Numerical integrators for the hybrid Monte Carlo method (Q6486744)

A methodology is given for constructing efficient methods for the numerical integration of the Hamiltonian differential equations that arise in a hybrid Monte Carlo method and related simulations. The approach is based on optimizing the behavior of a function over a relevant range of values of the step-length. New split-step integrators are constructed with two, three or four function evaluations per time-step It is claimed that the splitting formulae suggested here are more efficient than the standard Verlet method, especially if the number of dimensions is high.