cond-mat0502062/ht.tex
1: \documentclass[prl,twocolumn]{revtex4}
2: 
3: \usepackage{psfrag}
4: \usepackage[T1]{fontenc}
5: 
6: %\RequirePackage{FuturaLT}
7: \usepackage{times}
8: \usepackage{pifont}
9: \usepackage{ae,aecompl}
10: \usepackage{amsmath}
11: %\usepackage{concmath}
12: %\usepackage{epsf}
13: \usepackage{graphicx}
14: 
15: \newcommand{\dd}{\mbox{d}}
16: \newcommand{\DD}{\mbox{D}}
17: 
18: \begin{document}
19: 
20: 
21: \title{Fractal Structure of High-Temperature Graphs of O($N$) Models in
22:   Two Dimensions}
23: %
24: \author{Wolfhard Janke and Adriaan M. J. Schakel}
25: %
26: \affiliation{Institut f\"ur Theoretische Physik, Universit\"at Leipzig,
27:   Augustusplatz 10/11, 04109 Leipzig, Germany }
28: %
29: 
30: \begin{abstract}
31:   The critical behavior of the two-dimensional O($N)$ model close to
32:   criticality is shown to be encoded in the fractal structure of the
33:   high-temperature graphs of the model.  Based on Monte Carlo
34:   simulations and with the help of percolation theory, De Gennes'
35:   results for polymer rings, corresponding to the limit $N \to 0$, are
36:   generalized to random loops for arbitrary $-2 \leq N \leq 2$.  The
37:   loops are studied also close to their tricritical point, known as the
38:   $\Theta$ point in the context of polymers, where they collapse.  The
39:   corresponding fractal dimensions are argued to be in one-to-one
40:   correspondence with those at the critical point, leading to an analytic
41:   prediction for the magnetic scaling dimension at the O($N)$
42:   tricritical point.
43: \end{abstract}
44: 
45: \date{\today}
46: 
47: \maketitle
48: 
49: The high-temperature (HT) representation of the critical O($N$) spin
50: model naturally defines a loop gas, corresponding to a diagrammatic
51: expansion of the partition function in terms of closed graphs along the
52: bonds on the underlying lattice \cite{Stanley}.  In the limit $N\to 0$,
53: the loops reduce to closed self-avoiding random walks first considered
54: by de Gennes as a model for polymer rings in good solvents at
55: sufficiently high temperatures, so that the van der Waals attraction
56: between monomers is irrelevant \cite{deGennes}.  In his seminal paper,
57: de Gennes related the fractal structure of self-avoiding random walks to
58: the critical exponents of the O($N\to0$) model.  Invoking concepts from
59: percolation theory and recent Monte Carlo (MC) data
60: \cite{geoPotts,Dotsenko} of the HT representation of the two-dimensional
61: (2D) Ising model ($N=1$), we extend in this Letter de Gennes' result to
62: arbitrary $-2 \leq N \leq 2$ in 2D.  We consider the loops also close to
63: their tricritical point where they collapse, known as the $\Theta$ point
64: in the context of polymers.  We argue that the 2D fractal dimensions at
65: the tricritical point are in one-to-one correspondence with those at the
66: critical point, allowing us to also predict the magnetic scaling
67: dimension at the O($N)$ tricritical point.  We support our theoretical
68: prediction by comparing it with recent high-precision MC data
69: \cite{bloeteetal}.
70: 
71: A particularly simple representation of the O($N$) universality class is
72: specified by the partition function \cite{DMNS}
73: \begin{equation}
74:   \label{ZS}
75:   Z = \mathrm{Tr} \prod_{\langle \mathbf{x},\mathbf{x}' \rangle} (1 + K
76:   \mathbf{S}_\mathbf{x} \cdot \mathbf{S}_{\mathbf{x}'}),
77: \end{equation} 
78: where the product is over all nearest neighbor pairs, and the spins have
79: $N$ components $\mathbf{S}_\mathbf{x} = (S^1_\mathbf{x}, S^2_\mathbf{x},
80: \dots , S^N_\mathbf{x})$ and are of fixed length
81: $\mathbf{S}_\mathbf{x}^2=N$.  The trace Tr stands for the sum or
82: integral over all possible spin configurations.  The weighting factor is
83: obtained by truncating the more standard Boltzmann weight
84: $\exp(K\mathbf{S}_\mathbf{x} \cdot \mathbf{S}_{\mathbf{x}'})$.  This
85: choice mimics the weighting factor of the Ising model, where $\exp(\beta
86: S_\mathbf{x} S_{\mathbf{x}'}) \propto (1 + K S_\mathbf{x}
87: S_{\mathbf{x}'})$ with $K = \tanh \beta$.  When formulated on a
88: honeycomb lattice, which has coordination number $z=3$, the HT graphs of
89: the truncated model are automatically nonintersecting and self-avoiding. 
90: The partition function can then be written simply as a sum over all
91: possible closed graphs \cite{DMNS}, $Z = \sum_{\{G\}} K^b N^l$, with $b$
92: and $l$ the number of occupied bonds and separate loops forming the
93: graph $G$.  The parameter $K$ in the spin formulation (\ref{ZS}) appears
94: as bond fugacity in the loop model.  By mapping it onto a solid-on-solid
95: model, the critical exponents as well as the critical point were
96: determined exactly \cite{Nienhuis}.  
97: 
98: In the high-temperature phase, the HT graphs have a finite line tension
99: $\theta_\mathrm{G}$ and are exponentially suppressed.  A typical graph
100: configuration in this phase shows only a few small loops scattered
101: around the lattice.  Upon approaching the critical point from above, the
102: lattice starts to fill up with more and also larger graphs.  At the
103: critical point, the line tension vanishes, causing the exponential
104: suppression to disappear.  Graphs of all sizes now appear in the system
105: as they can grow without energy cost, i.e., the HT graphs proliferate. 
106: A graph spanning the lattice can be found irrespective of the lattice
107: size--much like the appearance of a spanning cluster at the percolation
108: threshold in percolation phenomena \cite{StauferAharony}.  The average
109: number density $\ell_b$ of graphs containing $b$ bonds takes
110: asymptotically a form similar to that of clusters in percolation theory,
111: \begin{equation}
112:   \label{ellb}
113:   \ell_b \sim b^{- \tau_\mathrm{G}} {\rm e}^{- \theta_\mathrm{G} b}, \quad 
114:   \theta_\mathrm{G} \propto (K-K_{\rm c})^{1/\sigma_\mathrm{G}},
115: \end{equation} 
116: with $\sigma_\mathrm{G}$ and $\tau_\mathrm{G}$ two exponents whose
117: values define the universality class.  The line tension vanishes upon
118: approaching the critical point at a pace determined by the exponent
119: $\sigma_\mathrm{G}$.  When present, this Boltzmann factor exponentially
120: suppresses large graphs.  The algebraic factor in the graph distribution
121: is an entropy factor, giving a measure of the number of ways a graph of
122: size $b$ can be embedded in the lattice.  The configurational entropy is
123: characterized by the exponent $\tau_\mathrm{G}$.  As in percolation
124: theory \cite{StauferAharony}, it is related to the fractal dimension
125: $D_\mathrm{G}$ of the HT graphs via
126: \begin{equation}
127:   \label{tau}
128:   \tau_\mathrm{G} = d/D_\mathrm{G} +1,
129: \end{equation} 
130: with $d=2$ the dimension of the lattice. 
131: 
132: When summed over all sizes, the graph distribution yields the scaling
133: part of the logarithm of the partition function,
134: \begin{equation}
135:   \label{lnZ} 
136:   \ln Z \sim \sum_b \ell_b. 
137: \end{equation} 
138: Each graph therefore contributes equally to the scaling part of the free
139: energy, irrespective of its size. 
140: 
141: 
142: %
143: %
144: \begin{table}
145:   \begin{tabular}{l|rc|r|rcccc|ccc}
146: %
147:     \hline \hline & & & & & & & & \\[-.4cm]
148: %
149:     Model & $N$ & $\bar\kappa_-$ & $c$ & $\alpha$ & $\beta$ & $\gamma$ & $\eta$
150:     & $\nu$ & $D_{\rm G}$ & $\tau_\mathrm{G}$ & $\sigma_\mathrm{G}$ \\[.1cm]
151: %
152:     \hline & & & & & & & & & & \\[-.4cm]
153: % 
154:     Gaussian & $-2$ & $\frac{1}{2}$ & $-2$ & $1$ & $0$ & $1$ & $0$ & $\frac{1}{2}$ &  $\frac{5}{4}$ & 
155:     $\frac{13}{5}$ &  $\frac{8}{5}$\\[.1cm]
156: %
157:     SAW & $0$ & $\frac{2}{3}$ & $0$ & $\frac{1}{2}$ & $\frac{5}{64}$ & 
158:     $\frac{43}{32}$ & $\frac{5}{24}$ & $\frac{3}{4}$  & $\frac{4}{3}$ &  $\frac{5}{2}$ &  $1$ \\[.1cm]
159: %
160:     Ising & $1$ & $\frac{3}{4}$ & $\frac{1}{2}$ 
161:     & $0$ & $\frac{1}{8}$ & $\frac{7}{4}$ &
162:     $\frac{1}{4}$ & $1$ &
163:     $\frac{11}{8}$ &  $\frac{27}{11}$ &  $\frac{8}{11}$ \\[.1cm]
164: % 
165:     XY & $2$ & $1$ & $1$  & $-\infty$ & $\infty$ & $\infty$ & $\frac{1}{4}$ & $\infty$ 
166:     & $\frac{3}{2}$ &  $\frac{7}{3}$ &  $0$\\[.1cm] \hline \hline
167: % 
168: %
169:   \end{tabular}
170:   \caption{Critical exponents of the two-dimensional critical O($N$) spin
171:     model, with $N=-2,0,1,2$, respectively, together with the parameter
172:     $\bar\kappa_-$ and the central charge $c= 1 - 6(1-\bar\kappa_-)^2/\bar\kappa_-$.  
173:     In addition, the fractal dimension
174:     $D_\mathrm{G}$ of the HT graphs  as well as the two exponents
175:     $\tau_\mathrm{G}, \sigma_\mathrm{G}$ characterizing their
176:     distribution are given.  For  $N=1$, the values of the latter two were recently 
177:     established numerically in Ref.~\cite{geoPotts}. 
178:     \label{table:On}}
179: \end{table}
180: %
181: %
182: 
183: 
184: Table \ref{table:On} summarizes the critical exponents and fractal
185: dimensions of the four most common O($N$) models.  The negative value
186: $N=-2$ corresponds to the noninteracting model, for which the critical
187: exponents take their Gaussian values \cite{BalianToulouse}.  In the
188: polymer limit $N\to 0$, first studied by de Gennes \cite{deGennes}, the
189: fractal dimension of the HT graphs is simply the inverse of the
190: correlation length exponent $\nu$.  In general, however, it follows from
191: Table \ref{table:On} that $D_\mathrm{G} \neq 1/\nu$.  To generalize de
192: Gennes' result, we note that in percolation theory a similar relation
193: between the fractal dimension $D$ of clusters and the correlation length
194: exponent involves the Fisher exponent $\sigma$, viz.\
195: \cite{StauferAharony}
196: \begin{equation}
197:   \label{nuD}
198:   D = 1/\sigma \nu. 
199: \end{equation} 
200: A closer look at de Gennes' derivation reveals that
201: $\sigma_\mathrm{G}=1$ in that case, implying that the result for
202: polymers in good solvents or self-avoiding random walks is consistent
203: with Eq.~(\ref{nuD}).  In a recent MC study of the HT graphs of
204: the 2D Ising model \cite{geoPotts}, we numerically found
205: the value $\sigma_\mathrm{G} = 0.732(6)$.  This estimate is within one
206: standard deviation from the value $\sigma_\mathrm{G} = 8/11 = 0.7273
207: \dots$ expected from Eq.~(\ref{nuD}), with $\nu=1$ and the fractal
208: dimension $D_\mathrm{G}=11/8$ appropriate for the Ising model. 
209: 
210: Parameterizing the 2D O($N$) models as \cite{Nienhuis_rev,SLE} $N = - 2
211: \cos(\pi/\bar\kappa_-)$, with $\frac{1}{2} \le \bar\kappa_- \le 1$, we
212: obtain from Eq.~(\ref{nuD}) $\sigma_\mathrm{G} = 8
213: (1-\bar\kappa_-)/(2+\bar\kappa_-)$, where use is made of the known
214: results \cite{Nienhuis} $1/\nu = 4 (1-\bar\kappa_-)$ and
215: \cite{Vanderzande}
216: \begin{equation}
217:   \label{dh}
218:   D_\mathrm{G} = 1+ \bar \kappa_-/2. 
219: \end{equation}   
220: The entropy exponent (\ref{tau}) follows similarly as $\tau_\mathrm{G} =
221: (6+ \bar\kappa_-)/(2+\bar\kappa_-)$, yielding $\tau_\mathrm{G}=5/2$ for
222: a self-avoiding random walk and $\tau_\mathrm{G}= 27/11 = 2.4546 \dots$
223: for the Ising model.  Through the exact enumeration and analysis of the
224: number of self-avoiding loops on a square lattice up to length 110, the
225: expected value $\tau_\mathrm{G}=5/2$ has been established numerically to
226: very high precision \cite{Jensen}.  In our MC study \cite{geoPotts}, we
227: numerically obtained the estimate $\tau_\mathrm{G} = 2.458(5)$ for the
228: Ising model in good agreement with the theoretical prediction.
229: 
230: The O($N$) spin-spin correlation function $G(\mathbf{x},\mathbf{x}') =
231: G(\mathbf{x}-\mathbf{x}')$ is represented diagrammatically by a modified
232: partition function, obtained by requiring that the two sites
233: $\mathbf{x}$ and $\mathbf{x}'$ are connected by an open HT graph
234: $\Gamma$ \cite{Stanley}.  On a honeycomb lattice, the scaling part of
235: the correlation function is given by the connected graphs
236: \begin{equation}
237: \label{GR}
238:   G(\mathbf{x},\mathbf{x}') \sim \sum_{\{\Gamma\}} K^b = \sum_b
239:   z_b(\mathbf{x},\mathbf{x}') K^b,
240: \end{equation} 
241: where $z_b(\mathbf{x},\mathbf{x}')$ is the number of (open)
242: nonintersecting and self-avoiding graphs along $b$ bonds connecting
243: $\mathbf{x}$ and $\mathbf{x}'$.  It is related to the graph distribution
244: (\ref{ellb}) through $ \ell_b = (1/V b) \sum_\mathbf{x}
245: z_b(\mathbf{x},\mathbf{x})$, with $V$ the lattice volume.  Since
246: $z_b(\mathbf{x},\mathbf{x})$ refers to closed graphs starting and ending
247: at $\mathbf{x}$, the factor $1/b$ is included to prevent overcounting as
248: a given loop can be traced out starting at any lattice point along that
249: loop.  Strictly speaking, $G(\mathbf{x},\mathbf{x}') < \sum_{\{\Gamma\}}
250: K^b$ as the cancellation of the disconnected graphs in the numerator and
251: $Z$ in the denominator, required for an equality, is not complete: For a
252: given open graph $\Gamma$, certain loops present in $Z$ are forbidden in
253: the modified partition function as they would intersect $\Gamma$, or
254: occupy bonds belonging to it, which is not allowed.  In other words, the
255: presence of an open graph influences the loop gas and \textit{vice
256: versa}.  Since each loop carries a factor $N$, the loop gas is absent in
257: the limit $N \to 0$.  The inequality in Eq.~(\ref{GR}) then becomes an
258: equality and the open graphs become ordinary self-avoiding random walks
259: with $D_\mathrm{G}=4/3$.  For $N>0$, the loops obstruct the formation of
260: graphs connecting the two endpoints, so that the fractal dimension of
261: these self-avoiding graphs on the honeycomb lattice is larger.
262: 
263: The magnetic susceptibility $\chi$ follows as $\chi = \sum_{\mathbf{x}'}
264: G(\mathbf{x},\mathbf{x}') \sim \sum_b z_b K^b$, with $z_b =
265: \sum_{\mathbf{x}'} z_b(\mathbf{x},\mathbf{x}')$ the number of open
266: graphs of size $b$ starting at $\mathbf{x}$ and ending at an arbitrary
267: lattice point.  For the susceptibility to diverge with the correct
268: exponent $\gamma$, it must behave close to the critical point as
269: \begin{equation}
270:   \label{asymp} 
271:   \chi \sim  \sum_b b^{\sigma_\mathrm{G} \gamma-1}
272:   \mathrm{e}^{-\theta_\mathrm{G} b},
273: \end{equation} 
274: for large $b$, where like in the closed graph distribution (\ref{ellb}), the
275: Boltzmann factor suppresses large graphs as long as the line tension
276: $\theta_\mathrm{G}$ is finite.  Indeed, replacing the summation over the
277: HT graph size $b$ with an integration, we find $\chi \sim
278: |K-K_\mathrm{c}|^{-\gamma}$.  The asymptotic form (\ref{asymp})
279: generalizes the de Gennes result for $N\to 0$ with $\sigma_\mathrm{G}=1$
280: to arbitrary $-2 \leq N \leq 2$ with $\sigma_ \mathrm{G} \neq 1$.  Note
281: that on account of Fisher's scaling relation $\gamma = (2-\eta) \nu$,
282: the combination $\sigma_\mathrm{G} \gamma$ in Eq.~(\ref{asymp})
283: satisfies $\sigma_\mathrm{G} \gamma = (2-\eta)/D_\mathrm{G}$.
284: 
285: The ratio of $z_b(\mathbf{x},\mathbf{x}')$ and $z_b$ defines the
286: probability $P_b(\mathbf{x},\mathbf{x}')$ of finding a graph connecting
287: $\mathbf{x}$ and $\mathbf{x}'$ along $b$ bonds \cite{Cloizeaux}.  On
288: general grounds, it scales at criticality as ($d=2$):
289: \begin{equation}
290:   \label{P}
291:   P_b(\mathbf{x},\mathbf{x}') = z_b(\mathbf{x},\mathbf{x}')/z_b \sim
292:   b^{-d/D_\mathrm{G}} \, \mathsf{P} \left(
293:     |\mathbf{x}-\mathbf{x}'|/b^{1/D_\mathrm{G}} \right),
294: \end{equation} 
295: with $\mathsf{P}$ a scaling function.  Since \textit{at} the critical
296: point, $z_b K^b_\mathrm{c} \sim b^{\sigma_\mathrm{G} \gamma-1}$
297: according to Eq.~(\ref{asymp}), we obtain for the correlation function
298: \begin{equation}
299:   \label{G}
300:   G(\mathbf{x},\mathbf{x}') \sim  \sum_b z_b
301:   P_b(\mathbf{x},\mathbf{x}') K_\mathrm{c}^b \sim 
302:   1/|\mathbf{x}-\mathbf{x}'|^{d-2 + \eta},
303: \end{equation} 
304: where use is made of Eq.~(\ref{nuD}) and Fisher's scaling relation.
305: Equation (\ref{G}) is the standard definition of the critical exponent
306: $\eta$, whose exact value is given by \cite{Nienhuis}
307: \begin{equation}
308:   \label{etakappa}
309:   \eta=2-3/4\bar\kappa_- - \bar\kappa_-,
310: \end{equation}  
311: and thus provides a consistency check.  Also, with $\nu =
312: (\tau_\mathrm{G}-1)/d \sigma_\mathrm{G}$, as is implied by
313: Eqs.~(\ref{nuD}) and (\ref{tau}), Eq.~(\ref{lnZ}) yields the scaling
314: relation $d \nu = 2 - \alpha$, where $\alpha$ determines the scaling
315: behavior of the free energy close to the critical point, $\ln Z \sim
316: |K-K_\mathrm{c}|^{2-\alpha}$.  Apart from $\nu$ and $\alpha$, the
317: exponents have no simple dependence on the graph distribution exponents
318: $\sigma_\mathrm{G}, \tau_\mathrm{G}$.  This is because the operator
319: whose scaling dimension $y_\mathrm{G}$ is given by the fractal dimension
320: $D_\mathrm{G}$ of the HT graphs is not a simple one, consisting of two
321: spin components at the same site $S_\mathbf{x}^i S_\mathbf{x}^j$ which
322: measures the tendency of spins to align \cite{Nienhuis}.
323: 
324: In our MC study \cite{geoPotts}, rather than determining the
325: scaling dimension $y_\mathrm{G}$ directly, we measured the so-called
326: percolation strength $P_\infty$, giving the fraction of bonds in the
327: largest graph.  This observable obeys the finite-size scaling relation
328: $P_\infty \sim L^{-\beta_{\rm G}/\nu}$, where the exponent $\beta_{\rm
329:   G}$ is related to $y_\mathrm{G}$ via $y_\mathrm{G} = d -
330: \beta_\mathrm{G}/\nu$.  We found $\beta_{\rm G} = 0.626(7)$ for the
331: Ising model in perfect agreement with the value $\beta_{\rm
332:   G}=5/8=0.625$, leading to $y_\mathrm{G} = 11/8$, which coincides with
333: the fractal dimension $D_\mathrm{G}$ of the HT graphs. 
334: 
335: We next extend our results for the HT graphs at the \textit{critical}
336: point to the \textit{tricritical} point.  It is generally accepted that
337: including vacancies in the O($N$) model gives in addition to critical
338: behavior rise to also tricritical behavior.  By gradually increasing the
339: activity of the vacancies, the continuous O($N$) phase transition is
340: eventually driven first order.  The endpoint where this happens is a
341: tricritical point.  In the context of polymers ($N \to 0$), the latter
342: obtains by lowering the temperature to the $\Theta$ point where the
343: increasingly important van der Waals attraction between monomers causes
344: the polymer chain to collapse.  Coniglio \textit{et
345:   al.\/}~\cite{Coniglioetal} argued that a polymer ring at the $\Theta$
346: tricritical point is equivalent to the hull of a percolation cluster.
347: With the known dimension for the percolation hull \cite{SDFK}, the
348: analogy then gives $D_\mathrm{G}^\mathrm{t}=\frac{7}{4}$ as fractal
349: dimension for a polymer chain at the $\Theta$ point (the superscript
350: ``t'' refers to the tricritical point).  It also implies that polymer
351: chains at the critical and tricritical point share the same central
352: charge $c$ because both the O($N \to 0$) model and percolation have
353: $c=0$ \cite{DStheta}.
354: 
355: To generalize the analogy found by Coniglio \textit{et
356:   al.\/}~\cite{Coniglioetal}, we consider the $Q$-state Potts model,
357: which in the limit $Q\to 1$ describes ordinary, uncorrelated percolation
358: \cite{Potts}.  In the Fortuin-Kasteleyn (FK) formulation \cite{FK}, the
359: model with $Q>1$ is mapped onto a \textit{correlated} percolation
360: problem.  Clusters are formed by lumping together with a certain
361: temperature-dependent probability nearest neighbor spins in the same
362: spin state.  These so-called FK clusters percolate at the critical
363: temperature and their fractal structure encodes the entire critical
364: behavior.  In 2D (and in 2D only), also the geometrical clusters, formed
365: by unconditionally lumping together nearest neighbor spins in the same
366: spin state, percolate at the critical temperature.  Their fractal
367: structure encodes the \textit{tricritical} $Q_\mathrm{t}$-state Potts
368: behavior, which emerges when enlarging the pure model to include
369: vacancies.  The tricritical point shares the same central charge $c$
370: with the critical point, but apart from $c=1$, $Q(c) \neq
371: Q_\mathrm{t}(c)$.  Both fractal structures and thus both critical
372: behaviors are intimately related, being connected by replacing
373: $\bar\kappa_-$ with $\bar\kappa_+ = 1/\bar\kappa_- \geq 1$ in the
374: appropriate expressions (for details, see Ref.~[\onlinecite{geoPotts}]
375: or Ref.~[\onlinecite{DBN}], where similar conclusions were reached
376: independently). This map conserves the central charge. With increasing
377: $Q$ or $c$, the critical and tricritical points approach each other
378: until merging at $c=1$, where $Q=Q_\mathrm{t}=4$ and the scaling
379: behaviors of FK and geometrical clusters coincide.  Beyond $Q=4$, the
380: transition becomes discontinuous. 
381: 
382: The hulls of the geometrical clusters of the $Q(c)$-state Potts model at
383: the same time represent the HT graphs of the critical O($N$) model, with
384: $0\leq N(c) \leq 2$, sharing the same central charge
385: \cite{DS89,geoPotts}.  Since the geometrical clusters encode the
386: tricritical Potts behavior, while the FK clusters encode the critical
387: Potts behavior (characterized by the same central charge), it is natural
388: to expect the FK hulls to represent the HT loop gas not at the critical,
389: but at the tricritical point.  For the special case of polymers ($N \to
390: 0$), this reproduces the result by Coniglio \textit{et
391: al.\/}~\cite{Coniglioetal}.  Note that in both Potts and O($N$) models,
392: including vacancies leads to tricritical behavior.  However, the
393: critical and tricritical points get interchanged when passing from one
394: model to the other.  The fractal dimension $D_\mathrm{G}^\mathrm{t}$ of
395: the tricritical loops follows by applying the central-charge conserving
396: map $\bar\kappa_- \to \bar\kappa_+ = 1/\bar\kappa_-$ to Eq.~(\ref{dh}).
397: We submit that the exponent $\eta_\mathrm{t}$ follows in the same way
398: from Eq.~(\ref{etakappa}). Given $\eta_\mathrm{t}$, the scaling
399: relations then yield values for the ratios
400: $\beta_\mathrm{t}/\nu_\mathrm{t}$ and
401: $\gamma_\mathrm{t}/\nu_\mathrm{t}$.
402: 
403: To verify these predictions we rewrite our results, given as a function
404: of $\bar \kappa_-$, as entries in the Kac table
405: \begin{equation}
406:   h_{p,q} = \frac{[(m+1)p -m q]^2 -1}{4m(m+1)}. 
407: \end{equation} 
408: Here, the parameter $m$ is related to $\bar\kappa_-$ and the central
409: charge via $\bar\kappa_- = m/(1+m)$ and $c = 1 - 6/m(m+1)$,
410: respectively, while the central-charge conserving map becomes $m \to
411: -m-1$.  Specifically, for $0 \leq c \leq 1$
412: \begin{equation}
413:   \label{Dt}
414:   D_\mathrm{G}^\mathrm{t} =  1+ 1/2\bar \kappa_- = 2 -2 h_{m,m},
415: \end{equation} 
416: leading to the correct result \cite{DStheta}
417: $D_\mathrm{G}^\mathrm{t}=2-2h_{2,2}=\frac{7}{4}$ for a polymer at the
418: $\Theta$ point ($m=2$), and
419: \begin{equation}
420: \label{etat}
421:   \eta_\mathrm{t} = 2-3\bar\kappa_-/4 - 1/\bar\kappa_- = 4 h_{m/2,m/2}. 
422: \end{equation} 
423: The right hand of this equation is in agreement with the results
424: $\eta_\mathrm{t}=4 h_{1,1}=0$ \cite{DStheta} for the $\Theta$ point
425: ($m=2$) and $\eta_\mathrm{t}=4 h_{2,2} = \frac{3}{20}$
426: \cite{Zamolodchikov} for the tricritical Ising model ($m=4$).  After
427: circulating a draft of this paper, we have been informed about a recent
428: high-precision MC study of the tricritical O($N$) model in
429: Ref.~\cite{bloeteetal} previously unavailable to us, in which the
430: authors extend earlier numerical work on the tricritical
431: O($\frac{1}{2}$) model \cite{GBN}.  In Fig.~\ref{fig:xh_c}, we compare
432: our theoretical prediction (\ref{etat}) for $\eta_\mathrm{t}$ with the
433: magnetic scaling dimension
434: $x^\mathrm{t}_h=2-y^\mathrm{t}_h=\eta_\mathrm{t}/2$ obtained numerically
435: in that study.  For $0 \leq c \leq \frac{7}{10}$, the MC data
436: are within one standard deviation of our prediction.  Beyond the
437: tricritical Ising model ($c=\frac{7}{10}$) the numerical data start
438: deviating from our analytic result.  A detailed future investigation is
439: required to clarify this discrepancy.
440: 
441: 
442: \begin{figure}
443: \centering
444: \includegraphics[width=0.45\textwidth]{xh_c.eps}
445: \caption{Our analytic prediction for the tricritical exponent
446:   $\eta_\mathrm{t}$ (solid line) as a function of the central charge $c$
447:   compared with high-precision MC data (marks with error bars)
448:   \cite{bloeteetal}. Our prediction extends only to $c=1$, indicated by
449:   the vertical line, while the numerical results continue to the
450:   tricritical O(2) model with $c=1.149(1)$ \cite{bloeteetal}.
451:   \label{fig:xh_c}}
452: \end{figure}
453: %
454: The tricritical HT graphs, representing simultaneously the hulls of FK
455: clusters, have a distribution again of the form (\ref{ellb}),
456: characterized by two exponents
457: $\sigma_\mathrm{G}^\mathrm{t}, \tau_\mathrm{G}^\mathrm{t}$.  Given our
458: result (\ref{Dt}) for the fractal dimension of these graphs,
459: $\tau_\mathrm{G}^\mathrm{t}$ is determined exactly through
460: Eq.~(\ref{tau}) with $\tau_\mathrm{G}$ and $D_\mathrm{G}$ replaced by
461: their tricritical counterparts $\tau^\mathrm{t}_\mathrm{G}$ and
462: $D^\mathrm{t}_\mathrm{G}$.
463: 
464: In conclusion, we have shown that for $-2 \leq N \leq 2$ the fractal
465: structure of 2D HT graphs of the O($N$) spin model encodes the O($N$)
466: critical behavior.  We thereby extended de Gennes' result for
467: self-avoiding loops in the limit $N \to 0$ to random loops for arbitrary
468: $-2 \leq N \leq 2$.  We studied the loops also close to the point where
469: they collapse, corresponding to the HT representation of the tricritical
470: O($N$) model.  The fractal structure of the tricritical loops was argued
471: to be in one-to-one correspondence with that of the critical loops,
472: allowing us to also predict the magnetic scaling dimension
473: $x^\mathrm{t}_h =\eta_\mathrm{t}/2 = 2 h_{m/2,m/2}$ at the O($N)$
474: tricritical point, in very good agreement with recent MC data in the
475: range $0 \leq c \lesssim 0.7$.
476: 
477: % \acknowledgments
478: %
479: The authors thank H. Bl\"ote, C. von Ferber, Y. Holovatch, and B. 
480: Nienhuis for useful discussions and communications.  This work is
481: partially supported by the DFG grant No. JA 483/17-3 and by the
482: German-Israel Foundation (GIF) under grant No.\ I-653-181.14/1999. 
483: 
484: \vspace{-.4cm}
485: 
486: \begin{thebibliography}{99}
487: 
488: \vspace{-.4cm}
489: 
490: \bibitem{Stanley} H. E. Stanley, \textit{Introduction to Phase
491:     Transitions and Critical Phenomena} (Oxford University Press, New
492:   York, 1971). 
493: \bibitem{deGennes} P. G. de Gennes, Phys. Lett. A \textbf{38}, 339
494:   (1972). 
495: \bibitem{geoPotts} W. Janke and A. M. J. Schakel, Nucl. Phys. B {\bf
496:     700} [FS], 385 (2004); Comp. Phys. Comm. {\bf 169}, 222 (2005).
497: \bibitem{Dotsenko} For an alternative numerical estimate of $\tau$,
498:   using Eq.~(2) \textit{at} criticality, see V. S. Dotsenko \textit{et
499:     al.}, Nucl. Phys. B {\bf 448} [FS], 577 (1995).
500:  \bibitem{bloeteetal} W. Guo, H. W. J. Bl\"ote, and Y.-Y. Liu, Commun. 
501:   Theor. Phys. (Beijing) {\bf 41}, 911 (2004). 
502: \bibitem{DMNS} E. Domany \textit{et al.}, Nucl.  Phys. B \textbf{190},
503:   279 (1981).
504: \bibitem{Nienhuis} B. Nienhuis, Phys. Rev. Lett. {\bf 49}, 1062 (1982);
505:   J. Stat. Phys. \textbf{34}, 731 (1984).
506: \bibitem{StauferAharony} D. Stauffer and A. Aharony, {\it Introduction
507:     to Percolation Theory}, 2nd edition (Taylor \& Francis, London,
508:   1994). 
509: \bibitem{BalianToulouse} R. Balian and G. Toulouse, Phys. Rev. Lett. 
510:   {\bf 30}, 544 (1973). 
511: \bibitem{Nienhuis_rev} B. Nienhuis, in: {\it Phase Transitions and
512:     Critical Phenomena}, edited by C. Domb and J. L. Lebowitz (Academic,
513:   London, 1987), Vol. 11, p. 1.
514: \bibitem{SLE} For consistency with Ref.~\cite{geoPotts}, we adopt here
515:   the notation (apart from normalization) of the stochastic Loewner
516:   evolution (SLE$_\kappa$) formalism which puts various known results
517:   cited below on a mathematical rigorous footing. Specifically,
518:   $\bar{\kappa}_- := \kappa/4$. For an overview on SLE$_\kappa$, see O.
519:   Schramm, Israel J. Math. {\bf 118}, 221 (2000); S. Rohde and O.
520:   Schramm, Ann. Math. {\bf 161}, 879
521:   (2005); B. Duplantier, J. Stat. Phys.  {\bf 110}, 691 (2003).
522: \bibitem{Vanderzande} C. Vanderzande, J. Phys. A {\bf 25}, L75 (1992). 
523: \bibitem{Jensen} I. Jensen, J. Phys. A {\bf 36}, 5731 (2003). 
524: \bibitem{Cloizeaux} J. des Cloizeaux, Phys. Rev. A {\bf 10}, 1665
525: (1974). 
526: \bibitem{Coniglioetal} A. Coniglio \textit{et al.}, Phys. Rev. B {\bf
527:     35}, 3617 (1987).
528: \bibitem{SDFK} B. Duplantier and H. Saleur, Phys. Rev. Lett. {\bf 58},
529:   2325 (1987). 
530: \bibitem{DStheta} B. Duplantier and H. Saleur, Phys. Rev. Lett. {\bf
531:     59}, 539 (1987); \textit{ibid.} {\bf 62}, 1368 (1989). 
532: \bibitem{Potts} R. B. Potts, Proc. Camb. Phil. Soc. {\bf 48}, 106
533:   (1952). 
534: \bibitem{FK} C. M. Fortuin and P. W.  Kasteleyn, Physica {\bf 57}, 536
535:   (1972). 
536: \bibitem{DBN} Y. Deng, H. W. J. Bl\"ote, and B. Nienhuis, Phys. Rev. E
537:   {\bf 69}, 026123 (2004). 
538: \bibitem{DS89} B. Duplantier and H. Saleur, Phys. Rev. Lett. {\bf 63},
539:   2536 (1989). 
540: \bibitem{Zamolodchikov} A. B. Zamolodchikov, Sov. J. Nucl. Phys. {\bf
541:     44}, 529 (1986). 
542: \bibitem{GBN} W. Guo, H. W. J. Bl\"ote, and B. Nienhuis, Int. J. Mod. 
543:   Phys. C {\bf 10}, 291 (1999). 
544: \end{thebibliography}
545: 
546: \end{document}
547: 
548: