Tensor Factorizations for Recommending Perovskite Crystallization Trials
Matrix factorizations have been used to build recommendation systems for over a decade, with their primary use case therein being the completion of unfilled entries in user-item ratings matrices via low rank approximation. Analogously, tensor factorization has been used to complete ratings tensors that model ratings as a function of more than two variables. This method has recently been applied to the recommendation of novel chemical reactions with promising results. In the present study we use tensor factorizations to recommend reactant concentrations at which to grow crystals of metal-halide perovskites. This method is evaluated through simulated campaigns of crystal trials based on the largest known dataset of metal-halide perovskite crystallization experiments. Results indicate a 19.8% improvement in success rate of crystal growth when following tensor-factorization based recommendations over randomly selected crystal trials. Further work will be dedicated to refining this method and deploying it in the laboratory.
Computer science|Computational chemistry|Materials science
Tynes, Michael, "Tensor Factorizations for Recommending Perovskite Crystallization Trials" (2020). ETD Collection for Fordham University. AAI28027265.