[lammps-users] reax/c problems

Hi! I have a problem with reax/c and would like to
find out what am I doing wrong...

When submitting my job, it seems to work for some time,
then it stops and I receive the message:

Warning: changed valency_val to valency_boc for X
step182-ran out of space on far_nbrs: top=18166, max=16782application called MPI_Abort(MPI_COMM_WORLD, step182\-ran out of space on far\_nbrs: top=18110, max=16768application called MPI\_Abort\(MPI\_COMM\_WORLD,
step182-ran out of space on far_nbrs: top=17850, max=16770application called MPI_Abort(MPI_COMM_WORLD, step182\-ran out of space on far\_nbrs: top=18250, max=16798application called MPI\_Abort\(MPI\_COMM\_WORLD,
mpiexec: Warning: tasks 0,2,4,6 exited with status 242.

My input file is the following:

# REAX potential
# MD test for MWCNT
dimension 3
units real
boundary p p p
atom_style full
newton on
read_data multip.data
pair_style reax/c NULL
pair_coeff * * ffield.reax 1 1
timestep 0.5
run_style respa 4 2 2 2 bond 1 dihedral 2 pair 3 kspace 4
velocity all create 500 312839 rot yes mom yes
fix 1 all langevin 500.0 500.1 50.0 46279
fix 2 all nve
fix 3 all qeq/reax 1 0.0 10.0 1.0e-6 reax/c
dump 1 all xyz 1000 dump.multi
restart 100000 restart.multi
run 500000

Is it possible to solve the problem ?

Thank you,
Ricardo

Aidan and Metin will have to answer this. Is your simulation
otherwise well-behaved? E.g. is the thermodynamics
output reasonable and have you visualized the system to
insure all is well?

Steve

I just noticed this:

timestep 0.5
run_style respa 4 2 2 2 bond 1 dihedral 2 pair 3 kspace 4

Why are you using rRESPA with ReaxFF? You have no bonds or dihedrals
or kspace defined, nor should you for ReaxFF. This will
invoke the pair reax at a timestep of 0.5/2 = 0.25 fmsec which
may be too large a timestep for ReaxFF.

I'm guessing you are doing bad dynamics.

Steve

I got a similar error when attempting to run an initial minimization on a Al2O3 structure. The error occurred after 22 steps of a minimization with fix box/relax. My (condensed) input and error message are:

Input

units real
timestep 0.5

pair_style reax/c NULL

pair_coeff * * /home/sxc033/Potentials/ffield.reax.AlO 7 3
fix 0 all qeq/reax 1 0.0 10.0 1.0e-6 reax/c

thermo 1
thermo_style $S
thermo_modify lost warn flush yes
fix 1 all box/relax tri 1.001 vmax 0.0001
min_modify line quadratic
minimize 0 0 1000 1000

fix 1 all ave/time 1 1 1 c_RDF file $A_Min.rdf mode vector

dump 1 all custom 1 $A_Min.dump id type x y z
dump_modify 1 flush yes
run 0
undump 1

Error Message

step24-ran out of space on far_nbrs: top=39473, max=39298step24-ran out of space on far_nbrs: top=39473, max=39298--------------------------------------------------------------------------
MPI_ABORT was invoked on rank 68 in communicator MPI_COMM_WORLD
with errorcode -14.

This looks like an error internal to the reax/c code, which would
be something for Metin to look into. Maybe Aidan wants
to take a look since it also involves fix box/relax. Two questions:

a) can you reproduce this problem with reax (not reax/c)?

b) is the simulation box or energy doing anything odd in
the steps leading up to the crash - e.g. is everything well
behaved in the previous iterations?

Steve

a) I will have to look into the original reax code more. I am having difficulties with this code because I exceed the defined MBONDDDEF even with the smallest box.

b) Yes, the volume, total energy, and pressure fluctuate less than 0.5%. The atom energy and lone-pair energy (as calculated with compute reax) fluctuate about 1.5%. The highest energy fluctuations are from the valence angle conjugation energy (4.8%).
– I’m calculating fluctuation percentages as standard deviation / average.

Thanks again,

Shawn P Coleman

University of Arkansas
Mechanical Engineering

You are using fix box/relax with reax/c. Sudden changes in box dimensions
can cause large changes in the number of neighbors of each atom. Apparently
this caused one of the internal reax/c neighbor lists to overflow, which the
code helpfully noticed. This may or may not be easy for Metin to fix. In the
meantime, I suggest you relax "more gently." LAMMPS provides many knobs for
doing this. Read the documentation on minimization and box relaxation and
experiment with the smallest possible system.

Aidan

Shawn,

We have heard about similar problems while doing energy minimization with reax/c. It looks like, during energy relaxations, systems are changing much faster than reax/c anticipates. We will try to address this problem, but I do not know how soon a fix can be made.

But currently, like Aidan said, you can play with fix box/relax parameters to relax your system "more gently". You can also try to decrease your timestep length, from 0.5 to 0.1 for example...

Metin

Thank you for your help.

Shawn P Coleman

University of Arkansas
Mechanical Engineering