I am simulating fully flexible polymer chains in the cubix box, periodic along x and y, and the repulsive wall along z-direction. I am having two problems in this setting:
1). I am getting error after running some steps: ERROR on proc 16: Bond atoms 19429 19430 missing on proc 16 at step 15899174 (src/ntopo_bond_all.cpp:61)
MPI_ABORT was invoked on rank 16 in communicator MPI_COMM_WORLD with error code 1.2). I am starting a simulation with all initial image flags zero. After running many steps, the image flags are not evolving from zero, though they are zero all the time while simulation is not breaking from the above error.
I am simulating fully flexible polymer chains in the cubix box, periodic along x and y, and the repulsive wall along z-direction. I am having two problems in this setting:
1). I am getting error after running some steps: ERROR on proc 16: Bond atoms 19429 19430 missing on proc 16 at step 15899174 (src/ntopo_bond_all.cpp:61)
MPI_ABORT was invoked on rank 16 in communicator MPI_COMM_WORLD with error code 1.
difficult to say. most likely explanation would be that atoms can eventually pick up enough kinetic energy to punch through the wall in z direction.
your input has an unusually small time step at 0.001 fs for using real units. if you run a larger time step, does this error happen sooner?
or what is the reason for this small timestep.
2). I am starting a simulation with all initial image flags zero. After running many steps, the image flags are not evolving from zero, though they are zero all the time while simulation is not breaking from the above error.
that depends on the specific geometry and connectivity. impossible to comment without seeing the data file and visualizing its included configuration.
why should those atoms diffuse through the box? that would be typical only for small molecules or atomic systems.
when I run your data file with the posted input file (please note that the wall fix cannot be used because the input is incomplete) under p p p boundary conditions, it cannot run.
what is highly suspicious is the very high potential energy. A closer look reveals that you have an extremely large bond energy.
for atom mass 1.0 in real units, this input should be able to run a stable time integration with time steps in the range 0.1fs to 0.5fs but that quickly leads to “bond atoms missing” errors.
upon closer inspection it seems that you have bond coefficients with an equilibrium distance of 4 angstrom, but your typical bond length is around 10 angstrom.
even changing that parameter, however, doesn’t stabilize the simulation, now the system quickly gets into other high potential energy configurations, this time for non-bonded interactions.
this means that basically both, your geometry/topology is inconsistent and ultimately bogus and also your force field parameters are not proper.
basically, this is a case of GI-GO (aka garbage in, garbage out) and the issue has nothing to do with the walls.
thus my advice:
set up a much smaller system to figure out and debug the force field and topology generating process. there obviously is a lot amiss there
run your tests with fully periodic boundary conditions
choosing the ridiculously small timestep (mostly) avoided the instabilities, but didn’t solve the fundamental problems. basically, you have been wasting a lot of your and some of my time.
the described procedure is a best practice when working with a new system and with tools you still need to gather experience with. just starting a simulation and adjusting parameter until it “runs” is no guarantee to get meaningful results. in fact, the opposite is the case.