[gmx-users] Effects of reducing timestep

[braffert at purdue.edu]

Hello,

We are studying the effects of applied forces during protein
unfolding.  In running simulations with varying timestep lengths, a
peculiar pattern emerges.  The timestep can be lowered from the
typical 2 fs down to 0.5 fs with only very minor changes in the
results, however, with a 0.25 fs timestep, a significant difference
can be seen in the resulting trajectory.  If the timestep is further
lowered to 0.125 fs, we observe a phenomenon that only occurs under
lower applied forces (the force has been held constant through these
timestep trials).

Is there any reason that a very small timestep value can have adverse
effects on the simulation results?  Could there be such a thing as a
minimum value that ensures accuracy?

Thanks,
Ben