1: \documentstyle[multicol,pre,aps,epsf]{revtex}
2: %\documentstyle[prl,aps,preprint,epsf]{revtex}
3: \begin{document}
4:
5: \title{Novel glassy behavior in a ferromagnetic p-spin model}
6:
7: \author{Michael R. Swift$^{1,3,}$\cite{a1}}
8: \address{$^1$School of Physics and Astronomy, University of
9: Nottingham, Nottingham NG7 2RD, UK}
10: \author{Hemant Bokil$^{2,3,}$\cite{a2}}
11: \address{$^2$Abdus Salam ICTP, Strada Costiera 11, Trieste 34100, Italy}
12: \author{Rui D.M. Travasso$^3$ and Alan J. Bray$^3$}
13: \address{$^3$Department of Theoretical Physics, University of
14: Manchester, Manchester M13 9PL, UK}
15:
16: \date{\today}
17: \maketitle
18:
19: \begin{abstract}
20: Recent work has suggested the existence
21: of glassy behavior in a ferromagnetic
22: model with a four-spin interaction. Motivated by
23: these findings, we have studied the dynamics of
24: this model using Monte Carlo simulations with particular
25: attention being paid to two-time quantities.
26: We find that the system shares many features in common with
27: glass forming liquids. In particular,
28: the model exhibits: (i) a very long-lived metastable state,
29: (ii) autocorrelation functions that
30: show stretched exponential relaxation, (iii) a non-equilibrium
31: timescale that
32: appears to diverge at a well defined
33: temperature, and (iv) low temperature
34: aging behaviour characteristic
35: of glasses.
36:
37: \end{abstract}
38:
39: \pacs{PACS numbers: 64.70.Pf, 75.10.Hk, 75.10.Nr}
40:
41: \begin{multicols}{2}
42:
43: \section{Introduction}
44: Many liquids when cooled below the melting temperature
45: do not crystallize. Instead,
46: as the temperature is lowered, the relaxation time
47: (or the viscosity) increases dramatically, eventually becoming
48: so large that the liquid appears frozen on experimental
49: timescales. For all practical purposes the system has solidified
50: yet there is no long range order. The system is said to have
51: become a glass. Despite over fifty years of work, a consensus is lacking on
52: a theory of this phenomenon\cite{deb}.
53: Even as basic a question as the existence of a genuine thermodynamic
54: glass transition is difficult to decide experimentally.
55: Therefore, much work in this field has been numerical. Simulations
56: in the continuum of glassy systems remain difficult, and,
57: despite the substantial progress in this direction\cite{kob},
58: there are still many unresolved issues.
59:
60: Because of these problems it would be useful to have simplified lattice
61: models that behave in some respects like glasses and which do not
62: have quenched disorder. Recently,
63: Lipowski\cite{lip1} and Lipowski and Johnston\cite{lip2,lip3}
64: studied such a model by Monte Carlo simulations and provided
65: evidence of glassy behavior.
66: The model is defined by the Hamiltonian
67: \begin{equation}
68: H= -J\sum_{\Box} S_i S_j S_k S_l,
69: \label{eqn1}
70: \end{equation}
71: where the spins sit on a cubic lattice and the interactions
72: are between the four spins at the corners of each plaquette.
73: The spin variables take values $S_i = \pm 1$ and
74: $J$ is the strength of the ferromagnetic coupling.
75: (We will take $J=1$ in all that follows.)
76: Their study was motivated
77: by the fact that the spin-glass version of
78: this model was found numerically
79: \cite{silvio} to have several features in common with structural glasses.
80: In addition, there is a close connection\cite{lip1} between the plaquette
81: model, Eq.~\ref{eqn1}, and the Ising model with competing interactions
82: studied in the context of glasses by Shore and Sethna\cite{ss}.
83: Lipowski and Johnston\cite{lip2} showed that, in equilibrium, the
84: plaquette model has
85: a first order transition
86: as the temperature is lowered but that
87: this transition is never observed in simulations within accessible timescales.
88: Instead, one sees what appears to be a
89: dynamical transition to a glassy state\cite{lip2,lip3}.
