1: %% LyX 1.4.3 created this file. For more info, see http://www.lyx.org/.
2: %% Do not edit unless you really know what you are doing.
3: \documentclass[11pt,english,a4paper]{article}
4: \usepackage{times}
5: \usepackage[T1]{fontenc}
6: \usepackage[latin1]{inputenc}
7: \usepackage{geometry}
8: \geometry{verbose,letterpaper,tmargin=2.3cm,bmargin=5.94cm,lmargin=2cm,rmargin=7.59cm}
9: \usepackage{graphicx}
10:
11: \makeatletter
12:
13: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% LyX specific LaTeX commands.
14: \newcommand{\noun}[1]{\textsc{#1}}
15: %% Bold symbol macro for standard LaTeX users
16: \providecommand{\boldsymbol}[1]{\mbox{\boldmath $#1$}}
17:
18:
19: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Textclass specific LaTeX commands.
20: \newenvironment{lyxlist}[1]
21: {\begin{list}{}
22: {\settowidth{\labelwidth}{#1}
23: \setlength{\leftmargin}{\labelwidth}
24: \addtolength{\leftmargin}{\labelsep}
25: \renewcommand{\makelabel}[1]{##1\hfil}}}
26: {\end{list}}
27:
28: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% User specified LaTeX commands.
29: %% LyX 1.3 created this file. For more info, see http://www.lyx.org/.
30: %% Do not edit unless you really know what you are doing.
31:
32:
33:
34:
35:
36: \makeatletter
37:
38: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% LyX specific LaTeX commands.
39: %% Bold symbol macro for standard LaTeX users
40:
41:
42:
43: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% User specified LaTeX commands.
44: \setlength{\textwidth}{15cm}
45:
46:
47: \makeatother
48:
49: \usepackage{babel}
50: \makeatother
51: \begin{document}
52: \begin{flushleft}\textbf{\noun{\LARGE burst statistics as a criterion
53: for imminent failure}}\par\end{flushleft}{\LARGE \par}
54:
55: \textbf{\Large \vskip.2in}{\Large \par}
56:
57: \noindent Srutarshi Pradhan$^{1}$, Alex Hansen$²$ and Per C. Hemmer$³$
58:
59: \vskip.1in
60:
61: \noindent \emph{Department of Physics, Norwegian University of Science
62: and Technology, N--7491 Trondheim, Norway, $¹$pradhan.srutarshi@ntnu.no,
63: $²$alex.hansen@ntnu.no, $³$per.hemmer@ntnu.no}
64:
65: \vskip.2in
66:
67: \begin{lyxlist}{00.00.0000}
68: \item [{Abstract:}] \noindent {\small The distribution of the magnitudes
69: of damage avalanches during a failure process typically follows a
70: power law. When these avalanches are recorded close to the point at
71: which the system fails catastrophically, we find that the power law
72: has an exponent which differs from the one characterizing the size
73: distribution of all avalanches. We demonstrate this analytically for
74: bundles of many fibers with statistically distributed breakdown thresholds
75: for the individual fibers. In this case the magnitude distribution
76: $D(\Delta)$ for the avalanche size $\Delta$ follows a power law
77: $\Delta^{-\xi}$ with $\xi=3/2$ near complete failure, and $\xi=5/2$
78: elsewhere. We also study a network of electric fuses, and find numerically
79: an exponent $2.0$ near breakdown, and $3.0$ elsewhere. We propose
80: that this crossover in the size distribution may be used as a signal
81: for imminent system failure. }{\small \par}
82: \end{lyxlist}
83: \vskip0.1in
84:
85: \begin{flushleft}Key words: Failure, Fiber bundle model, Fuse model,
86: Burst statistics, Crossover\par\end{flushleft}
87:
88:
89: \section{\noun{introduction}}
90:
91: Catastrophic failures \cite{Herrmann,Chakrabarti,Sornette,Sahimi,Pratip}
92: are abundant in nature: earthquakes, landslides, mine-collapses, snow-avalanches
93: etc. are well-known examples. A sudden catastrophic failure is a curse
94: to human society due to the devastation it causes in terms of properties
95: and human lives. Therefore a fundamental challenge is to detect reliable
96: precursors of such catastrophic events. This is also an important
97: issue in strength considerations of materials as well as in construction
98: engineering.
99:
100: During the failure process of composite materials under external stress,
101: avalanches of different size are produced where an avalanche consists
102: of simultaneous rupture of several elements. Such avalanches closely
103: correspond to the bursts of acoustic emissions \cite{AE-1,AE-2} which
104: are observed experimentally during the failure process of several
105: materials. If one counts all avalanches till the complete failure,
106: their size distribution is typically a power law. However we observed
107: recently that if one records avalanches not for the entire failure
108: process, but within a finite interval, a clear crossover behavior
109: \cite{Pradhan1,Pradhan2} is seen between two power laws, with a large
110: exponent when the system is far away from the failure point and a
111: much smaller exponent for avalanches in the vicinity of complete failure.
112: Therefore such crossover behavior in the burst statistics can be taken
113: as a criterion for imminent failure. In this report we discuss the
114: crossover behavior in two different fracture-breakdown models- \emph{fiber
115: bundle models} {[}10-18] which describe the failure of composite material
116: under external load and \emph{fuse models} \cite{Herrmann,Hansen2,Zapperi}
117: which demonstrate the breakdown of electrical networks. An analytic
118: derivation of the crossover behavior is given for fiber bundle models
119: and numerical study confirms similar behavior in a fuse model. Thus
120: we claim that such crossover behavior can be used as a signal of imminent
121: failure. A recent observation \cite{kawamura06} of the existence
122: of a crossover behavior in the magnitude distribution of earthquakes
123: within Japan has strengthened this claim.
124:
125:
126: \section{\noun{fiber bundle model}}
127:
128: Our fiber bundle model consists of $N$ elastic and parallel fibers,
129: clamped at both ends, with statistically distributed thresholds for
130: breakdown of individual fibers (Figure\ 1). The individual thresholds
131: $x_{i}$ are assumed to be independent random variables drawn from
132: the same cumulative distribution function $P(x)$ and a corresponding
133: density function $p(x)$: \begin{equation}
134: \mbox{Prob}(x_{i}<x)=P(x)=\int_{0}^{x}p(u)\; du.\label{1}\end{equation}
135:
136:
137: \begin{center}\includegraphics[width=2.2in,height=1.8in]{fbm-sp}\par\end{center}
138:
139: \noindent \emph{\small Figure 1}{\small . A fiber bundle of $N$ parallel
140: fibers clamped at both ends. The externally applied force $F$ corresponds
141: to a stretching by an amount $\delta$.}\\
142:
143:
144: Whenever a fiber experiences a force equal to or greater than strength
145: threshold $x_{i}$, it breaks immediately and does not contribute
146: to the strength of the bundle thereafter. The maximal load the bundle
147: can resist before complete breakdown is called the \textit{critical}
148: load and its value depends upon the probability distribution of the
149: thresholds. Two popular examples of threshold distributions are the
150: uniform distribution \begin{equation}
151: P(x)=\left\{ \begin{array}{cl}
152: x/x_{r} & \mbox{ for }0\leq x\leq x_{r}\\
153: 1 & \mbox{ for }x>x_{r},\end{array}\right.\label{uniform}\end{equation}
154: and the Weibull distribution \begin{equation}
155: P(x)=1-\exp(-(x/x_{r})^{\kappa}).\label{Weibull}\end{equation}
156: Here $x_{r}$ is a reference threshold, and the dimensionless number
157: $\kappa$ is the Weibull index (Figure\ 2.)
158:
159: \begin{center}\includegraphics[width=2in,height=1.8in]{rev-fig2A}\hskip.2in\includegraphics[width=2in,height=1.8in]{rev-fig2B}\par\end{center}
160:
161: \vskip.2in
162:
163: \noindent \emph{\small Figure 2}{\small . The uniform distribution
164: (A) and the Weibull distribution (B) with $\kappa=5$ (solid line)
165: and $\kappa=10$ (dotted line) .}\\
166:
167:
168: Fiber bundle models differ in the mechanism for how the extra stress
169: caused by a fiber failure is redistributed among the unbroken fibers.
170: The simplest models are the equal-load-sharing models, in which the
171: load previously carried by a failed fiber is shared equally by all
172: the remaining intact fibers in the system. In the present article
173: we study the statistics of burst avalanches for equal-load-sharing
174: models.
175:
176:
177: \subsection{Burst statistics}
178:
179: The \textit{burst distribution} $D(\Delta)$ is defined as the expected
180: number of bursts in which $\Delta$ fibers break simultaneously when
181: the bundle is stretched steadily until complete breakdown. Hemmer
182: and Hansen performed a detail statistical analysis \cite{Hemmer}
183: to find the burst distribution for this quasi-static situation. For
184: a bundle of many fibers they calculated the probability of a burst
185: of size $\Delta$ starting at fiber $k$ with threshold value $x_{k}$
186: as
187:
188: \begin{equation}
189: {\displaystyle \frac{\Delta^{\Delta-1}}{\Delta!}\left[1-\frac{x_{k}p(x_{k})}{Q(x_{k})}\right]\left[\frac{x_{k}p(x_{k})}{Q(x_{k})}\right]^{\Delta-1}\times\exp\left[-\Delta\frac{x_{k}p(x_{k})}{Q(x_{k})}\right],}\label{C}\end{equation}
190: where $Q(x)=1-P(x)$. Since a burst of size $\Delta$ can occur at
191: any point before complete breakdown, the above expression has to be
192: integrated over all possible values of $x_{k}$, i.e, from $0$ to
193: $x_{c}$ where $x_{c}$ is the maximum amount of stretching beyond
194: which the bundle collapses completely. Therefore the burst distribution
195: is given by \cite{Hemmer}
196:
197: \begin{flushleft}\begin{equation}
198: \frac{D(\Delta)}{N}={\displaystyle \frac{\Delta^{\Delta-1}e^{-\Delta}}{\Delta!}\int_{0}^{x_{c}}p(x)r(x)[1-r(x)]^{\Delta-1}\exp\left[\Delta\, r(x)\right]dx,}\label{D}\end{equation}
199: where \begin{equation}
200: r(x)=1-\frac{x\, p(x)}{Q(x)}=\frac{1}{Q(x)}\;\frac{d}{dx}\left[x\, Q(x)\right].\label{eq:8}\end{equation}
201: \par\end{flushleft}
202:
203: The integration yields the asymptotic behavior \begin{equation}
204: D(\Delta)/N\propto\Delta^{-\frac{5}{2}},\label{asymp}\end{equation}
205: which is universal under mild restrictions on the threshold distributions
206: \cite{Kloster}. As a check, we have done simulation experiments for
207: different threshold distributions. Figure 3 show results for the uniform
208: threshold distribution (\ref{uniform}) and the Weibull distribution
209: (\ref{Weibull}) with index $\kappa=5$.
210:
211: \begin{center}\includegraphics[width=2.2in,height=1.8in]{rev-fig3A}\hskip.2in\includegraphics[width=2.2in,height=1.8in]{rev-fig3B}\par\end{center}
212:
213: \noindent \emph{\small Figure 3}{\small . The burst distribution $D(\Delta)/N$
214: for the uniform distribution (A) and the Weibull distribution with
215: index 5 (B). The dotted lines represent the power law with exponent
216: $-5/2$. Both figures are based on $20000$ samples of bundles each
217: with $N=10^{6}$ fibers.}\\
218:
219:
220:
221: \subsection{Crossover behavior near failure point}
222:
223: \begin{center}\includegraphics[width=2.2in,height=1.8in]{rev-fig6}\par\end{center}
224:
225: \noindent \emph{\small Figure 4}{\small . The distribution of bursts
226: for thresholds uniformly distributed in an interval $(x_{0},x_{c})$,
227: with $x_{0}=0$ and with $x_{0}=0.9x_{c}$. The figure is based on
228: 50 000 samples, each with $N=10^{6}$ fibers.}{\small \par}
229:
230: \vskip.1in
231:
232: When all the bursts are recorded for the entire failure process, we
233: have seen that the burst distribution $D(\Delta)$ follows the asymptotic
234: power law $D\propto\Delta^{-5/2}$. If we just sample bursts that
235: occur near criticality, a different behavior is seen. As an illustration
236: we consider the uniform threshold distribution, and compare the complete
237: burst distribution with what one gets when one samples merely burst
238: from breaking fibers in the threshold interval $(0.9x_{c},x_{c})$.
239: Figure\ 4 shows clearly that in the latter case a different power
240: law is seen.
241:
242: From Eq. \ref{eq:8} we see that $r(x)$ vanishes at the point $x_{c}$.
243: If we have a situation in which the weakest fiber has its threshold
244: $x_{0}$ just a little below the critical value $x_{c}$, the contribution
245: to the integral in the expression (\ref{D}) for the burst distribution
246: will come from a small neighborhood of $x_{c}$. Since $r(x)$ vanishes
247: at $x_{c}${\small ,} it is small here, and we may in this narrow
248: interval approximate the $\Delta$-dependent factors in (\ref{D})
249: as follows
250:
251: \begin{eqnarray}
252: (1-r)^{\Delta}\, e^{\Delta\, r} & = & \exp\left[\Delta(\ln(1-r)+r)\right]\nonumber \\
253: & = & \exp[-\Delta(r^{2}/2+{\mathcal{O}}(r^{3}))]\approx\exp\left[-\Delta r(x)^{2}/2\right]\label{D3}\end{eqnarray}
254:
255:
256: \noindent We also have \begin{equation}
257: r(x)\approx r'(x_{c})(x-x_{c}).\end{equation}
258:
259:
260:
261:
262: \noindent Inserting everything into Eq.\ (\ref{D}), we obtain to
263: dominating order \begin{eqnarray}
264: \frac{D(\Delta)}{N} & = & \frac{\Delta^{\Delta-1}\, e^{-\Delta}}{\Delta!}\int_{x_{0}}^{x_{c}}p(x_{c})\; r'(x_{c})(x-x_{c})e^{-\Delta\, r'(x_{c})^{2}(x-x_{c})^{2}/2}\; dx\nonumber \\
265: & = & \frac{\Delta^{\Delta-2}\, e^{-\Delta}p(x_{c})}{\left|r'(x_{c})\right|\Delta!}\left[e^{-\Delta\, r'(x_{c})^{2}(x-x_{c})^{2}/2}\right]_{x_{0}}^{x_{c}}\nonumber \\
266: & = & \frac{\Delta^{\Delta-2}\, e^{-\Delta}}{\Delta!}\frac{p(x_{c})}{\left|r'(x_{c})\right|}\left[1-e^{-\Delta/\Delta_{c}}\right],\label{D3}\end{eqnarray}
267: with \begin{equation}
268: \Delta_{c}=\frac{2}{r'(x_{c})^{2}(x_{c}-x_{0})^{2}}.\label{Dc}\end{equation}
269:
270:
271: By use of the Stirling approximation $\Delta!\simeq\Delta^{\Delta}e^{-\Delta}\sqrt{2\pi\Delta}$,
272: the burst distribution (\ref{D3}) may be written as\begin{equation}
273: \frac{D(\Delta)}{N}=C\Delta^{-5/2}\left(1-e^{-\Delta/\Delta_{c}}\right),\label{D2}\end{equation}
274: with a nonzero constant \begin{equation}
275: C=(2\pi)^{-1/2}p(x_{c})/\left|r'(x_{c})\right|.\end{equation}
276: We can see from (\ref{D2}) that there is a crossover at a burst
277: length around $\Delta_{c}$:\begin{equation}
278: \frac{D(\Delta)}{N}\propto\left\{ \begin{array}{cl}
279: \Delta^{-3/2} & \mbox{ for }\Delta\ll\Delta_{c}\\
280: \Delta^{-5/2} & \mbox{ for }\Delta\gg\Delta_{c}\end{array}\right.\end{equation}
281:
282:
283: We have thus shown the existence of a crossover from the generic asymptotic
284: behavior $D\propto\Delta^{-5/2}$ to the power law $D\propto\Delta^{-3/2}$
285: near criticality, i.e., near global breakdown. The crossover is a
286: universal phenomenon, independent of the threshold distribution $p(x)$.
287:
288: \begin{center}\includegraphics[width=2.2in,height=1.8in]{rev-fig7}\hskip.2in\includegraphics[width=2.2in,height=1.8in]{rev-fig7B}\par\end{center}
289:
290: \noindent \emph{\small Figure 5}{\small . The distribution of bursts
291: for the uniform threshold distribution (left) with $x_{0}=0.80x_{\textrm{c}}$
292: and for a Weibull distribution (right) with $x_{0}=1$ (square) and
293: $x_{0}=1.7$ (circle). Both the figures are based on $50000$ samples
294: with $N=10^{6}$ fibers each. The straight lines represent two different
295: power laws, and the arrows locate the crossover points $\Delta_{c}\simeq12.5$}
296: {\small and $\Delta_{c}\simeq14.6$,} respectively.
297:
298: \vskip.1in
299:
300: For the uniform distribution $\Delta_{c}=(1-x_{0}/x_{c})^{-2}/2$,
301: so for $x_{0}=0.8\, x_{\textrm{c}}$, we have $\Delta_{c}=12.5$.
302: For the Weibull distribution $P(x)=1-\exp(-(x-1)^{10})$, where $1\leq x\leq\infty$,
303: we get $x_{c}=1.72858$ and for $x_{0}=1.7$, the crossover point
304: will be at $\Delta_{c}\simeq14.6$. Such crossover is clearly observed
305: (Figure 5) near the expected values $\Delta=\Delta_{c}=12.5$ and
306: $\Delta=\Delta_{c}=14.6$, respectively, for the above distributions.
307:
308: \begin{center}\includegraphics[width=2.2in,height=1.8in]{rev-fig8}\par\end{center}
309:
310: \noindent \emph{\small Figure 6}{\small . The distribution of bursts
311: for the uniform threshold distribution for a single fiber bundle with
312: $N=10^{7}$ fibers. Results with $x_{0}=0$, i.e., when all avalanches
313: are recorded, are shown as squares and data for avalanches near the
314: critical point ($x_{0}=0.9x_{c}$) are shown by circles. }{\small \par}
315:
316: \vskip.1in
317:
318: The simulation results we have shown so far are based on \textit{averaging}
319: over a large number of samples. For applications it is important that
320: crossover signal can be seen also in a single sample. We show in Figure\ 6
321: that equally clear crossover behavior is seen in a \textit{single}
322: fiber bundle when $N$ is large enough. Also, as a practical tool
323: one must sample finite intervals ($x_{i}$, $x_{f}$) during the fracture
324: process. The crossover will be observed when the interval is close
325: to the failure point \cite{Pradhan2}.
326:
327:
328: \section{\noun{burst statistics in the fuse model}}
329:
330: \begin{center}\includegraphics[width=2.2in,height=1.8in]{fusenet}\hskip.6in
331: \includegraphics[width=2.2in,height=1.8in]{fuse-cross-pre}\par\end{center}
332:
333: \begin{flushleft}\emph{\footnotesize Figure 7}{\footnotesize .} {\footnotesize A
334: fuse model and the burst distribution: System size is $100\times100$
335: and averages are taken for $300$ samples. On the average, catastrophic
336: failure sets in after $2097$ fuses have blown. The circles denote
337: the burst distribution measured throughout the entire breakdown process.
338: The squares denote the burst distribution based on bursts appearing
339: after the first $1000$ fuses have blown. The triangles denote the
340: burst distribution after $2090$ fuses have blown. The two straight
341: lines indicate power laws with exponents $\xi=3$ and $\xi=2$, respectively. }\par\end{flushleft}{\footnotesize \par}
342:
343: \vskip.1in
344:
345: To test the crossover phenomenon in a more complex situation than
346: for fiber bundles, we have studied burst distributions in the fuse
347: model \cite{Herrmann}. It consists of a lattice in which each bond
348: is a fuse, i.e., an ohmic resistor as long as the electric current
349: it carries is below a threshold value. If the threshold is exceeded,
350: the fuse burns out irreversibly. The threshold $t$ of each bond is
351: drawn from an uncorrelated distribution $p(t)$. The lattice is placed
352: between electrical bus bars and an increasing current is passed through
353: it. The lattice is a two-dimensional square one placed at $45{}^{\circ}$
354: with regards to the bus bars, and the Kirchhoff equations are solved
355: numerically at each node assuming that all fuses have the same resistance.
356:
357: When one records all the bursts, the distribution follows a power
358: law $D(\Delta)\propto\Delta^{-\xi}$ with $\xi=3$, which is consistent
359: with the value reported in recent studies \cite{Hansen2,Zapperi}.
360: We show the histogram in Figure 7. With a system size of $100\times100$,
361: $2097$ fuses blow on the average before catastrophic failure sets
362: in. When measuring the burst distribution only after the first $2090$
363: fuses have blown, a different power law is found, this time with $\xi=2$.
364: After $1000$ blown fuses, on the other hand, $\xi$ remains the same
365: as for the histogram recording the entire failure process (Figure
366: 7).
367:
368: In Figure 8, we show the power dissipation $E$ in the network as
369: a function of the number of blown fuses and as a function of the total
370: current. The dissipation is given as the product of the voltage drop
371: across the network $V$ times the total current that flows through
372: it. The breakdown process starts by following the lower curve, and
373: follows the upper curve returning to the origin. It is interesting
374: to note the linearity of the unstable branch of this curve. In Figure
375: 9, we record the avalanche distribution for power dissipation, $D_{d}(\Delta)$.
376: Recording, as before, the avalanche distribution throughout the entire
377: process as well as recording only close to the point at which the
378: system catastrophically fails, result in two power laws, with exponents
379: {\small $\xi=2.7$} and {\small $\xi=1.9$,} respectively. It is interesting
380: to note that in this case there is not a difference of unity between
381: the two exponents. The power dissipation in the fuse model corresponds
382: to the stored elastic energy in a network of elastic elements. Hence,
383: the power dissipation avalanche histogram would in the mechanical
384: system correspond to the released energy. Such a mechanical system
385: could serve as a simple model for earthquakes.
386:
387: \vskip.3in
388:
389: \begin{center}\includegraphics[width=2.2in,height=1.8in]{fig10} \hskip.4in
390: \includegraphics[width=2.2in,height=1.8in]{fig11}\par\end{center}
391:
392: \begin{flushleft}\emph{\small Figure 8}{\small . Power dissipation
393: $E$ as a function of the number of broken bonds (left) and as a function
394: of the total current $I$ flowing in the fuse model (right). }\par\end{flushleft}{\small \par}
395:
396: \vskip.1in
397:
398: The Gutenberg-Richter law \cite{Herrmann,Chakrabarti} relating the
399: frequency of earthquakes with their magnitude is essentially a measure
400: of the elastic energy released in the earth's crust, as the magnitude
401: of an earthquake is the logarithm of the elastic energy released.
402: Hence, the power dissipation avalanche histogram $D_{d}(\Delta)$
403: in the fuse model corresponds to the quantity that the Gutenberg-Richter
404: law addresses in seismology. Furthermore, the power law character
405: of $D_{d}(\Delta)$ is consistent with the form of the Gutenberg-Richter
406: law. It is then intriguing that there is a change in exponent {\small $\xi$}
407: also for this quantity when failure is imminent.
408:
409: \vskip.3in
410:
411: \begin{center}\includegraphics[width=2.2in,height=2in]{fig12}\par\end{center}
412:
413: \noindent \emph{\small Figure 9}{\small . The power dissipation avalanche
414: histogram} $D_{d}(\Delta)$ {\small for the fuse model. The slopes
415: of the two straight lines are $-2.7$ and $-1.9$, respectively. The
416: circles show the histogram of avalanches recorded through the entire
417: process, whereas the squares show the histogram recorded after $2090$
418: fuses have blown. }{\small \par}
419:
420:
421: \section{\noun{concluding remarks}}
422:
423: Establishing a signature of imminent failure is the principal objective
424: in our study of different breakdown phenomena. The same goal is of
425: course central to earthquake prediction scheme. As in different failure
426: situations, \emph{bursts} can be recorded from outside -without disturbing
427: the ongoing failure process, burst statistics are much easily available
428: and also contain reliable information of the failure process. Therefore,
429: any signature in burst statistics that can warn of imminent system
430: failure would be very useful in the sense of wide scope of applicability.
431: The crossover behavior in burst distributions, we found in the fiber
432: bundle models, is such a signature which signals that catastrophic
433: failure is imminent. Similar crossover behavior is also seen in the
434: burst distribution and energy distributions of the fuse model. Most
435: important is that this crossover signal does not hinge on observing
436: rare events and is seen also in a single system. Therefore, such signature
437: has a strong potential to be used as useful detection tool. It should
438: be mentioned that most recently, Kawamura \cite{kawamura06} has observed
439: a change in exponent values of the local magnitude distributions of
440: earthquakes in Japan, before the onset of a mainshock (Figure 10).
441: This observation has definitely strengthened our claim of using crossover
442: signals in burst statistics as a criterion for imminent failure.
443:
444: \begin{center}\includegraphics[width=3in,height=2.7in]{kawamura-cross}\par\end{center}
445:
446: \noindent \emph{\small Figure 10}{\small . Crossover signature in
447: the local magnitude distributions of earthquakes in Japan . The exponent
448: of the distribution during $100$ days before a mainshock is about
449: $0.60$, much smaller than the average value $0.88$} {\small \cite{kawamura06}}. {\small }{\small \par}
450:
451: \vskip.2in
452:
453: \begin{flushleft}\textbf{\noun{\Large Acknowledgment}}\par\end{flushleft}{\Large \par}
454:
455: \vskip.2in
456:
457: S. P. thanks the Research Council of Norway (NFR) for financial support
458: through Grant No. 166720/V30.
459:
460: \begin{thebibliography}{10}
461: \bibitem{Herrmann}\noun{H.\ J}.\ Herrmann and S.\ Roux, eds.\ {}\emph{Statistical
462: Models for the Fracture of Disordered Media\/{}} (Elsevier, Amsterdam,
463: 1990).
464:
465: \bibitem{Chakrabarti}B.\ K. Chakrabarti and L.\ G.\ Benguigui
466: \emph{Statistical Physics and Breakdown in Disordered Systems\/{}}
467: (Oxford University Press, Oxford, 1997).
468:
469: \bibitem{Sornette}D. Sornette, \emph{Critical Phenomena in Natural
470: Sciences}, Springer-Verlag, Berlin (2000).
471:
472: \bibitem{Sahimi}M. Sahimi, \emph{Heterogeneous Materials II: Nonlinear
473: and Breakdown Properties}, Springer-Verlag, Berlin (2003).
474:
475: \bibitem{Pratip}P. Bhattacharyya and B. K. Chakrabarti (Eds), \emph{Modelling
476: Critical and Catastrophic Phenomena in Geoscience}, Springer, Berlin
477: (2006).
478:
479: \bibitem{AE-1}A. Petri, G. Paparo, A. Vespignani, A. Alippi, and
480: M. Costantini, Phys. Rev. Lett \textbf{73} 3423 (1994).
481:
482: \bibitem{AE-2}A. Garcimartin, A. Guarino, L. Bellon, and S. Ciliberto,
483: Phys. Rev. Lett. \textbf{79} 3202 (1997).
484:
485: \bibitem{Pradhan1}S.\ Pradhan, A.\ Hansen, and P.\ C.\ Hemmer,
486: Phys.\ Rev.\ Lett. \textbf{95}, 125501 (2005).
487:
488: \bibitem{Pradhan2}S.\ Pradhan, A.\ Hansen, and P.\ C.\ Hemmer,
489: Phys.\ Rev.\ E \textbf{74}, 026106 (2006).
490:
491: \bibitem{Peirce}F.\ T.\ Peirce, J.\ Text.\ Ind.\ {}\textbf{17},
492: 355 (1926).
493:
494: \bibitem{Daniels}H.\ E.\ Daniels, Proc.\ Roy.\ Soc.\ London
495: \textbf{A183}, 405 (1945).
496:
497: \bibitem{Smith}R.\ L.\ Smith, Ann.\ Prob.\ {}\textbf{10}, 137
498: (1982).
499:
500: \bibitem{Phoenix}S.\ L.\ Phoenix and R.\ L.\ Smith, Int.\ J.\ Sol.\ Struct.\ {}\textbf{19},
501: 479 (1983).
502:
503: \bibitem{Hemmer}P.\ C.\ Hemmer and A.\ Hansen, ASME J.\ Appl.\ Mech.\ {}\textbf{59},
504: 909 (1992).
505:
506: \bibitem{Kloster}M. Kloster, A. Hansen, and P. C. Hemmer, Phys. Rev.
507: E \textbf{56}, 2615 (1997).
508:
509: \bibitem{Pradhan5}S.\ Pradhan, P.\ Bhattacharyya, and B.\ K.\ Chakrabarti,
510: Phys.\ Rev.\ E \textbf{66} 016116 (2002).
511:
512: \bibitem{Pradhan6}P.\ Bhattacharyya, S.\ Pradhan, and B.\ K.\ Chakrabarti,
513: Phys.\ Rev.\ E \textbf{67}, 046122 (2003).
514:
515: \bibitem{Hidalgo}R.\ C.\ Hidalgo, Y.\ Moreno, F.\ Kun, and H.\ J.\ Herrmann,
516: Phys.\ Rev.\ E \textbf{65}, 046148 (2002).
517:
518: \bibitem{Hansen2}A.\ Hansen and P.\ C.\ Hemmer, Trends in Statistical
519: Physics \textbf{1}, 213 (1994).
520:
521: \bibitem{Zapperi}S.\ Zapperi, P.\ Ray, H.\ E.\ Stanley, and A.\ Vespignati,
522: Phys.\ Rev.\ Lett.\ {}\textbf{85}, 2865 (2000).
523:
524: \bibitem{kawamura06}H. Kawamura, arXiv: cond-mat/0603335 (2006).
525: \end{thebibliography}
526:
527: \end{document}
528: