Infinite repeating SCF during cell opt #151
Replies: 5 comments 3 replies
-
With |
Beta Was this translation helpful? Give feedback.
-
I've spent a little time analysing this. If we relax with But using |
Beta Was this translation helpful? Give feedback.
-
I've encountered a similar issue, using I've confirmed that it converges as expected if |
Beta Was this translation helpful? Give feedback.
-
Your SCF tolerance is rather low for this kind of simulation. I ran the calculation with a 2x2x2 k-mesh for the sake of speed, which converged, and found warnings in the |
Beta Was this translation helpful? Give feedback.
-
Thanks for all this data! The warning will generally only be triggered when the change in cell size is small, as this will lead to small total energy changes. If the change in energy from the last step of the SCF cycle is comparable to the change in energy from the cell update then the minimisation is likely to fail (the SCF change will be like noise). My guess is that this is somewhat random: it's just that with the 8x8x8 mesh the stress after the second or third iteration is large enough that the optimisation doesn't stop, but small enough that the SCF noise affects it. We should maybe make the default maximum number of iterations different for cell optimisation and ionic optimisation; I will think about whether there is a way to detect this kind of noisy problem and abort or warn further. |
Beta Was this translation helpful? Give feedback.
-
Dear All,
Attached is a basic example of cell optimisation using Conquest.
This example was working properly using the develop branch before
last commits (I don't know when). It looks like the SCF is restarting
infinitely. I found that by increasing SCF tol. to 1e-8 things
get better but I feel that there might be something wrong.
Could you have a look.
Best regards,
Lionel
cq_example_rocksalt_optcell_cq.tar.zip
Beta Was this translation helpful? Give feedback.
All reactions