q734738781/structure-visual-inspection

structure-visual-inspection

Overview

Use rendered structure views as auxiliary evidence for geometry sanity checks, visual comparison, and reportable artifacts.

Quick Start

1. Render the structure with four standard views using render_structure_views(backend="ase", ...).
2. Inspect the panel directly first; only call image analysis when the visual question is still ambiguous.
3. Treat visual findings as supporting evidence.
4. Confirm critical geometric claims with numerical tools when available.

Allowed tools

• render_structure_views
• analyze_images
• read_file

Workflow

1. Render a default four-view panel first with backend="ase". Keep the first pass simple: default fit, default legend, no arbitrary camera tuning.
2. Increase supercell only when slab-site context is visually too local. Typical slab retries are (2,2,1) or (3,3,1), not larger by default.
3. Use analyze_images only with a narrow question: orientation sanity check, obvious clash, site-family ambiguity, or pre/post-relax comparison.
4. When comparing candidates, keep the rendering preset and visual context aligned across all images. Do not mix one candidate with a larger supercell or looser fit unless the report says so.
5. If a visual conclusion would affect ranking, site assignment, or a scientific claim, hand back to numerical geometry tools or file-based metadata before finalizing.

Method-critical defaults

• Visual evidence is never a substitute for exact bond lengths, coordination metrics, adsorption energies, or thermodynamic quantities.
• For slab-site interpretation, supercell context can materially change what looks like bridge/hollow/top. Make the chosen supercell explicit when it matters.
• Prefer the ASE backend for experiment-side inspection because it preserves more explicit geometric context; reserve OVITO for manuscript-facing aesthetics.
• If perspective or isometric views suggest a geometry issue, verify it with file-based structures or numeric geometry analysis before changing the workflow.
• Keep rendered legend and element identity explicit; do not infer species from color alone in the final answer.

Output Contract

• Return the main panel image path.
• If the analysis used multiple images, return all image paths referenced in the conclusion.
• State whether the visual conclusion is obvious, tentative, or requires numerical confirmation.
• When image analysis was used, summarize the concrete findings and the main uncertainty.

References

• Use the rendered panel for interpretable reporting artifacts, but keep the final scientific conclusion grounded in structure files and numerical results.