According to the gromacs manual, setting 'gen_vel = yes' for production generates initial velocities for particles, and setting it to 'no' doesn't. I set it to 'no' and my simulations still work as expected. Could it be that the initial velocities of particles were instigated just by forces, fluctuations in pressure/temp etc.? Or does 'gen_vel = no' generate initial velocities but without following the Maxwell distribution?
I also used 'gen_seed' along with 'gen_vel = no' to produce multiple replicas of the same system, and they all turned out fine, their dynamics vary significantly from one another, as expected. This doesn't seem to connect with what's mentioned in the manual. What am I missing? Any help is much appreciated.
If gen-vel is 'no,' velocities are read from the structure file. If gen-vel is 'yes,' velocities in the structure file are ignored and generated anew.
Production MD trajectories will be different with gen-vel set to 'no' and 'yes' because the initial velocities will be different.
Setting gen-vel to 'yes' for the production MD simulation will require a short re-equilibration of the system, and that initial part of the production trajectory should be discarded.
https://manual.gromacs.org/archive/4.6.5/online/mdp_opt.html#vel
Thanks a lot Martin Klvana !
On the other hand, If gen_vel=no and production velocities are read from the previous equilibration, wouldn't all my prod simulation replicas turn out exactly the same? In my simulations, all replicas gave different results as expected. How would you explain that? As far as I know, gen_seed is not considered if gen_vel=no. Please correct me if I'm wrong.
Repeating a simulation with the same initial velocities should produce identical trajectories. Unless there is another element of randomness (besides generating initial velocities) introduced into the simulation protocol.
In that case, there's clearly another source of randomness and I don't know what it is. I was wondering if it could be the 'slight' fluctuations in temp/pressure that's affecting the velocities in each simulation. Or do you reckon that these fluctuations would also be identical each time?
There are many sources of potential deviation that lead to non-deterministic trajectories. See https://manual.gromacs.org/current/user-guide/managing-simulations.html#reproducibility for examples.
What I understand is MD simulations are inherently chaotic in nature, there are too many underlying variables (you have already pointed a few) that can alter the trajectories of the replicates. You might get identical/similar trajectories if you repeat the simulation 10/20 times. Besides, the whole purpose of simulation replicates is to sample as many conformations as possible and get reliable results.
Read Justin Lemkul answer on this subject: https://www.researchgate.net/post/Why-does-performing-molecular-dynamics-simulation-on-different-CPUs-lead-to-different-results#:~:text=MD%20simulations%20are%20inherently%20chaotic,replicates%20are%20almost%20always%20required.
Check this paper out where the authors have done 100 replicas: Avoiding False Positive Conclusions in Molecular Simulation: The Importance of Replicas Article Avoiding False Positive Conclusions in Molecular Simulation:...
Thanks Rajat Kumar Jha for pointing out Justin's answer (I hadn't come across that before). It was insightful. Also, the JCTC paper is a good reference, it answers precisely my question and elaborates. Thanks for sharing.
- Badges
- Science topic