1: \documentstyle [12pt,amsfonts] {article}
2: \input epsf
3: \bibliographystyle{unsrt}
4: \topmargin=-0.4in
5: \oddsidemargin=-0.0in
6: \textheight=8.75in
7: \textwidth=6.5 in
8:
9: \renewcommand{\textfraction}{0}
10:
11: \newcommand{\beq}{\begin{equation}}
12: \newcommand{\eeq}{\end{equation}}
13: \newcommand{\beqs}{\begin{eqnarray}}
14: \newcommand{\eeqs}{\end{eqnarray}}
15: \newcommand{\lsim}{\mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle<}{\sim}$}}}
16: \newcommand{\gsim}{\mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle>}{\sim}$}}}
17: \newcommand{\gtwid}{\mathrel{\raise.3ex\hbox{$>$\kern-.75em\lower1ex
18: \hbox{$\sim$}}}}
19: \newcommand{\ltwid}{\mathrel{\raise.3ex\hbox{$<$\kern-.75em\lower1ex
20: \hbox{$\sim$}}}}
21: \newcommand{\bi}{\bibitem}
22:
23: \catcode`@=11
24: \@addtoreset{equation}{section}
25: \@addtoreset{equation}{subsection}
26: \def\theequation{\ifnum\value{section}=0 \arabic{equation}\ignorespaces
27: \else \ifnum\value{section}=-1 A.\arabic{equation}\ignorespaces
28: \else \ifnum\value{subsection}=0
29: \thesection.\arabic{equation}\ignorespaces
30: \else \thesection.\arabic{subsection}.\arabic{equation}\ignorespaces
31: \fi
32: \fi
33: \fi}
34: \catcode`@=12
35:
36: \newtheorem{th}{Theorem}[section]
37: \newtheorem{cor}{Corollary}[section]
38: \newtheorem{lemma}{Lemma}[section]
39: \newtheorem{defi}{Definition}[section]
40: \newtheorem{conj}{Conjecture}[section]
41:
42: \begin{document}
43:
44: \begin{titlepage}
45:
46: \begin{center}
47:
48: {\Large \bf Reliability Polynomials and their Asymptotic Limits for Families of
49: Graphs}
50:
51: \vspace{1.1cm}
52: {\large Shu-Chiuan Chang${}$\footnote{address after Sept. 2002:
53: Department of Applied Physics, Faculty of Science, Tokyo University of
54: Science, Tokyo 162-8601, Japan; email: chang@rs.kagu.tus.ac.jp}
55: and Robert Shrock${}$\footnote{email:
56: robert.shrock@sunysb.edu }}\\
57: \vspace{18pt}
58: C. N. Yang Institute for Theoretical Physics \\
59: State University of New York \\
60: Stony Brook, NY 11794-3840 \\
61: \end{center}
62:
63: \vskip 0.6 cm
64:
65: \begin{abstract}
66: \vspace{2cm}
67:
68: We present exact calculations of reliability polynomials $R(G,p)$ for lattice
69: strips $G$ of fixed widths $L_y \le 4$ and arbitrarily great length $L_x$ with
70: various boundary conditions. We introduce the notion of a reliability per
71: vertex, $r(\{G\},p) = \lim_{|V| \to \infty} R(G,p)^{1/|V|}$ where $|V|$ denotes
72: the number of vertices in $G$ and $\{G\}$ denotes the formal limit $\lim_{|V|
73: \to \infty} G$. We calculate this exactly for various families of graphs. We
74: also study the zeros of $R(G,p)$ in the complex $p$ plane and determine exactly
75: the asymptotic accumulation set of these zeros ${\cal B}$, across which
76: $r(\{G\})$ is nonanalytic.
77:
78: \end{abstract}
79:
80: \begin{flushleft}
81: keywords: reliability polynomial, Potts model, Tutte polynomial
82: \end{flushleft}
83:
84: \end{titlepage}
85:
86: \section{Introduction}
87:
88: With the development of progressively larger communications networks such as
89: those involving telecommunications and those connecting computers, from large
90: local area networks to the internet, the analysis of the reliability of these
91: networks has become an increasingly important area of study. Because many
92: types of failures of communication links are random, this study falls within
93: the area of statistical physics, as well as engineering and, as will be
94: evident, mathematical graph theory. A common approach to this study is to
95: factorize the reliability into a factor describing the components such as
96: message routers and computers at various nodes, on the one hand, and the links
97: between these nodes, on the other hand. One often assumes that nodal network
98: elements like message routers and computers have a certain probability
99: $p_{node}$ of (normal) operation. Since failures of these nodal components
100: generally occur independently, the factor in the overall reliability due to
101: these elements is then $p_{node}^n$, where $n$ is the number of nodes in
102: the network. The other factor contributing to the overall network reliability
103: involves the connecting links on the network and hence the connectivity
104: structure of the network. As mentioned above, many failures in these
105: connecting links are also random and independent and may thus be treated in a
106: probabilistic manner in which one describes a given link as having a
107: probability $p$ of operating and thus a probability $1-p$ of failing.
108: (Examples of nonrandom failures include earthquakes that affect only one
109: portion of a network; these are not considered here.) The network is thus
110: represented as a connected graph $G=(V,E)$ with vertex (= node) set $V$ and
111: edge (= bond) set $E$, in which each vertex represents either a message router
112: or switch or an endpoint such as a computer or terminal (or telephone in the
113: telecommunications example) and each edge represents the link between the
114: nodes. We shall be interested here in a commonly used simplified model of the
115: network in which each nodal element and link have, respectively, the same fixed
116: probabilities $p_{node}$ and $p$ of operation with $p_{node}$ and $p$ lying in
117: the interval $[0,1]$. We shall denote the number of vertices and edges in $G$
118: as $|V|=n$ and $|E|$. In the graphical representation of the network, one
119: regards a normally operating communications link as an edge that is present
120: (with probability $p$), and a malfunctioning communications link as an edge
121: that is absent (with probability $1-p$), the presence or absence being random
122: and uncorrelated. Defining a successfully operating network as one in which
123: all of links and nodes are operating normally, it follows that the node
124: reliability is, as noted above, $p_{node}^{|V|}$, independent of the
125: connectivity of the network. The factor due to the operation of the links is
126: much more difficult to calculate. It is therefore customary, in modelling the
127: reliability behavior of networks, to separate out the node-reliability factor
128: and concentrate on the contribution due to the structure of the network.
129:
130:
131: The all-terminal reliability polynomial $R(G,p)$ is defined as the probability
132: that there is an operating communications link between any two nodes in the
133: network, i.e. any two vertices in the set $V$ are connected (by a path
134: consisting of edges that are present) \cite{shannon}-\cite{colbourn}. In
135: passing, we note that one can also study $k$-terminal reliability probability
136: polynomials, but we shall restrict ourselves here to the all-terminal
137: reliability polynomial and hence shall henceforth omit the qualifier
138: ``all-terminal'' in the notation. The contributions to $R(G,p)$ arise from the
139: sum of connected spanning subgraphs of $G$. Here a spanning subgraph is a
140: subgraph $G^\prime=(V,E^\prime)$ with $E^\prime \subseteq E$, i.e. a graph with
141: the same vertex set and a subset of the edge set of $G$. Denote a connected
142: spanning subgraph as $H=(V,E_H)$. Each contribution to $R(G,p)$ is weighted by
143: a factor $p^{|E_H|}$ reflecting the probability that the $|E_H|$ edges are
144: present, multiplied by a factor $(1-p)^{|E|-|E_H|}$ reflecting the probability
145: that the edges in the complement set $\{e \in E; e \not\in E_H\}$, of
146: cardinality $|E|-|E_H|$, are absent. That is,
147: %
148: \beq
149: R(G,p)=\sum_{H \subseteq G} p^{|E_H|}(1-p)^{|E|-|E_H|} \ .
150: \label{R}
151: \eeq
152: %
153: Some pioneering studies of network reliability include
154: \cite{shannon}-\cite{kelmans}; a review is \cite{colbourn}. The calculation of
155: $R(G,p)$ for an arbitrary (connected) graph $G$ has been proved to be \#-P
156: complete \cite{colbourn,pb}; that is, roughly speaking, for a generic network
157: $G$, the time required to calculate $R(G,p)$ grows exponentially with the size
158: of the network. In the face of this difficulty, network designers have often
159: relied upon upper and lower bounds on $R(G,p)$ \cite{lp}- \cite{bc96}. It is
160: clearly of interest if one can obtain exact solutions for $R(G,p)$ for some
161: families of graphs $G$.
162:
163:
164: In this paper, we shall present a number of such exact solutions for
165: reliability polynomials, for recursive families of graphs $G$. We have
166: previously presented such solutions for a specific recursive family of graphs
167: in \cite{ka3}. A recursive family of graphs is one in which one constructs
168: successive members of the family in a recursive manner starting from an initial
169: member. Recursive families of graphs that are of particular interest here are
170: strips of regular lattices of a given width $L_y$ vertices and arbitrarily
171: great length $L_x$ vertices with some specified transverse and longitudinal
172: boundary conditions. We shall envision these strips as extending in the
173: horizontal (= longitudinal, $x$) direction and having transverse extent in the
174: vertical, $y$, direction. To see that these form recursive families, one can
175: picture a strip, say of the square lattice, of length $L_x$ and width $L_y$ as
176: being formed by gluing on a column of squares of height $L_y$ to the strip of
177: length $L_x-1$. The boundary conditions that we shall consider include free
178: (FBC), periodic (PBC), and twisted periodic (TPBC), by which is meant that the
179: longitudinal ends are identified with a reversal of orientation. We shall
180: denote the various combinations as (i) $(FBC_y,FBC_x)=$ free, (ii)
181: $(FBC_y,PBC_x)=$ cyclic, (iii) $(FBC_y,TPBC_x)=$ M\"obius, (iv)
182: $(PBC_y,FBC_x)=$ cylindrical, (v) $(PBC_y,PBC_x)=$ torus, and (vi)
183: $(PBC_y,TPBC_x)=$ Klein bottle. For an arbitrary graph $G$, one defines the
184: degree $d_i$ of a vertex $v_i$ as the number of edges connected to it. One
185: then defines the maximum degree as $\Delta = \max_{v_i \in V} d_i$. A
186: $\Delta$-regular graph is one in which all of the vertices have the same degree
187: $\Delta$. Some strip graphs, such as those with toroidal or Klein bottle
188: boundary conditions, are $\Delta$-regular graphs. For strip graphs that are
189: not $\Delta$-regular, it will still be useful to define an average or effective
190: vertex degree in the limit of infinite length, namely
191: %
192: \beq
193: d_{eff} = \lim_{|V| \to \infty} \frac{2|E|}{|V|} \ .
194: \label{delta_eff}
195: \eeq
196: %
197: Secondly, we shall introduce a definition of the reliability per node in the
198: limit of many nodes,
199: %
200: \beq
201: r(\{G\},p) = \lim_{|V| \to \infty} R(G,p)^{1/|V|}
202: \label{r}
203: \eeq
204: %
205: where $\{G\}$ denotes the formal limit of a given family, $\lim_{|V| \to
206: \infty} G$. We calculate this function exactly for a number of families and
207: study its properties. As we shall show, this function provides a convenient
208: quantitative measure of the reliability properties of large networks.
209:
210: A third part of our work involves the calculation of the zeros of $R(G,p)$ in
211: the complex $p$ plane and the determination of their asymptotic continuous
212: accumulation set, denoted ${\cal B}$ as the number of vertices $|V| \to \infty$
213: (in practice, for our strips, this is equivalent to the limit of infinite
214: length, $L_x \to \infty$). As we shall show, the function $r(\{G\})$ is
215: nonanalytic across the locus ${\cal B}$. We define $p_c$ as the maximal point
216: where ${\cal B}$ intersects (or coincides with) the real axis. For the
217: infinite-length limits of some lattice strips, the locus ${\cal B}$ does not
218: cross the real axis, so no $p_c$ is defined. In certain of these cases, there
219: are complex-conjugate arcs on ${\cal B}$ whose endpoints at $p_{end},
220: p_{end}^*$ are very close to the real axis; in these cases, it is useful to
221: define a $(p_c)_{eff} = Re(p_{end})$. Although the width $L_y \to \infty$
222: limit of loci ${\cal B}$ for the infinite-length lattice strips is not
223: necessarily the same as the continuous accumulation locus of the zeros of the
224: reliability polynomial for the two-dimensional thermodynamic limit $L_x \to
225: \infty$, $L_y \to \infty$ with $L_y/L_x$ fixed to a finite nonzero constant,
226: our exact results on infinite-length limits of lattice strip graphs can give
227: some insight into plausible behavior of the locus ${\cal B}$ for the
228: thermodynamic limits of the corresponding 2D lattices. We shall discuss this
229: further below.
230:
231:
232: There are several motivations for our study. Clearly, new exact calculations
233: of reliability polynomials are of interest from the point of view of both the
234: statistical physics of network theory and of mathematical graph theory. The
235: generalization of $p$ from the interval $p \in [0,1]$ to $p \in {\mathbb C}$ is
236: clearly necessary in order to analyze the zeros of the reliability polynomial
237: and their accumulation set as $|V| \to \infty$. Our results also further
238: illuminate the fascinating variety of applications of the Tutte polynomial, or
239: equivalently, the Potts model partition function. Our focus here is thus
240: somewhat more abstract than in a specific engineering context, where one is
241: trying to design the most reliable network subject to various constraints, such
242: as cost.
243:
244: \section{Some General Properties of the Reliability Polynomial}
245:
246: In this section we review some general properties of the reliability polynomial
247: that will be used in the rest of the paper. As is evident from its definition,
248: the reliability polynomial has the properties that
249: %
250: \beq
251: R(G,0)=0 \ , \quad R(G,1)=1,
252: \label{rgp01}
253: \eeq
254: %
255: %
256: \beq
257: {\rm If} \ \ p \in [0,1], \quad {\rm then} \quad R(G,p) \in [0,1]
258: \label{rgpbound}
259: \eeq
260: %
261: and
262: %
263: \beq
264: \frac{dR(G,p)}{dp} \ge 0 \quad {\rm for} \quad p \in [0,1] \ .
265: \label{deriv}
266: \eeq
267: %
268:
269:
270: In general, an edge can be classified as (i) a loop, i.e., an edge that
271: connects a vertex to itself, (ii) a bridge (= co-loop) with the property that
272: if it is deleted, this increases the number of connected components of the
273: graph by one (so that, if $G$ is connected, deleting the bridge breaks $G$ into
274: two disjoint parts), or (iii) neither a loop nor a bridge. We shall use the
275: standard mathematical notation that $G/e$ means the graph obtained by deleting
276: $e$ and identifying the two vertices that were connected by $e$ (called
277: contracting $G$ on $e$) and $G-e$ means the graph obtained by deleting $e$.
278: Now if $e$ is a loop, $e=e_\ell$, then the probability that all terminals are
279: connected with each other is independent of $e_\ell$, so
280: %
281: \beq
282: R(G,p)=R(G/e_\ell,p) \ .
283: \label{reloop}
284: \eeq
285: %
286: If $e=e_b$ is a
287: bridge, then the probability that it is present is an overall multiplicative
288: factor in $R$, so
289: %
290: \beq
291: R(G,p)=pR(G/e_b,p) \ .
292: \label{rebridge}
293: \eeq
294: %
295: If $e$ is neither a loop nor a bridge, then $R(G,p)$ satisfies the
296: deletion-contraction relation
297: %
298: \beq
299: R(G,p)=(1-p)R(G-e,p)+pR(G/e,p) \ .
300: \label{delcon}
301: \eeq
302: %
303: That is, if the edge $e$ is not present, then $R(G,p)$ has the value given by
304: the reliability polynomial of a graph $G-e$ times the factor $(1-p)$; if the
305: edge is present, then $R(G,p)$ has the same value as the reliability polynomial
306: of a graph contracted on $e$, multiplied by $p$, and since these two
307: possibilities ($e$ not present or present) are mutually exclusive and exhaust
308: the possibilities, the full polynomial $R(G,p)$ is the sum of these two
309: possibilities.
310:
311: We next comment on some general properties of our $r(\{G\},p)$ function. We
312: recall that a tree graph $T_n$ is a connected graph with $n$ vertices and no
313: circuits. Since $R(T_n,p)=p^{n-1}$, it follows that $r(\{T\},p)=p$, where
314: $\{T\} \equiv \lim_{n \to \infty} T_n$. However, in general, $r$ differs from
315: $p$ for network topologies that are more complicated than that of a tree graph.
316: The $r$ function can give insight into the reliability behavior of large
317: networks. From obvious arguments, it follows that any reliability function,
318: whether the $k$-terminal reliability function $R_k(G,p)$ or the all-terminal
319: reliability $R_A(G,p)$, is a monotonically increasing function of $p$ on the
320: relevant interval $p \in [0,1]$. It follows that $r(\{G\},p)$ is also a
321: monotonically nondecreasing function of $p$ on this interval.
322:
323:
324: From the basic definition (\ref{R}) one can generalize $p$ from the physical
325: interval $0 \le p \le 1$ to arbitrary real or, indeed, complex $p$. This
326: generalization is necessary when one calculates the zeros of $R(G,p)$ in the
327: complex $p$ plane. Using our calculations of $R(G,p)$ for a variety of
328: families of lattice strip graphs, we shall determine exactly the continuous
329: accumulation set of the zeros of $R(G,p)$ in the complex $p$ plane as $|V| \to
330: \infty$ for each family of graphs $G$. We shall denote this accumulation set
331: as ${\cal B}$. In this context we note that in 1992 Brown and Colbourn
332: conjectured that for an arbitrary connected graph $G$, all of the zeros of the
333: reliability polynomial $R(G,p)$ lie in the disk $|p-1| \le 1$ \cite{bc} (some
334: related work is \cite{wagner,sbounds,hpp}). Recently, Royle and Sokal have
335: reported counterexamples that show the Brown-Colbourn conjecture to be false
336: \cite{bcf}, although the maximal values of $|p-1|$ found so far are only
337: slightly greater than unity. Extending the counterexamples found in
338: \cite{bcf}, we have also found several recursive families of graphs for
339: which the conjecture is false.
340:
341: One can also address a related question for the continuous asymptotic
342: accumulation set ${\cal B}$: Consider a recursive family of arbitrary connected
343: graphs $G = G(V,E)$ and, as above, let ${\cal B}$ be the asymptotic
344: accumulation set of zeros of $R(G,p)$ in the limit $|V| \to \infty$. Is it
345: true that all points on ${\cal B}$ are contained in the disk disk $|p-1| \le
346: 1$? Clearly, although the Brown-Colbourn conjecture is false, the answer to
347: the above question could still be yes since the zeros of $R(G,p)$ that lie
348: outside the disk $|p-1|$ might, {\it a priori}, not accumulate to form part of
349: ${\cal B}$. However, we shall show that the answer to this question is no,
350: i.e. we shall exhibit recursive families of graphs for which ${\cal B}$ extends
351: outside the disk $|p-1|$. Using our exact results, we shall address a number
352: of other interesting questions concerning ${\cal B}$. It is also of interest
353: to note that Brown and Colbourn proved that all of the real roots of the
354: reliability polynomial lie in the set $\{0\} \cup (1,2]$ \cite{bc}.
355:
356: \bigskip
357:
358: Since we shall present calculations for strip graphs with free and periodic or
359: twisted periodic longitudinal boundary conditions, we observe the following
360: inequality, where $BC_y$ refers to any given transverse boundary condition
361:
362: \begin{th} For a strip graph $G_s$ of length $L_x$ and width $L_y$ vertices,
363: %
364: \beq
365: R(G_s, L_y \times L_x, BC_y,(T)PBCx,p) \ge R(G_s, L_y \times L_x, BC_y,FBC_x,p)
366: \label{rinequality}
367: \eeq
368: \end{th}
369: %
370: \medskip
371:
372: {\bf Proof} \quad These inequalities follow because for a given type of lattice
373: strip graph $G_s$ with a given length $L_x$ and width $L_y$, the strip with
374: periodic or twisted periodic longitudinal boundary conditions has a greater
375: connectivity than strip with free longitudinal boundary conditions. In
376: particular, for a given $G_s$, if all of the transverse edges at a fixed
377: longitudinal location are absent, this does not produce two disjoint graphs for
378: a cyclic or M\"obius strip but does for the free strip. $\Box$.
379:
380: \bigskip
381:
382: \section{Relations Between the Reliability Polynomial and the Tutte Polynomial
383: and Potts Model Partition Function}
384:
385:
386: We recall that the Tutte polynomial of a graph $G=(V,E)$ is
387: \cite{tutte54}-\cite{boll}
388: %
389: \beq
390: T(G,x,y)=\sum_{G^\prime \subseteq G} (x-1)^{k(G^\prime)-k(G)}
391: (y-1)^{c(G^\prime)}
392: \label{t}
393: \eeq
394: %
395: where $k(G^\prime)$ and $c(G^\prime)$ denote the number of components and
396: linearly independent circuits of $G^\prime$, with $c(G^\prime) =
397: |E^\prime|+k(G^\prime)-|V|$. Since we deal only with connected graphs here,
398: $k(G)=1$. (We follow standard notation for the variables $x$ and $y$ of the
399: Tutte polynomial and caution the reader that these have no connection with the
400: $x$ (longitudinal) and $y$ (transverse) axes of the strip graphs.) The
401: reliability polynomial $R(G,p)$ can be expressed in terms of the Tutte
402: polynomial $T(G,x,y)$ as \cite{colbourn}
403: %
404: \beq
405: R(G,p) = p^{|V|-1}(1-p)^{|E|-|V|+1}T(G,1,\frac{1}{1-p}) \ .
406: \label{rt}
407: \eeq
408: %
409: We shall use this relation to calculate reliability polynomials for various
410: families of graphs from our previous calculations of the Tutte polynomials for
411: these graphs \cite{a}-\cite{ts}. We have also used an iterative application
412: of the deletion-contraction relation (\ref{delcon}) to calculate reliability
413: polynomials directly.
414:
415: The Tutte polynomial is equivalent to the partition function of the Potts spin
416: model. On a lattice, or more generally, on a graph $G$, at temperature $T$,
417: this model is defined by the partition function \cite{wurev,wu87}
418: %
419: \beq
420: Z(G,q,v) = \sum_{ \{ \sigma_i \} } e^{-\beta {\cal H}}
421: \label{zfun}
422: \eeq
423: with the Hamiltonian
424: %
425: \beq
426: {\cal H} = -J \sum_{\langle i j \rangle} \delta_{\sigma_i \sigma_j}
427: \label{ham}
428: \eeq
429: %
430: where $\sigma_i=1,...,q$ are the spin variables on each vertex $i \in G$;
431: $\beta = (k_BT)^{-1}$; $\langle i j \rangle$ denotes pairs of adjacent
432: vertices; and we use the notation
433: %
434: \beq
435: K = \beta J \ , \quad a = e^K \ , \quad v = a-1 \ .
436: \label{kvdef}
437: \eeq
438: %
439: This partition function can be written as \cite{fk}
440: %
441: \beq
442: Z(G,q,v) = \sum_{G^\prime \subseteq G} q^{k(G^\prime)}v^{e(G^\prime)} \ .
443: \label{cluster}
444: \eeq
445: %
446: This formula enables one to generalize $q$ from ${\mathbb Z}_+$ to ${\mathbb
447: R}$ or, indeed, ${\mathbb C}$. From it one also directly deduces the
448: equivalence
449: %
450: \beq
451: Z(G,q,v)=(x-1)^{k(G)}(y-1)^{|V|}T(G,x,y)
452: \label{zt}
453: \eeq
454: %
455: where
456: %
457: \beq
458: x=1+\frac{q}{v}
459: \label{xqv}
460: \eeq
461: %
462: and
463: %
464: \beq
465: y=v+1=e^K
466: \label{yqv}
467: \eeq
468: %
469: so that
470: %
471: \beq
472: q=(x-1)(y-1) \ .
473: \label{qxy}
474: \eeq
475: %
476: Combining (\ref{rt}) and (\ref{zt}), we have, for a connected graph $G$, the
477: relation between the reliability polynomial and the Potts model partition
478: function
479: %
480: \beq
481: R(G,p) = (1-p)^{|E|} \ \lim_{q \to 0} \ q^{-1}Z(G,q,v=\frac{p}{1-p}) \ .
482: \label{rz}
483: \eeq
484: %
485: Note that $Z(G,q,v)$ always has an overall factor of $q$, which cancels the
486: factor of $q^{-1}$ in (\ref{rz}). The valuation $y=1/(1-p)$ in (\ref{rt}), or
487: equivalently, $v=p/(1-p)$ in (\ref{rz}) can also be expressed as
488: %
489: \beq
490: p=1-e^{-K} = \frac{v}{1+v}\ .
491: \label{pek}
492: \eeq
493: %
494: Thus the physical range of $p \in [0,1]$ for the network corresponds to the
495: physical range of temperature for the $q=0$ Potts ferromagnet, with $T=0 \
496: \Leftrightarrow \ p=1$ and $T=\infty \ \Leftrightarrow \ p=0$. The physical
497: range of temperature for the Potts antiferromagnet would correspond to the
498: unphysical interval $-\infty \le p \le 0$ for the network.
499:
500: \section{Structural Results}
501:
502: \subsection{General Recursive Families of Graphs}
503:
504: A general form for the Tutte polynomial for the strip graphs considered here,
505: or more generally, for recursively defined families of graphs $G_m$ comprised
506: of $m$ repeated subgraph units, is \cite{a}
507: %
508: \beq T(G_m,q,v) =
509: \frac{1}{x-1}\sum_{j=1}^{N_{T,G,\lambda}} c_{G,j}(\lambda_{G,j})^m
510: \label{tgsum}
511: \eeq
512: %
513: where the terms $\lambda_{G,j}$, the coefficients $c_{G,j}$, and the total
514: number $N_{Z,G,\lambda}$ depend on $G$ through the type of lattice, its width,
515: $L_y$, and the boundary conditions, but not on the length.
516:
517: We observe the following general result:
518:
519: \begin{th} \quad Consider a recursive family of graphs $G_m$ comprised of $m$
520: repeated subgraph units. Then the reliability polynomial $R(G_m,p)$ has the
521: general form
522: %
523: \beq
524: R(G_m,p) = \sum_{j=1}^{N_{R,G,\mu}} c_{R,G,j} (\mu_{G,j})^m
525: \label{rgsum}
526: \eeq
527: \end{th}
528: %
529: where the coefficients $c_{R,G,j}$ and terms $\mu_{G,j}$'s depend on
530: $G$ through the type of lattice, its width, $L_y$, and the boundary conditions,
531: but not on the length.
532:
533: \medskip
534:
535: {\bf Proof} \quad This is proved by combining (\ref{rt}) and (\ref{tgsum}).
536: Clearly, the $\mu_{G,j}$'s and $c_{R,G,j}$'s can be determined from the
537: $\lambda_{G,j}$'s and $c_{G,j}$'s. $\Box$ The reduction of the form
538: (\ref{tgsum}) to (\ref{rgsum}) is analogous to the reduction that we had
539: found for chromatic polynomials of recursive families of graphs
540: \cite{w}-\cite{k}.
541:
542: \medskip
543:
544: It is convenient to factor out a power of $p$ and write (\ref{rgsum}) as
545: %
546: \beq
547: R(G_m,p) = p^{a_{1,G}m + a_{0,G}}
548: \sum_{j=1}^{N_{R,G,\alpha}} c_{R,G,j} (\alpha_{G,j})^m
549: \label{rgsumfactored}
550: \eeq
551: %
552: with
553: %
554: \
555: %
556: \beq
557: \alpha_{G,j}=p^{-a_{1,G}}\mu_{G,j}
558: \label{betaalpha}
559: \eeq
560: %
561: where again the $a_{i,G}$, $i=0,1$, depend on $G$ through the type of lattice,
562: its width, $L_y$, and the boundary conditions, but not on the length.
563: Obviously, $N_{R,G,\alpha}=N_{R,G,\mu}$. We find that for $p=1$, for all of
564: the families that we have considered, all except one of the $N_{R,G,\alpha}$
565: $\alpha_{G,j}$'s vanish, and the nonzero term, which can be taken to be the
566: first, has the value unity:
567: %
568: \
569: \beq
570: \alpha_{G,1}=1 \ , \quad \alpha_{G,j}=0 \ , \quad 2 \le j \le N_{R,G,\alpha}
571: \quad {\rm for} \ \ p=1
572: \label{alphap1}
573: \eeq
574: %
575: For the strips of the square and triangular lattices considered here,
576: %
577: \beq
578: a_{1,G}=L_y \ , \quad G \ \ {\rm of} \ \ sq, t \ \ {\rm type}
579: \label{a1gsqtri}
580: \eeq
581: %
582: and for the strips of the honeycomb lattice,
583: %
584: \beq
585: a_{1,G}=2L_y \ , \quad G \ \ {\rm of} \ \ hc \ \ {\rm type}
586: \label{a1ghc}
587: \eeq
588: %
589: Note that for all of the strips of the square and triangular lattice with
590: periodic or twisted periodic longitudinal boundary conditions, $m=L_x$.
591: For the corresponding strips with free longitudinal boundary conditions, it is
592: convenient to set $m=L_x-1$; with this definition, it follows that for strips
593: with both free and periodic (or twisted periodic) longitudinal boundary
594: conditions, $m$ is equal to length measured in units of edges, with $L_x$
595: being the length in terms of vertices. The situation with the honeycomb strips
596: is different. For example, for a cyclic strip of the honeycomb lattice,
597: envisioned as a brick lattice, $L_x=2m$; that is, if the strip is $m$ bricks
598: long, then the length, measured in terms of vertices, is $2m$.
599:
600:
601: Following our earlier nomenclature \cite{w}, we denote a term $\alpha_{G,j}$ in
602: (\ref{rgsumfactored}) as leading (= dominant) if it has a magnitude greater
603: than or equal to the magnitude of other $\alpha_{G,j^\prime}$'s. In the limit
604: $n \to \infty$ the leading $\alpha_{G,j}$ in $R(G,p)$ determines the function
605: $r(\{G\},p)$. The continuous locus ${\cal B}$ where $r(\{G\},p)$ is
606: nonanalytic thus occurs where there is a switching of dominant $\alpha_{G,j}$'s
607: in $R$, respectively, and is the solution of the equation of degeneracy
608: in magnitude
609: %
610: \beq
611: |\alpha_{G,j}|=|\alpha_{G,j^\prime}|
612: \label{degeneq}
613: \eeq
614: %
615: between two dominant $\alpha$'s.
616:
617: \subsection{Cyclic Strips of Regular Lattices}
618:
619: Here we will derive some structural properties of reliability polynomials for
620: cyclic strips of regular lattices from corresponding properties of Tutte
621: polynomials, using eq. (\ref{rt}). In \cite{cf} it was shown that for cyclic
622: and M\"obius strips of the square lattice of fixed width $L_y$ and arbitrary
623: length $L_x$ (and also for cyclic strips of the triangular lattice) the
624: coefficients $c_j$ in the Tutte polynomial are polynomials in $q$ with the
625: property that for each degree $d$ there is a unique polynomial, denoted
626: $c^{(d)}$. Further, this was shown to be
627: %
628: \beq
629: c^{(d)} = U_{2d}(q^{1/2}/2) = \sum_{j=0}^d (-1)^j {2d-j \choose j}
630: q^{d-j}
631: \label{cd}
632: \eeq
633: %
634: where $U_n(x)$ is the Chebyshev polynomial of the second kind. A number of
635: properties of these coefficients were derived in \cite{cf}. The first few of
636: these coefficients are
637: %
638: \beq
639: c^{(0)}=1 \ , \quad c^{(1)}=q-1 \ , \quad c^{(2)}=q^2-3q+1 \ ,
640: \label{cd012}
641: \eeq
642: %
643: \beq
644: c^{(3)}=q^3-5q^2+6q-1 \ .
645: \label{cd3}
646: \eeq
647: %
648: Following our earlier work \cite{dg}, we define
649: %
650: \beqs
651: \kappa^{(d)} & = & c^{(d)} + c^{(d-1)} \cr\cr
652: & = & \sum_{j=0}^{d-1} (-1)^j { 2d-1-j \choose j} q^{d-j}
653: \label{kappad}
654: \eeqs
655: %
656: and
657: %
658: \beq
659: \bar\kappa^{(d)}=q^{-1}\kappa^{(d)}=
660: \sum_{j=0}^{d-1} (-1)^j{2d-1-j \choose j} q^{d-1-j} \ .
661: \label{kappabar}
662: \eeq
663: %
664: An important property of $\kappa^{(d)}$ is that it always has
665: the factor $q$; this is evident from (\ref{kappad}). The first few of these
666: polynomials are
667: %
668: \beq
669: \kappa^{(1)}=q \ , \quad \kappa^{(2)}=q(q-2) \ , \quad
670: \kappa^{(3)}=q(q-1)(q-3) \ .
671: \label{kapp123}
672: \eeq
673: %
674: The following special results for $q=0$, i.e., $x=1$, will be useful:
675: %
676: \beq
677: \lim_{q \to 0} \bar\kappa^{(d)} = (-1)^{d+1}d
678: \label{kappabarq0}
679: \eeq
680: %
681: \beq
682: c^{(d)} = (-1)^d \quad {\rm for} \quad q=0
683: \label{cdq0}
684: \eeq
685: %
686: and
687: %
688: \beq
689: \frac{dc^{(d)}}{dq} = (-1)^{d-1}\frac{(d+1)d}{2} \quad {\rm for} \ \ q=0
690: \label{diffcdq0}
691: \eeq
692: %
693: so that, using (\ref{qxy}) to express $c^{(d)}$ as a function of $x$ and $y$,
694: %
695: \beq
696: \frac{\partial c^{(d)}}{\partial x} = (-1)^{d-1}(y-1)\frac{(d+1)d}{2}
697: \quad {\rm for} \ \ x=1 \ .
698: \label{diffcdx1}
699: \eeq
700: %
701:
702: The terms $\lambda_{T,G_s,L_y,j}$ that occur in (\ref{tgsum}) can be classified
703: into sets, with the $\lambda_{T,G_s,L_y,j}(x,y)$ in the $d$'th set being
704: defined as those terms with coefficient $c^{(d)}$. In Ref. \cite{cf} the
705: numbers of such terms, denoted $n_T(L_y,d)$, were calculated. Labelling the
706: eigenvalues with coefficient $c^{(d)}$ as $\lambda_{T,G_s,L_y,d,j}$ with $1 \le
707: j \le n_T(L_y,d)$, the Tutte polynomial for a strip graph of length $L_x=m$ of
708: a regular lattice of type $G_s$ can be written in the form
709: %
710: \beq
711: T[G_s(L_y \times m; FBC_y,PBC_y),x,y] = \frac{1}{x-1}\sum_{d=0}^{L_y}
712: c^{(d)} \sum_{j=1}^{n_T(L_y,d)} (\lambda_{T,G_s,L_y,d,j})^m \ .
713: \label{tgsumcyc}
714: \eeq
715: %
716: The property that $T(G,x,y)$ is a polynomial in $x$, which is evident from its
717: definition (\ref{t}), is not manifest in (\ref{tgsumcyc}) because of the
718: $1/(x-1)$ prefactor; however, this prefactor is always cancelled in the
719: evaluation of the sum on the right-hand side of (\ref{tgsumcyc}) for any
720: particular type of lattice strip graph $G_m$. Nevertheless, the presence
721: of this prefactor means that if one calculates the reliability polynomial for a
722: cyclic strip graph by evaluating the Tutte polynomial for this graph at $x=1$,
723: one encounters an expression of the form $0/0$ and must use L'Hopital's rule to
724: evaluate it. Thus, differentiating numerator and the denominator $1/(x-1)$
725: from the prefactor and using eqs. (\ref{cdq0}) and (\ref{diffcdx1}), we have
726: %
727: \beqs
728: & & T[(G_s(L_y \times m; FBC_y,PBC_y),x=1,y] = \sum_{d=1}^{L_y} (-1)^{d-1}(y-1)
729: \frac{(d+1)d}{2}\sum_{j=1}^{n_T(L_y,d)}(\lambda_{T,G_s,L_y,d,j})^m \cr\cr
730: & & + m\sum_{d=0}^{L_y} (-1)^d \sum_{j=1}^{n_T(L_y,d)}
731: (\lambda_{T,G_s,L_y,d,j})^{m-1} \frac{\partial \lambda_{T,G_s,L_y,d,j}}
732: {\partial x}
733: \label{tgsumcycxeq1}
734: \eeqs
735: %
736: where $G_s$ denotes the type of lattice and the right-hand side is evaluated at
737: $x=1$. This provides one way of understanding the origin of certain factor of
738: $m$ in the exact results that we shall give below for reliability polynomials
739: of strip graphs of length $m$.
740:
741:
742: The number $n_T(L_y,d)$ of $\lambda$'s with a given coefficient $c^{(d)}$ is
743: \cite{a}
744: %
745: \beq
746: n_T(L_y,d)=\frac{(2d+1)}{(L_y+d+1)}{2L_y \choose L_y-d} \quad {\rm for} \ \
747: 0 \le d \le L_y
748: \label{ntlyd}
749: \eeq
750: %
751: and zero otherwise. The total number $N_{T,L_y,\lambda}$ of different terms
752: $\lambda_{T,L_y,j}$ in eq. (\ref{tgsum}) is \cite{cf}
753: %
754: \beq
755: N_{T,L_y,\lambda}=\sum_{d=0}^{L_y} n_T(L_y,d) \ .
756: \label{nttot}
757: \eeq
758: %
759: For cyclic and M\"obius strips of the square, triangular, and honeycomb
760: lattices, we calculated this to be \cite{cf,hca}
761: %
762: \beq
763: N_{T,L_y,\lambda}={2L_y \choose L_y} \ .
764: \label{nttotcyc}
765: \eeq
766: %
767:
768: For cyclic and M\"obius strips of regular lattices $G_s$ with arbitrary
769: $L_y$, eq. (\ref{ntlyd}) shows that there is a unique
770: $\lambda_{T,G_s,L_y,d}$ corresponding to the coefficient $c^{(d)}$ of
771: highest degree $d$, namely $d=L_y$. We have found \cite{a,ta,hca} that this
772: term is
773: %
774: \beq
775: \lambda_{T,G_s,L_y,d=L_y} = 1 \ .
776: \label{lamtutlyly}
777: \eeq
778: %
779: As indicated, since this eigenvalue is unique, it is not necessary to append a
780: third index, as with the other $\lambda$'s, and we avoid this for simplicity.
781: For each of the cyclic strips of regular lattices $G_s$ considered here,
782: one can derive a general formula for $\alpha_{G_s,L_y,L_y}$ from (\ref{rt})
783: in conjunction with (\ref{lamtutlyly}), viz.,
784: %
785: \beq
786: \alpha_{G_s,L_y,L_y}=(1-p)^{\rho_{G_s}}
787: \label{alphalyly}
788: \eeq
789: %
790: where
791: %
792: \beq
793: \rho_{G_s} = \frac{|E|-|V|}{m} \ .
794: \label{rho}
795: \eeq
796: %
797: For example, for the cyclic and M\"obius strips of the square (sq) lattice of
798: width $L_y$ vertices and length $L_x=m$ vertices, $|V|=L_ym$
799: and $|E|=(2L_y-1)m$, so that $\rho_{sq}=L_y-1$; for the corresponding strips
800: of the triangular (t) lattice, $|V|$ is the same and $|E|=(3L_y-2)m$, so that
801: $\rho_t=2(L_y-1)$. For the cyclic strip of the honeycomb (hc) lattice of width
802: $L_y$ vertices and length $L_x=2m$ vertices, we have $|V|=2L_ym$ and
803: $|E|=(3L_y-1)m$ so that $\rho_{hc}=L_y-1$.
804:
805: We next have
806:
807: \begin{th} \quad Consider the reliability polynomial for cyclic strips of the
808: square, triangular, or honeycomb lattices, denoted generically $G_s$, of
809: width $L_y$ and arbitrary length $L_x$. The total number of different terms in
810: (\ref{rgsum}) satisfies the inequality $N_{R,G_s,L_y,\alpha} \le
811: (1/2)N_{T,G_s,L_y,\lambda}$ and hence
812: %
813: \beq
814: N_{R,G_s,L_y,\alpha} \le \frac{1}{2}{2L_y \choose L_y} \ .
815: \label{nrtot}
816: \eeq
817: \label{ncyclic}
818: \end{th}
819: %
820:
821: \medskip
822:
823: {\bf Proof} \quad The proof makes use of the expression (\ref{rt}) of the
824: reliability polynomial $R(G_m,p)$ as a special case of the Tutte polynomial
825: $T(G_m,x,y)$ for $x=1$ and $y=1/(1-p)$, together with our earlier result
826: (\ref{tgsumcyc}) for the Tutte polynomial of a cyclic strip graph of the
827: square, triangular, or honeycomb lattice. Because the prefactor $1/(x-1)$ in
828: (\ref{tgsumcyc}) is singular at $x=1$ where one evaluates the Tutte polynomial,
829: it is necessary that
830: %
831: \beq
832: \left [ \sum_{d=0}^{L_y}c^{(d)} \sum_{j=1}^{n_T(L_y,d)}
833: (\lambda_{T,G_s,L_y,d,j})^m \right ]_{x=1} = 0
834: \label{numzero}
835: \eeq
836: %
837: This vanishing condition holds for arbitrary $m$, which means that it implies
838: pairwise relations among various terms $\lambda_{T,G_s,L_y,d,j}$. The only
839: such relations that can yield an overall factor of $(x-1)$ are relations of the
840: form $\lambda_{T,G_s,L_y,d,j}=\lambda_{T,G_s,L_y,d-\ell,j^\prime}$ for $1 \le d
841: \le L_y$ and odd $\ell$ in the range $1 \le \ell \le d$ for a given $d$,
842: since these yield expressions of the form
843: %
844: \beqs
845: c^{(d)}(\lambda_{T,G_s,L_y,d,j})^m +
846: c^{(d-\ell)}(\lambda_{T,G_s,L_y,d-\ell,j^\prime})^m
847: & = &
848: (c^{(d)}+c^{(d-\ell)})(\lambda_{T,G_s,L_y,d,j})^m
849: \label{cdcombination}
850: \eeqs
851: %
852: Using (\ref{cdq0}), one sees that since $\ell$ is odd, the combination
853: $c^{(d)}+c^{(d-\ell)}$ vanishes at $q=0$ and hence has a factor of $q$.
854: Using eq. (\ref{qxy}), one sees that this is sufficient to get an
855: overall factor of $x-1$ so as to yield the vanishing in eq. (\ref{numzero}) at
856: $x=1$ or equivalently to cancel the prefactor $1/(x-1)$. Because of the
857: pairwise equalities connecting each of the $\lambda$'s when evaluated at $x=1$,
858: the total number of these terms in the reliability polynomial is reduced to
859: half of the number for the Tutte polynomial, eq. (\ref{nttot}). Further
860: equalities are, in principle, possible and would further reduce the number of
861: different $\lambda$'s. This completes the proof of the theorem. $\Box$
862:
863: \medskip
864:
865: Two remarks are in order here. First, we note that in all of our calculations,
866: we find $\ell=1$; that is, the pairwise equalities among the $\lambda$'s occur
867: for adjacent values of $d$. Hence, the linear combination in
868: eq. (\ref{cdcombination}) is $c^{(d)}+c^{(d-\ell)}=\kappa^{(d)}$, where
869: $\kappa^{(d)}$ was given in (\ref{kappad}). Second, in all of our
870: calculations, we have observed that the inequality (\ref{nrtot}) is realized
871: as an equality. Let us illustrate how the pairwise equalities work in some
872: special cases. For $L_y=2$, we have $n_T(2,1)=3$ and $n_T(2,0)=2$. At $x=1$,
873: the single $\lambda_{T,2,2}$ is equal to one of the three $\lambda_{T,2,1,j}$,
874: which we denote $\lambda_{T,2,1,3}$. The other two of these three
875: $\lambda_{T,2,1,j}$'s are equal to the two respective $\lambda_{T,2,0,j}$'s:
876: $\lambda_{T,2,1,j}=\lambda_{T,2,0,j}$ for $j=1,2$. For $L_y=3$, we have
877: $n_T(3,3)=1$, $n_T(3,2)=5$, $n_T(3,1)=9$, and $n_T(3,0)=5$. At $x=1$,
878: $\lambda_{T,3,3}$ is equal to one of the five $\lambda_{T,3,2,j}$'s while the
879: other four are equal to four of the nine $\lambda_{T,3,1,j}$'s, and finally,
880: the remaining five of these $\lambda_{T,3,1,j}$'s are equal, respectively, to
881: the five $\lambda_{T,3,0,j}$'s.
882:
883: \subsection{Cyclic Strips of the Square Lattice with Self-Dual Boundary
884: Conditions}
885:
886: We have previously calculated the Tutte polynomial for cyclic strips of the
887: square lattice with self-dual boundary conditions \cite{jz,dg,sdg}. These
888: strip graphs have (i) a fixed transverse width $L_y$, (ii) arbitrarily great
889: length $L_x$, (iii) periodic longitudinal boundary conditions, and (iv) are
890: such that each vertex on one side of the strip, which we take to be the upper
891: side (with the strip oriented so that the longitudinal, $x$ direction is
892: horizontal) is joined by edges to a single external vertex. A strip graph of
893: this type will be denoted generically as $G_D$ (where the subscript $D$ refers
894: to the self-duality) and, when its size is indicated, as $G_D(L_y \times L_x)$.
895: For cyclic strips with self-dual boundary conditions we determined the general
896: structure of the Tutte polynomial in \cite{dg}. We showed that the
897: coefficients were precisely the $\kappa^{(d)}$ polynomials. The general form of
898: the Tutte polynomial is \cite{dg,sdg}
899: %
900: \beq
901: T(G_D[L_y \times m,cyc.],x,y) = \sum_{d=1}^{L_y+1} \bar\kappa^{(d)}
902: \sum_{j=1}^{n_T(G_D,L_y,d)} (\lambda_{T,G_D,L_y,d,j})^m
903: \label{tgsumcycsdg}
904: \eeq
905: %
906: where $\bar\kappa^{(d)}$ was defined in eq. (\ref{kappabar}) and
907: %
908: \beq
909: n_T(L_y,d)=\frac{2d}{L_y+d+1}{2L_y+1 \choose L_y-d+1}
910: \label{ntlydgd}
911: \eeq
912: %
913: so that for the total number of terms \cite{dg},
914: %
915: \beq
916: N_{T,G_D,L_y,\lambda} = { 2L_y+1 \choose L_y+1} \ .
917: \label{nttotgd}
918: \eeq
919: %
920: Note that, in contrast to the situation with the cyclic strips, because the
921: Tutte polynomial does not contain any prefactor of $1/(x-1)$, there is no need
922: to use L'Hopital's rule in evaluating this polynomial to get the reliability
923: polynomial. Related to this, the coefficients are $\bar\kappa^{(d)}$, i.e., do
924: not contain the factor of $q$ that the $\kappa^{(d)}$ coefficients do. In this
925: evaluation at $x=1$, i.e., $q=0$, the $\bar\kappa^{(d)}$ simply reduce to the
926: constants $(-1)^{d+1}d$. The reliability polynomial for the cyclic self-dual
927: strip, $R(G_D[L_y \times m],p]$, thus contrasts with the reliability polynomial
928: for the cyclic strip graphs in not having any terms proportional to $m$, a
929: consequence of the fact that one did not have to use L'Hopital's rule in
930: evaluating the corresponding Tutte polynomial at $x=1$.
931:
932:
933: Our next result is
934: \begin{th} The reliability polynomial of the self-dual strip graph
935: $G_D[L_y \times L_x]$ is (with $L_x = m$)
936: %
937: \beq
938: R(G_D[L_y \times m],p) = p^{L_ym} \sum_{d=1}^{L_y+1} (-1)^{d+1}d
939: \sum_{j=1}^{n_T(G_D,L_y,d)}(\alpha_{G_D,L_y,d,j})^m
940: \label{rgd}
941: \eeq
942: %
943: {\rm where}
944: %
945: \beq
946: \alpha_{G_D,L_y,d,j}=(1-p)^{L_y} \left . \lambda_{T,G_D,L_y,d,j} \right |_{
947: x=1,y=1/(1-p)} \ .
948: \label{alpha_lambda_gd}
949: \eeq
950: %
951: \end{th}
952:
953: \medskip
954:
955: {\bf Proof} \quad This result follows directly from (\ref{tgsumcycsdg}) and
956: (\ref{rt}), using (\ref{kappabarq0}) and the fact that the number of vertices
957: and edges for a graph in this family are $|V|=L_yL_x+1$ and $|E|=2L_yL_x$.
958: $\Box$
959:
960: \medskip
961:
962: From our explicit calculation of $R(G_D[L_y \times m],p)$ for several values
963: of the width $L_y$, we find that there is no reduction in the total number of
964: $\lambda$'s when one carries out the evaluation at $x=1$ for eq. (\ref{rgd}).
965: Hence, for the cases that we have studied, this total number of $\lambda$'s in
966: the reliability polynomial for the self-dual cyclic strips of the square
967: lattice is the same as the number for the Tutte polynomial (Potts model
968: partition function) for these graphs. Specifically, this is
969: $N_{R,G_D,L_y,\lambda}=3, 10, 35$ for $L_y=1,2,3$, respectively.
970:
971:
972: \subsection{Strips with Free Longitudinal Boundary Conditions}
973:
974:
975: As before for chromatic and Tutte polynomials, it is convenient to express the
976: results in terms of a generating function. For a strip graph of type $G_s$
977: (where this includes the specification of the transverse boundary conditions)
978: and length $m$ longitudinal edges, we have
979: %
980: \beq
981: \Gamma(G_s,p,z) = \sum_{m=0}^\infty R((G_s)_m,p)z^m
982: \label{gamma}
983: \eeq
984: %
985: where
986: %
987: \beq
988: \Gamma(G_s,p,z) = \frac{ {\cal N}(G_s,p,z)}{{\cal D}(G_s,p,z)} \ .
989: \label{gammazcalc}
990: \eeq
991: The numerator and denominator are rational functions of $z$ and $p$ of the form
992: %
993: \beq
994: {\cal N}(G_s,p,z) = \sum_{j=1}^{{\rm deg}_z({\cal N})} A_{G_s,j} z^j
995: \label{n}
996: \eeq
997: %
998: \beq
999: {\cal D}(G_s,p,z) = 1 + \sum_{j=1}^{{\rm deg}_z({\cal D})} b_{G_s,j} z^j \ .
1000: \label{d}
1001: \eeq
1002: %
1003: The property (\ref{alphap1}) is manifested in the relations
1004: %
1005: \beq
1006: b_{G_s,1}=-1 \ , \quad b_{G_s,j}=0 \ , 2 \le j \le N_{G_s,\alpha} \quad
1007: {\rm for} \ \ p=1
1008: \label{bjp1}
1009: \eeq
1010: %
1011: For the strips of the square and triangular lattices with free longitudinal
1012: boundary conditions,
1013: %
1014: \beq
1015: A_{G_s,0}=R(T_{L_y},p)= p^{L_y-1} \quad {\rm for} \quad G_s=sq,t
1016: \label{nstrip}
1017: \eeq
1018: %
1019:
1020: \subsection{Accumulation Set of Zeros of $R(G,p)$}
1021:
1022: Concerning the continuous accumulation set ${\cal B}$ of the zeros of $R(G,p)$
1023: for cyclic and M\"obius lattice strip graphs in the limit $L_x \to \infty$, we
1024: note
1025:
1026: \begin{th} \quad
1027: Consider the lattice strip graphs $G[L_y,L_x,cyc]$ and $G[L_y,L_x,Mb]$. The
1028: continuous accumulation set ${\cal B}$ defined in the limit $L_x \to \infty$
1029: is the same for both of these.
1030: \label{Bcycmb}
1031: \end{th}
1032:
1033: {\bf Proof} \quad This is a corollary of earlier theorems that we have proved
1034: for the accumulation set of the zeros of the Tutte polynomial or equivalent
1035: Potts model partition function, stating that these are the same for a given
1036: strip with cyclic or M\"obius longitudinal boundary conditions.
1037: $\Box$
1038:
1039: \medskip
1040:
1041: We next discuss the construction of families of graphs $G$ for which the
1042: reliability polynomial $R(G,p)$ has zeros that extend to the circle $|p-1|=1$.
1043: A useful lemma can be stated as follows \cite{colbourn,bc}:
1044: Let $G=(V,E)$ be a connected graph with no multiple edges. We also assume that
1045: $G$ has no loops; this assumption incurs no loss of generality since loops do
1046: not affect the reliability polynomial. Now let $G_\ell$ be the graph obtained
1047: from $G$ by replacing each edge by $\ell$ edges connecting the same pair of
1048: vertices. Then
1049: %
1050: \beq
1051: R(G_\ell,p) = R(G,p^\prime)
1052: \label{rgell}
1053: \eeq
1054: %
1055: where
1056: %
1057: \beq
1058: p^\prime = 1-(1-p)^\ell \ .
1059: \label{pprime}
1060: \eeq
1061: %
1062: In addition to the proofs in \cite{colbourn,bc}, this result can be understood
1063: easily from a physics viewpoint, using the expression of the reliability
1064: polynomial as a special case of the Potts model partition function. Replacing
1065: every edge in $G$ by $\ell$ edges connecting the same vertices has the effect
1066: of replacing the spin-spin coupling $J$ by $J^\prime = \ell J$ in the Potts
1067: model Hamiltonian ${\cal H}$, and hence $K \to K^\prime = \ell K$, so that,
1068: since $y=e^K$ (cf. eq. (\ref{yqv})),
1069: %
1070: \beq
1071: y \to y^\prime = y^\ell, \ \ i.e., \ \ v \to v^\prime = (v+1)^\ell-1
1072: \label{yell}
1073: \eeq
1074: %
1075: and
1076: %
1077: \beq
1078: Z(G_\ell,q,v)=Z(G,q,v^\prime)
1079: \label{zgell}
1080: \eeq
1081: %
1082: The result (\ref{rgell}) then follows via (\ref{rz}) (or (\ref{rt})).
1083:
1084: As noted in \cite{colbourn,bc}, using this result, one can generate families of
1085: graphs, namely those with replicated edges, that extend to $|p-1|=1$.
1086: The relation (\ref{zgell}) is equivalent to
1087: %
1088: \beq
1089: T(G_\ell,x,y) = \left ( \frac{y^\ell - 1}{y-1} \right )^{n-1}
1090: T(G,x^\prime,y^\prime)
1091: \label{tgell}
1092: \eeq
1093: %
1094: where $y^\prime$ was given in (\ref{yell}) and
1095: %
1096: \beq
1097: x^\prime = 1 + \frac{(x-1)(y-1)}{y^\ell - 1}
1098: \label{xell}
1099: \eeq
1100: %
1101: Now let $G=(V,E)$ be a simple graph (i.e. a graph without loops or multiple
1102: edges) that is connected and has at least one edge. As before, the exclusion
1103: of loops does not incur any loss of generality since loops have no effect on
1104: the reliability polynomial. Now let $G_\ell$ be the graph obtained from $G$ by
1105: replacing each edge by $\ell$ edges connecting the same pair of vertices. Then
1106: the zeros of $R(G_\ell,p)$ in the complex $p$ plane include the following:
1107: %
1108: \beq
1109: p = 1 - e^{\frac{2\pi i k}{\ell}} \ , \quad k=0,1,..., \ell-1
1110: \label{pzero_t2}
1111: \eeq
1112: %
1113: and these satisfy $|p-1| =1$. This follows by noting that in general, for the
1114: specified $G$ with at least one edge, $R(G,p)$ contains a factor $p^{|V|-1}$
1115: and hence $G_\ell$ contains a factor of $(p^\prime)^{|V|-1}$. Therefore
1116: $R(G_\ell,p)$ has at least one zero at $p^\prime =0$. Recalling the relation
1117: (\ref{pprime}) and solving for the $\ell$ roots of this equation in the $p$
1118: plane, we obtain (\ref{pzero_t2}), which satifies $|p-1|=1$.
1119:
1120: Let us give an example. Perhaps the simplest is to consider the graph
1121: $G=T_2$. Replacing the single edge in this graph by $\ell$ edges yields a graph
1122: commonly denoted the ``thick link'' or ``fat link'', $TL_\ell$. This is the
1123: (planar) dual to the circuit graph $C_\ell$; i.e., $TL_\ell = C_\ell^*$. From
1124: the elementary result $R(T_2,p)=p$ and our theorem (\ref{rgell}), one gets
1125: $R(FL_\ell,p)=p^\prime$, which has zeros at the points (\ref{pzero_t2}). The
1126: location of these zeros for $\ell=(a) 2, (b) 3, and (c) 4$ are (a) $p=0,2$; (b)
1127: $p=0$ and $p=3/2 \pm (i/2)\sqrt{3}$; (c) $p=0,2$, and $1 \pm i$. In the limit
1128: $\ell \to \infty$, one then obtains an accumulation set ${\cal B}$ that is the
1129: circle $|p-1|=1$. Note that in this example, although the number of edges
1130: goes to infinity, the number of vertices is fixed at 2. One can also consider
1131: families of graphs with a subset of the vertices replicated. The family of
1132: $L_y=2$ strips of the square lattice with cylindrical or torus boundary
1133: conditions are of this type, with double vertical and single horizontal edges.
1134: These families will be analyzed below.
1135:
1136:
1137: \section{Calculations for Specific Families of Graphs}
1138:
1139: \subsection{$L_y=1$ Strips of the Square Lattice}
1140:
1141: The reliability function for the free $L_y=1$ strip is a special case of the
1142: result for a tree graph $T_m$ with $m$ vertices, namely $R=p^{m-1}$.
1143: For the case of cyclic boundary conditions, the strip is the circuit graph,
1144: $C_m$. By an elementary application of the deletion-contraction theorem, one
1145: has
1146: %
1147: \beq
1148: R(C_m,p) = p^{m-1}[m(1-p)+p] \ .
1149: \label{Rcn}
1150: \eeq
1151: %
1152: Hence,
1153: %
1154: \beq
1155: r=p
1156: \label{rsqxy1}
1157: \eeq
1158: for the infinite-length limits of both the free and cyclic $L_y=1$ strips.
1159: The polynomial $R(C_m,p)$ has a zero of multiplicity $m-1$ at $p=0$ and another
1160: zero at $p=m/(m-1)$, which monotonically approaches $p=1$ from above as $m \to
1161: \infty$. The $r$ function (\ref{rsqxy1}) for these $L_y=1$ strips has the
1162: special property that $dr/dp=1$. As will be seen, the $r$ functions that we
1163: shall calculate for other strips have various nonzero slopes at $p=0$ but, in
1164: contrast to (\ref{rsqxy1}), they have the derivative $dr/dp=0$ at $p=1$.
1165:
1166: \bigskip
1167:
1168:
1169: \subsection{$L_y=2$ Strip of the Square Lattice with Free Boundary Conditions}
1170:
1171: The generating function for this strip is of the form (\ref{gammazcalc}) with
1172: ${\cal N}=p$ and
1173: %
1174: \beqs
1175: {\cal D}(p,z) & = & 1-p^2(4-3p)z+p^4(1-p)z^2 \cr\cr
1176: & = & (1-p^2 \alpha_{sq,2,0,1}z)(1-p^2 \alpha_{sq,2,0,2}z)
1177: \label{dengamma}
1178: \eeqs
1179: %
1180: where
1181: %
1182: \beq
1183: \alpha_{sq,2,0,j} = \frac{1}{2}\left [ 4-3p \pm \sqrt{R_2} \ \right ]
1184: \ , \quad j=1,2
1185: \label{alpha20j}
1186: \eeq
1187: %
1188: with
1189: %
1190: \beq
1191: R_2=12-20p+9p^2 \ .
1192: \label{r2}
1193: \eeq
1194: %
1195: The subscripts in the notation make reference to the structure of the
1196: $\alpha$'s for the cyclic strip; see below. We observe that $R(sq[2 \times
1197: m],p)$ has the factor $(4-3p)$ if $m \ge 1$ is odd. For the cases that we have
1198: examined, besides the real root at $p=0$, this reliability polynomial has a
1199: single real root at $p=4/3$ if and only if $m \ge 1$ is odd.
1200:
1201: The dominant term in the physical interval $p \in [0,1]$ is
1202: $\alpha_{sq,2,0,1}$, and hence the asymptotic reliability per vertex for $L_x
1203: \to \infty$ is
1204: %
1205: \beqs
1206: r(sq[2 \times \infty, free],p) & = & p(\alpha_{sq,2,0,1})^{1/2} \cr\cr
1207: & = & \frac{p}{\sqrt{2}} \left [ 4-3p + \sqrt{12-20p+9p^2} \ \right ]^{1/2} \ .
1208: \label{rsqxy2}
1209: \eeqs
1210: %
1211: This has the derivatives
1212: %
1213: \beq
1214: \left . \frac{dr}{dp} \right |_{p=0} = \frac{1+\sqrt{3}}{\sqrt{2}} = 1.93185...
1215: \label{drdp_sqxy2_p0}
1216: \eeq
1217: %
1218: and
1219: %
1220: \beq
1221: \left . \frac{dr}{dp} \right |_{p=1} = 0 \ .
1222: \label{drdp1}
1223: \eeq
1224: %
1225:
1226:
1227: \begin{figure}[hbtp]
1228: \centering
1229: \leavevmode
1230: \epsfxsize=4.0in
1231: \begin{center}
1232: \leavevmode
1233: \epsffile{sqpxy2.ps}
1234: \end{center}
1235: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
1236: of $sq(2 \times L_x)$ for either free, periodic, or twisted periodic
1237: longitudinal boundary conditions. For comparison, zeros of the reliability
1238: polynomial are shown for the cyclic strip with $L_x=30$ (i.e., $n=60$).}}
1239: \label{sqpxy2}
1240: \end{figure}
1241:
1242:
1243: The locus ${\cal B}$ for the $L_y=2$ free strips of the square lattice, shown
1244: in Fig. \ref{sqpxy2}, is formed as the continuous accumulation set of zeros of
1245: the reliability polynomial in the limit of infinite strip length, $m \to
1246: \infty$ and is the solution of the equation expressing the degeneracy of
1247: magnitudes $|\alpha_{sq,2,0,1}|=|\alpha_{sq,2,0,2}|$. This is an arc of a
1248: circle with radius 1/3 centered at 1,
1249: %
1250: \beq
1251: p = 1 + \frac{1}{3}e^{i\theta} \quad \theta
1252: \in [-\theta_{sq2}, \theta_{sq2}]
1253: \label{pcircsq2}
1254: \eeq
1255: %
1256: where
1257: %
1258: \beq
1259: \theta_{sq2} = {\rm arctan}(2\sqrt{2} \ ) \simeq 70.53^\circ \ .
1260: \label{theta_sq2}
1261: \eeq
1262: %
1263: This circular arc crosses the real $p$ axis at a single point, which is thus
1264: $p_c$, namely
1265: %
1266: \beq
1267: p_c = \frac{4}{3} \quad {\rm for} \quad sq, \ FBC_y, \ 2 \times \infty \ .
1268: \label{pcross_sq2}
1269: \eeq
1270: %
1271: This applies here for the case of free longitudinal boundary conditions, and we
1272: shall show below that it also applies for the case of periodic or twisted
1273: periodic longitudinal boundary conditions. We find that this property that the
1274: value of $p_c$ for a strip of a given type of lattice and a given set of
1275: transverse boundary conditions is independent of the longitudinal boundary
1276: conditions also holds for other strips, and we indicate this when listing
1277: these values of $p_c$ (or $(p_c)_{eff}$) by displaying only the transverse
1278: boundary condition ($FBC_y$ in eq. (\ref{pcross_sq2})). The arc has endpoints
1279: at the values of $p$ where the the expression $12-20p+9p^2$ in the square roots
1280: in $\alpha_{sq,2,0,j}$ vanishes so that $\alpha_{sq,2,0,1}=\alpha_{sq,2,0,2}$.
1281: These endpoints are
1282: %
1283: \beq
1284: p = 1 + \frac{1}{3}e^{\pm i\theta_{sq2}} =
1285: \frac{2}{9}(5 \pm \sqrt{2} \ i) \simeq 1.11111 \pm 0.31427i \ .
1286: \label{pendsqly2}
1287: \eeq
1288: %
1289: Recall that two roots of an algebraic equation are equal where the discriminant
1290: vanishes; for the equation $\alpha^2-(4-3p)\alpha+1-p=0$ that yields
1291: $\alpha_{sq,2,0,j}$, $j=1,2$ as its roots, this discriminant is $12-20p+9p^2$.
1292:
1293:
1294: \subsection{$L_y=3$ Strip of the Square Lattice with Free Boundary Conditions}
1295:
1296:
1297: The generating function for this strip follows from our calculation of
1298: $Z(G,q,v)$ for this strip \cite{s3a} and has the form (\ref{gammazcalc}) with
1299: %
1300: \beq
1301: {\cal N}=p^2\biggl [ 1 + p^4(1-p)z - p^6(1-p)^3z^2 \biggr ]
1302: \label{numsqy3}
1303: \eeq
1304: %
1305: and
1306: %
1307: \beqs
1308: {\cal D}(p,z) & = & 1+\sum_{j=1}^4 b_{sq3f,j}z^j \cr\cr
1309: & = & \prod_{j=1}^4(1-p^3\alpha_{sq3f,j}z)
1310: \label{densqy3}
1311: \eeqs
1312: %
1313: where
1314: %
1315: \beq
1316: b_{sq3f,1}= -p^3(15-24p+10p^2)
1317: \label{bsq3f1}
1318: \eeq
1319: %
1320: \beq
1321: b_{sq3f,2}=p^6(1-p)(32-66p+46p^2-11p^3)
1322: \label{bsq3f2}
1323: \eeq
1324: %
1325: \beq
1326: b_{sq3f,3}=-p^9(1-p)^3(15-20p+7p^2)
1327: \label{bsq3f3}
1328: \eeq
1329: %
1330: \beq
1331: b_{sq3f,4}=p^{12}(1-p)^5
1332: \label{bsq3f4}
1333: \eeq
1334: %
1335:
1336: \begin{figure}[hbtp]
1337: \centering
1338: \leavevmode
1339: \epsfxsize=4.0in
1340: \begin{center}
1341: \leavevmode
1342: \epsffile{sqpxy3.ps}
1343: \end{center}
1344: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
1345: of $sq(3 \times L_x)$ for either free, periodic, or twisted periodic
1346: longitudinal boundary conditions. For comparison, zeros of the reliability
1347: polynomial are shown for the cyclic strip of this width with $L_x=20$ (i.e.,
1348: $n=60$).}}
1349: \label{sqpxy3}
1350: \end{figure}
1351:
1352:
1353: In Fig. \ref{sqpxy3} we show the locus ${\cal B}$ for the $L_y=3$ strips
1354: of the square lattice with any longitudinal boundary conditions, together with
1355: zeros for the cyclic strip of this width with $L_x=20$. This locus consists of
1356: two complex-conjugate arcs which do not cross the real axis. From the arc
1357: endpoints closest to the real axis, one can infer the effective quantity
1358: $(p_c)_{eff} \simeq 1.335$. The locus is again concave to the left and
1359: roughly centered about the point $p=1$.
1360:
1361:
1362: \subsection{$L_y=2$ Cyclic and M\"obius Strips of the Square Lattice}
1363:
1364: For the reader's convenience, we shall give some details of our calculation for
1365: this family of graphs. The Tutte polynomials for these families of graphs were
1366: computed in \cite{a} and have $N_{T,2,\lambda}=6$, in accordance with the
1367: general formula (\ref{nttotcyc}). There are $n_T(2,0)=2$ terms with
1368: coefficient $c^{(0)}$, denoted $\lambda_{T,sq,2,0,j}$, $j=1,2$;
1369: $n_T(2,1)=3$ terms
1370: with coefficient $c^{(1)}$, denoted $\lambda_{T,sq,2,1,j}$, and the unique
1371: term
1372: with coefficient $c^{(2)}$, which term is $\lambda_{T,sq,2,2}=1$ as a special
1373: case of the general result (\ref{lamtutlyly}). Explicitly,
1374: %
1375: \beqs
1376: & & T(sq[L_y=2,L_x=m,cyc],x,y) = \frac{1}{x-1}\sum_{d=0}^2
1377: c^{(d)} \sum_{j=1}^{n_T(2,d)} (\lambda_{T,sq,2,d,j})^m \cr\cr
1378: & & = \frac{1}{x-1}\left [ \sum_{j=1}^2 (\lambda_{T,sq,2,0,j})^m
1379: + c^{(1)}\sum_{j=1}^3 (\lambda_{T,sq,2,1,j})^m
1380: + c^{(2)}(\lambda_{T,sq,2,2})^m \right ]
1381: \label{tlad}
1382: \eeqs
1383: %
1384: where $n_T(L_y,d)$ was recalled from \cite{cf} in eq. (\ref{ntlyd}) and
1385: \cite{a}
1386: %
1387: \beq
1388: \lambda_{T,sq,2,0,j} = \frac{1}{2}\biggl [ (1+y+x+x^2) \pm \Bigl (
1389: y^2 +2y(1+x-x^2) + (x^2+x+1)^2 \Bigr )^{1/2} \ \biggr ]
1390: \label{lam20j}
1391: \eeq
1392: %
1393: %
1394: \beq
1395: \lambda_{T,sq,2,1,j} = \frac{1}{2}\biggl [x+y+2 \pm \Bigl ( (x-y)^2 +
1396: 4(x+y+1) \Bigr )^{1/2} \ \biggr ]
1397: \label{lam21j}
1398: \eeq
1399: %
1400: with $j=1,2$ corresponding to $\pm$, and
1401: %
1402: \beq
1403: \lambda_{T,sq,2,1,3}=x \ .
1404: \label{lam213}
1405: \eeq
1406: %
1407: We have discussed above how equalities occur between certain
1408: $\lambda_{T,G,L_y,d,j}$'s when evaluated at $x=1$. We show this explicitly
1409: here. First, we have
1410: %
1411: \beq
1412: \lambda_{T,sq,2,0,j} = \lambda_{T,sq,2,1,j} = \frac{1}{2} \left [y+3 \pm
1413: \sqrt{y^2+2y+9} \ \right ] \quad {\rm for} \ \ x=1
1414: \label{lamlady}
1415: \eeq
1416: %
1417: where $j=1,2$ refer correspond to $\pm$ on the right-hand side,
1418: respectively. Furthermore,
1419: %
1420: \beq
1421: \lambda_{T,sq,2,1,3} = \lambda_{T,sq,2,2} = 1 \quad {\rm for} \ \ x=1 \ .
1422: \label{lamladlast}
1423: \eeq
1424: %
1425: Hence, $N_{R,sq,2,\lambda}=(1/2)N_{T,sq,2,\lambda} = 3$, in accordance with the
1426: inequality (\ref{nrtot}) (realized as an equality) in our theorem above. Using
1427: L'Hopital's rule to evaluate the Tutte polynomial (\ref{tlad}) at $x=1$, we
1428: obtain
1429: %
1430: \beqs
1431: & & R(sq[L_y=2,m,cyc],p)=p^{2m} \Biggl [
1432: (\alpha_{sq,2,0,1})^m+(\alpha_{sq,2,0,2})^m - 2(\alpha_{sq,2,2})^m \cr\cr & & +
1433: m p^{-1}(1-p)^2 \Biggl \{ (\alpha_{sq,2,0,1})^{m-1} \left (1 +
1434: \frac{3(1-p)}{\sqrt{R_2}} \right ) \cr\cr
1435: & & + (\alpha_{sq,2,0,2})^{m-1} \left (1 -
1436: \frac{3(1-p)}{\sqrt{R_2}} \right ) - (\alpha_{sq,2,2})^{m-1} \Biggr \} \Biggr ]
1437: \label{Rsqpxy2}
1438: \eeqs
1439: %
1440: where $R_2$ was defined above in eq. (\ref{r2}), the $\alpha_{sq,2,0,j}$ were
1441: given in (\ref{alpha20j}) above, and
1442: %
1443: \beq
1444: \alpha_{sq,2,2}=1-p
1445: \label{alpha22}
1446: \eeq
1447: %
1448: as a special case of (\ref{alphalyly}) above.
1449:
1450: For the $L_y=2$ M\"obius (Mb) strip of the square lattice \cite{a}
1451: %
1452: \beqs
1453: & & T(sq[L_y=2,L_x=m,Mb],x,y) =
1454: \frac{1}{x-1}\Biggl [ \sum_{j=1}^2 (\lambda_{T,sq,2,0,j})^m \cr\cr
1455: & & + c^{(1)}\biggl ( -(\lambda_{T,sq,2,1,1})^m + \sum_{j=2}^3
1456: (\lambda_{T,sq,2,1,j})^m \biggr ) - (\lambda_{T,sq,2,2})^m \Biggr ] \ .
1457: \label{tmb}
1458: \eeqs
1459: %
1460: From this we obtain
1461: %
1462: \beqs
1463: & & R(sq[L_y=2,m,Mb],p)=p^{2m} \Biggl [ (\alpha_{sq,2,0,1})^m +
1464: (\alpha_{sq,2,0,2})^m - (\alpha_{sq,2,2})^m \cr\cr
1465: & & + m p^{-1}(1-p)^2 \Biggl \{
1466: (\alpha_{sq,2,0,1})^{m-1} \left (1 + \frac{3(1-p)}{\sqrt{R_2}} \right ) +
1467: (\alpha_{sq,2,0,2})^{m-1} \left (1 - \frac{3(1-p)}{\sqrt{R_2}} \right )
1468: \cr\cr
1469: & & + (\alpha_{sq,2,2})^{m-1} \Biggr \} \Biggr ] \ .
1470: \label{Rsqpxy2mb}
1471: \eeqs
1472: %
1473:
1474:
1475: For both of these $L_y=2$ cyclic and M\"obius strips of the square lattice, the
1476: dominant term in the physical interval $p \in [0,1]$ is $\alpha_{sq,2,0,1}$ and
1477: hence the asymptotic reliability per vertex, $r$, is the same as that for the
1478: corresponding strip with free longitudinal boundary conditions, given by
1479: (\ref{rsqxy2}):
1480: %
1481: \beqs
1482: r(sq[2 \times \infty, cyc.],p) & = & r(sq[2 \times \infty, Mb.],p) \cr\cr
1483: & = & r(sq[2 \times \infty, free],p) \cr\cr
1484: & = & r(sq[2 \times \infty, FBC_y],p) \ .
1485: \label{r2sq}
1486: \eeqs
1487: %
1488: For the infinite-length limit of these strip graphs, and for the
1489: respective infinite-length limits of other strip graphs to be considered below,
1490: we find that the $r$ function is independent of the longitudinal boundary
1491: conditions. This is indicated in the last line of eq. (\ref{r2sq}) and
1492: analogous equations below by listing only the type of transverse boundary
1493: condition. The dominant term for large positive $p$ is $\alpha_{sq,2,0,2}$.
1494: Because the two dominant $\alpha$'s in the reliability polynomials for these
1495: cyclic and M\"obius strips are the same as the two $\alpha$'s that enter in the
1496: reliability polynomial for the corresponding strip with free boundary
1497: conditions, it follows that the locus ${\cal B}$ is identical for the $L_y=2$
1498: strips of the square lattice with free, cyclic, and M\"obius boundary
1499: conditions. This locus is shown in Fig. \ref{sqpxy2}.
1500:
1501: We find that the reliability polynomials $R(sq[2 \times L_x, cyc.],p)$ and
1502: $R(sq[2 \times L_x, Mb],p)$ have only $p=0$ as a real root if $L_x \ge 1$ is
1503: odd, while for even $L_x \ge 2$, they each have an additional real root. In
1504: Table \ref{sq2realroots} we list the values of this additional respective real
1505: root for the reliability polynomials of these two strips. The results in this
1506: table suggest that as $L_x \to \infty$, these two respective additional real
1507: zeros for the cyclic and M\"obius strips could approach the same limit. The
1508: value of this limit could be about 1.3.
1509:
1510:
1511: \begin{table}
1512: \caption{\footnotesize{Values of the real root (aside from $p=0$) of
1513: $R(sq[2 \times L_x,cyc.],p)$ and $R(sq[2 \times L_x,Mb],p)$ for even $L_x$.}}
1514: \begin{center}
1515: \begin{tabular}{|c|c|c|}
1516: \hline\hline $L_x$ & cyc. & Mb. \\ \hline\hline
1517: 2 & 2.000000 & 1.626538 \\ \hline
1518: 4 & 1.404813 & 1.388652 \\ \hline
1519: 6 & 1.349507 & 1.346559 \\ \hline
1520: 8 & 1.333333 & 1.332635 \\ \hline
1521: 10 & 1.327103 & 1.326919 \\ \hline
1522: 12 & 1.324413 & 1.324362 \\ \hline
1523: 14 & 1.323229 & 1.323214 \\ \hline
1524: 16 & 1.322761 & 1.322757 \\ \hline
1525: 18 & 1.322658 & 1.322657 \\ \hline
1526: 20 & 1.322749 & 1.322748 \\ \hline\hline
1527: \end{tabular}
1528: \end{center}
1529: \label{sq2realroots}
1530: \end{table}
1531:
1532:
1533: \subsection{$L_y=2$ Cylindrical Strips of the Square Lattice}
1534: \label{sqcyl2}
1535:
1536: This is the minimal-width strip of the square lattice with cylindrical boundary
1537: conditions and involves double vertical edges. The results for the
1538: reliability polynomial are conveniently expressed in terms of the generating
1539: function (\ref{gamma}) and (\ref{gammazcalc}). We find
1540: %
1541: \beq
1542: {\cal N} = R(C_2,p) = p(2-p)
1543: \label{numsqpy2}
1544: \eeq
1545: %
1546:
1547: %
1548: \beqs
1549: {\cal D} & = & 1-p^2(6-8p+3p^2)z+(1-p)^2p^4z^2 \cr\cr
1550: & = & \prod_{j=1}^2(1-p^2 \alpha_{sqcyl2,j}z)
1551: \label{densqpy2}
1552: \eeqs
1553: %
1554: where
1555: %
1556: \beq
1557: \alpha_{sqcyl2,j} = \frac{1}{2}\left [ 6-8p+3p^2 \pm
1558: \sqrt{(2-p)(4-3p)(4-6p+3p^2)} \ \right ] \ .
1559: \label{alpha_sqcyl2}
1560: \eeq
1561: %
1562: We observe that $R(sq[2 \times L_x,cyl.],p)$ always has the factor $(2-p)$;
1563: this is associated with the fact that this strip has double vertical edges.
1564: For odd $m \ge 1$, i.e., even $L_x \ge 2$, $R(sq[2 \times L_x,cyl.],p)$ also
1565: has the factor $(3p^2-8p+6)$, with roots $(1/3)(4 \pm \sqrt{2}i)$. For $m \ge
1566: 4$, we observe that $R(sq[2 \times L_x,cyl.],p)$ has, besides the real zeros at
1567: $p=0,2$, a set of real zeros in the interval $(1,2]$. Among other features, we
1568: note that as $m$ increases, (i) the number of real zeros in this interval
1569: increases; (ii) the smallest zero in this interval decreases monotonically
1570: toward 4/3; (iii) the largest zero increases monotonically toward 2 as $m$
1571: increases, and (iv) the density of zeros is largest in the regions slightly
1572: above 4/3 and slightly below 2. A plot of these zeros is shown in
1573: Fig. \ref{sqpxpy2}; as is evident from this plot, as $m \to \infty$, these real
1574: zeros away from $p=0$ accumulate to form the line segment on ${\cal B}$
1575: extending between $p=4/3$ and $p=2$.
1576:
1577: In the physical interval $p \in [0,1]$ the dominant term is
1578: $\alpha_{sqcyl2,1}$, so the asymptotic reliability per vertex is
1579: %
1580: \beqs
1581: & & r(sq[2 \times \infty,PBC_y],p) = p(\alpha_{sqcyl2,1})^{1/2} \cr\cr
1582: & = & \frac{p}{\sqrt{2}} \left [ 6-8p+3p^2 + \sqrt{(2-p)(4-3p)(4-6p+3p^2)} \
1583: \right ]^{1/2}
1584: \label{rsqpy2}
1585: \eeqs
1586: %
1587: This has derivatives
1588: %
1589: \beq
1590: \left . \frac{dr}{dp} \right |_{p=0} = 1+\sqrt{2} = 2.41421...
1591: \label{drdp_sqpy2_p0}
1592: \eeq
1593: %
1594: and $dr/dp=0$ at $p=1$.
1595:
1596:
1597: \begin{figure}[hbtp]
1598: \centering
1599: \leavevmode
1600: \epsfxsize=4.0in
1601: \begin{center}
1602: \leavevmode
1603: \epsffile{sqpxpy2.ps}
1604: \end{center}
1605: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
1606: of $sqcyl(2 \times L_x)$ for either free or (twisted) periodic longitudinal
1607: boundary conditions. For comparison, zeros of the reliability polynomial are
1608: shown for the torus strip with $L_x=30$ (i.e., $n=60$).}}
1609: \label{sqpxpy2}
1610: \end{figure}
1611:
1612:
1613: The locus ${\cal B}$, shown in Fig. \ref{sqpxpy2}, is given by the union of
1614: an arc of a circle with a line segment on the real axis:
1615: %
1616: \beq
1617: {\cal B}: \quad \{ p = 1 + \frac{1}{\sqrt{3}}e^{i\theta} \ , \quad \theta \in
1618: [-\pi/2, \pi/2] \} \ \cup \{ \frac{4}{3} \le p \le 2 \} \ .
1619: \label{b_sqcpxy2}
1620: \eeq
1621: %
1622: Hence,
1623: %
1624: \beq
1625: p_c=2 \quad {\rm for} \quad sq, \ PBC_y, \ 2 \times \infty
1626: \label{pc_sqpy2}
1627: \eeq
1628: %
1629: The circular arc intersects the real axis and line segment at
1630: $p=1+\frac{1}{\sqrt{3}} \simeq 1.577$. The endpoints of the arc are located at
1631: %
1632: \beq
1633: p_{sqce}, p_{sqce}^* = 1 \pm \frac{i}{\sqrt{3}}
1634: \label{psqce}
1635: \eeq
1636: %
1637: These points, together with the endpoints of the line segment, $p=4/3$ and
1638: $p=2$ are the zeros of the expression in the square root in
1639: $\alpha_{sqcyl2,j}$ in eq. (\ref{alpha_sqcyl2}), which expression is the
1640: discriminant of the equation $\alpha^2-(6-8p+3p^2)\alpha+(1-p)^2=0$ whose roots
1641: are these $\alpha$'s.
1642:
1643: \subsection{$L_y=2$ Torus and Klein Bottle Strips of the Square Lattice}
1644:
1645: Either using a direct calculation or as a special case of our previous
1646: calculation of the Tutte polynomial \cite{hca}, using (\ref{rt}),
1647: we find
1648: %
1649: \beqs & & R(sq[L_y=2,m,torus],p)=p^{2m} \Biggl [
1650: (\alpha_{sqtor2,1})^m+(\alpha_{sqtor2,2})^m - 2(\alpha_{sqtor2,3})^m
1651: \cr\cr & & -
1652: m p^{-1}(1-p)^2 \left \{ \frac{(\alpha_{sqtor2,1})^{m-1}}{2} \left (2p-3 +
1653: \frac{(p-2)(6p^2-13p+8)}{\sqrt{(p-2)(3p-4)(3p^2-6p+4)}} \right )
1654: \right. \cr\cr & &
1655: \left. + \frac{(\alpha_{sqtor2,2})^{m-1}}{2} \left (2p-3 -
1656: \frac{(p-2)(6p^2-13p+8)}{\sqrt{(p-2)(3p-4)(3p^2-6p+4}} \right )
1657: + (1-p)(\alpha_{sqtor2,3})^{m-1} \right \} \Biggr ] \cr\cr & &
1658: \label{Rsqpxpy2}
1659: \eeqs
1660: %
1661: where
1662: %
1663: \beq
1664: \alpha_{sqtor2,j}=\alpha_{sqcyl2,j} \quad j=1,2
1665: \label{sqcyltor}
1666: \eeq
1667: %
1668: with $\alpha_{sqcyl2,j}$, $j=1,2$, given above in (\ref{alpha_sqcyl2}) and,
1669: %
1670: \beq
1671: \alpha_{sqtor2,3}=(1-p)^2 \ .
1672: \label{alpha_sqc_22}
1673: \eeq
1674: %
1675:
1676: For the $L_y=2$ Klein bottle (Kb.) strip of the square lattice
1677: %
1678: \beqs
1679: & & R(sq[L_y=2,m,Kb.],p)=p^{2m} \Biggl [
1680: (\alpha_{sqtor2,1})^m+(\alpha_{sqtor2,2})^m - (\alpha_{sqtor2,3})^m
1681: \cr\cr & & -
1682: m p^{-1}(1-p)^2 \left \{ \frac{(\alpha_{sqtor2,1})^{m-1}}{2} \left (2p-3 +
1683: \frac{(p-2)(6p^2-13p+8)}{\sqrt{(p-2)(3p-4)(3p^2-6p+4)}} \right )
1684: \right. \cr\cr & &
1685: \left. + \frac{(\alpha_{sqtor2,2})^{m-1}}{2} \left (2p-3 -
1686: \frac{(p-2)(6p^2-13p+8)}{\sqrt{(p-2)(3p-4)(3p^2-6p+4)}} \right ) -
1687: (1-p)(\alpha_{sqtor2,3})^{m-1} \right \} \Biggr ] \cr\cr & &
1688: \label{Rsqpxpy2kb}
1689: \eeqs
1690: %
1691:
1692: In the physical region $p \in [0,1]$, the dominant term is $\alpha_{sqtor2,1}$
1693: for both of these families of strips, and hence $r(sq[2 \times
1694: \infty,torus/Kb.],p) = r(sq[2 \times \infty,cyl.],p)$. This equality has
1695: already been incorporated in our notation $r(sq[2 \times \infty,PBC_y],p)$ in
1696: eq. (\ref{rsqpy2}). This is another example of the feature that for a given
1697: type of strip graph $G_s$ with a given width and set of transverse boundary
1698: conditions, $r$ is independent of the longitudinal boundary conditions.
1699:
1700: Similarly, since the locus ${\cal B}$ is determined by the equality in
1701: magnitude of the two dominant eigenvalues $\alpha_{sqtor2,1}$ and
1702: $\alpha_{sqtor2,2}$, and since these are the same for the corresponding strip
1703: with cylindrical boundary conditions, it follows that this locus is the same as
1704: the locus for the $L_y=2$ cylindrical strip of the square lattice, shown in
1705: Fig. \ref{sqpxpy2}.
1706:
1707:
1708: \subsection{$L_y=3$ Cylindrical Strips of the Square Lattice}
1709: \label{sqcyl3}
1710:
1711:
1712: Again we express the results in terms of the generating function for the
1713: reliability polynomial, (\ref{gamma}) and (\ref{gammazcalc}). We find
1714: %
1715: \beq
1716: {\cal N} = R(C_3,p) + p^5(3-p)(1-p)^2z = p^2[3-2p+p^3(3-p)(1-p)^2z]
1717: \label{numsqpy3}
1718: \eeq
1719: %
1720: where $R(C_3,p)=p^2(3-2p)$, and
1721: %
1722: \beqs
1723: {\cal D} & = &
1724: 1-p^3(24-56p+46p^2-13p^3)z+p^6(1-p)^2(24-46p+30p^2-7p^3)z^2-p^9(1-p)^5z^3
1725: \cr\cr
1726: & = & \prod_{j=1}^3(1-p^3 \alpha_{sqcyl3,j}z) \ .
1727: \label{densqcyl}
1728: \eeqs
1729: %
1730: The largest of the roots of the associated equation
1731: %
1732: \beqs
1733: & & \xi^3-(24-56p+46p^2-13p^3)\xi^2 \cr\cr
1734: & & +(1-p)^2(24-46p+30p^2-7p^3)\xi-(1-p)^5=0
1735: \label{xieqsqcyl3}
1736: \eeqs
1737: %
1738: is the dominant $\alpha_{sqcyl3,max}$, so that
1739: %
1740: \beq
1741: r(sq[3 \times \infty, PBC_y],p)=p(\alpha_{sqcyl3,max})^{1/3} \ .
1742: \label{rsqcyl3}
1743: \eeq
1744: %
1745: This function has the property that $dr/dp=2.84207...$ at $p=0$ and $dr/dp=0$
1746: at $p=1$. We find that the reliability polynomial $R(sq[3 \times \infty,
1747: cyl.],p)$ has only $p=0$ as a real root if $m \ge 1$ is odd, i.e. $L_x \ge 2$
1748: is even, while for even $m \ge 0$, i.e., odd $L_x \ge 1$, it has an additional
1749: real root which decreases as $m$ increases and appears to approach a limit of
1750: roughly 1.4. In Table \ref{sqcyl3realroots} we list the values of this
1751: additional real root.
1752:
1753:
1754: \begin{table}
1755: \caption{\footnotesize{Values of the real root (aside from $p=0$) of
1756: $R(sq[3 \times \infty, cyl.],p)$ for odd $L_x$.}}
1757: \begin{center}
1758: \begin{tabular}{|c|c|}
1759: \hline\hline $L_x$ & root \\ \hline\hline
1760: 1 & 1.500000 \\ \hline
1761: 3 & 1.440337 \\ \hline
1762: 5 & 1.422909 \\ \hline
1763: 7 & 1.416297 \\ \hline
1764: 9 & 1.412887 \\ \hline
1765: 11 & 1.410809 \\ \hline
1766: 13 & 1.409409 \\ \hline
1767: 15 & 1.408403 \\ \hline
1768: 17 & 1.407644 \\ \hline
1769: 19 & 1.407051 \\ \hline
1770: 21 & 1.406575 \\ \hline\hline
1771: \end{tabular}
1772: \end{center}
1773: \label{sqcyl3realroots}
1774: \end{table}
1775:
1776: \begin{figure}[hbtp]
1777: \centering
1778: \leavevmode
1779: \epsfxsize=4.0in
1780: \begin{center}
1781: \leavevmode
1782: \epsffile{sqpxpy3.ps}
1783: \end{center}
1784: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
1785: of $sqcyl(3 \times L_x)$ for either free or (twisted) periodic longitudinal
1786: boundary conditions. For comparison, zeros of the reliability polynomial are
1787: shown for the toroidal strip with $L_x=20$ (i.e., $n=60$).}}
1788: \label{sqpxpy3}
1789: \end{figure}
1790:
1791: In Fig. \ref{sqpxpy3} we show a plot of ${\cal B}$ for the $L_x \to \infty$
1792: limit of the $L_y=3$ strip of the square lattice with cylindrical boundary
1793: conditions.
1794:
1795:
1796: \subsection{$L_y=4$ Cylindrical Strip of the Square Lattice}
1797:
1798: As before, we express the results for the $L_y=4$ strip of the square lattice
1799: with cylindrical boundary conditions in terms of the numerator and denominator
1800: of the generating function. In particular, for the denominator, we find
1801: %
1802: \beqs
1803: {\cal D} & = & 1 + \sum_{j=1}^6 b_{sqcyl4,j}z^j \cr\cr
1804: & = & \prod_{j=1}^6 (1-p^4 \alpha_{sqcyl4,j}z)
1805: \label{densqcyl4}
1806: \eeqs
1807: %
1808: where
1809: %
1810: \beq
1811: b_{sqcyl4,1}=-p^4(90-293p+369p^2-211p^3+46p^4)
1812: \label{bsqcyl4_1}
1813: \eeq
1814: %
1815: \beq
1816: b_{sqcyl4,2}=p^8(1-p)^2(735-2977p+5094p^2-4710p^3+2481p^4-706p^5+85p^6)
1817: \label{bsqcyl4_2}
1818: \eeq
1819: %
1820: \beqs
1821: b_{sqcyl4,3} & = & -p^{12}(1-p)^4(1548-7518p+15948p^2-19170p^3 \cr\cr
1822: & & +14104p^4-6351p^5+1621p^6-181p^7)
1823: \label{bsqcyl4_3}
1824: \eeqs
1825: %
1826: \beq
1827: b_{sqcyl4,4}=p^{16}(1-p)^7(735-3177p+5870p^2-5934p^3+3456p^4-1097p^5+148p^6)
1828: \label{bsqcyl4_4}
1829: \eeq
1830: %
1831: \beq
1832: b_{sqcyl4,5}=-p^{20}(1-p)^{11}(90-223p+208p^2-88p^3+15p^4)
1833: \label{bsqcyl4_5}
1834: \eeq
1835: %
1836: \beq
1837: b_{sqcyl4,6}=p^{24}(1-p)^{15}
1838: \label{bsqcyl4_6}
1839: \eeq
1840: %
1841: The degree of ${\cal D}$ is too high to obtain closed-form algebraic
1842: expressions for the $\alpha_{sqcyl4,j}$'s, but they may be computed
1843: numerically. Denoting the dominant $\alpha_{sqcyl4,j}$ in the physical
1844: interval $p \in [0,1]$ as $\alpha_{sqcyl4,max}$, one has, for the asymptotic
1845: reliability per vertex, $r(sq[4 \times \infty,
1846: PBC_y],p)=p(\alpha_{sqcyl4,max})^{1/4}$. In Fig. \ref{sqpy4} we show the locus
1847: ${\cal B}$ for the infinite-length limit of this strip. This locus includes
1848: two complex-conjugates arcs, a self-conjugate arc crossing the real axis at $p
1849: \simeq 1.364$, and a real line segment extending from $p=4/3$ to $p \simeq
1850: 1.384$, which defines the value of $p_c$ for this strip.
1851:
1852: \begin{figure}[hbtp]
1853: \centering
1854: \leavevmode
1855: \epsfxsize=4.0in
1856: \begin{center}
1857: \leavevmode
1858: \epsffile{sqpy4.ps}
1859: \end{center}
1860: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
1861: of $sqcyl(4 \times L_x)$ for either free or (twisted) periodic longitudinal
1862: boundary conditions. For comparison, zeros of the reliability polynomial are
1863: shown for the cylindrical strip with $L_x=21$ (i.e., $n=84$).}}
1864: \label{sqpy4}
1865: \end{figure}
1866:
1867: \bigskip
1868:
1869:
1870: \subsection{$L_y=2$ Strip of the Triangular Lattice with Free Boundary
1871: Conditions}
1872:
1873: In addition to studying how reliability polynomials and their asymptotic $r$
1874: functions depend on the width and boundary conditions, it is also of interest
1875: to explore how they depend on the lattice type. For this purpose we have
1876: carried out calculations of reliability polynomials for strips of the
1877: triangular and honeycomb lattice. We begin with the $L_y=2$ strip of the
1878: triangular lattice with free boundary conditions, for which we calculate the
1879: generating function with
1880: %
1881: \beq
1882: {\cal N}(p,z)=p\biggl [ 1+p(1-p)z \biggr ]
1883: \label{numgamma_txy2}
1884: \eeq
1885: %
1886: \beqs
1887: {\cal D}(p,z) & = & 1-p^2(7-10p+4p^2)z+p^4(1-p)^2z^2 \cr\cr
1888: & = & (1-p^2 \alpha_{t,2,0,1}z)(1-p^2 \alpha_{t,2,0,2}z)
1889: \label{dengamma_txy2}
1890: \eeqs
1891: %
1892: where
1893: %
1894: \beqs
1895: \alpha_{t,2,0,j} & = & \frac{1}{2}\left [
1896: 7-10p+4p^2 \pm (3-2p)\sqrt{5-8p+4p^2} \right ] \cr\cr
1897: & = & \left [ \frac{ 3-2p + \sqrt{5-8p+4p^2}}{2} \right ]^2
1898: \quad {\rm for} \quad j=1,2 \ .
1899: \label{alpha20j_tpxy2}
1900: \eeqs
1901: %
1902: We note that $R(p,z)$ for the strip of the triangular lattice of length $m$ has
1903: the factor $(8-11p+4p^2)$ if $m=1$ mod 3, where our convention is that $m=1$ is
1904: a single square with diagonal (i.e. two triangles), and so forth.
1905:
1906:
1907: The dominant term in the physical interval $p \in [0,1]$ is $\alpha_{t,2,0,1}$,
1908: and hence the reliability per vertex for $L_x \to \infty$ is
1909: %
1910: \beqs
1911: r(t[2 \times \infty, FBC_y],p) & = & p(\alpha_{t,2,0,1})^{1/2} \cr\cr
1912: & = & \frac{p}{\sqrt{2}} \left [ 7-10p+4p^2 \pm (3-2p)\sqrt{5-8p+4p^2}
1913: \right ]^{1/2}
1914: \label{rtrixy2}
1915: \eeqs
1916: %
1917: This has derivatives
1918: %
1919: \beq
1920: \left . \frac{dr}{dp} \right |_{p=0} = \frac{3 + \sqrt{5}}{2} = 2.61803...
1921: \label{drdp_txy2_p0}
1922: \eeq
1923: %
1924: and $dr/dp=0$ at $p=1$.
1925:
1926:
1927: \begin{figure}[hbtp]
1928: \centering
1929: \leavevmode
1930: \epsfxsize=4.0in
1931: \begin{center}
1932: \leavevmode
1933: \epsffile{tpxy2.ps}
1934: \end{center}
1935: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
1936: of $tri(2 \times L_x)$ for either free or (twisted) periodic longitudinal
1937: boundary conditions. For comparison, zeros of the reliability polynomial are
1938: shown for the cyclic strip with $L_x=30$ (i.e., $n=60$).}}
1939: \label{tpxy2}
1940: \end{figure}
1941:
1942:
1943: The locus ${\cal B}$, shown in Fig. \ref{tpxy2}, is given by an arc of a
1944: circle:
1945: %
1946: \beq
1947: {\cal B}: \quad p = 1 + \frac{1}{2}e^{i\theta} \ , \quad \theta \in
1948: [-\frac{\pi}{2}, \frac{\pi}{2}]
1949: \label{b_tpxy2}
1950: \eeq
1951: %
1952: This is a leftwardly concave circular arc that crosses the real axis at
1953: %
1954: \beq
1955: p_c = \frac{3}{2} \quad tri, \ FBC_y, \ 2 \times \infty,
1956: \label{pctpxy2}
1957: \eeq
1958: %
1959: and has endpoints at $p=1 \pm i/2$, where the factor $5-8p+4p^2$ in the
1960: discriminant $(5-8p+4p^2)(3-2p)^2$ of the equation
1961: $\alpha^2-(7-10p+4p^2)\alpha+(1-p)^2=0$ yielding the $\alpha$'s vanishes. We
1962: shall see that a rather different strip, the self-dual strip of the square
1963: lattice of width $L_y=1$, yields the same locus ${\cal B}$, although the
1964: reliability polynomials are different. Interestingly, the effective degree is
1965: the same, namely 4, for these two strips, although the $L_y=2$ cyclic strip of
1966: the triangular lattice is a $\Delta$-regular graph, with uniform vertex degree
1967: $\Delta=4$, whereas the self-dual strip of the square lattice with $L_y=1$ has
1968: two quite different kinds of vertex degrees - 3 for the outer rim and $m$ for
1969: the central vertex connected to each vertex on the rim by edges. However, it
1970: is not, in general, true that if two families of graphs have the same effective
1971: vertex degree then the resultant loci ${\cal B}$ are the same; for example,
1972: each of the self-dual cyclic strips of the square lattice of width $L_y$ has
1973: the same effective vertex degree in the limit $L_x \to \infty$, but they have
1974: different loci ${\cal B}$.
1975:
1976:
1977: \subsection{$L_y=2$ Cyclic Strip of the Triangular Lattice}
1978:
1979: We next consider the $L_y=2$ cyclic strip of the triangular lattice.
1980: Either using a direct calculation or as a special case of our previous
1981: calculation of the Tutte polynomial \cite{ta}, using (\ref{rt}), we find
1982: %
1983: \beqs & & R(t[L_y=2,m,cyc],p)=p^{2m} \Biggl [
1984: (\alpha_{t,2,0,1})^m+(\alpha_{t,2,0,2})^m - 2(\alpha_{t,2,2})^m \cr\cr & & +
1985: m p^{-1}(5-3p)^{-1}(1-p)^2 \left \{ (\alpha_{t,2,0,1})^{m-1} (1-p) \left
1986: (7-4p + \frac{(3-2p)(5-4p)}{\sqrt{4p^2-8p+5}} \right ) \right. \cr\cr & &
1987: + \left. (\alpha_{t,2,0,2})^{m-1} (1-p) \left (7-4p -
1988: \frac{(3-2p)(5-4p)}{\sqrt{4p^2-8p+5}} \right ) -
1989: 2(p^2-3p+2)(\alpha_{t,2,2})^{m-1} \right \} \Biggr ]
1990: \cr\cr & &
1991: \label{Rtripxy2}
1992: \eeqs
1993: %
1994: where $\alpha_{t,2,0,j}$ were given above for the free strip and
1995: %
1996: \beq
1997: \alpha_{t,2,2} = (1-p)^2 \ .
1998: \label{alpha22_tpxy2}
1999: \eeq
2000: %
2001: Because the dominant $\alpha$'s consist of $\alpha_{t,2,0,j}$, $j=1,2$, which
2002: are the same for the free, cyclic, and M\"obius strips, it follows that
2003: the locus ${\cal B}$ is the same for the $L_x \to \infty$ limits of these
2004: three strips.
2005:
2006:
2007: \subsection{$L_y=2$ Free Strips of the Honeycomb Lattice}
2008:
2009: For the $L_y=2$ strip of the honeycomb lattice with free boundary conditions we
2010: calculate a generating function with
2011: %
2012: \beq
2013: {\cal N} = p
2014: \label{numhc}
2015: \eeq
2016: %
2017:
2018: %
2019: \beqs
2020: {\cal D} & = & 1-p^4(6-5p)z+(1-p)p^8z^2 \cr\cr
2021: & = & (1-p^4 \alpha_{hc,2,0,1}z)(1-p^4 \alpha_{hc,2,0,2}z)
2022: \label{denhc}
2023: \eeqs
2024: %
2025: where
2026: %
2027: \beq
2028: \alpha_{hc,2,0,j} = \frac{1}{2}\left [ 6-5p \pm \sqrt{H_2} \
2029: \right ]
2030: \label{alpha_hc2}
2031: \eeq
2032: %
2033: with
2034: %
2035: \beq
2036: H_2=32-56p+25p^2 \ .
2037: \label{h2}
2038: \eeq
2039: %
2040: In the physical interval $p \in [0,1]$ the dominant term is
2041: $\alpha_{hc,2,0,1}$, so the asymptotic reliability per vertex is
2042: %
2043: \beqs
2044: r(hc[2 \times \infty,FBC_y],p) & = & p(\alpha_{hc,2,0,1})^{1/4} \cr\cr
2045: & = & \frac{p}{2^{1/4}} \left [6-5p + \sqrt{32-56p+25p^2} \right ]^{1/4}
2046: \label{rhcpxy2}
2047: \eeqs
2048: %
2049: This has derivatives
2050: %
2051: \beq
2052: \left . \frac{dr}{dp} \right |_{p=0} = (1+\sqrt{2})^{1/2} = 1.55377...
2053: \label{drdp_hc2_p0}
2054: \eeq
2055: %
2056: and $dr/dp=0$ at $p=1$.
2057:
2058:
2059: \begin{figure}[hbtp]
2060: \centering
2061: \leavevmode
2062: \epsfxsize=4.0in
2063: \begin{center}
2064: \leavevmode
2065: \epsffile{hpxy2.ps}
2066: \end{center}
2067: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
2068: of $hc(2 \times L_x)$ for either free or (twisted) periodic longitudinal
2069: boundary conditions. For comparison, zeros of the reliability polynomial are
2070: shown for the cyclic strip with $L_x=30$ (i.e., $n=120$).}}
2071: \label{hpxy2}
2072: \end{figure}
2073:
2074:
2075: The locus ${\cal B}$, shown in Fig. \ref{hpxy2}, is given by an arc of a
2076: circle:
2077: %
2078: \beq
2079: {\cal B}: \quad p = 1 + \frac{1}{5}e^{i\theta} \ , \quad \theta \in
2080: [-\theta_{hce}, \theta_{hce}]
2081: \label{b_hcpxy2}
2082: \eeq
2083: %
2084: where
2085: %
2086: \beq
2087: \theta_{hce} = {\rm arctan}(4/3) \simeq 53.13^\circ
2088: \label{theta_hce}
2089: \eeq
2090: %
2091: This is a leftwardly concave arc that crosses the real axis at
2092: %
2093: \beq
2094: p_c = \frac{6}{5} \quad {\rm for} \quad hc, \ FBC_y, \ 2 \times \infty,
2095: \label{pchcpxy2}
2096: \eeq
2097: %
2098: and has endpoints at
2099: %
2100: \beq
2101: p_{hce}, p_{hce}^* = \frac{4}{25}(7 \pm i)
2102: \label{phce}
2103: \eeq
2104: %
2105: where the discriminant $H_2$ vanishes.
2106:
2107:
2108: \subsection{$L_y=2$ Cyclic and M\"obius Strips of the Honeycomb Lattice}
2109:
2110: For the $L_y=2$ cyclic strip of the honeycomb lattice, either using a direct
2111: calculation or as a special case of our previous calculation of the Tutte
2112: polynomial \cite{hca}, using (\ref{rt}), we find
2113: %
2114: \beqs
2115: & & R(hc[L_y=2,m,cyc],p)=p^{4m} \Biggl [
2116: (\alpha_{hc,2,0,1})^m+(\alpha_{hc,2,0,2})^m - 2(\alpha_{hc,2,2})^m \cr\cr & & +
2117: m p^{-1}(1-p)^2 \left \{ (\alpha_{hc,2,0,1})^{m-1} \left (3 +
2118: \frac{16-15p}{\sqrt{H_2}} \right ) \right. \cr\cr & &
2119: \left. + (\alpha_{hc,2,0,2})^{m-1} \left (3 -
2120: \frac{16-15p}{\sqrt{H_2}} \right ) - 2(\alpha_{hc,2,2})^{m-1}
2121: \right \} \Biggr ] \cr\cr & &
2122: \label{Rhcpxy2}
2123: \eeqs
2124: %
2125: where $\alpha_{hc,2,0,j}$, $j=1,2$, were given above in (\ref{alpha_hc2}) and,
2126: as a special case of (\ref{alphalyly}), we have
2127: %
2128: \beq
2129: \alpha_{hc,2,2}=1-p \ .
2130: \label{alpha_hc_22}
2131: \eeq
2132: %
2133:
2134: For the $L_y=2$ M\"obius strip of the honeycomb lattice, using the same
2135: methods, we obtain
2136: %
2137: \beqs
2138: & & R(hc[L_y=2,m,Mb],p)=p^{4m} \Biggl [
2139: (\alpha_{hc,2,0,1})^m+(\alpha_{hc,2,0,2})^m - (\alpha_{hc,2,2})^m \cr\cr & & +
2140: m p^{-1}(1-p)^2 \left \{ (\alpha_{hc,2,0,1})^{m-1} \left (3 +
2141: \frac{16-15p}{\sqrt{H_2}} \right ) \right. \cr\cr & &
2142: \left. + (\alpha_{hc,2,0,2})^{m-1} \left (3 -
2143: \frac{16-15p}{\sqrt{H_2}} \right ) + 2(\alpha_{hc,2,2})^{m-1}
2144: \right \} \Biggr ] \ . \cr\cr & &
2145: \label{Rhcpxy2mb}
2146: \eeqs
2147:
2148: Since the locus ${\cal B}$ is determined by the equality in magnitude of the
2149: two dominant eigenvalues $\alpha_{hc,2,0,1}$ and $\alpha_{hc,2,0,2}$, and since
2150: these are the same for the strip with free and periodic or twisted periodic
2151: boundary conditions, it follows that this locus is the same as (\ref{b_hcpxy2})
2152: and (\ref{theta_hce}) for the $L_y=2$ cyclic or M\"obius strip of the honeycomb
2153: lattice. We show this locus in Fig. \ref{hpxy2}.
2154:
2155:
2156: \subsection{$L_y=1$ Cyclic Strip of the Square Lattice with Self-Dual Boundary
2157: Conditions}
2158:
2159: We have presented some general structural results above for reliability
2160: polynomials of cyclic strip graphs of the square lattice with self-dual
2161: boundary conditions. Recall that these are denoted $G_D$, of size $L_y \times
2162: L_x$; in subscripts we will use the symbol $sqdbc$. In this and the next
2163: section we give explicit calculations of the reliability polynomials. Note
2164: that $G_D[1 \times L_x]$ is the wheel graph with a central spoke vertex
2165: connected to $L_x$ outer vertices on the rim. Either using a direct
2166: calculation or as a special case of our previous calculation of the Tutte
2167: polynomial \cite{jz,dg}, using (\ref{rt}), we find (with $L_x=m$)
2168: %
2169: \beq
2170: R(G_D[1 \times m],p)=p^m\biggl [ (\alpha_{sqdbc,1,1,1})^m +
2171: (\alpha_{sqdbc,1,1,2})^m -2(\alpha_{sqdbc,1,2,1})^m \biggr ]
2172: \label{rwh1}
2173: \eeq
2174: %
2175: where
2176: %
2177: \beq
2178: \alpha_{sqdbc,1,1,j} = \frac{1}{2}\biggl [ 3-2p \pm \sqrt{5-8p+4p^2} \
2179: \biggr ] \quad j=1,2
2180: \label{rlam11j}
2181: \eeq
2182: %
2183: and, consistent with (\ref{alphalyly}),
2184: %
2185: \beq
2186: \alpha_{sqdbc,1,2,1}=1-p \ .
2187: \label{rlam121}
2188: \eeq
2189: %
2190: In accordance with our general results above, the total number of terms is the
2191: same as that for the Tutte polynomial, $N_{R,G_D,L_y=1,\lambda}=
2192: N_{T,G_D,L_y=1,\lambda}=3$.
2193:
2194: In the physical interval $p \in [0,1]$ the dominant term is
2195: $\alpha_{sqdbc,1,1,1}$ so that the asymptotic reliability per vertex is
2196: %
2197: \beqs
2198: r(G_D[1 \times \infty],p) & = & p \alpha_{sqdbc,1,1,1} \cr\cr
2199: & = & p \left [ \ \frac{3-2p + \sqrt{5-8p+4p^2}}{2} \ \right ]
2200: \label{rgd1}
2201: \eeqs
2202: %
2203: Interestingly, although the graphs are different and the reliability
2204: polynomials are different for the $G_D(1 \times m)$ and $tri(2 \times m)$
2205: strips (with either both free or both periodic longitudinal boundary
2206: conditions), the infinite-length limits of these four families of graphs yield
2207: the same $r$ function, i.e., (\ref{rgd1}) is identical to $r(tri[2 \times
2208: \infty,FBC_y],p)$ given above in (\ref{rtrixy2}).
2209:
2210:
2211: \begin{figure}[hbtp]
2212: \centering
2213: \leavevmode
2214: \epsfxsize=4.0in
2215: \begin{center}
2216: \leavevmode
2217: \epsffile{whpxy1.ps}
2218: \end{center}
2219: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
2220: of $G_D(1 \times L_x)$. For comparison, zeros of the reliability polynomial are
2221: shown for $L_x=40$.}}
2222: \label{whpxy1}
2223: \end{figure}
2224:
2225:
2226: The locus ${\cal B}$ is shown in Fig. \ref{whpxy1} and forms arc of a circle
2227: centered at $p=1$ of radius 1/2 with endpoints at $p=1 \pm (1/2)i$, i.e.,
2228: %
2229: \beq
2230: {\cal B}: \quad p = 1 + \frac{1}{2}e^{i\theta} \ , \quad \theta \in
2231: [-\frac{\pi}{2}, \frac{\pi}{2}]
2232: \label{b_whpxy1}
2233: \eeq
2234: %
2235: so that
2236: %
2237: \beq
2238: p_c = \frac{3}{2} \quad {\rm for} \quad G_D(1 \times \infty) \ .
2239: \label{pcwhpxy1}
2240: \eeq
2241: %
2242: The endpoints occur at the branch point singularities of the polynomial
2243: $5-8p+4p^2$ that appears in the square root in the terms (\ref{rlam11j}).
2244: The locus (\ref{b_whpxy1}) is identical to (\ref{b_tpxy2}), although the
2245: $\alpha$'s involved are different.
2246:
2247: \bigskip
2248:
2249:
2250: \subsection{$L_y=2$ Cyclic Strip of the Square Lattice with Self-Dual Boundary
2251: Conditions}
2252:
2253: For $G_D[2 \times m]$, our general formulas give $N_{R,G_D,2,\lambda}=10$ with
2254: $n_R(G_D,2,1)=5$, $n_R(G_D,2,2)=4$, and $n_R(G_D,2,3)=1$. We find that
2255: $R(G_D[2 \times m],p)$ has the form (\ref{rgd}),
2256: %
2257: \beq
2258: R(G_D[2 \times m],p)=p^{2m}\biggl [
2259: \sum_{j=1}^5 (\alpha_{sqdbc,2,1,j})^m -2 \sum_{j=1}^4
2260: (\alpha_{sqdbc,2,2,j})^m + 3(\alpha_{sqdbc,2,3,1})^m \biggr ]
2261: \label{rwh2}
2262: \eeq
2263: %
2264: where
2265: %
2266: \beq
2267: \alpha_{sqdbc,2,3,1}=(1-p)^2
2268: \label{lam_sqdbc2_231}
2269: \eeq
2270: %
2271: and the $\alpha_{sqdbc,2,2,j}$ are the roots of the degree-4 equation
2272: %
2273: \beq
2274: \xi^4+a_{db1}\xi^3+a_{db2}\xi^2+a_{db3}\xi+a_{db4}=0
2275: \label{dbcly2deg4eq}
2276: \eeq
2277: %
2278: with
2279: %
2280: \beq
2281: a_{db1}=-(1-p)(7-5p)
2282: \label{adb1}
2283: \eeq
2284: %
2285: \beq
2286: a_{db2}=(1-p)^2(13-17p+5p^2)
2287: \label{adb2}
2288: \eeq
2289: %
2290: \beq
2291: a_{db3}=-(1-p)^4(7-5p)
2292: \label{adb3}
2293: \eeq
2294: %
2295: \beq
2296: a_{db4}=(1-p)^6 \ .
2297: \label{adb4}
2298: \eeq
2299: %
2300: In eq. (\ref{rwh2}) the $\alpha_{sqdbc,2,1,j}$ are the roots of the
2301: degree-5 equation
2302: %
2303: \beq
2304: \xi^5+b_{db1}\xi^4+b_{db2}\xi^3+b_{db3}\xi^2+b_{db4}\xi+b_{db5}=0
2305: \label{dbcly2deg5eq}
2306: \eeq
2307: %
2308: where
2309: %
2310: \beq
2311: b_{db1}=-(12-18p+7p^2)
2312: \label{bdbc1coeff}
2313: \eeq
2314: %
2315: \beq
2316: b_{db2}=(1-p)(36-67p+41p^2-8p^3)
2317: \label{bdbc2coeff}
2318: \eeq
2319: %
2320: \beq
2321: b_{db3}=-(1-p)^2(36-75p+53p^2-13p^3)
2322: \label{bdbc3coeff}
2323: \eeq
2324: %
2325: \beq
2326: b_{db4}=2(1-p)^4(3-2p)(2-p)
2327: \label{bdbc4coeff}
2328: \eeq
2329: %
2330: \beq
2331: b_{db5}=-(1-p)^6 \ .
2332: \label{bdbc5coeff}
2333: \eeq
2334: %
2335:
2336: \begin{figure}[hbtp]
2337: \centering
2338: \leavevmode
2339: \epsfxsize=4.0in
2340: \begin{center}
2341: \leavevmode
2342: \epsffile{whpxy2.ps}
2343: \end{center}
2344: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
2345: of $G_D(2 \times L_x)$. For comparison, zeros of the reliability polynomial are
2346: shown for $L_x=30$ (i.e., $n=61$).}}
2347: \label{whpxy2}
2348: \end{figure}
2349:
2350: The locus ${\cal B}$ is shown in Fig. \ref{whpxy2} and consists of arcs, again
2351: concave to the left, that almost cross, but actually have endpoints near to,
2352: the real axis at $q \simeq 1.4$ and have endpoints at $q \simeq 0.7765 \pm
2353: 0.4302i$ and $q \simeq 1.406 \pm 0.036795i$.
2354:
2355:
2356:
2357: \subsection{$L_y=3$ Cyclic Strip of the Square Lattice with Self-Dual Boundary
2358: Conditions}
2359:
2360: For $G_D[3 \times m]$, our general formulas give $N_{R,G_D,3,\lambda}=35$ with
2361: $n_R(G_D,3,4)=1$, $n_R(G_D,3,3)=6$, $n_R(G_D,3,2)=14$, and $n_R(G_D,3,1)=14$.
2362: We have calculated the reliability polynomial from our earlier calculation of
2363: the Tutte polynomial for this strip \cite{dg,sdg}. Since the results for the
2364: $\alpha$'s are rather complicated, we do not list them in detail here (some
2365: examples are given in the appendix) but instead concentrate on the locus
2366: ${\cal B}$. This is shown in Fig. \ref{whpxy3}.
2367:
2368: \begin{figure}[hbtp]
2369: \centering
2370: \leavevmode
2371: \epsfxsize=4.0in
2372: \begin{center}
2373: \leavevmode
2374: \epsffile{whpxy3.ps}
2375: \end{center}
2376: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
2377: of $G_D(3 \times L_x)$. For comparison, zeros of the reliability polynomial
2378: are shown for $L_x=20$ (i.e., $n=61$).}}
2379: \label{whpxy3}
2380: \end{figure}
2381:
2382: \subsection{Some Families of Graphs with Multiple Edges}
2383:
2384: Consider a connected graph $G=(V,E)$. We take this graph to be loopless; this
2385: incurs no loss of generality since loops do not affect the reliability
2386: polynomial. Now consider a graph $H$ obtained from $G$ by adding another edge.
2387: Clearly, for the physical range $p \in [0,1]$, the inequality $R(H,p) \ge
2388: R(G,p)$ holds, with equality only at $p=1$. That is, adding further
2389: communications link(s) between two (already connected) nodes in a network
2390: increases the reliability of the network. A particularly simple modification
2391: in the graph is to replace each edge joining two vertices by $\ell$ edges
2392: joining these edges, thereby obtaining what we have denoted $G_\ell$ above.
2393: Given any such graph $G$, whether or not it is a member of a recursive family,
2394: our theorem above with (\ref{rgell}) and (\ref{pprime}) enables one to
2395: calculate $R(G_\ell,p)$. Here we shall present some illustrations of this.
2396: For definiteness, we consider the $L_y=2$ cyclic strip of the square lattice.
2397: Using our calculation of the reliability polynomial for the $L_y=2$ strip of
2398: the square lattice together with our theorem, we have calculated the
2399: reliability polynomial for the same strip with all edges replaced by
2400: $\ell$-fold multiple edges joining the same vertices.
2401:
2402: We have calculated the loci ${\cal B}$ in the $L_x \to \infty$ limit for these
2403: strips. The single point at $p=4/3$ where the locus ${\cal B}$ crosses the
2404: real $p$ axis for the $sq[2 \times \infty]$ case, shown in Fig. \ref{sqpxy2},
2405: corresponds to $\ell$ separate points for the $\ell$-fold edge replication of
2406: $sq[2 \times \infty]$. These can be calculated using (\ref{rgell}) with
2407: (\ref{pprime}), and we obtain
2408: %
2409: \beq
2410: p = 1 - \frac{1}{3^{1/\ell}}e^{\frac{i\pi (1+2k)}{\ell}} \ , \quad
2411: k=0,1,...,\ell-1
2412: \label{pcross_ell}
2413: \eeq
2414: %
2415: For example, for $\ell=2$, this yields the two values $p=1 \pm i/\sqrt{3}$, and
2416: so forth for higher values of $\ell$. The arc endpoints can be calculated in
2417: the same way.
2418:
2419:
2420: It is also of interest to study strip graphs in which a certain subset of the
2421: edges are replaced by $\ell$-fold replicated edges joining the same vertices.
2422: A simple example of this is already provided by the $L_y=2$ strips of the
2423: square lattice with cylindrical, torus, or Klein bottle boundary conditions,
2424: which have double transverse edges. We have calculated the reliability
2425: polynomials for the cyclic $L_y=2$ strips of the square lattice with
2426: $\ell$-fold multiple transverse edges. We denote these strip graphs as
2427: $sq[2 \times m, cyc., v\ell]$. We find
2428: %
2429: \beqs
2430: & & R(sq[L_y=2,m,cyc,v\ell],p)=p^{2m} \Biggl [
2431: (\alpha_{sqcyc2,v\ell,1})^m+(\alpha_{sqcyc2,v\ell,2})^m -
2432: 2(\alpha_{sqcyc2,v\ell,3})^m \cr\cr & & +
2433: m p^{-1}(1-p)^{\ell+1} \left \{ (\alpha_{sqcyc2,v\ell,1})^{m-1} \left (
2434: VL1(p,\ell) + VL2(p,\ell) \right ) \right. \cr\cr & &
2435: \left. + (\alpha_{sqcyc2,v\ell,2})^{m-1} \left ( VL1(p,\ell) - VL2(p,\ell)
2436: \right ) - (\alpha_{sqcyc2,v\ell,3})^{m-1} \right \} \Biggr ] \cr\cr & &
2437: \label{Rsqpxy2vl}
2438: \eeqs
2439: %
2440: where
2441: %
2442: \beqs
2443: & & \alpha_{sqcyc2,v\ell,j} = (2p)^{-1}\Biggl [ 2-p-(2-3p)(1-p)^\ell \cr\cr
2444: & & \pm [(1-(1-p)^\ell)((2-p)^2-(2-3p)^2(1-p)^\ell)]^{1/2} \Biggr ],
2445: \quad j=1,2
2446: \label{sqcycvl12}
2447: \eeqs
2448: %
2449: \beq
2450: \alpha_{sqcyc2,v\ell,3}=(1-p)^\ell
2451: \label{sqcycvl3}
2452: \eeq
2453: %
2454: and
2455: %
2456: \beq
2457: VL1(p,\ell)=\frac{(1-p)^{\ell-1}(2p-1)+1}{2p(1-p)^{\ell-1}}
2458: \label{vl1}
2459: \eeq
2460: %
2461: \beq
2462: VL2(p,\ell)=\frac{(2p-1)(3p-2)(1-p)^\ell-4+10p-7p^2
2463: +\frac{2-p}{(1-p)^{\ell-1} } }
2464: {2p [(1-(1-p)^\ell)((2-p)^2-(2-3p)^2(1-p)^\ell)]^{1/2}}
2465: \label{vl2}
2466: \eeq
2467: %
2468:
2469:
2470: \begin{figure}[hbtp]
2471: \centering
2472: \leavevmode
2473: \epsfxsize=4.0in
2474: \begin{center}
2475: \leavevmode
2476: \epsffile{sqpxy2v3.ps}
2477: \end{center}
2478: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
2479: of the $L_y=2$ strip of the square lattice with $\ell=3$-fold multiple
2480: transverse edges. For comparison, zeros of the reliability polynomial are
2481: shown for this strip with cyclic boundary conditions and $L_x=20$.}}
2482: \label{sqpxy2v3}
2483: \end{figure}
2484:
2485:
2486: \begin{figure}[hbtp]
2487: \centering
2488: \leavevmode
2489: \epsfxsize=4.0in
2490: \begin{center}
2491: \leavevmode
2492: \epsffile{sqpxy2v4.ps}
2493: \end{center}
2494: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to \infty$ limit
2495: of the $L_y=2$ strip of the square lattice with $\ell=4$-fold multiple
2496: transverse edges. For comparison, zeros of the reliability polynomial
2497: are shown for this strip with cyclic boundary conditions and $L_x=20$.}}
2498: \label{sqpxy2v4}
2499: \end{figure}
2500:
2501:
2502: \bigskip
2503:
2504: In Figs. \ref{sqpxy2v3} and \ref{sqpxy2v4} we show the loci ${\cal B}$ in the
2505: limit $L_x \to \infty$ for the cases $\ell=3,4$. In general, ${\cal B}$
2506: contains arcs protruding inward from points on the circle $|p-1|=1$ given by
2507: $p=1-\exp(2i \pi k/\ell)$ (cf. eq. (\ref{pzero_t2})) for $0 \le k \le \ell-1$.
2508: If and only if $\ell$ is even, then one of these points, namely the one
2509: corresponding to $k=\ell/2$, occurs at $p=2$, and the part of ${\cal B}$ that
2510: connects with this point is a line segment on the real axis extending from
2511: $p=2$ downward to an $\ell$-dependent lower limit.
2512:
2513:
2514: \subsection{$L_y=2$ Free Strip of the $sq_d$ Lattice with Multiple Transverse
2515: Edges}
2516:
2517: In this section we exhibit the first of several recursive families of lattice
2518: strip graphs that we have found with the property that some zeros of their
2519: reliability polynomials and, in the limit of infinite length, some portions of
2520: their accumulation sets ${\cal B}$, lie outside the disk $|p-1|$. This family
2521: is a one-parameter family generalizing the lowest-order graph found by Royle
2522: and Sokal. That graph is constructed as follows \cite{bcf}: start with the
2523: complete graph $K_4$ and choose two nonintersecting edges; replace each of
2524: these edges by six edges in parallel, i.e., joining the same pair of vertices.
2525: (Here the complete graph $K_n$ is defined as the graph with $n$ vertices such
2526: that every vertex is connected by one edge to every other vertex.) We had
2527: previously calculated Tutte and chromatic polynomials for strips composed of
2528: $K_4$ subgraphs \cite{w3,k,ka}. One of the equivalent ways of defining these
2529: strip graphs is to start with a strip of the square lattice and add edges
2530: connecting diagonally opposite vertices of each square, thereby replacing each
2531: square by a $K_4$; they were thus denoted strips of the $sq_d$ lattice, $sq_d(2
2532: \times L_x, BC_x)$, where $BC_x$ indicated the longitudinal boundary
2533: conditions. In a physical context, the Potts model on the $sq_d$ lattice is
2534: equivalent to the Potts model on the square lattice with next-nearest-neighbor
2535: spin-spin couplings. The longitudinal ($x$) and transverse ($y$) directions on
2536: this strip are taken to be horizontal and vertical, respectively. In
2537: particular, we consider a $sq_d$ strip with width $L_y=2$ and arbitrary length
2538: $L_x$. We then replace each vertical edge by six edges joining the same pair
2539: of vertices (leaving the horizontal and diagonal edges unchanged). An
2540: elementary proof \cite{k} shows that for free longitudinal boundary conditions
2541: this is a planar graph. We calculate the reliability polynomial and find for
2542: the numerator and denominator of the generating function
2543: %
2544: \beq
2545: {\cal N}(p,z)=p\biggl [ A_0 + A_1 z \biggr ]
2546: \label{numgamma_sqdv6}
2547: \eeq
2548: %
2549: where
2550: %
2551: \beq
2552: A_0 = -(p-2)(p^2-3p+3)(p^2-p+1)
2553: \label{a0sqdv6}
2554: \eeq
2555: %
2556: \beq
2557: A_1 = -p^2(p-1)^6(3p-4)
2558: \label{a1sqdv6}
2559: \eeq
2560: %
2561: \beq
2562: {\cal D}(p,z)=1 + b_1z + b_2z^2
2563: \label{densqdv6}
2564: \eeq
2565: %
2566: where
2567: %
2568: \beq
2569: b_1 = -p^2(6p^8-52p^7+200p^6-448p^5+644p^4-616p^3+391p^2-156p+32)
2570: \label{b1sqdv6}
2571: \eeq
2572: %
2573: \beq
2574: b_2 = -2p^5(p-2)(p-1)^8
2575: \label{b2sqdv6}
2576: \eeq
2577: %
2578: Using this exact calculation, one easily verifies that members of this family
2579: have reliability polynomials with zeros outside the disk $|p-1| \le 1$. The
2580: locus ${\cal B}$ for the infinite-length limit of this family is shown in
2581: Fig. \ref{k4y2v6k}. As is evident, ${\cal B}$ consists of arcs, and the tips of
2582: two complex-conjugate arcs on the upper and lower left end at the points
2583: %
2584: \beq
2585: p = 0.327752 \pm 0.747464i
2586: \label{k4y2v6k_endpoint}
2587: \eeq
2588: %
2589: which have
2590: %
2591: \beq
2592: |p-1| = 1.005296
2593: \label{k4y2v6kviolation}
2594: \eeq
2595: %
2596: These arc endpoints and their locally neighboring sections of their arcs thus
2597: lie outside of the disk $|p-1| \le 1$. Since ${\cal B}$ is the continuous
2598: accumulation set of the zeros of the reliability polynomial, it follows that as
2599: $L_x \to \infty$, infinitely many zeros of the reliability polynomial lie
2600: outside this disk. Interestingly, although the locus ${\cal B}$ does extend
2601: outside the disk $|p-1| \le 1$, the amount by which it does so is quite small,
2602: as is evident from the positions of the arc endpoints given in eq.
2603: (\ref{k4y2v6kviolation}). This property is also true of the other recursive
2604: families that we have found with zeros of $R(G,p)$ and limiting loci
2605: ${\cal B}$ lying outside the disk $|p-1| \le 1$.
2606:
2607: \begin{figure}[hbtp]
2608: \centering
2609: \leavevmode
2610: \epsfxsize=4.0in
2611: \begin{center}
2612: \leavevmode
2613: \epsffile{k4y2v6k.epsi}
2614: \end{center}
2615: \caption{\footnotesize{Singular locus ${\cal B}$ for the $L_x \to
2616: \infty$ limit of the $L_y=2$ strip of the $sq_d$ lattice with
2617: 6-fold multiple transverse edges. For comparison, zeros of the
2618: reliability polynomial are shown for this strip with free boundary
2619: conditions and $L_x=21$. Horizontal and vertical axes are $Re(p)$ and $Im(p)$
2620: with tick spacing 0.25, and the circle $|p-1|=1$ is shown with light dots.}}
2621: \label{k4y2v6k}
2622: \end{figure}
2623:
2624:
2625: We have also calculated the reliability polynomial for the analogous strip
2626: graphs with free longitudinal boundary conditions and with each longitudinal,
2627: rather than transverse, edge replaced by 6 edges joining the same pair of
2628: vertices. We find similar results. A plot of the zeros for $L_x=9$ is shown in
2629: Fig. \ref{k4y2h6k}. The arcs on ${\cal B}$ that extend outside the disk
2630: $|p-1| \le 1$ are again situated in the upper and lower left, and the endpoints
2631: of these arcs that extend outside this disk are
2632: %
2633: \beq
2634: p = 0.4346475 \pm 0.8266808i
2635: \label{k4y2h6k_endpoint}
2636: \eeq
2637: %
2638: with
2639: %
2640: \beq
2641: |p-1| = 1.001511
2642: \label{sqdh6_violation}
2643: \eeq
2644:
2645: \begin{figure}[hbtp]
2646: \centering
2647: \leavevmode
2648: \epsfxsize=4.0in
2649: \begin{center}
2650: \leavevmode
2651: \epsffile{k4y2h6k.epsi}
2652: \end{center}
2653: \caption{\footnotesize{Zeros of the reliability polynomial for the
2654: $L_y=2,L_x=9$ strip of the $sq_d$ lattice with 6-fold multiple longitudinal
2655: edges and free boundary conditions. Horizontal and vertical axes are $Re(p)$
2656: and $Im(p)$ with tick spacing 0.25, and the circle $|p-1|=1$ is shown with
2657: light dots.}}
2658: \label{k4y2h6k}
2659: \end{figure}
2660:
2661:
2662: \subsection{$L_y=2$ Free Strip of $K_4$ Subgraphs with Multiple Edges}
2663:
2664: We consider the strip shown in Fig. 8 of the Appendix of \cite{k}, but with
2665: free, instead of cyclic, longitudinal boundary conditions. In the notation of
2666: that figure, we replace each of the edges (labelled as $e_{ij}=(v_i,v_j)$)
2667: (1,2), (2,3), etc. on the top; (4,5), (5,6), etc. on the bottom, and (1,7),
2668: (5,8), (2,9), etc. in the middle with a 6-fold replicated edge. This is a
2669: family of planar graphs (as is the family in \cite{k} with cyclic boundary
2670: conditions). We calculate the reliability polynomial and find for the
2671: numerator and denominator of the generating function
2672: %
2673: \beq
2674: {\cal N}(p,z)=p\biggl [ A_0 + A_1 z \biggr ]
2675: \label{numgamma_sqd_appendix}
2676: \eeq
2677: %
2678: where
2679: %
2680: \beq
2681: A_0 = 1
2682: \label{k4k2e6_appendix0}
2683: \eeq
2684: %
2685: \beq
2686: A_1= p^2(p-1)(p-2)(p^2-p+1)(p^2-3p+3)(2p^6-14p^5+42p^4-70p^3+70p^2-42p+13)
2687: \label{k4k2e6_appendix1}
2688: \eeq
2689: %
2690: \beq
2691: {\cal D}(p,z)=1 + b_1z + b_2z^2
2692: \label{densqd_appendix}
2693: \eeq
2694: %
2695: where
2696: %
2697: \beqs
2698: & & b_1 = p^2(6p^{13}-86p^{12}+574p^{11}-2366p^{10}+6734p^9-14014p^8+22018p^7
2699: \cr\cr
2700: & & -26561p^6+24731p^5-17661p^4+9471p^3-3647p^2+918p-118)
2701: \label{b1_k4k2e6_appendix6}
2702: \eeqs
2703: %
2704: \beqs & & b_2 =
2705: p^4(1-p)^3(4p^{22}-100p^{21}+1201p^{20}-9224p^{19}+50876p^{18}-214544p^{17}
2706: \cr\cr & &
2707: +719020p^{16}-1965184p^{15}+4459758p^{14}-8511200p^{13}+13781747p^{12}-19045712p^{11}
2708: \cr\cr & &
2709: +22536020p^{10}-22849416p^9+19815434p^8-14628596p^7+9116492p^6
2710: \cr\cr & & -4732392p^5+2004288p^4-670184p^3+167360p^2-28048p+2401)
2711: \label{b2_k4k2e6_appendix6} \eeqs
2712: %
2713: Zeros of the reliability polynomial for the $L_x=9$ member of this family are
2714: shown in Fig. \ref{k4k2planark}. The tips of two complex-conjugate arcs, again
2715: on the upper and lower left, extend outside the disk $|p-1| \le 1$,
2716: ending at the points
2717: %
2718: \beq
2719: p = 0.4254334 \pm 0.8216255 i
2720: \label{k4k2_appendix6_endpoint}
2721: \eeq
2722: %
2723: which have
2724: %
2725: \beq
2726: |p-1| = 1.002594
2727: \label{k4k2_appendix6_violation}
2728: \eeq
2729: %
2730: Thus, as $L_x \to \infty$, an infinite number of zeros of the reliability
2731: polynomial accumulate to form part of the locus ${\cal B}$ which extends
2732: outside the disk $|p-1| \le 1$. As before, the amount by which some zeros and
2733: ${\cal B}$ lie outside the disk is small.
2734:
2735:
2736: \begin{figure}[hbtp]
2737: \centering \leavevmode \epsfxsize=4.0in
2738: \begin{center}
2739: \leavevmode \epsffile{k4k2planark.epsi}
2740: \end{center}
2741: \caption{\footnotesize{Zeros of the reliability polynomial for the
2742: $L_y=2, L_x=9$ free strip of $K_4$ subgraphs with 6-fold multiple
2743: longitudinal edges. Horizontal and vertical axes are $Re(p)$
2744: and $Im(p)$ with tick spacing 0.25, and the circle $|p-1|=1$ is shown with
2745: light dots.}}
2746: \label{k4k2planark}
2747: \end{figure}
2748:
2749:
2750:
2751: \section{Complete Graphs}
2752:
2753: The complete graph $K_n$ is the graph with $n$ vertices such that every vertex
2754: is connected by edges to every other vertex. The family of complete graphs is
2755: not a recursive family, and provides a contrast to the recursive families on
2756: which we have concentrated in this paper. The reliability polynomial for $K_n$
2757: is \cite{colbourn}
2758: %
2759: \beq
2760: R(K_n,p) = 1 - \sum_{j=1}^{n-1} {n-1 \choose j-1} (1-p)^{j(n-j)}R(K_j,p) \ .
2761: \label{rkn}
2762: \eeq
2763: %
2764: Since this family is not recursive, the reliability polynomial does not have
2765: the form (\ref{rgsum}) and our usual methods for determining an asymptotic
2766: accumulation set of zeros as the solution of the degeneracy in magnitudes of
2767: dominant $\alpha$ terms do not apply. However, we have studied the zeros of
2768: $R(K_n,p)$ for a wide range of values of $n$, and we find that these zeros
2769: typically form an oval-shaped pattern that is concave to the left and is
2770: roughly centered around $p=1$, somewhat similar to the exact results that we
2771: have established for our recursive families of graphs without multiple edges.
2772: As illustrations, we show in Fig. \ref{complete} the patterns of zeros for
2773: $n=5,10$, and 15. We observe that as $n$ increases, the oval moves outward
2774: from $p=1$. Note that since $K_n$ is a $\Delta$-regular graph with
2775: $\Delta=n-1$, it follows that the vertex degree $\Delta \to \infty$ as $n \to
2776: \infty$. This feature that $\Delta \to \infty$ as $n \to \infty$ is thus
2777: shared in common by the two families $K_n$ and the thick links $TL_n$. For the
2778: latter family, it is elementary that as $n \to \infty$, ${\cal B}$ is the
2779: circle $|p-1|=1$. However, our results show that it is not necessary for any
2780: vertex degree to go to infinity in order for a part of ${\cal B}$ to satisfy
2781: $|p-1| \ge 1$. Our results for the family of wheel graphs, $G_D(1 \times m)$
2782: also show that the property that a vertex degree goes to infinity as $|V| \to
2783: \infty$ is not sufficient for ${\cal B}$ to have a part with $|p-1| \ge 1$.
2784:
2785:
2786: \begin{figure}[hbtp]
2787: \centering
2788: \leavevmode
2789: \epsfxsize=4.0in
2790: \begin{center}
2791: \leavevmode
2792: \epsffile{complete.ps}
2793: \end{center}
2794: \caption{\footnotesize{Plot of zeros of the reliability polynomial
2795: $R(K_n,p)$ for $n=$ (a) 5 \ ($\Diamond$), (b) 10 \ ($\Box$), (c) 15 \
2796: ($\circ$).}}
2797: \label{complete}
2798: \end{figure}
2799:
2800:
2801: \section{Discussion of Results}
2802:
2803:
2804: In this section we discuss some general features of our results.
2805:
2806:
2807: \begin{itemize}
2808:
2809: \item
2810:
2811: For a given type of lattice and a given width and choice of transverse
2812: boundary conditions, we find that in the infinite-length limit, the reliability
2813: per vertex $r$ is the same for any choice of longitudinal boundary conditions.
2814:
2815: \item
2816:
2817: On general grounds, one expects that the higher the connectivity, and,
2818: equivalently for our present purposes, the higher the effective vertex degree
2819: $d_{eff}$, the larger the value of $r$ for a fixed $p$. In
2820: Fig. \ref{rcomparison} we plot $r$ for some strips of the square, triangular,
2821: and honeycomb lattices. We see that our expectation is confirmed by our exact
2822: results.
2823:
2824: \item
2825:
2826: Closely related to this, our results exhibit the property that the derivative
2827: $dr/dp$ at $p=0$ is an increasing function of $d_{eff}$. Except for the
2828: $L_y=1$ strips of the square lattice, i.e., the line and circuit graphs, we
2829: find that $dr/dp=0$ at $p=1$ for all strips that we have studied. In Table
2830: \ref{properties} we list some of the properties that we have found from our
2831: calculations.
2832:
2833: \item
2834:
2835: We can also comment on features of the zeros of the reliability polynomials and
2836: their continuous accumulation sets ${\cal B}$. We find that the zeros of the
2837: reliability polynomials for many strips studied do exhibit the property that
2838: $|p-1| \le 1$, as do their loci ${\cal B}$ in the limit of infinite length.
2839: However, motivated by \cite{bcf}, we have calculated reliability polynomials
2840: for strips such that the zeros of $R(G,p)$ violate the bound $|p-1| \le 1$ and
2841: such that, in the limit of infinite length, the continuous accumulation sets of
2842: zeros ${\cal B}$ extend outside of the disk $|p-1| \le 1$. An intriguing
2843: result is that the amounts by which some individual zeros, and the outer part
2844: of the loci ${\cal B}$, lie outside this disk are small.
2845:
2846: \item
2847:
2848: We find that in some cases, ${\cal B}$ crosses the real axis and hence defines
2849: a $p_c$, but in other cases, it does not. However, typically, even for
2850: families for which ${\cal B}$ does not cross the real axis, there are arc
2851: endpoints on ${\cal B}$ that lie close to this axis, thereby enabling one
2852: to define a $(p_c)_{eff}$ via extrapolation or just as the real part of the
2853: nearby endpoints. Our results have the property that these values of $p_c$ or
2854: $(p_c)_{eff}$ lie in the interval $p \in (1,2]$. It is interesting to discuss
2855: the relation between this result and the Brown-Colbourn theorem \cite{bc} that
2856: $R(G,p)$ does not have any real zeros in the region outside of the set $0 \cup
2857: (1,2]$. In general, the property that a polynomial has no zeros in an interval
2858: of the real axis does not exclude the possibility that as the degree of this
2859: polynomial goes to infinity, a resultant continuous accumulation set of zeros
2860: ${\cal B}$ crosses the real axis in this interval. A well-known case where
2861: this phenomenon does occur is provided by the Potts model partition function.
2862: This is a polynomial in $q$ and $a=e^K$ with non-negative coefficients. Thus,
2863: for fixed positive $q$, for either sign of the spin-spin coupling $J$, this
2864: partition function cannot have any zeros on the positive real $a$ axis.
2865: However, for the infinite limit of the square lattice, the locus ${\cal B}$ in
2866: this case does cross the positive real $a$ axis at the point $a_c =
2867: 1+\sqrt{q}$, which serves as the phase boundary between the paramagnetic and
2868: ferromagnetic phases occupying the respective intervals $1 \le a \le a_c$ and
2869: $a_c \le a \le \infty$.
2870:
2871: \item
2872:
2873: For all of the lattice strip graphs for which we have calculated $R(G,p)$ and
2874: ${\cal B}$, we find that ${\cal B}$ is independent of the longitudinal boundary
2875: conditions, although it depends on the transverse boundary conditions and the
2876: type of lattice. This requires that the dominant $\alpha$ in the case of
2877: cyclic strips must arise from the subset of the $\alpha$'s with coefficient
2878: $c^{(0)}$ that coincides with the set of $\alpha$'s for the corresponding strip
2879: with free longitudinal boundary conditions. The property that, for a given
2880: type of lattice strip, the locus ${\cal B}$ is independent of the longitudinal
2881: boundary conditions is quite different from what we found for the corresponding
2882: continuous accumulation sets of zeros of chromatic polynomials of lattice
2883: strips \cite{bcc}, \cite{w}-\cite{k}, which do depend on both longitudinal and
2884: transverse boundary conditions.
2885:
2886: \item
2887:
2888: For all of the lattice strip graphs for which we have calculated $R(G,p)$ and
2889: ${\cal B}$, we find that this locus consists of arcs (and in some cases a line
2890: segment) and does not enclose regions in the complex $p$ plane. In some simple
2891: cases ${\cal B}$ is connected, but in general, it may consist of several
2892: disjoint components. Again, this is quite different from what we found for the
2893: loci ${\cal B}$ for chromatic polynomials of lattice strips; for those, a
2894: sufficient (not necessary) condition that this locus encloses areas is that the
2895: strip has periodic longitudinal boundary conditions \cite{w,wcyl,bcc}.
2896:
2897: \item
2898:
2899: In all of the cases for which we have calculated exact results, whenever
2900: ${\cal B}$ consists of an arc of a circle, then this circle is centered at
2901: $p=1$. More generally, even if ${\cal B}$ is not an arc of a circle, it is
2902: roughly centered around $p=1$. For families without multiple edges, we find
2903: that the component(s) of ${\cal B}$ is (are) roughly concave toward the left.
2904:
2905: \item
2906:
2907: The width $L_y \to \infty$ limit of loci ${\cal B}$ for the infinite-length
2908: lattice strips is not necessarily the same as the continuous accumulation locus
2909: of the zeros of the reliability polynomial for the two-dimensional
2910: thermodynamic limit $L_x \to \infty$, $L_y \to \infty$ with $L_y/L_x$ fixed to
2911: a finite nonzero constant. This is clear since for any finite $L_y$ regardless
2912: of how large, the infinite-length strip is a quasi-one-dimensional system,
2913: characterized by the ratio $L_y/L_x = 0$. Thus, for example, the Potts model
2914: on these strips has a paramagnetic-ferromagnetic (PM-FM) phase transition point
2915: only at $K=\infty$, i.e., $v=\infty$, while for 2D lattices, it has such a
2916: PM-FM phase transition point at finite $v$. However, in our previous studies
2917: of continuous accumulation sets of loci of Potts model partition functions
2918: (e.g., \cite{a}-\cite{sdg} and our earlier work referred to therein on
2919: complex-temperature phase diagrams for Ising models) we have found that it is
2920: possible to gain some insight into certain features of the continuous
2921: accumulation set of zeros of the Potts model for the two-dimensional
2922: thermodynamic limit from studies of wider and wider infinite-length strips.
2923:
2924: In this context, we note that for the (infinite) square lattice, the PM-FM
2925: phase transition point for the Potts model is given by $v^2=q$ \cite{wurev}.
2926: This transition point is reflected in a nonanalyticity in the free energy of
2927: the Potts model at the corresponding temperature variable $v$. As we have
2928: discussed and studied in earlier papers, e.g., \cite{chisq,chitri}, when one
2929: generalizes $v$ to complex values, one sees that a singular point for a real
2930: physical value of $v$, such as the PM-FM transition point, is associated with
2931: the fact that a complex-temperature phase boundary crosses the real physical
2932: $v$ axis at this point. At the crossing points there are physical or
2933: complex-temperature singularities in the free energy \cite{chisq}-\cite{p2}.
2934: The boundaries on ${\cal B}$ (here as a function of $v$ for fixed $q$, but more
2935: generally, also as a function of $q$ for fixed $v$) arise as the continuous
2936: accumulation set of the complex-temperature zeros of the partition function,
2937: analogous to the origin of the locus ${\cal B}$ studied here as continuous
2938: accumulation set of the zeros of the reliability polynomial. Connected with
2939: this, the physical paramagnetic, ferromagnetic, and antiferromagnetic (AFM)
2940: phases have complex-temperature extensions, and these are separated from each
2941: other by the phase boundaries ${\cal B}$ in the $v$ plane. We recall that for
2942: the Potts model on the square lattice, the PM and FM phases occupy the
2943: intervals $0 \le v \le \sqrt{q}$ and $\sqrt{q} \le v \le \infty$, respectively.
2944: Analytically continuing $q$ to real values, using eq. (\ref{cluster}), we infer
2945: that in the limit $q \to 0$ that yields the reliability polynomial, as given by
2946: eq. (\ref{rz}), the PM phase therefore contracts to a point at $v=0$, which
2947: corresponds, via the relation $p=v/(1+v)$ in eq. (\ref{pek}), to the point
2948: $p=0$, while the FM phase expands to occupy the interval $0 \le v \le \infty$,
2949: corresponding to the interval $0 \le p \le 1$. As is evident, e.g., in the
2950: square-lattice Ising model and more generally for the 2D $q$-state Potts model,
2951: the complex-temperature extension of the FM phase includes part of the negative
2952: real $v$ axis extending to $v=-\infty$ (as well as the outerlying region $|v|
2953: \to \infty$ in the $v$ plane). This semi-infinite interval of large negative
2954: real values of $v$ is mapped, via the transformation $p=v/(1+v)$, to a finite
2955: interval of values of $p \ge 1$. Points in the outer region $v=Re^{i\theta}$
2956: with $R >> 1$ are mapped via this transformation to points in the vicinity of
2957: $p=1$, namely $p \simeq 1- R^{-1}e^{-i\theta}$. Since (in the case $q \to 0$
2958: relevant here) all of the points mentioned above in the $v$ plane, namely (i)
2959: the interval $0 \le v \le \infty$, (ii) the interval of large negative real
2960: $v$, and (iii) the semi-infinite annular region $v=Re^{i\theta}$ with $R \to
2961: \infty$ are in the complex-temperature extension of the FM phase of the Potts
2962: model, it follows that the images of these sets of points are all in the same
2963: analytically connected region of the $p$ plane as defined by the locus ${\cal
2964: B}$. We can draw on another source of information about this boundary as
2965: follows. Arguments have been given that for the Potts model on the (infinite)
2966: square lattice, the paramagnetic-antiferromagnetic (PM-AFM) transition occurs
2967: at a (physical) root of the equation $v(v+4)+q=0$ \cite{baxter82}. Setting
2968: $q=0$ and $v=p/(1-p)$ as in (\ref{rz}) and (\ref{pek}), one finds the solutions
2969: $v=0$ and $v=-4$, corresponding respectively to $p=0$, the solution already
2970: found above, and $p=4/3$. Hence for the square lattice, the image, under the
2971: mapping $v \to p=v/(1+v)$, of the complex-temperature extension of the FM phase
2972: includes the real interval $0 \le p \le 4/3$. The complementary intervals
2973: $-\infty \le p \le 0$ and $4/3 \le p \le \infty$ would be part of the image
2974: under this map of an unphysical phase denoted the ``O'' (for ``other'') phase
2975: in Ref. \cite{chisq} and our subsequent papers on complex-temperature phase
2976: diagrams. Using the fact that the complex-temperature phase boundary separates
2977: these various phases, we infer that for the reliability polynomial on the
2978: square lattice, in the thermodynamic limit, the zeros accumulate on a locus
2979: ${\cal B}$ that is a closed curve crossing the real axis at $p=0$ and $p=4/3$
2980: and separating the interior region from the exterior, which latter includes the
2981: intervals $-\infty \le p \le 0$ and $4/3 \le p \le \infty$. In an obvious
2982: nomenclature extending that of \cite{chisq}, we denote the regions in the
2983: interior and exterior of this closed curve as $R_{FM}$ and $R_O$, indicating
2984: the correspondence with the complex-temperature phases of the Potts model.
2985: Just as the free energy is an analytic function in the interior of a
2986: complex-temperature phase, the function $r$ that we have introduced in
2987: \cite{ka3} and here in eq. (\ref{r}) is an analytic function in the regions
2988: bounded by the locus ${\cal B}$. One expects qualitatively similar results to
2989: hold for the regions of analyticity of $r$ on (the thermodynamic limits of)
2990: lattice graphs of dimensionality $d \ge 3$. For the (infinite) square lattice,
2991: %
2992: \beq
2993: p_c(sq) = \frac{4}{3}
2994: \label{pcsq}
2995: \eeq
2996: %
2997: We shall discuss how this compares with our exact results for $p_c$ values on
2998: infinite-length finite-width strips below.
2999:
3000: Similar reasoning can be applied to obtain inferences for the loci ${\cal B}$
3001: and the corresponding region diagrams of analyticity of the function $r$ on
3002: other (infinite) 2D lattices. For the triangular and honeycomb lattices, the
3003: PM-FM phase transition point is given by a (physical) root of the equations
3004: $v^2(v+3)-q=0$ and $v^3-3qv-q^2=0$, respectively \cite{kj}. Evaluating these
3005: for $q=0$ and $v=p/(1-p)$ yields for the triangular lattice, the solutions
3006: $p=0$ and $p=3/2$, so that
3007: %
3008: \beq
3009: p_c(tri)=\frac{3}{2}
3010: \label{pctri}
3011: \eeq
3012: %
3013: For the honeycomb lattice, the same evaluation yields just the solution $p=0$
3014: and hence does not determine $p_c(hc)$ (By the reasoning given above, one can
3015: infer that $p_c(hc) > 1$). The fact that the point $p=0$ on ${\cal B}$ is
3016: common to all of the three lattices - square, triangular, and honeycomb -
3017: follows because it is the image under the map (\ref{pek}) of the point $v=K=0$.
3018: This point corresponds to infinite temperature, where, as one can see from
3019: eqs. (\ref{zfun}) and (\ref{ham}), the spin-spin interaction does not
3020: contribute to $Z(G,q,v)$, which reduces to $q^{|V|}$ for all of these lattices
3021: (indeed for any graph $G$).
3022:
3023:
3024: \item
3025:
3026:
3027: For a given type of lattice strip (square, triangular, honeycomb, etc.) and a
3028: given set of transverse boundary conditions, we may consider the sequence of
3029: loci ${\cal B}$ for each width, $L_y$, and inquire whether these approach a
3030: limiting locus as $L_y \to \infty$. Our results are summarized in Table
3031: \ref{properties} and are consistent with the hypothesis that such a limiting
3032: locus exists. As part of this analysis, we have investigated, for each type of
3033: strip family, how $p_c$ or $(p_c)_{eff}$ depends on $L_y$. We find that, for a
3034: given type of lattice strip and transverse boundary conditions, if a value of
3035: $p_c$ exists, i.e., if ${\cal B}$ crosses (intersects) the real axis, then
3036: $p_c$ decreases as $L_y$ increases. However, if, for a given type of lattice
3037: strip and transverse boundary conditions, one considers the set of values of
3038: both $p_c$ and $(p_c)_{eff}$ (the latter for widths where ${\cal B}$ comes
3039: close to but does not cross the real axis), then the dependence on $L_y$ does
3040: not appear to be monotonic. For example, for the infinite-length limit of the
3041: strip of the square lattice with free transverse boundary conditions (and any
3042: longitudinal boundary conditions), for the width $L_y=2$, we have shown that
3043: $p_c=4/3$. Interestingly, this is the same value as inferred above for the
3044: infinite square lattice defined via the usual thermodynamic limit. For the
3045: square-lattice strip with width $L_y=3$, the locus ${\cal B}$ does not cross
3046: the real axis, but, as is evident in Fig. \ref{sqpxy3}, the inner endpoints of
3047: the arcs on ${\cal B}$ are sufficiently close to the real axis that one can
3048: infer a $(p_c)_{eff}$, and it is $(p_c)_{eff} \simeq 1.335$, which is slightly
3049: greater than the value of $p_c$ for the $L_y=2$ strip and the inferred value of
3050: $p_c(sq)$. As is evident in Table \ref{properties}, the values of $p_c$ are not
3051: as close to $p_c(sq)$ for the square-lattice strips of the corresponding widths
3052: with periodic transverse boundary conditions or self-dual boundary conditions,
3053: as compared with free transverse boundary conditions, although they are again
3054: consistent with approaching $p_c(sq)=4/3$ as the width increases. For the
3055: $L_y=2$ strip of the triangular lattice, we find the exact result $p_c=3/2$.
3056: Again, interestingly, this is equal to the inferred value of $p_c(tri)$ for the
3057: infinite triangular lattice defined via the thermodynamic limit. We have
3058: encountered this sort of situation before. For example, for the chromatic
3059: polynomial $P(G,q) = Z(G,q,v=-1)$ and its asymptotic limiting function as $|V|
3060: \to \infty$, $W(\{G\},q)$ (the degeneracy per vertex for the Potts model), we
3061: found that for the infinite-length strip of the triangular lattice with width
3062: $L_y=3$ and toroidal or Klein-bottle boundary conditions, $q_c$ (defined as the
3063: maximal point where the continuous accumulation set of zeros of $P(G,q)$
3064: intersects the real $q$ axis) is $q_c=4$ \cite{tk}, the same value as for the
3065: infinite triangular lattice. Similarly, for infinite-length limits of
3066: self-dual strips of the square lattice, we found $q_c=3$ \cite{dg,sdg}, the
3067: same value as for the infinite square lattice.
3068:
3069: As regards the behavior of ${\cal B}$ near $p=0$, our results are also
3070: consistent with the possibility that for each type of lattice, as the strip
3071: width $L_y \to \infty$, the respective loci ${\cal B}$ have, on the left,
3072: complex-conjugate arc endpoints that come together and join at $p=0$, as
3073: motivated by the discussion above for the infinite 2D lattices. The observed
3074: feature that this occurs only as a limit, in contrast to the crossing on the
3075: right at the respective values of $p_c$, can be understood via the relation
3076: (\ref{rz}) and the aforementioned fact that the infinite-length finite-width
3077: strips are quasi-1D systems as far as their thermodynamic behavior is
3078: concerned. Since, as we are noted above, the interior of ${\cal B}$ for the
3079: reliability polynomial on the 2D lattice corresponds to the image of the
3080: complex-temperature extension of the FM phase of the 2D Potts model (for $q \to
3081: 0$), and since the argument that this must be separated from other phases
3082: relies on the existence of nonzero ferromagnetic long range order
3083: (magnetization), it follows that for quasi-1D systems, where for a short-range
3084: interaction like that in (\ref{ham}), the standard Peierls argument shows that
3085: there is no long range order, there is no FM phase. This yields a plausible
3086: explanation of the features that we observe in our exact solutions for the
3087: infinite-length, finite-width strip graphs, that the loci ${\cal B}$ do not
3088: enclose regions in the $p$ plane.
3089:
3090: \end{itemize}
3091:
3092:
3093: \begin{figure}[hbtp]
3094: \centering \leavevmode \epsfxsize=4.0in
3095: \begin{center}
3096: \leavevmode
3097: \epsffile{rcomparison.ps}
3098: \end{center}
3099: \caption{\footnotesize{Comparison of asymptotic reliability per vertex $r$ for
3100: the $L_x \to \infty$ limit of several lattice strips. From bottom to top, the
3101: curves refer to
3102: (a) $sq[1 \times \infty,FBC_y]$ with $d_{eff}=2$,
3103: (b) $hc[2 \times \infty,FBC_y]$ with $d_{eff}=2.5$,
3104: (c) $sq[2 \times \infty,FBC_y]$ with $d_{eff}=3$,
3105: (d) $sq[2 \times \infty,PBC_y]$ with $d_{eff}=4$,
3106: (e) $G_D[1 \times \infty]$, same $r$ as for
3107: $tri[2 \times \infty]$, with $d_{eff}=4$,
3108: (f) $G_D[2 \times \infty]$, with $d_{eff}=4$.}}
3109: \label{rcomparison}
3110: \end{figure}
3111:
3112:
3113: \begin{table}
3114: \caption{\footnotesize{Properties of reliability polynomials and resultant loci
3115: ${\cal B}$ including $p_c$ or $(p_c)_{eff}$ for various lattice strip
3116: graphs. The notation $sqdl$ and $D$ refer to the self-dual strips of the square
3117: lattice.}}
3118: \begin{center}
3119: \begin{tabular}{|c|c|c|c|c|c|c|}
3120: \hline\hline $G_s$ & $L_y$ & $BC_y$ & $BC_x$ & $d_{eff}$ & $N_\alpha$ &
3121: $p_c$ \\ \hline\hline
3122: sq & 1 & F & F & 2 & 1 & $-$ \\ \hline
3123: sq & 1 & F & P & 2 & 1 & $-$ \\ \hline
3124: sq & 2 & F & F & 3 & 2 & 4/3 \\ \hline
3125: sq & 2 & F & (T)P & 3 & 3 & 4/3 \\ \hline
3126: sq & 3 & F & F &10/3 & 4 & 1.335 \\ \hline
3127: sq & 3 & F & (T)P &10/3 & 4 & 1.335 \\ \hline\hline
3128: sq & 2 & P & F & 4 & 2 & 2 \\ \hline
3129: sq & 2 & P & (T)P & 4 & 3 & 2 \\ \hline
3130: sq & 3 & P & F & 4 & 3 & 1.402 \\ \hline
3131: sq & 3 & P & (T)P & 4 & 11 & 1.402 \\ \hline
3132: sq & 4 & P & F & 4 & 6 & 1.384 \\ \hline
3133: sqdl & 1 & D & F & 4 & 2 & 3/2 \\ \hline
3134: sqdl & 1 & D & P & 4 & 3 & 3/2 \\ \hline
3135: sqdl & 2 & D & F & 4 & 5 & 1.4 \\ \hline
3136: sqdl & 2 & D & P & 4 & 10 & 1.4 \\ \hline
3137: sqdl & 3 & D & F & 4 & 14 & 1.386 \\ \hline
3138: sqdl & 3 & D & P & 4 & 35 & 1.386 \\ \hline\hline
3139: tri & 2 & F & F & 4 & 2 & 3/2 \\ \hline
3140: tri & 2 & F & (T)P & 4 & 3 & 3/2 \\ \hline\hline
3141: hc & 2 & F & F & 5/2 & 2 & 6/5 \\ \hline
3142: hc & 2 & F & (T)P & 5/2 & 3 & 6/5 \\ \hline\hline
3143: \end{tabular}
3144: \end{center}
3145: \label{properties}
3146: \end{table}
3147:
3148:
3149: \section{Conclusions}
3150:
3151: In this paper we have presented exact calculations of reliability polynomials
3152: $R(G,p)$ for lattice strips $G$ of fixed width and arbitrarily great length
3153: with various boundary conditions. We have introduced the notion of a
3154: reliability per vertex, $r(\{G\},p)$ and have calculated this exactly for the
3155: infinite-length limits of various lattice strip graphs. We have also studied
3156: the zeros of $R(G,p)$ in the complex $p$ plane and determine exactly the
3157: asymptotic accumulation set of these zeros ${\cal B}$, across which $r(\{G\})$
3158: is nonanalytic. We have observed and discussed several general features of the
3159: $r$ functions and the loci ${\cal B}$ for these families of graphs.
3160:
3161: Acknowledgment: This research was partially supported by the NSF grant
3162: PHY-00-98527.
3163:
3164: \begin{thebibliography}{99}
3165:
3166: \bi{shannon}{E. F. Moore and C. E. Shannon, Reliable circuits using less
3167: reliable components, J. Franklin Inst. {\bf 262}, 191-208, 281-297 (1956).}
3168:
3169: \bi{moskowitz}{F. Moskowitz, The analysis of redundancy networks, AIEE
3170: Trans. Commun. Electronics {\bf 39}, 627-632 (1958).}
3171:
3172: \bi{mine}{H. Mine, Reliability of physical systems, IRE Trans. Circuit
3173: Theory CT-6, 138-151 (1959).}
3174:
3175: \bi{kelmans}{A. Kel'mans, Connectivity of probabilistic networks,
3176: Automation and Remote Control {\bf 26}, 564-573 (1965).}
3177:
3178: \bi{colbourn}{C. Colbourn, {\it The Combinatorics of Network
3179: Reliability} (Oxford Univ. Press, New York, 1987).}
3180:
3181: \bi{pb}{J. Provan and M. Ball, The complexity of counting cuts and of computing
3182: the probability that a graph is connected, SIAM J. on Computing, {\bf 12},
3183: 777-788 (1983).}
3184:
3185: \bi{lp}{M. Lomonosov and V. Polesskii, An upper bound for the reliability of
3186: information networks, Problems of Information Transmission {\bf 7}, 337-339
3187: (1971).}
3188:
3189: \bi{lp2}{M. Lomonosov and V. Polesskii, Lower bound of network reliability,
3190: Problems of Information Transmission {\bf 8}, 118-123 (1972).}
3191:
3192: \bi{kruskal}{J. Kruskal, The number of simplices in a complex, in
3193: R. Bellman, ed. {\it Mathematical Optimization Techniques} (Univ. California
3194: Press, 1963), 251-278.}
3195:
3196: \bi{stanley1}{R. P. Stanley, The upper bound conjecture and Cohen-Macaulay
3197: rings, Studies in Appl. Math. {\bf 14}, 135-142 (1975).}
3198:
3199: \bi{bp}{M. Ball and J. Provan, Bounds on the reliability polynomial for
3200: shellable independence systems, SIAM Journal on Algebraic and Discrete
3201: Methods {\bf 3}, 166-181 (1982).}
3202:
3203: \bi{bcr}{M. Ball, C. Colbourn, and J. Provan, ``Network reliability'', in {\it
3204: Handbook of Operations Research: Network Models} (Elsevier North-Holland,
3205: 1995), pp. 673-762.}
3206:
3207: \bi{bc96}{J. Brown and C. Colbourn, Non-Stanley bounds for network
3208: reliability, J. Alg. Combinatorics {\bf 5}, 18-36 (1996).}
3209:
3210: \bi{ka3}{S.-C. Chang and R. Shrock, Tutte polynomials and related asymptotic
3211: limiting functions for recursive families of graphs, in {\it Proceedings of
3212: the Workshop on Tutte Polynomials}, Centre de Recerca Matem\`{a}tica, Univ.
3213: Aut\'{o}noma de Barcelona, (Sept. 2001), Advances in Applied. Math., in press
3214: (math-ph/0112061).}
3215:
3216: \bi{bc}{J. Brown and C. Colbourn, Roots of the reliability polynomial,
3217: SIAM J. Discrete Math. {\bf 4}, 571-585 (1992).}
3218:
3219: \bi{wagner}{D. Wagner, Zeros of reliability polynomials and $f$-vectors of
3220: matroids, Combin. Prob., and Computing {\bf 9}, 167-190 (2000).}
3221:
3222: \bibitem{sbounds}{A. Sokal, Bounds on the complex zeros of (di)-chromatic
3223: polynomials and Potts-model partition functions, Combin. Prob. Comput. {\bf
3224: 10}, 41-77 (2001). See also \cite{hpp}.}
3225:
3226: \bi{hpp}{Y.-B. Choe, J. Oxley, A. Sokal, and D. Wagner, Homogeneous
3227: multivariate polynomials with the half-plane property, math.CO/0203034.}
3228:
3229: \bi{bcf}{After the submission of the original version of this manuscript
3230: (August, 2002), a referee informed us of the Royle-Sokal counterexamples
3231: to the Brown-Colbourn conjecture in Nov. 2002. We thank A. Sokal,
3232: C. Colbourn, and J. Brown for private communications on this. See
3233: G. Royle and A. Sokal, math.CO/0103199.}
3234:
3235: \bi{tutte54}{W. T. Tutte, A contribution to the theory of chromatic
3236: polynomials, Canad. J. Math. {\bf 6}, 80-91 (1954). Sadly, we must record
3237: here that W. T. Tutte died on 2 May, 2002, during the completion of this
3238: manuscript.}
3239:
3240: \bi{tutte66}{W. T. Tutte, On the algebraic theory of graph colourings,
3241: J. Combin. Theory {\bf 1}, 15-50 (1966).}
3242:
3243: \bi{tutte67}{W. T. Tutte, On dichromatic polynomials, J. Combin. Theory
3244: {\bf 2}, 301-320 (1967).}
3245:
3246: \bi{tutte84}{W. T. Tutte {\it Graph Theory}, vol. 21 of
3247: {\it Encyclopedia of Mathematics and its Applications}, ed. Rota,
3248: G. C. (Addison-Wesley, New York, 1984).}
3249:
3250: \bi{bo}{T. Brylawski and J. Oxley, The Tutte polynomial and its
3251: applications, Chap. 6 in {\it Matroid Applications, Encyclopedia of Mathematics
3252: and its Applications}, vol. 40 (ed. N. White) (Cambridge University Press,
3253: Cambridge, 1992), pp. 123-225.}
3254:
3255: \bi{bbook}{N. L. Biggs, {\it Algebraic Graph Theory} (2nd ed., Cambridge
3256: Univ. Press, Cambridge, 1993).}
3257:
3258: \bi{welsh}{D. J. A. Welsh, {\it Complexity: Knots, Colourings, and
3259: Counting}, London Math. Soc. Lect. Note Ser. 186 (Cambridge University Press,
3260: Cambridge, 1993).}
3261:
3262: \bi{boll}{B. Bollob\'as, {\it Modern Graph Theory} (Springer, New
3263: York, 1998).}
3264:
3265: \bi{a}{R. Shrock, Exact Potts model partition functions on ladder
3266: graphs, Physica A {\bf 283}, 388-446 (2000).}
3267:
3268: \bi{ta}{S.-C. Chang and R. Shrock, Exact Potts model partition
3269: functions on strips of the triangular lattice, Physica A {\bf 286},
3270: 189-238 (2000).}
3271:
3272: \bi{bcc}{R. Shrock, Chromatic polynomials and their zeros and asymptotic
3273: limits for families of graphs, in the {\it Proceedings of the 1999 British
3274: Combinatorial Conference, BCC99} (July, 1999), Discrete Math. {\bf 231},
3275: 421-446 (2001).}
3276:
3277: \bi{cf}{S.-C. Chang and R. Shrock, Structural properties of Potts
3278: model partition functions and chromatic polynomials for lattice strips,
3279: Physica A {\bf 296}, 131-182 (2001).}
3280:
3281: \bi{hca}{S.-C. Chang and R. Shrock, Exact Potts model partition
3282: functions on strips of the honeycomb lattice, Physica A {\bf 296}, 183-233
3283: (2001).}
3284:
3285: \bi{s3a}{S.-C. Chang and R. Shrock, Exact Potts model partition
3286: functions on wider arbitrary-length strips of the square lattice, Physica
3287: A {\bf 296}, 234-288 (2001).}
3288:
3289: \bi{ka}{S.-C. Chang and R. Shrock, Exact Partition function for the
3290: Potts model with next-nearest neighbor couplings on strips of the square
3291: lattice, Int. J. Mod. Phys. B {\bf 15}, 443-478 (2001).}
3292:
3293: \bi{jz}{S.-C. Chang and R. Shrock, Zeros of Jones polynomials for
3294: families of knots and links, Physica A {\bf 301}, 196-218 (2001).}
3295:
3296: \bi{dg}{S.-C. Chang and R. Shrock, Potts model partition functions for
3297: self-dual families of graphs, Physica A {\bf 301}, 301-329 (2001).}
3298:
3299: \bi{sdg}{S.-C. Chang and R. Shrock, Complex-temperature phase diagrams
3300: for the $q$-state Potts model on self-dual families of graphs and the
3301: nature of the $q \to \infty$ limit, Phys. Rev. E {\bf 64}, 066116 (2001).}
3302:
3303: \bi{ts}{S.-C. Chang, J. Salas, and R. Shrock,
3304: Exact Potts model partition functions for strips of the square lattice,
3305: J. Stat. Phys. {\bf 107} (2002) 1207-1253.}
3306:
3307: \bi{wurev}{F. Y. Wu, The Potts Model, Rev. Mod. Phys. {\bf 54}, 235-268
3308: (1982).}
3309:
3310: \bi{wu87}{F. Y. Wu, Potts model and graph theory, J. Stat. Phys. {\bf 52},
3311: 99-112 (1988).}
3312:
3313: \bibitem{fk}{C. Fortuin and P. Kasteleyn, On the random cluster model.
3314: I. Introduction and relation to other models, Physica {\bf 57}, 536-564
3315: (1972).}
3316:
3317: \bi{w}{R. Shrock and S.-H. Tsai, Asymptotic limits and zeros of chromatic
3318: polynomials and ground state entropy of Potts antiferromagnets,
3319: Phys. Rev. E {\bf 55}, 5165-5179 (1997).}
3320:
3321: \bi{strip}{M. Ro\v{c}ek, R. Shrock, and S.-H. Tsai, Chromatic polynomials
3322: for families of strip graphs and their asymptotic limits, Physica
3323: {\bf A252}, 505-546 (1998).}
3324:
3325: \bi{hs}{R. Shrock and S.-H. Tsai, Ground state entropy of Potts
3326: antiferromagnets on homeomorphic families of strip graphs, Physica
3327: {\bf A259}, 315-348 (1998).}
3328:
3329: \bi{strip2}{M. Ro\v{c}ek, R. Shrock, and S.-H. Tsai, Chromatic polynomials
3330: on $J(\prod H)I$ strip graphs and their asymptotic limits, Physica
3331: {\bf A259}, 367-387 (1998).}
3332:
3333: \bi{wcyl}{R. Shrock and S.-H. Tsai, Ground state degeneracy of Potts
3334: antiferromagnets on 2D lattices: approach using infinite cyclic strip
3335: graphs, Phys. Rev. E {\bf 60}, 3512-3515 (1999).}
3336:
3337: \bi{wcy}{R. Shrock and S.-H. Tsai, Exact partition functions for Potts
3338: antiferromagnets on cyclic lattice strips, Physica A {\bf 275}, 429-449
3339: (2000).}
3340:
3341: \bi{pm}{R. Shrock, $T=0$ partition functions for Potts antiferromagnets on
3342: M\"obius strips and effects of graph topology, Phys. Lett. {\bf A261},
3343: 57-62 (1999).}
3344:
3345: \bi{tk}{N. L. Biggs and R. Shrock,
3346: $T=0$ partition functions for Potts antiferromagnets on square lattice
3347: strips with (twisted) periodic boundary conditions,
3348: J. Phys. A (Letts) {\bf 32}, L489-L493 (1999).}
3349:
3350: \bi{k}{S.-C. Chang and R. Shrock, Ground state entropy of the Potts
3351: antiferromagnet with next-nearest-neighbor spin-spin couplings on strips of
3352: the square lattice, Phys. Rev. E {\bf 62}, 4650-4664 (2000).}
3353:
3354: \bi{w3}{R. Shrock and S.-H. Tsai, Ground state entropy of Potts
3355: antiferromagnets: bounds, series, and Monte Carlo measurements, Phys. Rev.
3356: E {\bf 56}, 2733-2737 (1997).}
3357:
3358: \bi{chisq}{V. Matveev and R. Shrock, Complex-Temperature singularities of the
3359: susceptibility in the $d=2$ Ising model: square lattice, J. Phys. A {\bf 28}
3360: 1557-1583 (1995).}
3361:
3362: \bi{chitri}{V. Matveev and R. Shrock, Complex-temperature singularities in
3363: the $d=2$ Ising model: triangular and honeycomb lattices, J. Phys. A {\bf 29},
3364: 803-823 (1996).}
3365:
3366: \bi{pfef}{V. Matveev and R. Shrock, Complex-Temperature Singularities in
3367: Potts Models on the Square Lattice, Phys. Rev. E {\bf 54}, 6174-6185 (1996).}
3368:
3369: \bi{hcl}{H. Feldman, R. Shrock, and S.-H. Tsai, A mapping relating complex
3370: and physical temperatures in the 2D $q$-State Potts model and some
3371: applications, J. Phys. A (Lett.) {\bf 30}, L663-668 (1997).}
3372:
3373: \bi{p2}{H. Feldmann, A. J. Guttmann, I. Jensen, R. Shrock, and S.-H. Tsai,
3374: Study of the Potts model on the honeycomb and triangular lattices:
3375: low-temperature series and partition function zeros, J. Phys. A {\bf 31}
3376: 2287-2310 (1998).}
3377:
3378: \bi{ss}{J. Salas and A. Sokal, Absence of phase transition for
3379: antiferromagnetic Potts models via the Dobrushin uniqueness theorem, J.
3380: Stat. Phys. {\bf 86}, 551-579 (1997).}
3381:
3382: \bi{baxter82}{R. J. Baxter, Critical antiferromagnetic square-lattice
3383: Potts model, Proc. Roy. Soc. London, Ser. A {\bf 383}, 43-54 (1982).}
3384:
3385: \bi{saleur}{The results obtained in Ref. \cite{baxter82} involved a
3386: mapping from the Potts model to a six-vertex model. A complication pertaining
3387: to a mismatch between the boundary conditions for the Potts and vertex model
3388: was noted there and in H. Saleur, Zeros of chromatic
3389: polynomials: a new approach to Beraha conjecture using quantum groups,
3390: Commun. Math. Phys. {\bf 132}, 657-679 (1990).}
3391:
3392: \bi{kj}{D. Kim and R. Joseph, Exact transition temperature of the Potts
3393: model with $q$ states per site for the triangular and honeycomb lattices,
3394: J. Phys. C {\bf 7}, L167-L169 (1974).}
3395:
3396: \end{thebibliography}
3397: \vfill
3398: \eject
3399: \end{document}
3400: