1: \documentstyle[aps,multicol,amstex,natbib]{revtex}

2: \input epsf

3: \def\etal{\textit{et al.}}

4: \begin{document}

5: \title{Competing Populations in Flows with Chaotic Mixing}

6: 

7: \author{Istv\'an Scheuring$^\ddag$,Gy\"orgy K\'arolyi$^\star$,

8: Zolt\'an Toroczkai$^\P$,

9: Tam\'as T\'el$^\S$,

10: and \'Aron P\'entek$^\dag$}

11: 

12: 

13: \address{

14: $^\ddag$Department of Plant Taxonomy and Ecology,

15: Research Group of Ecology and Theoretical Biology,

16: E\"otv\"os University,

17: Ludovika t\'er 2, H-1083 Budapest, Hungary\\

18: $^\star$Department of Structural Mechanics,

19: Budapest University of Technology and Economics, M\H{u}egyetem rkp.~3,

20: H-1521 Budapest, Hungary\\

21: $^\P$Theoretical  Division and Center for Nonlinear Studies,

22: Los Alamos National Laboratory,

23: Los Alamos,  New Mexico 87545, USA \\

24: $^\S$Institute for Theoretical Physics,

25:  E\"{o}tv\"{o}s

26: University, P. O. Box 32, H-1518 Budapest, Hungary\\

27: $^\dag$Marine Physical Laboratory, Scripps Institution

28: of Oceanography, University of California

29: at San Diego, La Jolla, CA 92093-0238, USA}

30: 

31: \date{\today}

32: \maketitle

33: \begin{abstract}

34: We investigate the effects of spatial heterogeneity on the

35: coexistence of competing species in the

36: case when the heterogeneity is dynamically generated

37: by environmental flows with chaotic mixing properties.

38: We show that one of the effects of chaotic advection on the

39: passively advected species (such as phytoplankton, or

40: self-replicating macromolecules) is the possibility of coexistence

41: of more species than that limited by the number of niches they occupy.

42: We derive a novel set of dynamical equations

43: for competing populations.

44: \end{abstract}

45: 

46: \vspace*{0.6cm}

47: 

48: \begin{multicols}{2}

49: 

50: \section*{Introduction}

51: 

52: One of the classical problems of ecology is the identification of the

53: mechanisms responsible for the coexistence of competing species.

54: It is an observational fact that in Nature numerous

55: species are able to coexist, all competing for a limited number

56: of resources. This observed coexistence is at odds with many

57: classical theories and empirical studies predicting

58: competitive exclusion of all but the

59: most perfectly adapted species for each limiting factor

60: \citep{Gau35,Har60}. However, one of the

61: key ingredients in these classical studies was the

62: assumption of a homogeneous, well mixed and non-structured

63: environment which leads to an equilibrium state in the system.

64: Thus, if coexistence is to persist over longer time periods,

65: it must have a nonequilibrium character.

66: %As presented in more detail below, the mechanisms responsible for the

67: %nonequilibrium nature of the coexistence are rooted in  the

68: %\emph{variability} of the niches and of the

69: %environmental conditions.

70: 

71: The coexistence problem is best illustrated in the case

72: of \emph{phytoplankton communities} as

73: was originally

74: presented by \citet{Hut61}.

75: Here a number of species coexist

76: in a relatively isotropic or unstructured environment, all

77: competing for the same sorts of materials, and the number of species exceeds

78: considerably the number of limiting factors. 

79: To solve this so called ``paradox of plankton'', Hutchinson 

80: put forward the 

81: idea that seasonal environmental 

82: changes prevent competitive 

83: exclusion in natural phytoplankton communities. 

84: Thus the species of the community, at least on 

85: the time scale of ecological observation, are 

86: in nonequilibrium coexistence. 

87: 

88: Since then numerous investigations revealed  

89: many different 

90: mechanisms, including spatial and 

91: temporal heterogeneity of habitat,

92: predation, disturbance, coevolution, etc.\ \citep{Wil90,Che00},

93: increasing

94:  the probability of competitive coexistence.

95: Naturally, under the word ``competition''  many different

96: biological phenomena are collected together, which influence

97: the coexistence of species in different ways.

98: 

99: Thus the original problem changed into finding the most relevant

100: mechanisms which maintain diversity

101: in particular situations \citep{Con78,Hus79,Wil90,Til93,Bar97}.

102: Despite the vivid debate in this field

103: of ecology, there is  by now a consensus

104: that climatic periodicities and fluctuations

105: play the main role in causing

106: species' persistence in phytoplankton

107: communities \citep{Gae86,Rey93,Som93}.

108: It is frequently argued that an intermediate disturbance

109: \citep{Con78} is the most adequate hypothesis for the

110: explanation of high diversity in aquatic

111: systems \citep[cf.][]{Rey98}.

112: 

113: One can meet a similar problem in \emph{early evolution of life}.

114: Since life evolves from the simple structured entities

115: to the most complex ones, there must have been

116: a stage in the evolution, when life was essentially no

117: more complex than what a collection  of self-replicating

118: nucleic acids present \citep{May95}.

119: They were competing for a few limiting resources

120: (such as mononucleids and

121: energy rich chemicals) and making copies of

122: themselves without any specific

123: enzyme. Without enzymes the copying accuracy could not be very high.

124: Estimating the selective superiority

125: of the best replicator and the copying accuracy per

126: nucleotide, it is concluded  that the maximum length of these

127: molecules is about 100 nucleotides \citep{Eig71}.

128: However, in a well mixed homogeneous environment, as the

129: prebiotic ocean  is supposed to have been,

130: there are only a few  winners of the

131: selection, namely the most fit macromolecule surrounded

132: by its closest mutants

133: \citep{Eig71,Eig79}.

134: But how can we surmount the gap  between

135: these primitive replicators with 100 nucleotides

136: and the most simple

137: RNA viruses with 4000--5000 nucleotides?

138: Specific replicase enzymes are needed to increase

139: the copying fidelity, and thus the length of the

140: replicator, but these replicators are too short to code

141: specific enzymes. This is the ``Catch~22'' of the

142: prebiotic evolution \citep{May83}: no genome without

143: an enzyme, however no enzyme without genomes.

144: This problem can be resolved if some mechanism maintains

145: the coexistence of several  different replicator molecules

146: and therefore the information necessary for coding a

147: replicase enzyme can be stored by the union of smaller

148: information carriers. In this situation the replication error

149: does not grow exponentially as in the case of a base-by-base

150: copying, it grows only linearly with the number of

151: smaller carriers.

152: 

153: \begin{figure}[htbp]

154: \vspace*{-0.5cm}

155: \hspace*{0cm}\epsfxsize = 3.2 in \epsfbox{phytoff.ps}

156: \vspace*{0.5cm}

157: \caption{ SEAWIFS image of a phytoplankton bloom at

158: Shetland Islands, May 12, 2000,

159: from the NASA archive. Plankton individuals (light grey)

160: move along a fractal set. }

161: \label{fig:satelite}

162: \end{figure}

163: 

164: Current theories point out coexistence of replicators moving on a

165: surface %in a constrained manner

166: \citep{Boe91,Cza00}, preferring thus the concept of

167: ``prebiotic pizza'' against

168: the concept of ``prebiotic soup'' \citep{Wac94}. However some

169: kind of cooperation among the replicator molecules are

170: built into these models,

171: consequently they are not completely competitive.

172: An alternative explanation assumes that both the

173: replicative and enzymatic

174: functions were co-evolved, thus the lenght of the

175: replicators and the accuracy

176: of enzymatic functions increased together

177:  \citep{Poo99,Sch00a}.

178: In both problems (i.e., in the paradox of plankton

179: and in the Catch~22

180: of prebiotic evolution) the  traditional population

181: dynamical equation

182: for two species $B_1$, $B_2$ competing for the

183: resource $A$ read:

184: \begin{eqnarray}

185: &&\frac{dN_1}{dt} = \alpha_1 N_1 - \delta_1 N_1, \\\label{trad0}

186: &&\frac{dN_2}{dt} = \alpha_2 N_2 - \delta_2 N_2.\label{trad}

187: \end{eqnarray}

188: Here $N_i$ is the instantaneous

189: number of individuals of species $B_i$ in a given range

190: of a well strirred region. The instantaneous

191: parameters $\alpha_i$, $\delta_i$

192: are positive and

193: %the replication rates $\alpha_i$

194: depend, in general,  on the concentration of the resource material

195: $A$, too.

196: 

197: \begin{figure}[htbp]

198: \begin{minipage}{3.3in}

199: \hspace*{-0.4cm}\epsfxsize = 3.3 in \epsfbox{T11.ps}

200: \end{minipage} \\

201: \begin{minipage}{3.3in}

202: \hspace*{-0.4cm} \epsfxsize = 3.3 in \epsfbox{T21.eps}

203: \end{minipage}

204: \vspace*{0.0cm}

205: \caption{The distribution of two populations (light gray

206: ($B$) and dark gray ($C$))

207: competing for the same resource material (white ($A$))

208: in the wake of

209: a cylinder. The flow is from left to right. The inset

210: in (a) shows the

211: time-dependence of the population numbers $n_B,n_C$ and

212: clearly indicates

213: the approach to a steady state of coexistence after about

214: $40$ time units which is

215: the period of the flow. A blowup of the region

216: indicated by a rectangle

217: in (a) is seen in (b).  Species distribution is

218: strikingly similar to many of the patterns

219: found on the NASA SEAWIFS satelite pictures of plankton blooms

220: (see Fig \ref{fig:satelite}).

221: After  \citet{Kar00}.}\label{fig:Compare}

222: \end{figure}

223: 

224: Independently of the particular form of this

225: dependence and the dynamical equation of $A$,

226: \emph{no} fixed points can exist in the system

227: in which both species would be in a steady state with

228: nonzero values of $N_i=N_i^*$ \citep{Gur98}.

229: The coexistence might, however,

230: be possible in \emph{imperfectly stirred environments}. 

231: As numerous remote sensing images demonstrate

232: phytoplankton are distributed along fractal

233: filaments in the oceans indicating strong but

234: imperfect mixing environment

235: (Fig \ref{fig:satelite}).

236: 

237: 

238: Recent development in the field of hydrodynamics

239: encouraged us

240: to revisit the population dynamics of competing species in \emph{open} 

241: aquatic

242: systems. In aquatic systems of large extension, on

243: the time scales characteristic to the life cycle

244: of microorganisms and replicators, the hydrodynamical flows are

245: locally open, i.e. there is a net current, transporting both

246: competitors and nutrients,

247: flowing through the typical observation region.

248: It is even more obvious that the flow

249: is open in the wake of islands surrounded by strong

250: ocean currents \citep{Ari97} and around the deep see hot springs

251: where the cradle of life probably swung \citep{Hol92}.

252: 

253: With the aid of numerical simulations we have

254: previously shown  that the

255: coexistence of passively

256: advected competing species is typical in open chaotic flows 

257: \citep{Kar00,Sch00b}. For simplicity, we have considered the

258: two-dimensional

259: flow around a cylindrical obstacle

260: placed into a uniform background flow.

261: For moderate inflow velocities

262: there is a periodic detachment of vortices in the wake of the obstacle

263: with period $T$, which forms the von K\'arm\'an

264: vortex street \citep{Sha91,Jun92,Som96}. 

265: The flow in the wake is time-dependent but still spatially

266: regular. Here individuals of two passively advected species compete for a

267: common limiting resource, see Fig.~\ref{fig:Compare}.

268: We argued that

269: coexistence is due to the fractal structures typically appearing in the

270: advection patterns of such flows, however, we have given  only a  heuristic

271: interpretation for the mechanisms maintaining coexistence in this

272: hydrodynamical system. In this article we present a  mathematical deduction to

273: explain coexistence of competitors in open chaotic flows.   The mathematical

274: problem is to investigate two coupled population dynamical processes

275: evolving on a \emph{fractal} support.

276: We shall present a new class of equations which describe this situation,

277: and allow for the coexistence of at least two species

278: competing for the same resource. A novel feature of these equations  

279: will be a singularly nonlinear 

280: (power-law) form of both the replication term and the 

281: coupling between the populations.

282: 

283: In the following section we summarize the

284: qualitative features of the

285: relevant physical process, followed by a  study of the

286: dynamics of a single population in an open chaotic flow. Consequently, the

287: coexistence of competitors is discussed by first giving

288: a qualitative argument based on the single population

289: picture, followed by a detailed mathematical model

290: leading to the aforementioned new type of population dynamical equations.

291: Next, this theory is compared with further numerical results carried out

292: on a simple map modelling the advection dynamics, on the so-called baker map.

293:  We conclude with a summary and outlook.

294: 

295: \section*{Passive advection in open flows}

296: 

297: Chaotic advection in open hydrodynamical flows is an 

298: ubiquitous phenomenon.  A flow is considered locally open 

299: if there is a  net current flowing through 

300: the observation region \citep{Lam32}. 

301: It became clear in the last decade 

302: that passive advection even in simple 

303: time-dependent flows is typically 

304: chaotic \citep{Pen96,Som96,Kar97} and possesses  complicated 

305: particle trajectories.

306: These flows, characterized by strong \emph{imperfect} mixing, 

307: lead to a fractal spatial distribution of

308: advected particles in a finite region of the flow.

309: This region is called the \emph{mixing region}.

310: In our terminology, a flow is chaotic if the advection dynamics

311: generated by the flow is chaotic.

312: 

313: In the case of several (three or more) types of passively advected tracers,

314: distinguished for example by their color, it was shown

315: \citep[and references therein]{Tor97} that

316: their distribution may follow a rather non-trivial topology on the

317: fractal, a property called the Wada property: every point \emph{on}

318: the fractal is lying on the boundary of at least three colors.

319: %Thus, when there are interactions among the species,

320: %then, due to the relatively long time the species spend on this

321: %fractal, and due to the largely increased interface between species due to the

322: %Wada property,

323: %one expects

324: %a non-trivial and novel type of behavior to emerge.

325: %, due to the

326: %complex nature of the mixing dynamics.

327: 

328: It is worth emphasizing that a complicated flow field (turbulence)

329: inside the mixing region

330: is not required for the flow to be chaotic, (i.e., for

331: complex advection dynamics or for the appearance of

332: fractal patterns).

333: Even simple forms of time dependence,

334: e.g.\ a periodic repetition

335: of the velocity field with some period $T$, is sufficient \citep{Are94}.

336: Thus, for sake of simplicity, 

337: we examine advection in time-periodic open flows.

338: 

339: The complicated form of trajectories implies a long time spent

340: in the mixing region. In other words, advected particles can be temporarily

341: trapped there.

342: It is even more surprising, however,  that

343: there is an \emph{infinity} of special

344: %for very special initial particle positions

345: nonescaping orbits.

346: %exist.

347: The simplest among these orbits are the periodic

348: ones with periods that are integer multiples of the

349: flow's period $T$.

350: All the nonescaping orbits are highly unstable, of saddle

351: type, and possess a

352: strictly positive local Lyapunov exponent

353: (which is the expanding eigenvalue of the unstable periodic orbit).

354: %, or saddle point).

355: Another important feature of these orbits is that

356: despite their infinite number

357: they are rather exceptional so that they cannot fill a finite portion

358: of the phase space. Indeed,

359: the union of all nonescaping orbits forms

360: a fractal ``cloud'' of points on any snapshot.

361: This fractal cloud moves

362: periodically with the flow and never leaves the mixing region.

363: 

364: Typical advected particle trajectories are not in the set of

365: %exactly reaching any of

366: the nonescaping orbits, but are, nevertheless, influenced by them.

367: They follow closely some periodic orbit

368: for a while and later turn to follow others.

369: This wandering amongst periodic (or, more generally, nonescaping)

370: orbits

371: results in the chaotic motion of passively advected particles.

372: Indeed, as long as the particles are in the mixing region,

373: their trajectories possess a positive average Lyapunov exponent $\lambda$.

374: Hence the union of all nonescaping orbits is called

375: the \emph{chaotic saddle}.

376: %It is to be emphasized that t

377: The flows relevant from our point of view

378: can be conisdered to be incompressible. This

379: results in a time-reversal invariant, area preserving

380: particle dynamics. Therefore, the negative average Lyapunov

381: exponent is exactly $-\lambda$, and it characterizes the

382: compression towards the chaotic saddle.

383: %It has a unique fractal dimension

384: %$D_{0}^{(saddle)}$.

385: %on a stroboscopic map, independent of the time instant at which

386: %the snapshot is taken.

387: 

388: While many of the particles spend a long time

389: in the mixing region, the overwhelming majority of

390: them leaves this region

391: sooner or later.

392: The decay of their number in a fixed frame is

393: typically \emph{exponential}

394: with a positive exponent

395: $\kappa$ ($ < \lambda$), which is independent of the frame, i.e.,

396: $N(t) = N(0) \exp(-\kappa t)$.

397: This quantity $\kappa$

398: is the \emph{escape rate} from the saddle (or from the mixing region).

399: The reciprocal of the escape rate can %also

400: be considered

401: as the average lifetime of chaos, and therefore the

402: chaotic advection of passive particles in open flows

403: is a kind of \emph{transient chaos} \citep{Tel90}.	

404: 

405: The chaotic saddle is the set of nonescaping orbits

406: which advected particles

407: may follow for an arbitrarily long time.

408: Each orbit of the set, and therefore the set as a whole,

409: has an  inflow  and an outflow curve, also called in the

410: mathematical jargon of chaos theory the stable and unstable

411: manifolds, respectively.

412: The \emph{inflow curve} is a set of points along which the

413: saddle can be reached after an infinitely long time.

414: The \emph{outflow curve} is the set along which

415: particles lying infinitesimally close to

416: the saddle will eventually leave it in the course of time.

417: By looking at different snapshots of these curves we can

418: observe that

419: they move periodically with the period $T$

420: of the flow. Their fractal dimension $D_0$ ($1< D_0 < 2$ in

421: two-dimensional flows) is, however, independent of the snapshot.

422: (The inflow and outflow curves

423: have identical fractal dimension due to the

424: advection dynamics' time reversal invariance.)

425: 

426: There is a unique relation between the fractal geometry and the

427: advection dynamics, expressed by the relation

428: \footnote{

429: By characterizing the dynamics by one single dimension $D_0$, we have assumed

430: that the advection process has a  monofractal geometry.

431: In reality, a set of dimensions $D_q$ is required for the

432: full description of the fractal aspects. It is for the $q=1$ dimension,

433: the so-called information dimension, $D_1$, for which

434: (\ref{KG}) is an exact equality.

435: In practice, however, the relative difference between $D_0$ and $D_1$ is

436: on the

437: order of a few  percents and therefore the use of a single dimension is

438: justified for practical purposes.}

439: \citep{Kan85,Hsu88,Tel90}:

440: \begin{equation}

441: D_0 = 2- \frac{\kappa}{\lambda}.

442: \label{KG}

443: \end{equation}

444: It says that the deviation of the dimension from that of the

445: plane is given by the ratio of two quantities characterizing the global

446: and the local instability of the dynamics. Relation (\ref{KG}) shows that

447: out of the three basic

448: characteristics ($\kappa$, $\lambda$ and $D_0$) only two are independent.

449: When speaking about population numbers in what follows, we shall use

450: the escape rate and the fractal dimension as independent parameters.

451: In the dynamics of the stripe widths (see next section), however, only the

452: average Lyapunov exponent appears.

453: 

454: The outflow curve plays a special role since it is the

455: only set which can be directly observed in an experiment.

456: Let us consider a droplet (ensemble)

457: of a large number of particles

458: which initially

459: overlaps with the inflow curve. As

460: the droplet is advected into the mixing region its

461: shape is strongly deformed, but the ensemble

462: comes closer and closer to the chaotic saddle as time goes on.

463: Since, however, only a small portion of particles can fall very

464: close to the inflow curve, the majority

465: does not reach the saddle and starts

466: flowing away from it along its outflow corve. Therefore,

467: % we conclude that 

468: in open flows  droplets of particles trace out

469: the outflow curve

470: of the chaotic saddle after a sufficiently long time of 

471: observation (in fact,

472: the populations in Fig.~\ref{fig:Compare} are distributed 

473: along the outflow curve

474: of the chaotic saddle present in the wake).

475: 

476: \section*{Dynamics of a single population}

477: 

478: In this section the mathematical derivation of

479: the dynamics of a single

480: population

481: living in an open  chaotic flow is briefly repeated 

482: \citep[][for more details see]{Tor98,Kar99,Tel00}.

483: Replication, competition for the limiting resources, and

484: spontaneous decay are taken into account in our population model,

485: while stage and age structure is neglected for simplicity.

486: We derive discrete and continuous-time models as well.

487: 

488: First we assume that the  intake of resource, multiplication and

489: decomposition are instantaneous and take place at integer

490: multiples of a time

491: lag $\tau$.

492: Here $\tau$ acts as an average time-scale on which the reproduction

493: takes place.

494: 

495: The basic observation is that after a sufficiently long time,

496: the filaments of the outflow curve are covered in narrow stripes

497: by individuals of species 

498: $B$  due to their replication \citep{Tor98,Kar99}. Individuals are thus

499: distributed on a fattened-up fractal

500: set. On linear scales larger than an average width $\varepsilon^*$ the

501: distribution of $B$

502: is a fractal of the same dimension $D_0$ as the outflow curve of

503: the chaotic saddle in the flow with passive advection

504: without biological activity. Let $\varepsilon^{(n)}$ 

505: denote the \emph{average} width

506: of  these stripes right before replication and decomposition takes place. It

507: is worth measuring this width in the unit of a characteristic  length scale

508: of the flow (e.g. in the cylinder radius in the example of 

509: Fig.~\ref{fig:Compare}). Thus,

510: $\varepsilon^{(n)}$ is a dimensionless variable.

511: %the following processes:  \emph{i)} replication, which increases the width by

512: %$\gamma$ (on average)

513: %and \emph{ii)} spontaneous decay due to deaths in the population,

514: %which decreases it by a distance $\mu$ (on average).

515: Since material $A$ is

516: available outside of these stripes,

517:  replication increases the width with some

518: constant distance $\gamma$,

519:  the replication range,

520:  while spontaneous decay

521: due to

522:  death of individuals decreases it with a distance $\mu$. The  net

523: effect of

524:  the replication and spontaneous decay is then a broadening of the

525: width by an

526:  amount proportional to the  difference $\sigma=\gamma-\mu$, the

527: effective

528:  replication range. Thus,

529: $\varepsilon^{(n)} \rightarrow \varepsilon^{(n)} + c \sigma$. Here $c$ is a

530: dimensionless number expressing geometrical effects. If the fattened-up

531: filaments do not overlap, then replication does occur on the both side of

532: stripes and  $c=2$. If there is  overlapping among some of the fattened-up

533: filaments, such as in the case of a fractal,   then $c \neq 2$. This

534: \emph{geometrical factor} turns out to be slightly time dependent

535: due to the pulsation of the flow, but for

536: simplicity it can be considered to be constant from the point of view of the

537: qualitative behaviour of the population \citep{Tor01}.

538: 

539: In the next period of length $\tau$ there is no replication and decomposition,

540: just contraction

541: towards the outflow curve.

542: The average contraction factor is $\exp{(-\lambda \tau)}$, where $(-\lambda$)

543: is the negative average Lyapunov exponent of the advection dynamics. Therefore,

544: the

545: width $\varepsilon^{(n+1)}$ right before the next replication can be given as

546: \begin{equation} 

547: \varepsilon^{(n+1)}=(\varepsilon^{(n)} + c \sigma) e^{-\lambda

548: \tau}. 

549: \label{rece} 

550: \end{equation}  

551: This is a recursive map for the actual

552: width of the $B$-stripes on snapshots taken with

553: multiples of the time lag $\tau$. The solution of (\ref{rece}) converges

554: for $n \rightarrow \infty$ to the fixed point

555: \begin{equation}

556: \varepsilon^*= \frac{c \sigma}{ e^{\lambda\tau}-1}. \label{eps*}

557: \end{equation}

558: 

559: In the time-continuous

560: limit

561: $\tau \rightarrow 0$,

562: $\sigma \rightarrow 0$, but keeping $\sigma/\tau \equiv v_r$ constant,

563: one obtains

564: the differential equation:

565: \begin{equation}

566: \frac{d \varepsilon}{d t}=c v_r -\lambda \varepsilon,

567: \label{epsdiff}

568: \end{equation}

569: which has a steady-state solution given by:

570: \begin{equation}

571: \varepsilon^* = \frac{c v_r}{\lambda}\;. \label{ec*}

572: \end{equation}

573: Here $v_r$ can be interpreted as the  net speed of replication.

574: 

575: Knowing the $\varepsilon$-dynamics and

576: that the individuals accumulate on a fractal

577: set in the mixing

578: region, the time evolution of the number $N$ of $B$ individuals

579: in that region   can be

580: calculated. First, note that the area ${\cal A}$ occupied by species

581: $B$  scales as ${\cal A} \approx \epsilon^{2-D_0}$

582: with $D_0$ as the fractal dimension of the outflow curve for any box size

583: $\epsilon$ not smaller than the width $\varepsilon$ of the $B$-stripes. We can

584: thus choose\footnote{In general, (\ref{scale}) also contains a proportionality

585: constant, called the Hausdorff volume. Since this only rescales  the constant

586: $q$ in equation (\ref{discr}), for clarity we took the Hausdorff valume

587: to be unity.}

588: \begin {equation}

589: \epsilon=\varepsilon \approx {{\cal A}}^{1/(2-D_0)}. \label {scale}

590: \end {equation}

591:  If the linear size of the area occupied by a single individual is

592: $\epsilon_0$, we have $N = \epsilon_0^{-2} {\cal A}$, and therefore we can

593: rewrite (\ref {rece}) or (\ref{epsdiff}) so that it represents an equation for

594: the  individuals in discrete and continuous cases,

595: respectively:

596: 

597: \begin {equation}

598: N^{(n+1)}=e^{-\kappa \tau}

599: \{[N^{(n)}]^{1/(2-D_0)}+q \sigma\}^{(2-D_0)}, \label{discr}

600: \end {equation}

601: and

602: \begin{equation}

603: \frac{d N}{dt}=-\kappa

604: N+q (2-D_0) {v_r} N^{-\beta}, \label{diff}

605: \end{equation}

606: with

607: \begin{equation}

608: q = c \;\epsilon_0^{-2/(2-D_0)}. \label{eq:cn}

609: \end{equation}

610:  Here (\ref{KG}) has been used, and

611: \begin{equation}

612: \beta \equiv \frac{D_0-1}{2-D_0}

613: \label{beta}

614: \end{equation}

615: appears as a nontrivial exponent.%

616: \footnote{For multifractal flows

617: one can show

618: \citep{Tel00} that exponent $\beta$ is that given by

619: (\ref{beta}) with $D_0$ replaced by the information dimension

620:  $D_1$.}

621: Since the fractal dimension of the outflow curve lies

622: between $1$ and $2$, exponent $\beta$ is positive. For $D_0=1$ the differential

623: equation (\ref{diff}) describes a classical surface reaction along a line with

624: front velocity $v_r$ in the presence of escape. For $1<D_0<2$ it represents a

625: novel form of dynamical equations containing also an enhancing biological

626: activity term

627: with a negative power of the

628: area occupied by $B$ due to the fractality of the outflow curve.

629: The less $B$ individuals are present, the more

630: effective the reproduction is, because the resolved perimeter is

631: larger. Consequently, in a competitive situation the subordinate species has

632: an advantage if it becomes rare compared to the dominant species. This

633: balancing mechanism can make  coexistence possible, as shown in the

634: the next sections.

635: 

636: As one can see from Eqs.~(\ref{discr},\ref{diff}), in both the discrete

637: and continuum pictures a steady state is reached after a sufficiently

638: long time if the geometrical factor $c$ (and therefore also $q$)

639: is constant \citep{Tor01}. In this case, the steady-state number of

640: individuals in the mixing region is

641: $N^* =\epsilon_0^{-2}(\varepsilon^*)^{2-D_0}$ where

642: $\varepsilon^*$ is given by

643: (\ref{eps*}) and (\ref{ec*}) for the discrete and continuum cases,

644: respectively.

645: 

646: 

647: \section*{A model of competition}

648: 

649: 

650: 

651: As in the single species case, we consider a simple

652: model of replication and competition with

653: passively advected point like individuals

654: of type $B_1$ and $B_2$, multiplying themselves instantaneously.

655: The resource material $A$ which the different species $B_1$

656: and $B_2$ compete for

657: is uniformly distributed on the

658: surface of the flow.

659: Similarly to the single species case, the parameters $\gamma_i$ and $\mu_i$

660: ($i=1,2$) are defined as the increase and decrease of the $B_i$ stripe width

661: due to replication and decomposition, respectively.

662: Therefore two

663: autocatalytic processes $A+B_1 \longrightarrow 2B_1$,

664: $B_1 \longrightarrow  A$ and $A+B_2 \longrightarrow 2B_2$,

665: $B_2 \longrightarrow A$

666: represent the  replication and competition

667: process in our model in an imperfectly mixed environment.

668: Similarly to the single species case, the parameters $\gamma_i$ and $\mu_i$

669: ($i=1,2$) are defined as the increase and decrease of the $B_i$ stripe width

670: due to replication and decomposition, respectively, so that

671: the effective replication distances are

672: $\sigma_i=\gamma_i-\mu_i$.

673: %There is an important comment to be made at this point on the valability

674: %of treating spontaneous deaths as an effective decrease in the average

675: %widths of the filaments.  Let us imagine again that we are in the

676: %co-moving frame of a small fluid element on one of the unstable filaments.

677: %Let us also  assume that this fluid element has both species, $B_1$ and $B_2$

678: %in it, i.e. there is an interface between the two. Due to spontaneous deaths,

679: %occasionaly empty spaces are created on the $B_1 - B_2$ interface. Obviously,

680: %these spaces will be taken over more frequently by the fittest species, and

681: %thus, after sufficiently long times, in the fluid element, the weaker species

682: %will be outcompeted. However, during this time the fluid element may have

683: %been washed out from the mixing region into the asymptotic, non-mixing

684: %region and thus it is of no interest any longer (we are interested

685: %in coexistence effects due to dynamical mixing and thus

686: %in the mixing region). It is therefore crucial to compare the relevant

687: %time-scales. In a typical situation spontanous death occurs after

688: %the species give birth several times (at least once), and thus the

689: %average lifetime of an individual $t_i^{(d)}$ is greater than the birth time,

690: %$t_i^{(b)}$, $t_i^{(d)} > t_i^{(b)}$. If we assume that the lifetimes

691: %$t_i^{(d)}$ are much longer than the average time of a trajectory spent

692: %in the mixing region, i.e., $t_i^{(d)} \gg 1/\kappa$, the spontaneous

693: %deaths on the $B_1 - B_2$ interface will lead to the extinction of the

694: %weaker species only long time after the fluid element left the mixing region

695: %along the unstable manifold, and thus it will not have a crucial effect

696: %on the distribution and composition of the fattened filaments in the

697: %mixing region. In this case, one may safely model the spontaneous

698: %death as an effective decrease of the filaments for each species.

699: 

700: As before, an important feature

701: of the advection dynamics is its  deterministic nature. Concerning the

702: population dynamics, this implies that  we work in the limit of weak diffusion

703: and assume that the mutual diffusion coefficients  between any pair of the

704: constituents is small.

705: 

706: Prior to discussing the consequences of the imperfect mixing generated by the

707: chaotic flow to this dynamics, it is worth briefly giving the

708: traditional equations governing the above defined autocatalytic processes

709: in a well-mixed environment. In a fixed region of observation

710: they are:

711: \begin {equation}

712: \frac{dN_1}{dt}=\gamma_1AN_1-\mu_1N_1,

713: \label {trad2}

714: \end {equation}

715: \begin {equation} 

716: \frac {dN_2}{dt}=\gamma_2AN_2-\mu_2N_2, 

717: \label {trad3}

718: \end {equation}

719: where $N_i$ denotes the number of individuals of

720: species $B_i$, and $A$ is the instantaneous

721: amount of the

722: resource material in the same region. Note that the meaning of the

723: replication and death rates are slightly different here from those in the  

724: discrete model (thus e.g., $\mu_i$ in eqs. (\ref{trad2}), (\ref{trad3})

725: is of dimension frequency, while the same quantity in the

726: discrete version is a distance).

727: If the dynamics of resource is much

728: faster than the dynamics of competing species, then

729: the former can be considered to be in a quasistationary state: $dA/dt=0$.

730: The equation for resource $A$ is then

731: \begin {equation}

732: \frac {dA}{dt} = 0 =l-\gamma_1AN_1 -\gamma_2AN_2, 

733: \label {trad1}

734: \end {equation}

735: and $l$ is the constant inflow of

736: resource $A$ into the region of observation.

737: Equations (\ref{trad2}), (\ref{trad3}) correspond to the general scheme

738: (\ref{trad0}) and (\ref{trad}) 

739: given in the Introduction by identifying $\mu_i$ with

740: $\delta_i$ and $\gamma_i A$ (where $A$ is given by the right hand side of

741: (\ref{trad1})) with $\alpha_i$. 

742: 

743: After analyzing (\ref {trad3}) and (\ref {trad1}),

744: one can easily see that species with lower ratio

745: $\gamma_i/\mu_i$ of replication and death rates

746: would be outcompeted, and thus stable coexistence is impossible.

747: 

748: \section* {Coexistence of competing species}

749: 

750: In the following, using a gedankenexperiment, we find

751: conditions

752:  under which the two species may coexist in the imperfectly mixing

753: environment

754:  characterized by the existence of a chaotic saddle and its

755: (fractal) outflow

756:  curve.

757:  First, let us assume that initially there is only

758: one species, for e.g., $B_2$

759:  in the mixing region. Also for the simplicity of

760: the writing, we will refer

761:  to the continuum version of the single species

762: dynamics, namely

763:  Eq.~(\ref{diff}).

764:  After a time the number $N_2$ of $B_2$

765: will be close to a steady state value,

766:  $N_{2}^*$. We now let a small quantity

767: of species $B_1$ invade the

768:  mixing region, so small

769:  that it cannot change

770: the steady state of $B_2$.

771:  The question is: under what conditions this

772: invasion

773:  will lead to a self-sustained $B_1$ population, coexisting with

774: $B_2$?

775:  

776: %Recall, that when deriving Eqs.~(\ref{diffA}) and (\ref{diff}) we made

777: %use of the following conditions: \emph{i)} there is growth of the

778: %covering material on the filament with a rate determined by the difference

779: %between the births and spontaneous deaths; \emph{ii)} the distribution

780: %of the filaments follows a fractal scaling. In the case of \emph{i)},

781: %the crucial component for growth is the length of the interface between

782: %resource $A$ and the species, since only at this interface the reactions

783: %are active. This quantity is introduced through the phenomenological

784: %geometrical factor, as discussed previously.

785: When we are letting species

786: $B_1$ invade the mixing region such that $B_2$ is already present, we must

787: ensure for the coexistence that there is not only $B_1 - B_2$ interface

788: present but also $B_1 - A$ interface, at all times. To

789: show that this is indeed  possible, we recall the Wada property of mixing on

790: the chaotic saddle,  mentioned previously. As shown by 

791: \citep{Tor97},

792: if species $B_1$ is inserted such that the $B_1 - A$ interface

793: \emph{intersects} 

794: %at least one of 

795: the inflow curve, then this boundary must

796: be present arbitrarily close to all points on the outflow curve. Therefore

797: this initial condition ensures the existence of the $B_1 - A$ interface

798: at all times on the saddle and its outflow curve.

799: Thus, when there are interactions among the species,

800: then, due to the relatively long time they spend on this

801: fractal, and due to the largely increased interfaces

802: a non-trivial and novel type of behavior can emerge.

803: % Since the time-scale

804: %of the spontaneous deaths is larger than the exit time from the saddle,

805: %the $B_1 - B_2$ interface will not advance within the mixing region

806: %to the detriment of the weaker species.

807: 

808: 

809: Species $B_1$  becomes thus trapped by		

810: the chaotic saddle, and will be distributed in very narrow filaments

811: along the surface  of some of the $B_2$ stripes close to the

812: $D_0$ dimensional outflow curve.

813: Since by assumption, the population number $N_1$ of species

814: $B_1$ is small, ${N_{1}}/{N_{2}} \ll 1$, there is no feedback on $B_2$

815: and this species remains in dynamical equilibrium:		

816: ${d N_{2}}/{d t}\approx 0$.

817: The dynamics of species $B_1$ can thus

818: be described by (\ref{diff}) written for $N_1$.

819: The

820: important nature of (\ref{diff}) is that the first term on the right side,

821: (which is responsible for the outflow from the fractal set)

822: tends to zero

823: for very small $N_{1}$. The second term, however,

824: which describes an autocatalytic process can be arbitrarily large if

825: $N_{1}$ is small,

826: due to the negative exponent $-\beta$.

827: Thus, if $N_{1}$ is sufficiently small then  ${d

828: N_{1}}/{d t} > 0$.

829: It means that the $B_1$ population always increases if it's

830: number is close to zero.

831: In other words, the fixed point $N_1^*=0$ is unstable.

832: 

833: %Similarly, if $B_2$ invades the fractal set where $B_1$ is already on

834: %dynamical equilibrium, then the number $N_{2}$

835: %of $B_2$ increases when it is sufficiently

836: %small. Consequently neither species can be excluded from the fractal set, they

837: %will be in coexistence. Certainly, if the invading species does not cross

838: %the inflow curve of the saddle, it will not be trapped by it, and thus

839: %it will be transported away from the mixing region.

840: 

841: Following the same argument,

842: if ${B_1}$ and ${B_2}$ are already in dynamical

843: equilibrium on the fractal set then a third species can invade this coalition

844: in the same manner.

845: %Even more generally, if  any of $n$ species out of $n+1$

846: %competing ones are in a steady state in the mixing region,

847: %then the $(n+1)$st can

848: %invade, and  $n+1$ species will be in coexistence.

849: This coexistence has indeed been demonstrated for

850: three species by numerical simulations \citep{Kar00}.

851: 

852: %While coexistence of competitors in open chaotic flows can be shown with the

853: %help of the above invasion analysis, the global dynamical description of these

854: %population

855: %valid for comparable population numbers will be delt with

856: %in the following section.

857: 

858: In this gedankenexperiment we assumed that the invading population

859: is so weak that it is distributed in narrow filaments

860: along the ``surface'' of the already existing population and does not

861: influence this population at all. After the number of the invading

862: population has started to grow, there is an increasing interaction

863: between the populations due to the competition for the same

864: resource. As a consequence, the originally steady population

865: is not

866: %does not feel itself to be 

867: in a stationary state and the dynamics

868: should be treated in a selfconsistent manner. This is

869: shown in the next section.

870: 

871: \section*{A mathematical model for the competition dynamics}

872: 

873: After sufficiently long time,

874: both species $B_1$ and $B_2$  will be distributed in

875: narrow stripes along the chaotic saddle's outflow curve as follows from the

876: passive advection dynamics. Due to the replication and decomposition,

877: however, the stripes have finite widths (cf.\ Fig.~\ref{fig:Compare}) 

878: which might depend on time.

879: Let $\varepsilon^{n}$ denote the dimensionless average width of the stripes

880: right before an instantaneous replication

881: %biologically active step 

882: takes place. These stripes are

883: defined by the fact that outside of them there is only background material

884: $A$ available. Inside the stripe of width

885: $\varepsilon^{(n)}$

886: there might be several narrow $B_1$ or $B_2$

887: filaments.

888: % and spots of $A$ (the latter due to sponatenous deaths).

889: %Since the creation of $A$ spots

890: %due to the spontaneous decay is slow compared to the dynamics in the mixing

891: %region, the $B_1 - B_2$ interface  will not be varying in a co-moving

892: %fluid element, as we mentioned earlier.

893: %Also,

894: The background material $A$

895: is eaten up sooner or later in the inside of any stripe, therefore,

896: for the sake of an easier presentation,  we assume that this is the case and

897: only material $B_1$ and $B_2$ are present. Let us

898: denote the total widths of all the filaments of a given material  within an

899: $\varepsilon^{(n)}$ stripe  by $\varepsilon_i^{(n)}$

900: with $i=1,2$ corresponding to $B_1$ and $B_2$, respectively.

901:  The sum of

902: these partial widths is of course the total one

903: $\varepsilon_1^{(n)} + \varepsilon_2^{(n)}=

904: \varepsilon^{(n)}$.

905: Our aim is to build up the dynamics of the partial widths

906: based on plausible assumptions, from which

907: the dynamics of the different populations follows.

908: 

909: We assume,

910: that the boundaries are occupied  by

911: species $B_1$ or $B_2$

912: with \emph{probabilities} $p_1$ and $p_2$, respectively.

913: In other words, a stripe-boundary picked at random from

914: the many filaments of the outflow

915: curve will have a probability  $p_i$ to be of type $B_i$,

916: $i=1,2$.

917: If mixing of the two species were

918: perfect along the fractal set, these probabilities would be equal to their

919: relative densities.

920: This is not the case, however.

921: The relative position of the species in the initial

922: distribution  to the inflow

923: curve determines which individual or patch of individuals will be

924: trapped by which orbit of the chaotic saddle.

925: The rest, i.e., the untrapped individuals will drift out of the

926: mixing region.

927: The trapped individuals, however, will stay there forever, and follow

928: their specific trapped orbit.

929: In the course of time, individuals give birth to others of the

930: same species, and patches of individuals are stretched along the

931: outflow curve specific to the trapping orbit of the chaotic saddle.

932: In either cases, we end up with long stripes of the two species

933: lined up along each other in an alternating manner, tracing out the

934: outflow curve.

935: Then the probability of one species to be on the edge of these lines,

936: and thus to be capable of reproduction, depends

937: on which trapping orbit will produce the filament of

938: outflow curve being on the edge of the stripe,

939: on the order in which the species are lined up across one stripe,

940: and on the actual width of the coverage of the filaments.

941: In other words, it is the complex chaotic dynamics which makes

942: the introduction of probabilistic concepts---on a somewhat

943: phenomenological level---unavoidable.

944: 

945: The probabilities $p_i$

946: depend on what the distribution of the species

947: inside the stripes is. % on the partial widths of each

948: %species inside the stripe.

949: Thus, the simplest possible assumption is that the probabilities

950: depend on the partial widths $\varepsilon_i^{(n)}$.

951: Their actual functional

952: form might also contain parameters of the flow and of the

953: biological activity.

954: 

955: 

956: Naturally the probabilities fulfill $0 \le p_1^{(n)} \le 1$ and

957: $p_2^{(n)}=1-p_1^{(n)}$.

958: They might have a general dependence

959: on the partial widths $\varepsilon_i^{(n)}$,

960: $i=1,2$: $p_i^{(n)}=p_i^{(n)}\left(\varepsilon_1^{(n)},\varepsilon_2^{(n)}

961: \right)$.

962: Due to dimensional reasons they can only depend on the

963: ratio $z^{(n)}\equiv \varepsilon_1^{(n)}/\varepsilon_2^{(n)}$. Thus

964: we write

965: \begin{equation}

966: p_1^{(n)} = g\left(z^{(n)}, \omega \right), \;\;\;\;\;

967: p_2^{(n)}=1-g\left(z^{(n)}, \omega \right). \label{pe}

968: \end{equation}

969: Here $\omega$ is a parameter of the distributions, and incorporates

970: the dependence on the replication rates.

971: We also made the plausible assumption, that $g$ has no explicit $n$ (or

972: time) dependence.

973: A general property of $g$ is that it vanishes in the origin

974: $g(0) = 0$ since this expresses the obvious fact that

975: if species $B_1$ is missing, then the probability to find it

976: in the filaments is zero. Similarly for infinitely large values of

977: $z$ it must be unity: $g(\infty)=1$ which corresponds to the absence

978: of $B_2$.

979: Also, due to the fact that $g(z)$ is a

980: probability distribution, we must have $0 \leq g(z) \leq 1$

981: for all $z \geq 0$. Furthermore, the functional form must be

982: symmetric by interchanging the role of the species. This implies

983: that one must have $g'(0) \geq 0$, where the prime denotes

984: derivation with respect to the argument.

985: This implies

986: \begin{equation}

987: p_2^{(n)}=g\left(1/z^{(n)},1/\omega \right),

988: \end{equation}

989: where the appearance of $1/\omega$ means that an interchange of the

990: species index brings the parameter in its reciprocal value,

991: as e.g.\ in the case when $\omega=\sigma_1/\sigma_2$ (the dependence 

992: on the ratio of the replication distances follows from dimensional

993: reasons).

994: The normalization of the probability implies

995: \begin{equation}

996: g\left(z,\omega \right)+

997: g\left(1/z,1/\omega \right)=1 \label{feltetel}

998: \end{equation}

999: This is a functional equation for $g$. With the above properties and

1000: boundary conditions we find that a family of solutions is given by

1001: the form:

1002: \begin{equation}

1003: g(z) = \frac{z^{\alpha}}{z^{\alpha} +\omega}, \label{pe2}

1004: \end{equation}

1005: with $\alpha$ and $\omega$ as two positive parameters. In the range

1006: of $0<\alpha<1$ the smaller population is less probable on the

1007: boundary but yet with a weight which is weaker than linear in the widths.

1008: For $\alpha=0$ there is no width-dependence at all, the probabilities

1009: $p_i$ are constant.

1010: The case $\alpha=1$ and $\omega=1$ corresponds to a homogeneous mixing within

1011: the stripe of width $\varepsilon$. For $\alpha>1$ a superdominance is

1012: described.

1013: %The %range where

1014: %parameters of $g(z)$ will be

1015: %determined by numerical simulations in the next section.

1016: In the next section we show that  the form

1017: (\ref{pe2}) of $g(z)$ is indeed in good agreement with

1018: numerical simulations, and determine values for parameters

1019: $\alpha$ and $\omega$.

1020: 

1021: The broadening of the average widths is then $c \sigma_1 p_1^{(n)}$ and

1022: $c \sigma_2 p_2^{(n)}$  due to species $B_1$ and $B_2$, respectively. Here

1023: the geometrical factor $c$

1024: and parameter $\sigma_i=\gamma_i-\mu_i$ have the same meanings  as

1025: in the single species problem defined previously.

1026: 

1027: 

1028: Thus, similar to (\ref{rece}) the partial width of

1029: $B_i$

1030: %and $B_2$

1031: after the $(n+1)$st

1032: %replication

1033: step

1034: %and right before the

1035: %$(n+2)$nd step are

1036: is

1037: \begin{equation}

1038: {\varepsilon_i^{(n+1)}}=\left[

1039: {\varepsilon_i^{(n)}}+

1040:  c \sigma_i p_i^{(n)} \right] e^{-\lambda \tau}

1041: \label{eq:epsi}

1042: \end{equation}

1043: for $i=1,2$.

1044: Note that in our theory, $c p_1$ and $c p_2$

1045: can also be interpreted as renormalized

1046: geometric factors for each species, due to the screening effects at the

1047: boundaries of the stripes.

1048: As a consequence, the total width of the stripes changes at a replication

1049: as

1050: \begin{equation}

1051: {\varepsilon^{(n+1)}}=\left[

1052: {\varepsilon^{(n)}}+ c

1053: ( \sigma_1 p_1^{(n)}+ \sigma_2 p_2^{(n)})

1054: \right]e^{-\lambda \tau}.

1055: \label{eq:eps}

1056: \end{equation}

1057: For simplicity, the explicit width-dependence (\ref{pe}) 

1058: of the probabilities

1059: has not been written out.

1060: For %$p_1=p_2=1/2$ and

1061: $\sigma_1=\sigma_2$ we recover the width dynamics

1062: of the single species problem, see (\ref{rece}).

1063: 

1064: Next we turn to the dynamics of the number of individuals.

1065: %(the number dynamics simply follows from the relation

1066: %$N_{B_i}^{(n)} = \big[\epsilon_{0}^{(i)} \big]^{-2} {\cal A}_i^{(n)}$,

1067: %where $\epsilon_{0}^{(i)}$ is the linear size of the space a single

1068: %individual of type $B_i$ occupies).

1069: On scales larger than or equal to $\varepsilon^{(n)}$,

1070: the total  number of individuals $N =N_1 + N_2$

1071: occupied by stripes appears to be a fractal of the same

1072: dimension $D_0$ as the outflow curve. For simplicity of writing we assume

1073: that both species have the same size $\varepsilon_0$ (an extension for

1074: different sizes is straightforward).

1075: 

1076: Since the relation between the $\varepsilon^{(n)}$ and the number of

1077: individuals $N^{(n)}$ is the same as in the single species

1078: model, we can use (\ref {scale}).

1079: Thus (\ref{eq:eps})

1080: implies a recursion  for the  area

1081: right before  replication as

1082: \begin{equation}

1083: N^{(n+1)}\!\!= \!e^{-\kappa \tau}\!\!

1084: \left\{ \!\left[ N^{(n)}\right]^{1/(2-D_0)} \! +\!

1085: q\left[\sigma_1 p_1^{(n)} + \sigma_2 p_2^{(n)} \right]

1086: \right\}^{2-D_0}\label{eq:mapauto}

1087: \end{equation}

1088: with $q$ given by (\ref{eq:cn}).

1089: Next, we derive the dynamics of the  number of

1090: individuals $N_i^{(n)}$ for 

1091: %the different 

1092: species $i$ contained in the stripes.

1093: The number of individuals of species $i$ is the 

1094: portion of the total

1095: number $N^{(n)}$ proportional to the partial widths:

1096: \begin{equation}

1097: N_i^{(n)}=N^{(n)}

1098: \frac{\varepsilon_i^{(n)}}{\varepsilon^{(n)}}. \label{eq:Ai}

1099: \end{equation}

1100: This is due to the fact that there is no

1101: fractal scaling below $\varepsilon^{(n)}$. Since

1102: $\varepsilon^{(n)}=[\epsilon_0^2 N^{(n)}]^{1/(2-D_0)}$,

1103: Eq.~(\ref{eq:Ai}) leads to

1104: \begin{equation}

1105: \varepsilon_i^{(n)} = N_i^{(n)} \epsilon_0^{2/(2-D_0)}

1106: \left[ N^{(n)} \right]^{\beta}. \label{aiei}

1107: \end{equation}

1108: As another consequence of (\ref{eq:Ai}), the ratio of the partial widths

1109: is the ratio of the population numbers:

1110: \begin{equation}

1111: z^{(n)} \equiv \frac{\varepsilon_1^{(n)}}{\varepsilon_2^{(n)}} =

1112: \frac{N_1^{(n)}}{N_2^{(n)}}.

1113: \label{pA}

1114: \end{equation}

1115: {From} Eqs.~(\ref{eq:epsi}) and (\ref{aiei})

1116: we therefore  obtain the dynamics of the population

1117: numbers as

1118: \begin{eqnarray}

1119: N_i^{(n+1)}[N^{(n+1)}]^\beta = e^{-\lambda\tau}

1120: \left\{ N_i^{(n)}\left[N^{(n)}\right]^\beta 

1121: \right. && \nonumber \\

1122: \left. +  q \sigma_i p_i^{(n)}

1123: \left( N_1^{(n)}/

1124: N_2^{(n)}\right) \right\} && \label{eq:mapautoA}

1125: \end{eqnarray}

1126: for $i=1,2$. Here exponent $\beta$ is the same expression

1127: (\ref{beta}) as in the case of the single species problem,

1128: and $q$ is given by (\ref{eq:cn}).

1129: 

1130: We observe that

1131: by dividing the rearranged

1132: (\ref{eq:mapautoA}) for $i=1$ by that with $i=2$, one obtains

1133: \begin{equation}

1134: \frac{\sigma_1 p^{(n)}_1}{\sigma_2 p^{(n)}_2}= M^{(n)},

1135: \end{equation}

1136: where

1137: \begin{equation}

1138: M^{(n)}=\frac{e^{\lambda}N_1^{(n+1)}[N^{(n+1)}]^{\beta}-

1139: N_1^{(n)}[N^{(n)}]^{\beta}}{

1140: e^{\lambda}N_2^{(n+1)}[N^{(n+1)}]^{\beta}-

1141: N_2^{(n)}[N^{(n)}]^{\beta}}.

1142: \end{equation}

1143: {From} this, $p_1=1-p_2$ is easily found as

1144: \begin{equation}

1145: p^{(n)}_1=\frac{M^{(n)}}{M^{(n)}+

1146: \frac{\sigma_1}{\sigma_2}},\label{eqfit}

1147: \end{equation}

1148: This relation provides us with a method for measuring

1149: how the probability $p_1(z) \equiv g(z)$ depends on the

1150: ratio $z\equiv N_1/N_2$ at any instant of time. We shall use

1151: this observation to extract the form of the $g$ function from

1152: numerical results. The ratio of fixed points,

1153: $z^* \equiv \frac{N_1^*}{N_2^*}$ can be calculated from (\ref{eqfit})

1154: assuming that $N_2^* \neq 0$ as

1155: \begin {equation}

1156: g(z^*)=\frac{M^{*}}{M^{*}+\frac{\sigma_1}{\sigma_2}},\label{fixdisc}

1157: \end{equation}

1158: 

1159: 

1160: The time continuous limit is obtained

1161: by letting  both the time lag and the effective replication

1162: ranges go to zero so that their

1163: ratios remain finite. Thus we define replication velocities

1164: \begin{equation}

1165: v_i = \lim_{\tau \rightarrow 0} \frac{\sigma_i}{\tau},

1166: \end{equation}

1167: with $i=1,2$ for species $B_1$, $B_2$, respectively.

1168: In the continuous time limit,

1169: the differential equations

1170: obtained for the partial widths from (\ref{eq:epsi}) read as

1171: \begin{equation}

1172: \frac{d \varepsilon_i}{dt}=-\lambda {{\varepsilon}_i}

1173: + c v_i p_i(\varepsilon_1 / \varepsilon_2)

1174: \label{epsi}

1175: \end{equation}

1176: where $p_1=g$, $p_2=1-g$.

1177: %Here $\lambda=\kappa/(2-D_1)$ is the Lyapunov exponent of the passive

1178: %advection's dynamics.

1179: 

1180: The differential equation for the number of all individuals follows from

1181: (\ref{eq:mapauto}) as

1182: \begin{equation}

1183: \frac{d N}{dt} = -\kappa N+ q (2-D_0) v N^{-\beta}.

1184: \label{diffeq}

1185: \end{equation}

1186: Here

1187: \begin{equation}

1188: v \equiv p_1 v_1 + p_2 v_2

1189: \label{avv}

1190: \end{equation}

1191: is an average velocity, but note that it is not a constant since

1192: the $p_i$ depend on the population numbers, and $q$ is given by (\ref{eq:cn}).

1193: 

1194: The differential equation for the number 

1195: $N_i$ of individuals of

1196: the two species can be derived from (\ref{epsi}) and the continuum

1197: version of (\ref{aiei}), i.e., 

1198: $\varepsilon_i = \epsilon_0^{2/(2-D_0)}N_i

1199: N^{\beta}$. The result is

1200: \begin{eqnarray}

1201: \frac{d N_i}{dt} = -\kappa N_i-q(D_0-1) v

1202: N^{-\beta-1} N_i+ &&\nonumber \\

1203: q v_i p_i \left( N_1 / N_2\right) N^{-\beta},&& \label{diffeqi}

1204: \end{eqnarray}

1205: with $N = N_1 + N_2$.

1206: Here (\ref{KG}) and (\ref{beta}) have been used.

1207: By summing over $i$ in (\ref{diffeqi})

1208: one recovers Eq.~(\ref{diffeq}).

1209: 

1210: An equivalent form is obtained after rearranging terms and 

1211: taking into account the

1212: definition of the average replication velocity (\ref{avv}). It reads

1213: \begin{eqnarray}

1214: \frac{d N_1}{dt} = -\kappa N_1 +

1215: qN^{-\beta-1}\left[ (2-D_0)v N_1+ \right. &&\nonumber \\

1216: \left.(v_1 p_1 N_2 -v_2 p_2 N_1) \right] &&,\label{diffeq1}

1217: \end{eqnarray}

1218: and an analogous expression for the second species obtained 

1219: from (\ref{diffeq1}) by interchanging the indices $1$ and $2$.

1220: It can be clearly seen that the first term of the

1221: bracket corresponds to the growth 

1222: of the total population, while the second describes the effect due to a

1223: weighted difference in the population numbers.

1224: Expression (\ref{diffeqi}) or (\ref{diffeq1})

1225: represents a strongly coupled set of nonlinear equations

1226: with  a novel type of power-law behavior (with negative exponent

1227: $-\beta$). This set of equations is the central result of our paper since it can 

1228: be

1229: considered as a population dynamics

1230: describing the coupling of two  populations {mixing} on a fractal,

1231: and as we show below,

1232: opens up the possibility to have a nontrivial

1233: coexistence.

1234: 

1235: If one of the species, say $B_2$, is not present, then $p_1=1$, 

1236: $N_2=p_2=v_2=0$ and

1237: hence $v=v_1=v_r$, $N=N_1$ and Eq.~(\ref{diffeq}) becomes equivalent to

1238: (\ref{diff}). The same happens if both species are equivalent, i.e., for

1239: $v_1=v_2$ when $v=v_r$. 

1240: 

1241: A simple further equivalent  form can be derived

1242: by using relative densities $c_i \equiv N_i/N$. 

1243: The equations describing the populations then become

1244: (by using (\ref{diffeq1}) and (\ref{diffeq}):

1245: \begin{equation}

1246: \frac{d c_1}{dt}  = q N^{-\beta - 1} (v_1 p_1 c_2- v_2 p_2 c_1) 

1247: \end{equation}

1248: with $c_1+c_2=1$. The temporal change of the densities is determined

1249: by the weighted relative difference in the densities.

1250:  Note that they  are {\em multiplicatively}

1251: coupled to $N^{-\beta -1}$ which is proportional to the average

1252: width of the filament covering. For $D_0 = 1$ this is just $1/N$ and

1253: it is the { spatial concentration} or the density of the total population.

1254: For $2 > D_0 > 1$ (fractals), this factor is the \emph{fractal} 

1255: spatial density of the population as a whole.

1256: 

1257: \section*{Coexistence analysis: fixed points and their stabilities}

1258: 

1259: It is simple and instructive to study the time-continuous

1260: dynamics of the widths

1261: $\varepsilon_i$, $i=1,2$ in and around steady states.

1262: Assuming, as before,  $c=\mbox{const.}$, we find from (\ref{epsi})

1263: \begin{equation}

1264: \lambda {{\varepsilon}_i^*}=

1265: c v_i p_i \left( {\varepsilon_1^*} 

1266: / {\varepsilon_2^*} \right).\label{fpi} 

1267: \end{equation}

1268: The weighted sum of the fixed points widths (\ref{fpi}) gives

1269: \begin{equation}

1270: \varepsilon_1^* v_2+

1271: \varepsilon_2^* v_1=  \frac{c v_1 v_2}{ \lambda}. \label{fp1}

1272: \end{equation}

1273: {From} here on, for the stability analysis, we shall use the

1274: explicit form (\ref{pe2}) for the function $g$.

1275: % and refer the

1276: %interested reader to Appendix A, where the analysis is performed

1277: %for an arbitrary $g$ function (which, however, still respects

1278: %Eqs.~(\ref{pe}-\ref{feltetel})). 

1279: Thus (\ref{fpi}) translates into:

1280: \begin{eqnarray}

1281: &&\lambda {{\varepsilon}_1^*}=

1282: c v_1 \frac{ {\varepsilon_1^*}^{\alpha}}{

1283: {\varepsilon_1^*}^{\alpha}+\omega

1284: {\varepsilon_2^*}^{\alpha}} \label{fpio1} \\

1285: && \lambda {{\varepsilon}_2^*}=

1286: c v_2  \frac{ {\omega\varepsilon_2^*}^{\alpha}}{

1287: {\varepsilon_1^*}^{\alpha}+\omega

1288: {\varepsilon_2^*}^{\alpha}} \label{fpio2}

1289: \end{eqnarray}

1290: 

1291: Formula (\ref{fp1}) shows that one species always survives. Without

1292: loss of generality, we can choose this to be species $B_2$, and

1293: everything remains valid with the indices 1 and 2 switched.

1294: %and this symmetry will also be reflected in the stability diagram.

1295: It is worth defining:

1296: \begin{equation}

1297: z^* \equiv \varepsilon_1^*/\varepsilon_2^*.  \label{zst}

1298: \end{equation}

1299: Since $\varepsilon_2^*$ is not zero, the ratio of the fixed point equations

1300: (\ref{fpio1},\ref{fpio2}) yields:

1301: \begin{equation}

1302: {z^*}^{1-\alpha} = \frac{v_1}{v_2 \omega}\;\;\;\mbox{or}\;\;\;

1303: {z^*} = 0\label{fp2}

1304: \end{equation}

1305: The first equality describes the $z^* \neq 0$ \emph{coexistence}

1306: fixedpoint while the second describes the \emph{non-coexistence} fixedpoint.

1307: 

1308: The  equations (\ref{epsi}) of the continuous case  written out 

1309: explicitely are as follows:

1310: \begin{eqnarray}

1311: &&\frac{d \varepsilon_1}{dt}=-\lambda {{\varepsilon}_1}

1312: + c v_1 \frac{\varepsilon_1^{\alpha}}{\varepsilon_1^{\alpha} + 

1313: \omega \varepsilon_2^{\alpha}} \label{flow2} \\

1314: &&\frac{d {\varepsilon}_2}{dt}=-\lambda {{\varepsilon}_2}

1315: + c v_2 \frac{\omega \varepsilon_2^{\alpha}}{\varepsilon_1^{\alpha} + 

1316: \omega \varepsilon_2^{\alpha}} \label{flow1}

1317: \end{eqnarray}

1318: The linear stability of a fixed point  $(\varepsilon_1^{*}>0,\varepsilon_2^{*}>0)$

1319: will be

1320: given by the eigenvalues of the stability matrix ${\bf E}$, calculated from

1321: (\ref{flow2}-\ref{flow1}) as follows (here we also used (\ref{fpio1}-\ref {fpio2}):

1322: \begin{eqnarray}

1323: {\bf E} \!= \!\lambda \left( \!\!

1324: \begin{array}{ll}

1325: \! -1 +\frac{\alpha \lambda}{c v_2}\varepsilon_2^* & \;\;

1326: -\frac{\alpha \lambda}{c v_2}\varepsilon_1^* \\

1327: & \\

1328: \! -\frac {\alpha \lambda}{c v_1} \varepsilon_2^* & \!\!

1329: -1 + \frac{\alpha \lambda}{c v_1}\varepsilon_1^*

1330: \end{array}\!\!

1331: \right) \nonumber

1332: \end{eqnarray}

1333: The eigenvalues

1334: of ${\bf E}$ are easily calculated:

1335: 

1336: \begin{equation}

1337: \Lambda_+=-\lambda (1-\alpha),\quad \Lambda_-=-\lambda . \label{coex}

1338: \end{equation}

1339: %See  the Appendix for a more general treatment.

1340: 

1341: One eigenvalue of the width dynamics is always the negative of the chaotic

1342: advection's positive Lyapunov exponent. As long as the parameter $\alpha$ is

1343: less than unity, the other eigenvalue is also negative.

1344: 

1345: We find that for

1346: $0< \alpha < 1$ \emph{coexistence is stable}, for $\alpha > 1$ it 

1347: becomes unstable.  

1348: 

1349: The case $\alpha=1$ is special.

1350: It follows from (\ref {fp2}) that

1351: for $\alpha=1$, and $\omega \neq v_1/v_2$, the

1352: non-coexsitence point is the only fixed point, and it is stable.

1353: (If $\omega > v_1/ v_2$ then 

1354: ($\varepsilon_1^* > 0, \varepsilon_2^* = 0$) is stable, if $\omega < v_1/ v_2$ then 

1355: ($\varepsilon_1^* = 0, \varepsilon_2^* > 0$) is stable fixed point.)

1356:    Having

1357: $\alpha=1$  with $\omega= v_1/v_2$ implies that all 

1358: points fulfilling (\ref{fp1})

1359: are fixed points of marginal stability.

1360: Thus, stable coexistence

1361: is found in the 

1362: \begin{equation}

1363: 0 < \alpha < 1

1364: \end{equation}

1365: regime.

1366: 

1367: Interestingly, the stability conditions are the same for

1368: discrete-time dynamics, as well.

1369: 

1370: In the next section we analyze coexistence in a simple

1371: chaotic dynamical system, the Baker map,

1372: where we show that  $g(z)$ is given by Eq.~(\ref{pe2})

1373: in this process,

1374: and determine the parameters $\alpha$ and $\omega$ from

1375: numerical experiments.

1376: 

1377: \section*{Numerical results}

1378: 

1379: In this section we present numerical verification of the

1380: new type of population dynamical equation we introduced before.

1381: We have already shown that coexistence in open

1382: flows is possible \citep{Sch00b,Kar00},

1383: so we deal only  with the  \emph{quantitative} verification

1384: of the theoretical results.

1385: 

1386: 

1387: 

1388: For computational simplicity, we use the so-called

1389: \emph{baker map} to model the flow \citep{Tor01}.

1390: This can be considered as a simplified model of

1391: stretching and folding in a chaotic flow observed

1392: periodically after specified time-intervals.

1393: Thus, in this case $\tau$, the time lag between instantaneos

1394: multiplications of the species, is an integer number denoting

1395: the number of snapshots taken of the flow between two

1396: consequtive multiplications.

1397: The baker map,

1398: acting on the unit square,

1399: gives the new location $(x',y')$ of an individual starting at

1400: point $(x,y)$:

1401: \begin{eqnarray}

1402: \begin{array}{ll}

1403: x'=ax+(1-a)\theta(y-1/2), & x\in[0,1],\\

1404: y'=\frac{1}{a} y-(\frac{1}{a}-1)\theta(y-1/2), & y\in[0,1],

1405: \end{array}

1406: \end{eqnarray}

1407: where $a<1/2$ is the parameter of the baker map,

1408: and $\theta(x)$ is the Heaviside step function.

1409: The action of the baker map is shown schematically in

1410: Fig.~\ref{figbaker}.

1411: The area preserving property of this baker map models

1412: the incompressibility of realistic hydrodynamical flows,

1413: while outflow is modelled by neglecting the area hanging over

1414: the edge of the unit square.

1415: 

1416: \begin{figure}[htbp]

1417: \vspace*{-0.5cm}

1418: \hspace*{-0.5cm}

1419: \epsfxsize=3.4 in \epsfbox{f2.eps}

1420: \vspace*{0.2cm}

1421: \caption{Two consecutive steps of the baker map and two replications

1422: ($\tau=1$) for the single species model.

1423: The bands of width $\sigma$ become occupied by $B$ in each replication.

1424: The material hanging over the unit square is discarded.

1425: }\label{figbaker}

1426: \end{figure}

1427: 

1428: Starting with any initial conditions, after a few steps of

1429: iterations both species will be distributed along

1430: narrow filaments parallel  to the $y$ axis.

1431: 

1432: \begin{figure}[htbp]

1433: \vspace*{-0.5cm}

1434: \protect\hspace*{-0.5cm}

1435: \epsfxsize=3.2 in

1436: \epsfbox{fig3.eps}

1437: \protect\vspace*{0.2cm}

1438: \caption{Reaching the equilibrium states with the coexistence of

1439: two species is shown. Initially, two patches of species were

1440: placed, one patch of $B_1$ in $x\in [0; 0.1]$, $y\in [0;1]$,

1441: and another

1442: patch of

1443: $B_2$ at $x\in [0.1; 1]$, $y\in [0;1]$.

1444: The parameter values are $a=0.4$ %, $b=2$

1445: for the baker map, and $\sigma_1=0.003$

1446: $\sigma_2=0.001$ for the competing species.

1447: The areas covered by the species is shown right after

1448: the multiplications

1449: taking place.

1450: After an initial transient (time-steps 1--4), we have a rapid

1451: convergence to the fixed point (time-steps 5--18), after that

1452: we have

1453: %some fluctuations around the

1454: an equilibrium setting in (time-steps 18--20).

1455: %, the fluctuations are so small they are not visible on this scale).

1456: }

1457: \label{fig:timeevol}

1458: \end{figure}

1459: 

1460: After $\tau$ baker steps, individuals of species $B_i$

1461: multiply and give birth into a vertical stripe of width

1462: $\sigma_i$ covered by resource $A$,

1463: lying along the borderline of the previously

1464: occupied region of species $B_i$ parallel to the $y$ axis.

1465: In the numerical experiments, we used $\tau=1$, that is,

1466: the species multiplied after each baker-step.

1467: Regions which are invaded by both species after

1468: instantaneous multiplication are divided between them in a ratio

1469: of $\sigma_1/\sigma_2$.

1470: It is expected that (\ref{eq:mapautoA})

1471: describes the time-evolution

1472: of the species, reaching the fixed-point (\ref{fixdisc}) eventually.

1473: Figure~\ref{fig:timeevol} shows in a typical case how the

1474: equilibrium state with coexistence is reached after about

1475: 18 baker steps.

1476: %This implies coexistence according to the theoretical result,

1477: %and was observed also in the numerical experiment.

1478: Similar results were obtained with various other parameter

1479: settings, in accordance with the theoretical results.

1480: %So, we measured the number of individuals after reaching the

1481: %equilibrium, and then calculated the probability

1482: %$p_1^*=g^*(z)$ in equilibrium.

1483: %

1484: We checked the validity of the form

1485: (\ref{fixdisc}) against the

1486: numerical results in steady states.

1487: $N_i^*$ is the fixed-point number of individuals of species

1488: $B_i$. The fixed point values are found to fulfill

1489: (\ref{fixdisc}), see

1490: Fig.~\ref{fig:fixedpoint}.

1491: 

1492: \begin{figure}

1493: \protect\hspace*{-0.5cm}

1494: \epsfxsize=3.2 in

1495: \epsfbox{fig4.eps}

1496: \protect\vspace*{0.2cm}

1497: \caption{The dependence of the probabilities $p_1$

1498: on $N_1/N_2$ in the nontrivial fixed points.

1499: The initial  positions do not influence the fixed point reached.

1500: %(as long as they overlap with the inflow curve).

1501: The curve

1502: $g(z)=z/(z+\sigma_1/\sigma_2)$ is

1503: shown with

1504: solid line for $\sigma_1=0.002$, $\sigma_2=0.001$,

1505: with dashed line for $\sigma_1=0.003$, $\sigma_2=0.001$,

1506: and with dotted line for $\sigma_1=0.004$, $\sigma_2=0.001$.

1507: All the measured fixed point values  fulfill

1508: $g(z^*)=z^*/(z^*+\sigma_1/\sigma_2)$.

1509: The fixed points are marked by crosses ($a=0.25$), black

1510: squares ($a=0.3$),

1511: starts ($a=0.35$), and circles ($a=0.4$ as the baker parameter).

1512: }

1513: \label{fig:fixedpoint}

1514: \end{figure}

1515: 

1516: Next we check

1517: the validity of (\ref{eq:mapautoA}) for the time-evolution

1518: before reaching the convergence. We measure the population numbers

1519: in discrete time $n$ and

1520: use relation (\ref{eqfit})   to extract the form of the

1521: probability distribution $g$.

1522: Figure~\ref{fig:figfit} shows $p_1$ as a function of $N_1/N_2$

1523: for fixed parameter values,

1524: but for various initial conditions.

1525: There is a single function covering the measured points

1526: which can well be fitted by the form

1527: $g_1(z)=z^{\alpha}/(z^{\alpha}+\omega$.

1528: %with $\alpha=...$.

1529: In all cases $\alpha<1$ was measured indicating that the coexistence

1530: fixed point is stable.

1531: Also note that $\omega\approx (\sigma_1/\sigma_2)^{\alpha}$

1532: was found in all experiments, which implies that (\ref{feltetel}) holds.

1533: 

1534: \begin{figure}

1535: \protect\hspace*{-0.5cm}

1536: \epsfxsize=3.2 in

1537: \epsfbox{fig5.eps}

1538: \protect\vspace*{0.2cm}

1539: \caption{The dependence of the probabilities $p_1$

1540: on $N_1/N_2$

1541: with $\sigma_1=0.003$, $\sigma_2=0.001$ values, and with five

1542: different initial conditions.

1543: The parameter of the baker map was $a=0.4$.

1544: The initial positions do not influence the fixed point reached.

1545: %(as long as they overlap with the inflow curve).

1546: Solid line shows the function $g(z)=z^{\alpha}/(z^{\alpha}+\omega)$

1547: with $\alpha=0.818\pm 0.002$ and $\omega=2.312\pm 0.006$.

1548: The parameter of the baker map was $a=0.4$.

1549: %Dashed line shows $g_1(z)=z^{\alpha}/(z^{\alpha}+

1550: %(\sigma_1/\sigma_2)^{\alpha})$ with $\alpha=.....$

1551: }

1552: \label{fig:figfit}

1553: \end{figure}

1554: 

1555: We also measured how $\alpha$ depends on the

1556: parameter of the baker map,

1557: or, on the fractal dimension $D_0$ of the

1558: outflow curve of the chaotic

1559: saddle.

1560: We found that $\alpha=0.79\ln a+1.54$, see Fig.~\ref{fig:fit}.

1561: Using the fact that $D_0=\ln 2/ln(1/a)$, we obtain

1562: $\alpha=1.54-0.55/D_0$.

1563: 

1564: \begin{figure}

1565: \protect\hspace*{-0.5cm}

1566: \epsfxsize=3.2 in

1567: \epsfbox{fit.eps}

1568: \protect\vspace*{0.2cm}

1569: \caption{

1570: Dependence of $\alpha$ on the parameter $a$ of the baker map.

1571: Various points are measured values for multiple $\sigma_1/\sigma_2$

1572: ratios. The dashed line is the function $\alpha(a)=0.79\ln a+1.54$.}

1573: \label{fig:fit}

1574: \end{figure}

1575: 

1576: \section* {Discussion}

1577: 

1578: We derived a novel type of coupled population dynamic equations for

1579: two populations competing on a fractal set provided by open chaotic flows.

1580: The equation for the number of species in a given fixed range

1581: of the flow can be written in the general scheme (cf.\ \ref{diffeq1})

1582: \begin{equation}

1583: \frac{dN_1}{dt} = \alpha_1\left(\frac{N_1}{N}\right) N_1^{-\beta} - 

1584: \kappa N_1,

1585: \end{equation}

1586: \begin{equation}

1587: \frac{dN_2}{dt} = \alpha_2\left(\frac{N_2}{N}\right) N_2^{-\beta} - 

1588: \kappa N_2.

1589: \end{equation}

1590: The coefficients $\alpha_i$ of the replication terms depend on the relative

1591: densitites (denoted by $N_i/N \equiv c_i$) only.

1592: Their explicit form follows from  (\ref{diffeq1}). For example,

1593: \begin{equation}

1594:  \alpha_1\left(\frac{N_1}{N}\right) =

1595:  q \left( \frac{N_1}{N} \right)^{\beta} \left[(1-D_0) v \frac{N_1}{N}- v_1 p_1

1596: \right].

1597: \end{equation} 

1598: The structure of these equations is similar to that of (\ref{trad0}),

1599: (\ref {trad}) or

1600: (\ref{trad2}, \ref {trad3}). The time derivative is the sum of a gain term

1601:  and a

1602: loss term, but now the gain term contains a nontrivial

1603:  negative power of the

1604: population number and is coupled

1605:  to the other population in a nonlinear way.

1606: These equations

1607:  describe  the population dynamics in an imperfectly mixed

1608: enviroment of dimension $1<D_0<2$. The fractality $D_0$

1609: of the mixing region (in our case of the outflow curve)

1610: appears in the power

1611: $\beta=(2-D_0)/(D_0-1)$.

1612: In this set of equations a phenomenological

1613: function ($p_1$) is also present characterizing the probability that a given

1614: population is on the surface of the fractal support

1615: with free access to the single

1616: available resource. Based on general arguments and a simple model,

1617: this function turned out to be a normalized power law distribution

1618: of the type of (\ref{pe2}).

1619: %It is worth mentioning that t

1620: This form

1621: expresses a kind of ``advantage of rarity'' principle:

1622: for exponent $0<\alpha < 1$ the derivative is infinite in the origin,

1623: a very small increase in the size of the weaker population

1624: leads to a drastical increase of the probability for being on

1625: the free surface and hence to grow.

1626: On the contrary, for $\alpha>1$,

1627: %not even

1628: only a relatively large

1629: population size has considerable growing probability, in this case

1630: the weaker population dies out. It is worth mentioning that in

1631: the gedankenexperiment of Section ``Coexistence of competing species'', 

1632: we did not see this effect since

1633: we had an initial stage at which the weaker population

1634: does not yet influence the stronger one. The probability of being

1635: on the surface was assumed to be constant. It is the interaction

1636: between the two populations which leads to the power law distribution.

1637: Its exponent is determined by the flow and the biological

1638: process.

1639: With exponents larger than unity this form does not allow for

1640: coexistence.

1641: The presented mathematical forms and the conditions

1642: for coexistence remain valid if $m >2$ species live in

1643: open chaotic flow,

1644: a numerical evidence for which 

1645: has been reported by \citet{Kar00}. It is natural

1646: to expect that the probabilities $p_i$, $i=1,...,m$ appear in the generalized

1647: form of

1648: $p_i=(\omega_i\varepsilon_i^{\alpha})/(\sum_{i=1}^m

1649: \omega_i\varepsilon_i^{\alpha})$, where $\varepsilon_i$ are the partial width

1650: of the species and

1651: $\omega_i$ are phenomenological constants. 

1652: 

1653: Although in the numerical simulation, based on the baker map

1654: as a model flow,

1655: we only found stable coexistence, we also

1656:  carried out simulations where the

1657: biological process was not

1658:  based on parallel stripes filled out homogeneously

1659: with

1660:  individuals, as assumed in our theory.

1661:  In these cellular automaton-like

1662: simulations

1663:  the replication and competition process is carried out

1664:  on a

1665: uniform

1666:  rectangular grid of lattice size

1667: $\epsilon_0$.

1668: This $\epsilon_0$ can be considered as the smallest distance

1669: between the individuals, or the linear size of a single individual

1670: below which there is hard-core exclusion among them.

1671: Individuals of each species can occupy the center of each grid-cell.

1672: When they are advected by the flow into another grid-cell

1673: during the time $\tau$, they are instantaneously placed to the

1674: center of that grid-cell.

1675: During reproduction, they give birth to new individuals in the

1676: surrounding empty grid-cells, whose centers are within a distance

1677: $\sigma_i$.

1678: If more than one species tries to give birth into the same grid-cell,

1679: only one of them will be able to do so according to one of

1680: the following rules:

1681: \emph{rule~I:}~both species

1682: can win this competition in each cell with equal

1683: probability, or

1684: \emph{rule~II:}~both species

1685: can win this competition in each cell with

1686: probability proportional to the number

1687: of individuals of the same species intending to give birth there, or

1688: \emph{rule~III:}~always the better competitor (with higher $\sigma$)

1689: wins.

1690: Our results show that the coexistence depends on which rule

1691: has been applied. In some cases one of the populations was

1692: competed out, but even in such cases the distribution function

1693: was found to be of the shown form, with an exponent

1694: $\alpha >1$, in full harmony with the theory.

1695: It is worth mentioning that with the same rules on the lattice, in

1696: previous simulations \citep{Kar00} for

1697: the more realistic fluid dynamical case

1698: of a flow around an obstacle we always found coexistence.

1699: This indicates that the boundary layer present around the

1700: obstacle enhances the chances of survival.

1701: 

1702: Our theory does not describe the effects of diffusion. Besides

1703: the fact that for individuals of small but macroscopic mass and

1704: size, like eg.  phytoplankton, diffusion is not believed

1705: to be important, it can be shown \citep{Tel00}

1706: that weak diffusion in such models only renormalizes

1707: the replication rates. As a consequence, the cut-off

1708: scale below which fractality cannot be observed is somewhat

1709: increased, but the population dynamical equations remain unchanged.

1710: 

1711: %Other shortcomings of

1712: In this theory the location dependence

1713: of the death and replication rates

1714: is not taken into account. Such effects can be studied in

1715: numerical simulations \citep{Kar00,San01} 

1716: and are not expected to change the essence of our findings.

1717: 

1718: In conclusion, we have shown that a particle-like (microscopic) model

1719: of individuals competing for a single resource around a fractal

1720: outflow curve of a chaotic flow leads, on the level of the total

1721: number of individuals, to dynamical equations with unusual

1722: singular terms. These describe enhanced competition due to

1723: inhomogeneous mixing and a kind of advantage of rarity property.

1724: The appearance of unusual population

1725: dynamical equations can be expected in general

1726: in all cases where

1727: the individual dynamics is not taking place on full

1728: compact regions of the space but are

1729: restricted to  fractal subsets of it.

1730: 

1731: 

1732: \section*{Acknowledgements}

1733: 

1734: Z.T. was supported by the Department of Energy 

1735: under contract W-7405-ENG-36.

1736: Support from the Hungarian Science Foundation (OTKA T032423, F029637)

1737: the US-Hungarian Joint Fund (Project No. 501) and the

1738: MTA/OTKA/NSF Fund (Project No. Int. 526) is acknowledged.

1739: G.K.  and I. Sch were supported by the Bolyai grant.

1740: The authors gratefully acknowledge discussions with C. Grebogi,

1741: E. Ben-Naim, T. Cz\'ar\'an and E. Szathm\'ary.

1742: 

1743: 

1744: 

1745: \begin{references}

1746: 

1747: \bibitem[Aref(1994)]{Are94}

1748: Aref, H. (Ed.) 1994:

1749: Special issue on Chaotic Advection.

1750: \emph{Chaos, Solitons, Fractals} \textbf{4}, No.~6.

1751: 

1752: \bibitem[Ar{\'\i}stegui \etal(1997)]{Ari97}

1753: Ar{\'\i}stegui, J.; Tett, P.; Hern\'andez-Guerra, A.; Basterretxea,

1754: G.; Montero, M.F.; Wild, K.; Sangr\'a, P.; Hern\'andez-Le\'on, S.;

1755: Cant\'on, M.; Garc\'\i{}a-Braun, J.A.; Pacheco, M. \& Barton, E.D. 1997:

1756: The influence of island-generated eddies on chlorophyll distribution:

1757: a study of mesoscale variation around Gran Canaria.

1758: \emph{Deep-See Research}

1759: \textbf{44}, 71--96.

1760: 

1761: \bibitem[Bartha \etal(1997)]{Bar97}

1762: Bartha, S.; Cz\'ar\'an, T. \& Scheuring, I. 1997:

1763: Spatiotemporal scales of non-equilibrium community dynamics:

1764: a methodological challenge.

1765: \emph{N. Z. J. of Ecol.}\ \textbf{21}, 199--206.

1766: 

1767: \bibitem[Boerlijst \&  Hogeweg(1991)]{Boe91}

1768: Boerlijst, M.C. \&  Hogeweg, P. 1991:

1769: \emph{Physica D} \textbf{48}, 17--28.

1770: 

1771: \bibitem[Chesson(2000)]{Che00}

1772: Chesson, P. 2000:

1773: Mechanism of maintenance of species diversity.

1774: \emph{Annu.\ Rev.\ Ecol.\ Syst.}\ \textbf{31}, 343--366.

1775: 

1776: \bibitem[Connell(1978)]{Con78}

1777: Connell, J.H. 1978:

1778: Diversity in tropical rain forests and coral reefs.

1779: \emph{Science} \textbf{199}, 1302--1310.

1780: 

1781: \bibitem[Cz\'ar\'an \& Szathm\'ary(2000)]{Cza00}

1782: Cz\'ar\'an, T. \& Szathm\'ary, E. 2000:

1783: In:

1784: Dieckmann, U.; Law, R. \& Metz, J.A.J. (Eds):

1785: \emph{The

1786: Geometry of Ecological Interactions: Simplifying Spatial Complexity}.

1787: Cambridge Univ.\ Press,

1788: Cambridge.

1789: 

1790: \bibitem[Eigen(1971)]{Eig71}

1791: Eigen, M. 1971:

1792: \emph{Naturwissenschaften} \textbf{58}, 465--523.

1793: 

1794: \bibitem[Eigen \&  Schuster(1979)]{Eig79}

1795: Eigen, M. \&  Schuster, P. 1979:

1796: \emph{The Hypercycle}.

1797: Springer, Berlin.

1798: 

1799: \bibitem[Gaedeke \& Sommer(1986)]{Gae86}

1800: Gaedeke, U. \& Sommer, U. 1986:

1801: The influence of the frequency of periodic disturbances on the maintenance

1802: of phytoplankton diversity.

1803: \emph{Oecologia} \textbf{71}, 98--102.

1804: 

1805: \bibitem[Gause \& Witt(1935)]{Gau35}

1806: Gause, G.F. \& Witt, A.A. 1935:

1807: Behavior of mixed populations and the problem of natural selections.

1808: \emph{Am.\ Nat.}\ \textbf{69}, 596--609.

1809: 

1810: \bibitem[Gurney \& Nisbet(1998)]{Gur98}

1811: Gurney, W.S.C. \& Nisbet, R.M. 1998:

1812: \emph{Ecological Dynamics}. 

1813: Oxford Univ.\ Press, Oxford.

1814: 

1815: \bibitem[Hardin(1960)]{Har60}

1816: Hardin, G. 1960: The competitive exclusion principle.

1817: \emph{Science} \textbf{131}, 1292--1298.

1818: 

1819: \bibitem[Holm(1992)]{Hol92}

1820: Holm, N.G. 1992: Marine hydrothermal systems and the origin of life.

1821: \emph{Origins Life Evol.\ Biosphere}

1822: \textbf{22}, 5.

1823: 

1824: \bibitem[Hsu \etal(1988)]{Hsu88}

1825: Hsu, G.H.; Ott, E \& Grebogi, C. 1988:

1826: \emph{Phys.\ Lett.\ A} \textbf{127}, 199.

1827: 

1828: \bibitem[Huston(1979)]{Hus79}

1829: Huston, M.A. 1979:

1830: General hypothesis of species diversity.

1831: \emph{Am.\ Nat.}\ \textbf{113}  81--101.

1832: 

1833: \bibitem[Hutchinson(1961)]{Hut61}

1834: Hutchinson, G.E. 1961:

1835: The paradox of the plankton

1836: \emph{Am.\ Nat.}\ \textbf{95}, 137--147.

1837: 

1838: \bibitem[Jung \& Ziemniak(1992)]{Jun92} 

1839: Jung, C. \& Ziemniak, E. 1992:

1840: Hamiltonian scattering chaos in a hydrodynamical system.

1841: \emph{J. Phys.\ A} \textbf{25}, 3929--3943.

1842: 

1843: 

1844: %\bibitem[Kadtke \& Novikov(1993)]{Kad93}

1845: %Kadtke, J.B. \& Novikov, E.A. 1993:

1846: %\emph{Chaos} \textbf{3}, 543.

1847: %\textbf{NEVER REFERENCED!}

1848: 

1849: \bibitem[Kantz \& Grassberger(1985)]{Kan85}

1850: Kantz, H. \& Grassberger, P. 1985:

1851: Repellers, semi-attractors, and long-lived chaotic transients.

1852: \emph{Physica D} \textbf{17}, 75.

1853: 

1854: \bibitem[K\'arolyi \& T\'el(1997)]{Kar97} 

1855: K\'arolyi, G. \& T\'el, T. 1997:

1856: Chaotic tracer scattering and fractal basin boundaries in a

1857: blinking vortex-sink system.

1858: \emph{Physics Reports} \textbf{290}, 125--147.

1859: 

1860: 

1861: \bibitem[K\'arolyi \etal(1999)]{Kar99}

1862: K\'arolyi, G.; P\'entek, \'A.; Toroczkai, Z.; T\'el, T. \&

1863: Grebogi, C. 1999:

1864: Chemical or biological activity in open chaotic flows.

1865: \emph{Phys.\ Rev.\ E} \textbf{59} 5468--5481.

1866: 

1867: \bibitem[K\'arolyi \etal(2000)]{Kar00}

1868: K\'arolyi, Gy.; P\'entek, \'A.;

1869: Scheuring, I.; T\'el, T. \& Toroczkai, Z. 2000: 

1870: Open chaotic flow: the physics

1871: of species coexistence. 

1872: \emph{Proc.\ Natl.\ Acad.\ Sci.\ USA} 

1873: \textbf{97} 13661--13665.

1874: 

1875: \bibitem[Lamb(1932)]{Lam32}

1876: Lamb, H. 1932:

1877: \emph{Hydrodynamics}.

1878: Cambridge University Press, Cambridge.

1879: 

1880: \bibitem[Maynard Smith(1983)]{May83}

1881: Maynard Smith, J. 1983:   Models of evolution.

1882: \emph{Proc.\ Roy.\ Soc.\ London B}

1883: \textbf{219}, 315--325.

1884: 

1885: \bibitem[Maynard Smith \& Szathm\'ary (1995)]{May95}

1886: Maynard Smith, J. \& Szathm\'ary, E. 1995:

1887: \emph{The major transitions in evolution}. 

1888: Freeman, Spektrum, Oxford.

1889: 

1890: %\bibitem[Ottino(1989)]{Ott89}

1891: %Ottino, J.M. 1989:

1892: %\emph{The kinematics of mixing: stretching, chaos and transport}.

1893: %Cambridge University Press, Cambridge.

1894: %\textbf{NEVER REFERENCED!!!}

1895: 

1896: \bibitem[P\'entek \etal(1996)]{Pen96} 

1897: P\'entek, \'A.; T\'el, T. \& Toroczkai, Z. 1996:

1898: Transient chaotic mixing in open hydrodynamical flows.

1899: \emph{Int.\ J. Chaos and Bifurcations} \textbf{6}, 2619--2625.

1900: 

1901: \bibitem[Poole \etal(1999)]{Poo99}

1902: Poole, A.; Jeffares, D. \& Penny, D. 1999:

1903: Early evolution: prokaryotes, the

1904: new kids on the block. 

1905: \emph{BioEssays} \textbf{21}, 880--889.

1906: 

1907: \bibitem[Reynolds(1993)]{Rey93}

1908: Reynolds, C.S. 1993: 

1909: Scales of disturbance and their role in plankton ecology.

1910: \emph{Hydrobiologia} \textbf{249}, 157--171.

1911: 

1912: \bibitem[Reynolds(1998)]{Rey98}

1913: Reynolds, C.S. 1998:

1914: The state of freshwater ecology.

1915: \emph{Freshwater Biology} \textbf{39}, 741--753.

1916: 

1917: \bibitem[Santoboni \etal(2001)]{San01}

1918: Santoboni, G.; Nishikawa, T.; Toroczkai, Z. \& Grebogi, C. 2001:

1919: Autocatalytic reactions of active particles with phase distribution. 

1920: \emph{Preprint}.

1921: 

1922: \bibitem[Scheuring(2000)]{Sch00a}

1923: Scheuring, I. 2000:

1924: Avoiding Catch-22 of early evolution by stepwise

1925: increase in copying fidelity. 

1926: \emph{Selection} \textbf{1--3}, 135--145.

1927: 

1928: \bibitem[Scheuring \etal(2000)]{Sch00b}

1929: Scheuring, I.; K\'arolyi, G.; P\'entek, \'A.; Toroczkai, Z. \& T\'el, T.

1930: 2000: 

1931: A Model  for

1932: resolving the plankton paradox: coexistence in open flow. 

1933: \textit{Freshwater Biology}

1934: \textbf{45}, 123--133.

1935: 

1936: \bibitem[Shariff \etal(1991)]{Sha91} 

1937: Shariff, K.; Pulliam, T.H. \&  Ottino, J.M. 1991:

1938: A dynamical systems analysis of kinematics in the time-periodic wake of

1939: a circular cylinder.

1940: \emph{Lect.\ Appl.\ Math.}\ \textbf{28}, 613--646.

1941: 

1942: \bibitem[Sommer \etal(1993)]{Som93}

1943: Sommer, U.; Padis\'ak, J.; Reynolds C.S. \& Juh\'asz-Nagy, P. 1993:

1944: Hutchinson's heritage: the diversity disturbance relationship

1945: in phytoplankton.

1946: \emph{Hydrobiologia} \textbf{249}, 1--8.

1947: 

1948: \bibitem[Sommerer \etal(1996)]{Som96}

1949: Sommerer, J.C.;  Ku, H.-C. \&  Gilreath,  H.E. 1996:

1950: Experimental evidence for chaotic scattering in a fluid wake.

1951: \emph{Phys.\ Rev.\ Lett.}\ \textbf{77}, 5055--5058.	

1952: 

1953: \bibitem[T\'el(1990)]{Tel90}

1954: T\'el, T. 1990: 

1955: In:

1956: Bai-Lin, H. (Ed.):

1957: \emph{Directions in chaos}, Vol.~3., pp.~149--211.

1958: World Scientific, Singapore.

1959: 

1960: \bibitem[T\'el \etal(2000)]{Tel00} 

1961: T\'el, T.; K\'arolyi, G.; P\'entek, \'A.; Scheuring, I.;  Toroczkai, Z.;

1962: Grebogi, C. \& Kadtke, J. 2000:

1963: Chaotic advection, diffusion, and reactions in

1964: open flows. 

1965: \emph{Chaos} \textbf{10}, 89--98.

1966: 

1967: \bibitem[Tilman \& Pacala(1993)]{Til93}

1968: Tilman, D. \& Pacala, S. 1993:

1969: In:

1970: Ricklefs, R.E. \& Schluter, D.:

1971: \emph{Species diversity in ecological communities:

1972: The maintenance of species richness in plant communities}.

1973: pp.~13--25, 

1974: Univ.\ of Chicago Press.

1975: 

1976: \bibitem[Toroczkai \etal(1998)]{Tor98}

1977: Toroczkai, Z.; K\'arolyi, G.; P\'entek, \'A.; T\'el, T.

1978: \& Grebogi, C. 1998:

1979: Advection of active particles in open chaotic flows.

1980: \emph{Phys.\ Rev.\ Lett.}\ \textbf{80}, 500--503.

1981: 

1982: \bibitem[Toroczkai \etal(1997)]{Tor97}

1983: Toroczkai, Z.; K\'arolyi, G.; P\'entek, \'A.; T\'el, T.;

1984: Grebogi, C. \& Yorke, J.A. 1997:

1985: Wada dye boundaries in open

1986: hydrodynamical flows.

1987: \emph{Physica A} \textbf{239}, 235--243.

1988: 

1989: \bibitem[Toroczkai \etal(2001)]{Tor01}

1990: Toroczkai, Z.; K\'arolyi, G.; P\'entek, \'A. \& T\'el, T. 2001:

1991: Autocatalytic reactions in systems with

1992: hyperbolic mixing: Exact results for the

1993: active baker map.

1994: \emph{J. Phys.\ A}, \textbf{34}, 5215 -- 5235.

1995: 

1996: \bibitem[W\"achtersh\"auser(1994)]{Wac94}

1997: W\"achtersh\"auser, G. 1994: Life in a ligand sphere.

1998: \emph{Proc.\ Natl.\ Acad.\ Sci.\ USA}

1999: \textbf{91}, 4283--4287.

2000: 

2001: \bibitem[Wilson(1990)]{Wil90} 

2002: Wilson, J.B. 1990:

2003: Mechanisms of species coexistence: twelve explanations for Hutchinson's

2004: `paradox of the plankton': evidence from New Zealand plant communities.

2005: \emph{N. Z. J. of Ecol.}\ \textbf{43}, 17--42.

2006: \end{references}

2007: 

2008: \end{multicols}

2009: 

2010: 

2011: 

2012: \end{document}

2013: 

2014: