1: \documentclass[12pt]{article}

2: \catcode`\@=11

3: \topmargin 0pt

4: \oddsidemargin 0pt

5: \headheight 0pt

6: \headsep 0pt

7: \textheight 9in

8: \textwidth 6.25in

9: \marginparwidth 0.875in

10: \parskip 4pt plus 1pt

11: \def\numberbysection{\@addtoreset{equation}{section}

12: \def\theequation{\thesection.\arabic{equation}}}

13: \numberbysection

14: \def\baselinestretch{1.2}

15: \newcommand{\abs}[1]{\vert#1\vert}

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

17: \newcommand{\beqa}{\begin{eqnarray}}

18: \newcommand{\comport}[2]{\mathrel{\mathop{#1}\limits_{#2}^{}}}

19: \renewcommand{\d}{{\rm d}}

20: \newcommand{\dq}{\!{\frac{\d^D\q}{(2\pi)^D}}}

21: \newcommand{\dpar}{\partial}

22: \newcommand{\e}{{\rm e}}

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

24: \newcommand{\eeqa}{\end{eqnarray}}

25: \newcommand{\eq}{{\rm eq}}

26: \newcommand{\eps}{\varepsilon}

27: \newcommand{\eqcr}{{\rm eq,c}}

28: \newcommand{\fd}{fluc\-tu\-a\-tion-dis\-si\-pa\-tion }

29: \newcommand{\frad}[2]{\displaystyle{\displaystyle#1\over\displaystyle#2}}

30: \newcommand{\F}{{\rm F}}

31: \renewcommand{\H}{{\cal H}}

32: \renewcommand{\i}{{\rm i}}

33: \renewcommand{\L}{{\rm L}}

34: \newcommand{\mf}{{\rm MF}}

35: \newcommand{\mean}[1]{\langle#1\rangle}

36: \newcommand{\qea}{q_{\rm EA}}

37: \newcommand{\reg}{{\rm reg}}

38: \newcommand{\sg}{{\rm sg}}

39: \newcommand{\taueq}{\tau_{\rm eq}}

40: \newcommand{\TRM}{{\rm TRM}}

41: \newcommand{\q}{{\bf q}}

42: \newcommand{\0}{{\bf 0}}

43: \newcommand{\x}{{\bf x}}

44: \newcommand{\X}{{\cal X}}

45: \newcommand{\y}{{\bf y}}

46: \newcommand{\z}{Z}

47: \begin{document}

48: \centerline{\Large\bf Response of non-equilibrium systems at criticality:}

49: \vspace{.3cm}

50: \centerline{\Large\bf Ferromagnetic models in dimension two and above}

51: \vspace{1cm}

52: \centerline{\large

53: by C.~Godr\`eche$^{a,}$\footnote{godreche@spec.saclay.cea.fr}

54: and J.M.~Luck$^{b,}$\footnote{luck@spht.saclay.cea.fr}}

55: \vspace{1cm}

56: \centerline{$^a$Service de Physique de l'\'Etat Condens\'e,

57: CEA Saclay, 91191 Gif-sur-Yvette cedex, France}

58: \vspace{.1cm}

59: \centerline{$^b$Service de Physique Th\'eorique,

60: CEA Saclay, 91191 Gif-sur-Yvette cedex, France}

61: \vspace{1cm}

62: \begin{abstract}

63: We study the dynamics of ferromagnetic spin systems quenched

64: from infinite temperature to their critical point.

65: We show that these systems are aging in the long-time regime,

66: i.e., their two-time autocorrelation

67: and response functions and associated \fd ratio

68: are non-trivial scaling functions of both time variables.

69: This is exemplified by the exact analysis of the spherical model

70: in any dimension $D>2$,

71: and by numerical simulations on the two-dimensional Ising model.

72: We show in particular that, for $1\ll s$ (waiting time)

73: $\ll t$ (observation time),

74: the \fd ratio possesses a non-trivial limit value $X_\infty$,

75: which appears as a dimensionless amplitude ratio,

76: and is therefore a novel universal characteristic of non-equilibrium

77: critical dynamics.

78: For the spherical model, we obtain $X_\infty=1-2/D$ for $2<D<4$,

79: and $X_\infty=1/2$ for $D>4$ (mean-field regime).

80: For the two-dimensional Ising model we measure $X_\infty\approx0.26\pm0.01$.

81: \end{abstract}

82: \vfill

83: \noindent To be submitted for publication to Journal of Physics A

84: \hfill S/00/006

85: \vskip -6pt

86: \noindent P.A.C.S.: 02.50.Ey, 05.40.+j, 61.43.Fs, 75.50.Lk

87: \hfill T/00/004

88: \newpage

89: 

90: \section{Introduction}

91: 

92: Consider a ferromagnetic model, without quenched randomness,

93: evolving from a disordered initial state, according to some dynamics

94: at fixed temperature~$T$.

95: In the high-temperature paramagnetic phase $(T>T_c)$,

96: the system relaxes exponentially to equilibrium.

97: At equilibrium, two-time quantities such as the

98: autocorrelation function $C(t,s)$ or the response function $R(t,s)$

99: only depend on the time difference $\tau=t-s$,

100: where~$s$ (waiting time) is smaller than~$t$ (observation time),

101: and both quantities are simply related to each other by the \fd theorem

102: \beq

103: R_\eq(\tau)=-\frac{1}{T}\,\frac{\d C_\eq(\tau)}{\d\tau}.

104: \label{fdt}

105: \eeq

106: In the low-temperature phase $(T<T_c)$ the system undergoes phase ordering.

107: In this non-equilibrium situation, $C(t,s)$ and $R(t,s)$

108: are non-trivial functions of both time variables,

109: which only depend on their ratio at late times,

110: i.e., in the self-similar domain growth (or coarsening) regime~\cite{bray}.

111: This behavior is usually referred to as aging~\cite{revue}.

112: Moreover, no such simple relation as eq.~(\ref{fdt}) holds

113: between correlation and response, i.e.,

114: $R(t,s)$ and $\dpar C(t,s)/\dpar s$ are no longer proportional.

115: It is then natural to characterize the distance to equilibrium

116: of an aging system by the so-called \fd ratio~\cite{revue,ckp,ck}

117: \beq

118: X(t,s)=\frac{T\,R(t,s)}{\,\frad{\dpar C(t,s)}{\dpar s}\,}.

119: \label{dx}

120: \eeq

121: In recent years,

122: several works~\cite{revue,ckp,ck,cd,x1,barrat,berthier,autres,zkh}

123: have been devoted to the study of the \fd ratio

124: for systems exhibiting domain growth,

125: or for aging systems such as glasses and spin glasses,

126: showing that in the low-temperature phase

127: $X(t,s)$ turns out to be a non-trivial function of its two arguments.

128: In particular, for domain-growth systems,

129: analytical and numerical studies indicate that the limit \fd ratio,

130: \beq

131: X_\infty=\lim_{s\to\infty}\lim_{t\to\infty}X(t,s),

132: \label{xinfdef}

133: \eeq

134: vanishes throughout the low-temperature phase~\cite{x1,barrat,berthier}.

135: 

136: However, to date, only very little attention has been devoted

137: to the response function $R(t,s)$, and

138: \fd ratio $X(t,s)$, for non-equilibrium systems {\it at criticality}.

139: {}From now on, we will only have in mind ferromagnetic systems

140: without quenched randomness.

141: For instance one may wonder whether there exists,

142: for a given model, a well-defined limit $X_\infty$ at $T=T_c$,

143: different from its trivial value $X_\infty=0$ in the low-temperature phase,

144: and to what extent $X_\infty$ is universal.

145: Indeed a priori, for a system such as a ferromagnet,

146: quenched from infinitely high temperature to its critical point,

147: the limit \fd ratio $X_\infty$ at $T=T_c$

148: (if it exists) may take any value between $X_\infty=1$ ($T>T_c$: equilibrium)

149: and $X_\infty=0$ ($T<T_c$: domain growth).

150: 

151: The only cases of critical systems for which the \fd ratio has been

152: considered are, to our knowledge, the models of ref.~\cite{ckp}

153: (random walk, free Gaussian field,

154: and two-dimensional X-Y model at zero temperature)

155: which share the limit \fd ratio $X_\infty=1/2$,

156: and the backgammon model, a mean-field model for which $T_c=0$,

157: where it has been shown that $X_\infty=1$,

158: up to a large logarithmic correction,

159: for both energy fluctuations and density fluctuations~\cite{fr97,gl99}.

160: 

161: In a recent companion paper~\cite{gl}, we have determined the \fd ratio

162: $X(t,s)$ for the Glauber-Ising chain, another model for which $T_c=0$.

163: In particular, the limit \fd ratio was found to be $X_\infty=1/2$

164: (see also ref.~\cite{zan}).

165: In the present work we investigate the non-equilibrium correlation

166: and response functions and the associated \fd ratio

167: in generic ferromagnetic models at their critical point.

168: We first present (in section~2) an analytical study of the spherical model

169: in arbitrary dimension.

170: We then turn (in section~3) to a scaling analysis of the generic case,

171: and to numerical simulations on the two-dimensional Ising model.

172: One salient outcome of these joint works is the realization

173: that the limit \fd ratio $X_\infty$

174: is a novel universal characteristic of critical dynamics,

175: intrinsically related to non-equilibrium initial situations.

176: 

177: The present paper is written in a self-contained fashion.

178: For the spherical model,

179: though our main intention lies in the study of non-equilibrium

180: dynamics at the critical point,

181: we shall present the three situations $T>T_c$, $T=T_c$, and $T<T_c$

182: in parallel.

183: The latter case has already been the subject of a number of investigations,

184: for both correlation and response~\cite{cd,zkh,bray}.

185: Results on the scaling behavior of the two-time autocorrelation

186: function at $T_c$ can be found in ref.~\cite{janssen}.

187: For the two-dimensional Ising model,

188: several numerical works have already been devoted

189: to its non-equilibrium dynamics in the low-temperature phase,

190: concerning both correlations~\cite{bray,fisher} and response~\cite{barrat}.

191: We will therefore restrict our numerical study to the dynamics at the

192: critical point.

193: 

194: \section{The spherical model}

195: 

196: \subsection{Langevin dynamics}

197: 

198: The ferromagnetic spherical model was introduced by

199: Berlin and Kac~\cite{berlin}, as an attempt to simplify the Ising model.

200: It is solvable in any dimension, yet possesses non-trivial critical

201: properties~\cite{berlin,baxter}.

202: Consider a lattice of points of arbitrary dimension~$D$,

203: chosen to be hypercubic for simplicity, with unit lattice spacing.

204: The spins $S_\x$, situated at the lattice vertices $\x$,

205: are real variables subject to the constraint

206: \beq

207: \sum_\x S_\x^2=N,

208: \label{constr}

209: \eeq

210: where~$N$ is the number of spins in the system.

211: The Hamiltonian of the model reads

212: \beq

213: \H=-\sum_{(\x,\y)}S_\x S_\y,

214: \label{ham}

215: \eeq

216: where the sum runs over pairs of neighboring sites.

217: 

218: Throughout the following, we assume that the system is homogeneous,

219: i.e., invariant under spatial translations.

220: This holds for a finite sample with periodic boundary conditions,

221: and (at least formally) for the infinite lattice.

222: We also assume that the initial state of the system

223: at $t=0$ is the infinite-temperature equilibrium state.

224: This state is fully disordered, in the sense that spins are uncorrelated.

225: The dynamics of the system is given by

226: the stochastic differential Langevin equation

227: \beq

228: \frac{\d S_\x}{\d t}=\sum_{\y(\x)}S_\y-\lambda(t)S_\x+\eta_\x(t).

229: \label{lan}

230: \eeq

231: The first term, where $\y(\x)$ denotes the $2D$ first neighbors

232: of the site $\x$, is equal to the gradient $-\dpar\H/\dpar S_\x$,

233: while $\lambda(t)$ is a Lagrange multiplier

234: ensuring the constraint~(\ref{constr}), which we choose to parameterize as

235: \beq

236: \lambda(t)=2D+z(t),

237: \label{zdef}

238: \eeq

239: and $\eta_\x(t)$ is a Gaussian white noise with correlation

240: \beq

241: \mean{\eta_\x(t)\eta_\y(t')}=2T\,\delta_{\x,\y}\,\delta(t-t').

242: \eeq

243: 

244: Equation~(\ref{lan}) can be solved in Fourier space.

245: Defining the spatial Fourier transform by the formulas

246: \beq

247: f^\F(\q)=\sum_\x f_\x\,\e^{-\i\q.\x},\qquad

248: f_\x=\int\dq\,f^\F(\q)\,\e^{\i\q.\x},

249: \eeq

250: where

251: \beq

252: \int\dq=\int_{-\pi}^{\pi}\frac{\d q_1}{2\pi}

253: \cdots\int_{-\pi}^{\pi}\frac{\d q_D}{2\pi}

254: \eeq

255: is the normalized integral over the first Brillouin zone, we obtain

256: \beq

257: \frac{\dpar S^\F(\q,t)}{\dpar t}=-[\omega(\q)+z(t)]S^\F(\q,t)+\eta^\F(\q,t),

258: \label{lanq}

259: \eeq

260: where

261: \beq

262: \omega(\q)=2\sum_{a=1}^D(1-\cos q_a)\comport{\approx}{\q\to\0}\q^2,

263: \label{omdef}

264: \eeq

265: and

266: \beq

267: \mean{\eta^\F(\q,t)\eta^\F(\q',t')}

268: =2T\,(2\pi)^D\,\delta^D(\q+\q')\,\delta(t-t').

269: \label{etavar}

270: \eeq

271: The solution to eq.~(\ref{lanq}) reads

272: \beq

273: S^\F(\q,t)=\e^{-\omega(\q)t-\z(t)}

274: \left(S^\F(\q,t=0)+\int_0^t\e^{\omega(\q)t_1+\z(t_1)}\eta^\F(\q,t_1)

275: \,\d t_1\right),

276: \label{sq}

277: \eeq

278: with

279: \beq

280: \z(t)=\int_0^t z(t_1)\,\d t_1.

281: \eeq

282: 

283: \subsection{Equal-time correlation function}

284: 

285: Our first goal is to compute the equal-time correlation function

286: \beq

287: C_{\x-\y}(t)=\mean{S_\x(t)S_\y(t)},

288: \label{crt}

289: \eeq

290: which is a function of the separation $\x-\y$, by translational invariance.

291: We have in particular

292: \beq

293: C_\0(t)=\mean{S_\x(t)^2}=1,

294: \label{c0t}

295: \eeq

296: because of the spherical constraint~(\ref{constr}), and

297: \beq

298: C_\x(t=0)=\delta_{\x,\0},

299: \label{cinit}

300: \eeq

301: reflecting the absence of correlations in the initial state.

302: In eq.~(\ref{crt}), the brackets denote the average

303: over the ensemble of infinite-temperature initial configurations

304: and over the thermal histories (realizations of the noise).

305: 

306: In Fourier space the equal-time correlation function is defined by

307: \beq

308: \mean{S^\F(\q,t)S^\F(\q',t)}=(2\pi)^D\,\delta^D(\q+\q')\,C^\F(\q,t).

309: \label{cqdef}

310: \eeq

311: Using the expression~(\ref{sq}),

312: averaging it over the white noise $\eta^\F(\q,t)$

313: with variance given by eq.~(\ref{etavar}), and imposing the condition

314: \beq

315: C^\F(\q,t=0)=1

316: \eeq

317: implied by eq.~(\ref{cinit}), we obtain

318: \beq

319: C^\F(\q,t)=\e^{-2\omega(\q)t-2\z(t)}

320: \left(1+2T\int_0^t\e^{2\omega(\q)t_1+2\z(t_1)}\,\d t_1\right).

321: \label{cq}

322: \eeq

323: 

324: At this point, we are naturally led to introduce two functions, $f(t)$ and

325: $g(T,t)$, which play a central role in the following developments.

326: 

327: The function $f(t)$ is explicitly given by

328: \beq

329: f(t)=\int\dq\,\e^{-2\omega(\q)t}=\left(\e^{-4t}I_0(4t)\right)^D

330: \comport{\approx}{t\to\infty}(8\pi t)^{-D/2},

331: \label{fdef}

332: \eeq

333: where

334: \beq

335: I_0(z)=\int\frac{\d q}{2\pi}\,\e^{z\cos q}

336: \comport{\approx}{z\to\infty}\,(2\pi z)^{-1/2}\,\e^z

337: \eeq

338: is the modified Bessel function.

339: 

340: The function

341: \beq

342: g(T,t)=\e^{2\z(t)}

343: \label{gdef}

344: \eeq

345: is related to $f(t)$ by the constraint~(\ref{c0t}), namely

346: \beq

347: \int\dq\,C^\F(\q,t)

348: =\frac{1}{g(T,t)}\left(f(t)+2T\int_0^tf(t-t_1)g(T,t_1)\,\d t_1\right)=1,

349: \label{constrq}

350: \eeq

351: which yields a linear Volterra integral equation for $g(T,t)$~\cite{cd}, namely

352: \beq

353: g(T,t)=f(t)+2T\int_0^tf(t-t_1)g(T,t_1)\,\d t_1.

354: \eeq

355: This equation can be solved using temporal Laplace transforms, denoted by

356: \beq

357: f^\L(p)=\int_0^\infty f(t)\,\e^{-pt}\,\d t.

358: \eeq

359: We obtain

360: \beq

361: g^\L(T,p)=\frac{f^\L(p)}{1-2Tf^\L(p)},

362: \label{fglap}

363: \eeq

364: with

365: \beq

366: f^\L(p)=\int\dq\,\frac{1}{p+2\omega(\q)}.

367: \label{fldef}

368: \eeq

369: The dependence of $g^\L(T,p)$ on temperature appears explicitly in

370: eq.~(\ref{fglap}).

371: 

372: We now present an analysis of the long-time behavior of the function $g(T,t)$,

373: considering successively the paramagnetic phase ($T>T_c$),

374: the ferromagnetic phase ($T<T_c$), and the critical point ($T=T_c$).

375: To do so, we shall extensively utilize eq.~(\ref{fglap}).

376: We therefore investigate first the function $f^\L(p)$,

377: as given in eq.~(\ref{fldef}).

378: This function has no closed-form expression, except in one and two dimensions:

379: \beqa

380: D=1:&&f^\L(p)=\frac{1}{\sqrt{p(p+8)}},\cr

381: D=2:&&f^\L(p)=\frac{2}{\pi\abs{p+8}}

382: \,{\bf K}\!\left(\frac{8}{\abs{p+8}}\right),

383: \eeqa

384: where ${\bf K}$ is the complete elliptic integral.

385: Together with the definition~(\ref{omdef}),

386: eq.~(\ref{fldef}) implies that $f^\L(p)$ is analytic in the complex $p$-plane

387: cut along the real interval $[-8D,0]$.

388: The behavior of $f^\L(p)$ in the vicinity of the branch point at $p=0$

389: can be analyzed heuristically as follows.

390: The asymptotic behavior of $f(t)$ given in eq.~(\ref{fdef})

391: suggests that its Laplace transform has a universal singular part:

392: \beq

393: f^\L_\sg(p)\comport{\approx}{p\to0}(8\pi)^{-D/2}\Gamma(1-D/2)p^{D/2-1},

394: \label{fsg}

395: \eeq

396: while there also exists a regular part of the form

397: \beq

398: f^\L_\reg(p)=A_1-A_2p+A_3p^2+\cdots,

399: \label{freg}

400: \eeq

401: where

402: \beq

403: A_k=\int\dq\,\frac{1}{(2\omega(\q))^k}

404: \eeq

405: are non-universal (lattice-dependent) numbers,

406: given in terms of integrals which are convergent for $D-2k>0$.

407: For instance $A_1$ only exists for $D>2$, and so on.

408: Equations~(\ref{fsg}) and~(\ref{freg}) jointly determine

409: the small-$p$ behavior of $f^\L(p)$, as a function of the dimensionality~$D$:

410: \beqa

411: D<2:&&f^\L(p)\approx(8\pi)^{-D/2}\Gamma(1-D/2)p^{-(1-D/2)},\cr

412: 2<D<4:&&f^\L(p)\approx A_1-(8\pi)^{-D/2}\abs{\Gamma(1-D/2)}p^{D/2-1},\cr

413: D>4:&&f^\L(p)\approx A_1-A_2p.

414: \label{flist}

415: \eeqa

416: These expressions can be justified by

417: more systematic studies~(see e.g. ref.~\cite{lprb}):

418: $f^\L(p)$ possesses an asymptotic expansion

419: involving only powers of the form $p^n$ and $p^{D/2-1+n}$, for $n=0,1,\dots$

420: Whenever $D=2,4,\dots$ is an even integer,

421: the two sequences of exponents merge, giving rise to logarithmic corrections,

422: which shall be discarded throughout the following.

423: 

424: In low enough dimension ($D<2$), $f^\L(p)$ diverges as $p\to0$.

425: As a consequence, for any finite temperature,

426: $g^\L(T,p)$ has a pole at some positive value of~$p$, denoted by $1/\taueq$,

427: away from the cut of $f^\L(p)$.

428: Hence

429: \beq

430: g(T,t)\comport{\sim}{t\to\infty}\e^{t/\taueq},

431: \label{ghigh}

432: \eeq

433: and therefore, as further analyzed below,

434: the system relaxes exponentially fast to equilibrium,

435: with a finite relaxation time $\taueq$.

436: The latter diverges as the zero-temperature phase transition is approached, as

437: \beq

438: \taueq\comport{\approx}{T\to0}

439: \left(2(8\pi)^{-D/2}\Gamma(1-D/2)T\right)^{-2/(2-D)}.

