prefactor calculation

[mr.precog]

Hello everybody!

I am studying the system bcc-tungsten with a trivacancy. The goal is to find the prefactor, so I first perform a NEB calculation in order to fix the saddle point. Next I run the dymseldsp script and the DISPLACECAR file is obtained, so I now proceed to run VASP again. Finally the dymmatrix script and the dymprefactor script are used and I get the prefactor.

However, when I use the dymseldsp script and the number of atoms to be displaced is specified I do not know how to choose it. First I choose the nearest neighbors around the trivacancy and 0.001 the displacement. My first election was 6 atoms (NSW = 19) and the prefactor obtained 1.67E13 Hz. Everything was alright because the equilibrium frequencies were all real and the saddle frequencies all real but first one (as indicated in VASP manual). Afterwards I did a new calculation with 12 atoms displaced (NSW = 37) and the new prefactor was 1.19E13 Hz so the difference is nearly 30 % if compared both results. My next try was using 18 atoms displaced (NSW = 55) but then two saddle point frequencies were imaginaries and no prefactor could be obtained. With NSW = 67 (22 atoms displaced) two imaginary saddle point frequencies were also obtained and again three imaginary saddle point frequencies for NSW = 91 (30 atoms displaced) in addition to one imaginary equilibrium frequency.

So these are my questions: Is it critical the number of atoms displaced to be chosen? How does the prefactor depend on that number? Am I losing my time when the NSW parameter increases?

Thanks in advance
Sergio L. Palacios

[graeme]

In order to trust a prefactor calculation it really needs to be converged with respect to the number of atoms included in the calculation. It doesn't make sense to select some special small number of atoms which give a reasonable prefactor if a larger number of atoms gives a different value.

Prefactor calculations can be tricky. My strongest advice is to make sure that your forces are converged using, for example, EDIFF=1e-8. Also, make sure that the subset of atoms included in the prefactor calculation are well-converged both at the minimum and saddle point. You should be able to reach a convergence criterion of 0.01 eV/Ang. If you are finding two negative modes at the saddle, you can focus on that problem; make sure that the saddle is converged well enough so that you get a single negative mode with a substantial number of atoms in the Hessian calculation.

[mr.precog]

Thank yoy very much for your fast response. I will try to perform my calculations taking into account your advice.

Regards,
Sergio L. Palacios