FermiSolver and ChemicalPotentialGrid
#46
Conversation
|
Looking good @alexsquires. I like the mix-in/abstract class approach. The Ofc the user could iterate over all possible vertex/limit combinations if they knew what they were doing, which would be a good stab at this and get you most of the way there I imagine (kind of like a 'band structure path' approach), but still is only covering certain 'high-symmetry' paths in chem pot space. So if a grid approach was also possible (where e.g. the user can just set the chempot spacing in eV, or total number of grid points), that would also be v nice I think – should be doable with some of the Sorry to be piling on the requests, but I guess the other main use case I'd imagine for this part of the code would be for the more complex defect thermodynamics analysis that Also just fyi, about your earlier |
|
Yeah, dw this is all on the list, should be ready to go by the end of the
week, I sort of trashed everything to start again (for about the 6th time
now), and just adding everything back in piecewise to make sure it's all
consistent with the latest thermodynamics object.
…On Thu, 15 Feb 2024 at 04:58, Seán Kavanagh ***@***.***> wrote:
Looking good @alexsquires <https://github.com/alexsquires>. I like the
mix-in/abstract class approach.
The _interpolate_chempots() function is quite nice, and I think will
likely be one of the mostly-used features from these additions. From what
we were discussing before in the #doped Slack, would it be possible to
also integrate a grid-scanning approach too? As in, the current function
linearly interpolates between two points (chem pot limits), which is nice
and v useful in many cases, but if for instance we have a 3D chemical
potential space, our min/max of the property of interest could also lie
somewhere in the middle of this space, rather than along any linear path
between two vertices/limits.
Ofc the user could iterate over all possible vertex/limit combinations if
they knew what they were doing, which would be a good stab at this and get
you most of the way there I imagine (kind of like a 'band structure path'
approach), but still is only covering certain 'high-symmetry' paths in chem
pot space. So if a grid approach was also possible (where e.g. the user can
just set the chempot spacing in eV, or total number of grid points), that
would also be v nice I think – should be doable with some of the scipy
grid space interpolation functions I think? And/or pymatgen chemical
potential diagram methods maybe?
If it was possible to implement relatively painlessly, could we add these
two options? ☝️ 🙏
Sorry to be piling on the requests, but I guess the other main use case
I'd imagine for this part of the code would be for the more complex defect
thermodynamics analysis that py-sc-fermi allows, where you want to fix
certain defect/species etc concentrations and perform some constrained
solutions. Is it possible to integrate this to the fermi_solver code?
Also just fyi, about your earlier _format_defect_name() question, yes it
is actually quite externally useful and I'll make it a public function in
the next (minor) release.
—
Reply to this email directly, view it on GitHub
<#46 (comment)>, or
unsubscribe
<https://github.com/notifications/unsubscribe-auth/AFMK74XGZEUFF3TPF7MZHH3YTXZ5JAVCNFSM6AAAAABBVDPFSSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNBVHE2TIMRVHE>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
|
@adair-nicolson , could you re-parse your Cu2SiSe3 data now that everything is relatively final and I can add the grid searching stuff back in? |
|
@kavanase proposing this as the full functionality - do you want to take a look at the example ipynb while I finish off the tests before I get to comfortable? |
# Conflicts: # examples/CdTe/CdTe_example_thermo.json
…trations, add tests
…of a defect in `py-sc-fermi`
…free_defects` and auto-switching to `py-sc-fermi` backend when required
…e code (and thus maintenance burden)
|
@alexsquires I'm aiming to merge this and release this weekend / early this week. I renamed the Some notes/thoughts just for reference:
I'll try get to these when I can find some time (lol), but if you had a chance at any point to try tackle some of these would be much appreciated! |
|
Sounds good. I should be able to work on some of these this week.
```
Future TODO: Should also allow just specifying an extrinsic element for
fixed_defects, to allow the user to specify the known concentration of a
dopant / over/under-stoichiometry of a given element (but with unknown
relative populations of different possible defects) -- realistically the
most commonly desired option (but would require a bit of refactoring from
the current py-sc-fermi implementation).
- The DefectThermodynamics JSONs in the repo for extrinsic-doped
Selenium (link
<https://chemrxiv.org/engage/chemrxiv/article-details/6706a93612ff75c3a1f8c995>)
would be a good test case for this
```
Never thought of this, nice idea. I was talking to Ben about some plans for
some new py-sc-fermi functionality and I can add this to the pile
(including proper metastable defects and site competition). Just to note,
as well as `per_charge` information in the concentration dict, you can get
at `per_site` concentrations via setting `per_volume` to false. Just in
case you get there first.
…On Sat, 18 Jan 2025 at 20:06, Seán Kavanagh ***@***.***> wrote:
@alexsquires <https://github.com/alexsquires> I'm aiming to merge this
and release this weekend / early this week.
I renamed the min_max_X method to optimise as I think this might be a bit
more intuitive / obvious as a user-facing function. For the tutorial, I
think we can probably remove the doped/py-sc-fermi comparisons; useful to
have for testing (I have added some TODO notes to the tests to use these
as handy integration tests, by generating these plots and testing with
pytest-mpl), but I think it's not needed in the tutorials and will just
confuse users. Any thoughts on this or all other changes lmk!
Some notes/thoughts just for reference:
- Per-charge outputs are currently not supported for FermiSolver, but
are for DefectThermodynamics, and so only total defect concentrations
are accessible. This is useful information to users in a lot of cases, so
would be good to include in future. I think this is possible with
py-sc-fermi, just needs to use the decomposed option with
concentration_dict?
- I think I'm still not sure how metastable defects (two defects of
the same type and charge state) are handled in py-sc-fermi's
functions. They're handled natively in doped with the appropriate
Boltzmann summation (from what I remember before). Anyway I guess this is
unlikely to matter significantly in most cases.
- Future TODO: Should also allow just specifying an extrinsic element
for fixed_defects, to allow the user to specify the known
concentration of a dopant / over/under-stoichiometry of a given element
(but with unknown relative populations of different possible defects) --
realistically the most commonly desired option (but would require a bit of
refactoring from the current py-sc-fermi implementation).
- The DefectThermodynamics JSONs in the repo for extrinsic-doped
Selenium (link
<https://chemrxiv.org/engage/chemrxiv/article-details/6706a93612ff75c3a1f8c995>)
would be a good test case for this.
- Future TODO: In future the fixed_defects, free_defects and
fix_charge_states options may be added to the doped backend (in theory
very simple to add)
- Future TODO: Add per-site option like in doped DefectThermodynamics,
should be quick to add
- Currently the fixed_defects, free_defects and fix_charge_states
options aren't really tested, I've added a TODO note for this.
- I have some examples now in the tests where the extremum position
for a defect/carrier concentration doesn't occur at a limiting chemical
potential (e.g. CdTe from my thesis, V_S in Sb2S3 just about (
https://pubs.acs.org/doi/10.1021/acsenergylett.4c02722)), which I
think will be good to show in the tutorial as this is the main case where
this function is particularly powerful.
- It will also be good to use the scan_X functions now in the main thermodynamics
tutorial
<https://doped.readthedocs.io/en/latest/thermodynamics_tutorial.html>
as this should now be the most convenient and recommended way of doing
this, unless extra control is needed e.g. to do the bandgap scissoring I
showed for CdTe.
I'll try get to these when I can find some time (lol), but if you had a
chance at any point to try tackle some of these would be much appreciated!
—
Reply to this email directly, view it on GitHub
<#46 (comment)>, or
unsubscribe
<https://github.com/notifications/unsubscribe-auth/AFMK74UZFSK2YF56RKG2XZL2LKX4DAVCNFSM6AAAAABBVDPFSSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKOJZHA3DAOJSGY>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
# Conflicts: # examples/YTOS/YTOS_example_defect_dict.json # examples/YTOS/YTOS_example_thermo.json
|
Ok cool. All |
kavanase
changed the title
FermiSolver and ChemicalPotentialGrid
defect_levels argument when annealing becomes a boolean