LB GPU uses uniform random numbers instead of normal distributed

[KaiSzuttor]

No description provided.

[RudolfWeeber]

I thought that could be switched at compile time? The CPU and GPU ones using different distributions might also be a cause of the different width of the temperature fluctuations

[KaiSzuttor]

Yeah but... check lb.cpp line 101

[KaiSzuttor]

Btw it's also not tested that LB thermostat gives the right maxwell distribution of velocities.

[RudolfWeeber]

I did a quick check. With Gauss random cut, the standard deviations in cpu and gpu are similar.
I'll open a pr fixing lbcpu with other noise types

[KaiSzuttor]

Why is that actually switchable by the user?

[mkuron]

As far as I know, the actual random number distribution does not matter as long as their mean and variance is correct. There might be a publication by Alexander Wagner on this topic, but I couldn't find it. The switch may have been added for testing purposes at some point.

[jonaslandsgesell]

Please cite this paper in a comment in the corresponding parts of the code. This question will pop up in each developer generation.

[KaiSzuttor]

Dünweg, Burkhard, and Wolfgang Paul. "Brownian dynamics simulations without Gaussian random numbers." International Journal of Modern Physics C 2.03 (1991): 817-827.

[jonaslandsgesell]

Please cite in source as documentation!

[KaiSzuttor]

Ladd, Anthony JC. "A fast random number generator for stochastic simulations." Computer Physics Communications 180.11 (2009): 2140-2142.