Adaptive and accurate force-based QM/MM calculations

The QM/MM method, which couples a quantum mechanical (QM) description of bonding in a localized region to a larger molecular mechanics (MM) simulation, is widely used for simulations of proteins [1-5]. The combination allows it to benefit from the strengths of both QM and MM approaches. The QM description gives accuracy on bond-breaking energetics and structures where it is needed, but it is too computationally expensive to be used beyond a few hundred atoms at most. The MM description is much faster computationally and makes it possible to describe large-scale structure and motion, solvation and entropic effects. In its usual formulation, the set of atoms described by QM is fixed during the simulation, but under some circumstances this requirement can be overly restrictive. One possible scenario is if protein residues move over a distance larger than the feasible QM region size or if describing the reaction requires reactants, products, or solvent molecules to be transported over large distances into or out of the QM region. Another possible scenario is if a large QM region, including one or more solvation shells around a reaction site, is needed simply for accuracy. If such a large QM region is needed, solvent molecules near its edge may diffuse away, and without adaptivity the diffusion would lead to a fragmented QM region. Instead of using adaptivity it would be possible to keep the QM molecules confined using some sort of ad hoc restraining potential, but that would require individual tuning for each system to maintain correct density in the QM region. The simulations we present here pertain to this latter situation