Sei sulla pagina 1di 28

Philosophy of Science Association

On the Brussels School's Arrow of Time in Quantum Theory


Author(s): Vassilios Karakostas
Source: Philosophy of Science, Vol. 63, No. 3 (Sep., 1996), pp. 374-400
Published by: The University of Chicago Press on behalf of the Philosophy of Science
Association
Stable URL: http://www.jstor.org/stable/188101
Accessed: 28/12/2008 17:45

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=ucpress.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.

The University of Chicago Press and Philosophy of Science Association are collaborating with JSTOR to
digitize, preserve and extend access to Philosophy of Science.

http://www.jstor.org
ON THE BRUSSELS SCHOOL'S ARROW OF TIME
IN QUANTUM THEORY*

VASSILIOS KARAKOSTAStt
Department of History and Philosophy of Science
Universityof Cambridge

This paper examines the problem of founding irreversibilityon reversible equa-


tions of motion from the point of view of the Brussels school's recent develop-
ments in the foundations of quantum statistical mechanics. A detailed critique of
both their 'subdynamics' and 'transformation' theory is given. It is argued that
the subdynamics approach involves a generalized form of 'coarse-graining' de-
scription, whereas, transformation theory cannot lead to truly irreversible pro-
cesses pointing to a preferred direction of time. It is concluded that the Brussels
school's conception of microscopic temporal irreversibility, as such, is tacitly as-
sumed at the macroscopic level. Finally a logical argument is provided which
shows, independently of the mathematical formalism of the theory concerned, that
statistical reasoning alone is not sufficient to explain the arrow of time.

1. The Arrow of Time in Physics. One of the long standing foundational


problems in theoretical physics is that of the temporal asymmetry, or 'ar-
row of time' in Eddington's phrase. The problem essentially can be seen
as arising from the contradictory behavior between the time asymmetric
evolution of macroscopic irreversible processes and the time symmetric
evolution of the reversible processes governing the underlying dynamics
of the atomic constituents. The so evident paradox is a direct consequence
of the symmetric nature of the basic laws of physics with regard to time
inversion.1
All of our fundamental theories, be it classical mechanics, as expressed
by Hamilton's equations of motion
AH AH
q=-, -- (1)

*Received March 1993; revised July 1995.


tI wish to thank Michael Redhead and Jeremy Butterfield for helpful comments on an
earlier draft of this manuscript, and an anonymous referee for useful remarks. I also thank
the Arnold Gerstenberg Fund and the British Academy for generous financial support while
this work was being done.
tSend reprint requests to the author, Department of History and Philosophy of Science,
University of Cambridge, Free School Lane, Cambridge CB2 3RH, England.
'More precisely, our present physical laws are invariant under simultaneous reversal of
charge, parity, and time. The CPT symmetry has been introduced into elementary particle
physics since 1968 in order to account for the decay of the K-meson in processes governed
by the weak interaction force, which however is not directly relevant in the context of this
paper.

Philosophyof Science,63 (September1996)pp. 374-400.0031-8248/96/6303-0003$2.00


Copyright1996by the Philosophyof ScienceAssociation.All rightsreserved.

374
THE BRUSSELS SCHOOL'S ARROW OF TIME 375

or quantum mechanics, as described by Schr6dinger's equation


aTP
i- =H T, (h = 1) (2)
at
remain invariant with respect to time reversal transformation t -> -t.
This can be seen in classical dynamics by reversing both the directions of
time and momenta, whereas the reversal of the direction of motion in the
quantum case requires the replacement of T by its complex conjugate T*.
(A lucid account of the time reversal invariance of the wave function is
given by Davies (1977, 154-157)).
As is well known (e.g. Balescu 1975), both of the dynamical equations
above can be unified through the Liouville-von Neumann equation for the
density operator
aw
i= LW (3)
at
with

ti{HiH
,
i
L }}-i ,
p aq q ap) (4)
[H, ]-H -
I-IOH
the braces designating the classical Poisson bracket and the square brack-
ets the quantal commutator. The von Neumann equation can be obtained
either by quantizing the classical Liouville equation or by starting from a
Schr6dinger equation and applying statistics to a fictitious ensemble of
wave functions. It may, therefore, be characterizedas the general equation
of motion in dynamics. One of its basic features is the so-called 'L-t in-
variance'. This simply means that under the successive operation of the
two transformations
L
IL- (5)
t -t -t
Eq. (3) remains invariant.
Direct application of the L-t symmetry, as expressed by the unitary
character of the dynamical evolution of Eq. (3), shows that the usual ex-
pressions for nonequilibrium entropy in classical or quantum mechanics
S-Kf Win Wdy (classical)
-K Tr Wln W (quantal)
are constants of motion. Hence, one may legitimately say that both clas-
sical and quantum dynamics describe a timeless picture of reality in which
there is no true evolution of matter and consequently no intrinsic distinc-
tion between past and future.
376 VASSILIOS KARAKOSTAS

This description, however, not only comes in flagrant contradiction with


our impression of the flow of time but, most importantly, is in disagree-
ment with a plethora of irreversiblephysical phenomena which point to a
privileged direction of time, the direction in which entropy monotonically
increases towards a maximum, dSldt2 0.
Nonetheless, the apparent incompatibility between the reversible laws
of the dynamical description on the one hand and the law of entropy
increase of the thermodynamic description on the other cannot be resolved
within the usual framework of dynamics, because the entropy concept as
captured by the second law is purely macroscopic; it has no microscopic
counterpart. Of course, the advent of statistical mechanics aimed at pro-
viding the thermodynamic description with a foundation at the micro-
scopic level. Within that framework, Boltzmann's H-theorem has been
introduced as a microscopic basis for the phenomenological second law.
However, the relation with the arrow of time is not generally reproduced:
under time-reversal, entropy increases (or the H-function decreases) in
both directions of time (Tolman 1979, Jancel 1963).2 There is nothing in
Boltzmann's H-theorem or any of its variants in both kinetic and statistical
theory that provides a physical explanation of the problem of the irre-
versibility of time. Thus, statistical mechanics cannot account for the or-
igin of irreversibility which is not to be found in the mathematics of the
theory. For these reasons the traditional point of view has taken the ap-
pearance of irreversibility in physical processes and the related arrow of
time as corresponding to some sort of supplementary approximation, such
as coarse-graining, superimposed upon the reversible laws of dynamics,
or even in considering it to be an illusion due to certain limitations in
human observation, such as a lack of knowledge about the exact dynam-
ical state of the system. Thus, for instance, Born (1949, 72) asserts that
"irreversibilityis a consequence of the explicit introduction of ignorance
into the fundamental laws", whereas, Einstein emphasises that "irrevers-
ibility is an illusion, a subjective impression, coming from exceptional in-
itial conditions" (quoted in Prigogine 1980, 203).
It is paradoxical, however, that such a fundamental aspect of our aware-
ness of the direction of time should not be recognised as an aspect of

2As is well known, the thermodynamic arrow of time is not a direct consequence of Boltz-
mann's H-theorem. To derive the second law of thermodynamics one needs an extra con-
dition, usually an initially low entropy which might be of cosmological origin. The H-the-
orem itself does not single out a direction in time; it only states that given an isolated system
whose entropy is not maximal at a certain time t,, the probability that the entropy at any
time t # to is greater than the entropy at to is enormous. One can therefore infer with over-
whelming probability that the entropy of the system is greater for t > tothan for to. But with
the same overwhelming probability one can show that the entropy was greater for t < to.
This contradicts the second law, which requires a smaller or at least equal entropy value for
the past. Hence, the second law can be regarded to be consistent with the H-theorem only
in the case of future entropy values of present known systems, never past entropy values.
THE BRUSSELS SCHOOL'S ARROW OF TIME 377

physical reality. Moreover, it can hardly be dismissed as being merely


subjective in origin. The arrow of time has objective counterparts in all
parts of modern physics. From radiation damping and the exponential
decay of unstable particles to the emergence of dissipative structures in
far from equilibrium conditions and the collapse of the wave function.
Notice that gravitation too has several aspects of irreversibility, particu-
larly in connection with the properties of black holes and 'big bang' cos-
mology (Zeh 1989, Nicolis et al. 1977, Penrose 1987).
In this light, Prigogine and his school at Brussels have attempted to
endow irreversibility with an objective dynamical meaning and to formu-
late the law of entropy increase as a fundamental dynamical principle
through a unified scheme of dynamics and thermodynamics. To this end,
they have been developing over the past twenty years or so a 'microscopic
theory of irreversibleprocesses', from which the physics of being (dynam-
ics) and the physics of becoming (thermodynamics) emerge as comple-
mentary modes of description. Within this new theory, still in a state of
further development, the reconciliation of the time asymmetric processes
of thermodynamics with the time reversal invariant laws of dynamics takes
the form of the existence of an appropriate 'mechanism' for breaking the
time-reversal symmetry of the underlying dynamics.
The construction of this symmetry-breakingmechanism into the frame-
work of mechanics involves two successive steps: firstly, the introduction
of the concept of subdynamicsin terms of a dynamics of correlations which
will be considered immediately below, and secondly, that of a non-unitary
transformationtheory (a so-called star-unitary transformation) which will
be examined afterwards. The information obtained therein will be directed
in the final part of the paper towards questioning the actuality of the
processes under consideration and especially the possibility of obtaining
within the Brussels school's theory truly irreversible processes, that is to
say, backwards deterministic and forwards completely probabilistic pro-
cesses. Particularly, in connection with the problem of the arrow of time,
I shall argue that no fundamental solution to the problem can be given
within the context of their theory unless they are able to introduce genuine
irreversibility,and thus also a privileged direction of time, entirely in terms
of the concepts of dynamics without any appeal to ill-defined concepts
such as observability and the observer's role as selector of extra-theoretical
conditions or without invoking some ad hoc assumptions such as coarse-
graining or contraction of description.

2. Subdynamics: Dynamics of Correlations. The subdynamics approach,


first introduced by Prigogine, George, Henin, and Rosenfeld (1973, and
more recently refined by Prigogine and Petrosky (1988), concerns the as-
ymptotic evolution towards equilibrium of large quantum systems exhib-
378 VASSILIOS KARAKOSTAS

iting the 'Poincare catastrophe' type of dynamical instability. Briefly, this


type of instability occurs in non-integrable systems for which the invari-
ants of motion, except for the Hamiltonian itself, cannot be written as
analytic functions of the action variables (e.g. Prigogine 1980). Non-
integrable systems, which have been associated with the famous three-
body problem, are of significant interest in physics. A description of ir-
reversible processes for such systems presents a special difficulty which
stems from the fact that the Poincare catastrophe, unlike the dynamical
instabilities of 'mixing' or 'Kolmogorov' flows in ergodic systems, is rep-
resentation dependent. It involves an explicit formulation of the system's
Hamiltonian H in a given set of coordinates. The most convenient starting
point is to decompose H into an unperturbed part Ho corresponding to
system's free motion and a perturbation term V associated with the mo-
lecular collisions
H = Ho + V, (7)
where X is a coupling constant characterizing the strength of the inter-
action.
In conformity with the introductory considerations the evolution of the
system can be described in the Liouville-space formalism through the von
Neumann equation3

ia- LW (8)
at
with L = Lo + kLv. The formal solution of Eq. (8) is,
W, = Ut Wo (9)
whose unitary operator Ut = e-it expresses in effect the time reversal
symmetry always present in the dynamical description.
To construct a theory of dissipative processes an analysis of the contri-
butions to Eq. (9) is required (Balescu 1975). This is usually done by adopt-
ing a Laplace-transform representation and writing Wtin terms of a con-
tour integral involving the resolvent of L, R(z) = (z - L)-1,

Wt = e-it W, = (2ni)-1 fe-iztR(z) Wodz (10)

3The formal analogy of the Liouville-von Neumann equation of motion in quantum and
classical dynamics, discussed in Section 1, enables Prigogine et al. to treat, from a mathe-
matical point of view, the evolution of both quantum and classical dissipative systems to-
wards equilibrium identically. Note however that conceptually the problem of irreversibility
is much more involved in quantum than in classical mechanics, because irreversibilityin the
quantum case can only appear when the Liouvilian has an absolutely continuous spectrum
so one must consider large quantum systems in the asymptotic thermodynamic limit, whereas
in the classical case, as Sinai has shown in 1962, a system of a finite number of hard spheres
(actually, more than two) enclosed within a finite box satisfies the K-flow condition. Details
revealing this disparity in irreversiblebehavior with regard to classical and quantum systems
can be found in Coveney and Highfield 1990, 272-288.
THE BRUSSELS SCHOOL'S ARROW OF TIME 379

where c is a contour taken parallel and above the real axis and z a complex
number not belonging to the spectrum of L, the latter being solely confined
to the real axis. Notice also that R(z) is an analytic function of z and its
singular points therefore compose exactly the spectrum of L. When L has
a discrete spectrum, the only possible singularities of the resolvent are real
poles. This is simply another way to express the quasi-periodic nature of
the time dependence. Since we are dealing however with a continuous
spectrum, we expect a cut along the real axis and an analytic continuation,
in some region of the lower half of the complex plane (Im z - 0), might
be possible. It is assumed in the Brussels school's approach that such a
continuation exists and that singularities appear off the real axis. The
problem then is to study the nature of these singularities, bearing in mind
that ultimately we will use Wtto predict the time behavior of the system
while approaching equilibrium. To this end, we will mostly follow the
treatment, first given by George (1973) (see also Prigogine et al. 1979 and
Balescu 1975, Ch. 14), in analyzing the dynamics of the system in terms
of correlations.
The first step is to introduce a pair of operators P and Q such that they
satisfy the following relations:4
a) P + Q = 1 completeness
b) P = Pt, Q = Qt hermiticity
c) P = P2, Q = Q2 idempotency
d) PQ = QP = 0 orthogonality
This decomposition of unity follows from the initial decomposition of
the Hamiltonian H into the unperturbed energy Ho and the interaction
energy V,
PH= H, QH = V. (11)
The significance of these projection operators resides in the fact that the
complete density matrix Wmay now appear as the sum of two components
W, and W,
W, = PW, W>= QW (12)
whose evolutions will be seen to be at every moment independent from
each other. Their definition depends on the choice of P and Q which pro-
vides the 'language' in which we describe the dynamical evolution of a
particular physical system. Generally P is chosen to project onto the di-
agonal elements in some given representation in which Ho is diagonal. As
4To discuss the subdynamics approach in general terms we must introduce a complete set
of orthogonal projectors Pv according to which the density operator W is analyzed into
components W(v).Of particular interest however (especially for the problem of irreversibility)
is the 'vacuum' subdynamics we consider here since it contains the asymptotic time contri-
butions.
380 VASSILIOS KARAKOSTAS

diagonal elements give no information about the correlations between par-


ticles, PW = Wois called the 'vacuum of correlations', whereas the set of
off-diagonal elements of the density matrix W corresponds to the 'corre-
lation component' W,.Physically, the evolution of the vacuum component
W, leads to a 'memory' loss of the system's past state. It is similar in this
respect to Boltzmann's original hypothesis of molecular chaos. W, how-
ever, is the memory term. Its evolution describes the creation of correla-
tions due to the dynamical development. By considering both W, and W,
Prigogine et al. aim at avoiding the charge that their approach involves
some arbitrary coarse-graining scheme. I will argue in the sequel that ul-
timately they do not succeed in doing so.
To proceed further we decompose the resolvent of L into a sum of partial
resolvents corresponding to the decomposition of unity into P and Q,
R(z) = (z - L)-1 = P(z - L)-1P + P(z - L)- .Q (13)
+ Q(z - L)-P + Q(z - L)-Q.
It can be shown(l) that further manipulations lead to the identity:
(z - L)-= [P + C(z)] [z - PLP- T(z)]-1
[P + D(z)] + 'P(z). (14)
In this way, we introduce the basic operators of the kinetic description of
the system, namely,
the propagation operator
P(z) = (z - QLQ)-1 Q
the destruction operator
D(z)= PLQ 7P(z)
the creation operator
C(z(z)= (z) QLP
and the collision operator
'(z) = PLQ P(z) QLP.
This terminology stems from the fact that, for dissipative systems of our
interest, the 'asymptotic future collision operator' ( + io) as determined
in the limit z -> + io (which corresponds to the limit of WVas t -- + co)
describes the effect of collision processes; it corresponds to a transition
from the vacuum of correlations, accompanied by a dynamical evolution
in the correlation space P(z), and finally followed by a return to the vac-
uum of correlations. Moreover, the decay of initial precollisional corre-
lations depends on the properties of D(z), while fresh postcollisional cor-
relations, created in the system because of interactions (collisions), are
characterised by C(z).
Keeping these definitions in mind, let us apply the Laplace-transform
THE BRUSSELS SCHOOL'S ARROW OF TIME 381

solution of the Liouville Eq. (10), separately, to the vacuum of correlations


and to the correlation component. We obtain in this way a system of
coupled equations for the evolution of W, and Wc:

W(t) = (2ni)- f e-izt {(olR(z)lo) W,(o) ?+ (olR(z)lc) Wc(o)}dz (15)


c

Wc(t) = (2i)-1 ' e-iz {(clR(z)o) W(o + E (o)(cR(z)lc') Wc(o)} dz (16)


c'

where, for example, (oIR(z)lo)is the matrix element of the resolvent relat-
ing the vacuum of correlations to the vacuum of correlations. It is the
objective of the theory of subdynamics to decouple these equations and
to show that each component evolves independently from each other; i.e.,
it obeys its own subdynamics.
To this end, Prigogine et al. separate the various contributions received
by the contour integrals (15) and (16) to those resulting from the 'essential'
singularity at z = 0 as compared to the other singularities of the resolvent
which, lying sufficiently far away from the real axis, depend upon the
specific nature of the Hamiltonian. It is then useful to analyze the complete
density matrix W, according to these two types of contributions and to
write,
Wo(t)= o(t) + O(t)
Wc(t)= T(t) + WC(t) (17)
where Wo(t)is the contribution to the vacuum resulting from the pole at
z = 0, and Wo(t)is the contribution from the remainder. The same applies
to the correlation components Wc(t)and WV(t).
To identify subdynamics, the basic step is to associate these two classes
of contributions with corresponding operators of motion each of which
obeys a separate evolution equation. This introduces the operators S(t)
and S(t), each satisfying the semigroup property:5
Z(tl) Z(t2) = Z(tl + t2) tl, t2 > 0. (18)
Equation (17) can be written in compact form as,
Wt = 7, + ,W,. (19)

Then, the evolution of W can be given in vectorial notation by:

c(t)J
-(
( W()W(o))
(0)
(20)
=W
5The demonstration of the semigroup property (18) is a particularly delicate and involved
problem. Balescu and Wallenborn (1971), for instance, have shown that it can be extended
to finite negative values of t, and t2.
382 VASSILIOS KARAKOSTAS

where,
e-iot (1 + DC)-' et (1 + DC)-' D\
( Ce-it(1 + DC)-1 Ce-iot(l + DC)-1Di (21)
with the evolution operator 0 being a functional of the collision operator
T, while D and C being essentially dependent upon the irreducible de-
struction and creation (of correlations) operators considered previously.
Similarly, for the time development of the complementary component W
we have,

(W()~ ^(t) W(o) (22)

We see therefore that the operator of motion ((t)corresponds to the


effect of the singularity for z = 0. So, it gives us the asymptotic evolution
of the system which can be obtained only for long times. However, F(t)
can be defined, at least formally, for all times together with the operator
;(t) which includes the effect of all other singularities of the resolvent.
What then has been achieved in this way is a decomposition of the com-
plete unitary operator of motion U(t) into two parts,

U(t) = ((t) + ;(t). (23)


To realize how each of the Eqs. (20), (22) satisfy an independent dy-
namics, we introduce the operator

[I [I'P Q 11') (24)

which can be defined in terms of the corresponding operator of motion


S(t) in the limit t -> + io. (In fact this limiting operation corresponds to
a similarity transformation of P to H-subspace which will be seen to con-
stitute one of the determining relations of the nonunitary transformation
A to be defined in section 3). Hence,

fthm V(A-( + DC)-' (1 + DC)'


l=lim (t)= C(1 + DC)- C(1 + DC)-' D(
D (25)

The semigroup property (18) for ?(t) then leads to,

fi2 = i. (26)
Thus, II is a
essentially projection operator associated with the singularity
of the resolvent at z = 0. fI enjoys other symmetry properties which do
THE BRUSSELS SCHOOL'S ARROW OF TIME 383

not concern us here. The point to stress however is that it commutes with
the Liouville operator,

HL = Lf. (27)
As a consequence of (25) and (27) we have,

((t) = fe-iLt = e-iLt I. (28)


The commutativity relation (27) in effect shows that the projected evolu-
tion on the subspace associated with the operator TI is stable under L.
This is why this evolution constitutes a subdynamics. Indeed, in this for-
mulation Wtcan be obtained by the action of the projector H on the total
density matrix Wt

ftJ = I W,. (29)


The evolution then of WUas generated by the operator of motion S(t) in
Eq. (20) defines a subdynamics, since
W = X(t) W(o) = le-iLt W(o) = e-iLt i W(o) = e-iLt W(o). (30)

Similar considerations hold for ((t).The corresponding operator

n = lim 2(t) (31)


t--+O

is such that, it satisfies the decomposition

[I + 1I = 1. (32)
From Eqs. (26) and (32) follows immediately that

f = l = o (33)
and
2 = fl. (34)
Thus I is also a projection operator and more importantly is orthogonal
to the operator fI. Relations (32) and (33) capture the essence of the sub-
dynamics concept. Their physical content allows the decomposition of the
dynamics in the total space of the density matrix Wtinto two completely
separated subdynamics associated with the mutually orthogonal subspaces
TIand nI. Using the decomposition (32), we may indeed write for all times

w = W + W = f W + ^W. (35)
Each of these parts satisfies a separate evolutionequation as a consequence
384 VASSILIOS KARAKOSTAS

of the orthogonality condition (33). Hence, in keeping with the preceding


analysis, we may associate to W and W separate dynamics and speak of
subdynamics generated by the operators of motion ;(t) and S(t), func-
tioning respectively in the complementary subspaces If and II.
According to Prigogine et al. (1973, 1979) the nI-subdynamics describes
the irreversibleapproach to equilibrium, whereas, the 11-subdynamicsde-
scribes the 'non-thermodynamic' evolution of correlations introduced in
the specification of the initial distribution state of the system. In their view,
equilibrium is attained when the I W component reaches the canonical
distribution in its subspace, provided that "the projection of W into the
complementary subspace II gives a vanishing contribution".
To say, however, that certain correlations (or, equivalently, elements of
the density matrix W) vanish is, at best, to make a statement about the
average values of certain microscopic variables. The probabilities involved
in the definition of the average are the ordinary probabilities of statistical
mechanics (as opposed to irreducible quantal probabilities) reflecting our
ignorance or lack of interest in microscopic details. Thus the statement
that the incoherent n W component vanishes simply means that the cor-
responding information disappears from our description of the system. In
principle the information could always be supplied if one was willing to
expend enough free energy to get it.
It is for this reason that we regard the projector fI as being equivalent
to a generalized form of 'coarse-graining'; it provides a contracted de-
scription of the system in that only the IH W portion of the information
available in the full ensemble W can be shown to exhibit kinetic or irre-
versible behaviour. A complete dynamical description would require in
addition the consideration of the evolution in the fI-subspace. Contrary
to this, Prigogine et al. proceed from the total unitary evolution U, by
eliminating a certain kind of initial conditions, for instance, those corre-
sponding to precollisional correlations in the antiscattering case, forms of
which however can be realized in time-reversal spin echo experiments as
well as in cases of plasma echo and turbulence. (Descriptions of spin echo
experiments where a sample of protons is prepared in such a state from
which the system appears to be running backwards in time can be found,
for example, in Rhim et al. 1971, Kreuzer 1981, 330-336, and Brewer et
al. 1984). In such situations, loosely speaking, velocity inversion would
necessarily lead the evolution of the system outside the fI-subspace as
persistent 'antirandomization' of momenta (or spin axes orientation)
would lead to increasing deviations from thermal equilibrium for t ->
- oo.This means that motion in the Il-subspace will now produce coher-
ent effects opposing the evolution in II. The effect of the time inversion
THE BRUSSELS SCHOOL'S ARROW OF TIME 385

operation will be extensively discussed in section 3 where I will consider


the possible constructions of the Brussels school's A-transformation with
respect to time change t -> - t.
Returning to the subdynamics theory, the fact that has to be noticed is
that the density matrix Wtwhen projected in any of the subspaces in Eq.
(35) still possesses a vacuum part W, and a correlated part Wcin accor-
dance with relation (17). Moreover, these various parts are not indepen-
dent from each other. By using the explicit form (25) of the operator nI
to calculate ft W, it can be shown that the correlations in fI appear as
linear functionals of the vacuum component

W(t) = Co(t). (36)


Thus in the subspace II only the vacuum elements Woare independent.
In the complementary subspace II, the situation is just the opposite.
there the vacuum component is a well-defined functional of the correla-
tions

Wo(t)= -D W(t). (37)


These correlations are thereby the independent elements in the fI-
subspace.
It becomes then natural to use in the decomposition (17) of the complete
density matrix as independent elements the 'privileged' components Wo
and W, since W, and Wocan be obtained from Eqs. (36) and (37) respec-
tively. It is now the purpose of the whole program on dissipative processes
adopted by the Brussels school to show that these privileged components
satisfy separate equations of motion.
To identify the structure of the various independent evolutions we must
first consider the noncanonical transformation developed by Prigogine
and his coworkers. We do not intend here to discuss fully this transfor-
mation (details can be found in George et al. 1985, Prigogine et al. 1983).
We will summarize only the essential steps and indicate those properties
which are closely connected with the irreversible behavior of quantal sys-
tems in the thermodynamic limit.

3. Star-Unitary Transformation Theory. The starting point is the intro-


duction of two operators x, and xc in the P and Q subspaces respectively,
such that

0xo* - PPIP = (1 + DC)-1 (38)


and

xX* = QIQ = (1 + CD)-1. (39)


386 VASSILIOS KARAKOSTAS

The asterisk denotes the so-called 'star-conjugation' symmetry of opera-


tors having a continuous spectrum in the Liouvillian space. This property
plays a central role in both subdynamics and transformation theory and
will be discussed further below together with other symmetry conjuga-
tions. The point on which we want to insist at present is that star-
conjugation endows the functional space of the dissipative density matrix
with a star-hermitianmetric, necessary for the study of processes involving
a change in the direction of time. Of course, as one might expect, if the
spectrum of the operator is discrete, this operation reduces to the usual
'hermitian conjugation'.
The need of introducing a further operator x is a direct consequence of
the fact that the evolution operator 0 which appears in the structure of
((t)does not possess well-defined hermiticity properties and hence may
have imaginary eigenvalues. (Note that most of the difficulties encoun-
tered in the Brussels school's approach in defining a nonequilibrium en-
tropy beyond the 'Boltzmann approximation' may be traced to this fact).
On the contrary, the operator x can be shown to satisfy definite symmetry
properties with respect to star-conjugation operation alluded to above. In
fact, the various components x,, xc constitute particular elements of an
evolution operator A (to be defined immediately below), where xo, xc are
respectively the P-P and Q-Q components of A
Xo = PAP, x, = QAQ (40)
and x*, x* those of its inverse A-1
x* = PA-1P, x* = QA-1Q. (41)
Now to construct A,6 observe that by virtue of Eqs. (38), (40), and (41)
the fl-subspace may be viewed as a transformation of the P-subspace
satisfying the similarity relation (we drop from now on the upper symbols
on the I projectors as we will be dealing in the sequel with the fl-subspace
alone, which defines according to George and Prigogine (1974, 341) the
'macroscopic level of quantum mechanics')
I = APA-1. (42)
Then from the formal structure of II, cf. Eq. (24), and the relations (40),
(41) defining x, the operator A can be written as
A (PAP PAQ \
QAP QAQ)
( -
Dx c
cx xC,
'
(43)'

6Recently, Prigogine et al. (1988), Petrosky et al. (1991) and Hasegawa et al. (1991) instead
of deriving A through relation (42), start with A and use (42) to derive n. These two pro-
cedures of course give identical results for the dynamics of correlations for a complete set of
orthogonal I projectors. It should be noted in the light of this equivalence that the latter
approach although simplifies presentation is nonetheless mathematically more abstract and
physically less intuitive.
THE BRUSSELS SCHOOL'S ARROW OF TIME 387

while its inverse A-~ may be written in terms of the star-conjugate x* as7

'
-X* C x*

The importance of the operator A stems from the fact that it relates the
original density matrix W

W= Wo) (45)

satisfying the unitary Liouville-von Neumann equation (8), to a new rep-


resentation W()

( W(P)) (46)

whose elements can be expressed as linear combinations of the initial priv-


ileged components Woand Wc

Wo) = x* (1 + DC) WI, W) = x* (1 + CD) WC (47)


through a non-unitary (star-unitary) transformation
W) = A-1 W, (48)
which will be shown to incorporate the entropy concept as a functional of
the redefined density matrix W(P).
It is in this so-called 'physical representation', hence the raised 'p', that
the new diagonal elements (the redefined vacuum WY))and the new off-
diagonal elements (the redefined correlations W?)) evolve independently
from each other. Recall that in the integral representation of the Liouville-
von Neumann equation, the temporal evolution of the initial Woand W,
components is linearly related to each other. To compare with the (p)
formulation of dynamics we rewrite the corresponding equations (15) and
(16) in differential form as,
aWo PLP Wo+ PLQ
i -?= Wc
at

a-at = QLP Wo+ QLQ Wc. (49)

It can now be shown (e.g. Prigogine et al. 1983) that the A-transformation
7Note that x* is not the inverse of x as the factorisation relations (38) and (39) indicate.
This is a necessary condition the 'dressing' operator x should satisfy in order for the trans-
formation A to be assigned a definite symmetry under star-conjugation and such that A-'
= A*. But more on this below.
388 VASSILIOS KARAKOSTAS

allows the above equations to be expressed in a diagonal form, i.e., a


separate equation for each type of correlation. Specifically, the system of
the coupled equations (49) can be replaced in the (p) representation by the
set of independent equations:
aWY)
at
i W())
=
i t- = c W?), (50)

where the new generator of motion () is now directly related to the Liou-
ville operator L via the transformation A by the relation
(D A-' LA. (51)
What has been achieved in this way is that once the solution of Eqs.
(50) is known the whole time evolution of the system is known. As to the
time variation of the elements of the original density matrix W, it can be
obtained from the transformation relation (48) and the Eqs. (36), (37)
which relate the privileged components WI, WIjto the other components
of W. In other words, Prigogine et al., by replacing W with W(P)and L
with (, bloc-diagonalize the initial equation in the 'physical representa-
tion'.
Before inquiring into the time-asymmetry behavior of the (p) dynamics,
it is important first to inquire into the nature of the transformation law
(48). To fix the ideas, the preceding analysis basically showed that through
a resolution of the singularities of the resolvent in the asymptotic limit z
-> + io the formal solution of the Liouville equation can be formulated
in terms of Wand W(). As both representations are considered equivalent
under transformation A (Misra et al. 1979), it is natural to require that
all average values of observables should be kept invariant in going to either
representation. That is,

(0), = Tr Ot W = Tr O(P)tW(P). (52)


As well known, the time evolution of observables is given by the Heisen-
berg equation
ao(t)
= - LO(t), (53)

which differs from the Liouville equation for the evolution of states by a
mere replacement of L by -L. (This replacement plays an essential role
in the theory because it is intimately connected, through the L - t invar-
THE BRUSSELS SCHOOL'S ARROW OF TIME 389

iance (5), to a change in the direction of time). We therefore require that


for an observable 0,
0 A)= A-1(-L) 0. (54)

Inserting (54) into (52) and comparing with the initial form of the trace
we obtain

A-1(L) = At(-L). (55)


Relation (55) defines what has been called by the Brussels school a 'star-
unitary transformation'. The introduction of the 'star' notation denotes the
combinedoperations of taking the hermitian conjugate and reversing L

A*(L) = At(-L). (56)

Accordingly, the above operator is called the 'star-Hermitian' operator


associated with A. Expression (56) indicates in effect that for star-unitary
transformations the inverse of the initial operator is equal to its star-
Hermitian conjugate. The important aspect of these transformations is
that they are nonunitary. They enable us to go from the initial dynamical
representation of motion determined by the reversible operator L to a new
(physical) representation in which the evolution is governed by the oper-
ator D exhibiting a kind of irreversible behavior, as will be shown imme-
diately below.
To this end, let us compare the Liouville equation for the density matrix
Win the initial representation

= -(iL)W (57)
at
with the time evolution for W in the (p) representation

= -(iD) W(). (58)


at
It is easily verified that both evolution operators (iL) and (iF) are star-
hermitian, i.e.,

(iL) = (iL)* (59)

(i0) = (ij)*. (60)


Recall that for an operator to be star-hermitian, it must be invariant with
respect to the combined operations
(adjunction)A (inversionL -> - L). (61)
390 VASSILIOS KARAKOSTAS

The star-hermiticity of an operator, therefore, can be realized in two pos-


sible ways: the operator may be either

(hermitian)A (even in L) (62)

or

(antihermitian)A (odd in L). (63)

Now the invariance (59) comes from the fact that (iL) is antihermitian

(iL) = -(iL)t (64)


and odd under the L-inversion. While the star-hermiticityproperty of (60)
is a direct consequence of the definition of the transformation law (48),
the star-unitary of A (56), and the usual hermiticity of L. The new im-
portant feature here is that the generator of motion (i0) in the 'physical
representation' can be shown to contain parts which are even (superscript
e) or odd (superscript o) under the L-inversion
e e o o
( (L) = ( (-L), ( (L)= -) (-L). (65)
It can thus be written as

iD = (i(D) + (iO). (66)


In accordance then with the general setting of operations (62) and (63)
and in order for the star-hermiticity of (iD) to be preserved, the even part
must be hermitian while the odd part must be antihermitian, i.e.,
e e o o

(iD) = (i0)t, (iD) = -(iDI)t. (67)


The possibility of decomposing the new evolution operator IDin this
way concludes the scope of the Brussels school's approach to the irre-
versibility problem. Now, the star-unitary transformation A, which relates
D to L through relation (51), can lead to a symmetry-breaking of the
L - t invariance present in the Liouville equation. To realize this, observe
that the star-hermiticity property of (i0>)can be satisfied in two different
ways (62), (63) with respect to L-inversion as compared to the unique way
(63) of satisfying star-hermiticity for (iL). It is the appearance of the even
part in (i0) which guarantees the possibility of a broken time-symmetry
description of the system's evolution. When therefore the 'dissipativity
condition'

(iO) 0 0
()e
(68)
THE BRUSSELS SCHOOL'S ARROW OF TIME 391

is satisfied, the evolution of the system as determined by Eq. (58) can be


split
ao W(P) e

- = -[(i@) + (iO)] W() (69)


at
o e

into a reversible part (i0D)and an irreversiblepart (iD). Moreover, as a


consequence of condition (68), we may associate to the system's devel-
opment a functional Qfof its state W(), such that, it satisfies the relations
-
Q(W(p))> 0, d?>/dt 0, (70)
and in terms of which a microscopic expression of the second law of
thermodynamics may be formulated. In fact, it can be shown (e.g. Pri-
gogine et al. 1977) that the Lyapounov function
= (W(), W()) = (W, A-ltA-' W) (71)
can only decrease in time and it may therefore be used as a reasonable
candidate for the entropy of the system.
In view of these considerations Prigogine and coworkers conclude that
"time's arrow, the entropy ... is now explicitly displayed ... [within] . .
a purely dynamical content" (George et al. 1985, 224; see also Hasegawa
et al. 1991, 263). For them, entropy, as resulting from irreversibility, is a
dynamic property based on the possibility of constructing a noncanonical
transformation A which enable us to go from the microscopic reversible
dynamics to the macroscopic irreversible thermodynamics. In this per-
spective the approach to equilibrium appears as the evolution of the 'trans-
formed Liouvillian' operator IDwhich acts on the (p) representation.
In the next section I shall criticize the Prigogine's school 'mechanical
theory' of irreversibility. I shall argue, in particular, that their account of
irreversible processes neither endows time with an intrinsic direction as
thermodynamically irreversible processes seem to point to, nor does it
provide a fundamental physical basis for the phenomenological direction-
ality of time.

4. The Brussels School's Second Law May Not Be Irreversible.To provide


a coherent criticism of the Brussels school's approach to the problem of
irreversibility let me first summarize the essential steps of their overall
program.
They demonstrate that for sufficiently unstable systems one can con-
struct a nonunitary transformation A which leads to the physical repre-
sentation (W() = A-1 W) by transforming the hermitian operator L to
the star-hermitian operator (D(( = A- 'LA) containing a dissipative even
part. The invertible character of the A-transformation essentially amounts
392 VASSILIOS KARAKOSTAS

to a similarity transition from the initial reversible description (W) defined


by the unitary group operator U, = e-iL'to a dissipative description (W(?))
governed by a contraction semigroup operator E, = e-iO' of a Markov
process exhibiting irreversible behavior. Through an analysis of the sin-
gularities of the resolvent R(z) = (z - L)-' in the limit z -> + io, they
are able to show that the physical evolution E, = A-1 UA describes the
approach of the system towards equilibrium in the future direction of time
(t - 0) with which they associate an entropy functional Q (W(P))of the
system's state.
The inefficacy of this approach, however, and as far as the problem of
irreversibility is concerned, becomes evident by considering the fact that
the time-reversal symmetry of the underlying dynamics allows us to con-
struct equally well another transformation A off the limit z -- - io, such
that, the new transformed states W'P) = A-1 W will evolve now under a
Markov process corresponding to the semigroup description t =
A-1 UA_ for the negative direction of time (t < 0).
The point that I want to stress here is that the reversibility of the original
dynamical motion implies that if symmetry breaking is possible for one
direction of time, it is also possible for the reverse direction. Hence, the
unitary group of evolution U, can be mapped through the two distinct
transformations A+ and A, corresponding respectively to forward and
backward directions of time, into two distinct contraction semigroups
2+ and SE approaching equilibrium accordingly for t -> + oo and t ->
- oo(in this connection see also Sklar 1993, 276).
The question then is whether there exists any intrinsic physical distinc-
tion between the two possibilities of symmetry breaking (corresponding
to the opposite directions of time) that would permit us to rationalise the
choice of the one semigroup of operators over the other. In other words,
on what physical grounds can the apparent degeneracy between A+ and
A_ (or +[ and Z-) be lifted? Or put it another way, why we ought to
adopt A+, as the Brussels school suggest, instead of A_?
It is not surprising that Prigogine and coworkers, like Boltzmann some
one hundred years ago, appeal to features of initial states in terms of the
type of 'correlations' involved in order to lift the degeneracy between the
two A transformations. (See, especially, Prigogine et al. 1983, George et
al. 1985, Petrosky et al. 1990. The corresponding case for classical dynam-
ical systems is discussed, for example, in Courbage et al. 1983, Prigogine
et al. 1987). The authors hope to establish a connection between what
kinds of initial states are observed in nature or can be prepared in the
laboratory and the subsequent thermodynamic evolution towards the de-
sired direction in the future.
To this end, they express the second law of thermodynamics on the
microscopic level as a selection principle on initial conditions by requiring
THE BRUSSELS SCHOOL'S ARROW OF TIME 393

that not all distribution functions but only a suitable subset of them which
tend to regular invariants for t -- + oo can correspond to the class of
physically realizable states.
In order for a function X to be a regular invariant, the asymptotic evo-
lution of the system in the limit z -> + io (which corresponds to t -
+ oo) has to satisfy the following two conditions:
lim zD(z)X, = 0 (72)
z-*+ io

and
lim zP(z)X, = 0 (73)
z- + io

The behavior of the destruction operator near the origin in Eq. (72) implies
the complete elimination of initial correlations around equilibrium, while
the stronger condition (73) prohibits the propagation in the correlational
space of precollisional correlations. If for a given distribution function
one of the above conditions does not hold, we speak of a singular invariant.
The basic physical difference between regular and singular invariants
as appears from Eqs. (72) and (73) is that regular invariants correspond
to quantities which are conserved by the result of collisions, whereas, sin-
gular invariants necessarily involve persistent correlations. According
therefore to Prigogine's et al. interpretation of the second law, only if the
initial (precollisional) correlations disappear asymptotically for t -- + oo,
do we have to accept normal thermodynamic behavior in the evolution of
physical systems towards equilibrium. As an illustration, the formulation
of the second law as a selector of initial conditions would exclude from
the set of physically realizable states precisely the future-directed corre-
lations of the type present in 'incoming spherical waves' (in the time-
inverted process of the scattering case) that would transform into 'out-
going plane waves' converging to a point in the infinite future.
Further on, Prigogine et al. set about to demonstrate that the selection
principle, so formulated, is propagated by dynamics. The point here is
that when the projector II acts on distribution functions W, it contains
generally in the asymptotic limit the quantity D W. Now, it can be shown
that when acting on a state W(+),containing postcollisional correlations
as a result of past collisions, condition (72), namely,
lim zD(z) W+) = 0 (74)
z- + io

is satisfied, and W(+)goes to equilibrium for t -> + oo.On the contrary,


when H acts on a state W-), containing initial (precollisional) correlations
that may cause future collisions
lim zD(z) W(-) = finite, (75)
z- + io
394 VASSILIOS KARAKOSTAS

the condition of Eq. (72) is violated, and WJ-) reaches equilibrium for
t- -- oo.
Then in accordance with the formulation of the second law acting as a
selection principle, Prigogine et al. conclude that the Wr-)states involving
precollisional correlations of long range have to be excluded from the class
of physically acceptable states.
However, through the same procedure but by considering the extraction
of singularities of the resolvent of L in the neighborhood of z = - io, we
may define equally well a 'time-oriented' projection operator f_ such that
when acting on W, the WV-)states tend now to a singular invariant in the
far future and the W(+) states to a regular invariant in the far past. The
approach to equilibrium according to this time ordering of events corre-
sponding to the sequence [correlations -> collisions] is but the time-
inverted process of the one examined previously and based on the sequence
[collisions -> correlations]. From the point of view of dynamics, collisions
can give rise to correlations and correlations to collisions. The relation
between them is symmetric with respect to time-reversal transformation
and consequently distribution functions can reduce to regular invariants
in the future or in the past alike, the direction of time being determined
by the nature of the initial conditions concerned. It is then hardly under-
stood how, if not by fiat, the Brussels school's formulation of the second
law as a selection principle restricts the class of physically realizable states
to distributions that tend to a regular invariant for t -> + oo and to a
singular one for t -> - ooat the expense of their dynamically permitted
time-inverse processes.
As to the repeatedly made claim, aiming presumably at promoting a
'mechanical' interpretation of entropy, that their selection principle is
propagated by the dynamical operator II = APA-1, it also holds true
that the opposite content of the so formulated second law, excluding now
W(+) as unrealizable states, would be propagated by the dynamics by
means of II_. The operator HI_is related with the same diagonal P-com-
ponent of the functional space of W through a transformation A_, ob-
tained from A by an L inversion. Although the A+ and A_ operators are
different, they are related nonetheless through the star-unitary property
(55), i.e.,
A* _ A-' = At (76)
Consequently, as one might expect, the II+ and H_ operators are star-
hermitian
fl* -lt = H+. (77)
The star-operation in II actually interchanges the sign of the analytic
continuation z -> ? io between the creation operator of correlations C
THE BRUSSELS SCHOOL'S ARROW OF TIME 395

and the destruction operator of correlations D. I.e., C and D in Prigogine's


(p) representation of dynamics are L (velocity)-inverted hermitian conju-
gates whose connection reflects the time change (t - - t) in usual dy-
namics. It is only in C and D that we have to distinguish between expres-
sions in terms of L and -L. Under these terms therefore the relation
between II+ and H_ as well as between A+ and A_ is time-reversal in-
variant; A_ is the mirror image of A+. It determines the asymptotic evo-
lution of the system towards the past, leaving the physics of the system
unaltered. Moreover to this type of evolution we may associate an entropy
functional Q_ which decreases now monotonically towards its minimum
value at equilibrium in the negative direction of time.
It is then evident that the symmetry breaking transition from dynamics
to thermodynamics as described in the subdynamics-based theory of ir-
reversible processes does not prevent us from constructing the 'time-re-
versed' analogues for each of A, Et, and Q yielding the same limiting results
except that now the limit is set at t -> -oo.
Prigogine et al., however, point out that their formulation of the second
law on the microscopic level as a selection principle requires that the two
transformations A+ and A do not operate "on the same time-symmetric
set of states but on two distinct sets of states" (Courbage and Prigogine
1983, 2413; recall the distinction between regular and singular invariants
of motion). In fact, this is the property which, they feel, lifts the degeneracy
between A+ and A_ or between the corresponding semigroups z+ and
~t-. The entropy Q+ selects A+. But, why choose Q+ over Q_ as the
selector of initial conditions? The answer to this seems to be that Q+ single
out states that go to equilibrium as t -> + oo. But clearly this is just to
beg the question, because a theoretical explanation of it is what the prob-
lem of macroscopic irreversibilityis all about.

5. Final Considerations.The Brussels school, instead of explaining why it


is that systems evolve to equilibrium in the future and not in the past,
presupposes that the 'correct' direction of time is known. To confront
however the irreversibility problem in a non-circular way, it is essential
first to assume that physical systems approach equilibrium backwards in
time in the same way as they do forwards in time and that our ordinary
inclination to favor dynamical evolution towards the future direction is
just one manifestation of the macroscopic time-asymmetry in question. In
sharp contrast, Prigogine et al. introduce temporal asymmetry right from
the beginning in the mathematical structure of the theory. The type of
analytic continuation employed in the construction of the subdynamic
operators virtually determines the direction of time. Following the cus-
tomary procedure for the study of the asymptotic evolution of non-equi-
librium systems, they associate to the positive direction of time (t > 0) an
396 VASSILIOS KARAKOSTAS

analytic continuation into the lower half plane (Imz < 0) and to the neg-
ative direction of time (t < 0) an analytic continuation into the upper half
plane (Imz > 0). In the first description the system approaches equilibrium
in the far future (z -> + io), whereas, in the second in the far past
(z -> - io). But clearly if the problem of the irreversibility of time is to
be solved, the formalism of the theory must incorporate past and future
on a strictly equal basis because otherwise the analysis is being rigged in
advance.
Thus the Brussels school's developments, despite their merits, cannot
provide a suitable basis for the characteristic feature of irreversible evo-
lution as it is expressed by the second law of thermodynamics. Their re-
formulation of dynamics in terms of the star-unitary transformation A
cannot possibly lead to processes that point to a preferred direction of
time. Indeed, as we showed, the symmetry breaking induced by A does
not determine whether the future orientation of time coincides with the
direction in which entropy increases. A can only distinguish structurally
between both orientations of time; it does not, however, allow us to tell
the past from the future. In this respect the star-dynamics mimics the usual
dynamical procedures described by quantum (or classical) mechanics
where the inversion of time has no effect on the physical situation. The
only difference being that the character of lower symmetry of the dynam-
ical evolution in the (p) representation (star-unitarity instead of unitarity),
renders possible the mathematical mapping of the unitary group Ut into
two contractive semigroups E+ and A-, reaching equilibrium for either
t- + oo or t - oo.
Now to use the second law as a selector on initial conditions in order
to lift the apparent degeneracy is unfortunate. For it is not permissible to
invoke an asymmetric argument with respect to time reversal, if one is to
explain temporal asymmetry. Prigogine et al., by interpreting the second
law as a selection principle, assume what they want to prove. They deter-
mine the one-wayness of time by supposing that its future direction co-
incides with the direction relatively to which regular invariants converge.
From a purely physical point of view, however, this determination of the
arrow of time is arbitrary; physically, there is no reason why the future
direction of time could not be defined as the direction relatively to which
singular invariants diverge. Prigogine et al. simply choose a proper subset
of distribution functions that tend to regular invariants when a future
situation is to be described as the consequence of a certain condition,
whereas, the singular solutions may be chosen to calculate the conse-
quences that same condition had in the past.
Moreover the problem cannot be resolved by pleading initial conditions;
if initial states of the type [collisions -> correlations] are allowed, then, the
reversibility of the underlying dynamics guarantees that the preparation
THE BRUSSELS SCHOOL'S ARROW OF TIME 397

of states based on the sequence [correlations - collisions] is also allowed.


They are velocity inverses of each other. What is then so special about
post-collisional correlations which according to Prigogine et al. permit
regular distribution functions to approach equilibrium in the future as
compared to the pre-collisional ones which prevent the system from doing
so? Why should it be physically impossible to find or suitably prepare a
system to be in one of these 'improbable' initial states? There is an old
saying in physics: anything that is not forbidden becomes compulsory.
One should find a criterion to distinguish between abnormal correlations
and normal correlations. Prigogine's exclusion principle by merely rec-
ognizing some correlations as thermodynamically legitimate and others as
thermodynamically deviant does not suffice; it requires further analysis.
For, it does not refer to some primitive, factual statement about nature
which defies any explanation, but instead it states that nature displays a
certain pattern by constantly prohibiting the creation of pre-collisional
correlations. It is precisely this pattern itself that needs further clarification
and explanation. In particular, the Brussels school needs a physical ac-
count to justify the move from denying the possibility of preparing pre-
collisional correlations to the impossibility of their being realized in na-
ture. In the absence of such an account, Prigogine et al. can make their
point only by appealing to our temporally asymmetric way of looking at
things and to regard it as natural for physical systems to evolve towards
equilibrium only in the far future. For them, therefore, the irreversibility
of time can only be founded in phenomenology, involving thereby an ap-
peal to the macroscopic experience of the direction of time by a conscious
observer.
In view of the above analysis, I feel that the parts of Prigogine's argu-
ments with regard to the consideration of regular as opposed to singular
invariants, post-collisional as against to pre-collisional correlations, en-
tropy as selector on initial conditions-despite their usefulness in the anal-
ysis of de facto dissipative processes-do not solve the real problem of
irreversibility, that is, the problem of determining a preferred direction of
time during the approach of systems to equilibrium. In concluding this
paper, I advance the following general argument which shows, indepen-
dently of the mathematical formalism of the theory concerned, that sta-
tistical reasoning alone is not sufficient to explain the arrow of time. In
other words, by adding probabilistic considerations to an underlying time-
reversible dynamics, this, by itself, is not going to produce a time asym-
metric feature of the kind that would normally serve as an explanation of
entropic time asymmetry.
To see this, let T stand for the Brussels school's subdynamics approach
or any other theoretical scheme whose purpose is to describe the time
evolution of physical systems towards equilibrium. Hence, by definition,
398 VASSILIOS KARAKOSTAS

T deals with an initial nonequilibrium state W(t) which is confined to a


certain region of its phase space at time t. The approach then of the system
to equilibrium at any time t' is determined by W(t'). The decisive question
is whether, in order to predict W(t'), T needs to take into account whether
t' is actually later or earlier than t, in the ordinary sense of macroscopic
time. If not, then T is symmetric with respect to time reversal, allows W
(t') to reach equilibrium either in the far future or in the far past alike,
and therefore incorporates no preferred direction of time. If, on the other
hand, T depends upon the further information of time ordering, then this
can only be as a consequence of the fact that T already treated the future
and past directions of time differently-for otherwise the extra informa-
tion would be redundant. In this case therefore any preferred direction of
time would have been built into the structure of T right from the begin-
ning, as it is the case with the Brussels school's microscopic theory of
irreversible processes.
This being so, the theory of the Brussels school nonetheless should be
commended for the mathematical ingenuity and physical insight that it
has introduced into the foundational problem of the irreversibilityof time.
By extending the formulation of dynamical systems to larger spaces so as
to incorporate consistently instabilities of motion, the Brussels school does
provide a new conceptual structure in which certain questions with rela-
tion to irreversibility of physical processes can now be considered, which
otherwise could not even be formulated within the framework of tradi-
tional quantum statistical mechanics.
To be sure, one cannot get time asymmetry from either the subdynamic
decomposition or the star-unitary transformation. As we have constantly
argued, for any thermodynamic subdynamics, one can construct a (time-
reversed) anti-thermodynamic subdynamics, and for each star-unitary
transformation going to equilibrium in the future in its fine-grained en-
tropy, one can construct a star-unitarilytransformed distribution function
going to equilibrium in the past in terms of its fine-grained entropy. The
theory of the Brussels school however does enlighten us about the ways
in which the kinetically behaving reduced descriptions of subdynamics and
the star-unitarily transformed representations derive their natures from
the underlying dynamics of sufficiently unstable systems. It is important
to be able to show on the basis of dynamical instabilities of motion that
the evolution of a system can be split into two contractive semigroups
exhibiting irreversible behavior in their approach to equilibrium. In the
light of the Brussels school's developments, this may be characterized as
the dynamical content of the second law of thermodynamics. This dynam-
ical content is of course absent in the conventional unitary representation
of dynamics. One may say then that the essence of the Brussels school's
approach to the irreversibility problem is the extension of dynamics to
THE BRUSSELS SCHOOL'S ARROW OF TIME 399

larger spaces where the split into two semi-groups is manifest. As well-
defined trajectories or wave functions are not in the domain of the ex-
tended Liouville space, irreversibility(as one may expect) is not a property
of pure states. The question therefore concerning de facto irreversiblepro-
cesses is for which classes of real physical systems the split (or dynamical
content) is feasible and what are the admissible densities of the ensembles
involved. As to the conceptual aspect of the irreversibility problem, one
still needs (as we have argued) to connect in a suitable, non-ad hoc manner,
the dynamical content of the second law with its empirical content ac-
cording to which just one of the two semi-groups is realized in the future
direction of time.

REFERENCES
Balescu, R. (1975), Equilibrium and Non-Equilibrium Statistical Mechanics. New York:
Wiley.
Balescu, R. and Wallenborn, J. (1971), "On the Structure of the Time-Evolution Process in
Many-Body Systems", Physica 54: 477-503.
Born, M. (1949), Natural Philosophy of Cause and Chance. Oxford: Clarendon Press.
Brewer, R. and Hahn, E. (1984), "Atomic Memory", Scientific American Dec. 251: 42-50.
Courbage, M. and Prigogine, I. (1983), "Intrinsic Randomness and Intrinsic Irreversibility
in Classical Dynamical Systems", Proceedings of the National Academy of the Sciences
of the United States of America 80; 2412-2416.
Coveney, P. and Highfield, R. (1990), The Arrow of Time. London: Allen.
Davies, P. C. W. (1977), The Physics of Time Asymmetry. Berkeley: University of California
Press.
George, C. (1973), "Dynamics of Correlations", Physica 65: 277-302.
George, C. and Prigogine, I. (1974), "Quantum Mechanics of Dissipative Systems and Non-
canonical Formalism", InternationalJournal of Quantum Chemistry8: 335-346.
George, C., Mayne, M., and Prigogine, I. (1985), "Scattering Theory in Superspace", in
I. Prigogine and S. Rice (eds.), Advances in Chemical Physics. New York: Wiley, pp.
223-299.
Hasegawa, H., T. Petrosky, I. Prigogine, and S. Tasaki (1991), "Quantum Mechanics and
the Direction of Time", Foundations of Physics 21: 263-281.
Jancel, R. (1963), Foundations of Quantumand Classical Statistical Mechanics. Oxford: Per-
gamon Press.
Kreuzer, H. (1981), NonequilibriumThermodynamicsand its Statistical Foundations. Oxford:
Oxford University Press.
Misra, B., I. Prigogine, and M. Courbage (1979), "From Deterministic Dynamics to Prob-
abilistic Descriptions", Physica A 98; 1-26.
Nicolis, G. and Prigogine, I. (1977), Self-Organisationin NonequilibriumSystems. New York:
Wiley.
Penrose, R. (1987), "Newton, Quantum Theory and Reality", in S. W. Hawking and
W. Israel (eds.), ThreeHundred Yearsof Gravitation.Cambridge:Cambridge University
Press, pp. 16-49.
Petrosky, T. and Prigogine, I. (1990), "Laws and Events: The Dynamical Basis of Self-
Organisation", CanadianJournal of Physics 68: 670-682.
Petrosky, T., I. Prigogine, and S. Tasaki (1991), "Quantum Theory of Non-Integrable Sys-
tems", Physica A 173: 175-242.
Prigogine, I. (1980), From Being to Becoming. San Francisco: Freeman.
Prigogine, I. and George, C. (1983), "The Second Law as a Selection Principle: The Micro-
scopic Theory of Dissipative Processes in Quantum Systems", Proceedings of the Na-
tional Academy of the Sciences of the United States of America 80: 4590-4594.
Prigogine, I. and Elskens, Y. (1987), "Irreversibility, Stochasticity and Non-Locality in
400 VASSILIOS KARAKOSTAS

Classical Dynamics", in B. Hiley and F. Peat (eds.), Quantum Implications. London:


Routlege, pp. 205-223.
Prigogine, I., C. George, F. Henin, and L. Rosenfeld (1973), "A Unified Formulation of
Dynamics and Thermodynamics", Chemica Scripta 4: 5-32.
Prigogine, I., F. Mayne, C. George, and M. Haan (1977), "Microscopic Theory of Irrevers-
ible Processes", Proceedingsof the National Academy of the Sciences of the UnitedStates
of America 74: 4152-4156.
Prigogine, I. and Grecos, A. (1979), "Topics in Nonequilibrium Statistical Mechanics", in
T. Francia (ed.), Problems in the Foundations of Physics. Amsterdam: North-Holland,
pp. 308-343.
Prigogine, I. and Petrosky, T. (1988), "An Alternative to Quantum Theory", Physica A 147:
461-486.
Rhim, W., A. Pines, and J. Waugh (1971), "Time-Reversal Experiments in Dipolar-Coupled
Spin Systems", Physical Review B 3: 684-695.
Sklar, L. (1993), Physics and Chance: Philosophical Issues in the Foundations of Statistical
Mechanics. New York: Cambridge University Press.
Tolman, C. R. (1979), The Principles of Statistical Mechanics. New York: Dover.
Zeh, H. (1989), The Physical Basis of the Direction of Time. New York: Springer-Verlag.

Potrebbero piacerti anche