[lammps-users] Different output with different LAMMPS versions

Dear LAMMPS users

Using two different LAMMPS versions i.e. 22 August, 2018 and 29 Oct., 2020, I am getting different results. The Gay-Berne model is used to simulate the aspherical particles. The simulations for both the versions begin by using the same restart file which is made in the older LAMMPS version i.e. 22 Aug., 2018. Using the GPU package with INTEL (parallel_studio_xe_2020_update1) and starting from the read_restart command, the system is allowed to run for a long time under the nvt ensemble using fix nvt/asphere command. In the older version of LAMMPS (i.e. 22 Aug., 2018), the phase coexistence is observed even after a very long run. But in the newer version (i.e. 29 Oct., 2020), there is no coexistence observed and the whole system results in a single phase even after a run of a few timesteps. I have looked at the changes made in LAMMPS during the period of Aug. 2018 to Oct., 2020. But I couldn’t find out the possible reason behind these unusual results. Please help me to know what changes in these two versions could lead to these different results? Thank you in advance.

The commands which I have used in my input script are :

newton off
package gpu 1
read_restart restart.filename

variable zlo equal zlo
variable zhi equal zhi
variable xlo equal xlo
variable xhi equal xhi
variable ylo equal ylo
variable yhi equal yhi
region box block {xlo} {xhi} {ylo} {yhi} {zlo} {zhi} units box
group Nall region box

variable dof equal count(all)
variable runtime equal ******

pair_style gayberne/gpu * * * *
pair_coeff * * * * * * * * * *

neighbor ** bin
thermo **
timestep **

compute q all property/atom quatw quati quatj quatk
fix 1 all nvt/asphere temp * * * mtk yes tchain *
fix_modify 1 energy yes
compute_modify 1_temp extra/dof {dof}* *compute pr all pressure 1_temp* *thermo_style custom step c_pr c_1_temp density pe ke etotal* <i>dump 1 all custom ** dump.filename id type x y z c_q[1] c_q[2] c_q[3] c_q[4]</i> *dump_modify 1 format line "%d %d %1.16le %1.16le %1.16le %1.16le %1.16le %1.16le %1.16le" sort id* _restart ** restart.1 restart.2_ *run {runtime}
undump 1
uncompute pr
unfix 1

Thanks and best regards,

Jagroop Kaur

Research Scholar,
School of Physics and Materials Science,

Thapar Institute of Engineering & Technology,

Patiala, Punjab, India.

LAMMPS binary restart files are in general not portable between LAMMPS versions so you may be reading corrupted data.

and in fact there has been a significant change in the restart file format. binary restart files prior to July 9th 2020 were missing data about aspherical particles and thus could not properly restart.
so any restart files created with older versions will be bogus.