Investigation of Lithium Manganese Oxides Using High-Dimensional Neural Networks
von Marco Eckhoff
Datum der mündl. Prüfung:2021-12-16
Erschienen:2022-03-03
Betreuer:Prof. Dr. Jörg Behler
Gutachter:Prof. Dr. Jörg Behler
Gutachter:Prof. Dr. Peter E. Blöchl
Gutachter:Prof. Dr. Eckhard Spohr
Zum Verlinken/Zitieren:
http://dx.doi.org/10.53846/goediss-9088
Dateien
Name:Dissertation_Marco_Eckhoff.pdf
Size:45.7Mb
Format:PDF
Lizenzbestimmungen:
Zusammenfassung
Englisch
Unraveling the atomic and electronic processes in modern energy materials is a key to advances in many important applications, from battery technology to heterogeneous catalysis. A prominent electrode material in lithium ion batteries is the lithium manganese oxide spinel LixMn2O4, with 0 ≤ x ≤ 2. Similar to many other transition metal oxides, the complex electronic structure of LixMn2O4 includes coexisting oxidation states and associated Jahn-Teller distortions. Therefore, electronic structure methods like hybrid density functional theory (DFT) are required for accurate calculations. However, hybrid DFT is computationally too demanding to study, for example, structural transitions, conductance, and reaction kinetics in extended simulations. To enlarge the length and time scales of first principles-quality simulations, machine learning potentials like high-dimensional neural network potentials (HDNNP) have emerged as a powerful tool. However, the representation of different oxidation states, Jahn-Teller distortions, and electron hopping poses a new level of complexity. Moreover, HDNNP-based simulations do not provide electronic structure information. To obtain atomic oxidation and spin states, the high-dimensional neural network spin prediction method is introduced in this work. Furthermore, to represent different magnetic interactions between collinear spin-polarized atoms, spin-dependent atom-centered symmetry functions are proposed taking the atomic spins into account. Applying these methods, a series of processes essential for battery applications of LixMn2O4 is investigated. For example, structural changes during charge and discharge as well as temperature induced changes as the orthorhombic to cubic transition are analyzed. Additionally, the lithium ion and electronic conduction are studied including an analysis of the charge ordering transition close to room temperature, which reduces the electrical conductivity. Further, the geometric and electronic structure of various LixMn2O4-water interfaces are characterized with regard to water dissociation and proton transfer processes that are important for catalysis of the oxygen evolution reaction. In addition, the performance of a magnetic HDNNP is demonstrated for manganese oxide, MnO, by the accurate prediction of the structural changes at the antiferromagnetic to paramagnetic phase transition.
Keywords: Theoretical Chemistry; Machine Learning; High-Dimensional Neural Network Potentials; Density Functional Theory; Atomistic Simulations; Battery Materials; Lithium Manganese Oxides; Solid-Liquid Interfaces; Phase Transitions; Jahn-Teller Effect; Oxidation States; Spin States; Magnetic Order; Electron Hopping; Ionic Conduction; Proton Transfer; Water Dissociation