My Community

Hi,
I'm trying to optimize a metal surface but I'm struggling to understand from the output what are the figures I have to consider in order to improve the calculation and modify the thresholds to reach the minimum as quickly as possible without wasting computational time.

In my INCAR file I have set
EDIFF = 1E-6
EDIFFG = -1E-2
IBRION = 2

and the last lines of my job-out.txt file are (please carry on reading until the bottom of the message where I'm asking specific questions):

69 F= -.96956373E+03 E0= -.96956567E+03 d E =-.845154E-04 mag= 2.0000
trial-energy change: -0.000085 1 .order -0.000091 -0.000109 -0.000074
step: 0.8537(harm= 0.8537) dis= 0.00311 next Energy= -969.563814 (dE=-0.168E-03)
bond charge predicted
N E dE d eps ncg rms rms(c)
DAV: 1 -0.969563291231E+03 0.43950E-03 -0.39237E-01 12160 0.809E-01 0.103E-01
RMM: 2 -0.969563794635E+03 -0.50340E-03 -0.65029E-03 12156 0.987E-02 0.578E-02
RMM: 3 -0.969563775590E+03 0.19045E-04 -0.18756E-04 12177 0.138E-02 0.342E-02
RMM: 4 -0.969563783799E+03 -0.82090E-05 -0.75296E-05 11750 0.876E-03 0.149E-02
RMM: 5 -0.969563789037E+03 -0.52387E-05 -0.13752E-05 11212 0.364E-03 0.571E-03
RMM: 6 -0.969563788622E+03 0.41537E-06 -0.89533E-07 7286 0.111E-03 0.343E-03
RMM: 7 -0.969563788903E+03 -0.28137E-06 -0.10808E-06 6975 0.114E-03 0.132E-03
RMM: 8 -0.969563789072E+03 -0.16904E-06 -0.13274E-07 6033 0.483E-04
70 F= -.96956379E+03 E0= -.96956573E+03 d E =-.142796E-03 mag= 2.0000
curvature: -10.34 expect dE=-0.413E-02 dE for cont linesearch -0.902E-07
trial: gam= 0.97828 g(F)= 0.400E-03 g(S)= 0.000E+00 ort =-0.912E-05 (trialstep = 0.392E+00)
search vector abs. value= 0.950E-02
writing wavefunctions
Application 2794415 resources: utime ~8367312s, stime ~65796s
--------------------------------------------------------------------------------

Resources requested: mpparch=XT,mppnppn=16,mppwidth=256,ncpus=1,place=pack,walltime=12:00:00
Resources allocated: cpupercent=0,cput=00:00:01,mem=5224kb,ncpus=1,vmem=39220kb,walltime=09:09:12

*** sacfm2 Job: 911674.sdb ends: 28/09/12 21:12:26 queue: par:16n_12h ***
*** sacfm2 Job: 911674.sdb ends: 28/09/12 21:12:26 queue: par:16n_12h ***
*** sacfm2 Job: 911674.sdb ends: 28/09/12 21:12:26 queue: par:16n_12h ***
*** sacfm2 Job: 911674.sdb ends: 28/09/12 21:12:26 queue: par:16n_12h ***

Which one of "F=..", "dE=..", "g(F)=.." relates to EDIFF and EDIFFG?
How can I keep checking that the calculation is getting anywhere near the minimum and then maybe it's time to change from a conjugate-gradient (INBRION = 2) to a quasi-Newton (IRBION = 1) algorithm?

Thank you.
Regards,
FIlippo

I am very interested in this as well. Please correct me if I am wrong, but I usually switch from CG to QN when values next to harm and opt step (in stdout) are similar. Usually, I let the CG algorithm to run until I get a ZBRENT warning and then switch to QN.