Pseudopotentials and Plane Wave Cutoff

The selection of proper pseudopotentials and the corresponding plane wave cutoff are crucial for obtaining good results with a CPMD calculation. The cutoff required is mainly determined by the type and the ``softness'' of a pseudopotential.
Ideally a pseudopotential for a specific atom type should be usable for all kinds of calculations (Transferability), but in practice one frequently has to make compromises between accuracy and impact on the computational effort when creating a pseudopotential. Therefore one always has to test pseudopotentials before using them on new systems. There are quite a large number of CPMD calculations published (see http://www.cpmd.org/cpmd_publications.html) which can serve as a guideline.

Since CPMD uses a plane wave basis, a concept of several different, formalized basis sets of different quality, like with gaussian basis set based quantum chemical software, does not apply here. Since plane waves are 'delocalized' in space, they provide the same 'quality' everywhere in space and one can increase the basis set almost arbitrarily by increasing the number of plane waves via the CUTOFF keyword. The cutoff has to be chosen in such a way, that all required quantities are reasonably converged with respect to the plane wave cutoff. For a molecular dynamics run this refers primarily to the atomic forces. For the calculation of other properties like the stress tensor a different, usually a much higher cutoff is required. It's always a good idea to make checks at some critical points of the calculations by increasing the cutoff.

Typical cutoff values range from 20-40 ry for Vanderbilt ultra-soft pseudopotentials, 60-100 ry for Troullier-Martins norm-conserving pseudopotentials to 80-200 ry for Goedecker pseudopotentials. Pseudopotentials of different types can be freely mixed, but the required plane wave cutoff is determined by the ``hardest'' pseudopotential. Support for Vanderbilt ultra-soft pseudopotentials is (mostly) limited to the basic functionality like molecular dynamics and geometry optimization.