Molecular dynamics

To run molecular dynamics with LAMMPS, appa provides automatic input generation through the command appa lammps, which takes an initial .xyz file (INITIAL) and generates a LAMMPS input file and LAMMPS .data initial structure file.

Usage: appa lammps [OPTIONS] INITIAL

  Write LAMMPS simulation inputs.

Options:
  --architecture TEXT  appa-supported architecture (mace-mliap, grace, mtt,
                       nequip...)  [required]
  --model TEXT         Path to model  [required]
  --steps INTEGER      Number of steps to run  [default: 1000]
  --temperature FLOAT  MD temperature (K)  [default: 300]
  --timestep FLOAT     MD timestep (ps)  [default: 0.0005]
  --dump-freq INTEGER  How many steps between saving frames to the dump file
                       [default: 20]
  --plumed-file TEXT   Path to PLUMED input file
  --help               Show this message and exit.

The PLUMED file is optional.

To run LAMMPS you can use a job like this:

#!/bin/bash
#SBATCH --job-name=lmp
#SBATCH --partition=gpu_a100
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=18
#SBATCH --gpus=1
#SBATCH --time=1-00:00:00

module purge
module load 2025
module load OpenMPI/5.0.7-GCC-14.2.0
module load CUDA/12.8.0
module load CMake/3.31.3-GCCcore-14.2.0
module load OpenBLAS/0.3.29-GCC-14.2.0
module load FFTW.MPI/3.3.10-gompi-2025a

source ~/.bashrc
conda activate grace
which python

WORK_DIR="${PWD}"
OUTPUT_PATH="${PWD}/${SLURM_JOB_ID}"
LOCAL_PATH="/scratch-local/$USER/${SLURM_JOB_ID}"

# --- Prepare scratch ---
mkdir -p $LOCAL_PATH
cd $LOCAL_PATH || exit 1
cp -r $WORK_DIR/* $LOCAL_PATH/

# --- Generate input files for LAMMPS ---
appa lammps initial.xyz --architecture grace --model ~/train/seed/1/final_model --steps 20000
srun /home/ldkam/lammps/build/lmp -in input.lmp

# --- Copy results back & clean ---
mkdir -p "$OUTPUT_PATH"
cp -r $LOCAL_PATH/* $OUTPUT_PATH
rm -rf $LOCAL_PATH

For MACE, you need a different command to run LAMMPS, and you need to convert the .model file to a MLIAP-LAMMPS interface model. The model conversion needs to be run on a GPU with CUDA available. So better include it in the job:

mace_create_lammps_model path/to/mymace.model --format=mliap
srun /home/ldkam/lammps/build/lmp -k on g 1 -sf kk -pk kokkos newton on neigh half -in input.lmp

If you installed NequIP without kokkos then you should be able to use the same command as for GRACE. Otherwise see the NequIP LAMMPS interface repo.

From the MD you get a lammps.dump file which you can analyze further. TODO: add a CLI tool to convert to XTC.

Output file conversion¶

appa contains a handy tool to convert big LAMMPS dump files and/or XYZ files to the compressed XTC format. The XTC file then only contains the atomic positions in 5-decimal precision (and always needs the corresponding topology file system.data to be interpreted). For XYZ trajectories, a system.data file is generated.

Usage: appa convert xtc [OPTIONS]

  Convert XYZ or LAMMPS dump trajectories to XTC.

Options:
  --pattern TEXT             Glob pattern for input files (e.g., './*/*.xyz').
                             [required]
  --format [xyz|lammpsdump]  Input format.  [required]
  --nprocs INTEGER           Number of parallel processes.  [default: 1]
  --help                     Show this message and exit.