Computational chemistry
NN potentials map gold–water ORR paths
Open access · cc by · source: Europe PMC
Equivariant neural-network potentials drive nanosecond metadynamics showing associative ORR on Au(100) in which *OOH is reduced to two *OH, with a ~0.3 eV barrier.
Study at a glance
- Design
- Computational / modelling — PaiNN active-learning NNPs enabling path-CV metadynamics of ORR on Au(100)–water
- N
- 2.5 ns production metadynamics after AIMD active learning — no sample N
- Population
- Au(100)–water models with O2/OH adsorbates
- Outcome
- Associative ORR pathway and ~0.3 eV O2-to-hydroxyl barrier
Structured fields used in claim comparison tables when every cited study has a complete layer.
Key findings
ORR follows an associative path: O2 is hydrogenated toward *OOH, which is reduced to two *OH with neighboring water rather than splitting into *OH + *O. The O2-to-hydroxyl barrier is about 0.3 eV, matching gold’s high experimental ORR activity. Active learning continued until MD could run beyond 5 ns without large force uncertainty.
Methodology
They trained a PaiNN ensemble on DFT (PBE-D3) data selected by CUR active learning from AIMD of Au(100)–water with O2 and OH adsorbates, then ran 2.5 ns path-CV metadynamics in water at 350 K. Path collective variables tracked O–O and O–H coordination so the reduction path did not have to be guessed in advance.
Limitations
The model is Au(100) in pure water without explicit cations or applied potential as in alkaline ORR cells. Facet dependence is left for future work. The 0.3 eV barrier is from the NNP/MetaD landscape, not a measured Tafel slope on the same electrode.
How this study connects
Role on claims
Each row is a claim on a concept or method page where this paper supports, challenges, or qualifies the statement. Roles are hand-checked — not a model guess.
Learned potentials trained on DFT can extend the timescale of a mechanism study. A PaiNN ensemble on PBE-D3 Au(100)–water data plus metadynamics gave an associative ORR path with an O₂-to-hydroxyl barrier of about 0.3 eV — a landscape from the neural-network potential, not a measured Tafel slope, and without explicit cations or applied potential.
Evidence for the claim as stated.
These papers do not agree on what DFT is for. One trains a statistical model to correct DFT barriers; one trains a net to emulate DFT spin states; others use a single-point or periodic calculation to rationalise a crystal, a gas-phase ion, or a ligand redox event. Quoting a kcal mol⁻¹ figure without saying whether it is raw DFT, a GP correction, or an NNP barrier mixes those jobs.
Evidence for the claim as stated.
A learned potential can push DFT-quality forces into nanosecond metadynamics. A PaiNN ensemble trained on PBE-D3 Au(100)–water data, then 2.5 ns path-CV metadynamics at 350 K, finds associative ORR: O₂ is hydrogenated toward *OOH, which is reduced to two *OH with neighbouring water rather than splitting into *OH + *O. The O₂-to-hydroxyl barrier is about 0.3 eV. The model is Au(100) in pure water without explicit cations or applied potential; the barrier is from the NNP landscape, not a measured Tafel slope.
Evidence for the claim as stated.
A short classical MD on a crystal pose can fail at ranking, while a neural-network AIMD-quality trajectory can propose a path. MM-PBSA could not rank lysozyme Lys-Me₂ sites; the Au(100) NNP metadynamics did propose an associative 0.3 eV ORR path — still without cations or bias. Success is Hamiltonian- and question-specific, not a property of 'having run MD'.
Evidence for the claim as stated.
Open questions
Tensions this paper is part of
From concept pages' “where studies disagree.” Disagreement means the same question; scope means different assays, populations, or outcomes.
These papers do not agree on what DFT is for. One trains a statistical model to correct DFT barriers; one trains a net to emulate DFT spin states; others use a single-point or periodic calculation to rationalise a crystal, a gas-phase ion, or a ligand redox event. Quoting a kcal mol⁻¹ figure without saying whether it is raw DFT, a GP correction, or an NNP barrier mixes those jobs.
A short classical MD on a crystal pose can fail at ranking, while a neural-network AIMD-quality trajectory can propose a path. MM-PBSA could not rank lysozyme Lys-Me₂ sites; the Au(100) NNP metadynamics did propose an associative 0.3 eV ORR path — still without cations or bias. Success is Hamiltonian- and question-specific, not a property of 'having run MD'.
- Supports · Calixarene docks dimethyllysine on lysozyme
Related papers in this topic
Same topic cluster — not a recommendation engine.