I hope that my question is not too obvious but I can’t find a satisfying answer. I’m studying a very simple box of 8000 Argon atoms in a Lennard-Jones potential. I want to study different samples (let’s say 10) to then average over them (I guess that’s what we could call replicates ?). I was thinking of using one equilibration of the system (NVT) and then run it for some time, and using this as my second equilibrated system. What I want to do after that is to run NVE simulations without velocity generation.
My question is: how long should I run the equilibration again for them to be decorrelated in time ? I was thinking 50ps, but I would like something optimal.
I’m not even sure if that’s the way to go, and maybe it’s better to generate new velocities and equilibrate from the start again ? I’m just trying to find what’s gonna be the most efficient.
I guess one way would be to output velocities (in this case probably with short output interval) and use gmx velacc - GROMACS 2024.4 documentation to calculate the velocity autocorrelation function, and then use twice that time (i.e., the time to reach a low normalized value - not necessarily 0) to ensure uncorrelated velocities.
I’m not sure if only velocity autocorrelation would matter.
Thank you. I obtain something like that. It reaches a low normalized value around 1ps after the start. Does it mean that 2ps is enough then ? My timestep is 2fs, so this would be only a thousand steps. This seems low, but I guess if the autocorrelation function is already around 0, there should be nothing to worry about ?
If you think uncorrelated velocities are enough, then it should be fine. I guess it would be a good enough requirement in such a simple system, but I haven’t really considered it much yet, so I’m not completely sure.
Since the time-span is so short anyhow, I don’t see any reason not to run at least 5000 or 10,000 steps to be on the safe side. These simulations would be very quick anyhow.
Yes, I think uncorrelated velocities should be enough because I’m working on velocity calculations. I shared the autocorrelation for a 1000 particles and not 8000. I will compute it again and check. But I will do as you advise and run a bigger amount of steps. I’m not so sure yet how many different equilibrations I will have to perform, and I thought that if it is a bigger amount, it would be better to do a minimum amount of steps.
