Role of forms of exchange and correlation used in generating pseudopotentials

We generate pseudopotentials using various treatments of exchange and correlation and test the pseudopotentials both for physical predictions that they make (with quantum Monte Carlo many-body calculations for the valence electrons) and for transferability. The calculated results for physical quantities (e.g., binding energies, ionization potentials, molecular dissociative energies, and bond lengths) are compared with each other and experiment for monatomic sodium, potassium, calcium, scandium, titanium, and silicon, and for diatomic sodium, potassium, and silicon. We find that pseudopotentials generated using Hartree-Fock exchange in conjunction with local-density correlation are more transferable and yield better physical ionic properties than those generated using either local-density exchange-correlation or pure Hartree-Fock exchange. For critical atoms like chromium and nickel we attribute the better transferability to the absence of the nonlinearity problem associated with local exchange. In particular, we find marked improvement in the 3d energies for calcium, scandium, and titanium. Systematically obtaining better pseudopotentials may require a many-body treatment of correlation effects in the full-atomic configurations from which the pseudopotentials are generated.