Segmentation fault during LAMMPS-MACE-CPU run

[fcannizzaro-byte]

I am encountering a segmentation fault while running a LAMMPS simulation using MACE. The error occurs in the PairMACE::compute() function. Below are the details and steps to reproduce the issue.

Steps to Reproduce:
Compile LAMMPS with the MACE potential as described in the MACE documentation.
Use the attached input files for the simulation setup.
Execute the LAMMPS simulation.

Error Output:
Signal: Segmentation fault (11)
Signal code: (-6)
Failing at address: 0x62bb003992ab

==== backtrace (tid:3773099) ====
0 0x0000000000054db0 __GI___sigaction() :0
1 0x00000000008bb323 LAMMPS_NS::PairMACE::compute() /path/to/lammps/src/ML-MACE/pair_mace.cpp:95
2 0x00000000005c21fa LAMMPS_NS::Verlet::setup() /path/to/lammps/src/verlet.cpp:140
3 0x000000000055963f LAMMPS_NS::Run::command() /path/to/lammps/src/run.cpp:171
4 0x000000000045c46c LAMMPS_NS::Input::execute_command() /path/to/lammps/src/input.cpp:853
5 0x000000000045cc3e LAMMPS_NS::Input::file() /path/to/lammps/src/input.cpp:302
6 0x000000000043c90d main() /path/to/lammps/src/main.cpp:105
7 0x000000000003feb0 __libc_start_call_main() ???:0
8 0x000000000003ff60 __libc_start_main_alias_2() :0
9 0x000000000043d8d5 _start() ???:0

Environment:

LAMMPS version: Large-scale Atomic/Molecular Massively Parallel Simulator - 28 Mar 2023 - Development
Git info (mace / 184b6fd)
Hardware: HPC CPU (Viper cluster from MPCDF)
Compiler: gcc/13

Attachments:
in.lmps (input file)
pair_mace.cpp (source file with the function causing the issue)
err.err (full error log)
test_my_mace_model.zip

[wcwitt]

It looks like you are using multiple MPI ranks, which is incompatible with no_domain_decomposition. We should probably detect this and produce an error.

In the meantime, does it work with --ntasks-per-node=1 in your slurm script?