1: \documentclass[aps,pre,showpacs,amsmath,amssymb,twocolumn]{revtex4}\usepackage{pslatex}

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

3: \usepackage{graphicx}

4: \usepackage{dcolumn}

5: \usepackage{natbib}

6: 

7: \newcommand{\ST}{\mathcal S}

8: 

9: \begin{document}

10: 

11: \title{Logarithmic Relaxation in a Colloidal System}

12: 

13: \author{M. Sperl}

14: \affiliation{Physik-Department, Technische Universit\"at M\"unchen, 

15: 85747 Garching, Germany}

16: 

17: \date{\today}

18: 

19: \begin{abstract}

20: The slow dynamics for a colloidal suspension of particles interacting with

21: a hard-core repulsion complemented by a short-ranged attraction is

22: discussed within the frame of mode-coupling theory for ideal glass

23: transitions for parameter points near a higher-order glass-transition

24: singularity. The solutions of the equations of motion for the density

25: correlation functions are solved for the square-well system in

26: quantitative detail by asymptotic expansion using the distance of the

27: three control parameters packing fraction, attraction strength and

28: attraction range from their critical values as small parameters. For given

29: wave vectors, distinguished surfaces in parameter space are identified

30: where the next-to-leading order contributions for the expansion vanish so

31: that the decay functions exhibit a logarithmic decay over large time

32: intervals. For both coherent and tagged particle dynamics the

33: leading-order logarithmic decay is accessible in the liquid regime for

34: wave vectors of several times the principal peak in the structure factor.

35: The logarithmic decay in the correlation function is manifested in the

36: mean-squared displacement as a subdiffusive power law with an exponent

37: varying sensitively with the control parameters. Shifting parameters

38: through the distinguished surfaces, the correlation functions and the

39: logarithm of the mean-squared displacement considered as functions of the

40: logarithm of the time exhibit a crossover from concave to convex behavior,

41: and a similar scenario is obtained when varying the wave vector.

42: \end{abstract}

43: 

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

45: 

46: \maketitle

47: 

48: %%%

49: \section{\label{sec:introduction}Introduction}

50: 

51: The dynamics in an interacting many particle system is conveniently

52: described by density autocorrelation functions $\phi_q(t)$ for time $t$

53: and wave vector $q$. These correlation functions can be measured in both

54: experiment and computer simulation \cite{Boon1980}.  Mode-coupling theory

55: for ideal glass transitions (MCT) discusses the transition from a liquid

56: to a glass as a bifurcation in the long-time limit of the correlator

57: $\phi_q(t)$ \cite{Goetze1991b}. In the liquid state, the correlation

58: function decays to zero. If a control parameter, say density, exceeds some

59: critical value, the long-time limit changes discontinuously from zero to a

60: finite value, a glass transition occurs \cite{Bengtzelius1984}. This

61: liquid-glass transition is identified with an $A_2$- or fold singularity

62: \cite{Arnold1986} in the equations of motion of MCT. The simplest example

63: for a liquid-glass transition is found in the hard-sphere system (HSS),

64: where the interaction potential among the particles is zero unless their

65: mutual separation becomes smaller than their diameter where the potential

66: becomes infinitely repulsive, thus preventing the particles from

67: overlapping. The HSS is the system MCT was applied to first

68: \cite{Bengtzelius1984}, and it is also the system for which the most

69: detailed predictions have been worked out \cite{Franosch1997,Fuchs1998}.

70: Close to the singularity, the equations of motion can be expanded in

71: asymptotic series. This yields a two-step decay with two related power

72: laws for the short-time and the long-time decay at a liquid-glass

73: transition \cite{Goetze1991b}. The HSS can be realized in colloidal

74: suspensions \cite{Pusey1986}. Experiments in these systems lead to the

75: conclusion that MCT is able to describe the main aspects of the glass

76: transition qualitatively and some aspects even quantitatively

77: \cite{Megen1995}.

78: 

79: MCT can also exhibit other singularities than the fold \cite{Goetze1991b}.  

80: These higher-order singularities were predicted recently to occur for

81: colloidal systems with short-ranged attraction where $A_3$- and

82: $A_4$-singularities were found that are also called cusp and swallowtail

83: \cite{Fabbian1999,Bergenholtz1999,Dawson2001}. In these systems, the

84: hard-core repulsion is supplemented by a short ranged attraction, e.g. in

85: the square-well system (SWS). A cusp singularity is the endpoint of a line

86: of glass-glass transitions that arises if two different mechanisms of

87: arrest are of the same importance. In the SWS the first mechanism is the

88: hard core repulsion that causes a transition as in the HSS via the well

89: known cage effect. The second mechanism leading to arrest is bond

90: formation introduced by the attractive part of the potential.  This latter

91: transition was proposed as relevant for the transition to a gel

92: \cite{Bergenholtz1999}. If the difference in the two mechanisms is less

93: pronounced, the glass-glass transitions vanish and give rise to an

94: $A_4$-singularity. In the SWS this happens as the range of the attraction

95: is increased \cite{Dawson2001}. The range of attraction considered here is

96: of order less than $20\%$ of the particle diameter and the strength is

97: about several $k_\text{B}T$. This is accessible in colloid-polymer

98: mixtures with nonadsorbing polymer which is well under control

99: experimentally \cite{Poon2002}. Higher-order singularities have also been

100: identified for a number of short-ranged potentials with shapes differring from 

101: the SWS yielding certain quantitative trends but no qualitative changes

102: \cite{Goetze2003b}.

103: 

104: In addition to the success of MCT for the description of the HSS, two

105: findings support the use of this theory for the description of colloids

106: with attraction. First, a reentry phenomenon was predicted by the theory

107: where a glass state is melted upon increasing the attraction

108: \cite{Fabbian1999,Dawson2001}. This was subsequently found in several

109: experiments \cite{Eckert2002,Pham2002} and computer simulation studies

110: \cite{Pham2002,Foffi2002b,Zaccarelli2002b,Puertas2003}. Second, there are

111: indications of logarithmic decay \cite{Puertas2002} and related anomalous

112: decays \cite{Mallamace2000,Zaccarelli2002b} that are consistent with

113: scenarios found within MCT \cite{Fabbian1999,Dawson2001}. To investigate

114: the dynamics in such systems, apart from computer simulation dynamic light

115: scattering has already been used to determine correlation functions

116: \cite{Mallamace2000,Eckert2002,Pham2002}. Direct imaging techniques are

117: available to determine also the mean-squared displacement (MSD) with high

118: precision \cite{Kegel2000,Weeks2000,Weeks2002}. The purpose of the present

119: paper is the application of the general theory for higher-order

120: singularities \cite{Goetze2002} to the SWS and the derivation of testable

121: quantitative predictions for the correlation functions and the MSD.

122: Certain scenarios have been discussed before for schematic models 

123: \cite{Goetze2002}. Some of these scenarios shall be identified also in the 

124: microscopic model in the following.

125: 

126: The paper is organized as follows. In Sec.~\ref{sec:asy} the equations of

127: motion and the asymptotic solution for the logarithmic decay are

128: summarized and the subdiffusive power law for the MSD is derived. The

129: theory is applied to the $A_4$-singularity in the SWS for the correlation

130: functions in Sec.~\ref{sec:corr} and for the MSD in Sec.~\ref{sec:MSD}.

131: Changes in the scenarios when moving from the $A_4$-singularity to

132: $A_3$-singularities are discussed in Sec.~\ref{sec:A3}, and

133: Sec.~\ref{sec:HCY} contains a comparison to results obtained for the

134: hard-core Yukawa system (HCY).  Section\ref{sec:conclusion} presents a

135: conclusion.

136: 

137: %%%

138: \section{\label{sec:asy}Asymptotic Solutions}

139: 

140: We shall consider a system of $N$ particles with diameter $d$ in a volume

141: $V$ interacting with a spherical potential. When at time $t$ the $j$th

142: particle is located at $\vec{r}_j(t)$ the density variables are defined as

143: $\rho_q(t)=\sum_j\exp[i\vec{q}\vec{r}_j(t)]$.

144: 

145: %%

146: \subsection{\label{subsec:eom}Equations of Motion}

147: 

148: The equations of motion for the normalized density correlators $\phi_q(t)=

149: \langle\rho_{{\vec{q}}}^{*}(t)\rho_{\vec{q}}\rangle/\langle|\rho_{\vec{q}}|^2

150: \rangle$ within MCT, when Brownian dynamics for the motion in colloids is

151: assumed, are given by

152: \cite{Bengtzelius1984,Goetze1991b,Szamel1991,Fuchs1993a},

153: \begin{subequations}\label{eq:mct:MCT}

154: \begin{equation}\label{eq:mct:phi_col}

155: \tau_q\partial_t\phi_q(t)+\phi_q(t)+\int_0^t

156: m_q(t-t')\partial_{t'}\phi_q(t')\,dt'=0\,.

157: \end{equation}

158: The initial conditions are $\phi_q (0)=1$. The microscopic time scale

159: reads $\tau_q=S_q/(D_0q^2)$. It is given by the short-time diffusion

160: coefficient, $D_0$, characterizing the Brownian motion and the static

161: structure factor $S_q=\langle|\rho_{\vec{q}}|^2\rangle$. The mode-coupling

162: approximation results in expressing the kernels $m_q(t)$ in terms of the

163: correlators $\phi_q(t)$ \cite{Goetze1991b},

164: \begin{equation}\label{eq:mct:kernel}

165: m_q (t) = {\cal F}_q \left[\mathbf{V}, \phi_k (t) \right] \,\, .

166: \end{equation}

167: As a consequence of the factorization into pair modes for the structural

168: relaxation in simple liquids, ${\mathcal F}_q$ is a bilinear functional of

169: the density correlators \cite{Bengtzelius1984},

170: \begin{equation}\label{eq:mct:Fdef}

171: {\mathcal F}_q[\tilde{f}]  = \frac{1}{2} \int \frac{{d}^3k}{(2 \pi )^3}

172: V_{\vec{q},\vec{k}} \tilde{f}_k \tilde{f}_{|\vec{q}-\vec{k}|}\,,

173: \end{equation}

174: and the vertex is determined completely by the static structure of the

175: liquid system \cite{Goetze1976b,Sjoegren1980b},

176: \begin{equation}\label{eq:mct:vertex}

177: V_{\vec{q},\vec{k}} =  S_q S_k S_{|\vec{q}-\vec{k}|}\, \rho

178: \left[ {\vec{q}} \cdot

179: \vec{k}\,{c_k} +\vec{q} \cdot

180: (\vec{q}-\vec{k})\,{{c_{|\vec{q}-\vec{k}|}} }

181:  \right]^2/q^4\,.

182: \end{equation}

183: \end{subequations}

184: The number density is given by $\rho=N/V$ and $c_q$ denotes the direct

185: correlation function which is related to the static structure factor $S_q$

186: in the Ornstein-Zernike relation, $S_q=1/[1-\rho\,c_q]$, both depend on

187: external control parameters like density or temperature \cite{Hansen1986}.

188: For the SWS with hard-core diameter $d$, depth of the potential $u_0$, and

189: range of the potential $\Delta$, we get three dimensionless control

190: parameters, the packing fraction $\varphi=d^3\rho\pi/6$, the attraction

191: strength $\Gamma=u_0/(k_\text{B}T)$ and the relative well width

192: $\delta=\Delta/d$. These can be combined to a control-parameter vector

193: $\mathbf{V}=(\varphi,\Gamma,\delta)$.

194: 

195: It is the long-time limit of the correlation function,

196: $\lim_{t\rightarrow\infty}\phi_q(t) = f_q$, that determines whether a

197: system is in the liquid regime, where $f_q=0$, or in an arrested state,

198: where $0< f_q\leqslant 1$. In the latter case, the values $f_q$

199: characterize the arrested glassy state and the $f_q$ are called glass-form

200: factors or Debye-Waller factors. In the long-time limit, the equation of

201: motion Eq.\ (\ref{eq:mct:MCT}) reduces to an equation involving only the

202: mode-coupling functional and the glass-form factors \cite{Goetze1991b},

203: \begin{equation}\label{eq:Feq}

204: f_q/(1-f_q) = {\cal F}_q[f]\,.

205: \end{equation}

206: 

207: Frequently studied is the dynamics of a single or tagged particle with the

208: single particle density $\rho_q^s(t)=\exp[i\vec{q}\vec{r}_s(t)]$. For the

209: correlation function of a tagged particle,

210: $\phi^s_q(t)=\langle\rho_{{\vec{q}}}^{s\,*}(t) \rho^s_{\vec{q}}\rangle$,

211: similar equations as Eqs.\ (\ref{eq:mct:MCT}) have been derived

212: \cite{Bengtzelius1984,Fuchs1998},

213: \begin{subequations}\label{eq:mct:tagged}

214: \begin{equation}\label{eq:mct:phis_col}

215: \tau^s_q\partial_t\phi^s_q(t)+\phi^s_q(t)+\int_0^t

216: m^s_q(t-t')\partial_{t'}\phi^s_q(t')\,dt'=0\,,

217: \end{equation}

218: with $\tau^s_q=1/(D^s_0q^2)$. The short-time diffusion coefficient for a

219: single particle, $D^s_0$, again specifies the Brownian dynamics. The

220: mode-coupling functional for the tagged particle motion,

221: \begin{equation}\label{eq:mct:Fsdef}

222: {\mathcal F}^s_q[f,f^s]  = \int \frac{{d}^3k}{(2 \pi )^3}  S_k

223: \frac{\rho}{q^4} {c^s_k}^2  (\vec{q}\vec{k})^2

224: f_k f^s_{|\vec{q}-\vec{k}|}\,,

225: \end{equation}

226: is also determined by the static structure of the liquid system where

227: $c^s_q$ is the single-particle direct correlation function

228: \cite{Hansen1986}.

229: \end{subequations}

230: 

231: The dynamics of the tagged particle is coupled to the coherent density

232: correlator $\phi_q(t)$ and for that reason $\phi^s_q(t)$ also displays the

233: bifurcation dynamics that is driven by $\phi_q(t)$. The equation for the

234: long-time limits of the tagged particle correlations function,

235: $\phi^s_q(t\rightarrow\infty)=f_q^s$, reads

236: \begin{equation}\label{eq:mct:fqs}

237: f_q^s/(1-f_q^s)

238: ={\cal F}^s_q[f,f^s]\,.

239: \end{equation}

240: In the following, the tagged particle will be assumed as of the same sort

241: as the host fluid. If the host particles are in the liquid state, $f_q=0$,

242: a tagged particle cannot be arrested, and in that case Eq.\

243: (\ref{eq:mct:fqs})  implies $f^s_q=0$.

244: 

245: The MSD is defined by 

246: $\delta r^2(t) = \langle |\vec{r}_s(t)-\vec{r}_s(0)|^2 \rangle$ and 

247: describes the average distance a particle has traveled within some time

248: $t$ \cite{Hansen1986}. It is obtained, e.g., as small wave-number limit of

249: the tagged-particle correlator in Eq.\ (\ref{eq:mct:tagged}),

250: $\phi^s_q(t)=1-q^2\delta r^2(t)/6+{\mathcal O}(q^4)$

251: \cite{Goetze1991b,Fuchs1998},

252: \begin{subequations}\label{eq:mct:MSD}

253: \begin{equation}\label{eq:mct:MSD_col}

254: \delta r^2(t) + D^s_0 \int_0^t m^{(0)}(t-t')\,\delta

255: r^2(t')\,dt'=6D^s_0t\,,

256: \end{equation}

257: $m^{(0)}(t)=\lim_{q\rightarrow 0}m^s_q(t) 

258: ={\mathcal F}_{MSD}[\phi(t),\phi^s(t)]$. The mode-coupling functional for 

259: the MSD reads

260: \begin{equation}\label{eq:mct:FMSDdef}

261: {\mathcal F}_{MSD}[f,f^s] = \int \frac{{d}k}{(6 \pi^2 )}\,\rho\, S_k

262: (c^s_k)^2 f_k f^s_k\,.

263: \end{equation}

264: \end{subequations}

265: A characteristic localization length $r_s$ is defined by the second moment

266: for the relaxation of the distribution of $\phi^s_q(t)$

267: \cite{Goetze1991b}, which can be identified with the functional in Eq.\

268: (\ref{eq:mct:FMSDdef}) $r_s^2=1/{\mathcal F}_{MSD}[f,f^s]$. It is the

269: long-time limit of the MSD. Its value at the critical point, $r_s^c$,

270: characterizes the arrested structure. The value $6r_s^{c\,2}$ represents

271: the plateau for the dynamics of $\delta r^2(t)$.

272: 

273: Equations~(\ref{eq:mct:MCT}) to (\ref{eq:mct:MSD}) are solved numerically

274: using algorithms introduced in \cite{Fuchs1991b,Goetze1996}. Details of

275: the implementation are found in \cite{Franosch1997,Dawson2001}. We use $d$

276: as unit of length, $d=1$, and choose the unit of time so that

277: $1/D_0=1/D_0^s=160$. The structure factors for the SWS and the HCY are

278: calculated in mean-spherical approximation \cite{Dawson2001,Cummings1979}.

279: The wave numbers shall be discretized to a grid of $M$ points with a

280: spacing $\Delta q=0.4/d$. The cutoff in the calculations is ranging from

281: $M=300$ for $\delta>0.04$ up to $M=750$ for $\delta=0.02$.

282: 

283: %%

284: \subsection{\label{subsec:log}Logarithmic Decay Laws}

285: 

286: The asymptotic solution at higher-order singularities shall be quoted from

287: Ref.~\cite{Goetze2002} where also further details can be found. The

288: asymptotic expansion is performed in small deviations of the correlation

289: function from the critical long-time limit $f^c_q$ involving the

290: coefficients

291: \begin{equation}\label{eq:asy:Aqk_def}

292: A_{q k_1 \cdots k_n}^{(n)} (\mathbf{V})  =  \frac{1}{n!} (1 - f_q^c)\,

293: \frac{\partial^n {\cal F}_q \left [\mathbf{V}, f_k^c \right ] }{ \partial

294: f_{k_1}^c \cdots\partial f_{k_n}^c } \,

295: (1 - f_{k_1}^c) \cdots (1 - f_{k_n}^c)   \,,

296: \end{equation}

297: 

298: which can be split into values at the singularity, 

299: $A_{qk_1\cdots k_n}^{(n)c}$, and remainders, 

300: $A_{q k_1 \cdots k_n}^{(n) } (\mathbf{V}) = A_{q k_1 \cdots k_n}^{(n)c} 

301: + \hat{A}_{q k_1 \cdots k_n}^{(n)} (\mathbf{V})$.

302: The Jacobian matrix of Eq.\ (\ref{eq:Feq}) is singular at the critical 

303: points and assumes the form $\bigl[\delta_{q k} - A_{q k}^{(1) c} \bigr]$. 

304: The non-negative left and right eigenvectors of matrix $A_{q k}^{(1) c}$ 

305: shall be denoted by $a^*_q$ and $a_q$ and can be fixed uniquely by 

306: requiring $\sum_q \, a^*_q \, a_q = 1$ and $\sum_q \, a^*_q \, a_q^2 = 1$. 

307: The reduced resolvent $R_{q k}$ of $A_{q k}^{(1) c}$ maps vectors 

308: orthogonal to $a^*_q$ to vectors orthogonal to $a_q$.

309: The leading-order result for Eq.\ (\ref{eq:mct:MCT}) is then given by

310: \begin{equation}\label{eq:asy:log_decay}

311: \phi_q (t) = f_q^c + h_q \left[ - B \ln (t / \tau) \right]  \,,\;

312:  B = \sqrt{\left[ - 6\varepsilon_1 (\mathbf{V}) / \pi^2 \right]}\,,

313: \end{equation}

314: with the critical amplitudes $h_q=(1-f^c_q)a_q$ and the separation 

315: parameter $\varepsilon_1(\mathbf{V})=a^*_q \hat{A}_q^{(0)}(\mathbf{V})$, 

316: which is restricted to negative values, $\varepsilon_1<0$. If 

317: $\varepsilon$ indicates the distance of the control parameters 

318: $\mathbf{V}$ from the critical point, the leading result is of order 

319: $\sqrt{\varepsilon}$ and correct up to ${\cal O}(\varepsilon)$. The 

320: next-to-leading-order approximation is

321: \begin{eqnarray}\label{eq:asy:G1G2q}

322: \phi_q (t) = (f_q^c + \hat{f}_q) &+& h_q  \bigl [  (- B + B_1)

323: \ln (t / \tau)  \nonumber

324: \\ && +  (B_2 + K_q B^2) \ln^2 (t / \tau)\nonumber

325: \\ && +     B_3 \ln^3 (t / \tau) +  B_4 \ln^4 (t / \tau)  \bigr ] \,,

326: \end{eqnarray}

327: what includes the terms of order $\varepsilon$ and neglects terms of order

328: $\varepsilon^{3/2}$. It involves corrections to the plateau values,

329: \begin{equation}\label{eq:asy:deltafq}

330: \hat{f}_q = (1 - f_q^c)  R_{q k} \left[ \hat{A}_{k }^{(0)} (\mathbf{V}) -

331: \epsilon_1 (\mathbf{V}) a_k^2 \right]\,,

332: \end{equation}

333: correction amplitudes,

334: \begin{equation}\label{eq:asy:Kq}

335: K_q =  R_{q k} \left[ A_{k k_1 k_2 }^{(2)c} a_{k_1} a_{k_2} 

336: - a_{k}^2 \right] / a_q  \,,

337: \end{equation}

338: and the prefactors,

339: \begin{subequations}\label{eq:asy:Bcoeffs}

340: \begin{equation}\label{eq:asy:Bcoeffs_B1}

341: B_1 =  (0.44425\, \zeta - 0.065381\, \mu_3)\,\varepsilon_1 (\mathbf{V})

342: - 0.22213 \,\varepsilon_2 (\mathbf{V}) \,\, ,

343: \end{equation}

344: \begin{equation}\label{eq:asy:Bcoeffs_B2}

345: B_2 = (0.91189\,\zeta + 0.068713\,\mu_3)\,\varepsilon_1 (\mathbf{V})

346: - 0.15198\,\varepsilon_2 (\mathbf{V}) \,\, ,

347: \end{equation}

348: \begin{equation}\label{eq:asy:Bcoeffs_B3}

349: B_3 = - 0.13504 \,\mu_3 \,\varepsilon_1 (\mathbf{V}) \,\, ,

350: \quad B_4 = - 0.046197 \,\mu_3 \,\varepsilon_1 (\mathbf{V}) \,\, .

351: \end{equation}

352: \end{subequations}

353: The numbers characterizing the higher-order singularities are

354: \begin{equation}\label{eq:asy:zeta}

355: \zeta = \sum_q a^*_q \left[ a_q^2 K_q +  a_q^3/2

356: \right] \,\, ,

357: \end{equation}

358: and

359: \begin{equation}\label{eq:asy:mu3}

360: \mu_3 = 2 \zeta - \sum_q a^*_q \left[ A_{q k_1 k_2 k_3}^{(3)c}

361: a_{k_1} a_{k_2} a_{k_3} + 2  A_{q k_1 k_2}^{(2)c} a_{k_1}a_{k_2}

362: K_{k_2}

363:  \right] \,\, .

364: \end{equation}

365: For the leading correction also an additional separation parameter is 

366: introduced,

367: \begin{equation}\label{eq:asy:epsilon2_A}\begin{split}

368: \varepsilon_2 (\mathbf{V}) =  \sum_q a^*_q

369:         \hat{A}_{q k}^{(1)} (\mathbf{V}) a_k

370: + 2 \varepsilon_1 (\mathbf{V}) \sum_q a^*_q a_q^2 K_q 

371: \\  +  2 \sum_q a^*_q \left[ A_{q k_1 k_2}^{(2)c} a_{k_1}

372: \hat{f}_{k_2}/(1-f_{k_2}^c) 

373: - a_q \hat{f}_q /(1-f_q^c) \right]  \,.\end{split}

374: \end{equation}

375: The time scale $\tau$ is determined by matching asymptotic approximation 

376: and numerical solution of $\phi_q(t)$ at the plateau $f_q^c$ or the 

377: rescaled plateau $f_q^c+\hat{f}_q$ for Eq.\ (\ref{eq:asy:log_decay}) and 

378: Eq.\ (\ref{eq:asy:G1G2q}), respectively.

379: 

380: %%

381: \subsection{\label{subsec:couple}Coupled Variables}

382: 

383: Inserting the asymptotic expansion of Eq.\ (\ref{eq:asy:G1G2q}) into the

384: long-time limit of Eq.\ (\ref{eq:mct:phis_col}), the approximation for the

385: tagged particle dynamics up to order $\varepsilon$ is obtained

386: \cite{Sperl2003},

387: \begin{eqnarray}\label{eq:asy:phis}

388: \phi^s_q (t) = (f_q^{s\,c} + \hat{f}^s_q) &+& h^s_q  \bigl [  (- B + B_1)

389: \ln (t / \tau)  \nonumber

390: \\ && +  (B_2 + K^s_q B^2) \ln^2 (t / \tau)\nonumber

391: \\ && +     B_3 \ln^3 (t / \tau) +  B_4 \ln^4 (t / \tau)  \bigr ]\,,

392: \end{eqnarray}

393: with the critical amplitudes $h^s_q=(1-f^{s\,c}_q)\,a_q^s$, the correction

394: amplitudes $K^s_q$ and the plateau corrections $\hat{f}^s_q$. The latter 

395: are derived from the functional~(\ref{eq:mct:Fsdef}) and the related 

396: coherent quantities by

397: \begin{subequations}\label{eq:asy:taggedTaylor}

398: \begin{equation}\label{eq:asy:taggedTaylor:asq}

399: \sum_k (\delta_{qk} - {A^{s\,c}_{q,k}})\, a_k^s = \sum_k  

400: {A^{s\,c}_{qk}} a_k\,,

401: \end{equation}

402: \begin{equation}\label{eq:asy:taggedTaylor:Ksq}\begin{split}

403: \sum_k (\delta_{qk} - {A^{s\,c}_{q,k}})\,a_k^s K^s_k = -

404: {a^s_q}^2

405: +\sum_k {A^{s\,c}_{qk}} a_k K_k\\

406: +\sum_{k,p} [

407: {A^{s\,c}_{q,kp}} a^s_k a^s_p + {A^{s\,c}_{qkp}} a_k a_p + 

408: {A^{s\,c}_{qk,p}} a_k a^s_p

409: ]\,,\end{split}

410: \end{equation}

411: \begin{equation}\label{eq:asy:taggedTaylor:fsq}

412: \sum_k (\delta_{qk} - {A^{s\,c}_{q,k}})\,a_k^s \hat{f}_k^s = 

413: -\varepsilon_1(\mathbf{V})\,{a^s_q}^2

414: + \sum_k {A^{s\,c}_{qk}}\, a_k\hat{f}_k

415: +\hat{A}^{s}_q(\mathbf{V})

416: \,.

417: \end{equation}

418: \end{subequations}

419: The derivatives with respect to the coherent and tagged particle 

420: glass-form factors  are denoted before and after the comma, respectively. 

421: The coefficients are 

422: \begin{equation}\label{eq:asy:Asqk_def}\begin{split}

423: &A^s_{q k_1 \cdots k_n, p_1 \cdots p_m} (\mathbf{V})  =  

424: \frac{1}{n!} \frac{1}{m!} (1 - {f^s_q}^c)\,

425: \frac{\partial^n \partial_m {\cal F}^s_q \left [\mathbf{V}, f_k^c 

426: ,{f^s_q}^c\right ] }{ \partial f_{k_1} \cdots\partial f_{k_n}

427: \partial f^s_{p_1} \cdots\partial f^s_{p_n} } \,

428: \\&\qquad\times

429: (1 - f_{k_1}^c) \cdots (1 - f_{k_n}^c)  

430: (1 - {f^s_{p_1}}^c) \cdots (1 - {f^s_{p_n}}^c) =\\

431: &\qquad={A^{sc}_{q k_1 \cdots k_n, p_1 \cdots p_m}} +

432: \hat{A}^{s}_{q k_1 \cdots k_n, p_1 \cdots p_m} (\mathbf{V})

433: \,\, .\end{split}

434: \end{equation}

435: 

436: Similar arguments as above yield the asymptotic expansion for the MSD up 

437: to order $\varepsilon$ ,

438: \begin{eqnarray}\label{eq:asy:MSD}

439: \frac{1}{6}\delta r^2 (t) = {r^c_{s}}\,^2 &-& {\hat{r}_{s}}^2 -  

440: h_\text{MSD}\, \bigl [  (- B + B_1)

441: \ln (t / \tau)  \nonumber

442: \\ && +  (B_2 + K_\text{MSD} B^2) \ln^2 (t / \tau)\nonumber

443: \\ && +     B_3 \ln^3 (t / \tau) +  B_4 \ln^4 (t / \tau)  \bigr ] \,,

444: \end{eqnarray}

445: with parameters

446: \begin{subequations}\label{eq:asy:MSDpar}

447: \begin{equation}\label{eq:asy:MSDpar:h}

448: h_\text{MSD} = {r_s^{c\,4}}\{{\cal F}^c_\text{MSD}[h_k,f_p^{s\,c}]+

449: {\cal F}^c_\text{MSD}[f_k^c,h_p^s]\}\,,

450: \end{equation}

451: \begin{equation}\label{eq:asy:MSDpar:K}\begin{split}

452: K_\text{MSD} = \,&r_s^{c\,4}\{{\cal F}^c_\text{MSD}[h_k,h_p^s]

453: +{\cal F}^c_\text{MSD}[h_k K_k,f_p^{s\,c}]

454: \\&\qquad+{\cal F}^c_\text{MSD}[f_k^c,h_p^s K_p^s]

455: \}/h_\text{MSD}

456: - 

457: h_\text{MSD}/r_s^{c\,2}\,,

458: \end{split}\end{equation}

459: \begin{equation}\label{eq:asy:MSDpar:df}\begin{split}

460: \hat{r}^2_{s} =\,& {r_s^{c\,4}}\{

461: {\cal F}^c_\text{MSD}[h_k\hat{f}_k,f_p^{s\,c}]

462: +{\cal F}^c_\text{MSD}[f_k^c,h_p^s\hat{f}_p^s]\\&\qquad

463: +{\cal F}^c_\text{MSD}[f_k^c,f_p^{s\,c}](\mathbf{V})

464: -{\cal F}^c_\text{MSD}[f_k^c,f_p^{s\,c}](\mathbf{V}^c)

465: \}/h_\text{MSD}\\&

466: -\varepsilon_1(\mathbf{V})h^2_\text{MSD}/r_s^{c\,2}\,.

467: \end{split}\end{equation}

468: \end{subequations}

469: For the generic liquid-glass transition, the asymptotic expansion was

470: carried out with a different convention as in Eq.\ (\ref{eq:asy:Aqk_def}),

471: however, the quantities $h_q$, $K_q$, $h^s_q$, $K^s_q$, $h_\text{MSD}$,

472: $K_\text{MSD}$, and $\zeta$ are the same in both descriptions

473: \cite{Franosch1997,Fuchs1998}.  The plateau corrections, $\hat{f}_q$,

474: $\hat{f}^s_q$, and $\hat{r}^2_{s}$ are different for liquid-glass

475: transitions and higher-order singularities. The expansions in Eqs.\

476: (\ref{eq:asy:G1G2q}), (\ref{eq:asy:phis}), and (\ref{eq:asy:MSD}) share

477: the coefficients $B$, $B_1$, $B_2$, $B_3$, and $B_4$. They differ in the

478: plateau and its correction, the critical amplitude and the correction

479: amplitude.

480: 

481: %%

482: \subsection{\label{subsec:MSD}Subdiffusive Power Law in the MSD}

483: 

484: The logarithmic decay laws shall be phrased for the MSD in a slightly

485: different form than in Eq.\ (\ref{eq:asy:MSD}). This is done in order to

486: account for the fact that the MSD is conveniently shown in a

487: double-logarithmic representation which is more sensitive to the detection

488: of power laws. The asymptotic approximation (\ref{eq:asy:MSD}) for the MSD

489: can be written as $z = a_0 + a_1\,y + a_2\,y^2+ a_3\,y^3 + a_4\, y^4$. 

490: Here, $z=\delta r^2(t)/6$ and $y=\ln(t/\tau)$ The constant term represents

491: the square of the corrected localization length,

492: $a_0=r_s^{c\,2}-\hat{r}_s^2$, the coefficients $a_1=h_\text{MSD}(B-B_1)$,

493: $a_2=-h_\text{MSD}(B_2+K_\text{MSD}B^2)$ as well as $a_3=-h_\text{MSD}B_3$

494: and $a_4-h_\text{MSD}B_4$ are the separation dependent prefactors for the

495: leading and next-to-leading order terms. This yields the expansion

496: \begin{subequations}\label{eq:asy:expox}

497: \begin{equation}\label{eq:asy:expoxexp}

498: \ln z = \ln r_s^{c\,2} - \hat{r}_s^2/r_s^{c\,2} 

499: + x' \,y + b_2\,y^2 + a_3/r_s^{c\,2}\, y^3 + a_4/r_s^{c\,2}\, y^4

500: + {\mathcal O}(\varepsilon^{3/2})\,,

501: \end{equation}

502: with

503: \begin{equation}\label{eq:asy:expoxpar}

504: x' = a_1/r_s^{c\,2}\,,\quad b_2= 

505: \frac{2 r_s^{c\,2} a_2 - a_1^2}{2 r_s^{c\,4}}\,.

506: \end{equation}

507: \end{subequations}

508: In leading order, one gets a power law for the MSD,

509: \begin{subequations}\label{eq:asy:powerlead}

510: \begin{equation}\label{eq:asy:expoxlaw}

511: \delta r^2(t)/6= r_s^{c\,2}\,(t/\tau)^{x}\,,

512: \end{equation}

513: with an exponent

514: \begin{equation}\label{eq:asy:expoxcalc}

515: x = h_\text{MSD}B/{r_s^c}\,^2\,.

516: \end{equation}

517: \end{subequations}

518: Exponent $x$ varies with the square-root in the separation parameter

519: $\varepsilon_1$, cf. Eq.\ (\ref{eq:asy:log_decay}). Including the

520: corrections of order $\varepsilon$ rescales the exponent to

521: \begin{subequations}\label{eq:asy:powercorr}

522: \begin{equation}\label{eq:asy:b1dash}

523: x' = h_\text{MSD}(B-B_1)/r_s^{c\,2}\,.

524: \end{equation}

525: and the next-to-leading order result reads

526: \begin{equation}\label{eq:asy:powercorrb2}\begin{split}

527: \delta r^2(t)/6 = (t/\tau)^{x'}&\{r_s^{c\,2}\,

528: -\hat{r}_s^2+ b_2\,r_s^{c\,2} \ln(t/\tau)^2\\&

529: +a_3\ln(t/\tau)^3+a_4\ln(t/\tau)^4\}\,.\end{split}

530: \end{equation}

531: \end{subequations}

532: 

533: %%%

534: \section{\label{sec:corr}Correlation Functions near an $A_4$-singularity}

535: 

536: Before we can apply the asymptotic expansion of Eq.\ (\ref{eq:asy:G1G2q}),

537: we need to specify the values for $\mu_3$ and $\zeta$ appearing in the

538: prefactors of Eqs.~(\ref{eq:asy:Bcoeffs}). The $A_4$-singularity is

539: characterized by $\mu_3=0$. This condition has been used to locate the

540: $A_4$-singularity at $\mathbf{V}=\mathbf{V}^*$ for the SWS by:

541: \begin{equation}\label{eq:gtd:A4.SWS.MSA2nd}

542: \varphi^* = 0.52768\,,\quad\Gamma^* = 4.4759\,,\quad \delta^* = 0.04381\,.

543: \end{equation}

544: The vanishing parameter $\mu_3$ implies a considerable simplification in

545: the preceding formulas since $B_3=B_4=0$ \cite{Goetze2002}. The deviations

546: of the control parameter values specifying the $A_4$-singularity from the

547: ones reported in Ref.~\cite{Dawson2001} originate from refined numerical

548: procedures used here and they do not exceed $6\%$. The characteristic

549: parameter was $|\mu_3|<5\cdot 10^{-4}$ at the control-parameter values

550: specified above. The parameter $\zeta$ varies regularly and is

551: $\zeta=0.122$ at the $A_4$-singularity. This is smaller than the value in

552: the HSS, $\zeta_\text{HSS}=0.269$ \cite{Franosch1997}.

553: 

554: \begin{figure}[htb] 

555: \includegraphics[width=\columnwidth]{fqA4}

556: \caption{\label{fig:swsasy:fqA4}Wave-vector dependent amplitudes 

557: characterizing the $A_4$-singularity, Eq.\ (\ref{eq:gtd:A4.SWS.MSA2nd}), 

558: for coherent and tagged particle correlators of the square-well system 

559: (SWS).

560: In the upper and middle panel the critical glass-form factors $f_q^*$,

561: Eq.\ (\ref{eq:Feq}), and the amplitudes $h_q^*$ are shown as full lines,

562: respectively. The dashed lines represent the values for $f_q^{s\,*}$, Eq.\

563: (\ref{eq:mct:fqs}), and $h_q^{s\,*}$, Eq.\

564: (\ref{eq:asy:taggedTaylor:asq}). For the hard-sphere system (HSS),

565: $f_q^{s\,c}$ and $h_q^{s}$ are shown dotted. The lower panel shows the

566: correction amplitudes $K_q^*$, Eq.\ (\ref{eq:asy:Kq}), and $K_q^{s\,*}$,

567: Eq.\ (\ref{eq:asy:taggedTaylor:Ksq}), as full and dashed lines,

568: respectively.  A square at $q=24.2$ indicates the corrections for the path

569: calculated for Fig.\ \ref{fig:swsasy:A4quadlines} and the correlators

570: shown in Fig.\ \ref{fig:swsasy:A4log}. The correction amplitudes $K_q$

571: ($-\cdot-$) and $K_q^{s}$ ($\cdots$) for the HSS are shown for comparison.  

572: The unit of length here and in the following figures is the hard-core

573: diameter $d$ of the particles.

574: }

575: \end{figure} 

576: 

577: The second prerequisite for the asymptotic description according to Eq.\

578: (\ref{eq:asy:G1G2q}) are the wave-vector dependent amplitudes $f_q^*$,

579: $h_q^*$ and $K_q^*$. These are shown for the $A_4$-singularity in Fig.\

580: \ref{fig:swsasy:fqA4} together with the related values for the

581: tagged-particle correlator, Eq.\ (\ref{eq:asy:phis}).  The quantities for

582: the tagged particle motion are close to the ones for the coherent

583: correlator $\phi_q(t)$ except for values of $q$ smaller than, say, $q=10$.

584: This difference was observed already for the HSS \cite{Fuchs1998}. Since

585: we will not be concerned with small $q$ in the following, we restrict the

586: discussion to the coherent dynamics and imply that the same is applicable

587: also to the incoherent part with only minor changes. In comparison to the

588: HSS the $f_q^*$, $h_q^*$, $f^{s\,*}_q$ and $h^{s\,*}_q$ are extended over

589: a broader $q$-range. The maximum in $h^s_q$ is shifted from $q\approx 13$

590: to $q\approx 25$ reflecting the smaller localization length in the SWS as

591: noticed before, cf. \cite{Dawson2001}. We see in the lower panel of Fig.\

592: \ref{fig:swsasy:fqA4} that the distributions of the correction amplitudes

593: $K_q$ and $K^s_q$ share that trend of becoming broader from the HSS to the

594: $A_4$-singularity of the SWS.  The zero in $K_q$ moves from around

595: $q\approx 14$ in the HSS to $q\approx 32$ in the SWS. In addition, the

596: amplitudes are shifted to lower values for small $q$.

597: 

598: \begin{figure}[htb] 

599: \includegraphics[width=\columnwidth]{A4quadlines}

600: \caption{\label{fig:swsasy:A4quadlines}Curves of vanishing quadratic 

601: correction in Eq.\ (\ref{eq:asy:G1G2q}) at the $A_4$-singularity of the 

602: SWS, $B_2(q)=0$ (dash-dotted), for $q=7.0$, $20.2$, $24.2$, $27.0$, and 

603: for $K_q=0$ as labeled. 

604: The full line shows a part of the glass-transition diagram for constant

605: $\delta=\delta^*$.  The lines of vanishing separation parameters

606: $\varepsilon_1(\mathbf{V})$ and $\varepsilon_2(\mathbf{V})$ are shown by a

607: broken and a dotted line, respectively. For the wave vector $q=24.2$, a

608: path on the curve $B_2(24.2)=0$ is marked ($+$) and labeled by $n$, for

609: which the correlators are shown in Fig.\ \ref{fig:swsasy:A4log}. State

610: $n=2$ is analyzed also in Fig.\ \ref{fig:swsasy:A4logqvar}. For the points

611: ($\blacktriangle$) labeled a, b, and c the decay is shown in Figs.\

612: \ref{fig:swsasy:A4logVvar} and \ref{fig:conclog}.

613: } 

614: \end{figure} 

615: 

616: Having specified the characteristic parameters for the $A_4$-singularity,

617: the solution at any point in the control-parameter space can be compared

618: to the asymptotic approximation as the control parameters are translated

619: into separation parameters $\varepsilon_1$ and $\varepsilon_2$. As done

620: for the schematic models in Ref.~\cite{Goetze2002}, we start by

621: determining the surfaces where the quadratic corrections in Eq.\

622: (\ref{eq:asy:G1G2q}) are zero, $B_2(q)=B_2+K_qB^2=0$. On these surfaces in

623: the control-parameter space the logarithmic decay is expected to show up

624: as straight line around the plateau $f_q^*$, as the cubic and quartic terms

625: in Eq.\ (\ref{eq:asy:G1G2q}) vanish because of $B_3=B_4=\mu_3=0$ at the

626: $A_4$-singularity, cf. Eq.\ (\ref{eq:asy:Bcoeffs_B3}). 

627: We get a different surface for each wave vector $q$ and show typical examples 

628: in Fig.\ \ref{fig:swsasy:A4quadlines} for a cut through the glass-transition

629: diagram for $\delta=\delta^*$. For $q=7.0$ one gets $K_q=-1.81$. The

630: solution of $B_2(\mathbf{V})=1.81\,B(\mathbf{V})^2$ yields the chain line

631: labeled $B_2(7.0)=0$ in Fig.\ \ref{fig:swsasy:A4quadlines} and is lying in

632: the arrested region close to the line of liquid-glass transitions. Since

633: the $K_q$ depend smoothly on $q$, the evolution of the curve where

634: $B_2(q)=0$, can be understood by inspecting the parameters $B$ and $B_2$.

635: The square $B^2$ is always positive and proportional to

636: $\varepsilon_1(\mathbf{V})$, cf. Eq.\ (\ref{eq:asy:log_decay}), therefore

637: $K_q\,B^2$ is proportional to $K_q\,|\varepsilon_1(\mathbf{V})|$ and

638: shares the sign of $K_q$. Inserting $\mu_3=0$ and $\zeta=0.1216$ into Eq.\

639: (\ref{eq:asy:Bcoeffs_B2}) yields

640: $B_2(\mathbf{V})=0.111\,\varepsilon_1(\mathbf{V})  

641: -0.152\,\varepsilon_2(\mathbf{V})$, which has to be positive to comply

642: with $B_2(q)=0$. The second separation parameter is negative,

643: $\varepsilon_2(\mathbf{V})<0$, below the dotted curve for

644: $\varepsilon_2=0$ in Fig.\ \ref{fig:swsasy:A4quadlines}. In addition, the

645: value $|\varepsilon_2(\mathbf{V})|$ on the line $B_2(7.0)=0$ is larger

646: than $|\varepsilon_1(\mathbf{V})|$ which we can also infer from the fact

647: that the line $\varepsilon_1=0$ is closer than the line $\varepsilon_2=0$.

