We use the term "small core" datasets to refer to sets constructed with at least 2 occupied bound states included in the basis set for a given angular momentum quantum number l. For example, a "small core" configuration of Cu, would include the following orbitals in the core: 1s 2s 2p, and the folowing orbitals in the valence: 3s 4s 3p 3d. In general the full basis set would include additional continuum basis functions in the l=1,2 channels. There are several advantages for using the small core basis sets. For example, for most small core configurations, it is relatively easy to find combinations of parameters to generate robust datasets without interference of ghost states. These datasets generally give excellent results for both elemental solids and ionic compounds. On the other hand, because both semicore and valence states are treated self-consistently, calculations using these datasets tend to be computationally demanding.

We use the term "large core" datasets to refer to sets constructed with at most 1 occupied bound state i included in the basis set for a given angular momentum quantum number l. For example, a "large core" configuration of Cu, would include the following orbitals in the core: 1s 2s 3s 2p 3p, and the folowing orbitals in the valence: 4s 3d. In general the full basis set would include additional continuum basis functions in the l=0,1,2 channels. For large core configurations, it is sometimes difficult to find ranges of parameters that generate datasets without ghost states. The advantage of large core datasets are they are designed to focus the computational effort on the important valence states of the system, in line with the original notions of pseudopotentials.