fix_nve_limit

hi
I've got an experimental data-file containing coordinates of spherical grains, however there are some wrong overlaps between these spherical grains, i have tried to import this data file into the liggghts and let the package relax, so the overlaps are gone,
I have used fix_nve/sphere but the overlaps make the grains going on an exploding manner, even if i reduce the time-step, the same thing happens but in a longer time,
I have used fix_nve_limit and everything looks fine: after a little displacement the grains settle down,

I have been wondering if it is logical to use fix_nve_limit instead of fix_nve/sphere, since fix_nve_limit treats grains like point particle not spherical particles? in the case of fix_nve_limit : does the friction play any role?
it's even more surprising that: when i relax the system with fix_nve_limit for a while and then unfix the nve_limit and again apply fix_nve/sphere the systems follow the same exploding manner.

the only thing that makes me doubt about fix_nve_limit is that if it's not simulating a real physical system in which grains have rotational degrees of freedom.

Thanks