I am trying to calculate some MD simulations using MACE ML potential by applying KOKKOS with LAMMPS. My running command is as
lmp -k on g 1 -sf kk -pk kokkos newton on neigh half comm host
when I use fix move, fix ave/atom, fix stress/atom, I receive warnings like
WARNING: Fix ave/atom not compatible with sending data in Kokkos communication (src/KOKKOS/comm_kokkos.cpp:934)
WARNING: Fix with atom-based arrays not compatible with sending data in Kokkos communication, switching to legacy exchange/border communication (src/KOKKOS/comm_kokkos.cpp:941)
Fix move not compatible with sending data in Kokkos communication (src/KOKKOS/comm_kokkos.cpp:934)
After these warnings the job is terminated. Therefore, I am reaching out to you to ask whether anybody knows how I can overcome these issues.
There are three specific notes about your message:
there is no fix stress/atom, you probably mean compute stress/atom
neither fix ave/atom nor fix move have yet been ported to KOKKOS, so the warnings you see about switching to legacy communication are to be expected
the important questions now are: a) does the run also terminate without messages when you remove these commands? b) does the run also terminate if you use a different pair style?
Some general remarks:
please see the sticky post about guidelines and suggestions for what additional information to provide when reporting LAMMPS not working as expected. You don’t provide anything here.
sometimes output is buffered and the buffers not flushed on error, especially if those errors orginate outside of the LAMMPS code. You can turn off buffering with the -nb command line flag. That can sometimes recover output that you don’t get to see otherwise
please note that MACE is not part of LAMMPS and maintained and distributed separately, so any issues that cannot be reproduced with a pair style that ships with LAMMPS are a topic to discuss with the MACE developers and post in their forum.
Thanks for your prompt and accurate reply, and apologize for my improper question writing format and style.
I used this version of LAMMPS with noted setup based on my .log file, as mentioned below:
# LAMMPS (30 Mar 2026 - Development - patch_30Mar2026-1074-g7f680de296-modified)
KOKKOS mode with Kokkos version 5.1.99 is enabled
using double precision
using view layout = legacy
will use up to 1 GPU(s) per node
using 1 OpenMP thread(s) per MPI task
package kokkos
package kokkos newton on neigh half comm host
In the case of fix stress/atom` I made a mistake and the correct command was compute stress/atom as you correctly surmised.
To answer your questions:
No, if I remove those command lines, the LAMMPS calculation will not be terminated and it will continue.
If I use another pair style with different potential type, It will properly work. Indeed, I checked it by replacing my MACE potential with EAM potential, and it worked without any problem.
That isolates the MACE potential as the trigger for this unexpected behavior. This issue thus falls into the responsibility of the MACE developers unless they can prove independent from using MACE that the flaw is in the LAMMPS source code.