1: \documentstyle [12pt,amsfonts] {article}

2: \input epsf

3: 

4: \topmargin -0.3in

5: \textwidth  6.5in

6: \textheight 8.5in

7: 

8: \oddsidemargin 0in

9: 

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

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

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

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

14: 

15: \catcode`@=11

16: \@addtoreset{equation}{section}

17: \@addtoreset{equation}{subsection}

18: \def\theequation{\ifnum\value{section}=0 \arabic{equation}\ignorespaces

19: \else \ifnum\value{section}=-1 A.\arabic{equation}\ignorespaces

20: \else \ifnum\value{subsection}=0 \thesection.\arabic{equation}\ignorespaces

21: \else \thesection.\arabic{subsection}.\arabic{equation}\ignorespaces

22:                            \fi

23:                       \fi

24:                  \fi}

25: %\@addtoreset{table}{section}

26: %\@addtoreset{table}{subsection}

27: %\def\thetable{\ifnum\value{section}=0 \arabic{table}\ignorespaces

28: %\else \ifnum\value{subsection}=0 \thesection.\arabic{table}\ignorespaces

29: %\else \thesection.\arabic{subsection}.\arabic{table}\ignorespaces

30: %                    \fi

31: %              \fi}

32: \catcode`@=12

33: 

34: \begin{document}

35: 

36: \def\thefootnote{\fnsymbol{footnote}}

37: 

38: \baselineskip 6.0mm

39: 

40: \vspace{4mm}

41: 

42: \begin{center}

43: 

44: {\Large \bf Ground State Entropy of the Potts Antiferromagnet on Triangular 

45: Lattice Strips}

46: 

47: \vspace{8mm}

48: 

49: \setcounter{footnote}{0}

50: Shu-Chiuan Chang\footnote{email: shu-chiuan.chang@sunysb.edu} and 

51: \setcounter{footnote}{6}

52: Robert Shrock\footnote{email: robert.shrock@sunysb.edu}

53: 

54: \vspace{6mm}

55: 

56: C. N. Yang Institute for Theoretical Physics  \\

57: State University of New York       \\

58: Stony Brook, N. Y. 11794-3840  \\

59: 

60: \vspace{10mm}

61: 

62: {\bf Abstract}

63: \end{center}

64: 

65: We present exact calculations of the zero-temperature partition function

66: (chromatic polynomial) $P$ for the $q$-state Potts antiferromagnet on

67: triangular lattice strips of arbitrarily great length $L_x$ vertices and of

68: width $L_y=3$ vertices and, in the $L_x \to \infty$ limit, the exponent of the

69: ground-state entropy, $W=e^{S_0/k_B}$.  The strips considered, with their

70: boundary conditions ($BC$) are (a) $(FBC_y,PBC_x)=$ cyclic, (b)

71: $(FBC_y,TPBC_x)=$ M\"obius, (c) $(PBC_y,PBC_x)=$ toroidal, and (d)

72: $(PBC_y,TPBC_x)=$ Klein bottle, where $F$, $P$, and $TP$ denote free, periodic,

73: and twisted periodic.  Exact calculations of $P$ and $W$ are also given for

74: wider strips, including (e) cyclic, $L_y=4$, and (f) $(PBC_y,FBC_x)=$

75: cylindrical, $L_y=5,6$. Several interesting features are found, including the

76: presence of terms in $P$ proportional to $\cos(2\pi L_x/3)$ for case (c).

77: The continuous locus of points ${\cal B}$ where $W$ is nonanalytic in the $q$

78: plane is discussed for each case and a comparative discussion is given of the

79: respective loci ${\cal B}$ for families with different boundary conditions.

80: Numerical values of $W$ are given for infinite-length strips of various widths

81: and are shown to approach values for the 2D lattice rapidly.  A remark is also

82: made concerning a zero-free region for chromatic zeros. 

83: 

84: \vspace{16mm}

85: 

86: \pagestyle{empty}

87: \newpage

88: 

89: \pagestyle{plain}

90: \pagenumbering{arabic}

91: \renewcommand{\thefootnote}{\arabic{footnote}}

92: \setcounter{footnote}{0}

93: 

94: \section{Introduction} 

95: 

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

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

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

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

100: W$.  Such nonzero ground state entropy is important as an exception to the

101: third law of thermodynamics.  There is a close connection with graph theory

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

103: $q$-state Potts antiferromagnet on a graph $G$ satisfies

104: \beq

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

106: \label{zp}

107: \eeq

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

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

110: adjacent vertices have the same color (for reviews, see

111: \cite{rrev}-\cite{bbook}).  The minimum number of colors necessary for such a

112: coloring of $G$ is called the chromatic number, $\chi(G)$. 

113: Thus\footnote{\footnotesize{At certain special

114: points $q_s$ (typically $q_s=0,1,.., \chi(G)$), one has the noncommutativity of

115: limits $\lim_{q \to q_s} \lim_{n \to \infty} P(G,q)^{1/n} \ne \lim_{n \to

116: \infty} \lim_{q \to q_s}P(G,q)^{1/n}$, and hence it is necessary to specify the

117: order of the limits in the definition of $W(\{G\},q_s)$ \cite{w}.  We use the

118: first order of limits here; this has the advantage of removing certain isolated

119: discontinuities in $W$.}}

120: \beq

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

122: \label{w}

123: \eeq 

124: where $n=v(G)$ is the number of vertices of $G$ and $\{G\} = \lim_{n \to

125: \infty}G$.  Since $P(G,q)$ is a polynomial, one can generalize $q$ from

126: ${\mathbb Z}_+$ to ${\mathbb C}$.  The zeros of $P(G,q)$ in the complex $q$

127: plane are called chromatic zeros; a subset of these may form an accumulation

128: set in the $n \to \infty$ limit, denoted ${\cal B}$, which is the continuous

129: locus of points where $W(\{G\},q)$ is

130: nonanalytic.~\footnote{\footnotesize{Although it does not happen in the cases

131: considered here, for some families of graphs ${\cal B}$ may be null, and $W$

132: may also be nonanalytic at certain discrete points.}}  The maximal region in

133: the complex $q$ plane to which one can analytically continue the function

134: $W(\{G\},q)$ from physical values where there is nonzero ground state entropy

135: is denoted $R_1$.  The maximal value of $q$ where ${\cal B}$ intersects the

136: (positive) real axis is labelled $q_c(\{G\})$.  This point is important since

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

138: axis where the Potts model (with $q$ extended from ${\mathbb Z}_+$ to ${\mathbb

139: R}$) exhibits nonzero ground state entropy (which increases with $q$,

140: asymptotically approaching $S_0 = k_B \ln q$ for large $q$, and which for a

141: regular lattice $\Lambda$ can be calculated approximately via large--$q$ series

142: expansions) from the interval $0 \le q \le q_c(\{G\})$ in which $S_0$ has a

143: different analytic form. 

144: 

145: 

146: 

147: \unitlength 1.3mm

148: %cyclic triangular lattice with L_y=3

149: \begin{picture}(100,20)

150: \multiput(0,0)(10,0){5}{\circle*{2}}

151: \multiput(0,10)(10,0){5}{\circle*{2}}

152: \multiput(0,20)(10,0){5}{\circle*{2}}

153: \multiput(0,0)(10,0){5}{\line(0,1){20}}

154: \multiput(0,0)(0,10){3}{\line(1,0){40}}

155: \put(0,10){\line(1,1){10}}

156: \multiput(0,0)(10,0){3}{\line(1,1){20}}

157: \put(30,0){\line(1,1){10}}

158: \put(-2,-2){\makebox(0,0){9}}

159: \put(8,-2){\makebox(0,0){10}}

160: \put(18,-2){\makebox(0,0){11}}

161: \put(28,-2){\makebox(0,0){12}}

162: \put(38,-2){\makebox(0,0){9}}

163: \put(-2,12){\makebox(0,0){5}}

164: \put(8,12){\makebox(0,0){6}}

165: \put(18,12){\makebox(0,0){7}}

166: \put(28,12){\makebox(0,0){8}}

167: \put(38,12){\makebox(0,0){5}}

168: \put(-2,22){\makebox(0,0){1}}

169: \put(8,22){\makebox(0,0){2}}

170: \put(18,22){\makebox(0,0){3}}

171: \put(28,22){\makebox(0,0){4}}

172: \put(38,22){\makebox(0,0){1}}

173: \put(20,-8){\makebox(0,0){(a)}}

174: 

175: %Mobius triangular lattice with L_y=3

176: \multiput(60,0)(10,0){5}{\circle*{2}}

177: \multiput(60,10)(10,0){5}{\circle*{2}}

178: \multiput(60,20)(10,0){5}{\circle*{2}}

179: \multiput(60,0)(10,0){5}{\line(0,1){20}}

180: \multiput(60,0)(0,10){3}{\line(1,0){40}}

181: \put(60,10){\line(1,1){10}}

182: \multiput(60,0)(10,0){3}{\line(1,1){20}}

183: \put(90,0){\line(1,1){10}}

184: \put(58,-2){\makebox(0,0){5}}

185: \put(68,-2){\makebox(0,0){6}}

186: \put(78,-2){\makebox(0,0){7}}

187: \put(88,-2){\makebox(0,0){8}}

188: \put(98,-2){\makebox(0,0){1}}

189: \put(58,12){\makebox(0,0){9}}

190: \put(68,12){\makebox(0,0){10}}

191: \put(78,12){\makebox(0,0){11}}

192: \put(88,12){\makebox(0,0){12}}

193: \put(98,12){\makebox(0,0){9}}

194: \put(58,22){\makebox(0,0){1}}

195: \put(68,22){\makebox(0,0){2}}

196: \put(78,22){\makebox(0,0){3}}

197: \put(88,22){\makebox(0,0){4}}

198: \put(98,22){\makebox(0,0){5}}

199: \put(80,-8){\makebox(0,0){(b)}}

200: \end{picture}

201: \vspace*{3cm}

202: 

203: %torus triangular lattice with L_y=3

204: \begin{picture}(100,30)

205: \multiput(0,0)(10,0){5}{\circle*{2}}

206: \multiput(0,10)(10,0){5}{\circle*{2}}

207: \multiput(0,20)(10,0){5}{\circle*{2}}

208: \multiput(0,30)(10,0){5}{\circle*{2}}

209: \multiput(0,0)(10,0){5}{\line(0,1){30}}

210: \multiput(0,0)(0,10){4}{\line(1,0){40}}

211: \put(0,20){\line(1,1){10}}

212: \put(0,10){\line(1,1){20}}

213: \multiput(0,0)(10,0){2}{\line(1,1){30}}

214: \put(20,0){\line(1,1){20}}

215: \put(30,0){\line(1,1){10}}

216: \put(-2,-2){\makebox(0,0){1}}

217: \put(8,-2){\makebox(0,0){2}}

218: \put(18,-2){\makebox(0,0){3}}

219: \put(28,-2){\makebox(0,0){4}}

220: \put(38,-2){\makebox(0,0){1}}

221: \put(-2,12){\makebox(0,0){9}}

222: \put(8,12){\makebox(0,0){10}}

223: \put(18,12){\makebox(0,0){11}}

224: \put(28,12){\makebox(0,0){12}}

225: \put(38,12){\makebox(0,0){9}}

226: \put(-2,22){\makebox(0,0){5}}

227: \put(8,22){\makebox(0,0){6}}

228: \put(18,22){\makebox(0,0){7}}

229: \put(28,22){\makebox(0,0){8}}

230: \put(38,22){\makebox(0,0){5}}

231: \put(-2,32){\makebox(0,0){1}}

232: \put(8,32){\makebox(0,0){2}}

233: \put(18,32){\makebox(0,0){3}}

234: \put(28,32){\makebox(0,0){4}}

235: \put(38,32){\makebox(0,0){1}}

236: \put(20,-8){\makebox(0,0){(c)}}

237: 

238: %Klein bottle triangular lattice with L_y=3

239: \multiput(60,0)(10,0){5}{\circle*{2}}

240: \multiput(60,10)(10,0){5}{\circle*{2}}

241: \multiput(60,20)(10,0){5}{\circle*{2}}

242: \multiput(60,30)(10,0){5}{\circle*{2}}

243: \multiput(60,0)(10,0){5}{\line(0,1){30}}

244: \multiput(60,0)(0,10){4}{\line(1,0){40}}

245: \put(60,20){\line(1,1){10}}

246: \put(60,10){\line(1,1){20}}

247: \multiput(60,0)(10,0){2}{\line(1,1){30}}

248: \put(80,0){\line(1,1){20}}

249: \put(90,0){\line(1,1){10}}

250: \put(58,-2){\makebox(0,0){1}}

251: \put(68,-2){\makebox(0,0){2}}

252: \put(78,-2){\makebox(0,0){3}}

253: \put(88,-2){\makebox(0,0){4}}

254: \put(98,-2){\makebox(0,0){5}}

255: \put(58,12){\makebox(0,0){5}}

256: \put(68,12){\makebox(0,0){6}}

257: \put(78,12){\makebox(0,0){7}}

258: \put(88,12){\makebox(0,0){8}}

259: \put(98,12){\makebox(0,0){1}}

260: \put(58,22){\makebox(0,0){9}}

261: \put(68,22){\makebox(0,0){10}}

262: \put(78,22){\makebox(0,0){11}}

263: \put(88,22){\makebox(0,0){12}}

264: \put(98,22){\makebox(0,0){9}}

265: \put(58,32){\makebox(0,0){1}}

266: \put(68,32){\makebox(0,0){2}}

267: \put(78,32){\makebox(0,0){3}}

268: \put(88,32){\makebox(0,0){4}}

269: \put(98,32){\makebox(0,0){5}}

270: \put(80,-8){\makebox(0,0){(d)}}

271: \end{picture}

272: \vspace*{3cm}

273: 

274: \begin{figure}[h]

275: %\vspace{-4cm}

276: %\centering

277: %\leavevmode

278: %\epsfxsize=6.0in

279: %\begin{center}

280: %\leavevmode

281: %\epsffile{trigraphsly3.ps}

282: %\end{center}

283: %\vspace{-2cm}

284: \caption{\footnotesize{Illustrative strip graphs of the triangular lattice with

285: width $L_y=3$ and length $L_x=4$ having the following

286: boundary conditions: (a) $(FBC_y,PBC_x)=$ cyclic, (b) $(FBC_y,TPBC_x)=$

287: M\"obius, (c) $(PBC_y,PBC_x)=$ toroidal, and (d) $(PBC_y,TPBC_x)=$ Klein

288: bottle.}}

289: \label{fig1}

290: \end{figure}

291: 

292:   In the present work we report exact calculations of $P(G,q)$, $W(\{G\},q)$,

293: and ${\cal B}$ for families of triangular lattice strips of fixed width $L_y=3$

294: vertices and arbitrary length $L_x=m$ vertices.  The longitudinal and

295: transverse directions are taken to be $\hat x$ and $\hat y$).  In Fig. 1 we

296: display some illustrative examples.  

297: A generic form for chromatic polynomials for recursively defined families of

298: graphs, of which strip graphs $G_s$ are special cases, is

299: \beq

300: P((G_s)_m,q) =  \sum_{j=1}^{N_\lambda} c_j(q)(\lambda_j(q))^m

301: \label{pgsum}

302: \eeq

303: where $c_j(q)$ and the $N_\lambda$ terms $\lambda_j(q)$ depend on the type of

304: strip graph $G_s$ but are independent of $m$. 

305: 

306: 

307:    Let us comment on the motivations for the current study and how it extends

308: previous work.  Calculations of the chromatic polynomials for the cyclic and

309: M\"obius strip graphs of the square lattice were carried out for $L_y=2$ in

310: \cite{bds} (see also \cite{bm,b}).  Subsequently, these calculations were

311: extended to the width $L_y=3$ for the cyclic \cite{wcy} and M\"obius \cite{pm}

312: square lattice strips. After studies of chromatic zeros in \cite{bkw79,bkw}

313: and, for $L_y=2$ strips in \cite{bds,readcarib88,read91}, the $W$ and ${\cal

314: B}$ defined in the infinite-length limit were determined for $L_y=2$ in

315: \cite{w} and for $L_y=3$ in \cite{wcy,pm}.  Similar calculations were carried

316: out for strips of various lattices with widths up to $L_y=4$ and free boundary

317: conditions in \cite{strip,strip2,hs,w2d}, and for cyclic and M\"obius graphs

318: involving homeomorphic expansions of square strips \cite{pg} and cyclic polygon

319: chains \cite{nec}.  An important question concerns the effect of boundary

320: conditions (BC's), and hence graph topology, on $P$, $W$, and ${\cal B}$. We

321: use the symbols FBC$_y$ and PBC$_y$ for free and periodic transverse boundary

322: conditions and FBC$_x$, PBC$_x$, and TPBC$_x$ for free, periodic, and twisted

323: periodic longitudinal boundary conditions.  The term ``twisted'' means that the

324: longitudinal ends of the strip are identified with reversed orientation.  These

325: strip graphs can be embedded on surfaces with the following topologies: (i)

326: (FBC$_y$,FBC$_x$): open strip; (ii) (PBC$_y$,FBC$_x$): cylindrical; (iii)

327: (FBC$_y$,PBC$_x$): cylindrical, denoted cyclic here; (iv) (FBC$_y$,TPBC$_x$):

328: M\"obius; (v) (PBC$_y$,PBC$_x$): torus; and (vi) (PBC$_y$,TPBC$_x$): Klein

329: bottle.\footnote{\footnotesize{These BC's can all be implemented in a manner

330: that is uniform in the length $L_x$; the case (vii) (TPBC$_y$,TPBC$_x$) with

331: the topology of the projective plane requires different identifications as

332: $L_x$ varies and will not be considered here.}}  (We recall that that unlike

333: graphs of type (i)-(v), the Klein bottle surface cannot be embedded without

334: self-intersection in ${\mathbb R}^3$.)

335: 

336: Several specific effects of different boundary conditions have been found (for

337: a general discussion, see \cite{bcc,tw99}).  First, from comparisons of exact

338: calculations of $W$ for the infinite-length limit of lattice strips with

339: boundary conditions of types (i)-(iv), it was observed that those with a global

340: circuit\footnote{\footnotesize{A global circuit is a route following a lattice

341: direction which has the topology of $S^1$ and a length $\ell_{g.c.}$ that goes

342: to infinity as $n \to \infty$.}}  lead to a locus ${\cal B}$ that separates the

343: $q$ plane into different regions, and it was inferred that the presence of

344: global circuits is a sufficient condition for ${\cal B}$ to have this property

345: \cite{strip,hs,pg,nec,wcy,pm}.\footnote{\footnotesize{This is not a necessary

346: condition, as was shown in \cite{strip2}.}}  For strip graphs of regular

347: lattices, this is equivalent to having $PBC_x$ or $TPBC_x$, i.e., periodic or

348: twisted periodic boundary conditions in the direction in which the strip length

349: goes to infinity as $L_x \to \infty$.  Further support for this inference was

350: adduced from the calculation of $P$, $W$, and ${\cal B}$ for the $L_y=3$ strip

351: of the square lattice with torus and Klein bottle boundary conditions

352: \cite{tk}.  Applications of a coloring matrix method \cite{b} for these

353: calculations has been discussed in \cite{matmeth,bgen}.  For the (homogeneous)

354: strip graphs with (FBC$_y$,FBC$_x$) in \cite{strip,hs}, ${\cal B}$ consists of

355: arcs that do not separate regions of the $q$ plane.  As the width of the strip

356: increases, these arcs tend to elongate, and their ends tend to move toward each

357: other, thereby suggesting that if one considered the sequence of strip graphs

358: of this type with width $L_y$ (having taken the limit $L_x \to \infty$ to

359: obtain a locus ${\cal B}$ for each member of this sequence), then in the limit

360: $L_y \to \infty$, the arcs would close to form a ${\cal B}$ that separated the

361: $q$ plane into two or more regions and passed through $q=0$.  The interesting

362: feature of the families of graphs with global circuits is that when the graphs

363: contain a global circuit, this separation of the $q$ plane into regions (with

364: ${\cal B}$ passing through $q=0$) already occurs for finite $L_y$. This means

365: that the $W$ functions of the infinite-length limit of these graphs with cyclic

366: and twisted cyclic longitudinal boundary conditions already exhibit a feature

367: which is expected to occur for the ${\cal B}$ for the $W$ function of the full

368: two-dimensional lattice.  This expectation is supported by the calculation of

369: $W$ and ${\cal B}$ for the 2D triangular lattice with cylindrical boundary

370: conditions by Baxter \cite{baxter}.  Thus, although calculations of chromatic

371: polynomials for lattice strips with global circuits are, in general, more

372: difficult than for the corresponding strips with free boundary conditions, such

373: calculations are worthwhile, since in the $L_x \to \infty$ limit, the resultant

374: locus ${\cal B}$ embodies the analytic property of ${\cal B}$ expected for the

375: full 2D lattice, viz., that it crosses the real axis at least at $q=0$ and a 

376: maximal point $q_c$, and separates the $q$ plane into different regions. 

377: 

378: 

379: 

380: Second, it has been found that for a given type of strip graph $G_s$ with

381: FBC$_y$, the chromatic polynomials for PBC$_x$ and TPBC$_x$ boundary conditions

382: (i.e., cyclic and M\"obius strips) have the same $\lambda_j$, although in

383: general different $c_j$.  It follows that the loci ${\cal B}$ are the same for

384: these two different longitudinal boundary condition choices \cite{pg,wcy,pm}.

385: In the case of PBC$_y$, the reversal of orientation involved in going from

386: PBC$_x$ to TPBC$_x$ longitudinal boundary conditions (i.e. from torus to Klein

387: bottle topology) can lead to the removal of some of the $\lambda_j$'s that were

388: present; i.e., $P$ for the (PBC$_y$,TPBC$_x$) strip may involve only a subset

389: of the $\lambda_j$'s that are present for the (PBC$_y$,PBC$_x$) strip.  For

390: example, for the $L_y=3$ strips of the square lattice with (PBC$_y$, PBC$_x$)

391: boundary conditions, there are $N_\lambda=8$ $\lambda_j$'s, but for the strip

392: with (PBC$_y$, TPBC$_x$) boundary conditions only a subset of $N_\lambda=5$ of

393: these terms occurs in $P$ \cite{tk}.  None of the three $\lambda_j$'s that are

394: absent from $P$ in the TPBC$_x$ case is leading, so that ${\cal B}$ is the same

395: for both of these families.  

396: 

397: 

398: Third, for a given type of strip graph $G_s$ containing a global circuit, it

399: has been found that, in the infinite-length limit where the locus ${\cal B}$ is

400: defined, it not only passes through the origin, $q=0$, but always crosses the

401: positive real axis at one or more points, the maximal one being denoted

402: $q_c(\{G\})$, as mentioned above.  In contrast, for strip graphs that do not

403: contain global circuits, ${\cal B}$ may not cross the real axis.  For example,

404: for the (infinite-length limit of the) strip of the triangular lattice with

405: free transverse and longitudinal boundary conditions, $(FBC_y,FBC_x)$, the

406: locus ${\cal B}$ crosses the real axis for width $L_y=3$ but does not cross it

407: for $L_y=4$ (see Fig. 5 of \cite{strip}).  For square strips with

408: $(FBC_y,(T)PBC_x)$ boundary conditions, it has been found that $q_c(\{G\})$ is

409: a nondecreasing function of $L_y$ for the cases studied so far, namely $q_c=2$

410: for $L_y=1$ and $L_y=2$, and $q_c \simeq 2.33654$ for $L_y=3$

411: \cite{wcy,pm,bcc}.  For the strips of the triangular lattice with

412: $(FBC_y,(T)PBC_x)$ boundary conditions, we only have one width for which ${\cal

413: B}$ has been determined, namely, $L_y=2$, for which $q_c=3$.  It is thus of

414: interest to explore what the value of $q_c$ is for cyclic and M\"obius strips

415: of the triangular lattice with larger widths.  For the $L_y=3$ torus and Klein

416: bottle strips of the square lattice, very interestingly, $q_c=3$, which is the

417: same value as for the 2D square lattice \cite{tk}.  Clearly it is worthwhile

418: to investigate what the value of $q_c$ is for torus and Klein bottle strips of

419: the triangular lattice, and we shall do this here.  

420: 

421: Fourth, for the strip graph $(G_s)_m$ with a given type of transverse boundary

422: conditions BC$_y$, the chromatic polynomial for PBC$_x$ has a larger value of

423: $N_\lambda$ than the chromatic polynomial for FBC$_x$, and the corresponding

424: loci ${\cal B}$ are different.

425: 

426: Fifth, one may ask how the $W$ functions in region $R_1$ compare for 

427: different boundary conditions.  It has been found \cite{pg,nec,wcy,pm,bcc}

428: that, for a given type of strip graph $G_s$, in the region $R_1$ (with its

429: left-hand boundary on the positive real $q$ axis at $q_c$) being 

430: defined for the PBC$_x$ boundary conditions, the $W$ function is the same for

431: FBC$_x$, PBC$_x$, and TPBC$_x$.  This includes the region of real $q$ greater

432: than this value of $q_c$, so that this result is somewhat reminiscent of the

433: statement of the existence of the thermodynamic limit in statistical mechanics,

434: i.e. the fact that in the disordered phase of a statistical mechanical 

435: system the thermodynamic functions are independent of the boundary conditions

436: and if an ordered phase exists, then the role of the boundary conditions is

437: only to set a preferred direction for the symmetry-breaking order parameter. 

438: 

439: 

440: Our new results on $P$, $W$, and ${\cal B}$ for the $L_y=3$ strip of the

441: triangular lattice with the various boundary conditions of type (iii)-(vi) give

442: insight concerning these five items, as do our additional results for $L_y=4$

443: cyclic strips and $L_y=5,6$ cylindrical strips. 

444: Calculations of these quantities

445: for the $L_y=2$ cyclic and M\"obius triangular strips were presented in

446: \cite{wcy}.  An early study of $P$ for the $L_y=2$ strip is in

447: \cite{sands}\footnote{\footnotesize{We note the following corrections in Sand's

448: result for $P$ for the $L_y=2$ cyclic strip of the triangular lattice, given on

449: p. 88 of \cite{sands}: the coefficients of the last two terms should be

450: reversed in sign to read ($z \equiv q$ in his notation) \ $(z-1)\Bigl [

451: [(1/2)(5-2z + \sqrt{9-4z})]^m + [(1/2)(5-2z - \sqrt{9-4z})]^m \Bigr ]$, as in

452: eq. (5.23) of \cite{wcy}.}} and a coloring matrix method to obtain $P$ has been

453: given in \cite{matmeth}.  $P$, $W$, and ${\cal B}$ have been calculated for

454: triangular lattice strips with free boundary conditions (i) for $L_y$ up to 4

455: \cite{strip}. (Some related calculations including work on cases with

456: noncompact ${\cal B}$, are \cite{read91,w}, \cite{wa}-\cite{sokal}.)

457: 

458: For strips with global circuits, and $L_x$ above the lowest values involving

459: degenerate cases, the triangular lattice strips with boundary conditions of

460: type (i)-(vi) have $n=L_yL_x$ vertices; the cyclic and M\"obius triangular

461: strips have $e=(3L_y-2)L_x$ edges while the torus and Klein bottle triangular

462: strips have $e=3L_yL_x$ edges.  Define $N_t$ to be the number of triangles in

463: the strips (i.e., excluding the triangles forming the cross sections of the

464: strips in the case $L_y=3$); then, again for $L_x$ and $L_y$ large enough to

465: avoid degenerate cases, the cyclic and M\"obius strips have $N_t=2(L_y-1)L_x$

466: such triangles, while the torus and Klein bottle strips have $2L_yL_x$ such

467: triangles.  For strips of type (i), $e = 3L_xL_y-2L_x-2L_y+1$ and

468: $N_t=2(L_y-1)(L_x-1)$ while for strips of type (ii), $e=3L_xL_y-2L_y$ and

469: $N_t=2L_y(L_x-1)$.

470: 

471: 

472:   Again, for $L_x=m$ and $L_y$ large enough to

473: avoid degenerate cases, the chromatic number for the triangular

474: lattice strips with cyclic or torus boundary conditions is

475: \beq

476: \chi(tri(L_y)_m,FBC_y,PBC_x) = \chi(tri(L_y)_m,PBC_y,PBC_x) = 

477: \cases{ 3 & if $m=0$ \ mod 3 \cr

478:         4 & if $m=1$ or 2 \ mod 3} 

479: \label{chitricyclictorus}

480: \eeq

481: and for the triangular lattice strips with M\"obius or Klein bottle boundary

482: conditions, 

483: \beq

484: \chi(tri(L_y)_m,FBC_y,TPBC_x) = \chi(tri(L_y)_m,PBC_y,TPBC_x) = 4 \quad \forall

485: \quad m \ge 3 \ . 

486: \label{chitrimbklein}

487: \eeq

488: For strips with $(FBC_y,FBC_x)$ or $(PBC_y,FBC_x)$, $\chi=3$. 

489: 

490: \section{$L_y=3$ Cyclic and M\"obius Triangular Strips} 

491: 

492: We calculate the chromatic polynomials by iterated use of the

493: deletion-contraction theorem, via a generating function approach 

494: \cite{strip,hs}.  In general, for a strip graph of type $G_s$ we have 

495: \beq

496: \Gamma(G_s,q,x) = \sum_{m=m_0}^{\infty}P((G_s)_m,q)x^{m-m_0}

497: \label{gamma}

498: \eeq

499: where, as before \cite{wcy}, we take $m_0=2$ for strips with periodic or

500: twisted periodic longitudinal boundary conditions. 

501: The generating functions $\Gamma(G_s,q,x)$ are rational functions of the form

502: \beq

503: \Gamma(G_s,q,x) = \frac{{\cal N}(G_s,q,x)}{{\cal D}(G_s,q,x)}

504: \label{gammagen}

505: \eeq

506: with

507: \beq

508: {\cal N}(G_s,q,x) = \sum_{j=0}^{d_{\cal N}} A_{G_s,j}(q) x^j

509: \label{n}

510: \eeq

511: and

512: \beq

513: {\cal D}(G_s,q,x) = 1 + \sum_{j=1}^{d_{\cal D}} b_{G_s,j}(q) x^j 

514: \label{d}

515: \eeq

516: where the $A_{G_s,i}$ and $b_{G_s,i}$ are polynomials in $q$ (with no common

517: factors) and

518: \beq

519: d_{\cal N} = deg_x({\cal N})

520: \label{dn}

521: \eeq

522: and

523: \beq

524: d_{\cal D} = deg_x({\cal D}) \ . 

525: \label{dd}

526: \eeq

527: The denominator ${\cal D}$ can be written in factorized form as 

528: \beq

529: {\cal D}(G_s,q,x) = \prod_{j=1}^{d_{\cal D}}(1-\lambda_{G_s,j}x) \ . 

530: \label{lambdaform}

531: \eeq

532: Equivalently, the $\lambda_{G_s,j}$ are roots of the equation

533: \beq

534: \xi^{d_{\cal D}}{\cal D}(G_s,q,1/\xi) = \xi^{d_{\cal D}} +

535: \sum_{j=1}^{d_{\cal D}} b_{G_s,j}\xi^{d_{\cal D}-j} \ . 

536: \label{xieq}

537: \eeq

538:  From the generating function, we calculate the chromatic polynomials in the

539:  form of eq. (\ref{pgsum}) by the use of the general formula, eq. (2.14), in 

540: \cite{hs},

541: \beq

542: P(G_m,q) = \sum_{j=1}^{d_{\cal D}} \Biggl [ \sum_{s=0}^{d_{\cal N}}

543: A_s \lambda_j^{d_{\cal D}-s-1} \Biggr ]

544: \Biggl [ \prod_{1 \le i \le d_{\cal D}; \ i \ne j}

545: \frac{1}{(\lambda_j-\lambda_i)} \Biggr ] \lambda_j^{m-m_0}

546: \label{chrompgsumlam}

547: \eeq

548: (where the conventions in eq. (\ref{gamma}) were chosen such that $m_0=0$ in 

549: \cite{hs}).  Note that some of the coefficients $c_{G_s,j}$ may vanish, so that

550: not all of the $\lambda_{G_s,j}$'s in ${\cal  D}(G_s,q,x)$ contribute to the

551: sum in (\ref{pgsum}) \cite{wcy}.  The formula  (\ref{pgsum}) for the chromatic

552: polynomial has the virtue of being a closed-form expression.  However, for a

553: given type of strip graph, for values of the width $L_y$ greater than the first

554: one or two values, the set of $\lambda_j$'s include nonpolynomial algebraic

555: roots.  These occur, of course, as symmetric functions of the roots,

556: so that, by a theorem on symmetric functions of roots of algebraic equations,

557: their resultant contributions to $P$ are expressible in terms of the

558: coefficients of these equations (which are polynomials in $q$) and hence are 

559: polynomials in $q$ \cite{pm}. However, in calculating chromatic polynomials

560: and the resultant chromatic zeros for large values of $L_x=m$, it can be 

561: convenient to use directly the expression for the generating function, since

562: both the numerator and denominator of this function, eqs. (\ref{n}), (\ref{d}),

563: are polynomials in $q$ so that one does not have to rely upon cancellations of

564: nonpolynomial algebraic expressions at intermediate stages in the calculation. 

565: 

566: 

567: \vspace{5mm}

568: 

569: The coefficients $c_{tri(L_y),j}$ that enter into the expressions for the

570: chromatic polynomial (\ref{pgsum}) for the cyclic triangular ($t$) strip of 

571: width $L_y$ are (like those for the cyclic square strip) certain polynomials 

572: that we denote $c^{(d)}$, given by \cite{cf}

573: \beq

574: c^{(d)} = \prod_{k=1}^d (q-q_{d,k})

575: \label{cd}

576: \eeq

577: where 

578: \beq

579: q_{d,k} \equiv 2+2\cos \Bigl ( \frac{2\pi k}{2d+1} \Bigr ) 

580: \quad {\rm for} \quad

581: k=1,2,...d

582: \label{cdzeros}

583: \eeq

584: with $0 \le d \le L_y$. We list below the specific $c^{(d)}$'s that 

585: appear in our results for the $L_y=3$ and $L_y=4$ cyclic strips of the 

586: triangular lattice 

587: \beq

588: c^{(0)}=1 \ , \quad c^{(1)}=q-1 \ , \quad c^{(2)}=q^2-3q+1 \ , 

589: \label{cd012}

590: \eeq

591: \beq

592: c^{(3)}=q^3-5q^2+6q-1 \ , 

593: \label{cd3}

594: \eeq

595: and

596: \beq

597: c^{(4)}=(q-1)(q^3-6q^2+9q-1) \ . 

598: \label{cd4}

599: \eeq

600: Where the coefficient $c^{(0)}$ appears in chromatic polynomials, we shall

601: simply write it as unity. 

602: 

603: \vspace{8mm}

604: 

605: For the cyclic $L_y=3$ triangular strip of length $L_x=m$, we obtain the 

606: result 

607: \beqs

608: & & P(tri(3 \times m),FBC_y,PBC_x,q)=\sum_{j=1}^{10} c_{t3,j} 

609: (\lambda_{t3,j})^m

610: \cr\cr

611: & = & c^{(3)}(-1)^m 

612: + c^{(2)}\Bigl [ (q-2)^m+(\lambda_{t3,3})^m + (\lambda_{t3,4})^m \Bigr ] 

613: \cr\cr

614: & + & c^{(1)}\Bigl [ [-(q-2)(q-3)]^m + (\lambda_{t3,6})^m + (\lambda_{t3,7})^m

615:  + (\lambda_{t3,8})^m \Bigr ] + \Bigl [ 

616: (\lambda_{t3,9})^m + (\lambda_{t3,10})^m  \Bigr ] \ . \cr\cr 

617: & & 

618: \label{ptcyclic}

619: \eeqs

620: where the short notation $t3$ is used in subscripts for $tri(L_y=3)$.  

621: The labelling of $c_{t3,j}$ and $\lambda_{t3,j}$ in eq. (\ref{ptcyclic}) is 

622: consecutive, so that $c_{t3,1}=c^{(3)}$, $\lambda_{t3,1}=-1$; 

623: $c_{t3,2}=c^{(2)}$, $\lambda_{t3,2}=q-2$, and so forth: 

624: \beq

625: \lambda_{t3,(3,4)}=\frac{1}{2}\Bigl ( -7+2q \pm \sqrt{25-8q} \ \Bigr ) \ , 

626: \label{lamt34}

627: \eeq

628: $\lambda_{t3,j}$, $j=6,7,8$ are the roots of the cubic equation 

629: \beq

630: \xi^3+b_{t3,1}\xi^2+b_{t3,2}\xi + b_{t3,3}=0 

631: \label{cubeq}

632: \eeq

633: where 

634: \beq

635: b_{t3,1}=2q^2-12q+19

636: \label{bt31}

637: \eeq

638: \beq

639: b_{t3,2}=(q-2)(q^3-9q^2+28q-29)

640: \label{bt32}

641: \eeq

642: \beq

643: b_{t3,3}=-(q-2)^4(q-3) \ , 

644: \label{bt33}

645: \eeq

646: and 

647: \beq

648: \lambda_{t3,(9,10)}=\frac{1}{2}\biggl [ q^3-7q^2+18q-17 \pm 

649: (q^6-14q^5+81q^4-250q^3+442q^2-436q+193)^{1/2} \biggr ] \ . 

650: \label{lamt910}

651: \eeq

652: Note that some $\lambda_{t3,j}$'s vanish if $q=2$ or $q=3$; specifically, if 

653: $q=2$, then $\lambda_{t3,3}=0$ and two of the roots of the cubic equation

654: (\ref{cubeq}) vanish, while for $q=3$, $\lambda_{t3,3}=0$ and one of the 

655: roots of eq. (\ref{cubeq}) vanish.  The vanishing of the chromatic polynomial

656: at $q=0,1,2$ and (i) $q=3$ for the cyclic strip with $m \ne 0$ mod 3 or (ii)

657: $q=3$ for the M\"obius strip discussed below involves both some terms

658: $\lambda_j$ vanishing and cancellations among others.  This is also true of

659: chromatic polynomials for strips of other lattices. 

660: 

661: Parenthetically, we note that the denominator of the generating function for 

662: this strip graph actually has $d_{\cal D}=16$:

663: \beq

664: {\cal D}(tri(L_y=3),FBC_y,PBC_x,q,x) = \sum_{j=1}^{16}(1-\lambda_{t3,j}(q)x)

665: \label{lambdaformt}

666: \eeq

667: where the $\lambda_{t,j}$ were given in eq. (\ref{ptcyclic}) and 

668: \beq

669: \lambda_{t,11}=\lambda_{t,1}=-1

670: \label{lamt11}

671: \eeq

672: \beq

673: \lambda_{t,12}=\lambda_{t,13}=q-3

674: \label{lamt1213}

675: \eeq

676: and $\lambda_{t,j}$, $j=14,15,16$ are the roots of the cubic equation 

677: \beq

678: \xi^3+(2q-5)(q-3)\xi^2+(q-2)^2(q-3)(q-4)\xi-(q-2)^4(q-3)=0 \ . 

679: \label{extracubic}

680: \eeq 

681: A new feature of these results that we have not encountered before in our

682: calculations of chromatic polynomials and generating functions for strip graphs

683: is the occurrence of factors in ${\cal D}$ that have multiplicity higher than

684: unity: specifically, the factors $(1+x)^2$, corresponding to

685: $\lambda_{t3,1}=\lambda_{t3,11}=-1$, and $[1-(q-3)x]^2$, corresponding to

686: $\lambda_{t3,12}=\lambda_{t3,13}=q-3$.  None of the $\lambda_{t3,j}$, 

687: $12 \le j \le 16$, contribute to $P$ (and $\lambda_{t3,11}$ is identical to 

688: $\lambda_{t3,1}$), so that $P$ only involves the ten $\lambda_{t3,j}$'s in

689: eq. (\ref{ptcyclic}).  We had found cases earlier of some $\lambda_j$'s in

690: ${\cal D}$ not contributing to $P$, e.g., for the $L_y=2$ cyclic strip of the

691: kagom\'e lattice and homeomorphic expansions of the $L_y=2$ cyclic strip of the

692: square lattice \cite{wcy}.  However, this is the first value of $L_y$ where we

693: have encountered this type of behavior for a strip of a homopolygonal

694: lattice.\footnote{\footnotesize{As in \cite{cmo,wn}, we define a homopolygonal

695: 2D lattice as a regular tiling of the plane involving a single type of polygon,

696: in which all vertices are equivalent, while a heteropolygonal lattice is the

697: same but involving two or more different types of regular polygons. In the case

698: of the kagom\'e lattice, these are triangles and hexagons.}}

699: 

700: The numerator of the generating function has degree 

701: $deg_x({\cal N})=14$.  The polynomials $A_j$, $j=1,.., 14$ comprising this

702: numerator can be calculated from the expressions for $P$ and ${\cal D}$, 

703: eqs. (\ref{ptcyclic}) and (\ref{lambdaformt}) and hence are not listed here. 

704: 

705:  From eq. (\ref{chitricyclictorus}), it follows that for $m=1$ or 2 mod 3 and 

706: $q=3$, the chromatic polynomial in eq. (\ref{ptcyclic}) vanishes; for $m=0$ 

707: mod 3 (and $m \ge 3$) and $q=3$ we find 

708: \beq

709: P(tri(L_y=3)_m,FBC_y,PBC_x,q=3)=6 \quad {\rm for} \quad m = 0 \quad mod \ 3 \ .

710: \label{ptq3m0mod3}

711: \eeq

712: This may be contrasted with the case of a $k$-critical graph $G_k$, where 

713: $\chi(G)=k$ and $P(G,q=\chi(G))=k!$ (e.g. $k=3$ for the triangular lattice). 

714: 

715: We define

716: \beq

717: C(G)=\sum_{j=1}^{N_{\lambda_G}} c_{G,j} \ . 

718: \label{cgsum}

719: \eeq

720: where the $G$-dependence in the coefficients is indicated explicitly.  Note

721: that for recursive graphs like the strip graphs considered here, the $c_{G,j}$

722: depend on $L_y$ and the boundary conditions, but not on $L_x$. Our results 

723: above give 

724: \beq

725: C=q(q-1)^2 \quad {\rm for} \quad G=tri(L_y=3,FBC_y,PBC_x) \ .

726: \label{csumtrily3cyc}

727: \eeq

728: in accord with the generalization \cite{wcy,pm} $C=P(T_{L_y},q)$ for this

729: strip, where $P(T_n,q)=q(q-1)^{n-1}$ and $T_n$ is the tree graph with $n$

730: vertices.  This was also true for the $L_y=3$ cyclic strip of the square

731: lattice \cite{wcy,pm} and is in accord with the coloring matrix approach

732: \cite{tk,matmeth}.

733: 

734: \vspace{6mm}

735: 

736: For the $L_y=3$ triangular lattice M\"obius strip (denoted by the subscript 

737: $t3Mb$) we calculate 

738: \beq 

739: P(tri(L_y=3)_m,FBC_y,TPBC_x,q) = \sum_{j=1}^{10}c_{t3Mb,j} (\lambda_{t3Mb,j})^m

740: \label{pmobius}

741: \eeq

742: where 

743: \beq

744: \lambda_{t3Mb,j} = \lambda_{t3,j} \ , \quad j=1,..,10 \ . 

745: \label{lamtm}

746: \eeq

747: As was true in the case of the L$_y=2$ M\"obius strip of the triangular 

748: lattice, the coefficients $c_j$ are not just of the $c^{(d)}$ polynomial 

749: form; instead, some coefficients are rational functions of $q$ and others are 

750: algebraic nonpolynomial functions of $q$: 

751: \beq

752: c_{t3Mb,1}=c^{(2)}

753: \label{c1tmb}

754: \eeq

755: \beq

756: c_{t3Mb,2}=-\frac{(q^2-4q+5)}{2(q-2)}

757: \label{c2tmb}

758: \eeq

759: 

760: \beq

761: c_{t3Mb,(3,4)} = \frac{1}{4(q-2)} \biggl [ (q-3)^2 \mp \frac{(q-1)(q-5)}

762: {\lambda_{t,3}-\lambda_{t,4}} \ \biggr ]

763: \label{c34tmb}

764: \eeq

765: 

766: \beq

767: c_{t3Mb,5} = \frac{(q-1)(q^2-4q+5)}{2(q-2)}

768: \label{c5tmb}

769: \eeq

770: 

771: \beq

772: c_{t3Mb,9}=c_{t3Mb,10}=c^{(0)}=1 \ . 

773: \label{c910tmb}

774: \eeq

775: The $j=3,4$ terms can be written in a manner which manifestly 

776: exhibits their property of being a symmetric function of $\lambda_{t3,3}$ and 

777: $\lambda_{t3,4}$:

778: \beq

779: c_{t3Mb,3}(\lambda_{t3,3})^m+c_{t3Mb,4}(\lambda_{t3,4})^m = 

780: \frac{(q-3)^2}{4(q-2)}\Bigl [ (\lambda_{t3,3})^m+(\lambda_{t3,4})^m \Bigr ] 

781: -\frac{(q-1)(q-5)}{4(q-2)}\biggl [ \frac{(\lambda_{t3,3})^m-(\lambda_{t3,4})^m}{

782: \lambda_{t3,3}-\lambda_{t3,4}} \biggr ] \ . 

783: \label{ctm34}

784: \eeq

785: It is convenient to leave the expressions for $c_{t3Mb,j}$, $j=6,7,8$, in the

786: general forms that follow from eq. (\ref{chrompgsumlam}):

787: \beq

788: c_{t3Mb,j}=\lambda_{t3,j}^{-2} \Biggl [ \sum_{s=0}^8 A_{tm3,s} 

789: (\lambda_{t3,j})^{9-s} \Biggr ]\Biggl [ \prod_{1 \le i \le 10; \ i \ne j}

790: \frac{1}{(\lambda_{t3,j}-\lambda_{t3,i})} \Biggr ] \ , \quad j=6,7,8

791: \label{ctmj}

792: \eeq

793: where the $A_{t3Mb,s}$ are given in the appendix. 

794: 

795: For the sum of the coefficients for the $L_y=3$ M\"obius strip of the 

796: triangular lattice we calculate 

797: \beq

798: C=\sum_{j=1}^{10}c_{t3Mb,j}=q(q-1) \quad {\rm for} \quad

799: G=tri(L_y=3,FBC_y,TPBC_x)

800: \label{csumtrily3mb}

801: \eeq

802: in agreement with a general formula that we have derived elsewhere for a

803: M\"obius strip graph $G_s$ of width $L_y$ of the square or triangular lattice 

804: \cite{cf},

805: \beq

806: C(G_s(L_y),Mb)=\cases{ 0 & for even $L_y$ \cr

807: P(T_{(\frac{L_y+1}{2})},q) & for odd $L_y$ \cr } \ . 

808: \label{csummb}

809: \eeq

810: 

811: \begin{figure}[p]

812: %\vspace{-4cm}

813: \centering

814: \leavevmode

815: \epsfxsize=4.0in

816: \begin{center}

817: \leavevmode

818: \epsffile{tpxy3.ps}

819: \end{center}

820: %\vspace{-2cm}

821: \caption{\footnotesize{${\cal B}$ and chromatic zeros for the $L_y=3$,

822: $L_x=m=20$ (i.e., $n=60$) cyclic triangular strip graph.}}

823: \label{tricyczeros}

824: \end{figure}

825: 

826: Chromatic zeros for the $L_y=3$, $L_x=m=20$ cyclic graph of the triangular

827: lattice are shown in Fig.  \ref{tricyczeros}; with this value of $m$, the

828: chromatic zeros for the triangular lattice M\"obius graph are quite similar and

829: both sets of chromatic zeros lie close to the boundary ${\cal B}$ and indicate

830: its position.  The locus ${\cal B}$ and the $W$ functions are the same for the

831: cyclic and M\"obius graph families.  We find

832: \beq

833: q_c(tri(L_y=3),FBC_y,(T)PBC_x)=3 \ . 

834: \label{qctcycly3}

835: \eeq 

836: This will be discussed further below when we give our result for $q_c$ for the

837: corresponding $L_y=4$ strip of the triangular lattice. 

838: 

839: 

840: The locus ${\cal B}$ has support for $Re(q) \ge 0$ and separates the $q$ plane

841: into five main regions.  The outermost one, region $R_1$, extends to infinite

842: $|q|$ and includes the intervals $q \ge 3$ and $q \le 0$ on the real $q$ axis.

843: The other regions are labelled consequtively moving leftward along the real

844: axis from $q=q_c=3$ to $q=0$, and then including complex conjugate pairs of

845: regions that have no overlap with the real axis.  Thus, region $R_2$ includes

846: the real interval $2 \le q \le 3$, while region $R_3$ includes the real

847: interval $0 \le q \le 2$.  There are also two complex-conjugate regions, $R_4,

848: R_4^*$ centered at approximately $q \simeq 3.35 \pm 0.7i$.  The triple points

849: occur at (i) $q \simeq 3.14+0.37i$ and (ii) $q \simeq 3.12+1.12i$, where $R_1,

850: \ R_2$, and $R_4$ are contiguous, and (iii) $q \simeq 2.25+ 1.66i$, where

851: $R_1$, $R_2$, and $R_3$ are contiguous, together with the complex conjugates of

852: these points.  In addition, there are two tiny regions $R_5, \ R_5^*$ that

853: touch the real axis at the single point $q_c=3$ and extend a very short

854: distance to the upper and lower right, with triple points at about $q \simeq

855: 3.02 \pm 0.13i$.  These are plotted in Fig. \ref{tricyczeros}. Thus, $q_c$ is a

856: multiple point where four branches of ${\cal B}$ intersect.  The occurrence of

857: extremely small regions was also found for the ($L_x \to \infty$ limit of the)

858: $L_y=3$ cyclic strip of the square lattice, as shown in Fig. 1 of \cite{wcy}.

859: The fact that such tiny regions can occur means that when one performs the

860: usual computer scan over the complex plane to map out the region diagram and

861: determine the dominant $\lambda_j$'s in each region, since this involves a

862: finite grid, one can only detect tiny sliver regions down to a certain

863: resolution set by the grid of the scan.  On the real axis, this is not a

864: serious complication, since the mapping is a one-variable problem, but in the

865: complex plane, one must a commensurately fine scanning grid to detect tiny

866: sliver regions.

867: 

868: The part of ${\cal B}$ and the associated chromatic zeros with largest

869: magnitude occur at roughly $q \simeq 3.53 + 0.69i$ and have $|q| \simeq 3.59$.

870: Comparing the present locus ${\cal B}$ for $L_y=3$ with the one obtained in

871: \cite{wcy} for $L_y=2$, we obtain further evidence supporting the conjecture

872: that for a strip graph of a regular lattice with periodic or twisted periodic

873: longitudinal boundary conditions (${\cal B}$ is the same for both of these),

874: the envelope curve of ${\cal B}$, i.e. the outermost portion of ${\cal B}$ for

875: a given $L_y$ surrounds the envelope curve for the ${\cal B}$ of the same strip

876: with a smaller value of $L_y$.  This was found to be true for the cyclic and

877: M\"obius strips of the square lattice for the cases $L_y=2,3$ and 4 (and, for

878: the cyclic case, also $L_y=1$) \cite{w,wcy,bcc}.  For strips of a regular

879: lattice $\Lambda$, as $L_y \to \infty$, one expects that the envelope curves

880: defined for each $L_y$ will approach a limiting curve, which is precisely the

881: envelope for the boundary ${\cal B}$ of the full 2D lattice $\Lambda$ defined

882: via this limit.

883: 

884: In region $R_1$, $\lambda_{t3,9}$ is the dominant $\lambda_j$ (with appropriate

885: choice of branch cut for evaluation away from the negative real axis), so 

886: \beq

887: W=(\lambda_{t3,9})^{1/3} \ , \quad q \in R_1 \ . 

888: \label{wt3r1}

889: \eeq

890: The fact that this is the same as $W$ for the (FBC$_y$,FBC$_x$) case 

891: \cite{strip} is a general result \cite{bcc}.  

892: In region $R_2$, $\lambda_{t3,4}$ is dominant, so

893: \beq

894: |W| = |\lambda_{t3,4}|^{1/3} \ , \quad q \in R_2

895: \label{wt3r2}

896: \eeq

897: (in regions other than $R_1$, only $|W|$ can be determined unambiguously

898: \cite{w}). In region $R_3$, $W$ is given by the largest (in magnitude) of the 

899: roots of the cubic (\ref{cubeq}), which we label $\lambda_{t3,6}$:

900: \beq

901: |W|=|\lambda_{t3,6}|^{1/3} \ , \quad q \in R_3 \ .

902: \label{wt3r3}

903: \eeq

904: In regions $R_4, R_4^*$, 

905: \beq

906: |W|=|\lambda_{t3,8}|^{1/3} \ , \quad q \in R_4 \ , R_4^* \ .

907: \label{wt3r4}

908: \eeq

909: where $\lambda_{t3,8}$ is the other among these cubic roots that has maximal

910: magnitude in this region.  In $R_5, \ R_5^*$, $|W|=|\lambda_{t3,6}|^{1/3}$. 

911: 

912: 

913: \section{$L_y=4$ Cyclic Triangular Strip}

914: 

915: We have succeeded in performing an exact calculation of the chromatic

916: polynomial for the cyclic triangular strip of arbitrarily great length and of

917: the next larger width, namely, $L_y=4$.  

918: The calculation is considerably more involved than for the $L_y=3$

919: cyclic strip, as is indicated by the number of $\lambda_j$ terms in

920: eq. (\ref{pgsum}), namely, $N_\lambda=26$, as contrasted with the value

921: $L_y=10$ for the $L_y=3$ cyclic strip.  Elsewhere we have proved that 

922: $N_\lambda$ is the same for the square and triangular strips of a given width 

923: and have given a general determination of $N_\lambda$ as a function of $L_y$ 

924: \cite{cf}.  To go beyond this width for this family of cyclic strips of the 

925: triangular lattice would be increasingly difficult, since the number of

926: terms $N_\lambda$ in (\ref{pgsum}) is 70, \ 192, \ and 534 for $L_y=5, \ 6,$ 

927: and 7, respectively. For $L_y=4$ we find the chromatic polynomial

928: \beq 

929: P(tri(4 \times m), FBC_y,PBC_x,q)=\sum_{j=1}^{26}

930: c_{t4,j}(\lambda_{t4,j})^m

931: \label{pt4}

932: \eeq

933: where one term is $\lambda_{t4,1}=1$ and the others, $\lambda_{t4,j}$, are 

934: the roots of

935: (i) the quartic equation (\ref{t4eq1}), for $2 \le j \le 5$; (ii) the quartic

936: equation (\ref{t4eq2}), for $6 \le j \le 9$; (iii) the degree-8 equation

937: (\ref{t4eq3}), for $10 \le j \le 17$; and (iv) the degree-9 equation

938: (\ref{t4eq4}), for $18 \le j \le 26$.  Since these equations are somewhat

939: lengthy, they are given in the appendix.  In contrast to the $L_y=2$ and

940: $L_y=3$ cyclic triangular strips, here it is not possible to solve for all of

941: the terms $\lambda_{t4,j}$ as algebraic roots since for $j=10$ through $j=26$

942: these are roots of equations of degree higher than quartic.  

943: The corresponding coefficients 

944: $c_{t4,j}$ are 

945: \beq

946: c_{t4,1}=c^{(4)}

947: \label{ct4j1}

948: \eeq

949: \beq

950: c_{t4,j}=c^{(0)} \quad {\rm for} \quad 2 \le j \le 5

951: \label{ct4j25}

952: \eeq

953: \beq

954: c_{t4,j}=c^{(3)} \quad {\rm for} \quad 6 \le j \le 9

955: \label{ct4j69}

956: \eeq

957: \beq

958: c_{t4,j}=c^{(2)} \quad {\rm for} \quad 10 \le j \le 17

959: \label{ct4j1017}

960: \eeq

961: \beq

962: c_{t4,j}=c^{(1)} \quad {\rm for} \quad 18 \le j \le 26  \ .

963: \label{ct4j1826}

964: \eeq

965: As before, one can, equivalently, list the generating function; however, since

966: all of the relevant information is present in (\ref{pt4}) with the requisite

967: definitions of the coefficients $c_{t4,j}$ and terms $\lambda_{t4,j}$, we shall

968: forego this. 

969: 

970: For the M\"obius $L_y=4$ strip of the triangular lattice, a general argument

971: \cite{pm} shows that the $\lambda_j$'s are the same as for the cyclic strip,

972: although the coefficients $c_j$ are different and are more complicated than

973: those for the cyclic strip, as was already shown by the $L_y=2$ triangular

974: strip, where two coefficients for the M\"obius case are algebraic nonpolynomial

975: functions of $q$.  As discussed before, since ${\cal B}$ is determined only by

976: the $\lambda_j$'s, it is the same for the cyclic and M\"obius strips of a given

977: lattice.

978: 

979: \begin{figure}[p]

980: %\vspace{-4cm}

981: \centering

982: \leavevmode

983: \epsfxsize=4.0in

984: \begin{center}

985: \leavevmode

986: \epsffile{tpxy4m16.ps}

987: \end{center}

988: %\vspace{-2cm}

989: \caption{\footnotesize{${\cal B}$ and chromatic zeros for the $L_y=4$,

990: $L_x=m=16$ (i.e., $n=64$) cyclic triangular strip graph.}}

991: \label{tricyczeros4}

992: \end{figure}

993: 

994: Chromatic zeros for the $L_y=4$, $L_x=m=16$ (hence $n=64$) cyclic graph of the

995: triangular lattice are shown in Fig. \ref{tricyczeros4}; the complex zeros give

996: an approximate indication of ${\cal B}$.  The locus ${\cal B}$ separates the

997: $q$ plane into six main regions, including four that contain intervals of the

998: real axis.  The outermost one, region $R_1$, extends to infinite $|q|$ and

999: includes the intervals $q \ge q_c$ and $q \le 0$ on the real $q$ axis.  Region

1000: $R_2$ includes the real interval $3 \le q \le q_c$; region $R_3$ includes the

1001: real interval $2 \le q \le 3$, and region $R_4$ includes the real interval $0

1002: \le q \le 2$.  There are also complex-conjugate regions $R_5$ and $R_5^*$

1003: centered at approximately $q \simeq 3.5 \pm 0.6i$.  There could also be other

1004: such complex-conjugate pairs of regions.  The density of zeros is observed to

1005: be smaller on the parts of ${\cal B}$ extending through $q=2$ and $q=3$ and

1006: also on the right-most bubble-like curves than on the rest of the of ${\cal

1007: B}$.  As before with the $L_y=3$ strip, it is straightforward to determine the

1008: various triple points.  The maximal point at which the locus ${\cal B}$ crosses

1009: the real axis is \beq q_c(tri(L_y=4),FBC_y,(T)PBC_x) \simeq 3.228126 \ .

1010: \label{qctcycly4}

1011: \eeq 

1012: Thus, for the triangular strips with $(FBC_y,(T)PBC_x)$ boundary

1013: conditions for which the chromatic polynomials have been calculated so far,

1014: i.e. for the widths $L_y=2,3$, and 4, $q_c$ is a nondecreasing function of

1015: $L_y$: $q_c(tri(2))=3$, $q_c(tri(3))=3$, and $q_c(tri(4))=3.228...$.  We have

1016: found that this is also true for the strips of the square lattice with the same

1017: cyclic or M\"obius boundary conditions: $q(sq(1))=q_c(sq(2))=2$, $q_c(sq(3))

1018: \simeq 2.34$, and $q_c(sq(4)) \simeq 2.49$.  This contrasts with the

1019: non-monotonic behavior of $q_c$ as a function of $L_y$ that we have found for

1020: strips with $(PBC_y,FBC_x)$ boundary conditions (see eq. (\ref{qctpy5}) below).

1021: It is anticipated that for cyclic and M\"obius strips of the triangular

1022: lattice, as $L_y$ increases beyond 3, $q_c$ will approach the 2D value

1023: $q_c(tri)=4$.

1024: 

1025: 

1026: Another important feature of the locus ${\cal B}$ for the $L_y=4$ cyclic 

1027: triangular strip, which is the same as was true of the $L_y=2$ and $L_y=3$ 

1028: cyclic triangular strips, is that it crosses the real axis at the points 

1029: $q=0,2$ and 3.  This leads us to the conjecture

1030: \beq

1031: Conjecture: \quad {\cal B} \supset \{q=0, \ 2, \ 3 \}  \ \  {\rm for} \ \ 

1032: tri(L_y,FBC_y,(T)PBC_x) \quad \forall \ L_y \ge 2

1033: \label{bcrossq023}

1034: \eeq

1035: The crossing at $q=0$ is present in a very wide class of families of graphs 

1036: that contain global circuits \cite{bcc}. 

1037: 

1038: The crossing at $q=2$ that we have found for $L_y=2,3,4$ signals a

1039: zero-temperature critical point of the Ising antiferromagnet on these

1040: triangular strips, and similarly for the generalization (\ref{bcrossq023}) to

1041: arbitrary $L_y$.  Note that this $T=0$ critical point involves frustration,

1042: since not all of the spin-spin interactions around a triangle can have their

1043: energies minimized.  Frustration is also responsible for the $T=0$ critical

1044: point of the Ising antiferromagnet on the full 2D triangular lattice

1045: \cite{wannier,stephenson}, but the nature of the critical singularities in the

1046: free energy, correlation length, etc. are different: they are algebraic for the

1047: triangular lattice, but are essential singularities for the $L_y \times \infty$

1048: strips \cite{ta}.  As will be shown below, this crossing of ${\cal B}$ at $q=2$

1049: is also found for the $L_y=2$ strips with torus or Klein bottle boundary

1050: conditions, i.e.  $(PBC_y,PBC_x)$ or $(PBC_y,TPBC_x)$.  Thus, it appears to be

1051: a general feature of the strip graphs of the triangular lattice with periodic

1052: or twisted periodic longitudinal boundary conditions.  Indeed, the respective

1053: boundaries ${\cal B}$ for the $L_y=2$ cyclic and M\"obius strips of the square,

1054: triangular, and kagom\'e lattices, and for the $L_y=3$ cyclic and M\"obius

1055: strips of the square lattice \cite{wcy} and the torus and Klein bottle strips

1056: of the square lattice \cite{tk} also cross the real $q$ axis at $q=2$ (as well

1057: as $q=0$).

1058: 

1059: In contrast, for strips with $(FBC_y,FBC_x)$ (open) or $(PBC_y,FBC_x)$

1060: (cylindrical) boundary conditions, ${\cal B}$ does not pass through $q=0$ and,

1061: while ${\cal B}$ for the $L_y=3$ open square strip crosses the real axis at

1062: $q=2$, the respective loci ${\cal B}$ for the $L_y=4,5$ open and cylindrical

1063: square strips and the open and cylindrical triangular strips with $L_y=3,4,5$

1064: do not, as discussed further below.

1065: 

1066: In regions $R_j$, $1 \le j \le 4$, the dominant terms are, respectively, the

1067: root of maximal magnitude of (1) the first quartic equation (\ref{t4eq1}), 

1068: the second quartic equation (\ref{t4eq2}), (3) the eighth-degree equation 

1069: (\ref{t4eq3}), (4) the ninth-degree equation (\ref{t4eq4}).  In region $R_1$, 

1070: we label the dominant term as $\lambda_{t4,R1}$, and so forth for the other

1071: regions. In the complex-conjugate regions $R_5, R_5^*$ 

1072: the dominant term is the largest-magnitude root of the ninth-degree equation 

1073: (\ref{t4eq4}), which we label as $\lambda_{t4,R5}$.  We have 

1074: \beq 

1075: W=(\lambda_{t4,R1})^{1/4} \ , \quad q \in R_1

1076: \label{wt4r1}

1077: \eeq

1078: \beq

1079: |W| = |\lambda_{t4,Rj}|^{1/4} \ , \quad q \in R_j \ , j=2,3,4

1080: \label{wt4r23}

1081: \eeq

1082: and 

1083: \beq

1084: |W| = |\lambda_{t4,R5}|^{1/4} \ , \quad q \in R_5, R_5^* \ . 

1085: \label{wt4r45}

1086: \eeq

1087: As usual, the boundaries of the regions are defined by the degeneracies, in 

1088: magnitude, of the leading $\lambda_{t4,j}$'s in these regions.  For example, 

1089: on the real axis, one has $|\lambda_{t4,R1}|=|\lambda_{t4,R2}|$ at $q_c$, 

1090: $|\lambda_{t4,R2}|=|\lambda_{t4,R3}|$ at $q=3$, $|\lambda_{t4,R3}|=

1091: |\lambda_{t4,R4}|$ at $q=2$, and $|\lambda_{t4,R4}|=|\lambda_{t4,R1}|$ at

1092: $q=0$.  

1093: 

1094: In contrast to the situation for the $L_y=2$ and $L_y=3$ cyclic strips of the

1095: triangular lattice, here $q_c$ is a regular instead of a multiple (=

1096: intersection) point on the algebraic curve forming ${\cal B}$, in the

1097: terminology of algebraic geometry (a multiple point on an algebraic curve is a

1098: point where several branches of the curve cross each other). 

1099: 

1100: Another qualitative difference is that for $L_y=2$ and $L_y=3$, ${\cal B}$

1101: contained support only for $Re(q) \ge 0$, and the only point on ${\cal B}$ with

1102: $Re(q)=0$ was the origin itself; however, for $L_y=4$, ${\cal B}$ extends

1103: slightly into the half-plane with $Re(q) < 0$.  Although these lattice strips

1104: involve global circuits, it was found in \cite{hs} that this is not a necessary

1105: condition for chromatic zeros and loci ${\cal B}$ of lattice strips (or

1106: homeomorphic expansions thereof) to have support for $Re(q) < 0$; this was also

1107: evident in results of \cite{baxter}, ruling out a conjecture in \cite{strip}.

1108: In the case of strips of the square lattice we had found that for $L_y=1$ and

1109: 2, ${\cal B}$ only had support for $Re(q) \ge 0$, but for $L_y=3$ and $L_y=4$

1110: it had support also for $Re(q) < 0$.

1111: 

1112: Finally, another general feature is that the motion of the outermost curves

1113: forming part of ${\cal B}$ for $L_y=2,3,4$ is consistent with the inference

1114: that as $L_y$ increases, these form a limiting curve.  

1115: 

1116: \section{$L_y=3$ Triangular Lattice Strips with Torus and Klein Bottle 

1117: Boundary Conditions}

1118: 

1119: For the $L_y=3$ triangular lattice strips with torus boundary conditions 

1120: (denoted with the subscript $tt$), we calculate, by the same methods as above, 

1121: \beq

1122: P(tri(L_y=3)_m,PBC_y,PBC_x,q) = \sum_{j=1}^{11} c_{tt,j} (\lambda_{tt,j})^m

1123: \label{ptorus}

1124: \eeq

1125: where

1126: \beq

1127: \lambda_{tt,1}=-2 \ , \qquad c_{tt,1} = \frac{1}{3}(q-1)(q^2-5q+3) \ ,

1128: \label{lamtt1}

1129: \eeq

1130: 

1131: \beq

1132: \lambda_{tt,2}=q-2 \ , \qquad c_{tt,2}=\frac{1}{2}(q-1)(q-2) \ ,

1133: \label{lamtt2}

1134: \eeq

1135: 

1136: \beq

1137: \lambda_{tt,3}=3q-14 \ , \qquad c_{tt,3}=\frac{1}{2}q(q-3) \ ,

1138: \label{lamtt3}

1139: \eeq

1140: 

1141: \beq

1142: \lambda_{tt,4}=-2(q-4)^2 \ , \qquad c_{tt,4}=q-1 \ , 

1143: \label{lamtt4}

1144: \eeq

1145: 

1146: \beq

1147: \lambda_{tt,5}=q^3-9q^2+29q-32 \ , \qquad c_{tt,5}=1  \ ,

1148: \label{lamtt5}

1149: \eeq

1150: 

1151: \beq

1152: \lambda_{tt,(6,7)}=e^{\pm 2\pi i/3} \ , \qquad c_{tt,6} = c_{tt,7} = 

1153: \frac{1}{6}q(q-1)(2q-7) \ ,

1154: \label{lamtt67}

1155: \eeq

1156: 

1157: \beq

1158: \lambda_{tt,(8,9)}=(q^2-5q+7)e^{\pm 2\pi i/3} \ , \qquad 

1159: c_{tt,8}=c_{tt,9}=q-1 \ , 

1160: \label{lamtt89}

1161: \eeq

1162: 

1163: \beq

1164: \lambda_{tt,(10,11))}=-(2q-7)e^{\pm 2\pi i/3} \ , \qquad c_{tt,10}=c_{tt,11}=

1165: \frac{1}{2}(q-1)(q-2) \ . 

1166: \label{lamtt1011}

1167: \eeq

1168: Note that although several of the coefficients are complex (and have

1169: coefficients that are not integers), the chromatic polynomial itself is, of

1170: course, a polynomial with integer coefficients. 

1171: The sum of the complex terms can be written as

1172: \beq

1173: \sum_{j=6}^{11} c_{tt,j} (\lambda_{tt,j})^m  = 

1174: (q-1)\Bigl [ \frac{q(2q-7)}{3} + 2(q^2-5q+7)^m + (q-2)[-(2q-7)]^m \Bigr ] 

1175: \cos \Bigl (\frac{2m \pi}{3} \Bigr ) 

1176: \label{osctermsum}

1177: \eeq

1178: None of these terms with $6 \le j \le 11$ can be dominant anywhere in the $q$

1179: plane since this would imply that the limit (\ref{w}) would not exist. 

1180: As compared with previously calculated strip graphs, this exhibits a number of

1181: qualitatively new features: (i) previously, only one $\lambda_j$ was a

1182: constant, independent of $q$, and this was always equal to either 1 or $-1$;

1183: here there are three, and none of these is equal to $\pm 1$; (ii) previously,

1184: all of the $\lambda_j$'s were either polynomials with real coefficients or

1185: algebraic functions of polynomials with real coefficients.  By general 

1186: arguments given before \cite{bcc}, the dominant $\lambda_j$ in region $R_1$ has

1187: as its highest power $q^{n/m}$, i.e., in the present case, $q^3$, and has

1188: coefficient 1.  We find that the generating function for this case has 

1189: $deg_x({\cal N})=9$ and $deg_x({\cal D})=11$, so that all of the $\lambda_j$'s

1190: in ${\cal D}$ contribute to $P$. 

1191: 

1192: Given that $\chi=4$ for $m=1$ or 2 mod 3 from eq. (\ref{chitrimbklein}), it 

1193: follows that for these $m$ values,

1194: and $q=3$, $P(tri(L_y=3)_m,PBC_y,PBC_x,q=3)=0$; however, $\chi=3$ for $m=0$ mod

1195: 3; in this case, we find (for $m \ge 3$)

1196: \beq

1197: P(tri(L_y=3)_m,PBC_y,PBC_x,q=3)=6 \quad {\rm for} \quad m = 0 \quad mod \ 3 \ .

1198: \label{pttq3m0mod3}

1199: \eeq

1200: This is analogous to eq. (\ref{ptq3m0mod3}) above. 

1201: 

1202: \vspace{6mm}

1203: 

1204: For the $L_y=3$ triangular lattice strips with Klein bottle topology

1205: (denoted with the subscript $tk$) we calculate 

1206: \beqs

1207: & & P(tri(L_y=3)_m,PBC_y,TPBC_x,q) = \sum_{j=1}^5 c_{tk,j} (\lambda_{tk,j})^m 

1208: \cr\cr

1209: & & = -(q-1)(-2)^m-\frac{1}{2}(q-1)(q-2)(q-2)^m 

1210: + \frac{1}{2}q(q-3)(3q-14)^m + \cr\cr

1211: & & (q-1)[-2(q-4)^2]^m + (q^3-9q^2+29q-32)^m \ . 

1212: \label{ptk}

1213: \eeqs

1214: Thus, $c_{tk,j}=c_{tt,j}$, $j=1,..5$, and the six $\lambda_{tt,j}$, 

1215: $j=6,..11$ that involve complex factors do not contribute to 

1216: $P(tri(L_y=3)_m,PBC_y,TPBC_x,q)$.  Since none of these six $\lambda_{tt,j}$'s

1217: is dominant anywhere, it follows that ${\cal B}$ is the same for the 

1218: $L_y=3$ triangular lattice strips with torus and Klein bottle topology, just as

1219: was true of the analogous square lattice strips \cite{tk}, and in accord with 

1220: the general discussion of Ref. \cite{bcc}.  For this case the generating

1221: function has $deg_x({\cal N})=2$ and $deg_x({\cal D})=5$, so that, as was true

1222: of the $L_y=3$ M\"obius and torus strips of the triangular lattice, all of 

1223: the $\lambda_j$'s in ${\cal D}$ contribute to $P$.

1224: 

1225: 

1226: For the sum of the coefficients we find, for the triangular $L_y=3$ strip with

1227: torus boundary conditions 

1228: \beq

1229: C=P(K_3,q)=q(q-1)(q-2) \quad {\rm for} \quad G=tri(L_y=3) \quad {\rm with} 

1230: \quad (PBC_y,PBC_x)

1231: \label{csumtt}

1232: \eeq

1233: and for this strip with Klein bottle boundary conditions

1234: \beq

1235: C=0 \quad {\rm for} \quad G=tri(L_y=3) \quad {\rm with} \quad (PBC_y,TPBC_x) 

1236: \ . 

1237: \label{csumtk}

1238: \eeq

1239: These are the same respective values as those calculated in \cite{tk} for the 

1240: corresponding $L_y=3$ strips of the square lattice with torus and Klein bottle

1241: boundary conditions. 

1242: 

1243: \begin{figure}[p]

1244: %\vspace{-4cm}

1245: \centering

1246: \leavevmode

1247: \epsfxsize=4.0in

1248: \begin{center}

1249: \leavevmode

1250: \epsffile{tpxpy3.ps}

1251: \end{center}

1252: %\vspace{-2cm}

1253: \caption{\footnotesize{${\cal B}$ and chromatic zeros for the $L_y=3$,

1254: $L_x=m=20$ (i.e., $n=60$) triangular strip graph with torus boundary

1255: conditions.}}

1256: \label{tritorzeros}

1257: \end{figure}

1258: 

1259: Chromatic zeros for the $L_y=3$, $L_x=m=20$ torus graph of the triangular

1260: lattice are shown in Fig. \ref{tritorzeros}; as before, with this value of

1261: $m$, the chromatic zeros for the torus and Klein bottle boundary conditions are

1262: generally similar and both sets of chromatic zeros lie close to the boundary

1263: ${\cal B}$ and indicate its position.  The locus ${\cal B}$ and the $W$

1264: functions are the same for the torus and Klein bottle graph families.  We find

1265: \beq

1266: q_c \simeq 3.7240756... \quad {\rm for} \quad \{G\} = tri(L_y=3) \quad 

1267: {\rm with} \quad (PBC_y,PBC_x) \ {\rm or} \ (PBC_y,TPBC_x) \ . 

1268: \label{qcttorus}

1269: \eeq

1270: This value of $q_c$ is given by the real root of the degeneracy equation 

1271: $q^3-9q^2+32q-46=0$. It is interesting that this is just 7 \% below the value

1272: for the infinite 2D triangular lattice, $q_c(tri)=4$.  Thus, for the width 

1273: $L_y=3$ strips where a comparison can be made, the 

1274: use of both periodic transverse boundary conditions and periodic or twisted 

1275: periodic longitudinal boundary conditions expedites the approach to the 2D

1276: thermodynamic limit, in the sense that it yields a value of $q_c$ that is

1277: substantially closer to the 2D value than was obtained for the same width strip

1278: with free transverse boundary conditions in eq. (\ref{qctcycly3}).  This is

1279: expected since for the torus or Klein bottle boundary conditions (a) the

1280: resultant graphs are $\Delta$-regular\footnote{\footnotesize{A $\Delta$-regular

1281: graph is a graph all of whose vertices have the same degree, $\Delta$, where

1282: the degree of a vertex is defined as the number of edges connected to it.}}

1283: with the degree (coordination number) $\Delta=6$ of the 2D triangular 

1284: lattice, and (b) there are no boundaries to the surface on which the graphs are

1285: embedded.  In contrast, for the boundary conditions of types (i)-(iv) (aside

1286: from degenerate cases) the graphs are not $\Delta$-regular and they do have

1287: boundaries.  

1288: 

1289: The locus ${\cal B}$ has support for $Re(q) \ge 0$ and

1290: separates the $q$ plane into three regions.  The outermost one, region $R_1$,

1291: extends to infinite $|q|$ and includes the intervals $q \ge q_c$ and $q \le 0$

1292: on the real $q$ axis.  Region $R_2$ includes the real interval $2 \le q \le

1293: q_c$ and extends upward and downward to the complex conjugate triple points on

1294: ${\cal B}$ at $q_t, q_t^* \simeq 4.0 \pm 1.7i$.  Region $R_3$ is the innermost

1295: one and includes the real interval $0 \le q \le 2$.  The maximum value of

1296: $|Im(q)|$ on the boundary between $R_1$ and $R_3$ is about 3.0, occurring at

1297: $Re(q) \simeq 2.6$.  The maximum value of $|q|$ on ${\cal B}$ is roughly 4.6,

1298: occurring at $q \simeq 3.9 \pm 2.5i$.  The boundary between $R_2$ and $R_3$

1299: curves to the right as one increases $|Im(q)|$, extending from $q=2$ upward to

1300: $q_t$ and downward to $q_t^*$.  As is evident in Fig. \ref{tritorzeros}, the

1301: density of chromatic zeros is highest on the $R_1-R_2$ boundary near $q_c$.

1302: The feature that ${\cal B}$ has support only for $Re(q) \ge 0$ is the same as

1303: was found \cite{tk} for the $L_y=3$ strips of the square lattice with torus and

1304: Klein bottle topology. 

1305: 

1306: 

1307: In region $R_1$, $\lambda_{tt,5}=\lambda_{tk,5}$ is the dominant 

1308: $\lambda_j$, so 

1309: \beq

1310: W = (q^3-9q^2+29q-32)^{1/3} \ , \quad q \in R_1 \ . 

1311: \label{wr1}

1312: \eeq

1313: The fact that this is the same as $W$ for the (PBC$_y$,FBC$_x$) case,

1314: eq. (\ref{wsqly3pbc}), is a general result \cite{bcc}.  The importance of the

1315: PBC$_y$ is evident from the fact that for the same width of three squares, the

1316: strip with (FBC$_y$,FBC$_x$) yields a different $W$ \cite{strip}.

1317: 

1318: In region $R_2$ $\lambda_{tt,3}=\lambda_{tk,3}$ is dominant, so

1319: \beq

1320: |W| = |3q-14|^{1/3} \ , \quad q \in R_2

1321: \label{wr2}

1322: \eeq

1323: (in regions other than $R_1$, only $|W|$ can be determined unambiguously 

1324: \cite{w}). In region $R_3$, $\lambda_{tt,4}=\lambda_{tk,4}$ is dominant, so 

1325: \beq

1326: |W|=|2(q-4)^2|^{1/3} \ , \quad q \in R_3 \ . 

1327: \label{wr3}

1328: \eeq 

1329: At all of the three points, $q=0,2$, and $q=q_c \simeq 3.72$ where

1330: ${\cal B}$ crosses the real $q$ axis, it does so vertically.  The present

1331: results are in accord with the inference \cite{strip,wcy} that for a recursive

1332: graph with regular lattice structure, a sufficient condition for

1333: ${\cal B}$ to separate the $q$ plane into two or more regions is that it

1334: contains a global circuit, i.e. a path along a lattice direction whose length

1335: goes to infinity as $n \to \infty$; here this is equivalent to PBC$_x$. 

1336: 

1337: Our calculations of the zero-temperature Potts antiferromagnet partition

1338: functions (chromatic polynomials) and exponential of the entropy, $W$, for

1339: $L_y=3$ strips of the triangular lattice with various boundary conditions

1340: including free and periodic or reversed-orientation periodic 

1341: thus elucidate the role that these boundary conditions and the

1342: associated topologies play.

1343: One particular feature of note is that the torus and Klein bottle graphs have 

1344: interestingly different chromatic polynomials, with different $N_\lambda$, 

1345: but the $W$ functions and hence the boundaries ${\cal B}$ are the same.

1346: 

1347: 

1348: \section{Width $L_y=5,6$ Strips of the Triangular Lattice with $(PBC_y,FBC_x)$}

1349: 

1350: 

1351: We first briefly review the cases $L_y=3,4$ \cite{strip2,w2d,bcc}.

1352: The chromatic polynomial for the $L_y=3$ triangular $(t)$ strip with 

1353: $n=3(m+2)$ vertices (following the labelling conventions in \cite{strip})

1354: and $(PBC_y,FBC_x)$ boundary conditions, denoted $t3PF$ in 

1355: the subscripts, has $N_\lambda=1$, 

1356: \beq

1357: \lambda_{t3PF}=q^3-9q^2+29q-32 \ , 

1358: \label{lamtpy3}

1359: \eeq

1360: \beq

1361: P(tri(3 \times m,PBC_y,FBC_x),q)=q(q-1)(q-2)(\lambda_{t3PF})^{m+1} \ , 

1362: \label{ptpy3}

1363: \eeq

1364: and 

1365: \beq

1366: W(tri(L_y=3),PBC_y,FBC_x,q) = (q^3-9q^2+29q-32)^{1/3}

1367: \label{wsqly3pbc}

1368: \eeq

1369: with ${\cal B}=\emptyset$.

1370: 

1371: For the $L_y=4$ strip with $(PBC_y,FBC_x)$ boundary conditions and 

1372: $n=4(m+2)$ vertices, one had \cite{strip2,w2d} $N_\lambda=2$ and 

1373: \beq

1374: P(tri(4 \times m,PBC_y,FBC_x),q)=\sum_{j=1}^2 c_{t4PF,j}

1375: (\lambda_{t4PF,j})^{m+1}

1376: \label{ptri4pf}

1377: \eeq

1378: where

1379: \beq

1380: \lambda_{t4PF,j} = \frac{(q-3)}{2}\biggl [ T_{t4PF} \pm

1381: (q-4)\sqrt{R_{t4PF}} \biggr ]

1382: \label{lamtpy4}

1383: \eeq

1384: where 

1385: \beq

1386: T_{t4PF} = q^3-9q^2+33q-48

1387: \label{tt4pf}

1388: \eeq

1389: \beq

1390: R_{t4PF}=q^4-10q^3+43q^2-106q+129 \ . 

1391: \label{rt4pf}

1392: \eeq

1393: In this case, as is evident in Fig. 4(a) of \cite{strip2}), ${\cal B}$ includes

1394: both arcs and a closed oval invariant under complex conjugation which crossed

1395: the real axis where the prefactor of the square root vanished, i.e. at

1396: $q_c=4$, and at the real root of $T_{t4PF}$, namely $q \simeq 3.481406$.  There

1397: are two regions, $R_1$ and $R_2$, the exterior and interior of the oval.  We

1398: have 

1399: \beq

1400: W(tri, 4 \times \infty, PBC_y,q) = (\lambda_{t4PF,1})^{1/4} \ , \quad q \in R_1

1401: \label{wt4pfr1}

1402: \eeq

1403: and

1404: \beq

1405: W(tri, 4 \times \infty, PBC_y,q) = (\lambda_{t4PF,2})^{1/4} \ , \quad q \in R_2

1406: \label{wt4pfr2}

1407: \eeq

1408: with appropriate choices of branch cuts.

1409: 

1410: 

1411: We have calculated new results for the width $L_y=5$ and $L_y=6$ strips with 

1412: $(PBC_y,FBC_x)$ boundary conditions and $n=L_y(m+2)$ vertices.  For $L_y=5$ we 

1413: find $N_\lambda=2$ and 

1414: \beq

1415: P(tri(5 \times m,PBC_y,FBC_x),q)=\sum_{j=1}^2 c_{t5PF,j}

1416: (\lambda_{t5PF,j})^{m+1}

1417: \label{ptri5pf}

1418: \eeq

1419: where

1420: \beq

1421: \lambda_{t5PF,j} = \frac{1}{2}\biggl [ T_{t5PF} \pm

1422: S_{t5PF}\sqrt{R_{t5PF}} \biggr ]

1423: \label{lamtpy5}

1424: \eeq

1425: where

1426: \beq

1427: T_{t5PF}=q^5-15q^4+98q^3-355q^2+711q-614

1428: \label{tt5pf}

1429: \eeq

1430: 

1431: \beq

1432: R_{t5PF}=q^4-8q^3+26q^2-60q+85

1433: \label{rt5pf}

1434: \eeq

1435: and

1436: \beq

1437: S_{t5PF}=q^3-11q^2+43q-58 \ . 

1438: \label{st5pf}

1439: \eeq

1440: As before, the coefficients $c_{t5PF,j}$ can be computed using

1441: eq. (\ref{chrompgsumlam}) in terms of the generating function.  This generating

1442: function is given in the appendix.

1443: 

1444: Just as was true for the $L_y=4$ triangular lattice strip with $(PBC_y,FBC_x)$,

1445: in the present $L_y=5$ case ${\cal B}$ includes both a pair of complex

1446: conjugate arcs and an oval that crosses the real axis at two points. 

1447: Analogously to the previous case, one of these points is the real zero of 

1448: $T_{t5PF}$ at

1449: \beq

1450: q_\ell \simeq  3.207224

1451: \label{qell}

1452: \eeq

1453: and the other is the real zero of the

1454: prefactor $S_{t5PF}$ in front of the square root, at

1455: \beq

1456: q= 3.284775 = q_c  \quad {\rm for} \quad \{G\}=tri(5 \times \infty,

1457: PBC_y,FBC_x) \ . 

1458: \label{qctpy5}

1459: \eeq

1460: Taking the approximate center of the oval as the average of the left and right

1461: crossings, $q_{center}=(q_\ell+q_c)$, one finds that

1462: \beq

1463: q_{center} \simeq 3.2460

1464: \label{qcenter}

1465: \eeq

1466: near to the Tutte-Beraha number $B_7 \simeq 3.246980..$, where \cite{tutte,bkw}

1467: \beq

1468: B_r = 2 + 2\cos \Bigl ( \frac{2\pi}{r} \Bigr ) \ . 

1469: \label{br}

1470: \eeq 

1471: Chromatic zeros near $B_7$ were also noted in \cite{baxter} for these

1472: types of strips.  Note that $q_c$ decreases from 4 to the above value in

1473: eq. (\ref{qctpy5}) as $L_y$ increases from 4 to 5.  Hence, in contrast to our

1474: calculations of strips of the square and triangular lattices with

1475: $(FBC_y,FBC_x)$, $(FBC_y,(T)PBC_x)$, and $(PBC_y,(T)PBC_x)$, where in all cases

1476: considered, the respective $q_c$'s were nondecreasing functions of $L_y$, here

1477: we find that this is not the case for boundary conditions of the type

1478: $(PBC_y,FBC_x)$.

1479: 

1480: 

1481: \begin{figure}[hbtp]

1482: \centering

1483: \leavevmode

1484: \epsfxsize=2.5in

1485: \begin{center}

1486: \leavevmode

1487: \epsffile{tpy5.ps}

1488: \end{center}

1489: \caption{\footnotesize{Locus ${\cal B}$ for the width $L_y=5$ strip (tube) of

1490: the triangular lattice with $(PBC_y,FBC_x)$ boundary conditions.  Thus, the

1491: cross sections of the tube form pentagons.  For comparison, chromatic zeros

1492: calculated for the strip length $L_x=m+2=16$ (i.e., $n=80$ vertices) are shown.

1493: }}

1494: \label{tpy5}

1495: \end{figure}

1496: 

1497: Chromatic zeros and ${\cal B}$ for this strip are shown in Fig. \ref{tpy5};

1498: ${\cal B}$ separates the $q$ plane into two regions, $R_1$, the exterior, 

1499: and $R_2$, the interior, of the oval shown in the figure. 

1500: Comparing the ovals for the $L_y=4$ strip (Fig. 4(a) of \cite{strip2}) and the

1501: $L_y=5$ strip (Fig. \ref{tpy5} here), one sees that the oval shrinks in size.

1502: We have 

1503: \beq

1504: W(tri, 5 \times \infty, PBC_y,q) = (\lambda_{t5PF,1})^{1/5} \ , \quad q \in R_1

1505: \label{wt5pfr1}

1506: \eeq

1507: and 

1508: \beq

1509: W(tri, 5 \times \infty, PBC_y,q) = (\lambda_{t5PF,2})^{1/5} \ , \quad q \in R_2

1510: \label{wt5pfr2}

1511: \eeq

1512: with appropriate choices of branch cuts. 

1513: 

1514: \vspace{6mm}

1515: 

1516: For the strip of with $L_y=6$ and $(PBC_y,FBC_x)$ we find $N_\lambda=5$ and

1517: \beq

1518: P(tri(6 \times m,PBC_y,FBC_x),q)=\sum_{j=1}^5 c_{t6PF,j}

1519: (\lambda_{t6PF,j})^{m+1}

1520: \label{ptri6pf}

1521: \eeq

1522: Here the $\lambda_{t6PF}$'s are the roots of a fifth-order equation, so we

1523: cannot solve for them in terms of algebraic expressions as was possible for

1524: previous strips of this type with $L_y=3,4,5$.  We give the generating function

1525: for this strip in the appendix.  

1526: 

1527: \begin{figure}[hbtp]

1528: \centering

1529: \leavevmode

1530: \epsfxsize=2.5in

1531: \begin{center}

1532: \leavevmode

1533: \epsffile{tpy6.ps}

1534: \end{center}

1535: \caption{\footnotesize{Locus ${\cal B}$ for the width $L_y=6$ strip (tube) of

1536: the triangular lattice with $(PBC_y,FBC_x)$ boundary conditions.  Thus, the

1537: cross sections of the tube form hexagons.  For comparison, chromatic zeros

1538: calculated for the strip length $L_x=m+2=16$ (i.e., $n=96$ vertices) are shown.

1539: }}

1540: \label{tpy6}

1541: \end{figure}

1542: 

1543: Chromatic zeros for this strip are shown in Fig. \ref{tpy6} for $L_x=m+2=16$, 

1544: i.e., $n=96$; this length is sufficiently great that the chromatic zeros 

1545: give an approximate indication of the location of the locus ${\cal B}$.  The

1546: locus thereby inferred appears to consist of a single connected set of curves 

1547: and crosses the real axis.  From our exact analytic expressions, we calculate 

1548: the this crossing point to be 

1549: \beq

1550: q_c = 3.252419 \quad {\rm for} \quad \{G\}=tri(6 \times \infty,PBC_y,FBC_x) \ .

1551: \label{qctpy6}

1552: \eeq 

1553: The morphology of chromatic zeros for this long $6 \times 16$ cylindrical

1554: strip is similar to that found for a $8 \times 8$ patch with cylindrical

1555: boundary conditions in \cite{baxter}.  In both cases, one can figuratively

1556: think of how the inferred loci ${\cal B}$ can be modified to yield the locus

1557: for the $L_y=\infty$ strip with $(PBC_y,FBC_x)$ cylindrical boundary

1558: conditions: one (i) pulls the left-hand endpoints further to the left (into the

1559: $Re(q) < 0$ half-plane) and around so that they meet in a cusp at the origin,

1560: $q=0$; (ii) pulls the right-hand endpoints of the prongs over and around so

1561: that they meet in a cusp at $q=4$; (iii) shifts the crossing curve slightly to

1562: the right so that it crosses the real axis at $q \simeq 3.82$; and (iv) makes

1563: minor shifts of the rest of the locus ${\cal B}$ so as to obtain ${\cal B}$ for

1564: the cylindrical strip with $L_y=\infty$.  In this respect, just as was

1565: discussed in \cite{w2d}, one sees that the periodic transverse boundary

1566: conditions help to minimize finite-size effects.  Clearly, for a fixed $L_y$,

1567: with $L_x \to \infty$, finite-size effects are reduced the most for $(PBC_y,

1568: (T)PBC_x)$, next-most for $(PBC_y,FBC_x)$ and $(FBC_y,(T)PBC_x)$, and least for

1569: $(FBC_y,FBC_x)$.  We find

1570: 

1571: \beq

1572: W(tri,6 \times \infty,PBC_y,q)=(\lambda_{t6PF,max})^{1/6} \quad {\rm for} \ \

1573: q \in R_1

1574: \label{wt6pf}

1575: \eeq

1576: where $\lambda_{t6PF,max}$ denotes the root with the maximal magnitude. 

1577: 

1578: It is of interest to use these results to study the

1579: approach of $W$ to the limit for the full infinite 2D triangular lattice,

1580: extending the work of \cite{w2d}.  In Table \ref{tpf} we list

1581: the various values of $W(tri(L_y \times \infty, PBC_y, FBC_x),q)$, denoted as

1582: $W(L_y,q)$ to save space, together with the corresponding values of $W$ for the

1583: full 2D triangular lattice, $W(tri,q)$, obtained via numerical evaluation of

1584: an infinite product representation in \cite{baxter} as checked by series

1585: expansions and rigorous bounds \cite{wn}, and the ratio $R_W(tri(L_y \times

1586: \infty),PBC_y,FBC_x,q)$, where 

1587: \beq

1588: R_W(tri(L_y \times \infty),BC_y,BC_x,q) = \frac{W(tri(L_y \times \infty),

1589: BC_y,BC_x,q)}{W(tri,q)} \ . 

1590: \label{rw}

1591: \eeq

1592: This extends the previous study in \cite{w2d}.  Recall that for $q \ge 4$, 

1593: $W(tri(L_y \times \infty),BC_y,BC_x,q)$ is independent of $BC_x$

1594: \cite{w2d,bcc}.  Evidently, the approach of 

1595: $W$ for the infinite-length finite-width triangular strips to the 2D 

1596: thermodynamic limit as $L_y$ increases is quite rapid; for moderate values of 

1597: $q$, say 6 or 7, the ratio $R_W(tri(5 \times\infty),PBC_y,FBC_x,q)$ is equal

1598: to 1 to within approximately $10^{-3}$ or better, and the ratio 

1599: $R_W(tri(6 \times\infty),PBC_y,FBC_x,q)$ is at least as close to 1.  As was 

1600: proved in \cite{w2d}, the approach is non-monotonic for $PBC_y$ (and 

1601: monotonic for $FBC_y$). 

1602: 

1603: \begin{table}

1604: \caption{\footnotesize{Comparison of values of $W(tri(L_y),PBC_y,q)$ with

1605: $W(tri,q)$ for $4 \le q \le 10$ and $BC_x=FBC_x$ or $(T)PBC_x$. For each value

1606: of $q$, the quantities in the upper line are identified at the top and the

1607: quantities in the lower line are the values of $R_W(tri(L_y),PBC_y,q)$.  The

1608: $PBC_y$ is symbolized as $P_y$ in the table.}}

1609: \begin{center}

1610: \begin{tabular}{|c|c|c|c|c|c|}

1611: \hline\hline

1612: $q$ & $W(tri(3),P_y,q)$ & $W(tri(4),P_y,q)$ & $W(tri(5),P_y,q)$ & 

1613: $W(tri(6),P_y,q)$ & $W(tri,q)$ \\

1614: \hline\hline

1615: 4  & 1.58740   & 1.18921    & 1.39252  & 1.49603  &  1.46100   \\

1616:    & 1.0865    & 0.8140     & 0.9531   & 1.0240   &  1         \\ \hline

1617: 5  & 2.35133   & 2.21336    & 2.26877  & 2.26894  &  2.26411   \\

1618:    & 1.0385    & 0.9776     & 1.0021   & 1.0021   &  1         \\ \hline

1619: 6  & 3.23961   & 3.185055   & 3.20718  & 3.20399  &  3.20388   \\

1620:    & 1.0112    & 0.9941     & 1.0010   & 1.0000   &  1         \\ \hline

1621: 7  & 4.17934   & 4.15965    & 4.16987  & 4.16805  &  4.16819   \\

1622:    & 1.0027    & 0.99795    & 1.0004   & 1.0000   &  1         \\ \hline

1623: 8  & 5.14256   & 5.13936    & 5.14446  & 5.14348  &  5.14358   \\

1624:    & 0.99980   & 0.9992     & 1.0002   & 1.0000   &  1         \\ \hline

1625: 9  & 6.11803   & 6.12324    & 6.12587  & 6.12533  &  6.12539   \\

1626:    & 0.99880   & 0.99965    & 1.0001   & 1.0000   &  1         \\ \hline

1627: 10 & 7.10059   & 7.11027    & 7.11161  & 7.11131  &  7.11134   \\

1628:    & 0.99849   & 0.99985    & 1.0000   & 1.0000   &  1       \\ \hline\hline

1629: \end{tabular}

1630: \end{center}

1631: \label{tpf}

1632: \end{table}

1633: 

1634: 

1635: \section{Width $L_y=5$ Strips of the Triangular Lattices with 

1636: $(FBC_y,FBC_x)$}

1637: 

1638: Previous studies have been published on strips of various lattices with

1639: $(FBC_y,FBC_x)$ boundary conditions \cite{strip,strip2,hs}.  In the case of the

1640: the square and triangular lattice, these went up to $L_y=4$ and involved

1641: $\lambda_j$'s that were roots of a cubic and quartic equation, respectively.

1642: Although one can calculate chromatic polynomials for wider strips, the analysis

1643: is more cumbersome if the equations defining the $\lambda_j$'s are higher than

1644: quartic, so that one cannot solve for these $\lambda_j$'s as analytic

1645: closed-form algebraic expressions.  A study of these on wider open strips is in

1646: \cite{ss}.  As an illustration of this sort of situation, here we present a

1647: calculation of chromatic polynomials for the strip of the triangular lattice

1648: with width $L_y=5$ and $(FBC_y,FBC_x)$.  For this width, the $\lambda_j$'s are

1649: solutions of a degree-9 equation and hence cannot be solved for as algebraic

1650: roots.  The method of calculation is again the iterated use of the

1651: deletion-contraction theorem.  In \cite{strip}, a given strip $(G_s)_m$ was

1652: constructed by $m$ successive additions of a subgraph $H$ to an endgraph $I$;

1653: here, $I=H$, so that, following the notation of \cite{strip}, the total length

1654: of the strip graph $(G_s)_m$ is $m+2$ vertices, or equivalently, $m+1$ edges in

1655: the longitudinal direction.  The results are conveniently expressed in terms of

1656: the coefficient functions in the generating function, as discussed above.  For

1657: the width $L_y=5$ open strip of the triangular lattice we find $deg_z({\cal

1658: D})= N_\lambda = 9$.  The coefficient functions $b_{t5FF,j}$ and $A_{t5FF,j}$

1659: (cf. eqs. (\ref{d}) and (\ref{n})) are listed in the Appendix, where $FF$ is

1660: short for $(FBC_y,FBC_x)$.

1661: 

1662: \begin{figure}[hbtp]

1663: \centering

1664: \leavevmode

1665: \epsfxsize=2.5in

1666: \begin{center}

1667: \leavevmode

1668: \epsffile{ty5.ps}

1669: \end{center}

1670: \caption{\footnotesize{Chromatic zeros for the $L_y=5$ open strip of the

1671: triangular lattice of length $m+2=16$ vertices (i.e. total number of vertices

1672: $n=80$).}}

1673: \label{ty5}

1674: \end{figure}

1675: 

1676: In Fig. \ref{ty5} we show a plot of chromatic zeros for the open strip of the

1677: triangular lattice with $L_y=5$ and length $m+2=16$ vertices, so that the strip

1678: has $n=80$ vertices in all.  From an analysis of the degeneracy of leading

1679: $\lambda_j$'s, we find the exact result 

1680: \beq

1681: q_c = 3 \quad {\rm for} \quad sq(5 \times \infty,FBC_y,FBC_x) \ . 

1682: \label{qcsqff}

1683: \eeq

1684: This is in agreement with the chromatic zeros for the finite strip shown in 

1685: Fig. \ref{ty5}.  Again, we can compare Fig. \ref{ty5} with the corresponding 

1686: plots for $L_y=2$ and $L_y=3$ (Fig. 5(a,b) of \cite{strip}), and, just as was

1687: true for the corresponding three open strips of the square lattice, this

1688: comparison shows that as $L_y$ increases, the arcs forming ${\cal B}$ elongate 

1689: and the arc endpoints nearest to the origin approach more closely to the 

1690: origin.  We recall that for the $L_y=4$ open triangular strip, no $q_c$ is

1691: defined since ${\cal B}$ does not cross the real axis.  

1692: 

1693: For $q > q_c$, we have, for the physical ground state degeneracy per site of

1694: the $q$-state Potts antiferromagnet,

1695: \beq

1696: W(t(5 \times \infty,FBC_x, FBC_y), q) = (\lambda_{tri,j,max})^{1/5}

1697: \label{wtri5}

1698: \eeq 

1699: where $\lambda_{sq,j,max}$ denotes the solution of eq. (\ref{xieq}) with

1700: the coefficients (\ref{btri1})-(\ref{btri9}) that has the maximal magnitude in

1701: region $R_1$.

1702: 

1703: As before, it is of interest to use this result to study the approach of $W$ to

1704: the limit for the full infinite 2D triangular lattice, extending the work of

1705: \cite{w2d} for this set of boundary conditions.  In Table \ref{tritable} we

1706: list the various values of $W(tri(L_y \times \infty, FBC_y, BC_x),q)$, denoted

1707: as $W(L_y,q)$ to save space, together with the corresponding values of $W$ for

1708: the full 2D triangular lattice, $W(tri,q)$ and the ratio

1709: $R_W(tri(L_y \times \infty),FBC_y,BC_x,q) = W(tri(L_y \times \infty),

1710: FBC_y,BC_x,q)/W(tri,q)$.  Recall that the value of $W$ for $q \ge 4$ is

1711: independent of $BC_x$ \cite{w2d}.  Again, the approach of $W$ for the

1712: infinite-length finite-width triangular strips to the 2D thermodynamic limit as

1713: $L_y$ increases is quite rapid.

1714: 

1715: 

1716: \begin{table}

1717: \caption{\footnotesize{Comparison of values of $W(tri(L_y \times \infty),q)$

1718: for $FBC_y$ and $BC_x=FBC_x$ or $(T)PBC_x$ with $W(tri,q) \equiv W(tri(\infty

1719: \times \infty),q)$ for $3 \le q \le 10$. For brevity of notation, we omit the

1720: $(FBC_y,BC_x)$ in the notation.  For each value of $q$, the quantities in the

1721: upper line are identified at the top and the quantities in the lower line are

1722: the values of $R_W(tri(L_y),q)$.}}

1723: \begin{center}

1724: \begin{tabular}{|c|c|c|c|c|c|c|}

1725: \hline\hline

1726: $q$ & $W(tri(2),q)$ & $W(tri(3),q)$ & $W(tri(4),q)$ & $W(tri(5),q)$ & 

1727: $W(tri(\infty),q)$

1728: & \\ \hline\hline

1729: 4  &   2    & 1.77173   & 1.67619   & 1.62270   & 1.46100  \\

1730:    & 1.369  & 1.213     & 1.147     & 1.111     & 1        \\ \hline

1731: 5  &   3    & 2.72998   & 2.60495   & 2.53251   &  2.26411 \\

1732:    & 1.325  & 1.206     & 1.151     & 1.1185    & 1        \\ \hline

1733: 6  &   4    & 3.71457   & 3.579715  & 3.50112   &  3.20388 \\

1734:    & 1.248  & 1.159     & 1.117     & 1.093     & 1        \\ \hline

1735: 7  &   5    & 4.70571   & 4.56515   & 4.48283   & 4.16819  \\

1736:    & 1.200  & 1.129     & 1.095     & 1.075     & 1        \\ \hline

1737: 8  &   6    & 5.69974   & 5.55530   & 5.47040   & 5.14358  \\

1738:    & 1.167  & 1.108     & 1.080     & 1.0635    & 1        \\ \hline

1739: 9  &   7    & 6.695395  & 6.54810   & 6.46129   & 6.12539  \\

1740:    & 1.143  & 1.093     & 1.069     & 1.055     & 1        \\ \hline

1741: 10 &   8    & 7.69208   & 7.54259   & 7.45430   & 7.11134  \\

1742:    & 1.125  & 1.082     & 1.061     & 1.048     & 1        \\ \hline\hline

1743: \end{tabular}

1744: \end{center}

1745: \label{tritable}

1746: \end{table}

1747: 

1748: \section{Comparative Discussion}

1749: 

1750: In this section we give a general discussion of the locus ${\cal B}$.  We have

1751: found several interesting features: 

1752: 

1753: \begin{enumerate}

1754: 

1755: \item

1756: 

1757: For the strips of the triangular lattice of width $L_y=3$ and $L_y=4$ with

1758: $(FBC_y,(T)PBC_x)$ and strips of width $L_y=3$ and $(PBC_y,(T)PBC_x)$ boundary

1759: conditions studied here, we have shown that the locus ${\cal B}$ encloses

1760: regions of the $q$ plane including certain intervals on the real axis and

1761: passes through $q=0$ and 2 as well as other possible points, depending on the

1762: family.  This extends the previous study of the $L_y=2$ strip of the triangular

1763: lattice with $(FBC_y,(T)PBC_x)$ \cite{wcy}.  While the strips

1764: with $(FBC_y,(T)PBC_x)$ have a locus ${\cal B}$ that passes through $q=3$, this

1765: is not the case with at least the $L_y=3$ strip with $(PBC_y,(T)PBC_x)$.  For

1766: the strips with $(FBC_y,(T)PBC_x)$ considered here we have shown that $q_c$ is

1767: a nondecreasing function of $L_y$.  As one increases $L_y$, the $L_y=4$ cyclic

1768: strip is the first one for which one can no longer solve for all of the

1769: $\lambda_j$'s as algebraic roots, since some of the equations involved are

1770: higher than quartic.

1771: 

1772: \item 

1773: 

1774: The crossing of ${\cal B}$ at the point $q=2$ for the (infinite-length limit

1775: of) strips with global circuits nicely signals the property that the Ising

1776: antiferromagnet has a frustrated zero-temperature critical point on these

1777: strips.  This has been discussed in \cite{ta} in the context of exact solutions

1778: for finite-temperature Potts model partition functions on the $L_y=2$ cyclic

1779: and M\"obius triangular strips.  In contrast, this connection is not, in

1780: general, present for strips with free longitudinal boundary conditions since

1781: ${\cal B}$ does not pass through $q=2$ (as recapitulated below).  The $q=3$

1782: Potts antiferromagnet also has a zero-temperature critical point on these 

1783: strips, and this is similarly manifested by the crossing of the singular locus

1784: ${\cal B}$ through the point $q=3$ for the cyclic and M\"obius strips with 

1785: $(FBC_y,(T)BC_x)$, but not for the $L_y=3$ torus and Klein bottle strips with

1786: $(PBC_y,(T)PBC_x)$. 

1787: 

1788: \item

1789: 

1790: An interesting feature of the cyclic strips of the triangular lattice (as well

1791: as the square and kagom\'e lattices) is that for all of cases that have been

1792: studied here and in \cite{w,wcy,pm}, there is a correlation between the

1793: coefficient $c_{G_s,j}$ of the respective dominant $\lambda_{G_s,j}$'s in

1794: regions that include intervals of the real axis.  Before, it was shown

1795: \cite{bcc} that the $c_{G_s,j}$ of the dominant $\lambda_{G_s,j}$ in region

1796: $R_1$ including the real intervals $q > q_c(\{G\})$ and $q < 0$ is $c^{(0)}=1$,

1797: where the $c^{(d)}$ were given in eqs. (\ref{cd}), (\ref{cdzeros}). 

1798: Here we extend this, observing that the $c_{G_s,j}$ that multiplies the

1799: dominant $\lambda_{G_s,j}$ in the region containing the intervals $0 < q < 2$

1800: is $c^{(1)}$.  For the cyclic strips of the triangular lattice that we have

1801: studied, namely, $L_y=2,3,4$, the $c_{G_s,j}$ multiplying the dominant

1802: $\lambda_{G_s,j}$ in the region containing the interval $2 < q < 3$ is

1803: $c^{(2)}$.  For the $L_y=4$ strip of the triangular lattice, there is another

1804: region containing the real interval $3 < q < q_c$, where $q_c$ for this strip

1805: was given in (\ref{qctcycly4}), and we find that $c^{(3)}$ multiplies

1806: the dominant $\lambda_{t4,j}$ containing this interval.  For the cyclic strips

1807: of the square lattice, although the values of $q_c$ are different, a similar

1808: correlation is observed; in particular, for the respective widths $L_y=3,4$, 

1809: $c^{(2)}$ multiplies the $\lambda_{G_s,j}$ that is dominant in the region

1810: containing the interval $2 < q < q_c$, where $q_c \simeq 2.34$ and 

1811: $q_c \simeq 2.49$ for these two widths \cite{s4}. 

1812: 

1813: \item 

1814: 

1815: For the strip of the triangular lattice of width $L_y=5$ with cylindrical

1816: $(PBC_y,FBC_x)$ boundary conditions, we find that ${\cal B}$ consists of arcs

1817: together with a (closed) oval. However, ${\cal B}$ does not pass through $q=0,

1818: 2$, or 3.  This is qualitatively the same morphology that was found for the

1819: corresponding strip of width $L_y=4$ (see Fig. 4 of \cite{strip2}).  For

1820: comparison, the triangular strip with width $L_y=3$ and $(PBC_y,FBC_x)$ had

1821: ${\cal B}= \emptyset$.  The point $q_c=4$ for $L_y=4$ and $q_c \simeq 3.28$ for

1822: $L_y=5$, which shows that for this family of strips, in cases where there is a

1823: $q_c$ (there is none for $L_y=3$), it is not, in general, a nondecreasing

1824: function of $L_y$.  This is somewhat reminiscent of the non-monotonicity of $W$

1825: that we showed in the case of $(PBC_y,BC_x)$ strips, where $BC_x=FBC_x$ or

1826: $(T)PBC_x$, in \cite{w2d}.  For the $L_y=6$ strip with $(PBC_y,FBC_x)$, we

1827: infer from the chromatic zeros that ${\cal B}$ consists of a single component,

1828: and from the analytic results we compute that $q_c \simeq 3.25$, which again

1829: shows the nonmonotonicity of $q_c$ as a function of $L_y$. The morphology of

1830: chromatic zeros for our $6 \times 16$ strip is similar to that found in

1831: \cite{baxter} for $8 \times 8$ patch, both with cylindrical boundary

1832: conditions.  For strips with torus or Klein bottle boundary conditions, $W$ and

1833: ${\cal B}$ have been calculated for only one $L_y$ value, namely, $L_y=3$ for

1834: the square lattice in \cite{tk} and for the triangular lattice here.  For the

1835: $L_y=3$ $(PBC_y,(T)PBC_x)$ square strip, $q_c$ is equal to the value 3 for the

1836: full 2D square lattice, but for the $L_y=3$ $(PBC_y,(T)PBC_x)$ triangular

1837: strip, our result (\ref{qcttorus}) shows that $q_c$ is less than the value of 4

1838: for the 2D triangular lattice.  

1839: 

1840: \item 

1841: 

1842: We have included an illustrative result for a wider strip with $(FBC_y,FBC_x)$

1843: boundary conditions, namely the strip with $L_y=5$.  As one increases $L_y$,

1844: this is the first value at which one can no longer solve analytically for the

1845: $\lambda_j$'s as algebraic roots.  The locus ${\cal B}$ is similar to the

1846: respective loci that were found earlier in studies of open strips

1847: \cite{strip,strip2,hs} in that it does not pass through $q=0$ or $q=2$ and does

1848: not separate the $q$ plane into regions containing intervals of the real axis.

1849: (It does pass through $q=3$, unlike the loci for the open strips with

1850: $L_y=3,4$.) Our results confirm the trend that was observed in earlier work

1851: \cite{strip,strip2}, namely that as $L_y$ increases, the arcs elongate and move

1852: closer together, and the arc endpoints nearest to the origin move toward this

1853: point.  In contrast to the strips with global circuits, these strips

1854: do not manifest the property that the $q=2$ (Ising) and $q=3$ Potts

1855: antiferromagnets have zero-temperature critical points since ${\cal B}$ does

1856: not pass through these respective points.  The simplest example of this is the

1857: Ising antiferromagnet on the infinite open line; in this case, although the

1858: model has a well-known zero-temperature critical point, this is not evident in

1859: the locus ${\cal B}$, which is the emptyset.

1860: 

1861: \item

1862: 

1863: For the strips with $(PBC_y,FBC_x)$ and $(FBC_y,FBC_x)$, while the endpoints of

1864: the arcs on ${\cal B}$ that lie closest to the origin tend to move toward the

1865: origin as $L_y$ increases, leading one to expect that in the limit $L_y \to

1866: \infty$, the limiting locus ${\cal B}$ would pass through $q=0$, no such motion

1867: toward $q=2$ is observed in the cases so far calculated.  This is in agreement

1868: with the fact that the locus found in \cite{baxter} for the triangular lattice

1869: constructed as the limit $L_x, L_y \to \infty$ with $(PBC_y,FBC_x)$ boundary

1870: conditions passes through $q=0$ and 4 (and at $q \simeq 3.82$) but not through

1871: $q=2$ or $q=3$.  Thus, assuming that our conjecture (\ref{bcrossq023}) is

1872: correct, it follows that the locus ${\cal B}$ for the triangular lattice

1873: depends on the boundary conditions used to define this lattice: if one

1874: constructs as the limit $L_y \to \infty$ with $(PBC_y,FBC_x)$ (cylindrical)

1875: boundary conditions, then ${\cal B}$ does not pass through $q=2$ or 3

1876: \cite{baxter}, while if one constructs it as the limit $L_y \to \infty$ with

1877: $(FBC_y,(T)PBC_x)$ (cyclic or M\"obius) boundary conditions, then, if the

1878: conjecture is valid, ${\cal B}$ would pass through $q=2$ and 3 in the limit

1879: just as it does for each of the values $L_y=2,3,4$ studied so far.  However,

1880: since the value of $q_c$ pertains directly to a physical quantity, as the

1881: minimal value of real $q$ above which $W(q)$ is analytic, one expects that

1882: $q_c$ should be independent of the boundary conditions used to define the 2D

1883: lattice.  All results obtained so far are consistent with this expectation.

1884: 

1885: \item

1886: 

1887: There have been a number of theorems proved concerning real chromatic zeros.

1888: An elementary result is that no chromatic zeros can lie on the negative real

1889: axis $q < 0$, since a chromatic polynomial has alternating coefficients.  It

1890: has also been proved that there are no chromatic zeros in the intervals $0 < q

1891: < 1$, and $1 < q < 32/27$ \cite{jackson}.  The bound of 32/27 in \cite{jackson}

1892: has been shown to be sharp; i.e., for any $\epsilon > 0$, there exists a graph

1893: with a chromatic zero at $q=32/27 + \epsilon$ \cite{thomassen}. See also 

1894: \cite{brown,sokal} and references therein.  Based on our studies of strips of

1895: the triangular (and square) lattices with the various boundary conditions

1896: considered, we make the following observation: for such strips, we have not

1897: found any chromatic zeros, except for the zero at $q=1$, in the interior of the

1898: disk $|q-1|=1$.  This motivates the conjecture that for these strips, there are

1899: no chromatic zeros with $|q-1| < 1$ except for the zero at $q=1$.  Assuming

1900: that this conjecture is valid, the bound would be a sharp bound, since the 

1901: circuit graph with $n$ vertices, $C_n$, has chromatic zeros lying precisely 

1902: on the circle $|q-1|=1$ and at $q=1$ \cite{wc}. Further work is needed to prove

1903: (or disprove) this conjecture. 

1904: 

1905: \end{enumerate} 

1906: 

1907: Some relevant features are summarized in Table \ref{proptable}.  The entries

1908: for $L_y=\infty$ with $(PBC_y,FBC_x)$ are from \cite{baxter}. 

1909: 

1910: \begin{table}

1911: \caption{\footnotesize{Comparative listing of properties of chromatic

1912: polynomials $P$, ground state degeneracy functions $W$, and nonanalytic loci

1913: ${\cal B}$ for strip graphs $G_s$ of the triangular (tri) lattice and their

1914: infinite-length limits.  New results in this work are marked with asterisks in

1915: the first column. The properties apply for a given strip of type $G_s$ of size

1916: $L_y \times L_x$; some apply for arbitrary $L_x$, such as $N_\lambda$, while

1917: others apply for the infinite-length limit, such as the properties of the locus

1918: ${\cal B}$.  For the boundary conditions in the $y$ and $x$ directions ($BC_y$,

1919: $BC_x$), F, P, and T denote free, periodic, and orientation-reversed (twisted)

1920: periodic, and the notation (T)P means that the results apply for either

1921: periodic or orientation-reversed periodic. The column denoted eqs. describes

1922: the numbers and degrees of the algebraic equations giving the $\lambda_{G_s,j}$

1923: in $P$; for example, $\{3(1),2(2),1(3)\}$ indicates that there are 3 linear

1924: equations, 2 quadratic equations and one cubic equation.  The column denoted

1925: BCR lists the points at which ${\cal B}$ crosses the real $q$ axis; the largest

1926: of these is $q_c$ for the given family $G_s$. The notation ``none'' in this

1927: column indicates that ${\cal B}$ does not cross the real $q$ axis.  Column

1928: labelled ``SN'' refers to whether ${\cal B}$ has \underline{s}upport for

1929: \underline{n}egative $Re(q)$, indicated as yes (y) or no (n).}}

1930: \begin{center}

1931: \begin{tabular}{|c|c|c|c|c|c|c|c|}

1932: \hline\hline $G_s$ & $L_y$ & $BC_y$ & $BC_x$ & $N_\lambda$ & eqs. & BCR & SN

1933: \\ \hline\hline

1934: tri & 2 & F & (T)P & 4  & \{2(1),1(2)\} & 3, \ 2, \ 0 & n  \\ \hline

1935: **tri & 3 & F & (T)P & 10 & \{3(1),2(2),1(3)\} & 3, \ 2, \ 0 & n \\ \hline

1936: **tri & 4 & F &   P & 26 & \{1(1),2(4),1(8),1(9)\} & 3.23, \ 3, \ 2, \ 0 & y \\

1937: \hline\hline

1938: tri & 3 & P   & F  & 1  & \{1(1)\}  & none  & $-$           \\ \hline

1939: tri & 4 & P   & F  & 2  & \{1(2)\}  & 4, \ 3.48  & n        \\ \hline

1940: **tri& 5 & P  & F  & 2  & \{1(2)\}  & 3.28, \ 3.21 & n     \\ \hline

1941: **tri& 6 & P  & F  & 5  & \{1(5)\}  & 3.25        & n      \\ \hline 

1942: tri&$\infty$& P& F  & $-$& $-$       & 4, \ 3.82, 0 & y     \\ \hline\hline

1943: **tri & 3 & P    & P & 11 & \{5(1),3(2)\} & 3.72, \ 2, \ 0 & n \\ \hline

1944: **tri & 3 & P   & TP & 5  & \{5(1)\}      & 3.72, \ 2, \ 0 & n \\ \hline\hline

1945: tri & 2 & F   & F & 1  & \{1(1)\} & none   & $-$          \\ \hline

1946: tri & 3 & F   & F & 2  & \{1(2)\} & 2.57   & n            \\ \hline

1947: tri & 4 & F   & F & 4  & \{1(4)\} & none   & n            \\ \hline

1948: **tri& 5 & F   & F & 9  & \{1(9)\} & 3     & n            \\ \hline\hline

1949: \end{tabular}

1950: \end{center}

1951: \label{proptable}

1952: \end{table}

1953: 

1954: \section{Conclusions}

1955: 

1956: In this paper we have presented exact calculations of the zero-temperature

1957: $q$-state Potts antiferromagnet partition functions (equivalently, chromatic

1958: polynomials $P$), on strips of the triangular lattice of width $L_y=3$ and with

1959: boundary conditions of four types: (a) $(FBC_y,PBC_x)=$ cyclic, (b)

1960: $(FBC_y,TPBC_x)=$ M\"obius, (c) $(PBC_y,PBC_x)=$ toroidal, and (d)

1961: $(PBC_y,TPBC_x)=$ Klein bottle, where $F$, $P$, and $TP$ denote free, periodic,

1962: and twisted periodic. In the infinite-length limits of these strips, exact

1963: results were given for the ground state degeneracy (exponential of the ground

1964: state entropy), $W$, and its analytic structure in the complex $q$ plane, in

1965: particular, the nonanalytic locus ${\cal B}$, was discussed.  Exact

1966: calculations of $P$ and $W$ and studies of ${\cal B}$ were also presented for

1967: wider strips, including (e) cyclic, $L_y=4$, (f) $(PBC_y,FBC_x)=$ cylindrical,

1968: $L_y=5,6$, and an illustrative $(FBC_y,FBC_x)=$ open case with $L_y=5$.  A

1969: comparative analysis of these results was included.  An interesting result of

1970: our calculations of $W$ on infinite-length strips with different widths and

1971: transverse boundary conditions is the observation that for the cases studied,

1972: ${\cal B}$ passes through $q=2$ (as well as $q=0$) for strips with periodic or

1973: twisted periodic longitudinal boundary conditions but does not for strips with

1974: free longitudinal boundary conditions.  Hence, in particular, if one uses

1975: periodic or twisted periodic longitudinal boundary conditions, the locus ${\cal

1976: B}$ nicely signals the existence of the zero-temperature critical point of the

1977: Ising antiferromagnet on these infinite-length, finite-width strips of the

1978: triangular lattice.  Numerical values of $W$ were given for infinite-length

1979: strips of various widths and were shown to approach values for the 2D lattice

1980: rapidly.  Some conjectures for the behavior of ${\cal B}$ for arbitrarily wide

1981: strips, and for a region in the $q$ plane free of chromatic zeros, were stated.

1982: These exact calculations of the $T=0$ Potts antiferromagnet partition function

1983: and ground state degeneracy on strips of the triangular lattice give valuable

1984: analytic knowledge of properties of Potts antiferromagnets.

1985: 

1986: \vspace{10mm}

1987: 

1988: 

1989: Note added: The original version of this paper, submitted in early Oct. 1999,

1990: contained calculations on $L_y=3$ strips of the triangular lattice.  In

1991: response to a request by a referee to perform calculations for wider strips, we

1992: have added the results on the $L_y=4$ strip with $(FBC_y,PBC_x)$, the

1993: $L_y=5,6$ strips with $(PBC_y,FBC_x)$, and the $L_y=5$ strip with 

1994: $(FBC_y,FBC_x)$.

1995: 

1996: \vspace{10mm}

1997: 

1998: Acknowledgment: The research of R. S. was supported in part by the U. S. NSF

1999: grant PHY-97-22101. 

2000: 

2001: \section{Appendix}

2002: 

2003: \subsection{Generating Function for $L_y=3$ M\"obius Strip of the Triangular

2004: Lattice} 

2005: 

2006: As noted in the text, it is convenient to leave the coefficients $c_{t3Mb,j}$, 

2007: $j=6,7,8$ in the general form (\ref{ctmj}).  For the evaluation of 

2008: these coefficients, we list here the generating function for this strip.  

2009: We have $d_{\cal N}=8$ and $d_{\cal D}=10$, and 

2010: \beq

2011: {\cal D}(tri(L_y=3),FBC_y,TPBC_x,q,x) = \prod_{j=1}^{10}

2012: (1-\lambda_{t3,j}(q)x)

2013: \label{lambdaformtmb}

2014: \eeq

2015: where the $\lambda_{t3,j}$ were given in eq. (\ref{ptcyclic}).  

2016: Since several of the $\lambda_{t3,j}$'s are algebraic, it is useful to 

2017: display the denominator in an explicitly polynomial form:

2018: \beqs

2019: & & {\cal D}(tri(L_y=3),FBC_y,TPBC_x,q,x)=(1+x)\Bigl [ 1-(q-2)x \Bigr ] \times

2020: \cr\cr

2021: & & \Bigl [1-(2q-7)x+(q-2)(q-3)x^2\Bigr ]

2022: \Bigl [ 1+(q-2)(q-3)x \Bigr ]F_{t3q3}F_{t3q2}

2023: \label{dtly3mb}

2024: \eeqs

2025: where

2026: \beq

2027: F_{t3q3}=\prod_{j=6,7,8}(1-\lambda_{t3,j}x)=1+b_{t3,1}x+b_{t3,2}x^2+b_{t3,3}x^3

2028: \label{ft3q3}

2029: \eeq

2030: with $b_{t3,j}$, $j=1,2,3$ given in eqs. (\ref{bt31})-(\ref{bt33}), and 

2031: \beq

2032: F_{t3q2}=\prod_{j=9,10}(1-\lambda_{t,j}x)=

2033:  1-(q^3-7q^2+18q-17)x+(q-2)^3(q-3)x^2 \ . 

2034: \label{ft3q2}

2035: \eeq

2036: For the numerator, extracting a common factor via the definition of the reduced

2037: coefficients $\bar A_{t3Mb,j}$, 

2038: \beq

2039: A_{t3Mb,j} \equiv q(q-1)(q-2)(q-3)\bar A_{t3Mb,j} \ , 

2040: \label{areduced}

2041: \eeq

2042: we have

2043: \beq

2044: \bar A_{t3Mb,0}=q^2-6q+10

2045: \label{atm0}

2046: \eeq

2047: \beq

2048: \bar A_{t3Mb,1}=(q-3)(q^3-12q^2+45q-55)

2049: \label{atm1}

2050: \eeq

2051: \beq

2052: \bar A_{t3Mb,2}=q^6-20q^5+170q^4-779q^3+2016q^2-2779q+1588

2053: \label{atm2}

2054: \eeq

2055: \beq

2056: \bar A_{t3Mb,3}=-(q-3)(5q^6-82q^5+574q^4-2185q^3+4745q^2-5536q+2691)

2057: \label{atm3}

2058: \eeq

2059: \beq

2060: \bar A_{t3Mb,4}=(q-2)(q-3)(9q^6-141q^5+930q^4-3303q^3+6651q^2-7176q+3224)

2061: \label{atm4}

2062: \eeq

2063: \beq

2064: \bar A_{t3Mb,5}=-(q-2)^2(q-3)^3(6q^4-55q^3+186q^2-277q+152)

2065: \label{atm5}

2066: \eeq

2067: \beq

2068: \bar A_{t3Mb,6}=-(q-2)^5(q-3)(q^4-12q^3+59q^2-138q+125)

2069: \label{atm6}

2070: \eeq

2071: \beq

2072: \bar A_{t3Mb,7}=(q-2)^6(q-3)^3 (3q^2-14q+17)

2073: \label{atm7}

2074: \eeq

2075: \beq

2076: \bar A_{t3Mb,8}=-(q-2)^9(q-3)^3 \ . 

2077: \label{atm8}

2078: \eeq

2079: 

2080: It should also be noted that there is a significant difference between the

2081: strips of the triangular lattice studied here and the analogous strips of the

2082: square lattice \cite{wcy,pm,tk}.  In general, a strip of the square lattice of

2083: width $L_y$ and length $L_x$ with any of the boundary conditions

2084: $(FBC_y,PBC_x)$ (cyclic), $(FBC_y,TPBC_x)$ (M\"obius), $(PBC_y,PBC_x)$ (torus),

2085: or $(PBC_y,TPBC_x)$ (Klein bottle) is invariant under a translation by one edge

2086: or vertex in the longitudinal direction.  However, in the case of the analogous

2087: strip of the triangular lattice, this is only true of the cases with cyclic and

2088: torus boundary conditions; the strips with M\"obius and Klein bottle boundary

2089: conditions have a ``seam'' along with the orientation of the triangles

2090: reverses.  This is discussed further in the appendix.  Thus, if one proceeds in

2091: a longitudinal direction along the triangular-lattice M\"obius strip, starting

2092: in a manner such that the triangles are formed by edges connecting the upper

2093: left and lower right vertices of squares (relative to one's direction of

2094: motion), then when one crosses this seam, the triangles will be formed by edges

2095: connecting the upper right and lower left vertices of the squares on the strip.

2096: Related to this, there are differences in the degrees of various vertices on

2097: these strips (where the degree of a vertex is defined as the number of edges

2098: that connect to this vertex).  If one avoids the lowest few values of $L_x$

2099: where the strips degenerate, then, in general, (i) for the cyclic triangular

2100: strip of width $L_y$ and length $L_x$, the $(L_y-2)L_x$ internal vertices have

2101: degree 6 while the $2L_x$ vertices on the upper and lower sides have degree 4;

2102: (ii) for the same strip as in (i) but with M\"obius instead of cyclic

2103: longitudinal boundary conditions, the $(L_y-2)L_x$ internal vertices have

2104: degree 6, the $2(L_y-1)$ vertices on the upper and lower sides except for those

2105: on the seam have degree 4, and, on the seam, the external vertices have degrees

2106: 5 and 3; (iii) for the same strip as in (i) but with torus or Klein bottle

2107: boundary conditions, all of the vertices have degree 6.

2108: 

2109: 

2110: \subsection{Equations for the Terms in the Chromatic Polynomial for the 

2111: Cyclic $L_y=4$ Strip of the Triangular Lattice}

2112: 

2113: Four of the $\lambda_{t4,j}$, which we label as $j=2,3,4,5$, are the same as 

2114: for the open $L_y=4$ strip of the triangular lattice.  These are solutions to 

2115: the quartic equation

2116: \beq

2117: \xi^4+b_{t4,1,1}\xi^3+b_{t4,1,2}\xi^2+b_{t4,1,3}\xi+b_{t4,1,4}=0

2118: \label{t4eq1}

2119: \eeq

2120: where the coefficients were given as $b_{t(4),j} \equiv b_{t4,1,j}$, 

2121: $j=1,..,4$ in eqs. (B.15)-(B.18) of \cite{strip}

2122: 

2123: The terms $\lambda_{t4,j}$, $j=6,7,8,9$ are solutions to the quartic equation 

2124: \beq

2125: \xi^4+b_{t4,2,1}\xi^3+b_{t4,2,2}\xi^2+b_{t4,2,3}\xi+b_{t4,2,4}=0

2126: \label{t4eq2}

2127: \eeq

2128: where

2129: \beq

2130: b_{t4,2,1}=4q-13

2131: \label{bt421}

2132: \eeq

2133: \beq

2134: b_{t4,2,2}=2(3q^2-18q+26)

2135: \label{bt422}

2136: \eeq

2137: \beq

2138: b_{t4,2,3}=(q-2)(4q^2-25q+38)

2139: \label{bt423}

2140: \eeq

2141: and

2142: \beq

2143: b_{t4,2,4}=(q-2)^2(q-3)^2 \ . 

2144: \label{bt424}

2145: \eeq

2146: 

2147: Another set of $\lambda_j$'s for $10 \le j \le 17$ are roots of an equation 

2148: of degree 8,

2149: \beq

2150: \xi^8 + \sum_{k=1}^8 b_{t4,3,k}\xi^{8-k}=0

2151: \label{t4eq3}

2152: \eeq

2153: where

2154: \beq

2155: b_{t4,3,1}=2(-3q^2+19q-31)

2156: \label{bt431}

2157: \eeq

2158: \beq

2159: b_{t4,3,2}=15q^4-186q^3+867q^2-1794q+1385

2160: \label{bt432}

2161: \eeq

2162: \beq

2163: b_{t4,3,3}=-20q^6+366q^5-2784q^4+11248q^3-25425q^2+30452q-15080

2164: \label{bt433}

2165: \eeq

2166: \beq

2167: b_{t4,3,4}=(q-2)(15q^7-334q^6+3174q^5-16676q^4+52294q^3-97852q^2+101138q-44528)

2168: \label{bt434}

2169: \eeq

2170: \beq

2171: b_{t4,3,5}=-(q-2)^2(q-3)^2(6q^6-126q^5+1076q^4-4804q^3+11861q^2-15378q+8185)

2172: \label{bt435}

2173: \eeq

2174: \beq

2175: b_{t4,3,6}=(q-2)^4(q-3)^3(q^5-25q^4+216q^3-868q^2+1670q-1246)

2176: \label{bt436}

2177: \eeq

2178: \beq

2179: b_{t4,3,7}=(q-2)^6(q-3)^4(2q^3-22q^2+80q-97)

2180: \label{bt437}

2181: \eeq

2182: \beq

2183: b_{t4,3,8}=(q-2)^8(q-3)^6 \ . 

2184: \label{bt438}

2185: \eeq

2186: 

2187: 

2188: A final set of $\lambda_{t4,j}$, $18 \le j \le 26$, are solutions to the 

2189: equation of degree 9

2190: \beq

2191: \xi^9 + \sum_{k=1}^9 b_{t4,4,k}\xi^{9-k}=0

2192: \label{t4eq4}

2193: \eeq

2194: where

2195: \beq

2196: b_{t4,4,1}=2(q-3)(2q^2-12q+21)

2197: \label{bt441}

2198: \eeq

2199: \beq

2200: b_{t4,4,2}=(q-3)(6q^5-90q^4+558q^3-1772q^2+2865q-1875)

2201: \label{bt442}

2202: \eeq

2203: \beqs

2204: & & b_{t4,4,3}=4q^9-111q^8+1380q^7-10071q^6+47476q^5-149742q^4+315652q^3

2205: \cr\cr

2206: & & -428385q^2+339300q-119368

2207: \label{bt443}

2208: \eeqs

2209: \beqs

2210: & & b_{t4,4,4}=(q-2)(q-3)^2(q^9-34q^8+491q^7-4032q^6+20961q^5-71954q^4 \cr\cr

2211: & & +163654q^3-238278q^2+201722q-75672)

2212: \label{bt444}

2213: \eeqs

2214: \beqs

2215: & & b_{t4,4,5}=-(q-2)^2(q-3)^3(3q^9-81q^8+983q^7-7029q^6+32609q^5-101701q^4 

2216: \cr\cr

2217: & & +213036q^3-288702q^2+229385q-81299)

2218: \label{bt445}

2219: \eeqs

2220: \beqs

2221: & & b_{t4,4,6}=(q-2)^4(q-3)^4(3q^8-70q^7+728q^6-4404q^5+16929q^4-42286q^3 

2222: \cr\cr

2223: & & +66933q^2-61296q+24830)

2224: \label{bt446}

2225: \eeqs

2226: \beq

2227: b_{t4,4,7}=-(q-2)^6(q-3)^6(q^6-20q^5+166q^4-734q^3+1833q^2-2462q+1393)

2228: \label{bt447}

2229: \eeq

2230: \beq

2231: b_{t4,4,8}=-(q-2)^8(q-3)^7(2q^4-21q^3+87q^2-165q+119)

2232: \label{bt448}

2233: \eeq

2234: \beq

2235: b_{t4,4,9}=-(q-2)^{12}(q-3)^8 \ . 

2236: \label{bt449}

2237: \eeq

2238: 

2239: 

2240: \subsection{Generating Function for the $L_y=5,6$ Strips of the Triangular

2241: Lattice with $(PBC_y,FBC_x)$}

2242: 

2243: For the $L_y=5$ strip we calculate a generating function of the form

2244: (\ref{gammagen}) with $d_{\cal D}=2$, $d_{\cal N}=1$ and, in the notation of

2245: eqs. (\ref{n}) and (\ref{d}), 

2246: \beq 

2247: b_{t5PF,1}=-q^5+15q^4-98q^3+355q^2-711q+614

2248: \label{bt5pf1}

2249: \eeq

2250: 

2251: \beq

2252: b_{t5PF,2}=(q-3)(3q^7-66q^6+619q^5-3205q^4+9877q^3-18065q^2+18078q-7588)

2253: \label{bt5pf2}

2254: \eeq

2255: 

2256: \beq

2257: A_{t5PF,0}=q(q-1)(q-2)(q-3)(q^6-14q^5+85q^4-290q^3+599q^2-723q+398)

2258: \label{at5pf0}

2259: \eeq

2260: 

2261: \beqs

2262: & & A_{t5PF,1}=-q(q-1)(q-2)(q-3)(q^2-2q+2)(3q^7-66q^6+619q^5-3205q^4 \cr\cr

2263: & & +9877q^3-18065q^2+18078q-7588) \ . 

2264: \label{at5pf1}

2265: \eeqs

2266: 

2267: For the $L_y=6$ strip we calculate a generating function of the form 

2268: (\ref{gammagen}) with $d_{\cal D}=5$, $d_{\cal N}=4$ and 

2269: \beq

2270: b_{t6PF,1}=-q^6+18q^5-145q^4+680q^3-1980q^2+3379q-2586

2271: \label{bt6pf1}

2272: \eeq

2273: 

2274: \beqs

2275: & & b_{t6PF,2} = 4q^{10}-128q^9+1868q^8-16352q^7+94977q^6-382031q^5+1076317q^4 

2276: \cr\cr & & -2093899q^3+2686606q^2-2047842q+702080

2277: \label{bt6pf2}

2278: \eeqs

2279: 

2280: \beqs

2281: & & b_{t6PF,3} = -2(q-3)^2(q^{12}-39q^{11}+695q^{10}-7493q^9+54509q^8

2282: -282283q^7+1068575q^6 \cr\cr 

2283: & & -2982861q^5+6098756q^4-8908956q^3+8820488q^2-5306146q+1462992)

2284: \label{bt6pf3}

2285: \eeqs

2286: 

2287: \beqs

2288: & & b_{t6PF,4} = -2(q-2)(q-3)^5(2q^{11}-62q^{10}+882q^9-7601q^8+44105q^7 

2289: -181018q^6 \cr\cr 

2290: & & +536536q^5-1149015q^4+1742334q^3-1779827q^2+1099188q-309188)

2291: \label{bt6pf4}

2292: \eeqs

2293: 

2294: \beqs

2295: & & b_{t6PF,5} = 4(q-2)^2(q-3)^8(q^2-5q+5)(2q^7-36q^6+277q^5-1179q^4+2990q^3

2296: \cr\cr & & -4505q^2+3728q-1310)

2297: \label{bt6pf5}

2298: \eeqs

2299: 

2300: \beqs

2301: & & A_{t6PF,0} = q(q-1)(q-2)(q^9-21q^8+199q^7-1121q^6+4159q^5-10623q^4 \cr\cr

2302: & & +18887q^3-22824q^2+17177q-6143)

2303: \label{at6pf0}

2304: \eeqs

2305: 

2306: \beqs

2307: & & A_{t6PF,1} = -q(q-1)(q-2)(4q^{13}-140q^{12}+2266q^{11}-22416q^{10}+

2308: 150973q^9

2309: \cr\cr & & -730186q^8+2607252q^7-6958852q^6+13899608q^5-20584349q^4 

2310: \cr\cr & & +22103679q^3-16461349q^2+7723994q-1748140) 

2311: \label{at6pf1}

2312: \eeqs

2313: 

2314: \beqs

2315: & & A_{t6PF,2} = 2q(q-1)(q-2)(q-3)^2(q^{15}-42q^{14}+816q^{13}-9734q^{12}

2316: +79793q^{11}

2317: \cr\cr & & -476549q^{10}+2144264q^9-7409966q^8+19852299q^7-41297346q^6

2318: +66301130q^5 \cr\cr 

2319: & & -80939629q^4+73099740q^3-46448750q^2+18742947q-3655548)

2320: \label{at6pf2}

2321: \eeqs

2322: 

2323: \beqs

2324: & & A_{t6PF,3} = 2q(q-1)(q-2)(q-3)^5(2q^{15}-72q^{14}+1212q^{13}-12651q^{12}

2325: +91556q^{11}

2326: \cr\cr & & -486599q^{10}+1962326q^9-6116898q^8+14870220q^7-28223745q^6

2327: +41554711q^5 \cr\cr & & -46735427q^4+39026466q^3-22975509q^2+8586616q-1545752)

2328: \label{at6pf3}

2329: \eeqs

2330: 

2331: \beqs

2332: & & A_{t6PF,4}= -4q(q-1)(q-2)^2(q-3)^8(q^2-5q+5)(q^4-5q^3+10q^2-10q+5) \cr\cr

2333: & & \times (2q^7-36q^6+277q^5-1179q^4+2990q^3-4505q^2+3728q-1310)

2334: \label{at6pf4}

2335: \eeqs

2336: 

2337: 

2338: \subsection{Generating Function for $L_y=5$ Open Strip of the Triangular

2339: Lattice}

2340: 

2341: For this strip we calculate a generating function of the form (\ref{gammagen})

2342: with $d_{\cal D}=9$, $d_{\cal N}=8$ and, in the notation of eqs. (\ref{n})

2343: and (\ref{d}),

2344: \beq

2345: b_{t5FF,1}=-(q-3)(q^4-10q^3+46q^2-113q+120)

2346: \label{btri1}

2347: \eeq

2348: 

2349: \beqs

2350: & &

2351: b_{t5FF,2}=6q^8-131q^7+1280q^6-7328q^5+26930q^4-65081q^3+100888q^2-91462q

2352: +36965

2353: \cr\cr

2354: & &

2355: \label{btri2}

2356: \eeqs

2357: 

2358: \beqs

2359: & & b_{t5FF,3}=-(q-2)(q-3)(15q^9-378q^8+4289q^7-28788q^6+126096q^5-374139q^4

2360: \cr\cr

2361: & & +752541q^3-989867q^2+772611q-272483)

2362: \label{btri3}

2363: \eeqs

2364: 

2365: 

2366: \beqs

2367: & & b_{t5FF,4}=(q-2)(q-3)^2(20q^{11}-607q^{10}+8429q^9-70702q^8+398115q^7

2368: \cr\cr

2369: & & -1580547q^6+4515585q^5-9285872q^4+13471537q^3-13131321q^2 \cr\cr

2370: & & +7738560q-2087938)

2371: \label{btri4}

2372: \eeqs

2373: 

2374: \beqs

2375: & & b_{t5FF,5}=-(q-2)^2(q-3)^3(15q^{12}-502q^{11}+7729q^{10}-72397q^9

2376: \cr\cr

2377: & & +459566q^8-2083176q^7+6915864q^6-16947196q^5+30430188q^4 \cr\cr

2378: & & -39053679q^3+34008163q^2-18041392q+4408580)

2379: \label{btri5}

2380: \eeqs

2381: 

2382: \beqs

2383: & & b_{t5FF,6}=(q-2)^3(q-3)^5(6q^{12}-203q^{11}+3149q^{10}-29630q^9

2384: +188440q^8 -853749q^7 \cr\cr

2385: & & +2826657q^6-6893466q^5+12293272q^4-15636918q^3+13466958q^2 \cr\cr

2386: & & -7049578q+1695556)

2387: \label{btri6}

2388: \eeqs

2389: 

2390: 

2391: \beqs

2392: & & b_{t5FF,7}=-(q-2)^5(q-3)^7(q^{11}-34q^{10}+516q^9-4647q^8+27718q^7

2393: -115308q^6 \cr\cr

2394: & & +342008q^5-724072q^4+1072964q^3-1060043q^2+628196q-169014)

2395: \label{btri7}

2396: \eeqs

2397: 

2398: \beqs

2399: & & b_{t5FF,8}=-(q-2)^7(q-3)^9(q^9-23q^8+241q^7-1505q^6+6145q^5-16929q^4

2400: \cr\cr

2401: & & +31319q^3-37359q^2+25972q-7987)

2402: \label{btri8}

2403: \eeqs

2404: 

2405: \beqs

2406: & & b_{t5FF,9}=(q-2)^{12}(q-3)^{11}(q^3-8q^2+21q-17)

2407: \label{btri9}

2408: \eeqs

2409: 

2410: We use the definition $A_{t5FF,j}=q(q-1)(q-2) \bar A_{t5FF,j}$ and have 

2411: \beq

2412: \bar A_{t5FF,0}=(q-2)^7

2413: \label{atri0}

2414: \eeq

2415: 

2416: \beqs

2417: & & \bar A_{t5FF,1}=-(q-2)^2(6q^8-113q^7+926q^6-4304q^5+12381q^4-22504q^3

2418: \cr\cr

2419: & & +25133q^2-15663q+4121)

2420: \label{atri1}

2421: \eeqs

2422: 

2423: \beqs

2424: & & \bar A_{t5FF,2}=(q-2)^2(q-3)(15q^{10}-363q^9+3936q^8-25141q^7+104572q^6

2425: \cr\cr

2426: & & -295342q^5+572184q^4-748554q^3+630336q^2-306798q+65006)

2427: \label{atri2}

2428: \eeqs

2429: 

2430: \beqs

2431: & &

2432: \bar A_{t5FF,3}=-(q-2)^2(q-3)^2(20q^{12}-587q^{11}+7867q^{10}-63587q^9

2433: +344801q^8 \cr\cr

2434: & & -1319670q^7+3650087q^6-7338694q^5+10623070q^4-10770336q^3+7237158q^2

2435: \cr\cr

2436: & & -2882235q+511741)

2437: \label{atri3}

2438: \eeqs

2439: 

2440: \beqs

2441: & &

2442: \bar A_{t5FF,4}=(q-2)^3(q-3)^3(15q^{13}-487q^{12}+7266q^{11}-65902q^{10}

2443: +405008q^9 \cr\cr

2444: & & -1779077q^8+5739974q^7-13753341q^6+24433426q^5-31724357q^4+29191727q^3

2445: \cr\cr

2446: & & -17973781q^2+6613085q-1093258)

2447: \label{atri4}

2448: \eeqs

2449: 

2450: \beqs

2451: & &

2452: \bar A_{t5FF,5}=-(q-2)^4(q-3)^4(6q^{14}-215q^{13}+3554q^{12}-35907q^{11}

2453: +247594q^{10}

2454: \cr\cr

2455: & & -1231849q^9+4556733q^8-12718548q^7+26883798q^6-42760478q^5+50287310q^4

2456: \cr\cr

2457: & & -42307379q^3+24007733q^2-8200007q+1267630)

2458: \label{atri5}

2459: \eeqs

2460: 

2461: \beqs

2462: & &

2463: \bar A_{t5FF,6}=(q-2)^6(q-3)^7(q^{12}-33q^{11}+485q^{10}-4225q^9+24378q^8

2464: -98266q^7 \cr\cr

2465: & & +283651q^6-589907q^5+875195q^4-900611q^3+607987q^2-240848q+42203) \cr\cr

2466: & &

2467: \label{atri6}

2468: \eeqs

2469: 

2470: \beqs

2471: & &

2472: \bar A_{t5FF,7}=(q-2)^{10}(q-3)^9(q^8-18q^7+145q^6-674q^5+1941q^4-3474q^3

2473: \cr\cr

2474: & & +3701q^2-2120q+499)

2475: \label{atri7}

2476: \eeqs

2477: 

2478: \beq

2479: \bar A_{t5FF,8}=-(q-1)^3(q-2)^{11}(q-3)^{11}(q^3-8q^2+21q-17) \ . 

2480: \label{atri8}

2481: \eeq

2482: 

2483: 

2484: \subsection{$L_y=2$ Cyclic and M\"obius Strips of the Triangular Lattice 

2485: with Odd $N_t$} 

2486: 

2487: In this part of the appendix we shall report some new results for $L_y=2$

2488: cyclic and M\"obius strips of the triangular lattice with an odd number $N_t$

2489: of triangles and compare these with the case of even $N_t=L_xL_y$.

2490: Some illustrative strip graphs are shown in Fig. \ref{figseam}.  

2491: 

2492: \vspace{10mm}

2493: 

2494: \unitlength 1.3mm

2495: 

2496: %cyclic triangular lattice with L_y=2

2497: \begin{picture}(100,10)

2498: \multiput(0,0)(10,0){5}{\circle*{2}}

2499: \multiput(0,10)(10,0){5}{\circle*{2}}

2500: \multiput(0,0)(10,0){5}{\line(0,1){10}}

2501: \multiput(0,0)(0,10){2}{\line(1,0){40}}

2502: \multiput(0,0)(10,0){4}{\line(1,1){10}}

2503: \put(-2,-2){\makebox(0,0){5}}

2504: \put(8,-2){\makebox(0,0){6}}

2505: \put(18,-2){\makebox(0,0){7}}

2506: \put(28,-2){\makebox(0,0){8}}

2507: \put(38,-2){\makebox(0,0){5}}

2508: \put(-2,12){\makebox(0,0){1}}

2509: \put(8,12){\makebox(0,0){2}}

2510: \put(18,12){\makebox(0,0){3}}

2511: \put(28,12){\makebox(0,0){4}}

2512: \put(38,12){\makebox(0,0){1}}

2513: \put(20,-8){\makebox(0,0){(a)}}

2514: 

2515: %Mobius triangular lattice with L_y=2

2516: \multiput(60,0)(10,0){5}{\circle*{2}}

2517: \multiput(60,10)(10,0){5}{\circle*{2}}

2518: \multiput(60,0)(10,0){5}{\line(0,1){10}}

2519: \multiput(60,0)(0,10){2}{\line(1,0){40}}

2520: \multiput(60,0)(10,0){4}{\line(1,1){10}}

2521: \put(58,-2){\makebox(0,0){5}}

2522: \put(68,-2){\makebox(0,0){6}}

2523: \put(78,-2){\makebox(0,0){7}}

2524: \put(88,-2){\makebox(0,0){8}}

2525: \put(98,-2){\makebox(0,0){1}}

2526: \put(58,12){\makebox(0,0){1}}

2527: \put(68,12){\makebox(0,0){2}}

2528: \put(78,12){\makebox(0,0){3}}

2529: \put(88,12){\makebox(0,0){4}}

2530: \put(98,12){\makebox(0,0){5}}

2531: \put(80,-8){\makebox(0,0){(b)}}

2532: \end{picture}

2533: \vspace*{3cm}

2534: 

2535: %cyclic odd-N triangular lattice with L_y=2

2536: \begin{picture}(100,10)

2537: \multiput(0,0)(10,0){4}{\circle*{2}}

2538: \multiput(0,10)(10,0){5}{\circle*{2}}

2539: \multiput(0,0)(10,0){4}{\line(0,1){10}}

2540: \put(0,0){\line(1,0){30}}

2541: \put(0,10){\line(1,0){40}}

2542: \multiput(0,0)(10,0){4}{\line(1,1){10}}

2543: \put(-2,-2){\makebox(0,0){5}}

2544: \put(8,-2){\makebox(0,0){6}}

2545: \put(18,-2){\makebox(0,0){7}}

2546: \put(28,-2){\makebox(0,0){5}}

2547: \put(-2,12){\makebox(0,0){1}}

2548: \put(8,12){\makebox(0,0){2}}

2549: \put(18,12){\makebox(0,0){3}}

2550: \put(28,12){\makebox(0,0){4}}

2551: \put(38,12){\makebox(0,0){1}}

2552: \put(20,-8){\makebox(0,0){(c)}}

2553: 

2554: %Mobius odd-N triangular lattice with L_y=2

2555: \multiput(60,0)(10,0){4}{\circle*{2}}

2556: \multiput(60,10)(10,0){5}{\circle*{2}}

2557: \multiput(60,0)(10,0){4}{\line(0,1){10}}

2558: \put(60,0){\line(1,0){30}}

2559: \put(60,10){\line(1,0){40}}

2560: \multiput(60,0)(10,0){4}{\line(1,1){10}}

2561: \put(58,-2){\makebox(0,0){5}}

2562: \put(68,-2){\makebox(0,0){6}}

2563: \put(78,-2){\makebox(0,0){7}}

2564: \put(88,-2){\makebox(0,0){1}}

2565: \put(58,12){\makebox(0,0){1}}

2566: \put(68,12){\makebox(0,0){2}}

2567: \put(78,12){\makebox(0,0){3}}

2568: \put(88,12){\makebox(0,0){4}}

2569: \put(98,12){\makebox(0,0){5}}

2570: \put(80,-8){\makebox(0,0){(d)}}

2571: \end{picture}

2572: 

2573: \vspace{10mm}

2574: 

2575: \begin{figure}[h]

2576: %\vspace{-4cm}

2577: %\centering

2578: %\leavevmode

2579: %\epsfxsize=6.0in

2580: %\begin{center}

2581: %\leavevmode

2582: %\epsffile{trigraphsly2.ps}

2583: %\end{center}

2584: %\vspace{-2cm}

2585: \caption{\footnotesize{Illustrative strip graphs of the triangular lattice with

2586: width $L_y=2$: (a) cyclic, $L_x=4$, even $N_t=2L_x$; (b)

2587: M\"obius, $L_x=4$, even $N_t=2L_x$; (c) cyclic, odd $N_t=7$;

2588: (d) M\"obius, odd $N_t=7$.}}

2589: \label{figseam}

2590: \end{figure}

2591: 

2592: 

2593: For the $L_y=2$ cyclic strip of the triangular lattice with even $N_t=2L_x$ 

2594: we calculated \cite{wcy} (see also \cite{sands,matmeth}) 

2595: \beq

2596: P(tri(L_y=2,cyc.),N_t=2m,q) = (q^2-3q+1) + [(q-2)^2]^m + (q-1)\Bigl [

2597: (\lambda_{t2,3})^m + (\lambda_{t2,4})^m \Bigr ] 

2598: \label{ptrinteven}

2599: \eeq

2600: where 

2601: \beq

2602: \lambda_{t2,(3,4)} = \frac{1}{2}\biggl [ 5-2q \pm \sqrt{9-4q} \ \biggr ] \ .

2603: \label{lamtly234}

2604: \eeq

2605: For the corresponding $L_y=2$ M\"obius strip of the triangular lattice with 

2606: even $N_t=2L_x$, we calculated \cite{wcy}

2607: \beq

2608: P(tri(L_y=2,Mb.),N_t=2m,q) = -1 + [(q-2)^2]^m -

2609: (q-1)(q-3)\frac{\Bigl [ (\lambda_{t2,3})^m - (\lambda_{t2,4})^m \Bigr ]}{

2610: \lambda_{t2,3}-\lambda_{t2,4}} \ . 

2611: \label{ptrimbnteven}

2612: \eeq

2613: Note that the M\"obius strip has a seam.  The function $W$ and the boundary 

2614: ${\cal B}$ were given in \cite{wcy}; ${\cal B}$ separates the $q$ plane into 

2615: three regions and crosses the real axis at $q=0,2$, and at $q_c=3$. 

2616: 

2617: We proceed to our new results.  For the $L_y=2$ cyclic strip

2618: of the triangular lattice containing an odd number $N_t=2m+1$ of triangles, 

2619: we calculate 

2620: \beqs

2621: & & P(tri(L_y=2,cyc.),N_t=2m+1,q) = -(q^2-3q+1) + (q-2)[(q-2)^2]^m + \cr\cr

2622: & & \frac{1}{2}(q-1)(q-3)\Biggl [\Bigl ( (\lambda_{t2,3})^m+ (\lambda_{t2,4})^m

2623: \Bigr ) + \frac{\Bigl ( (\lambda_{t2,3})^m - (\lambda_{t2,4})^m \Bigr )}{

2624: \lambda_{t2,3}-\lambda_{t2,4}} \Biggr ] \ . 

2625: \label{ptrintodd}

2626: \eeqs

2627: This strip has a seam. 

2628: 

2629: For the $L_y=2$ M\"obius strip of the triangular lattice with an odd number 

2630: $N_t=2m+1$ of triangles (which does not have a seam) we obtain 

2631: \beqs

2632: & & P(tri(L_y=2,Mb.),N_t=2m+1,q) = 1 +  (q-2)[(q-2)^2]^m + \cr\cr

2633: & & \frac{1}{2}(1-q)\Biggl [ \Bigl ( (\lambda_{t2,3})^m+ (\lambda_{t2,4})^m

2634: \Bigr ) + (9-4q)\frac{\Bigl ( (\lambda_{t2,3})^m - (\lambda_{t2,4})^m \Bigr )}

2635: {\lambda_{t2,3}-\lambda_{t2,4}} \Biggr ] \ . 

2636: \label{ptrimbntodd}

2637: \eeqs

2638: Note that the sum of the coefficients $C=0$ for the chromatic polynomials for

2639: both of the $L_y=2$ strips with odd $N_t$, eqs. (\ref{ptrintodd}) and 

2640: (\ref{ptrimbntodd}).  For comparison, for the $L_y=2$ even-$N_t$ strips, 

2641: $C=q(q-1)$ for the cyclic case, eq. (\ref{ptrinteven}) and $C=0$ for the

2642: M\"obius case, eq. (\ref{ptrimbnteven}). 

2643: 

2644: Equivalently, one may write these results in terms of generating functions, 

2645: with the definition analogous to (\ref{gamma}):

2646: \beq

2647: \Gamma(tri(L_y=2,BC_x),N_t=even,q,x) = \sum_{m=2}^{\infty}

2648: P(tri(L_y=2,BC_x),N_t=2m,q)x^{m-2} \ . 

2649: \label{gammatrintevengen}

2650: \eeq

2651: and 

2652: \beq

2653: \Gamma(tri(L_y=2,BC_x),N_t=odd,q,x) = \sum_{m=2}^{\infty}

2654: P(tri(L_y=2,BC_x),N_t=2m+1,q)x^{m-2}

2655: \label{gammatrintodd}

2656: \eeq

2657: where $BC_x=PBC_x$ or $TPBC_x$.  The generating functions for all of these

2658: strips have the same denominator, 

2659: \beqs

2660: & & {\cal D}(tri(L_y=2,cyc.))=\prod_{j=1}^4 (1-\lambda_{t2,j}x) \cr\cr

2661: & & = (1-x)\Bigl [ 1-(q-2)^2x \Bigr ] \Bigl [1-(5-2q)x+(q-2)^2x^2 \Bigr ] \ . 

2662: \label{dtri}

2663: \eeqs

2664: The numerators are easily worked out from the results that we have given for

2665: the chromatic polynomials and the denominator ${\cal D}$; for example, 

2666: \beq

2667: \Gamma(tri(L_y=2,cyc.),N_t=even,q,x) = 

2668: \frac{ q(q-1)(q-2)\Bigl [ q-3+(q-2)x-(q-2)^3x^2 \Bigr ]}

2669: {{\cal D}(tri(L_y=2,cyc.))}

2670: \label{gammatrinteven}

2671: \eeq

2672: \beq

2673: \Gamma(tri(L_y=2,cyc.),N_t=odd,q,x) = \frac{q(q-1)(q-2)(q-3)

2674: \bigl [ q-3+(q-2)^2x \bigr ]}{{\cal D}(tri(L_y=2,cyc.))} \ . 

2675: \label{gammatri}

2676: \eeq

2677: and so forth for the M\"obius strips. 

2678: 

2679: We remark on some general features of these results.  The chromatic polynomials

2680: for both even and odd $N_t$ and both cyclic and M\"obius strips have the same 

2681: four $\lambda_j$'s, and hence the same $W$ functions and boundary ${\cal B}$ 

2682: (given in \cite{wcy}).  For all cases, $q_c=3$.  These properties are in

2683: agreement with the general discussion in \cite{wcy,pm,bcc} on the effects of 

2684: boundary conditions on $W$ and ${\cal B}$. 

2685: 

2686: 

2687: \begin{thebibliography}{99}

2688: 

2689: \bibitem{potts}{Potts, R. B. 1952 Proc. Camb. Phil. Soc. {\bf 48}, 106.}

2690: 

2691: \bibitem{wurev}{Wu, F. Y. 1982 Rev. Mod. Phys. {\bf 54}, 235.} 

2692: 

2693: \bibitem{rrev}{Read, R. C. 1968 J. Combin. Theory {\bf 4}, 52.}

2694: 

2695: \bibitem{tutte}{Tutte, W. T. 1974 ``Chromials'', in Lecture Notes in

2696: Math. v. 411, p. 243.}

2697: 

2698: \bibitem{rtrev}{Read, R. C. and Tutte, W. T. 1988 ``Chromatic Polynomials'',

2699: in {\it Selected Topics in Graph Theory, 3}, (Academic Press, New York), 

2700: p. 15.} 

2701: 

2702: \bibitem{bbook}{Biggs, N. L. 1993 {\it Algebraic Graph Theory} (Cambridge 

2703: Univ. Press, Cambridge).}

2704: 

2705: \bibitem{w}{Shrock, R. and Tsai, S.-H. 1997 Phys. Rev. {\bf E55}, 5165.}

2706: 

2707: \bibitem{bds}{Biggs, N. L., Damerell, R. M. and Sands, D. A. 1972

2708: J. Combin. Theory B {\bf 12}, 123.}

2709: 

2710: \bibitem{bm}{Biggs, N. L. and Meredith, G. H. 1976 J. Combin. Theory B 

2711: {\bf 20}, 5.}

2712: 

2713: \bibitem{b}{Biggs, N. L. 1977 Bull. London Math. Soc. {\bf 9}, 54.} 

2714: 

2715: \bibitem{wcy}{Shrock, R. and Tsai, S.-H. 1999 Phys. Rev. {\bf E60}, 3512; 

2716: Physica A {\bf 275}, 429.} 

2717: 

2718: \bibitem{pm}{Shrock, R. 1999 Phys. Lett. {\bf A261}, 57.}

2719: 

2720: \bibitem{bkw79}{Beraha, S. and Kahane, J. 1979 J. Combin. Theory B {\bf 27},

2721: 1.} 

2722: 

2723: \bibitem{bkw}{Beraha, S., Kahane, J., and Weiss, N. 1980 J. Combin. Theory B

2724: {\bf 28}, 52.}

2725: 

2726: \bibitem{readcarib88}{Read, R. C. 1988 in Proc. 5th Caribbean Conf. on

2727: Combin. and Computing.} 

2728: 

2729: \bibitem{read91}{Read, R. C. and Royle, G. F. 1991 in {\it Graph Theory,

2730: Combinatorics, and Applications} (Wiley, NY), vol. 2, p. 1009.}

2731: 

2732: \bibitem{strip}{Ro\v{c}ek, M., Shrock, R., and Tsai, S.-H. 1998 Physica

2733: {\bf A252}, 505.}

2734: 

2735: \bibitem{strip2}{Ro\v{c}ek, M., Shrock, R., and Tsai, S.-H. 1998 Physica

2736: {\bf A259}, 367.}

2737: 

2738: \bibitem{hs}{Shrock, R. and Tsai, S.-H. 1998 Physica {\bf A259}, 315.}

2739: 

2740: \bibitem{w2d}{Shrock, R. and Tsai S.-H. 1998 Phys. Rev. {\bf E58}, 4332,

2741: cond-mat/9808057.}

2742: 

2743: \bibitem{pg}{Shrock, R. and Tsai, S.-H. 1999 J. Phys. A Lett. {\bf 32} L195.}

2744: 

2745: \bibitem{nec}{Shrock, R. and Tsai, S.-H. 1999 J. Phys. A {\bf 32} 5053.}

2746: 

2747: \bibitem{bcc}{Shrock, R. 1999, in the Proceedings of the British Combinatorial

2748: Conference (July, 1999); Shrock, R. 2000 Physica A, in press.}

2749: 

2750: \bibitem{ks}{H. Kl\"upfel and R. Shrock, unpublished; H. Kl\"upfel, Stony Brook

2751: thesis, July, 1999.}

2752: 

2753: \bibitem{tw99}{Shrock, R. 1999, in the Proceedings of Statphys-Taiwan-99

2754: (August, 1999).} 

2755: 

2756: \bibitem{tk}{Biggs, N. L. and Shrock, R. 1999 J. Phys. A (Letts.) {\bf 32}, 

2757: L489.} 

2758: 

2759: \bibitem{matmeth}{Biggs, N. L. LSE report LSE-CDAM-99-03 (May, 1999).}

2760: 

2761: \bibitem{bgen}{Biggs, N. L. LSE reports LSE-CDAM-99-05,06 (July, 1999).} 

2762: 

2763: \bibitem{baxter}{Baxter, R. J. 1987 J. Phys. A {\bf 20}, 5241.}

2764: 

2765: \bibitem{sands}{Sands, D. A., Ph.D. Thesis, Univ. of London, 1972 

2766: (unpublished). RS thanks Prof. N. L. Biggs for kindly sending him a copy of

2767: this thesis.} 

2768: 

2769: \bibitem{wa}{Shrock, R. and Tsai, S.-H. 1997 Phys. Rev. {\bf E56}, 3935.}

2770: 

2771: \bibitem{wa3}{Shrock, R. and Tsai, S.-H. 1998 J. Phys. A {\bf 31}, 9641.}

2772: 

2773: \bibitem{wa2}{Shrock, R. and Tsai, S.-H. 1999 Physica {\bf A265}, 186.}

2774: 

2775: \bibitem{sokal}{Sokal, A. Combin. Prob. Comput., in press.}

2776: 

2777: \bibitem{cf}{Shrock, R. and Chang, S.-C. YITP-SB-00-12.} 

2778: 

2779: \bibitem{cmo}{Matveev, V. and Shrock, R. 1995 J. Phys. A {\bf 28} 5235.}

2780: 

2781: \bibitem{wn}{Shrock, R. and Tsai, S.-H. 1997 Phys. Rev. {\bf E55}, 6791, 

2782: Phys. Rev. {\bf E56}, 4111.}

2783: 

2784: \bibitem{wannier}{Wannier, G. H. 1950 Phys. Rev. {\bf 79}, 357.}

2785: 

2786: \bibitem{stephenson}{Stephenson, J. 1964 J. Math. Phys. {\bf 5}, 1009.}

2787: 

2788: \bibitem{ta}{Chang, S.-C. and Shrock, R., YITP-SB-99-58} 

2789: 

2790: \bibitem{ss}{Salas, J. and Sokal, A., work in progress (private

2791: communication).} 

2792: 

2793: \bibitem{s4}{Chang, S.-C. and Shrock, R. YITP-SB-00-11.}

2794: 

2795: \bibitem{jackson}{Jackson, B. 1993 Combin. Prob. Comput. {\bf 2} 325.}

2796: 

2797: \bibitem{thomassen}{Thomassen, C. 1997 Combin. Prob. Comput. {\bf 6} 497.}

2798: 

2799: \bibitem{brown}{Brown, J. 1998 J. Combin. Theory {\bf B72}, 251.}

2800: 

2801: \bibitem{wc}{Shrock, R. and Tsai, S.-H. 1997 Phys. Rev. {\bf E56}, 1342.} 

2802: 

2803: \end{thebibliography}

2804: 

2805: \vfill

2806: \eject

2807: \end{document}

2808: