Topics in Quantum Dynamics by A. Jadczyk Published in: INFINITE DIMENSIONAL GEOMETRY, NONCOMMUTATIVE GEOMETRY, OPERATOR ALGEBRAS AND FUNDAMENTAL INTERACTIONS Proceedings of the First Caribbean Spring School of Mathematics and Theoretical Physics Saint-François-Guadeloupe 30 May - 13 June 1993 edited by R Coquereaux (CNRS, CPT-Marseille), M Dubois-Violette & P Flad (CNRS, Univ. Paris XI)

# Introduction

## The Two Kinds of Evolution

In these lectures I will discuss the two kinds of evolution of quantum systems:
• CONTINUOUS evolution of closed systems
• STOCHASTIC evolution of open systems.
The first type concerns evolution of closed, isolated1 quantum systems that evolve under the action of prescribed external forces, but are not disturbed by observations, and are not coupled thermodynamically or in some other irreversible way to the environment. This evolution is governed by the Schrödinger equation and is also known as a unitary or, more generally, as an automorphic evolution. In contrast to this idealized case (only approximately valid, when irreversible effects can be neglected), quantum theory is also concerned with a different kind of change of state. It was first formulated by J. von Neumann (cf. [30, Ch. V. 1] and is known as von Neumann - Lüders projection postulate. It tells us roughly this: if some quantum mechanical observable is being measured, then - as a consequence of this measurement - the actual state of the quantum system jumps into one of the eigenstates of the measured observable. 2 This jump was thought to be abrupt and take no time at all , it is also known as reduction of the wave packet. Some physicists feel quite uneasy about this von Neumann's postulate, to the extent that they reject it either as too primitive (not described by dynamical equations) or as unnecessary. We will come to this point later, in Chapter 3, when we will discuss piecewise deterministic stochastic processes that unite both kinds of evolution.

## The Schrödinger Equation

It is rather easy to explain to the mathematician the Dirac equation - it became already a part of the mathematical folklore. But Dirac's equation belongs to field theory rather than to quantum theory. 3 Physicists are being taught in the course of their education rather early that every attempt at a sharp localization of a relativistic particle results in creation and annihilation processes. Sometimes it is phrased as: "there is no relativistic quantum mechanics - one needs to go directly to Relativistic Quantum Field Theory". Unfortunately we know of none non-trivial, finite, relativistic quantum field theory in four dimensional space-time continuum. Thus we are left with the mathematics of perturbation theory. Some physicists believe that the physical ideas of relativistic quantum field theory are sound, that it is the best theory we ever had, that it is "the exact description of nature", that the difficulties we have with it are only temporary, and that they will be overcomed one day - the day when bright mathematicians will provide us with new, better, more powerful tools. Some other say: perturbation theory is more than sufficient for all practical purposes, no new tools are needed, that is how physics is - so mathematicians better accept it, digest it, and help the physicists to make it more rigorous and to understand what it is really about. Still some other4, a minority, also believe that it is only a temporary situation, which one day will be resolved. But the resolution will come owing to essentially new physical ideas, and it will result in a new quantum paradigm, more appealing than the present one. It should perhaps be not a surprise if, in an appropriate sense, all these points of view will turn out to be right. In these lectures we will be concerned with the well established Schrödinger equation, which is at the very basis of the current quantum scheme, and with its dissipative generalization - the Liouville equation. In these equations we assume that we know what the time is. Such a knowledge is negated in special relativity, 5 and this results in turn in all kinds of troubles that we are facing since the birth of Einstein's relativity till this day. 6
The Schrödinger equation is more difficult than the Dirac one, and this for two reasons: first, it lives on the background of Galilean relativity - which have to deal with much more intricate geometric structures than Einstein relativity. 7 Second, Schrödinger's equation is about Quantum Mechanics and we have to take care about probabilistic interpretation, observables, states etc. - which is possible for Schrödinger equation but faces problems in the first-quantized Dirac theory.

Let me first make a general comment about Quantum Theory. There are physicists who would say: quantum theory is about calculating of Green's Functions - all numbers of interest can be obtained from these functions and all the other mathematical constructs usually connected with quantum theory are superfluous and unnecessary! It is not my intention to depreciate the achievements of Green's Function Calculators. But for me quantum theory - as any other physical theory - should be about explaining things that happen outside of us - as long as such explanations are possible. The situation in Quantum Theory today, more than 60 years after its birth, is that Quantum Theory explains much less than we would like to have been explained. To reduce quantum theory to Green's function calculations is to reduce its explanatory power almost to zero. It may of course be that in the coming Twenty First Century humanity will unanimously recognize the fact that understanding'was a luxury, a luxury of the "primitive age" that is gone for ever. But I think it is worth while to take a chance and to try to understand as much as can be understood in a given situation. Today we are trying to understand Nature in terms of geometrical pictures and random processes8 More specifically, in quantum theory, we are trying to understand in terms of observables, states, complex Feynman amplitudes etc. In the next chapter, we will show the way that leads to the Schrödinger Equation using geometrical language as much as possible. However, we will not use the machinery of geometrical quantization because it treats time simply as a parameter, and space as absolute and given once for all. 9On the other hand, geometrical quantization introduces many advanced tools that are unnecessary for our purposes, while at the same time it lacks the concepts which are important and necessary.
Before entering the subject let me tell you the distinguishing feature of the approach that I am advocating, and that will be sketched below in Ch. 2: one obtains a fibration of Hilbert spaces over time. There is a distinguished family of local trivializations, a family parameterized by

• space-time observer
• gauge.
For each , the Hilbert space is a Hilbert space of sections of a complex line bundle over . A space-time observer (that is, a reference frame) allows us to identify the spaces for different -s, while a gauge allows us to identify the fibers. Schrödinger's dynamics of a particle in external gravitational and electromagnetic fields is given by a Hermitian connection in Solutions of Schrödinger equation are parallel sections of Thus Schrödinger equation can be written as
 (1)

or, in a local trivialization, as
 (2)

where will be a self-adjoint operator in . 10Gravitational and electromagnetic forces are coded into this Schrödinger's connection. Let us discuss the resulting structure. First of all there is not a single Hilbert space but a family of Hilbert spaces. These Hilbert spaces can be identified either using an observer and gauge or,better, by using a background dynamical connection. It is only after so doing that one arrives at single-Hilbert-space picture of the textbook Quantum Mechanics - a picture that is the constant source of lot of confusion. 11
In Quantum Mechanics we have a dual scheme - we use the concepts of observables and states. We often use the word measurement in a mutilated sense of simply pairing an observable with a state to get the expected result - a number
One comment is in place here: to compare the results of actual measurements with predictions of the theory - one needs only real numbers. However experience proved that quantum theory with only-real-numbers is inadequate. So, even if the fundamental role of in Quantum Theory is far from being fully understood - we use in Quantum Theory only complex Hilbert spaces, complex algebras etc.12However, usually, only real numbers are at the end interpreted.

Now, it is not always clear what is understood by states and observables. There are several possibilities:

As it was already said, it is usual in standard presentations of the quantum theory to identify the Hilbert spaces . There are several options there. Either we identify them according to an observer (+ gauge) or according to the dynamics. If we identify according to the actual dynamics, then states do not change in time - it is always the same state-vector, but observables (like position coordinates) do change in time - we have what is called the Heisenberg picture. If we identify them according to some background "free" dynamics - we have so called interaction picture. Or, we can identify Hilbert spaces according to an observer - then observables do not change in time, but state vector is changing - we get the Schrödinger picture.

However, there is no reason at all to identify the -s. Then dynamics is given by parallel transport operators:

## Dissipative Dynamics

The Schrödinger equation describes time evolution of pure states of a quantum system, for instance evolution of pure states of a quantum particle, or of a many body system. Even if these states contain only statistical information about most of the physical quantities, the Schrödinger evolution of pure states is continuous and deterministic . Under this evolution Hilbert space vector representing the actual state of the system13 changes continuously with time, and with it there is a continuous evolution of probabilities or potentialities, but nothing happens - the formalism leaves no place for events. Schrödinger equation helps us very little, or nothing at all, to understand how potential becomes real. So, if we aim at understanding of this process of becoming, if we want to describe it by mathematical equations and to simulate it with computers - we must go beyond Schrödinger's dynamics. As it happens, we do not have to go very far - it is sufficient to relax only one (but important) property of Schrödinger's dynamics and to admit that pure states can evolve into mixtures. Instead of Schrödinger equation we have then a so called Liouville equation that describes time evolution of mixed states. It contains Schrödinger equation as a special case.14It was shown in [11] that using the Liouville type of dynamics it is possible to describe coupling between quantum systems and classical degrees of freedom of measurement devices. One can derive also a piecewise deterministic random process that takes place on the manifold of pure states. In this way one obtains a minimal description of "quantum jumps" (or "reduction of wave packets") and accompanying, directly observable jumps of the coupled classical devices. In Ch. 3 simple models of such couplings will be discussed. The interested reader will find more examples in Refs. [8]-[11].15In particular in [11] the most advanced model of this kind, the SQUID-tank model is discussed in details.

# Geometry of Schrödinger's Equation

## Preliminaries

Galilean General Relativity is a theory of space-time structure, gravitation and electromagnetism based on the assumption of existence of an absolute time function. Many of the cosmological models based on Einsteins relativity admit also a distinguished time function. Therefore Galilean physics is not evidently wrong. Its predictions must be tested by experiments. Galilean relativity is not that elegant as the one of Einstein. This can be already seen from the group structures: the homogeneous Lorentz group is simple, while the homogeneous Galilei group is a semidirect product of the rotation group and of three commuting boosts. Einsteins theory of gravitation is based on one metric tensor, while Galilean gravity needs both: space metric and space-time connection. Similarly for quantum mechanics: it is rather straightforward to construct generally covariant wave equations for Einsteins relativity, while general covariance and geometrical meaning of the Schrödinger equation was causing problems, and it was not discussed in textbooks. In the following sections we will present a brief overview of some of these problems.

## Galilean General Relativity

Let us discuss briefly geometrical data that are needed for building up generally covariant Schrödinger's equation. More details can be found in Ref. [21].16
Our space-time will be a refined version of that of Galilei and of Newton, i.e. space-time with absolute simultaneouity. Four dimensional space-time is fibrated over one-dimensional time The fibers of are three-dimensional Riemannian manifolds, while the basis is an affine space over By a coordinate system on we will always mean a coordinate system adapted to the fibration. That means: any two events with the same coordinate are simultaneous, i.e. in the same fibre of
Coordinate transformations between any two adapted coordinate systems are of the form:

We will denote by the time form Thus in adapted coordinates .
is equipped with a contravariant degenerate metric tensor which, in adapted coordinates, takes the form

where is of signature We denote by the inverse matrix. It defines Riemannian metric in the fibers of
We assume a torsion-free connection in that preserves the two geometrical objects and . 17The condition is equivalent to the conditions on the connection coefficients. Let us introduce the notation Then is equivalent to the equations:
 (3)

Then, because of the assumed zero torsion, the space part of the connection can be expressed in terms of the space metric in the Levi-Civita form:
 (4)

>From the remaining equations:
 (5)

we find that the -symmetric part of is equal to otherwise the connection is undetermined. We can write it, introducing a new geometrical object , as
 (6)

 (7)

where is antisymmetric. Notice that is not a tensor, except for pure space transformations or time translations.

### The Bundle of Galilei Frames

A basis in is called a Galilei frame if , and if are unit space-like vectors. If and are two Galilei frames at the same space-time point, then they are related by a transformation of the homogeneous Galilei group :
 (8)

 (9)

where and is an orthogonal matrix. The bundle of Galilei frames is a principal bundle.

### The Bundle of Velocities

The homogeneous Galilei group acts on in two natural ways: by linear and by affine transformations. The first action is not effective one - it involves only the rotations:
 (10)

The bundle associated to this action can be identified with the vertical subbundle of - i.e. with the bundle of vectors tangent to the fibers of .
acts also on by affine isometries :
 (11)

To this action there corresponds an associated bundle, which is an affine bundle over the vector bundle . It can be identified with the subbundle of consisting of vectors tangent to , and such that or, equivalently, as the bundle of first jets of sections of . We will call it .
We will denote by the coordinates in corresponding to coordinates of .

## The Presymplectic Form

The connection can be also considered as a principal connection in the bundle of Galilei frames. It induces an affine connection in the affine bundle . As a result, it defines a natural -valued one-form on . It can be described as follows: given a vector tangent to at it projects onto . Then is defined as the difference of and the horizontal lift of . It is a vertical tangent vector to , and can be identified with an element of . In coordinates:
 (12)

There is another -valued one-form on , namely the canonical form . Given at , we can decompose into space- and time-component along . Then is defined as its space component. In coordinates:
 (13)

Then, because the fibers of are endowed with metric , we can build out the following important two-form on :
 (14)

Explicitly
 (15)

The following theorem, proven in [21], gives a necessary and sufficient condition for to be closed.

Theorem 1   The following conditions (i-iii) are equivalent:
(i)
(ii)
where is the curvature tensor of and
(iii)
. 18

## Quantum Bundle and Quantum Connection

Let be a principal bundle over and let be its pullback to . We denote by and the associated Hermitian line bundles corresponding to the natural action of on . There is a special class of principal connections on , namely those whose connection forms vanish on vectors tangent to the fibers of . As has been discussed in [29] specifying such a connection on is equivalent to specifying a system of connections on parameterized by the points in the fibers of . Following the terminology of [29] we call such a connection universal.19The fundamental assumption that leads to the Schrödinger equations reads as follows:
Quantization Assumption: There exists a universal connection whose curvature is .20

We call such an a quantum connection. >From the explicit form of one can easily deduce that is necessarily of the form

where parameterizes the fibres of ,

and is a local potential for .

## Schrödinger's Equation and Schrödinger's Bundle

As it is shown in [21], there exists a natural -invariant metric on of signature . Explicitly

Using this metric we can build out a natural Lagrangian for equivariant functions or, equivalently, for sections of the line bundle . The Euler-Cartan equation for this Lagrangian will prove to be nothing but the Schrödinger equation. Notice that the action of group on defines an Killing vector field for which is isotropic. Therefore the above construction can explain why the approach of [19] works.

More precisely, the construction leading to the generally covariant Schrödinger-Pauli equation for a charged spin particle in external gravitational and electromagnetic field can be described as follows.

The contravariant metric , ,

 (16)

can be obtained from the following Clifford algebra of complex matrices:
 (17)

One takes then 5-dimensional charged Dirac operator and considers spinors that are equivariant with respect to the fifth coordinate
 (18)

This first-order, four-component spinor (called Lévy-Leblond equation in Ref. [24]) equation reduces then easily to the second-order, two-component Schrödinger Pauli equation with the correct Landé factor.21

We finish this section with pointing to the Ref. [21], where Schrödinger's quantization is discussed in details and where a probabilistic interpretation of generally covariant Schrödinger equation is given using the bundle of Hilbert spaces. The parallel transport induced by the quantum connection is shown in [21] to be directly related to Feynman amplitudes.

# Coupled Quantum and Classical Systems

## Preliminaries

Replacing Schrödinger's evolution, which governs the dynamics of pure states, by an equation of the Liouville type, that describes time evolution of mixed states, is a necessary step - but it does not suffice for modeling of real world events. One must take, to this end, two further steps. First of all we should admit that in our reasoning, our communication, our description of facts - we are using classical logic. Thus somewhere in the final step of transmission of information from quantum systems to macroscopic recording devices and further, to our senses and minds, a translation between quantum and classical should take place. That such a translation is necessary is evident also when we consider the opposite direction: to test a physical theory we perform controlled experiments. But some of the controls are always of classical nature - they are external parameters with concrete numerical values. So, we need to consider systems with both quantum and classical degrees of freedom, and we need evolution equations that enable communication in both directions, i.e. :
• flow of information from quantum to classical
and
• control of quantum states and processes by classical parameters .

## Completely Positive Maps

We begin with a brief recall of relevant mathematical concepts. Let be a - algebra. We shall always assume that has unit . An element is positive, , iff it is of the form for some . Every element of a -algebra is a linear combination of positive elements. A linear functional is positive iff implies . Every positive functional on a -algebra is continuous and Positive functionals of norm one are called states. The space of states is a convex set. Its extremal points are called pure states. The canonical GNS construction allows one to associate with each state a representation of on a Hilbert space , and a cyclic vector such that Irreducibility of is then equivalent to purity of

Quantum theory gives us a powerful formal language and statistical algorithms for describing general physical systems. Physical quantities are coded there by Hermitian elements of a -algebra of observables, while information about their values (quantum algorithms deal, in general, only with statistical information) is coded in states of . Pure states correspond to a maximal possible information. For each state and for each the (real) number is interpreted as expectation value of observable in state while

is the quadratic dispersion of in the state It is assumed that repeated measurements of 22made on systems prepared in a state will give a sequence of values so that approximately and If is Abelian, then it is isomorphic to an algebra of functions Then pure states of are dispersion free - they are parameterized by points and we have This corresponds to a classical theory: all observables mutually commute and maximal possible information is without any statistical dispersion. In the extreme opposition to that is pure quantum theory - here defined as that one when is a factor, that is has a trivial centre. The centre of a -algebra is defined as In general If Z (A) = A - we have pure classical theory. If - we have pure quantum theory. In between we have a theory with superselection rules. Many physicists believe that the ''good theory'' should be a ''pure quantum'' theory. But I know of no one good reason why this should be the case. In fact, we will see that cases with a nontrivial are interesting ones. Of course, one can always argue that whenever we have an algebra with a nontrivial centre - it is a subalgebra of an algebra with a trivial one, for instance of - the algebra of all bounded operators on some Hilbert space. This is, however, not a good argument - one could argue as well that we do not need to consider different groups as most of them are subgroups of - the unitary group of an infinite dimensional Hilbert space - so why to bother with others?

Let be -algebras. A linear map is Hermitian if It is positive iff implies Because Hermitian elements of a -algebra are differences of two positive ones - each positive map is automatically Hermitian. Let denote the by matrix algebra, and let be the algebra of matrices with entries from Then carries a natural structure of a -algebra. With respect to this structure a matrix from is positive iff it is a sum of matrices of the form If is an algebra of operators on a Hilbert space , then can be considered as acting on Positivity of is then equivalent to or equivalently, to for all
A positive map is said to be completely positive or, briefly, CP iff defined by , is positive for all When written explicitly, complete positivity is equivalent to

 (19)

for every and In particular every homomorphism of algebras is completely positive. One can also show that if either or is Abelian, then positivity implies complete positivity. Another important example: if is a algebra of operators on a Hilbert space , and if then is a CP map The celebrated Stinespring theorem gives us a general form of a CP map. Stinespring's construction can be described as follows. Let be a CP map. Let us restrict to a case when Let be realized as a norm closed algebra of bounded operators on a Hilbert space . One takes then the algebraic tensor product and defines on this space a sesquilinear form by
 (20)

This scalar product is then positive semi-definite because of complete positivity of Indeed, we have
 (21)

Let denote the kernel of . Then is a pre-Hilbert space. One defines a representation of on by One shows then that is invariant under , so that goes to the quotient space. Similarly, the map defines an isometry We get then on the completion of

Theorem 2 (Stinespring's Theorem)   Let be a -algebra with unit and let be a CP map. Then there exists a Hilbert space a representation of on and a bounded linear map such that
 (22)

is an isometry iff is unital i.e. iff maps the unit of into the identity operator of If and are separable, then can be taken separable.

The space of CP maps from to is a convex set. Arveson [2] proved that is an extremal element of this set iff the representation above is irreducible.

## Dynamical Semigroups

A dynamical semigroup on a -algebra of operators is a strongly continuous semigroup of CP maps of A into itself. A semigroup is norm continuous iff its infinitesimal generator is bounded as a linear map We then have

 (23)

The right hand side is, in this case, a norm convergent series for all real values of however for negative the maps , although Hermitian, need not be positive.
Evolution of observables gives rise, by duality, to evolution of positive functionals. One defines Then preserves the unit of iff preserves normalization of states. A general form of a generator of a dynamical semigroup in finite dimensional Hilbert space has been derived by Gorini, Kossakowski and Sudarshan [22], and Lindblad [27] gave a general form of a bounded generator of a dynamical semigroup acting on the algebra of all bounded operators It is worthwhile to cite, after Lindblad, his original motivation:

"The dynamics of a finite closed quantum system is conventionally represented by a one-parameter group of unitary transformations in Hilbert space. This formalism makes it difficult to describe irreversible processes like the decay of unstable particles, approach to thermodynamic equilibrium and measurement processes []. It seems that the only possibility of introducing an irreversible behaviour in a finite system is to avoid the unitary time development altogether by considering non-Hamiltonian systems. "

In a recent series of papers [7,8,9,10] Ph. Blanchard and the present author were forced to introduce dynamical semigroups because of another difficulty, namely because of impossibility of obtaining a nontrivial Hamiltonian coupling of classical and quantum degrees of freedom in a system described by an algebra with a non-trivial centre. We felt that lack of a dynamical understanding of quantum mechanical probabilistic postulates is more than annoying. We also believed that the word ''measurement'' instead of being banned, as suggested by J. Bell [5,6], can be perhaps given a precise and acceptable meaning. We suggested that a measurement process is a coupling of a quantum and of a classical system, where information about quantum state is transmitted to the classical recording device by a dynamical semigroup of the total system . It is instructive to see that such a transfer of information can not indeed be accomplished by a Hamiltonian or, more generally, by any automorphic evolution23. To this end consider an algebra with centre Then describes classical degrees freedom. Let be a state of then denotes its restriction to Let be an automorphic evolution of and denote Each is an automorphism of the algebra and so it leaves its centre invariant: The crucial observation is that, because of that fact, the restriction depends only on as the evolution of states of is dual to the evolution of the observables in This shows that information transfer from the total algebra to its centre is impossible - unless we use more general, non-automorphic evolutions.
>From the above reasoning it may be seen that the Schrödinger picture, when time evolution is applied to states, is better adapted to a discussion of information transfer between different systems. The main properties that a dynamical semigroup describing time evolution of states should have are: should preserve convex combinations, positivity and normalization. One can demand even more - it is reasonable to demand a special kind of stability: that it should be always possible to extend the system and its evolution in a trivial way, by adding extra degrees of freedom that do not couple to our system.24That is exactly what is assured by complete positivity of the maps One could also think that we should require even more, namely that transforms pure states into pure states. But to assume that would be already too much, as one can prove that then must be dual to an automorphic evolution. It appears that information gain in one respect (i.e. learning about the actual state of the quantum system) must be accompanied by information loss in another one - as going from pure states to mixtures implies entropy growth.

We will apply the theory of dynamical semigroup to algebras with a non-trivial centre. In all our examples we will deal with tensor products of and an Abelian algebra of functions. The following theorem by Christensen and Evans [14] generalizes the results of Gorini, Kossakowski and Sudarshan and of Lindblad to the case of arbitrary -algebra.

Theorem 3 (Christensen - Evans)   Let be a norm-continuous semigroup of CP maps of a - algebra of operators Then there exists a CP map of into the ultraweak closure and an operator such that the generator is of the form:
 (24)

We will apply this theorem to the cases of being a von Neumann algebra, and the maps being normal. Then can be also taken normal. We also have so that We will always assume that or, equivalently, that Moreover, it is convenient to introduce then from we get and so Therefore we have
 (25)

where denotes anticommutator. Of particular interest to us will be generators for which is extremal 25. By the already mentioned result of Arveson [2] this is the case when is of the form
 (26)

where is an irreducible representation of on a Hilbert space and is a bounded operator (it must be, however, such that ).

## Coupling of Classical and Quantum Systems

We consider a model describing a coupling between a quantum and a classical system. To concentrate on main ideas rather than on technical details let us assume that the quantum system is described in an -dimensional Hilbert space and that it has as its algebra of observables Similarly, let us assume that the classical system has only a finite number of pure states Its algebra of observables is then isomorphic to For the algebra of the total system we take which is isomorphic to the diagonal subalgebra of Observables of the total system are block diagonal matrices:

where are operators in 26Both and can be considered as subalgebras of consisting respectively of matrices of the form and States of the quantum system are represented by positive, trace one, operators on States of the classical system are -tuples of non-negative numbers with States of the total system are represented by block diagonal matrices with and For the expectation value we have Given a state of the total system, we can trace over the quantum system to get an effective state of the classical system or we can trace over the classical system to get the effective state of the quantum system 27
Let us consider dynamics. Since the classical system has a discrete set of pure states, there is no non-trivial and continuous time evolution for the classical system that would map pure states into pure states. As for the quantum system, we can have a Hamiltonian dynamics, with the Hamiltonian possibly dependent on time and on the state of the classical system As we already know a non-trivial coupling between both systems is impossible without a dissipative term, and the simplest dissipative coupling is of the form where is an irreducible representation of the algebra in a Hilbert space and is a linear map. It is easy to see that such an is necessarily of the form:

where is an block matrix with only one non-zero entry. A more general CP map of is of the same form, but with having at most one non-zero element in each of its rows.
Let us now discuss desired couplings in somewhat vague, but more intuitive, physical terms. We would like to write down a coupling that enables transfer of information from quantum to classical system. There may be many ways of achieving this aim - the subject is new and there is no ready theory that fits. We will see however that a naive description of a coupling works quite well in many cases. The idea is that the simplest coupling associates to a property of the quantum system a transformation of the actual state of the classical system.
Properties are, in quantum theory, represented by projection operators. Sometimes one considers also more general, unsharp or fuzzy properties. They are represented by positive elements of the algebra which are bounded by the unit. A measurement should discriminate between mutually exclusive and exhaustive properties. Thus one usually considers a family of mutually orthogonal projections of sum one. With an unsharp measurement one associates a family of positive elements of sum one.
As there is no yet a complete, general, theory of dissipative couplings of classical and quantum systems, the best we can do is to show some characteristic examples. It will be done in the following section. For every example a piecewise deterministic random process will be described that takes place on the space of pure states of the total system 28 and which reproduces the Liouville evolution of the total system by averaging over the process. A theory of piecewise deterministic (PD) processes is described in a recent book by M. H. Davis [15]. Processes of that type, but without a non-trivial evolution of the classical system, were discussed also in physical literature - cf. Refs [13,17,18,20]. 29We will consider Liouville equations of the form
 (27)

where in general and the can explicitly depend on time. The will be chosen as tensor products , where act as transformations 30 on classical (pure) states.

## Examples of Classical-Quantum Couplings

### The Simplest Coupling

First, we consider only one orthogonal projector on the two-dimensional Hilbert space To define the dynamics we choose the coupling operator in the following way:
 (28)

The Liouville equation ( ) for the density matrix of the total system reads now
 (29)

For this particularly simple coupling the effective quantum state evolves independently of the state of the classical system. One can say that here we have only transport of information from the quantum system to the classical one. We have:
 (30)

The Liouville equation ( ) describes time evolution of statistical states of the total system.
Let us describe now a the PD process associated to this equation. Let be a one-parameter semigroup of (non-linear) transformations of rays in given by
 (31)

where
 (32)

Suppose we start with the quantum system in a pure state , and the classical system in a state (resp. ). Then starts to evolve according to the deterministic (but non-linear Schrödinger) evolution until a jump occurs at time . The time of the jump is governed by an inhomogeneous Poisson process with the rate function . Classical system switches from to (resp. from to ), while jumps to , and the process starts again. With the initial state being an eigenstate of , , the rate function is approximately constant and equal to . Thus can be interpreted as the expected time interval between the successive jumps.

### Simultaneous "Measurement" of Several Noncommuting Observables

Using somewhat pictorial language we can say that in the previous example each actualization of the property was causing a flip in the classical system. In the present example, which is a non-commutative and fuzzy generalization of the model discussed in [7], we consider in general fuzzy, properties The Hilbert space can be completely arbitrary, for instance -dimensional. We will denote The -s need not be projections, and the different -s need not to commute. The classical system is assumed to have states with thought of as an initial, neutral state. To each actualization of the property there will be associated a flip between and Otherwise the state of the classical system will be unchanged. To this end we take

The Liouville equation takes now the following form:

 (33)

 (34)

We will derive the PD process for this example in some more details, so that a general method can be seen. First of all we transpose the Liouville equation so as to get time evolution of observables; we use the formula
 (35)

In the particular case at hand the evolution equation for observables looks almost exactly the same as that for states:
 (36)

 (37)

Each observable of the total system defines now a function on the space of pure states of the total system
 (38)

We have to rewrite the evolution equation for observables in terms of the functions To this end we compute the expressions Let us first introduce the Hamiltonian vector field on the manifold of pure states of the total system:
 (39)

Then the terms can be written as We also introduce vector field corresponding to non-linear evolution:
 (40)

Then evolution equation for observables can be written in a Davis form:
 (41)

where is a matrix of measures, whose non-zero entries are:
 (42)

 (43)

while
 (44)

The symbol denotes here the Dirac measure concentrated at
We describe now PD process associated to the above semigroup. There are one-parameter (non-linear) semigroups acting on the space of pure states of the quantum system via

where

If initially the classical system is in a pure state , and quantum system in a pure state , then quantum system evolves deterministically according to the semigroup : . The classical system then jumps at the time instant , determined by the inhomogeneous Poisson process with rate function . If the classical system was in one of the states , then it jumps to with probability one, the quantum state jumps at the same time to the state . If, on the other hand, it was in the state , then it jumps to one of the states with probability . The quantum state jumps at the same time to Let

Then is the distribution of - the first jump time. More precisely, is the survival function for the state :

Thus the probability distribution of the jump is and the expected jump time is The probability that the jump will occur between and , provided it did not occur yet/, is equal to . Notice that this depends on the actual state . However, as numerical computation show, the dependence is negligible and approximately jumps occur always after time . 31

### Coupling to All One-Dimensional Projections

In the previous example the coupling between classical and quantum systems involved a finite set of non-commuting observables. In the present one we will go to the extreme - we will use all one-dimensional projections in the coupling. One can naturally discover such a model when looking for a precise answer to the question:

how to determine state of an individual quantum system ?
For some time I was sharing the predominant opinion that a positive answer to this question can not be given, as there is no observable to be measured that answers the question: what state our system is in ?. Recently Aharonov and Vaidman [1] discussed this problem in some details. 32The difficulty here is in the fact that we have to discriminate between non-orthogonal projections (because different states are not necessarily orthogonal), and this implies necessity of simultaneous measuring of non-commuting observables. There have been many papers discussing such measurements, different authors taking often different positions. However they all seem to agree on the fact that predictions from such measurements are necessarily fuzzy. This fuzziness being directly related to the Heisenberg uncertainty relation for non-commuting observables. Using methods and ideas presented in the previous sections of this chapter it is possible to build models corresponding to the intuitive idea of a simultaneous measurement of several non-commuting observables, like, for instance, different spin components, positions and momenta etc. A simple example of such a model was given in the previous section. After playing for a while with similar models it is natural to think of a coupling between a quantum system and a classical device that will result in a determination of the quantum state by the classical device. Ideally, after the interaction, the classical "pointer" should point at some vector in a model Hilbert space. This vector should represent (perhaps, with some uncertainty) the actual state of the quantum system. The model that came out of this simple idea, and which we will now discuss, does not achieve this goal. But it is instructive, as it shows that models of this kind are possible. I believe that one day somebody will invent a better model, a model that can be proven to be optimal, giving the best determination with the least disturbance. Then we will learn something important about the nature of quantum states.

Our model will be formulated for a -state quantum system. It is rather straightforward to rewrite it for an arbitrary -state system, but for we can be helped by our visual imagination. Thus we take for the Hilbert space of our quantum system. We can think of it as pure spin . Pure states of the system form up the manifold which is isomorphic to the -sphere . Let , denote the Pauli -matrices. Then for each the operator has eigenvalues . We denote by the projection onto the -eigenspace.
For the space of pure states of the classical system we take also - a copy of . Notice that is a homogeneous space for . Let be the invariant measure on normalized to . In spherical coordinates we have . We denote the Hilbert space of the total system, and by its von Neumann algebra of observables. Normal states of are of the form

where satisfies

We proceed now to define the coupling of the two systems. There will be two constants:
• - regulating the time rate of jumps
• - entering the quantum Hamiltonian
The idea is that if the quantum system is at some pure state , and if the classical system is in some pure states , then will cause the Hamiltonian rotation of around with frequency , while will cause, after a random waiting time proportional to , a jump, along geodesics, to the "other side" of . The classical transformation involved is nothing but a geodesic symmetry on the symmetric space . It has the advantage that it is a measure preserving transformation. It has a disadvantage because overjumps .
We will use the notation to denote the rotation of around . Explicitly:

For each we define by
 (45)

Using -s we can define Lindblad-type coupling between the quantum system and the classical one. To give our model more flavor, we will introduce also a quantum Hamiltonian that depends on the actual state of the classical system; thus we define
 (46)

Our coupling is now given by
 (47)

Notice that and . Now, being -invariant, it must be proportional to the identity. Taking its trace we find that

and therefore
 (48)

Explicitly, using the definition of , we have
 (49)

Notice that for each operator we have the following formula: 33
 (50)

If , that is if we neglect the Hamiltonian part, then using this formula we can integrate over to get the effective Liouville operator for the quantum state :

 (51)

with the solution
 (52)

It follows that, as the result of the coupling, the effective quantum state undergoes a rather uninteresting time-evolution: it dissipates exponentially towards the totally mixed state , and this does not depend on the initial state of the classical system.

Returning back to the case of non-zero we discuss now the piecewise deterministic random process of the two pure states and . To compute it we proceed as in the previous example, with the only change that now pure states of the quantum and of the classical system are parameterized by the same set - in our case. To keep track of the origin of each parameter we will use subscripts as in and . As in the previous example each observable of the total system determines a function by

The Liouville operator , acting on observables, can be then rewritten it terms of the functions :
 (53)

where is the Hamiltonian vector field
 (54)

and

 (55)

is known as the transition probability between the two quantum states.
The PD process on can now be described as follows. Let and be the initial states of the quantum and of the classical system. Then the quantum system evolves unitarily according to the quantum Hamiltonian until at a time instant a jump occurs. The time rate of jumps is governed by the homogeneous Poisson process with rate . The quantum state jumps to a new state with probability distribution while jumps to and the process starts again (see Fig. 2).

Acknowledgements: This paper is partially based on a series of publications that were done in a collaboration with Ph. Blanchard and M. Modugno. Thanks are due to the Humboldt Foundation and Italian CNR that made this collaboration possible. The financial support for the present work was provided by the Polish KBN grant No PB 1236 on one hand, and European Community program "PECO" handled by the Centre de Physique Theorique (Marseille Luminy) on the other. Parts of this work have been written while I was visiting the CPT CNRS Marseille. I would like to thank Pierre Chiapetta and all the members of the Lab for their kind hospitality. Thanks are due to the Organizers of the School for invitation and for providing travel support. I would like to express my especially warm thanks to Robert Coquereaux for his constant interest, criticism and many fruitful discussions, and for critical reading of much of this text. Thanks are also due to Philippe Blanchard for reading the paper, many discussions and for continuous, vital and encouraging interaction.

## Bibliography

1
Aharonov, Y. and Vaidman, L. : "Measurement of the Schrödinger wave of a single particle", Phys. Lett. A178 (1993), 38-42

2
Arveson, W. B. : "Subalgebras of -algebras", Acta Math. 123 (1969), 141-224

3
Asorey, M. , Cariñena, J. F. , and Parmio, M. : "Quantum Evolution as a parallel transport", J. Math. Phys. 23 (1982), 1451-1458

4
Barnsley, M. F. : Fractals everywhere, Academic Press, San Diego 1988

5
Bell, J. : "Against measurement", in Sixty-Two Years of Uncertainty. Historical, Philosophical and Physical Inquiries into the Foundations of Quantum Mechanics, Proceedings of a NATO Advanced Study Institute, August 5-15, Erice, Ed. Arthur I. Miller, NATO ASI Series B vol. 226 , Plenum Press, New York 1990

6
Bell, J. : "Towards an exact quantum mechanics", in Themes in Contemporary Physics II. Essays in honor of Julian Schwinger's 70th birthday, Deser, S. , and Finkelstein, R. J. Ed. , World Scientific, Singapore 1989

7
Blanchard, Ph. and Jadczyk, A. : "On the interaction between classical and quantum systems", Phys. Lett. A 175 (1993), 157-164

8
Blanchard, Ph. and Jadczyk, A. : "Strongly coupled quantum and classical systems and Zeno's effect", Phys. Lett. A 183 (1993), 272-276

9
Blanchard, Ph. and Jadczyk, A. : "Classical and quantum intertwine", in Proceedings of the Symposium on Foundations of Modern Physics, Cologne, June 1993, Ed. P. Mittelstaedt, World Scientific (1993)

10
Blanchard, Ph. and Jadczyk, A. : "From quantum probabilities to classical facts", in Advances in Dynamical Systems and Quantum Physics, Capri, May 1993, Ed. R. Figari, World Scientific (1994)

11
Blanchard, Ph. and Jadczyk, A. : "How and When Quantum Phenomena Become Real", to appear in Proc. Third Max Born Symp. "Stochasticity and Quantum Chaos", Sobotka, Eds. Z. Haba et all. , Kluwer Publ.

12
Canarutto, D. , Jadczyk, A. , Modugno, M. : in preparation

13
Carmichael, H. : An open systems approach to quantum optics, Lecture Notes in Physics m 18, Springer Verlag, Berlin 1993

14
Christensen, E. and Evans, D. : "Cohomology of operator algebras and quantum dynamical semigroups", J. London. Math. Soc. 20 (1978), 358-368

15
Davies, E. B. : "Uniqueness of the standard form of the generator of a quantum dynamical semigroup", Rep. Math. Phys. 17 (1980), 249-255

16
Davis, M. H. A. : Markov models and optimization, Monographs on Statistics and Applied Probability, Chapman and Hall, London 1993

17
Dalibard, J. , Castin, Y. and Mølmer K. : "Wave-function approach to dissipative processes in quantum optics", Phys. Rev. Lett. 68 (1992), 580-583

18
Dum, R. , Zoller, P. , and Ritsch, H. : "Monte Carlo simulation of the atomic master equation for spontaneous emission", Phys. Rev. A45 (1992), 4879-4887

19
Duval, C. , Burdet, G. , Künzle, H. P. and Perrin, M. : "Bargmann structures and Newton-Cartan theory", Phys. Rev. D31, 1841-1853

20
Gardiner, C. W. , Parkins, A. S. , and Zoller, P. : "Wave-function stochastic differential equations and quantum-jump simulation methods", Phys. Rev. A46 (1992), 4363-4381

21
Jadczyk, A. and Modugno, M. : Galilei general relativistic quantum mechanics, Preprint, pp. 1-185, Florence 1993

22
Gorini, V. , Kossakowski, A. and Sudarshan, E. C. G. : "Completely positive dynamical semigroups of N-level systems", J. Math. Phys. 17 (1976), 821-825

23
Itzikson, C. and Zuber, J. -B. " Quantum Field Theory, McGraw-Hill Book Co. , New York 1980

24
Künzle, H. P. and Duval, C. : "Dirac Field on Newtonian Space-Time", Ann. Inst. H. Poincaré 41 (1984), 363-384

25
Landsman, N. P. : "Algebraic theory of superselection sectors and the measurement problem in quantum mechanics", Int. J. Mod. Phys. A6 (1991), 5349-5371

26
Lasota, A. and Mackey, M. C.: Chaos, Fractals and Noise. Stochastic Aspects of Dynamics, Springer Verlag, New York 1994

27
Lindblad, G. : "On the Generators of Quantum Mechanical Semigroups", Comm. Math. Phys. 48 (1976), 119-130

28
Lüders, G. : "Über die Zustandsänderung durch den Messprozzess", Annalen der Physik 8 (1951), 322-328

29
Modugno, M. : "Systems of vector valued forms on a fibred manifold and applications to gauge theories", Lect. Notes in Math. 1251, Springer Verlag, 1987

30
von Neumann, J. : Mathematical Foundations of Quantum Mechanics, Princeton Univ. Press, Princeton 1955

31
Ozawa, M.: "Cat Paradox for -Dynamical Systems", Progr. Theor. Phys. 88 (1992), 1051-1064

32
Parthasarathy, K. R. : An Introduction to Quantum Stochastic Calculus, Birkhäuser Verlag, Basel 1992

33
Sniatycki, J. : Geometric Quantization and Quantum Mechanics, Springer Verlag, New York 1980

Topics in Quantum Dynamics

This document was generated using the LaTeX2HTML translator Version 99.1 release (March 30, 1999)

Copyright © 1993, 1994, 1995, 1996, Nikos Drakos, Computer Based Learning Unit, University of Leeds.
Copyright © 1997, 1998, 1999, Ross Moore, Mathematics Department, Macquarie University, Sydney.

The command line arguments were:
latex2html -split 0 -local_icons 9406204.tex

#### Footnotes

... isolated1
Emphasized style will be used in these notes for concepts that are important, but will not be explained. Sometimes explanation would need too much space, but sometimes because these are either primitive or meta-language notions.
... observable.2
Lüders [28] noticed that this formulation is ambiguous in case of degenerate eigenvalues, and generalized it to cover also this situation.
... theory.3
In these lectures, "quantum theory" usually means "quantum mechanics", although much of the concepts that we discuss are applicable also to systems with infinite number of degrees of freedom and in particular to quantum field theory.
... other4
Including the author of these notes.
... relativity,5
There exists however so called "relativistic Fock-Schwinger proper time formalism" [23, Ch. 2-5-4] where one writes Schrödinger equation with Hamiltonian replaced by "super-Hamiltonian, and time replaced by "proper time"
... day.6
One could try to "explain" time by saying that there is a preferred time direction selected by the choice of a thermal state of the universe. But that is of no help at all, until we are told how it happens that a particular thermal state is being achieved.
... relativity.7
In group theoretical terms: the proper Lorentz group is simple, while the proper Galilei group is not.
... processes8
The paradigm may however change in no so distant future - we may soon try to understand the Universe as a computing machine, with geometry replaced by geometry of connections, and randomness replaced by a variant of algorithmic complexity.
... all.9
Cf. e. g. Ref. [33, Ch. 9].
.... 10
A similar idea was mentioned in [3]. For a detailed description of all the constructions - see the forthcoming book [21]
... confusion.11
Notable exceptions can be found in publications from the Genevé school of Jauch and Piron.
... etc.12
Quaternionic structures, on the other hand, can be always understood as complex ones with an extra structure - they are unnecessary.
... system13
Some physicists deny "objectivity" of quantum states - they would say that Hilbert space vectors describe not states of the system, but states of knowledge or information about the system. In a recent series of papers (see [1] and references therein) Aharonov and Vaidman [1] attempt to justify objectivity of quantum states. Unfortunately their arguments contain a loophole.
... case.14
It should be noted, however, that Schrödinger equation describes evolution of state vectors. and thus contains direct information about phases. This information is absent in the Liouville equation, and its restoration (e. g. as it is with the Berry phase) may sometimes create a non-trivial task.
...bla5.15
Cf. also the recent (June 1994) paper "Particle Tracks, Events and Quantum Theory", by the author.
...jamo.16
The reader may also consult [19], where a different approach, using dimensional reduction along a null Killing vector, is discussed.
.... 17
Some of these assumptions are superfluous as they would follow anyhow from the assumption in the next paragraph.
.... 18
Notice that because is not a tensor, the last condition need not be, a priori, generally covariant.
... universal.19
i.e. universal for the system of connections.
... .20
We choose the physical units in such a way that the Planck constant and mass of the quantum particle are equal to 1.
... factor.21
For an alternative detailed derivation see [12]
... 22
Our point is that "measurement" is an undefined concept in standard quantum theory, and that the probabilistic interpretation must be, because of that, brought from outside. What we propose is to define measurement as a CP semigroup coupling between a classical and a quantum system and to derive the probabilistic interpretation of the quantum theory from that of the classical one.
... evolution23
For a discussion of this fact in a broader context of algebraic theory of superselection sectors - cf. Landsman [25, Sec. 4. 4]. Cf. also the no-go result by Ozawa [31]
... system.24
That requirement is also necessary to guarantee physical consistency of the whole framework, as we always neglect some degrees of freedom as either irrelevant or yet unknown to us.
... extremal25
It should be noticed, however, that splitting of into and is, in general, not unique - cf. e. g. Refs [15] and [32, Ch. III. 29-30].
... 26
It is useful to have the algebra represented in such a form, as it enables us to apply the theorem of Christensen-Evans.
... 27
One can easily imagine a more general situation when tracing over the classical system will not be meaningful. This can happen if we deal with several phases of the quantum system, parameterized by the classical parameter . It may then happen that the total algebra is not the tensor product algebra. For instance, instead of one Hilbert space we may have, for each value of a Hilbert space of dimension .
... system28
One may wonder what does that mean mathematically, as the space of pure states of a algebra is, from measure-theoretical point of view, a rather unpleasant object. The answer is that the only measures on the space of pure states of the quantum algebra will be the Dirac measures.
...car,dal,dum,gar. 29
Thanks are due to N. Gisin for pointing out these references.
... transformations30
Or, more precisely, as Frobenius-Perron operators. Cf. Ref.[26] for definition and examples of Frobenius-Perron and dual to them Koopman operators.
.... 31
This sequence of transformation on the space of pure states of the quantum system can be thought of as a nonlinear version of Barnsley's Iterated Function System (cf. e. g. [4]
... details.32
I do not think that they found the answer, as their arguments are circular, and they seem to be well aware of this circularity.
... formula:33
The formula is easily established for of the form , and then extended to arbitrary operators by linearity.

.