Grain boundary cell relaxation crashes

[akretschmer]

Hello,

I am trying to relax some grain boundaries of Al. Most calculations run fine, but for two structures the calculation crashes immediately.

Looking at the error in the code "WARNING: Sub-Space-Matrix is not hermitian in DAV" and other topics with this issue it seems that my input geometry is not "reasonable". The same structure has worked with other elements like Cu, Ni, and Pd, and the minimum distance between ions is about 2.7, close to the distance in fcc Al, so I do believe my structure is reasonable.

I have tried a few different settings like ALGO = Normal and Fast and IBRION = 2 and 3, and set POTIM = 0.05, but nothing worked so far.

Is there something else I can try?

[alexey.tal]

Dear akretschmer,

Thank you for your question.

It looks like the positions in your original structure are fine. The problem is that the electronic minimization doesn't converge, which you can see in your OUTCAR:

Code: Select all

Itertation 1: -76.88267563
Itertation 2: -302.53962035
Itertation 3: -8273.53259391
Itertation 4: -2000.27165936
Itertation 5: -1282.33783238
Itertation 6: -1438.39156576
Itertation 7: -1295.68991427
Itertation 8: -469.69406741
Itertation 9: -951.71946867
Itertation 10: -346.21164725
Itertation 11: -256.03758795
Itertation 12: -313.53726077
Itertation 13: -465.57675991
Itertation 14: -1377.01728903
Itertation 15: -2770.38572102
Itertation 16: -1120.10388090
Itertation 17: -5081.33693934
Itertation 18: -8564.84235648
Itertation 19: -8672.42474028
Itertation 20: -39720.31113036
Itertation 21: -51979.16731773
Itertation 22: -89457.99753009

That is why when at the last iteration a new structure is calculated, the positions of the atoms are probably too close.
I would suggest that you restart your calculation with ICHRAG=2 to start the calculation from the atomic charge density.
Also, you can try to use linear mixing by setting BMIX=0.0001 and BMIX_MAG=0.0001 and/or mix slowly, i.e., reduce AMIX and AMIX_MAG.

Best wishes,
Alexey

[akretschmer]

I have changed the mixing parameters and it worked for one of the two structures, which is now relaxed. The second structure still crashes, but the divergence is less pronounced now. First I set only BMIX and BMIX_MAG to the recommended value, then I also decreased AMIX and AMIX_MAG. How should I tweak the settings further?

sigma3-(1-1-2)_new.zip

[alexey.tal]

It looks like the issue with this calculation is related to the parallelization settings. I see that you set NCORE=24. However, it is not an optimal parallelization scheme as each band is calculated on 24 cores. You can find more information on how to set the optimal parallelization parameters here. I would suggest that you try a smaller NCORE and/or set KPAR>1 to improve the performance of this calculation.