Machine Learning in Magnetic Materials
Georgios Katsikas, C. Sarafidis, Joseph Kioseoglou
Abstract
The technological advancements of every era of human civilization owe themselves to the materials available at the time. Despite the substantial interest in the discovery of novel materials, materials research remains a very delicate and time‐exhaustive procedure. Over the last 30 years, ab initio computational methods based on density functional theory (DFT) have allowed researchers to explore materials with ease and expect above‐experiment measurement precision. However, DFT‐based detailed investigation of novel materials is generally computationally intensive and greatly time‐consuming. This review presents machine learning methods applied to DFT simulation data to uncover connections between material structure, chemical composition, and magnetization, a target property chosen for its great industrial demand. Models are developed that can partially circumvent the need for simulation, guiding researchers in the design of magnetic materials. Specifically, the Materials Project database is examined and it is concluded that Eu, Gd, Pu, Fe, Np, Mn, U, Cr, Co, and Ce are amongst the most common elements found in magnetic materials, and that materials of the same composition may have different magnetization depending on their space group. A neural network capable of predicting magnetization with a standard error of 8.3 × 10 −3 μ B Å −3 is created.