1: \documentclass[pre,letterpaper]{revtex4}
2: %\usepackage{ifpdf}
3: %\ifpdf
4: %\usepackage[pdftex]{graphicx}
5: %\DeclareGraphicsExtensions{.pdf,.PDF}
6: %\usepackage{hyperref}
7: %\else
8: \usepackage[dvips]{graphicx}
9: %\DeclareGraphicsExtensions{.eps,.EPS}
10: %\fi
11: \setkeys{Gin}{width=0.4\columnwidth}
12: \usepackage{amsmath}
13:
14: \newcommand{\dparcial}[2]{\frac{\partial\,#1}{\partial#2}}
15: \newcommand{\ddparcial}[2]{\frac{\partial^2\,#1}{\partial#2^2}}
16: \newcommand{\dtotal}[2]{\frac{d\,#1}{d #2}}
17:
18: %\newcommand{\consortium}{\affiliation{\mbox{Consortium of the Americas
19: % for Interdisciplinary Science and Department of Physics and
20: % Astronomy}, \mbox{University of New Mexico, Albuquerque, New
21: % Mexico 87131}}}
22:
23: \begin{document}
24: \title{Study of Arbitrary Nonlinearities in Convective Population Dynamics
25: with Small Diffusion}
26: \author{I. D. Peixoto}
27: \author{L. Giuggioli}
28: \author{V. M. Kenkre}
29: \affiliation{Consortium of the Americas
30: for Interdisciplinary Science and Department of Physics and
31: Astronomy, \\ University of New Mexico, Albuquerque, New
32: Mexico 87131}
33:
34: %\consortium
35:
36: %%%%%% ABSTRACT
37: \begin{abstract}
38: Convective counterparts of variants of the nonlinear Fisher equation
39: which describes reaction diffusion systems in population dynamics
40: are studied with the help of an analytic prescription and shown to
41: lead to interesting consequences for the evolution of
42: population densities. The initial value problem is solved explicitly
43: for some cases and for others it is shown how to find traveling wave
44: solutions analytically. The effect of adding diffusion to the convective
45: equations is first studied through exact analysis through a piecewise
46: linear representation of the nonlinearity. Using an appropriate small
47: parameter suggested by that analysis, a perturbative treatment is developed
48: to treat the case in which the convective evolution is augmented
49: by a small amount of diffusion.
50: \end{abstract}
51: \pacs{87.17.Aa, 87.17.Ee, 87.17.Jj}
52:
53: \date{\today} \maketitle
54:
55: \section{Introduction}
56: The Fisher equation, proposed originally to describe the spread of an
57: advantageous gene in a population \cite{fisher1937}, has found a
58: great deal of use in mathematical ecology, directly
59: \cite{murray,skellam,kk,bkk} as well as indirectly,
60: i.e., in modified forms. Examples of modifications are the
61: incorporation of internal states such as those signifying the presence
62: or absence of infection \cite{ak,akyp,kpasi},
63: the
64: introduction of temporal nonlocality in the diffusive terms
65: \cite{manne,abk}, of spatial nonlocality in the competition
66: interaction terms \cite{fkk,fk}, and the addition of convective terms
67: signifying `wind effects' \cite{nelson,ana,gk,lgthesis,k}. Most of
68: the modifications lead to equations which, as in the case of the
69: original Fisher equation, are not soluble analytically and require
70: approximate or numerical methods for their analysis. There is,
71: however, one modification \cite{gk}, that leads to an analytic
72: solution via a trivial transformation, and to the possibility of the
73: extraction of interesting information of potential use to topics such
74: as bacterial population dynamics. That modification consists of the
75: replacement of the diffusive term in the Fisher equation
76: \begin{equation} \label{eq:fisher}
77: \dparcial{u}{t}= a u \left(1-\frac{u}{K}\right) + D \ddparcial{u}{x}
78: \end{equation}
79: by a convective term. As is well-known, in the original Fisher
80: equation, $u$ represents a concentration or density, $a$ the growth
81: rate, $K$ the so called `carrying capacity', and $D$ the diffusion
82: constant. In spite of the fact that purely convective (non-diffusive)
83: nonlinear partial differential equations are much less sophisticated
84: than their diffusive counterparts, there are at least two reasons why
85: their study is important. There are clear physical situations
86: \cite{nelson,ana,gk} in which wind effects which add the convective
87: term to a diffusive equation are present. These arise, for instance,
88: when one changes one's reference frame to that of a moving mask in
89: bacterial population dynamics experiments \cite{nelson}. Because the
90: convective term comes from the macroscopic motion of the mask under
91: experimental control whereas the diffusive term comes from the
92: microscopic motion of the bacteria, it is possible to arrange the
93: system experimentally so that the diffusion effects are relatively
94: unimportant. This may be achieved for instance, by using viscous
95: environments, or by genetically engineering the (bacterial)
96: population. In such a case one arrives at the analysis of a largely
97: convective (negligibly diffusive) nonlinear evolution which may be
98: investigated in zeroth order as a purely convective equation
99: \cite{gk,k}. This fact, that experimental realization of the
100: convective nonlinear equation is indeed possible, provides one
101: motivation for these studies. The other, as explained in Ref.
102: \cite{gk}, is that it is possible to develop perturbation schemes
103: starting from the soluble convective equation to incorporate diffusive
104: effects. In Sections \ref{sec:application} and \ref{sec:travelfront},
105: we develop the theory for arbitrary nonlinearities in the absence of
106: the diffusion constant, and in Section \ref{sec:diffusion},
107: we focus on the effect of
108: adding the diffusive term.
109:
110: Recently, a simple prescription for generalizing the analysis of Ref.
111: \cite{gk} was given by one of the present authors \cite{k} to
112: arbitrary nonlinearities in the convective equation. The equation
113: considered there is
114: \begin{equation} \label{problem}
115: \dparcial{u}{t} + v \dparcial{u}{x} = auf(u),
116: \end{equation}
117: where $v$ is the medium or `wind' speed, $a$ is a growth rate and
118: $f(u)$ is a dimensionless \emph{arbitrary} nonlinearity. The
119: prescription provided in that analysis gives the full time and space
120: evolution of $u(x,t)$ in terms of its initial distribution
121: $u(x,0)=u_0(x)$, through the relation
122: \begin{equation} \label{solution}
123: u(x,t) = G^{-1} \big[G[u_0(x - v t)] - at\big]
124: \end{equation}
125: where the function $G(u)$ is obtained from the nonlinearity $f(u)$ by
126: the integration \cite{k}
127: \begin{equation} \label{eq:gdef}
128: G(u) = - \int \frac{1}{uf(u)} du.
129: \end{equation}
130: We have used a slightly changed notation here relative to ref.
131: \cite{k}. Our functions $G(u)$ and $f(u)$ are both dimensionless in
132: contrast to the dimensioned counterparts $g(u)$ and $F(u)$ in ref.
133: \cite{k}.
134:
135: In this paper we present several applications of the prescription
136: (\ref{solution}) and report a number of additional results from the
137: nonlinear convective equation (\ref{problem}). In Section \ref{sec:application}, we
138: examine features of the solutions obtained exactly from the
139: prescription for several cases of the nonlinearity including an
140: interesting `pyramid effect' that occurs for nonlinearity functions
141: $f(u)$ with \emph{multiple zeros}. In Section \ref{sec:travelfront}, we develop a general
142: analysis for finding traveling wave solutions, and to study
143: characteristics of the tails and shoulders of such solutions.
144: Applications of this analysis for the nonlinearities studied in
145: Section \ref{sec:application} are also made in Section \ref{sec:travelfront}.
146: In Section \ref{sec:diffusion}, we analyze in
147: detail the perturbative problem obtained by adding a small amount
148: of diffusion to the nonlinear convective equation.
149: Conclusions are presented in Section \ref{sec:conclusions}.
150:
151:
152: \section{ Applications of the Prescription for the Solution of the
153: Initial Value Problem}
154: \label{sec:application}
155: Applications of the prescription
156: (\ref{solution}) for the solution of the arbitrary initial value
157: problem were indicated in ref. \cite{k} for three cases: the ordinary
158: logistic nonlinearity for which $f(u)$ has a single zero at $u=K$, the
159: Nagumo nonlinearity for which $f(u)$ has two zeros, and the
160: trigonometric nonlinearity for which $f(u)$ has multiple zeros. We
161: will display new results for the latter two, and analyze other
162: generalizations of the logistic nonlinearity. In some of these cases (subsections A,D and E below)
163: $u(x,t)$ is presented in an implicit form, whereas in the others
164: (subsections B and C) the form
165: is given explicitly.
166:
167: \subsection{Allee effect from Nagumo nonlinearity}
168: In (\ref{problem}) let $f(u)$ be given by
169: \begin{equation}\label{nagumof}
170: f(u)= \left(1 - \frac{u}{K}\right)
171: \left(\frac{u}{K}-\frac{A}{K}\right)
172: \end{equation}
173: where $A<K$. Such an $f(u)$ represents the so called `Allee effect'.
174: This effect \cite{allee} is associated with the existence of an
175: additional zero (fixed point) in the nonlinearity relative to the
176: logistic case. Unlike in the logistic case, the zero-$u$ solution is
177: stable here. If $u$ is small initially, it is attracted to the
178: vanishing value; if large, it is attracted to the non-zero value
179: $u=K$. The demarcation point is the additional zero introduced in
180: this case, viz. $u=A$. The physical origin of the Allee effect in population
181: dynamics is the possible increase of survival fitness as a function of population size for
182: low values of the latter. Existence of other members of the species may induce
183: individuals to live longer whereas low densities may, through loneliness, lead to extinction.
184: There is ample evidence for such an effeect in
185: nature \cite{courchamp, SS, fowlerbaker}.To construct $G(u)$ from (\ref{eq:gdef}) is
186: straightforward,
187: \begin{equation}
188: G(u)=\frac{1}{A(K-A)}
189: \ln\left(\frac{(K-u)^{A}u^{K-A}}{|u-A|^{K}}\right),
190: \end{equation}
191: but to invert $G(u)$ is not. In Fig.~(\ref{fig:nagumo}) the
192: nonlinearity is displayed along with the time evolution for a certain
193: initial condition. The Allee effect causes a decay to zero of parts of
194: the solution, while other parts evolve to the saturation value K.
195: Thus, an initial condition with $u$ values belonging to both regions,
196: will show a growth for those points in space such that $u_{0}(x)>A$,
197: while those points such that $u_{0}(x)<A$ will decay to zero. The
198: initial condition in this case evolves into a step profile of width
199: given by the separation in space of the two points where $u_{0}(x)=A$.
200: \begin{figure}[!htbp]\includegraphics[width=\columnwidth]{figure1}
201: \caption{The Nagumo nonlinearity of Eq.~(\ref{nagumof}) and its effects
202: on evolution. The nonlinearity is plotted in (a). Shown in (b) is an
203: example of the evolution from an initial condition $\textrm{sech}(x/\sigma)$
204: (shown by the dotted line) whose profile belongs both to the region
205: where $0<u(x)<A$ and where $A<u(x)<K$. The portion in the first
206: (lower) region decays while that in the second (upper) region grows
207: to the saturation value $u(x)=K$. Here, $\sigma$ is an arbitrary
208: length. }\label{fig:nagumo}
209: \end{figure}
210:
211: \subsection{Trigonometric nonlinearity and the Pyramid Effect}
212: What would happen if the nonlinearity function $uf(u)$ had multiple zeros
213: rather than three as in the Nagumo case? The physical origin of such
214: behavior may lie in nonmonotonic response of survival fitness to population size.
215: Complex biological and sociological interactions among members of the species
216: could exist on the basis of the coupling of mating instincts with mere size effects.
217: Such interactions could, in principle, make the survival fitness optimum at several, rather than a
218: single, values of the population size. In order to address such situations, extensions of the normal
219: Allee considerations are necessary. For this purpose, the discussion in A above may be generalized as follows. For a given
220: nonlinearity with an arbitrary number of zeros, $uf(u)$ can be
221: sectioned among each subsequent unstable zero. The dynamics of each
222: point of the initial profile will evolves according to which of these
223: sections it belongs initially. In the Nagumo case there are only two
224: such regions: the first region is between $u=0$ and $u=A$, while the
225: second region is for $u>A$. All the points of the initial condition
226: such that $u_{0}(x)<A$ decay to zero, while all those points belonging
227: to the second region grow to the saturation value $u=K$. But when
228: there is more than one unstable zeros, as in the case of a
229: trigonometric nonlinearity, the propagating solution may evolve into a
230: pyramidal profile. To show this we choose a sinusoidal nonlinearity
231: given by
232: \begin{equation} \label{nlpyramid}
233: f(u)=\frac{K}{\pi }\frac{\sin \left( \frac{\pi u}{K}\right) }{u}
234: \end{equation}
235: where we took the opportunity to change the stability of $u=0$ compared with the Nagumo case presented in A.
236: From (\ref{eq:gdef}), $G(u)$ is given by
237: \begin{equation}
238: G(u)=-\ln \left[ \tan \left( \frac{\pi u}{2K}\right) \right]
239: \end{equation}
240: and the exact solution at all time is
241: \begin{eqnarray}
242: u(x,t) &=&\sum_{m=1}^{+\infty }K\left( 1-\frac{2}{\pi }\arctan \left[
243: e^{-at}\cot \left( \frac{\pi }{2K}\left\{ \Theta \left[ \frac{2mK}{\pi }%
244: -u_{0}\left( x-vt\right) \right] \right. \right. \right. \right. \nonumber
245: \\
246: &&\left. \left. -\Theta \left[ \frac{\left( 2m-2\right) K}{\pi }-u_{0}\left(
247: x-vt\right) \right] u_{0}\left( x-vt\right) \right) \right] +\left(
248: m-1\right) \times \nonumber \\
249: &&\left. \left\{ \Theta \left[ \frac{2mK}{\pi }-u_{0}\left( x-vt\right)
250: \right] -\Theta \left[ \frac{\left( 2m-2\right) K}{\pi }-u_{0}\left(
251: x-vt\right) \right] \right\} \right) .
252: \end{eqnarray}
253: Let us consider an initial condition such that $0<u_{0}(x)<2mK$ where
254: $m$ is a positive integer. All those initial points of the profile
255: that belong to the region delimited by $(2m-2)<u(x)<2mK$ can be
256: described by the following dynamics. The solution grows from
257: $u=(2m-2)K$ to the value $u=(2m-1)K$ and decays to it from $u=2mK$
258: following the evolution of an initial condition with compact support
259: in that region. As extensively studied in the case of the logistic
260: nonlinearity \cite{gk}, any initial condition with a bounded domain
261: evolves into a step profile of width given by its initial value. In the
262: sinusoidal nonlinearity, from each region a step profile with width given by
263: $%
264: \Theta \left[ 2mK/\pi -u_{0}(x)\right] -\Theta \left[ (2m-2)K/\pi %
265: -u_{0}(x)\right] $ eventually emerges. Figure (\ref{fig:sinusoidal})
266: shows an example for an initial condition that overlays three such
267: regions. It is clear from the figure that in the bottom region the
268: profile behaves as if the other regions did not exist.
269: \begin{figure}[!htbp]
270: \includegraphics[width=\columnwidth]{figure2}
271: \caption{The `pyramid effect' for a nonlinearity with multiple zeros. The sinusoidal nonlinearity of
272: Eq.~(\ref{nlpyramid}) is plotted in (a). The horizontal lines at
273: $u/K=1,3,5$ depict levels of separation of the regions
274: between stable zones. The corresponding evolution is plotted in (b)
275: for the initial profile $u_{0}(x)=5.9\,K\,\textrm{sech}(x/\sigma)$, where
276: $\sigma$ is an arbitrary length. The initial profile is shown as a
277: dotted line. The profile moved rightwards as
278: shown.}\label{fig:sinusoidal}
279: \end{figure}
280:
281: \subsection{Multi-individual struggle for environment resources: linear growth with power saturation}
282: One can envisage physical situations in which one encounters a simple generalization of the logistic nonlinearity that retains
283: the linear behavior near $u=0$ but accentuates the nonlinearity for
284: larger values of $u$. This would arise, e.g., if the elemental struggle of the individuals for resources involves
285: not two but multiple individuals. The struggle would then not be binary in nature. To address this possibility, we consider
286: \begin{equation} \label{eq:lgps}
287: f(u)= \left[ 1 - (u/K)^{n}\right].
288: \end{equation}
289: The saturation effect already seen in the logistic case, becomes more
290: abrupt as $n$ increases. Pictorial representations for several cases
291: of $n$ are provided in Fig.~(\ref{fig:pictorialgenlog}). Application
292: of the prescription (\ref{eq:gdef}) produces the explicit form for
293: $G(u)$
294: \begin{equation} G(u)
295: =\frac{1}{n} \ln \left[ \frac{1 - (u/K)^n}{(u/K)^n} \right].
296: \end{equation}
297: Its inverse can be written analytically. An arbitrary initial
298: condition $u_0 (x)=u(x,0)$ therefore via prescription (\ref{solution})
299: evolves according to
300: \begin{equation} u(x,t) =
301: \frac{u_0(x-vt)}{\left[\left(\frac{u_0(x-vt)}{K}\right)^n(1-e^{-ant})+e^{-ant}\right]^{1/n}}.
302: \end{equation}
303: \begin{figure}[!htbp]
304: \begin{center}
305: \includegraphics[width=\columnwidth]{figure3}
306: \caption{The nonlinearity with
307: power saturation and its effect on the time evolution. The
308: nonlinearity with $f(u)$ given by (\ref{eq:lgps}) is shown in (a) for two different
309: powers $n$. Near $u=0$ both functions are similar but for higher
310: values they differ dramatically since the case $n=100$ has an abrupt drop close
311: to $u=K$. Their corresponding time evolution is plotted in
312: (b) at a time $at=5$ starting from the same initial condition $K \textrm{sech}(x/\sigma)$
313: where $\sigma$ is an arbitrary length.}
314: \label{fig:pictorialgenlog}
315: \end{center}
316: \end{figure}
317:
318: \subsection{Nonlinear growth with power saturation as in bisexual reproduction}
319: Interesting consequences arise if the nonlinearity cannot be
320: approximated by a linear term near $u=0$, such as when the growth of
321: the species is nonlinear. A physically relevant example is the
322: representation of bisexual reproduction that would require a growth
323: term bilinear rather than linear (see for example Ref.~\cite{rosas2002}).
324: The generalization of the nonlinearity (\ref{eq:lgps}) to a
325: \emph{nonlinear growth} term can be expressed as
326: \begin{equation} \label{eq:nonlinearf}
327: f(u)= (u/K)^{m}\left[1-(u/K)^{n}\right]
328: \end{equation}
329: where $m+1$ is the first power in $u$ in a Taylor expansion near zero.
330: Prescription (\ref{solution}) gives
331: \begin{equation} \label{eq:nonlinearG}
332: G(u)=\frac{{}_{2}F_{1}\left[1,-\frac{m}{n},1-\frac{m}{n},
333: \left(\frac{u}{K}\right)^{n}\right]}%
334: {K m \left(\frac{u}{K}\right)^{m}}.
335: \end{equation}
336: where ${}_{2}F_{1}$ is the hypergeometric function. In general, Eq.
337: (\ref{eq:nonlinearG}) cannot be inverted and $G^{-1}$ has to be
338: calculated implicitly. However, analytical considerations regarding
339: the shape of tails and shoulders of traveling fronts can be easily
340: obtained through the help of Eq. (\ref{eq:nonlinearG}) as will be seen
341: below in Sec \ref{sec:travelfront}.
342:
343:
344: \subsection{Nonlinearity with provisional saturation}
345: The previous nonlinearities we have discussed above increase
346: (decrease) monotonically from one zero to a maximum (minimum) and
347: then monotonically decrease (increase) to the next zero. Our
348: prescription (\ref{solution}) permits the analysis of more interesting
349: situations that may arise if the mutual struggle for resources among individuals constituting
350: the species leads to saturation of the population size that is not permanent but only provisional.
351: Further increase in the population density in such a situation might lead to a returned
352: increase in the nonlinearity, followed by an eventual decrease. Such a state of affairs
353: is analogous to the case analyzed in subsection B above but with an important difference.
354: In the present case, $uf(u)$ does not dip below zero. As a consequence, there
355: are no multiple zeros at $u \neq 0$. In Fig.(\ref{fig:dip}) we present, as an example of such a
356: nonlinearity, a polynomial of the form,
357: \begin{equation} \label{eq:poly1}
358: u f(u)=9(u/K) - \frac{131\,(u/K)^2}{12} + \frac{9\,(u/K)^3}{2} -
359: \frac{7\,(u/K)^4}{12}
360: \end{equation}
361: and the corresponding time evolution. As time
362: evolves, the curvature of the solution reflects the specific
363: particulars of the ``reaction'' term.
364: In the solution the regions with $u$ values close to the top of
365: the reaction function have more inclination.
366: \begin{figure}[!htbp]
367: \begin{center}
368: \includegraphics[width=\columnwidth]{figure4}
369: \caption{Example of a nonlinearity with provisional saturation.
370: In (a) the reaction function $uf(u)/K$ of the polynomial
371: \ref{eq:poly1} is presented, while in (b) the corresponding evolution
372: for an initial condition is shown.}
373: \label{fig:dip}
374: \end{center}
375: \end{figure}
376: To study the effects of different dips, in Fig.(\ref{fig:dip2}) three similar $u f(u)$ are shown
377: which differ in the 'depth' of their valleys. Their construction is straightforward by means of the Lagrange interpolating polynomial, i.e. by taking
378: a polynomial of fourth order and requiring it to have specific values at the five points desired.
379: In the example of the figure, four of the five points that are required to be common are: $(0;0),(1;0),(1/4;1),(3/4;1)$
380: described in $(u/K;u f(u))$ plane. The other point will characterize the dip, and
381: for the example shown in the figures are $(1/2;9/10),(1/2;1/2),(1/2;1/10)$.
382: From the same initial condition,
383: the evolution will differ as shown in (b): the more pronounced
384: the dip in the nonlinearity, the flatter the solution close to the
385: center of the dip.
386: \begin{figure}[!htbp]
387: \begin{center}
388: \includegraphics[width=\columnwidth]{figure5}
389: \caption{Dependence of the depth of the dip for nonlinearities with provisional saturation. In (a) three different nonlinearities are plotted. The axis are interchanged respect to the previous figure.
390: This has been done to express clearly the correspondence between the nonlinearities in (a) and the evolution in (b).}
391: \label{fig:dip2}
392: \end{center}
393: \end{figure}
394:
395: \section{Traveling Front Solutions: Analytical expressions for spatial limits}
396: \label{sec:travelfront}
397: The analysis in the previous section has focused on the initial value
398: problem. Much interesting information in nonlinear problems may
399: be extracted, however, directly from traveling waves (see, e.g., extensive
400: reports in the soliton literature such as in \cite{soli}). The $G$-function appearing in our prescription
401: (\ref{solution}) can be used conveniently, and directly, to find
402: traveling wave solutions. As will be seen below, there are situations
403: in which this technique is useful even when explicit analytic
404: expressions cannot be found for the initial value problem. Imposing
405: the traveling wave ansatz $u(x,t) = U( z = x - c t)$ in
406: Eq.\eqref{problem},
407: \begin{equation} \label{eq:tfronteq}
408: -\beta \dparcial{U}{z}= a U f(U),
409: \end{equation}
410: where
411: \begin{equation}
412: \beta=c-v
413: \end{equation}
414: is the difference between the traveling front velocity
415: and the medium velocity.
416: Integrating (\ref{eq:tfronteq}) in $z$, leads to the solution
417: \begin{equation}\label{eq:tweq}
418: G\left[ U_{\beta }(z)\right] =a\frac{z}{\beta },
419: \end{equation}
420: where $G$ is defined in Eq.~(\ref{eq:gdef}).
421: If the function $G^{-1}$ can be written analytically, the
422: front solution can be obtained in explicit form:
423: \begin{equation}
424: U_{\beta }(z)=G^{-1}\left[ a\frac{z}{\beta }\right] .
425: \label{eq:tfteo}
426: \end{equation}
427: Otherwise it is defined implicitly through Eq. (\ref{eq:tweq}).
428:
429: Equation (\ref{eq:tfteo}) shows that, for a given nonlinearity, the
430: steepness of the front depends on the propagation velocity $c$ while
431: its shape is determined by the function $G^{-1}$. In order to show
432: how, for a given propagation velocity $c$, the shape of the front
433: changes for various nonlinearities (according to the simple relation
434: (\ref{eq:tfteo})), we provide Fig. (\ref{fig:tw2speed1}).
435: \begin{figure}[!htbp]
436: \begin{center} \includegraphics[height=0.4\columnwidth,width=0.45\columnwidth]{figure6}
437: \end{center}
438: \caption{Example of several traveling waves of the \emph{same speed, $\beta =1$}, but
439: for different nonlinearity functions $f$. The solid line is the
440: generalized logistic function, $u f_1(u)=u(1-u^{2})$, while the dotted
441: line comes from the first half period of the sinusoidal function, $u f_2(u)=sin(\pi
442: u)$. }
443: \label{fig:tw2speed1}
444: \end{figure}
445:
446: We now show how one may find analytic expressions for the tail and the
447: shoulder of the traveling wave through inspection of the limiting
448: values of the nonlinearity $auf(u)$, respectively, as $u\to
449: 0^{+}$ and $u\to K^{-}$, where $0$ and $K$ are both zeros of
450: the nonlinearity. The basic idea is to apply to Eq. (\ref{eq:tweq}) a
451: judiciously chosen function $h$, multiply by $U$ or $(K-U)$ as
452: appropriate, and then calculate the relevant limit of the product.
453: Thus, if we are interested in the spatial dependence of the tail of
454: the solution, we seek to obtain
455: \begin{equation} \label{taillimit}
456: \lim_{U\to 0^{+}} U(z) h_T\left( G\left(
457: U\right)
458: \right) =\lim_{z/\beta \to +\infty } U(z) h_T(az/\beta ) = C_T.
459: \end{equation}
460:
461: The function $h_T$ is chosen in such a way that the limit for
462: $U\to 0^{+}$ of the left hand side of the equation equals a
463: non zero constant depicted above by $C_T$. If the shoulder spatial
464: dependence is sought, we apply another function $h_S$, and multiply by
465: $(K-U)$ rather than by $U$ before taking the limit $U\to
466: K^{-}$,
467: the choice of $h_S$ being such that the limit is a constant
468: $C_S$.
469:
470: The spatial dependence of the traveling solution in the two extreme
471: limits is found to be
472: \begin{equation} \label{eq:taillimitsolution}
473: \lim_{z/\beta \to + \infty }U(z)\simeq \frac{C_{T}}{h_{T}\left(
474: az/\beta\right) }
475: \end{equation}
476: for the tail and
477: \begin{equation}
478: \lim_{z/\beta \to - \infty }U(z)\simeq K-\frac{C_{S}}{h_{S}\left(
479: az/\beta\right) } \label{shouldep}
480: \end{equation}
481: for the shoulder.
482:
483: The steepness of the front at the inflection point is another
484: important feature of the traveling front that can be obtained from the
485: shape of the nonlinearity even if the exact analytic form is not
486: known. From Eq. (\ref{eq:tfronteq}) we can simply relate
487: \begin{equation}
488: \frac{dU(z)}{dz}=-aU(z) \frac{f\left[ U(z)\right] }{\beta}.
489: \end{equation}
490: which clearly shows the relation between the maximum of $f(u)$ and the
491: front derivative at the inflection point. Comparing such values with
492: those cases of $f(u)$ in which the traveling front shape are known
493: analytically, allows to determine further qualitative characteristics
494: of the front shapes. We apply these various general considerations to
495: two specific cases below.
496:
497: \subsection{Linear growth with power saturation}
498: From the general expression (\ref{solution}), it is straightforward to
499: write the analytic form of the traveling fronts
500: \begin{equation}
501: U_{\beta}(z) = \frac{K}{\left(1 + \left(2^{n}-1\right)e^{an
502: \frac{z}{\beta}}\right)^{1/n}},
503: \label{travfrontsat}
504: \end{equation}
505: where we have chosen to center the traveling front in such a way that
506: $U_{\beta}(z=0)$ is half the saturation value of $U$.
507: Application of the limiting procedure described above shows that
508: regardless of the value of the power saturation $n$ the front tail is
509: given by
510: \begin{equation}
511: \lim_{z/ \beta\to +\infty}U_{\beta}(z) \propto e^{-a z/\beta}.
512: \end{equation}
513: The value $n$ does not play a role since the tail shape is determined
514: by the form of the nonlinearity close to $U=0$ where saturation
515: effects are negligible. On the other hand, the value $n$ does play a
516: role in determining the shoulder shape. The limiting procedure gives
517: in fact
518: \begin{equation}
519: \lim_{z/ \beta\to -\infty}U_{\beta}(z) \propto K-\frac{K}{n}e^{-a z/\beta}.
520: \end{equation}
521: which coincides with the one calculated directly from Eq.
522: (\ref{travfrontsat}).
523:
524: It is interesting to notice that in the limiting case corresponding to
525: $n\to \infty$ the traveling front solution can be written as
526: \begin{equation}
527: U_\beta(z)=\left\{\begin{array}{ll}
528: K & \textrm{if } z\leq 0\\
529: K e^{-z/\beta} & \textrm{if } z>0
530: \end{array}\right.
531: \end{equation}
532:
533: \subsection{Nonlinear growth}
534: For the case in which the growth near $u\sim 0$ is bilinear as in
535: (\ref{eq:nonlinearf}) for $m=2$, the prescription (\ref{solution}) is given in terms of
536: hypergeometric functions (see Ref.~\cite{hypergeometric}), whose inverses are not
537: known. However our limit prescriptions (\ref{eq:taillimitsolution}) and
538: (\ref{shouldep}) do provide analytical expressions for the spatial
539: limits. For $U \approx 0$ the following behavior is obtained
540: \begin{equation}
541: G(u) \propto \frac{1}{K m (U/K)^m}.
542: \end{equation}
543: The function $h_T$ of Eq.\eqref{taillimit} which makes the left limit
544: a constant is found to be
545: \begin{equation}
546: h_T(G)= G^{\frac{1}{m-1}}.
547: \end{equation}
548: Therefore, in the tail,
549: \begin{equation}
550: \lim_{z/\beta \to +\infty} U(z) \simeq \frac{C}{(z/\beta)^{\frac{1}{m-1}}}.
551: \end{equation}
552:
553: Indeed, we can determine the shape of the tail (shoulder) for any function $f$
554: that has a Taylor expansion near $U=0$ ($U=K$).
555:
556: \section{Incorporation of diffusion along with convection}
557: \label{sec:diffusion}
558: The analytic treatment we have given above of the reaction-convection
559: problem represented by Eq. (\ref{problem}) has been made possible by
560: the fact that the partial differential equation considered is of first
561: order. Once the transformation is made to convert (\ref{problem})
562: exactly into its linear counterpart, the method of characteristics \cite{char},
563: well-known in the context of linear equations, is what is
564: behind the analysis we have presented.
565: If diffusion
566: is added to (\ref{problem}), individual points of the $u(x)$ curve at
567: a given time no longer evolve independently as they do for systems
568: characterized by (\ref{problem}) but influence one another in the
569: evolution. Finding analytic solutions in the manner
570: detailed here and in Refs. \cite{gk,k} is then impossible. There is no
571: doubt from the practical viewpoint that the case $D=0$ is important to
572: study. This is so because negligible diffusion does occur in several
573: physical
574: situations as in the case of bacterial dynamics when the environment
575: or genetic engineering can be made to result in convection being far
576: more important than diffusion. Nevertheless it is important to ask
577: whether the introduction of diffusion can be considered a perturbation
578: or whether it alters the problem drastically, even when it is
579: relatively small in magnitude. If the latter were true, the analysis
580: presented in the previous sections would be uninteresting for all
581: realistic cases in which
582: finite diffusion exists. We therefore analyze the effects of adding
583: diffusion to the nonlinear convective problem:
584: \begin{equation}
585: \label{eq:driftdiffusion}
586: \dparcial{u}{t} + v \dparcial{u}{x} = a u f(u) + D \ddparcial{u}{x}.
587: \end{equation}
588: We do this in three parts, always focusing attention on traveling wave
589: solutions for simplicity. In the first part (subsection A below), we
590: obtain exact analytical expressions
591: for traveling wave solutions of the reaction diffusion problem
592: represented by equation (\ref{eq:driftdiffusion}) when $f$ is a
593: piecewise
594: linear function. Also, we compare the results between the absence of
595: diffusion to those in its presence (arbitrary amount of diffusion). In
596: the second part, when the amount of diffusion is small, we identify an
597: appropriate small parameter and develop
598: a perturbation scheme. In the third part, we apply the
599: perturbation treatment to one of the interesting nonlinearities
600: (sinusoidal) introduced at the beginning of this paper. Furthermore,
601: we analyze its validity by comparing it to the exact solution of (\ref{eq:driftdiffusion}) when
602: the nonlinearity is piecewise linear. The last two parts make up subsection B below.
603:
604: \subsection{Diffusion effects treated through a piecewise linear representation of the nonlinearity}
605:
606: Although \emph{exact} analytic solutions of the nonlinear equation we consider are
607: impossible in the presence of diffusion, they are possible for
608: traveling waves if we employ a piecewise linearization of the
609: nonlinearity. The piecewise linear approximation has been used
610: earlier in the analysis of the Fisher equation for the study of the effects of
611: transport memory \cite{manne} as well as pattern formation
612: \cite{horacio}. In both cases it was possible to extract useful
613: information through this resource, whose advantage is that once
614: the approximation of representing the given nonlinearity
615: by the piecewise linear form is made, no further approximation is invoked.
616: Let us begin with a piecewise representation of the logistic nonlinearity through
617: \begin{equation} \label{eq:linear}
618: f(u)=
619: \left\{\begin{array}{ll}
620: 1 & u/K\leq \alpha \\
621: \left(\frac{\alpha}{1-\alpha}\right) \left(\frac{1-u/K}{u/K}\right) & u/K\geq \alpha,
622: \end{array}\right.
623: \end{equation}
624: in which the first piece of $u f(u)$ connects the unstable zero at $u=0$ to the maximum at
625: $u=\alpha K$, while the second piece joins the maximum with the stable
626: zero at $u=K$. The parameter $0\leq\alpha\leq1$ in (\ref{eq:linear}) represents the
627: relative position of the maximum of the nonlinearity between the stable and
628: unstable zeros of the system. If $\alpha=1/2$, the derivatives of
629: the logistic nonlinearity and its piecewise representation become equal at $u=0$ and $u=K$. Since the tail of a front solution is determined by the limit (\ref{eq:taillimitsolution}), the choice $\alpha=1/2$ ensures that the spatial dependence of the tail (shoulder) front in the logistic nonlinearity
630: coincides with the tail (shoulder) front when $f(u)$ is given by (\ref{eq:linear}).
631: In the analytic treatment below, for the sake of generality, we keep the parameter $\alpha$ unspecified.
632:
633: Following the method of Ref. \cite{manne} it is straightforward to determine
634: the exact form of the traveling wave solution of Eq. (\ref{eq:driftdiffusion}) with $f$ given by \eqref{eq:linear}. Such a solution, denoted as $U(z,\gamma)$, where $\gamma=v_f/\beta$ and $v_f=2 \sqrt{D a}$ is given by
635: \begin{subequations} \label{eq:pwtwsolution}
636: \begin{align}
637: U(az/\beta\leq0,\gamma)&=K-K(1-\alpha)\, \exp\left(\frac{az}{\beta}/\lambda_{1}\right),\\
638: U(az/\beta>0,\gamma)&= B_{2}\,
639: \exp\left(-\frac{az}{\beta}/\lambda_{2}\right) + B_{3}\,
640: \exp\left(-\frac{az}{\beta}/\lambda_{3}\right), \label{eq:sumexp}
641: \end{align}
642: with
643: \begin{align}
644: \lambda_{1} &= (2/\gamma^2)
645: \left(\sqrt{1+\frac{\alpha}{1-\alpha}\gamma^2}-1\right), &
646: B_{2} &= (1/2) K \left(\alpha-\xi\right),\\
647: \lambda_{2} &= (2/\gamma^2) \left(1+\sqrt{1-\gamma^2}\right), &
648: B_{3} &= (1/2) K \left(\alpha+\xi\right),\\
649: \lambda_{3}&= (2/\gamma^2) \left(1-\sqrt{1-\gamma^2}\right), &
650: \xi&=-(1-\alpha)\frac{\sqrt{1+\frac{\alpha}{1-\alpha}\gamma^2}}{\sqrt{1-\gamma^2}}+\frac{1}{\sqrt{1-\gamma^2}}.
651: \end{align}
652: \end{subequations}
653: By requiring that $u$ is always positive or
654: zero, a condition
655: for \eqref{eq:pwtwsolution} emerges as
656: \begin{equation} \label{eq:twcondition}
657: |\gamma|\leq 1.
658: \end{equation}
659: This bound represents
660: a constraint between the Fisher speed $v_f$ and the relative (to the medium) speed of the front solution $\beta$. Equation (\ref{eq:twcondition}) is equivalent to the minimum speed requirement for traveling wave of reaction-diffusion systems and it expresses the fact that the front speed must be faster or equal to the diffusive speed.
661: \begin{figure}[!htbp]
662: \includegraphics[width=\columnwidth]{figure7}
663: \caption{Comparison of the shape of the traveling wave solution (\ref{eq:pwtwsolution}) with and without diffusion. On the left, in (a), a front solution with $\gamma=0$ and one with $\gamma=0.9$ are depicted.
664: On the right, in (b), $\Delta U(z,\gamma)=U(z,\gamma)- U(z,0)$ between Eq. (\ref{eq:pwtwsolution}) with $\gamma=0.1,0.5,0.9$ and the case with $\gamma=0$ are shown.}
665: \label{fig:pwdiff}
666: \end{figure}
667:
668: The parameter $\gamma$ is a measure of the relative diffusion in the system. In Fig. (\ref{fig:pwdiff}) we show how the front solution changes as diffusion increases. As in reaction diffusion equations, it is clear in Fig. (\ref{fig:pwdiff}b) that, as diffusion increases, the tail of $U(z,\gamma)$ acquires a shallower profile. The steepness of the tail as function of $\gamma$ can be determined from
669: Eq. (\ref{eq:sumexp}). The front tails are sums of two exponentials each one with characteristic lengths ($\lambda_{2}$ and $\lambda_{3}$) and multiplicative weighing factors ($B_{2}$ and $B_{3}$), all functions of $\gamma$. When $\gamma=0$, $B_{2}=0$ and $\lambda_{2}=0$. Eq. (\ref{eq:sumexp}) then contains \textit{only} one exponential term. The tail of the front solution is thus given by
670: \begin{equation} \label{eq:gamma0}
671: U\left(az/\beta>0,0\right)=B_3 \exp\left(-\frac{az}{\beta
672: \lambda_3}\right)=\alpha\,K\, \exp\left(-a z/\beta\right).
673: \end{equation}
674: At the other extreme, i.e., when $\gamma=1$, the characteristic lengths $\lambda_2$ and $\lambda_3$ are equal to each other, the divergent part of the coefficients, $\xi$, cancels in the sum, and the tail of the front solution becomes
675: \begin{equation} \label{eq:gamma1}
676: U\left(az/\beta>0,1\right)=(B_2+B_3) \exp\left(-\frac{az}{\beta 2}\right)+
677: =\alpha\, K\, \exp\left(-\frac{1}{2}a z/\beta\right).
