Explainable Recommendation Engines to Predict Complex Intermetallics: Synthesis and Characterization of Gd₁₀RuCd₃, a Neutron Absorption Material

The search for a neutron absorption material was aided by materials informatics tools that comprise the developed recommendation engines. A new member in the ternary RE₁₀MCd₃ series, Gd₁₀RuCd₃, was predicted and synthesized as an experimental validation. The recommendation engine is based on partial least-squares-discriminant analysis (PLS-DA) crystallographic site processing to classify the site preference of elements in compounds crystallizing in the Y₁₀RuCd₃-type structure and upon projecting other elements onto Principal Component map, the titled compound emerged as the top candidate. The distinguishing feature of the developed recommendation engine is in the three explainable methods utilized in the predictive framework: unrestricted, conservative, and cluster methods. Predictions are visualized with convenient property projection tools. The prediction was validated by high-temperature synthesis. The structure was confirmed with single-crystal and powder X-ray diffraction. The cold-water-quenched samples quenched at 800 °C have a smaller unit cell volume than the samples quenched at 600 °C annealing temperature. Transport property measurements show an unusually low thermal conductivity (5-7 W·m^-1 K^-1) and a trend change indicates a possible structure transition between 600 and 800 °C. The unusual low thermal conductivity was predicted with a machine-learning model that focuses on thermoelectric property prediction. DFT analysis reveals the presence of a 0D electride phenomenon. The neutron cross section and absorption based on the constituent elements put our material within the top 0.4% of neutron absorbers. The negative thermal expansion and high mass absorption suggest Gd₁₀RuCd₃ as an attractive neutron absorption material.