Common molecular dynamics algorithms revisited: Accuracy and optimal time steps of Stormer-leapfrog integrators

The Stormer-Verlet-leapfrog group of integrators commonly used in molecular dynamics simulations has long become a textbook subject and seems to have been studied exhaustively. There are, however, a few striking effects in performance of algorithms which are well known but have not received adequate attention in the literature. A closer view of these unclear observations results in unexpected conclusions. It is shown here that contrary to the conventional point of view, the leapfrog scheme is distinguished in this group both in terms of the order of truncation errors and the conservation of the total energy. In this case the characteristic square growth of fluctuations of the total energy with the step size, commonly measured in numerical tests, results from additional interpolation errors with no relation to the accuracy of the computed trajectory. An alternative procedure is described for checking energy conservation of leapfrog-like algorithms which is free from interpolation errors. Preliminary tests on a representative model system suggest that standard step size values used at present are lower than necessary for accurate sampling. (C) 1997 Academic Press.