Dear Lammps Users,
I am running a simulation to compute msd for 10ns. However, the simulation terminated at 5ns because of exceeding the time limit in the cluster. Is there a way that I can continue the simulation to compute the msd from where it terminated? (I have only a dump file)
Thanks.
Dear Lammps Users,
I am running a simulation to compute msd for 10ns. However, the simulation terminated at 5ns because of exceeding the time limit in the cluster. Is there a way that I can continue the simulation to compute the msd from where it terminated? (I have only a dump file)
you could restart the simulation itself by loading your original configuration and then update the coordinates via read_dump, provided you saved imageflags or unwrapped coordinates.
however, that cannot restart/continue the MSD calculation and - unless you saved the atom velocities as well - will not be a clean restart.
so most likely your only option is to consider the first 5ns a very thorough equilibration and then continue from the latest (unwrapped) geometry and this time use the “restart” command to regularly write out restart files during the run, so you can continue smoothly when your calculation is terminated unexpectedly the next time.
also, it is always a good idea to use the equilibration part of your simulation to gather information about the expected performance, so you can set up your simulation so that it runs in chunks that fit into your maximal allotted time on the cluster and you have much cleaner restart. in fact, for many queuing systems it may give you more throughput and shorter wait times, if you reduce the requested wall time for your jobs and run multiple smaller chunks instead of a few large chunks.
axel.
axel.
Thanks, Dr. Axel.
You need to periodically write out restart files, see https://lammps.sandia.gov/doc/restart.html. Then if the simulation terminates, you can pick up where you left off using the read_restart command.
Stan
Thanks, Stan. This time I followed that procedure when I ran the simulation.