nlin0205054/secIV.tex
1: 
2: 
3: \section{Resonance assisted tunneling in the kicked Harper model}
4: \label{sec:kh}
5: 
6: 
7: After the  general discussion in  the previous sections, we  shall now
8: consider in more  detail a particular system in  the nearly integrable
9: regime, namely the  kicked Harper model.  The purpose  of this section
10: will be to  check, for this particular case, the  accuracy of the final
11: semiclassical expression Eq.~(\ref{globresonsplit}) -- which gives the
12: splitting between  the quasi-energy of  a symmetric Floquet  mode and
13: its antisymmetric counterpart  -- as well as to verify the degree  of validity
14: of the various hypotheses that were made along the way of its derivation. 
15: 
16: 
17: The  classical  dynamics of  the  kicked  Harper  is governed  by  the
18: hamiltonian   Eq.~(\ref{eq:kh_H}),  yielding   the   stroboscopic  map
19: Eq.~(\ref{eq:kh_map}).  In the limit $\tau \to 0$, this dynamics is equivalent to
20: the  one  generated  by   the  time-independent  (integrable)  Harper
21: hamiltonian
22: \begin{equation} 
23:     H_0(p,q) = \cos{p} + \cos{q} \; .
24: \label{eq:harper_ham}
25: \end{equation}
26: In Eq.~(\ref{eq:kh_H}), $\tau$ is thus both  the period of the kick and the
27: perturbation  parameter (i.e.\  $\epsilon \equiv  \tau$).
28: 
29: 
30: 
31: 
32: Quantum mechanically, the map Eq.~(\ref{eq:kh_map}) can be associated
33: with the evolution operator
34: \begin{equation}
35:     \widehat{U} = \exp\left(\frac{i\tau\cos\widehat{p}}{\hbar}\right)
36:     \exp\left(\frac{i\tau\cos\widehat{q}}{\hbar}\right).
37: \end{equation}
38: The $2\pi$ periodicity in $\hat q$ and $\hat p$ makes the quantum treatment of
39: the kicked Harper particularly easy if
40: \begin{equation}
41: \hbar = \frac{2\pi}{N}
42: \end{equation}
43: with integer $N > 0$.
44: For these particular values of $\hbar$, the eigenfunctions $\psi$ of $\widehat{U}$
45: can be written as Bloch functions in both position and momentum -- i.e.,
46: \begin{eqnarray*}
47:   \langle q+2\pi | \psi \rangle 
48:   & = &\exp(i\varphi_q) \langle q |
49:   \psi\rangle \; ,\\
50:   \langle p+2\pi | \psi \rangle & = &\exp(i\varphi_p) \langle p |
51:   \psi\rangle
52: \end{eqnarray*}
53: for some pair of Bloch phases $0 \leq \varphi_q, \varphi_p < 2 \pi$, where $| q \rangle$ and 
54: $| p \rangle$ denote the eigenfunctions of the position and momentum operator,
55: respectively.
56: For each pair $(\varphi_q,\varphi_p)$ of Bloch phases, the corresponding subspace of the
57: Hilbert space  is {\em finite} dimensional and contains $N$ linearly
58: independent wave functions, spanned, e.g., by the basis states 
59: \begin{equation}
60: |n\rangle = \sum_{l=-\infty}^{\infty}\exp(i(l+n/N)\varphi_q) \ |q \equiv ( 2 \pi n + \varphi_p ) / N + 2 \pi l \rangle
61: \end{equation}
62: for $0 \leq n < N$ \cite{Leboeuf90}.
63: The eigenvectors $|\psi_k\rangle$ of $\widehat{U}$ and their eigenphases $\phi_k$ can
64: therefore be computed up to numerical (quadruple) precision, by diagonalizing
65: the $N$$\times$$N$ matrix $\langle n | \widehat{U} | n' \rangle$.
66: %The classical kicked Harper  Hamiltonian is a $2\pi$-periodic function
67: %of $p$  and $q$.  This  implies that the  Hilbert space, and  thus the
68: %eigenphases     $\phi_k$     and     eigenfunctions    $\phi_k$     of
69: %$\widehat{U}(0,\tau)$,    are   divided    in    independent   sectors
70: %characterized by  a pair of  Bloch phases $\varphi_q$  and $\varphi_p$
71: %such  that  for a  wave  function  $ |\psi  \rangle  $  in the  sector
72: %$(\varphi_q,\varphi_p)$
73: %\begin{eqnarray*}
74: %  \langle q+2\pi | \psi \rangle 
75: %  & = &\exp(i\varphi_q) \langle q |
76: %  \psi\rangle \; ,\\
77: %  \langle p+2\pi | \psi \rangle & = &\exp(i\varphi_p) \langle p |
78: %  \psi\rangle \; .
79: %\end{eqnarray*}
80: %Note that for these two conditions to be compatible, it is required that 
81: %\begin{equation}
82: %  \hbar = \frac{2\pi}{N}
83: %\end{equation}
84: %with the $N$  the number of states.  
85: In the following, we  shall consider only the two 
86: pairs $(0,0)$ and  $(0,\pi)$ of Bloch  phases, corresponding  to periodic
87: boundary conditions in momentum, and periodic or anti-periodic
88: boundary conditions in position.  This choice is equivalent
89: to restricting $p$  to the  interval $[-\pi,\pi]$  and $q$  to  the interval
90: $[-\pi,3\pi]$ with periodic boundary  conditions (see Fig.~\ref{fig:khmap}), and
91: to consider the even  and  odd symmetry  classes  with  respect  to the
92: inversion  $q \to -q$.
93: 
94: 
95: 
96: 
97: The  calculation of the  integrable approximation  $\widetilde{H}$ for
98: the  kicked  Harper is  performed  straightforwardly  by applying  the
99: formalism  of Appendix  \ref{app:lie}.   One obtains  for instance  as
100: zeroth      order     coefficient      the      Harper     Hamiltonian
101: Eq.~(\ref{eq:harper_ham}), and  Eq.~(\ref{Hint}) for the approximation
102: of order  three (recall that  $\epsilon \equiv \tau$).   In principle,
103: one may construct  $\widetilde{H}^{(n)}$ up to orders as  high as $n =
104: 20$  fairly easily  with symbolic  programs such  as {\tt  MAPLE}.  As
105: mentioned  in   Sec.~\ref{sec:quasiint},  however,  the   series  Eq.\
106: (\ref{seriesH}) of $\widetilde{H}^{(n)}$ tends to re-diverge beyond an
107: optimal order $n_0$, which, for $\tau = 1.$, is generally found around
108: $n_0 \simeq  6$.  This is illustrated  in Fig.~\ref{fig:asympH_n}: For
109: various  orders  $n$ of  the  integrable  approximation, $40$ randomly
110: distributed initial  phase  space  points  have  been
111: propagated during  a given time by  means of the kicked  Harper map as
112: well as by its integrable approximation $\widetilde{H}^{(n)}$, and the
113: distance in phase space between  the two resulting sets of final points 
114: is plotted
115: as a function of $n$, yielding a minimum at rather moderate values ($n
116: \simeq 6$ in  this particular example). We shall  therefore mainly use
117: $\widetilde{H}^{(n)}$ with $n = 6$ in the following.
118: 
119: 
120: \subsection{Resonances parameters}
121: 
122: Fig.~\ref{fig:khmap} and Fig.~\ref{fig:Htildemap} compare the phase  space
123: portraits of the kicked Harper and of  its integrable (6th order) approximation
124: in the near-integrable regime at $\tau =  1.0$.  In fact, one observes that the
125: only significant difference between the two Poincar{\'e}  sections  is the
126: presence of  the  resonances.  One  may further note the  relative importance
127: of $r$:$1$ resonances with $r = 10$ and $14$ as compared to the $8$:$1$ and
128: the $12$:$1$ resonances
129: (the absence of resonances with odd $r$  is an obvious consequence of the
130: rectangular symmetry  of the kicked  Harper).
131: As a matter of fact, these latter resonances, with $r$ a multiple of $4$,
132: are rather weakly developed at $\tau = 1$ and systematically exhibit $2 r$
133: (instead of $r$) islands in the Poincar{\'e} surface of section.
134: We conjecture that this behavior is a consequence of the initial square
135: symmetry of  the Harper Hamiltonian, which is  still relevant for  small
136: values of $\tau$.
137: As the period in the center of the regular region is already larger than $6$
138: and monotonously increases when moving towards the separatrix, $r$:$s$
139: resonances with $r \leq 6$ do not exist at $\tau = 1$.
140: 
141: To obtain  a quantitative  prediction for the  tunneling rates,  it is
142: necessary   to  characterize  the   resonances  through   the  Fourier
143: coefficients $V^{r:s}_{r.m}$.   This is done  in practice by  a direct
144: application of Eq.~(\ref{couplTPS}),  i.e.\ by Fourier transforming the
145: function $\delta I_{r:s}(\theta) \equiv \delta I_{r:s}(p(I_{r:s},\theta),q(I_{r:s},\theta))$
146: %where  $(p,q)(\theta) = (p,q)(\theta,I_{r:s})$,  and 
147: where $I_{r:s}$  is the
148: action of the resonant  torus ${\Gamma}_{r:s}$.  
149: %For a one-dimensional
150: %system,  
151: On this torus, the  angle  variable  is  given by  $\theta  =  \Omega_{r:s}  t$,  with
152: $\Omega_{r:s} = \partial H / \partial  I (I_{r:s}) = 2\pi s/ (r \tau) $.  
153: %Once
154: %an   origin  $(p_0,q_0)$  has   been  chosen   on
155: %${\Gamma}_{r:s}$ ,  the phase space point  $(p,q)(\theta)$ is obtained
156: %by classical  propagation of $(p_0,q_0)$ during  a time $t  = \theta /
157: %\Omega_{r:s}$.  
158: For a given $\theta$, $\delta I_{r:s}$ is computed
159: through  the following  successive steps:  
160: i)  Choose once  for all  a
161: reference point $(p_0,q_0)$ on the resonant torus ${\Gamma}_{r:s}$
162: of $\widetilde{H}^{(n_0)}$.  
163: ii)  Propagate $(p_0,q_0)$ under $\widetilde{H}^{(n_0)}$ dynamics during the
164: time $t =  (r/s) (\theta /2\pi)  \tau$.  
165: iii) Apply the time reverse of  the Poincar{\'e} map Eq.~(\ref{eq:kh_map}) on the
166: resulting point.   
167: iv) Compute the  difference between the action of this iterated point and 
168: the action $I_{r:s}$ of ${\Gamma}_{r:s}$.  
169: The values obtained
170: in   this   way   for   the   $10$:$1$ resonance  are   plotted   on
171: Fig.~\ref{fig:Vk_prl_tps}, for various orders $n$ of the
172: integrable approximation,  showing that  for $3 \leq n \leq 6$ the coefficients
173: $V^{r:s}_{r.m}$ do not depend  sensitively on $n$.  Also shown 
174: in this  figure are  the values obtained  by the method  introduced in
175: \cite{prlbsu}, which is based on a Fourier analysis of the (pseudo-)separatrix
176: structure that is associated with the resonance. 
177: 
178: 
179: 
180: 
181: Within  our setting for the kicked Harper,  the tunnel splitting is
182: defined as the difference
183: \begin{equation}
184:       \delta \phi_k = |\phi_k(\varphi_q=0) - \phi_k(\varphi_q=\pi)|.
185: \end{equation}
186: As already stated, the exact quantum values of $\delta \phi_k$ can be calculated up to
187: numerical precision.
188: %The  exact  values   of  the  $\delta  \phi_k$  can   be  obtained  by
189: %diagonalization  of the matrix  $\langle n  | \widehat{U}(0,\tau)  | m
190: %\rangle$,  for instance  in  the basis  $|m\rangle$  of the  functions
191: %proportional  to  the  Dirac  distributions  $\delta(q  -  (2\pi  m  +
192: %\varphi_p )/N)$  \cite{Leboeuf90}.  
193: Using  the coefficients   $V^{r:s}_{r.m}$ obtained in the above way, as well
194: as the unperturbed energies  $\tilde{E}_k$, the periods $T_{k}$ and the
195: tunneling actions $\sigma_{k}$ which are straightforwardly calculated from the
196: integrable approximation  $\widetilde{H}^{(n)}$ of the  kicked Harper,
197: these exact splittings can be  compared with the ones derived from our
198: semiclassical  expression   Eq.~(\ref{globresonsplit})  based  on  the
199: resonance-assisted tunneling mechanism.
200: 
201: 
202: 
203: Before performing this comparison, 
204: %it can be however usefull to 
205: let us first verify that  the qualitative description of the tunneling
206: mechanism we  gave in Sec.~\ref{sec:pendulum}  and \ref{sec:mechanism}
207: actually applies  in this particular  example.  To start with,  we can
208: check that  all the  resonance involved in  the tunneling  process are
209: well within the quantum perturbative regime.  Indeed, for the value of
210: the perturbation parameter we  consider, $\tau=1$, the largest Fourier
211: coefficients  for the  resonances coming  into play  are $V^{8:1}_{16}
212: \simeq 9.0  \cdot 10^{-7}$, (as already stated,  the $8$:$1$ resonance
213: exhibits  16  islands),  $V^{10:1}_{10}  \simeq  2.5  \cdot  10^{-4}$,
214: $V^{14:1}_{14} \simeq 9 \cdot 10^{-4}$, while, in the range of $\hbar$
215: we  consider, the  energy difference  between  quasi-degenerate states
216: with respect  to the  resonance is typically  of the order  of $\simeq
217: 10^{-2}$.  Furthermore,  taking into account the actual  values of the
218: $V^{r:s}_{m.r}$ we observe that as $\hbar=2\pi/N$ gets smaller, higher
219: orders  of the  quantum  perturbation theory  become  dominant in  the
220: calculation     of     the     transition    amplitudes     ${\mathcal
221: A}^{r:s}_{k,m}$. This  can be  specifically verified for  the $10$:$1$
222: resonance: For  this resonance,  the $k \to  k+20$ transitions  are of
223: order  one --  i.e.,  are dominated  by  the first-order  perturbative
224: coupling terms -- for $N  \leqsim 38$, but involve perturbation theory
225: of order two for  $N \geqsim 38$.  Similarly, we find that  the $k \to k+30$
226: transitions are of order one for  $N \leqsim 38$, of order two for $38
227: \leqsim N \leqsim  127$ and involve higher terms  beyond ($k \to k+10$
228: transitions are, of course, always of order one).
229: %considers for instance couplings due to the $10$:$1$ resonance, it can
230: %be checked that the transitions $k \to  k+10$ are, of course order one, the
231: %transition $k \to  k+20$ are order one for $N$  smaller than $38$, but
232: %order two  beyond, and  for $k  \to k+30$, order  one dominates  up to
233: %$N=38$, and  order two up  to $N=127$.  
234: Effectively, one finds here the (possibly unusual) situation which will
235: generally be encountered in  the semiclassical  limit -- namely that the 
236: lowest order terms of the perturbative expansion (which converges nevertheless
237: well) are not the dominating ones.
238: 
239: Figures~\ref{fig:split1} and  \ref{fig:split2} show for a varying value of
240: $\hbar$, i.e., a varying total number $N  = 2 \pi / \hbar$ of states, the eigenphase
241: splittings of the eigenmode of $\widehat{U}$ that corresponds  to a 
242: fixed classical torus, with action $I=\pi/4$ in Fig.~\ref{fig:split1} and with
243: action $I=\pi/6$ in Fig.~\ref{fig:split2}.
244: Evidently, these splittings can be calculated only for particular values of
245: $N$, namely for $N = 4 ( 2 k + 1 )$ and $N = 6 ( 2 k + 1 )$ with $k = 0,1,\ldots$
246: in Fig.~\ref{fig:split1} and Fig.~\ref{fig:split2}, respectively, for which
247: this torus is selected by semiclassical quantization and supports the
248: $k$th excited quasi-mode.
249: % but for  a varying value of $\hbar$, which means
250: %a varying  value of the  number of states  $N$.  Fig.~\ref{fig:split1}
251: %correspond to  the classical torus of classical  action $I=\pi/4$, and
252: %figure \ref{fig:split2}  with the classical torus  $I=\pi/6$.  
253: In both cases, the perturbation parameter  $\tau$ equals $1.0$.  
254: The resonance involved are the $8$:$1$,  $10$:$1$, and $14$:$1$.   
255: We observe  that the  agreement  between the
256: quantum  and  semiclassical  results   is  extremely  nice.   For  the
257: moderately small values of $\hbar$ that we consider, it is possible to 
258: try  all the possible  coupling paths  that participate at the tunneling
259: process,  and  in  Figs.~\ref{fig:split1}  and  \ref{fig:split2},  the
260: semiclassical   prediction   is   obtained   by  summing up  all   these
261: contributions.    However,   as  shown   on   the   lower  panel   of
262: Fig.~\ref{fig:split2},   where   the   action   coordinates   of   the
263: intermediate states that participate at  the dominant tunneling  path are
264: displayed,   we  see   here  that,   as  discussed   at  the   end  of
265: section~\ref{sec:mechanism},  this dominant path  is always  such that
266: the  number of steps is as large  as possible,  taking into  account the
267: contraints due to $\hbar$.
268: 
269: 
270: Finally we  show on Fig.~\ref{fig:split-ki} a comparison,  for a fixed
271: value of $\hbar$  and a variable  initial   torus,  between  the  exact  quantum
272: mechanical  splitting  and  the  one calculated  from  the  expression
273: corresponding to integrable tunneling, with no resonance coupling.  We
274: observe on this figure that, although the two curves strongly differ in
275: the interior of the regular region, they match perfectly as
276: one gets close  to the separatrix.  This shows  that the presence
277: of  the separatrix  does not  introduce any  additional effect (e.g.\ from a
278: small chaotic layer) to the tunneling mechanism.
279: 
280: 
281: 
282: 
283: 
284: \subsection{Singularities of the invariant manifold of the integrable
285: approximation} 
286: 
287: 
288: In addition to  the  numerical values  of  the  coefficients $V^{r:s}_{r.m}$,
289: needed to  obtain quantitative prediction  for the tunneling  rates, a
290: qualitative understanding of their  behaviour, and in particular their
291: asymptotic  properties for  large  $m$,  is, as  seen  for instance  in
292: section~\ref{sec:mechanism},  also   required  to  guaranty   that  the
293: tunneling mechanism we propose  is indeed the dominating one.  Since
294: the  $V^{r:s}_{r.m}$ are proportional  to the  Fourier coefficients  of the
295: function  $\delta  I_{r:s}  (\theta)$,  their  asymptotic  behaviour  is
296: related  to  the  singularities of this function,  for  complex  values  of
297: the  angle $\theta$.
298: 
299: 
300: Let us consider,  more generally, for fixed values  of the energy $E$
301: and the order $n$ of  the integrable approximation,  the invariant
302: manifold ${\Gamma}$ of  $\widetilde{H}^{(n)}$, defined by the equation
303: $\widetilde{H}^{(n)}(p,q) = E$ and characterized by the angular frequency
304: $\Omega_0$.
305: Let a function $f(\theta)$ be defined on ${\Gamma}$ as
306: \begin{equation}
307:   f(\theta) = \hat f (p(\theta),q(\theta)) \; ,
308: \end{equation}
309: where $\hat f(p,q)$ is an entire  function of the phase space variables.
310: %Because  $\hat  f$  is   an  entire  function,  
311: As a consequence,
312: the  singularities  of
313: $f(\theta)$  are the ones  of $(p,q)(\theta)$.
314: What we therefore  need to  study  are the  singularities  of the  analytic
315: continuation of  $(p,q)(\theta)$ for complex  angles $\theta$.  Due to the
316: linear relation between $\theta$ and $t$, this analytic continuation
317: is  straightforwardly constructed  by     propagation (under
318: $\widetilde{H}^{(n)}$) of some real initial point $(p_0,q_0)$ on ${\Gamma}$, taken
319: as the origin of the  angle axis, over complex time $t$.
320: %$\widetilde{H}^{(n)}$   and   for    complex   time   of   the   point
321: %$(p_0,q_0)$ on ${\Gamma}$, assumed  real, taken as  the origin of the  angle.  
322: A singularity of $(p,q)(\theta)$  is an angle $\theta^\xi$ such that for the
323: time $t^\xi  = \theta^\xi / \Omega$ the point $g^{\widetilde H}_{t} (p_0,q_0)$ goes  to
324: infinity.  Note that  because of the  existence of these 
325: singularities, $(p,q)(\theta)$  actually depend not only  on the final
326: time $\theta / \Omega_0$, but  also on the homotopy  class of the
327: path joining $t=0$ to $\theta / \Omega_0$ in  complex the time plane.  In other words,
328: $(p,q)(\theta)$ is a priori a multivalued function of $\theta$.
329: 
330: To search for the singularities of $(p,q)(\theta)$, the first step will
331: consist  in  finding  asymptotic  expression describing  the  manifold
332: $\widetilde{H}^{(n)} (p,q) = E$ when the imaginary part of $p$ and/or  $q$ goes
333: to infinity.  For this purpose, we introduce the variables
334: \begin{eqnarray}
335: X & = & \exp(ip) \; , \\
336: Y & = & \exp(iq) \; .
337: \end{eqnarray}
338: In these new variables, the integrable approximation of the kicked Harper
339: Hamiltonian takes the polynomial form
340: \begin{equation}
341:    \label{eq:HXY}
342:    \widetilde{H}^{(n)} = \sum_{i,j=-(n+1)}^{+(n+1)} a_{ij}^{(n)} X^i Y^j \; ,
343: \end{equation}
344: with  known   real  coefficient  $a_{ij}^{(n)}$.    For  $H^{(0)}$  for
345: instance, the non-zero  coefficients are $a_{-10}^{(0)} = a_{10}^{(0)}
346: = a_{0-1}^{(0)} =  a_{01}^{(0)} = 1/2$. 
347: 
348: 
349: 
350:  The manifold  ${\Gamma}$ is  invariant under the symmetries  $s_O: (X
351: \mapsto 1/X,  Y \mapsto 1/Y)$, $s_\Delta: (X  \leftrightarrow Y)$, and
352: $\bar s: (X \mapsto 1/\bar X, Y \mapsto 1/\bar Y)$.  Moreover, one can
353: check  easily that  if $\widetilde{H}^{(n)}  [\tau] (X,Y)  =  E$, then
354: $\widetilde{H}^{(n)}  [-\tau] (1/X,Y)  = E$.   We shall  call $s_\tau$
355: this transformation, although this is not properly speaking a symmetry
356: of $\Gamma$.  The asymptotic  regions  of ${\Gamma}$ -- i.e., the neighborhood of points
357: at infinity  on ${\Gamma}$ -- can be obtained by application of one of the above
358: transformation from one region  such that $\Im  [p] \to +\infty$,  (i.e.~ $X \to 0$  )
359: and $\Im [q]$ is either bounded  or goes to $+\infty$ (i.e.~ $Y$ bounded).
360: For such regions, one can assume an asymptotic expression of the form
361: \begin{equation}
362:     \label{eq:asymptot}
363:     Y^\xi(X) = \gamma^\xi_{0} + \gamma^\xi_{1} X + 
364:     \gamma^\xi_{2}   X^{2} + \cdots \; .
365: \end{equation}
366: where $\xi$ label the asymptotic region.
367: Introducing Eq.~(\ref{eq:asymptot}) in the expression
368: Eq.~(\ref{eq:HXY}) of the Hamiltonian to solve the equation
369: $\widetilde{H}^{(n)}(X,Y) = E$ yields a series of polynomial equations
370: for the coefficients $\gamma^\xi_l$, which can be solved order by order
371: to determine successively $\gamma^\xi_{0}$,  $\gamma^\xi_{1}$,
372: $\gamma^\xi_{2}$, etc..  Again for the zeroth order Hamiltonian
373: $H^{(0)}$, the set of equations obtained in this way are
374: \begin{eqnarray*}
375:   \gamma_{0} & = & 0 \\
376:   \frac{1}{2} \left( \gamma_1 + 1 + \gamma_{0} \right)  
377:   & = & E \gamma_{0} \\
378:   \gamma_2 / 2 + \gamma_1 \gamma_{0} + \gamma_{0}/2 
379:   & = &  E \gamma_1 \\
380:   \cdots && \cdots \; ,
381: \end{eqnarray*}
382: yielding $\gamma_{0} = 0$, $\gamma_{1} = -1$, $\gamma_{2} = -2E$,
383: $\cdots$.  In other words, for small $X$, the manifold defined by the
384: implicit expression $\widetilde{H}^{(0)}(X,Y) = E$ admits the explicit
385: asymptotic expression 
386: \begin{equation} \label{eq:asym_H0}
387: Y(X) = -X - 2E X^2 + \cdots \; .
388: \end{equation} 
389: 
390: 
391: Using the above equation with $X$  small enough allows to find a point
392: with    a    large    imaginary    part    for    $p$,    such    that
393: $\widetilde{H}^{(0)}(p,q)$ is  very close to  $E$.  This point  can be
394: brought  back to  the  energy $E$  by  following the  gradient of  the
395: Hamiltonian,    giving     a    point    $(p^\xi,q^\xi)$     on    the
396: $\widetilde{H}^{(0)}(p,q) =  E$ manifold and in  the asymptotic region
397: of  large  $\Im  [p]$.   From  this  point,  we  integrate  Hamilton's
398: equations  of motion  choosing the  path in  the complex  time  in two
399: different ways: i)  First we take a purely  imaginary direction, until
400: $t=i  t_I$  such  that   the  trajectory  crosses  the  real  manifold
401: ${\Gamma}_{\mathbb R}  = {\Gamma} \cap {\mathbb  R}^2$.  The imaginary
402: part of the angle coordinate of $(p^\xi,q^\xi)$ it then given by $-t_I
403: / \Omega$.   ii) Then we  start again from $(p^\xi,q^\xi)$  and choose
404: the  complex phase  of each  time step  $dt$ in  such a  way  that the
405: imaginary part of $p$ remains constant.  The time $t$ describes then a
406: small loop  in the complex  time plane that contains  the singularity.
407: This  gives  the order  of  magnitude  of  the time  distance  between
408: $(p^\xi,q^\xi)$ and  the singularity,  which is in  practice extremely
409: small as soon as $\Im(p)$ is taken reasonably large.  For $n=0$, there
410: is only one independent (i.e.   up to symmetries) singularity, and the
411: imaginary part of its time  coordinate is just half of $t^\sigma$, the
412: imaginary  time  required  to   go  from  ${\Gamma}_{\mathbb  R}  \cap
413: [-\pi,\pi] \times [-\pi,\pi]$  to ${\Gamma}_{\mathbb R} \cap [\pi,3\pi
414: ] \times [-\pi ,\pi]$.
415: 
416: 
417: 
418: 
419: Such  a procedure  can be  reproduced for  various orders  $n$  of the
420: integrable  Hamiltonian, and we  have performed  it explicitly  up to
421: $n=3$.   Although the method  we apply  is basically  the same,  a few
422: important differences may be noticed
423: \begin{itemize}
424: \item[i)]  The  number  of  singularities (i.e.   more  precisely,  of
425: asymptotic regions  of the manifold)  increases with the order  of the
426: Hamiltonian.  Counting only the  number of  independent singularities,
427: that  is the ones  that cannot  be deduced  one from  each other  by a
428: symmetry, there is only one  for $n=0$, but $(2n+2)$ for $n = 1, 2,3$.
429: \item[ii)] If one starts form  a point $(p^\xi,q^\xi)$ in an asymptotic
430: region  such as  Eq.~(\ref{eq:asymptot}) and propagates along a time path that
431: describes  a small closed  loop of infinitesimal  radius around  the
432: singularity  in time plane, one  can show that the  real part of the
433: resulting momentum is not Re$[p^\xi]$, but Re$[p^\xi] + 2\pi/\ell^\xi$, 
434: where $\ell^\xi$
435: is  an  integer  which depends  on  the  order  $n$ of  the  integrable
436: approximation   and    on   the   singularity    under   consideration
437: %\cite{olivier:thesis} 
438: ($\ell^\xi$  is equal  to one for  $H^{(0)}$ and
439: $H^{(1)}$,  to two  for $H^{(3)}$  and  four of  the singularities  of
440: $H^{(2)}$, but to three for the two remaining ones). If one identifies
441: $p$ and $p+  2\pi$, this means that for  $n \neq 0,1$, $(p,q)(\theta)$
442: are not meromorphic  functions.
443: Instead, the singularities  are of logarithmic
444: type.   More precisely,  there are  $\ell^\xi$ distinct  sheets  of the
445: manifold $(p,q)(\theta)$ around each singularity.
446: \item[iii)] As a consequence, when one tries to reach the complex torus
447: from the  neighborhood of  a singularity, one  should specify  on what
448: sheet  one  places  oneself.   Moreover,  this implies  that  not  all
449: singularities are ``visible'' from the real torus: assuming the best way
450: to compute  the Fourier integral Eq.~(\ref{couplTPS}) is  to shift the
451: integration contour  in the  imaginary direction, the  only singularities
452: that will be encountered in this way are the ones that can be reached
453: by  purely imaginary time  propagation from  the real  manifold.  For
454: $n=3$, only four out of the eight singularities are ``visible'' from the
455: real torus.
456: \item[iv)] Starting from the neighborhood of a ``visible'' singularity
457: and following the Hamiltonian flow,  one may, depending on whether time
458: runs in  the positive or  negative imaginary direction,  and depending
459: also on the chosen sheet of the manifold, cross the real manifold
460: ${\Gamma}_R$ in different cell $[(2l- 1) \pi, (2l+1)\pi] \times  [(2l'- 1)\pi,
461: (2l'+1)\pi]$. Depending on the final cell, the time can be $\pm i
462: t^\xi$ or $\pm i (t^\sigma - t^\xi)$
463: \end{itemize}
464: 
465: 
466: 
467: On  Figs.~\ref{fig:txi_vs_tau:2} and \ref{fig:txi_vs_tau:3},  we plot,
468: for the resonant torus $10$:$1$  and as a function of the perturbation
469: parameter $\tau$,  the imaginary  part of the  time coordinate  of the
470: ``visible'' singularities of $\widetilde{H}^{(n)}$ for $n = 2$ and $3$
471: respectively.  What we  are waiting for is that  the $k$ dependence of
472: the $V^{r:s}_{k}$ (for $k=rm$ as well as $k \neq rm$) is given by an
473: expression like Eq.~(\ref{varV}), with $t_{r:s}$ the imaginary part of
474: the time coordinate of the  singularity closest to the real torus.  On
475: Fig~\ref{fig:txi_vs_tau:2}  and \ref{fig:txi_vs_tau:3} are  also shown
476: the  values $\theta_\xi$  obtained  by fitting  the numerically  obtained
477: $V^{r:s}_{k}$  with  the  expression  Eq.~(\ref{varV}).  We observe that
478: for $n=2,3$  the variation of  the fitted values follows  nicely the
479: predicted ones.   For higher $n$,  and up to $n=6$,  the $V^{r:s}_{k}$
480: are insensitive  to the variation  of the order of  the approximation,
481: and therefore the fitted values  remains on the curve corresponding to
482: the $n=3$ closest singularity.
483: 
484: 
485: 
486: %\subsection{Discussion}
487: 
488: The      data     shown     in      Fig~\ref{fig:txi_vs_tau:2}     and
489: \ref{fig:txi_vs_tau:3}  give  a  pretty  convincing  picture,  which
490: justifies  to   use  confidently  the   expression  Eq.~(\ref{varV})  to
491: describe the behavior of the $V^{r:s}_{r.m}$ coefficients.  Although we
492: believe this to be true from a practical point  of view, one should, however,
493: resist the temptation to oversimplify this issue.   Indeed, it is, to start
494: with, a  priori  not  obvious  to  justify  on a  rigorous  basis  the  form
495: $(mr)^{\gamma} V^{r:s}_{\xi_0}$ we have written for the prefactor, and
496: this can only be taken as a sensible parameterization.  Moreover, even
497: if we did  not extend the analysis of the  location of all singularities
498: for  $n$  greater than  three  because  the  approach described  above
499: becomes  somewhat  cumbersome, it  is  still  possible  to locate  the
500: closest singularity  by a brute  force search in the  complex $\theta$
501: plane.  Doing this for $n=4,5,6$ for the the torus $10$:$1$ at $\tau =
502: 1$  shows  that  this  closest  singularity  slightly  drifts  as  $n$
503: increase, and  that its  imaginary part goes  from $0.8$ for  $n=3$ to
504: $0.6$  for $n=6$, 
505: %{\sf  [caution, one  should put  the numbers  here in
506: %concordance with what  is shown on the figure (ie  with or without the
507: %$\Omega$ factor)]}, 
508: in spite of the $V^{r:s}_{r.m}$ being not affected
509: by this change.   This drift, although moderate, is  still larger than
510: the numerical precision of our fit.
511: In accordance, if we take $\hat
512: f(p,q)  \equiv \cos(p)$  as  done of  Fig.~\ref{fig:fk},  we see  that,
513: contrary to  $\delta I_{r:s}$,  the Fourier coefficient  of $f(\theta)$
514: changes with  $n$ even when this  latter is greater than  three, and in
515: particular  follows  the  asymptotic  slope  $0.6$  for  $n=6$.   This
516: indicates  that  although  the basic picture  we  gave  to  interpret  the
517: asymptotic  behavior  of the  $V^{r:s}_{r.m}$  is certainly 
518: correct,  the complete  description  is presumably  more complicated  and
519: might involve, for instance, the  link between the kicked Harper map 
520: ${\mathcal T}$ and the  integrable Hamiltonians  $\widetilde{H}_{(n)}$, as
521: well  as a more  careful  analysis  of  the  different ranges  in  the
522: asymptotic behavior of the $V^{r:s}_{r.m}$.
523: