[gmx-developers] Re: gmx-developers digest, Vol 1 #324 - 6 msgs

[Jason DeJoannis]

> Subject: Re: [gmx-developers] insertion-deletion
> From: David van der Spoel <spoel at xray.bmc.uu.se>
> To: gmx-developers at gromacs.org
> Date: Fri, 23 Apr 2004 08:47:04 +0200
> Reply-To: gmx-developers at gromacs.org
>
> On Thu, 2004-04-22 at 22:36, Jason DeJoannis wrote:
> > Dear All,
> >
> > Consider a lennard-jones fluid. I want to perform insertions
> > and deletions provided
> >
> >             1 < n < n0
> >
> > where n  is the number of atoms
> >       n0 is the number of atoms in the starting config
> >
> >
> > All I do is change top->atoms.nr and all the other measures of
> > the number of atoms (i.e. inside mdatoms,top->atoms,top->blocks,nsb).
> > Directly afterwards a new neighbor search is performed.
> >
> > Unfortunately after a dozen or so steps, one or more of the forces
> > becomes "nan". Right now I am running in serial mode with 10 atoms to
> > keep things simple.
> >
> > Any idea what is killing my forces?
> How about the lengths of the force arrays, coordinate arrays and son on?
>
That is the point of keeping the number of particles less than or equal
to the starting config. The arrays have a fixed length of n0. At first
I tried using srenew(array,newsize) to expand the arrays but it proved
difficult to find everything that needs expansion.

I am looking into the neighbor search right now. I'm thinking that
maybe the search is still including the deleted particles somehow.
But even so, it should not produce a force error, right? I think the
appropriate lists are in fr->nlist_sr[eNL_VDW].jindex[*]? Since it
is an LJ-fluid there are no charges, and I am not using dual-range
cutoffs (i.e. rlist=rvdw).

Jason