Machine learning the magnetocaloric effect in manganites from compositions and structural parameters

Solid-state refrigeration techniques have drawn increasing attention due to their potential for improving the energy efficiency of refrigeration and temperature-control systems without using harmful gas as in conventional gas compression techniques. Research on magnetocaloric lanthanum manganites with near-room-temperature Curie temperature shows promising results for development of magnetic refrigeration devices. Chemical substitutions are one of the most effective methods to tune the magnetocaloric effect, represented by the maximum magnetic entropy change (MMEC), through the incorporation of various lanthanides, rare-earth elements, alkali metals, alkaline-earth metals, transition metals, and other elements. Some theories based on lattice distortions and double-exchange interactions show that ionic radii of the dopants and final compositions correlate with the MMEC, but the correlations are generally limited to A-site substitutions and become less applicable to multi-doped manganites than single-doped ones. In this work, the Gaussian process regression model is developed as a machine learning tool to find statistical correlations between the MMEC and structural parameters among lanthanum manganites. More than 70 lattices, cubic, pseudocubic, orthorhombic, and rhombohedral, with the MMEC ranging from 0.65 J kg−1 K−1 to 8.00 J kg−1 K−1 under a field change of 5 T are explored for this purpose. Structural parameters utilized as descriptors include ionic radii at both A- and B-sites, ⟨Mn–O⟩ bond length, ⟨Mn–O–Mn⟩ bond angle, and compositions consisting of up to six elements. The modeling approach demonstrates a high degree of accuracy and stability, contributing to efficient and low-cost estimations of the magnetocaloric effect.