Drift and trapping in biased diffusion on disordered lattices

We reexamine the theory of transition from drift to no-drift in biased diffusion on percolation networks. We argue that for the bias field B equal to the critical value B_c, the average velocity at large times t decreases to zero as 1/log(t). For B < B_c, the time required to reach the steady-state velocity diverges as exp(const/|B_c-B|). We propose an extrapolation form that describes the behavior of average velocity as a function of time at intermediate time scales. This form is found to have a very good agreement with the results of extensive Monte Carlo simulations on a 3-dimensional site-percolation network and moderate bias.