A note on pseudopotentials in CASTEP

Pseudopotentials are used by CASTEP to evaluate the energy of the core electrons of each atom. They are a very cheap way of doing this, and have improved over the years. However, this improvement means that they are different with each version of CASTEP, and so minimizations carried out with older versions of CASTEP are not compatible with those carried out with the latest versions.

The problem was first noted when Louise came to extend some of Rui's work, and the energies were not compatible. It is possible to request different pseudopotentials by altering the commented out section in the .cell file. Normally the section looks like:

# Commented out pseudopotential block for easy editing
#%BLOCK SPECIES_POT
# H   H_00.usp
# C   C_00.usp
# O   O_00.usp
# N   N_00.usp
#%ENDBLOCK SPECIES_POT

In this case, if the section were uncommented, the calculation would use the same type of potential for each atom, but it should be clear how different types of potential could be used for different atoms. It is also possible to just have one line in this block, containing the potential that you want to use for all atoms. This is what was done in testing the effect of the different available potentials.

The following Table shows the discrepancies. All calculations were carried out with CASTEP version 23.1 on ARCHER2, in September 2023, except for the second line, which was carried out with CASTEP version 16.11 on ARCHER, in July 2016. With the exception of the first and second rows, all calculations were single point energy calculations of Oxcarbazepine A20 (and files are stored in /space/slp/lsp/Archer_from_Louise/Carbamazepine_family/Oxcarbazepine/A20/potential_tests). A plane wave basis set cutoff of 1100 eV and a kpoint spacing of 0.05 was used. The potential type was set as the value in the "Potential" column.

The energy of the atomic cores was subtracted from the Unit cell energy, and the remaining energy was due to the covalent bonding and any intermoleular interactions. The differences between this energy and that calculated in the first row are reported. The full calculations are available in Psudopotentials.xlsx

In the above table, the first row was the calculation with the default pseudopotential in CASTEP version 23.1. The energy was taken from the first point of a geometry optimization, so there is no correction for the finite basis set. The second row was the calculation with the default pseudopotential in CASTEP version 16.11. The energy was taken from the final point of a geometry optimization, so there is no correction for the finite basis set.

Since the single point energies are calculated at 1090 eV and then refined at 1095 and 1100 eV, it is likely that this accounts for the difference between the energy in the first row and that with the C18 and C19 pseudopotentials. However, the difference between the end point of the CASTEP version 16.11 calculation (which was used as the starting point for all these calculations) and the calculation with any other pseudopotential is larger than would be expected. The difference in the N atom energy (already accounted for when calculating the energy differences) is somewhat troubling - it does not match any of the other pseudopotentials. The starting point for the C8, C9 and C17 calculations was generated by converting the output .cell to .res, running Platon to get a Z'=1 structure, and running cif2cell on that again. This may account for the discrepancy. If it does, that is still very big though.

Both the ultrasoft pseudopotential and the norm-conserving pseudopotential have been consistent in versions 18, 19 and the current versions of CASTEP, and the pseudopotentials in use in versions 8, 9 and 17 were also consistent.

More information is available in the CASTEP manual.

© UCL Chemistry Department 2022. This page was last updated on 17 August, 2022. If you have any problems with this page please email the WebMaster