[lammps-users] SKIN settings affects the results(without dangerous build)?

For the first problem, I did some careful tests, and find that it is the skin setting that really matters. With metal units ,AIREBO potential, carbon system, 1fs stepsize, initial velocity generate with random number, calculate on one processor of the same machine, equilibriate 1000steps, . the 1000step results are as follows( all without dangerous build )( 6DEC2008 lammps edition ):

Skin Step Atoms Temp PotEng KinEng TotEng

2.0 1000 9600 232.8161 -39095.203 190.16299 -38905.04

2.1 1000 9600 233.52098 -39092.148 190.73874 -38901.409

3.0 1000 9600 231.9985 -39074.087 189.49518 -38884.592

5.0 1000 9600 231.9985 -39074.087 189.49518 -38884.592

10.0 1000 9600 231.9985 -39074.087 189.49518 -38884.592

It seems that with the skin value increasing, the potential energy and total energy decrease, and when the skin is bigger than some value between 2.1 and 3.0. the results don’t change anymore. what it implies seems that when the skin is small, although there is no dangerous build, the result is not reliable, can you tell me why . I’m really puzzled, thank you!

I assume you mean the neighbor skin setting, not the cutoff
in the pair airebo style itself. These differences seem pretty minor.
It's possible after 1000 steps that they are just round-off numerics
with diverging trajectories. If you change the neighbor skin,
you change the binning for neighboring, which means neighbors
will be stored in different orders and forces computed in different
orders, which will lead to round-off issues. If you plot the PotEng
as a function of time from step 0 to 1000, does it slowly diverge?