1: \begin{abstract}

2: 

3: 

4: We investigate deterministic diffusion in periodic billiard

5: models, in terms of the convergence of rescaled distributions to

6: the limiting normal distribution required by the central limit

7: theorem; this is stronger than the usual requirement

8: that the mean square displacement grow asymptotically linearly in

9: time. The main model studied is a chaotic Lorentz gas where the

10: central limit theorem has been rigorously proved. We study

11: one-dimensional position and displacement densities describing the

12: time evolution of statistical ensembles in a channel geometry,

13: using a more refined method than histograms.  We find a

14: pronounced oscillatory fine structure, and show that this has its

15: origin in the geometry of the billiard domain. This fine structure

16: prevents the rescaled densities from converging pointwise to

17: gaussian densities; however, demodulating them by the fine

18: structure gives new densities which seem to converge uniformly. We

19: give an analytical estimate of the rate of convergence of the

20: original distributions to the limiting normal distribution, based

21: on the analysis of the fine structure, which agrees well with

22: simulation results. We show that using a Maxwellian (gaussian)

23: distribution of velocities in place of unit speed velocities does

24: not affect the growth of the mean square displacement, but changes

25: the limiting shape of the distributions to a non-gaussian one.

26: Using the same methods, we give numerical evidence that a

27: non-chaotic polygonal channel model also obeys the central limit

28: theorem, but with a slower convergence rate.

29: 

30: \end{abstract}

31: