Skip to content

Package 10.8.1–10.8.2: real-map corpus and scientific isosurface convergence gate - #77

Merged
aliorhun merged 10 commits into
mainfrom
codex/package-10-8-real-map-convergence
Oct 17, 2026
Merged

aliorhun merged 10 commits into
mainfrom
codex/package-10-8-real-map-convergence

Conversation

@aliorhun

@aliorhun aliorhun commented Oct 17, 2026 •

Copy link
Copy Markdown
Contributor

Package 10.8.1–10.8.2 — real-map corpus and required scientific convergence gate

Implementation: The current production volume-iso-worker.js final marching-tetrahedra algorithm is invoked in a new analytic/real-map numerical gate. The P10.7 octree remains preview-only.

  • Analytic reference: signed-distance sphere evaluated at 16³, 24³, 32³. Fail CI if maximum radial vertex error ≥0.025 Å, surface relative error ≥8%, enclosed-volume relative error ≥10%, or fine-resolution measures do not improve on coarse.
  • Independent crystallographic data: real 5i55 partial monoclinic CCP4 fixture, and real 5wkd C2 MTZ FWT/PHWT and DELFWT/PHDELWT coefficients converted to 64³ density maps by pinned Gemmi 0.7.5 (not by JavaScript FFT).
  • Real-map measurements: physically matched resampling at 16³/24³/32³, production triangle cap enforced, surface areas, voxel-quadrature occupied volume, source-field scalar residual, CUBE/XSF coordinate+voxel round trips, machine-readable JSON artifact.
  • Required PR CI: Validate and deploy crystal studio runs the gate; malformed, nonfinite, truncated, degenerate or analytically inaccurate results fail CI. Existing Gemmi MTZ/CCP4/Chromium gates are retained.

Measured experimental resolution sensitivity (16³ → 32³)

Real source Absolute relative surface-area change Thresholded volume change
5i55 partial CCP4 3.668% 2.116%
5wkd FWT/PHWT 46.270% 24.229%
5wkd DELFWT/PHDELWT 3.790% 1.685%

Scientific finding: The refinement-weighted 5wkd FWT surface is not geometrically converged at these tested resolutions. A passing CI means the implementations, analytic shape and reproducible measurement/reporting gates passed; it does not mean every experimental surface converged. These between-grid changes are not Hausdorff guarantees. The real-map threshold is chosen near signed density peaks to keep measurements within the production 75,000-triangle safety budget.

Open work: higher-resolution 5wkd convergence plus independent experimental CHGCAR/CUBE/XSF and wider MRC vendor corpora; manifold/oriented volume on real closed surfaces; multi-GPU/OS benchmark matrix.

Unmerged pending browser CI.

@aliorhun
aliorhun merged commit c56c30a into main Oct 17, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant