I am trying to run an Intel Accelerated simulation of lammps using a hybrid gayberne lj/coul/cut/long potential. After much troubleshooting, I’ve found that the combination of ellipsoids and the LJ/cut/intel potentials always causes the simulation to run normally for a while and then a particle will rapidly accelerate and break bonds or disappear (even with a tiny time step). The same code with just an LJ/cut (no intel) pair potential in the same compiled lammps program runs fine. The erratic behavior occurs much sooner for larger systems. I am using the 29Oct20 version compiled with intel and asphere packages.
Any suggestions?
Here is an example script from the simplest relevant system I could contrive. The non-intel pair potential completed successfully.
Run Command: mpirun -np 1 lmp -in atest.initialize > savefile_run.out
------- atest.initialize --------------
package intel 1 mode double
dimension 3
units lj
boundary p p p
atom_style hybrid full ellipsoid
pair_style lj/cut/intel 3.0 #cut applies to lj&coul
##################################################################
read_data “atest.data”
#############################
pair_coeff * * 0.1 1.0
#############################
group disks type 1
group spheres type 2
timestep 0.00001
dump 1 all custom 100000 ovito.trj id type xu yu zu
dump_modify 1 sort id
thermo_style custom step temp epair
thermo 100
#Run
fix 1 disks nve
fix 2 disks langevin 1.0 1.0 1 72281 angmom 1.0
fix 3 spheres nve
fix 4 spheres langevin 1.0 1.0 1 56782
run 100000000
write_data *.data
----------end file-------
----------atest.data---------
Lammps Description
2 atoms
0 bonds
0 angles
1 ellipsoids
2 atom types
0 bond types
0 angle types
0 5 xlo xhi
0 5 ylo yhi
0 5 zlo zhi
Atoms
1 1 0.0 0.0 0.0 1 0.0 1 2.0
2 2 2.5 0.0 0.0 1 0.0 0 1.0
Ellipsoids
1 1.5 1.0 1.5 1 0 0 0
-----end file--------