Missing atoms OPLS force fied

Hi,
I am trying to simulate polymer systems using OPLS force field.
The system consists of Polyethylene chains and peroxide molecules on top of them.
I build the structure PE and peroxide parts separately and merge them using read data command twice.
Everything works fine. I minimize the structure without any problem. The system is a slab, periodic in x-y directions only. I defined a LJ wall at top and bottom of the structure using :

fix bottomwall all wall/lj93 zlo 5.0 2 2 5
variable linear equal ramp(170,105.0)
fix topwall all wall/lj93 zhi v_linear 2 2 5
run 100000

The top wall moves down to 105A in 100000 steps to compress the system in order to set the correct density.To avoid close contacts I built the structure with low density at the beginning.
Still no problem here. Bottomwall is fixed.
I start data collection part, nvt simulation at 500K for 2000000 steps.
After 243398 steps I received this error:

ERROR on proc 2: Dihedral atoms 17112 17117 17118 17032 missing on proc 2 at step 243398 (../ntopo_dihedral_al
l.cpp:72)

I searched this error, it says usually bad construction of structure like close contacts of atoms etc. However I do not think thats the case here because the simulation runs more than 300 000 steps before getting this error and also minimizes without any problem. Here is my neighbor settings:

neighbor 7.0 bin
neigh_modify every 2 delay 2 check yes

I increased the skin size to see it helps but no luck. Increased the neighborlist build frequency but still error persists.
Finally my force field parameters:

pair_style lj/cut/coul/cut 10.0 8.0
bond_style harmonic
angle_style harmonic
dihedral_style opls
pair_modify mix geometric tail yes
special_bonds lj/coul 0.0 0.0 0.5

#force field parameters

Hi,
I am trying to simulate polymer systems using OPLS force field.
The system consists of Polyethylene chains and peroxide molecules on top
of them.
I build the structure PE and peroxide parts separately and merge them
using read data command twice.
Everything works fine. I minimize the structure without any problem. The
system is a slab, periodic in x-y directions only. I defined a LJ wall at
top and bottom of the structure using :

fix bottomwall all wall/lj93 zlo 5.0 2 2 5
variable linear equal ramp(170,105.0)
fix topwall all wall/lj93 zhi v_linear 2 2 5
run 100000

The top wall moves down to 105A in 100000 steps to compress the system in
order to set the correct density.To avoid close contacts I built the
structure with low density at the beginning.
Still no problem here. Bottomwall is fixed.
I start data collection part, nvt simulation at 500K for 2000000 steps.
After 243398 steps I received this error:

ERROR on proc 2: Dihedral atoms 17112 17117 17118 17032 missing on proc 2
at step 243398 (../ntopo_dihedral_al
l.cpp:72)

I searched this error, it says usually bad construction of structure like
close contacts of atoms etc. However I do not think thats the case here
because the simulation runs more than 300 000 steps before getting this
error and also minimizes without any problem. Here is my neighbor settings:

minimizing and ​running for a long time does not automatically guarantee,
that close contacts won't happen. they can also happen due to bad force
field parameters, e.g. missing or too weak repulsive terms or incorrect
charges or missing or incorrect bond/angle/dihedral interactions.

neighbor 7.0 bin

​a large neighbor skin, is not really helpful here and slows you down a
*lot*.
optimal performance is usually somewhere between a skin size of 1.0 and 2.0​
a better way to prevent "long" dihedrals to be stretched too far across the
ghost area, is to increase the communication cutoff using comm_modify.
however, that does not help, in the case of problematic force field
settings.

neigh_modify every 2 delay 2 check yes

​rather more helpful is probably to change this to:

neigh_modify every 1 delay 0 check yes

is the most conservative choice.

if you can reproduce the "atoms missing" error reliably, you should record
a trajectory around those time steps with high resolution and visualize it
to check if you see some indication of atoms "shooting" through your
system. also recording thermo data and monitoring kinetic/potential energy
frequently can help to identify the cause. in case of close contacts, you
should see an unusual spike.

axel.