Reaction Neb Analysis

• estimate_neb_image_count
• make_neb_geometry
• vasp_neb_prepare
• vasp_execute_batch
• analyze_vasp_neb_results

Workflow

1. Keep geometry and input assembly separate

• Use make_neb_geometry only for the interpolation/image tree.
• Prefer estimate_neb_image_count before setting n_images so the count follows the actual periodic endpoint displacement rather than intuition.
• Use vasp_neb_prepare only for NEB-ready VASP support files and protected INCAR controls.
• The initial and final endpoints must keep the same atom ordering; fix the structures first instead of interpolating between reordered atoms.
• A single NEB should describe one local elementary event. Do not use one band for long-distance diffusion across several equivalent sites.
• If the apparent reaction is really a long migration, remodel the final state onto the nearest equivalent site and compute that primitive hop instead.
• When no prior image-count estimate exists, use the practical first guess n_images ≈ ceil(sqrt(sum_i ||Δr_i||^2) / 0.8 Å) for the intermediate-image count, after the endpoint matching/PBC convention is fixed.
• Under periodic boundary conditions, compute Δr_i with the same minimum-image convention (mic=true) that will be used for interpolation. Do not infer the path length from wrapped endpoint coordinates by eye.
• For routine local events, prefer about 4-8 intermediate images unless the displacement heuristic points elsewhere.
• If that estimate exceeds about 8 intermediate images, treat it as a warning that the chosen endpoints may encode an unrelated long-range path rather than one local barrier event.
• If the periodic root-sum-squared displacement exceeds about 6 Å, treat that as a warning that the band is probably too long and should be remodeled into shorter primitive hops.

2. Run one pathway contract

• Keep endpoint method settings compatible with the pathway run.
• Do not mix image-count changes and INCAR changes inside the same comparison.
• Keep the chosen image count and climbing-image policy explicit in the run record.
• Prefer a two-stage NEB contract by default: first run plain NEB with climbing image disabled for coarse convergence, then rerun from the coarse-converged images with CI-NEB enabled for barrier refinement.
• If the workflow intentionally skips the coarse plain-NEB stage, say so explicitly instead of implying the standard two-stage pattern was followed.

3. Require barrier evidence

• After dispatch, the workflow is not complete until analyze_vasp_neb_results exports the barrier summary, CSV profile, and profile plot.
• If image energies are missing or partial, report that explicitly instead of inventing a barrier.
• If the profile shows wave-like repeated rises and falls instead of one dominant saddle region, treat that as evidence that the chosen path may pass through hidden metastable intermediates.
• Use those suspect local minima images as clues for rebuilding the problem into shorter primitive hops instead of insisting on one long NEB.

4. Keep the energy reference explicit

• State whether the reported barrier is forward, reverse, or both.
• Unless the task says otherwise, treat the forward barrier as referenced to the initial endpoint energy and the reverse barrier as referenced to the final endpoint.
• If the endpoints were evaluated under a different footing than the NEB images, say so explicitly instead of implying a fully self-consistent barrier contract.

Method-critical defaults

• Treat endpoint validation as part of the barrier contract; a bad endpoint pair gives a bad pathway no matter how clean the dispatch is.
• NEB-critical INCAR keys should stay under the wrapper unless the task explicitly needs a controlled override.
• Default to plain-NEB -> CI-NEB rather than starting directly from CI-NEB; direct climbing-image starts need an explicit reason because they are often harder to converge robustly.
• If the image count is not obvious, prefer the displacement-norm heuristic above over a blind fixed default.
• If the heuristic asks for more than about 8 intermediate images, reconsider whether the endpoints should be decomposed into a shorter primitive step before running NEB.
• If the converged or partially converged profile shows a wave-like multi-bump pattern, treat that as a strong hint that undiscovered intermediate states may exist and that the path should be split.
• Do not present a barrier as final if image energies are partial, the climbing-image logic is unclear, or the image set never produced a credible saddle region.
• Keep the barrier convention explicit in the final answer; “the barrier” is ambiguous otherwise.

Output Contract

Return:

• NEB root path
• image count
• execution state path
• NEB summary JSON, CSV, and plot path
• stated barrier convention (forward/reverse reference)

References

• Use transition-state-neb when only the primitive NEB setup side is needed.