KeyError of elemental references in pre-computed phase diagram

[peikai]

When I try to plot a pre-computed phase diagrams that are retrieved via get_phase_diagram_from_chemsys() method,

phaseDiagram = mpr.get_phase_diagram_from_chemsys('Li-Fe-O')
plotter = PDPlotter(phaseDiagram)

a KeyError raises as shown below, which arises from the inconsistence on object stored in el_refs dictionary, i.e., the keys stored in phaseDiagram.el_refs are str object (dict_keys(['O', 'Fe', 'Li'])), instead of the Element object (dict_keys([Element O, Element Fe, Element Li])).

The pre-computed phase diagrams in database might not be updated conveniently. So, I add some codes in pymatgen.analysis.phase_diagram.PhaseDiagram class, to update keys in elemental reference dictionary to Element object, in case it loads a pre-computed phase diagram. A corresponding test unit is also added to check the keys in el_refs dictionary,

KeyError message:

Traceback (most recent call last):
File "D:\Seafile\peikai\My Libraries\Repos\Volume-Planning-for-Anodes\phase_diagram_mixscheme.py", line 37, in
plotter = PDPlotter(phase_diagram, show_unstable=0, backend='matplotlib')
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\Anaconda\envs\pymatgen2023\Lib\site-packages\pymatgen\analysis\phase_diagram.py", line 2111, in init
self._min_energy = min(self._pd.get_form_energy_per_atom(e) for e in self._pd.stable_entries)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\Anaconda\envs\pymatgen2023\Lib\site-packages\pymatgen\analysis\phase_diagram.py", line 2111, in
self._min_energy = min(self._pd.get_form_energy_per_atom(e) for e in self._pd.stable_entries)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\Anaconda\envs\pymatgen2023\Lib\site-packages\pymatgen\analysis\phase_diagram.py", line 577, in get_form_energy_per_atom
return self.get_form_energy(entry) / entry.composition.num_atoms
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\Anaconda\envs\pymatgen2023\Lib\site-packages\pymatgen\analysis\phase_diagram.py", line 564, in get_form_energy
return entry.energy - sum(comp[el] * self.el_refs[el].energy_per_atom for el in comp.elements)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\Anaconda\envs\pymatgen2023\Lib\site-packages\pymatgen\analysis\phase_diagram.py", line 564, in
return entry.energy - sum(comp[el] * self.el_refs[el].energy_per_atom for el in comp.elements)
~~~~~~~~~~~~^^^^
KeyError: Element Li

[coveralls]

Coverage: 78.129% (-0.7%) from 78.807% when pulling cd146e3 on peikai:master into d9b8697 on materialsproject:master.

[munrojm]

@peikai, I appreciate you making this PR. I just patched the API client to temporarily resolve this issue yesterday as we were having plotting issues on the front-end with the pre-computed data. I also found an additional change that needs to be made to for some ternary compounds. Namely, qhull_data needs to be a numpy array, and not a list. Although, it looks like you don't have that issue. In any case, to fix your local problems before this is merged, you can update mp-api as well.

@janosh, this relates to the emmet issue I tagged you in regarding PhaseDiagram de-serialization problems with monty.

[janosh]

@peikai Thanks for this PR. Do we really need pymatgen/analysis/tests/pre_computed_phase_diagram_Li-Fe-O.json? Maybe you can re-use an existing test file in the test_files/ dir? The repo size is already approaching 1GB which we really need to get down.

[peikai]

@munrojm Thanks for the information. The updated mp-api works. And this PR would fix the issue in pymatgen.