{"id":{"repo_id":"tu-berlin","oai_identifier":"oai:depositonce.tu-berlin.de:11303/25622"},"canonical_url":"https://search.dev.ndltd.org/etd/tu-berlin/oai:depositonce.tu-berlin.de:11303/25622","repository":{"repo_id":"tu-berlin","name":"Technische Universität Berlin","base_url":"https://api-depositonce.tu-berlin.de/server/oai/request"},"display":{"title":"Modeling larger length and time scales in machine learning force fields","abstract":"Molecular dynamics simulations describe the collective motion of atoms over time, enabling the study of fundamental processes across physics, chemistry, and biology. This has established molecular dynamics as a cornerstone of modern science and improving its accuracy has a long standing history in the computational sciences. The accuracy of a molecular dynamics simulation is determined by the accuracy of the interatomic forces. These forces can be obtained from fast but approximate empirical force fields, or slow but accurate quantum mechanical methods. Machine learning force fields have emerged as a promising alternative, bridging this gap by approximating (\"learning\") the interatomic forces from quantum mechanical reference data. Despite great advances over the past years, many challenges remain on the path towards generally applicable machine learning force fields. A key limitation is the accurate treatment of long time scales and large length scales. However, both are crucial for the correct prediction of experimental outcomes, which is the ultimate measure of predictive usefulness. In this thesis, we address some challenges that are related to the modeling of long time scales and large length scales within machine learning force fields. We propose a deep neural network architecture, that reduces the computational cost while maintaining accuracy and reliability. This is accomplished by proposing a new design space for geometric neural network architectures, extending the time scales that are accessible in machine learning force field simulations. As a next step, we combine this model with physically inspired long-range corrections, enabling a meaningful description of large length scales. This approach allows simulating time and length scales relevant for biomolecular simulations, further demonstrating agreement with experimental data. Recognizing that current approaches often resort to physically inspired post-hoc correction for long-range interactions, we also explore a complementary direction: Learning all interactions without physically derived correction terms but in a purely data-driven way. We propose a computationally efficient algorithm capable of learning short- and long-range interactions with equal accuracy. This paves the way for machine learning models, that can simultaneously learn across all relevant length scales in atomistic systems.","abstract_html":"Molecular dynamics simulations describe the collective motion of atoms over time, enabling the study of fundamental processes across physics, chemistry, and biology. This has established molecular dynamics as a cornerstone of modern science and improving its accuracy has a long standing history in the computational sciences. The accuracy of a molecular dynamics simulation is determined by the accuracy of the interatomic forces. These forces can be obtained from fast but approximate empirical force fields, or slow but accurate quantum mechanical methods. Machine learning force fields have emerged as a promising alternative, bridging this gap by approximating ("learning") the interatomic forces from quantum mechanical reference data. Despite great advances over the past years, many challenges remain on the path towards generally applicable machine learning force fields. A key limitation is the accurate treatment of long time scales and large length scales. However, both are crucial for the correct prediction of experimental outcomes, which is the ultimate measure of predictive usefulness. In this thesis, we address some challenges that are related to the modeling of long time scales and large length scales within machine learning force fields. We propose a deep neural network architecture, that reduces the computational cost while maintaining accuracy and reliability. This is accomplished by proposing a new design space for geometric neural network architectures, extending the time scales that are accessible in machine learning force field simulations. As a next step, we combine this model with physically inspired long-range corrections, enabling a meaningful description of large length scales. This approach allows simulating time and length scales relevant for biomolecular simulations, further demonstrating agreement with experimental data. Recognizing that current approaches often resort to physically inspired post-hoc correction for long-range interactions, we also explore a complementary direction: Learning all interactions without physically derived correction terms but in a purely data-driven way. We propose a computationally efficient algorithm capable of learning short- and long-range interactions with equal accuracy. This paves the way for machine learning models, that can simultaneously learn across all relevant length scales in atomistic systems.","abstract_has_math":false,"creators":["Frank, Thorben"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Müller, Klaus-Robert"],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025","date_published":"2025","updated_at":"2026-07-27T21:28:29Z","subjects":[],"languages":["en"],"rights":[],"rights_urls":["https://creativecommons.org/licenses/by-sa/4.0/"],"identifier_entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://doi.org/10.14279/depositonce-24446"],"render_values":[{"text":"https://doi.org/10.14279/depositonce-24446","href":"https://doi.org/10.14279/depositonce-24446","code":true}]}]},"links":{"outbound_url":"https://depositonce.tu-berlin.de/handle/11303/25622","outbound_label":"Repository record","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Müller, Klaus-Robert"]},{"key":"dc:creator","label":"Author","values":["Frank, Thorben"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-09-26T08:46:14Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2025-09-26T08:46:14Z"]},{"key":"dc:date.issued","label":"Date","values":["2025"]},{"key":"dc:type","label":"Dc Type","values":["Doctoral Thesis"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]},{"key":"dc:rights.uri","label":"Rights URI","values":["https://creativecommons.org/licenses/by-sa/4.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://depositonce.tu-berlin.de/handle/11303/25622","https://doi.org/10.14279/depositonce-24446"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Molecular dynamics simulations describe the collective motion of atoms over time, enabling the study of fundamental processes across physics, chemistry, and biology. This has established molecular dynamics as a cornerstone of modern science and improving its accuracy has a long standing history in the computational sciences. The accuracy of a molecular dynamics simulation is determined by the accuracy of the interatomic forces. These forces can be obtained from fast but approximate empirical force fields, or slow but accurate quantum mechanical methods. Machine learning force fields have emerged as a promising alternative, bridging this gap by approximating (\"learning\") the interatomic forces from quantum mechanical reference data. Despite great advances over the past years, many challenges remain on the path towards generally applicable machine learning force fields. A key limitation is the accurate treatment of long time scales and large length scales. However, both are crucial for the correct prediction of experimental outcomes, which is the ultimate measure of predictive usefulness. In this thesis, we address some challenges that are related to the modeling of long time scales and large length scales within machine learning force fields. We propose a deep neural network architecture, that reduces the computational cost while maintaining accuracy and reliability. This is accomplished by proposing a new design space for geometric neural network architectures, extending the time scales that are accessible in machine learning force field simulations. As a next step, we combine this model with physically inspired long-range corrections, enabling a meaningful description of large length scales. This approach allows simulating time and length scales relevant for biomolecular simulations, further demonstrating agreement with experimental data. Recognizing that current approaches often resort to physically inspired post-hoc correction for long-range interactions, we also explore a complementary direction: Learning all interactions without physically derived correction terms but in a purely data-driven way. We propose a computationally efficient algorithm capable of learning short- and long-range interactions with equal accuracy. This paves the way for machine learning models, that can simultaneously learn across all relevant length scales in atomistic systems.","Molekulardynamik-Simulationen ermöglichen die Untersuchung fundamentaler physikalischer, chemischer und biologischer Prozesse aufgrund ihrer Fähigkeit, die kollektive Bewegung von Atomen über die Zeit zu beschreiben. Daher haben sie sich zu einem wichtigen Bestandteil der modernen Naturwissenschaft entwickelt und ihre Weiterentwicklung ist ein zentraler Forschungsschwerpunkt. Die Genauigkeit der Simulationen wird dabei maßgeblich durch die Genauigkeit der interatomaren Kräfte bestimmt. Lange wurden diese Kräfte mit Hilfe von schnellen aber approximativen empirischen Kraftfeldern oder durch den Einsatz von langsamen aber genauen quantenmechanischen Methoden berechnet. Erst in der jüngeren Vergangenheit haben sich maschinell gelernte Kraftfelder zu einer vielversprechenden Alternative entwickelt, da sie in der Lage sind, die Genauigkeit von quantenmechanischen Methoden mit der Geschwindigkeit von empirischen Kraftfeldern zu kombinieren. Trotz großer Fortschritte in den vergangenen Jahren verbleiben viele Herausforderungen auf dem Weg zu universell einsetzbaren, maschinell gelernten Kraftfeldern. Eine wesentliche Einschränkung ist dabei die zuverlässige Beschreibung von langen Zeitskalen und großen Längenskalen. Da diese eine tragende Rolle für die präzise Vorhersage von experimentellen Ergebnissen spielen, ist ihre korrekte Beschreibung von zentraler Bedeutung. Diese Arbeit widmet sich einigen der Herausforderungen, die im Kontext von maschinell gelernten Kraftfelden auftreten, wenn lange Zeitskalen und große Längenskalen modelliert werden sollen. Wir beginnen mit der Entwicklung einer tiefen neuronalen Netz-Architektur, welche die Rechenkosten senkt und gleichzeitig die Genauigkeit und Zuverlässigkeit aufrechterhält. Zeitskalen, die mit maschinell gelernten Kraftfeldern simuliert werden können, werden dadurch erweitert. Wir kombinieren die entwickelte Netz-Architektur mit physikalischen Korrekturtermen für langreichweitige Interaktionen und ermöglichen dadurch eine sinnvolle Beschreibung großer Längenskalen. Diese Kombination erlaubt biomolekulare Simulationen auf relevanten Zeit- und Längenskalen. Darüber hinaus untersuchen wir einen diametralen Ansatz: Das Lernen von langreichweitigen Interaktionen in einer datengetriebenen Art und Weise; ohne die Zuhilfenahme von physikalischen Korrekturtermen. Um dies zu realisieren, entwickeln wir einen kosteneffizienten Algorithmus, der in der Lage ist, sowohl kurz- als auch langreichweitige Wechselwirkungen zu modellieren. Dies ebnet den Weg für maschinelle Lernmodelle, die in der Lage sind, alle relevanten Längenskalen lernen können."]},{"key":"dc:title","label":"Title","values":["Modeling larger length and time scales in machine learning force fields"]}]}],"canonical_facts":{"dc:contributor.advisor":["Müller, Klaus-Robert"],"dc:creator":["Frank, Thorben"],"dc:date.accessioned":["2025-09-26T08:46:14Z"],"dc:date.available":["2025-09-26T08:46:14Z"],"dc:date.issued":["2025"],"dc:description.abstract":["Molecular dynamics simulations describe the collective motion of atoms over time, enabling the study of fundamental processes across physics, chemistry, and biology. This has established molecular dynamics as a cornerstone of modern science and improving its accuracy has a long standing history in the computational sciences. The accuracy of a molecular dynamics simulation is determined by the accuracy of the interatomic forces. These forces can be obtained from fast but approximate empirical force fields, or slow but accurate quantum mechanical methods. Machine learning force fields have emerged as a promising alternative, bridging this gap by approximating (\"learning\") the interatomic forces from quantum mechanical reference data. Despite great advances over the past years, many challenges remain on the path towards generally applicable machine learning force fields. A key limitation is the accurate treatment of long time scales and large length scales. However, both are crucial for the correct prediction of experimental outcomes, which is the ultimate measure of predictive usefulness. In this thesis, we address some challenges that are related to the modeling of long time scales and large length scales within machine learning force fields. We propose a deep neural network architecture, that reduces the computational cost while maintaining accuracy and reliability. This is accomplished by proposing a new design space for geometric neural network architectures, extending the time scales that are accessible in machine learning force field simulations. As a next step, we combine this model with physically inspired long-range corrections, enabling a meaningful description of large length scales. This approach allows simulating time and length scales relevant for biomolecular simulations, further demonstrating agreement with experimental data. Recognizing that current approaches often resort to physically inspired post-hoc correction for long-range interactions, we also explore a complementary direction: Learning all interactions without physically derived correction terms but in a purely data-driven way. We propose a computationally efficient algorithm capable of learning short- and long-range interactions with equal accuracy. This paves the way for machine learning models, that can simultaneously learn across all relevant length scales in atomistic systems.","Molekulardynamik-Simulationen ermöglichen die Untersuchung fundamentaler physikalischer, chemischer und biologischer Prozesse aufgrund ihrer Fähigkeit, die kollektive Bewegung von Atomen über die Zeit zu beschreiben. Daher haben sie sich zu einem wichtigen Bestandteil der modernen Naturwissenschaft entwickelt und ihre Weiterentwicklung ist ein zentraler Forschungsschwerpunkt. Die Genauigkeit der Simulationen wird dabei maßgeblich durch die Genauigkeit der interatomaren Kräfte bestimmt. Lange wurden diese Kräfte mit Hilfe von schnellen aber approximativen empirischen Kraftfeldern oder durch den Einsatz von langsamen aber genauen quantenmechanischen Methoden berechnet. Erst in der jüngeren Vergangenheit haben sich maschinell gelernte Kraftfelder zu einer vielversprechenden Alternative entwickelt, da sie in der Lage sind, die Genauigkeit von quantenmechanischen Methoden mit der Geschwindigkeit von empirischen Kraftfeldern zu kombinieren. Trotz großer Fortschritte in den vergangenen Jahren verbleiben viele Herausforderungen auf dem Weg zu universell einsetzbaren, maschinell gelernten Kraftfeldern. Eine wesentliche Einschränkung ist dabei die zuverlässige Beschreibung von langen Zeitskalen und großen Längenskalen. Da diese eine tragende Rolle für die präzise Vorhersage von experimentellen Ergebnissen spielen, ist ihre korrekte Beschreibung von zentraler Bedeutung. Diese Arbeit widmet sich einigen der Herausforderungen, die im Kontext von maschinell gelernten Kraftfelden auftreten, wenn lange Zeitskalen und große Längenskalen modelliert werden sollen. Wir beginnen mit der Entwicklung einer tiefen neuronalen Netz-Architektur, welche die Rechenkosten senkt und gleichzeitig die Genauigkeit und Zuverlässigkeit aufrechterhält. Zeitskalen, die mit maschinell gelernten Kraftfeldern simuliert werden können, werden dadurch erweitert. Wir kombinieren die entwickelte Netz-Architektur mit physikalischen Korrekturtermen für langreichweitige Interaktionen und ermöglichen dadurch eine sinnvolle Beschreibung großer Längenskalen. Diese Kombination erlaubt biomolekulare Simulationen auf relevanten Zeit- und Längenskalen. Darüber hinaus untersuchen wir einen diametralen Ansatz: Das Lernen von langreichweitigen Interaktionen in einer datengetriebenen Art und Weise; ohne die Zuhilfenahme von physikalischen Korrekturtermen. Um dies zu realisieren, entwickeln wir einen kosteneffizienten Algorithmus, der in der Lage ist, sowohl kurz- als auch langreichweitige Wechselwirkungen zu modellieren. Dies ebnet den Weg für maschinelle Lernmodelle, die in der Lage sind, alle relevanten Längenskalen lernen können."],"dc:identifier.uri":["https://depositonce.tu-berlin.de/handle/11303/25622","https://doi.org/10.14279/depositonce-24446"],"dc:language.iso":["en"],"dc:rights.uri":["https://creativecommons.org/licenses/by-sa/4.0/"],"dc:title":["Modeling larger length and time scales in machine learning force fields"],"dc:type":["Doctoral Thesis"]},"updated_at":"2026-07-27T21:28:29Z"}