Search

We present eight new parameterizations of the SM5.42R solvation model: in particular we present parameterizations for HF/MIDI!, HF/6-31G*, HF/6-31+G*, HF/cc-pVDZ, AM1, PM3, BPW91/MIDI!, and B3LYP/MIDI!. Two of the new cases are parameterized using the reaction-field operator presented previously, and six of the new cases are parameterized with a simplified reaction-field operator; results obtained by the two methods are compared for selected examples. For a training set of 2135 data for 275 neutral solutes containing H, C, N, O, F, S, P, Cl, Br, and I in 91 solvents (water and 90 nonaqueous solvents), seven of the eight new parameterizations give mean unsigned errors in the range 0.43–0.46 kcal/mol, and the eighth – for a basis set containing diffuse functions – gives a mean unsigned error of 0.53 kcal/mol. The mean unsigned error for 49 ionic solutes (containing the same elements) in water is 3.5–3.9 kcal/mol for the Hartree–Fock, Becke–Perdew–Wang-1991 and Becke three-parameter Lee–Yang–Parr cases and 4.1 and 4.0 kcal/mol for parameterized model 3 and Austin model 1, respectively. The methods are tested for sensitivity of solvation free energies to geometry and for predicting partition coefficients of carbonates, which were not included in the training set.

The SM5.42R solvation model, which is based on SM5-type atomic surface tensions, class IV point charges based on charge model 2, and rigid geometries, is parametrized for the intermediate neglect of differential overlap for spectroscopy (INDO/S) method, both in the original (INDO/S) and more recent (INDO/S2) versions. The parametrization is based on 2184 free energies of solvation of 275 neutral solutes and 49 ions in water and 90 organic solvents.

Analytical energy gradients have been derived for an SM5-type solvation model based on Hartree–Fock self-consistent reaction-field theory and CM2 atomic charges. The method is combined with an analytic treatment of the first derivatives of nonelectrostatic first-solvation-shell contributions to the free energy and implemented in the General Atomic and Molecular Electronic Structure System (GAMESS). The resulting equations allow one to use accurate class IV charges to calculate equilibrium geometries of solutes in liquid-phase solutions. The algorithm is illustrated by calculations of optimized geometries and solvation free energies for water, methanol, dimethyl disulfide, and 9-methyladenine in water and 1-octanol.

We extend the SM5 solvation model for calculating solvation free energies of a variety of organic solutes in both aqueous and organic solvents so that it can be employed in conjunction with high-level electronic structure calculations. The extension is illustrated by presenting three implementations based on density-functional theory (DFT). The three implementations are called SM5.42R/BPW91/MIDI!6D, SM5.42R/BPW91/DZVP, and SM5.42R/BPW91/6-31G*. They have the following features: (1) They utilize gradient-corrected DFT with polarized double zeta basis sets to describe the electronic structure of a solute. The particular exchange-correlation functional adopted is Becke’s exchange with the Perdew–Wang 1991 correlation functional, usually called BPW91. The MIDI!6D, DZVP, and 6-31G* basis sets are used. (2) They employ fixed solute geometries in solvation calculations. The model is designed to predict solvation free energies based on any reasonably accurate gas-phase solute geometry. (3) The electric polarization in the solute-solvent system is described by the generalized Born approximation with self-consistent reaction-field solute partial atomic charges obtained from the CM2 class IV charge model. (4) The solvation effects within the first solvation shell are included in the form of SM5-type atomic surface tensions. Both DFT parameterizations are developed using 275 neutral solutes and 49 ions with gas-phase Hartree–Fock/MIDI! geometries. These solutes contain a wide variety of organic functional groups which include H, C, N, O, F, P, S, Cl, Br, and I atoms. For 2135 free energies of solvation of the neutral molecules in water and 90 organic solvents, SM5.42R/BPW91/MIDI!6D, SM5.42R/BPW91/DZVP, and SM5.42R/BPW91/6-31G* yield mean unsigned errors in solvation free energies of 0.45 kcal/mol, 0.44 kcal/mol, and 0.43 kcal/mol, respectively. For 49 ions in water, SM5.42R/BPW91/MIDI!6D produces a mean unsigned error of 3.9 kcal/mol, while SM5.42R/BPW91/DZVP and SM5.42R/BPW91/6-31G* give 3.6 kcal/mol and 3.9 kcal/mol, respectively.

We propose a new formalism, Charge Model 2 (CM2), to obtain accurate partial atomic charges from a population analysis of wave functions by a parametrized mapping procedure, so that the resulting charges reproduce highly accurate charge-dependent observables. The new method, which produces class IV charges, is illustrated by developing CM2 mappings of Lowdin charges obtained from semiempirical and ab initio Hartree−Fock theory and density functional theory, in particular AM1, PM3, HF/MIDI!, HF/6-31G*, HF/6-31+G*, BPW91/MIDI!, BPW91/6-31G*, B3LYP/MIDI!, and BPW91/DZVP calculations. The CM2 partial charges reproduce experimental dipole moments with root-mean-square errors that are typically a factor of 7 better than dipole moments computed from Mulliken population analysis, a factor of 3 better than dipole moments computed by Lowdin analysis, and even a factor of 2 better than dipole moments computed from the continuous electron denisty. At the HF/6-31G* and B3LYP/MIDI! levels, the new charge model yields r...

The divide-and-conquer density-functional approach is implemented for electronic structure calculations of clean surfaces and surface adsorption. The method divides a semi-infinite solid-state system into subsystems and determines each subsystem density separately. We have applied the method to study two low index Li surfaces and the chemisorption of a monolayer of atomic hydrogen on the Li(100) surface. Surface and adsorption energies converge well with respect to the number of buffer atoms and surface layers. Our converged energies are in very good agreement with accurate theoretical and available experimental results.

Self-consistent Kohn–Sham density functional calculations have been carried out to study the structure of the ammonia dimer. The local-density approximation yields unusually large binding energy and short internitrogen distance compared with the experimental and more accurate theoretical data. The results from the Becke–Perdew gradient-corrected functionals are generally in good agreement with those at the SCFMP2 level when the geometry is fully optimized with various large basis sets. With our best estimation, the staggered quasi-linear structure (Cs) is 0.6 kcal/mol lower in energy than the symmetric cyclic one (C2h). The hydrogen-bonded N—H bond in the staggered quasi-linear structure is found to be 0.008 A longer than the N—H bond in ammonia. In our calculations, we could not find the minima on the energy surface corresponding to the two asymmetric cyclic structures suggested by microwave spectra and coupled pair functional calculations. © 1994 John Wiley & Sons, Inc.

Several local and nonlocal exchange‐correlation functionals are tested through accurate Kohn–Sham self‐consistent calculations for six nobel gas atoms and three isoelectronic series (with two, four, and ten electrons and atomic numbers up to 20). The calculated exchange, correlation, and total energies have been compared with those from accurate Hartree–Fock calculations and accurate estimates of correlation energies available recently. Improvements over local‐density approximation are found for all the tested nonlocal exchange‐correlation functionals. In particular, nonlocal correlation energy functionals are shown to perform significantly better than the local approximations when both are used in conjunction with the nonlocal Becke’s exchange energy functional. However, in most cases, all the functionals fail to predict the correct trends of exchange and correlation energies as the atomic charge increases.

The first-principles divide-and-conquer density-functional approach has been extended to solid-state systems. The method has the following features. (1) It divides a periodic solid-state system into equivalent primitive cells and further divides each cell into subsystems. The electron density of each subsystem is determined through the local representation of the one-electron Hamiltonian and used to form the total density per primitive cell. The method calculates the electronic structure of solids without involving the reciprocal space and its associated band structure. (2) It uses numerical atomic orbitals as basis functions with great variational flexibility. The Hamiltonian and other matrix elements are evaluated by numerical integration without any shape approximation to the effective one-electron potential. (3) This method, based on real space partition, can be applied to extended solid-state systems without translational symmetry, such as defects and surface chemisorption. As the first step, we have applied and tested the method to the electronic structure calculations of various crystalline solids: metallic lithium and copper, ionic sodium chloride, and covalent diamond and silicon. The self-consistently computed cohesive energies, structural properties, and density of states are in good agreement with those from the local-density approximation band-structure calculations and experimental results.