1: 

2: %\documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}

3: \documentclass[showpacs,preprintnumbers,amsmath,amssymb]{revtex4}

4: \usepackage{graphicx}

5: 

6: \begin{document}

7: 

8: \title {Angular velocity variations and stability of spatially explicit prey-predator systems}

9: \author{Refael Abta  and Nadav M. Shnerb}

10: 

11: \affiliation{   Department of Physics, Bar-Ilan University,

12: Ramat-Gan 52900 Israel}

13: 

14: 

15: 

16: 

17: \begin{abstract}

18: The linear instability of Lotka-Volterra orbits in the homogenous

19: manifold of a  two-patch system is analyzed. The origin of these

20: orbits instability in the absence of prey migration is revealed to

21: be the dependence of the angular velocity on the azimuthal angle; in

22: particular, the system desynchronizes at the exit from the slow part

23: of the trajectory. Using this insight, an analogous model of  a two

24: coupled oscillator is presented and shown to yield the same type of

25: linear instability. This  unables one to incorporate the linear

26: instability within a recently presented general framework that

27: allows for comparison of all known stabilization mechanisms and for

28: simple classification of observed  oscillations.

29: \end{abstract}

30: 

31: \pacs{05.45.-a, 05.45.Xt, 87.23.Cc, 02.50.Ey}

32: 

33: \maketitle

34: 

35: Population oscillations in prey predator systems, as predicted by

36: the Lotka-Volterra equations \cite{lotka}, are known to be unstable

37: with respect to additive and multiplicative  noise. This instability

38: must lead to the extinction of one of the interacting species, a

39: fact that has been confirmed in various experiments for well-mixed

40: populations \cite{gause}. The persistence of natural prey-predator

41: and host-parasitoid systems, thus, is commonly attributed to their

42: spatial structure, such that migration between desynchronized

43: patches yields an inward flow toward the coexistence fixed point and

44: is responsible for the sustainability of the oscillations

45: \cite{nic33}. In fact, spatially extended systems tend to support

46: finite amplitude oscillation \cite{kerr}. The stabilization of such

47: oscillations is considered to be a major factor affecting species

48: conservation and ecological balance \cite{levin}.

49: 

50: The main challenge, thus, is to understand the conditions for the

51: appearance of desynchronization in diffusively coupled patches,

52: since diffusion tends to synchronize these patches so after a while

53: the whole system flows to the  well mixed, unstable limit

54: \cite{hoopes}. One of the solutions to that problem was presented by

55: Jansen \cite{jansen,roos,zig}. It turns out that the trajectories

56: far from the fixed point become unstable if the inter-patch

57: migration rate of the predator is much larger than that of the prey.

58: Jansen used Floquet analysis to show that instability. In this

59: paper, we try to elucidate the underlying mechanism that yields

60: Jansen's instability, to generalize it in the framework of the

61: recently presented coupled oscillator model, and to discuss the

62: conditions under which one may observe the stabilizing effect of

63: Jansen's mechanism, like oscillation amplitude that grows under

64: noise until it reaches the  first unstable orbit.

65: 

66: 

67: 

68: The Lotka-Volterra predator-prey system is a paradigmatic model for

69: oscillations in population dynamics \cite{lotka}. It describes the

70: time evolution of  two interacting populations: a prey ($b$)

71: population that grows with a constant birth rate $\sigma$ in the

72: absence of a predator (the energy resources consumed by the prey are

73: assumed to be inexhaustible), while the predator population ($a$)

74: decays (with death rate $\mu$),  in the absence of a prey. Upon

75: encounter, the predator may consume the prey with  a certain

76: probability. Following a consumption event,  the predator population

77: grows and the prey population decreases. For a well-mixed

78: population, the corresponding partial differential equations are,

79: \begin{eqnarray}\label{basic}

80: \frac{\partial a}{\partial t} &=& - \mu a + \lambda_1 a b \\

81: \nonumber \frac{\partial b}{\partial t} &=& \sigma b - \lambda_2 a

82: b,

83: \end{eqnarray}

84: where $\lambda_1$ and $\lambda_2$ are the relative increase

85: (decrease) of the predator (prey) populations due to the interaction

86: between species, correspondingly.

87: 

88: The system admits two unstable fixed points: the absorbing state

89: $a=b=0$ and the state $a=0, \ \ b = \infty$. There is one marginally

90: stable fixed point at $\bar{a}  = \sigma/\lambda_2, \ \ \bar{b} =

91: \mu/\lambda_1$. Local stability analysis yields the eigenvalues $\pm

92:  i \sqrt{\mu \sigma}$ for the stability matrix. Moreover, even

93:  beyond the linear regime there is neither convergence nor

94:  repulsion. Using logarithmic variables $z = ln(a), \  q = ln(b)$

95:  eqs. (\ref{basic}) take the canonical form $\dot{z} = \partial

96:  H/\partial q, \ \ \dot{q} = -\partial

97:  H/\partial z$, where the conserved quantity $H$ (in the $ab$

98:  representation) is,

99:  \begin{equation}\label{H}

100:  H = \lambda_1 b + \lambda_2 a - \mu \ ln(a) - \sigma \  ln(b).

101:  \end{equation}

102: The phase space, thus, is segregated into a collection of nested

103: one-dimensional trajectories, where each one is characterized by a

104: different value of  $H$, as illustrated in Figure 3.  Given a line

105: connecting the fixed point to one of the "walls" (e.g., the dashed

106: line in the phase space portrait, Figure 3), $H$ is a monotonic

107: function on that line, taking its minimum $H_{min}$ at the

108: marginally stable fixed point (center) and diverging on the wall.

109: Without loss of generality, we employ hereon the symmetric

110: parameters $\mu = \sigma = \lambda_1 = \lambda_2 =1$. The

111: corresponding phase space, along with the dependence  of $H$ on  the

112: distance from the center and a plot of the oscillation period vs.

113: $H$, are represented in Figure 3).

114: 

115: 

116: 

117: \begin{figure}

118:   % Requires \usepackage{graphicx}

119:   \includegraphics[width=8cm]{fig1b.eps}

120:   \caption{The Lotka-Volterra phase space (left panel) admits a

121:   marginally stable fixed point surrounded by close trajectories (three of these

122:   are

123:   plotted). Each trajectory corresponds to single $H$ defined in Eq.

124:   (\ref{H}), where $H$ increases monotonically  along the (dashed) line

125:   connecting the center with the $a=0$ wall, as shown in the lower

126:   right panel. In the upper right  panel, the period of a cycle $T$ is

127:   plotted against $H$, and is shown to increase almost linearly from

128:   its initial value $T = 2 \pi / \sqrt{\mu \sigma}$ close to the

129:   center.}

130:   \label{fig3}

131: \end{figure}

132: 

133: 

134: 

135: Given the integrability of that system, the effect of noise is quite

136: trivial: if $a$ and $b$ randomly fluctuate in time (e.g., by adding

137: or subtracting small amounts of population during each time step),

138: the system wanders between trajectories, thus performing some sort

139: of random walk in  $H$  with "repelling boundary conditions" at

140: $H_{min}$ and "absorbing boundary conditions" on the walls (as

141: negative densities are meaningless, the "death" of the system is

142: declared when the trajectory hits the zero population state for one

143: of the species). This result was emphasized by  Gillespie

144: \cite{Gillespie} for the important case where intrinsic stochastic

145: fluctuations are induced by the discrete character of the reactants.

146: In that case, the noise is multiplicative (proportional to the

147: number of particles), and the system flows away from the center and

148: eventually hits one of the absorbing states at $0,0$ or $0,\infty$.

149: The corresponding situation for a single patch Lotka-Volterra system

150: with additive noise is demonstrated in Figure 4, where the survival

151: probability $Q(t)$ (the probability that a trajectory does not hit

152: the absorbing walls within  time $t$) is shown for different noise

153: amplitudes.

154: 

155: 

156: \begin{figure}

157:   % Requires \usepackage{graphicx}

158:   \includegraphics[width=7cm]{fig2bw.eps}

159:  \caption{The survival probability $Q(t)$ is plotted versus time for

160: a single-patch, noisy LV

161:   system. Eqs. (\ref{basic}) (with the symmetric parameters) were integrated numerically (Euler integration

162:   with time step $0.001$), where the initial conditions are at the fixed point $a=b=1$. At each time step, a small random number

163:   $\eta(t) \Delta t$ was added to each population density, where $\eta(t) \in [-\Delta,\Delta]$. A typical phase space trajectory, for

164:   $\Delta = 0.5$, is shown in the inset. The system "dies" when the trajectory hits the walls $a=0$ or $b=0$. Using 300 different noise histories, the

165:   survival probability is shown here for $\Delta = 0.5$ (full line), $\Delta = 0.3$ (dotted line) and $\Delta = 0.25$ (dashed line).

166:   Clearly, the survival probability decays exponentially at long

167:   times, $Q(t) \sim exp(-t/\tau)$. As expected for a random walk

168:   with absorbing boundary conditions, $1/ \tau$ scales with

169:   $\Delta^2$.}

170:   \label{fig2}

171: \end{figure}

172: 

173: The Lotka-Volterra system on spatial domains has been investigated,

174: usually in a form of diffusively coupled patches, during the last

175: decades.  Any patch is assumed to be well mixed, and the flow of the

176: reactants from one patch to its neighbors is governed by the density

177: gradient. Clearly, any system of that type, independent of its

178: spatial topology (either regular lattice of some dimensionality or

179: some sort of network without isolated nodes) admits an infinite

180: number of solutions that correspond to \emph{synchronous}

181: oscillations of the whole system along one of the $H$ trajectories,

182: where the diffusion has no role as there are no population

183: gradients. The simplest example is  the two-patch system, described

184: by:

185: 

186: \begin{eqnarray}\label{two}

187: \frac{\partial a_1}{\partial t} &=& - \mu a_1 + \lambda_1  a_1 b_1 +

188: D_a (a_2-a_1) \\  \nonumber \frac{\partial a_2}{\partial t} &=& -

189: \mu a_2 + \lambda_1   a_2 b_2 + D_a (a_1-a_2)\\  \nonumber

190: \frac{\partial b_1}{\partial t} &=& \sigma b_1 - \lambda_2 a_1  b_1

191: + D_b (b_2 - b_1) \\ \nonumber \frac{\partial b_2}{\partial t} &=&

192: \sigma b_2 - \lambda_2  a_2 b_2 + D_b (b_1 - b_2).

193: \end{eqnarray}

194: 

195: Here the invariant manifold is the two dimensional subspace

196: $a_1=a_2, \ \ b_1 = b_2$. The diffusion, of course, suppresses

197: fluctuations and stabilizes the invariant manifold; one may expect,

198: thus, that the single-patch dynamics also capture the main features

199: of the extended system, and that the system behaves like a random

200: walker in the invariant manifold (with a rescaled noise) and hits

201: the absorbing walls after some characteristic  time $\tau$, where

202: $\tau$ scales linearly with the noise strength $\Delta^2$.

203: 

204: As a first hint for a stabilizing mechanism, let us consider the

205: total $H$,

206: \begin{equation}\label{Htotal}\

207: H_{T} \equiv H_1 + H_2  = a_1 + a_2 + b_1 + b_2 - ln(a_1 a_2 b_1

208: b_2).

209: \end{equation}

210: With the deterministic dynamics (\ref{two}), $H_{T}$ is a

211: \emph{monotonously decreasing} quantity in the non-negative

212: population regime:

213: \begin{equation} \label{Hdynamics}

214: \frac{d H_T }{dt} = - D_a \left( \frac{(a_1 - a_2)^2}{a_1 a_2}

215: \right) - D_b \left( \frac{(b_1 - b_2)^2}{b_1 b_2} \right) < 0.

216: \end{equation}

217: Accordingly, if an orbit on the invariant manifold becomes unstable,

218: the flow will be inward and the population oscillations stabilizes.

219: 

220: While if $D_a = D_b$ the stability properties of an orbit on the

221: invariant manifold are identical to the stability properties of the

222: corresponding single-patch orbit \cite{Abarbanel}, if the diffusion

223: of both species is different, there is a possibility for unstable

224: orbits on the homogenous plane. This option was materialized by

225: Jansen \cite{jansen}, who considered the set of Equations \ref{two}

226: in the limit $D_b = 0$, so that only the predator undergos

227: diffusion. With the transformation:

228: \begin{eqnarray}\label{transform}

229: A=\frac{a_1 + a_2}{2} \qquad B=\frac{b_1 + b_2}{2}\\\nonumber

230: \delta=\frac{a_1 - a_2}{2} \qquad \theta=\frac{b_1 - b_2}{2},

231: \end{eqnarray}

232: one recognizes the homogenous $AB$ manifold and that  the $\delta$

233: $\theta$ coordinates measure the deviation from that manifold (the

234: heterogeneity of the population). In these coordinates the system

235: satisfies,

236: \begin{eqnarray}\label{lv2trans}

237: \frac{\partial A}{\partial t} &=& - \mu A + \lambda_1  A B + \lambda_1  \delta \theta\\

238: \nonumber \frac{\partial B}{\partial t} &=& \sigma B - \lambda_2

239: A  B + \lambda_2  \delta \theta \\

240: \nonumber \frac{\partial \delta}{\partial t} &=& - \mu \delta +

241: \lambda_1 A \theta + \lambda_1 B \delta -2 D_a \delta \\ \nonumber

242: \frac{\partial \theta}{\partial t} &=& \sigma B - \lambda_2  A

243: \theta - \lambda_2  B \delta -2 D_b \theta.

244: \end{eqnarray}

245: Linearizing around the homogenous manifold, The $AB$ dynamic is

246: equivalent to that of a  single patch,

247: \begin{eqnarray}\label{lv2translin}

248: \dot {A} &=& - \mu A + \lambda_1  A B\\

249: \nonumber \dot{B} &=& \sigma B - \lambda_2 A  B  \\ \nonumber

250: \end{eqnarray}

251: and the $\delta-\theta$ linearized dynamic is

252: \begin{eqnarray}\label{matrix}

253:  \frac{\partial}{\partial t} \left(

254:   \begin{array}{cc}

255:     \delta \\ \theta \\

256:   \end{array}

257: \right) =\left(

258:   \begin{array}{cc}

259:     -\mu+\lambda_1 B -2D_a & \lambda_1 A \\

260:     -\lambda_2 B & \sigma - \lambda_2 A -2D_b \\

261:   \end{array}

262: \right) \left(

263:   \begin{array}{cc}

264:     \delta \\ \theta \\

265:   \end{array}

266: \right).

267: \end{eqnarray}

268: One may thus calculate the eigenvalues of the Floquet operator for

269: one period along an orbit of the homogenous manifold

270: (\ref{lv2translin}). The resulting stability diagram, first obtained

271: by \cite{jansen}, is shown in Figure \ref{fig1}.

272: \begin{figure}

273:    %Requires \usepackage{graphicx}

274:   \includegraphics [width=8cm] {jansen_map.eps}\\

275:   \caption{Stability diagram for phase space orbits (ordered by their conserved quantity $H$)

276:   for different values of predator diffusion $D_a$, where Db=0.}

277:   \label{fig1}

278: \end{figure}

279: 

280: Our first mission is to intuitively explain  Jansen's results.

281: First, we notice that the angular velocity along a single

282: Lotka-Volterra orbit is not fixed. Figure \ref{ang} emphasizes the

283: angular velocity gradient along an orbit. While the inner

284: trajectories (close to the fixed point) are almost harmonic with

285: constant angular velocity, the eccentric large $H$ orbits admit

286: large variations. In particular, the motion in the dilute population

287: region [close to the unstable empty fixed point ($0,0$)] is very

288: slow, while in the dense population region the angular velocity is

289: large.

290: \begin{figure}

291:    %Requires \usepackage{graphicx}

292:   \includegraphics [width=8cm] {cart.eps}\\

293:   \caption{The angular velocity along some orbits of the  Lotka-Volterra

294:   dynamic. Fast regions marked in red, slow regions are blue.

295:    Clearly, the dynamics is slowest when the populations of both species are diluted, and fastest along the

296:    dense region in the upper-right "shoulder." Note that the velocity gradient along an orbit increases with $H$. }

297:   \label{ang}

298: \end{figure}

299: 

300: Following the caricature of an orbit in  Fig.  \ref{z1}, we can

301: explain the source of the instability. For a two-patch system, if

302: one patch is at point A along the orbit and the other patch at B,

303: since the A patch is moving faster along the line it will get closer

304: and closer to B during their flow toward the slow region. The

305: diffusion of the prey plays no role along this branch, since the

306: prey density is almost equal, while the predator diffusion may only

307: strengthen that effect. Thus, the two patches must (almost)

308: synchronize along this branch.

309: 

310: \begin{figure}

311:    %Requires \usepackage{graphicx}

312:   \includegraphics [width=8cm] {z1.eps}\\

313:   \caption{An orbit of the LV dynamics and its fast and slow regions. As explained in the text, with no prey migration  the

314:   two patches desynchronize in the $CD$ region, thus predator diffusion causes a flow toward the fixed point and stabilizes the oscillations.   }

315:   \label{z1}

316: \end{figure}

317: 

318: 

319: \begin{figure}

320:    %Requires \usepackage{graphicx}

321:   \includegraphics [width=8cm] {zzj.eps}\\

322:   \caption{Phase portrait of the inward  flow in the homogenous manifold (average prey density vs. average predator density)

323:   for two-patch LV system with no prey diffusion and $D_{predators} = 1$. Clearly, the inward flow happens in the $C' - D'$ region of

324:   Figure \ref{z1}, where the desynchronization along the $CD$ branch interferes with the predator diffusion. There is almost no inward motion along the rest of the

325:   orbit.    }

326:   \label{zzj}

327: \end{figure}

328: 

329: The situation is completely different in the exit from the slow

330: region. The patch at D moves much faster than that at C, so they

331: will \emph{desynchronize}. As the predator density along this branch

332: is almost constant, the only factor that may avoid desynchronization

333: is the prey migration. In the absence of prey migration, the two

334: patches reach the points C' and D', where the predator migration

335: produces an inward flow. Figure \ref{zzj} is now well understood:

336: the inward flow happens when the desynchronization interferes with

337: the predator diffusion, as explained.

338: 

339: Let us consider, now, Jansen's instability in the framework of the

340: coupled nonlinear oscillator toy model, recently presented \cite{us}

341: as a generic tool for the investigation of oscillation stability in

342: diffusively coupled metapopulations. With the intuition gathered

343: from the above example, we want to consider diffusively coupled

344: orbits where the angular velocity depends on the radial angle and

345: the diffusing species density is changing along the slowing branch.

346: The following equations,

347: \begin{eqnarray}\label{co2}

348:  \frac{\partial x_1}{\partial t} &=&  \omega (\theta_1) y_1 + D_x (x_2-x_1)\\

349:   \nonumber \frac{\partial x_2}{\partial t} &=&  \omega (\theta_2) y_2 + D_x (x_1-x_2) \\

350: \nonumber \frac{\partial y_1}{\partial t} &=&  -\omega (\theta_1) x_1 + D_y(y_2-y_1) \\

351:  \nonumber \frac{\partial y_2}{\partial t} &=&  - \omega (\theta_2) x_2 + D_y (y_1

352:  -y_2),

353:  \end{eqnarray}

354: will satisfy these conditions for $D_x = D, \  D_y = 0$ and

355: \begin{eqnarray}\label{omegadef}

356:  \omega &=& \omega_0 + \omega_1 \cos(\theta-\frac{\pi}{4}).

357:   \end{eqnarray}

358: 

359: Using ($i \in 1,2$)

360: \begin{eqnarray}\label{tran} r_i^2 = x_i^2 + y_i^2 &\qquad&

361:   \theta_i  = arctan (\frac{y_i}{x_i}) \\

362:  \dot{r} = \frac {(x \dot{x} + y \dot{y})}{r} &\qquad&

363:  \dot{\theta} = \frac { (x \dot{y} - y \dot{x})}{r^2}, \nonumber

364: \end{eqnarray}

365: and

366: \begin{eqnarray}\label{transformation}

367:  r = r_2-r_1  &\qquad&

368:   R = r_2+r_1\\

369:  \phi = \theta_2-\theta_1 &\qquad&

370:   \Phi = \theta_2+\theta_1, \nonumber

371: \end{eqnarray}

372: one finds that the flow in the invariant manifold satisfies:

373: \begin{eqnarray}\label{proximity2}

374:   \dot R &=& 0 \\

375:   \nonumber \dot\Phi &=& \omega(\theta_1) + \omega(\theta_2) \\  \nonumber

376:   \end{eqnarray}

377: while the linearized equations for the desynchronization amplitude

378: $r$ and the desynchronization angle $\phi$ satisfy:

379: 

380: 

381: 

382: 

383: \begin{eqnarray}\label{mat}

384: \frac{\partial}{\partial t} \left(

385:   \begin{array}{cc}

386:     \phi \\ r \\

387:   \end{array}

388: \right)= \left(

389:   \begin{array}{cc}

390:     2\omega'(\Phi/2)-2D_x \cos^2(\Phi/2) & \frac{D_x \sin\Phi}{R} \\

391:     \frac{-D_x R\sin(\Phi)}{2} & -2D_x \sin^2(\Phi/2) \\

392:   \end{array}

393: \right)\left(

394:   \begin{array}{cc}

395:     \phi \\ r \\

396:   \end{array}

397: \right)

398:  \end{eqnarray}

399: 

400: 

401: 

402: Using the Floquet operator technique  to analyze the stability of an

403:  orbit by integrating (\ref{mat}) along a close trajectory of

404:  (\ref{proximity2}) one finds the stability map presented in Figure

405:  \ref{toy}, where the parameter $H$ of Figure \ref{fig1} is now

406:  replaced by $\omega_1$, which measures the "eccentricity" of the

407:  angular velocity along a circular path. Here, two unstable regions

408:  appear, for large and small $D_x$.

409: 

410:  It is interesting to point out that in the high $D_x$ region, the unstable  eigenvalue

411:  is positive, while in the small $D_x$ unstable region it takes

412:  negative values. The reason for that is the effect of predator

413:  migration. If the effect of migration is large, in comparison with

414:  the intra-patch dynamics, the two patches should admit (almost) the same

415:  predator density, and the corresponding points in the 2d phase

416:  portrait  should stay on the same

417:  vertical  line (same "x" coordinate). The trajectories of the two

418:  points representing the patches along an orbit are thus similar to

419:  the transport of a vertical rod along a circle, keeping the center

420:  of the rod on the circular trajectory: the two ends switch their

421:  role \emph{twice} (the inner becomes the outer and vice versa) so

422:  the rod returns to its original state after  a full cycle (See Figure \ref{rod}, left panel). In the

423:  low diffusion range, on the other hand, the points remain on a

424:  vertical line only close to the slow portion of the orbit, where

425:  they switch once, but in the fast region they support different

426:  predator populations, so the "rod" completes its cycle in opposite

427:  "phase"  (see Figure \ref{rod}, right panel).

428: 

429: \begin{figure}[h]

430: \begin{center}

431: $\begin{array}{c@{\hspace{1in}}c} \multicolumn{1}{l}{\mbox{\bf (a)}}

432: &

433:     \multicolumn{1}{l}{\mbox{\bf (b)}} \\ [-0.53cm]

434: \includegraphics [width=8cm] {CO_exm1.eps} &

435:     \includegraphics [width=8cm] {CO_exm2.eps}

436: \end{array}$

437: \end{center}

438: \caption{An illustration of the trajectories of two

439: diffusively-coupled patches, with slightly different

440:   initial conditions, projected on the invariant manifold.  In the strong coupling case, (left

441:   panel) the strong predator diffusion forces the two points to be

442:   on the same vertical line (same predator concentration) along the

443:   orbit, hence the phase of the Floquet eigenvalue inverted twice

444:   along the trajectory, yielding a positive eigenvalue. In the small

445:   diffusion limit, the patches posses equal predator density only in

446:   the slow portion of the orbit, when the intra-patch dynamic is

447:   slow with regard to the migration. This leads to trajectories like

448:   those illustrated in the right panel (points connected by "rod"

449:   stand for the population density in equal times), where only one

450:   sign change happens and the Floquet eigenvalue is positive.}

451: \label{rod}

452: \end{figure}

453: 

454: 

455: 

456: 

457: 

458: 

459: \begin{figure}

460:    %Requires \usepackage{graphicx}

461:   \includegraphics [width=8cm] {CO_flo_map.eps}\\

462:   \caption{Stability diagram in the $\omega_1-D_x$ plane for the Floquet operator (same  as Figure \ref{fig1})

463:    for the coupled oscillator system described by Eqs. (\ref{co2})

464:    with $D_y = 0$. The two unstable regions correspond to different

465:    signs

466:    of the Floquet unstable eigenvalue, as explained in the text.}

467:   \label{toy}

468: \end{figure}

469: 

470: We now turn to our last point, a comparison of this stabilizing

471: mechanism with the nonlinear, noise induced mechanism recently

472: presented by us \cite{us}. The stability mechanism of \cite{us}

473: involved with the \emph{amplitude dependence} of the angular

474: velocity (see the upper right panel of Figure \ref{fig3}), works as

475: well for system of equal prey and predator migration rates and is

476: not based on a linear instability of an orbit. One may ask, thus,

477: how to make a distinction between these two mechanisms in real

478: systems.

479: 

480: In order to make a distinction between amplitude-induced stability

481: \cite{us} and angular-induced stability \cite{jansen} one should

482: compare the corresponding radius of oscillations, where the dominant

483: mechanism corresponds to the smaller radius. The amplitude

484: synchronization prediction is that the oscillation radius scales

485: like $D/ ( \omega')^2$, where $\omega' \equiv \partial \omega /

486: \partial r$ is the frequency gradient along the  oscillations amplitude

487: (See Figure \ref{fig3}, upper right panel). This result should be

488: compared with the instability radius of \cite{jansen}, and for small

489: migration rates ($D \sim 0.01$)  it is smaller in few orders of

490: magnitude. It seems, thus, that the angular induced instability will

491: be relevant only for relatively large diffusivities, where the

492: effect of amplitude-induced instability is suppressed by patches

493: synchronization.

494: 

495: In order to observe the phase instability, we have simulated the

496: two-patch LV system, where the effect of  demographic stochasticity,

497: an intrinsic noise that should appear in any system independent of

498: environmental factors, was introduced via a noise term proportional

499: to the square root of the population size. The results are presented

500: in Figure \ref{xyz}, for the two opposite cases: no predator

501: migration (where one should expect angular instability), and no prey

502: diffusion, where no such instability is present. Both cases were

503: simulated for $\mu = \sigma = 1$, so the population (number of

504: individuals in each of the species)  at the coexistence fixed point

505: is $1/\lambda$. The diffusion $D=1$ corresponds to the smallest

506: amplitude of the last stable orbit  and is way too large to allow

507: amplitude desynchronization (see \cite{us}).

508: 

509: 

510: Fig. \ref{xyz} clearly shows the stabilizing effect of

511: angular-induced desynchronization.  For small populations at the

512: fixed point (large $\lambda$) the prey diffusion systems reach the

513: absorbing state, while only  predator diffusion stabilizes the inner

514: orbits. Large populations, though, may be stable in both regimes,

515: but the instability cuts the tail of the distribution, leaving only

516: a peak close to the "reflecting boundary."

517: 

518: \begin{figure}[h]

519: \begin{center}

520: $\begin{array}{c@{\hspace{1in}}c} \multicolumn{1}{l}{\mbox{\bf (a)}}

521: &

522:     \multicolumn{1}{l}{\mbox{\bf (b)}} \\ [-0.53cm]

523: \includegraphics [width=8cm] {histPHj.eps} &

524:     \includegraphics [width=8cm] {histPHnj.eps}

525: \end{array}$

526: \end{center}

527: \caption{Histograms of the probability density as a function of $H$,

528: for a two-patch LV system with only

529:   prey diffusion  ($D_b = 1, \ D_a = 0$)(b) and only predator diffusion ($D_a = 1, \ D_b = 0$)(a). Both systems were subject to demographic stochasticity, modeled

530:   by a multiplicative noise proportional to the square root of the population density. In both cases, the probability density is concentrated

531:   around $H=0$; however, Jansen's instability manifests itself in the peak at the instability radius at the left panel, caused by the

532:   "reflection" from the unstable manifold (note the arrow that indicates the first unstable orbit). The log-log plots of that histogram

533:   (insets) show that the tail of the distribution is continuous at the right panel,

534:   but the probability to find the system with $H$ above the instability limit is  practically zero. The LV parameters are $\mu = \sigma =1$ and

535:   $\lambda = 10^{-5}$. }

536: \label{xyz}

537: \end{figure}

538: 

539: 

540: 

541: To conclude, it has been  shown that systems where only the predator

542: admits the ability to migrate [a canonical examples include

543: herbivore - plant or parasite insect - plant systems, like in the

544: famous example of biological control of the Prickly Pear cactus by

545: the moth Cactoblastis cactorum in eastern Australia \cite{cac}] may

546: support sustained oscillations in noisy environments. This

547: phenomenon has been explained here, and its cause was traced to the

548: dependence of the angular velocity on the azimuthal angle along an

549: orbit lying in the homogenous manifold. This insight allows us to

550: incorporate that phenomenon into a generic framework of coupled

551: nonlinear oscillators and to compare that mechanism with other

552: stabilizing effects. In a separate publication \cite{us2}, we intend

553: to put forward a general classification scheme for stable population

554: oscillations, a scheme that may be used to typify the observed

555: desynchronization-induced stable manifold according to its

556: underlying mechanism.

557: 

558: \begin{thebibliography}{1}

559: 

560: \bibitem{lotka}  Lotka A.J.   \emph{Proc. Natl. Acad. Sci. USA} \textbf{6},

561: 410 (1920); Volterra V. \emph{Lecon sur la Theorie Mathematique de

562: la Lutte pour le via, Gauthier-Villars}, Paris, 1931; Murray J.D.,

563: \emph{Mathematical Biology} (Springer, New-York, 1993).

564: \bibitem{gause} Gause G.F.  \emph{The struggle for existance}.

565: William  and Wilkins, Baltimore (1934); Pimentel D., Nagel W.P. and

566:  Madden J.L.,  American Naturalist, \textbf{97}, 141  (1963);

567:  Huffaker C.B. Hilgardia \textbf{27} 343 (1958).

568:  \bibitem{nic33} Nicholson, A.J.,  \emph{ J. Anim.

569: Ecol. }\textbf{2}, 132 (1933).

570: \bibitem{kerr} Luckinbill L.S. Ecology  1142 (1974); Kerr B. et. al.   \emph{Nature}

571: \textbf{442}, 75 (2006); Holyoak M. \& Lawler S.P. \emph{Ecology}

572: \textbf{77}, 1867 (2000).

573: \bibitem{levin} Earn D.J.D., Levin S.A. and Rohani P. (2000)  \emph{Science} \textbf{290}, 1360-1363;

574:  Blasius B., Huppert A. and  Stone L. (1999) \emph{Nature}

575: \textbf{399}, 354-359.

576: \bibitem{hoopes} See a recent review by  Briggs C.J. and  Hoopes M.F.

577:  \emph{Theoretical Population Biology} \textbf{65}, 299 (2004).

578:  \bibitem{jansen} Jansen, V.A.A.   \emph{Oikos} \textbf{74}, 384 (1995).

579: \bibitem{roos} Jansen V.A.A and de Roos A.M. In: \emph{The

580: Geometry of Ecological Interactions: Simplifying Spatial

581: Complexity,} eds. Dieckmann U., Law R and Metz J.A.J., pp. 183

582: Cambridge University Press, (2000).

583: \bibitem{zig} Jansen V.A.A. and Sigmund K., Theo. Pop. Bio. \textbf{54}, 195

584: (1998).

585: \bibitem{Gillespie} Gillespie  D. T.

586: \emph{Jour. Phys. Chem.}, \textbf{81}, 2340 (1977).

587: \bibitem{Abarbanel} See, e.g., Abarbanel H.D.I. \emph{Analysis of observed

588: chaotic data} Springer, Berlin (1995), p. 87.

589: \bibitem{us} Abta R., Schiffer M. and Shnerb N.M., cond-mat/0608108.

590: \bibitem{cac} Freeman D.B., Geographical Review \textbf{82}, 413 (1992).

591: \bibitem{us2} Abta R., Ben-Ishay A., Schiffer M. and Shnerb N.M., in

592: preparation.

593: 

594: \end{thebibliography}

595: 

596: \end{document}

597: