1: \documentclass[12pt]{iopart}
2:
3: \usepackage{graphicx}
4: \usepackage{amssymb}
5:
6: \begin{document}
7:
8: \letter{Critical Binder cumulant in two--dimensional anisotropic Ising
9: models}
10: \author{W Selke$^1$ and L N Shchur$^2$}
11: \address{$^1$ Institut f\"ur Theoretische Physik, RWTH Aachen, 52056 Aachen, Germany}
12: \address{$^2$ Landau Institute for Theoretical Physics, 142432 Chernogolovka, Russia}
13: \ead{selke@physik.rwth-aachen.de}
14: \submitto{JPA}
15:
16: \begin{abstract}
17:
18: The Binder cumulant at the phase transition of Ising models on
19: square lattices with various ferromagnetic nearest and
20: next--nearest neighbour couplings is determined using mainly
21: Monte Carlo techniques. We discuss the possibility to relate the
22: value of the critical cumulant in the isotropic, nearest neighbour
23: and in the anisotropic cases to each other by means of a scale
24: transformation in rectangular geometry, to pinpoint universal and
25: nonuniversal features.
26:
27: \end{abstract}
28:
29: \pacs{05.50.+q, 75.40.Cx, 05.10.Ln}
30:
31: \section{Introduction}
32:
33: In the field of phase transitions and critical phenomena,
34: the fourth order cumulant of the order parameter \cite{Binder}, the
35: Binder cumulant $U$, plays an important role. Among others, the
36: cumulant may be used to compute the critical exponent of the
37: correlation length, and thence to identify the universality class
38: of the transition.
39:
40: The value of the Binder cumulant at the transition
41: temperature, $U(T_c)$, the critical Binder cumulant, has received much
42: attention, too, indicating the universality
43: class as well \cite {Privman}. For instance, in the case of the
44: spin-1/2 Ising model with
45: ferromagnetic nearest--neighbour couplings the critical cumulant
46: has been determined very accurately in numerical work, applying
47: Monte Carlo techniques \cite{Bruce} and transfer-matrix methods
48: augmented by finite--size extrapolations to the
49: thermodynamic limit \cite{Bloete}. The resulting
50: value, $U(T_c)= 0.61069...$\cite{Bloete}, has been observed to hold
51: also for various other, Ising--type models on a square
52: lattice \cite{Bruce,Bloete}.
53:
54: Quite recently, it has been demonstrated by Chen and Dohm \cite{CD} that
55: the critical Binder cumulant may display non--universal features, when
56: the couplings on the square lattice are anisotropic. More
57: specifically, they have studied critical effects of the anisotropy
58: matrix of the Landau--Ginzburg--Wilson Hamilitonian, consisting
59: of the coefficients
60: in front of the leading, second--order gradient term of the
61: Hamiltonian. A nondiagonal matrix is shown to imply that the
62: isotropic case cannot be restored by a rescaling of lengths, i. e. in
63: choosing suitable aspect ratios on a rectangular geometry. Such a
64: scale transformation is possible in the case of a diagonal anisotropy
65: matrix, for instance, for nearest--neighbour Ising models with
66: two different couplings along the axes of the
67: square lattice \cite{Bloete,CD,BW}.
68:
69: In this Letter, we shall present results of rather extensive Monte Carlo
70: simulations for a variety of anisotropic Ising models on a square
71: lattice, corresponding to diagonal and nondiagonal anisotropy
72: matrices. Our data agree qualitatively with the recent results
73: by Chen and Dohm \cite{CD} on nonuniversality in the case
74: of a nondiagonal matrix. They confirm quantitatively the scale
75: transformation given by Kamieniarz and Bl\"ote \cite {Bloete} in
76: the case of a diagonal matrix.
77:
78:
79: \section{Results}
80:
81: We consider spin-1/2 Ising models on a square lattice with
82: anisotropic nearest and next--nearest neighbour interactions. The
83: Hamiltonian reads
84:
85: \begin{equation}
86: {\cal H} = -\sum\limits_{x,y} S_{x,y} ( J_h S_{x+1,y}
87: +J_v S_{x,y+1} + J_{d1} S_{x+1,y+1} + J_{d2} S_{x+1,y-1)})
88: \end{equation}
89:
90: \noindent
91: where $S_{x,y}= \pm 1$ is the spin at site $(x,y)$, see Fig. 1. In
92: most cases, we take
93: all couplings to be positive, i. e. ferromagnetic. In the Monte Carlo
94: simulations, systems with $K L$ spins, subject to full periodic
95: boundary conditions, are analysed, where $K (L)$ corresponds
96: to the $x (y)$--direction. The aspect ratio, $r$, is defined by
97: $r= K/L$.
