LIGGGHTS hertz mindlin model slower than the LAMMPS one?

LAMMPS LAMMPS Mailing List Mirror
1

Hi everybody,

I've performed a simulation of a vibrated granular system with LAMMPS and now i've reproduced it on LIGGGHTS almost exactly.

The granular media is poured into a container and then the container starts to vibrate.

Just the way to construct the container is different because i use the wall/gran command of liggghts with an imported mesh instead of the wall/gran/region command of lammps with a region defined in the script.

I've seen also that in liggghts the pair-style commands is slightly different and in particular the hertz-mindlin model has an additional square root of the overlap (the delta in the doc formulas) in the dumping coefficients.

I've done some tests and i've seen that the simulation with liggghts is the 50% more slower than the lammps one.

Simulations are runned on the same machine with 4 cores using mpirun command.

Does anyone knows general suggestion to follow in order to switch from lammps to liggghts?

Is there a general reason why liggghts could be slower?

Thanks a lot in advance

Andrea

2

The speed difference could be the cost of the sqrt term.
The LAMMPS output will tell you how many pair interactions
are computed and the cost of the pair portion of the computation.
If LIGGGHTS tells you the same thing and the # of pair
interactions (neighbors) are comparable and the pair cost
is higher, that could be the reason.

Re: switching between LAMMPS and LIGGGHTS, there
is no rule of thumb. If the model you want to run
can be run in either code (LIGGGHTS has many
add-ons for granular systems that LAMMPS does not
have), then just try your model in both.

Steve