valued functions on X which sends f into the function x Kx f is
linear, injective and compatible with the identification of KF0 with a dense subset of KF0 .
Thus we can regard KF0 as a Hilbert space RpKq of H-valued functions on X. We have
proved that RpKq satisfies ((a)) and ((b)); the uniqueness assumption clearly holds. RpKq
is called the reproducing-kernel Hilbert space determined by K.

Theorem 1.9. A kernel has a Kolmogorov decomposition iff it is positive-definite.
Proof. It follows form Example 1.2 that a kernel having a Kolmogorov decomposition
is positive-definite. If K is a positive-definite kernel, take V pxq Kx : H RpKq as in
the proof of Theorem 1.8; then Kpx, yq V pxq V pyq. Thus pK , RpKqq is a Kolmogorov
decomposition of K; from Theorem 1.8 it is minimal.

Remark 1.10. It follows from Theorem 1.8 that a positive-definite kernel is Hermitian
symmetric: Kpx, yq V pyq V pxq Kpy, xq.
Definition 1.11. The set K ` pX; Hq of positive-definite kernels in KpX; Hq forms a
cone; we define the induced partial ordering: put K K 1 iff K K 1 P K ` pX; Hq.
The next result says that R is functorial:
Theorem 1.12. Let K and K 1 be positive-definite kernels; then K K 1 iff there is a
(necessarily unique) contraction C : RpKq RpK 1 q s.t. Kx1 CKx for all x P X.
Proof. Let K, K 1 P K ` pX; Hq. Then K K 1 iff pKv, vq pK 1 v, vq for all
v P F0 pX; Hq; this holds iff hKv, Kvi hK 1 v, K 1 vi for all v P F0 pX; Hq. This is the
case iff there is a contraction C : RpKq RpK 1 q s.t. Kv CK 1 v for all v P F0 pX; Hq.
The result follows by considering the generating set thx : h P H, x P Xu in F0 pX; Hq, since
Kx h Khx CK 1 hx CKx1 h for all x P X and h P H.

Putting this result together with Lemma 1.4 we have:
Corollary 1.13. Let K and K 1 be positive-definite kernels with Kolmogorov decompositions V and V 1 , respectively. Then K K 1 iff there is a positive contraction T in BpHV q
s.t.
K 1 px, yq V pxq V pyq
for all x, y P X.
Theorem 1.14. Let K be in K ` pX; Hq; then for each 0 and each z P X we have
Kp, q Kp, zqp ` Kpz, zqq1 Kpz, q.
In particular, the Schwarz inequality holds:
Kp, q }Kpz, zq} Kp, zqKpz, q.
Proof. Let V be a minimal Kolmogorov decomposition for K; then we have

Kpx, zqp ` Kpz, zqq1 Kpz, yq V pxq V pzqp ` V pzq V pzqq1 V pzq V pyq
for all x, y, z P X. Thus by Theorem 1.9 it is enough to show that the operator
1
W p ` V pzq V pzqq 2 V pzq

is a contraction. But
1
W W p ` V pzq V pzqq 2 V pzq V pzqp ` V pzq V pzqq 2

p ` V pzq V pzqq1 V pzq V pzq 1
by the spectral theorem.
CHAPTER 2
Positive-definite functions
The principal results in this chapter are two well-known representation theorems: the
Naimark-Sz.-Nagy characterization of positive-definite functions on groups (Corollary 2.6)
and the Stinespring decomposition for completely-positive maps on Banach *-algebras (Theorem 2.13). We exploit the existence and uniqueness of minimal Kolmogorov decompositions
for certain functions on semigroups with involution.
Definition 2.1. Let S be a semigroup, and let J : S S be a map of S into itself s.t.
(i) J 2 iS , (ii) Jpabq JpbqJpaq for all a, b P S; then J is said to be an involution. An
element a of a semigroup with involution pS, Jq is said to be an isometry if
JpsqJpaqat Jpsqt
for all s, t P S. The set SJ of isometries in pS, Jq is a sub-semigroup.
(2.1)
Example 2.2.
1. Let S be a group and let Jpaq a1 for all a P S; then SJ S.
2. Let S be a *-algebra with unit, and let Jpaq a ; then SJ ta P S : a a 1u
so
that the elements of SJ are isometries in the usual sense, and the elements of
SJ JpSJ q are the unitaries.
Definition 2.3. Let H be a Hilbert space and let pS, Jq be a semigroup with involution;
then a function T : S BpHq is said to be positive-definite if the kernel a, b T pJpaqbq
is positive-definite. A Kolmogorov decomposition for a positive-definite function is a Kolmogorov decomposition for it associated kernel.
Example 2.4. Let pS, Jq be a group, as in Example 1. above. Let : S BpH q be a
unitary representation of S. Let W : H H be an isometry; then the function
T pgq W pgqW
is positive-definite and has a Kolmogorov decomposition V where V pgq U pgqW . We shall
see that every positive-definite function on a group can be put in this form.
Theorem 2.5. Let pS, Jq be a semigroup with involution, let T : S BpHq be a positivedefinite function on S, and let V be a minimal Kolmogorov decomposition for T . Then there
exists a unique homomorphism of SJ into the semigroup of isometries on HV , s.t.
pbqV paq V pbaq
for all b P SJ and all a P S. It follows that
T pJpaqbcq V paq pbqV pcq
for all b P SJ and all a, c P S, and that the restriction of to SJ
pbq pJpbqq.
10
JpSJ q is a *-map:
2. POSITIVE-DEFINITE FUNCTIONS
11
Moreover, if S is a topological semigroup then continuity in the weak operator topology of the
map a T paq entails the same for b pbq.
Proof. For all a, c P S we have V pbaq V pbcq T pJpbaqbcq T pJpaqcq V paq V pcq
whenever b P SJ . Hence, by Lemma 1.4, the minimality of V entails the existence of a unique
isometry pbq : HV HV , s.t.
pbqV pcq V pbcq
(2.2)
1
1
1
for all
c P S. It follows from (2.2) that pbqpb q pbb q for all b, b P SJ . Now suppose that
b P SJ JpSJ q; then for all a, c P S we have
V paq pbq V pcq rpbqV paqs V pcq V pbaq V pcq
T pJpbaqcq T pJpaqJpbqcq
V paq pJpbqqV pcq,
so that pbq pJpbqq by uniqueness. The continuity assertion is clear.
Corollary 2.6. Let G be a group, and let T : G BpHq be a positive-definite function

on G. Then there exists a Hilbert space H , a unitary representation : G BpH q and a
map V : H H in V P BpH, H q s.t.
T pgq V pgqV
(2.3)
for all g P G. If the decomposition (2.3) is minimal then it is unique up to unitary equivalence.
Definition 2.7. Let A be a *-algebra with involution Jpaq a . A map T : A BpHq
is said to be completely positive if it is linear and positive-definite. It follows that if V is a
minimal Kolmogorov decomposition for a completely positive map then V : A BpH, HV q
is linear.
Example 2.8.
1. Let W : H HW be an isometry, and let A be a *-subalgebra of BpHW q. Then
T : A BpHq given by T paq W aW is completely positive.
2. Let : A BpHq be a *-representation of a *-algebra A; then is completely
positive.
Definition 2.9.
An algebra S with involution J is said
to be a U*-algebra if it is the
linear span of SJ JpSJ q. If S has a unit, then u is in SJ JpSJ q iff Jpuqu 1 uJpuq.
Example 2.10. An element of Banach *-algebra with identity, A, can be expressed as a
linear combination of four unitaries in A; hence every such algebra is a U*-algebra.
Theorem 2.11. Let pS, Jq be a U*-algebra, let T : S BpHq be completely positive,
and let V be a minimal Kolmogorov decomposition for T . Then there exists a unique *representation : S BpHV q s.t.
paqV pcq V pacq
for all a, c P S. It follows that
T pb acq V pbq paqV pcq
for all a, b, c P S.
12
Proof. Let : SJ JpSJ q BpHV q be the *-homomorphism of Theorem 2.5. Then

for a
P S we have a ni1
zi ui , where z1 , . . . , zn are complex numbers and u1 , . . . , un are
in SJ JpSJ q; put paq ni1 zi pui q. Then paqV pcq V pacq for all c P S, so that is a
well-defined *-homomorphism from S into BpHV q.

From this follows the Stinespring decomposition for a completely positive map on a unital
U*-algebra.
Corollary 2.12. Let A be a unital U*-algebra and let T : A BpHq be completely
positive. Then there exists, uniquely up to unitary equivalence, a *-representation of A on
a Hilbert space HV and a bounded linear map V : H HV s.t.
T paq V paqV
for all a P A and HV
tpaqV h : a P A, h P Hu.
Stinespring decomposition can also be obtained for more general algebras (for example,
for some non-unital algebras) in such a way that the Stinespring representation is actually
defined on a larger algebra. Rather than give the details in very abstract situations, we give
an example of an extension of Stinesprings theorem. The result is quite adequate for our
needs; the proof illustrates the essential technique.
Theorem 2.13. Let A be a Banach *-algebra with approximate identity, and let T be
a completely positive map from A into BpHq. Then there exists, uniquely up to unitary
equivalence, a Hilbert space HV , a *-representation of A on HV , and a map V in BpH, HV q,
s.t.
T paq V paqV
for all a P A, and
HV
tpaqV h : a P A, h P Hu.
Proof. Let V be a minimal Kolmogorov decomposition for T , and let A1 denote the
unital Banach *-algebra obtained from A by adjoining an identity. Then A is an ideal in A1
and
V pxaq V pbcq T ppbaq pbcqq T pa cq V pa qV pcq
for all a, c P A and all unitaries b P A1 . Hence, since A1 is a U*-algebra, there exists a
unique representation 1 of A1 on HV s.t. 1 pbqV paq V pbaq for all b P A1 and a P A. Let
denote the restriction of 1 to A. It follows from 0.4 that T is bounded and hence so is
V pq, since }V pxq}2 }T px xq} for all x P A. We identify BpH, HV q with the dual of the
space of trace-class operators from HV into H. Let tu u be an approximate identity for A,
then the net tV pu qu is bounded in BpH, HV q and so has a weak *-limit V , say. We see that
paqV lim paqV pu q lim V pau q V paq for all a P A. The result follows.

Note that the above theorem applies to a non-unital C*-algebra and to the group algebra
L pGq of a locally compact group G. It is apt at this point to discuss the intimate relationship
between positive-definite functions on groups and completely positive maps on algebras,
and in particular the relationship between the Naimark-Sz.-Nagy representation and the
Stinespring decomposition. In the first place, consider a unital U*-algebra A, and let G
denote a subgroup of its group of unitaries s.t. spanpGq A. Clearly a completely positive
map on A restricts to a positive-definite function on G. Conversely, if T is a linear map
on A s.t. its restriction to G is positive-definite, then T is completely positive. For if ai ,
1
13
i 1, . . . , n, are elements
of A, then there exist complex numbers zip and elements gp of G,
p 1, . . . , m, s.t. ai p zip gp , since A LinpGq. From the linearity of T we have
T pai aj q
zip T gp1 gp zjq ;
p,q
regarding the right-hand side as a matrix-element of the product of three matrices, we see
that rT pa aqs is a positive matrix since rT pgp1 gq qs is. Moreover, T is a homomorphism if
and only if its restriction to G is a unitary representation. Thus the restriction map takes
the Stinespring decomposition into the Naimark-Sz.-Nagy representation.
This connection can be taken further. Suppose G is a locally compact group, and T is a
strongly continuous positive-definite function on G (acting on a Hilbert space H, say). Then
it is easy to verify that
1
f pgqT pgqdg,
T pf q
G
where dg is a left-invariant Haar measure on G, defines a completely positive map T 1 of

the Banach *-algebra L1 pGq into BpHq. Moreover it can be shown, using the existence of
an approximate identity for L1 pGq, that each completely positive map on L1 pGq arises in
this way; T 1 is a homomorphism of L1 pGq iff T is a unitary representation of G. Thus the
(minimal) Naimark-Sz.-Nagy representation of T on G (Corollary 2.6),
T pgq V U pgqV,
gives the (minimal) Stinespring decomposition on L1 pGq (Theorem 2.13),
T 1 pf q V U 1 pf qV,
and vice-versa.
CHAPTER 3
Dilations of Semigroup of Contractions

In this chapter we discuss some dilation theorems for semigroups of operators on Hilbert
spaces. They are of two kinds: one typified by Coopers Theorem 3.1, the other by Sz.-Nagys
Theorem 3.2. In Chapter 16 we will produce yet a third kind.
Theorem 3.1. Let tTs : s P R` u be a strongly continuous semigroup of isometries on a
Hilbert space H; then there exists a Hilbert space Hu , a unitary group tUs : s P Ru on Hu ,
and an isometry V : H Hu , s.t. V Ts Us V for all s P R` .
If we assume less about tTs u we get the weaker result:
Theorem 3.2. Let tTs : s P R` u be a strongly continuous semigroup of contractions on
a Hilbert space H; then there exists a Hilbert space Hu , a unitary group tUs : s P Ru on Hu
and an isometry V : H Hu , s.t. Ts V Us V for all s P R` .
We now discuss the extent to which the results of Theorems 3.1 and 3.2 generalize when
R is replaced by an arbitrary abelian semigroup S; we will obtain Theorem 3.1 as a special
case of Theorem 3.4 and Theorem 3.2 as a special case of Theorem 3.11. Finally, we show
(Theorem 3.13) that when the semigroup tTs u in the statement of Theorem 3.2 is strongly
contracting to zero, the unitary group tUs u satisfies an abstract Langevin equation. Only
Theorems 3.1, 3.2 and 3.13 will be required in the applications to irreversible evolutions.
In this chapter each abelian semigroup S is assumed to have a zero. We are given a
homomorphism T : S BpHq of S into the semigroup of isometries on a Hilbert space H.
We want to use T to construct a homomorphism U : S BpHu q of S into the group of
unitaries on some Hilbert space Hu , and to examine its relation to T . Now to each abelian
semigroup S there corresponds a group KpSq and a homomorphism : S KpSq which is
universal in the sense that every homomorphism of S into a group G factors through KpSq:
there exists a unique homomorphism s.t. the following diagram commutes.
`
KpSq
S
G
The first step, then us to use T to construct a homomorphism from KpSq into the group
of unitaries on some Hilbert space. It turns out that this is always possible. First we recall
one construction of KpSq.
Definition 3.3. Let S be an abelian semigroup. Let : S S S be the diagonal
map, and let : S S S S{pSq be the natural projection. Then S S{pSq is a
group (since ps, tq ` pt, sq p0, 0q, every element has an inverse), which is called the
Grothendieck group of S, and denoted KpSq.
The map s p0, sq is a homomorphism. which we denote by S : S KpSq. If S is
itself a group then S is an isomorphism. The construction is functorial: if : S S 1 is a
14
3. DILATIONS OF SEMIGROUP OF CONTRACTIONS
15
homomorphism of semigroups, then there is a unique homomorphism Kpq : KpSq KpS 1 q

s.t. the following diagram commutes.
S
KpSq
Kpq
S1
S 1
KpS 1 q
The universal property of pS , KpSqq follows from this. The homomorphism S is injective
iff the cancellation law holds in S: s ` u t ` u implies that s t. When S is a topological
semigroup we give KpSq the quotient topology; this makes S continuous.
Theorem 3.4. Let S be an abelian semigroup. Let : S KpSq be the canonical homomorphism of S into the Grothendieck group of S. Let T : S BpHq be a homomorphism
of S into the semigroup of isometries on a Hilbert space H. Then there is a positive-definite
function T 1 on KpSq s.t.
1
Ts Tt
Tptqpsq
(3.1)
for all ps, tq in S S.
Proof. Consider the function s, t Ts Tt on S S; since Tu is an isometry we have
Tu Tu 1 and the function is constant on pSq-cosets and determines a unique function
T 1 on KpSq s.t. (3.1) holds. To prove that T 1 is positive-definite, consider a fixed n-tuple
k1 , . . . , kn in KpSq and choose coset representatives psi , ti q of ki , i 1, . . . , n.
Put
s11 t1 ` s2 ` ` sn ,
s12 s1 ` t2 ` ` sn ,
..
...
.
s1n s1 ` s2 ` ` tn ;
then
kj ki ps1i , s1j q
so that
Tk1 j ki Ts1i Ts1j .
and it is clear that T 1 is positive-definite.
Definition 3.5. A semigroup homomorphism T : S BpHq of an abelian semigroup

into the bounded operators on a Hilbert space H s.t. T0 1 is said to have a unitary
dilation in the strong sense if there exists a Hilbert space HV , an isometry V : H HV ,
and a unitary representation U : KpSq BpHV q of the Grothendieck group of S s.t.
V Ts Upsq V.
(3.2)
The relation (3.2) implies the weaker

Ts V Upsq V
since V is an isometry. If (3.3) holds we say that S has a unitary dilation.
(3.3)
16
Theorem 3.6. Let S be an abelian semigroup and let T : S BpHq be a homomorphism

s.t. T0 1. Then T has a unitary dilation pU, V q in the strong sense iff Ts is an isometry
for all s P S. If pU, V q is minimal then it is unique up to a unitary equivalence. If S is a
topological semigroup then the continuity of s Ts in the weak-operator topology implies the
same for k Uk .
Proof. The only if part is obvious since the Upsq and V are isometries. If the Ts
are isometries then it follows from Theorem 3.4 that the associated function T 1 on KpSq is
positive-definite. The remainder of the proof follows the lines of that of Theorem 2.5 and its
Corollary, but the particular form (3.1) of T 1 yields more: (3.2) holds. It is enough to use
the minimal Kolmogorov decomposition T 1 associated with the reproducing kernel Hilbert
space RpT 1 q of the kernel T 1 . We take HV to be RpT 1 q and the isometry V : H HV to be
1
pV hqpk 1 q Tk
1h
for all k P KpSq. The representation U : KpSq BpHV q is given on V H by

1
pUk V hqpk 1 q Tkk
1 h;
using (3.1) we get (3.2) when k psq.
Turning to semigroups of contractions which are not necessarily isometries, we ask if they
have a unitary dilation (in the sense of (3.3)). To adapt the proof of Theorem 3.4 to this
case we have to assume more about S.
Remark 3.7. The following two properties of S are equivalent:
(i)
pSq X rpSqs t0u.
(ii) If s, t, u, v are in S and s ` u v and u t ` v, then
s ` w w for some w P S.
(3.4)
(3.5)
Theorem 3.8. Let S be an abelian semigroup for which (3.4) holds, and let T : KpSq
BpHq satisfy
(i) T0 1,
(ii) Tk Tk @k P KpSq,
(iii) Tk Tk1 Tk`k1 , whenever k, k 1 and k ` k 1 are not in rpSqs.
Then T is positive-definite iff Tk is a contraction for each k in KpSq; in which case T
has a unitary dilation.
Proof. Choose a fixed n-tuple of elements k1 , . . . , kn of KpSq, ordered so that kj ki
is not in rpSqs if i j. Consider the n n matrix with entries
tij Tkj ki ,
and define
#
tij
wij
0
if i j,
otherwise,
and
`
dij ij 1 ti1,i ti1,i , i 1,

d11 1.
We claim that
t w dw;
(3.6)
17
then t is positive iff d is, and d is positive iff the Tki , i 1, . . . , n, are contractions. It
remains to prove (3.6). Notice that Tij tjk tik whenever i j k, and that tij tjk . If
i j then
i
pw dwqij
pw qik pdwqkj .
k1
Thus for i 1 we have
pw dwq11 w11
d11 w1j t11 t1j t1j ;
for j i 1 we have
pw dwqij
pw qik dkk wkj
k1
i
tki dkk tkj
k1
t1i t1j
` t2i t2j t1i t1j

`
`
` tii tij ti1,i ti1,j

tij .
This establishes that T is positive-definite; the existence of the unitary dilation follows from
Corollary 2.6.

Remark 3.9. The following conditions on S are equivalent:
(i)
pSq Y rpSqs KpSq
(ii) Whenever s, t are in S there exist u, v, w in S s.t.
either
or
t ` u v, s ` u w ` v,
t ` u v ` w, s ` u v.
(3.7)
(3.8)
Definition 3.10. We say that an abelian semigroup S is totally ordered if (3.4) and
(3.7) hold.
Theorem 3.11. Let S be a totally ordered abelian semigroup, and let T : S BpHq
be a homomorphism satisfying (i) T0 1, (ii) }Ts } 1, and the cancellation law: (iii) if
h ` s h ` t then Ts Tt . Then there is a unique positive-definite function T 1 on KpSq s.t.
1
1
Tpsq
Ts and Tpsq
Ts
(3.9)
for all s in S; hence T has a unitary dilation.

Proof. Since (3.4) and (3.7) hold, there is a well-defined function T 1 on KpSq which is
uniquely determined by (3.9). It is easy to check that T 1 satisfies conditions (i), (ii) and (iii)
of Theorem 3.8; the result follows.

We began this chapter by looking at one extreme case of a semigroup of contractions,
where the contractions preserve the norm of each vector. We end the chapter with a look at
the opposite extreme, in which the norm of each vector goes to zero eventually under repeated
action of each contraction. In this case, a minimal unitary dilation of the semigroup satisfies
an abstract Langevin equation.
18
Definition 3.12. Let S be a locally compact semigroup, a semigroup tTs : s P Su of

contractions on a Hilbert space H is said to contract strongly to zero (at infinity) if for all
h P H we have
lim }Ts h} 0.
s8
First we require an alternative construction of a unitary dilation of a semigroup of contractions over R` , which contracts strongly to zero.
Theorem 3.13. Let tTt : t P R` u be a strongly continuous semigroup of contractions on
a Hilbert space H which contracts strongly to zero. Then there is a Hilbert space N and an
isometry W : H L2 pR; N q s.t.
Tt W Ut W,
t 0,
(3.10)
where tUt : t P Ru is the strongly continuous unitary group of right-translations on L2 pR; N q:

pUt f qpsq f ps tq.
Proof. Let B denote the infinitesimal generator of Tt . Since t }Tt h}2 is monotone
decreasing we have, for all h in DpBq,
d
hBh, hi ` hh, Bhi hTt h, Tt hi 0.
(3.11)
dt
t0
Let N0 denote the null space of this quadratic form

N0 th P DpBq : hBh, hi ` hh, Bhi 0u.
Let A be the quotient map of DpBq onto DpBq{N0 . Then by (3.11) and the Schwarz
inequality there exists an inner product h, iB on DpBq{N0 s.t.
hAh, AkiB hBh, ki hh, Bki
(3.12)
for all h, k P DpBq. Let N denote the separable Hilbert space got by completing DpBq{N0 .
Then, for all h P DpBq and t 0, we have by (3.11) and (3.12)
0
}ATs h}2B ds }h}2 }Tt h}2 .
(3.13)
t
Letting t 8, remembering that Tt contracts strongly to zero, we see that there is an

isometric embedding W of H in L2 pR ; N q given on DpBq by
pW hqpsq ATs h
(3.14)
for all s 0.
We regard L2 pR ; N q as a subspace of L2 pR; N q in the obvious way; then we have, for
each h in DpBq and t 0,
#
ATts H, s t,
pUt W hqpsq
0,
s t,
pW Tt hqpsq ` wt psq,
where ws is in L2 pR` ; N q W pHqK . Thus for each t 0 we have
Tt W Ut W,
so that Ut is a unitary dilation of Tt on Hv L2 pR; N q.
19
It will be shown later that this dilation is minimal. It is, in fact, a consequence of the
Langevin equation (3.17) which we now propose to study.
Let : R BpN, HV q be the map given by
#
r0,ts psqn,
t 0,
pt nqpsq
rt,0s psqn, t 0,
for each n P N , where ra,bs denotes the characteristic function of the interval ra, bs in R.
Then is a minimal Kolmogorov decomposition of the positive-definite kernel s, t ps^tq1N
on R R:
s t ps ^ tq1N ,
(3.15)
for all s, t P R. The following lemma is useful in proving Theorem 3.15:

Lemma 3.14. Let tTt : t P R` u be a strongly continuous semigroup of contractions on a
Hilbert space H, and let B be its generator. Then DpBq can be regarded as a Hilbert space
with respect to the norm given, for h P DpBq, by
|h|2 }h}2 ` }Bh}2 ;
(3.16)
in which DpB 2 q is dense.

Proof. Since the generator B is a closed operator, its domain DpBq is a Hilbert space
w.r.t. the norm (3.16). On it we define the semigroup St : m Tt m. The strong continuity
of t Tt implies the same for St ; hence the domain of the generator of St is dense, and the
proof is completed.

Theorem 3.15. Let tTt etB : tR` u be a strongly continuous semigroup of contractions,
contracting strongly to zero, on a Hilbert space H. Let tUt u be a minimal unitary dilation of
Tt . Then there exists:
(i) a Hilbert space N , and a bounded linear operator
A : pDpBq, |cdot|q N,
(ii) a map : R BpN, HU q satisfying
t s ps ^ tq1N
for s, t P R and
HU
ts n : s P R, n P N u,
s.t.
t
Ut W h Us W h
Uu W Bh du ` pt s qAh
(3.17)
for all h P DpBq.

Proof. Take for tUt u the dilation of Theorem 3.13; take for the map the minimal
Kolmogorov decomposition (3.15); then (3.17) is easily verified by integration-by-parts for
h P DpB 2 q and hence, by Lemma 3.14 for all h P DpBq. That the dilation tUt u is minimal
now follows from (3.17) and the minimality of .
20
Remark 3.16. It is also possible to treat the semigroup N using this procedure. In this
case, let T be a contraction on the Hilbert space H s.t. the semigroup tT n : n P Nu contracts
strongly to zero at infinity. We can show that
0
DT T j h2 }h}2 ,
(3.18)
j8
1
for all h P H, where DT p1 T T q 2 . We take N rDT Hs and A : H N the map

given by Ah DT h. We embed H isometrically in HU L2 pNq by
#
AT j , j 0
pehqpjq
0,
j 0.
The unitary group tU n : n P Zu is defined on L2 pNq by translation:
pU n f qpjq f pj nq,
for j, n P Z and f P HU . Then tU n u is a minimal unitary dilation for T . We now define
: N BpN, HU q, as in the continuous case, so that
m
n pm ^ nq1N
for all m, n P Z, and
HU
tm x : m P N, x P N u.
In this case we have the discrete Langevin equation
m1
U u erpT 1qhs ` pm n qAh,

U m eh U n eh
un
valid for all h P H.
(3.19)
CHAPTER 4
C -algebras and Positivity

The main results in this chapter concern a positive linear map T from one C -algebra
A into another C -algebra B. If either A or B is commutative, then T is completely positive (Theorem 4.3 and 4.2). This allows us to deduce certain Schwarz-type inequalities in
Corollary 4.4, and the identities of Broise in Corollary 4.5. In the proofs we make use if
a characterization of the positivity of an element of the matrix C -algebra Mn pAq over a
C -algebra A (Lemma 4.1).
We end by deriving the canonical decomposition of a normal completely positive map on
a von Neumann algebra (Theorem 4.6).
If A is a -algebra, and n is a positive integer, we let Mn pAq denote the -algebra of all
n n matrices over A under the natural operations. If teij : 1 i, j
nu is a system of
matrix units for Mn Mn pCq, then the -algebraic isomorphism raij s aij b eij allows us
to identify Mn pAq with the algebraic tensor product AbMn . If A is a C -algebra, represented
say on a Hilbert space H, then Mn pAq is also a C -algebra and can be faithfully represented
on H n H H H b Cn as follows:
ffn
n
raij sni,j1 rfj snj1

aij fj
; raij s P Mn pAq, rfj s P H n .
j1
i1
Let A and B be -algebras, and let T be a linear map from A into B; let Tn denote the
product mapping T b 1n from Mn pAq into Mn pBq where 1n denotes the identity mapping
on Mn pCq. Then Tn acts element-wise on each matrix over A:
Tn : raij s rT paij qs.
Suppose now B is a C -algebra. Then Tn is positive (0.4) iff Tn pa aq 0 for each a
n
in Mn pAq. But if a P raij s P Mn pAq, a a is the sum

api apj . Thus Tn is positive iff
p1
rT pai aj qs is a positive matrix for all a1 , . . . , an in A. In particular, T completely positive is

equivalent to Tn positive for all n 1. It would thus seem useful to study the order structure
of matrix algebras more closely:
Lemma 4.1. Let A be a C -algebra, and a raij s be an element of Mn pAq.
(a) The following conditions are equivalent:
(i) a 0.
(ii) a
isa finite sum of matrices, each of the form rbi bj s where b1 , . . . , bn P A.
(iii) ai aij aj 0, for all sequences a1 , . . . , an P A.
(b) If A
is commutative, then the above three conditions are also equivalent to:
(iv) aij i j 0, for all sequences 1 , . . . n P C.
(c) If for the C -algebra A condition (iv) is equivalent to conditions (i)-(iii), then A
must be commutative.
21
4. C -ALGEBRAS AND POSITIVITY
22
Proof.
(a) (i) (ii) has already been observed;
(ii) (iii) is trivial;
(iii) (i): If we represent A on a Hilbert space H, we can decompose H into
cyclic orthogonal subspaces. Thus we can assume A has a cyclic vector f P H.
Then
E
D
haij aj f, ai f i
ai aij aj f, f 0,
for all a1 , . . . , an P A. Thus, since f is cyclic, haij fj , fi i 0 for all f1 , . . . , fn P

H. That is, raij s is positive.
(b) (iv) (iii): Represent A as C0 pXq, the continuous functions vanishing at infinity
on a locally compact Hausdorff space X. Then
aij zi zj 0, @z1 , . . . , zn P C,
aij pxq
zi zj 0, @z1 , . . . , zn P C, x P X,
raij pxqs 0p in Mn pCqq, @x P X,
aij ai pxqaj pxq 0, @a1 , . . . , an P A, x P X,
aij ai aj 0, @a1 , . . . , an P A.
(i),(ii) (iv) is trivial.
(c) Suppose A has the property that if a P M2 pAq satisfies
aij zi zj 0 @z1 , z2 P C,
(4.1)
then a is positive. The C -algebra obtained from A by adjoining an identity has

the same property. Thus we can assume A is unital. Take b P A, and consider the
matrix
1 b
a
b bb
which clearly satisfies (4.1), so that a is positive. But

b
b 1 1 b
;
bb b b
b bb 1
and so bb b b, for all b P A. By symmetry each element of A is normal and so
A is commutative.

Theorem 4.2. Let A, B be C -algebras, with B commutative. Then automatically any
positive linear map from A into B is completely positive.
Proof. Suppose raij s P Mn pAq is positive. Then
aij zi zj 0 @z1 , . . . , zn P C.
Then if T is any positive map from A into B, and
T
aij zi zj 0 @z1 , . . . , zn P C,
hence
T paij q zi zj 0 @z1 , . . . , zn P C.
23
The conclusion follows from Lemma 4.1(b).
Positive linear maps whose domains are commutative to C -algebras automatically are
completely positive, as the following theorem shows:
Theorem 4.3. Let A, B be C -algebras with A commutative. Then any positive linear
map from A into B is completely positive.
Proof. By going to the second dual, we can assume that A is a W -algebra and that
the given positive linear map T from A into B is ultraweakly continuous. We represent A
as L8 p, q for some localizable measure space p, q, with predual L1 p, q, and we take
B to act on a Hilbert space H. Then for all f, g P H, the map
a hT paqf, gi
is ultraweakly continuous on L8 p, q. Hence there exists hpf, gq in L1 p, q s.t.
hT paqf, hi ha, hpf, gqi .
Moreover, f, g
hpf, gq is sesquilinear, and hpf, gq 0 since T is positive. Let f1 , . . . , fn
be elements of H; then for all z1 , . . . , zn P C:
zi zj hpfi , fj q h
zi fi , zj fj 0,
zi zj hpfi , fj qpq 0 a.e. P ,

(4.2)
rhpfi , fj qpqs 0 a.e.
Then, for all a1 , . . . , an P L8 p, q
T pai aj qfj , fi ai aj , hpfi , fj q ai pqaj pqhpfj , fi qpq dpq 0,
by (4.2).

Corollary 4.4. Let T be a positive linear map from a C -algebra A into another C algebra B. If a is a normal element of A, then
}T } T pa aq T paq T paq.
(4.3)
}T } T pa a ` aa q T paq T paq ` T paqT paq
(4.4)
More generally:
for all a P A.
Proof. If C is a commutative C -algebra generated by a normal element a, then the
restriction of T to C is completely positive, by Theorem 4.3. Hence we can apply the
Schwarz inequality of Theorem 1.14. If a is an arbitrary element of A, we can apply (4.3)
to the self-adjoin elements a ` a and ipa a q. The inequality in (4.4) then follows by
addition.

Corollary 4.5. Let T be a positive contraction from a C -algebra A into another C algebra B, and a a self-adjoint element of A, s.t. T pa2 q T paq2 . Then
T pab ` baq T paqT pbq ` T pbqT paq
(4.5)
T pabaq T paqT pbqT paq
(4.6)
and
for all b P A.
24
Proof. Fix , a state on B, and consider the sesquilinear form D on A.

D : px, yq rT pxy ` y xq T pxqT pyq T pyq T pxqs.
By Corollary 4.4, we have Dpx, xq 0 by the Cauchy-Schwarz inequality applied to D; hence
(4.5) holds. Then (4.6) follows easily from Jordan identities.

The Stinespring representation theorem can also be used to obtain a description of completely positive normal maps:
Theorem 4.6. Let A be a von Neumann algebra on a Hilbert space H, and let K be
another Hilbert space. If is a completely positive ultraweakly continuous map from A into
BpKq, then there exist tAi : i P Xu in BpK, Hq s.t., for all x P A,
pxq
Ai xAi .
If K is infinite-dimensional, we can choose X s.t. its cardinality is at most that of a complete
orthonormal set for K.
Proof. By the Stinespring decomposition, we can assume that is a normal representation with cyclic vector f. Then since hpqf, f i is a normal
state on A, there exist
2
}fi } 8, and hpxqf, f i hxfi , fi i for all x P A. Since
vectors tfi : i P Nu in H s.t.
}xfi } }pxqf } for all x P A, there exist contractions Ai from K into H s.t. Ai pxqf xfi .
Then, for all x, z P A, we have
hpxqpzqf, pzqf i hpz xzqf, f i

hz xzfi , fi i
hxzfi , zfi i
hxAi pzqf, Ai pzqf i
hAi xAi pzqf, pzqf i .
Since f is a cyclic vector for , we have pxq Ai xAi for all x P A; the series converges in
the ultraweak topology. The usual counting arguments in a Hilbert space give the cardinality
result.
CHAPTER 5
Conditional Expectations
As we mentioned in the introduction, we wish to define a class of C -algebraic maps
which generalize the class of conditional expectations of classical probability theory. In this
chapter, A will denote a unital C -algebra, and B a unital C -subalgebra of A. To merit
the description conditional expectation, we will require the following properties of a linear
map N of A onto B:
CE1 : N is a projection of norm one s.t. N p1A q 1B ;
CE2 : N pa1 N pa2 qq N pa1 qN pa2 q, for all a1 , a2 P A, or equivalently, N pabq N paqb for
all a P A, and b P B;
CE3 : N is completely positive.
It is easily verified that these properties hold in the following examples:
Example 5.1.
1. Let tpi : i Pu be a mutually orthogonal
family of projections
in a W -algebra A, let p
pi and let N pxq
pi xpi for all x P A; then
N is a projection of A onto the intersection of pAp with the relative commutant
tpi : i P uc tx P A : xpi pi x @i P u.
2. Let A and B be W -algebras, and identify B with 1 b B as a W -subalgebra of the
W -tensor product A b B. Let be a normal state of A, then b 1 is a projection
of A b B onto B; it is the dual of the injection of states:
b
@ P B .
(Similarly, for C -algebras with spatial or minimal tensor product.)

The main result (Theorem 5.3) is that CE1 entails both CE2 and CE3. We are thus led
to:
Definition 5.2. Let B be a unital C -subalgebra of a unital C -algebra A. A conditional
expectation N is a projection of norm one from A onto B such that N p1A q 1B .
Taking C B , we see in the following theorem that a conditional expectation is
automatically completely positive (CE3), and has the module mapping property (CE2).
Theorem 5.3. Let B be a unital C -subalgebra of a unital C -algebra A. Let N be
a linear map of norm one from A into a W -algebra C s.t. the restriction of N to B
is a homomorphism onto a weakly dense subalgebra of C, with N p1A q 1C . Then N is
completely positive, and N pabq N paqN pbq for all a P A and b P B.
Proof. That N is positive follows from 0.4. By going to the second duals we can assume
that A, B, C are all von Neumann algebras and N is normal. It is enough to consider C in
an irreducible representation, and so we may assume
that C BpHq for some Hilbert space
H. Let e be the central projection in B s.t. Ker N B is the two-sided ideal Bp1B eq. Then
N peq 1C . For the moment we will only consider the restriction N0 of N to eAe, so that
the restriction of N0 to eBe B0 is faithful. Via a spatial isomorphism, we may assume
25
5. CONDITIONAL EXPECTATIONS
26
B0 C b BpHq, A0 A b BpHq and N0 p1A b bq b, for all b P BpHq. Then, by Corollary

4.5, we have N0 pa b f q f N0 pa b 1qf for all a P A and all projection f P BpHq. After some
computation, we find that f N0 pa b 1q N0 pa b 1qf . Thus N0 pa b 1q lies in BpHq1 C and
hence N0 b 1, which is completely
N0 pa b 1q paq, where is a normal state on A;
positive, and N0 pabq N0 paqN0 pbq for all a P A0 and b P B0 . Then for all a P A, b P B, we
have
N pabq N pabq1C N peqN pabqN peq
N peabeq, by Corollary 4.5,
N peaebeq, since e is central in B,
N peaeqN pebeq N paqN pbq;
the theorem follows.
CHAPTER 6
Fock Space
In this chapter we recall some elementary results about Fock space, and show how the
Boson and Fermion Fock spaces arise naturally with the Kolmogorov decomposition of certain
positive-definite functions.
Let H be a Hilbert space; for each positive integer n, let Hn denote the n-fold tensor
product bn H, and let H0 denote the one-dimensional Hilbert space spanned by a single unit
vector , called the Fock vacuum vector. Fock space F pHq is then defined as
8
F pHq
Hn .
n0
Let T be a contraction from H to another Hilbert space K, let Tn denote the contraction
bn T from Hn into Kn , and put T0 1; we define F pT q to be the contraction from F pHq
into F pKq given by
8
F pT q
Tn .
n0
The assertions in the following lemma are then easily verified.

Lemma 6.1.
1. F is a functor on the category whose objects are Hilbert spaces and
whose morphisms are contractions:
F pST q F pSqF pT q,
F p1q 1.
(6.1)
2. F p0q is the projection on the Fock vacuum vector :

F p0q b .
(6.2)
F pT q F pT q .
(6.3)
3. F is a -map:
We will not be interested in the whole of Fock space, but only in two of its subspaces,
namely the Boson and the Fermion Fock spaces.
For each positive integer n, let Sn denote the group of all permutations on n symbols.
There is a natural unitary action of Sn on the Hilbert space Hn given by
pf1 b b fn q f1 p1q b b f1 pnq
for all P Sn and f1 , . . . , fn P H.
Remark 6.2. Let T be a contraction between Hilbert spaces H and K; then Tn intertwines the actions of Sn on Hn and Kn : Tn Tn for all P Sn .
Let Pn pn!q1 PSn ; then Pn is the projection from Hn onto the space Hns of symmetric tensors of degree n. Symmetric (or Boson) Fock space F s pHq is then defined by
8
s
F pHq
Hns .
n0
27
6. FOCK SPACE
28
Now let T : H K be a contraction; it follows from Remark 6.2 that Tn maps Hns into
Kns , and so F pT q induces a contraction F s pT q : F s pHq F s pKq. Note that F s inherits the
properties (6.1) to (6.3) of the functor F in Lemma 6.1.
Let pq denote the signature of the permutation , and let Qn pn!q1 PSn pq;
then Qn is the projection from Hn onto the space Hna of antisymmetric tensors of degree n
ovet H. Antisymmetric (or Fermion) Fock space F a pHq is defined by
8
a
F pHq
Hna .
n0
Again, if T : H K is a contraction, it follows from Remark 6.2 that Tn maps Hna into Kna ,
and so F pT q induces a contraction F a pT q : F a pHq F a pKq, and F a inherits the properties
(6.1) to (6.3) from the functor F .
For use later in the study of some algebras naturally associated with the Fock spaces, we
relate the Fock spaces to Kolmogorov decompositions of some positive-definite kernels.
First we look at Boson Fock space: Let h be a vector in the Hilbert space H, and let
hn denote the n-fold tensor product h b b h which lies in Hns , with h0 . Then
hhn , kn i hh, kin for all h, k P H; thus h hn is a minimal Kolmogorov decomposition of
the positive-definite kernel h, k hh, kin on H H. Now define Exp : H F s pHq by
8
1
Expphq
pn!q 2 hn .
n0
s
Theorem 6.3. The map Exp : H F pHq is a minimal Kolmogorov decomposition for
the positive-definite kernel h, k exp hh, ki on H H. Moreover, tExpphq : h P Hu is a
linearly independent total set of vectors for F s pHq.
Proof. That Exppq is a Kolmogorov decomposition for the kernel exp h, i follows by
computation:
hExpphq, Exppkqi exp hh, ki .
(6.4)
Minimality is a consequence of the relation
dn
pn!q 21 hn .
Exppthq
n
dt
t0
It remains to prove the
asserted
linear
independence.
Suppose h1 , . . . , hn in H and
n
z , . . . , zn in C satisfy j1 zj Expphj q 0. Then, by the reproducing property (6.4),
1 n
t
is an eigenvector of the linear
j1 zj exppt hhj , kiq 0 for all t P R and k P H. But e
d
operator dt corresponding to the eigenvalue , and eigenvectors corresponding to distinct
eigenvalues are linearly independent. Thus, for each k P H, we have hhi , ki hhj , ki for
some i j. Hence the set thi u cannot be distinct.

Corollary 6.4. There is a natural identification of F s pH Kq with F s pHq b F s pKq
under which
Expph kq Expphq b Exppkq
and
F s pS T q F s pSq b F s pT q.
Proof. This is a consequence of the uniqueness of a minimal Kolmogorov decomposition
(Lemma 1.4), Theorem 6.3, and the relation
hExpph1 q b Exppk1 q, Expph2 q b Exppk2 qi exp hh1 k1 , h2 k2 i .
6. FOCK SPACE
29

Next we consider Fermion Fock space: Let f1 , . . . , fn lie in the Hilbert space H, and
define f1 ^ ^ fn by
1
f1 ^ ^ fn pn!q 2 Qn pf1 b b fn q.
Then we have
hf1 ^ ^ fn , g1 ^ ^ gn i pn!q hQn pf1 b b fn q, g1 b b gn i
pq f1 p1q b b f1 pnq , g1 b b gn
pq f1 p1q , g1 f1 pnq , gn

detphfi , gj iq.
Thus the map tfi uni1 f1 ^ ^ fn of H n into Hna is a minimal Kolmogorov decomposition for the positive-definite kernel tfi u, tgi u detphfi , gj iq on H n H n .
In what follows we will drop the indices s and a when there is no risk of confusion arising.
CHAPTER 7
Representation of the Canonical Commutation Relations

In this chapter we recall some definitions and formulae associated with the canonical
commutation relations. The main result (Theorem 7.8) is a characterization of generating
functions.
Let H be a Hilbert space; in Theorem 6.3 we noted that Exppq is a minimal Kolmogorov
decomposition for the positive-definite kernel exp h, i on H H. Consider now the linearly
independent total set of normalized vectors
+
#
2
1
}h}
Cphq Exp 2 2 h exp
:hPH .
4
Then
}h k}2
hCphq, Cpkqi exp
4
Im hh, ki
exp i
2
for all h, k P H, so that Cpq is a minimal Kolmogorov decomposition for the positive-definite
kernel
}h k}2
Im hh, ki
h, k exp
.
(7.1)
exp i
4
2
In other words, F s pHq can be identified with the reproducing kernel Hilbert space for the
kernel (7.1). Note that the map
Im hh, ki
: h, k exp i
(7.2)
2
defines a multiplier in the sense of group representation theory.
Definition 7.1. Let pG, `q be a group. A multiplier b on G is a map from G G into
the unit circle tz P C : |z| 1u, such that
bpg, oq bpo, gq 1,
(7.3)
bpg, g 1 qbpg ` g 1 , g 2 q bpg, g 1 ` g 2 qbpg 1 , g 2 q,
(7.4)
for all g, g 1 , g 2 P G. A b-representation of a group G with multiplier b is a map U from G

into the unitary operators on some Hilbert space such that
U poq 1,
1
(7.5)
1
U pgqU pg q U pg ` g qbpg, g q,
(7.6)
for all g, g 1 P G. A projective representation is a b-representation for some multiplier b.

Remark 7.2. The properties (7.3) and (7.4) of a multiplier are merely consistency conditions for the existence of b-representations; for example (7.4) reflects the associative law.
30
7. REPRESENTATION OF THE CANONICAL COMMUTATION RELATIONS
31
Since tCphq : h P Hu is a linearly independent total set of normalized vectors, there is a

well-defined unitary W phq, for each h P H, such that
W phqCpkq Cph ` kqpk, hq
(7.7)
for all k P H. Moreover, W phq obeys the canonical commutation relations:

W phqW pkq W ph ` kqph, kq.
(7.8)
Definition 7.3. A representation of the CCR (canonical commutation relations) is a

projective representation of a Hilbert space H with multiplier given by (7.2). The C algebra generated by a representation W of the CCR is denoted by W pHq. Thus W pHq
is the norm-closed linear span of the unitaries tW phq : h P Hu. The representation of the
CCR defined by (7.7) is called the Fock representation. A representation W of the CCR is
said to be non-singular if the map t W pthq is weakly continuous on R for each h P H,
or equivalently, if W is strongly continuous on all finite-dimensional subspace of H. In this
case, by Stones theorem, there is for each h P H a self-adjoint operator Rphq, called a field
operator, such that W phq exp iRphq.
Remark 7.4.
1. The Fock representation is non-singular.
2. It is sometimes instructive to regard the field operators Rphq as the random variables of a non-commutative probability theory. They satisfy, at least formally, the
commutation relation
RphqRpkq RpkqRphq iIm hh, ki 1,
as a consequence of the W phq satisfying (7.8).
1
3. Defining the annihilation operator aphq by aphq 2 2 pRphq ` iRpihqq, and the
1
creation operator a phq by a phq 2 2 pRphq iRpihqq, we have
aphqa pkq a pkqaphq hk, hi 1.
4. The Weyl operator W phq eiRphq can be written in terms of annihilation and
creation operators as follows:
2
1
}h}
W phq exp i2 2 a phq exp i2 2 aphq exp
.
4
Definition 7.5. A representation W of the CCR over H is said to be cyclic if there
exists a (unit) vector P HW s.t.
HW
tW phq : h P Hu.
We then call the vacuum vector of the representation. The generating functional of
a cyclic representation W with vacuum vector is the function defined on H by
h phq hW phq, i .
Remark 7.6.
1. We shall see (Theorem 7.9) that the Fock representation is irreducible; hence every non-zero vector is cyclic. In particular, the Fock vacuum vector
is cyclic.
32
2. The generating functional is useful for the calculation of the expectation values of
various operators (such as polynomials in the field operators, in the case of nonsingular representations) in the vacuum state of a cyclic representation. For a
non-singular representation the generating functional is given by

phq eiRphq , ,
analogous to the characteristic function of a probability distribution. The analogy
will be strengthened in Theorem 7.8.
A generalization of the notion of cyclic representation has proved useful:
Definition 7.7. Let H, K be Hilbert spaces; a representation W of the CCR over H is
said to be K-cyclic if there exists a V P BpK, HW q s.t.
HW
tW phqV k : h P H, k P Ku.
Let pW, V q be a K-cyclic representation of the CCR over H, and define a map M : H
BpKq by
h M phq V W phqV.
Then M is called the generating function of pW, V q.
The following theorem, which is simply a projective version of the Naimark-Sz.-Nagy
representation theorem for groups, provides a characterization of generating functions:
Theorem 7.8. Let H, K be Hilbert spaces, and M a map from H into BpKq. Then there
exists a K-cyclic representation pW, V q having M as its generating function iff the kernel
h, k M pk hqpk, hq
(7.9)
is positive-definite on H H. In this case pW, V q is uniquely determined up to unitary

equivalence; the representation W is non-singular iff the map t M ph ` tkq is weakly
continuous on R for all h, k P H.
Proof. Let M be the generating function of a K-cyclic representation, pW, V q; then
M pk hqpk, hq V W phq W pkqV,
and so (7.9) is a positive-definite kernel. Conversely, suppose the kernel (7.9) is positivedefinite with a minimal Kolmogorov decomposition V pq, so that
V phq V pkq M pk hqpk, hq
for all h, k P H. Then, for all h, h1 , h2 P H, we have
V ph ` h2 q V ph1 ` h2 qph1 , h2 qph, h2 q M ph1 hqph ` h2 , h1 ` h2 qph1 , h2 qph, h2 q
M ph1 hqph1 , hq,
V phq V ph1 q.
Thus, by the uniqueness of the minimal Kolmogorov decomposition, there exists a welldefined unitary W ph2 q s.t.
W ph2 qV phq V ph ` h2 qph, h2 q.
It is readily seen that W is a representation of the CCR over H, with cyclic map V V poq,
s.t. M is the generating function of pW, V q. The remainder of the proof is clear.
33
Thus we see that the Fock representation of the CCR is determined by the generating
functional
}h}2
h hW phq, i exp
.
(7.10)
4
More generally, we have:
Theorem 7.9. For each 1 there exists a cyclic representation W of the CCR over
H, acting on a Hilbert space F pHq, with cyclic vector , and generating functional
given by
}h}2
phq exp
.
(7.11)
4
The representation W is irreducible.
Proof. We can check directly that is positive-definite, and then apply Theorem 7.9.
Alternatively, we can write down a cyclic representation W having (7.11) as generating
functional. We choose the second approach. Let J be a conjugation on H (that is, an
antilinear map satisfying J 2 1 and hJh, Jh1 i hh1 , hi for all h, h1 P H). Given 1,
choose , beta 0 s.t. 2 ` 2 , 2 2 1, and put
W phq W phq b W pJhq.
(7.12)
Then W , defined on
F pHq F pHq b F pHq,
is a cyclic representation of the CCR with cyclic vector b . Az esay calculation
shows that
}h}2
hW phq , i exp
.
4
To show that W is an irreducible representation for each 1, it is enough (by a tensor
product argument) to show this for the case where H is a one-dimensional Hilbert space,
which we identify with C or R2 . In this case, consider the Schrodinger representation of the
CCR over C, defined on L2 pRq as follows:
rW px, yqgspsq e
ixp2s`yq
2
gps ` yq
(7.13)
for g P L pRq. One verifies that this defines a representation of the CCR over C; moreover,
1
s2
by considering the cyclic vector psq 4 e 2 , one can see that the Schrodinger representation has the same generating functional (7.10) as the Fock representation; so that the
representations are unitarily equivalent. We show that the Schrodinger (7.13) on L2 pRq is
irreducible; a similar argument will show that W , given by (7.12), is an irreducible representation of the CCR over C on L2 pRq b L2 pRq L2 pR2 q.
Let T be an element of W pCq1 , where W is the Schrodinger representation (7.13). Then,
in particular, T commutes with W px, 0q for all x P R. But W px, 0q is multiplication by
the function s eixs ; a density argument shows that T commutes with multiplication
by an arbitrary bounded measurable function. In other words, T is in the commutant of
L8 pRq. But L8 pRq is a maximal abelian von Neumann algebra (0.3); hence T is itself a
multiplication operator. Moreover, T commutes with W p0, yq for all y P R. But W p0, yq is
a translation operator, and so T must be multiplication by a constant function; hence the
Schrodinger representation is irreducible.
CHAPTER 8
Representation of the Canonical Anti-Commutation Relations

In the Chapter 7 we studied the CCR field operators Rphq through their exponentials
W phq eiRphq . This was done for technical convenience, since the Rphq are necessarily
unbounded. Nevertheless, this procedure carries a bonus: the generating functions are very
useful in computations. In this chapter we turn to canonical anti-commutation relations,
where the situation is very different; the field operators are necessarily bounded, and there
is no useful analog of a generating function. However, there is an associated projective
representation of a discrete group (Theorem 8.6) which will prove useful in Chapter 9.
Definition 8.1. Let H be a Hilbert space. A representation of the canonical anticommutation relations over H is a conjugate linear map a from H into the bounded linear
operators on some Hilbert space, which satisfies the canonical anti-commutation relation
(CAR):
apf q apgq ` apgqapf q hf, gi 1,
(8.1)
apf qapgq ` apgqapf q 0,
(8.2)
for all f, g P H. The norm closure of the linear span of the monomials in taphq : h P Hu
and taphq : h P Hu is a C -algebra denoted by ApHq. As a Banach space, ApHq is linearly
generated by the Wick monomials
aph1 q aphn q aphn`1 q aphn`m q,
with h1 , . . . , hn`m P H, or alternatively, by the anti-Wick monomials
aph1 q aphn qaphn`1 q aphn`m q .
Remark 8.2. It follows from (8.1) that }aphq} }h}, since aphqaphq 0 so that
aphq aphq }h}2 . Consequently, h aphq isautomatically continuous. Moreover, if tfn u is
an orthonormal basis for H, we have aphq hfn , hi apfn q in the sense of norm convergence,
so that aphq can be recovered from the an , where an apfn q. (For notational convenience,
we assume that H is separable, but this is not necessary.)
Trivial computation yield:
Lemma 8.3. Let tan uN
n1 satisfy the discrete version of the CAR:
an am ` am an nm ,
(8.3)
an am ` am an 0.
(8.4)
For each n 0, put

then tUn u2N
n1
U2n1 ipan an q, U2n an ` an ;

is a sequence of unitaries satisfying
Un Um ` Um Un 2mn 1.
34
(8.5)
(8.6)
8. REPRESENTATION OF THE CANONICAL ANTI-COMMUTATION RELATIONS
35
Conversely, if tUn u2N

n1 is a sequence of unitaries satisfying (8.6), then the sequence tan
1
pU2n ` iU2n1 q : n 1, . . . , N u satisfies the relations (8.3) and (8.4).
2
Before going further, we look at an example: the Fock representation of the CAR.
Example 8.4. Let f, h1 , . . . , hn be elements of a Hilbert space H.
spanth1 , . . . , hn u and put f f1 ` f2 , where f1 is in L and f2 in LK . Then
Let L
f ^ h1 ^ ^ hn f2 ^ h1 ^ ^ hn ,
and so, by considering determinants,
}f ^ h1 ^ ^ hn }2 }f2 }2 }h1 ^ ^ hn }2 }f }2 }h1 ^ ^ hn }2 .
a
Thus there is a well-defined linear map, denoted by apf qn , from Hna to Hn`1
s.t.
apf qn ph1 ^ ^ hn q f ^ h1 ^ ^ hn ,
(8.7)
and
}apf qn } }f } .
Hence we can define a bounded linear operator apf q : F pHq F pHq which extends the
family tapf qn u. Now let f be a unit vector in H, and put M tf uK . Then apf qn maps
a
,
Mna (regarded as a subspace of Hna ) isometrically onto f ^ Hna and annihilates f ^ Hn1
a
a
a
the orthogonal complement of Mn in Hn . Thus, apf q maps F pM q isometrically onto
F a pHq a F a pM q and annihilates F a pHq a F a pM q. That is, apf q apf q ` apf qapf q 1, or
more generally, apf q apf q ` apf qapf q hf, f i 1, for all f P H. So by polarization
apf q apgq ` apgqapf q hf, gi 1
for all f, g P H. We also have
apf qapgq ` apgqapf q 0
for all f, g P H, since f ^ g ` g ^ f pf ` gq ^ pf ` gq 0. The representation of the CAR
determined by (8.7) is called the Fock representation.
Theorem 8.5. The Fock representation of the CAR is irreducible.
Proof. Consider the state (called the Fock state) on the algebra ApHq given by the
cyclic Fock vacuum vector : pxq hx, i. The Fock vacuum vector is annihilated by
every Wick monomial except the identity. Thus, if is any state on ApHq with , we
have px xq }x}2 0 for every wick monomial except the identity. Thus, by the Schwarz
inequality, annihilates every Wick monomial except the identity, and so clearly , and
so is a pure state.

Finally, we show how to transform a representation of the Car so that it looks like a
representation of the CCR.
Theorem 8.6. Let H be a Hilbert space, let a be a representation of the CAR over H,
and let ApHq be the C -algebra generated by a. Then there exists a projective representation
2N
of the group Z2 , where N dim H, which also generates ApHq.

1
8. REPRESENTATION OF THE CANONICAL ANTI-COMMUTATION RELATIONS
36
Proof. For notational convenience we will assume that H is separable, but this is not
necessary. Let thn : n 1, . . . , N u be an orthonormal basis for H, and put an aphn q.
Then, by Lemma 8.3, there is a sequence tUn u of unitaries which determine the an . If
2N
g tgn : n 1, . . . , N u is an element of G Z2 , gn 0 unless n is in a finite set on which

1
gn 1; define Ug for g P G by
Ug
2N
Ungn .
(8.8)
n1
Then we have
Ug Ug1 bpg, g 1 qUg`g1 ,
(8.9)
where b is a multiplier taking values 1. Thus ApHq is generated by the projective representation (8.8) of the discrete group G.
CHAPTER 9
Slawnys Theorem
In this chapter we study projective representation of groups, in order to prove that two
representations of the CCR (or of the CAR) over a fixed Hilbert space generate isomorphic
C -algebras.
Definition 9.1. Consider a locally compact abelian group G with continuous multiplier b. Throughout this chapter, we will restrict attention to strongly continuous brepresentations. This will involve no less of generality, since in applications the group G
is given the discrete topology. Let B be the map from G into the unitary operators on
L2 pGq given by
rBpgqf spg 1 q bpg 1 , gqf pg 1 ` gq.
Then B is a strongly continuous b-representation called the b-regular representation. It is
unitary, because the inner product on L2 pGq is taken with respect to Haar measure on
G, which is translationally invariant. The regular representation R of G is the b-regular
representation in the particular case in which bp, q 1.
Lemma 9.2. Let G be a locally compact abelian group, and b a continuous multiplier
for G. Let U be a strongly continuous b-representation for G on a Hilbert space H. Then
the b-representations R b U and B b 1H are unitarily equivalent, where R is the regular
representation, and B the b-regular representation.
Proof. Identify L2 pGq b H with L2 pG; Hq, as in 0.5. Define the unitary operator A on
L pG; Hq by pAf qpgq Ug f pgq; then a straightforward computation yields
2
A pR b U q pB b 1H q A.

p of continDefinition 9.3. Let G be a locally compact abelian group; then the space G
uous characters on G can be endowed with the structure of a locally compact abelian group.
p which on L1 X L2 is
The Fourier transform is the unitary map f fp of L2 pGq onto L2 pGq,
given by
fppq
f pgqpgqdg,
G
where dg is Haar measure on G. The Fourier transform implements a unitary equivalence

p of G on L2 pGq
p
between the regular representation R of G on L2 pGq and the representation R
given by
pg hqpq pgqhpq
pR
p
for all P G.
Lemma 9.4. Let G be a locally compact abelian group, and b a continuous multiplier for
p b U and the C -algebra generated
G. The the C -algebra generated by the b-representation R
by the b-regular representation B are isomorphic.
37
9. SLAWNYS THEOREM
38
Proof. The representation B and B b 1H generate isomorphic C -algebras; thus the

result follows from the remarks following Definition 9.3 and from Lemma 9.2, since unitarily
equivalent representations generate isomorphic C -algebras.

Definition 9.5. Let G be a locally compact abelian group, and b a continuous multiplier
p called the natural map,
for G. Then there is a canonical homomorphism from G into G,
given by
g phq bpg, hqbph, gq1 .
p is injective; then pGq is dense in
Lemma 9.6. Suppose that the natural map : G G
p
G.
Proof. Put H pGq; then pHq^ G{H 0 , where H 0 is the annihilator in G of H (or
p
of its closure H). But H0 t0u, since is injective, and so H G.

Lemma 9.7. Let G be a locally compact abelian group, and let b be a continuous multiplier
p is injective. Let U be a strongly continuous
for G, s.t. the associated natural map of G into G
b-representation of G; then the C -algebra generated by U is isomorphic to the C -algebra

p b U.
generated by R
p
Proof. We will show that there is an isomorphism of the C -algebra generated by RbU
p
onto the C -algebra generated by U s.t.
f pgqpR b U qg f pgqUg , for each function f
on G with finite support. The problem is to show that this map is well-defined. We have
p b U qg ess. sup f pgqpgqUg ,

f pgqpR
p
PG
ess. sup f pgqh pgqUg , by Lemma 9.6

hPG
ess. sup f pgqUh Ug Uh , since U is a b-representation,

hPG

ess. sup Uh
f pgqUg Uh
hPG
f pgqUg .

Putting together the conclusions of Lemma 9.4 and 9.7, we have:
Theorem 9.8. Let G be a locally compact abelian group, and b a continuous multiplier
p is injective. Let U 1 and U 2 be strongly
for G s.t. the associated natural map : G G
continuous b-representations of G, and let A1 and A2 be the C -algebras which they generate.
Then there exists a unique isomorphism form A1 onto A2 s.t. pUg1 q Ug2 .
We now apply Theorem 9.8 to the case in which G is a Hilbert space H; we give it the
discrete topology in order to make it locally compact.
Example 9.9. Take G to be a Hilbert space H endowed with the discrete topology, and
define b : H H C by
bpg, hq exppiIm hg, hi {2q.
p is given by
Then b is a multiplier; the associated natural map h h of G into G
h pgq exppiIm hh, giq,
9. SLAWNYS THEOREM
39
and is clearly injective.

Thus from Theorem 9.8 we have
Theorem 9.10. Let H be a Hilbert space, and let W 1 and W 2 be representations of the
CCR over H; let W 1 pHq and W 2 pHq be the C -algebras which they generate. Then there
exists a (necessarily unique) isomorphism : W 1 pHq W 2 pHq s.t.
rW 1 phqs W 2 phq
for each h P H.
2N
Example 9.11. Take G to be Z2 and b to be the multiplier defined in (8.9). Then the
1
natural map h h is given by h pgq p1q ij hi gj , and this is injective.

Thus from Theorem 9.8 we have:
Theorem 9.12. Let H be a Hilbert space and let a1 and a2 be representations of the CAR
over H; let A1 pHq and A2 pHq be the C -algebras which they generate. Then there exists a
(necessarily unique) isomorphism : A1 pHq A2 pHq s.t.
ra1 phqs a2 phq
for each h P H.
CHAPTER 10
Completely Positive Maps on the CCR Algebra

Now that we have completed the construction of the C -algebras of the CCR and CAR
over a Hilbert space H, we turn to the study of their morphisms, known as quasi-free maps,
which are induced by morphisms of the Hilbert space H. In this chapter we treat the CCR
algebra W pHq.
The following simple fact will prove to be useful:
Theorem 10.1. Let H be a Hilbert space, B a C -algebra, and M a map from H into
B. Then there exists a completely positive map T : W pHq B such that T pW phqq M phq
for all h P H, iff the following kernel is positive-definite on H H:
h, k M pk hqpk, hq.
Proof. The result follows from Theorems 7.8 and 9.10. Alternatively, noting that W pHq
is the closed linear span of the unitaries tW phq : h P Hu, one can argue as in 2.

The following is the most general result on quasi-free completely positive maps which we
will need:
Theorem 10.2. Let H, K be Hilbert spaces, A a linear map from H into K, and f a
map from H into C. Then there exists a completely positive map T : W pHq W pKq such
that
T rW phqs W pAhqf phq
for all h in H, if and only if the following kernel is positive-definite on H H:
h, k f pk hq
pk, hq
.
pAk, Ahq
(10.1)
Proof. Define M : H W pKq by M phq W pahqf phq. Then for all h, k P H we have
M pk hqpk, hq W pAhq W pAkqf pk hq
pk, hq
.
pAk, Ahq
Thus if the kernel (10.1) is positive-definite then so is the kernel h, k M pk hqpk, hq, and
the existence of the required completely positive map T is a consequence of Theorem 10.1.
Conversely, if the kernel h, k M pk hqpk, hq is positive-definite, it has a Kolmogorov
decomposition V pq, so that
f pk hq
pk, hq
W pAhqV phq V pkqW pAkq ,
pAk, Ahq
and the result follows.

40
10. COMPLETELY POSITIVE MAPS ON THE CCR ALGEBRA
41
In Theorem 7.9 we noted that for each Hilbert space H, and each 1, there exists a
cyclic representation pW , q of the CCR over H, with generating functional given by
#
+
}h}2
phq hW phq , i exp
.
4
(The Fock generating functional is got by putting 1.) The representation W acts on the
space F pHq and is irreducible. We will denote by W pHq the concrete C -algebra generated
by the representation W . Since ph kq phq pkq, it follows that we can identify
F pH Kq with F pHq b F pKq, and W ph kq with W phq b W pkq, and hence W pH Kq
with the spatial C -tensor product (0.5), written W pHq b W pKq, which is the C -algebra
generated by the algebraic tensor product W pHq d W pKq.
Theorem 10.3. Let 1 be fixed. Let H, K be Hilbert spaces; for each contraction
T : H K there is a completely positive map W pT q : W pHq W pKq of C -algebras s.t.
W pT qrW phqs W pT hqe 4 t}h}

for all h P H. Moreover, W is functorial:
W pST q W pSqW pT q;
}T h}2 u
(10.2)
W p1q 1.
It has the additional properties:

W pS T q W pSq b W pT q,
W p0q is the state determined by .
Proof. We apply Theorem 10.1, checking that the kernel which appears is positivedefinite, to prove that W pT q is completely positive. The rest of the proof is straightforward.

Corollary 10.4. The generating functional is invariant under W pT q for each contraction T .
Remark 10.5. In the case in which 1 (the Fock representation), there is a connection
between the functor W and the Fock functor F . To see this, recall that to each contraction
T : H K there corresponds a contraction F pT q : F pHq F pKq s.t.
F pT qW phq F pT qCphq
1 }h}2
F pT q Exp 2 2 h e 4
1 }h}2
Exp 2 2 T h e 4
CpT hqe
t}h}2 }T h}2 u
4
W pT hqe
t}h}2 }T h}2 u
4
But we have seen that there is a completely positive map W pT q s.t.

t}h}2 }T h}2 u
4
.
W pT qrW phqs W pT hqe
Thus, for all h P H, we have
F pT qW phq W pT qrW phqs.
42
There is an analogous contraction F pT q in the general case in which 1:

Theorem 10.6. Let 1 be fixed. Let H, K be Hilbert spaces; for each contraction
T : H K there is a contraction F pT q : F pHq F pKq s.t.
F pT qW phq W pT hq e 4 t}h}
}T h}2 u
(10.3)
for all h P H. Moreover, F is functorial:

F pST q F pSqF pT q;
F p1q 1.
It has the additional properties:

F pT q F pT q,
F pS T q F pSq b F pT q,
F p0q is the projection on the vacuum.
Proof. For each x P W pHq we have
}W pT qrxs }2 hW pT qrx sW pT qrxs , i
hW pT qrx xs , i
hx x , i
by the Schwarz inequality (Theorem 1.14)
by the invariance of Corollary 10.4
}x }2 .
Hence there is a well-defined contraction F pT q : F pHq F pKq s.t. F pT qrx s
W pT qrxs for all x P W pHq. The only remaining assertion which is not immediately apparent is that F pT q F pT q . This can be verified by calculating
hF pT qW phq , W pkq i and hW phq , F pT qW pkq i, using the definitions.

Thus we have a functor W from the category of Hilbert spaces and contractions to
the category of unital C -algebras and completely positive identity-preserving maps, and a
functor F on the category of Hilbert spaces and contractions; the functors W and F are
related by the following result:
Theorem 10.7. Let 1 be fixed. Let T : H K be a contraction; then the map
x W pT qrxs F pT qxF pT q
from W pHq into BpF pKqq is completely positive. We have W pT q F pT qpqF pT q iff T
is a co-isometry. Moreover, we have
pW pT qrxsyq pxW pT qrysq
for all x P W pHq and y P W pKq.
Proof. Suppose W pT q F pT qpqF pT q ; then, by evaluating at the identity, we see
that F pT T q F pT qF pT q 1, and so T T 1. Conversely, if T T 1, we can show
that W pT q F pT qpqF pT q by using (10.2) and (10.3) to evaluate W pT qrW phqsW pkq
and F pT qW phqF pT q W pkq for all h, k P H. Now let T : H K be a contraction; then
there exists a Hilbert space L abd isometries V1 : H L and V2 : K L s.t. T V2 V1 .
Then we have the following Stinespring decomposition for W pT q :
W pT q W pV2 V1 q W pV2 qW pV1 q
F pV2 q pW pV1 qrsqF pV2 q,
(10.4)
43
since V2 is a co-isometry. Moreover, we have F pT q F pV2 q F pV1 q; thus it is enough

to prove that W pT q F pT qpqF pT q is completely positive when T is an isometry. An
isometry can be factored into a unitary and an injection, and so it is enough to consider the
case in which T is the canonical injection T : H H H 1 , for some Hilbert space H 1 . In
this case we have W pT qrxs x b 1, for each element x P W pHq, where 1 is the identity on
F pH 1 q. On the other hand, we have F pT q b , for each P F pHq, where is the
vacuum vector in F pH 1 q. Thus we have
x W pT qrxs F pT qxF pT q x b p1 eq,
where e is the projection on , and the map x x b p1 eq is completely positive.
Finally, for all x P W pHq and y P W pKq, we have
pW pT qrxsyq hW pT qrxsy , i
hy , W pT qrx s i
hy , F pT qx i
hxF pT qy , i
pxW pT qrysq.

Remark 10.8. In the course of the proof we obtained a Stinespring decomposition (10.4)
for W pT q; if we identify H with a subspace of L, we have
W pT qrxs F pV2 q px b 1qF pV2 q
for all x P W pHq, and so W pT q has an ultraweak extension to a completely positive map on
BpF pHqq (which is, in fact, W pHq2 since the representation W is irreducible, by Theorem
7.9). Thus the ultraweak extension W pT q : BpF pHqq BpF pKqq is unique.
Remark 10.9. We have constructed a C -algebra W pHq b W pKq by taking the spatial
tensor product. It is interesting to note that the CCR-algebra is nuclear: given any C algebra B there is a unique way of completing the -algebra W pHq d B to get a C -algebra.
Theorem 10.10. For any Hilbert space H, the CCR algebra W pHq is nuclear.
Proof. Showing thar W pHq is nuclear is equivalent (see [CE77]) to showing that the
weak closure of the CCR algebra in any representation is injective (that is, given any representation W of the CCR, there is a projection of norm one of BpHw q onto W pHq2 ). But a von
Neumann algebra is injective iff its commutant is injective (see [CE77]). Thus, given any
representation W of the CCR, we seek a projection of norm one from BpHw q onto W pHq1 .
If h is an element of H, let phq denote the automorphism of BpHw q given by
phqx W phq xW phq
for all x P BpHw q. Then is a representation of the abelian group H on BpHw q; but any
abelian group is amenable (see [Gre69]), so there exists an invariant mean M for H. Then
N M rpqs is a projection from BpHw q onto the fixed algebra of , namely W pHq1 . Thus
W pHq1 is injective, and the result follows.

Remark 10.11. The CAR algebra ApHq is nuclear. If H is a finite-dimensional Hilbert
space, say of dimension n, then ApHq can be identified with the full matrix algebra M2n pCq.
It follows that, for any infinite-dimensional Hilbert space H, the CAR algebra ApHq is
44
uniformly hyperfinite (that is, it is an inductive limit of full matrix algebras), and hence is
nuclear (see [CE77]).
CHAPTER 11
Completely Positive Maps on the CAR Algebra

The results on quasi-free completely positive maps on the CAR are not as extensive as
those on the CCR algebra, because of the lack of any useful analogue of the generating
functional. However, we have the following analogue of Theorem 10.3:
Theorem 11.1. Let T : H K be a contraction between Hilbert spaces; then there exists
a completely positive map ApT q : ApHq ApKq, whose action on Wick monomials is given
by
aph1 q aphm q aphm`1 q aphm`n q apT h1 q apT hm q apT hm`1 q apT hm`n q (11.1)
Moreover, A is functorial:
ApST q ApSqApT q;
Ap1q 1.
We have the additional property:

Ap0q is the Fock state.
Proof. First, let T : H K be an isometry; then the map h apT hq is a representation of the CAR. Hence there is a faithful homomorphism ApT q : ApHq ApKq s.t.
ApT qraphqs apT hq.
Next, let T : H K be a co-isometry. Consider the completely positive map of ApHq
into ApKq given, in the Fock representation, by
x F pT qxF pT q ;
direct calculation on a total set of vectors in Fock space shows that, on Wick monomials, we
have
F pT qaph1 q aphm`n qF pT q apT h1 q apT hm`n q.
Finally, let T : H K be a contraction; then there exists a Hilbert space L and
isometries V1 : H L, V2 : K L s.t. T V2 V1 . Put
ApT qrxs F pV2 q ApV1 qrxsF pV2 q
(11.2)
for all x P ApHq; then ApT q is a completely positive map whose action on Wick monomials
is given by (11.1). The remaining assertions follow from this.

Remark 11.2. The relation between the functors A and W can be seen formally as
follows: we have
ff
2
1
}h}
2
2
W phq exp
exp i2 aphq exp i2 aphq .
4
The right-hand side is a sum of Wick monomials and, applying the rule of the A-functor to
them, we have
2
2
1
W phq W pT hqe 4 t}h} }T h} u ,
45
11. COMPLETELY POSITIVE MAPS ON THE CAR ALGEBRA
46
as for the W -functor.

In the Fock representation the functor A and F are related as follows:
Theorem 11.3. For each contraction T : H K between Hilbert spaces we have
F pT qx ApT qrxs
for all x P ApHq. We have ApT q F pT qpqF pT q iff T is a co-isometry, and ApHq is a
homomorphism iff T is an isometry. Moreover, for the Fock state we have
pApT qrxsyq pxApT qrysq
for all x P ApHq and y P ApKq.
Proof. As for Theorem 10.7.
CHAPTER 12
Dilations of Quasi-free Dynamical Semi-groups

We now use the Hilbert space dilation theory which we describe in Chapter 3, together
with the quasi-free completely positive maps constructed in Chapter 11 and 12, to obtain
examples of dilation of dynamical semi-groups at the C -algebraic level.
Example 12.1. Let tTt : t 0u be a strongly continuous semi-group of contractions on
a Hilbert space H. Then, by Theorem 3.2, there is an isometric embedding V of H into
another Hilbert space K, on which there is a semi-group tUt : t 0u of unitaries s.t.
Tt V Ut V,
t 0.
Hence, for each 1, there is a strongly continuous semigroup tW pTt q : t 0u of completely positive maps on W pHq s.t.
W pTt q W pV qW pUt qW pV q,
t 0.
Now W pV q isa an embedding of W pHq as a C -subalgebra of W pKq, and W pV q is a

conditional expectation of W pKq onto W pHq. Furthermore,
W pUt q F pUt qpqF pUt q
is a unitarily implemented group of automorphisms of W pKq. If we identify H as a subspace
of K, we have
W pTt qrxs p1 b qpF pUt qpx b 1qF pUt q q, t 0,
for all x P W pHq. In particular, we have
W pTt qrW phqs W pTt qe 4 t}h}
}Tt h}2 u
t 0, for all h P H.
Example 12.2. Let tTt : t 0u be a semi-group of isometries on a Hilbert space H.
Then, by Theorem 3.1, we have the stronger dilation
V Tt Ut V,
t 0.
In this case, at the C -algebraic level we have

W pV qW pTt q W pUt qW pV q,
t 0;
(12.1)
identifying H as a subspace of K, we have

W pTt qrxs b 1 F pUt qpx b 1qF pUt q ,
t 0,
for all x P W pHq. This is a very strong form of dilation: it transforms the semi-group of
homomorphisms tW pTt q : t 0u into the unitarily implemented group of automorphisms
tW pUt q : t P Ru.
47
12. DILATIONS OF QUASI-FREE DYNAMICAL SEMI-GROUPS
48
Example 12.3. Let tTt : t 0u be a semi-group of contractions on a Hilbert space H,

s.t. there is an isometric embedding V of H into a Hilbert space K on which there is a
strongly continuous semi-group of isometries tGt : t 0u and
V Tt Gt V,
t 0.
(In Chapter 16 we will show that such a co-isometric dilation exists for certain semi-groups.)
For the CCR algebra, we have the following interesting isometric representation:
W pV qW pTt q F pGt q W pV qrsF pGt q,
t 0;
identifying H with a subspace of K, this gives

W pTt qrxs b 1 F pGt q W pV qrxsF pGt q,
t 0,
for all x P W pHq.

Analogous results hold for the CAR algebra. In the remaining chapters we will be
concerned with finding dilations of more general dynamical semi-groups on operator algebras.
We notice, by using a crossed-product construction, that a dilation of the type (12.1) exists
trivially for any semi-group of homomorphisms. In the C -algebra case, this method gives a
dilation of a family of completely positive maps the subject of the next chapter.
CHAPTER 13
Dilation of Completely Positive Maps on C -algebras

In Chapter 12 we gave some examples of dilations in a C -algebraic setting. We now
take a more abstract approach. We show that a family of completely positive maps on a
C -algebra can be dilated to a group of C -automorphisms on a larger C -algebra.
Theorem 13.1. Let A be a unital C -algebra of operators on a Hilbert space H; let
tTg : g P Gu be a family of completely positive maps Tg : A A, indexed by the elements
of a locally compact group G, and strongly continuous in the sense that g Tg pxq is norm
continuous for all x P A and P H. Suppose that Te 1 and Tg p1q 1 for all g P G. Then
there exists a C -algebras B on a Hilbert space K, a strongly continuous representation U
of G on K s.t. Ug BUg B for all g P G, an isometric -homomorphism i : A B, and a
conditional expectation N of B onto A s.t.
Tg rxs N pUg irxsUg q
for all g P G and x P A.
Proof. Let H 1 L2 pG; Hq, and define a completely positive map T : A BpH 1 q by
pT rxsf qpgq Tg rxsf pgq.
Let U 1 be the strongly continuous unitary representation of G on H 1 , defined by pUg1 f qphq
f phgq, and let A1 be the C -algebra generated by T pAq and U 1 pGq. Let tf u be an L2 approximate identity on G; for each , define an isometric embedding V : H H 1 by
pV qpgq f pgq.
Then displaystyle lim8 V aV exists in the weak operator topology for all a P A1 , and
lim V Ug1 T rxsUg1 V Tg rxs.
Since T is completely positive, there exists a representation i of A on a Hilbert space K, and

an isometry V : H 1 K, s.t. T rxs V irxsV for all x P A, and i is faithful since
lim V V irxsV V x
for all x P A. Let Ug be the strongly continuous unitary representation of G on K defined

by
Ug V Ug1 V ` 1 V V
for all g P G. Let B be the C -subalgebra of BpHq generated by the set tUg irxsUg : g P
G, x P Au. Then we have V BV A1 ; thus, for each x P B, the limit N pxq lim V V xV V
8
exists in the weak operator topology, and

Tg rxs N pUg irxsUg q
for all x P A and g P G.

49
CHAPTER 14
Generators of Dynamical Semigroups

In this chapter we examine the generators of norm-continuous one-parameter semigroups
of positive maps and, in particular, of completely positive maps on C -algebras. We sharpen
the well-known result for reversible process: derivations generate automorphism groups.
Recall that a derivation on an algebra A is a map L, whose domain DpLq is a subalgebra
of A, s.t.
Lpabq Lpaqb ` aLpbq
for all a, b P A.
Theorem 14.1. Let tetL : t 0u be a strongly continuous semigroup on a Banach algebra
A. Then etL for each t 0 is a homomorphism iff L is a derivation.
Proof. Let L be a derivation, let x, y be elements of DpLq, and put
f ptq etL pxyq etL pxqy xetL pyq,
t 0,
then t f ptq is continuously differentiable,

f 1 ptq LetL pxyq LetL pxqetL pyq etL pxqLetL pyq,
t 0,
and for h P DpLq we have

d ptsqL
e
phq eptsqL Lh,
ds
0 s t.
Thus we have
t
tL
d ptsqL
e
f psq ds
0 ds
t
t
ptsqL
Lf psqds ` eptsqL f 1 psqds
e
0
0
t
(
eptsqL L esL pxqesL pyq LesL pxq esL y esL pxq LesL pyq ds
f ptq e f p0q
0,
since L is a derivation.
Thus if f ptq is identically zero, we have

etL pxyq etL pxqetL pyq,
t 0,
for all x, y P DpLq. The result follows, since DpLq is dense in A. The proof of the converse
is trivial.

Next we need analogous results for the generators of positive semigroups on C -algebras.
First recall that if S is a set of states on a C -algebra A, then S is said to be full if f pxq 0
for all f in S implies that x 0 whenever x is a self-adjoint element of A. Moreover, if f
50
14. GENERATORS OF DYNAMICAL SEMIGROUPS
belongs to S implies that y
51
f px yxq
belongs to S for all x P A s.t. f px xq 0, then f is
f px xq
said to be invariant.
Theorem 14.2. Let L be a bounded self-adjoint linear map on a unital C -algebra A.
Then the following conditions are equivalent:
(i) etL is positive for all positive t.
(ii) p Lq1 is positive for all sufficiently large positive .
(iii) If y is in A` , then ya 0 implies a Lpyqa 0.
(iv) For some full, invariant set of states S: if f P S and y P A` , then f pyq 0 implies
f pLpyqq 0.
(v) Lpx2 q ` xLp1qx Lpxqx ` xLpxq for all self-adjoint x P A.
(vi) Lp1q ` u Lp1qu Lpu qu ` u Lpuq for all unitary u P A.
Proof. (iv) (iii) Let S be a full, invariant set of states satisfying (iv); let y P A` and
a P A be s.t. ya 0. Then f pa yaq 0 for all f P S. Hence, by (iv) and the invariance of
S, we have f pa Lpyqaq 0 for all f P S, and so a Lpyqa 0 since S is full.
(iii) (ii) Let be greater than }L}. In order to show that p Lq1 0, it is enough
to show that x 0 whenever x is self-adjoint and p Lqx 0. Let x x` x with x`
and x positive and x` x 0. Then, by (iii), we have x Lpx` qx 0, so that

`
0 x 1 1 pxq x
x xx 1 x Lpxqx
px q3 1 x Lpx` qx ` 1 x Lpx qx .
3
Thus 0 px q3 1 x Lpx qx , and so }x }3 1 }L} }x1 } , since }a} }b} whenever

0 a b. Hence x 0, since 1
1.
}L}
n
t
tL
(ii) (i) We have e lim 1 L
.
n8
n
2
(i) (v) Let K Lp1q
, and put L2 pxq Kx ` xK. Then etL pxq etk xetK , so that
2
2
tetL ; t P Ru is a group of positive maps. Applying the Lie-Trotter formula to L1 L ` L2 ,
1
we have etL 0 for all t 0. Using Kadisons Schwarz inequality (Corollary 4.4) and the
1
1
1
fact that etL p1q 1, we have etL px2 q retL pxqs2 for t 0. Differentiating at t 0, we have
L1 px2 q L1 pxqx ` xL1 pxq for all self-adjoint x P A, and so the result follows on substituting
L1 L ` L2 .
1
1
2
(v) (iv) Let y be in A` , f P A` with f pyq 0. Then f y z f zy 2 0 for all
z P A, by the Schwarz inequality. Hence
1 1
1
1
1
1
2
2
2
2
2
Lpyq ` y Lp1qy L y y ` y L y 2
implies that f pLpyqq 0.
(i) (vi) By the reduction employed above, it is enough to prove
this
when Lp1q 0.
(i) (vi) Since etL 0 and etL p1q 1 for all t 0, we have etL 1 for all t 0.
Thus etL puq 1 for all unitaries u P A and all t 0. Hence etL pu qetL puq 1 for all t 0;
differentiating this inequality at t 0, we have Lpu qu ` u Lpuq 0 for all unitaries u P A.
(vi) (i) Since we have assumed that etL p1q 1 for all t 0, it is enough (by 0.4) to
}1 ` tL} 1
prove that etL is a contraction for all t 0. By 0.1, this is the case if lim
0.
t0
t
52
Moreover
}1 ` tL} supt}u ` tLpuq} : u unitaryu
(see 0.2). But if u is unitary and t 0, we have
}u ` tLpuq}2 1 ` trLpu qu ` u Lpuqs ` t2 Lpuq Lpuq
1 ` t2 Lpuq Lpuq
1 ` t2 }L}2 .
1
Thus }1 ` tL} r1 ` t2 }L}2 s 2 , and so

1
r1 ` t2 }L}2 s 2 1
}1 ` tL} 1
lim
0;
lim
t0
t0
t
t
hence etL is a contraction for each t 0.
A self-adjoint linear map on a C -algebra is automatically continuous if it satisfies condition (v) of Theorem 14.2; we prove the following:
Theorem 14.3. Let L be a self-adjoint linear map on a unital C -algebra A, with the
following property:
if y P A` , f P A` , and f pyq 0, then f pLpyqq 0.
tL
Then L is bounded, and so e
(14.1)
is positive for all t 0.
Proof. The map x Lpxq 21 rLp1qx ` xLp1qs satisfies condition (14.1) whenever L
does, so we may assume that Lp1q 0. We will show that, in this case, L is dissipative on
Ah (in the sense of 0.1):
}x} }x Lx} for all x P Ah and 0.
(14.2)
In order to prove thus for some self-adjoint x, we may assume that there exists a positive
f P A s.t. f pxq }x} and }f } 1. Then f p}x} xq 0, and so f pLp}x} xqq 0; that is,
we have f pLpxqq 0. Let be strictly positive, then f pxq f px Lxq }f } }x Lx}.
Hence }x} }f } }x Lx} for all self-adjoint x P A. It follows that L is closed on Ah ,
and so L is bounded: Let tfn P Ah u be a sequence satisfying fn 0, Lfn g; then for all
h P Ah , and lambda 0, we have
}fn ` h} }p Lqpfn ` hq} .
Letting n 8, we have }h} }ph gq Lphq}; as 8 we have }h} }h g} for
all h P Ah . Hence g 0. It then follows that etL is positive for all t 0. Alternatively,
this follows from (14.2) which shows that p1 1 Lq1 is a contraction for all }L}, and
hence is positive since it preserves the identity (see 0.4).

The result listed in Theorem 14.2 relate mainly to the Jordan structure of a C -algebra,
but they will be used to prove a result about its C -structure (Theorem 14.4). First we
consider an example: let A Mn pCq and let Lpxq xt x (where x xt is the transpose
mapping); then L satisfies the hypotheses of Theorem 14.3, but not those of Theorem 14.4.
Theorem 14.4. Let L be a bounded self-adjoint linear map on a C -algebra A. Then the
following conditions are equivalent:
1. etL px xq etL px qetL pxq, t 0, for all x P A.
2. Lpx xq Lpx qx ` x Lpxq for all x P A.
53
Proof. 1. 2. This follows by differentiating the inequality in 1. at t 0.

2. 1. Suppose 2. holds; adjoin an identity 1 to A, and extend L to the enlarged
algebra by putting Lp1q 0. Then, by Theorem 14.2, etL is positive on the enlarged algebra
for all t 0. Fix x P A and define
f ptq etL px xq etL px qetL pxq,
t 0.
Then
f 1 ptq LetL px xq LetL px q etL pxq etL px q LetL pxq ,

so that
t
d ptsqL
e
f psq ds
0 ds
t
t
d
ptsqL
Lf psqds ` eptsqL f psqds
e
ds
0
0
t
(
eptsqL LresL px qesL pxqs LesL px q esL pxq esL px q LesL pxq ds.
0
But, by hypothesis, L esL px qesL pxq LesL px q esL pxq ` esL px q LesL pxq for all x P A
and s 0. Moreover, eptsqL is positive for 0 s t; hence f ptq etL f p0q 0 for all t 0.
This means that
etL px xq etL px qetL pxq, t 0,
for all x P A.

tL
f ptq e f p0q
Before we go on to prove some characterizations of the generators of norm-continuous

one-parameter semigroup of completely positive maps on C -algebras, we will give a result
which has a slightly more general setting, and which we will need in the proof of Theorem
15.1.
Lemma 14.5. Let A be a C -subalgebra of a C -algebra B, and let L : A B be a
self-adjoint bounded linear map. Then the following conditions are equivalent:
1. For all a P A, the kernels
s, t Lps a atq ` s Lpa aqt Lps a aqt s Lpa atq
are positive-definite on A A.
2. The kernels
ps1 , s2 q, pt1 , t2 q Lps1 s2 t2 t1 q ` s1 Lps2 t2 qt1 Lps1 s2 t2 qt1 s1 Lps2 t2 t1 q
are positive-definite on pA Aq pA Aq.
3. The following holds for all n P N:
n
bi Lpai aj qbj 0
i,j1
for all a1 , . . . , an P A and b1 , . . . , bn P B for which
n
i1
ai bi 0.
Proof. 1. 2. This is trivial.
1. 3. Let a1 , . . . , an P A and b1 , . . . , bn P B satisfy ni1 ai bi 0. Then for all a P A

we have
bi Lpai a aaj q ` ai Lpa aqaj Lpai a aqaj ai Lpa aaj qbj 0;

i,j
thus
54
bi Lpai a aaj qbj 0,
i,j
since
ai b i
bi ai 0.
Taking a to be an approximate identity for A, we have
bi Lpai aj qbj 0.
i,j
3. 2. Suppose c1 , . . . , cn , e1 , . . . , en P A and f1 , . . . , fn P B are arbitrary. Define

#
ci
, 1 i n,
ai
cin ein , n i 2n,
and
#
ei fi
bi
fin
Then
2n
i1
, 1 i n,
, n i 2n.
ai bi 0, so that
2n
bi Lpai aj qbj 0;
i,j1
substituting for ai and bi , we have

n
n
n
n

fi Lpei ci cj ej qfj `
fi ei Lpci cj qej fj
fi Lpei ci cj qej fj `
fi ei Lpci cj ej qfj .
i,j1
Thus 2. holds.
i,j1
i,j1
i,j1
Definition 14.6. Let A be a C -subalgebra of a C -algebra B. A linear map L : A B

is said to be conditionally completely positive if it satisfies the conditions of Lemma 14.5.
We conclude this chapter with a characterization of the generators of quantum dynamical
semigroups:
Theorem 14.7. Let L be a self-adjoint bounded linear map on a C -algebra A. Then L
is conditionally completely positive iff etL is completely positive for all t 0.
Proof. Suppose L is conditionally completely positive; then L satisfies condition 1. of
Lemma 14.5. By going to the second dual (if necessary) we can assume that A is unital;
then, taking a 1, the result follows from the implication (iii) (i), and the converse
follows from the implication (i) (iii)
CHAPTER 15
Canonical Decomposition of Conditional Completely Positive

Maps
In Chapter 14 we gave a characterization of the generators of norm-continuous oneparameter semigroups of completely positive maps: they are the conditionally completely
positive maps, characterized by certain inequalities. For a large class of von Neumann
algebras, a more detailed description of conditionally completely positive maps can be given,
in terms of a canonical decomposition (Theorem 15.1). This result can be stated using
a cohomology theory for operator algebras, and one is tempted at this point to introduce
all the machinery of cohomology; resisting the temptation, we make use instead of a little
shorthand. Let A be a von Neumann subalgebra of a von Neumann algebra B; we write
H 1 pA, Bq 0 if the following is true: If W : A B is a derivation (that is, a linear map s.t.
x in B s.t. W pxq W
x x xW
x
W pxyq W pxqy ` xW pyq for all x, y P A), then there exists W
for all x P A.
Let A be a von Neumann subalgebra of a von Neumann algebra B, and let L : A B
be a -linear map s.t. both L and L are conditionally completely positive:
Lpa b cdq ` a Lpb cqd Lpa b cqd ` a Lpb cdq
for all a, b, c, d P A. Putting
1
L0 pxq Lpxq tLp1qx ` xLp1qu
2
for all x P A, we see that L0 is a derivation of A into B; if H 1 pA, Bq 0, there exists a
self-adjoint h in B s.t. L0 i Ad h. Hence we have Lpxq k x ` xk for all x P A, where
k 21 Lp1q ` ih. Conversely, if k is an element of B, then the map L : A B given by
Lpxq k x ` xk is such that both L and L are conditionally completely positive. It is
trivial that a completely positive map is conditionally completely positive. We are now ready
to describe the canonical decomposition for conditionally completely positive maps.
Theorem 15.1. Let A be a W -algebra. Then the following conditions on A are equivalent:
(i) Whenever A is faithfully represented as a W -algebra on a Hilbert space H we have
H 1 pA, BpHqq 0.
(ii) Whenever A is faithfully represented as a W -algebra on a Hilbert space H, and
L : A BpHq is a conditionally completely positive ultraweakly continuous -linear
map, there exists k P BpHq and a completely positive map : A BpHq s.t.
Lpxq pxq ` k x ` xk
for all x P A.
Proof. (i) (ii) Let A be faithfully represented on a Hilbert space H, and let L : A
BpHq be a -linear ultraweakly continuous map such that if D is the tri-linear map defined
55
15. CANONICAL DECOMPOSITION OF CONDITIONAL COMPLETELY POSITIVE MAPS
56
by
Dpx, y, zq Lpxyzq ` xLpyqz Lpxyqz xLpyzq
for all x, y, z P A, then the map pa1 , a2 q, pb1 , b2 q Dpa1 , a2 b2 , b1 q is positive-definite on
pA Aq pA Aq. Then, by the results of Chapters 1 and 2, there exists a Hilbert
space K, a normal representation of A on K, and a linear
map V : A BpH, Kq, s.t.
Dpx, y, zq V px q pyqV pzq for all x, y, z P A, and K tpaqV pbqh : a, b P A, h P Hu.
Then, for all x, y, a, b P A, we have
V px q pyqrV pabq paqV pbq V paqbs Dpx, y, abq Dpx, ya, bq Dpx, y, aqb 0.
Hence, by minimality of K, we have
V pabq paqV pbq ` V paqb
for all a, b P A. Let denote the following normal faithful representation of A on H K:
a
0
paq
,
0 paq
where we identify elements of BpH Kq with 2 2 matrices in the obvious way. Let W be
the following linear map of paq into BpH Kq:
0
0
W ppaqq
.
V paq 0
Then W ppaqpbqq paqW pabq ` W ppaqqpbq for all a, b P A. Hence, since H 1 ppaq, BpH
Kqq 0, there exists
p q
x
W
r s
x paq paqW
x.
in BpH Kq s.t. W ppaqq W
In particular, V paq paqr ra for all a P A. Then, for all x, y, z P A, we have
Lpxyzq ` xLpyqz Lpxyqz xLpyzq Dpx, y, zq V px q pyqV pzq
rpx qr rx s pyqrpzqr rzs
pxyzq ` xpyqz pxyqz xpyzq,
where is the completely positive map a r paqr. From the discussion preceding the
statement of the theorem, and since H 1 pA, BpHqq 0, we see that there exists K in BpHq
s.t. Lpxq pxq ` k x ` xK for all x P A.
(ii) (i) Let A be faithfully represented on a Hilbert space H, and let L : A BpHq be
a derivation. Put k0 eLpeqLpeqe, where e is the identity of A, and define L0 : A BpHq
by L0 pxq Lpxq k0 x xk0 ; then L0 peq 0. Thus, without loss of generality, we may
assume that Lpeq 0, and that L is a -map. Hence, by condition (ii), there is an element
k P BpHq, and a completely positive map : A BpHq, s.t. Lpxq pxq ` k x ` xk
for all x P A. Take a minimal Stinespring decomposition pxq r pxqr, where is a
representation of A on a Hilbert space K and r is an element of BpH, Kq. Then, as above,
we have
0 Lpxyzq ` xLpyqz Lpxyqz xLpyzq
rpx qr rx s pyqrpzqr rzs
15. CANONICAL DECOMPOSITION OF CONDITIONAL COMPLETELY POSITIVE MAPS
57
for all x, y, z P A. Hence we have pzqr rz for all z P A; in particular, putting p 12 peq,
we have pzq pz `zp for all z P A. But we can assume that ek k, so that k `k `peq
Lpeq 0 and k ` p k p h, say. Then we have
Lpxq pxq ` kx ` xk
1
1
p peq ` k qx ` xpk ` peqq
2
2
hx xh
for all x P A, so that H 1 pA, BpHqq 0.
Remark 15.2. Let A be a von Neumann algebra on a Hilbert space H, and let tTt : t 0u
be a norm-continuous semigroup of completely positive normal maps on A. Then it follows
from Theorems 14.7 and 15.1 that, under suitable conditions on the algebras, there exists
k P BpHq and : A BpHq a completely positive normal map s.t. the generator L of Tt
is given by
Lpxq pxq ` k x ` xk
for all x P A. If Tt preserves the identity of A, then Lp1q 0 and so k ` k 12 p1q 0.
Hence k is the generator of a contraction semigroup, tBt : t 0u say, on H. Let tSt : t 0u
be the contraction semigroup on BpHq given by St pxq Bt xBt for all x P BpHq. The
generator of St is the map x k x ` xk; by Banach space perturbation theory we have
t
Tt pxq St pxq ` Sts Ts pxqds, t 0,
0
for all x P A.
More generally, we make the following definition:
Definition 15.3. Let A be a von Neumann algebra on a Hilbert space H. A dynamical
semigroup of Lindblad type on A is a weakly continuous semigroup tTt : t 0u of normal
completely positive unital maps s.t. there exists a strongly continuous contraction semigroup
tBt : t 0u on H, and a completely positive normal map : A BpHq, s.t.
t
Tt pxq St pxq ` Sts Ts pxqds, t 0,
0
for all x P A, where St pxq
Bt xBt .
Remark 15.4. A dynamical semigroup of Lindblad type on A has an extension to a

dynamical semigroup of Lindblad type on BpHq.
CHAPTER 16
Isometric Representation of Quantum Dynamical Semigroups

In Chapter 3 the problem of dilating was considered at the Hilbert space level. The result
were used in Chapter 12, together wit the CAR and CCR functors, to obtain examples of
dilations of dynamical semigroups at the C -algebra level. We now begin consideration of
the general problem of dilating dynamical semigroups. As in the Hilbert space situation
(Chapter 3), and thus in the CAR and CCR algebras (Chapter 12), there are various ways
of formulating the concept of a dilation. The first general form which we treat for arbitrary
operator algebras is the isometric representation version (compare 12.2 and 12.3).
Theorem 16.1. Let A be a von Neumann algebra on a Hilbert space H, and let tTt : t
0u be a weakly continuous dynamical semigroup of Lindblad type on A. Then there exists a
Hilbert space K and a strongly continuous semigroup tGt : t 0u of isometries on H b K,
s.t.
Tt pxq b 1 Gt px b 1qGt , t 0,
for all x P A.
Proof. We can assume (see Chapter 15) that A BpHq, and that there exists a
contraction semigroup tBt : t 0u on H, and a normal completely positive map V on
BpHq, s.t.
t
Tt pmq St pmq ` Sts V Ts pmqds, t 0,
(16.1)
0
for all m P BpHq, where St pmq Bt mBt . The pre-adjoint semigroup Tt and St on the
pre-dual T pHq satisfy
t
t0
(16.2)
Tt pq St pq `
Ts V Sts pqds,
0
for all P T pHq. By Theorem 4.6, there exists a family tAx : x P Xu of bounded operators
on H s.t.
V pmq
Vx pmq, Vx pmq Ax mAx ,
(16.3)
x
for all m P BpHq. Because of the particular form (16.3) of the perturbation V , we can write
the Neumann series for (16.1) and (16.2) in an unfamiliar, but useful way.
Let X8 be the set
# of all sequences
+ tpxi , ti q P X p0, 8q : 0 t1 u, regarded as a
8
m

Borel subset of
X p0, 8q in an obvious way. Let Y8 be the Borel subset of X8
m0
n1
consisting of all sequences of finite length, and for each t 0 let Xt be the Borel subset of
X8 given by all finite sequences tpxi , ti q : 0 t1 tn tu. For each t 0, there is a
Borel isomorphism t : Xt X8 Y8 defined by
tpxi , ti quni1 , tpyj , sj qum
j1 px1 , t1 q, . . . , pxn , tn q, py1 , s1 ` tq, . . . , pym , sm ` tq.
58
16. ISOMETRIC REPRESENTATION OF QUANTUM DYNAMICAL SEMIGROUPS
59
The inverse map is given by

tpyi , si quni1 py1 , s1 q, . . . , pyp , sp q, pyp`1 , sp`1 tq, . . . , pyn , sn tq,
where p is the unique integer s.t. sp t sp`1 . We denote by X0 the subset consisting
of the single sequence z of zero length. We define the measure t on Xt to be the product
measure constructed from counting measure on each component X, and Lebesgue measure
on each p0, 8q; we assign Dirac measure to the point z P Xt . We define a measure 8 on
X8 in an analogous fashion. For each w P Xt , define p S V Sqpwq by
p S V Sqpwq St1 Vx1 St2 t1 Vx2 Vxn Sttn
where w tpxi , ti q : 0 t1 tn tu. Then the Neumann series
t
p St1 V Stt1 q pqdt1
Tt pq St pq `
0
t t1
p St1 V St2 t1 V Stt1 q pqdt1 dt2
`
0
(16.4)
`
can be written as
Tt pq
p S V Sqpwqpqdt pwq,
(16.5)
rp S V Sqpwqs pmqdt pwq.
(16.6)
Xt
and the adjoint series as
Tt pmq
Xt
We take K to be L2 pY8 q, and define the operator Gt on L2 pY8 ; Hq for t 0:

pGt f qpwq pBABqpwtqf pwt q,
1
where pwt, wt q 1
t pwq; the element pBABqpw q of BpHq is given by
pBABqpw1 q Bt1 Ax1 Bt2 t1 Ax2 Axn Bttn ,

for each w1 tpxi , ti q : 0 t1 tn tu in Xt .
We prove that Gt is a strongly continuous semigroup of isometries on L2 pY8 ; Hq. First
we check that Gt is an isometry, by using (16.6), and by observing that the measure 8 is
the product of the measures t and 8 under the Borel isomorphism t : Xt Y8 Y8 .
That is,
hGt f, Gt f i
hpBABqpwtqf pwt q, pBABqpwtqf pwt qi d8 pwq
Y8
hrp S V Sqpwtqs p1qf pwt q, f pwt qi d8 pwq

Y8
hrp S V Sqpwtqs p1qf pwt q, f pwt qi dt pwtqd8 pwt q

Y8 Xt
hTt p1qf pwt q, f pwt qi d8 pwt q

Y8
hf pwt q, f pwt qi d8 pwt q hf, f i .

Y8
16. ISOMETRIC REPRESENTATION OF QUANTUM DYNAMICAL SEMIGROUPS
60
Here we have used the normalization condition Tt p1q 1. Next we show that tGt : t 0u
is a semigroup. Indeed, we have
pGt1 pGt2 f qqpwq pBABqpwt1 qpGt2 f qpwt1 q
pBABqpwt1 qpBABqpwt1 ,t2 qf pwt1 ,t2 q
pBABqpwt1 `t2 qf pwt1 `t2 q pGt1 `t2 f qpwq,
where we have used the following immediate consequences of the definitions:
pBABqpwt1 qpBABqpwt1 ,t2 q pBABqpwt1 `t2 q,
wt1 ,t2 wt1 `t2 .
Now that we have shown that tGt : t 0u is a semigroup of isometries, it is enough
to verify that it is weakly continuous at t 0 on elements of the algebraic tensor product
L2 pY8 q d H; this we do by noting that t tXt ztzuu tet . Finally, we derive the isometric
representation property of Gt : taking x P BpHq and f P L2 pY8 ; Hq, we have
hGt px b 1qGt f, f i hpx b 1qGt f, Gt f i
hxpBABqpwtqf pwt q, pBABqpwtqf pwt qi d8 pwq
Y8
hp S V Sqpwtqs pxqf pwt q, f pwt qi d8 pwq

Y8
hrp S V Sqpwtqs pxqf pwt q, f pwt qi dt pwtqd8 pwt q
Y8 Xt
hTt pxqf pwt q, f pwt qi d8 pwt q, by (16.6),

Y8
hpTt pxq b 1qf, f i .

The theorem follows.
Theorems 16.1 and 15.1 together show that all norm-continuos dynamical semigroups on
a large class of W -algebras possess isometric representations. We have, as a by-product,
the following Hilbert space dilation theorem (mentioned in Chapter 3, and used in Chapter
12 to dilate some quasifree dynamical semigroups):
Theorem 16.2. Let H be a Hilbert space, and h a self-adjoint (possibly unbounded)
operator on H; let k be a positive bounded operator on H. Then there exists an isometric
embedding W of H into a Hilbert space K and a strongly continuous semigroup tGt : t 0u
of isometries on K, s.t.
W epihkqt Gt W, t 0.
(16.7)
Proof. Let Tt be the dynamical semigroup of Lindblad type on BpHq constructed from
the contraction semigroup Bt epihkqt on H, together with the completely positive map V
1
1
given by V pxq k 2 xk 2 for x P BpHq. By Theorem 16.1, there exists a strongly continuous
semigroup Gt of isometries on K L2 pY8 ; Hq such that Tt pxq b 1 Gt px b 1qGt . Consider
the isometric embedding W of H in K given by f z b f ; then, by the definition of Gt ,
we have pGt f qpzq Bt f pzq. Thus W Gt Bt W , and so Gt W W Bt .
CHAPTER 17
Unitary Dilations of Dynamical Semigroups

In Chapter 16 we obtained isometric representations for semigroups of Lindblad type
on W -algebras. We now investigate unitary dilations of such semigroups, using Coopers
unitary dilation of isometric semigroups (Theorem 3.1). In order to carry through the
construction we need to place further restrictions on either the algebra or the semigroup.
In the first place, we can handle injective von Neumann algebras; for simplicity, we give a
detailed discussion for BpHq only.
Theorem 17.1. Let H be a Hilbert space, and let tTt : t 0u be a weakly continuous
dynamical semigroup of Lindblad type on BpHq. Then there exists a von Neumann algebra
M on a Hilbert space L, an embedding e of BpHq as a von Neumann subalgebra of M ,
a conditional expectation N of M onto BpHq, and a strongly continuous unitary group
tUt : t P Ru on L, s.t.
Ut M Ut M @t P R,
and
Tt pmq N rUt epmqUt s,
t 0,
for all m P BpHq. Moreover, we have

Bt mBt N rUt epmqUt s,
t 0,
for all m P BpHq.

Proof. We use the notation of Theorem 16.1. Let tGt : t 0u be the semigroup of
isometries s.t. 1 b Tt pmq Gt p1 b mqGt for all m P BpHq. By Coopers Theorem (Theorem
3.1), there exists a Hilbert space, an isometric embedding W2 : L2 pY8 ; Hq L, and a
strongly continuous unitary group tUt : t P Ru on L s.t.
W2 Gt Ut W2 ,
t 0.
(17.1)
Let e1 : BpHq 1 b BpHq BpL2 pY8 q b Hq be the canonical embedding x 1 b x,

and let e2 : BpL2 pY8 ; Hqq BpLq be the embedding given by e2 pxq W2 W2 . Define a
conditional expectation N2 of BpLq onto BpL2 pY8 ; Hqq by N2 pxq W2 W2 . Let W1 be the
isometry from H into L2 pY8 ; Hq given by W1 f z bf ; then the map x N1 pxq W1 W1
is a conditional expectation of BpL2 pY8 ; Hqq onto BpHq.
[ Warning: e1 pq W1 pqW1 .] Finally, we take M to be BpLq, embedding BpHq in M
with e e2 e1 , and projecting back onto BpHq with N N1 N2 . For x P BpHq and t 0,
we have
N pUt epxqUt q W1 W2 Ut W2 p1 b xqW2 Ut W2 W1
W1 Gt p1 b xqGt W1
W1 1 b Tt pxqW1 Tt pxq.
61
17. UNITARY DILATIONS OF DYNAMICAL SEMIGROUPS
62
On the other hand, for y P B pL2 pY8 ; Hqq, we have

N pUt e2 pyqUt q W1 W2 Ut W2 yW2 Ut W2 W1
W1 Gt yGt W1 ,
Bt N1 pyqBt ,
by (17.1),
by (16.7).
This is more than enough to prove the theorem.
Remark 17.2. In the course of the proof of Theorem 17.1 we noted that the embedding
e1 : BpHq BpL2 pY8 ; Hqq given by x 1 b x is distinct from the embedding e11 given by
x W1 xW1 Pz b x, where Pz is the projection in L8 pY8 q given by the characteristic
function of the singleton tzu P Y8 . However, it turns out that the embedding e11 has its
uses, and that N1 is a conditional expectation w.r.t. e11 (as well as w.r.t. e1 ). Moreover, for
x P BpHq and t 0, we have
N1 pGt e1 pxqGt q W1 Gt W1 xW1 Gt W1
Bt xBt St pxq,
so that
St pxq N1 rGt pPz b xqGt s N rUt e2 pPz b xqUt s,
while
Tt pxq N1 rGt p1 b xqG1 s N rUt e2 p1 b xqUt s.
More generally, for each Borel subset E of Y8 and its associated projection PE P L2 pY8 q, we
have
N rUt e2 pPE b xqUt s N1 rGt pPE b xqGt s
rp S V Sqpwqs pxqdt pwq.

EXXt
Thus we have a simultaneous dilation of the Markov kernels in Davies noncommutative

probability theory.
In Theorem 17.1 we constructed an automorphism group, namely tUt pqUt : t P Ru, on
the algebra BpLq, which projects onto the given dynamical semigroup tTt : t 0u on BpHq.
In order to treat a von Neumann subalgebra M of BpHq which is globally invariant under
Tt , we must either project from BpHq onto M , or work with some subalgebra of BpLq. To
follow the first alternative, we need the concept of an injective von Neumann algebra.
Definition 17.3. A von Neumann algebra M is injective if, whenever M is embedded as
a von Neumann subalgebra of another von Neumann algebra M 1 , there exists a conditional
expectation (not necessarily normal) of M 1 onto M .
Thus we see that weakly continuous dynamical semigroups of Lindblad type on injective
von Neumann algebras possess unitary dilations in the sense of Theorem 17.1.
However, it is known (see [EL77]) that a von Neumann algebra is injective iff it is
hyperfinite (assuming the underlying Hilbert space is separable). Thus in general the first
alternative in not feasible. Turning to the second alternative, we seek a von Neumann
subalgebra M of BpLq which is at least invariant under tUt pqUt : t 0u, and contains
epM q. We also employ the following device: we do not attempt to project M directly onto
1 b M via the map N2 , but rather onto some algebra B b M , where B is a judiciously chosen
von Neumann subalgebra of BpL2 pY8 qq. The following diagram may clarify matters:
1bM
e1
BbM
BpL2 pY8 ; Hqq

N
2
M
1bM
BbM
This programme is performed in the following theorem:
e2
63
BpLq
Theorem 17.4. Let H be a separable Hilbert space; let tTt : t 0u be a weakly continuous
dynamical semigroup of Lindblad type on BpHq, so that there exists a strongly continuous
contraction semigroup Bt on H, and an ultraweakly continuous completely positive linear
map V on BpHq, s.t.
t
Tt pmq St pmq ` pTts V Ss qpmqds,
0
with St pmq
Bt mBt .
Suppose that V has a decomposition
V pmq
Ax mAx dpxq,
X
where pX, q is a -finite measure space, and x Ax is weakly measurable. If M is a von

Neumann algebra on H s.t.
Ax lies in M for almost all x P X,
(17.2)
Bt M Bt M @t 0,
(17.3)
then the dynamical semigroup tTt : t 0u on M has a unitary dilation. That is, there
exists a von Neumann algebra M on a Hilbert space L, a strongly continuous unitary group
tUt : t P Ru on L, an embedding e P M as a von Neumann subalgebra of M , and a normal
conditional expectation N of M onto M s.t.:
Ut M Ut M
Tt pmq N rUt epmqUt s
@t 0,
@m P M, t 0.
(17.4)
(17.5)
Proof. For clarity, we give the details of the proof for the case where M BpHq and
is a counting measure. We employ the notation and construction used in the proof of
Theorem 16.1; thus we have a strongly continuous isometric semigroup tGt : t 0u on
L2 pY8 ; Hq, and an isometric embedding W2 of L2 pY8 ; Hq into a Hilbert space L on which
there is a strongly continuous unitary group tUt : t P Ru, s.t. W2 Gt Ut W2 for t 0. Take
B to be the commutative von Neumann algebra L8 pY8 q, and take M 1 to be L8 pY8 ; M q,
which is a W -algebra with predual M1 L1 pY8 ; M q. The mapping f b a f pqa has a
unique extension to a W -isomorphism of L8 pY8 qbBpHq onto L8 pY8 ; M q (see 0.5).
Put M tUt e2 pM 1 qUt : t 0u2 , where e2 : M 1 BpLq is again defined as e2 pxq
W2 xW2 . We will show that N2 pM q M 1 where N2 : M BpL2 pY8 ; Hqq is defined as
N2 pxq W2 xW2 . For this, it is convenient to have the explicit form of the action of Gt on
a vector f . We get this by inspecting hGt g, f i for arbitrary g:

hGt g, f i
hpBABqpwtqgpwt q, f pt pwt, wt qqi dt pwtqd8 pwt q
Y8 Xt

hgpwt q, rpBABqpwtqs f pt pwt, wt qi dt pwtqd8 pwt q.

Y8
Xt
64
Hence Gt is given by
pGt f qpwq
rpBABqpw1 qs f pt pw1 , wqqdt pwq.
Xt
In what follows we use wt to denote t pw1 , wq, where w1 is a variable of integration running
through Xt ; we remark that wtt w1 , and wtt w. We claim that N2 pM q M 1 . For t 0
and x P M 1 , we have
N2 pUt e2 pxqUt q W2 Gt W2 xW2 Gt Wt
Gt xGt .
We take x P L8 pY8 ; M q and compute Gt xGt as an element of BpL2 pY8 ; Hqq, and show that
it lies in L8 pY8 ; M q; we have
pGt xGt f qpwq
rpBABqpw1 qs pxGt f qpwt qdt pw1 q
Xt
rpBABqpw1 qs xpwt qpBABqpwtt qf pwtt qdt pw1 q
Xt
rpBABqpw1 qs xpwt qpBABqpw1 qf pwqdt pw1 q
Xt
rp S V Sqpw1 qs xpwt qdt pw1 qf pwq.
Xt
Thus pGt xGt qpwq Xt rp S V Sqpw1 qs xpwt qdt pw1 q lies in L8 pY8 ; M q, and so Gt M 1 Gt
M 1 . For n 1 and ti 0, i 1, . . . , n, we define an by
an N2 rUt1 e2 px1 qUt1 Ut2 e2 px2 qUt2 Utn e2 pxn qUtn s.
It follows that
an Gt1 x1 Gt2 Gt1 x2 Gt3 Gt2 Gtn1 xn Gtn ,
observing that for all s, t 0 we have W2 Ut Us W2 Gs Gt , as a consequence of Theorem 3.1.
We have to show that an lies in M 1 . In order to state an induction hypothesis we introduce
bn defined by
bn Gt1 x1 Gt2 Gt1 x2 Gt3 Gt2 xn Gtn`1 Gtn ,
and notice that bn tn`1 0 an . By direct calculation of the kind used above, we have

b1 pw1 , w2 ; wqf pwtt1 t2 qdt pw1 qdt pw2 q
pb1 f qpwq
1
2
1
Xt1
Xt2
where
b1 pw1 , w2 ; wq rpBABqpw1 qs x1 pwt1 qrpBABqpw2 qs pBABqpwtt1 t2 q.
1
Suppose that, for n 1, we have
bn pw1 , . . . , wpn`1q ; wqf pwtt1 t2 ttn`1 qdt pw1 q dt pwpn`1q q;

pbn f qpwq
1
n`1
n
1
Xt1
Xtn`1
(17.6)
65
then
pbn`1 f qpwq
bn pw1 , . . . , wpn`1q ; wqpxn`1 G Gt f qpwtt1 t2 ttn`1 q

n`1
tn`2
n
1
Xt1
Xtn`1
dt1 pw1 q dtn`1 pwn`1 q
bn`1 pw1 , . . . , wpn`2q ; wqf pwtt1 t2 ttn`1 ttn`2 q

n
n`1
1
Xt1
Xtn`2
dt1 pw1 q dtn`2 pwn`2 q;

(17.7)
where
bn`1 pw1 , . . . , wpn`2q ; wq bn pw1 , . . . , wpn`1q ; wqxn`1 pwtt1 t2 ttn`1 q
n
1
t
rpBABqpwpn`2q qs rpBABqpwtt11 t2 tn`2

qs.
n`1
But (17.6) holds for n 1, and hence, by (17.7), for all n 1. Evaluating pbn f qpwq at
tn`1 0, we have
1
pnq
bn pw1 , . . . , wpnq , z; wqf pwtt1 t2 tn
q.
pan f qpwq
tn1 tn qdt1 pw q dtn pw
1
Xtn
Xt1
It follows directly from the definitions that wtt11 t2 ttnn1 tn w, so that

pan f qpwq an pwqf pwq,
where
an pwq
bn pw1 , . . . , wpnq , z; wqdt pw1 q dtn pwpnq q.

1
Xt1
Xtn
Thus an lies in M , so that N2 pM q M 1 by continuity. We complete the proof by taking

any conditional expectation N 1 of M 1 onto BpHq. For example, let be a normal state on
L8 pY8 q (that is, is an element of L1 pY8 q) and put N 1 b 1 : L8 pY8 qbM M . (If
we take z , then N 1 paq apzq for a P M 1 ; in fact, in the notation of Theorem 17.1,
the restriction of N1 pq W1 pqW1 to M 1 coincides with N 1 in this case.) We then put
e e2 e1 and N N 1 N2 , and we have
Tt pmq N rUt epmqUt s,
t 0, @m P M.
Remark 17.5. The map t Ut pqUt is weakly continuous. It cannot be normcontinuous, even though t Tt may be, unless Tt is a homomorphism of M . Indeed,
suppose t Tt is strongly continuous with generator L, t Ut pqUt is strongly continuous
with generator , and Z Dpq X M is a core for L, (that is, L pLZ q ). Then for x P Z
we have etL pxq N ret pxqs, and so x is in DpLq and Lpxq N rpxqs. Thus for x, y in the
subalgebra Z, we have
Lpxyq N rpxyqs N rpxqy ` xpyqs
N rpxqsy ` xN rpyqs
Lpxqy ` xLpyq,
and so L is a derivation if Z is a core for L; in this case it follows from Theorem 14.1 that
Tt is a semigroup of homomorphisms.
References
1.
The main result of this chapter is Theorem 1.9. For scalar-valued kernels on Z Z,
it was proven by [Kol41]; he showed that a kernel is the correlation kernel of a stochastic
process iff it is positive-definite ([PS72]). For operator-valued kernels, versions of Theorem
1.9, with various restrictive assumptions on X, can be found in the literature ([Pay64],
[Kun66], [Pon72], [ANW75]).
The idea of using the image-space rather than the quotient-space ([Nai43b]) goes back
to [Aro50]; it has been exploited by [Hal12] and [SU75] for Hilbert space dilations, and
by [Kun66] and [Car78] in group representation theory.
Remarks on the origins of Theorem 1.14 will be found in the notes on Chapter 14.
2.
The dilation theorem for positive-definite functions on groups (Corollary 2.6 is due
to [Nai43a]; it was extended to semigroups by [SNR55]. The canonical decomposition
of a completely positive scalar-valued map (that is, of a state) on a C -algebra is known as
the GNS construction ([GN43], [Seg47]). It was extended by [Sti55] to operator-valued
completely positive maps on unital C -algebras; the original proof was simplified by [Arv69],
and the result was extended to a larger class of unital -algebras by [Pow74]. [Lan76]
obtained the Stinespring decomposition for non-unital C -algebras by going to the second
dual. The result for Banach -algebras with approximate identities (Theorem 2.13) is due to
[Eva75]; for some related results, see [Pas73]. As can be seen from the proof of Theorem
2.13, the Stinespring decomposition for a completely positive map whose domain consists of
a subspace N N , where N is a left ideal in an algebra A, can be obtained in such a way that
it is constructed on the whole of A. This is the decomposition used by [Eva77b] to study
unbounded completely positive maps on C -algebras whose domains consist of hereditary
-subalgebras.
The relationship between the Stinespring decomposition for algebras and the Naimark
dilation for groups has been described several times in the literature [Suc73]. If G is a
locally compact group, there is a canonical bijection between completely positive maps on
L1 pGq and those on C pGq, the enveloping C -algebra of L1 pGq. If G is abelian, C pGq can
p the continuous functions vanishing at
be identified via the Fourier transform with C0 pGq,
p the dual of G.
infinity on G,
3.
The theory of dilation of continuous semigroups began with [Coo47] who discovered
Theorem 3.1; it is interesting to note that his motivation came from quantum mechanics
([Coo50a, Coo50b]). Theorem 3.2, on the dilation of semigroups of contractions, is due to
[Nag53]; it is a powerful tool in Hilbert space theory ([NFBK10]).
The idea of the proof of Theorem 3.8 comes from [SNR55], who discovered the connection between positive-definite functions on Z and -semigroups of contractions indexed
by N. This method was generalized by [Mla65] and [Suc73], and their work is the basis of
our exposition.
66
REFERENCES
67
The construction of a unitary dilation of a contraction semigroup contracting strongly

to zero (Theorem 3.13) is due to [LP90]; this method can be modified to give an alternative
proof of Theorem 3.2 ([NFBK10, 1.10.2]). The abstract Langevin equation in Theorem
3.13 was obtained by [LT74a] in connection with an analysis of the Ford-Kac-Mazur model
([LT75, LT74b]); see also [LP75].
4.
There is an extensive recent literature on completely positive maps on C -algebras
and the tensor-product construction; see the review by [EL77]. The equivalence of (i) and
(ii) in Lemma 4.1(a) occurs in the work of [St74] and [Pas73]. The proof given here
of Lemma 4.1(c) is due to Skau (private communication). [St63] showed that a positive
map from an arbitrary C -algebra into a commutative C -algebra is completely positive; he
used a slightly different method from the one given here (Theorem 4.2). That any positive
map from a commutative C -algebra into an arbitrary C -algebra is completely positive
was shown by [Nai43b, Nai43a], and by [Sti55]. The Schwarz inequality 4.3 in Corollary
4.4 was first obtained for self-adjoint elements by [Kad52], who used an entirely different
method. Corollary 4.4 and its proof were first recorded by [St63, Sto69] along with
the Schwarz inequality of Theorem 1.14 for completely positive maps (with essentially the
same proof as in Chapter 1). For other Schwarz-type inequalities, with various positivity
assumptions, see [Ara60], [Cho74], [Eva76a, EVA77a], [LR74]. Corollary 4.5 is due to
Broise (unpublished), and is recorded by [Sto69]. The proof given here is due to [EHK78]
and uses an observation of [EVA77a].
[Kra71] obtained the canonical decomposition of a normal completely positive map
on the von Neumann algebra of all bounded operators on a Hilbert space. [Cho75] showed
that if, in Theorem 4.6, H and K are finite-dimensional the decomposition can be chosen so
that the cardinality of the set X is at most dimpHq dimpKq.
5.
Conditional expectations on classical probability spaces were characterized by
[M` 54] in terms of positive maps with module property CE2. The study of analogues
of conditional expectations in the non-commutative setting was begun by [Ume54]. A detailed discussion of Examples 5.1.1. and 2. can be found in [Dav76c]: the first arises
in measurement theory, the second in the composition of quantum systems. Theorem 5.3
is due to [Tom57] and Broise (unpublished); the proof given here is taken verbatim from
[Sto69]. The definition of a conditional expectation adopted in this chapter is quite adequate for many purposes in non-commutative probability theory, but not for all; see [DL70]
and [Acc74, Acc76] for more general concepts.
6.-8. These chapters provide an exposition of some of the folk-lore of mathematical physics.
The fundamental paper on Fock space is by [Coo53]. The characterization of a generating
functional of the CCR is due to [Ara60] and to [Seg61] independently; the extension to the
operator-valued case was given by [Eva75]. The extremal universally invariant states (whose
generating functionals are of the form 7.11) were introduced by [S` 62]. Our treatment of
the CAR algebra and its representations is in the spirit of [HK75].
9. The main results of this section are due to [Sla72]; we follow the exposition of [Sim72].
The key Lemma (Lemma 9.2) is due to [Fel62]. The construction employed in its proof
has been used by [How74] and by [Eva76b] in the study of scattering by time-dependent
perturbations.
REFERENCES
68
10. Quasi-free dynamical semigroups associated with representations of the CCR were
investigated in the thesis of [Tho71]; see also [LT75]. In the algebraic context they were
studied by [Dav72b, Dav72a, Dav76a, Dav76b, Dav76c, Dav78] and also by [DVV77,
DVV79], [Emc76] [EAE78], [EL76b], and [Lin76a]. Necessity in Theorem 10.2 was
proved by [EL76b], whilst sufficiency was shown by [DVV77]. In fact, [DVV77] introduce
ph,kq
the multiplier ph, kq pAh,Akq
, and use it to construct a CCR algebra WA pHq over H; they
exploit the fact that the function f of Theorem 10.2 gives rise to a completely positive map
iff it is a generating functional of a state of the algebra WA pHq.
Theorems 10.3, 10.6, 10.7 are an elaboration of the work of [EL76b]. Essentially, the
proof of Theorem 10.10 is due to Strmer (private communication), who uses it to give an
elementary proof of the fact that any type I von Neumann algebra is injective.
11.
Theorem 11.1 appears in [HK75]; for related work see [Nel73b, Nel73a],[SU75].
12. Dilation of quasi-free dynamical semigroups induced by contraction semigroups can

be found in the FKM model ([FKM65],[Tho71], [LT75, LT74b]). They have been studied
in detail by [Dav72a], [Emc76], [EAE78], [EL76a, EL76b].
[Ara71] has shown that a one-parameter unitary group on a Hilbert space gives rise
to a norm-continuous group of automorphisms in the CAR algebra iff the generator at the
Hilbert space level is trace-class. [Dav77b] has studied quasi-free dynamical semigroups on
the CAR algebra in detail; it follows from his work that a contraction semigroup with a
trace-class generator on a Hilbert space induces a norm-continuous dynamical semigroup on
the CAR algebra.
13. Theorem 13.1 and its proof are due to [Dav76a, Dav76b, Dav78]. It is unclear how
to extend the construction to the category of W -algebras. [Eva75, Eva76a] had previously
obtained this result for discrete groups; there is no problem in modifying his construction to
deal with von Neumann algebras.
The C -algebra generated by T pAq and U 1 pGq is a generalization of the C -crossed
product of a C -algebra by a group of automorphisms ([Tur58],[DKR66]).
14. Theorem 14.1 was obtained by [Eva75], who generalized the well-known result for
strongly continuous one-parameter groups. That (i) implies (vi) in Theorem 14.2 is due to
[Tsu77], who observed also that (vi) implies (i) is implicite in the work of [Lin76c]. The
equivalences (i)-(v) of Theorem 14.2, and also Theorems 14.3, 14.4, are due to [EHO79].
Theorem 14.3 is an improvement on the work of [Kis76]; we use [Sul75]s proof of [LP61]
result: a densely defined dissipative linear map is closeable. Theorem 14.4 was first proved
for identity-preserving semigroups on unital C -algebras by [Lin76c]; he used a different
method.
The concept of conditionally completely positive maps was introduced by [EVA77a];
Lemma 14.5 is built on the work of [EVA77a], [Lin76b] and [Dav79]. Theorem 14.7 is a
strengthening of the result of [EVA77a] for unital C -algebras. For the analogous result for
semigroups of positive-definite functions on group, see [PS72].
For earlier work on the generators of dynamical semigroups, and dissipativity, see
[Kos72b, Kos72a, Kos73] and [Kos75]. For recent work on the generators of strongly
continuous dynamical semigroups, see [Dav76a, Dav76b, Dav76c, Dav78]. Note also
the characterization of the generators of positive semigroups in a function space context by
[Sim77] and [HSU77].
REFERENCES
69
15. The canonical decomposition of norm-continuous semigroups of completely positive

normal maps on a von Neumann algebra was first obtained independently by [GFV` 78] for
finite-dimensional matrix algebras, and by [Lin76c] for hyperfinite von Neumann algebras.
The implication (i) (ii) in Theorem 15.1 is an improved version of [Lin76b]; the converse
is due to [EVA77a].
If A is a von Neumann algebra on a Hilbert space H, it is known that H 1 pA, BpHqq 0
if: (i) A is type I or hyperfinite ([Joh72, Rin72]); (ii) A is properly infinite ([Chr78]). It
is widely conjectured that H 1 pA, BpHqq 0 for all von Neumann algebras.
16. 17. These chapters are an improved version of the work of [EL76a] which was inspired
by [Dav72a].
Bibliography
[Acc74]
Luigi Accardi, Noncommutative markov chains, Int. School of Mathematical Physics, Camerino
(1974), 268295.
[Acc76]
Luigi Accardi, Nonrelativistic quantum mechanics as a noncommutative markof process, Advances
in Mathematics 20 (1976), no. 3, 329 366.
[ANW75] Graham Allen, Francis Narcowich, and James Williams, An operator version of a theorem of
kolmogorov, Pacific Journal of Mathematics 61 (1975), no. 2, 305312.
[Ara60]
Huzihiro Araki, Hamiltonian formalism and the canonical commutation relations in quantum field
theory, Journal of Mathematical Physics 1 (1960), no. 6, 492504.
, On quasifree states of car and bogoliubov automorphisms, Publications of the Research
[Ara71]
Institute for Mathematical Sciences 6 (1971), no. 3, 385442.
[Aro50]
Nachman Aronszajn, Theory of reproducing kernels, Transactions of the American mathematical
society (1950), 337404.
[Arv69]
William B Arveson, Subalgebras of c*-algebras, Acta Mathematica 123 (1969), no. 1, 141224.
[Car78]
Alan L Carey, Group representations in reproducing kernel hilbert spaces, Reports on Mathematical Physics 14 (1978), no. 2, 247259.
[CE77]
Man-Duen Choi and Edward G. Effros, Injectivity and operator spaces, Journal of Functional
Analysis 24 (1977), no. 2, 156 209.
[Che74]
Paul R Chernoff, Product formulas, nonlinear semigroups, and addition of unbounded operators,
no. 140, American Mathematical Society, 1974.
[Cho74]
Man-Duen Choi, A schwarz inequality for positive linear maps on c -algebras, Illinois J. Math.
18 (1974), no. 4, 565574.
[Cho75]
Man-Duen Choi, Completely positive linear maps on complex matrices, Linear Algebra and its
Applications 10 (1975), no. 3, 285 290.
[Chr78]
Erik Christensen, Extension of derivations, Journal of Functional Analysis 27 (1978), no. 2, 234
247.
[Coo47]
JLB Cooper, One-parameter semigroups of isometric operators in hilbert space, Annals of Mathematics (1947), 827842.
[Coo50a]
, The characterization of quantum-mechanical operators, Mathematical Proceedings of the
Cambridge Philosophical Society 46 (1950), no. 04, 614619.
[Coo50b]
, The paradox of separated systems in quantum theory, Mathematical Proceedings of the
Cambridge Philosophical Society 46 (1950), no. 04, 620625.
[Coo53]
JM Cook, The mathematics of second quantization, Transactions of the American Mathematical
Society (1953), 222245.
[Dav72a] E B Davies, Some contraction semigroups in quantum probability, Zeitschrift f
ur Wahrscheinlichkeitstheorie und Verwandte Gebiete 23 (1972), no. 4, 261273.
[Dav72b] EB Davies, Diffusion for weakly coupled quantum oscillators, Communications in Mathematical
Physics 27 (1972), no. 4, 309325.
[Dav76a] E Brian Davies, Markovian master equations. ii, Mathematische Annalen 219 (1976), no. 2,
147158.
[Dav76b] EB Davies, The classical limit for quantum dynamical semigroups, Communications in Mathematical Physics 49 (1976), no. 2, 113129.
[Dav76c] Edward Brian Davies, Quantum theory of open systems, IMA, 1976.
[Dav77a] E Brian Davies, Master equations: A survey of rigorous results, Milan Journal of Mathematics
47 (1977), no. 1, 165173.
70
BIBLIOGRAPHY
[Dav77b]
71
EB Davies, Irreversible dynamics of infinite fermion systems, Communications in Mathematical

Physics 55 (1977), no. 3, 231258.
[Dav78]
, Dilations of completely positive maps, Journal of the London Mathematical Society 2
(1978), no. 2, 330338.
[Dav79]
E Brian Davies, Generators of dynamical semigroups, Journal of Functional Analysis 34 (1979),
no. 3, 421432.
[Dix77]
Jacques Dixmier, C*-algebras, North-Holland Publishing Co., 1977.
, von neumann algebras, North-Holland Publishing Co., 1981.
[Dix81]
[DKR66] Sergio Doplicher, Daniel Kastler, and Derek W Robinson, Covariance algebras in field theory and
statistical mechanics, Communications in Mathematical Physics 3 (1966), no. 1, 128.
[DL70]
E Brian Davies and John T Lewis, An operational approach to quantum probability, Communications in Mathematical Physics 17 (1970), no. 3, 239260.
[DS58]
Nelson Dunford and Jacob T Schwartz, Linear operators. part i: General theory, Interscience,
New York, 1958.
, Linear operators. part ii: Spectral theory. self adjoint operators in hilbert space, Inter[DS63]
science, New York, 1963.
[DVV77] Bart Demoen, P Vanheuverzwijn, and Andre Verbeure, Completely positive maps on the ccralgebra, Letters in Mathematical Physics 2 (1977), no. 2, 161166.
[DVV79] B Demoen, P Vanheuverzwijn, and A Verbeure, Completely positive quasi-free maps of the ccralgebra, Reports on Mathematical Physics 15 (1979), no. 1, 2739.
[EAE78] Gerard G Emch, Sergio Albeverio, and Jean-Pierre Eckmann, Quasi-free generalized k-flows,
Reports on Mathematical Physics 13 (1978), no. 1, 7385.
[EHK78] David E Evans and Raphael Hegh-Krohn, Spectral properties of positive maps on c*-algebras,
Journal of the London Mathematical Society 2 (1978), no. 2, 345355.
[EHO79] David E Evans and Harald Hanche-Olsen, The generators of positive semigroups, Journal of
functional analysis 32 (1979), no. 2, 207212.
[EL76a]
David E Evans and John T Lewis, Dilations of dynamical semi-groups, Communications in Mathematical Physics 50 (1976), no. 3, 219227.
[EL76b] DE Evans and JT Lewis, Completely positive maps on some c-algebras, Preprint (1976).
[EL77]
Edward G Effros and E Christopher Lance, Tensor products of operator algebras, Advances in
Mathematics 25 (1977), no. 1, 134.
[Emc76] Gerard G Emch, Generalized k-flows, Communications in Mathematical Physics 49 (1976), no. 3,
191215.
[Eva75]
DE Evans, Some problems in functional analysis., Ph.D. thesis, University of Oxford, 1975.
[Eva76a] David E Evans, Positive linear maps on operator algebras, Communications in Mathematical
Physics 48 (1976), no. 1, 1522.
[Eva76b]
, Time dependent perturbations and scattering of strongly continuous groups on banach
spaces, Mathematische Annalen 221 (1976), no. 3, 275290.
[EVA77a] DAVID E. EVANS, Conditionally completely positive maps on operator algebras, The Quarterly
Journal of Mathematics 28 (1977), no. 3, 271283.
[Eva77b] David E Evans, Irreducible quantum dynamical semigroups, Communications in Mathematical
Physics 54 (1977), no. 3, 293297.
[Fel62]
James MG Fell, Weak containment and induced representations of groups, Canad. J. Math 14
(1962), 237268.
[FKM65] GW Ford, M Kac, and P Mazur, Statistical mechanics of assemblies of coupled oscillators, Journal
of Mathematical Physics 6 (1965), no. 4, 504515.
[GFV` 78] Vittorio Gorini, Alberto Frigerio, Maurizio Verri, Andrzej Kossakowski, and ECG Sudarshan,
Properties of quantum markovian master equations, Reports on Mathematical Physics 13 (1978),
no. 2, 149173.
[GN43]
I.M. Gelfand and M.A. Naimark, On the imbedding of normed rings into the ring of operators in
hilbert space, Matematiceskij sbornik 54 (1943), no. 2, 197217.
[Gre69]
Frederick P Greenleaf, Invariant means on topological groups and their applications, vol. 16, Van
Nostrand Reinhold Co., 1969.
BIBLIOGRAPHY
[Gro55]
[Hal12]
[HK75]
[How74]
[HP57]
[HSU77]
[Joh72]
[Kad52]
[Kat95]
[Kis76]
[Kol41]
[Kos72a]
[Kos72b]
[Kos73]
[Kos75]
[Kra71]
[Kun66]
[Lan76]
[Lin76a]
[Lin76b]
[Lin76c]
[Loo53]
[LP61]
[LP73]
[LP75]
[LP90]
[LR74]
[LT74a]
[LT74b]
72
Alexandre Grothendieck, Produits tensoriels topologiques et espaces nucleaires, no. 16, American
Mathematical Society, 1955.
Paul Richard Halmos, A hilbert space problem book, vol. 19, Springer Science & Business Media,
2012.
NM Hugenholtz and RV Kadison, Automorphisms and quasi-free states of the car algebra, Communications in Mathematical Physics 43 (1975), no. 2, 181197.
James S Howland, Stationary scattering theory for time-dependent hamiltonians, Mathematische
Annalen 207 (1974), no. 4, 315335.
Einar Hille and Ralph Saul Phillips, Functional analysis and semi-groups, vol. 31, American
Mathematical Soc., 1957.
H Hess, R Schrader, and DA Uhlenbrock, Domination of semigroups and generalization of katos
inequality, Duke Mathematical Journal 44 (1977), no. 4, 893904.
Barry Edward Johnson, Cohomology in banach algebras, vol. 127, American Mathematical Soc.,
1972.
Richard V Kadison, A generalized schwarz inequality and algebraic invariants for operator algebras, Annals of Mathematics (1952), 494503.
Tosio Kato, Perturbation theory for linear operators, vol. 132, springer, 1995.
A Kishimoto, Dissipations and derivations, Communications in Mathematical Physics 47 (1976),
no. 1, 2532.
Andre Nikolaevich Kolmogorov, Stationary sequences in hilbert space, Bull. Math. Univ. Moscow
2 (1941), no. 6, 140.
Andrzej Kossakowski, On necessary and sufficient conditions for a generator of a quantum dynamical semi-group, Bull. Acad. Pol. Sci., Ser. Sci. Math. Astr. Phys 20 (1972), no. 12, 10211025.
, On quantum statistical mechanics of non-hamiltonian systems, Reports on Mathematical
Physics 3 (1972), no. 4, 247274.
A Kossakowski, On the general form of the generator of a dynamical semi-group for the spin 1/2
system, Bull. Acad. Pol. Sci., Ser. Sci. Math. Astr. Phys 21 (1973), no. 7, 649653.
R.S Ingarden; A Kossakowski, On the connection of nonequilibrium information thermodynamics
with non-hamiltonian quantum mechanics of open systems, Annals of Physics 89 (1975), 451485.
Karl Kraus, General state changes in quantum theory, Annals of Physics 64 (1971), no. 2, 311
335.
RA Kunze, Positive definite operator-valued kernels and unitary representations, Proc. Conf.
Functional Anal., Irvine, 1966.
Christopher Lance, Tensor products of non-unital c*-algebras, Journal of the London Mathematical Society 2 (1976), no. 2, 160168.
G
oran Lindblad, Brownian motion of a quantum harmonic oscillator, Reports on Mathematical
Physics 10 (1976), no. 3, 393406.
, Dissipative operators and cohomology of operator algebras, Letters in Mathematical
Physics 1 (1976), no. 3, 219224.
Goran Lindblad, On the generators of quantum dynamical semigroups, Communications in Mathematical Physics 48 (1976), no. 2, 119130.
Lynn H Loomis, Introduction to abstract harmonic analysis, Courier Dover Publications, 1953.
G
unter Lumer and Ralph S Phillips, Dissipative operators in a banach space, Pacific J. Math 11
(1961), no. 2, 679698.
J L Lebowitz and O Penrose, Modern ergodic theory, Physics Today 26 (1973), no. 2, 2329.
JT Lewis and JV Pul`e, Dynamical theories of brownian motion, International Symposium on
Mathematical Problems in Theoretical Physics, Springer, 1975, pp. 516519.
Peter D Lax and Ralph S Phillips, Scattering theory, vol. 26, Academic press, 1990.
Elliott H Lieb and Mary Beth Ruskai, Some operator inequalities of the schwarz type, Advances
in Mathematics 12 (1974), no. 2, 269273.
JT Lewis and LC Thomas, A characterization of regular solutions of a linear stochastic differential
equation, Probability Theory and Related Fields 30 (1974), no. 1, 4555.
, How to make a heat bath, Tech. report, Dublin Inst. Adv. Studies, 1974.
BIBLIOGRAPHY
[LT75]
73
J. T. Lewis and L. C. Thomas, On the existence of a class of stationary quantum stochastic

processes, Annales de linstitut Henri Poincar (A) Physique thorique 22 (1975), no. 3, 241248
(eng).
[M` 54]
Shu-Teh Chen Moy et al., Characterizations of conditional expectation as a transformation on
function spaces, Pacific J. Math 4 (1954), 4763.
[Mla65]
W Mlak, Unitary dilations in case of ordered groups, Annales Polonici Mathematici 3 (1965),
no. 17, 321328.
[Nag53]
Bela Sz Nagy, Sur les contractions de lespace de hilbert, Acta Sci. Math.(Szeged) 15 (1953),
8792.
[Nai43a] M. A. Naimark, Positive definite operator functions on a commutative group, Izvestiya Rossiiskoi
Akademii Nauk. Seriya Matematicheskaya 7 (1943), no. 5, 237244.
[Nai43b] Mo A Naimark, On a representation of additive operator set functions, Dokl. Akad. Nauk SSSR,
vol. 41, 1943, pp. 359361.
[Nel73a] Edward Nelson, Construction of quantum fields from markoff fields, Journal of Functional Analysis 12 (1973), no. 1, 97112.
, The free markoff field, Journal of Functional Analysis 12 (1973), no. 2, 211227.
[Nel73b]
[NFBK10] Bela Sz Nagy, Ciprian Foias, Hari Bercovici, and Laszlo Kerchy, Harmonic analysis of operators
on hilbert space, Springer Science & Business Media, 2010.
[Pas73]
William L Paschke, Inner product modules over *-algebras, Transactions of the American Mathematical Society 182 (1973), 443468.
[Pay64]
R Payen, Representation spectrale des fonctions aleatoires stationnaires `
a valeurs dans un espace
de hilbert, COMPTES RENDUS HEBDOMADAIRES DES SEANCES DE L ACADEMIE DES
SCIENCES 259 (1964), no. 22, 3929.
[Pon72]
O.I. Ponomarenko, Correlation kernels of random functions with values in hilbert space, Selected
translations in mathematical statistics and probability (M Arato and AG Kusnirenko, eds.),
vol. 10, The Rosen Publishing Group, 1972.
[Pow74]
Robert T Powers, Selfadjoint algebras of unbounded operators. ii, Transactions of the American
Mathematical Society 187 (1974), 261293.
[PS72]
Kalyanapuram R Parthasarathy and Klaus Schmidt, Positive definite kernels, continuous tensor
products, and central limit theorems of probability theory, Mass., Lexington Books, 1972.
[RD` 66] B Russo, HA Dye, et al., A note on unitary operators in c*-algebras, Duke Mathematical Journal
33 (1966), no. 2, 413416.
[Rin72]
JR Ringrose, Cohomology of operator algebras, Lectures on operator algebras, Springer, 1972,
pp. 355434.
[RS72]
Michael Reed and Barry Simon, Methods of modern mathematical physics: Functional analysis,
vol. 1, Academic press, 1972.
[RS75]
, Methods of modern mathematical physics: Fourier analysis, self-adjointness, vol. 2, Elsevier, 1975.
[S` 62]
IE Segal et al., Mathematical characterization of the physical vacuum for a linear bose-einstein
field, Illinois Journal of Mathematics 6 (1962), no. 3, 500523.
[Sak71]
Sh
oichir
o Sakai, C*-algebras and w*-algebras, vol. 60, Springer, 1971.
[Seg47]
Irving E Segal, Irreducible representations of operator algebras, Bulletin of the American Mathematical Society 53 (1947), no. 2, 7388.
[Seg61]
IE Segal, Foundations of the theory of dyamical systems of infinitely many degrees of freedom. ii,
Canadian Journal of Mathematics 13 (1961), 118.
[Sim72]
Barry Simon, Topics in functional analysis, Mathematics of contemporary physics (1972), 1776.
[Sim77]
, An abstract katos inequality for generators of positivity preserving, Indiana University
Mathematics Journal 26 (1977), no. 6, 10671073.
[Sin76]
Allan M Sinclair, Automatic continuity of linear operators, vol. 21, Cambridge University Press,
1976.
[Sla72]
Joseph Slawny, On factor representations and thec*-algebra of canonical commutation relations,
Communications in Mathematical Physics 24 (1972), no. 2, 151170.
[SNR55] Bela Sz
okefalvi-Nagy and Frigyes Riesz, Prolongements des transformations de lespace de hilbert
que sortent de cet espace, Akademiai Kiado, 1955.
BIBLIOGRAPHY
[Sti55]
[St63]
[Sto69]
[St74]
[SU75]
[Suc73]
[Sul75]
[Tho71]
[Tom57]
[Tro59]
[Tsu77]
[Tur58]
[Ume54]
[VD78]
[Yos71]
74
W Forrest Stinespring, Positive functions on c*-algebras, Proceedings of the American Mathematical Society 6 (1955), no. 2, 211216.
Erling Strmer, Positive linear maps of operator algebras, Acta Mathematica 110 (1963), no. 1,
233278.
E Stormer, Positive linear maps and jordan homomorphisms of c*-algebras, Oslo University
(1969).
Erling Strmer, Positive linear maps of c*-algebras, Foundations of quantum mechanics and
ordered linear spaces, Springer, 1974, pp. 85106.
Ro Schrader and Do A Uhlenbrock, Markov structures on clifford algebras, Journal of Functional
Analysis 18 (1975), no. 4, 369413.
I Suciu, Dilation and extension theorems for operator-valued mappings, The Theory of Operators
and Operator Algebras (1973), 93130.
Wayne G Sullivan, Markov processes for random fields, Commun. Dublin Inst. Adv. Stud., A
(1975).
Lyn Carey Thomas, Some stochastic processes in quantum theory, Ph.D. thesis, University of
Oxford, 1971.
J Tomiyama, On the projection of norm one in w*-algebra, Proc. Japan Acad 33 (1957), 608612.
Hale F Trotter, On the product of semi-groups of operators, Proceedings of the American Mathematical Society 10 (1959), no. 4, 545551.
Sze Kai Tsui, A note on generators of semigroups, Proceedings of the American Mathematical
Society 66 (1977), no. 2, 305308.
Takasi Turumaru, Crossed product of operator algebra, Tohoku Mathematical Journal, Second
Series 10 (1958), no. 3, 355365.
Hisaharu Umegaki, Conditional expectation in an operator algebra, Tohoku Mathematical Journal, Second Series 6 (1954), no. 2-3, 177181.
Alfons Van Daele, Continuous crossed products and type iii von neumann algebras, vol. 31, Cambridge University Press, 1978.
K Yosida, Functional analysis, Spring-Verlag, New York/Berlin, 1971.

Dilations of Irreversible Evolutions in Algebraic Quantum Theory

Uploaded by

Document Information

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

Dilations of Irreversible Evolutions in Algebraic Quantum Theory

Uploaded by

Copyright:

Available Formats

Dilation of Irreversible Evolutions in

Algebraic Quantum Theory

For a description of the present situation in an historical context, see [LP73].

Chapter 1. Positive-definite kernels

Chapter 2. Positive-definite functions

Chapter 3. Dilations of Semigroup of Contractions

Chapter 4. C -algebras and Positivity

Chapter 5. Conditional Expectations

Chapter 6. Fock Space

Chapter 7. Representation of the Canonical Commutation Relations

Chapter 8. Representation of the Canonical Anti-Commutation Relations

Chapter 9. Slawnys Theorem

Chapter 10. Completely Positive Maps on the CCR Algebra

Chapter 11. Completely Positive Maps on the CAR Algebra

Chapter 12. Dilations of Quasi-free Dynamical Semi-groups

Chapter 13. Dilation of Completely Positive Maps on C -algebras

Chapter 14. Generators of Dynamical Semigroups

Chapter 15. Canonical Decomposition of Conditional Completely Positive Maps

Chapter 16. Isometric Representation of Quantum Dynamical Semigroups

Chapter 17. Unitary Dilations of Dynamical Semigroups

a subsystem. We thus require a non-commutative analogue of the conditional expectation

(see [DS63] p.619,622, [Yos71] p.232,240). Conversely,

0.2. BANACH *-ALGEBRAS AND C*-ALGEBRAS

for all x P X (see [Tro59], [Che74]).

0.5. TENSOR PRODUCT

P ` P P and R` P P . An ordered vector space is a vector space V equipped with a

hf, gi hf pq, gpqi dpq.

The map f b x f pqx extends uniquely to a unitary map of L2 pq b H onto L2 p; Hq (see

0.5. TENSOR PRODUCT

is a faithful *-representation of L8 pq as a maximal abelian von Neumann algebra on L2 pq

hKpxi , xj qhj , hi i V pxj qhj 0,

X BpH, HV q a map s.t. Kpx, yq V pxq

We have yet to show the existence of a Kolmogorov decomposition for an arbitrary

Then Definition 1.1 may be reformulated as:

(a) RpKq tKp, xqh : x P X, h P Hu,

Proof. Since the kernel K is positive-definite the associated convolution operator K

Proof. Let V be a minimal Kolmogorov decomposition for K; then we have

W p ` V pzq V pzqq 2 V pzq

W W p ` V pzq V pzqq 2 V pzq V pzqp ` V pzq V pzqq 2

Corollary 2.6. Let G be a group, and let T : G BpHq be a positive-definite function

Proof. Let : SJ JpSJ q BpHV q be the *-homomorphism of Theorem 2.5. Then

where dg is a left-invariant Haar measure on G, defines a completely positive map T 1 of

Dilations of Semigroup of Contractions

3. DILATIONS OF SEMIGROUP OF CONTRACTIONS

homomorphism of semigroups, then there is a unique homomorphism Kpq : KpSq KpS 1 q

Definition 3.5. A semigroup homomorphism T : S BpHq of an abelian semigroup

The relation (3.2) implies the weaker

3. DILATIONS OF SEMIGROUP OF CONTRACTIONS

Theorem 3.6. Let S be an abelian semigroup and let T : S BpHq be a homomorphism

for all k P KpSq. The representation U : KpSq BpHV q is given on V H by

using (3.1) we get (3.2) when k psq.

dij ij 1 ti1,i ti1,i , i 1,

3. DILATIONS OF SEMIGROUP OF CONTRACTIONS

Thus for i 1 we have

pw qik dkk wkj

tki dkk tkj

` t2i t2j t1i t1j

` tii tij ti1,i ti1,j

for all s in S; hence T has a unitary dilation.

3. DILATIONS OF SEMIGROUP OF CONTRACTIONS

Definition 3.12. Let S be a locally compact semigroup, a semigroup tTs : s P Su of

where tUt : t P Ru is the strongly continuous unitary group of right-translations on L2 pR; N q:

Let N0 denote the null space of this quadratic form

Letting t 8, remembering that Tt contracts strongly to zero, we see that there is an

0.2. BANACH -ALGEBRAS AND C-ALGEBRAS

pq f1 p1q b b f1 pnq , g1 b b gn

pq f1 p1q , g1 f1 pnq , gn