Hello,
I am trying to obtain a MEAM potential that fits some experimental values, especifically the elastic constants, I have manage to almost reproduce the crystalline structure and the ground state energies, I have been thinking about how to do this fitting so any suggestion is welcome.
I have seen that there is a tool here https://icme.hpc.msstate.edu/mediawiki/index.php/MPCI but I have found no way to download the file, even that I am already a registered user.
I consider myself as a begginer user of LAMMPS so perhaps can someone tell me if there is a way of run a LAMMPS script backwards, this in order to imput the results and obtain the original parameters of the potential file, especially the screening parameters.
Thank you for your help
Mario Muralles
Materials, Theory and Technology Laboratory
Department of Mechanical Engineering
Kyung Hee University, South Korea
If by running “backwards” you mean use an MD code
to fit parameters for a potential, then yes people do that.
That is an entire field of research unto itself, using
various optimization and sensitivitiy studies. You should
look in the literature. You won’t learn how to do it from
a mail list.
Steve
to expand a little bit on steve’s remark, i would also recommend, that you first dig a little deeper into statistical mechanics and statistical thermodynamics and have a look at the relation between macroscopic properties of a material and atomistic simulations.
on important aspect of (empirical) force field parameterization, is that that you are reducing a very large amount of information to a few parameters. that means that when doing simulations with these parameters, you will not be able to reduce all aspects perfectly, since you cannot recover information that was lost. the trick is now to find a set of parameters that is “good enough” (i.e. still reproduces overall structure and trends) and “balanced” (i.e. the errors for most properties are comparatively small). but this also leads to the problem that, “good enough” and “balanced” are terms that (some) humans can handle, but computers are not really good at. thus a lot of legwork in deriving parameters is just trying things out and getting a feel for what works well and what not so well. there are different tools that follow different recipes, but a large amount of parameterization is done only semi-automatically by people running lots of test calculations on many small test systems while modifying parameters until they are satisfactory or it is declared impossible. that requires a lot of experience in running and particularly analyzing MD simulations. that is not something a beginner should try to do. best you partner up with somebody that does have experience in that field.
axel.