arXiv:nlin/0311056v1 [nlin.CD] 26 Nov 2003 oi rnnyeblc nhproi hoi scattering, chaotic hyperbolic In hyper either nonhyperbolic. as or characterized general, be bolic In can problem. dynamics the advection mer to the the approach about system skepticism dynamical some the to of led previous has particles, the active of of distribu importance filamental lack of the characterization This reducing phenomenological governing 1.6. not dimension about while relevant is relation, the reactions but collisional 2 and system be infective to cylinder” known ma a unstable is past the ifold of “flow been dimension the the has [3], in considered dynamics previously example, di advection For fractal underlying the lacking. of the value and observed the mension between relation clear a manifold, unstable patterns. fractal the observed con of the reaction to copy the rise giving fattened-up of products a the along result, centrate a t As of manifold saddle. unstable chaotic the re along (scattering) scattered mixing being a before in gion time w some scattering, chaotic spend of tracers context another incoming the such one in of cast dynamics with be advection can interact particles The flow. to but the particles modifying nonturbulent, active without is (Lagrangi the chaotic flow and be to The chaos) considered is dynamics diffusion. particle weak the of performed been limit have the flows, studies system incompressible the two-dimensional dynamical of time-dependent the Most for from 4]. interest [3, of community deal great a attracted [2]. oceans the and in [1] plankton atmosphere of the i dynamics in processes population depletion environmental ozone as of such react variety flows, the a open for for relevant catalysts is dynamical which as act T structures dynamics. chaotic fractal advection a the particles of active sets invariant of fractal distribution filamental a by acterized ∗ lcrncades [email protected] address: Electronic lhuhfiaetlpten aebe bevdi nature, in observed been have patterns filamental Although has flows chaotic open in processes active of study The char- are fluids in processes biological and chemical Many Py.Rv E Rev. [Phys. ASnmes 77.w 75.n 75.j 05.45.-a 47.52.+j, 47.53.+n, 47.70.Fw, numbers: PACS d verification. reactive numerical hyperbolic for studied maps previously embedding dissipative all includes it as eemndb h yeblccmoet fteudryn d underlying the of components hyperbolic the by determined o h iglrehneeto h rdcincue ythe by caused the production are quantities the invariant of dynamical enhancement and singular chemical, the biological, for in applications many for portant 1 a lnkIsiuefrtePyiso ope ytm,N¨o Systems, Complex of Physics the for Institute Planck Max nmlu ieiso netv eg,atctltc rea autocatalytic) (e.g., infective of kinetics Anomalous 68 3 nttt eFıia nvriaed ˜oPuo ax Pos Caixa S˜ao Paulo, de Universidade F´ısica, de Instituto 537(2003)] 056307 , eciednmc fieta atce nnnyeblcc nonhyperbolic in particles inertial of dynamics Reactive 2 dlo .Motter, E. Adilson eatet fMteais lcrclEgneig n P and Engineering, Electrical Mathematics, of Departments rzn tt nvriy ep,Aioa827 USA 85287, Arizona Tempe, University, State Arizona fetv rca dimension fractal effective 1, Dtd coe 1 2018) 31, October (Dated: ∗ igCegLai, Ying-Cheng tions hese long here ion, all he an of n- in it n - - - - ein(e eo) sarpeettv plcto fthes form of the of application reactions representative autocatalytic for show, a we As concepts, below). (see region where ae oethan more takes and set, fractal relevant h e ocpsi u rmwr r the are framework our dimension in flo concepts chaotic key nonhyperbolic The in particles inertial of dynamics dropped? are the (2) to and happens (1) what assumptions when then: dynamics A reactive is processes. importance environmental physical most of in expec question role and important nature an in play prevalent to are effects nonhyper f inertial Therefore, case and the sea. bolic the is in as zooplankton noninertial, of as species regarded many be to large partic too active often individual w the coexist addition, In typically motions. tracers regular of an motions obstacles chaotic from away boundaries, an Even chaos nonhyperbolicity. Lagrangian of of Ob origin origin the the obstacles. time same of the surface at are the stacles at conditions because boundary character nonhyperbolic no-slip a u is have fluids to in constrained tracers ally of dynamics ideali fu advection very the Indeed, represent as and situations inertia. rare quite and are size systems finite hyperbolic have particles si active realistic nonhyperbolic the typically in is However, dynamics Lagrangian the [5]. tions, hyperbolic dynamics is advection a particles the particles these (2) active and the tracers (1) point-like are: massless works such in Fundam assumptions orbits. tal hav periodic stable as marginally and characterized chaotic non- both frequently the while is space, counterpart phase hyperbolic the Kolmogorov no in are tori there (KAM) and Arnold-Moser unstable are orbits periodic the endas defined nmc saseilcs.W nrdc ls of class a introduce We case. special a as ynamics niomna cecs epeetasaigtheory scaling a present We sciences. environmental tosi pn ohproi hoi osaeim- are flows chaotic nonhyperbolic open, in ctions nti ril,w rsn cln hoyfrtereactive the for theory scaling a present we article, this In nvra,udryn rca atrs h key The patterns. fractal underlying universal, nmclivratst.Teter sgeneral is theory The sets. invariant ynamical hizrSrse3,017Dedn Germany Dresden, 01187 38, Strasse thnitzer and a 61,03590Sa al,Brazil S˜ao Paulo, 05315-970 66318, tal 2 N n es Grebogi Celso and ( ε fetv saerate escape effective and ) stenme of number the is fetv saerate escape effective n D κ eff eff = ( ( n hysics, ε ε ( R = ) = ) ε ( ) atcflows haotic n 3 tp oecp rmtemixing the from escape to steps hc r primarily are which , ) stefato fprilsthat particles of fraction the is − − ε sursnee ocvrthe cover to needed -squares d d ln ln d dn hc r respectively are which , ln R N ( ε ( n ε ) ) , , fetv fractal effective e are les A tua- and zed ws. ing en- su- ted (2) (1) ith lly or of of + re d d e - - - 2

B → 2B, that the area covered by B-particles in the steady ics is robust, nonhyperbolic chaotic scattering typically un- state obeys the following scaling law: dergoes a metamorphosis in the presence of arbitrarily small amount of dissipation. For nonhyperbolic scattering in open − 2 Deff Hamiltonian systems, particles can spend a long time in the A ∼ σ B − , (3) eµeff /(2 Deff ) − 1 neighborhood of KAM tori, resulting in an algebraic decay for the survival probability of particles in the scattering re- where µeff ≡ (κeff +κ ˜)τ, τ is the time interval between gion. As a consequence, the fractal dimension of the invariant successive reactions (time lag), σ is the reaction range, and manifolds is the phase-space dimension [11]. This should be κ˜ is the contraction rate due to dissipation. For nonhyper- contrasted with the hyperbolic case, whose decay is exponen- bolic flows in two dimensions, Deff < 2 and κeff > 0 are tial and fractal dimension is typically smaller. The dissipation, non-trivial functions of the scale ε, which can often be re- however, may convert marginally stable periodic orbits of the garded as constants over a wide interval, even though D = KAM islands into attractors. The survival probability then limε→0 Deff (ε)=2 and κ = limε→0 κeff (ε)=0. We becomes exponential, the dimension of the chaotic saddle be- find, surprisingly, that Deff and κeff are significantly differ- comes fractional, and the overall dynamics of the scattering ent from D and κ, respectively, not only for noninertial but process becomes hyperbolic. also for inertial particles, even though the advection dynamics To understand the meanings of Deff and κeff for fractal of the latter is hyperbolic, meaning that scars of the nonhy- sets arising in the transition from Hamiltonian nonhyperbolic perbolic conservative dynamics are observable in the hyper- to weakly dissipative chaotic scattering, we consider a Cantor bolic dynamics of slightly dissipative systems (κ˜ ≪ 1). The set, which is constructed in the interval [0, 1] according to the previous relations for noninertial particles in hyperbolic fluids rule that in the nth time step, a fraction ∆n = γ/(β + n)+ δ appear as a particular case of our results. is removed from the middle of each one of the N =2n−1 re- The nature of the chaotic scattering arising in the context of maining subintervals, where β, γ, and δ are constants. The particle advection in incompressible fluids may change fun- conservative case corresponds to δ = 0, which is charac- damentally as the mass and size of the particles are increased terized by an algebraic decay with n of the total length re- from zero. Physically, this happens because of the detachment maining, given by R(n) ∼ n−γ for n ≫ β, and by a unity of the particle motion from the local fluid motion. For spher- fractal dimension for the invariant set, D = 1. The re- ≡ ical particles of finite size, the particle velocity v dx/dt moved fraction ∆n decreases at each time step and, as a re- is typically different from the (time dependent) fluid velocity sult, a systematic change of scales is induced, resulting in a u = u(x,t) and, in first order, is governed by the equation [6] non-self-similar invariant set that becomes denser as we go to smaller scales. The relevant consequence is that the box- counting dimension converges slowly to 1, leading to a scale- dv du −1 − α = −a (v − u). (4) dependent effective fractal dimension, Deff ≈ 1 − γ/ ln ε dt dt for small ε. Similarly, the effective escape rate behaves as ≈ ≈ −1 The parameters are α = 3ρ /(ρ +2ρ ) and a = 2 α/St, κeff γ/n γ ln 2/ ln ε , where n = n(ε) is defined as f f p 3 the number of iterations needed to make the length of each where ρf and ρp are the densities of fluid and particle, re- spectively, and St is the Stokes number, which goes to zero remaining subinterval smaller than ε. in the limit of point-like particles [7]. For neutrally buoy- The limiting dynamics changes drastically and acquires ant particles, the mass ratio parameter is α = 1, while for properties of hyperbolic dynamics when a small amount of aerosols and bubbles, we have α < 1 and α > 1, respec- dissipation is allowed, which is modeled by 0 < δ ≪ γ/β. tively. The inertia parameter a determines the contraction rate In particular, the total length of the remaining intervals de- or dissipation in the phase space (x, v), which for incom- cays exponentially, R(n) ∼ (1 − δ)n for n ≫ γ/δ − β, pressible flows can be shown to be −2a. In the limit a → ∞, and the dimension of the invariant set becomes smaller than the dynamics is projected on a surface defined by v = u, one, namely, D = ln 2/ ln [2/(1 − δ)]. At finite time, how- which corresponds to the advection dynamics of point par- ever, the transition from the conservative to the dissipative − ticles. The configuration-space projection of the particle mo- case is much smoother. For ln ε 1 ≫ β, the effective frac- tion is strongly influenced by α [8]. For small inertia (large a), tal dimension and the effective escape rate are Deff (ε) ≈ − − ′ −1 −1 ∇ · ≈ 1 − ∇ · · ∇ 1 − 2 − 2 , ln 2/ ln [2/(1 − δ)]−γ / ln ε and κeff (ε) ≈ ln(1−δ) + v a (α 1) ((v )u) = a (α 1)(s ω ) ′ − ′ where s and ω are proportional to the strain rate and vortic- γ ln [2/(1 − δ)]/ ln ε 1, respectively, where γ ≡ γ/(1 − δ). ity of the fluid [9], respectively. The behaviors of bubbles The key feature is that unrealistically small scales are required and aerosols are then qualitatively different. For instance, to resolve the limiting values of the fractal dimension and along a closed orbit, aerosols are pushed outward, while bub- the escape rate, rendering them physically irrelevant. For in- −20 bles are pushed inward. We first consider bubbles, whose stance, to obtain Deff > 0.95, scales ε < 10 may be configuration-space dynamics is dissipative when the vorticity required. Thus, the physically important characteristics of the overcomes the strain rate. fractal set are the effective dimension and escape rate. Dynamically, the inertial effects are effectively those due to We now present a physical theory for the scaling law (3), dissipation, so that the transition to finite inertia is equivalent valid for autocatalytic reactions in two-dimensional time- to a transition from open Hamiltonian to dissipative dynamics. periodic flows. Consider the area covered by B-particles in It has been recently shown [10] that, while hyperbolic dynam- the open part of the flow (i.e., region where particles eventu- 3 ally escape to infinity) and, to be specific, that the time lag τ is nates. This class of embedding maps can be a paradigm to integer multiple of the flow’s period. After a sufficiently long address many problems in inertial advection dynamics as it time fromthe onset of the reaction, the reagent B is distributed captures the essential properties of Eq. (4). In particular, it is along stripes of approximately uniform width, mimicking the uniformly dissipative, with phase-space contraction rate equal −2a unstable manifold. The average width ǫ of the stripes changes to e ; the noninertial dynamics xn+1 = M(xn) is recov- aperiodically over time until the steady state is reached, when ered in the limit a → ∞; and the configuration-space contrac- it undertakes the periodicity τ of the reaction. We assume tion rate is proportional to e−a(α − 1) for e−a(α − 1) ≪ 