Re: [ESPResSo-users] Fence-like density distribution MD+LB

[Owen Hickey]

Hey Aleksei,

The behaviour you describe is the result of the coupling scheme with
the LB not producing the exact same friction for all positions within
a plaquette, see Fig. 5 on page 55 of
http://arxiv.org/pdf/0803.2826.pdf. There are three ways  that I know
of to reduce this effect:
1) switch the ratio of the bare friction parameter in the LB to
viscosity*density*agrid (this is what you've done by increasing agrid)
2) switch to the three point coupling scheme (although be wary, in the
current release it is buggy so you would need to download the new
master from Github)
3) switch the type of noise used in the LB, the default is currently a
flat distribution which produces a stable LB but seems to make this
problem worst from my experience.

One last thing, about which I am not entirely certain, is changing the
temperature of the LB itself may alter the magnitude of this effect.
As for moving the specific position of the LB boundary I am not sure I
have any advice except to keep in mind that on the GPU the boundary
will always be more or less at whole integer multiples of agrid
(unless this has changed since the last time I looked at the code).

I hope that helps!

Owen

On Tue, Mar 10, 2015 at 3:24 PM, Aleksei Kabedev
<address@hidden> wrote:
> Dear all,
>
> I observe some strange behaviour of MD particles, dissolved in the fluid,
> modelled by LB.
> 1)What I do is
> - create a box with periodicity 1 1 0 with boundaries for both md and lb
> - put some particles into it
> - turn on the lb thermostat
> - write down the positions of the particles (that don't interact to each
> other directly, just hydrodynamically) along the z-axis
> - make a histogram of these positions, which is basically a density
> distribution function.
> 2) What I expect to see is
> - equal distribution in the middle of the channel, drastic declines by the
> sides
> - flatness of surface in the middle is getting higher upon the growth of
> statistics amount.
>
> Instead of that I see fence-like curves with "top edges" exactly at the
> positions of cell edges (or exactly in the middle between them). That works
> even if I don't use any external flow. After I changed the agrid parameter
> from 1.0 to 2.0 it seemed to disappear. If I do exactly the same simulation
> with langevin thermostat, I see a flat surface without any "hills" on that.
> I am just wondering if someone knows why it happens? Thank you in advance!
> See the plot examples and a simple code attached.
>
> Actually, there is another question I should ask - what is going to happen
> if I simulate the particles inside the channel and the boundaries for fluid
> and particles are not exactly at the same place? Say, lbboundary is
> 0.5*agrid further from the particles, than md-constraint or vice versa?
> Thank you!
>
> --
>
> Kind regards,
> Aleksei Kabedev