| Abstract
| - One of the practical difficulties precluding the generalized development of nonadditive, polarizable modelsfor statistical simulations is rooted in the costly estimation of accurate induction energies, from which distributedpolarizabilities can be derived. From a finite perturbation (FP) perspective, mapping the induction energyover a grid of points implies as many distinct quantum chemical calculations of the molecule interacting witha polarizing charge as the total number of points. Here, two alternative routes for computing accurate inductionenergies in a time-bound fashion are explored. The first one is based upon second-order perturbation theoryand only involves a single quantum chemical calculation at the Hartree−Fock level of approximation to mapthe induction energy. The second one, less straightforward in its implementation, relies on a topologicalpartitioning of the response charge density, also evaluated from a single quantum chemical calculation, yetat virtually any level of sophistication. Critical comparison with reference FP computations reveals that onlyappropriate scaling of the perturbative (PT) induction energies can warrant a faithful description of polarizationphenomena. In the case of neutral molecules, a reasonable reproduction of molecular dipole polarizabilitiesis achieved when use is made of a simple scaling function that solely depends on the distance separating thepoints of the grid from the center of mass of the molecule. For anions, the marked anisotropy in the deviationof the PT induction energies from the target FP ones makes the definition of such a general scaling functionvirtually impossible. In sharp contrast, the approach based upon the topological partitioning of the responsecharge density does not require any adjustment or scaling, and, thus, constitutes a more robust and rigorousstrategy for the computation of induction energies. Examination of distinct protocols for mapping the inductionenergy emphasizes the necessity to sample the space around the molecule far enough from the nuclei toreproduce molecular dipole polarizabilities accurately. Compared to the spatial extent of the grid, the densityof points appears to be of lesser importance.
|
