Summary We conducted a quantum-based molecular dynamics simulation of HMX at a density of 1.9 g/cm3 and temperature of 3500 K for up to 55 picoseconds has been conducted. These are conditions similar to those encountered at the Chapman-Jouget detonation state. Thus, although we do not model the entire shock process, we can provide some insight into the nature of chemical reactivity under similar conditions. Under the simulation conditions HMX was found to be in a highly reactive dense supercritical fluid state. We estimated effective reaction rates for the production of H2O, N2, CO2, and CO to be 0.48, 0.08, 0.05, and 0.11 ps−1, respectively. The simulation is being extended until a steady state for the production of these products is reached. The simulation can serve as a basis for the construction of a global decomposition mechanism of HMX that, lacking experimental data, can be validated through standard ab initio quantum mechanical methods. The reported results can be more fully validated through comparison with experimental findings at similar conditions, which we hope this study will motivate. Since the SCC-DFTB method can be implemented in parallel mode  with great efficiency, the size of the system can be increased to a few thousands atoms without imposing severe limitations on the simulation time. Overcoming the shortcoming of system size will allow us to consider solid carbon formation, which is not accounted for in our current simulation due to the limited system size and simulation time. The reaction rates were implemented in a three-step chemical/hydrodynamic/thermal model of flame propagation, allowing for better agreement with recent experimental results on the flame speed.  Further, we find reasonable agreement for the concentration of dominant species with those obtained from thermodynamic calculations. Although the present work sheds much needed light on the chemistry of energetic materials under extreme conditions, there are methodological shortcomings that need to be overcome in the future. The demanding computational requirements of the present method limit the simulation time to less than 100 ps. In fact, the simulations reported here took over one year of simulation time on a modern workstation. This limits the methods applicability to short times and corresponding high temperature conditions. Due to experimental difficulty, however, most work that resolves chemical speciation is done at significantly lower temperatures. A second issue is that the SCC-DFTB method is not as accurate as more elaborate ab initio methods. The high temperature conditions of the present work ameliorate this difficulty. For instance, at 3500 K a 10 kcal/mol error in a reaction barrier leads to a factor of 4 error in the reaction rate. This is an acceptable error for qualitative study. At 600K, however, a comparable error in the barrier would lead to a factor of 4000 error in the reaction rate. There is no quantum molecular dynamics technique currently accurate enough to provide reliable reaction rates at 600K. Nonetheless, we find the present approach to be a promising direction for future research on the chemistry of energetic materials.