1, that the reaction is sufficiently close to the steady state so that in agreement with the distinct behavior expected for aerosols Deff (ǫ) and κeff (ǫ) can be considered constant over time. and bubbles. Therefore, for finite a, a rich higher dimensional This condition is not so restrictive because for many systems dynamics with α-dependent x-space projection is expected. Deff and κeff are essentially constant over several decades Next we consider such a dynamics for both α> 1 and α< 1. (see below). Therefore, for scales larger than ǫ, the area cov- To simulate the flow, we consider a two-dimensional area- ered by B-particles can be regarded as a fractal characterized preserving map that has a pronounced nonhyperbolic char- by dimension Deff (ǫ) and escape rate κeff (ǫ). acter [11]: → − 2 −1 2 , where − (x, y) [λ(x w /4), λ (y + w )] Let ǫ(n 1)(τ) and ǫ(n)(0) denote the average widths of the w ≡ x + y/4 and λ > 1 is the bifurcation parameter. The stripes right before and right after the nth reaction [3], respec- dynamics is nonhyperbolic for λ . 6.5. For λ =4, for exam- tively. Between successive reactions, the stripes shrink due to ple, there is a major KAM island in the -space, as shown − xy escape and dissipation as follows: ǫ(n)(τ)= ǫ(n)(0)e heff τ , in Fig. 1(a). Also, from Fig. 1(a), one can see tangencies be- where heff = (κeff +κ ˜)/(2 − Deff ) plays the role of an tween the stable and unstable manifolds in the neighborhood effective (contracting) Lyapunov exponent, while κ˜ accounts of the KAM island, which is a signature of nonhyperbolic- for the nonconservative contribution. When the reaction oc- ity. It is well established that, within the nonhyperbolic re- curs, the widening due to the reaction is proportional to the gion, the dimension of the invariant manifolds is D = 2 and reaction range: ǫ(n+1)(0) − ǫ(n)(τ) ∝ σ. The area covered by the escape rate is κ = 0 [10, 11]. When this map is em- A(n) B-particles right before the (n + 1)th reaction B satisfies bedded in Eqs. (5-6), for α > 1, the xy-projection of the re- − A(n) ∝ (n) 2 Deff sulting 4-dimensional map is dissipative in the mixing region B (ǫ (τ)) . These relations can then be com- bined to yield the following recursive relation for the area: (KAM islands and neighborhood). In this regime, the dissi- 2−D (n+1) − (n) − eff pation stabilizes marginally stable periodic orbits in the KAM A µeff A 1/(2 Deff ) B = e ( B ) + c σ , where islands of the conservative map, converting the KAM islands h i µeff = (κeff +κ ˜)τ and c is a constant geometric factor. and neighborhoodinto the correspondingbasin of attractionof From the condition A(n+1) = A(n) follows our main scal- the newly created attractors, as shown in Fig. 1(b). The basin B B itself extends around the mixing region mimicking the stable ing (3) for the area AB in the steady state [12]. This scaling holds for both noninertial and inertial particles, regardless of manifold of the conservative dynamics. As a result, the tan- whether the flow is hyperbolic or nonhyperbolic. The hyper- gencies between the invariant manifolds apparently disappear, bolic case with inertial particles is studied in Ref. [13]. The suggesting that the advection dynamics of bubble particles is scaling (3) represents a further step toward generality since it hyperbolic. For α < 1, on the other hand, the configuration- is also valid for nonhyperbolic flows. space projection expands in the mixing region and almost all To make possible a numerical verification of the scaling law the orbits eventually escape to infinity. However, for small in- (3), it is necessary at present to use discrete-time maps. To ertia and α close to 1, particles in the regions corresponding to construct a class of maps that captures all essential features of KAM islands of the conservative dynamics and neighborhood continuous-time chaotic flows, we note: (1) the fluid dynam- are almost trapped in the sense that the time it takes to escape ics, determined by dx/dt = u, is embedded in the particle’s is much larger in these regions than outside them. These re- advection equation and is recovered in the limit a → ∞; (2) gions are neglectedin our analysis of the openpart of the flow, the phase-space contraction is determined by a (irrespective as shown in Fig. 1(c), because filamental structures cannot be of α); (3) for small inertia, the configuration-space contrac- resolved inside them. tion is proportional to a−1(α−1). For an area preservingmap Numerical simulation of the autocatalytic reaction is per- xn+1 = M(xn), representingthe dynamics of a time-periodic formed by dividing the mixing region with a grid where the incompressible fluid, a possible choice for the corresponding size of the cells represents the reaction range σ. Particles are embedding map representing the inertial particle dynamics is placed in the center of the cells. When a reaction takes place −a xn+2 − M(xn+1)= e (αxn+1 − M(xn)), where the fac- in a cell occupied by a B-particle, all the cells adjacent to tors involving a and α are naturally imposed by the particle it are infected with B-particles [15]. We assume that A is dynamics [14]. This can be written as the background material and that the reaction takes place si- multaneously for all the particles at time intervals τ [16]. If xn+1 = M(xn)+ δn, (5) we start with a small seed of B-particles near the stable man- −a δn+1 = e (αxn+1 − M(xn)), (6) ifold, after a transient time a steady state is reached where B-particles are accumulated along a fattened-up copy of the where x and δ can be interpreted as the configuration-space unstable manifold, as shown in Fig. 1(d) for massless point coordinates and the detachment from the fluid velocity, re- particles, in Fig. 1(e) for bubbles, and in Fig. 1(f) for aerosols. spectively, so that (x, δ) represents the phase-space coordi- In the computation, particles A and B are set to have the same 4 mass ratio and inertia parameters. To compute the effective proach to the reactive dynamics in imperfectly mixed flows fractal dimension Deff of the unstable manifold, we use the also applies to realistic situations where nonhyperbolic and uncertainty algorithm [17] applied to the first order approx- inertial effects are relevant. The rate equations of reactive imation of the inverse map. The effective dimension turns processes are primarily governed by finite-time dynamics and out to be constant over many orders of magnitude of varia- as such change smoothly in the noninertial → inertial transi- tions in ε and it is approximately the same for both nonin- tion, which is in sharp contrast with the metamorphosisunder- ertial and slightly inertial bubble particles (Deff = 1.73 for gone by the long-term and asymptotic dynamics of reaction- ε > 10−15), while it is somewhat smaller for slightly inertial free particles. We have focused on autocatalytic reactions, −15 aerosol particles (Deff = 1.68 for ε > 10 ), as shown in but our results are expected to hold whenever the reaction Fig. 2(a). Strongevidenceof the scaling law (3) is presented in front mimics the underlying unstable manifold and the activ- Fig. 2(b) for two different values of the time lag τ, where the ity takes place along the boundary of a fattened-up fractal. scaling exponent is consistent with Deff = 1.73 for nonin- Examples of this kind of process include infective reactions ertial and bubble particles, and with Deff =1.68 for aerosol (e.g., combustion [18]) and collisional reactions in general particles. We see that even though the area AB changes with (e.g., A + B → 2C, where an unlimited amount of mate- the inertial properties of the particles, the scaling of AB re- rial A is present and material B is continuously injected in the mains essentially the same for bubbles, as expected from our vicinity of the stable manifold [3]). Finally, we observe that Cantor-set model. our analysis does not rely on the existence of a well-defined It is instructive to compare this result with the reaction-free fractal set in the advection dynamics. The results remain valid dynamics. The dimension of nonhyerbolic invariant sets can as long as effectivevalues for the fractal dimension and escape be arguedto be integerby mean of a zoom-intechnique, where rate can be properly defined and are approximately constant a fast numerical convergence is achieved by focusing on the over the relevant interval of observation. This is important for densest parts of the fractal [11]. The reaction, however, hasa environmental processes, whose underlying dynamics is only global character, as it takes place along the unstable manifold partially understood [13]. around the entire mixing region. This makes the convergence of the relevant effective dimension extremely slow, and that is AEM and YCL were supportedby AFOSR underGrant No. why the effective dimension is apparently constant. F49620-03-1-0290. CG was supported by Fapesp and CNPq. In summary, we have shown that the dynamical system ap- AEM thanks Tamas T´el for illuminating discussions.

[1] S. Edouard, B. Legras, F. Lefevre, and R. Eymard, Nature (Lon- mation of the bailout embedding map introduced in [J. H. E. don) 384, 444 (1996). Cartwright, M. O. Magnasco, and O. Piro, Phys. Rev. E 65, [2] E. R. Abraham, Nature (London) 391, 577 (1998). 045203 (2002)] to study the advection dynamics of neutrally [3] Z. Toroczkai, G. K´arolyi, A.´ P´entek, T. T´el, and C. Grebogi, buoyant particles. Phys. Rev. Lett. 80, 500 (1998). [15] For numerical convenience, the autocatalytic reaction is fol- [4] See, for example, Focus issue: Active chaotic flow, Chaos lowed by coalescence so that when more than one B-particle 12(2), 372 (2002); references therein. is found in the same cell, they are all replaced by a single B- [5] Inertial effects on reative particles have been previouly consid- particle in the center of the cell. Without coalescence, the num- ered for turbulent flows [R. Reigada, F. Sagu´es, and J. M. San- ber of B-particles diverges because of the shrinking due to dis- cho, Phys. Rev. E 64, 026307 (2001)] and for cellular flows [T. sipation, but the area AB still reaches a steady state. Since we Nishikawa, Z. Toroczkai, C. Grebogi, and T. T´el, Phys. Rev. focus on the area covered by B-particles, our results do not E 65, 026216 (2002); Z. H. Liu, Y.-C. Lai, and J. M. Lopez, depend on the coalescence. Chaos 12, 417 (2002)]. [16] The time lag τ is assumed to be on the order of or smaller [6] M. R. Maxey and J. J. Riley, Phys. Fluids 26, 883 (1983). than the relevant time scale of the advection dynamics τf = [7] Eq. (4) is an approximation of the Maxey-Riley equation [A. 1/κeff (σ). For τ ≫ τf , the reaction undergoes an emptying Babiano, J. H. E. Cartwright, O. Piro, and A. Provenzale, Phys. transition and the Cantori structures near the KAM tori can- Rev. Lett. 84, 5764 (2000)]. not be neglected. In this regime, almost all the particles along [8] I. J. Benczik, Z. Toroczkai, and T. T´el, Phys. Rev. Lett. 89, the unstable manifold escape in between successive reactions 164501 (2002). and the hypothesis that reagent B is distributed along approxi- [9] M. R. Maxey, Phys. Fluids 30, 1915 (1987); E. Balkovsky, G. mately uniform stripes breaks down outside the Cantori. How- Falkovich, and A. Fouxon, Phys. Rev. Lett. 86, 2790 (2001). ever, since Cantori are obstacles to the transport of particles, [10] A. E. Motter and Y.-C. Lai, Phys. Rev. E 65, 015205 (2002). for σ small enough, high concentration of particles can still be [11] Y.-T. Lau, J. M. Finn, and E. Ott, Phys. Rev. Lett. 66, 978 found inside the hierarchy of Cantori structures. For details we (1991). refer to [A. P. S. de Moura and C. Grebogi, preprint (2003)]. [12] A similar scaling law can be derived in the continuous-time [17] C. Grebogi, S. W. McDonald, E. Ott, and J. A. Yorke, Phys. limit τ → 0. Lett. A 99, 415 (1983). [13] T. T´el, T. Nishikawa, A. E. Motter, C. Grebogi, and Z. [18] I. Z. Kiss, J. H. Merkin, S. K. Scott, P. L. Simon, S. Kalliadasis, Toroczkai, Chaos 14, 72 (2004). and Z. Neufeld, Physica D 176, 67 (2003). [14] For α = 1, this equation reduces to the first-order approxi- 5

FIG. 1: For λ = 4, (a) KAM island (light grey), and stable (grey) and unstable (black) manifolds for a = ∞. (b) Fixed point attractor (black dot), basin of attraction (grey), and unstable manifold (black), for α = 1.05 and a = 1. (c) Stable (grey) and unstable (black) manifolds for α = 0.95 and a = 1, outside the region covered by the almost trapped orbits (light grey). Particles are launched with initial velocity matching the fluid velocity (δ0 = 0). (d-f) Corresponding area covered by B-particles in the “open” part of the flow right before the reaction, for τ = 5 and σ = 5.10−3.

FIG. 2: For λ = 4, (a) Effective dimension of the unstable manifold as computed from the uncertainty method, where f(ε) is the fraction of ε-uncertain points in the line x = 0, 0

http://arxiv.org/ps/nlin/0311056v1 This figure "fig2.gif" is available in "gif" format from:

http://arxiv.org/ps/nlin/0311056v1