Correlation consistent pseudopotentials for quantum Monte Carlo and other many-body electronic structure methods: 2nd row elements

We report on construction and benchmarking of a set of pseudopotentials (PPs) for high accuracy electronic structure calculations for 2nd row atoms. Despite including cases often considered "easy", such as Si, this row poses unique difficulties. In fact, there are number of challenges that stem from the customary Ne-core partition due to the small number of valence electrons and large cores. We found especially that ionic bonds (such as oxides) can exhibit significant errors with respect to all-electron CCSD(T) results, in particular, for non-equilibrium conformations. This applies not only to Na and Mg, but to some extent even to Al and Si. For example, all existing PPs underestimate the repulsive side of polar bonds (say, for systems such as MgO) by tenths of eV. We therefore map-out the possibilities for Ne-core PPs and discuss their accuracy limits. Furthermore, we investigate the possibility of improving PPs by a repulsive term that eliminates this deficiency with little additional computational cost. Finally, for cases where none of the above leads to satisfactory accuracies we construct He-core ECPs that reproduce energetics (spectra and molecular binding curves) with typical errors within 0.003 eV.