1: \documentclass[aps,twocolumn,showpacs]{revtex4}
2: % \documentclass[aps,preprint,superscriptaddress,showpacs,tightenlines]{revtex4}
3: % \documentclass[aps,preprint,prl,superscriptaddress,showpacs]{revtex4}
4:
5: \usepackage{graphicx}
6: \usepackage{latexsym}
7: \frenchspacing
8:
9: \begin{document}
10:
11: \title{Nature of phase transition(s) in striped phase of \\
12: triangular-lattice Ising antiferromagnet
13: }\author{S. E. Korshunov} \affiliation{L. D. Landau Institute for
14: Theoretical Physics, Kosygina 2, Moscow 119334, Russia}
15: \date{\today}
16:
17: \begin{abstract}
18:
19: Different scenarios of the fluctuation-induced disordering of the striped
20: phase which is formed at low temperatures in the triangular-lattice
21: Ising model with the antiferromagnetic interaction
22: of nearest and next-to-nearest neighbors are analyzed and compared.
23: The dominant mechanism of the disordering is related to the formation of
24: a network of domain walls, which is characterized by an extensive number
25: of zero modes and has to appear via the first-order phase transition.
26: % The temperature remains
27: % finite even when the interaction of nearest neighbors tends to infinity.
28: In principle, this first-order transition can be preceded by a continuous
29: one, related to the spontaneous formation of double domain walls
30: and a partial restoration of the broken symmetry, but the realization of such
31: a scenario requires the fulfillment of rather special relations between
32: the coupling constants.
33:
34:
35: \end{abstract}
36:
37: \pacs{05.50.+q, 64.60.Cn, 75.10.Hk, 75.50Ee}
38:
39: \maketitle
40:
41: \section{Introduction}
42:
43: An Ising model can be defined by the Hamiltonian
44: \begin{equation} \label{HIs}
45: H=\sum_{({\bf i},{\bf j})}J_{\bf ij}\sigma_{\bf i}\sigma_{\bf j}\,,
46: \end{equation}
47: where the fluctuating variables (spins), $\sigma_{\bf j}=\pm 1$,
48: are defined on the sites ${\bf j}$ of some regular lattice.
49: In the standard version of the model \cite{Ons,Hout,Wan} the coupling
50: constants, $J_{\bf ij}$, are assumed to be non-zero only when ${\bf i}$ and
51: ${\bf j}$ are the nearest neighbors of each other, but in a more general
52: case one can suppose that they depend on the distance between ${\bf i}$
53: and ${\bf j}$.
54: The models belonging to this class play an extremely important role in the
55: condensed matter physics, because they can be used for the description
56: of a huge variety of systems with a two-fold degeneracy of an order parameter.
57: The best known examples of such systems are ferromagnets and
58: antiferromagnets with strong easy axis anisotropy and absorbed monolayers.
59:
60: The exact solution of the Ising model on a triangular lattice
61: with the interaction of only nearest neighbors was found
62: % by Onsager in 1944 \cite{Ons}. This was the first example of an exactly
63: % solvable model in more than one dimension.
64: % Later a much more simple solution of the same model has been
65: % proposed by Vdovichenko \cite{Vdv}.
66: %
67: %The exact solution of the analogous model on triangular lattice followed
68: in 1950 \cite{Hout,Wan}. In the case of the isotropic antiferromagnetic
69: interaction it demonstrates rather unusual properties.
70: Namely, the system remains disordered at arbitrarily low temperature
71: \cite{Hout,Wan} and at zero temperature is characterized by an algebraic
72: decay of the correlation functions \cite{St64,St70} and a finite residual
73: entropy per site \cite{Wan}.
74: The ground states of this model can be mapped \cite{BH} onto the states of
75: a solid-on-solid (SOS) model describing the fluctuations of the (111) facet
76: of a crystal with a simple cubic lattice and are infinitely
77: degenerate.
78: % The behavior of the correlation functions of the antiferromagnetic Ising
79: % model on triangular lattice \cite{St} allows one to conclude that this SOS
80: % model is in the rough phase \cite{BH} and, therefore, height fluctuations
81: % can be described by a continuous Gaussian Hamiltonian.
82:
83: This degeneracy is not related to symmetry
84: and therefore in a physical situation its removal by the interactions
85: of more distant neighbors should be taken into account.
86: % If in addition to the antiferromagnetic interaction of nearest neighbors
87: % (characterized by a coupling constant $J_1>0$)
88: If the interaction of second neighbors (characterized by the coupling
89: constant $J_2$) is included into consideration, for both signs of $J_2$
90: the degeneracy of the ground states is reduced to a sixfold one \cite{Met}.
91: In terms of the SOS representation \cite{BH}, the ferromagnetic interaction
92: of second neighbors ($J_2<0$) corresponds to the positive energy of a step
93: and, therefore, leads to the stabilization of the flat phase at low enough
94: temperatures \cite{NHB}.
95: With the increase of temperature a roughening transition \cite{Weeks}
96: takes place, which at a higher temperature is followed by another phase
97: transition related to the dissociation of pairs of dislocations
98: \cite{NHB,Land}.
99: Both transitions belong to the Berezinskii-Kosterlitz-Thouless % \cite{BKT}
100: universality class.
101:
102: This scenario follows also from approximate mappings \cite{AP,FSK}
103: of the considered model onto a six-state clock model and has been confirmed
104: by numerous Monte-Carlo simulations \cite{Land,FSK,MC}.
105: As is typical for low-dimensional systems, a mean field analysis \cite{Mek}
106: leads to the wrong conclusions about the character of the intermediate phase
107: % (which is critical)
108: or the nature of phase transitions.
109:
110:
111: \begin{figure}[b]
112: \vspace{0mm}
113: \begin{center}
114: \includegraphics[width=30mm]{iaf-fig1.eps}
115: \end{center}
116: \caption{\label{1a}
117: The structure of the ground state for $J_{1,2}>0$.
118: }\end{figure}
119:
120: The present work is devoted to the antiferromagnetic Ising model on triangular
121: lattice in which % not only the interaction of nearest neighbors,
122: the interaction of second neighbors is also antiferromagnetic.
123: For brevity we shall call such a system
124: a triangular-lattice Ising antiferromagnet.
125: The structure of the ground state of the triangular-lattice Ising
126: antiferromagnet with the interaction of first and second neighbors
127: \cite{Met} is shown in Fig. \ref{1a}.
128: Here and below we use filled and empty circles
129: to denote the spins of opposite signs.
130:
131: Although this version of the Ising model have been also investigated by
132: different methods \cite{DSWG,KTK,SH,EH},
133: % and numerical \cite{GP,TM,NKL} approaches,
134: its properties are not as clearly understood
135: as those of the model with the ferromagnetic interaction of second neighbors.
136: In particular, Domany {\em et al.} \cite{DSWG} have demonstrated
137: that the formal construction of the Ginzburg-Landau functional describing
138: the formation of the state % whose structure is
139: shown in Fig. \ref{1a} reproduces the Hamiltonian of the Heisenberg
140: model with the cubic anisotropy, which under the renormalization is
141: transformed \cite{DR} into the discrete cubic model \cite{KLF}.
142: However, this hardly allows one to make any conclusions about
143: the properties of a triangular-lattice Ising antiferromagnet (i) because
144: the cubic model allows for {\em three} different scenarios of disordering
145: \cite{NRS}
146: (although sometimes only one of them is mentioned \cite{Sch}),
147: and (ii) because this approach does not take into
148: account the strong chiral asymmetry \cite{OH}
149: existing in the problem (see Sec. \ref{gs}).
150: Additionally, such a description cannot reproduce the well-known properties
151: of the model with only first-neighbor interaction. The same can be also said
152: about the mean-field approach of Kaburagi {\em et al.} \cite{KTK}, which
153: gives an unrealistic prediction of a three-sublattice intermediate phase.
154:
155: A more direct analysis of the fluctuations which can be responsible for
156: the disordering of the striped phase has been undertaken
157: by Hemmer {\em et al.} \cite{SH,EH}.
158: However, these authors have considered
159: the formation of only one type of domain walls, whose appearance
160: (as is discussed in Sec. \ref{ddw}) can lead
161: only to a {\em partial} restoration of the broken symmetry.
162: Thus, all conclusions of Ref. \onlinecite{SH} and Ref. \onlinecite{EH} are
163: based on the essentially incomplete physical picture and have to be reconsidered.
164:
165:
166: This article is devoted to comparison of different mechanisms and different
167: scenarios of disordering which are possible in the striped phase
168: of a triangular-lattice Ising antiferromagnet.
169: Although a number of numerical works \cite{GP,NKL,RRT} give evidence on
170: the existence of a single first-order transition
171: (for this or that choice of the relation between the coupling constants),
172: in a situation when different mechanisms of disordering compete with each
173: other one hardly can be sure about the universality of this result.
174:
175: Our analysis assumes that $J_1$ is much larger then all other
176: coupling constants, which allows us to use the analytical methods based on
177: the separation of different energy scales.
178: This limit is also of interest because it corresponds to the case of
179: strong chiral asymmetry and is the most relevant one
180: for many physical realizations of the model.
181: The outlook of the article is as follows.
182:
183: In Sec. \ref{gs} we discuss the symmetry of the ground states and
184: the structure of domain walls and estimate the temperature at which
185: the free energy of a single domain wall vanishes as a result
186: of thermal fluctuations of this wall.
187: In Sec. \ref{ddw} we show that the free energy of a double domain wall can
188: be expected to vanish at much lower temperature than that of a single wall,
189: and argue that this suggests a possibility of a two-transition scenario.
190:
191: In Sec. \ref{dwnw} the spontaneous formation of a network of single domain
192: walls is analyzed. In the limit when this network has to be diluted, it
193: becomes clear that it has to appear via a first-order phase transition.
194: When only the interaction of up to third % (and more close)
195: neighbors is taken into account,
196: the estimates for the temperatures of this transition and of
197: the spontaneous formation of double domain walls coincide with each other,
198: which suggests that there is only one phase transition in the system.
199: In Sec. \ref{htph} a phenomenological free-energy functional
200: is constructed which allows one to confirm that the transition
201: to the disordered phase has always to be of the first order.
202:
203: The interplay between the two main mechanisms of the disordering is analyzed
204: in Sec. \ref{md}, whereas the results are summarized in Sec. \ref{conc}.
205: The short Appendix is devoted to a formal derivation of an exact upper
206: boundary for the temperature at which the long-range order in
207: $\sigma_{\bf j}$ is destroyed by thermal fluctuations when $J_1=\infty$.
208:
209: \section{Ground states and domain walls\label{gs}}
210:
211: \begin{figure}[b]
212: \vspace{0mm}
213: \begin{center}
214: \includegraphics[width=35mm]{iaf-fig2.eps}
215: \end{center}
216: \caption{\label{1b}
217: The classification of neighbors on triangular lattice.
218: }\end{figure}
219:
220: The structure of the ground state of the triangular-lattice Ising model
221: with the antiferromagnetic interaction of nearest and second neighbors
222: \cite{Met} is shown in Fig. \ref{1a},
223: whereas Fig. \ref{1b} illustrates the classification of neighbors
224: on a triangular lattice according to their distance from a given cite
225: (denoted by zero).
226: In the following we assume that $J_3$, the coupling constant describing
227: the interaction of third neighbors, can also be non-zero, but
228: satisfies the constraint \cite{TM,KK}, $J_3<J_2/2$,
229: which is required for the stability of the striped ground state of
230: Fig. \ref{1a}. The role of the interactions of more distant neighbors
231: will be discussed in Sec. \ref{md}.
232: To make a situation more transparent we assume that the interaction
233: of nearest neighbors, $J_1$, is much larger then all other coupling constants.
234:
235: The sixfold degeneracy of the state shown in Fig. \ref{1a}
236: corresponds to the violation of the $Z_2\times Z_3$ symmetry,
237: where $Z_2$ is related to the possibility of interchanging positive and
238: negative spins, and $Z_3$ to three possible orientations of the stripes
239: formed by spins of the same sign.
240: % This form of degeneracy suggests that
241: Each of the six ground states
242: can be associated with a unit vector pointing either in positive or
243: negative direction along one of the three axes, which are perpendicular to
244: each other. These six directions can also be put into correspondence with
245: the six faces of a cube.
246: In such a representation the group $Z_2$ is related to the reflection
247: symmetry which transforms the opposite faces of a cube into each other,
248: whereas the group $Z_3$ corresponds to the cyclic permutations of the three axes.
249:
250: In systems with a discrete degeneracy the destruction of a long-range
251: order has to be driven by thermal activation of infinite domain walls.
252: The appearance of a sequence of more or less parallel walls is expected when
253: % the free energy of a single domain wall vanishes as the result of its
254: % thermal fluctuations, but
255: the intersections of walls with
256: different orientations are energetically unfavored \cite{BMVW}.
257: The alternative option consists on the appearance of a network of
258: intersecting domain walls with different orientations, which is would be
259: favored by a negative energy of domain wall intersections \cite{BMVW}.
260:
261: \begin{figure}[b]
262: \vspace{0mm}
263: \begin{center}
264: \includegraphics[width=70mm]{iaf-fig3.eps}
265: \end{center}
266: \caption{\label{2} Low energy domain walls: (a) a straight one;
267: (b) with a kink.}
268: \end{figure}
269:
270: Fig. \ref{2}(a) shows an example of a lowest-energy domain wall separating
271: two different ground states.
272: Each segment of such a wall connects two neighboring spins of
273: the opposite signs and separates two second neighbors of the same sign.
274: The energy of this wall per unit length,
275: \[
276: E_{\rm dw}^{}=2J_2-4J_3>0\,,
277: \]
278: does not depend on $J_1$, because its presence does not
279: lead to the violation of the constraint
280: \begin{equation} \label{cond}
281: \sigma_{\bf j}\sigma_{\bf j'}+\sigma_{\bf j'}\sigma_{\bf j''}+\sigma_{\bf j''}\sigma_{\bf j}=-1
282: \end{equation}
283: on any triangular plaquette. This condition is satisfied when
284: a plaquette contains spins of both signs.
285:
286: When crossing a domain wall the direction of the stripes formed
287: by the spins of the same sign changes by $60^{\circ}$ and, therefore,
288: the direction of a lowest-energy wall is uniquely determined by the
289: pair of ground states which it separates.
290: In other terms, the energy of a domain wall in the considered model is strongly dependent both
291: on its orientation and on which states it separates.
292: From the analysis of systems with a threefold degeneracy it is known
293: that such a property (the chiral asymmetry \cite{OH}) can lead to the change
294: of the universality class \cite{HF} or even of the order
295: \cite{Vil80a,Vil80b,CFHLB} of a phase transition.
296: This is the reason why the analogy with the cubic model
297: \cite{KLF,NRS} without chiral assymetry (which follows from the
298: Landau-Ginzburg analysis of Domany {\em et al.} \cite{DSWG})
299: is insufficient for understanding the properties of a triangular-lattice
300: Ising antiferromagnet with $J_{2,3}\ll J_1$.
301:
302: Any fluctuations of domain wall
303: are impossible without the violation of the constraint (\ref{cond}), and
304: therefore require the energies proportional to $J_1\gg E_{\rm dw}$.
305: Figure \ref{2}(b) shows an example of the simplest elementary defect (a kink)
306: which can be formed on a straight domain wall. The energy of such a
307: defect,
308: \[
309: E_{\rm k}=2J_1-4J_3\,,
310: \]
311: has to include a contribution proportional
312: to $J_1$, because the formation of a kink requires to have one
313: plaquette at which all three spins are of the same sign,
314: and therefore
315: \[%begin{equation}
316: \sigma_{\bf j}\sigma_{\bf j'}+\sigma_{\bf j'}\sigma_{\bf j''}
317: +\sigma_{\bf j''}\sigma_{\bf j}=3\,.
318: \]%end{equation}
319: In Fig. \ref{2}(b) this plaquette is shown by bold lines.
320:
321: At finite temperatures the free energy of a domain wall (per unit length)
322: can be estimated as the difference between its energy and the entropic
323: term related to the formation of kinks \cite{PD}.
324: For $T\ll E_{\rm k}$ this gives
325: \begin{equation} \label{Fdw}
326: F_{\rm dw}(T)\approx E_{\rm dw}-2T\exp(-E_{\rm k}/T)\,.
327: \end{equation}
328:
329: It is well known that when the intersections of domain walls are unfavored,
330: one can expect the formation of a dilute sequence of parallel
331: (on the average) walls when the free energy
332: of a single wall, $F_{\rm dw}(T)$, becomes equal to zero \cite{BMVW,PT}.
333: For $F_{\rm dw}(T)$ defined by Eq. (\ref{Fdw}) this takes place at
334: \begin{equation} \label{Tdw}
335: T=T_1\approx \frac{E_{\rm k}}{\ln(E_{\rm k}/E_{\rm dw})}
336: %\ll E_{\rm k}\;,
337: \end{equation}
338: where we have taken into account that $E_{\rm dw}\ll E_{\rm k}$ (as a
339: consequence of $J_2\ll J_1$).
340: Eq. (\ref{Tdw}) shows that for \makebox{$J_{2},J_3\ll J_1$} the temperature $T_1$
341: only weakly depends on $J_2$ and $J_3$ and is determined mainly
342: % by $E_{\rm k}$, that is
343: by $J_1$.
344:
345: In the limit $J_1\rightarrow\infty$ the fluctuation-induced vanishing
346: of the free energy of a single domain wall becomes impossible.
347: This manifests itself in the divergence of the expression for $T_1$
348: given by Eq. (\ref{Tdw}) for $E_{\rm k}\rightarrow\infty$.
349: Quite remarkably, even in this limit there still
350: remain possibilities for a fluctuation-induced destruction of the
351: long-range order.
352: They are related to the spontaneous formation of {\em double} domain walls
353: (which is discussed in Sec. \ref{ddw})
354: and of a domain-wall network (discussed in Sec. \ref{dwnw}).
355:
356: \section{Spontaneous formation of double domain walls\label{ddw}}
357:
358: \begin{figure}[b]
359: \vspace{0mm}
360: \begin{center}
361: \includegraphics[width=70mm]{iaf-fig4.eps}
362: \end{center}
363: \caption{\label{3} Double domain walls: (a) straight;
364: (b) with two corners.}
365: \end{figure}
366:
367: A double domain wall consists of two parallel single walls
368: [see Fig. \ref{3}(a)] and separates two ground states with
369: the same direction of spin stripes.
370: The energy of a straight double wall per segment,
371: $E_{\rm ddw}^{}$, is given simply by the energy of two single domain walls
372: from which it consists, $E_{\rm ddw}=2E_{\rm dw}$. The interaction between
373: two parallel single walls does not appear even if one takes into account
374: the interactions of spins with their fourth and fifth
375: % until $J_6$, the coupling of sixth
376: neighbors (see Fig. \ref{1b}). %is taken into account.
377: However, it turns out that the fluctuations of a double wall cost less energy
378: than those of a single wall, as a consequence of which its
379: free energy can easily become smaller than that of a single wall.
380:
381: Fig. \ref{3}(b) demonstrates that the fluctuations of a double domain wall
382: are possible without the violation of the constraint (\ref{cond}).
383: From this figure it is clear that each corner on a double wall
384: requires the appearance of an additional segment of a single domain wall.
385: The energy of such a defect, $E_{\rm c}=2J_2$, does not depend on $J_1$ and $J_3$.
386: At finite temperatures the expression for the free energy of a double wall
387: which takes into account the presence of thermally
388: activated corners is of the form \cite{SW,EH}
389: \begin{equation} \label{Fddw}
390: F_{\rm ddw}^{}(T)
391: =2E_{\rm dw}-T\ln[1+\exp(-E_{\rm c}/T)]\,.
392: \end{equation}
393: The spontaneous appearance of a diluted sequence of such walls can be
394: expected to take place when $F_{\rm ddw}^{}(T)$ becomes equal to zero.
395: The condition $F_{\rm ddw}^{}(T_{\rm })=0$ can be rewritten as
396: \begin{equation} \label{Fddw=0}
397: T_{\rm }=\frac{E_{\rm c}}{-\ln[\exp(2E_{\rm dw}/T_{\rm })-1]^{}}\;.
398: \end{equation}
399: Note that Eq. (\ref{Fddw=0}) and, therefore its solution,
400: $T_2$, do not depend on $J_1$.
401:
402: In the case of \makebox{$J_3=0$} the solution of Eq. (\ref{Fddw=0})
403: gives
404: % shows that the spontaneous formation of double domain
405: % walls can be expected to take place at
406: \begin{equation} \label{Tddw0}
407: T_2= \gamma_2 J_2
408: \end{equation}
409: where \cite{SW,EH}
410: \[
411: \gamma_2=\frac{1}{2} \ln\!\left[\!
412: \left(\frac{1}{2}+\sqrt{\frac{23}{108}}\right)^{1/3}\hspace*{-4mm}+
413: \left(\frac{1}{2}-\sqrt{\frac{23}{108}}\right)^{1/3}\right]
414: \approx 7.112 % \nonumber
415: \]
416: whereas for \makebox{$E_{\rm dw}\ll E_{\rm c}$} an expansion
417: of the exponent in the right-hand side of Eq. (\ref{Fddw=0})
418: allows one to find that
419: \begin{equation} \label{Tddw}
420: T_2
421: \approx \frac{E_{\rm c}}{\ln{(E_{\rm c}/2 E_{\rm dw})}}
422: \approx \frac{2J_2}{\ln{(J_{\rm 2}/2 E_{\rm dw})}}\;.
423: \end{equation}
424:
425: Comparison of Eq.(\ref{Tddw0}) and Eq. (\ref{Tddw}) with Eq. (\ref{Tdw})
426: demonstrates that for $J_2\ll J_1$ the destruction of the long-range order in
427: $\sigma_{\bf j}$ cannot be driven by the spontaneous formation of a sequence of
428: single domain walls, because the analogous sequence of double domain walls
429: can be expected to appear at much lower temperature.
430: A numerical calculation of $T_2$ for an arbitrary relation between $J_1$
431: and $J_2$ (at $J_3=0$) in terms of the one-dimensional SOS model which
432: takes into account also more complex fluctuations of a double domain wall
433: can be found in \makebox{Ref. \onlinecite{SH}.}
434:
435:
436: In the limit $J_1\rightarrow\infty$ at low temperatures %($T<T_2$)
437: the system has to be completely frozen in one of its
438: ground states,
439: % (unless a network of domain walls is formed at
440: % $T=T_{\rm nw}<T_2$, see Sec. \ref{dwnw}),
441: because any finite size fluctuation on the background of
442: the striped ground state requires the violation of the constraint
443: (\ref{cond}) on at least two plaquettes (see Fig. \ref{loops}).
444: Accordingly, the exact expression for the free energy of a double domain
445: wall cannot contain any additional contributions related
446: to the suppression of finite size fluctuations.
447: This has allowed Shi and Wortis \cite{SW} to conjecture that
448: in the limit $J_1\rightarrow\infty$ the condition
449: $F_{\rm ddw}(T)=0$ [with $F_{\rm ddw}$ given by Eq. (\ref{Fddw})] determines
450: the {\em exact} value of the transition temperature.
451: However, this conclusion can be valid only if the spontaneous formation of
452: a diluted sequence of double domain walls is not preceded by the spontaneous
453: formation of a network of single domain walls (see Sec. \ref{dwnw}).
454:
455: It is rather evident that the average direction of a fluctuating double wall
456: containing thermally activated corners will be perpendicular to the direction
457: of spin stripes on both its sides.
458: The spontaneous formation of a sequence of such walls
459: leads to the restoration
460: of the $Z_2$ symmetry between the two states with the same direction
461: of stripes and the reduction of the broken symmetry to $Z_3$.
462:
463: Since the concentration of walls, $\nu(T)$, is restricted by their collisions
464: (which are responsible for the reduction of the entropy of their
465: fluctuations), Shi and Wortis \cite{SW} have concluded that this phase
466: transition is continuous and belongs to the Pokrovsky-Talapov \cite{PT}
467: universality class. Accordingly, in the vicinity of $T_2$
468: one should have $\nu(T)\propto (T-T_2)^{1/2}$.
469: Note that % $\nu(T)$ determines
470: the value of the correlation length describing
471: the decay of the correlation function
472: $\langle \sigma_{\bf i}\sigma_{\bf j}\rangle$ in the direction
473: along the stripes is inversely proportional to $\nu(T)$.
474:
475: On the other hand, Einevoll and Hemmer \cite{EH} have argued that this phase
476: transition cannot be continuous, because the temperature at which
477: $F_{\rm ddw}(T)$ vanishes is different for different directions of a
478: double wall. In our opinion this conclusion is completely unjustified.
479: The dependence of $F_{\rm ddw}$ on the direction of the wall manifests
480: itself in the spontaneous formation of a sequence of walls with the same
481: (on the average) direction, and is a {\em necessary condition} for the
482: applicability of the Pokrovsky-Talapov theory \cite{PT} rather than
483: an obstacle for its validity.
484:
485: \begin{figure}[t]
486: \vspace{3mm}
487: \begin{center}
488: \includegraphics[width=65mm]{iaf-fig5.eps}
489: \end{center}
490: \caption{\label{loops} Closed loops formed by (a) single domain walls,
491: (b) double domain walls.}
492: \end{figure}
493:
494: If one assumes that the value of $J_1$ is finite, but large in comparison
495: with $J_2$, the fluctuations in the low temperature phase become
496: possible. Each point where a pair of single walls is created
497: (or annihilates) cost the energy close to $2J_1$, so at low temperatures
498: there will exist a finite concentration of highly anisotropic loops formed
499: by such walls [see Fig. \ref{loops}(a)]. However, at $T\approx T_2\ll J_1$
500: the average distance between them will be much smaller then their size,
501: as a consequence of which their presence can be neglected.
