A Reliable and Inexpensive Flexible Molecule Crystal Structure Prediction Protocol Based on First Principles

Crystal structure prediction (CSP) methods are of importance for pharmaceutical, electronic, agricultural, and energetic materials. Most CSPs are performed by minimizing lattice energies of quasi-randomly generated polymorphs using either atom-atom force fields (FFs) or dispersion-augmented periodic density functional theory (pDFT+D) calculations. In the former case, the FFs can be of empirical nature or tailor-fitted to results of ab initio calculations. It has been recently shown that intermonomer FFs fitted to symmetry-adapted perturbation theory interaction energies, inter-aiFFs, perform exceedingly well compared to empirical FFs (empFFs) for crystals with rigid monomers. Here, we show that empFF-based CSPs for crystals with flexible monomers are generally not reliable and design a method for developing intramonomer FFs fitted to ab initio calculations for monomers (intra-aiFFs). These were used together with inter-aiFFs in full-dimensional CSPs to predict the crystal structure of 2-acetamido-4,5-dinitrotoluene with 6 soft degrees of freedom. For the 1000 lowest lattice energy polymorphs predicted by such an aiFF-based approach, pDFT+D calculations were performed without optimizations of geometries. Next, the top-ranked 100 polymorphs were fully optimized using pDFT+D. This protocol resulted in the experimental crystal being ranked as number 2 at much lower costs than those of other reliable approaches. Our method of developing intra-aiFFs should also have important implications for biomolecular simulations.