Page 1 of 1

Criteria to estimate the convergence of force

Posted: Thu Apr 15, 2010 2:07 am
by 2008zdd
Dear Gaeme,
I have performed a NEB run (actually still running), now I get a result of force values (the last iteraction), it is pasted as follows:

TANGENT CHAIN-FORCE (eV/Angst)
-----------------------------------------------------------------------------------
0.00101 -0.00101 0.00000 0.000000 0.000000 0.000000
0.00080 0.01307 -0.01403 0.000000 0.000001 -0.000001
0.00077 -0.00077 0.00000 0.000000 0.000000 0.000000
-0.00101 -0.00101 0.00000 0.000000 0.000000 0.000000
-0.00078 -0.00101 0.00024 0.000000 0.000000 0.000000
-0.01304 0.01304 0.00000 -0.000001 0.000001 0.000000
0.00080 0.01307 0.01403 0.000000 0.000001 0.000001
-0.00078 -0.00078 0.00000 0.000000 0.000000 0.000000
0.00104 0.01285 0.01405 0.000000 0.000001 0.000001
-0.01282 0.01282 0.00000 -0.000001 0.000001 0.000000
0.00101 0.00101 0.00000 0.000000 0.000000 0.000000
-0.01287 -0.01310 -0.00022 -0.000001 -0.000001 0.000000
0.00078 0.00078 0.00000 0.000000 0.000000 0.000000
0.01310 0.01310 0.00000 0.000001 0.000001 0.000000
-0.00078 -0.00101 -0.00024 0.000000 0.000000 0.000000
-0.00101 -0.00078 0.00024 0.000000 0.000000 0.000000
0.01287 0.01287 0.00000 0.000001 0.000001 0.000000
0.00104 0.01285 -0.01405 0.000000 0.000001 -0.000001
-0.00101 0.00101 0.00000 0.000000 0.000000 0.000000
0.01285 0.00104 0.01405 0.000001 0.000000 0.000001
0.70487 0.70487 0.00000 0.000041 0.000041 0.000000
-0.01287 -0.01310 0.00022 -0.000001 -0.000001 0.000000
-0.00077 0.00077 0.00000 0.000000 0.000000 0.000000
-0.01310 -0.01287 -0.00022 -0.000001 -0.000001 0.000000
-0.00101 -0.00078 -0.00024 0.000000 0.000000 0.000000
0.01304 -0.01304 0.00000 0.000001 -0.000001 0.000000
0.01285 0.00104 -0.01405 0.000001 0.000000 -0.000001
0.01307 0.00080 -0.01403 0.000001 0.000000 -0.000001
0.01282 -0.01282 0.00000 0.000001 -0.000001 0.000000
-0.01310 -0.01287 0.00022 -0.000001 -0.000001 0.000000
0.01307 0.00080 0.01403 0.000001 0.000000 0.000001
-----------------------------------------------------------------------------------
CHAIN + TOTAL (eV/Angst)
----------------------------------------------
-0.00889 0.01138 0.00000
-0.00289 0.00040 0.00327
0.00884 -0.01135 0.00000
0.00537 0.00537 0.00000
0.00047 -0.00047 0.01706
0.01445 -0.02031 0.00000
-0.00289 0.00040 -0.00327
-0.00538 -0.00538 0.00000
0.00288 -0.00039 0.00329
-0.01448 0.02035 0.00000
-0.01192 -0.01192 0.00000
-0.00411 0.00413 -0.01645
0.01191 0.01191 0.00000
-0.00908 -0.00908 0.00000
0.00047 -0.00047 -0.01706
-0.00047 0.00047 0.01706
0.00912 0.00912 0.00000
0.00288 -0.00039 -0.00329
0.01138 -0.00889 0.00000
-0.00039 0.00288 0.00329
0.00000 0.00000 0.00000
-0.00411 0.00413 0.01645
-0.01135 0.00884 0.00000
0.00413 -0.00411 -0.01645
-0.00047 0.00047 -0.01706
-0.02031 0.01445 0.00000
-0.00039 0.00288 -0.00329
0.00040 -0.00289 0.00327
0.02035 -0.01448 0.00000
0.00413 -0.00411 0.01645
0.00040 -0.00289 -0.00327
----------------------------------------------

should it be converged? So how to decide whether it is converged from the above result?

Re: Criteria to estimate the convergence of force

Posted: Thu Apr 15, 2010 5:17 am
by graeme
If you have our scripts, you can run nebef.pl in the neb directory to get a list of the final energies and forces for each image (the largest force on any atom).

Re: Criteria to estimate the convergence of force

Posted: Thu Apr 15, 2010 8:33 am
by 2008zdd
Thanks, I got great help. Now I have another question: If I want to calculate prefactor, is it necessary to displace all the atoms in the system to get the accurate prefactor?

Re: Criteria to estimate the convergence of force

Posted: Thu Apr 15, 2010 4:39 pm
by graeme
You can select the atoms which have moved the most in your calculation, and displace those and their neighbors. We have a script for making this selection. Then you have to check that as you include more atoms (for example, with a smaller distance criteria) the prefactor converges.

In practice though, it is usually easier to just displace all moving atoms in the calculation. Even with 100 atoms, this is only 300 force calls which can be done in parallel.

Re: Criteria to estimate the convergence of force

Posted: Fri Apr 16, 2010 1:31 am
by 2008zdd
Thanks a lot for the suggestions. I am learning it.