98:
99:
100: \begin{figure}
101: \begin{center}
102: \includegraphics[width=0.5\linewidth]{fig1}
103: \end{center}
104: \label{fig1}
105: \caption{Sketch of the interactions in the anisotropic Ising model.}
106: \end{figure}
107:
108:
109:
110: The critical Binder cumulant is defined by \cite{Binder}
111:
112: \begin{equation}
113: U(T_c) = 1- <M^4>/(3 <M^2>^2)
114: \end{equation}
115:
116: \noindent
117: taking the thermodynamic limit, $L, K \longrightarrow \infty$;
118: $<M^2>$ and $<M^4>$ denote the second and fourth moments
119: of the magnetization, $M= \sum\limits_{x,y} S_{x,y}/(KL)$.
120:
121: We studied especially two types of anisotropy. In the first
122: case, where both diagonal next--nearest neighbour couplings
123: vanish, $J_{d1}= J_{d2}= 0$, the anisotropy matrix is diagonal \cite{CD}.
124:
125: In the second case, where $J_v= J_h, J_{d1}=J_d$, and
126: $J_{d2}= 0$, one encounters nondiagonality \cite{CD}. Indeed, we
127: consider the two--dimensional
128: variant of a three--dimensional model which had been analysed
129: by Chen and Dohm \cite{CD} and subsequently investigated using Monte
130: Carlo techniques \cite{Drope}. The simulations, however, may
131: have been hampered by the fact that the critical
132: phase transition temperature is not known exactly, in contrast
133: to the present situation.
134:
135: Note that data of high accuracy are needed. We
136: computed systems of sizes, $K, L$, ranging typically from 4 to 64
137: for square lattices, $K= L$. Using the standard Metropolis
138: algorithm (a cluster flip algorithm becomes significantly
139: more efficient for larger system sizes), Monte Carlo runs with
140: up to $5\times 10^8$
141: Monte Carlo steps per site, for the largest systems, were
142: performed, averaging then over several, about ten, of these
143: runs to obtain final estimates and to determine the statistical
144: error bars, shown in the figures. We computed not only the cumulant, but also
145: other quantities like energy and specific heat, to
146: check the accuracy of our data. Of course, sufficiently
147: small lattices may be solved easily by direct enumeration.
148:
149: \subsection{Diagonal anisotropy matrix}
150:
151: \begin{figure}
152: \begin{center}
153: \includegraphics[width=0.55\linewidth]{fig2}
154: \end{center}
155: \label{fig2}
156: \caption{Critical Binder cumulant $U(T_c)$ for Ising models of size
157: $L^2$ with $J_v/J_h$= 4.0, 3.0, and 2.0 (from bottom to top). Open
158: symbols denote Monte Carlo data, full symbols, at $1/L=0$, result
159: from the scale transformation \cite{Bloete} as described in the
160: text. Lines are guides to the eye.}
161: \end{figure}
162:
163: We consider the anisotropic Ising model Eq. (1), with
164: vanishing next--nearest neighbour interactions and
165: different horizontal, $J_h$, and vertical $J_v$, nearest--neighbour
166: interactions, see Fig. 1. Then the anisotropy matrix is
167: diagonal \cite{CD}. The exact transition temperature is given
168: by \cite{Onsager,Berker}
169:
170: \begin{equation}
171: \sinh(2J_h/k_BT_c)\sinh(2J_v/k_BT_c)= 1,
172: \end{equation}
173:
174: \noindent
175: where $k_B$ is the Boltzmann constant. Following Kamieniarz
176: and Bl\"ote \cite{Bloete}, the critical Binder cumulant for
177: the anisotropic model
178: on a square lattice, $K= L$, in the thermodynamic limit, can be obtained
179: from that of the isotropic case, $J_h= J_v$, on a rectangular
180: lattice with the aspect ratio $r= K/L$ using a 'scale
181: transformation':
182:
183: \begin{equation}
184: \sinh(2J_h/k_BT_c)= 1/\sinh(2J_v/k_BT_c)=r.
185: \end{equation}
186:
187: The critical cumulant $U(T_c)$ of the isotropic case with arbitrary aspect
188: ratio $r$ has been calculated before, extrapolating to the thermodynamic
189: limit, with a high degree
190: of accuracy resulting in a polynomial representation of $U(T_c)$ in
191: $r$ \cite{Bloete}. Note that this critical cumulant decreases
192: monotonically from its maximal value at $r=1$ to zero, when the aspect
193: ratio is increased to infinity \cite{Bloete} (or lowered to zero, because of
194: the obvious symmetry when replacing $r$ by $1/r$), i. e. when
195: approaching the one--dimensional limit of the model.
196:
197: In Fig. 2 we display results of our check on that scale
198: transformation, showing our Monte Carlo results of $U(T_c)$ for
199: anisotropic Ising systems, $J_h \ne J_v$, of small and moderate
200: sizes, $K=L$. We also include the values we obtained by evaluating
201: the previously reported polynomial results for the isotropic case with the
202: appropriate, see Eq. (4), aspect ratio $r$ \cite{Bloete}. We find
203: a very good, quantitative agreement between our Monte Carlo data
204: and the values based on the scale transformation and
205: the polynomial representation \cite{Bloete}.
206:
207: This observation confirms the universality of the critical
208: Binder cumulant in the case of a diagonal anisotropy matrix.
209:
210: \subsection{Nondiagonal anisotropy matrix}
211:
212: \begin{figure}
213: \begin{center}
214: \includegraphics[width=0.55\linewidth]{fig3}
215: \end{center}
216: \label{fig3}
217: \caption{Critical cumulant of the anisotropic triangular Ising model for
218: sizes $L^2$, with $J_d/J$= 0.0 (squares), 1.0 (triangles
219: up), 1.5 (triangles left), and 2.5 (triangles down). Open symbols denote
220: our Monte Carlo (as well as numerically exact data for small
221: lattices), full symbols, at $1/L=0$, denote previous
222: estimates following from finite--size extrapolations \cite{Bloete}.}
223: \end{figure}
224:
225: Now, we consider the case where $J_v= J_h= J$, and $J_{d1}= J_d$,
226: while $J_{d2}= 0$, see Fig. 1. The model has been called, albeit
227: being defined on a square lattice, the 'anisotropic triangular
228: model' \cite{Berker}, because of an obvious isomorphy. Its
229: anisotropy matrix is nondiagonal \cite{CD}. Again,
230: the exact transition temperature is known \cite{Berker,Houtappel}
231:
232: \begin{equation}
233: (\sinh(2J/k_BT_c))^2 +2\sinh(2J/k_BT_c)\sinh(2J_d/k_BT_c) = 1,
234: \end{equation}
235:
236: \noindent
237: where $J$ and $J_d$ are supposed to be ferromagnetic in the
238: following, unless stated otherwise (a weakly
239: antiferromagnetic $J_d$ does not destroy the ferromagnetic
240: order at low temperatures \cite{Berker}). Some of our
241: simulational results for the Binder cumulant at $T_c$ for square
242: lattices, $L=K$, with ferromagnetic couplings are
243: shown in Fig. 3, together with highly accurate
244: estimates based on finite--size extrapolations to the
245: thermodynamic limit \cite{Bloete} in
246: the cases $J_d=0$, being the isotropic nearest neighbour square Ising
247: model, and $J_d=J$, being the isotropic nearest neighbour
248: triangular Ising model. The Monte Carlo data
249: seem to allow a smooth extrapolation to the thermodynamic limit, providing
250: reliable and accurate estimates (we refrained from a
251: quantitative finite--size--analysis, because, in general, the form of the
252: corrections to scaling seems to be unknown).
253:
254: Results for extrapolations to the thermodynamic limit
255: are depicted in Fig. 4. Most
256: importantly, the critical Binder cumulant $U(T_c)$ is seen to increase
257: from its value in the isotropic, $J_d=0$, case when introducing
258: the diagonal interaction $J_d > 0$. $U(T_c)$ seems to reach a
259: maximum at about $J_d/J$= 1.0, decreasing then, crossing the
260: isotropic value at about $J_d/J \approx 1.74$ (possibly
261: at $\sqrt 3$, as one may speculate), and finally
262: approaching the one--dimensional limit, where $U(T_c)$
263: tends to go to zero, when $J$ becomes indefinitely weak
264: compared to $J_d$. As has been just emphasized, $U(T_c)$ is initially, for
265: $0 < J_d/J \lesssim 1.74$, larger than in the isotropic limit. Now, varying
266: the aspect ratio and keeping the rectangular geometry
267: in the isotropic Ising model, this can only lead to
268: a decrease in the critical Binder
269: cumulant \cite{Bloete}. Therefore, it
270: is not possible to obtain the critical Binder cumulant in
271: the anisotropic case, $K=L$ and $0 < J_d/J \lesssim 1.74$, by a
272: scale transformation from the isotropic case, in contrast
273: to the situation discussed in the
274: previous subsection. Obviously, this finding is
275: in accordance with the analysis by Chen and
276: Dohm \cite{CD}. Our findings do not rule out the
277: possibility that special cases of the triangular and square
278: models, choosing suitable
279: aspect ratios and anisotropies, may be mapped onto each
280: other, having the same critical cumulants \cite{Bloete}.
281:
282: Note that the critical cumulant seems to decrease monotonically
283: when taking a weak antiferromagnetic coupling $J_d$, as we find in
284: preliminary simulations.
285:
286: \begin{figure}
287: \begin{center}
288: \includegraphics[width=0.6\linewidth]{fig4}
289: \end{center}
290: \label{fig4}
291: \caption{Critical Binder cumulant, extrapolated to the
292: thermodynamic limit, see Fig. 3, for the anisotropic triangular
293: Ising model as function of the anisotropy $J_d/J$. Full symbols
294: denote estimates taken from Ref.\cite{Bloete}. The solid line corresponds
295: to the isotropic case on a square lattice, $J_d$= 0.}
296: \end{figure}
297:
298: Perhaps quite interestingly, the critical Binder cumulant displays
299: a non--monotonic behaviour when varying the aspect ratio for
300: the anisotropic triangular model, when $J_d/J$ is sufficiently
301: large. Starting with the square lattice, $r$=1, and
302: then, say, enlargening (or, equivalently due to symmetry, lowering) the
303: aspect ratio, $U(T_c)$ first becomes larger, before finally
304: decreasing again as one approaches the one--dimensional geometry.
305:
306: In summary, the Monte Carlo simulations show that for a diagonal
307: anisotropy matrix, the critical Binder cumulant in
308: the anisotropic and isotropic cases are connected by a scale
309: transformation keeping rectangular geometry. In marked
310: contrast, such a scale transformation does not exist, in
311: general, in the anisotropic triangular model described by a nondiagonal
312: anisotropy matrix, for reasons which have been discussed
313: recently by Chen and
314: Dohm, demonstrating non--universal features in
315: the critical Binder cumulant \cite{CD}. Thence, care
316: is needed in identifying universality
317: classes and in estimating phase transition temperatures in using
318: the critical Binder cumulant.
319:
320: \ack
321: We thank especially V. Dohm for informing us about his work with
322: X. S. Chen, which, in turn, motivated our work, as well as
323: for very illuminating discussions. We also thank D. Stauffer for
324: useful information on the three--dimensional variant of the model
325: considered in this Letter.
326:
327: \section*{References}
328: \begin{thebibliography}{1}
329: \bibitem{Binder} Binder K 1981 {\it Z. Physik} B {\bf 43} 119
330: \bibitem{Privman} Privman V, Aharony A, and Hohenberg P C 1991 in {\it Phase
331: Transitions and Critical Phenomena, vol 10}, ed by C Domb and J L Lebowitz
332: (New York: Academic)
333: \bibitem{Bruce} Nicolaides D and Bruce A D 1988 {\it J. Phys. A:
334: Math. Gen.} {\bf 21} 233
335: \bibitem{Bloete} Kamieniarz G and Bl\"ote H W J 1993 {\it J. Phys. A:
336: Math. Gen.} {\bf 26} 201
337: \bibitem{CD} Chen X S and Dohm V 2004 {\it Phys. Rev.} E {\bf 70}
338: 056136; 2005 {\it Phys. Rev.} E {\bf 71} 059901(E)
339: \bibitem{BW} Binder K and Wang J-S 1989 {\it J. Stat. Phys.} {\bf 55} 87
340: \bibitem{Drope} Schulte M and Drope C 2005 {\it Intern. J. Mod. Phys.} C in print
341: \bibitem{Onsager} Onsager L 1944 {\it Phys. Rev.} {\bf 65} 117
342: \bibitem{Berker} Berker A N and Hui K 1993 {\it Phys. Rev.} B {\bf 48} 12393
343: \bibitem{Houtappel} Houtappel R M F 1950 {\it Physica} {\bf 16} 425
344: \end{thebibliography}
345:
346: \end{document}