Multiscale algorithms for molecular dynamics

We seek to develop more efficient algorithms for highly accurate, atomistic molecular dynamics (MD), and for more approximate MD as well. Presently, when tackling slow processes, scientists often resort to ad hoc procedures that have neither error control nor predictive abilities. MD, on the other hand, has been severely limited in what it can study, due in great part to the small time steps that have to be used when including all atomic interactions in a large molecule. Time steps of femtoseconds are used, when processes of interest are often in the range of microseconds or seconds. Our procedure attempts to use MD to estimate average values of motions that are of secondary interest, and incorporate them into a scheme that can predict the motions that are of real interest.