1: \documentclass[12pt]{article}
2: \usepackage{eqsection,subeqnarray,indent,amsfonts,amssymb}
3: \usepackage{amsfonts}
4: %\usepackage{epic,eepic}
5: \usepackage{graphicx}
6:
7:
8: \footnotesep 14pt
9: \floatsep 28pt plus 2pt minus 4pt % Nominal is double what is in art12.sty
10: \textfloatsep 40pt plus 2pt minus 4pt
11: \intextsep 28pt plus 4pt minus 4pt
12:
13: % Somewhat wider and taller page than in art12.sty
14: \topmargin -0.4in \headsep 0.4in \textheight 9.0in
15: \oddsidemargin 0.25in \evensidemargin 0.25in \textwidth 6in
16:
17: \def\doublespace{ \renewcommand{\baselinestretch}{1.7} \large\normalsize }
18: \def\singlespace{ \renewcommand{\baselinestretch}{1} \large\normalsize }
19: % The \large\normalsize forces the new \baselineskip to take effect.
20:
21: \begin{document}
22:
23: \bibliographystyle{plain}
24:
25: %\date{June, 2006}
26:
27: \title{\vspace*{-2.5cm}
28: Exact Potts Model Partition Functions for Strips of the
29: Honeycomb Lattice}
30:
31: \author{
32: \\
33: {\small Shu-Chiuan Chang$^{a,b}$} \\[-0.2cm]
34: {\small\it $(a)$ Department of Physics} \\[-0.2cm]
35: {\small\it National Cheng Kung University} \\[-0.2cm]
36: {\small\it Tainan 70101} \\[-0.2cm]
37: {\small\it Taiwan } \\[-0.2cm]
38: {\small\it $(b)$ Physics Division} \\[-0.2cm]
39: {\small\it National Center for Theoretical Science} \\[-0.2cm]
40: {\small\it National Taiwan University} \\[-0.2cm]
41: {\small\it Taipei 10617} \\[-0.2cm]
42: {\small\it Taiwan } \\[-0.2cm]
43: {\small\tt scchang@mail.ncku.edu.tw} \\[5mm]
44: {\small Robert Shrock$^{c}$} \\[-0.2cm]
45: {\small\it $(c)$ C.~N.~Yang Institute for Theoretical Physics} \\[-0.2cm]
46: {\small\it State University of New York} \\[-0.2cm]
47: {\small\it Stony Brook, N.~Y.~11794-3840} \\[-0.2cm]
48: {\small\it USA } \\[-0.2cm]
49: {\small\tt robert.shrock@sunysb.edu} \\[-0.2cm]
50: {\protect\makebox[5in]{\quad}} % To force authors' names to be written
51: % vertically, one above another.
52: % (\author seems to put them side-by-side
53: % if there is room.)
54: \\
55: }
56:
57:
58: \maketitle
59:
60: \thispagestyle{empty} % Suppress page number on front page.
61:
62: \begin{abstract}
63:
64: We present exact calculations of the Potts model partition function $Z(G,q,v)$
65: for arbitrary $q$ and temperature-like variable $v$ on $n$-vertex strip graphs
66: $G$ of the honeycomb lattice for a variety of transverse widths equal to $L_y$
67: vertices and for arbitrarily great length, with free longitudinal boundary
68: conditions and free and periodic transverse boundary conditions. These
69: partition functions have the form $Z(G,q,v)=\sum_{j=1}^{N_{Z,G,\lambda}}
70: c_{Z,G,j}(\lambda_{Z,G,j})^m$, where $m$ denotes the number of repeated
71: subgraphs in the longitudinal direction. We give general formulas for
72: $N_{Z,G,j}$ for arbitrary $L_y$. We also present plots of zeros of the
73: partition function in the $q$ plane for various values of $v$ and in the $v$
74: plane for various values of $q$. Explicit results for partition functions are
75: given in the text for $L_y=2,3$ (free) and $L_y=4$ (cylindrical), and plots of
76: partition function zeros are given for $L_y$ up to 5 (free) and $L_y=6$
77: (cylindrical). Plots of the internal energy and specific heat per site for
78: infinite-length strips are also presented.
79:
80: \end{abstract}
81:
82: \bigskip
83: \noindent
84: {\bf Key Words:} Potts model, honeycomb lattice, exact solutions,
85: transfer matrix
86:
87: \clearpage
88:
89: %
90: \newcommand{\beq}{\begin{equation}}
91: \newcommand{\eeq}{\end{equation}}
92: \newcommand{\beqs}{\begin{eqnarray}}
93: \newcommand{\eeqs}{\end{eqnarray}}
94: \newcommand{\lsim}{\mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle<}{\sim}$}}}
95: \newcommand{\gsim}{\mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle>}{\sim}$}}}
96: %
97: \newtheorem{theorem}{Theorem}[section]
98: \newtheorem{corollary}{Corollary}[section]
99: \newtheorem{conjecture}{Conjecture}[section]
100: \newtheorem{lemma}{Lemma}[section]
101: %
102: %\newtheorem{theorem}{Theorem}[section]
103: %\newtheorem{proposition}[theorem]{Proposition}
104: %\newtheorem{lemma}[theorem]{Lemma}
105: %\newtheorem{corollary}[theorem]{Corollary}
106: %\newtheorem{conjecture}[theorem]{Conjecture}
107: %
108: \newcommand{\be}{\begin{equation}}
109: \newcommand{\ee}{\end{equation}}
110: %\renewcommand{\<}{\langle}
111: %\renewcommand{\>}{\rangle}
112: \newcommand{\widebar}{\overline}
113: \def\reff#1{(\protect\ref{#1})}
114: \def\spose#1{\hbox to 0pt{#1\hss}}
115: \def\ltapprox{\mathrel{\spose{\lower 3pt\hbox{$\mathchar"218$}}
116: \raise 2.0pt\hbox{$\mathchar"13C$}}}
117: \def\gtapprox{\mathrel{\spose{\lower 3pt\hbox{$\mathchar"218$}}
118: \raise 2.0pt\hbox{$\mathchar"13E$}}}
119: \def\textprime{${}^\prime$}
120: \def\proof{\par\medskip\noindent{\sc Proof.\ }}
121: \def\qed{\hbox{\hskip 6pt\vrule width6pt height7pt depth1pt \hskip1pt}\bigskip}
122: \def\proofof#1{\bigskip\noindent{\sc Proof of #1.\ }}
123: \def\half{ {1 \over 2} }
124: \def\third{ {1 \over 3} }
125: \def\twothird{ {2 \over 3} }
126: \def\smfrac#1#2{\textstyle{#1\over #2}}
127: \def\smhalf{ \smfrac{1}{2} }
128: \newcommand{\real}{\mathop{\rm Re}\nolimits}
129: \renewcommand{\Re}{\mathop{\rm Re}\nolimits}
130: \newcommand{\imag}{\mathop{\rm Im}\nolimits}
131: \renewcommand{\Im}{\mathop{\rm Im}\nolimits}
132: \newcommand{\sgn}{\mathop{\rm sgn}\nolimits}
133: \newcommand{\tr}{\mathop{\rm tr}\nolimits}
134: \newcommand{\diag}{\mathop{\rm diag}\nolimits}
135: \newcommand{\Gal}{\mathop{\rm Gal}\nolimits}
136: \newcommand{\mycup}{\mathop{\cup}}
137: \newcommand{\Arg}{\mathop{\rm Arg}\nolimits}
138: \def\hboxscript#1{ {\hbox{\scriptsize\em #1}} }
139: \def\zhat{ {\widehat{Z}} }
140: \def\phat{ {\widehat{P}} }
141: \def\qtilde{ {\widetilde{q}} }
142: \newcommand{\mod}{\mathop{\rm mod}\nolimits}
143: \renewcommand{\emptyset}{\varnothing}
144:
145: \def\scra{\mathcal{A}}
146: \def\scrb{\mathcal{B}}
147: \def\scrc{\mathcal{C}}
148: \def\scrd{\mathcal{D}}
149: \def\scrf{\mathcal{F}}
150: \def\scrg{\mathcal{G}}
151: \def\scrl{\mathcal{L}}
152: \def\scro{\mathcal{O}}
153: \def\scrp{\mathcal{P}}
154: \def\scrq{\mathcal{Q}}
155: \def\scrr{\mathcal{R}}
156: \def\scrs{\mathcal{S}}
157: \def\scrt{\mathcal{T}}
158: \def\scrv{\mathcal{V}}
159: \def\scrz{\mathcal{Z}}
160:
161: \def\q{{\sf q}}
162:
163: % Blackboard font
164: \def\Z{{\mathbb Z}}
165: \def\R{{\mathbb R}}
166: \def\C{{\mathbb C}}
167: \def\Q{{\mathbb Q}}
168:
169: % Transfer matrices: use sans-serif font
170: \def\T{{\mathsf T}}
171: \def\H{{\mathsf H}}
172: \def\V{{\mathsf V}}
173: \def\D{{\mathsf D}}
174: \def\J{{\mathsf J}}
175: \def\P{{\mathsf P}}
176: \def\QQ{{\mathsf Q}}
177: \def\RR{{\mathsf R}}
178:
179: %%\def\bsigma{{\mathbf \sigma}}
180: \def\bsigma{\mbox{\protect\boldmath $\sigma$}}
181: % \boldmath is fragile, and without the \protect we get screwed when
182: % we try to use \bsigma in a \caption.
183: \def\bone{{\mathbf 1}}
184: \def\vv{{\bf v}}
185: \def\uu{{\bf u}}
186: \def\w{{\bf w}}
187:
188:
189:
190: \section{Introduction}
191: \label{sec.intro}
192:
193: In this paper we present some theorems on structural properties of Potts model
194: partition functions on strips of the honeycomb ($hc$) lattice of arbitrary
195: width equal to $L_y$ vertices and arbitrarily great length. We also report
196: exact calculations of these partition functions for a number of
197: honeycomb-lattice strips of various widths and arbitrarily great lengths.
198: Using these results, we consider the limit of infinite length. For this limit
199: we calculate thermodynamic functions and determine the loci in the complex $q$
200: and temperature planes where the free energy is non-analytic. This is an
201: extension of our previous study for this lattice in \cite{hca}.
202:
203: We briefly review the definition of the model and relevant notation. Consider
204: a graph $G=(V,E)$, defined by its vertex set $V$ and edge set $E$. (Here we
205: keep the discussion general; shortly, we will specialize to strips of the
206: honeycomb lattice.) Denote the number of vertices and edges as $|V| \equiv n$
207: and $|E|$, respectively. On the graph $G$, at temperature $T$, the Potts model
208: is defined by the partition function $Z(G,q,v) = \sum_{ \{ \sigma_n \} }
209: e^{-\beta {\cal H}}$ with the (zero-field) Hamiltonian ${\cal H} = -J
210: \sum_{\langle i j \rangle} \delta_{\sigma_i \sigma_j}$ where
211: $\sigma_i=1,\ldots,q$ are the spin variables on each vertex $i \in V$; $\beta =
212: (k_BT)^{-1}$; $\langle i j \rangle \in E$ denotes pairs of adjacent vertices,
213: and $J$ is the spin-spin interaction constant. We use the notation $K = \beta
214: J$, $a = e^K$, and $v = e^K-1$. The physical ranges are thus (i) $a \ge 1$,
215: i.e., $v \ge 0$ corresponding to $\infty \ge T \ge 0$ for the Potts ferromagnet
216: with $J > 0$, and (ii) $0 \le a \le 1$, i.e., $-1 \le v \le 0$, corresponding
217: to $0 \le T \le \infty$ for the Potts antiferromagnet with $J < 0$. Let
218: $G^\prime=(V,E^\prime)$ with $E^\prime \subseteq E$. Then $Z(G,q,v)$ can be
219: defined for arbitrary $q$ and $v$ by the formula \cite{fk}
220: %
221: \beq
222: Z(G,q,v) = \sum_{G^\prime \subseteq G} q^{k(G^\prime)}v^{|E^\prime|}
223: \label{cluster}
224: \eeq
225: %
226: where $k(G^\prime)$ denotes the number of connected components of
227: $G^\prime$. We define a (reduced) free energy per site $f=-\beta F$,
228: where $F$ is the free energy, via
229: %
230: \beq
231: f(\{G\},q,v) = \lim_{n \to \infty} \ln [Z(G,q,v)^{1/n} ] \ ,
232: \label{ef}
233: \eeq
234: %
235: where the symbol $\{G\}$ denotes $\lim_{n \to \infty}G$ for a given family of
236: graphs $G$.
237:
238: Our exact results on $Z(G,q,v)$ apply for arbitrary $q$ and $v$. We consider
239: free and cylindrical strip graphs $G$ of the honeycomb lattice of width $L_y$
240: vertices and of arbitrarily great length $L_x$ vertices. Here, free boundary
241: conditions (sometimes denoted FF), mean free in both the transverse and
242: longitudinal directions (the latter being the one that is varied for a fixed
243: width), while cylindrical boundary conditions (sometimes denoted PF) mean
244: periodic in the transverse direction and free in the longitudinal direction.
245: We represent the strip of the honeycomb lattice in the form of bricks oriented
246: horizontally. For the honeycomb lattice with cylindrical boundary conditions,
247: the number of vertices in the transverse direction, $L_y$, must be an even
248: number, and the smallest value without degeneracy (multiple edges) is
249: $L_y=4$. Exact partition functions for arbitrary $q$ and $v$ have previously
250: been presented for strips of the honeycomb lattice with free boundary
251: conditions for width $L_y=2$ in \cite{hca}. Our new results include theorems
252: that describe the structure of the Potts model partition function for strips
253: with free and cylindrical boundary conditions, of arbitrary width and length
254: and explicit calculations using the transfer matrix method (in the
255: Fortuin--Kasteleyn representation \cite{bn}) for strips with free and
256: cylindrical boundary conditions with width $ L_y=3$ (free) and $L_y=4$
257: (cylindrical). We have carried out similar calculations for $L_y \leq 7$
258: (free) and $L_y = 6$ (cylindrical); these are too lengthy to include here. We
259: shall also present plots of partition function zeros in the limit of infinite
260: length, for widths $2 \le L_y \le 5$ (free) and $L_y = 4, 6$ (cylindrical).
261: Related calculations of Potts model partition functions for arbitrary $q$ and
262: $v$ on fixed-width, arbitrary-length strips of the square and triangular
263: lattices are \cite{a}-\cite{jrs05} and \cite{jrs05,ta,tt}, respectively.
264: Analogous partition function calculations for arbitrary $q$ and $v$ on finite
265: sections of 2D lattices with fixed width and length include \cite{kl,kc}. The
266: special case $v=-1$ is the zero-temperature limit of the Potts antiferromagnet,
267: for which $Z(G,q,-1)=P(G,q)$, where $P(G,q)$ is the chromatic polynomial
268: expressing the number of ways of coloring the vertices of the graph $G$ with
269: $q$ colors such that no two adjacent vertices have the same color.
270:
271: As part of our work, we calculate zeros of the partition function in the $q$
272: plane for fixed $v$ and in the $v$ plane for fixed $q$. In the limit of
273: infinite strip length, $L_x \to \infty$, there is a merging of such zeros to
274: form continuous loci of points where the free energy is nonanalytic, which we
275: denote generically as ${\cal B}$. For the limit $L_x \to \infty$ of a given
276: family of strip graphs, this locus is determined as the solution to an
277: algebraic equation and is hence an algebraic curve.
278:
279: There are several motivations for this work. Clearly, exact calculations of
280: Potts model partition functions with arbitrary $q$ and $v$ are of value in
281: their own right. This is especially true since there are no exact calculations
282: of the free energy $f(\{G\},q,v)$ for arbitrary $q$ and $v$ on an infinite
283: lattice of dimension two or higher. Exact calculations on lattice strips of
284: fixed width and arbitrarily great length thus provide a useful set of results
285: complementing other methods of analysis such as series expansions and Monte
286: Carlo simulations of the Potts model. Our structural theorems elucidate the
287: form of the partition function on these strips for arbitrarily great widths as
288: well as lengths. The honeycomb lattice is of interest since, together with the
289: square and triangular lattices, it comprises the third and last regular tiling
290: of the plane which is homopolygonal, i.e. composed of of a single type of
291: regular polygon. While critical properties describing the second-order phase
292: transition of the Potts ferromagnet are universal and independent of lattice
293: type, the behavior of the Potts antiferromagnet is sensitively dependent on
294: lattice type, so that studies of this model on different lattices and lattice
295: strips are valuable.
296:
297: There is a particular motivation for carrying out exact calculations of
298: Potts model partition functions with arbitrary $q$ and $v$, because this allows
299: one to investigate more deeply a unique feature of the model, which is
300: qualitatively different from the behavior on either the square or triangular
301: lattice, namely the property that the critical temperature of the Potts
302: antiferromagnet on the honeycomb lattice decreases to zero, i.e., the critical
303: $v$ decreases to $-1$, at a non-integral value, $q=(3+\sqrt{5})/2$ (as reviewed
304: below in connection with the criticality condition, eq. (\ref{hc_eq})). This
305: is formal, since for $q \not\in {\mathbb Z}_+$, the model (with either sign of
306: $J$) is only defined via the representation (\ref{cluster}), and in the
307: antiferromagnet case (i.e., for $-1 \le v < 0$), this formula can yield a
308: negative, and hence unphysical, result for the partition function. One
309: obviously cannot investigate this formal criticality using the Hamiltonian
310: formulation, which requires $q \in {\mathbb Z}_+$.
311:
312:
313: Moreover, calculations of complex-temperature zeros of the partition function
314: show how the physical phases can be generalized to regions in the plane of a
315: complex-temperature variable, as was discussed for the 2D Ising model on the
316: square lattice \cite{fisher,earlyct}. Calculations of partition function zeros
317: on long finite lattice strips, and the loci ${\cal B}$ in the infinite-length
318: limit, also yield interesting insights into properties of the corresponding
319: phase diagrams in the complex-temperature and complex-$q$ planes.
320:
321: \section{General Structural Theorems}
322: \label{sec.2}
323:
324: \subsection{Preliminaries}
325:
326: In this section we prove several general theorems that describe the structure
327: of the partition function for the honeycomb-lattice strips under consideration.
328: Let $m$ denote the number of bricks in the longitudinal direction for such a
329: strip. Then the length (number of vertices in the longitudinal direction) is
330: %
331: \beq
332: L_x=\cases{2m+1 & for odd $L_x$ \ , \cr\cr
333: 2m+2 & for even $L_x$ \ .}
334: \label{Lxm}
335: \eeq
336: %
337: For this type of strip graph, $Z(G,q,v)$ has the form
338: %
339: \beq
340: Z(G,q,v) = \sum_{j=1}^{N_{Z,G,\lambda}} c_{G,j}(\lambda_{Z,G,j})^m
341: \label{zgsum}
342: \eeq
343: %
344: where the coefficients $c_{G,j}$ and corresponding terms $\lambda_{G,j}$, as
345: well as the total number $N_{Z,G,\lambda}$ of these terms, depend on the type
346: of strip (width and boundary conditions) but not on its length. In the special
347: case $v=-1$, the numbers $N_{Z,G,\lambda}$ will be denoted $N_{P,G,\lambda}$.
348: We define $N_{Z,hc,BC_y \ BC_x,L_y,\lambda}$ as the total number of $\lambda$'s
349: for the honeycomb-lattice strip with the transverse and longitudinal boundary
350: conditions $BC_y$ and $BC_x$ of width $L_y$. Henceforth where no confusion
351: will result, we shall suppress the $\lambda$ subscript. The explicit labels
352: are $N_{Z,hc,FF,L_y}$ and $N_{Z,hc,PF,L_y}$ for the strips of the honeycomb
353: lattices with free and cylindrical boundary conditions.
354:
355:
356: \subsection{Case of Free Boundary Conditions}
357:
358:
359: \begin{theorem} \label{theorem1} For arbitrary $L_y$,
360: %
361: \beq
362: N_{Z,hc,FF,L_y} = \cases{ C_{L_y} & for odd $L_y$ \ , \cr\cr \frac{1}{2}
363: \left [ C_{L_y} + {L_y \choose L_y/2} \right ] & for even $L_y$ \ .}
364: \label{nzhcff}
365: \eeq
366: %
367: \end{theorem}
368:
369: {\sl Proof} \quad A honeycomb-lattice strip with free boundary conditions is
370: symmetric under reflection about the longitudinal axis if and only if $L_y$ is
371: even. Therefore, for odd $L_y$, the total number of $\lambda$'s in the Potts
372: model partition function, $N_{Z,hc,FF,L_y}$, is the same as the number for the
373: triangular lattice strips $N_{Z,tri,FF,L_y}$, namely, the number of
374: non-crossing partitions of the set $\{1,2,...,L_y\}$. This is the Catalan
375: number \cite{ss00,cf}, $C_{L} = (L+1)^{-1} {2L \choose L}$. For even $L_x$,
376: the reflection symmetry reduces $N_{Z,hc,FF,L_y}$ to the number for the square
377: lattice strips $N_{Z,sq,FF,L_y}$, which was given in Theorem 5 of \cite{ts}. \
378: $\Box$ We list the first few values of $N_{Z,hc,FF,L_y}$ in Table
379: \ref{nfreetable}.
380:
381:
382: \bigskip
383:
384: \begin{table}[htbp]
385: \caption{\footnotesize{Numbers of $\lambda$'s for the Potts model partition
386: function and chromatic polynomials for the strips of the honeycomb lattices
387: having free boundary conditions and various widths $L_y$.}}
388: \begin{center}
389: \begin{tabular}{|c|c|c|}
390: \hline\hline $L_y$ & $N_{Z,hc,FF,L_y}$ & $N_{P,hc,FF,L_y}$ \\
391: \hline\hline
392: 2 & 2 & 1 \\ \hline
393: 3 & 5 & 3 \\ \hline
394: 4 & 10 & 5 \\ \hline
395: 5 & 42 & 19 \\ \hline
396: 6 & 76 & 25 \\ \hline
397: 7 & 429 & 145 \\ \hline
398: 8 & 750 & 194 \\ \hline
399: 9 & 4862 & 1230 \\ \hline
400: 10 & 8524 & 1590 \\ \hline
401: 11 & 58786 & 11139 \\ \hline
402: 12 & 104468 & 14681 \\ \hline \hline
403: \end{tabular}
404: \end{center}
405: \label{nfreetable}
406: \end{table}
407:
408: \bigskip
409:
410:
411: We next discuss some combinatorics which will be used in our next theorem. Let
412: us denote the elements in the $k$'th column of the $j$'th row of Pascal's
413: triangle as $P(j,k)$, with the value $P(j,k)={j \choose k}$ \ (the binomial
414: coefficient). The relation between these elements is
415: $P(j,k)=P(j-1,k-1)+P(j-1,k)$ with $P(0,0)=1$, $P(j,-1)=0$, and $P(j,k)=0$ for
416: $j<k$. That is, each element is the sum of the two numbers immediately above
417: it. For the reader's convenience, we display the first few rows of Pascal's
418: triangle in Fig. \ref{pascaltriangle}. Now, $P(j,k)$ is the number of ways to
419: place $k$ black beads and $j-k$ white beads in a line. Similarly, what is known
420: as Losanitsch's triangle \cite{sl} is given by the number of ways to put $j$
421: beads of two colors in a line, modulo reflection symmetry. We denote the entry
422: in the $k$'th column of the $j$'th row of this triangle as $L(j,k)$. For
423: reference, the sequence formed by reading the entries in this triangle (from
424: left to right) by rows is listed as sequence $A034851$ in \cite{sl}. The
425: relation between these entries is essentially the same as for Pascal's
426: triangle, except when $j$ is even and $k$ is odd:
427: %
428: \beq
429: L(j,k)=L(j-1,k-1)+L(j-1,k)- \delta _{j \ {\rm even}, k \ {\rm odd}}
430: {j/2-1 \choose (k-1)/2}
431: \eeq
432: %
433: where $\delta _{j \ {\rm even}, k \ {\rm odd}}=1$ if $j$ is even and $k$ is odd
434: and zero otherwise. For reference, the first few rows of Losanitsch's triangle
435: are shown in Fig. \ref{Losanitschtriangle}. Now form a new triangle by
436: subtracting the entries in Losanitsch's triangle from the corresponding entries
437: in Pascal's triangle, and denote its elements as $PL(j,k)$. The first few rows
438: of this triangle are displayed in Fig. \ref{Pascal-Losanitsch}.
439:
440:
441: \begin{figure}[htbp]
442: \begin{center}
443: \begin{tabular}{ccccccccccccccccc}
444: & & & & & & & & 1 & & & & & & & & \\
445: & & & & & & & 1 & & 1 & & & & & & & \\
446: & & & & & & 1 & & 2 & & 1 & & & & & & \\
447: & & & & & 1 & & 3 & & 3 & & 1 & & & & & \\
448: & & & & 1 & & 4 & & 6 & & 4 & & 1 & & & & \\
449: & & & 1 & & 5 & & 10 & & 10 & & 5 & & 1 & & & \\
450: & & 1 & & 6 & & 15 & & 20 & & 15 & & 6 & & 1 & & \\
451: & 1 & & 7 & & 21 & & 35 & & 35 & & 21 & & 7 & & 1 & \\
452: 1 & & 8 & & 28 & & 56 & & 70 & & 56 & & 28 & & 8 & & 1
453: \end{tabular}
454: \end{center}
455: \caption{\footnotesize{Pascal's triangle.}}
456: \label{pascaltriangle}
457: \end{figure}
458:
459: \begin{figure}[htbp]
460: \begin{center}
461: \begin{tabular}{ccccccccccccccccc}
462: & & & & & & & & 1 & & & & & & & & \\
463: & & & & & & & 1 & & 1 & & & & & & & \\
464: & & & & & & 1 & & 1 & & 1 & & & & & & \\
465: & & & & & 1 & & 2 & & 2 & & 1 & & & & & \\
466: & & & & 1 & & 2 & & 4 & & 2 & & 1 & & & & \\
467: & & & 1 & & 3 & & 6 & & 6 & & 3 & & 1 & & & \\
468: & & 1 & & 3 & & 9 & & 10 & & 9 & & 3 & & 1 & & \\
469: & 1 & & 4 & & 12 & & 19 & & 19 & & 12 & & 4 & & 1 & \\
470: 1 & & 4 & & 16 & & 28 & & 38 & & 28 & & 16 & & 4 & & 1
471: \end{tabular}
472: \end{center}
473: \caption{\footnotesize{Losanitsch's triangle.}}
474: \label{Losanitschtriangle}
475: \end{figure}
476:
477: \begin{figure}[htbp]
478: \begin{center}
479: \begin{tabular}{ccccccccccccccccc}
480: & & & & & & & & 0 & & & & & & & & \\
481: & & & & & & & 0 & & 0 & & & & & & & \\
482: & & & & & & 0 & & 1 & & 0 & & & & & & \\
483: & & & & & 0 & & 1 & & 1 & & 0 & & & & & \\
484: & & & & 0 & & 2 & & 2 & & 2 & & 0 & & & & \\
485: & & & 0 & & 2 & & 4 & & 4 & & 2 & & 0 & & & \\
486: & & 0 & & 3 & & 6 & & 10 & & 6 & & 3 & & 0 & & \\
487: & 0 & & 3 & & 9 & & 16 & & 16 & & 9 & & 3 & & 0 & \\
488: 0 & & 4 & & 12 & & 28 & & 32 & & 28 & & 12 & & 4 & & 0
489: \end{tabular}
490: \end{center}
491: \caption{\footnotesize{Triangle formed by subtraction of elements of
492: Losanitsch's triangle from corresponding elements of Pascal's triangle.}}
493: \label{Pascal-Losanitsch}
494: \end{figure}
495:
496:
497: The next theorem concerns the number of $\lambda$'s $N_{P,hc,FF,L_y}$ in the
498: chromatic polynomial for the free $hc$ strip.
499:
500: \bigskip
501:
502: \begin{theorem} \label{theorem2} For arbitrary $L_y$,
503: %
504: \beq
505: N_{P,hc,FF,L_y}=
506: \cases{ \sum_{i=0}^{(L_y-1)/2} M_{L_y-1-i} {(L_y-1)/2 \choose i} &
507: for odd $L_y$ \ , \cr\cr
508: \sum_{i=0}^{L_y/2-1} N_{P,sq,FF,L_y-i} {L_y/2-1 \choose i} & \cr
509: - \frac12 \sum_{i=1}^{L_y/2-2} PL(L_y/2-1,i) N_{P,FP,[(L_y+1-i)/2]} &
510: for even $L_y$ \ .}
511: \label{nphcpf}
512: \eeq
513: %
514: where $M_L$ is the Motzkin number \cite{ds}, $N_{P,sq,FF,L_y}$ is the number of
515: $\lambda$'s for the square lattice with free boundary conditions given in
516: Theorem 2 of \cite{ts}, $N_{P,FP,L_y}$ is the total number of $\lambda$'s for
517: the square, triangular, or honeycomb lattices with cyclic boundary conditions
518: given in eq.(5.2) of \cite{cf}, and $PL(j,k)$ is the number given by the
519: subtraction of the elements of Losanitsch's triangle from the corresponding
520: elements of Pascal's triangle.
521: \end{theorem}
522:
523: {\sl Proof} \quad Consider odd strips widths $L_y$ first; for these, there is
524: no reflection symmetry. For each transverse slice containing $L_y$ vertices of
525: the honeycomb lattice with free boundary conditions, there are $(L_y-1)/2$
526: edges. Compared with the calculation of partitions for strips of the square
527: lattice, the same numbers of edges are removed, such that the end vertices of
528: each of these missing edges are allowed to have the same color. Firstly, all of
529: the possible partitions for the triangular-lattice strip with free boundary
530: conditions $N_{P,tri,FF,L_y}=M_{L_y-1}$ \cite{ss00} are valid for the honeycomb
531: lattice. Among $(L_y-1)/2$ missing edges, if one pair of end vertices has the
532: same color, the number of partitions is given by $N_{P,tri,FF,L_y-1}$. As an
533: example, for the $L_y=3$ slice of the honeycomb-lattice strip, there is an edge
534: connecting vertices 1 and 2 and no edge connecting vertices 2 and 3. It follows
535: that the set of partitions is comprised of $\{ 1, \delta_{1,3}, \delta_{2,3}\}$
536: in the shorthand notation used in \cite{ts,tt}. Similarly, if two pairs of end
537: vertices of missing edges separately have the same color, there are ${(L_y-1)/2
538: \choose 2}$ choices for the locations of missing edges and the number of
539: partitions for each choice is $N_{P,tri,FF,L_y-2}$. By including all the
540: possible missing edges with the same color on their end vertices, the first
541: line in eq. (\ref{nphcpf}) is established for the honeycomb lattice with odd
542: $L_y$.
543:
544: For the honeycomb lattice with even $L_y$, reflection symmetry must be taken
545: into account. We thus consider partitions with the transverse slice composed of
546: edges connecting vertices 1 and 2, vertices 3 and 4, ..., vertices $L_y-1$ and
547: $L_y$. That is, there are $L_y/2$ edges in each slice and $L_y/2-1$ missing
548: edges. Naively, one would expect that the number of partitions is
549: $\sum_{i=0}^{L_y/2-1} N_{P,sq,FF,L_y-i} {L_y/2-1 \choose i}$ by the above
550: argument. However, this would actually involve an overcounting, because certain
551: partitions with the same color on end vertices of missing edge(s) are
552: equivalent under the reflection symmetry. For example, the $L_y=6$ strip of the
553: honeycomb lattice has two missing edges, i.e., there is no edge between
554: vertices 2 and 3, and between vertices 4 and 5. There are $N_{P,sq,FF,5}=7$
555: partitions if vertices 2 and 3 have the same color, and separately seven
556: partitions if vertices 4 and 5 have the same color. Five partitions for the
557: first case $\delta_{2,3}$, $\delta_{2,3} \delta_{1,6}$, $\delta_{2,3,5}$,
558: $\delta_{2,3} \delta_{1,4,6}$ and $\delta_{2,3,5} \delta_{1,6}$ are equivalent
559: under reflection to the following partitions for the second case:
560: $\delta_{4,5}$, $\delta_{4,5} \delta_{1,6}$, $\delta_{2,4,5}$, $\delta_{4,5}
561: \delta_{1,3,6}$, and $\delta_{2,4,5} \delta_{1,6}$. In general, the
562: double-counted partitions are those symmetric partitions in $N_{P,sq,FF,L_y-i}$
563: such that $i$ pairs of end vertices of missing edges have the same color. This
564: number of symmetric partitions is $\frac12 N_{P,FP,[(L_y+1-i)/2]}$, given as
565: Theorem 1 in \cite{ts}. The double counting occurs when two choices of $i$
566: missing edges are reflection-symmetric to each other. This number of asymmetric
567: choices of missing edges is given by $PL(j,k)$. \ $\Box$ We list the first few
568: values of $N_{P,hc,FF,L_y}$ for the honeycomb lattice with free boundary
569: conditions in Table \ref{nfreetable}.
570:
571:
572:
573: \subsection{Case of Cylindrical Boundary Conditions}
574:
575:
576: For the honeycomb lattice with cylindrical boundary condition, only strips with
577: even $L_y$ can be defined. The number of $\lambda$'s can be reduced from the
578: number of non-crossing partitions $N_{Z,tri,FF,L_y}=C_{L_y}$ with length-two
579: rotational symmetry, and this first reduction will be denoted as
580: $N_{Z,hc,PF,L_y}^\prime$. This number can be further reduced with reflection
581: symmetry, and this further-reduced number will be denoted as
582: $N_{Z,hc,PF,L_y}$. We list the first few relevant numbers in the Potts model
583: partition function for the strips of the honeycomb lattices with cylindrical
584: boundary conditions in Table \ref{nztable}.
585:
586:
587: \begin{table}[htbp]
588: \caption{\footnotesize{Numbers of $\lambda$'s in the Potts model partition
589: function for the strips of the honeycomb lattices having cylindrical boundary
590: conditions and various even $L_y$.}}
591: \begin{center}
592: \begin{tabular}{|c|c|c|c|c|c|}
593: \hline\hline $L_y$ & $N_{Z,tri,FF,L_y}$ & $N_{Z,hc,PF,L_y}^\prime$ &
594: $N_{Z,hc,PF,L_y}$ & $2N_{Z,hc,PF,L_y}$ & $\frac{L_y}{2}N_{Z,hc,PF,L_y}^\prime$
595: \\
596: & & & & $-N_{Z,hc,PF,L_y}^\prime$ & $-N_{Z,tri,FF,L_y}$ \\
597: \hline\hline
598: 2 & 2 & 2 & 2 & 2 & 0 \\ \hline
599: 4 & 14 & 10 & 8 & 6 & 6 \\ \hline
600: 6 & 132 & 48 & 34 & 20 & 12 \\ \hline
601: 8 & 1430 & 378 & 224 & 70 & 82 \\ \hline
602: 10 & 16796 & 3364 & 1808 & 252 & 24 \\ \hline
603: 12 & 208012 & 34848 & 17886 & 924 & 1076 \\ \hline \hline
604: \end{tabular}
605: \end{center}
606: \label{nztable}
607: \end{table}
608:
609: \bigskip
610:
611: \begin{lemma} \label{lemma1} For arbitrary even $L_y$,
612: %
613: \beq
614: \frac{L_y}{2} N_{Z,hc,PF,L_y}^\prime - N_{Z,tri,FF,L_y} = \sum_{{\rm even} \
615: d|L_y; \ 1 \le d < L_y} \phi \left (\frac{L_y}{d} \right ) {2d \choose d}
616: \label{nzhcpfprime-nztriff}
617: \eeq
618: %
619: where $d|L_y$ means that $d$ divides $L_y$, and $\phi(n)$ is the Euler
620: totient function, equal to the number of positive integers not exceeding the
621: positive integer $n$ and relatively prime to $n$.
622: %
623: \end{lemma}
624:
625: {\sl Proof} \quad Because the honeycomb-lattice strip has length-two rotational
626: symmetry, the value $\frac{L_y}{2} N_{Z,hc,PF,L_y}^\prime$ must be larger than
627: $N_{Z,tri,FF,L_y}$. For strips with cylindrical boundary conditions, partitions
628: can be classified according to the periodicity $d$ modulo rotations. That is, a
629: partition could be transformed back to itself when the $L_y$ vertices are
630: rotated by length $d$, where $d$ denotes any of the positive integers that
631: divide $L_y$. The number of partitions which have periodicity $d$ modulo these
632: rotations was denoted as $2\alpha_d$ in the proof of Theorem 2.2 of \cite{tt},
633: and is given by
634: %
635: \beq
636: 2\alpha_d = \frac{1}{d} \sum_{d^\prime|d} \mu (d/d^\prime) {2d^\prime
637: \choose d^\prime}
638: \label{mobiusalpha3}
639: \eeq
640: %
641: where $\mu(n)$ is the M\"obius function, defined as $\mu(n)=-1$ if $n$ is
642: prime, $\mu(n)=0$ if $n$ has a square factor, and $\mu(n)=1$ for other $n$.
643: Now the periodicity $d$ can be either odd or even. For a partition with odd
644: periodicity $d$, the contribution to the excess of $(L_y/2)
645: N_{Z,hc,PF,L_y}^\prime$ relative to $N_{Z,tri,FF,L_y}$ is
646: $2\alpha_d(L_y/2-d)$. On the honeycomb-lattice strip with length-two rotational
647: symmetry, a partition with even periodicity $d$ and its length-one rotation are
648: not equivalent. Thus for a partition with even periodicity $d$ and its
649: length-one rotated counterpart, the contribution is $2\alpha_d(L_y-d)$. We have
650: %
651: \beqs
652: \frac{L_y}{2} N_{Z,hc,PF,L_y}^\prime-N_{Z,tri,FF,L_y} & = & \sum_{{\rm
653: even} \ d|L_y} 2\alpha_d(L_y-d) + \sum_{{\rm odd} \ d|L_y}
654: 2\alpha_d(\frac{L_y}{2}-d) \cr\cr & = & \sum_{d|L_y} 2\alpha_d(L_y-d) -
655: \sum_{{\rm odd} \ d|L_y} \alpha_d L_y \cr\cr & = & \sum_{d|L_y; \ 1 \le d <
656: L_y} \phi (L_y/d) {2d \choose d} - \sum_{{\rm odd} \ d|L_y} \alpha_d L_y
657: \label{nzhcpfprime}
658: \eeqs
659: %
660: where the last line follows from the proof of Theorem 2.2 in \cite{tt}. Using
661: eq. (\ref{mobiusalpha3}) for the second summation in eq. (\ref{nzhcpfprime}) is
662: %
663: \beqs
664: \sum_{{\rm odd} \ d|L_y} \alpha_d L_y & = & \sum_{{\rm odd} \ d|L_y}
665: \frac{L_y}{2d} \sum_{d^\prime|d} \mu (d/d^\prime) {2d^\prime \choose d^\prime}
666: \cr\cr & = & \sum_{{\rm odd} \ d^\prime|L_y} \sum_{{\rm odd} \ d|L_y;
667: d^\prime|d} \frac{L_y}{2d} \mu (d/d^\prime) {2d^\prime \choose d^\prime} \cr\cr
668: & = & \sum_{{\rm odd} \ d^\prime|L_y} \frac12 {2d^\prime \choose d^\prime}
669: \sum_{{\rm odd} \ d^{\prime\prime}|L_y^\prime}
670: \frac{L_y^\prime}{d^{\prime\prime}} \mu (d^{\prime\prime}) \ ,
671: \eeqs
672: %
673: where we change the variables to $d^{\prime\prime}=d/d^\prime$ and
674: $L_y^\prime=L_y/d^\prime$. (This is given as sequence A062570 in \cite{sl}.)
675: The summation of $d^{\prime\prime}$ is
676: %
677: \beq
678: \sum_{{\rm odd} \ d^{\prime\prime}|L_y^\prime}
679: \frac{L_y^\prime}{d^{\prime\prime}} \mu (d^{\prime\prime}) = \phi (2L_y^\prime)
680: = 2\phi (L_y/d^\prime) \ , \eeq
681: %
682: where that second identity follows because $L_y^\prime$ is even by the formula
683: for the Euler totient function, $\phi (n)=n \prod _{{\rm prime} \
684: p|n}(1-1/p)$. Therefore, we find
685: %
686: \beq
687: \sum_{{\rm odd} \ d|L_y} \alpha_d L_y = \sum_{{\rm odd} \ d|L_y} \phi (L_y/d) {2d \choose d} \ ,
688: \eeq
689: %
690: and the proof is completed. \ $\Box$
691:
692: \bigskip
693: \begin{lemma} \label{lemma2} For arbitrary even $L_y$,
694: %
695: \beq
696: 2N_{Z,hc,PF,L_y} - N_{Z,hc,PF,L_y}^\prime = N_{Z,FP,\frac{L_y}{2}}
697: \label{2nzhcpf-nzhcpfprime}
698: \eeq
699: %
700: where $N_{Z,FP,L_y}={2L_y \choose L_y}$ is the total number of $\lambda$'s for
701: the square or triangular or honeycomb lattices with cyclic boundary conditions,
702: given in eq.(5.6) of \cite{cf}.
703: %
704: \end{lemma}
705:
706: {\sl Proof} \quad We use the result in Theorem 2.1 of \cite{tt} that
707: $2N_{Z,sq,PF,L_y}-N_{Z,tri,PF,L_y} =N_{Z,FP,\frac{L_y}{2}}$ for even $L_y$ is
708: the number of partitions with both rotation and reflection symmetries. The
709: partitions that do not have reflection symmetry appear once in
710: $N_{Z,sq,PF,L_y}$ but twice (itself plus its mirror image) in
711: $N_{Z,tri,PF,L_y}$, so that they do not contribute to
712: $2N_{Z,sq,PF,L_y}-N_{Z,tri,PF,L_y}$. It was shown there that for even $L_y$,
713: there are two classes of reflection symmetries, denoted as type I and type
714: II. For type I partitions, the reflection axis does not go through any vertex;
715: while for type II partitions, the reflection axis goes through two
716: vertices. There are at least two partitions that belong to both type I and II
717: classes, namely, the partitions 1 (identity) and $\delta_{1,2,...,L_y}$ (unique
718: block). Denote the set of partitions of $N_{Z,tri,PF,L_y}$ as ${\bf
719: P}_{Z,tri,PF,L_y}$, that of $N_{Z,hc,PF,L_y}^\prime$ as ${\bf
720: P}_{Z,hc,PF,L_y}^\prime$, and that of $N_{Z,hc,PF,L_y}$ as ${\bf
721: P}_{Z,hc,PF,L_y}$. For example, ${\bf P}_{Z,tri,PF,4} = \{ 1, \delta_{1,2,3,4},
722: \delta_{1,2}, \delta_{1,3}, \delta_{1,2}\delta_{3,4}, \delta_{1,3,4} \}$. As
723: the cylindrical strip of the honeycomb lattice only has length-two rotation
724: symmetry, $N_{Z,hc,PF,L_y}^\prime$ is larger than $N_{Z,tri,PF,L_y}$. The
725: excess partitions in ${\bf P}_{Z,hc,PF,L_y}^\prime$ relative to ${\bf
726: P}_{Z,tri,PF,L_y}$ can be obtained by making length-one rotations for certain
727: partitions in ${\bf P}_{Z,tri,PF,L_y}$. For example, in addition to those in
728: ${\bf P}_{Z,tri,PF,L_y}$, ${\bf P}_{Z,hc,PF,4}^\prime$ contains $\delta_{2,3},
729: \delta_{2,4}, \delta_{2,3}\delta_{1,4}, \delta_{1,2,4}$. Notice that the
730: partitions that belong to both type I and II classes remain the same after the
731: length-one rotation (modulo length-two rotational symmetry), so that they
732: should be excluded in the doubling. For the honeycomb lattice, the reduction
733: from ${\bf P}_{Z,hc,PF,L_y}^\prime$ to ${\bf P}_{Z,hc,PF,L_y}$ occurs only for
734: type I partitions. Considering the $L_y=4$ strip again as an example, we
735: observe that $\delta_{1,3}$ and $\delta_{2,4}$ are equivalent, so that only one
736: of them should be kept for ${\bf P}_{Z,hc,PF,L_y}$, and similarly for
737: $\delta_{1,3,4}$ and $\delta_{1,2,4}$. Since the numbers of partitions in type
738: I and II classes are the same, the right hand side of
739: eq. (\ref{2nzhcpf-nzhcpfprime}) is the same as that in Theorem 2.1 of \cite{tt}
740: for even $L_y$. \ $\Box$
741:
742: The exact formula for $N_{Z,hc,PF,L_y}$ follows from Lemmas~\ref{lemma1}
743: and~\ref{lemma2}:
744:
745: \begin{theorem} \label{theorem3} For arbitrary even $L_y$,
746: %
747: \beq
748: N_{Z,hc,PF,L_y} = \frac{1}{2} {L_y \choose L_y/2} + \frac{1}{L_y}
749: \left[ C_{L_y} + \sum_{{\rm even} \ d|L_y; \ 1 \le d < L_y} \phi
750: \left (\frac{L_y}{d} \right ) {2d \choose d} \right] \ .
751: \label{nzhcpf}
752: \eeq
753: %
754: \end{theorem}
755:
756: \bigskip
757:
758: We list the first few relevant numbers in the chromatic polynomial for the
759: strips of the honeycomb lattices with cylindrical boundary conditions in Table
760: \ref{nptable}. Analogously to Lemmas~\ref{lemma1} and~\ref{lemma2}, we state
761: the following conjectures,
762:
763:
764: \begin{table}[htbp]
765: \caption{\footnotesize{Numbers of $\lambda$'s for the chromatic polynomial for
766: the strips of the honeycomb lattices having cylindrical boundary conditions and
767: various even $L_y$.}}
768: \begin{center}
769: \begin{tabular}{|c|c|c|c|c|c|}
770: \hline\hline $L_y$ & $n_P(hc,L_y,0)$ & $N_{P,hc,PF,L_y}^\prime$ &
771: $N_{P,hc,PF,L_y}$ & $2N_{P,hc,PF,L_y}$ & $\frac{L_y}{2}N_{P,hc,PF,L_y}^\prime$
772: \\
773: & & & & $-N_{P,hc,PF,L_y}^\prime$ & $-n_P(hc,L_y,0)$ \\
774: \hline\hline
775: 2 & 1 & 1 & 1 & 1 & 0 \\ \hline
776: 4 & 6 & 5 & 4 & 3 & 4 \\ \hline
777: 6 & 43 & 17 & 12 & 7 & 8 \\ \hline
778: 8 & 352 & 99 & 62 & 25 & 44 \\ \hline
779: 10 & 3114 & 626 & 346 & 66 & 16 \\ \hline
780: 12 & 29004 & 4907 & 2576 & 245 & 438 \\ \hline \hline
781: \end{tabular}
782: \end{center}
783: \label{nptable}
784: \end{table}
785:
786: \bigskip
787:
788: \begin{conjecture} \label{conjecture1} For arbitrary even $L_y$,
789: %
790: \beq
791: \frac{L_y}{2} N_{P,hc,PF,L_y}^\prime - n_P(hc,L_y,0) = \sum_{{\rm even} \
792: d|L_y; \ 1 \le d < L_y} \phi \left (\frac{L_y}{d} \right ) N_{P,hc,FP,d}
793: \label{nphcpfprime-nphcff}
794: \eeq
795: %
796: where $n_P(hc,L_y,0)$ is the number of $\lambda$'s for the cyclic strips of the
797: honeycomb lattice with level $d=0$, and $N_{P,hc,FP,L_y}$ is the total number
798: of $\lambda$'s for these strips given in \cite{hca}.
799: \end{conjecture}
800: %
801: Compared with Lemma \ref{lemma1}, the number of non-crossing partitions
802: $N_{Z,tri,FF,L_y}=C_{L_y}$ is now replaced by the corresponding number
803: $n_P(hc,L_y,0)$ for the honeycomb strip in the chromatic polynomial, and the
804: number ${2d \choose d} =N_{Z,hc,FP,d}$ is replaced by $N_{P,hc,FP,d}$.
805:
806: \begin{conjecture} \label{conjecture2} For arbitrary even $L_y$,
807: %
808: \beq
809: 2N_{P,hc,PF,L_y} - N_{P,hc,PF,L_y}^\prime =
810: \cases{ \frac{1}{2} N_{P,hc,FP,\frac{L_y}{2}} & for odd $L_y/2 > 1$ \ , \cr\cr
811: \frac{1}{2} \left [ N_{P,hc,FP,\frac{L_y}{2}} + N_{P,hc,FP,\frac{L_y-2}{2}}
812: \right ] & for even $L_y/2$ \ .}
813: \label{2nphcpf-nphcpfprime}
814: \eeq
815: %
816: \end{conjecture}
817:
818: Conjectures~\ref{conjecture1} and~\ref{conjecture2} imply an exact formula for
819: $N_{P,hc,PF,L_y}$.
820:
821:
822:
823: \section{Potts Model Partition Functions for Strips of the Honeycomb Lattice
824: with Free Boundary Conditions} \label{sec.free_bc}
825:
826: The Potts model partition function for a strip of the honeycomb lattice of
827: width $L_y$ and length $L_x$ vertices with free boundary conditions is given by
828: %
829: \beq
830: Z(L_y \times L_x,FF,q,v)
831: = \w^{\rm T} \cdot \T^m \cdot \uu_{\rm id}
832: \label{def_Z_free}
833: \eeq
834: %
835: where $\T=\V \cdot \H_2 \cdot \V \cdot \H_1$ is the transfer matrix. $\H_1$ and
836: $\H_2$ are matrices corresponding respectively to adding two kinds of
837: transverse bonds in a slice, and $\V$ corresponding to adding longitudinal
838: bonds in each slice. The number $m$ is related to $L_x$ as defined in
839: eq. (\ref{Lxm}), and the vector $w$ is given by
840: %
841: \beq
842: \w^{\rm T} = \cases{ \w^{\rm T}_{\rm odd} = \vv^{\rm T} \cdot \H_1 & for odd
843: $L_x$ \ , \cr\cr
844: \w^{\rm T}_{\rm even} = \vv^{\rm T} \cdot \H_2 \cdot \V \cdot \H_1 & for even
845: $L_x$ \ .}
846: \eeq
847: %
848: Hereafter we shall follow the notation and the computational methods of
849: \cite{ts,tt}.
850:
851: The matrices $\T$, $\V$, $\H_1$ and $\H_2$ act on the space of connectivities
852: of sites on the first slice, whose basis elements $\vv_{\cal P}$ are indexed by
853: partitions ${\cal P}$ of the vertex set $\{1,\ldots,L_y\}$. In particular,
854: $\uu_{\rm id} = \vv_{\{\{1\},\{2\},\ldots,\{L_y\}\}}$. We denote the set of
855: basis elements for a given strip as ${\bf P} = \{ \vv_{\cal P} \}$.
856:
857: An equivalent way to present a general formula for the partition function is
858: via a generating function. Labelling a lattice strip of a given type and width
859: as $G_m$, with $m$ the length, one has
860: %
861: \beq
862: \Gamma(G,q,v,z) = \sum_{m=0}^\infty z^m Z(G_m,q,v)
863: \label{gamma}
864: \eeq
865: %
866: where $\Gamma(G,q,v,z)$ is a rational function
867: %
868: \beq
869: \Gamma(G,q,v,z) = \frac{{\cal N}(G,q,v,z)}{{\cal D}(G,q,v,z)}
870: \label{gammaform}
871: \eeq
872: %
873: with
874: %
875: \beq
876: {\cal N}(G,q,v,z)=\sum_{j=0}^{\deg_z({\cal N})} A_{G,j}z^j
877: \label{numgamma}
878: \eeq
879: %
880: \beq
881: {\cal D}(G,q,v,z) = 1+\sum_{j=1}^{N_{Z,hc,BC,L_y}} b_{G,j}z^j
882: = \prod_{j=1}^{N_{Z,hc,BC,L_y}}
883: (1- \lambda_{Z,G,j}z)
884: \label{dengamma}
885: \eeq
886: %
887: where the subscript $BC$ denotes the boundary conditions. In the
888: transfer-matrix formalism, the $\lambda_{Z,G,j}$'s in the denominator of the
889: generating function, eq.~(\ref{dengamma}), are the eigenvalues of $\T$.
890:
891:
892: Strips of the honeycomb lattice with free boundary conditions are well-defined
893: for widths $L_y \ge 2$. The partition function $Z(G,q,v)$ was calculated, for
894: arbitrary $q$, $v$, and $m$, for the strip with $L_y=2$ and free boundary
895: conditions in \cite{hca}, using a systematic iterative application of the
896: deletion-contraction theorem. Here after re-expressing the results for $L_y=2$
897: in the present transfer matrix formalism, we shall report explicit results for
898: the partition function for strips with $L_y = 3$ and free boundary conditions.
899: For $4 \le L_y \le 7$, the expressions for $\T(L_y)$, $\w(L_y)$ and $\uu_{\rm
900: id}(L_y)$ are too lengthy to include here. They are available from the authors
901: on request.
902:
903: %
904: % L=2F
905: %
906: \subsection{$L_y=2$} \label{sec.2F}
907:
908: For the strip with width $L_y=2$, we only have to consider odd $L_x$. The
909: number of elements in the basis is equal to $C_2=2$: ${\bf P} = \{ 1,
910: \delta_{1,2} \}$. In this basis, the transfer matrix and the other relevant
911: quantities are given by
912: %
913: \begin{subeqnarray}
914: \T &=& \left( \begin{array}{cc}
915: R_{11} & D_1 F_2 R_{12} \\
916: v^5 & v^4 D_1
917: \end{array} \right) \\
918: \w_{\rm odd}^{\rm T} &=& q \left( F_1, D_1 \right)
919: \label{Tt2ffcompact}
920: \end{subeqnarray}
921: %
922: where
923: %
924: \begin{subeqnarray}
925: \slabel{def_Dk}
926: D_k &=& v+k \\
927: \slabel{def_Fk}
928: F_k &=& q+kv \\
929: \slabel{def_Ek}
930: R_{11} &=& q^4 + 5q^3v + 10q^2v^2 + 10qv^3 + 5v^4 \\
931: R_{12} &=& q^2 + 2qv + 2v^2 \ .
932: \end{subeqnarray}
933: %
934: In terms of this transfer matrix and these vectors, one calculates the
935: partition function $Z(G_m,q,v)$ for the strip with a given length $m$ via
936: eq.~(\ref{def_Z_free}). Equivalently, one can calculate the partition
937: function\ using a generating function, and this was the way in which the
938: results were presented in \cite{hca}, with
939: ${\cal D} = \prod_{j=1}^2(1-\lambda_{hcf2,j}z)$ and
940: %
941: \beq
942: \lambda_{hcf2,(1,2)} = \frac{1}{2} \biggl [ M_1 \pm \sqrt{M_2}
943: \ \biggr ]
944: \label{lams}
945: \eeq
946: %
947: where
948: %
949: \beq
950: M_1=q^4+5q^3v+10q^2v^2+10qv^3+6v^4+v^5
951: \label{t12}
952: \eeq
953: %
954: and
955: %
956: \beqs
957: M_2 & = & q^8+10q^7v+45q^6v^2+120q^5v^3+208q^4v^4-2q^4v^5+244q^3v^5-6q^3v^6
958: \cr\cr
959: & & +196q^2v^6-4q^2v^7+104qv^7+4qv^8+32v^8+8v^9+v^{10} \ .
960: \label{rs12}
961: \eeqs
962: %
963: The product of these eigenvalues, i.e., the determinant of $\T$, is
964: %
965: \beq
966: {\rm det}(\T) = v^4(1+v)(v+q)^4 = v^4 D_1 F_1^4 \ .
967: \label{det_tff2}
968: \eeq
969: %
970: The vanishing of this determinant at $v=-1$ and $v=-q$ occurs because in each
971: case one of the two eigenvalues is absent for, respectively, the chromatic and
972: flow polynomials \cite{hs,f}. Analogous formulas can be given for ${\rm
973: det}(\T)$ for higher values of $L_y$; we omit these for brevity.
974: %
975: %
976: % L=3F
977: %
978: \subsection{$L_y=3$} \label{sec.3F}
979:
980: For the $hc$ strip of width $L_y=3$ the number of elements in the basis for
981: enumerating partitions is given by the Catalan number $C_3=5$. This basis is
982: ${\bf P} = \{ 1, \delta_{1,2}, \delta_{1,3}, \delta_{2,3}, \delta_{1,2,3}
983: \}$. In this basis, the transfer matrices and the other relevant quantities are
984: %
985: \begin{subeqnarray}
986: \T &=& \left( \begin{array}{ccccc}
987: F_1 F_2 S_{11} & D_1 F_1 S_{12} & S_{13} S_{13}^\prime & S_{14} & D_1 S_{15} \\
988: v^5 F_1 F_2 & v^4 D_1 F_1 F_2 & v^5 S_{23} & v^5 S_{25} & v^4 D_1 S_{25} \\
989: v^6 F_1 & v^5 D_1 F_1 & v^4 S_{33} & v^5 S_{25} & v^4 D_1 S_{25} \\
990: v^3 F_1 F_2 R_{12} & v^3 D_1 F_1 S_{42} & v^3 F_2 S_{43} & v^3 S_{44} & v^3 D_1 S_{45} \\
991: v^7 F_1 & v^6 D_1 F_1 & v^6 S_{25} & v^7 D_3 & v^6 D_1 D_3 \\
992: \end{array} \right) \\
993: \w_{\rm odd}^{\rm T} &=&
994: q \left( q(q+v), q(1+v), q+v, q+v, 1+v \right) \\
995: \w_{\rm even}^{\rm T} &=&
996: q \left( F_1^5, D_1 F_1^4, F_1 X_1, F_1^2 X_2, D_1 F_1 X_2 \right) \\
997: \uu_{\rm id}^{\rm T} &=& \left( 1, 0, 0, 0, 0 \right)
998: \end{subeqnarray}
999: %
1000: where the $S_{ij}$ and $X_k$ are defined in a shorthand notation as
1001: %
1002: \begin{subeqnarray}
1003: S_{11} &=& q^4 + 5q^3v + 11q^2v^2 + 12qv^3 + 7v^4 \\
1004: S_{12} &=& q^4 + 6q^3v + 15q^2v^2 + 19qv^3 + 11v^4 \\
1005: S_{13} &=& q^2 + 4qv + 5v^2 \\
1006: S_{13}^\prime &=& q^3 + 4q^2v + 7qv^2 + 7v^3 + v^4 \\
1007: S_{14} &=& q^5 + 8q^4v + 28q^3v^2 + q^3v^3 + 54q^2v^3 + 5q^2v^4 + 59qv^4 + 9qv^5 + 32v^5 + 7v^6 \cr
1008: & & \\
1009: S_{15} &=& q^4 + 7q^3v + 21q^2v^2 + q^2v^3 + 33qv^3 + 4qv^4 + 24v^4 + 5v^5 \\
1010: S_{23} &=& 2q + 5v + v^2 \\
1011: S_{25} &=& q + 4v + v^2 \\
1012: S_{33} &=& q^2 + 4qv + 6v^2 + v^3 \\
1013: S_{42} &=& q^2 + 3qv + 3v^2 \\
1014: S_{43} &=& q^2 + 3qv + 5v^2 + v^3 \\
1015: S_{44} &=& q^3 + 6q^2v + q^2v^2 + 12qv^2 + 3qv^3 + 10v^3 + 3v^4 \\
1016: S_{45} &=& q^2 + 5qv + qv^2 + 7v^2 + 2v^3
1017: \end{subeqnarray}
1018: %
1019: \begin{subeqnarray}
1020: X_1 &=& q^3 + 4q^2v + 6qv^2 + 4v^3 + v^4 \\
1021: X_2 &=& q^2 + 3qv + 3v^2 +v^3
1022: \end{subeqnarray}
1023: %
1024: We will discuss properties of the resultant partition functions below.
1025:
1026:
1027:
1028: \section{Potts Model Partition Functions for Honeycomb-lattice Strips
1029: with Cylindrical Boundary Conditions}
1030:
1031: For the honeycomb lattice with cylindrical boundary conditions, the width must
1032: be even (and larger than two in order to avoid the degenerate situation of
1033: vertical edges forming emanating from and returning to a given vertex). The
1034: Potts model partition function $Z(G,q,v)$ for a honeycomb-lattice strip with
1035: cylindrical boundary conditions can be written in the same form as in
1036: eq.(\ref{def_Z_free}). Here either $\H_1$ or $\H_2$ should include the bond
1037: connecting the boundary sites in the transverse direction. The dimension of
1038: the transfer matrix can be reduced by the two symmetries discussed in Section
1039: \ref{sec.2}, namely the length-two translation symmetry along the transverse
1040: direction and reflection symmetry. This number is $N_{Z,hc,PF,L_y}$ and is
1041: given in terms of $L_y$ by eq.~\reff{nzhcpf} (see Table~\ref{nztable} for some
1042: numerical values). We consider the basis in the translation-invariant and
1043: reflection-invariant subspace to construct the transfer matrix and the
1044: corresponding vectors. To simplify the notation, we will still use $\T$, $\w$
1045: and $\uu_{\rm id}$ as in eq.(\ref{def_Z_free}), so that the partition function
1046: is given by the analogous equation, $Z(L_y \times L_x,PF,q,v)= \w^{\rm T} \cdot
1047: \T^m \cdot \uu_{\rm id}$. We have calculated the transfer matrix $\T(L_y)$ and
1048: the vectors $\w(L_y)$ and $\uu_{\rm id}(L_y)$ for $L_y=4$ and $L_y=6$. The
1049: explicit results for $L_y=4$ are given in the appendix; the results for $L_y =
1050: 6$ are too lengthy to present here, and are available upon request.
1051:
1052:
1053:
1054: \section{Partition Function Zeros in the $\lowercase{q}$ Plane}
1055:
1056:
1057: In this section we shall present results for zeros in the $q$-plane for the
1058: partition function of the Potts antiferromagnet on strips of the honeycomb
1059: lattice with free and cylindrical boundary conditions, for various values of
1060: the temperature-like variable $v$. Fig. \ref{figures_qplane_F} shows these
1061: zeros for strips of widths $2 \le L \leq 5$ and free boundary conditions. In
1062: the limit $L_x \to \infty$ the zeros merge to form sets of curves which,
1063: together, comprise the locus ${\cal B}$. As an illustration of this, in
1064: Fig. \ref{figures_qplane_F2} we show these zeros for the free strips of width
1065: $L_y=3$, together with the loci ${\cal B}$ for various values of $v$. One sees
1066: that the strip lengths that we use to calculate the zeros are sufficiently
1067: great that most of these zeros lie rather close to the infinite-length
1068: asymptotic loci. This behavior - that partition function zeros calculated for
1069: long strips generally lie close to the asymptotic loci ${\cal B}$ - is similar
1070: to what we found in our previous work. Hence, one can draw a reasonably good
1071: inference for many of the features of these loci ${\cal B}$ from the zeros.
1072: The corresponding partition-function zeros for honeycomb-lattice strips with
1073: $L_y=4,6$ and cylindrical boundary conditions are shown in
1074: Figure~\ref{figures_qplane_P}. Again, one could calculate the asymptotic loci
1075: ${\cal B}$, but since the zeros already give a reasonably good idea of the
1076: structure of these loci, they will suffice for our present purposes. Our
1077: Figs. \ref{figures_qplane_F} and \ref{figures_qplane_P} include calculations up
1078: to $L_y$=5 and 6, respectively. We previously studied the case $L_y=2$ in
1079: \cite{hca} (again for arbitrary $q$ and temperature). From our present
1080: results, we see that, for general values of $v$, as the width $L_y$ increases,
1081: the curve envelope moves outward somewhat and the arc endpoints on the left
1082: move slowly toward $q=0$. This behavior is consistent with the hypotheses that
1083: for a given $v$, as $L_y \to \infty$, (i) ${\cal B}_q$ would approach a
1084: limiting locus as $L_y \to \infty$ and (ii) this locus would separate the $q$
1085: plane into different regions, with a curve passing through $q=0$ as well as a
1086: maximal real value, $q_c(v)$. This is qualitatively similar to the behavior
1087: that was found earlier for the square-lattice strips \cite{a,s3a,ts} and the
1088: triangular-lattice strips \cite{tt}. As expected, the convergence to the limit
1089: $L_y \to \infty$ seems to be faster with cylindrical boundary conditions, as
1090: there are no surface effects when the length is made infinite. In this limit
1091: $L_y \to \infty$, one expects that the locus ${\cal B}$ will have the property
1092: that the maximal point at which it crosses the real axis, $q_c(v)$, is equal to
1093: the solution, eq. (\ref{qsol}) (with the plus sign for the square root), of the
1094: criticality condition on the infinite honeycomb lattice, eq. (\ref{hc_eq})
1095: below.
1096:
1097:
1098: We recall some results for the special case $v=-1$ corresponding to the
1099: zero-temperature Potts antiferromagnet. Chromatic zeros were calculated for a
1100: finite patch of the honeycomb lattice with free and cylindrical boundary
1101: conditions in (Fig. 8 of) \cite{baxter87}. Chromatic zeros and asymptotic loci
1102: for strips of this lattice were calculated for free longitudinal boundary
1103: conditions in \cite{hca,hs,strip} and for periodic longitudinal boundary
1104: conditions in \cite{hca,cf,pg,pt} (see, e.g., Fig. 6 of \cite{strip} for
1105: $L_y=3$ and free b.c., Fig. 1 of \cite{pg} for $L_y=2$ and cyclic b.c., Fig. 17
1106: of \cite{hca} and Fig. 7 of \cite{pt} for $L_y=3$ and cyclic b.c., and Figs. 8
1107: and 9 of \cite{pt} for $L_y=4$ and $L_y=5$ with cyclic b.c.). Comparisons of
1108: the arcs on ${\cal B}$ for $v=-1$ obtained for strips with free longitudinal
1109: boundary conditions led to the inference that in the limit $L_y \to \infty$
1110: these loci would separate the $q$ plane into regions including a curve passing
1111: through $q=0$ \cite{bcc}. The property that, for $v=-1$, ${\cal B}_q$
1112: separates the $q$ plane into regions with one of the curves on ${\cal B}_q$
1113: passing through the origin is also observed for lattice strips with finite
1114: width $L_y$ if one imposes periodic longitudinal boundary conditions
1115: \cite{pg},\cite{wcyl}-\cite{bcc}. In particular, for $v=-1$ and cyclic b.c.,
1116: we found the following values of $q_c(-1)$ (i) 2 for $L_y=2$ \cite{pg}, (ii) 2
1117: for $L_y=3$ \cite{hca}, (iii) $\simeq 2.1548$ for $L_y=4$ \cite{pt}, and (iv)
1118: $\simeq 2.2641$ for $L_y=5$ \cite{pt}. For $L_y \to \infty$, eq. (\ref{qsol})
1119: formally yields $q_c(-1)=(3+\sqrt{5})/2 \simeq 2.6180$. This is formal since
1120: the Potts antiferromagnet is, in general, only defined for $q \in {\mathbb
1121: Z}_+$, owing to the fact that the formula (\ref{cluster}) can yield a negative
1122: and hence unphysical value for $Z(G,q,v)$ if $v$ is negative and $q$ is not a
1123: non-negative integer. There are at least two different ways in which, as $L_y
1124: \to \infty$, the loci ${\cal B}$ could approach the limiting form containing
1125: $q_c(-1)$: (i) the endpoints of the complex-conjugate prongs farthest to the
1126: right could move down and pinch the real axis and/or (ii) the point at which
1127: the locus ${\cal B}$ for finite $L_y$ crosses the real axis could increase to
1128: this limiting value. These are not necessarily mutually exclusive; a
1129: combination of (i) and (ii) could presumably occur, so that the final locus
1130: would have a structure similar to that of the $L_y=2$ cyclic case shown in
1131: Fig. 1(a) of \cite{pg} or the $L_y=3$ case shown in Fig. 7 of \cite{pt}, in
1132: which several curves on ${\cal B}$ meet at $q_c(-1)$. Alternatively, the
1133: rightmost complex-conjugate prongs might bend back and intersect the rest of
1134: the boundary ${\cal B}$ away from the real axis, forming ``bubble'' regions, as
1135: we found for the cyclic $L_y=4,5$ strips (Figs. 8 and 9 in \cite{pt}).
1136:
1137:
1138: As regards the general distribution of zeros, for a given $L_y$, as $v$
1139: increases from $-1$ to 0 (i.e., $K=\beta J$ increases from $-\infty$ to 0,
1140: these zeros contract to a point at $q=0$. This is an elementary consequence of
1141: the fact that $K \to 0$, the spin-spin interaction term in the Potts model
1142: Hamiltonian, ${\cal H}$, vanishes, so that the sum over states just counts all
1143: $q$ possible spin states independently at each vertex, and $Z(G,q,v)$
1144: approaches the value $Z(G,q,0)=q^n$. More generally, one can inquire about the
1145: maximum modulus of a zero of $Z(G,q,v)$ in the $q$ plane as a function of $v$.
1146: It has been proved \cite{sokalzero} that (for a graph $G$ without loops), for
1147: the antiferromagnetic Potts model partition function with $|1+v| \le 1$, the
1148: zeros of $Z(G,q,v)$ lie in the disk $|q| < C r |v|$, where $C \simeq 7.964$ and
1149: $r$ is the maximal degree of a vertex, with $r=3$ for our honeycomb-lattice
1150: strips. Our zeros have moduli that lie considerably below this upper bound.
1151: For example, for $v=-1$, the zeros that we have calculated and displayed in
1152: Figs. \ref{figures_qplane_F}-\ref{figures_qplane_P} have moduli bounded above
1153: by about 1.2 and 1.3, respectively, while the above-mentioned inequality would
1154: give an upper bound of $|q| < 3C \simeq 23.9$.
1155:
1156:
1157: One can also plot ${\cal B}_q$ for the ferromagnetic region $0 \le v \le
1158: \infty$. Although we have not included these plots here, we note that an
1159: elementary Peierls argument shows that the Potts ferromagnet on
1160: infinite-length, finite-width strips has no finite-temperature phase transition
1161: and associated magnetic long range order. Hence, for this model ${\cal B}_q$
1162: does not cross the positive real $q$ axis for $0 < v < \infty$.
1163:
1164: \bigskip
1165: \bigskip
1166:
1167: %
1168: % FIGURE 4: zeros in the complex q-plane as a function of v for F,
1169: % antiferromagnet
1170: %
1171: \begin{figure}[hbtp]
1172: \vspace*{-1cm}
1173: \centering
1174: \begin{tabular}{cc}
1175: \includegraphics[width=170pt]{hcFv-1.eps} \qquad \qquad & \qquad \qquad
1176: \includegraphics[width=170pt]{hcFv-0.75.eps} \\
1177: \phantom{(((a)}(a) & \phantom{(((a)}(b) \\[5mm]
1178: \includegraphics[width=170pt]{hcFv-0.5.eps} \qquad \qquad & \qquad \qquad
1179: \includegraphics[width=170pt]{hcFv-0.25.eps} \\
1180: \phantom{(((a)}(c) & \phantom{(((a)}(d) \\
1181: \end{tabular}
1182: \caption[a]{\protect\label{figures_qplane_F} Partition-function zeros in the
1183: $q$ plane for the Potts antiferromagnet with (a)
1184: $v=-1.0$, (b) $v=-0.75$, (c) $v=-0.5$, and (d) $v=-0.25$ on strips with free
1185: boundary conditions and several widths $L_y$: 2 ($\Box$, black), 3 ($\circ$,
1186: red), 4 ($+$, green), and 5 ($\Diamond$, blue), where the colors refer to
1187: the online paper.}
1188: \end{figure}
1189:
1190:
1191: %\clearpage
1192: %
1193: % FIGURE 5: zeros in the complex q-plane as a function of v for F, with loci
1194: %
1195: \begin{figure}[hbtp]
1196: \vspace*{-1cm}
1197: \centering
1198: \begin{tabular}{cc}
1199: \includegraphics[width=170pt]{Ly3Fv-1.eps} \qquad \qquad & \qquad \qquad
1200: \includegraphics[width=170pt]{Ly3Fv-0.75.eps} \\
1201: \phantom{(((a)}(a) & \phantom{(((a)}(b) \\[5mm]
1202: \includegraphics[width=170pt]{Ly3Fv-0.5.eps} \qquad \qquad & \qquad \qquad
1203: \includegraphics[width=170pt]{Ly3Fv-0.25.eps} \\
1204: \phantom{(((a)}(c) & \phantom{(((a)}(d) \\
1205: \end{tabular}
1206: \caption[a]{\protect\label{figures_qplane_F2} Partition-function zeros and
1207: asymptotic loci ${\cal B}$ in the $q$ plane for the Potts antiferromagnet on
1208: the $hc$ strip with width $L_y=3$ and free boundary conditions, for (a)
1209: $v=-1.0$, (b) $v=-0.75$, (c) $v=-0.5$, (d) $v=-0.25$.}
1210: \end{figure}
1211:
1212:
1213: %
1214: % FIGURE 6: zeros in the complex q-plane as a function of v for P
1215: %
1216: \begin{figure}[hbtp]
1217: \vspace*{-1cm}
1218: \centering
1219: \begin{tabular}{cc}
1220: \includegraphics[width=170pt]{hcPv-1.eps} \qquad \qquad & \qquad \qquad
1221: \includegraphics[width=170pt]{hcPv-0.75.eps} \\
1222: \phantom{(((a)}(a) & \phantom{(((a)}(b) \\[5mm]
1223: \includegraphics[width=170pt]{hcPv-0.5.eps} \qquad \qquad & \qquad \qquad
1224: \includegraphics[width=170pt]{hcPv-0.25.eps} \\
1225: \phantom{(((a)}(c) & \phantom{(((a)}(d) \\
1226: \end{tabular}
1227: \caption[a]{\protect\label{figures_qplane_P} Partition-function zeros for (a)
1228: $v=-1.0$, (b) $v=-0.75$, (c) $v=-0.5$, and (d) $v=-0.25$ on strips with
1229: cylindrical boundary conditions and several widths $L_y$: 4 ($\Box$, black), 6
1230: ($\circ$, red), where the colors refer to the online paper.}
1231: \end{figure}
1232:
1233:
1234: \section{Partition Function Zeros in the $\lowercase{v}$ Plane}
1235:
1236: \subsection{General}
1237:
1238: In this section we shall present results for zeros in the $v$-plane, for
1239: various values of $q$, for the partition function of the Potts model on
1240: strips of the honeycomb lattice of widths $L_y \leq 5$ with free boundary
1241: conditions and of widths $L_y=4,6$ with cylindrical boundary conditions.
1242: We recall the possible noncommutativity in the
1243: definition of the free energy for certain special integer values of $q$,
1244: denoted $q_{sp}$, (see eqs.~(2.10), (2.11) of \cite{a}):
1245: %
1246: \beq
1247: \lim_{n \to \infty} \lim_{q \to q_{sp}} Z(G,q,v)^{1/n} \ne
1248: \lim_{q \to q_{sp}} \lim_{n \to\infty} Z(G,q,v)^{1/n} \ .
1249: \label{fnoncom}
1250: \eeq
1251: %
1252: As discussed in \cite{a}, because of this noncommutativity, the formal
1253: definition (\ref{ef}) is, in general, insufficient to define the free energy
1254: $f$ at these special points $q_{sp}$; it is necessary to specify the order of
1255: the limits that one uses in the above equation. We denote the two definitions
1256: using different orders of limits as $f_{qn}$ and $f_{nq}$: $f_{nq}(\{G\},q,v) =
1257: \lim_{n \to \infty} \lim_{q \to q_{sp}} n^{-1} \ln Z(G,q,v)$ and
1258: $f_{qn}(\{G\},q,v) = \lim_{q \to q_{sp}} \lim_{n \to \infty} n^{-1} \ln
1259: Z(G,q,v)$.
1260:
1261: As a consequence of this noncommutativity, it follows that for the special set
1262: of points $q=q_{sp}$ one must distinguish between (i) $({\cal
1263: B}_v(\{G\},q_{sp}))_{nq}$, the continuous accumulation set of the zeros of
1264: $Z(G,q,v)$ obtained by first setting $q=q_{sp}$ and then taking $n \to \infty$,
1265: and (ii) $({\cal B}_v(\{G\},q_{sp}))_{qn}$, the continuous accumulation set of
1266: the zeros of $Z(G,q,v)$ obtained by first taking $n \to \infty$, and then
1267: taking $q \to q_{sp}$. For these special points (cf.~eq.~(2.12) of \cite{a}),
1268: %
1269: \beq
1270: ({\cal B}_v(\{G\},q_{sp}))_{nq} \ne ({\cal B}_v(\{G\},q_{sp}))_{qn} \ .
1271: \label{bnoncom}
1272: \eeq
1273: %
1274: Here this noncommutativity will be relevant for $q=0$ and $q=1$.
1275:
1276: In Figure~\ref{figures_vplane_F} we show the partition-functions zeros in the
1277: $v$-plane, for fixed values of $q$, for strips with $2 \leq L_y \leq 5$ and
1278: free boundary conditions. We have displayed each value of $q$ on a different
1279: plot: (a) $q=0$, (b) $q \simeq 1$, (c), $q=2$, and (d) $q=3$. The corresponding
1280: partition-function zeros for honeycomb-lattice strips with $L_y$=4,6 and
1281: cylindrical boundary conditions are shown in Figure~\ref{figures_vplane_P}.
1282: Complex-temperature phase diagrams and associated partition function zeros were
1283: given in \cite{hca} for $L_y=2$ for free longitudinal boundary conditions. Our
1284: present calculations extend that work to greater strip widths.
1285:
1286:
1287: For our discussion we will need some results concerning the behavior of the
1288: Potts model on the infinite honeycomb lattice (defined via the 2D thermodynamic
1289: limit). The criticality condition for the $q$-state Potts model on this
1290: lattice is \cite{hccrit1}-\cite{wurev}
1291: %
1292: \beq
1293: v^3-3qv-q^2=0 \ .
1294: \label{hc_eq}
1295: \eeq
1296: %
1297: Since eq. (\ref{hc_eq}) is cubic in $v$, it is cumbersome to write the general
1298: solution for $v$ as a function of $q$. The following information will suffice:
1299: the equation has (i) one real root in $v$ for real $q < 0$ and $q > 4$ (ii)
1300: three degenerate real roots, $v=0$, at $q=0$, (iii) three real roots, two of
1301: which are degenerate, at $q=4$: $v=-2, \ -2, \ 4$; and (iv) three distinct real
1302: roots for $0 < q < 4$. A plot of these roots is given, e.g., as Fig. 4 of
1303: \cite{p}. In the interval $0 \le q \le 4$, the maximal root, $v_{hc3}$, which
1304: is the transition point between the paramagnetic (PM) and ferromagnetic (FM)
1305: phases, increases monotonically from $v_{hc3}=0$ at $q=0$ to $v_{hc3}=4$ at
1306: $q=4$, while the middle one, $v_{hc2}$, which is the transition point between
1307: the paramagnetic and antiferromagnetic (AFM) phases, decrease monotonically
1308: from $v_{hc2}=0$ at $q=0$, through $v_{hc2}=-1$ at $q=(3+\sqrt{5})/2$, and then
1309: to $v_{hc2}=-2$ at $q=4$. Only the interval $0 > v_{hc2} \ge -1$ corresponds
1310: to a physical PM-AFM transition; for $q=(3+\sqrt{3})/2$, this transition occurs
1311: at zero temperature, i.e., $v=-1$, and for larger values of $q$, the Potts
1312: antiferromagnet has no physical PM-AFM transition and is disordered and
1313: noncritical even at $T=0$ (e.g., \cite{p3afhc}). A rigorous result is
1314: that the $q$-state Potts antiferromagnet on the honeycomb lattice is disordered
1315: and noncritical even at $T=0$ if $q \ge 4$ \cite{sstheorem}. As noted above,
1316: the critical point for the Potts antiferromagnet for $q \not\in {\mathbb Z}_+$
1317: has a formal, rather than directly physical, significance. The lowest root,
1318: $v_{hc1}$, is unphysical; it is equal to 0 at $q=0$, decreases through $v=-2$
1319: at $q=2$ to a minimum of $-9/4$ at $q=27/8 = 3.375$ and then increases slightly
1320: to reach the value $-2$ again at $q=4$. In the interval $0 \le q \le 4$ it is
1321: convenient to write
1322: %
1323: \beq
1324: q = q(\theta) = 4\cos^2 \Big ( \frac{\theta}{2} \Big )
1325: \label{qt}
1326: \eeq
1327: %
1328: with $0 \le \theta \le \pi$. The main cases of interest here are
1329: contained within the discrete set of values $q=q_r \equiv q(\theta_r)$, where
1330: $e^{i\theta_r}$ is a certain root of unity given by
1331: %
1332: \beq
1333: \theta_r = \frac{2 \pi}{r} \ , \quad r \in {\mathbb Z}_+ \ .
1334: \label{qr}
1335: \eeq
1336: %
1337: These special values $q_r$ in were discussed by Tutte and Beraha in connection
1338: with zeros of chromatic polynomials \cite{t,bkw} and are also of interest since
1339: they correspond to roots of unity for the deformation parameter in the
1340: Temperley-Lieb algebra relevant for the Potts model \cite{saleur,mbook}. Some
1341: values are $q_r=4, \ 0, \ 1, \ 2, (3+\sqrt{5})/2, \ 3$ for $1 \le r \le 6$,
1342: respectively. For $q=q_r$, the solutions of eq. (\ref{hc_eq}) for $v$ have
1343: simple expressions in terms of trigonometric functions, which will be of use
1344: for our discussion below \cite{qv}:
1345: %
1346: \beq
1347: v_{hc1}(r) = -4\cos \Bigl ( \frac{\pi}{r} \Bigr )
1348: \cos \biggl [ \frac{\pi}{3}\Bigl ( \frac{1}{r}-1 \Bigr )\biggr ]
1349: \label{vhc1}
1350: \eeq
1351: %
1352: %
1353: \beq
1354: v_{hc2}(r) = -4\cos \Bigl ( \frac{\pi}{r} \Bigr )
1355: \cos \biggl [ \frac{\pi}{3}\Bigl ( \frac{1}{r}+1 \Bigr )\biggr ]
1356: \label{vhc2}
1357: \eeq
1358: %
1359: and
1360: %
1361: \beq
1362: v_{hc3}(r) = 4\cos \Bigl ( \frac{\pi}{r} \Bigr )
1363: \cos \Bigl ( \frac{\pi}{3r} \Bigr ) \ .
1364: \label{vhc3}
1365: \eeq
1366: %
1367: This set is dual, via the map $v \to q/v$, to the set of solutions of the
1368: corresponding criticality condition for the triangular lattice, $v^3+3v^2-q=0$,
1369: viz., $v_{t,\eta}=-1+2\cos [2(1+\eta r)\pi/(3r)]$ for $\eta=1,0,-1$, given in
1370: eqs. (27)-(29) of \cite{qv}. In previous studies such as \cite{hca}, it has
1371: been found that although infinite-length, finite-width strips are
1372: quasi-one-dimensional systems, and hence the Potts model has no physical
1373: finite-temperature transition for such systems, some aspects of the
1374: complex-temperature phase diagram have close connections with those on the
1375: (infinite) honeycomb lattice. We shall discuss some of these connections
1376: below. One can also express the solutions of eq. (\ref{hc_eq}) as values of
1377: $q$ for a given $v$; these are
1378: %
1379: \beq
1380: q = \frac{1}{2}\Big [ -3v \pm \sqrt{v^2(4v+9)} \ \Big ]
1381: \label{qsol}
1382: \eeq
1383: %
1384: and are real for $v \ge -9/4$.
1385:
1386:
1387: %\clearpage
1388: %
1389: % FIGURE 7: zeros in the complex v-plane as a function of q for F
1390: %
1391: \begin{figure}[hbtp]
1392: \centering
1393: \begin{tabular}{cc}
1394: \includegraphics[width=170pt]{hcFq0.eps} \qquad \qquad & \qquad \qquad
1395: \includegraphics[width=170pt]{hcFq1.eps} \\
1396: \phantom{(((a)}(a) & \phantom{(((a)}(b) \\[5mm]
1397: \includegraphics[width=170pt]{hcFq2.eps} \qquad \qquad & \qquad \qquad
1398: \includegraphics[width=170pt]{hcFq3.eps} \\
1399: \phantom{(((a)}(c) & \phantom{(((a)}(d) \\
1400: \end{tabular}
1401: \caption[a]{\protect\label{figures_vplane_F} Partition-function zeros, in the
1402: $v$ plane, of (a) $Z(G,q,v)/q$ for $q=0$, and $Z(G,q,v)$ for (b) $q=0.999$,
1403: (c) $q=2$, and (d) $q=3$ on strips with free boundary conditions and several
1404: widths $L_y$: 2 ($\Box$, black), 3 ($\circ$, red), 4 ($+$, green), and 5
1405: ($\Diamond$, blue), where the colors refer to the online paper. }
1406: \end{figure}
1407:
1408:
1409: %
1410: % FIGURE 8: zeros in the complex v-plane as a function of q for P
1411: %
1412: \begin{figure}[hbtp]
1413: \centering
1414: \begin{tabular}{cc}
1415: \includegraphics[width=170pt]{hcPq0.eps} \qquad \qquad & \qquad \qquad
1416: \includegraphics[width=170pt]{hcPq1.eps} \\
1417: \phantom{(((a)}(a) & \phantom{(((a)}(b) \\[5mm]
1418: \includegraphics[width=170pt]{hcPq2.eps} \qquad \qquad & \qquad \qquad
1419: \includegraphics[width=170pt]{hcPq3.eps} \\
1420: \phantom{(((a)}(c) & \phantom{(((a)}(d) \\
1421: \end{tabular}
1422: \caption[a]{\protect\label{figures_vplane_P} Partition-function zeros, in the
1423: $v$ plane, of (a) $Z(G,q,v)/q$ for $q=0$, and $Z(G,q,v)$ for (b) $q=0.999$,
1424: (c) $q=2$, and (d) $q=3$ on strips with cylindrical boundary conditions and
1425: several widths $L_y$: 4 ($\Box$, black), 6 ($\circ$, red), where the colors
1426: refer to the online paper. }
1427: \end{figure}
1428:
1429:
1430: %
1431: % FIGURE 9: zeros in the complex v-plane for q_5
1432: %
1433: \begin{figure}[hbtp]
1434: \centering
1435: \begin{tabular}{cc}
1436: \includegraphics[width=170pt]{Fq5.eps} \qquad \qquad & \qquad \qquad
1437: \includegraphics[width=170pt]{Pq5.eps} \\
1438: \phantom{(((a)}(a) & \phantom{(((a)}(b) \\
1439: \end{tabular}
1440: \caption[a]{\protect\label{figures_vplane_q5} Partition-function zeros, in the
1441: $v$ plane for $q=q_5=(1/2)(3+\sqrt{5} \ )$ for (a) free boundary conditions
1442: with $L_y=$ 2 ($\Box$, black), 3 ($\circ$, red), 4 ($+$, green), and 5
1443: ($\Diamond$, blue); (b) cylindrical boundary conditions and
1444: $L_y=$: 4 ($\Box$, black), 6 ($\circ$, red), where the colors refer to the
1445: online paper.}
1446: \end{figure}
1447:
1448:
1449: \subsection{$q=0$}
1450:
1451: From the cluster representation of $Z(G,q,v)$, eq.~(\ref{cluster}), it follows
1452: that this partition function has an overall factor of $q^{k(G)}$, where $k(G)$
1453: denotes the number of components of $G$, i.e., an overall factor of $q$ for a
1454: connected graph. Hence, $Z(G,q=0,v)=0$. In the transfer matrix formalism,
1455: this is evident from the overall factor of $q$ coming from the vector $\w$.
1456: However, if we first take the limit $n \to \infty$ to define ${\cal B}$ for $q
1457: \ne 0$ and then let $q \to 0$ or, equivalently, extract the factor $q$ from the
1458: left vector $\w$, we obtain a nontrivial locus, namely $({\cal
1459: B}_v(\{G\},0)_{qn}$. This is a consequence of the noncommutativity
1460: (\ref{bnoncom}) for $q=0$.
1461:
1462: With the second order of limits or the equivalent removal of the factor of $q$
1463: in $Z$, we obtain the zeros for $q=0$ shown in
1464: Figures~\ref{figures_vplane_F}(a) (free boundary conditions)
1465: and~\ref{figures_vplane_P}(a) (cylindrical boundary conditions). The zeros
1466: appear to converge to a roughly circular curve. We see in these figures that
1467: the limiting curves cross the real $v$-axis at $v\approx -6$. As $L_y$
1468: increases, the arc endpoints on the upper and lower right move toward the real
1469: axis. It is possible that these could pinch this axis at $v=0$ as $L_y \to
1470: \infty$, corresponding to the root of (\ref{hc_eq}) for $q=0$.
1471:
1472:
1473: \subsection{$q=1$}
1474:
1475: For $q=1$, the spin-spin interaction in ${\cal H}$ always has the Kronecker
1476: delta function equal to unity, and hence the Potts model partition function is
1477: given by
1478: %
1479: \be
1480: Z(G,q=1,v) = e^{K|E|} = (1+v)^{|E|}
1481: \label{Z_q=1}
1482: \ee
1483: %
1484: where $|E|$ is the number of edges in the graph $G$. This has a single zero at
1485: $v=-1$. But again, one encounters the noncommutativity (\ref{bnoncom}) for
1486: $q=1$. It is interesting to analyze this in terms of the transfer matrix
1487: formalism. At this value of $q$, both the transfer matrix and the left vector
1488: $\w$ are non-trivial. There is thus a cancellation of terms that yields the
1489: result \reff{Z_q=1}. The strip with $L_y=2$ and free boundary conditions is the
1490: simplest one to analyze: the eigenvalues and coefficients for $q=1$ are given
1491: by
1492: %
1493: \begin{subeqnarray}
1494: \lambda_1(1,v) &=& v^4 \ ; \qquad \quad c_1(1,0) = 0 \\
1495: \lambda_2(1,v) &=& (1+v)^5 \ ; \quad c_2(1,0) = (1+v)
1496: \end{subeqnarray}
1497: %
1498: Thus, only the second eigenvalue contributes to the partition function, and it
1499: gives the expected result $Z(2 \times m,FF,q=1,v) = (1+v)^{5m+1}$. The strip
1500: with $L_y=3$ and free boundary conditions is similar: there is a single
1501: eigenvalue $\lambda_1(1,v)=(1+v)^8$ with a non-zero coefficient $c_1(1,v)$
1502: equal to $(1+v)$ for odd $L_x$ and $(1+v)^5$ for even $L_x$. The other four
1503: eigenvalues including $v^4(1+v)^2$ and the roots of
1504: %
1505: \beq
1506: x^3 -2v^4(v^3+6v^2+4v+1)x^2 + v^8(1+v)^2(v^4+10v^3+15v^2+6v+1)x-v^{14}(1+v)^6
1507: =0
1508: \label{q1eq}
1509: \eeq
1510: %
1511: have identically zero coefficients. Thus, the partition function takes the form
1512: $Z(3\times m,FF,q=1,v) = (1+v)^{8m+k}=(1+v)^{|E|}$, where $k$=1 for odd $L_x$
1513: and 5 for even $L_x$.
1514:
1515: In general, we conclude that at $q=1$, only the eigenvalue
1516: $\lambda=(1+v)^{3L_y}$ contributes to the partition function for cylindrical
1517: boundary conditions, and only the eigenvalue $\lambda=(1+v)^{3L_y-1}$
1518: contributes to the partition function for free b.c. The other eigenvalues do
1519: not contribute because they have zero coefficients in eq. (\ref{zgsum}).
1520: This is analogous to what we found in earlier work for cyclic strips
1521: \cite{a,s3a,ta}.
1522:
1523: In our present case, in order to get insight into ${\cal B}_{qn}$, we have
1524: computed the partition function zeros for a value of $q$ close to 1, namely,
1525: $q=0.999$ (see Figures~\ref{figures_vplane_F}(b)
1526: and~\ref{figures_vplane_P}(b)). The patterns of zeros show less scatter for
1527: cylindrical, as contrasted with free, boundary conditions. Subsituting $r=3$,
1528: i.e., $q=q_3=1$ in the solutions (\ref{vhc1})-(\ref{vhc3}) of the criticality
1529: equation (\ref{hc_eq}), we have
1530: %
1531: \beq
1532: v_{hc1} = -4\cos (\pi/3) \cos(2\pi/9) = -1.5320888...
1533: \label{vhc1_q1}
1534: \eeq
1535: %
1536: \beq
1537: v_{hc2} = -4\cos (\pi/3) \cos(4\pi/9) = -0.34729635...
1538: \label{vhc2_q1}
1539: \eeq
1540: %
1541: \beq
1542: v_{hc3} = 4\cos (\pi/3) \cos(\pi/9) = 1.8793852...
1543: \label{vhc3_q1}
1544: \eeq
1545: %
1546: Our results are consistent with the inference that as $L_y \to \infty$, the
1547: locus ${\cal B}_v $ crosses the real $v$ axis at the values of $v_{hc1}$ and
1548: $v_{hc2}$ in eqs. There is evidently no indication of any crossing near the
1549: value $v_{hc3}$. A noteworthy feature of the patterns of zeros, at least for
1550: the case of cylindrical boundary conditions, is that they do not appear to
1551: exhibit the prongs that tend to occur for some other cases discussed here.
1552:
1553: \subsection{$q=2$}
1554:
1555: The zeros for the $q=2$ Ising case are displayed in
1556: Figures~\ref{figures_vplane_F}(c) and~\ref{figures_vplane_P}(c) for free and
1557: cylindrical boundary conditions, respectively. (Fig. \ref{figures_vplane_F}(c)
1558: extends our previous calculation presented in Fig. 3 of \cite{hca} to greater
1559: widths.) From our earlier work \cite{a,ta} one
1560: knows that the loci ${\cal B}_v$ are different for strips with free or periodic
1561: transverse boundary conditions and free longitudinal boundary conditions, on
1562: the one hand, and free or periodic transverse boundary conditions and periodic
1563: (or twisted periodic) longitudinal boundary conditions. One anticipates,
1564: however, that in the limit of infinite width, the subset of the
1565: complex-temperature phase diagram that is relevant to real physical
1566: thermodynamics will be independent of the boundary conditions used to obtain
1567: the 2D thermodynamic limit.
1568:
1569: In the 2D thermodynamic limit, one knows the complex-temperature phase diagram
1570: exactly for the $q=2$ (Ising) case. (This isomorphism involves the
1571: redefinition of the spin-spin exchange constant $J_{\rm Potts} = 2J_{\rm
1572: Ising}$ and hence $K_{\rm Potts} = 2K_{\rm Ising}$, where $K_{\rm Potts}$ is
1573: denoted simply $K$ here.) Since the honeycomb lattice is bipartite, the phase
1574: boundary separating the paramagnetic and ferromagnetic phases maps into that
1575: separating the paramagnetic and antiferromagnetic phases under the
1576: transformation $K \to -K$, i.e., $a \to 1/a$, where $a=e^K=v+1$. The total
1577: boundary locus ${\cal B}$ is invariant under this inversion $a \to 1/a$.
1578: Because of this symmetry, it is convenient to discuss the phase diagram first
1579: in terms of the variable $a$; the features in the $v$ plane then follow in an
1580: obvious manner. Following the calculation of the zero-field free energy $f$ of
1581: the Ising model on the square lattice \cite{on}, $f$ was calculated on the
1582: triangular and honeycomb lattices in \cite{hc}. The critical points separating
1583: the PM and FM phases are given by $a_c=2+\sqrt{3}$ and $1/a_c=2-\sqrt{3}$,
1584: respectively. In terms of $v$, these correspond to the values $v_{hc3}$ and
1585: $v_{hc2}$ with $r=4$ in eqs. (\ref{vhc3}) and (\ref{vhc2}). The
1586: complex-temperature phase diagram, with boundaries comprised by the locus
1587: ${\cal B}$, was given in the plane of the variable ${\rm tanh}(K) =
1588: (a-1)/(a+1)$ in (Fig. 3 of) \cite{abe} and in the variable $a$ in (Fig. 1(c)
1589: of) \cite{chitri}. The locus ${\cal B}$ separates the complex $a$ plane into
1590: three phases: (i) the physical PM phase occupying the interval $2-\sqrt{3} \le
1591: a \le 2+\sqrt{3}$, and its complex-temperature extension (CTE), where the $S_q$
1592: symmetry is realized explicitly ($S_q$ being the symmetric group on $q$
1593: numbers, the symmetry group of the Hamiltonian), (ii) the physical
1594: ferromagnetic phase occupying the interval $2+\sqrt{3} \le a \le \infty$ (and
1595: its CTE), and (iii) the physical antiferromagnetic phase occupying the interval
1596: $0 \le a \le 2-\sqrt{3}$ and its CTE. The boundary separating the CTE of the
1597: FM and AFM phases is an arc of the unit circle $a=e^{i\theta}$ with $\pi/3 \le
1598: \theta \le 5\pi/3$; it thus has endpoints at $a=e^{\pm i \pi/3}$ and crosses
1599: the real $a$ axis at $a=-1$. The rest of ${\cal B}$ is a closed curve crossing
1600: the real axis at $a=2 \pm \sqrt{3}$ and having intersection points with the
1601: above-mentioned circular arc at the points at $a=\pm i$. In \cite{hca} the
1602: locus ${\cal B}$ for an infinite-length free and cyclic strip with width
1603: $L_y=2$ were compared with this 2D phase diagram.
1604:
1605: Using our exact results, we can compare the loci ${\cal B}_v$ for various strip
1606: widths and either free or periodic transverse boundary conditions with the
1607: known complex-temperature phase diagram for the Ising model on the infinite 2D
1608: honeycomb lattice. This comparison is simplest for the case of free boundary
1609: conditions, so we concentrate on these results. For the finite values of $L_y$
1610: that we have considered, the loci ${\cal B}_v$ inferred from these zeros
1611: clearly contain a circular arc crossing the real $v$ axis at $v=-2$ and have
1612: intersection points at $v=-1 \pm i$, just as is the case with the exactly known
1613: locus ${\cal B}$ for the infinite 2D honeycomb lattice. Moreover, one sees
1614: that as $L_y$ increases, the endpoints of the complex-conjugate arcs move down
1615: toward the real axis. As $L_y \to \infty$, we expect that these arc endpoints
1616: will cross the real $v$ axis at the points $v=1 \pm \sqrt{3}$ that constitute
1617: the intersections, with the real $v$ axis, of the complex-temperature phase
1618: boundaries ${\cal B}$ for the Ising model on the infinite honeycomb lattice.
1619: As in our earlier studies, this comparison shows that, although the behavior of
1620: the asymptotic locus ${\cal B}$ for infinite-length lattice strips is
1621: qualitatively different from the locus for the thermodynamic limit of the
1622: two-dimensional lattice as regards the physical phase transitions (owing to its
1623: quasi-one-dimensional nature), its features for complex temperatures show many
1624: similarities with the exactly known features for the 2D thermodynamic limit.
1625:
1626: \subsection{$q=(3+\sqrt{5})/2$}
1627:
1628: One of the useful features of exact solutions for $Z(G,q,v)$ for arbitrary $q$
1629: and $v$ is that they allow one to analyze values of $q$ that are not positive
1630: integers and hence cannot be represented in Hamiltonian form, but instead
1631: via the relation (\ref{cluster}). Within the sequence of the $q_r$'s, the
1632: first such example is provided by the case $r=5$, i.e., $q=q_5=(3+\sqrt{5})/2
1633: \simeq 2.6180$. For this value, the criticality condition for the model on the
1634: (thermodynamic limit of the) honeycomb lattice has the solutions
1635: %
1636: \beq
1637: v_{hc3} = \frac{1}{2}\Big [ 1 + \sqrt{3(5+2\sqrt{5}) \ } \ \Big ] \simeq 3.165
1638: \label{vhc3q5}
1639: \eeq
1640: %
1641: %
1642: \beq
1643: v_{hc2} = -1
1644: \label{vhc2q5}
1645: \eeq
1646: %
1647: %
1648: \beq
1649: v_{hc1} = \frac{1}{2}\Big [ 1 - \sqrt{3(5+2\sqrt{5}) \ } \ \Big ] \simeq -2.165
1650: \label{vhc1q5}
1651: \eeq
1652: %
1653: The first of these is the PM-FM critical point, while the second formally
1654: corresponds to the critical temperature for the Potts antiferromagnet going to
1655: zero. The third is a complex-temperature singular point. In Figs.
1656: \ref{figures_vplane_q5} we plot zeros of the partition function in the $v$
1657: plane for $q=q_5$ and strips with free and cylindrical boundary conditions.
1658: From these zeros, one can infer that as $L_y \to \infty$, the rightmost
1659: complex-conjugate arc endpoints would move in and pinch the real axis at the
1660: PM-FM value $v_{hc3} \simeq 3.165$ given in eq. (\ref{vhc3q5}). The results
1661: are also consistent with crossings on ${\cal B}$ at the other two points
1662: $v_{hc2}$ and $v_{hc1}$ as well as the values $v \simeq -0.6$ and $v \simeq
1663: -4.4$ which are not roots of the criticality condition (\ref{hc_eq}).
1664:
1665:
1666: \subsection{$q=3$}
1667:
1668: In contrast to the $q=2$ case, the free energy of the $q$-state Potts model has
1669: not been calculated exactly for $q \ge 3$ on any 2D (or higher-dimensional)
1670: lattice, and hence the corresponding complex-temperature phase diagrams are not
1671: known exactly. For $q=3$, i.e., $r=6$, the solutions of the criticality
1672: equation (\ref{hc_eq}) given by eqs. (\ref{vhc1})-(\ref{vhc3}) are
1673: %
1674: \beq
1675: v_{hc3}=2\sqrt{3}\cos(\pi/18) =3.4114741...
1676: \label{vcq3}
1677: \eeq
1678: %
1679: corresponding to the physical PM-FM phase transition point, and two other roots
1680: at the complex-temperature values
1681: %
1682: \beq
1683: v_{hc2}=-2\sqrt{3}\cos(7\pi/18) = -1.1847925...
1684: \label{vq3point2}
1685: \eeq
1686: %
1687: and
1688: %
1689: \beq
1690: v_{hc1}=-2\sqrt{3}\cos(5\pi/18) = -2.2266815...
1691: \label{vq3point3}
1692: \eeq
1693: %
1694: Some discussions of the complex-temperature solutions of eq.~(\ref{hc_eq}) and
1695: their connections with the complex-temperature phase diagram have been given in
1696: \cite{mm}-\cite{p2}.
1697:
1698: The partition-function zeros in the $v$ plane for $q=3$ are displayed in
1699: Figures~\ref{figures_vplane_F}(d) and~\ref{figures_vplane_P}(d) for free and
1700: cylindrical boundary conditions, respectively. We expect that the pair of
1701: complex-conjugate endpoints in this regime will eventually converge to the
1702: ferromagnetic critical point $v_{hc3}$ as $L_y \to \infty$. However,
1703: obviously, an infinite-length strip of finite width $L_y$ is a
1704: quasi-one-dimensional system, so the Potts model has no physical
1705: finite-temperature phase transition on such a strip for any finite $L_y$. The
1706: $q=3$ Potts antiferromagnet is disordered on the honeycomb lattice for all
1707: temperatures $T$ including $T=0$, so there is no finite-temperature PM-AFM
1708: transition.
1709:
1710: In the complex-temperature interval $v < -1$, there are considerable
1711: finite-size and boundary condition effects. Because of this, in previous work,
1712: a combination of partition-function zeros and analyses of low-temperature
1713: series expansions was used \cite{p2}; these enable one at least to locate some
1714: points on the complex-temperature phase boundary. As regards the infinite 2D
1715: honeycomb lattice, because of a duality relation, the complete physical
1716: temperature interval $0 \le T \le \infty$, i.e., $0 \le a \le 1$ of the
1717: $q$-state Potts antiferromagnet on the triangular lattice is mapped to the
1718: complex-temperature interval $-\infty \le v \le -q)$ on the honeycomb lattice
1719: (and vice versa) \cite{hcl}. Ref. \cite{hcl} found that there is a
1720: complex-temperature singularity for the $q=3$ Potts model at
1721: $v_{tri,PM-AFM,q=3} = -0.79691 \pm 0.00003$. From duality, the corresponding
1722: singularity on the honeycomb lattice is $v_{hc,q=3} = 3/v_{tri,PM-AFM,q=3} =
1723: -3.76454 \pm 0.00015$. One anticipates that as $L_y \to \infty$ for the
1724: infinite-length, width-$L_y$ strips of the honeycomb lattice, the left-most
1725: arcs on ${\cal B}_v$ will cross the real $v$ axis at this point. There are
1726: several other cases of interest, such as $q=4$ and $q_r$ values with $r \ge 7$.
1727: For brevity, we do not consider these here.
1728:
1729:
1730: %
1731: \section{Internal Energy and Specific Heat}
1732:
1733: It is of interest to display some of the physical thermodynamic functions for
1734: the Potts model on the infinite-length limits of these strips.
1735: Having calculated the partition function, one obtains the free energy per
1736: site, $f(G,q,v)$ as
1737: %
1738: \be
1739: f(G,q,v) = {1\over n} \log Z(G,q,v)
1740: \label{def_free_energy}
1741: \ee
1742: %
1743: for finite $n$, with the $n \to \infty$ limit having been defined in
1744: eq.~(\ref{ef}) above. The internal energy per site, $E$, is
1745: %
1746: \beq
1747: E(G,q,v)=-\frac{\partial f}{\partial \beta} =
1748: -J(v+1)\frac{\partial f}{\partial v}
1749: \slabel{def_E}
1750: \eeq
1751: %
1752: and the specific heat per site, $C$, is
1753: %
1754: \beq
1755: C = \frac{\partial E}{\partial T} = k_B K^2(v+1)\left [
1756: \frac{\partial f}{\partial v} + (v+1)\frac{\partial^2 f}{\partial v ^2}
1757: \right ] \ .
1758: \label{def_C}
1759: \eeq
1760: %
1761: As the strip width $L_y \to \infty$, these approach the internal energy and
1762: specific heat for the infinite 2D honeycomb lattice. For convenience we define
1763: a dimensionless internal energy
1764: %
1765: \beq
1766: E_r = -\frac{E}{J} = (v+1)\frac{\partial f}{\partial v} \ .
1767: \label{er}
1768: \eeq
1769: %
1770: Note that $\sgn(E_r)$ is (i) opposite to $\sgn(E)$ in the ferromagnetic case
1771: where $J > 0$ for which the physical region is $0 \le v \le \infty$ and (ii)
1772: the same as $\sgn(E)$ in the antiferromagnet case $J<0$ for which the physical
1773: region is $-1 \le v \le 0$. Of course, the infinite-length limits of the
1774: honeycomb-strips considered here are quasi-one-dimensional systems, so that
1775: $f$, $E$, and $C$ are analytic functions of temperature for all finite
1776: temperatures.
1777:
1778:
1779: We recall the high-temperature (equivalently, small--$|K|$) expansion for an
1780: infinite lattice of dimensionality $d \ge 2$ with coordination number $\Delta$:
1781: %
1782: \beq
1783: -\frac{E}{J} = E_r = \frac{\Delta}{2}\left [ \frac{1}{q} + \frac{(q-1)K}{q^2} +
1784: O(K^3) \right ] \ .
1785: \label{ehightemp}
1786: \eeq
1787: %
1788: Here, $\Delta=3$ for the infinite honeycomb lattice. Again, we recall that in
1789: papers on the $q=2$ Ising special case, the Hamiltonian is usually defined as
1790: ${\cal H}_I = -J_I\sum_{\langle i j \rangle} \sigma_i \sigma_j$ with $\sigma_i
1791: = \pm 1$ rather than the Potts model definition of ${\cal H}$, so that one has
1792: the rescaling $2K_I=K$, where $K_I = \beta J_I$. Furthermore, $E_I = -J\langle
1793: \sigma_i \sigma_j \rangle$ raher than the Potts definition $E = -J\langle
1794: \delta_{\sigma_i\sigma_j} \rangle$, where $\langle i j \rangle$ are adjacent
1795: vertices. Hence, for example, for $q=2$, with the usual Ising model
1796: definitions, $E_I(v=0)=0$ rather than $E=-J\Delta/(2q)$ and the
1797: high-temperature expansion is $E_I = -J(\Delta/2)[K + O(K^3)]$ rather than the
1798: $q=2$ form of (\ref{ehightemp}). Similarly, for $T \to 0$, with the
1799: conventional Ising definition, $E_I \to -|J|$ for both the ferromagnetic and
1800: antiferromagnetic cases, while with our Potts-based definition, $E \to
1801: -(\Delta/2)J = -3J/2$, i.e., $E_r \to 3/2$ for the ferromagnet and $E \to 0$
1802: for the antiferromagnet.
1803:
1804: We next present plots of the the (reduced) internal energy $E_r$ per site and
1805: the specific heat per site $C$ on infinite-length honeycomb-lattice strips for
1806: three values of $q$ in increasing order, $q=2$, $q=(3+\sqrt{5})/2$, and $q=3$
1807: in the respective Figs. \ref{EC_q=2}-\ref{EC_q=3}. Each plot contains
1808: curves for $L_y$ from 2 to 5 for strips with free boundary conditions and for
1809: $L_y=4$ and $L_y=6$ for strips with cylindrical boundary conditions. The free
1810: energy and its derivatives with respect to the temperature are independent of
1811: the longitudinal boundary conditions in the limit $L_x \to \infty$, although
1812: they depend on the transverse boundary conditions \cite{a}. As expected, in
1813: the vicinity of the infinite-temperature point $v=0$, the results for the
1814: internal energy are well described by the first few terms of the
1815: high-temperature series expansion given in eq. (\ref{ehightemp}). One can see
1816: the approach of $E_r$ to its zero-temperature limit of 1.5 for the ferromagnet
1817: as $v$ increases through positive values. This approach is more rapid for the
1818: strips with cylindrical boundary conditions, as is understandable since these
1819: minimize finite-size effects in the transverse direction. One can also see the
1820: approach of $E_r$ to its zero-temperature limit of zero for the antiferromagnet
1821: as $v$ decreases toward $v=-1$.
1822:
1823: With regard to the specific heat, the plots show maxima which occur at
1824: values of $v$, denoted $v_m$, that depend on the values of $q$ and $L_y$ and
1825: the transverse boundary conditions. In the ferromagnetic case, these approach
1826: the critical values for the infinite honeycomb lattice, $v_{PM-FM}$, as $L_y$
1827: increases. For the results shown, this approach is from above (below) in $v$
1828: for the case of free (cylindrical) boundary conditions. The heights of the
1829: maxima increase as $L_y$ increases, in accordance with the fact that on 2D
1830: lattices, the specific heat diverges at the PM-FM critical point for the values
1831: of $q$ shown. (This divergence is logarithmic for $q=2$ \cite{on}; more
1832: generally, for the interval $0 \le q \le 4$ where the 2D Potts ferromagnet has
1833: a second-order transition, the specific heat exponent is given by
1834: $\alpha=\alpha'=(2/3)(\pi-2\theta)/(\pi-\theta)$ \cite{wurev,baxterbook}, where
1835: $\theta= 2 \, {\rm arccos}(q^{1/2}/2)$ as in eq. (\ref{qt}), so $\alpha=2/9$
1836: for $q=(3+\sqrt{5})/2$ and $\alpha=1/3$ for $q=3$.) This behavior as a
1837: function of increasing strip width $L_y$ is the analogue, for these
1838: infinite-length strips, of the standard finite-size scaling behavior of the
1839: specific heat on $L \times L$ sections of a regular lattice, for which $|v_m -
1840: v_{PM-FM}| \sim L^{-1/\nu}$ and $C(v=v_m) \sim L^{\alpha/\nu}$, where
1841: $\nu=\nu'$ is the correlation length critical exponent \cite{fss,cft} for the
1842: Potts ferromagnet on a two-dimensional lattice, with $\nu$ being related to
1843: $\alpha$ by the hyperscaling relation $d\nu=2-\alpha$, so that $\nu=1$ for
1844: $q=2$ and $\nu=5/6$ for $q=3$. Note that $\alpha < 0$ for $0 < q < 2$, so
1845: that, although the Potts ferromagnet has a PM-FM critical point for this range
1846: of $q$, with a divergent correlation length, the specific heat has only a
1847: finite, rather than divergent, nonanalyticity at the critical point.
1848:
1849: We next consider the plots of the specific heat in the case of the Potts
1850: antiferromagnet. For $q=2$, $C$ exhibits a maximum at a value of $v$ that
1851: approaches the value $v_{PM-AFM}=1-\sqrt{3}$ for the 2D lattice as $L_y$
1852: increases, and the height of the maximum increases; these are simply related,
1853: by $K \to -K$, to the behavior for the $q=2$ ferromagnet. For
1854: $q=(3+\sqrt{5})/2$ the curves for the
1855: specific heat in the antiferromagnetic region of Fig. \ref{EC_q=q5} exhibit a
1856: maximum not too far from the zero-temperature value $v=-1$. However, the
1857: height of the maximum does not increase very much as $L_y$ increases, for
1858: either the free or cylindrical boundary conditions. Finally, we show $E_r$ and
1859: $C$ for $q=3$, in Fig. \ref{EC_q=3}. For this value of $q$ the Potts
1860: antiferromagnet has no finite-temperature PM-AFM transition and is noncritical
1861: even at $T=0$; consistent with this, although $C$ has a maximum, the height of
1862: this maximum does not increase with increasing $L_y$.
1863:
1864: %
1865: % FIG. 10 E_r and C for q=2
1866: %
1867: \begin{figure}[hbtp]
1868: \centering
1869: \begin{tabular}{cc}
1870: \includegraphics[width=170pt]{EhcFq2.eps} \qquad \qquad & \qquad \qquad
1871: \includegraphics[width=170pt]{EhcPq2.eps} \\
1872: \phantom{(((a)}(a) & \phantom{(((a)}(b) \\[5mm]
1873: \includegraphics[width=170pt]{ChcFq2.eps} \qquad \qquad & \qquad \qquad
1874: \includegraphics[width=170pt]{ChcPq2.eps} \\
1875: \phantom{(((a)}(c) & \phantom{(((a)}(d) \\
1876: \end{tabular}
1877: \caption[a]{\protect\label{EC_q=2} Reduced internal energy $E_r=-E/J$ and
1878: specific heat $C/k_B$ as functions of the temperature-like variable $v$ for the
1879: $q=2$ Potts model on the honeycomb-lattice strips of width $2 \le L_y \le 5$
1880: with free boundary conditions (a) (c) and of width $L_y=4,6$ with cylindrical
1881: boundary conditions (b) (d). The four curves shown for the case of free
1882: boundary conditions correspond to $L_y=2,3,4,5$ as one moves upward and the two
1883: curves for cylindrical boundary conditions correspond to $L_y=4,6$ as one moves
1884: upward. The plot includes both the ferromagnetic and antiferromagnetic Potts
1885: models, for which the temperature ranges are $0 \le v \le \infty$ and $-1 \le v
1886: \le 0$, respectively.}
1887: \end{figure}
1888:
1889:
1890:
1891: %
1892: % FIG. 12 E_r and C for q=(3+sqrt(5))/2
1893: %
1894: \begin{figure}[hbtp]
1895: \centering
1896: \begin{tabular}{cc}
1897: \includegraphics[width=170pt]{EhcFq5.eps} \qquad \qquad & \qquad \qquad
1898: \includegraphics[width=170pt]{EhcPq5.eps} \\
1899: \phantom{(((a)}(a) & \phantom{(((a)}(b) \\[5mm]
1900: \includegraphics[width=170pt]{ChcFq5.eps} \qquad \qquad & \qquad \qquad
1901: \includegraphics[width=170pt]{ChcPq5.eps} \\
1902: \phantom{(((a)}(c) & \phantom{(((a)}(d) \\
1903: \end{tabular}
1904: \caption[a]{\protect\label{EC_q=q5} Reduced internal energy $E_r=-E/J$ and
1905: specific heat $C/k_B$ as functions of the temperature-like variable $v$ for the
1906: Potts model with $q=(3+\sqrt{5})/2$ on the honeycomb-lattice strips of width $2
1907: \le L_y \le 5$ with free boundary conditions (a) (c) and of width $L_y=4,6$
1908: with cylindrical boundary conditions (b) (d). Ordering of curves is as in
1909: Fig. \ref{EC_q=2}.}
1910: \end{figure}
1911:
1912:
1913: %
1914: % FIG. 12 E_r and C for q=3
1915: %
1916: \begin{figure}[hbtp]
1917: \centering
1918: \begin{tabular}{cc}
1919: \includegraphics[width=170pt]{EhcFq3.eps} \qquad \qquad & \qquad \qquad
1920: \includegraphics[width=170pt]{EhcPq3.eps} \\
1921: \phantom{(((a)}(a) & \phantom{(((a)}(b) \\[5mm]
1922: \includegraphics[width=170pt]{ChcFq3.eps} \qquad \qquad & \qquad \qquad
1923: \includegraphics[width=170pt]{ChcPq3.eps} \\
1924: \phantom{(((a)}(c) & \phantom{(((a)}(d) \\
1925: \end{tabular}
1926: \caption[a]{\protect\label{EC_q=3} Reduced internal energy $E_r=-E/J$ and
1927: specific heat $C/k_B$ as functions of the temperature-like variable $v$ for the
1928: $q=3$ Potts model on the honeycomb-lattice strips of width $2 \le L_y \le 5$
1929: with free boundary conditions (a) (c) and of width $L_y=4,6$ with cylindrical
1930: boundary conditions (b) (d). Ordering of curves is as in Fig. \ref{EC_q=2}.}
1931: \end{figure}
1932:
1933: %
1934: % ACKNOWLEDGEMENTS
1935: %
1936: \subsection*{Acknowledgment}
1937:
1938: This research was partially supported by the Taiwan NSC grant
1939: NSC-95-2112-M-006-004 and NSC-95-2119-M-002-001 (S.-C.C.) and the U.S. NSF
1940: grant PHY-03-54776 (R.S.).
1941:
1942: \newpage
1943:
1944:
1945: \section{Appendix: Transfer Matrix for $L_y=4$ Strip with Cylindrical Boundary
1946: Conditions}
1947: \label{sec.4P}
1948:
1949: The number of elements in the basis is eight: ${\bf P} = \{ \delta_{1,2,3,4},
1950: \delta_{1,2,3}+\delta_{1,2,4}+\delta_{1,3,4}+\delta_{2,3,4},
1951: \delta_{1,2}\delta_{3,4}, \delta_{1,2}+\delta_{3,4}, \delta_{1,3}+\delta_{2,4},
1952: \delta_{1,4}\delta_{2,3}, \delta_{2,3}+\delta_{1,4}, 1 \}$. The transfer
1953: matrix is given by
1954: %
1955: \footnotesize
1956: \begin{eqnarray}
1957: & & \T = \cr
1958: & & \left( \begin{array}{cccccccc}
1959: v^8 D_1^2 D_3^2 & 4v^8 D_1 D_3 S_{25} & v^8 D_1^2 T_{13} & 2v^8 D_1 T_{14} & 2v^8 S_{25}^2 & v^9 D_2 D_3^2 & 2v^9 D_3 S_{23} & v^9 T_{18} \\
1960: %
1961: v^6 D_1^2 D_3 S_{25} & v^6 D_1 T_{22} & v^6 D_1^2 T_{23} & v^6 D_1 T_{24} & v^6 S_{25} T_{25} & v^7 D_2 D_3 S_{25} & v^7 T_{27} & v^7 T_{28} \\
1962: %
1963: 0 & 0 & v^8 D_1^2 & 2v^9 D_1 & 0 & 0 & 0 & v^{10} \\
1964: %
1965: v^4 D_1^2 S_{25}^2 & 2v^4 D_1 S_{25} T_{25} & v^4 D_1^2 T_{43} & v^4 D_1 T_{44} & 2v^5 S_{23} S_{33} & v^5 D_2 S_{25}^2 & 2v^5 S_{25} T_{47} & v^5 T_{48} \\
1966: %
1967: v^4 D_1^2 S_{25}^2 & 2v^4 D_1 S_{25} T_{25} & v^5 D_1^2 T_{53} & 2v^5 D_1 T_{54} & v^4 T_{55} & v^5 D_2 S_{25}^2 & 2v^5 S_{25} T_{47} & v^6 T_{58} \\
1968: %
1969: v^6 D_1^2 T_{61} & 4 v^6 D_1 T_{62} & v^6 D_1^2 F_2^2 & 2v^6 D_1 F_2 S_{42} & 2v^6 F_2 T_{65} & v^6 T_{66} & 2v^6 T_{67} & v^6 S_{42}^2 \\
1970: %
1971: v^3 D_1^2 T_{71} & 2v^3 D_1 T_{72} & v^3 D_1^2 T_{73} & 2v^3 D_1 T_{74} & 2v^3 T_{75} & v^3 T_{76} & v^3 T_{77} & v^3 T_{78} \\
1972: %
1973: D_1^2 T_{23} T_{81} & 4D_1 T_{82} & D_1^2 T_{83} & 2D_1 T_{84} & 2T_{85} T_{85}^\prime & T_{86} & T_{87} & T_{88}
1974: %
1975: \end{array} \right) \cr
1976: & &
1977: \end{eqnarray}
1978: \normalsize
1979: %
1980: where the factors $D_k$ and $F_k$ are defined in eqs.
1981: \reff{def_Dk} and \reff{def_Fk}; the $T_{ij}$ are given by
1982: %
1983: \begin{subeqnarray}
1984: %\beqs
1985: T_{13} &=& 2q + 6v + v^2 \\
1986: T_{14} &=& q^2 + 4qv + 7v^2 + v^3 \\
1987: T_{15} &=& q + 8 v + 4 v^2 \\
1988: T_{18} &=& 2q^2 + 6qv + 8v^2 + v^3
1989: \end{subeqnarray}
1990: %
1991: \begin{subeqnarray}
1992: T_{22} &=& 5q^2 + q^2v + 34qv + 10qv^2 + 65v^2 + 33v^3 + 4v^4 \\
1993: T_{23} &=& q^2 + 5qv + 8v^2 + v^3 \\
1994: T_{24} &=& q^3 + 7q^2v + 17qv^2 + 19v^3 + 2v^4 \\
1995: T_{25} &=& q^2 + 6qv + 11v^2 + 2v^3 \\
1996: T_{27} &=& 5q^2 + q^2v + 28qv + 8qv^2 + 41v^2 + 19v^3 + 2v^4 \\
1997: T_{28} &=& q^3 + 6q^2v + 12qv^2 + 11v^3 + v^4
1998: \end{subeqnarray}
1999: %
2000: \begin{subeqnarray}
2001: T_{43} &=& q^3 + 6q^2v + 14qv^2 + 14v^3 + v^4 \\
2002: T_{44} &=& q^4 + 8q^3v + 26q^2v^2 + 42qv^3 + 33v^4 + 2v^5 \\
2003: T_{47} &=& q^2 + 5qv + 7v^2 + v^3 \\
2004: T_{48} &=& q^4 + 7q^3v + 20q^2v^2 + 28qv^3 + 19v^4 + v^5
2005: \end{subeqnarray}
2006: %
2007: \begin{subeqnarray}
2008: T_{53} &=& 2q^2 + 8qv + 10v^2 + v^3 \\
2009: T_{54} &=& q^3 + 6q^2v + 13qv^2 + 12v^3 + v^4 \\
2010: T_{55} &=& q^4 + 8q^3v + 32q^2v^2 + 2q^2v^3 + 68qv^3 + 12qv^4 + 61v^4 + 22v^5 + 2v^6 \\
2011: T_{58} &=& 2q^3 + 10q^2v + 18qv^2 + 14v^3 + v^4
2012: \end{subeqnarray}
2013: %
2014: \begin{subeqnarray}
2015: T_{61} &=& q + 6v + 2v^2 \\
2016: T_{62} &=& q^2 + 6qv + qv^2 + 10v^2 + 3v^3 \\
2017: T_{65} &=& q^2 + 4qv + 8v^2 + 2v^3 \\
2018: T_{66} &=& q^2 + 8qv + 3qv^2 + 21v^2 + 16v^3 + 3v^4 \\
2019: T_{67} &=& q^3 + 7q^2v + q^2v^2 + 18qv^2 + 4qv^3 + 17v^3 + 5v^4
2020: \end{subeqnarray}
2021: %
2022: \begin{subeqnarray}
2023: T_{71} &=& q^3 + 8q^2v + q^2v^2 + 25qv^2 + 6qv^3 + 32v^3 + 13v^4 + v^5 \\
2024: T_{72} &=& 2q^4 + 17q^3v + q^3v^2 + 62q^2v^2 + 8q^2v^3 + 117qv^3 + 25qv^4 + 98v^4 + 34v^5 + 2v^6 \cr
2025: & & \\
2026: T_{73} &=& q^4 + 7q^3v + 21q^2v^2 + 32qv^3 + 22v^4 + v^5 \\
2027: T_{74} &=& q^5 + 8q^4v + 28q^3v^2 + 54q^2v^3 + 58qv^4 + 30v^5 + v^6 \\
2028: T_{75} &=& q^5 + 9q^4v + 37q^3v^2 + q^3v^3 + 87q^2v^3 + 7q^2v^4 + 118v^4q + 19v^5q + 74v^5 + 21v^6 \cr
2029: & & + v^7 \\
2030: T_{76} &=& q^4 + 10vq^3 + 2v^2q^3 + 41v^2q^2 + 16v^3q^2 + v^4q^2 + 88v^3q + 52v^4q + 7qv^5 + 88v^4 \cr
2031: & & + 72v^5 + 17v^6 + v^7 \\
2032: T_{77} &=& 2q^5 + 19q^4v + q^4v^2 + 78q^3v^2 + 8q^3v^3 + 180q^2v^3 + 28q^2v^4 + 240qv^4 + 52qv^5 \cr
2033: & & + 149v^5 + 47v^6 + 2v^7 \\
2034: T_{78} &=& q^6 + 9vq^5 + 36q^4v^2 + 83q^3v^3 + 118q^2v^4 + 99qv^5 + 41v^6 + v^7
2035: \end{subeqnarray}
2036: %
2037: \begin{subeqnarray}
2038: T_{81} &=& q^3 + 5q^2v + 12qv^2 + qv^3 + 16v^3 + 4v^4 \\
2039: T_{82} &=& q^6 + 11q^5v + 55q^4v^2 + q^4v^3 + 162q^3v^3 + 9q^3v^4 + 302q^2v^4 + 34q^2v^5 + 345qv^5 \cr
2040: & & + 67qv^6 + qv^7 + 192v^6 + 61v^7 + 4v^8 \\
2041: T_{83} &=& q^6 + 10q^5v + 45q^4v^2 + 118q^3v^3 + 194q^2v^4 + 194qv^5 + qv^6 + 96v^6 + 4v^7 \\
2042: T_{84} &=& q^7 + 11q^6v + 55q^5v^2 + 163q^4v^3 + 313q^3v^4 + 396q^2v^5 + 313qv^6 + qv^7 + 126v^7 \cr
2043: & & + 4v^8 \\
2044: T_{85} &=& q^3 + 6q^2v + 14qv^2 + 13v^3 + v^4 \\
2045: T_{85}^\prime &=& q^4 + 6q^3v + 16q^2v^2 + 25qv^3 + qv^4 + 22v^4 + 4v^5 \\
2046: T_{86} &=& q^6 + 12q^5v + q^5v^2 + 65q^4v^2 + 12q^4v^3 + 206q^3v^3 + 63q^3v^4 + 2q^3v^5 + 412q^2v^4 \cr
2047: & & + 184q^2v^5 + 15q^2v^6 + 512qv^5 + 312qv^6 + 46qv^7 + qv^8 + 320v^6 + 256v^7 \cr & & + 60v^8 + 4v^9 \\
2048: T_{87} &=& 2q^7 + 24q^6v + 132q^5v^2 + 2v^3q^5 + 434q^4v^3 + 18q^4v^4 + 934q^3v^4 + 72q^3v^5 \cr
2049: & & + 1346q^2v^5 + 166q^2v^6 + 1232qv^6 + 230qv^7 + 2qv^8 + 560v^7 + 160v^8 + 8v^9 \cr & & \\
2050: T_{88} &=& q^8 + 12q^7v + 66q^6v^2 + 218q^5v^3 + 477q^4v^4 + 718q^3v^5 + 739q^2v^6 + 486qv^7 \cr
2051: & & + qv^8 + 165v^8 + 4v^9
2052: %\eeqs
2053: \end{subeqnarray}
2054: %
2055: The vectors $\w$ and $\uu_{\rm id}$ are given by
2056: %
2057: \begin{subeqnarray}
2058: \w_{\rm odd}^{\rm T} &=& q \left( D_1^2, 4D_1 F_1, q D_1^2, 2q D_1 F_1, 2F_1^2, q+2v+v^2, 2 F_1^2, q F_1^2 \right) \\
2059: \w_{\rm even}^{\rm T} &=& q \left( D_1^2 X_2^2, 4D_1 X_1 X_2, D_1^2 X_3, 2 D_1 X_4, 2X_1^2, (q+2v+v^2) X_2^2, 2 X_2 X_5, X_6 \right) \cr
2060: & & \\
2061: \uu_{\rm id}^{\rm T} &=& \left( 1, 0,0,0,0,0,0,0 \right)
2062: \end{subeqnarray}
2063: %
2064: where
2065: %
2066: \begin{subeqnarray}
2067: X_3 &=& v^6 + 6v^5 + 15v^4q + 20v^3q^2 + 15v^2q^3 + 6q^4v + q^5 \\
2068: X_4 &=& v^7 + 7v^6 + 21v^5q + 35v^4q^2 + 35q^3v^3 + 21q^4v^2 + 7vq^5 + q^6 \\
2069: X_5 &=& v^5 + 5v^4 + 10v^3q + 10v^2q^2 + 5vq^3 + q^4 \\
2070: X_6 &=& v^8 + 8v^7 + 28v^6q + 56v^5q^2 + 70q^3v^4 + 56q^4v^3 + 28v^2q^5 + 8q^6v + q^7 \cr
2071: & &
2072: \end{subeqnarray}
2073: %
2074:
2075: \newpage
2076:
2077: \begin{thebibliography}{99}
2078:
2079:
2080: \bibitem{hca}
2081: S.-C. Chang and R. Shrock, Physica A {\bf 296}, 183 (2001).
2082:
2083: \bibitem{fk}
2084: C. M. Fortuin and P. W. Kasteleyn, Physica {\bf 57}, 536 (1972).
2085:
2086: \bibitem{bn}
2087: H. W. J. Bl\"ote and M. P. Nightingale, Physica A {\bf 112}, 405 (1982).
2088:
2089: \bibitem{a}
2090: R. Shrock, Physica A {\bf 283}, 388 (2000).
2091:
2092: \bibitem{s3a}
2093: S.-C. Chang and R. Shrock, Physica A {\bf 296}, 234 (2001).
2094:
2095: \bibitem{s5}
2096: S.-C. Chang and R. Shrock, Physica A {\bf 316}, 335 (2002).
2097:
2098: \bibitem{ts}
2099: S.-C. Chang, J. Salas, and R. Shrock, J. Stat. Phys. {\bf 107}, 1207 (2002).
2100:
2101: \bibitem{zt}
2102: S.-C. Chang and R. Shrock, Physica A {\bf 347}, 314 (2005).
2103:
2104: \bibitem{zttor}
2105: S.-C. Chang and R. Shrock, Physica A {\bf 364}, 231 (2006).
2106:
2107: \bibitem{jrs05}
2108: % Complex-temp. phase diagram of Potts and RSOS models, cond-mat/0511059
2109: J. L. Jacobsen, J.-F. Richard, and J. Salas, Nucl. Phys. B {\bf 743},
2110: 153 (2006); J. L. Jacobsen, private communication.
2111:
2112: \bibitem{ta}
2113: S.-C. Chang and R. Shrock, Physica A {\bf 286}, 189 (2001).
2114:
2115: \bibitem{tt}
2116: S.-C. Chang, J. Jacobsen, J. Salas, and R. Shrock, J. Stat. Phys. {\bf 114},
2117: 763 (2004).
2118:
2119: \bibitem{fisher}
2120: M. E. Fisher, in {\it Lectures in Theoretical Physics} (Univ. of
2121: Colorado Press, Boulder, CO, 1965), vol. 7C, p. 1.
2122:
2123: \bibitem{earlyct}
2124: R. Abe, Prog. Theor. Phys. {\bf 38}, 322 (1967);
2125: S. Katsura, Prog. Theor. Phys. {\bf 38}, 1415 (1967);
2126: S. Ono, Y. Karaki, M. Suzuki, and C. Kawabata, J. Phys. Soc. Jpn. {\bf 25},
2127: 54 (1968); H. Brascamp and H. Kunz, J. Math. Phys. {\bf 15}, 65 (1974).
2128:
2129: \bibitem{kl}
2130: H. Kl\"upfel, Master's Thesis, SUNY Stony Brook, May, 1999; H. Kl\"upfel and
2131: R. Shrock, unpublished.
2132:
2133: \bibitem{kc}
2134: S.-Y. Kim and R. Creswick, Phys. Rev. E {\bf 63}, 066107 (2001).
2135:
2136: \bibitem{sl}
2137: N. J. A. Sloane, {\it The On-Line Encyclopedia of Integer
2138: Sequences}, http://www.research.att.com/~njas/sequences.
2139:
2140: \bibitem{ds}
2141: R. Donaghey, and L. Shapiro, J. Combin. Th. Ser. A {\bf 23},
2142: 291 (1977).
2143:
2144: \bibitem{ss00}
2145: J. Salas and A. Sokal, J. Stat. Phys., {\bf 104}, 609 (2001).
2146:
2147: \bibitem{cf}
2148: S.-C. Chang and R. Shrock, Physica A {\bf 296}, 131 (2001).
2149:
2150: \bibitem{hs}
2151: R. Shrock and S.-H. Tsai, Physica A {\bf 259}, 315 (1998).
2152:
2153: \bibitem{f}
2154: S.-C. Chang and R. Shrock, J. Stat. Phys. {\bf 112}, 815 (2003).
2155:
2156: \bibitem{baxter87}
2157: R. J. Baxter, J. Phys. A {\bf 20}, 5241 (1987).
2158:
2159: \bibitem{strip}
2160: M. Ro\v{c}ek, R. Shrock, and S.-H. Tsai, Physica A {\bf 252}, 505 (1998).
2161:
2162: \bibitem{pg}
2163: R. Shrock and S.-H. Tsai, J. Phys. A {\bf 32} (Lett.), L195 (1999).
2164:
2165: \bibitem{pt}
2166: S.-C. Chang and R. Shrock, Physica A {\bf 346}, 400 (2005).
2167:
2168: \bibitem{wcyl}
2169: R. Shrock and S.-H. Tsai, Phys. Rev. {\bf E60}, 3512 (1999);
2170: Physica A {\bf 275}, 429 (2000).
2171:
2172: \bibitem{pm}
2173: R. Shrock, Phys. Lett. A {\bf 261}, 57 (1999).
2174:
2175: \bibitem{bcc}
2176: R. Shrock, in the {\it Proceedings of the 1999 British
2177: Combinatorial Conference, BCC99} (July, 1999), Discrete Math.
2178: {\bf 231}, 421 (2001).
2179:
2180: \bibitem{sokalzero}
2181: A. D. Sokal, Combin., Prob.. and Comput. {\bf 10}, 41 (2001).
2182:
2183: \bibitem{hccrit1}
2184: D. Kim and R. Joseph, J. Phys. C {\bf 7} (1974) L167.
2185:
2186: \bibitem{hccrit2}
2187: T. W. Burkhardt and B. W. Southern, J. Phys. A {\bf 11} (1978) L247.
2188:
2189: \bibitem{wurev}
2190: F. Y. Wu, Rev. Mod. Phys. {\bf 54}, 235 (1982).
2191:
2192: \bibitem{baxterbook}
2193: R. J. Baxter, {\it Exactly Solved Models in Statistical Mechanics} (Academic,
2194: London, 1982).
2195:
2196: \bibitem{p}
2197: H. Feldmann, R. Shrock, and S.-H. Tsai, Phys. Rev. E {\bf 57}, 1335 (1998).
2198:
2199: \bibitem{p3afhc}
2200: R. Shrock and S.-H. Tsai, J. Math. Phys. A {\bf 30}, 495 (1997).
2201:
2202: \bibitem{sstheorem}
2203: J. Salas and A. D. Sokal, J.Stat. Phys. {\bf 86}, 551 (1997).
2204:
2205: \bibitem{t}
2206: W. T. Tutte, J. Combin. Theory {\bf 9}, 289 (1970).
2207:
2208: \bibitem{bkw}
2209: S. Beraha, J. Kahane, and N. Weiss, J. Combin. Theory B {\bf 27}, 1 (1979);
2210: {\it ibid.} {\bf 28}, 52 (1980).
2211:
2212: \bibitem{mbook}
2213: P. P. Martin, {\it Potts Models and Related Problems in
2214: Statistical Mechanics} (World Scientific, Singapore, 1991).
2215:
2216: \bibitem{saleur}
2217: H. Saleur, Commun. Math. Phys. {\bf 132}, 657 (1990);
2218: Nucl. Phys. B {\bf 360}, 219 (1991).
2219:
2220: \bibitem{qv}
2221: S.-C. Chang and R. Shrock, J. Phys. A {\bf 39}, 10277 (2006).
2222:
2223: \bibitem{on}
2224: L. Onsager, Phys. Rev. {\bf 65}, 117 (1944).
2225:
2226: \bibitem{hc}
2227: R. Houtappel, Physica {\bf 16}, 425 (1950); K. Husimi and I. Syozi,
2228: Prog. Theor. Phys. {\bf 5}, 177 (1950); I. Syozi, Prog. Theor. Phys.
2229: {\bf 5}, 341 (1950); G. Newell, Phys. Rev. {\bf 79}, 876 (1950).
2230:
2231: \bibitem{abe}
2232: R. Abe, T. Dotera, and T. Ogawa, Prog. Theor. Phys. {\bf 85}, 509 (1991).
2233:
2234: \bibitem{chitri}
2235: V. Matveev and R. Shrock, J. Phys. A {\bf 29}, 803 (1996).
2236:
2237: \bibitem{mm}
2238: P. P. Martin and J.-M. Maillard, J. Phys. A {\bf 19}, L547 (1986).
2239:
2240: \bibitem{hcl}
2241: H. Feldmann, R. Shrock, and S.-H. Tsai, J. Phys. A (Lett.)
2242: {\bf 30}, L663 (1997).
2243:
2244: \bibitem{p2}
2245: H. Feldmann, A. J. Guttmann, I. Jensen, R. Shrock, and S.-H.
2246: Tsai, J. Phys. A {\bf 31}, 2287 (1998).
2247:
2248: \bibitem{fss}
2249: M. Barber, in C. Domb and J. Lebowitz, eds., {\it Phase Transitions and
2250: Critical Phenomena} (Academic, New York, 1983), vol. 8, p. 146
2251:
2252: \bibitem{cft}
2253: J. L. Cardy, J. Phys. A {\bf 17}, L385, L961 (1984);
2254: H. W. J. Bl\"ote, and M. P. Nightingale, Phys. Rev. Lett. {\bf 56}, 742 (1986);
2255: I. Affleck, Phys. Rev. Lett. {\bf 56}, 746 (1986).
2256:
2257: \end{thebibliography}
2258:
2259: \vfill
2260: \eject
2261: \end{document}
2262:
2263: