1: %\Documentstyle[12pt,aps,prd,preprint]{revtex}
2:
3: \documentstyle[prl,floats,aps,twocolumn,epsf,graphicx]{revtex}
4: \begin{document}
5: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname
6: @twocolumnfalse\endcsname
7:
8: %\begin{document}
9: \title{Survival probabilities in time-dependent random walks}
10: \author{Ehud Nakar and Shahar Hod}
11: \address{The Racah Institute of Physics, The Hebrew University, Jerusalem 91904, Israel}
12: \date{\today}
13: \maketitle
14:
15: \begin{abstract}
16:
17: \ \ \ We analyze the dynamics of random walks in which the jumping probabilities
18: are periodic {\it time-dependent} functions. In particular, we determine the
19: survival probability of biased walkers who are drifted towards an absorbing boundary.
20: The typical life-time of the walkers is found to decrease with an increment of the
21: oscillation amplitude of the jumping probabilities.
22: We discuss the applicability of the results in the context of complex adaptive
23: systems.
24: \end{abstract}
25: \bigskip
26:
27: ]
28:
29: Random walk is one of the most ubiquitous concepts of statistical
30: physics. In fact, it finds applications in virtually every area of
31: physics (see e.g.,\cite{BaNi,Kam,FeFrSo,Wei,AvHa,DiDa} and
32: references therein). Random walks in the presence of absorbing
33: traps are much studied in recent years as models for
34: absorbing-state phase transitions \cite{MaDi,Hin}, polymer
35: adsorption \cite{BeLo}, granular segregation \cite{FaFu}, and the
36: spreading of an epidemic \cite{GrChRo}.
37:
38: One of the main characteristics of a random walk in the presence of
39: an absorbing boundary is the survival probability $S(t_0)$ of the the
40: walkers, the probability that a walker has not reached the
41: absorption point before $t_0$.
42: When the walkers are drifted towards the absorbing point, the survival
43: probability falls exponentially at asymptotically late-times
44: $S(t)\propto exp(-t/\tau_s)$, where the characteristic life-time $\tau_s$
45: depends on the drift velocity (see e.g. \cite{BaGoLu}).
46:
47: The problem of random walks with an absorption boundary has recently been
48: extended to include situation in which the jumping probabilities of the
49: walkers are {\it time-dependent} \cite{Hod03}. The solution
50: presented in \cite{Hod03} accounts for the binary (alternating) case,
51: namely $p(t)={1 \over 2}-\epsilon_p +(-1)^t A$, where $p(t)$ is the
52: probability to step to the right at time $t$. Here $A$ is the oscillations
53: amplitude, and $\epsilon_p>0$ represents a net drift towards the absorbing boundary (the
54: period of such binary oscillations is $T=2$).
55: It was found \cite{Hod03} that the characteristic life-time of the walkers $\tau_s$
56: depends on the amplitude $A$, in a non-trivial way.
57: In particular, $\tau_s$ was shown to decrease monotonically with
58: the increment of $A$. While this analysis provides a useful
59: insight into the behavior of such time-dependent random walks, it is
60: of one's interest to explore the general case time-dependent
61: jumping probabilities with $T \neq 2$. This is the main goal of the
62: present work.
63:
64: In addition to the intrinsic interest in such time-dependent random walks, our
65: work may find direct applications in many complex physical, biological, and
66: economical systems. In fact, the main motivation for the introduction of
67: time-dependent random walks in \cite{Hod03} was its applicability
68: in the flourishing field of complex adaptive systems.
69: In the well-studied model of the minority game (MG) \cite{ChaZha},
70: and its evolutionary version (EMG) \cite{John1}
71: (see also\cite{DhRo,BurCev,LoHuJo,HuLoJo,HaJeJoHu,BuCePe,LoLiHuJo,LiVaSa,SaMaRi,HodNak1,Coo,HodNak2,CaCe}),
72: it was found that the winning probabilities of the agents display
73: a periodic behavior in time. This implies that the survival probabilities of the
74: agents are well-described by a model of a periodic time-dependent random walk with an
75: absorbing boundary \cite{Hod03}. The analytical model presented in \cite{Hod03} (with $T=2$)
76: provides an elegant explanation for the intriguing phase-transition
77: observed in the EMG \cite{HodNak1} (from self-segregation to clustering as the prize-to-fine ratio drops
78: below some critical value).
79: However, numerical studies of the EMG \cite{HodNak2} have indicated that
80: the period of the oscillations (in the winning probabilities of the agents) depends on the
81: specific parameters of the system. Extending the analysis of \cite{Hod03} to the general case of
82: time-dependent jumping probabilities with $T \neq 2$ is therefore
83: highly motivated.
84:
85: We consider a time dependent random walker who is drifted towards an absorption boundary,
86: located at $-d$ (where $d>0$). The drift towards
87: the absorption point may be attributed to two distinct reasons: (i) A smaller (average)
88: probability to take a step to the right (away from the
89: absorption point). That is, $\langle p(t) \rangle=1/2-\epsilon_p$ where $\epsilon_p >0$,
90: and (ii) a smaller step size $r$ to the right, $0\leq r<1$
91: (where the step size to the left is scaled to $1$). This kind of drift is
92: characterized by the parameter $\epsilon_r \equiv 1/(1+r)-{1 \over 2}$.
93: In the unbiased case ($\epsilon_p=\epsilon_r=0$)
94: there is no net drift, and the survival probability is well
95: known to scale as an inverse power-law: $S(t)\propto t^{-1/2}$. In the
96: presence of a drift ($\epsilon_p >0$ and/or $\epsilon_r>0$), and with a
97: constant (time-{\it independent}) jumping probabilities, one finds
98: (see e.g. \cite{BaGoLu})
99:
100: \begin{equation}
101: \label{EQ S_t}
102: S(t)\propto t^{-{3 \over 2}} e^{-t/{\tau }_{s}}\ .
103: \end{equation}
104: This result holds true in the {\it binary} time-dependent model \cite{Hod03} as well,
105: in which case the jumping probability is given by the alternating
106: function $p(t)={1 \over 2}-\epsilon_p + (-1)^tA$.
107:
108: Below we explore the time-dependence of the survival probability $S(t)$ for a general
109: periodic $p(t)$. Let
110:
111: \begin{equation}
112: \label{eq r(t)}
113: p(t)={1 \over 2}+F(t)\ ,
114: \end{equation}
115: where $F(t)$ is a periodic function, with a period $T$.
116: The mean of $F$ (over one period) is given by
117:
118: \begin{equation}
119: \label{EQ F}
120: \langle F \rangle \equiv (1/T)\sum ^{T}_{i=1}F(i)=-\epsilon_p\ .
121: \end{equation}
122:
123: In the time-independent case it has been established
124: that the probability $B(t)$ for a random walker
125: (with {\it no} absorbing boundary) to be located on the
126: right hand side [$x(t)>0$] at the time $t$ is given by $B(t) \propto exp[-t/\tau_B]$, with
127: $\tau _B=\tau_s$ (see e.g., \cite{BaGoLu}).
128: Motivated by the equality $\tau _B=\tau_s$ in the time independent case,
129: we will find the asymptotic form of $\tau_B$ in the general case of
130: time-dependent jumping probabilities. We will establish
131: the relation $\tau_B(T=2)\approx \tau_s(T=2)$ analytically, and demonstrate numerically
132: that $\tau_B(T \neq 2)\approx \tau_s(T \neq2)$.
133:
134: We first consider the case in which the period $T$ of the oscillations is a natural
135: number. Let $\omega(t)$ be the number of right steps
136: taken by the walker out of $t$ steps. The walker is located at the
137: right hand side [$x(t)>0$] if $\omega>\omega_c$, where
138:
139: \begin{equation}
140: \label{EQ w_{c}}
141: \omega_c={{t-d} \over {1+r}}\simeq t({1 \over 2}+\epsilon_r)\ ,
142: \end{equation}
143: in the $t\gg d$, $T$ limit.
144:
145: Note that $\omega$ is a sum of $T$ independent binomial distributions with
146: a success probability $p_{i}=p(t=i)$. After time $t\gg T$ a step is chosen
147: $\simeq t/T$ times from each one of these distributions. Thus, for
148: $t\gg T$, $\omega$ is distributed normally with an average of
149: $\mu_w(t)=t({1 \over 2}-\epsilon_p)$, and a variance
150: $\sigma^2_{\omega}={t \over T}\sum^{T}_{i=1}p_{i}(1-p_{i})=t(1-4 \langle F^2 \rangle)/4$, where
151:
152: \begin{equation}
153: \label{EQ F2}
154: \langle F^2 \rangle \equiv (1/T)\sum ^{T}_{i=1}F^2(i)\ .
155: \end{equation}
156: Hence, $B(t)$ can be approximated by this
157: normal distribution, with the condition $\omega(t)>\omega_c(t)$\
158:
159: \begin{eqnarray}
160: \label{EQ B(t)}
161: B(t)=\frac{1}{\sqrt{2\pi \sigma ^{2}}}\int ^{\infty
162: }_{w_{c}}exp\left[ -\frac{(x-\mu )^{2}}{2\sigma ^{2}}\right]
163: dx\nonumber \\
164: =\frac{1}{\sqrt{\pi }}\int ^{\infty }_{\sqrt{t/\tau
165: _{B}}}e^{-x^{2}}dx=\frac{1}{2}erfc(\sqrt{t/\tau _{B}})\ ,
166: \end{eqnarray}
167: where
168:
169: \begin{equation}
170: \label{EQ tauB}
171: \tau _B=\frac{2\sigma ^2t}{(w_c-\mu)^{2}}=\frac{1-4\langle F^2\rangle}{2(\epsilon_p+\epsilon_r)^2}\ ,
172: \end{equation}
173: and \( erfc \) is the complementary error function. The complementary error function
174: can be approximated as $erfc(\sqrt{t/\tau _{B}})\simeq e^{-t/\tau _{B}}/\sqrt{\pi t/\tau_{B}}$ at
175: asymptotically large times. One therefore finds
176:
177: \begin{equation}
178: B(t)\propto e^{-t/\tau_{B}}\ .
179: \end{equation}
180:
181: The model of binary time-dependent jumping probabilities [$F(t)=-\epsilon_p+A(-1)^t$]
182: was solved {\it exactly} in \cite{Hod03}, the expression for $\tau_s$ is given by
183: Eq. ($16$) of \cite{Hod03}. Expanding this expression for small $\epsilon_p$ and
184: $\epsilon_r$, one finds
185:
186: \begin{equation}
187: \label{EQ tauB_T2}
188: \tau_s(T=2)\simeq {{1-4A^2} \over {2(\epsilon_p+\epsilon_r)^2}}\ .
189: \end{equation}
190: This agrees with Eq. (\ref{EQ tauB}) in the limit $\epsilon_p \ll 1$ (or equivalently,
191: $\tau_s \gg 1$).
192: One therefore finds $\tau_s(T=2) \simeq \tau_B(T=2)$ in the $\tau_s \gg 1$ limit.
193:
194: The generalization of the above analysis for an arbitrary period $T$ is
195: straightforward. If $T=m/n$ is a rational number, then $F(t)$ could be replaced by an
196: equivalent function with a period $m$ which is a natural number.
197: If $T$ is an irrational, then the all range of
198: values of $F(0<t<T)$ is sampled (over many periods of the oscillations).
199: In this case, one should replace Eqs. (\ref{EQ F}) and (\ref{EQ F2}) by
200:
201: \begin{equation}
202: \langle F \rangle=\frac{1}{T}\int_0^T F(t)dt=-\epsilon\ ,
203: \end{equation}
204: and
205:
206: \begin{equation}
207: \langle F^2 \rangle=\frac{1}{T}\int_0^T F^2(t)dt\ .
208: \end{equation}
209:
210: In order to confirm the analytical predictions (in the $T \neq 2$ case),
211: we perform numerical simulations of (discrete) random walks.
212: We first consider the case of an harmonic function $p(t)$ of the form
213:
214: \begin{equation}
215: \label{EQ r_sinus}
216: p(t)={1 \over 2}-\epsilon_p -Acos(\frac{2\pi t}{T})\ ;\ T>1\ .
217: \end{equation}
218: In this case $\langle F \rangle =-\epsilon_p$, and
219:
220: \begin{equation}
221: \label{EQ F2_sinus}
222: \langle F^2 \rangle=\frac{1}{4}\left\{
223: \begin{array}{c}
224: \epsilon^2_p+A^{2}\ ;\quad T=2\\
225: \epsilon^2_p+(A/\sqrt{2})^{2}\ ;\quad T\neq 2
226: \end{array}\right.
227: \end{equation}
228: This implies that for $T \neq 2$, $\tau _{B}$ is {\it independent} of the period $T$ of the
229: oscillations. Moreover, we find that
230: $\tau_B(T \neq 2,A/\sqrt{2})=\tau _B(T=2,A) $. As we know that our
231: approximation holds true for $T=2$ (The exact
232: shape of $F$ is irrelevant in this case), one should compare
233: the survival probabilities in the $T=2$ and $T \neq 2$ cases.
234: Figure \ref{Fig f1} depicts $t(S=10^{-5})$, the step number in which the survival
235: probability has fallen to $S=10^{-5}$, as a function of the oscillation period $T$
236: (and for three different values of the oscillation amplitude $A$).
237: As predicted by Eq. (\ref{EQ F2_sinus}), the survival probabilities are almost
238: independent of the period $T$, except in the unique case of binary oscillations with $T=2$.
239: Moreover, $S(t;T=2) < S(t;T \neq 2)$, in agreement with Eqs. (\ref{EQ tauB}) and (\ref{EQ F2_sinus})].
240: In addition, the numerical simulations indicate that the survival probability decreases monotonically with an increment
241: in the oscillation amplitude $A$, in agreement with the analytical prediction
242: Eq. (\ref{EQ tauB}).
243:
244: In Figure \ref{Fig f2} we confirm the relation $S(t,T\neq 2,A)=S(t,T= 2,\sqrt{2}A)$ in the
245: harmonic case. The figure displays the step number $t(S=10^{-6})$ for which the survival probabilities
246: have fallen to $10^{-6}$. It is clear that $t(S=10^{-6})$
247: for $T=2$ agrees with the corresponding value in the $T \neq 2$ case, provided one
248: uses the transformation $A\rightarrow \sqrt{2}A$.
249:
250: \begin{figure}[tbh]
251: \centerline{\epsfxsize=9cm \epsfbox{sdtfig1.eps}}
252: \caption{The step number for which the survival probability has fallen to $S=10^{-5}$,
253: as a function of the oscillation period $T$, and for
254: three different values of $A$. The jumping probabilities $p(t)$ are characterize by harmonic
255: oscillations, and are given by Eq. (\ref{EQ r_sinus}).
256: We use $N=10^8$ walkers, $\epsilon_p=0.1$, $r=1$, and
257: $d=2$. $S(t)$ is independent of the oscillation period $T$, except in the
258: unique case of binary oscillations with $T=2$ [in which case $S(t)$ is
259: smaller, in agreement with the analytical prediction.]
260: In addition, $S(t)$ decreases monotonically with the increment of the oscillations amplitude $A$.}
261: \label{Fig f1}
262: \end{figure}
263:
264: \begin{figure}[tbh]
265: \centerline{\epsfxsize=9cm \epsfbox{sdtfig2.eps}}
266: \caption{The step number in which $S=10^{-6}$ as a function of the oscillations
267: amplitude $A$, and for two different values of the period $T$. The results confirm
268: the analytical prediction according to which $S(t,T=2,A)=S(t,T\neq 2,\sqrt{2}A)$, where
269: $p(t)$ is an harmonic function. The parameters used are the same as in Fig. 1.}
270: \label{Fig f2}
271: \end{figure}
272:
273: We further check the validity of the analytical results for $F(t)$ in the form of
274: a square wave of amplitude $A$, and a mean value $-\epsilon_p$. In this case one finds
275: $\langle F^2 \rangle=\epsilon^2_p+A^2$ for any $T$. This implies that $\tau_B(T)$ should be
276: {\it independent} of the period $T$ of the oscilations (and in particular,
277: $\tau_s(T) =\tau_s(T=2)$, where $\tau_s(T=2)$ was derived analytically in \cite{Hod03}).
278: Figure \ref{Fig f3} depicts $t(S=10^{-6})$, the step number for which the survival probability
279: has fallen to $S=10^{-6}$, as a function of the period $T$. We display results for three
280: different values of the oscillation amplitude $A$. The numerical results confirm the finding according to
281: which $S(t;T)$ is independent of the value of $T$.
282: Furthermore, the survival probability $S(t;T)$ decreases with increasing amplitude $A$, in
283: agrrement with Eq. (\ref{EQ tauB}).
284:
285: \begin{figure}[tbh]
286: \centerline{\epsfxsize=9cm \epsfbox{sdtfig3.eps}}
287: \caption{The step number for which the survival probability has fallen to $S=10^{-6}$,
288: as a function of the oscillation period $T$, and for
289: three different values of $A$. $F(t)$ is a square wave with amplitude $A$,
290: and a mean value $-\epsilon_p$.
291: We use $N=10^9$ walkers, $\epsilon_p=0.1$, $R=0.9$, and
292: $d=2$. We find that in this case, $S(t)$ is independent of the oscillation period $T$,
293: in agreement with the analytical results.}
294: \label{Fig f3}
295: \end{figure}
296:
297: In summary, in this paper we studied the survival probabilities of biased (drifted)
298: random walkers, whose jumping probabilities are {\it time-dependent}. This is
299: an extension of the binary case (with $T=2$), studied in \cite{Hod03}.
300: The long-time asymptotic survival probability is dominated by an exponential
301: fall $S(t)\propto exp[-t/\tau_s]$.
302: We have found a simple approximation for $\tau_s$
303: [see Eq. (\ref{EQ tauB})], which solely depends on the first and the second moments
304: of the oscillations ($\langle F \rangle$ and $\langle F^2 \rangle$), and on the step-size drift
305: parameter $\epsilon_r$.
306:
307: Our analytical results imply that the characteristic life-time of the
308: walkers $\tau_s$ {\it decreases} with increasing $\langle F^2 \rangle$.
309: The larger are the temporal-oscillations in the jumping probabilities $p(t)$, the
310: smaller is the survival probability. This result generalizes the one derived
311: in \cite{Hod03} for the binary case (with $T=2$), in which case $\langle F^2 \rangle=A^2+{\epsilon^2_p}$.
312:
313: The qualitative nature of the anti-correlation between the survival probability $S(t)$ and
314: the oscillation amplitude $\langle F^2 \rangle$ is rather simple.
315: It is a direct consequence of the positive correlation between $\sigma_{\omega}(t)$
316: (the dispersion in the access of right-steps taken out of $t$ steps) and $S(t)$
317: in the presence of a negative drift (towards the absorption boundary).
318: In the presence of a net drift, $\mu_{\omega}$ (the average value of $\omega$)
319: is smaller than $\omega_c$. A large $\sigma_{\omega}(t)$ is therefore required
320: in order to survive, that is in order to have $\omega > \omega_c$ at asymptotically late-times \cite{Note}.
321:
322: Finally, the present analysis provides a direct explanation for the underlying mechanism responsible for
323: the intriguing phase-transition observed in the evolutionary
324: minority game \cite{HodNak1,John1}. In this model, the winning probabilities of the
325: agents $p(t)$ where shown to display temporal oscillations \cite{HodNak2}.
326: The amplitude and period of these oscillations depend
327: on the various parameters of the model (such as the price-to-fine ratio).
328: In \cite{Hod03} a toy-model of time-dependent jumping probabilities with a period of $T=2$ was
329: used to reveal the physical source for the phase-transition. In the present work we
330: have shown that the anti-correlation between the characteristic life-time $\tau_s$ and the
331: oscillations amplitude $\langle F^2 \rangle$ is a generic feature, {\it independent} of the period $T$
332: of the oscillations.
333: Thus, our analysis (for generic values of $T$) lends support for the general applicability
334: of the conclusions presented in \cite{Hod03}.
335:
336: \bigskip
337: \noindent
338: {\bf ACKNOWLEDGMENTS}
339: \bigskip
340:
341: The research of SH was supported by G.I.F. Foundation. The research of EN was supported by the
342: Horowitz foundation, and through the generosity of the Dan David Prize Scholarship 2003. We would also
343: like to thank Yonatan Oren.
344:
345: \begin{thebibliography}{99}
346:
347: \bibitem{BaNi} M. N. Barber and B. W. Ninham, {\it Random and Restricted Walks} (Gordon
348: and Breach, New York, 1970).
349:
350: \bibitem{Kam} N. G. van Kampen, {\it Stochastic Processes in Physics and
351: Chemistry} (North-Holland, Amsterdam, 1992).
352:
353: \bibitem{FeFrSo} R. Fernandez, J. Frohlich, and A. D. Sokal, {\it Random Walks, Critical
354: Phenomena, and Triviality in Quantum Field Theory} (Springer Verlag, Berlin, 1992).
355:
356: \bibitem{Wei} G. H. Weiss, {\it Aspects and Applications of the Random Walk} (North
357: Holland, Amsterdam, 1994).
358:
359: \bibitem{AvHa} D. ben-Avraham and S. Havlin, {\it Diffusion and Reactions in Fractals and
360: Disordered Systems} (Cambridge University Press, Cambridge, 2000).
361:
362: \bibitem{DiDa} R. Dickman and D. ben-Avraham, Phys. Rev. E. {\bf 64}, 020102(R) (2001).
363:
364: \bibitem{MaDi} J. Marro and R. Dickman, {\it Nonequilibrium Phase Transitions in Lattice Models},
365: (Cambridge University Press, Cambridge, 1999).
366:
367: \bibitem{Hin} H. Hinrichsen, Adv. Phys. {\bf 49}, 815 (2000).
368:
369: \bibitem{BeLo} K. De'Bell and T. Lookman, Rev. Mod. Phys. {\bf 65}, 87 (1993).
370:
371: \bibitem{FaFu} Z. Farkasa and T. Fulop, J. Phys. A: Math. Gen. {\bf
372: 34}, 3191 (2001).
373:
374: \bibitem{GrChRo} P. Grassberger, H. Chate, and G. Rousseau,
375: Phys. Rev. E {\bf 55}, 2488 (1997).
376:
377: \bibitem{BaGoLu} M. Bauer, C. Godreche and J. M. Luck,
378: J. Stat. Phys. {\bf 96}, 963 (1999).
379:
380: \bibitem{Hod03} S. Hod, Phys. Rev. Lett. {\bf 90}, 128701 (2003).
381:
382: \bibitem{ChaZha} D. Challet and C. Zhang, Physica A {\bf 246}, 407 (1997);
383: {\bf 256}, 514 (1998); {\bf 269}, 30 (1999).
384:
385: \bibitem{John1} N. F. Johnson, P. M. Hui, R. Jonson, and T. S. Lo, Phys. Rev.
386: Lett. {\bf 82}, 3360 (1999).
387:
388: \bibitem{DhRo} R. D`Hulst and G. J. Rodgers, Physica A {\bf 270}, 514 (1999).
389:
390: \bibitem{BurCev} E. Burgos and H Ceva, Physica {\bf 284A}, 489 (2000).
391:
392: \bibitem{LoHuJo} T. S. Lo, P. M. Hui and N. F. Johnson, Phys. Rev. E {\bf 62}, 4393 (2000).
393:
394: \bibitem{HuLoJo} P. M. Hui, T. S. Lo, and N. F. Johnson, e-print cond-mat/0003309.
395:
396: \bibitem{HaJeJoHu} M. Hart, P. Jefferies, N. F. Johnson and P. M. Hui, e-print cond-mat/0003486;
397: e-print cond-mat/0004063.
398:
399: \bibitem{BuCePe} E. Burgos, H. Ceva and R. P. J. Perazzo, e-print cond-mat/0007010.
400:
401: \bibitem{LoLiHuJo} T. S. Lo, S. W. Lim, P. M. Hui and N. F. Johnson, Physica {\bf 287A}, 313 (2000).
402:
403: \bibitem{LiVaSa} Y. Li, A. VanDeemen and R. Savit, e-print nlin.AO/0002004.
404:
405: \bibitem{SaMaRi} R. Savit, R. Manuca and R. Riolo, Phys. Rev. Lett. {\bf 82}, 2203 (1999).
406:
407: \bibitem{HodNak1} S. Hod and E. Nakar, Phys. Rev. Lett. {\bf 88}, 238702 (2002).
408:
409: \bibitem{Coo} A. C. C. Coolen, e-print cond-mat/0205262.
410:
411: \bibitem{HodNak2} E. Nakar and S. Hod, Phys. Rev. E {\bf 67}, 016109 (2003).
412:
413: \bibitem{CaCe} I. Caridi and H. Ceva, e-print cond-mat/0206515.
414:
415: \bibitem{Note} We note that in the unbiased case (with no drift), one finds $\mu_{\omega} > \omega_c$.
416: A small $\sigma_{\omega}(t)$ therefore implies large survival probability. Thus,
417: $\langle F^2 \rangle$ is positively correlated with $S(t)$ in the unbiased case, in agreement with
418: the results of \cite{Hod03} for the binary case.
419:
420: \end{thebibliography}
421:
422: \end{document}
423: