1: \documentclass[10pt,prl,aps,superscriptaddress,twocolumn,showpacs]{revtex4}
2: \usepackage{graphicx}
3:
4: \usepackage{amssymb}
5:
6: \begin{document}
7:
8: \title{Emergence of U(1) symmetry in the 3D XY model with Zq anisotropy}
9:
10: \author{Jie Lou}
11: \affiliation{Department of Physics, Boston University,
12: 590 Commonwealth Avenue, Boston, Massachusetts 02215}
13:
14: \author{Anders W. Sandvik}
15: \affiliation{Department of Physics, Boston University,
16: 590 Commonwealth Avenue, Boston, Massachusetts 02215}
17:
18: \author{Leon Balents}
19: \affiliation{Department of Physics, University of California, Santa
20: Barbara, CA 93106-4030}
21:
22:
23: \begin{abstract}
24: We study the three-dimensional XY model with a $Z_q$ anisotropic term. At temperatures $T < T_{\rm c}$
25: this dangerously irrelevant perturbation is relevant only above a length scale $\Lambda$, which diverges as a power of
26: the correlation length; $\Lambda \sim \xi^{a_{q}}$. Below $\Lambda$ the order parameter is U(1) symmetric. We derive
27: the full scaling function controlling the emergence of U(1) symmetry and use Monte Carlo results to extract the exponent
28: $a_q$ for $q=4,\ldots,8$. We find that $a_q \approx a_4 (q/4)^2$, with $a_4$ only marginally larger than $1$. We discuss
29: these results in the context of U(1) symmetry at ``deconfined" quantum critical points separating antiferromagnetic and
30: valence-bond-solid states in quantum spin systems.
31: \end{abstract}
32:
33: \date{\today}
34:
35: \pacs{75.10.Hk, 75.10.Jm, 75.40.Mg, 05.70.Fh}
36:
37: \maketitle
38:
39:
40: A salient feature of the recently proposed theory of "deconfined" quantum critical points, which separate N\'eel
41: and valence-bond-solid (VBS) ground states of antiferromagnets on the square lattice, is the emergence of U(1) symmetry
42: \cite{senthil}. The VBS is either dimerized on columns or forms a square pattern with plaquettes of four strongly
43: entangled spins \cite{levin,sachdevrmp}. In both cases there are four degenerate patterns and, thus, $Z_4$ symmetry is broken.
44: However, as the critical point is approached the theory predicts a length scale $\Lambda$, diverging faster than
45: the correlation length, $\Lambda \sim \xi^a$, $a>1$, below which the distinction between columnar and plaquette
46: VBS states disappears. The nature of the VBS state is manifested only when coarse-graining on length-scales
47: $l > \Lambda$, whereas for $l < \Lambda$ the $Z_4$ symmetry is unbroken and is replaced by an emergent U(1)
48: symmetry characterizing the fluctuations between columnar and plaquette order.
49:
50: Quantum Monte Carlo simulations \cite{sandvikvbs} of an $S=1/2$ Heisenberg model with four-spin couplings have
51: recently provided concrete evidence for a continuous N\'eel--VBS transition, and also detected U(1) symmetry
52: in the VBS order-parameter distribution $P(D_x,D_y)$, where $D_x$ and $D_y$ are VBS order parameters for horizontal
53: and vertical dimers. There is no trace of the expected $Z_4$ anisotropy in the VBS phase---the distribution
54: is ring shaped---although the finite-size scaling of the squared order parameter shows that the system
55: is long-range ordered. This can be interpreted as the largest studied lattice size $L=32 < \Lambda$.
56: A ring-shaped distribution was also found in simulations of an SU(N) generalization of the $S=1/2$ Heisenberg
57: model \cite{kawashima}---possibly a consequence of proximity of this system to a deconfined quantum-critical point.
58:
59: In order to better understand the U(1) features of these VBS states, and to guide future studies of them, we here
60: exploit a classical analogy. In the three-dimensional XY model including a $Z_q$-anisotropic term,
61: \begin{equation}
62: {\cal H}=-J\sum_{(i,j)}\cos(\theta_i-\theta_j)-h\sum_{i}\cos(q\theta_i),
63: \label{hamiltonian}
64: \end{equation}
65: the anisotropy is dangerously irrelevant for $q\ge 4$ \cite{jose,blankschtein,caselle,oshikawa,carmona}, i.e.,
66: the universality class is that of the isotropic XY model but the perturbation is relevant for $T < T_{\rm c}$
67: above a length-scale $\Lambda$. In the closely related $q$-state clock model, the anisotropy is dangerously
68: irrelevant for $q\ge 5$. While numerical studies \cite{scholten,miyashita,hove} have confirmed the irrelevance
69: of the anisotropy at $T_{\rm c}$, the associated $\Lambda$ has, to our knowledge, not been extracted numerically,
70: except for an analysis of the 3-state antiferromagnetic Potts model, which corresponds to $Z_6$
71: \cite{oshikawa,aharony}.
