Advancing magnetic material discovery through machine learning: Unveiling new manganese-based materials

Abstract

Magnetic materials are used in a variety of applications, such as electric generators, speakers, hard drives, MRI machines, etc. Discovery of new magnetic materials with desirable properties is essential for advancement in these applications. In this research article, we describe the development and validation of a machine-learning model to discover new manganese-based stable magnetic materials. The machine learning model is trained on the input data from the Materials Project database to predict the magnetization and formation energy of the materials. New hypothetical structures are made using the substitution method, and the properties are predicted using the machine learning model to select the materials with desired properties. Harnessing the power of machine learning allows us to intelligently narrow down the vast pool of potential candidates. By doing so, we deftly reduce the number of materials that warrant in-depth examination using density functional theory, rendering the task more manageable and efficient. The selected materials, seemingly promising with their magnetic potential, undergo a meticulous validation process using the Vienna Ab initio Simulation Package, grounded in density functional theory. Our results underscore the paramount significance of input data in the efficacy of the machine learning model. Particularly in the realm of magnetic materials, the proper initialization of atomic magnetic spins holds the key to converging upon the true magnetic state of each material.