LAMMPS tips¶

Choose where to compile LAMMPS¶

If the code must be accessed only by you, you can save it in your /home/ folder on Dahu/Bigfoot. If you want the code to be available to all the member from your group, you can use /bettik/PROJECTS/. For instance, for the members of tamtam:

/bettik/PROJECTS/pr-tamtam/

Download LAMMPS¶

From the LAMMPS webpage, download the version of your choice. For instance, from the terminal, download the last stable release:

wget https://download.lammps.org/tars/lammps.tar.gz
tar -xzvf lammps.tar.gz

Compile LAMMPS with make¶

For compiling in serial, simply navidate to the src/ folder, and type:

make serial

For compilling in parallel, you will need to load manually openmpi and gcc using NIX:

source /applis/site/nix.sh
nix-shell -p openmpi -p gcc

Then, from the src/ folder, type:

make mpi

Compile LAMMPS with cmake using guix environment¶

source your guix environment and install the following package(specified the openmpi version, some issues were observed on dahu cluster with openmpi 5)

source /applis/site/guix-start.sh
guix install cmake
guix install glibc-locales
guix install gcc-toolchain
guix install openmpi@4.1.6
guix install clang
guix install ffmpeg

navigate into your lammps home, create a folder build/. From the build/ folder, compile LAMMPS:

mkdir build
cd build
cmake ../cmake -D PKG_MOLECULE=on -D PKG_KSPACE=on -D PKG_RIGID=on

It creates an executable lmp in the directory. You can add as many packages as you want and you can create different build if you need it.

Run LAMMPS¶

Create a bash file named sub.sh with the following content:

#!/bin/bash
#OAR -n test
#OAR -l /nodes=1/cpu=1/core=4,walltime=12:00:00
#OAR --stdout log.out
#OAR --stderr log.err
#OAR --project tamtam

# Path to the LAMMPS executable
lmp=/path/lmp_mpi

mpirun -np 4 ${lmp} -in input.lmp

where input.lmp is your LAMMPS input file, and where the project was assumed to be tamtam (to adapt to your case). Here, 4 CPU cores are requested, as well as a total duration of 12 hours. Then, make the file file executable with chmod and launch it using:

chmod +x ./sub.sh
oarsub -S ./sub.sh

Launch multiple jobs using bash¶

Assuming that one has the job submission script named sub.sh with name lmp-myvariable-0, that launch a lammps input script named input.lmp by passing a variable named myvariable into it. Additionally, here the job ID is used as an input for the random seed, allowing for example different initial configurations:

#!/bin/bash
#OAR -n lmp-myvariable-0
#OAR -l /nodes=1/cpu=1/core=4,walltime=12:00:00
#OAR --stdout log.out
#OAR --stderr log.err
#OAR --project tamtam

# Path to the LAMMPS executable
lmp=/path/lmp_mpi

myvariable=0

# Run LAMMPS using MPI, with 4 processes, using the input from 'input.lmp'
mpirun -np 4 ${lmp} -in input.lmp -var nb2 ${myvariable} -var seedin $OAR_JOBID

If one wants to launch the current job, one simply have to type:

chmod +x sub.sh
oarsub -S ./sub.sh

and a single job with name lmp-myvariable-0 will be send. To launch multiple simulations with different values of myvariable, say 0, 1, and 2, one can create a second bash script, named multi-sub.sh, and containing:

#!/bin/bash
set -e

for myvariable in 0 1 2
do
    # deal with OAR -n
    newline='#OAR -n lmp-myvariable-'$myvariable
    oldline=$(cat sub.sh | grep '#OAR -n lmp-myvariable-')
    sed -i '/'"$oldline"'/c\'"$newline" sub.sh
    # deal with myvariable
    newline='myvariable='$myvariable
    oldline=$(cat sub.sh | grep 'myvariable =')
    sed -i '/'"$oldline"'/c\'"$newline" sub.sh
    chmod +x sub.sh
    oarsub -S ./sub.sh
done

The newline command creates a new line that will replace the line containing myvariable in the script sub.sh The oldline= command finds the current line in sub.sh that contains ‘myvariable =’, storing it in the variable oldline. This assumes there is exactly one such line, otherwise the behavior may be unexpected. Then, sed is used to replace the old line with the new line (newline) in sub.sh.

Then, simply run multi-sub.sh by typing:

bash multi-sub.sh