1: \documentclass[twocolumn,prl,aps,showpacs,floatfix]{revtex4}
2: %\documentclass[preprint,prl,aps,showpacs,floatfix]{revtex4}
3: \usepackage{graphicx,epsf}
4:
5: \newcommand{\lsim}{\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}}
6: \newcommand{\gsim}{\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}}
7:
8: \begin{document}
9:
10: \title{ Large-Scale Simulations of the Two-Dimensional Melting
11: of Hard Disks }
12:
13: \author{C. H. Mak}
14:
15: \affiliation{
16: Department of Chemistry,
17: University of Southern California, \\
18: Los Angeles, California 90089-0482, USA }
19:
20:
21: \date{\today}
22:
23:
24: \begin{abstract}
25:
26: Large-scale computer simulations involving more than a
27: million particles have been performed to study the melting transition in
28: a two-dimensional hard disk fluid.
29: The van der Waals loop previously observed in the pressure-density
30: relationship of smaller simulations is shown to be an artifact
31: of finite-size effects. Together with a detailed
32: scaling analysis of the bond orientation order, the new results provide
33: compelling evidence for the Halperin-Nelson-Young picture.
34: Scaling analysis of the translational order also yields a
35: lower bound for the melting density that is much higher than previously
36: thought.
37:
38: \end{abstract}
39: \pacs{64.60.Fr, 64.70.Dv}
40:
41: \maketitle
42:
43:
44:
45: A system of hard disks in two dimension (2D) is one of the
46: simplest models of a classical fluid. But beneath the deceptive
47: simplicity of this model, 2D hard disks exhibit a set of surprisingly
48: rich behaviors.
49: Unlike in three dimensions, a 2D solid possesses only
50: quasi-long-range translational order which
51: decays algebraically to zero at large distances \cite{66mer1133}.
52: Instead of the usual first-order transition in three dimensions,
53: a 2D solid is also expected to melt into a liquid via
54: two continuous transitions.
55: The intervening phase called the ``hexatic'' was predicted
56: by Halperin and Nelson \cite{78hal121, 79nel2457} and
57: Young \cite{79you1855} to possess quasi-long-range bond orientation
58: order but no long-range translational order.
59:
60: Given the simplicity of the hard disk model, it would seem easy to
61: either prove or disprove the Halperin-Nelson-Young (HNY) theory by
62: detailed computer simulation studies. But twenty-five year after the HNY
63: theory was first proposed, simulations that
64: could definitively identify the nature of the melting transition
65: are still lacking \cite{02bin2323}.
66: The first simulation of 2D hard disks
67: was carried out by Alder and Wainwright \cite{62ald359}.
68: Based on the appearance of a van der Waals loop in the
69: pressure, they concluded that the melting transition must be
70: first-order. Since then, as more computing power has become available,
71: simulations have been carried out with increasingly larger system sizes
72: \cite{68hoo3609, 89zol9518, 92zol11187, 92lee11190,
73: 95web14636, 97mit6855, 95fer3477, 97fer750,
74: 98jas277, 99jas134, 99jas2594, 00bat5223, 00sen6294, 02bin2323,
75: 02mit184202, 02wat041110, 03woj939, 04nie4115, 04jas120, 04wat045103},
76: but instead of clarifying the picture, these simulations have provided
77: conflicting conclusions about the nature of melting transition.
78: One consensus that did emerge from the more recent simulation
79: studies is that the 2D hard disk system is very sensitive to finite-size
80: effects near the melting transition. This is not unexpected
81: if the transition is continuous, but compared to fluids with a soft
82: potential \cite{96bag255} the hard disk system is much more prone
83: to finite-size errors and boundary effects.
84: In a simulation of up to $N = 128^2$ particles, Zollweg and
85: Chester \cite{92zol11187} observed that the
86: equilibration time increased dramatically for densities very close to the
87: melting transition -- systems of this size were apparently
88: not large enough to reach the scaling limit.
89:
90: \begin{figure}[b]
91: \includegraphics[width=1.0\columnwidth]{fig1}
92: \caption[]{
93: Pressure of the hard disk fluid as a function of density.
94: Solid squares are $N = 256^2$ data from Jaster \cite{99jas2594}, and
95: crosses and plus, respectively, are $N = 512^2$ and $N = 1024^2$
96: data from Jaster \cite{04jas120}.
97: Open circles and open squares are data from the present work for
98: $N = 512^2$ and $1024^2$, respectively, with error bars as indicated.