90:
91: The work of these authors concentrated on single time
92: quantities. To see clear signatures of glassy
93: behaviour and to understand which particular aspects of
94: glasses are present in this model, it is important
95: to study two time quantities such as autocorrelation
96: and response functions. We do just that in this paper and
97: our findings are organized as follows: in section
98: II, we summarize the results of Lipowski and Johnston
99: on the statics of the plaquette model and the
100: strong metastability that they observed. In section III, we examine
101: the metastable state in greater detail. In particular, we
102: calculate the timescale for nucleation and show that
103: it is indeed extremely large.
104: In section IV, we present results for the energy
105: auto-correlation function in the supercooled
106: phase which is unambiguously described by a stretched exponential.
107: We investigate the behavior of the associated timescale as
108: a function of temperature, showing that it diverges with
109: a power law at a temperature indistinguishable from the
110: lower limit of metastability. This temperature is that which
111: the authors of \cite{lip1,lip2} called
112: the glass transition temperature $T_g$.
113: In section V, we study the dynamical behavior
114: below $T_g$. The spin-spin autocorrelation function and overlap distribution
115: function exhibit aging of a form that is characteristic of
116: glasses, showing that the plaquette model has a complex free-energy
117: landscape.
118: These findings provide further evidence to support the conjecture
119: that the plaquette model exhibits an ideal glass transition.
120:
121: \section{The Plaquette Model}
122:
123: In equilibrium, the model defined by Eq.~\ref{eqn1} has a first order
124: transition at $T_c\simeq 3.6\cite{lip2,esp,beppe}$.
125: This system possesses a novel symmetry: flipping
126: all the spins in any plane of the cubic lattice leaves
127: the Hamiltonian invariant, leading to a degeneracy of ground
128: states $\sim 2^{dL}$ for a cubic system of size $L$
129: in $d$ dimensions.
130: Note, however, that the ground state entropy is finite.
131: Because of this symmetry, the magnetisation
132: is not a good order parameter.
133: Instead, one can use the internal energy per
134: spin, $E$, to investigate this model further.
135:
136: Lipowski and Johnston looked at the dynamics of $E$ following a quench
137: from a high temperature state to a temperature $T<T_c$\cite{lip2}.
138: They found different behaviour depending on the value of
139: $T$. In particular, for $3.4<T<T_c$, $E$ appears
140: not to relax to its value in the low temperature phase.
141: Rather it settles into a plateau at an energy
142: higher than that of the true low temperature phase at the
143: same temperature.
144: Furthermore, the time the system spends on the plateau (in a ``supercooled''
145: state) appears to increase at least exponentially with system size,
146: suggesting the existence of a genuine metastable phase.
147: Such behaviour , if present, would indeed be strange because
148: one does not expect true
149: metastability (i.e. two local minima of the free energy
150: separated by diverging free energy barriers)
151: in systems with finite range interactions. Moreover,
152: nucleation theory tells us that the larger the
153: system, the shorter the lifetime of a metastable
154: state, provided that
155: the system is large enough to support a nucleating droplet.
156: To understand whether the strong metastability
157: is indeed a finite size artifact, Lipowski and Johnston studied
158: a quench starting from a configuration with a droplet of the low
159: temperature phase in a high temperature background.
160: They reported that for sufficiently large seeds, the
161: system always relaxed to the low temperature phase,
162: concluding that ordinary nucleation was operative. However, this
163: does not explain why the supercooled phase appears
164: so stable on observable timescales. The
165: analysis we report in the next section provides an explanation
166: for this phenomenon.
167:
168: For quenches to below $T=3.4$ the behaviour was found to be
169: quite different. The
170: plateau disappeared and, after initial transients,
171: the energy decreased extremely slowly. Moreover,
172: it appeared to relax not to the value in the low temperature
173: phase, but to a value extensively higher.
174: They identified this state as
175: the glassy phase. Finally, they have recently shown that a characteristic
176: zero-temperature length increases very slowly with the inverse
177: cooling rate, as is seen in ordinary glasses\cite{lip3}.
178:
179: To summarize, this model has many
180: properties that are similar to those seen in real glasses:
181: a first order transition analogous to crystal melting,
182: strong metastability effects in the supercooled phase,
183: and a kinetic transition to a glassy state characterised
184: by slow dynamics.
185:
186: In the next section we will discuss the
187: anomalous metastability described above before
188: looking in some detail at the
189: dynamical behavior of this model by studying
190: the autocorrelation functions both in the supercooled phase
191: and in the ``glassy'' state.
