Density Functional Theory (DFT) has reshaped the field of computational chemistry over the past decades, which is mainly the merit of the attractive trade-off between the efficiency of DFT computations and the relative accuracy of the DFT potential energy surface with modern functionals. Despite these benefits, DFT computations are not always fast enough, especially when facing sampling problems on systems with many degrees of freedom, e.g. docking of a ligand on a protein surface. It is not only the scaling of the computational cost with system size, but also the increase in complexity that requires more samples and hence more computing time. There are two major approaches to reduce the burden of such sampling problems: (i) smarter sampling algorithms that extract the same information from less samples and (ii) faster methods to compute the potential energy of a molecular system. The latter option is the topic of this paper.
Force fields are the fastest models to evaluate the potential energy (and nuclear forces) for a given molecular geometry. They are unfortunately also known for their limited accuracy and the inability to make and break chemical bonds during a simulation. Another major roadblock is the determination of force-field parameters for new systems, due to the lack of systematic calibration procedures. In this paper we present our recent methodological advances to surmount the typical weaknesses of (polarizable) force fields.