Arxiv:Quant-Ph/0105127V3 19 Jun 2003

Home , Einselection, Open quantum system, Pointer state

arXiv:quant-ph/0105127v3 19 Jun 2003 iaiy e ymtyapasi hsstig Envi- setting: clas- or this invariance their in assisted of appears - measure ronment symmetry a new is A ‘fit- sense) their sicality. Darwinian as the of thought in be ness’ can states (which pointer environment of the records by in the envi- of e.g, the redundancy in The re-prepared, present ronment. being already without information the out” intercepting “found the be on can based existence property objective a under- tively as to reality’ pointer one ‘classical classical allows stand environment effectively the the throughout with states Spreading correlations information. the process and of abil- store, their acquire, by to only ity distinguished systems, quantum open in wavepacket”. its results terpretation the of of apparatus “collapse utility macroscopic effective the predictive the the with on correlations and restriction microscopic this is system isolated, power. quantum ap- predictive measured has the the that of When information observable store corre- pointer can classical preferred paratus the the with Only system lation. measured appa- the the and einselec- between ratus entanglement measurements, In quantum replaces trajectory. tion point classical a a of of idealizations the and to leads dynamics in einse- with of space lection Combination of Hilbert limit: quantum macroscopic structure appropriate the the from Classical the emerges of space non- phase states. majority flagrantly vast cat” “Schrödinger the the especially local eliminating on space, ban Hilbert by effective spite classicality an in enforces Universe Einselection imposing the environment. of can the rest They of the stable. with infor- are correlations of states retain loss pointer Einselected selective with mation. associated process quantum Contents matri- program einselection. density the and of reduced decoherence principles the of basic to justifying and ultimately and rules ces, environment, Born’s the system to the with of leads correlations state quantum the of to ignorance due of nature the on light to the between observables coherence states destroying certain system, monitors the effect of in which vironment CLASSICAL THE OF ORIGINS QUANTUM THE AND EINSELECTION, DECOHERENCE, Decoherence niomn-nue superselection environment-induced .INTRODUCTION I. orsodn oteregnaus hsleads This eigenvalues. their to corresponding hoyDvso,LN,Mi tpB288 Stop Mail LANL, 87545 Mexico New Alamos, Division, Los Theory Zurek Hubert Wojciech mle yeneeto ead bevr as observers regards einselection by implied scue yteitrcinwt h en- the with interaction the by caused is fteeneetdsae:They states: einselected the of envariance or einselection xseta in- Existential hd new a sheds pointer rela- a , 2 I.CASADLS FCORRESPONDENCE OF LOSS AND CHAOS III. I ISLCINADMEASUREMENTS AND EINSELECTION VI. V NIOMN NUE SUPERSELECTION INDUCED – ENVIRONMENT IV. I UNU MEASUREMENTS QUANTUM II. .ENEETO NPAESPACE PHASE IN EINSELECTION V. .Peitblt iv n islcin20 einselection and sieve Predictability D. .Cascllmti hs pc 26 space phase in limit Classical D. .EitnilItrrtto n QatmDarwinism’4 ‘Quantum and Interpretation Existential D. .Ojcieeitneo islce tts30 states einselected of existence Objective A. .Mdl feneeto 14 einselection of Models A. .QatmBona oin21 motion Brownian Quantum A. .Ls fteqatmcasclcrepnec 11 correspondence quantum-classical the of Loss A. .Qatmcniinldnmc 5 dynamics conditional Quantum A. .Tepolm ibr pc sbg2 big is space Hilbert problem: The A. .Aim fqatmmaueetter 32 theory measurement quantum of Axioms C. .Dchrne nageet ehsn,adnie19 noise and dephasing, entanglement, Decoherence, C. .Peitblt iv npaesae26 space phase in sieve Predictability C. .“olpe nlgei lsia esrmn 9 measurement classical a in analogue “Collapse” C. .Smtm fcrepnec os13 loss correspondence of Symptoms C. .Tentr fterslto n h oeo envariance4 of role the and resolution the of nature The C. .Maueet n eois31 memories and Measurements B. .Eneeto steslciels fifrain17 information of loss selective the as Einselection B. .Dchrnei unu rwinmto 23 motion Brownian quantum in Decoherence B. .MylbaktadLovleflw12 flow Liouville and bracket Moyal B. .Ifraintase nmaueet 9 measurements in transfer Information B. .Dchrneadeneeto 3 einselection and Decoherence B. .Dchrne ho,adteScn a 28 Law Second the and chaos, Decoherence, E. .Poaiiis islcinadrcrs34 33 33 32 records and einselection Probabilities, (iv) axiom 4. (iiib) repeatability, axiom Immediate – outcomes (iiia) 3. as axiom Eigenvalues – Hermitean 2. are Observables 1. .Etoypouto 29 30 28 horizon predictability Quantum 3. production Entropy correspondence 2. of Restoration 1. .Inrneisie ofiec ncasclt 28 27 classicality in confidence inspires limit Ignorance macroscopic The 3. approach: Physical 2. .Eneeto n otoldsit 16 15 16 shifts halo controlled quantum and a Einselection and domain 3. classical qubit The single a 2. of Decoherence 1. .Mta nomto n icr 18 discord and information Mutual 1. .Dchrnetmsae24 25 decoherence of view space Phase 2. timescale Decoherence 1. .Mteaia prah(¯ approach Mathematical 1. .Srcuestrto 13 13 saturation Structure 2. values Expectation 1. .Ato e bit per Action 1. .Maueet n otoldsit.7 6 Amplification shifts. controlled measurement 3. and bit-by-bit Measurements a and 2. not Controlled 1. .Mn olsItrrtto 3 2 Interpretation Worlds Many 2. Interpretation Copenhagen 1. h → )27 0) 30 21 11 9 7 5 14 2

D. Probabilities from Envariance 34 A. The problem: Hilbert space is big 1. Envariance 35 2. Born’s rule from envariance 36 3. Relative frequencies from envariance 38 The interpretation problem stems from the vastness of 4. Other approaches to probabilities 39 the Hilbert space, which, by the principle of superposition, admits arbitrary linear combinations of any states VII. ENVIRONMENT AS A WITNESS 40 as a possible quantum state. This law, thoroughly tested A. Quantum Darwinism 40 in the microscopic domain, bears consequences that defy 1. Consensus and algorithmic simplicity 41 2. Action distance 41 classical intuition: It appears to imply that the familiar 3. Redundancy and mutual information 42 classical states should be an exceedingly rare exception. 4. Redundancy ratio rate 43 And, naively, one may guess that superposition principle B. Observers and the Existential Interpretation 43 should always apply literally: Everything is ultimately C. Events, Records, and Histories 44 made out of quantum “stuﬀ”. Therefore, there is no 1. Relatively Objective Past 45 a priori reason for macroscopic objects to have deﬁnite 1 VIII. DECOHERENCE IN THE LABORATORY 46 position or momentum. As Einstein noted localization A. Decoherence due to entangling interactions 46 with respect to macrocoordinates is not just independent, B. Simulating decoherence with classical noise 47 but incompatible with quantum theory. How can one then 1. Decoherence, Noise, and Quantum Chaos 48 establish correspondence between the quantum and the C. Analogue of decoherence in a classical system 48 D. Taming decoherence 49 familiar classical reality? 1. Pointer states and noiseless subsystems 49 2. Environment engineering 49 3. Error correction and resilient computing 50 1. Copenhagen Interpretation IX. CONCLUDING REMARKS 51

X. ACKNOWLEDGMENTS 52 Bohr’s solution was to draw a border between the quantum and the classical and to keep certain objects – References 52 especially measuring devices and observers – on the classical side (Bohr, 1928; 1949). The principle of superposition was suspended “by decree” in the classical domain. I. INTRODUCTION The exact location of this border was difficult to pinpoint, but measurements “brought to a close” quantum events. The issue of interpretation is as old as quantum the- Indeed, in Bohr’s view the classical domain was more ory. It dates back to the discussions of Niels Bohr, fundamental: Its laws were self-contained (they could be Werner Heisenberg, Erwin Schrödinger, (Bohr, 1928; confirmed from within) and established the framework 1949; Heisenberg, 1927; Schrödinger, 1926; 1935a,b; see necessary to define the quantum. also Jammer, 1974; Wheeler and Zurek, 1983). Perhaps The first breach in the quantum-classical border ap- the most incisive critique of the (then new) theory was peared early: In the famous Bohr – Einstein double-slit due to Albert Einstein, who, searching for inconsisten- debate, quantum Heisenberg uncertainty was invoked by cies, distilled the essence of the conceptual difficulties of Bohr at the macroscopic level to preserve wave-particle quantum mechanics through ingenious “gedankenexper- duality. Indeed, as the ultimate components of classical iments”. We owe him and Bohr clarification of the sig- objects are quantum, Bohr emphasized that the bound- nificance of the quantum indeterminacy in course of the ary must be moveable, so that even the human nervous Solvay congress debates (see Bohr, 1949) and elucidation system could be regarded as quantum provided that suit- of the nature of quantum entanglement (Einstein, Podol- able classical devices to detect its quantum features were sky, and Rosen, 1935; Bohr, 1935, Schrödinger, 1935a,b). available. In the words of John Archibald Wheeler (1978; Issues identified then are still a part of the subject. 1983) who has elucidated Bohr’s position and decisively Within the past two decades the focus of the re- contributed to the revival of interest in these matters, search on the fundamental aspects of quantum theory has “No [quantum] phenomenon is a phenomenon until it is shifted from esoteric and philosophical to more “down to a recorded (observed) phenomenon”. earth” as a result of three developments. To begin with, many of the old gedankenexperiments (such as the EPR “paradox”) became compelling demonstrations of quantum physics. More or less simultaneously the role of de- 1 In a letter dated 1954, Albert Einstein wrote to Max Born “Let coherence begun to be appreciated and einselection was ψ1 and ψ2 be solutions of the same Schrödinger equation..... recognized as key in the emergence of classicality. Last When the system is a macrosystem and when ψ1 and ψ2 are not least, various developments have led to a new view ‘narrow’ with respect to the macrocoordinates, then in by far the of the role of information in physics. This paper reviews greater number of cases this is no longer true for ψ = ψ1 + ψ2. Narrowness with respect to macrocoordinates is not only inde- progress with a focus on decoherence, einselection and pendent of the principles of quantum mechanics, but, moreover, the emergence of classicality, but also attempts a “pre- incompatible with them.” (The translation from Born (1969) view” of the future of this exciting and fundamental area. quoted here is due to Joos (1986).) 3

This is a pithy summary of a point of view – known as Everett reinstated quantum mechanics as a basic tool in the Copenhagen Interpretation (CI) – that has kept many the search for its interpretation. a physicist out of despair. On the other hand, as long as a compelling reason for the quantum-classical border could not be found, the CI Universe would be governed by two B. Decoherence and einselection sets of laws, with poorly defined domains of jurisdiction. This fact has kept many a student, not to mention their Decoherence and einselection are two complementary teachers, in despair (Mermin 1990a; b; 1994). views of the consequences of the same process of environmental monitoring. Decoherence is the destruction of quantum coherence between preferred states associ- 2. Many Worlds Interpretation ated with the observables monitored by the environment. Einselection is its consequence – the de facto exclusion The approach proposed by Hugh Everett (1957a, b) of all but a small set, a classical domain consisting of and elucidated by Wheeler (1957), Bryce DeWitt (1970) pointer states – from within a much larger Hilbert space. and others (see DeWitt and Graham, 1973; Zeh, 1970; Einselected states are distinguished by their resilience – 1973; Geroch, 1984; Deutsch, 1985, 1997, 2001) was to stability in spite of the monitoring environment. enlarge the quantum domain. Everything is now repre- The idea that “openness” of quantum systems may sented by a unitarily evolving state vector, a gigantic su- have anything to do with the transition from quantum to perposition splitting to accommodate all the alternatives classical was resolutely ignored for a very long time, prob- consistent with the initial conditions. This is the essence ably because in classical physics problems of fundamen- of the Many Worlds Interpretation (MWI). It does not tal importance were always settled in isolated systems. In suffer from the dual nature of CI. However, it also does the context of measurements Gottfried (1967) anticipated not explain the emergence of classical reality. some of the latter developments. The fragility of energy The difficulty many have in accepting MWI stems from levels of quantum systems was emphasized by seminal pa- its violation of the intuitively obvious “conservation law” pers of Dieter Zeh (1970; 1973), who argued (inspired by – that there is just one Universe, the one we perceive. remarks relevant to what would be called today “deter- But even after this question is dealt with, , many a con- ministic chaos” (Borel, 1914)) that macroscopic quantum vert from CI (which claims allegiance of a majority of systems are in effect impossible to isolate. physicists) to MWI (which has steadily gained popular- The understanding of how the environment distills the ity; see Tegmark and Wheeler, 2001, for an assessment) classical essence from quantum systems is more recent eventually realizes that the original MWI does not ad- (Zurek, 1981; 1982; 1993a). It combines two observa- dress the “preferred basis question” posed by Einstein1 tions: (1) In quantum physics “reality” can be attributed (see Wheeler, 1983; Stein, 1984; Bell 1981, 1987; Kent, to the measured states. (2) Information transfer usually 1990; for critical assessments of MWI). And as long as associated with measurements is a common result of al- it is unclear what singles out preferred states, perception most any interaction with the environment of a system. of a unique outcome of a measurement and, hence, of a Some quantum states are resilient to decoherence. This single Universe cannot be explained either2. is the basis of einselection. Using Darwinian analogy, In essence, Many Worlds Interpretation does not ad- one might say that pointer states are most ‘fit’. They dress but only postpones the key question. The quantum survive monitoring by the environment to leave ‘descen- - classical boundary is pushed all the way towards the dants’ that inherit their properties. Classical domain of observer, right against the border between the material pointer states offers a static summary of the result of Universe and the “consciousness”, leaving it at a very quantum decoherence. Save for classical dynamics, (al- uncomfortable place to do physics. MWI is incomplete: most) nothing happens to these einselected states, even It does not explain what is effectively classical and why. though they are immersed in the environment. Nevertheless, it was a crucial conceptual breakthrough: It is difficult to catch einselection “in action”: Envi- ronment has little effect on the pointer states, as they are already classical. Therefore, it was easy to miss decoherence - driven dynamics of einselection by tak- 2 DeWitt, in the Many Worlds re-analysis of quantum measure- ing for granted its result – existence of the classical do- ments makes this clear: in DeWitt and Graham (1973), last main, and a ban on arbitrary quantum superpositions. paragraph of p. 189 he writes about the key ‘remaining prob- Macroscopic superpositions of einselected states disap- lem’: “Why is it so easy to find apparata in states [with a well defined value of the pointer observable]? In the case of macro- pear rapidly. Einselection creates effective superselection scopic apparata it is well known that a small value for the mean rules (Wick, Wightman and Wigner, 1952; 1970; Wight- square deviation of a macroscopic observable is a fairly stable man, 1995). However, in the microscopic, decoherence property of the apparatus. But how does the mean square devi- can be slow in comparison with the dynamics. ation become so small in the first place? Why is a large value of the mean square deviation of a macroscopic observable virtually Einselection is a quantum phenomenon. Its essence never, in fact, encountered in practice? . . . a proof of this does cannot be even motivated classically: In classical physics not yet exist. It remains a program for the future.” arbitrarily accurate measurements (also by the environ- 4 ment) can be in principle carried out without disturbing Universe will appear classical to observers who monitor the system. Only in quantum mechanics acquisition of it from within, using their limited capacity to acquire, information inevitably alters the state of the system – store, and process information. It allows one to under- the fact that becomes apparent in double-slit and related stand classicality as an idealization that holds in the limit experiments (Wootters and Zurek, 1979; Zurek, 1983). of macroscopic open quantum systems. Quantum nature of decoherence and the absence of Environment imposes superselection rules by preserv- classical analogues are a source of misconceptions. For ing part of the information that resides in the correlations instance, decoherence is sometimes equated with relax- between the system and the measuring apparatus (Zurek, ation or classical noise that can be also introduced by 1981, 1982). The observer and the environment compete the environment. Indeed, all of these effects often ap- for the information about the system. Environment – pear together and as a consequence of the “openness”. because of its size and its incessant interaction with the The distinction between them can be briefly summed up: system – wins that competition, acquiring information Relaxation and noise are caused by the environment per- faster and more completely than the observer. Thus, a turbing the system, while decoherence and einselection record useful for the purpose of prediction must be re- are caused by the system perturbing the environment. stricted to the observables that are already monitored by Within the past few years decoherence and einselection the environment. In that case, the observer and the en- became familiar to many. This does not mean that their vironment no longer compete and decoherence becomes implications are universally accepted (see comments in unnoticeable. Indeed, typically observers use environ- the April 1993 issue of Physics Today; d’Espagnat, 1989 ment as a “communication channel”, and monitor it to and 1995; Bub, 1997; Leggett, 1998 and 2002; Stapp, find out about the system. 2001; exchange of views between Anderson, 2001, and Spreading of the information about the system through Adler, 2001). In a field where controversy reigned for so the environment is ultimately responsible for the emer- long this resistance to a new paradigm is no surprise. gence of the “objective reality”. Objectivity of a state can be quantified by the redundancy with which it is recorded throughout Universe. Intercepting fragments of C. The nature of the resolution and the role of envariance the environment allows observers to find out (pointer) state of the system without perturbing it (Zurek, 1993a, Our aim is to explain why does the quantum Universe 1998a, and 2000; see especially section VII of this pa- appear classical. This question can be motivated only per for a preview of this new “environment as a witness” in the context of the Universe divided into systems, and approach to the interpretation of quantum theory). must be phrased in the language of the correlations be- When an effect of a transformation acting on a system tween them. In the absence of systems Schrödinger equa- can be undone by a suitable transformation acting on the tion dictates deterministic evolution; environment, so that the joint state of the two remains unchanged, the transformed property of the system is Ψ(t) = exp( iHt/¯h) Ψ(0) , (1.1) | i − | i said to exhibit “environment assisted invariance” or en- and the problem of interpretation seems to disappear. variance (Zurek, 2002b). Observer must be obviously There is no need for “collapse” in a Universe with no ignorant of the envariant properties of the system. Pure systems. Yet, the division into systems is imperfect. As entangled states exhibit envariance. Thus, in quantum a consequence, the Universe is a collection of open (in- physics perfect information about the joint state of the teracting) quantum systems. As the interpretation prob- system-environment pair can be used to prove ignorance lem does not arise in quantum theory unless interacting of the state of the system. systems exist, we shall also feel free to assume that an Envariance offers a new fundamental view of what is environment exists when looking for a resolution. information and what is ignorance in the quantum world. Decoherence and einselection fit comfortably in the It leads to Born’s rule for the probabilities and justifies context of the Many Worlds Interpretation where they the use of reduced density matrices as a description of a define the “branches” of the universal state vector. De- part of a larger combined system. Decoherence and ein- coherence makes MWI complete: It allows one to ana- selection rely on reduced density matrices. Envariance lyze the Universe as it is seen by an observer, who is also provides a fundamental resolution of many of the inter- subject to decoherence. Einselection justifies elements pretational issues. It will be discussed in section VI D. of Bohr’s CI by drawing the border between the quantum and the classical. This natural boundary can be sometimes shifted: Its effectiveness depends on the de- D. Existential Interpretation and ‘Quantum Darwinism’ gree of isolation and on the manner in which the system is probed, but it is a very effective quantum - classical What the observer knows is inseparable from what the border nevertheless. observer is: The physical state of his memory implies Einselection fits either MWI or CI framework: It sup- his information about the Universe. Its reliability de- plies a statute of limitations, putting an end to the pends on the stability of the correlations with the exter- quantum jurisdiction. . It delineates how much of the nal observables. In this very immediate sense decoher- 5 ence enforces the apparent “collapse of the wavepacket”: ronment (Zurek, 1983, 1998a, 2000). This new ‘quantum After a decoherence timescale, only the einselected mem- Darwinism’ view of environment selectively amplifying ory states will exist and retain useful correlations (Zurek, einselected pointer observables of the systems of interest 1991; 1998a,b; Tegmark, 2000). The observer described is complementary to the usual image of the environment by some specific einselected state (including a configu- as the source of perturbations that destroy quantum co- ration of memory bits) will be able to access (“recall”) herence of the system. It suggests the redundancy of only that state. The collapse is a consequence of einse- the imprint of the system in the environment may be a lection and of the one-to-one correspondence between the quantitative measure of relative objectivity and hence of state of his memory and of the information encoded in it. classicality of quantum states. It is introduced in Sec- Memory is simultaneously a description of the recorded tions VI and VII of this review. information and a part of the “identity tag”, defining Benefits of recognition of the role of environment in- observer as a physical system. It is as inconsistent to clude not just operational definition of the objective exis- imagine observer perceiving something else than what is tence of the einselected states, but – as is also detailed in implied by the stable (einselected) records in his posses- Section VI – a clarification of the connection between the sion as it is impossible to imagine the same person with a quantum amplitudes and probabilities. Einselection con- different DNA: Both cases involve information encoded in verts arbitrary states into mixtures of well defined possi- a state of a system inextricably linked with the physical bilities. Phases are envariant: Appreciation of envariance identity of an individual. as a symmetry tied to the ignorance about the state of Distinct memory/identity states of the observer (that the system was the missing ingredient in the attempts of are also his “states of knowledge”) cannot be superposed: ‘no collapse’ derivations of Born’s rule and in the prob- This censorship is strictly enforced by decoherence and ability interpretation. While both envariance and the the resulting einselection. Distinct memory states label “environment as a witness” point of view are only begin- and “inhabit” different branches of the Everett’s “Many ning to be investigated, the extension of the program of Worlds” Universe. Persistence of correlations is all that is einselection they offer allowes one to understand emer- needed to recover “familiar reality”. In this manner, the gence of “classical reality” form the quantum substrate distinction between epistemology and ontology is washed as a consequence of quantum laws. away: To put it succinctly (Zurek, 1994) there can be no information without representation in physical states. II. QUANTUM MEASUREMENTS There is usually no need to trace the collapse all the way to observer’s memory. It suffices that the The need for a transition from quantum determinism of states of a decohering system quickly evolve into mix- the global state vector to classical definiteness of states tures of the preferred (pointer) states. All that can be of individual systems is traditionally illustrated by the known in principle about a system (or about an observer, example of quantum measurements. An outcome of a also introspectively, e.g., by the observer himself) is its “generic” measurement of the state of a quantum sys- decoherence-resistant ‘identity tag’ – a description of its tem is not deterministic. In the textbook discussions einselected state. this random element is blamed on the “collapse of the Apart from this essentially negative function of a cen- wavepacket”, invoked whenever a quantum system comes sor the environment plays also a very different role of into contact with a classical apparatus. In a fully quan- a “broadcasting agent”, relentlessly cloning the informa- tum discussion this issue still arises, in spite (or rather tion about the einselected pointer states. This role of because) of the overall deterministic quantum evolution the environment as a witness in determining what exists of the state vector of the Universe: As pointed out by von was not appreciated until now: Throughout the past two Neumann (1932), there is no room for a ‘real collapse’ in decades, study of decoherence focused on the effect of the purely unitary models of measurements. the environment on the system. This has led to a multitude of technical advances we shall review, but it has also missed one crucial point of paramount conceptual A. Quantum conditional dynamics importance: Observers monitor systems indirectly, by intercepting small fractions of their environments (e.g., a To illustrate the ensuing difficulties, consider a quan- fraction of the photons that have been reflected or emit- tum system initially in a state ψ interacting with a S | i ted by the object of interest). Thus, if the understand- quantum apparatus initially in a state A0 : ing of why we perceive quantum Universe as classical A | i is the principal aim, study of the nature of accessibil- Ψ = ψ A = a s A | 0i | i| 0i i| ii | 0i ity of information spread throughout the environment Xi should be the focus of attention. This leads one away a s A = Ψ . (1) from the models of measurement inspired by the “von −→ i| ii| ii | ti Xi Neumann chain” (1932) to studies of information transfer involving branching out conditional dynamics and the Above, A and s are states in the Hilbert spaces {| ii} {| ii} resulting “fan-out” of the information throughout envi- of the apparatus and of the system, respectively, and ai 6 are complex coefficients. Conditional dynamics of such of the two qubits will be also unambiguous, as in the clas- premeasurement (as the step achieved by Eq. (2.1) is sical case. This expectation is incorrect. In the conjugate often called) can be accomplished by means of a unitary basis + , defined by: Schrödinger evolution. Yet it is not enough to claim that {| i |−i} a measurement has been achieved: Equation (2.1) leads = ( 0 1 )/√2 , (2.6) |±i | i ± | i to an uncomfortable conclusion: Ψt is an EPR-like entangled state. Operationally, this EPR| i nature of the state the truth table of Eq. (2.3) along with Eq. (2.6) lead to emerging from the premeasurement can be made more a new complementary truth table: explicit by re-writing the sum in a different basis: + + ; . (4) |±i| i −→ |±i| i |±i|−i −→ |∓i|−i Ψ = a s A = b r B . (2) | ti i| ii| ii i| ii| ii In the complementary basis + , roles of the con- Xi Xi trol and of the target are reversed:{| i |−i} The former (basis This freedom of basis choice – basis ambiguity – is guar- 0 , 1 ) target – represented by the second ket above – anteed by the principle of superposition. Therefore, if {|remainsi | i} unaffected, while the state of the former control one were to associate states of the apparatus (or the ob- (the first ket) is conditionally “flipped”. server) with decompositions of Ψt , then even before en- In the bit-by-bit case the measurement interaction is: quiring about the specific outcome| i of the measurement one would have to decide on the decomposition of Ψt ; Hint = g 1 1 = | i g | ih |S |−ih−|A the change of the basis redefines the measured quantity. 1 1 (1 ( 0 1 + 1 0 )) (5) 2| ih |S ⊗ − | ih | | ih | A

1. Controlled not and a bit-by-bit measurement Above, g is a coupling constant, and the two operators refer to the system (i.e., to the former control), and to The interaction required to entangle the measured sys- the apparatus pointer (the former target), respectively. It is easy to see that the states 0 , 1 of the system tem and the apparatus, Eq. (2.1), is a generalization of {| i | i}S the basic logical operation known as a “controlled not” are unaffected by Hint, since; or a c-not. Classical, c-not changes the state at of the [Hint,e0 0 0 + e1 1 1 ]=0 (2.10) target when the control is 1, and does nothing otherwise: | ih |S | ih |S The measured (control) observableǫ ˆ = e 0 0 + e 1 1 0c at 0c at ; 1c at 1c at (2.3) 0| ih | 1| ih | −→ −→ ¬ is a constant of motion under Hint. c-not requires inter- Quantum c-not is a straightforward quantum version of action time t such that gt = π/2. Eq. (2.3). It was known as a “bit by bit measurement” The states + , of the apparatus encode the {| i |−i}A (Zurek, 1981; 1983) and used to elucidate the connection information about phase between the logical states. They between entanglement and premeasurement already be- have exactly the same “immunity”: fore it acquired its present name and significance in the [Hint,f+ + + + f ]=0 (2.11) context of quantum computation (see e.g. Nielsen and | ih |A −|−ih−|A Chuang, 2000). Arbitrary superpositions of the control Hence, when the apparatus is prepared in a definite phase bit and of the target bit states are allowed: state (rather than in a definite pointer/logical state), it (α 0c + β 1c ) at will pass on its phase onto the system, as Eqs. (2.7) - | i | i | i (2.8), show. Indeed, Hint can be written as: α 0c at + β 1c at (3) −→ | i| i | i|¬ i Hint = g 1 1 Above “negation” at of a state is basis dependent; | ih |S|−ih−|A |¬ i = g (1 ( + + + )) (6) 2 − |−ih | | ih−| S ⊗ |−ih−|A (γ 0t + δ 1t )= γ 1t + δ 0t (2.5) ¬ | i | i | i | i making this “immunity” obvious. With A0 = 0t , A1 = 1t we have an obvious anal- This basis-dependent direction of information flow in ogy between| i the| ic-not| i and| ai premeasurement. a quantum c-not (or in a premeasurement) is a con- In the classical controlled not the direction of informa- sequence of complementarity. While the information tion transfer is consistent with the designations of the two about the observable with the eigenstates 0 , 1 trav- bits: The state of the control remains unchanged while it els from the system to the apparatus, in{| thei | comple-i} influences the target, Eq. (2.3). Classical measurement mentary + , basis it seems that the apparatus is need not influence the system. Written in the logical ba- measured{| byi the|−i} system. This remark (Zurek 1998a, b; sis 0 , 1 , the truth table of the quantum c-not is see also Beckman, Gottesman, and Nielsen, 2001) clari- essentially{| i | i} – that is, save for the possibility of superpo- fies the sense in which phases are inevitably “disturbed” sitions – the same as Eq. (2.3). One might have antici- in measurements. They are not really destroyed, but, pated that the direction of information transfer and the rather, as the apparatus measures a certain observable designations (“control/system” and “target/apparatus”) of the system, the system simultaneously “measures” 7 phases between the possible outcome states of the ap- The interaction Hamiltonian: paratus. This leads to loss of phase coherence: Phases become “shared property” as we shall see in more detail Hint = gsˆBˆ (2.19) in the discussion of envariance. The question “what measures what?” (decided by the is an obvious generalization of Eqs. (2.9) and (2.12), with direction of the information flow) depends on the initial g the coupling strength ands ˆ: states. In “the classical practice” this ambiguity does not N 1 arise. Einselection limits the set of possible states of the − sˆ = l s s (2.20) apparatus to a small subset. | lih l| Xl=0

In the A basis Bˆ is a shift operator, 2. Measurements and controlled shifts. {| ki} iN ∂ The truth table of a whole class of c-not like trans- Bˆ = . (2.21) formations that includes general premeasurement, Eq. 2π ∂Aˆ (2.1), can be written as: To show how Hint works, we compute:

sj Ak sj Ak+j (2.13) | i| i −→ | i| i exp( iH t/¯h) s A = − int | j i| ki Equation (2.1) follows when k = 0. One can therefore 1 N 1 sj N − 2 − exp[ i(jgt/¯h +2πk/N)l] Bl (7) model measurements as controlled shifts – c-shifts – | i l=0 − | i c-not P generalizations of the . In the bases sj and We now adjust the coupling g and the duration of the A , the direction of the information ﬂow{| appearsi} to {| ki} interaction t so that the action ι expressed in Planck be unambiguous – from the system to the apparatus . units 2π¯h is a multiple of 1/N; However, a complementary basis canS be readily deﬁnedA (Ivanovic, 1981; Wootters and Fields, 1989); ι = gt/¯h = G 2π/N . (2.23a) ∗ N 1 1 − 2πi For an integer G, Eq. (2.22) can be readily evaluated: B = N − 2 exp( kl) A . (2.14a) | ki N | li Xl=0 exp( iHintt/¯h) sj Ak = sj A k+G j . (8) − | i| i | i| { ∗ }N i Above N is the dimensionality of the Hilbert space. Anal- ogous transformation can be carried out on the basis This is a shift of the apparatus state by an amount G j proportional to the eigenvalue j of the state of the sys-∗ si of the system, yielding states rj . {| Orthogonalityi} of A implies: {| i} tem. G plays the role of gain. The index k + G j N is {| ki} evaluated modulo N, where N is the number{ of the∗ } pos-

Bl Bm = δlm . (2.15) sible outcomes, that is, the dimensionality of the Hilbert h | i space of the apparatus pointer : When G j>N, the pointer will just rotate throughA the initial∗ “zero”. The N 1 c-shift 1 − 2πi truth table for G = 1 deﬁnes a , Eq. (2.13), and A = N − 2 exp( kl) B (2.14b) | ki − N | li with k = 0 leads to a premeasurement, Eq. (2.1). Xl=0 The form of the interaction, Eq. (2.19), in conjunc- tion with the initial state decide the direction of infor- inverts of the transformtion of Eq. (2.14a). Hence: mation transfer. Note that – as was the case with the c-nots – the observable that commutes with the interac- ψ = α A = β B , (2.16) | i l| li k| ki tion Hamiltonian will not be perturbed: Xl Xk where the coeﬃcients βk are: [Hint, sˆ]=0 (2.25)

N 1 sˆ commutes with Hint, and is therefore a non-demolition 1 − 2πi βk = N − 2 exp( kl)αl . (2.17) observable (Braginsky, Vorontsov and Thorne, 1980; − N Xl=0 Caves et al, 1980; Bocko and Onofrio, 1996). Hadamard transform of Eq. (2.6) is a special case of the more general transformation considered here. 3. Amplification To implement the truth tables involved in premeasurements we define observable Aˆ and its conjugate: Amplification was often regarded as the process forcing N 1 N 1 quantum potentialities to become classical reality. Its − − Aˆ = k A A ; Bˆ = l B B . (2.18a, b) example is an extension of the model of measurement | kih k| | lih l| kX=0 Xl=0 described above. 8

Assume the Hilbert space of the apparatus pointer is States of the system that need to be distinguished should large compared with the space spanned by the eigenstates rotate the pointer of the apparatus to the correlated out- of the measured observables ˆ: come states that are approximately orthogonal. When the coeﬃcients α(k) are peaked around k = 0 with dis- N = Dim( ) Dim( )= n (2.26) persion ∆, this implies: HA ≫ HS Then one can increase ι to an integer multiple G of 2π/N. ∆ G . (2.33) This was implicit in Eqs. (2.23a) and (2.24). However, ≪ larger ι will lead to redundancy only when the appara- In the general case of an initial mixture, Eq. (2.29), tus the Hilbert space has many more dimensions in then one can evaluate the dispersion of the expectation value there are possible outcomes. Otherwise, only “wrapping” of the record observable Aˆ as: of the same record will ensue. The simplest example of c-not 2 such wrapping: ( ) is the identity. For N n, Aˆ2 Aˆ 2 = Trρ0 Aˆ2 (Trρ0 Aˆ)2 (2.34) however, one can attain gain: ≫ h i − h i A − A The outcomes are distinguishable when: G = Ngt/2π¯h . (2.23b) Aˆ2 Aˆ 2 G (2.35) The outcomes are now separated by G 1 “empty” eigen- h i − h i ≪ states of the record observable. In this− sense G 1 Interaction with the environment yields a mixture of the achieves redundancy, providing that wrapping of≫ the form of Eq. (2.29). Ampliﬁcation can protect measure- record is avoided, which is guaranteed when: ment outcomes from noise through redundancy.3 nG

Amplification is useful in presence of noise. For exam- 3 ple, it may be difficult to initiate the apparatus in A0 , The above model of amplification is unitary. Yet, it contains c-shift so the initial state may be a superposition; | i seeds of irreversibility. Reversibility of is evident: As the interaction continues, the two systems will eventually disentangle. For instance, it takes te = 2π¯h/(gN) (see Eq. (2.23b) a = α (k) A . (2.28a) | li l | ki with G = 1) to entangle S (Dim(HS ) = n) with an A with Xk Dim(HA) = N ≥ n pointer states. However, as the interaction continues, A and S disentangle. For a c-shift this recur- Indeed, typically a mixture of such superpositions; rence timescale is: tRec = Nte = 2π¯h/g. It corresponds to gain G = N. Thus, for an instant of less than te at t = tRec the 0 apparatus disentangles from the system, as {k + N ∗ j}N = k. ρ = wi ai ai (2.28b) A | ih | Reversibility results in recurrences of the initial state, but for Xi N ≫ 1, they are rare. may be the starting point for a premeasurement. Then: For less regular interactions (e.g, involving environment) recur- rence time is much longer. In that case, tRec is, in effect, a Poincarétime: tRec ∼ tP oincare´ ≈ N!te. In any case tRec ≫ te sk sk′ ρ = sk sk′ wl al al for large N. Undoing entanglement in this manner would be | ih | A | ih | | ih | Xl exceedingly difficult because one would need to know precisely when to look, and because one would need to isolate the appara- s s ′ w a a ′ (9) −→ | kih k | l| l+Gkih l+Gk | tus or the immediate environment from other degrees of freedom Xl – their environments. The price of letting the entanglement undo itself by waiting for where al+Gk obtains from al , Eq. (2.28a), through: an appropriate time interval is at the very least given by the cost | i | i of storing the information on how long is it necessary to wait. c-shift a = α (j) A , (2.30) In the special case this is ∼ log N memory bits. In sit- | l+Gki l | j+Gki uations when eigenvalues of the interaction Hamiltonian are not Xj commensurate, it will be more like ∼ log N! ≈ N log N, as the entanglement will get undone only after a Poincarétime. Both and the simplifying assumption about the coefficients; classical and quantum case can be analyzed using algorithmic information . For related discussions see Zurek, (1989), Caves, αl(j)= α(j l) (2.31) (1994); Schack and Caves (1996) and Zurek (1998b). − Amplified correlations are hard to contain. The return to pu- has been made. Its aim to focus on the case when the rity after tRec in the manner described above can be hoped apparatus states are peaked around a certain value l (e.g., for only when either the apparatus or the immediate environment E (i.e., the environment directly interacting with the sys- α (j) exp( (j l)2/2∆2), and where the form of their l ∼ − − tem) cannot “pass on” the information to their more remote distribution over A does not depend on l. ′ {| ki} environments E . The degree of isolation required puts a strin- A good measurement allows one to distinguish states gent limit on the coupling gEE′ between the two environments: of the system. Hence, it must satisfy: Return to purity can be accomplished in this manner only if ′ ′ tRec < te′ = 2π¯h/(N gEE′ ), where N is the dimension of the 2 Hilbert space of the environment E′. Hence, the two estimates al+Gk al+Gk′ ′ |h | i| of tRec translate into: gEE′ < g/N for the regular spectrum, 2 ′ = α(j + G(k k′))α∗(j) δ ′ . (10) and the much tighter; g ′ < g/N!N for the random case more | j − | ≈ k ,k EE P 9

B. Information transfer in measurements 1. Action per bit

Information transfer is the objective of the measure- An often raised question concerns the price of infor- ment process. Yet, quantum measurements were only mation in units of some other “physical currency”, (Bril- rarely analyzed from that point of view. As a result of louin, 1962; 1964; Landauer, 1991). Here we shall es- the interaction of the system with the apparatus , tablish that the least action necessary to transfer one bit S A their joint state is still pure Ψt , Eq. (2.1), but each of is of the order of a fraction ofh ¯ for quantum systems the subsystems is in a mixture:| i with two-dimensional Hilbert spaces. Information trans-

N 1 fer can be made cheaper on the “wholesale” level, when − 2 the systems involved have large Hilbert spaces. ρ = T r Ψt Ψt = ai si si ; (11) S A| ih | | | | ih | Consider Eq. (2.1). It evolves initial product state Xi=0 N 1 of the two subsystems into a superposition of product − 2 states, ( αj sj ) A0 αj sj Aj . The expec- ρ = T r Ψt Ψt = ai Ai Ai . (12) j | i | i −→ j | i| i A S | ih | | | | ih | tation value of the action involved is no less than: Xi=0 P P N 1 Partial trace leads to reduced density matrices, here ρ − S I = α 2 arccos A A (2.39) and ρ , important for what follows. They describe sub- | j | |h 0| j i| systemsA to the observer who, before the premeasurement, Xj=0 knew pure states of the system and of the apparatus, but When A are mutually orthogonal, the action is: who has access to only one of them afterwards. {| j i} Reduced density matrix is a technical tool of I = π/2 (2.40) paramount importance. It was introduced by Landau (1927) as the only density matrix that gives rise to the in Planck (¯h) units. This estimate can be lowered by correct measurement statistics given the usual formalism using the initial A , a superposition of the outcomes | 0i that includes Born’s rule for calculating probabilities (see Aj . In general, interaction of the form: e.g. p. 107 of Nielsen and Chuang, 2000, for an insight- | i N 1 N 1 ful discussion). This remark will come to haunt us later − − when in Section VI we shall consider the relation between H = ig sk sk ( Ak Al h.c.) (15) SA | ih | | ih |− decoherence and probabilities: In order to derive Born’s Xk=0 Xl=0 rule it will be important not to assume it in some guise! Following premeasurement, the information about the saturates the lower bound given by: subsystems available to the observer locally decreases. This is quantiﬁed by the increase of the entropies: I = arcsin 1 1/N , (2.42) p − N 1 − For a two-dimensional Hilbert space the average action 2 2 H = Trρ log ρ = ai log ai can be thus brought down to π¯h/4 (Zurek, 1981; 1983). S − S S − | | | | Xi=0 As the size of the Hilbert space increases, action in- = Trρ log ρ = H (13) volved approaches the asymptotic estimate of Eq. (2.40). − A A A The entropy of entanglement can be as large as log N As the evolution of the whole is unitary, the increase SA where N is the dimension of the Hilbert space of the of entropies in the subsystems is compensated by the smaller of the two systems. Thus, the least action per increase of the mutual information: bit of information decreases with the increase of N: N 1 − 2 2 I π ( : )= H + H H = 2 ai log ai (14) I S A S A − SA − | | | | ι = (2.43) Xi=0 log2 N ≈ 2log2 N It was used in quantum theory as a measure of entangle- This may be one reason why information appears “free” ment (Zurek, 1983; Barnett and Phoenix, 1989). in the macroscopic domain, but expensive (close toh ¯/bit) in the quantum case of small Hilbert spaces.

relevant for decoherence. C. “Collapse” analogue in a classical measurement In short, once information “leaks” into the correlations between the system and the apparatus or the environment, keeping it from spreading further ranges between very hard and next to Deﬁnite outcomes we perceive appear to be at odds impossible. With the exception of very special cases (small N, with the principle of superposition. They can neverthe- regular spectrum), the strategy of “enlarging the system, so that less occur also in quantum physics when the initial state it includes the environment” – occasionally mentioned as an ar- of the measured system is – already before the measure- gument against decoherence – is doomed to fail, unless the Uni- verse as a whole is included. This is a questionable setting (as ment – in one of the eigenstates of the measured observ- the observers are inside this “isolated” system) and in any case able. Then Eq. (2.1) will deterministically rotate the makes the relevant Poincar´etime absurdly long! pointer of the apparatus to the appropriate record state. 10

The result of such a measurement can be predicted by the discoverer) about the measurement outcome, will de- an insider – an observer aware of the initial state of the scribe the same process still diﬀerently: system. This a priori knowledge can be represented by the preexisting record Ai , which is only corroborated A A A A p σ σ | i | ρS ih ρS | | 0ih 0| i| iih i| −→ by an additional measurement: Xi

AρS AρS ( pi Ai Ai σi σi ) . (17) Ai A0 σi Ai Ai σi . (2.44a) −→ | ih | | ih | | ih | | i| i| i −→ | i| i| i Xi In classical physics complete information about the initial This view of the outsider, Eq. (2.44c), combines one-to- state of an isolated system always allows for an exact one classical correlation of the states of the system and prediction of its future state: A well-informed observer the records with the indefiniteness of the outcome. will be even able to predict future of the classical Universe We have just seen three distinct quantum-looking de- as a whole (“Laplace’s demon”). scriptions of the very same classical process (see Zurek, Any element of surprise (any use of probabilities) must 1989, and Caves, 1994 for previous studies of the insider be therefore blamed on partial ignorance. Thus, when the - outsider theme). They differ only in the information information available initially does not include the exact available ab initio to the observer. The information in initial state of the system, observer can use an ensem- possession of the observer prior to the measurement de- ble described by ρ – by a list of possible initial states termines in turn whether – to the observer – the evolution σ and their probabilitiesS p . This is the ignorance {| ii} i appears to be (a) a confirmation of the preexisting data, interpretation of probabilities. We shall see in section VI Eq. (2.44a), (b) a “collapse” associated with the informa- that – using quantum envariance – one can justify igno- tion gain, Eq. (2.44b) – and with the entropy decrease rance about a part of the system by relying on perfect translated into algorithmic randomness of the acquired knowledge of the whole. data (Zurek, 1989; 1998b) – or (c) an entropy-preserving Through measurement observer finds out which of the establishment of a correlation, Eq. (2.44c). All three de- potential outcomes consistent with his prior (lack of) in- scriptions are classically compatible, and can be imple- formation actually happens. This act of acquisition of mented by the same (deterministic and reversible) dy- information changes physical state of the observer – the namics. state of his memory: The initial memory state contain- In classical physics the insider view always exists in ing description AρS of an ensemble and a “blank” A0, principle. In quantum physics it does not. Every ob- AρS AρS A0 A0 , is transformed into record of a spe- server in a classical Universe could in principle aspire to cific| outcome:ih | | ihA | A A A . In quantum notation | ρS ih ρS | | iih i| be an ultimate insider. The fundamental contradiction this process will be described by such a discoverer as a between every observer knowing precisely the state of the random “collapse”: rest of the Universe (including the other observers) can be swept under the rug (if not really resolved) in the

AρS AρS A0 A0 pi σi σi Universe where the states are inﬁnitely precisely deter- | ih | | ih | | ih | Xi mined and the observer’s records (as a consequence of the

AρS AρS Ai Ai σi σi . (16) ¯h 0 limit) may have an infinite capacity for informa- −→ | ih | | ih | | ih | tion→ storage. However – given a set value ofh ¯ – informa- This is only the description of what happens “as reported tion storage resources of any finite physical system are fi- by the discoverer”. Deterministic representation of this nite. Hence, in quantum physics observers remain largely very same process by Eq. (2.44a) is still possible. In ignorant of the detailed state of the Universe as there can other words, in classical physics discoverer can be always be no information without representation (Zurek, 1994). convinced that the system was in a state σi already Classical “collapse” is described by Eq. (2.44b): The | i before the record is made in accord with Eq. (2.44b). observer discovers the state of the system. From then on, This sequence of events as seen by the discoverer looks the state of the system will remain correlated with his like a “collapse” (see also Zurek, 1998a,b): For instance, record, so that all the future outcomes can be predicted, insider who knew the state of the system before discov- in effect by iterating Eq. (2.44a). This disappearance of erer carried out his measurement need not notice any all the potential alternatives save for one that becomes a change of that state when he makes further “confirma- “reality” is the essence of the collapse. There need not tory” measurements. This property is the cornerstone be anything quantum about it. of the “reality” of classical states – they need not ever Einselection in observers memory provides many of change as a consequence of measurements. We empha- the ingredients of the “classical collapse” in the quan- size, however, that while the state of the system may tum context. In presence of einselection, one-to-one cor- remain unchanged, the state of the observer must change respondence between the state of the observer and his to reflect the acquired information. knowledge about the rest of the Universe can be firmly Last not least, an outsider – someone who knows about established, and (at least in principle) operationally ver- the measurement, but (in contrast to the insider) not ified: One could measure bits in observers memory and about the initial state of the system nor (in contrast to determine what he knows without altering his records – 11 without altering his state. After all, one can do so with of momentum given by the Lyapunov exponents (but see a classical computer. Existential interpretation recog- Boccaletti, Farini, and Arecchi, 1997). In the real chaotic nizes that the information possessed by the observer is system stretching and narrowing of the probability distri- reflected in his einselected state, explaining his percep- bution in both x and p occur simultaneously, as the initial tion of a single “branch” – “his” classical Universe. patch is rotated and folded. Eventually, the envelope of its effective support will swell to fill in the available phase space, resulting in the wavepacket coherently spread over III. CHAOS AND LOSS OF CORRESPONDENCE the spatial region of no less than; ∆x (¯h/∆p ) exp(Λt) (3.2) The study of the relationship between the quantum ∼ 0 and the classical has been – for a long time – focused until it becomes confined by the potential, while the almost entirely on measurements. However, the problem small-scale structure will continue to descend to ever of measurement is difficult to discuss without observers. smaller scales (Fig. 1). Breakdown of the quantum- And once observer enters, it is often hard to avoid its ill- classical correspondence can be understood in two com- understood anthropic attributes such as consciousness, plementary ways, either as a consequence of small ∆p awareness, and the ability to perceive. (see the discussion of Moyal bracket below), or as a re- We shall sidestep these presently “metaphysical” prob- sult of large ∆x. lems and focus on the information-processing underpin- Coherent exponential spreading of the wavepacket – nings of the “observership”. It is nevertheless fortunate large ∆x – must cause problems with correspondence. that there is another problem with the quantum - clas- This is inevitable, as classical evolution appeals to the sical correspondence that leads to interesting questions idealization of a point in phase space acted upon by a not motivated by measurements. As was anticipated by force given by the gradient ∂ V of the potential V (x) Einstein (1917) before the advent of modern quantum x evaluated at that point. But quantum wavefunction can theory, chaotic motion presents such a challenge. Full im- be coherent over a region larger than the nonlinearity plications of classical dynamical chaos were understood scale χ over which the gradient of the potential changes much later. The concern about the quantum-classical significantly. χ can be usually estimated by: correspondence in this modern context dates to Berman and Zaslavsky (1978) and Berry and Balazs (1979) (see χ ∂xV/∂xxxV , (3.3) Haake, 1991 and Casati and Chirikov, 1995a, for refer- ≃ p ences). It has even led some to question validity of quan- and is typically of the order of the size L of the system: tum theory (Ford and Mantica, 1992). L χ . (3.4) ∼ A. Loss of the quantum-classical correspondence An initially localized state evolving in accord with Eqs. (3.1) and (3.2) will spread over such scales after: The interplay between quantum interference and 1 ∆p0χ t Λ− ln . (3.5) chaotic exponential instability leads to the rapid loss of h¯ ≃ ¯h the quantum-classical correspondence. Chaos in dynamics is characterized by the exponential divergence of the It is then impossible to tell what is the force acting on classical trajectories. As a consequence, a small patch the system, as it is not located in any specific x. This es- representing the probability density in phase space is ex- timate of what can be thought of as Ehrenfest time – the ponentially stretching in unstable directions and to expo- time over which a quantum system that has started in nentially compressing in the stable directions. The rates a localized state will continue to be sufficiently localized of stretching and compression are given by positive and for the quantum corrections to the equations of motion obeyed by its expectation values to be negligible (Got- negative Lyapunov exponents Λi. Hamiltonian evolution demands that the sum of all the Lyapunov exponents be tfried, 1966) – is valid for chaotic systems. Logarithmic dependence is the result of inverting of the exponential zero. In fact, they appear in Λi pairs. Loss of the correspondence± in chaotic systems is a con- sensitivity. In the absence of the exponential instability sequence of the exponential stretching of the effective (Λ = 0) divergence of trajectories is typically polyno- mial, and leads to a power law dependence, t (I/¯h)α, support of the probability distribution in the unstable h¯ ∼ direction (say, x) and its exponential narrowing in the where I is the classical action. Thus, macroscopic (large complementary direction (Zurek and Paz, 1994; Zurek, I) integrable systems can follow classical dynamics for a 1998b). As a consequence, classical probability distribu- very long time, providing they were initiated in a local- tion will develop structures on the scale: ized state. For chaotic systems th¯ also becomes infinite in the limith ¯ 0, but that happens only logarithmically → ∆p ∆p0 exp( Λt) . (3.1) slowly. As we shall see below, in the context of quantum- ∼ − classical correspondence this is too slow for comfort. Above, ∆p0 is the measure of the initial momentum Another way of describing the “root cause” of the cor- spread. Λ is the net rate of contraction in the direction respondence breakdown is to note that after the timescale 12 of the order of th¯ quantum wave function of the system demonstrates – into states that are non-local, and, there- would have spread over all of the available space, and fore, extravagantly quantum, simply as a result of the is being forced to “fold” onto itself. Fragments of the exponentially unstable dynamics. Moreover, this hap- wavepacket arrive at the same location (although with pens surprisingly quickly, even for very macroscopic ex- different momenta, and having followed different paths). amples. Hyperion is not the only chaotic system. There The ensuing evolution critically depends on whether they are asteroids that have chaotically unstable orbits (e.g., have retained phase coherence. When coherence persists, Chiron), and even indications that the solar system as a complicated interference event decides the subsequent a whole is chaotic (Laskar, 1989; Sussman and Wisdom, evolution. And – as can be anticipated from the double 1992). In all of these cases straightforward estimates of slit experiment – there is a big difference between coth¯ yield answers much smaller than the age of the solar herent and incoherent folding in the configuration space. system. Thus, if unitary evolution of closed subsystems This translates into the loss of correspondence, which sets was a complete description of planetary dynamics, plan- in surprisingly quickly, at th¯ . ets would be delocalized along their orbits. To find out how quickly, we estimate th¯ for an obviously macroscopic Hyperion, chaotically tumbling moon of Saturn (Wisdom, 1985). Hyperion has a prolate shape B. Moyal bracket and Liouville flow of a potato and moves on an eccentric orbit with a period tO =21 days. Interaction between its gravitational Heuristic argument about breakdown of the quantum- quadrupole and the tidal field of Saturn leads to chaotic classical correspondence can be made more rigorous with 1 tumbling with Lyapunov time Λ− 42 days. the help of the Wigner function. We start with the von To estimate the time over which orientation≃ of Hyper- Neumann equation: ion becomes delocalized, we use a formula (Berman and i¯hρ˙ = [H,ρ] . (3.9) Zaslavsky, 1978, Berry and Balazs, 1979): It can be transformed into the equation for Wigner func- 1 LP 1 I t =Λ− ln =Λ− ln (3.6) tion W , which is defined in phase space: r ¯h ¯h 1 ipy y y W (x, p)= exp( )ρ(x , x + )dy . (3.10) Above L and P give the range of values of the coordinate 2π¯h Z ¯h − 2 2 and momentum in phase space of the system. Since L ≃ χ and P > ∆p0, it follows that tr th¯ . On the other The result is: hand, LP I, the action of the system.≥ ≃ W˙ = H, W . (3.11) The advantage of Eq. (3.6) is its insensitivity to initial { }MB conditions, and the ease with which the estimate can be Here ..., ... MB stands for Moyal bracket, the Wigner obtained. For Hyperion, a generous overestimate of the transform{ of} the von Neumann bracket (Moyal, 1949). classical action I can be obtained from its binding energy Moyal bracket can be expressed in terms of the Pois- EB and its orbital time tO: son bracket ..., ... , which generates Liouville flow in the classical phase{ space,} by the formula: I/¯h E t /¯h 1077 (3.7) ≃ B O ≃ i¯h ..., ... MB = sin(i¯h ..., ... ) . (3.12) The above estimate (Zurek, 1998b) is “astronomically” { } { } large. However, in the calculation of the loss of corre- When the potential V (x) is analytic, Moyal bracket can spondence, Eq. (3.6), only its logarithm enters. Thus, be expanded (Hillery et al, 1984) in powers ofh ¯:

Hyperion 77 ¯h2n( )n tr 42[days] ln 10 20[yrs] (3.8) W˙ = H, W + − ∂2n+1V ∂2n+1W . (18) ≃ ≃ { } 22n(2n + 1)! x p nX1 After approximately 20 years Hyperion would be in a ≥ coherent superposition of orientations that differ by 2π! The first term is just the Poisson bracket. Alone, it We conclude that after a relatively short time an ob- would generate classical motion in phase space. How- viously macroscopic chaotic system becomes forced into ever, when the evolution is chaotic, quantum corrections a flagrantly non-local “Schrödinger cat” state. In the (proportional to the odd order momentum derivatives of original discussion (Schrödinger, 1935a,b) an intermedi- the Wigner function) will eventually dominate the right ate step in which decay products of the nucleus were mea- hand side of Eq. (3.10). This is because the exponential sured to determine the fate of the cat was essential. Thus, squeezing of the initially regular patch in phase space it was possible to maintain that the preposterous super- (which begins its evolution in the classical regime, where position of the dead and alive cat could be avoided pro- the Poisson bracket dominates) leads to an exponential viding that quantum measurement (with the “collapse” explosion of the momentum derivatives. Consequently, it presumably induces) was properly understood. after a time logarithmic inh ¯, (Eqs. (3.5), (3.6)) the Pois- This cannot be the resolution for chaotic quantum sys- son bracket will cease to be a good estimate of the right tems. They can evolve – as the example of Hyperion hand side of Eq. (3.13). 13

The physical reason for the ensuing breakdown of the but also averages diverge at a time th¯ . quantum-classical correspondence was already explained In integrable systems rapid breakdown∼ of correspon- before: Exponential instability of the chaotic evolution dence may still occur, but only for very special initial delocalizes the wavepacket. As a result, the force acting conditions. It is due to the local instability in phase on the system is no longer given by the gradient of the space. Indeed, a double - slit experiment is an example potential evaluated at the location of the system: It is not of a regular system in which a local instability (splitting even possible to say where the system is, since it is in a of the paths) leads to correspondence loss, but only for superposition of many distinct locations. Consequently, judiciously selected initial conditions. Thus, one may dis- phase space distribution and even the typical observables miss it as a consequence of a rare pathological starting of the system noticeably differ when evaluated classi- point, and argue that the conditions that lead to discrep- cally and quantum mechanically (Haake, Ku´sand Sharf, ancies between classical and quantum behavior exist, but 1987; Habib, Shizume and Zurek, 1998; Karkuszewski, are of measure zero in the classical limit. Zakrzewski, and Zurek, 2002). Moreover, this will hap- In the chaotic case the loss of correspondence is typi- pen after an uncomfortably short time th¯ . cal. As shown in Fig. 2, it happens after a disturbingly short th¯ for generic initial conditions. The time at which the quantum and classical expectation values diverge in C. Symptoms of correspondence loss the example studied here is consistent with the estimates of th¯ , Eq. (3.5), but exhibits a significant scatter. This is Wavepacket becomes rapidly delocalized in a chaotic not too surprising – exponents characterizing local insta- system, and the correspondence between classical and bility vary noticeably with the location in phase space. quantum is quickly lost. Schrödinger cat states appear Hence, stretching and contraction in phase space will occur at a rate that depends on the specific trajectory. De- after th¯ and this is the overarching interpretational as well as physical problem: In the familiar real world we pendence of its typical magnitude as a function ofh ¯ is never seem to encounter such smearing of the wavefunc- still not clear. Emerson and Ballentine (2001a&b) study tion even in the examples of chaotic dynamics it is pre- coupled spins and argue that it is of order ofh ¯, but Fig. dicted by quantum theory. 2 suggests it decreases more slowly than that, and that it may be only logarithmic inh ¯ (Karkuszewski et al. 2002).

1. Expectation values 2. Structure saturation

Measurements often average out fine phase space struc- Evolution of the Wigner function exhibits rapid tures, which may be striking, but experimentally inac- buildup of interference fringes. These fringes become pro- cessible symptom of breakdown of the correspondence. gressively smaller, until saturation when the wavepacket Thus, one might hope that when interference patterns in is spread over the available phase space. At that time the Wigner function are ignored by looking at the coarse- their scales in momentum and position are typically: grained distribution, the quantum results should be in accord with the classical. This would not exorcise the dp =¯h/L ; dx =¯h/P (3.14, 3.15) ‘chaotic cat’ problem. Moreover, breakdown of correspondence can be also seen in the expectation values of where L (P ) defines the range of positions (momenta) of quantities that are smooth in phase space. the effective support of W in phase space. Trajectories diverge exponentially in a chaotic system. Hence, smallest structures in the Wigner function oc- A comparison between expectation values for a single tra- cur (Zurek, 2001) on scales corresponding to action: jectory and for a delocalized quantum state (which is how a = dx dp =h ¯ ¯h/LP =h ¯2/I , (3.16) the Ehrenfest theorem mentioned above is usually stated) × would clearly lead to a rapid loss of correspondence. One where I LP is the classical action of the system. Ac- may object to the use of a single trajectory and argue tion a ≃¯h for macroscopic I. that both quantum and classical state be prepared and Sub-Planck≪ structure is a kinematic property of quan- accessed only through measurements that are subject to tum states. It limits their sensitivity to perturbations, the Heisenberg indeterminacy. Thus, it should be fair and has applications outside quantum chaos or deco- to compare averages over an evolving Wigner function herene. For instance, Schrödinger cat state can be used with an initially identical classical probability distribu- as a weak force detector (Zurek, 2001), and its sensitivity tion (Ballentine, Yang, and Zibin, 1994; Haake, Ku´s& is determined by Eqs. (3.14)-(3.16). Sharf, 1987; Fox and Elston, 1994; Miller, Sarkar and Structure saturation on scale a is an important distinc- Zarum, 1998). These are shown in Fig. 2 for an exam- tion between quantum and classical. In chaotic systems ple of a driven chaotic system. Clearly, there is reason smallest structures in the classical probability density ex- for concern: Fig. 2 (corroborated by other studies, see ponentially shrink with time, in accord with Eq. (3.1) e.g. Karkuszewski, Zakrzewski, and Zurek, 2002, for ref- (see Fig. 1). Equation (3.16) has implications for deco- erences) demonstrates that not just phase space portrait herence, as a controls sensitivity of the systems as well 14 as environments (Zurek, 2001; Karkuszewski, Jarzyn- space alternatives – they retain correlations in spite of ski, and Zurek, 2002): As a result of smallness of a, their immersion in the environment. Eq. (3.16), and as anticipated by Peres (1993), quantum Three quantum systems – the measured system , systems are more sensitive to perturbations when their the apparatus , and the environment – and corre-S classical counterparts are chaotic (see also Jalabert and lations betweenA them are the subject ofE our study. In Pastawski, 2001). But – in contrast to classical chaotic pre-measurements and interact. Their resulting en- systems – they are not exponentially sensitive to infinites- tanglement transformsS intoA an effectively classical corre- imally small perturbations: Rather, the smallest pertur- lation as a result of the interaction between and . bations that can effective are set by Eq. (3.16). This triangle helps analyze decoherenceA E and SAE Emergence of Schrödinger cats through dynamics is a study its consequences. By keeping all three corners of challenge to quantum - classical correspondence. It is this triangle in mind, one can avoid confusion, and main- not yet clear to what extent one should be concerned tain focus on the correlations between, e.g., the memory about the discrepancies between quantum and classical of the observer and the state of the measured system. averages: The size of this discrepancy may or may not be The evolution from a quantum entanglement to the clas- negligible. But in the original Schrödinger cat problem sical correlation may be the easiest relevant theme to quantum and classical expectation values (for the sur- define operationally. In the language of the last part of vival of the cat) were also in accord. In both cases it is Section II, we are about to justify the “outsider” point ultimately the state of the cat which is most worrisome. of view, Eq. (2.44c), before considering the “discoverer”, Note that we have not dealt with dynamical localiza- Eq. (2.44b) and the issue of the “collapse”. In spite of tion (Casati and Chirikov, 1995a). This is because it this focus on correlations, we shall often suppress one 1 appears after too long a time ( ¯h− ) to be a primary of the corners of the triangle to simplify notation. SAE concern in the macroscopic limit,∼ and is quite sensitive All three parts will however play a role in formulating to small perturbations of the potential (Karkuszewski, questions and in motivating criteria for classicality. Zakrzewski, and Zurek, 2002).

A. Models of einselection IV. ENVIRONMENT – INDUCED SUPERSELECTION The simplest case of a single act of decoherence involves just three one-bit systems (Zurek, 1981; 1983). The principle of superposition applies only when the They are denoted by , , and in an obvious refer- quantum system is closed. When the system is open, ence to their roles. TheS A measurementE starts with the interaction with the environment results in an inces- interaction of the measured system with the apparatus: sant monitoring of some of its observables. As a result, pure states turn into mixtures that rapidly diagonalize A0 A1 , A0 A0 ; (4.1a, b) in the einselected states. These pointer states are cho- | ↑i| i −→ | ↑i| i | ↓i| i −→ | ↓i| i sen with the help of the interaction Hamiltonian and are where A0 A1 = 0. For a general state: independent of the initial state of the system. Their h | i (α +β ) A α A +β A = Φ . (4.2) predictability is key to the eﬀective classicality (Zurek, | ↑i | ↓i | 0i −→ | ↑i| 1i | ↓i| 0i | i 1993a; Zurek, Habib, and Paz, 1993). These formulae are an example of a c-not like the pre- Environments can be external (such as particles of measurement discussed in section 2. the air or photons that scatter oﬀ, say, the apparatus The basis ambiguity – the ability to re-write Φ , Eq. pointer) or internal (collections of phonons or other in- (4.2), in any basis of, say, the system, with the superposi-| i ternal excitations). Often, environmental degrees of free- tion principle guaranteeing existence of the correspond- dom emerge from the split of the original set of degrees ing pure states of the apparatus – disappears when an of freedom into the “system of interest” which may be a additional system, , performs a premeasurement on : collective observable (e.g., an order parameter in a phase E A transition), and the “microscopic remainder”. (α A1 + β A0 ) ε0 The set of einselected states is called the pointer ba- | ↑i| i | ↓i| i | i α A ε + β A ε = Ψ . (19) sis (Zurek, 1981) in recognition of its role in measure- −→ | ↑i| 1i| 1i | ↓i| 0i| 0i | i ments. The criterion for the einselection of states goes A collection of three correlated quantum systems is no well beyond the often repeated characterizations based longer subject to the basis ambiguity we have pointed on the instantaneous eigenstates of the density matrix. out in connection with the EPR-like state Φ , Eq. (4.2). What is of the essence is the ability of the einselected This is especially true when states of the| environmenti states to survive monitoring by the environment. This are correlated with the simple products of the states of heuristic criterion can be made rigorous by quantifying the apparatus - system combination (Zurek, 1981; Elby predictability of the evolution of the candidate states, and Bub, 1994). In Eq. (4.3) this can be guaranteed or of the associated observables. Einselected states pro- (irrespective of the value of α and β) providing that: vide optimal initial conditions: They can be employed for the purpose of prediction better than other Hilbert ε ε =0 . (4.4) h 0| 1i 15

When this orthogonality condition is satisfied, the state evolves into: of the pair is given by a reduced density matrix: A − S Φ(t) = a (t) + b (t) ; (4.9) | i | ⇑i|E⇑ i | ⇓i|E⇓ i ρ = T r Ψ Ψ = AS E 2 | 2ih | α A1 A1 + β A0 A0 (20) N | | | ↑ih↑ || ih | | | | ↓ih↓ || ih | igk t igk t (t) = (αke k + βke− k)= ( (24)t) . |E⇑ i | ↑i | ↓i |E⇓ − i containing only classical correlations. kY=1 If the condition of Eq. (4.4) did not hold – that is, if the orthogonal states of the environment were not corre- The reduced density matrix is: lated with the apparatus in the basis in which the original 2 ρ = a + ab∗r(t) A premeasurement was carried out – then the eigenstates of | | | ⇑ih⇑ | 2| ⇑ih⇓ | + a∗br∗(t) + b . (25) the reduced density matrix ρ would be sums of prod- | ⇓ih⇑ | | | | ⇓ih⇓ | ucts rather than simply productsAS of states of and . S A The coefficient r(t)= determines the relative size Extreme example of this situation is the pre-decoherence of the off-diagonal terms.hE⇑|E It⇓i is given by: density matrix of the pure state: N 2 2 2 Φ Φ = α A1 A1 + αβ∗ A1 A0 r(t)= [cos2gkt + i( αk βk )sin2gkt] . (26) | ih | | | | ↑ih↑ || ih | 2 | ↑ih↓ || ih | | | − | | +α∗β A A + β A A (21) kY=1 | ↓ih↑ || 0ih 1| | | | ↓ih↓ || 0ih 0| Its eigenstate is Φ . When expanded, Φ Φ contains For large environments consisting of many (N) spins at terms that are off-diagonal| i when expressed| ih in any| of the large times the off-diagonal terms are typically small: natural bases consisting of the tensor products of states N 2 N 2 2 2 in the two systems. Their disappearance as a result of r(t) 2− [1 + ( α β ) ] (4.13) | | ≃ | k| − | k| tracing over the environment removes the basis ambigu- kY=1 ity. Thus, for example, the reduced density matrix ρ , Eq. (4.5a), has the same form as the outsider descriptionAS The density matrix of any two-state system can be of the classical measurement, Eq. (2.44c). represented by a point in the 3-D space. In terms of In our simple model pointer states are easy to charac- the coefficients a, b, and r(t) coordinates of the point 2 2 terize: To leave pointer states untouched, the Hamilto- representing it are; z = ( a b ), x = (ab∗r), y = | | − | |2 2 ℜ2 (ab∗r). When the state is pure, x + y + z = 1: Pure nian of interaction H should have the same structure ℑ as for the c-not, Eqs.AE (2.9) - (2.10): It should be a states lie on the surface of the Bloch sphere (Fig. 3). Any conceivable (unitary or non-unitary) quantum function of the pointer observable; Aˆ = a0 A0 A0 + | ih | evolution can be thought of as a transformation of the a1 A1 A1 of the apparatus. Then the states of the environment| ih | will bear an imprint of the pointer states surface of the pure states into the ellipsoid contained inside the Bloch sphere. Deformation of the Bloch sphere A0 , A1 . As noted in section II: {| i | i} surface caused by decoherence is a special case of such [H , Aˆ]=0 (4.6) general evolutions (Zurek, 1982, 1983; Berry, 1995). De- AE coherence does not affect a or b . Hence, evolution due | | | | immediately implies that Aˆ is a control, and its eigen- to decoherence alone occurs in plane z=const. Such a states will be preserved. “slice” through the Bloch sphere would show the point representing the state at a fraction r(t) of its maximum distance. The complex r(t) can be| expressed| as the sum 1. Decoherence of a single qubit of complex phase factors rotating with the frequencies given by differences ∆ωj between the energy eigenvalues An example of continuous decoherence is afforded by a of the interaction Hamiltonian, weighted with the prob- two-state apparatus interacting with an environment abilities of finding them in the initial state: A of N other spins (Zurek, 1982). The two apparatus states N are , . For the simplest, yet already interest- 2 {| ⇑i | ⇓i} r(t)= p exp( i∆ω t) . (4.14) ing example the self-Hamiltonian of the apparatus disap- j − j pears, H = 0, and the interaction has the form: Xj=1 A The index j denotes the environment part of the energy H = ( ) AE | ⇑ih⇑ | − | ⇓ih⇓ | eigenstates of the interaction Hamiltonian, Eq. (4.7): gk( )k , (22) j = . . . . The corresponding ⊗ | ↑ih↑ | − | ↓ih↓ | | i | ↑i1 ⊗| ↓i2 ⊗ ⊗| ↑iN Xk differences between the energies of the eigenstates j | ⇑i| Ni and j are; ∆ωj = j H j . There are 2 Under the influence of this Hamiltonian the initial state: AE distinct| ⇓i|j i’s, and, barringh⇑ degeneracies, |h | | i| ⇓i the same number | i N of different ∆ωj ’s. Probabilities pj are: Φ(0) = (a + b ) (αk k + βk k) (23) | i | ⇑i | ⇓i | ↑i | ↓i p = j (t = 0) 2 , (4.15) kY=1 j |h |E i| 16 which is in turn easily expressed in terms of the appro- course of decoherence, pointer states tend to be recorded priate squares of αk and βk. redundantly and can be deduced by intercepting a very The evolution of r(t), Eq. (4.14), is a consequence of small fraction of the environment (Zurek, 2000; Dalvit, the rotations of the complex vectors pk exp( i∆ωjt) with Dziarmaga and Zurek, 2001). One can then define dis- different frequencies. The resultant r(t) will− then start tance using the fraction of the environment that needs to with the amplitude 1, and quickly “crumble”to be intercepted to distinguish between two pointer states of the system (Ollivier, Poulin, and Zurek, 2002). 2N 2 2 N r(t) p 2− . (4.16) h| | i∼ j ∼ Xj=1 3. Einselection and controlled shifts

In this sense, decoherence is exponentially effective – the Discussion of decoherence can be generalized to the sit- magnitude of the off-diagonal terms decreases exponen- uation where the system, the apparatus, and the environ- tially fast with the physical size N of the environment ment have many states, and their interactions are more effectively coupled to the state of the system. complicated. Here we assume that the system is isolated, We note that the effectiveness of einselection depends and that it interacts with the apparatus in a c-shift on the initial state of the environment: When is in the E manner discussed in Section II. As a result of that in- k’th eigenstate of H , pj = δjk, the coherence in the teraction the state of the apparatus becomes entan- system will be retained.AE This special environment state gled with the state of the system: ( i αi si ) A0 is, however, unlikely in realistic circumstances. α s A . This state suffers from the| basisi | ambigu-i −→ i i| ii| ii P Pity: The entanglement implies that for any state of eitherS there − A exists a corresponding pure state of its 2. The classical domain and a quantum halo partner. Indeed, when the initial state of is chosen to be one of the eigenstates of the conjugateS basis: Geometry of flows induced by decoherence in the Bloch N 1 sphere exhibits characteristics encountered in general: 1 − r = N − 2 exp(2πikl/N) s , (4.17) (i) The classical set of the einselected pointer states | li | ki ( , in our case). Pointer states are the pure Xk=0 {| ⇑i | ⇓i} states least affected by decoherence. c-shift could equally well represent a measurement of (ii) Classical domain consisting of all the pointer states the apparatus (in the basis conjugate to Ak ) by the and their mixtures. In Fig. 3 it corresponds to the section system. Thus, it is not just the basis which{| isi} ambigu- [-1,+1] of z-axis. ous: Also the roles of the control (system) and of the (iii) The quantum domain – the rest of the volume target (apparatus) can be reversed when the conjugate of the Bloch sphere – consisting of more general density basis is selected. These ambiguities can be removed by matrices. recognizing the role of the environment. Visualizing decoherence-induced decomposition of the Figure 4 captures the essence of the idealized decoher- Hilbert space may be possible only in the simple case ence process, that allows the apparatus to be – in spite of studied here, but whenever decoherence leads to classi- the interaction with the environment – a noiseless clas- cality, emergence of generalized and often approximate sical communication channel (Schumacher, 1996; Lloyd, version of the elements (i) – (iii) is an expected feature. 1997). This is possible because the c-shifts do As a result of decoherence the part of the Hilbert space not disturb pointer states. A−E outside of the classical domain is “ruled out” by einselec- The advantage of this caricature of the decoherence tion. The severity of the prohibition on its states varies. process as a sequence of c-shifts lies in its simplicity. One may measure “non-classicality” of (pure or mixed) However, the actual process of decoherence is usually states by quantifying their distance from this classical continuous (so that it can be only approximately bro- state with the rate of entropy production and compar- ken up into discrete c-shifts). Moreover, in contrast ing it with the much lower rate in the classical domain. to the c-nots used in quantum logic circuits, the record Classical pointer states would be then enveloped by a inscribed in the environment is usually distributed over “quantum halo” (Anglin and Zurek, 1996) of nearby, rel- many degrees of freedom. Last not least, the observ- atively decoherence - resistant but still somewhat quan- able of the apparatus (or any other open system) may tum states, with a more flagrantly quantum (and more be subject to noise (and not just decoherence) or its self- fragile) Schrödinger cat states further away. Hamiltonian may rotate instantaneous pointer states into By the same token, one can define einselection - in- their superpositions. These very likely complications will duced metric in the classical domain, with the distance be investigated in specific models below. between two pointer states given by the rate of entropy Decoherence is caused by a premeasurement - like pro- production of their superposition. This is not the only cess carried out by the environment : way to define a distance: As we shall see in Section VII, E redundancy of the record of a state imprinted on the en- Ψ ε0 = ( αj sj Aj ) ε0 | SAi| i | i| i | i vironment is a very natural measure of its classicality. In Xj 17

αj sj Aj εj = Φ (27) Eq. (4.18), this transition is represented by: −→ | i| i| i | SAE i Xj P ρ = αiαj∗ si sj Ai Aj SA | ih || ih | −→ Decoherence leads to einselection when the states of the Xi,j environment εj corresponding to different pointer states 2 D | i αi si si Ai Ai = ρ (29) become orthogonal: −→ | | | ih || ih | SA Xi ε ε = δ (4.19) h i| j i ij Einselection is accompanied by the increase of entropy: Then the Schmidt decomposition of the state vector ∆H(ρ )= H(ρD ) H(ρP ) 0 (4.24) Φ into a composite subsystem and yields SA SA − SA ≥ product| SAE i states s A as partners of theSA orthogonalE envi- | j i| j i and by the disappearance of the ambiguity in what was ronment states. The decohered density matrix describing measured (Zurek, 1981, 1993a). Thus, before decoher- pair is then diagonal in product states: For simplicity ence the conditional density matrices of the system ρ Cj weSA shall often discard reference to the object that S| i are pure for any state Cj of the apparatus pointer. They are defined using the| unnormalized:i D 2 ρ = αj sj sj Aj Aj SA | | | ih || ih | Xj ρ˜ Πj = T r Πj ρ (4.25) S| A SA = T r Φ Φ . (28) E | SAE ih SAE | where in the simplest case Πj = Cj Cj projects onto a 4 | ih | does not interact with the environment (here – the system pure state of the apparatus. ). Nevertheless, preservation of the correlations is Normalized ρ Πj can be obtained by using the prob- S| theS criterion defining the pointer basis.SA Invoking it would ability of the outcome: get rid of many a confusion (see, e.g, discussions in Hal- 1 ρ Πj = pj− ρ˜ Πj ; pj = T rρ˜ Πj . (30) liwell, Perez-Mercader, and Zurek, 1994; Venugopalan, S| S| S| 1994). The density matrix of a single object in contact Conditional density matrix represents the description of with the environment will be always diagonal in an (in- the system available to the observer who knows that stantaneous) Schmidt basis. This instantaneous diago- the apparatusS is in a subspace defined by Π . nality should not be used the sole criterion for classical- j Before decoherence,A ρP is pure for any state C : Cj j ity (although see Zeh, 1973, 1990; Albrecht, 1992&1993). S| i | i Rather, ability of certain states to retain correlations in P 2 P (ρ Πj ) = ρ Πj Cj (4.27a) spite of the coupling to the environment is decisive. S| S| ∀| i When the interaction with the apparatus has the form: providing the initial premeasurement state, Eq. (4.23), was pure as well. It follows that: H = gklmAE Ak Ak εl εm + h.c. , (4.21) AE | ih || ih | k,l,mX P H(ρ Cj )=0 Cj . (4.28a) SA| i ∀| i the basis Ak is left unperturbed and any correlation {| i} For this same case given by the initially pure ρP of with the states Ak is preserved. But, by definition, {| i} Eq. (4.23), conditional density matrices obtainedSA from pointer states preserve correlations in spite of decoher- D ence, so that any observable A co-diagonal with the in- the decohered ρ will be pure if and only if they are conditioned uponSA the pointer states A ; teraction Hamiltonian will be pointer observable. For, {| ki} when the interaction is a function of A, it can be ex- D 2 P (ρ Cj ) = ρ Cj = sk sk Cj = Aj ; (31) panded in A as a power series, so it commutes with A: S| i S| i | ih | ⇐⇒ | i | i

[H (A), A]=0 (4.22) D P AE H(ρ Aj )= H(ρ Aj ) . (4.28b) S| i S| i The dependence of the interaction Hamiltonian on the observable is an obvious precondition for the monitoring of that observable by the environment. This admits 4 existence of degenerate pointer eigenspaces of A. This can be generalized to projections onto multidimensional subspaces of the apparatus. In that case, purity of the conditional density matrix will be usually lost during the trace over the states of the pointer. This is not surprising: When the ob- B. Einselection as the selective loss of information server reads oﬀ the pointer of the apparatus only in a coarse- grained manner, he will forgo part of the information about the Establishment of the measurement-like correlation be- system. Ampliﬁcation we have considered before can prevent some of such loss of resolution due to coarse graining in the ap- tween the apparatus and the environment changes the paratus. Generalizations to density matrices conditioned upon P density matrix from the premeasurement ρ to the de- projection operator valued measures (POVM’s) (Kraus, 1983) cohered ρD , Eq. (4.20). For the initiallySA pure Ψ , are also possible. SA | SAi 18

This last equation is valid even when the initial states of with the help of the conditional entropy H(ρ ). Above, the system and of the apparatus are not pure. Thus, only subscript indicates the member of the correlatedS|A pair in the pointer basis the pre-decoherence strength of the that will beA the source of the information about its part- correlation will be maintained. In all other bases: ner. A symmetric counterpart of the above equation, SA ( : )= H(ρ ) H(ρ ), can be also written. D 2 D JS S A A − A|S T r(ρ Cj ) < Trρ Cj ; Cj / Aj (4.27c) In the quantum case, deﬁnition of Eq. (4.32) is so far S| i S| i | i ∈ {| i} incomplete, as a quantum analogue of the classical con-

P D ditional information has not been yet specified. Indeed, H(ρ Cj ) < H(ρ Cj ); Cj / Aj (4.28c) S| i S| i | i ∈ {| i} Eqs. (4.30a) and (4.32) jointly imply that in the case of entanglement a quantum conditional entropy H(ρ ) In particular, in the basis Bj conjugate to the pointer would have to be negative! For, in this case; S|A states A , Eq. (2.14),{| therei} is no correlation left with {| ji} the state of the system, that is: 2 2 H(ρ )= αi lg αi < 0 (4.33) S|A | | | | D 1 1 Xi ρ Bj = N − sk sk = /N (4.29) S| i | ih | Xk would be required to allow for ( : ) = ( : ). I S A JA S A where 1 is a unit density matrix. Consequently; Various quantum redefinitions of ( : ) or H(ρ ) have been proposed to address thisI S (Lieb,A 1975; Schu-S|A D 2 macher and Nielsen, 1996; Lloyd, 1997; Cerf and Adami, (ρ Bj ) = ρ Bj /N, (4.27d) S| i S| i 1997). We shall simply regard this fact as an illustration of the strength of quantum correlations (i. e., entangle- D P ment), which allow ( : ) to violate the inequality: H(ρ Bj )= H(ρ Bj ) lg N = lg N . (4.28d) S| i S| i − − I S A Note that, initially, conditional density matrices were ( : ) min(H , H ) (4.34) I S A ≤ S A pure also in the conjugate (and any other) basis, provided that the initial state was the pure entangled projection This inequality follows directly from Eq. (4.32) and the operator ρP = Ψ Ψ , Eq. (4.23). non-negativity of classical conditional entropy (see e.g. SA | SAih SA| Cover and Thomas, 1991). Decoherence decreases ( : ) to this allowed level I S A 1. Mutual information and discord (Zurek, 1983). Moreover, now the conditional entropy can be defined in the classical pointer basis as the average of partial entropies computed from the conditional ρD Selective loss of information everywhere except in the Ai pointer states is the essence of einselection. It is re- over the probabilities of different outcomes: S| i flected in the change of the mutual information which starts from: D H(ρ )= p Ai H(ρ Ai ) (4.35) S|A | i S| i P P P P Xi ( : ) = H(ρ )+ H(ρ ) H(ρ , ) I S A S A − S A Prior to decoherence, the use of probabilities would not = 2 α 2 lg α 2 (32) − | i| | i| have been legal. Xi For the case considered here, Eq. (4.18), the condi- As a result of einselection, for initially pure cases, it de- tional entropy H(ρ ) = 0: In the pointer basis there is S|A creases to at most half its initial value: a perfect correlation between the system and the apparatus, providing that the premeasurement Schmidt basis D D D D ( : ) = H(ρ )+ H(ρ ) H(ρ , ) and the pointer basis coincide. Indeed, it is tempting I S A S A − S A to define a good apparatus or a classical correlation by = α 2 lg α 2 (33) − | i| | i| insisting on such a coincidence. Xi The difficulties with conditional entropy and mutual This level is reached when the pointer basis coincides information are highly symptomatic. The trouble with with the Schmidt basis of Ψ . The decrease in the H(ρ ) arises for states that exhibit quantum correla- S|A mutual information is due| toSA thei increase of the joint tions – entanglement of Ψ being an extreme exam- | SAi entropy H(ρ , ): ple – and, thus, do not admit an interpretation based on S A probabilities. A useful sufficient condition for the classi- ∆ ( : ) = P ( : ) D( : ) cality of correlations is then the existence of an apparatus I S A I SD A − I P S A basis that allows quantum versions of the two classically = H(ρ , ) H(ρ , ) = ∆H(ρ , )(34) S A − S A S A identical expressions for the mutual information to coin- Classically, equivalent definition of the mutual informa- cide, ( : ) = ( : ), (Zurek, 2000b; 2002a; Ol- I S A JA S A tion obtains from the asymmetric formula: livier and Zurek, 2002): Equivalently, the discord

( : )= H(ρ ) H(ρ ) (4.32) δ ( ) = ( : ) ( : ) (4.36) JA S A S − S|A IA S|A I S A − JA S A 19 must vanish: Unless δ ( ) = 0, probabilities of the ing when the observer does not know the time-dependent distinct apparatus pointerIA S|A states cannot exist. classical perturbation Hamiltonian responsible for this We end this subsection with a part summary, part an- unitary, but unknown evolution. For example, in the ticipatory remarks: Pointer states retain undiminished pre-decoherence state vector, Eq. (4.18), random phase correlations with the measured system , or with any noise will cause a transition: other system, including observers. The lossS of informa- Ψ = ( αj sj Aj ) tion caused by decoherence is given by Eq. (4.31). It | SAi | i| i was precisely such as to lift conditional information from Xj (n) (n) the paradoxical (negative) values to the classically al- αj exp(iφj ) sj Aj = Ψ . (35) −→ | i| i | SAi lowed level, Eq. (4.33). This is equal to the information Xj gained by the observer when he consults the apparatus pointer. This is no coincidence – the environment has A dephasing Hamiltonian acting either on the system or ‘measured’ (became correlated with) the apparatus in the on the apparatus can lead to such an eﬀect. In this second very same pointer basis in which observers have to access case its form could be: to take advantage of the remaining (classical) correla- H(n) = φ˙(n)(t) A A . (4.38) A D j | j ih j | tion between the pointer and the system. Only when Xj observers and the environment monitor co-diagonal observables they do not get in each others way. In contrast to interactions causing premeasurements, en- In the idealized case, preferred basis was distinguished tanglement, and decoherence, HD cannot inﬂuence the by its ability to retain perfect correlations with the sys- nature or the degree of the correlations: HD does not imprint the state of or SAanywhere else in the Uni- tem in spite of decoherence. This remark shall serve as S A a guide in other situations: It will lead to a criterion verse: For each individual realization n of the phase noise (n) (n) – predictability sieve – used to identify preferred states (each selection of φj (t) in Eq. (4.37)) the state Φ { } | SAi in less idealized circumstances. For example, when the (n) remains pure. Given only φj one could restore pre- self-Hamiltonian of the system is non-trivial, or when the dephasing state on a case -{ by -} case basis. However, in commutation relation, Eq. (4.22), does not hold exactly absence of such detailed information, one is often forced for any observable, we shall seek states that are best in to represent by the density matrix averaged over the retaining correlations with the other systems. ensemble of noiseSA realizations:

2 ρ = Ψ Ψ = αj sj sj Aj Aj SA h| SAih SA|i | | | ih || ih | C. Decoherence, entanglement, dephasing, and noise Xj i(φ(n) φ(n)) + e j − k α α s s A A (36) j k| j ih k|| j ih k| In the symbolic representation of Fig. 4, noise is the Xj,k Xn process in which environment acts as a control, inscrib- ing information about its state in the state of the system, In this phase - averaged density matrix off-diagonal terms that assumes the role of the target. However, the direc- may disappear. Nevertheless, each member of the ensem- tion of the information flow in c-nots and c-shifts de- ble may exist in a state as pure as it was before dephas- pends on the choice of initial states. Control and target ing. NMR offers examples of dephasing (which can be switch roles when, for a given Hamiltonian of interac- reversed using spin echo). Dephasing is a loss of phase tion, one prepares input of the c-not in the basis con- coherence due to noise in phases. It does not result in an jugate to the logical “pointer” states. Einselected states information transfer to the environment. correspond to the set which – when used in c-nots or Dephasing cannot be used to justify existence of pre- c-shifts – minimizes the effect of interactions directed ferred basis in individual quantum systems. Neverthe- from the environment to the system. less, the ensemble as a whole may obey the same master Einselection is caused by the premeasurement car- equation as individual systems entangling with the envi- ried out by the environment on the pointer states. De- ronment. Indeed, many of the symptoms of decoherence coherence follows from the Heisenberg’s indeterminacy: arise in this setting. Thus, in spite of the light shed Pointer observable is measured by the environment. on this issue by the discussion of simple cases (Woot- Therefore, the complementary observable must become ters and Zurek, 1979; Stern, Aharonov, and Imry, 1989) at least as indeterminate as is demanded by the Heisen- more remains to be understood, perhaps by considering berg’s principle. As the environment and the systems en- implications of envariance (see Section VI). tangle through an interaction that favors a set of pointer Noise is an even more familiar and less subtle effect states, their phases become indeterminate (see Eq. (4.29) represented by transitions that break one-to-one corre- and discussion of envariance in section VI). Decoherence spondence in Eq. (4.39). Noise in the apparatus would can be thought of as the resulting loss of phase relations. cause a random rotation of states Aj . It could be mod- | i (n) Observers can be ignorant of phases for reasons that elled by a collection of Hamiltonians similar to HD but do not lead to an imprint of the state of the system on not co-diagonal with the observable of interest. Then, af- the environment. Classical noise can cause such dephas- ter an ensemble average similar to Eq. (4.39), one-to-one 20 correspondence between and would be lost. How- In general, it is useful to invoke a more fundamental ever – as before – the evolutionS isA unitary for each n, and predictability criterion (Zurek, 1993a). One can measure the unperturbed state could be reconstructed from the the loss of predictability caused by the evolution for every information observer could have in advance. pure state Ψ by von Neumann entropy or some other Hence, in case of dephasing or noise information about measure of| predictabilityi such as the purity: their cause obtained either in advance, or afterwards, suf- 2 fices to undo their effect. Decoherence relies on entan- ςΨ(t)= TrρΨ(t) . (4.42) gling interactions (although, strictly speaking, it need not In either case, predictability is a function of time and a invlove entanglement (Eisert and Plenio, 2002). Thus, neither prior nor posterior knowledge of the state of en- functional of the initial state as ρΨ(0) = Ψ Ψ . Pointer states are obtained by maximizing predictability| ih | func- vironment is enough. Transfer of information about the decohering system to the environment is essential, and tional over Ψ . When decoherence leads to classicality, good pointer| statesi exist, and the answer is robust. plays key role in the interpretation. We note that, while nomenclature used here seems Predictability sieve sifts all of the Hilbert space, or- dering states according to their predictability. The top most sensible to this author and is widely used, it is un- fortunately not universal. For example, in the context of of the list will be the most classical. This point of view allows for unification of the simple definition of the pointer quantum computation “decoherence” is sometimes used to describe any process that can cause errors (but see states in terms of the commutation relation Eq. (4.41), with the more general criteria required to discuss classi- related discussion in Nielsen and Chuang, 2000). cality in other situations. The eigenstates of the exact pointer observable are selected by the sieve: Eq. (4.41) D. Predictability sieve and einselection guarantees that they will retain their purity in spite of the environment, and are (somewhat trivially) predictable. Evolution of a quantum system prepared in a classi- Predictability sieve can be generalized to situations cal state should emulate classical evolution that can be where the initial states are mixed (Paraonanu, 2002). Of- idealized as a “trajectory” – a predictable sequence of ten whole subspaces emerge from the predictability sieve, objectively existing states. For a purely unitary evolu- naturally leading to “decoherence-free subspaces” (see e. tion, all of the states in the Hilbert space retain their g. Lidar et al., 1999) and can be adapted to yield “noise- purity and are therefore equally predictable. However, less subsystems” (which are a non-Abelian generalization in the presence of an interaction with the environment, of pointer states; see e.g. Zanardi, 2000; Knill, Laflamme, a generic superposition representing correlated states of and Viola, 2000). However, calculations are in general the system and of the apparatus will decay into a mix- quite difficult even for the initial pure state cases. ture diagonal in pointer states, Eq. (4.23). Only when The idea of the “sieve” selecting preferred to-be- the pre-decoherence state of is a product of a single classical states is novel and only partly explored. We SA shall see it “in action” below. We have outlined two cri- apparatus pointer state Ai with the corresponding outcome state of the system| (ori a mixture of such product teria for sifting through the Hilbert space in search for states), decoherence has no effect: classicality: von Neumann entropy and purity define, after all, two distinct functionals. Entropy is arguably an P D ρ = si si Ai Ai = ρ (4.40) obvious information-theoretic measure of predictability SA | ih || ih | SA loss. Purity is much easier to compute and is often used A correlation of a pointer state with any state of an iso- as a “cheap substitute”, and has a physical significance lated system is untouched by the environment. By the of its own. It seems unlikely that pointer states selected same token, when the observer prepares in the pointer A by the predictability and purity sieves could substantially state Ai , he can count on it remaining pure. One can differ. After all, Trρ ln ρ = Trρ (1 ρ) (1 ρ)2/2+... even think| i of s as the record of the pointer state of . − { − − − | ii A so that one can expect the most predictable states to Einselected states are predictable: they preserve correla- also remain purest (Zurek, 1993a). However, the expan- tions, and hence are effectively classical. sion, Eq. (4.43), is very slowly convergent. Therefore, a In the above idealized cases predictability of some more mathematically satisfying treatment of the differ- states follows directly from the structure of the relevant ences between the states selected by these two criteria Hamiltonians (Zurek, 1981). Correlation with a subspace would be desirable, especially in cases where (as we shall associated with a projection operator PA will be immune see in the next section for the harmonic oscillator) pre- to decoherence providing that: ferred states are coherent, and hence the classical domain forms relatively broad “mesa” in the Hilbert space. [H + H , PA]=0 . (4.41) A AE The possible discrepancy between the states selected In more realistic cases it is difficult to demand exact con- by the sieves based on the predictability and on the pu- servation guaranteed by such a commutation condition. rity raises a more general question: Will all the sensi- Looking for approximate conservation may still be a good ble criteria yield identical answer? After all, one can strategy. Various densities used in hydrodynamics are imagine other reasonable criteria for classicality, such as one obvious choice (Gell-Mann and Hartle, 1990, 1994). the yet-to-be-explored “distinguishability sieve” of Schu- 21 macher (1999) which picks out states whose descendants A. Quantum Brownian motion are most distinguishable in spite of decoherence. More- over, as we shall see in Section VII (also, Zurek 2000) Quantum Brownian motion model consists of an en- one can ascribe classicality to states that are most re- vironment – a collection of harmonic oscillators (co- E dundantly recorded by the environment. The menu of ordinates qn, masses mn, frequencies ωn, and coupling various classicality criteria already contains several posi- constants cn) interacting with the system (coordinate tions, and more may be added in the future. There is no x), with a mass M and a potential V (x). WeS shall often 2 2 a priori reason to expect that all of these criteria will lead consider harmonic V (x) = MΩ0x /2, so that the whole to identical sets of preferred states. It is nevertheless rea- is linear and one can obtain an exact solution. This sonable to hope that, in the macroscopic limit in which assumptionSE will be relaxed later. classicality is indeed expected, differences between vari- The Lagrangian of the system-environment entity is: ous sieves should be negligible. The same stability in the selection of the classical domain is expected with respect L(x, qn)= L (x)+ L (x, qn ); (5.1) to the changes of, say, the time of the evolution from S SE { } the initial pure state: Reasonable changes of such details The system alone has the Lagrangian: within the time interval where einselection is expected to be effective should lead to more or less similar preferred M 2 M 2 2 2 L (x)= x˙ V (x) = (x ˙ Ω0x ) . (5.2) states, and certainly to preferred states contained within S 2 − 2 − each others “quantum halo” (Anglin and Zurek, 1996). The effect of the environment is modelled by the sum As noted above, this seems to be the case in the exam- of the Lagrangians of individual oscillators and of the ples explored to date. It remains to be seen whether all system-environment interaction terms: criteria will agree in other situations of interest.

mn 2 2 cnx 2 L = q˙n ωn qn . (5.3) SE 2 − − m ω2 Xn n n This Lagrangian takes into account renormalization of potential energy of the Brownian particle. The inter- V. EINSELECTION IN PHASE SPACE action depends (linearly) on the position x of the harmonic oscillator. Hence, we expect x to be an instantaneous pointer observable. In combination with the har- Einselection in phase space is a special, yet very impor- monic evolution this leads Gaussian pointer states – well- tant topic. It should lead to phase space points, trajecto- localized in both x and p. An important characteristic of ries, and to classical (Newtonian) dynamics. Special role the model is the spectral density of the environment: of position in classical physics can be traced to the nature c2 of interactions (Zurek, 1981; 1982; 1991) that depend on C(ω) = n δ(ω ω ) . (5.4) 2m ω − n distance, and, therefore, commute with position (see Eq. Xn n n (4.22)). Evolution of open systems includes, however, the flow in phase space induced by the self-Hamiltonian. The effect of the environment can be expressed through Consequently, the set of preferred states turns out to be the propagator acting on the reduced ρ : a compromise, localized in both position and momentum S – localized in phase space. ρ (x, x′,t)= dx0dx0′ J(x, x′,t x0, x0′ ,t0)ρ (x0, x0′ ,t0). S Z | S Einselection is responsible for the classical structure of (5.5) phase space. States selected by the predictability sieve We focus on the case when the system and the environ- become phase space “points”, and their time-ordered se- ment are initially statistically independent, so that their quences are “trajectories”. In the underdamped, clas- density matrices in a product state: sically regular systems one can recover this phase space structure along with the (almost) reversible evolution. In ρ = ρ ρ . (5.6) chaotic systems there is a price to be paid for classicality: SE S E Combination of decoherence with the exponential diver- This is a restrictive assumption. One can try to justify gence of classical trajectories (which is the defining fea- it as an idealization of a measurement that correlates ture of chaos) leads to entropy production at a rate given with the observer and destroys correlations of with – in the classical limit – by the sum of positive Lyapunov S, but that is only an approximation, as realisticS mea- exponents. Thus, the dynamical second law can emerge surementsE leave partial correlations with the environment from the interplay of classical dynamics and quantum intact. Fortunately, pre-existing post-measurement cor- decoherence, with the entropy production caused by the relations lead only to minor differences in the salient fea- information “leaking” into the environment (Zurek and tures of the subsequent evolution of the system (Romero Paz 1994 & 1995a; Zurek, 1998b; Paz and Zurek 2001). and Paz, 1997; Anglin, Paz and Zurek, 1997). 22

Evolution of the whole ρ can be represented as: where ν(s) and η(s) are dissipation and noise kernels, SE respectively, deﬁned in terms of the spectral density:

ρ (x, q, x′, q′,t)= dx dx′ dq dq′ ρ (x , q , x′ , q′ ,t ) 0 0 0 0 0 0 0 0 0 ∞ SE Z SE ν(s)= dω C(ω) coth(¯hωβ/2) cos(ωs); (5.14) Z K(x,q,t,x0, q0)K∗(x′, q′,t,x0′ , q0′ ) (37) 0 Above, we suppress the sum over the indices of the in- ∞ dividual environment oscillators. The evolution operator η(s)= dω C(ω) sin(ωs) . (5.15) Z0 K(x,q,t,x0, q0) can be expressed as a path integral: With the assumption of thermal equilibrium at k T = i B K(x,q,t,x0, q0)= DxDq exp( I[x, q]) (5.8) 1/β, and in the harmonic oscillator case V (x) = Z ¯h 2 2 MΩ0x /2, the integrand of Eq. (5.10) for the propagator is Gaussian: The integral can be computed exactly, where I[x, q] is the action functional that depends on the and should also have a Gaussian form. The result can trajectories x, q. The integration must satisfy boundary be conveniently written in terms of the diagonal and oﬀ- conditions: diagonal coordinates of the density matrix in the position representation, X = x + x′, Y = x x′: x(0) = x0; x(t)= x; q(0) = q0; q(t)= q . (5.9) − J(X,Y,t X , Y ,t ) = The expression for the propagator of the density matrix | 0 0 0 can be now written in terms of actions corresponding to b3 exp i(b1XY + b2X0Y b3XY0 b4X0Y0) 2 − − 2 .(42) the two Lagrangians, Eqs. (5.1-3): 2π exp(a11Y +2a12Y Y0 + a22Y0 )

i The time-dependent coeﬃcients bk and aij are computed J(x, x′,t x0, x0′ ,t0)= Dx Dx′ exp (I [x] I [x′]) from the noise and dissipation kernels, that reﬂect the | Z ¯h S − S properties of the environment. They obtain from the

dq dq0 dq0′ ρ (q0, q0′ ) solutions of the equation: × Z E s i 2 Dq Dq′ exp (I [x, q] I [x′, q′]) . (38) u¨(s)+Ω0u(s)+2 dsη(s s′)u(s′)=0 . (5.17) Z ¯h SE − SE Z0 − The separability of the initial conditions, Eq. (5.6), was Two such solutions that satisfy boundary conditions used to make propagator depend only on the initial con- u1(0) = u2(t) = 1 and u1(t) = u2(0) = 0 can be used ditions of the environment. Collecting all terms contain- for this purpose. They yield the coeﬃcients of the Gaus- ing integrals over in the above expression leads to the sian propagator through; inﬂuence functionalE (Feynman and Vernon, 1963): b1(2)(t) =u ˙ 2(1)(t)/2 , b3(4)(t) =u ˙ 2(1)(0)/2 , (5.18a)

F (x, x′) = dq dq0 dq0′ ρ (q0, q0′ ) Z E 1 t t i aij (t)= ds ds′ui(s)uj (s′)ν(s s′). (43) Dq Dq′ exp (I [x, q] I [x′, q′])(39). 1+ δij Z0 Z0 − Z ¯h SE − SE The master equation can be now obtained by taking It can be evaluated explicitly for specific models of the the time derivative of the Eq. (5.5), which in effect re- initial density matrix of the environment. duces to the computation of the derivative of the propa- Environment in thermal equilibrium provides a useful gator, Eq. (5.16) above: and tractable model for the initial state. The density matrix of the n’th mode of the thermal environment is: J˙ = b˙ /b + ib˙ XY + ib˙ X Y ib˙ XY { 3 3 1 2 0 − 3 0 ˙ 2 2 mnωn mnωn ib4X0Y0 a˙ 11Y a˙ 12Y Y0 a˙ 22Y0 J .(44) ρ n (q, q′) = hω¯ exp hω¯ − − − − } E 2π¯h sinh( n ) × −n2π¯h sinh( n ) kB T kB T The time derivative of ρ can be obtained by multiply- S 2 2 ¯hωn ing the operator on the right hand side by an initial den- (qn + qn′ ) cosh( ) 2qnqn′ . (40) ×h kB T − io sity matrix and integrating over the initial coordinates X0, Y0. Given the form of Eq. (5.19), one may ex- The influence functional can be written as (Grabert, pect that this procedure will yield an integro-differential Schramm, and Ingold, 1988): (non-local in time) evolution operator for ρ . However, S t s time dependence of the evolution operator disappears as i ln F (x, x′)= ds (x x′)(s) du a result of two identities satisfied by the propagator: Z0 − Z0 b1 i η(s s′)(x + x′)(s′) iν(s s′)(x x′)(s′) (41), Y0J = Y + ∂X J , (5.20a) − − − − b3 b3 23

t b1 i 2 X0J = X ∂Y γ(t)= ds sin(Ωs)η(s) (5.23b) − b2 − b2 MΩ Z0 2a11 a12b1 a12 i + Y + ∂X J . (45) − b b b b b t 2 2 3 2 3 1 D(t)= ds cos(Ωs)ν(s) (5.24b) After the appropriate substitutions, the resulting equa- ¯h Z0 tion with renormalized Hamiltonian has a form: i 1 t ρ˙ (x, x′,t) = x [Hren(t),ρ ] x′ f(t)= ds sin(Ωs)η(s) (5.25b) S −¯hh | S | i −MΩ Z0 2 γ(t)(x x′)(∂x + ∂x′ ) D(t)(x x′) ρ (x, x′,t) − − − − S These coefficient can be made even more explicit when if(t)(x x′)(∂x + ∂x′ )ρ (x, x′,t). (46) a convenient specific model for the spectral density: − − S The calculations leading to this master equation are non- ω Γ2 C(ω)=2Mγ (5.27) trivial. They involve use of relations between the coeffi- 0 π Γ2 + ω2 cients bk and aij . The final result leads to explicit formulae for its coefficients: is adopted. Above, γ0 characterizes the strength of the interaction, and Γ is the high-frequency cutoff. Then: ˙ ˙ Ωren(t)/2= b1b2/b2 b1 , (5.22a) 3 − 2 2γ0Γ Ω Γt Ω˜ = 1 cosΩt sinΩt e− ; (5.22c) −Γ2 +Ω2 − − Γ γ(t)= b b˙ /2b , (5.23a) − 1 − 2 2 2 γ0Γ Γ Γt γ(t)= 1 cosΩt sinΩt e− . (5.23c) D(t) =a ˙ 11 4a11b1 Γ2 +Ω2 − − Ω − +˙a b /b b˙ (2a + a b /b )/b , (47) 12 1 3 − 2 11 12 1 3 2 Note that both of these coefficients are initially zero. They grow to their asymptotic values on a timescale set by the inverse of the cutoff frequency Γ. 2f(t) =a ˙ 12/b3 b˙2a12/(b2b3) 4a11 . (5.25a) − − The two diffusion coefficients can be also studied, but The fact that the exact master equation (5.21) is local it is more convenient to evaluate them numerically. In in time for an arbitrary spectrum of the environment is Fig. 5 we show their behavior. The normal diffusion co- remarkable. It was demonstrated by Hu, Paz, and Zhang efficient quickly settles to its long-time asymptotic value: (1992) following discussions carried out under more re- 1 2 2 2 D = Mγ0Ω¯h− coth(¯hΩβ/2)Γ /(Γ +Ω ) . (5.28) strictive assumptions by Caldeira and Leggett (1983), ∞ Haake and Reibold (1985), Grabert, Schramm, and In- The anomalous diffusion coefficient f(t) also approaches gold (1988), and Unruh and Zurek (1989). It depends on asymptotic value. For high temperature it is suppressed the linearity of the problem, that allows one to anticipate by a cutoff Γ with respect to D , but the approach to f (Gaussian) form of the propagator. is more gradual, algebraic rather∞ than exponential. Envi-∞ The above derivation of the exact master equation used ronments with different spectral content exhibit different the method of Paz (1994) (see also Paz and Zurek, 2001). behavior (Hu, Paz, and Zhang, 1992; Paz, Habib, and Explicit formulae for the time-dependent coefficients can Zurek, 1993; Paz, 1994; Anglin, Paz and Zurek, 1997). be obtained when one focuses on the perturbative master equation. It can be derived ab initio (see Paz and Zurek, 2001) but can be also obtained from the above results by B. Decoherence in quantum Brownian motion finding a perturbative solution to Eq. (5.17), and then substituting it in Eqs. (5.22a - 5.25a). The resulting The coefficients of the master equation we have just master equation in the operator form is: derived can be computed under a variety of different assumptions. The two obvious characteristics of the en- i iγ(t) ρ˙ = [H + MΩ(˜ t)2x2/2,ρ ] [x, p,ρ ] vironment one can change are its temperature T and S −¯h S S − ¯h { S} its spectral density C(ω). One general conclusion: In f(t) D(t)[x, [x, ρ ]] [x, [p,ρ ]] . (48) case of high temperatures, D(t) tends to a temperature- − S − ¯h S dependent constant, and dominates over f(t). Indeed, in this case all of the coefficients settle to asymptotic values Coefficients such as the frequency renormalization Ω,˜ the after an initial transient. Thus: relaxation coefficient γ(t), and the normal and anomalous diffusion coefficients D(t) and f(t) are given by: i ρ˙ = [Hren,ρ ] γ(x x′)(∂x ∂x′ )ρ S −¯h S − − − S 2 t 2Mγk T Ω˜ 2(t)= ds cos(Ωs)η(s) (5.22b) B 2 2 (x x′) ρ . (49) −M Z0 − ¯h − S 24

This master equation for ρ(x, x′) obtains in the unre- High temperature master equation (5.29) is a good ap- alistic but convenient limit known as the high tempera- proximation in a wider range of circumstances than the ture approximation valid when kB T is much higher than one for which it was derived (Feynman and Vernon, 1963; all the other relevant energy scales, including the energy Dekker, 1977; Caldeira and Leggett, 1983). Moreover, content of the initial state and the frequency cutoﬀ in our key qualitative conclusion – rapid decoherence in the C(ω) (see, e.g, Caldeira and Leggett, 1983). However, macroscopic limit – does not crucially depend on the ap- when these restrictive conditions hold, Eq. (5.29) can proximations leading to Eq. (5.29). We shall therefore be written for an arbitrary V (x). To see why, we give a use it in our further studies. derivation patterned on Hu, Paz, and Zhang (1993). We start with the propagator, Eq. (5.5), ρ (x, x′,t)= S J(x, x′,t x0, x0′ ,t0)ρ (x0, x0′ ,t0), which we shall treat as 1. Decoherence timescale if it were| an equationS for state vector of the two dimensional system with coordinates x, x′. The propagator is In the macroscopic limit (that is, whenh ¯ is small then given by the high-temperature version of Eq. (5.10); compared to other quantities with dimensions of action, 2 such as 2MkBT (x x′) in the last term) the high- i temperature masterh equation− i is dominated by: J(x, x′,t x0, x0′ ,t0)= DxDx′ exp IR(x) IR(x′) p | Z ¯hn − o t k T Mγ ds[xx˙ x′x˙ ′+xx˙ ′ x′x˙]+ 2 B [x x′]2 (x x′) 2 e− 0 − − h¯2 − . (50) ∂tρ (x, x′,t)= γ − ρ (x, x′,t) . (5.34) S − λ S R T The term in the exponent can be interpreted as an eﬀec- Above; tive Lagrangian of a two-dimensional system:

2 2 ¯h Leff (x, x′)= Mx˙ /2 VR(x) Mx˙ ′ /2+ VR(x′) λT = (5.35) − − √2MkBT 2MγkBT 2 +γ(x x′)(x ˙ +x ˙ ′)+ i (x x′) . (51) − ¯h2 − is thermal de Broglie wavelength. Thus, the density matrix looses off-diagonal terms in position representation: One can readily obtain the corresponding Hamiltonian; ′ γt( x−x )2 Heff =x∂L ˙ eff /∂x˙ +x ˙ ′∂Leff /∂x˙ ′ Leff . (5.32) ρ (x, x′,t) = ρ (x, x′, 0)e− λT . (5.36) − S S Conjugate momenta p = p = Mx˙ + γ(x x′) and p′ = x − while the diagonal (x = x′) remains untouched. p ′ = Mx˙ ′ + γ(x x′) are used to express the kinetic x − − Quantum coherence decays exponentially at the rate term of Heff . After evaluatingx ˙ andx ˙ ′ in terms of p given by the relaxation rate times the square of the dis- and p′ in the expression for Heff one obtains: tance measured in units of thermal de Broglie wavelength 2 2 (Zurek, 1984a). Position is the “instantaneous pointer H = p γ(x x′) /2M p′ γ(x x′) /2M eff − − − − − observable”: If Eq. (5.36) was always valid, eigenstates 2 2 +V (x) V (x′) i2MγkBT (x′ x) /¯h .(52) of position would attain the classical status. − − − The importance of position can be traced to the nature This expression yields the operator that generates the of the interaction Hamiltonian between the system and evolution of the density matrix, Eq. (5.29). the environment. According to Eq. (5.3): The coefficients of Eq. (5.21) approach their high- temperature values quickly (see Fig. 5). Already for H = x cnqn . (5.37) SE T well below what the rigorous derivation would demand Xn high-temperature limit appears to be an excellent approximation. The discrepancy – manifested by symptoms This form of H is motivated by physics (Zurek, 1982; SE such as some of the diagonal terms of ρ (x′, x) assum- 1991): Interactions depend on the distance. However, ing negative values when the evolutionsS starts from an had we endeavored to find a situation where a different initial state that is so sharply localized in position to form of the interaction Hamiltonian – say, a momentum- have kinetic energy in excess of the values allowed by the dependent interaction – was justified, the form and con- high-temperature approximation – is limited to the ini- sequently predictions of the master equation would have tial instant of order 1/Γ, and is known to be essentially been analogous to Eq. (5.36), but with a substitution of unphysical for other reasons (Unruh and Zurek, 1989; the relevant observable “monitored” by the environment Ambegoakar, 1991; Romero and Paz, 1997; Anglin, Paz, for x. Such situations may be experimentally accessible and Zurek, 1997). This short - time anomaly is closely (Poyatos, Cirac, and Zoller, 1996) providing a test of one tied to the fact that Eq. (5.33) (and, indeed, many of of the key ideas of einselection: the relation between the the exact or approximate master equations derived to form of interaction and the preferred basis. date) does not have the Lindblad form (Lindblad, 1976; The effect of the evolution, Eqs. (5.34) - (5.36), on see also Gorini, Kossakowski and Sudarshan, 1976, Alicki the density matrix in the position representation is easy and Lendi, 1987) of a dynamical semigroup. to envisage. Consider a superposition of two minimum 25 uncertainty Gaussians. Off-diagonal peaks represent co- Wigner function in this case is given by: herence. They decay on a decoherence timescale τD, or with a decoherence rate (Zurek, 1984a, 1991); W (x, p)= G(x + x0,p) + G(x x0,p) 1 2 2 2 2 2 − +(π¯h)− exp( p ξ /¯h x /ξ ) cos(∆xp/¯h) , (54) 1 x x′ 2 − − τD− = γ( − ) . (5.38) λT where;

Thermal de Broglie wavelength, λT , is microscopic for (x x )2/ξ2 (p p )2ξ2/h¯2 e− ∓ 0 − − 0 massive bodies and for the environments at reasonable G(x x0,p p0)= . (55) temperatures. For a mass 1g at room temperature and ± − π¯h for the separation x′ x =1cm, Eq. (5.38) predicts de- We have assumed the Gaussians are not moving (p0 = 0). − 40 coherence approximately 10 times faster than relax- The oscillatory term in Eq. (5.41) is the signature of ation! Even cosmic microwave background suffices to superposition. The frequency of the oscillations is pro- cause rapid loss of quantum coherence in objects as small portional to the distance between the peaks. When the as dust grains (Joos and Zeh, 1985). These estimates for separation is only in position x, this frequency is: the rates of decoherence and relaxation should be taken with a grain of salt: Often the assumptions that have led f = ∆x/¯h =2x0/¯h . (5.43) to the simple high temperature master equation (5.29) are not valid (Gallis & Fleming, 1990; Gallis, 1992; An- Ridges and valleys of the interference pattern are paral- glin, Paz & Zurek, 1997). For example, the decoherence lel to the separation between the two peaks. This, and rate cannot be faster than inverse of the spectral cutoff in the fact thath ¯ appears in the interference term in W Eq. (5.27), or than the rate with which the superposition is important for phase space derivation of the decoher- is created. Moreover, for large separations quadratic de- ence time. We focus on the dominant effect and direct pendence of decoherence rate may saturate (Gallis and our attention on the last term of Eq. (5.40). Its effect Fleming, 1990; Anglin, Paz, and Zurek, 1997) as seen on the rapidly oscillating interference term will be very in the “simulated decoherence” experiments of Cheng different from its effect on the two Gaussians: The inter- and Raymer (1999). Nevertheless, in the macroscopic ference term is dominated by the cosine: domain decoherence of widely delocalized “Schrödinger ∆x cat” states will occur very much faster than relaxation, Wint cos( p) . (5.44) which proceeds at the rate given by γ. ∼ ¯h This is an eigenfunction of the diffusion operator. Deco- herence timescale emerges (Zurek, 1991) from the corre- 2. Phase space view of decoherence sponding eigenvalue;

A useful alternative way of illustrating decoherence is 2 2 W˙ int D∆x /¯h Wint . (5.45) aﬀorded by the Wigner function representation: ≈ −{ }×

+ We have recovered the formula for τD, Eq. (5.38), from a 1 ∞ y y W (x, p)= dy e(ipy/h¯)ρ(x + , x ) . (53) diﬀerent looking argument: Eq. (5.40) has no explicit de- 2π¯h Z 2 − 2 −∞ pendence onh ¯ for linear potentials (in the nonlinear case ¯h enters through the Moyal bracket). Yet, the decoher- Evolution equation followed by the Wigner function ob- ence timescale contains ¯h explicitly –h ¯ enters through tains through the Wigner transform of the correspond- Eq. (5.43), that is through its role in determining the ing master equation. In the high temperature limit, Eq. frequency of the interference pattern W . (5.29) (valid for general potentials) this yields: int The evolution of the pure initial state of the type con-

∂tW = Hren, W MB +2γ∂p(pW )+ D∂ppW . (5.40) sidered here is shown in Fig. 6. There we illustrate evo- { } lution of the Wigner function for two initial pure states: The first term – Moyal bracket – is the Wigner transform Superposition of two positions and superposition of two of the von Neumann equation (see Section III). In the momenta. There is a noticeable difference in the rate linear case it reduces to the Poisson bracket. The second at which the interference term disappears between these term is responsible for relaxation. The last diffusive term two cases. This was anticipated. The interaction in Eq. is responsible for decoherence. (5.3) is a function of x. Therefore, x is monitored by the Diffusion in momentum occurs at the rate set by environment directly, and the superposition of positions D = 2MγkBT . Its origin can be traced to the continu- decoheres almost instantly. By contrast, the superposi- ous “measurement” of the position of the system by the tion of momenta is initially insensitive to the monitoring environment: In accord with the Heisenberg indetermi- by the environment – the corresponding initial state is al- nacy, measurement of position results in the increase of ready well localized in the observable singled out by the the uncertainty in momentum (see Section IV). interaction. However, a superposition of momenta leads Decoherence in phase space can be explained on the to a superposition of positions, and hence to decoherence, example of a superposition of two Gaussian wavepackets. albeit on a dynamical (rather than τD) timescale. 26

An intriguing example of a long-lived superposition For predictability sieve this term is usually unimportant, of two seemingly distant Gaussians was pointed out by as for a vast majority of the initially pure states its effect Braun, Braun and Haake (2000) in the context of super- will be negligible when compared to the first, decoherence radiance. As they note, the relevant decohering inter- - related term. Thus, in the case of pure initial states: action cannot distinguish between some such superpositions, leading to a ‘Schrödinger cat’ pointer subspace. d 4ηkBT Trρ2 = ( x2 x 2), (5.50) dt − ¯h2 h i − h i

C. Predictability sieve in phase space Therefore, the instantaneous loss of purity is minimized for perfectly localized states (Zurek, 1993a). The second Decoherence rapidly destroys non-local superpositions. term of Eq. (5.49) allows for the equilibrium. (Neverthe- Obviously, states that survive must be localized. How- less, early on, and for very localized states, its presence ever, they cannot be localized to a point in x, as this causes an (unphysical) increase of purity to above unity. would imply – by Heisenberg’s indeterminacy – an infi- This is a well-known artifact of the high-temperature ap- nite range in momenta, and, hence, of velocities. As a proximation (see discussion following Eq. (5.33)). result, a wave function localized too well at one instant To find most predictable states relevant for dynamics would become very non-local a moment later. consider entropy increase over a period of the oscillator. Einselected pointer states minimize the damage done For a harmonic oscillator with mass M and frequency Ω, by decoherence over the “timescale of interest” (usually one can compute purity loss averaged over τ =2π/Ω: associated with predictability or with dynamics). They can be found through the application of the predictability ∆ς 2π/Ω = 2D ∆x2 + ∆p2/(MΩ)2 . (5.51) |0 − sieve outlined at the end of Section IV. To implement it, we compute entropy increase or purity loss for all initially Above, ∆x and ∆p are dispersions of the state at the pure states in the Hilbert space of the system under the initial time. By the Heisenberg indeterminacy, ∆x∆p cumulative evolution caused by the self-Hamiltonian and ¯h/2. The loss of purity will be smallest when: ≥ by the interaction with the environment. It would be a tall order to carry out the requisite calculations for ∆x2 =¯h/2MΩ, ∆p2 =¯hMΩ/2 . (5.52) an arbitrary quantum system interacting with a general environment. We focus on the exactly solvable case. Coherent quantum states are selected by the predictabil- In the high temperature limit the master equations ity sieve in an underdamped harmonic oscillator (Zurek, (5.26) and (5.29) can be expressed in the operator form: 1993a; Zurek, Habib and Paz, 1993; Gallis, 1996; Tegmark and Shapiro, 1994; Wiseman and Vaccaro, 1 γ ηk T B 1998; Paraoanu, 2002). Rotation induced by the self- ρ˙ = [Hren,ρ]+ [ p, x ,ρ] 2 [x, [x, ρ]] i¯h i¯h { } − ¯h Hamiltonian turns preference for states localized in posi- iγ [x,ρp] [p,ρx] (56) tion into preference for localization in phase space. This − ¯h − is illustrated in Fig. 7. Above, η =2Mγ is the viscosity. Only the last two terms We conclude that for an underdamped harmonic os- can change entropy. Terms of the form: cillator coherent Gaussians are the best quantum theory has to offer as an approximation to a classical point. ρ˙ = [O,ρˆ ] (5.47) Similar localization in phase space should obtain in the reversible classical limit in which the familiar symptoms where Oˆ is Hermitean leave the purity ς = Trρ2 and the of the “openness” of the system – such as the finite relax- von Neumann entropy H = Trρ lg ρ unaffected. This ation rate γ = η/2M – become vanishingly small. This follows from the cyclic property− of the trace: limit can be attained for large mass M , while the viscosity η remains fixed and sufficiently−→ large ∞ to as- N sure localization (Zurek, 1991; 1993a). This is, of course, d N k 1 N k Trρ = (Trρ − [Oρˆ ]ρ − )=0 (5.48) not the only possible situation. Haake and Walls (1987) dt kX=1 discussed the overdamped case, where pointer states are still localized, but relatively more narrow in position. On 2 Constancy of Trρ is obvious, while for Trρ lg ρ it follows the other hand, ”adiabatic” environment enforces einse- when the logarithm is expanded in powers of ρ. lection in energy eigenstates (Paz and Zurek, 1999). Equation (5.46) leads to the loss of purity at the rate (Zurek, 1993a):

d 4ηk T D. Classical limit in phase space Trρ2 = B T r ρ2x2 (ρx)2 +2γTrρ2 (5.49) dt − ¯h2 − There are three strategies that allow one to simulta- The second term increases purity – decreases entropy – as neously recover both the classical phase space structure the system is damped from an initial highly mixed state. and the classical equations of motion. 27

1. Mathematical approach (¯h → 0) superposition should be – after a few decoherence times – well approximated by the probability distribution over This “mathematical” classical limit could not be imple- such Gaussian “points”. mented without decoherence, since the oscillatory terms associated with interference do not have an analytic ¯h 0 limit (see, e.g, Peres, 1993). However, in the 2. Physical approach: The macroscopic limit presence→ of the environment, the relevant terms in the 2 master equations increase as (¯h− ), and make the non- The possibility of theh ¯ 0 classical limit in pres- O analytic manifestations of interference disappear. Thus, ence of decoherence is of interest.→ But ¯h = 1.05459 27 × phase space distributions can be always represented by 10− erg s. Therefore, a physically more reasonable ap- localized coherent state “points”, or by the distributions proach increases the size of the object, and, hence, its over the basis consisting of such points. susceptibility to decoherence. This strategy can be im- This strategy is easiest to implement starting from plemented starting with Eq. (5.40). Reversible dynamics the phase-space formulation. It follows from Eq. (5.45) obtains as γ 0 while D =2MγkBT = ηkB T increases. that the interference term in Eq. (5.41) will decay (Paz, The decrease→ of γ and the simultaneous increase of Habib, and Zurek, 1993) over the time interval ∆t as: ηkBT can be anticipated with the increase of the size and mass. Assume that density of the object is indepen- 2 D∆x ∆x dent of its size R, and that the environment quanta scat- Wint exp ∆t cos p . (5.53) ∼ − ¯h2 ¯h ter from its surface (as would photons or air molecules). Then M R3 and η R2. Hence: As long as ∆t is large compared to decoherence timescale ∼ ∼ 2 2 τD ¯h /D∆x , the oscillatory contributions to the η (R2) , (5.56) Wigner≃ function W (x, p) shall disappear withh ¯ 0. ∼ O −→ ∞ Simultaneously, Gaussians representing likely locations→ of the system become narrower, approaching δ-functions γ = η/2M (1/R) 0 , (5.57) in phase space. For instance, in Eq. (5.42); ∼ O −→ as R : Localization in phase space and reversibility lim G(x x0,p p0) = δ(x x0,p p0) , (5.54) → ∞ h¯ 0 − − − − can be simultaneously achieved in a macroscopic limit. → Existence of macroscopic classical limit in simple cases providing half-widths of the coherent states in x and p has been pointed out some time ago (Zurek 1984a, 1991; decrease to zero as ¯h 0. This would be assured when, → Gell-Mann and Hartle, 1993). We shall analyze it in the for instance, in Eqs. (5.41)-(5.42): next section in a more complicated chaotic setting, where reversibility can no longer be taken for granted. In the ξ2 ¯h . (5.55) ∼ harmonic oscillator case approximate reversibility is effectively guaranteed, as the action associated with the Thus, individual coherent-state Gaussians approach 1-σ contour of the Gaussian state increases with time at phase space points. This behavior indicates that in a the rate (Zurek, Habib, and Paz, 1993): macroscopic open system nothing but probability distributions over localized phase space points can survive in kB T theh ¯ 0 limit for any time of dynamical or predictive I˙ = γ (5.58) ¯hΩ signiﬁcance.→ (Coherence between immediately adjacent points separated only by ξ, Eq. (5.55), can last longer. Action I is a measure of the lack of information about ∼ This is no threat to the classical limit. Small scale coher- phase space location. Hence, its rate of increase is a ence is a part of a “quantum halo” of the classical pointer measure of the rate of predictability loss. Trajectory is states (Anglin and Zurek, 1996).) a limit of the “tube” swept in phase space by the mov- The mathematical classical limit implemented by let- ing volume representing instantaneous uncertainty of the tingh ¯ 0 becomes possible in presence of decoherence. observer about the state of the system. Evolution is ap- → It is tempting to take this strategy to its logical con- proximately deterministic when the area of this contour clusion, and represent every probability density in phase is nearly constant. In accord with Eqs. (5.56)-(5.57) I˙ space in the point(er) basis of narrowing coherent states. tends to zero in the reversible macroscopic limit: Such a program is beyond the scope of this review, but the reader should be by now convinced that it is possible. I˙ (1/R) (5.59) Indeed, Perelomov (1986) shows that a general quantum ∼ O state can be represented in a sparse basis of coherent The existence of an approximately reversible trajectory- states that occupy sites of a regular lattice, providing like thin tubes provides an assurance that, having local- that the volume per coherent state “point” is no more ized the system within a regular phase space volume at than (2π¯h)d in the d dimensional conﬁguration space. t = 0, we can expect that it can be found later inside the In presence of decoherence− arising from the coordinate- Liouville - transported contour of nearly the same mea- dependent interaction, evolution of a general quantum sure. Similar conclusions follow for integrable systems. 28

1 3. Ignorance inspires confidence in classicality th¯ = Λ− ln χ∆p/¯h. Here Λ is the Lyapunov exponent, χ = Vx/Vxxx typically characterizes the dominant scale Dynamical reversibility can be achieved with einselec- of nonlinearitiesp in the potential V (x), and ∆p gives the tion in the macroscopic limit. Moreover, I/I˙ or other coherence scale in the initial wavepacket. The above es- measures of predictability loss decrease with the increase timate, Eq. (3.5) depends on the initial conditions. It 1 of I. This is especially dramatic when quantified in terms is smaller than, but typically close to, tr = Λ− ln I/¯h, of the von Neumann entropy, that, for Gaussian states, Eq. (3.6), where I is the characteristic action of the sys- increases at the rate (Zurek, Habib and Paz, 1993): tem. By contrast, phase space patches of regular systems undergo stretching with a power of time. Consequently, I +1 H˙ = I˙ lg (5.60a) loss of correspondence occurs only over a much longer I 1 t (I/¯h)α, that depends polynomially onh ¯. − r ∼ The resulting H˙ is infinite for pure coherent states (I = 1), but quickly decreases with increasing I. Similarly, the 1. Restoration of correspondence rate of purity loss for Gaussians is: Exponential instability spreads the wavepacket to a ˙ 2 ς˙ = I/I . (5.60b) “paradoxical” extent at the rate given by the positive (i) Again, it tapers off for more mixed states. Lyapunov exponents Λ+ . Einselection attempts to en- This behavior is reassuring. It leads us to conclude force localization in phase space by tapering off interference terms at the rate given by the inverse of the de- that irreversibility quantified through, say, von Neu- 1 2 mann entropy production, Eq. (5.60a), will approach coherence timescale τD = γ− (λT /∆x) . The two processes reach status quo when the coherence length ℓ of H˙ 2I/I˙ , vanishing in the limit of large I. When in c the wavepacket makes their rates comparable, that is: the≈ spirit of the macroscopic limit we do not insist on the maximal resolution allowed by the quantum indeter- τDΛ+ 1 (5.61) minacy, the subsequent predictability losses measured by ≃ the increase of entropy or through the loss of purity will This yields an equation for the steady-state coherence diminish. Illusions of reversibility, determinism, and ex- length and for the corresponding momentum dispersion: act classical predictability become easier to maintain in ℓ λ Λ /2γ ; (5.62) presence of ignorance about the initial state! c T + ≃ p To think about phase space points one may not even need to invoke a specific quantum state. Rather, a point σc =¯h/ℓc = 2D/Λ+ . (5.63) can be regarded as a limit of an abstract recursive proce- p Above, we have quoted results (Zurek and Paz, 1994) dure, in which phase space coordinates of the system are that follow from a more rigorous derivation of the co- determined better and better in a succession of increas- herence length ℓ than the “rough and ready” approach ingly accurate measurements. One may be tempted to c that led to Eq. (5.61). They embody the same physical extrapolate this limiting process ad infintessimum which argument, but seek asymptotic behavior of the Wigner would lead beyond Heisenberg’s indeterminacy principle function that evolves according to the equation: and to a false conclusion that idealized points and trajectories “exist objectively”, and that the insider view of ¯h2n( )n W˙ = H, W + − ∂2n+1V ∂2n+1W Section II can be always justified. While in our quantum { } 22n(2n + 1)! x p nX1 Universe this conclusion is wrong, and the extrapolation ≥ 2 described above illegal, the presence, within the Hilbert + D∂p W . (57) space, of localized wavepackets near the minimum uncertainty end of such imagined sequences of measurements The classical Liouville evolution generated by Poisson is reassuring. Ultimately, the ability to represent motion bracket ceases to be a good approximation of the de- in terms of points and their time - ordered sequences cohering quantum evolution when the leading quantum (trajectories) is the essence of classical mechanics. correction becomes comparable to the classical force: 2 2 ¯h ¯h Vx Wp VxxxWppp 2 2 (5.65) E. Decoherence, chaos, and the Second Law 24 ≈ 24 χ σc

The term ∂xV ∂pW represents the classical force in Pois- Breakdown of correspondence in this chaotic setting son bracket. Quantum corrections are small when; was described in section III. It is anticipated to oc- σ χ ¯h (5.66) cur in all non-linear systems, as the stretching of the c ≫ wavepacket by the dynamics is a generic feature, absent Equivalently, Moyal bracket generates approximately Li- only in a harmonic oscillator. However, exponential in- ouville ﬂow when the coherence length satisﬁes: stability of chaotic dynamics implies rapid loss of quantum - classical correspondence after the Ehrenfest time, ℓ χ (5.67) c ≪ 29

(i) This last inequality has an obvious interpretation: It is Eq. (5.61), may not be met simultaneously for all Λ+ ). a demand for the localization to within a region ℓc small Then the entropy production rate can accelerate (before compared to the scale χ of the nonlinearities of the po- subsiding as a consequence of approaching equilibrium). tential. When this condition holds, classical force will The timescales on which this estimates of entropy pro- dominate over quantum corrections. duction apply are still subject to investigation (Zurek and Restoration of correspondence is illustrated in Fig. Paz, 1995a; Zurek, 1998b; Monteoliva and Paz 2000 & 8 where Wigner functions are compared with classical 2001) and even controversy (Casati and Chirikov, 1995b; probability distributions in a chaotic system. The dif- Zurek and Paz, 1995b). The instant when Eq. (5.69) ference between the classical and quantum expectation becomes a good approximation corresponds to the mo- values in same chaotic system is shown in Fig. 9. Even ment when the exponentially unstable evolution forces relatively weak decoherence suppresses the discrepancy, the Wigner function to develop small phase space struc- helping reestablish the correspondence: D =0.025 trans- tures on the scale of the effective decoherence - imposed lates through Eq. (5.62) into coherence over ℓc 0.3, not “coarse graining”, Eq. (5.63). This will be a good ap- much smaller than the nonlinearity scale χ ≃1 for the ≃ proximation until the time tEQ at which equilibrium sets investigated Hamiltonian of Fig. 8. in. Both th¯ have a logarithmic dependence on the corresponding (initial and equilibrium) phase space volumes I0 and IEQ, so the validity of Eq. (5.69) will be limited 2. Entropy production 1 to t t Λ− ln I /I . EQ − h¯ ≃ EQ 0 There is a simple and conceptually appealing way to Irreversibility is the price for the restoration of extend the interval over which entropy is produced at the quantum-classical correspondence in chaotic dynamics. rate given by Eq. (5.69): Imagine an observer monitoring It can be quantified through the entropy production rate. a decohering chaotic system, finding out its state at time The simplest argument recognizes that decoherence re- intervals small compared to Λ 1, but large compared to stricts spatial coherence to ℓ . Consequently, as the ex- − c the decoherence timescale. One can show (Zurek, 1998b) ponential instability stretches the size L(i) of the distri- that the average increase of the size of the algorithmi- bution in directions corresponding to the positive Lya- (i) (i) (i) cally compressed records of measurement of a decohering punov exponents Λ+ , L exp(Λ+ t) the squeezing chaotic system (that is, the algorithmic randomness of mandated by the Liouville theorem∼ in the complemen- (i) (i) the acquired data, see e.g. Cover and Thomas, 1991) is tary directions corresponding to Λ will halt at σc , Eq. given – after conversion into bits from “nats” – by Eq. − (5.63). In this limit, the number of pure states needed to (5.69). This conclusion holds providing that the effect of represent resulting mixture increases exponentially the “collapses of the wavepacket” caused by the repeated measurements is negligible – i.e., the observer is “skill- N (i) L(i)/ℓ(i) (5.68) ≃ c ful”. A possible strategy a skillful observer may adopt is in each dimension. The least number of pure states over- that of indirect measurements, of monitoring a fraction of (i) lapped by W will be then = ΠiN . This implies: the environment responsible for decoherence to find out N the state of the system. As we shall see in more detail in H˙ ∂ ln Λ(i) . (5.69) the following sections of the paper, this is a very natural ≃ t N ≃ + Xi strategy, often employed by the observers. A classical analogue of Eq. (5.69) has been obtained by This estimate for entropy production rate becomes ac- Kolmogorov (1960) and Sinai (1960) starting from very curate as the width of the Wigner function reaches satu- (i) different, mathematical arguments that in effect relied on ration at σc . When a patch in phase space correspond- an arbitrary but fixed coarse graining imposed on phase ing to the initial W is regular and smooth on scales large (i) space (see Wehrl, 1978). Decoherence leads to a sim- compared to σc , evolution will start nearly reversibly ilar looking quantum result in a very different fashion: (Zurek and Paz, 1994). However, as squeezing brings the Coarse graining is imposed by the coupling to the envi- (i) extent of the effective support of W close to σc , diffusion ronment, but only in the sense implied by the einselec- bounds from below the size of the smallest features of W . tion. Its “graininess” (resolution) is set by the accuracy Stretching in the unstable directions continues unabated. of the monitoring environment. This is especially obvi- As a consequence, the volume of the support of W will ous when the indirect monitoring strategy mentioned im- grow exponentially, resulting in an entropy production mediately above is adopted by the observers. Preferred rate set by Eq. (5.69), the sum of the classical Lyapunov states will be partly localized in x and p, but (in contrast exponents. Yet, it has an obviously quantum origin in to the harmonic oscillator case with its coherent states) decoherence. This quantum origin may be apparent ini- details of this environment - imposed coarse graining will tially, as the rate of Eq. (5.69) will be approached from likely depend on phase space location, precise nature of above when the initial state is a non-local. On the other the coupling to the environment, etc. Yet, in the appro- hand, in a multidimensional system different Lyapunov priate limit, Eqs. (5.66)-(5.67), the asymptotic entropy exponents may begin to contribute to entropy produc- production rate defined with the help of the algorithmic tion at different instants (as the saturation condition, contribution discussed above (i.e., in a manner of the 30 physical entropy, Zurek (1989)) does not depend on the patch in the phase space will asymptote to I˙ const: strength or nature of the coupling, but is instead given ≃ by the sum of the positive Lyapunov exponents. H˙ = I/I˙ 1/t . (5.71) ∼ Von Neumann entropy production consistent with the above discussion has been now seen in numerical stud- Thus, initial conditions allow one to predict future of a ies of decohering chaotic systems (Shiokawa and Hu, regular system for time intervals that are exponentially 1995; Furuya, Nemes, and Pellegrino, 1998; Miller and longer than in the chaotic case. The rate of entropy pro- Sarkar, 1999; Schack, 1998; Monteoliva and Paz, 2000 & duction of an open quantum system is therefore a very 2001). Extensions to situations where relaxation matters, good indicator of the nature of its dynamics, as was con- as well as in the opposite direction – to where decoher- jectured some time ago (Zurek and Paz, 1995a), and as ence is relatively gentle – have been also discussed (Brun, seems born out in the numerical simulations (Shiokawa Percival, and Schack, 1995; Miller, Sarkar, and Zarum, and Hu, 1995; Miller, Sarkar and Zarum, 1998; Miller 1998; Pattanyak, 2000). An exciting development is the and Sarkar, 1999; Monteoliva and Paz, 2000 & 2001). experimental study of the Loschmidt echo using NMR techiques (Levstein, Usaj, and Pastawski, 1998; Levstein et al, 2000, Jalabert and Pastawski, 2001) which sheds a VI. EINSELECTION AND MEASUREMENTS new light on the irreversibility in decohering complex dynamical systems. We shall briefly return to this subject It is often said that quantum states play only an epis- in Section VIII, while discussing experimental investiga- temological role in describing observers knowledge about tions relevant for decoherence and quantum chaos. the past measurement outcomes that have prepared the system (Jammer, 1974; d’Espagnat, 1976, 1995; Fuchs and Peres, 2000). In particular – and this is a key ar- 3. Quantum predictability horizon gument against their objective existence (against their ontological status) – it is impossible to find out what the The cross section I of the trajectory-like tube contain- state of an isolated quantum system is without prior in- ing the state of the harmonic oscillator in phase space formation about the observables used to prepare it: Mea- increases only slowly, Eq. (5.58), at a rate which – once surement of observables that do not commute with this the limiting Gaussian is reached – does not depend on I. original set will inevitably create a different state. By contrast, in chaotic quantum systems this rate is: The incessant monitoring of the einselected observables by the environment allows pointer states to exist in much I˙ I Λ(i) . (5.70) the same way classical states do. This ontological role of ≃ + Xi the einselected quantum states can be justified operationally, by showing that in presence of einselection one A fixed rate of entropy production implies an exponential can find out what the quantum state is, without inevitably increase of the cross-section of the tube of, say, the 1- re-preparing it by the measurement. Thus, einselected σ contour containing points consistent with the initial quantum states are no longer just epistemological. In a conditions: Phase space support expands exponentially. system monitored by the environment what is – the eins- This quantum view of chaotic evolution can be com- elected states – coincides with what is known to be – what pared with the classical “deterministic chaos”. In both is recorded by the environment (Zurek, 1993a,b; 1998a). cases, in the appropriate classical limit – which may in- The conflict between the quantum and the classical was volve either mathematicalh ¯ 0, or a macroscopic limit originally noted and discussed almost exclusively in the → – the future state of the system can be in principle pre- context of quantum measurements (Bohr, 1928; Mott, dicted to a set accuracy for an arbitrarily long time. How- 1929; von Neumann, 1932; Dirac, 1947; Zeh, 1971, 1973, ever, such predictability can be accomplished only when 1993; d’Espagnat, 1976, 1995; Zurek 1981, 1982, 1983, the initial conditions are given with the resolution that 1991, 1993a&b, 1998a; Omnès 1992, 1994; Elby, 1993, increases exponentially with the time interval over which 1998; Butterfield, 1996; Donald, 1995; Giulini et al., the predictions are to be valid. Given the fixed value 1996; Bub, 1997; Saunders, 1998; Healey, 1998; Baccia- ofh ¯, there is therefore a quantum predictability horizon galuppi and Hemmo, 1998; Healey and Hellman, 1998). after which the Wigner function of the system starting Here I shall consider quantum measurements, and, more from an initial minimum uncertainty Gaussian becomes to the point, acquisition of information in quantum the- stretched to the size of the order of the characteristic di- ory from the point of view of decoherence and einselec- mension of the system (Zurek, 1998b). The ability to tion. predict the location of the system in the phase space is then lost after t th¯ , Eq. (3.5), regardless of whether evolution is generated∼ by the Poisson or Moyal bracket, A. Objective existence of einselected states or, indeed, whether the system is closed or open. The case of regular systems is closer to the harmonic To demonstrate objective existence of einselected states oscillator. The rate of increase of the cross-section of we now develop an operational definition of existence and phase space “trajectory tube” consistent with the initial show how, in the open system, one can find out what the 31 state was and is, rather than just “prepare” it. This for existence: There may be in principle a pure state point was made before (Zurek, 1993a; 1998a), but this is of the Universe including the environment, the observer, first time I discuss it in more detail. and the measured system. While this may matter to Objective existence of states can be defined opera- some (Zeh, 2000), real observers are forced to perceive the tionally by considering two observers. The first of them Universe the way we do: We are a part of the Universe, is the record keeper R. He prepares the states with the observing it from within. Hence, for us, environment- original measurement and will use his records to find out induced superselection specifies what exists. if they were disturbed by the measurements carried out Predictability emerges as a key criterion of existence. by other observers, e.g. the spy S. The goal of S is to The only states R can rely on to store the information are discover the state of the system without perturbing it. the pointer states. They are also the obvious choice for When an observer can consistently find out the state of S to measure. Such measurements can be accomplished the system without changing it, that state – by our op- without danger of re-preparation. Einselected states are erational definition – will be said to exist objectively. insensitive to measurement of the pointer observables – In absence of einselection the situation of the spy S is they have already been “measured” by the environment. hopeless: R prepares states by measuring sets of com- Therefore, additional projections Pi onto the einselected muting observables. Unless S picks, by sheer luck, the basis will not perturb the density matrix (Zurek, 1993a) same observables in case of each state, his measurements – it will be the same before and after the measurement: will re-prepare states of the systems. Thus, when R re- ρD = P ρD P . (6.1) measures using the original observables, he will likely find after i before i Xi answers different from his records. The spy S will “get caught”, because it is impossible to find out an initially Correlations with the einselected states will be left intact unknown state of an isolated quantum system. (Zurek 1981; 1982). ˆ In presence of environmental monitoring the nature Superselection for the observable A = i λiPi with of the “game” between R and S is dramatically altered. essentially arbitrary non-degenerate eigenvaluesP λi and Now it is no longer possible for R to prepare an arbitrary eigenspaces Pi can be expressed (Bogolubov et al., 1990) pure state that will persist or predictably evolve without through Eq. (6.1). Einselection attains this, guarantee- losing purity. Only the einselected states that are al- ing diagonality of density matrices in the projectors Pi ready monitored by the environment – that are selected corresponding to pointer states. These are sometimes by the predictability sieve – will survive. By the same called decoherence - free subspaces when they are degen- token, S is no longer clueless about the observables he erate; compare also nonabelian case of noiseless subsys- can measure without leaving incriminating evidence. For tems discussed in quantum computation; see Zanardi and example, he can independently prepare and test survival Rasetti, 1997; Zanardi 1998; Duan and Guo 1998; Lidar, of various states in the same environment to establish Bacon, and Whaley, 1999; Knill, Laflamme, and Viola, which states are einselected, and then measure appropri- 2000; Blanchard and Olkiewicz, 2000; Zanardi 2000). ate pointer observables. Better yet, S can forgo direct measurements of the system, and gather information indirectly, by monitoring the environment. B. Measurements and memories This last strategy may seem contrived, but indirect measurements – acquisition of information about the sys- The memory of an apparatus or of an observer can tem by examining fragments of the environment that be modelled as an open quantum , interacting with through a Hamiltonian explicitlyA proportional to the have interacted with it – is in fact more or less the only S 5 strategy employed by the observers. Our eyes, for exam- measured observables ˆ ple, intercept only a small fraction of the photons that ∂ Hint = gsˆBˆ sˆ . (6.2) scatter from various objects. The rest of the photons con- − ∼ ∂Aˆ stitute the environment, which retains at least as complete a record of the same einselected observables as we Von Neumann (1932) considered an apparatus isolated can obtain (Zurek, 1993; 1998a). from the environment. At the instant of the interaction The environment acts as a persistent observer, dom- inating the game withE frequent questions, always about the same observables, compelling both R and S to focus 5 ˆ on the einselected states. Moreover, can be persuaded The observables ˆ of the system and B of the apparatus memory E need not be discrete with a simple spectrum as was previously to share its records of the system. This accessibility of assumed. Even whens ˆ has a complicated spectrum, the outcome the einselected states is not a violation of the basic tenets of the measurement can be recorded in the eigenstates of the of quantum physics. Rather, it is a consequence of the memory observable Aˆ, conjugate of Bˆ, Eq. (2.21). For the case of fact that the data required to turn quantum state into discretes ˆ the necessary calculations that attain premeasurement an ontological entity – an einselected pointer state – are – the quantum correlation that is the first step in the creation of the record – were already carried out in section II. For the other abundantly supplied by the environment. situations they are quite similar. In either case, they follow the We emphasize the operational nature of this criterion general outline of the von Neumann’s (1932) discussion. 32

between the apparatus and the measured system this is a (iv) Immediately after a measurement that yields the convenient assumption. For us it suffices to assume that, value oi the system is in the eigenstate oi of Oˆ. at that instant, the interaction Hamiltonian between the | i system and the apparatus dominates. This can be ac- (v) If the system is in a normalized state ψ , then a | i complished by taking the coupling g in Eq. (6.2) to be measurement of Oˆ will yield the value oi with the g(t) δ(t t0). Premeasurement happens at t0: probability p = o ψ 2. ∼ − i |h i| i| ( α s ) A α s A . (6.3) The first two axioms make no reference to measure- i| ii | 0i −→ i| ii| ii Xi Xi ments. They state the formalism of the theory. Axioms (iii) - (v) are, on the other hand, at the heart of the In practice the action is usually large enough to accom- present discussion. In the spirit, they go back to Bohr plish amplification. As we have seen in section II, all this and Born. In the letter, they follow von Neumann (1932) can be done without an appeal to the environment. and Dirac (1947). The two key issues are the projection For a real apparatus interaction with the environment postulate, implied by a combination of (iv) with (iiib), is inevitable. Idealized effectively classical memory will and the (Born’s) probability interpretation, axiom (v). retain correlations, but will be subject to einselection: To establish (iiib), (iv) and (v) we shall interpret – in Only the einselected memory states (rather than their operational terms – statements such as “the system is superpositions) will be useful for (or, for that matter, ac- in the eigenstate” and “measurement will yield value ... cessible to) the observer: Decoherence timescale is very with the probability ...” by specifying what these state- short compared to the time after which memory states ments mean for the sequences of records made and main- are typically consulted (i. e., copied or used in informa- tained by an idealized, but physical memory. tion processing), which is in turn much shorter than the We note that the above Copenhagen-like axioms pre- relaxation timescale, on which memory “forgets”. sume existence of quantum systems and of classical mea- Decoherence leads to classical correlation: suring devices. This (unstated) “axiom (o)” comple- P ments axioms (i) - (v). Our version of axiom (o) demands ρ = αiαj∗ si sj Ai Aj SA | ih || ih | −→ that the Universe consists of systems, and asserts that a Xi,j composite system can be described by a tensor product 2 D αi si si Ai Ai = ρ (58) of the Hilbert spaces of the constituent systems. Some −→ | | | ih || ih | SA Xi quantum systems can be measured, others can be used as measuring devices and/or memories, and as quantum following an entangling premeasurement. Left hand side environments that interact with either or both. of Eq. (6.4) coincides with Eq. (2.44c), the “outsiders” view of the classical measurement. We shall see how and Axioms (iii) - (v) contain many idealizations. For in- to what extent its other aspects – including the insiders stance, in real life or in laboratory practice measurements Eq. (2.44a) and the discoverers Eq. (2.44b) – can be have errors (and hence can yield outcomes other than the understood through einselection. eigenvalues oi). Moreover, only rarely do they “prepare” the system in the eigenstate of the observable they are designed to measure. Furthermore, coherent states – of- C. Axioms of quantum measurement theory ten an outcome of measurements, e. g., in quantum optics – form an overcomplete basis. Thus, their detection Our goal is to establish whether the above model can does not correspond to a measurement of a Hermitean fulfill requirements expected from measurement in text- observable. Last not least, the measured quantity may books (that are, essentially without exception, written in be inferred from some other quantity (e.g. deflection of a the spirit of the Copenhagen Interpretation). There are beam in the Stern-Gerlach experiment). Yet, we shall not several equivalent ‘textbook’ formulations of axioms of go beyond the idealizations of (i) - (v) above. Our goal is quantum theory. We shall (approximately) follow Farhi, to describe measurements in a quantum theory “without Goldstone, and Gutmann (1989) and posit: collapse”, to use axioms (o), (i) and (ii) to understand the origin of the other axioms. Non-ideal measurements (i) The states of a quantum system are associated are a fact of life incidental to this goal. with the vectors ψ which are theS elements of the Hilbert space | thati describes . HS S (ii) The states evolve according to i¯h ψ˙ = H ψ where 1. Observables are Hermitean – axiom (iiia) H is Hermitean. | i | i In the model of measurement considered here the ob- (iii a) Every observable O is associated with a Hermitian servables are Hermitean as a consequence of the assumed operator Oˆ. premeasurement interaction, e.g. Eq. (2.24). In particular, Hint is a product of the to-be-measured observ- (iii b) The only possible outcome of a measurement of O able of the system and of the “shift operator” in the is an eigenvalue oi of Oˆ. pointer of the apparatus or in the record state of the 33 memory. Interactions involving non-Hermitean operators apparatus decoheres, Eq. (6.3), so that it satisfies the (e.g., Hint a†b+ab†) may be, however, also considered. superselection condition, Eq. (6.1), for Pi = Ai Ai . It is tempting∼ to speculate that one could dispose of (ii) Einselection restricts states that can be read| ofih “as if| the observables (and, hence, of the postulate (iiia)) alto- they were classical” to pointer states. gether in the formulation of the axioms of quantum the- Indeed, following decoherence only pointer states ory. The only ingredients necessary to describe measure- Ai of the memory can be measured without diminish- ments are then the effectively classical, but ultimately ing{| thei} correlation with the states of the system. Without quantum apparatus, and the measured system. Observ- decoherence, as we have seen in section II, one could use ables emerge as a derived concept, as a useful idealiza- entanglement between and to end up with an al- tion, ultimately based on the structure of the Hamilto- most arbitrary superpositionS statesA of either, and, hence, nians. Their utility relies on the conservation laws that to violate the letter and the spirit of (iiib). relate outcomes of several measurements. Most basic of Outcomes are restricted to the eigenvalues of measured these laws states that the system that did not (have time observables because of einselection. Axiom (iiib) is then a to) evolve will be found in the same state when it is re- consequence of the effective classicality of pointer states, measured: This is the content of axiom (iv). Other con- the only ones that can be “found out” without being servation laws are also reflected in the patterns of correla- disturbed. They can be consulted repeatedly and remain tion in the measurement records, that must in turn arise unaffected under the joint scrutiny of the observers and from the underlying symmetries of the Hamiltonians. of the environment (Zurek, 1981; 1993a; 1998a). Einselection should be included in this program, as it decides which of the observables are accessible and useful – which are effectively classical. It is conceivable that 3. Immediate repeatability, axiom (iv) also the “fundamental” superselection may emerge in this manner (see Zeh, 1970; Zurek, 1982; for early specula- This axiom supplies the second half of the “collapse” tions; Giulini, Kiefer, and Zeh, 1995; and Kiefer, 1996; postulate. It asserts that in the absence of (the time Giulini, 2000, for the present status of this idea). for) evolution quantum system will remain in the same state, and its re-measurement will lead to the same outcome. Hence, once the system is found out to be in a 2. Eigenvalues as outcomes – axiom (iiib) certain state, it really is there. As in Eq. (2.44b) observer perceives potential options “collapse” to a single This statement is the first part of the “collapse” postu- actual outcome. (The association of the axiom (iv) with late. Given einselection, (iiib) is easy to justify: We need the collapse advocated here seems obvious, but it is not to show that only the records inscribed in the einselected common: Rather, some form of our axiom (iiib) is usually states of the apparatus pointer can be read off, and that – regarded as the sole “collapse postulate”.) in a well-designed measurement – they correlate with the eigenstates (and, therefore, eigenvalues) of the measured Immediate repeatability for Hermitean observables observables ˆ. with discrete spectra is straightforward to justify on the With Dirac (1947) and von Neumann (1932) we as- basis of Schrödinger evolution generated by Hint of Eq. sume that the apparatus is built so that it satisfies the (6.2) alone, although its implications depend on whether obvious “truth table” when the eigenstates of the mea- the premeasurement is followed by einselection. Everett sured observable are at the input: (1957) used the “no decoherence” version as a foundation of his relative state interpretation. On the other

si A0 si Ai (6.5) hand, without decoherence and einselection one could | i| i −→ | i| i postpone the choice of what was actually recorded by To assure this one can implement the interaction in ac- taking advantage of the entanglement between the sys- cord with Eq. (6.2) and the relevant discussion in sec- tem and the apparatus, evident on the left hand side tion II. This is not to say that there are no other ways: of Eq. (6.3). For instance, a measurement carried out Braginski and Khalili (1996), Aharonov, Anandan and on the apparatus in a basis different from Ai would {| i} Vaidman (1993), and Unruh (1994) have all considered also exhibit a one-to-one correlation with the system: “adiabatic measurements”, that correlate the apparatus αi si Ai = βk rk Bk . This flexibility to re- i | i| i k | i| i with the discrete energy eigenstates of the measured sys- writeP wavefunctionsP in different bases comes at a price tem, nearly independently of the structure of H . of relaxing the demand that the outcome states rk int {| i} Truth table of Eq. (6.5) does not require collapse – for be orthogonal (so that there would be no associated Her- any initial si it represents a “classical measurement in mitean observable). However, as was already noted, co- quantum notation”| i in the sense of Section II. However, herent states associated with a non-Hermitean annihila- Eq. (6.5) typically leads to a superposition of outcomes. tion operator can also be an outcome of a measurement. This is the “measurement problem”. To address it, we Therefore (and in spite of the strict interpretation of ax- assume that the record states Ai are einselected. This iom (iiia)) this is not a very serious restriction. has two related consequences:{| (i) Followingi} the measure- In presence of einselection the basis ambiguity disap- ment, the joint density matrix of the system and the pears. Immediate repeatability would apply only to the 34 records made in the einselected states. Other appara- and its eigenstates are entangled (even when the above a b tus observables lose correlation with the state of the sys- ρ ∨ is separable, and can be expressed as mixture of ma- tem on decoherence timescale. In the effectively classical tricesAS that have no entangled eigenstates). This would limit it is natural to demand repeatability extending be- imply an ambiguity of what are the record states, pre- yond that very short time interval. This demand makes cluding probability interpretation of measurement out- the role of einselection in establishing axiom (iv) evident. comes. Indeed, such repeatability is – albeit in a more general Observer may nevertheless have records of a system context – the motivation for the predictability sieve. that is in an ambiguous situation described above. Thus;

a b a a b b ρ ∨ = wk Ak Ak (pkρ k + pkρ k ) AS | ih | S S 4. Probabilities, einselection and records Xk

Density matrix alone – without the preferred set of are admissible for an effectively classical correlated states – does not suffice as a foundation for a probability with a quantum . Now the discord δ ( A )=0. S AI S|A interpretation. For, any mixed state density matrix ρ Mixing of ensembles of pairs of correlated systems one S can be decomposed into sums of density matrices that of which is subject to einselection does not lead to am- add up to the same resultant ρ , but need not share the biguities discussed above. Discord δ ( ) disappears S same eigenstates. For example, consider ρa and ρb , rep- in the einselected basis of , and theA eigenvaluesS|A of the resenting two different preparations (i.e., involvingS S mea- density matrices can behaveA as classical probabilities as- surement of two distinct, non-commuting observables) of sociated with ‘events’ – with the records. The “menu” two ensembles, each with multiple copies of a system . of possible events in the sample space – e.g., records in S When they are randomly mixed in proportions pa and memory – is fixed by einselection. Whether one can re- pb, the resulting density matrix: ally justify this interpretation of the eigenvalues of the reducued density matrix is a separate question we are a b a a b b ρ ∨ = p ρ + p ρ about to address. S S S is the complete description of the unified ensemble (see Schrödinger (1936); Jaynes (1957)). a b a b D. Probabilities from Envariance Unless [ρ ,ρ ] = 0, the eigenstates of ρ ∨ do not coincide with theS eigenstatesS of the components.S This feature The view of ‘the emergence of the classical’ based on makes it difficult to regard any density matrix in terms the environment – induced superselection has been occas- of probabilities and ignorance. Such ambiguity would be sionally described as “for practical purposes only” (see, especially troubling if it arose in the description of an observer (or, for that matter, of any classical system). Ignorance interpretation – i.e., the idea that probabilities are observer’s way of dealing with uncertainty about the“quantumness” of correlations. It should disappear as a re- the outcome we have briefly explored in the discussion of sult of the classical equivalence of two definitions of the mutual the insider-outsider dichotomy, Eqs. (2.44), – requires at information, but is in general positive for quantum correlations, the very least that the set of events (“the sample space”) including, in particular, pre-decoherence ρSA. Discord is asym- exists independently of the information at hand – i.e, metric, δIA(S|A) 6= δIS (A|S). Vanishing of δIA(S|A) (i.e., of the discord in the direction exploring the classicality of the states independently of pa and pb in the example above. Eigen- of A, on which H(ρS|A) in the asymmetric JA(S : A), Eq. (4.32), states of the density matrix do not supply such events, is conditioned) is necessary for the classicality of the measure- since the additional loss of information associated with ment outcome (Zurek, 2000; Ollivier and Zurek, 2002, Zurek, mixing of the ensembles alters the candidate events. 2002a). δIA(S|A) can disappear as a result of decoherence in Basis ambiguity would be disastrous for record states. the einselected basis of the apparatus. Following einselection it is then possible to ascribe probabilities to the pointer states. Density matrices describing a joint state of the memory In perfect measurements of Hermitean observables discord van- and of the system : ishes “both ways”: δIA(S|A) = δIS (A|S) = 0 for the pointer A S basis and for the eigenbasis of the measured observable correlated a b a b ρ ∨ = paρ + pbρ with it. Nevertheless, it is possible to encounter situations when AS AS AS vanishing of the discord in one direction is not accompanied by would have to be considered. In absence of einselection its vanishing “in reverse”. Such correlations are “classical one a b way” (Zurek, 2002a) and could arise, for instance, in measure- the eigenstates of such ρ ∨ need not be even associated ments of non-hermitean observables (which – as we have already with a fixed set of recordAS states of the presumably clas- noted – happen, in spite of the axiom (iii)). a b 6 sical . Indeed, in general ρ ∨ has a non-zero discord, This asymmetry between classical A and quantum S arises from A AS the einselection. Classical record states are not arbitrary superpositions. Observer accesses his memory in the basis in which it is monitored by the environment. The information stored is effectively classical because it is being widely disseminated. States 6 As we have seen in section IV, Eqs. (4.30) – (4.36), of the observers memory exist objectively – they can be found discord δIA(S|A) = I(S : A) −JA(S : A) is a measure of out through their imprints in the environment. 35 e.g., Bell, 1990), to be contrasted with more fundamental of Wheeler and Zurek, 1983), but not developed untill (if nonexistent) solutions of the problem one could imag- now. Envariance (Zurek, 2002b) addresses the question ine (i.e., by modifying quantum theory; see Bell, 1987, of meaning of these probabilities by defining “ignorance”, 1990). This attitude can be in part traced to the reliance and justifies a relative frequency argument, although in of einselection on reduced density matrices: For, even a manner different from the previous attempts. when explanations of all aspects of the effectively classical behavior are accepted in the framework of, say, Everett’s MWI, and after the operational approach to objectivity 1. Envariance and perception of unique outcomes based on the existential interpretation explained earlier is adopted, one Environment - assisted invariance is a symmetry ex- major gap remains: Born’s rule – axiom (v) connecting hibited by the system correlated with the other system probabilities with the amplitudes, p = ψ 2 – has to be (usually ‘the environment’S ). When a state of the com- k k E postulated in addition to axioms (o) - (ii).| True,| one can posite can be transformed by u acting solely on the show that within the framework of einselection Born’s HilbertSE space , but the effect ofS this transformation HS rule emerges naturally (Zurek, 1998a). Decoherence is, can be undone with an appropriate u acting only on , however, based on on reduced density matrices. Now, so that the joint state ψ remainsE unaltered; HE | SE i since they were introduced by Landau (1927), it is known that “partial trace” leading to reduced density matrices u u ψ = ψ (6.6) S E | SE i | SE i is predicated on Born’s rule (see Nielsen and Chuang, such ψ is envariant under u . Generalization to 2000, for discussion). Thus, derivations of Born’s rule | SE i S that employ reduced density matrices are open to charge mixed ρ is obvious, but we shall find it easier to assume thatSE has been purified in the usual fashion – of circularity (Zeh, 1997). Moreover, repeated attempts SE 2 i.e., by enlarging the environment. to justify pk = ψk within the no - collapse MWI (Ev- erett, 1957 a&b;| DeWitt,| 1970; DeWitt and Graham, Envariance is best elucidated by considering an example – an entangled state of and . It can be expressed 1973; Geroch, 1984) have failed (see e.g., Stein, 1984; S E Kent, 1990; Squires, 1990): The problem is their circu- in the Schmidt basis as: larity: An appeal to the connection (especially in certain ψ = αk sk εk , (6.7) limiting procedures) between the smallnes of the ampli- | SE i | i| i tude and the vanishing of probabilities has to be made Xk to establish that relative frequencies of events averaged where αk are complex, while sk and εk are or- over branches of the universal state vector are consitent thonormal. For ψ (and, hence{| i} – given{| ouri} above re- with Born’s rule. In particular, one must a claim that mark about purification| SE i – for any system correlated with “maverick” branches of the MWI state vector that have the environment) it is easy to demonstrate: “wrong” relative frequencies are of measure zero because Lemma 6.1 their Hilbert space measures are small. This is circular, : Unitary transformations co-diagonal with the Schmidt basis of ψ leave it envariant. as noted even by the proponents (DeWitt, 1970). | SE i The proof relies on the form of such transformations: My aim here is to look at the origin of ignorance, information, and, therefore, probabilities from a very quan- sk iσk u{| i} = e sk sk , (6.8) tum and fundamental perspective: Rather than focus on S | ih | Xk probabilities for an individual isolated system I shall – in the spirit of einselection, but without employing its usual where σk is a phase. Hence; tools such as trace or reduced density matrices – con- sk iσk sider what the observer can (and cannot) know about a u{| i} ψ = αke sk εk , (6.9) S | SE i | i| i system entangled with its environment. Within this con- Xk text I shall demonstrate that Born’s rule follows from the very quantum fact that one can know precisely the state can be undone by: of the composite system and yet be provably ignorant of εk iǫk the state of its components. This is due to environment - u{| i} = e εk εk (6.10) E | ih | assited invariance or envariance, a hitherto unrecognised Xk symmetry I am about to describe. Envariance of pure providing that ǫ =2πl σ for some integer l . QED. states is conspicuously missing from classical physics. It k k − k k allows one to define ignorance as aconsequence of invari- Thus, phases associated with the Schmidt basis are ance, and thus to understand the origin of Born’s rule, envariant. We shall see below that they are the only probabilities, and ultimately the origin of information envariant property of entangled states. The transforma- through arguments based on assumptions different from tions defined by Eq. (6.8) are rather specific – they share Gleason’s (1957) famous theorem. Rather, it is based (Schmidt) eigenstates. Still, their existence leads us to: on the Machian idea of ‘relativity of quantum states’ en- Theorem 6.1: Local description of the system entan- tertained by this author two decades ago (see p. 772 gled with a causally disconnected environmentS must E 36 not depend on the phases of the coefficients αk in the αk = αj , the effect of a swap on the system can be Schmidt decomposition of ψ . |undone| | by| an obvious counterswap in the environment: | SE i all the measurable properties of the It follows that i(φkj +φk φj +2πlkj ) are completely specified by the list of pairs α ; s S. u (k j)= e− − εk εj + h.c. (6.12) {| k| | ki} S ↔ | ih | An equivalent way of establishing this phase envariance Swap can be applied to states that do not have equal theorem appeals even more directly to causality: Phases absolute values of the coefficients, but in that case it is of ψ can be arbitrarily changed by acting on alone | SE i E no longer envariant. Partial swaps can be also generated (e.g, by the local Hamiltonian with eigenstates εk , gen- ri e.g., by underrotating or by a u{| i}, Eq. (6.8), but erating evolution of the from of Eq. (6.9)). But| causal-i with the eigenstates r intermediateS between these of ity prevents faster than light communication. Hence, no i the swapped and the{| complementaryi} (Hadamard) basis. measurable property of can be effected. QED. Swap followed by a counterswap exchanges coefficients of Phase envariance theoremS will turn out to be the crux the swapped states in the Schmidt expansion, Eq. (6.7). of our argument. It relies on an input – entanglement and Classical correlated states can also exhibit some- causality – which has not been employed to date in dis- thing akin to envariance under a classical version of cussions of the origin of probabilities. In particular, this “swaps”. For instance, a correlated state of system and input is different and more “physical” than that of the an apparatus described by: ρ s s A A + succesfull derivation of Born’s rules by Gleason (1957). k k k k s s A A can be swappedSA ∼ and | counterswapped.ih || ih | We also note that information contained in the “data j j j j The| ih corresponding|| ih | transformations would be still given base” α ; s implied by the Theorem 6.1 is the k k by, in effect, Eqs. (6.11) - (6.12), but without phases, same as{| in| the| i} reduced density matrix of the system and swaps could no longer be generated by the rota- ρ : Although we do not yet know probabilities of var- S tions around the complementary basis. This situation ious s , preferred basis of has been singled out – k corresponds to the “outsiders view” of the measurement Schmidt| i states (sometimes regardedS as instantaneous process, Eq. (2.44c): Outsider can be aware of the cor- pointer states; see e.g. Albrecht, (1992; 1993)) play a relation between the system and of the apparatus, but special role as the eigenstates of envariant transforma- ignorant of their individual states. This connection be- tions. Moreover, probabilities can depend on α (but k tween ignorance and envariance shall be exploited below. not on the phases). We still do not know that p | =| α 2. k k Envariance based on ignorance may be found in the The causality argument we could have used to establis| | classical setting, but envariance of pure states is purely Theorem 6.1 applies of course to arbitrary transforma- quantum: Observers can know perfectly the quantum tions one could perform on . However, such transfor- joint state of , yet be provably ignorant of . Consider mations would not be in generalE envariant (i.e., could not a measurementSE carried out on the state vectorS of from be undone by acting on alone). Indeed – by the same the point of view of envariance: SE token – all envariant transformationsS must be diagonal in Schmidt basis: N N Lemma 6.2: All of the unitary envariant transforma- A0 sk εk Ak sk εk Φ (6.13) | i | i| i → | i| i| i ∼ | SAE i tions of ψ have Schmidt eigenstates. kX=1 Xk=1 | SE i The proof relies on the fact that other unitary trans- Above, we have assumed that the absolute values of the formations would rotate Schmidt basis, s s˜ . The | ki → | ki coefficients are equal (and omitted them for notational rotated basis becomes a new ‘Schmidt’, and this fact can- simplicity). We also have ignored phases (which need not not be affected by unitary transformations of – by state E be equal) since – by the phase envariance theorem – they rotations in the environment. But a state that has a will not influence the state (and, hence, the probabilities) different Schmidt decomposition from the original ψ | SE i associated with . is different. Hence, unitary transformation must be co- Before the measurementS observer with access to can- diagonal with Schmidt states to leave it envariant. QED. not notice swaps in the state (such as Eq. (6.13))S with equal absolute values of the Schmidt coefficients. This follows from envariance of the pre-measurement ψ 2. Born’s rule from envariance under swaps, Eq. (6.11). | SE i One could argue this point in more detail by compar- When absolute values of some of the coefficients in Eq. ing what happens for two very different input states; an (6.7) are equal, any orthonormal basis is ‘Schmidt’ in the entangled ψ with equal absolute values of Schmidt corresponding subspace of . This implies envariance | SE i S coefficients and a product state: of more general nature, e.g.H under a swap;

iφkj u (k j)= e sk sj + h.c. (6.11) ϕ = sJ εJ . S ↔ | ih | | SE i | i| i Swap can be generated by a phase rotation, Eq. (6.8), When observer knows he is dealing with ϕ , he knows but in a basis complementary to the one swapped. Its the state and can predict outcome of theSE corresonding envariance does not contradict Lemma 6.2, as any or- measurement on : Schrödinger equation or just the re- thonormal basis in this case is also ‘Schmidt’). So, when sulting truth table,S Eq. (6.5), imply with certainty that 37 his state – the future state of his memory – will be A . Thus, if another observer (‘Wigner’s friend’) was getting | J i Moreover, swaps involving sJ are not envariant for ϕ . ready to find out – either by direct measurement of They just swap the outcomes| i (i.e. when u (J L) pre-SE or by communicating with observer – the outcome ofS S ↔ A cedes the measurement, memory will end up in AL ). his measurement, he would be (on the basis of envari- By contrast, | i ance) provably ignorant of the outcome has detected, but could be certain of the correlation.A We shall em- N AS iφk ploy this joint envariance in the discussion of the case of ψ e sk εk | SE i∼ | i| i unequal probabilities immediately below. Xk=1 Note that our reasoning does not really appeal to the is envariant under swaps. This allows the observer (who “information lost in the environment” in a sense in which knows the joint state of exactly) to conclude that the SE this phrase is often used. Perfect knowledge of the com- probabilities of all the envariantly swappable outcomes bined state of the system and the environment is the basis must be the same. Observer cannot predict his memory of the argument for the ignorance od alone: For entan- state after the measurement of because he knows too gled , perfect knowlegde of isS incompatible with much; the exact combined state ofS . SE SE SE perfect knowledge of . This is really a consequence of For completeness, we note that when there are system indeterminacy – jointS observables with entangled eigenstates that are absent from the above sum – i.e., states states such as ψ simply do not commute (as the reader that ‘appear with zero amplitude’ – they cannot be en- is invited to verify)SE with the observables of the system variantly swapped with the states present in the sum. Of alone. Hence, ignorance associated with envariance is course, observer can predict with certainty he will not de- ultimately mandated by Heisenberg indeterminacy. tect any of the corresponding zero - amplitude outcomes. The case of unequal coefficients can be reduced to the This argument about the ignorance of the observer con- case of equal coefficients. This can be done in several cerning his future state – concerning the outcome of the ways, of which we choose one that makes use of the premeasurement he is about to perform – is based on his ceding discussion of envariance of the post-measurement perfect knowledge of a joint state of . state. We start with: Probabilities refer to the guess observerSE makes on the basis of his information before the measurement about N the state of his memory – “the future outcome” – after Φ αk Ak sk εk (6.17) | SAE i∼ | i| i| i the measurement. As the left hand side of Eq. (6.13) is Xk=1 envariant under swaps of the system states, probabilities of all the states must be equal. Thus, by normalisation; where αk √mk and mk is a natural number (and, by phase envariance∼ theorem, we drop the phases). To get pk =1/N . (6.14) an envariant state we “increase the resolution” of by A Moreover, probability of n mutually exclusive events that assuming that; all appear in Eq. (6.13) with equal coefficients must be: mk Ak = aj /√m (6.18) pk k ... kn = n/N . (6.15) k k 1∨ 2∨ ∨ | i | i jXk =1 This concludes discussion of the equal probability case. Our case rests on the independence of the state of An increase of resolution is a standard trick, used in clas- S entangled with from the phases of the coefficients in sical probability theory “to even the odds”. Note that we E the Schmidt representation – the Theorem 6.1 – which in assume that basis states such as Ak are normalised (as the case of equal coefficients Eq. (6.13), allows envariant they must be in a Hilbert space).| Thisi leads to: swapping, and yields Eqs. (6.14)-(6.15). After a measurement situation changes. In accord N mk a jk=1 jk with our preceding discussion we interpret presence of Φ √mk | i sk εk (6.19) | SAE i∼ P √m | i| i the term A in Eq. (6.13) as evidence that an out- kX=1 k | ki come sk can be (or indeed has been – the language here is somewhat| i dependent on the interpretation) recorded. We now assume that and interact (e.g., through c-shift A E Conversely, absence of some A ′ in the sum above im- a of section 2, with a truth table ajk εk | k i | i| i → plies that the outcome s ′ cannot occur. After a mea- ajk ejk where ejk are all orthonormal), so that af- | k i | i| i {| i} surement memory of the observer who has detected sk ter simplifying and re-arranging terms we get a sum, over | i a new ‘fine-grained’ index, with the states of that re- will contain the record Ak . Futher measurements of the S same observable on the| samei system will confirm that main the same within ‘coarse - grained cells’, i.e., inter- S is in indeed in the state sk . vals measured by mk: This post-measurement| statei is still envariant, but only N mk M under swaps that involve jointly the state of the system ˜ Φ sk ( ajk ejk )= sk(j) aj ej and the correlated state of the memory: | SAE i∼ | i | i| i | i| i| i kX=1 jXk =1 Xj=1 iφkj u (k j)= e sk, Ak sj , Aj + h.c. (6.16) (6.20) AS ↔ | ih | 38

Above, M = N m , and k(j)=1 for j m , k(j)= 1989; Aharonov and Resnik, 2002). In particular, in the k=1 k ≤ 1 2 for j m1P+ m2, etc. The above state is envariant context of the “no collapse” MWI relative frequency seem under combined≤ swaps; to offer the best hope of arriving at the Born’s rule and elucidating its physical significance. Yet, it is generally u (j j′) = exp(iφjj′ ) sk(j),aj aj′ ,sk(j′ ) + h.c. acknowledged that the MWI derivations offered to date SA ↔ | ih | have failed to attain this goal (Kent, 1990). Suppose that an additional observer measures in the SA We postpone brief discussion of these efforts to the obviously swappable joint basis. By our equal coefficients next section, and describe an approach to relative fre- argument, Eq. (6.14), we get p(sk(j),aj )=1/M. But the quencies based on envariance. Consider an ensemble of observer can ignore states aj. Then the probability of many ( ) distinguishable systems prepared in the same different Schmidt states of is, by Eq. (6.15); N S initial state: 2 p(sk)= mk/M = αk . (6.21) σ = α 0 + β 1 (6.23) | | | S i | i | i This is Born’s rule. We focus on the two state case to simplify the notation. The case with coefficients that do not lead to commen- We also assume that α 2 and β 2 are commensurate, so surate probabilities can be treated by assuming continu- that the state vector of| | the whole| | ensemble of correlated ity of probabilities as a function of the amplitudes and triplets after the requisite increases of resolution taking appropriate (and obvious) limits. This can be (see Eqs.SAE (6.18)-(6.20) above) is given by: physically motivated: One would not expect probabilities to change drastically depending on the infinitesimal m M changes of the state. One can also extend the strategy ΦN 0 aj ej + 1 aj ej ⊗N (6.24) | SAE i∼ | i| i| i | i| i| i outlined above to deal with probabilities (and probabil- Xj=1 j=Xm+1 ity densities) in cases such as s(x) , i.e. when the index save for the obvious normalisation. This state is envari- of the state vector changes continuously.| i This can be ac- ant under swaps of the joint states s,a , as they appear complished by discretising it (so that the measurement j with the same (absolute value) of| the amplitudei in Eq. of Eq. (6.17) correlates different apparatus states with (6.24). (By Theorem 6.1 we can omitt phases.) small intervals of x) and then repeating the strategy of After the exponentiation is carried out, and the result- Eqs. (6.17) - (6.21). The wavefuction s(x) should be ing product states are sorted by the number of 0’s and sufficiently smooth for this strategy to succeed. 1’s in the records, we can calculate the number of terms We note that the “increase of resolution” we have ex- n n with exactly n 0’s, ν (n)= N m (M m) . To get ploited, Eqs. (6.18)-(6.21), need not be physically imple- n N− probability, we normalise:N − mented for the argument to go through: The very possibility of carrying out these steps within the quantum n n m (M m)N− 2n 2( n) formalism forces one to adopt Born’s rule. For example, p (n)= N − = N α β N− . N n M n | | | | if the apparatus did not have the requisite extra resolu- N (6.25) tion, Eq. (6.18), interaction of the environment with a This is the distribution one would expect from Born’s still different ‘counterweight’ system that yields C rule. To establish the connection with relative frequen- N cies we appeal to the de Moivre - Laplace theorem (Gne- denko, 1982) which allows one to approximate above Ψ = √mk sk Ak εk Ck (6.22) | SAECi | i| i| i| i p (n) with a Gaussian: Xk=1 N 2 2 would lead one to the Born’s rule through steps similar 1 1 n−N |α| p (n) e− 2 √N |αβ| (6.26) to these we have invoked before, providing that Ck N ≃ √2π αβ mk | i} has the requisite resolution, Ck = cj /√m . N | | | i jk=1 | k i k Interaction resulting in a correlation, Eq. (6.22), can This last step requires large , but our previous discus- P N occur between and , and happen far from the sys- sion including Eq. (6.25) is valid for arbitrary . Indeed, N tem of interestE or fromC the apparatus. Thus, it will not equation (6.21) can be regarded as the = 1 case. N influence probabilities of the outcomes of measurements Nevertheless, for large relative frequency is sharply N carried out on or of the records made by . Yet, the peaked around the expected n = α 2. Indeed, in the h i N | | fact that it canS happen leads us to the desiredA conclusion. limit appropriately rescaled p (n) tends to a DiracNδ(υ →α ∞2) in relative frequency υ =Nn/ . This justifies the−| relative| frequency interpretation ofN the squares 3. Relative frequencies from envariance of amplitudes as probabilities in the MWI context. ‘Mav- erick universes’ with different relative frequencies exist, Relative frequency is a common theme studied with the but have a vanishing probability (and not just vanishing aim of elucidating the physical meaning of probabilities Hilbert space measure) for large . in quantum theory (Everett, 1957; Hartle, 1968; DeWitt, Our derivation of the physicalN significance of proba- 1970; Graham, 1970; Farhi, Goldstone, and Gutmann, bilities – while it led to the relative frequency argument 39

– was based on a very different set of assumptions than versions of this approach that allow for the measurement previous derivations. The key idea behind it is the con- postulates (iii) and (iv) (Hartle, 1968; Farhi, Goldstone nection between a symmetry (envariance) and ignorance and Guttmann, 1989). Nevertheless, for the infinite en- (impossibility of knowing something). The unusual fea- sembles considered in the above references, (where, in ture of our argument is that this ignorance (for an in- effect, the Hilbert space measure of the MWI branches dividual system ) is demonstrated by appealing to the that violate relative frequency predictions is zero) that perfect knowledgeS of the larger joint system that includes the eigenvalues of the frequency operator acting on a large as a subsystem. or infinite ensemble of identical states will be consistent S We emphasize that one could not carry out the ba- with the (Born formula) prescription for probabilities. sic step of our argument – the proof of the indepen- However, the infinite size of the ensemble necessary to dence of the likelihoods from the phases of the Schmidt prove this point is troubling (and unphysical) and tak- expansion coefficients – for an equal amplitude pure ing the limit starting from a finite case is difficult to state of a single, isolated system. The problem with: justify (Stein, 1984; Kent, 1990; Squires, 1990). More- 1 N ψ = N − 2 exp(iφk) k is the accessibility of the over, the frequency operator is a collective observable of | i k | i phases. Consider,P for instance; ψ 0 + 1 2 and the whole ensemble. It may be possible to relate observ- | i ∼ | i | i − | i ψ′ 2 + 1 0 . In absence of decoherence swap- ables defined for such an infinite ensemble supersystem | i ∼ | i | i − | i ping of k’s is detectable: Interference measurements (i.e., to the states of individual subsystems, but the frequency measurements of the observables with phase-dependent operator does not do it. This is well illustrated by the eigenstates 1 + 2 ; 1 2 , etc.) would have revealed gedankenexperiment envisaged by Farhi et al. (1989). | i | i | i − | i the difference between ψ and ψ′ . Indeed, given an en- To provide a physical implementation of the frequency | i | i semble of identical pure states an observer will simply find operator they consider a version of the Stern-Gerlach ex- out what they are. Loss of phase coherence is essential periment where all the spins are attached to a common to allow for the shuffling of the states and coefficients. lattice, so that – during the passage through the inhomo- Note that in our derivation environment and einselec- geneity of the magnetic field – the center of mass of the tion play an additional, more subtle role: Once a mea- whole lattice is deflected by an angle proportional to the surement has taken place – i.e., a correlation with the projection of the net magnetic moment associated with apparatus or with the memory of the observer was estab- the spins on the direction defined by the field gradient. lished – one would hope that records will retain validity The deflection is proportional to the eigenvalue of the over a long time, well beyond the decoherence timescale. frequency operator that is then a collective observable – This is a pre-condition for the axiom (iv). Thus, a “col- states of individual spins remain in superpositions, un- lapse” from a multitude of possibilities to a single reality correlated with anything outside. This difficulty can be can be confirmed by subsequent measurements only in addressed with the help of decoherence (Zurek, 1998a), the einselected pointer basis. but using decoherence without justifying Born’s formula first is fraught with danger of circularity. Measure of certainty seems to be a rather vague con- 4. Other approaches to probabilities cept. Yet, Cox (1946) has demonstrated that Boolean logic leads – after addition of a few reasonable assump- Gnedenko (1982), in his classic textbook, lists three tions – to the definition of probabilities that, in a sense, classical approaches to probability: appear as an extension of the logical truth values. How- a. Definitions that appeal to relative frequency of oc- ever, the rules of symbolic logic that underlie Cox’s theo- currence of events in a large number of trials. rems are classical. One can adopt this approach (Zurek, 1998a) to probabilities in quantum physics only after de- b. Definitions of probability as a measure of certainty coherence “intervenes” restoring the validity of the dis- of the observer. tributive law, which is not valid in quantum physics (Birkhoff and von Neumann, 1936). c. Definitions that reduce probability to the more primitive notion of equal likelihood. One can carry out equal likelihood approach in the context of decoherence (Zurek, 1998a). The problem is In the quantum setting, the relative frequency ap- – as pointed out before – the use of trace, and the dan- proach has been to date the most popular, especially gers of circularity. An attempt to pursue a strategy akin in the context of the “no collapse” MWI ’ (Everett, to equal likelihood in the quantum setting at the level 1957a&b; Graham, 1970; DeWitt, 1970). Counting the of pure states of individual systems has been also made number of the “clicks” seems most directly tied to the ex- by Deutsch in his (unpublished) “signalling” approach to perimental manifestations of probability. Yet, Everett in- probabilities. The key idea is to consider a source of pure terpretation versions were generally found lacking (Kent, states, and to find out when the permutations of a set of 1990; Squires, 1990), as they relied on circular reasoning, basis states can be detected, and, therefore, used for com- invoking without physical justification an abstract mea- munication. When permutations are undetectable, prob- sure of Hilbert space to obtain a physical measure (fre- abilities of the permuted set of states are declared equal. quency). Some of the criticisms seem relevant also for the The problem with this idea (or with its more formal ver- 40 sion described by DeWitt, 1998) is that it works only for ences therein) uses this impossibility of finding out what superposition that have all the coefficients identical, in- is an unknown state of an isolated quantum system. On cluding their phases. Thus, as we have already noted, for the other hand classical reality seems to be made up closed systems phases matter and there is no invariance of quantum building blocks: States of macroscopic sys- under swapping. In a recent paper Deutsch (1999) has tems exist objectively – they can be found out by many adopted a different approach based on decision theory. observers independently without being destroyed or re- The basic argument focuses again on individual states of prepared. So – the question arises – how can objective quantum systems, but – as noted in the critical comment existence — the “reality” of the classical states – emerge by Barnum et al. (2000) – seems to make appeal to some from “purely epistemic” wavefunctions? of the aspects of decision theory that do depend on prob- There is not much one can do about this in case of a abilities. In my view, it also leaves the problem of the single state of an isolated quantum system. But open phase dependence of the coefficients unaddressed. systems are subject to einselection and can bridge the Among other approaches, recent work of Gottfried chasm dividing their epistemic and ontic roles. The most (2000) shows that in a discrete quantum system coupled direct way to see this arises from the recognition of the with a continuous quantum system Born’s formula fol- fact that we never directly observe any system. Rather, lows from the demand that the continuous system should we discover states of macroscopic systems from the im- follow classical mechanics in the appropriate limit. A prints they make on the environment: A small fraction of somewhat different strategy, with a focus on the coin- the photon environment intercepted by our eyes is often cidences of the expected magnitude of fluctuations was all that is needed. States that are recorded most redun- proposed by Aharonov and Resnik (2002). dantly in the rest of the Universe (Zurek, 1983; 1998a; In comparison with all of the above strategies, ‘prob- 2000) are also easiest to discover. They can be found out abilities from envariance’ is the most radically quantum, indirectly, from multiple copies of the evidence imprinted in that it ultimately relies on entanglement (which is still in the environment, without a threat to their existence. sometimes regarded as ‘a paradox’, and ‘to be explained’: Such states exist and are real – they can be found out I have used it as an explanation). This may be the reason without being destroyed as if they were really classical. why it has not been discovered untill now. The insight Environmental monitoring creates an ensemble of “wit- offered by envariance into the nature ignorance and in- ness states” in the subsystems of the environment, that formation sheds a new light on probabilities in physics. allow one to invoke some of the methods of the statistical The (very quantum) ability to prove ignorance of a part interpretation (Ballentine, 1970) while subverting its ide- of a system by appealing to perfect knowledge of the ology – to work with an ensemble of objective evidence of whole may resolve some of the difficulties of the classical a state of a single system. From this ensemble of witness approaches. states one can infer the state of the quantum system that has led to such “advertising”. This can be done without disrupting the einselected states. VII. ENVIRONMENT AS A WITNESS Predictability sieve selects states that entangle least with the environment. Question about predictability si- Emergence of classicality can be viewed either as a con- multaneously lead to states that are most redundantly sequence of the widespread dissemination of the informa- recorded in the environment. Indeed, this idea touches tion about the pointer states through the environment, on the “quantum Darwinism” we have alluded to in the or as a result of the censorship imposed by decoherence. introduction: The einselected pointer states are not only So far I have focused on this second view, defining ex- best in surviving the environment, but, also, they broad- cast the information about themselves – spread out their istence as persistence – predictability in spite of the environmental monitoring. Predictability sieve is a way of “copies” – throughout the rest of the Universe: Amplified information is easiest to amplify. This leads to analogies discovering states that are classical in this sense (Zurek, 1993a&b; Zurek, Habib and Paz, 1993; Gallis, 1996). with “fitness” in the Darwinian sense, and suggests looking at einselection as a sort of natural selection. A complementary approach focusses not on the system, but on the records of its state spread throughout the environment. Instead of “the least perturbed states” one A. Quantum Darwinism can ask “what states of the system are easiest to discover by looking at the environment”. Thus, environment is no Consider the “bit by byte” example of Section IV. Spin longer just a source of decoherence, but acquires a role of - system is correlated with the environment: a communication channel with a basis - dependent noise S that is minimised by the preferred pointer states. ψ = a 00 . . . 0 + b 11 . . . 1 | SE i | ↑i| i | ↓i| i This approach can be motivated by the old dilemma: = a + b (59) | ↑i|E↑i | ↓i|E↓i On the one hand, quantum states of isolated systems are The basis , of is singled out by the redundancy purely “epistemic” (see e.g., Peres, (1993); Fuchs and of the record.{| ↑i By| ↓i} comparison,S the same ψ is: Peres (2000)). Quantum cryptography (Bennett and Di- | SE i Vincenzo, 2000; Nielsen and Chuang, 2000, and refer- ψ = (a 00 . . . 0 + b 11 . . . 1 )√2 | SE i |⊙i | i | i 41

+ (a 00 . . . 0 b 11 . . . 1 )/√2 in a different guise: Rather than search for predictable |⊗i | i− | i = ( + )/√2 (60) sets of states of the system we are now looking for the |⊙i|E⊙i |⊗i|E⊗i records of the states of the system in the environment. in terms of the Hadamard-transformed , . Sequences of states of environment subsystems correlated One can find out whether is {|⊙ior |⊗i}from a with pointer states are mutually predictable and, hence, small subset of the environmentS bits.| By ↑i contrast,| ↓i states collectively algorithmically simple. States that are pre- , cannot be easily inferred from the environ- dictable in spite of the interactions with the environment ment.{|⊙i |⊗i} States , are typically not even orthogo- are also easiest to predict from their impact on its state. nal, ={|Ea⊙2i |E⊗bi}2. And even when a 2 b 2 = 0, The state of the form of Eq. (7.1) can serve as an exam- the recordhE⊙|E⊗ ini the| | environment− | | is fragile: Only| | − one | | rela- ple of amplification. Generation of redundancy through tive phase distinguishes from in that case, in amplification brings about objective existence of the oth- contrast with multiple records|E⊙i of the|E⊗i pointer states in erwise subjective quantum states. States and of and . Remarks that elaborate this observation the system can be found out reliably from| ↑i a small| ↓ifrac- |Efollow.↑i They|E↓i correspond to several distinct measures of tion of the environment. By contrast, to find out whether ‘fitness’ of states. the system was in a state or one would need to detect all of the environment.|⊙i Objectivity|⊗i can be defined as the ability of many observers to reach consensus inde- 1. Consensus and algorithmic simplicity pendently. Such consensus concerning states and is easily established – many ( N/n) can independently| ↑i | ↓i ¿From the state vector ψ , Eqs. (7.1) and (7.2), measure fragments of the environment.∼ observer can find the state| ofSE thei quantum system just by looking at the environment. To accomplish this, the total N of the environment bits can be divided into sam- 2. Action distance ples of n bits each, with 1 n N. These samples ≪ ≪ can be then measured using observables that are the One measure of robustness of the environmental same within each sample, but differ between samples. records is the action distance (Zurek, 1998a). It is given They may correspond, for example, to different antipo- by the total action necessary to undo the distinction be- dal points in the Bloch spheres of the environment bits. tween the states of the environment corresponding to dif- In the basis 0 , 1 (or bases closely aligned with it) ferent states of the system, subject to the constraints {| i | i} the record inferred from the bits of information scattered arising from the fact that the environment consists of in the environment will be easiest to come by. Thus, subsystems. Thus, to obliterate the difference between starting from the environment part of ψ , Eq. (7.1), and in Eq. (7.1), one needs to “flip” one-by-one | SE i ↑ ↓ the observer can find out, with no prior knowledge, the N|E isubsystems|E i of the environment. That implies an ac- state of the system: Redundancy of the record in the en- tion – i.e., the least total angle by which a states must vironment allows for a trial-and-error ‘indirect’ approach be rotated, see Section IIB – of: while leaving the system untouched. π In particular, measurement of n environment bits ∆( , )= N ¯h . (7.5) in a Hadamard transform of the basis 0 , 1 |E↑i |E↓i h 2 · i {| i | i} yields a random-looking sequence of outcomes (i.e., By contrast a “flip” of phase of just one bit will reverse + , ,... ). This record is algorithmically {| i1 |−i2 |−in} the correspondence between the states of the system and random: Its algorithmic complexity is of the order of of the environment superpositions that make up and its length (Li and Vitànyi, 1994): in Eq (7.2). Hence: |E⊙i |E⊗i K( n +, ) n (7.3) π hE | −i ≃ ∆( , )=1 ¯h . (7.6) By contrast, algorithmic complexity of the measurement |E⊙i |E⊗i h 2 · i outcomes in the 0 , 1 basis will be small: {| i | i} Given a fixed division of the environment into subsys- K( n 0, 1 ) n , (7.4) tems the action distance is a metric on the Hilbert space hE | i ≪ (Zurek, 1998a). That is; since the outcomes will be either 00 . . . 0 or 11 . . . 1. Ob- server seeking preferred states of the system by looking ∆( ψ , ψ )=0 , (7.7) at the environment should then search for the minimal | i | i record size and, thus, for the maximum redundancy in ∆( ψ , ϕ ) = ∆( ϕ , ψ ) 0, (7.8) the environmental record. States of the system that are | i | i | i | i ≥ recorded redundantly in the environment must have sur- and the triangle inequality: vived repeated instances of the environment monitoring, and are obviously robust and predictable. ∆( ψ , ϕ ) + ∆( ϕ , γ ) ∆( ψ , γ ) (7.9) Predictability we have utilized before to devise a | i | i | i | i ≥ | i | i “sieve” to select preferred states is used here again, but are all satisfied. 42

In defining ∆ it is essential to restrict rotations to the subsystems (i.e., photons, atoms, etc.). It is then use- subspaces of the subsystems of the whole Hilbert space, ful to define redundancy of the record by the number of and to insist that the unitary operations used in defin- times the information about the system has been copied, ing distance act on these subspaces. It is possible to or by how many times it can be independently extracted relax constraints on such unitary operations by allowing, from the environment. for example, pairwise or even more complex interactions In the simple example of Eq. (7.1) such redundancy between subsystems. Clearly, in absence of any restric- ratio for the , basis will be given by N, the tions the action required to rotate any ψ into any ϕ numberR of environment{| ↑i | ↓i} bits perfectly correlated with the π | i | i would be no more than 2 ¯h. Thus, constraints imposed obviously preferred basis of the system. More generally, by the natural division of the Hilbert space of the en- but in the same case of perfect correlation we obtain: vironment into subsystems play an essential role. Pre- ferred states of the system can be sought by extremizing lg(Dim ) = HE = logDim (Dim )= N (7.10) action distance between the corresponding record states R lg(Dim ) HS HE HS of the environment. In simple cases (e.g., see “bit-by- where is the Hilbert space of the environment per- byte”, Eq. (4.7) and below) the action distance criterion HE for preferred states coincides with the predictability sieve fectly correlated with the pointer states of the system. definition (Zurek, 1998a). On the other hand, with respect to the , basis, the redundancy ratio for ψ of Eq.{|⊙i (7.2)|⊗i} is only 1 (see also Zurek, 1983; 2000):| SE Redundancyi measures the∼ number of errors that can obliterate the difference 3. Redundancy and mutual information between two records, and in this basis one phase flip is clearly enough. This basis dependence of redundancy The most direct measure of reliability of the environ- suggests an alternative strategy to seek preferred states. ment as a witness is the information-theoretic redun- To define in general we can start with mutual in- R dancy of einselection itself. When environment monitors formation between the subsystems of the environment k the system (see Fig. 4), the information about its state and the system . As we have already seen in section IV,E will spread to more and more subsystems of the environ- definition of mutualS information in quantum mechanics ment. This can be represented by the state vector ψ , is not straightforward. The basis-independent formula: Eq. (7.1), with increasingly long sequences of 0’s| SE andi 1’s in the record states. The record size – the number k = ( : k)= H( )+ H( k) H( , k) (7.11) N of the subsystems of the environment involved – does I I S E S E − S E not affect the density matrix of the system . Yet, it is simple to evaluate (although it does have some strange obviously changes accessibility abd robustnessS of the in- features; see Eqs. (4.30) - (4.36)). In the present context formation analogues of the Darwinian “fitness”. As an it involves joint density matrix: illustration, let us consider c-shift’s. One subsystem ρ k = T r / ρ (7.12) of the environment (say, ) with the dimension of the k E1 SE E E SE Hilbert space no less than that of the system where the trace is carried out over all of the environment except for its singled out fragment k. In the example of Dim 1 Dim E HE ≥ HS Eq. (7.1) for any of the environment bits 2 2 suffices to eradicate off-diagonal elements of ρ in the ρ k = a 0 0 + b 1 1 . control basis. On the other hand, when N subsystemsS of SE | | | ↑ih↑ | | ih | | | | ↓ih↓ | | ih | the environment correlate with the same set of states of Given the partitioning of the environment into subsys- , the information about these states is simultaneously tems, the redundancy ratio can be defined as: accessibleS more widely. While ρ is no longer changing, S spreading of the information makes the existence of the = ( : k)/H( ) . (7.13) RI({⊗HEk }) I S E S pointer states of more objective – they are easier to Xk discover without beingS perturbed. When is maximized over all of the possible partitions, Information theoretic redundancy is defined as the R difference between the least number of bits needed to max = R E (7.14) uniquely specify the message and the actual size of the RI {⊗H k} encoded message. Extra bits allow for detection and cor- obtains. Roughly speaking, and in the case when the rection of errors (Cover and Thomas, 1991). In our case, number of the environment subsystems is large, max is the message is the state of the system, and the channel the total number of copies of the information aboutRI (the is the environment. The information about the system optimal basis of) that exist in . Maximal redundancy S E will often spread over all of the Hilbert space that is ratio max is of course basis - independent. HE RI enormous compared to . Redundancy of the record The information defined through the symmetric k, of the pointer observablesH ofS selected systems can be also Eq. (7.11), is in general inaccessible to observers whoI huge. Moreover, typical environments consist of obvious interrogate environment one subsystem at a time (Zurek, 43

2002a). It makes therefore a lot of sense to consider the 4. Redundancy ratio rate basis-dependent locally accessible information and define the corresponding redundancy ratio using: The rate of change of redundancy is of interest as an- RJ other measure of ‘fitness’, perhaps closest to the defini- = ( : )= H( )+ H( ) (H( )+ H( )). Jk J S Ek S Ek − S Ek|S tions of fitness used in modeling natural selection. Re- (7.15) dundancy can increase either as a result of interactions Conditional entropy must be computed in a specific basis between the system and the environment, or because the of the system (see Eq. (4.32)). All of the other steps that environment already correlated with is passing on the have led to the definition of max can be now repeated information to more distant environments.S In this second using . In the end, a basisR -I dependent: Jk case ‘genetic information’ is passed on by the ‘progeny’ of the original state. Even an observer consulting the ( s )= ( k) (7.16) RJ {| i} RJ ⊗HE environment becomes a part of such a more distant en- obtains. ( s ) quantifies the mutual information vironment. Redundancy rate is defined as: RJ {| i} between the collection of subsystems k of the environment and the basis s of the system.HE We note that d {| i} ˙ = (7.20) the condition of non-overlapping partitions guarantees R dtR that all of the corresponding measurements commute, and that the information can be indeed extracted inde- Either basis-dependent or basis-independent versions of ˙ may be of interest. pendently from each environment fragment k. R Preferred basis of can be now defined byE maximizing In general, it may not be easy to compute either or R ( s ) with respectS to the selection of s : ˙ exactly. This is nevertheless possible in models (such RJ {| i} {| i} asR those leading to Eqs. (7.1) - (7.2)). The simplest max = max ( s ) (7.17) illustrative example corresponds to the c-not model of RJ s ; E k RJ {| i} {| i} {⊗H } decoherence in Fig. 4. One can imagine that the con- This maximum can be sought either by varying the basis secutive record bits get correlated with the two branches of the system only (as it is indicated above) or by varying (corresponding to 0 and 1 in the “control”) at discrete both the basis and the partition of the environment. moments of time. | i(t) would| i be then the total number R It remains to be seen whether and under what circum- of c-nots that have acted over the time t, and ˙ is the stances pointer basis “stands out” through its definition number of new c-nots added per unit time. R in terms of . The criterion for a well - defined set of Redundancy rate measures information flow from the RJ pointer states p would be: system to the environment. Note that, after the first {| i} c-not in the example of Eqs. (7.1) - (7.2), will jump max = ( p ) ( s ) (7.18) RI RJ RJ {| i} ≫ RJ {| i} immediately from 0 to 2 bits, while the basis-specific RJ where s are typical superpositions of states belonging will increase from 0 to 1. In our model this initial dis- to different{| i} pointer eigenstates. crepancy (which reflects quantum discord, Eq. (4.36), This definition of preferred states directly employs the between and ) will disappear after the second c-not. I J notion of multiplicity of records available in the environ- Finally, we note that and, especially, ˙ can be used ment. Since , it follows that: to introduce new predictabilityR criteria: TheR states (or J ≤ I the observables) that are being recorded most redun- max max (7.19) RJ ≤ RI dantly are the obvious candidates for the “objective”, The important feature of either version of that makes and therefore for the “classical”. them useful for our purpose is their independenceR on H( ): The dependence on H( ) is in effect “normalized out”S of . characterizes theS “fan-out” of the informa- B. Observers and the Existential Interpretation tion aboutR R the preferred basis throughout environment, without a reference to what is known about the system. Von Neumann (1932), London and Bauer (1939) and The usual redundancy (in bits) is then ( ), al- Wigner (1963) have all appealed to the special role of though other implementations of this program∼ R·H (Ollivier,S the conscious observer. Consciousness was absolved from Poulin, and Zurek, 2002) employ different measures of following unitary evolution, and, thus, could collapse the redundancy, which may be even more specific than the wavepacket. Quantum formalism has led us to a differ- redundancy ratio we have described above. Indeed, what ent view, that nevertheless allows for a compatible con- is important here is the genreal idea of measuring clas- clusion. In essence, macroscopic systems are open, and sicality of quantum states through the number of copies their evolution is almost never unitary. Records main- they imprint throughout the Universe. This is a very tained by the observers are subject to einselection. In a Darwinian approach: We define classicality related to binary alphabet decoherence will allow only for the two einselection in ways reminiscent of ‘fitness’ in natural logical states, and prohibit their superpositions (Zurek, selection: States that spawn most of the (information- 1991). For human observers neurons conform to this theoretic) progeny are the most classical. binary convention and the decoherence times are short 44

(Tegmark, 2000). Thus, even if a cell of the observer en- thing effectively classical) is continuously monitored by tangles through a premeasurement with a pure quantum the rest of the Universe. Its state is repeatedly collapsed state, the record will become effectively classical almost – forced into the einselected states – and very well (very instantly: As a result, it will be impossible to “read it redundantly) ‘known’ to the rest of the Universe. off” in any basis except for the einselected one. This cen- The ‘higher functions’ of observers – e.g., conscious- sorship of records is the key difference between the exis- ness, etc. – may be at present poorly understood, but it tential interpretation and the original Everett’s MWI. is safe to assume that they reflect physical processes in Decoherence treats observer as any other macroscopic the information processing hardware of the brain. Hence, system. There is, however, one feature distinguishing ob- mental processes are in effect objective, as they leave an servers from the rest of the Universe: They are aware of indelible imprint on the environment: The observer has the content of their memory. Here we are using aware no chance of perceiving either his memory, or any other in a down - to - earth sense: Quite simply, observers macroscopic part of the Universe in some arbitrary su- know what they know. Their information processing ma- perposition. Moreover, memory capacity of observers is chinery (that must underlie higher functions of the mind miniscule compared to information content of the Uni- such as “consciousness”) can readily consult the content verse. So, while observers may know exact state of labo- of their memory. ratory systems, their records of the Universe will be very The information stored in the memory comes with fragmentary. By contrast, the Universe has enough mem- ‘strings attached’. The physical state of the observer is ory capacity to acquire and maintain detailed records of described in part by the data in his records. There is no states of macroscopic systems and their histories. information without representation. The information observer has could be, in principle, deduced from his physical state. Observer is – in part – information. Moreover, C. Events, Records, and Histories this information encoded in states of macroscopic quantum systems (neurons) is by no means secret: As a result Suppose that instead of a monotonous record sequence of lack of isolation the environment – having redundant in the environment basis corresponding to the pointer states of the system , implied by Eq. (7.1) the copies of the relevant data – ‘knows’ in detail everything {| ↑i | ↓i} observer knows. Configurations of neurons in our brains, observer looking at the environment detects: while at present undecipherable, are in principle as ob- 000...0111...1000...0111... jective and as widely accessible as the information about the states of other macroscopic objects. Given appropriate additional assumptions, such se- The observer is what he knows. In the unlikely case of quences consisting of long stretches of record 0’s and 1’s a flagrantly quantum input the physical state of the ob- justify inference of the history of the system. Let us fur- servers memory will decohere, resulting almost instantly ther assume that observer’s records come from intercept- in the einselected alternatives, each of them representing ing a small fragment of the environment. Other observers simultaneously both observer and his memory. The ‘ad- will be then able to consult their independently accessi- vertising’ of this state throughout the environment makes ble environmental records, and will infer (more or less) it effectively objective. the same history. Thus, in view of the “preponderance of An observer perceiving the Universe from within is in evidence” history defined as a sensible inference from the a very different position than an experimental physicist available records can be probed by many independently, studying a state vector of a quantum system. In a labora- and can be regarded as classical and objective. tory, Hilbert space of the investigated system is typically The redundancy ratio of the records is a measure of tiny. Such systems can be isolated, so that often the this objectivity. Note that this relativelyR objective exis- information loss to the environment can be prevented. tence (Zurek, 1998a) is an operational notion, quantified Then the evolution is unitary. The experimentalist can by the number of times the state of the system can be know everything there is to know about it. Common crit- found out independently, and not some “absolute objec- icisms of the approch advocated in this paper are based tivity”. However, and in a sense that can be rigorously on an unjustified extrapolation of the above laboratory defined, relative objectivity tends to absolute objectivity situation to the case of the observer who is a part of the in the limit . For example, cloning of unknown Universe. Critics of decoherence often note that the dif- states becomesR −→ possible ∞ (Bruss, Ekert, and Macchiavello, ferences between the laboratory example above and the 1999, Jozsa, 2002) in spite of the “no cloning” theorem case of the rest of the Universe are ‘merely quantitative’ (Wootters and Zurek, 1982, Dieks 1982). In that limit, – the system under investigation is bigger, etc. So why and given same reasonable constraints on the nature of cannot one analyze – they ask – interactions of the ob- the interactions and on the structure of the environment server and the rest of the Universe as before, for a small which underlie the definition of , it would take infinite isolated quantum system? resources such as action, Eqs.R (7.5) - (7.9), to hide or In the context of the existential interpretation the anal- subvert evidence of such an objective history. ogy with the laboratory is, in effect, turned “upside There are differences and parallels between relatively down”: For, now the observer (or the apparatus, or any- objective histories introduced here and consistent histo- 45 ries proposed by Griffiths (1984, 1996), and investigated consistent histories, one can discuss sequences of events by Gell-Mann and Hartle (1990; 1993; 1997), Omnès in an evolving quantum system without logical contra- (1988; 1992; 1994), Halliwell (1999), and others (Dowker dictions. The “golden middle” is advocated by Griffiths and Kent, 1996; Kiefer, 1996). Such histories are de- and Omnès (1999) who regard consistent histories as a fined as time-ordered sequences of projection operators convenient language, rather than as an explanation of 1 2 n Pα1 (t1), Pα2 (t2),...,Pαn (tn) and are abbreviated [Pα]. classicality. Consistency is achieved when they can be combined into The origin of the effective classicality can be traced coarse grained sets (where the projectors defining coarse to decoherence and einselection. As was noted by Gell- - grained set are given by sums of the projectors in the Mann and Hartle (1990), and elucidated by Omnès (1992; original set) while obeying probability sum rules: Proba- 1994) decoherence suffices to ensure approximate consis- bility of a bundle of histories should be a sum of the prob- tency. But consistency is both not enough and too much abilities of the constituent histories. The corresponding – it is too easy to accomplish, and does not necessarily condition can be expressed in terms of the decoherence lead to classicality (Dowker and Kent, 1996). What is functional (Gell-Mann and Hartle, 1990); needed instead is the objectivity of events and their time- ordered sequences – histories. As we have seen above, D([P ], [P ]) = T r (P n (t ) ...P 1 (t )ρP 1 (t ) ...P n (t ) α β αn n α1 1 β1 1 βn n both can appear as a result of einselection. (7.21) Above, the state of the system of interest is described by the density matrix ρ. Griffiths’ condition is equivalent to the vanishing of the real part of the expression 1. Relatively Objective Past above, Re D([Pα], [Pβ] = pαδα,β. As Gell-Mann and Hartle (1990){ emphasize,} it is more convenient – and in We have already provided an operational definition of the context of and emergent classicality more realistic – relatively objective existence of quantum states. It is to demand instead D([Pα], [Pβ]) = pαδα,β. Both weaker easy to apply it to events and histories: When many and stronger conditions for the consistency of histories observers can independently gather compatible evidence were considered (Goldstein and Page, 1995; Gell-Mann concerning an event, we call it relatively objective. Rel- and Hartle, 1997). The problem with all of them is atively objective history is then a time-ordered sequence that the resulting histories are very subjective: Given of relatively objective events. an initial density matrix of the Universe it is in general Monitoring of the system by the environment leads to quite easy to specify many different, mutually incompat- decoherence and einselection. It will also typically lead to ible consistent sets of histories. This subjectivity leads redundancy and hence effectively objective classical ex- to serious interpretational problems (d’Espagnat, 1989; istence in the sense of ‘quantum Darwinism’. Observers 1995; Dowker and Kent, 1996). Thus, a demand for ex- can independently access redundant records of events and act consistency as one of the conditions for classicality is histories imprinted in the environmental degrees of free- both uncomfortable (overly restrictive) and insufficient dom. The number of observers who can examine evidence (as the resulting histories are very non-classical). More- etched in the environment can be of the order of, and is over, coarse-grainings that help secure approximate con- bounded from above by . Redundancy is a measure RJ sistency have to be, in effect, guessed. of this objectivity and classicality. The attitudes adopted by the practitioners of the con- As observers record their data, changes: Consider RJ sistent histories approach in view of its unsuitability for an observer who measures the ‘right observable’ of (i.e., E the role of the cornerstone of the emergent classicality the one with the eigenstates 0 , 1 in the example of Eq. | i | i differ. Initially – before difficulties became apparent – it (7.1)). Then his records and – as his records decohere, was hoped that such approach would answer all of the their environment – become a part of evidence, and are interpretational questions, perhaps when supplemented correlated with the preferred basis of the system. Conse- by a subsidiary condition, i.e. some assumption about quently, computed from Eq. (7.14) increases. Every RJ favored coarse-grainings. At present, some still aspire interaction that increases the number of the records also to the original goals of deriving classicality from consis- increases . This is obvious for the “primary” interac- RJ tency alone. Others may uphold the original aims of the tions with the system, but it is also true for the secondary, program, but they also generally rely on environment- tertiary, etc. acts of replication of the information ob- induced decoherence, using in calculations variants of tained from the observers who recorded primary state of models we have presented in this paper. This strategy the system, from the environment, from the environment has been quite successful – after all, decoherence leads of the environment, and so on. to consistency. For instance, special role of the hydrody- A measurement reveals to the observer “his” branch of namic observables (Gell-Mann and Hartle, 1990; Dowker the universal state vector. The correlations established and Halliwell, 1992; Halliwell, 1999; Brun and Hartle, alter observer’s state – his records – and “attach” him 1999) can be traced to their predictability, or to their to this branch. He will share it with other observers approximate commutativity with the total Hamiltonian who examined the same set of observables, and who have (see Eq. (4.41)). On the other hand, the original goals recorded compatible results. of Griffiths (1984, 1996) have been more modest: Using It is also possible to imagine a stubborn observer who 46 insists on measuring either the relative phase between berg uncertainty into the observable complementary to the two obvious branches of the environment in Eq. the one that is monitored. One can simulate such uncer- (7.2), or the state of the environment in the Hadamard- tainty without any monitoring environment by introduc- transformed basis + , . In either case the distinc- ing classical noise. In each specific run of the experiment tion between the two{| outcomesi |−i} could determine the state – for each realization of time-dependent noise – quantum of the spin in the , basis. However, in that ba- system will evolve deterministically. However, after av- sis = 1. Hence,{|⊙i while|⊗i} in principle these measure- eraging over different noise realizations evolution of the mentsRJ can be carried out and yield the correct result, density matrix describing an ensemble of systems may the information concerning , basis is not redun- approximate decoherence due to an entangling quantum dant, and, therefore, not objective:{|⊙i |⊗i} Only one stubborn environment. In particular, the master equation may be observer can access it directly. As a result will de- essentially the same as for true decoherence, although the crease. Whether the ( , ) will becomeRJ larger interpretational implications are more limited. Yet, using than ( , ) wasRJ before{|⊙i |⊗i} the measurement of the such strategies one can simulate much of the dynamics stubbornRJ observer{| ↑i | ↓i} will depend on detailed comparison of of open quantum systems. the initial redundancy with the amplification involved, The strategy of simulating decoherence can be taken decoherence, etc. further: Not just the effect of the environment, but also There is a further significant difference between the two the dynamics of the quantum system can be simulated by stubborn observers considered above. When the observer classical means. This can be accomplished when classical measures the phase between the two sequences of 0’s and wave phenomena follow equations of motion related to 1’s in Eq. (7.2), correlations between the bits of the Schrödinger equation. We shall discuss experiments that environment remain. Thus, even after his measurement fall into all of the above categories. one could find relatively objective evidence of the past Last not least, while decoherence – through einselec- event – past state of the spin – and, in more complicated tion – helps solve the measurement problem, it is also a cases, of the history. On the other hand, measurement major obstacle to quantum information processing. We of all the environment bits in the + , basis will shall thus end this section briefly describing strategies {| i |−i} obliterate evidence of such a past. that may allow one to tame decoherence. Relatively objective existence of events is the strongest condition we have considered here. It is a consequence of the existence of multiple records of the same set of A. Decoherence due to entangling interactions states of the system. It allows for such manifestations of classicality as unimpeded cloning. It implies einselec- Several experiments fit this category, and more have tion of states most closely monitored by the environment. been proposed. Decoherence due to emission or scatter- Decoherence is clearly weaker and easier to accomplish. ing of photons has been investigated by the MIT group “The past exists only insofar as it is recorded in the (Chapman et al., 1995; Kokorowski et al., 2001) using present” (dictum often repeated by Wheeler) may the atomic interferometry. Emission or scattering deposits best summary of the above discussion. Relatively ob- a record in the environment. It can store information jective reality of few selected observables in our familiar about the path of the atom providing photon wavelength Universe is measured by their fitness – by the redundancy is shorter than the separation between the two of the with which they are recorded in the environment. This atoms. In case of emission this record is not redundant, multiplicity of available copies of the same information as the atom and photon are simply entangled – 1 can be regarded as a consequence of amplification, and in any basis. Scattering may involve more photons,RJ and∼ as a cause of indelibility. Multiple records safeguard ob- the recent careful experiment has confirmed the satura- jectivity of our past. tion of decoherence rate at distances in excess of photon wavelength (Gallis and Fleming, 1990; Anglin, Paz, and Zurek, 1997). VIII. DECOHERENCE IN THE LABORATORY There is an intimate connection between interference and complementarity in the two-slit experiment on one The biggest obstacle in the experimental study of de- hand, and the entanglement on the other (Wootters and coherence is – paradoxically – its effectiveness. In the Zurek, 1979). Consequently, appropriate measurements macroscopic domain only the einselected states survive. of the photon allow one to restore interference fringes Their superpositions are next to impossible to prepare. in the conditional subensembles corresponding to a def- In the mesoscopic regime one may hope to adjust the inite phase between the two photon trajectories (see es- size of the system, and, thus, interpolate between quan- pecially Chapman et al. (1995), as well as Kwiat, Stein- tum and classical. The strength of the coupling to the berg and Chiao (1993); Pfau et al., 1994; Herzog, Kwiat, environment is the other parameter one may employ to Weinfurter and Zeilinger, (1995) for implementations of control the decoherence rate. this “quantum erasure” trick due to Hillery and Scully One of the key consequences of monitoring by the en- (1983)). Similar experiments have been also carried out vironment is the inevitable introduction of the Heisen- using neutron interferometry (see e.g. Rauch, (1998)). 47

In all of these experiments one is dealing with a very quality factor Q, and from the data about the coherent simplified situation involving a single microsystem and state initially present in the cavity. Decoherence rate is a single “unit” of decoherence ( 1) caused by a a function of separation of the two components of the cat single quantum of the environment.RJ ∼ Experiments on a . Experimental results agree with predictions. |⊎i mesoscopic system monitored by the environment are The discussion above depends on the special role obviously much harder to devise. Nevertheless, Sergé of coherent states. Coherent states are einselected in Haroche, Jean-Michel Raimond, Michel Brune and their harmonic oscillators, and, hence, in the underdamped colleagues (Brune et al., 1996; Haroche 1998, Raimond, bosonic fields (Anglin and Zurek, 1996). Thus, they are Brune, and Haroche, 2001) have carried out a spectac- the pointer states of the cavity. Their special role is rec- ular experiment of this type, yielding solid evidence in ognized implicitly above: If number eigenstates were ein- support of the basic tenets of the environment-induced selected, predictions would be obviously quite different. transition from quantum to classical. Their system is a Therefore, while the ENS experiment is focused on deco- microwave cavity. It starts in a coherent state with an herence rate, confirmation of the predicted special role of amplitude corresponding to a few photons. coherent states in bosonic fields is its important (albeit “Schrödinger cat” is created by introducing an atom implicit) corollary. in a superposition of two Rydberg states, + = 0 + 1 : The atom passing through the cavity puts| itsi refractive| i | i index in a superposition of two values. Hence, the phase B. Simulating decoherence with classical noise of the coherent state shifts by the amount correlated with the state of the atom, creating an entangled state: ¿From the fundamental point of view, the distinction ( 0 + 1 ) = 0 + 1 = ϑ (8.1) between cases when decoherence is caused by entangling | →i | i | i ⇒ | րi| i | ցi| i | i interactions with the quantum state of the environment Arrows indicate relative phase space locations of coherent and when it is simulated by a classical noise in the observ- states. States of the atom are 0 and 1 . “Schrödinger able complementary to the “pointer” is essential. How- kitten” is prepared from this entangled| i | statei by measur- ever, from the engineering point of view (adopted, e.g., ing the atom in the + , basis: by the practitioners of quantum computation, see Nielsen {| i |−i} and Chuang, 2000 for discussion) this may not matter. ϑ = ( + ) + + ( ) (6.2) | i | րi | ցi | i | րi − | ցi |−i For instance, quantum error correction techniques (Shor, 1995; Steane, 1996; Preskill, 1999) are capable of deal- Thus, atom in the state + implies preparation of a “pos- ing with either. Moreover, experimental investigations of itive cat” = +| i in the cavity. Such super- this subject often involve both. positions of|⊎i coherent| րi states| ցi could survive forever if there was no decoherence. However, radiation leaks out of the The classic experiment in this category was carried out cavity. Hence, environment acquires information about recently by David Wineland, Chris Monroe, and their the state inside. Consequences are tested by passing an- collaborators (Myatt et al, 2000; Turchette et al., 2000). other atom in the state + = 0 + 1 through the cavity. They use ion trap to study behavior of individual ions in In absence of decoherence| i the| statei | wouldi evolve as: a “Schrödinger cat” state (Monroe et al., 1996) under the influence of injected classical noise. They also embark on + = ( + )( 0 + 1 )= a preliminary study of “environment engineering”. |⊎i| i | րi | ցi | i | i ⇒ ( 0 + 1 ) + ( 0 + 1 ) Superpositions of two coherent states as well as of num- | ↑i| i | →i| i | →i| i | ↓i| i ber eigenstates were subjected to simulated high tem- = ( 0 + 1 )+ √2 + . (61) | ↑i| i | ↓i| i | →i| i perature amplitude and phase “reservoirs”. This was Above we have omitted the overall normalization, but done through time-dependent modulation of the self- retained the (essential) relative amplitude. Hamiltonian of the system. For the amplitude noise these For the above state detection of + in the first are in effect random fluctuations of the location of the (preparatory) atom implies the conditional| i probability minimum of the harmonic trap. Phase noise corresponds 3 to random fluctuations of the trap frequency. of detection of + , p+ + = 4 , for the second (test) atom. Decoherence will| i suppress| off-diagonal terms of the den- In either case, the resulting loss of coherence is well sity matrix, so that, some time after the preparation, described by the exponential decay with time, with the exponent that scales with the square of the separation be- ρcavity that starts, say, as becomes: |⊎ih⊎| tween the two components of the macroscopic quantum ρ = ( + )/2 superposition (e.g., Eq. (5.34)). The case of the ampli- cavity | րihր | | ցihց | + z( + )/2 . (62) tude noise approximates decoherence in quantum Brow- | րihց | | ցihր | nian motion in that the coordinate is monitored by the 1 When z = 0 the conditional probability is p(+ +) = 2 . environment, and, hence, the momentum is perturbed. In the intermediate cases intermediate values| of this (Note that in the underdamped harmonic oscillator ro- and other relevant conditional probabilities are predicted. tating wave approximation blurs the distinction between The rate of decoherence, and, consequently, the time- x and p, leading to einselection of coherent states.) The dependent value of z can be estimated from the cavity phase noise would arise in an environment monitoring the 48 number operator, thus leading to uncertainty in phase. success (Moore et al., 1994). More recently, attention of Consequently, number eigenstates are einselected. both Raizen and his group in Texas as well as of Chris- The applied noise is classical, and the environment tensen and his group in New Zealand has shifted towards does not acquire any information about the ion ( = 0). the effect of decoherence on quantum chaotic evolution Thus, following a particular realization of the noiseRI the (Ammann et al, 1998; Klappauf et al., 1998). state of the system is still pure. Nevertheless, an ensem- In all of the above studies the state of the chaotic ble average over many noise realizations is represented system (δ-kicked rotor) was perturbed by spontaneous by the density matrix that follows an appropriate mas- emission from the trapped atoms, that was induced by ter equation. Thus, as Wineland, Monroe, and their col- decreasing detuning of the lasers used to set up the opti- leagues note, decoherence simulated by the classical noise cal lattice. In addition, noise was occasionally introduced could be in each individual case – for each realisation into the potential. Both groups find that, as a result of – reversed by simply measuring the corresponding time- spontaneous emission, localization disappears, although dependent noise run either beforehand or afterwards, and the two studies differ in some of the details. More ex- then applying the appropriate unitary transformation to periments, including some that allow gentler forms of the state of the system. By contrast, in the case of en- monitoring by the environment (rather than spontaneous tangling interactions, two measurements – one preparing emission noise) appear to be within reach. the environment before the interaction with the environ- In all of the above cases one deals, in effect, with a large ment, the other following it – would be the least required ensemble of identical atoms. While each atom suffers for a chance of undoing the effect of decoherence. repeated disruptions of its evolution due to spontaneous The same two papers study decay of a superposition emission, the ensemble evolves smoothly and in accord of number eigenstates 0 and 2 due to an indirect cou- with the appropriate master equation. The situation is pling with the vacuum.| i This| proceedsi through entan- reminiscent of the “decoherence simulated by noise”. glement with the first order environment (that, in effect, Experiments that probe the effect of classical noise consists of the other states of the harmonic oscillator) on chaotic systems have been carried out earlier (Koch, and a slower transfer of information to the distant en- 1995). They were, however, analyzed from the point of vironment. Dynamics involving the system and its first view that does not readily shed light on decoherence. - order environment leads to non-monotonic behavior of A novel experimental approach to decoherence and to the off-diagonal terms representing coherence. Further irreversibility in open complex quantum systems is pur- studies of decoherence in the ion trap setting are likely sued by Levstein, Pastawski, and their colleagues (Lev- to follow, as this is an attractive implementation of the stein, Usaj, and Pastawski, 1998; Levstein et al., 2000). quantum computer (Cirac and Zoller, 1995). Using NMR techniques the investigate reversibility of dynamics by implementing a version of spin echo. This promising “Loschmidt echo” approach has led to renewed 1. Decoherence, Noise, and Quantum Chaos interest in the issues that touch on quantum chaos, decoherence, and related subjects (see e.g., Gorin and Selig- Following a proposal of Graham, Schlautmann, and man, 2001; Prosen, 2001; Jalabert and Pastawski, 2001; Zoller (1993) Mark Raizen and his group (Moore et al., Jacquod, Silvestrov and Beenakker, 2001). 1994) used a one-dimensional optical lattice to implement a variety of 1-D chaotic systems including the “standard map”. Various aspects of the behavior expected from C. Analogue of decoherence in a classical system a quantized version of a classically chaotic system were subsequently found, including, in particular, dynamical Both the system and the environment are effectively localization (Casati and Chirikov, 1995a; Reichl, 1992). classical in the last category of experiments, represented Dynamical localization establishes, in a class of driven by the work of Cheng and Raymer (1999). They have in- quantum chaotic systems, a saturation of momentum dis- vestigated behavior of transverse spatial coherence dur- persion, and leads to a characteristic exponential form ing the propagation of the optical beam through a dense, of its distribution (Casati and Chirikov, 1995a). Local- random dielectric medium. This problem can be mod- ization is obviously a challenge to the quantum-classical elled by a Boltzmann-like transport equation for the correspondence, since in these very same systems clas- Wigner function of the wave field, and exhibits charac- sical prediction has the momentum dispersion growing teristic increase of decoherence rate with the square of unbounded, more or less with the square root of time. the spatial separation, followed by a saturation at suf- α However, it sets in after tL ¯h− , where α 1 (rather ficiently large distances. This saturation contrasts with ∼ 1 ∼ than on the much shorter th¯ lnh ¯− we have discussed the simple models of decoherence in quantum Brownian ∼ in Section III) so it can be ignored for macroscopic sys- motion that are based on dipole approximation. How- tems. On the other hand, its signature is easy to detect. ever, it is in good accord with the more sophisticated Demonstration of dynamical Anderson localization in discussions which recognize that, for separations of the the optical lattice implementation of the δ-kicked rotor order of the prevalent wavelength in the environment, and related studies of quantum chaos are a significant dipole approximation fails and other more complicated 49 behaviors can set in (Gallis and Fleming, 1990; Anglin, in conception with the pointer subspaces introduced some Paz and Zurek, 1997; Paz and Zurek, 1999). Similar re- time ago (Zurek, 1982), and satisfy (exactly or approxi- sult in a completely quantum case was obtained using mately) the same Eqs. (4.22) and (4.41) or their equiva- atomic interferometry by Kokorowski et al. (2001). lents (given, e.g., in terms of “Krauss operators” (Krauss, 1983)) that represent non-unitary consequences of the interaction with the environment in the Lindblad (1976) D. Taming decoherence form of the master equation). DFS’s were (re)discovered in the context of quantum information processing. They In many of the applications of quantum mechanics the appear as a consequence of an exact or approximate sym- quantum nature of the information stored or processed metries of the Hamiltonians that govern the evolution of needs to be protected. Thus, decoherence is an enemy. the system and its interaction with the environment (Za- Quantum computation is an example of this situation. nardi and Rasetti, 1997; Duan and Guo, 1998; Lidar et Quantum computer can be thought of as a sophisticated al., 1998; Zanardi, 1998; 2000). interference device that works by performing in parallel An active extension of this approach aimed at finding a coherent superposition of a multitude of classical com- quiet corners of the Hilbert space is known as dynamical putations. Loss of coherence would disrupt this quantum decoupling. There the effectively decoupled subspaces parallelism essential for the expected speedup. are induced by time-dependent modifications of the evo- In the absence of the ideal – completely isolated abso- lution of the system deliberately introduced from the out- lutely perfect quantum computer, something easy for a side by time-dependent evolution and / or measurements theorist to imagine but impossible to attain in the labora- (see e.g. Viola and Lloyd, 1998; Zanardi, 2000). A fur- tory – one must deal with imperfect hardware “leaking” ther generalization and unification of various techniques some of its information to the environment. And main- leads to the concept of noiseless quantum subsystems taining isolation while simultaneously achieving a reason- (Knill, Laflamme, and Viola, 1999; Zanardi, 2000), which able “clock time” for the quantum computer is likely to may be regarded as a non-abelian (and quite non-trivial) be difficult since both are in general controlled by the generalization of pointer subspaces. same interaction (although there are exceptions – for ex- A sophisticated and elegant strategy that can be re- ample, in the ion trap proposal of Cirac and Zoller (1995) garded as a version of the decoherence free approach was interaction is in a sense “on demand”, and is turned on devised independently by Kitaev (1997a, b). He has ad- by the laser coupling internal states of ions with the vi- vocated using states that are topologically stable, and, brational degree of freedom of the ion chain). thus, that can successfully resist arbitrary interactions The need for error correction in quantum computation with the environment. The focus here (in contrast to was realized early on (Zurek, 1984b) but methods for much of the DFS work) is on devising a system with self- accomplishing this goal have evolved dramatically from Hamiltonian that – as a consequence of the structure of the Zeno effect suggested then to very sophisticated (and the gap its energy spectrum related to the “cost” of topo- much more effective) strategies in the recent years. This logically non-trivial excitations – acquires a subspace de is fortunate: Without error correction even fairly modest facto isolated from the environment. This approach has quantum computations (such as factoring number 15 in been further developed by Bravyi and Kitaev (1998) and an ion trap with imperfect control of the duration of laser by Freedman and Meyer (1998). pulses) go astray rapidly as a consequence of relatively small imperfections (Miquel, Paz, and Zurek, 1997). Three different, somewhat overlapping approaches 2. Environment engineering that aim to control and tame decoherence, or to correct errors caused by decoherence or by the other imperfec- This strategy involves altering the (effective) interactions of the hardware have been proposed. We summarize tion Hamiltonian between the system and the environ- them very briefly, spelling out main ideas and pointing ment or influencing the state of the environment to se- out references that discuss them in greater detail. lectively suppress decoherence. There are many ways to implement it, and we shall describe under this label variety of proposed techniques (some of which are not all that 1. Pointer states and noiseless subsystems different from the strategies we have just discussed) that aim to protect quantum information stored in selected The most straightforward strategy to suppress deco- subspaces of the Hilbert space of the system, or even to herence is to isolate the system of interest (e.g., quantum exploit pointer states induced or redefined in this fashion. computer). Failing that, one may try to isolate some of The basic question that started this line of research its observables with degenerate pointer subspaces, which – can one influence the choice of the preferred pointer then constitute niches in the Hilbert space of the infor- states? – arose in the context of ion trap quantum com- mation processing system that do not get disrupted in puter proposed by Cirac and Zoller (1995). The answer spite of its the coupling to the environment. Decoher- given by the theory is, of course, that the choice of the ence free subspaces (DFS’s for brevity) are thus identical einselected basis is predicated on the details of the situa- 50 tion, and, in particular, on the nature of the interaction reasons. To begin with, quantum states – and, hence, between the system and the environment (Zurek, 1981; quantum information – cannot be “cloned” (Dieks, 1982; 1982; 1993a). Yet, Poyatos, Cirac and Zoller (1996) have Wootters and Zurek, 1982). Moreover, quantum infor- suggested a scheme suitable for implementation in an ion mation is very “private”, and a measurement that is in- trap, where interaction with the environment – and, in volved in majority voting would infringe on this privacy accord with Eq. (4.41), the pointer basis itself – can be and destroy quantum coherence, making quantum infor- adjusted. The key idea is to recognize that the effective mation classical. Fortunately, both of these difficulties coupling between the vibrational degrees of freedom of can be simultaneously overcome by encoding quantum an ion (the system) and the laser light (which plays the information in entangled states of several qubits. Cloning role of the environment) is given by: turns out not to be necessary. And measurements can be carried out in a way that identifies errors while keeping

Ω iωLt iωLt + quantum information untouched. Moreover, error cor- Hint = (σ+e− +σ e ) sin[κ(a+a )+φ] . (8.5) 2 − rection is discrete – measurements that reveal error syn- dromes have “yes - no” outcomes. Thus, even though Above, Ω is the Rabi frequency, ωL the laser frequency, φ information stored in a qubit represents a continuum of is related to the relative position of the center of the trap possible quantum states (e.g., corresponding to a surface with respect to the laser standing wave, κ is the Lamb- of the Bloch sphere) error correction is discrete, allay- Dicke parameter of the transition, while σ (σ+) and + − ing one of the earliest worries concerning the feasibil- a (a ) are the annihilation (creation) operators of the ity of quantum computation – the unchecked “drift” of atomic transition and of the harmonic oscillator (trap). the quantum state representing information (Landauer, By adjusting φ and ωL and adopting appropriate set of 1995). approximations (that include elimination of the internal degrees of freedom of the atom) one is led to the master This strategy (discovered by Shor (1995) and Steane equation for the system – i.e, for the density matrix of (1996)) has been since investigated by many (Ekert and the vibrational degree of freedom: Macchiavello, 1996; Bennett et al., 1996; Laflamme et al., 1996) and codified into a mathematically appeal- + ρ˙ = γ(2fρf f+fρ ρf+f) (8.6) ing formalism (Gottesman, 1996; Knill and Laflamme, − − 1997). Moreover, first examples of successful implemen- Above, f is the operator with a form that depends on tation (see e.g. Cory et al., 1999) are already at hand. the adjustable parameters φ and ωL in Hint, while γ is a constant that depends also on Ω and η. As Poyatos et al. Error correction allows one, at least in principle, to compute forever, providing that the errors are suitably show, one can alter the effective interaction between the 4 small ( 10− per computational step seems to be the slow degree of freedom (the oscillator) and the environ- ∼ ment (laser light) by adjusting parameters of the actual error probability threshold sufficient for most error cor- H . rection schemes). Strategies that accomplish this encode int qubits in already encoded qubits (Aharonov and Ben-Or First steps towards realization of this “environment 1997; Kitaev 1997c; Knill, Laflamme and Zurek, 1996; engineering” proposals were taken by the NIST group 1998a, b; Preskill 1998). The number of layers of such (Myatt et al., 2000; Turchette et al., 2000). Similar tech- concatenation necessary to achieve fault tolerance – the niques can be employed to protect deliberately selected ability to carry out arbitrarily long computations – de- states from decoherence (Carvalho et al., 2000). pends on the size (and the character) of the errors, and on Other ideas aimed at channeling decoherence have the duration of the computation, but when error proba- been also explored in contexts that range from quan- bility is smaller than the threshold, that number of layers tum information processing (Beige et al., 2000) to preser- is finite. Overviews of fault tolerant computation are al- vation of Schrödinger cats in Bose-Einstein condensates ready at hand (Preskill, 1999; Nielsen and Chuang, 2000, (Dalvit, Dziarmaga, and Zurek, 2000). and references therein). An interesting subject related to the above discussion 3. Error correction and resilient computing is quantum process tomography, anticipated by Jones (1994), and described in the context of quantum infor- This strategy is perhaps most sophisticated and com- mation processing by Chuang and Nielsen (1997) and by prehensive, and capable of dealing with the greatest va- Poyatos, Cirac and Zoller (1997). The aim here is to char- riety of errors in a most hardware - independent manner. acterize completely a process – such as a quantum logical It is a direct descendant of the error correction techniques gate – and not just a state. First deliberate implemen- employed in dealing with the classical information, based tation of this procedure (Nielsen, Knill, and Laflamme, on redundancy: Multiple copies of the information are 1998) has also demonstrated experimentally that einse- made, and the errors are found and corrected by sophis- lection is indeed equivalent to an unread measurement ticated “majority voting’ techniques. of the pointer basis by the environment, and can be re- One might have thought that implementing error cor- garded as such from the standpoint of applications (i.e., rection in the quantum setting will be difficult for two NMR teleportation in the example above). 51

IX. CONCLUDING REMARKS Ehrenfest theorem, some version of the Copenhagen interpretation, and other evidence that quantum theory Decoherence, einselection, pointer states, and even pre- is really not all that different from classical – especially dictability sieve have become familiar concepts to many when systems of interest become macroscopic, and all one in the past decade. The first goal of this paper was to cares about are averages – is presented. review these advances and to survey, and – where pos- The message seems to be that the there is really no sible, to address – the remaining difficulties. The sec- problem and that quantum mechanics can be “tamed” ond related aim was to “preview” the future develop- and confined to the microscopic domain. Indeterminacy ments. This has led to considerations involving informa- and double slit experiment are of course discussed, but tion, as well as to the operational, physically motivated to prove peaceful coexistence within the elbow room as- discussions of seemingly esoteric concepts such as objec- sured by the Heisenberg’s principle and complementar- tivity. Some of the material presented (including the ity. Entanglement is rarely explored. This is quite con- ‘Darwinian’ view of the emergence of objectivity through sistent with the aim of the introductory quantum me- redundancy, as well as the discussion of envariance and chanics courses, which has been (only slightly unfairly) probabilities) are rather new, and a subject of research, summed up by the memorable phrase “shut up and cal- hence the word “preview” applies here. culate”. Discussion of measurement is either dealt with New paradigms often take a long time to gain ground. through models based on the CI “old orthodoxy”, or not Atomic theory of matter (which, until early XX century, at all. An implicit (and sometime explicit) message is: was ‘just an interpretation’) is the case in point. Some of Those who ask questions that do not lend themselves the most tangible applications and consequences of new to an answer through laborious, preferably perturbative ideas are difficult to recognize immediately. In the case calculation are “philosophers”, and should be avoided. of atomic theory, Brownian motion is a good example: The above description is of course a caricature. But Even when the evidence is out there, it is often difficult given that the calculational techniques of quantum the- to decode its significance. ory needed in atomic, nuclear, particle, or condensed Decoherence and einselection are no exception. They matter physics are indeed difficult to master, and given have been investigated for about two decades. They are that – to date – most of the applications had nothing to the only explanation of classicality that does not require do with the nature of quantum states, entanglement, and modifications of quantum theory, as do the alternatives such, the attitude of avoiding the most flagrantly quan- (Bohm, 1952; Leggett, 1980, 1988, 2002; Penrose, 1986, tum aspects of quantum theory is easy to understand. 1989; Holland, 1993; Goldstein, 1998; Pearle, 1976; 1993; Novel applications force one to consider questions Ghirardi, Rimini, and Weber, 1986; 1987, Gisin and Per- about the information content, the nature of “the quan- cival, 1992; 1993a-c). Ideas based on the immersion of the tum”, and the emergence of the classical much more di- system in the environment have recently gained enough rectly, with a focus on states and correlations, rather than support to be described (by sceptics!) as “the new ortho- on the spectra, cross sections and the expectation values. doxy” (Bub, 1997). This is a dangerous characterization, Hence, problems that are usually bypassed will come to as it suggests that the interpretation based on the recog- the fore: It is hard to brand Schrödinger cats and entan- nition of the role of the environment is both complete glement as “exotic” and make them a centerpiece of a and widely accepted. Neither is certainly the case. marketable device. I believe that as a result decoherence Many conceptual and technical issues (such as what will become a part of the textbook lore. Indeed, at the constitutes ‘a system’) are still open. As for the breadth graduate level there are already some notable exceptions of acceptance, “the new orthodoxy” seems to be an op- among monographs (Peres, 1993) and specialized texts timistic (mis-)characterization of decoherence and einse- (Walls and Milburn, 1994; Nielsen and Chuang, 2000). lection, especially since this explanation of the transition Moreover, the range of subjects already influenced by from quantum to classical has (with very few exceptions) decoherence and einselection – by the ideas originally mo- not made it to the textbooks. This is intriguing, and may tivated by the quantum theory of measurements – is be- be as much a comment on the way in which quantum ginning to extend way beyond its original domain. In ad- physics has been taught, especially on the undergradu- dition to the atomic physics, quantum optics, and quan- ate level, as on the status of the theory we have reviewed tum information processing (which were all mentioned and its level of acceptance among the physicists. throughout this review) it stretches from material sci- Quantum mechanics has been to date, by and large, ences (Karlsson, 1998; Dreismann, 2000), surface science, presented in a manner that reflects its historical devel- where it seems to be an essential ingredient explaining opment. That is, Bohr’s planetary model of atom is still emission of electrons (Brodie, 1995; Durakiewicz et al., often the point of departure, Hamilton-Jacobi equations 2001) through heavy ion collisions (Krzywicki, 1993) to are used to “derive” Schrödinger equation, and an over- quantum gravity and cosmology (Zeh, 1986, 1988, 1992; simplified version of the quantum - classical relationship Kiefer, 1987; Kiefer and Zeh, 1995; Halliwell, 1989; Bran- (attributed to Bohr, but generally not doing justice to his denberger, Laflamme and Mijic, 1990; Barvinsky and much more sophisticated views) with the correspondence Kamenshchik, 1990, 1995; Paz and Sinha, 1991, 1992; principle, kinship of commutators and Poisson brackets, Castagnino et al., 1993, Mensky and Novikov, 1996). 52

Given the limitations of space we have not done justice X. ACKNOWLEDGMENTS to most of these subjects, focusing instead on issues of principle. In some areas reviews already exist. Thus, John Archibald Wheeler has – quarter century ago Giulini et al. (1996) is a valuable collection of essays, – taught a course on the subject of quantum mea- where, for example, decoherence in field theories is ad- surements at the University of Texas in Austin. The dressed. Dissertation of Wallace (2002) offers a good (if questions raised then have since evolved into ideas pre- somewhat philosophical) summary of the role of deco- sented here, partly in collaboration with Juan Pablo herence with a rathe different emphasis on similar field- Paz, and through interactions with many colleagues, in- theoretic issues. Conference proceedings edited by Blan- cluding Andreas Albrecht, Jim Anglin, Charles Bennett, chard et al. (2000) and, especially, an extensive historical Robin Blume-Kohout, Carlton Caves, Ike Chuang, Diego overview of the foundation of quantum theory from the Dalvit, David Deutsch, David Divincenzo, Jacek Dziar- modern perspective by Auletta (2000) are also recom- maga, Richard Feynman, Murray Gell-Mann, Daniel mended. More specific technical issues with implications Gottesmann, Robert Griffiths, Salman Habib, Jonathan for decoherence and einselection have also been reviewed. Halliwell, Serge Haroche, James Hartle, Chris Jarzynski, For example, on the subject of master equations there Erich Joos, Manny Knill, Raymond Laflamme, Anthony are several reviews with very different emphasis including Leggett, Seth Lloyd, Gerard Milburn, Michael Nielsen, Alicki and Lendi (1987), Grabert, Schramm, and Ingold Harold Ollivier, Asher Peres, David Poulin, Rüdiger (1988), Namiki and Pascazio (1993), as well as – more Schack, Ben Schumacher, Kosuke Shizume, Bill Unruh, recently – Paz and Zurek (2001). In some areas – such as David Wallace, Eugene Wigner, Bill Wootters, Dieter atomic BEC’s – the study of decoherence has only started Zeh, Anton Zeilenger, and Peter Zoller. Moreover, Serge (Anglin, 1997; Dalvit, Dziarmaga, and Zurek, 2001). In Haroche, Mike Nielsen, Harold Ollivier, Juan Pablo Paz many situations (e.g, quantum optics) a useful supple- and David Wallace have provoided me with extensive ment to the decoherence view of the quantum - classical written comments on earlier versions of the manuscript. interface is afforded by ‘quantum trajectories’ – a study Its preparation was assisted by authors’ participation in of the state of the system inferred from the intercepted two ITP programs on decoherence - related subjects, and state of the environment (see Carmichael, 1993; Wise- was in part supported by a grant from the NSA. Last not man and Milburn, 1993; Gisin and Percival, 1993a-c). least, this paper has evolved in course of over a dozen This approach “unravels” evolving density matrices of years, along with the field, under a watchful eye of a open systems into trajectories conditioned upon the mea- sequence of increasingly impatient editors of RMP, their surement carried out on the environment, and may have feelings shared by my family. Perseverance of all afflicted – especially in quantum optics – intriguing connections was very much appreciated by the author (if thoroughly with the “environment as a witness” point of view (see tested by the process). Dalvit, Dziarmaga, and Zurek, 2001). In other areas – such condensed matter – decoherence phenomena have so many variations and are so pervasive that a separate “de- coherent review” may be in order, especially as intriguing experimental puzzles seem to challenge the theory (Mo- References hanty and Webb, 1997; Kravtsov and Altshuler, 1999). [4] Adler, S. L., 2001, quant-ph/0112095. [4] Aharonov, D, and M. Ben-Or, in Proceedings of the 29th Annual ACM Symposium on the Theory of Computing, pp. 176-188. [4] Aharonov, Y, J. Anandan, and L. Vaidman, 1993, Phys. Rev. A47, 4616. [4] Aharonov, Y, B. Resnik, 2002, Phys. Rev. A65, 052116. Physics of information and computation is a special [332] Albrecht, A., 1992, Phys. Rev. D46, 5504–5520. case. Decoherence is obviously a key obstacle in the [332] Albrecht, A., 1993, Phys. Rev. D48, 3768–3778. implementation of information processing hardware that [332] Alicki R., and K. Lendi, 1987, Quantum Dynamical takes advantage of the superposition principle. While we Semigroups and Applications, # 286 of Lecture Nores have not focused on quantum information processing, the In Physics (Springer, Berlin). discussion was often couched in the language inspired by [332] Anderson, P. W., 2001, Stud. Hist. Phil. Mod. Phys. 32, the information theory. This is no accident: It is the 487-494. belief of this author that many of the remaining gaps in [332] Ambegaokar, V., 1991, Ber. Bunsenges. Phys. Chem. 95 our understanding of quantum physics and its relation to , 400–404. [332] Ammann, H., R. Gray, I. Shvarchuk, and N. Chris- the classical domain – such as the definition of systems, tensen, 1998, Phys. Rev. Lett. 80, 4111. or the still mysterious details of the “collapse” – shall [332] Anglin, J. R., 1997, Phys. Rev. Lett. 79, 6 follow pattern of the “predictability sieve” and shall be [332] Anglin, J. R., and Zurek, W. H., 1996, Phys. Rev. D53, expanded into new areas investigation by considerations 7327–7335. that simultaneously elucidate nature of “the quantum” [332] Anglin, J. R., J. P. Paz, and W. H. Zurek, 1997, Phys. and of “the information”. Rev. A 53, 4041. 53

[332] Auletta, G., 2000, Foundations and Interpretation of [332] Borel, E., 1914, Introduction Géométrique àQuelques Quantum Theory (World Scientific, Singapore). Théories Physiques (Gauthier-Villars, Paris). [332] Bacciagaluppi, G., and M. Hemmo, 1998, pp. 95–114 in [332] Born, M., 1969, Albert Einstein, Hedwig und Max Born, Healey and Hellman, 1998 Briefwechsel 1916–1955, (Nymphenburger Verlagshand- [332] Ballentine, L. E., 1970, Rev. Mod. Phys. 42, 358. lung, München). [332] Ballentine, L. E., Y. Yang, and J. P. Zibin, 1994, Phys. [332] Braginski, V. B., Vorontsov, Y. I., and Thorne, K. S., Rev. A 50, 2854 1980, Science 209, 547–557; reprinted in Wheeler and [332] Barnett, S. M., and Phoenix, S. J. D., 1989, Phys. Rev. Zurek, 1983. A40, 2404–2409. [332] Braginsky, V. B., and F. Ya. Khalili, 1996, Rev. Mod. [332] Barnum, H., C. M. Caves, J. Finkelstein, C. A. Fuchs, Phys. 68, 1. and R. Schack, 2000, Proc. Roy. Soc. London A 456, [332] Brandenberger, R., R. Laflamme, and M. Mijic, 1990, 1175-1182. Mod. Phys. Lett., A5, 2311. [332] Barvinsky, A. O., and A. Yu. Kamenshchik, 1990, Class. [332] Braun, D., P. A. Braun, and F. Haake, 2000, Opt. Quant. Grav., 7, 2285–2293. Comm. 179, 411. [332] Barvinsky, A. O., and A. Yu. Kamenshchik, 1995, Phys. [332] Bravyi, S. B., and A. Y. Kitaev, 1998, Rev. D 52 743–757. quant-ph 9811052. [332] Beckman, D., D. Gottesman, and M. A. Nielsen, 2001, [332] Brillouin, L., 1962, Science and Information Theory, quant-ph/0102043. (Academic Press, New York). [332] Beige, A., D. Braun, D. Tregenna, P. L. Knight, 2000, [332] Brillouin, L., 1964, Scientific Uncertainty, and Informa- Phys. Rev. Lett. 85, 1762. tion, (Academic Press, New York). [332] Bell, J. S., 1981, in Quantum Gravity 2, edited by [332] Brodie, I., 1995, Phys. Rev. B 51, 13660-13668. C. Isham, R. Penrose and D. Sciama, (Clarendon Press, [332] Brun, T., and J. B. Hartle, 1999, Phys. Rev. D 6012 Oxford) pp. 611–637. 1165. [332] Bell, J. S., 1987, Speakable and Unspeakable in Quantum [332] Brun, T., I. C. Percival, and R. Schack, 1995, J. Phys. Mechanics, (Cambridge University Press). A 29, 2077. [332] Bell, J. S., 1990, “Against ‘Measurement’ ”, in Sixty- [332] Brune, M., E. Hagley, J. Dreyer, X. Maître, C. Wunder- Two Years of Uncertainty, edited by A. I. Miller, lich, J.-M. Raimond, and S. Haroche, 1996, Phys. Rev. (Plenum, NATO Series, New York). Lett 77, 4887. [332] Bennett, C. H., and D. P. DiVincenzo, 2000, Nature [332] Bruss, D., A. Ekert and C. Macchiavello, 1999, Phys. 404, 247. Rev. Lett. 81 2598. [332] Bennett, C. H., D. P. DiVincenzo, J. Smolin, and W. K. [332] Bub, J., 1997, Interpreting the Quantum World, Cam- Wootters, 1996, Phys. Rev. A 54, 3824. bridge University Press, Cambridge. [332] Berman, G. P., and G. M. Zaslavsky, 1978, Physica [332] Butterfield, J., 1996, Brit. J. Phil. Sci. 47, 200-221. 91A, 450. [332] Caldeira, A. O., and Leggett, A. J., 1983, Physica [332] Berry, M. V., and N. L. Balazs, 1979, J. Phys. A 12, 121A, 587–616. 625. [332] Carmichael, H. J., 1993, An Open Systems Approach to [332] Berry, M. V. 1995, in Fundamental Problems in Quantum Optics (Springer, Berlin). Quantum Theory, edited by Greenberger, D. M. and [332] Carvalho, A. R. R., P. Milman, R. L. de Matos Filho, Zeilinger, A. (N. Y. Academy of Sciences, New York) L. Davidovich, 2000, e-print 0009024. pp. 303-317. [332] Casati, G., and Chirikov, B. V., 1995a, Quantum Chaos: [332] Birkhoff, G., and J. von Neumann, 1936, Ann. Math. Between Order and Disorder (Cambridge University 37, 823-843 Press). [332] Blanchard, Ph., D. Giulini, E. Joos, C. Kiefer, I.-O. Sta- [332] Casati, G., and Chirikov, B. V., 1995b, Phys. Rev. Lett. matescu, 2000, Decoherence: Theoretical, Experimental, 75, 2507. and Conceptual Problems (Springer, Berlin) [332] Castagnino, M. A., A. Gangui, F. D. Mazzitelli, and I. I. [332] Blanchard Ph., and R. Olkiewicz, 2000, Phys. Lett. A Tkachev, 1993, Class. Quant. Grav. 10, 2495-2504. 273 223. [332] Caves, C. M., Thorne, K. S., Drever, R. W. P., Sand- [332] Boccaletti, S., A. Farini, and F. T. Arecchi, 1997, Phys. berg, V. P., and Zimmerman, M., 1980, Rev. Mod. Phys. Rev. E 55 4979. 52, 341–392. [332] Bocko, M. F., and R. Onofrio, 1996, Rev. Mod. Phys. [332] Caves, C. M., 1994, in Physical Origins of Time Asym- 68, 755. metry, edited by J. J. Halliwell, J. Perez-Mercader, and [332] Bogolubov, N. N., Logunov, A. A., Oksak, A. I., and W. H. Zurek (Cambridge University Press, Cambridge), Todorov, I. T, 1990, General Principles of Quantum p. 47. Field Theory, Kluwer, Dordrecht. [332] Cerf, N. and Adami, C., 1997, Phys. Rev. A 55, 3371– [332] Bohm, D., 1952, Phys. Rev. 85, 166–193; reprinted in 3374. Wheeler and Zurek, 1983. [332] Chapman, M. S., T. D. Hammond, A. Lenef, J. [332] Bohr, N., 1928, Nature 121, 580–590; [reprinted in Schmiedmayer, R. A. Rubenstein, E. Smith, and D. E. Wheeler and Zurek, 1983]. Pritchard, 1995, Phys. Rev. Lett. 75, 3783–3787. [332] Bohr, N., 1935, Phys. Rev. 48, 696-702; [reprinted in [332] Cheng, C. C., and M. G. Raymer, 1999, Phys. Rev. Lett Wheeler and Zurek, 1983]. 82, 4802. [332] Bohr, N., 1949, Albert Einstein: Philosopher-Scientist, [332] Chuang, I. L., and M. A. Nielsen, J. Mod. Opt. 44, edited by P. A. Schilpp, (Open Court Publishing, 2455-2467. Evanston, pp. 200–241); [reprinted in Wheeler and [332] Cirac, J. I., and P. Zoller, 1995, Phys. Rev. Lett. 74, Zurek, 1983]. 4091 54

[332] Cory, D. G., W. Mass, M. Price, E Knill, R. Laflamme, Phys. 192, 368. W. H. Zurek, T. F. Havel, and S. S. Somaroo, 1999, [332] Feynman, R. P., and Vernon, F. L., Jr., 1963, Ann. Phys. Rev. Lett. 81, 2152-2155. Phys. (N.Y.) 24, 118–173. [332] Cover, T. M, and J. A. Thomas, 1991 Elements of In- [332] Ford, J., and Mantica, G., 1992, Am. J. Phys. 60, 1086– formation Theory (Wiley, New York). 1098. [332] Cox, R. T., 1946, Am. J. Phys. 14, 1. [332] Fox, R. F., and T. C. Elston, 1994, Phys. Rev. E49, [332] Dalvit, D., J. Dziarmaga, and W. H. Zurek, 2000, Phys. 3683; ibid. E50, 2553. Rev. A 62, 013607. [332] Freedman, M. H., and D. A. Meyer, 1998, [332] Dalvit, D., J. Dziarmaga, and W. H. Zurek, 2001, Phys. quant-ph 9810055. Rev. Lett. 86, 373. [332] Furuya, K., M. C. Nemes, and G. Q. Pellegrino, 1998, [332] Dekker, H., 1977, Phys. Rev. A16, 2126–2134. Phys. Rev. Lett. 80, 5524 [332] d’Espagnat, B., 1976, Conceptual Foundations of Quan- [332] Fuchs, C. A., and A. Peres, 2000, Phys. Today 53, 70. tum Mechanics, W. A. Benjamin, (Reading, MA). [332] Gallis, M. R., 1992, Phys. Rev. A45, 47–53. [332] d’Espagnat, B., 1989, J. Stat. Phys. 58, 747–766. [332] Gallis, M. R., 1996, Phys. Rev. A53, 655–660. [332] d’Espagnat, B., 1995, Veiled Reality, Addison-Wesley, [332] Gallis, M. R., and Fleming, G. N., 1990, Phys. Rev. Reading, MA. [1, 17, 37, 167, 170, 171, 186] A42, 38–48. [332] Deutsch, D., 1985, Int. J. Theor. Phys. 24, 1. [332] Gell-Mann, M., and Hartle, J. B., 1990, Complexity En- [332] Deutsch, D., 1997, The Fabric of Reality, (Penguin, New tropy, and the Physics of Information, edited by W. H. York). Zurek, Addison-Wesley, Reading, pp. 425–458. [332] Deutsch, D., 1999, Proc. Roy. Soc. London A 455 3129. [332] Gell-Mann, M., and Hartle, J. B., 1993, Phys. Rev. [332] Deutch, D., 2001 quant-ph/0104033. D47, 3345–3382. [332] DeWitt, B. S., 1970, Physics Today, 23, 30-35. [332] Gell-Mann, M., and J. B. Hartle, 1997, in Quantum - [332] DeWitt, B. S., 1998, pp. 39-49 in Frontiers in Quantum Classical Correspondence, edited by Feng, D. H., and Physics, S. C. Lim, R. Abd-Shukor, and K. H. Kwek, Hu, B. L. (International Press, Cambridge, MA), p.3. eds. (Springer, Singapore). [332] Geroch, R., 1984, Noûs, XVIII, 617-634. [332] DeWitt, B. S., and Graham, N., 1973, The Many- [332] Ghirardi, G. C., Rimini, A., and Weber, T., 1986, Phys. Worlds Interpretation of Quantum Mechanics (Prince- Rev. D34, 470–491. ton University Press). [332] Ghirardi, G. C., Rimini, A., and Weber, T., 1987, Phys. [332] Dieks, D., 1982, Phys. Lett. A 92, 271. Rev. D36, 3287–3289. [332] Dirac, P. A. M., 1947, The Principles of Quantum Me- [332] Gisin, N., and Percival, I. C., 1992, J. Phys. A: Math. chanics (Clarendon Press, Oxford) Gen. 25, 5677–5691. [332] Donald, M. J., 1995, Found. Phys. 25, 529–571. [332] Gisin, N., and Percival, I. C., 1993a, J. Phys. A: Math. [332] Dowker, H. F., and Halliwell, J. J., 1992, Phys. Rev. Gen. 26, 2233–2243. D46, 1580–1609. [332] Gisin, N., and Percival, I. C., 1993b, J. Phys. A: Math. [332] Dowker, H. F., and Kent, A., 1996, J. Stat. Phys. 82, Gen. 26, 2245–2260. 1575–1646. [332] Gisin, N., and Percival, I. C., 1993c, Phys. Lett. A175, [332] Duan, L.-M., and G.-C. Guo, 1998, Phys. Rev. A 57, 144–145. 737-741. [332] Giulini, D., 2000, pp. 67-91 in Relativistic Quantum [332] Durakiewicz, T., S. Halas, A. Arko, J. J. Joyce, and Measurement and Decoherence, H.-P. Breuer and F. D. P. Moore, 2001, Phys. Rev. B 6404, 5101. Petruccione, eds. (Springer, Berlin). [332] Einstein, A., 1917, Verh. Deutsch. Phys. Ges. 19, 82. [332] Giulini, D., Kiefer, C., and Zeh, H. D., 1995, Phys. Lett. [332] Einstein, A., B. Podolsky, and N. Rosen, 1935, Phys. A199, 291–298. Rev. 47, 777–780; [reprinted in Wheeler and Zurek, [332] Giulini, D., E. Joos, C. Kiefer, J. Kupsch, I.-O. Sta- 1983]. matescu, and H. D. Zeh, 1996, Decoherence and the [332] Eisert, J., and M. B. Plenio, 2002, Phys. Rev. Lett., in Appearance of a Classica World in Quantum Theory press. (Springer, Berlin) [332] Ekert, A., and C. Macchiavello, 1996, Phys. Rev. Lett. [332] Gleason, A. M., 1957, Journ. Math. Mech. 6, 885–893. 77, 2585. [332] Gnedenko, B. V., 1982, The Theory of Probability, (MIR [332] Elby, A., 1993, in Symposium on the Foundations of Publishers, Moscow). Modern Physics, edited by P. Busch, P. Lahti, and P. [332] Goldstein, S., 1998, Physics Today, 52, #3, 42-46 (part Mittelstaedt (World Scientific, Singapore), p. 168. I) and #4, 38-42 (part II). [332] Elby, A., 1998, pp. 87-94 in Healey and Hellmann, 1998 [332] Goldstein, S., and D. N. Page, 1995, Phys. Rev. Lett. [332] Elby, A., and Bub, J., 1994, Phys. Rev. A49, 4213– 74, 3715-3719. 4216. [332] Gorin, T., and Seligman, T. H., 2001, [332] Emerson, J., and L. E. Ballentine, 2001a, Phys. Rev. quant-ph/01120030. A63, 052103. [332] Gorini, V., A. Kossakowski, and E. C. G. Sudarshan, [332] Emerson, J., and L. E. Ballentine, 2001b, Phys. Rev. 1976, J. Math. Phys. 17, 821-825. E64, 026217. [332] Gottesman, D., 1996, Phys. Rev. A 54, 1862. [332] Everett, H., III., 1957a, Rev. Mod. Phys. 29, 454–462; [332] Gottfried, K., 1966, Quantum Mechanics, section 20 reprinted in Wheeler and Zurek, 1983. (Benjamin, New York). [332] Everett, H., III, 1957b, The Theory of the Univer- [332] Gottfried, K., 2000, Nature 405, 533-536. sal Wave Function, Princeton University Dissertation; [332] Grabert, H., Schramm, P., and Ingold, G.-L., 1988, reprinted in DeWitt and Graham, 1973. Phys. Rep. 168, 115–207. [332] Farhi, E., J. Goldstone, and S. Guttmann, 1989, Ann. [332] Graham, N., 1970, The Everett Interpretation of Quan- 55

tum Mechanics, (University North Carolina, Chapel 4153. Hill). [332] Kitaev, A. Y., 1997a, quant-ph 9707021. [332] Graham, R., M. Schlautmann, and P. Zoller, 1992, [332] Kitaev, A. Y., 1997b, Russ. Math. Surv. 52, 1191-1249. Phys. Rev. A 45 R19-R22. [332] Kitaev, A. Y., 1997c, in Quantum Communication, [332] Griffiths, R. B., 1984, J. Stat. Phys. 36, 219–272. Computing, and Measurement, pp. 181-188. [332] Griffiths, R. B., 1996, Phys. Rev. A 54, 2759. [332] Klappauf, B. G., W. H. Oskay, D. A. Steck, and M. G. [332] Griffiths, R. B, and Omnès, R., 1999, Physics Today, Raizen, 1998, Phys. Rev. Lett. 81, 1203; erratum in 52, 26-31. 1999, Phys. Rev. Lett. 82, 241. [332] Haake, F., 1991, Quantum Signatures of Chaos, [332] Knill, E., and R. Laflamme, 1997, Phys. Rev. A 55, 900. Springer, Berlin. [332] Knill, E., R. Laflamme, and L. Viola, 2000, Phys. Rev. [332] Haake, F., and R. Reibold, 1985, Phys. Rev. A32, 2462. Lett. 84, 2525. [332] Haake, F., M. Ku´s, and R. Sharf, 1987, Z. Phys. B65, [332] Knill, E., R. Laflamme, and W. H. Zurek, 1996, 381. quant-ph/96010011. [332] Haake, F., and D. F. Walls, 1987, Phys. Rev. A36, 730– [332] Knill, E., R. Laflamme, and W. H. Zurek, 1998a, Science 739. 279, 324-345. [332] Habib, S., K. Shizume, and W. H. Zurek, 1998, Phys. [332] Knill, E., R. Laflamme, and W. H. Zurek, 1998b, Proc. Rev. Lett. 80, 4361. Roy. Soc. London A 454, 365-384. [332] Halliwell, J. J., 1989, Phys. Rev. D 39, 2912-2923 [332] Koch, P. M., 1995, Physica D 83, 178. [332] Halliwell, J. J., 1999, Phys. Rev. D 6010 5031. [332] Kokorowski, D., A. D. Cronin, T. D. Roberts, and D. E. [332] Halliwell, J. J., J. Péres-Mercader, and W. H. Zurek, Pritchard, 2001, Phys. Rev. Lett., in press. eds., 1994, Physical Origins of Time Asymmetry (Cam- [332] Kolmogorov, A. N., 1960, Math. Rev. 21, 2035. bridge University Press). [332] Kraus, K., 1983, States, Effects, and Operations [332] Haroche, S., 1998, Physica Scripta T76, 159. (Springer, Berlin). [332] Hartle, J. B., 1968, Am. J. Phys. 36, 704. [332] Kravtsov, V. E., and B. L. Altshuler, 2000, Phys. Rev. [332] Healey, R. A., 1998, pp. 52–86 in Healey and Hellmann Lett., 84, 3394. (1998). [332] Krzywicki, A., 1993, Phys. Rev. D48, 5190. [332] Healey, R. A., and G. Hellman, eds., 1998, Quan- [332] Kwiat, P. G., A. M. Steinberg, and R. Y. Chiao, 1992, tum Measurement: Beyond Paradox, University of Min- Phys. Rev. A 45, 7729-7739.. nesota Press Minneapolis. [332] Laflamme, R., C. Miquel, J. P. Paz, and W. H. Zurek, [332] Heisenberg, W., 1927, Z. Phys.43, 172–198; [English 1996, Phys. Rev. Lett. 77, 198. translation in Wheeler and Zurek, 1983]. [332] Landau, L., 1927, Zeits. Phys. 45, 430-441. [332] Herzog, T. J., P. G. Kwiat, H. Weinfurter, and A. [332] Landauer, R., 1991, Physics Today 44, 23. Zeilinger, 1995, Phys. Rev. Lett. 75 3034-3037 [332] Landauer, R., 1995, Phil. Trans. Roy. Soc. Lond. 353, [332] Hillery, M. A., and M. O. Scully, 1983, in Quantum Op- 367. tics, Experimental Gravitation, and Measurement The- [332] Laskar, J., 1989, Nature 338, 237. ory edited by P. Meystre and M. O. Scully (Plenum, [332] Leggett, A. J., 1980, Supp. Prog. Theor. Phys. 69, 80– New York), p. 65. 100. [332] Holland, P. R., 1993, The Quantum Theory of Motion [332] Leggett, A. J., 1998, in Quantum Measurement, edited (Cambridge University Press, Cambridge). by R. A. Healey and G. Hellman (University of Min- [332] Hu, B. L., Paz, J. P., and Zhang, Y., 1992, Phys. Rev. nesota Press, Minneapolis), p.1. D45, 2843–2861. [332] Leggett, A. J., 2002, Physica Scripta, in press. [332] Hu, B. L., Paz, J. P., and Zhang, Y., 1993, Phys. Rev. [332] Levstein, P. R., G. Usaj, H. M. Pastawski, J. Raya, and D47, 1576–1594. J. Hirschinger, 2000, J. Chem Phy. 113, 6285. [332] Ivanovic, I. D., 1981, J. Phys. A 14, 3241–3245. [332] Levstein, P. R., G. Usaj, and H. M. Pastawski, 1998, J. [332] Jacquod, P., Silvestrov, P. G., and Beenakker, C. W. J., Chem. Phys. 108, 2718. 2001, Phys. Rev. E 64, 055203 (R). [332] Lidar, D. A., D. Bacon, and K. B. Whaley, 1999, Phys. [332] Jalabert, R. A., and H. M. Pastawski, 2001, Phys. Rev. Rev. Lett. 82, 4556. Lett. 86 2490. [332] Lieb, E. H., 1975, Bull. AMS 81, 1. [332] Jammer, M., 1974, The Philosophy of Quantum Me- [332] Lindblad, G., 1976, Comm. Math. Phys. 48, 119. chanics, (Wiley, New York). [332] Lloyd, S., 1997, Phys. Rev. A55, 1613. [332] Jaynes, E. T., 1957, Phys. Rev. 106, 620-630. [332] London, F., and Bauer, E., 1939, La Théorie de [332] Jones, K. R. W., 1994, Phys. Rev. A 50, 3682-3699. l’Observation en Méchanique Quantique (Hermann, [332] Joos, E., 1986a, Ann. New York Acad. Sci. 480, 6–13. Paris); English translation in Wheeler and Zurek, 1983. [332] Joos, E., and Zeh, H. D., 1985, Z. Phys. B59, 223–243. [332] Mensky, M. B., and I. D. Novikov, 1996, Int. J. Mod. [332] Jozsa, R., 2002, quant-ph/02044153. Phys. D5, 1–27. [332] Karkuszewski, Z. P., C. Jarzynski, and W. H. Zurek, [332] Mermin, N. D., 1990a, Physics Today 43, 9–11. 2002, Phys. Rev. Lett, submitted. [332] Mermin, N. D., 1990b, Am. J. Phys. 58, 731–734. [332] Karkuszewski, Z. P., J. M. Zakrzewski, and W. H. [332] Mermin, N. D., 1994, Physics Today 47, 9–11. Zurek, 2002, Phys. Rev. A 65, 042113. [332] Miquel, C., J. P. Paz, and W. H. Zurek, 1997, Phys. [332] Karlsson, E. B., 1998, Physica Scripta T 76, 179-185. Rev. Lett. [332] Kent, A., 1990, Int. J. Mod. Phys., 5, 1745–1762. [332] Miller, P. A., S. Sarkar, and R. Zarum, 1998, Acta Phys. [332] Kiefer, C., 1987, Class. Quant. Grav. 4 1369-1382. Pol. B29, 4643. [332] Kiefer, C., 1996, in Giulini et al., (1996), p. 157. [332] Miller, P. A., and S. Sarkar, 1999, Phys. Rev. E 60, [332] Kiefer, C., and H. D. Zeh, 1995, Phys. Rev. D 51, 4145– 1542. 56

[332] Mohanty, P., and R. A. Webb, 1997, Phys. Rev. B 55, [332] Reichl, L. E., 1992, Transition to Chaos in Conservative R13 452 Classical Systems: Quantum Manifestations (Springer, [332] Monroe, C., Meekhof, D. M., King, B. E., and Berlin). Wineland, D. J., 1996, A ‘Schrödinger Cat’ superpo- [332] Romero, L. D., and J. P. Paz, 1997, Phys. Rev. A53, sition state of an atom, Science 272, 1131–1136. 4070. [332] Monteoliva, D., and J.-P. Paz, 2000, Phys. Rev. Lett. [332] Saunders, S., 1998, Synthese 114, 373-404. 85, 3373. [332] Schack, R., 1998, Phys. Rev. A 57, 1634 [332] Moore F. L. et al, 1994, Phys. Rev. Lett. 73, 2974. [332] Schack, R. and C. M., Caves, 1996, Phys. Rev. E 53, [332] Mott, N. F, 1929, Proc. Roy. Soc. London A 126, 79-84; 3387 reprinted in Wheeler and Zurek, 1983. [332] Schrödinger, E., 1926, Naturwiss 14, 664–666. [332] Moyal, J. E., 1949, Proc. Camb. Phil. Soc. 45, 99. [332] Schrödinger, E., 1935a, Nautrwiss 23, 807–812, 823– [332] Myatt, C. J., et al., 2000, Nature 403, 269. 828, 844–849; [English translation in Wheeler and [332] Namiki, M., and S. Pascazio, 1993, Phys. Rep. 232, 301. Zurek, 1983]. [332] Nielsen, M. A., and I. L. Chuang, 2000, Quantum Com- [332] Schrödinger, E., 1935b, Proc. Cambridge Phil. Soc. 31, putation and Quantum Information (Cambridge Univer- 555–563. sity Press, Cambridge) [332] Schrödinger, E., 1936, Proc. Cambridge Phil. Soc. bf 32, [332] Nielsen, M. A., E. Knill, and R. Laflamme, 1998, Nature 446-452. 396, 52-55. [332] Schumacher, B. W., 1996, Phys. Rev. A54, 2614–2628. [332] Ollivier, H., D. Poulin, and W. H. Zurek, 2002, in prepa- [332] Schumacher, B. W., and M. A. Nielsen, 1996, Phys. Rev. ration. A54, 2629. [332] Ollivier, H., and W. H. Zurek, 2002, Phys. Rev. Lett. [332] Schumacher, B. W., 1999, private communication con- 88, 017901. cerning distinguishability sieve. [332] Omnès, R., 1988, J. Stat. Phys. 53, 893–975. [332] Shiokawa, K., and Hu, B. L., 1995, Phys. Rev. E52, [332] Omnès, R., 1992, Rev. Mod. Phys. 64, 339–382. 2497–2509. [332] Omnès, R., 1994, The Interpretation of Quantum Me- [332] Shor, P. W., 1995, Phys. Rev. A 52 2493 chanics (Princeton University Press, Princeton). [332] Sinai, Ya., 1960, Math. Rev. 21 2036. [332] Paraoanu, G.-S. 2002, quant-ph 0205127. [332] Squires, E. J., 1990, Phys. Lett. A 145, 67–68 [332] Pattanyak, A. K., 2000, Phys. Rev. Lett. 83, 4526 [332] Stapp, H. P., 2001, Can. J. Phys, to appear, [332] Paz, J.-P., 1994, in Physical Origins of Time Asymme- quant-ph/0110148. try, pp. 213-220. [332] Steane, A. M., 1996, Proc. Roy. Soc. London A 452, [332] Paz, J. P., Habib, S., and Zurek, W. H., 1993, Phys. 2551–2576. Rev. D47, 488–501. [332] Stein, H., 1984, Noûs XVIII, 635-652. [332] Paz, J.-P., and S. Sinha, 1991, Phys. Rev. D 44, 1038. [332] Stern, A., Y. Aharonow, and Y. Imry, 1989, Phys. Rev. [332] Paz, J.-P., and S. Sinha, 1992, Phys. Rev. D 45, 2823. A 41 3436. [332] Paz, J.-P., and W. H. Zurek, 1999, Phys. Rev. Lett. 82, [332] Sussman, G. J., and J. Wisdom, 1992, Science 257, 56. 5181. [332] Tegmark, M., 2000, Phys. Rev. E61, 4194. [332] Paz, J.-P., and W. H. Zurek, 2001, pp. 533-614 in Co- [332] Tegmark, M., and Shapiro, H. S., 1994, Phys. Rev. E50, herent Atomic Matter Waves, Les Houches Lectures, R. 2538–2547. Kaiser, C. Westbrook, and F. David, eds. (Springer, [332] Tegmark, M., and Wheeler, J. A., 2001, Sci. American, Berlin). 284, 68-75. [332] Pearle, P., 1976, Phys. Rev. D13, 857–868. [332] Turchette Q. A., et al.,, 2000, Phys. Rev. A 62 053807. [332] Pearle, P., 1993, Phys. Rev. A48, 913–923. [332] Unruh, W. G., 1994, Phys. Rev. A50, 882. [332] Penrose, R., 1986, Quantum Concepts in Space and [332] Unruh, W. G., and Zurek, W. H., 1989, Phys. Rev. D40, Time, edited by C. J. Isham and R. Penrose, (Clarendon 1071–1094. Press) pp. 129–146. [332] Venugopalan, A., 1994, Phys. Rev. A50, 2742–2745. [332] Penrose, R., 1989, The Emperors New Mind, (Oxford [332] Venugopalan, A., 2000, Phys. Rev. A6001, 2102. University Press, Oxford). [332] Viola, L., and S. Lloyd, 1998, Phys. Rev. A58, 2733. [332] Perelomov, 1986, Generalized Coherent States and [332] von Neumann, J., 1932, Mathematische Grundlagen der Their Applications (Springer, Berlin) Quantenmechanik, (Springer, Berlin); reprinted 1981; [332] Peres, A., 1993, Quantum Theory: Concepts and Meth- English translation by R. T. Beyer, 1955: Mathematical ods (Kluwer, Dordrecht). Foundations of Quantum Mechanics, (Princeton Uni- [332] Pfau, T., Spälter, S., Kurtsiefer, C., Ekstrom, C. R., versity Press). and Mlynek, J., 1994, Phys. Rev. Lett. 73, 1223–1226. [332] Wallace, D., 2002, Issues in the Foundations of Rela- [332] Poyatos, J. F., Cirac, J. I., and Zoller, P., 1996, Phys. tivistic Quantum Theory, Merton College, University of Rev. Lett. 77, 4728. Oxford. [332] Poyatos, J. F., Cirac, J. I., and Zoller, P., 1997, Phys. [332] Walls, D. F., and Milburn, G. J., 1994, Quantum Optics, Rev. Lett. 78, 390-393. Springer, Berlin. [332] Preskill, J., 1998, Proc. Roy. Soc. London A 454, 385- [332] Wehrl, A., 1978, General properties of entropy, Rev. 410. Mod. Phys. 50, 221–260. [332] Preskill, J., 1999, Phys. Today 52 (#6) 24-30. [332] Wheeler, J. A., 1957, Rev. Mod. Phys. 29, 463-465. [332] Prosen, T., 2001, quant-ph/0106149. [332] Wheeler, J. A., 1978, Mathematical Foundations of [332] Raimond, J. M., M. Brune, and S. Haroche, 2001, Rev. Quantum Theory, edited by A. R. Marlow, (Academic Mod. Phys. 73, 565-582. Press, New York), pp. 9–48. [332] Rauch, H., 1998, Physica Scripta T76, 24. [332] Wheeler, J. A., 1983, in Wheeler and Zurek, 1983. 57

[332] Wheeler, J. A., and W. H. Zurek, 1983, Quantum The- Without Representation”, pp. 341-350 of From Statisti- ory and Measurement, (Princeton University Press). cal Physics to Statistical Inference and Back, P. Grass- [332] Wick, G. C., Wightman, A. S., and Wigner, E. P., 1952, berger and J.-P. Nadal, eds. (Plenum, Dordrecht, 1994). Phys. Rev. 88, 101–105. [332] Zurek, W. H., 1998a, Phil. Trans. Roy. Soc. Lond. [332] Wick, G. C., Wightman, A. S., and Wigner, E. P., 1970, A356, 1793. Phys. Rev. D1, 3267–3269. [332] Zurek, W. H., 1998b, Physica Scripta T76, 186. [332] Wightman, A. S., 1995, Il Nuovo Cimento 110B, 751– [332] Zurek, W. H., 2000, Ann. der Physik, 9, 855. 769. [332] Zurek, W. H., 2001, Nature, 412, 712-717. [332] Wigner, E. P., 1963, Am. J. Phys. 31, 6-15; reprinted [332] Zurek, W. H., 2002a, Phys. Rev. A, in press. in Wheeler and Zurek, 1983. [332] Zurek, W. H., 2002b, quant-ph/0211037. [332] Wisdom, J., 1985, Icarus, 63 272. [332] Zurek, W. H., Habib, S., and Paz, J.-P., 1993, Phys. [332] Wiseman, H. M., and G. J. Milburn, 1993, Phys. Rev. Rev. Lett. 70, 1187–1190. Lett. 70, 548. [332] Zurek, W. H., and Paz, J.-P., 1994, Phys. Rev. Lett. 72, [332] Wiseman, H. M., and J. A. Vaccaro, 1998, Phys. Lett. 2508–2511. A 250 241. [332] Zurek, W. H., and Paz, J.-P., 1995a, Physica D83, 300– [332] Wootters, W. K. and Fields, B. D., 1989, Ann. Phys. 308. (N. Y.) 191, 363-381. [332] Zurek, W. H., and Paz, J.-P., 1995b, Phys. Rev. Lett. [332] Wootters, W. K. and W. H. Zurek, 1979, Phys. Rev. D 72, 2508. 19, 473–484 Figure Captions [332] Wootters, W. K. and W. H. Zurek, 1982, Nature 299, 802. [332] Zanardi, P., 1998, Phys. Rev. A 57 3276. Fig. 1. Snapshots of the quantum (¯h = 0.16) Wigner [332] Zanardi, P., 2000, Phys. Rev. A, 63 40304, e-print function (a-c) and the classical probability distribution quant-ph9910016. in phase space (d) for the chaotic evolution generated [332] Zanardi, P., and M. Rasetti, 1997, Phys. Rev. Lett. 79, from the same initial Gaussian by the Hamiltonian: 3306. 79, 3306–3309. [332] Zeh, H. D., 1970, Found. Phys. 1, 69–76; reprinted in H = p2/2m κ cos(x l sin t) + ax2/2 Wheeler and Zurek, 1983. − − [332] Zeh, H. D., 1971, in Foundations of Quantum Mechan- For m = 1, κ = 0.36, l = 3 and a = 0 0.01 it ex- ics, edited by B. d’Espagnat, (Academic Press, New hibits chaos with Lyapunov exponent Λ = 0−.2. Quantum York) pp. 263–273. (a) and classical (d) are obtained at the same instant [332] Zeh, H. D., 1973, Found. Phys. 3, 109–116. [332] Zeh, H. D., 1986, Phys. Lett. A 116, 9-12. t = 20. They exhibit some similarities (i.e., the shape [332] Zeh, H. D., 1988, Phys. Lett. A 172, 311-317. of the regions of significant probability density, “ridges” [332] Zeh, H. D., 1990, in Complexity, Entropy, and the in the topographical maps of (a) and (d)), but the dif- Physics of Information, edited by W. H. Zurek, ference – the presence of the interference patterns with (Addison-Wesley, Redwood City) p. 405. W (x, p) assuming negative values (marked with blue) is [332] Zeh, H. D., 1992, The Physical basis of the Direction of striking. Saturation of the size of the smallest patches Time (Springer, Berlin) is anticipated already at this early time, as the ridges [332] Zeh, H. D., 1993, Phys. Lett. A172, 189–192. of the classical probability density are narrower than in [332] Zeh, H. D., 1997, in New Developments on Fundamental the corresponding quantum features. Saturation is even Problems in Quantum Physics , M. Ferrero and A. van more visible in (c) taken at t = 60 and (d), t = 100 der Merwe, eds. (Kluwer, Dordrecht). [332] Zeh, H. D., 2000, in Decoherence: Theoretical, Experi- (note change of scale). Sharpness of the classical features mental, and Conceptual Problems, edited by Ph. Blan- makes simulations going beyond t = 20 unreliable, but chard et al., (Springer, Berlin), p.19. quantum simulations can be effectivelly carried out much [332] Zurek, W. H., 1981, Phys. Rev. D24, 1516–1525. further, as the necessary resolution can be anticipated in [332] Zurek, W. H., 1982, Phys. Rev. D26, 1862–1880. advance from Eqs. (3.15) - (3.16). [332] Zurek, W. H., 1983, in Quantum Optics, Experimen- tal Gravitation, and Measurement Theory, edited by P. Fig. 2. Difference between the classical and quantum av- Meystre and M. O. Scully (Plenum, New York), p. 87. erage of the dispersion of momentum ∆2 = p2 p 2 is [332] Zurek, W. H., 1984a, Reduction of the wave packet: How plotted in (a) for the same initial condition, buth i three − h i dif- long does it take?, Los Alamos Preprint LAUR 84-2750, ferent values ofh ¯ in the model defined in Fig. 1, but with pp. 145-149 in G. T. Moore and M. T. Scully, eds. Fron- the parameter a = 0. The instant when the departure tiers in Nonequilibrium Statistical Physics (Plenum, between the classical and quantum averages becomes sig- New York, 1986) 53 nificant varies withh ¯ in a manner anticipated from Eqs. [332] Zurek, W H., 1984b, Phys. Rev. Lett. , 391. (3.5) and (3.6) for the Ehrenfest time, as is seen in the [332] Zurek, W. H., 1989, Phys. Rev. A40, 4731. [332] Zurek, W. H., 1991, Physics Today 44 (Oct.), 36–44. inset. Figure (b) shows the behaviors for the same value [332] Zurek, W. H., 1993a, Progr. Theor. Phys. 89, 281–312. ofh ¯, but for four different initial conditions. Inset ap- [332] Zurek, W. H., 1993b, Physics Today (letters) 46 (April), pears to indicate that the typical variance difference δ 13. varies only logarithmically withh ¯, although the large er- [332] Zurek, W. H., 1994, Decoherence and the Existential ror bars (tied to the large systematic changes of behavior Interpretation of Quantum Theory, or “No Information for different initial conditions) preclude one from arriving 58 at a firmer conclusion. (See Karkuszewski, Zakrzewski, in position and in momenta. and Zurek, 2002, for further details and discussion). Fig. 7. Predictability sieve in action. The plot shows pu- Fig. 3. Schematic representation of the effect of deco- rity Trρ2 for mixtures that have evolved from initial min- herence on the Bloch sphere. When the interaction with imum uncertainty wavepackets with different squeeze pa- the environment singles out pointer states located on the rameters s in an underdamped harmonic oscillator with poles of the Bloch sphere, pure states (which lie on its γ/ω = 10 4. Coherent states – which have the same surface) will evolve towards the vertical axis. This classi- − spread in position as in momentum, s = 1 – are clearly cal core is a set of all the mixtures of the pointer states. most predictable. Fig. 4. Information transfer in a c-not caricature of measurement, decoherence, and decoherence with noise. Fig. 8. Snapshots of a chaotic system with a double-well Bit-by-bit measurement is shown on the top. It is the potential: H = p2/2m + Ax4 Bx2 + Cx cos(ft). fundamental logical circuit used to represent decoherence In the example discussed here m −= 1, A = 0.5, B = 10, affecting the apparatus. Note that the direction of the f = 6.07 and C = 10 yielding the Lyapunov exponent information flow in decoherence – from the decohering Λ 0.45 (see Habib, Shizume, and Zurek, 1998). All apparatus and to the environment – differs from the in- figures≈ were obtained after approximately eight periods formation flow associated with noise. In short, as a result of the driving force. The evolution started from the of decoherence environment is perturbed by the state of same minimum uncertainty Gaussian, and proceeded ac- the system. Noise is – by contrast – perturbation in- cording to the quantum Moyal bracket (a), the Poisson flicted by the environment. Preferred pointer states are bracket (b), and (c) the Moyal bracket with decoherence selected so as to minimize the effect of the environment (constant D = 0.025 in Eq. (5.64)). In the quantum – to minimize the number of c-nots pointing from the cases ¯h =0.1, which corresponds to the area of the rect- environment at the expense of these pointing towards it. angle in the image of the Wigner function above. Inter- ference fringes are clearly visible in (a), and the Wigner Fig. 5. Time-dependent coefficients of the perturbative function shown there is only vaguely reminiscent of the master equation for quantum Brownian motion. The pa- classical probability distribution in (b). Even modest rameters used in these plots (where the time is measure decoherence (D =0.25 used to get (c) corresponds to coin units of Ω 1) are γ/Ω=0.05, Γ/Ω = 100, k T/¯hΩ= − B herence length ℓ = 0.3) dramatically improves the cor- 10, 1, 0.1. Plots on the right show the initial portion of c respondence between the quantum and the classical. The the plots on the left – the initial transient – illustrating remaining interference fringes appear on relatively large its independence of temperature (although higher tem- scales, which implies small scale quantum coherence. peratures produce higher final values of the coefficients). Plots on the right show that the final values of the co- Fig. 9. Classical and quantum expectation values of po- efficients strongly depend on temperature, and that the sition x as a function of time for an example of Fig. anomalous diffusion is of importance only for very low 8. Evolutionh i started from a minimum uncertainty Gaus- temperatures. sian. Noticeable discrepancy between the quantum and Fig. 6. Evolution of the Wigner function of the decoher- classical averages appears on a timescale consistent with ing harmonic oscillator. Note the difference between the the Ehrenfest time th¯ . Decoherence – even in modest rate at which the interference term disappears for the ini- doses – dramatically decreases differences between the tial superposition of two minimal uncertainty Gaussians expectation values. SystemSystem

ApparatusApparatus

Decoherence Environment 0.2 ln(1/ -h) Environment - Noise h=0.16 0 1 2 3 4 5 60.3 -h=0.064 30 - 0.2 h=0.0256 0.2 δ(-h) t 0.120 -h 0.1 0 -2 -1 10 10 -h 10 0 0 variance difference -0.1variance difference

-0.2 -0.1 0 10 10 20 20 30 time