My Community

I have a fairly large GW calculation, that will not fit the allotted max run time that we have on the cluster. Is there a way to restart the calculation if it "dies" due to timeout somewhere half way through? Note that this is not a "normal" restart in which a calculation nicely finished and one just does something new with the result - I would know how to do this. Or is there a possibility to calculate only for some k-points in one go?
Best regards,
Katharina

Hi,

GW consists of several steps: one or two DFT runs (that produce WAVECAR and WAVEDER) followed by a GW run. If the termination occurred during the GW run, then the DFT part doesn't need to be repeated, however the GW run needs to be restarted (in general, WAVECAR is written only at the very end of a calculation).

The DFT part is not the issue. Can the calculation not be split at all?

Besides parallelization, no. If not already done, try to figure out if parameters (e.g. ENCUT) can be reduced to save computational time.