Stress tensor not symmetrized

[matthewkuner]

Hello VASP people. I am running high-throughput VASP calculations on the order of tens of thousands. These calculations are across chemical and structural space and include most elements on the periodic table. All PBE and r2SCAN calculations. I have noticed that a surprising number of these calculations have non-symmetric stress tensors, which is unphysical. These calculations were ran on VASP 6.4.2 on a linux-based CPU cluster. I have attached an example calculation, but there are thousands like it. The OUTCAR shows a symmetrized stress tensor, but the vasprun.xml does not. I figured this may be of interest to you.

[michael_wolloch]

Dear Matthew,

Thanks a lot for the bug report. I initially tried to examine this with a simplified version of your calculation, using PBE and a lot fewer k-points and ALGO = CG to be able to converge. I cannot see the issue there (see attached files) as soon as I got the calculations to converge. I used 6.5.1 and 6.4.2 without much difference in symmetry, but the actual results do differ, which indicates that they are not well converged. I will run your full example over the weekend with 6.4.2 and see if I can reproduce the problem. Your example is not a finished calculation. Could the issue be related to non-optimal convergence? Do you have an example of a fully converged calculation?

As for the discrepancy between the OUTCAR and vasprun.xml files: The OUTCAR only reports XY, YZ, and ZX, not YX, ZY, or XZ. I guess your OUTCAR parser completes the stress tensor in a symmetric way automatically? Btw.: hdf5 output matches vasprun.xml as expected.

I am not finished testing this yet, and will get back to you next week, but I wanted to reply before the weekend to let you know that we are on the case.

Thanks again, and a happy weekend.
Michael