First Principles Investigations of Novel Condensed Matter Materials
Adeniyi, Adebayo Oluseun
MetadataShow full item record
The advent of very fast computing power has led to the positioning of theoretical investigations of condensed matter materials as a core part of research in this area. Often the results of such numerical and computational investigations serve as reliable guide for future experimental exploration of new materials and has led to the discovery of numerous materials. In this thesis, state-of-the-art first principles calculations have been applied to investigate the structural, electronic and dynamical properties of some novel condensed matter materials. The novelty of these compounds stems from the fact that they challenge our previous knowledge of the chemistry of chemical reactions that support the formation and stability of chemical compounds and can therefore expand our frontier of knowledge in the quest for scientific understanding of new atypical compounds in high pressure physics. In the first project, the long sought post-Cmcm phase of the cadmium telluride is characterized with the application of first principles metadynamics method. It has a monoclinic unit cell and the P21/m space group. Enthalpy calculation confirms this phase transition sequence and further predicts a P21/m to P4/nmm transition near 68 GPa. Interestingly, the enthalpies of CdTe compounds are found to be higher than the enthalpy sum of its constituents Cd and Te at pressures higher than 34 GPa which is an indication that the com-pound should decompose above this pressure point. However, dynamical stability revealed in the phonon dispersion relations prevents the decomposition of CdTe at high pressure. This suggests that CdTe becomes a high-enthalpy compound at high pressure. The second project is directed towards the prediction of stable helium-hydrogen compound. In spite of extensive experimental and theoretical work, a general agreement on the crystal structure and stability of the helium-hydrogen system is lacking. In this study, the possibility of helium forming stable compound with hydrogen is investigated by using first principles structure search method. A stable helium hydrogen compound formed at ambient conditions is found. It belongs to the triclinic P-1 space group, having He(H2)3stoichiometry. Topological analysis of electron density at the bond critical points shows there exists a quantifiable level of bonding interaction between helium and hydrogen in the P-1 structure. At ambient pressure, the compound is characterized and stabilized by interactions with strength typical of van der Waals interaction that increases with pressure. This current results provide a case of weak interaction in a mixed hydrogen-helium system, offering insights for the evolution of interiors of giant planets such as Jupiter and Saturn. In the final project, a machine learning potential is successfully created for sodium based on the Gaussian process regression method and weighted atom-centered symmetry functions representation of the potential energy surface. Here, sodium potential energy surface is described using different relevant data sets that represent several regions of the potential energy surface with each data set consisting of three element groups which are total energies, inter-atomic forces, and stress tensors of the cell, which were constructed from density functional theory calculations. It is demonstrated that by learning from the underlying density functional theory results, the trained machine learning potential is able to reproduce important properties of all available sodium phases with an exceptional accuracy in comparison to those computed using density functional theory. In combination with the metadynamics method, this well trained machine learning potential is applied to large simulation boxes containing1024 and 3456 sodium atoms in the cI16 phase. These large-scale simulations reveal a notable phase transition at 150 K and 120 GPa with an impressive capturing of the rearrangements of atomic configurations involved in the transition process that may not be evident in asmall-scale simulation. Without a doubt, this work shows that applying machine learning methods to condensed matter systems will lead to significant increase in our understanding of important processes such as atomic rearrangements, growth and nucleation process in crystal formation and phase transition.
DegreeDoctor of Philosophy (Ph.D.)
DepartmentPhysics and Engineering Physics
CommitteeChang, GapSoo; Tse, John; Xiao, Chijin; Pan, Yuanming
Copyright DateSeptember 2021
Density functional theory
Machine Learning Potential.