NEBwalk: A transparent and calculator-agnostic framework for reproducible nudged elastic band workflows with machine-learned potentials and firstprinciples calculators
5 affiliations
- 1Department of Chemical Engineering
- 2Dhaka University of Engineering & Technology (DUET
- 3Department of Chemical and Geological Sciences
- 4University of Modena and Reggio Emilia (UNIMORE
- 5Department of Mechanical Engineering
Abstract
The nudged elastic band (NEB) method is established, but a reproducible transition-state calculation requires more than the NEB equations. Path construction, calculator isolation, restart behaviour, convergence monitoring, provenance, failure handling, and the transfer from inexpensive potentials to first-principles checks must remain consistent throughout the workflow. We developed NEBwalk as an open-source, calculator-agnostic Python framework for NEB and climbing-image NEB (CI-NEB) calculations built around the Atomic Simulation Environment (ASE). NEBwalk v0.11.0 supports improved-tangent NEB forces, variable springs, FIRE optimisation, minimum-image-aware periodic geometry, linear, IDPP, and regularized geodesic-style interpolation, restartable runs, structured result export, fresh calculator construction for each image, MLIP-assisted image selection, DFT-labelled data export, an optional MACE fine-tuning and evaluation handoff, resumable campaign infrastructure, reproducibility bundles, and conservative Quantum ESPRESSO recovery logic. A September 2026 audit re-examined 20 historical system-calculator cases rather than treating old documentation as automatically valid: 13 were re-audited and reconfirmed from surviving output artefacts, while 7 are retained only as historical provenance and excluded from the current quantitative validation claim. Ten artefact-backed vacancy-migration cases also have DOI-traceable literature comparators for the same element, crystal structure, and migration-hop class; because the numerical settings differ, we use these only as physical-scale checks. In two matched-physical-input implementation tests, NEBwalk and ASE differ by no more than 0.04 meV in the barrier and about 0.002 Å in the converged path geometry. A MACEMP-0 H/Cu(111) calculation gives a converged fcc-to-hcp barrier of about 0.126 eV, close to published DFT values of about 0.13-0.15 eV, although the literature itself disagrees on the sign of the small fcc-hcp site splitting. NEBwalk is therefore not presented as a new NEB algorithm or as a replacement for ASE or first-principles engines. Its role is to keep path optimisation, MLIP screening, DFT validation, reproducibility, and failure-aware execution in one inspectable workflow.
Related
- Effect of starting materials on microstructure and densification process of ZnGa2O4 transparent ceramics
- Effect of fluorine addition on the microstructure and emission properties of Ho:Y2O3 transparent ceramics
- Ferroelectric polymer-based flexible and transparent transducers for biomedical ultrasound and photoacoustic imaging.
- Yb:YAG Transparent Ceramics by Double‐Step Sintering—A Comparison Between HIP and FAST Postsintering Treatments
- Transparent, Conductive and Hydrophobic Cellulose Paper Fabricated by Photopolymerization
BibTeX
@misc{khandaker2026nebwalk,
title = {{NEBwalk: A transparent and calculator-agnostic framework for reproducible nudged elastic band workflows with machine-learned potentials and firstprinciples calculators}},
author = {Md. Rifat Khandaker and Ali Ghiami-Shomami and Mohammad Asaduzzaman Chowdhury},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010372/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010372/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - NEBwalk: A transparent and calculator-agnostic framework for reproducible nudged elastic band workflows with machine-learned potentials and firstprinciples calculators
AU - Md. Rifat Khandaker
AU - Ali Ghiami-Shomami
AU - Mohammad Asaduzzaman Chowdhury
PY - 2026
DA - 2026/10/13
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010372/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010372/v1
AB - The nudged elastic band (NEB) method is established, but a reproducible transition-state calculation requires more than the NEB equations. Path construction, calculator isolation, restart behaviour, convergence monitoring, provenance, failure handling, and the transfer from inexpensive potentials to first-principles checks must remain consistent throughout the workflow. We developed NEBwalk as an open-source, calculator-agnostic Python framework for NEB and climbing-image NEB (CI-NEB) calculations built around the Atomic Simulation Environment (ASE). NEBwalk v0.11.0 supports improved-tangent NEB forces, variable springs, FIRE optimisation, minimum-image-aware periodic geometry, linear, IDPP, and regularized geodesic-style interpolation, restartable runs, structured result export, fresh calculator construction for each image, MLIP-assisted image selection, DFT-labelled data export, an optional MACE fine-tuning and evaluation handoff, resumable campaign infrastructure, reproducibility bundles, and conservative Quantum ESPRESSO recovery logic. A September 2026 audit re-examined 20 historical system-calculator cases rather than treating old documentation as automatically valid: 13 were re-audited and reconfirmed from surviving output artefacts, while 7 are retained only as historical provenance and excluded from the current quantitative validation claim. Ten artefact-backed vacancy-migration cases also have DOI-traceable literature comparators for the same element, crystal structure, and migration-hop class; because the numerical settings differ, we use these only as physical-scale checks. In two matched-physical-input implementation tests, NEBwalk and ASE differ by no more than 0.04 meV in the barrier and about 0.002 Å in the converged path geometry. A MACEMP-0 H/Cu(111) calculation gives a converged fcc-to-hcp barrier of about 0.126 eV, close to published DFT values of about 0.13-0.15 eV, although the literature itself disagrees on the sign of the small fcc-hcp site splitting. NEBwalk is therefore not presented as a new NEB algorithm or as a replacement for ASE or first-principles engines. Its role is to keep path optimisation, MLIP screening, DFT validation, reproducibility, and failure-aware execution in one inspectable workflow.
ER -