How to ensure the accuracy of polarizable force fields?

Toon Verstraelen, Steven Vandenbrande, and Paul W. Ayers, Ghent University, Zwijnaarde, Belgium

In many force-field simulations, host-guest interactions determine the relevant results, e.g. the spatial distribution of guest molecules in a porous material, or the binding free energy of a ligand in an enzyme. Because the accuracy of non-covalent interactions is so critical, there is a continued interest in explicit treatments of electronic polarization in force fields since the early nineties. These so-called polarizable force fields (PFFs) seek to improve the accuracy of non-covalent interactions with a modest extra computational cost. The mutual electronic polarization of two compounds leads to an attractive force that is unattainable with few-body terms. Despite intensive research, the development of a quantitatively accurate PFF is still a very complex and labor-intensive procedure: parameters are calibrated empirically with diverse sets of reference data, including hydration energies, NMR data and ab initio data.

To facilitate the development of accurate PFFs, we propose to eliminate the empirical factors and to construct these models rigorously from an ab initio description of the electronic linear response. To this end, we recently derived “Atom-Condensed Kohn-Sham DFT approximated to Second order” (ACKS2) and showed that classical PFFs are lacking important non-local contributions due to the electronic kinetic energy. In this work, we generalize the theoretical foundations of the ACKS2 model, building on other theories than KS-DFT and including atomic multipoles to arbitrary order. We validate our approach numerically, showing that the generalized ACKS2 model reproduces ab initio reference data without any empirical calibration of parameters.