Grid-based energy density analysis: implementation and assessment.

Grid-based energy density analysis (grid-EDA) that decomposes the total energy into atomic energies by a space-partitioning function is proposed. The kinetic energy, nuclear attraction, and exchange-correlation functional are evaluated on grid points and are split into atomic contributions. To reduce numerical errors in the conventional scheme of numerical integration, the electronic Coulomb and HF exchange interactions are evaluated by the pseudospectral method, which was first applied to an ab initio method by Friesner [Chem. Phys. Lett. 116, 39 (1985)], and are decomposed into atomic contributions. Grid-EDA using the pseudospectral method succeeds in ensuring less than 1 kcalmol error in total energies for small molecules and providing reliable atomic energy contributions for the problematic lithium cluster, which exhibits a strong basis-set dependence for Mulliken-type EDA. Also, site-dependent atomization energies are estimated by grid-EDA for cluster models such as Li(48), C(41)H(60), and Mg(32)O(32). Grid-EDA reveals that these models imitate crystal environments reasonably because atomization energies estimated from the inner atoms of the models are close to the experimental cohesive energies.