Improving the Accuracy of Quantum Mechanics/Molecular Mechanics (QM/MM) Models with Polarized Fragment Charges

This paper introduces an economical approach for improving
the
accuracy and convergence of quantum mechanics/molecular mechanics
(QM/MM) models. The approach is tested on a series of neutral and
charged amino acids embedded in a 160-water cluster, where their intramolecular
proton transfer energies (neutral amino acid → zwitterionic
amino acid) were previously obtained at the ωB97X-D/6-31G­(d)
level of theory. When the charges on the MM atoms were replaced with
those obtained at the same QM level of theory used to treat the QM
atoms, this significantly improved the accuracy and convergence of
the QM/MM models. In particular, the QM/MM model converged to within
1.4 kcal mol−1 of directly calculated DFT energies
for smaller (by as many as 20 waters) QM regions. The use of atomic
charges obtained from the natural population analysis yielded the
most significant improvement, while other charge schemes such as Mulliken,
electrostatic potential, or CM5 led to poorer outcomes. It is further
demonstrated that the QM atomic charges can be accurately estimated
in a highly efficient manner using an iterative fragmentation approach
based on the moving-domain QM/MM method. Similar observations were
made when the approach was used to predict the barrier of an SN2 reaction. Thus, the use of QM-quality atomic charges on
MM atoms represents a simple and easy-to-implement strategy for improving
the accuracy of QM/MM models.