72:
73: Here we report results of Monte Carlo simulations for $4\le q \le 8$ on periodic-boundary lattices with $N=L^3$ sites
74: and $L$ up to $32$. In addition to Metroplis single-spin updates, we also use Wolff cluster updates \cite{wolff} to reduce
75: critical slowing down. We sample the order-parameter distribution $P(m_x,m_y)$, where
76: \begin{equation}
77: m_x=\frac{1}{N}\sum_{i=1}^N \cos(\theta_i),~~~m_y=\frac{1}{N}\sum_{i=1}^N \sin(\theta_i).
78: \end{equation}
79: The standard order parameter can be defined as
80: \begin{eqnarray}
81: \langle m\rangle & = & \int_{-1}^1 dm_x\int_{-1}^1 dm_y P(m_x,m_y)\left ( m_x^2+m_y^2 \right )^{1/2} \nonumber \\
82: & = & \int_0^1 dr \int_0^{2\pi} d\theta r^2P(r,\theta).
83: \label{mag}
84: \end{eqnarray}
85: We will compare this with an order parameter $\langle m_q\rangle$ which is sensitive to the angular distribution;
86: \begin{equation}
87: \langle m_q\rangle = \int_0^1 dr \int_0^{2\pi} d\theta r^2P(r,\theta)\cos(q\theta).
88: \label{magstar}
89: \end{equation}
90:
91: \begin{figure}
92: \centerline{\includegraphics[width=6.75cm,clip]{fig1.eps}}
93: \caption{(Color online) $P(m_x,m_y)$ at $h/J=1$ for $q=4,8$, $L=4,32$.
94: The temperature $T/J=2.17$ for $Z_4$ and $1.15$ for $Z_8$; both less than $T_{\rm c}/J \approx 2.20$. The size of
95: the histograms corresponds to $m_{x,y} \in [-1,1]$. Angular distributions $P(\theta)$ with $\theta \in [0,2\pi]$
96: are shown above each histogram.}
97: \label{his}
98: \vskip-3mm
99: \end{figure}
100:
101: While the finite-size scaling of $\langle m\rangle$ is governed by the correlation length $\xi$,
102: $\langle m_q\rangle$ should instead be controlled by the U(1) length scale $\Lambda$ \cite{oshikawa}, becoming
103: large for a system of size $L$ only when $L > \Lambda$.
104: Fig.~\ref{his} shows magnetization histograms at $h/J=1$ for $Z_4$ and $Z_8$ systems with $L=4$ and $32$. The angular
105: distribution $P(\theta)=\int drrP(r,\theta)$ is also shown. The average radius of the distribution is the magnetization
106: $\langle m\rangle$, which decreases with increasing $L$. The anisotropy, on the other hand, increases with $L$. This is
107: particularly striking for $Z_8$, where the $L=4$ histogram shows essentially no angular dependence, even though $T$ is very
108: significantly below $T_{\rm c}$, whereas there are 8 prominent peaks for $L=32$. Thus, in this case the U(1) length scale
109: $4 < \Lambda < 32$. For the $Z_4$ system $T$ is much closer to $T_{\rm c}$ but still some anisotropy is seen for $L=4$;
110: it becomes much more pronounced for $L=32$.
111:
112: It is instructive to examine a spin configuration with $m_x\approx m_y$, i.e., $\theta \approx \pi/4$. Fig.~\ref{sc} shows
113: one layer of a $Z_4$ system with $L=10$ below $T_{\rm c}$. The spins align predominantly along $\theta=0$ and $\theta=\pi/2$, with
114: only a few spins in the other two directions. Clearly there is some clustering of spins pointing in the same direction---the system
115: consists of two interpenetrating clusters. Essentially, the configuration corresponds to a size-limited domain wall between
116: $\theta=0$ and $\theta=\pi/4$ magnetized states.
117:
118: Hove and Sudb{\o} studied the $q$-state clock model and performed a course graining at criticality \cite{hove}. They found that
119: the structure in the angular distribution diminished with the size of the block spins for $q \ge 5$, as would be expected if the
120: anisotropy is irrelevant. Here we want to quantify the length scale $\Lambda$ at which the anisotropy becomes relevant for
121: $T < T_{\rm c}$. Consider first what would happen in a course graining procedure for a single spin configuration of an infinite system
122: in the ordered state very close to $T_{\rm c}$. The individual spins will of course exhibit $q$ preferred directions, as is seen
123: clearly in Fig.~\ref{sc}, i.e., there would be $q$ peaks in the probability distribution of angles $\theta_i$. Constructing block spins
124: of $l^3$ spins, we would expect the angular dependence to first become less pronounced because of the averaging over spins pointing
125: in different directions (again, as is seen in Fig.~\ref{sc}). Sufficiently close to $T_{\rm c}$ we would expect the distribution
126: to approach flatness. However, since we are in an ordered state, one of the $q$ preferred angles eventually has to become predominant,
127: and thus one peak in the histogram will start to grow. This happens at $l \approx \Lambda$. We cannot simulate the infinite system and
128: instead we carry out an analogous procedure as a function of the lattice size $L$, sampling a large number of configurations.
129: We calculate the order parameters $\langle m\rangle$ and $\langle m_q\rangle$, defined in Eqs.~(\ref{mag},\ref{magstar}),
130: and analyze them using
131: \begin{eqnarray}
132: \langle m\rangle & = & L^{-\sigma} f(tL^{1/\nu}), \label{mscale} \\
133: \langle m_q\rangle & = & L^{-\sigma} g(tL^{1/\nu_q}). \label{mqscale}
134: \end{eqnarray}
135: Here (\ref{mscale}) is the standard finite-size ansatz with $\sigma=\beta/\nu$, and the XY exponents are $\beta \approx 0.348$
136: and $\nu\approx 0.672$ \cite{compostrini}. Eq.~(\ref{mqscale}) is an intuitive generalization of (\ref{mscale}), which was
137: proposed and used also in Ref.~\cite{oshikawa}, but we can actually also derive the scaling function $g(X)$ exactly.
138:
139: \begin{figure}
140: \includegraphics[width=3.75cm,clip]{fig2.eps}
141: \caption{(Color online) Spins in one layer of the $Z_4$ model with $L=10$ at $h/J=1,T/J=1.9 < T_{\rm c}$. Here
142: $m_x \approx m_y$, corresponding to $\theta \approx \pi/4$ in $P(r,\theta)$. Arrows are color-coded according to
143: the closest $Z_4$ angle; $n\pi/2$, $n=0,1,2,3$.}
144: \label{sc}
145: \vskip-3mm
146: \end{figure}
147:
148: Let us consider the scaling behavior of the order-parameter distribution $P(\vec{m})$.
149: It depends upon the system size $L$ and the size of scaling operators that perturb the
150: critical theory. Specifically, we consider the temperature deviation $t = T_{\rm c}-T$ and the
151: presumed irrelevant $q$-fold anisotropy strength $h$. By conventional scaling arguments,
152: we expect
153: \begin{equation}
154: \label{eq:1}
155: P(\vec{m};L,t,h) = L^{\sigma/2} \hat{P}(L^\sigma
156: \vec{m},tL^{1/\nu},H=h
157: L^{3-\Delta_q}),
158: \end{equation}
159: where $\Delta_q>3$ is the scaling dimension of the irrelevant anisotropy.
160: The prefactor above is determined from normalization of the
161: probability distribution. In the scaling regime, $|t|\ll 1, L\gg 1$, so
162: $H$ is small. When the first two arguments are $O(1)$, $\hat{P}$ can be
163: well-approximated by taking $H=0$ [with ``corrections to scaling'' of
164: $O(H)$, i.e. suppressed by $L^{3-\Delta_q}$ for a large system]. At
165: $H=0$, the distribution is fully XY symmetric, and the integral in Eq.~(\ref{magstar})
166: vanishes. Thus, in this regime $\langle m_q\rangle$ is small, $O(H)$,
167: and should be considered as arising from corrections to scaling. This simply
168: reflects the irrelevance of the anisotropy at the critical point.
169:
170: Because the anisotropy is {\sl dangerously} irrelevant, a larger
171: contribution, however, emerges when $tL^{1/\nu}\gg 1$, i.e. $L\gg \xi
172: \sim t^{-\nu}$. In this limit, the system can be regarded as possessing
173: long-range XY order, and the only significant fluctuations are the {\sl
174: global} fluctuations of the XY phase $\theta$. This is biased by the
175: anisotropy. The scale $\kappa$ of the total anisotropy (free) energy
176: can be estimated by its typical magnitude within an XY correlation
177: volume, $h\xi^{3-\Delta_q}$, multiplied by the number of correlation
178: volumes, $(L/\xi)^3$, i.e. $\kappa=h L^3 \xi^{-\Delta_q}$. Note that
179: although the energy per correlation volume is small (due to the
180: irrelevance of anisotropy at the critical point), the number of
181: correlation volumes becomes very large and more than compensates for
182: this smallness for $L/\xi$ sufficiently large.
183:
184: From this argument, we see that for $L/\xi\gg 1$, the distribution of
185: angles $\theta$ is just determined from a Boltzmann factor for a single
186: XY spin with the $q$-fold anisotropy energy $\sim -\kappa \cos
187: q\theta$. Furthermore, for $L/\xi\gg 1$, the magnitude
188: $|\vec{m}|\approx \langle m\rangle$ is approximately non-fluctuating.
189: Thus the distribution factors into the form $P(|m|,\theta) = \langle
190: m\rangle^{-2} \delta(|m|-\langle m\rangle) P(\theta)$, with
191: \begin{equation}
192: \label{eq:2}
193: P(\theta)=
194: \frac{1}{Z} e^{ \kappa \cos (q\theta)}.
195: \end{equation}
196: Here $Z=\int_0^{2\pi}\! d\theta\, e^{ \kappa \cos (q\theta)}$ is the
197: single-spin partition function. It is then straightforward to obtain
198: from Eq.~(\ref{magstar})
199: \begin{equation}
200: \label{eq:3}
201: \langle m_q\rangle = \langle m\rangle \frac{I_1(\kappa)}{I_0(\kappa)},
202: \end{equation}
203: where $I_n$ is the modified Bessel function of order $n$. Oshokawa obtained a similar
204: expression in a different way, but we disagree with his scaling variable. Comparing
205: this with the scaling form in Eq.~(\ref{mqscale}), we see that $\nu_q=\Delta_q
206: \nu/3$ ($a_q=\Delta_q/3$), $\kappa=h (t L^{1/\nu_q})^{3\nu_q}$, and
207: \begin{equation}
208: \label{eq:4}
209: g(X) \propto \frac{I_1(\tilde h X^{3\nu_q})}{I_0(\tilde h X^{3\nu_q})}.
210: \end{equation}
211: Here $\tilde h$ should be viewed as a non-universal scale factor. From the
212: above discussion, one sees that this form is valid for $L/\xi \gg 1$ but $L/\xi_q$
213: arbitrary. For $L/\xi$ of $O(1)$
214: or smaller, $L/\xi_q \ll 1$ (implying $\kappa,X\ll 1$), and the scaling
215: form for $\langle m_q\rangle$ becomes small and of order the expected correction
216: to scaling in the critical regime.
217:
218: \begin{figure}
219: \includegraphics[height=10.25cm,clip]{fig3.eps}
220: \caption{(Color online) The XY order parameter $\langle m\rangle$ (solid curves) and the $Z_q$ order parameter $\langle m_q\rangle$
221: (dashed curves) vs temperature for $q=4,5,6$. The system sizes are $L=8,10,12,14,16,24$, and $32$. The curves become
222: sharper (increasing slope) around $T_{\rm c}$ (indicated by vertical lines). The ratios $h/J$ used are indicated on the graphs.}
223: \label{mm}
224: \vskip-3mm
225: \end{figure}
226:
227: In Fig.~\ref{mm} we show results for the two order parameters for systems with $q=4,5,6$. We have studied several values of
228: $h/J$ and here show results for a different value for each $q$. We have extracted $T_{\rm c}$ using finite-size scaling of $\langle m\rangle$
229: with Eq.~(\ref{mscale}) and the XY exponents. This works very well for all $q$, confirming the irrelevance of $h$. The magnetization for
230: $T < T_{\rm c}$ is seen to decrease marginally with increasing $q$ in Fig.~\ref{mm}. The $Z_q$ order-parameter $\langle m_q\rangle$ changes
231: more drastically, being strongly suppressed close to $T_{\rm c}$ for large $q$. This is expected, as $\langle m_q\rangle$
232: should vanish for all $T$ in the XY limit $q\to \infty$. For $Z_4$, the $\langle m_q\rangle$ curves for different $L$ cross
233: each other, with the crossing points moving closer to $T_{\rm c}$ as $L$ increases. This is consistent with the above discussion
234: of course-graining: In the ordered state close to $T_{\rm c}$, $\langle m_q\rangle$ should first, for small $L$, decrease with
235: increasing $L$ as the $q$-peaked structure in $P(\theta)$ diminishes due to averaging over more spins. For larger $L$, $\langle m_q\rangle$
236: starts to grow with $L$ as the length-scale $\Lambda$ is exceeded. This behavior is more difficult to observe directly for $q=5,6$
237: because $\langle m_q\rangle$ is small and dominated by statistical noise close to $T_{\rm c}$ where the curves cross.
238:
239: Fig.~\ref{mszq} shows finite-size scaling of the $Z_q$ order-parameter $\langle m_q\rangle$, using the hypothesis
240: (\ref{mqscale}) and the XY value for $\sigma$. Adjusting $\nu_q=a_q\nu$ for each $q$ we find satisfactory data collapse using
241: $a_4=1.07(3)$, $a_5=1.6(1)$, $a_6=2.4(1)$, and, not shown in the figure, $a_8=4.2(3)$. These results are consistent with
242: the form $a_q = a_4(q/4)^2$, in qualitative agreement with the $\epsilon$-expansion by Oshikawa, which gave $a_q \to q^2/30$
243: for large $q$ \cite{oshikawa}. However, in the $\epsilon$-expansion there are significant deviations from the $q^2$ form in
244: the range of $q$ values considered here. Our $a_6$ is also smaller than the value $\approx 3.5$ obtained on the basis of the
245: 3-state Potts antiferromagnet \cite{oshikawa}.
246:
247: In Fig.~\ref{mszq} we also show the scaling function (\ref{eq:4}). It does not match exactly the collapsed
248: data, but the agreement improves as $q$ increases. As we have discussed above, the scaling function
249: represents the dominant behavior for $T < T_{\rm c}$, but exactly at $T_{\rm c}$ this contribution vanishes and the
250: critical-point scaling form $\langle m_q(T_{\rm c}) \rangle \sim L^{3-\Delta_q}$ becomes dominant. For $q=4$, $\Delta_4-3$ is
251: small; our estimate is $\Delta_4-3=0.21(9)$, in good agreement with previous estimates of the scaling dimension
252: \cite{caselle,carmona}. Thus it is clear that very large systems would be required for this contribution to become
253: invisible on the scale used in our graph. It is also clear that for $t < 0$, close to $t=0$, there will be a similarly
254: significant correction to the asymptotically dominant scaling function. As $q$ increases, we have seen that $\Delta_q=3a_q$
255: increases rapidly, and we thus expect significantly smaller correction to scaling. For $q=6$ the agreement is already
256: seen to be quite good, considering that our lattices are not very large.
257:
258: \begin{figure}
259: \includegraphics[height=10.25cm,clip]{fig4.eps}
260: \caption{(Color online) Scaling of the $Z_q$ magnetization. We use $\sigma=0.52$ for all $q$,
261: and $\nu_q$ as indicated in the plots. The colors of the curves correspond to $L$ as in Fig.~\ref{mm}.
262: The dashed curves are the predicted scaling functions with $\tilde h$ and prefactors adjusted to fit
263: the data approximately.}
264: \label{mszq}
265: \vskip-3mm
266: \end{figure}
267:
268: To conclude, we relate our results to the quantum VBS states discussed in the introduction. Returning to
269: Fig.~\ref{sc}, associating $\theta_i \approx 0$ arrows with two adjacent horizontal dimers on even-numbered columns and
270: $\theta_i \approx \pi/2$ with vertical adjacent dimers on even rows, $\langle \theta \rangle=0,\pi/2$ correspond to
271: columnar VBS states. A plaquette is a superposition of horizontal and vertical dimer pairs, whence a plaquette VBS
272: corresponds to $\langle \theta\rangle=\pi/4$ \cite{levin}. Rotating the arrows by $90^\circ$ corresponds to translating
273: or rotating a VBS. Either a columnar or plaquette VBS should obtain in the infinite-size limit, but close to a deconfined
274: quantum-critical point, for $L < \Lambda$, the system fluctuates among all mixtures of plaquette and columnar states.
275: This corresponds to a ring-shaped VBS order-parameter histogram. In numerical studies of quantum antiferromagnets
276: \cite{sandvikvbs,kawashima} no 4-peak structure was observed in the angular distribution, and hence it is not clear what
277: type of VBS finally will emerge (although a method using open boundaries favors a columnar state in \cite{sandvikvbs}).
278: It seems unlikely that the U(1) symmetry should persist as $L \to \infty$. In the classical $Z_4$ model we never observe a
279: perfectly U(1)-symmetric histograms far inside the ordered phase, in contrast to Refs.~\cite{sandvikvbs,kawashima}. On the
280: other hand, $a_q$ is larger for $q>4$, and in Fig.~\ref{his} we have shown a prominently U(1)-symmetric histogram for the
281: $Z_8$ model deep inside the ordered phase. Thus, the exponent $a$ may be larger for the $Z_4$ quantum VBS than $a_4 \approx 1$
282: obtained here for the classical $Z_4$ model. There is of course no reason to expect them to be the same, as the universality class
283: of deconfined quantum-criticality is not that of the classical $Z_4$ model \cite{senthil,sandvikvbs}. Future numerical studies
284: of VBS states and deconfined quantum-criticality can hopefully reach sufficiently large lattices to extract the U(1) exponent
285: using the scaling method employed here.
286:
287: We would like to thank Kevin Beach, Masaki Oshikawa, Andrea Pelissetto, and Ettore Vicari for useful
288: discussions and comments. This research is supported by NSF Grant No.~DMR-0513930.
289:
290: \null\vskip-9mm
291:
292: \begin{thebibliography}{00}
293:
294: \bibitem{senthil}
295: T. Senthil, A. Vishwanath, L. Balents, S. Sachdev, and M. P. A. Fisher, Science \textbf{303}, 1490 (2004).
296:
297: \bibitem{sachdevrmp}
298: S. Sachdev, Rev. Mod. Phys. \textbf{75}, 913 (2003).
299:
300: \bibitem{levin}
301: M. Levin and T. Senthil, Phys. Rev. B {\bf 70}, 220403 (2004).
302:
303: \bibitem{sandvikvbs}
304: A. W. Sandvik, Phys. Rev. Lett. {\bf 98}, 227202 (2007).
305:
306: \bibitem{kawashima}
307: N. Kawashima and Y. Tanabe, Phys. Rev. Lett. {\bf 98}, 057202 (2007).
308:
309: \bibitem{jose}
310: J. V. Jose, L. P. Kadanoff, S. Kirkpatrick, and D. R. Nelson,
311: Phys. Rev. B {\bf 16}, 1217 (1977).
312:
313: \bibitem{blankschtein}
314: D. Blankschtein, M. Ma, A. N. Berker, G. S. Grest, and C. M. Soukoulis, Phys. Rev. B {\bf 29}, 5250 (1984).
315:
316: \bibitem{caselle}
317: M. Caselle and M. Hasenbusch, J. Phys. A {\bf 31}, 4603 (1998).
318:
319: \bibitem{oshikawa}
320: M. Oshikawa, Phys. Rev. B {\bf 61}, 3430 (2000).
321:
322: \bibitem{carmona}
323: J. M. Carmona, A. Pelissetto, and E. Vicari, Phys. Rev. B {\bf 61}, 15136 (2000).
324:
325: \bibitem{scholten}
326: P. D. Scholten and L. J. Irakliotis, Phys. Rev. B {\bf 48}, 1291 (1993).
327:
328: \bibitem{miyashita}
329: S. Miyashita, J. Phys. Soc. Jpn. {\bf 66}, 3411 (1997).
330:
331: \bibitem{hove}
332: J. Hove and A. Sudb\o, Phys. Rev. E {\bf 68}, 046107 (2003).
333:
334: \bibitem{aharony}
335: A related $Z_6$ problem has been studied by A. Aharony, R. J. Birgeneau, J. D. Brock, and J. D. Litster,
336: Phys. Rev. Lett. {\bf 57}, 1012 (1986).
337:
338: \bibitem{wolff}
339: U. Wolff, Phys. Rev. Lett. {\bf 62}, 361 (1989).
340:
341: \bibitem{gottlob}
342: A. P. Gottlob and M. Hasenbusch, Physica A {\bf 201}, 593 (1993).
343:
344: \bibitem{compostrini}
345: M. Campostrini, M. Hasenbusch, A. Pelissetto, P. Rossi, and E. Vicari, Phys. Rev. B {\bf 63}, 214503 (2001).
346:
347: \end{thebibliography}
348:
349: \end{document}
350:
351: