Tracer accepts non-iterable galaxies; multi-plane allows z_lens > z_source

[rhayes777]

Title: Tracer accepts non-iterable galaxies; multi-plane allows z_lens > z_source

Repo: Jammy2211/PyAutoLens

Tag: @Jammy2211

---

Two small validation gaps in al.Tracer. Found while auditing
autolens == 2026.5.21.1.

B4. Tracer accepts a string (or anything) for galaxies

import autolens as al
t = al.Tracer(galaxies="not a list")
# silently constructed; explodes much later

The first usage (t.image_2d_from(grid=...)) does eventually fail, but with a
confusing downstream error rather than TypeError("galaxies must be an iterable of Galaxy") at construction.

Bonus. Multi-plane configurations with z_lens > z_source

l = al.Galaxy(redshift=1.0, mass=al.mp.IsothermalSph(einstein_radius=1.0))
s = al.Galaxy(redshift=0.5,
              bulge=al.lp.Sersic(intensity=1.0, effective_radius=0.1,
                                 sersic_index=2))
t = al.Tracer(galaxies=[l, s])
np.array(t.image_2d_from(grid=grid)).sum()
# 65.05 — finite, no warning

Also: redshift = -0.5 and redshift = 1e-12 are both accepted. This may be
deliberate (multi-plane lensing supports lots of geometries) but in practice
"lens behind source" tends to be a user mistake worth a warning.

Environment

• autolens == 2026.5.21.1, Python 3.12.8.

Repros at work/edge_case_tests/03_tracer.py and
work/edge_case_tests/09_more_edges.py in the audit workspace.

[Jammy2211]

Thanks @rhayes777, and apologies for the long silence on this.

Both items reproduce on current main (autolens @ 6567b3b1c):

• B4 — al.Tracer(galaxies="not a list") constructs silently. First use then fails with AttributeError: 'str' object has no attribute 'redshift', which tells the user nothing about what they actually did wrong.
• Bonus — z_lens=1.0 with z_source=0.5 produces a finite image with no warning, and Galaxy(redshift=-0.5) and Galaxy(redshift=1e-12) are both accepted.

B4 is a clear-cut validation gap and we'll add a TypeError at construction naming what galaxies expects.

On the redshift ordering, your instinct that it "may be deliberate" is right, and it's worth being explicit about why we're going to treat the two halves differently. Multi-plane lensing genuinely does support configurations that look wrong at a glance — sources in front of deflectors is a real thing when you have several planes and the "lens" and "source" labels are just our naming convention rather than a physical constraint. So we don't want a hard error there, and we don't want to bake our two-plane vocabulary into a check the multi-plane code has to fight.

A negative redshift is a different matter — that isn't a geometry we support under any reading, and it should be rejected outright. 1e-12 sits in between: technically valid, almost always a mistake.

So the plan is: hard ValueError on negative redshift, and a warning (not an error) on the lens-behind-source ordering, worded so it's obviously skippable if you meant it. If you've hit a multi-plane setup where even a warning would be noise, say so — you're closer to that use case than the warning's author will be, and I'd rather hear it now than ship something that nags you.

Tracked with the rest of your audit (PyAutoArray#332, PyAutoArray#333, PyAutoGalaxy#440, PyAutoLens#531) as one campaign; I'll link the tracking issue here shortly.

[Jammy2211]

As promised above — the tracking issue for this campaign is
PyAutoLabs/PyAutoArray#415, which carries the verification evidence for all 16
findings and the phase breakdown. Watch that one for overall progress; I'll keep
updating this thread as the parts touching it land.

Phase 1 (the three crash-level findings, which is where we're starting) is
PyAutoLabs/PyAutoArray#416.

[Jammy2211]

Both unblocked findings here are fixed and merged. Thanks @rhayes777.

B4 — PyAutoLens#696 (65183d1). Tracer(galaxies=...) now raises TypeError at construction naming galaxies and the type given. Your point about iterability was the right one to make: the guard checks element type, not isinstance(x, Iterable), precisely because a string passes the latter. Broader than reported, too — 42, None and dicts were all being accepted at construction, not just the string.

Negative redshift — PyAutoGalaxy#566 (a366f77), not here. al.Galaxy is ag.Galaxy, so both the class and its redshift assignment live in PyAutoGalaxy; guarding at Tracer level would have missed a bare al.Galaxy(redshift=-0.5), which is your actual reproduction. Zero and 1e-12 stay accepted — degenerate, but not invalid, and rejecting them would break single-plane work where the redshift is a label.

z_lens > z_source is still yours to call. Untouched, exactly as we said. The question from 2026-07-28 stands: in a real multi-plane setup, would even a warning be noise? No rush — but so the decision cannot drift while it is open, both repos now carry a control test asserting that a z_lens=1.0 / z_source=0.5 system still constructs and evaluates to a finite image. If anyone later tries to make that an error, those tests fail first.

---

Generated by Claude Code