Affected Repositories

• PyAutoGalaxy (primary)

Suggested branch: feature/area-magnification-lenscalc

Implementation Steps

1. autogalaxy/operate/lens_calc.py: three methods after magnification_2d_via_hessian_from; sub-pixel grid from aa.Grid2D.from_mask(mask, over_sample_size=sub_size).over_sampled, μ via magnification_2d_via_hessian_from on a Grid2DIrregular of the sub-pixel centres.
2. Segmentation: native boolean arrays of (unmasked & μ>0) and (unmasked & μ<0) at pixel level (pixel-centre μ), scipy.ndimage.label for connected regions → per-segment Mask2D.
3. Tests in test_autogalaxy/operate/test_lens_calc.py (or a new test_area_magnification.py).

Key Files

• autogalaxy/operate/lens_calc.py
• test_autogalaxy/operate/

Area magnification on LensCalc (LEGGOS Eq. 6-7): per-pixel inversion over an image-plane mask

Type: feature
Target: autogalaxy
Repos:

• PyAutoGalaxy
  Epic: cluster-strong-lensing
  Phase: 6a
  Parent: draft/feature/autolens/area_magnification_leggos.md
  Arc: draft/feature/autolens/source_cluster_arc.md
  Difficulty: medium
  Autonomy: supervised
  Priority: high
  Consequence: judge
  Filed: 2026-10-30
  Witness: LensCalc.area_magnification_from(mask) reproduces the analytic SIS annulus-segment area magnification to <0.5% at sub_size 4 with error falling as sub_size grows; area_magnification_list_from(mask) splits a mask covering several images into same-parity connected segments and returns one value per image; works for a Tracer via LensCalc.from_tracer(..., plane_j=); full PyAutoGalaxy + PyAutoLens suites pass.

Original request (human, 2026-10-30)

ship it when the tests pass, do all steps and then do phase 6

Phase-6 parent (verbatim user request, 2026-08-19): "Ability to get areas only Magnification as extension to
source science scripts. Look at LEGGOS paper for description of equations but basically area extension to the
above, seems like standard calculation in cluster lensing. [...] Remember also that area can link to source size."

Slice

Phase 6 is split: 6a library API (this prompt, PyAutoGalaxy LensCalc, which PyAutoLens reaches through
LensCalc.from_tracer); 6b autolens_workspace_test validation against simulated truth (uniform-disk source of
known area, resolution and segmentation controls, CI smoke); 6c workspace docs (source_science scripts:
area magnification + source size); 6d autolens_assistant LEGGOS II wiki/literature ingest. Forward ShapeSolver
stays out of scope (transferred to cluster-pointsolver-speed).

Work

1. LensCalc.source_plane_area_from(mask, sub_size=4, xp=np): A_src = sum over the mask's sub-pixels of
   (sub-pixel area / |mu|), mu from magnification_2d_via_hessian_from (LEGGOS Eq. 7). This is the source-plane
   area the image region maps from — the link to source size.
2. LensCalc.area_magnification_from(mask, sub_size=4, xp=np): mu_area = A_img / A_src (LEGGOS Eq. 6), for a
   mask covering ONE image. Document that a mask spanning several images of the same source double-counts A_src.
3. LensCalc.area_magnification_list_from(mask, sub_size=4): split the mask into images — connected regions of
   the same parity (sign of mu; the critical curve is the boundary) — and return one mu_area per segment
   (plus the segment masks), so a whole-arc mask gives per-image values.
4. Unit tests: analytic SIS annulus segment (outer image, theta in [r1, r2] maps to beta = theta - theta_E);
   sub_size convergence; parity split of a mask straddling the tangential critical curve; multi-plane via a
   mocked/Tracer-like deflection callable; NumPy path (JAX optional, importorskip).