648: We now chose a point on the line $B_2(7.0)=0$, keep the first separation

649: parameter fixed, say $\varepsilon_1=\varepsilon_1'$, and move to higher

650: values for $K_q$, e.g., for $q=20.2$ where $K_q=-0.966$. $B^2$ stays the

651: same and the term $K_qB^2$ increases. To ensure that $B_2(20.2)=0$, the

652: value $B_2(\mathbf{V})$ has to decrease. We can achieve that by moving

653: closer to the line $\varepsilon_2=0$. For fixed $\varepsilon_1'$ this

654: implies a shift to lower $\varphi$ and higher $\Gamma$. Consequently the

655: entire line, where $B_2(q)=0$, is rotating clockwise around the

656: $A_4$-singularity as $K_q$ increases. This is seen for the chain line

657: $B_2(20.2)=0$ in Fig.\ \ref{fig:swsasy:A4quadlines}. Since $K^s_q$ is

658: monotonic increasing with $q$ and $K_q$ has the same trend when neglecting

659: the small oscillations, Fig.\ \ref{fig:swsasy:fqA4}, the line

660: $B^s_2(q)=B_2(\mathbf{V}) +B(\mathbf{V})^2K^s_q=0$ also rotates clockwise

661: with increasing wave-vector $q$.

662: 

663: The variation of the lines $B_2(q)=0$ described above depends only on the

664: angle at which the lines $\varepsilon_1=0$ and $\varepsilon_2=0$ intersect

665: at the $A_4$-singularity. This intersection is in a sense generic that it

666: is shared by the close-by $A_3$-singularities of the SWS. It applies also

667: to the $A_4$-singularities of the other potentials which are similar to

668: the square well. This is so because the functionals determining the

669: separation parameters depend on quantities like the structure factors and

670: the glass-form factors which are similar for different potentials

671: \cite{Goetze2003b}. For a given wave vector $q$, the line $B_2(q)=0$ may

672: or may not lie in the liquid regime depending on $K_q$. For the SWS at

673: $\delta=\delta^*$ we get a range of $-1\lesssim K_q\lesssim 0.4$

674: corresponding to $20\lesssim q \lesssim 35$, where a line $B_2(q)=0$ is

675: found in the liquid regime. We illustrate this by adding lines for

676: $q=24.2$, $q=27.0$ and for $K_q=0$ to Fig.\ \ref{fig:swsasy:A4quadlines}.

677: The vanishing $K_q$ is corresponding to $q\approx 32.3$ yielding a line 

678: $B_2(q)=0$ still in the liquid. For $q\gtrsim 35$, the latter line rotates

679: further around the $A_4$-singularity and into the arrested regime beyond the 

680: almost horizontal line of liquid-glass transitions.

681: 

682: \begin{figure}[htb] 

683: \includegraphics[width=\columnwidth]{A4log}

684: \caption{\label{fig:swsasy:A4log}Logarithmic decay at the $A_4$-singularity 

685: in the SWS for $q=24.2$ on the path indicated in 

686: Fig.\ \ref{fig:swsasy:A4quadlines}.

687: The correlation functions are shown as full lines for the states

688: $n=1,\,2,\,3$ (see text) and at $\mathbf{V}=\mathbf{V}^*$ labeled

689: $\phi_q^*$. The horizontal line indicates the critical plateau value

690: $f_q^*$ for $q=24.2$, short lines the renormalized plateaus

691: $f_q^*+\hat{f}_q$. Broken lines show the approximation of Eq.\

692: (\ref{eq:asy:G1G2q}), $-(B-B_1)\,\ln(t/\tau)$, dotted lines the

693: approximation by Eq.\ (\ref{eq:asy:log_decay}). Filled and open symbols,

694: respectively, mark the points where the approximations deviate by $5\%$

695: from the solution. The cross indicates the time when the solution for

696: $n=3$ and the critical correlator $\phi_q^*$ differ by $5\%$. The unit of

697: time here and in the following figures is given by a short-time diffusion

698: coefficient of $D_0=1/160$.

699: } 

700: \end{figure}

701: 

702: We select a wave vector $q=24.2$ with $K_q=-0.596$ as indicated in Fig.\

703: \ref{fig:swsasy:fqA4} by a square and choose a path on the line

704: $B_2(24.2)=0$ marked by the plus symbols in Fig.\

705: \ref{fig:swsasy:A4quadlines}. For $n=1,\,2,\,3$, the control parameters

706: are $(\Gamma,\varphi)=(3.312,0.5125)$, $(4.271,0.5250)$, and

707: $(4.453,0.5274)$, respectively. The solutions are shown in Fig.\

708: \ref{fig:swsasy:A4log} together with the leading approximation, Eq.\

709: (\ref{eq:asy:log_decay}), (dotted) and the next-to-leading approximation,

710: Eq.\ (\ref{eq:asy:G1G2q}), (dashed). The time scales $\tau$ are matched at

711: the plateau $f_q^*$ for the leading approximation and at the renormalized

712: plateaus $f_q^*+\hat{f}_q$ for the first correction. We recognize that for

713: $n=3$, Eq.\ (\ref{eq:asy:G1G2q}) accounts for more than ten decades in

714: time with a relative accuracy better than $5\%$. The leading approximation is

715: acceptable on that level for nine decades. For $n=1$ two and more than one

716: decade are covered, respectively. Five and three orders of magnitude in

717: time are achieved for $n=2$. For $n=1,\,2,\,3$, the leading approximation

718: describes at least $30\%$ of the complete decay and when including the

719: correction, $65\%$ are covered on the chosen accuracy level. The distance

720: in the control parameter $\Gamma$ from the value at the $A_4$-singularity

721: is $25\%$ for $n=1$ and $4\%$ for $n=2$, so no fine-tuning was necessary

722: to obtain such large windows for the logarithmic decay. The curve $n=1$

723: requires about five decades for the complete decay which is well in the

724: reach of today's computer simulation techniques \cite{Kob2003pre}.

725: 

726: It was possible to describe part of the critical decay at an

727: $A_3$-singularity in a one-component model by the expansion in polynomials

728: in $\ln t$ at a point away from the singularity \cite{Goetze2002}. We

729: therefore compare the critical decay $\phi_q^*(t)$ with the decay for

730: $n=3$ and indicate the point at $t\approx 5000$ where both differ by $5\%$

731: in Fig.\ \ref{fig:swsasy:A4log}. With only the leading correction at our

732: disposal, a $2\%$-criterion was not fulfilled as for the one-component

733: model, where also the next-to-leading correction could be used

734: \cite{Goetze2002}. The dashed line for $n=3$ does not come closer to the

735: critical decay than $4\%$. Allowing for $5\%$, the interval from $t\approx

736: 20$ to $t\approx 4000$ could be described. However, at the

737: $A_4$-singularity the approximation in Eq.\ (\ref{eq:asy:G1G2q}) always

738: yields a straight $\ln t$-decay as approximation on the chosen path with

739: $B_2(q)=0$.  This disagrees qualitatively with the observed critical

740: decay.

741: 

742: \begin{figure}[htb] 

743: \includegraphics[width=\columnwidth]{A4logqvar}

744: \caption{\label{fig:swsasy:A4logqvar}Logarithmic decay at the 

745: $A_4$-singularity for varying wave vector $q$. 

746: The inset shows the correlation functions $\phi_q(t)$ at state $n=2$ from

747: Fig.\ \ref{fig:swsasy:A4quadlines} for wave vectors $q=4.2,\,24.2\,,$ and

748: $32.2$ from top to bottom and the short horizontal lines show the

749: corresponding critical plateau values $f_q^*$. The full panel shows the

750: same correlation functions rescaled according to $\hat{\phi}_q(t)=

751: (\phi_q(t)-f^*_q-\hat{f}_q)/h^*_q$ as full lines and labeled by the

752: respective wave vectors. Dashed lines show the asymptotic laws, Eq.\

753: (\ref{eq:asy:G1G2q}). The deviations of the approximations from the

754: solutions of $5\%$ are marked by the open symbols. Filled symbols for

755: $q=4.2$ ($\blacktriangledown$) and $q=32.2$ ($\blacksquare$) show the

756: $5\%$ deviation from the additional approximation of neglecting quadratic

757: terms in Eq.\ (\ref{eq:asy:G1G2q}) (see text).

758: } 

759: \end{figure} 

760: 

761: To identify correctly some decay that is linear in the $\phi_q(t)$ versus

762: $\log t$ diagram with the logarithmic decay predicted by the asymptotic

763: laws, Eq.\ (\ref{eq:asy:G1G2q}), we check if a different correlator with a

764: different correction amplitude $K_q$ is \textit{not} linear in $\ln t$ at

765: the same point in the control-parameter space. For a two-component model a

766: characteristic alternation of concave, linear and convex decay in $\ln t$

767: was found \cite{Goetze2002,Sperl2003}. Not both correlators could be

768: linear in $\ln t$ at the same time. For the SWS this check is performed at

769: the point $n=2$ from Fig.\ \ref{fig:swsasy:A4quadlines} by variation of

770: the wave vector $q$. For the wave vectors $q=4.2$ and $32.2$ the

771: correction amplitudes are $K_q=-1.400$ and $-0.0413$, respectively.

772: Therefore $B_2(4.2)<0$ and $B_2(32.2)>0$. We expect $\phi_q(t)$ to be

773: concave or convex, accordingly, as is demonstrated by the inset of Fig.\

774: \ref{fig:swsasy:A4logqvar}. The rescaled correlators $\hat{\phi}_q(t)$

775: displayed in the full panel allow for a more detailed analysis. We see

776: that the solutions as well as the approximations clearly exhibit increased

777: curvature for larger $q$. Since the coefficient linear in $\ln t$ is not

778: depending on $q$, cf. Eq.\ (\ref{eq:asy:G1G2q}), the middle dashed line

779: represents the leading correction to all three correlators when the

780: quadratic terms are neglected. For $q=24.2$ we observe good agreement over

781: almost 5 decades as before, cf. Fig.\ \ref{fig:swsasy:A4log}. For $q=4.2$

782: and $32.2$, however, the additional approximation reduces the range of

783: applicability to less than one decade as marked by the filled symbols.

784: Including the quadratic terms from the approximation~(\ref{eq:asy:G1G2q})

785: extends this range by half a decade to later times and to earlier times by

786: one and almost two decades for $q=4.2$ and $32.2$, respectively. The time

787: window defined by a $5\%$ deviation from the

788: approximation~(\ref{eq:asy:G1G2q}) is larger by one and two orders of

789: magnitude for $q=24.2$ than for $q=32.2$ and $q=4.2$, respectively, what

790: indicates that $q$-dependent higher-order corrections significantly

791: influence the range of applicability for the leading

792: correction~(\ref{eq:asy:G1G2q}).

793: 

794: The time scale $\tau$ in Fig.\ \ref{fig:swsasy:A4logqvar} was matched for

795: $q=24.2$, so the violation of scale universality inherent to an

796: approximation like in Eq.\ (\ref{eq:asy:G1G2q}) leads to different times

797: $\tau(q)$, where the correlators for different $q$ cross their respective

798: renormalized plateau $f_q^*+\hat{f}_q$ \cite{Goetze2002}. The

799: representation with the rescaled $\hat{\phi}_q(t)$ is particularly

800: sensitive to these deviations since the point where the plateau is crossed

801: is required to be zero, $\hat{\phi}_q(t/\tau)=0$. In Fig.\

802: \ref{fig:swsasy:A4logqvar} we see that the line crossing the zero is

803: slightly broader than a single curve. The deviations in $\tau(q)$ are

804: small enough to not exceed the numerical grid for the time axis which

805: around $\tau=2988$ is given by $\Delta t=172$.  So we interpolate to get

806: for $q=4.2,\,24.2$, and $32.2$, $\tau(q)=2899$, $2988$, and $3017$,

807: respectively. These differences do not introduce larger errors in the

808: analysis carried out above.

809: 

810: \begin{figure}[htb]

811: \includegraphics[width=\columnwidth]{A4logVvar}

812: \caption{\label{fig:swsasy:A4logVvar}Logarithmic decay at the 

813: $A_4$-singularity for the three states marked by triangles in 

814: Fig.\ \ref{fig:swsasy:A4quadlines}. 

815: The inset shows the correlation functions $\phi_q(t)$ for $q=24.2$. The

816: plateau value $f_q^*$ is indicated by the short horizontal line. The full

817: panel shows $\hat{\phi}_q(t)=(\phi_q(t)-f_q^*-\hat{f}_q)/h_q$ divided by

818: the respective values for $(B-B_1)$ at the three states specified. The

819: dashed curves show the result from Eq.\ (\ref{eq:asy:G1G2q}). Filled

820: squares and circles mark the points where curve a and c deviate by $5\%$

821: from $-\ln(t/\tau)$, respectively.  The deviation for curve~b

822: ($\triangle$) for short times is at $t/\tau\approx 10^{-4}$ and not

823: included in the figure.

824: }

825: \end{figure}

826: 

827: In order to change from convex to concave behavior we can also change the

828: control parameters. For states above the line $B_2(24.2)=0$, we expect

829: concave behavior, $B_2(q)<0$, for states below, convex decay,

830: $B_2(24.2)>0$. For a demonstration of this result, the rescaled

831: correlators $\hat{\phi}_q(t/\tau)$ at the states labeled a, b, c in Fig.\

832: \ref{fig:swsasy:A4quadlines} are divided by the prefactor $(B-B_1)$ of

833: Eq.\ (\ref{eq:asy:G1G2q}). This way the part of the decay that is linear

834: in $\ln t$ shows up as straight line with slope $-\ln 10$ in Fig.\

835: \ref{fig:swsasy:A4logVvar}. The approximations~(\ref{eq:asy:G1G2q}) are

836: shown as dashed lines for each state representing

837: $-\ln(t/\tau)+[B_2(24.2)/(B-B_1)]\ln^2(t/\tau)$. For state~b the

838: approximation is identical to $-\ln(t/\tau)$ and the solution follows that

839: line over 5 decades before $5\%$ deviation is reached. The states a and c

840: are chosen to have the same value for $B-B_1\approx 0.015$ and

841: $B_2(24.2)=\mp0.0020$, respectively. The solutions at state~a and~c follow

842: the $-\ln t$-law closely within a $5\%$ margin for two decades or one

843: decade, respectively, which is significantly less than found for state~b.

844: We can infer from Fig.\ \ref{fig:swsasy:A4quadlines} that at state~a the

845: quadratic corrections would vanish again if we went from $q=24.2$ to the

846: higher wave vector $q=27.0$. A scenario similar to the one shown in Fig.\

847: \ref{fig:swsasy:A4log} can be found.

848: 

849: The procedure outlined in Figs.~\ref{fig:swsasy:A4quadlines},

850: \ref{fig:swsasy:A4log}, \ref{fig:swsasy:A4logqvar}, and

851: \ref{fig:swsasy:A4logVvar} can be summarized as follows. From the

852: higher-order singularities there emanate surfaces in the control-parameter

853: space for a specific wave vector $\bar{q}$ where the quadratic term in

854: Eq.\ (\ref{eq:asy:G1G2q}) is zero, cf. Fig.\ \ref{fig:swsasy:A4quadlines},

855: and the decay is linear in $\ln t$. Moving closer to the singularity on

856: that surface, the window in time where the logarithmic decay is a valid

857: approximation increases, cf. Fig.\ \ref{fig:swsasy:A4log}. On a fixed

858: point on that surface the decay is concave for $q<\bar{q}$ and convex for

859: $q>\bar{q}$, cf. Fig.\ \ref{fig:swsasy:A4logqvar}. For fixed $\bar{q}$,

860: the change from concave to convex is achieved by crossing the mentioned

861: surface from above in the sense exemplified in Fig.\

862: \ref{fig:swsasy:A4logVvar}.

863: 

864: The coupled quantities share the leading asymptotic behavior of the

865: density correlators. As a consequence of the factorization theorem of MCT,

866: only the glass-form factors and the critical amplitudes $h_q$ are

867: different for the coupled quantities \cite{Goetze1985}. The leading

868: corrections imply a violation of a generalized factorization theorem.

869: These are proportional to the correction amplitude $K_q$. Since for large

870: wave vectors, say $q>10$, the quantities $f^s_q$, $h^s_q$, and $K^s_q$ are

871: close to the ones for the coherent correlator, the approximation for the

872: tagged particle correlation functions $\phi^s_q(t)$ for these large $q$ is

873: the same as for $\phi_q(t)$. So the discussion for $\phi^s_q(t)$ is

874: already exhausted by Fig.\ \ref{fig:swsasy:fqA4}. Not much could be gained

875: from repeating the discussion of the previous section for $\phi^s_q(t)$.

876: 

877: 

878: %%%

879: \section{\label{sec:MSD}Mean Squared Displacement near an $A_4$-singularity}

880: 

881: \begin{figure}[htb]

882: \includegraphics[width=\columnwidth]{MSDx}

883: \caption{\label{fig:swsasy:MSDx}Subdiffusive power law in the mean-squared 

884: displacement (MSD). 

885: The solutions for states 1, 2, and 3 in the inset are shown as full lines

886: in the full panel together with the leading (dotted) and next-to-leading

887: (dashed) approximation by Eq.~(\ref{eq:asy:MSD}). The long horizontal line

888: represents $6\,r_s^{*\,2}=0.01086$, the short horizontal lines the

889: corrections to the plateau, $6(r_s^{*\,2}-\hat{r}_s^2)$, cf. Eq.\

890: (\ref{eq:asy:MSDpar:df}). The straight full lines show the power law

891: $(t/\tau)^{x}$, Eq.\ (\ref{eq:asy:expoxlaw}), with exponents $x=0.365$,

892: $0.173$ and $0.0878$ for states $n=1,\,2,\,3$. The filled symbols show the

893: points where the solutions deviate by $5\%$ from the leading-order power

894: laws. The inset shows part of the glass-transition diagram for

895: $\delta=\delta^*$ and a chain line where $b_2=0$, 

896: cf. Eq.\ (\ref{eq:asy:expoxpar}), (see text).

897: }

898: \end{figure}

899: 

900: According to Eq.\ (\ref{eq:asy:expoxexp}), $\delta r^2(t)$ is expected to

901: exhibit power-law behavior around the plateau $6\,r_s^{c\,2}$ provided the

902: term $b_2$ vanishes. The power-law exponent $x$ is determined explicitly

903: in Eq.\ (\ref{eq:asy:expoxcalc}) by the localization length and the

904: critical amplitude, which are

905: \begin{equation}\label{eq:swsasy:A4rs}

906: r_s^{*}=0.04255\,,\quad h^*_\text{MSD}=0.004051 \,.

907: \end{equation}

908: 

909: The inset of Fig.\ \ref{fig:swsasy:MSDx} shows the line where $b_2$ from

910: Eq.\ (\ref{eq:asy:expoxexp}) vanishes. This line is almost identical to

911: the one for $B_2(24.2)=0$ shown in Fig.\ \ref{fig:swsasy:A4quadlines} for

912: the correlators $\phi_q(t)$. The MSD for three states on that line is

913: shown in the full panel. It is described well by the approximation in Eq.\

914: (\ref{eq:asy:MSD}). For states $n=1,\,2,\,3$, one, three and six decades

915: are covered with deviations less than $5\%$, so the approximation yields a

916: description of similar accuracy as for the correlation functions in Fig.\

917: \ref{fig:swsasy:A4log}. The leading result from Eq.\

918: (\ref{eq:asy:log_decay}) describes the relaxation proportional to $\ln t$

919: (dotted) which always has negative curvature in the double-logarithmic

920: representation and does not provide a valid description for $n=1$ and 2.

921: The reason for the qualitative difference between the solution for the MSD

922: and the leading logarithmic law is that the corrections proportional to

923: $K_\text{MSD}=-1.708$ are large, Therefore, $K_\text{MSD}B^2+B_2$ is never

924: close to zero in the liquid regime except very close to the

925: $A_4$-singularity. This is seen for $n=3$ in Fig.\ \ref{fig:swsasy:MSDx}

926: where $\ln t$ develops a straightened decay around the plateau.

927: 

928: The power law~(\ref{eq:asy:powerlead}) provides a different formulation of

929: a leading order approximation and is shown in Fig.\ \ref{fig:swsasy:MSDx}

930: as straight line for $n=1,\,2,\,3$.  For $n=1$ this describes the MSD for

931: more than a decade as indicated by the squares. For $n=2$ three decades

932: are covered and six decades of power-law behavior are identified for curve

933: $n=3$. So the accuracy is similar to the one provided by the approximation

934: in next-to-leading order by Eq.\ (\ref{eq:asy:MSD}). Both asymptotic

935: descriptions fall on top of each other around the plateau and therefore

936: corroborate that the reformulation~(\ref{eq:asy:expoxexp}) is justified.

937: The interpretation of the behavior of the MSD is then much simpler when

938: considering the power laws instead of the logarithms of time.  The

939: decreasing slope of the relaxation when approaching the $A_4$-singularity

940: as in Fig.\ \ref{fig:swsasy:MSDx} is just the exponent $x$ from Eq.\

941: (\ref{eq:asy:expoxcalc}) which decreases as $B$ with the square-root of

942: the separation parameter $\varepsilon_1$, cf.  Eq.\

943: (\ref{eq:asy:log_decay}). The same parameter $B$ is the prefactor of the

944: leading-order logarithmic decay in Eq.\ (\ref{eq:asy:log_decay}). In that

945: sense Fig.\ \ref{fig:swsasy:MSDx} is the analog of Fig.\

946: \ref{fig:swsasy:A4log}.

947: 

948: \begin{figure}[htb]

949: \includegraphics[width=\columnwidth]{MSDV}

950: \caption{\label{fig:swsasy:MSDV}Concave and convex deviations from the 

951: power law, Eq.\ (\ref{eq:asy:powerlead}) in the MSD. 

952: Solutions for the states a, b, and c are shown as full lines, the

953: approximation~(\ref{eq:asy:powerlead}) as straight full lines for

954: exponents $x=0.147$, $0.173$, and $0.285$, respectively. Filled symbols

955: denote the $5\%$ deviation of the solutions from the leading-order power

956: law. For state~b, the dashed line exhibits the corrected power law with

957: $x'=0.155$, Eq.\ (\ref{eq:asy:b1dash}), and the open triangle the $5\%$

958: deviations of the solution from it. Dashed lines show the approximation by

959: Eq.\ (\ref{eq:asy:powercorrb2}) for a and c with $b_2=0.00363$ and

960: $-0.00735$, respectively, and $x'=0.143$ and $0.214$. The open symbols

961: mark the $5\%$ deviations. The inset replots the one from Fig.\

962: \ref{fig:swsasy:MSDx} and shows by the crosses the state points a, b, and

963: c.

964: }

965: \end{figure}

966: 

967: The term $b_2$ in Eq.\ (\ref{eq:asy:expoxexp}) varies regularly in the

968: separation parameters $\varepsilon_1$ and $\varepsilon_2$, and $b_2$ is

969: positive above the line $b_2=0$ and negative below. Therefore, similar to

970: the case for the correlators in the linear-$\log$ representation, in the

971: double-logarithmic representation, the behavior of the MSD can be changed

972: from convex to concave when crossing the line of vanishing $b_2$. This is

973: demonstrated for three states in Fig.\ \ref{fig:swsasy:MSDV}. State~b is

974: identical to the state $n=2$ in Fig.\ \ref{fig:swsasy:MSDx} and obeys

975: $b_2=0$. The power law $(t/\tau)^{x}$ is shown as straight full line. The

976: time scale $\tau$ is matched at the plateau $6\,r_s^{*\,2}$. Moving to

977: state~c below the chain line, $(\Gamma, \varphi)=(3.42,0.525)$, a

978: relaxation is obtained which clearly exhibits negative curvature and is

979: consistent with the calculated value $b_2=-0.00735$. The leading-order

980: power law with exponent $x=0.285$ fulfills a $5\%$-deviation criterion for

981: two decades which accidentally extends to short times as the approximation

982: crosses the solution twice. Reducing the allowed deviation to $4\%$ would

983: reduce that interval to less than a decade. If we include the term

984: proportional to $b_2$ from Eq.\ (\ref{eq:asy:powercorrb2}) and renormalize

985: the exponent to $x'$, Eq.\ (\ref{eq:asy:b1dash}), the approximation agrees

986: with the solution for three decades. It is obvious from a comparison with

987: curve~1 in Fig.\ \ref{fig:swsasy:MSDx}, that the leading-order power law

988: describes that solution better than it describes the solution at state~c

989: in Fig.\ \ref{fig:swsasy:MSDV} for comparable values for $\tau$ and the

990: plateau correction $\hat{r}^2_{s}$ . The deviation to convex behavior is

991: demonstrated by the dashed line at curve~a,

992: $(\Gamma,\varphi)=(4.57,0.523)$. Again the range of validity is extended

993: to earlier times but for later times no improvement can be found.

994: 

995: In Fig.\ \ref{fig:swsasy:MSDx} the dashed line, which describes the

996: next-to-leading order approximation of Eq.\ (\ref{eq:asy:G1G2q}), deviates

997: from the leading order power law~(\ref{eq:asy:expoxlaw}) below the plateau

998: where the range of validity for the power law extends to much smaller

999: times than justified by its derivation.  We also recognize that the

1000: exponent $x$ overestimates the slope of the relaxation in

1001: Figs.~\ref{fig:swsasy:MSDx} and~\ref{fig:swsasy:MSDV}.  In Eq.\

1002: (\ref{eq:asy:expoxcalc}) only the term $B$ from the leading order

1003: approximation is present. Taking into account the renormalization of this

1004: prefactor to $B-B_1$ in Eq.\ (\ref{eq:asy:b1dash}) changes the exponent

1005: for state b from $x=0.173$ to $x'=0.155$. By comparing the full line for

1006: the leading result and the dashed line for the corrected one in Fig.\

1007: \ref{fig:swsasy:MSDV}, we find that the range of applicability is shifted

1008: to later times by one decade and extended by two decades. The corrected

1009: power law is valid from $t=10^2$ to $t=5\cdot10^6$ and comparison to Fig.\

1010: \ref{fig:swsasy:MSDx} shows that approximation~(\ref{eq:asy:G1G2q}) covers

1011: a similar range. The accidental extension to shorter times is removed. The

1012: approximation now covers the range also a naive power-law fit would yield.

1013: 

1014: In summary, the correction amplitude $K_\text{MSD}$ for the MSD does not

1015: vanish within the liquid regime. Therefore a logarithmic relaxation law

1016: can be detected only for states very close to the singularity. However,

1017: there is a line of vanishing corrections for the logarithm of the MSD.

1018: Here a logarithmic relaxation can be observed and this describes a

1019: subdiffusive power law of the MSD. We can interpret Fig.\

1020: \ref{fig:swsasy:MSDV} as the analog of Fig.\ \ref{fig:swsasy:A4logVvar}.

1021: Some quadratic correction to a leading-order linear behavior can be set to

1022: zero on a surface in control-parameter space. Departing from that surface

1023: in opposite directions introduces either positive or negative corrections

1024: and the linear behavior is changed to convex or concave.

1025: 

1026: %%%

1027: \section{\label{sec:A3}$A_3$-singularities}

1028: 

1029: \begin{figure}[htb]

1030: \includegraphics[width=\columnwidth]{fqA3}

1031: \caption{\label{fig:swsasy:fqA3}Glass-form factors $f_q^\circ$ and 

1032: $f_q^{s\,\circ}$, amplitudes $h_q^\circ$ and $h_q^{s\,\circ}$, and 

1033: correction amplitudes $K_q^\circ$ and $K_q^{s\,\circ}$ at the 

1034: $A_3$-singularity for $\delta=0.03$. 

1035: Line styles are the same as in Fig.\ \ref{fig:swsasy:fqA4}. The values at

1036: the $A_4$-singularity, $f_q^*$ (dotted), $h_q^*$ (dash-dotted), and

1037: $K_q^*$ (dash-dotted), are shown for comparison.  The values for $q=24.2$

1038: and $45.0$ are marked by diamonds. The inset shows $K_q$ for $4<q<11$ for

1039: $\delta=\delta^*$ (chain line), $0.03$ (full line) and $0.02$ (dotted

1040: line).

1041: }

1042: \end{figure}

1043: 

1044: An $A_3$-singularity is not located on a liquid-glass-transition line but

1045: is the endpoint of a glass-glass-transition line \cite{Goetze1991b}. The

1046: parameter $\mu_3$ is no longer vanishing and for $\delta=0.03$ we get

1047: $\mu_3=0.109$ and $\zeta=0.157$. For this $A_3$-singularity the

1048: $q$-dependent amplitudes are shown in Fig.\ \ref{fig:swsasy:fqA3}. No

1049: qualitative changes are obvious compared to the results shown in Fig.\

1050: \ref{fig:swsasy:fqA4} for the $A_4$. The smaller length scale

1051: $\delta=0.03$ for the attractive well introduces a smaller localization

1052: length, and this implies the broader distributions in wave-vector space.

1053: So the trend seen when changing from the HSS to the $A_4$-singularity of

1054: the SWS is continued when approaching $A_3$-singularities at smaller

1055: $\delta$. There are only two notable exceptions at smaller $q$. First, the

1056: value for $K_q$ at the position of the structure factor peak is minimal

1057: for the $A_4$, $-1.81=K_q^*<K_q^\circ=-1.72$. The inset shows this region

1058: enlarged for $\delta=\delta^*$, $0.03$ and $0.02$, demonstrating that

1059: $K_q$ at the peak is again larger for the $A_3$-singularity with smaller

1060: well width $0.02$, where $K_q=-1.69$. Second, the zero-wave-vector limit

1061: of $K^s_q$ is also smallest at the $A_4$-singularity. The respective

1062: values for $\delta=\delta^*$, $0.03$ and $0.02$ are $-1.71$, $-1.64$, and

1063: $-1.62$. Therefore, one experiences the strongest $q$-dependent

1064: corrections at the $A_4$-singularity.

1065: 

1066: \begin{figure}[htb]

1067: \includegraphics[width=\columnwidth]{A3quadlines}

1068: \caption{\label{fig:swsasy:A3quadlines}Curves of vanishing quadratic 

1069: correction in Eq.\ (\ref{eq:asy:G1G2q}) for the $A_3$-singularity

1070: ($\bigcirc$) at $\delta=0.03$. 

1071: The $\delta=0.03$ cut through the glass-transition diagram is displayed by

1072: full lines. The various lines are shown in the same style as in Fig.\

1073: \ref{fig:swsasy:A4quadlines} and labeled accordingly. The line $b_2=0$,

1074: cf. Eq.\ (\ref{eq:asy:expoxpar}), indicates the analogous line for the

1075: MSD, cf. inset of Fig.\ \ref{fig:swsasy:MSDx}.

1076: }

1077: \end{figure}

1078: 

1079: Figure~\ref{fig:swsasy:A3quadlines} shows the analog of Fig.\

1080: \ref{fig:swsasy:A4quadlines} for a cut through the glass-transition

1081: diagram at $\delta=0.03$. The lines $\varepsilon_1=0$ and

1082: $\varepsilon_2=0$ for the $A_3$-singularity are obtained from a smooth

1083: transformation of the corresponding lines at the $A_4$-singularity, and

1084: they appear in similar locations in the diagram. The line

1085: $\varepsilon_2=0$ is again very close to the almost horizontal line of

1086: transitions. Just below, we find again the line where $B_2(q)=0$ when

1087: $K_q=0$. However, this now represents $q\approx 57.5$, cf.  Fig.\

1088: \ref{fig:swsasy:fqA3}, which is a value almost twice as large as for the

1089: corresponding line in Fig.\ \ref{fig:swsasy:A4quadlines}. For the wave

1090: vector $q=24.2$ we find the line, where $B_2(24.2)=0$, completely in the

1091: glass state. Taking the same value for the correction amplitude as for

1092: $q=24.2$ at the $A_4$, $K^*_q\approx-0.6$, we obtain $q=45.0$, cf. Fig.\

1093: \ref{fig:swsasy:fqA3} and the line labeled accordingly in Fig.\

1094: \ref{fig:swsasy:A3quadlines}. Since the latter line comes close to the

1095: liquid-glass-transition line we take that as a reference and estimate the

1096: range of wave-vectors where the quadratic corrections can be put to zero

1097: in the liquid regime to $45\lesssim q \lesssim 70$. The lines where

1098: $B_2(q)=0$ can be rather sensitive to $q$-variation. This is demonstrated

1099: by the curve $B_2(46.2)=0$. Although the change in the wave vector is

1100: relatively small in comparison to $q=45.0$, the values for $K_q$ differ by

1101: more than $20\%$ for fixed $q$ and induce a rotation of the line

1102: $B_2(q)=0$ by quite a significant angle.

1103: 

1104: Having in mind the drastic changes in the lines where $B_2(q)=0$, it may

1105: come with some surprise that the line for the MSD, where $b_2=0$, stays

1106: rather robust and accessible in the liquid regime as seen in Fig.\

1107: \ref{fig:swsasy:A3quadlines}. The variation in $q$ for the amplitudes is

1108: reflected in changes of the localization lengths. For the

1109: $A_3$-singularity at $\delta=0.03$ we get

1110: \begin{equation}\label{eq:swsasy:A3rs}

1111: r_s^\circ=0.0243\,,\quad h^\circ_\text{MSD}=0.00136\,.

1112: \end{equation}

1113: From Eq.\ (\ref{eq:swsasy:A4rs}) one gets $r_s^*/r_s^\circ= 1.75$ and the

1114: square of the latter ratio, $r_s^{*\,2}/r_s^{\circ\,2}\approx 3$, is the

1115: same as $h^*_\text{MSD}/h^\circ_\text{MSD}$. Since only the fraction

1116: $h_\text{MSD}/r_s^2$ could introduce larger modifications in Eq.\

1117: (\ref{eq:asy:expoxexp}), the changes in $b_2$ cancel approximately and the

1118: line specified by $b_2=0$ experiences only minor deformations when

1119: $\delta$ is varied. The wave vector for which the lines $B_2(q)=0$ and

1120: $b_2=0$ are closest to each other, is $q=45.8$ at the $A_3$-singularity

1121: for $\delta=0.03$.

1122: 

1123: \begin{figure}[htb]

1124: \includegraphics[width=\columnwidth]{delta_var}

1125: \caption{\label{fig:swsasy:delta_var}Parameters for the asymptotic 

1126: description at the $A_3$-singularities of the SWS for varying $\delta$.

1127: Panel~A displays $\mu_3$ ($\blacktriangle$), Eq.\ (\ref{eq:asy:mu3}), and

1128: $\zeta$ ($\Diamond$), Eq.\ (\ref{eq:asy:zeta}). The dashed curve shows the

1129: asymptotic $\sqrt{\delta^*-\delta}$-law for the $\mu_3$. The localization

1130: length $r_s^\circ$ is shown in panel~B. The correction amplitudes

1131: $K_\text{MSD}^\circ$ and the ratios

1132: $h_\text{MSD}^\circ/r_\text{MSD}^{\circ\,2}$ are shown in panels~C and~D.

1133: }

1134: \end{figure}

1135: 

1136: To corroborate the finding for the MSD from the preceding paragraph, the

1137: parameters for the asymptotic description of the MSD at the

1138: $A_3$-singularities are shown in Fig.\ \ref{fig:swsasy:delta_var}. The

1139: $\mu_3$ vanish when we approach the $A_4$-singularity. The decrease close

1140: to $\delta^*$ is described asymptotically by a square-root variation,

1141: $\mu_3\propto \sqrt{\delta^*-\delta}$, shown by the dashed line

1142: \cite{Sperl2003}. The smallness of $\mu_3$ indicates that all the

1143: $A_3$-singularities are already influenced by the proximity of the

1144: close-by $A_4$-singularity. One can take advantage of this finding and

1145: conclude that the terms proportional to $\mu_3$ in Eq.\

1146: (\ref{eq:asy:G1G2q}) are small. Moreover, one may neglect $B_3$ and $B_4$

1147: in Eq.\ (\ref{eq:asy:Bcoeffs_B3}) entirely without introducing large

1148: additional errors. The leading correction to the logarithmic decay laws is

1149: then only quadratic also for the $A_3$-singularities.  Parameter $\zeta$

1150: varies regularly around a finite value at $\delta^*$ but shares the

1151: variation of $\mu_3$ at $\delta^*$ due to Eq.\ (\ref{eq:asy:mu3}). Panel~B

1152: shows the decrease of the localization length at the $A_3$-singularity

1153: when $\delta$ is reduced. A change of $40\%$ in $r^\circ_s$ from

1154: $\delta=\delta^*$ to $\delta=0.03$, cf. Eqs.~(\ref{eq:swsasy:A4rs})  

1155: and~(\ref{eq:swsasy:A3rs}), is reflected in the broadening of the

1156: distributions in $q$ seen in Figs.\ \ref{fig:swsasy:fqA4}

1157: and~\ref{fig:swsasy:fqA3}. This broadening is responsible for the large

1158: variation in $q$ when comparing Fig.\ \ref{fig:swsasy:A4quadlines} with

1159: Fig.\ \ref{fig:swsasy:A3quadlines}. It was noted in the discussion of the

1160: inset of Fig.\ \ref{fig:swsasy:fqA3} that $K_q$ introduces the strongest

1161: corrections for the correlation functions at the $A_4$-singularity. This

1162: is also true for the MSD as seen in panel~C for $K_\text{MSD}$ which is

1163: largest in absolute value at the $A_4$-singularity. The variation in

1164: $K_\text{MSD}$ with $\delta$ is however small and does not introduce

1165: significant changes to $a_2$ in Eq.\ (\ref{eq:asy:expoxpar}). The

1166: amplitude $h_\text{MSD}$ is the remaining parameter entering Eq.\

1167: (\ref{eq:asy:expoxpar}) that could alter the location of the line $b_2=0$

1168: in the glass-transition diagram. We noted above that only the ratio

1169: $h_\text{MSD}/r_s^{c\,2}$ needs to be considered which is shown in

1170: panel~D. From there one infers that the ratio varies only by less than

1171: $5\%$. We can conclude that the line of power law variation for the MSD

1172: stays in the liquid regime even when $\delta$ is changed significantly.

1173: 

1174: %%%

1175: \section{\label{sec:HCY}Hard Core Yukawa System}

1176: 

1177: The $A_l$-singularities occurring in MCT are topologically stable, smooth

1178: changes in the control parameters do not challenge their existence.

1179: Therefore, the results for the SWS can be applied also to other potentials

1180: with a short-ranged attraction. Nevertheless, the deformation of the

1181: potential might introduce changes large enough to be relevant for the

1182: detection of the higher-order singularities. Among several potentials the

1183: hard core Yukawa system (HCY) was found to differ by up to $20\%$ in

1184: certain properties at the $A_4$-singularity from the SWS

1185: \cite{Goetze2003b}. Since other potentials differ less we use that system

1186: as a second example for an $A_4$-singularity.

1187: 

1188: \begin{figure}[htb]

1189: \includegraphics[width=\columnwidth]{HCYquadlines}

1190: \caption{\label{fig:HCYquadlines}Cut through the parameter space for the 

1191: hard-core Yukawa system for $\delta=\delta^*$. 

1192: Lines styles are the same as in Fig.\ \ref{fig:swsasy:A4quadlines}. The

1193: wave vectors $q=15.0$, $21.4$, $27.0$, and $28.2$ are approximately

1194: equivalent to $q=7.0$, $20.2$, $24.2$, and $27.0$ in 

1195: Fig.\ \ref{fig:swsasy:A4quadlines}, respectively, after rescaling 

1196: $\Gamma$ by a factor of 2.98 (see text).

1197: }

1198: \end{figure}

1199: 

1200: Figure~\ref{fig:HCYquadlines} shows the analog of Fig.\

1201: \ref{fig:swsasy:A4quadlines} for the HCY. For a comparison, the

1202: $A_4$-singularity in the SWS was mapped on top of the $A_4$-singularity in

1203: the HCY by scaling in $\Gamma$ with a factor of $2.98$ and by a shift in 

1204: $\varphi$ of $0.0065$.

1205: The same transformation was applied to the lines where $B_2(q)=0$ in the

1206: SWS. Fig.\ \ref{fig:HCYquadlines} displays the $B_2(q)=0$ lines for the

1207: HCY that come closest to the ones shown in Fig.\ \ref{fig:swsasy:A4quadlines} 

1208: after the mapping. The correction amplitude $K_q$ for the HCY vanishes at 

1209: $q\approx 34$, and the range in wave vector for which $B_2(q)=0$ is lying in 

1210: the liquid regime is shifted to higher wave vectors, $21\lesssim q\lesssim 36$ 

1211: or $-0.9\lesssim K_q\lesssim 0.2$, in comparison to the SWS. 

1212: For $q=27.0$ we get the line $B_2(q)=0$ for the HCY that is closest to the 

1213: line $b_2=0$ for the MSD as compared to $B_2(24.2)=0$ in the SWS.

1214: 

1215: %%%

1216: \section{\label{sec:conclusion}Conclusion}

1217: 

1218: Logarithmic decay or, equivalently, $1/f$~noise in the fluctuation

1219: spectra, can arise in a number of situations and is explained by various

1220: approaches \cite{Weissman1988}. In the log-linear representation

1221: appropriate for the correlation functions, this decay exhibits a straight

1222: line. To discriminate the logarithmic decay laws originating from

1223: higher-order glass-transition singularities within MCT \cite{Goetze2002}

1224: from other possible scenarios one needs criteria to distinguish one from

1225: the other. The theory makes specific predictions where in the

1226: control-parameter space the logarithmic decay is expected and how the

1227: corrections introduce deviations from that behavior. In this paper, the

1228: scenarios are discussed in quantitative detail for an example relevant for

1229: studies of colloidal dynamics, the square-well system (SWS). To proceed,

1230: specific cuts through the three-dimensional parameter space are

1231: considered. Here, lines are identified where the corrections quadratic in

1232: the logarithm of time vanish for a chosen wave vector $q$, cf. Fig.\

1233: \ref{fig:swsasy:A4quadlines}. These lines emanate from the higher-order

1234: singularity and rotate clockwise around the higher-order singularity with

1235: increasing $q$. The correlation functions for states on these lines

1236: exhibit decays that are linear in the logarithm of time for several orders

1237: of magnitude in time, cf. Fig.\ \ref{fig:swsasy:A4log}. In leading order,

1238: the slope of the decay is given by the square-root of the distance from

1239: the higher-order singularity, Eq.\ (\ref{eq:asy:log_decay}).  The

1240: mean-squared displacement MSD displays a power law, Eq.\

1241: (\ref{eq:asy:expoxlaw}), that is valid on a similar line in the

1242: control-parameter space, cf. Fig.\ \ref{fig:swsasy:MSDx}. The exponent $x$

1243: of this subdiffusive behavior is also decreasing with the square-root of

1244: the distance. Both the logarithmic decay and the power law are accessible

1245: in the liquid regime. The logarithmic decay is predicted for wave vectors

1246: $q$, which are equivalent to values of about three to four times the first

1247: peak of the static structure factor.

1248: 

1249: In a semi-logarithmic representation for the correlation functions and a

1250: double-logarithmic plot of the MSD, characteristic convex and concave

1251: relaxation patterns are found when states are chosen that are off the

1252: specified lines, cf. Figs.~\ref{fig:swsasy:A4logVvar} and

1253: \ref{fig:swsasy:MSDV}. Due to the variation of the correction amplitude

1254: $K_q$ in Fig.\ \ref{fig:swsasy:fqA4}, a similar variation from convex to

1255: concave behavior is introduced by changes in the wave vector at a fixed

1256: point in control-parameter space, cf. Fig.\ \ref{fig:swsasy:A4logqvar}.

1257: These deviations from logarithmic behavior provide a test for the clear

1258: identification of dynamical scenarios that are consistent with Eq.\

1259: (\ref{eq:asy:G1G2q}) and hence originate from higher-order singularities.

1260: 

1261: When the localization at the higher-order singularity is changed by either

1262: deforming the shape of the potential or by moving to $A_3$-singularities

1263: at smaller ranges of the attraction, the logarithmic decay of the

1264: correlation functions is shifted to higher wave vectors.  Whereas the

1265: difference between the SWS and the hard-core Yukawa system at the

1266: $A_4$-singularity is modest, cf. Fig.\ \ref{fig:swsasy:A4quadlines} and

1267: Fig.\ \ref{fig:HCYquadlines}, the lines of vanishing quadratic correction

1268: change drastically at the $A_3$-singularity, cf. Fig.\

1269: \ref{fig:swsasy:A3quadlines}. In contrast, the line where the subdiffusive

1270: power law for the MSD is valid, is robust against changes of the well

1271: width and the potential shape, cf.  Figs.\ \ref{fig:swsasy:MSDx},

1272: \ref{fig:swsasy:A3quadlines}, and \ref{fig:HCYquadlines}.

1273: 

1274: \begin{figure}[htb]

1275: \includegraphics[width=\columnwidth]{conclog}

1276: \caption{\label{fig:conclog}Correlators for states a (panel a) and b 

1277: (panel b) from Fig.\ \ref{fig:swsasy:A4quadlines} for wave vectors 

1278: $q=4.2$, $20.2$, $24.2$, $27.0$, $32.2$, and $36.2$ from top to bottom. 

1279: Full lines show the solutions of the MCT equations for the SWS, dashed

1280: lines the approximation by Eq.\ (\ref{eq:asy:G1G2q}). Triangles mark the

1281: $5\%$ deviation of the correlator from the approximation for $q=27.0$ and

1282: $24.2$, respectively. The dotted vertical lines indicate the time scales

1283: $\tau$, the short horizontal lines the corrected plateau value

1284: $f_q+\hat{f}_q$ for $q=27.0$ and $24.2$, respectively.

1285: }

1286: \end{figure}

1287: 

1288: For comparing the solutions of the equations of motion, Eqs.\

1289: (\ref{eq:mct:MCT}), (\ref{eq:mct:tagged}), and (\ref{eq:mct:MSD}), with

1290: the asymptotic expansions, Eqs.\ (\ref{eq:asy:G1G2q}),

1291: (\ref{eq:asy:phis}), and (\ref{eq:asy:MSD}), all parameters are calculated

1292: explicitly except the time scale $\tau$ which is matched at the plateau.

1293: In an experiment or a computer simulation only the correlators are

1294: available directly. We show these in Fig.\ \ref{fig:conclog} for two

1295: states specified in Fig.\ \ref{fig:swsasy:A4quadlines} for different wave

1296: vectors. Since state b is closer to the $A_4$-singularity, the range of

1297: validity for the asymptotic approximation is larger than for state a.

1298: Especially the extension of the linear-$\log$ decay at some specific wave

1299: vector increases when moving closer to the singularity. As noted in

1300: connection with Fig.\ \ref{fig:swsasy:A4logqvar}, the range of validity

1301: for the approximation by Eq.\ (\ref{eq:asy:G1G2q}) may vary with $q$.  

1302: Partly for that reason a larger absolute curvature is attributed to the

1303: correlators by the approximation than a fit would do. A free fit could

1304: identify logarithmic behavior at state~b for $q=20.2$ from $t\approx 5$ to

1305: $t\approx 5\cdot 10^7$ with a deviation of at most $5\%$. In addition,

1306: fitting the correlator for $q=24.2$ also for $t\geqslant 10^5$ would yield

1307: positive curvature. Therefore, a free fit in that region of the

1308: control-parameter space tends to find the logarithmic decay at a somewhat

1309: lower wave vector than predicted by Eq.\ (\ref{eq:asy:G1G2q}). However,

1310: with a choice of the time scale $\tau$ that is reasonably close to the

1311: theoretical value, the concave and convex decay patterns can still be

1312: identified unambiguously in the correlators without invoking additional

1313: assumptions.

1314: 

1315: \begin{figure}[htb]

1316: \includegraphics[width=\columnwidth]{concfqA4}

1317: \caption{\label{fig:concfqA4}Comparison of $f_q$ and $h_q$ from the fit to 

1318: the simulation of two different states \cite{Sciortino2003pre} with the 

1319: values  $f_q^*$ and $h_q^*$ for the SWS from Fig.\ \ref{fig:swsasy:fqA4}. 

1320: For the comparison in the lower panel the theoretical values are 

1321: multiplied by $0.14$.

1322: }

1323: \end{figure}

1324: 

1325: A recent molecular dynamics study of a binary mixture of square-well

1326: particles identifies a power law with $x=0.28$ for the MSD over four

1327: decades and a related logarithmic decay of the correlation function at a

1328: wave vector $q=16.8$ \cite{Sciortino2003pre}. A scenario similar to Fig.\

1329: \ref{fig:conclog} was found for the correlation functions: Upon increasing

1330: $q$, a change from concave to convex decay is observed. For a second

1331: state, faster decay with larger prefactors for the logarithmic decay is

1332: reported together with a larger exponent, $x=0.44$, for the power law in

1333: the MSD. This finding is consistent with the assumption that this second

1334: state is further from the supposed higher-order singularity than the first

1335: state. Different from Fig.\ \ref{fig:conclog}, in the simulation $\delta$

1336: was changed to vary the distance while $\varphi$ and $\Gamma$ were kept

1337: fixed. The logarithmic decay was shifted to a higher wave vector for

1338: smaller $\delta$ \cite{Sciortino2003pre}. This is consistent with the

1339: expectation that can be inferred from Figs.\ \ref{fig:swsasy:A4quadlines}

1340: and \ref{fig:swsasy:A3quadlines} by observing, e.g., the rotation of the

1341: line $B_2(24.2)=0$. The analysis of the simulation data allowed for a fit

1342: of the values for $f_q^*$ and $h_q^*$ \cite{Sciortino2003pre}. These are

1343: shown in Fig.\ \ref{fig:concfqA4} together with the theoretical

1344: predictions for the SWS. The fitted parameters for both states almost fall

1345: on top of each other for $f_q^*$. The amplitude $h_q^*$ is deduced from

1346: the simulation data only up to some overall factor. It can be matched

1347: reasonably by a multiplication of the theoretical prediction for $h_q^*$.

1348: The extension in $q$ for the values obtained from MCT for the SWS are

1349: narrower, the width at half maximum for $f_q$ differs by $15\%$. A similar

1350: difference was observed for a binary mixture of hard spheres and agreement

1351: between theory and simulation could be improved by using the structure

1352: factor from the simulation as input to the MCT calculations

1353: \cite{Foffi2003pre}. For the amplitude $h_q^*$ the locations of the maxima

1354: disagree by $15\%$ and the width is different by $25\%$. The deviations

1355: for $q<7$ in both $f_q$ and $h_q$ can be attributed to the effects of

1356: mixing \cite{Foffi2003pre}.

1357: 

1358: In summary, scenarios for logarithmic decay near higher-order

1359: glass-transition singularities are presented in this work. Some essential

1360: predictions are supported by the results of computer simulations. This

1361: should motivate further investigations in colloidal systems with

1362: short-ranged attraction. In particular the power-law behavior for the MSD

1363: including the deviations might be accessible to experiments.

1364: 

1365: \acknowledgments

1366: I thank W.~G\"otze for valuable discussion. This work was supported by the 

1367: Deutsche Forschungsgemeinschaft Grant Go154/13-1.

1368: 

1369: \bibliographystyle{apsrev}

1370: 

1371: \begin{thebibliography}{38}

1372: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi

1373: \expandafter\ifx\csname bibnamefont\endcsname\relax

1374:   \def\bibnamefont#1{#1}\fi

1375: \expandafter\ifx\csname bibfnamefont\endcsname\relax

1376:   \def\bibfnamefont#1{#1}\fi

1377: \expandafter\ifx\csname citenamefont\endcsname\relax

1378:   \def\citenamefont#1{#1}\fi

1379: \expandafter\ifx\csname url\endcsname\relax

1380:   \def\url#1{\texttt{#1}}\fi

1381: \expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi

1382: \providecommand{\bibinfo}[2]{#2}

1383: \providecommand{\eprint}[2][]{\url{#2}}

1384: 

1385: \bibitem[{\citenamefont{Boon and Yip}(1980)}]{Boon1980}

1386: \bibinfo{author}{\bibfnamefont{J.-P.} \bibnamefont{Boon}} \bibnamefont{and}

1387:   \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Yip}},

1388:   \emph{\bibinfo{title}{Molecular Hydrodynamics}}

1389:   (\bibinfo{publisher}{McGraw-Hill}, \bibinfo{address}{New York},

1390:   \bibinfo{year}{1980}).

1391: 

1392: \bibitem[{\citenamefont{G\"otze}(1991)}]{Goetze1991b}

1393: \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}}, in

1394:   \emph{\bibinfo{booktitle}{Liquids, Freezing and Glass Transition}}, edited by

1395:   \bibinfo{editor}{\bibfnamefont{J.~P.} \bibnamefont{Hansen}},

1396:   \bibinfo{editor}{\bibfnamefont{D.}~\bibnamefont{Levesque}}, \bibnamefont{and}

1397:   \bibinfo{editor}{\bibfnamefont{J.}~\bibnamefont{Zinn-Justin}}

1398:   (\bibinfo{publisher}{North Holland}, \bibinfo{address}{Amsterdam},

1399:   \bibinfo{year}{1991}), vol. \bibinfo{volume}{Session LI (1989)} of

1400:   \emph{\bibinfo{series}{Les Houches Summer Schools of Theoretical Physics}},

1401:   pp. \bibinfo{pages}{287--503}.

1402: 

1403: \bibitem[{\citenamefont{Bengtzelius et~al.}(1984)\citenamefont{Bengtzelius,

1404:   G\"otze, and Sj\"olander}}]{Bengtzelius1984}

1405: \bibinfo{author}{\bibfnamefont{U.}~\bibnamefont{Bengtzelius}},

1406:   \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}}, \bibnamefont{and}

1407:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Sj\"olander}},

1408:   \bibinfo{journal}{J.~Phys.~C} \textbf{\bibinfo{volume}{17}},

1409:   \bibinfo{pages}{5915} (\bibinfo{year}{1984}).

1410: 

1411: \bibitem[{\citenamefont{Arnol'd}(1992)}]{Arnold1986}

1412: \bibinfo{author}{\bibfnamefont{V.~I.} \bibnamefont{Arnol'd}},

1413:   \emph{\bibinfo{title}{Catastrophe Theory}} (\bibinfo{publisher}{Springer},

1414:   \bibinfo{address}{Berlin}, \bibinfo{year}{1992}), \bibinfo{edition}{3rd} ed.

1415: 

1416: \bibitem[{\citenamefont{Franosch et~al.}(1997)\citenamefont{Franosch, Fuchs,

1417:   G\"otze, Mayr, and Singh}}]{Franosch1997}

1418: \bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Franosch}},

1419:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fuchs}},

1420:   \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}},

1421:   \bibinfo{author}{\bibfnamefont{M.~R.} \bibnamefont{Mayr}}, \bibnamefont{and}

1422:   \bibinfo{author}{\bibfnamefont{A.~P.} \bibnamefont{Singh}},

1423:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{55}},

1424:   \bibinfo{pages}{7153} (\bibinfo{year}{1997}).

1425: 

1426: \bibitem[{\citenamefont{Fuchs et~al.}(1998)\citenamefont{Fuchs, G\"otze, and

1427:   Mayr}}]{Fuchs1998}

1428: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fuchs}},

1429:   \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}}, \bibnamefont{and}

1430:   \bibinfo{author}{\bibfnamefont{M.~R.} \bibnamefont{Mayr}},

1431:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{58}},

1432:   \bibinfo{pages}{3384} (\bibinfo{year}{1998}).

1433: 

1434: \bibitem[{\citenamefont{Pusey and van Megen}(1986)}]{Pusey1986}

1435: \bibinfo{author}{\bibfnamefont{P.~N.} \bibnamefont{Pusey}} \bibnamefont{and}

1436:   \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{van Megen}},

1437:   \bibinfo{journal}{Nature} \textbf{\bibinfo{volume}{320}},

1438:   \bibinfo{pages}{340} (\bibinfo{year}{1986}).

1439: 

1440: \bibitem[{\citenamefont{van Megen}(1995)}]{Megen1995}

1441: \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{van Megen}},

1442:   \bibinfo{journal}{Transp.~Theory~Stat.~Phys.} \textbf{\bibinfo{volume}{24}},

1443:   \bibinfo{pages}{1017} (\bibinfo{year}{1995}).

1444: 

1445: \bibitem[{\citenamefont{Fabbian et~al.}(1999)\citenamefont{Fabbian, G\"otze,

1446:   Sciortino, Tartaglia, and Thiery}}]{Fabbian1999}

1447: \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{Fabbian}},

1448:   \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}},

1449:   \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Sciortino}},

1450:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Tartaglia}},

1451:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Thiery}},

1452:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{59}},

1453:   \bibinfo{pages}{R1347} (\bibinfo{year}{1999}); 

1454:   \textbf{\bibinfo{volume}{60}},

1455:   \bibinfo{pages}{2430(E)} (\bibinfo{year}{1999}).

1456: 

1457: 

1458: \bibitem[{\citenamefont{Bergenholtz and Fuchs}(1999)}]{Bergenholtz1999}

1459: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Bergenholtz}} \bibnamefont{and}

1460:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fuchs}},

1461:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{59}},

1462:   \bibinfo{pages}{5706} (\bibinfo{year}{1999}).

1463: 

1464: \bibitem[{\citenamefont{Dawson et~al.}(2001)\citenamefont{Dawson, Foffi, Fuchs,

1465:   G\"otze, Sciortino, Sperl, Tartaglia, Voigtmann, and

1466:   Zaccarelli}}]{Dawson2001}

1467: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Dawson}},

1468:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Foffi}},

1469:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fuchs}},

1470:   \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}},

1471:   \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Sciortino}},

1472:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Sperl}},

1473:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Tartaglia}},

1474:   \bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Voigtmann}},

1475:   \bibnamefont{and}

1476:   \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Zaccarelli}},

1477:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{63}},

1478:   \bibinfo{pages}{011401} (\bibinfo{year}{2001}).

1479: 

1480: \bibitem[{\citenamefont{Poon}(2002)}]{Poon2002}

1481: \bibinfo{author}{\bibfnamefont{W.~C.~K.} \bibnamefont{Poon}},

1482:   \bibinfo{journal}{J.~Phys.: Condens.~Matter} \textbf{\bibinfo{volume}{14}},

1483:   \bibinfo{pages}{R859} (\bibinfo{year}{2002}).

1484: 

1485: \bibitem[{\citenamefont{G\"otze and Sperl}(2003)}]{Goetze2003b}

1486: \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}} \bibnamefont{and}

1487:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Sperl}},

1488:   \bibinfo{journal}{J.~Phys.: Condens.~Matter} \textbf{\bibinfo{volume}{15}},

1489:   \bibinfo{pages}{S869} (\bibinfo{year}{2003}).

1490: 

1491: \bibitem[{\citenamefont{Eckert and Bartsch}(2002)}]{Eckert2002}

1492: \bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Eckert}} \bibnamefont{and}

1493:   \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Bartsch}},

1494:   \bibinfo{journal}{Phys.~Rev.~Lett.} \textbf{\bibinfo{volume}{89}},

1495:   \bibinfo{pages}{125701} (\bibinfo{year}{2002}).

1496: 

1497: \bibitem[{\citenamefont{Pham et~al.}(2002)\citenamefont{Pham, Puertas,

1498:   Bergenholtz, Egelhaaf, Moussa\"{\i}d, Pusey, Schofield, Cates, Fuchs, and

1499:   Poon}}]{Pham2002}

1500: \bibinfo{author}{\bibfnamefont{K.~N.} \bibnamefont{Pham}},

1501:   \bibinfo{author}{\bibfnamefont{A.~M.} \bibnamefont{Puertas}},

1502:   \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Bergenholtz}},

1503:   \bibinfo{author}{\bibfnamefont{S.~U.} \bibnamefont{Egelhaaf}},

1504:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Moussa\"{\i}d}},

1505:   \bibinfo{author}{\bibfnamefont{P.~N.} \bibnamefont{Pusey}},

1506:   \bibinfo{author}{\bibfnamefont{A.~B.} \bibnamefont{Schofield}},

1507:   \bibinfo{author}{\bibfnamefont{M.~E.} \bibnamefont{Cates}},

1508:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fuchs}}, \bibnamefont{and}

1509:   \bibinfo{author}{\bibfnamefont{W.~C.~K.} \bibnamefont{Poon}},

1510:   \bibinfo{journal}{Science} \textbf{\bibinfo{volume}{296}},

1511:   \bibinfo{pages}{104} (\bibinfo{year}{2002}).

1512: 

1513: \bibitem[{\citenamefont{Foffi et~al.}(2002)\citenamefont{Foffi, Dawson,

1514:   Buldyrev, Sciortino, Zaccarelli, and Tartaglia}}]{Foffi2002b}

1515: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Foffi}},

1516:   \bibinfo{author}{\bibfnamefont{K.~A.} \bibnamefont{Dawson}},

1517:   \bibinfo{author}{\bibfnamefont{S.~V.} \bibnamefont{Buldyrev}},

1518:   \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Sciortino}},

1519:   \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Zaccarelli}},

1520:   \bibnamefont{and}

1521:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Tartaglia}},

1522:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{65}},

1523:   \bibinfo{pages}{050802} (\bibinfo{year}{2002}).

1524: 

1525: \bibitem[{\citenamefont{Zaccarelli et~al.}(2002)\citenamefont{Zaccarelli,

1526:   Foffi, Dawson, Buldyrev, Sciortino, and Tartaglia}}]{Zaccarelli2002b}

1527: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Zaccarelli}},

1528:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Foffi}},

1529:   \bibinfo{author}{\bibfnamefont{K.~A.} \bibnamefont{Dawson}},

1530:   \bibinfo{author}{\bibfnamefont{S.~V.} \bibnamefont{Buldyrev}},

1531:   \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Sciortino}},

1532:   \bibnamefont{and}

1533:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Tartaglia}},

1534:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{66}},

1535:   \bibinfo{pages}{041402} (\bibinfo{year}{2002}).

1536: 

1537: \bibitem[{\citenamefont{Puertas et~al.}(2003)\citenamefont{Puertas, Fuchs, and

1538:   Cates}}]{Puertas2003}

1539: \bibinfo{author}{\bibfnamefont{A.~M.} \bibnamefont{Puertas}},

1540:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fuchs}}, \bibnamefont{and}

1541:   \bibinfo{author}{\bibfnamefont{M.~E.} \bibnamefont{Cates}},

1542:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{67}},

1543:   \bibinfo{pages}{031406} (\bibinfo{year}{2003}).

1544: 

1545: \bibitem[{\citenamefont{Puertas et~al.}(2002)\citenamefont{Puertas, Fuchs, and

1546:   Cates}}]{Puertas2002}

1547: \bibinfo{author}{\bibfnamefont{A.~M.} \bibnamefont{Puertas}},

1548:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fuchs}}, \bibnamefont{and}

1549:   \bibinfo{author}{\bibfnamefont{M.~E.} \bibnamefont{Cates}},

1550:   \bibinfo{journal}{Phys.~Rev.~Lett.} \textbf{\bibinfo{volume}{88}},

1551:   \bibinfo{pages}{098301} (\bibinfo{year}{2002}).

1552: 

1553: \bibitem[{\citenamefont{Mallamace et~al.}(2000)\citenamefont{Mallamace,

1554:   Gambadauro, Micali, Tartaglia, Liao, and Chen}}]{Mallamace2000}

1555: \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Mallamace}},

1556:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Gambadauro}},

1557:   \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{Micali}},

1558:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Tartaglia}},

1559:   \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Liao}}, \bibnamefont{and}

1560:   \bibinfo{author}{\bibfnamefont{S.-H.} \bibnamefont{Chen}},

1561:   \bibinfo{journal}{Phys.~Rev.~Lett.} \textbf{\bibinfo{volume}{84}},

1562:   \bibinfo{pages}{5431} (\bibinfo{year}{2000}).

1563: 

1564: \bibitem[{\citenamefont{Kegel and van Blaaderen}(2000)}]{Kegel2000}

1565: \bibinfo{author}{\bibfnamefont{W.~K.} \bibnamefont{Kegel}} \bibnamefont{and}

1566:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{van Blaaderen}},

1567:   \bibinfo{journal}{Science} \textbf{\bibinfo{volume}{287}},

1568:   \bibinfo{pages}{290} (\bibinfo{year}{2000}).

1569: 

1570: \bibitem[{\citenamefont{Weeks et~al.}(2000)\citenamefont{Weeks, Crocker,

1571:   Levitt, Schofield, and Weitz}}]{Weeks2000}

1572: \bibinfo{author}{\bibfnamefont{E.~R.} \bibnamefont{Weeks}},

1573:   \bibinfo{author}{\bibfnamefont{J.~C.} \bibnamefont{Crocker}},

1574:   \bibinfo{author}{\bibfnamefont{A.~C.} \bibnamefont{Levitt}},

1575:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Schofield}},

1576:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{D.~A.} \bibnamefont{Weitz}},

1577:   \bibinfo{journal}{Science} \textbf{\bibinfo{volume}{287}},

1578:   \bibinfo{pages}{627} (\bibinfo{year}{2000}).

1579: 

1580: \bibitem[{\citenamefont{Weeks and Weitz}(2002)}]{Weeks2002}

1581: \bibinfo{author}{\bibfnamefont{E.~R.} \bibnamefont{Weeks}} \bibnamefont{and}

1582:   \bibinfo{author}{\bibfnamefont{D.~A.} \bibnamefont{Weitz}},

1583:   \bibinfo{journal}{Phys.~Rev.~Lett.} \textbf{\bibinfo{volume}{89}},

1584:   \bibinfo{pages}{095704} (\bibinfo{year}{2002}).

1585: 

1586: \bibitem[{\citenamefont{G\"otze and Sperl}(2002)}]{Goetze2002}

1587: \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}} \bibnamefont{and}

1588:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Sperl}},

1589:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{66}},

1590:   \bibinfo{pages}{011405} (\bibinfo{year}{2002}).

1591: 

1592: \bibitem[{\citenamefont{Szamel and L\"owen}(1991)}]{Szamel1991}

1593: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Szamel}} \bibnamefont{and}

1594:   \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{L\"owen}},

1595:   \bibinfo{journal}{Phys.~Rev.~A} \textbf{\bibinfo{volume}{44}},

1596:   \bibinfo{pages}{8215} (\bibinfo{year}{1991}).

1597: 

1598: \bibitem[{\citenamefont{Fuchs}(1993)}]{Fuchs1993a}

1599: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fuchs}}, Ph.D. thesis,

1600:   \bibinfo{school}{TU M\"unchen} (\bibinfo{year}{1993}).

1601: 

1602: \bibitem[{\citenamefont{G\"otze and L\"ucke}(1976)}]{Goetze1976b}

1603: \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}} \bibnamefont{and}

1604:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{L\"ucke}},

1605:   \bibinfo{journal}{Phys.~Rev.~B} \textbf{\bibinfo{volume}{13}},

1606:   \bibinfo{pages}{3825} (\bibinfo{year}{1976}).

1607: 

1608: \bibitem[{\citenamefont{Sj\"ogren}(1980)}]{Sjoegren1980b}

1609: \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{Sj\"ogren}},

1610:   \bibinfo{journal}{Phys.~Rev.~A} \textbf{\bibinfo{volume}{22}},

1611:   \bibinfo{pages}{2883} (\bibinfo{year}{1980}).

1612: 

1613: \bibitem[{\citenamefont{Hansen and McDonald}(1986)}]{Hansen1986}

1614: \bibinfo{author}{\bibfnamefont{J.-P.} \bibnamefont{Hansen}} \bibnamefont{and}

1615:   \bibinfo{author}{\bibfnamefont{I.~R.} \bibnamefont{McDonald}},

1616:   \emph{\bibinfo{title}{Theory of Simple Liquids}}

1617:   (\bibinfo{publisher}{Academic Press}, \bibinfo{address}{London},

1618:   \bibinfo{year}{1986}), \bibinfo{edition}{2nd} ed.

1619: 

1620: \bibitem[{\citenamefont{Fuchs et~al.}(1991)\citenamefont{Fuchs, G\"otze,

1621:   Hofacker, and Latz}}]{Fuchs1991b}

1622: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fuchs}},

1623:   \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}},

1624:   \bibinfo{author}{\bibfnamefont{I.}~\bibnamefont{Hofacker}}, \bibnamefont{and}

1625:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Latz}},

1626:   \bibinfo{journal}{J.~Phys.: Condens.~Matter} \textbf{\bibinfo{volume}{3}},

1627:   \bibinfo{pages}{5047} (\bibinfo{year}{1991}).

1628: 

1629: \bibitem[{\citenamefont{G\"otze}(1996)}]{Goetze1996}

1630: \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}},

1631:   \bibinfo{journal}{J.~Stat.~Phys.} \textbf{\bibinfo{volume}{83}},

1632:   \bibinfo{pages}{1183} (\bibinfo{year}{1996}).

1633: 

1634: \bibitem[{\citenamefont{Cummings and Smith}(1979)}]{Cummings1979}

1635: \bibinfo{author}{\bibfnamefont{P.~T.} \bibnamefont{Cummings}} \bibnamefont{and}

1636:   \bibinfo{author}{\bibfnamefont{E.~R.} \bibnamefont{Smith}},

1637:   \bibinfo{journal}{Chem.~Phys.} \textbf{\bibinfo{volume}{42}},

1638:   \bibinfo{pages}{241} (\bibinfo{year}{1979}).

1639: 

1640: \bibitem[{\citenamefont{Sperl}(2003)}]{Sperl2003}

1641: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Sperl}}, Ph.D. thesis,

1642:   \bibinfo{school}{TU M\"unchen} (\bibinfo{year}{2003}).

1643: 

1644: \bibitem[{\citenamefont{Kob}(2003)}]{Kob2003pre}

1645: \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{Kob}}, in

1646:   \emph{\bibinfo{booktitle}{Slow Relaxations and Nonequilibrium Dynamics in

1647:   Condensed Matter}}, edited by \bibinfo{editor}{\bibfnamefont{J.-L.}

1648:   \bibnamefont{Barrat}},

1649:   \bibinfo{editor}{\bibfnamefont{M.}~\bibnamefont{Feigelman}},

1650:   \bibnamefont{and} \bibinfo{editor}{\bibfnamefont{J.}~\bibnamefont{Kurchan}}

1651:   (\bibinfo{publisher}{Springer}, \bibinfo{address}{Berlin},

1652:   \bibinfo{year}{2003}), vol. \bibinfo{volume}{Session LXXVII (2002)} of

1653:   \emph{\bibinfo{series}{Les Houches Summer Schools of Theoretical Physics}}.

1654: 

1655: \bibitem[{\citenamefont{G\"otze}(1985)}]{Goetze1985}

1656: \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}},

1657:   \bibinfo{journal}{Z.~Phys.~B} \textbf{\bibinfo{volume}{60}},

1658:   \bibinfo{pages}{195} (\bibinfo{year}{1985}).

1659: 

1660: \bibitem[{\citenamefont{Weissman}(1988)}]{Weissman1988}

1661: \bibinfo{author}{\bibfnamefont{M.~B.} \bibnamefont{Weissman}},

1662:   \bibinfo{journal}{Rev.~Mod.~Phys.} \textbf{\bibinfo{volume}{60}},

1663:   \bibinfo{pages}{537} (\bibinfo{year}{1988}).

1664: 

1665: \bibitem[{\citenamefont{Sciortino et~al.}(2003)\citenamefont{Sciortino,

1666:   Tartaglia, and Zaccarelli}}]{Sciortino2003pre}

1667: \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Sciortino}},

1668:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Tartaglia}},

1669:   \bibnamefont{and}

1670:   \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Zaccarelli}}

1671:   (\bibinfo{year}{2003}), \eprint{cond-mat/0304192}.

1672: 

1673: \bibitem[{\citenamefont{Foffi et~al.}(2003)\citenamefont{Foffi, G\"otze,

1674:   Sciortino, Tartaglia, and Voigtmann}}]{Foffi2003pre}

1675: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Foffi}},

1676:   \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{G\"otze}},

1677:   \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Sciortino}},

1678:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Tartaglia}},

1679:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Voigtmann}}

1680:   (\bibinfo{year}{2003}), \eprint{preprint}.

1681: 

1682: \end{thebibliography}

1683: 

1684: \end{document}

1685: 

1686: