Hi, I wanted to ask about the compatibility of MPRelaxSet generated inputs vs. the Materials Project public inputs. I notice that for some materials, the MPRelaxSet-generated INCAR does not agree with the public INCAR.
For instance, the ISMEAR parameter for ‘mp-1418’ and ‘mp-20643’ is -5 by MPRelaxSet, but I got 1 from mpr.materials.tasks.search. With ISMEAR=-5 I also got the warning saying that “Tetrahedron method does not include variations of the Fermi occupations, so forces and stress will be inaccurate”, so I wonder if MPRelaxSet generates a reasonable ISMEAR
Hey @mathsphy, we really use two input sets, MPRelaxSet for relaxations and MPStaticSet, both use ISMEAR=-5 by default.
The former set can change the smearing to Methfessel-Paxton (ISMEAR = 1 or 2) if a structure is identified to be a metal. To disable that behavior, you can initialize MPRelaxSet with auto_ismear = False.
The issue with tetrahedron is primarily doing relaxations in metals where the forces are incorrect, which is why the warning is thrown. In insulators, the tetrahedron method + Blöchl corrections (ISMEAR = -5) has correct forces and is suitable for relaxations
Hey @Aaron_Kaplan, thank you very much for the explanation! I am still confused about how to automate the creation of input files that are fully consistent with MP.
What I am doing is structure optimization, so MPRelaxSet is what I need. Were you suggesting to use MPRelaxSet(structure) with auto_ismear set to True or False? I see that the default setting is alreayd False.
For mp-1418 (FeNi3) for instance, if I set auto_ismear set to True, the resulting ISMEAR is 0, which is still different from the value on https://next-gen.materialsproject.org/materials/mp-1418/tasks/mp-1418. As this is a metal, it’s unclear why I don’t get ISMEAR = 1 or 2…
Thanks in advance
That may be because some of the metals were run with MPMetalRelaxSet, which just bumps up the k-point density and changes ISMEAR and SIGMA to 1 and 0.2, respectively
Any of the electronic smearing settings should give you comparable results (unless you do relaxations with tetrahedron in metals, or Methfessel-Paxton / ISMEAR >= 1 relaxations with insulators). You’re fine to use either MPRelaxSet or MPMetalRelaxSet for MP GGA-compatible relaxations in metals
Also, just to note: we’ve been transitioning all of our data to a higher level of theory (r2SCAN meta-GGA) over the PBE GGA data that composes most of MP. These will be more expensive calcs to run, but you can use the MP24RelaxSet to generate forward-looking data. For compatibility with our current data, use MPRelaxSet / MPMetalRelaxSet as applicable
1 Like