[gmx-users] Simulation timestep: how to assess?

K.A. Feenstra Feenstra at chem.vu.nl
Mon Jul 22 17:30:38 CEST 2002

Lieven Buts wrote:
>   I have run simulations of a system (one protein molecule and
> 8 Na+ ions in SPC water) using timesteps of 2 and 4 fs. In both
> cases the temperature, density and total energy, as well as the
> box volume and the RMSD of the protein with respect to the starting
> structure are stable. The temperature, density, volume and RMSD
> oscillate in the same ranges for both timesteps. The total energy
> fluctuates between -5.75e5 and -5.77e5 kJ/mol for the 2 fs timestep,
> and between -5.70e5 and -5.73e5 kJ/mol for the 4 fs timestep. The RMSD
> also seems to be a little bit higher for the longer timestep (0.13 on
> average instead of 0.10 nm). Is this sufficient to draw conclusions
> about the acceptability of a 4 fs timestep?
>   More generally, what should one look for in determining the optimal
> timestep for a simulation? Do you just have to make sure that the system
> does not explode, or are there more subtle artefacts to be on the
> lookout for?

Ah, this is one of my favorite subjects! Unfortunately it is not
at all trivial. First off, it depends on exactly what you are 
going to use your simulation for. Some things are more sensitive.
A good thing, therefore, is to take (one of) the key propertie(ys)
that you are going to look at and see if that has any timestep
dependency. Ofcourse, as you already did, for every timestep
you will at least want to have the pressure stable, reasonable
energy fluctuations and reasonable rms values. You should, however,
be aware that for example rms values for two simulations of the
same system but with e.g. different starting velocities, can 
be quite different, for example one might reacy a 'plateau'
value of 0.2 nm, bu the other might as well go up to 0.5 nm!
So, to be really thorough you should do several (~10) independent
determinations of the same parameter, and see if the average
values compare for different timesteps. That is also more or
less what I did in my 'Dummy Atoms' article:

K. Anton Feenstra, Berk Hess and Herman J. C. Berendsen. Improving
  Efficiency of Large Time-scale Molecular Dynamics Simulations of 
  Hydrogen-rich Systems. (1999) J. Comput. Chem. 20 (8), 786-798 

Sorry, but there just isn't any clearer answer; it just depends...
Anyway I hope I have given you enough pointers to get you on
your way. Please report back if you have made interesting and/or
significant progress; it's nice to also hear about successes in
addition to the problems ;-)



More information about the gromacs.org_gmx-users mailing list