Collective Brownian Motion Particle Simulation with stochastic Langevin-Equation

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,
I would like to simulate collective brownian motion using CUDA. Iam a beginner and just found the particle code in the sdk.

  1. Does a more detailed comment/tutorial then the particle.pdf exist, that makes it easier to understand the code?

  2. Does anybody know if a sample code for the simulation of langevin-equations with a chemotactic field already exists?

2

Hello,

I am interested also in Langevin dynamics but without chemotactic field. In this case the implementation depends on the type of interactions between particles. There are different implementation possible for long range (N or almost N interactions per particle) or for short-range interactions. How do yo add the chemotactic field. For beginning you can try to divide the work, do the chmotactic part on gpu and the rest on cpu and see how much speed up.

How do you add the chemotactic term?

3

Hi,

I dont know yet how to implement the chemotactic term. Iam starting with the Langevin Equations as I suppose this is easier to implement. My interactions will be short-ranged. What kind of Langevin dynamics are you simulating? Are you using the sorting algorithm proposed or atomics?

4

Hello,

I am planning to implement the BBK algorithm for a Lennard-Jones potential following this paper: JOURNAL OF CHEMICAL PHYSICS,VOLUME 114, NUMBER 5,1 FEBRUARY 2001,Langevin stabilization of molecular dynamics, Jesus A. Izaguirrea, Daniel P. Catarello, Justin M. Wozniak, and Robert D. Skeel). The most intensive part are anyway calculating the forces. I think , in the case with few interactions per particle, the most effective way to do this is with a neighbour list which contains the particle positions , but at his moment I am stucked at this point.I did not get yet the chance to look over the example from the SDK. I am planning to check seriously the existing work in 1-2 weeks.

5

i guess u should start from zero for easy maintenance in the future,

i’m also working for particle method but for continuum approximation, so i guess this paper is the one u’r looking for http://dx.doi.org/10…rco.2008.12.005 for the purpose of potential sum. just modify for your molecular dynamics purpose

in the case of nbody problem in force, just look for the nbody in cuda sdk, it calculate gravitational nbody problem. if u need furthermore speed, just look for either barnet-hut/tree-code or FMM. i think some are available as opensouce like exaFMM, or in specific case of molecular dynamics, fast Ewald method is more popular.

6

Thanks for the suggestion I will check it out.