Critical Behaviour of the Drossel-Schwabl Forest Fire Model

Abstract

We present high statistics Monte Carlo results for the Drossel-Schwabl forest fire model in 2 dimensions. They extend to much larger lattices (up to $65536\times 65536$) than previous simulations and reach much closer to the critical point (up to $\theta \equiv p/f = 256000$). They are incompatible with all previous conjectures for the (extrapolated) critical behaviour, although they in general agree well with previous simulations wherever they can be directly compared. Instead, they suggest that scaling laws observed in previous simulations are spurious, and that the density $\rho$ of trees in the critical state was grossly underestimated. While previous simulations gave $\rho\approx 0.408$, we conjecture that $\rho$ actually is equal to the critical threshold $p_c = 0.592...$ for site percolation in $d=2$. This is however still far from the densities reachable with present day computers, and we estimate that we would need many orders of magnitude higher CPU times and storage capacities to reach the true critical behaviour -- which might or might not be that of ordinary percolation.