Fig. 1: Workflow of the ShotgunCSP algorithm.
From: Shotgun crystal structure prediction using machine-learned formation energies
a Virtual screening using a machine-learning-based energy predictor. A virtual library is created using a machine-learning-based element-substitution method (ShotgunCSP-GT) or a Wyckoff position generator (ShotgunCSP-GW). b Construction of formation-energy predictor based on a CGCNN. The CGCNN was trained with the Materials Project database and fine-tuned for the energy prediction of pre-relaxed candidate crystal structures for a query composition X.
