GROMACS version: 2023.2
GROMACS modification: Compiled with plumed 2.9.0. single-precision
Hi,
I am running a simulation of a protein in membrane assembled by CHARMM-GUI. In equilibration phase I used v-rescale (thermostat) and c-rescale (barostat). When I started to run the production, I switched to Nose-Hoover and Parrinello-Rahman.
It worked perfectly fine for my protein in many simulations, but once I increased the box a bit, I got this note every some steps (big gaps between the steps):
“Step 2842101 Pressure scaling more than 1%. This may mean your system is not yet equilibrated. Use of Parrinello-Rahman pressure coupling during equilibration can lead to simulation instability, and is discouraged.”
I did try to extend equilibration but it didnt resolve the issue.
I have searched for answers, but I didnt really get clear answers (although there are probably no clear answers).
Can this note be ignored? I have run production for 150 nanoseconds like that, and this note kept on appearing throughout all of it. How do I know if I can ignore it? Simply if the system isnt collapsing?
here is a plot of the pressure (picoseconds and bar) for the last steps of equilibration:
Is changing Parrinello-Rahman barostat to c-rescale, under semi-isotropic pcouple type, and in MD production phase, is acceptable? I know it is for isotropic but I am not sure about semi.
If I want to keep using Parrinello-Rahman, the option to increase tau-p is valid? I read in the forums here that the recommended values are tau-t=1 and tau-p=5 for atomistic simulations (and in charmm-gui workflow). Does changing tau-p can invalidate the simulation? How do I know the results are good? Same as before, checking that the system doesnt collapse and that the values are not off the charts, is enough to trust the results?
In general, with Parrinello-Rahman, I would recommend tau-p of 5 ps, but would consider increasing it to 10 ps if the system is unstable. C-rescale is not a sensitive to the tau-p, but 5-10 ps is still usually a good choice. There’s no (specific, i.e., with all the usual caveats) reason to believe that results would not be trustworthy when switching barostats. When it comes to switching during the simulation, it all depends on how you would like to describe your simulations, presumably in a publication. It would not invalidate a simulation, but it might definitely make it more difficult to explain your methods and it might also mean that you would need some extra care during analysis. If I discovered that my production simulation was not quite stable I would consider it another step of equilibration; try to stop it at a reasonable number of steps (to avoid having to say that you did, e.g., 42.563 ns of equilibration); modify the parameters I think would help improve the stability; start a “new” production simulation from the end of the one that kept crashing and/or warning about instabilities.
I didnt (and wont) change the barostat during production. I am just not sure about using c-rescale as a production barostat. I have read about c-rescale and that it is “allowed” to be used in production, but the recommended barostat is always Parrinello-Rahman (which I always used until now). Is there any considerations to take into account between using PR and c-rescale (except the pcoupling type)?
I have tried 25 nanoseconds of production using c-rescale (instead of PR) and the note is gone for the duration without changing tau-p.
I will do one more try with PR, but start with weaker coupling only in the equilibration (tau-p=10 as you suggested) and then change it back (at the end of equilibration).
Nowadays we recommend c-rescale for equilibration and production in GROMACS (except if you’re doing anisotropic pressure coupling). There’s nothing more to keep in mind, I’d say. It’s not as sensitive to tau-p, but a value of 5-10 ps is still usually good.
I will then probably use c-rescale in production. Although I am still not sure how I already used the same system, protein and parameters, and suddenly this pressure note appeared. It happened just when I upgraded to version 2023.2
I have 1 more issue to ask about: I understand c-rescale is valid as production barostat, but why cant I find research papers about protein dynamics (either in water or in a membrane bilayer) that use this barostat in the production? Is it because c-rescale is fairly new? or that Parrinello-Rahman is much more preferred for some reason when dealing with proteins? Or I just didnt search it well enough?
