Affected Repositories

• PyAutoArray (primary)
• PyAutoGalaxy
• PyAutoLens
• autolens_workspace

Branch Survey

Suggested branch: feature/magnification-maps-source-plane

Implementation Steps

1. autoarray/inversion/mappers/abstract.py: Mapper.mesh_magnification → np.ndarray of length pixels: mapping_matrix.sum(axis=0) * pixel_area / mesh_geometry.areas_for_magnification (pixel area from self.mask.pixel_scales). Pixels with zero area or no mapped image area return 0 (documented). Tests in test_autoarray/inversion/pixelization/mappers/: area identity sum(mu * area) == masked image area (to quadrature tolerance), identity mapping gives mu ~ 1 away from the mesh edge; rectangular, Delaunay, KNN.
2. PyAutoGalaxy autogalaxy/galaxy/plot/galaxies_plots.py:
   • plot_magnification_source_plane(mapper, caustics=None, vmin=None, vmax=None, title=..., output_path=..., ax=None) — plot_inversion_reconstruction(pixel_values=log10(mesh_magnification), mapper=..., lines=caustics).
   • plot_magnification_source_plane_grid(mass_obj, grid, source_shape_native=(100, 100), source_extent=None, sub_size=8, contour_levels=(...), include_caustics=True, ...) — over-sample the image grid, trace with grid - deflections_yx_2d_from(grid), np.histogram2d the sub-pixel areas into square source pixels, mu = area / pixel area, plot log10 mu with the 8a contour helper.
   • Export from autogalaxy/plot/__init__.py; tests.
3. Witness script (scratch, numbers in the PR): simulated simple__no_lens_light-style lens, Delaunay reconstruction with the true mass model; compare mesh mu per pixel against the square-pixel truth sampled at each mesh pixel centre.
4. PyAutoLens autolens/plot/__init__.py exports; Tracer tests.
5. autolens_workspace: scripts/imaging/features/pixelization/source_science.py and scripts/imaging/source_science.py new sections; regenerate_navigator.py autolens; smoke.

Key Files

• autoarray/inversion/mappers/abstract.py — Mapper
• autoarray/inversion/mesh/mesh_geometry/{rectangular,delaunay}.py — areas_for_magnification
• autoarray/plot/inversion.py — plot_inversion_reconstruction
• autogalaxy/galaxy/plot/galaxies_plots.py — 8a's plot_magnification_2d, _magnification_contours

Magnification maps 8b: source-plane magnification per mesh pixel

Type: feature
Target: PyAutoArray
Repos:

• PyAutoArray
• PyAutoGalaxy
• PyAutoLens
• autolens_workspace
  Themes:
• visualization
• cluster
  Epic: cluster-strong-lensing
  Phase: 8b
  Parent: draft/feature/autolens/magnification_maps_visualization.md
  Difficulty: medium
  Autonomy: supervised
  Priority: high
  Status: formalised
  Consequence: judge
  Review-minutes: 25
  Unattended: no
  Filed: 2026-11-02

Context

Slice of phase 8. Per mesh pixel j: mu_j = (image-plane area mapped into pixel j: sum_i M_ij * pixel area) /
mesh_geometry.areas_for_magnification[j] (phase 4, all meshes). Plot via plot_mapper(mapper, solution_vector=mu).
Square-pixel variant for parametric / MGE sources: ray-trace a fine image-plane grid, histogram image area per
source-plane pixel over its area. Uses the 8a colour/contour conventions. Workspace: pixelization source_science.
Witness: mesh mu map vs the forward-traced truth on a simulated lens (median per-pixel agreement stated).

Original request (human, 2026-11-02)

start 8b
rtu frame should be done so autoarray free now?
yes go ahead, create the issue

Approved plan (2026-11-02)

1. PyAutoArray: Mapper.mesh_magnification — mu_j = (sum_i M_ij * image pixel area) /
   mesh_geometry.areas_for_magnification[j]; tests: sum(mu_j * area_j) = masked image area, no-lens identity
   mapping gives mu ~ 1 in the mesh interior, rectangular / Delaunay / KNN.
2. PyAutoGalaxy: plot_magnification_source_plane(mapper, ...) (log10 mu on the mesh via
   plot_inversion_reconstruction, 8a conventions, optional caustics) and
   plot_magnification_source_plane_grid(mass_obj, grid, ...) (square-pixel: ray-trace a fine image grid,
   histogram traced image area per source pixel / pixel area; constant-mu contours).
3. Witness: mesh mu map vs forward-traced square-pixel truth on a simulated lens; median per-pixel agreement stated.
4. PyAutoLens: aplt exports + Tracer tests.
5. autolens_workspace: source-plane map sections in imaging/features/pixelization/source_science.py (mesh) and
   imaging/source_science.py (square-pixel); catalogue regen; smoke.