Electronic structure and atomistic modelling for materials science
The Applied Physics group at LTU is a part of the Division of Materials Science, and we study the properties materials and their interfaces using electronic structure calculations, atomistic modelling and machine learning. The materials development in the information society is constantly moving in the direction of thinner layers and nano-structured materials, and measurements at smaller scales, in what is termed nanotechnology. In this regime information on the atomic scale is desirable and necessary, since atomic and molecular scale properties and processes govern material properties and their interactions. We model material properties on two levels of theory: electronic structure theory and atomistic theory, such as molecular mechanics molecular dynamics (MM-MD), where the atom is the smallest building block. This modelling is complemented with quantum mechanical modelling based on density functional theory (DFT), which is used to model hundreds (up to thousands) of atoms in processes that form and break chemical bonds, while MM-MD is used to study the dynamics of millions of atoms for several nanoseconds. These two theories are combined in quantum mechanical molecular dynamics (QMD) that is used for the study of dynamic processes where chemical bonds are reformed. Machine learning (ML) can combine the high accuracy of DFT with the low computational cost of MM. We use ML to train neural network potentials (NNPs) on DFT reference data which are then used to model processes that require both high accuracy and high throughput. To describe the materials within this proposal it is often necessary to treat the quantum mechanical nature beyond the state-of-the-art method DFT, specifically with inclusion of relativistic effects, explicit electron correlation, and with the inclusion of proper description of van der Waals interactions between molecules and material surfaces. The inclusion of these effects adds another level of complexity to the simulations, rendering them even more time-consuming.