Multiscale modeling of a polymeric system is a challenging task in polymer physics. Here we introduce a bottom-up and then top-down scheme for the simulation of polyethylene (PE). The coarse-grained numerical potential for PE is derived through an automatic updating program by mapping its radial distribution function (RDF) from the Lowe-Andersen temperature controlling (LA) simulation onto the one from detailed molecular dynamics (MD) simulation. This coarse-grained numerical potential can be applied in larger systems under the same thermodynamic conditions. We have tested the reliability of the derived potential in two ways. First, the blends of different linear low-density polyethylene (LLDPE) with high-density polyethylene (HDPE) have been simulated in LA with the coarse-grained numerical potentials and reasonable results are obtained. Moreover, Rouse scaling behavior is reproduced for monodispersed polymeric systems with different chain lengths. The atomistic details of the beads can be reintroduced into the coarse-grained HDPE and LLDPE/HDPE models, followed by a few MD runs to alleviate the local tension induced by this fine-graining procedure. The equilibrated large atomistic system can then be used for further studies.