I suspect that you are making the mistake of translating your expectations from macroscopic behavior and the way the visualization is done to the atomic scale without taking into account that things are different. In particular:
- bonds are not a barrier. They pull atoms together, but bonds can pass through each other easily, especially with classical models where no chemical reactions happen.
- atoms are not “hard spheres” but rather soft and squishy since the LJ potential (with default settings) is somewhat soft.
- if you pull hard enough, you will probably always observe the behavior that you are seeing. So you have to check whether the force that you are using to stretch the chains in realistic. In typical simulations of this kind, these forces are often exaggerated, so there is some movement within finite time, but experimentally, those may be orders of magnitude smaller and thus the sometimes not even accessible to MD simulation time scales.
To have chains not move through each other you have a number of options:
- make the atoms larger (increase sigma, or use pair style lj/expand) or more closely connected (reduce bond length)
- add repulsive particles that sit in the middle of each bond: pair_style srp command — LAMMPS documentation
Please note that your problem is related to this recent discussion: Help on Implementing Capsule-like Pair Interactions in LAMMPS