Page 1 of 1

Different convergence between serial and parallel versions

Posted: Tue May 01, 2012 6:47 pm
by quan.phung
Dear VASP users,

Recently, I used VASP to compute a rather big system with 160 atoms.
Because of the problem wit SCF converge, I tried to use block Davison (ALGO=Normal) and I find a problem with the energy when using the parallel version of VASP:

Code: Select all

???????N???????E?????????????????????dE?????????????d?eps???????ncg?????rms??????????rms(c)
DAV:???1????-0.289086412606E+06???-0.28909E+06???-0.10942E+06??2156???0.264E+03
DAV:???2????-0.141986952982E+08???-0.13910E+08???-0.63294E+07??2160???0.389E+03
DAV:???3????-0.197007540347E+08???-0.55021E+07???-0.49416E+07??2192???0.295E+03
DAV:???4????-0.117582327538E+08????0.79425E+07???-0.20942E+07??2192???0.280E+03
DAV:???5????-0.141737319460E+08???-0.24155E+07???-0.34681E+07??2192???0.313E+03????0.457E+02
DAV:???6????-0.249451827012E+08???-0.10771E+08???-0.83463E+07??2192???0.748E+03????0.301E+02
DAV:???7????-0.122967852718E+08????0.12648E+08???-0.82898E+07??2176???0.170E+04????0.361E+01
DAV:???8????-0.107553326574E+08????0.15415E+07???-0.61256E+07??2192???0.169E+04????0.223E+01
As you can see, the energy is exceptional negative (it should be about E+4) and the energy then never goes up.
Using the serial version with the same input, I obtained a smother convergence:

Code: Select all

???????N???????E?????????????????????dE?????????????d?eps???????ncg?????rms??????????rms(c)
DAV:???1?????0.264350857965E+05????0.26435E+05???-0.12876E+05??1878???0.259E+03
DAV:???2?????0.121159361964E+05???-0.14319E+05???-0.12278E+05??1822???0.106E+03
DAV:???3?????0.545421658016E+04???-0.66617E+04???-0.58185E+04??2126???0.902E+02
DAV:???4?????0.295002609650E+04???-0.25042E+04???-0.21311E+04??2126???0.964E+02
DAV:???5?????0.164293928512E+04???-0.13071E+04???-0.10739E+04??2174???0.839E+02????0.195E+02
DAV:???6?????0.291376823400E+04????0.12708E+04???-0.53815E+03??1881???0.115E+03????0.792E+01
DAV:???7?????0.702354590168E+02???-0.28435E+04???-0.26680E+04??2156???0.119E+03????0.857E+01
DAV:???8????-0.970547070458E+02???-0.16729E+03???-0.43826E+03??1132???0.132E+03????0.679E+01
DAV:???9????-0.850397331279E+03???-0.75334E+03???-0.10719E+04??2182???0.506E+02????0.525E+01
DAV:??10????-0.110292088176E+04???-0.25252E+03???-0.38337E+03??2108???0.403E+02????0.236E+01
DAV:??11????-0.142072276050E+04???-0.31780E+03???-0.20328E+03??2188???0.207E+02????0.708E+01
This is my INCAR file

Code: Select all

?System?=?LuAG
?PREC?=?Normal
?ENCUT?=?400
?ISMEAR?=?0;?SIGMA?=?0.1;
?LREAL?=?Auto????!?Projection?done?in?real?space
?ALGO?=?Normal
?IBRION?=?2???!?CG
?ISIF?=?2?????!?Change?internal?parameters
?NSW?=?0??????!?Number?of?ionic?step
?EDIFF?=?0.1E-04
?EDIFFG?=?-0.01
?NELM?=?40
Is this behaviour normal (I guess not)? Is there any way to solve it by just adding some keyword in INCAR, or I have to recompile the source?

Best regards,
Quan

Different convergence between serial and parallel versions

Posted: Wed May 02, 2012 9:47 am
by alex
Hi Quan,

first off all, some of your input(s) has to be wrong, the convergence is just not present.

Check http://cms.mpi.univie.ac.at/vasp-forum/ ... hp?3.11202 for your original question.

Cheers,

alex