Improved algorithm for geometry optimisation using damped molecular dynamics

We present a new algorithm for optimising the configuration of a system of particles using damped molecular dynamics. This new algorithm has a much faster rate of convergence to the ground-state structure than other MD-based schemes. We also show how it is possible to use this algorithm to select dynamically the optimum time step or to precondition the masses for the modified dynamics, so as to produce additional significant efficiency gains especially as convergence is approached. The performance of the new algorithm is illustrated with several model systems and also by relaxing a defect in bulk silicon. The defect calculation was also relaxed using a standard BFGS algorithm and the performance was found to be comparable. (C) 2003 Elsevier Science B.V. All rights reserved.