[gmx-users] Crashing NPT run, 1.5 ns

[Mark Abraham]

JMandumpal wrote:
> Dear list,
> 
> When I changed the time step to 1fs, the simulations were successful. 
> What could be wrong with timestep 2.0 fs. I am using GROMAXS version 3.3.3

It's a numerical simulation. The outcomes of these are often sensitive 
to the initial conditions and the fine detail of the algorithm. In this 
case, probably your initial structure is insufficiently well minimized 
and/or equilibrated to cope with a "long" timestep with the large forces 
that result from a non-equilibrium structure. A "short" timestep allows 
the atoms to move a shorter distance in response to a large force, to 
reach a position where they should experience less force. Thus over just 
one integration time step, the 1fs integrator will be closer to the true 
ensemble than a 2fs integrator will be after one timestep. This is even 
more true of both integrators after the same length of time, e.g. 2 and 
1 steps respectively.

Mark