I am running an equilibration for my system both with c-rescale and berendsen to compare and I noticed a difference in the magnitude of fluctuations of the box dimensions between the two (same system, same mdp options apart from the barostat).
Is this to be expected due to the stochastic component added in c-rescale? If yes, is there a rule of thumb to get the ‘most comparable’ fluctuations in c-rescale?
Hi,
the stochastic cell rescaling algorithm is a variant of the Berendsen algorithm that allows correct fluctuations to be sampled.
(see Molecular Dynamics — GROMACS 2021.2 documentation)
\Alessandra
Thank you for the reply. What you are suggesting is that these are the ‘correct’ fluctuations of the box size, or did I misinterpret?
The follow-up question that I have is how to ‘scale’ the tau-p from berendsen to c-rescale. I did have a look at the documentation but I did not seem to find any suggestions if one should use a different tau-p for berendsen and c-rescale. In the original paper I see they used 0.1 ps, while here I used 5 ps. The next step would be to run with 0.1 ps of course, I was just wondering if there is a general reccomended value.
Cheers,
Claudia
The Berendsen barostat, just like the Berendsen thermostat, dampens the fluctuations, because it only has a damping term. The Bussi barostat, and thermostat, add noise so you get the correct distribution of box volume. This is the concept of the fluctuation-dissipation theorem.
The value of tau-p can be chose rather freely. The only limitation the paper states is that it should be much larger than tau-t. The paper uses tau-t=0.1 ps, tau-p=0.5 ps.I would not think this is much larger. You value of 5 ps sounds reasonable to me. That would work well with a tau-t of 0.5 ps, which is a popular value.
I don’t know what you are referring to with “they used 0.1 ps”.
