Long-time convergence of an Adaptive Biasing Force method: the bi-channel case (1005.0206v1)

Abstract: We present convergence results for an adaptive algorithm to compute free energies, namely the adaptive biasing force (ABF) method. The free energy is the effective potential associated to a so-called reaction coordinate (RC). Computing free energy differences remains an important challenge in molecular dynamics due to the presence of meta-stable regions in the potential energy surface. The ABF method uses an on-the-fly estimate of the free energy to bias dynamics and overcome metastability. Using entropy arguments and logarithmic Sobolev inequalities, previous results have shown that the rate of convergence of the ABF method is limited by the metastable features of the canonical measures conditioned to being at fixed values of the RC. In this paper, we present an improvement on the existing results, in the presence of such metastabilities, which is a generic case encountered in practice. More precisely, we study the so-called bi-channel case, where two channels along the RC direction exist between an initial and final state, the channels being separated from each other by a region of very low probability. With hypotheses made on `channel-dependent' conditional measures, we show on a bi-channel model that we introduce, that the convergence of the ABF method is in fact not limited by metastabilities in directions orthogonal to the RC under two crucial assumptions: (i) exchange between the two channels is possible for some values of the RC and (ii) the free energy is a good bias in each channel.