1: 

2: \documentstyle[aps,twocolumn,prb,aps]{revtex}

3: 

4: \begin{document}

5: 

6: \wideabs{

7: \title{{\bf Nonlinear parametric instability in double-well lattices}}

8: \author{Jaroslav Riedel$^{1}$, Boris A. Malomed$^{2}$, and Eva Majern\'{\i}kov\'{a}$

9: ^{1}$}

10: \address{$^1$Department of Theoretical Physics, Palack\'{y} University,\\

11: T\v{r}. 17. Listopadu 50, 77207 Olomouc, Czech Republic\\

12: $^2$Department of Interdisciplinary Studies, Faculty of Engineering,\\

13: Tel Aviv University, Tel Aviv 69978, Israel}

14: \maketitle

15: 

16: 

17: \begin{abstract}

18: A possibility of a nonlinear resonant instability of uniform oscillations in

19: dynamical lattices with harmonic intersite coupling and onsite nonlinearity

20: is predicted. Numerical simulations of a lattice with

21: a double-well onsite anharmonic potential confirm the existence of the

22: nonlinear instability with an anomalous value of the corresponding power

23: index, $\approx 1.57$, which is intermediate between the values

24: $1$ and $2$ characterizing the linear and nonlinear (quadratic)

25: instabilities. The anomalous power index may be a result

26: of a competition

27: between the resonant quadratic instability and nonresonant linear

28: instabilities. The observed instability

29: triggers transition of the lattice into a chaotic dynamical state.

30: \end{abstract}

31: }

32: 

33: \narrowtext

34: Dynamical lattices with onsite nonlinearity and harmonic intersite coupling

35: constitute a vast class of models which have numerous physical application

36: and are an object of great interest in their own right, see, e.g., Ref. \cite

37: {general}. Among these models, the ones with a {\it double-well} onsite

38: potential [given, e.g., by the expression (\ref{potential}) below] have

39: special importance, as they directly apply to the description of structural

40: transitions in dielectrics, semiconductors, superconductors, and optical

41: lattices (see recent works \cite{doublewell} and references therein), and

42: find other applications \cite{others}.

43: 

44: The simplest dynamical state in lattices represents spatially homogeneous

45: oscillations. This state in conservative lattice models is sometimes stable,

46: and sometimes it is subject to linear modulational instabilities initiating

47: a transition to nontrivial dynamics \cite{MI}. An objective of the present

48: work is to demonstrate analytically, and verify by direct simulations, that

49: homogeneous oscillatory states in lattices with the double-well

50: anharmonicity may be subject to a specific {\em nonlinear} instability,

51: which triggers transition of the lattice into a chaotic dynamical state.

52: 

53: The nonlinear parametric instability, which is a subject of this work, is

54: inherently related to the phonon anharmonism in the lattice: the instability

55: is caused by a resonance involving the uniform oscillations of the lattice

56: and a second harmonic of the phonon mode. A possibility of a nonlinear

57: instability of an {\it intrinsic localized mode} \cite{IM}\ in the lattice

58: due to the phonon anharmonism was first considered in Ref. \cite{me}. An

59: opposite, and more common, type of the resonance, namely, between a strictly

60: linear phonon mode and a higher harmonic of an intrinsic localized mode is

61: well known to give rise to a slow decay of the localized mode into phonons

62: (see recent works \cite{Aubry} and references therein).

63: 

64: The general form of the lattice equation of motion is 

65: \begin{equation}

66: \ddot{u}_{n}+f\left( u_{n}\right) u_{n}=u_{n+1}+u_{n-1}-2u_{n},

67: \label{general}

68: \end{equation}

69: where $u_{n}$ are real dynamical variables on the lattice, the overdot

70: stands for $d/dt$, $f\left( u_{n}\right) $ is a polynomial function

71: accounting for the onsite nonlinearity (in fact, nonpolynomial functions can

72: be considered too), and the right-hand side of the equation accounts for the

73: intersite harmonic coupling. The linearized version of Eq. (\ref{general})

74: gives rise to phonon modes 

75: \begin{equation}

76: u_{n}=A\sin \left( kn-\omega t\right)  \label{mode}

77: \end{equation}

78: with an arbitrary infinitesimal amplitude $A$ and the dispersion relation $%

79: \omega =2\left| \sin (k/2)\right| $ inside the phonon band, $\omega \leq 2$

80: (by definition, the frequencies are positive).

81: 

82: A homogeneous oscillatory state $U_{0}(t)$ is a time-periodic solution to

83: the equation 

84: \begin{equation}

85: \ddot{U}_{0}+f\left( U_{0}\right) U_{0}=0  \label{U0}

86: \end{equation}

87: with a fundamental frequency $\Omega $. The linear stability of the

88: homogeneous state is determined by a linearized equation for small

89: perturbations $\delta u_{n}$, which is produced by the substitution of $%

90: u_{n}=U_{0}(t)+\delta u_{n}$ into Eq. (\ref{general}): 

91: \begin{eqnarray}

92: &&\delta \ddot{u}_{n}+\left[ f^{\,\,\prime }\left( U_{0}(t)\right)

93: U_{0}(t)+f\left( U_{0}(t)\right) \right] \delta u_{n}  \nonumber \\

94: &=&\delta u_{n+1}+\delta u_{n-1}-2\delta u_{n}\,.  \label{linearized}

95: \end{eqnarray}

96: \newline

97: In the mean-field approximation, one may describe phonon modes of the type (%

98: \ref{mode}) on top of the homogeneous oscillations, replacing the

99: coefficient in front of $\delta u_{n}$ on the left-hand side of Eq. (\ref

100: {linearized}) by its time-average value $\omega _{0}^{2}\equiv \left\langle

101: f^{\prime }\left( U_{0}(t)\right) U_{0}(t)+f\left( U_{0}(t)\right)

102: \right\rangle $. The corresponding dispersion relation for the phonon modes

103: acquires a {\it gap} $\omega _{0}$, so that 

104: \begin{equation}

105: \omega ^{2}=\omega _{0}^{2}+4\sin ^{2}(k/2),  \label{dispersion}

106: \end{equation}

107: which gives rise to the phonon band 

108: \begin{equation}

109: \omega _{0}^{2}\leq \omega ^{2}\leq 4+\omega _{0}^{2}  \label{band}

110: \end{equation}

111: (if $\omega _{0}^{2}<0$, the homogeneous oscillations are immediately

112: unstable).

113: 

114: Beyond this simple approximation, the Fourier decomposition of the

115: coefficient $f^{\,\,\prime }\left( U_{0}(t)\right) U_{0}(t)+f\left(

116: U_{0}(t)\right) $ in Eq. (\ref{linearized}) gives rise to parametrically

117: driven terms $\,\sim \cos (m\Omega t)\cdot \delta u_{n}$ with all integer

118: values of $m$. The linear parametric drive resonates with a perturbation

119: frequency $\omega $, i.e., it may give rise to a {\em resonant} linear

120: instability, under the condition $m\Omega -\omega =\omega $, or 

121: \begin{equation}

122: \omega =\omega _{{\rm res}}^{{\rm (lin)}}\equiv (m/2)\Omega .

123: \label{linearRes}

124: \end{equation}

125: If any resonant frequency $(m/2)\Omega $ gets into the phonon band (\ref

126: {band}), the homogeneous oscillatory state is expected to be modulationally

127: unstable, otherwise the resonant linear instability does not take place.

128: 

129: In the latter case, it makes sense to seek for nonlinear parametric

130: instabilities, the simplest of which may be generated by a cubic term, or

131: any higher-order one, in the onsite nonlinearity. Indeed, the cubic term

132: generates a nonlinear correction $\sim U_{0}(t)\cdot \left( \delta

133: u_{n}\right) ^{2}$ to Eq. (\ref{linearized}), which may be regarded as a

134: parametric drive that can give rise to a {\em nonlinear} parametric

135: resonance under the condition $m\Omega -2\omega =\omega $, or 

136: \begin{equation}

137: \omega =\omega _{{\rm res}}^{{\rm (nonlin)}}\equiv (m/3)\Omega ,

138: \label{nonlinRes}

139: \end{equation}

140: where $m$ is an arbitrary integer different from a multiple of $3$, cf. Eq. (%

141: \ref{linearRes}) (if $m$ is a multiple of $3$, the linear parametric

142: resonance takes place at the same frequency, so that the nonlinear resonance

143: is insignificant).

144: 

145: Of course, this description has a very approximate nature for two reasons.

146: First, the phonon band (\ref{band}) was defined in the framework of the

147: mean-field approximation, hence one cannot be sure in the accuracy of the

148: predictions based on the comparison of the resonant frequencies with this

149: band. Second, the full lattice model (\ref{general}) may give rise to other

150: instabilities, which are not related to the parametric resonance. Therefore,

151: the above consideration should only be considered as a qualitative clue, and

152: an actual possibility of dynamical regimes dominated by the nonlinear

153: resonance must be checked by direct simulations.

154: 

155: Continuing the consideration, we note that, if none of the linear-resonance

156: frequencies (\ref{linearRes}) gets into the renormalized band (\ref{band}),

157: but a nonlinear-resonance frequency (\ref{nonlinRes}) can be found inside

158: the band, an evolution equation for the amplitude of the corresponding

159: resonant-perturbation mode, $\delta u_{n}=A(t)\cos \left( \omega _{{\rm res}%

160: }t\right) \cdot v_{n}$ , with some spatial profile $v_{n}$ (it may be, for

161: instance, the above-mentioned localized intrinsic mode), has a general form 

162: \begin{equation}

163: dA/dt=CA^{2},  \label{nonlinear}

164: \end{equation}

165: where $C$ is a constant which depends on a particular form of Eq. (\ref

166: {general}) and the homogeneous solution $U_{0}(t)$; cf. a similar equation

167: governing the nonlinear instability of the so-called {\it embedded solitons} 

168: \cite{embedded}. A solution to Eq. (\ref{nonlinear}) is 

169: \begin{equation}

170: A=A_{0}/\left( 1-CA_{0}t\right) ,  \label{growth}

171: \end{equation}

172: where $A_{0}$ is the initial value of the perturbation amplitude. A drastic

173: difference of the perturbation growth law (\ref{growth}) from the

174: exponential growth in the case of the linear instability is that the

175: nonlinear instability is initially growing much slower than an exponential,

176: and a characteristic time scale of the growth, $\sim 1/\left( CA_{0}\right) $%

177: , depends on the initial perturbation $A_{0}$, while in the case of the

178: exponential growth it is a fixed constant. However, the nonlinear

179: instability is self-accelerating, and, as a manifestation of that, Eq. (\ref

180: {growth}) formally predicts a singularity at $t=1/\left( CA_{0}\right) $. In

181: reality, of course, the singularity may not occur, as the above

182: approximation, taking into regard the first nonlinear correction to Eq. (\ref

183: {linearized}), becomes irrelevant if $A(t)$ is too large. A natural

184: conjecture, that will be corroborated by direct simulations below, is that

185: the nonlinear instability leads to a chaotic dynamical state.

186: 

187: It is relevant to mention that, although nonlinear instabilities are less

188: common than the usual linear instability, they occur and play an important

189: role in many physical problems, as diverse as optical solitons in media with

190: competing quadratic and cubic nonlinearities \cite{embedded}, Bose gases,

191: plasma turbulence, contact lines in flows, etc. \cite{nonlininstab}. In this

192: work, we will check the possibility of the nonlinear instability of the

193: homogeneous oscillations in the lattice model (\ref{general}) with 

194: \begin{equation}

195: f(u_{n})=-u_{n}^{2}+\nu u_{n}^{4},\,\nu >0,  \label{f}

196: \end{equation}

197: which corresponds to the double-well onsite anharmonic potential, 

198: \begin{equation}

199: V(u_{n})=-u_{n}^{4}/4+\nu u_{n}^{6}/6.  \label{potential}

200: \end{equation}

201: In this case, Eq. (\ref{U0}) can be solved in terms of elliptic functions,

202: but an explicit result is very cumbersome.

203: 

204: As a typical example, we take homogeneous oscillations produced by Eq. (\ref

205: {U0}) with $\nu =0.01$ and initial conditions $U_{0}(0)=1$ and $\stackrel{.}{%

206: U}_{0}(0)=0$. The variable $U_{0}$ then performs strongly anharmonic

207: oscillations between the values $\left( U_{0}\right) _{\min }=1$ and $\left(

208: U_{0}\right) _{\max }=12.247$ at the fundamental frequency $\Omega

209: =\allowbreak 1.\,\allowbreak 694$, and the gap in the renormalized phonon

210: spectrum (\ref{dispersion}) is calculated to be $\omega _{0}=\allowbreak

211: 3.\,518$, so that the renormalized band (\ref{band}) is, in the mean-field

212: approximation, 

213: \begin{equation}

214: 3.518<\omega <4.\allowbreak 047.  \label{realband}

215: \end{equation}

216: Then, it is straightforward to check that all the linear-resonance

217: frequencies (\ref{linearRes}) do {\em not} get into this band (the band as

218: whole fits between the linear resonant frequencies $3.392$ and $4.240$,

219: which correspond to $m=4$ and $m=5$). On the other hand, the

220: nonlinear-resonance frequency (\ref{nonlinRes}) corresponding to $m=7$ is $%

221: \omega _{{\rm res}}^{{\rm (nonlin)}}=\allowbreak 3.\,\allowbreak 957\,3$,

222: which lies {\em inside} the band (\ref{realband}) [all the other frequencies

223: given by Eq. (\ref{realband}) are located outside the band].

224: 

225: To directly test the instability, small perturbations of the form 

226: \begin{equation}

227: \delta u_{n}(0)=A_{0}\cos (2\pi p_{0}n/N),  \label{p}

228: \end{equation}

229: where $N$ is the net number of sites in the lattice and $p_{0}$ is an

230: integer, were added to the homogeneous oscillatory state. The lattice

231: equations of motion were solved for $N=1000$ and periodic boundary

232: conditions by means of the eighth-order explicit Runge-Kutta scheme with a

233: stepsize control such that the time step was dynamically changed within the

234: range $0.05$ - $0.3$. It was checked that the relative (per site) error at

235: each step did not exceed $10^{-10}$.

236: 

237: The simulations were performed for the perturbations (\ref{p}) with $p_{0}$

238: taking values in the interval $1\leq p_{0}\leq 30$. In all the cases

239: considered, results were quite similar. Here, we demonstrate a typical

240: example with $p_{0}=20$. Long-time evolution initiated by the small

241: perturbation (\ref{p}) with $A_{0}=0.05$ is displayed is Fig. 1 in the form

242: of a set of plots showing the temporal development of several components in

243: the Fourier transform of the lattice field, which are defined as follows: 

244: \begin{eqnarray}

245: U_{p}(t) &=&\left( 2/N\right) \sum_{n=1}^{N}u_{n}(t)\exp \left( 2i\pi

246: pn/N\right) ,\,p\neq 0;  \nonumber \\

247: U_{0}(t) &=&\left( 1/N\right) \sum_{n=1}^{N}u_{n}(t)\,.  \label{Fourier}

248: \end{eqnarray}

249: It is obvious that the small perturbation triggers a transition of the

250: lattice into a chaotic state. Fully developed chaos, i.e., a state in which

251: all the lattice modes are involved into the chaotic motion, is attained at $%

252: t\approx 22$, when the phonon mode with $p=p_{0}+1$ gets chaotically excited

253: too, see Fig. 1. To further illustrate the transition to chaos, in Fig. 2 we

254: additionally show in detail, on the logarithmic scale, the growth of the

255: amplitude $|U_{p_{0}+1}(t)|$. More detailed studies of the established

256: chaotic state may be of interest in their own right, but this problem is

257: beyond the scope of the present work.

258: 

259: As concerns the nonlinear character of the instability, a crucial issue is

260: the growth of the perturbation at the initial stage. It is necessary to

261: check whether it is indeed essentially different from the familiar

262: exponential law, being, instead, close to the Eq. (\ref{growth}). To this

263: end, in Fig. 3 we display the best fit of the time evolution of the

264: numerically computed Fourier amplitude $|U_{p_{0}}(t)|$ to a function 

265: \begin{equation}

266: A_{{\rm fit}}(t)=\Delta \cdot (1-\gamma t)^{-\alpha },  \label{fit}

267: \end{equation}

268: where the parameters are found to be $\Delta =0.041$, $\gamma =0.560$ and $%

269: \alpha =1.750$.

270: 

271: Comparison of these results with Eq. (\ref{growth}) shows a difference in

272: the (most essential) power parameter $\alpha $. Note that the expression ( 

273: \ref{fit}) with the empirically found value $\alpha =1.750$ formally

274: corresponds to a solution to the nonlinear evolution equation $%

275: dA/dt=CA\allowbreak ^{\beta }$, with an anomalous value of the power index, $%

276: \beta \equiv 1+\alpha ^{-1}\approx 1.5714$, that should be compared to Eq. ( 

277: \ref{nonlinear}), valid in case of the ordinary nonlinear instability \cite

278: {embedded}. This anomalous value is sort of intermediate between $\beta =1$

279: and $\beta =2$, which are expected for the linear and nonlinear

280: instabilities, respectively. This result may suggest that, in fact, in the

281: present model we have a competition between the resonant nonlinear

282: instability, qualitatively considered above, and linear instabilities

283: against {\em nonresonant} perturbations, which were not taken into regard in

284: the above consideration. While an accurate analysis of the full linear

285: stability problem of the homogeneous oscillations is a technically complex

286: problem, that we do not aim to consider here, Fig. 3 clearly shows that the

287: resonant nonlinear instability dominates in the growth of the perturbations.

288: 

289: In conclusion, we have proposed a possibility of a nonlinear resonant

290: instability of homogeneous oscillations in harmonically coupled nonlinear

291: lattices, which is expected to play a dominant role, provided that no

292: resonant frequency accounting for the linear parametric resonant instability

293: gets into the renormalized phonon band, while a frequency that gives rise to

294: a quadratic parametric resonance is found in the band. Numerical simulations

295: of the lattice with a double-well onsite anharmonic potential confirm the

296: existence of nonlinear instability with an anomalous value of the power

297: index $\approx 1.57$, which is intermediate between the values $1$ and $2$,

298: characteristic of the linear and nonlinear instabilities. The onset of the

299: nonlinear instability triggers transition of the lattice into a chaotic

300: dynamical state.

301: 

302: A valuable discussion with P.G. Kevrekidis is acknowledged. B.A.M.

303: appreciates hospitality of the Department of Theoretical Physics at the

304: Palack\'{y} University (Olomouc, the Czech Republic). E.M. acknowledges a

305: partial support by grants No. 202/01/1450 from the agency GACR and by VEGA

306: No. 2/7174/20.

307: 

308: \begin{references}

309: \bibitem{general}  J. Leon and M. Manna, J. Phys. A {\bf 32}, 2845 (1999).

310: 

311: \bibitem{doublewell}  A.B. Shick, J.B. Ketterson, D.L. Novikov, and A.J.

312: Freeman, Phys. Rev. B {\bf 60}, 15484 (1999); M.B. Smirnov, \ Phys. Rev. B 

313: {\bf 59}, 4036 (1999); F. Cordero, R. Cantelli, M. Corti, A. Campana, and A.

314: Rigamonti, Phys. Rev. B {\bf 59},12078 (1999); D.L. Haycock, P.M. Alsing,

315: I.H. Deutsch, J. Grondalski, and P.S. Jessen, Phys. Rev. Lett. {\bf 85},

316: 3365 (2000).

317: 

318: \bibitem{others}  J.C. Comte, P. Marqui\'{e}, and M. Remoissenet, Phys. Rev.

319: E {\bf 60}, 7484 (1999).

320: 

321: \bibitem{IM}  S. Aubry, Physica D {\bf 103}, 201 (1996); D. Hennig, and G.P.

322: Tsironis, Phys. Rep. {\bf 307}, 335 (1999).

323: 

324: \bibitem{me}  B.A. Malomed, Phys. Rev. B {\bf 49}, 5962 (1994).

325: 

326: \bibitem{Aubry}  M. Johansson and S. Aubry, Phys. Rev. E {\bf 61}, 5864

327: (2000); P.G. Kevrekidis and M.I. Weinstein, Physica D {\bf 142}, 113 (2000).

328: 

329: \bibitem{MI}  J. Leon and M. Manna, Phys. Rev. Lett. {\bf 83}, 2324 (1999);

330: Y. Kosevich and S. Lepri, Phys. Rev. B {\bf 61}, 299 (2000).

331: 

332: \bibitem{Shukla}  P.K. Shukla, Phys. Rev. Lett. {\bf 84}, 5328 (2000).

333: 

334: \bibitem{embedded}  J. Yang, B.A. Malomed, and D.J. Kaup, Phys. Rev. Lett. 

335: {\bf 83}, 1958 (1999); A.R. Champneys, B.A. Malomed, J. Yang and D.J. Kaup,

336: Physica D {\bf 152-153}, 340 (2001).

337: 

338: \bibitem{nonlininstab}  S. Khlebnikov, Phys. Rev. D {\bf 62}, 043519 (2000);

339: J.A. Krommes, {\bf 41}, A641 (1999); S. Kalliadasis, J. Fluid Mech. {\bf 413}

340: , 355 (2000).

341: \end{references}

342: 

343: \newpage

344: 

345: \section*{Figure Captions}

346: 

347: Fig. 1. The time dependence for selected Fourier amplitudes $|U_{p}(t)|$,

348: defined as per Eq. (\ref{Fourier}). The results are shown for $p=0$, $%

349: p=p_{0} $, $p=2p_{0}$, and $p=p_{0}+1$, where $p_{0}=20$. The lattice size

350: is $N=1000 $ (with periodic boundary conditions), and the initial amplitude

351: of the perturbation is $A_{0}=0.05$.

352: 

353: Fig. 2. Details of the evolution of the Fourier amplitude $|U_{p_{0}+1}(t)|$%

354: , shown on the logarithmic scale.

355: 

356: Fig. 3. Fitting the time dependence of the amplitude $|U_{p_{0}}(t)|$ to the

357: function (\ref{fit}) with $\Delta =0.0411$, $\gamma =0.560$ and $\alpha

358: =1.750$. Diamonds stand for numerical data, and stars (which almost

359: completely overlap with the diamonds) show the closest values provided by

360: the fitting function.

361: 

362: \end{document}

363: