The issue of HMX phase transition under hydrostatic compression is not clear and experiments show conflicting results. Effective solution via first-principles simulation is challenged by difficulty of accurate prediction of Van der Waals interaction, which exists ubiquitously and is crucial for determining the structure of molecules and condensed matter. We have contributed to this by constructing a set of pseudopotentials and pseudoatomic orbital basis, specialized for molecular systems with C/H/N/O elements. The reliability of the method is verified from the interaction energies of 45 complexes (comparing to the results of coupled cluster with singles and doubles (Triple) (CCSD)(T)) and the crystalline structures of 7 typical explosives (comparing to experiments). Using this method, we complete the phase diagram of HMX under static compression up to 50 GPa. We make it clear that no β→δ/ε→δ phase transition occurs at 27 GPa, which has long been a hot debate in experiments. A possible γ→β phase transition is found at around 2.10 GPa in the environment of vapour. We have also predicted the equation of states for α-, δ-, and γ-HMX, which are experimentally absent.