GGA+U with no structure optimization in MP

Materials Project
1

For about 300 materials in MP I see that either:

  • they have had a structure relaxation done with GGA only, no Hubbard parameter, and that task was deprecated, and a subsequent GGA+U Static was done, and from there a GGA+U NSCF Uniform. Example record However, no GGA+U Structure Optimization was performed.
  • they have no structure optimization, only a static calculation only. Example record

For most other GGA+U calculated materials, typically a GGA+U Structure Optimization was performed before the other GGA+U calculations.

Do you anticipate that these materials may have subsequent GGA+U Structure Optimization calculations in the future? Should these materials not be treated with the same level of accuracy as materials which have had a full GGA+U Structure Optimization prior to their other associated tasks?

I have upload a list of all of the mp-IDs affected in this post.
gga_u_no_structure_optimization.csv (31.9 KB)

2

We’re working on recomputing (re-relaxing + single-point at the optimized geometry) all entries in MP with the r2SCAN meta-GGA, which will resolve these inconsistencies

Both structures you pointed to are very high in energy above the hull. That could be because they’re at a non-optimal geometry, or because they really are unstable.

If you want to check how close a structure is to its equilibrium configuration, you want to look at the maximum final interatomic force from the static task calculation (or relaxation if that’s the primary calculation).

I’m taking task mp-1324026 from your first example (material mp-754888):

import numpy as np
from mp_api.client import MPRester

with MPRester() as mpr:
    task_doc = mpr.materials.tasks.search(task_ids=["mp-1324026"])[0]
ionic_steps = task_doc.calcs_reversed[0].output.ionic_steps
max_force = max([np.linalg.norm(f) for f in ionic_steps[-1].forces])
ediffg_per_atom = None
if len(ionic_steps) > 1:
    ediffg_per_atom = abs(ionic_steps[-1].e_0_energy - ionic_steps[-2].e_0_energy)/len(task_doc.structure)

For our current calculations, we aim for max_force < 0.02 eV/Å. That particular task has max_force=0.257 eV/Å, so it’s likely not at its optimal geometry. For the legacy GGA / +U calculations, ediffg_per_atom < 5 \times 10^{-4}` eV/atom is the target (this is only meaningful for a relaxation with more than one ionic step).