Properties of materials are constituted by molecular structures. The molecular dynamics simulation technique has an advantage in clarifying the relationship between molecular structure and properties through analysis of atomistic structure and dynamics, which are difficult to measure by experiment. On the other hand, major problems with applying this technique to larger and more complicated molecules consist of controlling the calculation time, and the larger scale of molecules. In this study, first, a relationship between melting properties and molecular structure is investigated by using the atomistic molecular dynamics simulation to verify effectiveness of this technique. The relationship between the reciprocal of self diffusion coefficients obtained from the simulation and 1/2 FT, which is an experimental indicator for melting properties, shows good agreement. This result indicates that atomistic molecular behavior affects melting properties, and implies that polymers with Bisphenol A structure have higher molecular dynamics. Next, a reduction of calculation cost and an enlargement of molecular simulation are examined by following 2 approaches—the coarse-grained united-atom model and the supercomputer K computer. The coarse-grained united-atom model shows good parallel scalability. For polycarbonates, it has the equivalent accuracy of the full atomistic model, and is more than 15 times faster than the full atomistic model with the same number of molecules and cores. These techniques expanded the spatial scale of molecular dynamics simulation by one digit larger than ever before.