Post slurm requeue restarting without a checkpoint file but with .xtc file

User discussions
1

GROMACS version: 2023.3
GROMACS modification: No

I have come across posts that help with restarting without a cpt file (although I am still confused), but I wanted to check if it was okay to do so in my case, or if I should just start from the beginning again (although this simulation had already run for 18 days).

I am using a HPC with a slurm manager which paused my sbatch mdrun submission but then restarted it from 0. This led to my checkpoint being overwritten, but thankfully y’all programmed GROMACs to not overwrite the xtc, edr, and logs so I still have those files from the partially run mdrun.

The GROMACS sites notes that restarting from a checkpoint is the only recommended way of restarting a run, but I will be using a different “computer” to restart regardless because I am using slurm manager.

Keeping the above in mind, am I okay to extract a .gro file from my .xtc file and then to continue the simulation?
If so is gmx traj or gmx grommp the suggested method for this? Found posts pointing to both. Log is attached for background. Please let me know if you need any other information.
md.log (3.9 MB)

Thanks for your time and help.

Sincerely,

David

2

You can start the simulation from the last snapshop, but it wont be the excact continuation of the existing trajectory.

You may use the -noappend with mdrun so that everytime new files are written and checkpoint files are not overritten.

3

Thanks for your response! How do I best generate the snapshot with the .xtc/.edr/.log files but not .cpt file? Good to know about the -noappend option, will add that in the future. Thanks for your time and help.

4

I would choose two option depends on the type of simulations and what i need etc.
you can generate the last frame from the gmx trjconv and start simulation from that snapshop. You dont need cpt for generating snapshot

5

Thanks for the help! I ran the following scripts with md.mdp in the 2nd script being the original mdp file with init-step set to 25000000 (50 ns with a dt of 2 fs) does this seem right?

gmx trjconv -f md.xtc -s …/md.tpr -dump 50000 -o 50ns.gro
gmx grompp -f md.mdp -c 50ns.gro -o redo.md.tpr -p …/topol.top

Then just run the continuation with the following?
gmx mdrun -v -deffnm redo.md

And finally concatenate the run:
gmx trjcat -f md.xtc md_20.redo.md.xtc -o final.xtc