1: \documentstyle [12pt,eqsecnum,amsfonts,aps]{revtex}

2: \input epsf

3: \tighten

4: \draft

5: \widetext

6: \input epsf

7: \topmargin -0.3in

8: 

9: \textwidth  6.5in

10: \textheight 8.5in

11: \oddsidemargin 0in

12: 

13: \newcommand{\beq}{\begin{equation}}

14: \newcommand{\eeq}{\end{equation}}

15: \newcommand{\beqs}{\begin{eqnarray}}

16: \newcommand{\eeqs}{\end{eqnarray}}

17: \newcommand{\lsim}{\mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle<}{\sim}$}}}

18: \newcommand{\gsim}{\mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle>}{\sim}$}}}

19: \newtheorem{th}{Theorem}[section]

20: \newtheorem{cor}{Corollary}[section]

21: \newtheorem{lemma}{Lemma}[section]

22: \newtheorem{defi}{Definition}[section]

23: \newtheorem{conj}{Conjecture}[section]

24: 

25: \begin{document}

26: \draft

27: \baselineskip 6.0mm

28: 

29: \bigskip

30: \bigskip

31: 

32: \title{Transfer Matrices for the Zero-Temperature Potts Antiferromagnet on 

33: Cyclic and M\"obius Lattice Strips}

34: 

35: \bigskip

36: 

37: \author{

38: Shu-Chiuan Chang$^{a,b}$ \thanks{email: scchang@phys.ntu.edu.tw} \and

39: Robert Shrock$^{c}$ \thanks{email: robert.shrock@sunysb.edu}}

40: 

41: \bigskip

42: 

43: \address{(a) \ Department of Applied Physics, Faculty of Science \\

44: Tokyo University of Science \\

45: Tokyo 162-8601, Japan}

46: 

47: \address{(b) \ Physics Division \\

48: National Center for Theoretical Sciences at Taipei \\

49: National Taiwan University  \\

50: Taipei 10617, Taiwan }

51: 

52: \address{(c) \ C. N. Yang Institute for Theoretical Physics \\

53: State University of New York \\

54: Stony Brook, N. Y. 11794}

55: 

56: \maketitle

57: 

58: \bigskip

59: 

60: \begin{abstract}

61: 

62: We present transfer matrices for the zero-temperature partition function of the

63: $q$-state Potts antiferromagnet (equivalently, the chromatic polynomial) on

64: cyclic and M\"obius strips of the square, triangular, and honeycomb lattices of

65: width $L_y$ and arbitrarily great length $L_x$.  We relate these results to our

66: earlier exact solutions for square-lattice strips with $L_y=3,4,5$,

67: triangular-lattice strips with $L_y=2,3,4$, and honeycomb-lattice strips with

68: $L_y=2,3$ and periodic or twisted periodic boundary conditions.  We give a

69: general expression for the chromatic polynomial of a M\"obius strip of a

70: lattice $\Lambda$ and exact results for a subset of honeycomb-lattice transfer

71: matrices, both of which are valid for arbitrary strip width $L_y$.  New results

72: are presented for the $L_y=5$ strip of the triangular lattice and the $L_y=4$

73: and $L_y=5$ strips of the honeycomb lattice.  Using these results and taking

74: the infinite-length limit $L_x \to \infty$, we determine the continuous

75: accumulation locus of the zeros of the above partition function in the complex

76: $q$ plane, including the maximal real point of nonanalyticity of the degeneracy

77: per site, $W$ as a function of $q$.

78: 

79: \end{abstract}

80: 

81: \bigskip

82: \bigskip

83: 

84: 

85: \newpage

86: \pagestyle{plain}

87: \pagenumbering{arabic}

88: 

89: \section{Introduction}

90: 

91: The $q$-state Potts antiferromagnet (AF) \cite{potts,wurev} exhibits nonzero

92: ground state entropy, $S_0 > 0$ (without frustration) for sufficiently large

93: $q$ on a given lattice $\Lambda$ or, more generally, on a graph $G$.  This is

94: equivalent to a ground state degeneracy per site $W > 1$, since $S_0 = k_B \ln

95: W$.  There is a close connection with graph theory

96: here, since the zero-temperature partition function of the above-mentioned

97: $q$-state Potts antiferromagnet on a graph $G=G(V,E)$ defined by vertex and

98: edge sets $V$ and $E$ satisfies

99: \beq

100: Z(G,q,T=0)_{PAF}=P(G,q)

101: \label{zp}

102: \eeq

103: where $P(G,q)$ is the chromatic polynomial expressing the number of ways of

104: coloring the vertices of the graph $G$ with $q$ colors such that no two

105: adjacent vertices have the same color \cite{wup}-\cite{bbook}.  Thus

106: %

107: \beq

108: W(\{G\},q) = \lim_{n \to \infty} P(G,q)^{1/n}

109: \label{w}

110: \eeq

111: %

112: where $n=|V|$ denotes the number of vertices of the graph $G$ and the symbol

113: $\{G\}$ formally denotes the set $\lim_{n \to \infty} G$. The minimum

114: number of colors that is necessary to color a graph $G$ subject to this

115: constraint is called the chromatic number of $G$, $\chi(G)$.

116: 

117: We represent a given strip as extending longitudinally in the $x$ direction and

118: transversely in the $y$ direction, with width $L_y$ vertices.  Each strip

119: involves a longitudinal repetition of $m$ copies of a particular subgraph.  For

120: the square-lattice strips, this is a column of squares.  It is convenient to

121: represent the strip of the triangular lattice as obtained from the

122: corresponding strip of the square lattice with additional diagonal edges

123: connecting, say, the upper-left to lower-right vertices in each square.  In

124: both these cases, the length is $L_x=m$ vertices.  We represent the strip of

125: the honeycomb lattice in the form of bricks oriented horizontally.  In this

126: case, since there are two vertices in 1-1 correspondence with each horizontal

127: side of a brick, $L_x=2m$ vertices.

128: 

129: The general structure of the Potts model partition function for cyclic strips

130: of the square lattice of width $L_y$, as a sum of powers of eigenvalues of a

131: formal transfer matrix multiplied by certain coefficients $c^{(d)}$, $0 \le d

132: \le L_y$, was given in Refs. \cite{saleur} (see also \cite{saleurcmp}).  The 

133: present authors (unaware of this finding in \cite{saleur}) rediscovered the 

134: result in Ref. \cite{cf} and showed that it applies also to cyclic strips of

135: the triangular and honeycomb lattices \cite{hca}.  The coefficients are 

136: polynomials of degree $d$ in the variable $q$ given by \cite{saleur,cf}

137: %

138: \beq

139: c^{(d)} = U_{2d}(q^{1/2}/2) = \sum_{j=0}^d (-1)^j {2d-j \choose j}

140: q^{d-j}

141: \label{cd}

142: \eeq

143: %

144: with $U_n(x)$ being the Chebyshev polynomial of the second kind.  The first few

145: of these coefficients are $c^{(0)}=1$, $c^{(1)}=q-1$, $c^{(2)}=q^2-3q+1$, and

146: $c^{(3)}=q^3-5q^2+6q-1$.  The $c^{(d)}$'s play a role analogous to

147: multiplicities of eigenvalues $\lambda_{\Lambda,L_y,d,j}$, although this

148: identification is formal, since $c^{(d)}$ may be zero or negative for the

149: physical values $q=1,2,3$.  For example, as shown in eqs. (2.19), (2.20) of 

150: Ref. \cite{cf}, if $q=2$, then $c^{(d)}=-1$ for $d=2$ or 3 mod 4, and if $q=3$,

151: then $c^{(d)}=-1$ for $d=3$ or 5 mod 6 and $c^{(d)}=-2$ for $d=4$ mod 6.  In

152: general, $c^{(d)}$ vanishes at $q=q_{d,k}$, where $q_{d,k}=4\cos^2(\pi

153: k/(2d+1))$, $k=1,2,...,d$, so that $c^{(d)}$ is positive for $q \ge 4$ for

154: arbitrary $d$.  

155: 

156: This structure also holds for the chromatic polynomial, with the difference

157: that, for $L_y \ge 2$, the number of eigenvalues for each $d$ is smaller than

158: the number for the full partition function.  We determined this number for

159: strips of the square and triangular lattice in Ref. \cite{cf}, for strips of

160: the honeycomb lattice in Ref. \cite{hca}, and for self-dual strips of the

161: square lattice in Ref. \cite{dg} (where we also noted how these numbers fill

162: out the entries in the relevant Bratelli diagrams). The chromatic polynomial

163: for a cyclic strip of the regular lattice $\Lambda$ has the form

164: %

165: \beq

166: P(\Lambda,L_y \times m,cyc.,q)=\sum_{d=0}^{L_y} c^{(d)}

167: \sum_{j=1}^{n_P(\Lambda,L_y,d)} (\lambda_{\Lambda,L_y,d,j})^m

168: \label{pgsum}

169: \eeq

170: %

171: 

172: 

173: Let $G^\prime=(V,E^\prime)$ be a spanning subgraph of $G$, i.e. a subgraph

174: having the same vertex set $V$ and an edge set $E^\prime \subseteq E$. Then

175: $P(G,q)$ can be written as the sum \cite{kf,fk}

176: %

177: \beq

178: P(G,q) = \sum_{G^\prime \subseteq G} q^{k(G^\prime)}(-1)^{|E^\prime|}

179: \label{cluster}

180: \eeq

181: %

182: where $k(G^\prime)$ denotes the number of connected components of $G^\prime$

183: and $|E^\prime|$ denotes the number of edges in the set $E^\prime$. 

184: Since we only consider connected graphs $G$, we have $k(G)=1$. The formula

185: (\ref{cluster}) enables one to generalize $q$ from ${\mathbb Z}_+$ to ${\mathbb

186: R}_+$.  The zeros of $P(G,q)$ in the complex $q$ plane are called chromatic

187: zeros. We denote the continuous accumulation set of these zeros in the $n \to

188: \infty$ limit as ${\cal B}$, which is the continuous locus of points

189: where $W(\{G\},q)$ is nonanalytic. The maximal value of $q$ where ${\cal B}$

190: intersects the (positive) real axis is labelled $q_c(\{G\})$.  This locus

191: occurs as a solution to the degeneracy in magnitude of $\lambda$'s of maximal

192: magnitude \cite{bkw1}-\cite{read91}. 

193: 

194: 

195: In this paper we present transfer matrices $T_{\Lambda,L_y}$ for strips of the

196: square (sq), triangular (tri), and honeycomb (hc) lattices of fixed transverse

197: widths $L_y$ and arbitrarily great length with periodic (cyclic) longitudinal

198: boundary conditions.  We also give results for the corresponding strips with

199: twisted periodic (M\"obius) longitudinal boundary conditions.  We relate these

200: results to our earlier exact solutions for square-lattice strips with

201: $L_y=3,4,5$; triangular-lattice strips with $L_y=2,3,4$, and honeycomb-lattice

202: strips with $L_y=2,3$ and periodic or twisted periodic boundary conditions.  We

203: give a general expression for the chromatic polynomial of a M\"obius strip of a

204: lattice $\Lambda$ which is valid for arbitrary strip width $L_y$.  New results

205: are presented for the $L_y=5$ strip of the triangular lattice and the $L_y=4$

206: and $L_y=5$ strips of the honeycomb lattice.  Using these results and taking

207: the infinite-length limit $L_x \to \infty$, we determine certain properties of

208: the continuous accumulation locus of the zeros of the above partition function

209: in the complex $q$ plane, including the maximal point of nonanalyticity of the

210: degeneracy per site, $W$ as a function of $q$.  We also give a general trace

211: formula for the honeycomb strip which is valid for arbitrary $L_y$.  Chromatic

212: numbers for these lattice strips are given in the appendix. 

213: 

214: 

215: There are several motivations for this work. One is that, as noted above, the

216: Potts antiferromagnet has the interesting property of nonzero ground state

217: entropy for sufficiently large $q$, which is an exception to the third law of

218: thermodynamics \cite{al,cw}.  Via eq. (\ref{w}), exact calculations of $P(G,q)$

219: yield calculations of $W\{ G \},q)$ and thus give insight into this property.

220: The constraint that no two adjacent vertices have the same value of $q$ becomes

221: less and less important as $q \to \infty$, and in this limit, $W(\{ G \},q) \to

222: q$.  As $q$ decreases, $W(\{ G \},q)$ remains a real analytic function of $q$

223: down to the point $q_c(\{G \})$, where it is nonanalytic.  In earlier work, we

224: studied how $q_c(\{G \})$ depends on the boundary conditions used for various

225: lattice strips \cite{w}-\cite{s5}.  It was found that a convenient property of

226: strips with periodic (or twisted periodic) longitudinal boundary conditions was

227: that they always defined a value of $q_c(\{ G \})$, and, furthermore, for a

228: given type of lattice, this point was observed to be a monotonically

229: nondecreasing function of the strip width for all cases calculated.  By

230: carrying out exact calculations of $P(G,q)$ and determining the analytic

231: structure of $W(\{ G \},q)$, one can study how $q_c(\{ G \})$ varies with

232: increasing strip width and how it approaches the values expected for the

233: corresponding two-dimensional lattices, namely $q_c(sq)=3$ \cite{lenard} and

234: $q_c(tri)=4$ \cite{baxter86,baxter87}.  This interpolation property is an

235: especially interesting use of these exact results since the resultant values of

236: $q_c(\{ G \})$ interpolate between the value $q_c=2$ for one-dimension and

237: values on two-dimensional lattices.  Similar results were found for the ground

238: state entropy itself \cite{w,ww,w3,wn,w2d}.  This should be contrasted with

239: other properties; for example, the critical temperature of the Potts

240: ferromagnet on an infinite-length lattice strip is zero for any width $L_y$

241: regardless of how great.  

242: 

243: For the honeycomb lattice, formal arguments give $q_c(hc)=(3+\sqrt{5})/2 \simeq

244: 2.618$ \cite{ssbounds,p3afhc}.  The honeycomb value is only formal because if

245: $q$ is not a positive integer, then one cannot use the expression of the

246: partition function for the Potts antiferromagnet in terms of a sum of

247: positive-definite Boltzmann factors and instead must use the formula \cite{kf}

248: $Z(G,q,v)=\sum_{G^\prime \subseteq G} q^{k(G^\prime)} v^{|E^\prime|}$, where

249: $v=e^K-1$ with $K=\beta J$, where $\beta=(k_BT)^{-1}$ and $J$ is the spin-spin

250: coupling (which reduces to eq. (\ref{cluster}) for the $T=0$ antiferromagnet,

251: where $K=-\infty$).  However, for $-1 \le v < 0$ and non-integral $q$, this

252: does not, in general, define a Gibbs measure \cite{ssbounds,a}. This

253: non-integral property of $q_c(hc)$ for the two-dimensional honeycomb lattice

254: and the associated subtleties make it especially useful to have explicit

255: calculations of $q_c$ values for finite-width strips of this lattice, to check

256: that these are consistent with the expected behavior, according to which they

257: approach the above value of $q_c(hc)$ as $L_y$ gets large.  

258: 

259: A second motivation related to this is that when one generalizes $q$ from a

260: positive integer to a complex variable, one sees that the point $q_c(\{ G \})$

261: is the maximal point where a certain nonanalytic boundary ${\cal B}$ crosses

262: the real axis (discussed further below); thus, another purpose of these exact

263: calculations is to get further insight about this locus ${\cal B}$.  As is well

264: known, slight changes in coefficients of polynomials can have drastic changes

265: on the positions of the zeros of the polynomials, which means that one must

266: have the exact expression for $P(G,q)$ to study these zeros and their

267: accumulation set ${\cal B}$.  Since it was found that for (the $L_x \to \infty$

268: limit of) strips with free longitudinal boundary conditions, ${\cal B}$ does

269: not necessarily cross the real axis, so that no $q_c(\{ G \}, q)$ is defined

270: \cite{w,strip}, one cannot carry out this study in the same manner for these

271: strips (although for sufficiently wide strips with free longitudinal boundary

272: conditions, arcs on ${\cal B}$ have endpoints that are often close enough to

273: the real axis to allow one to extrapolate and define an effective $q_c$).

274: 

275: A third motivation is that certain structural features of the transfer matrices

276: can allow one to obtain general formulas that are applicable for arbitrarily

277: large strip widths.  We have already used this feature in Ref. \cite{s5}, where

278: we presented transfer matrices whose eigenvalues are $\lambda_{\Lambda,L_y,d}$

279: with degree $d=L_y-1$ for several lattice strips with periodic boundary

280: conditions, including those of the square and triangular lattices. Thus another

281: purpose of exhibiting explicit transfer matrices for $0 \le d \le L_y-2$ here

282: is to make them available for a wider community, leading, hopefully, to some

283: advances beyond those we have made with $d=L_y,L_y-1$ in the construction of

284: explicit formulas for $\lambda_{\Lambda,L_y,d}$ valid for arbitrary $L_y$.  All

285: of these degree-$d$ sectors contribute to the partition function (here, the

286: chromatic polynomial), so that they are of physical interest, although an

287: elementary result is that the degeneracy per site $W$ only depends on the

288: (maximal eigenvalue in the) $d=0$ sector.

289: 

290: The chromatic polynomial for the cyclic and M\"obius strips of the square

291: lattice were calculated (i) for $L_y=2$ in \cite{bds} (see also \cite{bm});

292: (ii) for $L_y=3$ cyclic in \cite{wcyl,wcy}, and M\"obius case in \cite{pm};

293: (iii) for $L_y=4$ cyclic/M\"obius in \cite{s4}, (iv) for $L_y=5$

294: cyclic/M\"obius in \cite{s5}.  The chromatic polynomials for the triangular

295: lattice strips were calculated for (i) $L_y=2$, cyclic/M\"obius in \cite{wcy},

296: (ii) $L_y=3,4$ in \cite{t}; and for the honeycomb lattice with (i) $L_y=2$,

297: cyclic/M\"obius in \cite{pg}, (ii) $L_y=3$ in \cite{hca}. (Refs. \cite{pg,nec}

298: showed that the structure (\ref{pgsum}) holds more generally for cyclic strip

299: graphs composed of iterated subgraphs that are not parts of regular lattices;

300: we shall not need this degree of generality here.)  We have proved that for

301: these strips the set of eigenvalues are the same for cyclic (cyc.) and M\"obius

302: (Mb.) boundary conditions and have given the transformation rules for how the

303: coefficients change for strips of the square lattice when one changes from

304: cyclic to M\"obius boundary conditions \cite{cf}.

305: 

306: In our previous works on chromatic polynomials we have sometimes given transfer

307: matrices (e.g. \cite{s5}) but often have expressed our results for the

308: $\lambda_{\Lambda,L_y,d,j}$ in terms of generating functions or solutions to

309: algebraic equations.  In one respect this is a maximally compact way of

310: presenting the results, since for a given set of $n_P(\Lambda,L_y,d)$

311: $\lambda_{\Lambda,L_y,d,j}$'s, it suffices to give the coefficients of the

312: algebraic equation of degree $n_P(\Lambda,L_y,d)$ and for the generating

313: function, it suffices to give $2n_P(\Lambda,L_y,d)$ coefficients, which are

314: polynomials in $q$.  Since the generating function is a rational function in

315: $q$, it has the appeal that one always deals with polynomials.  For the

316: expression of the chromatic polynomial in terms of transfer matrices, since

317: $T_{\Lambda,L_y,d}$ has dimension $n_P(\Lambda,L_y,d) \times

318: n_P(\Lambda,L_y,d)$, one must specify a greater number of polynomials, namely

319: $n_P(\Lambda,L_y,d)^2$ polynomials.  As $L_y$ increases, the number of

320: expressions that one must give thus grows quadratically rather than linearly,

321: as in the more compact approach using generating functions or algebraic

322: equations for the $\lambda_{\Lambda,L_y,d}$.  However, offsetting this

323: disadvantage, the transfer matrix method has the advantage that some of the

324: entries are rather simple, and, moreover, one can sometimes spot useful

325: patterns in the matrices that expedite or confirm the construction of general

326: formulas that are applicable for arbitrarily large strip widths $L_y$.

327: 

328: We mention some previous work.  Matrix methods and related recursive linear

329: algebraic techniques for calculating chromatic polynomials were used in early

330: work \cite{bds,bm,b} and more recent papers \cite{matmeth}-\cite{matmeth3}. The

331: coloring matrix technique of Ref. \cite{b} was applied in

332: Refs. \cite{ww}-\cite{wn} to obtain upper and lower bounds on $W$ for various

333: (infinite) two-dimensional lattices.  This application relied upon the fact

334: that these coloring matrices are non-negative, so that one could use the

335: Perron-Frobenius theorem in analyzing the eigenvalues.  Related recursive

336: linear algebraic methods were used in Refs. \cite{strip,strip2,hs} and in our

337: subsequent papers on chromatic polynomials.  Transfer matrices for chromatic

338: polynomials were used in Refs. \cite{baxter86,baxter87,klein}.  In Ref.

339: \cite{sqtran} transfer matrices for the Potts model were developed and were

340: applied there and in Refs. \cite{cyltran,tritran} to calculate chromatic

341: polynomials for strips of the square and triangular lattices with free

342: longitudinal boundary conditions and free or periodic transverse boundary

343: conditions.  These have been termed transfer matrices in the Fortuin-Kasteleyn

344: representation (see eq. (\ref{cluster})).  In Refs. \cite{ts,tt}, transfer

345: matrix methods were used to calculate full Potts model partition functions

346: with arbitrary $q$ and $v$ for strips of the square and triangular lattices

347: with free longitudinal boundary conditions.  Using transfer matrix methods

348: together with the sieve methods of Refs. \cite{matmeth}-\cite{matmeth3} (see

349: also \cite{dn,cprg,ka3}), we have proved theorems that determined

350: $T_{\Lambda,L_y,d}$ for $d=L_y-1$ and arbitrarily large $L_y$ for several

351: lattice strips with periodic boundary conditions \cite{s5}.

352: 

353: Some remarks are in order here concerning an important respect in which the

354: transfer matrices presented here differ from the conventional transfer matrices

355: in statistical mechanics and coloring matrices in mathematical graph theory.

356: Consider a strip graph, and let the spins on each transverse slice of this

357: strip be labeled $\sigma_{x,y}$ where the longitudinal position $x$ is fixed

358: and the transverse position $y$ varies.  As conventionally defined, the

359: transfer matrix of the (zero-field) Potts model at some temperature $T$ is

360: defined as the matrix $T_{x,x+1} = \langle \{ \sigma_{x,y} \} | e^{-\beta {\cal

361: H}} | \{ \sigma_{x+1,y} \} \rangle$, where ${\cal H} = -J \sum_{\langle i j

362: \rangle} \delta_{\sigma_i \sigma_j}$, $\sigma_i=1,...,q$ are the spin variables

363: on each vertex and $\langle i j \rangle$ denotes pairs of adjacent vertices.

364: Since there are $q^{L_y}$ spin configurations on each slice, this is a $q^{L_y}

365: \times q^{L_y}$ dimensional matrix. Although this dimension depends on $q$, the

366: individual elements themselves do not explicitly depend on $q$.  For physical

367: temperatures $0 \le T \le \infty$ this matrix is non-negative, and at nonzero

368: temperatures it is positive-definite.  For $T > 0$, on a strip of finite width,

369: for which this is a finite-dimensional matrix, one can then make powerful use

370: of the Perron-Frobenius theorem on positive-definite matrices to conclude that

371: there is a unique real positive eigenvalue of maximal magnitude.  Indeed, this

372: theorem provides a standard way to prove that the free energy of the Potts

373: model or other spin model with short-range interactions on a strip of infinite

374: length and finite width is analytic at all finite temperatures.  Similarly, the

375: coloring matrices used in Refs. \cite{b,ww,w3,wn} for studies of chromatic

376: polynomials and the degeneracy per site $W$ on various lattices are

377: non-negative matrices, and this property was necessary for the upper and lower

378: bounds derived in these papers.  In contrast, the transfer matrices in the

379: Fortuin-Kasteleyn representation are not positive-definite, and hence the

380: Perron-Frobenius does not apply to them.  The fact that it does not apply is

381: crucial for the property that the locus ${\cal B}$ crosses the positive real

382: axis and for the existence of $q_c(\{ G \})$, since this locus occurs where

383: there is a switch between at least two distinct eigenvalues of maximal

384: magnitude and hence non-uniqueness of such eigenvalues.

385: 

386:  From the exact expressions for $P(G,q)$, one can evaluate eq. (\ref{w}) to get

387: $W$ and analyze the accumulation locus ${\cal B}$ of zeros of $P(G,q)$ in the

388: complex $q$ plane.  We have done this for cyclic/M\"obius strips in a number of

389: previous works.  These loci are different for strips with different boundary

390: conditions, e.g., free, cylindrical, cyclic/M\"obius, and toroidal.  As we have

391: shown, since the $\lambda$'s are the same for the cyclic and M\"obius strips of

392: a given lattice, it follows that the locus ${\cal B}$ is the same for the

393: infinite-length limits of cyclic and M\"obius strips of a given lattice.  In

394: our previous work we have found a number of general properties of the locus

395: ${\cal B}$, including the property that for strips of regular lattices with

396: periodic or twisted periodic longitudinal boundary conditions the locus ${\cal

397: B}$ is comprised of closed curves that enclose various regions and pass through

398: $q=0$ and at a maximal real point $q_c$ which depends on the lattice, as well

399: as possible other intermediate points.  The point $q_c(\{G\})$ is important

400: since it separates the interval $q > q_c(\{G\})$ on the positive real $q$ axis

401: where the Potts model (with $q$ extended from ${\mathbb Z}_+$ to ${\mathbb R}$)

402: exhibits nonzero ground state entropy (which increases with $q$, asymptotically

403: approaching $S_0 = k_B \ln q$ for large $q$) from the interval $0 \le q \le

404: q_c(\{G\})$ in which $S_0$ has a different analytic form.  

405: 

406: 

407: \section{General Structure}

408: 

409: Our results for chromatic polynomials are obtained as special cases, for the

410: zero-temperature antiferromagnet, of general transfer matrices that we have

411: calculated for the full temperature-dependent Potts model partition function on

412: cyclic and M\"obius lattice strips \cite{js}.  The latter results, and the

413: general method, will be reported in a companion paper \cite{zt}. The

414: motivations for presenting the special cases of these results for the

415: zero-temperature Potts antiferromagnet (chromatic polynomial) have been given

416: in the introduction.  In addition to these physics motivations, a relevant

417: point is that the transfer matrices for the chromatic polynomials are

418: subtantially smaller in dimension and simpler in structure, depending only one

419: one variable instead of two, than the transfer matrices for the full Potts

420: model partition function.  Another reason for separating these analyses of the

421: full Potts model and the chromatic polynomial has to do with the determinants

422: $det(T_{Z,\Lambda,L_y,d})$ and $det(T_{P,\Lambda,L_y,d})$; in both the case of

423: the full Potts model and the chromatic polynomial, the individual eigenvalues

424: $\lambda_{X,\Lambda.L_y,d}$ for most strip widths are not expressible in

425: explicit algebraic form because the corresponding characteristic polynomials

426: are of fifth order or higher.  However, we have found a most interesting

427: property, that the products of the eigenvalues that define the determinant

428: $det(T_{Z,\Lambda,L_y,d})$ have a very simple form.  When one specializes to

429: $v=-1$, for all values of $L_y$ and $d=0,1,...L_y-1$, except the lowest case,

430: $L_y=1$, some of these eigenvalues vanish, so that $det(T_{Z,\Lambda,L_y,d})=0$

431: at $v=-1$, reflecting the fact that the transfer matrices $T_{P,\Lambda,L_y,d}$

432: are of smaller dimension than $T_{Z,\Lambda,L_y,d}$.  But in this $v=-1$ case,

433: when one removes the zero columns and corresponding rows of

434: $T_{Z,\Lambda,L_y,d}$ to form the transfer matrix $T_{P,\Lambda,L_y,d}$ for the

435: chromatic polynomial, the resultant determinant is, in general, nonzero.  And

436: again we find a very interesting feature, namely that although the eigenvalues

437: $\lambda_{P,\Lambda,L_y,d}$ themselves are roots of characteristic polynomials

438: that are often of fifth or higher order, precluding solutions in terms of

439: explicit algebraic expressions, the products that define

440: $det(T_{P,\Lambda,L_y,d})$ are often quite simple, especially for the strips of

441: the triangular and honeycomb lattices.  This is another motivation for our

442: presenting results for the chromatic polynomials separately from those for the

443: full Potts model partition function.

444: 

445: 

446: The chromatic polynomial for cyclic strips can be written in the form 

447: \cite{saleur,cf}

448: %

449: \beq 

450: P(\Lambda,L_y \times m,cyc.,q) = \sum_{d=0}^{L_y} c^{(d)}

451: Tr((T_{\Lambda,L_y,d})^m) 

452: \label{pgsumt} 

453: \eeq

454: %

455: where $T_{\Lambda,L_y,d}$ is the transfer matrix, with eigenvalues

456: $\lambda_{\Lambda,L_y,d}$. We denote the characteristic polynomial of

457: $T_{\Lambda,L_y,d}$ in the variable $z$ as $CP(T_{\Lambda,L_y,d},z)$.  We have

458: shown that the full transfer matrix $T_{\Lambda,L_y}$ has a block structure

459: formally specified by

460: %

461: \beq 

462: T_{\Lambda,L_y} = \bigoplus_{d=0}^{L_y} \prod

463: T_{\Lambda,L_y,d} 

464: \label{Tdirectsum} 

465: \eeq

466: %

467: where the product $\prod T_{\Lambda,L_y,d}$ means a set of square blocks, each

468: of dimension, $c^{(d)}$, of the form $\lambda_{\Lambda,L_y,d,j}$ times the

469: identity matrix. 

470: The dimension of the total transfer matrix, i.e., the total number

471: of eigenvalues $\lambda_{\Lambda,L_y,d,j}$ in eq. (\ref{pgsum}),

472: counting multiplicities, is thus

473: %

474: \beq 

475: dim(T_{\Lambda,L_y}) = \sum_{d=0}^{L_y}

476: dim(T_{\Lambda,L_y,d}) = \sum_{d=0}^{L_y} c^{(d)}

477: n_P(\Lambda,L_y,d) \ . 

478: \label{dimTL} 

479: \eeq

480: %

481: 

482: Each matrix $T_{\Lambda,L_y,d}$ is defined relative to a basis of $d$-color

483: assignments to a path graph with $L_y$ vertices corresponding to a transverse

484: slice across the strip.  Here one introduces partitions of the vertices in

485: these path graphs.  Using these as bases, we calculate the transfer matrix 

486: for each level (= degree) $d$, as discussed in our Ref. \cite{zt}. 

487: For the chromatic polynomials of the square and triangular

488: lattices, we have the additional requirement that adjacent vertices cannot

489: connect to each other. We list graphically all the possible partitions for

490: $L_y=2, 3, 4$ strips in Figs. \ref{L2partitions} to \ref{L4partitions}, where

491: white circles are the original $L_y$ vertices and each black circle corresponds

492: to a specific color assignment. We denote these partitions ${\cal P}_{L_y,d}$

493: for $2 \le L_y \le 5$ as follows:

494: %

495: \beq {\cal P}_{2,0} = \{ I \} \ , \qquad {\cal P}_{2,1} = \{ \bar

496: 2; \bar 1 \} \ , \qquad {\cal P}_{2,2} = \{ \bar 1, \bar 2 \}

497: \label{L2partitionlist} \eeq

498: %

499: \beqs {\cal P}_{3,0} & = & \{ I; 13 \} \ , \qquad {\cal P}_{3,1} =

500: \{ \bar 3; \bar 2; \bar 1; \overline{13} \} \cr\cr {\cal P}_{3,2}

501: & = & \{ \bar 2, \bar 3; \bar 1, \bar 3; \bar 1, \bar 2 \} \ ,

502: \qquad {\cal P}_{3,3} = \{ \bar 1, \bar 2, \bar 3 \}

503: \label{L3partitionlist} \eeqs

504: %

505: \beqs {\cal P}_{4,0} & = & \{ I; 13; 14; 24 \} \ , \qquad {\cal

506: P}_{4,1} = \{ \bar 4; \bar 3; \bar 2; \bar 1; 13, \bar 4;

507: \overline{13}; \overline{14}; \overline{24}; 24, \bar 1 \} \cr\cr

508: {\cal P}_{4,2} & = & \{ \bar 3, \bar 4; \bar 2, \bar 4; \bar 1,

509: \bar 4; \bar 2, \bar 3; \bar 1, \bar 3; \bar 1, \bar 2;

510: \overline{13}, \bar 4; \bar 1, \overline{24} \} \cr\cr {\cal

511: P}_{4,3} & = & \{ \bar 2, \bar 3, \bar 4; \bar 1, \bar 3, \bar 4;

512: \bar 1, \bar 2, \bar 4; \bar 1, \bar 2, \bar 3 \} \ , \qquad {\cal

513: P}_{4,4} = \{ \bar 1, \bar 2, \bar 3, \bar 4 \}

514: \label{L4partitionlist} \eeqs

515: %

516: \beqs {\cal P}_{5,0} & = & \{ I; 13; 14; 15; 24; 25; 35; 24,15;

517: 135 \} \cr\cr {\cal P}_{5,1} & = & \{ \bar 5; \bar 4; \bar 3; \bar

518: 2; \bar 1; 13, \bar 5; 13, \bar 4; \overline{13}; 14, \bar 5;

519: \overline{14}; \overline{15}; 24, \bar 5; \overline{24}; 24, \bar

520: 1; \overline{25}; 25, \bar 1; \overline{35}; 35, \bar 2; 35, \bar

521: 1; \cr\cr & & 24, \overline{15}; \overline{135} \} \cr\cr {\cal

522: P}_{5,2} & = & \{ \bar 4, \bar 5; \bar 3, \bar 5; \bar 2, \bar 5;

523: \bar 1, \bar 5; \bar 3, \bar 4; \bar 2, \bar 4; \bar 1, \bar 4;

524: \bar 2, \bar 3; \bar 1, \bar 3; \bar 1, \bar 2; 13, \bar 4, \bar

525: 5; \overline{13}, \bar 5; \overline{13}, \bar 4; \overline{14},

526: \bar 5; \overline{24}, \bar 5; 24, \bar 1, \bar 5; \cr\cr & & \bar

527: 1, \overline{24}; \bar 1, \overline{25}; \bar 2, \overline{35};

528: \bar 1, \overline{35}; 35, \bar 1, \bar 2 \} \cr\cr {\cal P}_{5,3}

529: & = & \{ \bar 3, \bar 4, \bar 5; \bar 2, \bar 4, \bar 5; \bar 1,

530: \bar 4, \bar 5; \bar 2, \bar 3, \bar 5; \bar 1, \bar 3, \bar 5;

531: \bar 1, \bar 2, \bar 5; \bar 2, \bar 3, \bar 4; \bar 1, \bar 3,

532: \bar 4; \bar 1, \bar 2, \bar 4; \bar 1, \bar 2, \bar 3;

533: \overline{13}, \bar 4, \bar 5; \bar 1, \overline{24}, \bar 5;

534: \cr\cr & & \bar 1, \bar 2, \overline{35} \} \cr\cr {\cal P}_{5,4}

535: & = & \{ \bar 2, \bar 3, \bar 4, \bar 5; \bar 1, \bar 3, \bar 4,

536: \bar 5; \bar 1, \bar 2, \bar 4, \bar 5; \bar 1, \bar 2, \bar 3,

537: \bar 5; \bar 1, \bar 2, \bar 3, \bar 4 \} \cr\cr {\cal P}_{5,5} &

538: = & \{ \bar 1, \bar 2, \bar 3, \bar 4, \bar 5 \}

539: \label{L5partitionlist}  \eeqs

540: %

541: where partitions are separated by a colon, and each overline corresponds to a

542: color assignment.  That is, for $d=0$ there is no explicit color assignments;

543: for $d=1$, one vertex, or one connected set of vertices has its color

544: specified; for $d=2$, two vertices or two separate connected sets of vertices

545: have their colors specified; and so forth for higher values of $d$.  Therefore,

546: in eqs. (2.1) to (2.4), there is no overline for $d=0$, one overline for each

547: partition of $d=1$, two overlines for each partition of $d=2$, etc.

548: 

549: The number of partitions $n_P(\Lambda,L_y,d)$ is the dimension of the transfer

550: matrix $T_{\Lambda,L,d}$,

551: %

552: \beq 

553: dim(T_{\Lambda,L_y,d}) = n_P(\Lambda,L_y,d) \ . 

554: \label{dimTLd}

555: \eeq

556: %

557: In Ref. \cite{cf} we determined the number $n_P(\Lambda,L_y,d)$ (labelled

558: simply as $n_P(L_y,d)$) for the lattices $\Lambda = sq, tri$). 

559: Some special cases for $\Lambda=sq,tri$ are \cite{cf} as follows;

560: analogous results for the honeycomb lattice were given in Ref. \cite{hca}.

561: %

562: \beq

563: n_P(\Lambda,L_y,L_y)=1

564: \label{nplyly}

565: \eeq

566: %

567: \beq

568: n_P(\Lambda,L_y,L_y-1)=L_y

569: \label{nplylym1}

570: \eeq

571: %

572: \beq

573: n_P(\Lambda,L_y,0)=n_P(\Lambda,L_y-1,1)=M_{L_y-1} 

574: \label{nply0}

575: \eeq

576: %

577: where $M_n$ is the Motzkin number in combinatorics, given by 

578: %

579: \beq

580: M_n = \sum_{j=0}^n (-1)^j C_{n+1-j} {n \choose j}

581: \label{motzkin}

582: \eeq

583: %

584: where $C_n=(n+1)^{-1}{2n \choose n}$ is the Catalan number. (We use the same

585: symbol $C_n$ for the circuit graph; the meaning will be clear from context).

586: For reference, the first few Motzkin numbers are $M_n=1,2,4,9,21,51$ 

587: for $n=1,..., 6$, and we use the formal definition $M_0=1$. 

588: 

589: 

590: Note that the number $n_P(tri,L_y,0)$ is the same as the dimension of the

591: corresponding transfer matrix for the triangular-lattice strip with free

592: boundary conditions obtained in Ref. \cite{sqtran}, while the transfer matrix

593: for the free strip of the square lattice has a dimension (for which a general

594: formula was given in Ref. \cite{ts}) which is smaller than $n_P(sq,L_y,0)$ for

595: $L_y \ge 4$.  Consistent with this, for $L_y \ge 4$, we find that the

596: characteristic polynomials for $T_{sq,4,0}$ and $T_{sq,5,0}$ factorize into

597: parts such that one factor is the characteristic polynomial for the transfer

598: matrix of the corresponding free strip, as will be seen below.

599: 

600: \begin{figure}

601: \unitlength 1mm \hspace*{5mm}

602: \begin{picture}(90,12)

603: \put(10,12){\makebox(0,0){{\small $d=0$}}}

604: \multiput(20,8)(0,4){2}{\circle{2}}

605: \put(40,12){\makebox(0,0){{\small $d=1$}}}

606: \multiput(50,4)(10,0){2}{\circle*{2}}

607: \multiput(50,8)(10,0){2}{\circle{2}}

608: \multiput(50,12)(10,0){2}{\circle{2}} \put(50,4){\line(0,1){4}}

609: \put(60,8){\oval(4,8)[l]} \put(80,12){\makebox(0,0){{\small

610: $d=2$}}} \multiput(90,0)(0,4){2}{\circle*{2}}

611: \multiput(90,8)(0,4){2}{\circle{2}} \put(90,4){\line(0,1){4}}

612: \put(90,6){\oval(4,12)[l]}

613: \end{picture}

614: 

615: \caption{\footnotesize{Partitions for the $L_y=2$ strip.}}

616: \label{L2partitions}

617: \end{figure}

618: 

619: 

620: \begin{figure}

621: \unitlength 1mm \hspace*{5mm}

622: \begin{picture}(90,12)

623: \put(10,12){\makebox(0,0){{\small $d=0$}}}

624: \multiput(20,4)(0,4){3}{\circle{2}}

625: \multiput(30,4)(0,4){3}{\circle{2}} \put(30,8){\oval(4,8)[l]}

626: \put(50,12){\makebox(0,0){{\small $d=1$}}}

627: \multiput(60,0)(10,0){4}{\circle*{2}}

628: \multiput(60,4)(10,0){4}{\circle{2}}

629: \multiput(60,8)(10,0){4}{\circle{2}}

630: \multiput(60,12)(10,0){4}{\circle{2}} \put(60,0){\line(0,1){4}}

631: \put(70,4){\oval(4,8)[l]} \put(80,6){\oval(4,12)[l]}

632: \put(90,0){\line(0,1){4}} \put(90,8){\oval(4,8)[l]}

633: \end{picture}

634: 

635: 

636: \vspace*{5mm} \hspace*{5mm}

637: \begin{picture}(70,20)

638: \put(10,20){\makebox(0,0){{\small $d=2$}}}

639: \multiput(20,0)(10,0){3}{\circle*{2}}

640: \multiput(20,4)(10,0){3}{\circle*{2}}

641: \multiput(20,8)(10,0){3}{\circle{2}}

642: \multiput(20,12)(10,0){3}{\circle{2}}

643: \multiput(20,16)(10,0){3}{\circle{2}} \put(20,4){\line(0,1){4}}

644: \put(20,6){\oval(4,12)[l]} \put(30,4){\line(0,1){4}}

645: \put(30,8){\oval(4,16)[l]} \put(40,8){\oval(4,8)[l]}

646: \put(40,8){\oval(6,16)[l]} \put(60,20){\makebox(0,0){{\small

647: $d=3$}}} \multiput(70,0)(0,4){3}{\circle*{2}}

648: \multiput(70,12)(0,4){3}{\circle{2}} \put(70,8){\line(0,1){4}}

649: \put(70,10){\oval(4,12)[l]} \put(70,10){\oval(6,20)[l]}

650: \end{picture}

651: 

652: \caption{\footnotesize{Partitions for the $L_y=3$ strip.}}

653: \label{L3partitions}

654: \end{figure}

655: 

656: 

657: \begin{figure}

658: \unitlength 1mm \hspace*{5mm}

659: \begin{picture}(60,12)

660: \put(10,12){\makebox(0,0){{\small $d=0$}}}

661: \multiput(20,0)(10,0){4}{\circle{2}}

662: \multiput(20,4)(10,0){4}{\circle{2}}

663: \multiput(20,8)(10,0){4}{\circle{2}}

664: \multiput(20,12)(10,0){4}{\circle{2}} \put(30,8){\oval(4,8)[l]}

665: \put(40,6){\oval(4,12)[l]} \put(50,4){\oval(4,8)[l]}

666: \end{picture}

667: 

668: \vspace*{5mm} \hspace*{5mm}

669: \begin{picture}(100,16)

670: \put(10,16){\makebox(0,0){{\small $d=1$}}}

671: \multiput(20,0)(10,0){9}{\circle*{2}}

672: \multiput(20,4)(10,0){9}{\circle{2}}

673: \multiput(20,8)(10,0){9}{\circle{2}}

674: \multiput(20,12)(10,0){9}{\circle{2}}

675: \multiput(20,16)(10,0){9}{\circle{2}} \put(20,0){\line(0,1){4}}

676: \put(30,4){\oval(4,8)[l]} \put(40,6){\oval(4,12)[l]}

677: \put(50,8){\oval(4,16)[l]} \put(60,0){\line(0,1){4}}

678: \put(60,12){\oval(4,8)[l]} \put(70,4){\oval(4,8)[l]}

679: \put(70,12){\oval(4,8)[l]} \put(80,0){\line(0,1){4}}

680: \put(80,10){\oval(4,12)[l]} \put(90,8){\oval(4,8)[l]}

681: \put(90,0){\line(0,1){4}} \put(100,8){\oval(4,8)[l]}

682: \put(100,8){\oval(6,16)[l]}

683: \end{picture}

684: 

685: 

686: \vspace*{5mm} \hspace*{5mm}

687: \begin{picture}(90,20)

688: \put(10,20){\makebox(0,0){{\small $d=2$}}}

689: \multiput(20,0)(10,0){8}{\circle*{2}}

690: \multiput(20,4)(10,0){8}{\circle*{2}}

691: \multiput(20,8)(10,0){8}{\circle{2}}

692: \multiput(20,12)(10,0){8}{\circle{2}}

693: \multiput(20,16)(10,0){8}{\circle{2}}

694: \multiput(20,20)(10,0){8}{\circle{2}} \put(20,4){\line(0,1){4}}

695: \put(20,6){\oval(4,12)[l]} \put(30,4){\line(0,1){4}}

696: \put(30,8){\oval(4,16)[l]} \put(40,4){\line(0,1){4}}

697: \put(40,10){\oval(4,20)[l]} \put(50,8){\oval(4,8)[l]}

698: \put(50,8){\oval(6,16)[l]} \put(60,8){\oval(4,8)[l]}

699: \put(60,10){\oval(6,20)[l]} \put(70,10){\oval(4,12)[l]}

700: \put(70,10){\oval(6,20)[l]} \put(80,4){\line(0,1){4}}

701: \put(80,6){\oval(4,12)[l]} \put(80,16){\oval(4,8)[l]}

702: \put(90,12){\oval(4,8)[l]} \put(90,4){\line(0,1){4}}

703: \put(90,10){\oval(6,20)[l]}

704: \end{picture}

705: 

706: 

707: \vspace*{5mm} \hspace*{5mm}

708: \begin{picture}(50,24)

709: \put(10,24){\makebox(0,0){{\small $d=3$}}}

710: \multiput(20,0)(10,0){4}{\circle*{2}}

711: \multiput(20,4)(10,0){4}{\circle*{2}}

712: \multiput(20,8)(10,0){4}{\circle*{2}}

713: \multiput(20,12)(10,0){4}{\circle{2}}

714: \multiput(20,16)(10,0){4}{\circle{2}}

715: \multiput(20,20)(10,0){4}{\circle{2}}

716: \multiput(20,24)(10,0){4}{\circle{2}} \put(20,8){\line(0,1){4}}

717: \put(20,10){\oval(4,12)[l]} \put(20,10){\oval(6,20)[l]}

718: \put(30,8){\line(0,1){4}} \put(30,10){\oval(4,12)[l]}

719: \put(30,12){\oval(6,24)[l]} \put(40,8){\line(0,1){4}}

720: \put(40,12){\oval(4,16)[l]} \put(40,12){\oval(6,24)[l]}

721: \put(50,12){\oval(4,8)[l]} \put(50,12){\oval(6,16)[l]}

722: \put(50,12){\oval(8,24)[l]}

723: \end{picture}

724: 

725: \vspace*{5mm} \hspace*{5mm}

726: \begin{picture}(20,28)

727: \put(10,28){\makebox(0,0){{\small $d=4$}}}

728: \multiput(20,0)(0,4){4}{\circle*{2}}

729: \multiput(20,16)(0,4){4}{\circle{2}} \put(20,12){\line(0,1){4}}

730: \put(20,14){\oval(4,12)[l]} \put(20,14){\oval(6,20)[l]}

731: \put(20,14){\oval(8,28)[l]}

732: \end{picture}

733: 

734: \caption{\footnotesize{Partitions for the $L_y=4$ strip.}}

735: \label{L4partitions}

736: \end{figure}

737: 

738: As in earlier work, we define $N_{P,\Lambda,L_y,\lambda}$ as the total number

739: of distinct eigenvalues of $T_{\Lambda,L_y}$, i.e.  the sum of the dimensions

740: of the submatrices $T_{\Lambda,L_y,d}$, modulo the multiplicity $c^{(d)}$:

741: %

742: \beq 

743: N_{P,\Lambda,L_y,\lambda} = \sum_{d=0}^{L_y}

744: n_P(\Lambda,L_y,d) \ . 

745: \label{nptot}

746: \eeq

747: %

748: Since we will use these results here, we include the relevant values in the

749: Appendix.  For cyclic or M\"obius strips of the square and triangular lattices,

750: %

751: \beq 

752: N_{P,\Lambda,L_y,\lambda}=2(L_y-1)! \

753: \sum_{j=0}^{[\frac{L_y}{2}]}\frac{(L_y-j)} {(j!)^2(L_y-2j)!} \quad

754: {\rm for} \ \ \Lambda=sq,tri 

755: \label{nptotform} 

756: \eeq

757: %

758: where $[\nu]$ denotes the integral part of $\nu$.  

759: 

760: For the corresponding strips of the honeycomb lattice, we calculated

761: $n_P(hc,L_y,d)$ and $N_{P,hc,L_y,\lambda}$ in Ref. \cite{hca}. The results that

762: are needed here are listed in the appendix. These correspond to the numbers of

763: partitions which allow the connections between vertices 2 and 3, between

764: vertices 4 and 5, etc. Therefore, in addition to the partitions given in eq.

765: (\ref{L3partitionlist}), the other partitions for the $L_y=3$ strip of the

766: honeycomb lattice are

767: %

768: \beq 

769: {\cal P}_{hc,3,0} = \{ 23 \} \ , \qquad {\cal P}_{hc,3,1} =

770: \{ \overline{23}; 23, \bar 1 \} \ , \qquad {\cal P}_{hc,3,2} = \{

771: \bar 1, \overline{23} \} \ . 

772: \eeq

773: %

774: In addition to the partitions given in eq.  (\ref{L4partitionlist}), the other

775: partitions for the $L_y=4$ strip of the honeycomb lattice are

776: %

777: \beqs 

778: {\cal P}_{hc,4,0} & = & \{ 23; 23, 14 \} \ , \qquad {\cal

779: P}_{hc,4,1} = \{ 23, \bar 4; \overline{23}; 23, \bar 1; 23,

780: \overline{14} \} \cr\cr {\cal P}_{hc,4,2} & = & \{ \overline{23},

781: \bar 4; 23, \bar 1, \bar 4; \bar 1, \overline{23} \} \ , \qquad

782: {\cal P}_{hc,4,3} = \{ \bar 1, \overline{23}, \bar 4 \} 

783: \eeqs

784: %

785: In addition to the partitions given in eq.  (\ref{L5partitionlist}), the other

786: partitions for the $L_y=5$ strip of the honeycomb lattice are

787: %

788: \beqs 

789: {\cal P}_{hc,5,0} & = & \{ 23; 45; 13, 45; 23, 14; 23, 15;

790: 23, 45; 145; 235; 245; 23, 145 \} \cr\cr {\cal P}_{hc,5,1} & = &

791: \{ 23, \bar 5; 23, \bar 4; \overline{23}; 23, \bar 1;

792: \overline{45}; 45, \bar 3; 45, \bar 2; 45, \bar 1; 13,

793: \overline{45}; 45, \overline{13}; 14, 23, \bar 5; 23,

794: \overline{14}; 23, \overline{15}; \cr\cr & & 23, \overline{45};

795: 45, \overline{23}; 23, 45, \bar 1; \overline{145}; \overline{235};

796: 235, \bar 1; \overline{245}; 245, \bar 1; 23, \overline{145} \}

797: \cr\cr {\cal P}_{hc,5,2} & = & \{ 23, \bar 4, \bar 5;

798: \overline{23}, \bar 5; 23, \bar 1, \bar 5; \overline{23}, \bar 4;

799: 23, \bar 1, \bar 4; \bar 1, \overline{23}; \bar 3, \overline{45};

800: \bar 2, \overline{45}; \bar 1, \overline{45}; 45, \bar 2, \bar 3;

801: 45, \bar 1, \bar 3; 45, \bar 1, \bar 2; \cr\cr & & \overline{13},

802: \overline{45}; 23, \overline{14}, \bar 5; \overline{23},

803: \overline{45}; 23, \bar 1, \overline{45}; 45, \bar 1,

804: \overline{23}; \bar 1, \overline{235}; \bar 1, \overline{245} \}

805: \cr\cr {\cal P}_{hc,5,3} & = & \{ \overline{23}, \bar 4, \bar 5;

806: 23, \bar 1, \bar 4, \bar 5; \bar 1, \overline{23}, \bar 5; \bar 1,

807: \overline{23}, \bar 4; \bar 2, \bar 3, \overline{45}; \bar 1, \bar

808: 3, \overline{45}; \bar 1, \bar 2, \overline{45}; 45, \bar 1, \bar

809: 2, \bar 3; \bar 1, \overline{23}, \overline{45} \} \cr\cr {\cal

810: P}_{hc,5,4} & = & \{ \bar 1, \overline{23}, \bar 4, \bar 5; \bar

811: 1, \bar 2, \bar 3, \overline{45} \} \ . 

812: \eeqs

813: %

814: 

815: For cyclic strips of the square and triangular lattices, general

816: properties include first \cite{cf}

817: %

818: \beqs 

819: dim(T_{\Lambda,L_y}) & = & P(\Lambda,L_y \times m

820: ,cyc.,q)_{m=0} =\sum_{d=0}^{L_y} c^{(d)}n_P(\Lambda,L_y,d) \cr\cr

821: & = & P({\rm Tree}_{L_y},q) = q(q-1)^{L_y-1} \ , \quad \Lambda=sq,tri 

822: \eeqs

823: %

824: where ${\rm Tree}_{L_y}$ denotes a tree graph (here a path graph) with

825: $L_y$ vertices; and

826: %

827: \beq 

828: Tr(T_{\Lambda,L_y}) = P(\Lambda,L_y \times m,cyc.,q)_{m=1} =

829: \sum_{d=0}^{L_y} c^{(d)}Tr(T_{\Lambda,L_y,d}) = 0 \ , \quad \Lambda=sq,tri 

830: \ . 

831: \label{trace_Ttotalzero} 

832: \eeq

833: %

834: For cyclic strips of the honeycomb lattice \cite{hca}

835: %

836: \beqs 

837: dim(T_{hc,L_y}) & = & P(hc,L_y \times m,cyc.,q)_{m=0}

838: =\sum_{d=0}^{L_y} c^{(d)}n_P(hc,L_y,d) \cr\cr & = &  \cases{

839: q\Bigl (q(q-1) \Bigr )^{\frac{L_y-1}{2}} & for \ $L_y$ \ odd \cr

840:           &   \cr

841:           \Bigl (q(q-1) \Bigr )^{\frac{L_y}{2}}  & for \ $L_y$ \ even }

842: \label{propfac} 

843: \eeqs

844: %

845: Here we give another general result for these strips of the honeycomb lattice,

846: governing the trace of the transfer matrix: 

847: %

848: \beqs 

849: Tr(T_{hc,L_y}) & = & P(hc, L_y \times m, cyc., q)_{m=1} =

850: \sum_{d=0}^{L_y} c^{(d)} Tr(T_{hc,L_y,d}) \cr\cr & = &

851: P({\rm Tree}_{2L_y},q) = q(q-1)^{2L_y-1} \ . 

852: \eeqs

853: %

854: This is proved by the same coloring matrix methods that we used in Ref.

855: \cite{cf} and \cite{hca}.  The reason for the difference relative to the

856: corresponding formula (\ref{trace_Ttotalzero}) for the cyclic strips of the

857: square and triangular lattices is that for $m=1$ on those strips the

858: longitudinal edges connect each vertex to itself, so the chromatic polynomial

859: vanishes, but, in contrast, for the strip of the honeycomb lattice, the

860: longitudinal edges have interior vertices of degree 2 associated with multiple

861: edges.  Because of this, and taking into account the elementary theorem that

862: the chromatic polynomial of a graph remains unchanged if one replaces any edge

863: by multiple copies of this edge, it follows that the resulting coloring is

864: described by the chromatic polynomial $P({\rm Tree}_{2L_y},q)$.  For the

865: reader's convenience in checking that our results satisfy these general trace

866: formulas, we shall list traces of our transfer matrices below.

867: 

868: In general,

869: %

870: \beq det(T_{\Lambda,L_y}) = \prod_{d=0}^{L_y}

871: [det(T_{\Lambda,L_y,d})]^{c^{(d)}} \ . 

872: \label{detTtotal} 

873: \eeq

874: %

875: 

876: 

877: We next discuss M\"obius strips.  One of the major results of the present paper

878: is a general formula, eq. (\ref{zgsum_transfermb}) together with

879: eqs. (\ref{cd0tran})-(\ref{cdoddtran}), for the partition function of the

880: zero-temperature $q$-state Potts antiferromagnet on the M\"obius strip of a

881: regular lattice $\Lambda$ of arbitrary width as well as arbitrary length.  This

882: formula involves a set of $\tilde T_{\Lambda,L_y,d}$ which occur once in each

883: $m$-fold product (with the $T_{\Lambda,L_y,d}$ thus raised to the power $m-1$)

884: and which differ from the $T_{\Lambda,L_y,d}$ in the interchange of certain

885: columns, encoding the M\"obius property of reversed-orientation periodic

886: longitudinal boundary conditions.  In our previous papers involving M\"obius

887: strips of the square and triangular lattices, we expressed the chromatic

888: polynomial in a form analogous to (\ref{pgsum}) but with different sets of

889: coefficients.  In Ref. \cite{cf} we derived a set of transformations that

890: governed how the coefficients for a certain $d$ change when one changes the

891: longitudinal boundary condition from cyclic to M\"obius for the square-lattice

892: strip.  However, as noted there, this did not apply to the triangular-lattice

893: strip; for that strip, if one expresses the chromatic polynomial in the same

894: manner for the cyclic and M\"obius cases, as a sum of $m$'th powers of

895: eigenvalues $\lambda_{tri,L_y,d,j}$, then some of the coefficients become

896: algebraic, rather than polynomial, functions of $q$.  This was evident in the

897: simplest nontrivial example, the $L_y=2$ strip \cite{wcy}.  The general

898: formalism that we introduce here for M\"obius strips has the advantage that the

899: coefficients stay in the set of $c^{(d)}$'s for triangular (and honeycomb) as

900: well as square lattice strips.  For $L_y \ge 3$, the $\tilde T_{\Lambda,L_y,d}$

901: matrices have eigenvalues that, in general, are not just related to those of

902: $\tilde T_{\Lambda,L_y,d}$ by possible sign changes.  In the limit of infinite

903: length, $m \to \infty$, because only the $T_{\Lambda,L_y,d}$ matrices are

904: raised to a power that goes to infinity, it is only the eigenvalues of these

905: matrices that determine locus ${\cal B}$.  Thus, the result of our previous

906: papers, that the locus ${\cal B}$ is the same for the infinite-length limit of

907: the strip of the lattice $\Lambda$ for cyclic or M\"obius boundary conditions,

908: is again evident in our present formulation.

909: 

910: We proceed to the details. For M\"obius strips one set of horizontal edges

911: reverses the order of connection. This corresponds to the exchange of the pair

912: of bases which switch to each other when the vertices reverse the order, i.e.,

913: the set of bases which do not have self-reflection symmetry with respect to the

914: center of the tree with $L_y$ vertices. For example, among the partitions for

915: the $L_y=2$ strip in Fig. \ref{L2partitions}, the first partition $\bar 2$ and

916: the second partition $\bar 1$ in ${\cal P}_{2,1}$ must be exchanged under this

917: reflection. The pairs of partitions for the $L_y=3$ strip in

918: Fig. \ref{L3partitions} are the first partition $\bar 3$ and the third

919: partition $\bar 1$ in ${\cal P}_{3,1}$, and the first partition $\bar 2, \bar

920: 3$ and the third partition $\bar 1, \bar 2$ in ${\cal P}_{3,2}$. This

921: corresponds to the exchange of these pairs of columns of $T_{\Lambda,L_y,d}$,

922: and these matrices will be denoted as $\tilde T_{\Lambda,L_y,d}$

923: \cite{zt}. For general $L_y$ we have the following changes of coefficients for

924: the square, triangular and honeycomb lattices \cite{cf}:

925: %

926: \beq 

927: c^{(0)} \to c^{(0)} \label{cd0tran} 

928: \eeq

929: %

930: \beq 

931: c^{(2k)} \to -c^{(k-1)} \ , \qquad 1 \le k \le \Bigl

932: [\frac{L_y}{2} \Bigr ] 

933: \label{cdeventran} 

934: \eeq

935: %

936: \beq 

937: c^{(2k+1)} \to c^{(k+1)} \ , \qquad 0 \le k \le \Bigl

938: [\frac{L_y-1}{2} \Bigr ]  \ . 

939: \label{cdoddtran} 

940: \eeq

941: %

942: We thus obtain the important general formula, in terms of transfer matrices, 

943: for the chromatic polynomial for a M\"obius strip of the lattice $\Lambda$,

944: which is valid for arbitrary width $L_y$:

945: %

946: \beqs

947: P(\Lambda, L_y \times m, Mb., q) & = &

948: c^{(0)}Tr[(T_{\Lambda,L_y,0})^{m-1} \tilde T_{\Lambda,L_y,0}] \cr\cr & & +

949: \sum_{d=0}^{[(L_y-1)/2]} c^{(d+1)} Tr[(T_{\Lambda,L_y,2d+1})^{m-1}

950: \tilde T_{\Lambda,L_y,2d+1}] \cr\cr & & - \sum_{d=1}^{[L_y/2]}

951: c^{(d-1)} Tr[(T_{\Lambda,L_y,2d})^{m-1} \tilde T_{\Lambda,L_y,2d}] \ . 

952: \label{zgsum_transfermb}

953: \eeqs

954: %

955: (where the factor $c^{(0)}=1$ is displayed explicitly for uniformity). 

956: 

957: For the square lattice, the eigenvalues of $\tilde T_{sq,L_y,d}$

958: are the same as those of $T_{sq,L_y,d}$ except for possible changes

959: of signs. The number of eigenvalues with sign changes is the

960: number of column-exchanges from $T_{sq,L_y,d}$ to $\tilde

961: T_{sq,L_y,d}$.  Denote the number of eigenvalues that are the

962: same for $T_{sq,L_y,d}$ and $\tilde T_{sq,L_y,d}$ as

963: $n_P(sq,L_y,d,+)$, and the number of eigenvalues with different signs

964: as $n_P(sq,L_y,d,-)$. It is clear that

965: %

966: \beq n_P(sq,L_y,d,+) + n_P(sq,L_y,d,-) = n_P(sq,L_y,d) \ . \label{npld}

967: \eeq

968: %

969: Define

970: %

971: \beq \Delta n_P(sq,L_y,d) \equiv n_P(sq,L_y,d,+) - n_P(sq,L_y,d,-)

972: \label{deltanpld} \eeq

973: %

974: which gives the number of partitions having self-reflection

975: symmetry. For example, among the partitions for the $L_y=2$ strip

976: in Fig. \ref{L2partitions}, the partition $I$ in ${\cal P}_{2,0}$

977: and the partition $\bar 1, \bar 2$ in ${\cal P}_{2,2}$ have

978: self-reflection symmetry. Among the partitions for the $L_y=3$

979: strip in Fig. \ref{L3partitions}, it includes the two partitions

980: in ${\cal P}_{3,0}$, the second partition $\bar 2$ and the fourth

981: partition $\overline{13}$ in ${\cal P}_{3,1}$, the second

982: partition $\bar 1, \bar 3$ in ${\cal P}_{3,2}$, and the partition

983: in ${\cal P}_{3,3}$. We list $\Delta n_P(sq,L_y,d)$ for $1 \le L_y

984: \le 10$ in Table \ref{nppmtable}. The total number of these

985: partitions for each $L_y$ will be denoted as $\Delta N_{P,L_y}$,

986: and we have $\Delta N_{P,2n-1} = \Delta N_{P,2n}$ for $n>0$. The

987: relations between $\Delta n_P(sq,L_y,d)$ are

988: %

989: \beqs 

990: \Delta n_P(sq,2n-1,0) & = & \Delta n_P(sq,2n,0) \ , \qquad

991: \mbox{for} \ 0<n \cr\cr \Delta n_P(sq,2n-1,2m-1) & = & \Delta

992: n_P(sq,2n,2m)  \ , \qquad \mbox{for} \ 1 \le m \le n \cr\cr 

993: \Delta n_P(sq,2n-1,2m) & = & \Delta n_P(sq,2n,2m-1) \ , \qquad \mbox{for} \ 1

994: \le m \le n \cr\cr 

995: \Delta n_P(sq,2n+1,0) & = & 2\Delta n_P(sq,2n,0) +

996: \Delta n_P(sq,2n,1) \ , \qquad \mbox{for} \ 0<n \cr\cr 

997: \Delta

998: n_P(sq,2n+1,m) & = & \Delta n_P(sq,2n,m-1) + \Delta n_P(sq,2n,m) + \Delta

999: n_P(sq,2n,m+1) \ , \cr\cr

1000: & & \mbox{for} \ 0<m \le 2n+1 \cr\cr 

1001: \Delta n_P(sq,2n+1,2m) & = & \Delta n_P(sq,2n+1,2m+1) \ , \qquad \mbox{for} \ 0

1002: \le m \le n \cr\cr 

1003: \Delta n_P(sq,2n,0) & = & \Delta n_P(sq,2n,2) \cr\cr

1004: \Delta n_P(sq,2n,2m-1) & = & \Delta n_P(sq,2n,2m+2) \ , \ \mbox{for} \

1005: 0<m \le n-1 \ .

1006: \eeqs

1007: %

1008: We also list $n_P(sq,L_y,d,+)$ and $n_P(sq,L_y,d,-)$ for $2 \le L_y \le 10$ in

1009: Table \ref{npctable}. Notice that $n_P(sq,L_y,0,+)$ is the number of

1010: $\lambda_{P,sq,FF,L_y}$ proved in \cite{ts}. Recall the numbers of

1011: $\lambda_{P,sq,L_y,j}$ for the M\"obius strips of the square lattice with

1012: coefficients $\pm c^{(d)}$, defined as $n_{P,Mb}(L_y,d,\pm) \equiv

1013: n_{P,Mb}(sq,L_y,d,\pm) $, have been given in \cite{cf}. With the eqs.

1014: (\ref{cd0tran}) to (\ref{cdoddtran}), the relations between $n_P(sq,L_y,d,\pm)$

1015: and $n_{P,Mb}(sq,L_y,d,\pm)$ are

1016: %

1017: \beqs 

1018: n_{P,Mb}(sq,L_y,0,\pm) & = & n_P(sq,L_y,0,\pm) + n_P(sq,L_y,2,\mp)

1019: \cr\cr n_{P,Mb}(sq,L_y,k,\pm) & = & n_P(sq,L_y,2k-1,\pm) +

1020: n_P(sq,L_y,2k+2,\mp) \ , \qquad 1 \le k \le \Bigl [ \frac{L_y+1}{2}

1021: \Bigr ] 

1022: \label{npmbnp} 

1023: \eeqs

1024: %

1025: We illustrate the application of this general formalism in the section where we

1026: give results for specific transfer matrices.  For M\"obius strips of the square

1027: lattice, the sum of the coefficients is given by

1028: %

1029: \beqs 

1030: P(sq,L_y \times m ,Mb.,q)_{m=0} & = & c^{(0)} \Delta

1031: n_P(sq,L_y,0) + \sum_{d=0}^{[(L_y-1)/2]} c^{(d+1)}\Delta

1032: n_P(sq,L_y,2d+1) \cr\cr & & - \sum_{d=1}^{[L_y/2]} c^{(d-1)}\Delta

1033: n_P(sq,L_y,2d) \cr\cr & = & c^{(0)} [\Delta n_P(sq,L_y,0) - \Delta

1034: n_P(sq,L_y,2)] \cr\cr & & + \sum_{d=1}^{[(L_y+1)/2]} c^{(d)} [\Delta

1035: n_P(sq,L_y,2d-1) - \Delta n_P(sq,L_y,2d+2)] \cr\cr & = & 

1036: \cases{ 0 & for

1037: \ $L_y$ \ even \cr & \cr P({\rm Tree}_{(L_y+1)/2},q)  & for \ $L_y$ \

1038: odd } \ 

1039: \eeqs

1040: %

1041: which agrees with Theorem 5 in \cite{cf} because of eqn.  (\ref{npmbnp}).  This

1042: $m=0$ equation also holds for the triangular lattice strips. If

1043: we take $m=1$ in eqn. (\ref{zgsum_transfermb}), the chromatic polynomial is

1044: zero for the square lattice with odd $L_y$ or for the triangular lattice with

1045: any $L_y$. This is what we expect since these M\"obius strips have at least one

1046: edge connect a vertex to itself.

1047: 

1048: For the honeycomb lattice with $L_y$ even, the magnitudes of

1049: eigenvalues of $T_{hc,L_y,d}$ and $\tilde T_{hc,L_y,d}$ are

1050: again the same. We denote the number of eigenvalues that are the

1051: same as $n_P(hc,L_y,d,+)$, and the number of eigenvalues with

1052: different signs as $n_P(hc,L_y,d,-)$. Similarly to eqns.

1053: (\ref{npld}) and (\ref{deltanpld}), we have

1054: %

1055: \beqs n_P(hc,L_y,d) & = & n_P(hc,L_y,d,+) + n_P(hc,L_y,d,-) \cr\cr

1056: \Delta n_P(hc,L_y,d) & \equiv & n_P(hc,L_y,d,+) - n_P(hc,L_y,d,-) \ . 

1057: \eeqs

1058: %

1059: The values of $\Delta n_P(hc,L_y,d)$ for even $L_y$ up to $L_y=10$ are listed

1060: in Table \ref{nphcpmtable}. The values $\Delta n_P(hc,L_y,d)$ can be obtained

1061: from $\Delta n_P(sq,L_y,d)$. For example,

1062: %

1063: \beqs \Delta n_P(hc,2,d) & = & \Delta n_P(sq,2,d) \cr\cr \Delta

1064: n_P(hc,4,d) & = & \Delta n_P(sq,3,d)+\Delta n_P(sq,4,d) \cr\cr \Delta

1065: n_P(hc,6,d) & = & \Delta n_P(sq,4,d)+\Delta n_P(sq,6,d) \cr\cr \Delta

1066: n_P(hc,8,d) & = & \Delta n_P(sq,5,d)+\Delta n_P(sq,6,d)+\Delta

1067: n_P(sq,7,d)+\Delta n_P(sq,8,d) \cr\cr \Delta n_P(hc,10,d) & = & \Delta

1068: n_P(sq,6,d)+2\Delta n_P(sq,8,d)+\Delta n_P(sq,10,d) 

1069: \eeqs

1070: %

1071: The relations between $\Delta n_P(hc,L_y,d)$ are

1072: %

1073: \beqs 

1074: \Delta n_P(hc,4n,0) & = & 4\Delta n_P(hc,4n-2,0) + 2\Delta

1075: n_P(hc,4n-2,1) \ , \qquad \mbox{for} \ 0<n \cr\cr 

1076: \Delta n_P(hc,4n,2m-1) & = & n_P(hc,4n,2m) \cr\cr 

1077: & = & 2(\Delta n_P(hc,4n-2,2m-1) + \Delta n_P(hc,4n-2,2m)) \cr\cr 

1078: & & + \Delta n_P(hc,4n-2,2m-2) + \Delta n_P(hc,4n-2,2m+1) \ , \cr\cr

1079: & & {\rm for} \ \ 1 \le m \le 2n \cr\cr 

1080: \Delta n_P(hc,4n+2,0) & = & \Delta

1081: n_P(hc,4n+2,2) \cr\cr 

1082: & = & \Delta n_P(hc,4n,0) + \Delta n_P(hc,4n,2) \ , \qquad \mbox{for} 

1083: \ 1 \le n \cr\cr \Delta

1084: n_P(hc,4n+2,2m-1) & = & n_P(hc,4n+2,2m+2) \cr\cr & = & \Delta

1085: n_P(hc,4n,2m-1) + \Delta n_P(hc,4n,2m+2) \ , \mbox{for} \ 1

1086: \le m \le 2n \cr\cr

1087: & & 

1088: \eeqs

1089: %

1090: $n_P(hc,L_y,d,+)$ and $n_P(hc,L_y,d,-)$ for even $2 \le L_y \le

1091: 10$ are listed in Table \ref{npchctable}. For M\"obius strips of

1092: the honeycomb lattice with even $L_y$, the sum of the coefficients

1093: is given by

1094: %

1095: \beqs 

1096: P(hc, L_y \times m ,Mb.,q)_{m=0} & = & c^{(0)} \Delta

1097: n_P(hc,L_y,0) + \sum_{d=0}^{[(L_y-1)/2]} c^{(d+1)}\Delta

1098: n_P(hc,L_y,2d+1) \cr\cr & & - \sum_{d=1}^{[L_y/2]} c^{(d-1)}\Delta

1099: n_P(hc,L_y,2d) \cr\cr & = & c^{(0)} [\Delta n_P(hc,L_y,0) - \Delta

1100: n_P(hc,L_y,2)] \cr\cr & & + \sum_{d=1}^{[(L_y+1)/2]} c^{(d)}

1101: [\Delta n_P(hc,L_y,2d-1) - \Delta n_P(hc,L_y,2d+2)] \cr\cr & = &

1102: \cases{ 0 & for \ $L_y/2$ \ odd \cr & \cr \Bigl ( q(q-1) \Bigr

1103: )^{L_y/4} & for \ $L_y/2$ \ even } 

1104: \eeqs

1105: %

1106: 

1107: 

1108: It will be convenient to introduce the following polynomial:

1109: %

1110: \beq

1111: D_n = \frac{P(C_n,q)}{q(q-1)} = \sum_{j=0}^{n-2} (-1)^j {n-1 \choose j}

1112: q^{n-2-j}

1113: \label{dk}

1114: \eeq

1115: %

1116: where $P(C_n,q)=(q-1)^n+(q-1)(-1)^n$ is the chromatic polynomial for the

1117: circuit graph with $n$ vertices, $C_n$. Various properties of these polynomials

1118: were given in Refs. \cite{hs,wa2}. The first few $D_n$'s are $D_1=0$, $D_2=1$,

1119: $D_3=q-2$, $D_4=q^3-3q+3$. In addition to the identities given in Refs.

1120: \cite{hs,wa2}, another identity is

1121: %

1122: \beq

1123: D_n+D_{n-1}=(q-1)^{n-2} \ , n \ge 2 \ . 

1124: \label{dnsum}

1125: \eeq

1126: %

1127: Certain linear combinations of these polynomials also occur in the entries of

1128: the transfer matrices.  These can be seen to arise as powers of basic linear

1129: factors such as $q-a$ plus integer constants; e.g., $q^2-4q+5=(q-2)^2+1$,

1130: $q^2-6q+10=(q-3)^2+1$, etc.  

1131: 

1132: 

1133: We shall write out the entries of the transfer matrices explicitly if this

1134: can be done in the space available.  In cases where this is not feasible, we

1135: shall express the entries in terms of the above polynomials, together with

1136: the shorthand notation

1137: %

1138: \beq

1139: q_a = q-a \ , \quad q_{a,b}=(q-a)(q-b)

1140: \label{qab}

1141: \eeq

1142: %

1143: and additional shorthand defined in the last appendix, including the

1144: polynomials $F_{m,n}$, $G_{m,n}$, and others. 

1145: 

1146: 

1147: \section{Some Transfer Matrix Results for Arbitrary $L_{\lowercase{y}}$}

1148: 

1149: \subsection{$\lowercase{d}=L_{\lowercase{y}}$}

1150: 

1151: For arbitrary $L_y$, eq. (\ref{nplyly}) shows that for $d=L_y$ the transfer

1152: matrix $T_{\Lambda,L_y,d=L}$ is one-dimensional, i.e., a scalar.  We have

1153: previously proved that \cite{s5}

1154: %

1155: \beq

1156: \lambda_{\Lambda,L_y,L_y}=(-1)^{L_y} \quad {\rm for} \quad \Lambda=sq,tri 

1157: \ . 

1158: \label{lamsqLL}

1159: \eeq

1160: %

1161: Our results also led to the inference that \cite{pg,hca}

1162: %

1163: \beq

1164: \lambda_{\Lambda,L_y,L_y}=1 \quad {\rm for} \quad \Lambda=hc \ . 

1165: \label{lamhcLL}

1166: \eeq

1167: %

1168: These results can also be seen to follow immediately from the transfer matrix

1169: method.

1170: 

1171: \subsection{$\lowercase{d}=L_{\lowercase{y}}-1$}

1172: 

1173: \subsubsection{Square lattice} 

1174: 

1175: For arbitrary $L_y$, the dimension of the transfer matrix is specified by

1176: eq. (\ref{nplylym1}) as $n_P(\Lambda,L_y,L_y-1)=L_y$ for $\Lambda=sq,tri$. For

1177: this case we have given the transfer matrices $T_{\Lambda,L_y,d=L_y-1}$ for

1178: cyclic strips of the square lattice in eqs. (7.1.1) and (7.1.4)-(7.1.6) of

1179: Ref. \cite{s5} and for cyclic strips of the triangular lattice in

1180: eqs. (7.3.1)-(7.3.4) of Ref. \cite{s5}.  General formulas for eigenvalues and

1181: determinants were also given in Ref. \cite{s5}.  Since we will need these here,

1182: we list them below:

1183: %

1184: \beq

1185: \lambda_{sq,L_y,d=L_y-1,j} = (-1)^{L_y+1}(q-a_{sq,L_y,j}) \quad , \ 1 \le j

1186: \le L_y

1187: \label{lamdlym1}

1188: \eeq

1189: %

1190: where

1191: %

1192: \beq

1193: a_{sq,L_y,j} = 1 + 4\cos^2 \left ( \frac{(L_y+1-j) \pi}{2L_y} \right )

1194: \quad , \ 1 \le j \le L_y \ .

1195: \label{asqlyj}

1196: \eeq

1197: %

1198: Hence, the determinant and trace of $T_{sq,L_y,L_y-1}$ are

1199: %

1200: \beq

1201: det(T_{sq,L_y,L_y-1}) = \prod_{j=1}^{L_y}(q-a_{sq,L_y,j})

1202: \label{detTsqLLminus1}

1203: \eeq

1204: %

1205: %

1206: \beq

1207: Tr(T_{sq,L_y,L_y-1}) = (-1)^{L_y+1}[2+L_y(q-3)] \ . 

1208: \label{traceTsqLLminus1}

1209: \eeq

1210: %

1211: Although our general result (\ref{lamdlym1}) exhibits an explicit factorization

1212: of the characteristic polynomial of $T_{sq,L_y,L_y-1}$ into linear factors,

1213: this involves, in general, transcendental numbers (trigonometric functions) as

1214: coefficients.  A few of these are of the form $\pm (q - k)$ where $k$ is an

1215: integer; e.g., for $L_y=2, \ d=1$, one has $\lambda_{sq,2,1,j}=1-q,3-q$ and for

1216: $L_y=3, \ d=2$, one has $\lambda_{sq,3,2,j}=q-1,q-2,q-4$.  Some others involve

1217: terms of the form $\pm (q - a)$ where $a$ is an algebraic number; e.g., for

1218: $L_y=4, \ d=3$, one has $\lambda_{4,3,j} =1-q,3-q,3 \pm \sqrt{2}-q$. In Table

1219: \ref{sqfactors} we indicate the factorization properties of the various

1220: characteristic polynomials of $T_{sq,L_y,d}$ in factors with integer

1221: coefficients; for the illustrative case $L_y=4,d=3$, this is

1222: $CP(T_{sq,4,3},z)=(z-1+q)(z-3+q)(z^2-2(3-q)z+q^2-6q+7)$, symbolized as

1223: $(1^2,2)$ in the table.  We observe that the factorizations often correspond to

1224: the numbers $n_P(sq,L_y,d,+)$ and $n_P(sq,L_y,d,-)$ in Table \ref{npctable},

1225: although sometimes there are additional factorizations.

1226: 

1227: 

1228: 

1229: \subsubsection{Triangular lattice}

1230: 

1231: For $T_{tri,L_y,L_y-1}$ we proved that \cite{s5} 

1232: %

1233: \beq

1234: det(T_{tri,L_y,L_y-1}) = (q-2)^2(q-3)^{L_y-2}

1235: \label{detTtriLLminus1}

1236: \eeq

1237: %

1238: \beq

1239: Tr(T_{tri,L_y,L_y-1}) = (-1)^{L_y+1}[3+L_y(q-4)] \ . 

1240: \label{traceTtriLLminus1}

1241: \eeq

1242: %

1243: Since a matrix and its transpose have the same eigenvalues, it is a convention

1244: which one is presented.  The transfer matrices obtained directly from our

1245: calculation of the analogous matrices for the full Potts model are the

1246: transposes of those given in \cite{s5}. This just amounts to a different

1247: convention for the ordering of the basis of configurational states.

1248: 

1249: \subsubsection{Honeycomb lattice}

1250: 

1251: Here we give an exact determination of $T_{hc,L_y,d}$ for $d=L_y-1$ which is

1252: valid for arbitrary $L_y$.  This goes beyond our results in Ref. \cite{s5},

1253: which presented such formulas for the square and triangular lattices.  We first

1254: observe that this matrix has dimension 

1255: %

1256: \beq

1257: dim(T_{hc,L_y,L_y-1}) = n_P(hc,L_y,L_y-1) = \cases{ 

1258:          \frac{3L_y-1}{2} & for \ $L_y$ \ odd \cr

1259:           &   \cr

1260:          \frac{3}{2}L_y-1  & for \ $L_y$ \ even }

1261: \label{nphclylyminus1}

1262: \eeq

1263: %

1264: This can be derived in a straightforward manner from results in \cite{hca}.  It

1265: is convenient to work with two auxiliary sets of matrices, each of dimension

1266: $n_Z(hc,L_y,L_y-1)=2L_y-1$, which are special cases of the transfer matrices

1267: for the full Potts model on strips of the honeycomb lattice,

1268: $S_{hc,L_y,L_y-1,j}$ for $j=1,2$.  The elements of these matrices will be given

1269: below.  We then form the product

1270: %

1271: \beq

1272: S_{hc,L_y,L_y-1}=S_{hc,L_y,L_y-1,2}S_{hc,L_y,L_y-1,1} \ .  

1273: \label{smatrix}

1274: \eeq

1275: %

1276: In this matrix certain columns are identically zero.  The number of

1277: such zero columns is

1278: %

1279: \beq

1280: n_Z(hc,L_y,L_y-1)-n_P(hc,L_y,L_y-1) =  \cases{

1281:          \frac{L_y-1}{2} & for \ $L_y$ \ odd \cr

1282:           &   \cr

1283:          \frac{L_y}{2}  & for \ $L_y$ \ even }

1284: \label{numberofzerocols}

1285: \eeq

1286: %

1287: The transfer matrix $T_{hc,L_y,L_y-1}$ is then the submatrix of

1288: $S_{hc,L_y,L_y-1}$ formed by removing these columns with vanishing entries and

1289: the corresponding rows, which operation we denote with a subscript $red.$ for

1290: ``reduced'':

1291: 

1292: %

1293: \beq

1294: T_{hc,L_y,L_y-1}=S_{hc,L_y,L_y-1,red.} \ . 

1295: \label{Thclylyminus1}

1296: \eeq

1297: %

1298: To determine $T_{hc,L_y,L_y-1}$, it thus suffices to give the 

1299: $S_{hc,L_y,L_y-1,j}$ matrices for $j=1,2$.  For the relevant range $L_y \ge 2$

1300: where the strips of the honeycomb lattice are defined we calculate 

1301: %

1302: \beq 

1303: (S_{hc,L_y,L_y-1,1})_{j,j}=2-q \quad {\rm for} \quad 1 \le j

1304: \le L_y-1 

1305: \label{TThc1x1} 

1306: \eeq

1307: %

1308: %

1309: \beq 

1310: (S_{hc,L_y,L_y-1,1})_{L_y,L_y}= \cases{ 2-q & for $L_y$ even \cr

1311:                                         1-q & for $L_y$ odd  }

1312: \label{TThc1xl} 

1313: \eeq

1314: %

1315: %

1316: \beq 

1317: (S_{hc,L_y,L_y-1,1})_{2j-1,2j}= (S_{hc,L_y,L_y-1,1})_{2j,2j-1}= 1

1318: \quad {\rm for} \quad 1 \le j \le [L_y/2] 

1319: \label{TThc1_1} 

1320: \eeq

1321: %

1322: %

1323: \beq 

1324: (S_{hc,L_y,L_y-1,1})_{L_y+2j-1,L_y+2j-1}=0 \quad {\rm for} \quad 1

1325: \le j \le [L_y/2] 

1326: \label{TThc1ya} 

1327: \eeq

1328: %

1329: %

1330: \beq 

1331: (S_{hc,L_y,L_y-1,1})_{L_y+2j,L_y+2j}=1 \quad {\rm for} \quad 1 \le

1332: j \le [(L_y-1)/2] 

1333: \label{TThc1yb} 

1334: \eeq

1335: %

1336: %

1337: \beq

1338: (S_{hc,L_y,L_y-1,1})_{2j-1,L_y+2j-1}=(S_{hc,L_y,L_y-1,1})_{2j,L_y+2j-1}=0

1339: \quad {\rm for} \quad 1 \le j \le [L_y/2]

1340: \label{TThc1wa} 

1341: \eeq

1342: %

1343: %

1344: \beq 

1345: (S_{hc,L_y,L_y-1,1})_{2j,L_y+2j}=(S_{hc,L_y,L_y-1,1})_{2j+1,L_y+2j}=-1

1346: \quad {\rm for} \quad 1 \le j \le [(L_y-1)/2]

1347: \label{TThc1wb} 

1348: \eeq

1349: %

1350: %

1351: \beq

1352: (S_{hc,L_y,L_y-1,1})_{L_y+2j-1,2j-1}=(S_{hc,L_y,L_y-1,1})_{L_y+2j-1,2j}=-1 

1353: \quad {\rm for} \quad 1 \le j \le [L_y/2] 

1354: \label{TThc1v} 

1355: \eeq

1356: %

1357: %

1358: \beq (S_{hc,L_y,L_y-1,2})_{1,1}=1-q 

1359: \label{TThc2x} 

1360: \eeq

1361: %

1362: %

1363: \beq 

1364: (S_{hc,L_y,L_y-1,2})_{j,j}=2-q \quad {\rm for} \quad 2 \le j \le L_y-1 

1365: \label{TThc2x1} 

1366: \eeq

1367: %

1368: %

1369: \beq 

1370: (S_{hc,L_y,L_y-1,2})_{L_y,L_y}= \cases{ 2-q & for $L_y$ odd \cr

1371:                                         1-q & for $L_y$ even  }

1372: \label{TThc2xl} 

1373: \eeq

1374: %

1375: %

1376: \beq 

1377: (S_{hc,L_y,L_y-1,2})_{2j,2j+1}= (S_{hc,L_y,L_y-1,1})_{2j+1,2j}= 1

1378: \quad {\rm for} \quad 1 \le j \le [(L_y-1)/2] 

1379: \label{TThc2_1} 

1380: \eeq

1381: %

1382: %

1383: \beq 

1384: (S_{hc,L_y,L_y-1,2})_{L_y+2j-1,L_y+2j-1}=1 \quad {\rm for} \quad 1

1385: \le j \le [L_y/2] 

1386: \label{TThc2ya} 

1387: \eeq

1388: %

1389: %

1390: \beq 

1391: (S_{hc,L_y,L_y-1,2})_{L_y+2j,L_y+2j}=0 \quad {\rm for} \quad 1 \le

1392: j \le [(L_y-1)/2] 

1393: \label{TThc2yb} 

1394: \eeq

1395: %

1396: %

1397: \beq

1398: (S_{hc,L_y,L_y-1,2})_{2j-1,L_y+2j-1}=(S_{hc,L_y,L_y-1,2})_{2j,L_y+2j-1}=-1

1399:  \quad {\rm for} \quad 1 \le j \le [L_y/2]

1400: \label{TThc2wa} 

1401: \eeq

1402: %

1403: %

1404: \beq 

1405: (S_{hc,L_y,L_y-1,2})_{2j,L_y+2j}=(S_{hc,L_y,L_y-1,2})_{2j+1,L_y+2j}=0

1406: \quad {\rm for} \quad 1 \le j \le [(L_y-1)/2]

1407: \label{TThc2wb} 

1408: \eeq

1409: %

1410: %

1411: \beq 

1412: (S_{hc,L_y,L_y-1,2})_{L_y+2j,2j}=(S_{hc,L_y,L_y-1,1})_{L_y+2j,2j+1}=-1

1413: \quad {\rm for} \quad 1 \le j \le [(L_y-1)/2] 

1414: \label{TThc2v}

1415: \eeq

1416: %

1417: with all other elements equal to zero. Using our general formulas, we find

1418: that

1419: %

1420: \beq

1421: det(T_{hc,L_y,L_y-1})=(q-1)^{L_y+1} \quad {\rm for \ odd} \ \ L_y \ge 3 \ . 

1422: \label{detThclylyminus1}

1423: \eeq

1424: %

1425: For even $L_y$, $det(T_{hc,L_y,L_y-1})$ is proportional to 

1426: $(q-1)^{L_y}F_{4,3}=(q-1)^{L_y}(q^2-4q+5)$. 

1427: 

1428: We next proceed to list our explicit calculations of transfer matrices. Since

1429: we have completely determined the transfer matrices in the cases $d=L_y$, where

1430: it is a scalar, and for the case $d=L_y-1$, we do not list the former for each

1431: specific $L_y$.

1432: 

1433: \section{Strips of the Square Lattice}

1434: 

1435: \subsection{$L_{\lowercase{y}}=2$}

1436: 

1437: We recall first that for $L_y=1$, an elementary calculation yields $P(sq,1

1438: \times m, cyc.,q)=(q-1)^m+(q-1)(-1)^m$ so $T_{sq,1,0}=q-1$.  The chromatic

1439: polynomials $P(sq,2 \times m,BC,q)$ for cyclic and M\"obius BC's were given in

1440: Ref. \cite{bds} One has $n_P(\Lambda,2,0)=1$, $n_P(\Lambda,2,1)=2$, 

1441: $n_P(\Lambda2,2)=1$ for $\Lambda=sq$ (or $tri$), $T_{sq,2,0}=q^2-3q+3$, and

1442: %

1443: \beq

1444: T_{sq,2,1} = -\left( \begin{array}{cc}

1445:              q-2 & -1  \\

1446:              -1  & q-2 \end{array} \right )

1447: \label{Tsq21}

1448: \eeq

1449: %

1450: yielding the eigenvalues $\lambda_{sq,2,1,1}=1-q$ and

1451: $\lambda_{sq,2,1,1}=3-q$.  The determinant $det(T_{sq,2,1})=(q-1)(q-3)$ and

1452: trace $Tr(T_{sq,2,1})=-2(q-2)$ follow as the $L_y=2$ special case of our

1453: results in \cite{s5}, viz., eqs. (\ref{detTsqLLminus1}) and

1454: (\ref{traceTsqLLminus1}).  For the present strip, the general methods discussed

1455: above yield, for the M\"obius strip, the matrix $\tilde T_{sq,2,1}$ which has

1456: the first and second columns interchanged. 

1457: 

1458: \subsection{$L_{\lowercase{y}}=3$}

1459: 

1460: The chromatic polynomials $P(sq,3 \times m,BC,q)$ were given in \cite{wcyl,wcy}

1461: for cyclic BC's and in \cite{pm} for M\"obius BC's.  We find 

1462: %

1463: \beq

1464: T_{sq,3,0}=  \left( \begin{array}{cc}

1465:     (q-2)(q^2-3q+4) & q^2-4q+5  \\

1466:          1          & q-2        \end{array} \right )

1467: \label{Tsq30}

1468: \eeq

1469: %

1470: 

1471: \bigskip

1472: 

1473: %

1474: \beq

1475: T_{sq,3,1}= \left( \begin{array}{cccc}

1476:       -(q^2-4q+5) &  q-2      &  -1         &  2-q    \\

1477:          q-2      & -(q-2)^2  &  q-2        &    1    \\

1478:          -1       &  q-2      &-(q^2-4q+5)  &  2-q    \\

1479:           1       &   1       &   1         &  q-2    \end{array} \right )

1480: \label{Tsq31}

1481: \eeq

1482: %

1483: 

1484: \bigskip

1485: %

1486: \beq

1487: T_{sq,3,2}= \left( \begin{array}{ccc}

1488:          q-2   &   -1    &   0   \\

1489:           -1   &   q-3   &  -1   \\

1490:            0   &  -1     &  q-2  \end{array} \right )

1491: \label{Tsq32}

1492: \eeq

1493: %

1494: Although $T_{sq,3,0}$ is the same as the analogous transfer matrix for the free

1495: strip of width $L_y=3$, $T_{sq,L_y,0}$ is larger than the free-strip transfer

1496: matrix for $L_y \ge 4$.  Note that the upper left-hand $3 \times 3$ block of

1497: $T_{sq,3,1}$ is symmetric.  The matrix $\tilde T_{sq,3,1}$ is obtained by

1498: exchanging columns 1 and 3 of $T_{sq,3,1}$, and $\tilde T_{sq,3,2}$ is obtained

1499: by exchanging columns 1 and 3 of $T_{sq,3,2}$. 

1500: 

1501: We have

1502: %

1503: \beq

1504: det(T_{sq,3,0})=(q-1)(q^3-6q^2+13q-11)

1505: \eeq

1506: %

1507: \beq

1508: det(T_{sq,3,1})=-(q-1)(q-2)^2(q^4-9q^3+29q^2-40q+22)

1509: \eeq

1510: %

1511: \beq

1512: det(T_{sq,3,2})=(q-1)(q-2)(q-4)

1513: \eeq

1514: %

1515: \beq

1516: Tr(T_{sq,3,0})=(q-2)(q^2-3q+5)

1517: \eeq

1518: %

1519: \beq

1520: Tr(T_{sq,3,1})=-(3q^2-13q+16) \ . 

1521: \eeq

1522: %

1523: $Tr(T_{sq,3,2})$ is given by the $L_y=3,d=2$ special case of the general

1524: formula (7.1.28) of \cite{s5}, given above as eq. (\ref{traceTsqLLminus1}).

1525: The characteristic polynomial $CP(T_{sq,3,1},z)$ factorizes into polynomials of

1526: degree 1 and 3 in $z$.  This is indicated in Table \ref{sqfactors}. The

1527: eigenvalue corresponding to the linear factor is $\lambda_{sq,3,1,1}=-(q-2)^2$

1528: \cite{wcyl}.

1529: 

1530: 

1531: \subsection{$L_{\lowercase{y}}=4$}

1532: 

1533: The chromatic polynomials $P(sq,4 \times m,BC,q)$ for BC=cyclic and M\"obius

1534: were given in \cite{s4}.  We find

1535: %

1536: \beq

1537: T_{sq,4,0}= \left( \begin{array}{cccc}

1538: s_{21} & (q-2)p_6   & r_{20}    & (q-2)p_6  \\

1539:   q-2  &  (q-2)^2   & 3-q       & 1        \\

1540:   -1   &   2-q      & q^2-5q+7  & 2-q      \\

1541:   q-2  &    1       & 3-q       & (q-2)^2  \end{array} \right )

1542: \label{Tsq40}

1543: \eeq

1544: %

1545: where the polynomials $s_{21}$, $p_6$, and $r_{20}$ are defined in the

1546: appendix.  

1547: 

1548: We have

1549: %

1550: \beqs

1551: & & det(T_{sq,4,0}) = (q-1)(q-3)(q^8-16q^7+112q^6-449q^5+1130q^4-1829q^3 \cr\cr

1552: & & +1858q^2-1084q+279)

1553: \eeqs

1554: %

1555: and

1556: %

1557: \beq

1558: Tr(T_{sq,4,0}) = q^4-7q^3+24q^2-45q+36 \ . 

1559: \eeq

1560: %

1561: As indicated in Table \ref{sqfactors}, the characteristic polynomial of

1562: $T_{sq,4,0}$ consists of factors of degree 1 and 3 in $z$. The eigenvalue which

1563: is the root of the linear factor is $\lambda_{sq,4,0,1}=(q-1)(q-3)$; the

1564: eigenvalues of the cubic factor are the same as the eigenvalues for the free

1565: strip of width $L_y=4$ determined in Ref. \cite{strip}.  The matrix 

1566: $\tilde T_{sq,4,0}$ is obtained by exchanging columns 2 and 4 of $T_{sq,4,0}$.

1567: 

1568: \bigskip

1569: 

1570: %

1571: \beq

1572: T_{sq,4,1}= \left( \begin{array}{ccccccccc}

1573: -r_{13} &F_{4,3}     & -q_2 & 1  &-G_{4,3}& q_2   &-G_{4,3} & -q_2^2 & q_2  \\

1574: F_{4,3} &-q_2F_{4,3} &q_2^2 &-q_2& q_2    &-q_2^2 & q_3   & q_2 & -1   \\

1575: -q_2  & q_2^2  & -q_2F_{4,3}  &F_{4,3}& -1  & q_2   & q_3 &-q_2^2 & q_2  \\

1576:  1  & -q_2     &F_{4,3}&-r_{13}       & q_2 & -q_2^2&-G_{4,3}&q_2 &-G_{4,3} \\

1577:  -1  & 0       & 0       & 0    &-q_2 & 0        & 0    & 0        & 0    \\

1578:  -1  & q_2     & q_2     & q_2  &-q_2 & q_2^2    &-q_3  & 1        & 1    \\

1579:  -1  & -1      & -1      & -1   &-q_2 &-q_2      & G_{4,3} &-q_2   &-q_2  \\

1580:  q_2 & q_2     & q_2     & -1   & 1    & 1       &-q_3   &q_2^2    &-q_2  \\

1581:  0   & 0       & 0       & -1   & 0    & 0        & 0    & 0       &-q_2

1582: \end{array} \right )

1583: \label{Tsq41}

1584: \eeq

1585: %

1586: The shorthand notation used in eq. (\ref{Tsq41}) and below, such as $F_{m,n}$

1587: and $G_{m,n}$, is defined in the last appendix.  Note that the upper

1588: left-hand $4 \times 4$ block of $T_{sq,4,1}$ is symmetric.  We calculate

1589: %

1590: \beqs

1591: & & det(T_{sq,4,1}) = (q-1)^3(q-3)^5(q^2-3q+3)^2 \cr\cr

1592: & \times & (q^8-17q^7+125q^6-520q^5+1342q^4-2206q^3+2261q^2-1325q+341)

1593: \label{detTsq41}

1594: \eeqs

1595: %

1596: \beq

1597: Tr(T_{sq,4,1}) = -4q^3+27q^2-69q+65 \ . 

1598: \eeq

1599: %

1600: The matrix $\tilde T_{sq,4,1}$ is obtained by exchanging columns 1 and 4, 2 and

1601: 3, 5 and 9, and 6 and 8 of $T_{sq,4,1}$.  It is ironic that although the

1602: chromatic polynomial is a special case of the full Potts model partition

1603: function, the determinants of the transfer matrices $T_{\Lambda,L_y,d}$ for the

1604: degree-$d$ subspaces for these square-lattice strips, such as

1605: eq. (\ref{detTsq41}), are more complicated than the simple expression that we

1606: found for the determinant of the transfer matrix $T_{Z,sq,L_y,d}$ in

1607: Ref. \cite{zt},

1608: %

1609: \beq

1610: det(T_{Z,sq,L_y,d}) = v^{L_y[ n_Z(L_y,d) - n_Z(L_y-1,d)]} 

1611: \biggl [ (v+qr)^{L_y} (v+1)^{L_y-1} \biggr ]^{n_Z(L_y-1,d)}

1612: \label{detTZsqld}

1613: \eeq

1614: %

1615: where here $n_Z(L_y,d) \equiv n_Z(\Lambda,L_y,d)$ for $\Lambda=sq,tri$. 

1616: Furthermore, the expression for $det(T_{sq,L_y,d})$ for these square-lattice

1617: strips gets more complicated as $L_y$ increases, in contrast to

1618: eq. (\ref{detTZsqld}) and also in contrast to the results that we find for 

1619: $det(T_{\Lambda,L_y,d})$ for $\Lambda=tri,hc$ (see below).  

1620: 

1621: As indicated in Table \ref{sqfactors}, the characteristic polynomial of

1622: $T_{sq,4,1}$ consists of factors of degree 4 and 5.

1623: 

1624: \bigskip

1625: 

1626: %

1627: \beq

1628: T_{sq,4,2}= \left( \begin{array}{cccccccc}

1629: F_{4,3} & 2-q     & 1        &  0    &  0       &   0   & q- 2   &  0    \\

1630:  2-q   & q^2-5q+6 & 3-q      & 2-q   &  1       &   0   & -1    &  0    \\

1631:   1    & 3-q      & p_8      & 1     & 3-q      &   1   & q-2   & q-2   \\

1632:   0    & 2-q      & 1        &(q-2)^2& 2-q      &   0   & 0     & 0     \\

1633:   0    & 1        & 3-q      & 2-q   & q_{2,3}   & 2-q   & 0     & -1    \\

1634:   0    & 0        & 1        & 0     & 2-q      &F_{4,3} & 0     & q-2   \\

1635:  -1    & -1       & -1       & 0     & 0        & 0     & 2-q   & 0     \\

1636:   0    & 0        & -1       & 0     & -1       & -1    & 0     & 2-q

1637: \end{array} \right )

1638: \label{Tsq42}

1639: \eeq

1640: %

1641: where $F_{m,n}$ and $p_8$ are defined in the last appendix.  The

1642: upper left-hand $6 \times 6$ block of $T_{sq,4,2}$ is symmetric.  The matrix

1643: $\tilde T_{sq,4,2}$ is obtained by exchanging columns 1 and 6, 2 and 5, and 7

1644: and 8 of $T_{sq,4,2}$. $CP(T_{sq,4,2},z)$ has factors of degree 3 and 5. We

1645: have

1646: %

1647: \beqs

1648: & & det(T_{sq,4,2})=(q-1)^2(q-3)^2(q^3-7q^2+15q-11) \cr\cr

1649: & \times & (q^7-16q^6+106q^5-378q^4+788q^3-967q^2+653q-189)

1650: \eeqs

1651: %

1652: \beq

1653: Tr(T_{sq,4,2}) = 6q^2-29q+38

1654: \eeq

1655: %

1656: 

1657: \bigskip

1658: 

1659: %

1660: \beq

1661: T_{sq,4,3}= -\left( \begin{array}{cccc}

1662:   q-2 & -1  &  0  &  0   \\

1663:    -1 & q-3 & -1  &  0   \\

1664:    0  & -1  & q-3 & -1   \\

1665:    0  & 0   & -1  & q-2  \end{array} \right )

1666: \label{Tsq43}

1667: \eeq

1668: %

1669: 

1670:  From the general formulas (\ref{detTsqLLminus1}) and (\ref{traceTsqLLminus1}),

1671: we have $det(T_{sq,4,3})=(q-1)(q-3)(q^2-6q+7)$ and

1672: $Tr(T_{sq,4,3})=-2(2q-5)$.

1673: 

1674: 

1675: \subsection{$L_{\lowercase{y}}=5$}

1676: 

1677: The chromatic polynomials $P(sq,5 \times m,BC,q)$ for BC=cyclic and M\"obius

1678: were given in \cite{s5}.  In addition to the general determination of

1679: $T_{sq,L_y,L_y-1}$, the transfer matrix for $T_{sq,5,3}$ was given in

1680: \cite{s5}. We display here our calculation of $T_{sq,5,0}$:

1681: %

1682: \beqs

1683: & & T_{sq,5,0}= \cr\cr & & \left(

1684: \begin{array}{ccccccccc} F_{4,3} r_{11} & s_{31} & s_{47} &

1685: s_{61} & q_2^2p_7 & s_{47} & s_{31}& r_{19} & q_2p_8\\

1686: F_{4,3}& q_2F_{4,3} & -q_{2,3} & 2q-5 & q_2 & q_3 & q_2 & 1 & q_2^2\\

1687: -q_2 & -q_2^2 & q_2G_{4,3} & -p_8 & -q_2^2 & q_3 & -1 & -q_2 & -q_2\\

1688: 1 & q_2 & -G_{4,3} & r_{20} & q_2 & -G_{4,3} & q_2 & G_{4,3} & q_2^2\\

1689: q_2^2 & q_2 & -q_{2,3} & 2q_3 & q_2^3 & -q_{2,3} & q_2 & q_{2,3}& 1\\

1690: -q_2 & -1 & q_3 & -p_8 & -q_2^2 & q_2G_{4,3} & -q_2^2 & -q_2 & -q_2\\

1691: F_{4,3} & q_2 & q_3 & 2q-5 & q_2 & -q_{2,3} & q_2F_{4,3} & 1 & q_2^2\\

1692: 0 & 0 & 0 & 1 & 0 & 0 & 0 & q_2 & 0\\

1693: -1 & -q_2 & -q_3 & 5-2q & 1 & -q_3 & -q_2 & -1 & -q_2^2

1694: \end{array} \right ) \cr\cr

1695: & & \label{Tsq50}

1696: \eeqs

1697: %

1698: where the various $p, r, s$ polynomials are defined in the appendix. 

1699: The characteristic polynomial of $T_{sq,5,0}$ consists of factors of 

1700: degree 2 and 7 in $z$. The eigenvalues that are the roots of the 

1701: degree-7 factor are the same as those found in Ref. \cite{strip} for the 

1702: corresponding $L_y=5$ free strip of the square lattice.  

1703: 

1704: The transfer matrices $T_{sq,5,d}$ for $d=1,2$ are both of dimension $21 \times

1705: 21$ and are too lengthy to present here; they are available from the

1706: authors. We find

1707: %

1708: \beqs

1709: & & det(T_{sq,5,0})=-(q-1)^3(q-2)^2(q^6-13q^5+69q^4-191q^3+292q^2-236q+79)

1710: \cr\cr

1711: & & \times (q^{15}-34q^{14}+538q^{13}-5259q^{12}+35541q^{11}-176036q^{10}

1712: \cr\cr

1713: & & +660682q^9-1914798q^8+4324155q^7-7615130q^6+10381339q^5 \cr\cr

1714: & & -10768339q^4+8235159q^3-4388527q^2+1459163q-228580) \ . 

1715: \eeqs

1716: %

1717: The determinant $det(T_{sq,5,1})$ factorizes as $(q-1)^6(q-2)^2P_{20}P_{29}$,

1718: where $P_{20}$ and $P_{29}$ are polynomials in $q$ of degree 20 and 29 in

1719: $q$. $det(T_{sq,5,2})$ factorizes as $(q-1)^6(q-2)P_{19}P_{23}$, where $P_{19}$

1720: and $P_{23}$ are polynomials in $q$ of degree 19 and 23. For $d=3$ we have

1721: %

1722: \beqs

1723: & & det(T_{sq,5,3})=(q-1)^3(q-2)(q^8-20q^7+171q^6-817q^5+2389q^4

1724: \cr\cr

1725: & & -4387q^3+4954q^2-3155q+869)(q^{11}-29q^{10}+375q^9-2853q^8+14188q^7

1726: \cr\cr

1727: & & -48439q^6+115934q^5-194762q^4+225461q^3-171683q^2+77617q-15835)

1728: \eeqs

1729: %

1730: and, from eq. (\ref{detTsqLLminus1}),

1731: %

1732: \beq

1733: det(T_{sq,5,4})=(q-1)(q^2-5q+5)(q^2-7q+11)

1734: \eeq

1735: %

1736: We obtain the traces

1737: %

1738: \beq

1739: Tr(T_{sq,5,0})=q^5-9q^4+42q^3-120q^2+195q-137

1740: \eeq

1741: %

1742: \beq

1743: Tr(T_{sq,5,1})=-5q^4+46q^3-180q^2+346q-269

1744: \eeq

1745: %

1746: \beq

1747: Tr(T_{sq,5,2})=10q^3-74q^2+198q-189

1748: \eeq

1749: %

1750: \beq

1751: Tr(T_{sq,5,3})=-10q^2+51q-69

1752: \eeq

1753: %

1754: with $Tr(T_{sq,5,4})$ being given by eq. (\ref{traceTsqLLminus1}).

1755: 

1756: 

1757: \section{Strips of the Triangular Lattice}

1758: 

1759: \subsection{$L_{\lowercase{y}}=2$}

1760: 

1761: The chromatic polynomials $P(tri,2 \times m,BC,q)$ for BC=cyclic and M\"obius

1762: were given in \cite{wcy}.  One has $\lambda_{tri,2,0}=(q-2)^2$, and, as a

1763: special case of our general result in eqs. (7.1.1) and (7.3.1)-(7.3.4) of

1764: \cite{s5},

1765: %

1766: \beq

1767: T_{tri,2,1} = -\left( \begin{array}{cc}

1768:               q-3 & q-2  \\

1769:               -1  & q-2 \end{array} \right )

1770: \label{Ttri21} \eeq

1771: %

1772: with eigenvalues

1773: %

1774: \beq

1775: \lambda_{tri,2,1,j} = \frac{1}{2}\biggl [ 5-2q \pm \sqrt{9-4q} \ \biggr ]

1776: \label{lamtri21}

1777: \eeq

1778: %

1779: Note that we use a different ordering convention for the basis configurations

1780: here than the one we used in Ref. \cite{s5}, so that $T_{tri,L_y,L_y-1}$ in

1781: terms of our present basis is the transpose of the corresponding matrix given

1782: in \cite{s5}. This has no effect on the eigenvalues, since the eigenvalues of a

1783: matrix $A$ and its transpose $A^T$ are the same.  We have

1784: $det(T_{tri,2,1})=(q-2)^2$ and $Tr(T_{tri,2,1})=5-2q$, in accordance with the

1785: general formulas (\ref{detTtriLLminus1}) and (\ref{traceTtriLLminus1}).  For

1786: the M\"obius strip, the $\tilde T_{tri,2,1}$ matrix is obtained from

1787: $T_{tri,2,1}$ by interchanging the two columns, as was true for the square

1788: lattice strip.  However, in contrast to the case of the square-lattice strip,

1789: where this just reverses the sign of one of the eigenvalues, here the $\tilde

1790: T_{tri,2,1}$ matrix has different eigenvalues than $T_{tri,2,1}$ (namely,

1791: $(1/2)(3-q \pm \sqrt{5q^2-22q+25} \ )$).  It is instructive to display how our

1792: general formula operates.  For comparison, we first display the explicit

1793: chromatic polynomial for the cyclic strip \cite{wcy}:

1794: %

1795: \beq

1796: P(tri,2 \times m,cyc.,q)=c^{(0)}(\lambda_{tri,2,0})^m +

1797: c^{(1)}\biggl [ (\lambda_{tri,2,1,1})^m + (\lambda_{tri,2,1,2})^m \biggr ]

1798:  + c^{(2)}

1799: \label{tpxy2}

1800: \eeq

1801: %

1802: If one expresses the chromatic polynomial for the

1803: M\"obius strip as a sum of powers of the same eigenvalues as for the cyclic 

1804: strip, the result is \cite{wcy}

1805: %

1806: \beq

1807: P(tri,2 \times m,Mb.,q)=c^{(0)}(\lambda_{tri,2,0})^m 

1808: -\frac{(q-1)(q-3)}{\sqrt{9-4q}}\biggl [ (\lambda_{tri,2,1,1})^m - 

1809: (\lambda_{tri,2,1,2})^m \biggr ] - c^{(0)}

1810: \label{tpxy2mb}

1811: \eeq

1812: %

1813: One sees that the coefficient of the second and third terms is no longer in the

1814: set of $c^{(d)}$'s and, indeed, is not a polynomial function of $q$.  Of

1815: course, the square root in the denominator of this coefficient cancels so that,

1816: as must be true, the chromatic polynomial is a polynomial in $q$.  As discussed

1817: in \cite{pm}, this can be seen as a consequence of the theorem on symmetric

1818: functions of roots of a polynomial equation whose coefficients are polynomials

1819: in $q$; although these roots are not, in general polynomial functions of $q$,

1820: the symmetric functions can be expressed in terms of polynomials in $q$.  In

1821: the present case, the second and third terms can be written in a manifestly

1822: symmetric manner by observing that the denominator of the coefficient is equal

1823: to $\lambda_{tri,2,1,1}-\lambda_{tri,2,1,2}$.  With our general formula

1824: (\ref{zgsum_transfermb}) we express the chromatic polynomial in a form that

1825: keeps the coefficients in the set of $c^{(d)}$ polynomials. 

1826: 

1827: \subsection{$L_{\lowercase{y}}=3$}

1828: 

1829: The chromatic polynomials $P(tri,3 \times m,BC,q)$ for BC=cyclic and M\"obius

1830: were given in \cite{t}.  We have

1831: %

1832: \beq

1833: T_{tri,3,0} = \left( \begin{array}{cc}

1834:      (q-2)(q^2-5q+7) & (q-3)^2  \\

1835:           2-q        & q-3 \end{array} \right )

1836: \label{Ttri30}

1837: \eeq

1838: %

1839: 

1840: \bigskip

1841: This yields the same eigenvalues that were earlier calculated for the free

1842: $L_y=3$ strip in Ref. \cite{strip} and can be taken to be the transfer matrix 

1843: for that strip. 

1844: %

1845: \beq

1846: T_{tri,3,1}= \left( \begin{array}{cccc}

1847:  -(q^2-6q+10)& -(q-2)(q-3)  & q-2         &  3-q  \\

1848:      q-3     & -(q-2)(q-3)  & -(q-2)(q-3) &  3-q  \\

1849:      -1      & q-2          & -(q-2)(q-3) &  3-q  \\

1850:      2       & 2-q          & 2-q     &  q-3  \end{array} \right )

1851: \label{Ttri31}

1852: \eeq

1853: %

1854: 

1855: \bigskip

1856: 

1857: The characteristic polynomial $CP(T_{tri,3,0},z)$ is a quadratic, while

1858: $CP(T_{tri,3,1},z)$ has a linear and cubic factor, as indicated in Table

1859: \ref{trifactors}.  We note that for odd $L_y$ the factorizations given in Table

1860: \ref{trifactors} correspond to the numbers $n_P(sq,L_y,d,+)$ and

1861: $n_P(sq,L_y,d,-)$ given in Table \ref{npctable}, but there are no nontrivial

1862: factorizations for even $L_y$.

1863: 

1864: 

1865: 

1866: \bigskip

1867: As a special case of our general formulas given in eqs. (7.3.1)-(7.3.4) of

1868: \cite{s5}, we have

1869: %

1870: \beq

1871: T_{tri,3,2}= \left( \begin{array}{ccc}

1872:          q-3   &  q-4     &  q-2   \\

1873:          -1    &  q-4     &  q-2   \\

1874:          0     &  -1      &  q-2  \end{array} \right )

1875: \label{Ttri32} \eeq

1876: %

1877: We calculate

1878: %

1879: \beq

1880: det(T_{tri,3,0})=(q-2)^3(q-3)

1881: \eeq

1882: %

1883: \beq

1884: det(T_{tri,3,1}) = -(q-2)^5(q-3)^2

1885: \eeq

1886: %

1887: \beq

1888: det(T_{tri,3,2})=(q-2)^2(q-3)

1889: \eeq

1890: %

1891: \beq

1892: Tr(T_{tri,3,0})=q^3-7q^2+18q-17

1893: \eeq

1894: %

1895: \beq

1896: Tr(T_{tri,3,1})=-3q^2+17q-25

1897: \eeq

1898: %

1899: \beq

1900: Tr(T_{tri,3,2})=3(q-3) \ . 

1901: \eeq

1902: %

1903: 

1904: 

1905: \subsection{$L_{\lowercase{y}}=4$}

1906: 

1907: The chromatic polynomials $P(tri,4 \times m,cyc.,q)$ were given in \cite{t}.

1908: The transfer matrices that we give below, together with our general procedure

1909: for obtaining $\tilde T_{\Lambda,L,d}$ from $T_{\Lambda,L,d}$, yields a

1910: new solution for $P(tri,4 \times m,Mb.,q)$. We find

1911: %

1912: \beq

1913: T_{tri,4,0}= \left( \begin{array}{cccc}

1914:  (q-2)(q-3)p_8   & (q-2)(q-3)(q-4) & (q-3)p_{15}  & (q-3)p_{10}   \\

1915: -(q-2)(q-3)      & (q-2)(q-3)      & 2(3-q)       & 3-q           \\

1916:   q-2            & (q-2)(q-3)      & q^2-7q+13    & 3-q           \\

1917: -(q-2)(q-3)      & (q-2)(q-3)      & (q-3)(q-5)   & (q-3)^2

1918: \end{array} \right )

1919: \label{Ttri40}

1920: \eeq

1921: %

1922: where $p_8$, etc. are defined in the appendix.  This yields

1923: the same characteristic polynomial and eigenvalues as those obtained for 

1924: the free $L_y=4$ strip in Ref. \cite{strip} and can be taken as the transfer

1925: matrix for that strip.  We have 

1926: %

1927: \beq

1928: det(T_{tri,4,0})=(q-2)^6 (q-3)^4

1929: \eeq

1930: %

1931: \beq

1932: Tr(T_{tri,4,0})=q^4-10q^3+42q^2-88q+76 \ . 

1933: \eeq

1934: %

1935: (We recall that the actual form of the matrix is basis-dependent; for example,

1936: for the free $L_y=4$ strip, a different matrix which, however, also has the

1937: same characteristic polynomial and eigenvalues originally calculated in

1938: Ref. \cite{strip}, was given in Ref. \cite{tritran}.)

1939: %

1940: \beq

1941: T_{tri,4,1}=\left( \begin{array}{ccccccccc}

1942: -r_{34}&-q_2p_{10} &q_{2,3}  &-q_2 &-q_{3,5} &-q_{2,3} &-p_{13} &-q_3^2 &q_3 \\

1943: p_{10} &-q_2p_{10} &-q_2q_3^2&q_{2,3}&q_3 &-q_{2,3} &2q_3 &-q_3^2 &q_3  \\

1944: -q_3   &q_{2,3}&-q_2q_3^2&-q_2p_{10} &q_3 &-q_{2,3} &-p_{13}&-q_3^2 &-q_{3,4} \\

1945: 1    &-q_2  &q_{2,3} &-q_2p_{10} &q_3 &-q_{2,3} &-p_{13} &q_3  &-q_{3,4} \\

1946: q_4  &q_2   & 0      & 0         &-q_3& 0       & 0      & 0   & 0       \\

1947: -2   &2q_2  & -q_{2,3}&-q_{2,3}  &-q_3& q_{2,3} & -2q_3  &-q_3 & -q_3    \\

1948: -2   &q_2   & q_2     & q_2      &-2q_3&q_{2,3} &p_{13}  &-q_3 & -q_3    \\

1949: 2q_3 &-q_{2,3}&-q_{2,3}&2q_2     &-2q_3&q_{2,3} &p_{14}  &q_3^2&-2q_3    \\

1950: 0    & 0    & 0        & q_2     & 0   & 0      & 1      & 0   & -q_3

1951: \end{array} \right )

1952: \label{Ttri41}

1953: \eeq

1954: %

1955: We calculate

1956: %

1957: \beq

1958: det(T_{tri,4,1})=(q-2)^{12}(q-3)^8

1959: \eeq

1960: %

1961: \beq

1962: Tr(T_{tri,4,1})=-2(q-3)(2q^2-12q+21)

1963: \eeq

1964: %

1965: 

1966: %

1967: \beq

1968: T_{tri,4,2}=\left( \begin{array}{cccccccc}

1969:  p_{10} & q_{3,4} & -q_4   & q_{2,3} & -q_2    & 0       & q_3 & 0    \\

1970:  -q_3   & q_{3,4} & p_{14} & q_{2,3} & q_{2,4} & -q_2    & q_3 & q_3  \\

1971:  1      & -q_4    & p_{14} & -q_2    & q_{2,4} & -q_2    & q_3 & q_3  \\

1972:  0      & -q_3    & -q_4   & q_{2,3} & q_{2,4} & q_{2,3} & 0   & q_3  \\

1973:  0      & 1       & -q_4   & -q_2    & q_{2,4} & q_{2,3} & 0   & q_3  \\

1974:  0      & 0       & 1      & 0       & -q_2    & q_{2,3} & 0   & q_3  \\

1975: -2      & q_4     & q_4    & q_2     & q_2     & 0       & -q_3& 0    \\

1976: 0       & 0       & -2     & 0       & q_2     & q_2     & 0   & -q_3

1977: \end{array} \right )

1978: \label{Ttri42}

1979: \eeq

1980: %

1981: %

1982: \beq

1983: det(T_{tri,4,2})=(q-2)^8(q-3)^6

1984: \eeq

1985: %

1986: \beq

1987: Tr(T_{tri,4,2})=6q^2-38q+62

1988: \eeq

1989: %

1990: 

1991: As a special case of our general result in \cite{s5}, we have

1992: %

1993: \beq

1994: T_{tri,4,3}= -\left( \begin{array}{cccc}

1995:   q-3  &  q-4  &  q-4 &  q-2   \\

1996:   -1   &  q-4  &  q-4 &  q-2   \\

1997:   0    &  -1   &  q-4 &  q-2   \\

1998:   0    &  0    &  -1  &  q-2

1999: \end{array} \right )

2000: \label{Ttri43}

2001: \eeq

2002: %

2003: The determinant and trace are the $L_y=4$ special case of our general formulas

2004: (\ref{detTtriLLminus1}) and (\ref{traceTtriLLminus1}), viz.,

2005: $det(T_{tri,4,3})=(q-2)^2(q-3)^2$ and $Tr(T_{tri,4,3})=13-4q$.

2006: 

2007: \subsection{$L_{\lowercase{y}}=5$}

2008: 

2009: We present here new results for $P(tri,5 \times m,BC,q)$ for BC=cyclic and

2010: M\"obius.  For $d=0$ we calculate

2011: %

2012: \beqs

2013: & & T_{tri,5,0}= \cr\cr

2014: & & \left( \begin{array}{ccccccccc}

2015: q_2 s_{82} & q_{2,3}p_{13} & q_2 r_{58} & s_{173} & q_{2,3}p_{13} & s_{148}

2016:   &q_3 r_{34}&q_3 p_{17}&q_5q_3^2\\

2017: -q_2 p_{10}&q_2q_3^2 &-2q_{2,3} & -q_4^2& -q_{2,3}&-p_{13} &-q_{3,4}&-q_3

2018:   &q_3^2\\

2019:  q_{2,3}& q_2q_3^2 &q_2p_{13} & -q_4^2& -q_{2,3}  & 2q_3  & q_3   &-q_3

2020:   &q_3^2\\

2021: -q_2   & -q_{2,3}  &q_2 p_{13}  &r_{47} & -q_{2,3} &-p_{13} & q_3 &q_{3,5}

2022:   &q_3^2\\

2023: -q_2q_3^2& q_2q_3^2&q_2q_4^2 &-q_4(2q-7)&q_2q_3^2&-2q_3^2&-q_3^2&q_{3,4}

2024:   &q_3^2\\

2025: q_{2,3}&-q_{2,3}&q_2p_{13}& r_{49} & q_2q_3^2 &q_3p_{13}&-q_3^2&q_3(2q-7)

2026:   &q_3^2 \\

2027: -q_2 p_{10} &-q_{2,3} &-2q_{2,3} &-q_4(2q-7) &q_2q_3^2 &r_{46} &q_3p_{10}

2028:   &q_{3,4}&q_3^2 \\

2029:   0    & 0        & -q_2   & -q_4   & 0        & 0    & 0     & q_3  & 0    \\

2030: 2q_2 & q_{2,3}   & 2q_{2,3} & q_{4,5} & q_{2,3}  &p_{14} &-2q_3  &2q_3

2031:   & -q_3^2

2032: \end{array} \right ) \cr\cr

2033: & &

2034: \label{Ttri50}

2035: \eeqs

2036: %

2037: where the various $p$, $r$, and $s$ polynomials are defined in the appendix.

2038: We find

2039: %

2040: \beq

2041: det(T_{tri,5,0})=-(q-2)^{14}(q-3)^{12}

2042: \eeq

2043: %

2044: \beq

2045: Tr(T_{tri,5,0})=(q-3)(q^4-10q^3+46q^2-112q+118) \ . 

2046: \eeq

2047: %

2048: The characteristic polynomial of $T_{tri,5,0}$ consists of factors of degree 2

2049: and 7.

2050: 

2051: The transfer matrices $T_{tri,5,d}$ for $d=1,2$ are both of dimension $21

2052: \times 21$ and are too lengthy to present here; they are available from the

2053: authors.  However, we do note the simple results

2054: %

2055: \beq

2056: det(T_{tri,5,1})=(q-2)^{30}(q-3)^{27}

2057: \eeq

2058: %

2059: \beq

2060: Tr(T_{tri,5,1})=-5q^4+62q^3-306q^2+708q-642

2061: \eeq

2062: %

2063: %

2064: \beq

2065: det(T_{tri,5,2})=(q-2)^{25}(q-3)^{24}

2066: \eeq

2067: %

2068: %

2069: \beq

2070: Tr(T_{tri,5,2})=10q^3-98q^2+332q-387 \ . 

2071: \eeq

2072: %

2073: Factorizations of $CP(T_{tri,5,j},z)$, $j=1,2$, are given in Table

2074: \ref{trifactors}.

2075: 

2076: For $d=3$ we calculate

2077: %

2078: \beqs

2079: & & T_{tri,5,3}= \cr\cr

2080: & & \left( \begin{array}{ccccccccccccc}

2081: -p_{10}&-q_{3,4}&q_4    &-q_{3,4}&q_4 &0  &-q_{2,3} &q_2 &0 &0 &-q_3 &0 & 0 \\

2082: q_3&-q_{3,4}&-p_{14}&-q_{3,4}&-q_4^2 &q_4 &-q_{2,3}&-q_{2,4}&q_2&0&-q_3&-q_3&0\\

2083: -1 & q_4 &-p_{14} &q_4 &-q_4^2 &q_4 &q_2 &-q_{2,4} &q_2 &0 &-q_3 &-q_3 & 0  \\

2084: 0&q_3&q_4&-q_{3,4}&-q_4^2&-p_{14} &-q_{2,3} &-q_{2,4}&-q_{2,4}&q_2&0&-q_3&-q_3\\

2085: 0&-1 &q_4 &q_4 &-q_4^2 &-p_{14} &q_2 &-q_{2,4}&-q_{2,4}&q_2 &0 &-q_3 &-q_3 \\

2086: 0 &0 &-1 &0 &q_4 &-p_{14} &0 &q_2 &-q_{2,4} &q_2 &0 &-q_3 &-q_3 \\

2087: 0 &0 &0 &q_3 &q_4 &q_4 &-q_{2,3} &-q_{2,4} &-q_{2,4} &-q_{2,3} &0 &0 &-q_3 \\

2088: 0 & 0 & 0 &-1 &q_4 &q_4  &q_2&-q_{2,4} &-q_{2,4} &-q_{2,3}& 0 & 0 &-q_3 \\

2089: 0 & 0 & 0 & 0 & -1 & q_4 & 0 & q_2 & -q_{2,4} & -q_{2,3} & 0 & 0 & -q_3 \\

2090: 0 & 0 & 0 & 0 & 0  & -1  & 0 & 0 & q_2 & -q_{2,3} & 0 & 0 & -q_3\\

2091: 2 & -q_4 & -q_4 & -q_4 & -q_4 & 0 & -q_2 & -q_2 & 0 & 0 & q_3 & 0 & 0 \\

2092: 0 & 0 & 2 & 0 & -q_4 & -q_4 & 0 & -q_2 & -q_2 & 0 & 0 & q_3 & 0 \\

2093: 0 & 0 & 0 & 0 & 0 & 2 & 0 & 0 & -q_2 & -q_2 & 0 & 0 & q_3

2094: \end{array} \right ) \cr\cr

2095: & &

2096: \label{Ttri53}

2097: \eeqs

2098: %

2099: We calculate

2100: %

2101: \beq

2102: det(T_{tri,5,3})=(q-2)^{11}(q-3)^{12}

2103: \eeq

2104: %

2105: \beq

2106: Tr(T_{tri,5,3})=-10q^2+67q-115 \ . 

2107: \eeq

2108: %

2109: The characteristic polynomial of $T_{tri,5,3}$ has factors of degree 5 and 8.

2110: 

2111: For $d=4$, our general theorem in \cite{s5} yields

2112: %

2113: \beq

2114: T_{tri,5,4}= \left( \begin{array}{ccccc}

2115:   q-3 & q-4 & q-4 & q-4 & q-2  \\

2116:   -1  & q-4 & q-4 & q-4 & q-2  \\

2117:   0   & -1  & q-4 & q-4 & q-2  \\

2118:   0   & 0   & -1  & q-4 & q-2  \\

2119:   0   & 0   & 0   & -1  & q-2   \end{array} \right )

2120: \label{Ttri54} \eeq

2121: %

2122: In accordance with our general results (\ref{detTtriLLminus1}) and

2123: (\ref{traceTtriLLminus1}), we have

2124: %

2125: \beq

2126: det(T_{tri,5,4})=(q-2)^2(q-3)^3

2127: \eeq

2128: %

2129: 

2130: \beq

2131: Tr(T_{tri,5,4})=5q-17 \ . 

2132: \eeq

2133: %

2134: The characteristic polynomial of $T_{tri,5,4}$ has factors of degree 2 and 3.

2135: 

2136: 

2137: \section{Strips of the Honeycomb Lattice}

2138: 

2139: \subsection{$L_{\lowercase{y}}=2$}

2140: 

2141: The chromatic polynomials $P(hc,2 \times m,BC,q)$ for BC=cyclic and M\"obius

2142: were given in \cite{pg}.  For $d=0$, $T_{hc,2,0}$ is a scalar,

2143: $T_{hc,2,0}=\lambda_{hc,2,0}=D_6=q^4-5q^3+10q^2-10q+5$. For $d=1$, from our

2144: general formulas above, we obtain 

2145: %

2146: \beq

2147: S_{hc,2,1,1}= \left( \begin{array}{ccc}

2148:            2-q  & 1   &  0  \\

2149:            1    & 2-q &  0  \\

2150:            -1   & -1  &  0  \end{array} \right )

2151: \label{Shc211}

2152: \eeq

2153: %

2154: \beq

2155: S_{hc,2,1,2}= \left( \begin{array}{ccc}

2156:            1-q  & 0   & -1  \\

2157:            0    & 1-q & -1  \\

2158:            0    & 0   &  1  \end{array} \right )

2159: \label{Shc212}

2160: \eeq

2161: %

2162: whence 

2163: %

2164: \beq

2165: T_{hc,2,1} = \left( \begin{array}{cc}

2166:            D_4      & -D_3  \\

2167:            -D_3     &  D_4   \end{array} \right ) =

2168:  \left( \begin{array}{cc}

2169:            q^2-3q+3      & 2-q  \\

2170:            2-q     &  q^2-3q+3   \end{array} \right )

2171: \label{Thc21}

2172: \eeq

2173: %

2174: yielding the relevant special case of the eigenvalues given in eq. (6), (7)

2175: of \cite{pg}, $\lambda_{hc,2,1,1}=D_4-D_3=F_{4,3}=q^2-4q+5$ and

2176: $\lambda_{hc,2,1,2}=D_4+D_3=(q-1)^2$.  We note that

2177: %

2178: \beq

2179: det(T_{hc,2,1})=(q-1)^2(q^2-4q+5)

2180: \label{detT_hc21}

2181: \eeq

2182: %

2183: and $Tr(T_{hc,2,1})=2D_4$

2184: 

2185: \subsection{$L_{\lowercase{y}}=3$}

2186: 

2187: Further, 

2188: %

2189: \beq

2190: T_{hc,3,0}= \left( \begin{array}{ccc}

2191:    q_{1,2}s_7 & q_2s_{3,5}   &  t_{5,5}   \\

2192:    q_1        & F_{4,3}       &  -q_3      \\

2193: -F_{6,2}      & -q_2p_4      & -F_{5,4}   \end{array} \right )

2194: \label{Thc30}

2195: \eeq

2196: %

2197: so that 

2198: %

2199: \beq

2200: det(T_{hc,3,0})=(q-1)^4(q-2)^2

2201: \eeq

2202: %

2203: \beq

2204: Tr(T_{hc,3,0})=q^6-8q^5+28q^4-56q^3+71q^2-58q+26 \ . 

2205: \eeq

2206: 

2207: For $d=1$, 

2208: %

2209: \beq

2210: T_{hc,3,1}= \left( \begin{array}{cccccc}

2211: q_{1,2}p_4  &-q_1 F_{4,3} & 1       & r_{13}  & -G_{4,3} & q_2 p_6    \\

2212: -q_1 q_2^2  &q_{1,2} D_4  &-D_5     &-q_{2,3} & q_3 D_4  & -2q_2^2    \\

2213: q_{1,2}     &-q_1D_4      & F_{6,5} &-q_{2,3} & q_3 D_4  &q_2 F_{4,3} \\

2214: -q_{1,2}    & q_1         & 0       &-q_3     & 1        & -q_2       \\

2215: -q_{1,2}    & q_1 D_4     &D_5      &-2q_2    & 2D_4     & q_{2,3}    \\

2216:   q_1       & q_1         & -1      &-q_3     & -q_3     & F_{4,3}

2217: \end{array} \right )

2218: \label{Thc31}

2219: \eeq

2220: %

2221:  from which it follows that 

2222: %

2223: \beq

2224: det(T_{hc,3,1})=(q-1)^8(q-2)^2

2225: \eeq

2226: %

2227: \beq

2228: Tr(T_{hc,3,1})=3q^4-18q^3+46q^2-60q+37 \ . 

2229: \eeq

2230: %

2231: 

2232: For $d=2$, applying our general formulas above, we calculate

2233: %

2234: \beq

2235: S_{hc,3,2,1}= \left( \begin{array}{ccccc}

2236:   2-q  &   1   &   0   &   0   &   0   \\

2237:   1    &  2-q  &   0   &   0   &  -1   \\

2238:   0    &  0    &  1-q  &   0   &   -1  \\

2239:   -1   &  -1   &   0   &   0   &   0   \\

2240:   0    &   0   &   0   &   0   &   1  

2241: \end{array} \right )

2242: \label{Thc321}

2243: \eeq

2244: %

2245: \beq

2246: S_{hc,3,2,2}= \left( \begin{array}{ccccc}

2247:   1-q  &   0   &   0   &  -1   &   0   \\

2248:    0   &  2-q  &   1   &  -1   &   0   \\

2249:    0   &   1   &  2-q  &   0   &   0   \\

2250:    0   &   0   &   0   &   1   &   0   \\

2251:    0   &  -1   &  -1   &   0   &   0

2252: \end{array} \right )

2253: \label{Th322}

2254: \eeq

2255: % 

2256: The factorizations for $\Lambda=hc$ are given in Table \ref{hcfactors}.  We

2257: note that for even $L_y$, the factorizations in this table correspond to the

2258: numbers $n_P(hc,L_y,d,+)$ and $n_P(hc,L_y,d,-)$ given in Table \ref{nphctable},

2259: when these numbers can be defined, but there are no nontrivial factorizations

2260: for odd $L_y$. We calculate

2261: %

2262: \beq

2263: det(T_{hc,3,2})=(q-1)^4

2264: \eeq

2265: %

2266: \beq

2267: Tr(T_{hc,3,2})=3q^2-10q+12

2268: \eeq

2269: %

2270: 

2271: 

2272: \subsection{$L_{\lowercase{y}}=4$}

2273: 

2274: For the $6 \times 6$ dimensional matrix $T_{hc,4,0}$ we calculate 

2275: %

2276: \beq

2277: det(T_{hc,4,0})=2(q-1)^8(q-2)^2(q^2-5q+7)

2278: (2q^6-20q^5+83q^4-185q^3+239q^2-175q+60)

2279: \eeq

2280: %

2281: \beq

2282: Tr(T_{hc,4,0})=q^8-11q^7+55q^6-165q^5+333q^4-480q^3+503q^2-362q+142 \ . 

2283: \eeq

2284: %

2285: The characteristic polynomial has factors of degree 1 and 5, as indicated in

2286: Table \ref{hcfactors}.  The eigenvalues are

2287: %

2288: \beq

2289: \lambda_{hc,4,0,1}=(q-1)^2(q^2-5q+7)

2290: \eeq

2291: %

2292: and the roots of the factor of degree 5.

2293: 

2294: 

2295: For the $13 \times 13$ dimensional matrix $T_{hc,4,1}$ we calculate 

2296: %

2297: \beqs

2298: & & det(T_{hc,4,1})=4(q-1)^{16}(q-2)^2

2299: (q^8-15q^7+100q^6-389q^5+974q^4-1623q^3 \cr\cr

2300: & & +1774q^2-1171q+360)(q^{12}-20q^{11}+185q^{10}-1048q^9+4061q^8

2301: \cr\cr

2302: & & -11385q^7+23784q^6-37468q^5+44352q^4-38630q^3+23631q^2-9200q+1750)

2303: \cr\cr

2304: & &

2305: \eeqs

2306: %

2307: \beq

2308: Tr(T_{hc,4,1})=4q^6-36q^5+147q^4-351q^3+529q^2-489q+229 \ . 

2309: \eeq

2310: %

2311: The characteristic polynomial factorizes into factors of degree 5 and 8.

2312: 

2313: For the $11 \times 11$ dimensional matrix $T_{hc,4,2}$ we find

2314: %

2315: \beqs

2316: & & det(T_{hc,4,2})=(q-1)^{12}(q^6-10q^5+44q^4-108q^3+160q^2-141q+60) \cr\cr

2317: & & \times (2q^8-29q^7+186q^6-690q^5+1634q^4-2564q^3+2647q^2-1670q+500)

2318: \eeqs

2319: %

2320: \beq

2321: Tr(T_{hc,4,2})=6q^4-39q^3+110q^2-157q+104

2322: \eeq

2323: %

2324: 

2325: For $T_{hc,4,3}$, applying our general formula (\ref{Thclylyminus1}), we

2326: calculate 

2327: %

2328: \beq

2329: T_{hc,4,3}= \left( \begin{array}{ccccc}

2330: D_4     & 2-q     & 0        & 0      & 0         \\

2331: 3-q     & F_{4,3} & 2-q      & 1      & q-3       \\

2332: 1       & 2-q     & F_{4,3}  & 3-q    & q-3       \\

2333: 0       & 0       & 2-q      & D_4    & 0         \\

2334: -1      & q-2     & q-2      & -1     & 2    \end{array} \right )

2335: \label{Thc43}

2336: \eeq

2337: %

2338: whence 

2339: %

2340: \beq

2341: det(T_{hc,4,3})=2(q-1)^4(q^2-4q+5)

2342: \eeq

2343: %

2344: \beq

2345: Tr(T_{hc,4,3})=2(2q^2-7q+9)

2346: \eeq

2347: %

2348: The eigenvalues are

2349: %

2350: \beq

2351: \lambda_{hc,4,3,1}=(q-1)^2

2352: \eeq

2353: %

2354: \beq

2355: \lambda_{hc,4,3,2}=F_{4,3}=q^2-4q+5

2356: \eeq

2357: %

2358: and the three roots of

2359: %

2360: \beq

2361: z^3-2(q^2-4q+6)z^2+(q^4-8q^3+26q^2-36q+21)z-2(q-1)^2=0 \ . 

2362: \eeq

2363: %

2364: 

2365: \subsection{$L_{\lowercase{y}}=5$}

2366: 

2367: We have calculated the $T_{hc,5,d}$ and find 

2368: %

2369: \beq

2370: det(T_{hc,5,0}) =(q-1)^{36}(q-2)^{12}

2371: \label{detThc50}

2372: \eeq

2373: %

2374: \beqs

2375: Tr(T_{hc,5,0}) & = & q^{10}-14q^9+91q^8-364q^7+1007q^6-2058q^5+3232q^4 \cr\cr

2376: & & -3950q^3+3662q^2-2350q+808

2377: \eeqs

2378: %

2379: %

2380: \beq

2381: det(T_{hc,5,1}) =(q-1)^{78}(q-2)^{24}

2382: \label{detThc51}

2383: \eeq

2384: %

2385: %

2386: \beqs

2387: Tr(T_{hc,5,1}) & = & 5q^8-60q^7+336q^6-1152q^5+2676q^4-4374q^3 \cr\cr

2388: & & +4999q^2-3720q+1429

2389: \eeqs

2390: %

2391: %

2392: \beq

2393: det(T_{hc,5,2})=(q-1)^{66}(q-2)^{16}

2394: \label{detThc52}

2395: \eeq

2396: %

2397: %

2398: \beq

2399: Tr((T_{hc,5,2})=10q^6-96q^5+422q^4-1084q^3+1742q^2-1686q+805

2400: \eeq

2401: %

2402: %

2403: \beq

2404: det(T_{hc,5,3})=(q-1)^{30}(q-2)^4

2405: \label{detThc53}

2406: \eeq

2407: %

2408: \beq

2409: Tr(T_{hc,5,3})=10q^4-68q^3+202q^2-302q+207

2410: \eeq

2411: %

2412: %

2413: \beq

2414: det(T_{hc,5,4})=(q-1)^6

2415: \label{detThc54}

2416: \eeq

2417: %

2418: \beq

2419: Tr(T_{hc,5,4})=5q^2-18q+24

2420: \eeq

2421: %

2422: The matrix $T_{hc,5,4}$ is sufficiently small that we can list it here:

2423: %

2424: \beq

2425: T_{hc,5,4}= \left( \begin{array}{ccccccc}

2426:  D_4   &  2-q    &   0    &   0    &  0     &  0     &  0      \\

2427:  3-q   &F_{4,3}  & 2-q    &   1    &  0     & q-3    &  0      \\

2428:   1    & 2-q     &F_{4,3} & 3-q    &  0     & q-3    &  0      \\

2429:   0    & 0       & 3-q    &F_{4,3} &  1-q   & 0      &  q-3    \\

2430:   0    & 0       &  1     & 2-q    & q_{1,2}& 0      &  q-3    \\

2431:  -1    & q-2     &  q-2   & -1     & 0      & 2      &  0      \\

2432:   0    & 0       &  -1    & q-2    & q-1    & 0      &  2

2433:     \end{array} \right )

2434: \label{Thc54}

2435: \eeq

2436: %

2437: Factorizations of characteristic polynomials are given in Table

2438: \ref{hcfactors}. 

2439: 

2440: 

2441: \section{Some Algebraic Properties of Evaluations of Transfer Matrices}

2442: 

2443: We have studied algebraic properties of evaluations of transfer matrices

2444: $T_{\Lambda,L_y,d}$ at special values of $q$.  We find many interesting

2445: results, and mention only a few here.  

2446: 

2447: First, from our exact results for the $T_{\Lambda,L_y,d}$ for $d=L_y-1$, we can

2448: calculate properties of powers of these matrices, in particular,

2449: when evaluated at special values of $q$. For example,

2450: %

2451: \beq

2452: [T_{sq,2,1}]^m = 2^{m-1}\left( \begin{array}{cc}

2453:    1 & 1  \\

2454:    1 & 1 \end{array} \right ) \quad {\rm for} \quad q=1, \quad n \ge 1 

2455: \label{Tsq21_q1_power}

2456: \eeq

2457: %

2458: and, for odd $m=2n+1$ and even $m=2n$, 

2459: \beq

2460: [T_{sq,2,1}]^{2n+1} = \left( \begin{array}{cc}

2461:     0 & 1  \\

2462:     1 & 0 \end{array} \right ) \quad {\rm for} \quad q=2, \quad n \ge 0 

2463: \label{Tsq21_q2_oddpower}

2464: \eeq

2465: %

2466: \beq

2467: [T_{sq,2,1}]^{2n} = \left( \begin{array}{cc}

2468:      1 & 0  \\

2469:      0 & 1 \end{array} \right ) \quad {\rm for} \quad q=2, \quad n \ge 1 

2470: \label{Tsq21_q2_evenpower}

2471: \eeq

2472: %

2473: and

2474: %

2475: \beq

2476: [T_{sq,2,1}]^m = (-1)^m 2^{m-1}\left( \begin{array}{cc}

2477:       1 & -1  \\

2478:      -1  & 1 \end{array} \right ) \quad {\rm for} \quad q=3, \quad m \ge 1 \ .

2479: \label{Tsq21_q3_power}

2480: \eeq

2481: %

2482: Two other identities are given below (where $n \ge 1$ and there is no sum on

2483: $j$):

2484: %

2485: \beq

2486: [(T_{sq,4,3})^{2n}]_{jj} = 2^{n-1}(1+2^{n-1})  \quad {\rm for} \quad q=3

2487: \eeq

2488: %

2489: %

2490: \beq

2491: [(T_{sq,6,5})^{2n}]_{jj} = \frac{1}{3}\Bigl [ 2^{2n-1} + 3^n + 1 \Bigr ]

2492: \quad {\rm for} \quad q=3 \ . 

2493: \eeq

2494: %

2495: 

2496: We next consider transfer matrices for strips of the triangular lattice. 

2497: In general, from eq. (\ref{detTtriLLminus1}), it follows that

2498: the evaluation of $T_{tri,L_y,L_y-1}$ at $q=2$ is noninvertible and the

2499: evaluation of this matrix at $q=3$ is noninvertible if $L_y \ge

2500: 3$. First, we observe that 

2501: %

2502: \beq

2503: [T_{tri,2,1}]^m = \left( \begin{array}{cc}

2504:     1 & 0  \\

2505:     1 & 0 \end{array} \right ) \quad {\rm for} \quad q=2 \ . 

2506: \label{Ttri21_q2}

2507: \eeq

2508: %

2509: The matrix $T_{tri,2,1}$ at $q=3$ and its powers form the cyclic group of order

2510: 3, \ ${\mathbb Z}_3$, with $[T_{tri,2,1}]^{m+3}=[T_{tri,2,1}]^m$ and 

2511: %

2512: \beqs

2513: & & 

2514: T_{tri,2,1} = \left( \begin{array}{cc}

2515:     0 & -1  \\

2516:     1 & -1 \end{array} \right ) \ ,  \quad 

2517: [T_{tri,2,1}]^2 = \left( \begin{array}{cc}

2518:    -1 & 1  \\

2519:    -1 & 0 \end{array} \right ) \ , \quad

2520: [T_{tri,2,1}]^3 = \left( \begin{array}{cc}

2521:     1 & 0  \\

2522:     0 & 1 \end{array} \right ) \cr\cr

2523: & & \quad {\rm for} \quad q=3

2524: \label{Ttri21_q3}

2525: \eeqs

2526: %

2527: Given that the transfer matrices are real, if such a matrix satisfies

2528: $(T_{\Lambda,L_y,d})^p=I_k$, where $I_k$ denotes the $k \times k$ identity

2529: matrix, this implies that $det(T_{\Lambda,L_y,d})= \pm 1$, and if $p$ is odd,

2530: as is the case here, with $p=3$, then $det(T_{\Lambda,L_y,d})=1$. Referring to

2531: our general result (\ref{detTtriLLminus1}), we see that this condition is,

2532: indeed, met with $k=2$ if $L_y=2$, $d=1$, and $q=3$.

2533: 

2534: The matrix $T_{tri,3,0}$ evaluated at $q=2$ satisfies the relation (denoting

2535: this by $T$ for short) $T^3=T$, so, together with the $2 \times 2$ identity

2536: matrix $I_2$, $(T_{tri,3,0})_{q=2}$, $[(T_{tri,3,0})_{q=2}]^2$ form a

2537: multiplicative semigroup of order 3.  (Here, we recall that a semigroup

2538: satisfies the same axioms as a group but without the axiom that inverses of

2539: elements must exist; this is also called a semigroup with unit or monoid

2540: \cite{jacobson}.  The same matrix evaluated at $q=3$ is idempotent, i.e.,

2541: satisfies $T^2=T$, so that $I_2$ and $(T_{tri,3,0})_{q=3}$ form a semigroup of

2542: order 2.  One surmises that connections can be made between these cycle lengths

2543: and the dependence of the chromatic number of the cyclic strip of the 

2544: triangular lattice on its length (see appendix), but we have not pursued this. 

2545: 

2546: The matrix $T_{tri,3,1}$ evaluated at $q=3$ satisfies the relation (again

2547: denoting this by $T$ for short) $T^4=T$ so that $I_4$, $(T_{tri,3,1})_{q=3}$,

2548: $[(T_{tri,3,1})_{q=3}]^2$, and $[(T_{tri,3,1})_{q=3}]^3$ form a semigroup of

2549: order 4.  The matrix $T_{tri,3,2}$ evaluated at $q=3$ satisfies the relation

2550: $T^3=T$, and hence the elements $I_3$, $(T_{tri,3,2})_{q=3}$, and

2551: $[(T_{tri,3,2})_{q=3}]^2$ form a semigroup of order 3.  (These are not groups

2552: since $(T_{tri,3,1})_{q=3}$ and $(T_{tri,3,2})_{q=3}$ are not invertible.)  We

2553: have found a number of similar algebraic properties of powers of the

2554: $T_{\Lambda,L_y,d}$ matrices evaluated for special values of $q$.

2555: 

2556: 

2557: \section{Discussion of Determinants}

2558: 

2559: \subsection{Square-Lattice Strips} 

2560: 

2561: Here we discuss further the determinants $det(T_{\Lambda,L_y,d})$ for

2562: $\Lambda=sq,tri,hc$, starting with the square lattice .  We first consider the

2563: zeros of the determinants, restricting to the range $0 \le d \le L_y-1$ since

2564: $T_{sq,L_y,L_y}=(-1)^{L_y}$ has no zeros. From our theorems in Ref. \cite{s5},

2565: it follows that $det(T_{sq,L_y,L_y-1})$ has (i) the factor $(q-1)$ always; (ii)

2566: the factor $(q-3)$ if $L_y=0$ mod 2; (iii) the factors $(q-2)$ and $(q-4)$ if

2567: $L_y=0$ mod 3; further, all of the zeros of $det(T_{sq,L_y,L_y-1})$ lie in the

2568: interval $1 \le q \le 5$ and, as $L_y \to \infty$, they become dense in this

2569: interval.  One may also investigate the properties of the zeros of

2570: $T_{sq,L_y,d}$ for other values of $d$ than $L_y-1$. Below we list some zeros;

2571: for multiple zeros, we indicate the multiplicity in parentheses, $(m=2)$, etc.

2572: %

2573: \beq

2574: det(T_{sq,2,0})=0 \quad {\rm at} \quad q=\frac{1}{2}(3 \pm i \sqrt{3} \ ) =

2575: 1.5 \pm 0.866i

2576: \label{detTsq20z}

2577: \eeq

2578: %

2579: \beq

2580: det(T_{sq,3,0})=0 \quad {\rm at} \quad q=1, \quad 1.659 \pm 1.1615i, \quad

2581: 2.682

2582: \label{detTsq30}

2583: \eeq

2584: %

2585: \beq

2586: det(T_{sq,3,1})=0 \quad {\rm at} \quad q=1, \quad 1.229 \pm 0.712i, \quad

2587: 2\ (m=2), \quad 3.271 \pm 0.449i

2588: \label{detTsq31z}

2589: \eeq

2590: %

2591: %

2592: \beq

2593: det(T_{sq,3,2})=0 \quad {\rm at} \quad q=1, \ 2, \ 4

2594: \label{detTsq32z}

2595: \eeq

2596: %

2597: \beq

2598: det(T_{sq,4,0})=0 \ \ {\rm at} \ \  q=1, \ 1.163 \pm 0.539i, \ 1.355, \

2599: 1.730 \pm 1.331i, \ 2.437, 3, 3.210 \pm 0.6918i

2600: \label{detTsq40z}

2601: \eeq

2602: %

2603: \beqs

2604: & & det(T_{sq,4,1})=0 \quad {\rm at} \quad q=1 \ (m=3), \ 1.257 \pm 0.578i, \

2605: 1.302, \ 1.5 \pm 0.866i \ (m=2), \cr\cr

2606: & & 1.807 \pm 0.936i,  2.736, \ 3 (m=5), \ 3.4169 \pm 0.6339i

2607: \label{detTsq41z}

2608: \eeqs

2609: %

2610: \beqs

2611: & & det(T_{sq,4,2})=0 \quad {\rm at} \quad q=1 \ (m=2), \ 1.180 \pm 0.600i, \

2612: 1.327, \ 1.580 \pm 0.606i, \cr\cr

2613: & &  2.198 \pm 0.573i, \ 3 \ (m=2), \ 3.839, \ 3.959 \pm 0.294i

2614: \label{detTsq42z}

2615: \eeqs

2616: %

2617: \beq

2618: det(T_{sq,4,3})=0 \quad {\rm at} \quad q=1, \ 1.586, \ 3, \ 4.414

2619: \label{detTsq43z}

2620: \eeq

2621: %

2622: and so forth for higher $L_y$ values.  We plot these zeros in

2623: Figs. \ref{zerodetsqpxy4}-\ref{zerodetsqpxy5}. (The figures are formatted such

2624: that, to save space, the vertical axis is shown intersecting the horizontal

2625: axis slightly to the left of $q=1$.) 

2626: 

2627: We observe that all of the zeros of $det(T_{sq,L_y,d})$ that we have calculated

2628: satisfy the condition $|q-3| \le 2$.  As is evident in eq. (\ref{detThc20z}),

2629: some zeros of $det(T_{hc,L_y,d})$ have real parts that lie outside the interval

2630: $1 \le Re(q) \le 5$.  It would be of interest to prove a theorem bounding the

2631: complex zeros of the determinants analogous to results that have been obtained

2632: for chromatic zeros, such as the theorem in Ref. \cite{sokalzeros}.  We also

2633: note that for $d \ne 0$ (as well as $d \ne L$), for the cases that we have

2634: calculated, $det(T_{sq,L_y,d})$ has the factor $(q-1)$. Although

2635: $det(T_{sq,2,0})=T_{sq,2,0}$ does not have this factor, we observe that for the

2636: cases we have calculated with $L_y \ge 3$, $det(T_{sq,L_y,0})$ contains the

2637: factor $(q-1)$.

2638: 

2639: 

2640: \begin{figure}[hbtp]

2641: \centering

2642: \leavevmode

2643: \epsfxsize=4.0in

2644: \begin{center}

2645: \leavevmode

2646: \epsffile{zerodetsqpxy4.eps}

2647: \end{center}

2648: \caption{\footnotesize{Zeros of $det(T_{sq,L})$ for $L=4$ in the complex $q$ 

2649: plane.}}

2650: \label{zerodetsqpxy4}

2651: \end{figure}

2652: 

2653: \begin{figure}[hbtp]

2654: \centering

2655: \leavevmode

2656: \epsfxsize=4.0in

2657: \begin{center}

2658: \leavevmode

2659: \epsffile{zerodetsqpxy5.eps}

2660: \end{center}

2661: \caption{\footnotesize{Zeros of $det(T_{sq,L})$ for $L=5$ in the complex $q$ 

2662: plane.}}

2663: \label{zerodetsqpxy5}

2664: \end{figure}

2665: 

2666: 

2667: \subsection{Triangular-Lattice Strips}

2668: 

2669: An interesting finding is that the determinants of

2670: $T_{tri,L_y,d}$ are the simplest among the regular homopolygonal lattices,

2671: square, triangular, and honeycomb.  Restricting to the range $0 \le d \le

2672: L_y-1$ to exclude $T_{tri,L_y,L_y}=(-1)^{L_y}$, which has no zeros, we observe

2673: that all of the results that we have obtained are consistent with the following

2674: formula, which we conjecture to hold for general $L_y$ (over the range 

2675: $L_y \ge 2$ where the triangular lattice strip is defined):

2676: %

2677: \beq

2678: det(T_{tri,L_y,d})= \eta (q-2)^a(q-3)^b

2679: \label{detTtrilyd}

2680: \eeq

2681: %

2682: where $\eta=\pm 1$ and 

2683: %

2684: \beq

2685: a= 2 n_P(tri,,L_y-1,d) + (L_y-2) n_P(tri,L_y-2,d)

2686: \label{aform}

2687: \eeq

2688: %

2689: and

2690: %

2691: \beq

2692: b=(L_y-2) n_P(tri,L_y-1,d) \ . 

2693: \label{bform}

2694: \eeq

2695: %

2696: Here the sign $\eta$ depends on $L_y$ and $d$, and we take $n_P(tri,1,d) \equiv

2697: n_P(sq,1,d)$, in accord with the general equality

2698: $n_P(tri,L_y,d)=n_P(sq,L_y,d)$ \cite{cf}.  (It is not necessary to give a

2699: formal definition of $n_P(tri,0,d)$ because the only time it appears in the

2700: equation for $a$, for the lowest nontrivial case $L_y=2$, it is multiplied by

2701: zero.)  Some structural properties of this conjecture are indirectly motivated

2702: by the exact formula that we have found in Ref. \cite{zt} for

2703: $det(T_{Z,tri,L_y,d})$, where $T_{Z,tri,L_y,d}$ is the transfer matrix, in the

2704: degree-$d$ subspace, for the full Potts model partition function on the given

2705: strip.  However, aside from the factors of $(v+1)$ in $T_{Z,tri,L_y,d}$ that

2706: cause it to vanish in the special case $v=-1$ corresponding to the chromatic

2707: polynomial, the rest of the expression that we have established for

2708: $det(T_{Z,tri,L_y,d})$ does not involve factors that reduce to powers of

2709: $(q-2)$ or $(q-3)$ when $v=-1$; instead, it involves powers of $(q+v)$ and $v$,

2710: which reduce to $(q-1)$ and $\pm 1$.  The main structural feature of our known

2711: result for $det(T_{Z,tri,L_y,d})$ that influenced our present conjecture is

2712: that the powers of these factors involve quantities such as $n_Z(L_y,d)$ and

2713: $n_Z(L_y-1,d)$, where $n_Z(L_y,d)$ is the dimension of $T_{Z,\Lambda,L_y,d}$

2714: (which is the same for $\Lambda=sq,tri,hc$) \cite{cf,hca}.  This led us to

2715: construct and test conjectured forms for the chromatic polynomial

2716: $det(T_{tri,L_y,d})$ that involve powers of the analogous dimensions

2717: $n_P(tri,L_y,d)$.

2718: 

2719:  The conjecture (\ref{detTtrilyd}) implies that the zeros of these determinants

2720: $det(T_{tri,L_y,d})$ occur at the two points $q=3$ and $q=2$ (for $0 \le d \le

2721: L_y-1$), in accordance with all of our explicit calculations.  It is

2722: interesting to observe that these are, respectively, the chromatic number

2723: $\chi(tri)$ and $\chi(tri)-1$ of the two-dimensional lattice.  For the special

2724: case $d=L_y-1$, eq. (\ref{detTtrilyd}) (with the plus sign) reduces to our

2725: previously proved result (\ref{detTtriLLminus1}).  For $L_y \ge 2$ and

2726: $d=L_y-2$, eq. (\ref{detTtrilyd}) can be written explicitly as

2727: %

2728: \beq

2729: det(T_{tri,L_y,L_y-2}) = \eta (q-2)^{3L_y-4}(q-3)^{(L_y-1)(L_y-2)}

2730: \label{detTtrilylyminus2}

2731: \eeq

2732: %

2733: Starting from the other end of the range of $d$, for $d=0,1$ (and $L_y \ge 2$),

2734: eq. (\ref{detTtrilyd}) can be written explicitly in terms of the Motzkin

2735: numbers $M_n$ defined in eq. (\ref{motzkin}) as 

2736: %

2737: \beq

2738: det(T_{tri,L_y,0})= \eta (q-2)^{2M_{L_y-2}+(L_y-2)M_{L_y-3}}

2739: (q-3)^{(L_y-2)M_{L_y-2}}

2740: \label{detTtrily0}

2741: \eeq

2742: %

2743: \beq

2744: det(T_{tri,L_y,1})= \eta (q-2)^{2M_{L_y-1}+(L_y-2)M_{L_y-2}}

2745: (q-3)^{(L_y-2)M_{L_y-1}}

2746: \label{detTtrily1}

2747: \eeq

2748: %

2749: Similar explicit formulas can be obtained from (\ref{detTtrilyd}) for other

2750: values of $d$. 

2751: 

2752: 

2753: \subsection{Honeycomb-Lattice Strips}

2754: 

2755: In the case of the honeycomb lattice, we have

2756: %

2757: \beq

2758: det(T_{hc,2,0})=0 \quad {\rm at} \quad 0.691 \pm 0.951i, \ 1.809 \pm 0.588i

2759: \label{detThc20z}

2760: \eeq

2761: %

2762: %

2763: \beq

2764: det(T_{hc,2,1})=0 \quad {\rm at} \quad 1 \ (m=2), \ 2 \pm i

2765: \label{detThc21z}

2766: \eeq

2767: %

2768: As is evident in our results above, for $0 \le d \le 2$, the zeros of 

2769: $det(T_{hc,3,d})$ occur at the two points $q=1$ and $q=2$. 

2770: Restricting to $0 \le d \le L_y-1$ to exclude the constant $T_{hc,L_y,L_y}=1$,

2771: and to odd widths $L_y$, we observe that all of our results are consistent 

2772: with the following formula, which we conjecture to hold in general: 

2773: %

2774: \beq

2775: det(T_{hc,L_y,d})=(q-1)^s(q-2)^t  \quad {\rm for \ odd} \quad L_y 

2776: \label{detThclyd}

2777: \eeq

2778: %

2779: where

2780: %

2781: \beq

2782: s=(L_y+1) n_P(hc,L_y-1,d)

2783: \label{cform}

2784: \eeq

2785: %

2786: and

2787: %

2788: \beq

2789: t=(L_y-1) n_P(hc,L_y-2,d)

2790: \label{dform}

2791: \eeq

2792: %

2793: where we take $n_P(hc,1,d) \equiv n_P(sq,1,d)$.  The motivations for this

2794: conjecture are based in part on the exact expression that we have found 

2795: elsewhere \cite{zt} for $det(T_{Z,hc,L_y,d})$, where $T_{Z,hc,L_y,d}$ is the

2796: transfer matrix, in the degree-$d$ subspace, for the partition function of 

2797: the full Potts model on the given strip. However, we note that 

2798: our exact expression for $det(T_{Z,hc,L_y,d})$ does not have a factor that

2799: reduces to a power of $(q-2)$ for the special case $v=-1$ that defines the

2800: chromatic polynomial.  Thus, as in the case of the other lattices, the detailed

2801: structure of the determinants of $T_{Z,\Lambda,L_y,d}$ for the full partition

2802: function and $T_{P,\Lambda,L_y,d}$, denoted here simply as $T_{\Lambda,L_y,d}$,

2803: for the special case $v=-1$ are different.  This is not surprising, since 

2804: the matrix $T_{P,\Lambda,L_y,d}$ does not arise, in general, from a block

2805: decomposition of $T_{Z,\Lambda,L_y,d}$ but instead by removing zero columns 

2806: and corresponding rows of the latter matrix. It is interesting that the

2807: zeros of $det(T_{hc,L_y,d})$ occur at $q=\chi(hc)$ and $q=\chi(hc)-1$, where 

2808: $\chi(hc)=2$ is the chromatic number for the honeycomb lattice.  For the

2809: special case $d=L_y-1$, eq. (\ref{detThclyd}) agrees with the result

2810: (\ref{detThclylyminus1}) that we have proved from our general calculation of

2811: $T_{hc,L_y,L_y-1}$.  For the special case $d=L_y-2$, eq. (\ref{detThclyd}) can

2812: be written explicitly as 

2813: %

2814: \beq

2815: det(T_{hc,L_y,L_y-2})=(q-1)^{(L_y+1)(3L_y-5)/2} (q-2)^{L_y-1}

2816: \quad {\rm for \ odd} \quad L_y \ . 

2817: \label{detThclylyminus2}

2818: \eeq

2819: %

2820: Similar explicit formulas can be obtained from (\ref{detThclyd}) for other

2821: values of $d$.  Both of our conjectured general formulas for

2822: $det(T_{\Lambda,L_y,d})$ have the form

2823: %

2824: \beq

2825: det(T_{\Lambda,L_y,d}) = \eta \biggl [q-(\chi(\Lambda)-1) \biggr ]^{p_1}

2826: \biggl [q-\chi(\Lambda) \biggr ]^{p_2}

2827:  \quad {\rm for} \ \ \Lambda=tri,hc

2828: \label{detform}

2829: \eeq

2830: %

2831: where $\chi(\Lambda)$ is the chromatic number of the respective two-dimensional

2832: lattices $\Lambda=tri,hc$; $\eta=\pm 1$ for $\Lambda=tri$ and $\eta=1$ for

2833: $\Lambda=hc$; $L_y \ge 3$ is odd for the honeycomb-lattice strips; and the

2834: powers $p_1$ and $p_2$ have been given in (\ref{aform}), (\ref{bform}),

2835: (\ref{cform}), and (\ref{dform}).

2836: 

2837: 

2838: \section{Accumulation Locus of Chromatic Zeros}

2839: 

2840: Although the locus ${\cal B}$ for a strip with a given width $L_y$ and free

2841: longitudinal boundary conditions (and free or periodic transverse boundary

2842: conditions) is different from that for the corresponding strip with periodic or

2843: twisted periodic longitudinal boundary conditions, one notices that in some

2844: ways these loci become more similar as $L_y$ increases.  One expects, for

2845: example, that in the limit $L_y \to \infty$, with any of these boundary

2846: conditions, one will obtain the same value of $q_c$ for the ${\cal B}$ for the

2847: Potts model on the resultant infinite 2D lattice.  From our new exact

2848: calculations, we find the following results. First, for the $5 \times \infty$

2849: cyclic/M\"obius strip of the triangular lattice, ${\cal B}$ crosses the real

2850: $q$ axis at $q=0,2,3$, and the maximal point

2851: %

2852: \beq

2853: q_c=3.33245.. \ , \quad \Lambda=tri, \quad L_y \times L_x = 5 \times \infty,

2854: \quad {\rm BC} \ = \ {\rm cyclic/Mobius} \ . 

2855: \label{qc_tpxy5}

2856: \eeq

2857: %

2858:  From our new calculations for the $L_y=4$ and $L_y=5$ cyclic/M\"obius strips

2859: of the honeycomb lattice and our analysis of the locus ${\cal B}$ in the $L_x

2860: \to \infty$ limit, we find that in both cases

2861: ${\cal B}$ crosses the real $q$ axis at the points $q=0$ and $q=2$, and at

2862: the maximal crossing points

2863: %

2864: \beq

2865: q_c=2.15476..  \ , \quad \Lambda=hc, \quad L_y \times L_x = 4 \times \infty,

2866: \quad {\rm BC} \ = \ {\rm cyclic/Mobius}

2867: \label{qc_hpxy4}

2868: \eeq

2869: %

2870: \beq

2871: q_c=2.26407...\ ,  \quad \Lambda=hc, \quad L_y \times L_x = 5 \times \infty,

2872: \quad {\rm BC} \ = \ {\rm cyclic/Mobius} \ . 

2873: \label{qc_hpxy5}

2874: \eeq

2875: %

2876: In Table \ref{qctable} we present a summary of these new results, together with

2877: the results that we have obtained for smaller values of $L_y$ in earlier work.

2878: Our present calculations are in agreement with our previous conjecture, based

2879: on all of the exact results that we had obtained, that for the infinite-length

2880: limit of a cyclic or M\"obius strip graph of a given lattice $\Lambda$, with a

2881: given width $L_y$, $q_c$ is a nondecreasing function of $L_y$.  For each type

2882: of lattice, the values of $q_c$ increase toward the values for the infinite 2D

2883: lattices, $q_c(sq)=3$ \cite{lenard}, $q_c(tri)=4$ \cite{baxter87}, and,

2884: formally, $q_c(hc)=(3+\sqrt{5})/2 \simeq 2.618$ \cite{ssbounds,p3afhc}.

2885: Although the values of $q_c$ for each of these types of lattice strips are less

2886: than the value of $q_c$ for the respective two-dimensional lattices, we have

2887: shown previously that this need not necessarily be the case.  For example, for

2888: (the $L_x \to \infty$ limit of ) the $L_y=3$ strip of the square lattice with

2889: toroidal boundary conditions \cite{tk}, $q_c=3$, equal to the value for the

2890: square lattice, and, indeed, for the cyclic self-dual strip of the square

2891: lattice, $q_c=3$ for each of the widths for which we obtained exact

2892: calculations \cite{jz,dg,sdg}.  Nevertheless, the observed monotonically

2893: nondecreasing behavior of the $q_c$ values provides a nice interpolation

2894: between the value $q_c=2$ for the one-dimensional Potts antiferromagnet and the

2895: values on the respective two-dimensional lattices.  As noted before, in

2896: general, if one uses strips with free longitudinal boundary conditions, the

2897: locus ${\cal B}$ does not necessarily cross the real $q$ axis \cite{strip}, and

2898: in cases where it does not, there is, strictly speaking, no $q_c(\{G \})$. In

2899: these cases, the way that one gains information is different; one examines the

2900: endpoints of the rightmost arcs on ${\cal B}$ that are closest to the real

2901: axis.  These are found to exhibit the general tendency to move gradually to the

2902: right with endpoints that move closer to the real axis as the strip width is

2903: increased \cite{strip,strip2,bcc,sqtran,cyltran,tritran,s4,s5,t}.  The results

2904: are consistent with the inference that as $L_y \to \infty$, the arcs will merge

2905: to form closed curves on ${\cal B}$ and the right-most part of ${\cal B}$ will

2906: cross the real axis at the value of $q_c$ for the corresponding two-dimensional

2907: lattice. (One can attempt to carry out a similar type of study for

2908: three-dimensional lattices, and exploratory work has been done in \cite{hd}.)

2909: 

2910: In Figs. \ref{tpxy5zeros}-\ref{hcpxy5zeros} we show the singular loci ${\cal

2911: B}$ for infinite-length strips of the triangular lattice with width $L_y=5$ and

2912: of the honeycomb lattice with widths $L_y=3,4,5$, together with zeros of the

2913: chromatic polynomials (chromatic zeros) for long finite-length strips of each

2914: type.  We first discuss the triangular-lattice strip (Fig. \ref{tpxy5zeros}).

2915: This should be compared with our earlier results for the $L_y=2$ strip in

2916: Fig. 3 in Ref. \cite{wcy} and for the $L_y=3$ and $L_y=4$ strips in Figs. 2 and

2917: 3 of Ref. \cite{t}.  As expected, the comparison is particularly close with the

2918: $L_y=4$ case.  For $L_y=3,4$ we found that there are two inner curves that run

2919: between points on the outer envelope and cross the real $q$ axis at $q=2$ and

2920: $q=3$, separating the interior of the envelope curve into regions that include

2921: the real segments (i) $0 \le q \le 2$, (ii) $2 \le q \le 3$, and (iii) $3 \le q

2922: \le q_c$.  We also found that the densities of zeros on these curves passing

2923: through $q=2$ and $q=3$ were slightly less than the densities of the zeros on

2924: the outer envelope curves.  Another feature that we found was the existence of

2925: two ``bubble'' regions protruding to the right from the main curve that passes

2926: through $q_c$. Finally, we observed that as $L_y$ increases from $L_y=3$ to

2927: $L_y=4$, there are more zeros, and more support for a part of ${\cal B}$ in the

2928: half-plane with $Re(q) < 0$. Our current results for $L_y=5$ exhibit these same

2929: features, and show an increase, relative to our $L_y=4$ results, in the portion

2930: of ${\cal B}$ extending into the half-plane with $Re(q) < 0$.  This is

2931: indicated in Table \ref{qctable}.  In this table we also characterize the form

2932: of the curve ${\cal B}$ as it passes through the point $q=0$; given that it

2933: passes vertically through this point, and is invariant under complex

2934: conjugation, there are two possibilities, viz., that it extends into the

2935: half-plane with (i) $Re(q) > 0$ or (ii) $Re(q) < 0$.  We denote these two

2936: possibilities as (i) convex and (ii) concave to the left.  Note that, {\it a

2937: priori}, the conditions that ${\cal B}$ has support in the half-plane with

2938: $Re(q) < 0$ and that it is concave to the left are not equivalent since it

2939: might be convex to the left, i.e., curve into the $Re(q) > 0$ half-plane in the

2940: neighborhood of the origin, but then curve back to the left at larger values of

2941: $|Im(q)|$ and cross over into the $Re(q) < 0$ half-plane.  However, we find

2942: that this does not happen for any of the strips for which we have done

2943: calculations, as is evident in Table \ref{qctable}.  Thus, for these strips,

2944: the properties that ${\cal B}$ has support for $Re(q) < 0$ and that it is

2945: concave to the left at $q=0$ are observed to occur together.  We also note that

2946: the zeros on the right-hand side of the locus ${\cal B}$ cross the real axis at

2947: a point consistent with the asymptotic result for $q_c$ given in

2948: eq. (\ref{qc_tpxy5}).  We observe small complex-conjugate bubble regions around

2949: $q = 2.6 \pm 2i$, where the curve passing through $q=2$ intersects the outer

2950: curve on ${\cal B}$.  We should also remark on the parts of the locus ${\cal

2951: B}$ near the points $q=1 \pm 2.1i$: without magnification, these appear to be

2952: arc endpoints, but in fact the end in small, very narrow, bubble regions.  This

2953: is consistent with our finding for the $L_x \to \infty$ limit of all strips of

2954: regular lattices with periodic longitudinal boundary conditions, that the

2955: respective loci ${\cal B}$ do not have arc endpoints, in contrast with the

2956: situation for strips with free longitudinal boundary conditions, for which the

2957: loci ${\cal B}$ generically do exhibit such arc endpoints.  In our earlier work

2958: we have found that loci ${\cal B}$ can exhibit tiny sliver regions, and we

2959: mention that for this case of the $L_y=5$ cyclic strip of the triangular

2960: lattice and the for the loci shown below for the strips of the honeycomb

2961: lattice, such tiny sliver regions, if small enough, would elude our analysis

2962: because of the finite size of the grid that we use for testing for equimodular

2963: dominant $\lambda$'s.  

2964: 

2965: In addition to the continuous accumulation set of chromatic zeros that forms

2966: ${\cal B}$, one may also investigate discrete chromatic zeros.  In general, for

2967: any graph with at least one edge, there is always a chromatic zero at $q=1$,

2968: since there is no proper coloring of such a graph with just one color.  For

2969: strips of the triangular lattice with width the $L_y=5$ and reasonably great

2970: length, we also find a chromatic zero near to the point $q=(3+\sqrt{5})/2

2971: \simeq 2.618$, which is a zero of $c^{(2)}$, where a

2972: degree $d=2$ $\lambda$ is dominant, and a chromatic zero near to the point 

2973: $q \simeq 3.2470$, a zero of $c^{(3)}$, where a degree $d=3$ $\lambda$ is

2974: dominant.

2975: 

2976: We next show the singular loci ${\cal B}$ for the infinite-length strips of the

2977: honeycomb lattice with $L_y=3,4,5$ in

2978: Figs. \ref{hcpxy3zeros}-\ref{hcpxy5zeros}, together with chromatic zeros for

2979: long finite-length strips of each respective width.  As background, we recall

2980: that in the $L_x \to \infty$ limit for the smaller widths $L_y=2$ (Ref.

2981: \cite{pg}) and $L_y=3$ (Ref. \cite{hca}), we found that six curves on ${\cal

2982: B}$, forming three branches, intersect at the point $q_c$ (which is equal to 2

2983: for those widths).  Hence, it was found that the region diagram included at

2984: least six regions (which comprised the totality of regions for $L_y=2$ and were

2985: augmented by two additional very small regions centered approximately at $q=0.5

2986: \pm 0.45i$ for $L_y=3$): (i) the outermost region $R_1$, extending infinitely

2987: far away from the origin and including the intervals $q > 2$ and $q < 0$ on the

2988: real axis; (ii) the innermost region $R_2$, which includes the interval $0 \le

2989: q \le 2$; (iii) a complex-conjugate (c.c.) pair of regions $R_3, \ R_3^*$

2990: forming upper and lower outer crescent-shaped areas adjacent to $q_c$, and (iv)

2991: the c.c. pair $R_4, \ R_4^*$ forming upper and lower inner crescent-shaped

2992: areas adjacent to $q_c$.  It was also found that the density of zeros on the

2993: inner two of the six curves passing through $q_c$ was somewhat lower than the

2994: density of the zeros on other parts of ${\cal B}$.  The locus ${\cal B}$ in the

2995: vicinity of the origin $q=0$ is concave to the left, which implies that it has

2996: support for values of $q$ in the half-plane with $Re(q) < 0$.  This concavity

2997: increases as the width increases.  In the loci ${\cal B}$ presented in

2998: Fig. \ref{hcpxy4zeros} and \ref{hcpxy5zeros}, we see that the point $q=2$ is an

2999: intersection point of four rather than six curves, and this feature is

3000: correlated with the fact that for these widths, $q_c$ exceeds 2.  Thus, in the

3001: plot for $L_y=4$ we see a small self-conjugate bubble region that includes the

3002: interval $2 \le q \lsim q_{hc4c}$, where $q_{hc4c} \simeq 2.155$ is the value

3003: of $q_c$ for this strip, given in eq. (\ref{qc_hpxy4}).  For $L_y=5$ we observe

3004: three small self-conjugate regions in ${\cal B}$ (see Fig. \ref{hcpxy5zeros})

3005: that include the three segments $2 \le q \le q_{hc5a}$, $q_{hc5a} \le q \le

3006: q_{hc5b}$, and $q_{hc5b} \le q \le q_{hc5c}$, where $q_{hc5a} \simeq 2.1997$,

3007: $q_{hc5b} \simeq 2.2468$, and $q_{hc5c} \simeq 2.26407$ is the value of $q_c$

3008: for this strip, given in eq. (\ref{qc_hpxy5}).  Thus, the locus ${\cal B}$ for

3009: the cyclic $L_y=5$ strip of the honeycomb lattice crosses the real $q$ axis

3010: $q=0, \ 2, \ q_{hc5a}, \ q_{hc5b}$, and $q_{hc5c}$.  These values are listed,

3011: together with those for other strips, in Table \ref{qctable}.  For the $L_y=3$

3012: strip, one sees in Fig. \ref{hcpxy3zeros} a complex-conjugate pair of narrow

3013: regions around $q \simeq 0.5 \pm 1.4i$; similarly, for the $L_y=4$ strip, one

3014: sees in Fig. \ref{hcpxy4zeros} a complex-conjugate pair of very narrow sliver

3015: regions around $q \simeq 0.1 \pm 1.3i$.  For the $L_y=5$ strip of the honeycomb

3016: lattice, we find arcs ending in tiny bubble regions at approximately $q = -0.1

3017: \pm i$. A general feature that we observe is increasing complexity of the loci

3018: ${\cal B}$ with increasing strip width.  As before, we note that regions that

3019: are extremely small could be missed by our grid used in testing for

3020: degeneracies between dominant eigenvalues that yield components of the locus

3021: ${\cal B}$. 

3022: 

3023: 

3024: \begin{figure}[hbtp]

3025: \centering

3026: \leavevmode

3027: \epsfxsize=4.0in

3028: \begin{center}

3029: \leavevmode

3030: \epsffile{tpxy5.eps}

3031: \end{center}

3032: \caption{\footnotesize{Singular locus ${\cal B}$ in the complex $q$ plane for

3033: the $L_x \to \infty$ limit of the cyclic strip of the triangular lattice with

3034: width $L_y=5$.  For comparison, the plot also shows zeros of $P(tri,L_y \times

3035: L_x,cyc.,q)$ for $L_y=5$ and a typical large value of the length, $L_x=20$.}}

3036: \label{tpxy5zeros}

3037: \end{figure}

3038: 

3039: 

3040: 

3041: \begin{figure}[hbtp]

3042: \centering

3043: \leavevmode

3044: \epsfxsize=4.0in

3045: \begin{center}

3046: \leavevmode

3047: \epsffile{hcpxy3.ps}

3048: \end{center}

3049: \caption{\footnotesize{Singular locus ${\cal B}$ in the complex $q$ plane for

3050: the $L_x \to \infty$ limit of the cyclic strip of the honeycomb lattice with

3051: width $L_y=3$.  For comparison, the plot also shows zeros of $P(tri,L_y \times

3052: L_x,cyc.,q)$ with $L_y=3$ and a typical large value of the length, $L_x=40$.}}

3053: \label{hcpxy3zeros}

3054: \end{figure}

3055: 

3056: 

3057: \begin{figure}[hbtp]

3058: \centering

3059: \leavevmode

3060: \epsfxsize=4.0in

3061: \begin{center}

3062: \leavevmode

3063: \epsffile{hcpxy4.eps}

3064: \end{center}

3065: \caption{\footnotesize{Singular locus ${\cal B}$ in the complex $q$ plane for

3066: the $L_x \to \infty$ limit of the cyclic strip of the honeycomb lattice with

3067: width $L_y=4$.  For comparison, the plot also shows zeros of $P(tri,L_y \times

3068: L_x,cyc.,q)$ with $L_y=4$ and a typical large value of the length, $L_x=24$.}}

3069: \label{hcpxy4zeros}

3070: \end{figure}

3071: 

3072: \begin{figure}[hbtp]

3073: \centering

3074: \leavevmode

3075: \epsfxsize=4.0in

3076: \begin{center}

3077: \leavevmode

3078: \epsffile{hcpxy5.eps}

3079: \end{center}

3080: \caption{\footnotesize{Singular locus ${\cal B}$ in the complex $q$ plane for

3081: the $L_x \to \infty$ limit of the cyclic strip of the honeycomb lattice with

3082: width $L_y=5$.  For comparison, the plot also shows zeros of $P(tri,L_y \times

3083: L_x,cyc.,q)$ with $L_y=5$ and a typical large value of the length, $L_x=20$.}}

3084: \label{hcpxy5zeros}

3085: \end{figure}

3086: 

3087: \bigskip

3088: 

3089: Acknowledgment

3090: 

3091: We thank J. Salas for discussions on transfer matrix methods during the work

3092: for Refs. \cite{ts,tt}, A. Sokal for related discussions, and N. Biggs for

3093: discussion on sieve methods.  The research of R.S. was partially supported by

3094: the NSF grant PHY-00-98527.  The research of S.C.C. was partially supported by

3095: the Nishina and Inoue Foundations, and he thanks Prof. M. Suzuki for further

3096: support.  The NCTS Taipei address for S.C.C. applies after April 12, 2004.

3097: 

3098: \newpage

3099: 

3100: \section{Appendix 1: Chromatic Numbers for Lattice Strips}

3101: 

3102: The following chromatic numbers apply for lengths that are greater than the

3103: lowest few where strips sometimes degenerate.  

3104: For cyclic strips of the square lattice, the chromatic number is given by 

3105: %

3106: \beq

3107: \chi(sq, L_y \times L_x,cyc.) = \cases{ 2 & for even $L_x$ \cr

3108:                3 & for odd $L_x$ \cr }

3109: \label{chi_sq_cyc}

3110: \eeq

3111: %

3112: independent of $L_y$.  For the M\"obius strips of the square lattice (e.g.,

3113: \cite{s4}) 

3114: %

3115: \beq

3116: \chi(sq, L_y \times L_x,Mb.) = \cases{ 2 & for $(L_x,L_y)=(e,o), \ (o,e)$ \cr

3117:                3 & for $(L_x,L_y)=(e,e), \ (o,o)$ \cr }

3118: \label{chi_sq_mb}

3119: \eeq

3120: %

3121: where $e$ and $o$ denote even and odd. 

3122: 

3123: 

3124: The cyclic and M\"obius strips of the triangular lattice have chromatic numbers

3125: (e.g. \cite{t}) 

3126: 

3127: \beq

3128: \chi(tri, L_y \times L_x, cyc.) = \cases{ 3 & if $L_x=0$ \

3129: mod 3 \cr 4 & if $L_x=1$ \ or $L_x =2$ mod 3 \cr }

3130: \label{chi_tri_cyc}

3131: \eeq

3132: %

3133: \beq

3134: \chi(tri, L_y \times L_x, Mb.) = 4

3135: \label{chi_tri_mb}

3136: \eeq

3137: %

3138: 

3139: Again for $L_x$ above the first few values ($L_x=1,2$ where strips can

3140: degenerate), the cyclic and M\"obius strips of the honeycomb lattice have

3141: chromatic numbers (e.g. \cite{hca})

3142: % 

3143: \beq 

3144: \chi(hc, L_y \times L_x,cyc.) = 2

3145: \label{chi_hc_cyc} 

3146: \eeq 

3147: % 

3148: \beq 

3149: \chi(hc, L_y \times L_x,Mb.) = 3 

3150: \label{chi_hc_mb}

3151: \eeq 

3152: %

3153: 

3154: \newpage

3155: 

3156: \section{Appendix 2: Tables}

3157: 

3158: 

3159: \begin{table}

3160: \caption{\footnotesize{Table of numbers $n_P(\Lambda,L_y,d)$ and their sums,

3161: $N_{P,\Lambda,L_y,\lambda}$ for cyclic strips of the square and triangular

3162: lattices.  Blank entries are zero.}}

3163: \begin{center}

3164: \begin{tabular}{|c|c|c|c|c|c|c|c|c|}

3165: $L_y \ \downarrow$ \ \ $d \ \rightarrow$

3166:    & 0 & 1   & 2   & 3   & 4   & 5  & 6  & $N_{P,\Lambda,L_y,\lambda}$

3167: \\ \hline\hline

3168: 1  & 1   & 1   &     &     &     &    &    & 2    \\ \hline

3169: 2  & 1   & 2   & 1   &     &     &    &    & 4    \\ \hline

3170: 3  & 2   & 4   & 3   & 1   &     &    &    & 10   \\ \hline

3171: 4  & 4   & 9   & 8   & 4   & 1   &    &    & 26   \\ \hline

3172: 5  & 9   & 21  & 21  & 13  & 5   & 1  &    & 70   \\

3173: \end{tabular}

3174: \end{center}

3175: \label{nptable}

3176: \end{table}

3177: 

3178: 

3179: \begin{table}[hbtp]

3180: \caption{\footnotesize{Table of numbers $n_P(hc,L_y,d)$ and their sums,

3181: $N_{P,hc,L_y,\lambda}$ for cyclic strips of the honeycomb lattice.

3182: Blank entries are zero.}}

3183: \begin{center}

3184: \begin{tabular}{|c|c|c|c|c|c|c|c|c|}

3185: $L_y \ \downarrow$ \ \ $d \ \rightarrow$

3186:    & 0 & 1   & 2   & 3   & 4   & 5  & 6  & $N_{P,hc,L_y,\lambda}$

3187: \\ \hline\hline

3188: 2  & 1   & 2   & 1   &     &     &    &    & 4    \\ \hline

3189: 3  & 3   & 6   & 4   & 1   &     &    &    & 14   \\ \hline

3190: 4  & 6   & 13  & 11  & 5   & 1   &    &    & 36   \\ \hline

3191: 5  & 19  & 43  & 40  & 22  & 7   & 1  &    & 132   \\

3192: \end{tabular}

3193: \end{center}

3194: \label{nphctable}

3195: \end{table}

3196: 

3197: 

3198: \begin{table}

3199: \caption{\footnotesize{Table of $\Delta n_P(sq,L_y,d)$ for strips of the square

3200: lattice. Blank entries are zero. The last entry for each value of $L_y$ is the

3201: total number of partitions with self-reflection symmetry.}}

3202: \begin{center}

3203: \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|c|c|}

3204: $L_y \ \downarrow$ \ \ $d \rightarrow$ & 0 & 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 & 

3205: 9 & 10 & $\Delta N_{P,L_y}$ \\

3206: \hline\hline 1  & 1  & 1  &    &    &    &    &   &   &   &   & &

3207: 2  \\ \hline 2  & 1  & 0  & 1  &    &    &    &   &   &   &   & &

3208: 2  \\ \hline 3  & 2  & 2  & 1  & 1  &    &    &   &   &   &   & &

3209: 6  \\ \hline 4  & 2  & 1  & 2  & 0  & 1  &    &   &   &   &   & &

3210: 6  \\ \hline 5  & 5  & 5  & 3  & 3  & 1  & 1  &   &   &   &   & &

3211: 18 \\ \hline 6  & 5  & 3  & 5  & 1  & 3  & 0  & 1 &   &   &   & &

3212: 18 \\ \hline 7  & 13 & 13 & 9  & 9  & 4  & 4  & 1 & 1 &   &   & &

3213: 54 \\ \hline 8  & 13 & 9  & 13 & 4  & 9  & 1  & 4 & 0 & 1 &   & &

3214: 54 \\ \hline 9  & 35 & 35 & 26 & 26 & 14 & 14 & 5 & 5 & 1 & 1 & &

3215: 162 \\ \hline

3216: 10 & 35 & 26 & 35 & 14 & 26 & 5 & 14 & 1 & 5 & 0 & 1 & 162 \\

3217: \end{tabular}

3218: \end{center}

3219: \label{nppmtable}

3220: \end{table}

3221: 

3222: 

3223: \begin{table}

3224: \caption{\footnotesize{Table of numbers $n_P(sq,L_y,d,\pm)$ for strips of the

3225: square lattice.  For each $L_y$ value, the entries in the first and second

3226: lines are $n_P(sq,L_y,d,+)$ and $n_P(sq,L_y,d,-)$, respectively. Blank entries

3227: are zero. The last entry for each value of $L_y$ is the total

3228: $N_{P,L_y,\lambda}$.}}

3229: \begin{center}

3230: \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|c|c|}

3231: $L_y \ \  (d,+) $

3232:    & $0,+$ & $1,+$ & $2,+$ & $3,+$ & $4,+$ & $5,+$ & $6,+$ & $7,+$ & $8,+$ &

3233:    $9,+$ & $10,+$ & \\

3234: \quad \ $(d,-) $

3235:    & $0,-$ & $1,-$ & $2,-$ & $3,-$ & $4,-$ & $5,-$ & $6,-$ & $7,-$ & $8,-$ &

3236:    $9,-$ & $10,-$

3237: & $N_{P,L_y,\lambda}$ \\ \hline\hline

3238: 2  & 1   & 1    & 1    &      &     &     &     &    &    &   &   &       \\

3239:    &     & 1    &      &      &     &     &     &    &    &   &   & 4     \\

3240:    \hline

3241: 3  & 2   & 3    & 2    & 1    &     &     &     &    &    &   &   &       \\

3242:    &     & 1    & 1    &      &     &     &     &    &    &   &   & 10    \\

3243:    \hline

3244: 4  & 3   & 5    & 5    & 2    & 1   &     &     &    &    &   &   &       \\

3245:    & 1   & 4    & 3    & 2    &     &     &     &    &    &   &   & 26    \\

3246:    \hline

3247: 5  & 7   & 13   & 12   & 8    & 3   & 1   &     &    &    &   &   &       \\

3248:    & 2   & 8    & 9    & 5    & 2   &     &     &    &    &   &   & 70    \\

3249:    \hline

3250: 6  & 13  & 27   & 30   & 20   & 11  & 3   & 1   &    &    &   &   &       \\

3251:    & 8   & 24   & 25   & 19   & 8   & 3   &     &    &    &   &   & 192   \\

3252:    \hline

3253: 7  & 32  & 70   & 77   & 61   & 34  & 15  & 4   & 1  &    &   &   &       \\

3254:    & 19  & 57   & 68   & 52   & 30  & 11  & 3   &    &    &   &   & 534   \\

3255:    \hline

3256: 8  & 70  & 166  & 199  & 163  & 106 & 49  & 19  & 4  & 1  &   &   &       \\

3257:    & 57  & 157  & 186  & 159  & 97  & 48  & 15  & 4  &    &   &   & 1500  \\

3258:    \hline

3259: 9  & 179 & 435  & 528  & 468  & 318 & 174 & 72  & 24 & 5  & 1 &   &       \\

3260:    & 144 & 400  & 502  & 442  & 304 & 160 & 67  & 19 & 4  &   &   & 4246  \\

3261:    \hline

3262: 10 & 435 & 1107 & 1405 & 1288 & 946 & 550 & 265 & 96 & 29 & 5 & 1 &       \\

3263:    & 400 & 1081 & 1307 & 1274 & 920 & 545 & 251 & 95 & 24 & 5 &   & 12092 \\

3264: \end{tabular}

3265: \end{center}

3266: \label{npctable}

3267: \end{table}

3268: 

3269: 

3270: \begin{table}

3271: \caption{\footnotesize{Table of $\Delta n_P(hc,L_y,d)$ for strips

3272: of the honeycomb lattice with even $L_y$. Blank entries are zero.

3273: The last entry for each value of $L_y$ is the total number of

3274: partitions with self-reflection symmetry.}}

3275: \begin{center}

3276: \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|c|c|}

3277: $L_y \ \ \downarrow$ \ \  $d \rightarrow$ 

3278: & 0 & 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 & 9 & 10 & $\Delta

3279: N_{P,hc,L_y}$ \\

3280: \hline\hline

3281: 2  & 1  & 0  & 1  &    &    &    &   &   &   &   & & 2  \\ \hline

3282: 4  & 4  & 3  & 3  & 1  & 1  &    &   &   &   &   & & 12 \\ \hline

3283: 6  & 7  & 4  & 7  & 1  & 4  & 0  & 1 &   &   &   & & 24 \\ \hline

3284: 8  & 36 & 30 & 30 & 17 & 17 & 6  & 6 & 1 & 1 &   & & 144 \\ \hline

3285: 10 & 66 & 47 & 66 & 23 & 47 & 7 & 23 & 1 & 7 & 0 & 1 & 288 \\

3286: \end{tabular}

3287: \end{center}

3288: \label{nphcpmtable}

3289: \end{table}

3290: 

3291: \begin{table}

3292: \caption{\footnotesize{Table of numbers $n_P(hc,L_y,d,\pm)$ for strips of the

3293: honeycomb lattice with even $L_y$.  For each $L_y$ value, the entries in the

3294: first and second lines are $n_P(hc,L_y,d,+)$ and $n_P(hc,L_y,d,-)$,

3295: respectively. Blank entries are zero. The last entry for each value of $L_y$ is

3296: the total $N_{P,hc,L_y,\lambda}$.}}

3297: \begin{center}

3298: \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|c|c|}

3299: $L_y \ (d,+) $

3300:    & $0,+$ & $1,+$ & $2,+$ & $3,+$ & $4,+$ & $5,+$ & $6,+$ & $7,+$ & $8,+$ &

3301:    $9,+$ & $10,+$ & \\

3302: \quad \ $(d,-) $

3303:    & $0,-$ & $1,-$ & $2,-$ & $3,-$ & $4,-$ & $5,-$ & $6,-$ & $7,-$ & $8,-$ &

3304:    $9,-$ & $10,-$

3305: & $N_{P,hc,L_y,\lambda}$ \\ \hline\hline

3306: 2  & 1   & 1    & 1    &      &     &     &     &    &    &   &   &       \\

3307:    &     & 1    &      &      &     &     &     &    &    &   &   & 4     \\

3308:    \hline

3309: 4  & 5   & 8    & 7    & 3    & 1   &     &     &    &    &   &   &       \\

3310:    & 1   & 5    & 4    & 2    &     &     &     &    &    &   &   & 36    \\

3311:    \hline

3312: 6  & 25  & 53   & 56   & 35   & 17  & 4   & 1   &    &    &   &   &       \\

3313:    & 18  & 49   & 49   & 34   & 13  & 4   &     &    &    &   &   & 358   \\

3314:    \hline

3315: 8  & 194 & 454  & 518  & 404  & 237 & 100 & 32  & 6  & 1  &   &   &       \\

3316:    & 158 & 424  & 488  & 387  & 220 & 94  & 26  & 5  &    &   &   & 3748  \\

3317:    \hline

3318: 10 & 1590& 4036 & 4953 & 4324 & 2947& 1565& 664 & 208& 51 & 7 & 1 &       \\

3319:    & 1524& 3989 & 4887 & 4301 & 2900& 1558& 641 & 207& 44 & 7 &   & 40404 \\

3320: \end{tabular}

3321: \end{center}

3322: \label{npchctable}

3323: \end{table}

3324: 

3325: 

3326: \begin{table}

3327: \caption{\footnotesize{Factorization properties of characteristic polynomials

3328: $CP(T_{sq,L_y,d},z)$ for cyclic strips of the square lattices. The notation

3329: $(m,n)$ means that $CP(T_{sq,L_y,d},z)$ factorizes into polynomials in $z$ of

3330: degree $m$ and $n$ with integer coefficients.  The notation $j^k$ indicates

3331: that there are $k$ factors of degree $j$.}}

3332: \begin{center}

3333: \begin{tabular}{|c|c|c|c|c|c|c|c|}

3334: $L_y \ \downarrow$ \ \ $d \ \rightarrow$

3335:    & 0       &   1    &   2        & 3        & 4  & 5 & $N_{P,sq,L_y,\lambda}$

3336: \\ \hline\hline

3337: 1  & (1)     & (1)    &            &          &       &    & 2    \\ \hline

3338: 2  & (1)     & $(1^2)$&  (1)       &          &       &    & 4    \\ \hline

3339: 3  & (2)     & (1,3)  & $(1^3)$    & (1)      &       &    & 10   \\ \hline

3340: 4  & (1,3)   & (4,5)  & $(3,5)$    & $(1^2,2)$& (1)   &    & 26   \\ \hline

3341: 5  & (2,7)   & (8,13) & (9,12)     & (5,8)    & $(2,3)$ &(1) & 70   \\

3342: \end{tabular}

3343: \end{center}

3344: \label{sqfactors}

3345: \end{table}

3346: 

3347: 

3348: \begin{table}

3349: \caption{\footnotesize{Factorization properties of characteristic polynomials

3350: $CP(T_{tri,L_y,d},z)$ for cyclic strips of the triangular lattices. The notation

3351: $(m,n)$ means that $CP(T_{tri,L_y,d},z)$ factorizes into polynomials in $z$ of

3352: degree $m$ and $n$ with integer coefficients. The notation $j^k$ indicates that

3353: there are $k$ factors of degree $j$.}}

3354: \begin{center}

3355: \begin{tabular}{|c|c|c|c|c|c|c|c|}

3356: $L_y \ \downarrow$ \ \ $d \ \rightarrow$

3357:    & 0       &   1    &   2     & 3      & 4     & 5  & $N_{P,tri,L_y,\lambda}$

3358: \\ \hline\hline

3359: 1  & (1)     & (1)    &         &        &       &     &   2   \\ \hline

3360: 2  & (1)     & (2)    & (1)     &        &       &     &   4   \\ \hline

3361: 3  & (2)     & (1,3)  & (1,2)   & (1)    &       &     &  10   \\ \hline

3362: 4  & (4)     & (9)    & (8)     & (4)    & (1)   &     &  26   \\ \hline

3363: 5  & (2,7)   & (8,13) & (9,12) & (5,8)  & (2,3)  & (1) &  70   \\

3364: \end{tabular}

3365: \end{center}

3366: \label{trifactors}

3367: \end{table}

3368: 

3369: 

3370: \begin{table}

3371: \caption{\footnotesize{Factorization properties of characteristic polynomials

3372: $CP(T_{hc,L_y,d},z)$ for cyclic strips of the honeycomb lattices. The notation

3373: $(m,n)$ means that the characteristic polynomial factorizes into polynomials in

3374: $z$ of degree $m$ and $n$ with integer coefficients. The notation $j^k$

3375: indicates that there are $k$ factors of degree $j$.}}

3376: \begin{center}

3377: \begin{tabular}{|c|c|c|c|c|c|c|c|}

3378: $L_y \ \downarrow$ \ \ $d \ \rightarrow$

3379:    & 0       &   1    &   2     & 3      & 4    & 5  & $N_{P,hc,L_y,\lambda}$

3380: \\ \hline\hline

3381: 2  & (1)     &$(1^2)$ & (1)     &        &       &     &   4    \\ \hline

3382: 3  & (3)     & (6  )  & (1,3 )  & (1)    &       &     &  14    \\ \hline

3383: 4  & (1,5)   & (5,8)  & (4,7)  &$(1^2,3)$& (1)   &     &  36    \\ \hline

3384: 5  & (19)    & (43)   & (40)   & (22)    & (1,6) &  1  &  132   \\ 

3385: \end{tabular}

3386: \end{center}

3387: \label{hcfactors}

3388: \end{table}

3389: 

3390: \begin{table}

3391: \caption{\footnotesize{Properties of the locus ${\cal B}$ for infinite-length

3392: cyclic/M\"obius strips with width $L_y$ of regular lattices $\Lambda$,

3393: including square (sq), triangular (tri),and honeycomb (hc) lattices.  The

3394: notation BCR denotes ${\cal B}$ crossings on the real $q$ axis, the greatest of

3395: which is $q_c$.  The notation SN indicates whether ${\cal B}$ has some support

3396: for values of $q$ with $Re(q) < 0$, marked as y (yes) or n (no).  The behavior 

3397: of ${\cal B}$ in the neighborhood of the origin is indicated as convex 

3398: or concave to the left, with the convention taken such that, for

3399: example, the circle $|q-1|=1$ is convex (to the left) at $q=0$.  The

3400: final column gives the reference.}}

3401: \begin{center}

3402: \begin{tabular}{|c|c|c|c|c|c|}

3403: $\Lambda$ & $L_y$ & BCR & SN & ${\cal B}$ at $q=0$ & Ref. \\

3404: \hline\hline

3405: sq  & 1 & 2,          \ 0  & n & convex  & $-$      \\ \hline

3406: sq  & 2 & 2,          \ 0  & n & convex  & \protect{\cite{w}}     \\ \hline

3407: sq  & 3 & 2.34,  \ 2, \ 0  & y & concave & \protect{\cite{wcyl}}  \\ \hline

3408: sq  & 4 & 2.49,  \ 2, \ 0  & y & concave & \protect{\cite{s4}}    \\ \hline

3409: sq  & 5 & 2.58,  \ 2, \ 0  & y & concave & \protect{\cite{s5}}  \\ \hline\hline

3410: tri & 2 & 3,     \ 2, \ 0  & n & convex  & \protect{\cite{wcy}}   \\ \hline

3411: tri & 3 & 3,     \ 2, \ 0  & n & convex  & \protect{\cite{t}}     \\ \hline

3412: tri & 4 & 3.23, \ 3,\ 2,\ 0 &y & concave &  \protect{\cite{t}}    \\ \hline

3413: tri & 5 & 3.33, \ 3,\ 2,\ 0 &y & concave &   here              \\ \hline\hline

3414: hc  & 2 & 2,    \ 0        & y & concave & \protect{\cite{pg}}    \\ \hline

3415: hc  & 3 & 2,    \ 0        & y & concave & \protect{\cite{hca}}   \\ \hline

3416: hc  & 4 & 2.155,\ 2,\ 0    & y & concave & here                  \\ \hline

3417: hc  & 5 & 2.26, \ 2.25, \ 2.20, \ 2, \ 0 & y & concave & here    \\ 

3418: 

3419: \end{tabular}

3420: \end{center}

3421: \label{qctable}

3422: \end{table}

3423: 

3424: \newpage

3425: 

3426: \section{Appendix 3}

3427: 

3428: 

3429: Here we list polynomials that occur in our explicit expressions for elements of

3430: transfer matrices.  It is convenient to define

3431: %

3432: \beq

3433: F_{m,n} = D_m - D_n

3434: \label{fmn}

3435: \eeq

3436: %

3437: One then sees that $q^2-4q+5$ is $F_{4,3}$.  Note that

3438: %

3439: \beq

3440: F_{m+2,m}=(q-1)^{m-1}(q-2)

3441: \eeq

3442: %

3443: \beq

3444: F_{m+4,m}=(q-1)^{m-1}(q-2)(q^2-2q+2)

3445: \eeq

3446: %

3447: \beqs

3448: F_{m+6,m} & = & (q-1)^{m-1}(q-2)(q^2-q+1)D_4 \cr\cr

3449: & = & (q-1)^{m-1}(q-2)(q^2-q+1)(q^2-3q+3)

3450: \eeqs

3451: %

3452: etc., so e.g., $F_{4,2}=(q-1)(q-2)$.  One can also define

3453: %

3454: \beq

3455: G_{m,n} = D_m - 2D_n

3456: \label{gmn}

3457: \eeq

3458: %

3459: so that another frequently occurring entry, $q^2-5q+7$ is seen to be

3460: $G_{4,3}$.  We have not tried to carry out such constructions in an exhaustive

3461: manner and often just give shorthand notation to commonly occurring

3462: polynomials. Because of the various equivalences, there are also several ways

3463: of writing a given polynomial.

3464: 

3465: 

3466: We also list a set of polynomials that recur frequently and are given shorthand

3467: names, $p_n$, $r_n$, and $s_n$ for quadratic, cubic, and quartic terms: 

3468: %

3469: \beq

3470: p_4=q^2-3q+4, \quad p_6=q^2-4q+6, \quad p_8=q^2-5q+8

3471: \eeq

3472: %

3473: \beq

3474: p_{10}=q^2-6q+10, \quad p_{13}=q^2-7q+13, \quad p_{14}=q^2-7q+14

3475: \eeq

3476: %

3477: \beq

3478: p_{15}=q^2-7q+15, \quad p_{17}=q^2-8q+17

3479: \eeq

3480: %

3481: 

3482: \bigskip

3483: 

3484: %

3485: \beq

3486: r_{11}=q^3-5q^2+11q-11, \quad r_{13}=q^3-6q^2+14q-13

3487: \eeq

3488: %

3489: \beq

3490: r_{19}=q^3-7q^2+19q-19, \quad r_{20}=q^3-7q^2+19q-20

3491: \eeq

3492: %

3493: \beq

3494: r_{34}=q^3-9q^2+29q-34, \quad r_{46}=q^3-11q^2+39q-46

3495: \eeq

3496: %

3497: \beq

3498: r_{47}=q^3-10q^2+36q-47, \quad r_{49}=q^3-10q^2+37q-49, 

3499: r_{58}=q^3-11q^2+43q-58

3500: \eeq

3501: %

3502: 

3503: \bigskip

3504: 

3505: %

3506: \beq

3507: s_7=q^4-5q^3+11q^2-12q+7, \quad s_{21}=q^4-7q^3+21q^2-32q+21

3508: \eeq

3509: %

3510: \beq

3511: s_{31}=q^4-8q^3+27q^2-45q+31, s_{47}=q^4-9q^3+34q^2-63q+47

3512: \eeq

3513: %

3514: \beq

3515: s_{61}=q^4-9q^3+35q^2-70q+61, s_{82}=q^4-11q^3+48q^2-99q+82

3516: \eeq

3517: %

3518: \beq

3519: s_{148}=q^4-13q^3+67q^2-160q+148, s_{173}=q^4-13q^3+68q^2-171q+173

3520: \eeq

3521: %

3522: \beq

3523: s_{3,5}=(q^2-2q+3)(q^2-4q+5), \quad t_{5,5}=(q^2-4q+5)(q^3-4q^2+6q-5)

3524: \eeq

3525: %

3526: 

3527: \newpage

3528: 

3529: \begin{thebibliography}{99}

3530: 

3531: \bibitem{potts}

3532: R. B. Potts, Proc. Camb. Phil. Soc. {\bf 48} (1952) 106.

3533: 

3534: \bibitem{wurev}

3535: F. Y. Wu, Rev. Mod. Phys. {\bf 54} (1982) 235.

3536: 

3537: \bibitem{wup}

3538: F. Y. Wu, J. Stat. Phys. {\bf 52} (1988) 99.

3539: 

3540: \bibitem{rrev}

3541: R. C. Read, J. Combin. Theory {\bf 4} (1968) 52.

3542: 

3543: \bibitem{rtrev}

3544: R. C. Read and W. T. Tutte, ``Chromatic Polynomials'',

3545: in {\it Selected Topics in Graph Theory, 3}, (Academic Press, New York, 1988),

3546: p. 15.

3547: 

3548: \bibitem{bbook}

3549: N. L. Biggs, {\it Algebraic Graph Theory} (Cambridge

3550: Univ. Press, Cambridge, 1st ed. 1974, 2nd ed. 1993).

3551: 

3552: \bibitem{saleur}

3553: H. Saleur, Nucl. Phys. B {\bf 360} (1991) 219.  

3554: 

3555: \bibitem{saleurcmp}

3556: H. Saleur, Commun. Math. Phys. {\bf 132} (1990) 657. 

3557: 

3558: \bibitem{cf}

3559: S.-C. Chang and R. Shrock, Physica A {\bf 296} (2001) 131.

3560: 

3561: \bibitem{hca}

3562: S.-C. Chang and R. Shrock, Physica A {\bf 296} (2001) 183.

3563: 

3564: \bibitem{dg}

3565: S.-C. Chang and R. Shrock, Physica A {\bf 301} (2001) 301.

3566: 

3567: \bibitem{kf}

3568: P. W. Kasteleyn and C. M. Fortuin, J. Phys. Soc. Jpn. (Suppl.) {\bf 26} (1969)

3569: 11. 

3570: 

3571: \bibitem{fk}

3572: C. M. Fortuin and P. W. Kasteleyn, Physica {\bf 57} (1972) 536.

3573: 

3574: \bibitem{bkw1}

3575: S. Beraha, J. Kahane, and N. Weiss, J. Combin. Theory B {\bf 27} (1979) 1.

3576: 

3577: \bibitem{bkw2}

3578: S. Beraha, J. Kahane, and N. Weiss, J. Combin. Theory B {\bf 28} (1980) 52.

3579: 

3580: \bibitem{read81}

3581: R. C. Read, {\it Proc. 3rd Caribbean Conference on

3582: Combinatorics and Computing} (1981) 23.

3583: 

3584: \bibitem{read91}

3585: R. C. Read and G. F. Royle, in {\it Graph Theory,

3586: Combinatorics, and Applications}, Y. Alavi et al., eds. (Wiley, NY, 1991),

3587: vol. 2, pp. 1009-1029.

3588: 

3589: \bibitem{al}

3590: M. Aizenman and E. H. Lieb, J. Stat. Phys. {\bf 24} (1981) 279.

3591: 

3592: \bibitem{cw}

3593: Y. Chow and F. Y. Wu, Phys. Rev. B {\bf 36} (1987) 285.

3594: 

3595: \bibitem{w}

3596: R. Shrock and S.-H. Tsai, Phys. Rev. E {\bf 55} (1997) 5165.

3597: 

3598: \bibitem{wc}

3599: R. Shrock and S.-H. Tsai, Phys. Rev. E {\bf 56} (1997) 1342.

3600: 

3601: \bibitem{wa}

3602: R. Shrock and S.-H. Tsai, Phys. Rev. E {\bf 56} (1997) 3935.

3603: 

3604: \bibitem{strip}

3605: M. Ro\v{c}ek, R. Shrock, and S.-H. Tsai, Physica A {\bf 252} (1998) 505.

3606: 

3607: \bibitem{strip2}

3608: M. Ro\v{c}ek, R. Shrock, and S.-H. Tsai, Physica A {\bf 259} (1998) 367.

3609: 

3610: \bibitem{hs}

3611: R. Shrock and S.-H. Tsai, Physica A {\bf 259} (1998) 315.

3612: 

3613: \bibitem{wa2}

3614: R. Shrock and S.-H. Tsai,  Physica A {\bf 265} (1999) 186. 

3615: 

3616: \bibitem{pg}

3617: R. Shrock and S.-H. Tsai, J. Phys. A Lett. {\bf 32} (1999) L195.

3618: 

3619: \bibitem{nec}

3620: R. Shrock and S.-H. Tsai, J. Phys. A {\bf 32} (1999) 5053.

3621: 

3622: \bibitem{wcyl}

3623: R. Shrock and S.-H. Tsai, Phys. Rev. E {\bf 60} (1999) 3512.

3624: 

3625: \bibitem{wcy}

3626: R. Shrock and S.-H. Tsai, Physica A {\bf 275} (2000) 429.

3627: 

3628: \bibitem{pm}

3629: R. Shrock, Phys. Lett. A {\bf 261} (1999) 57.

3630: 

3631: \bibitem{tk}

3632: N. L. Biggs and R. Shrock, J. Phys. A (Letts.) {\bf 32} (1999) L489.

3633: 

3634: \bibitem{k}

3635: S.-C. Chang and R. Shrock, Phys. Rev. E {\bf 62} (2000) 4650

3636: 

3637: \bibitem{bcc}

3638: R. Shrock, in {\it Proceedings of the 1999 British

3639: Combinatorial Conference, BCC99}, Discrete Math. {\bf 231} (2001) 421.

3640: 

3641: \bibitem{s4}

3642: S.-C. Chang and R. Shrock, Physica A {\bf 290} (2001) 402.

3643: 

3644: \bibitem{t}

3645: S.-C. Chang and R. Shrock, Ann. Phys. {\bf 290} (2001) 124.

3646: 

3647: \bibitem{tor4}

3648: S.-C. Chang and R. Shrock, Physica A {\bf 292} (2001) 307.

3649: 

3650: \bibitem{hd}

3651: J. Salas and R. Shrock, Phys. Rev. E {\bf 64} (2001) 011111.

3652: 

3653: \bibitem{s5}

3654: S.-C. Chang and R. Shrock, Physica A {\bf 316} (2002) 335.

3655: 

3656: \bibitem{lenard}

3657: A. Lenard (unpublished), cited in E. H. Lieb, Phys. Rev. {\bf 162} (1967) 162.

3658: 

3659: \bibitem{baxter86}

3660: R. J. Baxter, J. Phys. A {\bf 19} (1986) 2821.

3661: 

3662: \bibitem{baxter87}

3663: R. J. Baxter, J. Phys. A {\bf 20} (1987) 5241.

3664: 

3665: \bibitem{p3afhc}

3666: R. Shrock and S.-H. Tsai, J. Phys. A {\bf 30} (1997) 495.

3667: 

3668: \bibitem{ww}

3669: R. Shrock and S.-H. Tsai, Phys. Rev. E {\bf 55} (1997) 6791.

3670: 

3671: \bibitem{w3}

3672: R. Shrock and S.-H. Tsai, Phys. Rev. E {\bf 56} (1997) 2733.

3673: 

3674: \bibitem{wn}

3675: R. Shrock and S.-H. Tsai, Phys. Rev. E {\bf 56} (1997) 4111.

3676: 

3677: \bibitem{w2d}

3678: R. Shrock and S.-H. Tsai, Phys. Rev. E {\bf E58} (1998) 4332 

3679: (cond-mat/9808057). 

3680: 

3681: \bibitem{ssbounds}

3682: J. Salas and A. Sokal, J. Stat. Phys. {\bf 86} (1997) 551.

3683: 

3684: \bibitem{a}

3685: R. Shrock, Physica A {\bf 283} (2000) 388.

3686: 

3687: \bibitem{bds}

3688: N. L. Biggs, R. M. Damerell, and D. A. Sands, J. Combin. Theory

3689: B {\bf 12} (1972) 123.

3690: 

3691: \bibitem{bm}

3692: N. L. Biggs and G. H. Meredith, J. Combin. Theory B {\bf 20} (1976) 5.

3693: 

3694: \bibitem{b}

3695: N. L. Biggs, Bull. London Math. Soc. {\bf 9} (1977) 54.

3696: 

3697: \bibitem{matmeth}

3698: N. L. Biggs, J. Combin. Theory B {\bf 82} (2001) 19.

3699: 

3700: \bibitem{matmeth2}

3701: N. L. Biggs, Bull. London Math. Soc. {\bf 34} (2002) 129.

3702: 

3703: \bibitem{matmeth3}

3704: N. L. Biggs, London School of Economics Centre for Discrete and Applicable

3705: Mathematics, report LSE-CDAM-2000-04. 

3706: 

3707: \bibitem{klein} 

3708: D. Klein and W. Seitz, Studies in Physical and

3709: Theoretical Chemistry, vol. 63 (1988), pp. 155-166.

3710: 

3711: \bibitem{sqtran}

3712: J. Salas and A. Sokal, J. Stat. Phys. {\bf 104} (2001) 609. 

3713: 

3714: \bibitem{cyltran}

3715: J.Jacobsen and J. Salas, J. Stat. Phys. {\bf 104} (2001) 701.

3716: 

3717: \bibitem{tritran}

3718: J. Jacobsen, J. Salas, and A.D. Sokal, J. Stat. Phys. {\bf 112} (2003) 921.

3719: 

3720: \bibitem{ts}

3721: S.-C. Chang, J. Salas, and R. Shrock, J. Stat. Phys. {\bf 107} (2002) 1207.

3722: 

3723: \bibitem{tt}

3724: S.-C. Chang, J. Jacobsen, J. Salas, and R. Shrock,

3725: J. Stat. Phys. {\bf 114} (2004) 763.

3726: 

3727: \bibitem{dn}

3728: S.-C. Chang, Physica A {\bf 296} (2001) 495.

3729: 

3730: \bibitem{cprg}

3731: N. L. Biggs, M. Klin, and P. Reinfeld, in {\it Proc. Algebraic Graph Theory 

3732: Workshop}, European Journal of Combinatorics {\bf 25} (2004) 147. 

3733: 

3734: \bibitem{ka3}

3735: S.-C. Chang and R. Shrock, in {\it Proc. of the CRM Workshop on Tutte

3736: Polynomials, 2001}, Advances in Applied. Math. {\bf 32} (2004) 44.

3737: 

3738: \bibitem{js} J. Salas has kindly informed that he, A. Sokal, and J.  Jespersen

3739: have also generalized their transfer matrix method used for chromatic

3740: polynomials on free and cylindrical strips of the square and triangular lattice

3741: to cyclic strips of these lattices.

3742: 

3743: \bibitem{zt}

3744: S.-C. Chang and R. Shrock, cond-mat/0404524. 

3745: 

3746: \bibitem{jacobson}

3747: N. Jacobson, {\it Basic Algebra} (Freeman, San Francisco, 1974). 

3748: 

3749: \bibitem{sokalzeros}

3750: A. Sokal, Combin. Probab. Comput. {\bf 10} (2001) 41. 

3751: 

3752: \bibitem{jz}

3753: S.-C. Chang and R. Shrock,  Physica A {\bf 301} (2001) 196. 

3754: 

3755: \bibitem{sdg}

3756: S.-C. Chang and R. Shrock, Phys. Rev. E {\bf 64} (2001) 066116. 

3757: 

3758: \end{thebibliography}

3759: \vfill

3760: \eject

3761: \end{document}

3762: 

3763: