1: \documentclass[aps,pre,amsmath,amssymb,twocolumn,nofootinbib]{revtex4}

2: %\documentclass[preprint,aps,pre,amsmath,amssymb,nofootinbib,endfloats]{revtex4}

3: \usepackage{pslatex}

4: \usepackage{graphicx}

5: \usepackage{dcolumn}

6: \usepackage{bm}

7: \usepackage{natbib}

8: %\usepackage{showkeys}

9: 

10: \begin{document}

11: \title{The Cole-Cole Law for Critical Dynamics in Glass-Forming Liquids}

12: 

13: \author{M. Sperl}

14: \affiliation{Duke University, Department of Physics, Box 90305,

15: Durham, NC 27708, USA}

16: 

17: \date{\today}

18: \begin{abstract}

19: 

20: Within the mode-coupling theory (MCT) for glassy dynamics, the asymptotic 

21: low-frequency expansions for the dynamical susceptibilities at critical 

22: points are compared to the expansions for the dynamic moduli; this shows 

23: that the convergence properties of the two expansions can be quite 

24: different. In some parameter regions, the leading-order expansion formula 

25: for the modulus describes the solutions of the MCT equations of motion 

26: outside the transient regime successfully; at the same time, the leading- 

27: and next-to-leading order expansion formulas for the susceptibility fail. 

28: In these cases, one can derive a Cole-Cole law for the susceptibilities; 

29: and this law accounts for the dynamics for frequencies below the band of 

30: microscopic excitations and above the high-frequency part of the 

31: $\alpha$-peak. It is shown that this scenario explains the 

32: optical-Kerr-effect data measured for salol and benzophenone (BZP). For 

33: BZP it is inferred that the depolarized light-scattering spectra exhibit a 

34: wing for the $\alpha$-peak within the Gigahertz band. This wing results 

35: from the crossover of the von~Schweidler-law part of the $\alpha$-peak to 

36: the high-frequency part of the Cole-Cole peak; and this crossover can be 

37: described quantitatively by the leading-order formulas of MCT for the 

38: modulus.

39: 

40: \end{abstract}

41: 

42: \pacs{61.20.Lc, 82.70.Dd, 64.70.Pf}

43: 

44: \maketitle

45: 

46: %%%

47: \section{Introduction\label{sec:intro}}

48: 

49: During the past 15~years, several new spectrometers have been introduced 

50: for the study of the glassy dynamics of liquids. The evolution of this 

51: complex slow dynamics upon decreasing the temperature $T$ or increasing 

52: the density $\rho$ has been documented for many systems for times $t$, 

53: which exceed the natural time scale $t_\text{mic}$ for condensed-matter 

54: motions by three or more orders of magnitude. Similar progress has been 

55: made for molecular-dynamics simulations of liquid models. In parallel to 

56: these experimental activities, a theory for the evolution of glassy 

57: dynamics has been developed which is referred to as mode-coupling theory 

58: (MCT). This theory is based on regular equations of motion for a set of 

59: auto-correlation functions. The MCT equations lead to fold bifurcations 

60: for the correlators' long-time limits if some control parameter like $T$ 

61: reaches a critical value $T_c$; this bifurcation describes a transition 

62: from a liquid to an amorphous solid. The distance of the control parameter 

63: from the critical value, say $\varepsilon = (T_c-T)/T_c$, can be used for 

64: the discussion of the bifurcation dynamics as a small parameter. For 

65: $\varepsilon$ tending to zero and times increasing to infinity, it is 

66: possible to calculate asymptotic solutions of the MCT equations. The 

67: leading-order results provide a set of general formulas, which explain the 

68: qualitative features of the bifurcation scenario. Many fits of data with 

69: these general formulas have been studied in order to test the relevance of 

70: the MCT for the explanation of the experimental facts \cite{Goetze1999}.

71: 

72: A set of general MCT results, which is of main interest in this paper, 

73: concerns the critical dynamics. This dynamics is described by 

74: auto-correlation functions $\phi(t)$ for control parameters at the 

75: bifurcation point, say $T=T_c$. Equivalently, one can consider the 

76: corresponding loss spectra $\chi''(\omega)$. These are the products of the 

77: frequency $\omega$ and the Fourier-cosine transform of $\phi(t)$. The 

78: central asymptotic formula is specified by (1) a positive number $f^c$, 

79: which is called the plateau, (2) a positive amplitude, say $A$, and (3) 

80: the critical exponent $a$, obeying $0<a<0.396$, $\lim_{t\rightarrow\infty} 

81: t^a[\phi(t)-f^c] = A$. This formula is equivalent to 

82: $\lim_{\omega\rightarrow 0} \chi''(\omega)/\omega^a = \sin(\pi 

83: a/2)\Gamma(1-a)A$, with $\Gamma$ denoting the gamma function. For a given 

84: transition point, all correlators are specified by the same exponent $a$, 

85: but different critical points can differ in their value for $a$. The 

86: leading-order long-time result for the correlator describes a power-law 

87: decay: $\phi(t)-f^c \propto 1/t^a$. Equivalently, the leading-order result 

88: for the low-frequency critical loss spectrum is given by a power law 

89: variation $\chi''(\omega) \propto \omega^a$. For states near the 

90: transition point, $\phi(t)$ and $\chi''(\omega)$ can be replaced by their 

91: respective critical functions for short times, $t \ll t_\sigma$, or large 

92: frequencies, $\omega t_\sigma \gg 1$. The time scale $t_\sigma$ is the 

93: same for all correlators and diverges for states approaching the 

94: transition point. For $t \geqslant t_\sigma$ and $\omega t_\sigma 

95: \leqslant 1$, the correlators and spectra depend sensitively on 

96: $\varepsilon$; for shorter times, $t \ll t_\sigma$ and $\omega t_\sigma 

97: \gg 1$, the correlators and spectra depend on $\varepsilon$ smoothly. An 

98: $\omega^a$ spectrum was identified first for the glass-forming molten salt 

99: $0.4 Ca(NO_3)_2 0.6 K(NO_3)$ (CKN) in data obtained by neutron-scattering 

100: spectroscopy \cite{Knaak1988}. This system was also used to document for 

101: the first time the evolution of glassy dynamics within the full Giga-Hertz 

102: band \cite{Li1992}: Using depolarized light-scattering spectroscopy, a 

103: spectrum compatible with the $\omega^a$ law was found extending from 1GHz 

104: to 400GHz. The $t^{-a}$ decay in the time domain was measured first for 

105: density correlators by photon-correlation spectroscopy for a colloidal 

106: suspension of hard spheres \cite{Megen1993b} with the density as a control 

107: parameter.

108: 

109: Other glass-forming systems that can be studied experimentally are found 

110: in many van-der~Waals liquids. The natural time scale for inter-molecular 

111: vibrations is a picosecond. For normal liquid behavior, one expects 

112: correlations to decay to zero for times around some picoseconds. The 

113: normal-liquid excitation spectra extend from, say, 0.5THz to, say, 5THz. 

114: Glassy-dynamics spectra for several systems have been measured by 

115: depolarized light-scattering spectroscopy, and the data were shown to be 

116: consistent with the MCT bifurcation scenario. For example, spectra for 

117: toluene have been fitted successfully with leading-order asymptotic 

118: results for frequencies between 0.5GHz and 1THz and for temperatures 

119: decreasing from $T=T_c+140$K to $T_c+10$K; $T_c \approx 150$K 

120: \cite{Wiedersich2000}. For all these systems, the weight of the 

121: glassy-dynamics part of the spectra is large compared to the part of 

122: normal-liquid dynamics, i.e., the so-called $\alpha$-peak is large. In 

123: agreement with the MCT prediction for such situations, the amplitude $A$ 

124: for the $\omega^a$ spectrum is small, and the low-frequency contributions 

125: of the normal-liquid dynamics affect the region of the expected $\omega^a$ 

126: behavior of the spectra. As a result, no frequency interval has ever been 

127: identified for a van-der-Waals liquid, where an $\omega^a$ spectrum can be 

128: identified explicitly.

129: 

130: Torre \textit{et. al} \cite{Torre1998} have introduced optical-Kerr-effect 

131: (OKE) spectroscopy as a technique for the study of glassy dynamics. The 

132: measurement provides the response function $\chi(t) \propto 

133: -\partial_t\phi(t)$ for the same probing variable, which is studied in 

134: depolarized light-scattering experiments. The Fourier-sine transforms of 

135: $\chi(t)$ are proportional to the loss spectra $\chi''(\omega)$ mentioned 

136: in the preceding paragraph. The evolution of the glassy dynamics of 

137: m-toluidine was measured for temperatures decreasing from 295K to 250K. 

138: The response functions could be fitted well by the scaling-law results 

139: predicted by the leading-order asymptotic formulas for the MCT bifurcation 

140: \cite{Torre2000}. The analysis implies a critical temperature $T_c$ near 

141: 220K and a critical exponent $a$ near 0.3. However, lowering the 

142: temperature to 225K, the critical power-law decay $\chi(t) \propto 

143: 1/t^{1+a}$ was not observed \cite{Ricci2002}. The negative slope of the 

144: measured $\log\chi$-versus-$\log t$ curve is not $1+a$, rather it is a 

145: number smaller than unity, say $1-b'$. Decay laws $\chi(t) \propto 

146: 1/t^{1-b'}$ with exponents $b'$ around 0.2 have been identified by Cang 

147: \textit{et. al} \cite{Cang2003c} for a number of other van-der-Waals 

148: liquids. The OKE response of salol was studied by Hinze \textit{et. al}, 

149: and the data were shown to be consistent with the known MCT scaling-law 

150: formulas for the temperature decreasing from 340K to 266K 

151: \cite{Hinze2000b}. The glassy response for $T = 257$K was measured with an 

152: impressive accuracy for times increasing up to 0.5~$\mu$s; and the 

153: dynamics for $t >$ 100ps displays the behavior expected from the 

154: leading-order asymptotic results of MCT. However, the glassy dynamics for 

155: 2ps $<t<$ 20ps manifests itself by a $\chi(t) \propto 1/t$ decay. The 

156: corresponding correlator shows a logarithmic time dependence. All OKE 

157: response functions measured so far demonstrate a glassy dynamics that 

158: cannot be described by the general leading-order asymptotic formulas for 

159: the MCT bifurcation in the regime $t \lesssim 30$ps. In a recent 

160: alternative approach \cite{Berthier2005} is was possible to fit some of 

161: the OKE data for $t > 30$ps.

162: 

163: It was argued recently that the new facets of glassy dynamics discovered 

164: by OKE spectroscopy \cite{Ricci2002,Cang2003c,Hinze2000b} can be 

165: understood as generic implications of the Cole-Cole law for the critical 

166: dynamics \cite{Goetze2004}. In the following, this statement shall be 

167: explained in detail. Section~\ref{sec:general_MCT} summarizes the 

168: equations of motion and the known scaling-law results of MCT. It is 

169: explained in Sec.~\ref{sec:crit} that an asymptotic expansion of the 

170: modulus can have a much larger range of validity than the expansion for 

171: the susceptibility; the Cole-Cole law -- introduced in 1941 as empirical 

172: law \cite{Cole1941} -- is derived in full generality from the microscopic 

173: equations of motion. The relevance of these results is demonstrated for a 

174: schematic model in Sec.~\ref{sec:appl} using parameter values that 

175: describe the mentioned OKE data. The interplay between the Cole-Cole peak 

176: and the $\alpha$-peak is investigated in Sec.\ref{sec:alphabeta} where for 

177: specific parameter values the $\alpha$-peak displays a wing. 

178: Section~\ref{sec:sum} presents a conclusion.

179: 

180: %%% 

181: \section{Essential MCT Formulas\label{sec:general_MCT}}

182: \subsection{Equations of Motion\label{subsec:basic_eqs}}

183: 

184: Within the basic version of MCT, the dynamics of the system is described 

185: by $M$ correlators $\phi_q(t)$, $q=1,\dots, M$. These are real and even 

186: functions of the time $t$, which obey the initial conditions $\phi_q(t=0) 

187: = 1$, $\partial_t\phi_q(t=0)=0$. The corresponding set of normalized 

188: response functions is given by

189: 

190: \begin{subequations}\label{eq:susc}

191: \begin{equation}\label{eq:susc:der}

192: \chi_q(t) = -\partial_t \phi_q(t)\,.

193: \end{equation}

194: Laplace transforms map functions from the time domain, say $F(t)$, in the 

195: frequency domain. They shall be used with the convention 

196: $\mathrm{LT}[F(t)](z) = i \int_0^\infty\,dt\,\exp[izt] F(t)$, $z = \omega 

197: + i 0$. One gets $\mathrm{LT}[\chi_q(t)](z) = [1+\omega \phi_q(\omega)]$, 

198: where $\phi_q(\omega) = \mathrm{LT}[\phi_q(t)](z)$. The normalized 

199: dynamical susceptibilities are given by

200: 

201: \begin{equation}\label{eq:susc:FDT}

202: \chi_q(\omega) = 1+\omega\phi_q(\omega)\,.

203: \end{equation}

204: \end{subequations}

205: The loss spectra $\chi''_q(\omega) = \mathrm{Im}\, \chi_q(\omega)$ are 

206: related trivially to the fluctuation spectra $\phi_q''(\omega) = 

207: \mathrm{Im}\, \phi_q(\omega)$: $\chi_q''(\omega) = \omega 

208: \phi_q''(\omega)$.

209: 

210: The Zwanzig-Mori formalism provides a fraction representation of 

211: $\phi_q(\omega)$ in terms of a fluctuating-force correlator $M_q(\omega)$: 

212: $\phi_q(\omega) = -1/\{ \omega - \Omega_q^2 / [\omega + M_q(\omega)] \}$. 

213: The positive frequency $\Omega_q$ quantifies the initial decay of the 

214: correlator $\phi_q(t) = 1 - [\Omega_q t]^2/2 + {\cal O}(t^3)$. Within MCT, 

215: a white-noise term $\nu_q$ is split off from the kernel $M_q(\omega)$; the 

216: remainder is represented in terms of a dimensionless function $m_q(t)$: 

217: $M_q(\omega) = i\nu_q + \Omega_q^2 m_q(\omega)$; $\nu_q\geqslant 0$. Here, 

218: $m_q(t)$ and $m_q(\omega)$ are related by Laplace-transformation. The 

219: fraction representation is equivalent to the equations of motion

220: 

221: \begin{subequations}\label{eq:EOM}

222: \begin{equation}\label{eq:EOM:int}

223: \begin{split}

224: \partial_t^2 \phi_q(t) +& \nu_q\partial_t\phi_q(t)\hfill \\

225: +& \Omega_q^2 \left[

226: \phi_q(t)+\int_0^t\,dt'\,m_q(t-t')\partial_{t'}\phi_q(t')

227: \right] = 0\,.

228: \end{split}

229: \end{equation}

230: The essential approximation in MCT is the expression of $m_q(t)$ as a 

231: polynomial 

232: ${\cal F}_q$ of the correlators: 

233: \begin{equation}\label{eq:EOM:kernel}

234: m_q(t) = {\cal F}_q[\phi_1(t), \dots, \phi_M(t)]\,.

235: \end{equation}

236: \end{subequations}

237: There is no monomial contribution of order zero. The coefficients of the 

238: polynomial are called mode-coupling coefficients. They are the coupling 

239: constants of the theory and must not be negative. Within the microscopic 

240: theory, the polynomials are of second order and the coefficients are given 

241: by the equilibrium structure functions, which in turn are smooth functions 

242: of control parameters like the temperature T for the states considered.

243: 

244: At the generic transition mentioned in the preceding section, the 

245: correlator's long-time limits depend singularly on the distance parameter 

246: $\varepsilon$. For $\varepsilon < 0$, fluctuations disappear for long 

247: times, $\phi_q(t\rightarrow\infty) = 0$. For $\varepsilon \geqslant 0$, 

248: $\phi_q(t \rightarrow \infty) = f_q$, $0<f_q<1$. $q = 1,\dots, M$; the 

249: fluctuations arrest. Within the microscopic version of MCT, the arrested 

250: part $f_q$ has the meaning of the Debye-Waller factor of the solid 

251: amorphous state. The leading-order variation with changes of $\varepsilon$ 

252: for the arrested part is given by $f_q - f^c_q \propto 

253: \sqrt{\varepsilon}$, $\varepsilon \rightarrow 0^+$, $f_q^c > 0$, 

254: $q=1,\dots, M$. The regularity of the MCT equations implies the following. 

255: For $\varepsilon$ tending to arbitrarily small values, there appears an 

256: arbitrarily large time interval, where $\phi_q(t)$ is arbitrarily close to 

257: $f^c_q$. This critical arrested part $f^c_q$ has the meaning of a plateau 

258: for the $\phi_q(t)$-versus-$\log t$ curves for states near the transition. 

259: The corresponding plateau for the force correlators is $m_q(t \rightarrow 

260: \infty) = f^{m\,c}_q = {\cal F}^c_q[f^c_1, \dots, f^c_M]$. At the 

261: transition, $\phi_q(\omega)$ and $m_q(\omega)$ exhibit poles 

262: $-f^c_q/\omega$ and $-f^{m\,c}_q/\omega$, respectively. Hence, there is a 

263: region of small $|\varepsilon|$ and small $\omega$, where $\omega+i\nu_q$ 

264: can be neglected compared to $\Omega_q^2 m_q(\omega)$. This region is the 

265: one for the MCT glassy dynamics. The fraction representation of the 

266: correlators simplifies to $\phi_q(\omega) = -1/[\omega - 1/m_q(\omega)]$. 

267: This formula can also be noted as

268: 

269: \begin{subequations}\label{eq:MCTlong} 

270: \begin{equation}\label{eq:MCTlong:m} 

271: \omega m_q(\omega) = \omega \phi_q(\omega)/\left[ 1 + \omega\phi_q(\omega) 

272: \right]\,. 

273: \end{equation} 

274: Equation~(\ref{eq:susc:FDT}) yields the equivalent expression for the 

275: dynamical susceptibility:

276: 

277: \begin{equation}\label{eq:MCTlong:chi} 

278: \chi_q(\omega) = 1/\left[ 1-\omega m_q(\omega)  \right]\,. 

279: \end{equation} \end{subequations} 

280: Within the regime of glassy dynamics, $\left[1-\omega m_q(\omega)\right]$ 

281: has the meaning of a modulus for the response described by $\chi_q(t)$. 

282: The pair of Eqs.~(\ref{eq:EOM:kernel}, \ref{eq:MCTlong:m}) can fix the 

283: solution only up to some overall time scale $t_0$. The latter is 

284: determined by matching of the transient dynamics with the glassy dynamics. 

285: For further details and for a list of original papers, the reader can 

286: consult Ref.~\cite{Franosch1997}.

287: 

288: 

289: %%

290: \subsection{Scaling Laws\label{subsec:scaling_laws}}

291: 

292: There is a straight-forward recipe to calculate from the coupling 

293: coefficients at the transition point and from the critical arrested parts 

294: $f^c_q$ a number $\lambda$, $1/2 \leqslant \lambda < 1$. It is called the 

295: exponent parameter for the chosen transition point of the model under 

296: discussion. It fixes the critical exponent $a$, mentioned in 

297: Sec.~\ref{sec:intro}. It fixes a further exponent $b$, $0<b\leqslant 1$, 

298: which is called the von~Schweidler exponent. The equation for the two 

299: exponents reads $\Gamma(1-a)^2/\Gamma(1-2a) = \lambda = 

300: \Gamma(1+b)^2/\Gamma(1+2b)$. Parameter $\lambda$ also specifies a pair of 

301: equations for a pair of functions $g_\pm(\hat{t})$, which are defined for 

302: $\hat{t}>0$. The equations read $\pm 1 + \lambda g_\pm(\hat{t})^2 = 

303: (d/d\hat{t}) \int_0^{\hat{t}}\,d\hat{t}\, g_\pm(\hat{t}-\hat{t}') 

304: g_\pm(\hat{t}')$, and they have to be solved with the initial condition 

305: $\lim_{\hat{t}\rightarrow 0} \hat{t}^a g_\pm(\hat{t}) = 1$. Up to 

306: corrections of order $\hat{t}^a$, one gets for small $\hat{t}$:

307: 

308: \begin{subequations}\label{eq:gpm}

309: \begin{equation}\label{eq:gpm:ta}

310: g_\pm(\hat{t}\ll 1) = 1/\hat{t}^a\,.

311: \end{equation}

312: The function $g_+(\hat{t})$ approaches its long-time limit exponentially, 

313: $g_+(\hat{t}\gg 1) = 1/\sqrt{1-\lambda}$.

314: The function $g_-(\hat{t})$ exhibits a power-law divergence for large 

315: $\hat{t}$. Up to corrections of order $1/\hat{t}^b$, one gets

316: 

317: \begin{equation}\label{eq:gpm:B}

318: g_-(\hat{t}\gg 1) = -B\hat{t}^b\,.

319: \end{equation}

320: \end{subequations}

321: There are tables allowing the determination of $a$, $b$ and $B$ from a 

322: given $\lambda$. There are also tables to determine $g_\pm(\hat{t})$ with 

323: an accuracy sufficient for all practical purposes \cite{Goetze1990}.

324: 

325: The functions $g_\pm(\hat{t})$ are the shape functions for the first 

326: scaling law of MCT. For $\varepsilon$ tending to zero, there appears a 

327: time interval of diverging length, within which $\Delta_q(t) = |\phi_q(t) 

328: - f^c_q|$ is arbitrary small. The leading-order solution for the small 

329: parameter $\Delta_q(t)$ yields the first scaling-law. The solution 

330: assumes the form

331: 

332: \begin{subequations}\label{eq:phi_}

333: \begin{equation}\label{eq:phi_gpm}

334: \phi_q(t) = f^c_q + h_q \sqrt{|\sigma|} g_\pm(t/t_\sigma)\,,\quad 

335: \varepsilon \gtrless 0\,.

336: \end{equation}

337: The amplitudes $h_q > 0$, $q = 1, \dots, M$ are calculated from the 

338: mode-coupling coefficients at the critical point. The separation parameter 

339: $\sigma$ is defined similarly: it is a smooth function of the control 

340: parameters and can be linearized close to the transition point, $\sigma = 

341: C\varepsilon + {\cal O}(\varepsilon^2)$, with a positive coefficient $C$ 

342: that depends on the chosen control parameter. The first critical time 

343: scale $t_\sigma$ is given by the critical exponent $a$, the separation 

344: parameter $\sigma$, and the time scale $t_0$, which is defined by the 

345: short-time dynamics:

346: 

347: \begin{equation}\label{eq:sigma:tsigma}

348: t_\sigma = t_0/|\sigma|^\delta\,,\quad \delta = 1/(2a)\,.

349: \end{equation} \end{subequations}

350: The strong control-parameter dependence of the correlators near the 

351: plateau is described solely by that of the correlation scale 

352: $\sqrt{|\sigma|}$ and of the time scale $t_\sigma$.

353: 

354: From Eqs.~(\ref{eq:gpm:ta},\ref{eq:phi_gpm},\ref{eq:sigma:tsigma}), one 

355: gets the leading-order asymptotic law for the decay of the critical 

356: correlator discussed in Sec.~\ref{sec:intro}: $\phi_q(t) - f^c_q = h_q 

357: (t_0/t)^a$. From Eqs.~(\ref{eq:gpm:B}, \ref{eq:phi_gpm}, 

358: \ref{eq:sigma:tsigma}) one gets the von~Schweidler law for the decay of 

359: the liquid correlators below the plateau:

360: 

361: \begin{equation}\label{eq:vS:phi}

362: \phi_q(t) = f^c_q + h_q (t/t_{\sigma'})^b\,,\quad t_\sigma \ll t \ll 

363: t_{\sigma'}\,,\quad \varepsilon < 0 \,.

364: \end{equation}

365: Here, the second critical time scale of MCT reads $t_{\sigma'} = t_0/[ 

366: B^{1/b} |\sigma|^\gamma]$, with $\gamma = 1/(2a) + 1/(2b) \,. $ Up to 

367: errors of order $|\varepsilon|$, the decay of the liquid correlators below 

368: the plateau is described by the second scaling law of MCT

369: 

370: \begin{equation}\label{eq:alpha}

371: \phi_q(t) = \tilde{\phi}_q(t/t'_\sigma) \,,\quad

372: t'_\sigma \ll t\,,\quad \varepsilon <0\,.

373: \end{equation}

374: Here, $\tilde{\phi}_q(\tilde{t})$ is an $\varepsilon$-independent shape 

375: function. Its initial part is given by von~Schweidler's law: 

376: $\tilde{\phi}_q(\tilde{t}\ll 1) = f^c_q - h_q \tilde{t}^b$. Corrections of 

377: order $\sqrt{|\varepsilon|}$ modify the formula (\ref{eq:alpha}) for the 

378: below-plateau decay in a regime where $\phi_q(t)$ is close to the plateau. 

379: These corrections are described by Eq.~(\ref{eq:phi_gpm}) for $t \geqslant 

380: t_\sigma$. The below-plateau decay is referred to as $\alpha$-process, and 

381: Eq.~(\ref{eq:alpha}) formulates the superposition principle. Details of 

382: the derivation of the cited results, references to the original work, and 

383: a comprehensive demonstration for the hard-sphere system can be found in 

384: Refs.~\cite{Franosch1997,Fuchs1998}.

385: 

386: 

387: %%%

388: \section{Critical Dynamics and Cole-Cole law\label{sec:crit}}

389: 

390: In this section, correlators and susceptibilities shall be discussed for 

391: states at the transition point, say $T=T_c$. Focusing on the range of 

392: validity, the essential formulas are analyzed for the correlators in 

393: \ref{sec:crit:corr}, and for the susceptibilities in \ref{sec:crit:susc}. 

394: The Cole-Cole law is derived in \ref{sec:crit:cole}.

395: 

396: %%

397: \subsection{Power-Law Solution for Correlation 

398: Functions\label{sec:crit:corr}}

399: 

400: The leading-order scaling-law formulas~(\ref{eq:gpm:ta},\ref{eq:phi_gpm}) 

401: yield for the critical correlators

402: 

403: \begin{equation}\label{eq:crit:corr}

404: \phi_q(t) = f^c_q + h_q (t_0/t)^a\,.

405: \end{equation}

406: A first question, whose answer is not implied by the results cited in 

407: Sec.~\ref{subsec:scaling_laws}, concerns the range of validity of 

408: Eq.~(\ref{eq:crit:corr}) for short times. Let us define an onset time 

409: $t^*_q$ for the power law by the request, that Eq.~(\ref{eq:crit:corr}) 

410: describes $\phi_q(t)- f^c_q$ within a relative error of, say, 10\%, for $t 

411: \geqslant t^*_q$. For normal-liquid dynamics, one would expect the 

412: correlators to decay to zero for times around $t_\text{mic}$. A second 

413: question to be discussed is: How can one describe the critical correlators 

414: in cases where $t^*_q$ exceeds $t_\text{mic}$, i.e., when there is a gap 

415: between the end of the transient regime and the onset of the critical 

416: power law?

417: 

418: The asymptotic solution~(\ref{eq:crit:corr}) can be extended to an 

419: asymptotic series expansion in powers of $t^{-a}$. The result up to the 

420: next-to-leading term shall be noted as

421: 

422: \begin{subequations}\label{eq:critcorr}

423: \begin{equation}\label{eq:critcorr:phi}

424: \phi_q(t) = f^c_q + h_q (t_0/t)^a \left[ 1+ \hat{K}_q (t_0/t)^a \right]\,.

425: \end{equation}

426: Here, remainders which are given by $(t_0/t)^{3a}$ times some power of 

427: $\ln(t/t_0)$ are dropped. The Tauberian theorem yields an equivalent 

428: formula in the frequency domain: 

429: 

430: \begin{equation}\label{eq:critcorr:chi}

431: \begin{split}

432: \omega\phi_q(\omega) = -f^c_q - &\Gamma(1-a) h_q (-i\omega t_0)^a\\

433:  &- \Gamma(1-2 a) h_q \hat{K}_q (-i\omega t_0)^{2a}\,. 

434: \end{split}\end{equation} 

435: \end{subequations} 

436: In this formula, terms are dropped, which are proportional to 

437: $\omega^{3a}$ times some power of $\ln\omega$. The preceding two formulas 

438: are the starting point for the following derivations. It is a 

439: straight-forward procedure to calculate the correction amplitude 

440: $\hat{K}_q$ from the mode-coupling coefficients at the critical point 

441: $T=T_c$ \cite{Franosch1997}. From Eq.~(\ref{eq:critcorr:phi}), one can 

442: estimate an onset time of $t^*_q/t_0 = [10|\hat{K}_q|]^\frac{1}{a}$. This 

443: is an estimate based on the assumption that higher-order expansion terms 

444: in Eq.~(\ref{eq:critcorr:phi}) do not influence the results seriously for 

445: $t \geqslant t^*_q$. Typically for many systems, the critical exponent $a$ 

446: is around 0.3, and $1/a \gtrsim 3$. Hence, $t^*_q$ depends sensitively on 

447: the correction amplitude $\hat{K}_q$. As a result, $t^*_q$ can vary 

448: considerably for different $q$. As a relevant example, let us cite the 

449: results for the density-fluctuation correlators of a system of hard 

450: spheres of diameter $d$ \cite{Franosch1997}: The time $t_1^*$ referring to 

451: a wave-number $q_1d = 7.0$ exceeds the time $t_2^*$ for $q_2d = 10.6$ by a 

452: factor of around 100.

453: 

454: %%

455: \subsection{Power-Law Solution for the 

456: Susceptibilities\label{sec:crit:susc}}

457: 

458: From Eqs.~(\ref{eq:susc:der}, \ref{eq:crit:corr}), one obtains the 

459: long-time result for the response functions up to leading-order 

460: corrections,

461: \begin{equation}\label{eq:chit0}

462: \chi(t) t_0 = a h_q (t_0/t)^{1+a}\,.

463: \end{equation}

464: According to Eq.~(\ref{eq:critcorr:phi}), the leading order result in 

465: Eq.~(\ref{eq:crit:corr}) is valid for $t \geqslant t_q^*$, but the onset 

466: time for Eq.~(\ref{eq:chit0}) is later, $t \geqslant t^*_q 2^{1/a}$, where 

467: typically $2^{1/a} \approx 10$. Hence, the detection of the critical 

468: dynamics in its leading asymptotic form is more difficult for the response 

469: functions than for the correlators.

470: 

471: From Eq.~(\ref{eq:critcorr:chi}), one arrives at the expansion formula for 

472: the absorptive part of the dynamical susceptibility 

473: 

474: \begin{subequations}\label{eq:chi_power}

475: \begin{equation}\label{eq:chi_power:chi}\begin{split}

476: \chi_q''(\omega) = \left[ \Gamma(1-a) \sin\left( \pi a/2 \right) \right] 

477: &h_q (\omega t_0)^a \\&\times

478: \left[ 1+ k_a \hat{K}_q (\omega t_0)^a \right]\,,

479: \end{split}\end{equation}

480: 

481: \begin{equation}\label{eq:chi_power:ka}

482: k_a =  2 \Gamma(1-a) \cos\left( \pi a/2 \right)/\lambda\,.

483: \end{equation}

484: The leading-order power-law result reads

485: 

486: \begin{equation}\label{eq:chi_power:leading}

487: \chi_q''(\omega) = \left[ \Gamma(1-a) \sin\left( \pi a/2 \right) \right] 

488: h_q (\omega t_0)^a\,,

489: \end{equation}

490: \end{subequations}

491: which has an onset frequency of $\omega_q^* = 1/(t^*_q k_a^{1/a})$; for 

492: $\omega \leqslant \omega_q^*$, the leading-order result in 

493: Eq.~(\ref{eq:chi_power:leading}) describes the critical spectrum with an 

494: error smaller than 10\%. If $\lambda$ decreases from 1 to $1/2$, $k_a$ 

495: increases from 2 to near 5. For $\lambda$ near a typical value of 0.7, 

496: $k_a$ is above 3, and the onset frequency for the $\omega^a$-law is about 

497: 30 times smaller than $1/t^*_q$. Therefore, the detection of the critical 

498: power-law in the loss spectra is even more difficult then in the response 

499: function.

500: 

501: Including the leading-correction terms in the asymptotic formulas, as 

502: noted in Eqs.~(\ref{eq:critcorr:phi}, \ref{eq:chi_power:chi}), is an 

503: obvious manner to extend the range of applicability of the analytic 

504: description of the dynamics. This is demonstrated comprehensively for the 

505: MCT for the hard-sphere system in Refs.~\cite{Franosch1997, Fuchs1998}. 

506: But there are cases, where this procedure does not lead to satisfactory 

507: results. The mean-squared displacement is an example, where the 

508: description of the increase towards the plateau by the analog of 

509: Eq.~(\ref{eq:critcorr:phi}) cannot account for the glassy dynamics 

510: \cite{Fuchs1998}.

511: 

512: %%

513: \subsection{The Cole-Cole Law\label{sec:crit:cole}}

514: 

515: Equation~(\ref{eq:critcorr:chi}) is an asymptotic expansion in terms of 

516: powers of the small quantity $\xi = (-i\omega t_0)^a$ that holds up to 

517: errors of $\xi^3$. Substitution of this expansion into 

518: Eq.~(\ref{eq:MCTlong:m}) provides an analogous expansion for $\omega 

519: m_q(\omega)$. Let us indicate the non-trivial parts of the coefficients by 

520: a superscript $m$:

521: 

522: \begin{equation}\label{eq:mqw}\begin{split}

523: -\omega m_q(\omega) = f_q^{m\,c} +& \Gamma(1-a) h^m_q (-i \omega t_0)^a \\

524: &+ \Gamma(1-2a) h^m_q \hat{K}^m_q (-i \omega t_0)^{2a}\,.

525: \end{split}\end{equation}

526: Comparing coefficients of equal powers of $\xi$, one finds:

527: 

528: \begin{subequations}\label{eq:cmq}

529: \begin{equation}\label{eq:fqmc}

530: f_q^{m\,c} = f_q^c/(1-f_q^c)\,,\quad f_q^c =  f_q^{m\,c} /(1 +  

531: f_q^{m\,c} ) \,.

532: \end{equation}

533: \begin{equation}\label{eq:hmq}

534: h_q^m = h_q /(1-f_q^c)^2\,,\quad h_q = h_q^m/(1+ f_q^{m\,c} )^2\,.

535: \end{equation}

536: \begin{equation}\label{eq:cmq:cmq}\begin{split}

537: \hat{K}^m_q = \hat{K}_q + \lambda [h_q/(1-f_q^c)]\,,\quad\\

538: \hat{K}_q = \hat{K}^m_q -\lambda [h^m_q / (1 +  f_q^{m\,c} )]\,.

539: \end{split}

540: \end{equation}

541: \end{subequations}

542: 

543: Since the auto-correlation functions are normalized, $\phi_q(t=0) = 1$, 

544: one gets $f_q^c < 1$. The kernel $m_q(t=0)$ can have any positive value. 

545: Therefore, in principle, $f_q^{m\,c}$ can be any positive number. The 

546: different normalizations can be eliminated in the critical amplitudes by 

547: comparing the ratios $h_q/f_q^c$ and $h^m_q/f_q^{m\,c}$:

548: 

549: \begin{equation}\label{eq:hqfq}

550: [h^m_q/f^{m\,c}_q] = [h_q/f_q^c]/(1-f^c_q)\,.

551: \end{equation}

552: These ratios determine the relative amplitude of the dynamics around the 

553: plateau within the first scaling-law regime. Equation (\ref{eq:phi_gpm}) 

554: yields $(\phi_q(t)-f^c_q)/f^c_q = [h_q/f_q^c] \sqrt{|\sigma|} 

555: g_\pm(t/t_\sigma)$, and a corresponding identity holds for $m_q(t)$. From 

556: Eq.~(\ref{eq:hqfq}) one concludes that the relative amplitude of the 

557: first-scaling-law contribution is larger by $1/(1-f_q^c) > 1$ for the 

558: modulus than for the susceptibility; and $1/(1-f_q^c)$ increases with 

559: $f^c_q$, the $\alpha$-peak contribution of the loss spectrum. 

560: Consequently, within the range of validity of the first scaling law, it is 

561: easier to measure the first-scaling-law contribution for the modulus than 

562: that for the susceptibility.

563: 

564: The expansion~(\ref{eq:mqw}) can be substituted into 

565: Eq.~(\ref{eq:MCTlong:chi}) in order to obtain an asymptotic expansion for 

566: the inverse of the critical susceptibility, i.e., for the modulus. The 

567: result shall be noted in the form

568: 

569: \begin{equation}\label{eq:ccc}

570: \chi_q(\omega) = \chi_{0\,q}^{cc} /\left[ 1 + \left( 

571: -i\omega/\omega_q^{c}  \right)^a + \hat{K}^{cc}_q  \left( 

572: -i\omega/\omega_q^{c}  \right)^{2a}\right] \,.

573: \end{equation}

574: The expression $\chi_q(\omega)^{-1}$ is correct up to errors of the order 

575: $\omega^{3a}$. The three parameters specifying this formula can be 

576: expressed in terms of the coefficients in Eq.~(\ref{eq:mqw}) and, via 

577: Eqs.~(\ref{eq:cmq}), in terms of the coefficients 

578: specifying the correlators in Eq.~(\ref{eq:critcorr:phi}). One gets 

579: $\chi_{0\,q}^{cc} = 1/(1+f_q^{m\,c})$. This amplitude is the complement of 

580: the $\alpha$-peak strength $f^c_q$ of the normalized loss spectrum,

581: 

582: \begin{subequations}\label{eq:cccpar}

583: \begin{equation}\label{eq:cccpar:chi0}

584: \chi_{0\,q}^{cc} = 1- f^c_q\,.

585: \end{equation}

586: The characteristic frequency entering the new formula for the 

587: susceptibility, is given by $(\omega_q^{c}t_0)^a = (1 + f_q^{m\,c}) / 

588: \left( h^m_q \Gamma(1-a) \right)$ or by

589: 

590: \begin{equation}\label{eq:cccpar:wcc}

591: \omega_q^{c}t_0 = \left[ (1-f^c_q) / \left(h_q \Gamma(1-a)\right)  

592: \right]^{1/a}

593: \,.

594: \end{equation}

595: The correction amplitude reads 

596: $\hat{K}_q^{cc} = \hat{K}^m_q (1+f^{m\,c}_q)/(\lambda h^m_q)$,

597: which is equivalent to 

598: \begin{equation}\label{eq:cccpar:cqcc}

599: \hat{K}_q^{cc} = 1 + \left[(1-f^c_q)/h_q\right] \hat{K}_q / \lambda\,. 

600: \end{equation}

601: \end{subequations}

602: 

603: If the frequencies are so small that $|\hat{K}_q^{cc} 

604: (-i\omega/\omega_q^{c})| \ll 1$, Eq.~(\ref{eq:ccc}) simplifies to the 

605: transparent expression of the Cole-Cole law,

606: 

607: \begin{equation}\label{eq:ccleft:chi}

608: \chi_q(\omega) = \chi_{0\,q}^{cc}/\left[ 1 + (-i\omega/\omega_q^{c})^a 

609: \right]\,.

610: \end{equation}

611: The condition of validity for this formula means that the critical modulus 

612: can be described by the simple power law in the first line of 

613: Eq.~(\ref{eq:mqw}). The modulus spectrum obeys a formula analog to 

614: Eq.~(\ref{eq:chi_power:leading}): $\omega m_q''(\omega) = \Gamma(1-a) 

615: \sin\left(\pi a/2\right) h^m_q (\omega t_0)^a$. Using the 10\% criterion 

616: from above, the last term in Eq.~(\ref{eq:mqw}) can be dropped for 

617: frequencies $\omega t_0 \leqslant \left[ \lambda/\left( 10 \Gamma(1-a) 

618: |\hat{K}^m_q| \right) \right]^{1/a}$. The correlator corresponding to the 

619: loss spectrum in Eq.~(\ref{eq:ccleft:chi}) is given by the Mittag-Leffler 

620: function of index $a$: $M_a(x) = \sum_{n=0}^\infty x^n/\Gamma(1+n a)$,

621: 

622: \begin{equation}\label{eq:ML}

623: \phi_q(t) = f_q^c + \chi_{0\,q}^{cc} M_a\left[-(t\omega_q^{c})^a\right] 

624: \,.

625: \end{equation}

626: The Mittag-Leffler function can be calculated efficiently by Fourier-back 

627: transformation of $\chi''_q(\omega) / \omega$.

628: 

629: Equation~(\ref{eq:ccleft:chi}) describes a very broad peak for the loss 

630: spectrum $\chi_q''(\omega)$ with a maximum located at $\omega = 

631: \omega_q^{c}$. The spectrum is invariant under the interchange of 

632: $(\omega/\omega_q^{c})$ to $(\omega_q^{c}/\omega)$: the 

633: $\chi_q''(\omega)$-versus-$\log\omega$ curve is symmetric. The width of 

634: the peak increases strongly with decreasing exponent $a$. For $a \approx 

635: 0.3$, $\omega$ has to increase by more than a factor of $10^4$ in order to 

636: scan the interval of frequency where $\chi_q''(\omega) \geqslant 

637: \chi_q''(\omega_q^{c}) / 2$. This susceptibility formula known as 

638: Cole-Cole law, Eq.~(\ref{eq:ccleft:chi}), was introduced as an empirical 

639: formula for dielectric loss spectra in glassy systems \cite{Cole1941}. In 

640: particular, broad peaks located above the $\alpha$ peaks -- often called 

641: $\beta$-peaks in this context -- have been fitted by it.

642: 

643: The Cole-Cole formula, Eq.~(\ref{eq:ccleft:chi}), exhibits simple limits 

644: for small and large frequencies. For $(\omega/\omega_q^{c})^a\ll 1$, the 

645: Cole-Cole susceptibility reproduces the general power-law for the loss 

646: spectrum, Eq.~(\ref{eq:chi_power:leading}), $\chi_q''(\omega) \propto 

647: \omega^a$. Similarly, the corresponding Mittag-Leffler correlator 

648: reproduces the general long-time asymptote for the response, 

649: Eq.~(\ref{eq:chit0}). For $(\omega/\omega_q^{c})^a \gg 1$, the Cole-Cole 

650: spectrum describes a critical spectrum which decreases with increasing 

651: frequency: $\chi_q''(\omega) \propto 1/\omega^a$. This is similar to a 

652: von~Schweidler-law spectrum. The corresponding Mittag-Leffler correlator 

653: decreases according to the law $\phi(t) -\text{const.} \propto

654: - t^a$. This yields a response function $\chi(t) \propto 1/t^x$ with $x = 

655: 1-a <1$. Such behavior is consistent with the one detected by the OKE 

656: results for van-der Waals liquids \cite{Cang2003c}.

657: 

658: The crossover of the critical spectrum from the low-frequency wing to the 

659: high-frequency wing of the Cole-Cole loss peak has a counterpart for the 

660: Mittag-Leffler correlator that is most easily seen in a semilogarithmic 

661: plot. For $(t\omega_q^{c})^a \ll 1$, the $\left[ \phi(t) - \text{const.} 

662: \right]$-versus-$\log t$ curve is bent downward, as known for the 

663: von~Schweidler curves. For $(t\omega_q^{c})^a \gg 1$, there is the 

664: critical power-law decay which shows up as an upward-bent curve. Hence, 

665: the $\phi(t)$-versus-$\log t$ curve exhibits an inflection point. For an 

666: exponent $a$ around 0.3, the curve is nearly straight for an increase of 

667: $\log_{10} t$ by a factor of about 3. In this case, there is 

668: nearly-logarithmic decay of the critical correlator for a time variation 

669: over three orders of magnitude.

670: 

671: It depends on the size of $1/\omega_q^{c}$ relative to the microscopic 

672: time scale $t_\text{mic}$ which part of the Mittag-Leffler correlator 

673: dominates the critical dynamics in the time range of interest. The crucial 

674: question concerns the range of validity of the leading-order result for 

675: the susceptibility and for the modulus. These ranges are determined by the 

676: correction amplitudes. If $|\hat{K}_q|$ is large and $|\hat{K}^m_q|$ 

677: small, the Cole-Cole formula has a larger range of applicability than the 

678: general simple power-law formula~(\ref{eq:chi_power:leading}). But, if 

679: $|\hat{K}^m_q|$ is larger than $|\hat{K}_q|$, 

680: Eq.~(\ref{eq:chi_power:leading}) is a better approximation for the 

681: critical decay than the Cole-Cole spectrum. Since the brackets in 

682: Eqs.~(\ref{eq:cmq}) are positive, there are two obvious results. If 

683: $\hat{K}_q$ is positive, there holds $\hat{K}^m_q > \hat{K}_q$. In this 

684: case, the leading-order result for the susceptibility is a better 

685: description of the critical spectrum than the Cole-Cole spectrum. If 

686: $\hat{K}^m_q$ is negative, there holds $|\hat{K}^m_q| < |\hat{K}_q|$ and 

687: the application of the Cole-Cole law is superior to the one of the general 

688: leading-order result for the loss spectrum. For negative correction 

689: amplitudes $\hat{K}_q$, there is a trend for cancellation of the two 

690: contributions to $\hat{K}_q^m$ in Eq.~(\ref{eq:cmq:cmq}). It is a generic 

691: situation, that $|\hat{K}^m_q|$ is smaller than $|\hat{K}_q|$. Hence, for 

692: $\hat{K}_q < 0$, it is expected that the Cole-Cole law is a good 

693: description of the critical dynamics.

694: 

695: A detailed discussion of the MCT equations has shown that a large arrested 

696: part $f^c_q$ implies a large negative correction amplitude $\hat{K}_q$ for 

697: the critical decay \cite{Franosch1997}. In this case, the onset time 

698: $t^*_q$ may exceed the microscopic time scale $t_\text{mic}$ by several 

699: orders of magnitude. The time range for the applicability of the 

700: leading-order asymptotic description of the critical decay may be outside 

701: the accessible range of the existing spectrometers. But, this is the 

702: situation, where $|\hat{K}^m_q|$ may be so small that the Cole-Cole law or 

703: the Mittag-Leffler correlator can provide an adequate description of the 

704: critical dynamics.

705: 

706: 

707: %%%

708: \section{Application to Schematic Models\label{sec:appl}}

709: 

710: The preceding results shall be demonstrated in a simple schematic model 

711: for two different cases where the model parameters are adjusted to 

712: reproduce the data measured for benzophenone (BZP) and salol 

713: \cite{Hinze2000b,Cang2003c,Goetze2004}

714: 

715: The simplest MCT models deal with a single correlation function only. For 

716: these models, the formulas in the preceding sections simplify since the 

717: label $q$ can be dropped. The correlator, the critical arrested part, and 

718: the critical amplitude shall be denoted by $\phi(t)$, $f^c$, and $h$, 

719: respectively. The model is specified by the frequency $\Omega$ and the 

720: friction coefficient $\nu$ in the equation of motion~(\ref{eq:EOM:int}). 

721: Furthermore, there are the non-negative coefficients $v_l,\, l \geqslant 

722: 1$, which specify the mode-coupling monomial of order $l$ for ${\cal F}$ 

723: in Eq.~(\ref{eq:EOM:kernel}). In applications for data descriptions, the 

724: specified numbers are considered as smooth functions of the physical 

725: control parameters.

726: 

727: The simplest model for the polynomial ${\cal F}$, which can reproduce all 

728: possible values for the exponent parameter $\lambda$, is given by the 

729: following formula for the fluctuating-force kernel

730: 

731: \begin{equation}\label{eq:mF12}

732: m(t) = v_1\phi(t) + v_2 \phi(t)^2\,.

733: \end{equation}

734: The two coupling constants $v_1$ and $v_2$ specify the state of the system 

735: by a point in the first quadrant of the $v_1$-$v_2$ plane, cf. lower inset 

736: in Figs.~\ref{fig:BZP} and \ref{fig:salol}. The points $(v_1^c, v_2^c)$ 

737: for generic fold bifurcations are located on a piece of a parabola. The 

738: position of the specific transition point can be characterized by the 

739: value for $\lambda$ \cite{Goetze1984}:

740: 

741: \begin{subequations}\label{eq:vc}

742: \begin{equation}\label{eq:vc:vc}

743: v_1^c = (2\lambda-1)/\lambda^2 \,,\; v_2^c = 1/\lambda^2 \,,\; f^c = 

744: 1-\lambda \,,\; h=\lambda\,.

745: \end{equation}

746: The separation of some point $(v_1, v_2)$ from the transition point 

747: $(v_1^c, v_2^c)$, see upper insets Figs.~\ref{fig:BZP} and 

748: \ref{fig:salol}, is given by

749: \begin{equation}\label{eq:vc:sigma}

750: \sigma = \left[ \hat{v}_1 + \hat{v}_2 (1-\lambda) \right] \lambda 

751: (1-\lambda) \,,\quad \hat{v}_{1,\,2} = v_{1,\,2} - v^c_{1,\,2}

752: \,.

753: \end{equation}

754: \end{subequations}

755: The arrested part of the correlator for the glass state reads $\phi(t 

756: \rightarrow \infty) = h[\sigma/(1-\lambda)]^{1/2}+{\cal O}(\sigma^{3/2})$. 

757: The splitting of this value in an amplitude $h$ and a remainder is not 

758: unique. The cited value for the critical amplitude $h$ follows the 

759: conventions made in the preceding literature \cite{Franosch1997}. The 

760: general expression for the correction amplitude $\hat{K}$ for $M=1$ models 

761: \cite{Goetze1989c} is specialized easily to $\hat{K} = \kappa(x)$, where 

762: the functions $\kappa(x)$ is defined by

763: 

764: \begin{equation}\label{eq:kappa}

765: \kappa (x) = \frac{1}{2} \Gamma(1-x)^3/ \left[

766: \lambda \Gamma(1-3x) - \Gamma(1-x) \Gamma(1-2x)

767: \right] \,.

768: \end{equation}

769: 

770: Function $\kappa(x)$ increases monotonically with increasing $\lambda$: it 

771: decreases with increasing $x$. For $\lambda = 1/2$, i.e., for $a = 

772: 0.395\dots$, one gets $\kappa(a) = -0.169\dots$ . For $\lambda$ 

773: approaching 1, $a$ tends to zero and $\kappa(a)$ diverges. For $a = 1/3$, 

774: $\lambda = 0.684\dots$, the correction amplitude vanishes. All 

775: higher-order corrections for the critical decay outside the transient 

776: regime vanish as well for this special value of $\lambda$ 

777: \cite{Fuchs1999b}. Therefore, for $\lambda$ near 0.7, the simple power-law 

778: formulas of the leading-order asymptotic-expansion theory like 

779: Eqs.~(\ref{eq:crit:corr}, \ref{eq:chit0}, \ref{eq:chi_power:leading}) 

780: describe the critical dynamics very well.

781: 

782: For the specified model, there holds $f^c\leqslant 1/2$. Hence, this model 

783: cannot be used to describe glassy-dynamics data for systems where the 

784: $\alpha$-peak loss spectra have a weight $f^c$ which exceeds 50\% of the 

785: total weight. Moreover, the relation between the exponent parameter 

786: $\lambda$ and the critical arrested part $f^c$, as formulated by 

787: Eq.~(\ref{eq:vc:vc}), is an artifact of the model. The minimum requirement 

788: for a schematic model for data analysis is the freedom to adjust $\lambda$ 

789: and $f^c$ independently. This goal can be achieved by introducing a second 

790: correlator. The first correlator $\phi_{q=1}(t) = \phi(t)$ is used as a 

791: caricature of the density-fluctuation dynamics. It provides the exponent 

792: parameter. The second one describes the dynamics of some probing variable, 

793: say, $A$. The correlator shall be denoted by $\phi_{q=2}(t) = \phi_A(t)$. 

794: All quantities referring to this correlator shall be indicated by an index 

795: $A$. The equation of motion~(\ref{eq:EOM:int}) is specified by $\Omega_A > 

796: 0$, $\nu_A \geqslant 0$, and a kernel $m_A(t)$. The two frequencies 

797: quantify the transient dynamics. The kernel is a polynomial 

798: (\ref{eq:EOM:kernel}) of the two correlators involved. The simplest model 

799: describing the coupling of the probing variable to the density-fluctuation 

800: is given by \cite{Sjoegren1986}

801: 

802: \begin{equation}\label{eq:mA}

803: m_A(t) = v_A \phi(t) \phi_A(t)\,.

804: \end{equation}

805: The coupling of the probing variable to the density fluctuations is 

806: quantified by $v_A >0$. Since there is no influence of the dynamics of the 

807: probing variable on the density dynamics, the same scaling-law function 

808: $g_\pm(t/t_\sigma)$, Eq.~(\ref{eq:phi_gpm}), describes the leading-order 

809: near-plateau dynamics of $\phi_A(t)$ as it does for that of $\phi(t)$. 

810: Notice in particular that the parameters $\Omega_A$, $\nu_A$, and $v_A$ do 

811: not modify the time scale $t_0$. In applications for data descriptions, 

812: the model parameters $\Omega_A$, $\nu_A$, and $v_A$ are to be considered 

813: as smooth functions of the physical control parameters, cf. lower insets 

814: in Figs.~\ref{fig:BZP} and \ref{fig:salol}.

815: 

816: The loss spectrum $\chi_A''(\omega)$ develops a $\beta$-peak upon 

817: increasing the mode-coupling coefficient $v_A$ \cite{Buchalla1988}. For 

818: this special case and in the limit of infinite $v_A$, the Cole-Cole 

819: susceptibility and the Mittag-Leffler correlator have been derived for the 

820: critical dynamics of the probing variable in Ref. \cite{Goetze1989c}. The 

821: model specified by Eqs.~(\ref{eq:mF12}, \ref{eq:mA}) was used repeatedly 

822: for the description of experimental data 

823: \cite{Singh1998,Ruffle1999,Brodin2002,Wiebel2002,Goetze2004,Cang2005}. 

824: Large data sets for the evolution of the glassy dynamics of propylene 

825: carbonate have been analyzed in Ref.~\cite{Goetze2000b}; results measured 

826: for different probing variables $A$ obtained by neutron-scattering, 

827: depolarized-light-scattering, and dielectric-loss spectroscopy have been 

828: fitted by a common first correlator $\phi(t)$. The different probes have 

829: been characterized by adjusting the parameters for the second correlator 

830: $\phi_A(t)$ only.

831: 

832: Up to order $(t_0/t)^{2a}$, the kernel $m_A(t)$ can be calculated from 

833: Eq.~(\ref{eq:mA}) substituting the expression~(\ref{eq:critcorr:phi}) for 

834: $\phi(t)$ and the analog expression for the second correlator,

835: 

836: \begin{equation}\label{eq:phiA}

837: \phi_A(t) = f_A^c + h_a (t_0/t)^a \left[ 1 + \hat{K}_A (t_0/t)^a \right] 

838: \,.

839: \end{equation}

840: Laplace transform yields $\omega m_A(\omega)$ as an asymptotic power 

841: series in the small parameter $\xi = (-i\omega t_0)^a$. The coefficients 

842: are linear functions of $f^c_A$, $h_A$, and $\hat{K}_A$. This is 

843: substituted on the left-hand side of Eq.~(\ref{eq:MCTlong:m}) for $q=2$. 

844: Expanding the right-hand side up to orders $\xi^2$, one can compare the 

845: coefficients in order to arrive at:

846: 

847: \begin{equation}\label{eq:fAhA}

848: f_A^c = 1- \left[ 1/ (f^c v_A) \right] \,,\quad

849: h_A = \lambda / \left( f^{c\,2} v_A\right)\,.

850: \end{equation}

851: 

852: The correction amplitude shall be noted in the convention of the general 

853: theory \cite{Franosch1997}:

854: 

855: \begin{subequations}\label{eq:cAKA}

856: \begin{equation}\label{eq:cAKA:cA}

857: \hat{K}_A = \kappa(a) + K_A\,,

858: \end{equation}

859: 

860: \begin{equation}\label{eq:cAKA:KA}

861: K_A = \lambda v_a \left[ \frac{1}{v_A(1-\lambda)} -\lambda \right] / 

862: \left[ v_A(1-\lambda) - 1 \right]\,.

863: \end{equation}

864: \end{subequations}

865: This leads to the parameters for the Cole-Cole susceptibility:

866: 

867: \begin{equation}\label{eq:chi0ccA}

868: \chi^{cc}_{0\,A} = 1/\left(f^c v_A\right) \,,\quad 

869: \omega^{c}_A t_0 = \left[\left(1-\lambda\right)/\left(\lambda\Gamma(1-a) 

870: \right)\right]^{1/a} \,.

871: \end{equation}

872: Remarkably, the comparison of Eq.~(\ref{eq:chi0ccA}) with the general 

873: expression in Eq.~(\ref{eq:cccpar:wcc}) shows, that the Cole-Cole 

874: frequency $\omega_A^{c}$ does not depend on the coupling coefficient 

875: $v_A$. $\omega_A^{c}t_0$ decreases with increasing $\lambda$ from 

876: 0.37\dots for $\lambda=1/2$ to about $10^{-4}$ for $\lambda$ near 0.88. 

877: For $\lambda$ around 0.7, $\omega_A^{c}t_0$ is about 0.02.

878: 

879: %%

880: \subsection{Critical Relaxation for a Small Cole-Cole 

881: Frequency\label{subsec:BZP}}

882: 

883: %%Fig. 1

884: \begin{figure}

885: \includegraphics[width=\columnwidth]{BZP}

886: \caption{\label{fig:BZP}OKE response functions measured for BZP for $T/K = 

887: 251,\,260,\,290,\,320$ \cite{Cang2003c} (full lines from bottom to top)

888: and fits by the schematic-model functions $\chi_A(t)$ \cite{Goetze2004} 

889: (dotted lines). The straight dashed line has slope $-0.80$.

890: The dash-dotted lines labeled \textit{sc} are the 

891: approximations by the scaling functions Eq.~(\ref{eq:chiA_beta}), the thin 

892: dashed lines are fits with Eq.~(\ref{eq:chiA_beta}) for freely adjusted 

893: $s_\sigma$ and $t_\sigma$. The insets show the separation parameter 

894: $\sigma$ (right) and the fitting parameters (left); two additional state 

895: points ($+$ and $\times$) indicate extrapolations for which solutions are 

896: shown in Fig.~\ref{fig:BZPwing}.

897: }

898: \end{figure}

899: 

900: Figure~\ref{fig:BZP} reproduces OKE-response functions measured for 

901: benzophenone (BZP) \cite{Cang2003c} and fits to these data by the response 

902: $\chi_A(t)$ calculated for the model defined in the preceding section. The 

903: fit parameters for the mode-coupling coefficients $v_1$, $v_2$, and $v_A$ 

904: are specified in the lower inset. The analysis shall be done by 

905: anticipating a transition point for $\lambda=0.70$, which implies the 

906: exponents $a=0.33$ and $b=0.64$. The cross in the inset is close to this 

907: bifurcation point. The $\sigma$-versus-$T$ diagram is shown in the upper 

908: inset; and extrapolation to $\sigma=0$ suggests the critical temperature 

909: $T_c=235\text{K}$ with an estimated uncertainty of $\pm 5\text{K}$. The 

910: $251\text{K}$ result exhibits the expected von~Schweidler decay, 

911: $\log\chi(t) = \text{const.}-(1-b)\log t$ for times between about 0.3ns 

912: and about 6ns. Further details can be inferred from 

913: Ref.~\cite{Goetze2004}.

914: 

915: 

916: %%Fig. 2

917: \begin{figure}

918: \includegraphics[width=\columnwidth]{BZPAt}

919: \caption{\label{fig:BZPphiAt} Correlation functions for BZP.

920: The lower panel exhibits the correlators 

921: $\phi_A(t)$ underlying the fits in Fig.~\ref{fig:BZP} (heavy full lines 

922: from left to right for $T/K = 320,\,290,\,260,\,$ and $251$, 

923: respectively). The light full line with label \textit{c} shows the 

924: critical correlator calculated for $\lambda = 0.7$ and $v_A = 30$. The 

925: dotted line is the leading asymptotic law, Eq.~(\ref{eq:crit:corr}), with 

926: the time scale $t_0 = 0.3775$ps. The dashed line marked \textit{vS} shows 

927: a von~Schweidler law, Eq.~(\ref{eq:vS:phi}), with $b=0.64$ and a time 

928: scale adjusted to match the $T=251$K curve. The dashed line labeled 

929: \textit{K} is a fit by the Kohlrausch law, $\phi_A(t) = f_A^c 

930: \exp[-(t/\tau)^\beta],\,\beta=0.91$, with $\tau$ adjusted to match the 

931: $T=260$K curve. The times $t_0$ and $1/\omega^c_A = 15$ps are indicated by 

932: arrows. The upper panel shows the critical decay (\textit{c}) together 

933: with the leading-order ($t^{-a}$) and next-to-leading-order ($t^{-2a}$)

934: asymptotic power-law solution, Eq.~(\ref{eq:critcorr:phi}). The dotted 

935: curve labeled \textit{cc} displays the leading-order Cole-Cole solution, 

936: Eq.~(\ref{eq:ML}), with $\chi_0^{cc}=1/9$. The points of 10\% deviation of 

937: the approximations $t^{-a}$, $t^{-2a}$, and \textit{cc} from the critical 

938: decay are indicated by the diamond, circle, and triangle, respectively.

939: }

940: \end{figure}

941: 

942: %%Fig. 3

943: \begin{figure}

944: \includegraphics[width=\columnwidth]{BZPchi}

945: \caption{\label{fig:BZPchi} Spectra for BZP.

946: The lower panel shows the susceptibility spectra for the correlators from 

947: the lower panel of Fig.~\ref{fig:BZPphiAt}. The leading-order critical 

948: spectrum, Eq.~(\ref{eq:chi_power:leading}), is the dotted straight

949: line marked by 

950: $\omega^a$. A dashed straight line with slope $-0.2$ is fitted to the 

951: $T=251$K spectrum to indicate an $\omega^{-b'}$ law for $b' = -0.2$. 

952: In the upper panel, the dashed curve marked $\omega^{2a}$ shows the  

953: approximation by Eq.~(\ref{eq:chi_power:chi}). The approximation by the 

954: Cole-Cole function (\textit{cc}), Eq.~(\ref{eq:ccleft:chi}), is shown 

955: dotted, the inclusion of the correction, Eq.~(\ref{eq:ccc}) yields the 

956: dashed curve labeled \textit{ccc} for $\omega_A^c = 67$ns$^{-1}$.

957: The points of 10\% deviation of the approximations $\omega^{-a}$, 

958: $\omega^{-2a}$, and \textit{cc} from the critical spectrum are indicated 

959: by the diamond, circle, and triangle, respectively. The frequencies 

960: $\omega_A^c$ and $\nu = \omega/(2\pi) = 1$THz are marked by arrows.

961: }

962: \end{figure}

963: 

964: The MCT results describe the measured evolution of the glassy dynamics 

965: adequately for times exceeding $t_\text{mic} = 1\text{ps}$ with two 

966: exceptions. The $T = 251\text{K}$ data exhibit a pronounced oscillation 

967: near $1.1\text{ps}$, while the calculated curve shows this oscillation 

968: near $0.9\text{ps}$. Furthermore, there are some deviations between the 

969: theoretical results and the data for the $T = 260\text{K}$ results for 

970: times around $10\text{ns}$. The four correlators $\phi_A(t)$ and the four 

971: loss spectra $\chi''_A(t)$, which are shown in the lower panels of 

972: Figs.~\ref{fig:BZPphiAt} and~\ref{fig:BZPchi}, respectively, are the MCT 

973: results corresponding to the MCT responses $\chi_A(t)$ shown in 

974: Fig.~\ref{fig:BZP}. The data fits have been calculated for fixed relative 

975: values of the four frequencies, which specify the transient dynamics: 

976: $\Omega_A = \Omega$, $\nu_A = \nu = 5\Omega$. The scale $\Omega$ for the 

977: fits is chosen different for different temperatures $T$ \cite{Goetze2004}. 

978: The curves in Fig.~\ref{fig:BZPphiAt} and~\ref{fig:BZPchi} are presented 

979: with a $T$-independent scale corresponding to that one used for the fit 

980: for $T = 251$K: $\Omega = 1.67$ps$^{-1}$. With the mentioned reservations, 

981: these curves can be considered as the measured quantities for the 

982: OKE-probing variable $A$. The full lines marked $c$ in 

983: Figs.~\ref{fig:BZPphiAt} and~\ref{fig:BZPchi} show the critical correlator 

984: $\phi_A^c(t)$ and the critical loss spectrum ${\chi''_A}^c(\omega)$, 

985: respectively, for the transition point specified by $\lambda = 0.70$ and 

986: $v_A = 30$. The correlator relaxes to the plateau $f_A^c = 8/9$ with a 

987: critical amplitude $h_A = 7/27$.

988: 

989: The second scaling law for the decay below the plateau of the correlators, 

990: Eq.~(\ref{eq:alpha}), implies a corresponding scaling law for the 

991: response:

992: 

993: \begin{equation}\label{eq:alpha_chi}

994: \chi_A(t) = \tilde{\chi}_A(t/t'_\sigma)/t'_\sigma\,.

995: \end{equation}

996: The control-parameter independent shape function reads 

997: $\tilde{\chi}_A(\tilde{t}) = -\partial\tilde{\phi}_A(\tilde{t})  / 

998: \partial\tilde{t}$. The reader can check this superposition principle for 

999: the results in Fig.~\ref{fig:BZP} as follows. The $\log 

1000: \chi_A(t)$-versus-$\log t$ curve for $T = 320\text{K}$ and $t \geqslant 

1001: 2\text{ps}$ can be translated so that it collapses with the curves for 

1002: $T/\text{K} = 290,\,260,\,$ and $251$ for times exceeding $4\text{ps}$, 

1003: $90\text{ps}$, and $290\text{ps}$, respectively. Equivalently, the four 

1004: loss spectra in Fig.~\ref{fig:BZPchi} for $\chi''_A(\omega) \geqslant 0.1$ 

1005: are connected by the superposition principle: $\chi''_A(\omega) = 

1006: \tilde{\chi}''_A(\omega t'_\sigma)$. The von~Schweidler law describes a 

1007: part of the high-frequency wing of the loss peak as shown by the straight 

1008: dashed line with label \textit{vS} for the $T = 251\text{K}$ curve. The 

1009: Kohlrausch law for a stretching exponent $\beta = 0.91$ describes the 

1010: upper part of the $\alpha$-peak as shown by the dashed line with label $K$ 

1011: for the $T = 260\text{K}$ result.

1012: 

1013: There is an interval of times between the end of the transient dynamics 

1014: and the start of the von~Schweidler decays that deals with the relaxation 

1015: towards and through the plateau $f_A^c$. The dynamics causes the loss 

1016: spectra for $\chi''_A(\omega) \leqslant 0.1$ and $\omega/(2\pi) < 0.1$THz 

1017: shown in Fig.~\ref{fig:BZPchi}. The leading-order asymptotic solution of 

1018: the MCT equations of motion deals with this part of the dynamics together 

1019: with the von~Schweidler-law part by Eq.~(\ref{eq:phi_gpm}). This formula 

1020: yields the first scaling law for the response

1021: 

1022: \begin{equation}\label{eq:chiA_beta}

1023: \chi_A(t) = h_A s_\sigma \hat{\chi}(t/t_\sigma)\,;\quad

1024: s_\sigma = \sqrt{|\sigma|}/t_\sigma\,.

1025: \end{equation}

1026: 

1027: The shape function reads $\hat{\chi}(\hat{t}) = -\partial g_\pm(\hat{t}) / 

1028: \partial\hat{t}$, $T \gtrless T_c$. For $T > T_c$, Eq.~(\ref{eq:gpm:B}) 

1029: yields the von~Schweidler response for large rescaled times $\hat{t} = t / 

1030: t_\sigma$: $\hat{\chi}(\hat{t} \gg 1) =B\cdot b / \hat{t}^x\,,\quad x = 1- 

1031: b < 1$. For small rescaled times, Eq.~(\ref{eq:gpm:ta}) yields 

1032: $\hat{\chi}(\hat{t} \ll 1) = a/ \hat{t}^x\,,\quad x = 1 + a > 1$. The 

1033: crossover from one asymptote to the other occurs for times near 

1034: $t_\sigma$. The light dashed lines in Fig.~\ref{fig:BZP} exhibit fits of 

1035: the data by Eq.~(\ref{eq:chiA_beta}) for $\lambda = 0.70$. However, both 

1036: scales $s_\sigma$ and $t_\sigma$ are adjusted with the aim to achieve a 

1037: good match in the von~Schweidler-law regime. The $1/t^{1+a}$-law is not 

1038: exhibited by the results in Fig.~\ref{fig:BZP}. For $T = 320\text{K}$, the 

1039: crossover time $t_\sigma$ is located near $1\text{ps}$, i.e., it is within 

1040: or close to the transient regime. For the other three temperatures and 

1041: times within the interval $t_\text{mic} < t < t_\sigma$, the measured 

1042: response and the calculated functions $\chi_A(t)$ are below the values of 

1043: $h_A s_\sigma\hat{\chi}(t / t_\sigma)$.

1044: 

1045: The straight dashed line in Fig.~\ref{fig:BZP} has a slope of $-0.80$. It 

1046: demonstrates a pseudo-von~Schweidler decay of the $T = 251\text{K}$ 

1047: results for $2\text{ps} < t < 20\text{ps}$: $\phi_A(t) - \text{const.} 

1048: \propto -t^{b'}\,,\quad b' = 0.20$. The $\phi_A(t)$-versus-$\log t$ curves 

1049: in Fig.~\ref{fig:BZPphiAt} for $t > 2\text{ps}$ approach and cross the 

1050: plateau as downward-bent curves before entering the von~Schweidler decay 

1051: regime. There are no inflection points of the curves for some time near 

1052: $t_\sigma$, as implied by Eq.~(\ref{eq:phi_gpm}).

1053: 

1054: The first scaling law describes a loss minimum of some value 

1055: $\chi_\text{min}$ at some position $\omega_\text{min}$. The shape of the 

1056: $\log \chi_A''(\omega)$-versus-$\log\omega$ curves is independent of the 

1057: separation parameter $\sigma$ and the scales fix the position of the 

1058: curve: $\chi_\text{min} \propto \sqrt{|\sigma|}$, $\omega_\text{min} 

1059: \propto 1/t_\sigma$. The minimum is due to the crossover from the 

1060: von~Schweidler wing of the $\alpha$-peak, $\chi''(\omega) \propto 1 / 

1061: \omega^b$, to the power-law asymptote for the critical dynamics, 

1062: $\chi''(\omega) \propto \omega^a$. For most practical purposes, the 

1063: minimum can be approximated by the interpolation formula 

1064: \cite{Sjoegren1990b}:

1065: 

1066: \begin{equation}\label{eq:interpol}

1067: \chi''_A(\omega) / \chi_\text{min} = \left[

1068: b \left(\omega/\omega_\text{min}\right)^a

1069: + a \left(\omega_\text{min}/\omega\right)^b

1070: \right]/(a+b)\,.

1071: \end{equation}

1072: However, the loss spectra in the lower panel of Fig,~\ref{fig:BZPchi} do 

1073: not exhibit loss minima, which can be described by the interpolation 

1074: formula for the leading-order asymptotic result. The minima for 

1075: frequencies $\omega$ near $1\text{ps}^{-1}$ do not depend sensitively on 

1076: $\sigma$. With decreasing $|\sigma|$, the minimum position even shifts 

1077: slightly upwards rather than downwards. With increasing frequency, the 

1078: high-frequency part of the $\alpha$-peak exhibits a crossover from the 

1079: von~Schweidler decay, $\chi''(\omega) \propto 1/\omega^b$, to a pseudo 

1080: von~Schweidler decay, $\chi''(\omega) \propto 1/\omega^{b'}$. For $T = 

1081: 251$K, this high-frequency wing extends from $\omega \approx 

1082: 0.02\text{ps}^{-1}$ to about 0.4ps$^{-1}$ with $b' = 0.2$, as is 

1083: demonstrated by the dashed straight line. The observed minimum is due to 

1084: the crossover from the $1/\omega^{b'}$ wing to the spectral peak for the 

1085: normal-liquid dynamics. The latter is located near $\nu= 1$THz.

1086: 

1087: The scaling-law fits, which are shown in Fig.~\ref{fig:BZP} by the light 

1088: dashed lines, are misleading because the scales $s_\sigma$ and $t_\sigma$ 

1089: have been adjusted freely. Calculating these scales from 

1090: Eqs.~(\ref{eq:sigma:tsigma}, \ref{eq:vc:sigma}, \ref{eq:fAhA}), one 

1091: obtains for $T = 251\text{K}$ and 260K the dash-dotted lines marked by 

1092: $sc$. The two distance parameters $\varepsilon = (T_c -T) / T_c$ are -0.07 

1093: and -0.11, respectively. These values are too large for a leading-order 

1094: asymptotic formula to be applicable.

1095: 

1096: The correlator $\phi_A(t)$ for the lowest temperature under discussion is 

1097: close to the critical one for $t$ up to about $30$ps, as is shown in the 

1098: lower panel of Fig.~\ref{fig:BZPphiAt}. Hence, the $T = 251$K curves 

1099: exhibit critical glassy dynamics for the times between 1ps and 30ps. From 

1100: Eqs.~(\ref{eq:cAKA}) one obtains the correction amplitude $\hat{K}_A = 

1101: -1.51$. This large value yields an onset time $t_A^*$ for the $t^{-a}$-law 

1102: which is beyond $t = 10^3$ps as is marked by the diamond in the upper 

1103: panel of the figure. Including the leading correction term yields the 

1104: approximation by the dashed line denoted $t^{-2a}$. It improves the 

1105: description of the critical decay so that it can be understood for times 

1106: larger than about $t = 30$ps. But the expansion formula~(\ref{eq:phiA}) 

1107: cannot be used to describe that part of the critical decay, which is 

1108: measured for BZP and described by the solution of the two-component 

1109: schematic MCT model.

1110: 

1111: Substituting the cited values for $\lambda$, $f_A^c$, and $h_A$ into 

1112: Eqs.~(\ref{eq:fqmc}, \ref{eq:hmq}), one gets the plateau for the modulus,

1113: $f_A^{m\,c} = 8$, and the critical amplitude $h_A^m = 21$. Notice that the 

1114: relative strength of the first scaling law amplitude for the modulus is 

1115: much larger than that for the correlator: $h_A^m/f_A^{m\,c} = 2.6$ versus 

1116: $h_A/f_A^c = 0.29$. From Eq.~(\ref{eq:cmq:cmq}), one obtains the 

1117: correction amplitude $\hat{K}_A^m = 0.124$. As expected from the 

1118: discussions of 

1119: Sec.~\ref{sec:crit}, the onset time $t_A^{m\,*}$ for the leading-order 

1120: formula for the modulus, 

1121: 

1122: \begin{subequations}\label{eq:CC}

1123: \begin{equation}\label{eq:CC:crit_modulus}

1124: m_A(t) = f_A^{m\,c} + h_A \left(t_0 /t \right)^a\,,

1125: \end{equation}

1126: is smaller than $t_A^*$ by more than a factor $1000$. As a result, see the 

1127: upper panel of Fig.~\ref{fig:BZPphiAt}, the equivalent 

1128: formula~(\ref{eq:ML}) for the correlator,

1129: 

1130: \begin{equation}\label{eq:CC:ML}

1131: \phi_A(t) = f_A^c + (1-f_A^c) 

1132: \text{M}_a\left[-\left(t\omega_A^c\right)^a\right]\,,

1133: \end{equation}\end{subequations}

1134: describes the probing variable $A$ for all times exceeding $t = 1$ps. The 

1135: leading-order asymptotic result for the modulus explains the response for 

1136: $T = 251$K quantitatively within the interval $1$ps$\leqslant t \leqslant 

1137: 30$ps.

1138: 

1139: Figure~\ref{fig:BZPchi} shows that the leading-asymptotic description of 

1140: the critical loss spectrum, Eq.~(\ref{eq:chi_power:leading}), would be 

1141: relevant for the explanation of the loss minimum only in cases with 

1142: $\omega_\text{min} < 2\cdot10^{-5}$ps$^{-1}$. The dashed line labeled 

1143: $\omega^{2a}$ exhibits the asymptotic expansion for the critical loss up 

1144: to the leading correction, Eq.~(\ref{eq:chi_power:chi}). This formula is 

1145: relevant for $\omega < 10^{-3}$ps$^{-1}$. Even this expression is 

1146: unsatisfactory for the discussion of the BZP results because of the large 

1147: correction amplitude. However, the Cole-Cole spectrum describes the 

1148: critical loss reasonably for $\omega < 0.2$ps$^{-1}$. The correction 

1149: amplitude for the Cole-Cole law, Eq.~(\ref{eq:cccpar:cqcc}) is very small,

1150: $\hat{K}_A^{cc} = 0.076$. The Cole-Cole law with leading correction, 

1151: Eq.~(\ref{eq:ccc}), is shown in the upper panel of Fig.~\ref{fig:BZPchi} 

1152: by the curve marked $ccc$, and it slightly above the leading-order result 

1153: in the regime of large frequencies.

1154: 

1155: For the system under study, the Cole-Cole frequency reads $\omega_A^c = 

1156: 0.067$ps$^{-1}$. This frequency $\omega_A^c/(2\pi) \approx 10$GHz is small 

1157: compared to the $1$THz scale for the normal-liquid dynamics. The value 

1158: $-\log\omega_A^c$ is in the center of the $\log t$ interval studied by the 

1159: results of Fig.~\ref{fig:BZP}. Therefore, the critical spectrum relevant 

1160: for the understanding of the data differs qualitatively from the 

1161: leading-order power-law formula. As a result, the loss minima in 

1162: Fig.~\ref{fig:BZPchi}, which are caused by the superposition of the 

1163: critical spectra for $\omega$ near and above $\omega_A^c$ and the 

1164: von~Schweidler-law wing of the $\alpha$-peak, differ drastically from the 

1165: general low-frequency shape described by Eq.~(\ref{eq:interpol}). For $T = 

1166: 251$K, the superposition yields the pseudo-von~Schweidler law manifested 

1167: as a wing specified by the exponent $b' = 0.20$. This wing will be 

1168: analyzed further in Sec.~\ref{sec:alphabeta}.

1169: 

1170: 

1171: %%

1172: \subsection{Critical Relaxation for an Intermediate Cole-Cole 

1173: Frequency\label{subsec:salol}}

1174: 

1175: Five OKE response functions measured for salol \cite{Hinze2000b} and fits 

1176: by the functions $\chi_A(t)$ of the above explained schematic model are 

1177: reproduced in Fig.~\ref{fig:salol}. The filled dots in the insets specify 

1178: the mode-coupling coefficients $v_1$, $v_2$, and $v_A$ used for the 

1179: calculations. Further details can be found in Ref.~\cite{Goetze2004}, 

1180: where, however, the value for $\Omega$ was misprinted. The series of 

1181: states $(v_1, v_2)$ extrapolates to a transition point with $\lambda = 

1182: 0.73$, which implies a critical exponent $a=0.31$ and a von~Schweidler 

1183: exponent $b=0.59$. The extrapolation of the $\sigma$-versus-$T$ parameters 

1184: to $\sigma = 0$ suggests a critical temperature $T_c = 245$K with an 

1185: estimated uncertainty of $\pm 3$K. The MCT correlators $\phi_A(t)$ and 

1186: loss spectra $\chi''_A(\omega)$, which are equivalent to the response 

1187: $\chi_A(t)$ in Fig.~\ref{fig:salol}, are shown in the lower panels of 

1188: Figs.~\ref{fig:salolphiAt} and~\ref{fig:salolchiAw}, respectively. They 

1189: can be considered as the measured results for the glassy dynamics of 

1190: salol, since the calculated and measured curves in Fig.~\ref{fig:salol} 

1191: agree outside the transient regime. This holds with the reservation, that 

1192: fits and measurements for the $T = 270$K results exhibit some 

1193: discrepancies for times exceeding 20ns. The lines with label $c$ in 

1194: Figs.~\ref{fig:salolphiAt} and~\ref{fig:salolchiAw} show critical 

1195: correlators and critical loss spectra, respectively, calculated for 

1196: $\lambda = 0.73$ and $v_A = 55$. The correlators relax to the plateau 

1197: $f_A^c = 0.93$ with a critical amplitude $h_A = 0.18$.

1198: 

1199: %%Fig 4

1200: \begin{figure}

1201: \includegraphics[width=\columnwidth]{salol}

1202: \caption{\label{fig:salol}

1203: OKE-response functions measured for salol for $T/K = 247,\, 257,\, 270,\, 

1204: 300,\, 340$ (full lines from bottom to top) \cite{Hinze2000b}. The dotted 

1205: lines are fits by the schematic-model response $\chi_A(t)$ calculated with 

1206: $\Omega = 2\Omega_A = 10\nu_A = 15.9$ps$^{-1}$ \cite{Goetze2004}. Symbols 

1207: and curve styles are the same as in Fig.~\ref{fig:BZP}. The slope of the 

1208: straight dashed line is $-1.15$.

1209: }

1210: \end{figure}

1211: 

1212: 

1213: %%Fig 5

1214: \begin{figure}

1215: \includegraphics[width=\columnwidth]{phiAt}

1216: \caption{\label{fig:salolphiAt} Correlation functions for salol.

1217: The lower panel shows the correlators 

1218: $\phi_A(t)$ used for the response fits in Fig.~\ref{fig:salol} for $T/K = 

1219: 247,\,257,\,270,\,300,\,$ and $340$ (heavy full lines from right to left). 

1220: The light full line shows the critical correlator calculated for $\lambda 

1221: = 0.73$, and $v_A = 55$. The dashed line marked \textit{vS} exhibits a 

1222: von~Schweidler law, Eq.~(\ref{eq:vS:phi}), for a time scale chosen to 

1223: match the $T=247$K correlator. The dashed line marked \textit{K} is a 

1224: Kohlrausch-law fit for the $T=257$K curve with exponent $\beta = 0.95$. 

1225: The dotted line exhibits the formula~(\ref{eq:crit:corr}) with $t_0 = 

1226: 0.00246$ps. The upper panel shows the critical decay (\textit{c}) together 

1227: with the leading- ($t^{-a}$) and next-to-leading-order asymptotic 

1228: solution~(\ref{eq:critcorr:phi}) ($t^{-2a}$). The dotted curve labeled 

1229: \textit{cc} displays the leading-order solution from Eq.~(\ref{eq:ML}) 

1230: with $\chi_0^{cc}=0.067$ and $\omega_A^c = 7.25$ps. The points of 10\% 

1231: deviation of the approximations $t^{-a}$, $t^{-2a}$, and \textit{cc} from 

1232: the critical decay are indicated by the diamond, circle, and triangle, 

1233: respectively. The times $t_0$ and $1/\omega_A^c$ are marked by arrows.

1234: }

1235: \end{figure}

1236: 

1237: The test of the superposition laws for the long-time relaxation parts of 

1238: the correlators is left to the reader. This second scaling law of MCT 

1239: relates the $\alpha$-relaxation peaks for the loss spectra for 

1240: $\chi''_A(\omega) \geqslant 0.1$. Kohlrausch-law fits and von~Schweidler 

1241: asymptotes have been added to the data for $T = 257$K and $T = 247$K, 

1242: respectively, in order to emphasize that the results for the evolution of 

1243: the below-plateau-decay process follows the familiar pattern. Notice that 

1244: the response for $T = 257$K exhibits the von~Schweidler-law decay, 

1245: $\log\chi_A(t) = \text{const.} -(1-b) \log t$ for the large time interval 

1246: 0.1ns $\leqslant t \leqslant $ 10ns. The measurement for this temperature 

1247: permits a rather precise determination of the exponent $b$ and, thereby, 

1248: of $\lambda$.

1249: 

1250: %%Fig 6

1251: \begin{figure}

1252: \includegraphics[width=\columnwidth]{chiA}

1253: \caption{\label{fig:salolchiAw} Spectra for salol.

1254: The lower panel shows the susceptibility spectra $\chi_A''(\omega)$ for 

1255: the correlators shown in the lower panel of Fig.~\ref{fig:salolphiAt}. The 

1256: leading-order approximation to the critical spectrum, 

1257: Eq.~(\ref{eq:chi_power:chi}), is shown as dotted straight line marked 

1258: $\omega^a$. The dashed line labeled \textit{sc'} exhibits the 

1259: interpolation formula (\ref{eq:interpol}) with the von~Schweidler exponent 

1260: $b=0.59$ and the critical exponent $a$ replaced by $a'' = 0.24$. 

1261: The upper panel exhibits the critical spectrum as full line with label 

1262: \textit{c}, the leading-order asymptotic approximation, 

1263: Eq.~(\ref{eq:chi_power:leading}), (dotted, labeled $\omega^a$) and 

1264: Eq.~(\ref{eq:chi_power:chi}), (dashed, labeled $\omega^{2a}$). The 

1265: approximation by the Cole-Cole function for $\omega_A^c = 7.25$ps$^{-1}$ 

1266: (\textit{cc}), Eq.~(\ref{eq:ccleft:chi}), is shown dotted, the inclusion 

1267: of the correction, Eq.~(\ref{eq:ccc}), yields the dashed curve labeled 

1268: \textit{ccc}. The points of 10\% deviation of the approximations 

1269: $\omega^{-a}$, $\omega^{-2a}$, and \textit{cc} from the critical spectrum 

1270: are indicated by the diamond, circle, and triangle, respectively. The 

1271: frequencies $\omega_A^c$ and $\omega/(2\pi) = \nu = 1$THz are marked by 

1272: arrows.

1273: }

1274: \end{figure}

1275: 

1276: The light-dashed lines in Fig.~\ref{fig:salol} are fits of the data by the 

1277: first-scaling-law expression~(\ref{eq:chiA_beta}) with freely adjusted 

1278: scales $s_\sigma$ and $t_\sigma$. For $T \geqslant 270$K, these fits 

1279: describe the data outside the transient regime up to times within the 

1280: von~Schweidler-law region. A more detailed documentation of this result 

1281: can be found in Ref.~\cite{Hinze2000b}, where also consistency of the 

1282: fitted scales $s_\sigma$, $t_\sigma$, and $t'_\sigma$ with the 

1283: MCT-power-law formulas was demonstrated. The scaling-law results 

1284: calculated with the MCT values for the scales $h_A$, $s_\sigma$, and 

1285: $t_\sigma$ in Eq.~(\ref{eq:chiA_beta}) are shown in Fig.~\ref{fig:salol} 

1286: as dash-dotted lines marked $sc$ for the $T = 257$K and $T = 247$K data. 

1287: For $T = 257$K, $\varepsilon = -0.05$ is so large, that 

1288: Eq.~(\ref{eq:chiA_beta}) cannot account quantitatively for the scales 

1289: ruling the von~Schweidler-law decay. For $T \geqslant 270$K the distance 

1290: parameter $\varepsilon$ exceeds 10\%; the leading asymptotic expansion 

1291: formula for the plateau-crossing process is not applicable for such large 

1292: distances of the control parameter from the critical value. In contrast, 

1293: the scaling law~(\ref{eq:chiA_beta}) describes the $T = 247$K result well 

1294: for the large time interval 0.04ns $<t<$ 10ns. In this case, the distance 

1295: parameter $\varepsilon = -0.008$ is so small that the leading-order 

1296: expansion result for the plateau crossing, Eq.~(\ref{eq:phi_gpm}), 

1297: accounts for a major part of the long-time response -- readjusting the 

1298: scales $(h_As_\sigma)$, and $t_\sigma$ would not improve the fit. 

1299: 

1300: Figure~\ref{fig:salolphiAt} shows that the $T = 247$K response is close to 

1301: the critical one for $t \leqslant 1$ns. The straight dashed line with 

1302: slope $-1.15$, which is shown in Fig.~\ref{fig:salol}, demonstrates that 

1303: the critical response exhibits a power-law decay $\chi_A(t) \propto 

1304: 1/t^{1+a'}$, $a' = 0.15$, for 2ps $<t<$ 100ps. This is equivalent to a 

1305: power-law decay of the correlator: $\phi_A(t) - f_a^c \propto 1/t^{a'}$. 

1306: In qualitative agreement with the prediction by Eq.~(\ref{eq:phi_gpm}), 

1307: the crossover from the $t^{-a}$ decay to the $-t^b$ decay causes an 

1308: inflection point for the $\phi_A(t)$-versus-$\log t$ curve in 

1309: Fig.~\ref{fig:salolphiAt}. The corresponding crossover from the 

1310: $\omega^{-b}$ wing of the $\alpha$-peak to some $\omega^{a''}$ spectrum 

1311: causes the loss minimum located near $\log\omega = -2.5$ in 

1312: Fig.~\ref{fig:salolchiAw}. The dashed curve marked $sc'$ exhibits a 

1313: description of this minimum by the function 

1314: $\chi''_A(\omega)/\chi_\text{min} = \left[ b \left(\omega / 

1315: \omega_\text{min}\right)^{a''} + a'' \left(\omega_\text{min} / 

1316: \omega\right)^b \right]/(a''+b)$ which is suggested by 

1317: Eq.~(\ref{eq:interpol}). For salol, as opposed to BZP, the minimum 

1318: position decreases strongly with decreasing $|\varepsilon|$, in 

1319: qualitative agreement with the first-scaling-law results.

1320: 

1321: The inconsistent exponents for $T = 247$K in the regime 2ps $<t<$100ps 

1322: indicate that the leading-order asymptotic description is still not 

1323: applicable; the exponents $a' = 0.15$ obtained from the fit of the OKE 

1324: data and $a'' = 0.24$ obtained from the fit of the corresponding minimum

1325: are different and both are smaller than the correct value of $a = 0.31$. 

1326: The deviation of the critical correlator from 

1327: the leading-order power-law result~(\ref{eq:crit:corr}) 

1328: and~(\ref{eq:chit0}) within the specified time interval is caused by the 

1329: large value for the correction amplitude: $\hat{K}_A = -1.78$. This value 

1330: implies an onset time for the $t^{-a}$ decay of beyond 20ps as shown by 

1331: the diamond in Fig.~\ref{fig:salolphiAt}. The high-frequency part of the 

1332: loss minimum is located in the region 10ns$^{-1}<\omega<$ 1ps$^{-1}$. 

1333: Figure~\ref{fig:salolchiAw} demonstrates that the $\omega^a$ law does not 

1334: describe the critical loss there. Inclusion of the leading correction 

1335: terms for the analytic description of the critical dynamics, i.e., using 

1336: Eq.~(\ref{eq:critcorr:phi}) for the correlator and 

1337: Eq.~(\ref{eq:chi_power:chi}) for the loss spectrum, explains the result 

1338: for $t \geqslant 0.6$ps and $\omega \leqslant 25$ns$^{-1}$, as 

1339: demonstrated in Figs.~\ref{fig:salolphiAt} and~\ref{fig:salolchiAw}, 

1340: respectively.

1341: 

1342: From Eq.~(\ref{eq:cmq:cmq}), one derives the correction amplitude for the 

1343: modulus $\hat{K}_A^m = 0.177$. The leading-order formula 

1344: (\ref{eq:CC:crit_modulus}) for the relaxation kernel describes the kernel 

1345: $m_A$ outside the transient regime. The corresponding expression 

1346: (\ref{eq:CC:ML}) for the correlator and the equivalent Cole-Cole formula 

1347: for the loss spectrum account for the critical glassy dynamics, as shown 

1348: in Figs.~\ref{fig:salolphiAt} and~\ref{fig:salolchiAw}. The Cole-Cole 

1349: formula with correction term of amplitude $\hat{K}_A^c = 0.09$ yields a 

1350: slight improvement compared to the equation based on $\hat{K}_A^c = 0$, as 

1351: shown in Fig.~\ref{fig:salolchiAw} by the curve with label $ccc$. 

1352: Equation~(\ref{eq:cccpar:wcc}) leads to the Cole-Cole frequency for salol: 

1353: $\omega_A^c = 7.25$ps$^{-1}$. This value is close to the loss peak for the 

1354: normal-liquid dynamics: $\omega_A^c t_\text{mic} \approx 1$. The part of 

1355: the glassy critical loss spectrum which is relevant for the explanation of 

1356: the data in Fig.~\ref{fig:salol} deals with the regime $10^{-4} \leqslant 

1357: \omega/\omega_A^c \leqslant 10^{-1}$. Within this frequency interval, the 

1358: Cole-Cole spectrum increases smoothly with $\omega$. Therefore, the 

1359: leading-order formulas (\ref{eq:chit0}) or (\ref{eq:chi_power:leading}) 

1360: describe the dynamics qualitatively. However, the true critical spectrum 

1361: differs from its low-frequency asymptote, 

1362: Eq.~(\ref{eq:chi_power:leading}). Hence, the description of the dynamics 

1363: by the scaling law is not correct quantitatively. Within the specified 

1364: interval, the $\log\chi''_A(\omega)$-versus-$\log\omega$ spectrum can be 

1365: approximated reasonably by a straight line, $\log\chi''_A(\omega) \approx 

1366: \text{const.} + a'' \log\omega$, $a'' = 0.24$. As a result, the critical 

1367: dynamics is approximated well by a power law decay specified by an 

1368: exponent $a''$ smaller than the critical exponent $a$.

1369: 

1370: In the preceding section, the critical spectra have been discussed for 

1371: both small and intermediate Cole-Cole frequencies $\omega_q^c$; the case 

1372: of a large Cole-Cole frequency is found in the 

1373: depolarized-light-scattering spectra for CKN \cite{Li1992,Sperl2005bpre}. 

1374: In the fits with a schematic model by Alba-Simionesco and coworkers 

1375: \cite{Krakoviack1997,Krakoviack2002} one can identify a Cole-Cole peak for 

1376: the critical spectra, but the description with the power-law solution is 

1377: superior if terms up to order $\omega^{2a}$ are considered.

1378: 

1379: 

1380: %%%

1381: \section{Cole-Cole wing\label{sec:alphabeta}}

1382: 

1383: Figures~\ref{fig:BZP}--\ref{fig:BZPchi} and 

1384: \ref{fig:salol}--\ref{fig:salolchiAw} demonstrate the scenarios for the 

1385: evolution of the glassy dynamics for $\omega^c_q \ll t^{-1}_\text{mic}$ 

1386: and $\omega^c_q \approx t^{-1}_\text{mic}$, respectively, for the probing 

1387: variable described by the second correlator of a two-component schematic 

1388: MCT model. Solutions for this model, which exemplify a scenario as shown 

1389: in Fig.~\ref{fig:BZPchi}, have been discussed before by Cummins 

1390: \cite{Cummins2005}. He pointed out that the evolution of the wing 

1391: phenomenon obtained from the model is similar to the one known for 

1392: dielectric-loss spectroscopy for glycerol and several van-der-Waals 

1393: liquids \cite{Dixon1990}. He emphasizes also that the cited measurements 

1394: refer to temperatures $T$ below $T_c$, while the calculations are done for 

1395: $T > T_c$. Furthermore, the experimental data demonstrate the wing only 

1396: for frequencies below 10GHz$\approx 10^{-2}t^{-1}_\text{mic}$. So far, no 

1397: wing has been reported which occurs in the two-decade frequency window 

1398: adjacent to the microscopic excitation region.

1399: 

1400: In order to describe the wing phenomenon quantitatively, one can extend 

1401: the Cole-Cole formula so that the spectrum can be described also for small 

1402: but non-vanishing separation parameters $\sigma = C \varepsilon \propto 

1403: (T_c-T)/T_c$. The leading-order expression for the plateau-crossing 

1404: process of the fluctuating-force correlators reads in analogy to 

1405: Eq.~(\ref{eq:phi_gpm}): $m_q(t) = f_q^{m\,c} + h_q^m 

1406: g_\pm(t/t_\sigma),\,\varepsilon\gtrless 0$. Laplace transformation yields 

1407: $\omega m_q(\omega) = -f_q^{m\,c} + h_q^m C(\omega)$. Function $C(\omega)$ 

1408: obeys a scaling law,

1409: 

1410: \begin{equation}\label{eq:bet_C}

1411: C(\omega) = \sqrt{|\sigma|} c_\pm(\omega t_\sigma)\,,\quad 

1412: \varepsilon \gtrless 0\,,

1413: \end{equation}

1414: with the control-parameter independent shape functions $c_\pm(\hat{\omega})$ 

1415: given by the Laplace transforms $g_\pm(\hat{\omega})$ of the shape 

1416: functions $g_\pm(\hat{t})$:

1417: $c_\pm (\hat{\omega}) = \hat{\omega} g_\pm (\hat{\omega})$.

1418: Substitution of $\omega m_q(\omega)$ into Eq.~(\ref{eq:MCTlong:chi}) 

1419: yields the desired result. Expansion of the right-hand side in terms of 

1420: the small parameter $C(\omega)$ and comparison with the result for 

1421: $\chi_q(\omega)$ following from Eq.~(\ref{eq:phi_gpm}) reproduces the 

1422: relations (\ref{eq:fqmc},\ref{eq:hmq}) for the critical arrested parts and 

1423: critical amplitudes. One gets

1424: 

1425: \begin{subequations}\label{eq:ccbeta}

1426: \begin{equation}\label{eq:ccbeta_lead}

1427: \chi_q(\omega) = \left( 1- f_q^c \right) /

1428: \left\{ 1 - \left[h_q/(1-f_q^c)\right] C(\omega)\right\}\,.

1429: \end{equation}

1430: Equation~(\ref{eq:ccbeta_lead}) was found for the schematic model in 

1431: Ref.~\cite{Goetze1989c} for the special case of the second correlator 

1432: $\phi_A(t)$ and in the limit of infinite coupling strength $v_A$. It is 

1433: shown here, that Eq.~(\ref{eq:ccbeta_lead}) can be derived without such 

1434: restrictions. The leading correction to Eq.~(\ref{eq:ccbeta_lead}) can be 

1435: calculated from the asymptotic expansions in \cite{Franosch1997,Fuchs1998} 

1436: in a straightforward manner; however, the evaluation of the result for 

1437: fitting data is rather involved. Focusing on the critical decay, one can 

1438: extend the preceding formula by adding the large-frequency part of the 

1439: correction:

1440: 

1441: \begin{equation}\label{eq:ccbeta_corr}\begin{split}

1442: \chi_q(\omega) = \left( 1- f_q^c \right) /

1443: \left\{ 1 - \left[h_q/(1-f_q^c)\right] C(\omega)\right.\\\left.

1444: + \hat{K}_q^\text{cc}\left(-i\omega/\omega_q^c\right)^{2a}\right\}\,.

1445: \end{split}\end{equation}

1446: \end{subequations}

1447: This expression describes the small-$\varepsilon$ dynamics for $\omega 

1448: t_\text{mic}<1$ and frequencies extending down to the beginning of the 

1449: von~Schweidler-law decay. 

1450: 

1451: %% Fig 7

1452: \begin{figure}[ht]

1453: \includegraphics[width=\columnwidth]{BZPwing}

1454: \caption{\label{fig:BZPwing}Asymptotic description of the wing in BZP.

1455: The heavy full line reproduces the BZP loss spectrum for $T=251$K from 

1456: Fig.~\ref{fig:BZPchi} and the dashed one exhibits the spectrum of the 

1457: extended Cole-Cole susceptibility, Eq.~(\ref{eq:ccbeta_lead}). The 

1458: triangles mark the points of 10\% deviation between these spectra. 

1459: The dotted line \textit{sc} is the first-scaling-law result for the 

1460: susceptibility, and the line \textit{cc} shows the Cole-Cole spectrum for 

1461: the critical point. The dash-dotted line represents the 

1462: second-scaling-law result for the loss peak. The thin full lines are 

1463: solutions corresponding to the state points indicated by $+$ and $\times$ 

1464: in Fig.~\ref{fig:BZP}.

1465: }

1466: \end{figure}

1467: 

1468: BZP exemplifies the case of such a small $\hat{K}^{cc}_q$ that the 

1469: leading-order result for the modulus, Eq.~(\ref{eq:ccbeta_lead}), can be 

1470: used. Figure~\ref{fig:BZPwing} demonstrates that this result accounts for 

1471: the $T=251$K spectrum for the large dynamical range $10^{-3}\text{ps}^{-1} 

1472: < \omega < 1\text{ps}^{-1}$. The scaling-law result for the loss spectrum, 

1473: $\chi_A''(\omega) = h_A C_A''(\omega)$, is shown as dotted line 

1474: \textit{sc}; it can describe only a part of the von~Schweidler-law 

1475: spectrum for $\omega < 10^{-3}\text{ps}^{-1}$, but is inadequate for 

1476: larger frequencies. This observation for the dynamics in the frequency 

1477: domain is equivalent to the one demonstrated in Fig.~\ref{fig:BZP} for the 

1478: results in the time domain. The second-scaling-law result for the loss 

1479: spectrum is exhibited as dash-dotted line. It follows from 

1480: Eq.~(\ref{eq:alpha}), $\chi_A''(\omega) = (\omega t_\sigma') 

1481: \tilde{\phi}_A''(\omega t_\sigma')$, and accounts for the loss peak for 

1482: $\omega < 10^{-3}\text{ps}^{-1}$. Combining the result for the second 

1483: scaling law for the modulus with the result for the first scaling law for 

1484: the loss spectrum explains the BZP spectrum for $T = 251$K for $\omega < 

1485: 1\text{ps}^{-1}$ including in particular the $\alpha$-peak wing. The small 

1486: but systematic discrepancies between the loss spectrum in the 

1487: structural-relaxation regime and the asymptotic MCT expressions result 

1488: from the fact that the distance parameter $\varepsilon = (T-T_c)/T_c 

1489: \approx 0.07$ is so large, that the asymptotic results for the scales 

1490: still have noticeable errors while the shape functions already agree well 

1491: with the nontrivial spectra.

1492: 

1493: Starting from the state that fits the 251K data for BZP, a number of 

1494: extrapolations are possible within the schematic model. Keeping all 

1495: parameters but $(v_1, v_2)$ fixed, two additional states shall be 

1496: considered; they are indicated by $\times$ and $+$ in the left inset of 

1497: Fig.~\ref{fig:BZP}; and their spectra are marked accordingly in 

1498: Fig.~\ref{fig:BZPwing}. The first state point ($\times$) is close to the 

1499: transition point and is characterized by $\sigma = -0.003$ ($\varepsilon 

1500: =-0.01$). Solutions at this point are shown in Ref.~\cite{Goetze2004} for 

1501: $\chi(t)$ and $\chi_A(t)$. The respective spectrum $\chi_A''(\omega)$ in 

1502: Fig.~\ref{fig:BZPwing} exhibits nearly-constant loss for $0.4\, 

1503: 10^{-3}$ps$^{-1} < \omega < 0.4$ps$^{-1}$ within a 10\% margin. In this 

1504: frequency window the maximum around $\omega^c_A$ indicates an emerging 

1505: Cole-Cole peak; the minimum at $\omega = 10^{-3}$ps$^{-1}$ is caused by 

1506: the crossover of the high-frequency wing of the $\alpha$-peak and the 

1507: low-frequency wing of the Cole-Cole peak. This minimum is ruled by the 

1508: first scaling law and the divergent time scale $t_\sigma$. The second 

1509: state ($+$) is an interpolation between $\times$ and the solution for BZP; 

1510: the separation parameter is $\sigma = -0.01$ ($\varepsilon = -0.03$), and 

1511: the point is chosen similar to the ones used in Fig.~4 of 

1512: Ref.~\cite{Cummins2005}. The spectrum in Fig.~\ref{fig:BZPwing} shows a 

1513: wing, $\chi_A''(\omega) \propto \omega^{-0.1}$, for almost three orders of 

1514: magnitude in frequency. Hence, the crossover between $\alpha$- and 

1515: Cole-Cole peaks can be interpreted as a power-law wing $\omega^{-b'}$ for 

1516: some range in frequencies and control parameters without fine-tuning. The 

1517: evolution of the crossover progresses from a wing with gradually lower 

1518: exponents $b'$ to nearly-constant loss and eventually the emergence of a 

1519: Cole-Cole peak. While the extrapolations presented in this paragraph are 

1520: not at all implied by the fit of the data, the scenarios seem nevertheless 

1521: possible for realistic parameter values.

1522: 

1523: %% Fig 8

1524: \begin{figure}[ht]

1525: \includegraphics[width=\columnwidth]{BZPOKE}

1526: \caption{\label{fig:BZPOKE}Asymptotic description of the OKE data for BZP.

1527: The full line reproduces the BZP data for $T=251$K from \cite{Cang2003c}; 

1528: the dotted line reproduces the fit from Fig.~\ref{fig:BZP}. The dashed 

1529: curve, the dash-dotted curve \textit{sc}, and the dotted curve \textit{cc} 

1530: respectively show the Fourier backtransforms of the extended Cole-Cole 

1531: susceptibility in Eq.~(\ref{eq:ccbeta_lead}), of the first scaling law, 

1532: and of the Cole-Cole law in Eq.~(\ref{eq:ccleft:chi}), cf. 

1533: Fig.~\ref{fig:BZPwing}.

1534: }

1535: \end{figure}

1536: 

1537: To conclude the discussion of BZP, let us return to the OKE data for 

1538: $T=251$K as shown in Fig.~\ref{fig:BZPOKE}. In addition to data and fit 

1539: for this state, the Fourier backtransforms of the asymptotic curves from 

1540: Fig.~\ref{fig:BZPwing} are displayed. While the asymptotic solution at the 

1541: critical point (labeled \textit{cc}) covers only less than a decade, the 

1542: extended Cole-Cole susceptibility in Eq.~(\ref{eq:ccbeta_lead}) describes 

1543: the data over three orders of magnitude in time and thereby explains the 

1544: $t^{-0.8}$-law found empirically. The remaining discrepancies are again 

1545: due to the relatively large distance parameter for this state; Fig.~3 of 

1546: Ref.~\cite{Goetze2004} shows prefect agreement with the Cole-Cole law for 

1547: states closer to the transition including state $\times$ from 

1548: Fig.~\ref{fig:BZP}.

1549: 

1550: %Fig 9

1551: \begin{figure}[ht]

1552: \includegraphics[width=\columnwidth]{Salolwing}

1553: \caption{\label{fig:Salolwing}Asymptotic description of the minimum in 

1554: salol. The full line reproduces the salol loss spectrum for $T=247$K from 

1555: Fig.~\ref{fig:salolchiAw} and the dashed one exhibits the spectrum of the 

1556: extended Cole-Cole susceptibility, Eq.~(\ref{eq:ccbeta_lead}). The 

1557: triangles mark the points of 10\% deviation between these spectra. 

1558: The dotted line  \textit{cc} is the Cole-Cole spectrum for the critical 

1559: point, and the line \textit{sc} shows the first-scaling-law result for the 

1560: susceptibility; the diamonds mark the points where \textit{sc} deviates by  

1561: 10\% from the loss spectrum. The dash-dotted line represents the 

1562: second-scaling-law result for the loss peak.

1563: }

1564: \end{figure}

1565: 

1566: Different from the case for BZP, the lowest temperature available for 

1567: salol, $T = 247$K, is described by a much smaller distance parameter, 

1568: $\varepsilon \approx 0.01$, and therefore shares a larger frequency 

1569: regime, say $\omega\lesssim 0.1$ps$^{-1}$, with the critical loss 

1570: spectrum, see the lower panel of Fig.~\ref{fig:salolchiAw}. Being so close 

1571: to the critical point, the second scaling law in Fig.~\ref{fig:Salolwing} 

1572: -- labeled $\alpha$ -- is indistinguishable from the loss spectrum for 

1573: $\omega \leqslant 10^{-4}$ps$^{-1}$. The first scaling law \textit{sc} 

1574: describes the solution up to around $\omega\lesssim 0.005$ps$^{-1}$ which 

1575: is the same as for the solution at the critical point, cf. upper panel of 

1576: Fig.~\ref{fig:salolchiAw}. As for the critical spectrum, the leading-order 

1577: approximation in Eq.~(\ref{eq:ccbeta_lead}) improves the description to 

1578: larger frequencies by two decades up to $\omega\approx 0.1$ps$^{-1}$ 

1579: before crossing over to the Cole-Cole peak which in this case is hidden 

1580: under the microscopic excitations.

1581: 

1582: 

1583: %%%

1584: \section{Conclusion\label{sec:sum}}

1585: 

1586: Within the regime of glassy dynamics, susceptibilities $\chi_q(\omega)$ 

1587: and moduli $\omega m_q(\omega)$ are related by Eq.~(\ref{eq:MCTlong:chi}); 

1588: and at the critical point, susceptibilities and moduli are both described 

1589: by the universal power law $\omega^a$. The range of validity of this power 

1590: law is given by amplitudes for the correction $\omega^{2a}$ -- $\hat{K}_q$ 

1591: for the susceptibilities and $\hat{K}_q^m$ for the moduli, which are 

1592: related by Eq.~(\ref{eq:cmq:cmq}). For some parameter regions a large 

1593: value of $|\hat{K}_q|$ can render the power-law expansion for the 

1594: susceptibilities irrelevant, while the correction amplitude 

1595: $|\hat{K}_q^m|$ is so small that the universal power-law can be used for 

1596: the description of the modulus successfully. Typically, this occurs if 

1597: $\hat{K}_q$ is large and negative, and this can be expected if the plateau 

1598: value $f^c_q$ is high, cf. Eq.~(\ref{eq:cmq:cmq}). In this case, the 

1599: susceptibility can be described well by the expansion of its inverse, 

1600: Eq.~(\ref{eq:ccc}). For vanishing correction $\hat{K}_q^m = K_q^{cc} = 0$, 

1601: formula, the susceptibility at the critical point is given by the 

1602: Cole-Cole law~(\ref{eq:ccleft:chi}), a formula first introduced in 1941 

1603: \cite{Cole1941}.

1604: 

1605: The Cole-Cole frequency $\omega_q^c$ in Eq.~(\ref{eq:ccleft:chi}) 

1606: introduces a characteristic scale for assorting the critical dynamics into 

1607: three categories. (1) If $\omega_q^c$ is large compared to the scale 

1608: $t^{-1}_\text{mic}$ for the band of normal liquid excitations, the 

1609: Cole-Cole susceptibility reduces to the leading-order scaling-law formula 

1610: which is implied by Eq.~(\ref{eq:phi_gpm}). There is a control-parameter 

1611: sensitive minimum as discussed for Eq.~(\ref{eq:interpol}) that 

1612: interpolates between the von~Schweidler-law tail of the $\alpha$-peak, 

1613: $\chi_q''(\omega) \propto 1/\omega^b$, and the critical spectrum 

1614: $\chi_q''(\omega) \propto \omega^a$, where $a$ and $b$ are related via the 

1615: exponent parameter $\lambda$. As an example for such a scenario one can 

1616: cite the molten salt CKN. While the leading-order result alone is not 

1617: sufficient to describe the measured data quantitatively, the power-law 

1618: expansion in Eq.~(\ref{eq:chi_power:chi}) yields a satisfactory 

1619: description that is superior to the Cole-Cole solution 

1620: \cite{Sperl2005bpre}. (2) If $\omega^c_q$ is close to $t^{-1}_\text{mic}$, 

1621: one encounters a scenario demonstrated in Fig.~\ref{fig:Salolwing} for 

1622: salol. There is a control-parameter sensitive loss minimum. It originates 

1623: from the crossover between the von~Schweidler-law tail and the critical 

1624: spectrum; and the critical spectrum approaches the maximum of the 

1625: underlying Cole-Cole peak. A description by Eq.~(\ref{eq:interpol}) is 

1626: possible only, if the exponent $a$ is replaced by some effective one 

1627: $a''$, which is smaller than $a$, cf. Fig.~\ref{fig:salolchiAw}. (3) If 

1628: $\omega_q^c$ is smaller than the microscopic time scale, say $\omega_q^c 

1629: t_\text{mic} \approx 0.05$, we obtain the scenario seen in 

1630: Fig.~\ref{fig:BZPwing} for benzophenone (BZP). In this case, the 

1631: von~Schweidler-law part of the $\alpha$-peak, $\chi_q'' \propto 

1632: 1/\omega^b$, crosses over to some flatter wing, $\chi_q'' \propto 

1633: 1/\omega^{b'}$, $a < b' < b$. This wing is caused by approaching the 

1634: Cole-Cole spectrum for frequencies around the maximum at $\omega^c_q$. For 

1635: yet higher frequencies, one encounters a crossover -- from the 

1636: high-frequency part of the Cole-Cole spectrum for structural relaxation to 

1637: the spectrum due to normal-liquid excitations -- producing a different 

1638: kind of a minimum. The position of this minimum is control-parameter 

1639: insensitive. For the understanding of this minimum, both the scaling-law 

1640: (\ref{eq:chiA_beta}) and the interpolation formula (\ref{eq:interpol}) are 

1641: irrelevant. On the other hand, as shown in Sec.~\ref{sec:alphabeta} and 

1642: especially in Fig.~\ref{fig:BZPwing}, it is possible to describe the wing 

1643: rather accurately by the leading-order formula~(\ref{eq:ccbeta_lead}). If 

1644: the $\alpha$-peak is shifted to yet lower frequencies, the wing can give 

1645: way to a separate Cole-Cole peak.

1646: 

1647: All figures in this work have been prepared with model parameters that 

1648: reproduce the OKE-response functions of benzophenone (BZP) and salol. 

1649: Hence, Fig.~\ref{fig:BZPwing} implies that the BZP spectra for 

1650: temperatures near 250K, when measured by depolarized light-scattering in 

1651: backward direction, should exhibit an $\alpha$-peak wing for frequencies 

1652: $\nu$ between about 1GHz and about 100GHz. The crossover from the 

1653: von~Schweidler-law wing to the wing induced by the Cole-Cole peak is 

1654: expected to occur around 1GHz. This crossover position depends sensitively 

1655: on the temperature, because the von~Schweidler-law relaxation depends 

1656: sensitively on $T$.

1657: 

1658: The Cole-Cole law with correction, Eq.~(\ref{eq:ccc}), has been derived in 

1659: its fully general microscopic version in Sec.~\ref{sec:crit}; only in 

1660: Secs.~\ref{sec:appl} and~\ref{sec:alphabeta} the theory was specialized to 

1661: schematic models. While for the schematic models a number of parameters 

1662: can be fixed to describe experimental data, for microscopic models the 

1663: static structure of the model system determines these parameters uniquely. 

1664: A first example for Cole-Cole dynamics in a microscopic model has been 

1665: discussed recently for the mean-squared displacement $\delta r^2(t)$ of 

1666: the hard-sphere system \cite{Sperl2005}; including the Mittag-Leffler 

1667: function for the critical relaxation allows for an analytic description of 

1668: the mean-squared displacement for the full range of the dynamics similar 

1669: to BZP in Fig.~\ref{fig:BZPwing}. In addition, the Cole-Cole dynamics was 

1670: identified in the data measured by van~Megen \textit{et. al} 

1671: \cite{Megen1998} where Eq.~(\ref{eq:ccc}) accounts for the data for an 

1672: interval in time of three orders of magnitude adjacent to the transient 

1673: dynamics. Hence, the results discussed above seem relevant for both 

1674: molecular and colloidal glasses and can be expected to facilitate more 

1675: detailed investigations.

1676: 

1677: I want to thank W.~G\"otze for repeated help during the preparation of 

1678: this work; I am also grateful to L. Berthier, H.~Z.~Cummins, E.~R\"o\ss 

1679: ler, and Th.~Voigtmann for enlightening discussions. Support was provided 

1680: by DFG grant No. SP~714/3-1 and NSF grants DMR0137119 and DMS0244492.

1681: 

1682: %%% bib

1683: \bibliographystyle{apsrev}

1684: \bibliography{lit,add}

1685: 

1686: \end{document}

1687: 

1688: