Multiple trajectories + checkpointing with OpenMM

sbatch.sh

#!/bin/bash

#SBATCH --partition=long
#SBATCH --job-name=<job-name>
#SBATCH --output=<job-name>-%j.stdout
#SBATCH --error=<job-name>-%j.stderr
#SBATCH --time=24:00:00
#SBATCH --qos=normal
#SBATCH --nodes=1
#SBATCH --mail-type=ALL
#SBATCH --mail-user=<your_email>
#SBATCH --ntasks-per-node=10
#SBATCH --gres=gpu:6

export PYTHONUNBUFFERED="true"
module load cuda60/toolkit/6.0.37
cd $SLURM_SUBMIT_DIR
HOSTNAME=`hostname`

for i in `seq 0 5`; do
    cd $SLURM_SUBMIT_DIR
    mkdir -p $HOSTNAME-$i
    cd $HOSTNAME-$i
    python ../simulate.py $i ../setup/system.xml ../setup/state.xml ../setup/integrator.xml ../setup/initial.pdb \
	> simulate.$SLURM_JOB_ID.$i.stdout \
       2> simulate.$SLURM_JOB_ID.$i.stderr &
done

wait

simulate.py

from __future__ import print_function
import os
import re
import glob
import socket
import argparse
from simtk.openmm import app
import simtk.openmm as mm
from simtk import unit

for i in range(mm.Platform.getNumPlatforms()):
    print('Platform', mm.Platform.getPlatform(i).getName())


reportInterval = int((100*unit.picosecond) / (2*unit.femtosecond))
runLength = int((1*unit.microsecond) / (2*unit.femtosecond))


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument('gpuindex')
    parser.add_argument('system')
    parser.add_argument('state')
    parser.add_argument('integrator')
    parser.add_argument('topologypdb')
    args = parser.parse_args()


    with open(args.system) as f:
        system = mm.XmlSerializer.deserialize(f.read())
    with open(args.integrator) as f:
        integrator = mm.XmlSerializer.deserialize(f.read())
    with open(args.state) as f:
        state = mm.XmlSerializer.deserialize(f.read())
    topology = app.PDBFile(args.topologypdb).topology

    platform = mm.Platform.getPlatformByName('CUDA')
    properties =  {'CudaPrecision': 'mixed', 'CudaDeviceIndex': args.gpuindex}
    
    traj_number = 0
    simulation = app.Simulation(topology, system, integrator, platform=platform,
                                platformProperties=properties)
    simulation.context.setState(state)

    
    existing_trajectories = glob.glob('trajectory-*.dcd')
    if len(existing_trajectories) == 0:
        traj_number = 0
    else:
        last_traj = max(int(re.search('\d+', e).group()) for e in existing_trajectories)
        traj_number = last_traj + 1
        assert os.path.exists('checkpoint-%08d.chk' % last_traj)
        assert os.path.exists('trajectory-%08d.dcd' % last_traj)
        assert os.path.exists('statedata-%08d.log' % last_traj)
        with open('checkpoint-%08d.chk' % last_traj) as f:
            simulation.context.loadCheckpoint(f.read())

    simulation.reporters.append(app.CheckpointReporter('checkpoint-%08d.chk' % traj_number, reportInterval))
    simulation.reporters.append(app.DCDReporter('trajectory-%08d.dcd' % traj_number, reportInterval))

    f = open('statedata-%08d.log' % traj_number, 'w', 0)
    simulation.reporters.append(app.StateDataReporter(f, reportInterval,
        step=True, time=True, potentialEnergy=True, totalEnergy=True, temperature=True,
        density=True, remainingTime=True, speed=True, separator='\t', totalSteps=runLength))
    simulation.step(runLength)



if __name__ == '__main__':
    main()

I didn't know the CheckpointReporter. Is it possible to get a particular frame from the chk file to restart a simulation with?