Improved predictions of thermomechanical properties of molecular crystals from energy and dispersion corrected DFT

Thermal stability and pressure-dependent changes are key to molecular crystals and their properties. The determination of their thermal properties from ab initio methods is, however, a challenging task. While the low-frequency phonon spectrum related to intermolecular vibrations remains difficult to describe, the Quasi-Harmonic Approximation (QHA) also induces for molecular crystals a significant volume deviation, which makes their thermal behavior ill-determined. To overcome these difficulties, we consider a pragmatic energy correction (EC) that has long been used for atomic crystals, and we presently report the first ever use for molecular crystals. Applying the QHA in dispersion-corrected density functional theory (DFT-D) calculations with an ab initio parameterized EC, the resulting model can simultaneously and accurately derive thermal and mechanical properties of high-explosive molecular crystals. When compared to experiments, the mean absolute percent error of previous DFT-based thermomechanical models is 12% for mechanical and 31% for thermal properties. Our model performs significantly better and reduces these uncertainties to 4.1% and 9.8%, respectively. In particular, the agreement between our model and experiments for the thermal properties is three times better. This significant improvement greatly benefits the determination of thermomechanical properties such as the Grüneisen parameter and the shock properties. The method has been successfully applied to molecular crystals showing a large diversity of weak intermolecular interactions (β-1,3,5,7-tetranitro-1,3,5,7-tetrazoctane (HMX), α-1,1-diamino-2,2-dinitroethylene (FOX-7), Triaminotrinitrobenzene (TATB), ε-Hexanitrohexaazaisowurtzitane (CL20), and Pentaerythritol tetranitrate (PETN)-I). Due to its accuracy and transferability, our model is expected to work for a large class of computationally designed molecular crystals and co-crystals, providing a basis for a predictive framework.