cond-mat0111157/v12.tex
1: %  PORUKA:  privremeno je format onaj koji trosi manje papira
2: %
3: %  ispell v12.tex                       <- gruba provjera engl.
4: %  dvips -o v12.ps v12.dvi &             <- napravim *.ps fajl
5: %  dvips -p 9 -l 11 -o v12.ps v12.dvi &  <- napravim *.ps fajl
6: %  ghostview v12.ps  &                  <- pogledam ga
7: %
8: % zaglavlje uzeto iz clanka
9: %---------------------------
10: %  \documentstyle{elsart}
11:  \documentclass{elsart}          % za latex2e: \cal -> \mathcal   
12:  \usepackage{graphicx}      
13: %   \input /home/glumac/latex/croat.tex  
14: 
15:  \begin{document}
16: 
17:  \newcommand{\JPA}{J. Phys. A} 
18:  \newcommand{\JPC}{J. Phys. C} 
19:  \newcommand{\JSP}{J. Stat. Phys.}   
20:  \newcommand{\PA}{Physica A}   
21:  \newcommand{\PRB}{Phys. Rev. B}  
22:  \newcommand{\PRE}{Phys. Rev. E}  
23:  \newcommand{\PRL}{Phys. Rev. Lett.}   
24: 
25:  \newcommand{\be}{\begin{equation}}
26:  \newcommand{\bea}{\begin{eqnarray}}
27:  \newcommand{\cF}{{\mathcal F}}
28:  \newcommand{\cG}{{\mathcal G}}
29:  \newcommand{\cH}{{\mathcal H}}
30:  \newcommand{\cO}{{\mathcal O}}
31:  \newcommand{\cq}{complex-$q$ }
32:  \newcommand{\cS}{complex-$S$ }
33:  \newcommand{\ee}{\end{equation}}
34:  \newcommand{\eea}{\end{eqnarray}}
35:  \newcommand{\etl}{{\it et al.}\,\,}   % ? etal vec postoji u latex2e ??
36:  \newcommand{\req}{real-$q$ }
37:  \newcommand{\vfi}{\varphi}
38: 
39: 
40: % ==========  PROMIJENITI KASNIJE: ======================
41: % za pravi format: razkomentirati 3 linije i izbaciti 4 sljedece ( do =======)
42:    \renewcommand              
43: %% \baselinestretch 2              %  s proredom 
44: %% \baselineskip 24pt              %                     
45:    \baselinestretch 1              %   bez proreda 
46:    \baselineskip 15pt              %                     
47:   \setlength\textwidth{19cm}      % ve"ta "sirina teksta 
48:   \setlength\textheight{23cm}     % ve"ta visina teksta
49: %       \vsize=9.0truein
50: %       \hsize=6.3truein
51: % =======================================================
52: 
53:  \begin{frontmatter} 
54: 
55:  \title{Complex-$q$ zeros of the partition function of the 
56:         Potts model with long-range interactions}  
57: 
58:  \author{Zvonko Glumac\thanksref{corresp}}, 
59:  \thanks[corresp]{Corresponding author. tel.: (385)-031-208-900;
60:                                         fax:  (385)-031-208-905; 
61:                                         e-mail: zvonko@vrabac.ifs.hr.} 
62: 
63:  \address{Faculty of Electrical Engineering, 
64:          Kneza Trpimira 2B, 31 000 Osijek, Croatia}  
65:  \author{Katarina Uzelac}
66: 
67:  \address{ Institute of Physics, P.O.B. 304,
68:  Bijeni\v{c}ka 46, HR-10000 Zagreb, Croatia}
69: 
70: 
71:  \begin{abstract}
72: The zeros of the partition function of the ferromagnetic
73: $q$-state Potts model with long-range interactions in the \cq plane
74: are studied in the mean-field case, while preliminary numerical results are 
75: reported for the finite $1d$ chains with power-law decaying interactions. 
76:  In both cases, at any fixed temperature, the zeros lie on the 
77: arc-shaped contours, which cross the positive real axis at the
78: value for which the given temperature is transition temperature. 
79: For finite number of spins the positive real axis is free of zeros, which 
80: approach to it in the thermodynamic limit.
81:  The convergence exponent of the zero closest to the positive \req axis is 
82:  found to have the same value as the temperature critical exponent $1/\nu$.
83: 
84: 
85:  \leftline{PACS: 05.50.+q, 64.60.Cn} 
86:  \end{abstract}
87: 
88:  \begin{keyword}
89:  Phase transitions, $q$-state Potts model, Complex-$q$ zeros, 
90: Long-range interactions 
91:  \end{keyword} 
92: 
93: \end{frontmatter} 
94: 
95: 
96:   \section{Introduction}  \label{sec-int}  
97: 
98:  Several decades ago, it was shown that the study of distribution 
99:  of the partition-function zeros in the complex plane of an appropriate 
100: variable,  can provide relevant  informations on phase transitions.
101:  In their pioneering work, Lee and Yang \cite{YL52,LY52} have shown 
102:  the connection between the existence of phase transition and the 
103:  distribution of partition function zeros of the ferromagnetic Ising model 
104:  in the complex plane of a symmetry breaking field.
105:  Later on, \cite{F65,OKSK68},  similar connection was established for 
106:  the partition function zeros in the complex-temperature plane. 
107: In both cases, the singularities at the transition point 
108:  are closely related to the way that the zeros approach (either with 
109: temperature or with the size of the system) to the positive real axis.
110: 
111: The partition function zeros were investigated for number of models, 
112: including the Potts model both in complex-field \cite{KC98,KC99} and 
113: in complex-temperature  \cite{B90,M91,CHW96,KCCH99} plane.
114: In the Potts model a relevant quantity is also the number of states 
115: $q$, often considered as a continuous parameter, which is formally 
116: possible on the basis of its graphical representation \cite{FK69} as it 
117: will be discussed later in text.  
118: This gives the motivation to investigate the zeros for this model also 
119: in the plane of complex $q$.   
120: Such an idea was already formulated \cite{Bx86,Bx87} but mostly in 
121: context of chromatic polynomials in connection with studies of the 
122: ground state of the anti-ferromagnetic Potts model 
123:  \cite{ST97a,ST97b,RST98,ST99a,ST99b,S99,SS00}.
124: Only a few works on complex-$q$ plane address the ferromagnetic 
125: Potts model at finite temperatures.
126: The papers by Chang and  Shrock, \cite{CS00}, study the loci of 
127: partition function zeros in $2d$ Potts model with different boundary 
128: conditions.
129: In their very recent paper Kim and Creswick \cite{KC01} give a systematic
130: study of both Fisher and complex-$q$ zeros in the $2d$ Potts model with
131: short-range  interactions and find the similarity between the scaling property
132: of complex-$q$ zeros and the den Nijs expression for the thermal critical
133: exponent.
134: 
135:   In the present paper we investigate the zeros in the complex-$q$ plane
136:   in connection with the first- and second-order phase
137:   transitions in the ferromagnetic Potts model.
138:   To this purpose we consider here the Potts model with long-range 
139: power-law decaying interactions, which
140: already in one dimension displays diversity of critical regimes
141: and embraces most of the nontrivial aspects of phase
142: transitions, characteristic of the Potts model in general.
143: The limiting case of this model, where all the interactions are equal
144: (and which represents the mean-field case of the model) has the
145: advantage to be exactly solvable for arbitrary $q$.
146: 
147: In the next section we describe briefly the starting
148:  Hamiltonian and its graphical representation. In subsequent sections
149:  we present the results for the mean-field (Section 3) and the
150:  power-law decaying
151:  interactions case (Section 4).
152: 
153: 
154: 
155: \section{Hamiltonian}
156: 
157: The ferromagnetic Potts model with interactions of arbitrary range is 
158: described by the Hamiltonian
159:  \be \label{eq:lrham}
160:   H  = - \sum_{i,j}\; J_{i,j}\cdot
161:              \delta\,(s_i, s_{j}),   
162:  \ee  
163: where $s_i$ denotes the Potts spin at site $i$ of a 
164:  $d$-dimensional lattice, 
165: which can take $q$ values, and $J_{i,j}\ge 0$ denotes the ferromagnetic 
166: interaction between spins at sites $i$ and $j$. 
167: 
168: 
169: An impressive amount of studies of this model in the case of 
170: nearest neighbour interactions (for a review see Wu \cite{W82})
171: reveal its very complex critical behaviour which varies with $q$ and
172: comprises  transitions of different orders and belonging to various 
173: universality classes. 
174: 
175: The model with long-range interactions was studied less, 
176: and mostly in two special cases.
177: The first is the case when all the interactions are equal and 
178: it represents the mean-field limit of the model. 
179: It can be approached by the saddle point approximation 
180:  \cite{KMS54,SF73,MS74,PL76} and in the thermodynamic limit 
181: the exact expressions, in particular those for critical 
182: temperature and exponents may be derived for all values of $q$.
183: Second is the case with power-law interactions 
184: decaying with distance as $1/r^{d+\sigma}$. 
185: The type of phase transition there depends in addition 
186: on the parameter of range  $\sigma$, which in certain aspects has 
187: similar effects as the change of dimensionality \cite{LB96}. 
188: The phase transition is there present in one dimension
189:  \cite{D69,ACCN88,GU93} and by
190: varying $\sigma$, series of different regimes are met, which 
191: makes this $1d$ model a useful paradigm for most of important 
192: features of the model in general (including also the first-order 
193: phase transition).  
194: It was a subject of  number of recent studies
195: \cite{GU93,LB95,C95,UG97,GU98,M99,BDD99,KL00,LM01}. 
196: 
197: 	The graphical representation of the Potts model \cite{FK69} 
198: holds for arbitrary range of interactions.
199: The partition function for the Hamiltonian (\ref{eq:lrham}) 
200: may be expressed as 
201:   \be \label{eq:zlropci}
202:    Z_N = \sum_{{\rm all}\;\cG} \,
203:    \prod_{\rm {active\atop links}} v_{i,j} \;\,q^{n(\cG)},
204:    \ee
205: where the summation is taken over all possible graphs $\cG$.
206: Each graph represents one possible configuration of 
207: connections between the $N$ spins, called active links, 
208: each of which contributes with a factor $v_{i,j} = \exp(J_{i,j}/{k_B\,T}) - 1$. 
209: $n(\cG)$ denotes the number of  
210: of disconnected parts of the graph called clusters.
211: The product is taken over all the active links of a considered graph.
212: 
213: Whatever the interactions are, the number of clusters that a single graph 
214: contains ranges from one cluster  (low temperature limit, where 
215: all spins are interconnected), up to at most $N$ clusters
216: (high temperature limit where no active  links are present and each 
217: spin is a cluster of size one).
218: Thus, by collecting all the graphs with the same number of clusters, 
219: the partition function becomes simply an $N$-th order polynomial in $q$
220:    \be \label{eq:Npol}
221:    Z_N = \sum_{n = 1}^N\,a_n(T,J_{i,j},N)\;q^n.
222:    \ee
223: This made possible to treat the number of states $q$ as a continuous parameter.
224: It also gives the possibility to examine the zeros of the partition function
225: as the zeros of the $N$-th order polynomial in $q$.  
226: 
227: The coefficients $a_{n}$ are functions of temperature, the finite size 
228: and interactions, and are quite complicated in the general case 
229: (\ref{eq:lrham}), but can be written in more compact form for  
230: the special cases considered here.
231: 
232: 
233: \section {The mean-field case}
234: 
235: 
236: The simplest case of long-range interactions is the one where all the 
237: interactions are set to be equal, $J_{i,j}=J$, which reduces the  
238: Hamiltonian (\ref{eq:lrham})  to the form
239:  \be \label{eq:mfham}
240:  \cH_{MF} = H_{MF}/{k_B\,T} = -
241:   \frac{K}{N} \sum_{i=1}^{N-1}\;\sum_{j=1}^{N-i}\; 
242:              \delta\,(s_i, s_{i+j}),   
243:  \ee  
244: where $K=J/{k_B\,T}$, and division by $N$ is necessary to keep 
245: the summation finite. 
246: This model corresponds to the mean-field limit. It was solved by the 
247: saddle point approximation \cite{KMS54}. 
248: The phase transition is there of the first order for $q > 2$ 
249: and of the second order for $q \le 2$.
250: The exact expressions for the inverse of the transition temperature 
251: $K_t$  is given by 
252:  \be \label{eq:Kexact1}  
253:   K_t =  2\,\frac{q-1}{q-2}\,\ln (q-1), 
254: \hspace{1cm} 
255:  \ee 
256: for $q > 2$, 
257: and by $K_c = q$ for $q \le 2$.   
258: Exact analytical expressions may be derived also for the 
259: critical exponents matching the values obtained in the
260: mean-field regime of the short-range interaction model \cite{PL76}
261: (except those related to the correlation function, as it 
262: will be discussed later in text). 
263: 
264: The graph representation (\ref{eq:zlropci}) reduces in the mean-field limit to
265:    \be \label{eq:zmf}   
266:    Z_N = \sum_{{\rm all}\;\cG} \, v^{b(\cG)}\,q^{n(\cG)},
267:    \ee
268: where the active links have all the same strength $v = \exp(K/N) - 1$, 
269:   independently of distance between the spins they connect and
270:  $b(\cG)$ denotes the number of active links in a given graph $\cG$. 
271: Written as a  polynomial in $q$  the above equation is equal to
272:    \be \label{eq:Npolmf}
273:    Z_N = \sum_{n = 1}^N\,a_n(v, N)\;q^n,
274:    \ee
275:  where the coefficients $a_n$ are real and positive and depend on 
276: temperature and size $N$ only. 
277: (Notice, that, since the number of active bonds may reach up to $N\,(N-1)/2$, 
278: the above partition function could equally be written as a polynomial 
279: in temperature variable $v$, which would be of order $N\,(N-1)/2$.)
280: 
281: 
282: \subsection{Numerical results}
283: 
284: For relatively small sizes the evaluation of coefficients and 
285: finding the zeros of the polynomial (\ref{eq:Npolmf}) in the 
286: complex plane may easily be performed numerically.
287: Numerical results for sizes up to $N = 35$ spins are presented in
288:  Figs. 1 (a)-(c) for the three different temperatures,  
289:   $K = 0.5, 2, 4.540457$.   
290: These temperatures are the transition temperatures of the mean-field model
291: with $q = 0.5, 2$, and $8$ respectively.
292: Each figure displays the data obtained for several sizes.
293: 
294: One may observe that in all the three cases the zeros lie on contours
295: similar but different from circle
296: which cross the real axis near the value of $q$, for which the given 
297: temperature is the transition temperature. 
298: The zeros are absent from the positive real axis, but approach to 
299: it with increasing $N$.
300: 
301: We were thus interested to perform a more precise extrapolation of this
302: crossing point in the thermodynamic limit and studied the convergence
303: of the zero closest to the positive real axis 
304: with $N$.
305: Further, we have examined the possible connection of the convergence 
306: of the closest zero with $N$ to the scaling properties of the model 
307: near criticality.
308: 
309: To this purpose we have used Burlisch and Stoer (BST) extrapolation
310: procedure \cite{BS64,HS88}.
311: This algorithm was shown \cite{HS88} to converge rapidly, needs a modest 
312: amount  of data and, in  advantage over the similar algorithm  by van den 
313: Broeck and Schwartz (VBS) \cite{VBS79}, it is less  sensitive to the rounding 
314: errors.
315: 
316:    In short, BST algorithm extrapolates a sequence of the form
317:    \be \label{eq:seq}  
318:    A(N) = A_{\infty} + a_1\,\cdot\,N^{-\omega} + 
319:    a_2\,\cdot\,N^{-2\,\omega} 
320:    + \cdots  
321:    \ee 
322:    in the $N\,\to\,\infty$ limit by a set of transformations recursively 
323:    given by 
324:    \bea 
325:    A_m^{(N)} & = & A_{m - 1}^{(N + 1)}  \\
326:    & + & \left(A_{m - 1}^{(N + 1)}  - A_{m - 1}^{(N)}\right) \, \cdot \,
327:    \left[ \left(\frac{N + m}{N}\right)^{\omega}\, 
328:    \left( 1 - 
329:    \frac{A_{m - 1}^{(N + 1)} - A_{m - 1}^{(N)}}{A_{m - 1}^{(N + 1)} 
330:    - A_{m - 2}^{(N + 1)}}
331:    \right) - 1 \right]^{-1},  \nonumber 
332:    \eea  
333:    where $A_{-1}^{(N)} = 0$ and $A_0^{(N)} = q_{1, 2}^{(0)}(N)$ 
334:    are the input data and $\omega$ is a free parameter. 
335: 
336: If the original series is of the form:
337:    \be \label{eq:seq12}  
338:    A(N) = A_{\infty} + c_1\,\cdot\,N^{-\omega_1} + 
339:    c_2\,\cdot\,N^{-\omega_2} 
340:    + \cdots  
341:    \ee 
342: it incorporates the leading convergence exponent as $\omega$, 
343: while the higher order approximants converge as $N^{-\omega^{'}}$, 
344: where $\omega^{'}=min(2\omega_1,\omega_2)$.
345: Although it was shown, that the extrapolations are insensitive to the choice 
346: of $\omega$ in a wide range of values, it influences its convergence
347: speed. Thus, by maximising the convergence of higher approximants one can in 
348: the same time obtain both the extrapolated value and the leading convergence 
349: exponent. Also, when the extrapolated value is known in advance, 
350: $\omega$ yields the leading convergence exponent.
351: 
352: We have examined the convergence of the real and imaginary parts 
353: of the zero closest to the real axis, denoted as $q_{1, 2}^{(0)}(N)$ 
354: for several  temperatures using the numerical results ranging from 
355: $N = 21$ to $N = 35$. 
356: 
357:  In the extrapolation of  both real and imaginary part
358:  $q_{1, 2}^{(0)}(N)$,    
359:  the same parameter $\omega$ was used and was fixed by the constraint 
360: $q_2^{(0)}(N\,\to\,\infty)\,\to\,0$.  
361: 
362: The results of extrapolations are presented on Table 1. 
363: By $Q_0$ is denoted the value of $q$ for which the considered temperature
364: is known to be the transition temperature. 
365: All the data were obtained by requiring that condition $q_2^{(0)}(N)\,\to\,0$
366: should be fulfilled with precision of $10^{-6}$.
367: 
368:    \begin{table}[hbt] \label{tb:rez12}
369:    \caption{ The BST extrapolation results of the positions of 
370:              closest zeros.}
371:    \begin{center} \begin{tabular}{ccccccc}
372:    &  &  &  &  &  & \\
373:    \hline
374:   $K$ & 0.1 & 0.5 & 1.0 & 2.0 & 2.772589 & 4.540457  \\
375:   $Q_0$ & 0.1 & 0.5 & 1.0 & 2.0 & 3.0 & 8.0  \\
376:   $q_1$ & 0.09998 & 0.504 & 0.95 & 2.05 & 3.02 & 8.00009 \\
377:   $\omega$ & 0.34 & 0.33 & 0.33 & 0.55 & 0.91 & 1.19   \\
378:    \hline
379:    &  &  &  &  &  & \\
380:    \end{tabular} \end{center} \end{table}
381: 
382:    The presented results indicate that: \
383: 
384:    (i) loci of complex-$q$ zeros intersect the positive \req axis at 
385:     value $q_1 = Q_0$. The values agree up to a few percents.
386: 
387: (ii)  Power-law convergence of the closest zero  
388: is observed in all the cases considered. 
389: 
390: The convergence exponent $\omega$ represents the scaling with 
391: size of the distance from the real axis to the closest zero.
392: Before proceeding with the finite-size scaling analysis it is important to 
393: notice that here the scaling is not done with the linear size, but
394: rather with volume, more precisely, with number of spins,
395: since in the infinitely coordinated model the distance as well as the
396: dimensionality lose their meaning.
397: In this respect the considered mean-field limit differs from the standard
398: mean-field approach to the short-range interaction model.
399: As pointed out by Botet \etl \cite{BJP82,BJ83} the basic scaling 
400: quantity in the infinitely coordinated system is not the correlation length, 
401: but the "correlation number"
402: $N_c$, which scales with temperature with the critical exponent
403: $\nu^{*}=\nu_{MF} \cdot d_c$. The exponent $\nu_{MF}$ and $d_c$
404: are, respectively, the mean-field value of the critical exponent $\nu$ 
405:     and 
406: the upper critical dimension (the dimension above which the mean-field regime
407: sets on) of the short-range model.
408: 
409: Thus, the temperature critical exponent for the present model,
410: where only the scaling with the number of spins can be applied
411: will be $1/\nu^{*}$ and not $1/\nu_{MF}$.
412: In the Potts model  $\nu_{MF}=1/2$ both for $q < 2$ and $q=2$, while
413: $d_c=4$ for $q=2$, but $d_c=6$ for $q<2$ for symmetry  reasons.
414: This gives  $\nu^{*}=2$ for $q=2$ and $\nu^{*}=3$ for $q<2$.
415: 
416: The results presented in Table 1 clearly show that
417: in the regime of second-order phase transition ($q \leq 2$) the 
418: convergence exponent coincides up to a few percents 
419: with the temperature critical exponent  $1/\nu^{*}$.
420: This result should be compared to the one in complex temperature plane, 
421: where it was shown \cite{IPZ83}  that the distance of the zero 
422: closest to the real axis scales as the temperature critical exponent.
423: Similar behaviour in a complex-$q$ plane was recently observed in 
424: the $2d$ ferromagnetic Potts model by Kim and Creswick \cite{KC01}, 
425: but the precision of their results did not permit them to be conclusive
426: whether the obtained convergence exponent was indeed equal to the 
427: exponent $1/\nu$. 
428: 
429: In cases $q=3$ and $q=8$ we obtain approximately linear convergence, 
430: proportional to the system size, which is the form of scaling that is 
431: expected for the first-order transition.  
432: 
433: 
434: \subsection{Large $N$ expansion}
435: 
436:    In the limit $N \gg 1$ 
437:    some analytical results may easily be derived 
438:    by extending the mean-field solution by Kihara \etl 
439:    For the large number of spins $N$,
440:    the free energy density, without the terms which are constant 
441:    in the order parameter $S$, is given by  
442:    \bea \label{eq:slen}
443:     & \cF & (K, q, S)  = \\  & - &  \,S^2\,\frac{K}{2}\,\frac{q-1}{q} 
444:     + 
445:    \frac{1+(q-1)S}{q}\,\ln\,\left[1+(q-1)S\right]  +
446:    (q-1)\,\frac{1-S}{q}\,\ln(1-S). \nonumber  
447:    \eea
448:     The partition function is given by 
449:    \be 
450:     Z_N (K, q)  
451:     \sim 
452: %   = \, \exp\left[ NK/2q + N \ln q\right]    
453:     \,\int_0^1 \exp\,(- N\,\cF)\,dS,
454:    \ee
455:     where the proportionality sign stands for all the omitted constant terms.      
456: 
457: We limit our considerations here to the first-order regime of 
458: temperatures, where the free energy density at the transition 
459: shows two equal minima, and the partition function integral can be
460:    approximately solved  by a saddle point method
461:    \be
462:     Z_N \sim \, 1 + \exp\,[-\,N \cF(S_{min})\,]. \label{eq:pf}
463:    \ee
464: The first term on the right hand side denotes the minimum at $S=0$,
465:    and the second term is the 
466: contribution of the second minimum at $S=S_{min} > 0$,
467:    which is obtained from 
468:  the equation for the extremum 
469:    \be
470:    K S = \ln \frac{1+(q-1)S}{1-S}. \label{eq:ext}
471:    \ee
472: 
473:    The straightforward way to find the complex-$q$ zeros would consist
474:    of two steps:
475:     first, by using (\ref{eq:ext}) one should represent $\cF$ as a function 
476: of $K$ and $q$ only;
477:    second,  one should solve the equation $Z_N = 0$ 
478: by analytic continuation of $q$ to the complex plane.
479:    Since $\cF = \cF_R + i \cF_I$ then becomes complex, 
480: the vanishing of $Z_N$, as given by (\ref{eq:pf}), requires
481:    \be \label{eq:uvjeti}
482:    \cF_R   = \, 0, \hspace{1cm} N \cF_I = \, \pm \, (2 m + 1) \pi,
483:     \;\;\;\;m = 0, 1, \ldots
484:    \ee
485:    These two equations determine the position of complex conjugate
486:    pairs of zeros $(q_1^{(m)}, \pm\,q_2^{(m)})$
487:    in the complex-$q$ plane.
488: 
489:    The first of the above mentioned steps cannot be performed in the
490:    straightforward way.
491:    Instead, we will solve (\ref{eq:ext}) in variable $q$
492:    \be
493:    q = \frac{1-S}{S} [ \exp(K S) - 1 ],  \label{eq:qS}
494:    \ee
495: and represent $\cF$ as function of $S$ and $K$.
496:    \be \label{eq:FS}
497:    \cF (K, S) = \frac{K}{2}\frac{S^2}{1-S}
498:    \frac{(1-S)\exp(K S) + 1}{\exp(K S) - 1} + \ln (1-S).
499:    \ee
500:    The equations (\ref{eq:uvjeti}) 
501: are then solved 
502: in the \cS plane and, by equation (\ref{eq:qS}),
503:     mapped back into the complex-$q$ plane.
504:  
505: The solutions $q^{(m)}_{1,2}$ should comply with both equations  
506: (\ref{eq:uvjeti}).
507:   Since the first of equations (\ref{eq:uvjeti}) is $N$-independent, 
508:    it contains all the possible positions of zeros for arbitrary $N$ 
509: including the limit $N\to\infty$. The zeros of the system of
510: particular size $N$ are then determined as intersections of
511: this curve with solutions of the second equation in (\ref{eq:uvjeti}).
512: For large $N$ these intersections appear to be equally distributed along the
513: the curve  $\cF_R   =  0$, so that it can be drawn as the contour 
514: containing zeros.   (see Fig. 2 as an illustration).
515:   The curve crosses the positive \req axis
516:    at a point which is transition point for a given temperature.
517: 
518: Since the locus of zeros has a shape similar to a circle,
519: we draw the circular contour (long dashed line) in order  to stress
520: the difference. 
521: The parameters of this circle are determined by the facts, that its 
522: center lies on the \req axis (since zeros appear in complex conjugate 
523: pairs only) and that it intersects the positive \req axis at $q$ solution 
524: of (\ref{eq:Kexact1}) and the negative \req axis at $q = - |q|$, 
525: solution of
526:      \be
527:       K = 2\,\frac{|q|+1}{|q|+2}\,\ln (|q|+1).   \label{eq:Kexact3}
528:      \ee   
529: 
530: We were not able to give analytical expressions for all the zeros 
531: of the model, but the location of zeros closest to the positive real 
532: $S$ or $q$ axis is a relatively easy task.
533: 
534: Since we limit here to the first-order transitions, 
535:    the temperature will be fixed through the parameter $Q_0 > 2$
536:    \be \label{eq:KQ}
537:     K = 2\,\frac{Q_0-1}{Q_0-2}\,\ln (Q_0-1),
538:    \ee
539: In the thermodynamic limit the second minimum of $\cF (S)$
540:    appears at $S_0 = (Q_0-2)/(Q_0-1)$.
541:    For large but finite $N$, we will expand free energy (\ref{eq:FS})
542:    around the minimum, $S = S_0 + s$.
543:    The small $s \ll S_0$ vanishes as $N\,\to\,\infty$, so we
544:    expect to have, to leading order in $1/N$,
545:    \be \label{eq:s1s2}
546:    s = s_1 + i\,s_2, \hspace{1cm} s_1 = \frac{a}{N^{x_1}},
547:                      \hspace{0.5cm} s_2 = \frac{b}{N^{x_2}},
548:    \ee
549:    for real constants $a, b, x_1, x_2$.
550:    The expansion of (\ref{eq:FS}) and the constraint (\ref{eq:uvjeti})
551:    give the equation
552:    \bea
553:    \cF & = & (Q_0 - 1)\,\left[
554:     1 - \ln\,(Q_0 - 1)\,(A_1 - B_1)
555:          \right] \, s + \cO(s^2)
556:    = 0 \pm\,i\,\frac{2\,m + 1}{N}\,\pi, \nonumber \\
557:    A_1 & = &  \left(\frac{K - Q_0 + 1}{Q_0} +
558:      \frac{2}{Q_0 - 2} \right), \;
559:    B_1 = \frac{Q_0 - 1}{Q_0\,(Q_0 - 2)}\,K - 1.
560:    \eea
561:    The closest zero has the smallest imaginary
562:    part, so its index is $m = 0$.
563:    By inserting (\ref{eq:s1s2}) in the above equation,
564:    one reads
565:    \be \label{eq:konv1}
566:    a = 0, \hspace{1cm}
567:    b = \frac{\pi}{(Q_0 - 1) - \ln\,(Q_0 - 1)\,(A_1 - B_1)},
568:    \hspace{1cm} x_2 = 1,
569:    \ee
570:    while $x_1$ is not defined.
571: Mapping back to the complex-$q$ plane by using the equation (\ref{eq:qS})
572: gives
573:    \bea \label{eq:fp1qN}
574:    q_1^{(0)}(N) & = & Q_0 + \frac{(Q_0 - 1)^2}{Q_0 -2}\,
575:    (K - Q_0)\,\frac{a}{N^{x_1}} \nonumber \\
576:    & + & \frac{(Q_0 - 1)^2}{Q_0 -2}\,\left\{
577:    \frac{K}{2}\,[K - 2\,(Q_0 - 1)] + \frac{Q_0 - 1}{Q_0 - 2}\,(Q_0 - K)
578:    \right\}\,\left(
579:    \frac{a^2}{N^{2\,x_1}} - \frac{b^2}{N^{2\,x_2}}
580:    \right) \nonumber \\
581:    & + & \cO\left(N^{-3\,x}\right), \\
582:    q_2^{(0)}(N) & = &  \frac{(Q_0 - 1)^2}{Q_0 -2}\,
583:    (K - Q_0)\,\frac{b}{N^{x_2}} \nonumber \\
584:    & + & \frac{(Q_0 - 1)^2}{Q_0 -2}\,\left\{
585:    \frac{K}{2}\,[K - 2\,(Q_0 - 1)] + \frac{Q_0 - 1}{Q_0 - 2}\,(Q_0 - K)
586:    \right\}\,\frac{2\,a\,b}{N^{x_1 + x_2}}  \nonumber \\
587:    & + & \cO\left(N^{-3\,x}\right). \nonumber
588:    \eea
589:    Finally, by inserting values for $a, b, x_2$ from (\ref{eq:konv1}),
590:    one obtains the large-$N$ dependence of the 
591: distance of the closest zero from the real axis in the case of the 
592: first-order transitions.
593:    \bea \label{eq:q12N}
594:    q_1^{(0)}(N) & = & Q_0 - \frac{(Q_0 - 1)^2}{Q_0 -2}\,\left\{
595:    \frac{K}{2}\,[K - 2\,(Q_0 - 1)] + \frac{Q_0 - 1}{Q_0 - 2}\,(Q_0 - K)
596:    \right\}\,\frac{b^2}{N^2} \nonumber \\
597:    & + &  \cO\left(N^{-3\,x}\right), \nonumber \\
598:    q_2^{(0)}(N) & = & \frac{(Q_0 - 1)^2}{Q_0 -2}\,
599:    (K - Q_0)\,\frac{b}{N} + \cO\left(N^{-3\,x}\right),
600:    \eea
601:    The above expressions can be checked numerically. 
602: For illustration,  the numerical data of the closest zeros, 
603:    for the temperature
604: critical for $q=8$
605:    are presented in Table 2 
606: for different  sizes $N$.
607: 
608:     \begin{table}[htb]  \label{tb:Nq8}
609:     \caption{ The location of the closest zeros for the temperature 
610: given by the relation (\ref{eq:KQ}) for $Q_0 = 8$.  $N$ denotes the 
611: number of  the system considered.}
612:     \begin{center} \begin{tabular}{ccc}
613:     &  & \\
614:     \hline
615:      $N$ & $q_1^{(m = 0)}(N)$ & $q_2^{(m = 0)}(N)$  \\
616:    5000&7.99999543669504&$ -8.679091351481774\,\cdot\,10^{-3}$ \\
617:    6000&7.99999683103614&$ -7.232577480179549\,\cdot\,10^{-3}$ \\
618:    7000&7.99999767178073&$ -6.199352825621011\,\cdot\,10^{-3}$ \\
619:    8000&7.99999821745666&$ -5.424434119811201\,\cdot\,10^{-3}$ \\
620:    9000&7.99999859157044&$ -4.821719459791061\,\cdot\,10^{-3}$ \\
621:   10000&7.99999885917194&$ -4.339547669719317\,\cdot\,10^{-3}$ \\
622:   11000&7.99999905716679&$ -3.945043440975050\,\cdot\,10^{-3}$ \\
623:   12000&7.99999920775815&$ -3.616289894012384\,\cdot\,10^{-3}$ \\
624:   13000&7.99999932495365&$ -3.338113800831173\,\cdot\,10^{-3}$ \\
625:   14000&7.99999941794470&$ -3.099677139486969\,\cdot\,10^{-3}$ \\
626:   15000&7.99999949296515&$ -2.893032025972416\,\cdot\,10^{-3}$ \\
627:   \hline \end{tabular} \end{center}
628:    \end{table}
629: 
630: The data $q_{1, 2}^{(0)}(N)$ from Table 2 and analogous series of data for 
631: different temperatures corresponding to $q=3$ and $q=2.1$ were extrapolated 
632: to the thermodynamic limit by using the BST algorithm. 
633: 
634: The extrapolation  parameter $\omega$ was fixed by the constraint 
635: $q_2^{(0)}(N)\,\to\,0$.
636: After we have made the extrapolations of $q_{1, 2}^{(0)}(N)$
637: and found the values of $q_1(\infty)$ and $\omega$, we have performed 
638: alternative calculation of convergence exponents of  $q_{1, 2}^{(0)}(N)$.
639: By assuming the convergence of the $q_{1, 2}^{(0)}(N)$ to be of the form  
640:    \be \label{eq:q12alfa}
641:     q_1^{(0)}(N)  - 
642:     q_1(\infty) = C_1\,\cdot\,N^{-\alpha_1 } + \cdots,
643:     \hspace{0.9cm}
644:     q_2^{(0)}(N) = C_2\,\cdot\,N^{-\alpha_2 } + \cdots,
645:    \ee
646: the convergence exponents $\alpha_{1, 2}(N)$ for real and imaginary part 
647: of the closest zeros, can be expressed as
648:    \bea \label{eq:alfa12}
649:    \alpha_1(N) & = &  - \frac{\ln\left\{\left[q_1^{(0)}(N) - q_1(\infty)\right]/
650:    \left[q_1^{(0)}(N + 1) - q_1(\infty)\right]
651:    \right\}}{\ln[N/(N + 1)]},  \nonumber \\
652:    \alpha_2(N) & = & - \frac{\ln\left[
653:    q_2^{(0)}(N) / q_2^{(0)}(N + 1) \right] }{\ln[N/(N + 1)]} .
654:    \eea
655:    The above data for $\alpha_{1, 2}(N)$ are extrapolated
656: again by the BST algorithm.
657:    The complete set of extrapolated results is summarised 
658: in the Table 3. 
659: 
660:    \begin{table}[hbt] \label{tb:bst1ord}  
661:    \caption{ The results of the BST extrapolations performed on the
662:     closest zeros, $q_{1, 2}(\infty)$, followed by parameter
663:     $\omega$ and convergence exponents $\alpha_{1, 2}$,
664:     for set of temperatures determined by $Q_0$ and (\ref{eq:KQ}).
665:     The predicted convergence exponents $x_{1, 2}$ as given
666:     by (\ref{eq:q12N}) are in the
667:     last two columns. }
668:     \begin{center}
669:     \begin{tabular}{ cccccccc }
670:     & & & & & & & \\
671:     \hline
672:    $Q_0$ & $q_1(\infty)$ & $q_2(\infty)$ & $\omega$ &
673:    $\alpha_1$ & $\alpha_2$ & $x_1$ & $x_2$  \\
674:     8 & 8.0000000000001 & $3\,\cdot\,10^{-13}$ & 1.0000 &
675:     1.9999988 & 1.000000002 & 2 & 1 \\
676:     3 & 3.000000000005 & $4\,\cdot\,10^{-12}$ & 1.00000 &
677:     2.00003 & 0.99999997 & 2 & 1 \\
678:     2.1 & 2.1002 & $9\,\cdot\,10^{-6}$ & 0.98 & 1.85 & 0.95 & 2 & 1 \\
679:     \hline
680:     & & & & & & & \\
681:    \end{tabular}
682:    \end{center}
683:    \end{table}
684: 
685:   We observe a full agreement between the analytically given values for 
686: $q$ and convergence exponents on one side, and the numerically
687:   calculated and extrapolated values for $q$ and exponents on the other side.
688: Although the scaling in the case of the first-order phase transition 
689: is trivial and linearly proportional to system size, the results can be used
690: to check the convergence of numerical results. 
691: For $q=3$ and $q=8$ the convergence exponent has improved compared to 
692: the earlier numerical results given in Table 1, confirming that larger 
693: error in this case comes from the small size of considered systems and 
694: that scaling is indeed linear with size as expected for the first-order 
695: phase transition.
696:    The loss of precision in extrapolations persists only for the
697: case corresponding to $q=2.1$ and should be attributed to the 
698: crossover effects due to vicinity of  the border between the first- 
699: and second-order transitions.
700: 
701:      \vspace{1cm}
702: 
703:      For the partition function zeros lying on the curve in the appropriate
704:      complex plane, the normalised density of zeros $g$, is defined as
705:       \cite{LY52}
706:      \be
707:      \label{eq:defg}
708:      g = \frac{1}{N}\,\frac{d n}{d l}, \hspace{1cm} \int\,g\,dl = 1,
709:      \ee
710:      where $d n$ is the number of zeros inside the arc of the curve of
711:      the length $d l$. In the case of the field- or temperature-driven
712:      second-order transitions, the density of zeros vanishes 
713: at the transition with an exponent connected to
714:      critical exponents \cite{CK97}. On the contrary, at the first-order
715:      transition point, the density of zeros remains constant.
716:      To the end of this paragraph, we will examine the behaviour of 
717: density of complex-$q$ zeros in the vicinity of first-order transition point.
718:      To this purpose, we notice that the large-$N$ expansion
719:      of $q_{1, 2}(N)$ for the closest zeros (those with $m =0$)
720: presented earlier in text, 
721:  may be directly generalised to other values of $m$
722: as long as  the condition $(2 m + 1) \ll N$ is fulfilled. 
723: This leads to
724:    \bea
725:     q_1^{(m)}(N) - q_1(\infty) & = &
726:      C_1\,\cdot\,\left(
727:     \frac{2 m + 1}{N}\right)^{\alpha_1}  + \cdots,
728:      \nonumber \\
729:     q_2^{(m)}(N) & = & C_2\,\cdot\,\left(
730:     \frac{2 m + 1}{N}\right)^{\alpha_2} + \cdots,
731:    \eea
732:      with the constants $q_1(\infty), C_1$ and $C_2$ given by
733:      the relations (\ref{eq:q12N}), where $1/N$ is changed by
734:      $(2 m + 1)/N$.
735: 
736:      To calculate numerically density of zeros (\ref{eq:defg}), 
737:      we choose two neighbouring zeros, with $d l$ denoting 
738:      geometrical distance between them, \\
739:      $ d l = \left\{ \left[q_1^{(m + 1)} - q_1^{(m)}\right]^2 +
740:      \left[q_2^{(m + 1)} - q_2^{(m)}\right]^2\right\}^{1/2}$.
741:      In that case density of zeros near positive real-$q$ axis becomes
742:      \bea
743:      g(m) = & \frac{1}{N} & \,\left\{
744:      C_1^2\,\left(\frac{2 m + 1}{N}\right)^{2\,\alpha_1}\,\left[\left(1 +
745:      \frac{2}{2 m + 1}\right)^{\alpha_1} - 1\right]^2
746:      \right. \nonumber \\
747:      & + & \left.
748:      C_2^2\,\left(\frac{2 m + 1}{N}\right)^{2\,\alpha_2}\,\left[\left(1 +
749:      \frac{2}{2 m + 1}\right)^{\alpha_2} - 1\right]^2
750:      \right\}^{-1/2}.
751:      \eea
752:      The edge $(m = 0)$ density of zeros $g_0$, tends to constant value 
753:      in the thermodynamic limit.   
754:      \be \label{eq:gN1}
755:      g_0 = \frac{1}{2\,C_2}\,\left[
756:      1 - 8\,\frac{C_1^2}{C_2^2}\, \frac{1}{N^2} + \cdots
757:      \right] \;\to\; \frac{1}{2\,C_2},
758:      \ee
759: 
760: For example, for $Q_0 = 8$ and $K(Q_0 = 8) = 4.540457014462$ 
761: this constant is
762:      \be \label{eq:constg}
763:      \frac{1}{2\,C_2} = \left[ 2\,\frac{(Q_0 - 1)^2}{Q_0 -2}\,
764:      (K - Q_0)\,b \right]^{-1}
765:       =  0.011521939 \cdots   ,
766:      \ee
767: For illustration, we check this result numerically by calculating    
768:   for the values of $Q_0$ and $K$
769:      from above, and $N = 10\,000$,  the
770:      twenty zeros closest to the positive \req axis and corresponding densities.
771: They are presented in Table 4. 
772: 
773:    \renewcommand              
774: %% \baselinestretch 2              %  s proredom 
775: %% \baselineskip 24pt              %                     
776:    \baselinestretch 1              %   bez proreda 
777:    \baselineskip 15pt              %                     
778:    \begin{table} \label{tb:gq8}
779:    \caption{ The twenty zeros closest to the positive \req axis
780:    and the corresponding densities are
781:    shown at temperature determined from the relation
782:     (\ref{eq:KQ}) and $Q_0 = 8$ and for the $ N = 10\,000$ spins.}
783:    \begin{center} \begin{tabular}{cccc}
784:       & & & \\
785:    \hline 
786:   $m$ & $q_1^{(m)}$ & $q_2^{(m)}$ & $g(m)$ \\
787:   0&7.99999885917192&-0.004339547669719&0.0115219529910602 \\
788:   1&7.99998973259138&-0.013018627057442&0.0115219974088187 \\
789:   2&7.99997147966529&-0.021697658591626&0.0115220714335465 \\
790:   3&7.99994410086352&-0.030376610375500&0.0115221750581617 \\
791:   4&7.99990759689083&-0.039055450520507&0.0115223082726076 \\
792:   5&7.99986196868649&-0.047734147149695&0.0115224710640261 \\
793:   6&7.99980721742419&-0.056412668400979&0.0115226634165474 \\
794:   7&7.99974334451167&-0.065090982430549&0.0115228853117394 \\
795:   8&7.99967035159032&-0.073769057415945&0.0115231367280630 \\
796:   9&7.99958824053479&-0.082446861559540&0.0115234176412815 \\
797:  10&7.99949701345249&-0.091124363091702&0.0115237280243068 \\
798:  11&7.99939667268302&-0.099801530274076&0.0115240678473335 \\
799:  12&7.99928722079755&-0.108478331402754&0.0115244370775645 \\
800:  13&7.99916866059815&-0.117154734811648&0.0115248356796385 \\
801:  14&7.99904099511710&-0.125830708875537&0.0115252636152771 \\
802:  15&7.99890422761603&-0.134506222013369&0.0115257208435568 \\
803:  16&7.99875836158505&-0.143181242691353&0.0115262073207999 \\
804:  17&7.99860340074195&-0.151855739426127&0.0115267230005962 \\
805:  18&7.99843934903115&-0.160529680787899&0.0115272678338162 \\
806:  19&7.99826621062264&-0.169203035403569&0.0115278417687404  \\ 
807:    \hline 
808:       & & & \\
809:    \end{tabular} \end{center} \end{table}
810:    \renewcommand              
811: %%  \baselinestretch 2              %  s proredom 
812: %%  \baselineskip 24pt              %                     
813:    \baselinestretch 1              %   bez proreda 
814:    \baselineskip 15pt              %                     
815: 
816:      The limiting value of the density at the transition point, is extracted
817:       by the BST extrapolation procedure (\ref{eq:seq}) simply
818:      by changing variable $1/N$ into $(2\,m + 1)/N$.
819:      \be \label{eq:g0}
820:      g_m = g_0 +{\tilde g}\,\left(\frac{2\,m + 1}{N}\right)^{\omega} + \cdots.
821:      \ee
822:      The extrapolations give $g_0 = 1.152194\,\cdot\,10^{-2}$ for a 
823:      wide range of $0.3 < \omega < 3.0$, which reproduces the analytically 
824: obtained value (\ref{eq:constg}) in six significant digits. 
825: 
826:  We can conclude that the density of the partition function zeros in the
827:  complex-$q$
828:   plane behave in similar way as the density of zeros in the
829:   complex-temperature \cite{F65} or complex-field \cite{CK97} planes:
830:   in thermodynamic limit
831:   it has constant value at the first-order transition point.
832: 
833: \section{The power-law  decaying interactions} \label{sec-lr}  
834: 
835: The second case that we have considered is the ferromagnetic Potts model 
836: in one dimension with power-law decaying interactions. When taking the 
837: periodic boundary conditions, it is described by the reduced Hamiltonian
838:   \be
839:    \cH_{PL} = - \frac{H_{PL}}{k_B\,T} = 
840:    K \sum_{i=1}^{N-1}\;\sum_{j=1}^{N-i}\; \delta\,(s_i, s_{i+j}) 
841:    \left[\frac{1}{j^{1+\sigma}}+\frac{1}{(N-j)^{1+\sigma}}\right]. 
842:   \ee
843: Although in one dimension, the model is nontrivial and has a phase 
844: transition at nonzero temperature for all $q$ when 
845:  $0 < \sigma \leq 1$ \cite{D69,ACCN88,GU93}. 
846: This transition is of the mean-field type for low enough values of 
847: $\sigma$,  $\sigma< \sigma_c(q)$, and is a
848: non-trivial second-order phase transition for $\sigma > \sigma_c(q)$.
849: In the classical regime $\sigma < \sigma_c(q)$, the transition is of the 
850: first order when $q > 2$ \cite{UG97,GU98,BDD99}.
851: The exact position of the line separating the two regimes, $\sigma_c(q)$, 
852:  is a difficult and still open question \cite{GU98,BDD99,UG00}.
853: 
854:  In  the graph representation, the difference from the mean-field case is that 
855: the contribution of active links depends on distance between the spins 
856: they connect.
857: The contribution of active bonds may then have $N/2$ (for $N$ even) 
858:  or $(N - 1)/2$ (for $N$ odd) distinct values
859: given by $v_j = \exp[K/j^{1+\sigma}+K/(N-j)^{1+\sigma}] - 1$.
860: If we denote by  $b_j(\cG)$ the number of active links of the length 
861: $j$ or $N - j$ in a graph $\cG$, the partition function becomes  
862:    \be \label{eq:zlr}   
863:    Z_N = \sum_{{\rm all}\;\cG} \, v_1^{b_1(\cG)}\,v_2^{b_2(\cG)}\,\cdots
864:    \,q^{n(\cG)} 
865:    \ee
866:     or, in form of polynomial in $q$, 
867:    \be \label{eq:Npollr}   
868:    Z_N = \sum_{n = 1}^N\,a_n(v_j, N)\;q^n.
869:    \ee
870:    The real and positive coefficients $a_n$ depend on temperature and  $N$, 
871: but also on the parameter of range $\sigma$. 
872: This makes the calculations more demanding for the computer memory.
873: 
874:  We have performed exact numerical calculations on chains with 
875: $N \leq 9$ spins for two sets of parameters,  
876:  ($\sigma = 0.3$, $K = 0.576$) and ($\sigma = 0.8$ $K = 0.8230$). 
877: As in the previous case, the chosen temperatures are the critical
878: temperatures known for certain values of $q$. They correspond to 
879: $q = 5$ and  $q = 2$ respectively.
880: Remark, however, that in the present case the critical temperatures
881: are not known exactly. The values used here were obtained by the finite-range 
882: scaling approach \cite{GU93}.
883: The first and second set correspond respectively to the  to the first- 
884: and second-order phase transition.
885: 
886: The numerical results for the  loci of complex-$q$ zeros for sizes
887:  $N = 6, \cdots, 9$ 
888: are presented in Figures 3 and 4.
889:   In both examples are obtained the zeros lying on the arc-shaped curves,
890: while the positive real axis is free of zeros. The zeros closest to it
891: approach to it with increasing $N$.
892: 
893: As in the mean-field case, we have examined the convergence of the
894: closest zeros
895: to the thermodynamic limit  by the BST algorithm, assuming the power-law form.
896: The analysis included the data obtained for chains with $N = 5$ to $9$. 
897: The parameter $\omega$ was determined, as earlier, by requiring that
898:  $q_2 \,\to\,0$ in the thermodynamic limit. 
899: The results are presented in Table 5. 
900: 
901:    \renewcommand              
902: %% \baselinestretch 2              %  s proredom 
903: %% \baselineskip 24pt              %                     
904:    \baselinestretch 1              %   bez proreda 
905:    \baselineskip 15pt              %                     
906:    \begin{table}[hbt]  \label{tb:bstLR}
907:    \caption{ The loci of closest zeros and its BST extrapolation at
908:     $(K = 0.576, \sigma = 0.3)$ and $(K = 0.823, \sigma = 0.8)$} 
909:    \begin{center} \begin{tabular}{cccc} \hline
910:       & $ K = 0.576$ & $\sigma = 0.3$ &  \\ \hline  
911:   $N$ & $q_1^{(0)}(N)$ &  $q_2^{(0)}(N)$ & $\omega$  \\
912:     5 & -0.447413829093298& 2.78806303559461 & \\
913:     6 & 0.0734103695976816& 2.88316232503579 & \\
914:     7 & 0.486108948338677& 2.88942370075814 & \\
915:     8 & 0.818954244961081& 2.85269647435491 & \\   
916:     9 & 1.09269817849075& 2.79489503015230 & \\ 
917:  $\infty$ & 4.2        & $3\,\cdot\,10^{-3}$  & 0.89 \\ \hline 
918:     & $ K = 0.823$ & $\sigma = 0.8$  \\ \hline
919:   $N$ & $q_1^{(0)}(N)$ &  $q_2^{(0)}(N)$ & $\omega$ \\  
920:     5 &-0.418572384698087 & 3.69577294340190 & \\
921:     6 & 0.179124021423254 & 3.55212427658784 & \\    
922:     7 & 0.581821465484680 & 3.37723067427777 & \\   
923:     8 & 0.866098195024489 & 3.20669717997192 & \\
924:     9 & 1.074705295884560 & 3.05042285906340 & \\  
925:  $\infty$ & 2.01       & $3\,\cdot\,10^{-3}$  & 0.41 \\  
926:   \hline \end{tabular} \end{center} \end{table}
927:    \renewcommand              
928: %%  \baselinestretch 2              %  s proredom 
929: %%  \baselineskip 24pt              %                     
930:    \baselinestretch 1              %   bez proreda 
931:    \baselineskip 15pt              %                     
932:  
933: In the case of second order phase transition the expected 
934: crossing point with the real axis is obtained with precision better than 1\%.
935: The leading convergence exponent given by $\omega$, 
936: is equal to 0.41, 
937: which is close to the value for the temperature critical exponent 
938: $1/\nu =0.48$ obtained earlier by finite-range scaling
939: for $q=2$ and $\sigma=0.8$ \cite{GU93}. 
940: 
941: In the case of the first-order transition the fit is more difficult.
942: The discrepancy between the obtained crossing point and the
943: value expected for the considered temperature 
944: is larger, about 20\%, while the convergence exponent, which 
945: in context of finite-size scaling should be linear for a first-order 
946: phase transition, differs by about 11\%. 
947: Larger error in determination
948: of the value of $q$, may be attributed to both lower precision of
949: the data for critical temperature obtained by finite-range scaling
950:  \cite{GU93,GU98} 
951: when the values of $\sigma$ are low and to  the proximity of
952: the second-order transition regime in the $(q,\sigma)$ plane, 
953: which might induce crossover effects for small sizes.
954: 
955:  Since in these preliminary calculations we dispose with small number 
956: of data, we have not calculated the density of zeros in the case of 
957:   power-law model.  
958:         
959: 
960:   \section{Conclusion} \label{sec-zklj}  
961: 
962:   We have studied the \cq zeros of the partition function of the ferromagnetic
963:   Potts model with long-range interactions in context of the first- and
964:   second-order phase transitions in this model.
965:   Two cases are presented: the mean-field limit and a few examples
966:   of the case with power-law decaying interactions.
967: 
968:   We conclude that the positions of the zeros may be analysed in a 
969:   similar way as the zeros in the plane of complex field or 
970:   temperature and that they can give useful informations on the 
971:   phase transition in ferromagnetic Potts models.   
972: 
973: Numerical results for relatively small systems show the similar picture 
974: as in the case of zeros in the complex field or temperature plane. 
975: In all the cases considered the zeros lie on smooth curves
976: that intersect the real positive axis at value for which the given 
977: temperature is the transition temperature.
978: For finite systems the positive real axis is free of zeros, but they 
979: approach to it with increasing size.
980: 
981:  The distance of the closest zero from the real axis is found to scale
982:  with size as the temperature critical exponent $1/\nu$.
983:  In the mean-field case, this result could be obtained  with a good precision
984:  for series of different temperatures corresponding to different transition
985:  regimes. In the case of power-law interactions, the preliminary results
986:  obtained with less precision suggest the same behaviour.
987:  This result is similar to behaviour in the complex temperature plane,
988:  and it supports recent findings by Kim and Creswick \cite{KC01}
989:  in the short-range Potts model in $2d$.  
990: 
991: In the end, let us mention some technical advantages of the analysis 
992: by complex-$q$ zeros compared to the complex-temperature zeros
993: in the model with long-range interactions.
994:   The partition function of the mean-field model consisting of $N$
995:    spins can
996: be written as a $N$-th order polynomial in $q$, or roughly 
997: $N^2$-th order polynomial in temperature variable. Consequently, 
998: this allows better numerical precision of calculations in the complex-$q$
999: plane then those in the complex-temperature plane.
1000:   For a model with power-law decaying interactions the possibilities 
1001: in complex temperature plane are even more restricted. 
1002: Model with $N$ spins, has $N/2$ (for $N$ even) 
1003:  different parameters involving 
1004: temperature, so that the partition function cannot be defined as a 
1005: polynomial in some temperature variable at all.
1006: 
1007:  Further investigations of the complex-$q$ zeros should be done in
1008:  future in at least two directions.
1009:  We plan to perform a more extended numerical calculations of the
1010:  complex-$q$ zeros in  the case of power-law decaying interactions.
1011:  This should allow us to obtain the convergence exponent for the closest 
1012:  zero with more  precision. 
1013:  In addition, it would be interesting to examine a still
1014:  open question of the borderline $q_{c}(\sigma)$ dividing
1015:  the first- from the second-order transition regimes within the
1016:  formalism of complex-$q$ zeros.
1017:  Also, on the basis of the results obtained in this paper, a possible
1018:  connection between the zeros in \cq and complex-temperature
1019:  plane deserves further investigation.
1020: 
1021:  \newpage 
1022: 
1023:  \begin{thebibliography}{9} 
1024: %%  \baselineskip=12pt plus 0.2pt minus 0.2pt
1025:    \renewcommand              
1026: %%   \baselinestretch 2              %  s proredom 
1027: %%   \baselineskip 24pt              %                     
1028:    \baselinestretch 1              %   bez proreda 
1029:    \baselineskip 15pt              %                     
1030:  \bibitem{YL52} C. N. Yang and T. D. Lee, Phys. Rev. {\bf 87} (1952) 404. 
1031:  \bibitem{LY52} T. D. Lee and C. N. Yang, Phys. Rev. {\bf 87} (1952) 410.
1032:  \bibitem{F65} M. Fisher, in Lectures in theoretical physics vol. 12C p. 1 
1033:                (University of Colorado Press, Boulder, 1965).        
1034:  \bibitem{OKSK68} S. Ono, Y. Karaki, M. Suzuki and C. Kawabata, 
1035:                J. Phys. Soc. Japan {\bf 25} (1968) 54.    
1036:  \bibitem{KC98} S-Y. Kim and R. J. Creswick, \PRL\ {\bf 81} (1998) 2000.  
1037:  \bibitem{KC99} S-Y. Kim and R. J. Creswick, Physica A {\bf 281} (2000) 252. 
1038:  \bibitem{B90} G. Bhanot, \JSP\ {\bf 60} (1990) 55.  
1039:  \bibitem{M91} P. Martin, Potts models and related problems in
1040:                statistical mechanics, World Scientific Publishing
1041:                Co. Pte. Ltd., 1991, Ch. XI, and references therein.
1042:  \bibitem{CHW96} C-N. Chen, C-K. Hu and F. Y. Wu, 
1043:                        \PRL\ {\bf 76} (1996) 169. 
1044:  \bibitem{KCCH99} S-Y. Kim, R. J. Creswick, C-N. Chen and C-K. Hu, 
1045:                        Physica A {\bf 281} (2000) 262.  
1046:  \bibitem{FK69} P. W. Kasteleyn and C. M. Fortuin, J. Phys. Soc. Jpn. 
1047:                Suppl. {\bf 26} (1969) 11; 
1048:                C. M. Fortuin and P. W. Kasteleyn, Physica {\bf 57} (1972) 536.   
1049:  \bibitem{Bx86} R. J. Baxter, \JPA\ {\bf 19} (1986) 2821.
1050:  \bibitem{Bx87} R. J. Baxter, \JPA\ {\bf 20} (1987) 5241.
1051:  \bibitem{ST97a} R. Shrock and S-H. Tsai, \PRE\ {\bf 55} (1997) 5165. 
1052:  \bibitem{ST97b} R. Shrock and S-H. Tsai, \PRE\ {\bf 56} (1997) 1342.  
1053:  \bibitem{RST98} M. Ro\v cek, R. Shrock and S-H. Tsai, \PA\ {\bf 259}
1054: 	       (1998) 367. 
1055:  \bibitem{ST99a} R. Shrock and S-H. Tsai, \PA\ {\bf 275} (1999) 429. 
1056:  \bibitem{ST99b} R. Shrock and S-H. Tsai, \JPA\ {\bf 32} (1999) L195. 
1057:  \bibitem{S99} A. D. Sokal, Physica A {\bf 279} (2000) 324.
1058:  \bibitem{SS00} J. Salas and A. D. Sokal, cond-mat/0004330, 
1059:                       to be published in J. Stat. Phys.   
1060:  \bibitem{CS00} S-C. Chang and R. Shrock,
1061:                       Physica A {\bf 286} (2000) 189.
1062:  \bibitem{KC01} S-Y. Kim and R. J. Creswick, \PRE\ {\bf 63} (2001) 0066107.
1063:  \bibitem{W82} F. Y. Wu,  Rev. Mod. Phys. {\bf 54} (1982) 235.
1064:  \bibitem{KMS54} T. Kihara, Y. Midzuno and T. Shizume,
1065:       J. Phys. Soc. Jpn. {\bf 9} (1954) 681. 
1066:  \bibitem{SF73} J. P. Straley and M. E. Fisher, \JPA\ {\bf 6} (1973) 1310.
1067:  \bibitem{MS74} L. Mittag and J. Stephen, \JPA\ {\bf 7} (1974) L109.
1068:  \bibitem{PL76} R. G. Priest and T. C. Lubensky, \PRB\ {\bf 13} (1976) 4159.
1069:  \bibitem{LB96} E. Luijten and H. W. J. Bl\"ote, \PRL\ {\bf 76} (1996) 1557.
1070:  \bibitem{D69} F. J. Dyson, Commun. Math. Phys. {\bf 12} (1969) 91.  
1071:  \bibitem{ACCN88} M. Aizenman, J. T. Chayes, L. Chayes and C. M. Newman, 
1072:                   \JSP\ {\bf 50} (1988) 1. 
1073:  \bibitem{GU93} Z. Glumac and K. Uzelac, \JPA\ {\bf 26} (1993) 5267.  
1074:  \bibitem{LB95} E. Luijten and H. W. J. Bl\"ote, 
1075:                 Int. J. Mod. Phys. C {\bf 6} (1995) 359.
1076:  \bibitem{C95} S. A. Cannas, \PRB\ {\bf 52} (1995) 3034; 
1077:                S. A. Cannas, A. C. N. de Magalh\~aes and 
1078:                F. A. Tamarit, \PRB\ {\bf 61} (2000) 11521. 
1079:  \bibitem{UG97} K. Uzelac and Z. Glumac, Fizika B {\bf 6} (1997) 133.
1080:  \bibitem{GU98} Z. Glumac and K. Uzelac, \PRE\ {\bf 58} (1998) 4372. 
1081:  \bibitem{M99} J. L. Monroe, \JPA\ {\bf 32} (1999) 7803.   
1082:  \bibitem{BDD99} E. Bayong, H. T. Diep and V. Dotsenko, 
1083:                        \PRL\ {\bf 83} (1999) 14. 
1084:  \bibitem{KL00} M. Krech and E. Luijten, \PRE\ {\bf 61} (2000) 2058.
1085:  \bibitem{LM01} E. Luijten and H. Me{\ss}ingfeld, \PRL\ {\bf 86} (2001) 5305.
1086:  \bibitem{BS64} R. Bulirsch and J. Stoer, Numer. Math. {\bf 6} (1964) 413. 
1087:  \bibitem{HS88} M. Henkel and G. Sch{\" u}tz, \JPA\ {\bf 21} (1988) 2617.  
1088:  \bibitem{VBS79} J. M. Vanden Broeck and L. W. Schwartz, 
1089:         SIAM J. Math. Anal. {\bf 10} (1979) 658.  
1090:  \bibitem{BJP82} R. Botet, R. Jullien and P. Pfeuty, \PRL\ {\bf 49} (1982) 478.
1091:  \bibitem{BJ83} R. Botet and R. Jullien, \PRB\ {\bf 28} (1983) 3955.
1092:  \bibitem{IPZ83} C. Itzykson, R.B. Pearson, and J. B. Zuber, 
1093:           Nucl. Phys. B {\bf 220}, 415 (1983)
1094:  \bibitem{CK97} R. J. Creswick and S-Y. Kim, \PRE\ {\bf 56} (1997) 2418.  
1095:  \bibitem{UG00} K. Uzelac and Z. Glumac, \PRL\ {\bf 85} (2000) 5255.
1096: 
1097: \end{thebibliography}
1098: 
1099:   \newpage
1100: 
1101:   \centerline{\Large Figure caption}
1102: 
1103:   \vspace{1cm}
1104: 
1105:   {\bf Fig. 1}: The loci of complex-$q$ zeros of the partition function of
1106:               the mean-field Potts
1107:               model calculated at inverse temperatures
1108:               (a) $K = K_c(q = 0.5) = 0.5$, (b) $K = K_c(q = 2) = 2$ and 
1109:               (c) $K = K_t(q = 8.0) = 4.540457$ for 
1110:               $N = 25$ (open diamonds), $N = 30$ (filled triangles) 
1111:               and $N = 35$ (filled circles) spins.
1112: 
1113:   {\bf Fig. 2}: The locus of complex-$q$ zeros of the partition function of
1114:               the mean-field Potts
1115:               model calculated at inverse temperature
1116:               $K = K_{MF}(q = 8) = 4.540457$.     
1117:               The full line denotes loci of zeros in the limit $N\to\infty$, 
1118: obtained by the approach by the saddle point approach
1119: as a solution of the equation $\cF_R = 0$.
1120:    Long dashed line is a circle (described in text) drawn as a guide to the eye.
1121: 
1122:   {\bf Fig. 3}: The complex-$q$ zeros of the partition function of
1123:                 the $1 d$ 
1124: Potts model with power-law interactions at the $FRS$ estimate of first-order
1125:               transition temperature
1126:               $K = K_t(q = 5, \sigma = 0.3) = 0.576$,
1127:               for 
1128:               $N = 6$ (open diamonds), $N = 7$ (filled triangles),
1129:               $N = 8$ (open squares) and $N = 9$ (filled circles) spins.
1130: 
1131:   {\bf Fig. 4}: The complex-$q$ zeros of the partition function of the $1 d$ 
1132:         Potts model with power-law interactions at the $FRS$ 
1133:         estimate of second-order transition temperature  
1134:               $K = K_c(q = 2, \sigma = 0.8) = 0.8230$,
1135:               for 
1136:               $N = 6$ (open diamonds), $N = 7$ (filled triangles),
1137:               $N = 8$ (open squares) and $N = 9$ (filled circles) 
1138:               spins.   
1139: 
1140: 
1141: 
1142: \vfill
1143: \eject
1144: 
1145:    \centerline{\includegraphics{Fig1a.eps}}
1146: 
1147:    \centerline{\includegraphics{Fig1b.eps}}
1148: 
1149:    \centerline{\includegraphics{Fig1c.eps}}
1150: 
1151:    \centerline{\includegraphics{Fig2.eps}}
1152: 
1153:    \centerline{\includegraphics{Fig3.eps}}
1154: 
1155:    \centerline{\includegraphics{Fig4.eps}}
1156: 
1157:     \end{document}
1158:     
1159: 
1160: