nlin0012013/cat.tex
1: \documentstyle[aps,floats,psfig,multicol,pre,amstex]{revtex}
2: 
3: \newcommand{\at}{{\char '100}}
4: 
5: \newcommand{\DEF}{\stackrel{\mathrm{def}}{=}}
6: \newcommand{\imat}{{\mathrm{i}}}    
7: \newcommand{\hbare}{\hbar_{\mathrm{eff}}}  
8: 
9: \newcommand{\opp}{\hat{p}}
10: \newcommand{\opq}{\hat{q}}
11: \newcommand{\opk}{\hat{k}}
12: 
13: 
14:   
15:   
16: \begin{document}
17: 
18: %\widetext
19: \draft
20: %\twocolumn[\hsize\textwidth\columnwidth\hsize\csname
21: %\twocolumnfalse\endcsname 
22: 
23: \title{Chaos assisted tunnelling with cold atoms}
24: \author{
25:        A. Mouchet$^1$\thanks{mouchet\at celfi.phys.univ-tours.fr},
26:    C. Miniatura, R. Kaiser$^2$\thanks{miniat,kaiser\at inln.cnrs.fr},
27:      B. Gr\'emaud, D. Delande$^3$\thanks{gremaud,delande\at
28: spectro.jussieu.fr}
29:        }
30: \address{        $^1$Laboratoire de Math\'ematiques 
31:                             et de Physique Th\'eorique 
32:                             \textsc{(cnrs upres-a 6083)}.
33:                             Ave\-nue Monge, Parc de Grandmont 37200
34: Tours, 
35:                             France.\\
36:          $^2$Institut Non Linéaire de Nice \textsc{(cnrs umr 7618)},
37:                                 1361, route des Lucioles,
38:                                 Sophia Antipolis,
39:                                 F-06560 Valbonne, France.\\
40:  $^3$Laboratoire Kastler-Brossel \textsc{(cnrs umr 8552)}, Universit\'e
41:                                Pierre et Marie Curie, 4, place Jussieu,
42:                                F-75005 Paris,  France. 
43:         }
44: 
45: 
46: \date{\today}
47: 
48: 
49: \maketitle
50: 
51: \begin{abstract}
52: In the context of quantum chaos, both theory and numerical analysis
53:  predict large
54:  fluctuations of the tunnelling transition
55: probabilities when irregular dynamics is present at the classical level.
56: We consider here the non-dissipative quantum evolution of 
57: cold atoms trapped in a time-dependent modulated periodic potential 
58: generated by 
59: two laser beams.  We
60: give some precise guidelines for the observation of 
61:  chaos assisted tunnelling between invariant phase space structures 
62:  paired
63: by time-reversal symmetry. 
64: 
65: 
66: \end{abstract}
67: 
68: \pacs{PACS : 05.45.Mt   Semiclassical chaos (quantum chaos) \\
69: 05.60.Gg   Quantum transport \\
70: 32.80.Qk   Coherent control of atomic interactions with photons \\
71: 05.45.Pq   Numerical simulations of chaotic models 
72: }
73: 
74: \bigskip
75:                      
76: \section{Introduction}
77:  
78: During the seventies and the eighties, it became gradually clear that
79: classical Hamiltonian chaos profoundly affects 
80: the  temporal evolution and the spectral  properties 
81: of the corresponding quantum system as 
82: compared to the integrable case~\cite{Giannoni+91a}.
83: Some of these features 
84: (dynamical localization, 
85: scars of periodic orbits~\cite{Heller84a}, etc) 
86: share striking similarities with concepts
87:  originating from condensed matter such as weak and strong 
88: localization~\cite{Akkermans+95a}. In fact these
89: phenomena can be recast in terms of wave transport in disordered
90: media, (quasi)-randomness being of statistical or dynamical origin.
91: In this  context, it is important to 
92: understand the mechanisms 
93: underlying a key feature of wave propagation which has no
94: classical analog: tunnelling.
95: 
96: 
97: Tunnelling refers to any wave  process which is classically
98: forbidden to \emph{real} solutions of Hamilton equations. For one-dimensional
99: (1D)
100:  autonomous systems, it is well known that the quantum 
101: tunnelling probability 
102: through an
103: energetic barrier can be
104: evaluated semiclassically with the help of classical 
105: \emph{complex} solutions 
106: of Hamilton 
107: equations~\cite{Messiah65a,Balian/Bloch74a}. 
108: The direct generalization of this procedure to higher
109: dimensional  systems is straightforward for separable dynamics 
110: but is already subtle for  still integrable,
111:  but no longer separable,
112: dynamics~\cite{Wilkinson86b,Wilkinson/Hannay87a}. 
113: In the generic case 
114: of chaotic dynamics,  it even proves 
115:  extremely hard to
116: handle and the situation, until recently, seemed hopeless.
117: This is so because, in the presence of chaos, the analytical and
118: topological 
119: properties of the classical
120: \emph{complexified}  phase space are far from trivial.
121:   During the last ten years however, theoretical and numerical 
122: investigations on autonomous 2D and time-dependent 1D Hamiltonians
123: systems have 
124: started to highlight
125: some
126: mechanisms~\cite{Wilkinson86b,Wilkinson/Hannay87a,Lin/Ballentine90a,Bohigas+93a,Bohigas+93b,Tomsovic/Ullmo94a,Creagh/Whelan96a} and 
127: much insight
128: has been gained on the influence of such classical
129:  non-separable dynamics. Experimental
130:  evidence of such mechanisms, which is still lacking,
131:  would be of great interest especially in the light 
132: of the subtle interplay between interferences and disorder.
133:  
134: 
135:  In this paper, we consider 1D time-dependent 
136:  dynamics, one of the simplest case where
137: irregular motion can appear, and we study  
138: chaos assisted tunnelling.  Our effective Hamiltonian 
139: model, which is derived from an experimentally
140: achievable situation, exhibits three 
141: main properties. First, 
142: its classical dynamics is time-reversal invariant.
143: Second it is
144: controlled by a single real external parameter $\gamma$ (for $\gamma=0$
145:  the dynamics is integrable and chaos develops more and more in
146: phase space as $\gamma$ is increased). Third, there exists 
147: in phase space, for a whole  continuous range of~$\gamma$,
148:  a pair of stable 
149: islands~$\mathcal{I}_{+}$ and~$\mathcal{I}_{-}$ which are time
150: reversed images of each other. By stable islands we mean the set 
151: of regular classical 
152: trajectories in phase space which stay near 
153:  a stable equilibrium point or near a stable periodic orbit of the
154: system.
155: In this case,
156: no real classical orbit started in one of these islands can
157:  go into the other one. However, the quantum
158: dynamics of a wave-packet, initially prepared in one island, will
159: display a periodic behavior. The wave-packet jumps from one island to
160: its time reversed image~\cite{Averbukh+95a}.
161:   In the quantum spectrum
162: this tunnelling process appears via the 
163: existence of non
164: degenerate energy doublets whose splitting give the inverse of the
165: tunnelling time  between~$\mathcal{I}_{+}$ 
166: and~$\mathcal{I}_{-}$. Varying~$\gamma$ slowly modifies 
167: the geometry of the islands themselves.
168:  The crucial point is that it will drastically change the classical
169: dynamics
170: for some initial conditions lying between the islands. For~$\gamma$
171: small 
172: enough, the chaotic layers are too small to play a significant role
173: at~$\hbar$ scales and hence cannot influence the quantum behaviour of
174: the 
175: system
176: which is essentially still regular.
177: For larger values,  but still
178:  before the stable islands are completely destroyed, there is a chaotic
179: regime where varying~$\gamma$ or $\hbar$ 
180: (by $\hbar$ we mean here Planck's constant 
181: divided by some typical classical action)
182: alone induces large fluctuations, on several orders of magnitude,
183: of the doublet splittings around their mean value. This in turn 
184: means large fluctuations of the
185: tunnelling periods. These wild fluctuations
186: induced by small changes of any  parameter are a
187: signature of the so-called ``chaos assisted tunnelling'' regime.
188: It has been
189: extensively studied both theoretically and 
190: numerically in the situation described 
191: above~\cite{Lin/Ballentine90a,Bohigas+93a,Bohigas+93b,Tomsovic/Ullmo94a} 
192: but has not been yet observed in 
193: real experiments, the main reason being its extreme sensitivity 
194: to small changes in the
195: classical dynamics.
196:  Any \emph{uncontrolled} variation of $\gamma$, be it
197:  noise or dissipation, will dramatically swamp or destroy the signal.
198: The
199: observation of this highly fluctuating
200: tunnelling regime thus requires both an
201: accurate control of the dynamics, of the preparation 
202:  of the initial state and of the
203: analysis of the final state.
204: 
205: 
206: Atom cooling techniques~\cite{Arimondo+92a}
207:  provide systems which fulfill all these requirements. 
208:  They allow an accurate manipulation and
209: control of internal and external degrees of freedom and are
210: now a useful tool to produce situations where the wave character
211: of the atomic motion is 
212: essential~\cite{Berman96a}. 
213: A great variety of potentials can be produced to influence the
214: atomic motion, be it by means of inhomogeneous magnetic fields, material
215: gratings or laser light. Optical
216:  lattices with crystalline or 
217: quasi-crystalline
218: order~\cite{Weidemuller+95a,Guidoni+97a,Drese/Holthaus97a} 
219: can be easily
220: produced where atoms mimic 
221: situations usually
222: encountered in condensed 
223: matter~\cite{BenDahan+96a,Niu+95a}.
224:  Dissipation (spontaneous emission and atom-atom interaction) 
225: is easily controlled
226:  and coherence times of the order of 10 ms
227: can be achieved. This is why cold atoms are a unique tool to
228: study transport properties of waves, be it quantum
229: chaos~\cite{Moore+94a} 
230: or weak localization~\cite{Labeyrie+99a,Jonckheere+00a}.
231: 
232: 
233: The general organization of this paper is the following. 
234: In section~\ref{sec:experimentalsetup}  we
235: explain the origin of the effective Hamiltonian
236:  for the experimental situation
237:  under consideration. In section~\ref{sec:classicaldynamics} 
238: we study the corresponding classical dynamics 
239: and show why this effective Hamiltonian is relevant for chaos assisted
240:  tunnelling. In 
241: section~\ref{sec:quantumdynamics} we quickly review some of the
242: usual theoretical techniques when dealing with both space and time 
243: periodic quantum dynamics. We also
244: illustrate  how some quantum spectral properties 
245:  have a natural classical interpretation.
246: In section~\ref{sec:cat} we show, with the help of numerical 
247: experiments,
248:  how chaos assisted tunnelling
249: arises in our system and then explain how to observe it 
250: in a real experiment.
251: Section~\ref{sec:conclusion} is devoted to some concluding remarks.
252:    
253: 
254: 
255: \section{Effective Hamiltonian}
256: \label{sec:experimentalsetup}
257: \subsection{Light shifts}
258: 
259: The very basic physical mechanism underlying 
260: our forthcoming discussion 
261: is the following: when an atom is exposed to monochromatic light, 
262: its energy levels are shifted by the interaction. 
263:  These level shifts originate from the polarization 
264: energy of the atom in the incident light field and are called the
265: light shifts~\cite{CohenTannoudji+88a}. In the dipolar 
266: approximation, they only depend on the field intensity
267: value at the center-of-mass position of the atom. 
268: If the field intensity is space-time 
269: dependent, then a moving atom will experience dipolar forces: 
270: inhomogeneous light shifts act as potentials and alter the 
271: center-of-mass 
272: motion of the atom. By 
273: appropriately tailoring the space-time dependence of the 
274: light field, one can then produce a great variety of potentials 
275: for the external atomic motion as evidenced by the atom cooling 
276: industry. Note, however, that the atom-light interaction is also
277: responsible for a dissipative phenomenon (real absorption of a
278: photon followed by spontaneous emission) which shortens the
279: temporal coherence of the atomic wave function. By using a laser light
280: far detuned from any atomic resonance, it is possible to control
281: this stray phenomenon and maintain it at a reasonably low rate.
282: 
283: In the following, we shall describe a simple physical 
284: situation for atoms where chaos 
285: assisted tunnelling should show up.
286: 
287: 
288: \subsection{Experimental configuration}
289: 
290: Although the internal structure (hyperfine Zeeman sub-levels) are
291: of major importance in the atom cooling techniques, we will here for
292: simplicity model the atom by a two-level system (as indeed, only one
293: optical transition usually governs the dynamics). 
294: We consider a dilute sample of identical (but independent) 
295: two-level atoms propagating in the light field configuration
296:  created by two monochromatic standing waves 
297: with frequencies $\omega_\pm=\omega_L\pm\delta\omega/2$ 
298: where~$\delta\omega\ll\omega_L$.
299:  We denote
300:  by $|g\rangle$ and $|e\rangle$ the ground-state and 
301: the excited state of each atom, these levels being connected by 
302: an electric dipole transition of angular
303:  frequency $\omega_{\mathrm{at}}$ and width~$\Gamma$. 
304:   All atoms are supposed to be initially 
305: prepared in their ground state. 
306: Each standing-wave is produced along the $x$ axis by two
307: counter-propagating 
308: laser beams and we suppose all fields
309: to be linearly polarized along the $z$ axis (see
310: figure~\ref{fig:piegeos}). 
311: After a suitable choice of space-time origin, the total electrical 
312: field strength is
313: \begin{equation}
314:         E(x,t)=\big[E_+\cos(\omega_+t)+E_- \cos(\omega_-t)\big] \cos(k_Lx)
315: \end{equation}
316: where $E_\pm$ are the field strengths of the two standing-waves. 
317: At this point we have neglected the difference in wave-vectors 
318: of the standing-waves. 
319: For this to hold, it is sufficient to assume that
320: the atomic sample size is small enough. Typically, the difference in the
321: $k$ vectors will be of the order of $10^{-9}$ or less (see below),
322: so that this requires the atomic cloud to be smaller than typically
323: few kilometers, which is amply satisfied in a standard magneto-optical trap.
324: 
325: 
326: \subsection{Dimensionless effective Hamiltonian}\label{subsec:effham}
327: 
328: The effective Hamiltonian which describes the atomic motion is 
329: derived in appendix~\ref{app:effham} under some common and
330: well-controlled 
331: approximations. It acts in the Hilbert space of a one-dimensional system
332: which is simply the $x$ component (position) of the center
333: of mass of the atom
334: (which means that the internal degree of freedom as well as the
335: $y$ and $z$ coordinates can be eliminated, see Appendix). It reads: 
336: \begin{equation}\label{eq:ham_eff}
337:         H=\frac{p_x^2}{2M}-V_0\cos(2k_Lx)[\theta+\cos(\delta\omega\, t)]
338: \end{equation}
339: where $V_0\DEF-\hbar\Omega_+
340: \Omega_-/8\delta_L$ and 
341: $\theta\DEF(\Omega_+/2\Omega_-)
342: +(\Omega_-/2\Omega_+)$, with $\delta_L=\omega_L-\omega_{at}$ the
343: detuning with respect to the atomic frequency 
344: and $\Omega_{\pm}=dE_{\pm}/\hbar$ ($d$ being the atomic dipole strength).
345: Without loss of generality we will assume~$V_0$ to be positive since, if~$V_0$
346: is negative, it is sufficient to shift~$x$ by~$\pi/2k_L$ to recover this case.
347: 
348: In the following, it will prove convenient to work with dimensionless
349: quantities. Rescaling quantities through 
350: $\tau\DEF\delta\omega\, t$, $q\DEF2k_Lx$, 
351: $p\DEF(2k_L/M\delta\omega)p_x$, 
352: $\gamma\DEF(4k_L^2/M\delta\omega^2)V_0$ 
353: and $H_{\mathrm{eff}}\DEF(4k_L^2/M\delta\omega^2)H$, 
354: then yields the dimensionless effective Hamiltonian:
355: \begin{equation}\label{eq:ham_red}
356: H_{\mathrm{eff}}
357:         =\frac{p^2}{2}
358:         -\gamma \ (\theta + \cos\tau )\ \cos q\;.
359: \end{equation}
360: Such an Hamiltonian describes the dynamics of a periodically 
361: driven pendulum. The associated quantum canonical commutation 
362: relation is 
363: $[q,p]=i\hbar_{\mathrm{eff}}$ and we get 
364: $\hbar_{\mathrm{eff}}=8\omega_R/\delta\omega$ 
365: where $\omega_R=\hbar k_L^2/2 M$ 
366: is the atomic recoil frequency
367:  and  $\delta\omega$ is the beating frequency between the laser 
368: waves.
369: 
370: Such an effective Hamiltonian clearly exhibits two of the three 
371: properties mentioned in the introduction: the corresponding 
372: classical dynamics 
373: is governed by a single classical parameter, the dimensionless 
374: coupling strength 
375: $\gamma$, and is invariant 
376: under 
377:  time-reversal symmetry 
378: $(p,q,\tau)\mapsto (-p,q,-\tau)$. 
379: It is worth mentioning that the semiclassical limit 
380: $\hbar_{\mathrm{eff}} \to 0$ is realized here by increasing the beating
381:  frequency $\delta\omega$ 
382: between the two laser waves.
383:  
384: With our field configuration, only $\theta\ge 1$ can be achieved. 
385: As a slight generalization, we extend the range of 
386: $\theta$ to any positive value since one can
387:  design other field configurations where $\theta\le 1$ occurs. For
388: example, 
389: $\theta=0$ yields the Hamiltonian
390: studied in~\cite{Averbukh+95a} in a different 
391: context.
392: 
393: \subsection{Orders of magnitude}
394: 
395: Let us give some typical experimental parameters.
396:  For Rubidium atoms, the atomic parameters are $M=85\,\mathrm{amu}$, 
397: $\lambda_{\mathrm{at}}=2\pi c/\omega_{\mathrm{at}}=0.78\,\mathrm{\mu
398: m}$, 
399: $\Gamma/2\pi = 6\,\mathrm{MHz}$, 
400: $\omega_R/2\pi=3.8\,\mathrm{kHz}$ and saturation intensity 
401: $I_{\mathrm{sat}}=1.6\,\mathrm{mW/cm^2}$. 
402: Using far-detuned laser beams ($2\delta_L/\Gamma = 10^4$) 
403:  focussed down to $500\,\mathrm{\mu m}$ (power $100\,\mathrm{mW}$), 
404: with a beating 
405: frequency $\delta\omega/2\pi = 60\,\mathrm{kHz}$, lead to 
406: $\gamma = 0.4$ and $\hbar_{\mathrm{eff}} =0.05$. With such 
407: values, spontaneous emission can be neglected up to times of 
408: the order of few $\mathrm{ms}$. It is worth noticing the tiny energies
409: which 
410: come into play ($V_0\sim 5\,\mathrm{neV}$), by several orders of
411: magnitude
412: smaller than the typical ones for mesoscopic systems.
413: 
414:  
415: 
416: \section{Classical dynamics}\label{sec:classicaldynamics}
417: \subsection{Poincar\'e surface of section}
418: A Poincar\'e surface of section provides the usual tool for visualizing
419:  the classical dynamics~\cite{Lichtenberg/Lieberman83a}. 
420: As $H_{\mathrm{eff}}$ is $2\pi$-periodic both in
421: time and space, this surface of section simply consists 
422: in the whole phase space itself 
423: (which has the topology of a cylinder) where trajectories 
424: $(p(\tau),q(\tau))$ 
425: are seen stroboscopically at every time period $2\pi$. 
426: In the following, without any substantial loss of generality, 
427: we will restrict our analysis to the
428:  case~$\theta=1$ which is easily
429: experimentally achieved when the standing waves have the same 
430: field strengths.
431: 
432: Figure~\ref{fig:3yeux} shows stroboscopic plots
433:  of phase space orbits for different
434:  $\gamma$'s. 
435: For $\gamma=0,$ $p$ is a constant of motion, so that the
436: system is integrable and the surface of section is composed
437: of horizontal lines.
438: For $\gamma$ weak enough (fig.~\ref{fig:3yeux}-a), the orbits remain
439: confined 
440:  to invariant curves. These invariant curves stratify the whole 
441:  phase space and the dynamics appears regular. One can clearly see 
442:  well separated stability islands, each being bordered by a separatrix. 
443:  This is the situation encapsulated in the \textsc{kam} theorem 
444:  for near-integrable 
445:  motion: although no globally-defined constants of motion exist, 
446:  some invariant curves
447:  can still be 
448:  constructed which order the dynamics. As $\gamma$ is increased
449:  (fig.~\ref{fig:3yeux}-b), 
450:  some of more and more invariant curves are broken and  chaotic
451:  layers start to 
452:  spread around  separatrices. These layers fill some portion of phase 
453:  space but motion is still predominantly confined to invariant curves. 
454:  Above some coupling threshold (fig.~\ref{fig:3yeux}-c 
455: to~\ref{fig:3yeux}-e), stochastic orbits invade 
456:  phase space and the surviving stability islands are surrounded by a 
457:  connected chaotic sea. This occurs for $\gamma\sim0.1$. The phase space 
458:  structure in this regime is typical of a mixed dynamics where regular 
459:  orbits co-exist with stochastic ones. If $\gamma$ is 
460:  increased further (fig.~\ref{fig:3yeux}-f), the stability 
461: islands disappear (or are too 
462:  small to be seen at this scale)
463:  and one gets global chaos. We note 
464:  however that, even in this situation, the chaotic portion of phase
465: space 
466:  is still bounded by invariant curves which means that chaos can only
467: fully 
468:  develop within some range of momentum $p$.
469:  
470: 
471: \subsection{Resonances}\label{subsec:resonances}
472: 
473: At this stage, let us rewrite the effective 
474: Hamiltonian as follows:
475: 
476: \begin{equation}\label{eq:ham_red2}
477:         H_{\mathrm{eff}}=H_0+\gamma H_1
478:         =\frac{p^2}{2}\,-\gamma\cos q
479:          -\frac{\gamma}{2}\cos(q+\tau)
480:          -\frac{\gamma}{2}\cos(q-\tau)
481: \end{equation}
482: 
483: The physical interpretation of the various terms is rather simple:
484: the two counter-propagating laser beams at frequency $\omega_+$
485: create a stationary wave which, in turn, creates for the atom
486: an effective optical potential proportional to the
487: square of the modulus of the electric field in the standing wave,
488: hence the $\cos q$ dependence (it is actually rather
489: $1+\cos q$, but the constant term does not play any role in
490: the dynamics). The same effective potential is due to the
491: standing wave created by the two $\omega_-$ counter-propagating beams.
492: A pair of counter-propagating beams at frequencies $\omega_+$ and
493: $\omega_-$ does not create a standing wave in the lab frame. However,
494: in a frame sliding at constant velocity
495: $v_0=(\omega_+-\omega_-)c/2\omega_L,$
496: (=1 in rescaled units)
497: the two laser beams are shifted in frequency by Doppler effect
498: and appear to have equal frequency, building another stationary
499: wave and yet another effective optical potential. In the lab frame,
500: this appears as a modulated optical potential moving at velocity $v_0.$ 
501: By symmetry, there are two such effective potentials moving
502: either to the right or to the left. These are the $\cos(q\pm\tau)$
503: terms in the Hamiltonian.
504: 
505: This form of the Hamiltonian is in 
506: order to point out the perturbative terms which may be resonant
507: with the unperturbed frequencies. When~$\gamma=0$, the system is 
508: integrable since we recover free motion: $H_{\mathrm{eff}}$ reduces 
509: to $H_0=p^2/2$ and $(p,q)$~are exact action-angle variables. 
510: For $\gamma>0$, the absence of any 
511: constant of motion
512: generates chaos. Stroboscopic plots of phase-space trajectories 
513:  are no more constrained to follow lines of
514: constant~$H_0$ but generically fill densely a two dimensional volume 
515: in phase space. As long as $\gamma$ is small enough, these volumes
516: remain
517: thin enough not to be distinguished from regular lines at the scale
518: of finite precision of the measurements and/or the calculations 
519: (\textit{cf.}~figure~\ref{fig:3yeux}-a).
520:  Nevertheless, for higher values of~$\gamma$, some chaotic layers can be
521: seen
522: (\textit{cf.}~figure~\ref{fig:3yeux}-b)
523: between regular regions. They consist of portions of phase space where 
524: trajectories are exponentially sensitive on initial conditions. 
525: From classical 
526: first-order perturbation
527:  theory~\cite[chap.~2]{Lichtenberg/Lieberman83a}, 
528: we can infer that a term of the form $A \cos(sq-r\tau)$,
529: where $(s,r)$ are integers, will create a resonance of width 
530: $\Delta p = 4\sqrt{A}$ around the point $p=r/s$. In our case, $s=1$
531: and 
532: there exist 
533: only three such resonances. They are located at $p=0$ 
534: ($r=0$) and at 
535: $p=\pm1$ ($r=\pm1$). This can be seen in 
536: figure~\ref{fig:3yeux} (a to e). 
537: For each resonance there exists one stable and one unstable periodic 
538: orbit with period approximately $2\pi |r|.$ In the stroboscopic plot
539: of the surface of section, they appear as a stable and unstable fixed
540: points and give rise locally to the well-known phase-space portrait 
541: of a pendulum. In the following, we will 
542: denote by~${\mathcal{I}}_0,{\mathcal{I}}_+,{\mathcal{I}}_-$
543: the three stable islands associated with $r=0,+1,-1$ respectively. 
544: The physical interpretation of these three resonances is simple:
545: each resonance is associated with one of the modulated potentials
546: (either static or moving) described above. For example,
547: the fixed point at the center of the ${\mathcal{I}}_+$ resonance
548: is associated with a periodic orbit where the atom moves at almost
549: constant velocity $v_0,$ being is fact trapped in the minimum of the
550: moving optical potential. The other two components of the
551: potential appear along this orbit as rapidly varying potentials
552: which are adiabatically averaged to constant values.
553: As the atom can be trapped in any of the 3 modulated potentials, we
554: obtain three stable periodic orbits at the centers of the
555: 3 resonance islands.
556: 
557: For $\gamma$ small enough, the resonances are well separated and the
558: motion 
559: is quasi-integrable. Chaos will develop when the resonances start
560:  to overlap. This is the celebrated Chirikov's overlap 
561: criterion~\cite{Chirikov79a}
562:  and its evaluation gives 
563: $\gamma\,\simeq \,0.1$ in our case. 
564: Thus chaos develops in phase-space regions where the kinetic energy term 
565: and the perturbation are of the same order of magnitude. Taking into 
566: account higher perturbation orders in~$\gamma$ 
567: will shift the position in phase-space
568: of the previous resonances
569:  as well as the frequency around their stable points. For instance, 
570: it can be seen in figure~\ref{fig:3yeux}-e
571:  that the stable island~$\mathcal{I}_+$ is centered on a point having a 
572: momentum slightly larger than~$+1$.
573: Perturbation terms of higher order will also introduce
574:  other resonances of smaller size. It is precisely the 
575: overlap of the infinite cascade of such resonances which gives rise to 
576: the chaotic layers. Nevertheless, Chirikov's criterion already gives 
577: a good order of magnitude for the onset of chaos.
578: For higher $\gamma,$ the previous three resonant islands of stability
579: have shrunk inside a large chaotic sea and eventually disappear
580: completely
581: (\textit{cf.}~figure~\ref{fig:3yeux}-f). Nevertheless
582:  revival of some stable islands can still be observed for some narrow
583: windows
584: of high~$\gamma$'s.
585: In our situation, chaos cannot invade the whole phase space 
586: but is bounded by regular coasts. This is so because chaos develops 
587: where resonances overlap. Sufficiently far away from the resonances,
588: atoms move 
589: so fast that they experience an average time-independent potential. 
590: Then chaos is 
591: absent and one recovers (quasi)-free motion when $|p|\gg1$.
592:  
593: 
594: 
595: 
596: \subsection{Typical classical phase space portrait in 
597: the chaos assisted tunnelling regime}
598: The two resonant islands~$\mathcal{I}_{\pm}$,
599:  when they exist, are related by a
600: discrete symmetry: the time reversal invariance. 
601: As can be seen in figure~\ref{fig:3yeux_nonstrobo}, 
602: the atoms trapped 
603: in one island cannot classically 
604: escape from it: the boundaries of the islands play the role of a 
605: dynamical barrier which atoms cannot cross. 
606: Hence jumping from one island
607: to the other is a classically forbidden process
608: though it is expected to occur in quantum mechanics.
609: This is precisely the tunnelling situation we are interested in.
610: In fact,
611:  we will study the tunnelling
612:  between~$\mathcal{I}_+$ and $\mathcal{I}_-$ for~$\gamma$ varying from
613: 0.1 to
614:  0.3 since, in that range, classical chaos may play a revealing role
615: though
616: the two stable islands still occupy a significant volume in phase space.
617: 
618: Note that, in the physical situation described by $H_{\mathrm{eff}}$, 
619: tunnelling occurs in
620: momentum coordinate instead of space coordinate as it is usually 
621: presented 
622: in standard textbooks. The denomination of 
623: ``dynamical tunnelling''~\cite{Davis/Heller81a}, 
624: refers to this situation.
625:    The reason for investigating this situation is that manipulation 
626: of cold atoms allows for a better control (preparation and detection)
627:  of momentum rather than position.
628: 
629: 
630: 
631: 
632: 
633: \section{Quantum dynamics}\label{sec:quantumdynamics}
634: 
635: \subsection{Floquet-Bloch theory}
636: When an autonomous Hamiltonian is  spatially periodic, it is well 
637: known \cite{Ashcroft/Mermin76a} that its spectrum is organized in 
638: energy bands~$E_n(k)$. These bands are labelled by a set of integers, 
639: the band index~$n$, and depend continuously on a set of real numbers,
640: the Bloch numbers~$k$. 
641: As $E_{n}(k)$ and the associated eigenfunctions 
642: are periodic functions of the $k$'s, all the physical information
643: is contained in the first Brillouin zone. 
644: For 1D-systems, it is simply the interval
645: ~$[-\frac{\pi}{Q},\frac{\pi}{Q}[$, where $Q$ is the spatial period 
646: of the Hamiltonian. 
647: 
648: When the Hamiltonian is time-periodic, with period $T$, 
649: the analogous of Bloch theory is Floquet 
650: theory~\cite{Floquet1883a,Cherry27a,Shirley65a}. The eigenvalues of the
651:  evolution operator~$U(\tau+T,\tau)$
652:  over one period take the form ${\mathrm{e}}^{-\imat\epsilon T/\hbare}$. 
653: The $\epsilon$'s are $\tau$-independent real quantities which are 
654: called the quasi-energies of the system. 
655: Due to the time-periodicity, the quasi-energy spectrum
656: as well
657: as the associated eigenfunctions
658: are now invariant under~$\epsilon\mapsto\epsilon+2\pi\hbare/T$.
659:  
660: For $H_{\mathrm{eff}}$, the application of Bloch and Floquet theorems
661: with $Q=T=2\pi$ yields a spectrum made of quasi-energy bands
662: $\epsilon_n(k)$ 
663: where $n$ goes over the whole 
664: set of integers (for a detailed derivation see
665: appendix~\ref{app:FBformalism}). 
666: For brievety, we will 
667: define~$|{n,k,\tau}\rangle$ the ket at time~$\tau$
668:  with Bloch angle~$k$, with 
669: quasienergy~$\epsilon_n(k)$ and
670: which is a solution of the Schrodinger's equation 
671: (following the
672: notations of 
673: appendix~\ref{app:FBformalism}, we 
674: have
675: set~$|{n,k,\tau}\rangle\DEF|{\psi_{\epsilon_n(k),k}(\tau)}\rangle$).
676: We will also define~$|{n,k}\rangle\DEF|{n,k,\tau=0}\rangle$. 
677: 
678: As it can be seen in figure~\ref{fig:biperiodicite}, the band spectrum
679: has
680: the topology of a torus since it is 
681: both periodic in quasi-energy (with period $\hbare$) and 
682: in Bloch number (with period $1$).
683: 
684: \subsection{Numerical calculation of the Floquet eigenstates}
685: 
686: As derived in appendix \ref{app:FBformalism}, the Floquet
687: eigenstates can be obtained by diagonalization of the
688: Floquet-Bloch operator, $\tilde{K}$ 
689: \begin{equation}\label{eq:Ktilde2}
690: \tilde{K}(\opp,\opq,\tau, k) = \frac{(\opp+\hbare k)^2}{2} -\gamma\cos
691: \opq\ (1 + \cos\tau )  
692: -\imat\hbare\frac{d}{d\tau}\;,
693: \end{equation}
694: with periodic boundary conditions both in time and space.
695: The eigenvalues, which depend on the Bloch vector $k$, 
696: are the quasi-energies of the system. The band spectrum is
697: symmetric with respect to the axis~$k=0$ since
698: the operator~\eqref{eq:Ktilde2}
699:  is invariant
700: under the transformation~$k\mapsto-k$ and~$p\mapsto-p$. 
701: From the expression of the Floquet-Bloch operator 
702: and the boundary conditions, it is very natural
703: to expand the eigenstates on a basis set composed of products
704: of the type $\phi_{lm}(\tau,q)=\exp(\imat n\tau)\exp(\imat m q)$ 
705: which automatically obey 
706: the periodic boundary conditions.
707: In such a basis, the operator  $\tilde{K}$ has very strong selection
708: rules,
709: namely:
710: \begin{equation}
711: |\Delta n| \leq 1\ \ \ {\mathrm and}\ \ \ |\Delta m| \leq 1
712: \end{equation}
713: All matrix elements violating one of these selection rules is
714: zero. Hence, the matrix representing the operator $\tilde{K}$  in this
715: basis is sparse and banded, and all matrix elements have simple
716: analytical expressions. This is well suited for numerical
717: diagonalization (powerful algorithms exist, for example the
718: Lanczos algorithm). All the numerical results presented
719: here use this method.  We checked
720: that the effect of the truncation of the basis  is negligible: 
721: the size of the
722:  Floquet matrix is considered as sufficiently large
723:  when increasing it modifies the value of 
724: the quasi-energy on the scale of the  numerical noise only, say
725: $10^{-15}$
726:  in double precision. 
727: Not only this criterion is a proof of algorithmical
728:  convergence but also
729: it is a safeguard against numerical  discrepancies  since we are
730:  looking for exponential small quantities. 
731: 
732: \subsection{Husimi representation and classification of the quantum
733: states}
734: Classical dynamics is very illuminating when describing 
735:  the states~$|{n,k,\tau}\rangle$. In order to strengthen 
736: the correspondence between classical phase-space structures
737:  and quantum states, it is convenient to work with the Husimi 
738: representation of quantum states~\cite{Louisell73a}.
739: 
740: 
741: Such a representation associates to each quantum state 
742: $|{\psi}\rangle$ a phase space 
743: function~$\psi^{\mathrm{\scriptscriptstyle H}}(p,q)$ (where $p$ and 
744: $q$ are real numbers) defined by
745: \begin{equation}
746: \psi^{\mathrm{\scriptscriptstyle H}}(p,q) \DEF N_\psi\;|\langle{z}|{\psi}\rangle|^2        
747: \end{equation}
748: where~$|{z}\rangle$ is the normalized coherent state corresponding to
749: the 
750: complex number~$z=(q+\imat p)/\sqrt{2\hbare}$. $N_\psi$ is a 
751: $(p,q)$-independent normalization factor. Because~$|{z}\rangle$
752: is a minimal gaussian wave packet with average momentum $p$ and 
753: average position $q$, the Husimi function 
754: $\psi^{\mathrm{\scriptscriptstyle H}}(p,q)$
755: contains some information about the degree of localization 
756: of~$|{\psi}\rangle$ in phase space.
757: 
758: The minimal cell size in phase-space allowed by the Heisenberg
759: inequalities 
760: is $\hbare$. Let us see how classical phase space structures of typical
761: size
762: larger than~$\hbare$ are mirrored
763: at the quantum level.
764: In figure~\ref{fig:bandes_un} and~\ref{fig:bandes_deux} we have plotted 
765: some of the~$\epsilon_n(k)$ 
766: corresponding to the Hamiltonian~\eqref{eq:ham_red} for specified 
767: fixed values of~$\gamma$ and~$\hbare$. Von Neumann and Wigner 
768: arguments~\cite{vonNeumann/Wigner29a} claim that, 
769: generically, no exact degeneracy can occur: one rather gets 
770: avoided crossings.
771: Of course, this is relevant provided the minimal energy splitting
772: is greater than the  resolution in energy. 
773:  Some 
774: Husimi
775: functions 
776: are plotted in
777:  figure~\ref{fig:bandes_un} and~\ref{fig:bandes_deux} ($a$ to~$f$)
778: \footnote{A technicality should be mentioned here:
779: $|{u_{\epsilon,k}}\rangle$
780:         and~$|{\psi_{\epsilon,k}}\rangle$ obey some spatial boundary 
781:         conditions
782:         which are lost when working with their
783:          Husimi representations, essentially
784:          because the coherent
785:          states do not fulfill themselves these properties. 
786:         To deal with spatially (quasi-)periodic phase-space functions
787:         it is necessary to unfold the coherent
788: states\cite{Leboeuf/Voros92a} into
789:         \begin{equation}
790:                 |{z,k}\rangle\DEF\sum_{m\in{\mathbf{Z}}}
791:                         {\mathrm{e}}^{\imat mkQ}|{z+mQ}\rangle
792:         \end{equation}
793:         and define for instance
794:         \begin{equation}
795:                 \psi^{\mathrm{\scriptscriptstyle H}}_{n,k}(p,q,\tau)
796:                 \DEF N\;
797:                 |\langle{z,k}|{\psi_{n,k}(\tau)}\rangle|^2\;
798:         \end{equation}
799:          }.
800: 
801: 
802: In appendix~\ref{app:FBformalism}, it is
803: shown that we have 
804: \begin{equation}
805: \label{eq:averagev}
806: v_{n,k}\DEF\frac{1}{T}\int_0^T\langle{n,k,\tau}|\,\opp\,|{n,k,\tau}\rangle\,d\tau
807: =\frac{1}{\hbare} \frac{\partial \epsilon_n}{\partial k}\;, 
808: \end{equation}
809:  which
810: generalizes the velocity theorem~\cite[Appendix E]{Ashcroft/Mermin76a}
811: to
812: periodic time-dependent hamiltonians.
813: 
814: 
815: 
816: Figure~\ref{fig:bandes_un}-b shows an example of 
817: the Husimi representation of
818: a state with sufficiently high average velocity
819: to be alike a free eigenstate of~$H_0$. Far from quasi-degeneracies 
820: it is localized in a narrow 
821: strip of width~$\Delta p\sim2\pi\hbare/\Delta q\sim\hbare$ (since~$\Delta q$
822: covers~$2\pi$) and which is
823: centered on one of the two classical phase space trajectories of 
824: energy about~$v_{n,k}^2/2$ (compare with figure~\ref{fig:3yeux}-e).
825: Its quasi-energy band 
826: (figure~\ref{fig:bandes_un}) is an arc of the parabola of the free
827: motion but
828: can hardly be distinguished from a straight line of slope~$v_{n,k}$
829: if $k$ is restricted to one Brillouin zone. 
830: We will naturally call these states quasi-free states.
831: 
832: Some states have their Husimi functions localized in the resonant stable
833: islands (in ${\mathcal{I}}_0$ but also in~${\mathcal{I}}_\pm$).
834:  The number of these
835: states is semiclassically given by the volume of these islands divided 
836: by~$2\pi\hbare$. 
837: Far away from quasi-degeneracies,
838:  these states are at any time centered 
839:  on the 
840: stable  periodic orbit: it can be explained within a semiclassical
841: approach 
842: and can
843: be observed in figure~\ref{fig:bandes_un}-c for a state localized 
844: in~${\mathcal{I}}_0$ and 
845: in figures~\ref{fig:bandes_deux}-d,e for states in~${\mathcal{I}}_-$, 
846: ${\mathcal{I}}_+$ respectively.
847:  Their average velocity as well as their 
848: Husimi functions depend exponentially weakly on the Bloch parameter~$k$.
849: 
850: 
851: The last class of 
852: states which can be encountered corresponds to chaotic ones, that is
853: to say states whose Husimi functions are negligible on a typical
854: distance 
855: of~$\sqrt{\hbare}$ out of the chaotic
856: seas (\textit{cf.}~figure~\ref{fig:bandes_un}-a).
857:  Unlike the previous ones, their Husimi functions
858:  are very sensitive to any variation 
859: of~$k$ since they are delocalised in the whole chaotic sea which
860: spreads over all elementary  cells. The large classical distribution
861: of possible velocities is to be linked to the very fluctuating slopes of
862: the
863: quasi-energies as~$k$ is varied.
864:  
865: 
866: \subsection{Tunnelling states}\label{subsec:tunnellingstates}
867: 
868: 
869: Although the system as a whole is of course
870: time-reversal invariant, this is no longer true for its restriction
871: at a fixed value $k$ of the Bloch vector. Indeed, the operator
872: $\tilde{K}$ is not time-reversal invariant, because
873: of the crossed term $k\opp .$ In other words, the time-reversed partner
874: of a state with Bloch vector $k$ is a state with Bloch vector $-k.$
875: It is only at the special value $k=0$ (and also $k=1/2$ since $k$
876: is defined modulo 1)
877: that  $\tilde{K}$ is invariant under time reversal symmetry.
878: Therefore it corresponds to the typical situation of tunnelling 
879: between~$\mathcal{I}_{+}$ and~$\mathcal{I}_{-}$. Every island state
880: localised about~$|p|=1$ is quasi-degenerate with another one. These
881: doublets
882: represent 
883: a symmetric and an antisymmetric combination 
884: of states  localised in one island only 
885: (\textit{cf.}~figure~\ref{fig:bandes_deux}-f).
886:  The 
887: energy splitting $\Delta\epsilon_n$
888:  of these states for~$k=0$ is precisely the signature of tunnelling: 
889: it is $\pi\hbare$  
890: divided by the typical time an atom takes 
891: to jump from one island to its time-reversed image, \textit{i.e.}~to
892: reverse 
893: the sign of its velocity.   
894: 
895: 
896: \section{Chaos assisted tunnelling}\label{sec:cat}
897: 
898: 
899: \subsection{Large fluctuations}
900: 
901: After having selected the two quasi-energy bands corresponding to the 
902: two states which are
903: localized  the more deeply inside the islands~$\mathcal{I}_{\pm}$,
904: we are able to plot the splitting as a function of~$\hbare$.  
905: The great advantage of studying fluctuations when the 
906: effective Planck constant is varied is that it does not affect 
907: the classical dynamics. The behavior of the splitting is very
908: different  whether
909: chaos is present at the~$\hbare$ scale or not.
910: In the chaotic regime (\textit{cf.}~figure~\ref{fig:fluctuations_hbar} 
911: and~\ref{fig:fluctuations_gamma}), that is
912: when~$\hbare$ varies in a range where chaotic seas
913:  can be resolved, 
914: the splittings vary rapidly versus the change of any parameter, in our
915: case $\hbare.$ Moreover, the variations of the splittings, despite 
916: being perfectly deterministic, are apparently
917: erratic -- without any regular structure -- and cover
918: several orders of magnitude. They show that direct coupling to the chaotic 
919: sea is the key 
920: mechanism for their understanding and are a signature of chaos assisted 
921: tunnelling\cite{Zanardi+95a,Zakrzewski+98a}. In fact these huge fluctuations are 
922: reminiscent of the universal conductance fluctuations observed in 
923: mesoscopic systems \cite{Washburg/Webb86a,Feng/Lee91a}
924:  since 
925: tunnelling is nothing else than wave transport from one stability 
926: island to the other. 
927: In contrast,
928: in the regular regime where chaotic seas
929: are smaller than~$\hbare$, the splittings are expected 
930: to vary smoothly\cite{Wilkinson86b}.
931: 
932: In figure~\ref{fig:fluctuations_hbar}, we show the splittings of the
933: pair of
934: states
935: localized at the center of the resonances islands $\mathcal{I}_{\pm},$
936: as a 
937: function of $\hbare.$ They display huge fluctuations over about
938: 4 orders of magnitude while the general trend is a decrease as $\hbare
939: \to 0.$
940: Similarly, when plotted as a function of $\gamma,$ 
941: see figure \ref{fig:fluctuations_gamma},
942: they also display large fluctuations. The general trend is here a fast
943: increase of the typical splitting with $\gamma ;$ this is associated with
944: the
945: shrinking of the regular island when $\gamma$ increases which results
946: in a increasingly large tunnelling probability
947: The overlaps of the  regular states (still supported by the islands)
948: with the chaotic states increases. Therefore the  coupling between
949:  the two components 
950: of the tunnelling doublets which involves the chaotic states increases
951:  as well.
952:  
953: In order to understand both the general trend and the origin of the
954: fluctuations, two points of view can be used: a quantum point of view 
955: and a semiclassical one.
956: 
957: \subsection{Semiclassical interpretation}
958: If the chaotic sea is large, it is rather intuitive that it can be
959: easier to 
960: first tunnel from the center of the regular island to the chaotic sea, 
961: then propagate
962: freely in the chaotic sea to the vicinity of the symmetric island and
963: finally tunnel to the center of the symmetric island than directly
964: tunnelling
965: between the two islands. The crucial point is that because the chaotic 
966: component
967: is explored rapidly and densely, it does not cost anything to cross the
968: chaotic sea. Tunnelling trajectories can be viewed as complex
969: trajectories 
970: (i.e.
971: with complex position and momentum) which are real at both the starting
972: (in the initial island) and ending (in the symmetric island) points.
973: The tunnelling amplitude associated with a single tunnelling orbit
974: is essentially $\exp (-\mathrm{Im}(S)/\hbare)$ 
975: where $S$ is the complex action of
976: the tunnelling orbit. In a usual 1-dimensional system (like a
977: double-well),
978: there is only one such trajectory at each energy and the tunnelling rate
979: thus displays the well-known exponential decrease. In a chaotic system,
980: it may happen that there is a whole set of tunnelling trajectories whose
981: imaginary parts of the action are essentially identical. In such
982: conditions, 
983: the
984: actual tunnelling amplitude is the sum of all individual amplitudes (each
985: taken with its proper phase) which results in a very complicated
986: quantity
987: which fluctuates when parameters are changed. In some sense, this is
988: analogous
989: to the speckle pattern obtained when plenty of optical rays with various
990: geometries are randomly interfering. This is the very origin of the
991: (deterministic) fluctuations of the tunnelling rates and
992: consequently of the energy splittings. The general trend (exponential
993: decrease) is related to the typical imaginary part of the action of
994: the tunnelling trajectories.
995: 
996: 
997: \subsection{Quantum point of view}
998: A complementary quantum point of view is possible. One can divide the
999: eigenstates of the system in two subsets: the ``regular" states
1000: localized
1001: in the resonance islands and the ``chaotic" states localized in the
1002: chaotic sea.
1003: The two sets are only weakly coupled by tunnelling. Because there are 
1004: two symmetric 
1005: islands, the regular states are essentially doubly degenerate
1006: (neglecting 
1007: {\em direct}
1008: tunnelling). The chaotic sea also has the two-fold symmetry and states
1009: can be
1010: classified as even or odd. The two series of odd and even states are 
1011: ignoring each other.
1012: Hence, when by accident, a even chaotic state is almost  degenerate with
1013: a even regular state, they repell each other: at the same time, there 
1014: is usually
1015: no odd chaotic state with the same energy. Thus, the odd regular state
1016: is
1017: not significantly repelled. Hence, the splitting appears because
1018: of different shifts of the even and odd regular states. Close to any
1019: avoided crossing between either the odd or the even regular state and
1020: a corresponding chaotic state, a large splitting is obtained.
1021: On the contrary, far from any avoided crossing, the splitting is 
1022: small. Hence, the fluctuations are just associated with the
1023: existence of a large number of successive avoided crossings. The typical
1024: size of these fluctuations is related to the typical size
1025: of the avoided crossings while the typical parameter range of these
1026: fluctuations is the distance (in parameter space) between two
1027: consecutive 
1028: avoided crossings. A model implementing this idea (each regular state 
1029: is independently and randomly 
1030: coupled to the chaotic states of the same symmetry) has been
1031: proposed in~\cite{Leyvraz/Ullmo96a} and further used in \cite{Zakrzewski/Delande93a}.
1032: In this model, the chaotic states are modeled by a Hamiltonian
1033: belonging to the Gaussian Orthogonal Ensemble of random matrices
1034: while the coupling between the regular state and the chaotic
1035: state is also taken as a random Gaussian variable.
1036: With these assumptions, the splitting distribution can be
1037: calculated. Let  us denote by $\Delta$  the mean level spacing between
1038: chaotic states and by $\sigma$ the typical strength of the coupling between
1039: the regular states and the chaotic sea. Only if $\sigma \ll \Delta$
1040: is the regular state weakly coupled to the continuum (if this inequality
1041: is
1042: violated, the regular state is completely diluted in the chaotic sea
1043: by the strength of the coupling). We thus assume the inequality
1044: to be valid. Then, the distribution of splittings 
1045: $\Delta\epsilon$ is
1046: given by:
1047: \begin{equation}
1048: \left\{
1049: \begin{array}{l}
1050: \displaystyle P(\Delta \epsilon) = \frac{1}{\pi} \frac{s}{s^2+\Delta\epsilon^2}
1051: \ \ \ {\mathrm for}\ \ |\Delta\epsilon|<\sigma \\
1052: \displaystyle P(\Delta \epsilon) \simeq 0 \ \ \ {\mathrm for}\ \ |\Delta \epsilon| >
1053: \sigma
1054: \end{array}
1055: \right.
1056: \end{equation}
1057: where
1058: \begin{equation}
1059: s = \frac{\sqrt{2} \pi \sigma^2}{\Delta}
1060: \end{equation}
1061: The interpretation is rather simple. The maximum splitting is observed
1062: exactly
1063: at the avoided crossing where the levels are shifted by $\pm\sigma/2$ on
1064: both sides
1065: of their unpertubed positions. Hence, the splitting cannot be larger
1066: than about $\sigma.$
1067: In fact, there is an exponentially decreasing tail in the distribution
1068: $P(\Delta\epsilon)$
1069: (associated with the Gaussian fluctuations of $\sigma$)
1070: which we do not detail here because it is not relevant in our present
1071: case. 
1072: $s$ is the typical splitting one expects to observe: it corresponds to
1073: the
1074: shift typically due to the closest chaotic state. The full distribution
1075: is
1076: a (truncated) Cauchy distribution: it is obtained as the overall effect
1077: of all chaotic states lying above or below in energy.
1078: Note that, in the absence of the truncation of the Cauchy distribution,
1079: the average splitting is not defined because the corresponding integral
1080: diverges. Hence, it is better to discuss the typical splitting $s$
1081: rather
1082: than the average splitting. It is the slow decrease of the Cauchy
1083: distribution for large splitting which is responsible for the huge
1084: fluctuations
1085: in the splittings, which can be as large as $\sigma \gg s.$ This is
1086: reminiscent of random processes such as L\'evy flights where rare events 
1087: are dominant\cite{Bouchaud/Georges90a}.
1088: 
1089: In fig.~\ref{fig:LUlaw}, we show the statistical distribution of
1090: splittings
1091: that we obtain numerically in the chaotic regime (normalized to the typical splitting in order
1092: to work with~$s=1$). The distribution is
1093: shown
1094: on a double logarithmic scale and compared with the Cauchy distribution. 
1095: On can clearly see two regimes: for small $|\Delta\epsilon|,$
1096: $P(|\Delta\epsilon|)$
1097: is almost constant and decreases with a slope -2 for large
1098: $|\Delta\epsilon|.$
1099: The agreement with the Cauchy distribution is very good, which 
1100: proves that the model catches the essential part of the physics in this
1101: system.
1102: 
1103: The typical splitting $s$ is proportional to the square of the tunnelling
1104: matrix element from the initial state to the chaotic sea. Hence, it is
1105: expected
1106: to decrease roughly like $\exp(-2\mathrm{Im}(S)/\hbare)$ where $S$ is the complex
1107: action of tunnelling orbits (the mean level spacing scales
1108: as a power of $\hbare$ and is thus a correction to the main exponential
1109: decrease). This is roughly what is observed in
1110: fig.~\ref{fig:fluctuations_hbar}.
1111: Note however that there is a plateau in the range 
1112: $15\le\hbare^{-1}\le30.$
1113: A similar observation has been done in~\cite[fig. 3]{Roncaglia+94a}.
1114: Although this is not a crucial problem (the statistics of the 
1115: splitting distribution is not affected), a detailed explanation of this
1116: behaviour is still lacking, see however~\cite{Roncaglia+94a}.
1117: Finally, it should be interesting to calculate explicitely some
1118: of the complex tunnelling orbits in our specific system in order
1119: to compare the imaginary part of their actions with the slope
1120: in fig.~\ref{fig:fluctuations_hbar}. This work is currently under
1121: progress.
1122: 
1123: 
1124: 
1125: \section{Experimental signal}\label{sec:experimentalsignal}
1126: 
1127: 
1128: 
1129: As it can be seen in figure~\ref{fig:bandes_un}, the splittings we want 
1130: to measure
1131: correspond to very tiny scales among the other structures in the band
1132: spectrum. 
1133: For say~$\hbare\simeq0.1$, the tunnelling times are about 
1134: $\hbare/\Delta\epsilon\sim10^3$ 
1135: times the typical period (T$\sim \mathrm{few} \mu\mathrm{s}$). The observation of
1136: atoms
1137: having reversed their velocity  is therefore still possible since 
1138: during $1\ \mathrm{ms}$
1139: spontaneous emission has not begun to spoil our dynamical model.
1140: Nevertheless, 
1141: measuring tiny splittings which are hidden so deeply in the spectrum 
1142: is far from 
1143: being straightforward. 
1144: Several steps are needed: preparation of the initial state, the
1145: experiment
1146: itself (where chaos assisted tunnelling takes place) and the analysis of
1147: the
1148: final state. During the first step, one must prepare a state
1149: localized inside
1150: one resonance island, i.e. localized both in position and momentum
1151: spaces.  The idea is to use an adiabatic transfer from an initial state
1152: extended in position space by slowly branching the effective potential.
1153: Although the second step looks trivial (one just has to wait),
1154: the dispersion in the vector angle $k$ makes things much more difficult
1155: as only the $k=0$ states are related by time-reversal symmetry (see
1156: section~\ref{subsec:tunnellingstates}) and thus tunnel relatively fast. It is however possible
1157: to overcome this difficulty as explained below.
1158: Finally, the detection should be rather easy, using velocity dependent
1159: Raman transitions. We now explain in detail how the various steps
1160: can be worked out.
1161: 
1162: \subsection{Adiabatic preparation of the atoms in one
1163:  lateral stable island}
1164: 
1165: The first step consists in preparing an initial cloud of cold Rubidium
1166: atoms
1167: in order to have it 
1168: located in one of the stable island, say~$\mathcal{I}_{+}$, only.
1169: Using a standard magneto-optical trap, one can obtain a more or less
1170: thermal distribution of atoms with velocity of the
1171: order of few times the recoil velocity 
1172: $v_{\mathrm rec}=\hbar k_L/M=6~\mathrm{mm/s}$. 
1173: However -- as shown below --
1174: this is probably too much for a good measurement of the
1175: tunnelling splitting. Additional techniques (side-band
1176:  cooling\cite{Morinaga+99a}
1177: , Raman cooling\cite{Kasevich/Chu92a,Reichel+95a}) make it possible to obtain
1178: a sub-recoil velocity distribution, i.e. atoms 
1179: with an average momentum 
1180: $p_0$
1181: and a thermal dispersion~$\Delta p_x=M\Delta v=M \alpha
1182: v_{\mathrm{rec}}$
1183: with $\alpha$ significantly smaller than 1.
1184: We chose the initial momentum to be $M\delta\omega/2k_L$ so that, on
1185: average, the atoms exactly follow one of the sliding standing
1186: wave created by a pair $\omega_{\pm}$ of laser beams.
1187: 
1188: The next step is to slowly (i.e. adiabatically) switch on
1189: the standing waves. During this phase, the spatial periodicity
1190: is preserved, and the Bloch vector $k$ is thus a conserved quantity. 
1191: Initially, the momentum $p_x$ is nothing but
1192: the Bloch vector (modulo a integer multiple of the
1193: recoil momentum). Thus, by preparing a sub-recoil initial state,
1194: one populates only a small range of $k$ values and, for each 
1195: $k$ value populated, a single state (momentum eigenstate). 
1196: Otherwise stated, the initial momentum distribution has become a 
1197: statistical mixture of Bloch states with $\Delta k=\alpha/2$ 
1198: in a single energy band. 
1199: More generally, if the initial momentum distribution is not
1200: a sub-recoil one, but has a width equal to $\alpha$ recoil momenta
1201: (with $\alpha > 1$),
1202: about $\alpha$ bands will be populated.
1203: 
1204: Switching on the standing waves increases the optical potential 
1205: $V_0$ and therefore
1206: $\gamma$ from zero and  enlarges the
1207: resonance islands. In the sliding frame moving
1208:  with velocity~$p_0/M=\delta\omega/2k_L$,
1209: the atoms feel a pendulum like potential~$-\frac{1}{2}\gamma\cos q$ 
1210: (in scaled coordinates)
1211: in addition to 
1212:  some rapidly time varying terms. Consequently, they will adiabatically
1213:  localize in the potential minima, that is at the center of the
1214: resonance island.
1215: Increasing $\gamma$ localizes successively an increasing number of states
1216: in $\mathcal{I}_{+}.$
1217: The switching time must be sufficiently long as compared to the 
1218: beating period (in this case the discarded terms are still 
1219: rapidly oscillating) but also to the inverse of the minimum energy gap 
1220: (of the order of $\hbare^{-1}$ if $\Delta k$ is sufficiently narrow). 
1221: In order to trap all the initially populated states, $\gamma$ has to be
1222: sufficiently
1223: large. For a given Bloch angle, we want to localize the~$\alpha$ first
1224: states. 
1225: The quantum energies of a pendulum are given by the eigenvalues of the
1226:  Mathieu equation~\cite[chap.~20]{Abramowitz/Segun65a} (see 
1227: figures~\ref{fig:approxpendule} and~\ref{fig:approxpendulebis}).
1228:  For a pendulum whose hamiltonian
1229: is
1230: \begin{equation}
1231:         H_{\mathrm{pend}}=\frac{p^2}{2}-\frac{\gamma}{2}\cos q\;,
1232: \end{equation}
1233: the phase space volume enclosed by the separatrix is
1234: given~$16\sqrt{\gamma/2}$. 
1235: Semiclassically, it corresponds to~$16\sqrt{\gamma/2}/(2\pi\hbare)$
1236: states. 
1237: This number will be of the order of $\alpha$  when $\gamma$ reaches the
1238: value:
1239: \begin{equation}
1240:         \gamma_{\mathrm{adiab}}\simeq\frac{(\alpha \pi\hbare)^2}{128}\;.
1241: \end{equation}
1242: 
1243: Figures~\ref{fig:approxpendule} and \ref{fig:approxpendulebis}
1244: show the exact energy levels of the system together with the ones
1245: using the pendulum approximation and the Mathieu
1246: equation, as well as plots of selected Husimi representations for few
1247: eigenstates.
1248: This allows to check that :
1249: \begin{itemize}
1250: \item
1251: The pendulum approximation works well in the regime of interest, say
1252: up to $\gamma=0.1.$
1253: \item The semiclassical estimate of the number
1254: of  trapped states is sufficiently accurate for our purpose.
1255: \item The Husimi representations of the trapped states
1256: are well localized: especially, the states in
1257: figure~\ref{fig:approxpendulebis}
1258: have two well separated components in the $\mathcal{I}_{+}$ and
1259: $\mathcal{I}_{-}$ islands, meaning that we are in a real case of
1260: tunnelling.
1261: \item During the initial increase of $\gamma$, the ``ground" state is
1262: well
1263: isolated (in energy) from the other ones, which means that an adiabatic
1264: preparation is possible.
1265: \end{itemize}
1266: 
1267: 
1268: 
1269: Nevertheless, for this adiabatic preparation
1270: to be valid, we must make sure that~$\gamma$ has not reached
1271:  a range where chaos have non negligible effects
1272: on quantum properties. Physically, we want that chaotic layers 
1273: at~$\gamma=\gamma_{\mathrm{adiab}}$ have a small volume compared to the
1274: Planck 
1275: constant.
1276: From figure~\ref{fig:approxpendule},
1277:  we can obtain the upper bound limit of~$\gamma$  
1278: by estimating
1279: when the (small) avoided
1280: crossings become too large to be passed diabatically.
1281:   For~$1/50<\hbare<1/5$, 
1282: we observe that chaos have no influence 
1283: for~$\gamma<\gamma_{\mathrm{chaos}}\simeq0.04$. 
1284: 
1285: The adiabatic preparation of all the atoms in one stable island will be 
1286: achieved 
1287: if~$\gamma_{\mathrm{adiab}}<\gamma_{\mathrm{chaos}}$  that is if the
1288: atoms 
1289: are cold 
1290: enough  to have
1291: \begin{equation}\label{cond:adiabatic}
1292:                 \alpha<\frac{8\sqrt{2\gamma_{\mathrm{chaos}}}}{\pi\hbare}
1293:                 \simeq\frac{0.7}{\hbare}\;.
1294: \end{equation}
1295: In every situation considered in the following, we will have to check  
1296: that this condition 
1297: is fulfilled. Next we have to reach the desired value for $\gamma$ 
1298: ($\simeq 0.18$) in 
1299: the chaotic regime while preserving the state in the island. This will
1300: be achieved if $\gamma$ is increased sufficiently fast so that all 
1301: encountered avoided crossings with chaotic states are passed 
1302: diabatically.
1303: 
1304: Hence the whole preparation of the initial atomic state proceeds by 
1305: two steps: a first adiabatic increase of $\gamma$ at the very
1306: beginning 
1307: to significantly populate regular states in one island followed by 
1308: a diabatic increase of $\gamma$ to preserve them in the chaotic regime.
1309: 
1310: 
1311: 
1312: 
1313: \subsection{How to force tunnelling ?}
1314: 
1315: As discussed in section~\ref{subsec:tunnellingstates}, the tunnelling splitting is small only
1316: for Bloch vector $k=0.$ As it is not presently possible to prepare
1317: only this value of $k$ in a real experiment, it seems at first sight
1318: that only a small fraction of atoms (close to $k=0$) may effectively
1319: tunnel, hence considerably reducing the signal to noise ratio.
1320: A solution is to force all atoms to go through the $k\approx0$ region.
1321: The simplest idea is to impose a slow increase of the $k$ value, by
1322: adding
1323: a constant external force $F$ from outside. 
1324: Then, the full Hamiltonian which 
1325: governs the dynamics is:
1326: \begin{equation}
1327:         H'(p,q,\tau)=H(p,q,\tau)-qF\;.
1328: \label{drift}
1329: \end{equation}
1330: The potential~$V$ induces a dynamical drift in the  Bloch angle:
1331: 
1332: \begin{equation}\label{eq:kpoint3}
1333:         \frac{d}{d\tau} k(\tau)
1334:         =\frac{1}{\hbare}F\;.
1335: \end{equation} 
1336: It is shown in appendix~\ref{app:banddynamics} that this 
1337: relation, which is well-known in the time independent case 
1338: (see for instance~\cite[chap.6]{Callaway74a}), remains valid
1339: when~$H$ is periodic in time.
1340: 
1341: A convenient way of realizing experimentally such a constant force
1342: is to chirp the laser frequencies, that is make all the frequencies
1343: drift linearly in time. In an accelerated frame, the laser frequencies
1344: appear as constant, and we are back to our model. However, in this
1345: non inertial frame, the constant acceleration is translated in a
1346: constant
1347: force, hence the system is governed by equation~(\ref{drift}).
1348: This method has been used with cold atoms, see~\cite{BenDahan+96a}.
1349: 
1350: The global result is a slow drift of the $k$ distribution. This makes
1351: that the various $k$ classes successively come close to $k=0$ and
1352: are thus able to tunnel. 
1353: Whether the atom will effectively tunnel or not depends on the time
1354: scale on which $k$ changes. If $k$ varies rapidly, the avoided
1355: crossing at $k=0$ is crossed diabatically, i.e. the velocity
1356: distribution will not be modified. If $k$ varies slowly, it is crossed
1357: adiabatically (rapid adiabatic passage).  The 
1358: Landau-Zener formula\cite{Zener32a} 
1359: yields for the typical time scale for the crossover between
1360: diabatic and adiabatic crossing.
1361: 
1362: Figures~\ref{fig:suiviadiabatique} and~\ref{fig:suiviadiabatique2}
1363: illustrate,
1364:  for~$\hbare=0.2037$
1365:  the drift of  atoms
1366:  initially localized in~$\mathcal{I}_+$ (one energy band)
1367:  whose distribution in~$k$ covers
1368: 1/10 of the Brillouin zone from~$k\simeq-1/10$ to $k\simeq0$ 
1369: (sub-recoil initial velocity distribution with width of the
1370: order of $v_{\mathrm rec}/5$)
1371: Let us have the atoms evolving under the force~$F$ in order 
1372: to get a global
1373:  translation of~$1/10$ in~$k$.
1374: After a  sliding across the splitting at~$k=0$, if 
1375: the force~$F$ is weak to follow the energy level adiabatically,
1376:  the average momentum of the
1377:  atoms has reversed its sign as can be seen 
1378: on figure~\ref{fig:suiviadiabatique2}. 
1379: Measuring the critical value of~$F$
1380: for this Landau-Zener like transition to occur furnishes a mean to
1381: measure
1382: the splitting.
1383: 
1384: Another possibility would be to modulate the external force (and thus
1385: the
1386: $k$ values) periodically
1387: in time in order to induce a resonant transfer between~$\mathcal{I}_{+}$
1388: and~$\mathcal{I}_{-}$.
1389: 
1390: In any case, the method may work only if no other avoided crossing
1391: come into play. Numerical investigations show that there are mainly
1392: tiny avoided crossings along the energy curve of interest. However, as
1393: a general rule, there are usually few avoided crossings with similar
1394: or larger size than the avoided crossing of interest. If such
1395: an avoided crossing is also passed adiabatically, it will of course
1396: spoil the momentum distribution. Hence, it is crucial for the initial
1397: $k$ distribution to be sufficiently narrow to avoid this problem.
1398: Hence, a sub-recoil velocity distribution seems necessary.
1399: 
1400: \subsection{Detection}
1401: 
1402: After the atoms have interacted with the modulated waves, one can
1403: switch off the lasers either abruptly or adiabatically (in which case
1404: the atoms adiabatically leave the resonance island). In both cases,
1405: the atoms which have tunneled from $\mathcal{I}_{+}$
1406: to~$\mathcal{I}_{-}$ will end with a momentum close to $-p_0.$
1407: A standard time of flight technique should be enough to detect them.
1408: A more sophisticated technique, for example based on velocity
1409: sensitive Raman transitions~\cite{Vuletic+98a,Morinaga+99a} could also be
1410: used with a subrecoil resolution if needed
1411: \cite{BenDahan+96a}.
1412: 
1413: 
1414: 
1415:  
1416: \section{Conclusion}\label{sec:conclusion}
1417: 
1418: 
1419: In this paper we have proposed a simple and accessible 
1420: experimental configuration in which the observation of chaos 
1421: assisted tunnelling should be feasible. It consists in atoms 
1422: propagating in the light field of two far-detuned monochromatic 
1423: standing waves 
1424: with slightly different frequencies.
1425: Observing the tunnelling effect however requires sub-recoil cooling 
1426: techniques to conveniently prepare the atomic sample together 
1427: with a well-controlled experimental procedure (adiabatic preparation 
1428: of the 
1429: atomic state in one stable island followed by a drift of the 
1430:  Bloch vector). As the tunnelling period fluctuates over 
1431: several decades when the potential strength varies (see fig.8), 
1432: observing these fluctuations requires a stabilization of the laser 
1433: intensity at the level of a few percent.
1434: 
1435: 
1436: \acknowledgments Ch. M., R. K. and A. M. would like to thank 
1437: the Laboratoire Kastler Brossel
1438: for kind hospitality. CPU time on computers has been provided by IDRIS.
1439: Laboratoire Kastler Brossel de
1440: l'Universit\'e Pierre
1441: et Marie Curie et de l'Ecole Normale Sup\'erieure is
1442: UMR 8552 du CNRS.
1443: 
1444: \appendix
1445: \section{Derivation of the effective hamiltonian}
1446: \label{app:effham}
1447: 
1448: To derive the effective Hamiltonian~\eqref{eq:ham_eff}, 
1449: we basically proceed in three steps\cite{Graham+92a} :
1450: 
1451: 
1452: - We first assume that any dissipation process can be safely 
1453: ignored. Indeed for tunnelling to be observable, phase coherence 
1454: of the atomic wave-function must be preserved during all the 
1455: process. In our case, this means that spontaneous 
1456: emission must be negligible. It can be shown 
1457: \cite{CohenTannoudji90a} 
1458: that increasing 
1459: the laser detuning considerably decreases phase coherence 
1460: loss. Hence the evolution will be essentially Hamiltonian 
1461: provided each laser beam is sufficiently far-detuned from 
1462: the atomic resonance :
1463: \begin{equation}
1464: |\delta_\pm|=|\omega_\pm-\omega_{\mathrm{at}}|\gg\Gamma
1465: \end{equation}
1466: 
1467: When this holds, the total Hamiltonian operator for this 
1468: two-level system is the sum of 
1469: three terms: the kinetic energy operator describing the 
1470: center-of-mass motion of atoms of mass M %and momentum $p_x$,
1471:  the 
1472: energy operator for the internal degrees of freedom and the coupling 
1473: between internal and external degrees of freedom. 
1474: In the dipolar approximation, % the total Hamiltonian reads:
1475: the interaction is just~$dE(x,t)$ and does not depend on~$y$ and~$z$.
1476: The dynamics along~$y$ and~$z$ is just trivially described by free motion and
1477:  can be easily eliminated since the total quantum state factorizes as a
1478: plane wave in~$y$
1479: and~$z$. One is then left with the dynamics along~$x$ which is described
1480:  by the hamiltonian :
1481: \begin{equation}
1482:         H_{\mathrm{at}}= 
1483:         \frac{p_x^2}{2M}\big(|e\rangle\langle e|+|g\rangle\langle g|\big)
1484:                        +\hbar\omega_{\mathrm{at}}|e\rangle\langle e|
1485:                        -dE(x,t)\big(|e\rangle\langle g|+|g\rangle\langle e|\big)
1486: \end{equation}
1487: where~$p_x$ is the atomic momentum along~$x$. $|g\rangle$ and $|e\rangle$ 
1488: are the ground-state 
1489: and excited-state and 
1490: $d$ is the atomic dipole strength connecting them.
1491: 
1492: 
1493: - Second, we expand the total atomic state as: 
1494: \begin{equation}
1495: |\Psi\rangle = \psi_g(x,t)|g\rangle+\psi_e(x,t)
1496: \exp{(-\imat \omega_Lt)}|e\rangle
1497: \end{equation}
1498: and we drop high frequency 
1499: (optical) anti-resonant terms as far the amplitudes change slowly 
1500: during an optical period.
1501: This is known as the rotating wave approximation~\cite{Allen/Eberly87a}
1502: and features the averaging procedure to eliminate fast variables in 
1503: classical perturbation theory~\cite{Lichtenberg/Lieberman83a}.
1504: This yields the coupled amplitude equations:
1505: 
1506: \begin{subequations}\label{eq:rwa}
1507: \begin{eqnarray}
1508:  \imat\hbar\partial_t\psi_g     &=& -\frac{\hbar^2}{2M}\partial_{xx}^2\psi_g
1509:          -\frac{\hbar\Omega(x,t)}{2}\psi_e\;;\label{eq:rwaa}\\
1510:  \imat\hbar\partial_t\psi_e     &=&
1511:  -\frac{\hbar^2}{2M}\partial_{xx}^2\psi_e-\hbar\delta_L\psi_e
1512:          -\frac{\hbar\Omega^*(x,t)}{2}\psi_g\;.
1513:  \label{eq:rwab}
1514: \end{eqnarray}
1515: \end{subequations}
1516: 
1517: In these equations $\delta_L=\omega_L
1518: -\omega_{\mathrm{at}}$ is the mean
1519: laser detuning, the star denotes complex conjugation and  
1520: $\Omega(x,t)$ reads:
1521: \begin{equation}
1522: \Omega(x,t)=\big[
1523:            \Omega_+\exp(-i\delta\omega\,
1524: t/2)+\Omega_-\exp(i\delta\omega\, t/2)
1525:             \big]\cos(k_Lx)
1526: \end{equation}
1527: where $\Omega_\pm=d\,E_\pm/\hbar$ 
1528: are the Rabi frequencies of each standing wave. 
1529: 
1530: 
1531: - As a final step, we assume
1532: now that atoms initially prepared in their ground-state 
1533: mostly evolve in their ground-state. This means 
1534: that the whole atomic dynamics is solely determined by 
1535: the ground-state amplitude $\psi_g$. For this to hold, 
1536: adiabatic elimination of the excited-state 
1537: amplitude~\cite{CohenTannoudji90a} must be justified. 
1538: If the spatial partial derivatives were absent, equations 
1539: (\ref{eq:rwa}) would just describe the Rabi oscillation phenomenon. 
1540: It is then known that far off resonance, i.e. when the 
1541: frequency separation of the states is much larger than any 
1542: other frequencies, the Rabi oscillation 
1543: is very small in amplitude. A sufficient condition is :
1544: 
1545: \begin{equation}
1546:   |\delta_L| \gg \Omega_\pm, \delta \omega
1547: \end{equation}
1548: 
1549: If in addition we assume that the excited-state kinetic 
1550: energy is very small (which will be easily achieved with 
1551: cold atoms):
1552: 
1553: \begin{equation}
1554:         |\delta_L| \gg \left\langle\psi_e\left|
1555: \frac{p_x^2}{2M}\right|\psi_e\right\rangle
1556: \end{equation}
1557:  
1558: then adiabatic elimination of the excited-state 
1559: amplitude just amount to neglect the spatial and temporal 
1560: derivatives 
1561: of $\psi_e$ in equation~\eqref{eq:rwab} which is then 
1562: solved as $\psi_e\simeq-(\Omega^*/2\delta_L)\psi_g \ll \psi_g$. 
1563: It is then easy to see that the 
1564: ground-state amplitude $\psi_g$ 
1565: obeys an effective Schr\"odinger equation with Hamiltonian:
1566: \begin{equation}
1567: H=\frac{p_x^2}{2M}+\frac{\hbar|\Omega(x,t)|^2}{4\delta_L}\;.
1568: \end{equation}
1569: Eventually, up to an irrelevant purely time-dependent term, we get
1570: \begin{equation}
1571:         H=\frac{p_x^2}{2M}-V_0\cos(2k_Lx)[\theta+\cos(\delta\omega\, t)]
1572: \end{equation}
1573: where $V_0\DEF-\hbar\Omega_+
1574: \Omega_-/8\delta_L$ and 
1575: $\theta\DEF(\Omega_+/2\Omega_-)
1576: +(\Omega_-/2\Omega_+)$.
1577: 
1578: 
1579: \section{Floquet-Bloch formalism}
1580: \label{app:FBformalism}
1581: 
1582: In this appendix, we briefly recall the Floquet-Bloch formalism which is
1583: used for a quantum problem whose Hamiltonian~$H$ is periodic 
1584: both in space and in time. We will  denote $T$
1585: and $Q$ the temporal and spatial periods 
1586: respectively. 
1587: 
1588: 
1589: Let us first consider the time periodicity. We 
1590: define~\cite{Floquet1883a,Shirley65a,Zeldovich67a} 
1591: \begin{equation}\label{def:K}
1592:         K(\opp,\opq,\tau)\DEF-\imat\hbar\frac{d}{d\tau}+
1593:                              H(\opp,\opq,\tau)
1594: \end{equation}
1595: where $\opp$ and $\opq$ stand for canonical Hermitian operators whose 
1596: commutator is~$[\opp,\opq]=-\imat\hbar$.
1597: 
1598: If $U(\tau',\tau)$ denotes the unitary evolution operator from 
1599: $\tau$ to $\tau'$ associated with Hamiltonian $H,$
1600: the periodicity of the dynamics
1601: implies that~$U(\tau+T,T)=U(\tau,0)$ and
1602: \begin{equation}
1603:         U(\tau+T,\tau)=U(\tau+T,T)\;U(T,0)\;U(0,\tau)
1604:                     =[U(0,\tau)]^{-1}\;U(T,0)\;U(0,\tau)\;.
1605: \end{equation}
1606: This shows that $U(\tau+T,\tau)$ and $U(T,0)$
1607:   differ
1608:  by a unitary
1609: transformation and hence have the same spectrum (but of course
1610: different eigenvectors), independent of $\tau$.
1611: The eigenvalues of $U(\tau+T,\tau)$ have unit modulus
1612: and can be written as ${\mathrm{e}}^{-\imat\epsilon_n T/\hbar}$
1613: where $\epsilon_n$ is the so-called quasi-energy defined
1614: modulo $2\pi\hbar/T.$ If $|\psi_n(\tau)\rangle$ denotes the
1615: corresponding
1616: eigenvector, we can define the Floquet state:
1617: \begin{equation}
1618: \label{def:fstate}              
1619:                 |{\chi_n(\tau)}\rangle\DEF
1620:                 {\mathrm{e}}^{\imat\epsilon_n \tau/\hbar}|
1621:                 {\psi_n(\tau)}\rangle
1622: \end{equation}
1623: which is by construction periodic with period $T$. 
1624: 
1625: Inserting the definition of the Floquet state 
1626: in the time-dependent Schr\"odinger equation, we immediately obtain:
1627: \begin{equation}
1628:                 K(\opp,\opq,\tau)\,|{\chi_n(\tau)}\rangle
1629:                 =\epsilon_n\,|{\chi_n(\tau)}\rangle\;.
1630: \end{equation}
1631: which means that the quasi-energy spectrum is obtained by diagonalizing
1632: the Floquet Hamiltonian in the space of time-periodic functions.
1633: 
1634: The second step consists in making use of the invariance of $H$ under 
1635: spatial translations with period $Q$. 
1636: The unitary translation operator~$\widehat{T}_Q\DEF
1637: {\mathrm{e}}^{-\imat \opp Q/\hbar}$
1638: commutes with~$K$. We can then use the spatial counterpart of the
1639: Floquet 
1640: theorem, namely the Bloch theorem \cite{Ashcroft/Mermin76a}, 
1641: and label the eigenstates of $K$ with 
1642: the Bloch number~$k\in[-\pi/Q,\pi/Q[$ (the first Brillouin zone),
1643: which means diagonalising $K$ in each subspace with fixed $k.$
1644: If one defines the Floquet-Bloch states as:
1645: 
1646: \begin{equation}
1647:         |{u_{n,k}(\tau)}\rangle={\mathrm{e}}^{-\imat k
1648: \opq}|{\chi_{n,k}(\tau)}\rangle
1649:         ={\mathrm{e}}^{\imat\epsilon_{n}(k) \tau/\hbar}
1650:                              {\mathrm{e}}^{-\imat k \opq}
1651: |{\psi_{n,k}(\tau)}\rangle\;,
1652: \end{equation}
1653: where~$\{{\psi_{n,k}(\tau)}\rangle\}$ forms a complete orthogonal eigenbasis, it is easy to show that they can be obtained by diagonalizing the
1654: Floquet-Bloch Hamiltonian:
1655: \begin{equation}
1656: \tilde{K}(\opp,\opq,\tau, k) = K(\opp+\hbar k,\opq,\tau)
1657: \end{equation}
1658: on the subspace of time and space periodic functions.
1659: In our specific case, the
1660: Floquet-Bloch Hamiltonian reads:
1661: \begin{equation}\label{eq:tildeK}
1662: \tilde{K}(\opp,\opq,\tau, k) = \frac{(\opp+\hbar k)^2}{2} -\gamma
1663: \ (\theta + \cos\tau )\  \cos \opq
1664: -\imat\hbar\frac{d}{d\tau}\;.
1665: \end{equation}
1666: 
1667: The spatial periodicity of the Floquet-Bloch states leads to a discrete
1668: set of
1669: dispersion relations~$\epsilon_n(k)$ . 
1670: For fixed~$n$, the set of all 
1671: quasi-energies~$\epsilon_n(k)$ for
1672: $k$~in the first Brillouin zone~$[-\pi/Q,\pi/Q[$ is called 
1673: the~$n\mathrm{th}$ band of the system.
1674: 
1675: Let us now obtain the velocity
1676: theorem~\eqref{eq:averagev}. Using the above relations, we have
1677: $\langle\psi_{\epsilon,k}(\tau)|\,\opp\,|{\psi_{\epsilon,k}(\tau)}\rangle=
1678: \langle u_{\epsilon,k}(\tau)|\,(\opp+\hbar
1679: k)\,|{u_{\epsilon,k}(\tau)}\rangle= \langle
1680: u_{\epsilon,k}(\tau)|\big(\hbar^{-1}\partial \tilde{K}/\partial
1681: k\big)|{u_{\epsilon,k}(\tau)}\rangle $. The derivation with respect
1682: to~$k$
1683: of the relation~$\langle
1684: u_{\epsilon,k}(\tau)|\,\tilde{K}\,|{u_{\epsilon,k}(\tau)}\rangle=\epsilon(k)$
1685: leads to
1686:  \begin{multline}\label{eq:psippsi}
1687: \langle\psi_{\epsilon,k}(\tau)|\,\opp\,|{\psi_{\epsilon,k}(\tau)}\rangle\\=
1688:          \frac{1}{\hbar}\frac{\partial \epsilon}{\partial k}
1689:          +\imat\hbar\frac{d}{d\tau}\left(
1690:                                 \langle u_{\epsilon,k}(\tau)|
1691:                                 \frac{\partial}{\partial k}
1692:                                 |{u_{\epsilon,k}(\tau)}\rangle
1693:                                 \right)
1694:          -\left(
1695:                 \frac{1}{\hbar}\frac{\partial}{\partial k}\langle
1696: u_{\epsilon,k}(\tau)|
1697:           \right)
1698:           \tilde{K}\,|{u_{\epsilon,k}(\tau)}\rangle
1699:          -\left[\left(
1700:                 \frac{1}{\hbar}\frac{\partial}{\partial k}\langle
1701: u_{\epsilon,k}(\tau)|
1702:                  \right)
1703:                   \tilde{K}\,|{u_{\epsilon,k}(\tau)}\rangle
1704:           \right]^*\;. 
1705: \end{multline} 
1706: The two last terms of the the right hand
1707: side
1708: are opposite since the normalization of the $u$'s leads to
1709: $\partial\langle u_{\epsilon,k}|u_{\epsilon,k}\rangle /\partial k=0$.
1710: Moreover, after time-averaging~\eqref{eq:psippsi} over~$T$, the total
1711: $\tau$-derivative vanishes since the $u$'s are precisely~$T$-periodic
1712: while the time-independent $k$-derivative of the quasi-energy remains
1713: unchanged. Eventually,
1714:  \begin{equation}
1715:         \frac{1}{T}\int_0^T  
1716: \langle\psi_{\epsilon,k}(\tau)|\,\opp\,|{\psi_{\epsilon,k}(\tau)}\rangle
1717:         \,d\tau=\frac{1}{\hbar}\frac{\partial \epsilon}{\partial k}\;.
1718: \end{equation} 
1719: 
1720: 
1721: 
1722: \section{Bloch angle dynamics}
1723: \label{app:banddynamics}
1724: In this appendix we derive equation~(\ref{eq:kpoint3}) 
1725: which is valid for an arbitrary 
1726: strength of the constant force~$F$ provided that the potential~$V=-Fq$ remains
1727: strictly linear in~$q$.
1728: Let us choose a state~$|\psi(\tau)\rangle$ evolving under~$H'=H+V$ such 
1729: that it coincides with a Floquet-Bloch state at~$\tau=0$:
1730: 
1731: \begin{equation}
1732: 	\imat\hbare\frac{d\,|{\psi(\tau)}\rangle}{d\tau}
1733:         =H'(\opp,\opq,\tau)\,|{\psi(\tau)}\rangle
1734: \end{equation}
1735: and
1736: \begin{equation}
1737: 	|\psi(\tau=0)\rangle=|\psi_{n,k}(\tau=0)\rangle\;.
1738: \end{equation}
1739: In the interaction picture we have immediately
1740: \begin{equation}\label{eq:intpicture1}
1741: 	\imat\hbare\frac{d\,|{\psi^I(\tau)}\rangle}{d\tau}
1742:         =-F U^\dagger(\tau,0)\opq U(\tau,0)\,|{\psi^I(\tau)}\rangle
1743: \end{equation}
1744: where~$|{\psi^I(\tau)}\rangle\DEF U^\dagger(\tau,0)|\psi(\tau)\rangle$
1745: and where~$U$ denotes the evolution operator under~$H$.
1746: Let~$|\phi(\tau)\rangle$ be the ket defined by
1747: \begin{equation}\label{eq:intpicture2}
1748: 	|\phi(\tau)\rangle\DEF{\mathrm{e}}^{-\imat F\opq \tau/\hbare}|{\psi^I(\tau)}\rangle\;.
1749: \end{equation}
1750: It is straightforward to obtain its evolution :
1751: \begin{equation}
1752: 	\imat\hbare\frac{d\,|{\phi(\tau)}\rangle}{d\tau}
1753:         =G(\tau)\,|{\phi(\tau)}\rangle
1754: \end{equation}
1755: where
1756: \begin{equation}
1757: 	G(\tau)\DEF F\left(\opq-{\mathrm{e}}^{-\imat\tau F\opq/\hbare}
1758:                                 \,U^\dagger(\tau,0)\,\opq\, U(\tau,0)
1759:                      		{\mathrm{e}}^{\imat\tau F\opq/\hbare}
1760:                      \right)\;.
1761: \end{equation}
1762: Since~$U^\dagger(\tau,0)$ commutes with the translation 
1763: operator~$\widehat{T}_Q$, it can be checked that~$[G(\tau),\widehat{T}_Q]=0$.
1764: The evolution of~$|{\phi(\tau)}\rangle$ under~$G$ will therefore  preserve its initial 
1765:  quantum number~$k$ :
1766: \begin{equation}
1767: 	\widehat{T}_Q|{\phi(\tau)}\rangle={\mathrm{e}}^{-\imat k Q}|{\phi(\tau)}\rangle
1768: \end{equation}
1769: for all~$\tau$. Thus, making use of~\eqref{eq:intpicture1} and~\eqref{eq:intpicture2}, we have
1770: \begin{equation}
1771: 	\widehat{T}_Q|{\psi(\tau)}\rangle
1772:         ={\mathrm{e}}^{-\imat (k+F\tau/\hbare) Q}|{\psi(\tau)}\rangle
1773: \end{equation}
1774: which shows that~$|{\psi(\tau)}\rangle$ is actually a Bloch wave with
1775:  a Bloch angle given by~$k(\tau)=k(0)+F\tau/\hbare$ even if it spreads
1776:  among the quasi-energy bands.
1777: 
1778: 
1779: \begin{thebibliography}{10}
1780: 
1781: \bibitem{Giannoni+91a}
1782:  in {\em Chaos et Physique Quantique --- Chaos and Quantum Physics}, {L}es
1783:   {H}ouches, {\'e}cole d'{\'e}t{\'e} de physique th{\'e}orique 1989, session
1784:   {LII}, edited by M. Giannoni, A. Voros, and J. {Zinn-Justin}
1785:   ({N}orth-{H}olland, {A}msterdam, 1991).
1786: 
1787: \bibitem{Heller84a}
1788: E. Heller, Phys. Rev. Lett. {\bf 53,},  1515  (1984).
1789: 
1790: \bibitem{Akkermans+95a}
1791:  in {\em Mesoscopic quantum physics}, {L}es {H}ouches, {\'e}cole d'{\'e}t{\'e}
1792:   de physique th{\'e}orique 1994, session {LXI}, edited by E. Akkermans, G.
1793:   Montambaux, J. Pichard, and J. {Zinn-Justin} ({N}orth-{H}olland, {A}msterdam,
1794:   1995).
1795: 
1796: \bibitem{Messiah65a}
1797: A. Messiah, {\em M{\'e}canique Quantique~(2~vol.)} (Dunod, Paris, 1964),
1798:   english translation: {N}orth-{H}olland ({A}msterdam).
1799: 
1800: \bibitem{Balian/Bloch74a}
1801: R. Balian and C. Bloch, Ann. Physics {\bf 84},  559  (1974).
1802: 
1803: \bibitem{Wilkinson86b}
1804: M. Wilkinson, Physica D {\bf 21},  341  (1986).
1805: 
1806: \bibitem{Wilkinson/Hannay87a}
1807: M. Wilkinson and J.~H. Hannay, Physica D {\bf 27},  201  (1987).
1808: 
1809: \bibitem{Lin/Ballentine90a}
1810: W. Lin and L. Ballentine, Phys. Rev. Lett. {\bf 65},  2927  (1990).
1811: 
1812: \bibitem{Bohigas+93a}
1813: O. Bohigas, D. Boos{\'e}, R. Egydio~de Carvalho, and V. Marvulle, Nuclear Phys.
1814:   A {\bf 560},  197  (1993).
1815: 
1816: \bibitem{Bohigas+93b}
1817: O. Bohigas, S. Tomsovic, and D. Ullmo, Phys. Rep. {\bf 223},  43  (1993).
1818: 
1819: \bibitem{Tomsovic/Ullmo94a}
1820: S. Tomsovic and D. Ullmo, Phys. Rev. E {\bf 50},  145  (1994).
1821: 
1822: \bibitem{Creagh/Whelan96a}
1823: S.~C. Creagh and N.~D. Whelan, Phys. Rev. Lett. {\bf 77},  4975  (1996).
1824: 
1825: \bibitem{Averbukh+95a}
1826: V. Averbukh, N. Moiseyev, B. Mirbach, and H.~J. Korsh, Z. Phys. D {\bf 35},
1827:   247  (1995).
1828: 
1829: \bibitem{Arimondo+92a}
1830:  in {\em Laser and manipulation of atoms and ions}, Enrico Fermi international
1831:   summer school, Course CXVIII, 9-19 July 1991, edited by E. Arimondo, W.~D.
1832:   Phillips, and F. Strumia ({N}orth-{H}olland, {A}msterdam, 1992).
1833: 
1834: \bibitem{Berman96a}
1835: {\em Atom Interferometry}, edited by P.~R. Berman (Academic Press, {N}ew
1836:   {Y}ork, 1996).
1837: 
1838: \bibitem{Weidemuller+95a}
1839: M. Weidem{\"u}ller, A. Hemmerich, A. G{\"o}rlitz, T. Esslinger, and T.~W.
1840:   H{\"a}nsch, Phys. Rev. Lett. {\bf 75},  4583  (1995).
1841: 
1842: \bibitem{Guidoni+97a}
1843: L. Guidoni, C. Trich{\'e}, P. Verkerk, and G. Grynberg, Phys. Rev. Lett. {\bf
1844:   79},  3363  (1997).
1845: 
1846: \bibitem{Drese/Holthaus97a}
1847: K. Drese and M. Holthaus, Phys. Rev. Lett. {\bf 78},  2932  (1997).
1848: 
1849: \bibitem{BenDahan+96a}
1850: M. Ben~Dahan, E. Peik, J. Reichel, Y. Castin, and C. Salomon, Phys. Rev. Lett.
1851:   {\bf 76},  4508  (1996).
1852: 
1853: \bibitem{Niu+95a}
1854: Q. Niu, X.-G. Zhao, G.~A. Georgakis, and M.~G. Raizen, Phys. Rev. Lett. {\bf
1855:   76},  4504  (1996).
1856: 
1857: \bibitem{Moore+94a}
1858: F.~L. Moore, J.~C. Robinson, C. Bharucha, P.~E. Williams, and M.~G. Raizen,
1859:   Phys. Rev. Lett. {\bf 73},  2974  (1994).
1860: 
1861: \bibitem{Labeyrie+99a}
1862: G. Labeyrie, F. de~Tomasi, J. Bernard, C.~A. M{\"u}ller, C. Miniatura, and R.
1863:   Kaiser, Phys. Rev. Lett. {\bf 83},  5266  (1999).
1864: 
1865: \bibitem{Jonckheere+00a}
1866: T. Jonckheere, C.~A. M{\"u}ller, R. Kaiser, C. Miniatura, and D. Delande, Phys.
1867:   Rev. Lett. {\bf 85},  4269  (2000).
1868: 
1869: \bibitem{CohenTannoudji+88a}
1870: C. Cohen-Tannoudji, J. Dupont-Roc, and G. Grynberg, {\em Processus
1871:   d'interaction entre photons et atomes}, {\em Savoirs actuels}
1872:   ({I}nter{E}ditions/{E}ditions du \textsc{cnrs}, Paris, 1988), english
1873:   translation: {W}iley and sons ({N}ew {Y}ork).
1874: 
1875: \bibitem{Lichtenberg/Lieberman83a}
1876: A.~J. Lichtenberg and M.~A. Lieberman, {\em Regular and Stochastic Motion},
1877:   Vol.~38 of {\em Applied Mathematical Sciences} ({S}pringer-{V}erlag, {N}ew
1878:   {Y}ork, 1983).
1879: 
1880: \bibitem{Chirikov79a}
1881: B. Chirikov, Phys. Rep. {\bf 52},  263  (1979).
1882: 
1883: \bibitem{Davis/Heller81a}
1884: M.~J. Davis and E.~J. Heller, J. Phys. Chem. {\bf 85},  307  (1981).
1885: 
1886: \bibitem{Ashcroft/Mermin76a}
1887: N.~W. Ashcroft and N.~D. Mermin, {\em Solid State Physics} (Saunders College,
1888:   Philadelphia, 1976).
1889: 
1890: \bibitem{Floquet1883a}
1891: G. Floquet, Ann. de l'{\'E}c. Normale, {$2^{\mbox{\small e}}$} S{\'e}rie {\bf
1892:   12},  47  (1883), (in french).
1893: 
1894: \bibitem{Cherry27a}
1895: T.~M. Cherry, Proc. London Math. Soc. (2nd ser.) {\bf 26},  211  (1927).
1896: 
1897: \bibitem{Shirley65a}
1898: J.~H. Shirley, Phys. Rev. {\bf 138},  B979  (1965).
1899: 
1900: \bibitem{Louisell73a}
1901: W.~H. Louisell, {\em Quantum statistical properties of radiation} (Wiley,
1902:   chichester, 1973).
1903: 
1904: \bibitem{vonNeumann/Wigner29a}
1905: J. von Neumann and E. Wigner, Phys. Z. {\bf 30},  467  (1929), english
1906:   translation in \cite{Knox/Gold64a} pp. 167--172: {O}n the Behavior of the
1907:   Eigenvalues in Adiabatic Processes.
1908: 
1909: \bibitem{Leboeuf/Voros92a}
1910: P. Leb{\oe}uf and A. Voros,  in {\em Quantum chaos --- Between order and
1911:   disorder}, edited by G. Casati and B. Chirikov ({C}ambridge University Press,
1912:   {C}ambridge, 1992), pp.\ 507--533.
1913: See Ref.\ \cite{Casati/Chirikov95b}.
1914: 
1915: \bibitem{Zanardi+95a}
1916: E.~M. Zanardi, Guti{\'e}rrez, and J.~M. Gomez~Llorente, Phys. Rev. E {\bf 5},
1917:   4736  (1995).
1918: 
1919: \bibitem{Zakrzewski+98a}
1920: J. Zakrzewski, D. Delande, and A. Buchleitner, Phys. Rev. E {\bf 57},  1458
1921:   (1998).
1922: 
1923: \bibitem{Washburg/Webb86a}
1924: S. Washburg and R. Webb, Adv. in Phys. {\bf 35},  375  (1986).
1925: 
1926: \bibitem{Feng/Lee91a}
1927: S. Feng and P.~A. Lee, Science {\bf 251},  633  (1991).
1928: 
1929: \bibitem{Leyvraz/Ullmo96a}
1930: F. Leyvraz and D. Ullmo, J. Phys. A {\bf 29},  2529  (1996).
1931: 
1932: \bibitem{Zakrzewski/Delande93a}
1933: J. Zakrzewski and D. Delande, Phys. Rev. E {\bf 47},  1650  (1993).
1934: 
1935: \bibitem{Bouchaud/Georges90a}
1936: J.-P. Bouchaud and A. Georges, Phys. Rep. {\bf 195},  127  (1990).
1937: 
1938: \bibitem{Roncaglia+94a}
1939: R. Roncaglia, L. Bonci, F.~M. Izrailev, B.~J. West, and P. Grigolini, Phys.
1940:   Rev. Lett. {\bf 73},  802  (1994).
1941: 
1942: \bibitem{Morinaga+99a}
1943: M. Morinaga, I. Bouchoule, J.-C. Karam, and C. Salomon, Phys. Rev. Lett. {\bf
1944:   83},  4037  (1999).
1945: 
1946: \bibitem{Kasevich/Chu92a}
1947: M. Kasevich and S. Chu, Phys. Rev. Lett. {\bf 69},  1741  (1992).
1948: 
1949: \bibitem{Reichel+95a}
1950: J. Reichel, F. Bardou, M. Ben~Dahan, E. Peik, S. Rand, C. Salomon, and C.
1951:   Cohen-Tannoudji, Phys. Rev. Lett. {\bf 75},  4575  (1995).
1952: 
1953: \bibitem{Abramowitz/Segun65a}
1954: M. Abramowitz and I.~A. Segun, {\em Handbook of mathematical functions}
1955:   ({D}over publications, {N}ew {Y}ork, 1965).
1956: 
1957: \bibitem{Callaway74a}
1958: J. Callaway, {\em Quantum Theory of the Solid State} (Academic Press, {L}ondon
1959:   and {N}ew {Y}ork, 1974), Vol.~B.
1960: 
1961: \bibitem{Zener32a}
1962: C. Zener, Proc. Roy. Soc. London Ser. A {\bf 137},  696  (1932).
1963: 
1964: \bibitem{Vuletic+98a}
1965: V. Vuleti{\'c}, C. Chin, A.~J. Kerman, and S. Chu, Phys. Rev. Lett. {\bf 81},
1966:   5768  (1998).
1967: 
1968: \bibitem{Graham+92a}
1969: R. Graham, M. Schlautmann, and P. Zoller, Phys. Rev. A (3) {\bf 45},  R19
1970:   (1992).
1971: 
1972: \bibitem{CohenTannoudji90a}
1973: C. Cohen-Tannoudji,  in {\em Syst{\`e}mes fondamentaux en optique quantique ---
1974:   Fundamental Systems in Quantum Optics}, {L}es {H}ouches, {\'e}cole
1975:   d'{\'e}t{\'e} de physique th{\'e}orique 1990, session {LIII}, edited by J.
1976:   Dalibard, J.-M. Raimond, and J. {Zinn-Justin} ({N}orth-{H}olland,
1977:   {A}msterdam, 1990), pp.\ 1--164.
1978: 
1979: \bibitem{Allen/Eberly87a}
1980: A.~L. C. and J. Eberly, {\em Optical Resonance and Two Level Atoms} (Dover,
1981:   {N}ew {Y}ork, 1987).
1982: 
1983: \bibitem{Zeldovich67a}
1984: Y.~B. Zeldovich, Soviet Phys. JETP {\bf 24},  1006  (1967), translated from the
1985:   original article (1966) in russian.
1986: 
1987: \bibitem{Casati/Chirikov95b}
1988: {\em Quantum chaos --- Between order and disorder}, edited by G. Casati and B.
1989:   Chirikov ({C}ambridge University Press, {C}ambridge, 1995).
1990: 
1991: \bibitem{Knox/Gold64a}
1992: R.~S. Knox and A. Gold, {\em Symmetry in the solid state}, {\em Lecture notes
1993:   and supplements in physics} ({{W}. {A}. {B}enjamin, Inc}, {N}ew {Y}ork,
1994:   1964).
1995: 
1996: \end{thebibliography}
1997:             
1998: 
1999: \newpage \ 
2000: 
2001: \begin{figure}[!ht]%\ \vspace{-10cm}
2002: \center
2003: \psfig{file=piegeos.ps,width=18cm}%\input{3yeux.pstex_t}
2004: \caption{\label{fig:piegeos}Experimental configuration under
2005: consideration:
2006: a cloud of two-level atoms is exposed to two monochromatic standing waves with
2007: frequencies
2008: $\omega_\pm = \omega_L \pm \delta\omega/2$ $(\delta\omega\ll\omega_L)$.
2009:  All fields are
2010: linearly polarised along the same direction and are sufficiently
2011: far-detuned
2012: from the
2013: atomic resonance so that dissipation effects can be ignored.}
2014: \end{figure}
2015: 
2016: 
2017: \begin{figure}[!ht]%\ \vspace{-10cm}
2018: \center
2019: %\hspace{-3.5cm}
2020: \psfig{file=3yeux.ps,height=19cm}%\input{3yeux.pstex_t}
2021: \caption{\label{fig:3yeux} Stroboscopic plots of trajectories in phase
2022:  space for different initial conditions at~$\tau=0$ and
2023: different~$\gamma$'s. 
2024: The classical dynamics is governed by Hamiltonian~\eqref{eq:ham_red}
2025: with 
2026: $\theta=1$. At low $\gamma$ values, resonance islands are visible
2027: separated by quasi-free motion. As $\gamma$ increases, the resonance
2028: islands
2029: grow and chaos appears close to the separatrices. The situation of
2030: interest
2031: for chaos assisted tunnelling is when two symmetric islands are separated
2032: by
2033: a chaotic sea, as $\mathcal{I}_+$ and ${\mathcal{I}}_-$ in (d) and (e).
2034: }
2035: \end{figure}
2036: 
2037: \begin{figure}[!ht]%\ \vspace{-10cm}
2038: \center
2039: %\hspace{-3.5cm}
2040: %\center
2041: \psfig{file=3yeux_nonstrobo.ps}
2042: \caption{\label{fig:3yeux_nonstrobo} Plots of some trajectories in phase
2043:  space
2044: for different initial conditions ($\gamma=0.1$). In $A,$
2045: we display
2046: the usual Poincar\'e surface of section like in figure~\ref{fig:3yeux},
2047: that is 
2048: stroboscopic plots which are fold according to the spatial 
2049: periodicity. The small  black disks show some  initial conditions.
2050:    In~$B$ 
2051: and~$C,$ trajectories are plotted every $2\pi/100$ and,
2052: unlike in~$A$, we let evolve $q(\tau)$ continously 
2053: outside~$[-\pi,\pi[$.  
2054:  Trajectories~$b$ and $f$  are trapped 
2055: in the $\mathcal{I}_+$ and ${\mathcal{I}}_0$ resonance islands respectively.
2056: Trajectories $a$ and~$e$ are  two examples of regular quasi-free 
2057: motion.
2058: $c$ and~$d$ correspond to chaotic motion, their initial conditions 
2059: (at $\tau=0$) lie in the chaotic sea of figure~\ref{fig:3yeux}-d. 
2060:         }
2061: \end{figure}
2062: 
2063: \begin{figure}[!ht]%\ \vspace{-10cm}
2064: \center
2065: %\hspace{-3.5cm}
2066: \psfig{file=biperiodicite.ps}
2067: \caption{\label{fig:biperiodicite} For a Hamiltonian which is 
2068:  $Q$-periodic in space
2069:  and $T$-periodic in time, the quasi-energy spectrum is made of bands
2070: which are $2\pi/Q$-periodic in the Bloch numbers
2071:  and $2\pi\hbare/T$-periodic
2072: in the quasi-energies. Such a spectrum is shown here for
2073:  Hamiltonian~\eqref{eq:ham_red} ($Q=2\pi$, $T=2\pi$) with $\gamma=0.18$ 
2074: and $\hbare=3.1787$.}
2075: \end{figure}
2076: 
2077: \begin{figure}[!ht]
2078: \center
2079: \psfig{file=bandes_un.ps}
2080: \caption{\label{fig:bandes_un} Quasi-energy bands for~$\gamma=0.18$ 
2081: and~$\hbare=0.2037$ and some Husimi representations of  typical states. 
2082: In~$b,$ we
2083: show a quasi-free state with a well defined average velocity 
2084: (the derivative of the energy level with respect to $k$)
2085: localised in phase space on regular trajectories 
2086: (compare with figure~\ref{fig:3yeux}-e). On this
2087: scale, 
2088: the avoided crossings
2089: with other bands cannot be resolved.
2090: In~$c,$ we show a state 
2091: localised in the central stable island~${\mathcal{I}}_0$ 
2092: (actually the ``ground state'').
2093:  Far from quasi-degeneracies
2094: the average velocity of this state is zero.
2095: In~$a,$  we give an example of a chaotic state, 
2096:  whose Husimi function is
2097: localised in the chaotic sea (compare with figure~\ref{fig:3yeux}-e). 
2098: Unlike the former states, 
2099: the average velocity 
2100: is fluctuating when varying the Bloch angle~$k$. 
2101: The band spectrum is
2102: symmetric with respect to the axis~$k=0$ since
2103: the operator~\eqref{eq:Ktilde2}
2104:  is invariant
2105: under the transformation~$k\mapsto-k$ and~$\opp\mapsto-\opp$. 
2106: The tunnelling 
2107: situation
2108: due to the time reversal symmetry corresponds to the dashed squared zone 
2109: (around~$k=0$) which is enlarged
2110: in figure~\ref{fig:bandes_deux}.
2111:         }
2112: \end{figure}
2113: 
2114: \begin{figure}[!ht]
2115: \center
2116: \psfig{file=bandes_deux.ps,width=18cm}
2117: \caption{\label{fig:bandes_deux} Quasi-energy bands for~$\gamma=0.18$ 
2118: and~$\hbare=0.2037$ and some Husimi representations of  typical states.
2119: It is a zoom of the dashed squared zone in figure~\ref{fig:bandes_un}.
2120: For~$k\neq0$, one can find states, like in~$d$ or~$e,$ 
2121: whose Husimi function is
2122:  localised  in one stable island~${\mathcal{I}}_-$ 
2123: or~${\mathcal{I}}_+$ (compare with figure~\ref{fig:3yeux}-e). 
2124: In the frame where the center of the 
2125: island is fixed, these states correspond to the ``ground'' states 
2126: (some  excited states may of course exist if~$\hbare$ is small 
2127: enough as can be seen in figures 
2128: \ref{fig:suiviadiabatique} 
2129: and~\ref{fig:suiviadiabatique2}).
2130:  For~$k=0$, we recover time reversal 
2131: symmetry through the existence of quasidegenerate doublets of symmetric 
2132: or antisymmetric
2133: combinations. This is the typical tunnelling situation: following 
2134: adiabatically with~$k$
2135: the state in~$e$ (which has an average velocity
2136: about~$+1$), by decreasing~$k$
2137: we get the state in~$d$ which has a reversed velocity. 
2138: This reversal of the velocity
2139: is a classically forbidden process (compare with orbit~$b$ in 
2140: figure~\ref{fig:3yeux_nonstrobo}). 
2141:         }
2142: 
2143: \end{figure}
2144: \begin{figure}[!ht]
2145: \center
2146: \psfig{file=fluctuations_hbar.ps}
2147: \caption{\label{fig:fluctuations_hbar} Fluctuations of the 
2148: energy splittings~$\Delta\epsilon_n$ between pairs of
2149: symmetric/antisymmetric
2150: states localized in the~$\mathcal{I}_{\pm}$ resonance islands (here
2151: shown
2152: for the ``ground state" inside the island).
2153: The classical dynamics is fixed at~$\gamma=0.18$ 
2154: (compare with figure~\ref{fig:3yeux}-e). The existence 
2155: of large fluctuations
2156:  over several orders of magnitude is a signature of chaos assisted
2157: tunnelling.
2158: On the average, $\ln|\Delta_{\epsilon_n}|$
2159: appears to decrease more or less linearily with~$\hbare^{-1}$
2160:  except for
2161:  the plateau at $15\le\hbare^{-1}\protect\le30$,  
2162:         }
2163: \end{figure}
2164: 
2165: 
2166: 
2167: \begin{figure}[!ht]
2168: \center
2169: \psfig{file=fluctuations_gamma.ps}
2170: \caption{\label{fig:fluctuations_gamma} Fluctuations of the 
2171: energy splittings~$\Delta\epsilon_n$ between the pair
2172: of symmetric/antisymmetric states  
2173: ($\hbare^{-1}$ is fixed at~$19.309$) as a function of
2174: $\gamma.$  Again,
2175: the large fluctuations
2176:  over several orders of magnitude are a signature of chaos assisted
2177: tunnelling. The global increase with $\gamma$ is due to the growth
2178: of the chaotic sea as $\gamma$ increases, see figure~\ref{fig:3yeux}.
2179:         }
2180: \end{figure}
2181: 
2182: \begin{figure}[!ht]
2183: \center
2184: \psfig{file=LUlaw.ps,width=16cm,angle=-90}
2185: \caption{\label{fig:LUlaw} Statistical distribution of the
2186: energy splittings $\Delta\epsilon$ 
2187: (normalized to the typical splitting) between pairs of
2188: symmetric/antisymmetric states localized inside the~$\mathcal{I}_{\pm}$
2189: resonance
2190: islands ($\gamma=0.18$, $k=0),$ represented on a double logarithmic
2191: scale.
2192: On can clearly distinguish two regimes: constant at small  
2193: $\Delta\epsilon$
2194: followed by a $1/\Delta\epsilon^2$ decrease and finally a rapid cut-off
2195: (not shown in the figure). The solid line is the Cauchy distribution
2196: predicted
2197: by Random Matrix Theory.} 
2198: \end{figure}
2199: 
2200: 
2201: 
2202: \begin{figure}[!ht]
2203: \center
2204: \psfig{file=approxpendule.ps}
2205: \caption{\label{fig:approxpendule} Comparison between the quasi-energies
2206: of Hamiltonian~\eqref{eq:ham_red} (lines) and the energies (circles) 
2207: obtained from  the pendulum
2208: approximation in the stable island~${\mathcal{I}}_0$ 
2209: ($\hbar=.0976,k=0$). The avoided crossings  which  appear
2210: at~$\gamma\sim0.07$ illustrate the  influence of  classical narrow
2211: chaotic seas
2212: on the quantum properties. The Husimi distribution of the second
2213: excited state localised in~${\mathcal{I}}_0$ for~$\gamma=0.03$ is
2214: shown.
2215: }
2216: \end{figure}
2217: 
2218: \begin{figure}[!ht]
2219: \center
2220: \psfig{file=approxpendulebis.ps}
2221: \caption{\label{fig:approxpendulebis} Comparison between the
2222: quasi-energies
2223: of Hamiltonian~\eqref{eq:ham_red} (lines) and the energies (circles) 
2224: obtained from  the pendulum
2225: approximation in the stable islands~${\mathcal{I}}_\pm$  
2226: ($\hbar=.0976,k=0$). The three upper Husimi plots show how a 
2227: quasi-free state doublet becomes progressively localised in the stable
2228: island when
2229: $\gamma$ increases.
2230: The lower 
2231: Husimi plot  corresponds to the ``ground'' state
2232:  of the pendulum approximation
2233: in the stable islands~${\mathcal{I}}_\pm$  for~$\gamma=0.03$.
2234: }
2235: \end{figure}
2236: 
2237: 
2238: \begin{figure}[!ht]
2239: \center  
2240: \psfig{file=suiviadiabatique.ps}
2241: \caption{\label{fig:suiviadiabatique} In order to measure the
2242: tunnelling splitting at~$k=0$ (see figures~\ref{fig:bandes_un}
2243:  and \ref{fig:bandes_deux}), the atoms are prepared into states
2244: localised
2245: inside the~${\mathcal{I}}_+$ resonance island. The average momentum 
2246: distribution (top graph) is peaked
2247: at a value slightly larger than~$+1$ in agreement with
2248: figure~\ref{fig:3yeux}.
2249:  The initial states occupy 1/10 of the first Brillouin zone and are
2250: represented  
2251: by the circles in the lower graph. 
2252: ($\hbare=0.2037,\gamma=0.18)$. 
2253: After an adiabatic sliding of~$\Delta k\simeq0.1$, the states populated
2254: and the 
2255: average momentum distribution is shown in
2256: figure~\ref{fig:suiviadiabatique2}.
2257: }
2258: \end{figure}
2259: 
2260: \begin{figure}[!ht]
2261: \center  
2262: \psfig{file=suiviadiabatique2.ps}
2263: \caption{\label{fig:suiviadiabatique2} The final states obtained
2264: after drift of the Bloch vector by $\Delta k\simeq1/10$ 
2265: (initial state in 
2266: figure~\ref{fig:suiviadiabatique}). If the force
2267: governing the motion~\eqref{eq:kpoint3} is chosen such that the 
2268: sliding is adiabatic through the avoided crossing at~$k=0,$ the 
2269: momentum
2270: of the tunnelling atoms has just reversed its sign.
2271: If the initial states cover a small enough well-centered interval 
2272: of the Brilloin zone, only the $k=0$ avoided crossing
2273: is important. Other avoided crossings can be seen
2274: at~$k\simeq\pm0.09$ 
2275: and~$\epsilon\simeq0.43$. Those are responsible for the momentum 
2276: dispersion
2277: of the atoms and should be avoided as much as possible.
2278: }
2279: \end{figure}
2280: 
2281: \end{document}
2282: