Machine learning for two-dimensional materials' properties

Abstract

Materials are represented by their atoms, composition, and structure (ACS). Materials with different ACS descriptions present a different set of properties. However, there is a complex and intriguing relationship between the ACS and materials' properties. Machine learning can explore such relations and accelerate the evaluation of properties by using fingerprints of known materials to find an approximate function to predict the property of unknown materials. Topological insulators emerge as exotic systems featuring insulating bulk and symmetry-protected metallic boundary states with potential applications in spintronic and spin-orbitronic devices, low energy loss devices, and quantum computing. In real materials, the topological properties are sensible to structural modifications such as defects, temperature, chemical substitution, strain, and amorphous phases. These modifications are shown to transform topological materials into trivial ones but also turn trivial materials into topological ones. We propose to study the relationship between materials' structure and topological properties using machine learning and data-driven methods. We will investigate both global and local structure descriptors to gain insight into the role of structure and symmetries. Also, we will explore the local effects on global properties. Finally, this could allow us to apply our models to study disordered materials and realistic structures. (AU)