Download as pdf or txt
Download as pdf or txt
You are on page 1of 27

Foundations of Physics, VoL 10, Nos.

5/6, 1980

On the Reconstruction of the Macroobservable 1

Olaf Melsheimer ~

Received August 6, 1979

The physical implications of the embe&h'ng method recently proposed as the


basic concept for a formulation of the general relationship between classical
and quantum mechanical modes of description of macroscopic systems are
further examined, A conceptual analysis of the idea of a macroobservable is
given. A very general assumption on the macroobservables, called the thermo-
dynamic uncertainty principle, leads then to a proposal for a formal recon-
struction prescription for the macroobservables. The implications of this
reconstruction methodJbr the formal structure of quantum statistical mechanics
are discussed.

1. INTRODUCTION

The overwhelming majority of all attempts to provide a foundation of


statistical mechanics is based on the tacit assumption that many-body
quantum mechanics or, in certain cases, also many-body classical mechanics,
is the correct physical theory even for the description of systems with very
many degrees of freedom. These theories have to be extended by certain
approximation or averaging procedures in order to make them applicable
to a description of many-body systems. It is understood that these procedures
simply reflect the fact that in dealing with systems consisting of about 1020
subsystems each, one is not interested in the detailed behavior of such
systems. Rather, one has to be satisfied with a so-called macroscopic descrip-
tion which necessarily bears the blemish of supplying only a restricted gain
of knowledge of the systems under consideration. This point of view is taken
by Cercignani, who writes explicitly(l>: "The basic idea of statistical mechanics
is that of averaging over our ignorance (meaning the incapability of macro-

1 Supported by a grant from the Deutsche Forschungsgemeinschaft,


2 Fachbereich Physik der Phitipps-Universit~it Marburg, Arbeitsgruppe "Grundlagen der
Physik," MarburgfL., West Germany.

375
0015-901818010600-0375503.00/0 © 1980 Plenum Publishing Corporation
376 Melsheimer

scopic bodies to detect certain microscopic details of another macroscopic


body) and treating our system in a statistical sense." Sometimes this point
of view is manifested in the definition of the so-called information entropy,
which is interpreted as a measure of our ignorance on the system. In spite
of the numerous and important results that have been obtained within the
field of statistical mechanics, one runs into difficulties when trying to
formulate rigorously one a n s a t z or another for a proper foundation of
statistical mechanics. Regarding the "macroscopic" description merely as an
approximation procedure where one restricts himself to a restricted number
of so-called macroscopically relevant observables, one usually cannot cope
with the problem of irreversibility. When treating statistical mechanics as an
averaging procedure one encounters the following problem: in case of a
macroscopic statistical theory (e.g., the Brownian motion) one has no
evident compatibility between the statistics of the macroscopic theory and
the corresponding quantum statistics. For a far more detailed explanation
of these problems we refer to Ludwig. (2)
Going back to the origins of statistical mechanics, one gets the impression
that already the founder of statistical mechanics, Ludwig Boltzmann, was
opting for a conception which differs from what many of his followers
believe. The point of view that many-body quantum mechanics (or classical
mechanics) is the correct physical theory even for the description of macro-
systems would only be justified if one really could ascribe a physical reality
to the microscopic entities the macroscopic body is believed to be composed
of. However, this kind of physical reality was never claimed by Boltzmann.(8)
Instead, he treated the microscopic bits as being of hypothetical nature. He
even called them "Vorstellungseinheiten" (entities of imagination).
With all the new physical esperience at hand, Ludwig(~) again put forth
a criticism of the frequent point of view according to which quantum
mechanics (or classical mechanics) is the correct theory even for the descrip-
tion of the behavior of macroscopic systems. He starts with the observation
that the mere attempt to correlate many-body quantum mechanics with a
certain domain of its applicability goes beyond the limits of physical
experience. This argument alone--without any modification---would equally
well apply to many-body classical mechanics. Certain attempts to provide
quantum mechanics with a proper physical interpretation cast doubts upon
the idea that quantum mechanics is after all the appropriate physical theory
for the description of systems with many degrees of freedom. As follows
from Ludwig's(5) axiomatic approach to quantum mechanics, microsystems
can at best be regarded a some kind of a "universal" interaction between
macroscopic apparatus. There is no possibility to ascribe objective properties
to microsystems in the sense that these proporties would be independent of
the way in which these objects are physically prepared or measured. With
On the Reconstruction of the Macroobservable 377

respect to its formal structure, quantum mechanics can be regarded as a


formalism which in a way emerges from classical physics, namely from the
description of statistical laws for the occurrence of effects that are triggered
on macroscopic apparatus by so-called microchannels. The inerpretation of
this formalism is therefore inseparably linked to the existence of a theoretical
mode of description for macroscopic systems (apparatus). For a very clear-cut
explanation of these relationships we refer to the paper of Lanz. (m
Within this context it does not seem to be very likely that an extension
of quantum mechanics to a large number of degrees of freedom should be
the correct physical theory for the description of macrosystems. The fact
that there is no obvious restriction as to the number of degrees of freedom
for quantum mechanics should not be misinterpreted in the sense that there
cannot exist at all an extended theoretical framework by which such a
restriction could be provided. The unsatisfactory situation with respect to a
proper foundation of statistical mechanics seems to indicate the need for the
existence of such an extended theoretical scheme, which in a way would
have to contain as theoretical substructures not only quantum mechanics
(for microsystems with very few degrees of freedom), but also a theoretical
frame work for the description of macrosystems. The existence of such a
supertheory already presupposes that theoretical settings for the description
of macrosystems have a right of existence in their own.
There is one particular aspect of such theoretical settings that seems
to be of paramount importance whatever particular formalism for the
classical theory might be chosen. They must express the objective character
of reality ascribed to macroscopic systems, which, in classical theories, is
customarily embodied in the concept of a state of such a system. Classical
theories deal with so-called state parameters, which are assumed to yield a
(complete) description of a state of a macrosystem. The controversial question
of a proper foundation of statistical mechanics may therefore in principle be
reduced to the problem of how one can justify the existence of state para-
meters within the context of a many-body theory--either quantum mechanics
or classical mechanics. It is essentially this question that has given rise to a
dispute between Boltzmann and his opponents (Ref. 6, p. 17). One may
very strongly surmise that a many-body theory alone is not at all able to
account for the state concept. Rather, the hypothetical supertheory should
provide the rules with the aid of which the state concept within many-body
theories can emerge. More precisely, these rules must lead to an explanation
of how it can happen that with an increase of the number of degrees of
freedom new state parameters emerge in addition to the parameters that are
given by the superselection rules in ordinary quantum mechanics.
Since such a supertheory even in the weakest form is not very likely
to be constructed in the near future, the question naturally arises whether
378 Melsheimer

the hypothetical existence of such a theory is of any help for the purpose
of providing a rigorous foundation of statistical mechanics. Although we
think that the development of such a supertheory would be the proper way
to develop a foundation for statistical mechanics, we believe very strongly
that at present one should be satisfied with the solution of a far more modest
problem, namely, in which form classical physics can be set in correspondence
with many-body quantum mechanics. We would like to make the following
point well understood:
The idea of a supertheory encompassing, on the one hand, quantum
theory as a theory of specific "microchannet-type" interactions between
macroscopic apparatus and, on the other hand, also the macrotheory as a
theory of interactions between macroscopic systems of course just represents
a rudimentary idea within the framework of our fundamental conception of
statistical physics; in this conception, the meaning of the quantum mechanical
model and its representativity for statistical physics is essentially relativized,
though--now only in a restricted sense--it still is heuristically accepted as a
useful formal starting point for a new development of physically plausible
concepts and structures that possibly could prove more appropriate for the
description of macroscopic systems from a somewhat more realistic point
of view.
The knowledge of the "supertheory" is, of course, not needed for this
passage to new methods in statistical physics; as an idea in the basic scheme
of our approach, it serves naturally for the motivation of such research.
It is just our goal to select within the formalism of many-body quantum
mechanics a particular substructure which is appropriate for the description
of macrosystems. This is essentially done on the basis of the embedding
method as explained in Ref. 7. In this way we bypass the task of founding a
unified theory on the basis of the microstructure of macrosystems. Rather,
a method is proposed for incorporating the state parameter concept, which
strictly speaking is alien to the formalism of quantum mechanics, into this
very formalism.
This is done with the help of the so-called objectifying function (cf.
Section 3.2.4) (5, whose existence is motivated on the basis of a conceptual
analysis of the structure of the macroobservable. As has been shown in
Ref. 7, the regress to the so-called pretheoretical level either for quantum
mechanics or for a general framework for classical theories indicates the
existence of a compatibility relation between both theoretical schemes. It is
essentially" a further investigation of the implications of this compatibility
relation that we are aiming at in the first part of this paper. The third part
is then devoted to the problem of a reconstruction of this compatibility
condition starting from a set of state parameters for a particular type of
macrosystem and a many-body formalism which, on the basis of an atomic
On the Reconstruction of the Macroobservable 379

conception of the structure of the macrosystem, is believed to be put in


correspondence with the chosen set of state parameters. This task is essentially
what we shall call the reconstruction of the macroobservable.
We conclude this section with a few remarks concerning the general
layout of this paper. Since the main aim of this paper is a better under-
standing of the relationship between quantum mechanics and classical
physics, we have tried to stick to a rigorous mathematical setting and it is to
be hoped that this setting accounts for the main ingredients of the physical
situation. It is thereby assumed that a C*-algebra formalism is the proper
theoretical framework for quantum mechanics with practically infinitely
many degrees of freedom, in particular, that this is also the appropriate
theoretical setting for the description of systems far from equilibrium.

2. M O R E ON THE CLASSICAL C O N T E N T OF MANY-BODY


QUANTUM MECHANICS

In this section we mainly discuss the implications of results obtained


in Ref. 7, which made a global characterization of the relationship between
quantum mechanics and classical physics possible. It turned out essentially
that only certain domains of the quantum mechanical sets of observables
and ensembles, respectively, are needed for a quantum mechanical description
of macrosystems. We called this part of the quantum mechanical formalism
the classical content.

2.1. A Further Aspect of the Embedding Concept

In this section we deal (from a somewhat different point of view) with


the arguments employed in Ref. 7 for a justification of the a n s a t z for
establishing the relationship between the quantum mechanical and classical
modes of description of macrosystems. The embedding concept as discussed
in Ref. 7 rests on the hypothesis that both microphysics and macrophysics
are two domains of physical experience each of which has its own pecularities
requiring a specific form for its theoretical handling. The widespread belief
that quantum mechanics extended to a very large number of degrees of
freedom should after all provide a possibility for the theoretical description
of macrosystems is conjectured--roughly speaking--partly on the basis of
the hypothesis (we share the conception of Boltzmann C2~on the hypothetical
nature of the atomic bits) that all macrosystems are composed of microscopic
bits and partly on the fact that quantum mechanics does not contain any
restriction as to the number of degrees of freedom.

8zslio15/6-2
380 Melsheimer

In any case there remains the problem of how to put into correspondence
the quantities of many-body quantum mechanics with physical reality. This
problem should not be mixed up with the more technical problem of how to
make many-body physics usable for the theoretical treatment of macro-
systems. It concerns mainly the question of the form of statistical mechanics
that can be made a proper physical theory, which requires essentially the
setting up of a precise correspondence between quantities of the formalism
and certain quantities of physical reality.
It was this problem in the case of quantum mechanics itself that
motivated Einstein, Rosen, and Podolsky in presenting their famous
paradox) s) The realistic ansatz of Ludwig(4) within his axiomatic approach
to quantum mechanics has led to an interpretation of the statistical operators
(density matrices) as classes of sets of microobjects which might have empty
intersection and notwithstanding have to be related to one and the same
density matrix. The single set of such classes corresponds to appropriate
preparing or registering apparatus, which are assumed to prepare or register
the microobjects. Since the elements of each set are such as if they were
selected by such an apparatus, one terms such a set a selection procedure.
The statistical operators therefore no longer contain the information on the
actual process of the preparation, which is, however, urgently needed in
order to avoid the paradoxical feature of the ERP paradox.
It turns out that quantum mechanics can only be regarded to be a
formalism for the description of microsystems being the form of a certain
universal interaction between macroscopic systems (apparatus). The statistics
of quantum mechanics is somehow a direct reflection of this universality of
the microobjects. This interpretation of quantum mechanics is, however, not
at all adequate for the theoretical treatment of classical objects. It seems to
contradict common sense to consider an ensemble of macrosystems as a
class of sets of macrosystems whose elements, on the one hand, cannot be
distinguished by any physical method whatsoever and, on the other hand,
have to be regarded as different physical objects. The partitioning of physical
objects into classes of such objects, which is a pecularity of quantum
mechanics, seems to be alien to any theory of macroobjects. In order to
investigate the "physical" relevance of many-body quantum mechanics, it
seems therefore to be best to regress to the original selection procedures,
which in case of quantum mechanics give rise to either statistical operators
or simple observables. The fact that, among these selection procedures, there
occur also the real selection procedures of macrosystems, i.e., selection
procedures that can really be carried through for macrosystems, was the
basis of the embedding concept as discussed in Ref. 7.
On the Reconstruction of the Macroobservable 381

2.2. The Thermodynamic Limit

In this section we discuss in more detail the particular form of many-


body quantum mechanics we have used within the embedding concept. We
have adopted the point of view that the thermodynamic limit of many-body
quantum mechanics can suitably be described by a C*-algebra ~/acted upon
by a one-parameter group of automorphisms ~(t) representing the time
evolution. We shall refrain here from any attempt to justify this formalism
as the appropriate one that,enables one also to describe systems far away
from equilibrium. From our point of view the C*-algebra formalism is only
a proper idealization of the many-body formalism of quantum mechanics
that has the advantage of allowing a better representation of certain
characteristic features of a "macroscopic" description of classical systems.
However, we want to emphasize that from our point of view the C*-algebra
formalism is not yet the macroscopic level for the quantum mechanical
description of macrosystems. Rather, it has to be regarded as a suitable
mathematical formalism in which the macroscopic level has to be embedded.
We shall note a few more points in which we differ from the ordinary inter-
pretation of the C*-algebra d . We distinguished the set L containing all
positive Hermitian elements of s l whose norm falls into the interval [0, 1].
Following the above-mentioned idealization, this set was called the set of
quantum mechanical effects. ]-he elements of L must not be mixed up with
bounded obselwables. Rather, they are interpreted as generalized yes-no
measurements. For the relevance of these effects for the interpretation of
quantum mechanics we refer to Ref. 5. It will turn out in the sequel that in
particular the generalized yes-no measurements are of paramount im-
portance for a suitable description of the so-called trajectory registrations.
We have furhermore assumed that the set of states is not represented
by the whole of the positive part 5Y of the unit ball of d * , the topological
dual of ~s¢. Rather, a certain sublet 5~.p~,has to be chosen which has to fulfil
certain conditions given in Ref. 7. With regard to the particular importance
of ,~,~ for the derivation of the structure of the macroobservable, t7) it seems
necessary to discuss a bit more the physical relevance of the elements of 5P~,1o.
As physical systems with infinitely many degrees of freedom are beiieved
to be describable by the triplet ( ~ , 5Pvh, c~(t)) and this is a purely algebraic
formalism, there is no immediate way to apply this formalism to particular
physical situations. It is necessary to choose a specific representation of c~/
by an algebra of operators on a Hilbert space. As is well known, this can
easily be done by the aid of an element of 5~h in the framework of the
well-known GNS representation of .~¢. However, it turns out that not all
GNS representations of ~/ are physically equivalent. Certain unitarily
nonequivalent representations of d may, in fact, characterize different
382 Melsheimer

physical situations. <9 This situation is, of course, completely different from
the case of ordinary quantum mechanics for systems with finitely many
degrees of freedom. For the formalism ( d , 5P~7~, ~(t)), the elements of 5°~h
attain a new role in comparison with ordinary quantum mechanics. They
not only describe the statistics for the outcome of experiments, but also help
to shape the formalism itself. Unfortunately, there is no generally accepted
method for the selection of .5°~n fi'om 5e. A decisive but yet unknown
criterion for the choice of Y ~ is that the universal physical representation
~40 of ~¢ as described in Ref. 7 is rich enough to comprehend all essential
physical features of the macrosystems under consideration.

2.3. The Dynamical Laws

Besides the role of shaping the formalism, the set 5e~, has yet another
function within the macroscopic mode of description. Certain elements of
5~,,j~ are needed to fix the "dynamical laws" on the macroscopic level of
description. In order to explain what we precisely mean by "dynamical laws,"
we shall discuss in some detail the particular aspect of the framework for
classical theories which pertains to dynamics as developed in Ref. 10.
As already explained in Ref. 7, the basic structure term for a general
framework theory for classical systems are the spaces C(Y), the space of all
continuous functions over the trajectory space ~, and its topological dual
space C'(Y). The C(Y) contains as a subset the set L(Y) = {f(y) ]f(y) E C(Y),
0 <~f(y) -.<.1}, the set of the so-called trajectory effects, whereas C'(Y)
contains the set K~, i.e., the set of the macroensembles. If we then denote by
tz,,~(u, f(y)) with u ~ C'(Y) and f ( y ) ~ C(V) the bilinear form for the duaI
pair (C(Y), C'(Y)), the expression Ix,~(u, f(y)) with u ~ K~ and f(y) c L(Y)
gives approximately the experimental frequency for the registering of
trajectories of single systems contained within the ensemble u ~ K~ by means
of an apparatus described by the function f ( y ) e L(Y). In this way, the
classical framework theory has in principle a statistical form. The following
question naturally arises: how can one characterize the dynamical laws
within this kind of formalism ? Within the context of this framework theory,
we shall call "dynamical laws" all those possible statements that allow
predictions for the occurrence of ceI~tain trajectories. Within the context of
the theory sketched above this can only be done in terms of statistical
predictions. A proper analysis then shows (1°) that the dynamical laws may be
formulated entirely in terms of the extremal points of K~ if K~ happens to be
a compact set, or else in terms of so-called almost extremal points.
We want to establish the relationship between the dynamical laws and
certain elements of the set 5£~p~. From the naive point of view, one could
O n the Reconstruction of the Macroobservable 383

think that question of dynamics should be treated entirely in terms of the


so-called quantum mechanical time evolution

A(t) = V t A = ei/nm Ae -i/n~

with H the Hamiltonian of the system and A an observable, However, as


Ludwig has shown within his axiomatic approach to quantum mechanics,(4.5)
the transformation V, may be interpreted as a purely kinematical one. In fact,
i f F i s an effect standing for a registering apparatus which is open from t = 0
on with respect to the preparing apparatus, the effect V~F stands simply for
a registering apparatus that is open from t = ~- on. This is the correct
kinematical interpretation of the quantum mechanical time evolution, which,
does not attain dynamical aspects until there some quantum mechanical
ensembles are given. As has been pointed out in Ref. 7, within the formalism
of the classical framework theory there is also a kinematical time-translation
transformation which very closely resembles the quantum mechanical time
evolution transformation. In fact, if we define a time shifting transformation
T~- on trajectories, namely T , y ( t ) = y(t + "r)with z ( t ) = y E Y, then T~
induces on C(Y) an operator W , f ( y ) = f ( T , y ) . As has been shown by
Ludwig, ~°) W~ may be interpreted in close relation to V~: W , f ( y ) turns
out to be a registering procedure which registers trajectories only from
t--Ton.
If now the macroscopic description (within the context of the classical
framework theory) and the many-body quantum mechanical description are
compatible, which we justified by the regress to the so-called pretheoretical
level of the selection procedures, (v) we are led to the following compatibility
condition:
R: L(Y)--~L C d
S: ~ -+ 5P~h C ~/*

where R and S are (linear) mappings such that i~(u, W,f(y)) = tz(Su, ~ R f ( y )
holds for all u ~ K ~ , f ( y ) ~ L(Y), and T ~> 0. Here, /x denotes the bilinear
form for the dual pair (sS, d * } . To indicate the transition from the ordinary
quantum mechanical formalism to the algebraic one we have denoted the
time translation transformation by c~, instead of V~.
The above embedding condition states that all statistical predictions
should be equally well describable either by the formalism of the classical
theory or the formalism of quantum mechanics. In a way this condition
already expresses a compatibility of the dynamics. The intimate relationship
between the dynamical laws and the set 5P~ can, however, be made more
transparent.
384 Melsheimer

The image of K,~ under the mapping S may be termed the set of macro-
scopic ensembles.
Taking from Ref. 7 the N(Y) for the class of Borel sets of the trajectory
space Y, we get with

F: ~(v) -~ (RL(V)) ~ = ~/

the macroobservable, a mapping F': SK~--7 K,,~ defined by the relation


F'(u) =/z(u, F( - )) for all u ~ SK~,~. The dynamical laws can then be identified
in terms of the extremal or almost extremal points of F'SKm.

2.4. The Quantum Mechanical Trajectory Registration

tn order to avoid misinterpretations of the embedding concept, we


shall try to clarify some strange features of the embedding or more specifically,
the macroobservabte concept. The existence of the macroobservable
F: ~ ( Y ) --- ~ " does not mean that in the proper quantum mechanical context
there is an apparatus that measures together the effects F(B) with B ~ N(Y).
Trajectories y with y ~ V are measured. The macroobservable merely
describes the statistical outcome of a classical trajectory registration within
the quantum mechanical formalism. The precise meaning of this statement
is mathematically expressed by the above embedding concept.
Notwithstanding, the question arises whether the registration of a whole
trajectory is completely alien to the formalism of quantum mechanics.
From the naive quantum mechanical point of view, the registration of a
trajectory seems to be inadmissible. One is accustomed to regard the measure-
ment of a time-dependent observable A(t)as a measurement of this observable
at each instant of time. A close inspection of this statement, however, reveals
that this is in fact a very strong idealization, since any measurement--from
the point of view of the quantum mechanical process of measurement--takes
a certain amount of time. But, all the more, there are aiready within the
context of ordinary quantum mechanics physical situations where one in a
way has to rely on the concept of a "trajectory" registration in order to avoid
paradoxical features. We have put the word trajectory in quotation marks
in order to indicate that in quantum mechanics there is, of course, no
trajectory in the classical sense, We shall sketch such a situation. It is well
known that in general, the three components of the position observable for
various instants of time are not simultaneously measurable, since they do not
build in general a set of commuting operators. This would mean that the
"trajectory" of a particle is not measurable. It seems that such a measurement
is forbidden on the very basis of quantum mechanics. On the other hand,
there are measuring apparatus, i.e., a Wilson chamber, that measure, at least
On the Reconstruction of the Maexoobservable 385

in an approximate way, the "trajectory" of a particle. As Ludwig has pointed


out, (5) one can very easily get rid of this paradoxical feature by sticking to
the more general conception of yes-no measurements, namely, the use of
the effect operators represented by positive Hermetian operators F with
0 ~ F -~ 1. One can in fact specify a set of such generalized yes-no measure-
ments that measure whether a particle has passed through a certain space-
time region. It turns out that the corresponding operators are necessarily
generalized effect operators, i.e., not projection operators!

2.$. The Non-Coexistence of Trajectory Effects

In this section we mention another aspect of the so-called classical


content. According to the results of Ref. 7, the classical content ~//t of tile
universal physical representation d 0 of the C*-algebra d , which was defined
as d/l : (RL(Y))% namely as the e ( d 0 , [5~h])-closure of the image of all
trajectory effects, can be represented as all" : (con R(F))% the e-closure of
the convex range of the measure F that represents the macroobservable.
Since (RL(Y)) ~ is merely a topological extension of the set RL(Y) and as the
latter set is a set of effects that stands for classical registering procedures,
one could surmise that all elements of (RL(Y)) ~ are in fact commensurable
effects. According to the deep analysis of tile commensurability concept of
quantum mechanics within the axiomatic approach of quantum mechanics
of Ludwig, (5) this does not necessarily mean that all elements o f / / f are
commuting operators. It would be sufficient if all elements of d f were
so-called coexistent effects. (5) Certainly, the range R(F) of the measure F is
by the very definition a set of coexistent effects. One may ask whether this
property survives if one goes over to (con R(F))% As we have already
mentioned in Ref. 7, one can in fact always find an extended measure fi~ that
means an extension of the measure F to a Borel ring N containing N as a
subring such that R(_#) : (con R)F)) ~ holds. At first glance, this means that
J / i s a set of coexistent effects. However, one has to guarantee that/~ is a
macroobservable, which means that it stands for the registration of trajec-
tories. According to the results of Ref. 7, this can only be fulfilled if/~ were
in fact the macroobservable for a classical registration of "finer" state
parameters, which need not necessarily be the case. Hence, we arrive at the
somewhat surprising conclusion that the classical content is in general
neither a set of commuting operators nor a set of coexistent effects in the
sense of Ludwig. (4)

2.6. The Maeroensembles

Here we discuss within the context of the embedding concept the question
of the so-called macroensembles, which was already mentioned in Section 2.3,
386 Melsheimer

where we explained the relationship of the so-called dynamical laws and the
set SK~ called the set of all macroensembles. Since the set SK,~ is an essential
part of the embedding structure, the question arises whether this set may be
given a better characterization within the set of all quantum mechanical
ensembles. One may at first assume that SK~ or its weak closure forms a
simplex, (12)which accounts for the fact that, for macroensembles, there do not
exist so-called incompatible demixtures, m in contrast to pure quantum
mechanical ensembles. On the other hand, as the set SK~ determines the
dynamics of the classical theory--as we pointed out in Section 2.3--it turns
out to be of paramount importance to know the structure of the set o f
extremal points of (SKIn)% If, for instance, the set of extremat points is a set
of Dirac measures over the trajectory space ~g, the dynamics of the corre-
sponding classical theory is of purely deterministic character. The precise
knowledge of the set SK~, however, does not seem to be necessary to
reestablish the embedding structure, a problem which we shall deal with
in the next section. Once the macroobservable F is known by means of the
mapping F' (cf. Section 2.3), one can reestablish the relationship between
SK~, which has, of course, to be chosen in an appropriate way, and the
s e t / ~ . All the more, one may assume that the set SKm may be determined
in " a first approximation" completely in terms of the macroobservables.
Since the elements of K,~ stand for selection procedures of macrosystems and
these selection procedures are given by preparing procedures that take place
at a certain instant of time (i.e., t -----0), one may start from the assumption
that SK~ consists of all those ensembles that arise in registering state para-
meters within a vicinity of t = 0. We shall treat this problem in more detail
in another paper.

2.7. The Recovery of the Embedding Structure


In this section we discuss very briefly in which way the embedding
structure may serve for a proper foundation of statistical mechanics.
F r o m our point of view, the methods of statistical mechanics would be
justified if one really could reestablish the embedding structure that according
to the intimate relationship of classical and quantum mechanical modes of
description on the pretheoretical level (7) should govern the relationship
between classical phenomenological theories and many-body quantum
mechanics. The first step of a possible strategy is then as foliows: Given a
C*-algebra d , a set of states 5 ° over d , and a time-translation transformation
a~, which we assume to describe in a proper way the microscopic structure
of the macrosystem under consideration, we first look for a subset 50~ of 50
and with the help of 50~h we construct the algebra ~40, the universal physical
representation of d . We then face the difficult task of choosing a suitable
On the Reconstruction of the Macroobservable 387

subset d l of ~/0, the so-called classical content, and a subset SI~, C S~h
characterizing the dynamical laws of our classical theory.
In order to retrieve the embedding structure, one has then to find a
vector measure F over the Borel sets of the trajectory space Y with values
in J g such that J g is the closed convex range of the measure/:. A further task
is the construction of a projection operator P onto de' such that P a ' P = ~,
(for the meaning of ~ we refer to Ref. 7) is a semigroup for ~- >~ 0 and &,
is compatible with the classical time-shift transformation semigroup.
In what follows, we shall mainly concentrate on the reconstruction of
the macroobservable. We shall present a rigorous mathematical method
by the aid of which the macroobservable may, in principle, be fully recon-
structed.
We emphasize once more that no answer is given to the question of
which kind of state parameter can classically be measured on a macrosystem.
According to our point of view, the set of state parameters forming the
essential part of a phenomenological theory has to be determined on the
basis of a certain physical foreknowledge.

3. O N THE R E C O N S T R U C T I O N OF THE MACROOBSERVABLE

3.1. A New Mathematical Characterization of the Classical Content

= (RL(•)) ..... (con R(F))

Here we give a characterization of the set J/f which does not refer to
the trajectory space Y. This will essentially be done in close connection with
the results of Ref. 12. F o r this purpose, it is necessary to pass over from the
complex vector spaces to the "physical" real vector spaces. We shatt base
our way of proceeding on the results of Ref. 13.
According to Ref. 7, the yon Nemnann algebra ~ o is given as the dual
space of the space [5"~n], the linear space of the set of physically relevant
states. On the other hand, J 0 may be identified with the direct sum of atl
GNS representation of the C*-algebra s~' with respect to elements of J,n.
Therefore d o is given as a v o n Neumann algebra of operators acting on the
Hilbert space

~e.Y~h

Let e be the unit element of d 0 , which is assumed to be the unit element of


the whole of ~(ffi). The predual space d 0 , of d 0 is [SP~n]. In our case,
[ ~ , ] is a set of positive Hermitian operators in N. The elements cp ~ 5P~e,
388 Melsheimer

satisfy the condition /z@, e) = 1. Then [5e~i,] is the cone of all positive
normal linear functionals on d o . The pair ([5°~], Se~n) is a so-called
"complete base n o r m space. ''(la
Following the notation of Ludwig, (5,14 we denote the space ([Seth], Sash)
as B. The real dual space B ' of B is then a "complete order unit space. 'mS)
The space B' m a y be identified with the set of all Hermitian operators of
d 0 . The spaces B, B ' are considered to be the physically relevant spaces,
since all the statistical laws and predictions can be formulated entirely in
terms of elements of these spaces. Henceforth, we shall use these spaces
instead of the v o n N e u m a n n algebra e~'0 and the set of states '~Dh •
Let us consider now the vector measure

F: ~)(V) --+ J4' C B'

where ~ ( Y ) is the Borel ring over Y. The vector measure F is assumed to


describe a macroobservable as derived in Ref. 7.
Via duality, the measure F defines a linear mapping F', namely

a' ~ i~(a ', F ( . ) = F'(a')

from B into the set of finite, a-additive, real-valued measures over ~ ( Y )


with the norm 1l/~ l] -----I/z [(Y), where 1/z ] denotes the variation of/L. We
shall denote by ca(N'(Y)) the set of all finite, real-valued a-additive measures
over ~ ( Y ) . We therefore have

F' : B --+ c a ( ~ ( Y ) )

F ' may be extended to the whole of B", the topological dual space of B'.
All the more, F ' may even be extended to the vector lattice h(B') generated
by B". The h(B') is defined as the smallest vector lattice (with respect to
pointwise ordering and linear operations) that contains the function spaceB"
(with a" ~ B", a"(a'), a' ~ B', is the value of the function a" for the point
a' ~ B'). Each element x" of h(B') may be represented in the form

x" = V a"~ - a"i with a i" ~ B", i = 1..... e


i=l i=k+l

The extension of F ' to the whole of h(B') is given by

tt H
F ' ( x " ) = V ai ° F - ai o F
I=i i=l~+l
On the Reconstruction of the Macroobservable 389

where the lattice operations on the right-hand side are those of ca(N(Y)). ~5)
Let us now consider the following expression:

u~(x") ..... F ' ( x " ) ( V )

which is defined for all x" E h(B'). Obviously, u e ( ) is linear in x" ~ h(B').
Furthermore, for x" ~ h(B') and x" >~ 0 we have

u~(x") = F ' ( x " ) ( v ) >~ o

The mathematical object uv( • ) is called a conical measure (Ref. 16, Vol. I).
More generally, one calls the set M+(X) of all nonnegative linear
functionals on h(X), where X is a topological vector space, the set of conical
measures over X. The set M+(X) is a relatively complete lattice with respect
to the order v ~< u defined by v(x') ~ u(x') for all x' ~ h(X) with x' >~ O.
For u ~ M+(X) and x ~ X such that u(x') = x'(x) for all x' ~ X', one writes
x = r(u) and calls x = r(u) the resultant of y. Whenever x = r(u) exists,
it is uniquely determined.
Let now u ~ M+(X) be a fixed element of M+(X). If there exists for each
element v s M+(X) with v ~ u the resultant r(v) we group together all these
elements into the set

K~ = {r(v) i v ~ u, v ~ M+(X)}

K~ has the following properties (Ref. 16, VoI. III).

i. K~ is convex and contains the null dement.


2. For u, v E M+(X) it follows that K~+~ - K~. + K~.
3. For 2t >~ 0 and u ~ M+(X) one has K ~ = ?ur(~.

Convex sets of the type K~ are also called zonoforms (Ref. 16, Vol. tli).
With the help of the conical measure ue ~ M+(B ') one arrives at a character-
ization of the set J i which is, in fact, independent of the trajectory space Y.

Proposition L The relation K~. .......J2/holds.


For the p r o o f we simply observe that B' is a quasicomplete topological
vector space with respect to the topology cr(B', B). The assertion follows then
by Theorem VII.2.1. of Ref. 12. We remark in passing that ~ f is a ~r(B', B)-
compact set.
On the other hand, our final goal to represent a given ~-closed convex
set .AI C B' as the closed, convex range of a vector measure is formally
achieved by the following:
390 Melsheimer

Proposition 2. For K~ C B' C d 0 with u ~ M + ( B ') there exist a set T,


a a-algebra 50 of subsets of T, and a vector measure F: 5: -+ B' such that
U = H F .

The proof of this proposition is obtained at once by using Theorem


VII.2.2 of Ref. 12 if one again pays attention to the fact that B' is a quasi-
complete topological vector space.
Proposition 2 is, however, of little help for the solution of our problem,
since we are looking for a vector measure over a given trajectory space Y
such that the closed, convex range of this measure equals ~ ' . in fact, there
are in general many such representations of a given J / = K~ as a closed,
convex range of a vector measure. We really need an additional citerion
which enables us to select among all these various representations the
physically meaningful one. This will essentially be the subject of the next
section.

3.2. The Reconstruction of the Macroobservable

3.2.1. Formulation of the Reconstruction Problems. Here we formulate


in more detail the reconstruction Problem of the macroobservable already
mentioned at the end of Section 2.7.
We may now sketch the problem in the following way: Let us take a
C*-algebra formalism ( ~ , 5:~h, a,), which is believed to describe the
quantum mechanical structure of the macrosystem under consideration.
The triplet ( d , 5:~h, c~,) consists of a C*-algebra ~ , a set of states over d ,
and a time-translation group ~,. Following the spirit of Boltzmann's work/~)
we assume here that ( d , ~ , c~,) represents the quantum mechanical
many-body formalism for the microscopic bits which in the proper sense
are assumed even to be of hypothetical existence.
In a rough sense, the general task is to select those observables and
ensembles that are sufficient for a reproduction of the classical phenome-
nological theory for the macrosystems under consideration. We assume here
that this particular phenomenologicat theory has been given the form of
the general framework theory treated in Ref. 7, i.e., there is a uniform
(precompac0 trajectory space Y, a set of Boret measures K,,,~over Y, and a
semigroup T~, 7 ~> 0, of time shifting transformations on trajectories, which
means

T¢y = T , z ( t ) = z(t -? ,r) == y ' for all ¢ ~> 0 and all y ~ Y.

With the results of Ref. 7, we may proceed then as foIlows: Let -NO be
a universal physical representation of the C*-algebra J . Find a closed,
On the Reconstruction of the Macroobservable 391

convex set -#I C L C B' such that J/l" represents the classical content of the
given classical phenomenological theory. This means that one has to construct
an operator-valued measure over the Borel sets N(Y) of the trajectory space
Y with values in L such that

F: ~ ( Y ) -~ J {

and
(con R(F)) ~ = #/l

holds. Furthermore, one has to construct a projection mapping P onto JCl


such that ~, = Polyp is at least a semigroup for all z >/0. For the explanation
of the meaning of P and c~, we refer to Ref. 7. ~, has to be compatible
with the transformation T~ in the following way:

o~TF(B) = F(T~IB) for ~- ~> 0 and all B ~ B ~ N'(Y)

Last but not least, one has to select a subset of states SK,~ C 5P~h such that

holds. In what follows, we shall pay attention essentially to the following


problem: Starting fi'om a classical content J~, i.e., the set ~/d : - (con R(F)) °,
for a given macroobservable F we investigate what are the main structural
parts of the macroobservable that enable the reconstruction of the given
macroobservable. This kind of analysis will then lead us to a proposal for a
general reconstruction procedure for macroobservables.

3.2.2. The Thermodynamic Uncertainty Principle (TUP). Here we


formulate an assumption concerning a particular structure property of the
macroobservable which has already been treated by Ludwig.~2) Since in a way
is resembles the well-known uncertainty relation in ordinary quantum
mechanics, it is called the thermodynamic uncertainty principle. The precise
meaning of this statement will, however, be explained further below.
Let F: ~ ( g ) - - + L C B' denote a macroobservable for a given classical
theory, where ~ ( g ) denotes the Borel ring over the trajectory space Y. We
recall the fact ~7) that Y may be assumed to be a compact topological space
which is also metrizable. Each element y e Y has, therefore, a fundamental
system of neighborhoods of the form Bo(y ) with Bp(y) a closed ball around y
with radius p; p runs through the set of real numbers of the form 1In, n ~ N.
392 Melsheimer

Since B' is an order-unit space and at the same time the dual space of the
base norm space B, the order-unit norm on B' is given by (m

][FH = sup ]p,(u,f)[

The T U P may now be formulated as follows.

TUP. I f {Bo(y))} is a fundamental system of neighborhoods for any


element y ~ Y, then II F(Bo(Y)) 1[ tends to zero if p does so.
In what follows we present some facts that underline the possibilityof
this assumption, which, in fact, is of crucial importance for the reconstruction
procedure of the macroobservable we are going to describe in the next
sections. The T U P is related to the fact that for classical theories one has in
general the possibility of choosing so-called restrictions or extensions of the
state space. To be more precise, the nonvalidity of the T U P would lead to a
contradiction with the postulate that the embedding concept is compatible
with refinements or coarsenings of the state description of classical systems.
The mathematical structure resulting from extensions or restrictions of the
state space has been treated in detail in Section 4 of Ref. 7. If2~ is an extension
of the state space ~, we have the following embedding sequence (in accordance
with the compatibility postuiate mentioned above)

C(V) - L C(V~) ---,- L C B'

K~ --~ K,,~o-~ ~ h C B

where K,~ C C'(Y) and K% C C'(Y~) are the sets of macroensembles for the
two classical theories d¢'~- and J/e'%, respectively. In Ref. 7 it has been shown
that the corresponding macroobservables F and F, are related to each other
in the following way: F is just the measure-theoretic restriction of F~, which
means precisely that the Borel ring M(Y) may be identified with a certain
Borel subring M 0 of 2(Y~) and F is the restriction of the measure F~ to 2 0 .
Let K(Y) and K(Y~) denote the set of all normed regular Borel measure
on Y and Y~, respectively. Let us then consider the mapping F ' already
introduced in Section 2.3. We have
F~': 5:~7~--+ K(Y,) and F': .Sf~ --+ K(V)

Now, it is impossible that F'S:~h is equal to K(Y~) and at the same time
F'SP~h is equal to K(Y), or in a weaker form, that F~'5:~I~ is dense in K(Y~)
and in the same way F'5:,n is dense in K(Y) for the topologies a(C'(Y~),
C(Y~)) and ~r(C'(Y), C(Y)), respectively. In fact, if we denote by W' the
dual map of W, then we must have F~' = ~ ' o F' and this can in general not be
"onto."
On the Reconstruction of the Macroobservable 393

However at least the density condition must be fulfilled in order to


guarantee that II FXBe)II ~ 0 for all B, e N'(~'~) and [/F(B)I1 =~ 0 for all
~ ~(~').
A consequence of the TUP is that F'SK,,,(V) -~ K(V) must hold, which
means that not all the regular Borel measures over the trajectory space V
correspond to physical macroensembles. Within the framework of the
embedding structure, only a very limited set of measures K~ C K('~') is
physically realizable. It is this aspect that justifies calling the T U P an un-
certainty principle since, in a formal resemblance to the uncertainty principle
in ordinary quantum mechanics, it also implies a certain restriction as to
the set of physically preparable ensembles.
In what follows, we shall use a mathematically somewhat sharper
formulation of the TUP:

STUP. There exists a regular Borel measure ;~ on ~'(~/), not necessarily


normed, such that

if(u, F(B)) <~A(B) for all u ~ 6e~7~ and all B c ~(V)

holds. We remark in passing that the TUP is based on the fact that the
macroobservable is represented by a generalized operator-valued measure.
T h e TUP could not hold for a projector-valued measure. The same arguments
apply to the STUP.

3.2.3. The Localizability of the Conical Measure uF . The first


problem that one comes across when trying to solve the reconstruction
problem formulated in Section 3.2.1 is whether the underlying parameter
set V is uniquely determined, or more precisely, whether there could exist
another set ~/, a ~-sring ~(~/) of subsets of ~ , and anaeasure

such that o~/f = (con R(/~))%


It is well known (Ref. 12, p. 72) that the so-caUed closure of a given
measure F: N'(Y) --~ L C B' yields such a measure F. We shall very briefly
explain here the concept of the closure of the vector measure

F: ~(SY) -~ L C B'

Two elements /71, B2 ~ ~'(Y) are called F-equivalent if F(B)= 0 for all
B C B1 /~ B~, where B1 A B2 denotes the symmetrical difference between
Ba and Bz. The class of all F-equivalent sets B e N(Y) is again a Boolean
ring ~(F).
394 Melsheimer

One defines then a uniform structure (Ref. 18 p. 6) on ~ ( F ) in the


following way: if U ° denotes the polar set of a convex symmetric neighbor-
hood of zero in B ' with respect to the dual pair ( B ' , B " ) , (Ref. 18, p. 125),
one gets a pseudometric d~ on N'(F) by putting

d~(Ba , B2) = sup{r x " o F ( B 1 z~ B2) I; x " ~ U °}

for all B1, B2 e N'(F). The family of all such pseudometrics defines then a
uniform structure ~-(F) on M(F) and also on N'(Y).
The measure F is said to be closed if ~ ( F ) is a complete topological
space with respect to 7(F). It has been shown in Ref. 12, p. 72 that one can
always close a vector measure, which means that one can find a compact
space T D Y such that there exists a measure P: ~ ( T ) - - ~ L C B' with ~ ( T )
the Borel ring over T such that P is closed and (con R ( F ) ) ~ = con ( R ( P ) ) ~
holds.
In general, T will be an unphysical parameter space. This becomes clear
if one looks more closely at the actual closure procedure for a given vector
measure. The topological space T D V is obtained in taking ~ as the com-
pletion of 5Y(F) with respect to ~(F) and defining then T as the set of all
homomorphisms of the Boolean ring ~ into the Boolean ring {0, 1}, whereby
the maximal element o f ~ is mapped onto I. Since for any y ~ 3/the measure
3y can uniquely be extended to be such a homomorphism, 3 / c a n be identified
with a subset of T. Obviously

~ ( v ) = ~ n 3 / = {~ n 3/; ~ E &

The set T obtained in this way will, in general, be devoid of any physical
significance. The same reasoning applies to the set T occurring in Proposi-
tion 2 of Section 3.1, so that in fact this proposition helps very little to solve
the reconstruction problem.
But how then can the trajectory space 3 / b e characterized as the under-
lying parameter space for a vector measure that is intended to represent a
macroobservable for a given classical theory ?
We deem that a first step in obtaining an answer to this question can be
made by sticking to the localizability concept of a conical measure. A conical
measure u ~ M + ( B ') is called localizable if there exist a compact subset
M C B' and a R a d o n measure rn over M such that

u(z) = m ( z ( M ) ) for all z e h(B')

Proposition 3. I f a macroobservable F: N'(Y) --~ L C B' satisfies the


STUP, then the corresponding conical measure uF is localized on a compact
subset Jgo C Jg.
On the Reconstruction of the Maeroobservable 395

Let A be the regular Borel measure such that, according to the STUP,

u(u, F(B)) ~ A(B)

for all u ~ 5a, h and all B ~ ~ ( Y ) . First, we want to show that there exists a
A-integrable function g: Y --,- .A' (not necessarily uniquely defined) such that

F(B) = fB g(y) dA(y)

holds for all B ~ ~(~/).


For the p r o o f of this statement, we shall make use of a generalized
R a d o n - N y k o d y m theorem of Kupka ag) for vector-valued measures.
In order to fulfill the conditions of Theorem 4.9 of Ref. 19, we need
to have that {Y, N(Y), )t} is a so-called complete, decomposable measure
space such that F: ~ ( Y ) -+ L C B' is absolutely continuous with respect to 2t
and furthermore

{F(B)/A(B), B ~ ~ ( Y ) , A(B) # O} C d f

Since A is a regular Borel measure on a compact topological Hausdorff space,


the measure space {Y, ~(Y), A} has the required property. (19) The second
part of the condition is an immediate consequence of STUP. It can be seen
from the proof of Theorem 4.9 in Ref. 19 that g is not uniquely defined.
We put now

rig'0 = {g(y), Y ~ Y}~

which--as a closed subset of a compact set--also is a compact set. If we


denote now by ~(d/0) the Borel ring over d [ 0 , we may define a Borel measure
m on N'(J//0) by putting

m(E) = h(g-1(E)) for E ~ ~(Jgo)

Using then Lemma VII.3.1 of Ref. 12, we obtain

ur(z) -= fv z(g(y)) dh(y) = f-~/oz(b) dm(b)

for all z ~ h(B'), which is the assertion we need.


Now the question arises as to when a conical measure u e M+(B ") with
zonoform dg = K~ C L C B' can be localized on a subset Me'0C ~g. It is
answered partly by the following:

8251~o15/6-3
396 Melsheimer

Proposition 4. Let u ~ M+(B ') be a conical measure with ~g =


K~ C L C B'. I f there exists a closed subset dg 0 C ~g such that the restriction
of the functions z ~ h(B') to functions on J / o is an injective mapping, there
exists a positive R a d o n measure m c M+(J{o) such that

u(z) = f dt z(b) dm(b) for all z ~ h(B').


o

rn is unique.
For the p r o o f of this assertion, let us consider the set of functions

/ 4 = {z Idt0 ; z ~ h(B')}

which is a subset of the set of all real-valued continuous functions over dc 0 .


According to our assumption, h(B') is in this way represented as a set of
continuous functions over alto.
We want to apply Theorem 34.6 of Choquet (16) on H. For that purpose
we have only to show that H C C(•/o, IR) is an adapted space in the sense
of Choquet(16):

(i) We have H = H + -- H + since h(B') is a lattice of functions and


this property is preserved when passing to H.
(ii) F o r each x ~ - g o , there exists an f ~ H + such that f ( x ) > 0:
We have dd 0 C d / C B' and B' is the real dual vector space of 5~1,. Since
5°~7~ m a y be identified with a subset of H+, (7) and £f~ is separating on B',
one has the desired property.
(iii) Since J/Z0 is a compact set, the condition that every element of
H + is dominated by some element of H + is also fulfilled.

According to Theorem 34.6 of Choquet, there is then a positive R a d o n


measure m E M+(ddo) such that

u(z) = £go z(b) dm(b) for all z ~ h(B')

This measure rn is also unique: According to the lattice formulation of the


Stone-Weierstrass theorem (Ref. 16, Vol. I, p. 28) it follows that H is dense
in C ( ~ ' 0 , ~), which entails (Ref. 16, Vol. II, p. 278) the uniqueness of m.
Unfortunately, we have not found any easy way to characterize those
sets Jgo C ~ ' C L C B' for which Proposition 4 holds. For the purpose of the
reconstruction procedure of the macroobservable, it would be desirable to
know all those conical measures u ~ M+(B ') that may be localized on some
closed subset dd 0 C K ~ .
On ~hc Reconstruction of the Maeroobservable 397

However if for a given conical measure u ~ M~(B') with K~ --J/C,


~{o C ~//ZC L C B' is such that u is localized on d l 0 , one gets an immediate
representation of J / a s the closed, convex range of a vector measure

/~: ~ ( d f o ) --+ L C B'

One simply uses Theorem V!I.3.2 of Ref. 12 and obtains

F(E) = f~ e din(e)

for all E e N(dG). One has/7(E) c ~/1 for all E e ~ ( d / 0 ) and u = u~. Further-
more,/~ is automatically a closed vector measure, ~2) which has far-reaching
mathematical consequences. We remark once more in concluding this
section that the results obtained so far have to be regarded from the mathe
matical point of view as being only of preliminary character. The localizability
property of conical measures deserves a far more detailed investigation,
which will be carried through elsewhere.

3.2.4. The Objectifying Function ft. We have seen in the last section
that, for a conical measure u ~ M+(B ') with a suitable corresponding zono-
form K ~ , it is possible to represent K~ as the closed, convex range of a vector
measure
F: ~(~-G) -~ L C B'

where J/¢'o is a suitable closed subset of K ~ . The question now arises how a
vector measure
F: ~ ( v ) --. L c B '

may be obtained which has the same conical measure and represents the
desired marcoobservable. In the first part of the last section it became clear
that for a given macroobservable F: ~ ( k ' ) ~ L C B' the set J/l 0 is just given
by dd 0 = {g(y), y ~ Y} where g: Y -+ L is the mapping resulting from the
generalized R a d o n - N i k o d y m theorem. Now, taking formally the inverse
g-1 of g, one may then reconstruct the original macroobservable/7.. ~ ( ~ ' ) - 7
L C B' when P': ~(.~d0) --+ L C B' is given.
Motivated by this, we shall make the following assumption: Existence
of an objectifying function

which is required to be a continuous mapping. N originally need not be


defined on the whole of JfL0 , but merely on a dense subset of dd 0 .
398 Melsheimer

is called an objectifying function since it connects quantum mechanical


effects with the objectively given trajectories. One may consider C~ as the
proper tool that brings about the passage from quantum mechanics to the
structure of classical physics. In a way it replaces completely the map
R: C ( V ) ~ B'.
Once the function ~ has been found, the construction of the macro-
observable is performed in the following way:
F ( B ) = F(cg-l(B)) for all B ~ M(Y)

Since ~ is continuous and -~Z0 is a compact set, f&X(B) is a Borel set of Jr0
for every B ~ ( Y ) . According to Dinculeanu (Ref. 20, p. 402), such an
F: ~ ( Y ) - - + L is a ~-additive measure which is regular and of bounded
variation if/? has these properties.
We shall show that up = UF = U holds. According to Proposition 1, we
need only to show up = uF. However, this equality is a consequence of the
following relation:

up(z) = f d~o z(b) din(b) = P'(z)(~o) = P'(z)(~C-I(V))

= F'(z)(Y) = uv(z) for all z ~ h(B')

The measure F constructed in this way necessarily fulfills the STUP: We


define a measure A ( B ) = m(f&I(B)) for all B ~ ( Y ) , for which we get

~(u, F(~)) = t~(u, P(~-I(B))) = f~-l(~ t~(u, b) d~,(b)


m(fY-~(B)) = A(B) for all B E M(Y) and all u ~ S~,~7~

Note that tz(u, b) ~ 1 for all u E Se~ and all b ~ L.


We conclude this section with a short remark concerning the com-
patibility of the time-shift transformations T~ and ~ for the classical theory
and many-body quantum mechanics, respectively.
According to the results of Ref. 7, there has to exist a projection P onto
the set ~g such that ~ : ~ P is a semigroup for ~- ~ 0 and ¢~F(B) =
F(T~-a(B)) holds for all ~- ~ 0 and all B ~ ~ ( Y ) . Unfortunately, we have not
been able to state such conditions that the validity of the last equation could
be derived. It is easy to see that by putting
~ F ( B ) : B(T~-~(B)) for all B ~ ~(Y)
and ~- ~ 0, one defines a semigroup of transformations mapping d{ into
itself. The question then is when do a~ and ~, coincide ? This question mainly
concerns the existence of the projection mapping P, which--as one may see
from Ref. 7--also depends on the set of macroensembles. Since we have no
On the Reconstruction of the Macroobservable 399

definite results concerning the determination of the set of macroensembles,


it is impossible to solve the problem of how to obtain P here.

4. T H E F O R M A L S T R U C T U R E OF S T A T I S T I C A L M E C H A N I C S

We shall close this paper with a short discussion of the consequences


that the results of the last section have for an analysis of the formal structure
of statistical mechanics. There are a few particular aspects that seem to be
new for the foundation of statistical mechanics. From our general point of
view the structure of statistical mechanics has to be analyzed in terms of the
embedding concept as described in Section 2 of Ref. 7. It yields a clear-cut
mathematical prescription of how a classical theory has to be reproduced
within the somewhat formal scheme of many-body quantum mechanics.
All the more we obtain a physical interpretation of the relevant mathematical
objects that pertain to classical physics. These objects were identified as
so-called macroensembles and trajectory effects, which, in a more elaborate
setting, (v~ could fully be replaced by the macroobservable. The expression
i~(Su, F(B)) with u c K~,~, B ~ ~ ( Y ) (all notation as in Ref. 7) gives approxi-
mately the relative frequency for the statement that physical systems that
have been prepared according to u ~ I(,,~ have trajectores falling into the set B.
The statistical predictions refer in this way to the dynamics of the classical
systems. However, it becomes already clear from this context that for a fuI
description of the dynamics, we need to know not only the set of macro-
ensembles SKIn, but also the explicit form of the macroobservable F: ~ ( Y ) --~
L C B'. I f both SK,~ and F were known for a given classical system, we would
have in fact succeded in solving the reconstruction problem as formulated in
Section 3.2.1. The knowledge of the projection mapping P is only necessary
if we want to treat cases like a deterministic classical theory or a classical
theory with a so-called Markovian dynamics. This treatment will be given
elsewhere.
Within the scope of this paper, we were almost exclusively concerned
with the conceptual analysis of the formal structure of the macroobservable
pertaining to a given classical theory. As already mentioned, the knowledge
of the internal structure of the macroobservable is of paramount importance
for a proper foundation of statistical mechanics. The most intriguing new
concept that we were led to propose is the so-called objectifying function N,
which relates quantum mechanical effects to classical trajectories. It seems
now in order to discuss in more detail the conceptual relevance of this
function. As mentioned in Section 1, the most difficult problem for the
foundation of statistical mechanics is the justification of the existence of
objectively given state parameters (trajectories) within the context of quantum
400 Melsheimer

mechanics. In general, state parameters cannot be described by ensemble


averages of ordinary observables. Instead, the analysis in Section 3.2 clearly
shows that their relation to the formalism of quantum mechanics is of quite
a different nature. The objectifying function expresses the fact that many-body
quantum theory is by no means the correct physical theory for the description
of classical macroscopic systems, but is merely a very broad formal scheme
that allows the reproduction of the theoretical formalism for classical
systems as is revealed by the embedding structure. In a forthcoming paper,
we will treat explicitly the case of equilibrium statistical mechanics, where
we wilt present the form of the objectifying function and a suitable set of
effects that are related to the classical state parameters. As for the case of
nonequilibrium statistical mechanics, there seems to exist only the concept
o f the so-called Wigner distribution function, ~21,~2)which resembles in a way
the concept of the objectifying function. In contrast to the Wigner distribu-
tion function, we want, however, to stress the point that our objectifying
function is not intended to yield directly trajectories with the correct classical
time evolution. As pointed out in Section 3, the dynamical laws can only be
obtained through the complete knowledge of the macroobservable and the
set of macroensembles, both leading to the measures on the trajectory space
that describe the dynamical laws.
For the actual reconstruction of a macroobservable F: ~ ( Y ) --- L C B'
one will proceed along the following lines: First choose a suitable set of
effects J{0 and an appropriate objectifying function ~ which maps de" o
into Y. With the help of a suitable Radon measure m on d l o construct the
measure/?: ~ ( ~ 0 ) - ~ L C B' as described in Section 3.2.3. Combining the
measure P and the objectifying function ff, one then gets the desired measure
N: ~ ( Y ) --+ L C B' as has been shown in Section 3.2.4. However, this is only
a formal program. The physical arguments for the choice of d4"0 , N, and
the Radon measure m have to be given.

ACKNOWLEDGMENT

Part of this work was done while the author was staying at the Istituto
di Scienze Fisiche dell' UniversitS. di Milano. The author would like to thank
the members of this institute for their kind hospitality as well as for many
invaluable discussions and comments.

REFERENCES

1. Carlo Cercignani, Mathematical Methods in Kinetic Theory (Plenum Press, 1969).


2. G. Ludwig, Einfiihrung in die Grundlagen der Theoretisehen Physik, Vol. 4 (Bertelsmann
Universit~itsverlag).
On the Reconstruction of the Maeroobservable 401

3. L. Boltzmann, Uber die Unentbehrlichkeit der Atomistik in der Naturwissensehaft


(Populate Schriften, Ambrosius Barth, Leipzig, 1905).
4. G. Ludwig, in Proceedings of the International School of Physics "Enrico Fermi"
Varenna, Italy, 1960 (Academic Press, New York, 1961).
5. G. Ludwig, in The Uncertainty Principle and Foundation of Quantum Mechanics,
W. C. Price and S. S. Chissick, eds. (Wiley, London, 1977).
6. E. G. D Cohen and W. Thirring, eds., The Boltzmann Equation. Aeta Phys. Austriaca,
Suppl. X (Springer-Verlag, Vienna, 1973).
7. O. Melsheimer, Found. Phys. 9, 193 (1979).
8. A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47, 777 (1935).
9. H. Araki and E. J. Woods, J. Math. Phys. 4, 637 (1963).
10. G. Ludwig, Makroskopische Systeme und Qaantenmechanik (Notes in Mathematical
Physics, No. 5; University of Marburg, 1972).
11. L. Lanz, in The Uncertainty Principle and Foundation of Quantum Mechanics, W. C.
Price and S. S. Chissick, eds. (Wiley, London, 1977).
12. J. Kluvanek and G. Knowles, Vector Measures and Control Systems (North-Holland,
1976).
13. C. M. Edwards, Comm. Math. Phys. 20, 26 (1971).
14. G. Ludwig, Einfiihrung in die Grundlagen der Theoretisehen Physik, Vol. 3 (Braunsch-
weig, Vieweg, 1976).
15. J. Kluvanek, J. Funct. Anal. 31, 316 (1976).
16. G. Choquet, Leetures on Analysis, Vols. I-III (Benjamin, 1969).
17. E. A. Alfsen, Compact Convex Sets and Boundary Integrals (Springer Verlag, Berlin,
1971).
18. H. H. Sch/ifer, Topological Vector Spaces (Macmillan, New York, 1966).
19. J. Kupka, Trans. Am. Math. Soe. 169, 197 (1972).
20. N. Dinculeanu, Vector Measures (Pergamon Press, 1967).
21. S. Fujity, Introduction to Non-Equilibrium Quantum Statistical Meehanies (Saunders,
Philadelphia, 1966).
22. D. Massignon, Mdcanique statistique des fluides (Dunod, Paris, 1957).

You might also like