Re: [ESPResSo-users] lbfluid+virtual particles issue

[Andrew Jewett]

On Sun, Nov 29, 2015 at 6:11 PM, Aleksei Kabedev
<address@hidden> wrote:
> "background_errors 0 {999 force on particle was NAN.} {999 force on particle
> was NAN.} {999 force on particle was NAN.} {999 force on particle was NAN.}
> {999 force on particle was NAN.} {999 force on particle was NAN.} "

This looks like a generic numerical explosion, and my post to the list
is a generic response.  (Forgive that.  If you tried the suggestions
below, and they did not help, then post again.)

This kind of error could occur for many reasons.  But the most common
reason is that, at some moment, two atoms are too close together.
(You say it only happens when real and virtual atoms are too far
apart.  See below.)

Generally speaking, to solve this kind of problem, use some
combination of these strategies:

-visualize the system.
  Be sure to check periodic boundary conditions (see below).

-Try running the simulation with a much smaller timestep.

-Simplify the system by removing molecules or molecule types.
Keep simplifying until the error goes away.  That will help you locate
the problem.  (It helps to do this using a small system.)

 (For example, get rid of your "rasberrry" molecules, and see if the
error goes away.  If not, then try building a smaller system with
fewer LJ particles.  if you have any torsion/dihedral interactions,
get rid of them.  If you have any charges, set them to zero, etc...)

> during the integration cycle.

How long does the simulation run before you get this error?
(Does this happen during the first verlet iteration.
Or does the simulation run for a short while and then crash?)

> It happens if the distance between the central
> particle and the virtual ones is greater than ~2,

Not sure what to make of that.

But be sure to check periodic boundary conditions to rule out the
possibility that you don't have two particles too close together.

(In other words, if the boundary box is too small, it possible that
the atoms in different unit-cells are overlapping.  If you only look
at one copy of the unit cell, you might not see this problem.  If you
use VMD, you can test this by selecting the
"Graphics"->"Representations" menu option, clicking on the "Periodic"
tab, and click on the +x, -x, +y, -y, +z, -z boxes.  I am not familiar
with other visualization programs.)

> whereas the parameters of
> the flow look as follows
> lbfluid density 1.0 viscosity 0.5 agrid 1.0 tau 0.02 friction 10.0 ext_force
> 0.001 0.0 0.0;
> thermostat lb=1.0
> timestep=0.01
> Probably some of you have faced this problem before.
> Any help in these matters would be greatly appreciated.
> Thanks!

Don't know if that helped.
Good luck.

> Kind regards,
> Aleksei Kabedev