(623e) Molecular Crystal Structure Prediction with Genarris and Machine Learned Inter-Atomic Potentials

Genarris generates molecular crystal structures in all space groups compatible with the number of molecules per unit cell (Z) and the molecular symmetry, including in space groups with special Wyckoff positions. Genarris can be used to generate random structures for training machine learned inter-atomic potentials (MLIPs), to generate initial populations for genetic algorithms or swarm optimization, or as a standalone crystal structure prediction (CSP) workflow. We present a new version of Genarris, in which we have implemented the "Rigid Press" algorithm to generate close-packed structures. In addition, we have interfaced Genarris with MLIPs for geometry relaxation and energy ranking. Genarris 3.0 with system-specific AIMNet MLIPs was used successfully in the 7th CSP blind test.