1: \documentclass[aps,prl,twocolumn,superscriptaddress,showpacs]{revtex4}
2: \usepackage{psfig}
3: \usepackage{epsfig}
4: \usepackage{color}
5: \newcommand{\beq}{\begin{equation}}
6: \newcommand{\eeq}{\end{equation}}
7: \newcommand{\beqa}{\begin{eqnarray}}
8: \newcommand{\eeqa}{\end{eqnarray}}
9: \newcommand{\ba}{\begin{array}}
10: \newcommand{\ea}{\end{array}}
11:
12: \begin{document}
13:
14: \widetext
15: \title{$1/f^{\alpha}$ noise in spectral fluctuations of quantum systems}
16:
17: \author{J.M.G. G\'omez}
18: \author{A. Rela\~no}
19: \author{J. Retamosa}
20: \affiliation{Departamento de F\'{\i}sica At\'omica, Molecular y Nuclear,
21: Universidad Complutense de Madrid, E-28040 Madrid, Spain}
22:
23: \author{E. Faleiro}
24: \affiliation{Departamento de F\'{\i}sica Aplicada, EUIT Industrial,
25: Universidad Polit\'ecnica de Madrid,
26: E-28012 Madrid, Spain}
27:
28: \author{L. Salasnich}
29: \affiliation{INFM, Unit\`a di Milano Universit\`a, Dipartimento di
30: Fisica, Universit\`a di Milano, Via Celoria 16, I-20133 Milano, Italy}
31: \affiliation{CAMTP, University of Maribor, Krekova 2,
32: SI-2000 Maribor, Slovenia}
33:
34: \author{M. Vrani\v car}
35: \author{M. Robnik}
36: \affiliation{CAMTP, University of Maribor,
37: Krekova 2, SI-2000 Maribor, Slovenia}
38:
39: \begin{abstract}
40: The power law $1/f^{\alpha}$ in the power spectrum characterizes the
41: fluctuating observables of many complex natural systems. Considering
42: the energy levels of a quantum system as a discrete time series where
43: the energy plays the role of time, the level fluctuations can be
44: characterized by the power spectrum. Using a family of quantum
45: billiards, we analyze the order to chaos transition in terms of this
46: power spectrum. A power law $1/f^{\alpha}$ is found at all the
47: transition stages, and it is shown that the exponent $\alpha$ is
48: related to the chaotic component of the classical phase space of the
49: quantum system.
50: \end{abstract}
51:
52: \pacs{05.45.Mt, 05.40.a, 05.45.Pq, 05.45.Tp}
53:
54: \maketitle
55:
56: Of the different features which characterize complex physical systems,
57: perhaps the most ubiquitous, interesting and puzzling is the presence
58: of $1/f^{\alpha}$ noise \cite{Mandelbrot:99} in fluctuating physical
59: variables, i.e. the Fourier power spectrum $S(f)$ behaves as
60: $1/f^{\alpha}$ in terms of the frequency $f$. This kind of noise has
61: been detected in condensed matter systems, traffic engineering, DNA
62: sequence, quasar emissions, river discharge, human behavior,
63: heartbeat and dynamic images, among many others. Despite this
64: ubiquity, there is no universal explanation about this phenomenon. It
65: does not arise as a consequence of particular physical interactions,
66: but it is a generic manifestation of complex systems.
67:
68: Recently, it was conjectured that {\it the energy spectra of chaotic
69: quantum systems are characterized by 1/f noise} \cite{Relano:02} . The
70: original idea was that the sequence of discrete energy levels in a
71: quantum system can be considered as a discrete time series, where the
72: energy plays the role of time. In that case, the energy level
73: fluctuations can be studied using traditional methods of time series
74: analysis, like the study of the power spectrum. When the idea was
75: applied to typical chaotic quantum systems, the power spectrum showed
76: a very accurate $1/f$ behavior \cite{Relano:02}. Hence chaotic quantum
77: systems can be added to the long list of complex natural systems which
78: exhibit $1/f$ noise. However, this new point of view also raises new
79: questions. Is this a consequence of the universal behavior of
80: fluctuations in chaotic quantum systems? What happens in quantum
81: systems which are neither fully chaotic nor fully regular? In this
82: paper we try to find some answers using a quantum billiard to study
83: the power spectrum in the order--to--chaos transition. As shown below,
84: the ubiquitous $1/f^{\alpha}$ noise appears at all the transition
85: stages, with the exponent smoothly decreasing from ${\alpha}=2$ in a
86: regular system to ${\alpha}=1$ in a chaotic system. This is quite a
87: remarkable result indeed, since it contradicts the predictions of the
88: strict semiclassical limit \cite{Robnik:03}.
89:
90: The concept of quantum chaos, or wave chaos in more general terms
91: \cite{Stockmann:99}, has no unique precise definition as yet, but
92: definitely can be described as quantum or wave like signatures of
93: classical chaos. It is well known that there is a relationship between
94: the energy level fluctuation properties of a quantum system and the
95: dynamics of its classical limit. Classically integrable systems give
96: rise to uncorrelated adjacent energy levels, which are well described
97: by Poisson statistics \cite{Berry:77}. In contrast, spectral
98: fluctuations of a quantum system whose classical limit is fully
99: chaotic (ergodic) show a strong repulsion between energy levels and
100: follow the predictions of random matrix theory (RMT) \cite{Bohigas:84,
101: Casati:80}. In practice, quantum systems without classical limit are
102: assumed to be chaotic when their fluctuations coincide with RMT
103: predictions.
104:
105: The essential feature of chaotic energy spectra is the existence of
106: level repulsion and correlations (leading to strong spectral
107: rigidity), i. e. the spacing of two adjacent levels is unlikely to
108: deviate much from the mean spacing. This property is similar to the
109: {\it antipersistence} characteristic of some time series
110: \cite{Mandelbrot:99}. Antipersistence, with different intensity
111: degrees, appears in time series with $1/f^{\alpha}$ noise, with
112: $1<\alpha<2$. Could the analogue level repulsion feature be also
113: associated to $1/f^{\alpha}$ noise?
114:
115: To study the spectral fluctuations of quantum systems we follow the
116: method introduced in \cite{Relano:02}. We use the statistic
117: $\delta_n$ defined by
118:
119: \beq
120: \delta_n =
121: \sum_{i=1}^n\left(s_i-\left<s\right>\right)=\epsilon_{n+1}-\epsilon_1-n,
122: \eeq
123: %
124: where $\epsilon_i$ are the unfolded energy levels,
125: $s_i=\epsilon_{i+1}-\epsilon_i$, and $<s>=1$ is the average value of
126: $s_i$. Thus $\delta_n$ represents the fluctuation of the $n$th excited
127: state. Formally $\delta_n$ is similar to a time series where the
128: level order index $n$ plays the role of a discrete time. Therefore the
129: statistical behavior of level fluctuations can be investigated
130: studying the power spectrum $S(k)$ of the signal, given by
131:
132: \beq
133: S(k)=\left| {1\over \sqrt{M}} \sum_{n=1}^M \delta_n \exp{\left(
134: {-2\pi i k n\over M} \right)} \right|^2 \; ,
135: \eeq
136: where $M$ is the size of the series and $f=2{\pi} k/M$ plays the role
137: of a frequency.
138:
139: To investigate the behavior of $S(k)$ in the mixed regime between
140: integrability and chaos, we analyze it in the Robnik billiard
141: \cite{Robnik:83}. Quantum billiards are considered as a paradigm in
142: quantum chaos. They have a discrete spectrum with an infinite number
143: of eigenvalues, and therefore it is possible to reach high statistical
144: precision by computing a large number of them. Furthermore, they can
145: also be studied experimentally \cite{Stockmann:99,Veble:00}.
146:
147:
148: The boundary of the Robnik billiard is defined as the set of points
149: $w$ in the complex plane ${\bf C}$ which satisfy the equation
150: $w=z+\lambda z^2$, where $|z|=1$ and $\lambda$ is the deformation
151: parameter. It has been shown \cite{Robnik:83} that this billiard
152: exhibits a smooth transition from the integrable case ($\lambda =0$)
153: to an almost chaotic case ($1/4\leq \lambda \leq 1/2$). In order to
154: obtain a smooth analytic boundary, $\lambda$ must lie in the interval
155: $[0,1/2)$. The Robnik billiard is one of the best systems to
156: investigate the order--to--chaos transition \cite{Robnik:83,
157: Robnik:84, Prosen:93}. Compared to other quantum billiards, it has
158: the advantage that there are no bouncing ball orbits. For small values
159: of $\lambda$ the billiard is a typical KAM system, whereas for larger
160: values of $\lambda$ only one chaotic region dominates the phase space
161: with only few stability islands covered with invariant tori. The total
162: area in the bounce map (Poincar\'e surface of section) of these
163: invariant tori decreases monotonically with $\lambda$ and becomes
164: negligible when the shape of the billiard becomes non-convex, for
165: $\lambda > 1/4$. For $\lambda =1/2$ it has been shown rigorously by
166: Markarian \cite{Markarian:93} that the billiard is ergodic.
167:
168:
169: \begin{figure}
170: \centerline{\psfig{file=mmm-f1.eps,height=3.8in,angle=-90}}
171: %\vskip 0.5cm
172: \caption{Plot of the statistic $\delta_n$ for a set of 256 consecutive
173: energy levels of odd parity in the Robnik billiard, for several values
174: of the deformation parameter $\lambda$.}
175: \label{fig:delta_n}
176: \end{figure}
177:
178: The quantum energy levels $E_n$ of the Robnik billiard are numerically
179: calculated by solving the stationary Schr\"odinger equation of a free
180: particle whose wave function $\psi(w)$ is zero at the boundary of the
181: billiard. The billiard has reflection symmetry with respect to the
182: real axis, so there are two types of states: those with even parity
183: $\psi(w)=\psi(w^*)$ and those with odd parity $\psi(w)=-\psi(w^*)$;
184: odd and even parity states must be treated separately
185: \cite{Robnik:84, Prosen:93}. For each symmetry, our calculation
186: uses approximately 80,000 basis states, giving good eigenvalues for
187: about 65,000 levels for $\lambda=0$ and about 30,000 levels for
188: $\lambda=0.5$.
189:
190: Fig. \ref{fig:delta_n} shows the energy level fluctuations of the
191: Robnik billiard given by $\delta_n$. It illustrates the effect of
192: level repulsion in the order--to--chaos transition, and its
193: relationship with the antipersistence of $\delta_n$ considered as a
194: time series. For the regular system ($\lambda=0$), the levels are
195: uncorrelated and therefore $\delta_n$ is neither persistent nor
196: antipersistent. As $\lambda$ increases the system becomes more
197: chaotic, and $\delta_n$ looks like a typical antipersistent series in
198: the almost chaotic region for $\lambda>1/4$.
199:
200: \begin{figure}[h]
201: \centerline{\psfig{file=mmm-f2.eps,height=3.8in,angle=-90}}
202: %\vskip 0.5cm
203: \caption{Average power spectrum $\langle S(k) \rangle$ of the
204: statistic $\delta_n$ for the odd parity energy levels corresponding to
205: the shapes of the Robnik billiard inserted in the figures. The four
206: values of the deformation parameter $\lambda$ are the same as in
207: Fig. \ref{fig:delta_n}. The solid (red) line is the best fit to the
208: power law $1/k^{\alpha}$.}
209: \label{fig:s_k}
210: \end{figure}
211:
212:
213: Fig. \ref{fig:s_k} shows the power spectrum of $\delta_n$ obtained
214: from the energy levels $E_n$ of the Robnik billiard for several values
215: of $\lambda$. The contour of the billiard is shown as an inset in each
216: panel. We calculate an ensemble average of $S(k)$ in order to reduce
217: statistical fluctuations and clarify its main trend. The average
218: $\langle S(k)\rangle$ is calculated with $25$ sets of $256$
219: consecutive high energy levels of odd parity. It is clearly seen that
220: for each value of $\lambda$ it follows the scaling law
221: \beq
222: \langle S(k) \rangle \sim {1\over k^{\alpha}} \; ,
223: \eeq
224: where $\alpha$ depends on $\lambda$. In fact the fit of $\langle
225: S(k)\rangle $ to the power law $1/k^{\alpha}$ is excellent. In all
226: cases the error in the linear regression is less than $3\%$. For
227: $\lambda =0$ (integrable case) the exponent is $\alpha = 1.98$, as
228: expected for uncorrelated energy levels. As $\lambda$ increases the
229: exponent $\alpha$ decreases and becomes $\alpha \simeq 1 $ for
230: $\lambda \simeq 1/2$. Thus, $\alpha$ may serve as a measure of the
231: chaoticity of the system, since it changes from $\alpha=2$ for regular
232: systems to $\alpha=1$ for chaotic ones.
