Short-Range Δ-Machine Learning: A Cost-Efficient Strategy to Transfer Chemical Accuracy to Condensed Phase Systems DOI
Bence Balázs Mészáros,

András Szabó,

János Daru

и другие.

Journal of Chemical Theory and Computation, Год журнала: 2025, Номер unknown

Опубликована: Май 28, 2025

DFT-based machine-learning potentials (MLPs) are now routinely trained for condensed-phase systems, but surpassing DFT accuracy remains challenging due to the cost or unavailability of periodic reference calculations. Our previous work ( Phys. Rev. Lett. 2022, 129, 226001) demonstrated that high-accuracy MLPs can be within CCMD framework using extended yet finite Here, we introduce short-range Δ-Machine Learning (srΔML), a method starts from baseline MLP on low-level data and adds Δ-MLP correction based high-level cluster calculations at CC level. Applied liquid water, srΔML reduces required size (H2O)64 (H2O)15 significantly lowers number clusters needed, resulting in 50-200× reduction computational cost. The potential closely reproduces target accurately captures both two- three-body structural descriptors.

Язык: Английский

Fabrication of biomimetic giant waterlily cellulosic adsorption-catalytic material for efficient water purification DOI

Zhanlong Tan,

Mao Jun, Yuanyuan Hu

и другие.

Applied Catalysis B Environment and Energy, Год журнала: 2025, Номер 366, С. 125063 - 125063

Опубликована: Янв. 12, 2025

Язык: Английский

Процитировано

2
Modeling the impact of structure and coverage on the reactivity of realistic heterogeneous catalysts DOI Creative Commons
Benjamin W. J. Chen, Manos Mavrikakis

Nature Chemical Engineering, Год журнала: 2025, Номер unknown

Опубликована: Фев. 17, 2025

Язык: Английский

Процитировано

1
Δ-model correction of foundation model based on the model’s own understanding DOI
Mads-Peter V. Christiansen, Bjørk Hammer

The Journal of Chemical Physics, Год журнала: 2025, Номер 162(18)

Опубликована: Май 8, 2025

Foundation models of interatomic potentials, the so called universal may require fine-tuning or residual corrections when applied to specific subclasses materials. In present work, we demonstrate how such an augmentation can be accomplished via Δ-learning based on representation already embedded in potentials. The Δ-model introduced is a Gaussian Process Regression (GPR) model, and various types aggregation (global, species-separated, atomic) vector are discussed. Employing potential, CHGNet [Deng et al., Nat. Mach. Intell. 5, 1031 (2023)], global structure optimization setting, find that it correctly describes energetics “8” Cu oxide, which ultra-thin oxide film Cu(111). potential model even predicts more favorable compared with discussed recent density functional theory-based literature. Moving sulfur adatom overlayers Cu(111), Ag(111), Au(111), however, requires corrections. We these efficiently provided GPR-based formulated CHGNet’s own internal atomic embedding representation. need for tracked scarcity metal–sulfur environments materials project database trained on, leading overreliance sulfur–sulfur environments. Other potentials same data, MACE-MP0, SevenNet-0, ORB-v2-only-MPtrj, show similar behavior but varying degrees error, demonstrating general schemes models.

Язык: Английский

Процитировано

0
Accurate and efficient machine learning interatomic potentials for finite temperature modelling of molecular crystals DOI Creative Commons
Flaviano Della Pia, Benjamin X. Shi, Venkat Kapil

и другие.

Chemical Science, Год журнала: 2025, Номер unknown

Опубликована: Янв. 1, 2025

We fine-tune machine learning interatomic potentials to accurately model molecular crystals at finite temperature with the inclusion of nuclear quantum effects.

Язык: Английский

Процитировано

0
Short-Range Δ-Machine Learning: A Cost-Efficient Strategy to Transfer Chemical Accuracy to Condensed Phase Systems DOI
Bence Balázs Mészáros,

András Szabó,

János Daru

и другие.

Journal of Chemical Theory and Computation, Год журнала: 2025, Номер unknown

Опубликована: Май 28, 2025

DFT-based machine-learning potentials (MLPs) are now routinely trained for condensed-phase systems, but surpassing DFT accuracy remains challenging due to the cost or unavailability of periodic reference calculations. Our previous work ( Phys. Rev. Lett. 2022, 129, 226001) demonstrated that high-accuracy MLPs can be within CCMD framework using extended yet finite Here, we introduce short-range Δ-Machine Learning (srΔML), a method starts from baseline MLP on low-level data and adds Δ-MLP correction based high-level cluster calculations at CC level. Applied liquid water, srΔML reduces required size (H2O)64 (H2O)15 significantly lowers number clusters needed, resulting in 50-200× reduction computational cost. The potential closely reproduces target accurately captures both two- three-body structural descriptors.

Язык: Английский

Процитировано

0