[gmx-users] How to suppress the error "X particles communicated to PME node Y are more than a cell length out of the domain decomposition cell of their charge group
Mark Abraham
Mark.Abraham at anu.edu.au
Mon Dec 27 00:17:33 CET 2010
On 27/12/2010 9:22 AM, WU Yanbin wrote:
> Hi, Mark,
>
> Thanks for the reply.
>
> As suggested, I tried a "water in vacuum" case. Initially the water
> droplet is a cubic 4nm-by-4nm-by-4nm water box, located in the middle
> of the simulation box. Everywhere else is just vacuum. The simulation
> box size is 8nm by 8nm by 8nm. SPC/E model is used to describe
> interaction between water molecules.
>
> Such system is first equilibrated with "steep" option.
>
> The "mdrun" simulation goes with no problem with parallel running on
> 32 cpus, even though there are occasionaly one or two water molecules
> move very fast (Mean square displacement being 200 times fast than
> bulk water) in the vacuum.
OK, that sounds like behaviour that might be expected... some molecules
evaporate. Check the trajectory to be confident this is the situation.
>
> When I switch to 64 cpus, the error "X particles communicated to PME
> node Y are more than a cell length out of the domain decomposition
> cell of their charge group" appears.
The most plausible reason for this is that the above evaporated
molecules are moving so fast that they're doing what the error message
says - travelling more than width of a box in one integration step. The
DD implementation is predicated on that being impossible.
You might succeed by arranging for the largest possible internal DD cell
diameter, i.e. a 4x4x4 DD. mdrun -npme 0 might choose this, otherwise
use the hidden options to mdrun (use mdrun -hidden) to use that DD with
-npme 0.
Otherwise, you'll need to accept the fact that there are "engineering"
constraints on efficient parallelization algorithms, and not every
situation can be catered for. For example, a simulation of a bulk LJ
fluid would fail if you used so many processors that the cell diameter
was too small with respect to the distribution of particle speeds.
Mark
>
> Below is the parameter file I'm using.
>
> ------------------------------------
> integrator = md
> tinit = 0
> dt = 0.002
> nsteps = 5000000
>
> comm-mode = Linear
> nstcomm = 1
> comm-grps =
>
> energygrps =
>
> nstlist = 1
> ns_type = grid
> pbc = xyz
> rlist = 1.5
>
> coulombtype = PME
> rcoulomb-switch = 0
> rcoulomb = 1.5
> epsilon-r = 1
>
> vdw-type = Cut-off ;Switch
> rvdw-switch = 1.0
> rvdw = 1.5
>
> fourierspacing = 0.12
> fourier_nx = 0
> fourier_ny = 0
> fourier_nz = 0
> pme_order = 4
> ewald_rtol = 1e-05
> ewald_geometry = 3d
> epsilon_surface = 0
> optimize_fft = no
>
> Tcoupl = nose-hoover
> tc-grps = System
> tau_t = 1.0
> ref_t = 300
>
> Pcoupl = no ;Parrinello-Rahman ;no ;berendsen
> Pcoupltype = isotropic
> tau_p = 1.0
> compressibility = 4.5e-5
> ref_p = 1.0
>
> gen_vel = no
> gen_temp = 300
> gen_seed = 2008
>
> constraints = none
> constraint-algorithm = Lincs
> Shake-SOR = no
> shake-tol = 1e-04
> lincs-order = 4
> lincs-warnangle = 30
>
> energygrp_excl =
> ------------------------------------
>
> Is there any parameters configuration wrong with the simulation? Or is
> there any way to go around this error?
>
> Any tip is appreciated.
> Thank you.
>
> Best,
> Yanbin
>
>
>
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Thu, 02 Dec 2010 17:24:18 +1100
> From: Mark Abraham <Mark.Abraham at anu.edu.au
> <mailto:Mark.Abraham at anu.edu.au>>
> Subject: Re: [gmx-users] How to suppress the error "X particles
> communicated to PME node Y are more than a cell length
> out of the
> domain decomposition cell of their charge group"
> To: Discussion list for GROMACS users <gmx-users at gromacs.org
> <mailto:gmx-users at gromacs.org>>
> Message-ID: <4CF73B92.40003 at anu.edu.au
> <mailto:4CF73B92.40003 at anu.edu.au>>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>
> On 2/12/2010 4:16 PM, WU Yanbin wrote:
> > Dear GMXers,
> >
> > I'm running a simulation of water contact angle measurement on
> top of
> > graphite surface.
> > Initially a water cubic box is placed on two-layer graphite surface
> > with the rest of the box being vacuum. The water droplet is relaxed
> > during the simulation to develop a spherical shape.
> >
> > An error of "X particles communicated to PME node Y are more than a
> > cell length out of the domain decomposition cell of their charge
> > group" was encountered.
> > And I have read the suggested solutions at the link below
> >
> http://www.gromacs.org/Documentation/Errors#X_particles_communicated_to_PME_node_Y_are_more_than_a_cell_length_out_of_the_domain_decomposition_cell_of_their_charge_group.
> >
> > I guess the reason for this error in my case is because of the
> vacuum
> > such that the water molecules at the boundary of the droplet can
> move
> > fast. I have check the trajectory and the simulation is OK.
>
> I doubt the simulation is OK. This error message is one of several
> that
> can happen when the system is not well-enough conditioned for the
> MD to
> be stable. See
> www.gromacs.org/Documentation/Terminology/Blowing_Up
> <http://www.gromacs.org/Documentation/Terminology/Blowing_Up>.
> Here, you have atoms moving much faster than GROMACS was engineered to
> expect.
>
> You should be confident that a water drop in a vacuum, and your
> graphite
> surface are both stable on their own before you try the wetting
> simulation.
>
> >
> > For this situation, is there a way of suppressing this error? Or
> what
> > else can I do?
>
> Work out why it's poorly conditioned.
>
> Mark
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20101227/5c70f184/attachment.html>
More information about the gromacs.org_gmx-users
mailing list