|
From: | Vincent Ustach |
Subject: | Re: [ESPResSo-users] Setting up Lattice Boltzmann |
Date: | Thu, 9 May 2013 09:31:38 -0700 |
Hi,
That means that you also plan to throw out the DPD particles? OK. That sounds fair. Have you tried the GPU version of LB? It is incredibly fast!
On May 7, 2013, at 5:43 AM, Vincent Ustach <address@hidden> wrote:
> Stefan,
>
> Thanks for your reply! I have the code running but of course I still have a lot of work to do. I did not want the discussion to get stale, however, so I am replying now with the info that I have.
>
> I am abandoning the DPD thermostat and using LB for a thermostat and for the HIs. I was not using inter_dpd. Sorry that I was not clear.
That makes sense. However it is probably better to use the bounce back method for walls. One thing that we always wanted to do, is making it possible to flag single nodes as boundary nodes, but this is not implemented at the moment. Though quite easy, both for GPU and CPU. However I'm rather busy with my thesis right now, and thus would prefer to do it later :-). You can do it quite easily.
> I will look into the pore constraint, but I don't want to restrict myself to cylindrical and conical pores.
What is your mass and your Diffusion coefficient? Most likely you'll find out that if you put in the numbers, both your viscosity and your friction will be incredibly high.
> This is indeed a concentration gradient driven diffusion, however electrostatics will be involved later. Viscosity seems to be (0.8 Pa * s) *10^-3. What are the units for friction?
You can try. It depends on your system, if it is. This I can tell you for sure. And on you definition of "adequate" :-).
> I am working on printing the fluid velocity field to an output file and then hopefully it will shed some light on if the wall of particles serves as an adequate LB boundary.
Please don't think that you can get LB for free (not computationally, but in terms of carefully thinking about your system). Just as in DPD you have to worry seriously about getting parameters right. I have spent quite some time learning about that and I'm quite sure I can get helpful advice now. But you need to know: Will you have "macroscopic" fluid motion in your system, or is just correlations that matter? Probably in both cases it is necessary to fit the "effective particle radius" 1/(6 pi eta D) to the other lengths scales in you system. But in the first case it is OK if the radius is small, while in the second case it is important to have a high resolution to get the near field right.
Don't worry about the discussion getting stale, we can revive it later :-). My desk is also quite full.
Cheers and good luck
Stefan
[Prev in Thread] | Current Thread | [Next in Thread] |