678: \end{equation}
679: For values $0<\gamma<1$, both exponentials contribute to the tail shape but for small $\gamma$ values
680: $B_{2}$ is negligible and $\lambda_{3}\gg \lambda_{2}$ therefore making the second exponential in (\ref{eq:sumexp}) dominate. This dependence is plotted in Fig. \ref{fig:pltwdisplay}.
681: \begin{figure}[!htbp]
682: \includegraphics[width=0.6\columnwidth]{figure8}
683: \caption{Plot of the dependence on $\gamma$ of the parameters associated with the traveling front solution (\ref{eq:sumexp}). The characteristic lengths of the exponentials $\lambda_2$ and $\lambda_3$ are depicted between the values $\gamma=0$ and $\gamma=1$. In the inset the multiplicative factors $B_{2}$ and $B_{3}$ are also plotted.}
684: \label{fig:pltwdisplay}
685: \end{figure}
686:
687: The use of the piecewise representation has allowed us to conduct an exact analysis above. When such a representation is not used, a perturbation scheme must be employed. The behavior shown in Fig. \ref{fig:pltwdisplay} lends support to the idea that a quantity directly related to $\gamma$
688: should serve as the small parameter on which to base the development of a perturbation technique for traveling front solutions.
689: We develop such a scheme in the next subsection.
690:
691: \subsection{Perturbation treatment for traveling fronts}
692: The analysis presented here is a generalization of the technique proposed by Canosa
693: \cite{canosa1973} for studying traveling fronts of the Fisher equation.
694:
695: In the presence of diffusion, the traveling fronts of Eq. (\ref{eq:driftdiffusion})
696: are solutions of the following equation
697: \begin{equation}
698: \frac{\gamma^2}{4} U_{\zeta\zeta}+ U_{\zeta} + U f(U)=0,
699: \end{equation}
700: where $\zeta=az/\beta$ and $z=x-ct$ where $c$ is the
701: absolute speed of the traveling front.
702:
703: The quantity $\xi=(\gamma/2)^2 \le1/4$
704: is the perturbation parameter we will choose. A standard expansion of
705: the traveling front in powers of $\xi$ in Eq. (\ref{eq:driftdiffusion})
706: gives the identity
707: \begin{equation}
708: U_0'(\zeta) + \sum_{n=1}^\infty \xi^n \left( U_n'(\zeta) + U_{n-1}''(\zeta) \right)%
709: +F\left[U_0+\sum_{n=1}^\infty \xi^n U_n\right]=0,
710: \end{equation}
711: where $F(u)=uf(u)$. By Taylor-expanding the function $F(u)$, and rearranging each term in powers of $\xi$, it is possible to find the approximation of order $n$ in $\xi$ to the traveling front solution of Eq. (\ref{eq:driftdiffusion}). The zeroth order of such an expansion is given by the solution of
712: \begin{equation}
713: \frac{d U_0}{d \zeta} + F(U_0) = 0 \label{eq:zerothorder}
714: \end{equation}
715: which has been analyzed in the previous sections.
716: The next order is given by
717: \begin{equation}
718: \frac{d U_1}{d \zeta} + \frac{d^2 U_0}{d \zeta^2} + U_1 F'(U_0) = 0,
719: \end{equation}
720: whose solution is \cite{murray}
721: \begin{equation}
722: U_1(\zeta)=-U_0'(\zeta) \ln\left(4\, |U_0'(\zeta)|\right)
723: \end{equation}
724: The next order, the second, is given by
725: \begin{equation}
726: \frac{d U_2}{d \zeta} + U_2 F'(U_0) + \frac{d^2 U_1}{d \zeta^2} + \frac{1}{2!}F''(U_0) U_1^2 =0
727: \end{equation}
728: which is a linear equation for $U_2$, and all the other functions are given by solutions
729: of the previous orders.
730:
731: To arbitrary order $n$, we have
732: \begin{equation}
733: \frac{d U_n}{d \zeta} + \frac{d^2 U_{n-1}}{d \zeta^2} + U_n F'(U_0) + h(\{U_{i<n}\}) = 0
734: \end{equation}
735: where $h(\{U_{i<n}\}$ are the rest of the terms that come from the $n$th power of $\xi$.
736:
737: Canosa\cite{canosa1973} and Murray\cite{murray} have provided up to to first order for the Fisher case.
738: For the trigonometric nonlinearity with $f(u)$ given by (\ref{eq:lgps}) studied earlier in the present paper,
739: application of this perturbation techniques gives for the first order correction
740: \begin{equation}
741: U_1(z)= \frac{1}{\pi} \textrm{sech}(a z/\beta) \ln\left(\textrm{sech}(a z/\beta)\right).
742: \end{equation}
743:
744: To further analyze the validity of the perturbation scheme,
745: we use the exact expression for the traveling fronts of the piecewise
746: linear function and compare it with the results obtained from the zeroth,
747: first, and second order perturbation.
748: Fig. \ref{fig:exactandnosoexact} shows the differences between the solutions
749: for two different values of the diffusion constant represented by $\gamma=0.1$ and $\gamma=0.9$.
750: We see that, for small $a z/\beta$, the lower orders are \emph{more} accurate than the higher orders.
751: In space this corresponds to that part of the front profile close to the value where $U(z)=1/2$. However, for the rest of the $az/\beta$ range, the higher orders in the perturbation get closer to the actual solution. It is worth noticing that, for $\gamma=0.9$, which represents a high value of diffusion, the error committed is still low.
752: Perhaps unexpectedly, for low values of $az/\beta$, the lower orders are more accurate.
753: The convergence of the orders is fast: for $\gamma=0.1$ the second order is just overlaped by the first order.
754: \begin{figure}[!htbp]
755: \centering
756: \includegraphics[width=\columnwidth]{figure9}
757: \caption{Difference between exact solution of the piecewise linear function and its zeroth, first and second order approximations in our perturbative treatment for two different values: $\gamma=0.1$ and $\gamma=0.9$ in (a) and (b) respectively. Notice that, although not shown in the graph, as the order of the perturbation increases the curve difference converges rapidly to a curve which represents the error upper bound.}
758: \label{fig:exactandnosoexact}
759: \end{figure}
760:
761: The above analysis has uncovered a perhaps obvious but important point
762: regarding the perturbation parameter. In \eqref{eq:driftdiffusion}, there are two
763: characteristic velocities: the Fisher velocity $v_f = \sqrt{Da}$ and
764: the medium velocity $v$. It may be tempting to suppose that the
765: appropriate perturbation parameter is their ratio \cite{gk}. This is
766: not correct because the entire effect of the medium velocity $v$ can
767: be eliminated by a transformation of the frame of reference.
768: Our analysis above shows that it is crucial to take into consideration
769: the third velocity that exists in the traveling wave context in this
770: problem, viz. the velocity of the traveling wave $c$. The appropriate
771: perturbation parameter when we consider a traveling wave is
772: $v_f/(c-v)$ and our conclusion is that for diffusion to be
773: considered as a perturbation on the pure convective analysis that we
774: have presented in this paper, it is necessary that the velocity of the
775: traveling wave exceed the medium velocity by an amount larger than the
776: Fisher velocity.
777:
778:
779: \section{Summary} \label{sec:conclusions}
780: Several physically realizable situations exist in various areas of
781: nonlinear population dynamics wherein the behavior is controlled
782: primarily by the nonlinearities and the convective element of the
783: motion, the diffusive contribution to the motion being small. The
784: typical example is the behavior of bacteria in a petri dish
785: \cite{nelson,ana,gk}, where the so-called "wind" motion
786: induced by moving masks is made
787: stronger than the inherent diffusive motion. A study of such situations
788: has been provided in the present paper by first analyzing the case for
789: no diffusion (sections \ref{sec:application},\ref{sec:travelfront}) and then extending the analysis to small
790: diffusion (section \ref{sec:diffusion}). Explicit results have been obtained for a
791: variety of nonlinearities on the basis of a straightforward prescription
792: \cite{k}, for the full initial value problem. Traveling waves
793: have been analyzed explicitly for several cases. Interesting
794: consequences such as the "pyramid effect" for sinusoidal nonlinearities
795: have been shown to occur. The effect of small diffusion has been
796: incorporated through exact analysis for piecewise representation of the
797: nonlinearities on the one hand and through perturbative calculations on
798: the other.
799:
800: \begin{acknowledgments}
801: This work was supported in part by DARPA under grant no. DARPA-N00014-03-1-0900, by NSF/NIH Ecology of Infectious Diseases under grant no. EF-0326757, and by the NSF under grant no. INT-0336343.
802: \end{acknowledgments}
803:
804: \begin{thebibliography}{99}
805: \bibitem{fisher1937} R.A. Fisher, Ann. Eugen. \textbf{7}, 355 (1937).
806: \bibitem{murray} J.D. Murray, \textit{Mathematical Biology}, 2nd
807: edition (Springer, New York, 1993).
808: \bibitem{skellam} J.C. Skellam. Biometrika \textbf{38}, 196 (1951)
809: \bibitem{kk} V.M. Kenkre and M.N. Kuperman, Phys. Rev. E \textbf{67},
810: 051921 (2003).
811: \bibitem{bkk} M. Ballard, V.M. Kenkre, and M.N. Kuperman, Phys. Rev. E
812: \textbf{70}, 031912 (2004)
813: \bibitem{ak} G. Abramson and V.M. Kenkre, Phys. Rev. E \textbf{66},
814: 011912 (2003).
815: \bibitem{akyp} G. Abramson, V.M. Kenkre, T. Yates, and R.R. Parmenter,
816: Bull. Math. Bio. \textbf{65}, 519 (2003).
817: \bibitem{kpasi} V.M. Kenkre, in \textit{Modern Challenges in
818: Statistical Mechanics: Patterns, Noise, and the Interplay of
819: Nonlinearity and Complexity}, V. M. Kenkre and K. Lindenberg,
820: eds., AIP Proc. vol. \textbf{658} (2003), p. 63.
821: \bibitem{manne} K.K. Manne, A.J. Hurd, and V.M. Kenkre, Phys. Rev. E
822: \textbf{61}, 4177 (2000).
823: \bibitem{abk} G. Abramson, A.R. Bishop, and V.M. Kenkre, Phys. Rev. E
824: \textbf{64}, 66615 (2001).
825: \bibitem{fkk} M.A. Fuentes, M.N. Kuperman, and V.M. Kenkre, Phys. Rev.
826: Lett. \textbf{91}, 158104 (2003).
827: \bibitem{fk} M.A. Fuentes, M. Kuperman, and V.M. Kenkre, J. Phys. Chem. B \textbf{108}, 10505 (2004).
828: \bibitem{nelson} D.R. Nelson and N.M. Shnerb, Phys. Rev. E \textbf{58},
829: 1383 (1998); K.A. Dahmen, D.R. Nelson, and N.M. Shnerb, J. Math. Biology
830: \textbf{41}, 1 (2000).
831: \bibitem{ana} A.L. Lin, B.A. Mann, G. Torres-Oviedo, B. Lincoln, J.
832: Kas, and H.L. Swinney, Biophys. J. \textbf{87}, 75 (2004).
833: \bibitem{gk} L. Giuggioli and V.M. Kenkre, Physica D \textbf{183}, 245
834: (2003).
835: \bibitem{lgthesis} L. Giuggioli, Ph.D. Thesis (University of New
836: Mexico).
837: \bibitem{k} V.M. Kenkre, Physica A \textbf{342}, 242 (2004).
838: \bibitem{allee} W.C. Allee, \textit{The Social Life of Animals}, (Beacon Press, Boston,1938).
839: \bibitem{courchamp} F. Courchamp, T. Clutton-Brock, and B. Grenfell, Trends Ecol. Evol. \textbf{14}, 405 (1999).
840: \bibitem{SS} P.A. Stephens and W.J. Sunderland, Trends Ecol. Evol. \textbf{14}, 401 (1999).
841: \bibitem{fowlerbaker} C.W. Fowler and J.D. Baker, Rep. Int. Whaling Comm. \textbf{41}, 545 (1999).
842: \bibitem{rosas2002} A. Rosas, C.P. Ferreira, and J.F. Fontanari, Phys.
843: Rev. Lett. \textbf{89}, 188101 (2001).
844: \bibitem{soli} M. Remoissenet, \textit{Waves Called Solitons: Concepts and Experiments}, 3rd edition, (Springer, Berliln, 1999).
845: \bibitem{hypergeometric} Ed. M. Abramowitz and I.A. Stegun, \textit{Handbook of Mathematical Functions}, (Dover, New York, 1970)
846: \bibitem{char} I.G. Petrovsky, \textit{Lectures on Partial Differential Equations}, (Dover, New York, 1991)
847: \bibitem{horacio} G. Izus, R. Deza, C. Borzi, and H.S. Wio, Physica A \textbf{237}, 135 (1997).
848: \bibitem{canosa1973} J. Canosa, IBM. J. Res. and Dev. \textbf{17} 307 (1973).
849:
850: \end{thebibliography}
851:
852:
853: \end{document}
854:
855:
856:
857: