I am trying to simulate the detachment of grafted polymers from a surface. I am currently doing this with simple bead-spring polymers, with atoms bound to each other and the surface by FENE potentials, and using the bond/break fix to break any surface-polymer bonds that exceed their equilibrium length by a small margin. All pair interactions are cut-off LJ, with interactions for directly bonded atoms set to zero via special_bonds (as required for the FENE bond type).
However, my simulations produce segfaults related to an invalid pointer, often in combination with the "fix bond/break needs ghost atoms from further away" error message (despite already using large values for the neighbor skin distance and communicate cutoff). These issues primarily occur during the run, but sometimes when writing a data file afterwards. In the latter case, an apparently valid data file is produced despite the error.
The problem shows up more quickly in large or dense systems, but it also occurs for a minimal system of 2 interacting grafted polymers. For a given system and input file, it always happens at the same time step. The simulation works fine when I take out the bond/break fix. I am working with the lammps-daily executable on Ubuntu 14.04. The problem occurs considerably faster (in terms of timesteps) when running lammps in parallel compared to serially, but working serially does not eliminate the problem.
Any input on what might be going wrong and how to fix it would be fantastic, as I am quite baffled myself.
Thanks in advance,
Guido Ritsema van Eck