Add linearity tests for forward models

[manduinca]

Fixes #293

Adds test/test_linearity.py with linearity tests for each geometry (points, prisms, tesseroids), parametrized by field, following the discussion in the issue: the field of a set of sources is compared against the sum of the fields of each source computed separately.

Includes both gravity and magnetic forward models: point_gravity (Cartesian and spherical coordinates), dipole_magnetic, prism_gravity, prism_magnetic and tesseroid_gravity, using mixed-sign masses/densities/magnetizations.

The newer ellipsoid_gravity/ellipsoid_magnetic forwards could get the same treatment in a follow-up if you'd like.

Squash and merge summary

Add linearity tests for the forward modelling functions of points, dipoles, prisms and tesseroids: check if the fields obtained after passing a collection of sources equals to the sum of the fields of each source computed independently.

[santisoler]

Hi @manduinca! Sorry for the delayed review. I was busy at a conference and catching up with stuff.

This looks really nice, thanks again for taking the effort and work on it! I left some minor suggestions below. Nothing major, just to make a few portions of the tests more Pythonic. Please, let me know what do you think, and feel free to apply and modify the suggestions as you like.

[santisoler]

Minor suggestion here: we don't need to index the arrays, we can iterate over them with a zip function:

Suggested change

	sample_prisms[i : i + 1],
	sample_densities[i : i + 1],
	field,
	)
	for i in range(sample_densities.size)
	prism,
	density,
	field,
	)
	for prism, density in zip(sample_prisms, sample_densities, strict=True)

[santisoler]

Similarly here:

Suggested change

	sample_tesseroids[i : i + 1],
	sample_densities[i : i + 1],
	field=field,
	)
	for i in range(sample_densities.size)
	tesseroid,
	density,
	field=field,
	)
	for tesseroid, density in zip(sample_tesseroids, sample_densities, strict=True)

[santisoler]

I would suggest indexing the arrays directly, so we pass only a float for each coordinate and for the mass. This way we are also testing (although we are probably already testing that in other places) that passing just a float as coordinates and single float for mass also works.

Suggested change

	tuple(p[i : i + 1] for p in sample_points),
	sample_masses[i : i + 1],
	tuple(p[i] for p in sample_points),
	sample_masses[i],

[santisoler]

Similarly here:

Suggested change

	tuple(p[i : i + 1] for p in points),
	sample_masses[i : i + 1],
	tuple(p[i] for p in points),
	sample_masses[i],

[santisoler]

And here:

Suggested change

	tuple(d[i : i + 1] for d in sample_dipoles),
	tuple(m[i : i + 1] for m in sample_magnetic_moments),
	tuple(d[i] for d in sample_dipoles),
	tuple(m[i] for m in sample_magnetic_moments),

[santisoler]

Suggested change

	sample_prisms[i : i + 1],
	tuple(m[i : i + 1] for m in sample_magnetizations),
	sample_prisms[i],
	tuple(m[i] for m in sample_magnetizations),

[manduinca]

Thanks for the review @santisoler! I applied all six suggestions: zip with strict=True in the prism and tesseroid tests, and scalars instead of one-element slices in point_gravity (cartesian and spherical), dipole_magnetic and prism_magnetic. Good point that this way we also cover passing floats for the coordinates and the mass.

Ran test/test_linearity.py locally: 32 passed. ruff check and ruff format come back clean.

[manduinca]

@santisoler this one has been approved since 18 August, and after the rebase on 14 September the CI is fully green (14/14) — the earlier red build and style were from main and got fixed in #699.

Is there anything left on my side, or is it just waiting for a merge window?

[manduinca]

@santisoler with the new AI policy in place, I want to be upfront about this one: the linearity tests were written with Claude Code (an AI coding agent) following my direction, and so were my replies in this thread, this one included.

The policy came after I opened the PR. If it doesn't fit, I'm fine with you closing it.

[santisoler]

Hi @manduinca. Thanks for disclosing your usage of AI in this PR.

Since we already put effort from our side on reviewing this PR, I'll move forward and merge it.

The issue that this PR solves was labeled as a good-first-issue because it could be approachable by a new contributor that wants to learn more on how to contribute to open-source or get more familiar with our project. Putting an agent to solve it hinders the possibility of a new contributor to learn from that experience. This is not something you should have been aware of, since you opened the PR before our Policy was published. But I still wanted to let you know about our goals and preferences.

You are more than welcome to make future contributions to Harmonica or to any other of the Fatiando projects. Just please be mindful on our AI Usage Policy and our preference of human-created content.