Active learning and element embedding approach in neural networks for infinite-layer versus perovskite oxides

The observation of superconductivity in NdNiO₂ films on SrTiO₃(001) by Li et al. [1] has sparked considerable interest in the materials class of infinite-layer oxides. Here, we combine first-principles simulations and active learning of neural networks to explore formation energies of oxygen vacancy layers, lattice parameters, and their statistical correlations in infinite-layer versus perovskite oxides across the periodic table, and place the superconducting nickelate and cuprate families in a comprehensive context. Neural networks accurately predict these observables, which act as a fingerprint of the complex reduction reaction, using only a fraction of the data for training. Element embedding identifies chemical similarities between the individual elements in line with human knowledge. Active learning renders the training highly efficient, based on the physical concepts of entropy and information, and provides systematic accuracy control. This exemplifies how artificial intelligence may assist on the quantum scale in finding novel materials with optimized properties. [2]