We should parallelize the central loops in the cell list container, using either atom or spatial decomposition (see e.g. doi:10.1006/jcph.1995.1039).