Force accuracy

Message

dcurulla · #1 Post by **dcurulla** » Mon Jul 13, 2009 9:24 am

Hi, I am interested in estimating the error in the vibrational frequencies. For that, I need to know what the accuracy is in the calculation of the forces acting on the atoms. Could anyone tell me where I can get this info? Is it printed in the OUTCAR file? Is maybe the "convergence-correction" printed in the table "Forces acting on ions"?

I have a second question; in the VASP workshop documentation, there is a paper by Prof. Kresse (Accurate force calculations and VASP datasets) where he explains the calculation of phonons from finite differences. There, he gives an equation that I presume it gives the errors in the forces (f) given the fmax, f and the fractional error in forces 'e'. Then, he says that the attainable precision in e is 1e-4. Could anyone tell me how can 'e' be calculated from the OUTCAR file?

Thanks :?

#2 Post by **admin** » Thu Jul 16, 2009 9:15 am

the accuracy of the forces (given in eV/Ã…) can be chosen via EDIFFG (set to negative numbers), so you are free to determine the level of accuracy.
Please note that
1) the precision of the electronic convergence (EDIFF) should be at least one order of magnitude higher.
2) the frequencies are calculated within the harmonic limit. Therefore please choose a small POTIM (not larger than 0.015) in order to be sure there are no anharmonic contributions to the frequencies

dcurulla · #3 Post by **dcurulla** » Mon Aug 10, 2009 10:21 pm

Hi, as long as I understand, I can set up the convergence criteria for geometry optimizations in eV/A using EDIFFG and negative values. However, this does not correspond to the numerical error on the forces, does it? I mean, if I get a force of 0.007896 on atom A coordinate X using EDIFFG = -0.02, I know that my geometry calculation has converged, but what is the numerical error on 0.007896? e.g. Is it 0.0079 (+/-) 0.0001?
Thanks

pkroll · #4 Post by **pkroll** » Fri Aug 14, 2009 7:12 pm

First, as always there's an error of the method.
Second, your forces carry an error due to limited ENCUT and choice of k-points, electronic convergence but also due to the scheme (projector, real-space, etc) you are choosing.
While this is something you can estimate yourself through different calculations, the impact of this numerical errors will depend very much on the CHANGES in forces that come within the finite difference method. If these changes are much larger than your "error", you're safe ...
On the bottom line: high frequencies (wavenumbers) are ususally not much affected, since the corresponding mode has a large force constant.
Low frequencies have to be taken with care. How much care? Depends on what you want to do with them ....