1: \documentstyle[preprint,aps,eqsecnum]{revtex}

2: \input epsf

3: 

4: \newcommand{\eeq}{\end{equation}}

5: \newcommand{\vs}[1]{\rule[- #1 mm]{0mm}{#1 mm}}

6: \newcommand{\beq}{\begin{equation}}

7: \newcommand{\beql}{\begin{eqnarray}}

8: \newcommand{\eeql}{\end{eqnarray}}

9: \newcommand{\lp}{\left(}

10: \newcommand{\rp}{\right)}

11: \newcommand{\scr}{\scriptstyle}

12: \newcommand{\om}{\omega}

13: \newcommand{\al}{\alpha}

14: \newcommand{\lam}{\lambda}

15: \newcommand{\Lam}{\Lambda}

16: \newcommand{\eps}{\epsilon}

17: \newcommand{\dV}{\frac{d}{dV}}

18: \newcommand{\dxi}{\frac{dx_i}{dV}(p)}

19: \newcommand{\parV}{\frac{\partial}{\partial V}}

20: \newcommand{\sect}[1]{\setcounter{equation}{0}\section{#1}}

21: \newcommand{\NP}[3]{{\it Nucl. Phys. }{\bf B#1} (#2) #3}

22: \newcommand{\PL}[3]{{\it Phys. Lett. }{\bf B#1} (#2) #3}

23: \newcommand{\PR}[3]{{\it Phys. Rev. }{\bf #1} (#2) #3}

24: \newcommand{\PRL}[3]{{\it Phys. Rev. Lett. }{\bf #1} (#2) #3}

25: \newcommand{\IMP}[3]{{\it Int. J. Mod. Phys }{\bf #1} (#2) #3}

26: \newcommand{\MPL}[3]{{\it Mod. Phys. Lett. }{\bf #1} (#2) #3}

27: \newcommand{\JP}[3]{{\it J. Phys. }{\bf A#1} (#2) #3} 

28: \renewcommand{\theequation}{\thesection.\arabic{equation}}

29: 

30: \begin{document}

31: 

32: \draft 

33: 

34: 

35: \title{Spectral Statistics of Instantaneous Normal Modes in Liquids and Random Matrices}

36: 

37: \author{Srikanth Sastry$^1$, Nivedita Deo$^{1,3}$ and Silvio Franz$^2$}

38: 

39: \address{$^1$ Jawaharlal Nehru Center for Advanced Scientific Research, Bangalore 560064, India}

40: \address{$^2$ The Adbus Salam International Centre for Theoretical Physics,

41: Treieste, Italy}

42: \address{$^3$ Santa Fe Institute,

43: 1399 Hyde Park Road, Santa Fe, NM 87501, USA}

44: 

45: \maketitle

46: 

47: \begin{abstract}

48: We study the statistical properties of eigenvalues of the Hessian

49: matrix ${\cal H}$ (matrix of second derivatives of the potential

50: energy) for a classical atomic liquid, and compare these properties

51: with predictions for random matrix models (RMM). The eigenvalue

52: spectra (the Instantaneous Normal Mode or INM spectra) are evaluated

53: numerically for configurations generated by molecular dynamics

54: simulations. We find that distribution of spacings between nearest

55: neighbor eigenvalues, $s$, obeys quite well the Wigner prediction $~

56: s~exp(-s^2)$, with the agreement being better for higher densities at

57: fixed temperature. The deviations display a correlation with the

58: number of localized eigenstates (normal modes) in the liquid; there

59: are fewer localized states at higher densities which we quantify by

60: calculating the participation ratios of the normal modes. We confirm

61: this observation by calculating the spacing distribution for parts of

62: the INM spectra with high participation ratios, obtaining greater

63: conformity with the Wigner form. We also calculate the spectral

64: rigidity and find a substantial dependence on the density of the

65: liquid.

66: \end{abstract}

67: 

68: \pacs{PACS numbers: 75.10.Nr, 61.20.-p, 61.43.-j, 64.70.Pf, 05.40.-a, 05.45+b}

69: 

70: \section{Introduction:}

71: The local topography of the potential energy surface, as characterized

72: by the ensemble averaged spectrum (INM spectrum) of eigenvalues of the

73: second-derivative matrix (Hessian) of the potential energy function,

74: have been studied in recent years as an approach to the analysis of

75: dynamics in liquids\cite{keyes,Sciortino}. Broadly, the study

76: of the INM spectra has been directed at analysing short time dynamics 

77: as in studying solvation \cite{strattcho}, and at elucidating information

78: about pathways to long time relaxation in the form of potential energy 

79: barriers {\it etc}\cite{keyes,Sciortino}. 

80: 

81: Considerable effort has also been dedicated to developing analytical

82: theories for calculating the INM spectra within an equilibrium

83: description\cite{xustratt,wu,stratt,mona,cava}. The approach in much

84: of these attempts has been to formulate the problem of calculating the

85: INM spectrum as an exercise in random matrix theory. If one treats the

86: individual elements of the Hessian matrix as independent, and

87: distributed according to the appropriate Boltzmann weight, then the

88: Hessian may be viewed as a real, symmetric random matrix with a known

89: distribution of matrix elements. Two properties, however, distinguish

90: the Hessian from the standard corresponding case treated in random

91: matrix theory: (i) The diagonal entries of the Hessian are related to

92: the off-diagonal entries by the property ${\cal H}^{\alpha \beta}_{ii}

93: = - \sum_{j \neq i} {\cal H}^{\alpha \beta}_{ij}$, where $i,j$ label

94: the particles and $\alpha, \beta$ the spatial coordinates $x, y,

95: z$. (ii) For liquids with short-ranged interaction potentials, the

96: Hessian matrix is sparse, with the fraction of non-zero entries, $p$,

97: depending on the system size as $p \sim 1/N$.

98: 

99: In view of the above considerations, it is of interest to inquire to

100: what extent the INM spectrum displays universal features idenitied in

101: random matrix theory. In this paper, we address this question by

102: obtaining INM spectra numerically for a model atomic liqiud that has

103: been studied in the context of slow dynamics in supercooled

104: liquids\cite{sastryNature}. The statistics we consider are the

105: spacing statistics between nearest neighbor eigenvalues and the 

106: spectral rigidity, which we explain below. 

107: 

108: \noindent

109: \section{Eigenvalues of the Hessian}

110: The model liquid we study is a binary mixture\cite{kob} composed of

111: $80 \%$ of particles of type $A$ and $20 \%$ of type $B$, interacting

112: via the Lennard-Jones potential, with Lennard-Jones parameters

113: $\epsilon_{AB}/\epsilon_{AA} = 1.5$, $\epsilon_{BB}/\epsilon_{AA} =

114: 0.5$, $\sigma_{AB}/\sigma_{AA} = 0.8$, and $\sigma_{BB}/\sigma_{AA} =

115: 0.88$, and a ratio of masses $m_B/m_A = 1$. Lennard-Jones reduced

116: units are used to report all the quantities, in terms of the $A$

117: particle paramters $\epsilon_{AA}$, $\sigma_{AA}$ and $m_{A}$:

118: temperatures as $T^{*} = k_B T/\epsilon_{AA}$, densities as $\rho^{*}

119: = \rho/\sigma_{AA}^3$ and Hessian eigenvalues $\lambda^{*} = \lambda

120: {m_A \sigma_{AA}^2 \over \epsilon_{AA}}$. Further details may be found

121: in \cite{sastryNature}.  Molecular dynamics simulations of the liquid

122: are performed at ten reduced densities $\rho^{*}$ from $\rho^{*} =

123: 0.65 $ to $1.40$, at reduced temperature $T^{*} = 1.0$. Two hundred

124: sample configurations in each case are chosen from the equilibrated

125: trajectory for the INM analysis. For each of these configurations, the

126: Hessian is calculated and diagonalized numerically to obtain the

127: eigenvalues $\lambda_i$ as well as the eigenvectors ${\bf e}_i$. The

128: eigenvectors are used to calculate the localization properties of the

129: normal modes, {\it via} the participation ratio

130: 

131: \begin{equation}

132: P_i = \left[N \sum_{\alpha = 1}^{3N} ({\bf e}_i^\alpha.{\bf e}_i^\alpha)^2 \right]^{-1}. 

133: \end{equation}

134: 

135: The participation ratio thus defined is small (order of $1/N$) for localized modes and large (order of $1$) for extended modes. 

136: 

137: By constructing the histogram of eigenvalues $\lambda$ of the Hessian

138: for all configurations considered, we obtain the INM Density of States

139: (DOS) or spectrum\cite{fn1}. Figure 1 shows the INM spectrum

140: $D(\lambda)$ verses $\lambda$ for different densities. Note that the

141: DOS is very different from the Wigner semi-circle distribution,

142: obtained in RMM. We know from the literature\cite{bz} that the

143: correlation and spacing functions are universal in certain regions no

144: matter what the DOS is. Thus we use the corresponding statistics of

145: spacings between eigenvalues, Wigner-Dyson statistics, as the standard

146: of reference.

147: 

148: \noindent

149: \section{Unfolding The Spectrum}

150: The statistical analysis of the numerical data proceeds by first using

151: an {\it unfolding procedure}. The numerical calculation yields the

152: eigenvalues of the Hessian which is ordered and forms the sample

153: spectrum $\{\lambda_1,\lambda_2,...,\lambda_n\}$. In order to analyse

154: the spacing statistics, one must transform the eigenvalues $\lambda_i$

155: in such a way that the transformed eigenvalues $\zeta_i$ are uniformly

156: distributed.  That is, the spectral density $D(\zeta) = 1$. This

157: procedure is referred to as ``unfolding'' the

158: spectrum\cite{mehta,guhr}. In general such a transformation for a

159: spectral density function $D(\lambda)$ is most easily accomplished

160: through its cumulative distribution

161: \begin{equation}

162: C(\lambda) \equiv \int_{-\infty}^{\lambda} D(\lambda) d\lambda,

163: \end{equation}

164: by defining $\zeta(\lambda) = C(\lambda)$. For a discrete spectrum

165: such as the ones we consider here, the corresponding procedure would

166: be to extract from the sample spectrum the ``smooth'' part of the

167: ``staircase'' cumulative distribution\cite{guhr}. In this work, the

168: procedure we adopt to estimate the smooth part of the cumulative

169: distribution is to evaluate the cumulative distribution for the union

170: of all eigenvalues $\lambda_i$ obtained for $200$ independent

171: configurations at each density and temperature value. Such a smooth

172: cumulative spectrum along with the ``staircase'' cumulative spectrum

173: for a single configuration, is shown in Fig. 2, and the unfolding

174: procedure is indicated.

175: 

176: \noindent

177: \section{Spacing Distribution}

178: The spacing distribution $P(s)$ for the random matrix models is

179: defined as the probability of finding the next nearest neighbor

180: eigenvalue of the spectrum to be at a distance s i.e.  $

181: s_i={{\lambda_{i+1}-\lambda_i}\over {\Delta}} $ where $ \Delta $ is

182: the mean level spacing. In the present case where we use unfolded

183: eigenvalues, $s_i=\zeta_{i+1}-\zeta _{i}$.  Then $P(s)= A s^{\beta}

184: e^{-B s^2}$ where $\beta=1$ for the orthogonal random matrix model,

185: which is the ``Wigner surmise''\cite{mehta}.  This spacing

186: distribution arises in various quantum systems which show an

187: underlying classically chaotic behavior e.g. quantum billiards,

188: quantum dots, nuclear spectra, disordered mesoscopic systems. The

189: system we analyze, on the other hand, is a classical liquid, with a

190: disordered microscopic structure. The spacing distributions obtained

191: are shown in Fig. 3 for three densities of the liquid. We find that to

192: a first approximation, the cases considered display the universal

193: behavior according to the Wigner surmise, with the agreement being

194: better for the liquid at higher densities. We note here that in

195: Ref. \cite{cava} a sparse random matrix is used to calculate the INM

196: DOS analytically for a one-dimensional system which is very different

197: from the usual Wigner semi-circle, and displays qualitative features

198: very similar to the DOS we calculate numerically here.  It would be

199: very interesting to see whether the correlators of the sparse random

200: matrix proposed in ref. \cite{cava} captures the above behaviour for

201: different densities for the liquid.

202: 

203: \noindent

204: \section{Spectral Rigidity}

205: We next study the spectral rigidity, which measures the fluctuations

206: of the number of eigenvalues in a window of given size as a function

207: of the size of the window (or equivalently, the average number of

208: eigenvalues expected in the window). The number fluctuations are

209: plotted as a function of the mean number of eigenvalues in Fig. 4 for

210: the same three densities as in the preceding figures. For the case of

211: the Poisson spectrum and the harmonic oscillator\cite{mehta} the

212: number fluctuations are $ <\delta N^2> \sim <N> $ and $ <\delta N^2> =

213: constant $, while for the Gaussian random matrix ensembles $ <\delta

214: N^2> \sim ln <N> $ for $ N \gg 1 $.

215: 

216: 

217: The number fluctuations are of the form $ <\delta N^2> \sim

218: N^{\gamma} $ where $\gamma(\rho)$ depends on the liquid's density.

219: This situation is reminescent of that found for the mobility edge in

220: the quantum hall effect and in the Anderson metal-insulator

221: transition.  A detailed analysis of this awaits future work.

222: 

223: \noindent

224: \section{Participation Ratio}

225: In order to get some insight into the reason for the systematically

226: better agreement with increasing density of the spacing distribution,

227: we consider the localization properties of the normal modes in the

228: liquid.  In the standard random matrix case for orthogonal matrices,

229: the eigenvectors are all extended, while from previous numerical and

230: analytical studies we know that a fraction of the INM eigenmodes are

231: localized. As described earlier, we use the participation ratio to

232: quantify the localization of modes, averaging over modes corresponding

233: to eigenvalues in each histogram bin, for the unfolded eigenvalues.

234: In Fig. 5, the participation ratio is plotted as a function of the

235: unfolded eigenvalues for $T^{*} =1.00$, for values of the density

236: $\rho^{*} =0.65, 1.00$ and $1.35$. One notes that for the highest density

237: $\rho=1.35$ the participation ratios are highest overall, while for the 

238: lower densities the participation ratios are quite small for a substantial 

239: fraction of the eigenmodes, indicating a large number of localized modes. 

240: 

241: 

242: 

243: Next, we calculate the spacing distribution for $\rho^{*} = 1.35$ for

244: unfolded eigenvalues between $0.1$ and $0.6$ for which the

245: participation ratio is high ($> 0.75$) and relatively unchanging

246: (Fig. \ref{fig:prat}). The resulting spacing distribution is shown in

247: Fig. 6, along with the spacing distribution for the entire eigenvalue 

248: spectrum and the expectation based on the RMM result. The data shown 

249: clearly demonstrate that the spacing distribution is practically 

250: identical to the standard RMM result, confirming the speculation that 

251: the increasing fraction of localized states at lower densities are 

252: responsible for the deviations at these densities from the standard 

253: RMM result. Indeed, this observation has been used, in a different context

254: to locate the mobility edge in disordered systems\cite{carpena}

255: 

256: \noindent

257: \section{Conclusions}

258: We have presented the spacing statistics and spectral rigidity for

259: numerically calculated INM spectra. The spacing statistics is seen to

260: conform better with increasing density (at fixed temerature), with the

261: predictions for random matrix models. We demonstrate that the source

262: of deviations from RMM predictions is related to the presence of

263: localized instantaneous normal modes in the liquid, whose number is

264: greater for lower density. As the two features distinguishing the INM

265: spectra from the standard random matrix case are the nature of the

266: diagonal elements and the sparseness of the Hessian matrix, further 

267: understanding of the non-universality of the INM spectra are to be

268: sought in the manner in which these aspects affect the INM spectral 

269: statistics. 

270: 

271: 

272: \noindent

273: 

274: \begin{flushleft}

275: 

276: \underline{Acknowledgements}:\\ 

277: 

278: We would like to thank S. Jain, V. E. Kravtsov, A. Cavagna,

279: I. Giardina, P. J. Garrahan and P. Carpena for very useful discussions

280: during the course of this work. SF thanks the Jawaharlal Nehru Center

281: for Advanced Scientific Research for hospitality.

282: \end{flushleft}

283: 

284: \begin{figure}

285: %\hbox to\hsize{\epsfxsize=0.6\hsize\hfil\epsfbox{sdf-fig1.eps}\hfil}

286: \caption{INM Density of States shown for three densities. Inset shows

287: the DOS on a logarithmic scale. Eigenvalues $\lambda$ are expressed in

288: units of ${\epsilon_{AA} \over m_A \sigma_{AA}^2}$, and $D(\lambda)$ in

289: units of ${m_A \sigma_{AA}^2 \over \epsilon_{AA}}$.}

290: \label{fig:inmdos}

291: \end{figure}

292: 

293: \begin{figure}

294: %\hbox to\hsize{\epsfxsize=0.6\hsize\hfil\epsfbox{sdf-fig2.eps}\hfil}

295: \caption{A portion of the staircase cumulative spectrum from a single

296: configuration is shown, along with the estimate of the smooth part of

297: the cumulative distribution, obtained from combining eigenvalues from

298: 200 configurations, for density $\rho^{*} = 1.0$ and temperature

299: $T^{*} = 1.0$. The arrows indicate the mapping of any given eigenvalue

300: $\lambda_i$ to the unfolded eigenvalue $\zeta_i$. The inset shows the

301: cumulative spectra for the full range. Eigenvalues $\lambda$ are

302: expressed in units of ${\epsilon_{AA} \over m_A \sigma_{AA}^2}$, and 

303: cumulative probabilities $\zeta$ are dimensionless.}

304: \label{fig:unfold}

305: \end{figure}

306: 

307: \begin{figure}

308: %\hbox to\hsize{\epsfxsize=0.6\hsize\hfil\epsfbox{sdf-fig3.eps}\hfil}

309: \caption{The level spacing distribution $P(s)$ for densities $\rho =

310: 0.65, 1.0, 1.35$. Level spacings $s$ are expressed in units of

311: ${\epsilon_{AA} \over m_A \sigma_{AA}^2}$, and $P(s)$ in units of

312: ${m_A \sigma_{AA}^2 \over \epsilon_{AA}}$.}

313: \label{fig:pofs}

314: \end{figure}

315: 

316: 

317: \begin{figure}

318: %\hbox to\hsize{\epsfxsize=0.5\hsize\hfil\epsfbox{sdf-fig4.eps}\hfil}

319: \caption{Spectral rigidity: 

320: Lines are fits to the form $<N^2> - <N>^2 \sim <N>^\gamma$, and the

321: values of $\gamma$ are $1.036, 0.85, 0.765$ respectively for $\rho^{*} = 0.65, 1.00, 1.35$.  }

322: \label{fig:rigid}

323: \end{figure}

324: 

325: 

326: \begin{figure}[t]

327: %\hbox to\hsize{\epsfxsize=0.5\hsize\hfil\epsfbox{sdf-fig5.eps}\hfil}

328: \caption{Participation ratio as a function of unfolded eigenvalues

329: $\zeta$ (see caption of Fig. 2).}

330: \label{fig:prat}

331: \end{figure}

332: 

333: 

334: \begin{figure}[t]

335: %\hbox to\hsize{\epsfxsize=0.5\hsize\hfil\epsfbox{sdf-fig6.eps}\hfil}

336: \caption{The level spacing distribution shown for $\rho^{*} = 1.35$ for

337: the full eigenvalue spectrum (open rhombs), for the range of

338: eigenvalues with high participation ratio ($\> 0.75$) (filled rhombs),

339: along with the RMM prediction. Level spacings $s$ are expressed in units of

340: ${\epsilon_{AA} \over m_A \sigma_{AA}^2}$, and $P(s)$ in units of

341: ${m_A \sigma_{AA}^2 \over \epsilon_{AA}}$.}

342: \label{fig:pofs2}

343: \end{figure}

344: 

345: 

346: \begin{thebibliography}{99}

347: 

348: \bibitem[] {} Email: {$^1$sastry@jncasr.ac.in, $^2$ndeo@jncasr.ac.in,

349: $^3$franz@ictp.trieste.it}.

350: 

351: \bibitem{keyes} T. Keyes, {\it J. Chem. Phys.} {\bf A 101},

352: 2921, (1997) and references therein.

353: 

354: \bibitem{Sciortino}

355: F. Sciortino and P. Tartaglia, Phys. Rev. Lett. {\bf 78}, 2385 (1997).

356: 

357: \bibitem{strattcho} R. M. Stratt and M. Cho, {\it J. Chem. Phys.} {\bf 100},

358: 6700, (1994).

359: 

360: \bibitem{xustratt} B.-C. Xu and R. M. Stratt, {\it J. Chem. Phys.} {\bf 92}, 1923 (1990).

361: 

362: \bibitem{wu} T. M. Wu and R. F. Loring, {\it J. Chem. Phys.} {\bf 97},

363: 8568, (1992).

364: 

365: \bibitem{stratt} Y. Wan and R. M. Stratt, {\it J. Chem. Phys.} {\bf 100},

366: 5123, (1994).

367: 

368: \bibitem{mona} G. Biroli and R. Monasson, {\it J. Phys. A: Math. Gen.}

369: {\bf 32}, L255 (1999).

370: 

371: \bibitem{cava} A. Cavagna, I. Giardina and G. Parisi,

372: {\it Phys. Rev. Lett.} {\bf 83}, 108 (1999).

373: 

374: \bibitem{kob} W. Kob and H. C. Andersen, {\it Phys. Rev. E} {\bf 51}, 4626 (1995); K. Vollmayr, W. Kob and K. Binder {\it J. Chem. Phys.} {\bf 105}, 4714 (1996).

375: 

376: \bibitem{sastryNature} S. Sastry, P. G. Debenedetti and

377: F. H. Stillinger, {\it Nature} {\bf 393}, 554 (1998).

378: 

379: \bibitem{fn1} Note that we have not followed the standard practing in

380: the INM literature of plotting the DOS against $\omega \equiv sign(\lambda)\times \sqrt{|\lambda|}$.

381: 

382: \bibitem{guhr} T. Guhr, A. Muller-Groeling and H. A. Weidenmuller,

383: Physics Reports, {\bf 299}, Number 4-6, (1998) and references therein.

384: 

385: \bibitem{bz} E. Br\'ezin and A. Zee, Nucl. Phys. B {\bf 402}, 613 (1993),

386: Phys. Rev. E {\bf 49}, 2588 (1994).

387: 

388: \bibitem{mehta} M. L. Mehta (1991) Random Matrices, Academic Press, San Diego.

389: 

390: \bibitem{bouch} P. Cizeau and J. P. Bouchaud, {\it Phys. Rev. E} {\bf 50},

391: 1810, (1994).

392: 

393: \bibitem{bray} A. J. Bray and G. J. Rodgers, {\it Phys. Rev. B} {\bf

394: 38}, 11461, (1988); G. J. Rodgers and A. J. Bray, {\it Phys. Rev. B}

395: {\bf 37}, 3557 (1988).

396: 

397: \bibitem{dedom} G. J. Rodgers and C. De Dominicis, {\it J. Phys. A}

398: Math. Gen. {\bf 23}, 1567 (1990).

399: 

400: \bibitem{mirlin} A. D. Mirlin and Y. V. Fyodorov, J. Phys. A: Math.

401: Gen. {\bf 24},2273 (1991); Y. V. Fyodorov and A. D. Mirlin, {\it

402: Phys. Rev. Lett.} {\bf 67}, 2049 (1991).

403: 

404: \bibitem{mezard} M. Mezard, G. Parisi and A. Zee,

405: http://arXiv.org/abs/cond-mat/9906135.

406: 

407: \bibitem{carpena} P. Carpena and P. Bernaola-Galva\'n {\it

408: Phys. Rev. B} {\bf 60}, 201 (1999).

409: 

410: \end{thebibliography}

411: 

412: \end{document}

413:                                                                     

414: 

415: 

416: 

417: 

418: