FAQs#
My LAMMPS run dies immediately with a pair_style / pair_coeff error.
The most common cause is that the pair_coeff element list does not match the element list (or is in a different order). For multi-element systems calphy enforces strict ordering: element[i] is mapped to LAMMPS atom type i+1, and the elements at the end of pair_coeff must appear in the same order. Reorder both lists so they agree.
calphy_kernel complains that the simulation folder already exists.
calphy refuses to overwrite an existing run folder so that data is never silently lost. Either remove the offending mode-lattice-temp-pressure directory or set folder_prefix to a unique tag.
My liquid free-energy calculation reports MSD = 0 / spring constant warnings.
This happens when the structure has not actually melted during the equilibration cycle. Increase temperature_high (calphy will overheat to this value before quenching) and/or set melting_cycle: True.
My liquid free-energy calculation loses atoms during the integration (ERROR: Lost atoms).
This is most common for molecular liquids (e.g. water) where the single-component Uhlenbeck–Ford reference cannot represent two well-separated length scales. There are two distinct failure modes:
Detonation — the UFM energy spikes just before the crash. The single
sigmais larger than a bonded distance, so bonded pairs sit inside the UFM repulsive core and are blown apart. Use the two-leg path with a per-element-pairsigmadict and a small cross-term (e.g.H_O: 0.9) so bonded pairs stay outside the core.Drift — atoms wander off slowly with no energy spike. The UFM is purely repulsive, so once the real potential is switched off the reference must confine the atoms on its own; a too-small
p(the confinement ispin units ofkB·T) lets light atoms escape. Increasep(the default50is usually sufficient).
The two effects are independent: sigma controls detonation, p controls drift. See the uhlenbeck_ford_model block for the two-leg setup.
How do I run calphy with the GPU / KOKKOS package?
Use the md.cmdargs keyword to pass the standard LAMMPS accelerator command-line arguments:
md:
cmdargs: "-k on g 1 -sf kk -pk kokkos newton on neigh half"