I am trying to find elastic constants of a triclinic structure using Lammps which I have compiled with kokkos and cuda support and pair nequip allegro.
I am using the default Elastic Constant at 0 K script given by lammps.
But when running lammps with the trained potential, for every run, I get different elastic constant values and the Elastic Constants calculated are either very high value or negative values.
Please tell a solution for this issue.
This is not part of LAMMPS, so you need to contact the developers of that pair style.
Your description is far too vague to give any kind of advice and the fact that you are using a non-canonical pair style makes it impossible for any user of the standard LAMMPS distribution to reproduce your issue, even if you would make the sufficient data available.
Does running Lammps on GPU make any difference in the accuracy of calculations or the precision as compared to running on CPU?
For the GPU styles that are supported by LAMMPS there is a discussion in the manual. But since you are not using a pair style that is part of LAMMPS, your posts are off-topic here.
Thus I am closing this discussion. Please consult with the developers of the pair style you are using.