[lammps-users] Timestep Selection

Dear Lammps Users,

I am simulating friction between metals (eam potential) in metal units at 300K temperature. My results are very much dependent on the timestep size. What would be the time-step selection criteria? Is there a reason for keeping the default timestep for metal units to 0.001ps in Lammps?

Thanks,

Bhavin Vadgama
Graduate Student
Auburn University

Dear Lammps Users,

I am simulating friction between metals (eam potential) in metal units at 300K temperature. My results are very much dependent on the timestep size. What would be the time-step selection criteria? Is there a reason for keeping the default timestep for metal units to 0.001ps in Lammps?

which results in what respect are depending how much on what choices
of time step?

the time step selection criterion would be a small error in the integration
of the equation of motion, typically indicated by a good conservation of
the total energy in the NVE ensemble for an already well equilibrated
configuration. if the time step is too small, you not only waste (CPU) time,
but also accumulate more numerical noise. so there is some optimal
range, which may be very problem dependent.

axel.

A 1 fmsec timestep is reasonable for many potentials, e.g with EAM.
If you want to change it, use the timestep command. But as Axel
said, your results should be relatively insensitive to timestep size
unless you make it too big. I presume you realize you need to run
2x as many timesteps if you half the timestep size, to see the same
results.

Steve

Bhavin - one good criterion for most potentials is that the timestep should not be more than about 1% of the period of the fastest vibrational mode. One thing to look at (in simulation of an underdamped system, e. g. a ‘quiescent’ crystal) is the first zero-crossing time of the velocity-autocorrelation function - that’s often a decent measure for fast vibrations. Of course, for a ‘violent’ process like two metals rubbing one another (i. e. if your friction is large enough to ‘damage’ the surfaces), you may need a smaller timestep.

Best,
Rob

Agreed - except I would say 5-10%. I think a C-H bond is 10 fmsec,
and people routinely use a 1 fmsec timestep for those systems.

Steve

Thank you Rob/Steve for all the suggestions and help.

Bhavin N. Vadgama
Auburn University