Fix NEB IS/FS independent wrapping producing spurious atom "jumps"

[Azeezakinyemi999]

build_neb_images() and write_ase_neb_script()'s template (the one that generates every real run_neb.py) wrapped IS and FS independently before NEB.interpolate(). An atom whose real relaxation crosses a cell boundary in one endpoint but not the other got mapped to opposite sides of the cell, and NEB.interpolate() (raw Cartesian, no periodic awareness) treated that as a literal path to traverse.

Confirmed on the real reference pair (Hastelloy_N_1234_supercell, s_145__s_73+s_144): 3 metal atoms showed ~12.6 Å naive IS->FS displacement (essentially the full 12.619 Å cell x-length) vs ~0.01 Å true (minimum-image) displacement -- almost certainly a major contributor to that pair's abnormally high starting fmax (1118 eV/A) and the slow convergence fought all session.

Fix: align FS to IS's frame via find_mic() (minimum-image convention) instead of wrapping independently. IS still gets a normal .wrap() as the anchor frame; FS is repositioned to the periodic image of each atom closest to its IS counterpart. H atoms' own displacement is unaffected (verified identical naive vs MIC on the real pair), so this only corrects the spuriously-wrapped metal atoms.

Hardening added on top of the core fix:

• pbc=(True, True, False) passed to find_mic (not to is_raw/fs_raw's own .pbc, which is left untouched to avoid changing calculator behaviour elsewhere in the NEB run) -- these are slabs with a vacuum gap, not periodic in z; a blanket pbc=True would let find_mic treat a real z-motion (e.g. toward desorption) as if it wrapped through the vacuum.
• Half-cell-length warning (find_mic cannot distinguish a genuine displacement > L/2 from a shorter wrap-around candidate), gated by the same pbc mask so an unambiguous large z-motion doesn't false-positive.
• IS/FS cell-mismatch guard (find_mic uses is_raw.cell only) raises RuntimeError before any silent misuse.

Verified via 13 new synthetic tests (numeric reproduction of the real bug plus each hardening case) and content-assertion tests on the generated template, plus re-running the real reference pair after each change (identical result: 3 flagged atoms, same magnitudes, zero warnings) to confirm the hardening is additive, not a behaviour change. Full suite: 1585/1585 passing.

Separately identified but deliberately out of scope for this fix: is_raw/fs_raw's own .pbc is left at ASE's lammps-data reader default (True, True, True), meaning MACE's periodic neighbour search also treats z as periodic during actual NEB force evaluations. With r_max=6-8 A and MACE's standard 2-layer architecture, the effective receptive field (12-16 A) is close to or may exceed the configured 15 A vacuum gap -- a real, separate, force/energy-level concern (distinct from this purely geometric interpolation fix) to be investigated on its own branch.

[Copilot]

Pull request overview

This PR fixes a NEB path construction artifact caused by independently wrapping the initial state (IS) and final state (FS) before interpolation. It realigns FS into IS’s periodic frame using ASE’s minimum-image convention (find_mic), preventing spurious near–cell-length “jumps” in Cartesian interpolation, and adds guards/warnings to make misuse more detectable.

Changes:

• Update build_neb_images() and the generated run_neb.py template to align FS to IS using find_mic(..., pbc=(True, True, False)) instead of calling fs_raw.wrap().
• Add hardening: IS/FS cell mismatch check (raises) and a half-cell displacement ambiguity warning (periodic axes only).
• Add/extend regression tests: numeric reproduction + template-content assertions, and expand functional guards to accept find_mic(...) realignment as wrap-equivalent.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
MHI_Nickel/models/ase_neb.py	Replaces independent FS wrapping with MIC-based IS→FS alignment; adds cell guard + periodic-axis ambiguity warning; updates generated script template similarly.
MHI_Nickel/tests/test_ase_neb.py	Adds template assertions and numeric tests reproducing the boundary-wrap “jump” bug plus hardening behaviors (cell mismatch, z non-periodic, warning gating).
MHI_Nickel/tests/functional/test_ft_script_generation.py	Updates .wrap() regression guard to allow find_mic(...) realignment and expands the post-read inspection window.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.