Oksidiperovskiittien mekaanisten ominaisuuksien laskeminen koneoppimisen keinoin

Chemistry, materials science, and the energy sector require the development of new materials with reduced toxicity and improved mechanical properties. In addition, industry-specific functionalities, such as piezoelectricity for electronic devices including actuators and energy harvesting systems, are increasingly important. One promising group of materials that may address these demands are oxide perovskites. Oxide perovskites comprise a wide range of elements from the periodic table, offering significant compositional flexibility. They have been studied for over hundred years; however, the growing de-mand for new materials may further accelerate research in this field, particularly with continued advances in atomic-scale simulation methods. The investigation of mechanical properties does not rely solely on experimental approaches. For several decades, Density Function Theory (DFT) has been widely used for atomic simulation purposes. More recently, Machine Learning Interatomic Potential (MLIP), a subset of artificial intelligence methods, have emerged as promising tools for accelerating DFT-level calculations. MLIPs are trained on large DFT-generated databases to approximate potential energy surfaces for atomistic simulations. MLIP enable faster calculations and simulations involving larger numbers of atoms reducing significantly computational costs. In this master’s thesis, the performance of Mattersim MLIP was evaluated. The calculated elastic constants and derived mechanical properties were compared with previously published results obtained using the same DFT functional on which the MLIP was trained. The results of the thesis indicate that the MLIP is currently best suited for large-scale screening studies involving extensive material sets. However, it remains susceptible to inconsistencies that directly affect the predicted properties. For applications requiring higher accuracy, conventional DFT calculations are therefore recommended.