Hello everybody,
I have decided to switch from si to nano units because I am working with nanoparticles. From the beginning I had the feeling that the simulations are way slower in nano units than in si units. Now I have made the test and ran equal simulations.
I am sorry that I cannot give a complete example here, because I had to make some modifications in the source code. Here a short case scenario:
3000 Particles in a very porous structure (approx 98 %), 1000 steps.
boundary p p s
Walls above and below the particles, upper wall is moving down (compressing the particles)
I found that the simulation are 8 times slower in nano units. The reason seems to be in the Neigh time:
Is 6.97182 (86.82 %) nano vs 0.137387 (13.76 %) si
I know the absolute time doesn't help too much but it shows the difference.
Neighs: 4087 ave 4087 max 4087 min is the same for both (seems to be quite high)
I am using:
units nano
neighbor 3.0e+00 bin (3e-9 for si, respectively)
neigh_modify delay 0
while 3 nm is the diameter of my smallest particle (polydispersed, 3-23 nm).
But this is what I have found: As soon as the top wall is moving, the neigh time is shooting up. When the walls hold still it is pretty much the same for both unit systems. This is independent from the direction of the motion, even if I move it away from the particles.
I am really sorry that I cannot give an example, there is a restriction in the source code (multi_node_mesh_parallel_I.h:555) that doesn't allow wall dimensions below 1e-4, and it is suggested to switch the unit system. Therefore it is difficult to give an easy example here.
But did anyone ever experience a similar issue?
Regards,
VBaric