HSE calculations

Message

apple · #1 Post by **apple** » Sat Oct 09, 2010 1:04 am

hi,
I am trying to run single point HSE calculations for a system of 48 atoms. The calculations are extremely slow, and so I wonder if someone could suggest what flags to use to speed it up a little bit. Has anyone performed some systematic study of how different parameters affect the convergence time? Lastly, can I start it with the WAVECAR of a PBE calculation (ISTART=1) to run my HSE jobs?
I would appreciate if someone would respond. Thank you.

Oscar_rp · #2 Post by **Oscar_rp** » Tue Oct 12, 2010 12:28 pm

Hello,
Please take a look to the manual and the keywords
NPAR = number of nodes
NSIM = 4

This may improve the speed of the calculation,
More feedback is welcome.
Thanks

apple · #3 Post by **apple** » Mon Oct 18, 2010 12:43 am

Thank you. I played with these flags, but no significant improvement was achieved.

alex · #4 Post by **alex** » Mon Oct 18, 2010 6:44 am

Hello apple,

that all sounds sound. Maybe some other strategy helps: Try to figure out with plain DFT if you can decrease the overall accuracy of your calculation by decreasing for example cut-off energy, k-points and mesh precision (,...) and keeping your results not far from the 'precise' ones. Go to the cheapest set and use this for your hybrid calculations.

One other tag to save something in hybrid calculations is NKRED.

Cheers,

alex

apple · #5 Post by **apple** » Tue Oct 19, 2010 3:23 pm

Thanks Alex. Good point, but my calculations are still extremely slow.