1: \documentstyle[aps, prb, epsfig]{revtex}
2: %\documentstyle[aps, prl, epsfig]{revtex}
3: %\newcommand{\figdir}{Figures/}
4: \newcommand{\figdir}{./}
5: \newcommand{\pins}{p_{I}}
6: \newcommand{\psc}{p_{S}}
7: \newcommand{\sins}{\sigma_{I}}
8: \newcommand{\ssc}{\sigma_{S}}
9:
10: \newcommand{\spar}{\sigma_{{\small M},\|}}
11: \newcommand{\sper}{\sigma_{M,\perp}}
12:
13: \newcommand{\beq}{\begin{equation}}
14: \newcommand{\unbeq}{\end{equation}}
15: \newcommand{\beqa}{\begin{eqnarray}}
16: \newcommand{\unbeqa}{\end{eqnarray}}
17:
18: \begin{document}
19: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname
20: @twocolumnfalse\endcsname
21: \title{
22: Magnetoresistance of Three-Constituent Composites:
23: Percolation Near a Critical Line
24: }
25: \author{Sergey V.\ Barabash$^1$, David J.\ Bergman$^{1,2}$, D.\ Stroud$^1$}
26: \address{$^1$Department of Physics,
27: The Ohio State University, Columbus, Ohio 43210;\\
28: $^2$School of Physics and Astronomy, Raymond and Beverly Sackler
29: Faculty of Exact Sciences,\\Tel Aviv University,
30: Tel Aviv 69978, Israel}
31:
32: \date{\today}
33:
34: \maketitle
35:
36: \begin{abstract}
37: Scaling theory, duality symmetry, and numerical simulations of a random
38: network model are used to study the magnetoresistance of a
39: metal/insulator/perfect conductor composite with a
40: disordered columnar microstructure.
41: The phase diagram is found to have a critical line which separates regions
42: of saturating and non-saturating magnetoresistance. The percolation
43: problem which describes this line is a generalization of anisotropic
44: percolation. We locate the percolation threshold and determine the
45: values of the critical exponents
46: $t_{\|} = t_{\perp} = s_{\|} = s_{\perp} = 1.30 \pm 0.02$,
47: $\nu=4/3 \pm 0.02$, which are the same as in two-constituent 2D
48: isotropic percolation.
49: %and $\delta_{\|} = \delta_{\perp}= 1.00 \pm 0.07$.
50: We also determine the exponents which characterize the critical
51: dependence on magnetic field, and confirm numerically that $\nu$ is
52: independent of anisotropy.
53: We propose and test a complete scaling description of the
54: magnetoresistance in the vicinity of the critical line.
55: \end{abstract}
56: \vskip1.5pc]
57:
58: \section{Introduction}
59:
60: Magnetotransport in composite conductors has attracted
61: increased attention due to the discovery of new effects, like
62: the appearance of non-saturating magnetoresistance in a
63: metal/insulator columnar composite (denoted by $M/I$), induced
64: by the Hall effect in the $M$ (i.e., the metallic) constituent.
65: \cite{BS1} This means that, even when the $M$ constituent
66: has no {\em intrinsic} magnetoresistance but only a Hall resistivity,
67: an induced magnetoresistance appears in the mixture and continues
68: to increase as ${\bf B}^2$ ({\bf B} is the applied magnetic field)
69: {\em indefinitely} whenever the Hall-to-Ohmic resistivity ratio
70: of that constituent is much greater than 1. By contrast, in
71: a columnar composite of normal conductor and perfect
72: conductor (denoted by $M/S$), while there also appears an
73: induced magnetoresistance, it saturates when {\bf B} is
74: comparably large. \cite{BS1} Related new effects were also found in
75: periodic microstructures of either the $M/I$ type or the
76: $M/S$ type, where the induced
77: magnetoresistance often exhibits a strong anisotropy.
78: \cite{BergStrelPRB94,TornowWeissKlitzingEberlBergStrelPRL96}
79:
80: In this paper, we study the magnetoresistance of
81: three-constituent composites with a {\em random columnar microstructure}.
82: Specifically, we analyze
83: a composite consisting of cylindrical $I$
84: and $S$ inclusions, though not necessarily circular-cylinders,
85: in an $M$ host film, with cylinder
86: axes perpendicular to the film, and with both a
87: magnetic field and a uniform macroscopic or volume averaged electric current
88: applied {\em in the film plane}.
89: We assume that the $M$ constituent has a
90: Hall effect, with a Hall-to-transverse-Ohmic resistivity ratio
91: $H \equiv \rho_{Hall}/\rho_{Ohmic} = \mu|{\bf B}|$,
92: where $\mu$ is the Hall mobility.
93: We are concerned with the effective resistivity tensor $\hat \rho_e$ of this
94: system at {\it large} $H$
95: ($\hat \rho_e$ is
96: defined by $\langle {\bf E}\rangle = \hat \rho_e\langle {\bf J}\rangle$, where
97: $\langle {\bf E}\rangle$ and $\langle {\bf J}\rangle$ are the volume averaged
98: electric field and current density.\cite{SSP})
99:
100: This system might appear to be inherently three-dimensional (3D),
101: because the Hall effect will generate an
102: electric field with a component perpendicular to the film even with an
103: in-plane applied current. However, the electric field perpendicular to
104: the film plane vanishes because of the columnar $S$ inclusions.
105: Moreover, this 3D problem can be
106: reduced\cite{BS1} to that of calculating the effective conductivity of
107: a {\em two-dimensional} (2D)
108: composite of perfect insulator $I$ with $\sins = 0$,
109: perfect conductor (or superconductor) $S$ with $\ssc=\infty$,
110: and anisotropic 2D metal $M$ with conductivity tensor
111: \beq
112: \label{2Dsigmas}
113: \hat{\sigma}_M
114: \equiv\left(
115: \begin{array}{cc}
116: \sper & 0 \\
117: 0 &\spar \\
118: \end{array}
119: \right)
120: = \frac{1}{\rho_M} \left(
121: \begin{array}{cc}
122: \frac{1}{1+H^2}& 0 \\
123: 0 & \frac{1}{\lambda} \\
124: \end{array}
125: \right).
126: \unbeq
127: The conductivities $\spar$ and $\sper$ correspond to the
128: in-plane directions
129: parallel and perpendicular to the applied magnetic field $\bf B$.
130: This transformation underlies our further discussion.
131: The form assumed for $\hat{\sigma}_M$ means that the $M$ constituent
132: remains an isotropic conductor, even in a magnetic field.
133: This excludes ``open orbit'' conductors, but includes the
134: possibility of some intrinsic magnetoresistance:
135: In that case both the transverse and the longitudinal Ohmic
136: resistivities, $\rho_M$ and $\lambda\rho_M$, would depend upon
137: $\bf B$. Our subsequent discussion will assume, for simplicity,
138: that $\lambda=1$ and $\rho_M$ are both independent of $\bf B$,
139: as would be the case if $M$ were a free electron or free hole
140: conductor. In that case, $H$ is simply proportional to $|{\bf B}|$.
141: The three-constituent
142: problem in 2D, but with a scalar $\hat{\sigma}_M$, was
143: treated previously in Ref.\ \onlinecite{kogut79}. Scaling and
144: simulational studies, of magnetotransport in 3D, two-constituent
145: $M/I$ composites with a random isotropic microstructure, were
146: also performed previously. \cite{SarBergStrelPRB93}
147:
148: \section{Magnetoresistance and percolation}
149:
150: The relation of this problem to percolation can be understood
151: by considering very large $H$.
152: In this limit, $\sper\rightarrow 0$, and within the metallic constituent
153: the in-plane current prefers to flow in the $\spar$ direction.
154: For large enough volume fraction $\psc$ of the $S$ constituent,
155: and small enough volume fraction $\pins$ of $I$,
156: there always exist current paths between $S$ grains that are
157: everywhere parallel to the high-conductivity direction in $M$
158: [cf.\ Fig.\ \ref{fig:percolation}(a)].
159: Therefore, in this regime,
160: at large $H$ the effective composite conductivity (and hence resistivity)
161: saturates at some finite value. In the opposite case
162: (small $\psc$, large $\pins$), the current
163: will be forced in some regions to flow in the low conductivity direction
164: [cf.\ Fig.\ \ref{fig:percolation}(b)]. In this case,
165: the macroscopic or effective conductivity of the system will be
166: proportional to $\sper$;
167: hence, the effective resistivity $\rho_e$ satisfies
168: $\rho_e \sim 1/\sper \sim H^2$, i.e., $\rho_e$ will never saturate.
169:
170: Next, we qualitatively discuss the expected ``phase diagram'' of this
171: composite. The effective conductivity $\hat\sigma_e$ of the 2D mixture is
172: a $2\times 2$ tensor
173: (like the effective 2D resistivity $\hat \rho_e = \hat \sigma_e^{-1}$),
174: whose components depend on the applied magnetic field.
175: If the in-plane microstructure is isotropic,
176: then the principal axes of $\hat \sigma_e$ are determined by
177: ${\bf B}$. At large $H$, both $\sigma_{e,\|}$
178: and $\sigma_{e,\perp}$
179: can either decrease as $H^{-2}$ for arbitrarily large $H$,
180: or else saturate at some finite value.
181: In principle, there could be a region in the phase diagram where
182: the resistivity saturates along one principal axis, but not along the other.
183: But we expect no such region in an infinite system.
184: This conclusion, as well as the location of the critical line, follow from
185: the relation between the present problem and the anisotropic percolation
186: model in 2D, as we now explain.
187:
188: \begin{figure}[b]
189: \centerline{
190: \epsfig{file=\figdir fig1a.eps, width=1.4in}
191: \epsfig{file=\figdir fig1b.eps, width=1.4in}
192: }
193: \vskip0.5pc
194: \caption[]
195: {
196: Schematic of transport in a 2D composite of a perfect conductor $S$,
197: insulator $I$, and a very anisotropic metal $M$
198: (the host medium in this figure).
199: The current prefers to flow between $S$ grains
200: only in the direction of high metallic conductivity $\spar$,
201: i.e., parallel to ${\bf B}$.
202: At high $S$ volume fraction $\psc$, such a path always exists,
203: as shown schematically in (a). At smaller $\psc$ or larger
204: $\pins$, the current must sometimes flow through the
205: metal in the low-conductivity direction ($\sper\sim H^{-2}$), as shown in (b).
206: This behavior leads to a non-saturating effective resistivity.
207: }
208: \label{fig:percolation}
209: \end{figure}
210:
211: Anisotropic percolation is usually defined in terms of a random resistor
212: network (RRN), where the bond occupation probability depends on
213: the bond orientation. In our 2D network model those probabilities
214: will be chosen as $p_\parallel=p_M+p_S$ for bonds along {\bf B},
215: the high conductivity direction of the $M$ constituent,
216: and $p_\perp=p_S$ for bonds
217: perpendicular to {\bf B}, the low conductivity direction. For
218: an infinite network, the percolation threshold for both directions
219: occurs when \cite{sykes63}
220: \begin{equation}
221: \label{anisotropicPT}
222: 1=p_\|+p_\perp=2p_S+p_M\Rightarrow\pins=\psc.
223: \end{equation}
224: This threshold separates the regimes of saturating and non-saturating
225: magnetoresistance. It
226: agrees with a prediction based on the
227: effective medium approximation (EMA), \cite{BergPRLsub,BergmanPRB2001}
228: as well as with
229: numerical results below.
230: In a composite with less symmetry between the $I$ and $S$ constituents,
231: the threshold will usually differ from $p_I=p_S$.
232:
233:
234: The predicted phase diagram is shown in Fig.~\ref{fig:diagram}.
235: For $\psc>0.5$, the $S$ constituent percolates, hence the composite
236: is a perfect conductor. Similarly, if $\pins>0.5$ then the
237: combination of $M$ and $S$ constituents does not percolate, hence
238: the composite is a perfect insulator. These behaviors are independent
239: of how much of the $M$ constituent is present.
240: The remainder of the phase diagram is divided into saturating and
241: non-saturating regimes by the line
242: $\pins=\psc$. In both regimes, our numerical results indicate that it is
243: irrelevant whether the $M$ constituent percolates by itself
244: (as in the area below the dotted line), or whether only the
245: aggregate of $M$ and $S$ constituents percolates.
246:
247: \begin{figure}[h]
248: \centerline{ \epsfig{file=\figdir fig2.eps,width=2.2in} }
249: \vskip0.5pc
250: \caption[]
251: {
252: Schematic phase diagram for an $M/I/S$ composite
253: with random columnar microstructure.
254: $\pins$ and $\psc$ are the volume fractions of $I$ and $S$;
255: the $M$ volume fraction is $p_M=1-\pins-\psc$.
256: In the shaded regions, either the $S$ or the $I$ constituent
257: percolates. The critical line $\pins=\psc$ separates regions
258: with saturating and non-saturating magnetoresistance. The dotted line
259: corresponds to percolation of the metallic constituent alone
260: ($p_M>0.5$ or $\pins+\psc<0.5$); no critical behavior appears to
261: be associated with this line.
262: }
263: \label{fig:diagram}
264: \end{figure}
265: %%%%%%%%%%%%%%%%%%%%%%%%%%% Critical exponents %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
266: \newcommand{\gs}{g_{sat,i}}
267: \newcommand{\gns}{g_{non-sat,i}}
268: %\section{Critical exponents for the problem}
269: \section{Critical exponents and scaling properties}
270:
271: The critical behavior on approaching the critical line
272: is described by a number of critical exponents. One of these governs the
273: two correlation lengths,
274: $\xi_\|$ and $\xi_\perp$, which diverge at the critical line. But we
275: expect that the divergence of both will be governed by {\em the same}
276: critical exponent $\nu$, because their ratio $\xi_\|/\xi_\perp$ is determined
277: by some function of the constituent volume fractions which
278: remains finite on crossing the critical line. \cite{note1,sarychev79}
279: Thus we write
280: \begin{equation}
281: \label{def:xi}
282: \xi_\| \sim \xi_\perp \sim (\pins-\psc)^{-\nu}.
283: \end{equation}
284: Moreover, since the correlation length is just a geometrical property
285: of a percolating system, it should have the same value $\nu_{a}$
286: as in anisotropic percolation, i.e.,
287: $\nu = \nu_{a}$. An earlier renormalization group analysis predicts
288: that anisotropic percolation is governed by the isotropic
289: fixed point. \cite{anisotropic} We therefore expect
290: that $\nu_{a}$ has the same value
291: as in 2D isotropic percolation, which has been shown
292: both analytically\cite{nijs} and numerically\cite{derrida} to equal 4/3.
293: However, a similar proof has not been given for anisotropic
294: percolation; \cite{anisotropic_nu} moreover, some time ago it was reported that
295: $\nu$ does depend on anisotropy. \cite{shapiro}
296: Our numerical results below give
297: $\nu = 4/3 \pm 0.02$, independent of anisotropy.
298:
299: The magnetoresistance is described by other critical indices.
300: At any point such that $\pins<\psc$
301: (the region below the critical line in Fig.~\ref{fig:diagram}),
302: the effective conductivity saturates at some finite value
303: ($i=\|$ or $i=\perp$):
304: \begin{equation}
305: \label{def:s}
306: \gs \equiv \lim_{H \rightarrow \infty} \sigma_{sat, i}(H).
307: \end{equation}
308: As we approach the critical line (say, along the line $p_M =$ const),
309: $\gs$ must tend to 0,
310: since on the other side of that line the conductivity vanishes at large $H$ as
311: $H^{-2}$. Hence we can introduce the critical exponents $t_{\|}$ and $t_{\perp}$
312: according to
313: \begin{equation}
314: \label{def:gs}
315: \gs \sim (\psc-\pins)^{t_i}.
316: \end{equation}
317: In the region $\pins>\psc$, the effective conductivity is
318: proportional to $1/H^2$. Therefore, we can define
319: the finite limiting value
320: \begin{equation}
321: \label{def:gns}
322: \gns \equiv \lim_{H \rightarrow\infty} H^2\sigma_{non-sat,i}(H).
323: \end{equation}
324: Since $\gns$ must diverge as the critical line is approached, we can define
325: the critical exponents $s_{||}$ and $s_{\perp}$, according to
326: \begin{equation}
327: \label{def:t}
328: \gns \sim (\pins-\psc)^{-s_{i}}.
329: \end{equation}
330: Just as in the case of $\nu$, we expect $s_{\|}= s_{\perp}$,
331: $t_{\|}=t_{\perp}$.
332: We might expect that the actual values of $t$ and $s$ are
333: determined, once again, by the connection to anisotropic percolation.
334: However, this connection is more subtle than for $\nu$,
335: because $t$ and $s$ refer to different
336: physical quantities in the present problem, than in anisotropic
337: percolation.
338: Nonetheless, it is reasonable to expect that both problems
339: belong to the same universality class, implying
340: $s_i = t_i$ = 1.30, the values of the two conductivity exponents in
341: conventional 2D percolation problems. \cite{t2D,YDelta,t2Dnew}
342: Note also that EMA yields $s_i=t_i=1$, \cite{BergPRLsub,BergmanPRB2001}
343: and that continuum composites sometimes belong to a different
344: universality class of percolation, with different values of
345: the critical exponents, depending upon the microstructural details.
346: \cite{FengHalperinSenPRB87}
347:
348: %%%%% NEW RESULTS
349:
350:
351: Finally, consider systems exactly at the percolation threshold, i.e., at
352: $p_I = p_S$. Any finite-size sample inevitably falls into either the
353: percolating class, with $\rho_e \sim H^0$ as $H \rightarrow \infty$, or the
354: non-percolating class, with $\rho_e \sim H^2$.
355: On increasing the size of the system,
356: one finds that the field where a saturating sample achieves saturation also
357: increases. The same size effect also holds for non-saturating
358: samples: the characteristic field at which $\rho_e$ starts to vary as $H^2$
359: increases with the size of the system. Below that field, or at any field
360: in the limit of a system of infinite size, the behavior of the
361: magnetoresistance is described by yet another critical exponent
362: $\delta_i$, defined by
363: \begin{equation}
364: \label{def:delta}
365: \rho_{e,i}\sim|H|^{\delta_i}.
366: \end{equation}
367: The EMA analysis\cite{BergPRLsub,BergmanPRB2001} yields $\delta_i=1$.
368:
369: In fact, the result $\delta_i = 1$
370: also follows from a duality argument, if one assumes that
371: $\delta_\| = \delta_\perp$. The duality principle\cite{SSP} gives
372: %\beqa
373: $$1=\sigma_{e,\|}\left(
374: \sigma_{M\|},\sigma_{M\perp},\sigma_S,\sigma_I
375: \right)
376: \sigma_{e,\perp}\left(
377: \frac{1}{\sigma_{M\perp}},\frac{1}{\sigma_{M\|}},
378: \frac{1}{\sigma_S},\frac{1}{\sigma_I}
379: \right).$$% \nonumber \\
380: %&&=1.
381: %\label{eq:duality}
382: %\unbeqa
383: It immediately follows that the anisotropic percolation thresholds
384: for the two directions $\parallel$, $\perp$ must coincide, for
385: otherwise the singular behaviors of $\sigma_{e,\|}$ and $\sigma_{e,\perp}$
386: would not cancel, as required by this equation. It also follows,
387: rigorously, that $t_\parallel=s_\perp$ and $t_\perp=s_\parallel$.
388: Using the homogeneity of $\sigma_{e,\|}$, $\sigma_{e,\perp}$
389: as functions of the various constituent conductivities, and noting
390: that $\sigma_I=0$ and $\sigma_S=\infty$, we can rewrite this equation as
391: $$\frac{\sigma_{M\perp}}{\sigma_{M\|}}=
392: \sigma_{e,\|}\left(1,\frac{\sigma_{M\perp}}{\sigma_{M\|}},\infty,0\right)
393: \sigma_{e,\perp}\left(1,\frac{\sigma_{M\perp}}{\sigma_{M\|}},0,\infty\right).$$
394: Note that the relation $p_I=p_S$ for the percolation threshold follows
395: rigorously from this equation if the microstructure is invariant
396: under the interchange of the $S$ and $I$ constituents.
397: At the percolation threshold, and when $|H|\gg 1$, this reduces to
398: $$\frac{1}{H^2}\propto\frac{1}{|H|^{\delta_\parallel+\delta_\perp}}
399: \Rightarrow\delta_\parallel+\delta_\perp=2.$$
400: The assumption that $\delta_\| = \delta_\perp$, and hence the final
401: result $\delta_\| = \delta_\perp=1$, are supported by the physical
402: picture of the anisotropic percolation process, and is consistent with
403: the expectation that $t_\parallel=t_\perp$ and $s_\parallel=s_\perp$.
404: These equalities also follow if a simple scaling description applies
405: to both $\sigma_{e,\|}$ and $\sigma_{e,\perp}$ with {\em the same
406: scaling variable}. Thus, for an $M/I/S$ columnar composite precisely
407: at the critical composition (i.e., on the phase boundary line between
408: saturating and non-saturating magnetoresistance), we expect to
409: find
410: $$\sigma_{e,\|} \sim \sigma_{e,\perp} \sim\frac{1}{|H|},\;\;\;
411: \rho_{e,\|} \sim \rho_{e,\perp} \sim |H|\Rightarrow\delta=1.$$
412:
413: Near the transition, where $|p_I-p_S|\ll 1$ and
414: $\sigma_{M,\perp}/\sigma_{M,\|}\cong 1/H^2\ll 1$,
415: we expect that a scaling description of the critical behavior is applicable.
416: In view of the preceding discussion, that description can be formulated
417: as follows:
418: \begin{eqnarray}
419: \frac{\rho_{e,i}}{\rho_{M,\|}}&\cong&{\rm sign}\Delta p\,|\Delta p|^{-t}
420: F_i(Z);
421: \label{def:F}\\
422: \Delta p&\equiv& p_I-p_S,\;\;\;
423: i=\|,\,\perp,\;\;\;
424: Z\equiv |H||\Delta p|^t\,{\rm sign}\Delta p.
425: \nonumber
426: \end{eqnarray}
427: The scaling functions $F_i(Z)$ should have the following asymptotic
428: forms for extreme values of the scaling variable $Z$:
429: \begin{equation}
430: F_i(Z)\sim\left\{\begin{array}{lll}
431: -Z^0 & {\rm for}\;Z\ll-1,& {\rm i.e.}\;p_S>p_I,\\
432: Z^2 & {\rm for}\;Z\gg 1,& {\rm i.e.}\;p_S<p_I,\\
433: Z & {\rm for}\;|Z|\ll 1,& {\rm i.e.}\;p_S\stackrel{>}{\scriptstyle <}p_I.
434: \end{array}\right.
435: \label{F_i_Z}
436: \end{equation}
437: The first two lines in this expression follow from the fact that
438: we expect to have $\rho_{e,i}\sim H^0$ in the saturating regime
439: and $\rho_{e,i}\sim H^2$ in the non-saturating regime. The
440: third line results from the necessity to cancel the dependence
441: of $\rho_{e,i}$ upon $\Delta p$ when $\Delta p\rightarrow 0$.
442: This kind of behavior was already found previously using
443: EMA, \cite{BergPRLsub,BergmanPRB2001} where the incorrect value
444: $t=1$ was found, as is usual when that approximation is invoked.
445: %%%CHANGE
446: Note that, even though $F_i(Z)$ has a different
447: analytic form for large $Z$, depending on the sign of $Z$ or $\Delta p$,
448: these functions are expected to vary smoothly when $Z$ passes
449: through 0.
450:
451: Finally, note that duality is a general symmetry property of 2D systems,
452: and does not require that the $I$ and $S$ inclusions have similar
453: shapes or spatial distributions. Thus, the conclusions regarding
454: values of $s$, $t$, $\delta$ are valid even if the line of
455: critical compositions differs from the simple straight line
456: $p_S=p_I$, which is valid only for the case where the $I$ and $S$
457: constituents appear in the composite in a symmetric fashion.
458:
459:
460: \section{Results}
461: %%%%%%%%%%%%%%%%%%%%%%%%%% Results %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
462:
463: We carried out calculations on simple-square-lattice RRN's, choosing the
464: resistors in accordance with the model described above.
465: The calculation was done using the Y-$\Delta$ bond elimination
466: algorithm. \cite{YDelta} Besides being very efficient in 2D,
467: this algorithm allows the inclusion of both
468: perfectly insulating and perfectly conducting
469: bonds without any approximations.
470: This technique is much more efficient than the techniques available
471: for 3D networks. That is why the results obtained here for columnar
472: systems are much more detailed and accurate than the results obtained
473: previously for 3D isotropic systems by simulations of 3D random
474: network models. \cite{SarBergStrelPRB93}
475:
476: The networks for our results were constructed as follows:
477: each bond was independently and randomly
478: chosen to be insulating, perfectly conducting,
479: or metallic, with appropriate probabilities. A metallic bond was
480: assigned an appropriate conductance, depending on whether it was oriented
481: parallel or perpendicular to the magnetic field. To calculate
482: $\gs$, as can be seen from Eqs.\ (\ref{2Dsigmas}) and (\ref{def:s}),
483: the conductances of the constituents should be taken as
484: \begin{equation}
485: \label{Ssigmas}
486: \sins=0 \text{ , }
487: \ssc=\infty \text{ , } \
488: \spar=1 \text{ , }
489: \sper=0.
490: \end{equation}
491: But in order to calculate
492: $\gns$, one needs to multiply all the conductances by $H^2$ and take the limit
493: $H\rightarrow \infty$, which leads to
494: \begin{equation}
495: \label{NSsigmas}
496: \sins=0 \text{ , }
497: \ssc=\infty \text{ , } \
498: \spar=\infty \text{ , }
499: \sper=1.
500: \end{equation}
501:
502:
503: The definitions (\ref{def:gs}) and (\ref{def:t}) for the exponents $s$ and $t$
504: are exactly valid only for an infinite system.
505: Instead of using these definitions directly, we adopted the
506: finite-size scaling approach, as described below.
507: For a finite-size system generated {\em exactly at}
508: the percolation threshold $\pins=\psc$,
509: the system size $L\times L$ is always smaller than the (infinite) correlation
510: length $\xi$. From Eq.\ (\ref{def:xi}) it follows that such a system
511: behaves like a system at another volume fraction satisfying
512: $|\pins-\psc| \sim L^{-1/\nu}$.
513: Therefore, the average of $\gs$ or $\gns$ over many such systems
514: should scale as $\left< \gs\right> \sim L^{-t_i/\nu}$ or
515: $\left<\gns\right>\sim L^{s_i/\nu}$. (Due to the appearance of
516: percolating samples, the latter average is actually
517: infinite. A finite result is obtained by averaging
518: only over non-percolating samples. This procedure
519: does not change the scaling form, because {\em at the percolation threshold}
520: the fraction of samples which are non-percolating is
521: asymptotically size-independent, as discussed below).
522:
523: Using RRN's with $L$ ranging from 100 to 2000 and $p_I=p_S$,
524: we estimated $s_i$ and $t_i$ for $p_M=0.2$, $0.5$, $0.8$, and, with
525: less accuracy, for several other values of $p_M$.
526: (At $p_M=0.8$ we could estimate only $t_\|$ and $s_\perp$, since these systems
527: usually percolated only in one direction but not in the other.)
528: The finite size scaling assumption is supported very well, as demonstrated
529: in Fig.~\ref{fig:scaling} for $p_M=0.2$.
530: Assuming $\nu=4/3$, we find for all values of $p_M$ studied that
531: %\begin{equation} \label{result:ts}
532: $t_{||} =t_{\perp} =s_{||}=s_{\perp}=1.30\pm 0.02$.
533: %\end {equation}
534: This value supports the hypothesis that the problem belongs to the same
535: universality class as both isotropic and anisotropic percolation in 2D.
536:
537:
538: The assumption that $\nu=4/3$ can be tested for our problem
539: by considering percolation in finite-size systems. For such systems,
540: the ``percolation threshold'' in the direction $i$
541: is naturally defined as the value of $p_S$ for which exactly
542: one half of the systems percolate in that direction:
543: $P_i\equiv N_{sat,i}/N_{total}=1/2$, where $P_i$ is the probability
544: of finding saturating behavior in the direction $i$.
545: Fig.~\ref{fig:probabilistic} then shows
546: that the percolation threshold depends on the system size $L$. In
547: fact, a finite system has
548: two ``percolation thresholds'', ${\psc}_{0,\perp}$ and ${\psc}_{0,\|}$, located
549: respectively above and below the percolation threshold of an infinite system,
550: and approaching $p_I$ as $|{\psc}_{0,i} - p_I|\sim L^{-1/\nu}$
551: with increasing $L$. \cite{PercThresh}
552: We have checked these hypotheses numerically by calculating
553: $| {\psc}_{0,\perp} - {\psc}_{0,\|} |$ for four linear
554: sizes $L$ at $p_M=0.2$ and $0.5$,
555: and verified that $\nu=4/3 \pm 0.02$, independent of anisotropy.
556: %
557: \begin{figure}
558: \epsfig{file=\figdir fig3.eps, width=3.in}
559: \vskip0.3pc
560: \caption[]
561: {
562: Calculated $\left< \gs \right>$ and $\left< \gns \right>$
563: [cf.\ definitions (\ref{def:s}) and (\ref{def:gns})] plotted on a log-log
564: plot versus linear size $L$ of the RRN for $p_M=0.2$ and
565: $\pins=\psc=0.4$.
566: Error bars are smaller than the dot sizes.
567: The slopes of the two lines on this plot equal $t_i/\nu$ and $-s_i/\nu$,
568: and are consistent with the value 0.975, corresponding to $t=s=1.30$
569: and $\nu=4/3$.
570: }
571: \label{fig:scaling}
572: \end{figure}
573: %%%%%NEW RESULTS
574:
575: Fig.\ \ref{fig:probabilistic} also makes clear that the
576: fraction of samples which are non-percolating
577: is size-independent at the percolation threshold.
578: Specifically, Fig.~\ref{fig:probabilistic} shows
579: the fraction of percolating samples of a given size vs.\ concentration $p_S$.
580: The lines corresponding to different sample sizes $L$ intersect
581: at the percolation threshold $p_I=p_S$ (vertical line).
582: Thus, as $L \rightarrow \infty$,
583: the fraction of the samples percolating at $p_I=p_S$ approaches
584: a certain limiting value determined by the other parameters of the system,
585: namely, the direction of percolation and the concentration $p_M$ of the metal.
586: This limiting value can be related to the spanning probability in the case
587: of {\em isotropic} percolation in a system of rectangular shape,
588: which was found to depend on the aspect ratio
589: of the rectangle. \cite{SpanningProbability}
590: In the present paper, we only consider systems
591: with the aspect ratio equal to unity; however, the {\em correlation lengths}
592: are different in the two directions. Thus, $\xi_\perp / \xi_\|$
593: may be thought of as an effective aspect ratio for the present problem.
594:
595: In order to test the scaling behavior at finite values of $H$ and
596: $\Delta p$, we also simulated random networks at $p_S=p_I$
597: with large but finite values of both $H$ and $L$.
598: %%%CHANGE
599: Since the RRN's required for such calculations
600: contain bonds with widely different
601: conductivities [cf.\ Eq.\ (\ref{2Dsigmas})], special care must be taken
602: when performing these calculations on large systems. \cite{note:CPP}
603: For that reason we wrote special code for storing numbers as large (small)
604: as roughly $10^{\pm 6\times 10^8}$.
605: Using this code, we studied RRN's of size up to
606: $4000\times 4000$ in fields ranging from 0 up to $H^2=10^8$.
607: One useful way to
608: exhibit those results is as plots of $\ln[(H^2+1)\sigma_{e,i}]$
609: vs.\ $\ln(H^2+1)$ for different values of the linear size $L$ --- see
610: Figs.\ \ref{fig:ThirdIndexSameLine} and \ref{fig:ThirdIndex}.
611: Fig.~\ref{fig:ThirdIndex} shows $\sigma_{e,\|}$ and $\sigma_{e,\perp}$
612: vs. magnetic field in systems with sizes ranging from 100 to 4000.
613: For the larger systems, one can clearly see a ``critical'' range of
614: fields, in which the magnetoresistance is consistent with the scaling
615: form of Eq.\ (\ref{def:delta})
616: with $\delta_\perp = \delta_\| = 1.00 \pm 0.07$.
617: For fields within and below that critical range,
618: the behavior of $\sigma_{e,i}$
619: is independent of $L$, and is also independent of
620: whether its value does or does not saturate at higher fields,
621: as demonstrated in Fig.~\ref{fig:ThirdIndexSameLine}.
622: %\newline
623: %\noindent
624: However, the intercepts of those linear dependencies
625: are different for the different directions:
626: $\sigma_{e,i} \sim (|H|- H_{0,i})^{\delta_i}$, with positive $H_{0,\perp}$
627: and negative $H_{0,\|}$. This is due to the fact that the system always
628: conducts better in the direction parallel to the applied magnetic field,
629: as can be surmised from Eq.\ (\ref{2Dsigmas}).
630: Figs.\ \ref{fig:ThirdIndexSameLine} and \ref{fig:ThirdIndex}
631: bear out the forms hypothesized earlier for $F_i(Z)$---see
632: Eq.\ (\ref{F_i_Z}). In particular, Fig.\ \ref{fig:ThirdIndexSameLine}
633: bears out the expectation that, for small $|Z|$, $F_i(-Z)=-F_i(Z)$,
634: and thus that the scaling functions are smooth at $Z=0$.
635: Figs.\ \ref{fig:ThirdIndexSameLine} and \ref{fig:ThirdIndex} also
636: show that both $\sigma_{e,\|}$ and $\sigma_{e,\perp}$ are
637: independent of $L$ when $|H|\ll L^{t/\nu}$, in agreement with
638: Eqs.\ (\ref{F_i_Z}) and (\ref{rho_e_L}) below.
639: %
640: \begin{figure}%[h]
641: \centerline{\epsfig{file=\figdir fig4.eps,width=3.2in}}
642: \vskip2pc
643: \caption[]{
644: Fraction $P_i\equiv N_{sat}/N_{total}$ of samples which
645: saturate in the direction $i$, plotted versus
646: $S$ volume fraction $\psc$,
647: at fixed $M$ volume fraction $p_M=0.2$, for two different sizes $L$.
648: The vertical line denotes the percolation threshold in
649: an infinite system, as predicted by Eq.\ (\protect\ref{anisotropicPT}).
650: For a finite system, the ``percolation threshold'' is
651: the point where one half of the samples percolate: $P_i=0.5$.
652: Such thresholds are unequal in the directions parallel and
653: perpendicular to ${\bf B}$, but
654: approach the same infinite-size value,
655: ${\psc}_0=0.4$,
656: as $L\rightarrow \infty$.
657: }
658: \label{fig:probabilistic}
659: \end{figure}
660:
661: %%%CHANGE
662: In order to find the form of the scaling functions $F_i(Z)$
663: [defined by Eq.\ (\ref{def:F})] from these numerical results,
664: it is convenient to also
665: invoke the finite-size-scaling hypothesis:
666: \cite{AharonyStauffer92} In a system of finite
667: linear size $L$, when the correlation length $\xi$ satisfies
668: $\xi\gg L$, $|\Delta p|$ should be replaced by $C_1/L^{1/\nu}$
669: in all the expressions of Eq.\ (\ref{def:F}):
670: \begin{equation}
671: \frac{\rho_{e,i}}{\rho_{M,\|}}\cong L^{t/\nu}{\rm sign}\Delta p\,F_i(Z),\;\;\;
672: Z\equiv{\rm sign}\Delta p\,|H|/L^{t/\nu}.\label{rho_e_L}
673: \end{equation}
674: Note that the constant $C_1$ has been absorbed into the definitions of
675: $Z$ and $F_i(Z)$. The sign of $\Delta p$,
676: appearing in Eq.\ (\ref{rho_e_L}), should now be determined from the actual
677: behavior (i.e., percolating vs.\ non-percolating, or saturating
678: vs.\ non-saturating, in the direction $i$)
679: of {\em each particular sample}.
680:
681:
682: The plots of $L^{-t/\nu}\rho_{e,i}/\rho_{M,\|}$ vs.\ $|H|L^{-t/\nu}$,
683: shown in Fig.\ \ref{scaling}, clearly demonstrate that
684: the results obtained for $\rho_{e,i}$, using different values
685: of $H$ and $L$, collapse onto a single curve (for a given $i$)
686: when scaled in
687: accordance with Eq.\ (\ref{rho_e_L}). These figures constitute
688: a quantitative graphical representation of the scaling functions
689: $F_\|(Z)$ and $F_\perp(Z)$ for the special case $p_M=0.5$.
690: Note that these two scaling functions appear to have a similar shape,
691: up to a constant coefficient. That is qualitatively consistent with the
692: results of EMA, which lead to \cite{BergmanPRB2001}
693: $$\frac{F_\perp(Z)}{F_\|(Z)}=
694: \left(1+p_M\over 1-p_M\right)^2.$$
695: In the case of the systems featured in Fig.\ \ref{scaling},
696: where $p_M=0.5$, this ratio should be 9 according to EMA. By contrast, the
697: simulation results plotted in that figure indicate that this
698: ratio is between 4.5 and 5. As stated earlier, such quantitative
699: discrepancies should come as no surprise, in view of the known
700: deficiencies of EMA in the critical region near a percolation threshold.
701:
702: %\vskip2.5pc
703: \begin{figure}[t]
704: \centerline{\epsfig{file=\figdir fig5.eps,width=3.5in}}
705: \vskip1pc
706: \caption[]{
707: $\ln [(H^2+1) \sigma_{e,i}]$ vs. $\ln (H^2+1)$ for systems with
708: $L=100$ and $p_M=0.5$ at the percolation threshold $p_I=p_S$.
709: Averaging over different realizations is performed separately for
710: saturating and non-saturating samples.
711: Open symbols correspond to $\sigma_{sat,\perp}$ and $\sigma_{non-sat,\perp}$
712: and the filled symbols to $\sigma_{sat,\|}$ and $\sigma_{non-sat,\|}$.
713: For values of $H$ within and below the ``critical'' range
714: of fields (as defined in the text), the conductivity in a given direction
715: does not depend on whether or not it saturates at stronger fields $H$.
716: }
717: \label{fig:ThirdIndexSameLine}
718: \end{figure}
719:
720: \section{Discussion}
721: We have demonstrated that the phase diagram of an
722: $M/I/S$ random columnar composite in an in-plane magnetic field exhibits
723: a critical line between saturating and non-saturating regimes
724: of magnetoresistance.
725: The behavior near this critical line was interpreted
726: using an anisotropic percolation model.
727: The critical exponents $t=s$ and $\nu$ were found, numerically,
728: to be the same as in the more conventional 2D isotropic
729: random percolating networks. Also found were the exponents
730: which govern the dependence on magnetic field in the critical
731: region. The existence of scaling behavior in that region was
732: confirmed and the forms of the scaling functions were obtained numerically.
733:
734: \begin{figure}[t]
735: \centerline{\epsfig{file=\figdir fig6.eps,width=3.6in}}
736: \vskip1pc
737: \caption[]{
738: $\ln [(H^2+1) \sigma_{e,i}]$ vs. $\ln (H^2+1)$ for systems of size
739: $L=$ 100 (triangles), 1000 (squares) and 4000 (circles), with
740: $p_M=0.5$, $p_I=p_S=0.25$. Open symbols correspond to $\sigma_{e,\perp}$
741: and the filled symbols to $\sigma_{e,\|}$. The lines are drawn as
742: guides for the eye.
743: At large fields, all systems exhibit
744: either saturating ($H^2 \sigma_i\sim H^2$) or non-saturating
745: ($H^2 \sigma_i\sim 1$) magnetoresistance.
746: In the ``critical'' range of fields,
747: clearly seen for the larger systems, the magnetoresistive behavior is
748: given by Eq.\ (\ref{def:delta}) with
749: $\delta_\perp = \delta_\| = 1.00 \pm 0.07$.
750: Note that, for the sake of visual clarity, we include only those systems
751: in which $\sigma_{e,\perp}$ does not saturate and
752: $\sigma_{e,\|}$ does saturate [cf. Fig.~\ref{fig:ThirdIndexSameLine}].
753: }
754: \label{fig:ThirdIndex}
755: \end{figure}
756:
757: The critical line discussed here could
758: be studied experimentally using a doped semiconductor film as
759: the $M$ constituent,
760: with a random collection of etched perpendicular holes as the
761: $I$ constituent, and a random collection of perpendicular
762: columnar inclusions, made of a high conductivity normal metal,
763: playing the role of $S$.
764: Extremely low temperatures or very clean single crystals
765: would not be required in order to observe this critical line.
766: What would be necessary is a large contrast at each stage of
767: the following chain of inequalities
768: %\begin{equation}
769: $\rho_S\ll\rho_M\ll H^2\rho_M\ll\rho_I$. %\label{inequalities}
770: %\end{equation}
771: If Si-doped GaAs is used as the $M$ host, with a negative charge
772: carrier density of $1.6\times 10^{18}$ cm$^{-3}$ and a mobility
773: $\mu=2500$ cm$^2/$V\,s at a temperature of 90\,K,
774: as in the experiment described in Ref.\
775: \onlinecite{TornowWeissKlitzingEberlBergStrelPRL96},
776: then a magnetic field of 40\,Tesla
777: would result in $H=-10$. Such a material would have an Ohmic
778: resistivity of $1.6\times 10^{-3}\,\Omega$\,cm, about 1000
779: times greater than that of Cu. Thus, using Cu for the $S$
780: inclusions and etched holes for the $I$ inclusions, all the above inequalities
781: could be satisfied without difficulty.
782:
783: This is apparently the first experimentally accessible and technologically
784: promising system in which anisotropic percolation is relevant.
785: Our numerical results for the exponents $\nu$, $t$, $s$, and $\delta$
786: are consistent with the assumption that this problem
787: belongs to the same universality class as the usual 2D isotropic
788: percolation problem, and confirm that $\nu$ is independent of anisotropy.
789:
790:
791: %
792: \begin{figure}[t]
793: \centerline{
794: %\rotatebox{-90}{
795: \epsfig{file=\figdir fig7.eps,width=4.25in}
796: %}
797: }
798: \vskip-1pc
799: \caption[]{Plots of $F_i(Z)\equiv
800: {\rm sign}\Delta p\,L^{-t/\nu}\rho_{e,i}/\rho_{M,\|}$
801: vs.\ $Z\equiv{\rm sign}\Delta p\,|H|L^{-t/\nu}$ for networks with $p_M=0.5$ and
802: various large but finite values of $H$ and $L$. The top frame shows
803: $F_i(Z)$ in the non-saturating regime, when $Z>0$, while the
804: bottom frame shows $F_i(Z)$ in the saturating regime, when $Z<0$.
805: The shapes of the points correspond to the value of $L$, as in
806: Fig.\ \ref{fig:ThirdIndex}. Open symbols correspond to $F_\perp(Z)$,
807: while filled symbols correspond to $F_\|(Z)$. The horizontal lines surrounding
808: the symbols are parts of error bars. Inset: expanded
809: view of $F_i(Z)$ for small values of $|Z|$ and both signs of $Z$,
810: showing the smooth linear behavior of $F_i(Z)$.
811: }
812: \label{scaling}
813: \end{figure}
814:
815: %\vskip1pc
816: %%%CHANGE
817: %\noindent
818:
819: \acknowledgements
820:
821: We are grateful to Y. Kantor for providing us with the idea
822: regarding the $L$ dependence of $P_i$ and the appropriate references.
823: This work was supported in part by NSF Grant DMR97-31511, and by
824: grants from the
825: US-Israel Binational Science Foundation and
826: the Israel Science Foundation.
827:
828: \vspace{-0.5 cm}
829: \begin{thebibliography}{5}
830: \vspace{-1.6 cm}
831:
832: \bibitem{BS1} D.\ J.\ Bergman and Y.\ M.\ Strelniker, Phys.\ Rev.\ B
833: {\bf 60}, 13016 (1999).
834:
835: \bibitem{BergStrelPRB94} D.J. Bergman and Y.M. Strelniker,
836: Phys.\ Rev.\ B {\bf 49}, 16256 (1994); Phys.\ Rev.\ B {\bf 51},
837: 13845 (1995); Phys.\ Rev.\ B{\bf 59}, 2180 (1999).
838:
839: \bibitem{TornowWeissKlitzingEberlBergStrelPRL96} M. Tornow, D. Weiss,
840: K. von Klitzing, K. Eberl, D.J.
841: Bergman, and Y.M. Strelniker,
842: Phys.\ Rev.\ Lett.\ 147 (1996).
843:
844: \bibitem{SSP}
845: D.\ J.\ Bergman and D. Stroud, Sol. St. Phys. {\bf 46}, 178 (1992).
846:
847: \bibitem{kogut79}
848: P.\ M.\ Kogut and J.\ P.\ Straley, J. Phys. C: Solid State Phys.,
849: {\bf 12}, 1 (1979).
850:
851: \bibitem{SarBergStrelPRB93} A. K. Sarychev, D. J. Bergman,
852: and Y. M. Strelniker, Phys.\ Rev.\ B {\bf 48}, 3145 (1993).
853:
854: \bibitem{sykes63}
855: M.\ F.\ Sykes and J.\ W.\ Essam, Phys.\ Rev.\ Lett. {\bf 10}, 3 (1963).
856: % proof that $$p_\|+p_\perp = 1$$
857:
858: \bibitem{BergPRLsub}
859: D.J. Bergman, Phys.\ Rev.\ B {\bf 62}(Brief Reports), 13820 (2000).
860:
861: \bibitem{BergmanPRB2001}
862: D.J. Bergman, Phys.\ Rev.\ B {\bf 64}, 024412-1 (2001).
863:
864: \bibitem{note1}
865: For example, in the limit $p_M \rightarrow 1$, the RRN version of
866: the composite at $H \rightarrow \infty$
867: becomes a set of one-dimensional parallel ``wires'' composed of
868: $\spar$ and $\ssc$ bonds. These wires will be broken by
869: $\sins$ bonds into units with average length $a/\pins$. These longitudinal
870: wires will be connected
871: by an average of one transverse $\ssc$ bond of length $a$ to neighboring
872: wires (here $a$ is the size of a single bond).
873: As a critical point is approached, since the $\sigma_I$ and $\sigma_S$ bonds appear
874: with {\em equal probabilities}, the size of a critical cluster will
875: grow in both directions in a similar fashion.
876:
877: \bibitem{sarychev79}
878: A.\ K.\ Sarychev and A.\ P.\ Vinogradoff,
879: J. Phys. C: Solid St. Phys., {\bf 12}, L681 (1979).
880: % anisotropy remains finite in 2D
881:
882: \bibitem{anisotropic}
883: H.\ Nakanishi, P.\ J.\ Reynolds and S.\ Redner, J. Phys. A: Math. Gen.,
884: {\bf 14}, 855 (1981);
885: % the anisotropic percolation is governed
886: % by the fixed point of an isotropic system;
887:
888: \bibitem{nijs}
889: M.\ P.\ M.\ den Nijs, J. Phys. A: Math. Gen., {\bf 12}, 1857 (1979).
890: %analytical proof \nu=4/3
891:
892: \bibitem{derrida}
893: B.\ Derrida and L.\ De Seze, J. Physique {\bf 43}, 475 (1982).
894: %numeric confirmation \nu=4/3
895:
896: \bibitem{anisotropic_nu}
897: P.\ M.\ Castro de Oliveria, Phys.\ Rev.\ B {\bf 25}, 2034 (1982).
898: % Refines the work by Nakanishi, yet gives very bad estimate \nu=1.099.
899:
900: \bibitem{shapiro}
901: D.\ Arovas, R.\ N.\ Bhatt and B.\ Shapiro, Phys. Rev. {\bf B 28}, 1433 (1983).
902:
903: \bibitem{t2D}
904: R.\ Fogelholm, J. Phys. C: Solid St. Phys., {\bf 13}, L571 (1980);
905: J.-M.\ Normand, H.\ J.\ Herrmann, and M. Hajjar,
906: J. Stat. Phys., {\bf 52}, 441 (1988).
907:
908: \bibitem{YDelta}
909: D.\ J.\ Frank and C.\ J.\ Lobb, Phys.\ Rev.\ B {\bf 37}, 302 (1988).
910:
911: \bibitem{t2Dnew}
912: P.\ Grassberger, Physica A {\bf 262}, 251 (1999).
913: %NB: they report t=1.310 !
914:
915: \bibitem{FengHalperinSenPRB87}
916: S. Feng, B.I. Halperin, and P.N. Sen, Phys.\ Rev.\ B {\bf 35}, 197 (1987).
917:
918: \bibitem{AharonyStauffer92} A. Aharony and D. Stauffer, {\em
919: Introduction to Percolation Theory}, 2nd edition, Taylor and
920: Francis, London, 1992.
921:
922: \bibitem{PercThresh}
923: P.\ J.\ Reynolds, H.\ E.\ Stanley and W.\ Klein,
924: Phys.\ Rev.\ B {\bf 21}, 1223 (1980);
925: M.\ E.\ Levinshtein, B.\ I.\ Shklovskii, M.\ S.\ Shur, and A.\ L.\ Efros,
926: ZhETF {\bf 69}, 386 (1975) [Soviet Physics JETP {\bf 42}, 197 (1976)].
927:
928: \bibitem{SpanningProbability}
929: J.\ L.\ Cardy, J.\ Phys. A {\bf 25}, L201 (1992);
930: J.\ -P.\ Hovi and A.\ Aharony, Phys.\ Rev.\ {\bf E 53}, 235 (1996);
931: R.\ P.\ Langlands, C.\ Pichet, Ph.\ Pouliot and Y.\ Saint-Aubin,
932: J.\ Stat.\ Phys. {\bf 67}, 553 (1992).
933:
934: \bibitem{note:CPP}
935: Indeed, application of the Y-$\Delta$ algorithm\cite{YDelta} to such a system
936: leads to generation of intermediary bonds with rapidly growing
937: exponents, both positive and negative. It is important to keep track of such
938: very large and very small numbers, especially in a system that is
939: close to a percolation threshold:
940: a single replacement of a very weak bond by a bond of zero conductance
941: changes the topology of the network, possibly
942: driving it from percolating to non-percolating.
943:
944: \end{thebibliography}
945:
946:
947: \end{document}
948: