Repeating an exact simulation with a higher sampling rate

GROMACS version: 23.2
GROMACS modification: No
Dear community,

I was repeating a simulation with an external pulsed electric field induction many thousands of times on explicitly water soluted protein, looking for an induced conformational transition. After much searching I found several convincing transitions. I would now want to analyse the speed and temperature distribution of these trajectories across the protein, however my velocity write-out frequency (nstvout) was not sufficient for the timescale of the transition. I would need to repeat the exact same simulation, but with a higher write-out frequency.
I have looked at the mdrun -rerun option, but it does not produce more frames, just calculates energies and forces at existing frames.
Could someone help me with setting up an exact repeat of a simulation please?

Best Regards,
Istvan Lorincz

Dear @Elvis,

I think the -rerun option is going through the trajectory to compute energies and so on, it is not really re-running the MD simulation. Also, I think it is pretty much impossible to run two really identical simulations, that is, in which each frame is identical, even if you start from the same reference. Floating point arithmetic, stochastic therms (c-rescale thermostat for example) and other algorithms will be sources of small differences along the trajectory which eventually will propagate and make the trajectories different. You can prove that trajectories actually diverge exponentially and they tend to become uncorrelated extremely quickly.

There is one option that might help you here, which is the reproducible mode. However, it still relies on forcing the simulation from the beginning to be reproducible, and it still depends on the hardware, as if you change hardware then the flag is going to be useless.

Dear @obZehn,

How can I use the reproducible mode to repeat an existing simulation? I’m running on an HPC cluster, getting the same machine is manageable. As far as I understand (and tbh. that’s not really far) reproducible mode would help if I used it with an existing simulation (and would want to reproduce it afterwards).
To make matters worse, I don’t have the random seed set, and it is not in the log file, only the ld-seed is in the log file.

Thank you for your help and comments,

Dear @Elvis,

Actually you should use gmx mdrun -reprod ... when you run the simulation so that GROMACS makes sure to keep the sources of “irreproducibility” to zero, so there is no way you can now with the one you have, as this should have been part of the original mdrun command that you used to run that simulation. You should however also request the same identical node. Regarding the seed, I think 99% of the info you need is contained in the .tpr file, you can access it with gmx dump, take a look at the manual.

On a side note, I never used the reprod flag and, as per manual, it really seems mostly related to debugging or to check specific properties of the integrator/software etc. I wouldn’t try to force any of this type of behavior. Are you sure you won’t get any transition with another normal run?

Dear @obZehn ,

Thank you for your answer. I get partial transitions for about every 1000 simulations and 1’ve got a very nice undisputable one out of 10000 runs. (a lot gpu hours burnt for it). So in theory, yes, I could hunt for another, in practice… I would have really liked to have used the trajectory I’ve already got.