GROMACS version: .tpr with 2020.4 , setup with older version.
GROMACS modification: Yes
Here post your question : I am doing a 900ns simulation , max time allowed in my server is 2 days ; so I am dividing it up into 20ns simulations . Is there an easier way to start the next 20 nano second simulation without having to rename the mdp names ? for the next 20 ns simulation do I use the previous .cpt file as an input for mdrun -deffnm ?
Hi,
You can continue a simulation from its last checkpoint with the mdrun flag -cpi. You should then also ensure that Gromacs stops before the alloted time with -maxh 48 (to stop it before 48 hours have passed). With this you won’t have to manually split the simulation into smaller parts, just let Gromacs do that work for you.*
So, simply change your run command to gmx mdrun -cpi -maxh 48 (plus any other flags).
Petter
* If you absolutely do want to split the simulation into 20 ns steps by yourself you can use the final saved checkpoint as input for your new simulation when running grompp, use the -t flag to select the .cpt file from your previous step (by default this is state.cpt). But unless you really want to do this I’d say it’s much worse than just extending the simulation with -cpi: it’s much easier to make a mistake when running grompp, and you need to keep track of all simulation files by yourself and manually stitch trajectories and energy files.
If one does split the simulation into 20 ns steps and use grompp -t *.cpt file from the previous step, is the trajectory somehow inaccurate?
In the GROMACS 2022 documentation under gmx grompp, it is stated that " gmx grompp can be used to restart simulations (preserving continuity) by supplying just a checkpoint file with -t. However, for simply changing the number of run steps to extend a run, using gmx convert-tpr is more convenient than gmx grompp. You then supply the old checkpoint file directly to gmx mdrun with -cpi. If you wish to change the ensemble or things like output frequency, then supplying the checkpoint file to gmx grompp with -t along with a new .mdp file with -f is the recommended procedure. Actually preserving the ensemble (if possible) still requires passing the checkpoint file to gmx mdrun -cpi."
Why does preserving the ensemble require passing the checkpoint file to gmx mdrun? I am just wondering why wanting to change something like output frequency would mean that it is necessary to sacrifice the correctness of the ensemble. Would it still be suitable for equilibrating a system to break a simulation down into 20 ns simulations with gmx grompp -t new mdp file without passing -cpi to gmx mdrun?