My Community

Dear all:

I would like to perform a relaxation on the bulk monoclinic WO3 with tilting, but am currently running into problems with the forces converging to below 1 meV/angstrom; for the most part, they decrease but small oscillations seem to occur with visual inspection of the OUTCAR file. I've tried combinations of ALGO = Normal/All, PREC = Normal/Accurate, changing the values of EDIFFG/EDIFF (typically around -0.001 and -0.0001 respectively), trying ISMEAR = 0 and -5. I've also attempted using a sequence of ISIF = 2 to 3, but the forces seem to have trouble falling below 1 meV/atom. I've also attempted letting the number of ionic steps run for typically up to 75 steps to see if it is just slow; the energy does continually decrease but the forces still have some trouble. Using a higher ENCUT = 800 yields similar results.

Any ideas for how to make relaxation successfully happen would be helpful.
Thanks!

Below are the parameters for the sequence runs of ISIF = 2 and then ISIF = 3, as these had the more converged forces.
================================
INCAR

ALGO = All
PREC = Accurate
ENCUT = 500

NSW = 75
EDIFFG = -0.001
EDIFF = -0.0001
NELMIN = 5
ISMEAR = 0
SIGMA = 0.1
---------------------------------------------
POSCAR

is taken from experimental coordinates taken from ICSD database (Col. No 17003)
---------------------------------------------
KPOINTS

MP 4 x4x4
---------------------------------------------
POTCAR

PAW_PBE W and PAW_PBE O
---------------------------------------
Last iteration of OUTCAR Summary

Under "FORCES acting on ions", forces are up to 10^2 order of magnitude for individual atoms, although it appears that overall these are more of 10^-11 eV

For total-force these range from 0.01 to 0.00006 eV/Angstrom; there is no drift

Last few steps in OSZICAR- this one had a lot of ionic steps

84 F= -.29066792E+03 E0= -.29066792E+03 d E =-.251451E-03
85 F= -.29066791E+03 E0= -.29066791E+03 d E =-.248082E-03
86 F= -.29066792E+03 E0= -.29066792E+03 d E =-.255511E-03
87 F= -.29066805E+03 E0= -.29066805E+03 d E =-.128477E-03
88 F= -.29066812E+03 E0= -.29066812E+03 d E =-.195313E-03
89 F= -.29066828E+03 E0= -.29066828E+03 d E =-.161508E-03
90 F= -.29066834E+03 E0= -.29066834E+03 d E =-.217507E-03
91 F= -.29066846E+03 E0= -.29066846E+03 d E =-.128363E-03
92 F= -.29066859E+03 E0= -.29066859E+03 d E =-.256580E-03
93 F= -.29066878E+03 E0= -.29066878E+03 d E =-.185902E-03
94 F= -.29066886E+03 E0= -.29066886E+03 d E =-.264711E-03
95 F= -.29066900E+03 E0= -.29066900E+03 d E =-.143284E-03
96 F= -.29066906E+03 E0= -.29066906E+03 d E =-.206996E-03
97 F= -.29066926E+03 E0= -.29066926E+03 d E =-.192183E-03
98 F= -.29066928E+03 E0= -.29066928E+03 d E =-.220865E-03
99 F= -.29066943E+03 E0= -.29066943E+03 d E =-.144637E-03
100 F= -.29066945E+03 E0= -.29066945E+03 d E =-.162620E-03
101 F= -.29066958E+03 E0= -.29066958E+03 d E =-.134546E-03
102 F= -.29066970E+03 E0= -.29066970E+03 d E =-.257064E-03
103 F= -.29066982E+03 E0= -.29066982E+03 d E =-.113589E-03
104 F= -.29066986E+03 E0= -.29066986E+03 d E =-.155561E-03
105 F= -.29066986E+03 E0= -.29066986E+03 d E =-.153788E-03
106 F= -.29066986E+03 E0= -.29066986E+03 d E =-.156669E-03
107 F= -.29066991E+03 E0= -.29066991E+03 d E =-.510264E-04
108 F= -.29066999E+03 E0= -.29066999E+03 d E =-.124988E-03
109 F= -.29067011E+03 E0= -.29067011E+03 d E =-.125673E-03
110 F= -.29067023E+03 E0= -.29067023E+03 d E =-.248318E-03
111 F= -.29067030E+03 E0= -.29067030E+03 d E =-.618591E-04
112 F= -.29067031E+03 E0= -.29067031E+03 d E =-.793090E-04
113 F= -.29067032E+03 E0= -.29067032E+03 d E =-.872425E-04
114 F= -.29067032E+03 E0= -.29067032E+03 d E =-.895582E-04
115 F= -.29067032E+03 E0= -.29067032E+03 d E =-.901980E-04
116 F= -.29067035E+03 E0= -.29067035E+03 d E =-.233948E-04
117 F= -.29067039E+03 E0= -.29067039E+03 d E =-.628110E-04
118 F= -.29067044E+03 E0= -.29067044E+03 d E =-.114139E-03
119 F= -.29067053E+03 E0= -.29067053E+03 d E =-.882051E-04
120 F= -.29067058E+03 E0= -.29067058E+03 d E =-.137226E-03

It looks like your wavefunction is not good enough. Try a tighter criterium for EDIFF.

Cheers,

alex

in addition to alex' posting:
the ionic convergence criterium of 1meV/A is very strict, usually 10meV/A (as your calculations have reached already), would be sufficient for a ionic configuration to be regarded as 'converged'
you might try to continue with a different ionic relaxation algorithm (IBRION). As your cell must be very close to the converged geometry,
IBRION=1 (quasi-Newton) should work well, it may also help to set
ADDGRID=.True. in addition..

General advice: Figure it out by yourself.
For fast networks (like Infiniband or comparable stuff) it's good to start with NPAR = 1. Double it from one calculation to the next and compare the timings. If you see huge drops you'll have to adjust further. You'll go through a minimum of execution time with increasing NPAR (or NPAR = 1 is already the fastest ...).

Cheers,

alex