Fix rigid in flowing system

Dear LAMMPS users,

I have problem in simulation with rigid bodies.

My system consists of FENE chains and small planer rigid bodies. The system was bounded to two real walls in both z boundaries. The simulation was running ok for tens millions steps (LJ unit), until I started to shear the system by moving top and bottom walls at opposite directions.

Missing bonded atom on processors was reported after about 4 millions steps of continuously shearing. I have tried hard to debug this issue,

Step Temp utemp Press poswall PotEng TotEng Volume

19276053 0.99642222 1.1341181 -0.25016504 36.794468 12.045919 13.520054 15807.753

19276054 71.692212 82.415256 224849.69 36.794468 12.045768 118.10928 15807.753

From the thermo output above, it’s a “bad dynamics” problem. But I can’t tell what cause it.

As I said, it’s running fine until about 4 millions steps at shearing velocity 0.48.

  1. I tried lower velocity, 0.12 leads to fail at 16 million steps, 0.24 leads to fail at about 8 million steps. Which seems like relevant to how far the system have flow.

  2. The initial configuration are most likely fine, I checked image flags for rigid body (they are tightly arranged). The system also equilibrates well for 30 millions steps.

  3. I also tried to double the system, it still fail at around 4 millions steps.

  4. I have applied fix nvt with partial temperature (exclude the shearing direction) to the chains. And have tried both fix rigid/nve and fix rigid/nvt to the rigid bodies, it make no difference.

  5. Tried reduce time step from 0.005 to 0.002, failure happened at almost same time.

  6. System without rigid body works just fine.

  7. Tried increase communication cutoff from 2.8 to 4.0, doesn’t help.

After desperately investigating for weeks, I decided to post the problem here. Any suggestion would be greatly appreciated.

Thank you,

Di

Dear LAMMPS users,

I have problem in simulation with rigid bodies.

My system consists of FENE chains and small planer rigid bodies. The
system was bounded to two real walls in both z boundaries. The simulation
was running ok for tens millions steps (LJ unit), until I started to shear
the system by moving top and bottom walls at opposite directions.

Missing bonded atom on processors was reported after about 4 millions
steps of continuously shearing. I have tried hard to debug this issue,

​...and yet you fail to provide the most elementary and important piece of
information: the exact version of LAMMPS you are using and what platform
you are running on.​

Step Temp utemp Press poswall PotEng TotEng Volume

19276053 0.99642222 1.1341181 -0.25016504 36.794468 12.045919
  13.520054 15807.753

19276054 71.692212 82.415256 224849.69 36.794468 12.045768
  118.10928 15807.753

From the thermo output above, it's a "bad dynamics" problem. But I can't
tell what cause it.

As I said, it's running fine until about 4 millions steps at shearing
velocity 0.48.

1. I tried lower velocity, 0.12 leads to fail at 16 million steps, 0.24
leads to fail at about 8 million steps. Which seems like relevant to how
far the system have flow.

2. The initial configuration are most likely fine, I checked image flags
for rigid body (they are tightly arranged). The system also equilibrates
well for 30 millions steps.

3. I also tried to double the system, it still fail at around 4 millions
steps.

4. I have applied fix nvt with partial temperature (exclude the shearing
direction) to the chains. And have tried both fix rigid/nve and fix
rigid/nvt to the rigid bodies, it make no difference.

5. Tried reduce time step from 0.005 to 0.002, failure happened at almost
same time.

6. System without rigid body works just fine.

7. Tried increase communication cutoff from 2.8 to 4.0, doesn't help.

After desperately investigating for weeks, I decided to post the problem
here. Any suggestion would be greatly appreciated.

​have you checked the image flags right before it crashes? are they getting
close to +/- 512 for some atoms?

if atoms pass through the box too many times, the image flags will
overflow. they are usually 3 signed 10-bit integers packed into a single
32-bit integer. the fact that the number of steps when your calculations
fail and the velocity have an inverse proportional relation (point 1.)
supports this suspicion. you may need to update to a current version of
LAMMPS and compile it with -DLAMMPS_BIGBIG to increase the number of
available bits per image flag.

axel.

Hi, Axel,

-DLAMMPS_BIGBIG do solve the issue. Thanks a lot

Di