Synthesis-condition recommender system discovers novel inorganic oxides

Accurately predicting successful synthesis conditions to prepare new pseudo-binary oxides remains a challenge despite extensive research. This study presents a synthesis-condition recommender system to efficiently explore a wide chemistry space composed of 10 206 pseudo-binary oxide compositions. As a training dataset, we systematically performed 1542 synthesis experiments by a polymerized complex method. The results were organized into a tensor-type database. Then we scored 66 150 unexperimented synthesis conditions by the tensor-decomposition method and arranged the results in the order of increasing predicted score. We selected 300 synthesis conditions from the top predictions of unexperimented compositions and verified the predictive performance by conducting additional experiments. The fraction of successful syntheses was approximately proportional to the predicted scores, validating the synthesis-condition recommender system. Additionally, the synthesis experiments for the high-scored conditions led to the discovery of two yet-to-be-found pseudo-binary oxides: La4V2O11 and La7Sb3O18. La4V2O11 and La7Sb3O18 have similar crystal structures to gamma-Bi4V2O11 and La7Ru3O18, respectively. The synthesis-condition recommender system gave higher predicted scores to the successful synthesis conditions of those materials.