440: \eeq

441: 

442: In high enough dimension ($D>2$), $f^\L(p=0)=A_1$ is finite,

443: so that the pole of $g^\L(T,p)$ hits the cut of $f^\L(p)$

444: at $p=0$ at a finite critical temperature

445: \beq

446: T_c=\frac{1}{2A_1}=\left(\int\dq\,\frac{1}{\omega(\q)}\right)^{-1}.

447: \label{tc}

448: \eeq

449: As $T\to T_c^+$, the relaxation time $\taueq$ diverges according to

450: \beqa

451: 2<D<4:&&\taueq\comport{\approx}{T\to

452: T_c^+}\left(\frac{2(8\pi)^{-D/2}\abs{\Gamma(1-D/2)}T_c^2}

453: {T-T_c}\right)^{2/(D-2)},\cr\cr

454: D>4:&&\taueq\comport{\approx}{T\to T_c^+}\frac{2A_2T_c^2}{T-T_c}.

455: \label{taueqas}

456: \eeqa

457: \setcounter{footnote}{0}

458: Note that these equations can be recast into the form

459: $\taueq\sim(T-T_c)^{-\nu z_c}$,

460: where $\nu$ is the critical exponent of the correlation length,

461: equal to $1/(D-2)$ for $2<D<4$ and to $1/2$ for $D>4$~\cite{baxter},

462: while $z_c$ is the dynamic critical exponent, equal to 2 in the present

463: case.\footnote{A summary of the values of static and dynamical exponents

464: appearing in this work is given in Table~1.}

465: 

466: We now discuss the asymptotic behavior of the function $g(T,t)$

467: according to temperature.

468: Throughout the following, we will assume that $D>2$,

469: so that the model has a ferromagnetic transition at a finite $T_c$,

470: given by eq.~(\ref{tc}).

471: 

472: \begin{itemize}

473: 

474: \item In the paramagnetic phase $(T>T_c)$,

475: $g(T,t)$ still grows exponentially, according to eq.~(\ref{ghigh}).

476: 

477: \item In the ferromagnetic phase $(T<T_c)$,

478: a careful analysis of eq.~(\ref{fglap}) yields

479: \beq

480: g(T,t)\comport{\approx}{t\to\infty}\frac{f(t)}{M_\eq^4}

481: \approx\frac{(8\pi t)^{-D/2}}{M_\eq^4},

482: \label{glow}

483: \eeq

484: where the spontaneous magnetization $M_\eq$ is given by~\cite{baxter}

485: \beq

486: M_\eq^2=1-\frac{T}{T_c}.

487: \eeq

488: 

489: \item At the critical point $(T=T_c)$, we obtain

490: \beqa

491: 2<D<4:&&g(T_c,t)\comport{\approx}{t\to\infty}

492: (D-2)(8\pi)^{D/2-1}\sin[(D-2)\pi/2]\frac{t^{-(2-D/2)}}{T_c^2},\cr

493: D>4:&&g(T_c,t)\comport{\to}{t\to\infty}\frac{1}{4A_2T_c^2}.

494: \label{gcrit}

495: \eeqa

496: 

497: \end{itemize}

498: 

499: Finally, eqs.~(\ref{fglap}), (\ref{fldef}), (\ref{flist}),

500: and~(\ref{tc}) yield the following identities:

501: \beqa

502: \int_0^\infty f(t)\,\d t&=&\frac{1}{2T_c},\cr\cr

503: \int_0^\infty f(t)\,\e^{-t/\taueq}\,\d t&=&\frac{1}{2T}

504: {\hskip 1.8cm}(T>T_c),\cr\cr

505: \int_0^\infty g(T,t)\,\d t&=&\frac{1}{2T_cM_\eq^2}

506: {\hskip 1cm}(T<T_c).

507: \label{fgid}

508: \eeqa

509: 

510: We are now in a position to discuss the temporal behavior of

511: the equal-time correlation function in the different phases.

512: Its expression~(\ref{cq}) in Fourier space reads

513: \beq

514: C^\F(\q,t)=\frac{\e^{-2\omega(\q)t}}{g(T,t)}

515: \left(1+2T\int_0^t\e^{2\omega(\q)t_1}g(T,t_1)\,\d t_1\right),

516: \label{cqfg}

517: \eeq

518: using the definition~(\ref{gdef}) of $g(T,t)$.

519: We shall consider in particular the dynamical susceptibility

520: \beq

521: \chi(t)=\frac{1}{T}\sum_\x\mean{S_\0(t)S_\x(t)}=\frac{C^\F(\q=\0,t)}{T},

522: \eeq

523: for which eq.~(\ref{cqfg}) yields

524: \beq

525: \chi(t)=\frac{1}{g(T,t)}\left(\frac{1}{T}+2\int_0^t g(T,t_1)\,\d t_1\right).

526: \eeq

527: The asymptotic expressions~(\ref{ghigh}), (\ref{glow}),

528: and~(\ref{gcrit}) of $g(T,t)$ lead to the following predictions.

529: 

530: \begin{itemize}

531: 

532: \item In the paramagnetic phase $(T>T_c)$, the correlation function converges

533: exponentially fast to its equilibrium value, which has the Ornstein-Zernike

534: form

535: \beq

536: C^\F_\eq(\q)=\frac{T}{\omega(\q)+\xi_\eq^{-2}},

537: \label{cfeq}

538: \eeq

539: where the equilibrium correlation length $\xi_\eq$ is given by

540: \beq

541: \xi_\eq^2=2\taueq.

542: \eeq

543: The corresponding value of the equilibrium susceptibility is

544: $\chi_\eq=\xi_\eq^2=2\taueq$.

545: Eq.~(\ref{cfeq}) implies an exponential and isotropic fall-off

546: of correlations, of the form $C_{\x,\eq}\sim\e^{-|\x|/\xi_\eq}$,

547: at large distances and for $\xi_\eq$ large, i.e., $T$ close enough to $T_c$.

548: 

549: \item In the ferromagnetic phase $(T<T_c)$,

550: using the third of the identities~(\ref{fgid}),

551: we obtain a scaling form for the correlation function, namely

552: \beq

553: C^\F(\q,t)\approx M_\eq^2\,(8\pi t)^{D/2}\,\e^{-2\q^2t},

554: \eeq

555: or equivalently,

556: \beq

557: C_\x(t)\approx M_\eq^2\,\e^{-\x^2/(8t)},

558: \eeq

559: in the regime where $\x$ is large (i.e., $\q$ is small) and~$t$ is large.

560: Both the Gaussian profile of the correlation function,

561: and its scaling law involving one single diverging length scale

562: \beq

563: L(t)\sim t^{1/2},

564: \label{lt}

565: \eeq

566: reflect the diffusive nature of the coarsening process.

567: The growing length $L(t)$ can be interpreted as the characteristic

568: size of an ordered domain.

569: The dynamical susceptibility,

570: \beq

571: \chi(t)\approx\frac{M_\eq^2}{T}(8\pi t)^{D/2},

572: \label{chilow}

573: \eeq

574: grows as $\chi(t)\sim L(t)^D$,

575: or else as the volume explored by a diffusive process.

576: 

577: \item At the critical point $(T=T_c)$,

578: the equilibrium correlation function reads

579: \beq

580: C^\F_\eq(\q)\approx\frac{T_c}{\q^2},

581: \eeq

582: i.e.,

583: \beq

584: C_{\x,\eq}\approx\frac{\Gamma(D/2-1)}{4\,\pi^{D/2}}

585: \,\frac{T_c}{\abs{\x}^{D-2}}.

586: \eeq

587: These limiting expressions are reached

588: according to scaling laws of the form

589: \beqa

590: C^\F(\q,t)&\approx&C^\F_\eq(\q)\,\Phi(\q^2t),\cr

591: C_\x(t)&\approx&C_{\x,\eq}\,\Psi(\x^2/t),

592: \label{cxsca}

593: \eeqa

594: with

595: \beqa

596: 2<D<4:&&\Phi(x)=2x\int_0^1\e^{-2x(1-z)}\,z^{D/2-2}\,\d z,\cr

597: &&\Psi(y)=\e^{-y/8},\cr\cr

598: D>4:&&\Phi(x)=1-\e^{-2x},\cr

599: &&\Psi(y)=\frac{1}{\Gamma(D/2-1)}\int_{y/8}^\infty

600: \e^{-z}\,z^{D/2-2}\,\d z.

601: \eeqa

602: 

603: The second expression of eq.~(\ref{cxsca}) has the general

604: scaling form for the equal-time correlation function

605: (see eq.~(\ref{cxtcrit})),

606: with the known value of the static exponent of correlations $\eta=0$

607: for the spherical model~\cite{baxter}, and with $z_c=2$, already found above.

608: 

609: The dynamical susceptibility grows linearly with time,

610: as $\chi(t)\approx\Phi'(0)\,t$, i.e.,

611: \beqa

612: 2<D<4:&&\chi(t)\approx\frac{4}{D-2}\,t,\cr\cr

613: D>4:&&\chi(t)\approx2t.

614: \eeqa

615: 

616: \end{itemize}

617: 

618: \subsection{Two-time correlation function}

619: 

620: We now consider the two-time correlation function

621: \beq

622: C_{\x-\y}(t,s)=\mean{S_\x(t)S_\y(s)},

623: \eeq

624: with $0\le s$~(waiting time) $\le t$~(observation time).

625: Its Fourier transform $C^\F(\q,t,s)$ is defined as in eq.~(\ref{cqdef}).

626: Using eq.~(\ref{sq}), we obtain

627: \beq

628: C^\F(\q,t,s)=\frac{\e^{-\omega(\q)(t+s)}}{\sqrt{g(T,t)g(T,s)}}

629: \left(1+2T\int_0^{s}\e^{2\omega(\q)t_1}g(T,t_1)\,\d t_1\right),

630: \label{cqts}

631: \eeq

632: or else

633: \beq

634: C^\F(\q,t,s)=C^\F(\q,s)\,\e^{-\omega(\q)(t-s)}\sqrt\frac{g(T,s)}{g(T,t)},

635: \label{cqts2}

636: \eeq

637: using the expression~(\ref{cqfg}) for $C^\F(\q,s)$.

638: 

639: In the following, we shall be mostly interested in the two-time

640: autocorrelation function

641: \beq

642: C(t,s)\equiv C_\0(t,s)=\mean{S_\x(t)S_\x(s)}=\int\dq\,C^\F(\q,t,s),

643: \eeq

644: for which eq.~(\ref{cqts}) yields

645: \beq

646: C(t,s)=\frac{1}{\sqrt{g(T,t)g(T,s)}}

647: \left[f\!\left(\frac{t+s}{2}\right)

648: +2T\int_0^{s}f\!\left(\frac{t+s}{2}-t_1\right)g(T,t_1)\,\d t_1\right].

649: \eeq

650: The autocorrelation with the initial state assumes the simpler form

651: \beq

652: C(t,s=0)=\frac{f(t/2)}{\sqrt{g(T,t)}}.

653: \label{czero}

654: \eeq

655: 

656: The asymptotic expressions~(\ref{fdef}), (\ref{ghigh}), (\ref{glow}),

657: and~(\ref{gcrit}) of the functions $f(t)$ and $g(T,t)$

658: lead to the following predictions.

659: 

660: \begin{itemize}

661: 

662: \item In the paramagnetic phase $(T>T_c)$, as $s\to\infty$

663: with $\tau=t-s$ fixed, the system converges to its equilibrium state,

664: where the correlation function only depends on~$\tau$:

665: \beq

666: C(s+\tau,s)\comport{\to}{s\to\infty}

667: C_\eq(\tau)=T\int_\tau^\infty f(\tau_1/2)\,\e^{-\tau_1/(2\taueq)}\,\d\tau_1.

668: \label{ceq}

669: \eeq

670: This equilibrium correlation function decreases exponentially to zero

671: as $\e^{-\tau/(2\taueq)}$ when $\tau\to\infty$.

672: The initial value $C_\eq(\tau=0)=1$ is ensured by the second identity

673: of eq.~(\ref{fgid}).

674: 

675: \item In the ferromagnetic phase $(T<T_c)$, two regimes need to be considered.

676: In the first regime ($s\to\infty$ and $\tau$~fixed, i.e., $1\sim\tau\ll s$),

677: using again the identities~(\ref{fgid}), we obtain

678: \beq

679: C(s+\tau,s)\approx M_\eq^2+(1-M_\eq^2)C_\eqcr(\tau),

680: \label{cbeta}

681: \eeq

682: where we have set

683: \beq

684: C_\eqcr(\tau)=T_c\int_\tau^\infty f(\tau_1/2)\,\d\tau_1.

685: \label{cc0}

686: \eeq

687: This function, which corresponds to the $T\to T_c$ limit of eq.~(\ref{ceq}),

688: decreases only algebraically to zero when $\tau\to\infty$, as

689: \beq

690: C_\eqcr(\tau)\comport{\approx}{\tau\to\infty}

691: \frac{2(4\pi)^{-D/2}}{D-2}\,T_c\,\tau^{-(D/2-1)},

692: \label{cc1}

693: \eeq

694: as implied by eq.~(\ref{fdef}).

695: The first identity of eq.~(\ref{fgid}) ensures that

696: $C_\eqcr(\tau=0)=1$.

697: 

698: In the second regime,

699: where~$s$ and~$t$ are simultaneously large (i.e., $1\ll s\sim\tau$),

700: with arbitrary ratio

701: \beq

702: x=\frac{t}{s}=1+\frac{\tau}{s}\ge1,

703: \label{dxts}

704: \eeq

705: the correlation function obeys a scaling law of the form

706: \beq

707: C(t,s)\approx M_\eq^2\left(\frac{4ts}{(t+s)^2}\right)^{D/4}

708: \approx M_\eq^2\left(\frac{4x}{(x+1)^2}\right)^{D/4}.

709: \label{calpha}

710: \eeq

711: When $x\gg1$, this expression behaves as

712: \beq

713: C(t,s)\approx A\,M_\eq^2\,x^{-\lambda/2},

714: \eeq

715: which can be recast into

716: \beq

717: C(t,s)\sim M_\eq^2\,\left(\frac{L(t)}{L(s)}\right)^{-\lambda},

718: \label{fish}

719: \eeq

720: where $L(t)$ is the length scale defined in eq.~(\ref{lt}),

721: and $\lambda$ is the autocorrelation exponent (see section~3),

722: which is equal to $D/2$ in the present case,

723: in agreement with the result found in the $n\to\infty$ limit of the $O(n)$

724: model (see ref.~\cite{bray}, p.~386, and references therein).

725: 

726: Between these two regimes, the correlation function takes a plateau value

727: \beq

728: \qea=\lim_{\tau\to\infty}\lim_{s\to\infty}C(s+\tau,s)

729: =M_\eq^2=1-\frac{T}{T_c},

730: \label{qeadef}

731: \eeq

732: known as the Edwards-Anderson order parameter (see e.g. ref.~\cite{mezard}).

733: 

734: Hereafter we shall refer to the first regime ($1\sim\tau\ll s$)

735: as the {\it stationary} regime, and to the second one ($1\ll s\sim\tau$)

736: as the {\it scaling} (or {\it aging}) regime.

737: In the former, the system becomes stationary,

738: though without reaching thermal equilibrium,

739: because the system is coarsening.

740: In the latter regime, as said above, the system is aging.

741: It is possible to match these two kinds of behavior,

742: corresponding respectively to eq.~(\ref{cbeta}) and~(\ref{calpha}),

743: into a single expression:

744: \beq

745: C(t=s+\tau,s)\approx

746: (1-M_\eq^2)C_\eqcr(\tau)

747: +M_\eq^2\left(\frac{4ts}{(t+s)^2}\right)^{D/4},

748: \eeq

749: which is the sum of a term corresponding to the stationary contribution,

750: and a term corresponding to the aging one.

751: Let us finally recall that, in the context of glassy dynamics,

752: in a low-temperature phase, the first regime, where $C(t,s)>\qea$, is usually

753: referred to as the $\beta$ regime, while the second one,

754: where $C(t,s)<\qea$, is referred to as the $\alpha$ regime~\cite{revue}.

755: 

756: \item At the critical point $(T=T_c)$,

757: the same two regimes are to be considered.

758: However their physical interpretation is slightly different, since

759: the order parameter $M_\eq$ vanishes, and symmetry

760: between the phases is restored.

761: 

762: In the first regime $(1\sim\tau\ll s)$, the system again becomes stationary,

763: the autocorrelation function behaving as the $T\to T_c$ limit

764: of eq.~(\ref{ceq}), that is

765: \beq

766: C(s+\tau,s)\comport{\to}{s\to\infty}C_\eqcr(\tau),

767: \label{cc}

768: \eeq

769: which decreases algebraically to zero

770: when $\tau\to\infty$ (cf. eq.~(\ref{cc1})).

771: In the second regime ($1\ll s\sim\tau$),

772: the correlation function obeys a scaling law of the form

773: \beq

774: C(t,s)\approx T_c\,s^{-(D/2-1)}\,F(x),

775: \label{csca}

776: \eeq

777: where the scaling function $F(x)$ reads

778: \beqa

779: 2<D<4:&&F(x)=\frac{4(4\pi)^{-D/2}}{(D-2)(x+1)}\,x^{1-D/4}(x-1)^{1-D/2},\cr\cr

780: D>4:&&F(x)=\frac{2(4\pi)^{-D/2}}{D-2}\left((x-1)^{1-D/2}-(x+1)^{1-D/2}\right).

781: \label{f}

782: \eeqa

783: 

784: In this regime the system is still aging, in the sense that $C(t,s)$

785: bears a dependence in both time variables.

786: However, the scaling of expression~(\ref{csca}) is different from that

787: found in the low-temperature phase (see eq.~(\ref{calpha})),

788: which depends on the ratio $x=t/s$ only.

789: The presence in eq.~(\ref{csca}) of an additional $s$-dependence

790: through the factor $s^{-(D/2-1)}$

791: can be interpreted as coming from the anomalous

792: dimension of the field $S_\x$ at $T_c$.

793: In the critical region one has indeed

794: $M_\eq\sim(T-T_c)^{\beta}\sim\xi_\eq^{-\beta/\nu}$.

795: Replacing $\xi_\eq$ by $s^{1/z_c}$ implies the replacement of

796: $M_\eq^2$ by $s^{-2\beta/\nu z_c}\sim s^{-(D-2+\eta)/z_c}$.

797: With $\eta=0$ and $z_c=2$, the factor $s^{-(D/2-1)}$ is thus recovered.

798: Note that the static hyperscaling relation $2\beta/\nu=D-2+\eta$

799: holds for $D<4$, while it is violated for $D>4$ (see Table~1).

800: 

801: Two limiting regimes are of interest.

802: First, for $x\to1$, i.e., $1\ll\tau\ll s$,

803: eq.~(\ref{csca}) matches eq.~(\ref{cc1}).

804: Second, for $x\gg1$, i.e., $1\ll s\ll t$, one gets

805: \beq

806: F(x)\approx B\,x^{-\lambda_c/z_c},

807: \label{fxxgd}

808: \eeq

809: where the autocorrelation exponent $\lambda_c$ (see section~3)

810: is equal to $3D/2-2$ if $2<D<4$,

811: and to~$D$ above four dimensions, in agreement with the result

812: found in ref.~\cite{janssen}.

813: 

814: We also quote for later reference the scaling law of the derivative

815: \beq

816: \frac{\dpar C(t,s)}{\dpar s}\approx T_c\,s^{-D/2}\,F_1(x),

817: \label{dcsca}

818: \eeq

819: with

820: \beq

821: F_1(x)=-\frac{D-2}{2}\,F(x)-x\,F'(x),

822: \eeq

823: i.e.,

824: \beqa

825: 2<D<4:&&F_1(x)=(4\pi)^{-D/2}\frac{(D-2)(x+1)^2+2(x-1)^2}{(D-2)(x+1)^2}

826: \,x^{1-D/4}(x-1)^{-D/2},\cr\cr

827: D>4:&&F_1(x)=(4\pi)^{-D/2}\left((x-1)^{-D/2}+(x+1)^{-D/2}\right).

828: \label{f1}

829: \eeqa

830: 

831: \end{itemize}

832: 

833: \subsection{Two-time response function}

834: 

835: Suppose now that the system is subjected to a small magnetic field $H_\x(t)$,

836: depending on the site $\x$ and on time $t\ge0$ in an arbitrary fashion.

837: This amounts to adding to the ferromagnetic Hamiltonian~(\ref{ham})

838: a time-dependent perturbation of the form

839: \beq

840: \delta\H(t)=-\sum_\x H_\x(t)S_\x(t).

841: \eeq

842: The dynamics of the model is now given by the modified Langevin equation

843: \beq

844: \frac{\d S_\x}{\d t}=\sum_{\y(\x)}S_\y-\lambda(t)S_\x+H_\x(t)+\eta_\x(t).

845: \label{lanh}

846: \eeq

847: 

848: Causality and invariance under spatial translations imply that we have,

849: to first order in the magnetic field $H_\x(t)$,

850: \beq

851: \mean{S_\x(t)}=\int_0^t\d s\sum_\y R_{\x-\y}(t,s)H_\y(s)+\cdots

852: \label{sxt}

853: \eeq

854: This formula defines the two-time response function $R_{\x-\y}(t,s)$

855: of the model.

856: A more formal definition reads

857: \beq

858: R_{\x-\y}(t,s)=\left.\frac{\delta\mean{S_\x(t)}}

859: {\delta H_\y(s)}\right\vert_{\{H_\x(t)=0\}}.

860: \eeq

861: 

862: The solution to eq.~(\ref{lanh}) reads, in Fourier space,

863: \beq

864: S^\F(\q,t)=\e^{-\omega(\q)t-\z(t)}

865: \left(S^\F(\q,t=0)+\int_0^t\e^{\omega(\q)t_1+\z(t_1)}

866: \left[H^\F(\q,t_1)+\eta^\F(\q,t_1)\right]\,\d t_1\right).

867: \eeq

868: It can be checked that the Lagrange function $\lambda(t)$,

869: and hence $z(t)$ and $\z(t)$,

870: remain unchanged, to first order in the magnetic field.

871: As a consequence, the two-time response function reads, in Fourier transform,

872: \beq

873: R^\F(\q,t,s)=\left.\frac{\delta\mean{S^\F(\q,t)}}{\delta H^\F(\q,s)}

874: \right\vert_{\{H_\x(t)=0\}}

875: =\e^{-\omega(\q)(t-s)}\sqrt\frac{g(T,s)}{g(T,t)}

876: \label{rqts}

877: \eeq

878: (cf. eq.~(\ref{cqts2})).

879: In the following, we shall be mostly interested in the diagonal component

880: of the response function, corresponding to coinciding points:

881: \beq

882: R(t,s)\equiv R_\0(t,s)

883: =\left.\frac{\delta\mean{S_\x(t)}}{\delta H_\x(s)}\right\vert_{\{H_\x(t)=0\}}

884: =\int\dq\,R^\F(\q,t,s).

885: \eeq

886: With the notations~(\ref{fdef}), (\ref{gdef}), eq.~(\ref{rqts}) yields

887: \beq

888: R(t,s)=f\!\left(\frac{t-s}{2}\right)\sqrt\frac{g(T,s)}{g(T,t)}.

889: \eeq

890: The response function at zero waiting time

891: assumes the simpler form~(cf.~eq.~(\ref{czero}))

892: \beq

893: R(t,s=0)=C(t,s=0)=\frac{f(t/2)}{\sqrt{g(T,t)}}.

894: \eeq

895: 

896: The asymptotic expressions~(\ref{fdef}), (\ref{ghigh}), (\ref{glow}),

897: and~(\ref{gcrit}) of $F(t)$ and $g(T,t)$ lead to the following predictions.

898: 

899: \begin{itemize}

900: 

901: \item In the paramagnetic phase $(T>T_c)$,

902: at equilibrium, the response function only depends on $\tau$, according to

903: \beq

904: R_\eq(\tau)=f(\tau/2)\,\e^{-\tau/(2\taueq)}.

905: \label{req}

906: \eeq

907: Moreover, it is related to the equilibrium correlation function

908: $C_\eq(\tau)$ of eq.~(\ref{ceq})

909: by the fluctuation-dissipation theorem~(\ref{fdt}), as it should.

910: 

911: \item In the ferromagnetic phase $(T<T_c)$,

912: the two regimes defined in the previous section for the case of

913: the autocorrelation function are still to be considered.

914: In the stationary regime ($1\sim\tau\ll s$),

915: the response function behaves as the $T\to T_c$ limit of eq.~(\ref{req}),

916: namely

917: \beq

918: R_\eqcr(\tau)=f(\tau/2)=-\frac{1}{T_c}\frac{\d C_\eqcr(\tau)}{\d\tau},

919: \label{rc}

920: \eeq

921: so that the \fd theorem is valid.

922: 

923: On the contrary, in the scaling regime ($1\ll s\sim t$),

924: the response function has the form

925: \beq

926: R(t,s)\approx(4\pi(t-s))^{-D/2}(t/s)^{D/4}=

927: (4\pi s)^{-D/2}\,(x-1)^{-D/2}x^{D/4},

928: \label{ralpha}

929: \eeq

930: which, when compared to

931: the corresponding expression~(\ref{calpha}) for the autocorrelation function,

932: demonstrates the violation of the \fd theorem (see section~2.5).

933: 

934: \item At the critical point $(T=T_c)$,

935: in the stationary regime $(1\sim\tau\ll s)$,

936: the response function still behaves as in eq.~(\ref{rc}),

937: so that the \fd theorem still holds.

938: In the scaling regime ($1\ll s\sim t$),

939: the response function obeys a scaling law of the form

940: \beq

941: R(t,s)\approx s^{-D/2}\,F_2(x),

942: \label{rsca}

943: \eeq

944: where the scaling function $F_2(x)$ reads

945: \beqa

946: 2<D<4:&&F_2(x)=(4\pi)^{-D/2}x^{1-D/4}(x-1)^{-D/2},\cr

947: D>4:&&F_2(x)=(4\pi)^{-D/2}(x-1)^{-D/2}.

948: \label{f2}

949: \eeqa

950: 

951: Again two limiting regimes are of interest.

952: For $x\to1$, the scaling result~(\ref{f2}) matches eq.~(\ref{rc}).

953: For $x\gg 1$, one finds the same power-law fall-off

954: for the functions $F(x)$, $F_1(x)$, and $F_2(x)$, that is

955: \beq

956: F(x)\sim F_1(x)\sim F_2(x)\sim x^{-\lambda_c/z_c}

957: \eeq

958: (see sections~2.5 and~3).

959: 

960: \end{itemize}

961: 

962: \subsection{Fluctuation-dissipation ratio}

963: 

964: As already mentioned in the introduction,

965: the violation of the \fd theorem~(\ref{fdt}) out of thermal equilibrium

966: can be characterized by the \fd ratio $X(t,s)$, defined in eq.~(\ref{dx}).

967: In the case of the spherical model,

968: the results derived so far yield at once the following predictions.

969: 

970: \begin{itemize}

971: 

972: \item In the paramagnetic phase $(T>T_c)$,

973: the system converges to an equilibrium state, where the \fd theorem holds.

974: In other words, the \fd ratio converges toward its equilibrium value

975: \beq

976: X_\eq=1.

977: \eeq

978: 

979: \item In the ferromagnetic phase $(T<T_c)$,

980: the \fd theorem~(\ref{fdt}) is only valid in the stationary regime

981: ($1\sim\tau\ll s$).

982: On the contrary, in the scaling regime ($1\ll s\sim\tau$),

983: the results~(\ref{calpha}) and~(\ref{ralpha})

984: imply that the \fd ratio falls off as

985: \beq

986: X(t,s)\approx\frac{(8\pi)^{-D/2}}{D}\,\frac{4T}{M_\eq^2}\,

987: \left(\frac{x+1}{x-1}\right)^{D/2+1}s^{-(D/2-1)}.

988: \label{vanis}

989: \eeq

990: In particular, the limit \fd ratio introduced in eq.~(\ref{xinfdef}) reads

991: \beq

992: X_\infty=0.

993: \eeq

994: 

995: \item At the critical point $(T=T_c)$,

996: the scaling laws~(\ref{dcsca}) and~(\ref{rsca})

997: imply that the \fd ratio $X(t,s)$ becomes asymptotically

998: a smooth function of the time ratio $x=t/s$:

999: \beq

1000: X(t,s)\comport{\approx}{t,s\to\infty}\X(x)=\frac{F_2(x)}{F_1(x)},

1001: \label{xsca}

1002: \eeq

1003: i.e., explicitly,

1004: \beqa

1005: 2<D<4:&&\X(x)=\frac{1}{1+\frad{2}{D-2}\left(\frad{x-1}{x+1}\right)^2},\cr\cr

1006: D>4:&&\X(x)=\frac{1}{1+\left(\frad{x-1}{x+1}\right)^{D/2}}.

1007: \label{x}

1008: \eeqa

1009: The scaling law~(\ref{xsca}) interpolates between the equilibrium behavior

1010: \beq

1011: \X(x)\comport{\to}{x\to1}X_\eq=1

1012: \eeq

1013: in the stationary regime of relatively short time differences,

1014: and a non-trivial limit value

1015: \beq

1016: \X(x)\comport{\to}{x\to\infty}X_\infty

1017: \eeq

1018: at large time differences, given by

1019: \beqa

1020: 2<D<4:&&X_\infty=\frac{D-2}{D},\cr\cr

1021: D>4:&&X_\infty=\frac{1}{2}.

1022: \label{xexpl}

1023: \eeqa

1024: 

1025: \end{itemize}

1026: 

1027: Further comments on the scaling behavior of the \fd ratio

1028: will be made in section~3.2.

1029: 

1030: \def\figun{

1031: \vskip 8.5cm{\hskip .8cm}

1032: \special{psfile=f1.ps hoffset=0 voffset=0 hsize=360 vsize=250}

1033: 

1034: \noindent {\bf Figure~1:}

1035: Log-log plot of the critical autocorrelation function $C(t,s)$

1036: of the two-di\-men\-si\-o\-nal Ising model,

1037: against time ratio $x=t/s$, for several values of the waiting time~$s$.

1038: Data are multiplied by $s^{2\beta/\nu z_c}$,

1039: in order to demonstrate collapse into the scaling function $F(x)$

1040: of eq.~(\ref{gc}).

1041: Straight line: exponent $-\lambda_c/z_c\approx-0.73$

1042: of the fall-off at large~$x$.

1043: }

1044: 

1045: \def\figde{

1046: \vskip 8.5cm{\hskip .8cm}

1047: \special{psfile=f2.ps hoffset=0 voffset=0 hsize=360 vsize=250}

1048: 

1049: \noindent {\bf Figure~2:}

1050: Log-log plot of the critical integrated response function $\rho(t,s)$

1051: of the two-dimensional Ising model,

1052: against time ratio $x=t/s$, for several values of the waiting time~$s$.

1053: Data are multiplied by $s^{2\beta/\nu z_c}$,

1054: in order to demonstrate collapse into the scaling function $F_3(x)$

1055: of eq.~(\ref{grint}).

1056: Straight line: exponent $-\lambda_c/z_c\approx-0.73$

1057: of the fall-off at large~$x$.

1058: }

1059: 

1060: \def\figtr{

1061: \vskip 8.5cm{\hskip .8cm}

1062: \special{psfile=f3.ps hoffset=0 voffset=0 hsize=360 vsize=250}

1063: 

1064: \noindent {\bf Figure~3:}

1065: Parametric plot of the integrated response $\rho(t,s)$

1066: against the autocorrelation $C(t,s)$, using the data of Figures~1 and~2.

1067: Symbols: data for integer time differences $\tau=t-s=0,\dots,8$.

1068: Full line: unit slope corresponding to the \fd theorem

1069: in the stationary regime.

1070: Dashed line: limit slope $X_\infty=0.26$ (see text and Figures~4 and~5).

1071: }

1072: 

1073: \def\figqu{

1074: \vskip 8.5cm{\hskip .8cm}

1075: \special{psfile=f4.ps hoffset=0 voffset=0 hsize=360 vsize=250}

1076: 

1077: \noindent {\bf Figure~4:}

1078: Parametric plot of the ratio $\rho/C$ against~$C$.

1079: Straight lines: constrained least-square fit with common intercept,

1080: yielding $X_\infty\approx0.262$.

1081: }

1082: 

1083: \def\figci{

1084: \vskip 8.5cm{\hskip .8cm}

1085: \special{psfile=f5.ps hoffset=0 voffset=0 hsize=360 vsize=250}

1086: 

1087: \noindent {\bf Figure~5:}

1088: Parametric plot of the combination $(\rho-C^2)/(C(1-C))$ against~$C$.

1089: Straight lines: constrained least-square fit yielding $X_\infty\approx0.260$.

1090: }

1091: 

1092: \section{The generic situation}

1093: 

1094: \subsection{Aging below $T_c$}

1095: 

1096: Let us first briefly sketch the description

1097: of the dynamical behavior of a ferromagnetic system

1098: quenched from a disordered initial state

1099: to a temperature $T<T_c$~\cite{langer,bray,cd,barrat,berthier}.

1100: 

1101: In the scaling regime ($1\ll s\sim t$), the autocorrelation $C(t,s)$

1102: is expected to be a function of the ratio $L(t)/L(s)$ only,

1103: where the length scale $L(t)\sim t^{1/z}$ is the characteristic size

1104: of an ordered domain, and~$z$ is the growth exponent,

1105: equal to 2 for non-conserved dynamics.

1106: More precisely,

1107: \beq

1108: C(t,s)=M_\eq^2\,f(t/s),

1109: \label{lo1}

1110: \eeq

1111: where the scaling function~$f$ is temperature independent.

1112: Furthermore we have, for $x=t/s\gg1$, i.e., $1\ll s\ll t$,

1113: \beq

1114: f(x)\approx A\,x^{-\lambda/z},

1115: \label{lo2}

1116: \eeq

1117: where $\lambda$ is the autocorrelation exponent~\cite{fisher}.

1118: For the spherical model, eqs.~(\ref{calpha}) and~(\ref{fish})

1119: match eqs.~(\ref{lo1}) and~(\ref{lo2}), with $\lambda=D/2$ and $z=2$.

1120: 

1121: As a consequence, we have

1122: \beq

1123: \frac{\dpar C(t,s)}{\dpar s}\approx\frac{M_\eq^2}{s}\,f_1(x),

1124: \label{lo3}

1125: \eeq

1126: with $f_1(x)=-xf'(x)$, so that, when $x\gg1$,

1127: \beq

1128: f_1(x)\approx A_1\,x^{-\lambda/z},

1129: \label{lo5}

1130: \eeq

1131: with $A_1=A\,\lambda/z$.

1132: 

1133: Although the situation of the response $R(t,s)$ is less clear-cut,

1134: it is however reasonable to make the scaling assumption

1135: \beq

1136: R(t,s)\approx s^{-1-a}\,f_2(x),

1137: \label{gr2}

1138: \eeq

1139: where $a>0$ is an unknown exponent, and again with the behavior

1140: \beq

1141: f_2(x)\approx A_2\,x^{-\lambda/z}

1142: \eeq

1143: when $x\gg1$.

1144: 

1145: The scaling law~(\ref{gr2}) holds for the spherical model,

1146: with $a=D/2-1$, as can be seen from eq.~(\ref{ralpha}).

1147: Furthermore, for non-conserved dynamics,

1148: at least in the case of a discrete broken symmetry,

1149: like e.g. in the Ising model,

1150: it has been argued~\cite{berthier,bray2} that the integrated response

1151: $\rho(t,s)$ (to be defined in eq.~(\ref{rhodef})),

1152: scales as $\rho(t,s)\sim L(s)^{-1}\,\varphi(L(t)/L(s))$.

1153: This corresponds to eq.~(\ref{gr2}) with $a=1/z=1/2$.

1154: 

1155: The scaling laws~(\ref{lo3}), (\ref{gr2}) imply

1156: \beq

1157: X(t,s)\approx s^{-a}\,g(x),

1158: \eeq

1159: with $g(x)=(T/M_\eq^2)\,f_2(x)/f_1(x)$,

1160: in agreement with eq.~(\ref{vanis}) for the spherical model, and especially

1161: \beq

1162: X_\infty=0.

1163: \eeq

1164: 

1165: \subsection{Aging at $T_c$}

1166: 

1167: Let us now turn to the situation where a ferromagnetic

1168: system is quenched from a disordered initial state to its critical point.

1169: 

1170: In such a circumstance, spatial correlations develop in the system,

1171: just as in the critical state,

1172: but only over a length scale which grows like $t^{1/z_c}$,

1173: where $z_c$ is the dynamic critical exponent.

1174: For example the equal-time correlation function has the scaling form

1175: \beq

1176: C_\x(t)=|\x|^{-2\beta/\nu}\,\phi\!\left(|\x|/t^{1/z_c}\right),

1177: \label{cxtcrit}

1178: \eeq

1179: where $\beta$ and $\nu$ are the usual static critical exponents.

1180: The scaling function $\phi(x)$ goes to a constant for $x\to0$,

1181: while it falls off exponentially to zero for $x\to\infty$,

1182: i.e., on scales smaller than $t^{1/z_c}$ the system looks critical,

1183: while on larger scales it is disordered.

1184: This behavior is illustrated in the case of the spherical model

1185: by eq.~(\ref{cxsca}),

1186: corresponding to $2\beta/\nu=D-2$ and $z_c=2$ in eq.~(\ref{cxtcrit}).

1187: 

1188: In the scaling region of the two-time plane,

1189: where both times~$s$ and~$t$ are large and comparable ($1\ll s\sim t$),

1190: with arbitrary ratio $x=t/s$,

1191: the two-time autocorrelation function $C(t,s)$

1192: is expected to obey a scaling law of the form

1193: (see the discussion below eq.~(\ref{f}), and ref.~\cite{janssen})

1194: \beq

1195: C(t,s)\approx s^{-2\beta/\nu z_c}\,F(x).

1196: \label{gc}

1197: \eeq

1198: When both time scales are well separated $(1\ll s\ll t$, i.e., $x\gg1$),

1199: the scaling function $F(x)$ falls off as

1200: \beq

1201: F(x)\approx B\,x^{-\lambda_c/z_c},

1202: \label{gfas}

1203: \eeq

1204: where $\lambda_c$ is the critical autocorrelation exponent~\cite{huse},

1205: related to the (magnetization) initial-slip critical exponent

1206: $\Theta_c$~\cite{janssen} by $\lambda_c=D-z_c\Theta_c$.

1207: 

1208: We thus have

1209: \beq

1210: \frac{\dpar C(t,s)}{\dpar s}\approx s^{-1-2\beta/\nu z_c}\,F_1(x),

1211: \label{gdc}

1212: \eeq

1213: with

1214: $F_1(x)=-(2\beta/\nu z_c)F(x)-xF'(x)$, so that, when $x\gg1$,

1215: \beq

1216: F_1(x)\approx B_1\,x^{-\lambda_c/z_c},

1217: \label{gf1as}

1218: \eeq

1219: with

1220: \beq

1221: B_1=\frac{\nu\lambda_c-2\beta}{\nu z_c}\,B.

1222: \label{bb1}

1223: \eeq

1224: 

1225: For the spherical model, eqs.~(\ref{csca}), (\ref{fxxgd}), and~(\ref{dcsca})

1226: respectively match eqs.~(\ref{gc}), (\ref{gfas}), and~(\ref{gdc}), with

1227: $\lambda_c=3D/2-2$ if $D<4$, and $\lambda_c=D$ if $D>4$ (see Table~1).

1228: 

1229: The similarity between the results~(\ref{dcsca}) and~(\ref{rsca}),

1230: obtained in the case of the spherical model,

1231: strongly suggests that $\dpar C(t,s)/\dpar s$

1232: and $R(t,s)$ behave similarly in the generic case,

1233: i.e., one is lead to hypothesize

1234: that a scaling law of the form~(\ref{gdc}),

1235: with the same power-law fall-off~(\ref{gf1as}),

1236: holds for the response, that is

1237: \beq

1238: R(t,s)\approx\frac{1}{T_c}\,s^{-1-2\beta/\nu z_c}\,F_2(x),

1239: \label{gr}

1240: \eeq

1241: with, when $x\gg1$,

1242: \beq

1243: F_2(x)\approx B_2\,x^{-\lambda_c/z_c}.

1244: \label{gf2as}

1245: \eeq

1246: 

1247: The scaling laws~(\ref{gdc}) and~(\ref{gr}) imply then that

1248: the \fd ratio only depends on the

1249: time ratio~$x$ throughout the scaling region:

1250: \beq

1251: X(t,s)\approx\X(x)=\frac{F_2(x)}{F_1(x)},

1252: \label{xgen}

1253: \eeq

1254: where the scaling function $\X(x)$ is universal.

1255: It appears indeed as a dimensionless combination of scaling functions.

1256: In turn, the hypothesis~(\ref{gf2as}) implies that the limit \fd ratio reads

1257: \beq

1258: X_\infty=\X(\infty)=\frac{B_2}{B_1}.

1259: \label{xr}

1260: \eeq

1261: This number thus appears as a dimensionless amplitude ratio,

1262: in the usual sense of critical phenomena.

1263: It is therefore a novel universal quantity

1264: of non-equilibrium critical dynamics, as already claimed in ref.~\cite{gl}.

1265: 

1266: In the case of the spherical model,

1267: the analytical treatment of section~2 corroborates the above analysis,

1268: and yields the quantitative predictions~(\ref{x}) and~(\ref{xexpl}).

1269: 

1270: In order to perform a numerical evaluation of $X_\infty$,

1271: one needs to measure the response.

1272: A convenient way to do so is to

1273: measure instead the dimensionless integrated response function

1274: \beq

1275: \rho(t,s)=T\int_0^s R(t,u)\,\d u.

1276: \label{rhodef}

1277: \eeq

1278: By eq.~(\ref{sxt}), this quantity is proportional to the thermoremanent

1279: magnetization $M_\TRM$,

1280: i.e., the magnetization of the system at time~$t$

1281: obtained after applying a small magnetic field~$h$, uniform and constant,

1282: between $t=0$ and $t=s$:

1283: \beq

1284: M_\TRM(t,s)\approx\frac{h}{T}\,\rho(t,s).

1285: \eeq

1286: The thermoremanent magnetization is a natural quantity

1287: to measure experimentally in spin glasses~\cite{revue},

1288: and it is also accessible to numerical simulations,

1289: for systems with and without quenched randomness (see section~3.3).

1290: 

1291: The scaling law~(\ref{gr}) for the response function implies

1292: \beq

1293: \rho(t,s)\approx s^{-2\beta/\nu z_c}\,F_3(x),

1294: \label{grint}

1295: \eeq

1296: with $F_2(x)=-(2\beta/\nu z_c)F_3(x)-xF_3'(x)$, so that, when $x\gg1$,

1297: \beq

1298: F_3(x)\approx B_3\,x^{-\lambda_c/z_c},

1299: \label{gf3as}

1300: \eeq

1301: with

1302: \beq

1303: B_3=\frac{\nu z_c}{\nu\lambda_c-2\beta}\,B_2.

1304: \label{b2b3}

1305: \eeq

1306: 

1307: A clear representation of the evolution of $X(t,s)$ in time is provided by

1308: the parametric plot of $\rho(t,s)$ against $C(t,s)$,

1309: obtained by varying~$t$ at fixed~$s$~\cite{x1,barrat,berthier}.

1310: For well-separated times in the scaling regime (i.e., $1\ll s\ll t$),

1311: the common power-law behavior ~(\ref{gfas}), (\ref{gf1as}), (\ref{gf2as}),

1312: and~(\ref{gf3as}) implies that the limit \fd ratio

1313: has the alternative expression

1314: \beq

1315: X_\infty=\frac{B_3}{B},

1316: \label{xrho}

1317: \eeq

1318: which is equivalent to eq.~(\ref{xr}),

1319: due to eqs.~(\ref{bb1}) and~(\ref{b2b3}).

1320: In other words, the relationship~(\ref{dx}) also holds in integral form,

1321: that is

1322: \beq

1323: \rho(t,s)\approx X_\infty\,C(t,s),

1324: \label{rhosca}

1325: \eeq

1326: in the regime $1\ll s\ll t$.

1327: The limit \fd ratio can thus be measured as the slope of the parametric plot

1328: in the scaling region, i.e., near the origin of the $C-\rho$ plane.

1329: Eq.~(\ref{rhosca}) is expected to hold as long

1330: as~$C$ and $\rho$ are much smaller than the crossover scale

1331: \beq

1332: C^*(s)=C(2s,s)\sim s^{-2\beta/\nu z_c},

1333: \label{sizecrit}

1334: \eeq

1335: corresponding to $\tau=s$.

1336: This quantity provides a measure of the size of the critical region,

1337: giving thus a quantitative definition of the critical analogue of $M_\eq^2$,

1338: involved in the discussion below eq.~(\ref{f}).

1339: 

1340: \subsection{The two-dimensional Ising model: numerical simulations}

1341: 

1342: In order to check the validity of the scaling analysis

1343: made in the previous section, beyond the case of the spherical model,

1344: we have performed numerical simulations on the ferromagnetic Ising model

1345: on the square lattice,

1346: evolving under heat-bath (Glauber) dynamics at its critical temperature

1347: $T_c=2/\ln(1+\sqrt{2})\approx2.2692$, starting from a disordered initial state.

1348: The rules of the dynamics are as follows.

1349: Consider a finite system, consisting of $N=L^2$ spins $\sigma_\x=\pm1$

1350: situated at the vertices $\x$ of a square lattice,

1351: with periodic boundary conditions.

1352: The Ising Hamiltonian reads

1353: \beq

1354: \H=-\sum_{(\x,\y)}\sigma_\x\sigma_\y,

1355: \eeq

1356: where the sum runs over pairs of neighboring sites.

1357: Heat-bath dynamics consists in

1358: updating the spins $\sigma_\x(t)$ according to the stochastic rule

1359: \beq

1360: \sigma_\x(t)\to\left\{\matrix{

1361: +1&\mbox{with prob.}\,\frad{1+\tanh(h_\x(t)/T_c)}{2},\hfill\cr\cr

1362: -1&\mbox{with prob.}\,\frad{1-\tanh(h_\x(t)/T_c)}{2},\hfill\cr}\right.%}

1363: \eeq

1364: where the local field $h_\x(t)$ acting on $\sigma_\x(t)$ reads

1365: \beq

1366: h_\x(t)=\sum_{\y(\x)}\sigma_\y(t),

1367: \label{hlocdef}

1368: \eeq

1369: with $\y(\x)$ denoting the four neighbors of site $\x$.

1370: 

1371: Let us give a brief summary of known facts on the dynamics of the Ising model.

1372: For $T<T_c$, numerical studies have shown that the scaling

1373: forms~(\ref{lo1}) and~(\ref{lo2}) hold,

1374: with $z=2$ (non-conserved dynamics) and $\lambda\approx1.25$~\cite{fisher}.

1375: The integrated response function (in another form,

1376: known as the ZFC magnetization) has been measured in ref.~\cite{barrat}.

1377: At $T=T_c$, the dynamic critical exponent reads $z_c\approx2.17$~\cite{blote},

1378: and the autocorrelation exponent $\lambda_c\approx1.59$~\cite{huse,grass}.

1379: 

1380: Our aim is now to verify the hypotheses made in section~3.2,

1381: especially the scaling laws~(\ref{gr}) and~(\ref{grint})

1382: for the response function,

1383: and to demonstrate the existence of a non-trivial limit $X_\infty$.

1384: 

1385: Computing $C(t,s)$ with good statistics is rather easy,

1386: while the computation of $\rho(t,s)$ requires more effort.

1387: We have followed the lines of the method introduced in ref.~\cite{barrat}.

1388: In order to isolate the diagonal component of the response function,

1389: a quenched, spatially random magnetic field,

1390: is applied to the system from $t=0$ to $t=s$.

1391: This magnetic field is of the form $H_\x=h_0\eps_\x$,

1392: with a constant small amplitude $h_0$, and a quenched random modulation,

1393: $\eps_\x=\pm1$ with equal probability, independently at each site $\x$.

1394: The heat-bath dynamical rules are modified by adding up

1395: the magnetic field $H_\x$ to the local field $h_\x(t)$ of eq.~(\ref{hlocdef}).

1396: We have then

1397: \beq

1398: \overline{\langle\eps_\x\sigma_\x(t)\rangle}

1399: =h_0\int_0^s R(t,u)\,\d u=\frac{h_0}{T}\,\rho(t,s)=M_{\rm TRM}(t,s),

1400: \eeq

1401: where the bar means an average with respect to

1402: the distribution of the modulation $\eps_\x$ of the magnetic field.

1403: 

1404: \figun

1405: 

1406: We have first checked the validity of the scaling laws~(\ref{gc}),

1407: and especially~(\ref{grint}).

1408: Figures~1 and~2 respectively show log-log plots of the autocorrelation function

1409: $C(t,s)$ and of the corresponding integrated response function $\rho(t,s)$,

1410: against the time ratio $x=t/s$, for several values of the waiting time~$s$.

1411: For each value of~$s$, the simulations are run up to $t/s=10$,

1412: and data are averaged over at least 500 independent samples

1413: of size $300\times300$.

1414: For the response function, the amplitude of the quenched magnetic field

1415: reads $h_0=0.05$.

1416: Multiplying the data by $s^{2\beta/\nu z_c}$,

1417: with $2\beta/\nu z_c\approx0.115$, gives good data collapse,

1418: thus producing a plot of the scaling functions $F(x)$ and $F_3(x)$.

1419: The data follow a power-law fall-off at large values of~$x$,

1420: with a slope in good agreement with the value

1421: $-\lambda_c/z_c\approx-0.73$, shown on the plots as a straight line.

1422: 

1423: We then turned to an investigation of the parametric plot

1424: of these data in the $C-\rho$ plane.

1425: At the qualitative level, this plot, shown in Figure~3

1426: for several values of the waiting time~$s$, confirms our expectations.

1427: The stationary regime $(1\sim\tau\ll s$, i.e., roughly speaking, $C>C^*(s))$,

1428: corresponds to the right part of the plot.

1429: The symbols show the data for small integer values of the time difference,

1430: $\tau=t-s=0,\dots,8$, illustrating the fast decay of correlation

1431: and integrated response in the stationary regime.

1432: The rightmost points, corresponding to $\tau=0$, i.e., $C=C(s,s)=1$,

1433: are compatible with the scaling law

1434: $1-\rho(s,s)\sim C^*(s)\sim s^{-2\beta/\nu z_c}$.

1435: The validity of the \fd theorem is testified

1436: by the unit slope of this part of the plot, shown as a full straight line.

1437: The aging regime $(1\ll s\sim t$, i.e., roughly speaking, $C<C^*(s))$,

1438: corresponds to the left part of the plot.

1439: As expected, the data crossover toward a non-trivial slope,

1440: equal to the limit \fd ratio $X_\infty$.

1441: The dashed line shows the slope $X_\infty=0.26$,

1442: obtained by the analysis described below.

1443: 

1444: \figde

1445: 

1446: In order to obtain a quantitative prediction of the limit \fd ratio $X_\infty$,

1447: we have followed two approaches.

1448: Figure~4 depicts the local slope of the plot of Figure~3,

1449: i.e., the ratio $\rho/C$, against~$C$, in the aging regime.

1450: The data for the largest available waiting time $s=200$ have been discarded

1451: from the analysis because they appear as too noisy on that scale.

1452: The data look pretty linear all over the range presented in the plot.

1453: This precocious scaling is due to the fact that the exponent

1454: $2\beta/\nu z_c\approx0.115$ is small.

1455: Hence the size of the critical region,

1456: given by the estimate~(\ref{sizecrit}), is very large,

1457: at least for waiting times~$s$ accessible to computer simulations.

1458: We have indeed, for example, $C^*(100)=C(200,100)\approx0.24$.

1459: The straight lines show a constrained least-square fit of the three

1460: series of data, imposing a common intercept.

1461: The value of this intercept yields the prediction $X_\infty\approx0.262$.

1462: 

1463: \figtr

1464: 

1465: We have also followed an alternative approach,

1466: aiming at subtracting most of the deviations of the ratio $\rho/C$

1467: with respect to its limit $X_\infty$ at $C\to0$.

1468: This can be done by incorporating the known limit

1469: of the stationary regime, i.e., $\rho\approx1$ as $C\to1$,

1470: into a quadratic phenomenological formula:

1471: $\rho\approx X_\infty C +(1-X_\infty) C^2$.

1472: This formula can be rewritten as $X_\infty\approx(\rho-C^2)/(C(1-C))$,

1473: suggesting to plot $(\rho-C^2)/(C(1-C))$ against~$C$,

1474: instead of the mere ratio $\rho/C$.

1475: This has been done in Figure~5.

1476: As expected, the vertical scale has been considerably enlarged.

1477: In return this procedure increases the statistical noise on the data points.

1478: The straight lines again show a constrained least-square fit,

1479: yielding $X_\infty\approx0.260$.

1480: 

1481: We can conclude from this numerical analysis that we have

1482: \beq

1483: X_\infty=0.26\pm0.01

1484: \eeq

1485: for the ferromagnetic Ising model in two dimensions.

1486: 

1487: \figqu

1488: 

1489: \section{Discussion}

1490: 

1491: In the present work we dealt with the dynamics of ferromagnetic

1492: spin systems quenched from infinite temperature to their critical state.

1493: This study, exemplified by the exact analysis of

1494: the spherical model in any dimension $D>2$,

1495: and by numerical simulations on the two-dimensional Ising model,

1496: complements that of the Glauber-Ising chain,

1497: presented in a companion paper~\cite{gl}.

1498: The main results obtained in this work can be summarized as follows.

1499: 

1500: In such a non-equilibrium situation, these systems are aging

1501: in the sense that their correlation and response functions

1502: depend non-trivially on the waiting time~$s$

1503: as well as on the observation time~$t$,

1504: whenever these two times are simultaneously large.

1505: The corresponding scaling laws (see eqs.~(\ref{gc}), (\ref{gdc}), (\ref{gr}),

1506: and~(\ref{grint})), involve powers of~$s$,

1507: related to the static anomalous dimension of the magnetization,

1508: and universal scaling functions of the ratio $x=t/s$.

1509: In the regime of large time separations, i.e., $1\ll s\ll t$ (or $x\gg1$),

1510: these scaling functions fall off algebraically

1511: with the common exponent $\lambda_c/z_c$.

1512: 

1513: The \fd ratio $X(t,s)$, characterizing the violation of the \fd theorem,

1514: has a universal scaling form $\X(x)$,

1515: and, for well-separated times in the aging regime,

1516: it assumes a limit value $X_\infty$ equal to a dimensionless amplitude ratio

1517: (see eqs.~(\ref{xr}) and~(\ref{xrho})).

1518: Therefore, as announced in ref.~\cite{gl},

1519: $X_\infty$ is a novel universal characteristic of critical dynamics,

1520: which is intrinsically related to the non-equilibrium initial condition

1521: of a critical quench from a disordered state.

1522: 

1523: \figci

1524: 

1525: The ferromagnetic models studied in the present work

1526: turn out to have values of $X_\infty$ in the range

1527: \beq

1528: 0\le X_\infty\le\frac{1}{2}.

1529: \eeq

1530: We have indeed $X_\infty=1-2/D$

1531: if $2<D<4$, and $X_\infty=1/2$ for $D>4$, for the spherical model,

1532: and $X_\infty\approx0.26\pm0.01$ for the two-dimensional Ising model.

1533: Let us mention that preliminary simulations

1534: on the three-dimensional Ising model yield $X_\infty\approx0.40$.

1535: The backgammon, for which $X_\infty=1$~\cite{fr97,gl99},

1536: thus belongs to another class of models.

1537: 

1538: The mean-field value

1539: \beq

1540: X_\infty^\mf=\frac{1}{2},

1541: \eeq

1542: obtained for the spherical model in dimension $D>4$,

1543: also holds for a variety of models which are not mean-field-like,

1544: including the Glauber-Ising chain~\cite{gl}

1545: and the two-dimensional X-Y model at zero temperature~\cite{ckp}.

1546: 

1547: Let us finally discuss a few open questions.

1548: It would be interesting to know whether there is an analogue for the present

1549: case of the results found for models with discontinuous spin-glass transitions,

1550: where the violation of the \fd theorem

1551: is related to the configurational entropy~\cite{config}.

1552: One would also like to know the status of the quantity $X_\infty$

1553: for non-equilibrium systems with quenched disorder,

1554: or for systems defined by dynamical rules without detailed balance.

1555: 

1556: In principle the limit \fd ratio $X_\infty$ could be calculated

1557: by field-theoretical renormalization-group methods,

1558: generalizing the computations done for universal amplitude ratios

1559: in usual static critical phenomena~\cite{zinn},

1560: as series in either $\eps=4-D$, or in $1/n$ for the $n$-component

1561: Heisenberg model, the spherical model corresponding to the $n\to\infty$ limit.

1562: The dimensionless time ratio $x=t/s$,

1563: appearing in the two-time autocorrelation and response functions and \fd ratio,

1564: is a temporal analogue of aspect ratios,

1565: which play an important role in static critical phenomena

1566: and finite-size scaling theory~\cite{fss}.

1567: One may therefore wonder whether the latter,

1568: and especially its latest developments involving conformal

1569: and modular invariance,

1570: could be used in order to put constraints on non-equilibrium critical dynamics.

1571: Generalized symmetry groups, such as those introduced in ref.~\cite{henkel},

1572: may also play a role in this issue.

1573: 

1574: \subsubsection*{Acknowledgements}

1575: 

1576: It is a pleasure for us to thank A.J. Bray and S.~Franz for interesting

1577: discussions.

1578: 

1579: \newpage

1580: \section*{Table and caption}

1581: 

1582: \vskip .6cm

1583: \begin{center}

1584: \begin{tabular}{|c|c|c|c|}

1585: \hline

1586: exponent&spherical $(2<D<4)$&spherical $(D>4)$&Ising $(D=2)$\\

1587: \hline

1588: $\eta$&$0$&$0$&$1/4$\\

1589: $\beta$&$1/2$&$1/2$&$1/8$\\

1590: $\nu$&$1/(D-2)$&$1/2$&$1$\\

1591: \hline

1592: $z$&$2$&$2$&$2$\\

1593: $\lambda$&$D/2$&$D/2$&$\approx1.25$\\

1594: \hline

1595: $z_c$&$2$&$2$&$\approx2.17$\\

1596: $\lambda_c$&$3D/2-2$&$D$&$\approx1.59$\\

1597: $\Theta_c$&$1-D/4$&$0$&$\approx0.19$\\

1598: \hline

1599: \end{tabular}

1600: \end{center}

1601: 

1602: \vskip .6cm

1603: \noindent{\bf Table~1:}

1604: Static and dynamical exponents of the ferromagnetic spherical model

1605: and of the two-dimensional Ising model.

1606: First group: usual static critical exponents $\eta$, $\beta$, and $\nu$

1607: (equilibrium).

1608: Second group: zero-temperature dynamical exponents~$z$ and $\lambda$

1609: (coarsening below $T_c$).

1610: Third group: dynamic critical exponents $z_c$, $\lambda_c$, and $\Theta_c$

1611: (non-equilibrium critical dynamics).

1612: 

1613: \newpage

1614: \begin{thebibliography}{99}

1615: 

1616: \bibitem{bray} A.J. Bray, Adv. Phys. {\bf 43}, 357 (1994).

1617: 

1618: \bibitem{revue} For recent reviews, see: E. Vincent, J. Hammann, M. Ocio,

1619: J.P. Bouchaud, and L.F. Cugliandolo,

1620: in {\it Complex Behavior of Glassy Systems},

1621: Springer Lecture Notes in Physics {\bf 492}, 184 (1997),

1622: preprint cond-mat/9607224;

1623: J.P. Bouchaud, L.F. Cugliandolo, J. Kurchan, and M.~M\'ezard,

1624: in {\it Spin Glasses and Random Fields},

1625: Directions in Condensed Matter Physics, vol.~{\bf 12},

1626: edited by A.P. Young (World Scientific, Singapore, 1998),

1627: preprint cond-mat/9702070.

1628: 

1629: \bibitem{ckp} L.F. Cugliandolo, J. Kurchan, and G. Parisi, J. Phys. I (France)

1630: {\bf 4}, 1641 (1994).

1631: 

1632: \bibitem{ck} L.F. Cugliandolo and J. Kurchan, J. Phys. A {\bf 27}, 5749 (1994).

1633: 

1634: \bibitem{cd} L.F. Cugliandolo and D.S. Dean, J. Phys. A {\bf 28}, 4213 (1995).

1635: 

1636: \bibitem{x1} L.F. Cugliandolo, J. Kurchan, and L. Peliti, Phys. Rev. E

1637: {\bf 55}, 3898 (1997).

1638: 

1639: \bibitem{barrat} A. Barrat, Phys. Rev. E {\bf 57}, 3629 (1998).

1640: 

1641: \bibitem{berthier} L. Berthier, J.L. Barrat, and J. Kurchan,

1642: Eur. Phys. J. B {\bf 11}, 635 (1999).

1643: 

1644: \bibitem{autres} S. Franz, M. M\'ezard, G. Parisi, and L. Peliti,

1645: Phys. Rev. Lett. {\bf 81}, 1758 (1998); J. Stat. Phys. {\bf 97}, 459 (1999).

1646: 

1647: \bibitem{zkh} W. Zippold, R. K\"uhn, and H. Horner, preprint cond-mat/9904329.

1648: 

1649: \bibitem{fr97} S. Franz and F. Ritort, J. Phys. A {\bf 30}, L 359 (1997),

1650: and references therein.

1651: 

1652: \bibitem{gl99} C. Godr\`eche and J.M. Luck, J. Phys. A {\bf 32}, 6033 (1999),

1653: and references therein.

1654: 

1655: \bibitem{gl} C. Godr\`eche and J.M. Luck, preprint cond-mat/9911348,

1656: J. Phys. A {\bf 33} (2000), to appear.

1657: 

1658: \bibitem{zan} E. Lippiello and M. Zannetti, preprint cond-mat/0001103,

1659: Phys. Rev. E (2000), to appear.

1660: 

1661: \bibitem{janssen} H.K. Janssen, B. Schaub, and B. Schmittmann, Z. Phys. B

1662: {\bf 73}, 539 (1989).

1663: 

1664: \bibitem{fisher} D.S. Fisher and D.A. Huse, Phys. Rev. B {\bf 38}, 373 (1988).

1665: 

1666: \bibitem{berlin} T.H. Berlin and M. Kac, Phys. Rev. {\bf 86}, 821 (1952).

1667: 

1668: \bibitem{baxter} R.J. Baxter, {\it Exactly Solved Models in Statistical

1669: Mechanics} (Academic Press, London, 1982).

1670: 

1671: \bibitem{lprb} J.M. Luck, Phys. Rev. B {\bf 31}, 3069 (1985).

1672: 

1673: \bibitem{mezard} M. M\'ezard, G. Parisi, and M. Virasoro, {\it Spin-glass

1674: Theory and Beyond} (World Scientific, Singapore, 1987).

1675: 

1676: \bibitem{langer} J.S. Langer, in {\it Solids far from Equilibrium}, edited by

1677: C. Godr\`eche (Cambridge University Press, 1991).

1678: 

1679: \bibitem{bray2} A.J. Bray, ICTP Summer School on ``Statistical Physics of

1680: Frustrated Systems'',

1681: \texttt{http://www.ictp.trieste.it/$\sim$pub\_off/sci-abs/smr1003/index.html}

1682: (1997).

1683: 

1684: \bibitem{huse} D.A. Huse, Phys. Rev. B {\bf 40}, 304 (1989).

1685: 

1686: \bibitem{blote} M.P. Nightingale and H.W.J. Bl\"ote,

1687: Phys. Rev. Lett. {\bf 76}, 4548 (1996).

1688: 

1689: \bibitem{grass} P. Grassberger, Physica A {\bf 214}, 547 (1995);

1690: Erratum: Physica A {\bf 217}, 227 (1995).

1691: For a recent review, see: B.~Zheng, preprint cond-mat/9910504,

1692: Physica A, to appear.

1693: 

1694: \bibitem{config} S. Franz and M.A. Virasoro, preprint cond-mat/9907438.

1695: 

1696: \bibitem{zinn} J. Zinn-Justin, {\it Quantum Field Theory and Critical

1697: Phenomena} (Clarendon, Oxford, 1989).

1698: 

1699: \bibitem{fss} J. Cardy~(ed.), {\it Finite-Size Scaling} (North Holland,

1700: Amsterdam, 1988); V. Privman~(ed.), {\it Finite-Size Scaling and Numerical

1701: Simulation of Statistical Systems} (World Scientific, Singapore, 1990).

1702: 

1703: \bibitem{henkel} M. Henkel, Phys. Rev. Lett. {\bf 78}, 1940 (1997).

1704: 

1705: \end{thebibliography}

1706: \end{document}

1707: