Machine-learning interatomic potential for radiation damage and defects in tungsten

Physical review. B./Physical review. B, Journal Year: 2019, Volume and Issue: 100(14)

Published: Oct. 17, 2019

Tungsten will be used as a plasma-facing material in fusion power reactors, where the absorption of high-energy neutrons leads to permanent damage crystal structure. A comprehensive understanding atom-level tungsten has been limited by slowness quantum simulations and insufficient accuracy classical simulations. This study bridges gap between two developing machine-learning interatomic potential that allows simulation extreme environments with accuracy.

Language: Английский

---

Understanding X-ray Spectroscopy of Carbonaceous Materials by Combining Experiments, Density Functional Theory, and Machine Learning. Part I: Fingerprint Spectra

Chemistry of Materials, Journal Year: 2019, Volume and Issue: 31(22), P. 9243 - 9255

Published: Oct. 28, 2019

Carbonaceous materials, especially tetrahedral amorphous carbon (ta-C), can form complex functionalized surface structures and are thus promising candidates for applications in biomedical devices electrochemistry. Functional groups at ta-C surfaces have been widely studied by spectroscopic techniques; however, interpretation of the experimental data is extremely difficult, case X-ray photoelectron spectroscopy (XPS) absorption (XAS). The assignments XPS XAS signals normally based on references obtained from molecular or crystalline samples, which simplified approximations far more structures. Here, we use extensive density functional theory (DFT) simulations to predict signatures carbon-based materials realistic environments, building large sets structural models generated a machine-learning (ML) interatomic potential. results indicate clear signatures: individual fingerprint spectra distinctive binding energy distributions, both terms center broadness signal, chemically different groups. point out what kind information cannot be extracted with spectroscopy. This study will enable deeper physicochemical understanding ultimately theory-based identification quantification carbonaceous materials.

Language: Английский

---

Machine learning for predicting thermal transport properties of solids

Materials Science and Engineering R Reports, Journal Year: 2021, Volume and Issue: 146, P. 100642 - 100642

Published: Sept. 14, 2021

Language: Английский

---

Structure and Pore Size Distribution in Nanoporous Carbon

Chemistry of Materials, Journal Year: 2022, Volume and Issue: 34(2), P. 617 - 628

Published: Jan. 4, 2022

We study the structural and mechanical properties of nanoporous (NP) carbon materials by extensive atomistic machine-learning (ML) driven molecular dynamics (MD) simulations. To this end, we retrain a ML Gaussian approximation potential (GAP) for recalculating a-C database Deringer Csányi adding van der Waals interactions. Our GAP enables notable speedup improves accuracy energy force predictions. use to thoroughly structure pore-size distribution in computational NP samples. These samples are generated melt-graphitization-quench MD procedure over wide range densities (from 0.5 1.7 g/cm3) with structures containing 131 072 atoms. results good agreement experimental data available observables provide comprehensive account (radial angular functions, motif ring counts, X-ray diffraction patterns, pore characterization) (elastic moduli their evolution density) properties. show relatively narrow distributions, where peak position width distributions dictated mass density materials. allow further work on characterization materials, particular energy-storage applications, as well suggest future carbon-based

Language: Английский

---

Global optimization of atomic structure enhanced by machine learning

Physical review. B./Physical review. B, Journal Year: 2022, Volume and Issue: 105(24)

Published: June 9, 2022

Global optimization with first-principles energy expressions (GOFEE) is an efficient method for identifying low-energy structures in computationally expensive landscapes such as the ones described by density functional theory (DFT), van der Waals enabled DFT, or even methods beyond DFT. GOFEE evolutionary algorithm, that order to explore configuration space creates several candidates parallel. These are treated approximately using a machine learned surrogate model of energies and forces, trained on fly, eliminating need relaxations methods. Eventually, Bayesian statistics, chooses one candidate treats at full level. In this paper we elaborate importance use Gaussian kernel two length scales process regression model. We further role lower confidence bound relaxation selection structures. addition, present details sampling scheme obtaining parent evolution. Using learning clustering entire pool ever calculated, choosing most stable member from each cluster, ensures highly diverse sample plays population. The versatility demonstrated applying it identify gas-phase fullerene-type 24-atom carbon clusters dome-shaped 18-atom supported Ir(111).

Language: Английский

---