1: \documentclass[prd,eqsecnum,twocolumn,amsfonts,showpacs]{revtex4}
2:
3: \input epsf
4:
5: \usepackage{graphicx}
6:
7: \usepackage{bm}
8:
9: \setlength{\unitlength}{1cm}
10:
11: \newcommand{\beq}{\begin{equation}}
12: \newcommand{\eeq}{\end{equation}}
13: \newcommand{\beqs}{\begin{eqnarray}}
14: \newcommand{\eeqs}{\end{eqnarray}}
15: \newcommand{\lsim}{\mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle<}{\sim}$}}}
16: \newcommand{\gsim}{\mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle>}{\sim}$}}}
17: %\newtheorem{th}{Theorem}[section]
18: \newtheorem{cor}{Corollary}[section]
19: \newtheorem{lemma}{Lemma}[section]
20: \newtheorem{defi}{Definition}[section]
21: \newtheorem{conj}{Conjecture}[section]
22:
23: \begin{document}
24:
25: \title{Zeros of the Potts Model Partition Function in the Large-$q$ Limit}
26:
27: \author{Shu-Chiuan Chang$^{a,b}$ \thanks{email: scchang@mail.ncku.edu.tw}
28: Robert Shrock$^{c}$ \thanks{email: robert.shrock@sunysb.edu}}
29:
30: \bigskip
31:
32: \affiliation{(a) \ Department of Physics \\
33: National Cheng Kung University \\
34: Tainan 70101, Taiwan}
35:
36: \bigskip
37:
38: \affiliation{(b) \ Physics Division \\
39: National Center for Theoretical Science \\
40: National Taiwan University \\
41: Taipei 10617, Taiwan}
42:
43: \bigskip
44:
45: \affiliation{(c) \ C. N. Yang Institute for Theoretical Physics \\
46: State University of New York \\
47: Stony Brook, N. Y. 11794 }
48:
49: \begin{abstract}
50:
51: We study the zeros of the $q$-state Potts model partition function
52: $Z(\Lambda,q,v)$ for large $q$, where $v$ is the temperature variable and
53: $\Lambda$ is a section of a regular $d$-dimensional lattice with coordination
54: number $\kappa_\Lambda$ and various boundary conditions. We consider the
55: simultaneous thermodynamic limit and $q \to \infty$ limit and show that when
56: these limits are taken appropriately, the zeros lie on the unit circle
57: $|x_\Lambda|=1$ in the complex $x_\Lambda$ plane, where $x_\Lambda=v
58: q^{-2/\kappa_\Lambda}$. For large finite sections of some lattices we also
59: determine the circular loci near which the zeros lie for large $q$.
60:
61: \pacs{05.20.-y, 64.60.Cn, 75.10.Hk}
62:
63: \end{abstract}
64:
65: \maketitle
66:
67: \newpage
68: \pagestyle{plain}
69: \pagenumbering{arabic}
70:
71: \section{Introduction}
72: \label{sectionI}
73:
74: In this paper we shall study the $q$-state Potts model in the limit $q \to
75: \infty$ and shall present some new results on the complex-temperature zeros of
76: the partition function in this limit. The Potts model has served as a valuable
77: model for the study of phase transitions and critical phenomena
78: \cite{wurev,baxterbook}. The $q \to \infty$ limit of the model is exactly
79: solvable on any lattice \cite{pg}, as can be seen from the fact that in this
80: limit the model essentially reduces to a single-site problem. Since the $q \to
81: \infty$ limit of the Potts model is solvable, it is natural to investigate the
82: properties of the model for large $q$. Indeed, although we will focus on
83: partition function zeros rather than physical applications, we note that the
84: Potts model with large $q$ has been used in modelling the kinetic behavior of
85: soap froths \cite{gag}.
86:
87: The (zero-field) Potts model is defined, for temperature $T$ on a lattice
88: $\Lambda$, or more generally, a graph $G$, by the partition function
89: %
90: \beq
91: Z(G,q,v) = \sum_{ \{ \sigma_n \} }
92: \exp ( K\sum_{\langle i j \rangle} \delta_{\sigma_i \sigma_j} )
93: \label{z}
94: \eeq
95: %
96: where $\sigma_i=1,...,q$ are the classical spin variables on each vertex (site)
97: $i \in G$, $\langle i j \rangle$ denotes pairs of adjacent vertices, $K=\beta
98: J$ where $\beta = (k_BT)^{-1}$, and $J$ is the spin-spin coupling. We use the
99: notation $a=e^K$, and $v=a-1=e^K-1$. For the Potts ferromagnet, $J > 0$ so $v
100: \ge 0$, while for the antiferromagnet, $J < 0$ so that $-1 \le v \le 0$. The
101: graph $G$ is formally defined by its set of vertices $V$ and its set of edges
102: (= bonds) $E$; we denote the number of vertices of $G$ as $n=n(G)=|V|$ and the
103: number of edges of $G$ as $e(G)=|E|$. The complex-temperature zeros of
104: $Z(G,q,v)$, i.e., the zeros in the complex $v$ plane, have the property that on
105: a regular lattice graph with dimensionality greater than the lower critical
106: dimensionality, $d \ge 2$, in the thermodynamic limit $L_j \to \infty$ with
107: $L_j/L_k$ a finite nonzero constant, $1 \le j,k \le d$, they merge to form a
108: locus ${\cal B}$ consisting of curves that bound regions of different analytic
109: behavior for the corresponding free energy $F = -k_BTf$, where the
110: dimensionless function $f$ is given by $f = \lim_{n \to \infty} n^{-1} \ln Z$.
111: For the thermodynamic limit of a graph which is a section of a regular lattice
112: graph $\Lambda$ with some boundary conditions, we denote $\kappa_\Lambda$ as
113: the coordination number satisfying
114: $\kappa_\Lambda = 2 \lim_{n(G) \to \infty} \ e(G)/n(G)$
115: in this limit (independent of boundary conditions). Reviews of the Potts model
116: include \cite{wurev,baxterbook}.
117:
118: Let $G^\prime=(V,E^\prime)$ be a spanning subgraph of $G$, i.e. a subgraph
119: having the same vertex set $V$ and a subset of the edge set, $E^\prime
120: \subseteq E$. Then a useful representation of $Z(G,q,v)$ is \cite{kf}
121: %
122: \beq
123: Z(G,q,v) = \sum_{G^\prime \subseteq G} q^{k(G^\prime)}v^{e(G^\prime)}
124: \label{cluster}
125: \eeq
126: %
127: where $k(G^\prime)$ denotes the number of connected components of $G^\prime$.
128: Since we only consider connected lattice graphs $G$, we have $k(G)=1$. For the
129: ferromagnet, eq. (\ref{cluster}) allows one to generalize $q$ from the positive
130: integers to the positive reals, ${\mathbb R}_+$. As is evident from
131: eq. (\ref{cluster}), $Z(G,q,v)$ is a polynomial in $q$ with minimal and maximal
132: degrees 1 and $n(G)$, and a polynomial in $v$ with minimal and maximal degrees
133: 0 and $e(G)$. Thus, for a finite $G$, $Z(G,q,v)$ is completely determined by
134: its zeros and can be written as $Z(G,q,v) =\prod_{i=1}^{e(G)}[v-v_i(q)]$.
135:
136: We first recall some previous related work. For the special case $q=2$ on the
137: square lattice, where one has the exact Onsager solution for the partition
138: function, it was found that (in the thermodynamic limit) the locus ${\cal B}$
139: consists of the union of circles $|a \pm 1|=\sqrt{2}$ \cite{fisher}. The locus
140: ${\cal B}$ has also been determined exactly for $q=2$ on various other $d=2$
141: lattices (e.g., \cite{abe,cmo}). However, aside from the special case $q=2$,
142: $d=2$, one does not, in general, know ${\cal B}$ for lattice graphs of
143: dimensionality $d \ge 2$. Indeed, studies of complex-temperature zeros of the
144: Potts model partition function for finite sections of various lattices have
145: found that they exhibit a considerable scatter, especially in the half-plane
146: with $Re(a) < 0$ \cite{mm}-\cite{kc}. A complementary approach has been to
147: obtain exact solutions for the partition function and finite-width strips of
148: lattices with various boundary conditions and hence determine exactly the locus
149: ${\cal B}$ in the infinite-length limit (e.g., \cite{a}-\cite{ts}. These
150: systems are quasi-one-dimensional, so that ${\cal B}$ does not cross the
151: positive real $a$ axis, i.e., the free energy is analytic for all nonzero
152: temperatures. With the definition $x_{sq} = v/\sqrt{q}$ for the square
153: lattice, it was found the complex-temperature zeros calculated for finite
154: sections of this square lattice with duality-preserving boundary conditions
155: have the property that a subset lies exactly on an arc of the unit circle
156: $|x_{sq}|=1$ in the complex $x_{sq}$ plane for $Re(x_{sq}) \gsim 0$
157: \cite{mbook,chw,wuetal,sdg}. For physical temperature the coefficients of
158: powers of $a=e^K$ are positive, so there are no zeros on the positive real axis
159: $Re(a) > 0$ for any finite lattice. However, in the thermodynamic limit, the
160: phase boundary on ${\cal B}$ crosses this axis at the ferromagnetic phase
161: transition point, $a_c=1+\sqrt{q}$, i.e. $x_c=1$. A number of interesting
162: results have been obtained by F. Y. Wu and collaborators: in Ref. \cite{chw} a
163: conjecture was made that for finite sections of the square lattice with
164: self-dual boundary conditions and for the same lattice with free or periodic
165: boundary conditions in the thermodynamic limit, the complex-temperature zeros
166: of the partition function in the $Re(x_{sq}) > 0$ half-plane are located on the
167: unit circle $|x_{sq}|=1$. In Ref. \cite{wuetal}, it was shown, using Euler's
168: identity for integer partitions, that for the Potts model on the self-dual
169: square lattice, if one takes the limit $q \to \infty$, all of the zeros of the
170: partition function are located on this unit circle $|x_{sq}|=1$ for any lattice
171: size. In Refs. \cite{wuetal,wu3,huangwu} it was shown that partition function
172: for the Potts model on a finite $d$-dimensional Cartesian lattice ${\mathbb
173: E}^d$ with $d \ge 2$ and appropriate boundary conditions, in the limit $q \to
174: \infty$, divided by an appropriate power of $q$, is given by the generating
175: function of restricted integer partitions in $d-1$ dimensions, with expansion
176: variable $t=v^d/q$.
177:
178: In the present paper we present some results on the zeros of the Potts model
179: partition function for large $q$ on general regular $d$-dimensional lattices,
180: both in the thermodynamic limit and for finite lattices, with various boundary
181: conditions. We shall give a simple proof for an arbitrary regular lattice of
182: dimension $d \ge 2$ that, when the thermodynamic limit and the $q \to \infty$
183: limit are taken in an appropriate way, the zeros of the partition function lie
184: on the circle $|x_\Lambda|=1$. This generalizes the corresponding result of Wu
185: and coworkers from a Cartesian lattice ${\mathbb E}^d$ to an arbitrary regular
186: $d$-dimensional lattice. We also consider finite sections of some regular
187: lattices and determine the circular loci near which the zeros lie for large
188: $q$. We shall often call this the ``approximating circle'' for the zeros.
189:
190:
191: \section{General Results for Large $q$}
192: \label{sectionII}
193:
194: In this section we analyze the pattern of partition function zeros of the Potts
195: model for large $q$. Let us first arrange the terms in eq. (\ref{cluster}) in
196: order of decreasing power of $v$, from $v^{e(G)}$ to 1. There are ${e(G)
197: \choose j}$ terms in the coefficients of $v^{e(G)-j}$ and $v^j$, so that in the
198: special case $q=1$ the partition function reduces to $Z(G,q=1,v)=(v+1)^{e(G)}$.
199: These ${e(G) \choose j}$ terms can have different powers in $q$ except for very
200: small or very large $j$, as follows. The term with the highest power of $v$
201: arises from the contribution of $G'=G$ in eq. (\ref{cluster}), where all of the
202: edges of $G$ are present, and the power of $v$ is $e(G)$; since this graph
203: consists of a single connected component, the coefficient is $q$ and the term
204: is $qv^{e(G)}$. In considering other terms in $Z(G,q,v)$, we assume that the
205: graph $G$ has the property that the removal of a few edges does not cut it into
206: disconnected components. This excludes the cases of a one-dimensional lattice
207: and quasi-one-dimensional lattice strips of narrow widths; we comment on these
208: in the appendix. For example, for a regular $d$-dimensional lattice graph, we
209: require that $d \ge 2$ and that the lengths $L_k$ with $1 \le k \le d$ are
210: great enough so that this condition is satisfied. Next, the term in $Z(G,q,v)$
211: with one lower power of $v$ is obtained from the set of $G^\prime$ each of
212: which has an edge set $E'$ with one less edge than the full edge set $E$. This
213: term is ${e(G) \choose 1} \, q \, v^{e(G)-1}$, where ${\ell \choose
214: m}=\ell!/[m!(\ell-m)!]$. Similarly, the first few terms with descending powers
215: of $v$ below the maximal power are given by ${e(G) \choose j} \, q \,
216: v^{e(G)-j}$, where $j$ edges are removed in $G^\prime$ and the linear
217: dependence on $q$ reflects the property that this edge removal maintains the
218: connectedness of the resulting $G'$, so $k(G')=1$. In general, this holds for
219: $j < \delta(G)$, where $\delta(G)$ and $\Delta(G)$ denote the minimal and
220: maximal degrees of a vertex in $G$. (The degree of a vertex is defined as the
221: number of edges connected to it). Thus, for example, a term involving $q^2$
222: begins to appear in the coefficient of $v^{e(G)-3}$ for cyclic strips of the
223: square lattice with $L_1 \equiv L_x > 2$ and $L_2 \equiv L_y \ge 2$ since the
224: degree of vertices on the upper and lower boundaries is three, but it appears
225: in the coefficient of $v^{e(G)-2}$ for the same strip with $L_x=2$. At the
226: other end of the polynomial $Z(G,q,v)$ ordered as powers of $v$, the term with
227: the power of $v$ equal to zero arises from the contribution of $G'$ with
228: $E'=\emptyset$, i.e., the spanning subgraph containing no edges, leaving the
229: $n(G)$ vertices as disconnected components; this term is thus $q^{n(G)}$. The
230: term linear in $v$ results from the contributions of spanning subgraphs
231: $G^\prime$ with one edge and is $e(G) \, q^{n(G)-1} \, v$. In general, the
232: first few such terms are given by ${e(G) \choose j} \, q^{n(G)-j} \, v^j$,
233: where $G^\prime$ consist of $j$ edges such that $j < g(G)$, with $g(G)$
234: denoting the girth of $G$, i.e., the length of minimum closed circuit on $G$.
235: For example, for cyclic strips of the square lattice again, the term
236: $q^{n(G)-3}$ appears first in the coefficient of $v^4$ for $L_x > 3$ because
237: the girth is four, but term $q^{n(G)-2}$ appears first in the coefficient of
238: $v^3$ if $L_x=3$ and the term $q^{n(G)-1}$ appears in the coefficient of $v^2$
239: if $L_x=2$. Thus, for graphs $G$ satisfying the above-mentioned conditions, we
240: have
241: %
242: \beqs
243: & & Z(G,q,v) = \sum_{j=0}^{e(G)} c_j(q) v^j \cr\cr
244: & = & qv^{e(G)} + e(G)qv^{e(G)-1} + {e(G) \choose 2}qv^{e(G)-2} + ... \cr\cr
245: & + & {e(G) \choose 2}q^{n(G)-2}v^2 + e(G)q^{n(G)-1}v + q^{n(G)} \cr\cr
246: & &
247: \label{Zexp}
248: \eeqs
249: %
250: where the coefficients $c_j(q)$ are polynomials in $q$.
251:
252: Now let us consider the limit $q \to \infty$. Evidently, in this limit, for a
253: given finite graph $G$, there is a single dominant term in $Z(G,q,v)$, namely
254: $q^n$, arising from $G'$ with $E'=\emptyset$. This term is what one would get
255: in the evaluation of $Z$ if there were no spin-spin interactions, but instead
256: just single-site contributions. This reduction helps to understand the fact
257: that the Potts model is exactly solvable \cite{pg} in the limit $q \to \infty$.
258:
259: Next, let us examine the terms in $Z(G,q,v)$ in more detail. For a given graph
260: $G$, we define
261: %
262: \beq
263: x_G = \frac{v}{q^{n(G)/e(G)}}
264: \label{xg}
265: \eeq
266: %
267: and, in the thermodynamic limit,
268: %
269: \beq
270: x_\Lambda = \frac{v}{q^{2/\kappa_\Lambda}} \ .
271: \label{x}
272: \eeq
273: %
274: Considering $q$ to be large, we focus on the first two and last two of the
275: terms in eq. (\ref{Zexp}); we can write these as
276: %
277: \beqs
278: \lefteqn{Z(G,q,v)} \cr\cr
279: & = & q^{n(G)+1}\Bigl [x_G^{e(G)}\, \{1+e(G)[x_G \, q^{n(G)/e(G)}]^{-1} \}
280: + ... \cr\cr
281: & + & q^{-1} \, \{ e(G) \, x_G \, q^{(n(G)/e(G))-1} + 1 \} \Bigr ] \ .
282: \label{Zexpx}
283: \eeqs
284: %
285: We can approximate $Z(G,q,v)$ by including just the first and last terms if
286: several conditions are met. In the expression multiplying $x_G^{e(G)}$, the
287: second term is negligible compared with the first if $x_G \sim O(1)$ and the
288: condition
289: %
290: \beq
291: q \gg e(G)^{e(G)/(n(G)-1)}
292: \label{qcon1}
293: \eeq
294: %
295: is satisfied. In the expression multiplying $q^{-1}$, the first term is
296: negligible compared with the second if $x_G \sim O(1)$ and the condition
297: %
298: \beq
299: q \gg [e(G)]^{e(G)/(e(G)-n(G)+1)}
300: %
301: \label{qcon2}
302: \eeq
303: %
304: is satisfied. Let us denote the degree of $c_j(q)$ in $q$ as $p_j$. Among the
305: ${e(G) \choose j}$ terms in $c_j(q)$, only a few have this power except
306: for small and large $j$ as discussed above. Now we require that
307: $c_j(q)v^j$ be small compared with $q^n$, so that in terms of the
308: power of $q$,
309: %
310: \beq
311: p_j + j \left ( \frac{n(G)-1}{e(G)} \right ) < n(G)
312: \label{pj1}
313: \eeq
314: %
315: or
316: %
317: \beq
318: p_j < \frac{(e(G)-j)(n(G)-1)}{e(G)} + 1 \ .
319: \label{pj2}
320: \eeq
321: %
322: Further, if these conditions hold and $x_G \sim O(1)$, then terms arising from
323: the third, fourth, and subsequent terms of $Z(G,q,v)$ are small relative to the
324: first, and the third, fourth, and earlier terms counting in from the last, are
325: negligible relative to the last, with the ordering specified by ascending
326: powers of $v$, as in eq. (\ref{Zexp}). Provided that $x_G \sim O(1)$ and the
327: above conditions hold, one can approximate $Z(G,q,v)$ by keeping only the first
328: and last terms in eq. (\ref{Zexp}):
329: %
330: \beq
331: Z(G,q,v) \sim qv^{e(G)} + q^{n(G)} = q^{n(G)+1}[x_G^{e(G)}+q^{-1}] \ .
332: \label{Zapprox}
333: \eeq
334: %
335: The zeros of this approximation to $Z(G,q,v)$ are given by
336: %
337: \beq
338: x_G = (-q^{-1})^{1/e(G)} \ .
339: \label{xgeq}
340: \eeq
341: %
342: Now consider the limit $e(G) \to \infty$ and $n(G) \to
343: \infty$ with the ratio $e(G)/n(G)=\kappa_G/2$ finite. If $q$ also goes to
344: infinity, sufficiently fast to satisfy the conditions (\ref{qcon1}) and
345: (\ref{qcon2}) but more slowly than the exponential $e^{b e(G)}$ with
346: $b$ a real positive constant, then
347: %
348: \beq
349: \lim_{q \to \infty} \lim_{e(G) \to \infty} \, |q|^{1/e(G)} = 1 \ ,
350: \label{qecon}
351: \eeq
352: %
353: and these zeros merge onto the unit circle
354: %
355: \beq
356: |x_G|=1 \ .
357: \label{xg1}
358: \eeq
359: %
360: If $q$ were to grow more rapidly, as $q \sim e^{b e(G)}$ where $b$ is an
361: arbitrary positive real constant, then the zeros would merge onto the circle
362: $|x_G|=e^{-b}$.
363:
364: In particular, one may consider the thermodynamic limit of a graph $G$ which is
365: a section of the regular lattice $\Lambda$. Then the conditions
366: corresponding to eqs. (\ref{qcon1}) and (\ref{qcon2}) are, respectively,
367: %
368: \beq
369: q \gg e(G)^{\kappa_\Lambda/2}
370: \label{qcon1p}
371: \eeq
372: %
373: and
374: %
375: \beq
376: q \gg e(G)^{\kappa_\Lambda/(\kappa_\Lambda-2)} \ .
377: \label{qcon2p}
378: \eeq
379: %
380: The inequality $\kappa_\Lambda/2 \ge \kappa_\Lambda/(\kappa_\Lambda-2)$ is
381: equivalent to the inequality $\kappa_\Lambda \ge 4$. Since $\kappa_\Lambda \ge
382: 4$ for all of the lattices that we consider except for the honeycomb lattice,
383: a consequence is that for these former lattices with $\kappa_\Lambda \ge 4$,
384: condition (\ref{qcon1p}) implies condition (\ref{qcon2p}). Provided that
385: $x_\Lambda \sim O(1)$ and the above conditions hold, so that one can
386: approximate $Z(G,q,v)$ with the first and last terms in eq. (\ref{Zexp}), it
387: follows that in this thermodynamic limit, with $q$ also going to infinity
388: sufficiently fast to satisfy the conditions (\ref{qcon1p}) and (\ref{qcon2p})
389: but more slowly than the exponential form given above, the zeros of $Z(G,q,v)$
390: merge onto the circle
391: %
392: \beq
393: |x_\Lambda|=1
394: \label{xcircle}
395: \eeq
396: %
397: where $x_\Lambda$ is given by eq. (\ref{x}). Our result (\ref{xcircle}) holds
398: for any regular lattice with dimension $d \ge 2$. For example, for the
399: $d$-dimensional Cartestian lattice ${\mathbb E}^d$, $\kappa_{{\mathbb
400: E}^d}=2d$, while for the $d$-dimensional body-centered cubic lattice,
401: $\kappa_{bcc^d}=2^d$.
402:
403:
404: Since the free energy is nonanalytic at $x_\Lambda=1$, eq.
405: (\ref{xcircle}) yields the asymptotic relations
406: %
407: \beq
408: v_{c,\Lambda} \sim q^{2/\kappa_\Lambda} \quad {\rm as} \ \ q \to \infty
409: \label{vc}
410: \eeq
411: %
412: and hence
413: %
414: \beq
415: K_{c,\Lambda} \sim \frac{2}{\kappa_\Lambda} \ln q \quad {\rm as} \ \
416: q \to \infty
417: \label{kc}
418: \eeq
419: %
420: for the values of $v$ and $K$ where the Potts model on the lattice $\Lambda$
421: has a phase transition from a paramagnetic (PM) high-temperature phase to a
422: ferromagnetic (FM) low-temperature phase in the limit of large
423: $q$. We note that this agrees with the large-$q$ limit of the mean-field
424: theory result \cite{mft}
425: %
426: \beq
427: K_{c,MFT} = \frac{2(q-1)}{\kappa_\Lambda(q-2)} \ \ln (q-1) \ ,
428: \label{kc_mft}
429: \eeq
430: %
431: viz.,
432: %
433: \beqs
434: K_{c,MFT} & \sim & \frac{2}{\kappa_\Lambda} \Bigl [
435: \{ 1 + q^{-1} + O(q^{-2}) \} \ln q \cr\cr
436: & - & \{ q^{-1} + O(q^{-2}) \} \Bigr ]
437: \label{kc_mftseries}
438: \eeqs
439: %
440: as $q \to \infty$. In eqs. (\ref{kc}) and (\ref{kc_mft}), the property that
441: $K_c$ increases with increasing $q$ can be understood as a consequence of the
442: fact that as $q$ gets large, each spin has more possible values (is
443: ``floppier''), and hence one must go to a lower temperature for the
444: ferromagnetic long-range order to occur. The feature that $K_c$ increases
445: asymptotically like $\ln q$ as $q \to \infty$ can be understood since an
446: order-disorder transition involves a balance between minimizing the
447: configurational energy and maximizing the entropy terms in the free energy per
448: site, $F=U-TS$, and in this limit, the entropy per site is $S \to k_B \ln q$.
449:
450: Next, we recall the known exact equations for the PM-FM phase transition on the
451: 2D lattices \cite{wurev,baxterbook}, which we write in a convenient manner for
452: the discussion of the large-$q$ limit,
453: %
454: \beqs
455: \frac{q}{v^2}=1 & & \qquad {\rm for} \ \ \Lambda=sq \cr\cr
456: \frac{q}{v^3}=1+\frac{3}{v} & & \qquad {\rm for} \ \ \Lambda=tri \cr\cr
457: \frac{q^2}{v^3}=1-\frac{3q}{v^2} & & \qquad {\rm for} \ \ \Lambda=hc \ ,
458: \eeqs
459: %
460: where $sq$, $tri$, and $hc$ denote the square, triangular, and honeycomb
461: lattices. As is evident, in the limit of large $q$ and $v$
462: such that $q=v^2$ for the square lattice, $q=v^3$ for the triangular
463: lattice and $q=v^{3/2}$ for the honeycomb lattice, these equations are all in
464: agreement with the general relation (\ref{vc}), which also holds for
465: higher-dimensional lattices.
466:
467:
468: \section{Partition Function Zeros for Finite Lattice Sections}
469: \label{sectionIII}
470:
471: \subsection{General Structure}
472:
473: It is also of interest to consider zeros of $Z(G,q,v)$ on finite lattice graphs
474: in the limit of large $q$. In general, if the conditions specified in the
475: previous section are satisfied, one can approximate $Z(G,q,v)$ by keeping the
476: first and last terms in eq. (\ref{Zexp}). We have studied these zeros in
477: detail for sections of various 2D lattices and have found that a sufficient
478: criterion that these conditions can be satisfied for these and
479: higher-dimensional lattices is that one uses fully periodic boundary
480: conditions. Accordingly, when $q$ is large, the zeros in the $x_{\Lambda}$
481: plane are located close to a circle. For finite lattice graphs with fully
482: periodic boundary conditions, $e(G)/n(G)=\kappa_\Lambda/2$ so
483: that the two definitions in eqs.(\ref{xg}) and (\ref{x}) are
484: equivalent. However, for finite graphs with other boundary conditions, $x_G$
485: varies from one graph to another. Therefore, we will plot zeros in the
486: $x_\Lambda$ plane for the lattice $\Lambda$ with different boundary
487: conditions. We first rewrite the approximate expression (\ref{Zapprox}) in an
488: equivalent form that will be convenient for our discussion; when $q$ is large,
489: %
490: \beqs
491: & & Z(G,q,v) \sim qv^{e(G)} + q^{n(G)} \cr\cr
492: & = & q^{(2e(G)/\kappa_\Lambda)+1}
493: [x_\Lambda^{e(G)}+q^{n(G)-(2e(G)/\kappa_\Lambda)-1}]
494: \label{Zapprox2}
495: \eeqs
496: %
497: The zeros of this approximation to $Z(G,q,v)$ are located
498: on a circle in the $x_\Lambda$ plane with radius
499: %
500: \beq
501: r(\Lambda,BC) = q^{p(\Lambda,BC)}
502: \label{rcircle}
503: \eeq
504: %
505: where
506: %
507: \beq
508: p(\Lambda,BC)=\frac{n(G)-1}{e(G)}-\frac{2}{\kappa_\Lambda} \ .
509: \label{ppower}
510: \eeq
511: %
512: (This power $p$ should not be confused with the powers $p_j$ in
513: eqs. (\ref{pj1}), (\ref{pj2}). The radius $r(\Lambda,BC)$ is less (greater)
514: than unity if $p(\Lambda,BC)$ is negative (positive). For lattices with
515: periodic boundary conditions in all directions, $n(G)/e(G)=2/\kappa_\Lambda$
516: and hence
517: %
518: \beq
519: p(\Lambda,PBC)=-\frac{1}{e(G)}
520: \label{ppbc}
521: \eeq
522: %
523: as in eq. (\ref{xgeq}).
524:
525:
526: In our explicit calculations of zeros for various lattices, we have found that
527: for large but finite $q$, the approximating circles near which these zeros lie
528: may have centers that are slightly shifted to the left or right of the origin
529: of the $x_\Lambda$ plane and move in toward the origin as $q \to \infty$.
530: The following analysis provides an understanding of these shifts. As
531: discussed in Section \ref{sectionII}, there are $\delta(G)$ terms in the
532: partition function given by ${e(G) \choose j}qv^{e(G)-j}$ and there are $g(G)$
533: terms given by ${e(G) \choose j}q^{n(G)-j}v^j$. For the finite lattice graphs
534: with large vertex degree and small girth, a specific approximation for the
535: partition function $Z(G,q,v)$ when $q$ is large is
536: %
537: \beq
538: Z(G,q,v) \sim q(v+1)^{e(G)}+q^{n(G)}
539: \label{zlargedegree}
540: \eeq
541: %
542: so that the center of the circle, denoted by $c(\Lambda,BC)$, is
543: not at the origin but at
544: %
545: \beq
546: c(\Lambda,BC)=-\frac{1}{q^{2/\kappa_\Lambda}} \ .
547: \label{c_left}
548: \eeq
549: %
550: Of course, this approaches zero as $q \to \infty$, in agreement with our
551: result above, eq. (\ref{xcircle}). The triangular lattice is an example
552: for which the center of the circle is shifted to the left, as in
553: eq. (\ref{c_left}) and will be discussed below.
554:
555: On the other hand, for finite lattice graphs with small vertex degree and
556: large girth, a specific approximation for $Z(G,q,v)$ when $q$ is large is
557: %
558:
559: %
560: \beqs
561: & & Z(G,q,v) \sim qv^{e(G)}+q^{n(G)-e(G)}(q+v)^{e(G)} \cr\cr
562: & = & q^{n(G)+1}[x_G^{e(G)}+q^{-1}(1+q^{(n(G)/e(G))-1}x_G)^{e(G)}] \cr\cr
563: & &
564: \label{zlargegirth}
565: \eeqs
566: %
567: The circle near which the zeros lie is now shifted to the right. Let
568: us first consider the case with fully periodic boundary conditions so that
569: $x_G = x_\Lambda$. In the complex $x_\Lambda$
570: plane, the circle has radius
571: %
572: \beq
573: r(\Lambda,PBC)=\frac{q^{1/e(G)}}{q^{2/e(G)}-q^{(2n(G)/e(G))-2}}
574: \label{rlargegirth}
575: \eeq
576: %
577: and center at
578: %
579: \beq
580: c(\Lambda,PBC)=\frac{q^{n/e(G)-1}}{q^{2/e(G)}-q^{(2n(G)/e(G))-2}} \ .
581: \label{c_right}
582: \eeq
583: %
584: For the lattices with $e(G)$ larger than $n(G)$, the term $q^{(2n(G)/e(G))-2}$ in the
585: denominator is negligible when $q$ is large, and the radius of the circle is
586: approximately $q^{-1/e(G)}$, as in eq. (\ref{rcircle}). For large $q$, the
587: center of the circle in the $x_\Lambda$ plane is
588: %
589: \beq
590: c(\Lambda,PBC)=q^{-s(\Lambda,BC)}
591: \eeq
592: %
593: where
594: %
595: \beq
596: s(\Lambda,PBC)=1+\frac{2-n(G)}{e(G)} \ .
597: \label{cqpowerpbc}
598: \eeq
599: %
600: When $e(G)$ is large, so that $2/e(G) \ll 1$, the position of center can be
601: further approximated as $c(\Lambda,PBC) \simeq q^{(2/\kappa_\Lambda)-1}$,
602: which approaches the origin as $q \to \infty$.
603:
604: For the lattice sections with other boundary conditions, a specific
605: approximation for $Z(G,q,v)$ for large $q$ is
606: %
607: \beqs
608: & & Z(G,q,v) \sim q^{(2e(G)/\kappa_\Lambda)+1} \times \cr\cr
609: & & [x_\Lambda^{e(G)}+
610: q^{n(G)-(2e(G)/\kappa_\Lambda)-1}(1+q^{(2/\kappa_\Lambda)-1}x_\Lambda)^{e(G)}]
611: \cr\cr
612: & &
613: \eeqs
614: %
615: In the $x_\Lambda$ plane, the resultant approximating circle has
616: radius
617: %
618: \beq
619: r(\Lambda,BC)=\frac{q^{1/e(G)-(n(G)/e(G))+2/\kappa_\Lambda}}
620: {q^{2/e(G)-(2n(G)/e(G))+(4/\kappa_\Lambda)}-q^{(4/\kappa_\Lambda)-2}}
621: \eeq
622: %
623: and center
624: %
625: \beq
626: c(\Lambda,BC)=\frac{q^{(2/\kappa_\Lambda)-1}}
627: {q^{2/e(G)-(2n(G)/e(G))+(4/\kappa_\Lambda)}-q^{(4/\kappa_\Lambda)-2}}
628: \ .
629: \eeq
630: %
631: For lattices with $\kappa_\Lambda>2$, $q^{4/\kappa_\Lambda-2}$ in the
632: denominator is negligible when $q$ is large, and the radius of the circle is
633: the same as eqs. (\ref{rcircle}) and (\ref{ppower}). We also have
634: %
635: \beq
636: s(\Lambda,BC)=1+\frac{2-2n(G)}{e(G)}+\frac{2}{\kappa_\Lambda}
637: \label{cqpower}
638: \eeq
639: %
640: for large $q$. The honeycomb lattice is an example for which the circle near
641: which the zeros lie shifts to the right of the origin and will be discussed
642: below. Another example is the square lattice with next-nearest-neighbor
643: interactions that have the same strength as the nearest-neighbor interactions,
644: which we have also analyzed. From the above discussion, we expect that the
645: center of the circle is the origin for lattice sections with vertex degree
646: roughly equal to girth.
647:
648:
649: \subsection{Square Lattice Sections}
650:
651: For the square lattice with toroidal (tor) boundary conditions,
652: $e(sq,tor)=2L_xL_y$, $\kappa_{sq} = 4$, and
653: %
654: \beq
655: p(sq,tor)=-\frac{1}{2L_xL_y} \ .
656: \label{psqtor}
657: \eeq
658: %
659: Hence the radius of the approximating circle increases as the area of the
660: lattice increases, and approaches unity from below in the thermodynamic
661: limit. The zeros of the partition function for the Potts model on the toroidal
662: strip of the square lattice with $L_y=4$ and $L_x=9$ in the $x_{sq}$ plane when
663: $q=1000$ are shown in Fig. \ref{sqq1000}(a). The radius of the circle is about
664: 0.9085. The zeros lie very close to this circle for the above value of $q$, and
665: we find that they get closer to the circle given by eqs. (\ref{rcircle}) and
666: (\ref{ppower}) when $q$ increases.
667:
668:
669: \begin{figure}[hbtp]
670: \epsfxsize=1.5in
671: \epsffile{sqtory4x9q1000.eps}
672: (a) \\
673: \epsfxsize=1.5in
674: \epsffile{sqcycy4x9q1000.eps}
675: (b) \\
676: \epsfxsize=1.5in
677: \epsffile{sqcyly4x9q1000.eps}
678: (c) \\
679: \epsfxsize=1.5in
680: \epsffile{sqfreey4x9q1000.eps}
681: (d) \\
682: %\vspace*{-2cm}
683: \caption{\footnotesize{Zeros of the Potts model partition function, plotted in
684: the $x_{sq}$ plane, for a section of the square lattice with $L_y=4$ and
685: $L_x=9$, when $q=1000$, for the following boundary conditions: (a) toroidal;
686: (b) cyclic; (c) cylindrical; (d) free. For comparison, the unit circle and
687: circles where the zeros cluster are also shown.}}
688: \label{sqq1000}
689: \end{figure}
690:
691:
692: For the square lattice with cyclic boundary conditions we have
693: $e(sq,cyc)=L_x(2L_y-1)$. By eq.(\ref{ppower}),
694: %
695: \beq
696: p(sq,cyc)=\frac{L_x-2}{2L_x(2L_y-1)} \ .
697: \eeq
698: %
699: In the thermodynamic limit the radius of the approximating circle decreases and
700: approaches unity from above. Note that if one were to keep $L_y$ fixed and
701: take $L_x \to \infty$, thereby violating the premises of our analysis, the
702: radius of this approximating circle would be $q^{1/2(2L_y-1)}$, which would
703: diverge as $q \to \infty$. The zeros of the cyclic strip with $L_y=4$ and
704: $L_x=9$ in the $x_{sq}$ plane when $q=1000$ are shown in
705: Fig. \ref{sqq1000}(b). The radius of the approximating circle is about
706: 1.47. The zeros are not as close to the circle as for the corresponding
707: toroidal case. On the other hand, if one keeps $L_x$ fixed and takes $L_y \to
708: \infty$, corresponding to cylindrical boundary conditions, the radius of the
709: approximating circle approaches unity from above. The zeros of the cylindrical
710: strip with $L_y=4$ and $L_x=9$ in the $x_{sq}$ plane when $q=1000$ are shown in
711: Fig. \ref{sqq1000}(c). The radius of this approximating circle is about
712: 1.11. The zeros are closer to the circle than the cyclic strip with the same
713: $L_y$ and $L_x$ because the number of boundary vertices (i.e. vertices with
714: coordination number three) is reduced.
715:
716:
717:
718: For the square lattice with free boundary conditions,
719: $e(sq,free)=2L_xL_y-L_x-L_y$. By eq.(\ref{ppower}),
720: %
721: \beq
722: p(sq,free)=\frac{L_x+L_y-2}{2(2L_xL_y-L_x-L_y)} \ .
723: \eeq
724: %
725: In the thermodynamic limit the radius of the circle decreases and approaches
726: unity from above. Again, however, if one were to keep $L_y$ fixed and take $L_x
727: \to \infty$, the radius of the approximating circle would be $q^{1/2(2L_y-1)}$,
728: (the same as the cyclic case), which would diverge as $q \to \infty$. Since for
729: a given lattice size the number of boundary vertices is larger than for other
730: boundary conditions, and especially since there are four vertices with
731: coordination number equal to two, it is natural to expect that the zeros will
732: lie farther from the asymptotic circle than for these other boundary
733: conditions. The zeros of the free strip with $L_y=4$ and $L_x=9$ in the
734: $x_{sq}$ plane when $q=1000$ are shown in Fig. \ref{sqq1000}(d). The radius of
735: the approximating circle is about 1.90.
736:
737:
738: \subsection{Triangular lattice}
739:
740:
741: For the triangular lattice with toroidal boundary conditions,
742: $e(tri,tor)=3L_xL_y$ and $\kappa(tri,tor)=6$. Hence, for a finite section of
743: the triangular lattice with toroidal boundary conditions, as for the infinite
744: triangular lattice,
745: %
746: \beq
747: x_{tri}=\frac{v}{q^{1/3}}
748: \eeq
749: %
750: and
751: %
752: \beq
753: p(tri,tor)=-\frac{1}{3L_xL_y} \ ,
754: \eeq
755: %
756: as given by eq.(\ref{xgeq}). This radius increases as the size of the lattice
757: increases, and approaches unity from below in the thermodynamic limit. There is
758: a noticeable shift of the circle to the left. The center of the approximating
759: circle is at $x_{tri}=-q^{-1/3}$, approaching zero as $q \to
760: \infty$. The zeros of the toroidal strip with $L_y=3$ and $L_x=9$ in the
761: $x_{tri}$ plane when $q=1000$ are shown in Fig. \ref{triq1000}(a). The radius
762: of this approximating circle is about 0.918.
763:
764: \begin{figure}[hbtp]
765: \epsfxsize=1.5in
766: \epsffile{tritory3x9q1000.eps}
767: (a) \\
768: \epsfxsize=1.5in
769: \epsffile{tricycy4x9q1000.eps}
770: (b) \\
771: \epsfxsize=1.5in
772: \epsffile{tricyly4x9q1000.eps}
773: (c) \\
774: \epsfxsize=1.5in
775: \epsffile{trifreey4x9q1000.eps}
776: (d) \\
777: %\vspace*{-2cm}
778: \caption{\footnotesize{Zeros of the Potts model partition function in the
779: $x_{tri}$ plane for the triangular lattice when $q=1000$, for the following
780: boundary conditions and sizes: (a) toroidal with $L_y=3$ and $L_x=9$; (b)
781: cyclic with $L_y=4$ and $L_x=9$; (c) cylindrical with $L_y=4$ and $L_x=9$; (d)
782: free with $L_y=4$ and $L_x=9$. For comparison, the unit circle and
783: approximating circles are also shown.}}
784: \label{triq1000}
785: \end{figure}
786:
787: For a section of the triangular lattice with cyclic boundary conditions,
788: $e(tri,cyc)=L_x(3L_y-2)$. When $q$ is large, we calculate
789: %
790: \beq
791: p(tri,cyc)=\frac{2L_x-3}{3L_x(3L_y-2)}
792: \eeq
793: %
794: for the approximating circle. In the thermodynamic limit, the radius of this
795: circle decreases, approaching unity from above. However, if one were to keep
796: $L_y$ fixed and take $L_x \to \infty$, then the radius would be
797: $q^{2/3(3L_y-2)}$, which would diverge as $q \to \infty$. The zeros of the
798: cyclic strip with $L_y=4$ and $L_x=9$ in the $x_{tri}$ plane when $q=1000$ are
799: shown in Fig. \ref{triq1000}(b). The radius of this approximating circle is
800: about 1.47 as for the corresponding case of the square lattice.
801: If one keeps $L_x$ fixed and takes $L_y \to \infty$, corresponding to
802: cylindrical boundary conditions, the radius of the approximating circle
803: approaches unity from above. The zeros of the cylindrical strip with $L_y=4$
804: and $L_x=9$ in the $x_{tri}$ plane when $q=1000$ are shown in
805: Fig. \ref{triq1000}(c). The radius of this approximating circle is about 1.12.
806:
807: In general, the approximation in eq.(\ref{Zapprox}) is not valid for the
808: triangular lattice with free boundary conditions. The reason is that there are
809: two vertices with degree two, so that the coefficient of $v^{e(tri,free)-2}$
810: contains the term $2q^2$ and the coefficient of $v^{e(tri,free)-4}$ contains
811: the term $q^3$ for $L_x, L_y > 2$. Therefore, when $q$ is large there are two
812: pair of roots around $v \sim \pm i \sqrt{q}$, or equivalently around $\pm
813: i q^{1/6}$ in the $x_{tri}$ plane. We have $e(tri,free)=3L_xL_y-2L_x-2L_y+1$.
814: The partition function zeros for the free strip of the triangular lattice with
815: $L_y=4$ and $L_x=9$ when $q=1000$ are shown in
816: Fig. \ref{triq1000}(d). Although the approximation (\ref{Zapprox}) does not
817: apply here, one can see that many of these zeros lie close to an
818: approximating circle with radius 1.84, which is similar to what one would get
819: by formally using $p(tri,free)=2(L_x+L_y-2)/[3(3L_xL_y-2L_x-2L_y+1)]$.
820: We observe that there are two almost overlapping complex-conjugate pairs
821: of zeros which have relatively large magnitude and are close to the imaginary
822: axis.
823:
824:
825: \subsection{Honeycomb lattice}
826:
827: For the honeycomb lattice with toroidal boundary conditions where both $L_x$
828: and $L_y$ are even, $e(hc,tor)=3L_xL_y/2$ so that $n(hc,tor)/e(hc,tor)=2/3$ for
829: any $L_x$ and $L_y$. Thus,
830: %
831: \beq
832: x_{hc}=\frac{v}{q^{2/3}}
833: \eeq
834: %
835: and
836: %
837: \beq
838: p(hc,tor)=-\frac{2}{3L_xL_y} \ ,
839: \eeq
840: %
841: as given in eq.(\ref{xgeq}). This radius increases as the size of the lattice
842: increases, and approaches unity from below in the thermodynamic limit. There is
843: a shift of the circle to the right. By eq.(\ref{cqpowerpbc}),
844: %
845: \beq
846: s(hc,tor)=\frac{1}{3} + \frac{4}{3L_xL_y}
847: \eeq
848: %
849: which approaches $1/3$ as the size of the lattice section becomes large. The
850: zeros of the toroidal strip with $L_y=4$ and $m=9 (L_x=18)$ in the $x_{hc}$
851: plane when $q=1000$ are shown in Fig. \ref{hcq1000}(a). The radius of this
852: approximating circle is about 0.938.
853:
854:
855: \begin{figure}[hbtp]
856: \epsfxsize=1.5in
857: \epsffile{hctory4x9q1000.eps}
858: (a) \\
859: \epsfxsize=1.5in
860: \epsffile{hccycy4x9q1000.eps}
861: (b) \\
862: \epsfxsize=1.5in
863: \epsffile{hccyly4x9q1000.eps}
864: (c) \\
865: \epsfxsize=1.5in
866: \epsffile{hcfreey4x9q1000.eps}
867: (d) \\
868: %\vspace*{-2cm}
869: \caption{\footnotesize{Zeros of the Potts model partition function in the
870: $x_{hc}$ plane for the honeycomb lattice when $q=1000$, for the following
871: boundary conditions and sizes: (a) toroidal with $L_y=4$ and $L_x=18$; (b)
872: cyclic with $L_y=4$ and $L_x=18$; (c) cylindrical with $L_y=4$ and $L_x=9$; (d) free boundary conditions with $L_y=4$ and $L_x=9$.
873: For comparison, the unit circle and approximating circles are also shown.}}
874: \label{hcq1000}
875: \end{figure}
876:
877: For a section of the honeycomb lattice with cyclic boundry conditions we have
878: $e(hc,cyc)=L_x(3L_y-1)/2$, where $L_x$ must be even. By eq.(\ref{ppower}),
879: %
880: \beq
881: p(hc,cyc)=\frac{2(L_x-3)}{3L_x(3L_y-1)} \ .
882: \eeq
883: %
884: In the thermodynamic limit, the radius of the approximating circle
885: decreases, approaching unity from above. For the shift of the circle, by
886: eq. (\ref{cqpower}),
887: %
888: \beq
889: s(hc,cyc)=\frac{1}{3} + \frac{4(3-L_x)}{3L_x(3L_y-1)}
890: \eeq
891: %
892: The zeros of the cyclic strip with $L_y=4$ and $m=9 (L_x=18)$ in the $x_{hc}$
893: plane when $q=1000$ are shown in Fig. \ref{hcq1000}(b). The radius of this
894: approximating circle is about 1.42.
895:
896: Since we represent the honeycomb lattice in the form of bricks oriented
897: horizontally, the results for cylindrical boundary conditions cannot be
898: obtained from cyclic boundary conditions by simply switching $L_x$ and
899: $L_y$. We have $e(hc,cyl)=L_y(3L_x/2-1)$ for the honeycomb lattice with
900: cylindrical boundary conditions where $L_y$ must be even. When $q$ is large, we
901: calculate
902: %
903: \beq
904: p(hc,cyl)=\frac{2(2L_y-3)}{3L_y(3L_x-2)} \ .
905: \eeq
906: %
907: and
908: %
909: \beq
910: s(hc,cyl)=\frac{1}{3} + \frac{4(3-2L_y)}{L_y(3L_x-2)} \ .
911: \eeq
912: %
913: The zeros of the cylindrical strip with $L_y=4$ and $L_x=9$ in the $x_{hc}$
914: plane when $q=1000$ are shown in Fig. \ref{hcq1000}(c). The radius of this
915: approximating circle is about 1.26. We have also considered other boundary
916: conditions, as shown in Fig. \ref{hcq1000}.
917:
918: \subsection{Kagom\'e lattice}
919:
920: As an example of a heteropolygonal two-dimensional lattice (an Archimedean
921: lattice comprised of more than one type of regular polygon), we consider the
922: kagom\'e lattice. For a section of this lattice with toroidal boundary
923: conditions, $n(kag,tor)=3L_xL_y$ and $e(kag,tor)=6L_xL_y$ so that
924: $n(kag,tor)/e(kag,tor)=1/2$ as for the square lattice. Hence,
925: %
926: \beq
927: x_{kag}=\frac{v}{\sqrt{q}}
928: \eeq
929: %
930: and
931: \beq
932: p(kag,tor)=-\frac{1}{6L_xL_y} \ .
933: \eeq
934: %
935: The radius of the approximating circle increases as the size of the lattice
936: increases and approaches unity from below in the thermodynamic limit. The
937: zeros of the toroidal strip with $L_y=2$ and $L_x=9$ in the $x_{kag}$ plane
938: when $q=1000$ are shown in Fig. \ref{kagq1000}(a). The radius of this
939: approximating circle is about 0.938. We have also considered other boundary
940: conditions as shown in Fig. \ref{kagq1000}.
941:
942:
943: \begin{figure}[hbtp]
944: \epsfxsize=1.5in
945: \epsffile{kagtory2x9q1000.eps}
946: (a) \\
947: \epsfxsize=1.5in
948: \epsffile{kagcycy3x9q1000.eps}
949: (b) \\
950: \epsfxsize=1.5in
951: \epsffile{kagcyly2x9q1000.eps}
952: (c) \\
953: \epsfxsize=1.5in
954: \epsffile{kagfreey3x9q1000.eps}
955: (d) \\
956: %\vspace*{-2cm}
957: \caption{\footnotesize{Zeros in the $x_{kag}$ plane for Potts model partition
958: function on the Kagom\'e lattice when $q=1000$, for the following boundary
959: conditions and sizes: (a) toroidal with $L_y=2$ and $L_x=9$; (b) cyclic with
960: $L_y=3$ and $L_x=9$; (c) cylindrical with $L_y=2$ and $L_x=9$; (d) free with
961: $L_y=3$ and $L_x=9$. For comparison, the unit circle and approximating circles
962: are also shown.}}
963: \label{kagq1000}
964: \end{figure}
965:
966: \section{Summary}
967:
968: In summary, we have presented some new results on the distribution of
969: complex-temperature zeros of the partition function of the $q$-state Potts
970: model for $q \to \infty$. Generalizing previous work of Wu and coworkers for
971: Cartesian lattices, we have shown that with an appropriate definition of the
972: thermodynamic limit and the limit $q \to \infty$ on an arbitrary regular
973: lattice $\Lambda$ of dimensionality $d \ge 2$, the zeros lie on the circle
974: $|x_\Lambda|=1$. We have also studied the distribution of zeros for finite
975: sections of various two-dimensional lattices for large $q$, showing how they
976: also lie approximately on circles.
977:
978: \bigskip
979:
980: Acknowledgments: The research of R.S. was partially
981: supported by the NSF grant PHY-00-98527. The research of S.C.C. was partially
982: supported by the NSC grant NSC-93-2119-M-006-009, NSC-94-2112-M-006-013 and NSC-94-2119-M-002-001.
983:
984: \newpage
985:
986: \section{Appendix I: 1D Lattice and Quasi-1D Lattice Strips}
987:
988: Our derivation of eq. (\ref{xcircle}) assumed that the lattice graph $G$ has
989: the property that removing a few edges does not cut it into disconnected parts.
990: This property does not hold for a 1D lattice or quasi-1D lattice strips of
991: sufficiently small width. Here we comment briefly on this. We consider first a
992: one-dimensional lattice (line graph) with free or periodic boundary conditions,
993: labelled FBC and PBC, respectively. We denote such graphs with $n$ vertices as
994: $L_{n,F}$ and $L_{n,P}$. The coordination number $\kappa=2$ for $L_{n,P}$, and
995: this is the effective value of the coordination number also for $L_{n,F}$ when
996: $n \to \infty$, so that in both cases, $x_{1D}=v/q$. From
997: elementary calculations one has the exact results
998: %
999: \beq
1000: Z(L_{n,F},q,v) = q^n(1+x_{1D})^{n-1}
1001: \label{ZL1f}
1002: \eeq
1003: %
1004: and
1005: \beq
1006: Z(L_{n,P},q,v) = q^n[(1+x_{1D})^n + (q-1)x_{1D}^n]
1007: \label{ZL1p}
1008: \eeq
1009: %
1010: As in the text, we focus on large $n$ and large $q$. Aside from the
1011: zero at $q=0$, which is not relevant in this case, the zeros of
1012: $Z(L_{n,F},q,v)$ occur at the single point
1013: %
1014: \beq
1015: x_{1D}=-1 \quad (FBC) \ .
1016: \label{zeros_1d_fbc}
1017: \eeq
1018: %
1019: Note that the number of edges for this graph is $e(L_{n,F})=n-1$. Because the
1020: degree in $q$ of the coefficient $c_j(q)$ defined in eq. (\ref{Zexp}) is
1021: $p_j=n-j$, all terms have the same order and none can be neglected.
1022:
1023: In the limit $n \to \infty$, followed by $q \to \infty$, the zeros of
1024: $Z(L_{n,P},q,v)$ are given by the solution to the equation
1025: $|1+x_{1D}|=|x_{1D}|$, which is the vertical line $Re(x_{1D})=-1/2$ in the
1026: $x_{1D}$ plane, i.e.,
1027: %
1028: \beq
1029: x_{1D} = -\frac{1}{2} + iy \quad (PBC) \ .
1030: \label{zeros_1d_pbc}
1031: \eeq
1032: %
1033: where $-\infty \le y \le \infty$. As these exact results show, the 1D case is
1034: different from the thermodynamic limit of lattices with dimensionality $d \ge
1035: 2$ in that the locus of zeros is strongly dependent upon the choice of boundary
1036: conditions and, moreover, is not the circle $|x_\Lambda|=1$.
1037:
1038: In contrast, for strips of the square lattice with various types of boundary
1039: conditions that maintain the duality of the infinite square lattice, we have
1040: shown via exact solutions in Ref. \cite{sdg} that as $q \to \infty$, the zeros
1041: do lie exactly on the unit circle $|x_\Lambda|=1$. For these strips we also
1042: found that for large but finite $q$, the locus ${\cal B}$ consists of the union
1043: of (i) a self-conjugate portion of this unit circle given by
1044: $x_{sq}=e^{i\theta}$ with $\theta_0 \le \theta \le \pi$ and $-\pi \le \theta
1045: \le -\theta_0$ (where the angle $\theta_0$ depends on the strip) with (ii) a
1046: line segment $-\rho \le x_{sq} \le -1/\rho$ on the negative real axis, where
1047: $\rho > 1$ is a positive real constant (see the plots given in
1048: Ref. \cite{sdg}). The arc of the unit circle and the line segment intersect at
1049: $x_{sq}=-1$. As $q \to \infty$, $\rho \to 1$ and $\theta_0 \to 0$. Thus, for
1050: large but finite $q$, ${\cal B}$ is almost the unit circle except for a small
1051: line segment centered at $x_{sq}=-1$ and a gap in the circle in the vicinity of
1052: $x_{sq}=1$. As we discussed in Ref. \cite{sdg}, these deviations can be
1053: understood from the fact that the large-$q$ expansion of the dominant
1054: eigenvalues of the transfer matrix contain poles at $x_{sq}=\pm 1$, showing
1055: that regardless of how large $q$ is, the large-$q$ expansion breaks down at
1056: these two points.
1057:
1058: More generally, we have carried out similar large-$q$ expansions for the
1059: dominant eigenvalues of the transfer matrix for quasi-one-dimensional strips of
1060: several lattices with different boundary conditions. We find that for the
1061: strips we have considered, these expansions exhibit singularities on the unit
1062: circle $x_\Lambda=1$ for strips with periodic transverse boundary conditions
1063: but not for strips with free transverse boundary conditions. For example, we
1064: find that this expansion for the $L_y=2$ cylindrical strip of the square
1065: lattice has branch-point singularities at $x_{sq}=\pm e^{i\pi/6}$ and
1066: $x_{sq}=\pm e^{-i\pi/6}$. Similarly, for the $L_y=2$ cylindrical strip of the
1067: triangular lattice we find that the large-$q$ expansion of the dominant
1068: eigenvalues has branch-point singularities at $x_{tri}=\pm 1$, $x_{tri}=\pm
1069: e^{i \pi/3}$, and $x_{tri}= \pm e^{-i\pi/3}$.
1070:
1071: These deviations are characteristic of certain quasi-one-dimensional strips,
1072: which do not satisfy the premises of our general analytic results in Section
1073: II. We do not find such deviations for the thermodynamic limits of sections of
1074: regular lattices of dimensionality $d \ge 2$ in the limit of large $q$.
1075:
1076:
1077:
1078: \begin{thebibliography}{99}
1079:
1080: \bibitem{wurev}
1081: F. Y. Wu, Rev. Mod. Phys. {\bf 54} (1982) 235.
1082:
1083: \bibitem{baxterbook}
1084: R. J. Baxter, {\it Exactly Solved Models} (Academic Press, New York, 1982).
1085:
1086: \bibitem{pg}
1087: P. Pearce and R. B. Griffiths, J. Math. Phys. A {\bf 13}, 2143 (1980).
1088:
1089: \bibitem{gag}
1090: J. Glazier, M. Anderson, and G. Grest, Philos. Mag. B {\bf 62}, 615 (1990);
1091: Y. Jiang and J. Glazier, Philos. Mag. Lett. {\bf 74}, 119 (1996).
1092:
1093: \bibitem{kf}
1094: P. W. Kasteleyn and C. M. Fortuin, J. Phys. Soc. Jpn. {\bf 26} (Suppl.), 11
1095: (1969); C. M. Fortuin and P. W. Kasteleyn, Physica {\bf 57}, 536 (1972).
1096:
1097: \bibitem{fisher}
1098: M. E. Fisher, {\it Lectures in Theoretical Physics}
1099: (Univ. of Colorado Press, 1965), vol. 7C, p. 1.
1100:
1101: \bibitem{abe}
1102: R. Abe, T. Dotera, and T. Ogawa, Prog. Theor. Phys. {\bf 85}, 509 (1991).
1103:
1104: \bibitem{cmo}
1105: % Complex-Temperature Properties of the 2D Ising Model on Heteropolygonal
1106: % Lattices
1107: V. Matveev and R. Shrock, J. Phys. A {\bf 28}, 5235 (1995).
1108:
1109: \bibitem{mm}
1110: P. Martin and J. M. Maillard, J. Phys. A {\bf 19}, L547 (1986).
1111:
1112: \bibitem{mbook}
1113: P. P. Martin, {\it Potts Models and Related Problems in
1114: Statistical Mechanics} (World Scientific, Singapore, 1991).
1115:
1116: \bibitem{chw}
1117: %Partition function zeros of the square lattice Potts model
1118: C. N. Chen, C. K. Hu, and F. Y. Wu, Phys. Rev. Lett. {\bf 76}, 169 (1996).
1119:
1120: \bibitem{wuetal}
1121: % Directed compact lattice animals, restricted partitions of numbers, and the
1122: % infinite-state Potts model
1123: F. Y. Wu, G. Rollet, H. Y. Huang, J. M. Maillard, C. K. Hu,
1124: and C. N. Chen, Phys. Rev. Lett. {\bf 76}, 173 (1996).
1125:
1126: \bibitem{pfef}
1127: % Complex-Temperature Singularities in Potts Models on the Square Lattice
1128: V. Matveev and R. Shrock, Phys. Rev. {\bf E54}, (1996) 6174.
1129:
1130: \bibitem{p}
1131: % Complex-temperature partition function zeros of the Potts model on the
1132: % honeycomb and kagome lattices
1133: H. Feldmann, R. Shrock, and S.-H. Tsai, Phys. Rev. E {\bf 57}, 1335 (1998),
1134:
1135: \bibitem{p2}
1136: % Study of the Potts model on the honeycomb and triangular lattices:
1137: % low-temperature series and partition function zeros
1138: H. Feldmann, A. J. Guttmann, I. Jensen, R. Shrock, and S.-H. Tsai, J. Phys. A
1139: {\bf 31}, 2287 (1998).
1140:
1141: \bibitem{ks}
1142: H. Kluepfel and R. Shrock, unpublished; H. Kluepfel, Stony Brook
1143: thesis ``The $q$-State Potts Model: Partition Functions and Their Zeros in the
1144: Complex Temperature and $q$ Planes'' (July, 1999).
1145:
1146: \bibitem{kc}
1147: S.-Y. Kim and R. Creswick, Phys. Rev. {\bf E63}, 066107 (2001).
1148:
1149: \bibitem{a}
1150: % Exact Potts Model Partition Functions for Ladder Graphs
1151: R. Shrock, Physica A {\bf 283}, 388 (2000).
1152:
1153: \bibitem{ta}
1154: % Exact Potts Model Partition Functions on Strips of the Triangular Lattice
1155: S.-C. Chang and R. Shrock, Physica {\bf A 286}, 189 (2000).
1156:
1157: \bibitem{hca}
1158: % Exact Potts Model Partition Functions on Strips of the Honeycomb Lattice
1159: S.-C. Chang and R. Shrock, Physica A {\bf 296}, 183 (2001).
1160:
1161: \bibitem{s3a}
1162: S.-C. Chang and R. Shrock, Physica A {\bf 296}, 234 (2001).
1163:
1164: \bibitem{sdg}
1165: % Complex-Temperature Phase Diagrams for the $q$-State Potts Model on
1166: % Self-Dual Families of Graphs and the Nature of the $q \to \infty$ Limit
1167: S.-C. Chang and R. Shrock, Phys. Rev. E {\bf 64}, 066116 (2001).
1168:
1169: \bibitem{ts}
1170: S.-C. Chang, J. Salas, and R. Shrock, J. Stat. Phys. {\bf 107} 1207 (2002).
1171:
1172: \bibitem{wu3}
1173: % The Infinite-State Potts Model and Restricted Multidimensional Partitions
1174: % of an Integer
1175: F. Y. Wu, J. Mathematical and Computer Modelling {\bf 26}, 269 (1997).
1176:
1177: \bibitem{huangwu}
1178: % The Infinite-State Potts model and solid partitions of an integer
1179: H. Y. Huang and F. Y. Wu, Int. J. Mod. Phys. B {\bf 11}, 121 (1997).
1180:
1181: \bibitem{mft}
1182: L. Mittag and M. Stephen, J. Phys. A {\bf 7}, L109 (1974).
1183:
1184: \end{thebibliography}
1185: \vfill
1186: \eject
1187: \end{document}
1188:
1189: