Restarting GROMACS writes prefix.colvars.traj/prefix.colvars.meta.hills.traj from restart point

[sathishdasari]

I am trying to use clover module in GROMACS to perform metadynamics. when I am trying to restart/extend the simulation I observe that the files prefix.colvars.traj/prefix.colvars.meta.hills.traj are written from where I restart the simulation but not having the information from the beginning.

I am using following restart/extend command for GROAMCS

gmx_mpi mdrun -deffnm prod -cpi prod.cpt -append
My colvars.in is the following

colvarsTrajFrequency 500
colvarsRestartFrequency 500
colvar {
    name dist
    width 0.1
    lowerBoundary 0.0
    upperBoundary 2.5
    distance {
        group1 {
            atomNumbers 318 943 1563
        }
        group2 {
            atomNumbers 43 46
        }
    }
}
metadynamics {
   name  meta
   useGrids             on
   colvars              dist
   hillWeight           0.3
   hillwidth            0.1
   newHillFrequency     500
   writeHillsTrajectory on
   wellTempered         on
   biasTemperature      2000
}

I have another question

How do we get the .pmf files with time after the simulation is done instead of writing during the simulation to check the convergence like we do in plumed where we can post process the hills file to write the free energy profile with simulation time.

[giacomofiorin]

Hi @sathishdasari! Thanks for reaching out.

I am trying to use clover module in GROMACS to perform metadynamics. when I am trying to restart/extend the simulation I observe that the files prefix.colvars.traj/prefix.colvars.meta.hills.traj are written from where I restart the simulation but not having the information from the beginning.

Can you please clarify what information is missing? Are you saying that the bias accumulated at the previous state of the simulation is not included?

I am using following restart/extend command for GROAMCS

gmx_mpi mdrun -deffnm prod -cpi prod.cpt -append

This should be the correct way to restart: I am really not a fan of the -append flag: however, that will affect the name of the files that will be written. The CPT file will include the restart information. Look for the following lines in prod.log:

colvars: Loading state from unformatted memory.
colvars:   Restarting collective variable "dist" from value: XXX
colvars:   Restarted metadynamics bias "meta" with step number XXX.

My colvars.in is the following

colvarsTrajFrequency 500

This is okay.

colvarsRestartFrequency 500

This is bad: you are asking Colvars to checkpoint way more often than you would ever use. Please delete that line, and let GROMACS do the checkpointing.

colvar {
    name dist
    width 0.1
    lowerBoundary 0.0
    upperBoundary 2.5

Please make sure that this is well above the upper limit of the distances, otherwise you will need a wall restraint
https://colvars.github.io/gromacs-2025/colvars-refman-gromacs.html#sec:colvarbias_meta_boundaries

I have another question

How do we get the .pmf files with time after the simulation is done instead of writing during the simulation to check the convergence like we do in plumed where we can post process the hills file to write the free energy profile with simulation time.

If you want to save the .pmf files at intermediate stage, use the keepFreeEnergyFiles option:
https://colvars.github.io/gromacs-2025/colvars-refman-gromacs.html#sec:colvarbias_meta_output
Alternatively, since you have writeHillsTrajectory on you will get a file with the extension .hills.traj that is the portion of the "HILLS" file for that job. Concatenate all such files in order and feel free to feed the result to sum_hills.

Hope that helps!

[sathishdasari]

Hi @giacomofiorin
Thank you very much for your response.
I simulated a metadynamics for 100 ns and then I wanted to extend the simulation to 200 ns as I did not see the transition that I am looking for. When I extended the simulation I do not see the information of the previous simulation in those files I mentioned.

prod.colvars.traj

# step         dist
    50000000    2.12371980394234e-01
    50000500    1.91679041070942e-01
    50001000    2.66810160619816e-01
    50001500    2.24911695120361e-01
    50002000    2.01805702861412e-01
    50002500    9.42138976351065e-02
    50003000    2.35211271787780e-01
    50003500    1.43194783015365e-01
    50004000    1.51295107779582e-01

prod.colvars.meta.hills.traj

    50000500     1.91679041070942e-01    5.00000000000000e-03   7.35675306144874e-03
    50001000     2.66810160619816e-01    5.00000000000000e-03   2.00768290491940e-02
    50001500     2.24911695120361e-01    5.00000000000000e-03   2.00767439249154e-02
    50002000     2.01805702861412e-01    5.00000000000000e-03   2.00767439249154e-02
    50002500     9.42138976351065e-02    5.00000000000000e-03   8.47534521138502e-03
    50003000     2.35211271787780e-01    5.00000000000000e-03   2.00767439249154e-02
    50003500     1.43194783015365e-01    5.00000000000000e-03   7.35675306144874e-03
    50004000     1.51295107779582e-01    5.00000000000000e-03   7.35546416475063e-03
    50004500     1.98157121994592e-01    5.00000000000000e-03   7.35231863167059e-03
    50005000     1.51237492403210e-01    5.00000000000000e-03   7.35231863167059e-03

This is what I see in the log file

colvars: The final output state file will be "prod.colvars.state".
colvars: ----------------------------------------------------------------------
colvars: Loading state from unformatted memory.
colvars:   Restarting collective variable "dist" from value: 0.212372
colvars:     successfully read the biasing potential and its gradients from grids.
colvars:     successfully read 0 explicit hills from state.
colvars:   Restarted metadynamics bias "meta" with step number 50000000.
colvars: ----------------------------------------------------------------------

Thank you

[giacomofiorin]

Thank you for checking. It looks like the information you are looking for should be in #prod.colvars.traj.1#

We should say in the manual that Colvars does not support the combination of -deffnm and -append. The way that mdrun manages the writing of the output files (i.e. deleting what was written since the last checkpoint in case of crash) is very difficult to extend to code like Colvars that is called differently. Therefore, we opted for the simpler approach of keeping the existing data by backing up the file being overwritten.

In your case, can you use -deffnm and -noappend?

[sathishdasari]

Hi @giacomofiorin,
I tried -deffnm and -noappend, it writes new files. I think later we have to concatenate all the files to get one single file.

Thank you