99: The dotted line is a guide to the eye through Jaster's data for $N = 256^2$.
100: The dashed and solid line are guides to the eye through the $N = 512^2$
101: and $1024^2$ data in the present work.
102: Note the presence of an apparent
103: van der Waals (vdW) loop between $\rho$ = 0.895 and 0.910,
104: which becomes shallower for increasingly larger size simulations.
105: For $N = 1024^2$, the van der Waals loop between $\rho$ = 0.895 and 0.905 has
106: disappeared completely, with a small decrease in the pressure still visible
107: for $\rho$ = 0.910.
108: The two arrows indicate the approximate locations of the isotropic-hexatic and
109: hexatic-solid boundaries.
110: }
111: \end{figure}
112:
113: The largest simulation that has been performed to date was carried
114: out by Jaster with up to $N = 256^2$ particles \cite{99jas134,99jas2594},
115: and more recently for two higher densities with up to
116: $N = 1024^2$ \cite{04jas120}.
117: Even though a van der Waals loop was observed in the pressure
118: at densities between $\rho$ = 0.895 and 0.910 (solid squares in Fig.~1),
119: Jaster showed using a scaling analysis
120: that his data were also compatible with the HNY scenario.
121: A van der Waals loop is often the sign of a first-order transition,
122: but it may also arise from finite-size errors.
123: To definitively rule out a first-order scenario, one must
124: demonstrate that the van der Waals loop is a finite-size artifact,
125: i.e. it must be shown to disappear with larger size simulations.
126: Curiously, the same van der Waals loop was observed for
127: two different sizes in Jaster's data -- the pressure for $N = 128^2$
128: (not shown in Fig.~1) and $256^2$ coincide almost completely.
129:
130: In this letter, we describe a Monte Carlo study of 2D hard disks for
131: up to $N = 1024^2 = 1048576$ particles.
132: The calculations were carried out in the
133: canonical ensemble, in a square box with periodic boundary
134: condition and in a rectangular box with aspect ratio $\sqrt{3}:2$
135: for the higher densities.
136: We worked with densities in the range $\rho = 0.880$ to $0.920$, which
137: according to previous estimates should span the transition region
138: \cite{62ald359, 92zol11187, 95web14636, 99jas2594, 02bin2323}.
139: Densities $\rho$ are given in reduced units where the hard disk diameter
140: is one.
141:
142: While the rationale for going to larger system size is to
143: eliminate finite-size effects, larger simulations also take longer
144: to equilibrate.
145: We have focused
146: on $N = 512^2$ to try to carry out detailed simulations covering a large
147: range of densities between $\rho$ = 0.880 and 0.920.
148: At this size, one run at each density took several months of CPU time.
149: Additional larger simulations with $N = 1024^2$ were performed for
150: four densities between $\rho$ = 0.895 and 0.910 in the vicinity of
151: the van der Waals loop previously observed in smaller simulations.
152: In contrast, Jaster's recent simulations \cite{04jas120} focuses on
153: a different region in the phase diagram,
154: offering data for $\rho$ = 0.918 at $N = 1024^2$
155: and two densities, $\rho$ = 0.914 and 0.918, for $N = 512^2$.
156:
157: Two different types of Monte Carlo moves were used for our simulations.
158: The first is a conventional Metropolis move, where each particle is displaced
159: in a random direction by a random amount.
160: A second Monte Carlo move based on the cluster
161: algorithm proposed by Dress and Krauth \cite{95dre597} and
162: Liu and Luiijten \cite{04liu035504}
163: was also used.
164: At the densities we worked with, neither algorithm is
165: particularly efficient in causing very large rearrangements in the system
166: configuration. But by mixing two different algorithms that have vastly
167: different properties, we hope to minimize equilibration problems
168: characteristic of any single algorithm. One Monte Carlo step (MCS) in
169: our simulation is defined as having moved each particle on the average
170: once using the Metropolis algorithm, plus having made one global cluster
171: update.
172: The simulations reported here were carried out with no fewer than
173: 5 million MCS for each density. Depending on the equilibration rate,
174: results from the last 1 to 3 million MCS were used to collect statistics.
175: Two to four independent simulations were carried out for each density for
176: simulations with a square box, and five to six for those with a
177: rectangular box.
178:
179: The pressure $P$ was calculated using the virial formula $PA_0/NkT =
180: [1 + \pi\rho g(1^+)/2] \sqrt{3}\rho /2$, where $g(1^+)$ is the
181: contact value of the pair correlation function and $A_0 = \sqrt{3}N/2$
182: is the closed-packed area of the system.
183: The calculated pressure $P$ is shown in Fig.~1 as a function of
184: density $\rho$ for $N = 512^2$ (open circles) and for $N = 1024^2$
185: (open triangles).
186: Comparing the $N = 512^2$ and $1024^2$ data to
187: those from Jaster's simulation with $N = 256^2$, the two sets of data
188: are almost identical for $\rho \leq 0.890$, but inside the range $\rho$
189: = 0.895 to 0.910, the larger size simulations produced
190: a smaller pressure for $\rho$ = 0.895 but larger
191: pressures for $\rho$ = 0.900 to 0.910.
192: It is therefore clear that the apparent van der Waals loop in the pressure
193: is a result of finite-size effects, and using even larger size simulations,
194: this slight nonmonotonic decrease in the pressure should eventually vanish
195: altogether.
196: For the $N = 1024^2$ simulations, the van der Waals loops has
197: completely disappeared between $\rho$ = 0.895 and 0.905, with a slight
198: dip in $P$ still visible for $\rho$ = 0.910.
199: As expected, finite-size
200: effects are indeed very pronounced in the transition region even for
201: simulations of this magnitude.
202:
203: Even though the evidence in Fig.~1 is compelling that the
204: van der Waals loop is an artifact of finite-size effects, these data
205: alone cannot definitively rule out a first-order melting transition.
206: For this, wee need to carefully analyze the finite size effects.
207: To disentangle the finite-size effects, a
208: detailed scaling analysis must be performed on the simulation data.
209: We have found a subblock scaling analysis
210: \cite{95web14636, 96bag255} to be useful for this purpose.
211: With this method, a single large size simulation provides information on
212: multiple length scales simultaneously.
213: The subblock scaling analysis was applied to the bond orientation
214: order as well as the translational order.
215:
216: The bond orientation order is given by
217: $\psi_6^2 = \vert (6N)^{-1} \sum_l \sum_j \exp(6i\theta_{lj})\vert^2$,
218: where the sum goes over each particles $l$ and its
219: nearest neighbors $j$ and $\theta_{lj}$ is the angle between
220: the line from $l$ to $j$ with some fixed reference axis.
221: According to HNY theory, $\psi_6$
222: should have only short-range order in the isotropic phase and
223: quasi-long-range order in the hexatic phase with exponent $\eta_6 \leq 1/4$.
224: We calculated $\psi_6$ for subblock sizes of $L_B = L/64$, $L/32$
225: $\ldots$ $L$, where $L$ is the full length of the box and plot the
226: results in Fig.~2. For $\rho \leq 0.895$, $\psi_6$ clearly
227: scales to zero, but for $\rho \geq 0.900$, $\psi_6$ appears to scale to a
228: finite value.
229:
230: \begin{figure}
231: \includegraphics[width=1.0\columnwidth]{fig2}
232: \caption[]{
233: The bond orientation order parameter $\psi_6$ derived from the subblock
234: analysis as a function of density for the $N = 512^2$ simulations.
235: For $\rho \leq 0.895$, $\psi_6$ scales
236: to 0 with larger system sizes. For $\rho \geq 0.905$, $\psi_6$ appears
237: to scale to a nonzero value.
238: }
239: \end{figure}
240:
241: To establish the precise scaling behavior, we plot
242: $\ln \psi_6^2$ vs. the natural log of the length of the subblock $L_B$
243: in Fig.~3 for the $N = 512^2$ simulations.
244: According to HNY theory, this plot should show a slope $-\eta_6$ in
245: the hexatic phase and $-2$ in the isotropic phase where there is only
246: short-range order.
247: Figure~3 shows that for both $\rho$ = 0.880 (solid triangles) and 0.890
248: (open diamonds), the bond
249: orientation order has no long-ranged correlations in the long
250: length scale limit, and the size of the simulations was large enough to
251: reach the scaling limit.
252: We can safely conclude that
253: densities $\rho \leq 0.890$ are in the isotropic phase.
254: On the other hand, for the highest densities $\rho$ = 0.905
255: (solid diamonds) and 0.910 (open squares),
256: the bond orientation order shows an algebraic decay with an exponent
257: $\eta_6$ much smaller than 1/4. This is consistent with the
258: interpretation that these densities are either inside the hexatic or the
259: solid phase.
260:
261: \begin{figure}
262: \includegraphics[width=1.0\columnwidth]{fig3}
263: \caption[]{
264: Subblock scaling analysis for the bond orientation order parameter
265: for the $N = 512^2$ simulations.
266: The dotted line corresponds to a slope of $-2$ and the dashed line
267: a slope of $-1/4$. The inset shows an expanded view
268: for $\rho$ = 0.895, 0.900 and 0.905 in the large length scale region.
269: }
270: \end{figure}
271:
272: For the two densities $\rho$ = 0.895 and 0.900, the interpretation of the
273: subblock scaling plots is more involved.
274: The inset in Fig.~3 shows an expanded
275: view of their behaviors in the large length scale limit.
276: For $\rho$ = 0.900 (open circles), the bond orientation
277: order shows a slope that is very close to $-1/4$ at large length scales.
278: In the HNY scenario, this is consistent with a density inside the hexatic
279: phase, very close to the hexatic-isotropic boundary $\rho_i$.
280: These evidence suggest that $\rho_i \lsim 0.900$.
281:
282: For $\rho$ = 0.895 (closed squares), the subblock scaling plot
283: changes slope twice, first at $L/2$ and then more gradually
284: between $L/4$ and $L/8$.
285: The first abrupt slope change at $L/2$ is
286: an artifact of the subblock scaling analysis which has been discussed
287: by Weber, Marx and Binder \cite{95web14636}.
288: The reason for this sudden slope change
289: is that the subblocks and the full box actually
290: belongs to two different ensembles -- the
291: canonical for the full box and something resembling the grand
292: canonical for the subblocks.
293: It is therefore
294: possible for the full box to exhibit a different scaling behavior
295: compared to the subblocks when the correlation length exceeds the
296: size of the simulation box, in which case the full box data point must be
297: excluded from the scaling analysis. When this is done, the scaling behavior
298: suggests that the orientation order decays
299: algebraically with an exponent larger than $1/4$. But clearly the
300: scaling limit has not been reached, so it is possible that this
301: exponent will continue to increase with lengths beyond the size of
302: the present simulation.
303: These evidence suggest that $\rho$ = 0.895 must still be inside the
304: isotropic phase but is very close to the isotropic-hexatic boundary.
305:
306: Taken together, the pressure data and the subblock scaling
307: analysis of the bond orientation order reveal a consistent picture.
308: For densities $\rho \leq 0.895$, the
309: system is in the isotropic phase.
310: The van der Waals loop in the pressure between $\rho$ = 0.895 and 0.910
311: observed in previous simulations is most certainly due to finite-size effects.
312: The bond orientation
313: correlation length increases when
314: the isotropic-hexatic boundary $\rho_i$ is approached from below
315: and it changes
316: from short-range correlation to an algebraic decay with
317: $\eta_6$ close to 1/4 at $\rho_i \lsim 0.900$,
318: which is consistent with previous estimates \cite{99jas2594}.
319: Above $\rho_i$, the exponent $\eta_6$ decreases quickly from 1/4
320: to zero when the hexatic-solid boundary $\rho_m$ is approached from below.
321: These findings are consistent with the HNY scenario.
322:
323: The fact that $\eta_6 \to 0$ for $\rho \to 0.910$ has been used
324: previously to suggest that the hexatic-solid boundary is
325: at $\rho_m \approx 0.910$ \cite{95web14636,99jas2594}.
326: The recent data of Jaster, however,
327: have placed $\rho_m$ at a much higher value $\approx$ 0.933 \cite{04jas120}.
328: To more accurately locate the hexatic-solid boundary $\rho_m$, we turn to
329: a subblock scaling analysis of the translational order
330: $\psi_t^2 = \vert N^{-1} \sum_l \exp(i \vec{k}\cdot\vec{r}_l)\vert^2$,
331: where the wavevector $\vec{k}$ has magnitude $2\pi/(\sqrt{3}/2\rho)^{1/2}$.
332: In the solid phase, $\psi_t^2$ is expected to decay algebraically with
333: exponent $\eta_t$ = 1/3.
334: The results for three densities, $\rho$ = 0.900, 0.910 and 0.920, are
335: shown in Fig.~4 for the $N = 512^2$ simulations in both a square and a
336: rectangular box with a $\sqrt{3}:2$ aspect ratio.
337: For $\rho$ = 0.900 (triangles) and 0.910 (squares),
338: the results are consistent with no long-range
339: translational order in the large length scale limit for both box
340: geometries. This indicates that both of these densities are
341: inside the hexatic phase.
342: On the other hand, for $\rho$ = 0.920,
343: the translational order shows apparently
344: different scaling behaviors for the two box geometries -- no long-range
345: order in the square box but quasi-long-range
346: translational order with an apparent exponent $\eta_t > 1/3$ for the
347: rectangular box. In fact,
348: comparing the two different box geometries,
349: we found that the rectangular box simulations at this density
350: were much slower to equilibrate,
351: leading to the larger error bars on the right panel of Fig.~4
352: for $\rho$ = 0.920.
353: Since the translational correlation length is expected
354: to diverge according to HNY theory as $\rho$ approaches the melting
355: density $\rho_m$ \cite{79nel2457},
356: the rectangular box used for the simulations at $\rho$ = 0.920 was
357: probably too small to reach the scaling limit.
358: Therefore, we believe that $\rho = 0.920$ is most likely still
359: inside the hexatic phase and has not yet reached the hexatic-solid boundary.
360: This establishes a lower bound for $\rho_m$,
361: one that is much higher than the value previously
362: suggested \cite{95web14636,99jas2594}. But this new
363: lower bound is consistent with the recent estimate provided by Jaster based
364: on simulations with $N$ up to $1024^2$ \cite{04jas120}.
365: Since the pressure at $\rho$ = 0.920 (see Fig.~1) is much higher than
366: the pressure inside the apparent van der Waals loop,
367: this new lower bound for $\rho_m$ also
368: provides evidence corroborating the conclusion we have
369: drawn from the size-dependence of the pressure-density data in Fig.~1
370: that the apparent van der Waals loop in the pressure is not related to a
371: first-order transition.
372:
373: \begin{figure}
374: \includegraphics[width=1.0\columnwidth]{fig4}
375: \caption[]{
376: Subblock scaling analysis for the translational order parameter for
377: the $N = 512^2$ simulations in a square box and a rectangular box.
378: The dotted line corresponds to a slope of $-2$ and the dashed line
379: a slope of $-1/3$.
380: }
381: \end{figure}
382:
383: In conclusion, we have shown using
384: large-scale computer simulations with more than a
385: million particles that the apparent van der Waals loop observed
386: previously in smaller simulations is an artifact of finite-size effects.
387: In conjunction with a detailed scaling analysis, the data provide
388: compelling evidence for a continuous isotropic-hexatic transition
389: as predicted by HNY theory at $\rho_i \lsim 0.900$.
390: Scaling analysis of the translational order also yields a
391: lower bound for the melting density, $\rho_m > 0.920$, one that is much higher
392: than previously thought, providing additional evidence that the apparent
393: van der Waals loop is not due to a first-order transition.
394:
395:
396: \begin{acknowledgments}
397:
398: This work was supported by the National Science Foundation under grant
399: CHE-9970766. The author has benefited from helpful discussions with
400: Hans C. Andersen.
401:
402: \end{acknowledgments}
403:
404:
405: \begin{thebibliography}{99}
406:
407:
408:
409:
410: \bibitem{66mer1133}
411: N.D. Mermin and H. Wagner,
412: {\em Phys. Rev. Lett.}
413: {\bf 17}, 1133 (1966).
414:
415:
416:
417: \bibitem{78hal121}
418: B.I. Halperin and D.R. Nelson,
419: {\em Phys. Rev. Lett.}
420: {\bf 41}, 121 (1978).
421:
422:
423:
424: \bibitem{79nel2457}
425: D.R. Nelson and B.I. Halperin,
426: {\em Phys. Rev. B}
427: {\bf 19}, 2457 (1979).
428:
429:
430:
431: \bibitem{79you1855}
432: A.P. Young,
433: {\em Phys. Rev. B}
434: {\bf 19}, 1855 (1979).
435:
436:
437:
438: \bibitem{02bin2323}
439: K. Binder, S. Sengupta and P. Nielaba,
440: {\em J. Phys. Condens. Matt.}
441: {\bf 14}, 2323 (2002).
442:
443:
444:
445: \bibitem{62ald359}
446: B.J. Alder and T.E. Wainwright,
447: {\em Phys. Rev.}
448: {\bf 127}, 359 (1962).
449:
450:
451:
452: \bibitem{68hoo3609}
453: W.G. Hoover and F.H. Ree,
454: {\em J. Chem. Phys.}
455: {\bf 49}, 3609 (1968).
456:
457:
458:
459: \bibitem{89zol9518}
460: J.A. Zollweg, G.V. Chester and P.W. Leung,
461: {\em Phys. Rev. B}
462: {\bf 39}, 9518 (1989).
463:
464:
465:
466: \bibitem{92zol11187}
467: J.A. Zollweg and G.V. Chester,
468: {\em Phys. Rev. B}
469: {\bf 46}, 11187 (1992).
470:
471:
472:
473: \bibitem{92lee11190}
474: J. Lee and K.J. Strandburg,
475: {\em Phys. Rev. B}
476: {\bf 46}, 11190 (1992).
477:
478:
479:
480: \bibitem{95web14636}
481: H. Weber, D. Marx and K. Binder,
482: {\em Phys. Rev. B}
483: {\bf 51}, 14636 (1995).
484:
485:
486:
487: \bibitem{97mit6855}
488: A.C. Mitus, H. Weber and D. Marx,
489: {\em Phys. Rev. E}
490: {\bf 55}, 6855 (1997).
491:
492:
493:
494: \bibitem{95fer3477}
495: J.F. Fernandez, J.J. Alonso and J. Stankiewicz,
496: {\em Phys. Rev. Lett.}
497: {\bf 75}, 3477 (1995).
498:
499:
500:
501: \bibitem{97fer750}
502: J.F. Fernandez, J.J. Alonso and J. Stankiewicz,
503: {\em Phys. Rev. E}
504: {\bf 55}, 750 (1997).
505:
506:
507:
508: \bibitem{98jas277}
509: A. Jaster,
510: {\em Europhys. Lett.}
511: {\bf 42}, 277 (1998).
512:
513:
514:
515: \bibitem{99jas134}
516: A. Jaster,
517: {\em Physica A}
518: {\bf 264}, 134 (1999).
519:
520:
521:
522: \bibitem{99jas2594}
523: A. Jaster,
524: {\em Phys. Rev. E}
525: {\bf 59}, 2594 (1999).
526:
527:
528:
529: \bibitem{00bat5223}
530: M.A. Bates and D. Frenkel,
531: {\em Phys. Rev. E}
532: {\bf 61}, 5223 (2000).
533:
534:
535:
536: \bibitem{00sen6294}
537: S. Sengupta, P. Nielabe and K. Binder,
538: {\em Phys. Rev. E}
539: {\bf 61}, 6294 (2000).
540:
541:
542:
543: \bibitem{02mit184202}
544: A.C. Mitus, A.Z. Patashinski, A. Patrykiejew and S. Sokolwski,
545: {\em Phys. Rev. B}
546: {\bf 66}, 184202 (2002).
547:
548:
549:
550: \bibitem{02wat041110}
551: H. Watanabe, S. Yukawa, Y. Ozeki and N. Ito,
552: {\em Phys. Rev. E}
553: {\bf 66}, 041110 (2002).
554:
555:
556:
557: \bibitem{03woj939}
558: K.W. Wojciechowski, K.V. Tretiakov, A.C. Branka and M. Kowalik,
559: {\em J. Chem. Phys.}
560: {\bf 119}, 939 (2003).
561:
562:
563:
564: \bibitem{04nie4115}
565: Binder K, Chaudhuri D, Franzrahe K, Henseler P, Lohrer M, Ricci A, Sengupta S, Strepp W Nielaba P,
566: {\em J. Phys. Cond. Matt.}
567: {\bf 16}, S4115 (2004).
568:
569:
570:
571: \bibitem{04jas120}
572: A. Jaster,
573: {\em Phys. Lett. A}
574: {\bf 330}, 120 (2004).
575:
576:
577:
578: \bibitem{04wat045103}
579: H. Watanabe, S. Yukawa, Y. Ozeki and N. Ito,
580: {\em Phys. Rev. E}
581: {\bf 69}, 045103 (2004).
582:
583:
584:
585: \bibitem{96bag255}
586: K. Bagchi, H.C. Andersen and W. Swope,
587: {\em Phys. Rev. Lett.}
588: {\bf 76}, 255 (1996).
589:
590:
591:
592: \bibitem{95dre597}
593: C. Dress and W. Krauth,
594: {\em J. Phys. A}
595: {\bf 28}, L597 (1995).
596:
597:
598:
599: \bibitem{04liu035504}
600: J.W. Liu and E. Luiijten,
601: {\em Phys. Rev. Lett.}
602: {\bf 92}, 035504 (2004).
603:
604:
605:
606:
607:
608: \end{thebibliography}
609:
610:
611:
612:
613:
614: \end{document}
615:
616:
617:
618: