zeh-2002-decoherence- basic concepts and their interpretation.9506020v3

8/10/2019 Zeh-2002-Decoherence- Basic Concepts and Their Interpretation.9506020v3

1/42

arXiv:quant-ph/9506020v3

30Jun2002

2 Basic Concepts and their Interpretation

H. D. Zeh (www.zeh-hd.de)

2.1 The Phenomenon of Decoherence

2.1.1 Superpositions

The superposition principle forms the most fundamental kinematical con-cept of quantum theory. Its universality seems to have first been postulated

by Dirac as part of the definition of his ket-vectors, which he proposed asa complete1 and general concept to characterize quantum states regardlessof any basisof representation. They were later recognized by von Neumannas forming an abstract Hilbert space. The inner product (also needed to de-fine a Hilbert space, and formally indicated by the distinction between braand ket vectors) is not part of the kinematics proper, but required for theprobability interpretation, which may be regarded as dynamics (as will bediscussed). The third Hilbert space axiom (closure with respect to Cauchyseries) is merely mathematically convenient, since one can never decide em-pirically whether the number of linearly independent physical states is infinitein reality, or just very large.

According to this kinematical superposition principle, any two physicalstates,

|1

and

|2

, whatever their meaning, can be superposed in the form

c1|1+c2|2, with complex numbersc1andc2, to form anew physical state(tobe be distinguished from a state of information). By induction, the principlecan be applied to more than two, and even an infinite number, of states, andappropriately generalized to apply to a continuum of states. After postulat-ing the linear Schrodinger equation in a general form, one may furthermore

Chapter 2 of D. Giulini, E. Joos, C. Kiefer, J. Kupsch, I.-O. Stamatescu, andH. D. Zeh: Decoherence and the Appearance of a Classical World in QuantumTheory, 2nd edn. (Springer-Verlag, 2003). In this draft version, references tootherchapters refer to the first edition (Springer, 1996).

1 This conceptualcompleteness does not, of course, imply that all degrees of free-dom of a considered system are always known and taken into account. It onlymeans that, within quantum theory (which, in its way, is able to describe allknown experiments), no more complete description of the system is required or

indicated. Quantum mechanics lets us even understand whywe may neglect cer-tain degrees of freedom, since gaps in the energy spectrum often freeze themout.
http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3http://arxiv.org/abs/quant-ph/9506020v3


2/42

2 H.D. Zeh (www.zeh-hd.de)

conclude that the superposition of two (or more) of its solutionsforms againa solution. This is the dynamicalversion of the superposition principle.

Let me emphasize that this superposition pinciple is in drastic contrast

to the concept of the quantum that gave the theory its name. Superposi-tions obeying the Schrodinger equation describe a deterministically evolvingcontinuum rather than discrete quanta and stochastic quantum jumps. Ac-cording to the theory of decoherence, these effectiveconcepts emerge as aconsequence of the superposition principle when universally and consistentlyapplied.

A dynamical superposition principle (though in general with respect toreal numbers only) is also known from classical waves which obey a linearwave equation. Its validity is then restricted to cases where these equationsapply, while the quantum superposition principle is meant to be universal andexact. However, while the physical meaning of classical superpositions is usu-ally obvious, that of quantum mechanical superpositions has to be somehowdetermined. For example, the interpretation of a superposition dq eipq |

q

asrepresenting a state of momentumpcan be derived from quantization rules,valid for systems whose classical counterparts are known in their Hamiltonianform (see Sect. 2.2). In other cases, an interpretation may be derived fromthe dynamics or has to be based on experiments.

Dirac emphasized another (in his opinion even more important) differ-ence: all non-vanishing components of (or projections from) a superpositionare in some sense contained in it. This formulation seems to refer to an en-sembleof physical states, which would imply that their description by formalquantum states is notcomplete. Another interpretation asserts that it isthe (Schrodinger)dynamicsrather than the concept of quantum states whichis incomplete. States found in measurements would then have toarisefrom aninitial state by means of an indeterministic collapse of the wave function.

Both interpretations meet serious difficulties when consistently applied (seeSect. 2.3).In the third edition of his textbook, Dirac (1947) starts to explain the su-

perposition principle by discussing one-particle states, which can be describedby Schrodinger waves in three-dimensional space. This is an important appli-cation, although its similarity with classical waves may also be misleading.Wave functions derived from the quantization rules are defined on their clas-sical configuration space, which happens to coincide with normal space onlyfor a single mass point. Except for this limitation, the two-slit interferenceexperiment, for example, (effectively a two-state superposition) is known tobe very instructive. Diracs second example, the superposition of two basicphoton polarizations, no longer corresponds to a spatial wave. These twobasic states contain all possible photon polarizations. The electron spin,

another two-state system, exhausts the group SU(2) by a two-valued repre-sentation of spatial rotations, and it can be studied (with atoms or neutrons)by means of many variations of the SternGerlach experiment. In his lecture


3/42

2 Basic Concepts and their Interpretation 3

notes (Feynman, Leighton, and Sands 1965), Feynman describes the masermode of the ammonia molecule as another (very different) two-state system.

All these examples make essential use of superpositions of the kind

|

=

c1|1 + c2|2, where the states|1,|2, and (all)|can be observed as phys-ically differentstates, and distinguished from one another in an appropriatesetting. In the two-slit experiment, the states|1 and|2 represent the par-tial Schrodinger waves that pass through one or the other slit. Schrodingerswave function can itself be understood as a consequence of the superposi-tion principle by being viewed as the amplitudes (q) in the superpositionof classical configurations q (now represented by corresponding quantumstates|qor their narrow wave packets). In this case of a system with a knownclassical counterpart, the superpositions|= dq (q)|q are assumed todefine all quantum states. They may represent new observable properties(such as energy or angular momentum), which are not simply functions ofthe configuration,f(q), only as a nonlocal whole, but not as an integral overcorresponding local densities (neither on space nor on configuration space).

Since Schrodingers wave function is thus defined on (in general high-dimensional) configuration space, increasing its amplitude does not describean increase of intensity or energy density, as it would for classical wavesin three-dimensional space. Superpositions of the intuitive product states ofcomposite quantum systems may not only describe particle exchange sym-metries (for bosons and fermions); in the general case they lead to the fun-damental concept ofquantum nonlocality. The latter has to be distinguishedfrom a mereextensionin space (characterizing extended classical objects). Forexample, molecules in energy eigenstates are incompatible with their atomsbeing in definite quantum states themselves. Although the importance of thisentanglement for many observable quantities (such as the binding energyof the helium atom, or total angular momentum) had been well known, its

consequence of violating Bells inequalities (Bell 1964) seems to have sur-prised many physicists, since this result strictly excluded all local theoriesconceivably underlying quantum theory. However, quantum nonlocality ap-pears paradoxical only when one attempts to interpret the wave functionin terms of an ensemble of local properties, such as particles. If realitywere definedto be local (in space and time), then it would indeed conflictwith the empirical actuality of a general superposition. Within the quantumformalism, entanglement also leads to decoherence, and in this way it ex-plainsthe classical appearance of the observed world in quantum mechanicalterms. The application of this program is the main subject of this book (seealso Zurek 1991, Mensky 2000, Tegmark and Wheeler 2001, Zurek 2001, orwww.decoherence.de).

The predictive power of the superposition principle became particularly

evident when it was applied in an ingenious step to postulate the existenceof superpositions of states with different particle numbers (Jordan and Klein1927). Their meaning is illustrated, for example, by coherent states of dif-


4/42


ferent photon numbers, which may represent quasi-classical states of the elec-tromagnetic field (cf. Glauber 1963). Such dynamically arising (and in manycases experimentally confirmed) superpositions are often misinterpreted as

representing virtual states, or mere probability amplitudes for the occur-rence of real states that are assumed to possess definite particle number.This would be as mistaken as replacing a hydrogen wave function by theprobability distribution p(r) =|(r)|2, or an entangled state by an ensem-bleof product states (or a two-point function). A superposition is in generalobservably different from an ensemble consisting of its components with cor-responding probabilities.

Another spectacular success of the superposition principle was the pre-diction of new particles formed as superpositions of K-mesons and their an-tiparticles (Gell-Mann and Pais 1955, Lee and Yang 1956). A similar modeldescribes the recently confirmed neutrino oscillations (Wolfenstein 1978),which are superpositions of energy eigenstates.

The superposition principle can also be successfully applied to states thatmay be generated by means of symmetry transformations from asymmet-ric ones. In classical mechanics, a symmetric Hamiltonian means that eachasymmetricsolution(such as an elliptical Kepler orbit) implies other solu-tions, obtained by applying the symmetry transformations (e.g. rotations).Quantum theory requires in additionthat all their superpositions also formsolutions (cf. Wigner 1964, or Gross 1995; see also Sect. 9.6). A complete setof energy eigenstates can then be constructed by means of irreducible linearrepresentationsof the dynamical symmetry group. Among them are usuallysymmetric ones (such as s-waves for scalar particles) that need not have acounterpart in classical mechanics.

A great number of novel applications of the superpositon principle havebeen studied experimentally or theoretically during recent years. For exam-

ple, superpositions of different classical states of laser modes (mesoscopicSchrodinger cats) have been prepared (Monroe et al. 1996), the entangle-ment of photon pairs has been confirmed to persist over tens of kilometers(Tittel et al. 1998), and interference experiments with fullerene moleculeswere successfully performed (Arndt et al. 1999). Even superpositions of amacroscopic current running in opposite directions have been shown to ex-ist, and confirmed to be different from a state with two (cancelling) currents(Mooij et al. 1999, Friedman et al. 2000). Quantum computers, now underintense investigation, would have to perform parallel (but notspatiallysep-arated) calculations, while forming one superposition that may later have acoherent effect. So-called quantum teleportation requires the advanced prepa-ration of an entangled state of distant systems (cf. Busch et al. 2001 for aconsistent description in quantum mechanical terms). One of its components

may then later be selected by a localmeasurement in order to determine thestate of the other (distant) system.


5/42


Whenever an experiment was technically feasible, all components of asuperposition have been shown to act coherently, thus proving that theyexistsimultaneously. It is surprising that many physicists still seem to regard

superpositions as representing some state of ignorance (merely characterizingunpredictable events). After the fullerene experiments there remains buta minor step to discuss conceivable (though hardly realizable) interferenceexperiments with a conscious observer. Would he have one or many minds(when being aware of his path through the slits)?

The most general quantum states seem to be superpositions of differ-ent classical fields on three- or higher-dimensional space.2 In a perturbationexpansion in terms of free particles (wave modes) this leads to terms cor-responding to Feynman diagrams, as shown long ago by Dyson (1949). Thepath integral describes a superpositionof paths, that is, the propagation ofwave functionals according to a generalized Schrodinger equation, while theindividual paths under the integral have no physicalmeaning by themselves.(A similar method could be used to describe the propagation of classicalwaves.) Wave functions will here always be understood in the generalizedsense ofwave functionals if required.

One has to keep in mind this universality of the superposition princi-ple and its consequences for individuallyobservable physical properties inorder to appreciate the meaning of the program of decoherence. Since quan-tum coherence is far more than the appearance of spatial interference fringesobserved statistically in series of events, decoherence must not simply beunderstood in a classical sense as their washing out under fluctuating envi-ronmental conditions.

2.1.2 Superselection Rules

In spite of this success of the superposition principle it soon became evi-dent that not all conceivable superpositions are found in Nature. This ledsome physicists to postulate superselection rules, which restrict this prin-ciple by axiomatically excluding certain superpositions (Wick, Wightman,

2 The empirically correct pre-quantum configurations for fermions are given byspinor fields on space, while the apparently observed particles are no more thanthe consequence of decoherence by means of local interactions with the environ-ment (see Chap. 3). Field amplitudes (such as (r)) seem to form the generalarguments of the wave function(al) , while space points r appear as their in-dices not as dynamical position variables. Neither a second quantization nora wave-particle dualism are required. N-particle wave functions may be obtainedas a non-relativistic approximation by applying the superposition principle (asa quantization procedure) to these apparent particles instead of the correct

pre-quantum variables (fields), which are not directly observable for fermions.The concept of particle permutations then becomes a redundancy(see Sect. 9.6).Unified field theories are usually expected to provide a general (supersymmetric)pre-quantum field and its Hamiltonian.


6/42


and Wigner 1970, Streater and Wightman 1964). There are also attempts toderivesome of these superselection rules from other principles, which can bepostulated in quantum field theory (see Chaps. 6 and 7). In general, these

principles merely exclude unwanted consequences of a general superposi-tion principle by hand.

Most disturbing in this sense seem to be superpositions of states withinteger and half-integer spin (bosons and fermions). They violate invarianceunder 2-rotations (see Sect. 6.2), but such a non-invariance has been exper-imentally confirmed in a different way (Rauch et al. 1975). The theory ofsupersymmetry (Wess and Zumino 1971) postulates superpositions of bosonsand fermions. Another supposedly fundamental superselection rule forbidssuperpositions of different charge. For example, superpositions of a protonand a neutron have never been directly observed, although they occur inthe isotopic spin formalism. This (dynamically broken) symmetry was latersuccessfully generalized to SU(3) and other groups in order to characterizefurther intrinsic degrees of freedom. However, superpositions of a proton anda neutron may exist within nuclei, where isospin-dependent self-consistentpotentials may arise from anintrinsic symmetry breaking. Similarly, superpo-sitions of different charge are used to form BCS states (Bardeen, Cooper, andSchrieffer 1957), which describe the intrinsic properties of superconductors.In these cases, definite charge values have to be projected out (see Sect. 9.6)in order to describe the observed physical ob jects, which do obey the chargesuperselection rule.

Other limitations of the superposition principle are less clearly defined.While elementary particles are described by means of wave functions (thatis, superpositions of different positions or other properties), the moon seemsalways to be at a definite place, and a cat is either dead or alive. A generalsuperposition principle would even allow superpositions of a cat and a dog (as

suggested by Joos). They would have to define a new animal analogousto a Klong, which is a superposition of a K-meson and its antiparticle. In theCopenhagen interpretation, this difference is attributed to a strict conceptualseparation between the microscopic and the macroscopic world. However,where is the border line that distinguishes an n-particle state of quantummechanics from an N-particle state that is classical? Where, precisely, doesthe superposition principle break down?

Chemists do indeed know that a border line seems to exist deep in themicroscopic world (Primas 1981, Woolley 1986). For example, most molecules(save the smallest ones) are found with their nuclei in definite (usually ro-tating and/or vibrating) classical configurations, but hardly ever in super-positions thereof, as it would be required for energy or angular momentumeigenstates. The latter are observed for hydrogen and other smallmolecules.

Even chiral states of a sugar molecule appear classical, in contrast to itsparity and energy eigenstates, which correctly describe the otherwise analo-gous maser mode states of the ammonia molecule (see Sect. 3.2.4 for details).


7/42


8/42


experience that the interference disappears also when the passage is mea-sured without registration of a definite result. The latter may be assumedtohave become a classical fact as soon the measurement has irreversibly oc-

curred. A quantum phenomenon may thus become a phenomenon withoutbeing observed (in contrast to this early formulation of Bohrs, which is inaccordance with Heisenbergs idealistic statement about a trajectory cominginto being by its observation while Bohr later spoke of objective irreversibleevents occurring in the counter). However, what presicely is an irreversiblequantum event? According to Bohr, it can notbe dynamically analyzed.

Analysis within the quantum mechanical formalism demonstrates nonethe-less that the essential condition for this decoherence is that complete infor-mation about the passage is carried away in some physicalform (Zeh 1970,1973, Mensky 1979, Zurek 1981, Caldeira and Leggett 1983, Joos and Zeh1985). Possessing information here means that the physical state of theenvironment is now uniquely quantum correlated(entangled) with the rele-vant property of the system (such as a passage through a specific slit). Thisneednothappen in a controllable form (as in a measurement): the informa-tion may as well be created in the form of noise. However, in contrast tostatistical correlations, quantum correlations definepure(completly defined)nonlocal states, and thus individual physicalproperties, such as the total spinof spatially separated objects. Therefore, one cannotexplainentanglement interms of the concept of information (cf. Brukner and Zeilinger 2000). Thisterminology would mislead to the popular misunderstanding of the collapseas a mere increase of information (which would require an initial ensem-ble describing ignorance). Since environmental decoherence affects individualphysical states, it can neitherbe the consequence of phase averaging in anensemble, norone of phases fluctuating uncontrollably in time (as claimedin some textbooks). For example, nonlocal entanglement exists in the static

quantum state of a relativstic physical vacuum (even though it is then oftenvisualized in terms of particles as vacuum fluctuations).When is unambiguous information carried away? If a macroscopic ob-

ject had the opportunity of passing through two slits, we would always beable to convince ourselves of its choice of a path by simply opening our eyesin order to look. This means that in this case there is plenty of light thatcontains information about the path (even in a controllable manner that al-lows looking). Interference between different paths never occurs, since the

order to remain indistinguishable from quantum theory). The fact that these twoquite different situations (closing slit 2 or measuring the passage through slit 1)lead to exactly the same subsequent frequencies, which differ entirely from thosethat are definedby this theory when not measured or selected, emphasizes itsextremely artificial nature (see also Englert et al. 1992, or Zeh 1999). The predic-

tions of quantum theory are here simply reproduced by leaving the Schr odingerequation unaffected and universally valid, identical with Everetts assumptions(Everett 1957). In both these theories the wave function is (for good reasons)regarded as a realphysical object (cf. Bell 1981).


9/42


path is evidently continuously measured by light. The common textbookargument that the interference pattern of macroscopic ob jects be too fine tobe observable is entirely irrelevant. However, would it then not be sufficient

to dim the light in order to reproduce (in principle) a quantum mechanicalinterference pattern for macroscopic objects?

This could be investigated by means of more sophisticated experimentswith mesoscopic objects (see Bruneet al.1996). However, in order to preciselydetermine the subtle limit where measurement by the environment becomesnegligible, it is more economic first to apply the established theory which isknown to describe such experiments. Thereby we have to take into accountthe quantum nature of the environment, as discussed long ago by Brillouin(1962) for an information medium in general. This can usually be done easily,since the quantum theory of interacting systems, such as the quantum the-ory of particle scattering, is well understood. Its application to decoherencerequires that one averages over all unobserved degrees of freedom. In tech-nical terms, one has to trace out the environment after it has interactedwith the considered system. This procedure leads to a quantitative theory ofdecoherence (cf. Joos and Zeh 1985). Taking the trace is based on the prob-ability interpretation applied to the environment (averaging over all possibleoutcomes of measurements), even though this environment is notmeasured.(The precise physical meaning of these formal concepts will be discussed inSect. 2.4.)

Is it possible to explain all superselection rules in this way as an effectinduced by the environment4 including the existence and position of theborder line between microscopic and macroscopic behaviour in the realmof molecules? This would mean that the universality of the superpositionprinciple could be maintained as is indeed the basic idea of the program ofdecoherence(Zeh 1970, Zurek 1982; see also Chap. 4 of Zeh 2001). If physical

states are thus exclusively described by wave functions rather than by pointsin configuration space as originally intended by Schrodinger in space bymeans of narrow wave packets instead of particles then no uncertaintyrelations are availablefor statesin order to explain the probabilistic aspectsof quantum theory: the Fourier theorem applies to a givenwave function(al).

As another example, consider two states of different charge. They inter-act very differently with the electromagnetic field even in the absence ofradiation: their Coulomb fields carry complete information about the totalchargeat any distance. The quantum state of this field would thus decoherea superposition of different charges if considered as a quantum system in aboundedregion of space (Giulini, Kiefer, and Zeh 1995). This instantaneousaction of decoherence at an arbitrary distance by means of the Coulomb fieldgives it the appearance of a kinematical effect, although it is based on the

4 It would be sufficient, for this purpose, to use an internalenvironment (un-observed degrees of freedom), but the assumption of a closed system would ingeneral be unrealistic.


10/42

10 H. D. Zeh (www.zeh-hd.de)

dynamical law of charge conservation, compatible with a retardedfield thatwould measure the charge (see Sect. 6.4).

There are many other cases where the unavoidable effect of decoherence

can easily be imagined without any calculation. For example, superpositionsof macroscopically different electromagnetic fields, f(r), may be describedby a field functional [f(r)]. However, any charged particle in a sufficientlynarrow wave packet would then evolve into different packets, depending onthe field f, and thus become entangled with the state of the quantum field(Kubler and Zeh 1973, Kiefer 1992, Zurek, Habib, and Paz 1993; see alsoSect. 4.1.2). The particle can be said to measure the quantum state of thefield. Since charged particles are in general abundant in the environment, nosuperpositions of macroscopically different electromagnetic fields (or differentmean fields in other cases) are observed under normal conditions. Thisresult is related to the difficulty of preparing and maintaining squeezedstates of light (Yuen 1976) see Sect. 3.3.3.1. Therefore, the field appearstobein one of its classical states (Sect. 4.1.2).

In all these cases, this conclusion requires that the quasi-classical states(or pointer states in measurements) are robust (dynamically stable) undernatural decoherence, as pointed out already in the first paper on decoherence(Zeh 1970; see also Diosi and Kiefer 2000).

A particularly important example of a quasiclassical field is the metricof general relativity (with classical statesdescribed by spatial geometries onspace-like hypersurfaces see Sect. 4.2). Decoherence caused by all kindsof matter can therefore explain the absence of superpositions of macroscop-ically distinct spatial curvatures (Joos 1986, Zeh 1986, 1988, Kiefer 1987),while microscopic superpositions would describe those hardly ever observ-able gravitons.

Superselection rules thus arise as a straightforward consequence of quan-

tum theory under realistic assumptions. They have nonetheless been dis-cussed mainly in mathematical physics apparently under the influence ofvon Neumanns and Wigners orthodox interpretation of quantum mechan-ics (see Wightman 1995 for a review). Decoherence by continuous measure-ment seems to form the most fundamental irreversible process in Nature. Itapplies even where thermodynamical concepts do not(such as for individualmolecules see Sect. 3.2.4), or when any exchange of heat is entirely negligi-ble. Its time arrow of microscopic causality requires a Sommerfeld radiationcondition for microscopic scattering (similar to Boltzmanns chaos), viz., theabsence of any dynamically relevant initial correlations, which would definea conspiracy in common terminology (Joos and Zeh 1985, Zeh 2001).


11/42


2.2 Observables as a Derived Concept

Measurements are usually described by means of observables, formally rep-

resented by Hermitian operators, and introduced in addition to the conceptsof quantum states and their dynamics as a fundamental and independentingredient of quantum theory. However, even though often forming the start-ing point of a formal quantization procedure, this ingredient should not beseparately required if physical states are well described by these formal quan-tum states. This understanding, to be further explained below, complies withJohn Bells quest for the replacement of observables with beables (see Bell1987). It was for this reason that his preference shifted from Bohms theoryto collapse models (where wave functions are assumed to completely describereality) during his last years.

Let| be an arbitrary quantum state, defined operationally (up to acomplex numerical factor) by a complete preparation procedure. The phe-nomenological probability for finding the system during an appropriate mea-surement in another quantum state|n, say, is given by means of their innerproduct aspn=|n||2 (where both states are assumed to be normalized).The state|nis here defined by the specific measurement. (In a position mea-surement, for example, the number n has to be replaced with the continuouscoordinates x, y, z, leading to the improper Hilbert states|r.) For measure-ments of the first kind (to which all others can be approximately reduced see Sect. 2.3), the system will again be found in the state|nwith certaintyif the measurement is immediately repeated. Preparationscan be regarded assuch measurements whichselecta certain subset of outcomes for further mea-surements. n-preparations are therefore also calledn-filters, since all not-nresults are thereby excluded from the subsequent experiment proper. Theabove probabilities can also be written in the form pn =| Pn| , withan observable Pn :=|nn|, which is thus derived from the kinematicalconcept of quantum states.

Instead of these special n or not-n measurements (with fixed n), onecan also perform more general n1 or n2 or . . . measurements, with allnismutually exclusive (ni|nj= ij). If the states forming such a set{|n} arepure and exhaustive (that is, complete,

Pn= 1l), they represent a basis of

the corresponding Hilbert space. By introducing an arbitrary measurementscale an, one may construct general observables A =

|nann|, whichpermit the definition of expectation values| A| = pnan. In thespecial case of a yes-no measurement, one has an = nn0, and expectationvalues become probabilities. Finding the state|n during a measurement isthen also expressed as finding the value an of an observable. A changeof scale, bn = f(an), describes the samephysical measurement; for position

measurements of a particle it would simply represent a coordinate transfor-mation. Even a measurement of the particles potential energy is equivalentto a position measurement (up to degeneracy) if the function V(r) is given.


12/42


According to this definition, quantum expectation values must not beunderstood as mean values in an ensemble that represents ignorance of theprecise state. Rather, they have to be interpreted as probabilities for poten-

tially arising quantum states|n regardless of the latters interpretation.If the set{|n} of such potential states forms a basis, any state| can berepresented as a superposition|= cn|n. In general, it neither forms ann0-state nor any not-n0 state. Its dependence on the complex coefficients cnrequires that states which differ from one another by a numerical factor mustbe different in reality. This is true even though they represent the sameray in Hilbert space and cannot, according to the measurement postulate,be distinguished operationally. The states|n1+|n2and|n1 |n2could notbe physically different from another if|n2 and|n2 were the samestate.(Only aglobalnumerical factor would be redundant.) For this reason, pro-

jection operators|nn| are insufficient to characterize quantum states (cf.also Mirman 1970).

The expansion coefficients cn, relating physically meaningful states forexample those describing different spin directions or different versions of theK-meson must in principle be determined (relative to one another) by ap-propriate experiments. However, they can often be derived from a previouslyknown (or conjectured) classical theory by means of quantization rules.In this case, the classical configurations q(such as particle positions or fieldvariables) arepostulatedto parametrize a basis in Hilbert space,{|q}, whilethe canonical momenta p parametrize another one,{|p}. Their correspond-ing observables, Q =

dq|qqq| and P = dp |ppp|, are required to obey

commutation relations in analogy to the classical Poisson brackets. In thisway, they form an important tool for constructing and interpreting the spe-cific Hilbert space of quantum states. These commutators essentially deter-mine the unitary transformationp| q (e.g. as a Fourier transform e ipq) thus more than what could be defined by means of the projection operators|qq| and|pp|. This algebraic procedure is mathematically very elegantand appealing, since the Poisson brackets and commutators may representgeneralized symmetry transformations. However, the conceptof observables(which form the algebra) can be derived from the more fundamental one ofstate vectors and their inner products, as described above.

Physical states are assumed to vary in time in accordance with a dynam-ical law in quantum mechanics of the form it| = H|. In contrast,a measurement device is usually defined regardless of time. This must thenalso hold for the observable representing it, or for its eigenbasis{|n}. Theprobabilitiespn(t) =|n| (t)|2 will therefore vary with time according tothe time-dependence of the physical states|. It is well known that this(Schrodinger) time dependence is formally equivalent to the (inverse) time

dependence of observables (or the reference states |n). Since observablescorrespond to classical variables, this time dependence appeared sugges-tive in the HeisenbergBornJordan algebraic approach to quantum theory.


13/42


However, the absence ofdynamical states|(t)from this Heisenberg picture,a consequence of insisting on classicalkinematical concepts, leads to para-doxes and conceptual inconsistencies (complementarity, dualism, quantum

logic, quantum information, and all that).An environment-induced superselection rule means that certain superpo-

sitions are highly unstable with respect to decoherence. It is then impossiblein practice to construct measurement devices for them. This empiricalsitua-tion has led some physicists todeny the existenceof these superpositions andtheir corresponding observables either by postulate or by formal manipu-lations of dubious interpretation, often including infinities. In an attempt tocircumvent the measurement problem (that will be discussed in the follow-ing section), they often simply regardsuch superpositions as mixtures oncethey have formed according to the Schrodinger equation (cf. Primas 1990).

Whileanybasis{|n} in Hilbert space defines formal probabilities, pn =|n||2, only a basis consisting of states that are not immediately destroyedby decoherence defines a practically realizable observable. Since realizableobservables usually form a genuine subset ofallformal observables (diagonal-izable operators), they must contain a nontrivial center in algebraic terms.It consists of those of them which commute with all the rest. Observablesforming the center may be regarded as classical, since they can be mea-sured simultaneously with allrealizable ones. In the algebraic approach toquantum theory, this center appears as part of its axiomatic structure (Jauch1968). However, since the condition of decoherence has to be considered quan-titatively (and may even vary to some extent with the specific nature of theenvironment), this algebraic classification remains an approximate and dy-namically emerging scheme.

These classical observables thus define the subspaces into which super-positions decohere. Hence, even if the superposition of a right-handed and a

left-handed chiral molecule, say, couldbe prepared by means of an appropri-ate (very fast) measurement of the first kind, it would be destroyed beforethe measurement may be repeated for a test. In contrast, the chiral statesof all individual molecules in a bag of sugar are robust in a normal envi-ronment, and thus retain this property individuallyover time intervals whichby far exceed thermal relaxation times. This stability may even be increasedby the quantum Zeno effect (Sect. 3.3.1). Therefore, chirality appears notonly classical, but also as an approximate constant of the motion that has tobe taken into account in the definition of thermodynamical ensembles (seeSect. 2.3).

The above-used description of measurements of the first kind by meansof probabilities for transitions| |n(or, for that matter, by correspond-ing observables) is phenomenological. However, measurements should be de-

scribed dynamically as interactions between the measured system and themeasurement device. The observable (that is, the measurement basis) shouldthus be derived from the corresponding interaction Hamiltonian and the ini-


14/42


tial state of the device. As discussed by von Neumann (1932), this interactionmust be diagonal with respect to the measurement basis (see also Zurek 1981).Its diagonal matrix elements are operators which act on the quantum state of

the device in such a way that the pointer moves into a position appropriatefor being read,|n|0 |n|n. Here, the first ket refers to the system,the second one to the device. The states|n, representing different pointerpositions, must approximately be mutually orthogonal, and classical in theexplained sense.

Because of the dynamical superposition principle, an initial superpositioncn|ndoesnotlead to definite pointer positions (with their empirically ob-

served frequencies). If decoherence is neglected, one obtains their entangledsuperposition

cn|n|n, that is, a state that is different from all poten-

tial measurement outcomes|n|n. This dilemma represents the quantummeasurement problem to be discussed in Sect.2.3. Von Neumanns inter-action is nonetheless regarded as the first step of a measurement (a pre-measurement). Yet, a collapse seems still to be required now in the mea-surement device rather than in the microscopic system. Because of the en-tanglement between system and apparatus, it then affects the total system.5

If, in a certain measurement, a whole subset of states|n leads to thesame pointer position|n0, these states are not distinguished in this mea-surement. The pointer state|n0now becomes dynamically correlated withthe wholeprojectionof the initial state,

cn|n, on the subspace spanned by

this subset. A correspondingcollapsewas indeed postulated by Luders (1951)in his generalization of von Neumanns first intervention (Sect. 2.3).

In this dynamical sense, the interaction with an appropriate measuringdevicedefines an observable up to arbitrary monotoneous scale transforma-tions. The time dependence of observables according to the Heisenberg pic-ture would thus describe an imaginary time dependence of the states of this

device (its pointer states), paradoxically controlled by the intrinsic Hamilto-nian of the system.The question of whether a formal observable (that is, a diagonalizable

operator) can be physically realized can only be answered by taking into ac-count the unavoidable environment of the system (while the measurementdevice is alwaysasssumed to decohere into its macroscopic pointer states).However, environment-induced decoherence by itself does not solve the mea-surement problem, since the pointer states|nmay be assumed to includethe total environment (the rest of the world). Identifying the thus arising

5 Some authors seem to have taken the phenomenological collapse in the micro-scopic system by itself too literally, and therefore disregarded the state of themeasurement device in their measurement theory (see Machida and Namiki 1980,Srinivas 1984, and Sect. 9.1). Their approach is based on the assumption that

quantum states must always exist for all systems. This would be in conflict withquantum nonlocality, even though it may be in accordance with early interpre-tations of the quantum formalism.


15/42


global superposition with an ensembleof states, represented by a statisticaloperator , that merely leads to the same expectation valuesA = tr(A)for a limited set of observables

{A

}would obviously beg the question. This

argument is nonetheless found wide-spread in the literature (cf. Haag 1992,who used the subset of all local observables).

In Sect. 2.4, statistical operators will be derived from the concept ofquantum states as a tool for calculating expectation values, while the latterare defined, as described above, by means of probabilities for the occurrenceof new states in measurements. In the Heisenberg picture, is often regardedas in some sense representing the ensemble of potential values for all ob-servables that are here postulated to formally replace the classical variables.This interpretation is suggestive because of the (incomplete) formal analogyof to a classical phase space distribution. However, the prospective valueswould be physically meaningful only if they characterized different physicalstates (such as pointer states). Note that Heisenbergs uncertainty relationsrefer to potential outcomes which may arise in different (mutually exclusive)measurements.

2.3 The Measurement Problem

The superposition of different measurement outcomes, resulting according tothe Schrodinger equation (as discussed above), demonstrates that a naiveensemble interpretation of quantum mechanics in terms of incomplete knowl-edge is ruled out. It would mean that a quantum state (such as

cn|n|n)

represents an ensemble of some as yet unspecified fundamental states, ofwhich a subensemble (for example represented by the quantum state|n|n)may be picked out by a mere increase of information. If this were true, thenthe subensemble resulting from this measurement could in principle be traced

back in time by means of the Schrodinger equation in order to determine alsothe initial state more completely (to postselect it see Aharonov and Vaid-man 1991 for an inappropriate attempt). In the above case this would lead tothe initial quantum state|n|0that is physically differentfrom and thusinconsistent with the superposition (

cn|n)|0 that had been prepared

(whatever it means).In spite of this simple argument, which demonstrates that an ensemble

interpretation would require a complicated and miraculous nonlocal back-ground mechanism in order to work consistently (cf. Footnote 3 regardingBohms theory), the ensemble interpretation of the wave function seems toremain the most popular one because of its pragmatic (though limited) value.A general and rigorous critical discussion of problems arising in an ensembleinterpretation may be found in dEspagnats books (1976 and 1995).

A way out of this dilemma in terms of the wave function itself requiresone of the following two possibilities: (1) a modification of the Schrodingerequation that explicitly describes a collapse (also called spontaneous local-


16/42


ization see Chap. 8), or (2) an Everett type interpretation, in which allmeasurement outcomes are assumed to coexist in one formal superposition,but to beperceivedseparately as a consequence of their dynamical decoupling

under decoherence. While this latter suggestion may appear extravagant(as it requires myriads of coexisting parallel quasi-classical worlds), it issimilar in principle to the conventional (though nontrivial) assumption, madetacitly in all classical descriptions of observations, that consciousness is local-ized in certain (semi-stable and suffiently complex)spatial subsystemsof theworld (such as human brains or parts thereof). For a dispute about which ofthe above-mentioned two possibilities should be preferred, the fact that en-vironmental decoherence readily describes precisely the apparentlyoccurringquantum jumps or collapse events (as will be discussed in great detailthroughout this book) appears most essential.

The dynamical rules which are (explicitly or tacitly) used to describe theeffective time dependence of quantum states thus represent a dynamicaldualism. This was first clearly formulated by von Neumann (1932), whodistinguished between the unitary evolution according to the Schrodingerequation (remarkably his zweiter Eingriff or second intervention),

ih

t|= H| , (2.1)

valid for isolated (absolutely closed) systems, and the reduction or collapseof the wave function,

|=

cn|n |n0 (2.2)

(remarkably his firstintervention). The latter was to describe the stochas-tic transitions into the new state|n0during measurements. Their dynamicaldiscontinuity had been anticipated by Bohr in the form of quantum jumpsbetween his discrete electron orbits. Later, the time-dependentSchrodingerequation (2.1) for interacting systems was often regarded merely as a methodof calculating probabilities for similar (individually unpredictable) discontin-uous transitions between energy eigenstates (stationary quantumstates) ofatomic systems (Born 1926).6 However, there does not seem to be any mean-ingful difference between quantum jumps into new states or subspaces andthe occurrence of values for certain observables (cf. Sect. 2.2).

6 Thus also Bohr (1928) in a subsection entitled Quantum postulate and causal-ity about the quantum theory: . . . its essence may be expressed in the so-called quantum postulate, which attributes to any atomic process an essentialdiscontinuity, or rather individuality, completely foreign to classical theories andsymbolized by Plancks quantum of action (my italics). The later revision of

these early interpretations of quantum theory (required by the important role ofentangled quantum states for much larger systems) seems to have gone unnoticedby many physicists.


17/42


In scattering theory, one usually probes only partof quantum mechanicsby restricting consideration to asymptotic states and their probabilities (dis-regarding their superpositions). All quantum correlations between them then

appear statistical (classical). Occasionally even the unitary scattering am-plitudesmout|nin=m| S|nare confused with the probabilityamplitudesm|nwhich describe measurements to find a state|min an initial|n.In his general S-matrix theory, Heisenberg temporarily speculated about de-riving the latter from the former. Since macroscopic systems never becomeasymptotic because of their dynamical entanglment with the environment,they can not be described by an S-matrix at all.

The Born/von Neumann dynamical dualism was evidently the major mo-tivation for an ignorance interpretation of the wave function, which attemptsto explain the collapse not as a dynamical process in the system, but asan increase ofinformationabout it (the reduction of an ensemble of pos-sible states). However, even though the dynamics of ensembles in classicaldescription uses a formally similar dualism, an analogous interpretation inquantum theory leads to the severe (and apparently fatal) difficulties indi-cated above. They are often circumvented by the invention of new rules oflogic and statistics, which are notbased on any ensemble interpretation orincomplete information.

If the state of a classical system is incompletely known, and the cor-responding point p,q in phase space therefore replaced by an ensemble (aprobability distribution) (p, q), this ensemble can be reduced by a newobservation that leads to increased information. For this purpose, the systemmust interact in a controllable manner with the observer who holds theinformation (cf. Szilard 1929). His physical state of memory must therebychange in dependence on the property-to-be-measured of the observed sys-tem, leaving the latter unchanged in the ideal case (no recoil). Accord-

ing to deterministic dynamical laws, the ensemble entropy of the combinedsystem, which initially contains the entropy corresponding to the unknownmicroscopic quantity, would remain constant if it were defined to include theentropy characterizing the final ensemble of different outcomes. Since the ob-server is assumed to know (to be aware of) his own state, this ensembleis reduced correspondingly, and the ensemble entropy defined with respect tohis state of informationis lowered.

This is depicted by the first step of Fig. 2.1, where ensembles of statesare represented by areas. In contrast to many descriptions of Maxwells de-mon, the observer (regarded as a device) is here subsumed into the ensembledescription.Physical entropy, unlike ensemble entropy, is usually understoodas alocal(additive) concept, which neglects long range correlations for beingirrelevant, and thus approximately defines anentropy density. Physical and

ensemble entropy are equal in the absence of correlations. The informationI, given in the figure, measures the reduction of entropy according to theincreased knowledge of the observer.


18/42


a A O

b B

o r (by reduction of ensemble)

O

B'

A'

b

aO

reset

( forget)

Sensemble= So

Sphysical = So + kln2

I = 0

a

bO

Sensemble=: So

Sphysical = So

I = 0

measurement

O

system observer environment

Sensemble= So - kln2 (each)

Sphysical = So - kln2

I = kln2

Fig.2.1. Entropy relative to the state of information in an ideal classical mea-surement. Areas representsetsof microscopic states of the subsystems (while thoseof uncorrelated combined systems would be represented by their direct products).During the first step of the figure, the memory state of the observer changes de-terministically from 0 to A or B, depending on the state a or b of the system tobe measured. The second step depicts a subsequent reset, required if the measure-ment is to be repeated with the samedevice (Bennett 1973). A andB are effectswhich must thereby arise in the thermal environment in order to preserve the to-tal ensemble entropy in accordance with presumed microscopic determinism. Thephysical entropy (defined to add for subsystems) measures the phase space ofall microscopic degrees of freedom, including the property to be measured, whiledepending on givenmacroscopic variables. Because of its presumed additivity, this

physical entropy neglects all remaining statistical correlations (dashed lines, whichindicate sums of products of sets) for being irrelevant in the future henceSphysical Sensemble.Iis the amount of information held by the observer. The min-imum initial entropy,S0, isk ln 2 in this simple case of two equally probable valuesaand b.

This description does not necessarily require a conscious observer (al-though it may ultimately rely upon him). It applies to any macroscopic mea-surement device, since physical entropy is not only defined to be local, butalso relative to given macroscopic properties (as a function of them). Thedynamical part of the measurement transforms physical entropy (here theensemble entropy of the microscopic variables) deterministically into entropyof lacking information about controllable macroscopic properties. Before theobservation is taken into account (that is, before the or is applied), bothparts of the ensemble after the first step add up to give the ensemble entropy.When it istaken into account (as done by the numbers given in the figure),


19/42


the ensemble entropy is reduced according to the information gained by theobserver.

Any registration of information by the observer must use up his memory

capacity (blank paper), which represents non-maximal entropy. If the samemeasurement is to be repeated, for example in a cyclic process that could beused to transform heat into mechanical energy (Szilard, l.c.), this capacitywould either be exhausted at some time, or an equivalent amount of entropymust be absorbed by the environment (for example in the form of heat) inorder toresetthe measurement or registration device (second step of Fig. 2.1).The reason is that two different states cannot deterministically evolve intothe same final state (Bennett 1973).7 This argument is based on an arrow oftime of causality, which requires that all correlations possess local causesin their past (no conspiracy). The irreversible formation of irrelevantcorrelations then explains the increase ofphysical(local) entropy, while theensemble entropy is conserved.

The unsurmountable problems encountered in an ensemble interpretationof thewave function(or of any other superposition, such as|a + |b) are re-flected by the fact that there is no ensemble entropy that would represent theunknown property-to-be-measured (see the first step of Fig. 2.2 or 2.3 cf.also Zurek 1984). The ensemble entropy is nowdefinedby the correspond-ing expression Sensemble =ktr{ ln } (but see Sect. 2.4 for the meaningof the density matrix ). If the entropy of observer plus environment is thesame as in the classical case of Fig. 2.1, the total initial ensemble entropy isnow lower; in the case of equal initial probabilities fora and b it isS0k ln2.It would even vanish for pure states and of observer and environment,respectively: (|a+|b)|0|0. The Schrodinger evolution (depicted in Fig.2.3) would then be described by three dynamical steps,

(

|a

+

|b

)

|0

|0

(

|a

|A

+

|b

|B

)

|0

|a |A |A+|b |B |B(|a |AA+|b |BB) |0 , (2.3)

with an irrelevant (inaccessible) final quantum correlation between systemand environment as a relic from the initial superposition. In this unitary evo-lution, the two branches recombine to form a nonlocalsuperposition. Itexists, but it is not there. Its local unobservability characterizes an ap-parent collapse (as will be discussed). For a genuine collapse (Fig. 2.2),the final correlation would be statistical, and the ensemble entropy wouldincrease, too.

As mentioned in Sect. 2.2, the general interaction dynamics that is re-quired to describe ideal measurements according to the Schrodinger equa-

tion (2.1) is derived from the special case where the measured system is7 Inhisdefinitions, Bennett did not count the entropy of the microscopic ensemblea/bas physical entropy, because this variable is here assumed to be controllable,in contrast to the thermal (ergodic or uncontrollable) property A/B.


20/42


b B 0

a A 0

or (by reduction of

resulting ensemble)

A'aO

b B'collapse

Sensemble= So


I = 0

reset

( forget)

a

+b

A

+B

O

Sensemble= So - kln2

(Sphysical = So + kln2)

I = ?

a + b O



I = 0

O

system observer environment

'measurement'

interaction

Sensemble= So- kln2 (each)


I = kln2

Fig.2.2.Quantum measurement of a superposition|a+ |b by means of a collapseprocess, here assumed to be triggered by the macroscopic pointer position. Theinitial entropy is smaller by one bit than in Fig. 2.1 (and may in principle vanish),since there is no initial ensemblea/b for the property to be measured. Dashed linesbefore the collapse now represent quantum entanglement. (Compare the ensembleentropies with those of Fig. 2.1!) Increase of physical entropy in the first step isappropriate only if the arising entanglement is regarded as irrelevant. The collapseitself is often divided into two steps: first increasing the ensemble entropy by re-placing the superposition with an ensemble, and then lowering it by reducing theensemble (applying the or for macroscopic pointers only). The increase of en-semble entropy, observed in the final state of the Figure, is a consequence of thisfirst step of the collapse. It brings the entropy up to its classical initial value of

Fig. 2.1

prepared in an eigenstate|nbefore measurement (von Neumann 1932),|n|0 |n|n . (2.4)

Here,|n corresponds to|a or|b in the figures, the pointer positions|nto the states|A and|B. (During non-ideal measurements, the state|nwould change, too.) However, applied to an initial superposition,

cn|n,

the interaction according to (2.1) leads to an entangled superposition,cn|n

|0

cn|n|n . (2.5)

As explained in Sect. 2.1.1, the resulting superposition represents an individ-ual physical statethat is different from all components appearing in this sum.While decoherence arguments teach us (see Chap. 3) that neglecting the envi-ronment of (2.5) is absolutely unrealistic if|ndescribes the pointer state of


21/42


a A A' '

b B

+ or (by branching)

B''decoherence

A' A''

+

B' B''

a

+

b

O


Sphysical = So + 2kln2

I = 0

reset

( forget)

a

+b

A

+B

O



I = ?

a + b O



I = 0

O

system observe env ironmen

"measurement"

interaction


Sphysical = So (if A''/B'' irrelevant)

I = kln2

Fig.2.3. Quantum measurement of a superposition by means of branchingcaused by decoherence (see text). The increase of physical entropy during the secondstep applies if the distinction between environmental degrees of freedom A/B,responsible for decoherence, is irrelevant (uncontrollable). After the last step, al lentanglement has irreversibly become irrelevant in practice. Since the whole super-position is here assumed to exist forever (and may have future consequences inprinciple), the branching is meaningful only with respect to a local observer.

a macroscopic apparatus, this superposition remains nonetheless valid if isdefined to include the rest of the universe, such as|n=|n|n, with anenvironmental state

|

. This powerful consequence of the Schrodinger equa-

tion holds regardless of all complications, such as decoherence and other,in practice irreversible, processes (which need not even be known). There-fore, it does seem that the measurement problem can only be resolved if theSchrodinger dynamics (2.1) is supplemented by a nonunitary collapse (2.2).

Specific proposals for such a process will be discussed in Chap. 8. Remark-ably, however, there is no empirical evidence yet on where the Schr odingerequation may have to be modified for this purpose (see Joos 1986, Pearle andSquires 1994, or dEspagnat 2001). On the contrary, the dynamical superpo-sition principle has been confirmed with phantastic accuracy in spin systems(Weinberg 1989, Bollinger et al.1989).

The Copenhagen interpretation of quantum theory insists that the mea-surement outcome has to be described in fundamental classical terms ratherthan as a quantum state. While according to Pauli (in a letter to Einstein:Born 1969), the appearance of an electron position is a creation outside ofthe laws of Nature (eine ausserhalb der Naturgesetze stehende Schopfung),Ulfbeck and Bohr (2001) now claim (similar to Ludwig 1990 in his attempt


22/42


to derive the Copenhagen interpretation from fundamental principles) thatit is theclick in the counterthat appears out of the blue, and without anevent that takes place in the source itself as a precurser to the click. Together

with the occurrence of this, thusnot dynamically analyzable, irreversible eventin the counter, the wave function is then claimed to lose its meaning (pre-cisely where it would otherwise describe decoherence!). The Copenhagen in-terpretation is often hailed as the greatest revolution in physics, since it rulesout the general applicability of the concept of objective physical reality. Iam instead inclined to regard it as a kind of quantum voodoo: irrational-ism in place of dynamics. The theory of decoherence describes events in thecounter by means of a universal Schrodinger equation as a fast and for allpractical purposes irreversible dynamical creation of entanglement with theenvironment (see also Shi 2000). In order to remain politically correct, someauthors have recently even re-definedcomplementarity in terms of entangle-ment (cf. Bertet et al. 2001), although the latter has never been a crucialelement of the Copenhagen interpretation.

The Heisenberg cut between observer and observed has often beenclaimed to be quite arbitrary. This cut represents the borderline at which theprobability interpretation for the occurrence of events is applied. However,shifting it too far into the microscopic realm would miss the readily observedquantum aspects of certain large systems (SQUIDs etc.), while placing itbeyond the detector would require the latters decoherence to be taken intoaccount anyhow. As pointed out by John Bell (1981), the cut has to be placedfar enough from the measured object in order to ensure that our limitedcapabilities of investigation (such as those of keeping the measured systemisolated) prevent us from discovering any inconsistencies with the assumedclassical properties or a collapse.

As noticed quite early in the historical debate, the cut may even be placed

deep into the human observer, whose consciousness, which may be located inthe cerebral cortex, represents the final link in the observational chain. Thisview can be found in early formulations by Heisenberg, it was favored by vonNeumann, later discussed by London and Bauer (1939), and again supportedby Wigner (1962), among others. It has even been interpreted as an objectiveinfluence of consciousness on physical reality (e.g. Wigner l.c.), although itmay be consistent with the formalism only when used with respect to onefinal observer, that is, in a strictly subjective (though partly ob jectivizable)sense (Zeh 1971).

The undivisible chain between observer and observed is physically rep-resented by a complex interacting medium, or a chain of intermediary systems|(i), in quantum mechanical terms symbolically written as


23/42


systemn (1)0

(2)0

(K)0

obs0

systemn

(1)n

(2)0

(K)0 obs0

...

systemn

(1)n (2)n

(K)n obsn , (2.6)

instead of the simplified form (2.4). This chain is thus assumed to act dy-namically step by step (cf. Zeh 1973). While an initial superposition of theobserved system now leads to a superpositionof such product states (similarto (2.5)), we know empirically that a collapse must be taken into accountby the conscious observer before (or at least when) the information arrivesat him as the final link. If there are several chains connecting observer andobserved (for example via other observers, known as Wigners friends), thecorrectly applied Schrodinger equation warrants that each individual compo-

nent (2.6) describes consistent (objectivized) measurement results. Fromthe subjective point of view of the final observer, all intermediary systems(Wigners friends or Schrodingers cats, including their environments)could well remain in a superposition of drastically different situations untilhe observes (or communicates with) them!

Environment-induced decoherence means that an avalanche of other causalchains unavoidably branch off from the intermediary links of the chain as soonas they become macroscopic (see Chap. 3). This might even trigger arealcol-lapse process (to be described by hypothetical dynamical terms), since themany-particle correlations arising from decoherence would render the totalsystem prone to such as yet unobserved, but nevertheless conceivable, non-linear many-particle forces (Pearle 1976, Diosi 1985, Ghirardi, Rimini, andWeber 1986, Tessieri, Vitali, and Grigolini 1995; see also Chap. 8). Deco-

herence by amicroscopicenvironment has been experimentally confirmed tobe reversible in what is now often called quantum erasure of a measure-ment (see Herzog et al. 1995). In analogy to the concept of particle creation,reversible decoherence may be regarded as virtual decoherence. Real de-coherence, which gives rise to the familiar classical appearance of the macro-scopic world, is instead characterized by its unavoidability and irreversibilityin practice.

In an important contribution, Tegmark (2000) was able to demonstratethat neuronal and other processes in the brain also become quasi-classicalbecause of environmental decoherence. (Successful neuronal models are in-deed classical.) This seems to imply that at least objectiveaspects of humanthinking and behavior can be described by conceptually classical (though notnecessarily deterministic) models of the brain. However, since no precise lo-

calization of consciousness within the brain has been found yet, the neuralnetwork (just as the retina, say) may still be part of the external worldwith respect to the unknown ultimate observer system (Zeh 1979). Because


24/42


of Tegmarks arguments, this problem may not affect an objectivetheory ofobservation any longer.

Even real decoherence in the sense of above must be distinguished from

a genuine collapse, which is defined as the disappearanceof all but one com-ponents from reality (thus representing an irreversiblelaw).8 As pointed outabove, a collapse could well occur much later in the observational chain thandecoherence, and possibly remain less fine-grained. Nonetheless, it shouldthen be detectable in other situations if its dynamical rules are defined.Environment-induced decoherence (the dynamically arising strong correla-tions with the rest of the world) leads to the important consequence that,in a world with no more than few-particle forces, robust factorstates

obsn are not affected by what goes on in the other branches that have formedaccording to the Schrodinger equation.

In order to represent a subjective observer, a physical system must be ina definite state with respect to properties of which he/she/it is aware. Thesalvation of a psycho-physical parallelism of this kind was von Neumannsmain argument for the introduction of his first intervention (2.2): the col-lapse. As a consequence of the above-discussed dynamical independence ofthe different individual components of type (2.6) in their superposition, onemay instead associateal larising factor wave functions

obsn (different onesin each component) with separatesubjective observers (that is, with differ-ent states of consciousness). This approach, which avoids a collapse as a newdynamical law, is essentially identical with Everetts relative state inter-pretation (so called, since the worlds observedby these observer states aredescribed by their corresponding relative factor states). Although also calleda many worlds interpretation, it describes onequantum universe. Becauseof its (essential and non-trivial) reference to conscious observers, it may moreappropriately be called a multi-consciousness or many minds interpreta-

tion (Zeh 1970, 1971, 1979, 1981, 2000, Albert and Loewer 1988, Lockwood1989, Squires 1990, Stapp 1993, Donald 1995, Page 1995).9

8 Proposed decoherence mechanisms involving event horizons (Hawking 1987, Ellis,Mohanty and Nanopoulos 1989) would either requirea fundamental violation ofunitarity, or merely represent a specific kind of environmental decoherence (en-tanglement beyond the horizon). The most immediate consequence of quantumentanglement is that quantum theory can be consistently applied only to thewhole universe.

9 As Bell (1981) pointed out, Bohms theory would instead require consciousnessto be psycho-physically coupled to certain classicalvariables (which this theorypostulatesto exist). These variables are probabilistically related to the wave func-tion by means of a conserved statistical initial condition. Thus one may arguethat the many minds interpretation merely eliminates Bohms unobservableand therefore meaningless intermediary classical variables and their trajectories

from this psycho-physical connection. This is possible because of the dynami-cal autonomy of a wave function that evolves in time according to a universalSchrodinger equation, and independently of Bohms classical variables. These


25/42


Because of their dynamical independence, all these different observers (orrather, different versions of the same observer) cannot find out by means ofexperiments whether or not the other components have survived. This conse-

quence of the Schrodinger equationthus leads to the impression(for separateobservers) that all other components have hurried out of existence assoon as decoherence has become irreversible for all practical purposes. Thenit remains a pure matter of taste whether Occams razor is applied to thewave function (by adding appropriate but not directly detectable collapse-producing nonlinear terms to its dynamical law), or to the dynamical law (byinstead adding myriads of unobservable Everett components to our concep-tion of reality). Traditionally (and mostly successfully), consistency of thelaw has been ranked higher than simplicity of the facts.

Fortunately, the dynamics of decoherence can be discussed without giv-ing an answer to this question. A collapse (real or apparent) has to be takeninto accountregardless of its interpretation in order to describe the dynam-ics of that wave function which represents our observed quasi-classical world(the time-dependent component which contains our observer states

obsn ).Only specificdynamical collapse models could be confirmed or ruled out byexperiments, while Everetts relative states, on the other hand, depend inprinciple on the definition of the observer system.10 (No other systemshave to be specified in principle. Their density matrices, which describe de-coherence and quasi-classical concepts, are merely convenient.)

In contrast to Bohms theory or stochastic collapse models, nothing hasbeen said yet (or postulated) about probabilitiesof measurement outcomes.For this purpose, the Everett branches have to be given statistical weights in away that appearsad hocagain. However, these probabilities are meaningful toan observer only as frequencies in seriesof equivalent measurements. Thesemeasurements must be performed in his branch (and would in general be

performed and lead to different results in other branches). Graham (1970)was able to show that the norm of the superposition of all thoseEverettbranches (arising from such series of measurements) which contain frequenciesof results that substantially differfrom Borns probabilities vanishes in thelimit of infinite series. Although the definition of the norm, which is used inthis argument, is exactly equivalent to Borns probabilties, it can be selected

variables thus cannot, by themselves, carry memories of their surrealistic his-tory. Memory is solely in the quasi-classical wave packet that effectively guidesthem, while the other myriads of empty Everett world components (criticizedfor being extravagant by Bell) existas well in Bohms theory. Barbour (1994,1999), in his theory of timelessness, effectively proposed astaticversion of Bohmstheory, which eliminates the latters formal classical trajectories even though itpreserves a concept of memories without a history (time capsules see also

Chapt. 6 of Zeh 2001).10 Another aspect of this observer-relatedness of the observed world is the conceptof a presence, which is not part ofphysicaltime. It reflects the empirical factthatthe subjective observer is local in space andtime.


26/42


against other definitions of a norm by its unique property of being conservedunder the Schrodinger dynamics.

To give an example: an isolated decaying quantum system may be de-

scribed as a superposition of its metastable initial state and the outgoingchannel wave function(s) for all its decay products. On a large but finiteregion of space and time the total wave function may then approximatelydecrease exponentially and coherently in accordance with the Schrodingerequation (with very small late-time deviations from exponential behaviorcaused by the dispersion of the outgoing waves). For a system that decaysby emitting photons into a reflecting cavity, a superposition of different de-cay times has in fact been confirmed in the form of coherent state vectorrevival (Rempe, Walther and Klein 1987). An even more complex exper-iment exhibiting coherent state vector revival was performed by means ofspin waves (Rhim, Pines and Waugh 1971). In general, however, the decayfragments would soon be monitored by surrounding matter. The resultingstate of the environment must then depend (contain information) on thedecay time, and the superposition will decohere into dynamically independentcomponents corresponding to different(approximately defined) decay times.From the point of view of a local observer, the system may be assumed tohave decayed at a certain time(within the usually very narrow limits set bythe decoherence time scale see Sect. 3.3.2), even though he need not haveobserved the decay. This situation does not allow coherent state vector re-vival any more. Instead, it leads to an exponential distribution of decay timesin the arising apparent ensemble, valid shortly after the decaying state hasbeen produced (see Joos 1984).

However, as long as the information has not yet reached the observer,n

cn systemn

(1)n

(2)n

(K)n

obs0

, (2.7)

he may as well assume (from his subjective point of view) that the nonlo-cal superposition still exists. According to the formalism, Schrodingers cat(represented by

(2), say) would then become dead oralive only whenhe becomes aware of it. On the other hand, the property described by the

state|systemn (just as the cats status of being dead or alive,(2)n

) can

also be assumed to have become real as soon as decoherence has becomeirreversible in practice. Therefore, decoherence must also corrupt any control-lable entanglement that would give rise to a violation of Bells inequalities (asit does see Venugopalan, Kumar and Gosh 1995). If, instead of taking noticeof the result, the observer would decide to perform another measurement ofthe system (by using a new observational chain), he could not observe any

interference between different ns, since, as a local system, he cannot performthe required global measurements. All predictions which this observer cancheck are consistent with the assumption that the system was in oneof thestates|systemn (with probability|cn|2) before the second measurement (see


27/42


also Sect. 2.4). This justifies the interpretation that the cat is determinedtodie or not yet to die as soon as irreversible decoherence has occurred some-where in the chain (which will in general be the case before the poison is

applied).

2.4 Density Matrix, Coarse Graining, and Events

The theory of decoherence uses some (more or less technical) auxiliary con-cepts. Their physical meaning will be recalled and discussed in this section,as it is essential for a correct interpretation of what is actually achieved withthis theory.

In classical statistical mechanics, incomplete knowledge about the realphysical state of a system is described by ensembles of states, that is, byprobability distributions (p, q) on phase space (in Fig. 2.1 symbolically in-dicated by areas of uniform probability). Such ensembles are often called

thermodynamic or macroscopic states. Mean values of state functionsa(p, q) (that is, physical quantities that are determined by the microscopicstate p,q), defined with respect to this ensemble, are then given by the expres-sion

dp dq (p, q)a(p, q). The ensemble(p, q) itself could be recovered from

the mean values of a complete set of state functions (such as all -functions),while a (smaller) set, that may be realized in practice, determines only acoarse-grained probability distribution.

Ifall states which form such an ensemble are assumed to obey the sameHamiltonian equations, their probability distribution evolves according tothe Liouville equation,

t ={H , } , (2.8)

with Hamiltonian H and Poisson bracket{

,}

. However, this assumptionwould be highly unrealistic for a many-particle system. Even if the funda-mental dynamics is assumed to be given, the effectiveHamiltonian for theconsidered system depends very sensitively on the state of the environment,which cannot be assumed to be known better than that of the system itself.Borel (1914) showed long ago that even the gravitational effect resulting fromshifting a small rock at the distance of Sirius by a few centimetres wouldcompletely change the microscopic state of a gas in a vessel here on earthwithin seconds after the retarded field has arrived (see also Chap. 3). In asimilar connection, Ernst Mach spoke of the profound interconnectednessof things. This surprising result is facilitated by the enormous amplificationof the tiny differences in the molecular trajectories, caused by the slightlydifferent forces, during subsequent collisions with other molecules (each time

by a factor of the order of the ratio of the mean free path over the molecularradius). Similarly, microscopic differences in the state of the gas will imme-diately disturb its environment, thus leading in turn to slightly different ef-fective Hamiltonians for the gas, with in general grossly different (chaotic)


28/42


effects on the microscopic states of the original ensemble. This will inducestrongstatisticalcorrelations of the gas with its environment, whose neglectwould lead to an increase of ensemble entropy.

This effective local dynamical indeterminism can be taken into account(when calculating forward in time) by means of stochastic forces (using aLangevin equation) for the individual state, or by means of a correspondingmaster equationfor an ensemble of states, (p, q). The increase of the localensemble entropy is thus attributed to an uncertain effective Hamiltonian.In this way, statistical correlations with the environment are regarded asdynamically irrelevant for the futureevolution. An example is Boltzmannscollision equation (where the arising irrelevant correlations are intrinsic tothe gas, however). The justification of this time-asymmetric procedure formsa basic problem of physics and cosmology (Zeh 2001).

When appying the conventional quantization rules to the Liouville equa-tion (2.8) in a formal way, one obtains thevon Neumann equation(orquantumLiouville equation),

iht

= [H , ] , (2.9)

for the dynamics of statistical operators or density operators . Simi-larly, expectation valuesA = tr(A)/tr() of observables A formally re-place mean values a =

dp dq a(p, q)(p, q) of the state functions a(p, q).

Expectation values of a restrictedset of observables would again representa generalized coarse graining for the density operators. The von Neumannequation (2.9) is unrealistic for similar reasons as is the Liouville equation,although quantitative differences may arise from the different energy spec-tra mainly at low energies. Discrete spectra have relevant consequences formacroscopic systems in practice only in exceptional cases, while they oftenprevent mathematically rigorous proofs of ergodic or chaos theorems which

are valid in excellent approximation. However, whenever quantum correla-tions do form in analogy to classical correlations (as is the rule), they lead tofar more profound consequences than their classical counterparts.

In order to explain these differences, the concept of a density matrix hasto be derived from that of a (pure) quantum state instead of being postulatedby means of quantization rules. According to Sect. 2.2, the probability for astate|n to be found in a state| in a corresponding measurement isgiven by|n||2. Its meanprobability in an ensemble of states{|} withprobabilitiesprepresenting incomplete information about the initial stateis, therefore,pn=

p|n||2 = tr{Pn}, where =

|p|and Pn=|nn|. This result remains true for general observablesA = anPn in placeofPn. The ensemble of wave functions|, which thus defines a density matrixas representing astate of information, need not consist of mutually orthogonal

states, although the density matrix can always be diagonalized in terms of itseigenbasis. Its representation by a general ensemble of states is therefore farfrom unique in contrast to a classical probability distribution. Nonetheless,the density matrix can still be shown to obey a von Neumann equation if


29/42


all states contained in the ensemble are assumed to evolve according to thesame unique Hamiltonian.

However, the fundamental nonlocality of quantum states means that the

state of a local system does not existin general: it cannot be merely unknown.Accordingly, thereis noeffective local Hamiltonian that would allow (2.9) toapply in principle (see Kubler and Zeh 1973). In particular, a time-dependentHamiltonian would in general require a (quasi-)classical environment. Thisspecific quantum aspect is easily overlooked when the density matrix is intro-duced axiomatically by quantizing a classical probability distribution onphase space.

Quantum nonlocality means that the generic state of a composite system(system and environment, say),

|=m,n

cmnsystemm environmentn , (2.10)

does not factorize. The expectation values of all local observables,

A= Asystem 1lenvironment , (2.11)

have then to be calculated by tracing out the environment,

| A | tr {A | |}= tracesystem

Asystemsystem

. (2.12)

Here, the density matrix system, which in general has nonzero entropy evenfor a pure (completely defined) global state|, is given by

system m,m

systemm systemmm systemm

:= traceenv{| |} m,m

systemm

n

cmnc

mn

systemm

. (2.13)

It represents the specific coarse-grainingwith respect to all subsystem ob-servables only. This reduced density matrix can beformallyrepresented byvarious ensembles of local states (including its eigenrepresentation or diagonalform), although it does here characterize one pure but entangledglobalstate.A density matrix thus based on entanglement has been called an impropermixture by dEspagnat (1966). It can evidently not explainensemblesof def-inite measurement outcomes. If proper and improper mixtures were identifiedfor operationalistic reasons (that are based on the measurement postulate),then decoherence would indeed completely solve the measurement problem.

Regardless of its origin and interpretation, the density matrix can be

replaced by its partial Fourier transform, known as the Wigner function(seealso Sect. 3.2.3):


30/42


W(p, q) : = 1

e2ipx(q+ x, q x) dx

1

2

qz + z

2

eip(zz)

(z, z

) dzdz

= : trace{p,q} . (2.14)

The second line is here written in analogy to the Bloch vector, i = trace{i},since

p,q(z, z) :=

1

2eip(zz

)

q z + z

2

(2.15)

is a generalization of the Pauli matrices (with indexp, qinstead ofi = 1, 2, 3 see also Sect. 4.4 of Zeh 2001). Although the Wigner function is formallyanalogous to a phase space distribution, it does, according to its derivation,notrepresent an ensemble of classical states (phase space points). This is re-flected by its potentially negative values, while even a Gaussian wave packet,

which does lead to a non-negative Wigner function, is nonetheless one(pure)quantum state.

The degree of entanglement represented by an improper mixture (2.13)is conveniently measured by the latters formal entropy, such as the

zeh-2002-decoherence- basic concepts and their interpretation.9506020v3

Documents