ABSTRACT
From pioneering in vacuo, picosecond-timescale investigations of
proteins [1], atomistic simulations have gradually matured into a
scientific workhorse for (bio)molecular systems [2-9]. By averaging
over the electronic degrees of freedom, atomistic models idealize
the system by using a set of empirical interactions potentials [10-
13]. Though approximate, atomistic force fields have increasingly
become exquisitely finely tuned to reproduce ab initio and ex-
perimental properties [14-17]. Recent developments in the field
have highlighted their many successes, e.g., insight and predictions
in drug discovery [18], accurate thermodynamic calculations of
organic molecules [19], or beyond-microsecond-timescale protein
simulations [20]. Yet, atomistic simulations have also revealed their
limitations. Recent access to powerful computers exhibits force-
field inaccuracies that have long-time, spurious repercussions [21].
Beyond the quality of the parametrization of the potential energy
surfaces (PESs), their functional forms are based on crucial assump-
tions. Most current-generation force fields represent intermolecular
interactions via pairwise Lennard-Jones interactions and point-
charge (PC) electrostatics [14-17]. For instance, polarizable force
fields, which reproduce the response of a charge distribution to a
local change in the electric field, have become increasingly popular
for key systems [22], e.g., cation-π interactions [23] only recently.