502: On the other hand, the size of the closed
503: loops formed by double domain walls [see Fig. \ref{loops}(b)]
504: diverges when $T\rightarrow T_2$.
505: From the theory of the commensurate-incommensurate
506: transition it is known \cite{BPT} that this is accompanied by the
507: change of the type of the phase transition
508: from the Pokrovsky-Talapov universality class \cite{PT} to that of
509: the Ising model. However, the behavior will be changed only in a
510: narrow region around $T_2$, which will be exponentially small in $J_1/T_2$.
511:
512: Since the considered phase transition is related to the restoration of
513: $Z_2$ symmetry, the emergence of the Ising critical behavior looks
514: rather natural. The different universality class in the case of $J_1=\infty$
515: can be explained by the extremely anisotropic nature of domain walls in
516: that limit, which prevents the merging of different domains of the same state.
517:
518: If the spontaneous formation of a sequence of double domain walls indeed
519: takes place as a separate phase transition it has to be followed
520: (with a further increase of temperature) by a second phase transition
521: related to the restoration of $Z_3$ symmetry.
522: The completely disordered phase above this transition will look like a
523: mixture of finite domains of all six ground states.
524: Since double walls do not change the orientation of stripes,
525: the second phase transition requires the formation of single domain walls
526: of all possible orientations.
527: Above we have found that the spontaneous formation of double domain walls
528: takes place when the free energy of a single wall is still much larger
529: then temperature, so it looks rather plausible that the two phase transition
530: may be well separated from each other.
531: However, to check if it is really so, the formation of a
532: network of single domain walls has to be studied in more detail.
533:
534:
535:
536: \section{Spontaneous formation of a domain-wall network\label{dwnw}}
537:
538: \begin{figure}[b]
539: \vspace{3mm}
540: \begin{center}
541: \includegraphics[width=65mm]{iaf-fig6.eps}
542: \end{center}
543: \caption{\label{5} Three low-energy domain walls with
544: different orientations can merge or intersect with each other.}
545: \end{figure}
546:
547: Like in the previous section it will be convenient to start
548: by considering the case of $J_1=\infty$.
549: In this limit all single domain walls have to be straight due to
550: the absence of kinks.
551:
552: Fig. \ref{5} shows how such walls can intersect or merge with each other
553: without violating the constraint (\ref{cond}).
554: The energy of these intersections does not depend on $J_1$ or $J_2$.
555: In particular, the energy of the $120^{\circ}$ junction shown
556: in Fig. \ref{5}(a) is simply equal to zero, $E_{\rm a}=0$,
557: whereas for the $60^{\circ}$ junction shown in Fig. \ref{5}(b)
558: it is given by $E_{\rm b}=4J_3$.
559: The intersection shown in Fig. \ref{5}(c) can be considered as an
560: overlap of two $60^{\circ}$ junctions.
561: The energy of this object is equal to $E_{*}=12J_3$.
562:
563: An important feature (already mentioned in Sec. \ref{gs}), which plays
564: a crucial role in determining the structure of a domain-wall network
565: for $J_1=\infty$, is that the direction of each wall is uniquely
566: determined by the states which it separates.
567: A possible structure of a network which is formed by straight walls
568: and satisfies this criterion is schematically shown in Fig. \ref{4}.
569: Here the letters A, B and C are used to denote the domains with three
570: different orientations of stripes.
571: Note that all walls between A and B are parallel to each other.
572: The same is true for all walls between B and C,
573: as well as for all walls between C and A.
574:
575: For the sake of clearness we have not shown in Fig. \ref{4} which of
576: the two versions of A, of B, or of C (related to the change of sign
577: of all spins) is realized in each particular domain.
578: This depends on the exact positions of domain walls.
579:
580: The structure of the network shown in Fig. \ref{4} has been chosen to
581: maximize the entropy for the given total length of the walls.
582: The network of such a kind is characterized by a large number of zero
583: modes which do not change its energy.
584: For example, each domain of the type A can be moved to the left or to the
585: right. This changes the areas of all domains of the types B and C
586: which are adjacent to it, but the total length of domain walls (and,
587: therefore, the total energy of the network) is conserved.
588: Analogously, all domains of the types B and C can be moved in the two other
589: directions. When a domain is moved by one lattice unit in such a way,
590: the signs of all spins inside it are reversed.
591:
592: \begin{figure}[b]
593: %\vspace{4mm}
594: \begin{center}
595: \includegraphics[width=80mm]{iaf-fig7.eps}
596: \end{center}
597: \vspace{3mm}
598: \caption{\label{4} A possible structure of a domain-wall network}
599: \end{figure}
600:
601: A combination of all three types of zero modes allows to change the size
602: of a three-domain complex (a bubble) formed by
603: neighboring domains of three different types without changing its position.
604: An example of such a bubble is shown in Fig. \ref{4} by the bold line.
605: The zero modes of this particular type are called the breathing modes
606: \cite{CFHLB}.
607:
608: The existence of breathing modes has been discovered by Villain
609: \cite{Vil80a} when studying the formation of a honeycomb network
610: in which each domain has the shape of a hexagon.
611: In such a network the size of each hexagon can be changed without
612: changing the total length of domain walls.
613: A honeycomb network is formed in a system with a threefold degeneracy
614: in which a domain wall of a given type
615: (for example, a wall between A and B) can have only
616: three particular orientations out of six that are generally allowed.
617: In these terms, in our problem a wall of each type can have only two
618: orientations out of six, whereas in the three-state Potts model on
619: a triangular lattice all six orientations are allowed for domain walls of
620: any type.
621:
622: The entropy which can be associated with the existence of zero modes can be
623: estimated as $\ln M$ per mode, where $M$ is the typical number of the
624: configurations which can be spanned by a given zero mode.
625: It is clear that in a diluted network $M$ has to be proportional to $L$,
626: the typical distance between the centers of neighboring bubbles
627: (in lattice units). This allows one to estimate the free energy
628: (per unit area) of a network shown in Fig. \ref{4} as
629: \begin{equation} \label{Fnw}
630: F_{\rm nw}(L)\approx \frac{2}{\sqrt{3}}\left[4\frac{E_{\rm dw}}{L}+3\frac{E_{\rm mp}}{L^2}
631: -3T\frac{\ln L}{L^2}+O\left(\frac{1}{L^3}\right)\right]\,,
632: \end{equation}
633: where
634: \[
635: %E_1=E_{\rm dw}\,,~~~
636: E_{\rm mp}= E_{\rm a}+E_{\rm b}=4J_3\,.
637: \]
638: The first term in the right-hand side of Eq. (\ref{Fnw}) describes
639: the energy of domain walls, the second term - the energy of merging points
640: and the third term is related to the entropy of zero modes.
641: The expression (\ref{Fnw}) has the same structure as the one proposed by
642: Villain \cite{Vil80a,Vil80b} for a honeycomb network.
643:
644: The variation of $F_{\rm nw}(L)$ with respect to $L$ reveals that
645: this function has two minima, one of which is situated at $L=\infty$ and
646: corresponds to the absence of any network and another at $L=L_0<\infty$.
647: The free energies of these two minima become equal to each other
648: when %\cite{Vil80a}
649: \begin{equation} \label{L}
650: L_0=\frac{3T}{4E_{\rm dw}}\;,~~~\ln L_0=1+\frac{E_{\rm mp}}{T}\;.
651: \end{equation}
652: This shows that the applicability of Eq. (\ref{Fnw}), which assumes
653: $L\gg 1$, requires to have $E_{\rm dw}\ll T\ll E_{\rm mp}$, that is
654: $E_{\rm dw}\ll J_2$.
655: With the increase of $E_{\rm dw}/J_2$ the value of $L_0$ at the transition
656: is decreased and for $E_{\rm dw}\sim J_2$ it becomes comparable with 1, which
657: means that Eq. (\ref{Fnw}) is no longer applicable.
658:
659: In the limit of $E_{\rm dw}\ll J_2$ (when $E_{\rm mp}\approx 2J_2$)
660: the temperature of the first-order
661: phase transition related to the formation of a domain-wall network, which
662: follows from Eqs. (\ref{L}), can be estimated as
663: \begin{equation} \label{Tnw}
664: T_{\rm nw}\approx \frac{E_{\rm mp}}{\ln(E_{\rm mp}/E_{\rm dw})}
665: \approx\frac{2J_2}{\ln(J_2/E_{\rm dw})}\;,
666: \end{equation}
667: whereas the value of $L_0$ (which determines the correlation radius for the
668: fluctuations of $\sigma_{\bf j}$) at the transition point is given by
669: \begin{equation} \label{Lc}
670: L_{c} % \approx \frac{3}{4}\frac{E_2/E_1}{\ln(E_2/E_1)}
671: \approx \frac{3}{2}\frac{J_2/E_{\rm dw}}{\ln(J_2/E_{\rm dw})}\,.
672: \end{equation}
673:
674: With the decrease of $J_3$ the ratio $E_{\rm mp}/E_{\rm dw}$ is decreased, which
675: leads to the decrease of $L_c$. For $J_3\sim J_2$ (that is $E_{\rm dw}\sim E_{\rm mp}$)
676: the value of $L_c$ following from Eqs. (\ref{L}) becomes comparable
677: with 1, which means that the approach based on the minimization of
678: $F_{\rm nw}(L)$ is no longer applicable. However, since the decrease of
679: $J_3$ makes the first-order nature of the transition more and more
680: pronounced, one can expect that it will remain of the first order even when
681: the formation of a domain-wall network does not
682: allow for a quantitative description.
683: In the next section this conclusion is confirmed with the help
684: % first-order nature of the transition to the disordered phase follows also
685: of a phenomenological analysis which does not take
686: into account any details of a domain-wall network structure
687: and, therefore, is applicable in a wide interval of the values of $J_3$
688: (including $J_3=0$).
689:
690: The finiteness of $J_1$ cannot be expected to be
691: of any importance for the phase transition related with the spontaneous
692: formation of a domain-wall network. It allows for fluctuations of
693: single domain walls, which no longer have to be straight, but, as has
694: been shown by Villain \cite{Vil80b} for a hexagonal network, this
695: does not lead to any qualitative changes.
696:
697: \section{High-temperature phenomenology\label{htph}}
698:
699: Like above, it will be convenient to start the analysis
700: by considering the case of $J_1=\infty$.
701: % and only later discuss the consequences of a finite $J_1$.
702: In this limit the manifold of the allowed states coincides with the
703: manifold of the ground states of the system with only nearest neighbor
704: interaction,
705: which can be put into correspondence with the states of the (111) facet
706: of a crystal with a simple cubic lattice \cite{BH} .
707:
708: In particular, the ground states whose structure is shown in Fig.
709: \ref{1a} map onto the states with the maximal possible slope, corresponding
710: to the transformation of the (111) facet into one of the three facets of
711: the (100) family. The direction of the slope is
712: determined by the direction of stripes in spin representation.
713: % In the framework of the SOS representation it is immediately clear that
714: These states do not allow for the formation of
715: any finite size defects, because this would require locally the further
716: increase of a slope.
717:
718: In the limit of $T\rightarrow\infty$ all terms in the partition function
719: become equal to each other, therefore in this limit all correlation
720: functions in the system with $J_1=\infty$ have exactly the same form as
721: in the model with the interaction of only nearest neighbors at zero
722: temperature.
723: The corresponding phase is characterized by the zero slope and
724: a logarithmical divergence of fluctuations \cite{BH} of the discrete
725: variable $n$ describing the position of a surface. One can expect
726: that the same phase will be also stable at large but finite $T$.
727:
728: In this phase the large-scale fluctuations
729: of $n$ can be described by a continuous free energy functional,
730: \begin{equation} \label{f2}
731: F_{\rm eff}\{n\}=\int d^2{\bf r} f_2\{n\}\,,~~
732: f_2\{n\}=\frac{K}{2}(\nabla n)^2\;,
733: \end{equation}
734: in which the discreteness of $n$ is neglected \cite{BH}.
735: At $T=\infty$ the dimensionless effective rigidity $K$ (which is of
736: entropic origin) is equal to $K_0=\pi/9$.
737: In terms of the SOS representation the energy of a step (per unit length)
738: is equal to $-2J_2$, therefore for $J_2<0$ the decrease of
739: temperature is accompanied by a monotonic growth of $K$ \cite{NHB},
740: which at $K=\pi/2$ leads to the phase
741: transition to the smooth phase.
742:
743: We are now considering the opposite case of $J_2>0$, when the decrease
744: of $T$ from \makebox{$T=\infty$} should be accompanied by the decrease of
745: $K$ from $K=K_0$.
746: Since we know that at lower temperatures the triply degenerate
747: phase with a finite slope has to be formed, the phenomenological
748: functional (\ref{f2}) has to be replaced by a more complex one,
749: \begin{equation} \label{Feff}
750: F_{\rm eff}\{n\} = \int d^2{\bf r}(f_2\{n\}+f_3\{n\} +f_4\{n\})\;,
751: \end{equation}
752: where the second term in the integrand,
753: \begin{equation} \label{f3}
754: f_3\{n\}=-K_3[({\bf e}_1\nabla)n][({\bf e}_2\nabla)n][({\bf e}_3\nabla)n]\,,
755: \end{equation}
756: favors a finite slope in one of the three equivalent directions set by
757: the three unit vectors ${\bf e}_\alpha$ (where $\alpha=1,2,3$) forming
758: the angles of $120^{\,\circ}$ with each other.
759: The last term in the integrand, $f_4\{n\}$, is required to stabilize a finite value
760: of a slope. On general grounds, one can expect that the expansion of
761: $f_4\{n\}$ in powers of $\nabla n$ starts from the fourth-order contribution:
762: \begin{equation} \label{f4}
763: f_4\{n\}=\frac{K_4}{4}(\nabla n)^4+\ldots\;.
764: \end{equation}
765:
766: Note that the free energy functional (\ref{Feff}) has nothing in common
767: with the Ginzburg-Landau functional constructed by Domany {\em et al.}
768: \cite{DSWG}. In particular, in the latter the third-order term can appear
769: only in the absence of the particle-hole symmetry (which in terms of the spin
770: representation corresponds to % changing the sign of all variables
771: $s_{\bf j}\Rightarrow -s_{\bf j}$).
772: In the model that we consider this symmetry is, naturally, always present, but
773: $F_{\rm eff}\{n\}$ % the functional (\ref{Feff})
774: has to contain the third-order term just as a consequence
775: of the symmetry of the problem in terms of the SOS representation.
776:
777: From the form of $F_{\rm eff}\{n\}$ it is clear that a phase
778: transition between the phases with zero and finite slopes cannot occur in
779: a continuous way.
780: The three equivalent auxiliary minima of $F_{\rm eff}\{n\}$ are formed at
781: finite values of $|\nabla n|$, and a first-order phase transition takes place
782: when the decrease of $K$ makes the free energy in these minima lower
783: than in the central minimum at $|\nabla n|=0$.
784:
785: Thus we have obtained an additional confirmation of the conclusion that
786: a phase transition from the disordered phase with a zero slope to a phase
787: with a finite slope has to be of the first order.
788: Naturally, the construction of a phenomenological functional does not allow
789: one to distinguish whether it has to be a direct transition to the completely
790: frozen phase with the maximal possible slope,
791: or a transition to an intermediate phase with a smaller slope (that is,
792: with a finite concentration of spontaneously formed double domain walls),
793: which at lower temperatures will be followed by the second phase transition.
794:
795: At finite $J_1$ the height variable $n$ can no longer be uniquely defined.
796: In terms of $n$ each plaquette at which the condition (\ref{cond}) is
797: violated corresponds to a screw dislocation on going around which $n$
798: changes by $\pm 6$ \cite{NHB}.
799: The core energy of such dislocations is close to $2J_1$, whereas their
800: logarithmic interaction is too weak to keep them bound in pairs \cite{NHB}.
801: In such a situation the effective free energy should be a functional
802: not of a multivalued variable $n$, but of its derivatives,
803: \[
804: m_\alpha=({\bf e}_\alpha\nabla)n\,,
805: \]
806: which in the presence of free dislocations
807: do not have to satisfy the condition
808: \begin{equation} \label{Summ}
809: m_1+m_2+m_3=0\,.
810: \end{equation}
811:
812: This can be taken into account by making in Eqs.
813: \makebox{(\ref{f2})-(\ref{f4})} a replacement,
814: \[
815: (\nabla n)^2\Rightarrow \frac{2}{3}(m_1^2+m_2^2+m_3^2)\,,
816: ~~~
817: ({\bf e}_\alpha\nabla)n\Rightarrow m_\alpha\,,
818: \]
819: and adding to it a new contribution,
820: \[
821: f_{\rm D}=\frac{K_{\rm D}}{2}(m_1+m_2+m_3)^2
822: \]
823: (where $\ln K_{\rm D}\propto 2J_1/T$), which controls the fluctuations of
824: the difference between the densities of positive and negative dislocations.
825: However, the minimums of this new functional, which instead of the two
826: variables encoded in $\nabla n$ depends on the three variables $m_\alpha$, will
827: be achieved when they satisfy the condition (\ref{Summ}), and therefore
828: the conclusion on the first order of the transition
829: (obtained at $J_1=\infty$) will hold also at large but finite $J_1$.
830:
831:
832: \section{Comparison of different scenarios of disordering \label{md}}
833:
834: In the two previous sections we have demonstrated that the phase transition
835: to the completely disordered phase (which can be associated with the
836: formation of a network of single domain walls) has to be of the first
837: order.
838: This still leaves possibilities for two different scenarios of disordering.
839: The formation of a network of domain walls can either happen at
840: $T=T_{\rm nw}>T_2$ as a separate phase transition, or at $T_{\rm nw}\leq T_2$.
841: In the latter case at $T>T_{\rm nw}$ the system is already in the
842: disordered phase (in which the domains of all six ground states are intermixed
843: with each other) and nothing special can be expected to happen at $T=T_2$.
844: In that case the only phase transition takes place at $T=T_{\rm nw}$.
845:
846: In the Appendix we demonstrate how at $J_1=\infty$ one can construct
847: $T_c^+$, an exact upper boundary for $T_c$, the temperature of a phase
848: transition from the completely frozen phase, $T_c=\min\{T_2,T_{\rm nw}\}$.
849: The expression we obtain, Eq. (\ref{Tc+}), is applicable for an arbitrary
850: relation between $J_2$ and $J_3$.
851: In the case $J_3=0$ it gives \makebox{$T_c^{+}\approx 7.54\,J$,} which is
852: only 6\%
853: above the value of $T_2(J_3=0)=\gamma_2 T_2$ discussed in Sec. \ref{ddw}.
854: Since there are no reasons for this boundary from above
855: to give an extremely accurate estimate of $T_c$,
856: one can expect that the real value of the temperature at which
857: the fluctuations on the background of the completely frozen phase do
858: appear will be even lower than $T_2$, which means that for $J_3=0$
859: the single-transition scenario is realized.
860:
861: Comparison of Eq. (\ref{Tddw}) with Eq. (\ref{Tnw}) shows that in the
862: region of parameters in which one can rather accurately estimate
863: both $T_2$ and $T_{\rm nw}$, these two temperatures coincide with each
864: other with the logarithmical accuracy.
865: This gives a hint that both mechanisms may be different manifestations of
866: the same phenomenon.
867:
868: This idea looks even more plausible when one notices that the fluctuations of
869: a double domain wall considered in Sec. \ref{ddw} can also be discussed
870: in terms of zero modes. That is, the parallelogram in the middle
871: of Fig. \ref{3}(b) also can be considered as a domain which can be moved
872: % in parallel to itself (that is,
873: along the direction of domain walls that are adjacent
874: to it without changing the total length of the walls.
875: Apparently, a typical network will have a less regular structure than the
876: network shown in Fig. \ref{4} and will incorporate some fragments looking
877: like finite pieces of a double wall. However, the contribution to
878: the free energy from each zero mode has to be of the form
879: \[
880: cE_{\rm dw}L+(E_{\rm a}+E_{\rm b})-T\ln L\,,
881: \]
882: where $c\sim 1$, which after summation over all modes will reproduce
883: the general structure of Eq. (\ref{Fnw}).
884: This suggests that there should be only one phase transition
885: % (at $T=T_{\rm nw}$),
886: which is of the first order and is related to the
887: appearance of a network formed both by single and double domain walls.
888:
889: Comparison of Eq. (\ref{Tddw}) with Eq. (\ref{Tnw}) shows that
890: the realization of the two-transition scenario with $T_2<T_{\rm nw}$ requires
891: \makebox{$E_{\rm c}\ll E_{\rm mp}$}.
892: When only $J_1$, $J_2$ and $J_3$ are assumed to be non-zero, $E_{\rm c}$
893: satisfies the relation
894: \[
895: E_{\rm c}=E_a+E_b+E_{\rm dw}>E_{\rm mp}\,,
896: \]
897: and, therefore, any prerequisites for the phase transition splitting are
898: absent.
899: Nonetheless, they may appear when one takes into account the interaction
900: of more distant neighbors.
901:
902: In particular, if one includes into
903: consideration the interaction of fifth neighbors (see Fig. \ref{1b}),
904: it turns out that %the parameters describing the properties of domain walls
905: $E_{\rm dw}$, $E_{\rm ddw}$ and $E_{\rm c}$ remain unchanged,
906: and therefore $T_2$ should not depend on $J_5$.
907: On the other hand, $E_{\rm mp}$ % in Eq. (\ref{Fnw})
908: is increased by $8J_5$, which according to
909: Eq. (\ref{Tnw}) will shift the value of $T_{\rm nw}$ upwards.
910: Thus, one can expect that the increase of $J_5$ will lead
911: to the splitting of the phase transition into two.
912:
913: However, the positiveness of $J_4$ works in the opposite direction and
914: therefore in a realistic system in which the coupling constants
915: continuously depend on the distance between sites, the influence of $J_5$
916: is likely to be compensated by the influence of $J_4$ (unless $J_4$ is,
917: for some reasons, of the opposite sign). Nonetheless, the possibility of phase
918: transition splitting is not entirely prohibited, although its realization
919: in some physical system is not very probable.
920:
921:
922: \section{Conclusion\label{conc}}
923:
924: In the present work we have investigated the scenarios of disordering
925: of the striped phase which is formed
926: in a triangular-lattice Ising model with the antiferromagnetic interaction
927: of nearest and next-to-nearest neighbors.
928: % (satisfying the condition \makebox{$2J_3<J_2\ll J_1$}).
929: % In the limit of $J_1\rightarrow\infty$ such a system at low temperatures
930: % is frozen in one of its ground states.
931: Our analysis has shown that the destruction of such an ordering
932: has to take place via a single first-order phase transition.
933:
934: The nature of this transition becomes more transparent in the case
935: $J_2-2J_3\ll J_2$, when it can be discussed in terms of the
936: formation of a {\em diluted} network of domain walls,
937: which is characterized by an extensive number of zero modes.
938: In this limit one can find how the transition temperature and the value of
939: the correlation radius at the transition point depend on the parameters of
940: the model [see Eq. (\ref{Tnw}) and Eq. (\ref{Lc})].
941: In the opposite limit of $J_3\rightarrow 0$
942: the transition temperature has to be proportional to $J_2$, the only energy
943: scale which is relevant for $J_2\ll J_1$, whereas the correlation radius
944: at the transition point has to be comparable with 1.
945:
946: We also have shown that the formation of a domain-wall network could
947: be anticipated by a transition to the intermediate phase, in which
948: the long-range order in terms of $\sigma_{\bf j}$ is destroyed
949: by the spontaneous formation of a sequence of parallel (on the average)
950: double domain walls, whereas the long-range order in the orientation of
951: stripes formed by the spins of the same sign still exists.
952: This transition would be characterized by a combination of the Ising (in
953: the very narrow vicinity of the transition temperature) and the
954: Pokrovsky-Talapov (in a more wide temperature interval) critical behaviors.
955: However, in a system with only three coupling constants ($J_1$, $J_2$ and
956: $J_3$) no prerequisites for the realization of such a scenario can be
957: found.
958:
959: Nonetheless, they may appear when the interaction of more distant
960: neighbors is taken into account. In particular, the splitting of the phase
961: transition into two is favored by the increase of $J_5$. However, the
962: possibility of the realization of such a scenario depends on the fine interplay
963: between different coupling constants $J_k$ with $k\geq 4$ and in a
964: system with a monotonic dependence of $J_k$ on $k$ is not very probable.
965:
966: An additional mechanism favoring the two-transition scenario may be
967: related to quantum fluctuations, whose role in decreasing
968: the energy of double walls may be more prominent than in decreasing
969: the energy of a diluted network due to a smaller size of moving objects
970: in the former case.
971:
972: The numerical simulations of the triangular-lattice Ising antiferromagnet
973: have been performed in Refs. \makebox{\onlinecite{GP,NKL,RRT,TM}}.
974: In particular, Glosli and Plischke \cite{GP} have studied the system
975: with first and second neighbor interactions satisfying
976: $J_2/J_1=0.1$, Rastelli {\em et al.} \cite{RRT} - with
977: \makebox{$J_2/J_1=0.1,\,0.5,\,1$}, whereas Novikov {\em et al.} \cite{NKL}
978: have assumed that the interaction decays with the distance
979: exponentially. The results of these groups give evidence for the
980: existence of a single first-order transition,
981: which is consistent with our conclusions.
982: %which according to our analysis is the most probable scenario.
983: In the simulations of Takagi and Mekata \cite{TM} the disordering
984: of the striped phase has been investigated in the system with
985: \makebox{$J_2/J_1=0.2$} or $0.5$ and $J_3/J_1=-0.75$, but these authors make no
986: conclusions about the type of the single phase transition which they
987: observe.
988:
989: In addition to more traditional applications mentioned in the Introduction,
990: the considered version of the Ising model can be used for the description
991: of a triangular array of quantum dots at half-filling \cite{NKL}
992: and (at sufficiently low temperatures) of a Josephson junction
993: array with the dice lattice geometry and one-third of the flux quanta
994: per plaquette \cite{xdt}.
995: % Both these systems have recently been the subject
996: % of experimental investigations \cite{Bawendi,TTM}.
997: In the latter case the role of $J_k$ with $k>1$ is played by the magnetic
998: interactions of currents in the array \cite{xdt,fxe}.
999: The results of this work may also be of help for understanding the nature
1000: of phase transition(s) in the fully frustrated $XY$ model on a honeycomb
1001: lattice \cite{ShS85,fxh}, in which the fluctuation-induced vortex pattern
1002: \cite{fxh} has the same structure as in Fig. \ref{1a}.
1003:
1004: % The author is grateful to L.S. Levitov for interesting discussion.
1005: This work has been supported in part by the Program ``Quantum
1006: Macrophysics" of the Russian Academy of Sciences and
1007: by the Program ``Scientific Schools of the Russian Federation"
1008: (grant No. 1715.2003.2).
1009: % and by the Swiss National Science Foundation.
1010:
1011:
1012: %\newpage
1013: \appendix
1014:
1015:
1016: \section{}
1017:
1018: In the limit $J_1\rightarrow\infty$
1019: it is convenient to split the Hamiltonian
1020: (\ref{HIs}) into two terms, \[H=H_0+(H-H_0)\,,\] the first of which, $H_0$,
1021: includes only the infinite interaction of nearest neighbors and restricts
1022: the summation in the partition function to the states in which the
1023: constraint (\ref{cond}) is satisfied on all triangular plaquettes,
1024: whereas the second term, $H-H_0$, includes all other interactions.
1025: It is well known \cite{Feyn} that the application of a variational procedure
1026: allows one to use such a splitting to demonstrate that the free energy
1027: of the system, $F$, is bounded from above by
1028: \begin{equation} \label{Fplus}
1029: F^{+}\equiv F_0+\langle H-H_0\rangle_0 \,,
1030: \end{equation}
1031: where $F_0$ is the free energy of the system whose Hamiltonian is equal to
1032: $H_0$, whereas the angular brackets denote the average calculated with
1033: the help of $H_0$.
1034:
1035: In our case in the thermodynamic limit
1036: \begin{eqnarray*} % \label{F0}
1037: F_0 & = & -N(J_1+Ts_0)\,, \\
1038: %\]%end{equation} \[
1039: \langle H-H_0 \rangle_0 & = & 3N(J_2g_2+J_3g_3)\,,
1040: \end{eqnarray*}
1041: where $N$ is the total number of sites,
1042: \[%begin{equation} \label{s0}
1043: s_0\approx 0.323066
1044: \]%end{equation}
1045: is the residual entropy \cite{Wan}, whereas
1046: $%\[
1047: g_{k}=\langle \sigma_{\bf i} \sigma_{\bf j}\rangle_0
1048: $ %\]
1049: is the correlation function of the variables ${\sigma}$ on the sites
1050: ${\bf i}$ and ${\bf j}$ which are the $k$th neighbors of each other,
1051: calculated for the system with only nearest neighbor interaction at $T=0$.
1052: According to Stephenson \cite{St64},
1053: \[
1054: g_2= \frac{1}{9}+\frac{2}{\sqrt{3}\pi}\,,~~~ %\approx 0.479
1055: g_3= \frac{1}{9}-\frac{3\;}{\pi^2}\,. %\approx -0.193
1056: \]
1057:
1058: The comparison of $F^{+}(T)$
1059: % The temperature dependent boundary from above defined by Eq. (\ref{Fplus})
1060: % can be compared
1061: with the free energy of a completely frozen ground state,
1062: which, naturally, coincides with its energy,
1063: \[%\begin{equation} \label{E0}
1064: E_0=-N(J_1+J_2-3J_3)\,,
1065: \]%end{equation}
1066: allows one to conclude that the temperature $T_c$, at
1067: which a phase transition from a completely frozen state
1068: to some phase with more developed fluctuations takes
1069: place, cannot be larger than
1070: \begin{equation} \label{Tc+}
1071: T_c^{+}(J_2,J_3)=C_2 J_2-C_3J_3\,,
1072: \end{equation}
1073: where
1074: \begin{eqnarray} \label{C2}
1075: C_2 & = & {(1+3g_2)}/{s_0}\approx 7.54 \,,~
1076: \\ %\end{equation} \begin{equation} \label{C3}
1077: C_3 & = & {3(1-g_3)}/{s_0}\approx 11.08\,.
1078: \end{eqnarray}
1079: Naturally, this approach does not allow to distinguish if the phase
1080: transition at $T=T_c<T_c^+$ is a direct transition into the disordered
1081: phase or a transition to the intermediate phase in the framework of
1082: a two-transition scenario.
1083:
1084: For $J_3=0$ the value of $T_c^+$ following from Eqs. (\ref{Tc+}) and (\ref{C2})
1085: is rather close to $T_2$, the temperature of the spontaneous
1086: formation of double domain walls given by Eq. (\ref{Tddw0}),
1087: whereas in the limit of $E_{\rm dw}\rightarrow
1088: 0$ (that is \makebox{$J_3\rightarrow J_2/2$}) one gets
1089: \[
1090: T_c^+\approx 2.00\,J_2\,,
1091: \]
1092: which is compatible with an estimate for the temperature
1093: of the phase transition given by Eqs. (\ref{Tddw}) and (\ref{Tnw}).
1094:
1095:
1096: %\newpage $~$ \newpage
1097:
1098: \begin{thebibliography}{99}
1099: \bibitem{Ons} % H. A. Kramers and G. H. Wannier,
1100: % Phys. Rev. {\bf 60}, 252 (1941);
1101: L. Onsager, Phys. Rev. {\bf 65}, 117 (1944);
1102: G. H. Wannier,
1103: Rev. Mod. Phys. {\bf 17}, 50 (1945).
1104:
1105: % \bibitem{Vdv} N. V. Vdovichenko,
1106: % Zh. Eksp. Teor. Fiz. {\bf 47}, 715 (1964)
1107: % [Sov. Phys. JETP {\bf ??}, ??? (196?)].
1108: \bibitem{Wan} G.H. Wannier, Phys. Rev. {\bf 79}, 357 (1950);
1109: Phys. Rev. B {\bf 7}, 5017E (1973).
1110: \bibitem{Hout} R.M.F. Houtappel, Physica {\bf 16}, 425 (1950); %\\
1111: G.F. Newell, Phys. Rev. B {\bf 79}, 876 (1950); %\\
1112: K. Husimi and Y. Sy\^{ozi}, Prog. Theor. Phys. {\bf 5},
1113: 177 and 341 (1950).
1114: \bibitem{St64} J. Stephenson, J. Math. Phys. {\bf 5}, 1009 (1964).
1115: \bibitem{St70} J. Stephenson, J. Math. Phys. {\bf 11}, 413 (1970).
1116: \bibitem{BH} H.W.J. Bl\"{o}te and H.J. Hilhorst,
1117: J. Phys. A {\bf 15}, L631 (1982).
1118: \bibitem{Met} B.D. Metcalf, Phys. Lett. {\bf 46A}, 325 (1974);
1119: M. Kaburagi and J. Kanamori,
1120: Jpn. J. Appl. Phys., Suppl.: pt. 2, 145 (1974).
1121: \bibitem{NHB} B. Nienhuis, H.J. Hilhorst and H.W.J. Bl\"{o}te,
1122: J. Phys. A {\bf 17}, 3559 (1984).
1123: \bibitem{Weeks}J. D. Weeks, in {\em Order in Strongly Fluctuating
1124: Condensed Matter Systems}, edited by T. Riste
1125: (Plenum, New York - London, 1980), p. 293.%-317.
1126: \bibitem{Land}D.P. Landau, Phys. Rev. B {\bf 27}, 5604 (1983).
1127: % \bibitem{BKT} V.L. Berezinskii, Zh. Eksp. Teor. Fiz. {\bf 59}, 907 (1970)
1128: % [Sov. Phys. JETP {\bf 32}, 493 (1971)];
1129: % {\em ibid.} {\bf 61}, 1144 (1971) [{\bf 34}, 610 (1972)];
1130: % J.M. Kosterlitz and D.J. Thouless, J. Phys. C {\bf 5}, L124 (1972);
1131: % {\em ibid.} {\bf 6}, 1181 (1973);
1132: % J.M. Kosterlitz, {\em ibid.} {\bf 7}, 1046 (1974).
1133: \bibitem{AP} S. Alexander and P. Pincus,
1134: J. Phys. A {\bf 3}, 263 (1980).
1135: \bibitem{FSK} S. Fujiki, K. Shutoh, and S. Katsura,
1136: J. Phys. Soc. Jpn. {\bf 53}, 1371 (1984).
1137: \bibitem{MC} S. Fujiki, K. Shutoh, Y. Abe and S. Katsura,
1138: J. Phys. Soc. Jpn. {\bf 52}, 1531 (1983);
1139: H. Takayama, K. Matsumoto, H. Kawahara and K. Wada,
1140: {\em ibid.} {\bf 52}, 2588 (1983);
1141: S. Fujiki, K. Shutoh, S. Inawashiro, Y. Abe and S. Katsura,
1142: {\em ibid.} {\bf 55}, 3326 (1986);
1143: S. Miyashita, H. Kitatani and Y. Kanada,
1144: {\em ibid.} {\bf 60}, 1523 (1991);
1145: X. Qian and H.W.J. Bl\"{o}te,
1146: Phys. Rev. E {\bf 70}, 036112 (2004).
1147: \bibitem{Mek} M. Mekata, J. Phys. Soc. Jpn. {\bf 42}, 76 (1977);
1148: M. Kaburagi, T. Tonegawa and J. Kanamori,
1149: {\em ibid.} {\bf 51}, 3857 (1982).
1150: \bibitem{DSWG} E. Domany, M. Schick, J.S. Walker and R.B. Griffiths,
1151: Phys. Rev. B {\bf 18}, 2209 (1978).
1152: %\bibitem{ch-as} The chiral assymerty consists in the dependence of
1153: % domain-wall energy both on its orientation and on which states are separated by this wall.
1154: \bibitem{KTK} M. Kaburagi, T. Tonegawa and J. Kanamori,
1155: J. Magn. Magn. Mat. {\bf 31-34}, 1037 (1983).
1156: \bibitem{SH} P.A. Slotte and P.C. Hemmer,
1157: J. Phys. C {\bf 17}, 4645 (1984).
1158: \bibitem{EH} G. Einevoll and P.C. Hemmer,
1159: J. Phys. C {\bf 21}, L615 (1988).
1160: \bibitem{DR} E. Domany and E.K. Riedel,
1161: Phys. Rev. Lett. {\bf 40}, 561 (1978);
1162: Phys. Rev. B {\bf 19}, 5817 (1979).
1163: \bibitem{KLF} D. Kim, P.M. Levy and L.F. Uffer,
1164: Phys. Rev. B {\bf 12}, 989 (1975);
1165: A. Aharony, J. Phys. A {\bf 10}, 389 (1977).
1166: \bibitem{NRS} E.K. Riedel, Physica A {\bf 106}, 110 (1981);
1167: B. Nienhuis, E.K. Riedel and M. Schick,
1168: Phys. Rev. B {\bf 27}, 5625 (1983).
1169: \bibitem{Sch} M. Schick, Surf. Sci. {\bf 125}, 94 (1983).
1170: \bibitem{OH} S. Ostlund, Phys. Rev. B {\bf 24}, 398 (1981); % \\
1171: D.A. Huse, Phys. Rev. B {\bf 24}, 5180 (1981).
1172: \bibitem{GP} J. Glosli and M. Plischke,
1173: Can. J. Phys. {\bf 61}, 1515 (1983).
1174: \bibitem{NKL} D.S. Novikov, B. Kozinsky and L.S. Levitov,
1175: cond-mat/0111345 (2001).
1176: \bibitem{RRT} E. Rastelli, S. Regina and A. Tassi,
1177: Phys. Rev. B {\bf 71}, 174406 (2005).
1178:
1179: \bibitem{TM} T. Takagi and M. Mekata, {\bf 64}, 4609 (1995).
1180: \bibitem{KK} M. Kaburagi and J. Kanamori,
1181: J. Phys. Soc. Jpn. {\bf 44}, 718 (1978).
1182: \bibitem{BMVW} P. Bak, D. Mukamel, J. Villain and K. Wentowska,
1183: Phys. Rev. B {\bf 19}, 1610 (1979).
1184: \bibitem{PD} R.E. Peierls,
1185: Proc. Cambridge Phil. Soc., {\bf 32}, 477 (1936);
1186: C. Domb, in {\em Phase Transitions and Critical Phenomena},
1187: vol. 3, ed. by C. Domb and M. S. Green
1188: (New York: Academic Press, 1974).
1189: \bibitem{PT} V.L. Pokrovsky and A.L. Talapov,
1190: Phys. Rev. Lett. {\bf 42}, 65 (1979);
1191: Zh. Eksp. Teor. Fiz. {\bf 78}, 269 (1980)
1192: [Sov. Phys. JETP {\bf 51}, 134 (1980)].
1193: \bibitem{HF} D.A. Huse and M.E. Fisher,
1194: Phys. Rev. Lett. {\bf 49}, 793 (1982);
1195: Phys. Rev. B {\bf 29}, 239 (1984).
1196: \bibitem{Vil80a}J. Villain, in {\em Order in Strongly Fluctuating
1197: Condensed Matter Systems}, edited by T. Riste
1198: (Plenum, New York - London, 1980), p. 221.
1199: \bibitem{Vil80b}J. Villain, Surf. Sci. {\bf 97}, 219 (1980); % \\
1200: J. Villain and M.B. Gordon, Surf. Sci. {\bf 125}, 1
1201: (1983).
1202: \bibitem{CFHLB} S.N. Coppersmith, D.S. Fisher, B.I. Halperin,
1203: P.A. Lee and W.F. Brinkman,
1204: Phys. Rev. Lett. {\bf 46}, 549 and 869E (1981);
1205: Phys. Rev. B {\bf 25}, 349 (1982).
1206: \bibitem{SW} A.-C. Shi and M. Wortis,
1207: Phys. Rev. B {\bf 37}, 7793 (1988).
1208: \bibitem{BPT} T. Bohr, V.L. Pokrovsky and A.L. Talapov,
1209: Pis'ma ZhETF {\bf 35}, 165 (1982)
1210: [JETP Lett. {\bf 35}, 203 (1982)];
1211: T. Bohr, Phys. Rev. {\bf 25}, 6981 (1982).
1212: \bibitem{xdt} S.E. Korshunov, Phys. Rev. Lett. {\bf 94}, 087001 (2005).
1213: \bibitem{fxe} S.E. Korshunov, Phys. Rev. B {\bf 71}, 174501 (2005).
1214: %\bibitem{Bawendi}
1215: % C.B. Murray, D.J. Norris and M.G. Bawendi,
1216: % J. Am. Chem. Soc. {\bf 115}, 8706 (1995);
1217: % C.B. Murray, C.R. Kagan and M.G. Bawendi,
1218: % Science {\bf 270}(5240) p. 1335 (1995).
1219: % C.A. Leatherdale, C.R. Kagan, N.Y. Morgan, S.A. Empedocles,
1220: % M.A. Kastner and M.G. Bawendi,
1221: % Phys. Rev. B {\bf 62}, 2669 (2000);
1222: % {\bf 63}, 039901E (2001).
1223: % \bibitem{TTM} M. Tesei, R. Th\'{e}ron, and P. Martinoli,
1224: % unpublished (2004).
1225: \bibitem{ShS85} W.Y. Shih and D. Stroud,
1226: Phys. Rev. B {\bf 32}, 158 (1985);
1227: S.E. Korshunov, J. Stat. Phys. {\bf 43}, 17 (1986).
1228: \bibitem{fxh} S.E. Korshunov and B. Dou\c{c}ot,
1229: Phys. Rev. Lett. {\bf 93}, 097003 (2004).
1230: \bibitem{Feyn} R.P. Feynman, {\em Statistical Mechanics: A Set of Lectures}
1231: (Benjamen, Reading, 1972).
1232:
1233: \end{thebibliography}
1234: \end{document}
1235: