Instructions on how to install lammps using cmake are here: https://lammps.sandia.gov/doc/Build_cmake.html and all sorts of package options are here: https://lammps.sandia.gov/doc/Build_package.html

In general, it follows the very same (git) recipe for any installation:

cd programs/
git clone  https://github.com/lammps/lammps.git
cd lammps/
mkdir build
cd build/
cmake ../cmake -DCMAKE_INSTALL_PREFIX="~/Programs/lammps-19Sep2019" -DPKG_MOLECULE=yes -DPKG_OPT=yes
make install

Lammps stores its cmake files NOT in the main directory but in a subdirectory called cmake, so ../cmake has to be used to tell cmake where to look for them. Use ccmake . after you executed cmake to see all the options ( -DPKG_…=yes) there are for configuring lammps packages.

You can do make install -j4 to use four cores for compiling, it should be faster. The lammps executable will be in the folder '~/Programs/lammps-19Sep2019/bin' and is commonly called 'lmp'.

Lammps will need MPI to run efficiently, so load the appropriate modules on the cluster. For compiling Lammps with MPI add the flag -DBUILD_MPI=yes to the cmake above. When running the cmake command, it should include output like

-- Found MPI_CXX: /usr/local/Cellar/open-mpi/4.0.5/lib/libmpi.dylib (found version "3.1") 
-- Found MPI: TRUE (found version "3.1") 

After compiling, the executable now should be able to be run in parallel

 mpirun -n 2 ~/Programs/lammps-19Sep2019/bin/lmp -i lammps.init

without modificiations to the input scripts. It should tell you 1 by 1 by 2 MPI processor grid and 97.1% CPU use with 2 MPI tasks x no OpenMP threads or similar when you run it.

Even though installation and cmake outputs seem normal mpirun -np 4 may run 4 of the same simulation instead of parallelizing the same simulation by 4. If that is the case below can be the solution:

• Remove the current lammps build
• Reinstall/install open-MPI with sudo apt-get install openmpi-bin libopenmpi-dev
• Rebuild lammps as before

To get started, this is a typical lammps input file:

# Create simulation box and initialize system
units       	lj
atom_style 	atomic
boundary    	p p p
pair_style lj/cut 1.12246
pair_modify 	shift yes
region box 	block 0 13.6798 0 13.6798 0 13.6798 units box
create_box	1 box

# Set model parameters
mass 		1 1
pair_coeff 	1 1 1.0 1.0 1.12246

# Run settings
timestep      	0.005
run_style     	verlet

# Create atoms
create_atoms 	1 random 2048 59175 box

# Output and dump settings
thermo 		10000
thermo_style 	custom step temp epair pe etotal lx ly lz vol pxx pyy pzz press atoms
thermo_modify 	norm no
fix 		thermo_print all print 10000 "$(step) $(temp) $(epair) $(pe) $(etotal) $(lx) $(ly) $(lz) $(vol) $(pxx) $(pyy) $(pzz) $(press) $(atoms)" append thermo.out screen no title "# step temp epair pe etotal lx ly lz vol pxx pyy pzz press atoms"

dump 		1 all custom 100000 tg.lmpdump id type x y z
dump_modify 	1 sort id append yes
restart 	50000 restart.1 restart.2

# Apply initial velocities, minimize initial config, define integrator
neigh_modify every 1 delay 0 check yes
velocity 	all create 1.0 73241 dist gaussian units box
minimize 	1.0e-4 1.0e-6 1000000 100000000

# Equilibrate
fix 		nvt all nvt temp 1.0 1.0 1.0
fix 		mom all momentum 1 linear 1 1 1 rescale
write_data 	data.init
run 		1000000
write_data 	data.inter
write_restart 	restart.inter

It will run a WCA (i.e, LJ with cutoff 1.12246 shifted to zero at cutoff) fluid with 2048 particles in a 13.6798^3 box.

Useful introduction to run the first simulation and understand input scripts/outputs : https://icme.hpc.msstate.edu/mediawiki/index.php/LAMMPS_Polymer.html