Chaotic scattering from pinball machine like systems

Constantia Alexandrou¹  and Iacovos Kyprianou²
Department of Natural Sciences, University of Cyprus, CY-1678 Nicosia, Cyprus

Abstract

Note: The software of this article can be downloaded from: http://www.buffalo.edu/~kypriano/ . On the same site you can find a postscript version of the article with all the fonts included, in case your browser does not support HTML 4.0, or you need higher quality figures.

1  Introduction

Inelastic scattering processes are abundant in nature and a common phenomenon in classical physics. Energy is usually lost upon collision through various mechanisms as for example friction. An important example where dissipation plays a crucial role is the chaotic motion in the Solar system due to the gas drag force [1]. In addition, the behaviour of systems under an external driving force, usually taken to be harmonic, is of practical and theoretical interest. It is the purpose of this article to study the strange behaviour observed in the scattering of particles from pinball machine like systems i.e. systems subjected to an external force. The scattering region, S, of our system consists of a geometric arrangement of hard disks which exert a radial constant force when hit by a point particle. The difference of this system from the well studied problem of hard disk scattering is the fact that the velocity of the moving particle is no longer constant but depends on the number of collisions, giving rise to multifractal scaling laws. Since an external force is involved the system is not Hamiltonian. This study thus focuses on the chaotic behaviour as a function of the external force. This complements other studies on chaotic scattering of such systems which up to now where performed using inert hard disks or Hamiltonian potentials, as for example in Refs. [2,3,4,5,6,7,8,9,10]. In these studies a quantitative description of the chaotic properties of conservative systems was carried out: In Refs. [2,3] for instance a study of dynamical systems exhibiting particle diffusion such as the Lorentz gas was presented. Gaspard and Rice presented analytic results for the decay rate and Lyapunov exponents for the three and four disk systems [4] and Yalçinkaya and Lai [5], following their work, studied the chaotic behaviour of the hard disk repellor. Antillón et al [6] studied the time dependent two disk problem. Refs. [7,8,9] presented studies of hamiltonian models with mostly repelling potentials. Daniels et al [10] considered a chaotic model using two attracting valley potentials. A comprehensive description of these results is given in a review article on chaotic scattering by Ott and Tél [11]. In the same reference further applications and issues of chaotic scattering can be found.

The quantitative and theoretical results obtained in the aforementioned articles are needed for a quantitative investigation of chaotic scattering and will be extensively used in this work, like for instance, the pressure function P(b), the uncertainty dimension D_U [12] and the multifractal scaling functions. Results on these quantities are used to quantify our study and to compare our work with previous results of similar nature.

2  The dynamics of pinball machine like systems

We consider a number of hard disks fixed on a pinball machine table. When hit by the moving hard ball a radial force is exerted on the ball giving it a constant amount of additional speed independent of the ball's hitting momentum. In this process the energy of the ball always increases and therefore it can never stop in the scattering region. This makes this study simpler than if we had considered a dissipative force. However because of time reversal invariance of our model, its chaotic properties are equivalent to a dissipative one run backwards in time. To study the pinball machine it suffices to consider hard disk scattering of a point particle in two dimensions with a radial velocity given to the particle upon reflection from the hard disks. We thus focus on the chaotic behavior of a system of hard scatterers as a function of the radial force exerted by them upon each collision. Like in previous studies of hard disk chaotic scattering[2,3,4,5,6,13,14,15] with elastic collisions we also here investigate the dependence of the results on the geometry of the hard disks. In particular we will present results for systems with infinite and finite horizons.

We consider a point particle of unit mass which scatters inelastically on hard disks, fixed in the plane r = (x,y). All the hard disks have the same radius a. The center of the m^th disk has coordinates r_c^m = (x_c^m, y_c^m). The motion of the point particle follows a straight line path between collisions. At each collision its trajectory can be found using momentum conservation. At the k^th collision let the incoming velocity of the particle be u_k. The particle hits the m^th disk on its circumference at point C with coordinates r_k as shown in Fig. 1. We take the incident angle q_k to be the angle that the incoming velocity of the particle makes with the x-axis counterclockwise. At C the particle receives radial momentum u_e as it bounces off the hard disk and its final velocity is thus changed to u_k+1. This velocity is the vector sum of u_e and the velocity it would have, had the collision been elastic. Using the notation of Fig. 1 we find

3  The configuration of the scatterers

We study chaotic scattering under various configurations of the hard disks. Like in the studies of the Lorentz gas we distinguish configurations with finite and infinite horizons.

Following the pioneer work of Gaspard and Rice [4] as well as Refs. [14,13,15], we fix the disks on a triangular lattice with their centers a distance d apart. The ratio of the intercenter distance to the radius of the disks, d/a, determines the diffusive character of the Lorentz gas in an infinite lattice. In the high-density regime where the intercenter distance is

In the low-density regime, where d exceeds the critical value 4a/Ö3, there exists the possibility for the point particle to travel without collisions throughout the whole lattice. This can only occur along special trajectories given by the discrete values of the incident angle q₀ = 0, p/3,2p/3,p, ¼. Along these special trajectories of zero Liouville measure the particle emerges from the scattering region without deviation. This situation is known as the infinite horizon configuration. Along an arbitrarily trajectory, the particle will suffer inelastic collisions. We also expect in this configuration a modification in the behaviour from that seen in the studies of the Lorentz gas where only elastic collisions take place. We use configurations of three to six hard disks with d = 2.5 to study the low-density regime of our model. As in the case of the Lorentz gas, we also expect that for an infinite lattice the diffusion coefficient to be infinite due to the existence of the special trajectories where collisions do not occur.

In addition to the triangular lattice we studied scattering on a square body centered lattice. We take the size of the square to be 2d. This configuration is displayed in Fig. 3 along with some trajectories. For a square lattice, the particle has an infinite horizon for any d > Ö2a. For this study we also take a = 1 and d = 10/3 Ö3, and we consider systems of three up to nine disks. This choice of d give us a disk density slight higher than the d = 2.5-configuration of the triangular lattice.

It is well known, that the dynamical mechanism for the occurrence of chaotic scattering, is the existence of an invariant set of infinite unstable periodic orbits in the scattering region. When scattering a particle from two or more disks there are unstable periodic orbits. They can be represented by the sequence of symbols d₁,d₂,¼,d_m, ¼ where d_m = j denotes that the particle hits disk j at the mth collision. That is when a particle starts in one of these orbits it never leaves the scattering region S and it bounces on disks ¼,d_m-1, d_m, d_m+1, ¼. Thus the sequence of symbols

A feature that still remains for u_e ¹ 0 is that the scattering is still hyperbolic and there are no KAM tori present. The periodic orbits are unstable in the sense that only a slight perturbation of the initial conditions sends the particle away exponentially fast implying that there is at least one positive Lyapunov exponent. The set of these periodic orbits is called the repeller A_e. This behaviour is common for both the low and high-density regime and it applies to all numbers of disks.

4  Probes of chaotic scattering

4.1  The deflection function

In this work we give a detailed comparison of the deflection function for different values of u_e. The u_e = 0 case has been studied extensively [5] and it is used for comparison. In Figs. 4 we show the results for the deflection function for u_e = 0,  1 and 2 using initial conditions

4.2  The time delay function

Another probe of chaotic behaviour is the time that the particle remains in the scattering region S known as the time delay function [17]. If a particle crosses the boundary ¶S at time t_in entering the scattering region and then exits S at time t_out then the time delay function is defined as T(b,u_e) = t_out-t_in. We expect that the periodic orbits will produce singularities in the delay function, due to the almost infinite time that the particle spends in S. We also expect that the density of these singularities is the same as that observed for the scattering function Q(b,u_e). This behaviour is demonstrated in Figs. 4 for the three-disk system in the high density regime. Plotting in Fig. 4 the maximum time T(b,u_e) for various values of u_e we see that one main difference for non-zero u_e is an overall decrease in the value of the maximum time. In other words the time scale of the scattering process depends on u_e. But the actual values of the T(b,u_e) scale are of no importance; the interesting part is where the singularities are situated. The singularities of T(b,u_e) for all the values of u_e are located at the same b values as the singularities of the corresponding deflection functions, i.e. the region containing the singularities is compressed as the inelasticity increases in the same manner that it does for the deflection function.

Our main conclusion here is that the singularities will be compressed as u_e increases no matter what configuration we use and no matter what number of disks we configure. We will demonstrate this explicitly for various cases in the following section.

5  Multifractal behaviour and scaling functions

5.1  Scaling and fractal boundaries

In order to study the behaviour of the singularities we define the two basins [21] of the repeller; the first being the range q Î [0,p), and the second when q Î [p, -p). The line between the two basins is called the fractal basin boundary, with a non integer dimension. In Fig. 5(a,b) we mark with white the initial conditions (u_e, b)_W producing trajectories that result in basin 1 and with black the initial values (u_e, b)_B whose trajectories end in basin 2.

The regions with solid white or black stripes are regions having regular behaviour, i.e. Q(u_e,b) is a smooth function. The regions containing the singularities are the non smooth regions in the (u_e,b) plane and are seen to compress as u_e increases. The areas of smooth flow are bounded by fractal boundaries which converge along a critical line of flow at b_crit = (d-2a)/2 = 0.15 for the three-disk triangular lattice (using d = 2.3 and a = 1). Magnifying a small region as shown in Fig. 5(b) reveals the internal structure of the flow lines, a feature characteristic of a strange attractor seen in many chaotic systems as for example in the Hénon Map. Thus the singularities of the deflection function show multiscale similarities a typical fractal property. The fractal boundaries encountered here are Cantor-like and are locally disconnected [22]. This means that the areas of smooth flow (smooth rivers) follow other smooth rivers separated by fractal boundaries, but the rivers can not cross each other. As demonstrated in Fig. 5(a) the flow of the fractal boundaries as u_e increases for b > 0.15 is compressed along the critical boundary at b_crit exponentially.

This may come of no surprise given that the velocity of the particle is expected to be an exponential function of time⁴ for the singular values of b. To understand this consider two particles having the same initial conditions, (b₀ > b_crit,q₀), but for the first particle let u_e⁽¹⁾ = 0 and for the second let u_e⁽²⁾ = 1. The first particle scatters elastically following the reflection laws and its speed remains constant. On the other hand as the second particle hits more disks its momentum increases exponentially with the result of ending up in a basin that the first particle with b < b₀ would have chosen. This of course happens only for a small number of collisions. When the number of collisions is very high then the particles speed is so high that it scatters almost elastically, so for the exponential behaviour only the first few collisions matter. When the particle is initiated in the range b Î [0,b_crit], it suffers only one or two collisions, so the probability to change its basin as u_e increases, is very small.

That an exponential dependence describes the main features of the flow of singularities observed in the scattering process is best seen by plotting the points along a fractal boundary starting, e.g. at u_e = 0,  b = 0.239094 as done in Fig. 5(c), and fitting these points to the form c₁-c₂e^{-c₃ u_e}. The quality of the fit indicates that the singularity flow dependence on u_e is consistent with an exponential behaviour. Using such a fit one can follow a fractal boundary flow for a while and foresee by extrapolating whether a certain initial condition will choose a certain basin.

5.2  The particle decay rate g

A probe for the existence of a nonattracting chaotic invariant set is the decay rate. For a hyperbolic chaotic set the remaining number of particles N(t) in the scattering region S at time t, decays exponentially,

The decay rate was studied extensively for u_e = 0 in Refs. [4,15] for various values of d for the three-disk system and the Lorentz gas for open and closed systems.

Here we study the scaling properties of the decay rate as a function of u_e for our two geometries. The algorithm we use for the computation of g is the following: The system is initiated with typically 10000 particles entering the scattering region, with initial conditions described by Eq. (7). In regular time steps Dt we count the number of particles N(t) left in the scattering region, and plot N(t) on a semi logarithmic scale. The slope of the linearly fitted line is the decay rate g.

The results for g extracted from the slope as a function of the external momentum u_e are shown in Fig. 6 for the body centered lattice and in Fig. 7 for the triangular lattice. For the body centered lattice, g increases with increasing u_e in all cases, that is the mean time, t, the particle remains in the scattering area S is decreasing roughly linearly, as the velocity of the particle increases. As the number of disks increases the particles are trapped longer inside the scattering region and the rate of increase of g with u_e decreases until the number of disks reaches about six beyond which the decay rate becomes independent of the number of disks.

For the triangular lattice the escape rate also increases monotonically with a slope that decreases with the number of disks until again about six beyond which g becomes independent of the number of disks. In fact for the infinite horizon configuration the rate of increase of g with u_e for six disks is practically the same as that observed for the body centered lattice showing that scaling of g becomes independent of the geometry of the scatterers. For the three disks in both the finite and infinite horizon configurations we obtain an interesting behaviour: g first increases with u_e and then quickly reaches a constant value. This is most clearly seen in the finite horizon configuration, d = 2.1. The fact that this behaviour is also observed for five disks but not with four diks indicates that it has to do with the geometric arrangement of the scatterers. For the two disk system we argued that D = {...,j-1, j, j-1,...} is independent of u_e. In the case of three disks we can identify long trajectories which for u_e > 0.2 are almost independent of u_e. For these long trajectories there is an increased probability for the particle to bounce back and forth on the same disks for a relatively large number of times.

5.3  The scaling of the uncertainty dimension D_U

A more compact way of describing the quantitative characteristics of chaotic scattering is by determining the fractal dimension of the repellor A_e obtained from the geometric density of the singularities of the scattering and delay functions. Here we consider the uncertainty dimension D_U as defined in Ref. [12], to characterize fractal basin boundaries arising commonly in dissipative chaotic systems with multiple attractors. There are many examples where D_U is used to characterize chaotic scattering (see e.g. Refs. [17,21,5,16]).

The algorithm for computing D_U is as follows: we set x₀ = c, c < 0 and y₀ = b, where b is chosen randomly in the range given in Eq. (7) for each configuration. For a certain b we calculate the orbit and then initiate the particle from a nearby orbit with b = b+e where e is very small. If the number of bounces in the scattering region is the same for both trajectories then we call the value of b certain. If the number of bounces is not the same then b is uncertain. For a given e we compute the fraction of uncertain initial conditions f(e) for many b. We can then fix the number of uncertain b - values we want to consider, say n_u, and compare it with the total number of initial b's denoted by n_tot. f(e) scales with e as

We computed D_U for a number of disks for u_e Î [0, 1] for both the triangular and the body centered configurations in order to examine the effect of geometry on the scaling behaviour. The fractal dimension is a measure of the density of singularities and thus, following the reasoning of Sec. 5.1, we assume an exponential dependence of the form

The dependence of D_U on u_e for the infinite horizon configurations is shown in Fig. 6 for the body centered and in Fig. 7 for the triangular with d = 2.5. For this low density regime the exponential scaling of D_U with u_e is nicely displayed. This is what is expected since the low density of the disks gives room to singularity compression starting from lower values of D_U in contrast to the high density regime where the starting values are close to the limiting value. As the number of the disks increases the set of singularities becomes more dense and hence D_U reaches the asymptotic value almost immediately. Increasing the disk number increases the singularity density but it is also responsible for the increase in the asymptotic value of D_U which approaches unity from below. For the finite horizon configuration d = {2.1,2.3} shown in Fig. 7 D_U can be considered independent of u_e for configurations with N > 3. This is understood by the fact that the particle is trapped in the scattering region giving rise to a ``denser'' singularity set thus reaching the asymptotic value for any number of disks independently of the value of u_e. Thus for this case, an infinite lattice of pinballs would not differ in its properties, from an infinite lattice of hard disks. For the highest density configuration (d = 2.1) D_U has the largest asymptote but always stays below unity.

6  The Lyapunov exponents and the pressure function

In this section we will try to characterize the chaotic nature of the repeller. We calculate quantities that deal with the statistics of particle scattering and the amount of information we can obtain from our system. We will define the system's pressure function, and hence calculate the generalized dimensions of the repeller, the entropies and other quantities, as the system becomes inelastic.

The Lyapunov exponents and the pressure function can be calculated in terms of the Sinai-Bunimovich curvature k_u(G_t) of the horocycle of a trapped trajectory G_t. The horocycle is the geometrical wave front associated with the reference trajectory. A generalization of the result for the curvature derived in Refs. [3,23] for elastic hard disk scattering, is obtained here for inelastic scattering. We show in the Appendix that the curvature of the front between the kth and the (k+1)th collisions is given by the following continuous fraction

6.1  The average Lyapunov exponent

The stretching factor of the expanding front for an initial condition G₀ that will produce a total path length L is defined by

Using the same averaging procedure we can calculate the variance of the Lyapunov exponent [15]

6.2  The pressure function

The pressure function is a generalization of the Kolmogorov-Sinai (KS) entropy and was introduced into the thermodynamics formalism of dynamical systems by Bowen and Ruelle. It was named pressure in analogy to the standard pressure of statistical mechanics.

Following [15,13] we define the pressure function for open systems as

The second term of Eq. (21) has in its definition the particle's equation of motion, and thus is calculated with the same procedure used for the average Lyapunov exponent. It is expected that the pressure function is a convex function; either monotonically decreasing or constant with b. Our results shown in Fig. 8 comply with this expectation for all values of u_e. It was shown in Refs. [25,4,23,15] that many important properties of the chaotic scattering system can be calculated directly from the pressure function:

1. From Eq. (16) the average Lyapunov exponent is given by the derivative of the pressure function at b = 1 with constant u_e, i.e. l= -[(¶P/¶b)_ue]_{b = 1}.

2. The KS entropy per unit time is h_KS = l - g, where

3. the decay rate is given by g = -P(u_e,1),

4. and the topological entropy is h_top = P(u_e,0).

5. The partial Hausdorff dimension, d_H is given by the root of the pressure function, P(u_e,d_H) = 0. Because of the time reversal symmetry of classical Hamiltonian systems, the partial Hausdorff dimensions along the stable and unstable manifolds are equal. When the system becomes bounded, P(u_e,1) = 0, so d_H approaches unity, and the fractal repeller starts to fill the whole phase space. Since we are dealing with hyperbolic dynamics it is convenient to introduce the Hausdorff codimension as the defect of d_H with respect to one, D_H = 1-d_H, the same way as the uncertainty dimension from Sec. 5.3. In this manner when the system becomes bounded D_H will be zero.

6. The information dimension D_I is defined again by the partial information dimension d_I = h_KS/ l, as D_I = 1-d_I, following the same reasoning as in v.

As we see from g = l-h_KS = l(1-d_I), the escape rate vanishes as the system becomes bounded, a property showing in the graphs for the smaller values of d in Sec. 5.3.

In Table 2 we collect the results from these quantities for u_e = 0 as well as for non-zero values up to u_e = 0.8 in order to investigate the variation in the behaviour of these quantities as a function of u_e and to compare with the previous direct calculations. The displayed values of the escape rate g, the mean Lyapunov exponent l and its variance s² are calculated directly by Eq. (9), Eq (16), Eq. (17) respectively, whereas the other quantities are obtained from the pressure function. For comparison reasons (since this is the first time that some of these results are obtained directly from the pressure function) the escape rate g was obtained in two ways using Eq. (9) and from the pressure function -P(u_e,1) of Eq. (20) yielding consistent results. The same confirmation is also done for the average Lyapunov exponent, l, using Eq. (16) and -[(¶P/¶b)_ue]_{b = 1} from Eq. (20). Once the values of g and l are known the KS entropy and D_I can also be confirmed.

Observing Table (2) we conclude that the mean Lyapunov exponent as well as the two entropies increase as a function of u_e indicating that the system becomes more chaotic. These properties are also shown on the P-b-u_e surface plot of Fig 8(a) and (b). The two dimensions obtained by the pressure function are initially equal but later deviate when inelasticity is imposed, following the inequality D_H £ D_I (from d_I £ d_H). The two entropies increase as a function of u_e following (with in our precision) the inequality h_top ³ h_KS.

7  Conclusions

We have presented a study of the scaling properties of a scattering system of hard disks with inelasticity (pinball machine) in the finite horizon regime as well as in the infinite horizon regime for different number of scatterers. The two dimensional deflection function Q(b,u_e) exhibits Cantor-like fractal boundaries which are locally disconnected. These fractal boundaries are compressed exponentially as the inelasticity u_e increases. This lead to a scaling behaviour of the uncertainty fractal dimension which is consistent with an exponential dependence on u_e. For large enough number of scatterers the decay rate increases monotonically with u_e independently of the geometry and configuration of hard disks.

By generalizing the Sinai-Bunimovich curvature of the horocycle, known for elastic scattering, to the inelastic case we were able to study important properties of the chaotic inelastic scattering system, such as the Lyapunov exponent, the KS and the topological entropies. The increase of the Lyapunov exponents, the KS and the topological entropies as a function of u_e points to a relation between the compression of the singularities (the increase of the fractal dimension), and the measure of chaos in the pinball system. As u_e increases the uncertainty fractal dimension of the system approaches unity in an exponential manner, while the system becomes more chaotic. Further more this study raises questions on other similar systems that might be studied like for instance a system with randomly or periodically fluctuating u_e.

A  Derivation of the horocycle curvature

The horocycle is the geometrical front associated with the given trajectory. Given a reference particle we consider the front formed by accompanied particles initialized from the same point with different initial velocity angles. Around the initial point the front is circular with a radius which grows linearly with time so that the curvature is given by

Suppose we initialize a particle trajectory at point O_- which hits the hard disk at I,with a velocity u_I making an angle j_I with the normal at the impact point. On collision the disk exerts a radial force on the particle sending it away with an outgoing velocity u_I¢ making an angle j_I¢ with the normal. If the radius of the hard disk is a its curvature at I is 1/a. The incoming ray has a known curvature k_u^(-). Let the curvature of the outgoing ray be denoted by k_u⁽⁺⁾. We now show how to relate k_u⁽⁺⁾ to k_u^(-).

The center of the outgoing front curvature O₊ is found by considering another ray initialized from O_- making an infinitesimal angle d_- with respect to O_-I as seen in Fig. 9. This ray will be reflected at point J. The extrapolations of the reflected rays at J and I cross at O₊. Taking the front of the outgoing rays to be circular⁵ the curvature of the outgoing horocycle is given by k_u⁽⁺⁾ = |O₊I|^-1. Referring to Fig. 9 we have the following definitions: a (a¢) is the angle between O_-I  (O₊I) and the tangent to the disk at I; b (b¢) is the angle between O_-J  (O₊J) and the tangent to the disk at J; e is the angle at the center of the disk formed by the arc IJ; j_J  (j_J¢) is the angle between the incident (outgoing) ray at J. With these definitions using triangle identities we have

Since d_± and e are very small, in what follows, we will expand to first order in these quantities. For the angles using the relations given in Eqs. (24) we find

After successive collisions, combining the results of Eqs. (30) and (23) with r_k = 1/k_u⁽⁺⁾ we find for the curvature of the front between the kth and (k+1)th collisions the expression given in Eq. (13).

References

[1]

J. J. Lissauer, Rev. Mod. Phys. 71 (1999) 835.

[2]

J. Machta and R. Zwanzig, Phys. Rev. Lett. 50 (1983) 1959.

[3]

L. A. Buminovich and Ya. G. Sinai, Commun. Math. Phys. 78 (1980) 479.

[4]

P. Gaspard and S. A. Rice, J. Chem. Phys. 90 (1989) 2225.

[5]

T. Yalçinkaya and Y.C. Lai, Comp. in Phys. 9 (1995) 511.

[6]

A. Antillón, J. José and T.H. Seligman, cond-mat/9803512.

[7]

Y.C. Lai and C. Grebogi, Phys. Rev. E 49 (1994) 3761.

[8]

M. Ding, C. Grebogi, E. Ott and J.A. Yorke, Phys. Rev. A 42 (1990) 7025.

[9]

T. Tél, C. Grebogi and E. Ott, Chaos 4 (1993) 495.

[10]

V. Daniels and M Valliéres, Chaos 3 (1993) 475.

[11]

E. Ott and T. Tél, Chaos 3 (1993) 417; additional papers in the issue of Chaos 3 (1993).

[12]

C. Grebogi, S.W. McDonald, E. Ott and J.A. Yorke, Phys. Rev. Lett. 99A (1983) 415.

[13]

P. Gaspard and D.A. Ramirez, Phys. Rev. A 45 (1992) 8383.

[14]

P. Gaspard and G. Nicolis, Phys. Rev. Lett., 65 (1990) 1693.

[15]

P. Gaspard and F. Baras, Phys. Rev. E 51 (1995) 5332.

[16]

Y.T. Lau, J.M. Finn and E. Ott, Phys. Rev. Lett. 66 (1991) 978.

[17]

S. Bleher, C. Grebogi and E. Ott, Physica D 46 (1990) 87.

[18]

A. Beeker and P. Eckelt, Chaos 3 (1993) 487.

[19]

E. Ott, T. Sauer and J.A. Yorke, Phys. Rev. A 39 (1989) 4212.

[20]

C. GRebogi, E. Ott and J.A. Yorke, Phys. Rev. A 37 (1988) 1711.

[21]

S. Bleher, C. Ott and R. Brown, Phys. Rev. A 32 (1988) 930.

[22]

S. McDonald, C. Grebogi, E. Ott and J.A. Yorke, Physica D 17 (1985) 125.

[23]

P. Gaspard in Quantum Chaos, Proceedings of The International School of Physics ``Enrico Fermi", Course CXIX, Varenna, 1991, edited by G. Casati, I. Guarnri, and U. Smilansky (North-Holland, Amsterdam, 1993), pp. 307-383.

[24]

P. Gaspard and D.A. Ramirez, Phys. Rev. A 45, 8383 (1992).

[25]

P. Szepfalusy and T. Tél, Phys. Rev. A 32, 2520 (1986).

Footnotes:

¹ alexand@pauli.ns.ucy.ac.cy

² kypriano@buffalo.edu

³ An analytic derivation of such a scaling law is in general not possible [16,18,10]. There are however a few examples in chaotic scattering that the scaling function can be described by analytical relations [17,20,19].

⁴ The more disks the particle hits the more momentum it gains. With more momentum the particle will hit more disks for a given time, getting even faster. Thus we expect the rate of change of the velocity of the particle, for the singular values of b, to be given by ¶_t u(t) = bu(t) or u(t) = ce^bt where b = f(u_e).

⁵ The front is not strictly circular since now the outgoing rays from I and J have velocities which differ in magnitude leading to a correction which we neglect for small d_-.