Problematic static energy

Materials Project
1

Dear MP developers,

I would like to report that the total energy of the static run for mp-558822 is probably wrong.
The total energy of the static run for this compound is -260.137 eV. (https://materialsproject.org/tasks/mp-558822#mp-1387333) Whereas the total energy of its double relaxation is -272.597 eV (https://materialsproject.org/tasks/mp-558822#mp-566867) The MP website uses this static energy for ehull/formation energy calc, which leads to an unreasonable prediction of phase stability.

Though the kpts are different (2x2x3(double relaxation) vs 3x3x4(static)), the total energy difference is too huge (~12eV). I redid these calculations using the MPRelaxSet and MPStaticSet and found that the static run with MP defaults parameters was converged to a different (probably wrong) magnetic state. (I obtained the same results as MP) After playing with INCAR params, I got total energy of -272.567eV for the static run, which is very close to that of double relaxation.

Is the MP moving forward to use the more accurate energy (denser kpoints) from static run instead of double relaxation for all compounds? If so, it is better to check their magnetic states to make sure the calculation is “well-converged”.

Is there a way to only query double relaxation entries using MPRester?

Thanks for your help!

Best,
XG

2

Hi @xig126

This is a definite issue we’ve noticed. We recently switched to static calculations that also include aspherical corrections, which are expected to higher quality calculations. This can account for the big difference in energy. We needed the aspherical corrections for our switch to computing multiple magnetic orderings for materials. There’s roughly 10K magnetic orderings calculated and we plan on computing multiple orderings for all potentially magnetic materials in MP.