We describe our studies of two methods for generating bias potentials for use in hyperdynamics simulations. In the first method, first reported by Steiner, et al, the bias potential is the additional energy needed to keep the total potential energy above a fixed level. This potential exerts a negligible computational load and is very easy to code. The second technique involves computing the iterative Hessian-based bias potential as usual, but including in those calculations only the atoms in a small “active region” that surrounds the area in which a state transition is expected to occur. This “subspace hyperdynamics” method is less costly than full HD. The extra computational effort required for HD scales as the size of the active region(s), instead of with the size of the entire simulation domain.

We have carried out extensive tests of both methods on two problems, the diffusion of adatoms on silver surfaces and the migration of vacancies in bulk silver. Using the first method, to obtain large boosts the bias potential must be so high that many transitions are prevented from occurring. Our results for subspace hyperdynamics are promising; we obtain the same results as with full hyperdynamics, but with considerably less computational effort.