233:
234: It is worth to compare this result with the behavior of more
235: conventional statistics, like the nearest neighbor spacing
236: distribution $P(s)$, which measures short range correlations, and the
237: Dyson $\Delta_3(L)$ statistic appropriate for correlations of length
238: $L$ \cite{Mehta:91}. Fig. \ref{fig:nnsd} displays $P(s)$ for several
239: values of $\lambda$. At $\lambda=0$ the histogram follows the
240: predicted curve for regular systems (Poisson limit). For
241: $\lambda=0.15$, $P(s)$ deviates from Poisson toward the RMT limit. As
242: we shall see below, this behavior of $P(s)$ reflects that the
243: underlying classical dynamics is neither regular nor ergodic
244: (chaotic). Finally, for $\lambda=0.25$ and $\lambda=0.4$ the system
245: exhibits short range correlations characteristic of chaotic systems
246: (RMT limit). Fig. \ref{fig:delta3} shows the spectral average
247: $\langle\Delta_3(L)\rangle$ for energy intervals of length $L$ ranging
248: from $L=2$ to $L=50$, for several values of $\lambda$. The spectral
249: average is calculated using $25$ sets of $L$ consecutive high energy
250: levels to avoid, as far as possible, the influence of short periodic
251: orbits. The evolution of this statistic with $\lambda$ is analogous to
252: that of $P(s)$. When $\lambda=0$, $\left<\Delta(L)\right>$ falls near
253: the Poisson prediction for regular systems, and for $\lambda=0.25$ and
254: $0.4$ it is almost indistinguishable from the RMT prediction for
255: chaotic systems. Therefore, $P(s)$ and $\langle\Delta_3(L)\rangle$
256: have a smooth behavior in terms of $\lambda$. Nevertheless, they move
257: faster than $\delta_n$ toward the RMT limit as $\lambda$
258: increases. For instance, both $P(s)$ and $\langle\Delta_3(L)\rangle$
259: coincide with RMT predictions for $\lambda=0.25$, while $\delta_n$
260: still points to an intermediate regime between regularity and chaos.
261:
262: \begin{figure}[h]
263: \centerline{\psfig{file=mmm-f3.ps,height=3.6in,angle=-90}}
264: %\vskip 0.5cm
265: \caption{Nearest neighbor spacing distributions for the spectra of
266: four different shapes of the Robnik billiard. The solid line is the
267: Poisson distribution and the dotted line corresponds to the Wigner
268: distribution predicted by RMT.}
269: \label{fig:nnsd}
270: \end{figure}
271:
272: \begin{figure}[h]
273: \centerline{\psfig{file=mmm-f4.ps,height=2.8in,angle=-90}}
274: %\vskip 0.5cm
275: \caption{Spectral average $\langle\Delta_3(L)\rangle$ values
276: calculated numerically for the Robnik billiard using the same
277: deformation parameters as in previous figures. The results of the
278: Poisson statistics (solid line) and RMT (dotted line) are also
279: displayed.}
280: \label{fig:delta3}
281: \end{figure}
282:
283: Let us now compare the evolution of the parameter $\alpha$ as a
284: function of $\lambda$ with the fraction $\rho_1^{cl}$ of regular
285: classical trajectories in the phase space, and with the Brody
286: parameter $\omega$ \cite{Brody:81}. This is an {\em ad hoc} parameter,
287: without any known physical meaning, which quantifies to some extent the
288: chaoticity of the system.
289: It nevertheless captures well the important feature of fractional
290: power law level repulsion, that is the behavior of $P(s)$ at small $s$
291: \cite{Robnik:84, Prosen:93}. For $\omega=0$ we get the Poisson
292: distribution of regular systems, and at the other extreme,
293: $\omega=1$, we obtain the Wigner distribution predicted by RMT for
294: chaotic systems. Fig. \ref{fig:alpha_rho_omega} shows the behavior of
295: $\alpha$, $\omega$ and $\rho_1^{cl}$. There is a clear correlation
296: among these three variables, although the transition is smoother for
297: $\omega$ and especially for $\alpha$ than for $\rho_1^{cl}$. However,
298: while the fraction of regular classical trajectories is almost zero
299: near $\lambda=0.25$, the power spectrum, and to a lesser extent the
300: $P(s)$ distribution, indicate an intermediate situation between
301: regular and chaotic motion. This clearly shows that $\omega$ and
302: $\alpha$ are not only functions of $\rho_1^{cl}$, but depend on finer
303: details of the underlying classical mechanics as well.
304:
305: \begin{figure}[h]
306: \centerline{\psfig{file=mmm-f5.eps,height=2.6in}}
307: %\vskip 0.5cm
308: \caption{Behavior of the power spectrum exponent $\alpha$,
309: the Brody parameter $\omega$, and the
310: fraction $\rho_1^{cl}$ of regular classical trajectories
311: of the Robnik billiard as functions of the deformation
312: parameter $\lambda$.}
313: \label{fig:alpha_rho_omega}
314: \end{figure}
315:
316: It is well known that in the strict semiclassical limit the quantum
317: eigenstates of a quantum system with generic (mixed) classical dynamics
318: can be classified as regular and irregular following the
319: original proposition of Percival \cite{Percival:73}.
320: This has been further developed and raised
321: to a Principle of Uniform Semiclassical Condensation
322: (PUSC) of Wigner functions of the eigenstates \cite{Berry:77b}.
323: From this it follows that in the strict semiclassical limit
324: the regular and irregular level sequences are statistically
325: independent, but for themselves have Poisson or RMT level statistics,
326: respectively. This theory has been excellently confirmed in hard numerical
327: calculations \cite{Prosen:94}.
328: Nevertheless, if the system is not deep enough
329: in the semiclassical regime one can see substantial deviations from
330: such a behavior, manifested in the fractional power law level
331: repulsion \cite{Robnik:84, Prosen:93}
332: in $P(s)$ at small $s$. A similar recent
333: analysis \cite{Robnik:03} has demonstrated that in the strict
334: semiclassical limit we should not expect a power law
335: behavior for the power
336: spectrum but something more complicated. Therefore it is
337: quite an unexpected result of the present paper that
338: the power spectrum $S(k)$ is a power law at all $k$ in
339: mixed systems at low energies. Indeed, when going sufficiently
340: deep into the semiclassical regime the theory of reference
341: \cite{Robnik:03} should be expected and confirmed.
342:
343: In conclusion, the analogy between quantum energy spectra and time
344: series opens a new and fruitful perspective on the universal
345: properties of quantum level fluctuations. The $\delta_n$ function
346: gives the level fluctuations, and its power spectrum $S(k)$ is an
347: intrinsic characteristic of the quantum system. The important point is
348: that it exhibits a power law behavior, similar to the well known
349: $1/f^{\alpha}$ noise found in many complex systems.
350:
351: In the order to chaos transition, the chaoticity of a quantum system
352: is usually qualitatively assessed by how close to Poisson or RMT its
353: fluctuation properties are. In the present power spectrum approach,
354: the exponent changes smoothly from $\alpha=2$ for a regular system to
355: $\alpha=1$ for a chaotic system. Contrary to the Dyson $\Delta_3(L)$
356: statistic, that must be plotted for different values of $L$, the
357: exponent $\alpha$ quantifies the chaoticity of the system in a single
358: parameter.
359: Moreover, $\alpha$ has a physical meaning. It is a natural measure of the
360: fluctuation properties of a quantum system through the power spectrum,
361: and provides an intrinsic quantitative measure of the regular and chaotic
362: dynamical features.
363:
364: The origin of the universal power law behavior $S(k) \sim
365: 1/k^{\alpha}$ is now understood in the integrable case ($\alpha =2$)
366: and in the fully chaotic case ($\alpha =1$) on the basis of RMT
367: \cite{Faleiro:04} and semiclassical periodic orbit theory
368: \cite{Robnik:03}. The origin of the $1/f^{\alpha}$ power law in the
369: mixed regime still remains as an important open problem.
370:
371: This work is supported in part by Spanish Government grants
372: BFM2003-04147-C02 and FTN2003-08337-C04-04. This work is also
373: supported by the Ministry of Education, Science and Sports of the
374: Republic of Slovenia.
375:
376: \begin{thebibliography}{99}
377:
378: \bibitem{Mandelbrot:99} B.B. Mandelbrot, {\it Multifractals and $1/f$ Noise}
379: (Springer, New York, 1999).
380:
381: \bibitem{Relano:02} A. Rela\~no, J.M.G. G\'omez, R.A. Molina, J. Retamosa, and
382: E. Faleiro, Phys. Rev. Lett. {\bf 89}, 244102 (2002).
383:
384: \bibitem{Robnik:03} M. Robnik, preprint CAMTP/December 2003, to be submitted.
385:
386: \bibitem{Stockmann:99} H.-J. St\"ockmann, {\it Quantum Chaos: an Introduction}
387: (Cambridge University Press, Cambridge, 1999).
388:
389: \bibitem{Berry:77} M.V. Berry and M. Tabor,
390: Proc. R. Soc. London A {\bf 356}, 375 (1977).
391:
392: \bibitem{Bohigas:84} O. Bohigas, M.J. Giannoni, and C. Schmit,
393: Phys. Rev. Lett {\bf 52}, 1 (1984).
394:
395: \bibitem{Casati:80} G. Casati, F. Valz-Gris, and I. Guarneri,
396: Lett. Nuovo Cimento {\bf 28}, 279 (1980).
397:
398: \bibitem{Robnik:83} M. Robnik, J. Phys. A: Math. Gen. {\bf 31}, 3971 (1983).
399:
400: \bibitem{Veble:00} G. Veble, U. Kuhl, M. Robnik, H.-J. St\"ockmann,
401: J. Liu, and M. Barth, Prog. Theor. Phys. Suppl. (Kyoto)
402: {\bf 139}, 283 (2000).
403:
404: \bibitem{Robnik:84} M. Robnik, J. Phys. A: Math. Gen. {\bf 17}, 1049 (1984);
405: T. Prosen and M. Robnik, J. Phys. A: Math. Gen. {\bf 27}, 8059 (1994);
406: M. Robnik and T. Prosen, J. Phys. A: Math. Gen. {\bf 30}, 8787 (1997).
407:
408: \bibitem{Prosen:93} T. Prosen and M. Robnik, J. Phys. A:
409: Math. Gen. {\bf 26}, 2371 (1993).
410:
411: \bibitem{Markarian:93} R. Markarian, Nonlinearity {\bf 6}, 819 (1993).
412:
413: \bibitem{Mehta:91}
414: M. L. Mehta, {\it Random Matrices}, (Academic Press, 1991)
415:
416: \bibitem{Brody:81} T. A. Brody, J. Flores, J. B. French, P. A.
417: Mello, A. Pandey, S. S. M. Wong, Rev. Mod. Phys. {\bf 53}, 385, (1981).
418:
419: \bibitem{Faleiro:04} E. Faleiro, J.M.G. G\'omez, R.A. Molina,
420: L. Mu\~noz, A. Rela\~no, and J. Retamosa, e-preprint nlin.CD/0402028.
421:
422: \bibitem{Percival:73} I.C. Percival, J. Phys. B: At.Mol.Phys.
423: {\bf 6}, L229 (1973).
424:
425: \bibitem{Berry:77b} M.V. Berry, Phil. Trans. Roy.Soc.A {\bf 287}, 237 (1977);
426: M.V. Berry and M. Robnik, J. Phys. A: Math.Gen. {\bf 17}, 2413 (1984);
427: M. Robnik, in {\em Atomic Spectra and Collisions in External Fields},
428: eds. K.T. Taylor et al (Plenum, 1988) pp265-274;
429: M. Robnik, Nonlinear Phenomena in Complex Systems (Minsk)
430: {\bf 1}, No.1, 1 (1998);
431: M. Robnik, J. Phys. Soc. Jpn. Suppl. C {\bf 72}, 81 (2003).
432:
433: \bibitem{Prosen:94} T. Prosen and M. Robnik,
434: J. Phys.A: Math.Gen. {\bf 27}, 8059 (1994);
435: T. Prosen, J. Phys. A: Math.Gen. {\bf 31}, 7023 (1998);
436: T. Prosen and M. Robnik, J.Phys.A: Math.Gen. {\bf 32}, 1863 (1999);
437: J. Malovrh and T. Prosen, J.Phys.A: Math.Gen. {\bf 35}, 2483 (2002).
438:
439: \end{thebibliography}
440:
441: \end{document}
442:
443:
444:
445:
446: