Scribd Logo
Upload

Welcome to Scribd!

  • Asher Peres

    Uploaded by

    Franz Schindler
    47 views5 pages
    Physik

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document
    Physik

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    47 views5 pages

    Asher Peres

    Uploaded by

    Franz Schindler
    Physik

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    of 5
    PHYSICAL REVIEW D VOLUME 22, NUMBER 4 15 AUGUST 1980 ‘Can we undo quantum measurements? ‘Asher Peres* Center for Theoretical Physics. The Univers of Texas ot Austin, Austin, Texas 78712 (Received 1 October 197; revised manuscript received 14 Apri 1980) ‘The Schringer equation cannot convert a pure sate into a mixture Gust as Newton's equations cannot display irreversibly). However, to observe phase relationships between macroscopically distinguishable sates, one has to measure very peculiar operator. Aa example, constructed explicily, shows thatthe classical analog of such an ‘operator cannot be measured, because to do so would volte clusical irreversibility. Ths result justifies von ‘Neumann's measurement theory, without any hypothesis on the role ofthe observer. ‘The measurement process in quantum physics ‘was analyzed long ago by von Neumann‘ who showed that it could formally be described as the trans- formation of a pure state ¥=2ic,0, into a mixture p=Dic,|*P, Here, the @, are eigenstates of the ‘dynamical variable being measured, and the P, are the corresponding projection operators. ‘This irreversible transformation, commonly called the “collapse of the wave packet," cannot follow from the Schrédinger equation, since the latter generates a unitary mapping of the Hilbert space of states. In fact, the coupling of the eigen- states of the measured system to those of the mea~ suring apparatus is a perfectly reversible pro- cess as long as we are willing to measure cor- relations between the two. For these reasons, ‘von Neumann's theory has been considered un- satisfactory, or at least incomplete. ‘There have been several attempts* to prove ‘von Neumann's conjecture by supplementing quan- tum theory with superselection rules forbidding the measurement of operators of a certain type (those which eonneet macroscopically different states of the apparatus). The purpose of this paper is to show that systems with many degrees of freedom are indeed subject to such superselection rules. A general proof of this assertion would bbe very diffieult, but the following model is typical enough to convey belief in the result. ‘Consider a macroscopic apparatus designed to measure the z component of the spin of an elec tron. This apparatus has a pointer (center-of- mass coordinate q, conjugate momentum p) int tially localized around q=0. ‘The pointer is to move through a macroscopic distance 1 to the Fight or the left depending on whether =! or =}. This ean be achieved by the coupling H =2V(0s, p, where V0) is a large velocity, 80 large indeed that we can neglect all the other terms {in the Hamiltonian during the brief duration of the coupling.” Before the measurement, the state of the elee= tron is (7) and that of the apparatus is ¥(¢, 425435 14x) Where g,,43,--- ,4y are the other, “ir relevant," degrees of freedom. Naturally, Nis avery large number, say 10. (it would be more realistic to assume a density matrix instead of the pure state Y, but this refinement is not needed at the present stage.) After completion of the coupling, the combined state is (rome a where L=/ Vat. Since 9 is peaked around ¢ =0, etPhy is peaked around q=L. Thus, the sign of 4 3s correlated to that of s, and 6) =Csiga(a))= (lal? - 181"). @ We have performed what von Neumann calls a ‘measurement. (As we shall soon see, a better word would be “premeasurement.") ‘The question is whether this process is revers~ ible and, in particular, whether the relative phase ‘a/8 is still observable. At the present stage, it 4s, as can be seen by measuring the expectation values of the operators, Ay=s,cos2Lp +5, 8in2Lp (sa) and Ay"s,Sin2Lp ~s,c0s2Lp. (90) [To measure A, and A, we divide the electrons in ‘wo identical but disjoint ensembles. After each electron passage through the apparatus, we first measure p (modulo #/L) then the component of F in the direction of tan"(2Lp) oF cot"\(-2Lp). Note that the eigenvalues of A, and A, are 25.] ‘These operators ean conveniently be combined 0 0) czy : oy @) A straightforward calculation yields (4) fasts o an {© 1980 The American Physical Society 80 ASHER PERES which, together with Bq. (2), gives a and 8 sep- arately, up to a common phase. However, if we wait some time, the state (1) will evolve into ea(Qheon-C ens). ® where His the Hamiltonian of the electron and apparatus. Assuming for simplicity that the two spin states have the same energy, we obtain (Amage fleritetinyeettretitertrgaty. (6) But o¥7e%te®> is simply eM 0D (i.e,, H with the q coordinate shifted by L) and (6) can'be writ- ten as (A) caaeettttenig-ten tony, o ‘The coefficient of a@* would still be one if H did rot depend on 7, but there is no reason to expect that. As the pointer moves with respect to some fixed seale on the apparatus, its energy may vary Somewhat from place to place and the coefficient of a8* may be less than one in absolute value. For small f we get ettetenigrtentet)) 221 26H e+e, (Ba) where On? = (L(g ~L) ~Hq +L) ~ Glq —L) ~H(q +L). (@) Moreover, if the other degrees of freedom of the apparatus are ina mixed state, this coefficient will quickly fall to zero," because of the random- ness of the phases. ‘The time needed to erase (A) is of the order of 1/6H. Tt is therefore in~ versely proportional to the strength of the coupling of the macroscopic degree of freedom q, used for the measurement, with the other degrees of freedom of the apparatus. In the present model, this time could be of the order of the size of the pointer divided by the speed of sound (a few micro- seconds). ‘This neat distinetion between the reversible premeasurement—Bq. (1)—and the ensuing r= reversible process is admittedly unrealistic in ‘most instances. In practice, a macroscopic ap- paratus has almost always an amplification mech- anism based on a metastable initial state! and irreversibility appears at the very outset of the process. However, the amplification requirement is not essential and it obscures the true nature of the irreversibility of quantum measurements, which is explained below. (The reversal of an ‘idealized premeasurement is illustrated in Fig. ) ‘The above discussion of Eq. (7)—or some simi- Jar argument!**—is usually considered as a proof that the relative phase of the two branches of Eq. (6) is “lost” after some finite time. However, such arguments are not convineing, because Ea. (6) represents a pure state (what else could it bbe?) and this ean be shown by measuring the ex- pectation value of another operator, namely, Alseting elit, ) Indeed, we trivially have (A’) =aB*, since the #0" factors in’ cancel those of the wave func tons. However, the operator A’ has very peculiar properties. (it is not of course the Heisenberg picture of A, the later being o!#4.e""", mn fact, ve are always working in the SehrBdinger picture.) This A’ operator is explicitly time dependent and is also a constant of the motion To verify that it 1s a constant of the motion, st is enough to observe that its matrix elements be- tween any two Schr8dinger states are constant, or simply to go to the Heisenberg picture, where Af looks like Ag, without any time dependence, ‘ow, these explicitly time-dependent constants of the motion are very familiar in elassieal me- chanics. For example, for a free particle, q ~it/m is such a constant. Its meaning simply is the initial value of g. For an tarmonie oseillator, such a constant would be tan(msg/p) ~ast, the ‘meaning of which isthe Initial value of the phase. Ingeneral, for a system with N degrees of free- dom, there are 2N constants of the motion, a few of which may be explicitly time independent (the total energy, momentum, ete.), but almost all FIG. 1. idealized premensurement, using the recoil of a rigid double mirror. Ifa particle is reflected from, the first mirror, a correlation is established between the momentum of the particle and that of the instrument (this ts the premeasurement). That correlation is then reversibly destroyed when the particle {a reflected {rom the second mizror. (Note that if we wish to complete ‘the measurement and to observe the recoil of the double mirror between the two reflections, the latter must be prepared with ap But if we forego observing the recoil, the interference pattern fs restored because the same Ag 18 added and Subtracted at both reflections.) 2 CAN WE UNDO QUANTUM MEASUREMENTS? a of which include the time explicitly. Their phys~ ‘cal meaning is to give the 2N initial positions and ‘momenta as explieit functions of the positions and ‘momenta at some future time f. The structure of these constants of the motion is of eourse hope lessly complicated for large N and finite f. It leaves us no choice but to replace Newtonian me~ chanics by statistical mechanies. It is our in- ability to make use of these constants of the mo- tion which is the cause of irreversibility. In the present ease, we must measure, instead of A given by Ba. (4), are(? eur 10 where p’=e" pelt is the value of p immediately after the premeasurement, expressed as a function of p and 241 other variables at a later time f. Im classical physies, we would say that this is so complicated that only @ “Maxwell demon” can mea~ sure all these variables and then compute p" (as suming H is known). In quantum physics, the task {s even more difficult because the 2N variables o not commute. Therefore, the Maxwell demon rust contrive a single measurement” for p’, which is an ineredibly complicated function of 28 noncommuting variables and of ¢. Tm other words, we see that not every self- adjoint operator corresponds to an observable, simply because not every classical dynamical variable is observable. It isthe inobservability of these operators which makes pure states appear as mixtures and causes the irreversibility of quan- tum measurements. Tm conclusion, let us summarize the assumptions used in the derivation of this result. First, we note that the macroscopic degree of freedom used for the measurement—here, the center of mass ‘of the pointer —is not completely isolated from the other degrees of freedom of the apparatus.” (Wve could, of course, have treated these other degrees of freedom as an external reservoir, but then our result would have been trivial. It is es~ sential that our system be a closed one.) ‘The Second assumption is the impossibility of measuring the elassieal analog of the operator p (There is also a tacit assumption that ifa classical measurement is impossible, the same is true for the corresponding quantum measurement.) Here, it may be objected that as long as the number of dogrees of freedom is finite itis not impossible to measure p’, only very difticult. In principle, a measurement of p” should always be possible at the cost of a great increase of entropy of the rest of the world."" From this point of view, as Tong as we are able to pay the price,” we definitely can undo quantum measurements,* exeept in the ‘unattainable mathematical limit of an infi apparatus."* However, if we admit that a finite system may appear irreversible (if the time needed for a Poincaré reeurrence is longer than the Universe lifetime), the present paper shows how the irreversibility of quantum measurements is ‘rooted in the familiar classical irreversibility. 1am very grateful to J. A. Wheeler for the warm hospitality of the University of Texas and to E. C. G, Sudarshan for many stimulating dis- cussions. ‘The final version of this paper has benefited from comments by J. 8. Bell (CERN) and A. Ron (Technion). This work was supported {in part by the Center for Theoretical Physics, ‘The University of Texas at Austin, Austin, Texas ‘78712, and also by NSF Grant No. PRYTS-26592. APPENDIX: A MORE REALISTIC MODEL ‘The simple model discussed above involves an explicitly time-dependent coupling ¥(?), supposedly ‘switched on and off by an external agent. ‘This may give the impression that we are dealing with an open (Le., incompletely described) system, for which the transformation of a pure state into a mixture would be trivial In a real-life Stern-Gerlach experiment, this time dependence is of course due to the trans- Iational degree of freedom of the electron, which ‘was arbitrarily ignored in our model. A more realistic deseription of what happens follows, We write the complete Hamiltonian as HHH, $203 pul, —2ule =), cay where H, refers to the apparatus, H, to the free electron (mass m, momentum &, position x), V 4s a coupling constant, w is the unit step function, and x, and x, are the entrance and exit points of the electron as it passes through the apparatus. ‘The pointer is assumed massive enough so that its velocity p/iM is negligible when the electron is outside the apparatus. When it is inside, the pointer velocity is +. ‘The measurement process can be described as a scattering of the electron and the apparatus, Before the “collision,” the electron has momen- tum &, When it reaches the apparatus, it meets an energy barrier of height 4 Vp and thickness %4-%,. Inside the barrier, its momentum is b =" (2 £2m¥p)'*~ b emVp/ie, where we have as sumed that 4 > 2mVp, so that most electrons are transmitted (a reflected electron would mean an ‘unsuccessful experiment). The outgoing electron still has momentum 2, but has been subject to a hase shift (k ~t)(x,~x,)=1Vp where + *m(x, =»,)/l is the classical time of passage through the apparatus. We now identify L=7V and the a2 ASHER PERES om [Ceens Cen] , (02) whence the discussion proceeds as before. However, several remarks are in order. First, wwe have treated p as a constant during the colli- ion, Le., we assumed that p=ail,/oq =0. This is of course incompatible with a nontrivial SH (see Eq. (@b)). However, we can make the change {np arbitrarily small by inereasing V and (keeping their ratio constant, so that L remains unchanged). This does not affect 6, but makes arbitrarily small. ‘The condition is easily seen to be 7<<1/0H, 1.e., the premeasurement must be very brief, ‘compared to the time required to make the measurement irreversible. ‘To avoid a possible misunderstanding, it must bbe emphasized that OH is much smaller than the energy uncertainty aif=((iF) ~(i)*¥*, Indeed there must be many different energy eigenstates involved to make the measurement possible.” tn particular, the incoming electron must have ‘ail 6H because the two branches of the outgoing electron will not interfere if 47> aH (Wat is, we Would need an operator much more complicated than A to display their interference”). ‘The above remark is closely related to overall energy conservation. We have assumed hitherto that the outgoing electron ind the same energy as the incoming one. This cannot, of course, be rigorously exact if H(q~1)#i(q-+L). A more correct treatment follows. First, assume that initially the apparatus is in an eigenstate of energy 2, and thatthe elvetron too is monochromatie with energy K=22/2m. ‘Then obviously there is no irreversibility since the operator ein Eg. (A2) becomes a phase factor expl-it(B,+4)] and (A) Is constant. The energy picked up or released by the electron ex- actly compensates the energy difference in the final state of the apparatus. It is thus important to understand why I(A)! may decrease if we have a superposition (or mixture), rather than an ener ay eigenstate -Byen if |,) is an eigenstate of H,, the states 278.) uoually are not. We can write *0n sabbatical leave from Techaion-Israel Institute of ‘Technology, Half 5. yon Neumann, Mathematical Foundations of Quantum ‘Mechanics (Princeton University Press, Princeton, 1955). 2p. A, Moldauer, Phys. Rev. D 5, 1078 (1972). SA, Pores, Am. J. Phys, 22, 866 (1974), 102) =f ott BADE, where the coefficients c, depend also on K, because L=mlx,=%|/k- By virtue of energy conservation, the scattering process can therefore be written VB y4)~ f€\(BpB, NIE, K +85 ~ EME, where the 4 subscripts refer to 5, Now let the electron be initially in a superposi- tion / gl] K)dK (the apparatus may still be ini~ tially in an energy eigenstate”). ‘The outgoing states become J s6t0e Egy By ONE K + By ~ BNE AK, In order to compute (A), we first note that (B',K +B, = Bet] B+, ED (Ee EDO” BE! -K+E), s0 that A) saa f etter Ve (6y,B,Ket(Eay Es) x (E'|e%29) B)8(K’ — BY -K + Bd Ed k'dK dk’ We now make two essentia! physical assump- tons. One is tat AK

    You might also like