192:
193: \section{Metastability}
194:
195: We begin by summarizing the basics of nucleation theory\cite{nuc}.
196: The idea is that if the system is in a metastable
197: state, the process of finding the true free-energy minimum involves the
198: formation of a droplet of the stable phase which can
199: grow and take over the whole system. Phenomenologically,
200: one can express the free energy barrier to forming a droplet of linear
201: dimension
202: $R$ as $\Delta = A\sigma R^{d-1} - B\delta f R^d$ in $d$ dimensions.
203: Here $\sigma$ is the surface tension between the metastable
204: and stable phases and $\delta f$ is the difference in
205: the bulk free energy densities of the two phases. A and B are
206: constants characterizing the geometrical shape of the droplet.
207: Because
208: the surface term is a cost and the bulk term
209: is a gain, one can maximize $\Delta$ to find a critical radius
210: $R^{*} = (A(d-1)/Bd)(\sigma/\delta f)$. Droplets of a radius
211: greater than $R^{*}$ can grow and take over the whole
212: systems while those of radius less that $R^{*}$ will
213: shrink. At low temperatures and
214: in three dimensions, the timescale for nucleation of
215: a droplet is then given by $\tau_{\rm nuc} \sim
216: \exp(4 A^3 \sigma^3/27 B^2 \delta f^2 T)$.
217:
218:
219: For the plaquette problem, the free energy difference between the two
220: phases can be estimated from the data in \cite{lip2}. We find
221: $\delta f \approx 0.5 |T-T_c|$. The surface tension $\sigma$ at $T_c$
222: can be calculated from the size dependence of the mean time
223: it takes the system to flip between the high and low temperature
224: phases. This time $\tau_{\rm flip}$ is given by
225: $\tau_{\rm flip} \sim e^{{2\sigma L^2}/ {T_c}}$, where $L$ is the system
226: size and the factor of $2$ comes from using period boundary conditions.
227: We have performed standard single spin-flip Monte Carlo simulations
228: to determine this transition time numerically for systems of size
229: $L=4,5,6,7,8$. The transitions are identified from jumps in the energy
230: between the values appropriate to the high and low temperature phases.
231: For each system size, an `effective transition temperature', $T_c(L)$, is
232: determined by the requirement that the system spend equal time in both
233: phases, i.e.\ at each of the two possible values of the energy.
234: This gives $T_c(L) \simeq 4.45$, 4.19, 4.03, 3.92 and 3.86 for
235: $L=4$, 5, 6, 7 and 8 respectively. The mean transition time,
236: $\tau_{\rm flip}(L)$, is measured at $T_c(L)$, and the surface tension,
237: $\sigma$, obtained by plotting $\ln \tau_{\rm flip}(L)$ against
238: $L^2/T_c(L)$. The data is presented in Fig.~\ref{Fig1}. For large $L$,
239: the points should approach a straight line with slope $2\sigma$. Since
240: there is clear curvature in the data for the small values of $L$
241: available, we attempt an extrapolation to large $L$ by plotting
242: $[T_c(L)/L^2]\ln \tau_{\rm flip}$ against $T_c{L}/L^2$, as shown in the
243: inset to Fig.~\ref{Fig1}. The intercept on the vertical axis gives $2\sigma$.
244: While the extrapolation to $L=\infty$ is necessarily subjective, from
245: the curvature of the data it is clear that any reasonable procedure
246: will give a value of $2\sigma$ greater than the value $\simeq 0.62$
247: obtained from a linear extrapolation of the last two points, and less than
248: the value $\simeq 0.69$ given by the $L=8$ point. To obtain lower bounds
249: on critical drop sizes and nucleation times we use the estimate
250: $\sigma \simeq 0.31$ obtained from the linear extrapolation.
251: Assuming that $\sigma$ depends only weakly on $T$ close to $T_c$, and
252: taking $T_c=3.6$, we estimate the critical droplet
253: radius at $T=3.5$ to be $R^{*} \approx 25$ for a cubic droplet
254: ($A=6$, $B=1$) and $\approx 12$ for a spherical droplet
255: ($A=4\pi$, $B=4\pi/3$). The corresponding time to nucleate
256: a cubic droplet is of the order of $10^{47}$ MCS (and $\sim 10^{25}$
257: MCS for a spherical droplet).
