1: %2345678901234567890123456789012345678901234567890123456789012345678901234567890
2: \documentclass[floatfix,preprint,showemail,preprintnumbers,amsfonts,pre]{revtex4}
3: %\documentclass[showpacs,preprintnumbers,pre]{revtex4}
4: \usepackage{graphicx}% Include figure files
5: %\usepackage{dcolumn}% Align table columns on decimal point
6: \usepackage{bm}% bold math
7: %\usepackage{epsf}
8: %\usepackage[T1]{fontenc}
9: %\def\baselinestretch{2.0}
10: %\usepackage{amssymb}
11: \def\baselinestretch{1.0}
12: \pagestyle{plain}
13:
14: \begin{document}
15:
16: \title{Self-optimization, community stability, and fluctuations
17: in two individual-based models of biological coevolution
18: }
19:
20: \author{Per Arne Rikvold}\email{rikvold@scs.fsu.edu}
21:
22: \affiliation{School of Computational Science,
23: Center for Materials Research and Technology,\\
24: National High Magnetic Field Laboratory, and Department of
25: Physics,\\
26: Florida State University, Tallahassee, Florida 32306-4120,
27: USA\\
28: and Department of Fundamental Sciences,
29: Faculty of Integrated Human Studies,
30: Kyoto University, Kyoto 606, Japan
31: \\
32: }
33:
34: \date{\today}
35: % The correct dates will be entered by the editor
36:
37:
38: \begin{abstract}
39: We compare and contrast the long-time dynamical properties of
40: two individual-based models of biological
41: coevolution. Selection occurs via multispecies, stochastic
42: population dynamics with reproduction probabilities that
43: depend nonlinearly on the population densities of all species
44: resident in the community. New species are introduced through
45: mutation. Both models are amenable to exact
46: linear stability analysis, and we compare the analytic results
47: with large-scale kinetic Monte Carlo simulations, obtaining the
48: population size as a function of an average interspecies
49: interaction strength. Over time,
50: the models {\it self-optimize\/} through mutation and
51: selection to approximately maximize a community
52: fitness
53: %potential
54: function, subject only to constraints internal to the particular model.
55: If the interspecies
56: interactions are randomly distributed on an interval including
57: positive values, the system evolves toward self-sustaining,
58: {\it mutualistic\/} communities.
59: In contrast, for the predator-prey case the matrix of interactions is
60: antisymmetric, and a nonzero population size must be sustained by an
61: external resource.
62: Time series of the diversity and population size for both
63: models show approximate $1/f$ noise and power-law distributions for the
64: lifetimes of communities and species.
65: For the mutualistic model, these two lifetime distributions have the
66: same exponent, while their exponents
67: are different for the predator-prey model.
68: The difference is probably due to greater resilience
69: toward mass extinctions in the
70: food-web like communities produced by the predator-prey model.
71: \end{abstract}
72:
73: \maketitle
74:
75: %2345678901234567890123456789012345678901234567890123456789012345678901234567890
76:
77: \section{Introduction}
78: \label{sec:Int}
79:
80: Traditionally, problems in ecology and evolution have been addressed at
81: very different levels of resolution. Typically, ecological problems are
82: studied on a timescale of generations and often at the level of
83: individual organisms, while issues in evolution are considered on much
84: longer, often geological, timescales and usually at the level of
85: species or higher-level taxa. However, in recent years it has been
86: recognized that processes at the ecological and evolutionary scales can
87: be strongly linked \cite{DROS01B,THOM98,THOM99,YOSH03}.
88: Several models have therefore been proposed, which aim to
89: model the complex problem of coevolution in a fitness
90: landscape that changes with the composition of the community,
91: while spanning the disparate scales of both temporal and taxonomic resolution.
92: Early steps in this direction were simulations of parapatric and
93: sympatric speciation \cite{CROS70} and the coupled $NK$ model with
94: population dynamics \cite{KAUF93,KAUF91}.
95: More recent contributions include the Webworld model
96: \cite{CALD98,DROS01B,DROS04},
97: the Tangled-nature model
98: \cite{CHRI02,COLL03,HALL02}
99: and simplified versions of the latter
100: \cite{RIKV03,SEVI06,ZIA04},
101: as well as network models
102: \cite{CHOW05,CHOW03A}.
103: Recently, large individual-based simulations have also been performed of
104: parapatric and sympatric speciation \cite{GAVR98,GAVR00} and of
105: adaptive radiation \cite{GAVR05}.
106: Many of these models are deliberately quite simple, aiming to
107: elucidate {\it universal\/} features that are largely independent of the
108: finer details of the ecological interactions and the evolutionary
109: mechanisms. Such universal
110: features may include lifetime distributions for species and communities,
111: as well as other aspects of extinction statistics,
112: statistical properties of fluctuations in diversity and population
113: sizes, and
114: the structure and dynamics of food webs that develop and change
115: with time. By changing
116: specific features of such simplified models, one hopes to learn
117: which aspects influence the observed properties of the
118: resulting communities and their development with time.
119:
120: In this paper we compare and contrast two stochastic
121: coevolution models that combine
122: features from some of those mentioned above.
123: In each, selection is provided by an individual-based
124: population dynamics, while new species are provided by a low rate of
125: mutation. The models are studied
126: both analytically by linear stability theory, and numerically
127: by large-scale kinetic Monte Carlo simulations.
128: The first model (Model A) allows direct mutualistic
129: interspecies interactions, and some of its properties were discussed
130: previously \cite{RIKV03,SEVI06,ZIA04}.
131: It is a simplified version of the Tangled-nature model of Jensen and
132: coworkers \cite{CHRI02,COLL03,HALL02}.
133: The second model (Model B) is a predator-prey model. For both models we
134: obtain exactly the fixed-point mean population sizes and stability
135: properties for any given community of species in the limit of vanishing
136: mutation rate, using linear stability theory. These results
137: enable us to define a community
138: fitness
139: %potential
140: function, cubic in the
141: mean total population size, that is maximized at the fixed point
142: for a given community of species. The fact that this much information
143: can be obtained analytically makes these models
144: uniquely well suited as benchmarks for
145: more elaborate, but less tractable, nonlinear models.
146:
147: The analytical results are followed by numerical simulations of the
148: dynamics of both models for nonzero mutation rates. We focus
149: on very long simulations in a regime where both diversity and
150: population size are statistically stationary, albeit with very large,
151: strongly correlated fluctuations and an intermittently
152: vigorous turnover of species. We thus
153: study the {\it intrinsic\/} dynamics of extinction and origination of
154: species and communities in the absence of external perturbations.
155: Understanding of these intrinsic fluctuations in the stationary state
156: should enable one to estimate the community's response to external
157: perturbations \cite{SATO03} in a way analogous to the fluctuation-dissipation
158: relations of statistical mechanics \cite{PATH96}.
159:
160: Three main conclusions emerge from this combined analytical and numerical
161: investigation.
162: \begin{itemize}
163: \item
164: Mutations enable the models to evolve communities that tend to
165: maximize the community
166: fitness
167: %potential
168: function, subject only to
169: constraints that are internal to the specific model. Although there is
170: vigorous turnover of species and communities, all communities that
171: persist for a significant time remain in the vicinity of this maximum.
172: \item
173: The simulated systems exhibit power-law distributions in the lifetimes of
174: individual species, as well as of communities, and the fluctuations
175: in diversity and population size are characterized by approximate $1/f$ noise.
176: \item
177: The two models show distinct differences in the community turnover,
178: which are reflected in differences between the power-law exponent of the
179: community-lifetime distribution. In Model A, extinctions of species tend
180: to be highly synchronized, such that a whole community collapses in a mass
181: extinction on a relatively short time scale. In Model B, on the other hand,
182: extinctions are more often limited to a subset of the resident species.
183: This can be explained by differences between
184: the structures of the interaction networks characterizing
185: long-lived communities in the two models. For Model B, this network
186: amounts to a simple food web.
187: \end{itemize}
188:
189: The rest of this paper is organized as follows.
190: The models are introduced in Sec.~\ref{sec:Mod}.
191: In Sec.~\ref{sec:Lin} we perform a full analytical linear stability analysis,
192: and the results are compared with large-scale kinetic Monte Carlo simulations.
193: The simulations are described in further detail in
194: Sec.~\ref{sec:Dyn}, where lifetime distributions and power spectral
195: densities also are reported. A concluding summary is presented in
196: Sec.~\ref{sec:Conc}. Some technical details of the derivation of the
197: community
198: fitness
199: %potential
200: function are given in Appendix~\ref{sec:AppA},
201: an analytical result for the optimum population size of Model A is
202: obtained in Appendix~\ref{sec:AppX}, and
203: a discussion of the consequences of changes in several of the model
204: parameters is found in Appendix~\ref{sec:AppB}.
205:
206:
207: \section{The Models}
208: \label{sec:Mod}
209:
210: In both models, selection is provided by the reproduction rates in
211: an individual-based, simplified, stochastic
212: population-dynamics model with nonoverlapping generations.
213: This interacting birth-death process is
214: augmented to enable evolution of new species by a mutation mechanism.
215: The mutations act on a haploid, binary ``genome" of length $L$, as
216: introduced in Eigen's model for molecular evolution
217: \cite{EIGE71,EIGE88}. This bit string defines the species, which are
218: identified by the integer label $I \in [0,2^L-1]$.
219: Typically, only a few of these $2^L$ potential species
220: are resident in the community at any one time.
221:
222: Individual organisms of species $I$ reproduce asexually at the
223: end of each generation, each giving rise to $F$ offspring individuals
224: with probability $P_I$ before dying.
225: With probability $(1-P_I)$, they die without offspring. No individual
226: thus survives beyond one generation.
227: (For simplicity, the fecundity $F$ is assumed fixed, independent of both
228: species and individual.)
229: %In fact, all individuals of a particular species
230: %are assumed to be identical in every way.)
231:
232: During reproduction, each gene in an offspring individual's genome
233: may undergo mutation ($0 \rightarrow 1$ or $1 \rightarrow 0$)
234: with a small probability, $\mu/L$.
235: % (typically, $\mu = 10^{-3}$ is used in the simulations).
236: The mutation thus corresponds to diffusional moves from corner to corner
237: along the edges of an $L$-dimensional hypercube \cite{GAVR99,GAVR04}.
238: A mutated individual
239: is assumed to belong to a different species than its parent, with
240: different properties. Genotype and phenotype are thus in one-to-one
241: correspondence in these models.
242: This is clearly a highly idealized picture, and it is introduced
243: to maximize the pool of different species available within the computational
244: resources. The approximation is justified by
245: a large-scale computational study of
246: a version of Model A, in which species that differ by as many as
247: $L/2$ bits have correlated properties \cite{SEVI06}.
248: Quite remarkably, this study reveals that the correlated model has
249: long-time dynamical properties very similar to
250: the uncorrelated Model A studied here.
251:
252: The reproduction probability $P_I(t)$ for an individual of species $I$
253: in generation $t$ depends on the individual's ability to
254: utilize the amount of external resources available, $R$,
255: and on its interactions with the population sizes $n_J(t)$
256: of all the species present in the community at that time.
257: %(We emphasize that the population size
258: %$n_J(t)$ is the number of individuals of species $J$
259: %in generation $t$. It is thus restricted to being an integer $\ge 0$.)
260: The dependence of $P_I$ on the set of $n_J$ is determined by an {\it
261: interaction matrix\/} $\bf M$ \cite{SOLE96B}
262: with elements ${M}_{IJ} \in [-1,1]$ in a
263: way defined specifically in the next paragraph.
264: We emphasize that $\bf M$ is chosen randomly at the beginning of each
265: simulation run and is subsequently kept {\it constant\/} throughout the run.
266: If ${M}_{IJ}$ is positive and
267: ${M}_{JI}$ is negative, then $I$ is a predator and $J$ its prey, and {\it vice
268: versa\/}. If both matrix elements are positive, the species interact
269: directly in a mutualistic way, while both elements
270: negative implies direct competition.
271:
272: Specifically, the reproduction probability for species $I$, $P_I(t)$,
273: depends on $R$ and the set $\{n_J(t)\}$ through the nonlinear form,
274: \begin{equation}
275: P_I(t) = \frac{1}{1 + \exp[-\Delta_I(R,\{n_J(t)\})]} \;,
276: \label{eq:PI}
277: \end{equation}
278: with the density dependent argument
279: \begin{equation}
280: \Delta_I(R,\{n_J(t)\}) = - b_I + \eta_I R / N_{\rm tot}(t)
281: + \sum_J M_{IJ} n_J(t) / N_{\rm tot}(t) - N_{\rm tot}(t)/N_0
282: \;.
283: \label{eq:Delta}
284: \end{equation}
285: Here $b_I$ can be seen as the ``cost" of reproduction (always
286: positive), and $\eta_I$ (positive for primary producers or autotrophs,
287: and zero for consumers or heterotrophs) is the ability of individuals of
288: species $I$ to utilize the external resource $R$. The latter
289: %is an abiotic resource that
290: is renewed at the same level each generation.
291: It does {\it not\/} have independent dynamics.
292: The total population size is $N_{\rm tot}(t) = \sum_J n_J(t)$,
293: %[In contrast, the total number of {\it species\/} present in generation $t$
294: %(the species richness) will be defined as $\mathcal{N}(t)$.]
295: and the constant $N_0$ is an environmental carrying capacity
296: \cite{MURR89,VERH1838}
297: that prevents $N_{\rm tot}(t)$ from diverging to infinity.
298: It may be seen as representing implicit resource limitations not
299: explicitly included in $R$, such as available space.
300:
301: For large positive $\Delta_I$,
302: the individual almost certainly reproduces, giving rise to $F$
303: offspring, while for large negative $\Delta_I$,
304: it almost certainly dies without offspring.
305: The reproduction probability $P_I$, together with its argument
306: $\Delta_I$, play the role of a functional response for
307: this class of models \cite{DROS01B,KREB01}.
308: The two externally determined parameters that influence the population
309: size, the resource term $R$ and the carrying capacity $N_0$, play very
310: different roles. From Eq.~(\ref{eq:Delta}) it is seen that $R$ encourages
311: population growth, especially for small total population sizes, while it
312: has very little effect for large $N_{\rm tot}$. It can thus be thought
313: of as representing a finite amount of available food. The carrying
314: capacity $N_0$, on the other hand, always discourages growth, but its
315: effect is only appreciable for large population sizes. It cannot maintain
316: a nonzero population by itself and is thus best thought of as
317: representing an overall limitation, such as finite available space.
318: If $R=0$, a nonzero population can
319: therefore
320: only be maintained by
321: mutually positive interspecies interactions. This rather unrealistic
322: aspect is shared by the Tangled-nature model \cite{CHRI02,COLL03,HALL02}.
323:
324: In this work,
325: the model parameters are chosen to represent the realistic situation that
326: the number of species resident in the community at any time
327: is much smaller than the number of potential species
328: (i.e., that $\mathcal{N}(t) \ll 2^L$), and also that
329: $\mathcal{N}(t) \ll N_{\rm tot}(t)$.
330:
331: An analytic approximation describing the time development of the mean
332: population sizes (averaged over independent realizations),
333: $\langle n_I(t) \rangle$, can be
334: written as a set of coupled difference equations,
335: \begin{eqnarray}
336: \langle n_I(t+1) \rangle
337: &=& \langle n_I(t) \rangle FP_I(R,\{\langle n_J(t)\rangle \})[1-\mu ]
338: \nonumber\\
339: && +(\mu/L)F\sum_{K(I)}\langle n_{K(I)}(t)
340: \rangle P_{K(I)}(R,\{\langle n_J(t) \rangle \}) + O(\mu^2)
341: \;,
342: \label{eq:MF}
343: \end{eqnarray}
344: where $K(I)$ is the set of species that can be generated from species $I$
345: by a single mutation (``nearest neighbors" of $I$ in genotype space).
346:
347:
348: \subsection{Model A}
349: \label{sec:modA}
350: Model A was introduced and studied in \cite{RIKV03,ZIA04}.
351: In this model, the $M_{IJ}$ for $I \neq J$ are
352: stochastically independent and uniformly distributed on $[-1,+1]$,
353: while the intra-species interactions
354: $M_{II} = 0$. The external resource $R$ and the reproduction
355: costs $b_I$ are equal to zero, and the total population size $N_{\rm
356: tot}(t)$ is limited only by the carrying capacity $N_0$.
357: The model is found to evolve through a succession of quasi-stable,
358: mutualistic communities.
359:
360: \subsection{Model B}
361: \label{sec:modB}
362: Model B is a predator-prey model.
363: This is implemented by making the off-diagonal
364: part of $\bf M$ antisymmetric. In order to keep the connectance of
365: the resulting communities consistent with food webs
366: observed in nature \cite{DUNN02,GARL04},
367: the $(M_{IJ},M_{JI})$ pairs are
368: chosen nonzero with probability $c = 0.1$. The nonzero
369: elements in the upper triangle of $\bf M$ are
370: chosen independently and uniformly on $[-1,+1]$.
371: This model does not include a population-limiting carrying capacity
372: [i.e., formally, $N_0 = \infty$ in Eq.~(\ref{eq:Delta})],
373: and the community is supported by a constant external resource, $R$.
374: Only a proportion $p$ of the $2^L$ potential species are producers
375: that can directly utilize the resource
376: (for the numerical data reported here, we use $p = 0.05$).
377: Thus, with probability $(1-p)$ the resource coupling $\eta_I = 0$,
378: representing consumers, while with probability $p$ the $\eta_I$
379: are independently and uniformly distributed on $(0,+1]$, representing
380: producers of varying efficiency.
381: In addition to these constraints on $\bf M$, we require
382: that producers ($\eta_I>0$) always are the prey of consumers ($\eta_I=0$).
383: In other words: the case $\eta_I>0$ and $\eta_J=0$ with $M_{IJ} \equiv -
384: M_{JI} >0$ is forbidden.
385: Whenever it occurs, this situation is corrected during the
386: construction of $\bf M$ by reversing the signs of $M_{IJ}$ and
387: $M_{JI}$ for the pair in question.
388: Consumers at higher trophic levels are allowed, and they are indeed
389: observed in the resulting communities (see further discussion in
390: Sec.~\ref{sec:Dyn}).
391: The population sizes are limited by
392: independent reproduction costs $b_I$ that are uniformly distributed on
393: $(0,+1]$, and by negative intra-species interactions $M_{II}$
394: independently and uniformly distributed on $[-1,0)$.
395: The model is found to evolve through a succession of quasi-stable,
396: food-web like communities.
397: Some preliminary numerical results were presented in \cite{RIKV05A}.
398:
399:
400: \section{Linear Stability Analysis}
401: \label{sec:Lin}
402:
403: The form of $\Delta_I$ in Eq.~(\ref{eq:Delta})
404: represents frequency-dependent interactions that describe
405: universal competition and absence of adaptive foraging, and so is not
406: very realistic. However, the normalization by $N_{\rm tot}(t)$
407: has the advantage that it turns Eq.~(\ref{eq:MF}) with $\mu = 0$
408: into a set of linear equations for the average
409: population sizes in the fixed-point
410: community corresponding to a particular set of species.
411: The set of equations can be solved analytically to give
412: the average total population size, as well as
413: the average population size of each species
414: and the stability properties of the community.
415: This mean-field level analysis also enables us to construct a community
416: fitness
417: %potential
418: function that for a given set of species
419: is maximized by the fixed-point community, and
420: whose width is proportional to the size of the population fluctuations
421: around the fixed point, caused by the statistical birth-death process.
422: (For a detailed discussion of the fluctuations in Model A in the absence
423: of mutations, see \cite{ZIA04}.)
424: For small mutation rates, this picture remains valid as a description of
425: the population fluctuations on relatively short time scales, where the
426: species composition of the community remains constant, except for small
427: populations of unsuccessful mutants.
428:
429: \subsection{Fixed-point communities}
430: \label{sec:fpc}
431:
432: To obtain the stationary solution of Eq.~(\ref{eq:MF}) with $\mu = 0$
433: for a community of $\mathcal{N}$ species, we require $P_I=1/F$
434: for all $\mathcal{N}$ species.
435: Equations (\ref{eq:PI}) and~(\ref{eq:Delta}) then yield the
436: $\mathcal{N}$ linear relations
437: \begin{equation}
438: - \tilde{b}_I + \eta_I \frac{R}{N_{\rm tot}^*}
439: + \sum_J M_{IJ} \frac{n_J^*}{N_{\rm tot}^*}
440: - \frac{N_{\rm tot}^*}{N_0} =0
441: \;,
442: \label{eq:ss}
443: \end{equation}
444: where $\tilde{b}_I = b_I - \ln(F-1)$. (For simplicity, we
445: have dropped the $\langle \; \rangle$ notation for the average
446: population sizes. The asterisk superscripts denote fixed-point
447: solutions.) In a vector notation where $\vec{v}$ is a column
448: vector and $\vec{v}^T$ its conjugate row vector,
449: $\vec{n}^*$ is the column
450: vector composed of the $\mathcal{N}$ nonzero $n_I^*$, while
451: $\vec{1}^T$ is an $\mathcal{N}$-dimensional
452: row vector composed entirely of ones.
453: Thus, the total population size is given by
454: $N_{\rm tot}^* \equiv \sum_I n_I^* = \vec{1}^T\vec{n}^*$, and
455: Eq.~(\ref{eq:ss}) takes the matrix form
456: \begin{equation}
457: - \vec{\tilde{b}} N_{\rm tot}^*
458: + \vec{\eta} R
459: + {\hat{{\bf M}}} \vec{n}^*
460: - \vec{1} (N_{\rm tot}^*)^2/N_0 = 0
461: \;.
462: \label{eq:ss2}
463: \end{equation}
464: Here, $\vec{\tilde{b}}$ is the column vector whose elements are
465: $\tilde{b}_I$,
466: $\vec{\eta}$ is the column vector whose elements are $\eta_I$
467: (in both cases including only those $\mathcal{N}$
468: species that have nonzero $n_I^*$),
469: $\hat{\bf M}$ is the corresponding $\mathcal{N}
470: \times \mathcal{N}$ submatrix of $\bf M$, and $\vec{1}$ is an
471: $\mathcal{N}$-dimensional column vector of ones.
472: The solution for $\vec{n}^*$ is
473: \begin{equation}
474: \vec{n}^* = - {\hat{\bf M}}^{-1} \left[ \vec{\eta} R
475: - \vec{\tilde{b}} N_{\rm tot}^*
476: - \vec{1} (N_{\rm tot}^*)^2 /N_0 \right]
477: \;,
478: \label{eq:ssn}
479: \end{equation}
480: where ${\hat{\bf M}}^{-1}$ is the inverse of ${\hat{\bf M}}$.
481: (See below for a discussion of the effects of a singular $\hat{\bf M}$.)
482: To find each $n_I^*$, we must first obtain
483: $N_{\rm tot}^* \equiv \vec{1}^T \vec{n}^*$.
484: Multiplying Eq.~(\ref{eq:ssn}) from the left by $\vec{1}^T$, we
485: obtain the quadratic equation for $N_{\rm tot}$,
486: \begin{equation}
487: R \mathcal{E} + \Theta N_{\rm tot}^* - (N_{\rm tot}^*)^2/N_0 = 0
488: \;.
489: \label{eq:quad}
490: \end{equation}
491: The coefficients,
492: \begin{eqnarray}
493: \Theta =
494: \frac{1 - \vec{1}^T {\hat{\bf M}}^{-1} \vec{\tilde{b}}}
495: {\vec{1}^T {\hat{\bf M}}^{-1} \vec{1} }
496: & \;\; \mbox{\rm and} \;\; &
497: \mathcal{E} = \frac{\vec{1}^T {\hat{\bf M}}^{-1} \vec{\eta}}
498: {\vec{1}^T {\hat{\bf M}}^{-1} \vec{1} }
499: \;,
500: \label{eq:quad2}
501: \end{eqnarray}
502: can be viewed as an effective interaction strength and an effective
503: coupling to the external resource, respectively. Approximate expressions
504: for $\Theta$ and $\mathcal{E}$ that are less accurate but
505: more intuitive are obtained in Appendix~\ref{sec:AppA}.
506: The nonnegative solution of Eq.~(\ref{eq:quad}) is
507: \begin{equation}
508: N_{\rm tot}^* = \frac{\Theta N_0}{2}
509: + \sqrt{\left(\frac{\Theta N_0}{2} \right)^2
510: + R \mathcal{E} N_0 }
511: \;.
512: \label{eq:sol}
513: \end{equation}
514: Figure~\ref{fig:ntot}(a) shows $N_{\rm tot}^*$ as a function of
515: $\Theta$ for two choices of $N_0$ and $R$ at fixed $\mathcal{E}$.
516: Special cases of the solution are
517: \begin{equation}
518: N_{\rm tot}^* =
519: \left\{
520: \begin{array}{lllll}
521: \Theta {N_0}
522: & \mbox{for} & R=0 &\mbox{and}& \Theta \ge 0 \nonumber\\
523: 0
524: & \mbox{for} & R=0 &\mbox{and}& \Theta \le 0 \nonumber\\
525: \sqrt{R \mathcal{E} N_0 }
526: & \mbox{for} & \Theta = 0
527: & \mbox{and} & \mathcal{E} \ge 0
528: \nonumber\\
529: - R \mathcal{E} / \Theta
530: & \mbox{for} & N_0 = \infty & \mbox{and/or}&
531: \vec{1}^T \hat{\bf M}^{-1} \vec{1} = 0
532: \end{array}
533: \right.
534: \;.
535: \label{eq:spec}
536: \end{equation}
537: To find each $n_I^*$ separately, we now only need to insert
538: $N_{\rm tot}^*$ in Eq.~(\ref{eq:ssn}).
539:
540: \begin{figure}[tbp]
541: \begin{center}
542: \vspace*{0.3truecm}
543: \includegraphics[angle=0,width=.47\textwidth]{Phasefiga2.eps}
544: \hspace{0.5truecm}
545: \includegraphics[angle=0,width=.47\textwidth]{Phasefigb2.eps}
546: \end{center}
547: \caption[]{
548: (Color online.)
549: Solutions for {\bf (a)}
550: the total fixed-point population size $N_{\rm tot}^*$
551: and {\bf (b)} the maximum community
552: fitness
553: %potential
554: function
555: $\Phi_{\rm max} = \Phi(N_{\rm tot}^*)$. Both are shown vs $\Theta$ for fixed
556: $\mathcal{E} = 0.61$. The solid vertical lines represent the absolute
557: maximum value for $\Theta$ in Model A,
558: while the dotted vertical lines represent the analogous limit for
559: Model B. The points marked ``Model A" and ``Model B" were obtained
560: from large-scale Monte Carlo simulations as described in the text.
561: The calculation of the point marked ``Model A theoretical" is described in
562: Appendix~\protect\ref{sec:AppX}.
563: %See further discussion in the text.
564: %XXX
565: %Files in .../rikvold/scratch/ModABMFA
566: %XXX
567: }
568: \label{fig:ntot}
569: \end{figure}
570: Only those $\vec{n}^*$ that have all positive elements can represent a
571: {\it feasible\/} community \cite{ROBE74}.
572: If ${\hat{\bf M}} = {\bf 0}$ or is otherwise singular,
573: the set of equations (\ref{eq:ss2}) is
574: inconsistent for $\mathcal{N} >1$, unless $\tilde{b}_I$ and $\eta_I$
575: both are independent of $I$.
576: %(this case is equivalent to $\mathcal{N}=1$).
577: The only possible stationary community then consists of one
578: single species, the one with the largest value of $-\tilde{b}_I$ for
579: Model A or the one with the largest $\eta_I/\tilde{b}_I$ for Model B.
580: This is a trivial example of competitive
581: exclusion \cite{ARMS80,DENB86,HARD60}, and stable multispecies
582: coexistence in these models requires a nonsingular interaction
583: matrix \footnote{Neutral versions of these models can be constructed by
584: setting ${\hat{\bf M}} = {\bf 0}$, and then
585: requiring $-\tilde{b}_I$ or $\eta_I/\tilde{b}_I$ to be independent
586: of $I$ for Model A or Model B, respectively. This makes all species
587: equivalent, and these neutral models thus
588: maintain biodiversity through genetic drift in the sense of Hubbell's neutral
589: model \cite{ALON04,HUBB01,VOLK05,WILL06}.}.
590:
591: \begin{figure}[tbp]
592: \begin{center}
593: \vspace*{0.3truecm}
594: \includegraphics[angle=0,width=.47\textwidth]{phi3.eps}
595: \end{center}
596: \caption[]{
597: (Color online.)
598: The community
599: fitness
600: %potential
601: function $\Phi$, shown vs $N_{\rm
602: tot}$ for Model A (black) and Model B (gray, red online).
603: For Model A the parameters are: $F=4$, $R=0$, and $N_0=2000$.
604: For Model B they are: $F=2$, $R=2000$, $\mathcal{E} = 0.61$,
605: and $N_0=\infty$. The values of $\Theta$ used are $1.61$, corresponding
606: to the average value taken by Model A in the simulations,
607: and $-0.15$, corresponding to the average value taken by Model B.
608: $\Phi(N_{\rm tot})$ has a nontrivial maximum for Model A at positive
609: $\Theta$ (solid black curve marked A), and for Model B at negative $\Theta$
610: (solid gray curve marked B (red online)).
611: The circular and square data points are the results of Monte Carlo
612: simulations and correspond to the
613: equally shaped points in Fig.~\ref{fig:ntot}. They lie very close
614: to the maximum of $\Phi(N_{\rm tot})$ for each model.
615: The dashed curves show the physically inaccessible cases of Model A with
616: $\Theta < 0$ (stable, absorbing state at $N_{\rm tot} = 0$) and Model B with
617: $\Theta > 0$ (monotonically increasing $\Phi(N_{\rm tot})$).
618: %XXX
619: %Files in .../rikvold/scratch/Modified/PREDPREY/ETA/FIGS2005
620: %XXX
621: }
622: \label{fig:phi3}
623: \end{figure}
624: Equation~(\ref{eq:quad}) can be seen as a maximization condition
625: for a community
626: fitness
627: %potential
628: function,
629: \begin{equation}
630: \Phi(N_{\rm tot})
631: = \left( 1 - \frac{1}{F} \right)
632: \left(
633: R \mathcal{E} N_{\rm tot}
634: + \frac{\Theta}{2} N_{\rm tot}^2
635: - \frac{1}{3 N_0}N_{\rm tot}^3
636: \right)
637: \;,
638: \label{eq:fit}
639: \end{equation}
640: which is cubic in $N_{\rm tot}$ \footnote{Here we have
641: obtained Eq.~(\protect\ref{eq:fit}) simply by integration of
642: Eq.~(\protect\ref{eq:quad}).
643: A derivation that also
644: explains the prefactor $(1 - 1/F)$ and provides intuitive approximations
645: for $\Theta$ and $\mathcal{E}$ is given in
646: Appendix~\protect\ref{sec:AppA}.}.
647: The dependence of $\Phi$ on $N_{\rm tot}$ is shown in
648: Fig.~\ref{fig:phi3} for Models A and B at two different values of
649: $\Theta$. In each model, the maximum of $\Phi$ represents the
650: fixed-point value of $N_{\rm tot}$, while its width gives a measure of
651: the extent of the fluctuations about the fixed-point population in a
652: community of fixed composition \cite{ZIA04}.
653: The maximum value of $\Phi$ with respect to $N_{\rm tot}$,
654: $\Phi_{\rm max} = \Phi(N_{\rm tot}^*)$, is shown in Fig.~\ref{fig:ntot}(b)
655: vs $\Theta$ for two different values of $N_0$ at fixed $\mathcal{E}$.
656:
657: From Eq.~(\ref{eq:spec}) and Fig.~\ref{fig:ntot}(a) it is seen that
658: Model A (finite $N_0$ and $R=0$) undergoes a transcritical
659: bifurcation or exchange of stabilities \cite{CRAW91,STRO94}
660: at $\Theta = 0$. For $\Theta \le 0$, $N_{\rm tot}$ vanishes, while
661: for $\Theta > 0$ it increases linearly with $\Theta$ \footnote{In the presence
662: of a positive external resource ($R > 0$) the bifurcation for Model A
663: would become {\it imperfect\/}
664: \protect\cite{STRO94}, yielding positive
665: $N_{\rm tot}$ for all $\Theta$, as is easily seen from
666: Eq.~(\protect\ref{eq:sol}).}. Such behavior as a
667: function of a control parameter is common in many nonequilibrium
668: systems. In addition to population dynamics and the logistic map
669: (to which the present models with $\mu = 0$ are closely related),
670: these also include lasers and autocatalytic chemical reactions
671: \cite{HAKE77,STRO94}.
672: On the other hand, for Model B ($N_0 = \infty$ and $R>0$),
673: $N_{\rm tot}$ diverges to infinity as $\Theta$ approaches zero from below,
674: and it would be infinite for $\Theta \ge 0$.
675:
676: The results discussed in the previous paragraphs would not be
677: of evolutionary relevance
678: if $\Theta$ were just an externally fixed control parameter, as it would
679: be in the absence of extinctions and mutations.
680: However, mutations enable the community not only to
681: maximize $\Phi(N_{\rm tot})$ for fixed parameters, {\it but also
682: to change the parameters in response to new mutations, leading to the
683: possibility of further increasing $\Phi_{\rm max}$\/}.
684: Numerical results from the large-scale kinetic Monte Carlo simulations
685: discussed in Sec.~\ref{sec:Dyn} show that, on average,
686: the communities initially progress toward,
687: and later settle down to fluctuate near, the
688: maximum value of $\Theta$ (and thus the maximum value of $\Phi_{\rm max}$)
689: compatible with the constraints on $\bf M$ and $\tilde{\vec b}$.
690: These constraints depend on the specific model as follows.
691: For Model A, which allows direct mutualistic interactions
692: (i.e., $M_{IJ}$ and $M_{JI}$ both positive), $\Theta$ can be positive, thus
693: enabling communities with nonzero population size, even for $R=0$.
694: For the predator-prey
695: Model B, on the other hand, the antisymmetric structure of $\bf M$ forces
696: $\Theta$ to be nonpositive.
697: A detailed discussion of the time evolution of the mean-field parameters,
698: $\Theta$, $\mathcal{E}$, $N_{\rm tot}^*$, and $\Phi_{\rm max}$, is given
699: for both models in Sec.~\ref{sec:Dyn1}.
700:
701: In Monte Carlo simulations of Model A (in which the off-diagonal
702: elements of $\bf M$ are uncorrelated and
703: uniformly distributed on $[-1,+1]$), it was found that,
704: after the initial period characterized by an average positive trend in
705: the mean-field parameters,
706: the community spent most of its time in a succession of
707: quasi-steady states (QSSs), separated by brief bursts of intense
708: evolutionary activity \cite{RIKV03}.
709: All the QSSs studied in detail were found to be
710: mutualistic, with $\overline{M_{IJ}} = 0.78\pm0.03$.
711: [Here, the overbar represents averages over all the ten QSSs listed in
712: Table~I of \cite{RIKV03}.] The average of $N_{\rm tot}$, taken
713: over all the 16 realizations of $2^{25}$ generations that
714: were studied, was ${\overline{N_{\rm tot}}} = 3201\pm8$.
715: (The average over only the ten QSSs agrees with the total average to within the
716: statistical errors, showing that the periods when the system is not
717: in a QSS contribute negligibly to the overall time averages.)
718: From these averages we can use Eqs.~(\ref{eq:spec}) and~(\ref{eq:fit})
719: to estimate the average mean-field quantities,
720: $\overline{\Theta} = 1.61\pm0.01$ and
721: $\overline{\Phi_{\rm max}} = (2.09 \pm 0.06)\times 10^6$.
722: In Fig.~\ref{fig:ntot}(a), ${\overline{N_{\rm tot}}}$ is shown vs
723: $\overline{\Theta}$ as a black dot, while the
724: corresponding value of $\overline{\Phi_{\rm max}}$ is shown
725: the same way in Fig.~\ref{fig:ntot}(b).
726: %A typical time series of $N_{\rm tot}(t)$ for Model A is shown in
727: %Fig.~\ref{fig:timeser}(a).
728:
729: The results for Model B that are included in Fig.~\ref{fig:ntot}
730: were also obtained from Monte Carlo simulations of 2$^{25}$ generations.
731: (See Sec.~\ref{sec:Dyn} for further details.)
732: The parameter values used in the figure
733: were extracted as the average values from thirteen
734: QSS communities identified at late times in twelve independent simulation runs:
735: $\overline{\mathcal{E}} = 0.61\pm0.04$
736: %data from /rikvold/scratch/Modified/PREDPREY/ETA/1X_1Xeta_Run02.dat
737: and $\overline{\Theta} = - 0.15\pm0.01$.
738: The value of the total population size
739: given in the figure is the total time average
740: over the simulations, averaged over all twelve runs,
741: ${\overline{N_{\rm tot}}} = 7726 \pm 303$.
742: (Like for Model A, averages over only the
743: particular observed QSS communities
744: agree with this total average to within the error bars.)
745: %XXX Data in /rikvold/scratch/ModABMFA
746: The large uncertainty in ${\overline{N_{\rm tot}}}$
747: is a direct consequence of the steep slope of $N_{\rm tot}^*$ versus
748: $\Theta$ closely below $\Theta = 0$
749: for Model B (see Fig.~\ref{fig:ntot}(a)).
750: From these estimates and Eq.~(\ref{eq:fit}) we further get the estimate
751: $\overline{\Phi_{\rm max}} = (2.49 \pm 0.18)\times 10^6$.
752:
753:
754: \subsection{Stability of fixed-point communities}
755: \label{sec:stab}
756:
757: The internal stability of an $\mathcal{N}$-species
758: fixed-point community is obtained from the matrix of partial derivatives,
759: \begin{equation}
760: \left. \frac{\partial n_I(t+1)}{\partial n_J(t)} \right|_{\vec{n}^*}
761: =
762: \delta_{IJ} + \Lambda_{IJ} \;,
763: \label{eq:stab}
764: \end{equation}
765: where $\delta_{IJ}$ is the Kronecker delta and
766: $\Lambda_{IJ}$ are elements of the {\it community matrix\/}
767: $\bf \Lambda$ \cite{MURR89}. Straightforward differentiation yields
768: \begin{equation}
769: \Lambda_{IJ} = \left( 1 - \frac{1}{F} \right) \frac{n_I^*}{N_{\rm tot}^*}
770: \left[ M_{IJ}
771: - \frac{R \eta_I +({\hat{\bf M}}\vec{n}^*)_I}{N_{\rm tot}^*}
772: - \frac{1}{N_0}
773: \right]
774: \;,
775: \label{eq:lam}
776: \end{equation}
777: where $({\hat{\bf M}}\vec{n}^*)_I$ is the element of
778: ${\hat{\bf M}}\vec{n}^*$ corresponding to species $I$.
779: In order for deviations from
780: the fixed point to decay monotonically in magnitude,
781: the magnitudes of the eigenvalues of the matrix of partial derivatives
782: in Eq.~(\ref{eq:stab}), ${\bf \Lambda} + {\bf 1}$ where $\bf 1$ is the
783: $\mathcal{N}$-dimensional unit matrix, must be less than unity.
784: The values of the
785: fecundity $F$ that are used in this work (4 for Model A and 2 for
786: Model B) were chosen to satisfy this requirement for $\mathcal{N}=1$.
787:
788: Since new species are created by mutations, we must also study the
789: stability of the fixed-point community toward ``invaders."
790: Consider a mutant invader $i$.
791: Then its multiplication rate, in the limit that
792: $n_i \ll n_J$ for all $\mathcal{N}$
793: species $J$ in the resident community, is given by
794: \begin{equation}
795: \frac{n_i(t+1)}{n_i(t)}
796: =
797: \frac{F}{1 + \exp \left[ -\Delta_i(R,\{n_J^*\}) \right]}
798: \;.
799: \label{eq:inv}
800: \end{equation}
801: The Lyapunov exponent, $\ln[ n_i(t+1)/n_i(t) ]$, is the {\it
802: invasion fitness\/} of the mutant with respect to the resident community
803: \cite{DOEB00,METZ92}.
804: A characteristic feature of the QSS communities
805: observed in both Model A and Model B is that only about 1\%
806: of the mutants that are separated from the resident community
807: by a single mutation (``nearest-neighbor species") have multiplication rates
808: above unity (and most of those are between 1.0 and 1.1).
809: In fact, this is true only to a slightly lesser degree for
810: mutants separated by two or three mutations from the resident species
811: (``next-nearest neighbors" and ``third-nearest neighbors") \cite{RIKV03}.
812: Thus, a string of rather unsuccessful, or at best neutral,
813: mutations is necessary to bring significant change
814: to a QSS community -- a fact that to a large extent accounts for their
815: high degree of stability.
816: %This effect is illustrated for both models in Fig.~\ref{fig:invader}.
817:
818: \begin{figure}[t]
819: \begin{center}
820: \vspace*{0.3truecm}
821: \includegraphics[angle=0,width=.47\textwidth]{ThetafigA.eps}
822: \hspace{0.5truecm}
823: \includegraphics[angle=0,width=.47\textwidth]{NPhifigA.eps}
824: \end{center}
825: \caption[]{
826: (Color online.)
827: Typical time series of the mean-field, fixed-point parameters for Model A
828: ($N_0=2000$, $R=0$, $F=4$, $L=13$, and $\mu = 10^{-3}$),
829: emphasizing the transient early-time regime by using a logarithmic time axis.
830: {\bf (a)}
831: The effective interaction strength $\Theta$. The solid, horizontal line
832: is the absolute upper bound on $\Theta$ for Model A, $(1+\ln 3) \approx 2.10$,
833: while the dashed, horizontal line is the most probable value, 1.618.
834: Both are derived in Appendix~\protect\ref{sec:AppX}.
835: {\bf (b)}
836: The fixed-point population size $N_{\rm tot}^*$ and the corresponding
837: maximum value of the community
838: fitness
839: %potential
840: function, $\Phi_{\rm max}$. The
841: dashed, horizontal lines are the corresponding
842: values obtained from the most probable
843: value of $\Theta$. For comparison, the simulated total population size
844: (sampled on a much finer time scale) is shown light gray (green online)
845: in the background.
846: %This figure is found in
847: %/home/r3/rikvold/scratch/33M/NEW/FIXPTS
848: }
849: \label{fig:transA}
850: \end{figure}
851: \begin{figure}[ht]
852: \begin{center}
853: \vspace*{0.3truecm}
854: \includegraphics[angle=0,width=.47\textwidth]{ThetaEfig.eps}
855: \hspace{0.5truecm}
856: \includegraphics[angle=0,width=.47\textwidth]{NPhifig.eps}
857: \end{center}
858: \caption[]{
859: (Color online.)
860: Typical time series of the mean-field, fixed-point parameters for Model B
861: ($N_0=\infty$, $R=2000$, $F=2$, $L=13$, and $\mu = 10^{-3}$),
862: emphasizing the transient early-time regime by using a logarithmic time axis.
863: {\bf (a)}
864: The effective interaction strength $\Theta$ and the effective
865: coupling to the external resource, $\mathcal{E}$.
866: The solid, horizontal line at zero is the absolute upper bound on
867: $\Theta$ for Model B,
868: while the dashed, horizontal lines are the averages over
869: late-time communities, also shown as points in Figs.~\protect\ref{fig:ntot}
870: and~\protect\ref{fig:phi3}.
871: {\bf (b)}
872: The fixed-point population size $N_{\rm tot}^*$ and the corresponding
873: maximum value of the community
874: fitness
875: %potential
876: function, $\Phi_{\rm max}$.
877: The dashed, horizontal lines are obtained the same way as in (a).
878: For comparison, the simulated total population size
879: (sampled on a much finer time scale) is shown light gray (green online)
880: in the background.
881: %This figure is found in
882: %/home/r3/rikvold/scratch/Modified/PREDPREY/ETA
883: }
884: \label{fig:transB}
885: \end{figure}
886:
887:
888: \section{Comparison of Dynamical Features}
889: \label{sec:Dyn}
890:
891: In this section we go beyond the mean-field treatment of the previous
892: sections to compare and contrast
893: some of the dynamical features observed in long
894: kinetic Monte Carlo simulations of Model A
895: and Model B. For both models we performed multiple
896: simulations of $2^{25} = 33\,554\,432$~generations \footnote{The
897: simulation length was chosen as a power of two to enable use of the
898: Fast Fourier Transform algorithm in calculating power spectral densities
899: \protect\cite{PRES92}.}.
900: with a genome of length $L=13$ ($2^{13} = 8192$ potential species)
901: and a mutation rate per individual of $\mu = 10^{-3}$. The
902: fecundity $F$ was set to 4 for Model A and 2 for Model B.
903:
904: The model parameters used in our simulations are chosen for
905: computational feasibility with a view to keeping the system in the
906: realistic regime where $\mathcal{N}(t) \ll 2^L$ and $\mathcal{N}(t) \ll
907: N_{\rm tot}(t)$ at all times. At the same time, we want to
908: study the dynamics during the late-time, statistically
909: stationary regime, which should
910: therefore be reached in a reasonable time. As a result of these
911: restrictions, $L$, $R$, and $N_0$ are quite small, while $\mu$ is quite
912: large, compared to natural populations. Results of tests for Model B with
913: larger $L$ and $R$ and smaller $\mu$ are summarized in Appendix~\ref{sec:AppB}.
914: Analogous, but less exhaustive tests for Model A (not reported in detail
915: here, but see brief summary in Appendix~\ref{sec:AppB}) yield similar results.
916: These tests indicate that the products $R \mu$ for Model B and $N_0 \mu$
917: for Model A are
918: proportional to the average number of mutant individuals produced per
919: generation, $\overline{N_{\rm tot}} \mu$. We also note that this number
920: is half of
921: the {\it universal biodiversity number\/} in Hubbell's neutral theory of
922: biodiversity \cite{HUBB01}. Even though the species in the
923: models studied here are strongly interacting, this parameter appears to
924: play a similar role in determining the average diversity
925: as it does in the neutral theory. Furthermore, the
926: power laws that are observed and discussed in Sec.~\ref{sec:Dyn2}
927: are found to be robust, even against parameter changes that change $R \mu$.
928:
929:
930: \subsection{Evolution of mean-field parameters in the transient regime}
931: \label{sec:Dyn1}
932:
933: The simulation runs were started with 100 individuals of
934: a single species (or of a single producer species for Model B) and allowed to
935: evolve under the dynamics described in Sec.~\ref{sec:Mod}. At time
936: points sampled at approximately constant intervals on a logarithmic
937: scale, the ``stable core" of the community was identified as follows.
938: The fixed-point population sizes for the species present in the
939: community at that time were calculated according to Eq.~(\ref{eq:ssn}),
940: and species corresponding to negative population sizes were excluded
941: (starting from the one with the smallest simulated population size)
942: until a stable, feasible community was obtained. This procedure was found to
943: essentially correspond to removing species with populations below 100,
944: which mostly were unsuccessful mutants of the ``core" species. Once the
945: core community was identified, the corresponding
946: mean-field parameters $\Theta$,
947: $\mathcal{E}$ (for Model B only), $N_{\rm tot}^*$, and $\Phi_{\rm max}$
948: were calculated from Eqs.~(\ref{eq:quad2}) through (\ref{eq:fit}).
949:
950: Typical time series of these mean-field parameters are shown in
951: Figs.~\ref{fig:transA} and~\ref{fig:transB}, in which the early-time
952: transient regime is emphasized by using a logarithmic time axis.
953: For both models a clear tendency is seen for the mean-field parameters to
954: increase during the transient regime, for so to fluctuate randomly about
955: their long-time averages during the the later-time, statistically
956: stationary regime. Both models thus evolve from an initial
957: low-population community, toward a sequence of
958: QSS communities that optimize their population sizes,
959: subject to the constraints inherent in the respective models.
960:
961: \subsection{Long-time dynamics in the statistically stationary regime}
962: \label{sec:Dyn2}
963:
964: Here we concentrate on comparing and contrasting the dynamical properties
965: of the two models in the statistically stationary regime that follows
966: the transient regime discussed above. The emphasis will be on the
967: temporal fluctuations of the total population sizes and the diversities
968: of the resulting communities.
969: \begin{figure}[t]
970: \begin{center}
971: \vspace*{0.3truecm}
972: \includegraphics[angle=0,width=.47\textwidth]{PopfigA_JTB.eps}
973: \hspace{0.5truecm}
974: \includegraphics[angle=0,width=.47\textwidth]{PopfigB_JTB.eps}
975: \end{center}
976: \caption[]{
977: (Color online.)
978: Typical time series of the total population sizes, $N_{\rm tot}(t)$,
979: for Monte Carlo runs of $2^{25} = 33\,554\,432$ generations each.
980: To reduce the size of the figure file, the data are sampled only
981: once every 2048 generations. While this slightly reduces the
982: apparent range of the short-time fluctuations, the general shapes of
983: the time series are preserved. The data are from the same runs shown in
984: Figs.~\protect\ref{fig:transA} and~\protect\ref{fig:transB}.
985: {\bf (a)}
986: Model A.
987: %: $N_0=2000$, $R=0$, $F=4$, $L=13$, and $\mu = 10^{-3}$.
988: {\bf (b)}
989: Model B.
990: %: $N_0=\infty$, $R=2000$, $F=2$, $L=13$, and $\mu = 10^{-3}$.
991: %See discussion in the text.
992: %XXX
993: %This figure should be Pops-underbar-talk.eps, found in
994: %/home/r3/rikvold/scratch/Modified/PREDPREY/ETA/SPIEfigs/Pops-underbar-talk.xmg
995: }
996: \label{fig:timeser}
997: \end{figure}
998:
999: Typical time series of the total population sizes
1000: $N_{\rm tot}(t)$ are shown in Fig.~\ref{fig:timeser}.
1001: They are characterized by QSSs on different time scales,
1002: separated by periods of high evolutionary activity.
1003: In addition to the total population sizes,
1004: we also studied the diversities, defined as
1005: the number of major resident species. In order to obtain an
1006: approximation for the number of major, or ``core"
1007: species (which can be thought of as the wildtypes in a quasi-species model
1008: \cite{EIGE71,EIGE88}), we filter out the low-population
1009: species that are most likely unsuccessful mutants of the
1010: wildtypes. This is achieved by using the exponential
1011: Shannon-Wiener diversity index
1012: \cite{KREB89},
1013: \begin{equation}
1014: D(t) = e^{S[\{n_I(t)\}]} \;,
1015: \label{eq:div}
1016: \end{equation}
1017: where
1018: \begin{equation}
1019: S[\{n_I(t)\}] = - \sum_{\{ I | \rho_I(t) > 0 \}}
1020: \rho_I(t) \ln \rho_I(t)
1021: \label{eq:ent}
1022: \end{equation}
1023: with $\rho_I(t) = n_I(t)/N_{\rm tot}(t)$ is the
1024: information-theoretical entropy \cite{SHAN48,SHAN49}.
1025: Typical time series for $D(t)$ in the two models are shown in
1026: Fig.~\ref{fig:timeserD}. Just like in the time series for the total
1027: population sizes, the intermittent
1028: structure consisting of QSSs on different timescales,
1029: separated by periods of high evolutionary activity, is clearly seen.
1030: \begin{figure}[t]
1031: \begin{center}
1032: \vspace*{0.4truecm}
1033: \includegraphics[angle=0,width=.47\textwidth]{DiversityfigA_JTB.eps}
1034: \hspace{0.5truecm}
1035: \includegraphics[angle=0,width=.47\textwidth]{DiversityfigB_JTB.eps}
1036: \end{center}
1037: \caption[]{
1038: (Color online.)
1039: Typical time series of the exponential Shannon-Wiener diversity index $D(t)$
1040: in the same $2^{25}$-generations Monte Carlo runs shown in
1041: Fig.~\protect\ref{fig:timeser}.
1042: The data were downsampled as in Fig.~\protect\ref{fig:timeser}.
1043: %See discussion in the text.
1044: {\bf (a)}
1045: Model A.
1046: {\bf (b)}
1047: Model B.
1048: %XXX
1049: %This figure should be Diversityfig_talk.eps, found in
1050: %/home/r3/rikvold/scratch/Modified/PREDPREY/ETA/SPIEfigs/Diversityfig_talk.xmg
1051: }
1052: \label{fig:timeserD}
1053: \end{figure}
1054:
1055: \begin{figure}[t]
1056: \begin{center}
1057: \vspace*{0.3truecm}
1058: \includegraphics[angle=0,width=.47\textwidth]{FigHISTexpS33M.eps}
1059: \hspace{0.5truecm}
1060: \includegraphics[angle=0,width=.47\textwidth]{FigQuiet_02_DS_talk.eps}
1061: \end{center}
1062: \caption[]{
1063: (Color online.)
1064: {\bf (a)}
1065: Normalized histograms representing
1066: the probability density of the logarithmic derivative
1067: of the diversity, ${\rm d}S(t)/{\rm d}t$. The data were averaged over 16
1068: generations in each run, and then averaged over 16 independent runs for
1069: Model A (solid, marked A) and 12 runs for Model B (dashed, marked B).
1070: The central parts of both histograms are well fitted by a
1071: Gaussian (dotted, marked G).
1072: %After \protect\cite{RIKV05A}.
1073: {\bf (b)}
1074: Log-log plot of normalized
1075: histograms representing the probability density of the
1076: durations of QSSs, estimated as
1077: the periods between times when $|{\rm d}S(t)/{\rm d}t|$ exceeds a cutoff $y_c$
1078: (${\rm d}S(t)/{\rm d}t$ was averaged over 16 generations as in part (a)).
1079: Model A (solid)
1080: and Model B (dashed). Results for Model A were averaged over 16 runs, and
1081: for Model B over 12 runs. The error bars are standard errors, based on the
1082: spread between runs. The two dot-dashed straight lines represent time$^{-2}$
1083: and time$^{-1}$ power laws, respectively.
1084: %After \protect\cite{RIKV05A}.
1085: }
1086: \label{fig:quiet}
1087: \end{figure}
1088: Statistical information for characteristic time intervals
1089: that describe the dynamics can be extracted from our data. One such time
1090: interval is the duration of a QSS. One way to determine
1091: this is to use a cutoff on the magnitude of the diversity fluctuations, whose
1092: probability densities for the two models are
1093: shown in Fig.~\ref{fig:quiet}(a). For both models, the probability densities
1094: consist of a Gaussian central part representing the fluctuations during the
1095: QSS periods \cite{ZIA04},
1096: flanked by ``wings" that correspond to the large fluctuations during the
1097: evolutionarily active periods. We choose
1098: a cutoff $y_c = 0.015$ for Model A and 0.010 for Model B, in both
1099: cases corresponding to the transition region between the Gaussian
1100: central peak and the wings. The
1101: duration of a single QSS then corresponds to the time interval
1102: between consecutive times when $|{\rm d}S(t)/{\rm d}t|$ exceeds $y_c$.
1103: Log-log plots of the resulting probability densities for the durations of
1104: QSSs in the two models are shown together in Fig.~\ref{fig:quiet}(b).
1105: While both show approximate power-law behavior over
1106: five decades or more in time,
1107: there is an important difference: the exponent
1108: for Model A is near $-2$, while for Model B it is closer to $-1$.
1109:
1110: \begin{figure}[t]
1111: \begin{center}
1112: \vspace*{0.3truecm}
1113: \includegraphics[angle=0,width=.47\textwidth]{AvHistLifeAB_talk.eps}
1114: \end{center}
1115: \caption[]{
1116: (Color online.)
1117: Log-log plot of normalized
1118: histograms representing the probability density of the
1119: lifetime of a particular species. Model A (solid)
1120: and Model B (dashed). Results for Model A were averaged over eight runs,
1121: and for Model B over 12 runs.
1122: The error bars were calculated as in Fig.~\protect\ref{fig:quiet}(b).
1123: The dot-dashed straight line represents a time$^{-2}$ power law.
1124: %After \protect\cite{RIKV05A}.
1125: }
1126: \label{fig:life}
1127: \end{figure}
1128: A different time interval
1129: of interest is the lifetime of a particular species,
1130: defined as the time between its origination and eventual
1131: extinction. Log-log plots of histograms of the species lifetimes in the
1132: two models are shown in Fig.~\ref{fig:life}. In contrast to the case of
1133: the QSS durations, the species-lifetime distributions are very close for the
1134: two models, both showing approximate time$^{-2}$ behavior over near
1135: seven decades in time.
1136: The observed exponent
1137: is significantly different from $-3/2$, which would correspond to the
1138: simple hypothesis that the lifetime distributions simply
1139: correspond to the first-return-time distribution for a random walk of
1140: $n_I$ \cite{NEWM05,NEWM03}.
1141: On the other hand, a lifetime distribution with an exponent of $-2$ is
1142: consistent with a stochastic branching process \cite{PIGO05}.
1143: (Simulations of neutral versions of both models also yield
1144: species lifetime distributions that decay approximately as
1145: time$^{-2}$, but with a sharp cutoff near 10$^4$ generations.)
1146: Lifetime distributions for marine genera that are compatible
1147: with a power law with an exponent in the range $-1.5$ to $-2$
1148: have been obtained from the fossil record \cite{NEWM03,NEWM99B}.
1149: However, the possible power-law behavior in the fossil record
1150: is only observed over about one decade in
1151: time -- between 10 and 100 million years -- and other fitting functions,
1152: such as exponential decay, are also possible.
1153: Nevertheless, it is reasonable to conclude that the numerical results
1154: obtained from complex, interacting evolution models that extend
1155: over a large range of time scales support interpretations of the fossil
1156: lifetime evidence in terms of nontrivial power laws.
1157:
1158: \begin{figure}[t]
1159: \begin{center}
1160: \vspace*{0.3truecm}
1161: \includegraphics[angle=0,width=.47\textwidth]{species_33M_S.eps}
1162: \hspace{0.5truecm}
1163: \includegraphics[angle=0,width=.47\textwidth]{Species_02Dprod.eps}
1164: \end{center}
1165: \caption[]{
1166: (Color online.)
1167: Plots showing the labels $I$ of the most highly populated species vs time.
1168: The data are for the same $2^{25}$-generation Monte Carlo runs shown in
1169: Figs.~\protect\ref{fig:transA}--\protect\ref{fig:timeserD}.
1170: Gray (green online): $n_I \in [101,1000]$.
1171: Black: $n_I \ge 1001$.
1172: {\bf (a)} Model A.
1173: {\bf (b)} Model B. For clarity, only producer species are shown
1174: for this model. The consumer species give a similar picture.
1175: See further discussion in the text.
1176: %XXX Redraw from xmg files of same or similar names
1177: % in .../scratch/Modified/PREDPREY/ETA/SPIEfigs
1178: % and in .../scratch/Modified/PREDPREY/ETA/FIGS2005
1179: % XXX
1180: }
1181: \label{fig:specs}
1182: \end{figure}
1183: \begin{figure}[t]
1184: \begin{center}
1185: \vspace{0.5truecm}
1186: \includegraphics[angle=0,width=.47\textwidth]{species_33M_S_magn.eps}
1187: \hspace{0.5truecm}
1188: \includegraphics[angle=0,width=.47\textwidth]{Species_02Dprod_magn.eps}
1189: \end{center}
1190: \caption[]{
1191: (Color online.)
1192: Magnified version of Fig.~\protect\ref{fig:specs}, showing times
1193: between $1.0\times 10^7$ and $1.6\times 10^7$ generations.
1194: In addition to the species shown in Fig.~\protect\ref{fig:specs},
1195: the ones with $n_I \in [11,100]$ are also shown (light gray,
1196: yellow online).
1197: {\bf (a)} Model A.
1198: {\bf (b)} Model B.
1199: See further discussion in the text.
1200: %XXX Redraw from xmg files of same or similar names
1201: % in .../scratch/Modified/PREDPREY/ETA/SPIEfigs
1202: % and in .../scratch/Modified/PREDPREY/ETA/FIGS2005
1203: % XXX
1204: }
1205: \label{fig:specs2}
1206: \end{figure}
1207: The difference in the power laws for the QSS durations and the species
1208: lifetimes is a puzzling result. A likely explanation can be gleaned
1209: from the data shown in Figs.~\ref{fig:specs} and~\ref{fig:specs2}.
1210: These show the species labels
1211: of highly populated species as functions of time for both models. From
1212: Figs.~\ref{fig:specs}(a) and~\ref{fig:specs2}(a)
1213: it is seen that all or most of the horizontal lines representing
1214: populated species at a given time for Model A start
1215: and stop almost simultaneously, indicating that species originations
1216: and extinctions in this model are highly synchronized.
1217: In other words: whole communities in Model A tend to go extinct and be
1218: replaced with an entirely new community within a short time.
1219: In contrast, in Figs.~\ref{fig:specs}(b) and~\ref{fig:specs2}(b)
1220: the different horizontal species
1221: lines for Model B stop and start at different times. This indicates that
1222: communities in this model are much more robust, and extinction
1223: events seldom wipe out more than a part of the total community. Thus, QSSs
1224: would be expected to be more long-lived (but also less clearly defined)
1225: for Model B, than for Model A. This theory is supported by the
1226: strikingly different QSS community structures produced by the two
1227: models, shown in Fig.~\ref{fig:Webs}. The typical QSS community
1228: shown for Model A is a small cluster of mutualistically interacting
1229: species, while the typical community shown for Model B has the
1230: character of a simple food web. Extinctions in the latter are
1231: likely to be confined to a single branch of the web.
1232: A detailed discussion of the community structures generated by Model B,
1233: including comparisons with data from real food webs, is found in \cite{RIKV06B}.
1234:
1235: \begin{figure}[t]
1236: \begin{center}
1237: \vspace*{0.3truecm}
1238: \includegraphics[angle=0,width=.47\textwidth]{ModelARun10_fig.eps}
1239: \hspace{0.5truecm}
1240: \includegraphics[angle=0,width=.47\textwidth]{ModelBRunD1_fig.eps}
1241: \end{center}
1242: \caption[]{
1243: (Color online.)
1244: Typical QSS core communities observed in simulations of the two models.
1245: The numbers are the species labels $I$, and the sizes of the circles
1246: represent the individual species' fixed-point populations,
1247: obtained from Eq.~(\protect\ref{eq:ssn}).
1248: A black arrow pointing toward species $I$ indicates a positive
1249: $M_{IJ}$, with magnitude indicated by the line thickness and size
1250: of the arrow head. The thin gray lines without arrowheads
1251: (red online) connect nearest neighbors in genotype space.
1252: (a)
1253: Model A. This particular QSS community is the one included in
1254: Table 2 of \protect\cite{ZIA04} and line 10 of Table I of
1255: \protect\cite{RIKV03}.
1256: (b)
1257: Model B. This is the food web representing the
1258: QSS community existing between 15 and 18 million
1259: generations in the simulation shown in
1260: Figs.~\protect\ref{fig:transB}, \protect\ref{fig:timeser}(b),
1261: \protect\ref{fig:timeserD}(b),
1262: \protect\ref{fig:specs}(b), and~\protect\ref{fig:specs2}(b).
1263: It has three trophic levels above the resource node.
1264: %XXX
1265: % in .../scratch/Modified/PREDPREY/ETA/SPIEfigs
1266: % XXX
1267: }
1268: \label{fig:Webs}
1269: \end{figure}
1270:
1271: \begin{figure}[t]
1272: \begin{center}
1273: \vspace*{0.3truecm}
1274: \includegraphics[angle=0,width=.47\textwidth]{AvPSD_fig_talk.eps}
1275: \hspace{0.5truecm}
1276: \includegraphics[angle=0,width=.47\textwidth]{AvPSD_pop_fig_talk.eps}
1277: \end{center}
1278: \caption[]{
1279: (Color online.)
1280: Log-log plots of
1281: PSDs of {\bf (a)} the diversities $D(t)$ and {\bf (b)}
1282: the total population sizes $N_{\rm tot}(t)$ for Model A and Model B.
1283: The PSDs are averaged over each octave in frequency, and then
1284: over 16 runs for Model A and 12 runs for Model B.
1285: The error bars were calculated as in Fig.~\protect\ref{fig:quiet}(b).
1286: The dashed straight lines represent $1/f$ power-law behavior.
1287: %XXX
1288: % in .../scratch/Modified/PREDPREY/ETA/SPIEfigs
1289: % XXX
1290: }
1291: \label{fig:PSD}
1292: \end{figure}
1293:
1294: The arbitrariness inherent in the cutoff that must be used to extract
1295: QSS duration distributions from fluctuations in the diversity or other time
1296: series can to some extent be eliminated by mapping distributions obtained
1297: with different cutoffs onto a common scaling function
1298: \cite{PACZ96,RIKV05A}. However, an analysis method that completely
1299: avoids cutoffs is that of calculating power spectral densities
1300: (the square of the temporal Fourier transform). Power spectra
1301: (PSDs) are therefore shown in
1302: Fig.~\ref{fig:PSD} for both models. The PSDs for the diversity are
1303: shown in Fig.~\ref{fig:PSD}(a), and for the total population size in
1304: Fig.~\ref{fig:PSD}(b).
1305: Although there are clear deviations, the overall behaviors for both quantities
1306: and for both models are compatible with a $1/f$ power law over many
1307: decades in frequency. In the
1308: high-frequency regime the population-size PSDs have a significant background of
1309: noise, presumably caused by the rapid population fluctuations due to the
1310: birth and death of individual organisms.
1311: For very low frequencies there is little reason to believe that there should
1312: be large differences between the behaviors of the two quantities for the same
1313: model. We therefore think it is reasonable to consider the difference between
1314: the slopes of
1315: the diversity and population-size PSDs as an indication of the true
1316: uncertainty in the PSDs at the lowest frequencies. Better estimates in this
1317: regime would require several orders of magnitude longer simulations.
1318:
1319: \section{Discussion and Conclusions}
1320: \label{sec:Conc}
1321:
1322: In this paper we have compared and contrasted the long-time evolutionary
1323: dynamics of two individual-based models of biological coevolution,
1324: using both analytic linear stability calculations and large-scale
1325: kinetic Monte Carlo simulations.
1326: These models involve universal competition and ignore
1327: important effects such as adaptive foraging. They are therefore not
1328: highly realistic. However, the fact that the results of numerical
1329: simulations can be compared with exact
1330: analytical results make these simplified
1331: models ideal as benchmarks for simulations of more realistic
1332: coevolution models in the future \cite{RIKV06C}.
1333:
1334: A central result of the analytic study is that, in the
1335: absence of mutations, the total
1336: population size of a fixed-point community, $N_{\rm tot}^*$,
1337: is described by a community
1338: fitness
1339: %potential
1340: function that is cubic in $N_{\rm tot}$.
1341: In addition to the present application, such a model
1342: is also applicable to nonequilibrium phase transitions in
1343: such diverse systems as epidemics,
1344: lasers, and autocatalytic chemical reactions. However, these evolution
1345: models differ from those kinds of systems by the
1346: important effect that, in the {\it presence\/} of mutations,
1347: the model parameters $\Theta$ and $\mathcal{E}$ are no longer
1348: externally imposed constraints, but rather {\it evolve\/}
1349: as far in the direction of positive $\Theta$ and large $\mathcal{E}$
1350: as allowed by the internal constraints of the particular model.
1351: As a result, Model A, in which the elements of the
1352: interspecies interaction matrix $\bf M$ are randomly distributed on
1353: an interval that is symmetric about zero,
1354: evolves to produce communities that
1355: are heavily biased toward mutualism. The effective interaction variable
1356: $\Theta$ adjusts to a positive value.
1357: %, where a community of nonzero population size can exist
1358: %{\it without\/} an external resource.
1359: In contrast, the predator-prey Model B, in which the
1360: interspecies interactions
1361: are antisymmetric, is constrained to nonpositive values of
1362: $\Theta$, for which a nonzero population size can
1363: only be sustained through the external resource $R$.
1364: These results are illustrated in Fig.~\ref{fig:ntot}(a).
1365:
1366: In a recent study of a similar, but different model for coevolution
1367: \cite{TOKI03}, the emergence of strongly mutualistic
1368: communities from initially unbiased conditions was also observed.
1369: In that model, mutants
1370: are very similar to their parents, except for their interactions with a
1371: few other species (``local mutations"),
1372: and the authors suggest that the evolution of mutualism
1373: is related to this feature of their model.
1374: However, the mutations in our models would be ``global" in the language of
1375: \cite{TOKI03}, which leads us to the conclusion that the emergence of
1376: mutualism is common in models where direct
1377: mutualistic interactions are allowed.
1378: Rather remarkably, we have found little difference in the dynamics between
1379: the version of Model A studied in this paper and a version with
1380: strongly correlated interactions, and thus more ``local" mutations
1381: \cite{SEVI06}.
1382: It remains an important problem to reconcile this tendency for evolution
1383: of mutualism with the obvious requirement that biomass cannot be created
1384: without energy input.
1385: While predator-prey interactions are easy to reconcile
1386: with energy conservation, direct mutualistic interactions
1387: are more difficult to interpret in an energy framework \cite{BRON94}
1388: (although effective mutualism is
1389: common in nature \cite{BASC06,BRON94,KAWA93,KREB01}).
1390: We believe this emphasizes the importance of distinguishing between
1391: {\it direct\/} mutualistic interactions, as in Model A, and more realistic
1392: {\it effective\/} mutualisms, which merely mean that a pair of elements in
1393: the community matrix, $\Lambda_{IJ}$ and
1394: $\Lambda_{JI}$, are both positive.
1395: The unrealistic emergence of direct mutualism in Model A is shared by
1396: the Tangled-nature model \cite{CHRI02,COLL03,HALL02},
1397: of which it is a simplified version.
1398:
1399: Beyond the mean-field studies and the simulation results for average
1400: population sizes,
1401: we have also studied the temporal fluctuations of both the
1402: diversity and total population size for Models A and B. We find that
1403: the probability distributions of the lifetimes of individual species in
1404: both models are very similar,
1405: showing power-law decay with an exponent near
1406: $-2$ over near seven decades in time, as seen in Fig.~\ref{fig:life}.
1407: This exponent value is consistent
1408: with some interpretations of the available
1409: data for the lifetimes of marine genera
1410: in the fossil record \cite{NEWM03,NEWM99B},
1411: but other interpretations of the fossil evidence are also possible.
1412: Similarly, power spectra for the diversity, as well as for the total
1413: population size,
1414: show reasonable (although not perfect) $1/f$ behavior over many decades
1415: in frequency, as seen in Fig.~\ref{fig:PSD}.
1416: It is therefore very interesting that the probability distributions of the
1417: durations of individual QSS periods in the two models also both show
1418: reasonable power-law decay, but with {\it different\/} exponents:
1419: near $-2$ for Model A and close to $-1$ for Model B, as seen in
1420: Fig.~\ref{fig:quiet}(b).
1421: This result, which we found quite surprising at first, makes sense in
1422: light of the observation that the extinctions of major species are
1423: highly synchronized in Model A, while they are much less so in Model B.
1424: While communities in Model A tend to collapse completely when
1425: an aggressive mutant arrives and/or a major species goes extinct,
1426: communities in Model B are much more resilient and extinctions most often
1427: only extend to one or a few branches
1428: of the resident food web. This effect is illustrated in
1429: Figs.~\ref{fig:specs}, \ref{fig:specs2}, and \ref{fig:Webs}.
1430: The long-time correlations that give rise to these extended power
1431: laws are clearly due to the interspecies interactions. Indeed, our
1432: exploratory simulations of neutral versions of both models (not shown here)
1433: exhibit no correlations beyond about 10$^4$ generations.
1434:
1435: Our observation of the high resilience of Model B against complete
1436: extinction of communities is consistent with observations of
1437: extinction avalanches of limited size in the Web-world model by
1438: Drossel et al.\
1439: \cite{DROS01B}, who argue that their model is
1440: therefore {\it not\/} self-organized
1441: critical. Together with our observation of the self-optimization of
1442: the evolution models studied here to points {\it away from\/}
1443: the transcritical bifurcation point, these observations may support a
1444: conclusion that models of coevolution that take reasonable account of the
1445: dynamics at the ecological level (even if they are extremely simplified)
1446: are not in general self-organized critical. Such a conclusion, which
1447: appears reasonable, would be in
1448: disagreement with a number of recent theories of extinction
1449: \cite{BAK93,NEWM03,PACZ96}.
1450:
1451: The discussion in the preceding paragraphs brings out some of the
1452: differences and similarities between the models studied here, and some
1453: of the other physics-inspired evolution models that have been introduced
1454: in recent years. Like the Tangled-nature model \cite{CHRI02,COLL03,HALL02},
1455: our models are individual-based, and Model A also shares with that model
1456: the unrealistic emergence of mutualistic communities. The power-law
1457: behaviors are also reminiscent of species-based (rather than
1458: individual-based) extremal-dynamics evolution
1459: models like the Bak-Sneppen model and its generalizations
1460: \cite{BAK93,NEWM03,PACZ96}, and in \cite{RIKV05A} we speculated
1461: that Model B might correspond to a zero-dimensional extremal-dynamics
1462: model \cite{DORO00}. However, more accurate exponent estimates for Model
1463: B \cite{RIKV06B}, as well as the above comparison of Model B with the
1464: Web-world model \cite{DROS01B}, make this conjecture less likely to hold.
1465:
1466: In conclusion, the results presented here indicate that further work on
1467: models of macroevolution that are based on events on ecological time scales,
1468: with comparisons of the results with
1469: data from the fossil record, as well as from laboratory experiments and
1470: extant food webs, is
1471: highly desirable. In a separate paper we consider in detail the
1472: structure and dynamics of the food webs that develop within Model B,
1473: and we compare these with data for real food webs \cite{RIKV06B}.
1474:
1475:
1476: \begin{acknowledgements}
1477: The author thanks R.~K.~P.\ Zia, B.~Schmittmann, P.~Beerli,
1478: G.~J.~P.\ Naylor, and V.~Sevim
1479: for useful discussions and comments on the manuscript,
1480: and V.~Sevim for the data on Model A included in Fig.~\ref{fig:life}.
1481: He also gratefully acknowledges hospitality at Kyoto University
1482: by H.~Tomita in the
1483: Department of Fundamental Sciences, Faculty of Integrated Human Studies,
1484: and H.~Fujisaka in the
1485: Department of Applied Analysis and Complex Dynamical Systems,
1486: Graduate School of Informatics.
1487:
1488: This work was supported in part by U.S.\ National Science Foundation Grants
1489: No.\ DMR-0240078 and DMR-0444051, and by Florida State University
1490: through the School of Computational Science, the Center for Materials
1491: Research and Technology, the National High Magnetic Field Laboratory,
1492: and a COFRS summer-salary grant.
1493: \end{acknowledgements}
1494:
1495: \appendix
1496:
1497:
1498: \section{Derivation of the Community Fitness Function}
1499: \label{sec:AppA}
1500:
1501: In this Appendix we provide a conventional derivation of the cubic
1502: form of the community
1503: fitness
1504: %potential
1505: function $\Phi(N_{\rm tot})$ in a simple
1506: mean-field approximation. The derivation provides an explanation for the
1507: prefactor $(1 - 1/F)$ in Eq.~(\ref{eq:fit}), as well as intuitively
1508: clear approximations for the coefficients $\Theta$ and $\mathcal{E}$. It
1509: also provides justification that the equation to be integrated
1510: to obtain $\Phi(N_{\rm tot})$ is indeed Eq.~(\ref{eq:quad}), rather than
1511: this equation multiplied or divided by some power of $N_{\rm tot}$.
1512: The derivation is based on the time-dependent Ginzburg-Landau equation
1513: for a system with nonconserved order parameter
1514: \cite{GOLD92,HOHE77}, which for our current systems takes the form
1515: \begin{equation}
1516: \frac{\partial N_{\rm tot}}{\partial t}
1517: =
1518: \frac{\partial \Phi}{\partial N_{\rm tot}}
1519: \;.
1520: \label{eq:tdgl}
1521: \end{equation}
1522: %(The fitness $\Phi$ is the negative of the Landau free energy
1523: %customarily considered in physics applications.)
1524:
1525: Identifying $\partial n_I(t) / \partial t$ with
1526: $n_I(t+1) - n_I(t)$, we obtain from Eqs.~(\ref{eq:PI}--\ref{eq:MF})
1527: in the absence of mutations:
1528: \begin{equation}
1529: \frac{\partial n_I(t)}{\partial t}
1530: =
1531: n_I(t) \left\{
1532: \frac{F}
1533: {1 + \exp \left[ b_I - \frac{\eta_I R}{N_{\rm tot}(t)}
1534: - \sum_J M_{IJ} \frac{n_J(t)}{N_{\rm tot}(t)} + \frac{N_{\rm tot}(t)}{N_0}
1535: \right]}
1536: -1 \right\}
1537: .
1538: \label{eq:dndt}
1539: \end{equation}
1540: Expanding this nonlinear equation of motion around its fixed point, we get
1541: \begin{equation}
1542: \frac{\partial n_I(t)}{\partial t}
1543: \approx
1544: \left( 1 - \frac{1}{F} \right)
1545: n_I(t)
1546: \left[ - \tilde{b}_I + \frac{\eta_I R}{N_{\rm tot}(t)}
1547: + \sum_J M_{IJ} \frac{n_J(t)}{N_{\rm tot}(t)} - \frac{N_{\rm tot}(t)}{N_0}
1548: \right]
1549: .
1550: \label{eq:dndt2}
1551: \end{equation}
1552: To obtain the simplest mean-field approximation for ${\partial N_{\rm
1553: tot}}/{\partial t}$ (exact for $\mathcal{N}=1$), we set
1554: $n_I \approx N_{\rm tot}/\mathcal{N}$,
1555: $\tilde{b}_I \approx \mathcal{N}^{-1}\sum_I' \tilde{b}_I
1556: \equiv \langle \tilde{b} \rangle$,
1557: $\eta_I \approx \mathcal{N}^{-1}\sum_I' \eta_I \equiv \langle \eta \rangle$,
1558: and
1559: $M_{IJ} \approx \mathcal{N}^{-2} \sum_{IJ}' M_{IJ} \equiv \langle M
1560: \rangle $, where the primes on the sums indicate that they are
1561: restricted to the $\mathcal{N}$ species with $n_I > 0$.
1562: This yields
1563: \begin{equation}
1564: \frac{\partial N_{\rm tot}}{\partial t}
1565: \approx
1566: \left( 1 - \frac{1}{F} \right)
1567: \left[
1568: R \langle \eta \rangle
1569: + \left( \langle M \rangle - \langle \tilde{b} \rangle \right) N_{\rm tot}
1570: - \frac{(N_{\rm tot})^2}{N_0}
1571: \right]
1572: .
1573: \label{eq:dNdt}
1574: \end{equation}
1575: Integrating the right-hand side as prescribed by Eq.~(\ref{eq:tdgl}),
1576: we find
1577: \begin{equation}
1578: \Phi(N_{\rm tot})
1579: \approx
1580: \left( 1 - \frac{1}{F} \right)
1581: \left(
1582: R \langle \eta \rangle N_{\rm tot}
1583: + \frac{\langle M \rangle - \langle \tilde{b} \rangle}{2} N_{\rm tot}^2
1584: - \frac{1}{3 N_0}N_{\rm tot}^3
1585: \right)
1586: .
1587: \label{eq:PhiMFA}
1588: \end{equation}
1589: This result has the same cubic form as Eq.~(\ref{eq:fit}), with the
1590: approximate coefficients
1591: $\langle \eta \rangle \approx \mathcal{E}$ and
1592: $(\langle M \rangle - \langle \tilde{b} \rangle ) \approx \Theta$.
1593: The exact fixed-point solution for $N_{\rm tot}$ requires the use of
1594: $\mathcal{E}$ and $\Theta$, but the approximate forms
1595: obtained here may provide
1596: a more intuitive understanding of the significance of these coefficients.
1597:
1598:
1599: \section{Derivation of Optimum Point for Model A}
1600: \label{sec:AppX}
1601:
1602: We can obtain a theoretical estimate for the point
1603: ($\overline{\Theta},\overline{N_{\rm tot}})$ in Model A.
1604: We assume for simplicity that all the off-diagonal $M_{IJ}$
1605: have the same value, $a$. Then, using the definition of $\Theta$,
1606: and remembering that $F=4$ for Model A, so that
1607: $\tilde{b}_I = -\ln 3$ for all $I$, one can show
1608: that $\Theta = (1 - 1/\mathcal{N})a + \ln 3$, where
1609: $\mathcal{N}$ is the total number of species in the community.
1610: This yields the absolute maximum value for $\Theta$
1611: equal to $(1+\ln 3) \approx 2.10$
1612: for $\mathcal{N} = \infty$ (shown by vertical
1613: full lines in Fig.~\ref{fig:ntot}). However, it has been
1614: shown \cite{RIKV03} that the most probable number of species in a
1615: community within this model is finite and given by
1616: \begin{equation}
1617: \mathcal{N}^\dag \approx L \ln2 /\ln (1/q) \;,
1618: \label{eq:Ndag}
1619: \end{equation}
1620: where $q$ is the probability of finding a pair of interactions,
1621: $M_{IJ}$ and $M_{JI}$, conducive to this community.
1622: With the $M_{IJ}$ independently distributed on $[-1,+1]$,
1623: the probability of drawing a
1624: pair that are both larger than some given value $m \in [-1,+1]$,
1625: is $q = [(1-m)/2]^2$. In our approximate formula for
1626: $\Theta$, we now replace $\mathcal{N}$
1627: by $\mathcal{N}^\dag$ with this value of $q$, and $a$ by
1628: the average of a variable uniformly distributed over $[m,1]$,
1629: which is $(1+m)/2$. This yields
1630: \begin{equation}
1631: \Theta
1632: \approx
1633: \left[
1634: 1 - \frac{2 \ln \left( \frac{2}{1-m} \right)}{13 \ln 2}
1635: \right]
1636: \frac{1+m}{2} + \ln 3
1637: \label{eq:1x1}
1638: \end{equation}
1639: for $L=13$.
1640: This is a concave function of $m$ with a single maximum at $m_{\rm
1641: max}$, which can be found numerically. The result is
1642: $m_{\rm max} \approx 0.512$, which yields
1643: $\Theta_{\rm max} \approx 1.618$,
1644: $N_{\rm tot} \approx 3236$, and $\overline{M_{IJ}} = (1+m_{\rm max})/2
1645: \approx 0.756$. These values are in excellent agreement with those
1646: obtained from the Monte Carlo simulations, and they are
1647: shown as gray dots (turquoise online) in Fig.~\ref{fig:ntot}.
1648:
1649:
1650: \section{Trials with Other Parameter Values}
1651: \label{sec:AppB}
1652:
1653: In addition to the parameter set used in the main simulations
1654: presented above, we also performed trial simulations for Model B
1655: with $L=20$ (1\,048\,576 potential species), $\mu = 2.5 \times
1656: 10^{-4}$, and $R = 8000$. As each simulation run with $R=2000$ took
1657: about two weeks of CPU time, and each run with $R=8000$ took at
1658: least four weeks, relatively few trial runs were performed.
1659: No qualitative differences were observed
1660: in the quantities reported on the basis of the main simulation series.
1661:
1662: To obtain estimates of the effects of $L$, $\mu$, and $R$ on the
1663: the mean stationary levels of the total
1664: population size, $\overline{N_{\rm tot}}$, and diversity, $\overline{D}$,
1665: and the mean time to reach stationarity, $\overline{\tau}$,
1666: we fitted these variables by the
1667: exponential function $a[1 - \exp(-t/\tau)]$.
1668: Numerical results for $\overline{\tau}$,
1669: $\overline{D}$, and $\overline{N_{\rm tot}}$, based on
1670: twelve runs for
1671: $L=13$, $R=2000$, and $\mu = 10^{-3}$
1672: (the original runs used in the main part of this paper);
1673: twelve runs for
1674: $L=20$, $R=2000$, and $\mu = 10^{-3}$;
1675: nine runs for
1676: $L=20$, $R=2000$, and $\mu = 2.5 \times 10^{-4}$; and five runs for
1677: $L=20$, $R=8000$, and $\mu = 2.5 \times 10^{-4}$ are compiled in
1678: Table~\ref{table:I}.
1679:
1680: \subsection{Effects of the genome size}
1681:
1682: With $L=13$, the whole pool of potential species is visited in a
1683: typical $2^{25}$-generation simulation, while only about 10\%
1684: are typically visited for $L=20$ with $\mu=10^{-3}$ and $R=2000$. However, the
1685: ``revivals" of extinct species, seen frequently with $L=13$
1686: \cite{RIKV03} and much less
1687: so with $L=20$, has no effect on the observed power laws.
1688: Twelve full runs were performed for this parameter set.
1689: All the subsequent trial simulations were performed with $L=20$.
1690:
1691: As expected from Eq.~(\ref{eq:Ndag}), $\overline{D}$ appears to be
1692: proportional to $L$. The significant increase in $\overline{N_{\rm tot}}$
1693: when $L$
1694: is increased from 13 to 20 (not seen in trial runs with $L=20$ for Model A)
1695: is probably due to the
1696: increased probability of finding species with smaller $b_I$ and thus
1697: closer to the population divergence for Model B at $\Theta = 0$
1698: (see Fig.~\ref{fig:ntot}(a)).
1699:
1700: \begin{table}[t]
1701: \caption[]{Numerical results for the main simulations of Model B with
1702: $L=13$, compared with trial runs with $L = 20$ and varying $R$ and $\mu$.
1703: The unit for $R$ and $\overline{N_{\rm tot}}$ is individuals, for
1704: $\mu$ it is mutations per individual offspring per generation, for
1705: $\overline{\tau}$ it is $10^6$ generations, and for $\overline{D}$ it is
1706: species. Thus, $R \mu$ is proportional to the total number of
1707: mutated offspring per generation. The uncertainties are standard
1708: errors, based on the spread between individual runs.
1709: $\sigma_\tau$ is the standard deviation of $\tau$ over the
1710: independent runs. A ratio $\overline{\tau}/\sigma_\tau \approx 1$
1711: indicates that $\tau$ may be approximately exponentially distributed,
1712: while the even lower ratio for $R\mu = 0.5$ indicates
1713: very large variations from run to run.
1714: }
1715: \begin{center}
1716: \begin{tabular}{|c|c|c|c|c|c|c|c|c|}
1717: \hline
1718: $L$ & $R$ & $\mu$ & $R \mu$ & Runs & $\overline{\tau}$ &
1719: $\overline{\tau}/\sigma_\tau$ & $\overline{D}$
1720: & $\overline{N_{\rm tot}}$
1721: \\
1722: \hline\hline
1723: 13 & 2000 & $10^{-3}$ & 2.0 & 12 & 0.35$\pm$0.07 & 1.08 & 10.9$\pm$0.3
1724: & 7\,700$\pm$300
1725: \\
1726: \hline
1727: 20 & 2000 & $10^{-3}$ & 2.0 & 12 & 0.8$\pm$0.2 & 0.93 & 14.7$\pm$0.3
1728: & 13\,700$\pm$400
1729: \\
1730: \hline
1731: 20 & 8000 & $2.5 \times 10^{-4}$ & 2.0 & 5 & 1.2$\pm$0.3 & 0.90 & 20.4$\pm$0.6
1732: & 58\,000$\pm$7\,000
1733: \\
1734: \hline
1735: 20 & 2000 & $2.5 \times 10^{-4}$ & 0.5 & 9 & 5$\pm$2 & 0.53 & 10.3$\pm$0.2
1736: & 12\,000$\pm$2\,000
1737: \\
1738: \hline
1739: \end{tabular}
1740: \end{center}
1741: \label{table:I}
1742: \end{table}
1743:
1744: \subsection{Effects of reduced $\mu$ and increased $R$}
1745:
1746: We found $\overline{\tau}$ to be approximately inversely proportional
1747: to $\mu$ and $R$, while the mean diversity increases (but apparently
1748: sublinearly) with both $\mu$ and $R$. Within the uncertainty,
1749: $\overline{\tau}$ was the same for the diversity and
1750: population size, and the reported results are therefore based on both
1751: quantities. The mean total population size $\overline{N_{\rm tot}}$
1752: appears to be roughly independent of $\mu$ and approximately linearly
1753: dependent on $R$. This latter proportionality
1754: (which is also expected from the analytical result, Eq.~(\ref{eq:sol}))
1755: means that
1756: $R \mu \propto \overline{N_{\rm tot}} \mu$,
1757: the average total number of mutant individuals per
1758: generation. Even though the models studied here have strong interspecies
1759: interactions, we note that $\overline{N_{\rm tot}} \mu$
1760: is one half of the {\it fundamental biodiversity
1761: number\/} in Hubbell's neutral theory of biodiversity \cite{HUBB01}.
1762: This parameter appears to play an analogous role in determining the diversity
1763: in the present models as it does in the neutral theory.
1764:
1765: However, more important than the above results is that none of the observed
1766: power laws changed under parameter variation within this range. In
1767: particular, the exponent for the PDF of individual species
1768: lifetimes remained near $-2$ for all parameter sets, while the
1769: exponent for the QSS duration PDF remained close to $-1$. In fact,
1770: allowing for the larger overall intensity in the population PSD for
1771: $R=8000$ and to within
1772: the numerical accuracy, both PDFs and PSDs for all three parameter
1773: sets with $L=20$
1774: can be overlaid graphically with the ones presented
1775: for $L=13$, $\mu = 10^{-3}$, and $R=2000$ elsewhere in this paper.
1776: Based on these trials we therefore conclude that the parameters used in the
1777: main study are also representative for systems with larger $L$ and $R$ and
1778: smaller $\mu$ and are well chosen to study the stationary dynamics of
1779: the models on long time scales.
1780:
1781: Analogous, but less exhaustive, tests for Model A give similar results.
1782: Runs with $N_0 = 16\,000$ and $\mu = 1.25 \times 10^{-4}$ ($N_0 \mu = 2$)
1783: gave $\overline{N_{\rm tot}} \approx 1.5\,N_0$ as in the main simulations
1784: with $N_0 =2000$, and $\overline{D} \approx 3.6$, a statistically
1785: insignificant increase from the value of approximately 3.3 for the
1786: simulations with $N_0 =2000$ and $\mu = 10^{-3}$ \cite{RIKV03}.
1787:
1788:
1789: % BibTeX users please use
1790: %\bibliographystyle{spmpsci}
1791: %\bibliography{evol} % name your BibTeX data base
1792:
1793: % Non-BibTeX users please use
1794: %\begin{thebibliography}{10}
1795: %\end{thebibliography}
1796:
1797: %\begin{authorbiography}{example.eps}{A.N. Author}\
1798: %biography will be printed here if needed. Whether an author
1799: %biography is included per author is set
1800: %per journal. A photo is optional.
1801: %\end{authorbiography}
1802:
1803: \begin{thebibliography}{10}
1804: \providecommand{\url}[1]{{#1}}
1805: \providecommand{\urlprefix}{URL }
1806: \expandafter\ifx\csname urlstyle\endcsname\relax
1807: \providecommand{\doi}[1]{DOI~\discretionary{}{}{}#1}\else
1808: \providecommand{\doi}{DOI~\discretionary{}{}{}\begingroup
1809: \urlstyle{rm}\Url}\fi
1810:
1811: \bibitem{ALON04}
1812: Alonso, D., McKane, A.J.: Sampling Hubbell's neutral theory of biodiversity.
1813: \newblock Ecol.\ Lett. \textbf{7}, 901--910 (2004)
1814:
1815: \bibitem{ARMS80}
1816: Armstrong, R.A., McGehee, R.: Competitive exclusion.
1817: \newblock Am.\ Naturalist \textbf{115}, 151--170 (1980)
1818:
1819: \bibitem{BAK93}
1820: Bak, P., Sneppen, K.: Punctuated equilibrium and criticality in a simple model
1821: of evolution.
1822: \newblock Phys.\ Rev.\ Lett. \textbf{71}, 4083--4086 (1993)
1823:
1824: \bibitem{BASC06}
1825: Bascompte, J., Jordano, P., Olesen, J.M.: Asymmetric coevolutionary networks
1826: facilitate biodiversity maintenance.
1827: \newblock Science \textbf{312}, 431--433 (2006)
1828:
1829: \bibitem{DENB86}
1830: den Boer, P.J.: The present status of the competitive exclusion principle.
1831: \newblock Trends Ecol.\ Evol. \textbf{1}, 25--28 (1986)
1832:
1833: \bibitem{BRON94}
1834: Bronstein, J.L.: Our current understanding of mutualism.
1835: \newblock Quart.\ Rev.\ Biol. \textbf{69}, 31--51 (1994)
1836:
1837: \bibitem{CALD98}
1838: Caldarelli, G., Higgs, P.G., McKane, A.J.: Modelling coevolution in
1839: multispecies communities.
1840: \newblock J.\ theor.\ Biol. \textbf{193}, 345--358 (1998)
1841:
1842: \bibitem{CHOW05}
1843: Chowdhury, D., Stauffer, D.: Evolutionary ecology in-silico: Does mathematical
1844: modelling help in understanding the generic trends?
1845: \newblock J.\ Biosciences \textbf{30}, 277--287 (2005), and references
1846: therein
1847:
1848: \bibitem{CHOW03A}
1849: Chowdhury, D., Stauffer, D., Kunwar, A.: Unification of small and large time
1850: scales for biological evolution: Deviations from power law.
1851: \newblock Phys.\ Rev.\ Lett. \textbf{90}, 068101 (2003)
1852:
1853: \bibitem{CHRI02}
1854: Christensen, K., di~Collobiano, S.A., Hall, M., Jensen, H.J.: Tangled-nature: A
1855: model of evolutionary ecology.
1856: \newblock J.\ theor.\ Biol. \textbf{216}, 73--84 (2002)
1857:
1858: \bibitem{COLL03}
1859: di~Collobiano, S.A., Christensen, K., Jensen, H.J.: The tangled nature model as
1860: an evolving quasi-species model.
1861: \newblock J.\ Phys.\ A \textbf{36}, 883--891 (2003)
1862:
1863: \bibitem{CRAW91}
1864: Crawford, J.D.: Introduction to bifurcation theory.
1865: \newblock Rev.\ Mod.\ Phys. \textbf{63}, 991--1037 (1991)
1866:
1867: \bibitem{CROS70}
1868: Crosby, J.L.: The evolution of genetic discontinuity: Computer models of the
1869: selection of barriers to interbreeding between subspecies.
1870: \newblock Heredity \textbf{25}, 253--297 (1970)
1871:
1872: \bibitem{DOEB00}
1873: Doebeli, M., Dieckmann, U.: Evolutionary branching and sympatric speciation
1874: caused by different types of ecological interactions.
1875: \newblock Am.\ Naturalist \textbf{156}, S77--S101 (2000).
1876: \newblock And references therein
1877:
1878: \bibitem{DORO00}
1879: Dorogovtsev, S.N., Mendes, J.F.F., Pogorelov, Yu.G.:
1880: Bak-Sneppen model near zero dimension.
1881: \newblock Phys.\ Rev.\ E \textbf{62}, 295--298 (2000)
1882:
1883: \bibitem{DROS01B}
1884: Drossel, B., Higgs, P.G., McKane, A.J.: The influence of predator-prey
1885: population dynamics on the long-term evolution of food web structure.
1886: \newblock J.\ theor.\ Biol. \textbf{208}, 91--107 (2001)
1887:
1888: \bibitem{DROS04}
1889: Drossel, B., McKane, A., Quince, C.: The impact of non-linear functional
1890: responses on the long-term evolution of food web structure.
1891: \newblock J.\ theor.\ Biol. \textbf{229}, 539--548 (2004)
1892:
1893: \bibitem{DUNN02}
1894: Dunne, J., Williams, R.J., Martinez, N.D.: Network structure and diversity loss
1895: in food webs: Robustness increases with connectance.
1896: \newblock Ecol.\ Lett. \textbf{5}, 558--567 (2002)
1897:
1898: \bibitem{EIGE71}
1899: Eigen, M.: Selforganization of matter and evolution of biological
1900: macromolecules.
1901: \newblock Naturwissenschaften \textbf{58}, 465 (1971)
1902:
1903: \bibitem{EIGE88}
1904: Eigen, M., McCaskill, J., Schuster, P.: Molecular quasi-species.
1905: \newblock J.\ Phys.\ Chem. \textbf{92}, 6881--6891 (1988)
1906:
1907: \bibitem{GARL04}
1908: Garlaschelli, D.: Universality in food webs.
1909: \newblock Eur.\ Phys.\ J.\ B \textbf{38}, 277--285 (2004), and
1910: references therein
1911:
1912: \bibitem{GAVR99}
1913: Gavrilets, S.: Dynamics of clade diversification on the morphological
1914: hypercube.
1915: \newblock Proc.\ R.\ Soc.\ Lond.\ B \textbf{266}, 817--824 (1999)
1916:
1917: \bibitem{GAVR04}
1918: Gavrilets, S.: Fitness landscapes and the origin of species.
1919: \newblock Princeton University Press, Princeton and Oxford (2004)
1920:
1921: \bibitem{GAVR98}
1922: Gavrilets, S., Boake, C.R.B.: On the evolution of premating isolation after a
1923: founder event.
1924: \newblock Am.\ Naturalist \textbf{152}, 706--716 (1998)
1925:
1926: \bibitem{GAVR00}
1927: Gavrilets, S., Li, H., Vose, M.D.: Patterns of parapatric speciation.
1928: \newblock Evolution \textbf{54}, 1126--1134 (2000)
1929:
1930: \bibitem{GAVR05}
1931: Gavrilets, S., Vose, A.: Dynamic patterns of adaptive radiation.
1932: \newblock Proc.\ Natl.\ Acad.\ Sci.\ USA \textbf{102}, 18,040--18,045 (2005)
1933:
1934: \bibitem{GOLD92}
1935: Goldenfeld, N.: Lectures on Phase Transitions and the Renormalization Group.
1936: \newblock Addison-Wesley, Reading, MA (1992)
1937:
1938: \bibitem{HAKE77}
1939: Haken, H.: Synergetics -- An Introduction.
1940: \newblock Springer-Verlag, Berlin (1977)
1941:
1942: \bibitem{HALL02}
1943: Hall, M., Christensen, K., di~Collobiano, S.A., Jensen, H.J.: Time-dependent
1944: extinction rate and species abundance in a tangled-nature model of biological
1945: evolution.
1946: \newblock Phys.\ Rev.\ E \textbf{66}, 011904 (2002)
1947:
1948: \bibitem{HARD60}
1949: Hardin, G.: The competitive exclusion principle.
1950: \newblock Science \textbf{131}, 1292--1297 (1960)
1951:
1952: \bibitem{HOHE77}
1953: Hohenberg, P.C., Halperin, B.: Theory of dynamic critical phenomena.
1954: \newblock Rev.\ Mod.\ Phys. \textbf{49}, 435--479 (1977)
1955:
1956: \bibitem{HUBB01}
1957: Hubbell, S.P.: The Unified Neutral Theory of Biodiversity and Biogeography.
1958: \newblock Princeton University Press, Princeton (2001), Chap.\ 5
1959:
1960: \bibitem{KAUF93}
1961: Kauffman, S.A.: The origins of order. Self-organization and selection in
1962: evolution.
1963: \newblock Oxford University Press, Oxford (1993)
1964:
1965: \bibitem{KAUF91}
1966: Kauffman, S.A., Johnsen, S.: Coevolution to the edge of chaos: Coupled fitness
1967: landscapes, poised states, and coevolutionary avalanches.
1968: \newblock J.\ theor.\ Biol. \textbf{149}, 467--505 (1991)
1969:
1970: \bibitem{KAWA93}
1971: Kawanabe, H., Cohen, J.E., Iwasaki, K.: Mutualism and Community Organization.
1972: \newblock Oxford University Press, Oxford (1993)
1973:
1974: \bibitem{KREB89}
1975: Krebs, C.J.: Ecological Methodology.
1976: \newblock Harper \& Row, New York (1989).
1977: \newblock Chap.~10
1978:
1979: \bibitem{KREB01}
1980: Krebs, C.J.: Ecology. The Experimental Analysis of Distribution and Abundance.
1981: Fifth Edition.
1982: \newblock Benjamin Cummings, San Francisco (2001), Chaps.\ 13 and 14
1983:
1984: \bibitem{METZ92}
1985: Metz, J.A.J., Nisbet, R.M., Geritz, S.A.H.: How should we define `fitness' for
1986: general ecological scenarios?
1987: \newblock Trends Ecol.\ Evol. \textbf{7}, 198--202 (1992)
1988:
1989: \bibitem{MURR89}
1990: Murray, J.D.: Mathematical Biology.
1991: \newblock Springer-Verlag, Berlin (1989)
1992:
1993: \bibitem{NEWM05}
1994: Newman, M.E.J.: Power laws, Pareto distributions and Zipf's law.
1995: \newblock Contemporary Physics \textbf{46}, 323--351 (2005)
1996:
1997: \bibitem{NEWM03}
1998: Newman, M.E.J., Palmer, R.G.: Modeling Extinction.
1999: \newblock Oxford University Press, Oxford (2003)
2000:
2001: \bibitem{NEWM99B}
2002: Newman, M.E.J., Sibani, P.: Extinction, diversity and survivorship of taxa in
2003: the fossil record.
2004: \newblock Proc.\ R.\ Soc.\ Lond.\ B \textbf{266}, 1583--1599 (1999)
2005:
2006: \bibitem{PACZ96}
2007: Paczuski, M., Maslov, S., Bak, P.: Avalanche dynamics in evolution, growth, and
2008: depinning models.
2009: \newblock Phys.\ Rev.\ E \textbf{53}, 414--443 (1996)
2010:
2011: \bibitem{PATH96}
2012: Pathria, R.K.: Statistical Mechanics, Second Edition.
2013: \newblock Butterworth-Heinemann, Oxford (1996), Chaps.\ 11 and 14
2014:
2015: \bibitem{PIGO05}
2016: Pigolotti, S., Flammini, A., Marsili, M., Maritan, A.:
2017: Species lifetime distribution for simple models of ecologies.
2018: \newblock Proc.\ Natl.\ Acad.\ Sci.\ USA \textbf{102}, 15747--15751 (2005)
2019:
2020: \bibitem{PRES92}
2021: Press, W.H., Teukolsky, S.A., Vetterling, W.T., Flannery, B.P.: Numerical
2022: Recipes, Second Ed.
2023: \newblock Cambridge University Press, Cambridge (1992)
2024:
2025: \bibitem{RIKV06B}
2026: Rikvold, P.A., Sevim, V.:
2027: An individual-based predator-prey model for biological coevolution:
2028: Fluctuations, stability, and community structure.
2029: \newblock Phys.\ Rev.\ E, in press.
2030: \newblock E-print arXiv:q-bio.PE/0611023
2031:
2032: \bibitem{RIKV06C}
2033: Rikvold, P.A.: Complex behavior in simple models of biological coevolution.
2034: \newblock Submitted to Int.\ J.\ Mod.\ Phys.\ C.
2035: \newblock E-print arXiv:q-bio.PE/0609013
2036:
2037: \bibitem{RIKV05A}
2038: Rikvold, P.A.: Fluctuations in models of biological macroevolution.
2039: \newblock In: L.B. Kish, K.~Lindenberg, Z.~Gingl (eds.) Noise in Complex
2040: Systems and Stochastic Dynamics III, pp. 148--155. SPIE, The International
2041: Society for Optical Engineering, Bellingham, WA (2005).
2042: \newblock E-print arXiv:q-bio.PE/0502046
2043:
2044: \bibitem{RIKV03}
2045: Rikvold, P.A., Zia, R.K.P.: Punctuated equilibria and $1/f$ noise in a
2046: biological coevolution model with individual-based dynamics.
2047: \newblock Phys.\ Rev.\ E \textbf{68}, 031913 (2003)
2048:
2049: \bibitem{ROBE74}
2050: Roberts, A.: The stability of a feasible random ecosystem.
2051: \newblock Nature (London) \textbf{251}, 607--608 (1974)
2052:
2053: \bibitem{SATO03}
2054: Sato, K., Ito, Y., Yomo, T., Kaneko, K.: On the relation between fluctuation
2055: and response in biological systems.
2056: \newblock Proc.\ Natl.\ Acad.\ Sci.\ USA \textbf{100}, 14,086--14,090 (2003)
2057:
2058: \bibitem{SEVI06}
2059: Sevim, V., Rikvold, P.A.: Effects of correlated interactions in a biological
2060: coevolution model with individual-based dynamics.
2061: \newblock J.\ Phys.\ A \textbf{38}, 9475--9489 (2005)
2062:
2063: \bibitem{SHAN48}
2064: Shannon, C.E.: A mathematical theory of communication.
2065: \newblock Bell Syst.\ Tech.\ J. \textbf{27}, 379--423; 628--656 (1948)
2066:
2067: \bibitem{SHAN49}
2068: Shannon, C.E., Weaver, W.: The Mathematical Theory of Communication.
2069: \newblock University of Illinois Press, Urbana (1949)
2070:
2071: \bibitem{SOLE96B}
2072: Sol{\'e}, R.V., Bascompte, J., Manrubia, S.: Extinction: Bad genes or weak
2073: chaos?
2074: \newblock Proc.\ R.\ Soc.\ Lond.\ B \textbf{263}, 1407--1413 (1996)
2075:
2076: \bibitem{STRO94}
2077: Strogatz, S.H.: Nonlinear Dynamics and Chaos.
2078: \newblock Westview Press, Boston (1994)
2079:
2080: \bibitem{THOM98}
2081: Thompson, J.N.: Rapid evolution as an ecological process.
2082: \newblock Trends Ecol.\ Evol. \textbf{13}, 329--332 (1998)
2083:
2084: \bibitem{THOM99}
2085: Thompson, J.N.: The evolution of species interactions.
2086: \newblock Science \textbf{284}, 2116--2118 (1999)
2087:
2088: \bibitem{TOKI03}
2089: Tokita, K., Yasutomi, A.: Emergence of a complex and stable network in a model
2090: ecosystem with extinction and mutation.
2091: \newblock Theor.\ Popul.\ Biol. \textbf{63}, 131--146 (2003)
2092:
2093: \bibitem{VERH1838}
2094: Verhulst, P.F.: Notice sur la loi que la population suit dans son
2095: accroissement.
2096: \newblock Corres.\ Math.\ et Physique \textbf{10}, 113--121 (1838)
2097:
2098: \bibitem{VOLK05}
2099: Volkov, I., Banavar, J.R., He, F., Hubbell, S.P., Maritan, A.: Density
2100: dependence explains tree species abundance and diversity in tropical forests.
2101: \newblock Nature \textbf{438}, 658--661 (2005)
2102:
2103: \bibitem{WILL06}
2104: Wills, C., Harms, K.E., Condit, R., King, D., Thompson, J., He, F.,
2105: Muller-Landau, H.C., Ashton, P., Losos, E., Comita, L., Hubbell, S.,
2106: LaFrankie, J., Bunyavejchevin, S., Dattaraja, H.S., Davies, S., Esufali, S.,
2107: Foster, R., Gunatilleke, N., Gunatilleke, S., Hall, P., Itoh A., John, R.,
2108: Kiratiprayoon, S., de Lao, S.L., Massa, M., Nath, C., Nur Supradi Noor, M.,
2109: Kassim, A.R., Sukumar, R., Suresh, H.S., Sun, I.F., Tan, S., Yamakura, T.,
2110: Zimmerman, J.: Nonrandom processes maintain diversity in tropical forests.
2111: \newblock Science \textbf{311}, 527--531 (2006)
2112:
2113: \bibitem{YOSH03}
2114: Yoshida, T., Jones, L.E., Ellner, S.P., Fussmann, G.F., Hairston, N.G.: Rapid
2115: evolution drives ecological dynamics in a predator-prey system.
2116: \newblock Nature \textbf{424}, 303--306 (2003)
2117:
2118: \bibitem{ZIA04}
2119: Zia, R.K.P., Rikvold, P.A.: Fluctuations and correlations in an
2120: individual-based model of biological coevolution.
2121: \newblock J.\ Phys.\ A \textbf{37}, 5135--5155 (2004)
2122:
2123: \end{thebibliography}
2124:
2125: \end{document}
2126: