RPA calculation for GW

[sponza]

Hallo.
I'm running G0W0@HSE03 on a pretty big semiconductor (108 electrons).
I am splitting the calculation in four parts:
1) a PBE calculation of occupied states only,
2) a HSE03 calculation of occupied states only,
3) a HSE03 calculation with empty bands (LOPTICS=TRUE)
4) the actual GW calculation.
For some reason I get stuck at point 3): the simulation stops before the calculation of chi0 without a clear reason.
I have to say that I used the same recipe for other calculations without problems, so I don't get where the error may be.

I post in the following my input file and the last part of the OUTCAR of step 3).

INCAR:

Code: Select all

???NPAR???=?16


Start?parameters?:
???ISTART?=?1

Electronic?relaxation?:
???PREC???=?Normal
???ENCUT??=?800.00?eV
???EDIFF?=?0.1E-4
???NBANDS?=?500

???LWAVE?=?TRUE
???LCHARG?=?FALSE
???LOPTICS?=?True

DOS?related?values:
???NELM?=?1
???ISMEAR?=?0?;?SIGMA?=?0.05
???CSHIFT?=?0.1
Exchange?correlation?treatment
????LHFCALC?=?.TRUE.??;?HFSCREEN?=?0.3
????ALGO?=?Exact
????PRECFOCK=?F

????NKREDX?=?2?;?NKREDY?=?2?;?NKREDZ?=?2
    MAXMEM?=?4000

last part of the OUTCAR:

Code: Select all

--------------------------------------------------------------------------------------------------------


?Initial?charge?from?wavefunction
?First?call?to?EWALD:??gamma=???0.316
?Maximum?number?of?real-space?cells?3x?3x?3
?Maximum?number?of?reciprocal?cells?3x?3x?3

?FEWALD?executed?in?parallel
?FEWALD:?cpu?time???0.00:?real?time???0.00


-----------------------------------------?Iteration????1(???1)?---------------------------------------


????POTLOK:??cpu?time????0.18:?real?time????0.18
????SETDIJ:??cpu?time????4.69:?real?time????4.70

?total?amount?of?memory?used?by?VASP?on?root?node??1916708.?kBytes
========================================================================

???base???????:????????30000.?kBytes
???nonl-proj?:????????68874.?kBytes
???fftplans???:??????????1377.?kBytes
???grid???????:??????????4409.?kBytes
???one-center:??????????304.?kBytes
???HF?????????:??????????211.?kBytes
???nonlr-proj:??????????2259.?kBytes
???wavefun???:????1809274.?kBytes

Thanks a lot for any suggestion!

<span class='smallblacktext'>[ Edited ]</span>

[sponza]

I finally found how to do it properly.
The calculation is split into two steps.
1) compute GW putting in NBANDSGW as many bands as needed. Now you have a WAVECAR file where first eigenstates are the GW0 eigenstates.
2) Now feed this WAVECAR and the WAVEDER used to compute the step 1) into a ALGO=CHI (see manual) calculation.
This will give both RPA w/o local fields and RPA with local fields.
Be careful that if you are interested in the local fields you will maybe need to converge on the NBANDSGW value of the step 1).


The most important thing to notice is that if you just make ALGO=NONE ; LOPTICS=TRUE on top of the GW calculation, you will have a spectrum that has the right poles but not the right intensity.