This PR fixes #77 and #78. Thanks to @bruneval for reporting these.
In #77, the Dij was normalized by the magnitude of the nonlocal channel. For practical reasons, we were using a zeroed-out channel for He, Li_reg, and Be_reg which resulted in 1/0. Now, this is corrected by setting Dij = 0.
All *.upf and *.rpt files have been regenerated due to incorrect grid size as noted in #78.
Additional Changes
I tested all the upf files by comparing Pyscf/PBE vs QE/PBE energies and added this data as energies.txt.
All cases agree within a few mHa. ECPs which didn't agree within 3 mHa due to too high KE cutoff were indicated as deprecated.
Currently, this covers H-Kr. I will make a separate PR for the newer elements.
Changes
This PR fixes #77 and #78. Thanks to @bruneval for reporting these.
In #77, the Dij was normalized by the magnitude of the nonlocal channel. For practical reasons, we were using a zeroed-out channel for He, Li_reg, and Be_reg which resulted in 1/0. Now, this is corrected by setting
Dij = 0.All
*.upfand*.rptfiles have been regenerated due to incorrect grid size as noted in #78.Additional Changes
I tested all the upf files by comparing Pyscf/PBE vs QE/PBE energies and added this data as
energies.txt. All cases agree within a few mHa. ECPs which didn't agree within 3 mHa due to too high KE cutoff were indicated as deprecated.Currently, this covers H-Kr. I will make a separate PR for the newer elements.