Failure-Experiment-Supported Optimization of Poorly-Reproducible Synthetic Conditions for Novel Lanthanide Metal-Organic Frameworks

A series of novel metal organic frameworks with lanthanide double-layer-based inorganic subnetworks (KGF-3) was synthesized assisted by machine learning. Pure KGF-3 was difficult to isolate in the initial screening experiments. The synthetic conditions were successfully optimized by extracting the dominant factors for KGF-3 synthesis using two machine-learning techniques. Cluster analysis was used to classify the obtained PXRD patterns of the products and to decide automatically whether the experiments were successful or had failed. Decision tree analysis was used to visualize the experimental results, with the factors that mainly affected the synthetic reproducibility being extracted. The water adsorption isotherm revealed that KGF-3 possesses unique hydrophilic pores, and impedance measurements demonstrated good proton conductivities (σ = 5.2 × 10⁻⁴ S cm⁻¹ for KGF-3(Y)) at a high temperature (363 K) and high relative humidity (95%).