[gmx-users] un-coupled simulations and traj recovery
gk237 at cam.ac.uk
Tue May 13 17:26:01 CEST 2003
>/ thank you for this ref. The simulation i want to run i pretty long (40
/>/ ns) , it is actually a simulation has been done in the past by the
/>/ Gromacs group, the folding of a beta-hairpin, from the extended
/>/ configuration at 360 K. The reasons that i dont want to use coupling are
/>/ technical and not physical. Due to a special kind of statistical
/>/ analysis we want to apply, we need to be able to go back to the
/>/ simulation and reproduce several frames of critical moments. I
/>/ understood (and experienced) that this is not possible when coupling
/>/ (either T or P) is on, of course when i use coulomb. When i run an
/>/ equilibration run and then switch the coupling off,explosion takes
/>/ place. Then i used PME, i can now reproduce the frames (i.e. get same
/>/ coordinates and velocities) but i get the energy drift and the temp is
/>/ dropping down...
/>/ I was not familiar with the shifted potentials, but if the coupling
/>/ stays off, then i think they might be a solution to our problems
! Be aware that using shifted potentials introduces artefacts in your
simulation. (I think there was a ref about that in my JCC paper?)
I would defenetely *not* reccommend that for any type of 'serious'
simulation. The only reason I used it, was to get a good estimate
of the energy drift due to the integration errors (related to time
step size), and hence had to get rid of all other energy drift sources.
I don't expect that you could get energy conservation so well that
you can run an un-coupled simulation for nanoseconds. You could
invest some time to hack a version of mdrun that writes additional
velocity frames, and the scaling factors (or whatever is exactly
needed), to get the exact reproduction of frames you need.
Couldn't you use the other thermo-/barostats? From what I understand,
the influence Nose/Hoover and Parinello/Rahman have on the statistics
is well-known and could therefore possibly be corrected for(?).
i have tried in the past to dedicate some time hacking the code but to be honest i was kind of lost... so i gave up...
(actually, do you have any reference for a source code navigator? :-) ....)
So, i am using un-coupled simulations and try to minimise the energy drift (... not very succesfully i have to say!)
It is very important for us to reproduce the trajectries... This is why we are using PME.
Do you think is there another way of doing it?
More information about the gromacs.org_gmx-users