Details
Originalsprache | Englisch |
---|---|
Seiten (von - bis) | 4461-4473 |
Seitenumfang | 13 |
Fachzeitschrift | Journal of Chemical Theory and Computation |
Jahrgang | 19 |
Ausgabenummer | 14 |
Frühes Online-Datum | 13 Apr. 2023 |
Publikationsstatus | Veröffentlicht - 25 Juli 2023 |
Extern publiziert | Ja |
Abstract
Nanodiamonds have a wide range of applications including catalysis, sensing, tribology, and biomedicine. To leverage nanodiamond design via machine learning, we introduce the new data set ND5k, consisting of 5089 diamondoid and nanodiamond structures and their frontier orbital energies. ND5k structures are optimized via tight-binding density functional theory (DFTB) and their frontier orbital energies are computed using density functional theory (DFT) with the PBE0 hybrid functional. From this data set we derive a qualitative design suggestion for nanodiamonds in photocatalysis. We also compare recent machine learning models for predicting frontier orbital energies for similar structures as they have been trained on (interpolation on ND5k), and we test their abilities to extrapolate predictions to larger structures. For both the interpolation and extrapolation task, we find the best performance using the equivariant message passing neural network PaiNN. The second best results are achieved with a message passing neural network using a tailored set of atomic descriptors proposed here.
ASJC Scopus Sachgebiete
- Informatik (insg.)
- Angewandte Informatik
- Chemie (insg.)
- Physikalische und Theoretische Chemie
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
in: Journal of Chemical Theory and Computation, Jahrgang 19, Nr. 14, 25.07.2023, S. 4461-4473.
Publikation: Beitrag in Fachzeitschrift › Artikel › Forschung › Peer-Review
}
TY - JOUR
T1 - Machine Learning Frontier Orbital Energies of Nanodiamonds
AU - Kirschbaum, Thorren
AU - von Seggern, Börries
AU - Dzubiella, Joachim
AU - Bande, Annika
AU - Noé, Frank
N1 - Funding Information: We thank Prof. Gabriel Bester for providing preoptimized structures of nanodiamonds and Prof. Philipp Marquetand, Dr. Félix Musil, and Dr. Kristof T. Schütt for helpful discussions. T.K., J.D., and F.N. acknowledge support from the Helmholtz Einstein International Berlin Research School in Data Science (HEIBRiDS). F.N. acknowledges support from European Commission (ERC CoG 772230), The Berlin Mathematics center MATH+ (AA2-8), and the Berlin Institute for the Foundations of Learning and Data (BIFOLD). Computing resources were kindly provided by the Freie Universität Berlin hpc cluster Curta and by the Helmholtz-Zentrum Dresden-Rossendorf.
PY - 2023/7/25
Y1 - 2023/7/25
N2 - Nanodiamonds have a wide range of applications including catalysis, sensing, tribology, and biomedicine. To leverage nanodiamond design via machine learning, we introduce the new data set ND5k, consisting of 5089 diamondoid and nanodiamond structures and their frontier orbital energies. ND5k structures are optimized via tight-binding density functional theory (DFTB) and their frontier orbital energies are computed using density functional theory (DFT) with the PBE0 hybrid functional. From this data set we derive a qualitative design suggestion for nanodiamonds in photocatalysis. We also compare recent machine learning models for predicting frontier orbital energies for similar structures as they have been trained on (interpolation on ND5k), and we test their abilities to extrapolate predictions to larger structures. For both the interpolation and extrapolation task, we find the best performance using the equivariant message passing neural network PaiNN. The second best results are achieved with a message passing neural network using a tailored set of atomic descriptors proposed here.
AB - Nanodiamonds have a wide range of applications including catalysis, sensing, tribology, and biomedicine. To leverage nanodiamond design via machine learning, we introduce the new data set ND5k, consisting of 5089 diamondoid and nanodiamond structures and their frontier orbital energies. ND5k structures are optimized via tight-binding density functional theory (DFTB) and their frontier orbital energies are computed using density functional theory (DFT) with the PBE0 hybrid functional. From this data set we derive a qualitative design suggestion for nanodiamonds in photocatalysis. We also compare recent machine learning models for predicting frontier orbital energies for similar structures as they have been trained on (interpolation on ND5k), and we test their abilities to extrapolate predictions to larger structures. For both the interpolation and extrapolation task, we find the best performance using the equivariant message passing neural network PaiNN. The second best results are achieved with a message passing neural network using a tailored set of atomic descriptors proposed here.
UR - http://www.scopus.com/inward/record.url?scp=85154070317&partnerID=8YFLogxK
U2 - 10.48550/arXiv.2210.07930
DO - 10.48550/arXiv.2210.07930
M3 - Article
C2 - 37053438
AN - SCOPUS:85154070317
VL - 19
SP - 4461
EP - 4473
JO - Journal of Chemical Theory and Computation
JF - Journal of Chemical Theory and Computation
SN - 1549-9618
IS - 14
ER -