1: \documentclass[aps,prl,twocolumn,superscriptaddress,footinbib,floatfix,showpacs]{revtex4}
2:
3: \usepackage{graphicx}
4:
5: \begin{document}
6:
7: \title{The Quantum Emergence of Chaos
8: % \vbox to 0pt{\vss
9: % \hbox to 0pt{\hskip-50pt\rm LA-UR-00-XXXX\hss}
10: % \vskip 25pt}
11: }
12: \preprint{LA-UR-00-XXXX}
13: \author{Salman Habib}
14: \affiliation{MS B285, Theoretical Division, The University of California,
15: Los Alamos National Laboratory, Los Alamos, New Mexico
16: 87545}
17:
18: \author{Kurt Jacobs}
19: \affiliation{MS B285, Theoretical Division, The University of California,
20: Los Alamos National Laboratory, Los Alamos, New Mexico
21: 87545}
22: \affiliation{Centre for Quantum Computer Technology, Centre for
23: Quantum Dynamics, School of Science, Griffith University, Nathan 4111,
24: Australia}
25:
26: \author{Kosuke Shizume}
27: \affiliation{Institute of Library and Information Science, University
28: of Tsukuba, 1-2 Kasuga, Tsukuba, Ibaraki 305-8550, Japan}
29:
30: \begin{abstract}
31:
32: The dynamical status of isolated quantum systems, partly due to the
33: linearity of the Schr\"odinger equation is unclear: Conventional
34: measures fail to detect chaos in such systems. However, when quantum
35: systems are subjected to observation -- as all experimental systems
36: must be -- their dynamics is no longer linear and, in the appropriate
37: limit(s), the evolution of expectation values, conditioned on the
38: observations, closely approaches the behavior of classical
39: trajectories. Here we show, by analyzing a specific example, that
40: microscopic continuously observed quantum systems, {\em even far from
41: any classical limit}, can have a positive Lyapunov exponent, and thus
42: be truly chaotic.
43:
44: \end{abstract}
45:
46: \pacs{03.65.Bz,05.45.Ac,05.45.Pq}
47:
48: \maketitle
49:
50: There can be no chaos in the dynamics of isolated or closed quantum
51: systems, a result which follows primarily from the linearity of the
52: Schr\"odinger equation~\cite{b1} and the linear Hilbert space
53: structure of the theory which, by virtue of the uncertainty principle,
54: prevents the formation of fine-scale structure in phase space
55: precluding chaos in the sense of classical trajectories. This leads to
56: a widely recognized difficulty, as classical mechanics, which
57: manifestly exhibits chaos, must emerge from quantum mechanics in an
58: appropriate macroscopic limit~\cite{b3}. The key to the resolution of
59: this apparent paradox lies in the fact that all experimentally
60: accessible situations necessarily involve measured, open systems: the
61: central importance of such situations in the context of chaos was
62: first emphasized by Chirikov~\cite{b4}. In a closely connected
63: development, continuous quantum measurement theory~\cite{b5} has led
64: to the successful understanding of the emergence of classical dynamics
65: from the underlying quantum physics~\cite{b8,b12,b13,b14}, and
66: inequalities have been derived that encapsulate the regime under which
67: classical motion, and thus classical chaos, exists~\cite{b12}. The
68: transition to classical mechanics results from the localization of the
69: quantum density matrix due to the information continuously provided by
70: the measurement (itself mediated by an environmental interaction), and
71: the balancing of this against the unavoidable noise from the quantum
72: backaction of the measurement. For a macroscopic system, the
73: Ehrenfest theorem holds as a result of localization and,
74: simultaneously, the backaction noise is negligible, resulting in a
75: smooth classical trajectory.
76:
77: While it has been established that observed quantum systems can be
78: chaotic when they are macroscopic enough that classical dynamics has
79: emerged, can they be chaotic outside this limit? This is the question
80: we address here. By defining and computing the Lyapunov exponent for
81: an observed quantum system deep in the quantum regime, we are able
82: show that the system dynamics is chaotic. Further, the Lyapunov
83: exponent is not the same as that of the classical dynamics that
84: emerges in the classical limit. Since the quantum system in the
85: absence of measurement is not chaotic, this chaos must emerge as the
86: strength of the measurement is increased, and we examine the nature
87: of this emergence.
88:
89: The rigorous quantifier of chaos in a dynamical system is the maximal
90: Lyapunov exponent~\cite{b21}. The exponent yields the (asymptotic)
91: rate of exponential divergence of two trajectories which start from
92: neighboring points in phase space, in the limit in which they evolve
93: to infinity, and the neighboring points are infinitesimally close. The
94: maximal Lyapunov exponent characterizes the sensitivity of the system
95: evolution to changes in the initial condition: if the exponent is
96: positive, then the system is exponentially sensitive to initial
97: conditions, and is said to be chaotic. We apply this notion below to
98: the observation-conditioned evolution of quantum expectation values.
99:
100: The evolution of a simple single-particle quantum system under an
101: ideal continuous position measurement is given by the nonlinear
102: stochastic master equation (SME) for the system density
103: matrix~\cite{b15}:
104: \begin{eqnarray}
105: d\rho &=& - \frac{i}{\hbar} [H,\rho]dt - k[x, [x, \rho]]dt
106: \nonumber\\
107: & & + 4k ( x\rho + \rho x - 2 \langle x \rangle )
108: (dy - \langle x \rangle dt) \,,
109: \label{sme}
110: \end{eqnarray}
111: where the first term on the right hand side is due to unitary
112: evolution, $H$ being the Hamiltonian, and the second term represents
113: diffusion from ``quantum noise'' due to the unavoidable quantum
114: backaction of the measurement. The position operator is $x$, and the
115: parameter $k$ characterizes the rate at which the measurement extracts
116: information about the observable, and which we will refer to as the
117: {\em strength} of the measurement~\cite{b20}. The final term
118: represents the change in the density matrix as a result of the
119: information gained from the measurement. Here, $dy$ is the
120: infinitesimal change in the continuous output of the measuring device
121: in the time $dt$. The continuous output of the measuring device,
122: $y(t)$, referred to usually as the {\em measurement record}, is
123: determined by $dy = \langle x \rangle dt + dW/{\sqrt{8k}} $ where $dW$
124: is the Wiener increment, describing driving by Gaussian white
125: noise~\cite{footnote1}. The noise $dW$ is due to the fact that the
126: results of the measurement are necessarily random. (Note that the
127: backaction and $dW$ are uncorrelated with each other.) Thus on a given
128: experimental run, the system will be driven by a given realization of
129: the noise process $dW$. We will label the possible noise realizations
130: by $s$.
131:
132: \begin{figure}
133: \includegraphics[width=8.5cm,height=5.2cm]{figsal.eps}
134: \caption[example]{ \label{fig0} Position distribution for
135: the Duffing oscillator with measurement strengths $k=0.01$ (red)
136: and $k=10$ (green), demonstrating measurement-induced
137: localization ($k=10$). The momentum distribution behaves similarly.}
138: \end{figure}
139:
140: A single quantum mechanical particle is in principle an infinite
141: dimensional system. However, for the purpose of defining an
142: observationally relevant Lyapunov exponent, it is sufficient to use a
143: single projected data stream: Here we choose the expectation value of
144: the position, $\langle x(t)\rangle$. The important quantity is thus
145: the divergence, $\Delta(t) = |\langle x(t)\rangle - \langle
146: x_{\mbox{\scriptsize fid}}(t) \rangle |$, between a fiducial
147: trajectory and a second trajectory infinitesimally close to it. It is
148: important to keep in mind that the system is driven by noise. Since we
149: wish to examine the sensitivity of the system to changes in the
150: initial conditions, and not to changes in the noise, we must hold the
151: noise realization fixed when calculating the divergence. The Lyapunov
152: exponent is thus
153: \begin{equation}
154: \lambda \equiv \lim_{t\rightarrow\infty}
155: \lim_{\Delta_s(0)\rightarrow 0} \frac{\ln \Delta_s (t)}{t}
156: \equiv \lim_{t\rightarrow\infty}\lambda_s(t)
157: \label{Lyap}
158: \end{equation}
159: where the subscript $s$ denotes the noise realization. This definition
160: is the obvious generalization of the conventional ODE definition to
161: dynamical averages, where the noise is treated as a drive on the
162: system. Indeed, under the conditions when (noisy) classical motion
163: emerges, and thus when localization holds (Fig.~\ref{fig0}), it
164: reduces to the conventional definition, and yields the correct
165: classical Lyapunov exponent. To combat slow convergence, we measure
166: the Lyapunov exponent by averaging over an ensemble of finite-time
167: exponents $\lambda_s(t)$ instead of taking the asymptotic long-time
168: limit for a single trajectory.
169:
170: A key result now follows: In unobserved, i.e., isolated quantum
171: dynamical systems, it is possible to prove, by employing unitarity and
172: the Schwarz inequality, that $\lambda$ vanishes; the finite-time
173: exponent, $\lambda(t)$, decays away as $1/t$~\cite{tocome}. This
174: theorem codifies the expectation that, since the evolution is linear,
175: any measure of chaos applied to it should yield a null result. As we
176: have emphasized earlier, however, once measurement is included the
177: evolution becomes nonlinear and the Lyapunov exponent need not
178: vanish. We now address this crucial question for a specific example.
179:
180: %(Note
181: %that our definition of the Lyapunov exponent is drastically different from
182: %other definitions of Lyapunov-like exponents for quantum systems which
183: %saturate at long times, and thus do not quantify chaos in the classical
184: %sense of the word.)
185:
186: The system we consider is the quantum Duffing
187: oscillator~\cite{b26}, which is a single particle in a double-well
188: potential, with sinusoidal driving. The Hamiltonian for the Duffing
189: oscillator is
190: \begin{equation}
191: H=p^2/2m + B x^4 - A x^2 + \Lambda x\cos(\omega t)
192: \label{lbham}
193: \end{equation}
194: where $p$ is the momentum operator, $m$ the particle mass, and $A$,
195: $B$ and $\Lambda$ determine the potential and the strength of the
196: driving force. We fix the values of the parameters to be $m=1$,
197: $B=0.5$, $A=10$, $\Lambda=10$ and $\omega=6.07$. The action of a
198: system relative to $\hbar$ can be varied either by changing parameters
199: in the Hamiltonian, or by introducing scaled variables so that the
200: Hamiltonian remains fixed, but the effective value of $\hbar$ becomes
201: a tunable parameter. Here we employ the latter choice as it captures
202: the notion of system size with a single number; the smaller $\hbar$
203: the larger the system size, and vice versa.
204:
205: To examine the emergence of chaos we will first choose
206: $\hbar=10^{-2}$, which is small enough so that the system makes a
207: transition to classical dynamics when the measurement is sufficiently
208: strong. In this way, as we increase the measurement strength, we can
209: examine the transformation from essentially isolated quantum evolution
210: all the way to the (known) chaos of the classical Duffing
211: oscillator. To examine the emergence of chaos, we simulate the
212: evolution of the system for $k= 5\times 10^{-4}, 10^{-3}, 0.01, 0.1,
213: 1, 10$. When $k\leq 0.01$, the distribution is spread over the entire
214: accessible region, and Ehrenfest's theorem is not
215: satisfied. Conversely, for $k=10$, the distribution is well-localized
216: (Fig.~\ref{fig0}), and Ehrenfest's theorem holds throughout the
217: evolution. Since the backaction noise, characterized by the momentum
218: diffusion coefficient, $D_p=\hbar^2k$, remains small, at this value of
219: $k$ the motion is that of the classical system, to a very good
220: approximation.
221:
222: Stroboscopic maps help reveal the global structural transformation in
223: phase space in going from quantum to classical dynamics
224: (Fig.~\ref{fig1}). The maps consist of points through which the system
225: passes at time intervals separated by the period of the driving
226: force. For very small $k$, $\langle x\rangle$ and $\langle p\rangle$
227: are largely confined to a region in the center of phase
228: space. Somewhat remarkably, at $k=0.01$, although the system is
229: largely delocalized, as shown in Fig.~\ref{fig0}, nontrivial structure
230: appears, with considerable time being spent in certain outer
231: regions. By $k=1$ the localized regions have formed into narrower and
232: sharper swirling coherent structures. At $k=10$ the swirls disappear,
233: and we retrieve the uniform chaotic sea of the classical map (the
234: small `holes' are periodic islands). The swirls in fact correspond to
235: the unstable manifolds of the classical motion. Classically, these
236: manifolds are only visible at short times, as continual and repeated
237: folding eventually washes out any structure in the midst of a uniform
238: tangle. In the quantum regime, however, the weakness of the
239: measurement, with its inability to crystallize the fine structure, has
240: allowed them to survive: we emphasize that the maps result from
241: long-time integration, and are therefore essentially time-invariant.
242:
243: \begin{figure}
244: \includegraphics[width=8.5cm,height=7.2cm]{fig2comp.ps}
245: \caption[example]{ \label{fig1} Phase space stroboscopic maps shown
246: for 4 different measurement strengths, $k=5\times 10^{-4}$, 0.01
247: (top), and 1, 10 (bottom). Contour lines are superimposed to provide a
248: measure of local point density at relative density levels of
249: $0.05,~0.15,~0.25,~0.35,~0.45,$ and $0.55$.}
250: \end{figure}
251:
252: To calculate the Lyapunov exponent we implement a numerical version of
253: the classical linearization technique~\cite{b22}, suitably generalized
254: to quantum trajectories. The method was tested on a classical noisy
255: system with comparison against results obtained from solving the exact
256: equations for the Lyapunov exponents~\cite{b23}. The calculation is
257: very numerically intensive, as it involves integrating the stochastic
258: Schr\"{o}dinger equation equivalent to the SME (\ref{sme}) over
259: thousands of driving periods, and averaging over many noise
260: realizations; parallel supercomputers were invaluable for this task.
261:
262: We find that as $t$ is increased, for nonzero $k$, the value obtained
263: for $\lambda(t)$ falls as $1/t$, following the behavior expected for
264: $k=0$, until a point at which an asymptotic regime takes over,
265: stabilizing at a finite value of the Lyapunov exponent as
266: $t\rightarrow\infty$. This behavior is shown in Fig.~\ref{fig2} for
267: three different values of $k$. The Lyapunov exponent as a function of
268: $k$ is shown in Fig.~\ref{fig3}. The exponent increases over two
269: orders of magnitude in an approximately power-law fashion as $k$ is
270: varied from $5\times 10^{-4}$ to $10$, before settling to the
271: classical value, $\lambda_{Cl}=0.57$. The results in Figs.~\ref{fig2}
272: and \ref{fig3} show clearly that chaos emerges in the observed quantum
273: dynamics well before the limit of classical motion is obtained.
274:
275: \begin{figure}
276: \includegraphics[width=8.5cm,height=7.7cm]{fig3.eps}
277: \caption[example]{ \label{fig2} Finite-time Lyapunov exponents
278: $\lambda(t)$ for measurement strengths $k=5\times 10^{-4},~0.01,~10$,
279: averaged over 32 trajectories for each value of $k$ (linear
280: scale in time, top, and logarithmic scale, bottom; bands indicate
281: the standard deviation over the 32 trajectories). The (analytic) $1/t$
282: fall-off at small $k$ values, prior to the asymptotic regime, is
283: evident in the bottom panel. The unit of time is the driving period.}
284: \end{figure}
285:
286: We now compute the Lyapunov exponent for the quantum system when its
287: action is sufficiently small that smooth classical dynamics cannot
288: emerge, even for strong measurement. Taking a value of $\hbar=16$, we
289: find that for $k=5\times 10^{-3}$, $\lambda=0.029\pm 0.008$, for
290: $k=0.01$, $\lambda=0.046\pm 0.01$ and for $k=0.02$, $\lambda=0.077\pm
291: 0.01$. Thus the system is once again chaotic, and becomes more
292: strongly chaotic the more strongly it is observed. From these results,
293: it is clear that there exists a purely {\em quantum} regime in which
294: an observed system, while behaving in a fashion quite distinct from
295: its classical limit, nevertheless evolves chaotically with a finite
296: Lyapunov exponent, also distinct from the classical value.
297:
298: \begin{figure}[t]
299: \includegraphics[width=8.5cm,height=5.5cm]{fig4.eps}
300: \caption[example]{ \label{fig3} The emergence of chaos: The
301: Lyapunov exponent $\lambda$ as a function of measurement strength
302: $k$. Error-bars follow those of Fig.~3, taken at the final time.}
303: \end{figure}
304:
305: It is worth pointing out that an analogous analysis can also be
306: carried out for a continuously observed classical system. First one
307: notes that an {\em unobserved} probabilistic classical system also has
308: provably zero Lyapunov exponent: the average of $x$ for an ensemble of
309: classical particles does not exhibit chaos, due to the linearity of
310: the Liouville equation~\cite{tocome}. If we consider a noiseless
311: observed chaotic classical system -- possible since classical
312: measurements are by definition passive (no backaction noise) -- then
313: even the weakest meaningful measurement will, over time, localize the
314: probability density, generating an effective trajectory limit, and
315: thus the classical Lyapunov exponent,
316: $\lambda_{Cl}$~\cite{tocome}. Noise can always be be injected into
317: classical systems as an external drive, nevertheless, in the limit of
318: weak noise, the system will once again possess the noiseless exponent
319: $\lambda_{Cl}$: In a classical system the external noise is not
320: connected to the strength of the measurement, so one can
321: simultaneously have strong measurement and weak noise, which is
322: possible in the quantum theory only under specific
323: conditions~\cite{b12}. As one way to understand this case, we can
324: employ the quantum result as an intermediate step. Consider the
325: quantum Lyapunov exponent at a fixed value of $k$ (where $\lambda <
326: \lambda_{Cl}$) as in Fig.~\ref{fig3}. If the value of $\hbar$ is now
327: reduced, the dynamics of the system must tend to the classical limit
328: as the quantum-classical correspondence inequalities of
329: Ref.~\cite{b12} are better satisfied. Thus the Lyapunov exponent in
330: the classical limit of quantum theory -- which, to a very good
331: approximation, is just classical dynamics driven by weak noise -- must
332: tend to $\lambda_{Cl}$. If, however, the noise is not weak, an
333: observed classical system, like a quantum system outside the classical
334: regime, will also not be localized, and may well have an exponent
335: different from $\lambda_{Cl}$. In addition, one may expect the
336: non-localized quantum and classical evolutions to have quite different
337: Lyapunov exponents, especially when $\hbar$ is large on the scale of
338: the phase space, as quantum and classical evolutions generated by a
339: given nonlinear Hamiltonian are essentially different~\cite{Hnonlin}.
340: The nature of the Lyaunov exponent for non-localized classical systems,
341: and its relationship to the exponent for quantum systems is a very
342: interesting open question.
343:
344: Finally, we emphasize that the chaos identified here is not merely a
345: formal result - even deep in the quantum regime, the Lyapunov exponent
346: can be obtained from measurements on a real system as in
347: next-generation cavity QED and nanomechanics
348: experiments~\cite{exp}. Experimentally, one would use the known
349: measurement record to integrate the SME (\ref{sme}); this provides the
350: time evolution of the mean value of the position. From this fiducial
351: trajectory, given the knowledge of the system Hamiltonian, the
352: Lyapunov exponent can be obtained by following the procedure described
353: here.
354:
355: We thank Tanmoy Bhattacharya, Daniel Steck, and James Theiler for
356: helpful suggestions. Supercomputing resources were made available by
357: the LANL Institutional Computing Initiative and the Queensland
358: Parallel Supercomputing Facility. This work was supported by the DOE,
359: the ARC, and the state of Queensland.
360:
361: \begin{thebibliography}{99}
362:
363: \bibitem{b1}
364: R.~Kosloff and S.A.~Rice, J. Chem. Phys. {\bf 74}, 1340 (1981);
365: J.~Manz, J. Chem. Phys. {\bf 91}, 2190 (1989).
366:
367: \bibitem{b3}
368: See, e.g., A.~Peres, {\em Quantum Theory: Concepts and Methods}
369: (Kluwer, London, 1993).
370:
371: \bibitem{b4}
372: B.V.~Chirikov, Chaos {\bf 1}, 95 (1991).
373:
374: \bibitem{b5}
375: H.J.~Carmichael, {\em An Open Systems Approach to Quantum Optics}
376: (Springer, Berlin, 1993); C.W.~Gardiner and P.~Zoller, {\em
377: Quantum Noise} (Springer, Berlin, 2000); M.~Orszag, {\em Quantum
378: Optics} (Springer, Berlin, 2000).
379:
380: \bibitem{b8} T.P.~Spiller and J.F.~Ralph, Phys. Lett. A {\bf 194}, 235
381: (1994); T.A.~Brun, I.C.~Percival and R.~Schack, J. Phys. A {\bf 29},
382: 2077 (1996); I.C.~Percival and W.T.~Strunz, J. Phys. A {\bf 31},
383: 1801 (1998); {\bf 31}, 1815 (1998).
384:
385: \bibitem{b12} T.~Bhattacharya, S.~Habib and K.~Jacobs,
386: Phys. Rev. Lett. {\bf 85}, 4852 (2000); Phys. Rev. A {\bf 67},
387: 042103 (2003). See also, S.~Ghose et al., Phys. Rev. A {\bf 69},
388: 052116 (2004).
389:
390: \bibitem{b13}
391: A.J.~Scott and G.J.~Milburn, Phys. Rev. A {\bf 63}, 042101 (2001).
392:
393: \bibitem{b14}
394: Y. Ota and I. Ohba, Eprint: quant-ph/0308154.
395:
396: \bibitem{b21}
397: J.-P.~Eckmann and D.~Ruelle, Rev. Mod. Phys. {\bf 57}, 617 (1985).
398:
399: \bibitem{b26}
400: W.A.~Lin and L.E.~Ballentine, Phys. Rev. Lett. {\bf 65}, 2927
401: (1990); S.~Habib, K.~Shizume and W.H.~Zurek, Phys. Rev. Lett. {\bf
402: 80}, 4361 (1998).
403:
404: \bibitem{b15}
405: L.~Diosi, Phys. Lett. {\bf 129A}, 419 (1988); V.P.~Belavkin and
406: P.~Staszewski, Phys. Lett. {\bf 140A}, 359 (1989); Y.~Salama and
407: N.~Gisin, Phys. Lett. {\bf 181A}, 269 (1993); H.M.~Wiseman and
408: G.J.~Milburn, Phys. Rev. A {\bf 47}, 642(1993); G.J.~Milburn,
409: Quantum Semiclass. Opt. {\bf 8}, 269 (1996).
410:
411: \bibitem{b20}
412: A.C.~Doherty, K.~Jacobs and G.~Jungman, Phys. Rev. A {\bf 63},
413: 062306 (2001).
414:
415: \bibitem{footnote1}
416: This idealized measurement has an output record with infinite
417: bandwidth. It is a good approximation to real measurements so long as
418: the bandwidth of the measuring device is large compared to that of the
419: system dynamics.
420:
421: \bibitem{tocome}
422: S.~Habib, K.~Jacobs and K.~Shizume, in preparation.
423:
424: \bibitem{b22}
425: A.~Wolf, J.B.~Swift, H.L.~Swinney and J.A.~Vastano, Physica {\bf
426: 16D}, 285 (1985).
427:
428: \bibitem{b23}
429: S.~Habib and R.D.~Ryne, Phys. Rev. Lett. {\bf 74}, 70 (1995).
430:
431: \bibitem{Hnonlin}
432: S.~Habib, K.~Jacobs, H.~Mabuchi, R.~Ryne, K.~Shizume and
433: B.~Sundaram, Phys. Rev. Lett. {\bf 88}, 040402 (2002).
434:
435: \bibitem{exp}
436: H.~Mabuchi and A.C.~Doherty, Science {\bf 298}, 1372 (2002);
437: M.D.~LaHaye, O. Buu, B. Camarota, and K.C.~Schwab, Science {\bf
438: 304}, 74 (2004).
439:
440: \end{thebibliography}
441:
442: \end{document}
443: