Re: [potfit-users] Inconsistent cut-off in the embedding functions after fit

[Daniel S. <sch...@gm...>]

Hi Sayan,

I am sorry for the delayed answer.

Please note that writing LAMMPS potentials is a not well
tested feature, especially for tabulated potentials. And potfit
allows certain freedoms which are not supported by LAMMPS,
so this makes converting the potentials a little bit tricky.
If you find any issues please let us know.

I am not sure what you are describing here. Maybe sharing you 
in- and output files might help me better understand your situation.

Also I am not aware of any tricks for tabulated potentials. For analytic
potentials there is something like a cutoff function but not for tabulated 
potentials. You can set the gradient at the cutoff (which you probably did) to 
0 and that should usually give you something like a vanishing potential
at the cutoff distance.

Rescale is definitely something to help fix the gauge degrees of freedom and
the punishments also tries to do this. They are not mutually exclusive, even
when rescaling is enabled we still place some restrictions on the potential
and punish it for bad behavior.

Daniel


Am Montag, 15. März 2021, 21:01:43 CET schrieb Sayan Samanta:
> Hi,
> 
> I'm running potfit-20201014 with mpi, stress and rescale enabled. At the
> end of the fit, potfit is unable to write LAMMPS potential because of
> inconsistent cut-offs.
> 
> I checked the final potfit potential and indeed the last x value for the
> embedding function is not the same (not 1.0).  I began with the starting
> potential that did not have that inconsistency, in fact, another fit
> without the rescale option worked fine.
> 
> Did I do something wrong? I could share the output files that might help
> you diagnose a fix. Also is that 0 energy value at a large distance (for
> the pair and transfer function) a trick to ensure smooth cut-off to zero in
> the tabulated case?
> 
> Lastly, am I correct to think that both rescale and punishment are
> different tactics to fix the gauge of the potential and that if we are
> rescaling we don't need punishment and vice versa?
> 
> Best
> Sayan
> 
> Sayan Samanta
> Ph.D candidate
> Materials Science | School of Engineering
> Brown University
> Providence, RI 02912 USA
> +1 (401) 699-1340