258:
259: \begin{figure}
260: \narrowtext
261: \centerline{\epsfxsize\columnwidth\epsfbox{fig1.eps}}
262: \caption{Logarithm of the flip time between high and low
263: temperature phases at $T_c$ as a function of system size.
264: The errors are smaller than the symbols plotted. The inset shows the
265: extrapolation for large $L$.}
266: \label{Fig1}
267: \end{figure}
268:
269: These findings provide an explanation for the
270: results of \cite{lip2}, in which it was noted that the relaxation time
271: of the metastable phase grows rapidly with system size.
272: In \cite{lip2}, the system sizes simulated were all less than,
273: or of the same order as, $R^{*}$.
274: In this limit, the free energy barrier to nucleation is set by the
275: system size alone, as a droplet of the equilibrium phase cannot fully form.
276: Consequently,the lifetime of the metastable state will
277: grow exponentially with the size of the system\cite{lip2} as reported.
278: We thus conclude that the supercooled state is indeed metastable
279: but its lifetime is many orders of magnitude larger than
280: the times accessible in Monte Carlo simulations.
281:
282: \section{Dynamics in the supercooled phase}
283:
284: The philosophy underlying the search for models without
285: disorder that show glassy behavior is that in glasses,
286: inhomogeneities are self-generated, even though the microscopic Hamiltonian
287: has no intrinsic disorder. Thus there is
288: a spontaneous
289: separation of the degrees of freedom into variables that exhibit
290: either slow or fast relaxation.
291: One expects that it is the slow degrees of
292: freedom that freeze as the glass transition is approached.
293: However, this separation
294: is not necessarily a priori obvious, and one has to
295: be careful about which quantities one studies. It might happen,
296: for example, that the spin-spin autocorrelations
297: do not show any anomalous behavior, but that some complicated
298: functions of the spins show a dramatic slowing down. For this
299: reason it is often useful to look at autocorrelation functions
300: of global quantities because these presumably pick out the
301: longest timescale or the slowest degree of freedom.
302:
303: For the system defined by Eq.~\ref{eqn1},
304: one such quantity is the energy per spin, $E$, and we now discuss
305: our findings for the energy autocorrelation function
306: in the range of temperatures $T_g<T<T_c$.
307: Our results have been obtained from Monte Carlo simulations
308: using single spin-flip Glauber dynamics. We have considered
309: cubic systems of linear size $L$, with $L$ ranging from $12$ to $64$.
310: Most of the data presented here is for a system size of $L=48$.
311: To measure the energy autocorrelation function at some temperature $T$
312: we start from a random configuration (infinite temperature)
313: and quench to the temperature $T$. We then wait for a time $t_w$
314: and measure the autocorrelation function $A(t,t_w)$
315: for subsequent times, $t$, where
316: $A(t,t_w) = <E(t_w) E(t_w+t)>$. The angular brackets denote
317: averages over random initial conditions.
318: In equilibrium one expects
319: $A(t,t_w)$ to be independent of $t_w$ and it is only in equilibrium
320: that one can define a meaningful timescale associated with relaxation. In
321: the supercooled phase, we found that equilibrium is easily
322: achieved in terms of this definition even though the system
323: is in a metastable state. In particular, for any
324: temperature $T_g<T<T_c$, the autocorrelation function becomes
325: independent of $t_w$ once $t_w$ becomes of the order of $10^3$ MCS. Note
326: that this time is at least $10$ orders of magnitude smaller than the lifetime
327: of this metastable state.
328:
329: In Fig.~\ref{Fig2} we show $A(t,t_w)$ for three different temperatures
330: below $T_c$ and above $T_g$, together with attempted fits
331: to stretched exponentials of the form
332: \begin{equation}
333: A(t,t_w)= A_0\, e^{-\left(\frac{t}{\tau}\right)^\beta}.
334: \label{eqn2}
335: \end{equation}
336: As can be seen from the figure, the fits are extremely good and
337: we can identify a timescale for relaxation, $\tau$, at each temperature.
338: The exponent $\beta$ is also temperature dependent and is found
339: to lie in the range $0.5 < \beta < 1$.
340: In Fig.~\ref{Fig3} we plot the behaviour of $\tau$
341: as a function of $T$. It appears to increase sharply as $T$ is reduced
342: and can be fit quite accurately by a power-law divergence ,
343: \begin{equation}
344: \tau=\frac{2.23}{T-3.39},
345: \label{eqn3}
346: \end{equation}
347: as shown in the figure.
348: This gives a ``critical'' temperature very close to that which was identified
349: in \cite{lip2} as the lower limit of stability, $T_g$, and
350: interpreted as the glass transition temperature.
351: Both stretched exponential relaxation functions and a diverging
352: relaxation time are features of real glasses. We also tried a Vogel-Fulcher
353: (stretched-exponential) fit
354: to $\tau$ but found that the corresponding ``critical
355: temperature'' is significantly lower than $T_g$. As we
356: will show in the next section, $T_g$ signifies the onset of aging phenomena
357: and it is thus impossible to define a timescale below this temperature.
358: (We have also measured the spin-spin autocorrelation
359: function but found that the behaviour was much harder to interpret.
360: In fact, the difficulty of interpreting relaxational behavior
361: of local quantities has been noted before in molecular
362: dynamics studies of Lennard-Jones systems in two dimensions
363: \cite{klein}.)
364: \begin{figure}
365: \centerline{
366: \epsfxsize=0.9\hsize{\epsfbox{fig2.eps}}}
367: \narrowtext{\caption{The energy autocorrelation
368: function $A(t,t_w)$ in the supercooled phase. Three temperatures
369: are shown: $T=3.6 (\circ), 3.5 (\triangle), 3.42 (\diamond)$. The lines
370: are fits to stretched exponentials, Eq.~\ref{eqn2}, with
371: parameters: $\tau=11, \beta=0.7$ at $T=3.6$,
372: $\tau=21, \beta=0.66$ at $T=3.5$ and $\tau=72, \beta=0.6$ at $T=3.42$.}
373: \label{Fig2}}
374: \end{figure}
375: A possible explanation for this divergence is to
376: note that in mean-field
377: theory, the plaquette model has a first order transition at
378: a temperature quite close to $T_c$\cite{beppe}. In any mean-field
379: first order transition, one expects a region of metastability
380: delimited by the spinodal temperature. Above this temperature,
381: there is genuine metastability and one expects timescales of
382: autocorrelation functions to diverge as the spinodal
383: temperature is approached
384: When one moves away from mean-field,
385: genuine metastability is no longer present and instead
386: the spinodal gets rounded out. Our
387: results for the autocorrelation functions, and those of the authors
388: of \cite{lip2} for the energy relaxation, suggest that
389: the spinodal is actually remarkably well-defined in this
390: short-range system. It is therefore plausible that what we are calling
391: the glass transition
392: temperature is really a remnant of the mean-field spinodal. This too is
393: something that is seen in Lennard-Jones systems in two dimensions
394: \cite{klein}. However, the stretched exponential form of
395: $A(t, t_w)$ would not be expected in any simple mean-field treatment.
396:
397: Thus, the short-range plaquette
398: model has a first order transition analogous to crystal melting, but
399: because of the extraordinarily long-lived supercooled phase, it also
400: exhibits timescales for relaxation that diverge at $T_g$.
401: We would like to emphasize that in this model
402: $T_g$ is not strictly well-defined. In fact, we expect that as one gets
403: closer and closer to $T_g$, or waits for longer in the metastable
404: state, the divergence we are seeing will
405: get rounded out. Numerically, however, it is not possible to see this
406: rounding in any reasonable amount of time.
407: This again is a feature in common with glass forming liquids.
408:
409: \begin{figure}
410: \centerline{
411: \epsfysize=0.8\columnwidth{\epsfbox{fig3.eps}}}
412: \narrowtext{\caption{The timescale of relaxation $\tau$ as a function
413: of temperature. The curve is the best fit to a power-law
414: divergence given by Eq.~\ref{eqn3}.}\label{Fig3}}
415: \end{figure}
416:
417: \section{Dynamics in the glassy phase}
418:
419: For a quench to a temperature $T>T_g$, as we have seen above, the
420: system becomes trapped in a metastable state for times up to $10^{24}$
421: MCS. Within this time window, two-time correlation functions
422: become waiting-time independent and the system behaves as if
423: in equilibrium. This is no longer the case if $T<T_g$. The system
424: now continues to evolve slowly for all accessible times and two-time
425: quantities exhibit aging. To investigate this behaviour we have
426: measured the spin-spin autocorrelation function,
427: \begin{equation}
428: C(t,t_w)=\frac{1}{N} \sum_i S_i(t_w) \, S_i(t+t_w),
429: \label{eqn4}
430: \end{equation}
431: for a range of waiting-times $t_w$.
432: Our results are shown in Fig.~\ref{Fig4}. There is a strong dependence
433: on $t_w$ for all times and temperatures we have considered.
434: In the long time regime, $C(t, t_w)$ decays as a power of $t$
435: with an exponent of around $0.35$.
436: Similar behaviour has been noted in MD simulations of
437: binary Lennard-Jones mixtures\cite{kob2}.
438: We have also attempted to collapse the data at a fixed temperature
439: with a scaling form
440: $C(t,t_w)=\tilde C(t/\tau(t_w))$ where $\tau(t_w) \sim t_w^{\alpha}$.
441: While the fit is not perfect, our best estimate for $\alpha$ is
442: $\alpha=2$. This type of behaviour has been termed super-aging\cite{bouchaud}
443: as the characteristic timescale for relaxation, $\tau(t_w)$, grows
444: faster than the weighting time itself.
445: Note that the spin-glass version of this model exhibits sub-aging with
446: $\alpha \approx 0.77$\cite{silvio}.
447:
448: \begin{figure}
449: \centerline{
450: \epsfysize=0.8\columnwidth{\epsfbox{fig4.eps}}}
451: \narrowtext{\caption{The spin autocorrelation
452: function $C(t,t_w)$ in the glassy phase at temperature $T=3.2$.
453: Eight different different waiting
454: times are shown. From bottom to top $t_w= 10$, $100$, $200$, $500$, $1000$,
455: $2000$, $5000$, $10000$.}
456: \label{Fig4}}
457: \end{figure}
458:
459: However, simply having a strong waiting-time dependence in a
460: two-time correlation function does not imply glassy behaviour.
461: Indeed, simple coarsening systems exhibit aging\cite{alan}. To distinguish
462: this type of dynamics, type I aging\cite{barat} , from glassy dynamics,
463: type II aging, we have considered the overlap distribution function
464: \begin{equation}
465: Q(t_w+t,t_w+t)=\frac{1}{N} \sum_i S_i^{(1)}(t+t_w) \, S_i^{(2)}(t+t_w).
466: \label{eqn5}
467: \end{equation}
468: This is measured
469: by first relaxing the system from a disordered state for a time
470: $t_w$. Then two copies of the system are made,
471: $\{S_i^{(1)}\}$ and $\{S_i^{(2)}\}$,
472: and each is evolved for a further time $t$ with independent thermal noise.
473:
474: In Fig.~\ref{Fig5} we show a plot of $Q(t_w+t,t_w+t)$ vs. $C(t,t_w)$
475: for three different waiting
476: times. In each case $Q(t_w+t,t_w+t) \to 0$ as $C(t,t_w) \to 0$, indicating
477: that the two copies of the system continue to move apart from
478: each other irrespective of when the copies were made. This behaviour
479: is indicative of type II, glassy aging\cite{barat}.
480: Note that
481: this behaviour is not an automatic consequence of
482: the existence of infinitely many ground states. For example,
483: we expect that for a triangular lattice anti-ferromagnet, $Q(t_w+t,t_w+t)$
484: would tend to a constant as $C(t,t_w)$ tends to zero, despite
485: the non-zero ground-state entropy per spin.
486: Finally, we have observed that $C(t, t_w)$ and $Q(t_w+t, t_w+t)$ are related by
487: $C(2t, t_w)=Q(t_w+t, t_w+t)$ over a wide range of timescales. This
488: relation is known to hold in equilibrium, where both functions
489: are independent of $t_w$\cite{barat}. It has also been verified in the aging regime
490: of some trap models as well as in the spin-glass version of the
491: plaquette model\cite{bouchaud}.
492:
493: These results provide evidence
494: that the free-energy landscape of the ferromagnetic plaquette model is indeed
495: complex and has many mutually inaccessible minima.
496: This property is similar to what
497: is seen in structural glasses and in mean field spin-glass models.
498:
499: \begin{figure}
500: \centerline{
501: \epsfysize=0.8\columnwidth{\epsfbox{fig5.eps}}}
502: \narrowtext{\caption{Plot of $Q(t_w+t,t_w+t)$ versus $C(t,t_w)$.
503: Three weighting times are shown: $t_w=100 (\circ), 1000 (\triangle),
504: 5000 (\diamond)$.}
505: \label{Fig5}}
506: \end{figure}
507:
508: \section{Conclusions}
509:
510: We have shown that the ferromagnetic plaquette model has many
511: characteristics in common with glass forming liquids. We
512: provided an explanation for the long-lived nature of the supercooled
513: phase in terms of standard nucleation theory.
514: By measuring the two-time correlation
515: functions, we have observed
516: two distinct kinds of behavior: (i) in the supercooled phase, the system
517: appears to be stationary and the timescale of the energy autocorrelation
518: functions diverges at a temperature identified in
519: earlier work as a glass transition temperature $T_g$, and (ii) below
520: $T_g$ the spin-spin
521: autocorrelation functions exhibit aging of a form characteristic
522: of glasses. Our findings, along with the work reported
523: in \cite{lip2,lip3}, indicate that a deeper understanding of
524: this model would indeed be worthwhile.
525:
526: \medskip
527:
528: \begin{center}
529: \begin{small}
530: {\bf ACKNOWLEDGEMENTS}
531: \end{small}
532: \end{center}
533:
534: This work was supported by the Engineering and Physical Sciences Research
535: Council (MS, HB and AB) and by the Funda\c{c}\~{a}o para a Ci\^{e}cia e a
536: Technologia (RT).
537:
538: \begin{references}
539:
540: \bibitem[*]{a1} Permanent address: School of Physics and Astronomy,
541: University of Nottingham.
542:
543: \bibitem[**]{a2} Present address: ICTP, Trieste, Italy.
544:
545: \bibitem{deb} For a review see Pablo Debenedetti,``Metastable Liquids''
546: (Princeton University Press, 1996).
547:
548: \bibitem{kob} W. Kob, J. Phys. Cond. Mat. {\bf 11}, 85 (1999)
549: and references therein.
550:
551: \bibitem{lip1} A. Lipowski, J. Phys. A{\bf 30}, 7365 (1997).
552:
553: \bibitem{lip2} A. Lipowski and D.A. Johnston, (cond-mat/9812098).
554:
555: \bibitem{lip3} A. Lipowski and D.A. Johnston, (cond-mat/9910370).
556:
557: \bibitem{silvio} D. Alvarez, S. Franz and F. Ritort, Phys. Rev. B
558: {\bf 54}, 9756 (1996).
559:
560: \bibitem{ss} J. D. Shore, and J. P. Sethna, Phys. Rev. B {\bf 43},
561: 3782 (1991); See also J. D. Shore, M. Holzer, and J. P. Sethna,
562: Phys. Rev. B {\bf 46}, 11376 (1992).
563:
564: \bibitem{esp} D. Espriu, M. Baig, D. A. Johnston, and
565: R. P. C. K. Malmini, J. Phys. A {\bf 30}, 405 (1997).
566:
567: \bibitem{beppe} E. N. M. Cirillo, G. Gonnella, D. A. Johnston,
568: A. Pelizzola, Phys. Lett. A {\bf 226}, 59 (1997).
569:
570: \bibitem{nuc} P. A. Rikvold, H. Tomita, S. Miyashita, and
571: S. W. Sides, Phys. Rev. E {\bf 49}, 5080 (1994) and references therein.
572:
573: \bibitem{klein}A. I. Mel'cuk, R. A. Ramos, H. Gould, W. Klein,
574: and R. D. Mountain,Phys. Rev. Lett. {\bf 75} 2522 (1995).
575:
576: \bibitem{kob2} W. Kob and J.-L. Barrat, Physica A {\bf 263}, 234 (1999).
577:
578: \bibitem{alan} A. J. Bray, Adv. in Phys. {\bf 43}, 357 (1994).
579:
580: \bibitem{barat} A. Barrat, R. Burioni and M. Mezard, J. Phys. A {\bf 29}
581: 1311 (1996).
582:
583: \bibitem{bouchaud} J.-P. Bouchaud, L. F. Cugliandolo,
584: J. Kurchan, and M. M{\' e}zard, in ``Spin Glasses and
585: Random Fields'', ed. A. P. Young (World Scientific, 1998).
586:
587: \end{references}
588:
589: \end{multicols}
590: \end{document}
591:
592: