1:
2: %\documentclass[11pt,a4paper]{article}
3: %\documentclass[showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
4: \documentclass[showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
5: %\documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
6: %\documentclass[twocolumn,showpacs]{revtex4}
7: %\documentclass[reqno]{article}
8: %\documentclass{tran-l}
9: \usepackage{graphicx}
10:
11:
12:
13: \begin{document}
14: \title{Nonlocal competition and logistic growth: patterns, defects and fronts}
15: \author{Yosef E. Maruvka and Nadav M. Shnerb}
16: \affiliation{Department of Physics, Bar-Ilan University, Ramat-Gan
17: 52900 Israel}
18: \begin{abstract}
19: Logistic growth of diffusing reactants on spatial domains with
20: long range competition is studied. The bifurcations cascade
21: involved in the transition from the homogenous state to a
22: spatially modulated stable solution is presented, and a
23: distinction is made between a modulated phase, dominated by single
24: or few wavenumbers, and the spiky phase, where localized colonies
25: are separated by depleted region. The characteristic defects in
26: the periodic structure are presented for each phase, together
27: with the invasion dynamics in case of local initiation. It is
28: shown that the basic length scale that controls the bifurcation is
29: the width of the Fisher front, and that the total population
30: grows as this width decreases. A mix of analytic results and
31: extensive numerical simulations yields a comprehensive examination
32: of the possible phases for logistic growth in the presence of
33: nonlocal competition.
34: \end{abstract}
35:
36: \pacs{87.17.Aa,05.45.Yv,87.17.Ee,82.40.Np}
37:
38: \maketitle
39:
40: \section{Introduction}
41:
42: \noindent Recently, there is a growing interest in the spatial
43: properties of logistic growth with nonlocal interactions
44: \cite{Sakai, Doebeli, Bolker, Tokita, Hoopes, Anderson, Sayama,
45: Fuentes, Birch, shnerb, Garcia}. A variety of models have been
46: introduced, including various types of interaction kernels,
47: deterministic and stochastic evolution and growth or death rate
48: that depends on the local population. A common feature found in
49: all these models is the \emph{segregation transition}, i.e., for
50: small enough diffusion and for certain interaction kernels the
51: homogenous state of the system becomes unstable and the steady
52: state is spatially heterogenous. This feature turns out to be
53: stable against the stochasticity induced by the discrete nature of
54: the reactants, and the total carrying capacity (per unit volume)
55: of the stochastic system depends on the details of the spatial
56: segregation \cite{Birch, Garcia}.
57:
58: In previous work \cite{shnerb}, the general conditions for the
59: integral kernel to allow for spatial segregation have been
60: presented, and the existence of topological defects between
61: ordered domains has been analyzed in detail for a logistic growth
62: on a one dimensional array of patches with nearest-neighbor
63: competition. Here, a comprehensive study of this
64: reaction-diffusion equation is presented: short-range
65: interactions are shown to yield spatial modulation of arbitrary
66: large wavelength and different type of defects, the total
67: population of the system admits nontrivial dependence upon the
68: diffusion rate, and the dynamics of the system is studied, both
69: for global initiation and for local initiation. The appearance of
70: domains with different order parameter and the features of the
71: boundaries between them is considered in detail for various
72: situations.
73:
74: Our starting point is the well-investigated Fisher-KPP equation
75: \cite{fisher, kol}, first introduced by Fisher to describe the
76: spread of a favored gene in population:
77: \begin{equation}
78: \frac{\partial c(x,t)}{\partial t}= D \nabla^2
79: c(x,t)+ac(x,t)-bc^2(x,t).
80: \end{equation}
81: Clearly, this equation is a straightforward generalization of the
82: logistic growth to spatial domains, and allows for two steady
83: states: an unstable state with $c(x)=0 \ \forall x$ and the
84: stable steady state $c(x)=a/b$. It was shown that, for any local
85: initiation of the instability (i.e., $c(x)\neq 0 $ on a compact
86: domain) the invasion of the stable phase into the unstable region
87: takes place via a front that moves in a constant velocity $v_F = 2
88: \sqrt{Da}$. The stability of this solution, the fact that the
89: velocity is determined by the leading edge ("pulled front") and
90: the corrections to this expression due to stochastic noise
91: associated with the discrete nature of the reactants
92: \cite{derrida} has been reviewed, recently, by various authors
93: \cite{saarloos}.
94:
95: The FKPP equation is the simplest equation that describes the
96: transition from unstable to stable steady state on spatial
97: domains, and as such it fits many situations, from the spread of a
98: disease by infection to the advance of a fire or new technology.
99: Accordingly, this model has been widely studied from many points
100: of view and has been generalized in many directions such as
101: modified interaction terms, non linear diffusion and so on.
102:
103: The process considered here, logistic growth with nonlocal
104: competition, is described by the generalized FKPP equation:
105: \begin{equation} \label{eq2}
106: \frac{\partial c(x,t)}{\partial t}= D \nabla^2
107: c(x,t)+ac(x,t)-c(x,t)\int^\infty_{- \infty} \gamma(x,y) c(y,t)dy,
108: \end{equation}
109: where $\gamma(x,y)$ is the interaction kernel, and the original
110: FKPP process corresponds to the limit $\gamma(x,y)=\delta(x-y)$.
111:
112: The motivation for the study of this process comes from one of
113: the basic mechanisms in population growth, namely, the competition
114: for common resource. In any autocatalytic system the
115: multiplication of agents depends on various resources (energy,
116: chemicals, water etc.). If there is only limited amount of the
117: resource, its consumption leads to extinction, so generally any
118: crucial resource should be deposited, and its availability
119: dictates the saturation value for the population. As a concrete
120: example let us look at vegetation \cite{merron, lavee}: the common
121: resource needed for vegetation is water, and the rain corresponds
122: to deposition of this resource. If the resource dynamics is much
123: faster than that of the agents (shrubs, trees etc.), there is, at
124: any time, a soil moisture profile that reflects the instantaneous
125: vegetation configuration, and there is a depletion of this
126: moisture at the spatial region around a biomass unit. Accordingly,
127: the environmental conditions for a new agent at this region
128: becomes hostile. Following arguments of this type one suggests
129: that \emph{competition for common resource induces long range
130: competition among agents via the depletion of the resource in the
131: surroundings of an agent}. Another examples may involve the
132: competition for light \cite{huisman} and cooperation among agents
133: (symbiosis) that may yield "negative competition" among the
134: reactants.
135:
136: Numerical simulations of the dynamics corresponds to Eq.
137: (\ref{eq2}) require space and time discretization. In this work
138: the time evolution of the system is generated via forward Euler
139: integration, where the time step is taken small enough such that
140: further reduction of it do not effect the results. We simulate a
141: system of discrete patches, where the hopping rate is proportional
142: to the diffusion constant.
143:
144: Let us present some a-priori considerations related to this
145: system. There are few basic types of steady state solutions:
146: first, it may happens that the steady state is \emph{homogenous}:
147: this may be the case if the long range competition is too small,
148: or if the interaction kernel do not allow for the instability to
149: occur \cite{shnerb, Sayama}. At some point in the parameter region
150: a bifurcation may occur, and the homogenous states becomes
151: unstable to modulation of wavelength $2 \pi /k$. Right above the
152: bifurcation one expects, though, to see an inhomogeneous
153: (modulated) steady state. Far from the bifurcation point there are
154: many unstable wavelengthes and some sort of mode competition
155: takes place. In the other limit, i.e., very strong competition,
156: one may expect that "life" at a single patch forces all the other
157: patches at finite range to be (almost) empty, so the steady state
158: is sort of "spiky" phase, where many active wavelengthes
159: participate in the formation of localized bumps.
160:
161: As we are looking at a dynamic system with no noise, few stable
162: steady states may exist simultaneously, each admits its own basin
163: of attraction in the space of possible initial conditions.
164: Numerically, however, it turns out that only one important
165: distinction should be made, namely, between local and global
166: initiation: the initiation is "local" if at $t=0$ there is finite
167: support to the colony, while if the system begins with random
168: small biomass that spreads all around it corresponds to global
169: initiation. Within each of these subclasses, the numerics
170: suggests that a generic initial condition will flow into a unique
171: steady state.
172:
173:
174: This paper is organized as follows: in the second section the
175: stable spatial configurations (steady states stable against small
176: fluctuations) are presented: the conditions for an instability of
177: the homogenous solution are reviewed and discussed, and the
178: properties of the final state are identified in different
179: parameter regions, leading to a characteristic "phase diagram". In
180: the next section the appearance of defects (separating spatial
181: regions with different order parameter phase) is studied. The
182: fourth section deals with the "spiky" phase, where many excited
183: modes superimposed to yield a pattern of spikes and the typical
184: defect is a combination of two depletion regions. In the fifth
185: section there is a brief description of phases and defects in two
186: spatial dimensions, and next the effect of the spatial
187: segregation on the global population is considered. In the seventh
188: section the dynamic properties of the model are discussed,
189: including the velocity of the primary and the secondary Fisher
190: fronts and the appearance of topological defect in the invaded
191: region. Some comments and conclusions are presented at the end.
192:
193: \section{Static properties}
194:
195:
196:
197:
198: In this section we consider the steady state solutions for
199: equation (\ref{eq2}) on spatial domain of coupled, identical
200: patches. The initiation is assumed to be \emph{global}, i.e., the
201: initial conditions are small, randomly spread, reactant population
202: at each spatial patch. The model considered here allows for
203: nontrivial spatial organization even in the absence of diffusion,
204: due to the long range competition, and global initiation helps to
205: see these features within reasonable simulation times. The
206: differences, if any, between global and local initiation will be
207: considered in the last section.
208:
209:
210:
211: \subsection{The bifurcation cascade}
212:
213: Let us consider the spatially discretized version of (\ref{eq2}),
214: i.e., an infinite one dimensional array of identical patches
215: coupled to each other by diffusion and long range competition.
216: The time evolution of the reactant density at the n-th site,
217: $\widetilde{c}_n$, is given by:
218:
219: \begin{eqnarray}
220: \label{eq:1} \frac{\partial \widetilde{c}_n(t)}{\partial
221: t}&=&\frac{\widetilde{D}}{l_0^2}[-2\widetilde{c}_n(t)+\widetilde{c}_{n+1}(t)+\widetilde{c}_{n-1}(t)]+
222: a \widetilde{c}_n(t) \nonumber \\ &-&b \widetilde{c}_n^{2}(t) -
223: \widetilde{c}_n(t) \sum_{r=1}^\infty \widetilde{\gamma}_r
224: [\widetilde{ c}_{n+r}(t)+\widetilde{c}_{n-r}(t)].
225: \end{eqnarray}
226: where $\widetilde{D}$ is the diffusion constant and
227: $a,b,\widetilde{\gamma}$ are the corresponding reaction
228: coefficients. One may precede to define the dimensionless
229: quantities,
230: \begin{equation} \tau=at, \qquad
231: c=b\widetilde{c}/a, \qquad \gamma_r=\widetilde{\gamma}_r/b,
232: \qquad
233: D=\frac{\widetilde{D}}{a l_0^2}.
234: \end{equation}
235: Note that the new "diffusion constant" is $D = W^2/l_0^2$,
236: where $W \equiv \sqrt{D/a}$ is the width of the Fisher front, so
237: the dimensionless diffusion is determined by the ratio between
238: the front width and the lattice constant. The continuum limit,
239: though, is the limit where the front width is large in units of
240: lattice spacing. With these definitions Eq. (\ref{eq:1}) takes its
241: dimensionless form,
242: \begin{eqnarray}
243: \label{eq:5} \frac{\partial c_n}{\partial \tau}&=& D
244: [-2c_n+c_{n+1}+c_{n-1}] \nonumber \\ &+& c_n \left( 1-c_n-
245: \sum_{r=1}^\infty \gamma_r [c_{n+r}+ c_{n-r}] \right),
246: \end{eqnarray}
247: that may be expressed in Fourier space [with $A_k \equiv \sum_n
248: c_n e^{iknl_0}$] as,
249: \begin{equation} \label{eq3}
250: \dot{A}_k = \alpha_k A_k - \sum_q \beta_{k-q} A_q A_{k-q},
251: \end{equation}
252: where
253: \begin{eqnarray}
254: \alpha_k \equiv 1-2D[1-cos(kl_0)] \\
255: \beta_k \equiv 1+2\sum_{r=1}^\infty \gamma_r cos(rkl_0).
256: \end{eqnarray}
257:
258: \noindent Following \cite{ns}, one observes that $c_n$ is
259: positive semi-definite so $A_0$ is always "macroscopic". Any mode
260: is suppressed by $A_0$; accordingly, for small $\gamma_r$ one
261: expects only the zero mode to survive. If, on the other hand,
262: $\gamma_r$ increases above some threshold, bifurcation may occur
263: with the activization of some other $k$ mode(s), and the
264: homogenous solution becomes unstable. The condition for occurrence
265: of such bifurcation is that:
266: \begin{equation}\label{bif}
267: g(k)=\beta_k + 2 \beta_0 D [1-cos(kl_0)]< 0
268: \end{equation}
269: fulfilled by some k. This is the situation where patterns appear
270: and translational symmetry breaks. Right above the bifurcation
271: there is only one active $k$ mode that dictates the modulation of
272: the system. As $g(k)$ decreases further there are many active
273: modes that compete with each other via the nonlinear terms of
274: (\ref{eq3}), and the linear stability analysis of the homogenous
275: state may be irrelevant to the final spatial configuration.
276:
277: \subsection{Nearest neighbors interactions}
278:
279: In previous work, the properties of the system have been
280: considered for the extreme case where the competition takes place
281: only between neighboring sites ($\gamma_r = \gamma$ for $r=1$ AND
282: $\gamma_r=0$ if $r>1$). For nearest neighbors (n.n) interaction of
283: that type the only stable wave number is $k=\pi /l_0$, where
284: $l_0$ is the lattice constant, and the bifurcation takes place at
285: $\gamma<1/2$. The spatial state at this wave number is
286: $u_n=A_0+A_\pi cos(n \pi / l_0)$ and the spatial structure is of
287: the form ...ududud... (u=up, large amount of biomass, d=down,
288: small amount). In the absence of diffusion spatial segregation
289: takes the form 101010…, i.e., only the even (odd) sites are
290: populated. Obviously, starting from generic random state different
291: domains are crated with odd or even "order parameter" and kinks
292: (domain walls) emerge between different domains. As shown in
293: \cite{shnerb}, the structure of these topological defects,
294: including their size (that diverges at the segregation transition)
295: and their exact form may be calculated analytically.
296:
297: \subsection{Next nearest neighbors (n.n.n)}
298:
299: \noindent Quiet surprisingly, the increase of the competition
300: radius by a single site takes us to a completely different
301: regime. While in the case of nearest neighbors interaction the
302: spatial modulation length and the competition length are the same,
303: next nearest neighbors competition (and, accordingly, any
304: interaction of longer range) may yield, upon tuning the
305: parameters, spatial modulation of arbitrary large wavelength. This
306: situation resembles the case of magnetic systems, e.g., an Ising
307: chain: if the exchange interaction is only between nearest
308: neighbors the equilibrium state admits only an up-down
309: modulations, while n.n.n. interaction may yield large solitons, as
310: shown by \cite{bak}. In that sense the next n.n. case demonstrate
311: the essential features of the long range competition model in a
312: generic way, while at least part of the results may be inferred
313: analytically.
314:
315: The most general form of next nearest neighbors interactions is
316: given by Eq. (\ref{eq:5}) with:
317:
318: \begin{equation}
319: \gamma_r = \left\{ \begin{array}{cc}
320: \gamma_1 & r = 1 \\
321: \gamma_2 & r = 2 \\
322: 0 & \texttt{else} \\
323: \end{array} \right.
324: \end{equation}
325:
326: The bifurcation threshold is defined now by the equation:
327:
328: \begin{equation}\label{wave}
329: g(k)=1+2\gamma_1 cos(kl_0)+\gamma_2 cos(2kl_0)\beta_k + 2 \beta_0
330: D [1-cos(kl_0)]= 0.
331: \end{equation}
332:
333: \noindent where $g$ has extremum points at $k_{1,2}=0, \pi $ and
334: \begin{equation} \label{eqk}
335: k_3=arcos(\frac{-\gamma_1+\beta_0D}{4\gamma_2}).
336: \end{equation}
337: If a real wavevector $k_3$ exists (i.e., at
338: $|\frac{-\gamma_1+\beta_0D}{4\gamma_2}|<1$) it will be the minimum
339: of $g(k)$ while $k_{1,2}$ are maxima. For the range of parameters
340: where $k_3$ is imaginary the minima may be at $k= \pi$ and the
341: modulation is of "up-down" type, or $k=0$, where the homogenous
342: state is stable. The resulting phase diagram, in the
343: $\gamma_1-\gamma_2$ plane with zero diffusion, is presented in
344: figure (\ref{fig3}): In region I the homogenous state is stable,
345: while in region II the bifurcation takes the system to the up-down
346: mode, like the situation for n.n. interaction. In region III,
347: however, $k_3$ dominate and modulations of any size may occur. The
348: bifurcation line is given (in the presence of diffusion) by the
349: two branches of the equation:
350:
351: \begin{equation}
352: \gamma_2=\frac{1+2D-D^2+\gamma_1(5D-2D^2) \pm
353: \sqrt{1+4D+14D\gamma_1-2\gamma_1^2+12D\gamma_1^2}}{2D^2+4-16D}
354: \end{equation}
355: that reduces, at the $D=0$ case, to the simple form:
356: \begin{equation}
357: \gamma_2=\frac{1 \pm \sqrt{1-2\gamma_1^2}}{4}.
358: \end{equation}
359:
360:
361: \begin{figure}
362: % Requires \usepackage{graphicx}
363: \includegraphics[width=7cm]{phase_diagram1.eps}\\
364: \caption{Phase diagram for next nearest neighbors competition at
365: $D=0$. Region I is the homogenous, while II marks the up-down
366: stable solution region similar to the n.n. case. In region III
367: the wavevector $k_3$ (defined in the text) is stable and various
368: wavenumbers may be active.
369: }\label{fig3}
370: \end{figure}
371:
372: \subsubsection{wavelength selection, mode competition and the spiky phase}
373:
374: \noindent From (\ref{eqk}) it seems that the bifurcation
375: wavelength is bounded from above by the interaction length. This,
376: however, is not the actual situation on a discrete lattice: the
377: wavelength inferred from Eq. (\ref{eqk}), although bounded, is
378: generically incommensurate with the lattice constant, and the
379: system should choose a commensurate one. It turns out that, if the
380: wavelength is rational (i.e., if $k_3 = 2 \pi m/n$, where $m$ and
381: $n$ admits no common denominator) the spatial modulation repeats
382: itself after $n$ lattice sites. A typical example is the steady
383: state obtained numerically for the case $m=7,n=20$ where a
384: period-20 modulation appears, as demonstrated in Figure
385: (\ref{fig4}). At finite system the maximal $n$ allowed is of order
386: of the system size, and only in an infinite system all rational
387: fractions may be activated. Note that in an infinite system any
388: change of the interaction parameters yields different wavelength,
389: a phenomena that resembles the "devil staircase" situation in spin
390: systems \cite{bak}.
391:
392:
393: \begin{figure}
394: % Requires \usepackage{graphicx}
395: \includegraphics[width=7cm]{wavenumber7over10.EPS}\\
396: \caption{ Spatial structure of wavevector $k=
397: 14\pi/20 l_0$. According to Eq. (\ref{eqk}) one expects the modulation length
398: to be $\lambda=20 l_0/7$, but
399: but discreteness of the
400: lattice allows only for commensurate periodicity of 20 sites.
401: }\label{fig4}
402: \end{figure}
403:
404:
405: For finite system, though, there is a set of points along the
406: bifurcation line that correspond to the allowed wavelengthes.
407: Numerical simulation indicates that there is a basin of attraction
408: around each of these points, i.e., if the interaction parameters
409: $\gamma_1$ and $\gamma_2$ yields a prohibited wavelength the
410: system flows into one of the closest allowed modulations. The
411: overall structure is demonstrated in figure (\ref{fig5}): close to
412: an isolated point there is a basin of attraction, but further away
413: from the bifurcation line these regions begin to overlap, and the
414: system flows into some mixture of the closest allowed states,
415: depending on its initial conditions. Deep in region III many
416: wavenumbers are involved; the interaction parameters are
417: relatively large, and instead of simple harmonic modulation the
418: system flows, generically, into a spiky steady where the "living
419: colonies" are separated by the interaction length and are not
420: effected by the competition between patches.
421:
422:
423: \begin{figure}
424: % Requires \usepackage{graphicx}
425: \includegraphics[width=7cm]{basin_of_attraction.eps}\\
426: \caption{Basins of attraction for allowed states close to the
427: bifurcation line (sketched). The straight line is an enlarged
428: portion of the bifurcation line (I - III interface) of figure
429: (\ref{fig3}). The bold points on this line correspond to an
430: allowed states, i.e., states with wavelength commensurate with
431: the lattice size. Each possible wavevector admits a basin of
432: attraction, like those denoted by A and B in the figure. Starting
433: from $\gamma_1$,$\gamma_2$ values inside region A, for example,
434: the system flows to the modulation corresponds to the bold point
435: inside the triangle. Away from the bifurcation line (region C) few
436: basins of attraction overlap and mode competition takes place.
437: Even further away, deep in region III, the system is in the spiky
438: phase: many active modes exist and their superposition yields the
439: "wave packet" characteristics of the spikes}\label{fig5}
440: \end{figure}
441:
442:
443: Although the numerical examples presented here are for a system
444: with next nearest neighbor competition and without diffusion, it
445: is easy to extract from it the properties of the steady state in
446: general. The effect of diffusion is to increase the size of the
447: stable region so the bifurcation line of Figure (\ref{fig3}) moves
448: outward together with the pure and the spiky states. For
449: interactions of longer range the parameter space is of higher
450: dimensionality but all other features are essentially the same.
451:
452:
453:
454: \section{Defects}
455:
456: The transition from the homogenous to the modulated state involves
457: spontaneous breakdown of translational symmetry, and upon global
458: initiation one may expect domain walls, or kinks, that separate
459: spatial regions with different order parameter. The presence of
460: these defect and their character is crucial for the understanding
461: of the system response functions, e.g., its behavior under small
462: noise: as there is no preference to one phase of the order
463: parameter the kinks may move freely, while the "bulk" of the
464: domain is much more stiff. In the following paragraphs the
465: characteristic defects for various phases are presented.
466:
467:
468: \subsection{Domain walls}
469: As mentioned above, the nearest neighbors competition leads, above
470: the bifurcation threshold, to to appearance of an up-down
471: modulation ($k=\pi$), and if there is no diffusion the steady
472: state is the 0101010 configuration. Clearly there are two
473: equivalent segregation of this type, namely, filled odd sites and
474: empty even sites and vise versa. Accordingly, in case of global
475: initiation (random "seeds" are spread all around) one finds
476: domains of the stable patterns with different parity, and domain
477: walls (technically known as kinks or solitons) that separate these
478: regions, as seen in figure (\ref{kink}). The nearest neighbor
479: interaction is simple enough to allow for an analytic solution for
480: the kink, and the numerical results confirm the predictions
481: \cite{shnerb}.
482:
483:
484: \begin{figure}
485: % Requires \usepackage{graphicx}
486: \includegraphics[width=7.7cm]{fig2nadav.eps}\\
487: \caption{A typical kink of length $L=20$, an outcome of forward Euler integration of Eq. (\ref{eq:5})
488: (with n.n. competition) on 1024 lattice points with periodic boundary conditions and random initial conditions at
489: $\gamma = 0.505$ (just above the bifurcation). } \label{kink}
490: \end{figure}
491:
492: In the presence of diffusion there is a "smearing" of the above
493: results: the homogenous state is stable for larger $\gamma$, and
494: above the segregation threshold the steady state is "smeared" from
495: $...01010...$ to an "up - down - up -down" form, and the kinks are
496: not of finite size but admit exponentially decaying tails. See
497: \cite{shnerb} for details.
498:
499:
500: \subsection{Phase shift}
501:
502: \noindent Unlike the nearest neighbor case, competition of longer
503: range leads to instabilities with wavelength of more than one
504: site, i.e., $c_n=A_0+A_k cos(n k l_0)$ with general $k$. This
505: opens the problem of defects between ordered regions. Inspired by
506: the nearest neighbors example one may expect another types of
507: kinks that separate different regions of ordered state.
508: Surprisingly, this is not the case. Instead of getting kinks
509: between different oriented regions of the activated wavenumber,
510: one gets \emph{single} oriented region with \emph{phase shift},
511: namely, the spatial structure is of the form $c_n=A_0+ B cos(n k
512: l_0+\phi)$, where $\phi$ is the phase shift between the actual
513: solution and the predicted modulation $c_n=A_0+A_k cos(n k l_0)$
514: and $B = A_k/cos \phi$.
515:
516: \begin{figure}
517: % Requires \usepackage{graphicx}
518: \includegraphics[width=7cm]{unit_cycle.eps}\\
519: \caption{ The set of different values of population size at different site is
520: presented on the unit circle where the linear analysis predicts
521: an instability with wavenumber $k=3\pi/5l_0$.
522: The filled circles are the values of $cos(3n \pi/5l_0)$, the solution predicted by the
523: naive argument, and this is indeed a stable solution with finite
524: basin of attraction [see Figure (\ref{pure})].
525: It turns out, however, that generic initial conditions flow into a phase shifted solution where the
526: population of the form $cos(3n\pi/5l_0+\phi)$ (shown in Figure (\ref{shift}).
527: The value of $\phi$ is half of the angular distance between two close
528: sites, here corresponds to the open triangles on the unit cycle.
529: }\label{cycle}
530: \end{figure}
531:
532: On the unit cycle (Figure [\ref{cycle}]) the meaning of this
533: additional phase is a shift of all point by $\phi$. This shift
534: may reduce the number distinct values in one cycle by one, as
535: indicated in the example of Figure (\ref{cycle}): here, instead
536: of six distinct values taken by $c_n$ along one wavelength,
537: there are only five. Both numerical simulations of the system
538: dynamics, starting from random initial conditions, and stability
539: analyzes of the possible steady state for arbitrary $\phi$
540: indicates that, although any $\phi$ corresponds to locally stable
541: solution, the most stable $\phi$ equals to half of the angular
542: distance between two adjusting points on the circle. In Figure
543: (\ref{cycle}) the actual phase shifted pattern is shown for $k=3
544: \pi /5 l_0$, while Figure (\ref{stability}) indicates that the
545: most stable phase corresponds to $\phi/\phi_0 = 1/2$. As the
546: Lyapunov exponent of any $\phi$ is negative, small perturbations
547: around any $\phi$ value (in particular, $\phi =0$) decay. Figure
548: (\ref{pure}) shows the corresponding stable mode with $\phi=0$
549: where the initial conditions are small perturbation around it.
550: Figure (\ref{shift}), on the other hand, shows the final state
551: with generic initial conditions, where the system flows to the
552: most stable pattern with $\phi/\phi_0 = 1/2$.
553:
554:
555: \begin{figure}
556: % Requires \usepackage{graphicx}
557: \includegraphics[width=7cm]{stability_analysis.EPS}\\
558: \caption{The Lyapunov exponent (in arbitrary units) for states of the form $cos(nk+\phi)$
559: is shown against $\phi$ (in units of $k/2$) for various wavenumbers. While the steady state is
560: stable for any $\phi$, the most stable state corresponds to $k/2$, and }\label{stability}
561: \end{figure}
562:
563: \begin{figure}
564: % Requires \usepackage{graphicx}
565: \includegraphics[width=7cm]{pure_mode.EPS}\\
566: \caption{The steady state for the same parameters of figure (\ref{cycle}), where the initial
567: conditions are close to the $\phi=0$ solution, i.e.,
568: $c_n(t=0) = A_0+A_k cos(3\pi n/5) + \delta_n$, where $\delta_n$ is a small random number.
569: The system flows into the $\phi = 0$ case, in agreement with the local stability analysis
570: presented in
571: Figure (\ref{stability}). }\label{pure}
572: \end{figure}
573:
574: \begin{figure}
575: % Requires \usepackage{graphicx}
576: \includegraphics[width=7cm]{mode_plus_phase.EPS}\\
577: \caption{Same as Figure (\ref{pure}), but now the initial conditions are generic $c_n(t=0)=\delta_n$.
578: The system flows to the most stable steady state that corresponds, in this case, to
579: $c_n=A_0+B cos(3 n \pi/5 + \phi)$, with $\phi = 3 \pi / 10$. }\label{shift}
580: \end{figure}
581:
582:
583:
584:
585: \section{The spiky phase}
586:
587: Deep in region III of the phase diagram [Figure (\ref{fig3})] many
588: wavevectors are excited, with strong mode competition between
589: them, and the linear analysis picture based on Fourier
590: decomposition becomes ineffective. Better insight into the system
591: comes from a real space analysis: deep into region III the long
592: range competition is strong, and within the effective interaction
593: range new colony can not develop in the presence of a fully grown
594: one. Accordingly, this phase is characterized by fully developed
595: colonies separated by "dead regions" of constant length that
596: reflects the effective interaction length. In Fourier space, this
597: corresponds to many active modes that build together a periodic
598: structure of "bumps".
599:
600: In case of global initiation, of course, defects may appear in the
601: stable steady state as the system flows to different order
602: parameter in different regions. Again, it is better to use the
603: real space picture in order to describe these defects. The
604: situation is close to what observed in the case of random
605: sequential adsorption \cite{ads,lavee}: while an "optimal"
606: filling of the system admits a periodic structure of living
607: patches with periodicity of, say, $L$ lattice points, it may
608: happens that the distance between two fully developed sites is
609: between L and 2L, and all the site in between should remain empty
610: due to the long range competition. The emerging spatial
611: configuration is of ordered regions (with coherence size that
612: depends upon the dynamics) separated by "domain walls", where the
613: width of these walls is taken from some distribution function
614: between zero and the interaction effective length.
615:
616: \section{Two dimensional system}
617:
618: Although all the analysis presented was in one dimension, the
619: basic picture is the same for higher dimensionality. In
620: particular, the bifurcation condition is similar, nearest neighbor
621: interactions yields a "checkerboard" phase above the bifurcation
622: line, and the spiky phases is also observed.
623:
624: For nearest neighbors interaction kinks between different regions
625: (checkerboard parity) occurs. Because of the two dimensionality of
626: the lattice the kinks might have any arbitrary spatial line,
627: rather then straight line, as shown at figure (\ref{kink1d}).
628: Those kinks are de-facto one dimensional topological defects,
629: because of the periodic boundary conditions. On the other hand,
630: the domain walls of Figure (\ref{kink2d}) seems to admit a real 2d
631: features, although their topological character is not clear.
632:
633: \begin{figure}
634: % Requires \usepackage{graphicx}
635: \includegraphics[width=7cm]{Two_d_one_dimension_defects.eps}\\
636: \caption{Spatial domains in two dimensional system, for logistic growth with nearest neighbors
637: competition. The parameters are chosen to be above the bifurcation threshold, and the stable steady
638: state is a "checkerboard" with alternating filled and empty sites. Denoting a site by its coordinates $i,j$, there are
639: two possible phases of the solution, correspond to filled $i+j \ \ odd$, empty $i+j \ \ even$ and vice versa.
640: Here, the results of an Euler integration of the process for a $2d$ sample of $50x50$ sites with
641: periodic boundary conditions is presented, where only the kinks separating regions of different order parameters are
642: colored. The kinks here are non-contractible on the torus and correspond to one dimensional topological defects.
643: The simulation parameters are $D=0, \ \gamma_1 = 0.2505$. Initial conditions are "seed" population at each site taken
644: randomly from a square distribution between $[0,0.01]$. }\label{kink1d}
645: \end{figure}
646:
647: \begin{figure}
648: % Requires \usepackage{graphicx}
649: \includegraphics[width=7cm]{Two_d_two_dimension_defects.eps}\\
650: \caption{The same system and parameters as in Figure (\ref{kink1d}), for other choice of random initial
651: conditions. Here the domain wall
652: is contractible on the torus and the order parameter phases are different
653: between the inside and the outside of the kink. }\label{kink2d}
654: \end{figure}
655:
656: \section{Global properties}
657:
658: \subsection{Upper critical diffusion}
659:
660: Let us turn back to the bifurcation condition, Eq. (\ref{bif}), in
661: different representation:
662: \begin{equation}\label{bif1}
663: \frac{\beta_k}{\beta_0} + 2 D [1-cos(kl_0)]< 0
664: \end{equation}
665: where the $k$ considered is the one for which $\beta_k$ admits a
666: global minimum. Clearly, this $k_{min}$ depends only on the form
667: of the interaction kernel and is independent of its strength (if
668: one multiplies all $\gamma_r$ by constant factor, the value of
669: $k_{min}$ remains the same). Since the negative term in the
670: instability condition $\frac{\beta_k}{\beta_0}$ cannot exceed
671: $(-1)$, the absolute value of the right hand term should be even
672: smaller to allow a periodic modulation of the stable steady state.
673:
674: Assume, now, that the wavelength of the modulation is much larger
675: than the lattice constant (as already required as one approaches
676: the the continuum limit). In that case the approximation $2D
677: [1-cos(kl_0)] \approx D k_{min}^2 l_0^2$ holds, and since $D =
678: (W_F/l_0)^2$, this term is proportional to $(W_F/\lambda)^2$,
679: where $\lambda$ is the period of the modulation. This implies
680: that, independent of the strength of the long-range competition,
681: \emph{bifurcation never takes place if the width of the Fisher
682: front is larger than the period of the modulation}. This statement
683: holds up to a numerical factor (between zero and one) which
684: determined by the form of the competition kernel.
685:
686: Simple example that demonstrate these considerations is the case
687: of nearest neighbors interaction. Here
688: \begin{equation}
689: g(k)=1+2\gamma cos(k)+2(1+2\gamma)D[1-cos(kl_0)]
690: \end{equation}
691: and the global minima is $k=\pi$. $g(k_{min})$ is
692: \begin{equation}
693: g(\pi)=1-2\gamma+4(1+2\gamma)D
694: \end{equation}
695: so for any $\gamma$ there is an upper critical D
696: \begin{equation}
697: D_c=\frac{2\gamma-1}{4(1+2\gamma)}
698: \end{equation}
699: above which no bifurcation takes place. This upper critical
700: diffusion constant converges to a global value as $\gamma \to
701: \infty$
702: \begin{equation}
703: D_c^g \equiv {D_c,}_{\gamma->\infty} = \frac{1}{4}.
704: \end{equation}
705: and no bifurcation takes place when the width of the Fisher front
706: is of order of the modulation length. Intuitively this result may
707: be understood as follows: suppose that the system is in its 010101
708: state, and suppose that the dynamics is discrete in time. If
709: $D=1/4$ it implies that each filled site contribute $1/4$ to any
710: of its neighbors, and then the system is frozen in its homogenous
711: state with amplitude $1/2$ at each site. Generalizing this
712: intuition to periodic modulation of arbitrary wavelength yields
713: the same result, where the Fisher front width stands as a
714: definition of an "effective site".
715:
716:
717: \subsection{Spatial segregation and total population}
718:
719: Given a system with long range competition, one may ask how the
720: \emph{total} population (integrated over all the spatial domain)
721: or the average population density, depend on the phase of the
722: system. Naively one expects the total population to grow with the
723: diffusion constant, as faster spatial wandering helps an
724: individual reactant to escape from the depleted region of an
725: already existing colony. This, however, is not the case, as
726: pointed out by \cite{Birch} and \cite{Garcia}: the size of the
727: total population depends on the efficiency of segregation: strong
728: segregation implies higher population (on average, since there are
729: empty regions and living patches). Thus, decrease of diffusion
730: implies higher total population density.
731:
732: Clearly, the total population is given by the amplitude of the
733: zero mode in Fourier decomposition of the population, (See Eq.
734: \ref{eq3}). As long as the system is in its homogenous phase this
735: quantity is diffusion independent and the total population depends
736: only on the strength of the interaction, $A_0=1/\beta_0$. Right
737: above the bifurcation, when only one excited mode ($k$) exists,
738: the total population is proportional to $A_0 = \alpha_k/(\beta_0 +
739: \beta_k)$, and since $\alpha_k$ increases as $D$ decreases, so is
740: the total population. In the case of one dimensional lattice with
741: nearest neighbors interaction, for example, the dependence of the
742: total occupancy of the sample on the diffusion constant may be
743: calculated explicitly, since there is only one excited mode
744: $k=\pi/l_0$. Here even far from the bifurcation point the
745: amplitude of the zero mode is given by $A_0 = \alpha_k/(\beta_0 +
746: \beta_k)$. The total sum versus diffusion is, accordingly,
747: \begin{equation}
748: A_0 = \left\{ \begin{array}{cc}
749: (1-4D)/2 & D < D_c \\
750: 1/(1+2 \gamma) & D > D_c.
751: \end{array} \right.
752: \end{equation}
753:
754: Figure (\ref{sum}) shows the total sum versus diffusion for few
755: situations. The numerical results indicate that the decay of
756: average population is approximately linear. Note that, for the
757: "top hat" competition presented here, there seems to be a
758: discontinuity at $D_c$ in two dimensions, while in $1d$ the total
759: population is continuous at the transition.
760:
761:
762: \begin{figure}
763: % Requires \usepackage{graphicx}
764: \includegraphics[width=7cm]{Total_sum_normalized.EPS}\\
765: \caption{Total population versus diffusion coefficient for several situations. (A): $1d$ with nearest
766: neighbors interaction (squares). (C): $1d$, n.n.n. interaction (circles). (B): $2d$,
767: top hat interaction (triangles). The "top hat" is constant interaction with all sites inside a circle of radius $3
768: l_0$, and zero outside. In order to present all the results in the
769: same panel, the population has been normalized, for each system,
770: by its homogenous solution. }\label{sum}
771: \end{figure}
772:
773:
774:
775:
776: \section{Local initiation: the dynamics of invasion and segregation}
777:
778: Along this paper, an analysis of the stable steady states of the
779: logistic growth with long range competition was presented. As few
780: stable steady solutions may exist simultaneity for the same set of
781: parameters, the generic situation was identified numerically using
782: global initiation, i.e., small random population at each site. In
783: this section, the dynamics of growth is analyzed, where the
784: initial conditions are a colony with compact support. For local
785: logistic growth this problem was considered years ago by Fisher
786: \cite{fisher} and Kolomogorov \cite{kol}. The invasion of the
787: stable solution into the unstable one takes place via a front (the
788: Fisher front) that propagates in constant velocity. This problem
789: was considered by many authors in different contexts and was
790: generalized to other cases of invasion into unstable state, see
791: comprehensive review by van Saarloos \cite{saarloos}.
792:
793:
794: As emphasized above, the system considered here may admit [in
795: region II and III of figure (\ref{fig3})] two instabilities: the
796: empty state is unstable against the homogenous one, while the
797: homogenous solution breaks and yields a spatial modulation.
798: Accordingly, if the system is initiated locally from a small
799: colony of compact support one expects that two fronts propagate
800: into the empty region: first the front associated with the
801: homogenous state, and then the modulation (secondary instability)
802: front \cite{hohenberg}. These two fronts travel in different
803: velocities. Generally, it is known that the Fisher velocity is
804: determined by the leading edge ("pulled" fronts) and is related
805: to the Lyapunov exponent that characterizes the relevant
806: instability. Accordingly, the dynamics of our system is determined
807: by two velocities: $v_p$, the velocity of the primary front
808: (that interpolates between the empty and the homogenous state) and
809: the modulation velocity $v_s$. While $v_p$ is $\gamma$
810: independent, the secondary front velocity $v_s$ depends on the
811: characteristics of the long range competition. By tuning of
812: $\gamma$, though, one may change the relative velocity between the
813: primary and the secondary front. Both velocities may be calculated
814: analytically using a saddle point method and taking into account
815: the discreteness of the lattice points, as discussed in Appendix
816: A. Generically, there are two possible scenarios for the takeover
817: of an empty region by spatially modulated steady state: in the
818: first case $v_p
819: > v_s$ (see Figure \ref{vp}) and the homogenous region between the
820: primary and the secondary front grows linearly in time. This
821: situation is very sensitive, as small perturbations (induced by
822: the leading front) lead to spontaneous bifurcation of the
823: homogenous region, a process that yields many structural defects
824: (e.g., kinks) along the chain.
825:
826:
827: \begin{figure}
828: % Requires \usepackage{graphicx}
829: \includegraphics[width=7cm]{Two_invasion_waves.eps}\\
830: \caption{Snapshot of the one dimensional system, initiated locally from the left, where the primary
831: velocity is higher then the secondary velocity. The two fronts are clearly shown, and the homogenous
832: region between them is widening as time elapsed. The simulation assumes nearest neighbors competition with:
833: D=0.04 and $\gamma=0.8$. Along time, defects (not shown) are
834: generated at the tip of the secondary front due to the noise
835: induced by the primary front.
836: }\label{vp}
837: \end{figure}
838:
839: In the second case the situation is different: if $v_p < v_s$
840: there is no homogenous region, and only one front exists. Its
841: velocity is determined, of course, by the primary front velocity,
842: but its shape is different (see Figure \ref{vs}). In that case
843: the sensitive homogenous region never exists, and the pattern
844: formation process is robust, with no defects associated with the
845: front kinetics.
846:
847: \begin{figure}
848: % Requires \usepackage{graphicx}
849: \includegraphics[width=7cm]{secondary_fisher_wave_alone.eps}\\
850: \caption{Same as Figure \ref{vp}, but now the velocity of the secondary front is
851: higher than the velocity of the primary front. Since the secondary instability may appear only
852: after the primary, the velocity of the whole front is determined by $v_p$.
853: The parameters used are: D=0.04, $ \gamma=2 $. } \label{vs}
854: \end{figure}
855:
856:
857:
858:
859:
860:
861: \section{Conclusions and remarks}
862:
863:
864: This paper attempt to present the various phases associated with
865: the steady states of the logistic process on spatial domains with
866: nonlocal competition. The main feature is, of course, the
867: segregation transition that happens, as was shown, where the width
868: of the Fisher front (associated with the homogenous solution)
869: becomes shorter than the instability wavelength. Right above the
870: bifurcation one finds a pattern dominated by a single wavelength,
871: while far away from the bifurcation line the stable steady state
872: becomes spiky. Each phase is associated with its own defects:
873: phase shift close to the bifurcation, empty regions in the spiky
874: phase, and domain walls (kinks) for the up-down phase of the
875: nearest-neighbors interaction. It turns out that the segregation
876: transition increase the overall carrying capacity per unit volume.
877: In $1d$ the population is continues at the transition while in two
878: dimensions discontinuity might occur.
879:
880: Upon local initiation the system dynamics is governed by the
881: relations between the velocities of the primary (empty to
882: homogenous) and the secondary (homogenous to modulated) fronts.
883: The numerics suggests that, while global initiation may yield
884: "disordered" structure with many defects per unit length, local
885: initiation with the same parameters yields ordered structure
886: unless the secondary front velocity is smaller that the primary
887: one.
888:
889:
890: While in this work only rate equations of reaction-diffusion type
891: has been considered, in recent numerical works of Birch and Young
892: \cite{Birch} and Garcia et. al. \cite{Garcia} the stochastic
893: motion of the individual reactants is taken into account. These
894: stochastic models add two ingredients to the description presented
895: here. First, the introduction of individual reactants ("Brownian
896: bugs" \cite{young2}) implies a \emph{threshold} on the reactant
897: concentration on single patch. Secund, there is a multiplicative
898: noise associated with the stochastic motion of individual
899: reactants. As shown in this work, many of the features associated
900: with long range competition are independent of the discrete nature
901: of individual reactants.
902:
903:
904:
905: \section{acknowledgements}
906: The authors thank Prof. David Kessler for many helpful
907: discussions. This work was supported by the Israeli Science
908: Foundation, grant no. 281/03, and by Yeshaya Horowitz Fellowship.
909:
910: \section{Appendix A}
911:
912: In this appendix the analytic expression for the secondary front
913: velocity on a discrete lattice is obtained, via the saddle point
914: argument (see \cite{levin}). For the sake of simplicity, only the
915: case of nearest neighbors interaction is considered. In order to
916: preform the same calculations for competition beyond the n.n.
917: limit, one should first find numerically the steady state
918: modulation and then follow the same procedure.
919:
920: The evolution of a population is given by:
921: \begin{equation}\label{21}
922: \frac{\partial c_n}{\partial t}= D [-2c_n+c_{n+1}+c_{n-1}] + c_n -
923: c_n^2 + c_n\gamma(c_{n+1}+c_{n-1}).
924: \end{equation}
925: Denoting by $\delta_n$ the deviations from the homogenous
926: solution, $c_n=A_0+\delta_n$, Eq. (\ref{21}) is linearized to
927: yield:
928: \begin{equation}\label{22}
929: \frac{\partial \delta_n}{\partial t}= \alpha \delta_n +\beta
930: (\delta_{n+1}+\delta_{n-1})
931: \end{equation}
932: Where $\alpha= a-2bA_0+2A_0\gamma-2D/l_0^2$ and $
933: \beta=D/l_0^2-A_0\gamma$. Assuming modulated solution of the
934: form,
935: \begin{equation} \label{23}
936: \delta_n = \left\{ \begin{array}{cc}
937: Ae^{ikl_0n+\Gamma(k)t} & \ \ n \ odd \\
938: Be^{ikl_0n+\Gamma(k)t} & \ \ n \ even \\
939: \end{array} \right.
940: \end{equation}
941: and plugging (\ref{23}) into (\ref{22}) one gets
942: \begin{equation}
943: \Gamma(k) \left[ \begin{array}{c}
944: A \\
945: B \\
946: \end{array} \right] =
947: \left[ \begin{array}{cc}
948: \alpha & \beta \cos(kl_o) \\
949: \beta \cos(kl_o) & \alpha \\
950: \end{array}\right]\left[ \begin{array}{c}
951: A \\
952: B \\
953: \end{array} \right].
954: \end{equation}
955: The dispersion relations is given by:
956: \begin{equation}
957: \Gamma(k)=\alpha + \beta \cos(kl_0).
958: \end{equation}
959: where the plus sign is chosen for the unstable modes. The
960: solutions are of the form
961: \begin{equation}
962: \left[ \begin{array}{c}
963: c_n \\
964: c_{n+1} \\
965: \end{array} \right] =\left[ \begin{array}{c}
966: A \\
967: B \\
968: \end{array} \right] e^{ikx+\Gamma(k)t}.
969: \end{equation}
970:
971: If a solution represents a travelling front with velocity $v$ it
972: is useful to define the coordinate system in the moving frame,
973: $\zeta=x-vt$, to get
974: \begin{equation}
975: \left[ \begin{array}{c}
976: c_n \\
977: c_{n+1} \\
978: \end{array} \right] = \left[ \begin{array}{c}
979: A \\
980: B \\
981: \end{array}\right] e^{ik\zeta+ikvt+\Gamma(k)t}.
982: \end{equation}
983: Using the saddle point method \cite{saarloos} the two equations
984: that determine the velocity are
985: \begin{equation}\label{33}
986: f \equiv ivk+\alpha+2\beta cosh(kl_0)=0
987: \end{equation}
988: and
989: \begin{equation}\label{34}
990: \frac{\partial f}{\partial k} = iv+ 2 \beta_0 sinh(kl_0) = 0.
991: \end{equation}
992: In case of finite time steps one should replace $ivk$ by $
993: (e^{-ikv dt}-1)/ dt$ to get the appropriate corrections. Figure
994: (\ref{vp2}) shows the perfect fit between the solution of
995: (\ref{33}) and (\ref{34}) and the numerical solution.
996:
997: \begin{figure}
998: % Requires \usepackage{graphicx}
999: \includegraphics[width=7cm]{Secondary_fisher_wave.EPS}\\
1000: \caption{Comparison of the numerical simulation (triangles) and the
1001: theoretical prediction based on the saddle point method [Eqs. (\ref{33}) and (\ref{34}), solid line]
1002: for the velocity of the secondary front as a function of the interaction strength. The diffusion used is
1003: $D=0.04$ and the lattice constant is $l_0 =1$, $dt = 0.01$ }\label{vp2}
1004: \end{figure}
1005:
1006: \section{Acknowledgements}
1007:
1008: We thank Prof. David Kessler for many helpful discussions and
1009: comments. This work was supported by the Israel Science Foundation
1010: (grant no. 281/03) and by the Yeshaya Horowitz Fellowship.
1011:
1012:
1013: \begin{thebibliography}{}
1014:
1015: \bibitem{Sayama} H. Sayama, M.A.M. de Aguiar, Y. Bar-Yam and M.
1016: Baranger, Phys. Rev. {\bf E 65}, 051919 (2002).
1017: \bibitem{Fuentes} M. A. Fuentes, M. N. Kuperman and V.M. Kenkre, Phys. Rev. Lett. \textbf{91}, 158104 (2003).
1018: \bibitem{shnerb} N.M.Shnerb, Phys. Rev. \textbf{E 69}, 061917
1019: (2004).
1020: \bibitem{Garcia} E. H. Garcia and C. Lopez. , Phys. Rev. \textbf{E 70}, 016216 (2004).
1021: \bibitem{Sakai} A. Sakai, Jour. Theor. Biol. \textbf{186 }, 415 (1997).
1022: \bibitem{Doebeli} M. Doebeli and T. Killingback, Theo. Population Biology \textbf{64}, 397
1023: (2003).
1024: \bibitem{Bolker} B.M. Bolker, Theoretical Population Biology \textbf{64}, 255
1025: (2003).
1026: \bibitem{Tokita} K. Tokita and A. Yasutomi, Theoretical Population Biology \textbf{63}, 131
1027: (2003).
1028: \bibitem{Anderson} K. Anderson and C. Neuhauser, Ecological modeling
1029: \textbf{155}, 19 (2002).
1030: \bibitem{Hoopes} F. Hoops at el., Jour. Theor. Biol. \textbf{210},
1031: 201 (2001).
1032: \bibitem{Birch} D. Birch , W.R. Young, Private communication.
1033: \bibitem{fisher} R. A. Fisher, Ann. Eugenics \textbf{7}, 353, (1937).
1034: \bibitem{kol} A. Kolomogoroff, I. Petrovsky and N. Piscounoff,
1035: Moscow Univ. Bull. Math. \textbf{1},1 (1937).
1036: \bibitem{derrida} E. Brunet, B. Derrida, J. Stat. Phys. \textbf{103}, 269-282
1037: (2001).
1038: \bibitem{saarloos} See, e.g., W. van Saarloos, Physics Reports {\bf 386},
1039: 29,(2003).
1040: \bibitem{merron} J.B. Wilson and A.D.Q. Agnew, Adv. Ecol. Res. {\bf 23}, 263 (1992); R. Lefever and
1041: O. Lejeune, Bull. Math. Biol. {\bf 59}, 263 (1997); J. von
1042: Hardenberg, E. Meron, M. Shachak and Y. Zarmi, Phys. Rev. Lett.
1043: {\bf 87} 198101 (2001).
1044: \bibitem{lavee} N.M. Shnerb, P. Sarah, H
1045: Lavee and S. Solomon, Phys. Rev. Lett. 90, 038101 (2003).
1046: \bibitem{huisman} F.J. Weissing and J. Huisman, Jour. Theo.
1047: Biol. \textbf{168}, 323 (1994).
1048: \bibitem{bak} P. Bak and R. Bruinsma, Phys. Rev. Lett. \textbf{49}, 249
1049: (1982).
1050: \bibitem{ns} D.R. Nelson and N.M. Shnerb, Phys. Rev. \textbf{E 58}, 1383
1051: (1998), Appendix B.
1052: \bibitem{ads} See, e.g., J. Talbot, G. Tarjus, P. R.Van-Tassel, and P.Viot,
1053: Colloids Surf. \textbf{A 165}, 287 (2000), and references therein.
1054: \bibitem{Garcia2} E. H. Garcia, C. Lopez. , Physica. \textbf{D 199}, 223-234
1055: (2004).
1056: \bibitem{hohenberg} See, e.g., M.C. Cross and P. Hohenberg, Rev. Mod. Phys. \textbf{65}, 3, 851
1057: (1993) and references therin.
1058: \bibitem{young2} W.R. Young , A.J. Roberts and G. Stuhne , Nature
1059: \textbf{412} 328 (2001).
1060: \bibitem{levin} L. Pechenik, H Levine. cond-mat/9811020,
1061: (1998).
1062:
1063:
1064: \end{thebibliography}
1065:
1066: \end{document}
1067: