Compare final geometries, not input site labels
Open every relaxed adsorption structure in the same orientation, first from above and then from the side. Relabel migrated states by their final geometry; record coverage, lateral cell, adsorbate count, reconstruction, spin state, and key distances. Place thumbnails or saved views beside the energy table so a human can distinguish a physical coverage effect from a periodic-cell or identity change. Use visual tools and specialist adsorption tools for common routes.
Optional grouping check: the state grid is a synthetic classification aid. Use it only after inspecting real relaxed structures; it contains no atomistic geometry and cannot establish a preferred site or coverage trend.
Run the bounded synthetic grouping fixture:
python3 examples/practical-guides/adsorption_state_grid.pyThe command retains every invented row, ranks states only inside matched groups, records a start-to-final site change, and compares two cells at one declared coverage. It does not read DFT outputs or perform an adsorption search.
What this guide verifies
Generate candidates, then relabel final states. Use the ASE surface builder, the pymatgen adsorption API, or another documented builder. For molecules, vary anchoring atom, orientation, conformer, dissociation state, and initial height when they can reach different basins. These tools create candidates; they do not identify the final minimum.
After relaxation, inspect the saved geometry and assign a final-state label. One fixture row starts at a top site and ends at a bridge site; keeping the starting label would corrupt the ranking. Deduplicate equivalent final structures but retain distinct metastable states and the path from each start.
For each real row, record structure and output hashes, surface identity, final site, cell transformation, one-face area, coverage definition, constraints, charge and spin, method identity, energy ledger, and convergence decision. Compare energies only when reaction, normalization, surface state, coverage, and numerical protocol match.
Separate coverage from periodic-cell sensitivity
The fixture’s top, bridge, and hollow curves share one invented cell family and coverage definition. A second bridge row changes the commensurate cell at the same declared coverage and produces a 0.04 eV fixture difference. That value is a deterministic diagnostic, not a universal tolerance.
One adsorbate in a larger cell changes both coverage and image separation. To test finite-cell sensitivity at fixed coverage, preserve adsorbate density and ordering in commensurate cells. To test physical coverage dependence, vary the occupied-site pattern while keeping the surface and reaction ledger controlled. Neugebauer and Scheffler and the CMR benchmark show why these are different comparisons.
Successful execution verifies invented grouping, relabelling, arithmetic, and rendering only. It does not establish a real site search, coverage convergence, slab convergence, lateral interaction, or global adsorption minimum.