Associate Professor Frank Jensen

Department of Chemistry, University of Aarhus

Molecular dynamics simulations of biomolecular systems have become widely used tools for obtaining information, especially in drug discovery research, by providing a link between structure and function. Such simulations often treat systems with tens or hundreds of thousand atoms at the nano- or micro-second timescale. Given that the fundamental timestep is a few femto-seconds, this implies that only parameterized energy functions are realistic, and these are usually denoted force fields. Most force fields in common use were designed several decades ago and have essentially only changed by parameter updates. As computational resources have improved, it is increasingly becoming clear that the errors arising from the force field may be larger than the errors from the sampling in the time-dimension. A key limitation in common force fields is the lack of electric polarization, and modelling this is an active research area. A related issue is that force fields have a massive redundancy in the parameter space, and the many years of parameter refinements have built in lots of error cancellations, making it difficult to improve existing force fields. Given that accurate electronic structure methods are now possible for quite large molecules, perhaps it is time to rethink how force fields are designed and parameterized, and perhaps make a fresh start

Recording: