quant-ph0211036/qc3.tex
1: \documentclass[aps,twocolumn,footinbib,floatfix,showpacs]{revtex4}
2: 
3: \usepackage{graphicx}
4: 
5: \newcommand\kbar{{\bar k}}
6: 
7: \begin{document}
8: %\draft
9: \pacs{03.65.Bz,05.45.Ac,05.45.Pq} 
10: \title{Continuous Quantum Measurement and the Quantum to Classical Transition
11:        \vbox to 0pt{\vss
12:                     \hbox to 0pt{\hskip-50pt\rm LA-UR-02-5289\hss}
13:                     \vskip 25pt}
14: }
15: \preprint{LA-UR-00-XXXX}
16: \author{Tanmoy Bhattacharya}
17: \email{tanmoy@lanl.gov}
18: \homepage{http://t8web.lanl.gov/t8/people/tanmoy/}
19: \author{Salman Habib}
20: \email{habib@lanl.gov}
21: \homepage{http://t8web.lanl.gov/t8/people/salman/}
22: \author{Kurt Jacobs}
23: %\email{k.jacobs@lanl.gov}
24: \homepage{http://t8web.lanl.gov/t8/people/kaj/}
25: \affiliation{T-8, Theoretical Division, MS B285, Los Alamos
26: National Laboratory, Los Alamos, New Mexico 87545}
27: %\affiliation{A boulangerie in cheesy France}
28: 
29: \begin{abstract}
30: While ultimately they are described by quantum mechanics, macroscopic
31: mechanical systems are nevertheless observed to follow the
32: trajectories predicted by classical mechanics. Hence, in the regime
33: defining macroscopic physics, the trajectories of the correct
34: classical motion must emerge from quantum mechanics, a process
35: referred to as the quantum to classical transition. Extending previous
36: work [Bhattacharya, Habib, and Jacobs, Phys.~Rev.~Lett.~{\bf 85}, 4852
37: (2000)], here we elucidate this transition in some detail, showing
38: that once the measurement processes which affect all macroscopic
39: systems are taken into account, quantum mechanics indeed predicts the
40: emergence of classical motion.  We derive inequalities that describe
41: the parameter regime in which classical motion is obtained, and
42: provide numerical examples.  We also demonstrate two further important
43: properties of the classical limit.  First, that multiple observers all
44: agree on the motion of an object, and second, that classical
45: statistical inference may be used to correctly track the classical
46: motion.
47: \end{abstract}
48: 
49: \maketitle
50: 
51: %\begin{multicols}{2}
52: 
53: \section{Introduction}
54: Macroscopic mechanical systems are observed to obey classical
55: mechanics to within experimental error.  However, the atoms which
56: ultimately make up these systems certainly obey quantum
57: mechanics.  Therefore, the question of how the observed classical
58: mechanics emerges from the underlying quantum mechanics arises
59: immediately.  This emergence, referred to as the quantum to classical
60: transition, is particularly curious in the light of the fact that the
61: equations of motion for the trajectories of classical mechanics are
62: nonlinear, and can therefore exhibit chaos, whereas even a proper
63: quantification of chaos in quantum mechanics has been difficult to
64: obtain~\cite{noqc}.
65: 
66: Note that the task of explaining the quantum to classical transition
67: (the QCT) is essentially a practical question: it is a question of
68: explaining why real systems, such as nonlinear pendulums, baseballs,
69: and other systems which can be be built and observed in the laboratory
70: obey classical mechanics (at least to within any experimental error).
71: It is not a question of obtaining classical mechanics precisely as a
72: formal limit of quantum mechanics.  In fact, due to the absence of
73: chaos in closed quantum systems~\cite{noqc}, and the non-commutativity
74: of the twin limits $\hbar\rightarrow 0$ (the semi-classical limit) and
75: $t\rightarrow\infty$ (the long-time limit necessary to describe chaos)
76: efforts to extract classical chaos as a formal limit have been less
77: than successful.
78: 
79: If one describes macroscopic objects sufficiently realistically using
80: quantum mechanics, then it should be possible to predict the (often
81: chaotic) trajectories of classical dynamics.  In order to do this, it
82: is important to realize that {\em all} real classical systems are
83: subject to interaction with their environment.  This interaction does
84: at least two things.  First, it subjects the system to noise and
85: damping~\cite{CL,qnoise} (as a consequence all real classical systems
86: are subject to noise and damping --- even if small), and second, the
87: environment provides a means by which information about the system can
88: be extracted (effectively continuously if desired), providing a
89: measurement of the system~\cite{cqm1,cqm1a,cqm2,cqm3,cqm4}.  
90: 
91: Two levels of description have been used to discuss the QCT. The first
92: utilizes the decoherence resulting from tracing over the environment
93: to suppress quantum interference \cite{deco}: it is assumed that no
94: dynamical information about the individual system has been extracted
95: from its environment. In many circumstances this alone can lead to an
96: effectively classical evolution of a phase space distribution
97: function~\cite{HSZ1998}. As mentioned above, a more fine-grained
98: description is achieved when the environment is taken to be a meter
99: that is continuously monitored, leading to a `quantum trajectory
100: unraveling' of the system density operator conditioned on the
101: measurement record. If one averages over all possible measurement
102: results, the description reverts to that at the level of phase space
103: distributions. However, the fine-grained description which explicitly
104: incorporates monitoring of the environment is required to understand
105: the QCT at the level of extracting classical trajectories from the
106: quantum substrate.
107: 
108: An example of an environment that naturally provides a measurement is
109: that of the electromagnetic field which surrounds the system.
110: Monitoring this environment consists of focusing the light which is
111: reflected from the system, allowing the motion to be observed.  If the
112: environment is not being monitored, then the evolution is simply given
113: by averaging over all the possible motions of the system.  Classically
114: this means an average over any uncertainty in the initial conditions,
115: and over the noise realizations.
116: %This is the sense in which environmental interaction 
117: %and measurement are the same thing.  (this said more clearly later)
118: However, in the absence of explicit observation (monitoring the
119: environment and recording the evolution) it is impossible to obtain
120: classical trajectories: the system must be 
121: described by an ever broadening probability (or pseudo-probability) 
122: distribution in phase space.  This is an experimental truism, and 
123: therefore applies regardless of whether the system is being treated 
124: by a classical or quantum mechanical theory.
125: 
126: Since all classical systems are subject to environmental interactions,
127: and since measurement is necessary to deduce the trajectories of
128: classical motion,
129: %(this becomes absolutely crucial in chaotic systems, because of the extreme
130: % sensitivity to any uncertainty in initial conditions - and also because
131: % trajectories are all but essential in obtaining Lyapunov exponents), 
132: it may be expected that such environmental interaction, and the associated measurement process, will need to be included in a treatment 
133: that is adequate enough to predict the emergence of classical motion from
134: quantum mechanics. Indeed, recent work by a number of authors has 
135: made it increasingly clear that this provides a natural explanation
136: for the emergence of classical motion, and, therefore, a resolution of
137: the problem of the emergence of classical
138: chaos~\cite{Spiller,Graham,Percival,BHJ1,Milburn}.
139: Fortunately, the quantum theory of
140: environments and continuous measurement is now sufficiently well
141: developed that their effects can be treated in a fairly
142: straightforward manner, the emergence of classical dynamics verified,
143: and the mechanism of the quantum to classical transition elucidated.
144: 
145: Detailed studies of the QCT are particularly timely because current
146: experiments in quantum and atomic optics and condensed matter physics
147: are beginning to probe this transition directly in both ensemble and
148: individual system cases~\cite{expts}. Our approach here is to present
149: a general formalism for understanding the transition: more focused
150: analyses appropriate to specific experimental situations can easily be
151: developed based on this general approach.
152: 
153: In the following we examine the QCT in some detail.  In
154: Section~\ref{macobj}, we examine how macroscopic systems may be
155: treated, including environmental interactions and measurement.  In
156: Section~\ref{ineq}, we derive inequalities that describe the regime
157: under which classical motion emerges.  In Section~\ref{classest}, we
158: show that, in addition, classical state-estimation will 
159: work in the classical limit.  
160: In Section~\ref{numeg}, we provide two specific numerical
161: examples showing that classical motion is indeed obtained in the
162: regime predicted in Section~\ref{ineq}.  We finish with some
163: concluding remarks in Section~\ref{conc}.
164: 
165: \section{Describing the motion of macroscopic objects}\label{macobj}
166: A macroscopic object is composed of a very large number of quantum
167: degrees of freedom.  For example, we can consider the motions of the
168: atoms which comprise a massive object, and these are all coupled
169: together by the inter-atomic forces.  The equations of classical
170: mechanics are supposed to describe the dynamics of macroscopic
171: quantities, such as the center of mass; classical motion is not
172: observed in the entire many particle phase space.  Hence, one should
173: consider a change of variables, so as to write a Hamiltonian in terms
174: of the center-of-mass coordinate, $X$ (with conjugate momentum $P$).
175: This coordinate is coupled to all the other coordinates $x_i$ (in a
176: solid we might refer to these as the internal phonon modes, for
177: example).  Under the assumption that none of these environmental modes
178: is strongly perturbed by the dynamics, it is sufficient to treat them
179: as harmonic oscillators, and to couple them to the center-of-mass
180: motion via the Hamiltonian~\cite{CL,qnoise}
181: \begin{equation}
182:   H = H_{\mbox{\scriptsize cm}}(X,P) + \sum_{i} \left[ \frac{1}{2m_i}
183:           p_i^2 + \frac{\kappa_i}{2}(x_i - X)^2 \right]\, ,
184: \end{equation}
185: where $p_i$ are the momenta of the internal modes, and $\kappa_i$
186: gives the strength of the coupling between the center of mass and the
187: internal degrees of freedom.  The center of mass now constitutes
188: effectively an open system interacting with an environment consisting
189: of a large number of harmonic oscillators with a correspondingly large
190: range of frequencies.  This is the starting point for a treatment of
191: quantum Brownian motion, as developed by Caldeira and
192: Leggett~\cite{CL}.  Under general conditions, the phonon environment
193: can be treated as a heat bath, and in the double limit of weak
194: coupling to this bath and high temperature, it is possible to write a
195: very simple master equation for the center-of-mass
196: motion~\cite{qnoise}:
197: \begin{equation}
198:   \dot{\rho} = -\frac{i}{\hbar}[H_{\mbox{\scriptsize cm}}(X,P),\rho] -
199:                 k_{env} [X,[X,\rho]]\, ,
200: \label{nomeas}
201: \end{equation}
202: where $k_{env}$ is determined by the $\kappa_i$ and the temperature.
203: This provides a simple description of the effects of the internal
204: degrees of freedom upon the macroscopic motion of an object for which
205: frictional effects are negligible, and the heating due to the noise is
206: not significant over time scales of interest.  If one wanted to treat
207: damped classical systems, then one would relax the weak coupling
208: approximation so as to give a master equation that explicitly contains
209: damping.  However, for simplicity, we will restrict our attention here
210: to classical Hamiltonian systems.
211: 
212: Another important environment which we need to consider is the quantum
213: electromagnetic field.  This interacts with the object, and provides a
214: natural mechanism for measurement of the center-of-mass position $X$.
215: In general, macroscopic objects are bathed in light from all
216: directions, and the light that is reflected may be monitored by a
217: large number of observers.  Since we are considering a one-dimensional
218: system, and since we wish to use the simplest description which
219: captures the essential aspects of the measurement process, we restrict
220: ourselves to interaction with an electromagnetic field in one
221: dimension.  In particular, we consider a laser reflected from the
222: object such that the phase shift provides information about $X$.
223: Performing an analysis of such a measurement, one finds that the
224: evolution of the system, conditioned upon the measurement record, may
225: be written as a stochastic master equation in the It\^o
226: formalism~\cite{Itoref} as~\cite{afm,measx}
227: \begin{eqnarray}
228: d\rho &=& - \{ \frac{i}{\hbar} [H_{\mbox{\scriptsize cm}}(X,P),\rho] + k
229:                ([X^\dag X,\rho]_+ - 2 X \rho X^\dag) \} dt \nonumber\\ 
230:       && + \frac{\sqrt\kbar}2 \{ (X\rho + \rho X^\dag) - \rho \mathop{\rm Tr} 
231: 	\rho (X + X^\dag) \}  
232: 	dW \nonumber\\
233:       &=& - \{ \frac{i}{\hbar} [H_{\mbox{\scriptsize cm}}(X,P),\rho] + k
234: 	       [X, [X, \rho]] \} dt \nonumber\\
235:       && + \frac{\sqrt\kbar}2 \{ [X,\rho]_+ - 2 \rho \mathop{\rm Tr} \rho X \}
236:         dW \,,
237: \label{sme}
238: \end{eqnarray}
239: where the observed measurement record is given by
240: \begin{equation}
241:  dy = \mathop{\rm Tr} \rho X dt + \frac1{\sqrt\kbar} dW\, .
242: \label{measrec}
243: \end{equation}
244: In these equations, \(dW\) is a white noise generating a Wiener
245: process, $k$ gives the strength of the interaction between the light
246: and the object and is proportional to the power of the laser, whereas
247: \(\kbar\) gives the rate at which information about the system is
248: obtained.  When no information is obtained ({\it i.e.}, \(\kbar =
249: 0\)), or if the measurement record is averaged over, the stochastic
250: master equation (\ref{sme}) reduces to the ordinary master equation
251: (\ref{nomeas}).
252: 
253: The ratio \(\eta \equiv \kbar / 8 k\), called the efficiency of the
254: measurement~\cite{cqm4}, is a measure of the fraction of the reflected
255: light that is actually detected by the observer in making the
256: measurement.  As will be clear from our discussion in Sec.~\ref{ineq},
257: though both \(k\) and \(\kbar\) arise from the interaction of the
258: system with the measurement environment, they play very different
259: roles: whereas \(\hbar^2k\) represents a noise on the system that
260: leads to spreading out in phase space, \(\kbar\) provides information
261: about the system leading to localization around individual
262: trajectories.  The fact that Eq.~(\ref{sme}) leads to a (completely)
263: positive evolution for all initial conditions if and only if \(\eta
264: \le 1\)~\cite{Barchielli} is a particular case of the general
265: information-disturbance principles in quantum mechanics: any process
266: that leads to information about a system must produce at least a
267: minimal unavoidable disturbance.
268: 
269: If there exist multiple observers dividing the available reflected
270: light up among them, then each sees an evolution with a value of
271: $\eta_i < 1$ (and, for positivity, \(\sum_i \eta_i \le 1\)), with a
272: different noise realization for each observer~\cite{MOcomment}. This is
273: certainly the case in reality, where each observer usually captures
274: only a small fraction of the available light.  In the regime in which
275: classical motion is obtained (which we will refer to as the {\em
276: classical limit}), all observers must agree on the motion of the
277: system to within experimental error, and we consider this question at
278: the end of the next section, and in our numerical examples.
279: 
280: Since the form of the equation resulting from interaction with the
281: internal modes is the same as that which results from failing to
282: monitor the light which is being used to probe the system, we can take
283: this environment into account in the same way that we take multiple
284: observers into account, that is, by taking an appropriate value of
285: $\eta<1$.  (The measurement constant $k$ is then adjusted to include
286: the contribution from $k_{env}$).
287: 
288: The stochastic master equation~(\ref{sme}) constitutes our description
289: of the evolution of the center-of-mass of a macroscopic object.  In
290: the following we will show that this description, while very simple,
291: is sufficiently realistic to obtain the correct classical motion in
292: the classical limit.  It should also be noted that while we have
293: chosen to measure the position $X$, the analysis which follows
294: suggests that the extraction of the classical limit is not sensitive
295: to the precise observables which are measured; as long as the
296: measurement provides sufficient information about the location of the
297: system in phase space, the classical limit will be obtained.  For
298: example, a continuous measurement of momentum will suffice, as long as
299: the forces on the system are spatially dependent.  In fact, other
300: authors have provided numerical support for this view by showing that
301: quantum state diffusion (using a measurement interaction which
302: includes damping)~\cite{Spiller} or a simultaneous measurement of
303: position and momentum~\cite{Milburn} are sufficient to induce the QCT
304: in the same manner.
305: 
306: If the correct classical mechanics is to be obtained, two conditions need
307: to be satisfied.  First, it must be possible to observe the system so
308: that its center of mass (and all other degrees of freedom considered
309: classical) is known sufficiently accurately on the scale of the
310: potential and relevant dynamical timescales.  Second, these
311: observed values, which we might identify as noisy counterparts of the
312: means of the sufficiently well localized distribution, $x\equiv\langle
313: X\rangle$ and $p\equiv\langle P\rangle$, should evolve according to
314: the classical Hamiltonian $H_{\mbox{\scriptsize cm}}(x,p)$ which has
315: the same functional form as the quantum Hamiltonian
316: $H_{\mbox{\scriptsize cm}}(X,P)$, with deviations small compared to
317: the classical scales.  In other words, the existence of the quantum to
318: classical transition implies that in the classical limit we can
319: replace the quantum operators with effective classical dynamical
320: variables,
321: \begin{equation}
322:    H(X,P) \rightarrow H(x,p)\, .
323: \end{equation}
324: 
325: 
326: \section{Inequalities governing the classical limit}\label{ineq}
327: We now ask for the parameter regime in which the evolution reduces to
328: classical motion.  As explained in the previous section this means
329: that the quantum distribution remains sufficiently localized (such
330: that the system can be said to be executing a trajectory), and that
331: this trajectory, characterized by $x\equiv\langle X\rangle$ and
332: $p\equiv\langle P\rangle$, follows that of the classical motion,
333: generated by $H_{\mbox{\scriptsize cm}}(x,p)$.
334: 
335: We proceed by first writing down the equations of motion for the first
336: and second moments of $X$ and $P$.  From Eq.~(\ref{sme}) these become
337: \begin{eqnarray}
338:    dx & = & \frac{p}{m} dt + \sqrt{\kbar} V_x dW \,, \label{eqx} \\
339:    dp & = & \langle F(X)\rangle dt + \sqrt{\kbar} C_{xp} dW \,,
340:             \label{eqp} 
341: \end{eqnarray}
342: and
343: \begin{eqnarray}
344:    dV_x    & = & \left[ \frac{2}{m} C_{xp} - \kbar V_x^2 \right] dt +
345:                  \sqrt{\kbar} K_{xxx} dW\,, \label{c1} \\ 
346:    dV_p    & = & \left[ 2\hbar^2 k - \kbar C_{xp}^2 + 2\partial_x F
347:                  C_{xp} \right] dt \nonumber  \\
348:            &   & + \partial_x^2 F K_{xxp} dt + \sqrt{\kbar} K_{xxp}
349:                    dW\,, \label{c2} \\  
350:    dC_{xp} & = & \left[ \frac{1}{m} V_p - \kbar V_xC_{xp} +
351:                            \partial_x F V_x \right] dt \nonumber \\  
352:            &   & + \frac{1}{2}\partial_x^2 F K_{xxx} dt +
353:                    \sqrt{\kbar} K_{xpp} dW \label{c3}\,, 
354: \end{eqnarray}
355: where
356: \begin{eqnarray}
357:    V_x    & = & \langle X^2 \rangle - \langle X\rangle \,, \\
358:    V_p    & = & \langle P^2 \rangle - \langle P\rangle \,, \\
359:    C_{xp} & = & \frac{1}{2}\langle XP + PX \rangle - \langle X\rangle
360:                 \langle P\rangle \,, 
361: \end{eqnarray}
362: are the second cumulants, and the $K$'s are the third cumulants defined by
363: \begin{eqnarray}
364:    K_{abc} &=& \langle :ABC: \rangle - \langle :AB: \rangle 
365:              \langle C \rangle\nonumber - 
366:              \langle A \rangle \langle :BC: \rangle \\
367:    &&\qquad\qquad {} 
368: 	     - \langle :AC: \rangle \langle B \rangle
369:              + 2 \langle A \rangle \langle B \rangle \langle C \rangle \nonumber\,,
370: \end{eqnarray}
371: where $A,B,C$ can be $X$ or $P$, and
372: the colons denote Weyl ordering of the operator products.  In the
373: above equations we use the simplified notation $F \equiv F(x) =
374: F(\langle X\rangle)$ and expand $F$ in a Taylor series about $X=x$
375: truncated to second order.  Without this truncation higher derivatives
376: of $F$ would appear in the equations for $V_p$ and $C_{xp}$,
377: multiplied by higher powers of the widths or by higher cumulants.
378: Truncating the power series for $F$ in this way is a good
379: approximation so long as the distribution is sufficiently localized
380: about $x$ and $p$.  Examining Eq.~(\ref{eqx}), one sees that to
381: maintain classical motion for $x$ one needs $\langle F(X) \rangle
382: \approx F(x)$, which happens when the system is localized enough so
383: that $V_x \partial_x F(x) \ll 2F(x)$.  It is the task of the
384: measurement to maintain such localization, and numerical
385: studies~\cite{BHJ1} have shown that it can indeed do so.
386: 
387: At this point, it is perhaps instructive to look at the origin of
388: localization of the individual trajectories.  The density matrix
389: obtained by solving Eq.~(\ref{sme}) is conditioned on the measurement
390: record Eq.~(\ref{measrec}), or equivalently, by the noise realization
391: \(dW\).  Averaging over these realizations results in the density
392: matrix of the unobserved system which can also be obtained by solving
393: Eq.~(\ref{nomeas}).  The second cumulants \(\sigma^2_{xx}\),
394: \(\sigma^2_{xp}\), and \(\sigma^2_{px}\) of that distribution are
395: related to the corresponding cumulants for each trajectory by the
396: relations:
397: \begin{eqnarray}
398:    \sigma^2_{xx} &=& \langle V_x \rangle_W + \mathop{\rm var}\nolimits_W
399:                         (x,x)\,, \nonumber\\
400:    \sigma^2_{xp} &=& \langle C_{xp} \rangle_W + \mathop{\rm var}\nolimits_W
401: 			(x,p)\,, \nonumber\\
402:    \sigma^2_{pp} &=& \langle V_p \rangle_W + \mathop{\rm var}\nolimits_W
403: 			(p,p)\,,
404: \end{eqnarray}
405: where \(\langle \cdot \rangle_W\) and \(\mathop{\rm
406: var}\nolimits_W(\cdot,\cdot)\) represent respectively the means and
407: (co-\nobreak)variances of the quantities, when considered as
408: distributions over trajectories.  The Wiener process damps the first
409: term on the right hand side of each equation by a term proportional to
410: \(-\kbar\), at the same time compensating this with a growth of the
411: last term in each equation.  As discussed in Sec.~\ref{classest}, this
412: is precisely the way in which classical measurement also selects well
413: defined trajectories out of an ensemble spreading out in phase space.
414: 
415: At the level of these second cumulants, the exact form of the damping
416: is, thus, immaterial.  The fact, however, that we derived these from
417: the stochastic master equation~(\ref{sme}) gives us not only a
418: theoretical `unraveling' of the master equation, but also provides a
419: physical meaning to each trajectory.  Furthermore, it guarantees that
420: the underlying evolutions of density functions are completely
421: positive, and, therefore, not only does the covariance matrix stay
422: positive, but also these equations can be completed into a hierarchy
423: of cumulant equations that automatically satisfy the appropriate
424: reality conditions.  Furthermore, as discussed in Sec.~\ref{macobj},
425: such a measurement process leads to the unavoidable noise proportional
426: to \(\hbar^2 k\) apparent on the right hand side of Eq.~(\ref{c2}); in
427: contrast, in the classical discussion in Sec.~\ref{classest}, the
428: corresponding noise term is unrelated to the measurement process and
429: can even be set to zero.  In fact, the truncation to the second
430: cumulants implies that no truly quantum effects of dynamics come into
431: play~\cite{habib} in our approximation, and the quantum scale
432: \(\hbar\) appears in our equations purely from this
433: information-disturbance consideration~\cite{footnote1}.
434: 
435: We will make two self consistent approximations in order to examine in
436: what regime classical dynamics emerges.  The first is to truncate the
437: power series in $F$ to second order.  The second is to neglect third
438: and higher cumulants in the equations for the second cumulants.  An
439: examination of the equations of motion for the third cumulants
440: appearing in Eqs.~(\ref{c1}) to (\ref{c3}) shows that indeed these are
441: damped by the measurement, again with damping coefficients
442: proportional to $\kbar$.  The fact that the wavefunction stays close
443: to Gaussian is also borne out by numerical simulations~\cite{BHJ1}.
444: 
445: Setting the third cumulants to zero in the equations for the second
446: cumulants, we solve for the stable steady-state:
447: \begin{eqnarray}
448:     V_x^{\mbox{\scriptsize ss}} & = & \sqrt{\frac{2
449:                 C_{xp}^{\mbox{\scriptsize ss}}}{m\kbar}}\, , \label{Vxss} \\ 
450:    V_p^{\mbox{\scriptsize ss}} & = & mV_x^{\mbox{\scriptsize ss}} (
451:                 \kbar C_{xp}^{\mbox{\scriptsize ss}} - \partial_x F)
452:                 \,, \\ 
453:     C_{xp}^{\mbox{\scriptsize ss}} & = & \frac{\partial_x F}{\kbar}
454:                 + \mbox{sgn}(m) \sqrt{\left( \frac{\partial_x
455:                 F}{\kbar} 
456: 	           \right)^2  +
457:                 \frac{\hbar^2}{4\eta}} \,. \label{Cxpss}
458: \end{eqnarray}
459: where $\mbox{sgn}(m)$ is the sign of $m$ which we shall henceforth
460: take to be positive~\cite{footnote2} and $\partial_x F$ is taken to be
461: evaluated at a typical point in phase space.  Now, there are three
462: conditions that must be satisfied in order for the classical limit to
463: be obtained.  First, localization such that $V_x \partial_x F(x) \ll
464: 2F(x)$, as discussed above, must be maintained, second, the noise
465: introduced by the measurement should be negligible compared to the
466: classical motion, and third, that the measurement record should follow
467: the motion of the position with sufficient accuracy.
468: 
469: Before examining these conditions in turn, two points are in
470: order.  First, localization and low noise are not really independent
471: constraints: in fact to provide effective damping for the covariance
472: matrix, the noise has to increase with increasing width of the state.
473: Conversely, noise also effects a spread in phase space of any
474: uncertain state, especially near unstable points.  It is convenient,
475: however, to treat the direct effect of the finite width of the state
476: on the deterministic evolution in a nonlinear potential as a question
477: of localization, and the rest as a question of low noise on the
478: trajectories.
479: 
480: As a second point, it is important to emphasize that the deviations of
481: the quantum trajectories from the classical ones can be different in
482: different parts of the phase space.  Nevertheless, in most
483: experimental situations, `classical' quantities evaluated are of
484: similar orders of magnitude almost everywhere, and, so, we shall
485: ignore these differences and consider them evaluated at a `typical'
486: point around the trajectory in question.
487: 
488: \subsection{Localization}
489: We start by noting that for the deterministic part of the equations of
490: motion for the quantum mean values $x$ and $p$ to match the classical
491: equations of motion, we need
492: \begin{equation}
493:     \langle F(X) \rangle = F(x) + \frac{1}{2}V_x \partial^2_x F(x) +
494:     \ldots
495: \end{equation}
496: to very closely approximate $F(x)$.  That is, we need
497: \begin{equation}
498:     r \equiv \left| \frac{\partial_x^2 F V_x}{2F} \right| \ll 1 \,.
499: \label{smallr}
500: \end{equation}
501: Replacing $V_x$ with its typical steady state value
502: [Eqs.~(\ref{Vxss},\ref{Cxpss})], we see that $r$ is the positive
503: solution to
504: \begin{equation}
505:     \frac{2m\kbar^2 F^2}{(\partial_x^2 F)^2} r^2 - (\partial_x F) =
506:         \sqrt{(\partial_x F)^2 + \frac{\hbar^2 \kbar^2}{4 \eta}} \,.  
507:      \label{loc1}
508: \end{equation}
509: Since this equation implies that $r$ is a monotonically decreasing
510: function of $\kbar$, Eq.~(\ref{smallr}) can provide a lower limit for
511: $\kbar$.  We examine this possibility in the following discussion.
512: 
513: Due to the positivity of its right hand side,
514: Eq.~(\ref{loc1}) implies that at the unstable point $\partial_x F >
515: 0$, we must have  
516: \begin{equation}
517:    2 m \kbar^2 F^2 r^2 > (\partial_x^2 F)^2 \partial_x F\, .
518: \end{equation}
519: This alone means that to have $r\ll1$, it is necessary that
520: \begin{equation}
521:    \kbar^2 \gg \frac{(\partial_x^2 F)^2 |\partial_x F|}{mF^2} \,.
522:       \label{ineq13}
523: \end{equation}
524: 
525: Squaring Eq.~(\ref{loc1}) one sees that $r$ is the algebraically
526: largest solution of
527: \begin{equation}
528:    \kbar^2 = \frac{(\partial_x^2 F)^2}{mF^2 r^4} \left(
529:                \frac{(\partial_x^2 F)^2 \hbar^2}{16\eta m F^2} +
530:                \partial_x F r^2 \right) \,.  
531: \end{equation}
532: For this solution to be small would generically require 
533: \begin{equation}
534:    \kbar \gg \frac{(\partial_x^2 F)^2 \hbar}{4\sqrt{\eta} m F^2}
535:        \,, \label{ineq14} 
536: \end{equation}
537: except in the typical case of small nonlinearity
538: \begin{equation}
539:   (\partial_x^2 F)^2 \ll \frac{16 \eta m F^2|\partial_x F|}{\hbar^2}
540:             \, , \label{locineq1}
541: \end{equation}
542: when the width stays small at the stable points independent of the
543: value of $k$.  
544: 
545: Since, when the nonlinearity is large enough to violate
546: Eq.~(\ref{locineq1}), Eq.~(\ref{ineq14}) is stronger than
547: Eq.~(\ref{ineq13}), we can summarize these results as follows: If the
548: nonlinearity, characterized by $\partial_x^2 F$, is sufficiently weak
549: to satisfy Eq.~(\ref{locineq1}), then, at the unstable points
550: ($\partial_x F > 0$) one needs
551: \begin{equation}
552:    8\eta k \gg \sqrt{\frac{(\partial_x^2 F)^2 |\partial_x
553:              F|}{2mF^2}} \,.  \label{locineq2}
554: \end{equation}
555: In the case of strong nonlinearity, we need
556: \begin{equation}
557:    8\eta k \gg \frac{(\partial_x^2 F)^2
558:       \hbar}{4\sqrt{\eta} m F^2}  \,,  \label{locineq3}
559: \end{equation}
560: to hold at all points.
561: 
562: \subsection{Low Noise}
563: In this subsection we consider the noise component of these equations.
564: In the classical limit the effect of this noise must be negligible on
565: the scale of the deterministic dynamics.  To compare the random noise
566: with the deterministic dynamics, we need to average over an
567: appropriate time scale: during a time $T$ upon which the dynamics is
568: effectively linear, the noise $dW$ provides an {\em rms} contribution
569: of $\sqrt{T}$.  We will define `low noise' to mean that the noise
570: contribution on this time scale is small compared to the deterministic
571: contribution.  The time scales upon which the dynamics is linear for
572: $x$ and $p$ are those in which the terms in the respective equations
573: do not change appreciably, and we will use the deterministic terms to
574: obtain these time scales.  The deterministic motion for $x$ is driven
575: by $p/m$, so appreciable changes occur when the change in momentum,
576: $\Delta p$, is of the order of $p$.  The resulting time scale for this
577: change is $T_x \sim |p/F|$.  The deterministic motion for $p$ is
578: driven by $\langle F(X)\rangle \approx F$.  Changes in $F$ are due to
579: changes in $x$: in particular $\Delta F \approx \partial_x F\Delta x$.
580: Hence the time scale for changes in $F$ is $T_p \sim
581: (m|F|)/(|p\partial_x F|)$.  Demanding that the change in $x$ and $p$
582: from the noise is small compared to that due to the deterministic
583: motion in these time intervals gives the two inequalities
584: \begin{eqnarray}
585:   \sqrt{\kbar} V_x & \ll & \frac{p}{m} \sqrt{T_x} =
586:                \sqrt{\frac{|p^3|}{m|F|}} \label{ncon1} \,, \\
587:   \sqrt{\kbar} C_{xp} & \ll & F \sqrt{T_p} = \sqrt{\frac{m|F^3|}{|p
588:                \partial_x F|}}\,.  \label{ncon2} 
589: \end{eqnarray} 
590: We will now examine these two inequalities in turn.
591: 
592: Considering the first inequality, and replacing $V_x$ with its typical
593: steady state value, given above, we have
594: \begin{equation}
595:    C_{xp}^{\mbox{\scriptsize ss}} \ll \frac{E|p|}{|F|} \, ,
596: \end{equation}
597: where $E = p^2/(2m)$ is the typical energy of the system.  Noting that
598: $E|p|/|F|$ has units of action, to simplify the following analysis we
599: will now define a dimensionless action $s$ by $s\hbar \equiv
600: E|p|/(4|F|)$.  Conceptually $s$ may be identified with the typical
601: action of the system in units of $\hbar$.  Using Eq.~(\ref{c3}) to
602: write $C_{xp}^{\mbox{\scriptsize ss}}$ in terms of $\kbar$ the
603: inequality becomes
604: \begin{equation}
605:    \sqrt{\left( \frac{\partial_x F}{\kbar} \right) +
606:       \frac{\hbar^2}{4\eta}} \ll 4 \hbar s - \frac{\partial_x
607:       F}{\kbar} \,.  \label{ineq1}
608: \end{equation}
609: The positivity of the left hand side immediately gives us the condition
610: \begin{equation}
611:     \frac{\partial_x F}{\hbar \kbar} < 4s \,.  \label{ineq6} 
612: \end{equation}
613: Now squaring both sides of Eq.~(\ref{ineq1}), and rearranging, we
614: obtain 
615: \begin{equation}
616:     \frac{\hbar^2}{4\eta} \ll 16\hbar^2 s^2 - \frac{8\hbar s
617:               \partial_x F}{\kbar} \,.  \label{ineq2}
618: \end{equation}
619: Because of Eq.~(\ref{ineq6}), this condition reduces to
620: \begin{equation}
621:    s \gg \frac{1}{8\sqrt{\eta}} \,, \label{ineq3}
622: \end{equation}
623: except at the unstable points ($\partial_x F > 0$) where one requires,
624: in addition, 
625: \begin{equation}
626:    \frac{\hbar\kbar}{\partial_x F} \gg \frac{32\eta s}{64\eta s^2 -
627:                1} \approx \frac{1}{2s} \,, \label{ineq8}
628: \end{equation}
629: where the approximate equality is implied by the inequality in
630: Eq.~(\ref{ineq3}).  
631: 
632: We now consider the inequality given by Eq.~(\ref{ncon2}). Replacing
633: $C_{xp}$ with its typical steady state value, and performing some
634: rearrangements we obtain
635: \begin{equation}
636:    \xi \ll 4\eta \left( s' - \mbox{sgn}(\partial_x F)
637:           \sqrt{\frac{4s'}{\xi}} \right) \,, \label{ineq4}
638: \end{equation}
639: where for compactness we have written
640: \begin{eqnarray}
641:    \xi &\equiv& \frac{\hbar\kbar}{|\partial_x F|}\nonumber\\
642:    \hbar s' &\equiv& \frac{mF^2|F|}{(\partial_x F)^2 |p|}\,.
643: \end{eqnarray}
644: Here $s'$ is a dimensionless quantity, which we will once again take
645: to be an estimate of the typical action of the system in units of
646: $\hbar$.  For $\partial_x F > 0$, the condition
647: \begin{equation}
648:       \xi \ll 4\eta \left( s' - \sqrt{\frac{4s'}{\xi}} \right)\, ,
649:       \label{ineq5}
650: \end{equation}
651: is satisfied whenever
652: \begin{equation}
653:    \frac{16}{s'} \ll \xi \ll 2\eta s' \, ,
654:      \label{ineq9}
655: \end{equation}
656: whereas, for $\partial_x F < 0$ (the condition is not useful when
657: $\partial_x F = 0$), it is sufficient that
658: \begin{equation}
659:    \frac{\hbar\kbar}{|\partial_x F|} \ll 4 \eta s' \,.
660:       \label{ineq10}
661: \end{equation}
662: Collecting all the inequalities in this subsection, ({\it i.e.} 
663: Eqs.~(\ref{ineq6}), (\ref{ineq3}), (\ref{ineq8}),
664: (\ref{ineq9}) and (\ref{ineq10})), we find that they are all
665: implied by
666: \begin{equation}
667: %  \fbox{ $
668: 	\frac{ 2|\partial_x F| }{ \eta \bar s } \ll \hbar k \ll 
669:    	    \frac{ |\partial_x F|\bar{s} } {4}
670: %	 $ }
671:    \,,
672:   \label{lnineq}
673: \end{equation}
674: where $\bar{s}\equiv\mbox{min}(s,s')$.  
675: %This condition demonstrates
676: %that as the action of the system increases with respect to $\hbar$,
677: %the accuracy or strength of the measurement, as determined by $k$, can
678: %take an increasingly wide range of values.  For everyday macroscopic
679: %objects, in which $s$ is very large indeed, $k$ is able to vary over a
680: %very large range of values and still extract classical behavior.
681: 
682: \subsection{Faithful Tracking}
683: In the previous subsections we have been considering the motion of the
684: centroid of the quantum wave packet, $(x,p)$.  This centroid
685: represents the observer's true best-estimate of the mean value of
686: position and momentum at the current time, given the measurement
687: record.  To obtain this best estimate the observer must know the
688: dynamics of the system, given by $H_{\mbox{\scriptsize cm}}$, and then
689: integrate the full stochastic master equation, where the correct $dW$
690: is obtained continuously from the measurement record.
691: 
692: In practice, it is often merely the measured value of position which
693: is taken as the estimated value.  Hence, we need to find conditions
694: under which this value tracks the true best-estimate with sufficient
695: accuracy.  Since the measurement record in our formulation contains
696: white noise, the simplest way to model a realistic macroscopic
697: measuring apparatus is to low-pass filter, or {\em band limit} the
698: measurement record to obtain the continuous estimate of the position
699: (this is equivalent to making the reasonable assumption that all real
700: measuring devices have a finite response time).  This is achieved by
701: averaging the measurement record over some finite time $\Delta t$.  To
702: obtain an accurate estimate, $\Delta t$ must be short compared to the
703: dynamical time scale of the system.
704: 
705: If we assume that the change in $x$ over time $\Delta t$ is
706: negligible, then the error in the estimate of $x$ resulting from
707: averaging the measurement record $y(t)$ over $\Delta t$ is
708: \begin{equation}
709:    \sigma_{T}(x) = (8\eta k \Delta t)^{-1/2} \,.
710: \end{equation}
711: Hence, if to accurately track a classical dynamical system we require
712: a spatial resolution of $\Delta x$, and a temporal resolution of
713: $\Delta t$, then we must have
714: \begin{equation} 
715: %   \fbox{ $
716:   8 \eta k \ge \frac{1}{\Delta t (\Delta x)^2} 
717: % $ } 
718: \,.
719: \end{equation}
720: We also note, however, that in the observation of classical systems,
721: classical estimation theory is, in fact, often used to obtain the
722: classical equivalent of the quantum best-estimates provided by the SME
723: (Eq.~(\ref{sme})).  Such a procedure is most often used in classical
724: feedback-control applications.  In the classical limit, therefore,
725: such classical estimation procedures must work effectively, and we
726: will verify this in the next section.
727: 
728: \subsection{Summary}
729: We have now derived a set of inequalities which, when satisfied, lead
730: to the emergence of classical mechanics.  Consider first the
731: inequalities which come from the localization condition.  In the
732: macroscopic regime, which applies to common mechanical devices one
733: would build in the laboratory, the right hand side of inequality
734: (\ref{locineq1}) is extremely large compared to the typical
735: nonlinearity.  Consequently this inequality is satisfied, and the
736: resulting condition for $k$ is given by (\ref{locineq2}).  Note that
737: $\hbar$ does not appear in this inequality.  In fact, this is actually
738: a classical inequality, similarly required for classical continuous
739: measurement on classical systems.  In that case, the observer's state
740: of knowledge of the system is given by a classical probability density
741: in phase space, and this evolves as the system evolves and as
742: information is continuously obtained.
743: 
744: If the system is sufficiently small, and the nonlinearity
745: sufficiently large on the quantum scale so that inequality
746: (\ref{locineq1}) is not satisfied, then the condition for $k$ is
747: replaced by inequality (\ref{locineq3}).  This does contain $\hbar$,
748: and is, therefore, a uniquely quantum condition.  It appears due to
749: the unavoidable quantum noise which affects the dynamics strongly if the
750: nonlinearity is large on the quantum scale.
751: 
752: The left inequality in Eq.~(\ref{lnineq}), again is a classical
753: condition (the $\hbar$ arises because we chose to measure the action
754: $\bar s$ in units of $\hbar$): it reflects the observation that if the
755: measurement does not localize the motion, the state estimate changes
756: from moment to moment essentially randomly, or in other words, the
757: noise is large.  The right hand inequality in Eq.~(\ref{lnineq}) is
758: the direct effect of the irreducible noise coming from the measurement
759: process and is thus a quantum effect.  Together, as the action
760: increases, these low noise conditions put ever decreasing constraints
761: on the required measurement strength.
762: 
763: The faithful tracking condition is once again purely classical, in
764: that it also applies to classical observation.  It is simply the
765: condition on the accuracy of the measurement so that the measurement
766: record itself, as opposed to the estimated state, accurately tracks
767: the motion of the system from which the localization condition is
768: derived.
769: 
770: It is worth noting that the above inequalities also determine the
771: regime in which multiple observers agree on the motion of an object,
772: which is clearly an important property of the classical limit.  As
773: discussed in Section~\ref{macobj}, multiple observers can be taken into
774: account by giving each observer, $i$, a value of $\eta = \eta_i$ such
775: that $\sum_i \eta_i \le 1$, and giving each a different noise
776: realization, $dW_i$.  Furthermore, it is clear from the derivation of
777: the stochastic master equation~\cite{Barch93,DDZ} that the state
778: conditioned by the measurements made by all of the observers is
779: narrower than and consistent (in probability) with the state estimate
780: of each observer; {\it ipso facto}, the estimates of the different
781: observers must agree within errors.  Since the conditions derived in
782: this section can be satisfied with $\eta < 1$ (even with $\eta \ll
783: 1$), and since these imply localization and accurate tracking of the
784: measurement record, under these conditions all observes will agree
785: upon the motion of the system to errors small on the classical scale.
786: 
787: \section{Classical Estimation in the Classical Limit}\label{classest}
788: When a classical system is subject to noise and continuous
789: observation, a classical theory of continuous state-estimation may be
790: developed to describe the continuous acquisition of information
791: regarding the system~\cite{Maybeck}.  Consider an observed classical
792: system whose dynamics is given by
793: \begin{equation}
794:     \left( \begin{array}{c} dx \\ dp \end{array} \right) = \left(
795:     \begin{array}{c} p/m \\ F_{\mbox{\scriptsize c}}(x) \end{array}
796:     \right) dt + \left( \begin{array}{c} 0 \\ \sqrt{2 g_p} \; dW_p
797:     \end{array} \right) 
798: \end{equation}
799: with measurement record 
800: \begin{equation}
801:     dy_{\mbox{\scriptsize c}} = x dt + \frac{dV}{\sqrt{g_m}} \,,
802: \end{equation}
803: {\it i.e.}, we consider a system with purely additive momentum noise
804: being observed continuously and with random errors. Here, $dW_p$ and
805: $dV$ are Wiener noises with $dV$ possibly correlated with the $dW_p$,
806: and $g_p$ and $g_m$ are positive real numbers.  Then the evolution of
807: the state of knowledge of the observer, described by a probability
808: density $P(x,p,t)$ obtained by averaging over $dW_p$ and conditioning
809: by $dy_c$, is~\cite{Maybeck,DHJMT}
810: \begin{eqnarray}
811:    dP & = & \left[ -(p/m)\partial_x - (F_{\mbox{\scriptsize c}} - g_p
812:             \partial_p)\partial_p \right] P dt \nonumber \\ 
813:       &   & + \sqrt{g_m} (x - \langle x \rangle) P dW \,,
814: \end{eqnarray}
815: where $dW = \sqrt{g_m}(x - \langle x \rangle) dt + dV$, and turns out
816: to be a Wiener noise, uncorrelated with the conditional probability
817: $P$.  Note that we can then write the measurement record for the
818: classical measurement as 
819: \begin{equation}
820:    dy_{\mbox{\scriptsize c}} = \langle x \rangle dt +
821:                  \frac{dW}{\sqrt{g_m}} \,, 
822: \end{equation}
823: and we see that this can be viewed as directly analogous to the
824: quantum measurement record.  The equations of motion for the classical
825: best estimates $\langle x \rangle_{\mbox{\scriptsize c}}$ and $\langle
826: p \rangle_{\mbox{\scriptsize c}}$, and the second order moments are 
827: \begin{eqnarray}
828:    d \langle x\rangle_{\mbox{\scriptsize c}} & = & \frac{\langle
829:                  p\rangle_{\mbox{\scriptsize c}}}{m} dt + \sqrt{g_m}
830:                  V_x dW \,, \label{eqxc} \\ 
831:    d\langle p\rangle_{\mbox{\scriptsize c}} & = & \langle
832:                  F_{\mbox{\scriptsize c}}(X)\rangle dt + \sqrt{g_m}
833:                  C_{xp} dW \,, \label{eqpc} 
834: \end{eqnarray}
835: and
836: \begin{eqnarray}
837:    dV_x    & = & \left[ \frac{2}{m}C_{xp} - g_m V_x^2 \right] dt +
838:                  \sqrt{g_m} K_{xxx} dW\,, \label{cc1} \\ 
839:    dV_p    & = & \left[ 2 g_p - g_m C_{xp}^2 + 2\partial_x F C_{xp}
840:                  \right] dt \nonumber  \\
841:            &   & + \partial_x^2 F K_{xxp} dt + \sqrt{g_m} K_{xpp} dW
842:                   \,, \label{cc2} \\  
843:   dC_{xp}  & = & \left[ \frac{1}{m}V_p - g_m V_xC_{xp} + \partial_x F
844:                  V_x \right] dt \nonumber \\ 
845:            &   & + \frac{1}{2}\partial_x^2 F K_{xxx} dt + \sqrt{g_m}
846:                  K_{xpp} dW \,.  \label{cc3} 
847: \end{eqnarray}
848: Identifying $g_m = \kbar$ and $g_p = \hbar^2 k$, we see that these
849: equations are identical to the quantum equations governing the
850: continuously estimated state [Eqs.~(\ref{eqx})-(\ref{c3})], and the
851: only way that quantum mechanics enters is in enforcing $\hbar^2 g_m
852: \leq 8 g_p$. Even though this is the case, it should be noted that
853: when the potential is nonlinear, the equations of motion for the
854: third and higher cumulants are not the same in the quantum and
855: classical cases, so in general the evolutions of the classical and
856: quantum estimates differ.  In the classical limit, however, the
857: conditional probability, or the state in the quantum case, is Gaussian
858: to a very good approximation so that the third cumulants can be set to
859: zero, and as a result they no longer feed into the equations for the
860: second order cumulants.  Consequently, the evolution of the classical
861: best estimates and second cumulants are identical to the quantum
862: estimates for the same measurement record, and as a result classical
863: estimation may be used to track dynamical systems in the classical
864: limit.
865: 
866: \section{Numerical examples}\label{numeg}
867: In this section we provide numerical support for the arguments in the
868: previous section.  We present two examples, and show that under the
869: conditions derived in the previous sections, the quantum wave packet
870: remains localized, the evolution of the centroid follows the classical
871: motion with negligible noise, and both the measurement record
872: (suitably band limited) and the classical state-estimate accurately
873: track the motion of the system for each of a set of observers.
874: 
875: To derive the equation of motion for the wavefunction of the
876: continuously observed system, assuming $N$ observers, one can first 
877: write down the Stochastic Schr\"odinger equation for the unormalised 
878: wavefunction for a single observer, making $N$ measurements. If the 
879: interaction strength for measurement $i$ is $\eta_i k$, then this is
880: \begin{eqnarray}
881:    d|\psi\rangle & = & \left[ -\frac{1}{\hbar}(iH(t) + \hbar k X^2) dt
882:                         \right.  \nonumber \\  
883:                  &   & \left.  + \sum_{i=1}^N 4 \eta_i k dr_i \right]
884:                         \vert\psi\rangle \,, 
885:    \label{dpsiN}
886: \end{eqnarray}
887: where the record for each measurement is given by 
888: \begin{equation}
889:   dr_i = \langle X\rangle dt + \frac{dW_i}{\sqrt{8\eta_i k}} .
890: \end{equation}
891: Now we let each observer have access to just one of the measurement 
892: records. In addition, we choose $\sum_i\eta_i = 1$, so that $\eta_i$ 
893: represents the fraction of the total measurement interaction strength 
894: $k$ used by each observer.  The evolution 
895: of the state-of-knowledge for any particular observer (who only has 
896: access to her measurement record) can be calculated by averaging over 
897: the noise realizations for all the other observers while keeping the 
898: measurement record for the observer in question fixed. The resulting 
899: equation of motion for the state-of-knowledge of observer $i$, given 
900: the measurement record $dr_i$ generated by the stochastic Schr\"odinger 
901: equation (\ref{dpsiN}) is the stochastic master 
902: equation~\cite{Barch93,DDZ}
903: \begin{eqnarray}
904:   d\rho & = & -\frac{i}{\hbar}[H,\rho] dt - k [X,[X,\rho]] dt
905:         \nonumber \\ & & + ( [X,\rho]_+ - 2 \rho \mathop{\rm Tr}[\rho
906:         X]) \sqrt{2\eta_i k} dV_i \label{smemult}
907: \end{eqnarray}
908: where
909: \begin{equation}
910:   dV_i = \sqrt{8\eta_i k}(dr_i - \mathop{\rm Tr}[\rho X] dt) .
911: \end{equation}
912: Note that this is, in fact, just Eq.(\ref{sme}), because as far as 
913: observer $i$ is concerned, all the other observers are simply gathering 
914: part of the environment to which $i$ has no access. In addition, 
915: the fractions $\eta_i$ are the respective measurement efficiencies. 
916: %Thus, multiple simultaneous measurements, measurements by multiple 
917: %observers, and inefficient detection are all essentially the same 
918: %thing.
919: 
920: To simulate multiple observations on a given system, we first
921: integrate the stochastic Schr\"odinger equation which generates a set
922: of measurement records, one for each observer.  We then integrate the
923: corresponding stochastic master equations using the measurement record
924: for each observer.  The state-of-knowledge of each observer over time
925: can then be compared to the `actual' evolution of the system state
926: vector given by Eq.~(\ref{dpsiN}). The stochastic Schr\"odinger and
927: master equations were integrated in time using a spectral
928: split-operator method.  Since the classical limit is obtained when the
929: extent of the wavefunction is small compared to the range of motion of
930: the centroid, the algorithm is designed so that the computational grid
931: follows the wavefunction in both position and momentum space, and this
932: is crucial for efficient computation.
933: 
934: In treating systems of different sizes and actions it is convenient to
935: choose units for the system variables to keep the numerical value of
936: the action close to unity.  Due to this system dependent choice of
937: units, the fixed quantity $\hbar$ has a system dependent numerical
938: value; and indeed we expect the classical limit when $\hbar \ll 1$ in
939: these units.  This is what we demonstrate below.
940: 
941: \begin{figure}
942: \leavevmode\includegraphics[width=0.9\hsize]{duffV}
943: \caption{The standard deviations of the state estimates for each of
944: the three observers for the Duffing oscillator, plotted over a
945: duration of $t=5$.  Solid line: observer with $\eta=0.5$; dashed line:
946: observer with $\eta=0.3$; dash-dot line: observer with $\eta=0.2$.}
947: \label{fig1}
948: \end{figure}
949: 
950: \subsection{The Duffing oscillator}
951: The Duffing oscillator is a sinusoidally driven double-well potential,
952: with Hamiltonian
953: \begin{equation}
954:    H(t) = \frac{P^2}{2m} + BX^4 - AX^2 + \Lambda X \cos(\omega t) \,.
955: \end{equation}
956: We choose $m=1$, $A=10$, $B=0.5$, $\Lambda=10$ and $\omega=6.07$.  At
957: times when the driving is zero, this puts the minima of the two
958: potential wells at $\sim\pm 3.2$, with a central barrier height of
959: $50$.  We choose $\hbar=10^{-5}$ and $k=10^5$, which is sufficient to
960: satisfy the inequalities derived in Section~\ref{ineq} for all but
961: tiny values of $\eta$, and therefore puts the system in the classical
962: regime.  We now evolve the system with three observers, and set their
963: measurement efficiencies to be $0.5,0.3,0.2$ respectively.  We first
964: calculate the position variance $V_X$ of the wave-function given by
965: evolving Eq.~(\ref{dpsiN})), and verify that this remains sufficiently
966: small.  Running the simulation for a duration of $t=5$, the maximum
967: value of $\sqrt{V_X}$ is $2.7\times 10^{-3}$, and the {\em rms} value
968: over the evolution is $1.4\times 10^{-3}$.  The localization condition
969: is therefore well satisfied, and an inspection of the evolution of the
970: centroid shows that the noise is indeed negligible.  Showing that the
971: evolution is indeed the classical evolution is more nontrivial since
972: the system is chaotic: any small difference in the noise on two
973: trajectories will cause them to diverge rapidly, and one cannot
974: therefore simply compare the trajectory to the equivalent noise-free
975: classical trajectory.  In Ref.~\cite{BHJ1}, the classical dynamics was
976: verified by comparing the stroboscopic map and the largest Lyapunov
977: exponent obtained from the quantum evolution and their classical
978: equivalents.  Here we calculate the continuously estimated state, both
979: quantum and classical, for the different observers, and show that
980: these agree, and agree between observers.
981: 
982: \begin{figure}
983: \hbox{%
984: \hfill
985: \includegraphics[width=1.0\hsize]{duff_err}
986: }
987: \caption{Plot of the error standard deviation (blue) and the
988: difference between the estimated and true means for a single noise
989: realization (red) for the simulation of the Duffing oscillator using
990: the stochastic master equation with $\eta=0.5$. The green and purple
991: curves plot the same quantities for the Gaussian estimator.}
992: %\caption{Here we plot the standard deviation of the difference between
993: %the estimated mean and the mean of the wave-function (the "true" mean)
994: %for the observer with $\eta=0.5$, for the QDKR, along with the actual
995: %deviation in the Gaussian estimate and the estimate of the SME
996: %simulation for a single realization.} 
997: \label{fig2}
998: \end{figure}
999: 
1000: We now calculate the quantum state-estimate for each observer,
1001: obtained by integrating Eq.~(\ref{smemult}), and compare this with the
1002: classical (Gaussian) estimate for each observer, obtained by
1003: integrating Eqs.~(\ref{eqxc})-(\ref{cc3}).  In Figure~\ref{fig1} we
1004: plot the uncertainty in position (characterized by $\sqrt{V_X} =
1005: \sqrt{\mbox{Tr}[\rho X^2] - \mbox{Tr}[\rho X]^2}$) for the quantum
1006: state estimated by each observer over the duration of the run.  All
1007: these remain small.  The {\em rms} of $\sqrt{V_x}$ for each observer
1008: over the duration of the run is given in Table~\ref{tab1}.  The
1009: evolution of the uncertainty in position for the Gaussian
1010: state-estimate is essentially identical to the quantum estimate, and
1011: the {\em rms} of $\sqrt{V_X}$ for this estimate is also given in
1012: Table~\ref{tab1}.  Note that the position variances for each observer
1013: are, as expected, larger than the variance of the wave-function
1014: calculated using the stochastic Schr\"odinger equation. Since the
1015: solution to Eq.~(\ref{dpsiN}) can be viewed as an `unraveling' of the
1016: stochastic master equation in Eq.~(\ref{smemult}), the difference
1017: between the variances of the `true' state estimate from the former and
1018: the individual observers' state estimate from the latter provides an
1019: estimate of the amount by which their means differ, averaged over
1020: noise realizations for all the observers. We will refer to this,
1021: for want of a better name, as the error variance, and its square root
1022: as the ``error standard deviation''.  The {\em rms} value of this
1023: error standard deviation is $1.2\times 10^{-3}$, $1.7\times 10^{-3}$
1024: and $2.1\times 10^{-3}$ for observers with $\eta=0.5$, $0.3$ and $0.2$
1025: respectively for both the stochastic master equation and Gaussian
1026: simulations.  In Figure~\ref{fig2} we plot the evolution of the error
1027: standard deviation, and also the actual difference between the
1028: estimated mean and the mean of the wave-function, for both the master
1029: equation simulation and the Gaussian estimator. The small difference
1030: between these last two is most probably due solely to the fact that
1031: the mean of the computationally intensive master equation simulation
1032: has not completely converged at the value of the time step
1033: employed. (This difference is not seen for the computationally simpler
1034: delta-kicked rotor system discussed in the next subsection.)
1035: 
1036: Each observer may also track the position simply by averaging her
1037: measurement record over a suitable time period ({\it i.e.}, by low
1038: pass filtering the measurement record).  Naturally this period should
1039: be as long as possible so as to filter out the noise, but short enough
1040: so as not to filter out the deterministic motion.  For this system we
1041: use a time period of $2.5\times 10^{-2}$ for the filtering.  The
1042: average {\em rms} deviation of this estimate from the mean position of
1043: the wave-function is also given in Table~\ref{tab1} for each observer.
1044: From this we see that all observers can effectively track the motion
1045: of the particle (up to an error in position of about $10^{-2}$) using
1046: their measurement records directly.
1047: 
1048: \begin{table}
1049: \caption{{\em rms} standard deviation of state-estimates for the
1050: Duffing oscillator and the {\em rms} deviation of the averaged
1051: measurement record.}
1052: \begin{tabular}{|l@{\hspace{0.2cm}}|l@{\hspace{0.2cm}}|l@{\hspace{0.2cm}}|l@{\hspace{0.2cm}}|}
1053:    \hline
1054:    {\bf Observer's $\eta$}& $\eta=0.5$ & $\eta= 0.3$ & $\eta = 0.2$ \\ 
1055:    \hline 
1056:    {\bf Quantum } & $1.9\times 10^{-3}$ & $2.3\times 10^{-3}$ &
1057:                            $2.6 \times 10^{-3}$ \\
1058:    {\bf Gaussian } & $1.9\times 10^{-3}$ & $2.3\times 10^{-3}$ & 
1059:                            $2.6 \times 10^{-3}$ \\
1060:    {\bf Averaged Record} & $8.2\times 10^{-3}$ & $9.3\times 10^{-3}$ &
1061:                            $1.1\times 10^{-2}$ \\ 
1062:    \hline
1063: \end{tabular}
1064: \label{tab1}
1065: \end{table}
1066: 
1067: 
1068: \subsection{The delta-kicked rotor}
1069: 
1070: The delta-kicked rotor obeys the Hamiltonian
1071: \begin{equation}
1072:    H(t) = \frac{P^2}{2m} + \kappa \cos(X) \sum_{n=0}^\infty\delta(t-n)
1073:    \,.
1074: \end{equation}
1075: It is, thus, a free particle, which experiences regular kicks from the
1076: potential of a nonlinear pendulum. For a wide range of parameters, the
1077: quantum behavior of this system (by which we mean the evolution of the
1078: closed system) is very different from the classical motion.  In
1079: particular, after a few kicks the average energy of the closed
1080: classical system increases linearly with time.  In the closed quantum
1081: system, however, the average energy reaches a maximum value and after
1082: that point remains fairly constant. This is termed dynamical
1083: localization. We now simulate the evolution of the observed
1084: wave-function for this system, with the same values of $\hbar$ and $k$
1085: as we used for the Duffing oscillator, and with the same three
1086: observers.  For the system parameters we will choose $\kappa=10$ and
1087: $m=1$, and integrate for a time period of 30 kicks.  First we check
1088: the localization of the wave-function given by integrating the
1089: stochastic Schr\"odinger equation, and find that the average value of
1090: $\sqrt{V_X}$ is $2.1\times 10^{-3}$, and the maximum value obtained
1091: during the run is $3.2\times 10^{-3}$.  We check that the mean energy
1092: is indeed behaving in a classical fashion by averaging this energy
1093: over many realizations, and comparing this to the classical value.  In
1094: Figure~\ref{fig3} we plot the average energy of the observed quantum
1095: system, using $\hbar=0.1$ and $k=10$, along with both the
1096: classical result and the quantum result for $\hbar=0.1$.
1097: 
1098: \begin{figure}
1099: \leavevmode\includegraphics[width=0.9\hsize]{dkrQvsC}
1100: \caption{The average kinetic energy for the delta-kicked rotor as a
1101: function of time.  The classical value is obtained by averaging over
1102: 10,000 trajectories.  The observed quantum value was obtained by
1103: averaging over 1000 trajectories.}
1104: \label{fig3}
1105: \end{figure}
1106: 
1107: \begin{figure}
1108: \hbox{%
1109: \hfill
1110: \includegraphics[width=1.0\hsize]{qdkr_dev}
1111: \hfill
1112: }
1113: \caption{Plot of the error standard deviation (green) and the
1114: difference between estimated and true means for a single noise
1115: realization (red) for a simulation of the delta-kicked rotor using a
1116: stochastic master equation with $\eta=0.5$. The results for the
1117: Gaussian estimator are indistinguishable on this scale.}
1118: %\caption{ Here we plot the standard deviation of the difference
1119: %between the estimated mean and the mean of the wave-function (the
1120: %"true" mean) for the observer with $\eta=0.5$, for the QDKR, along
1121: %with the actual deviation for a single realization. The equivalent
1122: %plot for the Gaussian estimator is virtually indistinguishable.}
1123: 
1124: %QDKR: SME and Gaussian. diff is small. max diff SME 0.00194902488668.
1125: % max diff Gaussian 0.00194440467018.}
1126: \label{fig4}
1127: \end{figure}
1128: 
1129: We next compare the position uncertainties in the state-estimates of
1130: the different observers, as above for the Duffing oscillator, and
1131: present these results in Table~\ref{tab2}. The Gaussian estimator
1132: agrees with the stochastic master equation, and the uncertainties are
1133: small, so that the observers effectively all agree on the motion.  The
1134: averaged measurement record also tracks the motion effectively.  The
1135: {\em rms} value of the error standard
1136: deviation is $1.9\times 10^{-3}$, $2.9\times 10^{-3}$ and $3.7\times 10^{-3}$
1137: for observers with $\eta=0.5$, $0.3$, and $0.2$ respectively. In
1138: figure~\ref{fig4} we plot the evolution of the error standard
1139: deviation for the observer with $\eta=0.5$ and also the actual
1140: difference between the estimated mean and the mean of the
1141: wave-function for a single realization of the stochastic master
1142: equation simulation. The equivalent plots for the Gaussian estimator
1143: are virtually indistinguishable.
1144: 
1145: To conclude, we see from the above simulations that (i) in the
1146: classical regime the full quantum state-estimation reduces to Gaussian
1147: state-estimation, and hence classical state-estimation may be used,
1148: (ii) even without the use of true (and therefore optimal)
1149: state-estimation, low pass filtering of the measurement record alone
1150: provides adequate tracking of the system, and (iii) since the errors
1151: in the respective estimates are small, all observers effectively agree
1152: upon the motion of the system.
1153: 
1154: \begin{table}
1155: \caption{Average deviation of state-estimates for the delta-kicked
1156: rotor and the {\em rms} deviation of the averaged measurement record.}
1157: \begin{tabular}{|l@{\hspace{0.2cm}}|l@{\hspace{0.2cm}}|l@{\hspace{0.2cm}}|l@{\hspace{0.2cm}}|}
1158:     \hline 
1159:     {\bf Observer's $\eta$}& $\eta=0.5$ & $\eta= 0.3$ & $\eta = 0.2$ \\
1160:     \hline
1161:     {\bf Quantum } & $2.9\times 10^{-3}$ & $3.6\times 10^{-3}$ &
1162:              $4.3 \times 10^{-3}$ \\ 
1163:     {\bf Classical } & $2.9\times 10^{-3}$ & $3.6\times 10^{-3}$ & 
1164:              $4.3 \times 10^{-3}$ \\ 
1165:     {\bf Averaged Record } & $8.6\times 10^{-3}$ & $1.0\times 10^{-2}$ &
1166:              $1.1\times 10^{-2}$ \\ \hline
1167: \end{tabular}
1168: \label{tab2}
1169: \end{table}
1170: 
1171: \section{Conclusion}\label{conc}
1172: 
1173: The emergence of classical dynamics remains a central issue in
1174: understanding the predictions of quantum mechanics, especially now
1175: that experiments are becoming available to probe this transition
1176: directly~\cite{expts}.  In this paper, by deriving general
1177: inequalities which determine when classical mechanics will emerge, and
1178: by providing numerical examples, we have presented very substantial
1179: evidence that quantum measurement theory provides a completely
1180: satisfactory answer to the question of how classical mechanics, and
1181: hence classical chaos, emerges in a quantum world.  In doing so we
1182: have shown in some detail how the mechanism for this transition can be
1183: understood as a result of localization and noise suppression in the
1184: classical regime.
1185: 
1186: While the emergence of classical dynamics for a single motional degree 
1187: of freedom now appears to be well understood, the quantum to classical 
1188: transition as yet holds many unanswered questions. What happens, for 
1189: example, to the dynamics of a system as it passes ``through'' the 
1190: transition? How do systems behave when they are neither fully quantum 
1191: nor fully classical? For example, it is known that the delta kicked rotor demonstrates a complex behavior in the transition region~\cite{BHJS}. 
1192: Further questions include how classical dynamics emerges for other 
1193: degrees of freedom, such as spin, and what happens, for example, when 
1194: spin and motional degrees of freedom are coupled? Must all the subsystems 
1195: have a large action (we note that this has recently been investigated, see~\cite{GADBHJ}), and must all the degrees of freedom be 
1196: continuously measured, or will a subset suffice? For a spin system, 
1197: must one measure all the components of spin, or will a single 
1198: component suffice? Fortunately we are now at the point where one can 
1199: not only pose these questions, but expect that solid answers will soon 
1200: be forthcoming. 
1201: 
1202: 
1203: \begin{thebibliography}{99}
1204: %\vspace{-1.5cm}
1205: \bibitem{noqc} 
1206:   H.J.~Korsch and M.V.~Berry, Physica D {\bf 3}, 627 (1981); 
1207:   T.~Hogg and B.A.~Huberman, Phys.\ Rev.\ Lett.\ {\bf 48}, 711 (1982);
1208:   R.L.~Ingram, M.E.~Goggin and P.~W.~Milonni, in {\it Coherence
1209:       and Quantum Optics VI}, edited by J.H.~Eberly (Plenum, New York,
1210:       1990). 
1211: 
1212: \bibitem{CL} 
1213:   A.O.~Caldeira and A.J.~Leggett, Physica A {\bf 121} 587 (1983). 
1214: 
1215: \bibitem{qnoise} 
1216:   C.W.~Gardiner and P.~Zoller, {\it Quantum Noise}, 2nd edition
1217:       (Springer, Berlin, 2000).  
1218: 
1219: \bibitem{cqm1} Early work includes 
1220:   A.~Barchielli, L.~Lanz, and G.M.~Prosperi, Nuovo Cimento {\bf 72B},
1221:       79 (1982);
1222:   N.~Gisin, Phys.\ Rev.\ Lett.\ {\bf 52}, 1657 (1984);
1223:   L.~Diosi, Phys.\ Lett.\ A {\bf 114}, 451 (1986);
1224:                for a review see, 
1225:   A.~Barchielli, Int.\ J.\ Theor.\ Phys.\ {\bf 32}, 2221 (1993).
1226: 
1227: \bibitem{cqm1a}
1228:   V.P.~Belavkin and P.~Staszewski, Phys.\ Lett.\ A {\bf 40}, 359
1229:       (1989). 
1230: 
1231: \bibitem{cqm2} 
1232:   C.M.~Caves and G.J.~Milburn, Phys.\ Rev.\ A {\bf 36}, 5543 (1987). 
1233: 
1234: \bibitem{cqm3} 
1235:   H.~Carmichael, {\it An Open Systems Approach to Quantum Optics}
1236:       (Springer-Verlag, Berlin, 1993). 
1237: 
1238: \bibitem{cqm4} 
1239:   H.M.~Wiseman and G.J.~Milburn, Phys.\ Rev.\ A {\bf 47}, 642 (1993).
1240: 
1241: \bibitem{deco} 
1242:   K.~Hepp, Helv. Phys. Acta {\bf 45}, 237 (1972);
1243:   W.H.~Zurek, Phys. Rev. D {\bf 24}, 1516 (1981); 
1244:   {\em ibid} {\bf 26}, 1862 (1982); 
1245:   E.~Joos and H.D.~Zeh, Z. Phys. B {\bf 59}, 223 (1985).
1246: 
1247: \bibitem{HSZ1998} 
1248:   See, e.g., S.~Habib, K.~Shizume, and W.H.~Zurek, 
1249:   Phys.\ Rev.\ Lett.\ {\bf 80}, 4361 (1998).
1250: 
1251: \bibitem{Spiller} 
1252:   T.P.~Spiller and J.F.~Ralph, Phys.\ Lett.\ A {\bf 194}, 235 (1994).
1253: 
1254: \bibitem{Graham} 
1255:   M.~Schlautmann and R.~Graham, Phys.\ Rev.\ E {\bf 52}, 340 (1995).
1256: 
1257: \bibitem{Percival}
1258:   R.~Schack, T.A.~Brun, I.C.~Percival, J.~Phys.\ A {\bf 28}, 5401 (1995); 
1259:   T.A.~Brun, I.C.~Percival, and R.~Schack, J.~Phys.\ A {\bf 29}, 2077 (1996); 
1260:   I.C.~Percival and W.T.~Strunz, J.~Phys.\ A, {\bf 31}, 1815 (1998); 
1261:   J.~Phys.\ A, {\bf 31}, 1801 (1998);
1262: 
1263: \bibitem{BHJ1} 
1264:   T.~Bhattacharya, S.~Habib, and K.~Jacobs, Phys.\ Rev.\ Lett.\ {\bf
1265:      85}, 4852 (2000).
1266: 
1267: \bibitem{Milburn} 
1268:   A.J.~Scott and G.J.~Milburn, Phys.\ Rev.\ A {\bf 63}, 042101 (2001),
1269: 
1270: \bibitem{expts} 
1271:   M.~Brune, E.~Hagley, J.~Dreyer, X.~Maitre, A.~Maali, C.~Wunderlich,
1272:      J.M.~Raimond, and S.~Haroche, Phys.\ Rev.\ Lett.\ {\bf 77}, 4887
1273:      (1996);
1274:   C.J.~Hood, M.S.~Chapman, T.W.~Lynn, and H.J.~Kimble, 
1275:      {\it ibid.}\ {\bf 80}, 4157 (1998); 
1276:   H.~Ammann, R.~Gray, I.~Shvarchuck, and N.~Christensen, 
1277:      {\it ibid.}\ {\bf 80}, 4111 (1998); 
1278:   B.G.~Klappauf, W.H.~Oskay, D.A.~Steck, and M.G.~Raizen, 
1279:      {\it ibid.}\ {\bf 81}, 1203 (1998). 
1280: 
1281: \bibitem{Itoref} C.W. Gardiner, {\em Stochastic Methods} 
1282:                  (Springer-Verlag, Berlin, 1990).
1283: 
1284: \bibitem{afm} 
1285:   G.J.~Milburn, K.~Jacobs, and D.F.~Walls, Phys.\ Rev.\ A {\bf 50}
1286:      5256 (1994).   
1287: 
1288: \bibitem{measx} 
1289:   A.C.~Doherty and K.~Jacobs, Phys.\ Rev.\ A {\bf 60}, 2700 (1999).
1290: 
1291: \bibitem{Barchielli} The `if' part of the statement is clear from
1292:   early work in Ref.~\protect\cite{cqm1}.  The `only if' is clear by
1293:   noting that if \(\rho\) is an one-dimensional projector (so that
1294:   \(\mathop{\rm Tr} \rho^2 = 1\)) that is not into an eigenspace of
1295:   \(X^\dag X\), \(\eta > 1\) implies \(d \mathop{\rm Tr} \rho^2/dt\)
1296:   averaged over the noise \(d W\) is positive definite.
1297: 
1298: \bibitem{MOcomment} 
1299:   This follows most simply by treating each observer as having 
1300:   access to a separate output channel, or environment, but it is 
1301:   clear from~\cite{HMW} that one could obtain the same result 
1302:   by dividing up the available light using a sequence of beam 
1303:   splitters. 
1304:   %In fact, it is worth noting that simultaneous measurements by 
1305:   %multiple observers and simultaneous measurements of multiple 
1306:   %observables are essentially the same thing, because both are 
1307:   %examples of multiple output channels. 
1308:   For an explicit treatment of multiple observers see 
1309:   reference~\cite{Barch93}. A recent and less mathematical treatment 
1310:   is also given in reference~\cite{DDZ}.
1311:   
1312: \bibitem{HMW}
1313:   H.M Wiseman, PhD Thesis (University of Queensland)
1314: 
1315: \bibitem{Barch93} 
1316:   A.~Barchielli, Int.\ J.~Theor.\ Phys.\ {\bf 32}, 2221(1993).
1317:   
1318: \bibitem{DDZ}
1319:   J. Dziarmaga, D.A.R Dalvit, and W.H. Zurek quant-ph/0107033.
1320: 
1321: \bibitem{habib} S.~Habib, {\it in preparation}.
1322: 
1323: \bibitem{footnote1} Positivity at this second cumulant level does not
1324:   require the noise term: the covariance matrix stays positive even
1325:   without the addition of this piece.
1326: 
1327: \bibitem{footnote2} Since complex conjugation along with $P \to -P$
1328:   leaves the algebra $[X,P] = i\hbar$ unchanged, the momentum flip
1329:   along with $H \to -H$ ({\it i.e.}, $m\to -m$ and $F \to -F$) and
1330:   $\rho \to \rho^*$ is an invariance of Eq.~\protect\ref{sme}.  Thus
1331:   our results also apply to the negative mass case {\it mutatis
1332:   mutandis}.
1333: 
1334: \bibitem{Maybeck}
1335:   P.S.~Maybeck, {\it Stochastic Models, Estimation and Control},
1336:      (Academic Press, New York, 1982).  
1337: 
1338: \bibitem{DHJMT} 
1339:   A.C.~Doherty, S.~Habib, K.~Jacobs, H.~Mabuchi, and S.M.~Tan, 
1340:      Phys.\ Rev.\ A {\bf 62}, 012105 (2000).
1341:      
1342: \bibitem{BHJS}
1343:   T.~Bhattacharya, S.~Habib, K.~Jacobs and K.~Shizume, 
1344:      Phys. Rev. A 65, 032115 (2002) 
1345:      
1346: \bibitem{GADBHJ}
1347:   S.~Ghose, P.M.~Alsing, I.H.~Deutsch, T.~Bhattacharya, S.~Habib, K.~Jacobs, 
1348:      quant-ph/0208064.
1349: 
1350: \end{thebibliography}
1351: 
1352: %\end{multicols}
1353: \end{document}
1354: 
1355: 
1356: \bibitem{semiclass} M.C.~Gutzwiller, J.~Math. Phys. {\bf 12}, 343
1357: (1971); See also E.J.~Heller and S.~Tims, Phys. Today {\bf 46}, 38
1358: (1993).
1359: \bibitem{attempts} See, {\it e.g.}, A.~Peres, Phys. Rev. A {\bf 30},
1360: 1610 (1984); M.~Toda and K.~Ikeda, Phys. Lett. A {\bf 124}, 165
1361: (1987); Y.~Gu, {\em ibid} {\bf 149}, 95 (1990); R.~Schack and
1362: C.M.~Caves, Phys Rev. E. {\bf 53}, 3257 (1996), Eprint: quant-ph/9506008.
1363: \bibitem{expts} M.~Brune, E.~Hagley, J.~Dreyer, X.~Maitre, A.~Maali,
1364: C.~Wunderlich, J.~M.~Raimond, and S.~Haroche, Phys. Rev. Lett. {\bf
1365: 77}, 4887 (1996); C.J.~Hood, M.S.~Chapman, T.W.~Lynn, and H.J.~Kimble,
1366: {\em ibid} {\bf 80}, 4157 (1998); H.~Ammann, R.~Gray, I.~Shvarchuck,
1367: and N.~Christensen, {\em ibid} {\bf 80}, 4111 (1998); B.G.~Klappauf,
1368: W.H.~Oskay, D.A.~Steck, and M.G.~Raizen, {\em ibid} {\bf 81}, 1203
1369: (1998). 
1370: \bibitem{corresp} L.E.~Ballentine, Y.~Yang, and J.P.~Zibin,
1371: Phys. Rev. A {\bf 50}, 2854 (1994); B.S.~Helmkamp and D.A.~Browne,
1372: Phys Rev. E {\bf 49}, 1831 (1994); R.F.~Fox and T.C.~Elston, {\em
1373: ibid} {\bf 49}, 3683 (1994); {\em ibid} {\bf 50}, 2553 (1994).
1374: \bibitem{eomav} J.K.~Breslin and G.J.~Milburn, Phys. Rev. A {\bf
1375: 55}, 1430 (1997); J.~Halliwell and A.~Zoupas, Phys. Rev. D {\bf 52},
1376: 7294 (1995), Eprint: quant-ph/9503008.
1377: \bibitem{Doherty} A.C.~Doherty, S.M.~Tan, A.S.~Parkins and D.F.~Walls,
1378: to appear in Phys. Rev. A, Eprint: quant-ph/9903030.
1379: \bibitem{linbal} W.A.~Lin and L.E.~Ballentine, Phys. Rev. Lett. {\bf
1380: 65}, 2927 (1990).
1381:  
1382: 
1383: Since macroscopic systems obey classical mechanics, and since
1384: ultimately one believes that these systems are described by quantum
1385: mechanics, then in the regime in which macroscopic systems exist, the
1386: trajectories of the correct classical motion must emerge from quantum
1387: mechanics. This claim seems particularly curious when one considers
1388: that qauntum mechanics (at least for closed systems) is linear, and
1389: therefore does not exhibit the chaos which characterizes the motion of
1390: many classical systems. However, one must remember that in order to
1391: even know the trajectories that describe classical systems, they must
1392: be observed. Therefore, to describe accurately the motion of a
1393: classical system one must take into account not only the quantum
1394: dynamics given by Schr\"{o}dingers equation, but also the measurement
1395: process, which unavoidably effects the evolution of the
1396: system. Furthermore, since all classical systems interact with their
1397: environment, and since environmental interactions are equivalent to an
1398: average over a measurement process, even if they are not explicitly
1399: observed, all classical systems are subject to an effective
1400: measurement process. Here we elucidate this process in some detail,
1401: extending previous work [Bhattacharya, Habib and Jacobs (XXXX)]
1402: showing that once the measurement process is taken into account,
1403: quantum mechanics does indeed predict the emergence of classical
1404: motion, and elucidate this process in some detail. We derive
1405: inequalities that describe the parameter regime in which classical
1406: motion is obtained, and provide numerical examples. We also show that
1407: two further important properties of the classical limit. First, that
1408: multiple observers all agree on the motion of an object, and that
1409: classical statistical inference (ie. the Kalman filter) may be used to
1410: correctly track the classical motion (That is, quantum inference
1411: reduces to classical inference in the classical limit).
1412: 
1413: 
1414: 
1415: