1: \documentclass[12pt]{iopart}
2: %\documentclass{iopart}
3: \usepackage{amssymb}
4: \usepackage{epsf}
5: %\usepackage{subeqnar}
6:
7: %\def\received#1{\vspace*{5pt}\address{Received #1}}
8:
9: \begin{document}
10:
11: \title[Two-dimensional Ising model in a random surface
12: field]{Logarithmic corrections in the
13: two-dimensional Ising model in a random surface field}
14:
15: \author[M Pleimling \etal]{M Pleimling$^1$, F A Bagam\'ery$^{2,3}$,
16: L Turban$^3$\ and F Igl\'oi$^{2,4}$}
17:
18: \address{$^1$\ Institut f\"ur Theoretische Physik I, Universit\"at
19: Erlangen-N\"urnberg, D-91058 Erlangen, Germany}
20:
21: \address{$^2$\ Institute of Theoretical Physics,
22: Szeged University, H-6720 Szeged, Hungary}
23:
24: \address{$^3$\ Laboratoire de Physique des Mat\'eriaux, Universit\'e Henri
25: Poincar\'e (Nancy 1), BP~239, F-54506 Vand\oe uvre l\`es Nancy Cedex, France}
26:
27: \address{$^4$\ Research Institute for Solid State Physics and Optics,
28: H-1525 Budapest, P.O.Box 49, Hungary}
29:
30: \ead{turban@lpm.u-nancy.fr}
31:
32: %\received{}
33:
34: \begin{abstract}
35: In the two-dimensional Ising model weak random surface field is predicted to be
36: a
37: marginally irrelevant perturbation at the critical point. We study this question
38: by extensive Monte Carlo simulations for various strength of disorder. The
39: calculated effective (temperature or size dependent) critical exponents fit with
40: the field-theoretical results and can be interpreted in terms of the predicted
41: logarithmic corrections to the pure system's critical behaviour.
42: \end{abstract}
43:
44: \pacs{05.70Jk, 05.50.+q}
45:
46:
47: \section{Introduction}
48:
49: In an inhomogeneous system the local critical behaviour near
50: localized or extended defects may differ considerably from the
51: bulk critical behaviour in the regular lattice (for a review,
52: see~\cite{igloi93}). One possible source of inhomogeneity is quenched
53: (i.e., time-independent) randomness, which can be localized at
54: the surface of the system (fluctuating surface coupling
55: constants~[2--6],
56: microscopic terraces at the surface~\cite{selke02})
57: or at a grain boundary in the bulk of the system. It is known
58: experimentally~[8--10] that impurities may diffuse
59: from inside the sample and segregate on the surface or at
60: grain boundaries. In adsorbed systems, quenched disorder is naturally
61: present since adatoms may bind randomly on equivalent surface
62: sites~\cite{cardy91}.
63:
64: In a theoretical description of the local critical behaviour of
65: these systems, close to the bulk critical point, one can use a
66: coarse-grained picture in which quenched randomness couples to some
67: local operator. The local operator considered in this paper
68: is the surface order parameter, hence the perturbation is described by
69: the introduction of random fields (RFs) localized at the surface. Usually
70: the RF has zero mean and its variance is used to characterize
71: the strength of disorder.
72:
73: In the weak-disorder limit, the relevance or irrelevance of the
74: perturbation can be analyzed by making use of a Harris-type
75: criterion~\cite{harris74}. The condition for the irrelevance
76: of RFs on a defect with dimension $d-1$ can be expressed
77: in terms of the decay exponent for the local order parameter
78: correlations in the pure system~\cite{diehl90a} as
79: %
80: \begin{equation}
81: \eta_{\parallel} \ge 1\,.
82: \label{irrelevance}
83: \end{equation}
84: %
85:
86: A plane of RFs in the bulk often constitutes a relevant perturbation as
87: it is the case for the two-dimensional (2D) Ising model with $\eta=1/4$.
88: Thus a new fixed point appears, which controls the local critical
89: behaviour. This fixed point is expected to be a surface one since RFs
90: tend to destroy the local order and the bulk defect then acts as an
91: effective cut.
92:
93: Surface RFs are irrelevant for the 3D Ising model as noted and
94: demonstrated through Monte Carlo (MC) simulations~\cite{mon88}. Curiously, in
95: the
96: case of a system with continuous symmetry, like the 3D Heisenberg model, surface
97: RFs destroy the bulk long-range order~\cite{feldman02} according to Imry-Ma
98: arguments~\cite{imry75}, although the perturbation is irrelevant at the ordinary
99: surface transition according to~(\ref{irrelevance}). Among 2D systems the Ising
100: model represents the borderline case, since $\eta_{\parallel}=1$~\cite{McCoyWu}.
101: For this model, field-theoretical investigations~\cite{cardy91,igloi91} predict
102: that weak surface RF is a marginally irrelevant perturbation. Consequently, the
103: surface critical properties of the random model are characterized by the
104: critical
105: singularities of the pure model supplemented by logarithmic corrections to
106: scaling.
107:
108: These theoretical predictions have not yet been confronted with the results of
109: numerical calculations. In general, the observation and characterization of
110: logarithmic corrections to scaling by numerical methods are notoriously
111: difficult
112: tasks, particularly in systems with quenched disorder. In this respect,
113: a well-known example is the diluted 2D ferromagnetic Ising model, for which the
114: accurate form of the singularities was long debated~[18--23].
115:
116: In this paper we present the results of a numerical study of the surface
117: critical
118: behaviour of the 2D Ising model in the presence of random surface fields. In
119: section~2 we present the model and the known results about its critical
120: properties. Then, through intensive Monte Carlo simulations, we determine
121: effective surface magnetization exponents in two different ways. In section~3,
122: they are obtained as a function of the deviation from the critical temperature.
123: In section~4, we use a small homogeneous surface field at the critical point to
124: deduce size-dependent exponents from the magnetization profiles. In section~5 we
125: discuss the agreement between theoretical and numerical results.
126:
127: \section{The model and its predicted surface critical properties}
128:
129: We consider the Ising model on a $L\times M$ square lattice with the Hamiltonian
130: %
131: \begin{eqnarray}
132: &&{\cal H}=-J\sum_{i=1}^{L-1}\sum_{j=1}^M
133: (s_{i,j}s_{i+1,j}+s_{i,j}s_{i,j+1})
134: -\sum_{j=1}^M(h_1(j)s_{1,j}+h_L(j)s_{L,j})
135: \nonumber\\
136: &&s_{i,M+1}=s_{i,1}\,,\qquad
137: h_i(j)=\left\{
138: \begin{array}{ll}
139: h_{\rm s}+h & \mbox{with probability}\quad p=1/2\\
140: h_{\rm s}-h & \mbox{with probability}\quad p=1/2
141: \end{array}
142: \right.
143: \label{hamilt}
144: \end{eqnarray}
145: %
146: where $s_{i,j}=\pm 1$. $J$ is the first-neighbour exchange interaction. The RF
147: $h_{\rm s}\pm h$ acts on the surface spins in the columns at $i=1$ and $i=L$ and
148: periodic boundary conditions are used in the vertical direction. Our main
149: interest is to calculate the averaged magnetization per column, $m_i=\langle
150: |\sum_j s_{i,j}| \rangle / M$.
151:
152: For the pure system, i.e. with vanishing surface field, the surface
153: magnetization, $m_{\rm s}=m_1=m_L$ is exactly known in the thermodynamic limit
154: ($L,M \to \infty$)\cite{McCoyWu}:
155: %
156: \begin{equation}
157: m_{\rm s, pure} = \left[ \coth(2K)\, \frac{\sinh(2K) - 1}{\cosh(2K) -
158: 1}\right]^{1/2}\;,
159: \label{ms_pure}
160: \end{equation}
161: %
162: in terms of $K=J/k_{\rm B}T$, where $T$ is the temperature. At the critical
163: point
164: with $\sinh(2K_{\rm c})=1$ the surface magnetization vanishes as
165: %
166: \begin{equation}
167: m_{\rm s, pure} \approx m_0 t^{1/2}\;,
168: \label{ms_pure1}
169: \end{equation}
170: %
171: in terms of the reduced temperature, $t=(T_{\rm c}-T)/T_{\rm c}$, and
172: $m_0^2=4(\sqrt{2}+1) K_{\rm c} \ln(2K_{\rm c})$.
173: The relevant length scale is the bulk correlation length which, for $t>0$, is
174: given by~\cite{baxter}
175: %
176: \begin{equation}
177: \xi=\left[2 \ln(\sinh(2K))\right]^{-1}\;,
178: \label{xi}
179: \end{equation}
180: %
181: with the lattice constant for unit length. The correlation length diverges at
182: the
183: critical point as $\xi\approx [2\sqrt{2} K_{\rm c} t]^{-1}$. Thus the reduced
184: temperature and the length scale are related by
185: %
186: \begin{equation}
187: -\ln t \simeq .913 + \ln \xi\;.
188: \label{t_L}
189: \end{equation}
190: %
191: In the presence of random surface fields there are no exact results available.
192: In
193: this case one can use the replica trick to transform the semi-infinite system
194: with a random surface field into $n$ semi-infinite replicas, coupled two-by-two
195: through their surface spins, via nearest-neighbour interactions proportional to
196: $h^2$. The average properties of the random system are obtained in the limit $n
197: \to 0$.
198:
199: The surface critical properties of the system have been studied via two
200: different
201: methods, both using the differential renormalization group (RG) techniques
202: where the lengths are rescaled by a factor $\e^{l}$. The surface coupling
203: between
204: the replicas transforms as
205: %
206: \begin{equation}
207: h^2(l)=\frac{h^2}{1+\kappa h^2 l}\;.
208: \label{rg1}
209: \end{equation}
210: %
211:
212: In the first approach~\cite{igloi91} a conformal mapping is used at the bulk
213: critical point, with $h_{\rm s}=0$, to transform the $n$ semi-infinite replicas
214: into $n$ infinite strips with width $L$, which are coupled to each other at both
215: surfaces through $h^2$. The behaviour of the inverse correlation length is
216: studied using degenerate perturbation theory to second order in $h^2$. From the
217: transformation of the inverse correlation length on the strips under rescaling
218: by
219: $\e^{l}=L$, with $n=0$, and using the gap-exponent relation~\cite{cardy84}, one
220: can identify the $L$-dependent, effective decay exponent $\eta_{\parallel}$
221: which
222: is given by
223: %
224: \begin{equation}
225: \eta_{\parallel} =1+\frac{1}{\ln L}\;,
226: \label{eta}
227: \end{equation}
228: %
229: to leading logarithmic order.
230:
231: In the second approach~\cite{cardy91} the behaviour under rescaling of the
232: homogeneous part of the surface field is determined as
233: %
234: \begin{equation}
235: h_{\rm s}(l)=\frac{h_{\rm s} \e^{l/2}}{(1+\kappa h^2 l)^{1/2}}\;.
236: \label{rg2}
237: \end{equation}
238: %
239: The surface free energy density transforms as
240: %
241: \begin{equation}
242: f_{\rm s}(t,h_{\rm s},h^2)=\e^{-l}f_{\rm s}[\e^{l/\nu}t,h_{\rm s}(l),h^2(l)]\;,
243: \label{fs}
244: \end{equation}
245: %
246: where $\nu$ is the correlation length exponent. Using~(\ref{rg2}), the surface
247: magnetization reads
248: %
249: \begin{equation}
250: m_{\rm s}(t,h^2)=\left.\frac{\partial f_{\rm s}}{\partial h_{\rm
251: s}}\right|_{h_{\rm s}=0}
252: =\frac{\e^{-l/2}}{(1+\kappa h^2 l)^{1/2}}\, m_{\rm s}[\e^{l/\nu}t,h^2(l)]\;.
253: \label{msl}
254: \end{equation}
255: %
256: With $\e^{l}=\xi\sim t^{-\nu}$ and $\nu=1$ according to~(\ref{xi}),
257: ignoring higher order corrections, one obtains
258: %
259: \begin{equation}
260: m_{\rm s}(t,h^2) \sim \frac{t^{1/2}}{(1-\kappa_1 h^2 \ln t)^{1/2}}\;,\quad 0<t \ll
261: 1\;.
262: \label{ms}
263: \end{equation}
264: %
265: Thus the critical singularity of the pure model is supplemented by a logarithmic
266: corrections to scaling. From a practical point of view, one can define
267: temperature-dependent effective exponents through
268: %
269: \begin{equation}
270: \beta_{\rm s}(t)=\frac{\ln[m_{\rm s}(t(1+\delta))/m_{\rm
271: s}(t(1-\delta))]}{\ln[(1+\delta)/(1-\delta)]}=\frac{1}{2}\left(1+\frac{1}{|\ln
272: t|}+\cdots\right)\,,
273: \label{betas}
274: \end{equation}
275: %
276: with $\delta \to 0$. The last expression gives the leading logarithmic
277: correction
278: following from equation~(\ref{ms}).
279:
280: Taking into account the scaling relation $\eta_{\parallel} = 2 \beta_{\rm
281: s}/\nu$ with $\nu=1$, the effective exponents in
282: equations~(\ref{betas}) and~(\ref{eta}) correspond in terms of the relevant
283: length scales, $L \rightarrow \xi \sim 1/t$.
284:
285: The following two sections are devoted to a numerical test of the validity of
286: these theoretical results.
287:
288:
289: \section{Effective exponents}
290:
291: The surface critical exponents are deduced from the temperature dependence of
292: the magnetization $m_i$ in the surface layers (for a review,
293: see~\cite{pleimling04}). We set the
294: homogeneous surface field to zero, $h_{\rm s}=0$, the strength of the random
295: surface field ranging from $h=0.6$ to $h=1.5$, and take a finite reduced
296: temperature, $t>0$. Systems of square and rectangular shapes
297: containing $L \times M$ spins, with $L$ and $M$ ranging from 50 to 1000, have been
298: studied using the standard single-spin-flip method. Although systems with
299: rectangular shapes ($M<L$) lead to reduced finite-size effects, the fraction of
300: surface spins is smaller than for a square system and more runs are needed to
301: achieve the same accuracy for the surface magnetization which is self-averaging.
302: Thus we worked with square systems to spare computer time.
303: The final data are obtained after averaging over at least 1000 different runs with
304: different realizations of the random surface field. For every run, time average has
305: been taken over a few $10^4$ Monte Carlo steps per spin after equilibration.
306:
307: As an illustration we present $m_i$ at $t=0.05$ and $t=0.02$ in
308: figure~\ref{fig1} for different strengths of the RF. The profile of the pure
309: system is shown for comparison. For a given $t$ and different values of $h$,
310: $m_i$ displays a plateau around $i=L/2$ for large enough systems. It corresponds
311: to the bulk magnetization, $m_{\rm b}$, since its height is independent of $h$.
312: If we approach the surfaces close enough, $i,L-i < \xi$, we enter in the surface
313: region where the value of the magnetization is rapidly decreasing to its
314: surface
315: value, $m_{\rm s}$. As seen in figure~\ref{fig1}, for a given $t$ the surface
316: magnetization $m_{\rm s}$ and the inverse size of the surface region are
317: decreasing with increasing disorder strength $h$.
318:
319: %
320: \begin{figure}[tbh]
321: \vglue3mm
322: \epsfxsize=9cm
323: \begin{center}
324: \mbox{\epsfbox{fig1-rsf.eps}}
325: \end{center}
326: \vglue-3mm
327: \caption{Magnetization profiles with random surface fields $h=0$,
328: $0.5$ and $1$ at $t=0.05$ and $t=0.02$. The data have been obtained
329: for a system with $300 \times 300$ spins.}
330: \label{fig1}
331: \end{figure}
332: %
333:
334: According to finite-size scaling, in a large but finite system, sufficiently
335: close to its critical point, $m_{\rm s}(t)$ behaves as
336: $(-t)^{\beta_{\rm s}}f(\xi/L)$, where the scaling function, $f(x)$, tends to a
337: constant for small values of its argument. In the actual calculations, we
338: approach the transition point only to such a distance that the finite-size
339: effects remain negligible and effective surface exponents are calculated
340: using~(\ref{betas}). In practice, finite-size effects have been circumvented by
341: adjusting the size of the sample in a standard approach~\cite{pleimling98}.
342: For a given value of $t$, data obtained for different system sizes are compared.
343: Away from the critical point these data agree as long as the correlation
344: length is less than the extent of the smaller system. Closer to $T_c$ the
345: correlation length increases and at some stage it gets comparable to the
346: size of the smaller system. Finite-size effects then show up by a characteristic
347: fast drop of the effective exponent~\cite{pleimling98}. The smaller system is
348: then discarded and the procedure is continued with two system sizes which still
349: yield identical data at that temperature. This approach is somehow cumbersome
350: but assures that the final data are essentially free of finite-size effects.
351:
352: The effective exponents
353: are shown in figure~\ref{fig2} for different values of the strength
354: of the RF. Here, in order to check the form of the logarithmic corrections in
355: equations~(\ref{eta}) and~(\ref{ms}), $\beta_{\rm s}$ is plotted as a
356: function of $1/|\ln t|$. For a given $t$ the effective exponents are
357: increasing with $h$ as expected since RFs decrease local order. On the
358: other hand, for a given $h$, $\beta_{\rm s}(t)$ first shows a monotonic
359: increase when $t$ decreases and its value passes over the pure system's
360: surface exponent, $\beta_{\rm s}=1/2$. Then, by further decreasing the
361: reduced temperature, $\beta_{\rm s}(t)$ seems to approach a maximum value.
362: This saturation effect is more evident for large values of $h$. Unfortunately,
363: the size limitation did not allow us to approach the transition point close
364: enough to follow the predicted decrease of the effective exponent.
365:
366: Here, in order to compare the numerical results with the theoretical predictions
367: and to extrapolate our data to $t \to 0$, we use the following expression,
368: %
369: \begin{equation}
370: m_{\rm s}(t)=m_0 t^{1/2}\frac{1+a\, t}{(1+b \ln t )^{1/2}}\;,
371: \label{pheno}
372: \end{equation}
373: %
374: which contains the leading analytic correction which follows from
375: equation~(\ref{ms_pure}) and the leading logarithmic corrections to the
376: fixed-point singularity given by~(\ref{ms}). Note that the two corrections have
377: different signs and their competing effect results in the non-monotonic
378: temperature dependence of the effective exponent, $\beta_{\rm s}(t)$. For a
379: given
380: RF, we have fitted the surface magnetization data to the form given
381: in~(\ref{pheno}) with the amplitudes $a$ and $b$ as free parameters. From this,
382: the effective surface magnetization exponent was calculated and used to
383: extrapolate the data points in figure~\ref{fig2}. The data extrapolate to the
384: value of the pure system $\beta_{\rm s, pure}=1/2$ after an ``overshooting
385: effect''. A quite similar tendency was observed in~\cite{selke90} for
386: the bulk magnetization exponent of the 2D random bond Ising model. In
387: our calculations, however, the maximum value is almost reached for $h
388: \ge 1$, which was not possible for the random bond model.
389:
390: In order to check the leading logarithmic correction to the effective
391: exponent, we have calculated the difference of $\beta_{\rm s}(t)$ and its
392: value in the pure system, $\beta_{\rm s, pure}(t)$, as calculated
393: from~(\ref{ms_pure}) via equation~(\ref{betas}). This difference, which is
394: plotted in the inset of figure~\ref{fig2}, no longer contains the leading
395: analytic correction; therefore we expect to be able to compare it with
396: the theoretical prediction in equation~(\ref{betas}). As seen in the inset for
397: the random surface field, $h=1$, the corrections are compatible with theory,
398: although much larger systems are needed in order to reach the asymptotic regime.
399:
400: %
401: \begin{figure}[tbh]
402: \vglue3mm
403: \epsfxsize=9cm
404: \begin{center}
405: \mbox{\epsfbox{fig2-rsf.eps}}
406: \end{center}
407: \vglue-3mm
408: \caption{Effective surface exponent for different strengths of the
409: random surface field. The two first points (grey circles) for $h=1$ were deduced
410: from the short-distance
411: behaviour of the critical profiles discussed in section~4. The broken lines
412: fitting the numerical data
413: correspond to the formula in equation~(\protect\ref{pheno}),
414: with $a=-0.60$, $b=-0.23$ ($h=0.6$), $a=-0.53$, $b=-0.40$ ($h=0.8$),
415: $a=-0.40$, $b=-0.47$ ($h=1.0$) and $a=-0.47$, $b=-0.72$ ($h=1.5$). The
416: inset gives the difference between the effective exponents for the
417: random ($h=1$) and the pure systems. The straight line gives the
418: predicted leading logarithmic correction.}
419: \label{fig2}
420: \end{figure}
421: %
422:
423: %
424: \begin{figure}[hbt]
425: \vglue3mm
426: \epsfxsize=9cm
427: \begin{center}
428: \mbox{\epsfbox{fig3-rsf.eps}}
429: \end{center}
430: \vglue-3mm
431: \caption{Square of the ratio $r(t,h)$ of the surface magnetizations in the pure
432: and in the disordered Ising model as a function of $\ln t $. The slopes are
433: proportional to $h^2$ as shown in the inset. Error bars are much smaller than
434: the symbol sizes.}
435: \label{fig3}
436: \end{figure}
437:
438:
439: The form of the logarithmic corrections has been analyzed in still another way
440: by
441: forming the ratio, $r(t,h)=m_{\rm s, pure}/m_{\rm s}$, of the surface
442: magnetizations in the pure and in the disordered systems. In this way the
443: leading
444: analytic correction to scaling is eliminated. As shown in figure~\ref{fig3}, the
445: square of $r(t,h)$ has an asymptotic linear dependence on $\ln t$, the slope of
446: which is proportional to $h^2$, as shown in the inset of figure~\ref{fig3}. This
447: is in complete agreement with the theoretical prediction in equation~(\ref{ms}).
448:
449: \section{Critical profiles}
450:
451: In this section we study the system at the critical point, $t=0$; however, in the
452: presence of a small homogeneous surface field, $h_{\rm s} \ll h$. A typical
453: magnetization profile in the system with $M \ll L$ is shown in the inset of
454: figure~\ref{fig4}. These data have been obtained with the Swendsen-Wang
455: algorithm
456: with a layer of ghost spins next to the surface~\cite{cr}. At least 1000 runs
457: with different realizations of the random surface field were performed, the time
458: average of every run resulting from typically $3 \cdot 10^5$ MC updates.
459:
460: As known
461: from an analysis of the non-random system~\cite{cr} the profile first increases
462: close to the surface and then decreases in the bulk. The surface critical
463: exponent, $\beta_{\rm s}$, influences the form of the initial part and can be
464: extracted from it. The non-vanishing surface field $h_{\rm s}$ introduces a new
465: surface length scale, $l_{\rm s}$, which in 2D is given by
466: $l_{\rm s}\sim h_{\rm s}^{-1/(1-\eta_{\parallel}/2)}$ and scales as
467: $\sim h_{\rm s}^{-2}$ for the Ising model.
468:
469: %
470: \begin{figure}[tbh]
471: \vglue3mm
472: \epsfxsize=9cm
473: \begin{center}
474: \mbox{\epsfbox{fig4-rsf.eps}}
475: \end{center}
476: \vglue-3mm
477: \caption{Ratio of the initial part of the critical magnetization profiles
478: calculated in the random and in the pure system in the presence of a small
479: homogeneous magnetic field. The initial slope in the log-log plot corresponds to
480: the difference in the effective magnetization exponents (see the text). Inset: one
481: half of the magnetization profile in the random system.}
482: \label{fig4}
483: \end{figure}
484: %
485:
486: The initial part of the profile is restricted to $i \ll M, l_{\rm s}$ and,
487: according to finite-size scaling theory it behaves as
488: %
489: \begin{equation}
490: m_i(M,l_{\rm s})=i^{-x_{\rm m}} g(i/M,i/l_{\rm s})\;,
491: \label{m_k}
492: \end{equation}
493: %
494: where $x_{\rm m}=\beta/\nu=1/8$ is the bulk magnetization scaling dimension.
495: For small values of its arguments the scaling function $g$ is expected to
496: factorize as $g_0(i/M) g_{\rm s}(i/l_{\rm s})$. For the pure 2D Ising model the
497: second term is logarithmic~\cite{bariev}, $g_{\rm s}(i/l_{\rm s}) \sim
498: \ln(i/l_{\rm s}) \sim \ln(i h_{\rm s}^2)$, which makes the numerical analysis
499: difficult. For the Ising model with RF on the surface, the first term is
500: expected to behave as
501: %
502: \begin{equation}
503: g_0(y) \sim \frac{y^{1/2}}{(1+\kappa h^2 \ln y)^{1/2}}\;.
504: \label{g_y}
505: \end{equation}
506: %
507: This form incorporates logarithmic corrections which, with $i=1$ and $M\sim
508: \xi$,
509: are in agreement with the form of the surface magnetization given in
510: equation~(\ref{ms}). Now, in analogy with~(\ref{betas}), an effective surface
511: magnetization exponent can be defined as
512: %
513: \begin{equation}
514: \frac{\beta_{\rm s}(y)}{\nu}
515: =\frac{\ln[g_0(y(1-\delta))/g_0(y(1+\delta))}{\ln[(1+\delta)/(1-\delta)]}\,
516: \label{betas_y}
517: \end{equation}
518: %
519: which, finally when $y\ll 1$, will be a function $\beta_{\rm s}(M)$ of the
520: characteristic length of the problem.
521:
522: In order to get rid of the logarithmic factor, $g_{\rm s}(i/l_{\rm s})$, we have
523: calculated the ratio $m_i(M)/m_{i, {\rm pure}}(M)$ of the initial profiles in
524: the
525: random and pure systems. Assuming that $g_{\rm s}$ has the same logarithmic
526: singularity for both, we arrive to the conclusion that the ratio of profiles
527: scales with the difference of the effective exponents,
528: $\Delta \beta_{\rm s}(M)=\beta_{\rm s}(M)-\beta_{\rm s,pure}(M)$.
529:
530: In the actual calculation we set $h_{\rm s}=.01$ and $h=1$, and performed MC
531: simulations on systems with sizes $M=16$ and $64$ and different aspect ratios
532: $\alpha=L/M$. For $\alpha=4$ and $8$, the initial part of the profiles
533: turned out to be indistinguishable. The ratios calculated for the largest
534: $\alpha$ are presented in figure~\ref{fig4} and, from the extrapolated initial
535: slopes in a log-log plot, the differences of the effective exponents take the
536: values $\Delta \beta_{\rm s}(16)=0.08(1)$ and $\Delta \beta_{\rm
537: s}(64)=0.07(1)$.
538: In order to compare these estimates with the effective exponents obtained in
539: section~3 with a finite $t$, we use the correspondence of equation~(\ref{t_L})
540: with $M=\xi$. The data points obtained in this way are inserted in
541: figure~\ref{fig2}. They seem to fit very well with the predicted theoretical
542: curve and are located in the descending part of the curve. Therefore this
543: calculation gives further support to the theoretical results about the form of
544: the logarithmic corrections.
545:
546: \section{Discussion}
547:
548: Marginally irrelevant operators are responsible for logarithmic corrections
549: to scaling, the form of which can be often predicted by field theory and
550: conformal invariance. According to second-order perturbation theory, the random
551: surface field is expected to be such a marginally irrelevant operator at the
552: surface fixed point of the 2D Ising model. This conjecture, which is made on the
553: basis of the replica trick and in the weak disorder limit, is confronted here
554: with the results of extensive MC simulations for varying strength of the
555: disorder. The calculated effective surface magnetization exponent, which depends
556: either on the distance $t$ from the critical point or on the finite size $M$ of
557: the critical system, varies with these parameters. Since this variation is
558: non-monotonic, a direct extrapolation to the fixed-point values cannot be made
559: from the data available on finite systems. However, the variation of the
560: effective exponents is in good agreement with the theoretical form,
561: which contains the predicted logarithmic correction to scaling to the pure
562: systems critical behaviour. For the largest systems and for the strongest random
563: fields, the numerical results are not too far from that obtained perturbatively
564: in linear order. Therefore we interpret our results as numerical evidence in
565: favour of the validity of the field-theoretical predictions.
566:
567: \ack
568:
569: We thank D E Feldman for bringing the results of~\cite{feldman02} to
570: our attention.
571: FAB thanks the French Ministry of Foreign Affairs for a research grant. This
572: work has been supported by the Hungarian National Research Fund under grant no
573: OTKA TO34183, TO37323, MO28418 and M36803, by the Ministry of Education under
574: grant no FKFP 87/2001, by the EC Centre of Excellence (no~ICA1-CT-2000-70029).
575: Simulations have been done on the IA32 cluster of the Regionales Rechenzentrum
576: Erlangen and at CINES Montpellier under project pnm2318. The Laboratoire de
577: Physique des Mat\'eriaux is Unit\'e Mixte de Recherche CNRS~7556.
578:
579: \Bibliography{99}
580:
581: \bibitem{igloi93}
582: Igl\'oi F, Peschel I and Turban L 1993 {\it Adv. Phys.}
583: {\bf 42} 683
584:
585: \bibitem{diehl90a}
586: Diehl H W and N\"usser A 1990 \ZP B {\bf 79} 69
587:
588: \bibitem{diehl90b}
589: Diehl H W and N\"usser A 1990 \ZP B {\bf 79} 79
590:
591: \bibitem{pleimling98}
592: Pleimling M and Selke W 1998 {Eur. Phys. J.} B {\bf 1} 385
593:
594: \bibitem{diehl98}
595: Diehl H W 1998 {Eur. Phys. J.} B {\bf 1} 401
596:
597: \bibitem{chung00}
598: Chung M-C, Kaulke M, Peschel I, Pleimling M and Selke W 2000
599: {Eur. Phys. J.} B {\bf 18} 655
600:
601: \bibitem{selke02}
602: Selke W, Pleimling M, Peschel I, Kaulke M, Chung M-C
603: and Catrein D 2002 \JMMM {\bf 240} 349
604:
605: \bibitem{weller85}
606: Weller D, Alvarado S F, Gudat W, Schr\"{o}der K and
607: Campagna M 1985 \PRL {\bf 54} 1555
608:
609: \bibitem{rau86}
610: Rau C and Eichner S 1986 \PR B {\bf 34} 6347
611:
612: \bibitem{rau87}
613: Rau C and Robert M 1987 \PRL {\bf 58} 2714
614:
615: \bibitem{cardy91}
616: Cardy J 1991 \JPA {\bf 24} L1315
617:
618: \bibitem{harris74}
619: Harris A B 1974 \JPC {\bf 7} 1671
620:
621: \bibitem{mon88}
622: Mon K K and Nightingale M P 1988 \PR B {\bf 37} 3815
623:
624: \bibitem{feldman02}
625: Feldman D E and Vinokur V M 2002 \PRL {\bf 89} 227204
626:
627: \bibitem{imry75}
628: Imry Y and Ma S K 1975 \PRL {\bf 35} 1399
629:
630: \bibitem{McCoyWu}
631: McCoy B M and Wu T T 1973 {\it The Two-Dimensional Ising Model}
632: (Cambridge: Harvard University Press) p 132
633:
634: \bibitem{igloi91}
635: Igl\'oi F, Turban L and Berche B 1991 \JPA {\bf 24} L1031
636:
637: \bibitem{dotsenko81}
638: Dotsenko Vik S and Dotsenko Vl S 1981 {\it Sov. Phys.---JETP} {\bf 33} 37
639:
640: \bibitem{shalaev84}
641: Shalaev B N 1984 {Sov. Phys. Solid State} {\bf 26} 1811
642:
643: \bibitem{selke90}
644: Wang J-S, Selke W, Dotsenko Vl S and Andreichenko V B 1990
645: {\it Physica} A {\bf 164} 221
646:
647: \bibitem{kuhn94}
648: K\"uhn R 1994 \PRL {\bf 73} 2268
649:
650: \bibitem{plechko98}
651: Plechko V N 1998 \PL A {\bf 239} 289
652:
653: \bibitem{shchur02}
654: Shchur L N 2002 \PR E {\bf 65} 016107
655:
656: \bibitem{baxter}
657: Baxter R J 1982 {\it Exactly Solved Models in Statistical Mechanics}
658: (London: Academic) p~118
659:
660: \bibitem{cardy84}
661: Cardy J 1984 \JPA {\bf 17} L385
662:
663: \bibitem{pleimling04}
664: Pleimling M 2004 \JPA {\bf 37} R79
665:
666: \bibitem{cr}
667: Czerner P, and Ritschel U 1997 Int. J. Mod. Phys. B {\bf 11} 2075
668:
669: \bibitem{bariev}
670: Bariev R Z 1988 Theo. Math. Phys. {\bf 77} 1090
671:
672: \endbib
673:
674: \end{document}
675:
676:
677: