Elastic Tensor, strain criteria


Hopefully someone can answer me here, I figured I would keep the discussion open for others to see.

I have been trying to use the pymatgen elastic methods with Quantum Espresso and while trying to get Silicon elastic properties, I noticed that my shear elastic element (C44) was differing from your database by about 25 GPa. After checking convergence criteria, and pseudopotentials, I believe the difference may be in the strains used. The rest of the tensor is within 15% of yours.

I know the filters you apply force different strains to be studied, in your source code, you can specify normal strain and shear strain. However, in your paper you mention that you vary the strain percentages between 0.5% and 1.25% in order to get a consistent elastic tensor. How much do you tend to vary the shear strain (default values in pymatgen are 6%)?

If the suggested criteria of checking ranges of strain and comparing the fits to each other does not work, what other options do I have (besides increasing the kpoints)?

Just to check, you did use the conventional cell for the calculation correct ?

Hi cpashartis, thanks for the question.

Yes, the shear strain value is varied by roughly 3-6%, and is more explicitly outlined in the current elastic data documentation. There’s an updated version of the workflow that is slightly different (primarily in that the perturbations are applied to the strain tensor, rather than the deformation gradient) that’s implemented in atomate.

Also, we did use the conventional cell here. Have you tried using a higher k-point density? Some of our non-metallic compounds have used 1000 kppra, but we find that most of them need higher k-point densities, so that was set as the default. I’m not 100% sure that Si used the higher standard 7000 kppra, but I suspect it did.

1 Like

Ah, I simply assumed you applied to the strain tensor (which you input to the deformation generator).

I varied my kpoint mesh from 1000 to 3000 and didn’t see a significant shift. I don’t have access to VASP, but I can try a different code. I still have no idea what is causing this issue in any case. Oddly enough, calculating Carbon (diamond) does work sufficiently well…

I tried to vary my shear strain ranging from 3% to 8%. I have no difference in my shear elastic elements after I reach about 4%. If I go lower it drastically drops by more than 15 GPa.