Preliminary

Clone the GitHub repository:

git clone git@github.com:HaoZeke/nebmmf_repro.git

About

Contains the reproduction details for the publication on the performance and logs for the RONEB algorithm / OCI-NEB for seeking accelerated transition state.

Reference

If you use this repository or its parts please cite the corresponding publication or data source.

Preprint

1. Goswami, M. Gunde, and H. Jónsson, “Enhanced climbing image nudged elastic band method with hessian eigenmode alignment,” Jan. 22, 2026, arXiv: arXiv:2601.12630. doi: 10.48550/arXiv.2601.12630.

Replication data

Users must inflate the archives from the MaterialsCloud record into the repository structure. Place the following archives into the root or the designated subdirectories before execution:

Archive Destination Description

data.tar.xz data/ Stan models, Parquet data, and CSV benchmarks. case_baker_results.tar.xz case_studies/molecular_systems Parameter ablation results for the Baker benchmark set. optbench_results.tar.xz case_studies/optbench Platinum heptamer benchmark for surface systems from OptBench. case_static_results.tar.xz case_studies/static_switchover Baker test set, one-shot NEB followed by Dimer. eonRuns_results.tar.xz eonRuns/ Full NEB/RONEB trajectory logs and plots for the Baker.

Inflate the results archive using tar, e.g.

tar -xvf eonRuns_results.tar.xz -C eonRuns/

Reusing models

The repository offers both F.A.I.R.-compliant artifacts and native R objects for Bayesian analysis via brms.

1. From R

   The data/models/ directory stores serialized .rds objects, readable with readRDS. These require the brms library for further analysis:

   library('brms')
   model <- readRDS("data/models/brms_efficiency_scaling_v5.rds")
   

   More helper functions for generating and using these models and predictions are in the Github repository.

   The repository employs pixi for dependency management and organizes content by function:

   Directory Description

   case_studies/ Specific molecular systems and optbench configurations. data/ Raw benchmarks (.csv), Stan models, and Parquet data. docs/ Documentation, including .org files for visualization and model definitions. eonRuns/ Workflow configurations and results for the eOn runs. scripts/ Python utilities for generating YAML configs and parsing simulation outputs.

   The results directories (eonRuns/results) implement the following hierarchy:

   • ****00~models~****: Contains the PyTorch (.pt) Machine Learning Potentials used for the PES evaluations.
   • ****01~endpoints~****: Atomic coordinates for the reactant and product states for each system (e.g., 01_hcn, 09_parentdielsalder).
   • ****02~idpp~****: Initial guesses for the reaction paths.
   • ****03~neb~****: Optimized transition state paths using the RONEB and standard NEB algorithms.
   • ****04~plots~****: Resultant visualizations and energy profiles for each chemical system.