Accelerating crystal structure prediction by machine-learning interatomic potentials with active learning

In this letter we propose a new methodology for crystal structure prediction,\nwhich is based on the evolutionary algorithm USPEX and the machine-learning\ninteratomic potentials actively learning on-the-fly. Our methodology allows for\nan automated construction of an interatomic interaction model from scratch\nreplacing the expensive DFT with a speedup of several orders of magnitude.\nPredicted low-energy structures are then tested on DFT, ensuring that our\nmachine-learning model does not introduce any prediction error. We tested our\nmethodology on a problem of prediction of carbon allotropes, dense sodium\nstructures and boron allotropes including those which have more than 100 atoms\nin the primitive cell. All the the main allotropes have been reproduced and a\nnew 54-atom structure of boron have been found at very modest computational\nefforts.\n