To speed up GW0 calculations

Message

huangj3 · #1 Post by **huangj3** » Mon Feb 18, 2013 8:51 pm

GW0 calculations are very expensive and slow. ENCUTGW significantly speeds up the calculation and the Eg values are only slightly affected. So is PRECFOCK=fast for the tests I did. However, the manual suggests that for RPA calculations it is recommended to set PRECFOCK= normal to avoid numerical errors.

From the webpage of LRPA at http://cms.mpi.univie.ac.at/vasp/vasp/L ... l_RPA.html, it seems that no matter what you use, either LRPA=.TRUE. or LRPA=.FALSE. (default), local field effects are always included. Only that in the .TRUE. case, the local field effects are included only on the Hartree level while in the .FALSE. case, they are included on both the Hartree level and the exchange correlation level. So in what situation (LRPA=.TRUE. or .FALSE.) one should set PRECFOCK= normal to avoid numerical errors?

Besides, I tried v5.3.3 and v5.2.12, LRPA=.TRUE. and LRPA=.FALSE. gave the same results. No such line of MACROSCOPIC STATIC DIELECTRIC TENSOR (including local field effects in DFT) in the case of LRPA=.FALSE. as described in the vasp manual at http://cms.mpi.univie.ac.at/vasp/vasp/L ... l_RPA.html.

Thanks.

JSH

#2 Post by **admin** » Tue Feb 19, 2013 5:11 pm

PRECFOCK=normal is the default setting. You will avoid numerical errors when not using lower precisions than normal in ACFDT-RPA calculations.
On the other hand, PRECFOCK=fast is safe to use in any type of GW calculations.

huangj3 · #3 Post by **huangj3** » Tue Feb 19, 2013 5:58 pm

Dear Admin, thank you for your quick response. So you meant PRECFOCK should be set to normal for ACFDT-RPA but fast is OK for GW? Good to know that PRECFOCK = fast is safe to use for any types of GW calculations.

Just another quick question. Why does LRPA = .TRUE. and .FALSE. give the same GW0 results? Therefore it seems LRPA= .FALSE. didn't work.
<span class='smallblacktext'>[ Edited Tue Feb 19 2013, 07:31PM ]</span>