Discrepancy in Prefactor Calculations using LAMMPS and Our MD Code

To whom it may concern,

Recently we have been trying to calculate the prefactors using Vineyard equation for single vacancy diffusion in Ion. The potential we used is Mendelev 2003 (DOI: 10.1080/14786430310001613264). We have been using both LAMMPS and our MD code (Loughborough Molecular Dynamics code, LBOMD) to evaluate the second derivatives in the Hessian matrix. However, when we increased the inclusion radius (the larger the inclusion radius, the more atoms are included in the evaluation of Hessian matrix), we cannot get convergence in LAMMPS. However, we managed to get the convergence using LBOMD. The attachment is the results we got. Do you have any idea that what might lead to this discrepancy?

Best regards,

Ying

prefactor.pdf (28.6 KB)

Ying,

Perhaps LBOMD uses an analytical EMA, while LAMMPS uses the spline version? Are you doing quasi-harmonic calculations?

BTW, is LBOMD publicly available?

Bruce

To whom it may concern,

Recently we have been trying to calculate the prefactors using Vineyard equation for single vacancy diffusion in Ion. The potential we used is Mendelev 2003 (DOI: 10.1080/14786430310001613264). We have been using both LAMMPS and our MD code (Loughborough Molecular Dynamics code, LBOMD) to evaluate the second derivatives in the Hessian matrix. However, when we increased the inclusion radius (the larger the inclusion radius, the more atoms are included in the evaluation of Hessian matrix), we cannot get convergence in LAMMPS. However, we managed to get the convergence using LBOMD. The attachment is the results we got. Do you have any idea that what might lead to this discrepancy?

results without the specifics of the methods used are pretty much
useless. for example, how did you compute the hessian matrix in
LAMMPS? are these serial or parallel calculations?
also, while i know a lot about LAMMPS, i know nothing about your
in-house code, so i cannot say much about the differences. OTOH,
LAMMPS is publicly available and extensively documented, so you would
be in a much better situation to do that comparison than people here.

axel.