[lammps-users] bond atoms missing error

hello all,
I have the following type of error message when running a bead spring polymer simulation with 2 types of bonds:
ERROR on proc 16: Bond atoms 108815 28773 missing on proc 16 at step 77432
the timestep it crashes and the bond atoms vary in the different runs. The problematic bond is always of type nonlinear, FENE are fine.
my pair potential and bond parameters are:

pair_style lj/cut 1.12
pair_coeff 1 1 1.0 1.0 1.12
pair_coeff 1 3 1.0 1.0 1.12
pair_coeff 3 3 1.0 1.0 1.12
pair_coeff 1 2 0.82 0.6875 0.77
pair_coeff 2 3 0.82 0.6875 0.77
pair_coeff 2 2 0.0 1.0 0.0
neigh_modify exclude type 2 2
pair_modify shift yes
bond_style hybrid fene nonlinear
bond_coeff 1 fene 30.0 1.5 1.0 1.0
bond_coeff 2 nonlinear 1.5 0.8 0.8
special_bonds 0.0 1.0 1.0

nonlinear bonds are always between atom types 2 and 3.
Looking at the error messages description on the web site I found:
One or more of 2 atoms needed to compute a particular bond are missing on this processor. Typically this is because the pairwise cutoff is set too short or the bond has blown apart and an atom is too far away.

I don't think my bond is physically exploding, I successfully simulated similar systems with global LJ cutoff of 2.5.
I thaught that bond neighbors were on a different bond neighbor list and were not affected by the pair cutoff (well, I guess I was wrong). So I don't quite understand which pairwise cutoff I should boost and by how much. Do I need to boost the global cutoff or just the one between the atom types belonging to the exploding bond type? I tried increasing types 2 3 cutoff up to 1.1 and it is still happening. Maybe cutoff+skin should be greater or equal to the maximum extension of the bond (i.e. 1.6)? Or is it the global cutoff that has to fulfill that condition?
The problem is that I would like to keep my cutoff low to gain speed and moreover, I don't want attractive interactions. I was thinking of changing the skin distance but it would affect all atom types and slow down my simulation. So what are the solutions and which one would be the fastest?

thanks in advance,

Maybe cutoff+skin should be greater or equal to the maximum extension
of the bond (i.e. 1.6)?

This is the solution. You don't have to change the pair potential
cutoffs, but you need to insure that those cutoffs + skin will always
acquire the bonded atoms (with a bit of extra). So I would try
setting your skin to 1.12 + 0.5 or something like that.