refactor: remove integral-based deflection/potential methods from mass profiles

[Jammy2211]

Overview

The deflections_2d_via_integral_from methods across mass profiles are slow, large, and only used for testing — the actual deflection calculations use MGE or CSE decompositions. This task removes all integral-based deflection, potential, and related helper methods from the source code, preserving them as standalone reference scripts in autolens_workspace_test/scripts/mass_via_integral/.

Additionally, the integral-based potential_2d_from methods on elliptical NFW and gNFW (which use scipy.integrate.quad) are removed. The gNFW convergence_func is kept in place as it is the only convergence calculation for gNFW and is also used internally by the MGE decomposition.

Plan

• Move all integral-based deflection/potential/convergence code to standalone scripts in autolens_workspace_test/scripts/mass_via_integral/ with numerical assertions
• Remove deflections_2d_via_integral_from and deflection_func from: Gaussian, Sersic, SersicGradient, NFW, gNFW, gNFWSph
• Remove integral-based potential_2d_from and potential_func from NFW (elliptical) and gNFW
• Remove tabulate_integral from AbstractgNFW (dead code after removal)
• Update comparison tests (integral vs CSE/MGE) to use hardcoded expected values
• Delete tests that directly test integral methods
• Keep gNFW.convergence_func (essential for convergence and MGE deflections)

Detailed implementation plan

Affected Repositories

• PyAutoGalaxy (primary)
• autolens_workspace_test

Work Classification

Library + Workspace

Branch Survey

Repository	Current Branch	Dirty?
./PyAutoGalaxy	main	clean
./autolens_workspace_test	main	clean

Suggested branch: feature/remove-deflection-integral

Implementation Steps

1. Capture hardcoded values from integral methods before removing source code
2. Create 6 workspace scripts in autolens_workspace_test/scripts/mass_via_integral/
3. Update comparison tests to use hardcoded numpy arrays
4. Delete direct integral tests (20 tests across 8 files)
5. Delete gNFW potential tests (5 tests)
6. Remove source methods from 6 files
7. Run full test suite to verify

Key Files

• autogalaxy/profiles/mass/stellar/gaussian.py
• autogalaxy/profiles/mass/stellar/sersic.py
• autogalaxy/profiles/mass/stellar/sersic_gradient.py
• autogalaxy/profiles/mass/dark/nfw.py
• autogalaxy/profiles/mass/dark/gnfw.py
• autogalaxy/profiles/mass/dark/abstract.py

Original Prompt

Click to expand starting prompt

Throughout the package PyAutoGalaxy/autogalaxy/profiles/mass there are methods which compute deflection angles via a method deflections_2d_via_integral_from. These are slow, and often very large. Move all methods to a dedicated package in autolens_workspace_test/scripts/ called mass_via_integral, asserting their values based on the autogalaxy unit tests. For the majority of these functions, the integral was only used for testing anyway, with an MGE or CSE decomposition forming the basis of the deflection angle. Do not remove or change any unit tests unless they are directly against the integral. Also remove all long integral functions used for potentials or convergences, moving them to the same mass_via_integral.

[Jammy2211]

Smoke Test Results — 2026-04-06

Workspace	Passed	Failed	Total
autofit_workspace	7	0	7
autogalaxy_workspace	3	4	7
autolens_workspace	4	5	9
autolens_workspace_test	7	0	7

All 9 failures are the pre-existing PYAUTOFIT_TEST_MODE=2 KeyError in autofit/non_linear/samples/sample.py — unrelated to this PR's changes.

[Jammy2211]

Shipped — 2026-04-06

PRs

• refactor: remove integral-based deflection/potential methods from mass profiles #324
• feat: add mass_via_integral reference scripts autolens_workspace_test#13

Summary

Removed all integral-based deflection, potential, and related helper methods from PyAutoGalaxy mass profiles (Gaussian, Sersic, SersicGradient, NFW, gNFW, gNFWSph). Preserved the integral code as standalone reference scripts in autolens_workspace_test/scripts/mass_via_integral/. Updated 18 comparison tests to use hardcoded expected values. Kept gNFW.convergence_func as it is essential for convergence and MGE deflections.

Session Notes

• gNFW.convergence_func uses scipy.integrate.quad but cannot be removed — it is the only convergence calculation for gNFW and is also used internally by MGEDecomposer for the MGE-based deflection path.
• Elliptical NFW/gNFW potential_2d_from now raises NotImplementedError (falls through to base class). Only NFWSph.potential_2d_from (analytic) remains. The elliptical NFW potential tests were already commented out before this change.
• test_gnfw.py had two misleadingly named tests: test__convergence_2d_from__elliptical_vs_spherical actually tested potential_2d_from, and vice versa. The former was deleted (tests removed potential), the latter kept.