Open rkingsbury opened 1 year ago
Ah, I have just noticed that there is logic in FEFFDictSet
that changes COREHOLE
to RPA
when a reciprocal space calculation is invoked.
In my opinion, there is still value in considering a change to the .yaml value to minimize confusion for other users. I know that I (and I would suspect many others) look at the .yaml files as a clear, quick synopsis of what calculation settings are used by different input sets, so if it's likely that COREHOLE
would be overwritten in many cases of practical interest, I'd advocate changing the .yaml to reflect the more likely scenario.
We could always reverse the logic in FEFFDictSet
if it is desired to default to FSR for real-space calculations (although from what I can tell it's not clear that FSR is superior to RPA in this case; rather that RPA is definitely preferred for k-space calcs).
@rkingsbury I agree with everything you're saying here. I think though what's most important is that the default settings and behavior be clearly exposed in the documentation. They're currently not, as they're buried in the MPXANESSet.yaml
files and related. As I mentioned in #2611 I think this really needs to change. I am happy to implement those PR's if I get approval, then it will be much easier to specify the default behaviors and clarify these confusions.
Both studies (cited below) that describe the development of high-throughput XANES workflows state that the 'RPA' method was chosen for screening of the core-hole potential. However
MPXANESSet.yaml
uses the default 'FSR' setting.The ML paper (Zhenget al. 2018) seems to indicate that RPA gave more accurate results, and according to the user guide this is the preferred method if using reciprocal-space calculations.
I personally do not have a strong opinion one way or another, but if
MPXANESSet
is going to deviate from what is in the publications, I think it would be good to document that somewhere. The other settings in the yaml file appear to be generally consistent with the benchmarking work described in the studies.I propose that we either revise the .yaml to use the RPA method, or add something to the docstrings to explain when FSR is preferable.
https://www.nature.com/articles/sdata2018151#citeas
https://www.nature.com/articles/s41524-018-0067-x#citeas (in Supporting Information Fig 3B)
The FEFF10 user guide states