View
250
Download
3
Category
Preview:
Citation preview
Stochastic Analysis: hClassical and Quantum P u s p u t i v u o f W h i t a N o i s t T h e o r y
This page intentionally left blankThis page intentionally left blank
I: t ochas tic Analysis: Classical and Quantum P u s p u t i v u o f W h i t 4 N o i s 4 T h e o r y
Meijo University, Nagoya, Japaneditor
Wo World Scientificnew jersy - london - singapore - beijing - shanghai - hong kong - taipet - chennai
Published by
World Scientific Publishing Co. Re. Ltd.
5 Toh Tuck Link, Singapore 596224
USA once: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
LIK ofice: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.
STOCHASTIC ANALYSIS: CLASSICAL AND QUANTUM Perspectives of White Noise Theory
Copyright 0 2005 by World Scientific Publishing Co. Re. Ltd.
All rights reserved. This book, or parts thereol; may not be reproduced in any form or by any means, electronic or mechanical, includingphotocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to
photocopy is not required from the publisher.
ISBN 981-256-526-4
Printed in Singapore by B & JO Enterprise
Preface
We felt that time has come for a new epoch in stochastic analysis. Indeed various fields in mathematics as well as related fields in science continue to cross- fertilize each other, while keeping good relationships with probability theory.
There, dominant roles have been played by stochastic analysis (classical and
quantum).
It therefore seems to be a good opportunity to organize a conference on
important topics in stochastic analysis. Almost three decades has passed since
white noise analysis was launched, we thus plan to have perspectives of the
theory on this occasion.
Consequently, the conference “Stochastic Analysis: Classical and Quantum - Perspectives of White Noise Theory” took place at Meijo University, Nagoya,
Japan for the period of November 1-5, 2004.
The organizers of the conference were extremely happy to see many emi- nent mathematicians having contributed to the success of the conference and
cultivated new ideas. To our great pleasure, important papers presented at the
conference are published in the Proceedings of the conference. As such, we are grateful to the respective authors and to the referees of those papers.
We acknowledge gratefully the general support of Meijo University and Min-
istry of Education, Culture, Sports, Science and Technology for the conference. Special thanks are also due to Professors M. Rockner, L. Streit and T. Shimizu
who gave financial support together with me for the publication of this Pro- ceedings. Finally, I wish to note the great help given by members of the local
organizing committee: Professors M. Hitsuda, S. Ihara, K. Saito and Si Si. In
particular, it is to be mentioned that this Proceedings would not appear without the help of Professor Si Si who handled the aspect of editing.
July 2005 Takeyuki Hida
V
Organizing Committee of the Conference
Luigi Accardi
Takeyuki Hida
Hui-Hsiung Kuo
Masanori Ohya
Michael Rockner Ludwig Streit
vi
Contents
Preface V
Part I
White Noise Functional Approach to Polymer Entanglements
C. C. Bernido and M. V. Carpio-Bernido 1
White Noise Analysis, Quantum Field Theory, and Topology A . Hahn 13
A Topic on Noncanonical Representations of Gaussian Processes Y, Hibino 31
Integral Representation of Hilbert-Schmidt Operators on Boson
Fock Space
U. C. Ji
The Dawn of White Noise Analysis
I. Kubo
White Noise Stochastic Integration
H.-H. Kuo
Connes-Hida Calculus and Bismut-Quillen Superconnections
R. Le'andre and H. Ouerdiane
A Quantum Decomposition of L6vy Processes
Y.-J. Lee and H.-H. Shih
Generalized Entanglement and its Classification
T. Matsuoka
A White Noise Approach to Fractional Brownian Motion
D. Nualart
35
46
57
72
86
100
112
vii
Adaptive Dynamics in Quantum Information and Chaos
M. Ohya
Micro-Macro Duality in Quantum Physics
I . Ojima
White Noise Measures Associated to the Solutions of Stochastic
Differential Equations
H. Ouerdiane
A Remark on Sets in Infinite Dimensional Spaces with Full or
Zero Capacity
J . Ren and M. Rockner
An Infinite Dimensional Laplacian in White Noise Theory
K. Saitd
Invariance of Poisson Noise
Si Si, A. Tsoi and Win Win Htay
Nonequilibrium Steady States with Bose-Einstein Condensates
S. Tasaki and T. Matsui
Multidimensional Skew Reflected Diffusions
G. Trutnau
On Quantum Mutual Type Entropies and Quantum Capacity
N . Watanabe
Part I1
White Noise Calculus and Stochastic Calculus
L. Accardi and A. Boukas
127
143
162
177
187
199
211
228
245
260
viii
White Noise Functional Approach to Polymer Entanglements
Christopher C. Bernido and M. Victoria Carpio-Bernido’ Research Center for Theoretical Physics Central Visayan Institute Foundation
Jagna, Bohol 6308, Philappines
A b s t r a c t
The Hida-Streit white noise path integral is used to investi-
gate the entanglement probabilities of two chainlike macro-
molecules where one polymer lies on a plane and the other perpendicular to it. To simulate the data contained in the lin-
eal structure of a polymer of length L which lies on the plane,
a potential, V = f(s) ‘19, is introduced where, f = df/ds, 0 5 s 5 L, and f(s) a modulating function. Using the T- transform in white noise calculus, entanglement probabilities are calculated which show a significant influence of chirality
or the “handedness” of the polymer. The freedom to choose the modulating function f(s) , which gives rise to different entanglement probabilities, allows one to control and predict the coiling behavior of polymers. As examples, we consider
two cases: (a) f (s) = kcos (vs), and (b) f (s) = ksp.
1 Introduction
Investigations in biochemistry reveal that protein molecules are able to carry out their biological functions only when they are folded into spe-
cific three-dimensional structures [l]. For instance, enzymes, which are essentially protein molecules, have highly specific shapes which allow them to receive their targets as a lock receives a key. Understanding
this molecular recognition process, which depends on the structure of proteins, acquires importance since almost all chemical processes within a living organism rely on enzyme catalysis. What are the rules involved in forming protein structures? What are the factors which determine
the manner in which proteins fold? One essential factor which has been
identified is the one-dimensional sequence of data embodied in the re- peating units of macromolecules. It has been noted that the genetic code
is translated from DNA sequences to amino acid sequences, and this
‘Electronic mail: cbernido@mozcom.com
1
one-dimensional sequence of data influences the highly specific shapes of proteins. Moreover, chirality, or “handedness,” of macromolecules also
plays an important role in the globular structure of proteins. It is known, for instance, that amino acids in proteins are “left-handed,” and that the chirality of amino acids manifests in the helical structures of the proteins they form. These observations give rise to more specific questions. How can the sequence of data in the lineal structure of macromolecules allow
us to predict or determine a protein’s threedimensional structure? To what extent does chirality influence the folding or unfolding of proteins?
In this paper, we present a simpIe model which may shed light into
these questions. In particular, we look at a model which incorporates the following features observed in a macromolecule:
(1) the ability of a macromolecule to use the one-dimensional sequence of data in its repeating units to influence its glob- ular structure, and,
(2) the chirality of the polymer which influences the three- dimensional structure of a macromolecule.
We start with a polymer entanglement scenario originally studied by Edwards [2] and Prager and Frisch [3]. We then extend this system [4] by simulating the data contained in the repeating units of the entangled
polymer using a potential of the form, V = f(s) 6, where f = d f / d s , and 19 is an angular variable about the z-axis. Here, f (s) is a modulat- ing function where, 0 5 s 5 L, and L is the length of the polymer. We shall see that for any modulating function f (s), the “handedness” of the
winding polymer has a significant effect on the entanglement probabil-
ities. In particular, we look at two cases (a) f(s) = kcos(vs), and (b) f (s) = ksp, where k is a positive constant and p = fl , f 2 , f 3 , .... Our
calculations are greatly facilitated by first parametrizing the probabil- ity function in terms of the white noise variable, as was done by Hida
and Streit [5] for the case of quantum propagators. The white noise functional can then be evaluated in a straightforward manner using the T-transform of white noise calculus [7]-[9].
2 Two Chainlike Macromolecules
In 1967, S. F. Edwards [2] and, independently, S. Prager and H. L. Frisch [3], solved the entanglement problem of two chainlike macromolecules in the absence of intermolecular forces. The problem consists of a polymer on a plane whose motion is constrained by a straight polymer orthog- onal to the plane, since the macromolecules cannot cross each other.
The polymer on the plane which starts at ro and ends at rl has fixed
2
endpoints, and can be viewed as a random walk with paths that can en-
tangle, clockwise or counterclockwise, around the straight polymer which
intersects the origin of the plane. Employing polar coordinates r =(r, 6) for this problem, S. F. Edwards [2] used the Wiener representation of
the random walk in which the probability is represented by,
where the integral is taken over all paths r(s) such that r(0) = ro and
r(L) = rl. Here, we represent the polymer of length L as consisting of N freely hinged individual molecules, each of length 1 such that L = N1. In view of the point singularity, a set of topologically equivalent
configurations can be characterized by a winding number n, where n =
0, kl, f2, ..., indicating the number of times the polymer turns around the straight polymer intersecting the plane at the origin ( n 2 0, signifies
n turns counterclockwise, and n 5 -1 means (n + 11 turns clockwise).
3 Entanglement with an Intermolecular Potential v (4
In 1977, F. W. Wiegel [S] extended this entanglement problem to include an intermolecular force where the repeating units of the entangled poly- mer interact with the straight polymer. For any potential V( r ) which has a minimum at some radius R, Wiegel obtained a low-temperature
limit for the entanglement probabilities given by,
W(n) = ( R / l ) W e x p (-47r2n2R2/N12) ; ( N >> 1). (3.1)
For example, the potential of the form, V = Cr2+D/r2 (C > 0, D > 0),
was considered where R = (D/C)'I4 is a radius where the potential has a minimum. With this potential, the force is repulsive at short distances and attractive at large distances. Wiegel then obtained the entangle- ment probabilities for this harmonically bound polymer to be that of
Eq. (3.1). For low temperatures, he also noted that the configurations of the polymer are confined to a narrow strip in the immediate vicinity of a circle around the origin with radius R . Below, we shall use these
observations of Wiegel which we refer to as the generic case.
4 White Noise Path Integral Approach
Let us now familiarize ourselves with the white noise path integral ap- proach by using it to arrive at Eq. (3.1). Since we are interested in
3
the number of possible windings around the origin that the polymer on the plane undergoes, we can simplify the calculation by fixing the radial
variable to T = R, i.e., r =(R, 6) , and use 6 to track the number of turns,
clockwise or counterclockwise, around the origin. As mentioned in the previous section, a fixed radial part describes the entanglement scenario
in the low temperature limit [6] for any polymer interaction potential V ( T ) which has a minimum at some value T = R. For the generic case,
Eq. (2.1) reduces to,
L
P(61,60) =/exp [ - + / R 2 ($ ) ' ds ] 'D [Rdd ] , (4.1) 0
with, d1 = 6(L) and 60 = d(0). The paths 6 can be parametrized as,
6(L) =60 + ( h / R ) B(L ) L
=60 + (&/R) / w ( s ) ds, (4.2) 0
where B(s ) is a Brownian motion parametrized by s, and w(s ) a random
white noise variable. With Eq. (4.2), the integrand in P(61, 60) becomes,
Noting that the polymer can wind n times, clockwise or counterclock- wise, we use the Donsker delta function
to fix the endpoint 61, where n = 0, fl , 5 2 , ... . Since P(t91,60) is
now expressed as a white noise functional, the integration over D[R dt9] becomes an integration over, Nu d"w = exp [(1/2) Sw(s)' ds] dp(w), where dp(w) is the Gaussian white noise measure. Eq. (4.1) can now be written as,
where
4
10 = Nexp (-: )w(s). ds) .
The evaluation of P(61,60) is facilitated by using the Fourier represen- tation of the 6 - function, i.e.,
Observing that the integration over dp(w) is just the T-transform of 10 [7]-[9], we obtain,
x exp (-X2ZL/4R2) dX
n=-w
Here, the Pn is the probability function for polymer configurations which entangle n - times around the origin. The remaining integral in Pn is a Gaussian integral over A. We have,
= m e x p [- (R2/1L) (60 - 61 + 27rn)2] . (4.9)
Also, applying Poisson's sum formula,
to Eq. (4.8), we get,
5
x exp [ i X (60 - 61) - X2(ZL/4R2)] dX
1 +O0 = -
27r exp [-im (60 - dl) - m2(1L/4R2)] . (4.11)
m=--00
For an arbitrary initial starting point we may set, 60 = 61, and the probability that the polymer winds n - times is,
(4.12) d m e x p [- (27rn~)’ /ZL]
- - +m C exp [-m2(1L/4R2)]
2?r m=-m
For a very long polymer, L = N1 >> 1, the dominant term in the denominator is for m = 0. Hence,
W(n) = ( R / l ) W e x p (-47r2n2R2/N12) ; ( N >> l), (4.13)
which agrees with the result, Eq. (3.1), obtained by Wiegel [6].
5 Length-dependent Potentials
We shall now generalize the system discussed in the previous section by
adding a length-dependent potential, V = f(s) 6, acting on the polymer on the plane as it entangles around the second straight polymer at the
origin [4]. Here, f = df /ds, where f (s) is the modulating function. The potential V is added to the “kinetic part” of Eq. (4.1) such that the
probability function becomes,
(5.1)
The nature of the potential may be understood in the following way. Firstly, one may associate with it a length-dependent force given by,
F = -VV = - f (s ) /R. Secondly, the effect of the potential term may
also be understood by rewriting it as,
6
L
= f ( L ) S ( L ) - f ( 0 ) 6 ( 0 ) - 1 fS ds . (5.2)
0
The first two terms are constants given by the values of f and 6 at the endpoints. The last term, on the other hand, shows that one essentially
has a “velocity-dependent potential” in view of s. Moreover, from Eq.
(5.2), one may have the case, f(s) = 0 with f # 0, such that the nonzero f may still manifest in the probability function. For the case, f = 0, one
obtains the results of the generic case discussed in the previous section.
An example of a constant nonzero f may be illustrated if one takes,
f = (q@0/27r), where q is the net charge of each repeating unit of the
polymer which winds around the straight polymer that contains a con-
stant magnetic flux @O oriented along the z-axis. This choice leads to an
effective potential, f $ = qA . i, which resembles that of an Aharonov-
Bohm setup where A is the vector potential for the constant magnetic
Using again the parametrization Eq. (4.2), we obtain an expression similar to Eq. (4.7) but modified by the potential Eq. (5.2) of the form,
flux @o [4, 101.
L
x Jexp { i s ( d i / ~ ) (X - if ) w ( s ) ds
0
XIO dp(w) dX . (5.3)
The integration over dp(w) is again just the T - transform of I0 which yields,
7
+m
= c Pn. (5.4)
The Gaussian integral over X in Pn can be evaluated to give,
Employing the Poisson sum formula to Eq. (5.4), and integrating X yields,
+m
L L m21L iml d s + s $ f 1 ' d s } . (5.6) 4R2 2R2
0
From these, we obtain the probability that the polymer entangles n times
as, (setting, 60 = &) ,
8
where &(u) is the theta function [ll],
+W
e3(4 = 1 + 2 C qmz cos(2mu), m=l
with u = (1/4R2) j” f ds, and q = exp (-N12/4R2).
6 Chirality of Entangled Polymers
Let us now consider the effect of the “handedness” of a polymer on the coiling probabilities of a macromolecule. As is normally the case, we define “handedness” in a way that a “right-handed” polymer would have a mirror-image which is “left-handed.” We may write Eq. (5.7) as,
where W,, symmetric in n, is of the form,
From these equations the following observations may be made:
(1) It is clear from Eq. (6.1) that the entanglement prob-
ability W(n) significantly changes depending on whether n is a positive or a negative number. If we designate clock- wise winding (n < -1) as “right-handed,” and anti-clockwise winding (n 2 0) as “left-handed,” then for f > 0, a “right- handed” polymer is more likely to have configurations with large values of winding number n than “left-handed” ones. In particular, for winding numbers kn, the corresponding “right” and “left-handed” entanglements differ by an expc-
nential factor, i.e., W(-n)/W(n) = exp [ ( 4 m / L ) J f ds ] .
(2) We also see from Eq. (6.1), that a change in sign of the modulating function, from f to - f , gives a similar effect as that of n to -n, i.e., from “left-handed” to “right-handed”.
Since the role of the modulating function f is that of simu- lating the data contained in the lineal structure of the poly- mer, it appears that the choice of the sign of f is one way
of incorporating the observation that the “handedness” of
9
the constituent monomers manifests in the chirality of the
polymer.
(3) When V = 0, chirality has no effect on the entanglement probabilities. This holds true for the generic case where chi-
rality of the polymer has no influence on the way the macro- molecule entangles.
From these, one sees that the model embodies certain features of the “handedness” of a polymer which may lead us to a better understand-
ing of chirality and its influence on the three-dimensional structure of
macromolecules.
7 Examples of a Modulating Function
An advantage of this model is that the potential, V = f ( s ) 6, where
0 5 s 5 L with L the length of the polymer, can be chosen to simulate
the data contained in the one-dimensional structure of the entangled polymer. In particular, one can choose the form of the modulating func-
tion f (s ) . We, therefore, have a framework where an effective potential,
chosen to represent the effect of the sequence of monomers making up the macromolecule, can be used in order to determine the over-all structure
in three-dimensions such that it departs from the generic case. In short,
we have as our input the f (s), and our output the winding probabilities
W(n). We now consider two cases below.
7.1 This choice leads to the potential, V = -kv sin (us) 6, where we could
take u as the frequency of the repeating unit in the polymer. From Eq. (5.7) the winding probability becomes,
f(s) = k cos (VS)
For a very long macromolecule, L = Nl >> 1, the denominator ap- proaches unity and we have,
R 47r -47r2n2 R2 w(n)z T G e x P ( N12 ) 4NR2v2 1 ‘
2n-nk . k2 sin2 (vL) sin (vL) -
(7.2) . , As can be seen in Eq. (7.2), the second exponential modifies the entan-
glement probabilities for the generic case. The effect of the second expo- nential may also be viewed by using the expansion, exp (.) = xn/n! .
10
Note, however, that when the frequency has the value v = nn/L, where n = 0, f l , f 2 , ..., Eq. (7.2) reduces to the generic case.
7.2 f(s) = ksp
For this choice of the modulating function we can take k to be a positive constant and the possible values of p to be, p = f l , f 2 , f 3 , ... . When
p = -1, -2, -3, ..., the integral s f (s) ds in Eq. (5.7) becomes infinite
thus damping out the exponential. This implies that W(n) = 0, for
k > 0. Physically, this would correspond to a stretched or uncoiled
polymer.
f (s) ds = IcLpfl/ ( p + l), and we have from Eq. (5.7),
On the other hand, for p = +l,+2,+3, ..., we get,
(2sn.+ m ) 2 ] lkLp+l
(7.3) R 4n
1 0 3 (4%pp=:])
For a very long macromolecule, L = N1 >> 1, this becomes,
The second exponential which modifies the generic case inhibits, in gen- eral, the coiling of the polymer.
belongs to a class of modulating functions that can inhibit the coiling of
polymers where, V = f (s) 19, becomes a stretching potential.
8 Conclusion
In this paper, we employed the white noise path integral to investi- gate the morphology of macromolecules. Starting from the entangle-
ment scenario originally studied by Edwards [2] and Prager and Fkisch [3], we simulate the one-dimensional information embodied by an en-
tangled polymer by introducing the potential, V = f(s) 19. We showed
that different choices of the modulating function f (s) give rise to differ- ent entanglement probabilities. Ideally, one should choose a modulating
function f (s) that would best simulate the biochemical data contained in the one-dimensional structure of the entangled polymer to predict the coiling behavior of polymers. The model studied also makes explicit the effect of chirality on winding probabilities. The present study should
Theseobservations imply that, f(s) = ksP, ( I c > 0, p = fl, f 2 , *3, ...),
11
then lead to more detailed models which could provide additional in- sights on the study of protein folding and the role of chirality in the globular structure and morphology of macromolecules.
Acknowledgement
helpful comments.
References
The authors would like to thank L. Streit and F. W. Wiegel for their
[l] See, e.g., P. Ball, Designing the Molecular World (Princeton Univ.
[2] S. F. Edwards, Proc. Phys. SOC. London 91 (1967) 513-519.
[3] S. Prager and H. L. F’risch, J. Chem. Phys. 46 (1967) 1475. [4] C. C. Bernido and M. V. Carpio-Bernido, J. Phys. A: Math. Gen.
[5] L. Streit and T. Hida, Stoch. Proc. Appl. 16 (1983) 55-69.
[6] F. W. Wiegel, J. Chem. Phys. 67 (1977) 469-472 [7] T. Hida, H. H. Kuo, J. Potthoff and L. Streit, White Noise. An
[8] N. Obata, White Noise Calculus and Foclc Space, Lecture Notes in
[9] H. H. Kuo, White Noise Distribution Theory (CRC, Boca Raton,
[lo] See, also, C. C. Bernido and M. V. Carpio-Bernido, J. Math. Phys.
[ll] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
Press, Princeton, 1994).
36 (2003) 4247-4257.
Infinite Dimensional Calculus (Kluwer, Dordrecht , 1993).
Mathematics, Vol. 1577 (Springer, Berlin, 1994).
FL, 1996).
43 (2002) 1728-1736.
Products, 5th ed. (Academic Press, San Diego, 1994) p. 927.
12
White Noise Analysis, Quantum Field Theory,
and Topology
Atle Hahn Institut fur Angewandte Mathematik der Universitat Bonn
WegelerstraBe 6, 53115 Bonn, Germany E-Mail: hahnQuni-bonn.de
1 Introduction
Topological quantum field theories provide some of the most interesting exam- ples for the usefulness of path integrals. One of the best known of these examples was discovered in [32] where one particular topological quantum field theory, Chern-Simons theory, was studied and the so-called “Wilson loop observables” (WLOs) were computed explicitly. These WLOs are heuristic path integrals and the interesting thing about the expressions obtained in [32] is that they in- volve highly non-trivial link invariants like the Jones polynomial, the HOMFLY polynomial, and the Kauffman polynomial, cf. [26, 161. A more thorough study of the WLOs by other methods [18, 8, 5, 171 later led to a breakthrough in knot theory, the discovery of the universal Vassiliev invariant [27].
Unfortunately, it has not yet been possible to establish the aforementioned connection between path integrals and knot polynomials at a rigorous level. In the special case, however, where the base manifold M of the Chern-Simons model considered is of product form the situation looks much more promising and as we will show in the present paper it is reasonable to expect that, at least for some of these special manifolds M, it will eventually be possible
1. to obtain a rigorous definition of the WLOs in terms of Hida distributions (Task 1) and
2. to prove that the values of the rigorously defined WLOs are indeed given by the explicit formulae in 1321 (Task 2).
The present paper is organized as follows. In Sec. 2 we briefly describe what Chern-Simons models are and why, on a heuristic level, they give rise to link invariants.
In Sec. 3 we will summarize some recent results for the manifold M = W3 R2 x R, for which Task 1 has already been carried out successfully. This manifold
13
has the drawback of being noncompact and for this reason one cannot he sure that the values of the WLOs are given by the formulae in 1321.
Fortunately, there is at least one compact manifold for which Task 1 can also he carried out, namely the manifold M = S2 x S1. In Sec. 4 we will give an overview over the results obtained so far for this manifold and we will sketch what remains to he done in order to complete Tasks 1 and 2.
2 Chern-Simons models, heuristic path inte- grals, and topological invariants
2.1 Chern-Simons models
A (pure) Chern-Simons model is a gauge theory on a 3-dimensional connected oriented manifold M without (!) Riemannian metric. The manifold M is often assumed to be compact but sometimes also the noncompact case is studied. The structure group G of the Chern-Simons gauge theory is usually assumed to be compact and connected. Without loss of generality we will assume in the sequel that G is a (closed) subgroup of the group U ( N ) with fixed N 2 1. Once the base manifold M and the group G of the model are known the only free parameter of the theory is the so-called “charge” k E B\{O}, which in the case of
compact M is assumed to be an integer. We set X := and call X the “coupling constant” of the model. The Lie algebra of G, which we will identify with the obvious Lie subalgebra of u ( N ) , will be denoted by g. Using this identification we obtain a real (!) bilinear form (., : g x g 3 (A, B ) H - Tr(A . B) E B on g, which can be shown to he a scalar product.
The space of gauge fields, i.e. the space of all g-valued smooth 1-forms on M , will he denoted hy A. What makes Chern-Simons theory special is the fact that the action functional SCS : A -+ C of a Chern-Simons model does not involve a Riemannian metric. It is given by
As M was assumed to be oriented and 3-dimensional, the integral on the right- hand side is well-defined even though no measure is involved.
2.2
From the definition of SCS it is obvious that SCS is invariant under (orientation- preserving) diffeomorphisms. Thus, at a heuristic level, we can expect that the heuristic integral (the “partition function”)
Heuristic path integrals and topological invariants
Z ( M ) := exp(iScs(A))DA s is a topological invariant of the 3-manifold M . With DA above we mean the informal “Lebesgue measure” on the space A.
14
Similarly, we can expect that the mapping which maps every sufficiently “regular” link L = ( 1 1 ~ 1 2 , . . . , I n ) in M to the heuristic integral (the “Wilson loop observable” associated to L )
is a link invariant. Here we have used the standard physicists notation P exp (hi A) for the holonomy of A around the loop li.
2.3 A “Paradox”
For Abelian G the path integral expressions for the WLOs are well understood, see, e.g., [2, 301. For Non-Abelian G the WLOs were evaluated explicitly for the first time in [32]. The explicit expressions obtained in [32] contain the famous Jones polynomial and its two-variable generalizations, the so-called HOMFLY and Kauffman polynomials. These polynomials are highly non-trivial link in- variants which were discovered only a few years before [32] was written. Later, a more thorough study of the WLOs by other methods (see Subsec. 2.4 below) led to a breakthrough in knot theory, the discovery of the universal Vassiliev invariant in [27]. Thus we are in a somewhat “paradoxical” situation. On the one hand the heuristic integral expressions (2.1) contain some very deep math- ematics. On the other hand it is absolutely not clear whether it is possible to give a rigorous mathematical meaning to these heuristic expressions. As we will try to demonstrate in the sequel things look better if one considers only special base manifolds M and introduces a suitable gauge fixing procedure. Before we do this in Secs. 3 and 4 let us briefly summarize the approaches that have been developed so far for the computation of the WLOs.
2.4
To our knowledge the following approaches have been worked out for the com- putation of the WLOs for general G.
The computation of the WLOs: A short overview
i) Firstly, there is the original approach in [32] which uses arguments from Conformal Field Theory and surgery operations on the base manifold. This approach is the most elegant one. It is non-perturbative, does not involve a gauge fixing procedure and works for arbitrary base manifolds M . The drawback of this approach is that it is not very “explicit”.
ii) Another approach is the approach developed in [17] (see also [28]) for the special manifold M = R3. This approach is based on light-cone gauge fixing (which is equivalent to axial gauge fixing) and an additional “com- plexification” of the coordinates of the points in M . It is non-perturbative and rather explicit. The drawbacks of this approach are, firstly, that it does not work for general M and, secondly, that it leads to the correct values of the WLOs only when certain “ad hoc” correction factors are
15
introduced. Until today there seems to be no convincing explanation why these correction factors have to be used.
iii) There is also an approach for the special manifold M = P3 which is based on axial gauge fixing without “complexification” of the coordinates of the
points in M . Until 1997 this approach was much less developed than the other three approaches listed here and only preliminary results were
available, cf. [15], [29], [13]. Then, in [3] it was suggested that white noise analysis should be used for a successful treatment of the axial gauge setting. Finally, in [21, 22, 231 it was shown that the suggestion in [3] is indeed correct and that with the help of white noise analysis the WLOs
can indeed be defined and computed rigorously in the axial gauge setting. The advantages of the the axial gauge approach (without complexification) are that this approach is simple and rigorous. The essential drawback is that the corrections which have to be made if one wants to obtain the correct values of WLOs are even more serious than in the “complexified”
approach by [ 171, cf. [23].
iv) Finally there is the “Lorentz gauge fixing” approach initiated in 118, 8) and later elaborated by [7, 6, 5, 12, 41. This approach works for arbitrary M and is rather explicit. The drawback of this approach is that it is perturbative and rather complicated.
2.5 Open Questions
When comparing the four approaches listed above a couple of questions arise rather naturally:
- Why does the application of the “non-covariant” gauges, i.e. light-cone and axial gauge (with or without complexification) only lead to the correct values of the WLOs when certain “ad hoc” corrections are carried out later? Are the deviations from the correct values of the WLOs to be seen as defects of these non-covariant gauges or is there another explanation?
- Is there an approach which is simple, explicit, and correct? Is there an
approach with these three properties which is even rigorous? If so, this does not necessarily mean that the “paradox” of Subsec. 2.4 is solved’ but it would certainly be an important step towards the resolution of it.
It is mainly these two questions that will concern us during the rest of this
paper.
lNote that by applying a gauge fixing procedure one modifies the heuristic path inte-
gral expressions (2.1) so even if one succeeds in making sense of the modified path integral expressions this does not mean that one has also made sense of the original path integral expressions.
16
3 White noise analysis applied to Chern-Simons models on M = R3 in axial gauge
3.1 The basic idea
Let dxi, i = 0,1,2, denote the standard 1-forms on R3. For every A E A we will denote the coordinates w.r.t. (dxi)i=0,1,2 by (Ai)i=0,1,2. We will call A E A axial iff A2 = 0 and we will set A”” := {A E A I A2 = 0) . The space A”” has two nice features:
1. The so-called “Faddeev-Popov determinant” AFadpop which is associated to axial gauge fixing is a constant. By “definition” AFadPop is the unique function on A”“ with the property
for every gauge-invariant function x : A --t C. Here DA”” is the (informal) “Lebesgue measure” on A“”. As AFadpop(Aax) is constant we obtain for the special case x(A) = nin(’Pexp(Sli A)) exp(iScs(A)) where L =
( I I , . . . , I n ) is a fixed link in R3
N 1 n T r ( P e x p ( 1 A”“)) exp(iScs(Aa“))DAax. A”” i 1;
Here N denotes “equality up to a multiplicative constant”.
2. For every A““ E A”” we have A”” A A”” A A”“ = 0. Thus we obtain
Taken together these two features “imply” that the heuristic “measure”
dpFs(A””) := %h exp(iScs(Aa”))DAax
where Z a X ( M ) := ance operator”
exp(iScs(A”“))DA”” is of “Gaussian type” with “covari-
-1
c:= 2 T i X . (-oa, 2) .
17
3.2
If one can make sense of the operator C above as a continuous operator N ---t
N’ where N := Sgeg(W3) 2 A”” then < . >Fs:= s. . .dpFs can be defined
rigorously as an element of (N)” corresponding to the Gelfand triple N c H c N’ where H := Lie,(W3, dz). We make the following Ansatz:
Step 1: Making sense of J’. - . dpFs
where 8;’ : Sg(W3) + Cr(W3,g) is a left-inverse of the operator 8 2 , i.e. fulfills
8;’ = idsg(p). It is not difficult to see that every such left-inverse 8;’ must
be a linear combination of the form s + (1 - s) . a;’, s E W, where
(&‘f)(z) = 1“’ f(zo,zl,t)dt for f E Sg(R3)
(%‘f)(z) = -1 f (zo,z l ,W for f E S,(R3).
--oo
05
” a
Each operator C,, s E W, which is obtained from Eq. (3.1) above when inter-
in the way just described is indeed a con-
tinuous operator N -+ W . Moreover, one can show that although C, depends
on the choice of the parameter s, the quadratic form Qax(j) :=< j,C,j >>H,
j E N c H, does not. As n/ 3 j H exp(-+Qa”(j)) E C is clearly a U-functional the Characterization Theorem allows us to define < . >Fs:= s. . .dpFs rigor-
ously as the unique element of (N)* such that
preting the operator (-oa,
1
2 < exp(i(.,j)) >Fs= exp(--Q””(j)) for all j E N
where on the left-hand side (., .) denotes the canonical pairing n/* x N + W.
3.3 Step 2: Making sense of < n,Tr(Pexp(Li(-))) >Fs After having succeeded in making sense of the heuristic integral functional
j” dpFs we can now ask whether it is also possible to make sense of the
whole expression < niTr(Pexp(h,( . ) ) ) >Fs. This can indeed be achieved with the help of two regularization procedures, “loop smearing” and “framing”.
These two regularization procedures are described in detail in [21, 22, 231. Here we will content ourselves with a very brief sketch of how they are used:
Loop smearing: We regularize ni Tr (P exp(J, (.))) by using “smeared loops”
1: (later we eliminate the variable E by letting E + 0.) One can show that for fixed E > 0 the function n,Tr(Pexp(Jl:(.))) is in the domain ( N ) of < . >Fs.
Framing: In order to implement the “framing procedure” we first fix a family ($s)s>o of diffeomorphisms of W3 with certain properties. In particular, we
18
demand that ds + idas as s + 0 in a very weak sense. Each diffeomorphism q5s gives rise to a “deformed” version < . >g of < . >E% (later, we let s + 0).
Of course, one has to prove that the double limit
W L O ( L ; ~ ) := lim lim < n n ( p e x p ( s-+o € 4 0
i
really exists. This is part of the next step.
3.4 Step 3: Existence proof and computation of WLO(L; 4 ) For simplicity we will consider only two special cases, namely G = U(1) and
G = SU(N) . Let us start with the case G = U(1). Before we state the
corresponding theorem let us briefly recall the definition of the linking number LK(1,l’) of two given loops 1 , I’ which do not intersect each other. LK(1,l’) is given by
LK(1,l‘) := f c € ( p ) P E C T ( l , l ‘ )
where cr(1,l’) is the set of all mutual crossings of the planar loops which are obtained by (orthogonally) projecting 1 and 1’ to the xo-xl-plane. To each crossing p E cr(1,l’) a “sign” ~ ( p ) E {-1, l} is associated according to the following pictures:
Figure 1: ~ ( p ) = -1 Figure 2: ~ ( p ) = 1
Theorem 1 Let G = U(1). “admissible” framing q5 := ( q 5 s ) s > ~ the double limit
Then for every “admissible” link L and every
exists and we have
where lk j = lims+0 LK( l j , d5 o l j ) .
19
For an exact definition of the notion “admissible” for links and framings see 121, 221.
The “wraith” w ( L ) of a link L = ( 1 1 ~ 1 2 , . . . , I n ) is given by
w(L) := c 4 P ) PEWL)
where V(L) is the set of all (mutual and self) crossings of the planar loops
which are obtained by (orthogonally) projecting the loops Z1, 12, ..., 1, to the $0-xl-plane.
Two loops I , 1’ which are very ‘Lclose” to each other but do not intersect can be considered to be the boundary of a “ribbon” obtained by interpolating l ( t ) and l’(t) for every t E [0,1]. The number of “twists” of this ribbon will be denoted by twist(1,l’).
Theorem 2 Let G = S U ( N ) . Then for every “strongly admissible’’ link L and every “strongly admissible” framing 4 := ( q 5 s ) s > ~ the double limit
exists. wLO(L; 4) is independent of the “loop smearing axis” which was fixed in the course of the loop smearing procedure af and only i f X E 22. In this case we have
WLO(L; 4) = N # ~ exp(-+ t i ) exp(-+ w(L)). ( 3 4 jsn
Here #L is the number of components of L and t j := lirn,-o twist(lj, 4s 0 l j ) .
For an exact definition of the “loop smearing axis” and the notion “strongly admissible” for links and framings, see [23].
3.5 Comparison with the results obtained by the other approaches
According to the standard literature, cf., e.g., [18], [17], we should have
WLO(L; 4) = HOMFLYL(exp(XrzN), 2isin(Xr)) x
x fl expO\ri + t j ) exp(X7ri - NX1 w(L)) (3.3) i 5n
for every X E A := {t*,&&, ...}. Here HOMFLYL is the HOMFLY polynomial which is associated to the link L. The set A consists of those values of X for which the charge k = is an integer and for which equation (3.3) gives
rise to values for the Wilson loop observables which are compatible with the “unitarity” of the theory (cf. pp. 168f in [17]).
20
In order to compare equation (3.3) with (3.2) let us introduce the function
fL : R\Z 3 X H HOMFLYL(exp(XniN), 2isin(Xn)) x E C
fL is a well-defined function on R\Z which can be extended uniquely to a continuous function f~ on all of R. It is easy to see that ~ L ( X ) = N # L if
X E 22. Clearly, for X E 2 2 we have e x p ( - F ) = exp(Xnz(*)). Thus equation (3.2) can be considered to be the “special case” of equation (3.3) for
X E 225. Theorem 2 raises the question whether one should perhaps replace the set A
by 22. For all X E 2 2 the theory should again be unitary. Of course, the charge
demand k E 2 if the base manifold M of the Chern-Simons model considered is noncompact like in the case M = R3. This leads us to the following conjecture.
Conjecture. The problems that appear when applying light-cone gauge and axial gauge fixing to Chern-Simons models on R3 have nothing to do with these gauges but with the non-compactness of R3.
Fortunately, there is a good chance of finding out wether this conjecture is true or not. This is because there is at least one compact manifold, namely M = S2 x S’, for which a gauge fixing is available that is very similar to axial
gauge in the case of It3. This gauge was called “Torus gauge” in [9]. In Sec. 4 we will show how, using torus gauge fixing, one can find a rigorous
representation of the WLOs in terms of Hida distributions also for Chern-Simons
models on S2 x S’. We expect that by computing the WLOs explicitly one will get an answer to the question whether the conjecture above is true or not.
k = l will then not be an integer, but we doubt whether it makes sense to
4 White noise analysis applied to Chern-Simons models on A4 = S2 x S1 in torus gauge
4.1 Torus Gauge
In order to make the similarities between axial gauge and torus gauge (to be
defined below) more explicit let us first consider the manifolds of the form M = C x R. Let t denote the global coordinate M --f R given by t(u,s) = s,
u E C, s E R. The global coordinate t gives rise to a 1-form dt on M . By lifting the constant vector field on R taking only the value 1 to the manifold M = C x R with the help o f t we also obtain a vector field on M . This vector field will be denoted by &. Clearly, we have d t (& ) = 1. Let us now introduce the subspace A l
A’ = { A E dl A(&) = 0 )
of A. Clearly, every A E A can be written uniquely in the form
A = A’ + Atd t
21
with A’ E A’ and At E C“(M, g). Note that in the special case where C = R2 and consequently M = R2 x R 2 R3 the space A’ coincides with A”” defined in Sec. 3. In this case the following three statements are clearly equivalent: A is axial, A = A’ and At = 0.
After these preparations let us now consider manifolds M of the form M = C x S’. Even though the mapping t : C x S’ --t S1 with t (o ,s ) = s is not a global coordinate it can be used to “lift” the standard 1-form dt and the standard vector field & on 5’’ to a 1-form resp. vector field on C x S’. The
lifted 1-form resp. vector field on M will again be denoted by dt resp. 4. As before we can now introduce the space
A’ := { A E A I A(&) = O}.
Again every A E A can be written uniquely in the form
A = A’ + Atdt
with A’ E A’ and At E Cm(M,g) . However, there is a crucial difference between the case M = C x R and the
case M = C x S’. For M = C x R the condition At = 0 “defines” a gauge. More precisely: Every 1-form A E A is gauge equivalent to a 1-form in A’- = { A E A I At = 0) . By contrast for M = C x 5’’ the condition At = 0 does not define a gauge. There are 1-forms which are not gauge equivalent to any 1-form in A’. For example this is the case for any 1-form A for which the holonomy Pexp(Jco A ) around the loop I , , : S’ 3 s H (00 , s) E M where oo E C is a fixed
point is not equal to 1. This follows immediately from the two observations that, firstly, the holonomies are invariant under gauge transformations and, secondly, we clearly have P exp( A’) = 1 for every A’ E A’.
Thus, in order to obtain a proper gauge we have to weaken the condition At = 0. There are two natural candidates for such a weakened condition.
1-0
1. Option: Instead of demanding At(o, s) = 0 for o, t we just demand that At(o, s) is independent of the second variable s, i.e. we demand that At = B holds where B E C“ ( C , g) (“Quasi-axial gauge”)
2. Option (better): We demand, firstly, that At(a, s) is independent of the second variable and, secondly, that it takes values in the Lie algebra t of a fixed maximal torus T c G (“Torus gauge”),
Thus we arrive at the following definition
Definition 1 Let T be a maximal torus of G. A 1-form A E A is said to be “in the T-torus gauge” i f f A = A’ + Bdt holds.
In the next subsections
there is a A’ E A’ and a B E C”(C,t) such that
we will restrict ourselves to the special case C = S 2 .
22
4.2 The Faddeev-Popov-Determinant of Torus gauge fix-
From now on we will set C := S2. The aim of this subsection is to identify the
Faddeev-Popov-Determinant a F a d p o p which is associate to torus gauge fixing.
By “definition” A F a d p o p is the unique mapping on A‘ x Cm(s2, t) such that
ing
(4.1)
holds for every gauge invariant function x : A 4 C. Here DA’ denotes the (in- formal) “Lebesgue measure” on A’ and D B the (informal) “Lebesgue measure”
on Cm(S2, t). It is possible to compute A F a d p o p explicitly and if one does so one obtains
the following heuristic equation:
AFadPop(A’ + Bdt) = 1 det(& + ad(B)) I det(lgu - exP(ad(B))lgu)
Here go denotes the (., .),-orthogonal complement of t in g. The special case of Eq. (4.1) in which we are interested is the case obtained
by taking x(A) = ni Tr(Pexp(h, A ) ) exp(iScs(A)) where L = (11,. . . , I n ) is a
fixed link in S2 x S1. In this special situation we have
4.3 A Formula for Scs(A‘- + Bdt)
Let us identify the space A’ with the space C”(S1,dsz) of “smooth”2 ds2-
valued mappings on S1 in the obvious way. Here As2 denotes the space of g-valued 1-forms on S2. Moreover, we introduce the bilinear form
< ., . >sz: dsz x Asz 3 (0, 01’) H n(01 A 01’) E C L One can show that for all A’ E A’- and all B E Cm(S2, t) we have
Scs(A’ + Bdt)
= - & l l [ < A L ( t ) , ( $ + a d ( B ) ) . A L ( t ) >sz -2<A’-(t),dB>sz dt (4.3) 1 2For a definition of the notion “smooth” here, see [24]
23
Consequently, for fixed B, the mapping
dl 3 A' Scs(A' + Bdt) E C
is quadratic. This point will be of crucial importance in the sequel.
4.4
Let us now fix an auxiliary Riemannian metric g on S2 and let pg denote the Riemannian volume measure on S2 which is induced by g. Obviously, the bilinear form
Introduction of a scalar product
is a scalar product on A'. Here (.,.), denotes the fibre metric on the bundle Hom(TC, g) E TC' @g which is induced by the metric g and the scalar product
The Hodge star operator * : dsz 4 dsz induces a linear automorphism
C" (S', As*) which will also be denoted by * and which is explicitly
(.,.)* on g.
of A' given by
(*A')@) = *(AL(t)) Vt E S1
With the help of << ., . >> and * we can now rewrite Eq. (4.3) in the form
1 Scs(A'+Bdt) = -& << A', (*o(&+ad(B))).A' >> -2 << A',*.dB >>
(4.4)
It is tempting to conclude from (4.4) that
Scs(A' + Bdt)
= - & << (A'- (&+ad(B))-l.dB), (*o( &+ad(B))). (A' - (&+ad(B))-l .dB) >>
(4.5)
If one could make sense of the latter equation, one could conclude at an informal level that the (informal) 'Lmeasure" exp(iScs(A'+ Bdt))DAL is of "Gaussian type" with "mean" (2 + ad(B))-l . d B and a "covariance operator"
- y ( * o (& +ad(B)))- '
4.5 The decomposition dl = d' @
1 However, there are two problems with Eq.
does not exist because & + ad(B) is not injective, and secondly, the operator
* 0 (g + ad(B)) is not symmetric w.r.t. the scalar product << ., . >>.
(4.5): Firstly, (& + ad(B)))-
24
Both problems can be solved by introducing the decomposition A' = d' @ d$ where
d' := {A' E A' I rdSZ,t(A'(tO)) = 0)
d: := {A' E A' I Vt E S' : A'(t) = A'(t0) E dsz,t}
where t o is a fixed point in S1 and 7Tds2,1 : dsz 2 dSz,go @ dsz,t 3 dSz,t the
canonical projection. Here dSz,gO (resp. dsz,t) denotes the space of smooth
go-valued (resp. t-valued) 1-forms on C. It can be shown that the restriction (& + ad(B)),al of the operator (& +
ad(B)) onto d' is injective and that the operator * o (6 + ad(B)) IAI is sym-
metric w.r.t. the scalar product << ., . >. Finally, by extending (& + ad(B)) in a suitable way to the space
A' := d' @ {A: . (iit(.) - 1/2) I A: E dc,t}
where is1 : [0,1) 3 t H exp(2ri(to + t ) ) E S' c C one can achieve that the extended operator, which will also be denoted by (& + ad(B)) , is a bijection
A' + A'. Thus
is a well-defined element of A'. It is not difficult to show that
m(B) := (& + ad(B))-' . dB (4.6)
(4.7)
(4.8)
Scs(A' + A: + Bdt) = Scs(A' + Bdt) + & < A:, dB >sz, k
4K Scs(A' + Bdt) = -- << A' - m(B), (*a (& + ad(B))) . (A' - m(B)) >>
holds, which means that the heuristic integral functional
1.. .exp(iScs(A'+ B ~ ~ ) ) D A '
C(B) =-Y(*o(-&+ad(B)))- ' (4.9)
is of "Gaussian type" with "mean" m(B), "covariance operator"
and "mass" I d e t ( g +ad(B))/-'/' (up to a multiplicative constant, independent
of B).
4.6
Combining Eqs. (4.2), (4.7), (4.8) we arrive at
A preliminary heuristic formula for WLO(L)
WLO(L)
25
There is a curious thing about this equation: In the expression I det( & +ad(B)) I which appears above the operator & + ad(B) denotes the obvious operator P ( S 2 x S',g) --f C"(S2 x S',g). Now, at a heuristic level, the determinant of this operator should equal3 the root of the determinant of the operator
(& +ad(@) : Cw(S1,dSZ) 4 C"(S1,dsz).
Thus, heuristically, the "measure"
d f i i (A ' ) := 1 det(& + ad(B))I exp(iScs(A' + Z3dt))DAl
has mass 1 and we can rewrite Eq. (4.10) as
x det(l,, - exp(ad(B))l,,) exp(i& < Ag,dB >sz)DA: @ DB. (4.11)
4.7
So far we have neglected one topological subtlety. Above we claimed that torus gauge fixing is a proper gauge k i n g . However, strictly speaking, this is only true if the oriented surface C in Subsec. 4.1 is non-compact. If the surface C is compact like in the case C = S2 we can not expect Eq. (4.11) to hold without modification. The modification that takes care of the topological subtleties which we have just mentioned involves a summation CnEISZ,G,Tl over the set
[S', G/T] of free homotopy classes of mappings from S2 to G/T and, for each n E [S2, G/T] , n # 0, the inclusion of a 1-form A$,,,(n) with a singularity in a
fixed point 00 of C. More precisely, the modification of Eq. (4.11) is given by
The final heuristic formula for WLO(L)
WLO(L)
exp(i& < A:,dB >p)DA: 8 DB. (4.12)
4.8 The Program
The heuristic equation (4.12) can be used as the starting point for the search of a rigorous definition of the WLOs in terms of Hida distributions. In order to make rigorous sense of the right-hand side of Eq. (4.12) one can proceed in 5 steps:
3 ~ f . 19, 101; note that we consider the special cme C = S2 where the Euler characteristic x(C) equals 2
26
0 Step 1: Make rigorous sense of the heuristic integral functional s.. * d f i i ( a i )
0 Step 2: Make rigorous sense of the whole expression SJ. ni Tr(P exp(h, A *I + A: + Ak,,(n) + Bdt) )df i i (A l )
0 Step 3: Make rigorous sense of the heuristic integral functional
J d$xCw(S2, t )
. . . e x p ( i 3 < A t , dB >Sz)DA,I 8 DB
as a Hida distribution of “Gaussian type”.
0 Step 4: Make rigorous sense of the total expression on the right-hand side of Eq. (4.12)
Step 5: Compute the expression in Step 4 explicitly.
In [24, 191 we have already completed Steps 1-3 for arbitrary G using “loop smearing” and “framing” in a similar way as in Sec. 3 and, additionally, Steps
4 and 5 in the special case where G is Abelian. We plan to complete the last two steps also for Non-Abelian G in the near future, cf. [20].
4.9
Let 7 - 1 ~ denote the Hilbert space L2-r(Hom(TC,g),pg) of L2-sections of the bundle Hom(TC,g) w.r.t. the measure pg and let H denote the Hilbert space L$& (S’, dt) of square-integrable HE-valued functions on S’. Moreover, let us
identify the spaces AL and dl with the obvious subspaces of 7-1. Then the
operator C(B) : A l --* dl can be considered as a densely defined bounded symmetric operator on 7-t = L&=(S’, dt). Setting N := A’ we obtain a Gelfand
triple N c ‘H c N*. The informal integral functional s . . . d f i i can now be defined rigorously as the unique element @A of (N)* such that
Some Details for Step 1
@i(exp(z(.,j))) = exp(i << j ,m(B) >>x)exp(-i << j , C(B) j >>x) (4.13)
holds for all j E N with m(B) and C(B) given as in (4.6), (4.9). Here (., .) : N* x N -+ W denotes the canonical pairing and << ., . >>x the scalar product of ‘H.
5 Conclusions and Outlook In this paper we have explained how white noise analysis can be applied suc- cessfully to the study of Chern-Simons models on W3 W2 x W and S2 x S1. If it turns out after the completion of the 5 steps in Subsec. 4.8 that the torus gauge fixing approach of Sec. 4 applied to Non-Abelian G will lead to the correct knot
polynomial expressions for the WLOs this will clarify most of the open ques- tions listed in Subsec. 2.5. In particular, it would provide a strong argument
27
in favor of the conjecture made in Subsec. 3.5 and it would demonstrate that it is indeed possible to establish the relations discovered in [32] between the knot polynomials and the heuristic path integral expressions for the WLOs at a mathematically rigorous level.
Acknowledgements: I would like to express my gratitude to Prof. Dr. T. Hida for giving me the opportunity to contribute to the very stimulating Conference in Nagoya last November and to the Proceedings.
References
[l] S. Albeverio, A. Hahn, and A. N. Sengupta. Chern-Simons theory, Hida distributions and state models. Inf. Dim. Anal. Quant. Prob., 6, Supple- mentary volume 1:65-81, 2003.
[2] S. Albeverio and J. Schafer. Abelian Chern-Simons theory and linking
[3] S. Albeverio and A.N. Sengupta. A Mathematical Construction of the Non-Abelian Chern-Simons F'unctional Integral. Commun. Math. Phys.,
numbers via oscillatory integrals. J . Math. Phys., 36(5):2135-2169, 1994.
186:563-579,1997.
[4] D. Altschuler and L. Freidel. Vassiliev knot invariants and Chern-Simons perturbation theory to all orders. Comm. Math. Phys., 187:261-287, 1997.
[5] S. Axelrod and I.M. Singer. Chern-Simons perturbation theory. In Catto, Sultan et al., editor, Differential geometric methods in theoretical physics. Proceedings of the 20th international conference, June 3-7, 1991, New York City, NY, USA, volume 1-2, pages 3-45. World Scientific, Singapore, 1992.
J . [6] S. Axelrod and I.M. Singer. Chern-Simons perturbation theory. 11. Differ. Geom., 39(1):173-213, 1994.
[7] D. Bar-Natan. On the Vassiliev knot invariants. Topology, 34:423-472,
[8] D. Bar-Natan. Perturbative Chern-Simons theory. J . Knot Theory and its
1995.
Ramijcations, 4:503-547, 1995.
[9] M. Blau and G. Thompson. Derivation of the Verlinde Formula from Chern- Simons Theory and the G/G model. Nucl. Phys., B408(1):345-390,1993.
[lo] M. Blau and G. Thompson. Lectures on 2d Gauge Theories: Topological Aspects and Path Integral Techniques. In E. Gava et al., editor, Proceedings of the 1993 Weste Summer School on High Energy Physics and Cosmology, pages 175-244. World Scientific, Singapore, 1994.
1111 M. Blau and G. Thompson. On Diagonalization in Mup(M, G). Commun. Math. Phys., 171:639-660, 1995.
28
[12] R. Bott and Taubes C. On the self-linking of knots. J. Math. Phys., 35( 10):5247-5287,1994.
[13] A. Cattaneo, P. Cotta-Ramusino, J. Frohlich, and M. Martellini. Topolog- ical B F theories in 3 and 4 dimensions. J. Math. Phys., 36(11):6137-6160, 1995.
[14] M. de Faria, J. Potthoff, and L. Streit. The Feynman integrand as a Hida distribution. J. Math. Phys., 32(8):2123-2127, 1991.
[15] Shmuel Elitxur, Gregory Moore, Adam Schwimmer, and Nathan Seiberg. Remarks on the canonical quantization of the Chern-Simons-Witten theory. Nuclear Phys. B, 326(1):108-134, 1989.
[16] P. Freyd, J. Hoste, W. Lickorish, K. Millett, A. Ocneau, and D. Yetter. A new polynomial Invariant of Knots and Links. Bulletin of the AMS, 12( 2) ~239-246, 1985
[17] J. F'rohlich and C. King. The Chern-Simons Theory and Knot Polynomials. Cornrnun. Math. Phys., 126:167-199, 1989.
[18] E. Guadagnini, M. Martellini, and M. Mintchev. Wilson Lines in Chern- Simons theory and Link invariants. Nucl. Phys. B, 330:575-607, 1990.
[191 A. Hahn. Chern-Simons models on S2 x S', torus gauge fixing, and link invariants 11. In Preparation.
[20] A. Hahn. Geometric derivation of the R-Matrices of Jones and lhraev. In Preparation.
[21] A. Hahn. Chern-Simons Theory on R3 in axial Gauge. Ph.D. Thesis,
[22] A. Hahn. Chern-Simons theory on W3 in axial gauge: a rigorous approach.
[23] A. Hahn. The Wilson loop observables of Chern-Simons theory on R3 in
[24] A. Hahn. Chern-Simons models on S2 x S1, torus gauge fixing, and link
Bonner Mathematische Schriften Nr. 345, 2001.
J. Funct. Anal., 211(2):483-507, 2004.
axial gauge. Cornrnun. Math. Phys., 248(3):467-499, 2004.
invariants I. J. Georn. Phys., 53(3):275-314, 2005.
[25] T. Hida, H.-H. Kuo, J. Potthoff, and L. Streit. White Noise. An infinite dimensional Calculus. Dordrecht: Kluwer, 1993.
[26] L. Kauffman. Knots. Singapore: World Scientific, 1993
[27] M. Kontsevich. Vassiliev's knot invariants. Adw. in Sow. Math, 16(2):137- 150, 1993.
29
[28] J.M.F. Labastida. Chern-Simons gauge theory: ten years after. In Trends in theoretical physics 11 (Buenos Aims 1998), pages 1-40. AIP Conf. Proc., Amer. Inst. Physics, Woodbury, New York, 1999.
[29] G. Leibbrandt and C. P. Martin. Perturbative Chern-Simons theory in the light-cone gauge. The one-loop vacuum polarization tensor in a gauge- invariant formalism. Nuclear Phys. B, 377(3):593-621, 1992.
Simons Path Integrals using White Noise Analysis. 1301 P. Leukert and J. Schafer. A Rigorous Construction of Abelian Chern-
Rev. Math. Phys., 8( 3) :445-456, 1996.
[31] L. Streit and T. Hida. Generalized Brownian functionals and the Feynman integral. Stochastic Process. Appl., 16(1):55-69, 1984.
[32] E. Witten. Quantum Field Theory and the Jones Polynomial. Commun. Math. Phys., 121:351-399, 1989.
30
A topic on noncanonical representations of
Gaussian processes
Dedicated to Professor T. Hida for his 77th birthday
YUJI HIBINO Faculty of Science and Engineering] Saga University
840-8502] Saga, JAPAN
Abstract Give a noncanonical representation of a Gaussian process X. If the Hilbert spaces H t ( X ) = L S { X ( s ) ; s 5 t } satisfy a certain condition, then we can form the Brownian motion that is used for the canonical representation of X.
Keywords: Gaussian processes; Canonical representations; Non- canonical representations; Brownian motions.
AMS Subject Classification: 60G15, 60G35
1 Introduction
Let a Gaussian process X = { X ( t ) ; t 2 0) be given by a Wiener integral
X ( t ) = F( t , u)dB(u), (1) l where B is a Brownian motion and F is a nonrandom function. In the
representation (l), Bt(X)(r a { X ( s ) ; s 5 t } ) is smaller than or equal to
&(B) for each t 2 0. If &(X) = &(B) holds, the representation (1) is
canonical with respect to the Brownian motion B (see [4]). We remark that
&(X) = &(B) is equivalent to H t ( X ) = Ht (B) , where H t ( X ) is a closed
linear span of { X ( s ) ; s 5 t } , since X is Gaussian.
The concept of canonical representation was originated by Lkvy [4]. For
the canonical representation, we understand that the information generated
by the past of X up to time t is equal to that of B. In this case, the
randomness of X is equally undertaken by B, which is called the innovation of X. It is known that the canonical representation is essentially unique, if
it exists (see [2, 31). Generally speaking, a centered Gaussian process is determined by its
covariance. However, it is difficult to obtain the canonical representation
31
or the multiplicity directly from the given covariance, except for some spe-
cial cases such as stationary processes and so on. Even if a representation
like (1) is given, it may not be canonical. We occasionally met noncanonical
representations (see [4], [5], etc.) In the joint paper [l], it has been given how to construct a noncanonical
representation of a Brownian motion so as to be independent of a given
finite-dimensional subspace. The class includes examples of noncanonical
representations given by LBvy. In this article we shall propose a method of constructing the canoni-
cal representation when we are given a noncanonical representation of the
form (1) having a finite-dimensional orthogonal complement of H t ( X ) in
Ht(B). The fact shows a kind of usefulness of noncanonical representations.
The base of Ht(B) 8 H t ( X ) is obtained by solving the integral equation
derived from the Hida criterion [2].
2 Noncanonical representations
It is known in [l] that, for any N E N, we can construct a noncanonical rep-
resentation of a Brownian motion having a given N-dimensional orthogonal
complement as in the following way:
Let 91, g2, . . . , g~ E Lfo,[O, m) be linearly independent in L2[0, t ] for any t > 0. Define a Volterra-type integral operator Kg : L2[0,co) + L2[0,m) bv
where r(t)-' = (rij(t)) = (Ji g i (u )g j (u)du) -'. It is noted that the matrix
r(t) is invertible for any t > 0, since g ( t ) = (g l ( t ) , ga ( t ) , . . . , g N ( t ) } is
linearly independent.
Theorem 1 ([l, Theorem 2.11) Define a Gaussian process B, by
Then B, is a Brownian motion and has a noncanonical representation with respect to B:
H t ( B ) = Ht (Bg) EB LS g j ( u ) d B ( u ) ; j = 1,2, . . . {I' where K l is the formal adjoint operator of K,.
32
We prefer to write (3) symbolically
d,(t) = ( I - K,)B(t), Bg(0) = 0.
By using the theorem above, the canonical representation of X can be obtained from a noncanonical representation having a finite-dimensional or-
thogonal complement Ht (B) e H t ( X ) as follows:
Theorem 2 Let a Gaussian process X be given by (1). canonical representation satisfying
If it is a non-
then X has unit multiplicity, and has the canonical representation
X ( t ) = l- F(t , u)dB,(u)
with respect to B,, where B, is defined by (3) and F(t,.) = ( I - K,)F(t,.).
Proof: The Brownian motion B, defined by (3) satisfies (4). Since ( 5 ) holds, we have Ht(B,) = H t ( X ) for any t > 0. Because of the uniqueness
of canonical representation, B, is an innovation of X . Thus the multiplicity
of X is one.
Suppose the representation (6) of X is canonical. Since
t
X ( t ) = F ( t , U ) B , ( U ) d U J O
= Jd 'F( t , " ) ( I - K,)B(u)du
= I ' ( I - Kl )F( t , u)B(u)du,
wesee ( I -K i )F ( t , . ) = F(t , . ) . Since ( I -K,)( I -K;) = I , wecanconclude
0 F( t , .) = ( I - K,)F(t, .).
If X defined by (1) is noncanonical with respect to B, clearly H t ( B ) 2 H t ( X ) for some t > 0. When the orthogonal complement H t ( B ) 0 H t ( X ) is finite-dimensional and we can get ( 5 ) , then Theorem 2 can be applied in order to obtain the canonical representation of X.
33
Remark If we could construct, for a system of functions g1,$2, .. . E
Lfo,[O, oo), a noncanonical representation of a Brownian motion B having an infinite-dimensional orthogonal complement:
then, even for a given noncanonical representation of X having an infinite-
dimensional orthogonal complement, the innovation and the canonical rep-
resentation of X would be obtained by applying the idea of Theorem 2.
Acknowledgments. The author would like to express sincere gratitude to
Professor T. Hida for his constant encouragement. Thanks are due to Dr.
H. Muraoka for fruitful discussion with him.
References
[l] Y. Hibino, M. Hitsuda and H. Muraoka; Construction of noncanonical representations of a Brownian motion. Hiroshima Math. J . 27 (1997), 439-448.
[a] T. Hida; Canonical representations of Gaussian processes and their ap- plications. Mem. Coll. Sci. Univ. Kyoto 33 (1960), 10S155.
[3] T. Hida and M. Hitsuda; Gaussian Processes, Representation and Ap-
plications, Amer. Math. SOC. (1993).
[4] P. L&y; A special problem of Brownian motion and a general theory of .Gaussian random functions. Proc. of 3rd Berkeley Symp. Math. Stat.
and Prob. 2 (1956), 133-175.
[5] H. P. McKean, Jr.; Brownian motion with several-dimensional t ime, Theor. Probability Appl. 8 (1963), 335-354.
34
Integral Representation of Hilbert-Schmidt Operators
on Boson Fock Space'
U N CIG JI DEPARTMENT OF MATHEMATICS
RESEARCH INSTITUTE OF MATHEMATICAL FINANCE CHUNGBUK NATIONAL UNIVERSITY
CHEONGJU 361-763 KOREA E-MAIL: uncigj iQcbucc . chungbuk. ac . kr
Abstract
An integral representation of Hilbert-Schmidt operators on Boson Fock space is proved with explicit forms of integrands which is a quantum version of (classical) Clark-Haussmann-Ocone formula.
Mathematics Subject Classifications (2000): Primary: 60H40; Secondary: 81925.
Key words: white noise theory, Fock space, integral kernel operator, operator symbol,
Hilbert-Schmidt operator, chaotic expansion
1 Introduction
The quantum stochastic calculus of It6 type formulated by Hudson and Parthasarathy
[9] has been extensively developed in [20], [27] and the references cited therein. In particular, the stochastic integral representations of quantum martingales have been studied by many
authors (see [2], [lo], [12], [19], [20], [28], etc). An integral representation of Hilbert-Schmidt martingales was studied in [ll]. The chaotic expansion of Hilbert-Schmidt operators was established in [3] and more general theory was developed in [14].
On the other hand, the white noise theory initiated by Hida [7] has been considerably
developed as an infinite dimensional distribution theory with wide applications: to stochastic
analysis, Feynman path integral, quantum physics and so on, see [S], (181 and (221. Recently,
white noise approach to quantum stochastic calculus is successively studied in [13], [15], [23]- [26]. Specially, in [ l6], explicit forms of integrands in the integral representation of quantum
martingales was obtained by using the quantum white noise derivatives. In this paper, by using the white noise theory we study an integral representation of
Hilbert-Schmidt operators on Boson Fock space with explicit forms of integrands. This paper is organized as follows: In Section 2 we recall the basic notions in white
noise theory. In Section 3 we review the theory of white noise operators with its chaotic
expansions. In Section 4 we study an integral representation of Hilbert-Schmidt operators on Fock space.
'This work was supported by grant (No. R05-2004-000-11346-0) from the Basic Research Program of the Korea Science & Engineering Foundation.
35
2 Preliminaries
2.1 White Noise Triplet
Let H = L2(R+,dt) be the (complex) Hilbert space of L2-functions on R, = [O,m) with respect to the Lebesgue measure dt and the norm of H is denoted by 1 . lo. Let A be a
selfadjoint operator (densely defined) in H satisfying that there exist a sequence
m
~ < X O ~ X ~ < X Z < . . . , I (A- ' I I ; ,=CXy2<m, j=O
and an orthonormal basis { e j } g o of H such that Aej = Aje j . For p E R we define
m
I € I; = I APE I: = X?I (€, e j ) 1 2 , E E H. j = O
Now, for p 2 0, setting Ep = {< E H ; I E 1, < 03) and defining E-, to be the completion of
H with respect to 1 . I-,, we obtain a chain of Hilbert spaces {E, ; p E R}. Define their limit spaces:
E = proj lim E,, E* = ind lim E-,, p-m P-m
where E* is the strong dual space of E. Identifying H with its dual space, we have
E C E, c H = L 2 ( R + , d t ) c E-, c E*, p 2 0. (2.1)
From more general study in [22], we assume that
(Al) for each function E E E there exists a unique continuous function on R+ such that
€( t ) = F(t) for almost all t 2 0;
(A2) for each t 2 0 the evaluation map 6, : < H E(t), E E E, belongs to E' and the map
R+ 3 t H ht E E* is continuous with respect to the strong dual topology of E".
Let p E R and r(E,) be the (Boson) Fock space over E,, i.e.,
1 m
F(E,)= + = ( f " ) r = O ; f " E E ~ , II+Il;=~n!lf"I;<.. , { n=O
where EP is the symmetric n-tensor power of E,. By taking Fock space from (2.1), we have the natural inclusions:
r(E,) c r ( H ) c r(E-,), P 2 0.
By the general duality theory, r(E-,) is the strong dual space of r(E,). Setting
r ( E ) = proj limF(E,), r ( E ) * = indlimr(E-,), ,-a P-m
we have a complex Gelfand triple:
r ( E ) c r ( H ) c r (E)* (2.2)
which is referred to as the Hzda-Kubo-Takenaka space [17], see also [7, 8, 18, 221. The
canonical C-bilinear form on r ( E ) * x r ( E ) is denoted by ((., .)).
36
2.2 Admissible Triplet
For p E R we define
m
111 4 111; = Cn!e2pnIfnI& 4 = (fn) E r ( ~ ) . n=O
For p 2 0 we put Gp = {4 = (fn) E r ( H ) ; 111 4 I l lp < ca} and G-, to be the completion of r ( H ) with respect to (11 . I I I - p . Then {Gp ; p E R} form a chain of Hilbert spaces satisfying
G = proj lim C Gp C = r ( H ) c LP c G* = ind lim GPp. P-m P-m
Note that B is a countable Hilbert space and that G and G+ are mutually dual spaces. Then,
we have
r ( E ) C B c r ( H ) c G* c r (E ) *
of which the proof is immediate from the definition of the norms. The canonical C-bilinear
form on 8' x G is denoted by ((., .)) again.
3 White Noise Operators
A continuous linear operator from r ( E ) into r (E ) * is called a white noise operator. The
space of all white noise operators is denoted by L ( r ( E ) , r ( E ) * ) equipped with the bounded convergence topology.
3.1 Operator Symbol and Integral Kernel Operators
For each < E E, 4 < = (l,+. '>$ P ,...)
is called an exponential vector or a coherent vector. Note that the exponential vectors
{ 4 ~ ; < E E } span a dense subspace of r ( E ) , hence of Gp for all p E R and of 9'. Thus
each E E L( r (E) , r (E)* ) is uniquely specified by its matrix elements with respect to the exponential vectors. We put
h
(3.1) - =(& 11) = (Wt, 4 4 > ( 7 11 E
which is called the symbol of Z.
Theorem 3.1 [21, 221 A C-valued function 0 on E x E is the symbol of a white noise operator Z E L( r (E) , r (E)*) zf and only i f
(i) 0 is Giteaux entire;
(ii) there exist C 2 0, K 2 0 and p 2 0 such that
P(E, d I I C ~ X P K ( I E I: + I 7 I:), E , 7 E E .
37
For each given qrn E (E@('+m))*, 1,m 2 0, an integral kernel operator
- - - = = =l,m(qm) = =l,m(Kl,m)
is in L(F(E), r(E)*) satisfying
8(<,q) = (Kl,rn<@m, q@l)e((,q) = (qm, q@1C3<@m)e(E39), <,q E E , (3.2)
where K I , ~ E L(E@", (E@')*) corresponds to tqm E (E@('++"))* under the kernel theorem.
The existence of the integral kernel operator is immediate from Theorem 3.1 since Z given in (3.2) satisfies the conditions (i) and (ii) in Theorem 3.1. For each y E E*, the annihilation operator ay and the creation operator a; satisfy
h
&(<,d = (Y,<)e(S'9)? a3y(<,17)= (Y, 17)e(€'", <,oe E
Then for any ( E E , a< and a; are continuous linear operators acting on r ( E ) (or G). Also, a( has a unique extension to a continuous linear operator acting on r (E)* and a; E
L( r (E) , r (E ) ) . For each t 6 R+, we write at = and a: = a;*. Note that for each @ E G*, Dt@ = at@ is well-defined for almost all t E R+. Moreover, we have the following
Proposition 3.2 ( [ l ] , [15]) For each @ = (F,) E G*, Dt@ = ((n+l)F,+l(t, .))Fz0 for almost all t E R+. Moreover, if @ E Gp and q < p - log 4, then Dt@ E Gq for almost all t 6 R+.
Remark 3.3 The kernel distribution of an integral kernel operator El,rn(+,) is uniquely
determined whenever taken from the subspace
(E@('+m))* sym(l,m) = {K l , rn E (E@('+~))* ; si,rn(Kt,m) = nl,m} 7
where sl,+ is the symmetrizing operator with respect to the first 1 and the second m variables independently.
3.2 Chaotic Expansion of Hilbert-Schmidt Operators
Let p , q E R. For each K I , ~ E L(Efm, Eqg') there exists a unique operator &,(Kl,J E L( r (Ep) , r (Eq) ) such that
4,m(K1,~)-(<, V ) = ( K I , ~ < @ ~ , v@') I t, 1) E E (3.3)
and 1) ~+(KI,,,,) I J o p 5 only if Z I , ~ ( K ~ , ~ ) E Lz(r(Ep),I'(Eq)), in that case
) I K I , ~ \ l o p , see [14]. Moreover, K I , ~ E L2(Efm, E p ) if and
11 h,T71(K1,Wl) IIHS = (1 Kl,m (IHS,
where &(X, 9) is the space of Hilbert-Schmidt operators from a Hilbert space X into another
Hilbert space 9 of which the norm is denoted by ( 1 [ I H S .
38
Theorem 3.4 [14] Let p,q E R. For any Ei E L( r (Ep) , r (Eq ) ) there exists a unique family of operators Li,,,, E L(Efm, EP) , l ,m 2 0, such that
m
which is called the chaotic expansion of 2, where the series converges weakly in the sense that
05
((54, +)) = C ((IL,m(Ll,m)4> +)) , 4 E r ( E p ) , 4 E r ( w . l,m=O
Theorem 3.5 [I41 Let p , q E R. Given E E Lz(l'(Ep), r(Eq)) let m
be the chaotic expansion. converges in L2(r(Ep), l?(Eq)). Moreover, we have
Then K L , ~ E L2(Efm,EP) and the right hand side of (3.5)
m m
4
4.1 Quantum Stochastic Integrals
A family {=: t } t& C L( r (E) , r (E)*) of white noise operators is called a quantum stochastic process, where T C R+ is a (finite or infinite) interval. For a quantum stochastic process
{ E t } t G T , if t H ((E& +)) is integrable on T for any d ,+ E r ( E ) and if there exists ZT E
L(l?(E),r(E)*) such that
Integral Representation of Hilbert-Schmidt Operators
((ET4r $)) = / ((st47 $)) dt, 4, '$ E r(E),
then the process {5 t } tET is said to be integrable on T . In that case, we write
and call it the white noise integral of {Et}tGT on T .
Lemma 4.1 Let {Et} tE~ be a quantum stochastic process, where T C R is an interval. Assume
(i) for any pair <, 17 E E, the map t H Et(<, 7) is measurable on T ;
(ii) there exist constants a 2 0, p 2 0, a locally integrable nonnegative function K E L;,,(T) and a null set N c T such that
l ~ t ~ < , ~ ~ l ~ ~ ~ ~ ~ ~ ~ ~ { ~ ~ l < lE , s € E , t E T \ N
39
Then for a , t E T , the white noise integrals are defined:
=,a,ds. l- a:Z,ds, I' PROOF.
As a simple application of Lemma 4.1, the white noise integrals:
It is straightforward by applying Theorem 3.1, see also [23].
At = a,ds, A; = l a : d s
(4.1)
I
are well-defined and called the annihilation process and the creation process, respectively.
Thus the white noise integrals in (4.1) are called the quantum stochastic integrals against
the creation process and the annihilation process, respectively, and we write
If {Es} is an adapted process (see [6, 23]), we write
dA:Z, = Z,dA:. l I ' For more study of white noise approach to quantum stochastic integrals, we refer to [23].
4.2 Integral Representation
Let E E L2( r (H) , r ( H ) ) with the chaotic expansion
W
z = C zl,m(Kl,m), ~ 1 , ~ E L~(H",H'"). I,m=O
Then for each 1, m 2 0 by the kernel theorem there exists ~ 1 , ~ E H8' @I Horn such that
Therefore, we write
l,m=O
m and consider
Mt = EtSEt = C I ~ , m ( l E ~ $ m K ~ , m ) , t 2 0, l,m=O
where E t , t 2 0, is the conditional expectation (see also [24], [25]) defined by
Et@ = (1;31C), @ = (F,) E B'
(4.2)
40
and then Et E C(Gp,Gp) for any t 2 0 and p E R. On the other hand, for each l,m 2 1 we have
Hence there exists a null set N c R+ such that for all t E R+ \ N
Therefore,
Since for s E R+ \ N
we put
and
Then E J - ~ , ~ ( S ) and f i ,m-l(s) satisfy the conditions (i) and (ii) in Lemma 4.1 and so the
E1,,-1(S) = &I,,(s) = 0, s E N . ...
integrals
are well-defined, and EI,,,-l(s) and FI-~,,(S) are adapted. Hence by (4.3) we have
41
Let (4,) be a complete orthonormal basis for l?(H). Then for any p E R, {l?(e-PI)$,} is a complete orthonormal basis for Gp, where r(e-PI) is the second quantization of e-PI. Therefore, for each 2 E L 2 ( r ( H ) , T ( H ) ) and p , q 2 0 we have
IlZl/HS;p,-q = l l ~ ( e - q ~ ) 2 r ( e - p ~ ) l l H ~ ,
where I I ~ I I H S ; ~ , - ~ is the Hilbert-Schmidt norm of Z on L2(Gp, G-J and 11 . 1 1 ~ ~ = 11 . [IHS;O,O.
In particular, for any E B H G ~
e 114,rn(X~,rn)IlkS IIl',m(Xl,m)II~s;p,-g - - cZpm -2q1
where s = ( ~ 1 , . . . , s l ) and t = (t l , . . . , tm). Therefore, we have
Hence the maps
R + 3 s ++ E(s) = C G , m - ~ ( s ) , R+ 3 s H F(s ) = C & - I , ~ ( S ) (4.6) l,m l,m
belong to L2(R+, L2(Gp, Lq) for any p , p > 0 and then from (4.2) and (4.5) we have
More generally, we have the following theorem
Theorem 4.2 Let p , q E R and let 2 E C2(Gp,Gq). T h e n there exist adapted processes {E(s)},?o and { F ( S ) } ~ ~ O contained in L2(R+,L2(Gp+,,Gq-s) f o r any r , s > 0 such that Mt = EtEEt admits the integral representation (4.7).
Theorem 4.3 [4, 261 Let {Z,}r=.=, and 2 be in L( r (E ) , r (E ) * ) . Then En converges to S in L(l?(E), r (E)* ) i f and only zf the following conditions hold:
(i) z,(C, 7) converges to z(c, 7) f o r each t, 7 E E .
(ii) There exist p 2 0, a > 0 and K > 0 such that
lzn(t,q)l 5 Kea(lElg+lol~), (,7 E E , n E N.
By applying Theorems 4.2 and 4.3 we have the following
Theorem 4.4 Let p , q E R and let 2 E L2(GP,Gq). (E(s)},>o and {F(s)},>o contained in L2(R+,L2(Gp+,,Gq-s) f o r any r, s > 0 such that
T h e n there exist adapted processes
m m
E = ( ( ~ 4 ~ , 40)) r + J E ( s ) d A , + 1 F ( s ) d A : . (4.8)
42
4.3 Integrands in Integral Representation
@((, 7 ) = ( E , 11) e(c,v),
The number operator N is uniquely specified by the symbol
t , 7 E E
and its Wick exponential wexp(-N) is defined by
In general, for two integral kernel operators Zl,,(K) and Zlt,m~(K') their Wick product or
normal-ordered product is defined by
ZI,,(K) 0 ZV,d(K') = Zl+l',m+m'(K k3 K') ,
see [5]. Let p , q E R and Kl,,,, E L(Efm, E?). Then from (3.2), (3.3) and (4.9) we have
where the series converges in L(r(E(pvq)+r), I?(&)) for any r > 0 with (2p'/2))/(-r10gp) < 1. For the proof, we refer to [14].
Let c E E. Then by Proposition 4.1 in [5] and (4.10) we have
a<I~,m(K~,rn) = lI1-1,m(c * Kl,m)> Il,m(Kl,m)u: = mI~,m-l(K~,m * 0, where * K I , ~ E L(Efm, Ef('-') ) and Kl,m * c E L(Ef("-'), E Y ) satisfying
( c * Kl,naE@m, P ( 1 - 1 ) ) = (Kl,mpm, 63 c ) , ((K1,m * oE@' " - " , P ) = (Kl,m(€@(m-i) 63 0, P) , € 3 17 E E.
On the other hand, if q m E Hol k3 Horn corresponds to K I , ~ , then for almost all t E R+ 6t*Kl,m and Kl,,,,*bt are well-defined as operators corresponding to Kl ,m( t , .; .) and ~ 1 , ~ ( . ; t , .), respectively. In that case, we write
DtI~,m(%n) = &1,m(61,m(t, .; .)).
Then we have
Hence from (4.4) we have
( o t h , m ( Q , m ) * ) * = mIl ,m-l(Kl ,m(. ; t , .)).
G,m-i(S) = (DsEsI~,m(~l,m)*Es). , ~ - I , ~ ( S ) = DsEsIl,m(Kl,m)Esr s E R+ \N.
Therefore, for each 5 E LZ(I?(H),r(H)) , the maps E(s ) and F ( s ) in (4.6) are given by
E(s) = (D,E,Z*E,)' , F(s) = D,E,ZE,. (4.11)
Finally, by (4.11) and Theorem 4.4 we have the following theorem
Theorem 4.5 Let p , q E R and let 5 E L2(Gp,Gq). T h e n E admits the following integral representation:
W
3 = ((940, do)) I + 1 (DsE3E*E8)*
43
References
[l] K. Aase, B. 0ksenda1, N. Privault and J. Ubere: White noise generalizations of the Clark-Haussmann-Ocone theorem with application to mathematical finance, Finance Stochast. 4 (2000), 465496.
[2] S. Attal: A n algebra of non-commutative bounded semimartingales: square and angle quantum brackets, J. Funct. Anal., 124 (1994), 292-332.
[3] S. Attal: Non-commutative chaotic expansion of Halbert-Schmidt operators on Fock space, Commun. Math. Phys. 175 (1996), 43-62.
[4] D. M. Chung, T. S. Chung and U. C. Ji: A simple proof of analytic characterization theorem for operator symbols, Bull. Korean Math. SOC. 34 (1997), 421436.
[5] D. M. Chung, U. C. Ji and N. Obata: Higher powers of quantum white noises in terms of integral kernel operators, Infinite Dimen. Anal. Quantum Probab. Rel. Top. 1 (1998), 533-559.
[6] D. M. Chung, U. C. Ji and N. Obata: Quantum stochastic analysis via white noise operators in weighted Fock space, Rev. Math. Phys. 14 (2002), 241-272.
[7] T. Hida: “Analysis of Brownian Functionals,” Carleton Math. Lect. Notes, no. 13, Carleton University, Ottawa, 1975.
[8] T. Hida: “Brownian Motion,” Springer-Verlag, 1980.
[9] R. L. Hudson and K. R. Parthasarathy: Quantum Ito’s formula and stochastic evolu- tions, Commun. Math. Phys. 93 (1984), 301-323.
[lo] R. L. Hudson and J. M. Lindsay: A non-commutative martingale representation theorem for non-Fock quantum Brownian motion, J . Funct. Anal. 61 (1985), 202-221.
[ll] R. L. Hudson, J . M. Lindsay and K.R. Parthasarathy: Stochastic integral representation of some quantum martingales in Fock space, in “From Local Times to Global Geome- try, Control and Physics,” Proc. Warwick Symposium 1984/1985, pp. 121-131, Pitman RNM, 1986.
[12] U. C. Ji: Stochastic integral representation theorem for quantum semimartingales, J. Func. Anal. 201 (2003), 1-29.
[13] U. C. Ji and N. Obata: Quantum white noise calculus, in “Non-Commutativity, Infinite-
Dimensionality and Probability at the Crossroads (N. Obata, T. Matsui and A. Hora, Eds.),” pp. 143-191, World Scientific, 2002.
[14] U. C. Ji and N. Obata: A role of Bargmann-Segal spaces in characterization and ex- pansion of operators on Fock space, J. Math. SOC. Japan. 56 (2004), 311-338.
[15] U. C. Ji and N. Obata: Admissible white noise operators and their quantum white noise derivatives, to appear in “Infinite Dimensional Harmonic Analysis (H. Heyer, T. Kawazoe and K. SaitB, Eds.),” World Scientific.
[16] U. C. Ji and N. Obata: Annihilation-derivative, creation-derivative and representation of quantum martingales, preprint, 2003.
[17] I. Kubo and S. Takenaka: Calculus on Gaussian white noise I, Proc. Japan Acad. 56A (1980), 376-380.
44
[18] H.-H. Kuo: “White Noise Distribution Theory,” CRC Press, 1996.
1191 J. M. Lindsay: F e m i o n martingales, Probab. Theory Related Fields 71 (1986), 307- 320.
[20] P.-A. Meyer: “Quantum Probability for Probabilists,” Lect. Notes in Math. Vol. 1538,
Springer-Verlag, 1993.
[21] N. Obata: An analytic characterization of symbols of operators on white noise function- als, J . Math. SOC. Japan 45 (1993), 421-445.
[22] N. Obata: “White Noise Calculus and Fock Space,” Lect. Notes in Math. Vol. 1577,
Springer-Verlag, 1994.
[23] N. Obata: Generalized quantum stochastic processes on Fock space, Publ. RIMS, Kyoto Univ., 31 (1995), 667-702.
[24] N. Obata: Conditional expectation in classical and quantum white noise calculi, RIMS Kokyuroku 923 (1995), 154-190.
1251 N. Obata: White noise appmach to quantum martingales, in “Probability Theory and Mathematical Statistics (S. Watanabe et al. Eds.),” pp. 379-386, World Scientific, 1996.
[26] N. Obata: Wick product of white noise operators and quantum stochastic differential equations, J. Math. SOC. Japan, 51 (1999), 613-641.
[27] K. R. Parthasarathy: ”An Introduction to Quantum Stochastic Calculus,” Birkhauser, 1992.
[28] K. R. Parthasarathy and K. B. Sinha: Stochastic integral representation of bounded quantum martingales in Fock space, J. Funct. Anal., 67 (1986), 126-151.
45
THE DAWN OF WHITE NOISE ANALYSIS
IZUMI KUBO
Department of Environmental Design, Faculty of Environmental Studies, Hiroshima Institute of Technology
2-1-1 Miyake, Saeki-ku, Hiroshima 731-5193 Japan bbo@cc.zt-hiroshama. ac.jp
Abstract
The author gives a historical overview of Brownian motion, Srstly. Then he is going to see how white noise analysis was established in time of its dawn. Further, some comments, which are related with primitive questions, will be stated.
$1. Short history of Brownian motion
Titus Lucretius Carus (BC50 [l]) is the first scholar who mentioned the importance of
random movement of particles. He watched irregular movement of particles in sunbeam in dark house. By the observation, he could conclude the existence of atoms, which are invisible and move very rapidly and randomly’.
In 1827, Robert Brown observed vital molecules contained in pollen which move ac- tively in water under a microscope. He treated pollen grains of many kinds of flowers;
Clarckia pulchella, (Enothera, Gramineae, Onagrariae, Asclepiadeae, Periploceae, Apoc- ineae, Orchideae, Cryptogamous plants, Mosses, Equisetum, Equisetd and Phaenogamous plants etc.
After him, we call the movement Brownian motion2. Such moving organic particles had been found already by Buffon, Needham and Spallanzani in 18th century as mentioned
by himself and by Jean Perrin [8]. They thought that the vitality of particles causes the movement.
However, Brown (1828 [2]) went on further. He examined various products of organic
and inorganic bodies; for examples, the gum-resins, pit-coal, dust, soot, a fragment of window-glass, rocks of all ages, a fragment of Sphinx, aqueous rock, igneous rock, lava, travertine, stalactites, obsidian, pumice, manganese, nickel, plumbago, bismuth, anti- mony, arsenic and so on. Small particles, in which organic remains had never been found,
moved in water randomly!
Albert Einstein (1905 [4]), gave a kinetic theory of Brownian motion, the motion of small particles in liquid, firstly. Einstein’s model, which is so called Brwonian mo- tion today, had been already studied by Louis Bachelier in the theory of speculation. His fluctuation-dissipation theorem proposed a method to determine Avogadro number.
Marian Smoluchowski (1906 [5]) also investigated a similar formula.
lLouis George Gouy pointed out the worth of Lucretius’ atoms as referred in [8] ‘In this section, the author denotes the motion of Brown’s particle in liquid by bold style Brownian
motion. Italic style Brownian motion means the stochastic process as its mathematical model established by Wiener and L6vy. The priority of research of the process belongs to L. Bachlier.
46
Perrin (1909 [7]) made experiments of various particles in various liquid, and got
Avogadro number by Einstein’s formula. One of the best results was 6.88 x (the newest one is 6.0221367 x This is a fruit of statistical physics developed by Maxwel, Boltzmann and others, and the victory of Atomistik against Energitik.
Brownian motion as a stochastic process was firstly investigated by Louis Bachelier
(1900 [3]) as thesis of financial theory for the doctorate in mathematical sciences. His spec-
ulation model assumes that logarithmic stock prices are exchanged by random walks. He studied also Brownian motion as their limit process. He discussed the non-differentiability
of its paths and other properties without modern probability theory, which was established
later.
Norbert Wiener (1923 [9]) and Paul L6vy (1940 [la]) constructed Brownian motion and researched in the framework of modern probability theory. In particular, each of them
constructed Brownian motion and established analysis of functions of Brownian motion by his own way. Wiener gave a method of analysis of non-linear functions of Brownian motion, which is so called Wiener chaos (1938 [ l l ] ) .
Kiyosi It6 (1944 [l3]) wrote the his most famous article on stochastic integral, which
serves very important basic theorems in many fields including mathematical finance theory started by Bachelier. He gave also a very beautiful and useful theory of Wiener-It6 expansion (1951 [14]).
For a physical model of motion of a particle in liquid, Bachelier-Einstein’s Brownian motion does not take resistance force of liquid into the equation. Considering Stoke’s force, Paul Langevin (1909 [6]) gave an equation of the velocity of Brownian motion and Gorge Eugene Uhlenbeck and Leonard Salomon Ornstein (1930 [lo]) discussed the
corresponding process. They are so called Langevin equation and Ornstein-Uhlenbeck’s Brownian motion, respectively.
Since Stoke’s force can be applicable only to the case that a particle moves with
constant velocity, Ryogo Kubo (1966 [Zl]) used Boussinesq’s force with memory and his
linear response theorem. His formula fits to the experiment of Brownian motion by Kohji Obayashi, Tomoharu Kohono and Hiroyasu Uchiyama (1983 [29]).
In the following section, the author will see the root to white noise analysis with over
view of several key theories.
$2. Hida’s research before 1975
In 1975, the new approach of white noise analysis was proposed by Takeyuki Hida.
How did he get its idea? Here some of his results, which consist the basis of the theory3,
will be picked up. White noise is originally understood as the derivative of Brownian motion. Bachelier asserted and LBvy discussed precisely, non-differentiability of Brownian motion. It is interesting that Hida also showed a precise continuity of Brownian motion with multi-parameter in his early paper (1958 [15]). By the reason of non-differentiability, a rigorous treatment of white noise by Izrail’ Moiseevich Gel’fand’s generalized process
[17] is necessary. The use of reproducing kernel space is one of his important idea. He mentioned it
already in 1960 for the study of canonical representation problems. In 1967, he and Ikeda
3Many papers referred here are selected in [30].
47
applied it to study generalized process. It allowed a free calculus on non-linear functionals
of white noise. The study of infinite rotation group was begun in 1964 and was discussed in terms of white noise analysis later.
The following notations will be used often later. Let Eo be a Hilbert space with
norm 11 . 110 and E C Eo c E‘ be a nuclear triplet. Then Bochner-Minlos’ theorem (see
[17]) guarantees the existence of Gaussian measure p on E’ with characteristic functional
C( ( ) = e~p[--JI(l(~], ( E E . The measure p is called the standard Gaussian measure.
The infinite dimensional rotation group O(E’) is defined by
1
2
O(&’) = {h’ : h is a homeomorphism of &, llhfllo = l l f l lo for f E E } ,
where h* is the dual homeomorphism of h defined by (h“z, .$) = (z, he) for ( E E . Then
h* E O(&’) preserves the measure p and gives a unitary operator uh on (L2) = LZ(&’, p ) by (Uhf)(Z) = f (h*z) for z E E‘. For simplicity we note O(w) = O(&’) and call it
m-rotation group.
1. Properties of Paths
LBvy’s Brownian motion (1958 [IS]).
Theorem A. {r (2N( log r ( + clog I log rl}1/2. Then
Hida proved the upper and lower estimates for uniform continuity of N dimensional
Let { B ( x ) ,x E RN} be Liwy’s Brownian mot ion and put cpc(r) =
(2) If c > 8N + 1, for almost every w, there exits a positive number p = p(w) such that
r = )x - y1 < p implies JB(x) - B(y)l _< cpc(r)
(ai) If c < 1, for almost every w , there exits a pair (z, y) for any p > 0 such that
(0 < r = Ix - yI < p) . lB(x) - W Y ) I > cpdr)
2. Canonical Representation
A stationary Gaussian process X ( t ) can be represented in the form
X ( t ) = J” k( t - U ) d B ( U ) -m
by stochastic integral of Brownian motion {B( t ) : t E R} under a suitable condition.
Hida (1960 [IS]) showed that reproducing kernel Hilbert spaces are useful for studying
representation and causal analysis of Gaussian processes, which are linear functionals of
Brownian motion.
Let { X ( t ) } be a continuous Gaussian process of mean 0 with the covariance function
r(t, s) and M t ( X ) = linear span of { X ( s ) : s 5 t } . We see his results for the case that n M t + h ( X ) = M , ( X ) and that n M , ( X ) = (0 ) . h>O t
48
Theorem B. There exist Gaussian random measures {B(’)(t)} independent of each other and kernels {F,(t,u)} such that
For the analysis, he used the reproducing kernel Hilbert space 31 = 31(r) with the reproducing kernel r ( t , s ) . Put at = N t ( r ) = linear span of {r(s, .) : s 5 t } and let E( t ) be the projection form 31 onto 3tt. Applying Hellinger-Hahn theorem to the resolution
of unit {E( t ) } , he obtained the theorem. Prediction theory is its application.
3. Projective limit of spheres
Hida and Hisao Nomoto (1964 [18]) discussed the measure space given by the projective limit of spheres with uniform measures. Let S, be the n-dimensional sphere given by
x: +xi +. . . + = n + 1 and p,, be the uniform probability measure of S,. For m > n, define projection rn,, by
n f l rn,rn(Zl> ~ 2 , . . . zrn+~) = (X I , 22,. . . , xn+l) & + Z; + . . . + x:+1
Then we have a projective system {(Sn,p,,,~n,rn)}. Let us denote by (Sm,p,) its projec- tive limit space and let r,, be the projection from S, onto S,. They showed
Theorem C. (Sm, p,) is a standard Gaussian measure space; that is, {&(x) = x,} is a independent Gaussian sequence subject to N(0, l ) for x = (XI, 22,. . . , xn+l,. . .) E S, with respect t o p,.
Rotational flows on S, approximate flows on S,, which are called finite dimensional flows and are members of m-rotation group (Nomoto 1966, 1967 [19]). Further, Hida-
Nomoto (1967 [20]) constructed band limit Gaussian processes by the approximation.
4. Non-linear Analysis
For non-linear analysis of Gaussian and Poisson processes, Hida and Nobuyuki Ikeda
(1967 [22]) presented a method by means of reproducing kernel Hilbert spaces in Berkley
Symposium. They showed that the Wiener-It6 decompositions are given very naturally.
In the paper, they used theory of Laurant Schwartz’ distributions or generalize functions of Gel’fand [17] successfully.
Since this fact is directly related to white noise analysis, here its brief sketch is given.
We observe Schwartz’ S, S‘ and the standard Gaussian measure p. Then, in the sense of
Gel’fand’s generalized process, Brownian motion is identified with B(7) = /q(u)dB(u) =
Their first point is that C(c - 7) is positive definite and can be a reproducing kernel of
a Hilbert space 3. The second is that functional derivatives operate on the Hilbert space
and that Taylor expansion corresponds to Wiener-It6 decomposition. The 7-transform4
( G V ) , 11 E S.
40riginally, 7 was denoted by T .
49
defined by
is an isomorphism from (L2) = L2(E', p) onto F. The Taylor expansion
m 1 C(E - 7)) = exp [ - $ - 111121 = c C(E)KP(OC(17)7
KJE, 7) = ; ( / E ( t ) S ( t ) q
p=o
corresponds to Wiener-It6 decomposition. Let Fp be the Hilbert space with reproducing kernel K,(J,v). For f in a symmetric L2-space E2(Rn), I p ( f ) is the multiple Wiener integral o f f :
Ip(f) = / . . . / f ( ~ , u z , . . . , . " )dB(~ , )dB(~z )dB(un ) .
Put Jp(f; E ) = iPC([)I;(f : E ) with
I,*(f; E ) = 1. . . f (h , t z , . . . , t p ) E ( t i ) E ( t z ) . . . E(tp)dt idtz . . . d t p E
Set 31, = Ip(EZ(Rp)) c 31 and Fp = Jp(E2(Rp)) c F.
Theorem D. Ez(Rp) is isomorphic to 31, and F b the isomorphisms Ip(.) and Jp(.),
respectively. T h e orthogonal decompositions (Lz ) = c @ 31, and F = c @ Fp hold. p , Y m
p=o p=a It is important that both direct sums are realized by sums of functions:
m m
p=o p=o 9 b ) = c P P ( 4 and (79)(0 = c JP(fP)(E)?
and the Taylor expansion of ('Ty)(E) exp[fllElli] corresponds to its orthogonal decompo- sition.
Why we need S (S') and how differentiability of basic functions in S can be use? Is the necessary property only the nuclearity of linear spaces? Answer was given his new
approach.
5. Infinite Dimensional Rotation Group
Hida-Kubo-Nomoto-Yosizawa (1969 [23]) wished to realize LBvy's projective invari-
ance of Brownian motion as point wise transformations on a basic measure space. For the realization, they introduced a testing function space Do of special decay order at 03,
which is nuclear with a-Hilbertian norms :
'The author introduces J , ( f ; E ) , which was not used in Hida's paper, only for convenience of expla- nation.
50
For E = DO, we have the co-rotation group O(D6).
Each g = (: i ) E SL(2, R) acts as a linear homeomorphism gcp defined by
which preserves Lz-norm [ I .110. Therefore, its dual homeomorphism g* belongs to O(Db). A natural Brownian motion defined on the probability space (Db,p) satisfies point
wise projective invariance. Further, each one-parameter subgroup of SL(2, R) gives a
whisker in co-rotation group O(Db). Flows of Brownian notion and Ornstein-Uhlenbeck process are such examples.
Hida (1970 [24], 1972 [25], 1973 [26]) investigated infinite dimensional Laplacian and harmonic analysis relating to infinite dimensional unitary operators arising from O(E').
$3. White Noise Analysis Hida (1975 [27]) gave a series of lectures in Carleton University and published a lecture
note which is the first stage of white noise analysis. The author was informed the theory by Hida himself in Erlangen (1976) and understood its importance immediately. The new
idea is that a generalized process X ( t , w ) is considered as a generalized functional in w for fixed t , not a distribution in t for fixed w . As well known, the idea gives us much freedom to analyze Brownian functionals and can be applied to many fields in stochastic analysis.
Let p be the standard Gaussian measure on S'. Let
7 and Jp be the transformations stated in subsection 4 of 52. Observe the inclusions
G(n+1)/2(Rn) C Z2(R'') C g-("+')l2(Rn) of symmetric Sobolev spaces. The inclusions can be transformed6 to 3p) c Fn C FA-") by J,. Put 'lip) = 7-'F?), 31, = T-'Fn and Xi-") = 'T-'Fi-n), then we have inclusions:
Basic concepts are as follows.
CQ m m
( L z ) + = c $ 3 - 1 ~ ) c ( L z ) = c $ 3 1 n c ( L z ) - = c $ 3 1 ~ - " ) . n=O n=O n=O
Hida derivative is defined7 by aB(t)
Brownian motion B(t ) is considered as B(t ) = Z,(X[O,~~) E (L') for t 2 0. The derivative
1
h-tO h B(t) = lim -(B(t + h) - B(t ) ) = ZP(&) in (L2)-
has meaning in the sense of generalized Brownian functional; that is, an element of ( L z ) - . This is an answer to the question asked for Hida-Ikeda's paper in subsection 4 of 52. The
6The transformations Jp and 7-' must be extended properly. 7The original definition of Hida derivative was given for each term of Wiener-It6 decomposition
51
differentiability of test functions guarantees the differentiability of Brownian motion in t for each fixed w . In the frame work, one can treat L6vy’s Laplacian, Laplace-Bertrami operator, Feynman path integral etc. well.
Later, Kubo-Takenaka modified (1980 [28]) 7-transform to S-transform slightly:
where (E ’ ) is defined below. By S-transform, Hida derivative is described as
The relation of two definition is clear by observing the analytic continuation from the imaginary axis to the real axis. They constructed functional spaces by using reproducing kernels fully just as Hida’s idea. Let KO be the reproducing kernel Hilbert space with
kernel exp[((,v)]. Then A nuclear triplet (K) c KO c (K’) can be introduced by using
the Hilbertian norms of the basic space E (see [28]). Since S(Lz) = KO, one can define triplet ( E ) c (L2) c (E’) as the inverse image of the triplet by S-transform.
$4. Comments The author had spent with Hida in Nagoya University from 1967 to 1983 and watched
the creation of white noise analysis, which might be called Hida Calculus. The following
comments are based on questions in his mind in those days.
1. O(c0) In Hida and Nomoto’s projective limit of spheres, SO(n) acts on S, naturally. We call
Oj(S,) = U,SO(n) the finite dimensional rotation group of S,. The closure of Oj(c0) is not so clear. Of course, Oj(c0) is a small group of really finite rotation. However its closure under a suitable topology is big enough.
For example, the shift transformation of iid Gaussian random variables is really infinite
dimensional. It can be approximated by elements (g2,) of u,SO(n), where 92, rotates as 92, : (z1,z2,. . . , ~ 2 ~ ) + ( Q , z ~ , xs,x2, 27, 2 4 , . . . , X Z , , ~ ~ - 1 ) . This idea may be extended to any g E O(c0).
2n + 2n - 2 -+ ... -+ 4 -+ 2 + 1 -+ 3 -+ 5 + ... -+ 2n - 1 4 (2n)
t -1
2. Sobolev Norms
fn(u) on R” is defined by
Hida introduced Sobolev norms to control (Lz)+ C (L2) C (L2) - . The norm l l f n l l n for
52
where fn(v) is the Fourier transform of fn(u). By @, we denote the space of functions with finite norm. The function space is convenient to treat trace. Set
m m m
(L')+ = C ex?), II'PII' = C c ~ I I J o ~ J I E for 'P = C p n ? ' ~ n E .tit' n=O n=O n=O
by a suitable increasing sequence {c,,}.
This norm adjusts well to the trace properties of kernel functions fn E En. But it
has some difficulties for calculus. For example, it is hard to see that exponential function belongs to (L')+ or not and that for y, I,/J E (L2)+ does yI,/J belong to (L')+ or not. Its modification works sufficiently. We need to research more Sobolev type norms.
3. Generalized Random Variables
random variable. We say that CP E (E') is ordinal, if CP E (L ' ) = L'(&', p). CP E (E ' ) may be called a generalized random variable. If CP E (L'), then @ is a usual
(i) Is there any useful criterion for that CP is ordinal?
(ii) Suppose that {an} and @ are ordinal and and an converges to @, Is there any criterion for that CP, converges to @ in distribution sense?
(iii) Suppose that {@,} and (qn} are ordinal, and that @,, + @ and an + Q as n + 03.
Is there any criterion for that the product converges in (€').
(iv) Can we give any natural definition of the product @* using (iii) for suitable class of generalized random variables?
4. Renormalization Hida introduced a scheme of renormalization as follows. Let us observe Brownian
motion on the interval [0, TI. For the classical path y(t) = y(t ,z, a) from (0, T) to (T, a ) , define a perturbed path
by Brownian bridge. He expected that the propagator with potential V is given by
The renormalization process for : : is given by &Yo ' i
exP [ !!!! (a,) Ak - 7~ V(Ya(tk--l))& t k < t
2fi t k < t
lim I A l W A k Y a
53
for partition A : t o = 0 < tl < t2 < . . . < t , = T , (A( = max{tk - tk - l : 1 5 k 5 n}.
The author prefers this process and hopes for some one to research in this direction.
One of the purpose of Kubo-Takenaka ([as]) was to treat it in their framework. This approach is deeply related to the problem in previous subsection.
5. Path Hida and Si Si are studying variational calculus of random fields. Its origin is found
in investigation of LBvy’s Brownian motion B(x) with multidimensional parameter. Let
us suppose that the parameter space is R3. Let C(t) be a sphere with radius t and center xg. Further the origin 0 is outside of C(t). A method of interpolation B(x) of B(x) by
the information of B(.) in germ of C(t) . We can give 6(x) by
d an’
with smooth kernels KO, K1 and the normal derivative - symbolically. But B(z) is not
differentiable in path level as seen in Theorem A. However the following expression holds rigorously in the usual probability theory:
2
t B(x) = --
Integral of B(y) on the sphere C(t) with smooth kernel becomes smooth in t. The dif- ferentiability of such kind was observed in LCvy’s M(t)-processes as discussed also in
[16l. There are several questions:
(i) Do paths of the Brownian motion have such properties
generally? Is it possible to show that by some precise
discussion on path similar to Hida [15] ?
(ii) What is the characterization of a function space of such functions?
(iii) In the framework of white noise analysis, can we show that integrals of aB(y ) (in Hida’s sense) on smooth surfaces are ordinal random variable?
(iv) More generally, do we have any good criterion for that an element E (E’) is an ordinal random variable?
Acknowledgement The author thanks to the organizer of the international conference for giving him
the chance of this article, particularly to Professor Hida, who had guided him to this interesting field in the dawn of white noise analysis. The author also thanks to Masuyuki Hitsuda, who taught him an essay of Torahiko Terada in which Lucretius’ molecule is
described.
54
References
[l] Titus Lucretius Carus; De Rerum Natura, (BC 50). (Nature of things, translated by
William Ellery Leonard', Dover Publications, Incorporated (2004))
[2] Robert Brown; A brief account of microscopical observations made in the months of June, July, and August, 1827, on the particles contained in the pollen of plants; and on the general existence of active molecules in organic and Inorganic bodies, The Philosophical Magazine and Annals of Philosophy, Series 2, 4. No.21 (1828),
[3] Louis Bachelier; Thhorie de la spkulation. Annales Scientifiques de 1'Ecole Normale
161-173.
SupBrieure, 111-17 (1900), 21-86. Thesis for the Doctorate
[4] Albert Einstein;Uber die von der molekularkinetischen Theorie der Warme gefordete Bewegung von in ruhenden Flussigkeiten suspendierten Teilchen. Annalen der Physik, 17 (1905), 549-560.
[5] Marian Smoluchowski; Bulleitn de L'Acad. des Sc. de Cracovie. Juillet, Libraire FBlix
Alcan (1906)
[6] Paul Langevin; Sur la theorie du mouvement brownien, C. R. Acad. Sci. (Paris) 146
(1908), 530-533
[7] Jean Perrin; Movement Brownien et Rkartik Molkculaire, Annales de Chimie et Py-
isique, 8"" sBries, Septembre (1909)
[8] Jean Perrin; Les Atomes. Felix Alcan, Paris, Nouvelle collection scientifique, directeur
Emile Bore1 (1912)
[9] Norbert Wiener; Differential-Space, J . Math. and Phys., 2 (1923), 131-174
[lo] Gorge Eugene Uhlenbeck and Leonard Salomon Ornstein; On the theory of Brownian motion, Physical Review 36 (1930), 823-841.
1111 Norbert Wiener; The homogeneous chaos, Amer. Math. J . , 60 (1938), 897-936
[12] Paul LBvy; Le mouvement brownien plan, Amer. J . Math., 62 (1940), 487-550.
[13] Kiyosi It6; Stochastic Integral, Proc. Imp. Acad. Tokyo 20 (1944), 519-524.
[14] Kiyosi 1 6 ; Multiple Wiener Interal, Journ. Math. SOC. Japan 3 (1951), 1-51.
[15] Takeyuki Hida ; On the uniform continuity of Wiener process with amultidimensional
parameter, Nagoya Math. J . 13 (1958) 53-61.
[16] Takeyuki Hida; Canonical representations of Gaussian processes and their applica- tions, Memoirs Coll. Sci., Univ. Kyoto A33 (1960) 109-155.
'This translation is available from the web site - htpp://classics.mit.edu/Carus/nature-things.htm/
55
[17] Izrail’ Moiseevich Gel’fand and Naum Jakovlevich Vilenkin ; Generalized function 4, Some applications of harmonic analysis. Equipped Hilbert spaces, (1961)
[18] Takeyuki Hida and Hisao Nomoto; Gaussian measures on the projective limit space
of spheres, Proc. Japan Acad. 40 (1964) 301-304.
[19] Hisao Nomoto ; Finite dimensional approximations to some flows on the projective limit space of spheres. I, II., Nagoya Math. J . 28 (1966),167-177. ibid. 29 (1967), 127-135.
[20] Takeyuki Hida and Hisao Nomoto ; Finite dimensional approximation to band limited white noise, Nagoya Math. J . 29 (1967), 211-216.
[21] Ryogo Kubo; The fluctuation-dissipation theorem, Rep. Prog. Physi. 29 (1966), 235
[22] Takeyuki Hida and Nobuyuki Ikeda; Analysis on Hilbert space with reproducing kernel arising from multiple Wiener integral, Proc. 5th Berkeley Symp. on Math. Stat. Probab. 2 (1967), 117-143.
[23] Takeyuki Hida, Izumi Kubo, Hisao Nomoto and Hisaaki Yosizawa; On projective
invariance of Brownian motion, Publ. RIMS Kyoto Univ. A4 (1969), 595-609.
[24] Takeyuki Hida; Note on the infinite dimensional Laplacian operator, Nagoya Math.
J. 38 (1970), 13-19.
[25] Takeyuki Hida; A probabilistic approach to infinite dimensional unitary group, Proc.
Japan-USSR Probab. Symp. (1972) 66-77.
[26] Takeyuki Hida; A role of Fourier transform in the theory of infinite dimensional
unitary group, J. Math. Kyoto Univ. 13 (1973) 203-212.
[27] Takeyuki Hida; Analysis of Brownian Functionals, Carleton University, Carleton Math. Lecture Notes 13 (1975).
[28] Izumi Kubo and Shigeo Takenaka; Calculus on Gaussian white noise. I, 11, I11 & IV,
Proc. Japan Acad., 56A (1980), 376-380, ibid. 56A (1980), 411-416, ibid. 57A (1981), 433-437, ibid. 58A (1982), 186-189.
[29] Kohji Obayashi, Tomoharu Kohono and Hiroyasu Uchiyama; Photon correlation spectroscopy of the non-Markovian Brownian motion of spherical particles, Physi- cal Review A, 27, No.5 (1983), 2632-2641.
[30] Takeyuki Hida; Selected papers of Takeyuki Hida, Eds. L. Accardy, H.-H. Kuo, N. Obata, K. Saitb. Si Si and L. Streit (2001).
56
WHITE NOISE STOCHASTIC INTEGRATION
HUI-HSIUNG KUO
Department of Mathematics Louisiana State University
Baton Rouge, LA 70803, USA E-mail: kuoOmath. lsu. edu
White noise is often regarded as the informal nonexistent derivative B(t ) of a
Brownian motion B(t ) . Before K. It6 introduced the stochastic integral in 1944, white noise had been used as a random noise which is independent at different times and has large fluctuation. It was an innovative idea of It6 to consider the product
of white noise B(t) and the time differential dt as a Brownian motion differential dB(t), a quantity to serve as an integrator in the It6 theory. In 1975 T. Hida
introduced white noise theory which provides a rigorous mathematical definition
of B(t) as a generalized function defined on the space of tempered distributions on the real line. The white noise B(t) can further be regarded as a multiplication operator and B(t) = 6’t + 6’; with 6’t being the white noise differentiation and
6’; its adjoint. In this paper we will give a brief survey of Hida’s theory of white noise and its applications to stochastic integration. We will use the operator 6’; to study stochastic integrals with anticipative integrands and stochastic differential
equations with anticipative initial conditions and integrands. We will also point
out and describe several perspectives for further applications of white noise theory
to stochastic integration.
1. Brownian motion and white noise
Let B(t) be a Brownian motion. Its informal derivative B( t ) , which does
not exist in the ordinary sense, is called a white noise. We give three
simple examples of differential equations to motivate the need to establish
the theory of white noise.
Example 1.1. Consider the differential equation with white noise B(t)
- a x , + B(t ) , xo = zo, dXt dt --
where a E R. By regarding this equation as an “ordinary” differential
equation, we can derive the informal solution t
Xt = zOeat + ea( t -s) )B(s) ds.
57
But what is the integral s,' e"(t-s))B(s) ds? Observe that by the mean value
theorem n n
i=l i=l
which converges in L2(S2) to a Gaussian random variable with mean 0 and
variance
t Thus we can interpret the informal integral so e"(t-s)B(s) ds as a Wiener
integral so e"(t-s) dB(s) . Then the solution X , is given by t
t
X t = zOeat + Jd ea(t-s) dB(s) .
Example 1.2. Consider a differential equation with noise B(t )B( t )
dXt d t - = ax, + B(t)B(t), xo = 20.
Again, by regarding this equation as an "ordinary" differential equation,
we can derive the informal solution
Xt = zOeQt + ea(t-S)B(s)B(s) ds. (1) l
jDt
But what is the integral s," e"(t-s)B(s)B(s) ds? There are two ways to
interpret this integral. The first way is to combine B ( s ) and ds together
and regard B(s ) ds = dB(s) as an integrator for It6 integral17. Then the
solution is given by
X t = zOeat + e"(t-S)B(s) dB(s) .
The second way to interpret the integral s," e"(t-S)B(s)B(s) ds is to regard
B ( s ) as an operator 8: in the Hida theory of white noise12y13>25132. Then
the solution is given by
t
Xt = xgemt + 8; (e"(t-s)B(s)) ds. (2)
We will explain the operator 8; in Section 2 and define the white noise
stochastic integral s,' 8 , " ( f ( s ) ) ds in Section 4.
58
Example 1.3. Consider a linear stochastic differential equation
dXt = a( t )X t d t + P(t)Xt dB(t), X O = ZO, (3)
where a(t) and P ( t ) are deterministic functions. There are two methods to
solve this equation.
Method 1 (It6 theory of stochastic integration). We need to guess that
the integrating factor
Then use It6’s formula to derive the solution
Method 2 (Hida theory of white noise). Rewrite Equation (3) as
dXt = a( t )X t d t + 8; (,B(t)Xt) d t , Xo = zo . (5)
Then take the S-transform Ft(() = (SXt)(<), which we will explain in
Section 2, to get the following ordinary differential equation for each test
function <:
d d t -Ft(E) = a(t)Ft(E) + E(t)P(t)Ft(E), Fo(E) = Zo.
The solution is given by (there is no need to guess as in Method l!)
Then use the following fact from white noise theory
to get the same solution of Equation (3) as given in Equation (4).
In the above examples the equations and solutions have different degree
of involvement with the white noise B(t) . In Example 1.1 the equation is
solved as an “ordinary” differential equation and the solution requires the
concept of Wiener integral. In Example 1.2 the equation is again solved as
an “ordinary” differential equation, but the solution is interpreted as either
an It6 integral or a white noise integral. In Example 1.3 the equation
is interpreted as either a stochastic differential equation of It6 type or a
59
white noise stochastic differential equation, while the solution is derived by
different methods.
White noise had been used informally before K. It617 introduced the
theory of stochastic integration in 1944, in which B(t) and dt are combined
together to form dB(t) as in integrator. In 1975 T. Hida12 introduced the
theory of white noise so that B(t) itself is a rigorous mathematical object.
The progress can be illustrated by the following diagram:
K. It6 T. Hida AMS “B(t)” - dB(t) - B(t) ------+ 60H40 1944 1975 2000
One of the many motivations for T. Hida to introduce the theory of
white noise is to extend the It6 theory of stochastic integration. In this
paper we will briefly review the development in this direction and point
out several perspectives for further investigation.
2. White noise theory
Let S(R) denote the Schwartz space of rapidly decreasing functions on R.
The dual space of S(R) is denoted by S’(R). It is well-known that S(R) is
a nuclear space and we have a Gei’fand triple S(R) c L2(R) C S’(R). Let
p be the Gaussian measure on S’(R) with the characteristic function
where I . 1 0 is the L2(R)-norm. Then (S’(R), p ) is a probability space. The
stochastic process B(t, z) = (z, l p t l ) , t 2 0, z E S’(R), is a Brownian
motion. Then we have B(t) = z(t). Thus the elements of S’(R) can be
regarded as the “sample functions” of white noise.
For convenience, let ( L 2 ) denote the complex Hilbert space L2(S’(R), p) . By the Wiener-It6 decomposition theorem, each ‘p in (L2) can be uniquely
decomposed into the sum
n=O
where In is the multiple Wiener integral
above Brownian motion B(t) and 2p(Wn) symmetric L2-functions on R”. Moreover,
1 0 0
E 21(R”),
of order n with respect to the
is the complex Hilbert space of
the (L2)-norm of is given by
60
For each p 2 0 and cp given by Equation (6), define
\ 112
where A is the operator A = -d2/dx2 + x2 + 1. By using these norms we
can introduce the following spaces
(S,) = {‘p E ( L 2 ) ; IIcpllP < 00)
(S) = projective limit space of { (S,) 1 p 2 O}
( the space of test funct ions)
(S,)* = dual space of (S,)
(S)* = inductive limit space of { (S,)* 1 p 2 O}
( the space of generalized funct ions)
It is a factz5 that (S) is a nuclear space and so we have a Gel’fand triple
(S) C (L2) c (S)*. If < E &(a), the complexification of S(W), then the function
belongs to the space (S). The S-transform of a generalized function @ in
(S)* is defined to be the function
(s@)(<) = ((@,:e(’zt):)), < E sC(a), where ((., -)) is the bilinear pairing of (S)* and (S).
in terms of their S-transforms.
The next theorem characterizes generalized functions in the space (S)*
Theorem 2.1. (P~tthoff-Streit~~ 1991) A complex-valued function F on Sc(W) is the S-transform of a generalized function in (S)* if and only i f it satisfies the following conditions:
(1) For any <, q E Sc(W), the function F(z< + q) is an entire function
( 2 ) There exist constants K , a , p 2 0 such that of z E c.
I F ( J ) I I Kealtl:, V< E S~(IW>.
There are other spaces of generalized functions and the corresponding
characterization theorems, see the references in the bookz5 and the survey
p a p e r ~ ~ ~ J ~ .
61
Let y E S'(R). For 'p E (S) represented by Equation (3), we define
00
n=l
The operator Dy, differentiation in the direction y, is a continuous linear
operator from (S) into itself. The adjoint Dj is a continuous linear operator
from (S)* into itself. In particular, when y = &, we have the operators Da,
and Di,, which are often denoted by at and a,*, respectively. The operator
at is called the white noise differentiation operator. For y E S'(R), the multiplication operator
Qy = ( a , Y)V, 'P E (S).
is a continuous linear operator from (S) into (S)* and Qy = Dy + Dj. In
particular, for y = &, we have B(t) = at + 8: as continuous operators from
( S ) into (S)* . Moreover, the following commutation relationships hold:
[as, at] = 0, [a,., a;] = 0, [as, a;] = &(t)'.
S ( q y ) ( E ) = ( Y , W @ ( O , E E Sc(R).
s(a;@)(E) = E(t)S@(E), E E SC(R),
The S-transform of DE@ is given by
In particular, when y = &, we have
which is very useful for white noise stochastic integrals. On the other hand,
the S-transform of at'p for 'p E (S) is given by
d s(at'p)(E) = &WE + ZSt)l,,O' E E Sc(R).
The S-transform is not so useful to handle the white noise differentiation
at'p of a test function 'p in (S).
3. White noise stochastic integrals
A function @ : [a,b] .--) (S)* is called Pettis integrable if for any 'p E (S),
the function ((a(.), 'p)) is measurable and belongs to L1[a, b]. In that case,
there exists a unique Q E (S)* such that
b
( P , c p ) ) = J ((@(tL'p)) dt , 'p E (S).
The unique Q, denoted by J: @(t) dt , is called the Pettis integral of @(t).
Theorem 3.1. Suppose @ : [a, b] -+ (S)* satisfies the conditions:
62
(a) S@(.)(<) is measurable for any < E S,(R).
(b) There exist constants K , X,p 2 0 such that
I” Is@(t ) (<) ld t 5 KeAIcIi, v< E S,(R).
Then @(t) is Pettis integrable and
A function 4? : [a, b] 3 (S)* is said to be Bochner integrable if it satisfies
the conditions:
(a) For any cp E (S), the function ((@(.), cp)) is measurable.
(b) Thereexistsp 2 0 such that a(.) E (Sp)* a.e. and l ( @ ( . ) ( I p E L1[a, b].
In that case, there exists a unique Q E (S,)* such that
b
((Q, cp)) = 1 ((@(t), cp)) cp E (S) .
The unique Q, denoted by Jt @(t) dt , is called the Bochner integral of @(t).
Theorem 3.2. Suppose @ : [a, b] + (S)* satisfies the conditions:
(1) S+(.)(<) is measurable f o r any < E Sc(R).
( 2 ) There exist constants A, p 2 0 and a nonnegative integrable function L on [a , b] such that
IS@(.)(<>I 5 L(.)eXIElz, V < E S,(R).
Then @(t) is Bochner integrable and
for any q > p such that 2Xe2((A-(4-p)((& < 1.
If @ : [a , b] --+ (S)* is Pettis integrable, then the function t ++ a:@(t) is
also Pettis integrable and
Moreover, If @ : [a,b] 4 (S)* is Bochner integrable, then the function
t H a:@(t) is also Bochner integrable and Equation (7) holds.
63
4. Hitsuda-Skorokhod integrals
In Example 1.2 we interpreted the multiplication by B(t) in Equation (1) as the operator a,* in Equation (2). In Example 1.3 we interpreted the
integrator d B ( t ) in Equation (6) as 8; dt in Equation (5). The reason for this interpretation is given by the following theorem.
Theorem 4.1. (Kubo-TakenakaZ3 1981) Let p(t) be nonanticipating and J” Il‘p(t)IIz dt < 00. Then the function t H d;‘p(t) is Pettis integrable and
b 6” a,* (‘pW dt = 1 d t ) d B ( t ) ,
where the right-hand side is the A S integral of ‘p(t).
Observe that in the white noise integral J: 8; (cp(t)) dt there is no need
to require that the stochastic process ‘p(t> to be nonanticipating. Therefore,
it provides an extension of the It6 integral. However, the extension would
be more meaning if the integral Jf a,* ( ‘ p ( t ) ) dt defines a random variable
instead of just a generalized function in (S)*.
Thus we make the next definition. Recall a fact from the book^'^^^' that ul<plm W S ’ ( W , p ) c (S)*.
Definition 4.1. By a Hitsuda-Skorokhod integral we mean a white noise
integral s,” a,* (‘p(t)) dt when it defines a random variable in LP(S’(R), p ) for some 1 < p 5 00.
The integral J: 8; ( ‘p ( t ) ) dt was introduced by Hitsuda14 in 1972 and by
S k ~ r o k h o d ~ ~ in 1975 by different methods and notations.
Example 4.1. Jt d,*B(l) dt = B( l )2- l . On the other hand, the extension
by It618 gives a different value Jt B(1) d B ( t ) = B(1)’.
Theorem 4.2. Suppose ‘p : [u,b] - (S)* satisfies the condition that Jf II(N + 1)1/2‘p(t) l l idt < 00, ( N : number operator). Then S,ba,.io(t)dt is a Hitsuda-Skorokhod integral and
Il‘p(t>ll2, dt + J” J” ((at(P(4, a8‘p(t>>>, dsdt, 11 l 8t*‘p(t> dt$ = l a a
where ((., .))o is the inner product on (L’). Moreover,
64
5. Extensions of It6's formula
Let 8 be a C2-function. ItS's formula states that t
8(B(t)) = O(B(a)) +I e'(B(s)) d B ( s ) + a
The white noise formulation of this equality is given by
t
q q t ) ) = e(B(a)) +I a:el(B(s)) ds + e l p ( s ) ) ds. a l l
In this formulation, It6's formula can be generalized to Hitsuda-Skorokhod
integrals and generalized functions.
Theorem 5.1. ( H i t s ~ d a ' ~ ~ ~ ~ 1972) Suppose 9(z, y ) is a C2-function and assume that the functions
all belong to L2( [a, b]; ( L 2 ) ) . Then f o r any t E [a, b], the white noise integral 1," d,* (g (B(s ) , B ( b ) ) ) ds is a Hitsuda-Skorokhod integral and
Theorem 5.2. (Kubolg 1983) Let 8 E S'(R). Then fo r any 0 < a 5 b, the following equality holds in the space (S)*
where 9' and 8" are derivatives in the distribution sense.
It6's formula can also be generalized to 9(Xt ) with X , being a stochastic
process given by a white noise forward integra125>30. But the conditions are
rather complicated and technical. For the detail, see the papers by Asch
and P o t t h o P ~ ~ ? ~ and by P ~ t t h o f f ~ ~ .
6. Stochastic integral equations
Let X a be measurable with respect to u{B(t) It 5 a } and f and g satisfy
the Lipschitz and linear growth conditions. Then the stochastic integral
65
equation
x(t) = x a + I’ f(s, ~ ( s ) ) d ~ ( s ) + s” g(s, ~ ( s ) ) ds,
x(t) =xa+ l a : f ( s , ~ ( s ) ) d s + g (s ,X (s ) )ds , (9)
(8)
has a unique continuous solution.
Suppose the initial condition X a is not measurable with respect to
a{B( t ) 1 t 5 u} . Then we cannot use the iteration method to obtain the
solution within the It8 theory. However, by using white noise theory we
can rewrite the Equation (8) as t
where the first integral is a Hitsuda-Skorokhod integral. In many cases we
can use the S-transform to solve this equation as follows:
(1) Take the S-transform of Equation (9) to get an ordinary differential
(2) Solve the new equation for &(<) for each < E SC(R). (3) Take the inverse S-transform of &(<) to get the solution X ( t ) .
We give several examples to illustrate this method and to show new and
equation for Ft(J) = S X ( t ) ( J ) .
interesting phenomena beyond the It6 theory of stochastic integration.
Example 6.1. Buckdahn‘ considered the Skorokhod form of the following
stochastic integral equation
~ ( t ) = sgn(B(1)) + 1 a,*x(s) ds, o 5 t 5 1.
He used a rather complicated method to find the solution. However, it is
very easy to use the S-transform method to derive the solution
~ ( t ) = sgn(B(1) - t ) eB(t)-it.
Observe that the discontinuity of the sample paths of X ( t ) is due to the
anticipative initial condition.
Example 6.2. Consider the stochastic integral equation
t
where t E [0, I]. We can use the S-transform method to derive the solution
~ ( t ) = eB(t)-ft( l + tsgn(B(1) - t ) ) .
Observe that the discontinuity of the sample paths of X ( t ) is due to the
anticipative integrand of the second integral.
66
Example 6.3. There are simple equations such as
X ( t ) = 1 + a,* ( B ( l ) X ( s ) ) ds, (10) I’ which cannot be directly solved by the S-transform method. This equation
written in the Skorokhod form was considered by Buckdahn5. We can use
the next theorem to derive the solution of Equation (10)
Theorem 6.1. Suppose f is a deterministic function in Loo[a, b]. Then the stochastic integral equation
t
X ( t ) = 20 + 1 a,* ( f ( s )B (b )X(s ) ) ds, a I t I b, a
has a unique solution in L2([a, b] ; (L2)) and is given by
x(t) = 20 exp [B (b ) Ju f ( s ) e-J.t f ( ~ ) d ~ d ~ ( s ) t
Example 6.4. Consider the stochastic differential equation25 ,29
X ( t ) = 1 + a,*(B(l) o X ( S ) ) ds, 0 5 t 5 1, Jo” where o denotes the Wick product. The solution is given by
X ( t ) = d- 1 exp [-2(1+ 1 t + t 2 ) (tB(l)Z-2(l+t)B(l)B(t)+B(t .
Example 6.5. Compare the stochastic integral equations
X ( t ) = B(t) + I’ B ( l ) X ( s ) ds,
Y( t ) = B(t ) + I t B(1) o Y ( s ) ds.
Their solutions are given by
X ( t ) = B(t) + ~ ( 1 ) e( t -s)B( l )B(s) cis, t l
y(t) = I d,*e(t-S)B(1)-i(t-S)2 ds.
67
7. Perspectives of white noise stochastic integration
We briefly describe several directions and topics for further research on
white noise stochastic integration.
(1) Hitsuda-Skorokhod integral
We have the fact that ul<p~mLP(S' (R) ,p) c (S)*. But how can
we tell that a generalized function is a random variable belonging
to LP(S'(R),p) for some p > l? It is very desirable to obtain a
characterization theorem to answer this question. Such a theorem
will be very useful to study the Hitsuda-Skorokhod integral and
white noise stochastic differential equations.
(2) Solutions of white noise stochastic integral equations
Prove a regularity theorem to obtain a strong solution from a weak
solution of a white noise stochastic integral equation X ( t ) = X, + s," f(s, X ( s ) ) ds, in particular, a stochastic integral equation of the
Hitsuda-Skorokhod type X ( t ) = X , + J," d,*f(s, X ( s ) ) ds.
(3) Extensions of ItB's formula
Itb's formula for O(B(t),B(b)), a 5 t 5 b, has been extended to
O(X( t ) , B(b)) for a Wiener integral X ( t ) = J," f(s) dB(s) in the
How about when X ( t ) is a Hitsuda-Skorokhod integral?
(4) Clark-Ocone formula
The Clark-Ocone formula has been formulated and generalized in
terms of white noise theory by de Faria-Oliveira-Streit", namely,
'p = E'p + SRE(at IFt)dB(t) for any 'p E W1/2. This formula
has been further generalized by Ngobi-Stan31 by using the second
quantization operator I?( l(-m,tj). Aase-Bksendal-Privault-Ub~e
have also generalized this formula. It would be useful to extend
their results to the case when 'p is a generalized function in the
space (S)" , in particular, when 'p is a random variable in the union
Ul<p+LP(S'(R),p) c (S)*.
(5) Girsanov theorem
Recall the probability space (S'(R), p ) and the Brownian motion
B(t) defined in Section 2. Let h(t) , 0 5 t 5 T, be a nonanticipating
stochastic process such that E, exp (i 5: h(t)2 dt ) < m. Define a
68
book.
probability measure Q on S’(R) by
T
ds] dP.
The Girsanov theorem states that the It6 process X ( t ) = B(t ) + h(s) ds is a Brownian motion with respect to Q. It would be
very interesting to generalize this theorem to the case when h(t) is not assumed to be nonanticipating. In this case, the It8 integral
h(s) dB(s) needs to be replaced by a Hitsuda-Skorokhod integral.
Such a formula will be useful to study the Black-Scholes theoory for
anticipative markets.
Acknowledgments
The author would like to thank the Academic Frontier in Science of Meijo
University for financial supports and to give my deepest appreciation to
Professors Y. Hara-Mimachi, T. Hida, K. Nishi, and K. Sait6 for the warm
hospitality during my many visits to Meijo University.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
Awe, K., Bksendal, B., Privault, N., and Ubae, J.: A white noise general-
ization of the Clark-Haussmann-Ocone theorem, with application to mathe- matical finance; Finance and Stochastics 4 (2000) 465-496 Asch, J. and Potthoff, J.: A generalization of It& lemma; Proc. Japan Acad. 63 (1987) 289-291 Asch, J. and Potthoff, J.: White noise and stochastic analysis; in: Stochastics, Algebra and Analysis in Classical and Quantum Dynamics, S . Albeverio et al. (eds.), Kluwer Academic Publishers (1990) Asch, J. and Potthoff, J.: It6’s lemma without non-anticipatory conditions; Probab. Th. Rel. Fields 88 (1991) 17-46 Buckdahn, R.: A linear stochastic differential equation with Skorohod inte-
gral; Markov processes and control theory, Math. Res. 54 (1988) 9-15 Buckdahn, R.: Anticipating linear stochastic differential equations; Lecture Notes in Control and Information Sciences 136 (1989) 18-23, Springer- Verlag Buckdahn, R.: Anticipative Girsanov transformations; Probab. Theory Re- lated Fields 89 (1991) 211-238 Buckdahn, R.: Skorohod stochastic differential equations of diffusion type; Probab. Theory Related Fields 93 (1992) 297-323 Buckdahn, R.: Anticipative Girsanov transformations and Skorohod stochas- tic differential equations; Mem. Amer. Math. SOC. 111 (1994) no. 533 Buckdahn, R. and Nualart D.: Skorohod stochastic differential equations with boundary conditions; Stochastic Stochastic Rep. 45 (1993) 211-235
69
11. de Faria, M., Oliveira, M. J., and Streit, L.: A generalized Clark-Ocone formula; Random Oper. Stoch. Eqs. 8 (2000) 163-174
12. Hida, T.: Andpis of Brownian Functimals. Carleton Mathematical Lecture Notes 13, 1975
13. Hida, T., Kuo, H.-H., Potthoff, J., and Streit, L.: White Noise: A n Infinite Dimensional Calculus. Kluwer Academic Publishers, 1993
14. Hitsuda, M.: Formula for Brownian partial derivatives; Second Japan- USSR Symp. Probab. Th. 2 (1972) 111-114
15. Hitsuda, M.: Formula for Brownian partial derivatives; Publ. Fac. Integrated Arts and Sciences, Hiroshima University, Series III 4 (1978) 1-15
16. Holden, H., 0ksenda1, B., Ubpre, J., and Zhang, T.: Stochastic Partial Dij- ferential Equations. Birkhauser, 1996
17. Its, K.: Stochastic integral; Proc. Imp. Acad. Tokyo 20 (1944) 519-524 18. It6, K.: Extension of stochastic integrals; Proc. Intern. Symp. Stochastic
Differential Equations, K. It6 (ed.) (1978) 95-109, Kinokuniya 19. Kubo, I.: It6 formula for generalized Brownian functionals; Lecture Notes in
Control and Information Scis. 49 (1983) 156-166, Springer-Verlag 20. Kubo, I. and Kuo, H.-H.: Finite dimensional Hida distributions; J. Funct.
Anal. 128 (1995) 1-47 21. Kubo, I. and Takenaka, S.: Calculus on Gaussian white noise I; Proc. Japan
Acad. 56A (1980) 376-380 22. Kubo, I. and Takenaka, S.: Calculus on Gaussian white noise 11; Proc. Japan
Acad. 56A (1980) 411-416 23. Kubo, I. and Takenaka, S.: Calculus on Gaussian white noise 111; Proc. Japan
Acad. 57A (1981) 433-437 24. Kubo, I. and Takenaka, S.: Calculus on Gaussian white noise IV; Proc. Japan
Acad. 58A (1982) 186-189 25. Kuo, H.-H.: White Noise Distribution Theory. CRC Press, 1996 26. Kuo, H.-H.: Anticipatory It& formula and Hitsuda-Skorokhod integral; in
Skorokhod’s Ideas in Probability Theory, V. Korolyuk et al. (eds.), Proc. Institute of Math., National Academy of Sciences of Ukraine, Vol. 32 (2000)
27. Kuo, H.-H.: White noise theory; in Handbook of Stochastic Analysis and Applications, D. Kannan and V. Lakshmikantham (eds.) 107-158, Marcel
Dekker, 2002 28, KUO, H.-H.: A quarter century of white noise theory; in: Quantum Injorma-
tion IV, T. Hida and K. Sait6 (eds.) (2002) 1-37, World Scientific 29. Kuo, H.-H. and Potthoff, J.: Anticipating stochastic integrals and stochastic
differential equations; White Noise Analysis-Mathematics and Applications, T. Hida et al. (eds.) (1990) 256-273, World Scientific
30. Kuo, H.-H. and Russek, A.: White noise approach to stochastic integration;
J . Multivariate Analysis 24 (1988) 218-236 31. Ngobi, S. and Stan, A.: An extension of the Clark-Ocone formula; Interna-
tional J. Math and Math Sciences 1463-1476 (2004) 32. Obata, N.: White Noise Calculus and Fock Space. Lecture Notes in Math.
1577, Springer-Verlag, 1994
248-255
70
33. Potthoff, J.: Stochastic integration in Hida’s white noise analysis; in: Stochastic Processes, Physics and Geometry, S . Albeverio et al. (eds.), World Scientific (1990)
34. Potthoff, J. and Streit, L.: A characterization of Hida distributions; J. Funct.
35. SaitB, K.: Its’s formula and Lkvy’s Laplacian; Nagoya Math. J. 108 (1987)
36. SaitB, K.: ItB’s formula and Lkvy’s Laplacian 11; Nagoya Math. J. 123 (1991)
37. Skorokhod, A. V.: On a generalization of a stochastic integral; Theory
Anal. 101 (1991) 212-229
67-76
153-169
Probab. Appl. 20 (1975) 219-233
71
CONNES-HIDA CALCULUS A N D BISMUT-QUILLEN SUPERCONNECTIONS
R. LQandre Institut dc Mathbmatiques. Universitk de Bourgogne
21000. Dijon. FRANCE email:Flemi.leandreOu-bourgogne.fr
H. Ouerdiane Dkpartement de Mathbmatiques. Facultk des Sciences de Tunis
1060.Tunis. TUNISIA email:habib.ouerdiane@fst.rnu.tn
Abstract: We show that the J.L.O. cocycle for the Index theorem for a family of Dirac operators, by
Mathematics subject classification (2000): 58520. 58B34. 58J65. 60H40. Key words: J.L.O. cocycle. White noise analysis. Bismut-Quillen Superconnection.
using Bismut-Quillen superconnection formalism, is a white noise distribution.
I. Introduction
Let us remark that there are 3 infinite dimensional distribution theories: -1) White noise analysis ([B.K], [B], [HI, [H.K.P.S], [St]), which uses the Fock space and which is
-2)Watanabe’s distribution theory [Wall, which is a part of Malliavin Calculus ([Mall), and which is
-3)Non-commutative differential geometry of Connes ([Coz]) which is algebraic. The goal of this paper is to show that the first one is very similar to the third one, if we disregard the
operations (the algebra of annihilation and creation operators for white noise analysis and the cyclic complex for non-commutative differential geometry) for the particular case of the algebra of functions on a manifold.
The auxiliary bundle associated to a twisted Dirac operator plays an important role in the Index theorem. Let us recall that any complex bundle [ on a compact manifold endowed with a connection V is isomorphic to a subbundle of a trivial complex bundle Cd endowed with the projection connection. So [ is given by p(Cd) where p is a smooth complex projector. If we consider the Dirac operator D which acts on the sections of the spin bundle S of a spin manifold, the twisted Dirac operator is pDp which acts on the section of S@[. There are two ways to study the algebraic properties related to it:
-l)Bismut ([Bill) introduced its Chern Character over the free loop space, showed it is related to the equivariant complex on the free loop space. A beginning of rigorous interpretation in Watanabe’s sense of the works of Bismut is given in [J.L], [Ld], [Ls], [Le], [LT], [Lg] which give a probabilistic interpretation of the considerations of [G.J.P]. A beginning of interpretation in the white noise sense of the works of Bismut is given in [Ls] and [Llo] starting from the work of Getzler ([GI).
-2)The introduction of the cyclic complex and the Chern character in cyclic homology ((CO~], [G.S], (J.L.01, [J]) explains too the algebraic properties of the auxiliary bundle. These considerations are a part of non-commutative differential geometry.
In -1) and -2), the authors try to give a meaning to the Index theorem for a twisted Dirac operator as a current, which operates in -1) on forms on loop space and in -2) on the cyclic complex of the algebra of smooth complex functions on the manifold. In [Ls] and in [L~o], we have shown that the symmetric Fock space plays a big role in order to understand Witten’s current, by using Chen forms. There are others Fock spaces than bosonic Fock space and Fermionic Fock space. It is for instance the case of the interacting Fock space of Accardi-Boaejko ([A.B]]. We consider the interacting Fock space of Accardi-Bozejko in order
algebraic.
probabilistic.
72
to understand the natural criterium of convergence of series which appears in entire cyclic cohomology of Connes ([Col]). We recover the algebra of smooth functions on a compact manifold as the intersection of some Hilbert-Sobolev spaces on the manifold, by using the Sobolev imbedding theorem.
We are concerned in this paper by the algebraic properties associated to the Index theorem for families, by using Bismut-Quillen superconnection formalism. In the case of non commutative differential geometry, it is studied by [Nil, [N.T] and [PI, We are motivated to interpret the previous considerations in the framework of white noise analysis. In the case where we use Bismut-Chern character, it is the purpose of [Llo]. The main novelty of this work is the following:
Main theorem:the J.L.O. cocycle for a family of Dirac operators is a white noise distribution with values in the space of smooth forms on the basis manifold.
We use probabilistic technics in order to estimate densities. The reader interested by various probabilistic proofs of the Index theorem, including Bismut's proofs can see the surveys of LQandre [Lz], Kusuoka ([Ku]) and Watanabe ([Waz]) about that topic. The reader interested by probabilistic proofs of the Index theorem for families, besides the original paper of Bismut ([Biz]), can see the pedagogical paper of LBandre ([Ll]) with a simplified proof.
The second author thanks the warm hospitality of the Institut Elie Cartan where this work was done.
11. Cyclic homology and white noise analysis
Let us consider a compact Riemannian manifold M of dimension d endowed with the Riemannian measure d m ~ . Let A be the Laplace-Beltrami operator over it. A = dzvgrad where gradf(z) is the vector associated to the 1-form d f ( 2 ) via the Riemannian structure. A is self-adjoint, densely defined over the Hilbert space HO = L 2 ( M ) of complex functions on M . We consider the Sobolev space Hk (k E N ) of complex functions 4 on M such that /,(A + l ) k 4 & h ~ < M. The Laplace-Beltrami operator A has eigenvalues A, 2 0 (numbered in increasing order) associated to the eigenfunctions 4,. Only A0 = 0 and XO is associated to 1. Moreover, An - Cn' when n -+ co for some integer T ([Gi]). Let k E N,C > 0. We consider the space Wk,C constituted of elements F
where F,, belongs to Hk equipEed with its natural Hilbert structure. f f k = Hk/C, and we assimilate an element $ of Hk to its image in Hk (We work on the complex numbers). Moreover, Hk+l c Hk. We define over Wk,c the following Hilbert structure:
Wk,C is called an interacting Fock space ( [A.B], [A.N], [G.H.O.R], [Ou], [O]).
projective topology. (See [HI). Definition 11.1: The space of white noise test functionals is &CN,C>OWk,C = W endowed with the
W is a nuclear Frechet space, because the Sobolev imbedding theorem. We consider the Hochschild boundary (See [Lo] for instance)
n-1
(2.3) b(4o 8 41 8 .. 8 4,) = x( -1 ) ' 40 8 .. 8 da4i+i 8 .. 8 4n + (-1)"&4o 8 41.. 8 4"-i i =O
Theorem 11.2: The Hochschild boundary b is a continuous linear application from W into W . Proof: The proof is very similar to the proof of ThBoreme 11.5 of [Ld]. Let us write:
(2.4)
73
where I = (io, ._,in) and 111 = n. We have
(2.5)
We use (2 .3 ) and Sobolev imbedding theorem (See [Gi]) in order to find a kl > k and CI > C such that
We consider some big kz > kl and some small CZ such that
(2.7)
if kz is big enough. We deduce by Cauchy-Schwartz inequality that for some big C3 > 0
(2.9) IlbF/lk3C 5 AIIF11Zki,C3
Therefore the result.
0 We consider the Connes operator ([Coz], [Lo])
n
(2.10) B(4o @ 41 @ .. @ 4n) = c(-1)"1@ 4 i . . @ 4 n @ 40 @ 41.. @ 4 i - I
i=O
We have the following theorem, whose proof is omitted because it is similar to the proof of Theorem 11.2: Theorem II.S.:B is a linear continuous map from W into W . As classical (See [Lo], [COZ] for instance), we have the relations
(2.11) b2 = B2 = bBf Bb = 0
such that b + B is a complex.
Definition 11.4.: b + B operating on W is called the cyclic complex in white noise sense. Let us give some classical examples of elements belonging to the cyclic complex (See [G.S]).
We denote
(2.12) Hm- = nk>oHk.
By the Sobolev imbedding theorem, Hm- is nothing else that the algebra of smooth functions on the manifold.
Let p = ( p i , ) ) belonging to some Mn(Hm-) ( the space of (n x n) matrices with component in H m - ) such that p 2 = p. It defines a complex bundle over M . Reciprocally, each complex bundle ( endowed with
a connection V can be given by some projector belonging to some M,,(Hm-) endowed with the projection connection (See [N.R]).
We denote (See [GS] p 346):
(2.13) p n = Tr@@ .. @ P ) = C pio, i , @ P i l . i z . . @p i - - , , i , , @Pi", io
;a,. . .in
74
(2.15)
since p 2 = p and since the image of 1 in pk is equal to 0. We denote (See [GS] Proposition 1.1)
(2.16) k = l
It is clear that Ch,(p) projector belongs to the space W of white noise test functionals for every complex
projector p and that
(2.17) ( b + B)Ch*(p) = 0
(See [G.S] Proposition 1.1 for the proof of this identity which comes from (2.15)). In order to show that Ch,(p) belongs to W , we remark that:
and that
(2.19)
111. T h e J.L.O. cocycle as a white noise distribution
Let us consider a fibration ?r : M + B of compact Riemannian manifolds:a-'(O) - 0 x V for some
open neighborhood of each point y in B. The generic element of B is denoted by y and 2 denotes the generic element of the fiber V,. We denote by A(B) the exterior bundle on B: we consider ?r*(A(B)) the pullback bundle on B endowed with the trivial connection V. We suppose that the fiber V, is spin, and we consider
the spin bundle S, = S,' CB S; over V,. Let us recall what we mean by that. We consider the double cover
Spin(d) of SO(d), where the dimension d of V, is supposed even. We consider SO(V,) the frame bundle of V, supposed ocientable. We would like to get a lift of the principal SO(d) bundle SO(V,) by Spin(d), called Spin(V,). It is not always possible. We have to suppose that some topological constructions are satisfied. Moreover, we have the spin representation of Spin(d) on S+ CB S-. We consider the associated
bundle S,' @I S; on V, supposed spin. The construction of Spin(V,) is not canonical, and supposed some choices. We suppose that the bundles S, fit together in a complex bundle S on M . Let us recall that the Clifford algebra of the Euclidean space Rd endowed with its canonical Euclidean structure <, . > is the algebra constructed from Rd where we have the relation for e and e' in Rd:
(3.1) e.e' + e'.e = 2 < e, e' > .
75
and
Since the construction of the Clifford algebra on Rd does not involve any choice, there is no problem to construct the Clifford bundle CI, on V,. We assimilate the Clifford bundle with its complexification. The Clifford bundle CI, acts over S,, and the product by an element of the tangent bundle, considered as a subbundle of the Clifford bundle, is odd relatively to the natural graduation on S,. We consider the Levi- Civita connection V' on V,, which passes to S, because Spin(V,) is a lift of SO(V,) by Z/2Z and to CI,. We consider the family of Dirac operators D,. In local coordinate, D, = C e;,,V&, where ei,, is a local orthonormal basis of T(V,).
We consider the Levi-Civita connection VM on M . According Bismut ([Biz]), we introduce another connection V' on M . We have V',Y = IIVxY if Y is a vector field on V,, where II is the orthonormal projection from the tangent bundle of M on the tangent bundle on Vu. In particular, if X is a vector field on Vu, V>Y = VgY. If Y is the pullback of a vector field on B, V',Y = 0 if Y is a vector field on V,. Moreover Vk.xr*Y is the pullback of V$Y where VB is the Levi-Civita connection on B. In order to do these considerations, we have chosen the metric on M such that the orthogonal bundle of V, is isometric to the tangent space of B by the derivative of r. We get by using this orthonormal decomposition of the tangent bundle of M into the tangent bundle of V, and its orthogonal the notion of pullback of a vector field on B in a vector field on M .
We have, since connections differ by one form
(3.2) VM = V'+ s
where S is a one form with values in the tensorial operator on T ( M ) .
coordinate, it is equal to Let E, be the bundle on V, r *A(B)6Sy. Over Xu, we consider the connection V v = 0' + S. In local
where e, denotes a local orthonormal basis of T(V,) which acts by Clifford multiplication on S, and f, an orthonormal basis of T(B) which acts by exterior multiplication of A(B) (See [Ll] (3.41)).
Let us consider Hm = H+" fB H_" the infinite dimensional bundle on B of smooth sections of S,. Let us define a trivialization of Hm. Let 0 be a small open ball centered in yo in B. There is a unique geodesic joining yo to y in 0 called IB(YO,Y). Since r*T(B) is supposed orthogonal to T(V,) in T ( M ) , we can lift this curve in a curve lyo,,(z). $0 -+ I,,,,,(~o) realizes a diffeomorphism from V,, to V, if y is close enough from yo. The connection V' preserves the orthogonal decomposition of T ( M ) intp T(V,) and r *T(B) . Therefore, the parallel transport along lye,, realizes an isomorphism between H E and H F which preserves
the Z/ZZ graduation. We consider the bundle on B A(B)6Hm. We consider a Z/ZZ graded tensor product. r*Au(B)6CIu acts on r*A,(B)6H,m, but we have to take care that we consider 2 / 2 2 tensor products. We have namely:
(3.4)
if ey = eh..e;" where n = 111.
(u6e;)(u6$,) = ( - l ) l r l d e g o u A u16ey$'
If u is a smooth form on B and $Ju a section of S,, we write:
(3.5) D y ( u 6 $ ~ ) = (-l)deg"u6Du$JY
Let <B be a Z/ZZ-graded finite dimensional complex bundle on B. Let A ( T ( B ) ) ~ < B be the graded tensor product of the two bundle A(T(B)) and <B. A superconnection ([QI]) is locally an odd operator acting on the tensor product of the two previous bundles written as:
(3.6) d s + C d f , , A , , = Vs
where a, = (cY,~, ..,a,,) and dfo, == df,., A .. A df,., (T = larl). If IarI is even, df,, is an even operator such that A,, is an odd operator acting on the local sections on <B which do not involve derivatives of these
76
sections. If larl is odd, df,, is an odd operator (which acts by wedge multiplication according the sign rules (3.4)) such that A,, is an even operator acting on the local sections on E E , which acts without to take some derivatives of these sections. We put if A,, is even:
where A,, is given in diagonal form as (A&,A;,) . If A,, is odd trsA,, = 0. The curvature R, = 0: is a
tensorial operator, which is even, and which operates on A(T(B))6,EE (we have to take care with conventions of signs analog to (3.4) in order to do the product). The big difference with the traditional theory, where the curvature of a connection of a bundle is a 2-form, is that we can consider forms of any degrees in R..
Bismut ([Biz]) has extended this formalism to the case where we replace E B by Hm. If a1 = I, A8 = D, and if ( ( X I ( 2 1, A,, is an odd operator which is in some sense tensorial: it acts fibers by fibers on *'A(T(B))6Su without to take derivatives of the section of this bundle. In order to do the asymptotics Bismut used a special superconnection. We follow him. We consider the Bismut superconnection, called by Bismut the Levi-Civita superconnection:
(3.8)
(See [Biz], [Lil). We consider over the space of operators on A(TB)&IH" the natural 2 / 2 2 graduation, such that V r is
an odd operator. We can consider the curvature R? of 0:. It is an operator which is tensorial in y and a differential operator in z whose part of second order is (Du)2.
Let us consider a unitary complex bundle E, on V, endowed with some unitary connection Vc*. We can define the associated horizontal Laplacian acting on sections of this bundle. Let V' be the Levi-Civita connection on V, and ey be a local orthonormal basis of T(V,). Let & be a section of E Y . We have:
(3.9)
AV'" is a densely defined self-adjoint operator acting on the La sections of ,E,. We will consider in the sequel
the special case of the bundle r'A(T(BG))&IS, endowed with the connection 9,. Let K"z) be the scalar curvature on V, for the Riemannian metric on V,. We get the Lichnerowicz
formula (See [B,], [Ll] Theoreme 111.1):
(3.10)
where A: is the horizontal Laplacian -TT(@')~ which acts on the section of the bundle E, on V,.
equation in the Stratonovitch sense (See [El], [Em], [I.W])
RTy = &f + 1/4K'
Let zr(z) be the Brownian motion on V, starting from z. It is the solution of the stochastic differential
(3.11) dZ:(Z) = TF(Z)d& .
where Bt is a flat Brownian motion in Tz(V,), T?(Z) the stochastic parallel transport on V, along the Brownian path for the Levi-Civita connection on V,. We can use another way by introducing the bundle of direct orthonormal frames SO(%) and the canonical vector fields X:(z) on it. Let us recall what are these objects. The frame bundle SO(V,) is the space of oriented isornetries u u from the canonical Euclidean space Rd into T(V,(z)) . It inherits from the Riemannian structure the Levi-Civita connection. Let e, he the canonical oriented orthonormal basis from Rd. u Y ( e , ) constitutes a direct orthonormal basis of T(V,(z)) . We consider the horizontal lifts X:(u) of uY(e,). The set of X:(u) constitutes the set of canonical vector fields of SO(V,). Let A, the projection from O(V,) on V, and let us consider the following Stratonovitch equation (See [LW]) starting from u:
(3.12)
77
with Bf a fixed Brownian motion in Rd. We have a representation of the Brownian motion in z:(z) =
Let <, be the complex bundle on V, considered as before. Let T ~ ( z ) be the parallel transport along the Brownian path t + zy(z) for the bundle E, endowed with the connection Vcu. We have the following
stochastic representation to the semi-group associated to A"'" :
(3.13)
+(.l(.)).
exp[-t/2Av'"]@(z) = E[(TjY (z))-'$b (zy(z))] . We deduce a stochastic realization of the semi-group associated to Rr, by using the Feynmann-Kac
formula. The semi-group exp[-t/2RFu] acting on smooth sections p' of Ey is given by the following prob- abilistic representation (See [Biz] [LI]):
where i / ( z ) is the stochastic parallel transport along the Brownian path t -t z:(z) for the connection V Y .
Let F belonging to W . Let us introduce the J.L.O. cocycle Ch*(Vy) by
where [ . ,I denotes the supercommutator. Let us recall that if A and B are two operators acting on the sections of E considered as bundle on M ,
(3.15)
as the usual supercommutator. Since Q is even,
[A ,B ] = AB - (-l)degAdegBBA
(3.16) IVY, 41 = V F 4 - 4vr . Let us denote by dzq5 = xe:(z)&Q(z,) in a local normal coordinate system 2, of V,. Let us denote by
dub = C fe&4 in a local orthonormal basis of B after trivializing the fibration M + B. In some sense we
have identified A ( T ( M ) ) to x*A(T(B))6CL(V). We have
(3.17) P ? > Q l = d d + d g 4 .
Unlike the traditional J.L.O. cocycle for a single operator, our J.L.0 cocycle takes its values in the form on B. Let us recall in order to understand the definition that the supertrace Tr, of an operator transforming h,(B)&,+(z) into A,(B)&S$(z) is its partial trace in S$(x ) . If the operator interchanges S+ and S-, its
supertrace is equal to 0. If we consider an operator from AV(B)6S;(z) into himself, its supertrace is the opposite of its partial trace in S;(z). Let us recall that the supertrace of a supercommutator is equal to 0
([Qil)
Moreover, exp[-tR'$'] has an heat kernel qy(z,z'). We have:
4; exp[-slR?][VF, Q;] exp(-(sz - sdRF]. .[VY, $3 exp[-(l- sdR,"Id~~(z)
(3.18)
where dmv, is the Riemannian measure on V,
78
Therefore,
T ~ . { & e x ~ [ - s i R r ] [ V r , Sil exp[-(sz - sdRr]...[v?, 4:1 exp[-(l- ~ n ) R r l l
(3.19) = s,-+> T..{~x(.)4:,(z,zl)(dz4i + ~ ' 4 3 b l ) . . .
(&4: + dy4:)(zn)qY-s,(Zn, Z)}~~V,(Z)...~~V"(Z,) . Over A ( B ) , we consider the Laplacian dLds + dkds = A, and for k belonging to N , the various Soholev spaces:
(3.20) ~ ~ d ~ ( , q , h = (Ak, f l ) U , O > dmB
Ch norms of u can be estimated by the systems of norms (3.20) by Sobolev imbedding theorem ([Gi]). We have:
Lemma III.1:Let HYl,.,,sn($o, ..,+") (SI < sz < .. < s, < 1) be the operator acting on the sections of E,:
(3.21) $' + 4'exp[-siRT']4'..4"exp[-(l - s,)R~,]$J'
where the 4, are tensorial operators acting on section of the bundle E, considered as a bundle on M . Then
(3.22) H:,,.+" . . I 4")3'(4 = q:l,.,,sn(40, .., P ) ( z , z')$JYz')dmv,(z') . L Moreover the covariant derivatives in y, z and z' of the kernel qf,,,,,s,,($o, ..,@"'(z, z') can be estimated by
C"n 11q411k for an integer k which depends only from the order of the derivative. Proof: We consider a probabilistic representation of HY,,,,,sn($o, ..,@"). It is given by
where ?:,#, (z) if s < s' is the stochastic parallel transport for the bundle E, along the Brownian path z:(z)
runned in the opposite sense from zy,(z) to z:(z). We neglect the fact in (3.23) that the Brownian motion
represents the semi-group associated to RTv/2 and not the semi-group associated to R?'. We would like to apply Malliavin Calculus to estimate the density of HY,,,,,sJ@"' ..,@'). Let H be the
Hilbert space of functions h from [0,1] into Rd such that h(0) = 0 and such that Ji Id/dsh(s)12ds < w. It is the reproducing Hilbert space of the Brownian motion t + B;. In order to define the Sobolev norms of a Wiener functional F , we take its derivatives in the direction of H . Its derivative of order T is realized as a random element V'(s1, .., sr) of the symmetric rth-tensor product of H . The Sobolev norms of Malliavin Calculus are given by:
We work in a trivialization 0 x V of the fibration M . &ur(z) belongs to all the Sobolev spaces of
Malliavin Calculus, after imbedding V in a l i ea r space R" and trivializing SO(V,) by imbedding sod into
Gl(R").The derivatives in z, y of the functionals which are considered are bounded in all the Sobolev spaces of Malliavin Calculus by Cnll@llh for some k (We use Sobolev imbedding theorem). The same results holds
namely for &?Y,,82(z) after trivializing the family of bundle Ey considered as a bundle on M . Moreover,
the Malliavin matrix of z:(z) is uniformly bounded in all the Lp in z and y. The results holds by Malliavin
79
Calculus (See [Nu], [I.W], [L1] p 394.). Namely for any vector fields X I , .., X , on M in a local trivialization 0 x V of the fibration M , we have:
where 11.1103 denotes the supremum norm.
0 Remark Instead of using the non intrinsic Malliavin Calculus, we could use the geometrical Malliavin
Calculus on a manifold developed by Bismut and LQandre in order to state this lemma (See for instance [L5]
and [L6]). Lemma 111.1 allows us to state the following theorem: Theorem 111.2: The J.L.O. cocycle Ch*(Vy) is a white noise distribution (an element of the topo-
Proof: We write: logical dual W' of W) with values in L I ( B ) ~ - .
(3.26)
Then
(3.28)
for some k' by Lemma 111.1. But, on the other hand,
(3.29) Il4*> Ilk, = (A,, + 1)"'* ,
By proceeding as in Theorem 111.2, we deduce that there exists kl and Cl independent of F such that
(3.30)
0
V and V' first order operators acting on E,. Then
II < Ch*(V?),F > I I A ( B ) , ~ 5 CIIFllk,,cl
Lemma III.3:Let q:(x, z') be the heat kernel associated to the heat-semi group associated to RFu and
(3.31)
(3.33)
where d$(., .) denotes the Riemannian distance on V, and where t 5 1
80
Proof: We follow the method of the proof of [La] of this fact for the scalar heat kernel. Let h be a
mollifier function equal to zero outside a small convex neighborhood of z. Then the kernel associated to the operator
is bounded as well as its derivatives by exp[-C/t] by using the tools of Malliavin Calculus, because by exponential inequality P{d,(z,z:z)) > 6 > 0) 5 exp[-C/t] and because the inverse of the Malliavin matrix associated to z:(z) is bounded in LP by Ct-n(P) when t + 0. So it is enough to study the density of
(3.35) + E[h(zf(z))$,&) ~ y ( ~ ~ ( ~ ) ) ~ ~ / ~ l $ y ( ~ ~ ( ~ ) ) l .
We do the traditional time scaling, in order to come back at time 1 by replacing dB; by t1/2dBB in order to replace the short time asymptotic by a the study of a diffusion in time 1 which depends from a small parameter 4 (See [Mo]). But we don’t change the notation, in order to me more succinct. On the other
hand, there is a C such that e~p[C~;(~’;‘(’))] as soon as z f (z ) is close enough of has Sobolev norms in Lz of each order as its derivatives in z and y bounded (See [La]). Namely, we have the following large deviations estimates (See [F.W], [L3]) when t + 0:
where Int 0 denotes the interior of the borelian subset 0 of V, and clos 0 its closure. We consider the operator p: (z):
It has a kernel by proceeding as in [Lz] bounded by & with first derivatives bounded by &. We operate as in [La] to do that. We work in normal coordinates around z in V,. We consider the rescaled operator ur (z):
with some natural notations, because E, is locally trivial around z. It has a bounded density as well as its derivatives when t + 0. Let us denote by this density. The supremum norm of the density of p:(z) is bounded by t-d/zsup,, lQr(z, Analogously statements works for the derivatives of the density
of P:(x).
0. We conclude as in [La].
This lemma allows to show: Theorem 111.4:(b + B)Ch*(VY) = 0.
RemarkThis means that < Ch’(VY), ( b + B)F >= 0 for all F belonging to W . ProoEThe proof is exactly the same than in the Proof of Theorem A of [G.S], the bound of Lemma
111.3 allowing to justify the algebraic computations. It is enough to apply these bounds, and the Kolmogorov relation in the classical bound of the heat kernel p:(z, associated to the Laplace-Beltrami operator on V, as in [J.L]:
(3.39)
for t 5 1
81
0 Let us consider for T E [0,1] the superconnection VF7:
(3.40) vTT=v:+'d2P-1)1v?,P]
(See [G.S] p 357.). It has a curvature RF7.
associated to R F in Lemma 111.3. Lemma 111.5: The heat kernel associated to RFT satisfies the same estimates than the heat kernel
Proof: Let us write RFT = RF +A where A is a first order operator. We apply the Volterra expansion:
(3.41) exp[-tR:] + (-1)"exp[-siR~]Aexp[-(sz - s~)R:]A..Aexp[-(t - s,)R;]dsI..d~, .
By using (3.38), Lemma 111.3 and Kolmogorov formula, the kernel q?(z, z') of I , satisfies to
1 (3.42) 14;(% z')l 5 P%A% .') dsl..dsn .
l < s l < s 2 < . . < s n < t J s l ( s z - sl)..(t - Sn)
But (See [J.L] p 144.)
C" dsi..ds, 5 -
1
a . (3.43) %./SI(SZ - S l ) . . ( t - Sn)
The result arises when we apply (3.39) to p&(z , z'). We estimate the first derivatives of the kernel of I,, by
using the same arguments.
0 We can d e h e Ch*(VF7) by using Lemma 111.4 and show that it is a Hida distribution, if we disregard
the smoothness. The estimates of the first derivatives of the heat kernel associated to R G allow to justify
the algebraic considerations of [GS] and to show that:
Lemma III.fkd/dT(< Ch'(VTT),Ch.(p) 2) = 0. We deduce that: Let us state the main lemma, analogously to Lemma 2.2. of [GS] in order to show Theorem 111.4 and
We consider some tensorial operators @ acting on sections of E. considered as a bundle on M . We Lemma 111.6.
consider the expression:
(3.44) < H'($O, ..,$") >= J Tr, { do exp[-sl RT7]$'.. .$" exp[- (1 - s,)RTT]}dsl. .ds, O < S l < . . < S , < l
The difference with the delinition of [GS] p 348, is that < H'($O, ..,@') > takes its values in the space on forms on B unlike the scalars. We remark first of all that:
-i)[VFT,$] is a first order operator. -ii)[RFT,$] is the sum of composition of first order operators.
-iii)[VcT, d / d ~ V ? ~ ] is a first order operator. These remarks and the bounds given in Lemma 111.5 allow us to state the analog of Lemma 2.2. of [GS]
Lemma 111.7: Let us consider some tensorial operators @ acting on the sections of E. considered as a with values in forms unlike scalars, because the supetrace of a supercommutator is equal to 0:
bundle on M . We have:
(3.45) < ~ ~ ( 6 0 , , , , p n ) >= ( - l ) ( d e s i ' + . . . + d e s e ' - ' ) ( d ~ s e ' + . . + d . s d " ) < H'(d', ..$",$O, . . , $$ - I ) >
82
(3.48) < H'(&, .., [ R G , 41, ,., 6") >=< H'(cj0, ..,@-'#, ..,$") > - < HT(40, __, @-' ,@f+ ' , . .qY" ' >
n
(3.49) dldr < H'(d0, .., 6") > + C < H'($', .., @, [VrT,d/drVTT],@+', ..,qbn) > . *=O
Let us recall that KerpD,p can get some jumps of dimension when y is moving. CokerpD,p can get too some jumps of dimension. This means that y + KerpD,p does not in general define a bundle on B. But since the kernel of pD,p and its cokernel have the same jumps of dimension, this justifies that
1ndpD.p = Ke7pD.p - C0kerpD.p is a virtual bundle in complex K-theory sense. It is the virtual Index bundle of Atiyah and Singer associated to the family of twisted Dirac operator pD,p.
If <B is a complex bundle endowed with a connection VtB, we can define its curvature RCB. The Chern
character Ch(<s) has a representative given by
(3.50)
The Chern character does not depend in cohomology of the connection on E B . The Chern character can be extended to the complex K-theory as follows: the Chern character of the difference of two bundle is equal to the difference of the Chern character of each bundle. Atiyah-Singer's theorem for families compute the Chern character of the Index bundle of a family of Dirac operators.
Theorem 111.8: < Ch*(Vy),Ch.@) >= Ch(lndpD.p) where Ind(pD p) is the Index bundle on B of
the family of twisted Dirac operators y + pD,p.
Proof: We remark that if T = 1, it is nothing else than the theorem of Bismut ([Biz], [Ll])
(3.51) < Ch*(VFl), Ch,(p) >= Ch(lndpD.p)
where Ch is the Chern character in real phase of the Index bundle, because in such case p commute with VTl such that (See [GS] p 356.)
(3.52)
The result arises by Lemma 111.6.
0
< Ch*(Vrl),Ch,(p) >= Tr,pexp[-Rr1] = Tr,pexp[-Rr1]p = Ch(1npD.p).
IV. References
[A.B] Accardi L. Bozejko M.: Interacting Fock spaces and gaussianization of probability measures. Inf. Dim.
Ana. Quant. Probab. Rel. Top 1 (1998), 663-670. [A. N] Accardi L. Nahmi M.: Interacting Fock spaces and Orthogonal Polynomials in Several Variables.
In "Non-commutativity, Infinite-Dimensionality and Probability at the Crossroads". Obata N., Matsui T., Hora A. Edt. World Scientific (2003), 192-205. [At] Atiyah M.: Circular symmetry and stationary phase approximation. In "Colloque en I'honneur de L. Schwarta" Asterisque 131 (1985), 43-59. B] Bereaansky Y.M.: Infinite dimensional non-Gaussian analysis connected with generalized translation
0perators.h "Analysis on infinite-dimensional Lie groups and algebras". H. Heyer J. Marion edt. World Scientific (1998), 22-46. [B.K] Berezansky Y.M., Kondratiev Y.O.: Spectral methods in infinitedimensional Analysis. Vols 1, 2, Kluwer (1995).
83
[B.T] Berezansky Y.M. Tesko V.A.: the spaces of test and generalized functions, connected with generalized translation. Ukrainian Math. J 55 (2003), 1587-1657. [B.V] Berline N. Vergne M.: A proof of Bismut local index theorem for a family of Dirac operators. Topology 26 (1987),435-463. [Bill Bismut J.M.: Index theorem and equivariant cohomology on the loop space. Com. Math. Phys. 98 (1985), 213-237. [Biz] Bismut J.M.: The Atiyah-Singer for families of Dirac operators: Two heat equations proofs. Invent. Math. 63 (1986), 91-151. [Bi3]Bismut J.M.: Localization formulas, superconnections and the Index theorem for families. Com. Math. Phys. 103 (1986), 127-166. [C.K.S] Cochran W.G., Kno H.H., Sengupta A,: A new class of white noise generalized functions. Inf. Dim. Ana. Quant. Probab. rel. Top. 1 (1998), 43-67. [Col] Connes A.: Entire Cyclic cohomology of Banach algebras and characters of 0-summable Fredholm modules. K-Theory 1 (1988), 519-548. [Coz] Connes A.: Non-commutative differential geometry. Academic Press (1994). [Do] Dowelly H.: Local index for families. Michi. J. Maths. 35 (1988), 11-20. [El] Elworthy K.D.: Stochastic differential equations on manifolds. L.M.S. Lectures Notes Series 20. Cam- bridge University Press (1982). [Em] Emery M.: Stochastic Calculus in Manifolds. Springer (1989). [F.W] Freidli M. Wentzell A,: Random perturbations of dynamical systems. Springer (1984). [G.H.O.R.] Gannoun R., Hachaichi R., Ouerdiane H. Rezgui A.: Un theoreme de dualit6 entre espaces de fonctions holomorphes a croissance exponentielle. J. h n c t . Ana. 171 (2000), 1-14. [GI Getzler E.: Cyclic homology and the path integral of the Dirac operator. Preprint (1988). [G.J.P] Getzler E. Jones J.D.S. Petrack S.: Differential forms on a loop space and the cyclic bar complex. Topology 30 (1991), 339-371. [G.S] Getzler A. Szenes A,: On the Chern character of a theta-summable Fredholm module. J. h n c t . h a . 84 (1989), 343-357. [Gi] Gilkey P.: Invariance theory, the heat equation and the Atiyah-Singer theorem. C.R.C. press (1995). [HI Hida T.: Analysis of Brownian Functionals. Carleton. Math. Lect. Notes. 13 (1975). [H.K.P.S.] Hida T. Kuo H.H. Potthoff 3. Streit L.: White noise: an infmbe dimensional Calculus. Kluwer (1993). [LW] Ikeda N. Watanabe S.: Stochastic differential equations and diffusion processes. North Holland (1981). [J] Jaffe A,: Quantum harmonic analysis and geometric ionvariants. Adv. Maths. 143 (1999), 1-110. [J.L.O] Jaffe A. Lesniewski A. Osterwalder K.: Quantum K-theory. The Chern character. Com. Math. Phys. 118 (1988), 1-14. [J.L] Jones J.D.S. LQandre R.: LP Chen forms on loop spaces. In ”Stochastic analysis” Barlow M. Bmgham N. edit. Cambridge University Press (1991), 104-162. [Ku] Kusuoka S.: More recent theory of Malliavin Calculus. Sugaku 5. (1992), 155-173. [Ll] LBandre R.: Sur le theoreme de l’indice des familles. SBminaire de ProbabilitBs XXII. AzBma J, Meyer P.A. Yor M. edit. L.N.M. 1322 (1988), 348-413. [Lz] LBandre R.: Applications quantitatives et qualitatives du Calcul de Malliavin. In ”Col. Franco-Japonais” MBtivier M. Watanabe S. edit. L.N.M. 1322 (1988), 109-133. English translation: In ”Geometry of Random motion” Durrett R. Pinsky M. edit. Contemp. Maths. 73 (1988), 173-197. [L3] LBandre R.: A simple proof of a large deviation theorem. In ”stochastic analysis” Nualart D. SanzSolB M. edit. Prog. Prob. 32 Birkhauser (1993), 72-76. [L4] LBandre R.: Cohomologie de Bismut-Nualart-Pardoux et cohomologie de Hochschild entiere. SBminaire de Probabilitb XXX in honour of P.A. Meyer and J. Neveu. AzBma J Emery M. Yor M. edit. L.N.M. 1626 (1996), 68-100. [Ls] LBandre R.: Brownian cohomology of an homogeneous manifold. In ”New trends in Stochastic Analysis” Elworthy K.D. Kusuoka S. Shigekawa I. Edit. World Scientific (1997), 305-347. [LB] LBandre R.: Stochastic Adams theorem for a general compact manifold. Rev. Math. Phys. 13 (2001), 1095-1133. [L,] Mandre R.: Stochastic equivariant cohomology and cyclic cohomology. To appear Ann. Prob.
84
[Ls] LBandre R.: Theory of distribution in the sense of Comes-Hida and Feynman path integral on a manifold. Inf. Dim. Ana., Quant. Probab. Rel. Top.6 (2003), 505-517 [Lg] LBandre R.: Hypoelliptic diffusion and cyclic cohomology. To appear in: ”Stochastic analysis”, R. Dalang, M. Dozzi, F. Russo edt. [Llo] Lkandre R.: White noise analysis, filtering equation and the Index theorem for families. To appear in: ”Infinite dimensional harmonic analysis”, H. Heyer, H. Sait8 edt. [Lo] Loday J.L.: Cyclic homology. Springer (1998). [MalIMalliavin P.: Stochastic Calculus of variations and hypoelliptic operators. In ”Stochastic analysis” K. It8 edi. Kinokuyina (1978), 155-263. [Maz] Malliavin P.: Stochastic Analysis. Springer (1997). [M.Q] Mathai V. Quillen D.: Superconnections, Thom classes and equivariant differential forms. Topology 25 (1986), 85-110. [Mo] Molchanov S.: Diffusion processes and Riemannian geometry. Russ. Math. Surveys. 30 (1975), 1-63. [N.R] Narasimhan M.S. Ramanan S.: The existence of universal connection. Amer. J. Math. 83 (1961), 535-581. [N.T] Nest R. Tsygan B.: Algebraic Index for families. Adv. Math. 113 (1995), 151-205. [Nil Nistor V.: Super-Connections and Non-Commutative Geometry. In ”Cyclic Cohomology and Noncom- mutative Geometry”. Cuntz J.R. Khalkhali Edt. Fields Inst. Comm. 17 (1997), 115-136. [Nu] Nualart D.: The Malliavin Calculus and Related Topics. Springer (1995). [0] Obata N. White noise analysis and Fock space. L.N.M. 1577 (1994). [Ou] Ouerdiane H.: Fonctionnelles analytiques avec conditions de croissance et application a l’analyse Gaussi- enne. Japan. J. Math 20 (1994), 187-198. [PI Perrot D.: A Bivariant Chern Character for families of spectral triples. Com. Math. Phys. 231 (2002), 45-95. [QI] Quillen D. Superconnections and the Chern character. Topology 24. (1985), 89-95. [Qz] Quillen D.: Algebra cochains and cyclic cohomology. Publi. Math. I.H.E.S. 68 (1988), 139-174. [St] Streit L.: An introduction to white noise analysis. In ”Stochastic analysis and applications in physics”. Cardoso A. SknBor R. Steit L. edit. Nato Asie Series. 449. (1994), 415-441. [Sz] Szabo R.: Equivariant cohomology and localization of path integrals in physics. Lec. Notes. Physics 63. Springer (2000). [Wall Watanabe S.: Analysis of Wiener functionals and its applications to heat kernels. Ann. Probab. 15 (1987), 1-39. [Waz] Watanabe S.: Stochastic Analysis and its applications. Sugaku 5 (1992), 51-71.
85
A Quantum Decomposition of L6vy Processes
Yuh-Jia Lee* Hsin-Hung Shiht
Department of Applied Mathematics
National University of Kaohsiung
Kaohsiung, TAIWAN 811
Department of Accounting
Kun Shan University of Technology
Tainan, TAIWAN 710
Abstract
The existence of the Lkvy white noise measure A as a Bore1 measure on X’ is
proved. Then it is shown that a LQvy process can be represented by
Regarding X ( t ; x) as a linear operator in L(L,L’), the quantum decomposition of Lkvy process and the5enormalised” L6vy process are given, where L is the space of
test Lkvy white noise functionals with dual space L‘.
AMS Mathematics Subject Classification 2000 : 60H40
1 Introduction
Let X = { X ( t ) : t E W) be a LBvy process, which is an additive process with stationary
increaments on a probability space (O,F,P) with X ( 0 ) = 0 almost surely. By the well-
known LBvy-Khintchine formula,
for T E Ri and --co < s < t < +m, where fx is called the Livy function of X which is uniquely expressed as
where p is a real constant, and ,B is a positive finite measure on W with u2 = p((0)). Conversely, for a given triple (p , a, p) with p, u E W and a positive finite measure p on W, there exists a Ldvy process X such that the equality (1.1) holds. We will say that X is a
Lkvy process generated by the triple (p , u, p).
*Research supported by the National Science Council of Taiwan
+Research supported by the National Science Council of Taiwan
86
In our previous paper, we have shown that the Lkvy white noise measures A exists
and defined as a Borel measure on the space S' of tempered distributions under the the
assumption that p has finite absolute first moment. Under this assumption, Gaussian,
Poisson, Gamma processes and processes associated with Meixner class are included in
our investigation, but the stable processes are excluded. We are desirable to extend the
theory developed in [14] to stable processes. The study along this direction is by no
means complete, in this paper we devote ourselves in improving our previous results in 114) without assuming that p has finite absolute first moment.
The contents of the paper is arranged in the following manner. In section 2, the
Lkvy process is represented as a function defined on K' and it is proved that the L6vy
white noise measures always exists and is defined as Borel measure on the space K' of
distributions without further assuming the moment conditions on p. It is always asked
that if the measurable support is contained in S', we answer this question by given a criterion that the measurable support of a Lkvy white noise measure is contained in S'. In section 3, we introduced the space L of test Lkvy functionals, and the space L' of
generalized Lkvy functionals, there the Lkvy white noise derivative and creation operator
are defined and studied. In section 6, an connection between Lkvy white noise analysis and
quantum probability theory are studied. There X ( t ; z) is regarded as a linear operator
acting on L and the quantum decomposition of X ( t ; z) is given for the case where TI = 1.1 + s_"udp(u) < 00; while if TI = 1.1 + s_"udp(u) = ca, we obtain the quantum decomposition for the renomalized Lkvy process X ( t ) -.rlt. The former includes the cases
such as Gaussian processes, Poisson processes, Gamma processes and the processes in the
Meixner class; while the latter includes stable processes as special cases.
2
It is well-known that the sample paths of the Lkvy process X may be realized in the space
D(P) of all right continuous functions #J having left limits and # J ( O ) = 0. Let 00 be the
set of all continuous functions 19 of R onto itself that are strictly increasing with 19(0) = 0.
Then D(P) is a Polish space under the Skorohod metric p defined by
Levy white noise measure and a representation of LBvy processes
for any I$,$ E D(R), where kn(t) = 1 on -n 5 t 5 n, 0 at JtJ 2 n + 1, -t + n + 1 at
n 5 t 5 n + 1, and t + n + 1 at -n - 1 5 t 5 -n (see [lo]). Moreover, the Borel o-field B(!ID(W)) of the metric space (D(W), p) equals the c-field generated by all cylinder sets B of the form
B = {#J E D(R) : (#J(tl), . . . , #J(tn)) E E, E E f?(W)}, tl < ' ' ' < t,; 12 E w.
i(B)= P ( { w E R : ( X ( t l ; w ) , . . . , X( t , ;w) ) E E } ) .
Now, let i be a cylinder set measure in D(R) induced from the measure P by setting
87
By using the Kol_mogorov extension theorem, X has a unique extension to B(D(R)). Then
(D(R), B(D(R)), A) is a probability space on which
Yx = { Y x ( t ; 4) = 4( t ) : t E w, 4 E D(R)} (2.2)
is a LBvy process with the same distribution as X on (O,F,P). Next, let K ( a ) be the space of all infinitely differentiable functions on R having compact
supports in the interval [-a,a], a > 0. Then K ( a ) is a nuclear space [3]. Let K be the
union of the spaces K ( a ) , endowed with the inductive limit topology and K’ the dual of
K with the weak topology. Denote the K’-K pairing by (., .).
Proposition 2.1. The mapping T f r om the Skorokhod space (D(R),p) into the space K’ by T(4) = 4 i s injective and continuous, where 4 i s the distributional derivative of 4.
Proof. First we point out that the mapping T is well defined since each element in D(R) is locally integrable on R. The injectivity of T is an immediate consequence of the Du
Bois-Raymond lemma (see [4]). To prove the continuity of T, we assume that a sequence
{A } of D(R) converges to 4 in D(R) under the Skorokhod metric p. Then there is a
sequence {On} of 0 0 such that
SUPtat IWt) - tl -+ 0 { ::; SUP-r<t<r I4n(t) - 4(8n(t))l -+ 0
for any r E M, as n -+ +oo (see [lo, ~2921). Observe that, for any 9 E K ( a ) , a > 0,
+W
l(dn - 4,v)l I J I M ~ ) - +(t)~t l i ( t)~ d t -W
I t ~ t ) - 4( f i n ( t ) ) l l l i ( t )~d t + J a 1 4 ( ~ t ) ) - ~ ~ t l i ( t ) t d t -a
5 ( SUP 14n(t) - 4(On(t ) ) I ) t l i(t)Idt + 14(%(t)) - 4( t ) l l l i ( t ) ld t . -a<tja --a -a
So, by using (a) and (b) above, we deduce that (&, 9) converges to (d, 9). Hence T is continuous. 0
Corollary 2.2. For any Borel set B of (’D(R),p), the image T(B) of B under the mapping T as above is also a Borel set of K’. In particular, T(D(R)) i s a Borel set of K’ .
Proof. We note that the space (!ID@), p) and K’ are both standard measurable spaces; that
is, each of which is Borel isomorphic to a Borel subset of R viewed as a measurable space,
because that (D(R), p) is a Polish space and K’ is a projective limit of a sequence of dual countably Hilbertian spaces K(n)‘, n E M (see [7, 221). Then, by [7, Theorem 2.1.11 and
0
by the mapping T,
Proposition 2.1, the result immediately follows.
Let A be the probability measure on (K ’ ,B (K ‘ ) ) induced from
that is,
A ( E ) = X(T-’(E)), E E B(K’) .
88
Consider the Fourier transform CA of the measure A, which is a complex-valued functional
defined on K by
CA(T) = exp [i (x, 1711 A(dx ) . kl Then we have the following result.
Theorem 2.3. For any r j in K,
cA(17) = exP [J_&, f X ( ‘ d t ) ) d t ] ,
where f x i s the LLvy function of X as in (1.i).
Proof. From the definition of A,
Suppose that the compact support of + is contained in the interval [-a,a]. Then
+00
4( t ) +(t) dt = - s_: rj(t) L n-1
(2.4) - - - 12-00 lim C r j ( t n , j ) (4 ( tn , j+ l ) - 4 ( t n , j ) ) >
j=O
where tn, j = a(2j - n)/n, j = 0,1 ,2 , . . . , n - 1, for any n E M. Replacing the integral in
the bracket of (2.3) by the formula (2.4), we see that
n-1
cA(r j ) = ii% S,,,, exp[il)(tn,j)(Yx(tn,j+l; 4) - y X ( t n , j ; 4 ) ) 1 i ( d4 ) j=O
rn-1
17
The measure A will be called the LBvy white noise measure on K’ (generated by the
It is always asked under what conditions, the measurable support supp(A) c S’. An triple (P, ( T I P ) ) .
answer is given in the following theorem.
89
Theorem 2.4. [17] In order that the measurable support of the L6vy white noise measure
A is contained in the space S‘ of tempered distributions on R, it is necessary and suficient
that the complex-valued function log@(.,q)(l), 7 E K is continuous at zero in the topology
of the Schwartz space S.
It is easy to see that if the measure p satisfies the moment condition
+m
lulndp(u) < +co,
then the condition in the above theorem is satisfied so that A is supported by S‘.
which is equipped with the metric p defined by
V n E M, 1, By MIX we denote the set of all random variables on the probability space (K’, B(K’), A)
For any Y E Mx, we denote its characteristic function by @ y .
condition:
Let 8 be the class of all real-valued L’ n L2-functions 4 on R satisfying the following
+m
I fx(r+(s))l ds < +co for all r E R. (2 .5) L For each 4 E 8, it can be show that there exists a sequence (4%) of elements in the space
K such that
(i) {a} converges to 4 in L1 n L2(R,dt);
(ii) The sequence {(,, &)} converges to a certain random variable Y in the space (MIX, p),
+W such that
@y(r ) = exp( 1, f ,y(r+(s))ds) for any r E R.
Denote Y by (., 4). In particular, when 6 = l (s,t~, the indicator of (s, t], the charac-
teristic function a(.,+) of (., 4) is exactly the same as the one in (1.1). Therefore the L6vy
process X on (K’, B(lc’), A) can be represented in the following form:
3 Test and generalized LBvy white noise functionals
Segal-Bargmann transform
Let Bb(R2) be the class of all bounded Bore1 subsets E of R: = R2 \ {(t,O) : t E R} and define the product measure dv(t,u) = dpo(u) dt on B(R2).
The well-known LBvy-It8 decomposition (see 1211) asserts that there exist a Poisson
measure N ( E ; .) with intensity measure Y and a Brownian motion B(t) such that, for
90
b > a,
X ( b ) - X ( a ) = p ( b - a) + o (B(b) - B ( a ) ) U +i;% 1 1 [ udN( t ,u ) - - 1+u2
a < t j b n-'<lul<n
where B = {B( t ) : t E R} is independent of the system of { N ( E ) : E E ab(@)}. Let X be a positive measure on B(R2) defined by d X ( t , u) = (1 + u2) @(u) dt. Define a L2(K', A)- valued function M on { E E l?(R2) : A(E) < +m} by
r+m
Then the system of { M ( E ; z) : E E f3(R2) with X(E) < $03; z E K'} forms an inde-
pendent random measure with zero mean such that E[M(E) M ( F ) ] = X(E i l F ) for all
E, F E B(R2). The multiple stochastic integral with respect to A4 was firstly introduced
by It6 in [6]. In notation, we define In(g) by,
where g E sc((Rz)n,A@n), the closed subspace of Lz((R2)n,X@n) consisting of all sym-
metric complex-valued functions in L:((R2)", Awn) Then we have
We are ready to state Lkvy-It6 decomposition theorem for square integrable Lkvy white
noise functional.
Theorem 3.1. [12] Let 'p be given in L2(K ' ,A) . Then h
( i ) there exist uniquely a series of kernel junctions & E L2((R2)"', A@"'), n E M u {0 } , such that 'p is equal to the orthogonal direct sum CFy0 @In(&). In notation, we
write 'p - (&).
(ii) L2(Ic', A) is isomorphic to the symmetric Fock space .Fs(Lz(R2, A)) over Lf(R2, A) bycarryingcp-(+,) i n to ( f iq5Ol f iq51 ,..., fi& ,... ).
For 'p E L2(K ' ,A) , say 'p N (&), the Segal-Bargmann (or the 5'-) transform of 'p is a
complex-valued functional on L:(R2, A) by
Then S-transform is a unitary operator maps from L2(K' , A) onto the Bargmann-Segal-
Dywer space .F1(L:(R2, A)) over L:(R2,X). Moreover, we have
91
where D is the FrBchet derivative of Sv and 11 . l l L ( , , ) ( H ) denotes the Hilbert-Schmidt
operator norm of an n-linear functional on a Hilbert space H . (2)
The Segal-Bargmann transform enjoys an integral representation introduced in 1121.
Let A = -d2/dt2 + (1 + t2) be a densely defined self-adjoint operator on L2(R, d t ) and
{en : n E No} be eigenfunctions of A with corresponding eigenvalues 2n + 2, n E N U (0). {en : n E No} forms a complete orthonormal basis (CONS, in abbreviation) of L2(R2,dt) . For m y p E W and q E L2(W, d t ) , define Iqlp := I A P q 1 p ( ~ , d ~ ) and let S, be the completion of the class (77 E L2(B,dt) : )q), < +a} with respect to 1 . Jp-norm. Then S, is a real
separable Hilbert space and we have the continuous inclusions:
S = I ~ p ~ o S p ~ S p c S g c L 2 ( W , d t ) C S - , ~ S - p C S ' = ~ p > O S - p , p > q > O .
Similarly consider the real Hilbert space L2(R,y), where dy(u) = (1 + u2)dp(u) is a
0-finite measure on (R, a@)). Choose a CONS ((0, <I, . . .} of L2(R, y). Define a linear
operator Ap densely defined on L2(R,y) by Ap Cn = (2n + 2) Cn for n = 0,1,. . .. For each
p 2 0, let Ep be the set of all C E L2(R,y) with lA;C1+2(~,~) < +co. Then Ep is a real
separable Hilbert space with the inner product (., .)p,p given by
and induced norm by I . lp,p := d E .
of L2(W,y) with the inner product (.,.)-,,p and I . I-,p-norm by
Denote by E-, the dual of Ep. Then E-, is isometrically isomorphic to the completion
where we identify z E E- , with En(%, In ) In in which (., .) is the E-p-€p pairing. Let
E = lim &-,. E c L2(R,y) c E' forms a Gel'fand triple. Additionally, we assume that 1 E E'.
Remark 3.2.
Ep. Then E is a nuclear space with the dual E' = +p>o
(1) If the measure p satisfies the absolute moment condition of all order, i.e., J:z lul" dp(u) < +co for all n E N, then we can apply the method of Gram-Schmidt orthogonalization
to {1,u,u2,. . .} to obtain a CONS (<0,[1,. . .} of L2(W,y). In this case, 1 E E c E'.
(2) Let X be an a-stable process with 0 < (Y _< 1 such that
d ~ ( u ) = ci I u I ~ - ~ l(-w,o)(u) d~ + ~2 2 ~ ~ - ~ l(o,+,)(u) du, ~ 1 , c2 > 0.
We can choose a CONS {Q, ( 1 , . . .} of L2(R, y) by setting
where hn's are the Hermite functions on R. Then
and it is obvious that E E E' with ( (u) = un, u E R, for any n E N U (0).
92
For p E R, denote by Np the Hilbert space tensor product S, C3 EP with 1 . Ip-norm
defined by lei @ Cj Ip = lei\, I<jIp,p. So, No = I;:(B2, A). Let N = S @ E with the dual N' = S' @ E'.
Then N c L2(Rz,A) c N' forms a Gel'fand triple. Moreover, the inclusions N c Np c N, c L2(R2, A) c N-, c N-, c N', p > q 2 0, are all continuous.
In the following, we denote I . I , :=I. I - NZZ
for simplicity, where CONS
denotes the symmetric tensor product. In addition, we relabel the
{en @ Cm : n,m = 0,1, ... } by {fo,fi ,... 1, in such a way that fo = e o @
For p E W and 'p E L2(S', A), define
and let Aj = I ( A @ A p ) f j b for j E N U (0).
We next proceed to construct the spaces of test and generalized functions as follows.
and let L, be the completion of the class { 'p E L2(S' ,h) : 1\'pl), < +co} with respect to 1) . Ilp-norm. Then L,, p E R, is a Hilbert space with the inner product ((., .)), induced by 11 . Il,-norm. For p , q E R with q 2 p , L, c L, and the embedding L, - L, is of Hilbert-
Schmidt type, whenever q - p > l /2 . Let L = l&p,o L,. Then L is a nuclear space. L will serve as the space of test functions and the dual space L' of 13 the space of generalized functions. The members of L' are called generalized Lkvy white noise functionals. In this
way, we obtain a Gel'fand triple L c L2(S', A) c L' and have the continuous inclusion:
L c L, c L, c L2(S',h) c Lb c LL c 13' = 9,,, L;, p 2 q > 0.
In what follows, the dual pairing of L' and L will be denoted by ((. , .)) .
Example 3.3. For 17 E Ic, 11(7 8 1) E L' since 17 @ 1 E N', where 11(h) is the Lkvy-It6
integral with the kernel function h E L2(R2, A).
For g E Lz(B2,A), it is easy to see that l lEM(g)l lp = e(1/2)lgg for any p > 0. Hence
EM(g) E L, if and only if g E JfP,= for p > 0. We define the S-transform for F E 13' by
SF(g) = ((F, & M ( 9 ) ) ) , 9 E N .
Annihilation and creation operators
Let F E L, and ( E N-p,c, p E B. The Ggteaux derivative (d/dz)l,,o SF(.+zE) in the direction ( is an analytic function on N-p,c. In fact, by using the Cauchy integral formula
and the characterization theorem [14], one can show that S- l ( (d/dz)l,=, SF( . + z l ) ) E
Lp-z. Define
a, F = S-'( (d/dz)l,_, SF( . + z < )).
Then we have a, F in L,-1.
93
It is clear that 8, is continuous from C into itself. Its adjoint operator 8; is then
defined from by
((8; F, p)) := ((F, 8 ~ 9 ) ) for F E C‘ and p E L.
8, is called the annihilation operator and 6’; is called the creation operator.
It can be shown that, for p > 1/2,
Let Ap be a maximal Bore1 subset of W2 such that Cj”=, Ifj(t,u)121fj1!p is finite for
( t ,u) E Ap. Then A(W2 \ A,) = 0 and A, C A, as p 5 q. We note that ( t ,u) E A, for
any t E W, whenever p({u}) > 0. Let
A = u AP. p E N; p > 1/2
Then A(W2 \ A) = 0. Define b(t,u) be the functional on Np, p > 1/2, by
if ( t , u) E A; otherwise, b(t,u) = 0, where the sum in (3.6) is absolutely convergent in N,. It is easy to see that if F E L, ( p 2 2), then, for [A] almost all ( t ,u) E R2,
8(t,u) F = &(,,+, F in L2(K’, A).
and, for p > 1/2 and ( t l , ul) , . . . , (tn,un) E W2 \ A, we have
- A
(b(t l ,ul)@. . . @ 6(t,,,u,,), 9) = g((tl,u1), . . . , (tn,un)), g E NtF,
where (., .) is the N!&-Nfp pairing.
If the L6vy white noise measure A is analytic, then L2(S’, A) includes square integrable
analytic functionals. Let &Ain(SL) be the space of the projective limit of {&;,,(SL)} for
which &j,k(SL) consists of all analytic functionals p on SL such that
sup{l‘p(z)l e-( l ’k ) ’+p : z E S-,,=} < +m.
In this case supp(A) c S’, and, by [14, Theorem 2.71, &Ain(SL) c L2(S’,A). Then the cre-
ation and annihilation operators enjoy respectively the following integral representations.:
(ii)
8; ‘p = uh(t, u) p(. - ubt) dN(t , u) - u q t , u) dv(t, u) L? (L? 1
94
4 Quantum decomposition of LBvy processes
For a fixed p 2 0, denote by M , the class consisting of all functions h in N, so that the
associated multiplication operator Mh, which is defined by Mh(g) = h* g for g E Np,,, acts
continuously frornJ& into Lt(R2,X), where h*(t,u) = uh(t,u), (t ,u) E R2. For h EM,, let d r h be the differential second quantization of Mh, i.e., for 91,. . . ,gn E N,,,,
mh(g1G . . . G g n ) = Mh(g1) G g2G . ‘6 gn + 91% ~ h ( g 2 ) G 93 G . . G gn
+ . . .+ glG...Ggn-lGMh(gn) .
Now, let 8; be the linear operator on the linear space spanned by In(gl%.. .Ggn), 91,. . . ,gn E N,,, and n E M, defined by
8; In(g1G . ’ 3 gn) = In(drh(giG . . . G gn)) .
For ‘P N ( 4n ) E L,, 4 2 0, let
Let h 6 M,, p L. 0 and p E L, with q - p 2 1. Then
{% (k n=O ln(d’n,k))}w k=O
is a Cauchy sequence in L2(K’, A).
This leads us to the following
Definition 4.1. Define
It follows immediately from the Definition 4.1 that we have
where “sym” means “the symmetrization of ”. Moreover,
11% (PI10 5 lllP11q1
where IlMhll is the operator norm of Mh. 8: is called the conservation operator indexed by h.
95
Theorem 4.2. Let h E M , with p 2 0 and 'p N (& ) E L, with q - p 2 1. Then 8, 'p, 8; 'p, and 8; 'p are in Lz(S', A). Moreover, for [A] almost all x E K',
I l ( h ) ( x ) Lp(X) = ah ' f ( X ) + 8; dz) + 8; dx). (4.1)
In particular, when q 8 1 E M , for q E K, and $-'," u dp(u) < +03,
for [A] almost all x E K', where 71 = p + $-'," udp(u )
The identity (4.1) is called the quantum decomposition of the process I l (h) . If 7 1 is
finite the identity (4.2) is called the quantum decomposition of the process (x,q).
Proof. (Sketch) We verify the identity (4.1) for 'p = Im(gBn)
By the product formula [13], we have
I i ( h ) L ( g B n ) = m( h, g)L- l (gom- l ) + L + l ( h G g B m ) +mIm((hg)*%(g"- '))
= ah Im(gBm) + 8; Im(g@'") + 8; Im(gBm).
The second assertion follows from the fact that
0
In the rest of this paper, we assume that there is a fixed positive number po such that
for any q E K , the associated operator by carrying g( t ,u ) into uq( t )g ( t ,u ) , (t,") E R2, is continuous from Npo,c into Lz(R2, A). Then the definition of the conservation operator
8; can be extended to h E K 8 1 so that the related properties stated as above hold. For notational convenience, we identify 8; with
The conservation operator can also be written as the following more familiar form: for
q E K and 'p E L,
where the integral exists in the sense of Bochner (see[l4]).
lemma.
Lemma 4.3. Let 'p E C and the sequence {qn} of K converge to 4 in L1 nL2(R ,d t ) . Then
To derive the quantum decomposition for the L6vy process, we need the following
exists. We denote such a limit by I i ( 4 ) 'p.
96
Proof. Obviously {8,,,+n,nl p} and {13,f,,,~[-,,,~ p} are convergent to 8, p and 8; ‘p in L’ respectively.
On the other hand, since for q - po 2 1 and q E lC,
M
j = O
m
which implies that
then we have
Apply the above estimation and use the quantum decomposition of I1(vn), the lemma
follows immediately. 0
Finally apply the above lemma for q5 = l[o,tl, we derive the quantum decomposition of
LBvy processes as follows.
97
Theorem 4.4. The "renormalized" L h y process X ( t ) -rlt is a continuous operator from L into L' and we have
( X ( t ) - Tit) 'p = &[o,y81 'p + ~ ~ [ o , c l @ l 'p
+ I' 1: 'u. q,,") a&(,,") P (1 + u2) dP(u)dt. (4.3)
If r1 is finite (a case which excludes the a-stable process with 0 < a 5 l), we easily obtain the quantum decomposition for X ( t ) from the identity (4.3).
Example 4.5. Let X = { X ( t ) : t E R} be an a-stable process with 0 < a 5 1 such that
(1 + u2) d P ( U ) = C I I U ( ' - ~ l ( - m , o ) ( ~ ) du + c2 u ~ - ~ ~ ( O , + ~ ) ( U ) du,
Rom the Remark in the Section 3 we can obtain that for any g E N,,
~ 1 , c2 > 0.
3 4 2 1 ) = vm%7(t,4 l(O,+m)(.) + & - i T = g ( t , U ) l ( - w , O ) ( 4 E S@S.
Let p , q > 0 so that 1@ u2 E S-,@S-,, and Ihlcl, 5 Const. Ihlq Ilcl, for any h , k E S,@S,. Then, for q E Ic,
~ ~ q ( t ) ~ ]g( t ,u) I2 dX(t,u) = ~ ~ q ( t ) ~ Ic(t,u)12dudt s, L 2
I 171k I1 @ 4 - p . l?lP
- < Const. 11 @ u21-, 1 ~ 1 , " . Thus, by Theorem 4.4,
References
[l] Berezansky, Y. M., Kondratiev, Yu. G., Spectral Methods in Infinite Dimensional Analysis, (in Russian), Naukova Dumka, Kiev, 1988. English translation, Kluwer
Academic Publishers, Dordrecht, 1995.
[2] Doob, J. L., Stochastic Processes, Wiley, New York, 1953.
[3] Gel'fand, I. M., Vilenkin, N. Y., Generalized Functions, Vol 4. Academic Press, 1964.
(41 Giaquinta, M., Hildebrandt, S., Calculus of Variations, Vol. I, Springer, Berlin, 1996.
[5] Hida, T., Kuo, H.-H., Potthoff, J., and Streit, L.: White Noise: An Infinite Dimen- sional Calculus, Kluwer Academic Publishers (1993).
[6] ItB, K., Spectral Type of Shift Transformations of Differential Process with Stationary Increments, Trans. Amer. Math. SOC. 81 (1956) 253-263.
98
[7] It6, K., Foundations of Stochastic Differential Equations in Infinite Dimensional Spaces, CBMS-NSF Regionnnal Conf. series in Applied Math., Vol. 47, SIAM,
Philadelphia, 1984
[8] Ito, Y., Generalized Poisson functionals, Probab. Theory Relat. Fields 77 (1988) 1-28.
[9] Ito, Y., Kubo, I., Calculus on Gaussian and Poisson White Noises, Nagoya Math. J. 111 (1988) 41-84.
[lo] Jacod, J., Shiryaev, A. N., Limit Theorems for Stochastic Processes, Springer, Berlin,
1987.
[ll] Kuo, H.-H., White Noise Distribution Theory, CRC Press, 1996.
[12] Lee, Y.-J., Shih, H.-H., The Segal-Bargmann Transform for Lkvy Functionals, J. h n c t . Anal. 168 (1999) 46-83.
[13] Lee, Y.-J., Shih, H.-H., The Product Formula of Multiple Lkvy-It6 Integrals, Bull. Inst. Math. Academia Sinica 32, No. 2(2004), 71-95.
[14] Lee, Y.-J., Shih, H.-H., Analysis of Generalized Lkvy White Noise Functionals, J. Funct. Anal. 211(2004), 1-70
[15] Lee, Y.-J., Shih, H.-H., A Characterization of Generalized L6vy Functionals, Quantum Information and Complexity, World Scientific, 2005, 321-339
1161 Lee, Y.-J., Shih, H.-H., The Adjoint of Lkvy White Noise Derivetive, in preparation, 2005
[17] Lee, Y.-J., Shih, H.-H., On the Support Property of Lkvy White Noise Measures, preprint, 2005
[18] Lytvynov, E. W., Polynimials of Meixner's Type in Infinite Dimensions-Jacobi Fields
and Orthogonality Measures, J. Funct. Anal. 200 (2003) 118-149
[19] Meyer, P. A., Quantum Probability for Probabilistss, Lecture Notes in Math. 1538,
Spring-Verlag, 1993.
[20] Parthasarathy, K. R., A n Introduction to Quantum Stochastic Calculus, Basel/ Boston/ Berlin, Birkauser (1992).
[21] Sato, K., L h y Processes and Infinitely Divisible Distributions, Cambridge University
Press, 1999.
[22] Yamasaki, Y., Measures on Infinite Dimensional Spaces, Series in Pure Mathematics,
Vol. V, World Scientific, Singapore, 1985.
99
GENERALIZED ENTANGLEMENT AND ITS CLASSIFICATION
T. MATSUOKA Faculty of Management of Administration and Information
Tokyo University of Science, Suwa Chino City, Nagano 391-0292, Japan E-mail: matsuoka@rs.suwa.tus.ac.jp
Recently the quantum entanglement has been focused to study in quantum infor-
mation theory. Several classifications of separable and entangled states have been studied extensively by many authors. Belavkin and Ohya (MO) gave a rigorous
construction of quantum entangled state by means of Hilbert-Schmidt operator and gave a finer classification of separable and entangled state. In this article we re- view their entangling operator approach. The information degree of entanglement is applied for the entangled PPT state.
1. Introduction
The entanglement was introduced by Schrodinger in 1935 out of the need
to describe correlations of quantum states not captured by mere classical,
statistical correlations which are always the convex combinations of non-
correlated states. There have been various studies of the entanglement in
[1]-[8], in which the entangled state of two quantum system is defined by a
state not written as a form
with any states pk and g k . A state written as above is called a separable
state, so that an entangled state is a state not belonged to the set of all
separable states in the naive definition above. However the mathematical
characterization of entanglement is not yet fully understood except for some
simple cases, for example pure states or low dimensional mixed states. Fur-
ther it is obvious that there exist several types of correlated states, written
as separable forms. Such correlated states have been considered in several
contexts in the fields of quantum information and quantum probability such
100
as quantum measurement and filtering[9, lo], quantum compound states[ll,
121 and lifting[l3]. Belavkin and Ohya studied a rigorous operational struc-
ture of quantum entangled state by means of Hilbert-Schmidt operator (it
is called the entangling operator) and gave a wider definition of quantum
entanglement [14,15].
In this article we review how to construct the entangling operator H for a given compound state w according to [14, 151. We also discuss the
information degree of entanglement given in [14, 151 and its application
which is studied in [16]. The criterion is applied for entangled PPT states
in the C3 8 C3 model of Horodeckis 171. We report some results obtained
in [lS].
2. Operational structure of quantum entanglement
2.1. Pure state
Let K be a (separable) Hilbert space and B (K) be the set of all bounded
linear operators on K. A normal state cp on B (K) can be expressed as
cp(B) = trxH*BH = trxBa, B E B (K) , (1)
where H is a linear Hilbert-Schmidt operator from an another Hilbert space
'H to K (i.e. XI, IIH Ix1,)11~ < +oo for any complete orthogonal system
(CONS for short) (1x1,)) in 'H) and a is the density operator corresponding
to the normal state cp. This H is called the amplitude operator, which can
always be considered on 7-t = K as the square root of the operator HH* (i.e. H = a+), and it is called just the amplitude if 'H is one dimensional
space C, corresponding to the pure state cp (B) = (<I B I<) here H is given
as a vector I<) E K with 1 1 1 < ) 1 1 2 = 1.
The auxiliary Hilbert space 'H and the amplitude operator H are not
unique; however H is defined uniquely up to unitary equivalence in 'H. 'H can always be taken to be minimal by identifying it with the support K, 3
E,K for a, defined as the closure of aK (E, is the minimal orthoprojection
in B (K) such that aE, = a). In this case the amplitude operator H can
be represented as
n
where {le,)} is a CONS in 'H and IH,) is the orthogonal eigen-amplitudes
such that IHn) E 'H, (H,IH,) = p,Sn,, with the eigenvalues p , of the
101
density operator u. Note that u = En lHn)(Hnl is a Shatten decomposi-
tion, i.e., a one-dimensional orthogonal decomposition of u. If 'H is not one
dimensional, then the dim% must not be less than rank H* = rank u and
the dimension of the ran H*H coincide with that of the support ' H H ~ H . We can always equip 'H with the tilde operation "-" ('H-transposition)
x = JA*J given by an anti-unitary operator J on { / e n ) } in (2) as Jle,) = le,) and (JzI J y ) = (yI z) for z, y E 'H. Thus J is the standard complex con-
jugation in an eigen-representation of H*H (i.e. J ; C c,le,) H C c l e n ) , (em] xle,) = (en! A(emA(emIZ21(en) = (e,l A*Jem) with 71 = J A J ) such
that H*H = H*H = H*H. Given the amplitude operator H from 'H to K , one can define not only
the state 'p but also the normal state
-
?1, (A ) = t rxHXH* 5 t rxAp, A E B ('H) (3) hl
on B ('H), where the corresponding density operator p is given by H*H, as
the marginal of the following pure c m p o u n d state w
defined on B ('H @ K) on the composite system 'H @ K. The above bilinear form is uniquely expressed by the vector 10) E 'H@K,
which is given by (C @ 7]R) = (71HJC) for all I<) E 3.1 and 17) E K. In fact,
it holds that
where { Ikn)} is a CONS in K.
Lemma 2.1. For any A E B ('H) and B E B (K) ,
w ( A @ B) = t rXXH*BH = (RIA @ BlR)
holds.
102
Note that if H is represented by (2), then [R) becomes (R) =
C le, 18 Hn) as follows n
n
n
n n
Now let us move to the entangled mixed state and its entangling oper-
ator expression.
2.2. Classification of states via entangling operator
We can extend the argument of the normal state cp on B (K) to any normal
compound state on B (3.1 @ K) . Let w be a normal compound state on
B ( X @ K )
is described by an amplitude operator V from an another Hilbert space 3 to 3-1 K, satisfying the condition
VV* E B ('H @ K) , trwBKVV* = 1.
F can always be taken as 3 2 (3.1 @ K ) e = Ee (3.1 @ K) for 6 = VV* with
an anti-unitary operator J on eigen-representation of V*V. Then we can
define an entangling operator H : 31 H F@ K uniquely by f iU = V , where
(C @ rll fit[) = (Jt@ 'I1 H I J C ) , VC E 3.1, rl E K , c E F7 (6)
up to a unitary transformation U of 9. We have the following theorem.
Theorem 2.1. The normal compound state w in (5) can be achieved as an entanglement
w ( A @ B) = t r w x H * ( I @ B) H = t rFBKHxH* ( I @ B ) (7) h-
with its marginal states p = H * H on B (3.1) and 0 = t r 3 H H * on B ( K ) .
103
Note that the lemma 2.1 can be obtained as a special case 3 = C in
the theorem 2.1.
Now let us see how the entangling operator can be used to clas-
sify the entangled states. We introduce the mapping 4* from B ('FI) to the predual B (K), by 4* (A) f trx ( A 81) f3 = t r ~ H x H * for any
A E B (X) and its dual mapping 4 from B (lc) to the predual B ('FI)* by
4 (B ) = trK: ( I 8 B) f3 = H* ( I 8 B) H for any B E B (K). Then the normal
compound state w is written as
-
w ( A 8 B ) = trxA4 ( B ) = trK:B4* ( A ) . (8)
The map B (E B (K)) - 4 ( B ) :
4 ( B ) = H* ( I 8 B) H (E B ('FI)*)
is the complete positive map (CP for short) written in the Steinspring form,
and the map A (E B ('FI)) - $* x * ( 1 . 4* (x) = c (@I 8 0 HAH* (Ik) @ I > (E B (K)*)
k
is also CP written in the Kraus-Sudarshan form. Both maps 4 and $*
are positive, but they are not necessarily CP, unless B(K) = B(K) or -
104
Any other choice of V is unitary equivalent to H in the minimal sapce
the eigen sepresentation of VV and the marginal P, and be a CONS
Thus
B (3-1) = B (H) (i.e., B (X) or B ('FI) is Abelian). Both 4 and #* are called
complete co-positive.
We define the true quantum entanglement [14, 151.
Definition 2.1. The dual map q5* : B(7-l) + B(K) * of a complete co-
positive map 4 : B (K) --f B ('If)*, normalized as tr7-14 ( I ) = 1, is called
the (generalized) entanglement of the state p z q5 ( I ) on B (3-1) to the state
E $* ( I ) on B (K) .The entanglement 4* is called true quantum if it is
not complete positive.
Let { I f n ) } and {le,)} be a CONS in 3 and H corresponding to
the eigen-representation of VV* and the marginal p. In these eigen-
representation the entangling operator H = V becomes H = C IHn) (en\ =
c I f k 8 h k (n)) (en\, thus vc I v k ) ( f k l = [en 8 hk (n)) ( f k l , where the k ,n k k , n vectors Ihk (n)) E K: is defined by (en 8 .I V I f k ) . Note that the eigen-basis
of p is characterized by weak orthogonal condition
- n
tr?=@K IHm) (Hnl = And: = (HmIHn) (9)
where An are the eigenvalues of p. We summarize some notations for the sequel use and introduce a clas-
sification of quantum compound states. An entangled state w with its
marginal states p and IS is expressed by a density operator 0 on H@X; that
is, w (-) = tr .8, and 8 is written by the following forms due to the strength
of the correlation between two marginal states in eigenrepresentation of p.
(1) q-entang1ement:We denote true quantum entanglement by 4: (i.e., 4: is not CP) and q-compound state by 0;
with weak orthogonal condition
The set of all true quantum entanglement by EQ.
105
(2) d-entanglement: Let denote d-entanglement by #J$ and its com-
pound state by e$. Then
n
n
with strong orthogonal condition
The set of all d-entanglement by E d .
(3) c-entanglement : The entanglement #J* is called c-entanglement if
it has the same form as d-entanglement, but {on} are commutative.
We denote c-entanglement by 4; and its compound state by 0;. €, is the set of all c-entanglements.
It is clear that &d and €, are belonged to the set of all not true quantum
states. However there exist several important applications with quantum
correlated state written as d-entanglement, such as quantum measurement
and filtering, quantum compound state, and lifting. So that it is useful to
classify the quantum separable state and the classical one.
We show the necessary condition for separability. The state 0 is written
as the convex combinations
n n
of tensor products of pure or mixed densities pn E B (X)* and un E B (K), , then
n n
which are given as the convex combinations of maps A H untrxApn and
B H pntr&on. Such maps $* and #J are not only complete co-positive
but also CP as it follows from the positive-definiteness of operator-matrics
[A:Aj] , VAi E B (X) ,
So we have the following theorem.
106
Theorem 2.2. If a density operator I3 of a normal compound state w is
separable, then both q$ and q$* are CP.
The sufficient conditions in Theorem 2.2 will be discussed in a preparing
paper [17].
3. Degree of entanglement via quantum mutual entropy
In this section we review the classification of states by the information
degree of entanglement in [14, 151 and apply it to entangled PPT states in
the C3 @ C3 model of Horodeckis[7].
3.1. Characterization of a pure state by degree of entanglement
Entanglement degree for mixed states has been studied by some entropic
measures such as quantum relative entropy and quantum mutual entropy.
As an example of such a degree was defined in [18] by the relative entropy
S(I3,e0) rtr8(log13-log00) as
D (8) = min {S (8 ,Bo) ; 80 E D} , (10)
where D is the set of all separable states. Since this measure has to take
a minimum over all disentangled state, it is difficult to compute it analyt-
ically. Thus another degree of entanglement was introduced by Belavkin
and Ohya[l4, 151.
Definition 3.1. Let w be the entangled state of p and a.Let q$ be the
entanglement associated with w,and 86 is the density operator for w.
(1) The quantum mutual entropy of p and a w.r.t $ is defined by
(2) The q-entropy of a is defined by 3 (a) = sup {I+ ( p , a) ; q6* (I) = a}.
(3) D (q$;p,o) = {S (p ) + S (a)} - I6 (p , a) is called the degree of entanglement( DEM for short). 41 has stronger entanglement than
4 2 iff
Id ( p , 0) = trod (log Oc - logp @ 0).
(4) A compound state is said to be essentially entangled if D (4; p1 a) < 0.
107
If the subalgebra A of B(IC) is abelian and u is a normal state on
A, then s (u ) is equal to von Neumann entropy = - t ra loga. Moreover if
dimIC < +oo, then
s ( u ) 5 logdimd, S(a) I IogdimK.
The above D (4; p, u) can be negative. If 8 on H @ IC is entangled pure
state with the marginal states p, u, then von Neumann entropy S (8) = 0.
Moreover, from the Araki-Lieb inequality:
1s ( P ) - s (0) I I s (8) I s (P) + s (4 (11)
we have S ( p ) = S (u). It follows
lo ( p , G) = tr8 (log 8 - log p @ u)
= s ( p ) + s(G) - s(e) = 2s (P>
That is, for entangled pure state, the q-entropy becomes twice of von Neu-
mann (reduced) entropy [16].
Theorem 3.1. For a pure state w with the marginal state p and u, (1) w i s separable ifl D (4 ; p , u) = 0, and (2) w is inseparable (entangled) ifl D (4 ; p, u) = - f r { S ( p ) + S (u)} < 0. A pure inseparable state can be called as an essential entangled state.
3.2. Degree of entanglement for entangled PPT states
The positive partial transpose criterion (PPT criterion for short) proposed
by Peres [19] which is a necessary condition of separability but not a suffi-
cient one generally.
The IC-partial transpose operation of a compound density operator 8 on
H @ K: is denoted by BTK such that
8 - P K = ( I @ T ) 8 ,
where dim 'FI @ K: is finite and T is the transpose operation on B (IC). For
example, 8 is decomposable as 8 = C,,, [em) (e,l €3 B,,, where {le,)} is
a standard base in H and B,, E B (K). Then
e - e T K = C le,) (en[ @ B:,. m,n
108
Definition 3.2. A compound density operator 8 on ?-I @ K is called a PPT state if BTK is positive. Non PPT state is simply is called NPT state.
It is easy to show that the PPT condition is a necessary condition
for a separable density operator. If a state 8 is separable and written as
X k p k 8 crk, then eTK = x k X k p k 8 u; is positive because cr; is positive. However, in general, it is known that the converse statement does not hold.
In the low-dimensional case (C2 @I C2 and C2 63 C3) Horodeckis showed
that the PPT criterion gives a necessary and sufficient condition of separa-
ble states [ S ] . They also introduced the following state w, (.) = tr 8, as
an example of the entangled PPT states on C3 63 C3[7]: a 5 - a e-, 2 1 a 1 5 e - - IS+) (@+I + ?e+ + - 2
, - 7 7 where
1 la+) = - (1.1 63 el) + le2 63 e2) + le3 @I 4) , v5
1 8, = 2 (1.1) (ell 63 1.2) (e21 1.2) (e21 8 (e3) (e3( + le3) (e3( 8 1.1) (ell),
8- = - 3 ( I 4 (e218 1.1) (ell + 1 % ) ( ~ 1 8 1.2) ( ~ 1 + 1.1) (ell 8 1 % ) (e31), 1
{ 1.1) ,Ie2) , 1-23)) is a standard base in C3. The operator 0, has the following classification:
(1) 8, is a separable state for 2 5 a 5 3. (2) 8, is entangled but a PPT state for 3 < a 1 4. (3) 0, is a NPT state for 4 < IY I 5.
We give another classification of 8, by means of the DEM [16]. First
we remind the following theory [20].
Theorem 3.2. For a density operator p given as the convex combination
n n
of densities p n E B (‘H)*, the following inequality holds:
n n
The equality holds if pn I pm f o r n # m.
109
The decomposition of 0, can be regarded as a convex combination of
orthogonal states with the marginal states p and a given as p = (T =
5 ( le i ) ( e l l + lez) (ezl + le3) (%I). Then 1
1 D (6,; P, 0) = 5 {S (PI + s(a)) - Ie, (P , a)
1 = s (6,) - 5 {S (PI + s (4)
2 a a 5 - a 5 - a
7 7 7 7 7 (13) - - -- log3 - - log - - - log -
We obtain the change of the value D (Oa; p ,a ) w.r.t. a as shown in
Figure. I.
N P T '
D (Oa; p, 0) and classification of 0- Figure 1.
In Fig.1 we observe that the value a0 satisfying D ( O f f 0 ; p, a) = 0 is in
between 3 and 4. Then we conclude an alternative classification of 8, as
follows:
(1) 0, is not an essential entangled state for 2 _< a 5 ao. (2) 8, is an essential entangled state for a0 < a 5 5.
The strength of entanglement can be read the change of the value
This degree can be used to find several models showing different types
D (6,; p, a), that is the curve of the value.
of entanglement which will be discussed in [17].
110
References
1. R. F. Werner, Phys. Rev., A 40, 4277 (1989). 2. C. H. Bennett, G. Brassard, C. Crepeau, R. Joza, A. Peres, W. K. Wootters
Phys. Rev. Lett, 70, 1895 (1993). 3. C. H. Bennett, G. Brassard, S. Popescu, B. Schumacher, A. J. Smolin, W.
K. Wootters Phys. Rev. Lett, 76, 722 (1996). 4. A. Ekert, Phys. Rev. Lett., 67, 661 (1990). 5. R. Joza, B. Schumacher, J. Mod. Opt., 41, 2343 (1994). 6 . M. Horodecki, P. Horodecki, R. Horodecki, Phys. Lett., A 223, 1 (1996). 7. M. Horodecki, P. Horodecki, R. Horodecki, ”Mized-State entanglement and
quantum communication” in Quantum Information, Springer Tracts. in Mod- ern Physics 173 (2001)
8. B. Schumacher, Phys. Rev., A 51, 2614 (1993); Phys. Rev., A 51, 2738 (1993).
9. V. P. Belavkin, Radio Eng. Electron Phys, 25, 1445 (1980). 10. V. P. Belavkin, Found. Phys., 24, 685 (1994). 11. M. Ohya, IEEE Info. Theory, 29, 770 (1983). 12. M. Ohya, Nuovo Cimento., 38, 402 (1983). 13. L. Accardi, M. Ohya, J . Appl. Math. Optim., 39, 33 (1999). 14. V. P. Belavkin, M. Ohya, Infinite Dimensional Analysis, Quantum Probability
and Related Topics, 4, No.2, 137 (2001). 15. V. P. Belavkin, M. Ohya, Proc. R. SOC. Lond., A 458, 209 (2002). 16. T. Matsuoka, M. Ohya, to appear in Proc. of Foundations of Probability and
Physics, AIP Proceedings. 17. A. Jamiolkowski, T. Matsuoka, M. Ohya, in preparation. 18. L. Henderson, V. Vendal, Phys. Rev. Lett., 84, 2014 (2000). 19. A. Peres, Phys. Rev. Lett., 77, 1413 (1996). 20. M. Ohya, V. D. Petz, Quantum Entropy and Its Use., Springer (1993).
111
A white noise approach to fractional Brownian
motion
David Nualart Universitat de Barcelona
Facultat de Matembtiques Gran Via 585, 08007 Barcelona, Spain
Abstract
We show that the derivative in time of a class of Volterra processes, that includes the fractional Brownian motion, is a Hida distribution. We review some facts about the divergence integral with respect to the frac- tional Brownian motion and we interpret this integral form the point of
view of white noise analysis.
1 Introduction The fractional Brownian motion (fl3m) is a centered Gaussian process B H = { BF, t 2 0 ) with the covariance function
(1) 1 2
The parameter H E (0 , l ) is called the Hurst parameter. The fractional Brown-
ian motion has the following self-similar property: For any constant a > 0, the processes { a - N B z , t 2 0 } and { B y , t 2 0} have the same distribution.
From (1) we can deduce the following expression for the variance of the increment of the process in an interval Is, t]:
RH ( t i S) = E(B,HB,") = - ( s ~ ~ + t2H - It - ~ 1 ~ ~ ) .
E (IB; - Bf I2 ) = It - ~ 1 (2)
This implies that fBm has stationary increments. Furthermore, by Kolmogorov's continuity criterion, we deduce that fBm has a version with a-Holder continuous trajectories, for any a > H .
For H = i, the process BY is an ordinary Brownian motion. However, for H # i, the process BH does not have independent increments, and, further- more, it is not a semimartingale. Let r(n) := E [BB(B:+l - B:)]. Then, r(n) behaves as Cn2H-2, as n tends to infinity (long-memory process). In particular, if H > i, then En IT-(.)\ = 00 (long-range dependence) and if H < i, then, En Ir(n)l < 00 (short-range dependence).
112
The self-similarity and long memory properties make the fractional Brow-
nian motion a suitable input noise in a variety of models. Recently, fBm has been applied in connection with financial time series, hydrology and telecommu- nications. In order to develop these applications there is a need for a stochastic calculus with respect to the Bm. Because the fBm is neither a semimartingale nor a Markov process if H # i, new tools are required in order to develop a stochastic calculus.
In this note we first construct the fBm in the white noise space and show that its derivative in time is a distribution in the sense of Hida. More generally, in Section 3 we introduce a class of Volterra processes that include fl3m and show that their time derivatives are Hida distributions. Section 5 reviews the
approach to stochastic calculus for the fBm based on the Malliavin calculus, and in Section 6 we establish its relation with white noise analysis. We refer to Bender [4] and to Biagini et al. [5] for related works on this subject.
2 White noise analysis and Hida distributions In this section we present some preliminaries on white noise analysis. We refer to [9] and [13] for complete expositions of these notions. Let (O,F,P) be the white noise space. That is, R is the space of tempered distributions S’(R), 7 is the Bore1 0-field (with respect to the strong topology of S’(R)) and P is the Gaussian probability measure determined by
1 E(exp(i(w,E)) = exP ( -5 Itell:) 9
for any rapidly decreasing function E E S(R), where 1 1 . 1 1 2 denotes the norm in L2(W). The pairing (w,Q can be extended using the norm of L 2 ( 0 ) to any function E 6 L2(R). In particular, the process Wt = ( . , lp t l ) is a Brownian motion.
The trajectories of the Brownian motion Wt are nowhere differentiable al- most surely. The theory of generalized functions introduced by Hida allows to give a rigorous meaning to the derivative %. To this aim we briefly recall the main aspects of this theory.
Set (L2) = L2(R, F, P). Any random variable F E (L2) can be developed into a series of multiple stochastic integrals:
n=O
where fn E L2(Rn) is a symmetric square integrable kernel. Consider the oper-
ator A = -& + x 2 + 1 and define its second quantization as
n=O
113
Notice that the operators A and r ( A ) are densely defined in L2(R) and ( L 2 ) , respectively, and they are invertible and the inverse operators are bounded. For any p E R and F in the domain of I'(A)p, define
llFllp := (E [ ( I ' (A )pF)2 ] )1 /2 .
If p 2 0, we denote by (S), the space of random variables F E ( L 2 ) such that llF/lp < 00 equipped with the norm I l . l l p . If p < 0, we denote by ( S ) - p the
completion of ( L 2 ) with respect to the norm 11.11,. The projective limit of the spaces (S),, p 2 0, is called the space of test functions and is denoted by (S). The inductive limit of the spaces (S)-p, p 2 0, is called the space of Hida distributions and is denoted by (S)*.
Consider the Hermite functions defined for any n 2 0 as
E ~ ( ~ ) = n-1/4(2nn!)-1/2e-f22~n(2), (3)
where Hn(s) is the nth Hermite polynomial, Hn(z) = (-l)ner2&e--22. The Hermite functions form an orthonormal basis of L2(B), and A& = (2n+2)En for
any n 2 0 , because f k = f l C n - 1 - @&+I (with the convention (-1 = 0).
Moreover, there exists a constant K > 0 such that
As a consequence of this estimate one can show that the derivative of the Brownian motion W'(t) belongs to (S)-, for any p > A, and W ( t ) = (., &). The pairing (., bt) can be interpreted as the Wiener integral of the distribution bt, that is, I1(&).
3 Derivative of a Volterra process
Consider a Volterra process of the form X t = Ji K( t , s)dW,, where K( t , s) is a square integrable kernel such that K(t , s) = 0 if t < s, defined in the white noise probability space. That is, we assume that for any t 2 0
1' K ( t , s)2ds < 00.
In this section we will show that the derivative X'( t ) is an Hida distribution, under very general conditions on the kernel K( t , s).
The covariance function of this process is
t A s
R(t, S) = E ( X , X , ) = 1 K(t, r )K(s,r)dr.
We denote by &T the set of step functions on [0, TI. The space &m is the union of all ET. Let 'HT be the Hilbert space defined as the closure of &T with respect
to the scalar product
( I [ O , t ] , 1[0,4),T = R(t, s).
114
The mapping l ~ ~ , ~ ] -+ X t can be extended to an isometry between ZT and the Gaussian space H T ( X ) associated with { X t , t E [O,T]}. We will denote
this isometry by 'p --t X( 'p ) . The random variable X( 'p ) can be interpreted
as the Wiener integral of 'p with respect to the Gaussian process X, that is,
X(p) = so 'p tdXt, provided 'p can be identified as a function. Consider the mapping K- : & --+ L2(0 ,T) defined by K-l[0,~1 = K(t, . ) .
This mapping can be extended to a linear isometry between 'FtT and L2(0 ,T) because
T
The operator K- allows to write the following transfer rule:
B(P) = W(K-'p), (5)
for any 'p E ZT. We will make the following hypotheses on the kernel K ( t , s):
H1) K( t , s) is continuously differentiable on (0 < s < t < oo}, and its partial derivatives verify the following integrability condition:
for any 0 < E < T < 03. Moreover, t + StAb g(t, s ) ( t A b - s V a)+& is
continuous on (0, oo), for all 0 5 a < b.
H2) The function k ( t ) = Jot K( t , s)ds is continuously differentiable on (0, 03).
Then, for any step function 'p E &T, we can write deduce the following expression for the operator K-
In fact, it suffices to check (6) for an indicator function l[o,,], where a I T , and in this case, the right-hand side of (6) clearly vanishes for t > a, and for t < a yields
K(T, t ) - lT %(., t)dr = K(a, t ) = (K-l[o,a]) (t) .
As a consequence, Hypothesis H1) implies that the set of continuously differen-
tiable functions C1([O,T]) is contained in ZT, and for any 'p E ZT, we have
115
We will denote by K+ the adjoint of the operator K- in L2(O,0o). That is,
for any step function ‘p E &,
Hence,
An alternative expression for the operator K+ is as follows
In fact, if cp = l[0,~1, the right-hand side of (8) is
k’(t) l [o,al(t) + g ( t , r ) [ l [ o , a ] ( r ) - ~ [ o , a l ( t ) ] dr.
For t > a this is s,” g(t, r )d r = 6 s,” K( t , r ) dr = (K+ lpa ] ) ( t ) , and for t < a we obtain
t
K( t , S ) ~ S = (K+lp+1) (t) .
As a consequence, the space of continuously differentiable functions which are bounded together with its derivative Ci (0, co) is included in the domain of K+ and
(9) dK
I(K+cp) (t)l I IIcpllm + l I c p / l L J l+-)l (t - r )dr . 0
The following result provides an estimate for K+&, where En are the Hermite functions defined in (3).
Lemma 1 For any n 2 0, K+&, is continuous on (0, co) and
I(K+En) (t)I 5 ctn5/12, (10)
where Ct is bounded on [&,TI, for all 0 < E < T < 00.
Proof. From (8) we have
dK
Clearly, k’(t)En(t) is continuous on (0,co) by hypothesis H2). On the other hand, we can write
116
and this is a continuous function of t by Hypothesis Hl), because it can be approximated uniformly on compacts of (0, m) by continuous functions. The estimates (9), (4) and Hypothesis H1) yield
aK I(K+tn) ( t ) ~ 5 ~k’(t)~ Iltnllm + IIGII~J 0 I z ( t i r ) l ( t - r ~
5 ctn5fI2,
where Ct is bounded on [E, TI, for all 0 < E < T < co. rn The above lemma allows us to prove the following result.
Proposition 2 K(t, .) = K-lp,t] is differentiable from t E (0, co) into S’(R), and
Proof. Using the orthonormal basis of L2(R) formed by the Hermite func- tions we obtain
Hence,
Using the estimate (10) we can bound each term of the above series by (2n + 2)-2n5f6 Supt+<t+h C,” which is convergent. On the other hand, the continuity of K+tn implies that (11) tends to zero as n tends to infinity. rn
The preceding proposition together with the following result (see [4]) will allow as to compute the derivative of the Volterra process X t .
Proposition 3 Let F : (0,m) + S’(R) be a differentiable function. 11 (F ( t ) ) = (., F( t ) ) is a diflerentiable stochastic distribution process and
Then,
[I1 (F(t))l’ = 11(F’@)).
Combining propositions 2 and 3 we see that X t is differentiable a t any t > 0, and
X’( t ) = I l (& 0 K+),
117
where St o K+ is the distribution defined by (& o K+, f) = (K+f) (t), for any
f E S’(R). In fact,
m M d
&OK+ = C ( & o K + , E n ) E n = x (K+En)(t)E,=dtK-l[o,t] n=O n=O
We can also write formally
X’( t ) = (& o K+),W,’ds = (K+W’) (t). (12) 1- Example Consider the Volterra process X associated with the kernel K ( t , s ) = I ’ (H+i)- l ( t -s)H-$, where H E (0,l) . This process is related to the fractional
Brownian motion with Hurst parameter H . Clearly, the kernel K( t , s) verifies
properties H1) and H2). The operator K defined by (Kp) (t) = K ( t , s)cp(s)ds
coincides with the fractional integral operator I z ’ p . As a consequence, from
(7) we obtain
Therefore,. the derivative of the Volterra process X is given by
that is, the derivative of this process is expressed as the fractional integral if H > $ (or fractional derivative if H < k) of the white noise.
4 Fractional Brownian motion The fE3m is a Volterra process (see, for instance, [8])
B? = h’ KH (t , s ) d w ~
associated with the kernel KH(t, s) given by
verifies
-(t, ~ K H s) = CH(H - -)(t 1 - S ) ~ - Z (I> 4 - H , at 2
118
shere eg ius the normalizing constat c
and, as a consequence, Hypotheses H1) and H2) are satisfied.
integrals: If H > i, the operator K- on 'HT can be expressed in terms of fractional
In this case, the scalar product of 'HT has the simpler expression
where CYH = H(2H - l), and ZT contains the Banach space IHT~ of measurable
functions 'p on [O,T] such that
We have the following continuous embeddings (see [14]):
LB(o,T)) c I'HTI c ZT.
For H < 3, the operator K- on 'HT can be expressed in terms of fractional derivatives:
(16) (K-'p) ( S ) = CHr(H + 2 ) S $ - H 1 ( D * I H U H - " p ( U ) ) (S).
In this case, 'HT = I$ IH(L2) (see 181) and
y-Holder continuous, provided y > f - H . 'HT contains functions which are
Using the fractional integration by parts formula (see [IS]) we obtain
H - + u T - ~ ~ - - ' 1 CHr(H - ; )SH-+ (Io+ z 'p(u)) (s) if H > f C H r ( H + +H-++-H 'p(u)) ( s ) if H < f .
Hence, we obtain the following formula for the fractional white noise:
5 Stochastic calculus with respect to the fBm
Suppose now that u = {ut, t E [0, TI} is a random process. By the transfer rule
( 5 ) we can write
119
However, even if the process u is adapted to the filtration generated by the fBm (which coincides with the filtration generated by W ) , the process K-u is no longer adapted because from (14) and (16) we deduce that the operator K- does not preserves the adaptability. Therefore, in order to define stochastic integrals of random processes with respect to the fBm we need anticipating integrals.
In the case of an ordinary Brownian motion, the divergence operator co- incides with an extension of It6's stochastic integral to anticipating processes introduced by Skorohod in [17] and Hitsuda [lo], [ll]. Thus, we could use this
anticipating integral in formula (17), and in that case, the integral JTutdBF coincides with the divergence operator in the Malliavin calculus with respect to the a m BH. The approach of Malliavin calculus to define stochastic integrals with respect to the fl3m has been introduced by Decreusefont and Ustunel in [8], and further developed by several authors (Carmona and Coutin [6], Albs, Mazet and Nualart [2], Albs and Nualart [3], Albs, Le6n and Nualart [l], and Hu [12]).
5.1 Stochastic calculus of variations with respect to fESm
We review here some basic facts of the Malliavin calculus for the fJ3m B H . Let S be the set of smooth and cylindrical random variables of the form
F = f ( ~ ~ ( 4 1 ) , . . . , B ~ ( 4 n ) ) , (18)
where n 2 1, f E CF (Rn) (f and all its partial derivatives are bounded), and
The derivative operator D of a smooth and cylindrical random variable F d'i E XT.
of the form (18) is defined as the XT-valued random variable
The derivative operator D is then a closable operator from L2(fl) into L2(R; XT). We denote by D'v2 is the closure of S with respect to the norm
The divergence operator b is the adjoint of the derivative operator. That
is, we say that a random variable in L2(f12;X~) belongs to the domain of the divergence operator, denoted by Dom 6, if
for any F E S. In this case 6(u) is defined by the duality relationship
120
for any F E D’i2. We will denote 6(u) by ut6B2. We have D1v2(’H~) CDom 6 and for any u E D’~2( ’H~)
E (W2) = E ( llull;T) + E ( ( D % 1 (20)
where (Du)* is the adjoint of (Du) in the Hilbert space ‘HT @ HT.
5.2
The following result (see [3]) provides a relationship between the divergence operator and the symmetric stochastic integral introduced by Russo and Vallois
in [15].
Proposition 4 Let u = {ut, t E [0, TI} be a stochastic process in the space D’v2(3-I~). Suppose that
Divergence and symmetric integrals for H >
E (11~11;x,1 + llD~ll;xTlOIHT,) <
and
(21)
T Then the symmetric integral so u t d B 2 , defined as the limit in probability as E
tends to zero of T
H (2El-l J U S ( B ~ + + T - B ( ~ - ~ ) ~ ~ w ,
0
exists and we have
Remark The symmetric integral can be replaced by the forward or backward integrals in the above proposition.
5.3 ItB’s formula for the divergence integral for H > If F is a function of class C2, and H > 3, the path-wise Riemann-Stieltjes
integral F ’ ( B f ) d B f exists for each t E [0, TI by the theory of Young [18]. Moreover the following change of variables formula holds:
F ( B F ) = F(0) + F ’ ( B 7 ) d B y . l Suppose that F is a function of class C2(R) such that
121
where c and X are positive constants such that X < &. Then, the process
F’(Bp) satisfies the conditions of Proposition 4. As a consequence, we obtain
F‘(Bf)SBf + H F ” ( B f ) ~ ~ ~ - ’ d s . (25) = I ’ 6’ Therefore, putting together (23) and (25) we deduce the following It8’s formula
for the divergence process
A more general version of It8’s formula has been proved in [3]
5.4
The extension of the previous results to the case H < is not trivial and new
difficulties appear. For instance, the forward integral BpdBp in the sense of
Russo and Vallois does not exists, and one is forced to use symmetric integrals.
A counterpart of Proposition 4 in the case H < f and It6’s formula (26) has
been proved in [l] for < H < a. The reason for the restriction 4 < H is
the following. In order to define the divergence integral s:F’(BF)6BF, we
need the process F’(BB) to belong to L2(R; ‘ F I T ) . This is clearly true, provided F satisfies the growth condition (24), because F’(B:) is Holder continuous of
order H - E > f - H if c < 2H - f . If H 5 a, one can show (see [7]) that
Stochastic integration with respect to fBm for H <
P(BH E ‘ F I T ) = 0,
and the space D1,2(‘FI~) is too small to contain processes of the form F’(BF). In [7] a new approach is introduced in order to extend the domain of the
divergence operator to processes whose trajectories are not necessarily in the
space ‘ F IT . The basic tool for this extension of the divergence operator is the
adjoint of the operator K- in L2(0,T) that we have denoted by K+. Set
7f2 = (K-)-’ (K+)-’ (L2(0, T ) ) and denote by Sx, the space of smooth and cylindrical random variables of the form
F = f(BH(41), . 1 BH(4n)) , (27)
where n 2 1, f E Cp (Eln), and q5i E ‘FI2.
Definition 5 Let u= {ut , t E [0, TI} be a measurable process such that Es, uTdt < 00. We say that u E Dom*b if there exists a random variable 6(u) E L 2 ( 0 ) such that for all F E Sx, we have
T
s, E(u,K+K-D,F)dt = E(6(U)F) .
122
This extended domain of the divergence operator satisfies the following ele-
mentary properties:
1. Domb c Dom*b, and 6 restricted to Domb coincides with the divergence
operator.
2. If u E Dom*b then E(u) belongs to XT.
3. If u is a deterministic process, then u E Dom*b if and only if u 6 Xi-.
This extended domain of the divergence operator leads to the following ver- sion of It8’s formula for the divergence process, established by Cheridito and
Nualart in [7].
Theorem 6 Suppose that F is a function of class C2(R) satisfying the growth condition (24). Then for all t E [O,T], the process {F’(B$’)lro,tl(s)} belongs to Dom*6 and we have
6 White noise analysis and divergence integrals
In this section we will introduce a general definition of the anticipating integral
using the Wick product and the approach of white noise analysis (see [5] and [13]) and we will show that it includes the divergence integral studied in Section
5. A fundamental tool in white noise analysis is the S-transform. For any
generalized random variable F E (S)* , the S-transform is defined by
S F ( ( ) = ( F , e w ( ~ ) - ~ ~ ~ ~ ~ ~ ~ ) , 6 E S(R)
which is well defined because e w ( ~ ) - * l ~ ~ l l ~ E (S). properties of the S-transform.
Here are some well-known
(i) The S-transform S F characterizes the generalized random variable F.
(ii) For F,G E (S)* there exists a unique element F o G E (S)* such that for
all 6 E S(R), S ( F o G ) ( t ) = SF( ( )SG( ( ) . This element is called the Wick
product of F and G.
(iii) Let a stochastic distribution process X : [0, TI -+ (S)* be differentiable.
Then S(Xl)(O = ( S X ( 0 ) : .
If X is a Volterra process satisfying conditions H1) and H2), then S X t ( 6 ) =
(6, K( t , .))z and SXl (6) = (K+[) ( t ) , where K+ is given in (7). We say a stochastic distribution process X : [O,T] + (S)* is integrable
if for all 6 E S(R), S X ( 6 ) is measurable, S X ( [ ) E L1(O,T) and J:SXt(E)dt
is the S-transform of a Hida distribution that will be denoted by SoTXtdt. A
123
sufficient condition for the integrability of a stochastic distribution process is the fact that for all ( E S(R), S X ( ( ) is measurable and
ISXt(0I I W ) exp (; IIAPFII;) (294
for some p E M and L E L1(O,T).
distribution processes such that X is differentiable and the integral
Suppose that u = {ut , t E [O,T]} and X = {Xt , t E [O,T]} are stochastic
T 1 ut OX;&
exists. Then, we say that u is Wick integrable with respect to X . A particular case of this integral has already appeared in (12). The next proposition provides an example of a Wick integral with respect to a Volterra process.
Proposition 7 Suppose that X t = K( t , s)dW, is a Volterraprocess such that the kernel K( t , s) satisfies Hypotheses H1) and H2). Assume that the function
Ct = Ik'(t)l + l$(t,r)I (t - r )dr 0
belongs to L'(0,T). Let u = {ut,t E [O,T]} be a stochastic process such that EJ:u:dt < 00. Then u is Wick integrable with respect to X .
Proof. It suffices to show that Sut (K+E), satisfies (29a). But
1sutl = E utew(O-iIIcIIi) I 5 ( ~ u : ) f . eIIEIIz,
and on the other hand (K+E), can be bounded on any b i t e interval [O,T] by
I ( (9):
As a consequence,
Then, sc (Eu:)' Ctdt I 4- < 00, and
n = O
provided p > $, and (29a) holds for p > g. w On the other hand, the next proposition shows that the Wick integral (30)
is an extension of the divergence integral we have introduced before using the techniques of Malliavin calculus.
124
Proposition 8 Suppose first that H > f , then any process u in the domain of
the divergence is Skorohod integrable and
Also, if H < i, any process in the extended domain is Wick integrable and (31) holds.
Proof. The fact that u is Wick integrable follows from Proposition 7. On the other hand, for any E E S(W), taking into account that OW(<) = (K-)-' [, and using the duality relationship (19) we obtain
References
[l] Albs, E., Le6n, J. A. and Nualart, D. Stratonovich stochastic calculus with respect to fractional Brownian motion with Hurst parameter less than 1/2. Taiwanese Journal of Mathematics 5 (2001) 609-632.
[2] Albs, E., Mazet, 0. and Nualart, D. Stochastic calculus with respect to
[3] Albs, E. and Nualart, D. Stochastic integration with respect to the frac- tional Brownian motion. Stochastics and Stochastics Reports 75 (2003) 129- 152.
Gaussian processes. Annals of Probability 29 (2001) 766-801
[4] Bender, C. An It6 formula for generalized functionals of a fractional Brow- nian motion with arbitrary Hurst parameter. Stochastic Process. Appl. 104 (2003) 81-106.
[5] Biagini, F., (aksendal, B., Sulem, A. and Wallner, N. An introduction to white noise theory and Malliavin calculus for fractional Brownian motion. Preprint.
[6] Carmona, P. and Coutin, L. Stochastic integration with respect to frac- tional Brownian motion. Ann. Institut Henri Poincare' 39 (2003) 27-68.
125
which inplies the desired results because
[7] Cheridito, P. and Nualart, D. Stochastic integral of divergence type with respect to the fractional Brownian motion with Hurst parameter H < i. Ann. Institut Henri Poincark To appear.
[8] Decreusefond, L. and Ustiinel, A. S. Stochastic analysis of the fractional
Brownian motion. Potential Analysis 10 (1998), 177-214.
[9] Hida, T., Kuo, H. H., Potthoff, J. and Streit, L. White noise. A n infinite- dimensional calculus. Mathematics and its Applications, 253. Kluwer Aca- demic Publishers Group, Dordrecht, 1993.
[lo] Hitsuda, M. Formula for Brownian partial derivatives. Second Japan-USSR Symposium on Probability Theory, Kyoto, 1972, 111-114.
[ll] Hitsuda, M. Formula for Brownian partial derivatives. Proceedings of Fac- ulty of Integrated Arts and Sciences. Hiroshima University 111-4 (1978) 1-15.
[12] Hu, Y. Integral transformations and anticipative calculus for fractional Brownian motions. Reprint.
[13] Kuo, H. H. White noise distribution theory. Probability and Stochastics Series. CRC Press, Boca Raton, FL, 1996.
[14] Pipiras, V. and Taqqu, M. S. Are classes of deterministic integrands for fractional Brownian motion on a interval complete? Bernoulli 7 (2001) 873-897.
[15] Russo, F. and Vallois, P. Forward, backward and symmetric stochastic integration. Probab. Theory Rel. Fields 97 (1993) 403-421.
[16] Samko S.G., Kilbas A.A. and Marichev 0.1. Fractional Integrals and Derivatives. Theory and Applications, Gordon and Breach, 1993.
[17] Skorohod, A. V. On a generalization of a stochastic integral. Theory Probab. Appl. 20 (1975) 219-233.
[18] Young, L. C. An inequality of the Holder type connected with Stieltjes integration. Acta Math. 67 (1936) 251-282.
126
Adaptive Dynamics in Quantum Information
and Chaos
Masanori Ohya Department of Information Sciences, Tokyo University of Science,
Noda City, Chiba 278-8510, Japan
Abstract
The concept of adaptive dynamics is explained and its application to quantum information and chaos is discussed. In particular, we explain how it is used to understand chaos.
1 Introduction
There exist several reports saying that one can observe chaos in nature, which
are very much related to how one observes the phenomena in specified condi-
tions, for instance, scale, direction. It has been difficult to find a satisfactory
theory (mathematics) to explain such chaotic phenomena. An idea describing
chaos of a phenomenon is to find some divergence of orbits produced by the dy-
namics explaining the phenomenon. However to explain such divergence from
the differential equation of motion describing the dynamics is often difficult, so
that one takes (makes) a difference equation from that differential equation, for
which one has to take a certain time interval r between two steps of dynamics,
that is, one needs a processing discretizing time for observing the chaos. In lab-
oratory, any observation is done in finite size for both time and space, however
one believes that natural phenomena do not depend on these sizes how small
they are, so that most of mathematics (theory) has been made as free from the
sizes taken in laboratory. Therefore mathematical terminologies such as "lim",
"supll, "inf" are very often used to define some quantities measuring chaos, and
many phenomena showing chaos have been remained unexplained.
In the paper with Kossakowski and Togawa [8], we took the opposite position,
that is, any observation will be unrelated or even contradicted to such limits.
Observation of chaos is a result due to taking suitable scales of, for example,
time, distance or domain, and it will not be possible in the limiting cases.
In other words, it is very natural to consider that observation itself plays a
similar role of "noesis" of Husserl and the mode of its existence is a "being- for-itself", that is, observation itself can not exist as it is but it exists only
127
through the results (phenomena) of objects obtained by it. Phenomena can not
be phenomena without observing them, so to explain the phenomena like chaos it is necessary to find a dynamics with observation.
We claimed that most of chaos are scale-dependent phenomena, so the defi- nition of a degree measuring chaos should depends on certain scales taken and more generally it is important to find mathematics containing the rules (dynam- ics) of both objects and observation, which will be called "Adaptive dynamics".
2 Adaptive Dynamics
In adaptive dynamics, it is essential to consider in which states and by which ways we see objects. That is, one has to select phenomena and prepare mode for observation for understanding the whole of a system. Typical adaptive dynamics are the dynamics for state-adaptive and that for observable-adaptive.
State-adaptive dynamics is that the dynamics of a system depends on a state at one instant in which the interaction is switched on, or that in a composite system the interaction depends on the instant state of at least one of sub-
system. Examples of such adaptivity are seen in a compound state (or nonlinear lifting) studying quantum communication [6, 151 and in an algorithm solving NP complete problem in polynomial time with stochastic limit [l, 2, 4, 51.
Observable-adaptive dynamics is that the dynamics of a system depends
on observables to be observed and the measurement depends on an observable chosen. Examples of this adaptivity are used to understand chaos [8, 111 and examine violation of Bell's inequality [3].
There exists a deep relation between the adaptive dynamics and Information dynamics [7, 8, 161, which we will not discuss here. In this paper it is discussed how we can understand chaos in the adaptive scheme.
3 Entropic Chaos Degree In quantum systems, for von Neumann entropy S ( p ) and quantum mutual en- tropy I (p ; *) with a linear channel *, we define D (p ; *) = S ( *p ) - I (p ; *). Then it is easy to check that
where {En} is the set of one dimensional projections giving a Schatten decompo- sition of p. Therefore the above quantity D ( p ; *) is interpreted as the complexity produced through the channel *. We apply this quantity D (p ; *) t o study chaos even when the channel describing the dynamics is not linear. D (p ; *) is called the entropic chaos degree in the sequel.
In order to describe more general dynamics such as in continuous systems,
we define the entropic chaos degree in C*-algebraic terminology. This setting
128
will not be used in the sequel application, but for mathematical completeness we will discuss the C*-algebraic setting.
Let (A,(S) be an input C* system and (Ale) be an output C* system; namely, A is a C* algebra with unit I and (S is the set of all states on A. We assume 2 = A for simplicity. For a weak* compact convex subset S (called the reference space) of 8, take a state cp from the set S and let
--
cp = J wdp, S
be an extremal orthogonal decomposition of cp in S, which describes the degree of mixture of cp in the reference space S. In more detail this formula reads
c p ( 4 = 1 W(A)dP&), A E -4 s
The measure p, is not uniquely determined unless S is the Choquet simplex, so that the set of all such measures is denoted by M, (S) .
Definition 1 The entropic chaos degree with respect to cp E S and a channel * is defined by
1 Ds (p; *) = inf { Ss (*w) dp ; p E M, (S)
where Ss (*cp) is the mixing entropy (or S-entropy) of a state cp in the reference space S .
Before stating the theorem, we review the definition of the mixing S-entropy. For a state cp E S c G(A), put
D,(S) =
where 6(p) is the delta measure concentrated on {cp}, and put
for a measure p E D,(S). Then the S-entropy of a state cp E S is defined as
ss(cp) = { ; 2 H ( P ) ; P E D, (S)) when D, (S) # Q) otherwise
When S =8, Ds (cp; *) is simply written as D (cp; *) . This Ds (cp; *) contains the classical chaos degree and the quantum above.
The classical entropic chaos degree is the case that A is abelian and cp is the probability distribution of a orbit generated by a dynamics (channel) *;
129
1 ( k = j ) 0 ( k # . i )
p = E k p k b k , where b k is the delta measure such as 6k ( j ) = Then the classical entropic chaos degree is
Dc (9; *) = C p k S ( * b k ) k
with the entropy S. Summarize that Information Dynamics can be applied to
the study of chaos in the following way:
Definition 2 When p E S changes to *p, the entropic chaos degree associated to this state change(dynamics) * is given by
Ds (p;*) = inf { l S s ( * p ) d p ; p E Mv ( S ) } .
Definition 3 A dynamics * produces chaos iff Ds (p; *) > 0.
4 Algorithm Computing Entropic Chaos Degree
In order to observe a chaos produced by a dynamics, one often looks at the
behavior of orbits made by that dynamics, more generally, looks at the behavior
of a certain observed value. Therefore in our scheme we directly compute the
chaos degree once a dynamics is explicitly given as a state change of system.
However even when the direct calculation does not show a chaos, a chaos will
appear if one focuses to some aspect of the state change, e.g., a certain observed
value which may be called orbit as usual. In the later case, algorithm computing
the chaos degree for classical or quantum dynamics consists of the following two cases [Ill:
c RN : First
find a difference equation zn+l = F (zn) with a map f on I = [a, bIN C RN into itself, secondly let I = U k A k be a finite partition with A i f l Aj = 8 ( i # j ) . Then the state p(n) of the orbit determined by the difference equation is defined
by the probability distribution ( p p ) ) , that is, p(n) = where for an
initial value z E I and the characteristic function 1~
N (1) Dynamics is given by 2 = Fb (z) with z E I = [a, b]
Now when the initial value z is distributed due to a measure v on I , the above
pin) is given as
The joint distribution ( p."n+') :J ) between the time n and n + 1 is defined by
130
or
Then the channel *, at n is determined by
(,,,+I)
: transition probability ==+ p(,+') = *,cp(,),
and the entropic chaos degree is given by
We can judge whether the dynamics causes a chaos or not by the value of D a s
D > 0 u chaotic
D = 0 stable.
This chaos degree was applied to several dynamical maps such as logistic map, Baker's transformation and Tinkerbel map, and it could explain their chaotic characters. This chaos degree has several merits compared with usual measures such as Lyapunov exponent as explained below.
Therefore it is enough to find a partition {A,+} such that D is positive when the dynamics produces chaos.
(2) Dynamics is given by pt = f:po on a Hilbert space: Similarly as making a difference equation for (quantum) state, the channel *, at n is first deduced from F:, which should satisfy p("+l) = *,p(,). By using this constructed chan- nel, (a) we compute the chaos degree D directly according to the definition of ECD or (p ) we take a proper observable X and put x, = p(")(X), then go back to the algorithm (1).
The entropic chaos degree for quantum systems has been applied to the analysis of quantum spin system and quantum Baker's transformation[9].
Note that the chaos degree D above does depend on a partition A taken, which is somehow different from usual degree of chaos. This is a key point of our understanding of chaos, which will be discussed in the following sections.
131
4.1 Logistic Map
Let us apply the entropy chaos degree (ECD) to logistic map. Chaotic behav- ior in classical system is often considered as exponential sensitivity to initial condition.
The logistic map is defined by
X,+l = U2" (1 - 2") , x , E [O, 11 , o I u 5 4
The solution of this equation bifurcates as shown in Fig.1.
x,
a 3.2 3.4 3.6 3.8 4
Fig.1. The bifurcation diagram for logistic map
In order to compare ECD with other measure describing chaos, we take Lyapunov exponent for this comparison and remind here its definition.
<Lyapunov exponent X (f)> Let f be a map on R, and let 20 E R. Then the Lyapunov exponent
Xo(f)exponent of the orbit (3 = ( f " ( 2 0 ) ~ f o ... o f ( x o ) : n = 0 , 1 , 2 , . . . ) is defined by
When f = ( f i , f2,... , fm) is a map on R" and r0 E R". The Jacobi matrix J, = D f" (TO) at TO is defined by
132
0 . 5 .
0 . 4 .
0.3 .
0 . 2
0 . 1 .
0 : 3 3 . 2 3 . 4 3.6 3 . 8 4
Fig.2. Chaos degree for logistic map
Then, the Lyapunov exponent A 0 (f) off for the orbit 0 = {f" (zo) ; n = 0 ,1 ,2 , . . .} is defined by
~0 (f) = log ,GI, f i k = lim ( p ~ )
Here, p: is the Ic-th largest square root of the m eigenvalues of the matrix J,J:.
(~c = 1 , . . . , m) . n-co
A 0 (f) > 0 + Orbit 0 is chaotic.
A 0 (f) 5 0 + Orbit 0 is stable.
The properties of the logistic map depend on the parameter a. If we take a particular constant a , for example, a = 3.71, then the Lyapunov exponent and the entropic chaos degree are positive, the trajectory is very sensitive to the initial value and one has the chaotic behavior. It is important to notice that if the initial value zo = 0, then z, = 0 for all n.
Tinkerbell map
Let us compute the CD for the following two type Tinkerbell maps fa and fb
on I = [-1.2,0.4] x [-0.7,0.3].
133
A
0 . 5
a
- 0 . 5
-1
- 1 . 5
3 3.2 3.4 3.6 3.8 4
Fig.3. Lyapunov exponent for logistic map
) = ( ( X i " ' ) 2 - (XI"') + a$) + CZXI"), 2 2 p X p + c 3 2 p + c 4 z p ,
= ( (x?')' - 1 7
2
fa (z'"') = fa (zI.',zI"'
fb (d"') = fb (Xp ' ,z I" ' )
+ c~x?) + C Z Z ~ ' , 22P)xp ' + bzp) + c42p)
where (z?',zI"') E I , -0.4 I a I 0.9, 1.9 I b 5 2.9, ( C I , c 2 , c 3 , c 4 ) = (-0.3,
-0.6, 2.0, 0.5) and (z?),~?') = (0.1,O.l).
Let us plot the points (z?',zI")) for 3000 different n's between 1001 and
4000. Fig.4 and Fig.5 are examples of the orbits of fa and fb in a chaotic domain.
134
$1 0.6 ,
::: 1 -1b
Fig.4. Orbits of fa.
-0.7 1 Fig.5. Orbits of fb
135
0.8
0.7
0.6 0.5
0.4
0.3
0.2
0.1
0 -02 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 a
Fig.6. ECD of Tinkerbell map fa
ECD r
1.4
1.2
1 .o 0.8
0.6
0.4
0.2
0 a 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0
Fig.7. ECD of Tinkerbell map f b
The ECD of Tinkerbell map fa and f b are shown in Fig. 6 and Fig. 7. Here we took 740 different a’s between -1.2 and 0.9 and 740 different b’s
between 1.9 and 2.9 with
-- 100 100
(i = -120, -119,. . . ( j = -70, -69,. .. , -l,O, l,.. . ,28,29)
-1, 0,1, . . . ’38’39)
n = 100000.
From the above examples it is seen that Lyapunov exponent and chaos degree have clear correspondence, moreover the ECD can resolve some inconvenient properties of the Lyapunov exponent as follows [lo]:
136
Lyapunov exponent takes negative and sometimes -m, but the ECD is always positive for any a 2 0.
It is difficult to compute the Lyapunov exponent for some maps like Tin- kerbell map f because it is difficult to compute f" for large n. On the other hand, the ECD of f is easily computed.
Generally, the algorithm for the ECD is much easier than that for Lya- punov exponent.
ECD with memory
Here we generalize the above explained ECD to take the memory effect into account. Although the original ECD is based upon the choice of the base space C := {1 ,2 , . . . , N } , we here take another choice. Em, instead of C, is a new base space. On this base space, a probability distribution is naturally defined as
(n,n+l > , . , ,n+m). The with its mathematical idealization, pioil...,, := limn+cx, pioil nel A:, over Ern is defined by a transition probability,
~ j ~ i ~ , , . i , + ~ ~ i ~ j ~ . . . &,,,j,,, = P ( ~ I , ~ z , * . . , i r n , i r n + l I . i ~ , . i ~ , . . . ,.irn)~jo,jl...j,.
Thus it derives the ECD with m-steps memory effect,
It is easy to see that this quantity coincides with the original CD when m = 1. This memory effect shows an interesting property, that is, the longer the
memory is, the closer the ECD is to the Lyapunov exponent for its positive part.
The entropic chaos degree can be used to study some quantum chaos [9, 131.
5 Description of Chaos by Adaptive Dynamics Our discussion of this section is based on a recent work (81, which was a trial to explain several chaos proposed in various experiments, so that our formulation can be applied to dynamics in finite systems.
First of all we examine carefully when we say that a certain dynamics pro- duces a chaos. Let, us take the logistic map as an example. The original differ- ential equation of the logistic map is
dx dt _ - - U Z ( ~ -x),O 5 u 5 4
137
4.2
with initial value xo in [0,1]. This equation can be easily solved analytically, whose solution (orbit) does not have any chaotic behavior. However once we
make the equation above discrete such as
x,+1 = ax,(l - X,),O 5 a 5 4.
This difference equation produces a chaos.
Taking the discrete time is necessary not only to make a chaos but also to
observe the orbits drawn by the dynamics. Similarly as quantum mechanics, it
is not possible for human being to understand any object without observing it,
for which it will not be possible to trace a orbit continuously in time.
Now let us think about finite partition A={A,; k = 1,. . . , N} of a proper set
I = [a, bIN c RN and equi-partition Be = {Bi; k = 1 , . . . , N } of I . Here "equi" means that all elements Bg are identical. We denote the set of all partitions
by P and the set of all equi-partitions by P e . In the section 4, we specified
a special partition, in particular, an equi-partition for computer experiment
calculating the ECD. Such a partition enables to observe the orbit of a given
dynamics, and moreover it provides a criterion for observing chaos. There exist
several reports saying that one can observe chaos in nature, which are very
much related to how one observes the phenomena, for instance, scale, direction,
aspect. It has been difficult to find a satisfactory theory (mathematics) to explain such chaotic phenomena. In the above difference equation we take some
time interval r between n and n + 1, if we take r -+ 0, then we have a complete
different dynamics. If we take coarse graining to the orbit of xt for time interval
r; 2, z $ sc-l)T xtdt, we again have a very different dynamics. Moreover it
is important for mathematical consistency to take the limits n + 00 or N (the
number of equi-partitions)+ 00 , i.e., making the partition finer and finer, and
consider the limits of some quantities as describing cham, so that mathematical
terminologies such as "lim", "sup", "int" are very often used to define such
quantities. Let us take the opposite position, that is, any observation will be
unrelated or even contradicted to such limits. Observation of chaos is a result
due to taking suitable scales of, for example, time, distance or domain, and it
will not be possible in the limiting cases.
It is claimed in [8] that most of chaos are scale-dependent phenomena, so the
definition of a degree measuring chaos should depend on certain scales taken.
Such a scale dependent dynamics is nothing but adaptive dynamics.
Taking into consideration of this view we modify the definitions of the chaos
degree given in the previous sections as below.
Going back to a triple (A, (5, a! (G)) considered in Section 2 and we use this
triple both for an input and an output systems. Let a dynamics be described
by a mapping rt with a parameter t E G from (5 to (5 and let an observation
be described by a mapping 0 from (A, (5, a! (G)) to a triple (a, 2, /3 (G)). The
triple (B,T,p(G)) might be same as the original one or its subsystem and
the observation map 0 may contains several different types of observations,
that is, it can be decomposed as 0 = O,..Ol.Let us list some examples of
observations.
138
For a given dynamics f = F ( c p t ) , equivalently, cpt = rtcp, one can take
several observations. Example: Time Scaling (Discretizing): 0, : t -+ n, f (t) --t cpn+l, so that
3 = F ( c p t ) + cpn+l = F (pt) and cpt = r,*cp + pn = l?Lcp. Here T is a unit time needed for the observation.
Let (B,%,p(G)) be a subsystem of ( A , G , a ( G ) ) , both of which have a cer- tain algebraic structure such as C*-algebra or von Neumann algebra. As an example, the subsystem (B , Z,@ (G)) has abelian structure describing a macro- scopic world which is a subsystem of a non-abelian (non-commutative) system (A, 8, a (G) ) describing a micro-world. A mapping 0c preserving norm (when it is properly defined) from A to B is, in some cases, called a conditional ex- pectation. A typical example of this conditional expectation is according to a
projection valued measure {Pk; PkPj = Pk&j = P;&j 2 0, Ck Pk = I } its-
sociated with quantum measurement (von Neumann measurement) such that
0~ ( p ) = Ck PkpPk for any quantum state (density operator) p . When B is a von Neumann algebra generated by {Pk} , it is an abelian algebra isometri- cally isomorphic to Loo (0) with a certain Hausdorff space 0, so that in this case 0 c sends a general state cp to a probability measure (or distribution) p . Similar example of 0c is one coming from a certain representation (selection) of a state such as a Schatten decomposition of p ; p = ( 3 ~ p = x k X k E k by
one-dimensional orthogonal projections {& } associated to the eigenvalues of p with x k E k = I . Another important example of the size scaling is due to a finite partition of an underlining space 0, e.g., space of orbit, defined as
d
Example: Size Scaling (Conditional Expectation, Partition):
0 P (a)={ Pk Pk n Pj = pk6k . j ( k , j = 1, ' ' ' N ) , u:=, Pk = 0) .
5.1 Chaos degree with adaptivity
We go back to the discussion of the entropic chaos degree. Starting from a given dynamics cpt = rfcp, it becomes cpn = rLcp after handling the operation 0,. Then by taking proper combinations 0 of the size scaling operations like UC, 0~ and Q p , the equation pn = r;cp changes to 0 (cpn) = 0 (I'Lcp), which will be written by 09, = Ol?~O-lOp or cpz = rLocpo. Then our entropic chaos degree is redefined as follows:
Definition 4 The entropic chaos degree of I?* with a n initial state cp and ob- servation 0 i s defined by
where po is the measure operated b y 0 to a extremal decomposition measure of
cp.
Definition 5 The entropic chaos degree of I?" with a n initial state cp i s defined b y
139
D (9; I?*) = inf {Do (p; r*) ; O E SO} , where SO is a proper set of observations naturally determined by a given dy- namics.
In this definition , SO is determined by a given dynamics and some con-
ditions attached to the dynamics, for instance, if we start from a difference
equation with a special representation of an initial state, then SO excludes 0, and OR.
Then one judges whether a given dynamics causes a chaos or not by the
following way.
Definition 6 (I) A dynamics I?* is chaotic for a n initial state cp in a n obser- vation O iff Do (p; I?*) > 0.
(2) A dynamics I?* is totally chaotic for an initial state p i f f D (cp; I?*) > 0. The idea introducing in this section to understand chaos can be applied
not only to the entropic chaos degree but also to some other degrees such as dynamical entropy, whose applications and the comparison of several degrees
will be discussed in the forthcoming paper.
In the case of logistic map, z,+1 = az,(l - z,) = F (zn), we obtain this
difference equation by taking the observation 0, and take an observation O p
by equi-partition of the orbit space 0 = {z,) so as to define a state (probability distribution). Thus we can compute the entropic chaos degree as was discussed
in Section 3. It is important to notice here that the chaos degree does depend on the
choice of observations. As an example, we consider a circle map
= fv(e,) = en + w (mod 2 4 ,
where w = 2 m ( O < v < 1). If v is a rational number N / M , then the orbit (0,) is periodic with the period M . If v is irrational, then the orbit (0,) densely fills
the unit circle for any initial value 00; namely, it is a quasiperiodic motion.
Theorem 7 Let I = [0,27r] be partioned into L disjoint components with equal length; I = B1 n B2 n . . . n BL.
P = {Bk; k = 1,. . . , M ) implies Do (Q0; fv) = 0. (1) If v is rational number N / M , then the finite equi-partition
(2) If v is irrational, then Do (00; fv) > 0 for any finite partition P={BI;). Note that our entropic chaos degree shows a chaos to quasiperiodic circle
dynamics by the observation due to a partition of the orbit, which is different from usual understanding of chaos. However usual belief that quasiperiodic
circle dynamics will not cause a chaos is not at all obvious, but is realized in a
special limiting case as shown in the following proposition.
Theorem 8 For the above circle map, if v is irrational, then D (00; f v ) = 0.
140
Such a limiting case will not take place in real observation of natural ob- jects, so that we claim that chaos is a phenomenon depending on observations, surrounding or periphery, which results the definition of chaos as above.
The details of this paper was discussed in [8] and will be discussed in [16].
References
(11 Ohya M., Masuda N. “NP problem in quantum algorithm”, Open Systems and Information Dynamics, Vo1.7, No.1, 33-39, (2000)
[2] Accardi L., Sabbadini R. “On the Ohya-Masuda quantum SAT Algorithm”, in: Proceedings International Conference UMC’O1, Springer (2001)
[3] Accardi L., Imafuku K., Refoli M.:On the EPR-Chameleon experiment, Infinite Dimensional Analysis, Quantum Probability and Related Topics
Vol. 5, No. 1 (2002) 1-20
[4] Ohya M., Volovich I.V. “New quantum algorithm for studying NP-complete problems” , Rep. Math . P hys . ,52, No. 1,25-33, (2003) and “Quant um comput- ing and chaotic amplifier”, J.Opt.B, 2003.
[5] Accardi L., Ohya M. “A Stochastic limit approach to the SAT problem”, Proceedings of VLSI 2003, and Open systems and Information Dynamics, 2004
[S] Accardi L., Ohya M., ”Compound channels, transition expectations, and
[7] Ohya M., Volovich I.V. “Mathematical Foundations of Quantum Informa-
[8] A.Kossakowski, M.Ohya and Y.Togawa (2003), How can we observe and describe chaos? Open System and Information Dynamics, 10(3):221-233
[9] K.Inoue, M.Ohya and 1.V.Volovich (2002) Semiclassical properties and chaos degree for the quantum baker’s map, J. Math. Phys., 43-2, 734-755.
[lo] K.Inoue, M.Ohya and K.Sato (2000) Application of chaos degree to some
Ill] M.Ohya: (1998), Complexities and their applications to characterization of
[12] R.S.Ingarden, A. Kossakowski and M. Ohya: (1997), Information Dynamics
[13] K.Inoue, M.Ohya and A.Kossakowski, A Description of Quantum Chaos,
liftings”, Appl. Math. Optim., Vo1.39, 33-59, 1999.
tion and Computation”, to be published in Springer-Verlag
dynamical systems, Chaos, Soliton & Fractals, 11, 1377-1385
chaos, International Journal of Theoretical Physics,Vol.37, No.1, 495-505.
and Open Systems, Kluwer Publ. Comp.
Tokyo Univ. of Science preprint (2002)
[14] Ohya M., Petz D.(1993), Quantum Entropy and its Use, Springer-Verlag
141
[15] Ohya M. (1983) On compound state and mutual information in quantum
[16] Ohya M. Adaptive dynamics its use in understanding of chaos, TUS
information theory, IEEE Trans. Information Theory, 29, No.5, 770-774.
preprint
142
Micro-Macro Duality in Quantum Physics*
Dedicated to Professor Tdeyuki Hida
on the occasion of his 77th birthday
Izumi Ojima
RIMS, Kyoto University, Kyoto, Japan
Abstract
Micro-Macro Duality means here the universal mutual relations between the microscopic quantum world and various macroscopic clas- sical levels, which can be formulated mathematically as categorical ad- junctions. It underlies a unified scheme for generalized sectors based upon selection criteria proposed by myself in 2003 to control differ- ent branches of physics from a unified viewpoint, which has played essential roles in extending the Doplicher-Haag-Roberts superselection theory to various situations with spontaneously as well as explicitly broken symmetries.
Along this line of thought, the state correlations between a system and a measuring apparatus necessary for measurements can canoni- cally be formulated within the context of group duality; the obtained measurement scheme is not restricted to the quantum mechanical sit- uations with finite number of particles but can safely be applied to quantum field theory with infinite degrees of freedom whose local sub- algebras are given by type I11 von Neumann algebras.
1 Why & what is Micro-Macro Duality?
- Vital roles played by Macro -
In spite of their ubiquitous (but implicit) relevance to quantum theory,
the importance of macroscopic classical levels is forgotten in current
trends of microscopic quantum physics (owing to the overwhelming belief in
the ultimate unification at the Planck scale?). Without those levels, how-
ever, neither measurement processes nor theoretical descriptions of micro-
scopic quantum world would be possible! For instance, a state w : B + C as one of the basic ingredients of quantum theory is nothing but a micro- macro interface assigning macroscopically measurable expectation value
'Invited talk at International Conference on Stochastic Analysis, Classical and Quan- tum held in Nagoya in November 2004
143
w ( A ) to each microscopic quantum observable A E M. Also physical inter- pretations of quantum phenomena are impossible without vocabularies (e.g.,
spacetime 2, energy-momentum p , mass m, charge q, particle numbers n; entropy S, temperature T, etc., etc.), whose communicative powers rely on
their close relationship with macroscopic classical levels of nature.
- Universality of Macro due to Micro-Macro duality -
Then one is interested in the question as to why and how macroscopic
levels play such essential roles: the answer is found in the universality of
“Macro” in the form of universal connections of a special Macro with generic Micro’s. To equip this notion with a precise mathematical formulation we
introduce the notion of a categorical adjunction Q Z C which controls the
mutual relations between [unknown generic objects Q (: microscopic quan-
tum side) to be described, classified and interpreted] and [special familiar model C (: macroscopic classical side) for describing, classifying and inter- preting], related by a pair of functors E(: c-iq) and F ( : q-c) , mutually
inverse up to homotopy I 3 E F , F E 5 I , via a natural isomorphism:
F
E
EaF(.) Q ( w , E ( a ) ) 2 C(F(w) ,a ) 7
E(.)%
so that
an ‘equation’ E ( a ) N w in Q to compare an unknown object w with
controlled ones E ( a ) specified by known parameters a in C can be
‘solved’ to give a solution a N F ( w ) which allows w to be interpreted
in the vocabulary a in C in the context and up to the accuracy specified,
respectively, by (E , F ) and (7, E ) .
Abstract mathematical essence of “Micro-Macro Duality” can be seen in
this notion of adjunction, whose concrete meanings are seen in the following discussion. What to be emphasized before going into details is the vast
freedom in the choices of categories Q, C and functors E, F which are not to be fixed but adjusted and modified flexibly so that our descriptions are
adapted to each focused context of given physical situations and to the
aspects to be examined. This point should be contrasted to the rigidity
inherent to the ultimate “Theory of Everything”. The simplest example of
duality is given by the Gel’fand isomorphism,
CommC*Alg(M, Co(M)) N HausSp(M, Spec(%)), (1)
between a commutative C*-algebra and a Hausdorff space defined by [cp*(x)](A) :=
[cp(A)](z) for [% 3 Co(M) ] 2 [M 5 Spec(%) = {x : M + C ; x: character
144
s.t. x(AB) = x(A)x(B)}] and for A E %,a: E M . Through our discus-
sion on the Micro-Macro duality below, we will encounter various kinds of
fundamental adjunctions appearing in quantum physics as follows:
1) Basic duality between algebras/ groups and states / representations
“Micro-Macro Duality” underlies “a unified scheme for generalized sec-
tors based upon selection criteria” [14] proposed by myself in 2003 to control
various branches of physics from a unified viewpoint. Extracted from a new
general formulation of local thermal states in relativistic QFT (Buchholz,
I0 and Roos [2]), this scheme has played essential roles in my recent work to
extend the Doplicher-Hag-Roberts superselection theory [5, 61 to recover a
field algebra 5 and its (global) gauge group G from the G-invariant observ-
able algebra % = SG and its selected family of states, according to which its
range of applicability restricted to unbroken symmetries has been extended
to not only spontaneously but also explicitly broken symmetries [15].
2) Adjunction as a selection criterion to select states of physical relevance to a specific physical situation, which ensures at the same time the phys-
ical interpretations of selected states. This is just the core of the present
approach to Micro-Macro Duality between microscopic quantum and
macroscopic classical worlds formulated mathematically by categorical ad- junctions:
q-c
c-q (generic:) Micro 2 Macro (: special model space with universality),
where c -+ q (q -+ c ) means a c + q (q -+ c) channel to transform classical
states into quantum ones (vice versa).
3) Symmetry breaking patterns constituting such a hierarchy as unbro- ken / spontaneously broken / explicitly broken symmetries: the adjunction
relevant here describes and controls the relation between [broken Z un-
broken], playing essential roles in formulating the criterion for symmetry
breakings in terms of order parameters. Through a Galois extension, an
augmented algebra can be defined as a composite system consising of the
object physical system and of its macroscopic environments including ex-
ternalized breaking terms, where broken symmetries are “recovered” and
the couplings with external fields responsible for symmetry breaking are
naturally described.
4) If we succeed in extrapolating this line of thoughts to attain an adjunc-
tion between [irreversible historical process] ?=! [stabilized hi- erarchical domains with reversible dynamics] through enough controls over
mutual connections among different physical theories describing different
domains of nature, we would be able to envisage a perspective towards a
homotopical dilation
145
theoretical framework to describe the historical process of the cosmic evolu-
tion.
2
2.1
Basic scheme for Micro/Macro correspondence
Definition of sectors and order parameters
In the absence of an intrinsic length scale to separate quantum and classical
domains, the distinctions between Micro and Macro and between quantum
and classical are to some extent ‘independent’ of each other, admitting such
interesting phenomena as “macroscopic quantum effects”. Since this kind of ’‘mixture”’ can be taken as ‘exceptional’, however, we put in parallel
micro//quantum//non-commutative and macro//classical//commutative, re-
spectively, in generic situations. The essence of Micro/Macro correspon-
dence is then seen in the fundamental duality between non-commutative
algebras of quantum observables and their states, where the latter transmit
the microscopic data encoded in the former at invisible quantum levels into
the visible macroscopic form. While the relevance of duality is evident from
such prevailing opposite directions as between maps cp : 8 1 + U2 of algebras
and their dual maps of states, cp* : Ea2 3 w +-+ cp*(w) = w o cp E Enl, their
relation cannot, however, be expressed in such a simple clear-cut form as
the Gel’fand isomorphism Eq.(l) valid for commutative algebras, because of
the difficulty in recovering algebras on the micro side from the macro data
of states. The essence of the following discussion consists, in a sense, in the
efforts of circumventing this obstacle for recovering Micro from Macro. Starting from a given C*-algebra U of observables describing a Micro
quantum system, we find, as a useful mediator between algebras and states,
the category Rep, of representations 7r = ( 7 r , f i r ) of U with intertwiners
T , T7r1(A) = 7r2(A)T (VA E a), as arrows E Rep,(nl,n2), which is nicely connected with the state space E, of U via the GNS construction: w E
(7rw,4j,) E Rep, with 0, E 4, s.t. w ( A ) = (0, I r,(A)R,) E, unitaFequiv.
( V A E U) and T,(U)R, = jj,. Two representations ~ 1 , 7 7 3 without (non-
zero) connecting arrows are said to be disjoint and denoted by 7r1 A 7r2, i.e.,
7r1 b 7r2 e Repm(n1,nz) = (0). The opposite situation to disjointness
can be found in the definition of quasi-equivalence, 7r1 = 7r2, which can be
simplified into
1:l up to
def
7r1 M 7r2 (: unitary equivalence up to multiplicity)
u 7r1(8)” N “2(U)” u C ( T 1 ) = c(r2) u W*(7r1)* = W*(7r2)*.
To explain the central support c(7r) of a representation 7r, we introduce the
universal enveloping W*-algebra U** N 7ru(U)” := W*(U) of C*-algebra U which contains all (cyclic) representations of U as W*-subalgebras W*(7r) :=
146
T(%)” c W*(%). In the universal Hilbert space fj, := $wE~mfjw, W*(U) and
W* ( T ) are realized, respectively, by the universal representation (T,, fj,), T , := & , E ~ m ~ w , and by its subrepresentations T(A) := P(T)T,(A) rp(,) (VA E a) in fj, = P ( T ) ~ ~ with P(T) E W*(%)’. W*(%) is characterized by
universality via adjunction,
W*Alg(W*(%), M ) N C*Alg(%, E ( M ) ) ,
between categories C*Alg, W*Alg of C*- and W*-algebras (with forgetful
functor E to treat M as C*-algebra E ( M ) forgetting its W*-structure due
to the predual M,) with a canonical embedding map % z E(W*(!Z)), so
that any C*-homomorphism V p : % + E ( M ) is factored p = E($) o qa through qa with a uniquely existing W*-homomorphism $I : W*(U) -+ M :
U 7% 1 0 \vv
E(W*(%)) i;;) E ( M )
In this situation, the central support C(T) of the representation T is defined
by the minimal central projection majorizing P(T) in the centre 3(W*(%)) := W*(%) n W*(%)’ of w*(%).
i) Basic scheme for Micro-Macro correspondence in terms of sec- tors and order parameters: The Gel’fand spectrum Spec(S(W* (a))) of the centre 3(W*(%)) := W*(%) n I+’*(%)’ can be identified with a factor
spectrum % of 8: h
h
Spec(S(W*(M))) N % := Fa/ =: factor spectrum,
defined by all quasi-equivalence classes of factor states w E Fa (with trivial centres ~ ( W * ( T ~ ) ) := W*(rw) n W*(T~) ’ = Cln, in the GNS representa-
tions ( ~ , , f j ~ ) ) .
Definition 1 A sector of observable algebra U is defined by a quasi- equivalence class of factor states of %.
In view of the commutativity of 3(W*(%)) and of the role of its spectrum, we can regard
- 0 Spec(S(W*(U))) N % as the classifying space of sectors to distin-
guish among different sectors, and
0 3(W*(%)) as the algebra of macroscopic order parameters to spec- ify sectors.
147
Then the map
h
Micro: 8* 2 Em -H Prob(G) c LOO(%)* : Macro, h
defined as the dual of embedding 3 ( W * ( % ) ) N LOO(%) L) W*(a), can be
interpreted as a universal q-+c channel, transforming microscopic quantum
states E Ea to macroscopic classical states E Prob(B) identified with prob-
abilities. This basic q + c channel,
h
E m 3 w - P~ = w” t3(wym))E E3(w*(m)) = M 1 ( s p 4 3 ( W * ( n ) ) ) ) = Prob(G) ,
describes the probability distribution pu of sectors contained in the central
decomposition of a state w of M: h
8 2 A - w” (xa ) = pu(A) = Prob(sector E A I w),
where wll denotes the normal extension of w E Em to W*(8). While it tells us as to which sectors appear in w , it cannot specify as to precisely which
representative factor state appears within each sector component of w.
ii) [MASA] To detect this intrasectorial data, we need to choose a m a - imal abelian subalgebm (MASA) n of a factor iM, defined by the condition
= n 2 Lm(Spec (n ) ) . Using a tensor product iM@% (acting on the
Hilbert-space tensor product c(7r)fi, @ L2(Spec(%))) with a centre given by
3(iM @ n) = 3(m) n = 1 @ L m ( S p e c ( n ) ) ,
we find a conditional sector structure described by spectrum S p e c ( n ) of a
chosen MASA n.
iii) [Measurement scheme as group duality] Since the W*-algebra n is generated by its unitary elements U(n), the composite algebra M@% can
be seen in the context of a certain group action which can be related with a
coupling of iM with the probe system 8 as seen in my simplified version [14] of Ozawa’s measurement scheme [17]. To be more explicit, a reformulation in
terms of a multiplicative unitary [l] can exhibit the universal essence of the problem. In the context of a Hopf-von Neumann algebra M ( c B( f i ) ) [7] with
a coproduct r : M + M @ M , a multiplicative unitary V E U( (M@M*) - ) c U(f i @ fi) implementing I?, r(o) = V * ( l 8 z)V, is characterized by the
pentagonal relation, v 1 2 & 3 h 3 = h 3 & 2 , on fi @ fi 8 3, expressing the
coassociativity of r, where subscripts i , j of Vij indicate the places in fi@fi@ fi on which the operator V acts. It plays fundamental roles as an intertwiner,
V(X @ L ) = (A @ X)V, showing the quasi-equivalence between tensor powers
of the regular representation X : M , 3 w - X(w) := (i @ w ) ( V ) E &a
generalized Fourier transform, X(wl * w2) = X(wl)X(w2), of the convolution
148
algebra M,, w1 * w2 := w1 8 w2 o I?. On these bases the duality for Kac
algebras as a generalization of group duality can be formulated. In the case
of M = Lw(G,dg) with a locally compact group G with the Haar measure
dg, the multiplicative unitary V is explicitly specified on L2(G x G) by
(Vt)(s, t ) := t (s ,s- l t ) for 4 E L2(G x G) , s, t E G, (2)
or symbolically in the Dirac-type notation,
V(S, t ) = Is, st) . (3)
Identifying M with the Hopf-von Neumann algebra Lw(G) = % correspond-
ing to G := U(%) given by the character group of our abelian group U(%) (assumed to be locally compact), we adapt this machinery to the present
context of the MASA ‘Jz, by considering a crossed product M G := [C 8 A(G)”] v a(M) [9] defined as the von Neumann algebra generated
by C 8 A(G)” = C 8 6 and by the image a(M) of M under an iso-
morphism a of M into M €3 Lw(G) N Lm(G,9X) N M 8 ‘Jz given by
[a(B)](y) := Ad,(B) = $,B4;, y E G, B E M where 4, is an action
of y E G on M. By definition, 332 ) a L G = 9X8 6 is evident for the triv-
ial G-action L with L(M) = M. The crossed product M )aor G is generated
by the representation 4(V) = lGdE(y) 8 A, of V on L2(M) 8 L2(G) with
the spectral measure E(A) = E ( x ~ ) of n (for Bore1 sets A in Spec(%)) defined by the embedding homomorphism E : % LC”(G) - M of % into
M, as seen from (w €3 i)(4(V)) = A(E*w) E C 8 6 and (i 8 n)(+(V)) = lGdE(y)R(Xy) E a(M). The action of 4(V) corresponding to Eq.(3) can be
expressed by
-
4(V)(t, 8 Ix)) = t, €3 Irx) for Y,X E G, (4)
satisfying the modified version of the pentagonal relation, 4(V)l24(V)13V23 = &34(v)12, or equivalently, V234(V)12V& = 4(V)124(V)13. Under the as- sumption that U(’Jz) is locally compact, the spectral measure E consti-
tutes an impn’mitivity system, &(E(A))d; = E(yA), w.r.t. a represen-
tation 4 of G on L2(M), from which the following intertwining relation
follows: 4(V)(4, 8 I) = (4, 8 A,)+(V), for y E G. While the role of a multiplicative unitary is to put an arbitrary representation p in quasi-
equivalence relation M with the regular representation A by tensoring with
A: ,o8 A U,(L 8 A)UJ M A, the above relation allows us to proceed further to
4 M 4(V)($8 L)4(V)* = 4 8 A = U,(L 8 A)u; M A.
The important operational meaning of Eq.(4) can clearly be seen in the
case where G is a discrete group which is equivalent to the compactness of the group U(’Jz) in its norm topology (or, the almost periodicity of functions on
149
it). In the present context of group duality with G as an abelian group gen-
erated by Spec(!Yt), the unit element L E G naturally enters to describe the
neutral position of measuring pointer in addition to Spec(’JI), in contrast
to the usual approach to measurements. Then Eq.(4) is seen just to create
the required correlation (“perfect correlation” due to Ozawa [18]) between
the states [, of microscopic system fm to be observed and that 17) of the
measuring probe system (n coupled to the former: +(V)(C, @ J L ) ) = I, 8 17) for Vy E G. Applying it to a generic statel E = CrEG cy&, of fm, an initial
uncorrelated state E @ I L ) is transformed by 4(V) to a correlated one:
The created perfect correlation establishes a one-to-one correspondence be-
tween the state [, of the system fm and the measured data y on the pointer,
which would not hold without the maximality of !Yt as an abelian subalge-
bra of fm. On these bases, we can define the notion of an instrument 3 unifying all the ingredients relevant to a measurement as follows:
In the situation with a state WE = ( ( 1 (-)Q of fm as an initial state of the system, the instrument describes simultaneously the probability p(A1wc) = 3(Alq)(1) for measured values of observables in !Yt to be found in a Bore1
set A and the final state J(Alwe)/p(Alwc) realized through the detection
of measured values [17]. While this measurement scheme of Ozawa’s is for-
mulated originally in quantum-mechanical contexts with finite degrees of
freedom where fm is restricted to type I, its applicability to general situ-
ations without such restrictions is now clear from the above formulation
which applies equally to non-type I algebras describing such general quan- tum systems with infinite degrees of freedom as QFT. Since instruments
do not exclude “generalized observables” described by “positive operator-
valued measures (POM)”, it may be interesting to examine the possibility
to replace the spectral measure d E ( y ) with such a POM as corresponding to
a non-homomorphic completely positive map for embedding a commutative
subalgebra into fm. In what follows, the above new formulation will be seen to provide a pro-
totype of more general situations found in various contexts involving sectors,
such as Galois-Fourier duality in the DHR sector theory and its extension
to broken symmetries with augmented algebras (see below). It is impor-
tant there to control such couplings between Micro (fm) and Macro (‘TI as
‘Note that any normal state of !M in the standard form can be expressed as a vectorial
state without loss of generality.
150
measuring apparatus) as 4(V) E M M GI whose Lie generators in infinitesi-
mal version consist of Ai E M and their “conjugate” variables to transform
G 3 x F+ yix E G. This remarkable feature exhibited already in von Neu-
mann’s measurement model, is related with a Heisenberg group as a central
extension of an abelian group with its dual and is found universally in such a form as Onsager’s dissipation functions, (currents) x (external forces), as a linearized version of general entropy production [ll], etc. To be precise,
what is described here is the state-changing processes caused by this type
of interaction terms 4(V) between the observed system M and the probing
external system ‘32, with the intrinsic (= %zperturbed”) dynamics of the for- m e r being neglected. While the validity of this approximation is widely taken
for granted (especially in the context of measurement theory), the problem
as to how to justify it seems to be a conceptually interesting and important
issue which will be discussed elsewhere.
iv) [Central measure as a c+q channel] Here we note that, from
the spectral measure in iii), a central measure p is defined and achieves a
central decomposition of M 8 % = L”(Spec(%), M) = J&,c%) MM,dp(a ) , where p ( A ) := wo(E(A)) with wo a state of % supported by Spec(%) being faithful to ensure the equivalence p(A) = 0 e E ( A ) = 0.
A central measure p is characterized as a special case of orthogonal mea-
sures by the following relations according to a general theorem due to Tomita
(see [3] Theorem 4.1.25): for a state w E Ea of a unital C*-algebra U there is
a 1-1 correspondence between the following three items, 1) (sub)central mea-
sures p on En s.t. w = Ell u’dp(w’) and [JEs,s w’dp(w’)] [Js u’dp(w’)] for
V A : Bore1 set in En, 2) W*-subalgebras B of the centre: B c 3(W*(7rW)) =
n 7r,(U)”, 3) projections P on 4, s.t. PR, = R,, P7r,(U)P c {PT,(Q)P}’. If p, B, P are in correspondence, they are related mutually
as follows:
1. 93 = {P}’ n 3(W*(r,));
2. P = [BR,];
3. ,(a,&. . . a,) = (Owl 7rw(A1)P 7rU(A2)P.. . P~F,(A,)R,), where a E
C(Ea) for A E U denotes a map A(q ) = cp(A) for ‘p E Ea;
4. B is *-isomorphic to the image of K , ~ : L”(Ea,p) 3 f - sp(f) E
rU(U)‘ defined by (0, I n,(f)r,(A)R,) = Jdp(w’) f (w’ )w’ (A) , and,
for A, B E U, K , ~ ( A ) ~ ~ ~ ( B ) R , = 7r,(B)P7rU(A)R2,.
When B = {P}’n3(W*(7r,)) = 3(W*(7rU)), or equivalently, ~(W*(?T,)) C
{P}’, p is called a central measure, for which we can derive the following
result from the above fact:
151
Proposition 2 ([16]) A map A, defined by
A p : 7ru(M)” 3 7ru(A) H ~ , (a ) E 3(W*(7ru))
is a conditional expectation characterized by
To summarize, we have established the following logical connections:
1) As dual of embedding 3 ( W * ( % ) ) 9 W*(!?l) of the centre, we obtain
a basic q+c channel En ++ Prob(Spec(3(W*(%))) = Prob(M) with a factor
spectrum M = Fa/ = as the classifying space of sectors.
2) A central measure pu with a barycentre w = sEB w’dpu(w’) E Ea specifies a conditional expectation Apw : W*(7ru) 3 7ru(A) - ~ , ~ ( a ) = [Spec(3(W*(7ru))) 3 w’ H w’ (A) ] E 3(W*(7ru)), whose dual
h
h
A;w : Prob(Spec(3(W*(7ru))) + Ew*(7rW)
h
defines a c+q channel given by S p e c ( 3 ( W * ( 7 r u ) ) ) [ ~ MI 3 y - wy := AEw(6,) = 6, 0 A, E supp(p,) c Fg[C Ez] as a (local) section of the
bundle Fa -+ [Fa/ M ] = M. 3 ) Operationally, this corresponds just to a choice of a selection cri-
terion to select out states of relevance and we have realized that the more internal structure to be detected, the larger algebra we need, which requires
the Galois extension scheme just in parallel with DHR sector theory and
with my propsal of general augmented algebra, as seen below.
h
2.2 Selection criteria to choose an appropriate family of sec- tors
Now we come to a “unified scheme for generalized sectors based on selection criteria” [13,14], extracted from a new general formulation of local thermal
states in relativistic QFT [2, 121. What I have worked out so far in this
direction can be summarized as follows:
A) Non-equilibrium
local states:
continuous sectors 1
C) Sector structure of
broken symmetry:
1 1
B) DHR sector theory of
unbroken internal symmetry:
discrete sectors
D) Unified scheme for
Micro-Macro based [ discrete & continuous 1 : [ on selection criteria
[
A) General formulation of non-equilibrium local states in QFT [2, 12, 131;
152
B) Reformulation [14] of DHR-DR sector theory [5,6] of unbroken internal
symmetry;
C) Extension of B) to spontaneously or explicitly broken symmetry [14, 151.
The results obtained in A), B) and C) naturally lead us to
D) Unified scheme for describing Micro-Macro relations based on
selection criteria [12, 13, 141:
1 q : generic states c : reference model system with +ll)
i, [ of object system ] " [ classifying space of sectors
iii) a map to compare i) with ii)
h 4
1 state preparation tk classification & selection criterion:
ii) a i) interpretation of ,
i) w.r.t. ii): i) ii) Q- c c-Q
which can be seen as a natural generalization of
Example 3 The formulation of a manifold M based on local charts { (Ux, cpx : Ux + an)} consisting of i)= local neighbourhoods lJx of M constituting a covering M = W x , ii)= model space R", iii)= local homeomorphisms cpx : Ux --t R", iv)= interpretation of the atlas in terms of geometrical invariants such as homology, cohomology, homotopy, K-groups, characteristic classes, etc., etc.
Example 4 Non-equilibrium local states in A ) [2, 12, 131 are character- ized by localizing the following generalized equilibrium states with fluctuating thermal parameters: i) = the set Ex of states w at a spacetime point x satisfying certain energy bound locally [w((l + Ho)") < 00 with "local Hamiltonian" Ho], i i) = the space BK of thermodynamic parameters (p, p ) to distinguish among different thermodynamic pure phases and the space M + ( B K ) =: Th of prob- ability measures p on BK to describe fluctuations of (p, p ) , iii) = comparison of an unknown state w with members of standard states wp = C*(p ) = sBK d p ( p , p ) ~ p , ~ with parameters p belonging to reference sys- tem, in terms of the criterion w = C*(p ) through "quantum fields at x" E I,
(justified by energy bound in i)). iv) = adjunction
z
153
with q-+c channel as a Yefl adjoint” to the c-+q channel C* (from the classical reference system to generic quantum states): as a localized form of
the zeroth law of thermodynamics, this adjunction achieves simultaneously the two goals of identifying generalized equilibrium local states and of giving the thermal interpretation (C*)- ’(w) p of a selected generic state w in
the vocabulary of a standard known object p E Th.
= C*c?i)
What we have discussed so far can be summarized as follows:
1. Classification of quantum states/representations by quasi-equivalence (= unitary equivalence up to multiplicity): achieved by means of sec- tors labelled by macroscopic order parameters as points in the
spectrum of centre, where a sector is defined by a quasi-equivalence class of factor states w E Fa with trivial centres 3(W*(ru)) := W*(ru) n W*(ru)‘ = C1fiw. In short, a sector = all density-matrix states within a factor representation = a folium of a factor state.
U
2. A mixed phase = non-factor state = non-trivial centre 3(W*(U)) # C1,j: allows “simultaneous diagonalization” as a central decomposition arising from non-trivial sector structure.
===+ 3(W*(U)): the set of all macroscopic order parameters to
distinguish among different sectors;
Spec(3(W*(%))) : a classifying space to parametrize sectors com- pletely in the sense that quasi-equivalent sectors correspond to one and the same point and that disjoint sectors to the different points.
U
3. Micro-macro relation: Intersector level controlled by 3 ( W * ( U ) ) : macroscopic situations pre-
vail, which are macroscopically observable and controllable;
Inside a sector: microscopic situations prevail (e.g., for a pure state in
a sector, as found in the vacuum situations, it represents a “coherent subspace” with superposition principle being valid).
4. Selection criterion = physically and operationally meaningful char-
acterization as to how and which sectors should be picked up for dis-
cussing a specific physical domain. E.g., DHR criterion for states w with localizable charges (based upon “Behind-the-Moon” argument)
nu r q q g T O ra(o/) in reference to the vacuum representation TO.
A suitably set up criterion determines the associated sector structure so that natural physical interpretations of a theory are provided in a
physical domain specified by it.
154
3 Sectors and symmetry: Galois-Fourier duality
To control the relations among algebras with group actons, their extensions
and corresponding representations, we need the Galois-Fourier duality as an important variation of our main theme Micro-Macro Duality. The essence
of DHR-DR theory [5 , 61 of sectors associated with an unbroken internal
symmetry can be seen in this duality which enables one to reconstruct a
field algebra 5 as a dynmaical system 5 OG with the action of an internal
symmetry group G from its fixed-point subalgebra Q = TG consisting of
G-invariant observables in combination with data of a family 7 of states
E Ea specified by the above DHR selection criterion:
Invisible micro Visible macro
In my recent reformulation, its applicability range restricted to unbroken
symmetries has been extended to not only spontaneously but also explicitly
broken symmetries. In B) DHR-DR sector theory, we see
1. Sector structure:
2. s(T(u)”) = @*C(I,, €3 IV,) = P(G); G = s ~ ~ c ( ~ ( T ( Q ) / ’ ) ) ==+ -@
vocabulary for interpretation of sectors in terms of G-charges.
3. ( ~ ~ ~ 4 ~ ) : sector of U (y,V,) E G : equiv. class of irred. unitary
representations of a compact Lie group G of unbroken internal sym- metry of field algebra 5 := U €3 c3d with a Cuntz algebra generated
09 by isometries.
4. (T, U, 4): covariant irred. vacuum representation of C*-dynamical sys-
tem 5 .A G, s.t. T ( T ~ ( ( F ) ) = U(g)T(F)U(g)*. 7
5 . U, G, 5: triplet of Galois extension 5 of Q = EG by Galois group
G = Gal(S/U), determining one term from two. How to solve two unknowns G & 5 from U?: DHR selection criterion
155
TannahKrein ==s duality
==+ I (C End(%)): DR tensor category E RepG G ==+
g r U x G .
-broken: G IJ H : unbroken T E! RepH - H ~ - - t IISHEG/HgHg-’ 1 n n
Ud = zH G/H: II U
- ~ M ( H T G ) = ~ ~ X G sector bundle
Similar schemes hold also for C) with spontaneously and/or explicitly
broken symmetries. For instance, in the case of SSB, we have [14]
with 3,(Ud) = L“(H\G;dg)@3,(Ud) = Lw(H\G; dg)@lw(A) and the base
space G/H of the sector bundle, Spec(3,(%’)) = ugHEc/Hgfig-’ -+ G/H, corresponds mathematically to the “roots” in Galois theory of equations and
physically to the degenerate vacua characteristic to SSB.
3.1 Hierarchy of symmetry breaking patterns and augmented algebras
Extension of B) to broken symmetries [14, 151: In my attempts to extend
DHR-DR sector theory with unbroken symmetries to the broken cases, the
adjunction,
Broken z Unbroken, augmented algebra
has been important, as seen in my criterion of symmetry breaking:
Definition 5 ([14]) A symmetry described by a (strongly continous) auto- morphic G-action r : G n 5(: field algebra), is unbroken in a given repre-
sentation ( ~ , 4 j ) of 5 i f the spectrum Spec(3,(3)) of centre 345) := s(5)”n ~ ( 5 ) ’ is pointwise invariant (p-a.e. the central measure p which decomposes T into factor representations) under the G-action induced on Spec(3,(5)). If the symmetry is not unbroken in ( ~ , 4 j ) ) it is said to be broken there.
7
w.r.t.
Remark 6 Since macroscopic order parameters Spec(3,(5)) emerge in low- energy infrared regions, a symmetry breaking means the “infrared(=Macro) instability” along the direction of G-action.
Remark 7 Since a representation 7~ with broken symmetry can still contain unbroken and broken subrepresentations, further decomposition of Spec(3,(5)) is possible into G-invariant domains. A minimal G-invariant domain is characterized by G-ergodicity which means central ergodicity. ==+ ?r is
156
decomposed into a direct sum (or, direct integral) of unbroken factor rep- resentations and broken non-factor representations, each component of which is centrally G-ergodic. phase diagram on Spec(3,(8)).
Thus the essence of broken symmetry is found in the conflict between factoriality and unitary implementability. In the usual approaches,
the former is respected at the expense of the latter. Taking the opposite
choice to respect implementability, we encounter a non-trivial centre which
provides convenient tools for analyzing sector structure and flexible treat-
ment of macroscopic order parameters to distinguish different sectors.
Namely, the adjunction holds between [Broken i 2 Unbroken2], controlled by a canonical homotopy 7 fro? [$ .A G with non-implementable
broken symmetry G in a pure phase] to [Z .A G with unitarily implementecj
symmetry G + U(G) in a mixed phase with a non-trivial centre], where 8 is an augmented algebra [14] defined by $ := 5 x (H\G), as a crossed
product of 8 by the coaction of H\G (: degenerate vacua) arising from the
symmetry breaking from G to its unbroken subgroup H . Note here that the above criterion does not touch upon the relation
between the symmetry group G and the dynamics of the physical system
described by the algebra 8 in relation with spacetime; if the latter is pre-
served by the former, the breakdown of symmetry G is called spontaneous (SSB for short). Otherwise, it is explicit, associated with some parameter changes involving changes of physical constants appearing in the specifi-
cation of a physical system. For instance, we can formulate such an ex- plicitly broken symmetry as broken scale invariance associated with temperature as order parameter [15], where augmented algebra of ob-
servables & = !2l x (S0(3)\(B+ x L i ) is the scaling algebra due to Buch-
holz and Verch [4] to accommodate the notion of renormalization group
(in combination with components arising from SSB of Lorentz boost sym-
metry due to thermal equilibrium [lo] to accommodate relative velocity
u!-’ := p!-’/p E S0(3)\Li). What is scaled here is actually Boltzmann con- stant kB!! In this way, we are led to the hierarchy of symmetry break- ing patterns ranging from unbroken symmetries, spontaneous and explicit
breakdown of symmetries, the latter of which would be related with more general treatments of transformations, such as semigroups or groupoids.
An eminent feature emerging through the hierarchy of symmetry break- ing patterns is the phenomena of ezternalization of internal degrees of free-
dom in the form of order parameters and breaking parameters, along which
external degrees of freedom coupled to the system are incorporated through Galois extension into the augmented algebra: it describes a composite
system consisting of the microscopic object system and its macroscopic “en-
non-trivial centre
*To be precise, “unbroken” should be understood as “unitarily implemented”.
157
vironments” , which canonically emerge at the macroscopic levels consist-
ing of macroscopic order parameters classifying different sectors and of
symmetry breaking terms such as mass rn and kg, etc. This formulation
allows us to describe the coupling between the system and external fields in a universal way (e.g., measurement couplings).
4 From [thermality 2 geometry] towards [history of Nature]
Although the modular structure of a W*-algebra in standard form has not
been explicitly mentioned so far, it plays fundamental roles almost every-
where in the above discussion, responsible for the homotopical extension mechanism: this is crucial, for instance, in the formulation of group duality
and of scaling as well as conformal aspects. From the viewpoint that the
notion of quasi-equivalence fundamental to our whole discussion is just a
form of homotopy, we show here the Galois-theoretical aspects of modular
structure !TI Z arising from canonical homotopy 7, : 7r -+ roo to move
to standard form.
I ,
Theorem 8 ([lS]) i ) I n the universal representation ( 7 r u , f j u = @ 3,)
of a C*-algebra %, we define the maximal representation no disjoint from a representation 7r = (.,fir) E Repa by
WEE%
7r’ := sup{p E Repa; p 5 xu, p 7r).
Then we have the following relations in terms of the projection P(7r) E
W*(%)’ on the representation space sj, of rr and its central support c(n) :
, I , , I
7r1 5 7r2 ===+ 7r; 2 7r;, 7ro = nooo and 7r 5 roo,
P(7r’) = .(.)I := 1 - c ( n ) ,
, I
P(7r””) = .(.)I1 = c(7r) = v uP,u* E P(3(W*(8))). UEU(?F@)’)
i i) Quasi-equivalence n1 = 7r2(* 7rl(%)” N 7 r 2 ( 8 ) ” c(r1) = c ( ~ 2 ) W*(7r1)* = W * ( q ) * ) is equivalent to
7rY0 = 7rg0.
( 1 , I
i i i) The representation (7roo,c(7r)fiu) of W*-algebra W*(7r) N 7roo(%)” in
the Hilbert space c(n)f iu = P(7rr”’)fi, gives the standard f o m of W*(r ) associated with a normal faithful semifinite weight p and the corresponding
158
Tomita- Takesaki modular structure ( J,,,, A,,,). It is characterized by the universality:
Std(7roo, a) N Repn(.rr,a),
where Std denotes the caterogy of representations of Q in standard form; according to this relation, any intertwiner T : 7r -+ a to a representation
(a,&,) in standard form of W*(a) is uniquely factored T = Toooqr through I ,
the canonical homotopy qT : T 3 roo with a uniquely determined intertwiner
Too : roo + a. iv) The quasi-equivalence relation 7r1 M 7r2 defines a classifying groupoid F a consisting of invertible intertwiners in the catego y Repn of representa- tions of U, which reduces on each T E Repa to Fa(7r,7r) N Isom(W*(7r),),
the group of isometric isomorphisms of predual W*(7r), as a Banach space. The modular structure in i i i) of W*-algebra W*(T) =: M in the standard
form in (7roo,c(7r)fj,) can be understood as the minimal implemention by the unitary group U(M’) of a normal subgroup Gm := I s o m ( M * ) m Q
I som(M,) f i ing M pointwise: namely, for y E Gml there exists U; E
\ \ , ,
$ 8
U(M’) s.t.
(yw,x) = (w,y*(x)) = (w,U?xU;) fo r w E M, ,
and UyxU!, = x J x E M. For M of type 1111 we can verify Galois-type relations involving crossed product by a coaction of the group Gm N U(m’) as follows:
= M V = M M G: Galois extension of M,
M = (M V ; fized-point subalgebra under Gm,
Gm = Gal(3(M)’/M): Galois group of M ~f
according to which factoriality 3(M) = C1 of M can be seen as the ergod- icity ofm under Aut(!Bl) or Gm:
CI = M n = n u(M’)/ = (mml)Gm 2 (m’)Aut(m).
In view of the dominant roles of thermal or modular-theoretical notions
mentioned above, this theorem suggests possible paths from thermality to ge- ometry to explain different geometries at macroscopic classical levels emerg-
ing from the invisible microscopic quantum world; it would explain the origin
of universality of Macro put in Micro-Macro Duality in our theoretical de-
scriptions of physical worlds. A typical example of this sort can be seen in
the formulation of group duality which exhibits its essence as a homotopi- cal duality involving interpolation spaces IS]. Moreover, we can develop
a framework to go into a step from the above modular homotopy to the
generalized version of classifying spaces or classifying toposes [16]. Along
159
this line of thoughts, we can envisage such a perspective that theoretical
descriptions of physical nature can be mapped into a “categorical bundle
of physical theories” over a base category consisting of selection criteria to
characterize each theory as a fibre, which are mutually connected by meta- morphisms of intertheory deformation arrows parametrized by fundamental
physical constants like ti, c, Ice; n, e, etc., controlled by the “method of vari-
ations of natural constants” (work in progress). One of the most important
virtues of the above augmented algebra is found in the possibility that such physical constants can be treated on the same footing as various physical
variables responsible for changing the symmetry properties of the systems; in
such contexts, they represent controlling parameters of deformations among different selection criteria to determine theories corresponding to stabilized hierarchical domains. Then the most crucial step will be to formulate each
selection criterion as an integrability condition in terms of generalized cat-
egorical connections, through which the framework can accommodate such
an adjunction as
[ historical process I = [ with reversible dynamics
irreversible
to be found among such adjunctions as to put a generic category with non-
invertible arrows (describing an irreversible open system in a historical pro- cess) in a relation adjoint to a groupoid with invertible arrows (correspond-
ing to a reversible closed system with repeatable dynamics in a specific
hierarchical domain). This kind of theoretical framework would provide
an appropriate stage on which the natural history of cosmic evolution be
developed .
1 homotopical stabilized hierarchical domains
References
[l] Baaj, S. and G. Skandalis, Ann. Scient. Ecole Norm. Sup. 26 (1993),
[2] Buchholz, D., Ojima, I. and ROOS, H., Ann. Phys. (N.Y.) 297 (2002),
425-488.
219 - 242.
[3] Bratteli, 0. and Robinson, D.W., Operator Algebras and Statistical Me- chanics, vol. 1, Springer-Verlag (1979).
[4] Buchholz, D. and Verch, R., Rev. Math. Phys. 7 (1995), 1195-1240.
[5] Doplicher, S., Haag, R. and Roberts, J.E., Comm. Math. Phys. 13 (1969), 1-23; 15 (1969), 173-200; 23 (1971), 199-230; 35 (1974), 49-85.
[6] Doplicher, S. and Roberts, J.E., Comm. Math. Phys. 131 (1990), 51-
107; Ann. Math. 130 (1989), 75-119; Inventiones Math. 98 (1989),
157-218.
160
[7] Enock, M. and Schwartz, J.-M., Kac Algebras and Duality of Locally
Compact Groups, Springer, 1992.
[8] Maumary, S. and Ojima, I., in preparation.
[9] Nakagami, Y. and Takesaki, M., Lec. Notes in Math. 731, Springer,
1979.
[fO] Ojima, I., Lett. Math. Phys. 11 (1986), 73-80.
[ll] Ojima, I., J. Stat. Phys. 56(1989), 203-226; Lec. Notes in Phys. 378, pp.164178, Springer, 1991.
[12] Ojima, I., pp. 48-67 in Proc. of Japan-Italy Joint Workshop on F’unda-
mental Problems in Quantum Physics, Sep. 2001, eds. Accardi, L. and
Tasaki, S., World Scientific (2003).
[13] Ojima, I., pp.365-384 in “A Garden of Quanta”, World Scientific (2003); e-print: cond-mat/0302283.
[14] Ojima, I., Open Systems and Information Dynamics, 10 (2003), 235-
279; math-ph/0303009.
1151 Ojima, I., Publ. RIMS 40, 731-756 (2004).
[16] Ojima, I., in preparation.
[17] Ozawa, M., J. Math. Phys. 25, 79-87 (1984); Publ. RIMS, Kyoto Univ.
21, 279-295 (1985); Ann. Phys. (N.Y.) 259, 121-137 (1997).
[18] Ozawa, M., quant-phys/0310072, to appear in Phys. Lett. A.
161
White noise measures associated to the solutions
of stochastic differential equations
Habib Ouerdiane
University of Tunis El Manar Faculty of Sciences of Tunis.
Campus universitaire, 1060 Tunis.Tunisia, E-mail: habib.ouerdiane@fst.rnu.tn
1 Introduction
Let N be a complex Frkchet nuclear space with topology given by an increasing family of Hilbertian norms ( 1 . I n , n E N}. It is well known that N may be represented as N = nnEWN, where the Hilbert space N, is the completion of N with respect to 1.1,. By the general duality theory N' is given by N' = UnE&, where Let 8 : R+ + R+ be a continuous convex strictly increasing function such that
= NA is the topological dual of N,.
lim - = 03, O(0) = 0. 2-00 z
Such functions are called Young functions. For a Young function 8 we define
o * ( ~ ) = sup(tz - e( t ) ) , (2) t2o
This is called the polar function associated to 8. It is known that 8* is again a Young function and (0.). = 8. For every p E Z and m > 0, we denote by Ezp(p(Np, 8, m) the space of entire functions f on the complex Hilbert space NP such that
l I f l l ~ , p , m := SUP If(z)le-e(m'zlp) < +03. (3) Z€Np
We fm a Young function 8. Then {F',m(N-p) := Ezp(N-,, 8, m ) ; p E N, m > O} becomes a projective system of Banach spaces and we put
Fe(N') = proj lim Ezp(N-,, 8, m) (4) p+w;m+O
162
which is called the space of entire functions on N’ with an 8-exponential growth
of minimal type. On the other hand {Ezp(N,, 8, m); p E N, m > 0} becomes an inductive system of Banach spaces and we put
This is called the space of entire functions on N with 8-exponential growth of arbitrary type. Then .Fo(N’) equipped with the projective limit topology is our test function space. The corresponding topological dual, equipped with the inductive limit topology, is denoted by 3z(N’) which is the generalized functions space, see [8] for more details. In particular, if N = &(R) (the complexified of the Schwartz test function space S(R)) and 8(z) = z2, then .Fo ( (N ’ ) is nothing than the analytic version of the Kubo-Takenaka test functions space and the corresponding topological dual is the Hida distributions space, see [9]. The test functions space of Kondratiev-Streit type ( S ) p , /3 E [0, 1) are obtained
choosing 8(z) = z h , see [14], [15], [21], [23]. More recently, it was introduced a two-variable version of the above spaces, see [ll]. In fact for arbitrary k E N, we can replace the nuclear space N by the product
NI x . . . X N k ,
and 8 by (81,. . . 8 k ) where 8i are Young functions and n/i is a complex nuclear Frbchet space, 1 5 i 5 k . Then it is possible to extend all the results obtained in [8] in the mulivariable case. In particular, the Laplace transform L: induces the following topological isomor- phism
and where (Nl x . . . x N ~ ) is the space of entire functions on N1 x . . . x Nk with 8*-exponential growth of arbitrary type with respect to 8* = (8;, . . . ,89, where 8; is the polar function corresponding to &. Another important result in [5] and [6] is the characterization theorem for convergent sequences of distributions in .F;(Ni x . . . x NL). Using this result, we can directly define for any given continuous stochastic process X ( t ) E Fi(N; x . . . x N;) the integral
F;(N{ x . . . x N;) Gfj*(Nl x * . . x N k )
Very useful in applications is the convolution product on 3; (N’), see [4], and [6] for details. In fact, we define the convolution of two distributions @, Q E Fi(N’) bY
which is well defined because Go* ( N ) is an algebra under pointwise multiplica- tion. We can define for any generalized function @ E Fi(N’) the convolution exponential of @ denoted by exp* @ as a generalized function on 3;*,@. ). (N’) Note that for a generalized function E (S)& the Wick exponential of @ de- noted by expo @ does not belong to (S)&, but it belongs to a bigger space of
@ * Q = c-l(c@. c q (7)
163
distributions (S)-’ called Kondratiev distribution space, see [13]. In this work, we do not restrict ourselves to the theory of Gaussian (White noise) and non-Gaussian analysis studied for example in [2], [9], [lo], [13], [14] and [15] but we develop a general infinite dimensional analysis. First, we give a decomposition of convolution operators from Fe(N’) into itself,
into a sum of holomorphic derivation operators. Then, we establish a topolog- ical isomorphism between the space L(Fe(N’), Fe(N’)) of operators and the space Fe(N’)&p (N) of holomorphic functions. Next, we develop a new convolution calculus over L(Fg(N’),Fe(N’)) and we give a sense to the expression eT := &o 5 for some class of operators T. As an application of this theory we solve some linear quantum stochastic differential equations. Finally using a recent result obtained in [18] and concerning white noise measures satisfying an exponential decay property, we give a asymptotic estimates of solutions of stochastic differential equations.
2 Preliminaries
For any n E M we denote by N On the n-th symmetric tensor product of N equipped with the 7r-topology and by N:” the n-th symmetric Hilbertian tensor product of Np. We will preserve the notation [ . I p and I . l - p for the norms on
NF” and N?; respectively. We denote by (., .) the C-bilinear form on N’xN connected to the inner product (.I.) of H =NO, i.e. ( z , t ) = (ZIE) , z E H , t E N. By definition f E Fe(N’) and g E Be(N) admit the Taylor expansions:
m
f (z) = C ( Z B n , f n ) , Z E N ’ , fn EN’’’ (8 ) n=O
00
g(E) = C(gn,Pn), F E N , gn E (Nan)’
where we used the common symbol (., .) for the canonical bilinear form on
(No”)’ x Nan for all n. In order to characterize Fe(N’) and Be(N) in terms of the Taylor expansions, we introduce weighted Fock spaces Fe,m(Np) and Gs,m(N-p) . First we define a sequence (0,) by
n=O
4
Suppose a pair p E N, m > 0 is given. Then, for f = (fn)F.o with fn E N p we put
00
n=O
m
n = O
164
Accordingly, we put
Finally, we define
Fe(N) = projlim Fe,,(Np),and Go(”) = indlim Ge,,(N-,). (10) p--too;mlO p-+w;m--tw
It is easily verified that Fe(N) becomes a nuclear RCchet space. By definition,
Fe(N) and Ge(N’) are dual each other, namely, the strong dual of Fe(N) is identified with Go (N’) through the canonical bilinear form:
n=O
The Taylor series map T (at zero) associates to any entire function the sequence of coefficients. For example, if the Taylor expansion o f f E Fo(N’) is given as
in (8), the Taylor series map is defined by If = f = (fn). In particular,
for every z E N‘, the Dirac mass 6, defined by << 6,,(p >:= p(z) , belongs to
Fi(N’). Moreover, b, coincide with the distribution associated to the formal
series 6, := ( ~ ) , Q O .
Theorem 1 [8] The Taylor series map T gives two topological isomorphisms Fe(N’) -+ Fe(N), Be*(N) + Ge(N’).
+
-+ 8-
3 Application to white noise analysis
For some functions 8, the spaces Fe(N’) and Be(N) play an important role in the theory of Gaussian and non Gaussian analysis. In fact let X c H c X’ be a real F’rkchet nuclear triplet. Let y be the standard Gaussian measure on
(XI, B) where B is the a-Borelian algebra on X’, determined via the Bochner- Minlos theorem by the characteristic function:
and l lEll i = (6, e ) ~ is the Hilbertian norm in the space F. By complexification
of the real triplet X c H c X’ we obtain N c 2 c N where N = X + iX and 2 = H + iH. Suppose that lim < 03 . Then Fe(N’) can by densely topologically embedded in the Hilbert space L2(X’,y) and we can construct the following Gelfand Triplet
Fe(N’) c L 2 ( X ’ , y ) c FG(N’) (13)
165
3.1 S-Transform
Let 0 be a Young function. Denote by Fi(N’) the strong dual of the test functions space Fe(N’). From condition (1) we deduces that for every E E N , the exponential function ec defined by ec(z) = e(’,t),z E N’ belongs to the space Fe(N’). The Laplace transform L of a distribution 4 E Fi(N’) is defined
by
By composition of the Taylor series map with the Laplace transfoLm, we deduce that 4 E .Fi(N’) if and only if there exists a unique formal series 4 = (q5n)n20 E
Ge (N ) such that ?(I) = Cn>O(E*n, &). Then, the action of the distribution
4 on a test function cp(z) = is given by
L(4)(I) = &I) = ((4,ec)), I E N . (14)
<< 4, ‘p x=-= c n!(4n, ‘pn) . (15) n > O
In the white noise Analysis we use the S-transform
(16) 1
S(4)(E) := L 4 ( 0 exp(-,E2),I E N , $ E Fe(N’).
Let now k , be given nuclear gaussian spaces ( X j c Hj c X$,y) and 0 = (el,&, ..., 0,) be a multivariable Young function, i.e., &,&, ..., BI, are k given Young functions and denote by
where Nj = X j + iXj ... @ y the k-fold tensor product of the standard gaussian measure. The next result give a characterization of new Gelfand triplet.
Theorem 2 I f we suppose’ that limz+a, < 00 for every 1 5 j 5 k , then Fo(N’) can be densely topologically embedded in the space L2(X ’ ,yBk) and we can construct the following Gelfand triplet: Fe((N’) c L2(X ’ ,yBk) c .Fi(N’). Moreover the chaotic transform (S-Transform) realizes a topological isomor- phism of nuclear triplets :
Zj = H j + iHj. Setting yBk = y @ y
To(”) c L2(X ’ ,yBk) c 3:(N’) 1 1 Is 1s
Fe(N’) c Fock(2’”) c Be*(N)
where I s is the Wiener -ItB-Segal isometry and Fock(Zk) is the Bosonic Fock space on Zk and e* = (el, e2, ..., &)* = (e;, e;, ..., e;).
166
3.2 Relation of this theorem with previous results
1. If k = 1 we obtain the results of [8]. In particular if e(z) = $ , a > 1
then @*(z) = with 1 + - - 1 and we obtain in this case the usual
space of entire functions of exponential type, see e.g., [21], [22] and [23]. For every f~ Fe(N) we have
a a
If a = 2 and X is the Schwartz space S(R), the space Fj,(S(R)) is the Hida distributions space, see [lo] and [9].
2. The Potthoff-Streit characterization theorem, see [24], is a particular case
of the general topological isomorphism: F:(N‘) + 90- (N ) where k = 1, O ( t ) = t2 and X = S(R).
3. In the particular case where k = 1, and N is a arbitrarily Banach complex space B and e(t) = ta, a 1 1 the spaces Fe(N’), Fe(N), L&(N), G,(N’) are introduced first by the author in [20], and the analog of Theorem 1 is given in this case.
4. In [7] Cochran-Kuo-Sengupta introduce the CKS space of distributions [v]: where a = (CY, ) ,~N is a positive sequence and G,(t) = En>,, a(n) f is an analytic function. If we put 8*( t ) = Log(G,(t2)) then [v]: = Fi(N). The hypothesis of the analycity of the function G,(t) in [7] is not necessary in our case, moreover we here obtain explicitly the space test functions and also a characterization theorem for this space.
-
4 Convolution calculus
In the next we develop a new convolution calculus over generalized functionals
space .Fi(N’). Unlike the Wick calculus studied by many authors, see [9], [15], [16], [14] and [23], the convolution calculus is developed independently of the Gaussian Analysis. In fact for 4 E Fi(N‘) and ‘p E Fe(N‘) the convolution of 4 and cp is defined by
( 4 * Cp)(Z) :=<< 4, T - z p >> , 2 6 N’ (17)
where T-, is the translation operator, i.e., T-,(P(z) = cp(z + z), z E N’ and for every t E N’, the linear operator T-= is continuous from Fo(N’) into itself. A direct calculation shows that 4 * ‘p E Fo(N’). Let 4 1 , 4 2 E Fi(N‘), we define the convolution product of 41 and 4 2 , denoted by 41 * 4 2 by
<< 41 * 4 2 , cp >> := [4i * ( 4 2 * cp)](O) , cp E .Fe(N’)
167
4.1 Convolution operators
In infinite dimensional complex analysis, a convolution operator on the test space Fe(N') denoted for simplicity by Fe is a continuous linear operator from Fe into itself which commutes with translation operators. It was proved in [4] that T is a convolution operator on Fe if and only if there exists 4~ E F; such that
Tcp = 4~ * cp , V cp E 3 e . (18)
Moreover, if the distribution 4~ is given by &- = (4m)m>o E Go and p(z) = CnzO(z@n, p n ) E Fe then
where (q5m, cpm+n)m denotes the right contraction of dm and cpm+n of order m, see [15]. In particular, we have
T(eE)(z) = dT * eE(z) = &<)eE(z).
Let 0 be a Young function, y E N' and p(z) = Cn,O(z@n,cpn) E 3 0 , then we define the holomorphic derivative of cp at the point F E N' in a direction y by
~ v c ~ ( z ) := C(n + l ) ( z m n , (y,cpn+l)l). n20
For each m E N the m-linear operator D : N' x ... x N' - C(F0, 3 0 ) defined
bY
is symmetric and continuous, hence it can be continuously extended to N'Dm, i.e.,
D : &, E dam H D+- E L(Fe,Fe)
The action of the operator Dbrnon a test function cp(z) = CnlO(~@n,cpn) is given by
(YI, ..., Ym) - DyI...Dyrn
(20) (n + m)! Z@n
Db,(cp)(Z) = c 7 ( 9 (4ml Pn+m)m). n > O
Then, in view of (18), (19) and (20), we give an expansion of convolution oper-
ators in terms of holomorphic derivation operators.
Proposition 3 Let T E C ( 3 ~ , 3 ~ ) . Then T is a convolution operator if and only if there exists 4 = ($m)m20 E GO such that
m>O
168
Let T+ = Cm20 D+,,, be a convolution operator and n E W. Then equality (18) shows that T$ := Tb o ... o Tb = Tp-. In particular
n
4.2 Symbols of operators
We denote by L(30, 3 8 ) the space of continuous linear operators from 3 0 into itself, equipped with the topology of bounded convergence. In this subsection we define the symbol map on the space L(30,30). Then we give an expansion of such operators in terms of multiplication and derivation operators.
Definition 4 Let T E L(30,F0), the symbol a(T) of the operator T is a com- plex -valued function defined by
a(T)(z, E ) := e-(*gE)T(eC)(z) , z E N' , E N.
Similar definitions of symbols have been introduced in various contexts, see [ll], [12], [15], [16], and [22]. In the general theory, if we take two nuclear F'rQchet spaces X and V then the canonical correspondence T - KT given by
(Tu,w) = ( K T , u @ v ) , u E X,W E V',
yields a topological isomorphism between the spaces L ( X , V ) and X'GV. In particular if we take X = V = 3 0 which is a nuclear FrQchet space, then we get
L(30,30) F,G30. So, the symbol u(T) of an operator T can be regarded as the Laplace transform of the kernel KT
a(T)(z , ( ) = KT(ec @ 6,) , z E N' , E E N. (21)
Moreover, with the help of equality (21) and theorem 2, we obtain the fallowing theorem.
Theorem 5 The symbol map yields a topological isomorphism
L(30,Fe) -+ FeGGi.
More precisely, we have the following isomorphisms
S.T .C (F~ ,F~) FeGGi - F ~ G G ~
T H a(~)(z,t) = C (Kl,m, 2 ~ ' B t ~ m ) H 2 = (Kl,m)l,m>o.
l,m>0
Example 6 1) The symbol of a convolution operator T+ = bu
D+- is given -
I69
2) If we denote by M f the multiplication operator by the test funct ion f . I ts symbol is given by
u(Mf)(z,E) = e- ( "pE) ( fec) (z ) = e-("yc) f (z)ec(z) = f ( z ) .
Let 2 E F&Ge and assume that 2 = f@ $ = ( f l @ 4m) l ,m>~ . Then the
operator T associated to K satisfies -+
T = MfT4, (22)
where f (2) = C120(z@'1, f l ) and T4 is the convolution operator associated to
the distribution 4 given by $. Moreover, we have
T = MfT+ = a(Mf) (z , D)a(T+)(z, D ) = u(T)(z , D).
Thus, using the density of Fe @ Ge in F&Ge, we obtain the following result.
Proposition 7 The vector space generated by operators of type (22) is dense in C(30, Fe).
4.3 Convolution product of operators
Let T1,T2 two operators in C(Fe,Fe), the convolution product of TI and T2, denoted by TI * T2, is uniquely determined by a(T1 * T2) = a(Tl)u(T2). If the operators TI and T2 are of type (22), i.e., TI = MflT+, and T2 = Mf,T+, then
TI * T2 = MflfiT',*cp,.
In particular, if T = MfT+ then for every n E N we have T*" = MfnT4.n. Let
T4 (resp. M f ) be a convolution (resp. multiplication) operator. Then for every
n EN TQIfn = T4.n = TT and M;" = M f n = Mf".
Proposition 8 Let T E Lo; then the operator e*T :=
L(F(.O*)* 1 F e e ) .
Proof. Let T E C(.Fe,F') and put S, = EL=, $. Then using the Laplace
transform isomorphism we shows that a(&) converges in Fee@G,e* to eU(*), from which the assertion follows. H
Let T E L(F9, Fe) and consider the linear differential equation
belongs to
A
d E dt - = T E , E(0) = I.
Then the solution is given informally by : E(t ) = etT, t E R. In the particular case, where T is a convolution or a multiplication operator; the solution E(t ) = etT is well defined since eT = erT. If T is not a convolution or a multiplication operator then the following theorem gives a sufficient condition on T to insure the existence of its exponential eT.
170
+ Theorem 9 Let K = (Kl,,) E FeGG,q satisfying (Kl,m, K ~ t , ~ , ) k = 0 f o r every m, I' 2 1, m', 1 2 0 and 1 5 k 5 m A 1' and denote by T the operator associated t o d. Then T" = T*", V n E N. Moreover
eT = e*T E L(F(,~* ). , ~ ~ 0 ) .
Proof. Using the last proposition, it will be sufficient to assume that K L , ~ =
( f i @ dm)) i.e.,
T = MfT4 = c M f l D h , l ,m>O
where f i ( z ) = (z@', f i ) Assume that f i = q@', q E N and $m = yBm, y E N' Then it is easy to see that
D L M f l = MflD4- + c k!CEeC~(Y,77)kMfi-hD4,-b, k=l
an equality on Fe. The assumption (&,m, &' ,m' )k = 0 implies that ( y , q) = 0. Then
Thus, using the density of the vector space generated by {q@l, q E N } in the
space N Oland the density of the vector space generated by {y@", y E N' } in " O m , we can extend the last equality to every f i E N O1 and dm E N'O" such that (&, f i ) k = 0 , V 1 5 k 5 I Am. Hence, we obtain
D4-Mfr = MflDbm *
MfT4 = c MflD4- = c DdmMfl = T 4 M f . l ,m>O l ,m>O
Then we have for every n E M
T" = (MfT4)" = ( M f ) " (T+)" = MfnTpn = T*".
This completes the proof. H
5 Applications to stochastic differential equa- t ions
A one parameter quantum stochastic process with values in L(F0, Fe) is a fam- ily of operators {Et, t E [O,T]} c L(Fe,Fe) such that the map t - Et is continuous.
Theorem 10 Let t E [O,T] H f ( t ) E Fe and t E [O,T] H d ( t ) E F; be two continuous processes and set Lt = Mf(t)T4(t) Then the linear differential equation
(23) _ - '2 - Mf(t)EtT+(t) I Eo = I
has a unique solution Et E L ( ~ ( , o * )* , Fe,s)given by
Et = .*(So" L a d s ) ,
171
Proof. Applying the symbol map to equation (23) to get
-- - a(Lt)u(Et) , a(1) = 1. dt
Then a(Et) = e-fJ‘(Ls)ds which is equivalent to Et = e*(- fJLsds). Finally, we conclude by the last proposition that Et E t(F(,e* ). , Fee).
Theorem 11 Let Lt be a quantum stochastic process with values in t(Fe,Fe) such that
a ( L t ~ s d s ) ( t , ~ ) = l ~ ~ ( K i . m ( t ) , 2 ~ 1 W ~ ~ m )
and assume that for every t E [O,T], m’,l 2 0 and m,l‘ 2 1 we have
(Ki,m(t), Kp,mi(t))k = 0 , V 1 5 k 5 m A 1 ’ .
Then the following differential equation
= L t E , E(0) = I , dE dt -
has a unique solution in .C(F(,~*)*,F,e) given by E(t ) = esi
5.1 Asymptotic estimates for white noise measures
In the sequel we take N = X+iX the complexification of a nuclear Fkechet space X. Let FO(N’)+ denote the cone of positive test functions, i.e. f E FQ(N’)+ if f ( y + i0) 2 0 for all y in the topological dual X’ of X.
Definition 12 The space 3O(N’) ; of positive distributions is defined as the space of 4 E FO(N’)* such that
(4 , f ) L 0, f E WW+.
We recall the following results on the representation of positive distributions.
Theorem 13 /19] Let 4 E Fo(N’):. There exists a unique Radon measure pb on X‘, such that
4 ( f ) = 1 f ( y + i O ) d ~ + ( y ) ~ f E Fe(N’). X‘
Conversely let p be a finite, positive Bore1 measure on X’. Then p represents a positive distribution in Fs(N’); i f and only i f p is supported by some X - p , p E W, and there exists some m > 0 such that
ee(mlYl-P)dp(y) < 00.
172
Given E E X and x E R, let At," = {y E X' : ( y , E ) > x } , denote the half-plane in X' associated to 6, x.
Theorem 14 Let 4 E Fe(N')* such that r$ defines a (positive) Radon measure p~d, on X'. For all E E X and x > 0 there exists m > 0 and p E N* such that:
where c = t1311e,m,p Proof. Using propriety of the Laplace transform of 4 E .Fe(N')* we have the following growth condition
I ~ ( < ) I 5 Cee'(mlElp), E X ,
for some m > 0 and p E *. For all t 2 0 we have the Chernoff type inequality:
hence p$(Ac,=) 5 Ce-(t"-e*(mtl~I~)), t 2 0.
Minimizing in t 1 0 we get, since (8.). = 8:
From Theorem 13, the result of Theorem 14 holds in particular for all positive distributions r$ E FO(N'):. Applying Theorem 13 and Theorem 14 we obtain a deviation result under an exponential integrability assumption, see [18].
Corollary 15 Let p be afinite, positive Bore1 measure on X' supported by some X - p , p E N* . Assume that for some m > 0,
L-p ee("'Y'-P)dp(y) < m.
Then for all E X we have:
173
5.2 Tail estimates for solutions of stochastic differential equations
Let c j : [O,T] -+ Fe(N’)* and M : [O,T] -+ Fe(N’)* be two continuous general- ized processes, and consider the initial value problem
(24) -- dXt - dt * X t + Mt, X o E Fo(N’)*. d t
In the particular case where dt = ado, cy E R, X = S(R) and (Mt)tE(o,T] is a
Gaussian white noise on [0, TI, (24) is a classical Omstein-Uhlenbeck equation.
Theorem 16 ([4) The stochastic differential equation (24) has a unique soh- tion in F(,v -l). ( N ) * , given by
X t = X o * e* 9 a d s + e* .rat +udu * M J s . I‘ From the relation (41 * 4 2 , l ) = ( $ 1 , 1 ) ( 4 2 , l), the expectation of X t satisfies
1fdt,Mt E F ~ ~ ( N ’ ) $ forallt ER+, t h e n e * S ~ ~ s d s a n d e * ~ ~ 9 s d s * M t E Fo;(N’): and we have the following corollary of Theorem 14 and Theorem 16.
Corol lary 17 Let Q; be such that Q;(r) 5 (e“ - l )*(~) for all r large enough, and assume that Xo,$t , Mt E Fo;(N’):, t > 0. Then the solution X t of (24) belongs to 30; (N’) : and the associated Radon measure (denoted by p ~ ~ ) satis- fies
for some Ct,rnt,pt > 0, t E R+
References
[l] L. Accardi and M. Bozejko:Interacting Fock spaces and gaussianization of probability measures. Infinite Dimensional Analysis, Quantum Probabilty and Related Topics, Vol. 1, no. 4 (1998), 663-670.
[2] S. Albever io , Yu. L. Daletsky, Yu. G . Kondra t iev and L. Stre i t : Non-Gaussian infinite dimensional Analysis. J. Funct. Analysis. 138, (1996), 311-350.
[3] N. Asa i , I. Kubo and H-H. Kuo:General characterization theorems and intrinsic topologies in white noise analysis, Hirochima Math. Journal, 31 (2001), 229-330.
174
[4] M. Ben Chrouda, M. El O u e d and H . Ouerd iane : Convolution Calculus and Applications to Stochastic Differential Equations. Soochow Journal of Math . 28(4):375-388, 2002.
[5] M. Ben Chrouda, M. El O u e d and H . Ouerd iane: Quantum stochas- tic processes and Applications, to appear.
[6] M. Ben Chrouda and H. 0uerdiane:Algebms of Operators on holomor- phic functions and Applications. Math. Physics, Analysis and Geometry, 5 (ZOOZ), 65- 76.
[7] W. G. Cochran, H-H. Kuo and A. Sengupta :A new class of white noise generalized functions, Infinite Dimensional Analysis, Quantum Prob- abilty and Related Topics, Vol.1, (1998), 43-67.
[8] R. Gannoun, R. Hachaichi, H. Ouerdiane and A. R e z g u i : Un the'ordme de dualite' entre espace de fonctions holomorphes ci croissance exponentielle. J. Funct. Analysis, Vol. 171, No. 1, (ZOOO), 1-14.
[9] T. Hida, H.-H. K u o , J . Potthof and L. Stre i t : White noise, an infinite-dimensional calculus, Kluwer Academic Publishers Group, Dor- drecht, 1993.
[lo] T. Hida: Brownian Motion, Berlin-Heidelberg-New York, Springer Verlag,
[ll] U. C. J i , N. Obata and H . 0uerdiane:Analytic characterization of gen- eralized Fock space operators as two-variable entire functions with growth conditions, Infinite Dimensional Analysis, Quantum Probabilty and Related Topics, Vol. 5, No. 3 (ZOOZ), 395-407.
(1980).
[12] P. Kr6e and R. Raczka:Kernels and symbols of operators in quantum field theory. Ann. I. H. P. Section A , Vol. 18, No 1, (1978), 41-73.
[13] Yu. G. Kondra t iev , L. Stre i t , W. W e s t e r k a m p and J.-A. Yan: Generalized functions in infinite dimensional analysis. Hiroshima Math. Journal, 28 (1 998), 21 3-260.
[14] H.-H. Kuo: White noise distribution theory, CRC Press, Boca Raton, New York, London and Tokyo, 1996.
[15] N. Obata: White noise calculus and Fock space. Lect. Notes in Math. Vol. 1577, Springer Vedag (1 994).
(161 N. Obata: Wick product of white noise operators and quantum stochastic differential equations. J. Math. SOC. Japan Vol. 51, No. 3, (1999), 613-641.
[17] N. 0bata:Multivariable White Noise Functions: Standard Setup Revisited, Publ. R.I.M.S, Kokyuroku 1227 (ZOOl) , 184-198.
175
[IS] H. Ouerdiane and N. Privault: Asymtotic estimates for white noise distributions. C. R. Acad. Sci. Paris, Ser. I338 (2004), 799-804
[19] H. Ouerdiane and A. Rezgui: Reprksentation intkgrale de fonctionnelles analytiques positives. Canadian Mathematical Proceedings( ZOOO), 28283- 290
[20] H. Ouerdiane: Dualite' et opkrateurs de convolution dans certains espaces de fonctions entieres nucleaires c i croissance exponentielle. Abhandlungen aus der Math. Seminar Hamburg, Band 54 (1983), 276-283.
[21] H. Ouerdiane: Fonctionnelles analytiques avec condition de croissance et applications ri l'analyse gaussienne. Japanese Journal of Math. Vol. 20, No.1, (1 994), 187-1 98.
[22] H. Ouerdiane: Noyaux et symboles d'opkrateurs sur des fonctionnelles analytiques gaussiennes. Japanese Journal of Math. Vol 21. No.1, (1995), 223-234 .
[23] H. Ouerdiane: Algkbres nuclkaires de fonctions entikres et equations aux derivkes partielles stochastiques. Nagoya Math. Journal. Vol. 151, (1998), 107-1 27.
[24] J. Potthoff and L. Streit: A characterization of Hida Distributions, J. Funct. Analysis, Vol. 101 (1991), 212-229.
176
A REMARK ON SETS IN INFINITE DIMENSIONAL SPACES WITH FULL OR ZERO CAPACITY
JIAGANG REN School of Mathematics and Computational Science,
Zhongshan University, Guangzhou, Guangdong 510275, P.R. China
MICHAEL ROCKNER Fakultiit fur Mathematik, Universitiit Bielefeld, 3361 5 Bielefeld, Germany
Abstract We give a simple proof that for classical Dirichlet forms on infinite di- mensional linear state spaces the intrinsic closure of a set of full measure has full capacity. Furthermore, we show that the C,,,-capacity of a set, enlarged by adding the linear span of a basis in the generalized Cameron- Martin space remains zero if it was zero for slightly bigger capacities a priori.
1. INTRODUCTION, FRAMEWORK AND A RESULT ON SETS WITH FULL CAPACITY
In infinite dimensional analysis the question whether a given set has zero or full capacity (in the sense that its compliment has zero capacity) is much less studied than in finite dimensions. This question is of importance, since roughly speaking capacity zero sets are not hit by the underlying process whereas a set of full capacity carries the process for all times. The first aim of this paper is to give a simple analytic proof for the fact that the intrinsic closure of a set of full measure has full capacity (cf. Theorem 1.4 below). This fact is essentially known to experts. We refer e.g. to [7] where this result was proved for a class of Dirichlet forms with non-flat underlying state space. But there is no reference for this result for general classical Dirichlet forms of gradient type on linear state spaces. In this case there is quite an easy proof which we present below. The second aim of this paper is to prove a result one would expect, but appears to be new. Namely, we prove that the Cl,,-capacity of a set, enlarged by adding all finite linear combinations of a basis in the generalized Cameron-Martin space, remains zero if it was zero for (slightly bigger) C,,,-capacities, r > 1, p > q, a priori (cf. Theorem 3.3 below). Let us first describe our framework, in which we strictly follow [2].
Let E be a separable Banach space over R. Let E' denote its dual and B(E) its Bore1 u-algebra. Let (H, (,)) be a Hilbert space such that H C E continuously and densely. Identifying H with its dual H' by Riesz's isomorphism, we have
E ' c H c E (1.1)
where both embeddings are continuous and dense. In particular, it follows for the dual- ization E,(, ) E : E' x E + R that
E,(l, h)E = ( I , h)H for all 1 E E', h E H.
177
Furthermore
H = { z E El sup{ ~ ' ( l , z ) ~ 1 1 E E' with lllll.q ,< l} < co }.
FCF := {g(ll , . . . , l ~ ) 1 N E M,g E CF(WN), I I , . . . , l ~ E E'},
(1.3)
The norm in E and H we denote by 11 I I E and 11 1J.q respectively. Let
(1.4)
where C?(WN) denotes the set of all infinitely differentiable bounded functions with all derivatives bounded. For u E .FCr(RN) and z E E we define Vu(z) E H by
(1.5) all d
dt (Vu(z), h)H = -&) := -u(z + th)lt=o.
Let p be a probability measure on (E ,B(E) ) and denote the corresponding real LP- spaces by U ( E , p ) , p E [I, 031, and define
For a set D of B(E)-measurable functions on E we denote the corresponding p-classes by
'6'. Throughout this paper we assume that the following hypothesis is fulfilled
(Hl) If u E FC? such that u = 0 p a x . , then Vu = 0 p-a.e. and the (thus on L2(E, p) well-defined) positive definite symmetric bilinear form
(&, 3 C r ) is closable on L2(E, p) . -P
Under condition (Hl) the Hilbert space H is sometimes called generalized Cameron- Martin space of p. We refer e.g. to [5] for the definition of closability and denote the
closure of (€,,FC," ) on L2(E ,p ) by (€P, H i 3 2 ( E , p ) ) . Then ( € P , H t g 2 ( E , p ) ) is a sym- metric Dirichlet form (see e.g. [5]).
-P
Remark 1.1 (i) For sufficient conditions for (Hl) we refer to [2]. We note that those conditions are also necessary, if one requires all partial derivatives to be closable separately (see [2] for details).
(ii) Closability of the form ( € P , F C ~ ) on L 2 ( E , p ) is equivalent to the closability of the operator
We denote its closure (whose domain is, of course, H,"'(E, p)) again by V.
-P
-P v : FC? c L2(E,p) + L2(E + H , p ) .
If this operator is closable, then also for p 2
-P V : FC? c LP(E,p) + LP(E + H , p )
is closable. Indeed, if un + 0 in P ( E , p) as TI + 00 and (VU,),~N is a Cauchy sequence in P ( E + H , p ) , then the same holds in L 2 ( E , p ) and L2(E + H , p ) respectively. By assumption it follows that Vu, + 0 in L2(E -+ H , p) as n + 00, hence in p-measure, so
by Fatou's Lemma
/11~u,ll~~dp < liminf I I V ~ , - ~u,~~gcip m-m s
But the right hand side can be made arbitrarily small.
178
(iii) Assuming that for p > 1
-P V : FCp c LP(E,p) + LP(E 4 H , p ) is closable, (1.7)
we can prove all what follows for p > 1 instead of p = 2 with entirely similar proofs. For simplicity we restrict, however, to the case p = 2. The definition of capacities, however,
we give for all p > 1 below. (iv) We refer t o [2] and [l] for examples for p satisfying (Hl). These examples include
the white noise measure on E , i.e. the centered Gaussian measure on ( E , B ( E ) ) with Cameron-Martin space H. But many other Gaussian measures and moreover Gibbs mea-
sures from statistical mechanics are included.
If for p E [1,co) condition (1.7) holds, we denote the closure by (V, H;’”(E,p)). For
notational convenience we then set as usual for p > 1
Now we recall the definition of capacity and intrinsic metric.
Definition 1.1. (i) For U C E , U open, and p E [l, co), we set
Cl,,(U) := inf{lluIl;,p I u E H , ~ ~ ( E , ~ ) , ~ > 1 p-a.e. on U)
Cl,,,(A) := inf{CIJU) I A c V } .
and for arbitrary A c E
CI,“(A) is called capacity of A . ($a) A function f : A H R, A c E, is called C1,p-quasicontinuous if there exist closed
sets A, C A, n E M, such that f I*,, is continuous for all n E M and limn+- Cl,”(E\A,) = 0.
Definition 1.2. For x , y E E set
AX, Y ) := sup{f ( x ) - f ( Y ) I f E FCr with llvf I I H < 1).
p is called intrinsic metric of (&p, Hil,’(E, p ) ) .
Lemma 1.3. Let x , y E E. Then The following is well-known. The proof is easy and included for the reader’s convenience.
A x , Y) = 115 - Y l l H ,
where we set l l z l l~ := f c o if z E E \ H .
Proof. Let f E FCP with I(V f I I H < 1 and assume x - y E H. Then
I lx -y l le . Here D f denotes the Frrkhet derivative of f . So,
179
So, by (1.3)
For A c E as usual we set
p ~ ( x ) := inf{p(x,y) I y E A } , z E E.
Now we can formulate the main result of this section which we shall prove in the next section.
Theorem 1.4. Assume hypothesis ( H l ) holds. Let A E B(E) such that p(A) = 1. Then Cl,Z(PA > 0) = 0, i.e., the p-closure of A has full C17z-capacity.
2. PROOF OF THEOREM 1.4
Throughout this section hypothesis (Hl) is assumed to hold. Before we can prove Theorem 1.4, we need the following lemma.
Lemma 2.1. Let K C E be ( 1 . 11s-compact and c E (0, co). Then p~ is B(E)-measurable and
p~ A c E H,”’(E,p) and I IV(~K A c ) [ ~ H < 1.
Furthermore, p~ A c is C1,z-quasicontinuous.
Proof. Let {ei I i E N} C E’ be an orthonormal basis of H separating the points of E , and for n E M define Pn : E H En := span{el,. . . ,en} by
n
P,Z := C E,(ei, z ) ~ ei, z E E. i=l
Fix y E E. By a simple approximation argument on EN we see that
u,(z) := IIP,z - Pnyll~ A c , z E E ,
is a function in H;”(E,p) with I]Vu,ll~ < 1. Clearly
un(x) T vv(x) := 11% - y l l ~ A c for all x E E. (2.1)
Hence by [5, Chap. I, Lemma 2.121 for y E E, vy E H t Y 2 ( E , p ) with [Ivv,))~ < 1 and
un + v, weakly in H i Z 2 ( E , p) , hence the Cesaro mean of a subsequence converges strongly
in H:’,’(E,p). A standard argument of Egorov type for capacities (cf. [5, Chap.111, Sect. 31) implies that selecting a subsequence if necessary, this Cesaro-mean converges C1,z- quasiuniformly, i.e., uniformly on closed sets whose compliments have arbitrarily small
Cl,z-capacity. Hence by (2.1) vy is C ~ J quasicontinuous.
180
Claim. Let dim E < 00. Then the assertions of the lemma hold even without assuming K to be compact.
Since dimE < 00, we have H = E and )I I I H and 1) JIE are equivalent norms. Let
{yn I n E M} be a countable 11 . IlH-dense subset of K and defining
vN:=inf{vy,,...vyN}, N E N ,
we have f K A c = inf V N on E.
N
Furthermore, (cf. e.g. [5, Chap. IV, Sect 41) V N E H; (E ,p ) with
~ / V v h ’ ~ ~ H G SUP{/IVVY1/lff7.” ! I ~ V v y ~ ~ ~ H ) r
hence
( (VVN(~H < 1 for all N E N. Therefore, the claim follows by the same arguments as above.
Now we go back to the general case. First we show that for all z E E
PPnK(Pnx) T PK(x) n W.
(So, in particular, p~ is B(E)-measurable.) Let z E E . Obviously, pp,K(Pnz) is increasing with n and
S~PPP,K(J‘~X) < PK(Z) n
(cf. (2.1)). To prove the dual inequality we may assume that supnppn~(Pnz) < 00. Let a E (0, co) such that
Then there exist kn E K such that
supPP,K(PnZ) C: a. n
IIPnx - PnknllH < a for all n E M. (2.3) Since balls in H are weakly compact and K is compact in E , we can find a subsequence
such that knl - k E K w.r.t. 11 ) I E and Pn,z-Pnjknj - h E H weakly in H . Hence for all i E M by (1.2)
Et(ei,h)E= Jim E, (e i ,PnJx -Pn jL j )E
= E,(ei,z - k ) E .
llz - kJIH 6 liminf IIPnjz - Pn,knjIIH < a.
j -m j-+m
3-m
j-m = lim E,(ei,z - knJ)E
Since {ei I i E M} separates the points of E , it follows that z - k = h and by (2.3) that
(2.4) I - -
In particular, pK(z) < 00.
Now suppose that for some E E: (0, p ~ ( z ) )
supPp,K(Pnz) < PK(z) - E E .
n
Then applying the above with a := p ~ ( z ) - E , we get a contradiction from (2.4). So, (2.2) is proved. Now the assertions follow from the claim and the same arguments used for its proof. 0
181
Proof of Theorem 1.4. By inner regularity there exist compact sets K, c A, n E M, such that p( K,) T p( A) = 1. Let
U, := PK, A 1, n E M.
Then
n (2.5) n-m u := inf un = lim u, 2 P A A 1 on E.
Furthermore, u = 0 on Unal K,, hence u = 0 p-a.e.
By Lemma 2.1 and the same arguments as in its proof we obtain that u is Cl,z-quasi- continuous, hence by [5, Chap. 111, Proposition 3.91 C1,2({u > 0)) = 0. But by (2.5),
0 {PA > 0) C {u > 0) and the assertion is proved.
3. A RESULT ON SETS WITH ZERO CAPACITY
Let { hl , hz, ' ' } be an ONB of H and let E, denote the linear span of { hl , . . . , h,} and set K := &En. A sufficient condition on A for p ( A + K ) = 0 is given in [4]. Now we look for a condition which implies C,,l(A + K) = 0. But we have to work under an additional quasi-invariance hypothesis.
For k E H define
T ~ ( z ) : = z - ~ , z E E . (3.1)
(H2) For all k E H , p is k-quasi-invariant, i.e., p o r,kl we assume the Radon-Nikodym derivatives
p for all s E R, and
to have the following properties: (H2a) afk 6 nq21Lq(E;p), for all s E R, and for all q E [l,co) the
function s H llufkll, is locally bounded on R (H2b) For all compact C C R
ds < 03 for p-a.e. z E E. Lm Here ds denotes Lebesgue measure on P.
Choosing appropriate versions by [l, Prop. 2.41 we may always assume that is
jointly measurable in s and z and that (H2b) holds for all z E E (rather than only p-a.e.
z E E ) . For examples of measures p satisfying condition (H2) we refer to 16, Section 31. As
shown in [2] hypothesis (H2) implies (Hl). We need Sobolev spaces with differentiability index higher than 1. Analogue to the
gradient operator V, we define the iterated gradient V2 on F C r by
Assume that -P
(H3) V2 : FCr C L P ( E , p ) + LP(E + H 8 H , p ) is closable for p 2 1. ( 3 . 3 )
182
We define
and the fractional Sobolev spaces H:?' (1 < p < 03, 1 < r < 2) are defined by real
interpolation as follows.
Definition 3.1. For 1 < p < co, 1 < r < 2 we define
H," = (HA", Hi'p)r-l,p (3.5)
where (., .) denotes the real interpolation space, see e.g. [3, 91.
The norm in H,"P is given by the discrete K-method:
where
K(t, f) = inf{llfllll,p + Ellf2/12,", fl + fz = f, f l E Hi,", fz E Hi,"} (3.7)
It follows by a standard interpolation argument that H:" is uniformly convex (see e.g.
[4]) and we know from the denseness of FC," in HixP and [3, Th. 3.4.21 that FC? is dense in H:?'. A combination of these two facts implies that every u E H,'," has a C,,,-quasicontinuous redefinition which we denote by C.
For 1 < p < 00, 1 < r < 2 we define, for a [0, co]-valued lower semi-continuous functions h on E
C,,"(h) := inf{llull:,p; u E H:?',U > h,a.e.},
and for an arbitrary [-co, co]-valued function f on E
C,,"(f) := inf{C,,P(h); h is 1.s.c.and h(z) 2 if(z)I Vz}.
This definition is an extension of the previous one for sets in the sense that for any B c E,
C,,p(B) = C r , p ( l B ) .
The following result is parallel to Shigekawa [8] which is stated for Bessel capacities.
We omit the proof which is the same as in 181. Note that (3.8) is implicit in IS].
Theorem 3.2. Fix q E (1, co), r E (1,2], P E (0,1] and 0 < y < P / q . Then there exists a constant C = C(q, r, P , y) such that I f 5 : [O, TI" x ,!? - R, ( t , z) H & ( z ) is measurable and i f
then {&} has a version {it} such that it is C,,,-quasi-continuous for every t E [0, TI" and
Now we state the main result of this section
183
Theorem 3.3. Suppose (H2) and (H3) hold and let A C E , p > 1. If for any n there edsts a pair (rn, pn) E (1,2] x ( 1 , p ) with r, > np;' + 1 such that
Cr,,p,(A) = 0, \Jn, then Cl,p(A + K ) = 0.
Proposition 3.5 below. First we need a lemma. Since capacities are continuous from below Theorem 3.3 immediately follows from
Fix n E M and define for t E [ -M,MIn and f : E ++ R a function f ( t ) := f ( . + Cy=l tihi).
Lemma 3.4. Fzx n E M and let p > q > 1 and r E (1,2] . Then there exzsts a constant C := C(p, q, r, T ) such that for all f E HOr,'
IIf(t) - f(s)lIl,q < CIlf IIr,pIt - 4 - l (3.9)
for all s, t E [0, TIn.
Proof. By the same argument as in [6], we can prove that for any T > 0, p > 1, q E (0,p) there exists a constant C1 = C,(p, q,T) such that for all f E H;lp and (s, t) E [-T,T]" x [-T, TIn
n n
Now let f E H;' \ (0): By (3.6), there exists a sequence ( f & ~ c H P such that
(3.10)
(3.11)
Choose the unique n E M such that
and the assertion is proved. 0
Proposition 3.5. Let T > 0, n E N, p > 1 and r E (1,2] such that r > np-' + 1. Let q E (1 ,p ) such that r > nq-' + 1. Then there exists a constant C = C(n,p,q,r,T) > 0 such that for any A c W we have
3
G,q(A + M ( T ; hl,. . . , hn)) < C . C&(A) (3.13)
where
184
Proof. Set sh := c r = l s,h,. By changing signs we only need to prove
G,,( u ( A + sh)) < C . Cv?p(A). (3.14)
Let I denote the set of all rational points in [0, TI". If 0 c W is a% opei set then so is 4 sE[O,T]"
U s ~ [ O , T ] " ( O + S h ) and We have
u (0 + sh) = U(0 + sh) sE[O,T]" S E I
Let eo denote the (r,p)-equilibrium potential of 0. Since eo 2 lo, we have
< Cl,,(suP eo(. + th)) tEI
We set f ( t ) := eo(. + th). Applying Lemma 3.4 gives
Ilf(t) - f(s)lll,q < Clleollr,plt - slr-l. (3.15) For y E [0, T - 1 - nq-I), by Theorem 3.2 there exists a C1,,-quasicontinuous modification
{&(.), t E [0, TI"} of {eo(. + th), t E [0, TI"} such that
for some constant C = C(n,p,q,r,T) which may be different from that in (3.15). In
particular, taking y = 0 we obtain
Cl,,(SUP IEt - Ed < lleoll,4,p. s f t
Hence
< (C + 1) ' Ileoll:,p
(C + 1) ' C&(O). =
Thus (3.13) is proved for open sets. For general A, we have'for 0 3 A, 0 open
G,,( u (A+sh)) < C1,,( u (O+sh) ) sE[O,T]" sE[O,T]"
< (C + 1) . C&(O) Consequently,
c1,,( U ( A + sh)) < (C + 1) . inf{Crtp(0), o 3 A ) sE[O,T]"
= (C + 1) . Cr&(A), as desired.
(3.16)
(3.17)
0
185
ACKNOWLEDGEMENT
We would like to thank Terry Jegaraj for spotting a gap in the proof of Lemma 2.1 in the first version of this paper.
Financial support of the BiBoS research center and the DFG through the Research Group “Spectral analysis, asymptotic distributions and stochastic dynamics”, as well as
the DFG and the MPG through the project “SPDE’s with non-Gaussian white noise” is gratefully acknowledged.
REFERENCES
[l] Albeverio, S., Kondratiev, Y. G. and Riickner, M. Ergodicity for the stochastic dynamics of quasi- invariant measures with applications to Gibbs states. J. Funct. Anal. 149 (1997), no. 2, 415-469.
[2] Albeverio, S. and Rockner, M.: Classical Dirichlet forms on topological vector spaces--closability
and a Cameron-Martin formula. J. Funct. Anal. 88 (1990), no. 2, 395-436. [3] Bergh, J.; Lofstrom, J. Interpolation spaces. A n introduction. Grundlehren der Mathematischen
Wissenschaften, No. 223. Springer-Verlag, Berlin-New York, 1576. [4] Hu, J. and Ren, J,: Infinite Dimensional Quasi Continuity, Path Continuity and Ray Continuity of
Functions with Fractional Regularity. J. Math. Pures Appl. 80,1(2001), pp.131-152.
151 Ma, 2-M. and Rockner, M. Introduction to the theory of (nonsymmetric) Dirichlet forms. Universi- text. Springer-Verlag, Berlin, 1992
[6] Ren, J. and Rockner, M.: Ray Holder-continuity for fractional Sobolev spaces in infinite dimensions
and applications, Prob. Theory and Rel. Fields, 117 (ZOOO), n.2, 201-220. [7] Rockner, M. and Schied, A,: Rademacher’s theorem on configuration spaces and applications. J .
Funct. Anal. 169 (1999), no. 2, 325-356. [8] Shigekawa, 1.: Sobolev spaces of Banach-valued functions associated with a Markov process, Probab.
Th. Relat. Fields, 99 (1954), pp. 425-441. 191 Triebel, H.: Interpolation Theory, Function Spaces, Diflerential Operators, North-Holland Publishing
Company, Amsterdam-New York-Oxford, 1978.
186
AN INFINITE DIMENSIONAL LAPLACIAN IN WHITE NOISE THEORY
DEDICATED TO PROFESSOR TAKEYUKI HIDA ON THE OCCASION OF HIS 77TH BIRTHDAY
KIMIAKI SAITO
Department of Mathematics
Meijo University
Nagoya 468-8502, Japan
E-mail: ksaitoOccmfs.meij0-u.ac..jp
The Ldvy Laplacian is formulated as an operator acting on a class in the white noise L2 space consisting of functionals of Gaussian noise and Poisson noise. This class includes regular functionals in terms of Gaussian noise and it is large enough to discuss the stochastic process generated by the Ldvy Laplacian. This formulation is slightly outside of the usual white noise distribution theory, while the Levy Laplacian has been discussed within the framework of white noise analysis. From Cauchy processes an infinite dimensional stochastic process is constructed, of which the generator is the Ldvy Laplacian.
Mathematics Subject CIassifications (2000): 60H40
Introduction
In 1922 P. L6vy l7 introduced an infinite dimensional Laplacian on the space
L2(0, 1). Since then this exotic Laplacian has been studied by many authors from various aspects see [l-6,18,20,23] and references cited therein. In this pa-
per, generalizing the methods developed in the former works [16,19,24,28,29], we construct a new domain of the Lkvy Laplacian acting on some class of functionals of Gaussian noise and Poisson noise, and associated infinite di- mensional stochastic processes.
This paper is organized as follows. In Section 1 we summarize basic el-
ements of white noise theory based on stochastic processes given as linear combinations of Brownian motions and Poisson Processes. In Section 2, fol- lowing the recent works Kuo-Obata-Sait6 16, Obata-Sait8 24,Sait6 30 and
Sait8-Tsoi 31, we formulate the L6vy Laplacian acting on a space consisting
of white noise functionals of linear combinations of Gaussian noise and Pois- son noise, and give an equi-continuous semigroup of class (CO) generated by
the Laplacian. This situation is further generalized in Section 3 by means of a direct integral of Hilbert spaces. The space is enough large to discuss the
stochastic process generated by the Ldvy Laplacian. It also includes regular
187
functionals (in the Gaussian sense) as a harmonic functions in terms of the LCvy Laplacian. In Section 4, based on infinitely many Cauchy processes, we give an infinite dimensional stochastic process generated by the Lkvy Lapla- cian.
1 White noise functionals of Gaussian noise and Poisson noise
Let E = S(R) be the Schwartz space of rapidly decreasing R-valued functions on R. There exists an orthonormal basis {e,},~o of L2(R) contained in E such that
d2 du2
Ae, = 2(v+ l)e,, Y = 0 ,1 ,2 , . . . , A = -- +u2 + 1.
For p E R define a norm 1 * 1, by I f / , = (AP f lp (R) for f E E and let Ep be the completion of E with respect to the norm I . 1,. Then Ep becomes a real separable Hilbert space with the norm 1 . 1, and the dual space E; is identified with E-, by extending the inner product ( a , a ) of L2(R) to a bilinear form on
E-, x E,. It is known that
E = projlimp,,Ep, E* = indlimE-,. ,-+a,
The canonical bilinear form on E* x E is also denoted by (., .). We denote the complexifications of L2(R), E and Ep by L&(R), Ec and Ec,,, respectively.
Let {B(t)}t>o be a Brownian motion and {N(t)}t>o a standard Poisson process which is-independent of the Brownian motion, For any u 2 0 and X E R, we set A,,x(t) = uB(t) + XN(t), t 2 0. Then we have
n
~ [ e ' " * - ~ ~ ( t ) ] = eth(z), h(z ) = --z2 uL + e iX% - 1, t 2 0. 2
Set C([) = exp {JR h ( < ( u ) ) d u } , [ E E. Then by the Bochner-Minlos Theo- rem, there exists a probability measure po,x on E* such that
s,* exp{i(z,E)} dcLu,x(z) = W, E E E .
Let (L2>,,x = L2(E*, p , , ~ ) be the Hilbert space of C-valued square-integrable functions on E* with L2-norm 1 1 . II,,x with respect to p , , ~ . The Wiener-It6 decomposition theorem says that:
n=O
188
where Hn is the space of multiple Wiener integrals of order n E N and HO = C. According to (1.1) each 'p E (L2),,,x is represented as
n=O
where L&(R)63" denotes the n-fold symmetric tensor power of Lg(R) (in the sense of a Hilbert space).
An element of ( L 2 ) , , , ~ is called a white noise functional. We denote by
( (e7 .)),,,A the inner product on ( L 2 ) , , ~ . Then, for 'p and $J E (L2),,,0 we have
where the canonical bilinear form on L&(R)@'" x L&(R)"" is denoted also by
( . 7 .). For cp and $J E (L2)0,x we also have
00 00 00
n=O n=O n=O
The U-transform of cp E (L2),,,x is defined by
Theorem 1.1 26 (see also 9 9 1 4 9 2 2 ) Let F be a complex-ualued function defined on E. Then F is a U-transform of some white noise functional in (L2),, ,~ i f and only if there exists a complex-valued function G defined on E c such that
1) f o r any < and q in E c , the function G(zJ + q) is an entire function of z E c,
2) there exist nonnegative constants K and a such that
IG(t)I I Kexp [alrl:] 7 vt E Ec,
3) F ( [ ) = G(io2( + X(eiAt - 1)) f o r all t E E.
189
2 The LQvy Laplacian acting on white noise functionals
Consider F = U p with 'p E (L2),,x. By Theorem 1.1, for any t,rl E E x E the function z H F([ + zq ) admits a Taylor series expansion:
where F(") ( [ ) : E x . . x E + C is a continuous n-linear functional. Fixing a
finite interval T of R, we take an orthonormal basis c E for L2(T) which is equally dense and uniformly bounded (see e.g. ' * I ' ~ ) . Let V L denote
the set of all 'p E (L2),,x such that the limit
exists for any E E E x E and i ~ ( U ' p ) is in U[(L2),,x]. The LQvy Laplacian
A, is defined by
ALP = U- lZ i~U 'p , 'p E V)t.
Given (I 2 0,X E R, n E N and f E L&(R)6n, we consider 'p E (L2)u,x of the form:
r
The U-transform U'p of 'p is given by
n
where E o , x ( [ ) ( u j ) = ia2[(uj) +XeiXt(uj) . For any (I 2 0, X E R and n E N let E,,x,, denote the space of 'p which admits an expression as in (2.1), where f
belongs to Et" and supp f c T". Then, since the space E , , A , ~ is included in
(L2),,x, we define E,,x,~ by the completion of E u , ~ , n with respect to 1 1 . II,,x. Set E,,x,O = C for any (I 2 0, X E R. Then E,,A,~ is a closed linear subspace of (L2),,x. Using a similar method as in 31, we get the following
Theorem 2.1 31 (see also l6l2') For each (I 2 0 , n E N and X E R the LLvy Laplacian AL becomes a scalar operator on E,,o,,, U Eo,~,, such that
A L ' p = 0 , 'p € E'o,o,n; (2.2)
1 90
Proposition 2.2 31 Let X E R be fixed. Consider two white noise functions of the form:
M M
n=O n=O
If 'p = .IC, in (L2)0,x, then pn = &, for all n E NU (0).
Taking (2.2) and (2.3) into account, we put
For N E N and X E R let D Y be the space of 'p E (L2)o,x which admits an expression
00
'p = C q n , n=l
pn E Eo,x,n,
such that
00
n=l
By the Schwartz inequality we see that D Y is a subspace of (L2)0,x and
becomes a Hilbert space equipped with the new norm 1 1 1 . I I I N , o , x defined in
(2.4). Moreover, in view of the inclusion relations:
(L2)o,x 3 D:)x 3 . . . 3 D Y 2 D$tl 3 . . . ,
we define
00
D%x = projlimN,OO DOJ = n D y . N=l
Note that for any X E R we have
00
IJ Eo,x,n c ~2~ c ( ~ ~ ) o , x . n=l
191
O X Then AL becomes a continuous linear operator defined on D i + l into D$x satisfying
IllALcplllN,O,A 5 IJIcpIIIN+l,O,X, cp E DLA, N E N. (2.5)
Summing up, we have the following
Theorem 2.3 l 6 t 3 l The operator AL is a self-adjoint operator densely defined in D$x for each N E N and X E R.
It follows from (2.5) that AL is a continuous linear operator on D%x. In
view of the action of (2.3), for each t 2 0 and X E R we consider an operator G; on DLx defined by
00 00
n=l n=l
We also define G: on ( L 2 ) g , ~ as an identity operator I by Icp = cp, cp E
(L2 )u,o.
Theorem 2.4 16,30 Let X E R. Then the family of operators {G; ; t 2 0 ) on DZA is an equi-continuous semigroup of class ((70) of which the infinitesimal generator is A,.
3 The LQvy Laplacian acting on WNF-valued functions
Let dv(X) be a finite Bore1 measure on R satisfying
Fix N E N. Let DO, be the space of (equivalent classes of) measurable vector
functions cp = (pX) with cpx = C,"=l cp; E D Y for all X E R \ {0}, and
cpo E (L2),,0, such that
Then !DO, becomes a Hilbert space with the norm given in (3.1).
Take a functional
192
j = ( j l , . . . ,jn) E (N u { O } Y , where an orthonormal basis {[n}r=o c E for L2(T) which is equally dense
and uniformly bounded. Since a linear span of { 'p t f ) ;n E N u {O} , j =
(j l , . . . , jn) E (N U { O } ) n } is dense in D Y and a linear span of {cpzf'; n E
N U { 0 } , j = (jl, . . . , jn) E (N U {O})n} is dense in (L2),,,0, the spaces 9% is nothing but a direct integral Hilbert space:
Let 9: be the space of white noise functionals 'po + C,"==, sR\{o} q: dv(X)
with C,"=l 'p: E (L2)0,x for all X E R \ {0}, and 'po E ( J ~ ' ) ~ , O , such that
Then 9; also becomes a Hilbert space with the norm given in (3.2).
Proposition 3.1 The map
is a continuous linear map and a bijection from 9% into 9;.
In view of the natural inclusion: 9%+1 c 9% for N E N, which is obvious from construction, we define
00
9.0, = projlimN+009G = n 9%. N=l
The LQvy Laplacian AL is defined on the space 9& by
ALcp = (AL'pX),
Gtcp = ( G W ) ,
cp = ((PA) E 9:.
cp = ((PA) E 9.0,.
Then A, is a continuous linear operator from D.", into itseIf. Similarly, for t 1 0 we define
Then we have the following:
Theorem 3.2 The family {Gt ; t 2 0) is an equi-continuous semigroup of
class (CO) on D.", whose generator is given by AL.
193
4 An infinite dimensional stochastic process associated with the LBvy Laplacian
For p E R let EF be the linear space of all functions X H Ex E E,, X E R, which are strongly measurable. An element of EF is denoted by E = ( E A ) A ~ R . Equipped with the metric given by
the space EF becomes a complete metric space. Similarly, let CR denote the linear space of all measurable function X H zx E C equipped with the metric
defined by
Then CR is also a complete metric space.
ER = projlim,,, EF.
by
for any cp = ((P’)X€R E 92. The space U[!D&] is endowed with the topology
induced from D& by the U-transform. Then the U-transform becomes a
homeomorphism from D& onto U[D&]. The transform Ucp of cp E 9: is a
continuous operator from ER into CR. We denote the operator by the same notation U p .
In view of d, 5 d, for p 2 q, we introduce the projective limit space
The U-transform can be extended to a continuous linear operator on D&
UP(() = (U(pX(&+))XR, E = (EX)XER E ER,
Let Et be an operator defined on U[DZ] by
Et = UGtU-I, t 2 0.
Then by Theorem 3.3, {Ct; t 2 0) is an equi-continuous semigroup of class
(Co) generated by the operator LL. Let {Xi}, j = 1,2 , be independent Cauchy processes with t running over
[0, oo), of which the characteristic functions are given by
E[ei”xi] = e-tlzl , Z E R , j = 1 , 2 .
Take a smooth function VT E E with VT = 1/ITI on T. Set
XitqT if X 2 0,
X ? , , T ~ , otherwise. =
194
Define an infinite dimensional stochastic process {Yt; t 2 0 } starting at < = ( 6 X ) X E R E E R by
Yt = ( E X + q X ) X E R , t 2 0.
Then this is an ER-valued stochastic process and we have the following
Theorem 4.1 If F is the U-transform of an element in DZ, we have - GtF(S) = E[F(Yt ) lYo = < I , t 2 0. (4.1)
PROOF. We first consider the case when F E U[D&] is given by
F(<) = (FX(&))XER, Fo E u[(L2)c,0],
with f E E$". Then we have
X E R - = ( G;FX ( h ) ) X E R - = GtF(<).
Next let F = (FoGx,o -t Cz=l F,")XER E U[9:] . Then for v-almost all X E R
and for any n E N, F," is expressed in the following form:
n
F , X ( J X ) = lim A" N-CC
Since Fo E U [ ( L 2 ) c , ~ ] and F," E U[DLX], there exist 'po E ( L 2 ) o , ~ and E DZX such that Fo = U[p0] and F," = U['pA] for v-almost all X and each n. By the Schwarz inequality, we see that
195
n=l W
where 'pcx = C(&,)-lei('*~X) for v-almost all X E R and each N E N. Therefore
by the continuity of G;, X E R, we get that -
/ W \
- = GtF(5).
Thus we obtain the assertion. I
Acknowledgments
This work was written based on a talk at International Conference on Stochas- tic Analysis: Classical and Quantum - Perspectives of White Noise Theory. This work was partially supported by JSPS grant 15540141. The author is grateful for the support.
References
1. L. Accardi and V. Bogachev: The Ornstein-Uhlenbeck process associated with the LLvy Laplacian and its Dirichlet form, Prob. Math. Stat. 17 (1997), 95-114.
196
2. L. Accardi, P. Gibilisco and I. V. Volovich: Yang-Mills gauge fields as harmonic functions for the Lkvy Laplacian, Russ. J. Math. Phys. 2
3. L. Accardi and 0. G. Smolyanov: Trace formulae for Levy-Gaussian mea- sures and their application, in “Mathematical Approach to Fluctuations Vol. I1 (T. Hida, Ed.),” pp. 31-47, World Scientific, 1995.
4. D. M. Chung, U. C. Ji and K. Sait6: Cauchy problems associated with the Le‘vy Laplacian in white noise analysis, Infin. Dimen. Anal. Quantum Probab. Rel. Top. 2 (1999), 131-153.
5. M. N. Feller: Infinite-dimensional elliptic equations and operators of Lkuy type, Russ. Math. Surveys 41 (1986), 119-170.
6. K. Hasegawa: Le‘vy’s functional analysis in terms of an infinite dimen- sional Brownian motion I , Osaka J. Math. 19 (1982), 405-428.
7. T . Hida: “Analysis of Brownian Functionals,” Carleton Math. Lect. Notes, No. 13, Carleton University, Ottawa, 1975.
8. T. Hida: A role of the L h y Laplacian in the causal calculus of generalized white noise functionals, in “Stochastic Processes,” pp. 131-139, Springer-
Verlag, 1993. 9. T. Hida, H.-H. Kuo, J . Potthoff and L. Streit: “White Noise: An Infinite
Dimensional Calculus,” Kluwer Academic, 1993. White noise analysis and the LLvy Laplacian,
in “Stochastic Processes in Physics and Engineering (S. Albeverio et al.
11. E. Hille and R. S. Phillips: “Functional Analysis and Semi-Groups,” AMS Colloq. Publ. Vol. 31, Amer. Math. SOC, 1957.
12. I. Kubo and S. Takenaka: Calculus on Gaussian white noise I-IV, Proc. Japan Acad. 56A (1980) 376-380; 56A (1980) 411-416; 57A (1981)
13. H.-H. Kuo: O n Laplacian operators of generalized Brownian functionals, Lect. Notes in Math. Vol. 1203, pp. 119-128, Springer-Verlag, 1986.
14. H.-H. Kuo: “White Noise Distribution Theory,” CRC Press, 1996. 15. H.-H. Kuo, N. Obata and K. Sait6: Lkuvy Laplacian of generalized func-
tions on a nuclear space, J. Funct. Anal. 94 (1990), 74-92. 16. H.-H. Kuo, N. Obata and K. Sait6: Diagonalization of the Lkuy Laplacian
and related stable processes, Infin. Dimen. Anal. Quantum Probab. Rel.
17. P. LQvy: “Lecons d’Analyse Fonctionnelle,” Gauthier-Villars, Paris, 1922. 18. R. Lkandre and I. A. Volovich: The stochastic Lkvy Laplacian and Yang-
Mills equation on manifolds, Infin. Dimen. Anal. Quantum Probab. Rel.
(1994), 235-250.
10. T. Hida and K. Sait6:
Eds.),” pp. 177-184, 1988.
433-436; 58A (1982) 186-189.
TOP. 5 (2002), 317-331.
TOP. 4 (2001) 161-172.
197
19. K. Nishi and K. Sait6: An infinite dimensional stochastic process and the Lkvy Laplacian acting on WND-valued functions, to appear in “Quantum
Inofrmation and Complexity” World Scientific, 2004.
20. K. Nishi, K. Sait6 and A. H. Tsoi: A stochastic expression of a semi- group generated by the Lkvy Laplacian, in “Quantum Information I11 (T. Hida and K. Sait6, Eds.),” pp. 105-117, World Scientific, 2000.
21. N. Obata: A characterization of the Lkvy Laplacian in terms of infinite dimensional rotation groups, Nagoya Math. J. 118 (1990), 111-132.
22. N. Obata: “White Noise Calculus and Fock Space,” Lect. Notes in Math.
Vol. 1577, Springer-Verlag, 1994.
23. N. Obata: Quadratic quantum white noises and Lkvy Laplacian, Nonlin-
ear Analysis 47 (2001), 2437-2448.
24. N. Obata and K. Sait6: Cauchy processes and the Lkvy Laplacian, Quan-
tum Probability and White Noise Analysis 16 (2002), 360-373.
25. E. M. Polishchuk: “Continual Means and Boundary Value Problems in Function Spaces,” Birkhauser, Basel/Boston/Berlin, 1988.
26. J. Potthoff and L. Streit: A characterization of Hida distributions, J. Funct. Anal. 101 (1991), 212-229.
27. K. Sait6: Its’s formula and Lkvy’s Laplacian I, Nagoya Math. J. 108
(1987), 67-76; 11, ibid. 123 (1991), 153-169.
28. K. Sait6: A (Co)-group generated by the Ldvy Laplacian II, Infin. Dimen.
Anal. Quantum Probab. Rel. Top. 1 (1998) 425-437.
29. K. Sait6: A stochastic process generated by the Lkvy Laplacian, Acta
Appl. Math. 63 (2000), 363-373. 30. K. Sait6: The Lkvy Laplacian and stable processes, Chaos, Solitons and
Fractals 12 (2001), 2865-2872. 31. K. Saitb and A. H. Tsoi: The Lkvy Laplacian as a self-adjoint operator,
in “Quantum Information (T. Hida and K. Sait6, Eds.),” pp. 159-171,
World Scientific, 1999. 32. K. Sait6 and A. H. Tsoi: The Lkvy Laplacian acting on Poisson noise
functionals, Infin. Dimen. Anal. Quantum Probab. Rel. Top. 2 (1999),
33. K. Sait6 and A. H. Tsoi: Stochastic processes generated by functions of the Lkvy Laplacian, in “Quantum Information I1 (T. Hida and K. Sait6,
Eds.),” pp. 183-194, World Scientific, 2000.
503-510.
34. K. Yosida: “Functional Analysis (3rd Edition) ,” Springer-Verlag, 1971.
198
INVARIANCE OF POISSON NOISE
SI SI
Graduate School of Information Science and Technology Aichi Prefectural University
Nagakute, Aichi-ken 480-1198, Japan
ALLANUS TSOI
Department of Mathematics University of Missouri
Columbia, MO 65211, U.S.A.
WIN WIN HTAY
Department of Computational Mathematics University of Computer Studies
Yangon, Myanmar
Dedicated to Professor Talceyulci Hida on the occasion of his 77th birthday
AMS Mathematics Subject Classification 2000 : 60H40
We discuss some transformation groups under which Poisson noise is kept invariant,
so that we can give a characterization. The idea is similar to the case of Gaussian
white noise, however the actual transformation group is quite different. In fact, symmetric group plays important role in the case of a Poisson noise.
1. Introduction
Our main purpose is to find out a characterization of Poison noise from the
viewpoint of white noise analysis. For this purpose we refer to the char-
acterization of white noise, using the infinite dimensional rotation group
O(E). See e.g. [5]. We are therefore suggested to find a suitable trans-
formation group acting on the sample function space of a Poisson noise.
199
Symmetric groups of all orders are fitting for our purpose. In addition,
some other groups that give invariance of Poisson noise tell us finer prop-
erties of Poissson noise and Poisson process.
2. Preliminaries
Introduce a Poisson noise by taking the parameter space to be Rd. A
construction is as follows. Let El and E2 be suitably chosen Hilbert spaces
that are topologized by Hilbertian norms 1 1 111 and ( 1 . 112, respectively.
Assume that 11 5 1 1 . 112 and that there exist consistent injections Ti,i =
1 , 2 , such that Ti from Ei to Ei-1 for i = 1 ,2 , is of Hilbert-Schmidt type,
where EO = L2(Rd). Then, we have inclusions of the form
E2 C El c Eo c E; c El. (2.1)
Recall that the characteristic functional of Poisson noise with Rd- parameter is given by
where we take unit intensity.
It can be proved that C$ ( I ) is continuous on El. It is positive definite
and Cg(0) = 1. The Bochner Minlos theorem guarantees the existence of
a probability measure p p on Ei such that its characteristic functional is
equal to the given C$(<). Thus a Poisson noise measure space (E;, p p ) is
given.
To fix the idea, let the time parameter space be a compact set, say
I = [0,1]. In this case, C;(<) is continuous in EO c L2( I ) , so that there is
a Gklfand triple of the form
El c L2(1) Eo c E;. (2.2)
A Poisson measure is now introduced on the space E;.
Define P( t ,z ) = (z,xpt]),O 5 t 5 1, z E Ei, by a stochastic bilinear
form. Then, P(t, x) is a Poisson process with parameter set [0,1].
Let A, be the event on which there are n jump points over the time
interval I. That is
A, = {z E E; ;P( l ,z ) = n}, (2.3)
200
where n is any non-negative integer.
Then, the collection {A,, n 2 0) is a partition of the entire space E;. Namely, up to measure 0, the following relations hold:
A n n A , = d , n # m ; U A n = E ; . (2.4)
The conditional probability on A, is
For C c Ak, the probability measure is defied as
on a probability measure space (Ak, Bk, &), where Bk is the sigma field
generated by measurable subsets of Ak, determined by P(t , x).
For k = 0, the measure space is (Ao, Bo, &) is trivial, where
Bo = {d,Ao} mod & and &(A) = 1.
We now recall a general notions about a probability measure space.
Definition A probability space (0, B , P ) is a Lebesgue space if
1) I = {I,} is a countable base such that
B = o{I, ;n E 2).
2) Let A, = I , or 1:. Then n,A, # 4. 3) For 5 # y, there exists I,, I , such that I,nI, = 4 , 5 E I,, y E I,.
It is known that if (R, B , P ) be a Lebesgue space then it is isomorphic to
([O, a] , Leb) u {atoms}.
Lemma 1. On a Lebguse space a measure preserving set transformation
implies a measure preserving point transformation.
Coming back to a Poisson noise with parameter set [0,1].
Proposition 1. (Ak, Bk, &), k 2 1 is a Lebesgue space without atom.
201
Let x E An,n 2 1, and let 7i = ~i(x),i = 1 ,2 ,..., n, be the order
statistics of jump points of P(t) :
0 = 70 < 7-1 < . . . < 7, < 7,+1 = 1.
(The 7's are strictly increasing almost surely.) Set
xi(.) = 7i(X) - Ti-i(X),
so that
n+l cxi = 1. 1
Proposition 2. On the space A,, the (conditional) probability distribu-
tion of the random vector (X1,Xz, ..., X,+l) is uniform on the simplex
n+l
j= l
Cororally The probability distribution function of each Xj is
1 - (1 - U ) n , 0 I u I 1.
Proposition 3. The conditional characteristic functional
Cp,,(E) = E[ei (PIE) IAn]
is obtained as
3. Invariance of p p under transformations on parameter
We are going to find a transformation group which is acting on the Poisson
noise space (E* , ,up) and which keeps the ,up invariant.
x E E*, of B(t) , such as
space
Note that in Gaussian case the actions are taken on sample functions
g* : x 4 g*x.
202
For the case of Poisson noise, in contrast with the Gaussian case, we
begin with measure preseving set transform, then we come to transforma-
tions of sample functions. Also, it is noted that our probabilistic approach
is also different.
With the study of invariance of Poisson noise measure we aim at har-
monic analysis of functionals of Poisson noise, where the invariance of Pois-
son noise will play interesting roles. In particular, the family of symmetric
groups acting on the conditional probability space is a characteristic of
Poisson noise.
There are two ways to introduce such transformation group. One uses
the vector space structure of E* with the help of a base of E*. Another way
comes from the transformation acting on the parameter space. This means
that the second method, unlike the first method, depends on the geometric
structure of the space in question.
3.1. Transformations in the Rd-parameter case
We give illustrative examples in the followings.
Class 1. Use of an orthogonal system
Let {Vn} be a system of unit cubes such that UV, = Rd and that Vn's are disjoint except boundaries.
~ ( t ) = C E n ( t ) , a.e.7 E E (3.1) n
where En(t) = E ( ~ ) x v , ( ~ ) . Let 7r be permutation of finitely many n. and gn be the transformation
such that 00
(g7rE)(t) = CElr(n)(t) E L2(R"). (3.2) -m
Then we can see that the characteristic functional of Poisson noise sat-
isfies
Hence the two characteristic functionals define the same Poisson measure.
203
Case 2. Group of motions
We can easily see that the time shift leaves the probability distribution
p p of a Poisson noise invariant, since
cp(s,"<) = c p ( < ) , (st<)(u) = <(u- tek) , E Rd, (3.3)
where ek is the unit vector in the k-th direction of Rd.
Obviously, orthogonal group O ( d ) acting on the parameter space Rd
Thus we have
presents invariance of Poisson measure.
Proposition 4. Poisson noise is invariant under
The probability distribution p p of an Rd-parameter
1) the rotation group S O ( d ) acting on Rd, and
2) the shifts.
Note. Interesting property can be seen in the dilation of Rd-vectors:
rt : u -.+ ueat, u E Rd, a > 0. (3.4)
Then the characteristic functional Cp(c) changes to
Namely, the intensity changes from X to Xe-dat, but remains to be the
Poisson distribution.
3.2. Td-parameter case
Automorphisms of Td
Take the parameter space to be a compact domain, say T d , d - dimensional Torus instead of Rd in (2.1). Then we have a corresponding
GBlfand triple
El c L2(Td) = Eo c E;. (3.6)
The characteristic functional of Poisson noise with Td parameter is
204
In particular, if we take d = 1 then Td is S1 , a circle and the character-
istic functional is
where the intensity X is taken to be 1.
Let r be a transformation defined by
(r)(t) = 2t(mod 2n), t E S1,
and define the automorphism a of E by
(a<)(u) = t(7(u))*
CP,1(05) = CP,1(5).
We can see that
It means that the characteristic functional is a-invariant and then measure
p p is a*-invariant.
Proposition 5. The mapping 0 is an automorphism of El and is contin-
uous. There esists a* acting on EF such that
and
D*PP = PP
u n * p p = p p .
4. Transformations that preserve conditional Poisson measures
Let Sn+l = {7rn(i), 1 5 i 5 (n + l)!} be the symmetric group and B be the
set of Bore1 subsets of simplex An+, where
n+l
An+, = {(x11x21*.* ,xn+ll Cxi = 110 I xi 5 1). 1
Let
X = (X1,X21... lxn)
205
and set
X7rn(i) = (xTy(i),XT;(i), .*‘,xT_n+l(i)),
where 7rn(i) = ( ~ r ( i ) , 7 r g ( i ) , . . . ,~:+~(i)) E &+I. Let g7rn(i) be the set
transformation on E* such that
g T n ( i ) : x-’(B) -+ X$(il(B), B E B(&+i).
(X-l(B),X;:(i)(B) c A,.) We can see that
P;(X-l(B)) = P;(x;:(i)(B)),
and gTn( i ) defines an automorphism of (A,, p;).
gTn( i ) on A,, n = 1 , 2 , . . . , such that
Then an automorphism gT( i ) on E* is defined by the automorphisms
gT(i)a: = g , p ( i ) Z , z E A,, n = 0 ,1 ,2 , ....
The space ( A k , Bk, p$) is a Lebesgue space by Proposition 1. Thus
by Lemma 1, gTk( i ) defines a measure preserving point transformation on
( A k , Bk, p$) . The same for g r ( i ) . Then we have
Proposition 6. The transformation g T ( i ) is a pp-measure preserving point
transformation on E*.
Theorem 1. The group G defines a symmetry of Poisson noise.
Applications to unitary representation can be given to develop a harmonic
analysis, and a comparison with the Gaussian case will be discussed in the
forthcoming paper by the author.
Remark We can play the same game if the parameter set is [ O , l l d .
206
5. A characterization of Poisson noise
Start with a probability space (G,,P,) and let {Y,(t ,w),w E G,} be a
stochastic process on (G,, P,), such that
n
Yn(t,~)=C&~(~)(t), tE [O, l ] , O = t o < t l < . . . < t , = l . (5.1) 1
Theorem 2. Assume that
(1) the joint probability distribution of n singular points t j ( w ) , j =
(2) {Y,(t, w ) } and {Y,(T(~), w ) } have the same probability distribution,
1, ..n, are absolutely continuous,
where T is the transformation defined as
~ ( t ) = 2t (mod 1). (5.2)
Then the probability distribution of singular points of Yu(t, w ) on the space
G, is the uniform distribution.
(Note that in Section 3.2, T is defined by mod 27r. Now we use mod 1.)
Proof. Let p(t1, t 2 , . . , t d ) be the joint probability distribution of singular
points t l , t 2 , + , t d . Use the transformation T , for all of t l , . . . , t d , for which
p(t1,. . . , t d ) is invariant, then we have
After N iteration, letting N tend to infinity we have
P ( t l , . . . , t d ) = i l i l P ( U l ,... ,Ud)dUl’..dW,
Thus the assertion is valid. The proof is the same for any n > 1.
Proposition 7. The characteristic functional C,(E) is invariance under the
transformation G, where CJ is defined by (a()(u) = <((.‘(u)) in which 7‘ be
a transformation defined by l ( t ) = (T(tl), . . . , ~ ( t d ) ) for t = (t l . . . . , t d ) E
[0, lId such that
~ ( t j ) = 2tj(mod l), t j E [0,1].
207
Theorem 3. space G, with the parameter space (time interval) [0,1], is
The characteristic functional of Yn(u), on the probability
(5.4)
We now consider the parameter space [0, TI instead of [0,1], at the aim
of extending the parameter space to [ O , o o ) .
Corollary The characteristic functional of Yn (t) , on the probability space
G, with the parameter space [0, TI, is expressed in the form
Set R = En G,. Let us define a stochastic process Y( t ) on (0, P ) such
that
In order to have Y( t ) well defined, it is necessary to determine the
probabdity of G,.
Let P(Gn) = en. The characteristic functional functional C(<;O,s) of
Y( t ) on R, with the parameter space [0, s], is
since Cn([; 0, S ) = E[eic(tJGn]. From the above assumption (ii), Y( t ) is additive then we have
C(<; 0, s1 + s2) = C(<; 0, s1)C(I; S1,Sl + s2). (5.7)
The relation (5.5), (5.6) and (5.7) yields
208
is additive and Levy process.
Consequently, we have
It is written as
where
Equating the coefficient of a1bn-', we have
n!dn(sl + SZ) = ( l !d l (s l ) ) ( ( ~ - L ) ! & - ~ ( S Z ) ) *
By letting h k = k!dk( t ) , we obtain that ho = 1,hk = Xk,k = 1 , 2 , . . . , where X is a constant, Hence, we are given
Since
bution. Thus we have
Ck = e-xt, we have to normalize them to be a probability distri-
That is, Y( t ) has a Poisson distribution.
Theorem 4. noise.
Under the above assumptions (i) and (ii), Y( t ) is a Poisson
Proof. From (5.6) and (5.8) the characteristic functional of Y( t ) is
209
References
1, L. Accardi, T. Hida and Si Si, Innovations for some stochastic processes. Volterra Center Notes, 2002.
2, T. Hida and Si Si, Innovation approach to some random fields, Application
of white noise theory, World Scientific (2004). 3, T. Hida , Canonical representation of Gaussian processes and their applica-
tions, Memoirs Coll. Sc., Kyoto A33 (1960) 109-155 4. T. Hida, Stationary Stochastic Processes, Math. Notes, Princeton Univ.
Press. 1970. 5. T. Hida, Brownian motion, Springer-Verlag, 1980. 6. P. Lkvy, Fonctions aleatoires A corrklation linebire, Illinois J. of Math. 1
7. S. Mataramvura, B. Oksendal and F. Proske, The Donsker delta fuction of a Lkvy process with applicatio to chaos expansion of local time, Annales de
Institut H. Poincark, Elsevere SAS, 40 (2004) 553-567. 8. V. A. Rohlin, On the fundamental ideas of measure theory, AMS Transaction
series 1. vol. 10 (1962), 1-54. 9. Si Si, Effective Determination of Poisson noise, Infinite Dimensional Analysis
and Quantum Information vo1.6. Number 4, World Scientific Pub. Co. (2003),
10. Si Si, Note on Poisson noise, Quantum Information and Complexity, World
Scientific 2004, 411-425. 11. Si Si and Win Win Htay, Structure of Linear processes, Proceeding of Inter-
national conference on Quantum Information, (to appear).
12. Win Win Htay, Note on Linear process, Quantum Information and Complex-
ity, World Scientific 2004, 449-455.
(1957), 217-258.
609-617.
210
NONEQUILIBRIUM STEADY STATES WITH BOSE-EINSTEIN CONDENSATES
S. TASAKI* AND T. MATSUIt
*Advanced Institute for Complex Systems and Department of Applied Physics, Waseda University, 3-4-1 Okubo, Shinjuku-ku, Tokyo 169-8555, Japan
Graduate School of Mathematics, Kyushu University, 1-10-6 Hakozaki, Fukuoka 812-8581, Japan
Nonequilibrium steady states (NESS) of bosonic system with Bose-Einstein con-
densate are investigated with the aid of the C*-algebraic method. The system con- sists of two free bosonic reservoirs coupled with each other. Initially the reservoirs are prepared to be in equilibrium with different temperatures and local densities. NESS are constructed as the t + +a limits of such initial states. Josephson
currents are studied as well.
1 Introduction
The understanding of irreversible phenomena including nonequilibrium steady
states is a longstanding problem of statistical mechanics. Various theories
have been developed so far and one of promising approaches deals with in-
finitely extended dynamical systems1 ,2. Not only equilibrium properties, but also nonequilibrium properties has been rigorously investigated. Those include
analytical studies of nonequilibrium steady states of harmonic crystal^^)^, a
one-dimensional gas5, unharmonic chains6, an isotropic XY-chain7, a one-
dimensional quantum conductor', systems with asymptotic a b e l i a n n e san interacting fermion-spin systemll, fermionic junction systems12, a quasi-
spin model of superconductors13 and a bosonic junction system without the
Bose-Einstein condensate14. Entropy production has been rigorously studied as well (see [9,15-191 and the references therein).
In this article, we add more examples of nonequilibrium steady states
(NESS), namely bosonic systems with Bose-Einstein condensate. The paper is arranged as follows: In the next section, the a bosonic junction system is described. In Sec. 3, its NESS are constructed and their properties are studied. In Secs. 4 and 5, NESS for free Bosons on lattice with different joints
are investigated. The Josephson currents at such NESS are studied as well.
2 Bosonic Junction System
In this and the next sections, we consider a bosonic version of a typical junc- tion system studied in [la]. The model is defined on a tensor product of a
211
Hilbert space L 2 ( R ) and two Fock spaces ‘H, (n = 1,2), both of which are
constructed from L2(R3) : ‘H = L2(R)@’H1@’H2. The spaces ‘Hn (n = 1,2) de- scribe the two reservoirs and L 2 ( R ) describes the junction between the reser- voirs. In terms of standard annihilation operators a, an,k satisfying canonical
commutation relations:
[‘&at] = 1, [an,k, a A t , k , ] = sn,n’d(k - k’) 7 (2.1) the Hamiltonian is formally given by
where (h.c.) means Hermitian conjugate, W k = lkI2/2, k-integrations are taken over R3 and gn(k) E L2(R3) (n = 1,2). Strictly speaking, the free field parts
s dkwkaL,kan,k (n = 1,2) should be understood as the second quantization of
the multiplicative self-adjoint operator p(k) + wkp(k) defined on L 2 ( R 3 ) to
the Fock spaces ‘Hn (n = 1,2), respectively.
The Hilbert space ‘H is a Boson Fock space over a Hilbert space
equipped with the inner product
The CCR algebra A, which corresponds to a set of observables, is generated by Weyl operators (cf. Theorem 5.2.8 of [2] ):
W(f) = exP (i@(f)) (2.5)
where @(f) = @s(f) + xn=1,2 @ b , n ( f ) is a linear map from the Hilbert space
L 2 ( 3 f ) to the space (3 @(f)) of unbounded operators on 7-l with
In the above, the overline stands for the closure. On A, the Hamiltonian H generates a time evolution automorphism Tt : A + A.
In order to specify the states over A, it is enough to investigate the
average values of the Weyl operators. Indeed, any element of A can be approximated with arbitrary precision by a finite linear combination of fi- nite products of Weyl operators. And a product of Weyl operators reduces
212
to a single Weyl operator because of the identity (Theorem 5.2.4 of [2]):
We assume that the two fields and the junction are initially decoupled and, in the initial states WO, the two reservoirs are in equilibrium with different temperatures pF1, and densities p l , p2. Nonequlibrium steady states are constructed as t -+ 03 limits of the state wo o ~t at time t. Since we have
studied the case where both initial reservoir states are normal14, we consider the case where at least one of the initial reservoir states has condensate.
Equilibrium states with Bose-Einstein condensate were firstly constructed by Araki-Woods2' and were investgated further by Cannon21 and Lewis- Pu1&22. In these approaches, the states are prepared by taking an appropriate infinite volume limit. Starting from a finite cubic box of volume V where N particles are contained and obey the periodic boundary condition, the canoni- cal average of the Weyl operator is shown to converge to a well-defined limit in a thermodynamic limit where V -+ 00 and N -+ 00 while keeping the density p = N / V constant20>21'22. When p exceeds a critical value, the condensate appears and the thermodynamic limit of the average of the Weyl operator acquires a factor represented by the 0th order Bessel function Jo(z). With the aid of the formula Jo(z) = J'T, ge izs in (e+a) , the thermodynamic limit of the canonical state is shown to be decomposed into gauge noninvariant pure equilibrium states where the phase of the condensate wave function is fixed. According to these results, the average of the reservoir Weyl operator at an initial state wo can be taken as
W.f)W(g) = W(f 3- 9) exp (-W.f, 9)).
When the density pn is less than the critical value pc(,L?n) / & epn2k -I,
On(+,) = 0 , where the chemical potential p, is a unique solution of
1 Pn = J &ePn(y*-Pn) - 1 *
On the other hand, when pn 2 pc(Pn) , one has
where _= ,on -pc(,&) and arg an are the density and the phase of the con- densate, respectively. Note that (2.6) is meaningful only for Weyl operators, where +,(k) (n = 1,2) are continuous.
213
3 Nonequilibrium Steady States
To construct NESS as t -+ 00 limits of w o o q, we assume the followings:
(A) The initial state satisfies
(wo(ab1abz . . .&)I 5 n!K, (3.1)
where abj = a or a+ and IS,(> 0) satisfies limn+m K,+l/K, = 0.
(B) The form factor g n ( k ) is a continuous function of (kJ and C?(R3). Then
is in L1(R+) and uniformly Holder continuous with index 1/2.
(C) There exists no real solution for q ( z ) = 0, where
(3.3)
and l / q - ( w ) = l / q ( w - i0) is bounded.
The assumptions (A) and (B) are posed to simplify the investigation, while
(C) plays an essential role as it guarantees the existence of the steady states.
As mentioned before, time evolution of the states is fully specified by the
behavior of the Weyl operators wg o Tt(W(f)). Since the Hilbert space 3-1 is a Boson Fock space over the Hilbert space 12, the bilinear Hamiltonian H is the second quantization of the Hamiltonian h densely defined on k?:
(3.4) 1 . flc + Cn=1,2 .f dkgn(k)*&(k)
h f = h ( & ) = ( "- 'k$l(k) -k Agl(k)C wk$'Z(k) + Ag2(k)c
The group Tt of time-evolution automorphisms generated by H satisfies (cf.
the argument before Proposition 5.2.27 of [2] )
Tt (W(.f)) = W ( e Y ) 7 (3.5)
and, under the condition (C), one has
214
Then, since the two fields and the junction system are independent at the
initial state, the average value of the Weyl operator at time t is evaluated as
(3.7) wo 0 Tt ( ~ ( f ) ) = wo ( ~ X P ( i + ~ s ( e Z ~ ~ f ) ) ) n wo (exp ( i+b,n(eiht f ) ) ) n=1,2
By evaluating each factor of the above equation, one obtains
Theorem 3.1 For the Weyl operator where each & ( k ) is Cr(R3), we have
where A(f) and A(f) are defined by
(3.10)
This implies that the steady state w+ exists and that it is quasi-free with
(3.12)
I n the above, the convention +n = 0 f o r pn 5 pc (pn ) i s used. Under the time evolution Tt , mass is conserved. Namely, one has
where Jn (n = 1 , 2 ) are mass flows from the reservoirs:
Jn = iX / dk (gn(k)aA,ka - gn(k)*atUn,k)
(3.14)
(3.15)
And, as a straightforward corollary, one has
215
Corollary 3.2 The average mass flow at NESS is a sum of the normal and Josephson currents:
The rest of this section is devoted to the proof o f the proposition. Firstly, by the same arguments as in [14] , one can show
lim wo (exp ( i+.s(eihtf))) = 1 t-00
Next we consider the quadratic part of log wo (exp ( i@b,n(e ih t f ) ) ) :
where
The time-dependent terms IL”(t) (n = 1,2 : j = 1,2) are shown to vanish in the limit o f t + +co with the aid of the following Lemma.
Lemma 3.3 Let W ( W ) be uniformly Holder continuous with index 1/2, i.e., ]v(w’) - v(w)l 5 Klw’ - w11/2, v(0) = 0, suppv c [0, R], then the integral
@ I ) ,i (w’ -w) t
w - WI - io J ( w , t ) = dw’
satisfies, for an arbitrary R’ > 0, sup J J ( w , t ) J -+ 0 (t + +co) . O<w<R’
Indeed, I ( w ; t ) can be rewritten as
00 v(wl)ei(w’-w)t
I ( w ; t ) = dw‘ w - w’ - io
216
where w(w’) satisfies all the assumptions of Lemma because gn
the uniformly Holder continuity ofl/q+(w) and s d&,(k, f) .
lim sup II(w;t)l = 0 (‘R’ > 0) ~-+CC OswSR‘
E Cr(R3) and
Thus,
(3.17)
On the other hand, in terms of u,(w) = g;(&) [ J d i cp,(k, f)]
R’ such that supp gz c [0, R’ ] , one has
and Ikl=&
which leads to
The integral converges as the integrand may have, at most, a 1/&-singularity at the origin and, thus, I,?’(t) -+ 0 as t + 00. Similarly, limt+cC I , (2) ( t ) = 0.
Finally, we consider the linear part of logwo (exp ( i@h,,(eihtf))) :
where
s;, o+(w,) (rn) [ ~ d k ’ c p n , ( k ’ , f ) l l k , = ~ is continuous. As is 47 and G(W’) = En, integrable, Riemann-Lebesgue theorem23 implies S$J,(O, t ) + 0 (t -+ GO).
4 Boson on Lattice
In this section, we consider the free Boson on the three dimensional lattice Z3. Let aj and a; ( j in Z”) be the creation and annihilation operators satisfying [a j , a;] = 6jk. Smeared Boson operators, a * ( f ) and a ( f ) , are defined as
217
where f j is a rapidly decreasing function on Z3. fj is rapidly decreasing if and
only if its Fourier transform f(k) is smooth( infinitely many differentiable). Set
We denote d(Z3) by the Weyl CCR C*-algebra generated by W(f). Next we consider equilibrium states above the critical density. For our
purpose, it is convenient to consider the Bosons on a general infinite graph.
Let I? = {V,E} be a graph where V is the set of vertices and E is the set of edges. We suppose that the Graph r is connected and it is a union of finite connected graphs rn , r = UF=lFn. The standard discrete Laplacian hr on
r is defined by
where the sum k j is taken for all vertices Ic connected to j by an edge of
and c ( j ) is the number of edges connected to j. The discrete Laplacian hr, for the finite sub-graph is defined in the same manner. If c ( j ) is bounded as
a function on vertices hr is a bounded selfadjoint operator on Z2(r). When
the graph is finite, the constant function 1 is the eigenvector for the smallest eigenvalue. h r l = 0. Furthermore, the Perron Frobenius theorem for positive matrices implies that hr is a positive operator and the constant function 1 is the unique(up to multiplicative constant) ground state for hr. The free Hamil-
tonian for Boson on r is the second quantization of the standard Laplacian hr:
Hr = ' jJc( j )a;aj - 1/2 C ( a ; a , + a ~ a j ) ) . (4.3) j a - k+ j
Then,
[Hr, a*(f)l = a*(hrf)
We can introduce the rapidly deceasing functions for arbitrary graph by using
the distance of the graph. (We omit the detail.) The Weyl CCR C*-algebra on I? can be introduce as before. Then, the time evolution of an observable Q in the Weyl CCR C*-algebra d(r) on r is determined by
In particular,
218
For each finite sub-graph rn, let vL!’) be the equilibrium state at the inverse
temperature ,B and the density p:
(4.4)
where the fugacity z, (0 < zn < 1) is determined by
We define l ; ( r ) = {( fj) E 12(r) xjcr f j = 0). When r is a finite graph,
EO will be the projection from 12(r) to l;(r). We set Tro(Q) = Tr(EoQE0) . Proposition 4.1 (i) Suppose that
1
Then, for any p, there exists z , such that limn z, = z , < 1 and
(ii) Suppose that the following limit exists:
(4.8)
If p < @), limn z , = z , < 1 and the equations (4.6) and (4.7) are valid. If P 2 p(P),
n = P - p w (4.9) zn
limz, = 1 , lim n Irnl(l - zn)
Furthermore, we assume that any f = (fj) with compact support is in the domain of h,’/2. Then, any rapidly decreasing f and g are in the domain of
h,‘/2 (hence in the form domain of 1 - e-phr) and
py’P)(W(f)) = li&fP)(W(f))
= exP(-2(P -/I(P))lxr(f)t2)exp(-i(f’ 1 - e-phr f ) ) 1 1 1+e-Dhr
(4.10)
219
where
(4.11)
(4.12)
The Bose condensation occurs for the second case of the above proposition. Due to the equation (4.10) and (4.11), we have off-diagonal long range oder
and the U(l) gauge symmetry breaking for the state ‘pk’”) with high density.
The decomposition of into factor states (pure phase) is as follows. Set
+ip’e’(W(f)) = expji(p - - p ( P ) ) 1 / 2 ( e i e x r ( f ) + e-zeXr(f>)I
Then,
(4.13)
(4.14)
Now we turn to the integer lattice. As we can compute eigenfunctions
of discrete Laplacians on r = Z3 as well as half infinite lattices, it is easy to see that Bose condensation occurs. The spectrum of the discrete Laplacians for r = Z3 and half infinite cases are same but eigenfunctions are slightly different. Proposition 4.2 Set
r < N = { j = (jl,j’L?,j3) E z31jl 5 N } , r N < = { j = ( j l , j 2 , j 3 ) E Z3iN 5 jl}, r = z3, rR = rol, rL = (4.16)
O n these lattices, the Bose condensation occurs, namely, all the assumptions of Proposition 4.1 (ii) are valid.
220
The proof is same as that for the Newmean boundary condition in thecontinuous space R3. See[2].
5 NESS
Next we consider NESS(non-equilibrium steady states) and the current. The Hamiltonian we consider is bilinear in creation and annihilation operators but it contains interaction with inpurity or with different particles in the middle layer. More precisely, we consider the following situations. (a) The Initial state is the product of the condensate states on the left infinite lattice l?L and the right r R . The time evolution is the discrete Laplacian which allows Boson hopping from left to right and vice versa. (b) We introduce interaction with impurity in the middle. (c) The lattice is composed of three different parts,
and the interaction between the middle layer and left and right infinite parts translationally invariant in the tangential direction of the junction.
In all of these cases the total space is the three dimensional lattice r = Z3
(a) Free Boson. As before we set rR = ro<, r L = l?<-l, r = Z3 = l?L U ~ R . The Weyl CCR C*-algebra A = d(r)is the tensor product of the left and right Weyl CCR C*-algebras, d R = d ( r R ) and AL == A(rL), A = AL 63 d R . The Hamiltonians we consider here are
HL = Hr, , HR = HrR , HF = Hr hL = h p , , hR = hrR , h F = hr.
Thus the Hamiltonian HF allows hopping from left to right and vice versa. The initial state 90 is the product state of condensed states for HL and HR.
‘PO = CPL 63 ‘PR , ‘PL = G ( p L , p L , o L ) ‘PR = 4 ( p R , p R , o R ) (5.1)
where the ‘ p ~ and p~ are the factor states ?JIL!c’eL*R) on r L , R introduced in
(4.13). Theorem 5.1 For any rapidly decreasing f on Z3 the following limit exists:
t-w lim ‘Po(n(W(f))) = ‘Pw(W(f))
J--i i 8 L = e x p { T [ e ( P L -f(pL))1’2 + eisR(PR -@R)) ”2)Xr ( f ) -k C.C.]}
(5.2) x exp{-q(f, 1 (1 +.-‘)(I - e- B -1 f))
where C.C. refers to the complex conjugate and H is the multaplication operator on L2(T3) (via Fourier transform) defined by :
(5.3) p ~ w ( I c ) f (Ic), pRbJ(k) f (k),
when -7r < Icl < 0, when 0 < k l < 7r,
221
w(k) = 6 - 2 ( ~ 0 ~ k l + C O S ~ ~ + c o s ~ ~ ) (5.4)
To show (5.2) we use the stationary phase method and the following
formulae;
Next we consider the particle current from right to left. As is in other non- equilibrium steady states we define the current via the following procedure. First we consider the finite volume Hamiltonian Hr(N) defined by
r ( N ) = { j = (jl , j 2 , j 3 ) E rl Ijil 5 N 1 l j z l 5 N 1 1331 I N),
and the number operator,
QZ = C agaj , FaN) = (0 I ji I N, Ij2l I N , Ij3l 5 I N ) j E r a N )
The current from right per area (of junction surface) j~ is defined by
Then, by (5.2),
In particular the current vanishes when ,LIL = PR, and the equation (5.6) reads that no effect of condensation appears in the current in bulk.
(b) Impurity Scattering First fu positive integers N , M , and we set
r = z3 = r<-M - u r s u r N < -
rs = { j = ( j i , j 2 , j 3 ) E Z31 - M + 1 I j 1 5 N - 1)
Thus we consider the system composed of three parts, two half infinite (con-
densed) Boson reservoirs on r < - M - and r N < - and two dimensional layer rs
222
between these reservoirs and we suppose that the intermediate layer rs con-
tains impurity. The total Hamiltonian H consist of the hopping term and the impurity
term as follows
H(X) = HF + Xka;ak. (5.7) k E r s
The time evolution rt is determined as before, r t ( Q ) = eitH(X)Qe--i tH(X) For simplicity we assume that xkErs ( X k ( is finite. At the level of the one-particle
space l 2 (Z3), the above Hamiltonian corresponds to the following Schrodinger
operator:
h(X)( f )(k) = h F f (k) + X k f (k). (5.8)
( We set X I , = 0 if k is not in rs.) It turns out that h(X) of (5.8) is a trace class perturbation of h H if xkErs lXk l is finite, and the following wave operator v exists on f in the absolute continuous subspace of h(X);
lim e - - i t h F e i t h ( X ) f = V f . t -LX
(5.9)
As the initial state, we consider the product of two condensed states on r 5 - M 1 r N < and the equilibrium state on I'S associated with the discrete Hamilto-
nian hs, the inverse temperature PS and the the density ps. (Note that the
condensation cannot occur on rs due to two dimensionality.)
(5.10)
Theorem 5.2 W e assume that xkErs IXkl is finite and the Schrodinger op- erator does not have point spectrum. Then, the wave operator V in (5.9) exists on any f in 1 2 ( Z 3 ) and it is unitary. For any rapidly decreasing f on z3, Vf is in 11(Z3).
Let cp be any state quasi equivalent to cpo of (5.10). The following limit exists for any rapidly decreasing f on Z3 :
PO = PL 8 PS 8 P R ~ PL = V ! J ( ~ ~ L , ~ L , O L ) , PR = V ! J ( D R , ~ R , O R )
lim P(Tt(W(f))) = V)m(W(f)) t-m
1 e-v*Hv)(l - e-V* f iV) - l f)} (5.11) x exp{-q(fl (1 +
We focus on the case when P = PL = PR , p = p~ = ,OR, and QL # OR. Then
the equation (5.11) has the following form:
223
(5.12)
The phase factor (1 + e i ( e L - e R ) ( 2 is consistent with the Josephson formula which tells us that the current is proportional to a function of the phase difference (3L - OR.
The total current JR from r N < defined by
contains a non-vanishing term proportional to 11 + e i ( e L - e R ) ) 2 at the level of the 1st order perturbation theory.
( c ) Two-Dimensional Layer Next we proceed one step further. The interaction between the intermedi-
ate layer and the reservoirs is tranlationally invariant in the direction tangent to the junction. We divide the three dimensional lattices into three parts as before, I? = Z3 = r 5 - M U rs U r N g . The Hamiltonian is
H(X) = HF + c (a*(frc)arc + a * W f d ) . (5.13)
The real constant X is the coupling constant and the smearing function fk is called the form factor. We assume that the Fourier transform of f k is smooth and
k E r s
f ( k l , k z l k 3 ) ( j l , j Z , j 3 ) = f ( k 1 , 0 , 0 ) ( h , j 2 - b , j 3 - h) (5.14)
for any intergers IC2 and k 3 . This condition (5.14) implies that the total Hamiltonian H(X) is translationally invariant in the 2nd and 3rd directions
and we say H(X) is translationally invariant in the vertical direction. The corresponding Schrodinger operator is denoted by
h(X) = h F + XK (5.15)
The interaction K is no longer of trace class, however, due to translational invariance (in the vertical direction), the Hamiltonian h(X) is simultaneously
224
diagonalizable with the vertical shift operators on Z3. As a consequence, the
Hamiltonian h(X) is the direct integral of a smooth family of one-dimensional discrete Schrodinger operators with finite rank interaction. This observation shows the asymptotic completeness of the wave operator for the pair h(X) and h F and the absence of singular continuous spectrum for h(X).
For our study of condensed states, we require the 11(Z3) convergence of
the wave operator. Theorem 5.3 W e assume the following T 2 condition for the interaction K in (5.15):
(5.16)
f o r any rapidly decreasing f. Assume further that h(X) does not have any point spectrum. Then, the
wave operator V = limt e-ithFeith(X) maps the set of rapidly decreasing func- tions to l ’ (Z3 ) and all other conclusion of Theorem 5.2 i s valid. We have several examples of the form factor fk for which the T 2 condition
(5.16) can be verified. For example if N 5 M = 1 f(-l,o,o) = f(l,o,o) =
-2f(0,0,0) and the Fourier transform f(o,o,o)(Ic) of f(o,o,o) vanishes in a neigh- borhood of the origin, the T 2 condition can be verified.
We can consider the current j~ per unit area as defined in (5.5). Unfor-
tunately, for all the examples satisfying the T 2 condition which we are aware of, the current vanishes. Our physical interpretation of vanishing current in bulk is as follows. The T 2 condition is an infrared cut-off for the form factor and the infra-particles can not interact under the T 2 condition. As a result, there is no friction in condensed states and the current vanishes.
Acknowledgments
One of the authors (ST) thank Professor T. Hida for his invitation and hos- pitality at “International Conference on Stocahstic Analysis; Classical and Quantum” (Meijo University, 1-5 November, 2004). This work is partially supported by Grant-in-Aid for Scientific Research (C) from the Japan Society of the Promotion of Science, by a Grant-in-Aid for Scientific Research of Prior- ity Areas “Control of Molecules in Intense Laser Fields” and the 21st Century COE Program at Waseda University “Holistic Research and Education Center for Physics of Self-organization Systems” both from the Ministry of Education, Culture, Sports, Science and Technology of Japan and by Waseda University Grant for Special Research Projects, Individual Research (2004A-161).
225
Appendix
A Proof of Lemma
Let R1 = 2 max(R, R’) = Zk, then one has
ei(w‘-W)t
I”’ dw’ w ’ - w + i o
- J ( w , t ) = - J ( w , t ) - V ( W )
By changing the contour from [0, R1] to [0, i] U [i, R1 +i] U [R1+ i, RI ] , the
On the other hand, one has second term of (A.l) is shown to vanish as t -+ 03 uniformly for w E (0, k].
It is easy to show that the second and third terms are bounded by 4 K f i and, thus, vanish as t + 00. Because of
K G ( ~ ~ + ( W ’ - W1/2
Iw’ - WI Iw‘ - w + $ 1 a n d f l + l ~ ’ - w + ~ 1 ~ / ~ 5 1 + ( R + R ’ + l ) ~ / ~ = K ’ f o r t > l r , w e
and, thus,
226
where the integral in the last line can be easily seen t o converge.
Therefore, J ( w , t ) converges to 0 as t --+ +cm uniformly for w E [0, E] .
References
1. 0. Bratteli and D.W. Robinson, Operator Algebras and Quantum Statistical Mechanics vol.1 (Springer, New York, 1987).
2. 0. Bratteli and D.W. Robinson, Operator Algebras and Quantum Statistical Mechanics v01.2, (Springer, New York, 1997).
3. H. Spohn and J.L. Lebowitz, Commun. Math. Phys. 54, 97 (1977) and references therein.
4. J. Bafaluy and J.M. Rubi, Physica A153, 129 (1988); ib id . 153, 147 (1988). 5. J. Farmer, S. Goldstein and E.R. Speer, J . Stat. Phys. 34, 263 (1984). 6. J.-P. Eckmann, C.-A. Pillet and L. Rey-Bellet, Commun. Math. Phys. 201,
657 (1999); J. Stat. Phys. 95, 305 (1999); L. Rey-Bellet and L.E. Thomas, Commun. Math. Phys. 215, 1 (2000) and references therein.
7. T.G. Ho and H. Araki, Proc. Steklov Math. Institute 228, (2000) 191.
8. S. Tasaki, Chaos, Solitons and Fractals 12 2657 (2001); in Quantum Informa- tion 111, eds. T. Hida and K. Saito, 157 (World Scientific, Singapore,2001).
9. D. Ruelle, Comm. Math. Phys. 224 , 3 (2001). 10. S. Tasaki, T. Matsui, in Fundamental Aspects of Quantum Physics eds.
11. V. JakBiE, C.-A. Pillet, Commun. Math. Phys. 226, 131 (2002). 12. J. F’rohlich, M. Merkli, D. Ueltschi, Ann. Hen& Poincare‘ 4, 897 (2003).
13. J. Lauwers, A. Verbeure, A microscopic model for Josephson currents, J. Phys.
14. S. Tasaki, L. Accardi, submitted (2004). 15. I. Ojima, H. Hasegawa and M. Ichiyanagi, J. Stat. Phys. 50, 633 (1988). 16. I. Ojima, J . Stat. Phys. 56, 203 (1989); in Quantum Aspects of Optical Com-
17. V. JakSiE and C.-A. Pillet, Commun. Math. Phys. 217, 285 (2001). 18. V. JakSiE and C.-A. Pillet, J. Stat. Phys. 108, 269 (2002). 19. D. Ruelle, J. Stat. Phys. 20. H. Araki, E. J. Woods, J . Math. Phys. 4, 637 (1963). 21. J. T. Cannon, Commun. Math. Phys. 29, 89 (1973). 22. J. T. Lewis, J. V. Pulit, Commun. Math. Phys. 36, 1 (1974). 23. E. C. Titchmarsh, Introduction to the Theory of Fourier Integrals, Clarendon
L.
Accardi, S. Tasaki, World Scientific, (2003) p.100.
A : Math. Gen.37 3577 (2004)
munications, (LNP 378, Springer,l991).
98, 57 (2000).
Press, Oxford, (1948).
227
MULTIDIMENSIONAL SKEW REFLECTED DIFFUSIONS
GERALDTRUTNAU
Center for Interdisciplinary Research, Wellenberg 1, 33615 Bielefeld, Germany e-mail: trutnau@math.uni- bielefeld. de
Summary. Let G c Rd be (non-empty) open with euclidean closure # Rd. We assume that G is the interior of G. Let p E L1(Rd,dz), p > 0
dz-a.e., 0 5 a 5 1, and pa := ( C Y ~ G + (1 - a ) lwd \~)p , m, := padx. Let
D, = R d \ E i f a = 0, D, = Rd if 0 < a < 1, D, = G if a = 1. Let A = (ai j ) be symmetric, and globally uniformly strictly elliptic, p satisfying some mild regularity assumption. Non-symmetric conservative
diffusion processes X t E D, which inside D, \ BG obey the generator L given in the following suggestive form
d d
where pi l (B1, ..., Bd) E L'(D,;@,m,) satisfies xf=l so, Biaifdz = o for all f E Cr(a,), are constructed and analyzed. The coefficients are in
general neither continuous nor locally bounded. For dG, p, uij, sufficiently
regular, a Skorokhod type decomposition for X t is given in Theorem 5.2.
Indeed, due to the jump of pa along aG, the generator L has to be regarded with boundary conditions, or as integration by parts show, with additional
surface measure terms. To these surface measures, there are uniquely asso-
ciated multidimensional local times similarly to the case of reflected Brow-
nian motion, where the local time is associated to the Dirac measure in
zero. Under the irreducibility assumption of the semigroup corresponding to the symmetric part of L we show in Theorem 6.5 that X t is recurrent for
q.e. starting point. Because of the non-symmetry (a, 1 - a) we can speak
about skew reflection. The extreme cases are a = 1, a = 0. In particular, if
a = 4 there is no reflection.
228
The constructed process reminds one of a multidimensional generalized ana- logue of the a-skew Brownian motion (see [5 ] , [9]), which corresponds to the case d = 1, G = Rf, p = 1, Bi = 0, aij = 6ij.
2000 Mathematics Subject Classification: 6OJ60,60555,31C15,31C25, 35525.
Key words: Diffusion processes, local time and additive functionals, po- tential and capacities, Dirichlet spaces, boundary value problems for second order elliptic operators.
1 The generalized Dirichlet form and its ca- pacity
The following is an extension result corresponding to the diffusion con- structed in [13].
Let G c Rd be (non-empty) open with euclidean closure # Wd. Since we are not interested in the geometry of G, we may assume that G is the interior of E. Then dG = d c = d(Rd \E). Let dx be the Lebesgue measure on Rd, p E L1(Rd,dz). Let 0 5 a 5 1, and put
Rd\E f o r a = O
{ G f o r a = l . I W ~ f o r ~ < a < 1 D, :=
We equip D, with the usual Euclidean norm 1 . 1 = Let p, = ( a l c p + (1 - a ) l w q c p ) with p > 0 dx-a.e., and m, := padx. Let 1.11, 1.12,
I . loo, denote respectively the L1-norm, L2-norm, L"-norm, w.r.t. m, on D,. For V c L1(D,,m,) let VI, := DnLoo(D,,m,), V+ := {f E Vlf 2 0 m, - a.e.}. Let A = (a i j ) l s i , j g be measurable, symmetric, and uniformly globally strictly elliptic on D,, i.e. 3X > 0 such that
x-' xt? 5 C u i j ( z ) ~ j I ACE? V(G, ...,t d ) E I K ~ , ma-a.e. z E D,. (1)
For k E N U {m}, let Ck(Rd) be the k-times continuously differentiable functions on Rd, and @(ad) those functions in Ck(Rd) which have compact support. For D c Rd closed let
d d d
a= 1 +=I i= 1
C(y (D) := {U lD : u E C,-(Rd)}.
229
Denote by D, the euclidean closure of D,. Let us from now on assume that
€"(ti, w) := 1 l , ( A V u , Vv) d m , ; u, w E CF(D,),
is closable in L2(D,,m,). Conditions in order to have closability are very
mild. For instance, if p satisfies the Hamza type condition in D,, or &p E Lto,(Bd, d z ) , 1 5 i 5 d , then (E', Cr(n,)) is closable in L2(D,, ma). This
follows easily by adapting arguments from [13]. Let us denote the closure of (€", Cr(D,)) on L2(D,, m,) by (E', D(E')). (E",D(€')) is a symmetric Dirichlet form (cf. [4]). (€',D(€")) needs not
to be regular (compare [4,p.6] for the definition) on D,. Let for instance Q = 1, then Co(D,) n D(E') is in general not dense in D(E'). It becomes a regular Dirichlet form on n, with the identifications specified in the following remark.
Remark 1.1 If 0 < Q < 1, then C0(B5,) no(€') is dense in D(€'). Obvi- ously, - CO(n,)nD(ET) is dense in Co(D,). Therefore (E ' , D(E")) is regular on D, = D,. For Q = 0, or a = 1, we can make it a regular Dirichlet form on n, through the following identification. W e first extend m, to n5, by setting m,(dG) := 0, thus identifying LP(D,, m,) with Lp(Da, ma), p 2 1.
W e then may regard (E" , D(€')) as a Dirichlet fo rm on L2(na,ma). I n this case (E" , D(E')) is regular. Henceforth, we will make these identifications.
Let (G;)O,~ be the resolvent associated to (€", D(€")) (see [4, Chapter 1.3]),
and (., .) be the inner product in L2(D,, ma). For p > 0 let €;(., .) :=
E " ( . , . ) + p(., .), I . I D ( E ' ) := E ; ( . , .)* be the norm corresponding to E'. Then D(€') is the completion of C?(Da) w.r.t. 1 . ID(^). We denote by (L", D(L") ) the generator associated to ( E T , D(E')), i.e.
D(L') = {f E D(E') : g H E'( f ,g) is continuous w.r.t. (.,.I* on D(E")),
and for f E D(L"), L" f is the unique element in L2(D,, m,) such that ( -L" f ,g)m = € " ( f , g ) for all g E D(€"). Since p E L 1 ( R d , d z ) , one can easily see that lg, E D(&"), and €'(lDa, f ) = 0 for any f E D(&"). Hence 16, E D(LT) and L'16, = 0, i.e. L" is conservative. In particular
Gila, = (1 - L')-'lna = (1 - L')-'(I - L')I- = 1- Dcv D,'
Remark 1.2 Suppose - that as in Theorem 5.2 below, everything is SUciently smooth. Let A = (Zij)I<i, j<d, where &j is a 4.c. version of aij . Let f E Cr(D,), Q # i. Then f is in D(L'), if
(XV f , 17) = 0 Tr(p)da-a.e.
230
Indeed, it sufices to look at the proof of Theorem 5.2.
We consider a measurable vector field B : D, - EXd, which is ma-square
integrable on D,, i.e. JD IBI2 dm, < CQ, and such that
(B, V u ) dm, = 0 for all u E Cr(n,). (2) La Note, that since Cr(n,) c D(&') densely w.r.t. I . I D ( ~ ) , and because of
strict ellipticity (l), (2), extends to all of D(&'). Furthermore,
L a ( B , V u ) v d m , = - (B, Vw)udm, for all u, v E D(&')a. (3)
Exactly as in [13, Proposition 1.41 we have the following:
Proposition 1.3 (i) The operator L'u + ( B , V u ) , u E D(L')b, is dissi- pative, hence in particular closable in L1(D,,m,). The closure (T, D ( z ) ) generates a sub-Markovian Co-semigroup of contractions (Tt)t>o.
(ii) D(z )b c D(E') and
In particular,
E'(u, u) = - LUU dm, for ewenJ u E D(Z)b. La - (5)
(iii) (Tt)t>O - is Markowian.
(Tt)t?o can be restricted to a Co-semigroup (Tt)tlo on L2(D,,mQ). The corresponding generator ( L , D(L) ) is the part of (z, D ( z ) ) on L2(D,, ma), i.e. D ( L ) = {u E D ( Q n L2(Du ,m, ) l L E L2(D,,m,)}, and Lu := Eu, u E D(L) . Let (L ' ,D(L')) be the adjoint of ( L , D ( L ) ) in L2(Da,m,), and (T,')t>o the corresponding semigroup. According to [lo, Examples 4.9(ii)], the generalized Dirichlet form corresponding to (L , D (L ) ) is
(-~u,v) for u E D ( L ) , v E L2(D,,m,) { (-L'v,u) for v E D(L') , u E L2(D,,m,). E(u,v) :=
Define &p(., .) := &(., a ) + p(., .). As in [13] one extends & as
&(u, w ) = I r (u, v) - ( B , Vu )vdm, for every u, v E D(&')b. (6) I D a
23 1
As a consequence of (3) we then have
&'(u, u) = &(u, u) for every u E D(E')b. (7)
(7) suggests that the capacities related to & and &' are equivalent. This will be shown further below. Let ( G p ) ~ > o (resp. (G&)p>o) the strongly continuous contraction resolvents
on L2(D,,m,) related to (L ,D(L) ) (resp. (L', D(L'))) . As usual let fu de- note the 1-reduced function off on U w.r.t. &. By abuse of notation (Gicp)~ will always denote the 1-coreduced of Gicp on U. Let e7u, f E D(E'), de-
note the 1-reduced function of f on U. If U = O, we simply write e;
instead of e?"", and e f instead of fBm. Let Cap be the capacity associated to E' as defined in [4, p.641. Recall that an increasing sequence (Fk)kE~ of closed subsets of n, is called an (€'-) nest on D,, if limk,, Cap(Fz) = 0.
-
We fix throughout cp E L1(D,, ma), 0 < cp 5 1. The cp-capacity related to & is determined by
Cap,(U) = E1(Glcp, (Gicp)~) for U c O,, U open.
An increasing sequence of closed subsets ( F k ) k > l is called an E-nest, if limk,, Cap,(Fz) = 0. A subset N c D, is cafied &-exceptional if there
is an &-nest (Fk)kll such that N C nk>lD, - \ Fk. The &'-exceptional sets are defined similarly.
Theorem 1.4 (a) ( F k ) k E ~ i s an &'-nest on a, in the sense of [4,p. 6'4 iJ andonly if (Fk)kE~ is an E-nest.
(i i) A subset N of n, is &-exceptional iJ and only i f it is exceptional w.r.t. E' in the sense of [4,p. 1341.
Proof It is enough to show (i). Let ( F k ) k E ~ be an &'-nest. Then (see
[4]) er'FkC -+ 0 in D(E'). Since Glcp 5 lDa it follows e::; + 0 in D(E').
For p > 0 let (Gicp& be the unique solution f E D(L') to (1 - L' ) f =
P(f - ~ F , c ) - . It is known that (Gicp)& t (G',cp)~; and strongly in
L2(D,, m,). It is easy to see that (G',(P)$~ also converges weakly in D(E")
li%
232
to (Gicp)~; as p t 00. Then using in particular (1)
we obtain
and consequently
k+oo
0 Having shown that E' and & have equivalent capacities, we can pronounce
exceptional, quasi-continuous, quasi-everywhere, etc. , unambiguously and without specifying whether it is meant w.r.t. Er or &, and so we will do.
2 Construction of an associated diffusion
The results obtained in [13] are derived by using the following ingredients: strict ellipticity (l), finiteness of the reference measure ma, and m,-square
233
where (G) is the resolvent associated to E Since
It follows that (F) is an E-nest Converesely, suppose that (F) is an
integrability of the vectorfield B. This is the reason why the results of [13] carry over to our situation. We will list below the results corresponding to
[13]. The interested reader may then refer to [13]. In order to show the existence of an associated process we have to check quasi-regularity of & and condition D3 of [lo, IV.21. Doing the identifications - of Remark 1.1, the quasi-regularity of the generalized Dirichlet form E on
D, is in view of Theorem 1.4 a consequence of the regularity of the classical
Dirichlet form E’ on D,. Since exactly as in [13] y := D(z)b is a linear space such that Y c L”(D,,m,), Y n D ( L ) c D(L) dense, limp,, ef -oGpf = 0
in L2(Dar m,) for all f E y, and such that f A p E y(=the closure of y in
Loo(D,, ma)) i f f E 7, p 2 0, the existence of an associated process follows
from a general result in [lo]. In particular one also shows that D( r )b is an
algebra. Finally, we can also show that E is local in the sense of generalized Dirichlet
forms and obtain the following.
Theorem 2.1 There exists a Hunt process M = (0, ( .F) t>o, (Xt) tzo, (Pz)zsuch that the resolvent Rp f ( x ) := Ez[~ow e-pt f ( X t ) d t ] is a quasi-continuous ma-version of Gpf for any f E L2(D,,ma)b, p > 0. In particular M is conservative, i.e. P R p l ~ ~ ( x ) = 1 for quasi every x E Dff, and there exists an exceptional set N c such that
P, (t H X , is continuous on [o, m[) = 1 f o r every x E D, \ N .
Remark 2.2 Since -B satisfies the same assumptions as B, exactly as we have co_structed the process M associated to L one can construct the coprocess M associated to L’. The coprocess will have exactly the same properties than M.
3 The Revuz correspondence
Definition 3.1 A positive measure v on (n,,l?(nff)) charging no excep- tional set is called smooth if there exists a nest (Fk )kEN of compact subsets of D,, such that
The smooth measures are denoted by S.
The positive continuous additive functionals (PCAF’s) of M are defined as in [12]. The following theorem accomplishes together with [12, Theorem 3.11 the so-called Revuz correspondence for the generalized Dirichlet form E . It can be shown directly and exactly as in [13] (please see also the following
Remark 3.3).
< m for all k E N.
234
Theorem 3.2 Let u E S. Then there exists one, and only one PCAF (At ) t lo of M such that for any positive, Bore1 measurable f , and any quasi-continuous ma-version i7 of v E pb = {u E L2(D,, m a ) b I PG&+lu I u 'd,B 2 0}, we have
h
v f d u = lim ,B/ i7(x)Ex [lW e- (p+ ' ) t f (X t )dAt ] m,(dx). (8) s,, - p+Oo D,
I n this case we write u = V A .
Remark 3.3 A consequence of Remark 2.2 is that the process M is in weak duality with the coprocess M. I n particular m, is an excessive measure for the process M. Theorem 3.2 could then also follow from results in [6, $91 where properties of Revuz measures are discussed in the weak duality context. The interested reader is invited to consult the given reference, and to compare carefully the fine (process) topology of [6], with our analytic capacity.
4 Semimartingale characterization
From [12,Theorem 4.5(i)] we know that for any f E D ( L ) with quasi-
continuous ma-version f we have a unique decomposition
t 2 0,
-
ALfl := T ( X t ) - T ( X 0 ) = Mif1 + Nif1, (9)
where M[ f l is a martingale additive functional (MAF) of M of finite energy and N[ f l is a continuous additive functional (CAF) of M of zero energy. The energy of an additive functional A of M is defined by
1 00
e ( A ) = f hlP2 L, Ex [I e-OtA?dt m a ( d x )
whenever this limit exists in [O,m]. The equality (9) is to be understood in
the sense of equivalence of additive functionals 0,f M. If AIfl decomposes in the sense of (9) for some f with f E L2(D,, ma) we
write f E L2(D,, ma)dec. Using the extension theorem [12, Theorem 4.5(ii)]
it can be shown exactly as in [13] that D(E')b U D(L) C L2(D,, ma)dec.
Remark 4.1 For f E D(E') we obtain also the decomposition (9) except that we do not know whether e(NIf1) = 0 (cf. Remark 4.2 in [13]).
If u E D(L) then clearly N/"] = s," Lu(X, )ds is of bounded variation, i.e. it can be represented as the difference of two PCAF's of M. For general
235
u E D(&’)b we have roughly the following. Nt(ul is of bounded variation, if and only if there exist v1, vz E S, such that
for “enough”quasi-continuous w E D(&’)b. In this case Nt(ul = A: - At, where A’, A’, are the PCAF’s associated with v1, vz. For the precise mean- ing of “enough”we refer to Theorem 4.5. in [13].
5 Identification of the process
Throughout this section we assume that G c Wd is a bounded Lipschitz domain, i.e. G is open, bounded, and its boundary dG is locally the graph of a Lipschitz function. Let G be the (compact) closure of G in Wd equipped with the usual Euclideannorm 1 . 1 = (., .) ‘ I2. Let R c Wd open. Let H’*P(R), p E [l,co[, denote the classical Sobolev spaces of order one in LP(R,dx), i.e. H’J’(S2) := {u E LP(R,dx)ldiu E LP(R,dx), 1 5 i 5 d } . We will give some kind of Skorokhod representation of Xt when p E H1>l(Rd), aij E
D(E‘). Note that p E H1?l(Wd) implies the closability of (E‘, Cr(D,)) in
Let (T be the surface measure on dG. Recall that since G is a bounded Lipschitz domain there exists a bounded linear operator
L2(D,, ma).
Tr : H1’”(G) --f LP(dG,a), (11)
called the trace on dG, and TT( f) = f on dG for any f E H1”(G) n C@). Furthermore, the weak Gauss-Green theorem holds, i.e. if f E H’il(G), 1 5 i I: d , then
S, d i f dx = - S,, Tr(f)qi d o
where q = (71, ..., q d ) is the inward normal of F on dG (see for instance [3]). Let v = (vl, ..., vd) denote the inward normal of Rd \ G on dG. We
have vi = -qi, 1 5 i 5 d. Let g E CF(D,), and f E Cr(Rd) such that f ID, = 9. Let pm E Cr(Wd) such that pm 4 p in H1y’(Wd). Then by (11)
236
and Gauss-Green theorem F
= - lG gTr(p)vi do.
For p E H1>'(Rd) we show that the weighted surface measure Tr(p)da, or equivalently l{T,(,),o}da, on dG is smooth.The proof is similar to the cor- responding proof in [13] but we include because of some subtle differences.
Theorem 5.1 Let p E H1vl(Wd). Then Tr(p)da E S, In particular for any w E D(ET)a and quasi-continuous ma- version 6 we have
- lG GTr(p)qi d a = k di(wp) dx = - ai(wp) dx = 1, GTr(p)vi da.
Proof If h E C1(G) it is well-known (see e.g. [3, p.134, 3. (* * *), (***)I) that there exists a universal constant C depending only on the Lipschitz
domain such that
for any p E [1,m[. Let us choose (pk)&N c C ~ ( R ~ ) with pk -+ p in H1>'(Rd) as k 4 m. By (11) p k --f Tr(p) in L1(dG,a) as k 4 00. Let K c Da be a compact set. Let f E Cp(Da), f 2 1 everywhere on K. Then, using (14), Lebesgue's theorem, and the Cauchy-Schwarz inequality,
5 max(a, 1 - a)
Assume Cap(K) = 0. By [4, Lemma 2.2.7(ii)]
237
where CK = {u E CF(D,) I.(.) 2 1,Vx E K } . Hence, there exists
(fn)nEN c CF(D,), fn(x) 2 1, for every n E N, x E K , such that IfnlD(EP) + 0 as n + 00. Since normal contractions operate on D(E') we may assume that supnENs~pZERd I fn(x)( 5 C. Selecting a subsequence if necessary we may also assume that limn--roo I f n l = 0 ma-a.e., hence dx- a.e. on D,. Suppose a # 1, then D, 3 G. Consequently, using Lebesgue's
theorem we obtain
)imL IfnllVpldx = 0,
and therefore JKnaG Tr(p) da = 0. Since T r ( p ) d c , as well as Cap are inner
regular the first assertion now follows for a # 1. If a = 1, we choose a compact domain v with smooth boundary such that c c V and dGndV = 8. Then U := V \ G is a bounded Lipschitz domain which is contained in D,. Let K denote its surface measure. As before we can then show that
T r ( p ) d n is smooth. Note that 1aGTr(p)dK, = Tr (p )do . Thus T r ( p ) d a is also smooth. The second assertion is clear by (13) and since T r ( p ) d a is finite and smooth as we just have shown.
0 We present here below the identification of the process in Theorem 2.1 for a special class of p, aij, dG.
Theorem 5.2 Let p E H 1 ? ' ( R d ) , p > 0 dx-a.e. Let A = (a i j ) satisfy ( 1 ) with aij E D(E'), and q.c. m,-versions Zi j , 1 5 i , j 5 d. Let 0 =
( a i j ) I < i , j < d be the positive square root o f the mat r ixA. Let B = p-l(B1, ..., z1be a ma-square integrable vector field satisfying (2). Let G be a bounded Lipschitz domain. Let q = (q l , ..., r ]d) (resp. v = (v1, ..., V d ) ) be the unit inward normal vector field of on dG (resp. of Rd \ G on dG). Let & q ) k := xy, lak jq j , (resp. A(v)k := c jd_ lak jv j ) , 1 I k I d, be the inward normal of G (resp. Rd \ G) associated with A . The conservative dif- fusion X = ( X ' , ..., X d ) of Theorem 2.1 is a semimartingale and has the following Skorokhod decomposition for 1 5 k 5 d:
- w -
t 2 0, P,-a.s. fo r q.e. z = (z1, ..., Z d ) E D,, where w = (w', ..., w ~ ) is a d-dimensional standard BM starting from zero, ( l f ) t 2 0 denotes the unique PCAF associated to the weighted surface measure T r ( p ) d u E S through
238
Theorem 3.2. I n particular
r t
Proof The coordinate functions pk(x1, ..., xd) := xk are not in D(E')b. But they are locally in in D(E')b, i.e. p k f E cr(D,) C D(Er)b for any
f E Cr(E,). Let f E Cr(Ea), (Ml f l ) , be the square bracket of Mjfl.
Then an easy calculation gives that the energy measure of Mif1, i.e. the
Revuz measure of (M[ f l ) , , is
P ( M [ f l ) = (AV f , V f ) d m a .
Thus ( M [ f l ) , = sof(AVf(X,), V f ( X , ) ) d s by Theorem 3.2. Let w E D(ET)b. Then by the previous results of this section
Let fl E Cr(D,), f l = 1 on K ~ ( o ) := { x E D, : 1x1 5 l } , 1 2 I. Using the above, Theorem 5.1, (lo), Theorem 3.2, we easily derive the decompo- sition (9) for p k f l . One can also easily see that ~ K ~ ( O ) P ( ~ [ ~ ~ ~ ~= 0
t - C K I ( O ) C .- inf{t > 0 : Xt E Kl(0)c}. Since obviously A P f i l = AIPkfml t 'dt < - Q K I ( O ) C ,
and o ~ ~ ( o ) ~ 00, by letting 1 --+ 00 we get the identification of the process.
Since by (10) NFk f i l is of bounded variation for any I , the process is a
semimartingale. Of course l f = s," laGn{Tr(p)>O)(Xs)&$, t 2. 0, by Theo-
rem 3.2, because la~n{T, (p)>o}Tr(p)da = T r ( p ) d a since supp(a) c dG.
0
for any m 2 1. This further implies that MFkf i l = M[pkfml 'dt < ._
Remark 5.3 W e would like to describe shortly the decomposition of The- orem 5.2. The diffusion has a symmetric dri j l part corresponding to the logarithmic derivative of p associated to the diffusion matrix A. I t has a
239
purely non-symmetric part p;’B, and two reflection parts. (15) tells us that ef’ behaves like a multidimensional local t ime on some part of the boundary, i.e. Cf’ only grows when X t meets the boundary d G at those points where Tr(p) > 0. At that t ime, Xt is reflected at, or passes through (see Remark 6.6), d G in normal direction associated with A. Finally note that if cx = there is no reflection.
6 Recurrence
Let f E L1(Da,m,)+. Then
is uniquely determined at least for ma-a.e. z E 0,. Recall that 1 ~ - E
D(L)b and p G p 1 ~ - = 1~ for any p > 0. Therefore
G16, = 00 ma-a.e.
We will need the following Hopf’s maximal ergodic inequality (cf. [4, Lemma 1.5.21, [8, Lemma 1.5.41). It can be shown exactly as in the sectorial case.
Lemma 6.1 Let h E L1(D,,m,), p > 0, and let Ep := {z E B, I supnL1 Ggh(z) > 0). Then
n
Hopf ’s maximal ergodic inequality is essentially sufficient in order to prove
the following recurrence lemma.
Lemma 6.2 Let f E L1(D,,m,)+. Then { G f = co) U {Gf = 0) = D,, and { f > 0) c {Gf = co}, up to an ma-negligible set.
Proof For arbitrary f E L1(D,,m,)+, a > 0, we set h := lg- - af in
(16). Since B := {Gf < co) = {GlB0 = c o ) n { G f < 00) c Ep for any ,B > 0, up to a ma-negligible set we obtain
for any N 2 1. Dividing by a, and letting a 00, N --f 00, we get sB PGp f dm, = 0, as well as sB G p f dm, = 0. Letting in the first case p 4 00, and in the second p 4 0, we get
l{Gf<oo) * f = 0 a~ well as l { ~ f < ~ } . Gf = 0.
240
l { ~ f < ~ } . G f = 0 implies ({Gf < m}n{Gf > O } ) c = {Gf = m}U{Gf =
0 ) = D, up to ma-negligible set. l {Gf<oo}- f = 0 implies {f > 0) c { G f = m}. This concludes the proof.
0
-
Lemma 6.3 Let f E L1(Da,m,)+. Then: (i) { G f < m} is invariant, {Gf = m} is co-invariant, i.e. for any p > 0, t > 0, h E L2(Da,ma), we have
(zi) l {Gf=oo} is excessive, l {G f<oo) is co-excessive, and l { ~ f < ~ } , l { ~ f =E
(iii) Let u, v E D(&')b. Set B := { G f < 00). Then l gu , l g v E D(E' )b, and
D(&').
&'(u,v) = &'( lgu, lgv) + &'(lgcu, lgcv) . (17)
In particular
& ' ( l g U , V ) = &'(lgu, 1gv) = ET(u, l g v ) ,
and B , hence also BC, is an invariant set w.r.t. E' in the sense of [4, Lemma 1.6.1 .].
Proof (i) We may assume that f is bounded. Put B = {Gf < m}, B" = {Gf = co}. Let h E L2(Da,ma)+. Then for any n 2 1
241
and
Thus 1gCT[(hl{Gf5n)) = 0, and therefore l g ~ T [ ' ( g l ~ ) = 0 for any g E
L2(D,,m,). It follows ( lgTt( lBch),g) = (h,IgcT;(1Bg)) = 0 and there- fore lgTt( lBch) = 0. We conclude that (i) holds. We now show (ii). By (i) ,f?Gplgc = lgePGplgc I l g c P G p l ~ ~ = l g = , and
E'(PGplgc,PGplgc) = E(PGpli+,PGplBc) I E ( P G ~ ~ { G ~ = ~ ) , 10,) = 0.
Hence lgc E D(€') easily follows from [7, I. Lemma 2.121. The proof cor- responding to 1~ works similarly.
(iii) Let p ( M [ w l ) , be the energy measure related to (MIzo]),, w E D(E' )b. Suppose that w is constant pu(M[wl)-a.e. on some Bore1 set A. Then it is
well-known that J A d y ( M ( w l ) = 0 (see [ l l , Lemma 3.8. (iii)], but also [l,
Corollaire 61, [2, equation ( S ) ] in the symmetric case). Note that p(M[wl) = (AVw, Vw) dm,. It follows that ET( lgu, lgcv) = E'(lg=u, l g v ) = 0. In-
deed, by Cauchy-Schwarz
€'( lgcu, l g v ) = 0 follows in the same way. Thus €'(u, v) = € ' ( ~ B u , 1gw)+ ET( lgCu, lgCv) . Replacing first u by l g u , and then v by l g v in (17), the second assertion of (iii) follows. For the last assertion, exactly as in the proof of [4, Theorem 1.6.1.1, we use the second assertion of (iii) in order to
see that for any h,g E L2(D,,m,)b,
From this, the invariance of B follows immediately.
0
Proposition 6.4 Suppose I' is irreducible (an the sense of (4, p . 481). Then Gf = m ma-a,e. for any f E L1(D,,m,)+ with ma({ f > 0)) > 0.
Proof By our assumption JD, f d m , > 0. Since { f > 0) c {G f = m} by
Lemma 6.2, it follows so, f dm, = s,, fl{Gf=..) dm,, hence m,({Gf = m}) > 0. Since {Gf = m} is €'-invariant by Lemma 6.3 (iii), and E' is irreducible, we must have m,({Gf = 00)~ ) = 0. Therefore the assertion
follows.
0
242
Theorem 6.5 Suppose E' is irreducible (in the sense of [4, p. 481). For r > o let ~ ' ( z ) := { y E D, : 111: - yI < r } , and cDr(,) := inf{t > OlXt E
DT(z) } . Let (6t)t>0 be the shift operator correponding to X t . Then
(18) P , ( ~ D ~ ( ~ ) 0 6, < 00, ~n 2 0) = 1 for q.e. z E 0,.
Proof The proof is similar to the proof of Theorem 4.6.6.(ii) in [4], but we
include it in order to point excactly out the subtle differences. Let B c 0, be an arbitrary Bore1 set, and (pt) t20 the transition semigroup of X t . The
Markov property implies that f(z) := P,((TB < m) is excessive, since
p t f ( z ) = P,(CJB 0 6t + t < m) 5 P,(oB < m). In particular due to the boundedness of f , standard arguments then imply that f E D(ET)b , and f is q.c. On the other hand for any positive g E CF(D,) with sD g dm, > 0 we have G'g = limp,o Gbg = 00 ma-a.e. Indeed, this follows immediately
from Remark 2.2 and the co-version of Proposition 6.4. Then, using the
resolvent equation
0 5 (G&g,f-aRaf) = (g,Rpf-aRpRaf) = (g1Raf-PRpRa.f) 5 (9,Ra.f) < m.
Letting /3 ---f 0 we conclude aR, f = f , hence L f = 0 and thus f ( X t ) in
(9) is a P,-martingale for every z E 0, \ N where N is some exceptional
set. Let E := { f = l}, y E [0, l), and E := { f = l}, c-, := c~~ be the first hitting time of E-, := { f 2 y}. Note that E-, is finely closed. Now, for any
z E E \ N, T > 0, we have by the optional sampling theorem
1 = f (X) = E z [ f (XuyAT)]
= Ez[f(XuJ;e-/ I T I + E z [ f ( X T ) ; q >TI
I yP,(a-, 5 T ) + Pz(a-, > T ) ,
which means that P,(e-, I T ) = 0. Thus
Pz(cp < m) = 0 for any z E E \ N ,
and E is invariant w.r.t. E . Exactly as in the proof of Lemma 6.3(iii) one shows that E is then invariant w.r.t. E'. Owing to the irreducibility of E' we must have either m,(E) = 0 or m,(EC) = 0. Finally we let B = DT(z). Then E 3 DT(z) and m,(E) 2 ma(DT(z)) > 0, thus m,(EC) = 0. It follows f(z) = 1 for ma-a.e. z. But f is quasi-continuous, and therefore
f = 1 q.e. (18) now follows from the Markov property.
0
Remark 6.6 W e have seen in Theorem 6.5 that i f E' is irreducible, then Xt is recurrent q.e. in the classical sense. In particular, i f additionally 0 < a < 1 one can conclude that the process passes infinitely often through dG.
243
References
[11 Bouleau, N.: DBcomposition de 1’6nergie par niveau de potentiel, ThBorie du potentiel (Orsay, 1983), 149-172, Lecture Notes in Math., 1096, Springer, Berlin, 1984.
[2] Bouleau, N., Hirsch, F.: Dirichlet forms and Analysis on Wiener
[3] Evans, L.C., Gariepy, R.F.: Measure theory and fine properties of
141 F’ukushima, M.,Oshima, Y., Takeda, M.: Dirichlet forms and Sym- metric Markov processes. Berlin-New York: Walter de Gruyter 1994.
[5] Harrison, J.M., Shepp, L.A.: On skew Brownian motion, Ann. Prob.,
[6] Getoor, R.K., Sharpe, M.J. Naturality, standardness, and weak du- ality for Markov processes, Z. Wahrscheinlichkeitstheor. Verw. Geb.
space, Walter de Gruyter, Berlin, 1991.
functions, Boca Raton: CRC Press 1992.
Vol. 9, N0.2, 309-313, (1981).
67, 1-62 (1984).
[7] Ma, Z.M., Rijckner, M.: Introduction to the Theory of (Non- Symmetric) Dirichlet Forms. Berlin: Springer 1992.
[8] Oshima, Y.: Lectures on Dirichlet spaces, Universitat Erlangen- Nurnberg, (1988).
Springer Verlag, (1999). [9] Revuz, D., Yor, M.: Continuous martingales and Brownian motion,
[lo] Stannat, W.: The theory of generalized Dirichlet forms and its ap-
plications in analysis and stochastics, Mem. Amer.Math. SOC., 142 (1999), no. 678.
[11] Trutnau, G.: On a class of non-symmetric diffusions containing fully nonsymmetric distorted Brownian motions. Forum Math. 15 (2003), no. 3, 409-437.
[12] Trutnau, G.: Stochastic calculus of generalized Dirichlet forms and applications to stochastic differential equations in infinite dimen- sions, Osaka J. Math. 37 (2000), 315-343.
[13] Trutnau, G.: Skorokhod decomposition of reflected diffusions on bounded Lipschitz domains with singular non-reflection part. Probab. Theory Related Fields 127 (2003), no. 4, 455-495.
244
ON QUANTUM MUTUAL TYPE ENTROPIES AND QUANTUM CAPACITY
NOBORU WATANABE
Department of Information Sciences Tokyo University of Science
Yam.azaki 2641, Noda City, Chiba 278-8510, Japan E-mail: watanabe@is.noda.tus.ac.jp
The mutual entropy (information) denotes an amount of information transmitted correctly from the input system to the output system through a channel. The (semi-classical) mutual entropies for classical input and quantum output were d e
fined by several researchers. The fully quantum mutual entropy, which is called Ohya mutual entropy, for quantum input and output by using the relative entropy was defined by Ohya in 1983, and he extended it to general quantum systems by
means of the relative entropy of Araki and Uhlmann. The capacity shows the abil-
ity of the information transmission of the channel, which is used as a measure for
construction of channels. The fully quantum capacity is formulated by taking the
supremum of Ohya mutual entropy with respect to a certain subset of the initial state space. One of the most important theorems in quantum communication t h s
ory is coding theorem. The quantum coding theorems are discussed by using the mutual entropy - type measures introduced in the studies of quantum information.
In this paper, we compare with these mutual entropy-type measures in order to obtain most suitable one for discussing the information transmission for quantum communication processes.
There exist several different types of quantum mutual entropy. The
classical mutual entropy was introduced by Shannon to discuss the infor-
mation transmission from an input system to an output system. It denotes
an amount of information correctly transmitted from the input system to
the output system through a channel. Kolmogorov, Gelfand and Yaglom
gave a measure theoretic expression of the mutual entropy by means of
the relative entropy defined by Kullback and Leibler. The Shannon's ex-
pression of the mutual entropy is generalized to one for finite dimensional
quantum (matrix) case by Holev~,Ingarden,Levitin~~~~~~. Ohya took the
measure theoretic expression by (KYG) Kolmogorov-Gelfand-Yaglom and
defined Ohya mutual entropy la by means of quantum relative entropy of
Umegaki 37~13 in 1983, he extended it 2o to general quantum systems by
245
using the relative entropy of Araki and Uhlmann 38. Recently Shor 36
and Bennet et a1 6,4 took the coherent entropy and defined the mutual type
entropy to discuss a sort of coding theorem for communication.
In this paper, we briefly review quantum channels. We briefly explain
three type of quantum mutual type entopies and compare with these mu-
tual types entropies in order to obtain most suitable measure to discuss
the information transmission for quantum communication processes. Fi-
naly, we show some results 2 1 1 2 3 1 2 4 7 2 5 1 2 8 9 2 9 3 2 6 of quantum capacity defined
by taking the supremum of Ohya mutual entropy according to the subsets
of the input state space. The capacity means the ability of the information
transmission of the channel, which is used as a measure for construction of
channels.
1. Quantum channels
In development of quantum information theory, the concept of channel
has been played an important role. In particular, an attenuation channel
introduced in l8 has been paid much attention in optical communication.
A quantum channel is a map describing the state change from an initial
system to a final system, mathematically. Let us consider the construction
of the quantum channels.
Let 3.11 , 'Fl2 be the complex separable Hilbert spaces of an input and
an output systems, respectively, and let B (3.1k) be the set of all bounded
linear operators on 3.1k. 6 (3.1k) is the set of all density operators on "rtk
( I c = 112) : 6 (3.1k) { p E B ( ' F l k ) ; p 2 0, p = p* , t r p = 1)
(1) A map A* from the input system to the output system is called a
(2) The quantum channel A* satisfying the affine property (i.e.,
(purely) quantum channel.
h = 1 ( v h 2 0) * A* (Ck & P k ) = XI, XI, A* ( P k ) 1 v p k 6 (3.11)) is called a linear channel.
A map A from B (3.12) to B (3.11) is called the dual map of A* : 6 (3.11) -+
6 (3.12) if A satisfies
trpA (A) = trh* (p ) A (1)
for any p E G (3.11) and any A E B ( 'F l2) .
(3) A* from 6 (3.11) to 6 ( X 2 ) is called a completely positive (CP)
246
channel if its dual map A satisfies n
j , k = l
for any n E N, any Bj E B ('HI) and any AI, E B ('H2) I
One can consider the quantum channel for more general systems. The
input system denoted by (A, 6(A) ) and an output system by (A, B(A)) , where A (resp. 2) is a C*-algebra, 6(d) (resp. 6(A) ) is the set of all
states on A (resp. 2). When some outside effects should be considered in a certain physical
process such as noise and an effect of reservoir, it is convenient to extend
the system A to A @ B, where B describes the outside system. In such
cases, the concept of lifting introduced in
--- --
is useful.
(4) A lifting from A to A @ B is a continuous map
&* : 6 ( d ) + 6 ( d @ B ) . (3)
(5) A lifting €* is linear if it is affine.
(6) A lifting &* is nondemolition for a state 'p E 6 (A) if E * ' p ( A @ I ) = 'p(A) for any A E A.
This compound state (lifting) can be extensively used in the sequel
sections.
The concept of lifting came from the above cmpound state (nonlinear
lifting) l8 and the dual of a transition expectation (linear lifting) l , hence
it is a natural generalization of these concepts.
Let (0, Sn, P (0)) , (6, Sn, P (6)) be input and output probability
spaces, where Sn (resp. 86) is a a-algebra of R (resp. fi) and P(R) ,
P (6) are the sets of regular probability measures on s2 and 6. A channel E* transmitted from a probability measure to a quantum
state is called a classical-quantum (CQ) channel, and a channel S* from a
quantum state to a probability measure is called a quantum-classical (QC)
channel. The capacity of both CQ and QC channels have been discussed
in several papers 9321i26. A channel from the classical input system to the
classical output system through CQ, Q and QC channels is now denoted
-
by
P ( R ) 5 6 ( ' H ) I f B 7? 3 P R ( ) =* (7 One of the examples of the CQ channel Z* is given in 29 as follows:
247
CQ channel: Let C(R) be the set of all continuous functions on
0. For each w E R, we assume that pw E 6 (3-1) is - measurable.
Then the CQ channel E* is denoted by
JCl
for any p E P (0). Namely, one of the QC channel is given in 2 9 .
QC channel: Let {ITI,} be the set of all non negative Hermite
operators with C, ITn = I, which is called a positive operator val-
ued measure (POV). The QC channel Z* given by the measurement
process using {n,} is obtained by
x
n
for any u E 6 3-1 . (-1 1.1. Noisy optical channel
To discuss the communication system using the laser signal mathematically,
it is necessary to formulate the quantum communication theory being able
to treat the quantum effects of signals and channels. In order to discuss
influences of noise and loss in communication processes, one needs the fol-
lowing two systems 18. Let K l , K 2 be the separable Hilbert spaces for the
noise and the loss systems, respectively.
Quantum communication process is described by the following scheme 18
6 (3-11) + A* + 6 (3-12)
1 t
1 t Y* a*
6 ( ? i i B K i ) - - + r * --+ B ( 3 - 1 2 B K 2 )
The quantum channel A* is given by the composition of three mappings
a*, T*, y* such as
A* 3 a* o r * oy* . (6)
a* (a) = trlc2r (7)
a* is a CP channel from 6 (3-12 @I K 2 ) to 6 (3-12) defined by
248
for any cr E B (3-12 @ Kz), where trKz is a partial trace with respect to
K2. r* is the CP channel from 8 ('HI @ K1) to 6 ('Ha 8 I c 2 ) depending on
the physical property of the device. y* is the CP channel from B (3-12) to
B (3-11 @ I c l ) with a certain noise state E E B (&) defined by
7 * ( p ) = L J @ t (8)
for any p E B ('HI). The quantum channel A* with the noise < is written
by
A* (PI = t r K z r* (P 8 <) (9)
for any p E B ( 'H I ) .
a noisy optical channel if 7r* and E above are given by
Here we briefly review noisy optical channel 27. A channel A* is called
E = Im) (mi and r* (.) = V (-) V*, (10)
where Im) (ml is m photon number state in 'HI and V is a linear mapping
from X1 @ K 1 to 'H2 8 K 2 given by
n+m
v (in) 8 Im)) = C cj"ym l j ) 8 1n + m - j > , (11) j=O
K dn!m!j! (n + m - j ) ! (ym- j+2r n+j-2r Cj"'" = c (-1)"- (4 r! (n - r ) ! ( j - r ) ! (m - j + r ) !
r=L (12)
for any In) in 3-11 and K 3 min{j,n), L = max(j - m, 0), where a and
p are complex numbers satisfying laI2 + IpI2 = 1 , and 7 = IaI2 is the
transmission rate of the channel. In particular, p @ is given by the tensor
products of two coherent states (0) (81 @ 16) (61, then r* ( p 8 E ) is obtained
by
T* ( ~ 8 0 = p + p K ) (ae + p ~ j 8 I-,& + G K } (-Be +SK;J .
7r* is called a generalized beam splittings. It means that one beam
comes and two beams appear after passing through r*. Here we remark
that the attenuation channel At; l8 is derived from the noisy optical channel
with m = 0. That is, the attenuation channel At; was formulated in1' on
1983 such as
Ag(P) = t rKc ,K; ( P @3 63) (13)
(14) = trlczVo ( p @ 10) (00 G,
249
where 10) (01 is the vacuum state in B(Ic1), VO is the mapping from 7 d 1 8 l C 1
to 7-l~ 8 I c 2 given by
This attenuation channel is most important channel for discussing the opti-
cal communication processes. After that, Accardi and Ohya reformulated
it by using liftings, which is the dual map of the transition expectation
by mean of Accardi. It contains the concept of beam splittings, which is
extended by Fichtner, F'reudenberg and Libsher concerning the mappings
on generalized Fock spaces. The generalized beam splittings on symmetric
Fock space was formulated by using the compound Hida-Malliavin deriva-
tive and the compound Skorohod integral.
2. Ohya S-mixing entropy and Ohya mutual entropy
In order to discuss some physical phenomena, for instance, phase transi-
tions, we had better start without Hilbert space, so that we need to formu-
late the entropy of a state in a C* system 20J2J5.
Let (A, G ( d ) , a(R)) be a C*-dynamical system and S be a weak* com-
pact and convex subset of e(d). For instance, S = e(d); S = I ( a ) , the
set of all invatiant states for a; S = K ( a ) , the set of all KMS states.
Every state 'p E S has a maximal measure p pseudosupported on exS (the set of all extreme points of S) such that
$0 = s, W d P . (17)
The measure p giving the above decomposition is not unique unless S is a
Choquet simplex, so that we denote the set of all such measures by M,(S). Put
D, ( S ) = { M , ( S ) ; 3 , u k c IR+ and { p k } c exS (18)
k k
where b((p) is the Dirac measure concentrated on an state (p, and define
H ( p ) - p k 1% PIC (20) k
250
for a measure p E DIp(S). Then the entropy of a state 'p E S w.r.t. S is
defined by
which is called Ohya S-mxinig entropy. This entropy is an extension of
von Neumann's entropy 16, and it depends on the set S chosen. Hence it
represents the uncertainty of the state measured from the reference system
S. When S = B(d), we simply denote S6(d)(p) by S('p) in the sequel.
When A =B ('HI), the quantum entropy is denoted by
S ( p ) = -trplogp
for any density operators p in 6 ( 'H I ) , which was introduced by von Neu-
mann around 1932 16. It denotes the amount of information contained in
the quantum state given by the density operator.
The mutual entropy was first discussed by Shannon to study the in-
formation transmission in classical systems and its fully general quantum
version was formulated in 'O. The classical mutual entropy is determied
by an input state and a channel, so that we denote the quantum mutual
entropy with respect to the input state p and the quantum channel A* by
I (p ; A*) . This quantum mutual entropy I (p ; A*) should satisfy the follow-
ing three conditions:
(1) If the channel A* is identity map, then the quantum mutual entropy
equals to the von Neumann entropy of the input state, that is,
( 2 ) If the system is classical, then the quantum mutual entropy equals
( 3 ) The following fundamental inequalities are satisfied:
I (p ; id) = s (p) .
to the classical mutual entropy.
0 I I ( p ; A*) 5 s (P ) .
In order to define such a quantum mutual entropy, we need the quantum
relative entropy and the joint state, which is called Ohya compound state, describing the correlation between an input state p and the output
state A*p through a channel A*. Ohya compound state D E (corresponding
to joint state in CS) of p and R*p was introduced in l8l1', which is given
by
251
where E derives from a Schatten decomposition 32 (i.e., one dimension or-
thogonal decomposition of p ) { p = En XnEn} of p. Ohya mutual entropy with respect to the input state p and the quantum channel A* was defined
in l8 such as
I (P ; A*) = SUP s (w, g o ) , (23) E
where 00 = p 8 A * p and S (GE, g o ) is the quantum relative entropy defined
by Umegaki 37,13 as follows:
t r p (log p - log G) (when ranp c EEF) (otherwise)
There were several trials to extend the relative entropy to more general
quantum systems and apply it to some other fields 2338322. This mutual
entropy I (p ; A*) satisfies all conditions (1)-(2) mentioned above, and
it satisfies also condition (3) the Shannon's type inequality as follows:
0 5 I ( p , A * ) 5 min{S(p) , S ( A * p ) } . It represents the amount of infor- mation correctly transmitted from the input quantum state p to the out-
put quantum system through the quantum channel A*. It is easily shown
that we can take orthogonal decomposition instead of the Schatten-von
Neumann decomposition 32. Ohya mutual entropy is completely quantum,
namely, they describe the information transmission from a quantum input
to a quantum output. When the input system is classical, the state p is
a probability distribution and the Schatten-von Neumann decomposition
is unique with delta measures 6, such that p = En Andn. In this case we
need to code the classical state p by a quantum state, whose process is a
quantum coding described by a channel r* such that r*6, = crn (quan-
tum state) and 0 E F * p = En Xngn. Then Ohya mutual entropy I ( p ; A*) becomes Holevo's one, that is,
I (p; A* o I'*) = S (A*g) - C XnS (A*G,) n
when En XnS (A*crn) is finite.
Let (A, B(d), Q (G)) be a unital C*-system and S be a weak* compact
convex subset of B(A). For an initial state 'p E S and a channel A* :
6 (d) 4 B (B), two fundamental compound states are
= 1 w 8 A*wdp, (25)
'p 8 R*p. (26)
252
Gf is called Ohya compound state, which expresses the correlation
between the input state cp and the output state A*cp. Then I*
(27)
(28)
(29)
&*cp = Gp S
I f (p; A*) L= S (a:, Go)
Is (cp; A*) E sup { 1; (9; A*) ; p E Mp (S)} .
I (cp; A*) = C phS(A*w, A*cp).
is a nondemolition lifting. Ohya mutual entropy w.r.t. S and p is
and Ohya mutual entropy w.r.t. S is defined as
When a state cp is expressed as cp = xk pkcpk (fixed), the mutual entropy
is given by
(30) k
If A : B 4 A is a unitial completely positive mapping between the
algebras A and B, that is, the dual A* is a channeling transformation from
the state space of A into that of B, then
S(A*cpl, A*cpz) 5 S(cp1, cpd (31)
Let A : B + A be completely positive unitial mapping and cp be a state of
23. So 'p is an initial state of the channel A*. The quantum mutual entropy
is defined after l8 as
where the least upper bound is over all orthogonal extremal decompositions.
Now we show two theorems proved in 26 according to the Holevo bound.
Let us define an extension of the functional of the relative entropy. If A and B are two positive Hermitian operators (not necassarily the states, i.e.
not necessarily with unit traces) then we set
S(A, B) = trA (log A - log B ) . (33)
There is the following Bogoliubov inequality.
Theorem 2.1. 26 One has
S(A, B) 2 trA (log trA - log trB) (34)
253
The following theorem gives us the bound of the mutual entropy
I (p ; A* o y*) , where y* is a classical -quantum channel.
Theorem 2.2. 26 For a probability distribution p = { A , } and a quantum coded State U = y*p = 1, One has the following inequality for any quantum channel decomposed as A* = A; o A; such that A ~ u s x i EpE i by a projection valued measure {E i } :
&Uk , 2 0, ck
In the case that the channel A*, is trivial; A*,a = 0, the above inequality
reduces to the bound obtained by Holevo ':
k
3. Comparison of various quantum mutual type entropies
3.0.1. Coherent Entropy and Lindblad entropy
Let us discuss the entropy exchange 3 4 9 1 4 . For a state p, a quantum channel
A* is defined by an operator valued measure {A j } such as
j
Then define a matrix W = (Wij)i,j with
254
by which the entropy exchange is defined by
Se ( p , A*) = -trW log W.
Using the above entropy exchange, two mutual type entropies are de-
fined as below and they are applied to the study of quantum version of
Shannon's coding theorem 3 3 9 1 0 1 1 1 @ 1 4 1 3 1 3 3 6 . The first one is called the co- herent entropy IC ( p ; A*) 35 and the second one is called the Lindblad entropy IL ( p ; A*) 6 , which are defined by
IC ( P ; A*) E S (A*p) - s e ( P I A*) 1
I L ( P ; A*) S ( P ) + s (A*P) - Se ( P , A*) .
By comparing these mutual entropies for quantum information commu-
nication processes, we have the following theorem 30:
Theorem 3.1. When { A j } is a projection valued measure and dim Aj = 1
for arbitrary state p we have (1) 0 5 I ( p , A*) 5 min {S (p) , S (A*p)} (Ohya mutual entropy), (2) IC ( p , A*) = 0 (coherent entropy), (3) IL ( p , A*) = S ( p ) (Lindblad entropy).
From this theorem, Ohya mutual entropy I (p , A*) only satisfies the
inequality held in classical systems, so that only Ohya mutual entropy can
be a candidate as quantum extension of the classical mutual entropy. Other
two entropies can describe a sort of entanglement between input and output,
such a correlation can be also described by quasi-mutual entropy, a slight
generalization of I ( p , A*) ,discussed in ' O s 5 .
4. Quantum capacity
The capacity of purely quantum channel was studied in 2 3 1 2 1 1 2 6 1 2 4 , 2 8 , 2 9 .
Let S be the set of all input states satisfying some physical conditions.
Let us consider the ability of information transmition for the quantum chan-
nel A*. The answer of this question is the capacity of quantum channel
A* for a certain set S ~6 (XI) defined by
Cf (A*) = sup { I (p ; A*) ; p E S } (39)
255
When S =6 (XI), the capacity of quantum channel A* is denoted by
C, (A*) . Then the following theorem for the attenuation channel was proved
in 23.
Theorem 4.1. 23 For a subset S, = { p E 6 (3-11) ;dims ( p ) = n} , the ca- pacity of the noisy optical channel A* satisfies
C? (A*) = logn,
where s ( p ) is the support projection of p.
When the mean energy of the input state vectors { I d k ) } can be taken
infinite, i.e.,
the above theorem tells that the quantum capacity for the noisy optical
channel A* with respect to S, becomes logn. It is a natural result, how-
ever it is impossible to take the mean energy of input state vector infinite.
Therefore we have to compute the quantum capacity
C,S. (A*) = SUP { I (pi A*) ; p E Se} (40)
under some constraint Se = { p E S; E ( p ) < e} on the mean energy E ( p ) of the input state p. In 2 0 9 2 3 , we also considered the pseudo-quantum capacity Ct; (A*) defined by
cP"p. (A*) = SUP { l p (P; A*) ; P E se} (41)
with the pseudo-mutual entropy Ipp (p ; A*)
Ipq ( p ; A*) = sup c ( A * p k , A*p) ; p = c x k p k , finite decomposition , k I
(42) { k
where the supremum is taken over all finite decompositions instead of all
orthogonal pure decompositions for purely quantum mutual entropy. A pseudo-quantum code is a probability distribution on B(7-L) with finite
support in the set of product states. so { ( A h ) , ( p k ) } is a pseudo-quantum
code if ( A , ) is a probability vector and p k are product states of B(3-1). The quantum states p k are sent over the quantum mechanical media, for
example, optical fiber, and yield the output quantum states A * p k . The per-
formance of coding and transmission is measured by the pseudo-mutual entropy (information)
I p q ( ( A k ) , ( P k ) ; A * ) = Ipq @;A*) (43)
256
with p = ck &Pk. Taking the supremum over certain classes of pseudo-
quantum codes, we obtain various capacities of the channel. The supremum
is over product states because we have mainly product (that is, memoryless)
channels in our mind. Here we consider a subclass of pseudo-quantum
codes. A quantum code is defined by the additional requirement that
{ p k } is a set of pairwise orthogonal pure states 18. However the pseudo-
mutual entropy is not well-matched to the conditions explained in Sec.3,
and it is difficult to compute numerically 24. From the monotonicity of the
mutual entropy 22, we have
In order to estimate the quantum mutual entropy , we introduce the
concept of divergence center. Let {wi : i E I } be a family of states and
R > 0. We say that the state w is a divergence center for {wi : i E I } with
radius 5 R if
S(w i ,w) I R for every i E I .
In the following discussion about the geometry of relative entropy (or diver-
gence as it is called in information theory) the ideas of can be recognized
very well.
Lemma 4.1. 23 Let ((A,), ( P k ) ) be a quantum code fo r the channel A* and w a divergence center with radius 5 R for { A * p k } . Then
Ipr?((~k), ( P k L A*) 5 R.
Lemma 4.2. 23 Let $0, $1 and w be states of B(K) such that the Hilbert space K is finite dimensional and set $Jx = (1 - X)$O + X$1 (0 5 X I 1). If S(&, w ) , S($J1, w ) are finite and
S($JX, w> 2 S($l , w ) (0 6 I 1)
S ( $ l , W ) + S ( $ O , $ J l ) I S($O,W).
then
Lemma 4.3. 23 Let {wi : i E I } be af in i te set of states of B(K) such that the Hilbert space K is finite dimensional. Then the exact divergence center is unique and it is in the convex hull on the states w i .
Theorem 4.2. 23 Let A* : G(3-1) + 6(K) be a channel with finite dimen- sional K. Then the capacity Cp( A*) is the divergence radius of the range of A*.
257
References
1. L. Accardi and M. Ohya, Compond channnels, transition expectation and lzftings, Appl. Math, Optim. 39, 33-59, 1999.
2. H. Araki, Relative entropy fo r states of von Neumann algebras, Publ. RIMS Kyoto Univ. 11, 809-833, 1976.
3. I. Csiszar, I - divergence geometry of probability distributions and minimiea- t ion problems, Ann. Probab. 3, 146-158, 1975.
4. H. Barnum, M.A. Nielsen and B.W. Schumacher, Information transmission through a noisy quantum channel, Physical Review A, 57, No.6, 4153-4175, 1998.
5. V.P. Belavkin and M. Ohya, Quantum entanglements and entangled mutual entropy, Proc. R. SOC. Lond. A. 458, 209-231, 2002.
6. C.H. Bennett, P.W. Shor, J.A. Smolin and A.V. Thapliyalz, Entanglement- Assisted. Capacity of a Quantum Channel and the Reverse Shannon Theorem, quant-ph/0106052.
7. K.H. Fichtner, W. Freudenberg and V. Liebscher, Beam splittings and t ime evolutions of Boson systems, Fakultat fur Mathematik und Informatik, Math/ Inf/96/ 39, Jena, 105, 1996.
8. R.S.Ingarden, Quantum information theory, Rep. Math. Phys., 10, 43-73, 1976.
9. A.S.Holevo, Some estimates fo r the amount of information transmittable by a quantum communication channel ( in Russian), Problemy Peredachi Infor- macii, 9, 3-11, 1973.
10. R. Jozsa and B.W. Schumacher, A new proof of the quantum noiseless coding theorem, Journal of Modern Optics, 41, 2343-2349, 1994.
11. C. King and A. Lesniewski, Quantum source and quantum coding theorem, Journal of Mathematical Physics, 39 ( l ) , 88-101, 1998.
12. L.B.Levitin, Physical information theory f o r 30 years: basic concepts and results, Springer Lect. Note in Phys., 378, 101-110, 1991.
13. G.Lindblad, Entropy, information and quantum measurements, Commun. Math. Phys., 33, 111-119, 1973.
14. S. Lloyd, Capacity of the noisy quantum channel, Physical Review A, 55, 1613, 1997.
15. N.Muraki, M.Ohya and D.Petz, Note o n entropy of general quantum systems, Open Systems and Information Dynamics, 1, No.1, 43-56, 1992.
16. von Neumann, Die Mathematischen Grundlagen der Quantenmechanik, Springer-Berlin, 1932.
17. M.Ohya, Quantum ergodac channels in operator algebras, J. Math. Anal.
18. M. Ohya, O n compound state and mutual information in quantum informa- t ion theory, IEEE Trans. Information Theory, 29, 770-777, 1983.
19. M.Ohya, Note o n quantum probability, L. Nuovo Cimento, 38, 402-406, 1983. 20. M. Ohya, Some aspects of quantum information theory and their applications
to irreversible processes, Rep. Math. Phys. 27, 19-47, 1989. 21, M. Ohya, Fundamentals of quantum mutual entropy and capacity, Open Sys-
Appl. 84, 318-327, 1981.
258
tems and Information Dynamics, 6. No.1, 69-78, 1999. 22. M. Ohya and D. Petz, Quantum Entropy and Its Use, Springer, 1993. 23. M. Ohya, D. Petz and N. Watanabe, On capacity of quantum channels, Prob-
ability and Mathematical Statistics, 17, 179-196, 1997. 24. M. Ohya, D. Petz and N. Watanabe, Numerical computation of quantum
capacity, International J. Theor. Phys., 38 , No.1, 507-510, 1998. 25. M. Ohya, D. Petz and N. Watanabe, Quantum capacity of a noisy optical
channel, TUS preprint. 26. M. Ohya, I. Volovich, On quantum capacity and its bound, Infinite Dimen-
sional Analysis, Quantum Probability and Related Topics, 6, No.2, 301-310, 2003.
27. M. Ohya and N. Watanabe, Construction and analysis of a mathematical model in quantum communication processes, Electronics and Communica- tions in Japan, Part 1, 68, No.2, 29-34, 1985.
28. M. Ohya and N. Watanabe, Quantum capacity of noisy quantum channel, Quantum Communication and Measurement, 3, 213-220, 1997.
29. M. Ohya and N. Watanabe, Foundation of Quantum Communication Theory (in Japanese), Makino Pub. Co., 1998.
30. M. Ohya and N. Watanabe, Comparison of mutual entropy - type measures, TUS preprint.
31. D. Petz and M Mosonyi, Stationary quantum source coding, Journal of Math- ematical Physics, 42, 4857-4864, 2001.
32. R. Schatten; Norm Ideals of Completely Continuous Operators, Springer- Verlag, 1970.
33. B.W. Schumacher, Quantum coding, Physical Review A, 51, 2738-2747, 1995. 34. B.W. Schumacher, Sending entanglement through noisy quantum channels,
Physical Review A, 54, 2614, 1996. 35. B.W. Schumacher and M.A. Nielsen, Quantum data processing and error
correction, Physical Review A, 54, 2629, 1996. 36. P. Sho, The quantum channel capacity and coherent information, Lecture
Notes, MSRI Workshop on Quantum Computation, 2002. 37. H. Umegaki, Conditional expectations in an operator algebra IV (entropy and
information), Kodai Math. Sem. Rep. 14, 59-85, 1962. 38. A.Uhlmann, Relative entropy and the Wigner- Yanase-Dyson-Lieb concavity
in interpolation theory, Commun. Math. Phys., 54, 21-32, 1977.
259
WHITE NOISE CALCULUS AND STOCASTIC CALCULUS
L. ACCARDI, ANDREAS BOUKAS*
AMS Mathematics Subject Classification 2000 : 60H40
1. Introduction
The plan of this paper is the following.
In section (2), after recalling some basic points of the structure theory
of classical stochastic process, we illustrate how quantum probability can
substantiate Hida’s vision about the ”elementality” of white noise, i.e. that
any other (purely non deterministic) process can be in some sense “built”
from it.
In section (3) we briefly mention how the idea of white noise approach
to stochastic calculus was suggested by the stochastic limit of quantum
theory and how it lead to the idea of developing a calculus for the non
linear powers of white noise. In section (4 this idea is illustrated in the case
of the square of white noise).
In section (5) we show how the classical processes, unified by the square
of white noise emerged in different contexts of classical probability, math-
ematical physics, and statistics.
In section (6 ) we describe our attempt, developed in past years, to
extend the results obtained for the square to higher powers of white noise.
The obstructions which make this dream hard to achieve (the so called no
go theorems) are quickly reviewed in section (7).
Finally in the Appendix, following the original paper [Meix34], we give
a proof of Meixner’s classification theorem.
‘centro vito volterra, universith di roma tor vergata, accardivolterra.mat.uniroma2.it web page: http://volterra.mat.uniroma2.it Work done within the european union research training network (rtn): “quantum probability - applications” web site: http://hyperwave.math-inf.uni- greifswald.de/algebra/qpapplications
260
2. Elemental processes
The structure theory of classical stochastic processes is one of the highlights
of classical stochastic analysis and it has been accomplished through the
work of generations of probabilists. Cornerstones in this long and broad
avenue have been the Wald decomposition; the De Finetti, Kolmogorov,
Levy, Khintchin structure theory of independent increment processes; P. Levy martingale representation theorem and his innovation equation
J X ( t ) = @ ( X ( S ) ; S 5 t ,Y (s ) , t , d t )
which opened the way to Ito stochastic calculus and the corresponding
theory of stochastic differential equations; the Hida-Cramer representation
theory for Gaussian processes; the Kunita-Watanabe extension of Levy
martingale representation theorem to non continuous trajectory martin-
gales; the systematic work by Doob, Meyer and many others, . . . The intuitive picture emerging from these developments is that any
"generic" stochastic process (i.e. whose trajectories are not too irregular)
can be built starting from:
(i) deterministic processes
(ii) distribution derivatives of stationary independent increment pro-
in the sense that it can be decomposed as a sum of integrals of such
Moreover, any stationary independent increment process { Zt >, in its
cesses {&>
processes.
turn, has the following LQvy-Ito decomposition :
Zt=mt+aBt+Xt
where
m is a constant
Bt is a Brownian motion
X t is a compound Poisson process
Finally any compound Poisson process X t can be expressed as an inte-
gral of independent Poisson processes P,,t with intensity of jumps equal to
U
and where the baricentric measure of this decomposition d P ( u ) is called the
Levy measure and has support in R \ (0). These decompositions justify Hida's terminology according to which the
i.e.r.v's (idealized elemental random variables ) are:
261
{&) the distribution derivatives of the standard Brownian motion
{Pu,t} the distribution derivatives of the standard Poisson process with
and this fact can be symbolically expressed by the formula:
intensity u
xt = f ( t , { B ) , {Pu)) = f ( t , {& , s E R) ; {Pu,s , s E R, u E IW \ (0 ) ) ) (1)
expressing (in a unique way) the stochastic process X t as a (non-
(i) the standard deterministic processes: t H t (ii) the standard white noise: B, (iii) the standard Poisson processes with intensity u E R \ (0): Pu,s, Clearly equation (1) is purely symbolic because both 8, and Pu+ are
random variable valued distributions and there is no natural way to define
a nonlinear function of a distribution.
Notice moreover that, at this level of development, the standard white
noise: B, and the standard Poisson densities P,,, appear on a totally equal level!
In the mid 1970’s Hida initiated his programme on white noise analysis
from which slowly it begun to emerge a more radical vision, namely:
there is only one elemental process: standard white noise! I call this a ”vision” because it was never formulated as a precise math-
ematical conjecture, nevertheless the outstanding role of white noise is a
constant in Hida’s mathematical thought [HidaOl]. This is indeed a bold
vision, because apparently the noises B, and P,,, look like totally disjoint
objects!
The first substantial support to Hida’s vision came from Hudson and
Parthasarathy discovery [HuPa84b] that, from the quantum probabilistic
point of view, the standard Poisson process with intensity u can be consid-
ered as a ”function” of the quantum white noise. More precisely (cf. the
end of this section for the definitions involved):
Theorem: Let P,,t denote the classical scalar valued standard Poisson
process with intensity u. Then
random) functional of:
P(2L,t) = &LBt + &B,f + B$Bt - This achievement was an important event for probability and created a
great impression in classical probabilists such as P.A. Meyer, because it
showed that the Brownian motion and the Poisson process which, in clas-
sical probability, are not apparently connected, in quantum probability be-
come expressible as sums of three fundamental objects (two if one consid-
262
ers, as usual in physics, the pair {&, B,f} as a single object: the quantum
Brownian motion)
’Notice however that the number process and the quantum Brownian
motion are independent in the very strong sense that a sum of stochas-
tic integrals over these processes is zero if and only if each integrand is
zero. In other terms: the white noise point of view is necessary in order to
achieve the full reductionistic programme (Hida’s vision) of expressing the
two building blocks of classical stochastic processes (Wiener and Poisson)
as functions of a single more fundamental object: the quantum white noise.
Notice in addition that, in order to achieve this reduction one needs the
”square modulus of the white noise” B$Bt = (which is for the num-
ber process what the white noise is for Brownian motion). This ”quadratic
functional of the white noise” is the first example of the basic role played
by the ”nonlinear powers the white noise”.
Using the above reduction Parthasarathy [Parth92] proved that any
infinitely divisible distribution can be represented as vacuum distribution
of a stochastic process in the space of the Boson Fock Brownian motion.
This result has now been developed into one of the basic new ideas
of quantum probability: the quantum decomposition of a classical random variable [AcBo97].
Another important step in the direction of Hida’s vision will be discussed
in the following section.
3. White noise approach to classical and quantum stochastic calculus
The white noise approach to classical and quantum stochastic calculus came
out from the experience accumulated in the solution of many concrete prob-
lems in physics and in mathematics. The main steps in this development
are:
1) The development of quantum stochastic calculus: Hudson and
Parthasarathy (1982)
2) The proof that quantum stochastic differential equations are ”fast
time” (t 4 t /A2) limits of Hamiltonian equations: Accardi, Frigerio, Lu
(1987)
3) The proof that white noise Hamiltonian equations are ”fast time”
limits of Hamiltonian equations: Accardi, Lu, Volovich (1993)
4) The combination of 2) and 3) into the statement that (classical and
quantum) stochastic equations can be expressed as (causally) normally or-
263
dered forms of (classical and quantum) white noise Hamiltonian equations
[AcLuVo99], IAcLuVo021.
The equivalence between white noise Hamiltonian equations and
stochastic differential equations is quite nontrivial and the connection be-
tween the coefficients of the two types of equations is strongly nonlinear.
The advantage of the white noise Hamiltonian equations over their (clas-
sical or quantum) stochastic equivalent is that the formal unitarity condi-
tions are the obvious ones, used by every physicist, which correspond to the
formal self-adjointness of the Hamiltonian. The relationship expressing the
coefficients of the stochastic equation as (nonlinear) functions of the coeffi-
cients of the corresponding white noise Hamiltonian equation explains the
deep origins of the Hudson-Parthasarathy unitarity conditions and gives a
"microscopic" interpretation of them in terms of the original (non stochas-
tic neither white noise) Hamiltonian equations of which the stochastic and
white noise equations are approximations.
A systematic development with several new results can be found in the
PhD thesis of Wided Ayed [AyedOS].
Recall that the standard scalar valued, classical white noise is a classical
mean zero, Gaussian, operator valued distribution process with variance
(W,Wt) = 6( t - s ) . This implies that the standard white noise is a stationary, additive, inde-
pendent increment process.
Definition: A quantum stochastic process is a family
z = {z( t ) : t E Rd}
of Hilbert space operators. Such a process is said to be classical if (i) For all t 2 0, each z(t) is a QM "observable" i.e
z(t) = z(t)* ; w
[z(t), z(s)] = z(t)z(s) - z(s)z(t) = 0
(ii) For all t , s,
the commutator (here and in the following) will always be meant weakly
on some dense domain.
Definition: A boson Fock (&dimensional) white noise (equivalently
called, in the physical literature, a free boson Fock field over L2(Rd) ) is
a pair bt, b$ of operator valued distributions acting on a Hilbert space 3-1, satisfying the commutation relations:
[bt , b:] = 6( t - s) t E Rd [algebra]
264
and such that there exists a unit vector 6, E ‘FI satisfying:
bt6, = 0 [Fock prescription].
For the precise meaning of these expressions, including domains, etc. ... cf.
[AcAyOuOS].
The definition above best illustrates one of the basic general principles
of QP namely that:
algebra implies statistics. In fact the above definition of white noise, which is purely algebraic,
implies the Brownian motion statistics in the sense that it implies that
the random variables
Wt-W, := dU(b:+b,) ; Wo = O I’ mutually commute, are independent on disjoint intervals and
(a, eiWt+j = (a , , iS , tds(b , f+bs)+) = e - 7 t 2
equivalently:
Wt = Bt + B$ = d s ( b z + b,) I’ is the increment process of a classical Brownian motion (this is the quantum
decomposition of the classical Brownian motion). Similarly one has the
quantum decomposition of the classical white noise
W t = bt + b,f
In this sense one can say that in classical probability wt is ”elemental”
(atomic), but in quantum probability it is not.
The connection between (the increment process of) quantum Brownian
motion and quantum white noise is the same as in the classical case:
t t B$ = I dsb: , Bt = I dsb,.
Conversely a naive approach would suggest the conjecture that the relation
between quantum Brownian motion and quantum white noise is:
d , btf := -B$. d d t d t
bt := -Bt
However application of this rule without constraints leads to the wrong
results as illustrated by the Hamiltonian white noise equations.
265
4.
Given the identification between the standard (d-dimensional) quantum
white noise and the free boson Fock fields over L2(Rd), one can formulate
what has been one of the fundamental unsolved problems of theoretical
physics since over 30 years:
to give a reasonable construction of ”local powers of free quan- tum fields”.
This means that one would like to associate well defined mathematical
objects to symbolic expressions of the form:
Nonlinear powers of white noise
in such a way to preserve as many properties as possible of the discrete
approximations of these objects (which are well defined!). In other terms we
would like to give a meaning to the ill defined ”powers of an operator valued
distributions”. Since these objects are very singular some ”renormalization”
is needed to achieve this goal.
In the literature one can find many ”renormalization techniques”. The
usual technique consists in introducing appropriate cut-offs and then trying
to remove them with some limiting procedure and after having subtract
some quantities tending to m. The main problem met up to now with this
technique is the following dichotomy. After renormalization the resulting
field is:
(i) either trivial (Gaussian or simple perturbation thereof)
(ii) or completely uncontrollable
This dichotomy is well illustrated in the paper by Segal [Sega’lOa] which
is devoted to the simplest renormalization problem: that involving the
squares of quantum fields.
Segal proves that the square of the usual time zero, scalar, Fock, Klein-
Gordon field on E d , cannot correspond to a self-adjoint operator acting on
the same Fock space of the field unless d = 1. However his techniques are
not able to provide:
- any information where this operator lives
- any information on its spectral distribution in interesting states
Thus, even for squares of fields the situation was very obscure until
1999. In 1999 a new idea was introduced in this problem. The main point
of this new idea was to postulate new, renormalized, commutation relations
for the higher powers of white noise and then construct some Hilbert space
realization of them. More precisely: one starts from the standard boson
266
algebra
[bt, b:] = 6( t - S )
and computes formally the higher powers commutators
[b!, b:kl
combining the 1-st order commutation relations with the renormalization
rule:
a(t - s ) ~ = cn-16(t - s) V n E N \ (01 Then one takes the result of this manipulation as the definition of the Lie
algebra of the renormalized higher powers of white noise.
The second much more difficult step consists in trying to construct some
Hilbert space realization, for example by introducing the Fock prescription:
b;‘“b:@=O ; h > l
which, also in this case, uniquely determines the statistics.
For the second powers, in the Fock case, this programme was realized
in [AcLuVo99] and led to the definition of the renormalized square of white
noise (RSWN) as the Lie algebra with generators
b$ lL=” /dtcp(t)b: ; b, = (b$)+
nrp lL=” / dtcp(t)b$bt
where cp is some test function, and Lie brackets given by the commutation
relations:
[b b$I = Y(cp7 $) + nq*
b,, b;ZI = 2 q *
(b$)+ = b , ; n: = nv
where y is a strictly positive constant called ”the renormalization constant”.
The Fock representation of this Lie algebra consists in realizing these
operators on a Hilbert space ‘H with a unit vector @ satisfying:
b,@=O.
267
Theorem: (Accardi, Lu, Volovich 1999) The Fock representation of the
renormalized square of white noise (RSWN) exists.
Soon after this result it was recognized by Accardi, Franz and Skeide
[AcFrSkOO] that the Lie algebra of the renormalized square of white noise
is isomorphic to (a central extension of) sZ(2, R). This has 3 generators
B - , B + , M
and relations
[B-,B+] = M
[M, B*] = f2B*.
The paper [AcFrSkOO], combining the well developed theory of unitary rep-
resentations of sZ(2, R) with Schiirmann’s representation theorem for quan-
tum independent increment processes [Schii93], obtains:
(i) a classification of all the unitary representations of this current alge-
bra satisfying an irreducibility condition
(ii) a classification of all the classical sub-processes of the RSWN as well as the identification of their vacuum distributions.
An unexpected fall out of this identification was an unexpected connec-
tion with the Meixner classes [Meix34].
In order to illustrate this connection let us first consider the case of
the 1-st order quantum white noise b$, bt and the family of classical (self-
adjoint) processes that can be obtained by linear combinations from it and
the numbers process bzbt. A simple calculation shows that these processes
are of the form
z(t) = a + z B(t ) + + PN(t)
where a, p E R and z E C. In more usual white noise notations:
k ( t ) = Q! + z bt + .Zb$ + ,f3b$bt . The constant part a is central and we omit it. Also the case z = 0 is trivial
and we omit it. Then, up to the ”time rescaling” t H t / lz l , the ”gauge
transformation” bt H eiebt, where z = lzleie and a renaming of p, we are
left with the family
k ( t ) = bt + b z + ,BN(t) . Remembering the Hudson-Parthasarathy decomposition of the white and
Poisson noises, we see that, depending on the parameter p, this family only
268
1 critical case, namely: p = 0 which corresponds to classical scalar valued
standard Brownian motion
x ( t ) = b( t ) + b+(t).
A=+
z(t) = JS;B(t) + A B + ( t ) + N( t )
If /3 # 0 then, fixing a complex square root of
after another rescaling and gauge transformation we obtain:
which is the classical scalar valued standard Poisson process with intensity
A. A similar calculation applied to the RSWN shows that the family of
classical processes that can be obtained by linear combinations from the
2-d order processes b z 2 , bf and the numbers process b$bt has the form:
z(t) = a + zb,2 + Eb,+2 + pb,+bt
where a , P E R and z E C. Again up to centering, rescaling and a gauge
transformation:
xp( t ) := bZ2 + b," + pbzb, = B$ + Bt + PNt
where p is a real number. It can be proved that here there are 2 critical
cases:
p = f 2
the value +2 corresponding to the square position (classical) white noise,
i.e.
I b,++bt 1 2 = b,+2+b,2+b,+bt+btbt++ = by+b,2+2b;bt+s(o) 3 b,+2+b,2+2b,+bt
and the value -2 to the renormalized square of the momentum white noise,
i.e.
(b,+ - b t ) / i
the vacuum distribution of both these critical processes is the Gamma-
distribution
whose parameter mo > 0 is uniquely determined by the choice of the uni-
tary representation of sZ(2, R) corresponding to the representation of the
renormalized square of white noise algebra [ACFRSKOO] .
269
In this functional realization the number vectors become the Laguerre
polynomials which are orthogonal for the gamma distribution.
Since the Gamma-distributions are precisely the distributions of the
X2-random variables, this result confirms the naive intuition that the dis-
tribution of the renormalized square of white noise should be a Gamma-
distributions
For IpI < 2 the jumps are not strong enough and one still has a density
where C is a normalization constant. The orthogonal polynomials corre-
sponding to this probability measure are the of the second kind, or Meixner-
Pollaczek polynomials.
For mo integer or half-odd there are explicit formulae for the densities
due to Grigelionis:
7rx(1+ x2). . . ( (n - 1 ) 2 + x2) Ir (n + ix)12 = , n = 1 , 2 , . . . , , X € R sinh(nx)
2 IT
, x € R l r ( i + i x ) l = cosh(7rx)
.(a + 2). . . ( (n - 1)n + a + z2) , n = 1 , 2 ) . . . ; 2 E R . 1
cosh( ~ x )
Finally, for > 2 the jumps dominate and the probability measure is
atomic, namely the negative binomial (Pascal) distribution:
where ( r n ~ ) ~ denotes the Pochammer symbol
(mo)n = mo(m0 + 1) . . . (mo + n - 1)
and
270
i f p > + 2
if /3 < -2.
binomial) distribution are the Meixner polynomials of the first kind:
The orthogonal polynomials associated to a centered Pascal (negative
if /3 > +2 and
ir n+mo- l
k=l
5. Emergence of the square of white noise in different contexts
The connection between the gamma processes and the current representa-
tions of SL(2, R) was studied in [TsiVeYoOl] independently of [AcFrSkOO]
(where this connection was established for all the Meixner classes).
The Lkvy processes, corresponding to the Pascal measures were intro-
duced by Bruss and Rogers [BruRoSl] in the context of optimal selection
strategies based on relative ranks, when the total number of options is
unknown.
In the paper [GrigOl] Grigelionis uses the term “Meixner distribution” to
denote the class of probability measures on R whose characteristic function
(Fourier transform of the probability density) has the form
with z E R, -7r < p < 7r, S > 0 , p E R. This class of probability measures
was introduced by Schoutens and Teugels [SchTeu98] who established their
connection with the Meixner-Pollaczek polynomials and proved that the
measures in this class correspond to Levy processes (the explicit construc-
tion of the Fock representation in [AcLuVoSg] can be considered a different
proof of this result).
The papers by Nualart and Schoutens [NUSCHOO] and by Schoutens
and Teugels [SchTeu98] study the gamma, Pascal, and Meixner processes as
main examples of generalized chaotic representation for squareintegrable
271
random variables in terms of the orthogonalized Teugels martingales (which
are the centered power jump processes related to the original process). They
use the onedimensional polynomials of Meixner’s type in order to carry
out the orthogonalization procedure of the Teugels martingales . We refer to [GrigOl], [Grig99], [GrigOOc] for several interesting properties
of these distributions and explicit formulae related to them.
In particular, in [GrigOl], the Meixner process was proposed as a model
for risky assets and an analogue of the Black and Sholes formula was es-
tablished for them.
The infinite dimensional and multidimensional analogues of orthogonal
polynomials associated to a given measure have been widely studied both
in the Gaussian case and in the Poisson case (CHIHA, KOKUOL).
The programme to extend this analysis to more general probability mea-
sures was developed by Berezansky [BEREZb98], [BEREZa97], [BEK094],
[BELILY951 who introduced in this connection the notion of Jacobi field of
operators, and his school [LYTVO2a], [LYTVOSb], [LYTV95c].
An infinite-dimensional analogue of the Laguerre polynomials and the
associated Jacobi fields, corresponding to the gamma case, i.e. to the class
(111) in Meixner’s classification, was studied in [KonLitOO], [KoSiStr97].
In conclusion it should be added that the square of white noise
(RSWN) was introduced as an example of interacting Fock space and in
the attempt to extend to infinite dimensions the canonical connection be-
tween orthogonal polynomials and interacting Fock spaces established, in
the 1-dimensional case, in [AcBo97] and, in the 1-dimensional case, in
[AcKuSt02]. I
6. Higher powers of white noise
The next step of our programme is: to extend, if possible, the results ob-
tained for the square to higher powers of white noise. We are developing
this programme jointly with Andreas Boukas including collaborations with
Uwe Franz, in the attempt to overcome the obstructions posed by the no-go
theorems (cf. below), and with Rene Schott and Massimo Regoli, on Lie al-
gebra and algorithmic aspects: among other things we are trying to extend
the symbolic programme developed by Feinsilver and Schott [FeinScho93]
for calculations on Lie algebras.
Our programme is to characterize those independent increment station-
ary processes (SIIP or Levy processes) which arise as renormalized higher
powers of the standard (Fock) quantum white noise. For a short period, in
272
2003, we believed we had realized this dream but, as I will try to explain
below, the situation is more subtle and the problem is related to some long
standing open problems in the classical theory of SIIP.
Definition The renormalized boson Fock white noise ( simply RBF white
noise in the following ) over a Hilbert space 'FI with vacuum (unit) vector @ is the locally finite Lie *-algebra canonically associated to the associative
unital *-algebra of operator-valued distributions on 'FI with generators
b;", b t , k, EN, t E R d
and relations
[bt , bf] = 6( t - s)
p,t , b 9 = lbt , bsl = 0
(b,)* = bi
b t @ = O
6(t)' = c1 - '6 ( t ) , c > 0 , 1 = 2,3, ....
Lemma The Lie algebra, associated to the RBF white noise (renormalized
boson Fock white noise), is the Lie algebra with generators
b,tkb2" =: bE(t)
(bf"2")t = (bt")b!
central element b:b,t" =: E and relations
where: k = 0,1,2, ...,
k( l ) = k(k - l)(k - 2) . . . (k - I + 1)
k(') = o if I > IC
These conditions guarantee that no negative powers of the white noise
functionals appear.
273
In terms of the smeared generators
with involution
(B,"(fN* = mf) and central elements
the relations become
Now let us deduce some necessary conditions for the existence of the Fock
representation.
Lemma (Boson Independent increments) Suppose that in the scalar
product
the supports of any two test functions either coincide or are disjoint.
in the scalar product. Then the above scalar product is equal to
Denote by Z the family of all supports of all the test functions appearing
where if {A : s u p p ( 4 ~ ) = I } = 8 we interpret
as 1.
274
7. No go theorems
The main result of [AcLuVo99] was the existence of the Fock representation
for the second order white noise. In [AcFrSkOO] it was shown, among other
things, that this representation can be interpreted as a representation of the
current algebra over the Lie algebra sL(2, W). The analogue representation
for the first order white noise, which corresponds to the Heisenberg-Weyl
algebra, had been known in physics for over 70 years.
Now: a current algebra over a Lie algebra is a functional version of
the Lie algebra itself. More precisely it is an algebra of functions on some
measure space ( X , v), (which in the case of [AcLuVo99] was E X d ) with values
in this Lie algebra (such algebras were introduced and widely studied in the
1960's and, in the more recent mathematical literature, they are sometimes
called " Kac-Moody algebras").
Now it might seem, at first glance, natural to conjecture that, if a Lie
algebra has a Fock (lowest weight) representation, then the associated cur-
rent algebra too have one. This is certainly true if the measure space ( X , v) has a finite number of points because in this case the current representation
is a finite tensor product of the original one.
For example the Lie algebra generated by the Heisenberg-Weyl algebra
and sL(2,R), called the Schrodinger algebra, has been widely studied in
the literature and, since the Schrodinger representation exists in any finite
dimension, the associated current algebra over ( X , v ) has a Fock represen-
tation for any space X with a finite number of points.
In the paper [Snia99], devoted to the extension of the results of
[AcLuVo99] to the free case, Sniady proved the following result.
Theorem The joint Fock representation of the first and second order
white noise, i.e. of the Schrodinger algebra, cannot exist.
This theorem was generalized in [AcFrSkOO] and further generalized
by Accardi, Boukas and Franz [AcBouFk05] whose result, reported below,
destroyed the hopes of a naive generalization, the higher powers of white
noise, of the results obtained for the second power.
Theorem 7.1. In the notation (2), denote
where X I is the characteristic function of the interval I & W (taking value 1 on I and 0 elsewhere). Let C be a Lie *-algebra with the following prop- erties:
275
( i) L contains B,", and Bin (i i) the BE satisfy the higher power commutation relations .
Then L does not have a Fock representation i f the interval I is such that
where c denotes the renormalization constant.
This theorem means that we cannot hope to have a single representation
including all the higher powers of white noise: the best one can hope is to
form, for each n, the smallest Lie algebra generated by B," and B: and look
for a representation of it.
The difficulty with this programme is that, as soon as n 2 3 these Lie
algebras are infinite dimensional and not so widely studied. In particular
one cannot apply the general methods of [AcFrSkeOO], which heavily used
the known theory of irreducible unitary representations of s l (2 , R), and one
has to go back to the direct method of [AcLuVoSg] which however, in these
cases is much more complex due to the more complex structure of the higher
order commutation relations.
At the moment we do not know if such a representation exists even in
the case n = 3. The following considerations show that this difficulty is related to and
old open problem of classical probability.
In the case of 2-d order noise and of higher orders with a single renormal-
ization constant, the current algebra restricted to a single block Liespan
{ B , h ( ~ ~ ~ , ~ l ) } is isomorphic to the 1-mode Lie algebra
Lie-span-{a+'ak}
Lemma 7.1. Let (b:) be the Boson Fock scalar white noise. Suppose that the k-th power of white noise exists for some natural zn-
teger k and admit a Pock representation. then the process
{wi,t],a,-OO < s < t < +00}
defined formally by some renormalization of
should be a stationary additive independent increment process on R.
276
Lemma 7.2. The map
with the [AcBouFrO5] renormalization, is a Lie algebra isomorphism.
Corollary. The Fock statistics of (B::),]) is the same as that of (uk).
Proof. This statistics is uniquely determined by the Lie algebra structure.
Lemma 7.3. The vacuum distribution of 1 1 1 ( b t + bt)"dt = 1 wpdt
coincides with that of
(u+ + a)n
Proof. The statistics is uniquely determined by the Lie algebra structure
Corollary. If the Fock representation of the n-th power of white noise
exists, then the vacuum distribution of
(a+ + a)n
must be infinitely divisible.
Proof. From Lemma (7.3) it follows that the distribution of (a+ + a)n is
the same as the distribution of Ji dt(b? + bt)" and from Lemma (7.1) we
know that this is infinitely divisible.
Theorem 7.2. A necessary condition for the existence of the n-th power of white noise, renormalized as in [AcBouFrOS] is that the n-th power of a classical Gaussian random variable is infinitely divisible.
In classical probability the n-th powers of the standard (1-dimensional,
mean 0, variance 1) Gaussian random variable and their distributions have
been widely studied. In particular it is known that, Vk _> 1, y Z k is infinitely
divisible.
However it is not known if, Vk 1 1, yZk+' is infinitely divisible.
This suggests the conjecture that the above mentioned programme
might be realizable if one starts from even powers (which fortunately are
closed under Lie brackets).
277
Appendix: Meixner’s classification theorem
8. Orthogonal generating functions
The purpose of this appendix is to give an exposition of the problem studied
by Meixner and of its method of solution. To this goal we begin with some
general definitions.
Definition 8.1. Let p be a probability measure on R with moments of any
order and let Pn(x) (x E R; n E N), denote the orthogonal polynomials of
p normalized so that
&(x) = 1 ; leading term of P,(z) = 1. (4)
A function
is called an orthogonal generating function if
where the series in (6) converges weakly in L2(R, p).
Problem. Under which conditions is a function F : R x R the orthogonal
generating function of some probability measure on R? It is easy to verify that a necessary condition is that, denoting (., .) the
scalar product in L2(R, p), one has
(1 , JY . , t ) ) = (Po,F(. , t ) ) = 1 *
In other words
Theorem 8.1. Now suppose that condition (7) is satisfied with a function F ( x , t ) of the special form
~ ( z , t ) = e z u ( t ) f ( t ) (8)
where u : IR -+ R is an invertible function such that
u(0) = 0 (9)
278
and f is a function such that
u’(0) = 1 . (11)
If one assumes that u is invertible, then the Laplace transform of p is uniquely determined in i ts domain, by the formula
P I
Proof. Then (7) becomes
f ( t ) p ( d z ) = 1 .
Introducing the change of variable
U ( t ) =: 7 ; t = U-’(‘T)
the identity (15) or equivalently
becomes (12).
Remark. The meaning of Theorem (8.1) is that a probability measure p, satisfying (7) and (8), is uniquely determined by the pair ( f lu ) provided
that u is invertible.
In his paper [Meix34] Meixner:
(i) determines all pairs of functions (f, u ) satisfying conditions (7) and
(ii) shows that for each such pair (f, u), u is invertible
(iii) explicitly determines all the corresponding probability measure.
This justifies the following
(8) for some probability measure p
Definition 8.2. A probability measure p on R is called a Meixner mea- sure if
(i) p admits an orthogonal generating function F ( z , t ) (ii) F ( z , t ) has the form (8) for some pair of functions ( f , u ) called the
associated pair.
279
Finally let us prove that the ansatz (8) is coherent, i.e. that the series
expansion of its right hand side has the form (6) with the P,(x) satisfying
condition (4). This follows from the following:
Lemma 8.1. Let f ( t ) be a formal power series with constant term
f (0) = 1
and let u(t) be a formal power series with constant term u ( 0 ) = 0 and with linear term coefficient u1 = 1, i.e.
u(t) = t ( l + [t] + . . . )
Pu(x) = xn + an,lzn-' + . . . + an,,
Then there exist polynomials P,(x), with leading coefficient equal to 1,
such that the following formal expansion holds
Proof. By assumption
u(t) = c U,t" = t + i i 2 ( t ) ; uo = 0 ; u1 = 1 . "20
We define the degree of a formal power series Cn,Oanzn, the smallest
n E N such that a , # 0. For example, f has degree 5 , u has degree 1 and
i i 2 degree 2 2. Moreover
Our assumption on u implies that
u(t)" = [t" + i i 2 " ( t ) ]
deg i i z , " ( t ) 2 n + 1 (u2")
and
f (9 = [I + with deg f ( t ) 2 1. Therefore
u(t)" f ( t ) = [t+iiz(t)]"[l+f(t)] = [tn+ii2,,(t)l[1+f(t)] = tn+G2,n+t"f+62,,f
280
with
deg tnf 2 n + 1 ; deg i i 2 ,n f _> 2(n + 1)
therefore
u( t ) " f ( t ) = tn + Bn(t)
with deg B,(t) 2 n + 1. Therefore
with deg Cn+l(t) 2 n + 1. Therefore
which proves that the leading coefficient of Pn(x) is equal to 1.
9. The equations for f and for 21 = 21-l
Denote v the inverse formal power series of u, i.e. by definition
u(v(t)) = v(u(t)) = t
and denote
d dx
D := -
Then the following identity is clearly satisfied:
v( D)e""(t)f( t ) = v(u(t))e2"(t) f ( t ) = te""@)f(t)
v (D)a;ezu(t) f ( t ) = a: (tezu(t) f ( t ) = ta,"eTu(t) f ( t ) + na,"-l ezU@) f ( t )
(17)
Taking a; of both sides of (17) one finds
evaluating this a t t = 0 and keeping (17) into account, one finds
v(D)Pn(x) = nPn-l(x). (18)
On the other hand the Pn(x) are the orthogonal polynomials of some mea-
sure II, on R if and only if there exist two sequences ( I n ) and (kn) of real
numbers such that
k n I O ; b'n
and
281
with the convection that
P-1(2) = 0
[V(D), .I = W )
Denoting x the multiplication by x and using the identity
we find, combining (18) and (19):
v(D)Pn+1 (z) = (n+l)Pn(~) = (z+ln+l)nPn-l(z)+v’(D)Pn(2)+Icn+l (n-l)Pn-z(z)
(20)
(21)
while the usual Jacobi relation (19) is
nPn(x ) = (x + ln)nPn-l(x) + nknpn-z . Subtracting (21) from (20) one finds
or equivalently
Applying v (D) to both sides and dividing by n + 1 one obtains
Now, P,(x) cannot be identically zero because its leading coefficient is
equal to 1. Therefore comparing (22) and (23) we conclude that there exist
constants A, K such that
ln+l - 1, = x * ln+l = nx + 1 (24)
(25) kn+1 kn - - - = K * kn+l = n((n - 1). + kz ) . n n - 1
Notice that, since the k, are all negative, also K must be negative.
Moreover, given (24) and (25), (19) becomes:
Pn+l(z) = (x + 11 + nX)Pn(x) + n ( h + (n - l ) K ) P n - l ( X ) .
282
We know that k2 5 0 and it cannot be = 0, otherwise $ is a multiple of
a &measure, hence in (16) Pn(x) = 0 , V n > 1 and, since PO(,) = 1 by
assumption, equation (16) becomes
f ( t ) = e-ZU@)
which can be satisfied for any x , t E R if and only if
u(t) E 0 , f ( t ) = 1
which corresponds to a trivial solution. Thus for all non trivial solutions
one must have
Since
it follows that
tn tn-1 C nXPn(0)- = At c ~ ~ ( 0 ) - = Xtf’(t) n! (n - l)!
n20 n>l
tn (n - I)! (n - I)! n!
tn tn-1 C k2pn-1(0)- = kzt C Pn-l(0)- = k2t c Pn(0)- n>l n2l n > O
= Kt2f’(t)
From these identities one deduces that
f ’ ( t ) = Zif(t) + Mf’(t) + kzt f ( t ) + Kt2f’(t)
283
or equivalently f satisfies the equation
Moreover from (18) and (22) we find
(1 - v’(D))P,(z) = Av(D)P,(s) + KW(D)~P,(~) ; V n (27)
Therefore, as operators on Liol(R, p )
1 - v’(D) = Av(D) + K V ( D ) 2
or equivalently
v’ = 1 - Av - nu2 . Therefore the pair (f, v ) (equivalently (f, u) ) is uniquely determined by the
solutions of the equations (26), (28) respectively. Notice that ‘the same
polynomial
1 - At - K t 2 (29)
appears in both equations. According to the various possible values of the
parameters A, K , we distinguish 5 possibilities:
(I) A = K = 0 ((29) has degree 0)
(11) K = 0; A # 0 ((29) has degree 1)
(111) A2 = - 4 ~ # 0 (29) has degree 2 and one non zero root of multiplicity
(IV) A2 > - 4 ~ > 0 (29) has degree 2 and 2 distinct non zero real roots)
(V) 0 < A2 < -46 (29) has degree 2 and 2 non zero complex conjugate
2)
roots)
The five Meixner classes are defined by t the solutions of equations (as), (26) corresponding to the values of the parameters ( A , K ) in the classes
defined above.
Remark. In fact Meixner ([Meix34], Section 6 ) calls class (11) what we
have called class (111) and conversely.
Moreover Meixner does not classify his five classes in terms of the pa-
rameters (A, K ) but in terms of two auxiliary parameters (a, p), related to
(A, K ) by the equations
a + P = A
ap = - K .
284
In the following section we will describe the translation code between our
parametrization and Meixner’s.
10. Meixner’s parametrization
Theorem 10.1.
( i) For any real numbers X,K there exist complex numbers a ,P such that the following identity holds
(30)
a + p = x (31)
1 - At - K t 2 = (1 - at)(l - p t )
or equivalently
ap = -K (32)
(ii) The pair (A,&) uniquely determines the pair (a lp ) up to the per- mutation
(a, P) + (P, a)
(iii) If in addition
K I O (33)
X = K = O * a = P = O (34)
then there are only four possibilities
(35) K = 0 ; x # 0 H (a, p) = (A, 0)
(37) where, in all the above identities (a, P ) has been identified to (p, a) and the square roots are the positive ones.
Moreover the last possibility (37) splits into three according to the following situations:
285
i.e. only one real solution
i. e. two distinct real solutions
i.e. two complex conjugate solutions.
of the pair (A, 6):
(iv) The five Meixner classes are characterized by the following values
(I) x = r; = 0 (11) A, n # 0, x = f 2 1 4 1 / 2
(111) n = 0; x # 0 (IV) A, r; # 0; A2 > 41nI (V) A, n # 0; A2 < 4)nI.
Let us first discuss the equation
0 = 1 - A t - r;t2
If
X = r C = O
there are no solutions.
If
n = o ; A # O
there is only one solution
1
x t,, = - *
If
n # O ; x = o
0 = 1 - nt = 1 + Inlt2
then (41) becomes
2
which has only 2 purely imaginary complex conjugate solutions:
(43)
(44)
t & f = f i - * /1:1
286
If both
then equation (41) can be written
x 1 1
IKI O = ( K . l t 2 - X t + l ~ O = t 2 - - t + - = t - - --
14 ( 2;1)2 4K12 +m which has exactly one solution if and only if (38) holds.
In this case the solution is
x t l , = -
214
and, due to the relation (38) there are 2 possibilities
giving rise to the solutions
1 t1%* = */K11/2
The two remaining possibilities, beyond (38) are (39) and (40). Condition (39) corresponds to two distinct real solutions
(46)
(47)
Condition (40) corresponds to two complex conjugate solutions
Now let us consider the identity (30) which is equivalent to
1 - At - Kt2 = 1 - (a + p>t + apt2
It is clear that the pair (a, p) is a solution if and only if the pair (p, a) is.
Equating coefficients we find (31), (32). Fkom these we deduce
-6 = (A - p)p = xp - p2
i.e.
287
This gives the solutions
which satisfy the condition
(P+,"+) = ("-1P-1.
Let us discuss the possible solutions of the system (32), (31) corresponding
to the various possibilities for the parameters X and K .
(34) is obvious.
Clearly (42) holds if and only if
a = p = o .
Now suppose that (44) holds then
P + = X # O ; p - = o
a + = o ; a - = X # O
that is, exactly one number, in the pair (alp) is # 0.
Conversely, if this is the case, then X must be # 0, otherwise
a* = -P* and it is impossible that exactly one is # 0. Moreover this condition can
be fulfilled only if X2/4 + IC is real.
In this case one has always
p + > o ; " - > O .
Thus the condition that exactly one in the pair (a, P ) is different from zero
can be fulfilled only if either
- + K = o 2
or
Thus the two conditions coincide and are both equivalent to
K = O .
288
This proves (35).
becomes
Now suppose that condition (44) holds. Then the system (32), (31)
a=-@ (51)
(52) 2
K = p
Since K # 0, this means that @ must be purely imaginary and # 0. Con-
versely, if this is the case and (51) holds, then (45) holds. This proves
If condition (45) holds, then the system (32), (31) has 2 distinct solutions (36).
satisfying
(a+,P+),(a-,P-) = (P+,a+) (53)
Conversely, if this is the case, then (45) must hold because, if either X or K
are zero, then (53) cannot define two distinct solutions.
Finally note that the above discussion is valid in both cases when the
solutions of (41) are real or complex, i.e. if either condition (39) or (40)
hold.
This proves (37).
This completes the proof of (iii).
The 1-st Meixner class is clearly characterized by the condition
X = n = O
The condition characterizing the 2-nd Meixner class is equivalent to the
case (37) under condition (38), i.e. when equation (41) has a unique non
zero real solution.
The 3-rd Meixner class is equivalent to the case (35).
The 4-th Meixner class is equivalent to the case (37) under the condition
The 5-th Meixner class is equivalent to the case (37) under the condition
(39), corresponding to two distinct real nonzero solutions.
(40), corresponding to two complex conjugate solutions.
11. Solutions of the equation for 21
In the present section we discuss the solutions of equation (28) correspond-
ing to the various Meixner classes.
Class I: A = n = 0. In this case equation (28) becomes
v' = 1 (54)
289
By assumption
u(t) = t + tZuz(t)
where uz(t) is an arbitrary formal power series. Moreover
0 = v(0) = 210 (55)
and the unique solution of (54) with initial condition (55) is
v(7) = 7- . (56)
Class 11: A, tc # 0, X = f 2 1 ~ 1 ' / ~ . In this case equation (28) becomes
w'(7) = 1 2)K11/2v(7-) + v(T)21K.I = (1 lK1'/221(7-))2.
This gives
which is of the form
with
b = O ; a = l ; c = l t p *
Therefore the solution is
This gives
or
1 - = v(t) 1
(t + C ) l K [ l K p 2
and condition (55) is satisfied if and only if
290
Thus
This gives
Thus
1 t + c = -- ln(1- Xu) x
2, = - 1 (1 - e-x'e-xc) ,
X The condition
v(0) = 0
fixes c = 0, so that
1
x = - (1 - e-XT) ,
Class IV: A, K # 0; X2 > 41~1 In this case equation (28) becomes
= I.l(v - t+)(v - t - )
or equivalently
(t+ +t- )-I J (v - t+)(v - t - ) JKJ t + c =
with
14 (t+ + t J1 = - x - This gives
291
In this case equation (28) becomes
v - t+ e A t e w l n l = ___
v - t-
and the condition v(0) = 0 fixes
eAc/lnl = 5 t-
A t A t At e t+(v - t - ) = t-(v - t+) ($ (e t+ - t-)v = e t+t- - t+t-
eAt - 1 1 eAt - 1
e%+ - t- l l ~ l e + - t- v = t+t- = - Att
In conclusion
eAT - 1
eATr+ - r- v(r) =
where
-r* := 2 2 (1 f /-). Class V: X , K # 0; X2 < 4161 The result is the same as in the case of Class IV, i.e. (59).
12. Solutions of the equation for f
Class I
X = r ; = O .
In this case
and the condition
f (0) = 1
is satisfied.
Class I1 Let us consider the case
R f O
(59)
292
A2 = 4(n( . In this case we have
A # O ; & = o
then
Therefore in case (64)
x
!?+-A) f ( t ) = e t y (; - t )
Classes IV and V
These classes are characterized by the condition
x2 # 4 1 ~ 1 f 0 (65)
(real roots, class IV; complex conjugate roots, class V). In this case the
characteristic polynomial 1 - At - fit2 has 2 different roots t f such that
Therefore the solution of equation (26) is
The solution of (67) satisfying
0 = In f(1) = In 1
293
is given by
It is convenient to write
With these notations
A = A1 + ~ z t + ; B = -(A1 + ~ z t - ) (70)
and (68) becomes equivalent to
(1 - t/t+) (1 - t/t+)t+ nz
f(t) = [(I - [ (1 - t/t+ I . Remark. In the paper p] (pg. 10) Meixner assumes that
11 = 0
therefore, due to (69), in order to recover his expression for f ( t ) , one has
to put A1 = 0 in (71).
13. The equations for u
Now notice that, if
U ( t ) = 7 U-’(T) = t
then
1 t = u-’(u(t)) = v(u(t)) =+ 1 = v’(u(t))u’(t) H v‘(u(t)) = -
u’(t> .
Therefore if ~ ’ ( 7 ) = F(v(7) ) then
- = v’(u(t)) = F(v(u( t ) ) ) = F ( t ) u’(t)
1
and in our case this becomes
1 - = 1 - Aw(u(t)) - 6 2 ( U ( t ) ) = 1 - A t - 6t2 u’(t)
294
or
1
1 - x - Kt2 u'(t) =
Therefore in the case (61)
u(t) = d t = t J because u(0) = 0.
In the case (64)
u(t) = - . X # O . Ji"",t 1
Therefore
1 u(t) = In
(1 - At) ' /X - In case (63), i.e.
In case (65) the polynomial (29) has two roots given by (66). Therefore
1/14 1 t - t , t - t+ IKI t - t-
-- - In-=ln(-) t - t- .
I Q = P = A = K = O
Q$(x) = ex2 /2kdx .
Hermite's polynomials.
I I a = P # O
295
confluent hypergeometric polynomials
I11 ff # 0, p = K = 0
Charlier's polynomials
IV Q! # p, rc # 0, a l p real
ex . Iff1 > IPI
v ff # p, K # 0 E = p
14. Moments of the Meixner measures
In this section we derive a simple formula which expresses the moments of
a Meixner measure in terms of the associated Meixner pair.
Taking &derivatives of both sides of (15) one obtains
or equivalently, writing
u(")(t) := dtu(n- l ) ( t ) ; u(O)(t) := u(t) (72)
In particular, putting t = 0 in (73) and using (9) one finds the first moment
of p, i.e.
296
Taking derivatives of both sides of (73) one finds
or equivalently
Now, considering u(l ) as a multiplication operator in L2(Iw, p ) and at as an
operator in the same space, one can introduce the notation
(76) 1
U(1) A, := -Lit
So that, if cp is another multiplication operator in L2(Iw, p ) :
(77) 1
nuP(t) = - (8tcpHt) u(1) ( t )
z2e""( t )p(dx) = A:- ( t )
In these notations (75) becomes
(78) 1 s, f
which gives the second moment of p by evaluating (78) at t = 0
mz(p) = L x 2 p ( d x ) = A:- 1 (0) .
f Now suppose by induction that,
znex" ( t )p (dz ) = A:- 1 ( t )
f then, taking derivatives of both sides, one finds
1 xn+lu ' ( t )e"" ( t )p(dx) = &A:- ( t )
f or equivalently
1 1 1 8th:- ( t ) = A:+'- ( t )
f p(dz) = -
U ( l ) ( t ) f
(79)
and therefore (80) holds for each n E N. In particular, taking t = 0 in (80)
one finds the n-th moment of u:
(81) m,(p) = L x n p ( d x ) =A,"- 1 (0) ; V n E N .
f In other words:
297
Theorem 14.1. Suppose that a probability measure p on 0% has an orthog- onal generating function of the form (8) for a pair of Cw-functions (u, f ) from Iw to R. Then p is polynomially determined by the pair (u, f ) through formula (81).
15. Bibliography
[AcLuVo02] Accardi L., Lu Y.G., Volovich I.: Quantum Theory and its
Stochastic Limit, Springer Verlag (2002)
[AcLuVo99] Accardi L., Lu Y.G., Volovich I.V.: White noise approach
to classical and quantum stochastic calculi, Lecture Notes of the Volterra
International School of the same title, Trento, Italy, 1999, Volterra Preprint
N. 375 July (1999)
[AcFrSkOO] Accardi L., Franz U., Skeide M.: Renormalized squares of
white noise and other non- Gaussian noises as Levy processes on real Lie
algebras, Comm. Math. Phys. 228 (2002) 123-150 Preprint Volterra, N.
423 (2000)
Accardi L.: The white noise approach to stochastic calculus: motiva-
tions, applications, developments, in: Colloque SMT 2000, p. 1-18 Plenary
Lecture given at: Colloque SociCtC MathCmatique de Tunisie, Sousse, 27-30
mars 2000
[AcBo97] Luigi Accardi, Marek Bozejko: Interacting Fock spaces and
Gaussianization of probability measures Infinite dimensional analysis, quan-
tum probability and related topics, 2, N. 4 (1998) Volterra Preprint (1997)
N. 321
[AcKuSt02] Accardi L., Kuo H.-H., Stan A.: Orthogonal polynomials
and interacting Fock spaces, Infinite Dimensional Anal. Quantum Probab.
Related Topics (IDA-QP) 7 (4) (2004) 485-505
[AcBouFr05] Luigi Accardi, Andreas Boukas, Uwe Franz: Non existence
of the Fock representation for the renormalized higher powers of white noise,
to be submitted to IDA-QP (2005)
[AcAyOuOS] Luigi Accardi, Wided Ayed, Habib Ouerdiane: White
Noise Approach to Quantum Stochastic Calculus, talk given at the:
COllOQUE INTERNATIONAL DE MATHEMATIQUES, ANALYSE ET PROBABILITES, 20-25 Octobre 2003, Hammamet, (Tunisie) to appear
[Ayed05] Wided Ayed: PhD thesis (to appear)
[BEREZb98] Yu.M. Berezansky: Commutative Jacobi fields in Fock
[BEREZa97] Yu.M. Berezansky: On the direct and inverse spectral space, J . Func. Anal. 138 (1998) 163-190.
298
problem for Jacobi fields, Algebra i Analys (in russian) 9 (6) (1997), english
transl. St. PetersburgMath. J. 9 (6) (1997).
[BEK094] Yu.M. Berezansky, Yu,G. Kondratiev: Spectral Methods in
infinite dimensional analysis, Kluwer (1994).
[BELILY951 Yu.M. Berezansky, V.O. Livinsy,, E.W. Lytvynov: A gen-
eralization of Gaussian white noise analysis, Methods Func. Anal. and
[BruRoSl] F.T. Bruss, L.C. Rogers: Pascal processes and their charac-
[FeinSchogS] P. Feinsilver, R. Schott: Algebraic Structures and Oper-
I: Representations and Probability Theory, Kluwer
[GrigOl] B. Grigelionis: Generalized z-distributions and related stochas-
[Grig99] B. Grigelionis: Processes of Meixner type, Lithuanian Math.
[GrigOOc] B. Grigelionis: On generalized z-diffusions, Preprint Vilnius
[HidaOl] Hida T.: Selected papers, World Scientific (2001)
[HiKuPoStr93] Hida T., Kuo H.-H., Potthoff J., Streit L.: White Noise.
An Infinite Dimensional Calculus, Kluwer Academic Publishers (1993) 185-
231
[Hi751 Hida T.: Analysis of Brownian Functionals, Carleton Mathemat-
ical Lecture notes 13 (1975); 2nd ed., 1978
[HuPa84b] Hudson R.L., Parthasarathy K.R.: Construction of quan-
tum diffusions, in: Quantum Probability and applications to the quantum
theory of irreversible processes, L. Accardi, A. Frigerio, V. Gorini (eds.)
Springer LNM N 1055 (1984)
[KonLitOO] Y. Kondratiev, E. Lytvynov Operators of gamma white noise
calculus Infin. Dimen. Anal. Quant. Prob. Rel. Top. 3 303-335 2000
[KoSiStr97] Y. Kondratiev, J.L. Silva, L. Streit Generalized Appell Sys-
tems, Methods Funct. Anal. Topology 3 (3) (1997) 28-61
[LYTV02a] Eugene Lytvynov: Polynomials of Meixner’s type
in infinite dimensions Jacobi fields and orthogonality measures,
arXiv:math.CA/0203026 v5, 27 May (2002).
[LYTV02b] Eugene Lytvynov: Orthogonal decompositions for Levy pro-
cesses with an application to the gamma, Pascal and Meixner processes,
arXiv:mathPR/0204087 v5, 27 May (2002).
[LYTV95c] E. Lytvynov: Multiple Wiener integrals and non-Gaussian
Topol. 1 (1995) 28-55.
terization, Stochastic Process. Appl. 37 (1991) 331-338
ator Calculus, Vol.
Academic Publishers, Dordrecht (1993)
tic processes, Lithuanian Math. J . 41 (2001) 303-319
J . 39 (1999) 33-41.
(2000).
299
white noises: a Jacobi field approach, Methods Func. Anal. and Topol. 1
[Meix34] Meixner J . : Orthogonale Polynomsysteme mit einen beson-
deren Gestalt der erzeugenden Funktion, J.Lond. Math. SOC. 9 (1934)
[NUSCHOO] D. Nualart, W. Schoutens: Chaotic and predictablerepre-
sentations for LQvy processes, Stochastic Process. Appl. 90 (2000) 109-122
[Parth92] Parthasarathy K.R.: A quantum stochastic approach to Ito’s
formula for LQvy processes, C.R. Acad. Sci. Paris 315 SQries I (1992)
[SchouOO] W. Schoutens: Stochastic Processes and Orthogonal Poly-
nomials, Lecture Notes in Statist., Vol. 146, Springer-Verlag, New York
[SchTeu98] W. Schoutens, J.L. Teugels: LQvy processes, polynomials
and martingales, Commun. Statististics-Stochastic Models 14 (1998) 335-
349
[Schii93] Schiirmann, M.: White noise on bialgebras. Springer Lect
Notes Math: 1544 (1993)
[Sega7Oa] Segal I.E.: Transformations in Wiener space and squares of
quantum fields, Adv. in Math. 4 (1970) 91-108
[Snia99] P. Sniady: Quadratic bosonic and free white noises, Commun.
Math. Phys. 211 (3) (2000) 615-628 Preprint (1999)
[TsiVeYoOl] N. Tsilevich, A. Vershik, M. Yor: An infinite-dimensional
analogue of the Lebesgue measure and distinguished properties of the
gamma process, J. Funct. Anal. 185 (2001) 274-296
(1995) 61-85.
6-13
1417-1420
(2000).
300
This page intentionally left blankThis page intentionally left blank
Recommended