LAMMPS WWW Site - LAMMPS Documentation - LAMMPS Mailing List Archives
Re: [lammps-users] compute aggregate/atom stucks with empty neighbor list
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [lammps-users] compute aggregate/atom stucks with empty neighbor list


From: Axel Kohlmeyer <akohlmey@...24...>
Date: Mon, 6 Nov 2017 13:54:15 -0500

quick notice. i can reproduce the behavior, but this is not so straightforward to sort out.
will let you know whether this can be fixed or whether there is some problem with your approach.

axel.

On Mon, Nov 6, 2017 at 8:06 AM, Michael King via lammps-users <lammps-users@lists.sourceforge.net> wrote:
Hello dear LAMMPS developers,

SHORT:

*compute aggregate/atom* gets stuck when run with mpirun and cutoff is too short and special_bonds is 0.0 0.0 0.0 (default)
supposed reason: Empty neighbor lists for some processors -> get stuck in the while loop of compute_aggregate_atom.cpp

LAMMPS Version: 22 Sep 2017

LONG:

We were interested in using aggregate/atom to study the cluster formation of our polymers.
Therefore, we did the simulation then using a rerun script to calculate the property and dump it into a file.
When we reduced the cutoff under a certain value, so that only the neighboring chains with a 'real' contact would be assigned to a cluster we ran in the following problem:

The rerun script runs in serial or mpirun -n 1, but not with more cores.

We can fix this behavior by a greater cutoff (unwanted) or special_bonds 1.0 1.0 1.0 (as stated in the doc).

So i believe that this is due to the while loop in compute_aggregate_atom.cpp. We don't have any clusters in the beginning of the simulation and when due to *special_bonds 0.0 0.0 0.0* the bonded atoms are removed from the neighbor list we get stucked in this loop since we iterate over 0 atoms and can not fulfill the condition for the break.

A test system is shown below.


Since this error can be fixed by appropriated special_bonds settings (see docs) i think its a minor problem,
but you can also create a case where special_bonds wont fix it, see below.
So we want to report it anyway.


TEST SYSTEM to reproduce the error:
LAMMPS Version: 22 Sep 2017

1.) use ${LAMMPSFOLDER}/examples/micelle/in.micelle and uncomment the dump command.

2.) run the simulation

3.) do a rerun with the following script (cluster.lammps):

<script>

dimension    2

atom_style    bond
read_data    data.micelle

pair_style      zero 2.5
pair_coeff * *
bond_style      zero nocoeff
bond_coeff *

# special_bonds lj 1.0 1.0 1.0

group alkane type 2 3 4
compute aggr alkane aggregate/atom 1.0
dump calc_agg alkane custom 1 aggregate.dat id c_aggr
dump_modify calc_agg sort id

rerun dump.micelle every 1 dump x y scaled yes box yes wrapped yes format native

</script>

should work: lmp -in cluster.lammps
should work: mpirun -n 1 lmp -in cluster.lammps
should not work: mpirun -n 4 lmp -in cluster.lammps

4.) uncomment * special_bonds lj 1.0 1.0 1.0*

should work: lmp -in cluster.lammps
should work: mpirun -n 1 lmp -in cluster.lammps
should work: mpirun -n 4 lmp -in cluster.lammps

5.) if i change my *aggregate/atom 1.0* cutoff to some value <= bond_length (0.75), i will get the same error but this time special_bonds wont help.

To track down where it went wrong (i know it changed the results of my cluster analysis) i just added

      if (inum == 0)
        change=1
        break;

after the second while(1) command in *compute_aggregate_atom.cpp* to check if the error comes from an empty neighbor list.
Then it will not stuck in the loop, but that quick&dirty test will change the results compared to the original code run in serial.


Greetings,
Michael King


------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
lammps-users mailing list
lammps-users@...655....net
https://lists.sourceforge.net/lists/listinfo/lammps-users



--
Dr. Axel Kohlmeyer  akohlmey@...92......  http://goo.gl/1wk0
College of Science & Technology, Temple University, Philadelphia PA, USA
International Centre for Theoretical Physics, Trieste. Italy.