Add support for PseudoDojo PAW pseudos (alternative) #42
Conversation
|
An even safer alternative suggested by @sponce24 is to set the unknown cutoffs to 1.5 ~ 2.0 times the maximum in the same stringency. |
aiida_pseudo/cli/install.py
Outdated
| max_cutoff_rho = max([str_cutoffs[element]['cutoff_rho'] for element in str_cutoffs]) | ||
| for element, cutoff in str_cutoffs.items(): | ||
| if cutoff['cutoff_wfc'] <= 0: | ||
| cutoffs[stringency][element]['cutoff_wfc'] = max_cutoff_wfc * 2.0 |
There was a problem hiding this comment.
If we are already taking the maximum of the stringency set, should we really even double it? Can't find your comment anymore about what the PD team recommended about cutoffs for lanthanides. I have the feeling this is a bit excessive. If we give a warning that they are not officially tested, maybe just using the max itself should be fine?
There was a problem hiding this comment.
They recommended 20 Ha as a rule of thumb, but they also mentioned that the density (PAW double grid) cutoff may need to be larger than what is generally required. As the elements that are missing are only the Lanthanides, it could make sense to do something "too strict" to be safe.
There was a problem hiding this comment.
ok, but what is the max cutoff in the strict set? Is that more or less than 20 Ha. Because it if iis more, and we times two that, it will be a lot more than 20 Ha, so maybe a bit excessive.
There was a problem hiding this comment.
I think the max in the strict set is 25, so I agree that 2.0 would be a bit extreme. Perhaps a lower multiplier would be better.
…ssage, and change NC dual to 4.0
|
Note: I've changed the |
|
@sphuber I can't remember which solution we decided upon in the end, and Marnik has run into an issue with the incorrect dual value which was fixed in this pull. Could we think about merging one of the solutions? |
|
We agreed on this variant which indeed already fixes the dual factor. The only thing remaining was the behavior for setting recommended cutoffs for those that don't have one. I thought that twice the max may be a bit much (see the only remaining open comment). Should we set the factor to 1.5 ? If you agree and update, then I will merge |
…f for the same stringency
Fixes #40
Alternative to PR #41
The range of recommended cutoffs in the PseudoDojo PAW families is relatively small (min 10, 10, 10; max 20, 20, 25 for low, normal, high stringency). So, it seems fairly reasonable to replace placeholder cutoff values (i.e. -1) with the maximum for the corresponding stringency. This is an alternative solution to dropping the pseudos entirely, as proposed in #41