Here’s a simple way to reproduce the bug: take the latest version of lammps, compiled with the USER-OMP package. Use the standard LJ liquid benchmark input script: http://lammps.sandia.gov/inputs/in.lj.txt
Then simply change the pair style form lj/cut to lj/sf. Now run with and without the “-sf omp” flag and you’ll get very different thermodynamics. I think the lj/sf results are correct and the lj/sf/omp are wrong.
The error also occurs without the “-sf omp” flag, when “lj/sf/omp” and “package omp 1 force” are used in the inputs script, i.e. with serial neighbour lists. That’s why I think the error is in the lj/sf/omp pair style.
Dr. Jürg Diemand, Theoretical Physics and Computational Science, University of Zurich.