Differences in the stress tensor when using primitive or mesh walls

Hi, I ran some simulations with dense granular material where the particles are stored in different containers ( cube, cylinder and annular cylinder) and the only external force on them is gravity. I model the particle-particle and the particle-wall contacts with the Hertz contact model with the history model enabled (without adding cohesion nor rolling friction). Every case has been run once using .stl meshes and replicated using primitive walls. Despite the geometries are the same and the simulations have been run under the same conditions, the stress tensor calculated with "compute peratom all stress/atom" has different results between simulations with .stl meshes and simulation with primitive walls. The stress tensor calculated with primitive walls shows more reliable results (Janssen equation is fulfilled when the height of the bed of particles is high enough) meanwhile the stress tensor using .stl meshes shows unrealistic results (sigma_zz bigger than the weight of the bed in some regions...). I should run other cases with more complex geometries which doesn't allow me to keep using primitive walls. It seems that the particle-wall contact is solved in different ways depending on the type of wall in LIGGGHTS code. Then, I would like to know if:
- is it true that the collisions between particles and walls for primitive and mesh walls is solved in different ways?
- Can one change the code to model the mesh collisions in the same way that the primitive walls?
- Any suggestions about a simplification or a way to calculate the collisional stress tensor in a simulation with .stl meshes?
Many thanks for your help