Peridynamics EPS model is broke

MD_Simulation · July 4, 2016, 8:53am

Hello Users,

I am trying to use the peridynamics eps pair_style, however, I am finding it to be unstable if the geometry is smaller than 1 meter.

For instance, if I use the following settings, the simulation is stable:

“”"

units si
lattice sc 0.25
region r_block block -0.5 0.5 -2.0 2.0 -0.5 0.5 units box
“”"

But if I simply divide by 10 and use the following settings, the simulation is no longer stable, even with a timestep 10 times smaller:

“”"

units si

lattice sc 0.025
region r_block block -0.05 0.05 -0.20 0.20 -0.05 0.05 units box
“”"

I have attached two input files that yield stable and unstable runs with the two different geometry sizes. The inputs are simulating a tensile test in the y-axis

I don’t believe it is a timestep issue (CFL gives a timestep around 1E-6 should be stable, I’m using 1E-7)

I believe there is something wrong with the actual EPS code.

If anyone has any ideas that might help, I would love to hear.

Thanks,

Abby

stable.lmp (1.45 KB)

unstable.lmp (1.36 KB)

sjplimp · July 8, 2016, 2:22pm

I think the PMB model in LAMMPS is one of the original PD models we put

in LAMMPS.

I don’t know stabilitiy criteria for PD. I’m CCing 2 people, who

have both developed PD models in LAMMPS, Rezwan Rahman (UTexas)

and John Mitchell here at Sandia.

Steve

rezwan_rahman · July 8, 2016, 2:41pm

To my knowledge there is no issue in the code which could raised this problem. I would suggest to debug the code by printing different components of force state. For simplicity you can do this on single processor. Try to see what happens when you scale down the system. Probably the force state becomes very high. I didn’t see any memory issue before while storing the dynamic parameters, i.e. elastic-plastic strain components. However, debugging might help with finding the source of the problem.