Erhard, Linus Carl (2024)
Atomistic Modelling of Structure Formation and Phase Transitions in Si-Ox Compounds using Machine-Learning Interatomic Potentials.
Technische Universität Darmstadt
doi: 10.26083/tuprints-00028191
Dissertation, Erstveröffentlichung, Verlagsversion
Kurzbeschreibung (Abstract)
Silica is used in a wide range of applications from catalysis to construction to microelectronics. The related silicon monoxide is promising for applications as an anode material in lithium batteries. Although these materials have been extensively studied for more than a century, there are still many open questions. For example, the high-pressure transformations of silica are not fully understood. Moreover, in the case of silicon monoxide, there is not even an atomistic structure model that captures the complexity of the structure. In this work, we use atomistic modelling to investigate these problems. For this purpose, we developed several machine learning interatomic potentials (MLIP). First, we developed a Gaussian approximation potential (GAP) model based on a database with focus on bulk silica. Later, we switched to the atomic cluster expansion (ACE) framework. The final ACE potential is fitted to a more comprehensive training database labeled with energies and forces from strongly constrained and appropriately normed (SCAN) exchange-correlation density functional theory (DFT) data. The database covers a wide range of structures, including amorphous and crystalline silica, silica surfaces, high-pressure silica, and silicon-silica interfaces. Several approaches were used to build the database including ‘batch’ learning and active learning. Moreover, we present an active learning technique that extracts DFT feasible small-scale images from large-scale simulations (Chapter 3). The MLIPs are extensively tested in reproducing the thermodynamics of the systems and show excellent behavior, outperforming existing classical models. Nevertheless, to generate realistic amorphous structures of silica, we rely on a ‘hybrid’ protocol using a combination of our MLIP and a classical interatomic potential (Chapter 4). We apply the ACE potential to two cases. First, we study the high-pressure be- havior of amorphous silica and quartz under shock (Chapter 5). We find that there is an intermediate structure between the amorphous state and the crystalline stable state of stishovite. This phase is based on the defective nickel arsenide (d-NiAs) structure. The structure has a disordered silicon sublattice and an ordered hexagonal close-packed (HCP) oxygen sublattice. While the oxygen lattice appears to form fast on the molecular dynamics (MD) time scales, the ordering of the silicon and hence the formation of stishovite takes significantly longer. Moreover, we found that a direct transition between quartz and rosiaite-structured silica is also possible, which seems to require certain strain boundary conditions. Second, we generate structural models of silicon monoxide using melt-quench simulations (Chapter 6). These models show the same nanoscale segregation of silicon and silica as observed in experiment. Moreover, the energetics, grain sizes and X-ray structure factors of these models are in excellent agreement with the experiment. Using 20 ns annealing simulations, we are able to partially crystallize these structures and generate structural models with crystalline silicon in an amorphous silica matrix.
Typ des Eintrags: | Dissertation | ||||
---|---|---|---|---|---|
Erschienen: | 2024 | ||||
Autor(en): | Erhard, Linus Carl | ||||
Art des Eintrags: | Erstveröffentlichung | ||||
Titel: | Atomistic Modelling of Structure Formation and Phase Transitions in Si-Ox Compounds using Machine-Learning Interatomic Potentials | ||||
Sprache: | Englisch | ||||
Referenten: | Albe, Prof. Dr. Karsten ; Deringer, Prof. Dr. Volker | ||||
Publikationsjahr: | 8 Oktober 2024 | ||||
Ort: | Darmstadt | ||||
Kollation: | VIII, 126 Seiten | ||||
Datum der mündlichen Prüfung: | 2 September 2024 | ||||
DOI: | 10.26083/tuprints-00028191 | ||||
URL / URN: | https://tuprints.ulb.tu-darmstadt.de/28191 | ||||
Kurzbeschreibung (Abstract): | Silica is used in a wide range of applications from catalysis to construction to microelectronics. The related silicon monoxide is promising for applications as an anode material in lithium batteries. Although these materials have been extensively studied for more than a century, there are still many open questions. For example, the high-pressure transformations of silica are not fully understood. Moreover, in the case of silicon monoxide, there is not even an atomistic structure model that captures the complexity of the structure. In this work, we use atomistic modelling to investigate these problems. For this purpose, we developed several machine learning interatomic potentials (MLIP). First, we developed a Gaussian approximation potential (GAP) model based on a database with focus on bulk silica. Later, we switched to the atomic cluster expansion (ACE) framework. The final ACE potential is fitted to a more comprehensive training database labeled with energies and forces from strongly constrained and appropriately normed (SCAN) exchange-correlation density functional theory (DFT) data. The database covers a wide range of structures, including amorphous and crystalline silica, silica surfaces, high-pressure silica, and silicon-silica interfaces. Several approaches were used to build the database including ‘batch’ learning and active learning. Moreover, we present an active learning technique that extracts DFT feasible small-scale images from large-scale simulations (Chapter 3). The MLIPs are extensively tested in reproducing the thermodynamics of the systems and show excellent behavior, outperforming existing classical models. Nevertheless, to generate realistic amorphous structures of silica, we rely on a ‘hybrid’ protocol using a combination of our MLIP and a classical interatomic potential (Chapter 4). We apply the ACE potential to two cases. First, we study the high-pressure be- havior of amorphous silica and quartz under shock (Chapter 5). We find that there is an intermediate structure between the amorphous state and the crystalline stable state of stishovite. This phase is based on the defective nickel arsenide (d-NiAs) structure. The structure has a disordered silicon sublattice and an ordered hexagonal close-packed (HCP) oxygen sublattice. While the oxygen lattice appears to form fast on the molecular dynamics (MD) time scales, the ordering of the silicon and hence the formation of stishovite takes significantly longer. Moreover, we found that a direct transition between quartz and rosiaite-structured silica is also possible, which seems to require certain strain boundary conditions. Second, we generate structural models of silicon monoxide using melt-quench simulations (Chapter 6). These models show the same nanoscale segregation of silicon and silica as observed in experiment. Moreover, the energetics, grain sizes and X-ray structure factors of these models are in excellent agreement with the experiment. Using 20 ns annealing simulations, we are able to partially crystallize these structures and generate structural models with crystalline silicon in an amorphous silica matrix. |
||||
Alternatives oder übersetztes Abstract: |
|
||||
Status: | Verlagsversion | ||||
URN: | urn:nbn:de:tuda-tuprints-281919 | ||||
Sachgruppe der Dewey Dezimalklassifikatin (DDC): | 500 Naturwissenschaften und Mathematik > 530 Physik 500 Naturwissenschaften und Mathematik > 540 Chemie 500 Naturwissenschaften und Mathematik > 550 Geowissenschaften |
||||
Fachbereich(e)/-gebiet(e): | 11 Fachbereich Material- und Geowissenschaften 11 Fachbereich Material- und Geowissenschaften > Materialwissenschaft 11 Fachbereich Material- und Geowissenschaften > Materialwissenschaft > Fachgebiet Materialmodellierung |
||||
TU-Projekte: | PTJ|03XP0174A|FestBatt-Daten | ||||
Hinterlegungsdatum: | 08 Okt 2024 12:04 | ||||
Letzte Änderung: | 09 Okt 2024 08:17 | ||||
PPN: | |||||
Referenten: | Albe, Prof. Dr. Karsten ; Deringer, Prof. Dr. Volker | ||||
Datum der mündlichen Prüfung / Verteidigung / mdl. Prüfung: | 2 September 2024 | ||||
Export: | |||||
Suche nach Titel in: | TUfind oder in Google |
Frage zum Eintrag |
Optionen (nur für Redakteure)
Redaktionelle Details anzeigen |