Questions that occured to me while coding a PR

[rjmoerland]

Hi @HarrisonKramer , I'm close to marking my PR ready for review. During testing, I encountered two things that I'm not sure of, but that do not directly (have to) affect the PR:

1. When testing flipping a single paraxial lens, I got the wrong sign for lens.paraxial.f2() (positive where it should be negative). I went into the code and saw that , be.abs([f2]) is returned. Should f2 always be positive? I tried returning plain f2, and all pytests passed. Also, I obtained the result I expected. Then I tried a system with three paraxial surfaces, but then I got the wrong sign (I believe) for f2 (negative where I expected positive). What could be happening? Here is the code that gives me a negative f2() where I expected it to be positive. If I make the thickness of the last surface before the image plane zero, I get the number that I expect for F2() (about 4.43 mm), with the correct sign.

llens = optic.Optic()
lens.add_surface(index=0, thickness=np.inf, material="Air")
lens.add_surface(
    index=1,
    thickness=45 + 75,
    f=45,
    surface_type="paraxial",
    is_stop=False,
)
lens.add_surface(
    index=2,
    thickness=75 + 3.33,
    f=75,
    surface_type="paraxial",
    is_stop=False,
)
lens.add_surface(
    index=3,
    thickness=0,
    f=3.33,
    surface_type="paraxial",
    material=IdealMaterial(1.33),
    is_stop=True,
)
lens.add_surface(index=4, material=IdealMaterial(1.33))
lens.set_aperture(aperture_type="EPD", value=4.0)
lens.set_field_type(field_type="angle")
lens.add_field(y=0)
lens.add_wavelength(value=0.55, is_primary=True)
print(lens.paraxial.f2())
print(lens.paraxial.F2())

2. This is more of a "design philosophy" question: calling Optic.flip() flips curvatures in sign and swaps surfaces around such that positive bending becomes negative bending and vice versa, f2 becomes f1 and vice versa, etc. However, a positive lens remains a positive lens and a negative lens remains a negative lens. This is all exactly what I expect. However, flipping a thin lens (paraxial type) changes the type of lens from positive to negative and vice versa. It has been my use case to use flip() to simulate the system with, effectively, "light from the right". However, if there are paraxial lenses in the system, the system fundamentally changes. Is this by design?
   Thanks for any insight!

[rjmoerland]

A third one:
3. Calculating the image surface for a negative lens gives a negative value. The code then gives the image surface a negative position along the optical axis:
Before lens.image_solve():

After lens.image_solve():

Is this the intended behavior?

[rjmoerland]

4. The material of the image surface affects the outcome of the paraxial items. I'm not exactly sure how it works, but it seems to only give the correct results when the surface before the image surface and the image surface itself are of the same material (which makes sense). Should there be a warning, or perhaps even an enforcement of this?

[rjmoerland]

5. Optic.image_solve() will run even without a stop, but it does not give the correct answer it seems (tested with an immersed lens). With an explicit Optic.update_paraxial(), it will fail without a stop, but the user gets a warning. Would that be better?

[HarrisonKramer]

Hi @rjmoerland,

Thanks a lot for the detailed notes and for testing these cases thoroughly. Very helpful observations.

On f2(), you’re right. This is just the effective focal length, and I honestly don’t recall why we used abs() there. EFFL can indeed be negative, so that can be removed, especially if all tests still pass. I vaguely remember changing this, but don't recall why. We can investigate if really needed.

For flip(), I agree it’s inconsistent that paraxial lenses change sign while physical ones don’t. That wasn’t intentional, so I’m fine with adjusting that if it doesn’t break existing behavior.

For negative image positions, yes, that’s currently intended but could be made clearer in the docs.

On the image surface material, we can add a note or warning. I remember there were some nuances here that we checked against commercial tools, e.g., ZOS. I believe it was during the validation & verification of the FFT PSF. If we have solid references or a clearer understanding now, I’m open to updating it. It is not a "law" that it must exactly match commercial tools, especially if we have different conventions and it's documented well.

And on the stop issue, I agree. If there’s no stop, most (if not all) calculations should fail. update_paraxial relies on marginal and chief rays, so without a stop that should definitely raise an error. Making things stricter sounds like a good move. Defaulting to the first surface could be an option, but probably best not to assume that automatically.

Thanks again. These are really solid catches.

Best,
Kramer

[rjmoerland]

Thank you for the detailed response!

On f2(), you’re right. This is just the effective focal length, and I honestly don’t recall why we used abs() there. EFFL can indeed be negative, so that can be removed, especially if all tests still pass. I vaguely remember changing this, but don't recall why. We can investigate if really needed.

I've investigated a little further, and am nearly ready to open a PR with what I think is a fix for the cases I can think of. I'll follow up there and tag you.

For flip(), I agree it’s inconsistent that paraxial lenses change sign while physical ones don’t. That wasn’t intentional, so I’m fine with adjusting that if it doesn’t break existing behavior.

I'll check and open a PR if it can be changed without repercussions

For negative image positions, yes, that’s currently intended but could be made clearer in the docs.

Good to know, absolutely OK if it was intended.

On the image surface material, we can add a note or warning. I remember there were some nuances here that we checked against commercial tools, e.g., ZOS. I believe it was during the validation & verification of the FFT PSF. If we have solid references or a clearer understanding now, I’m open to updating it. It is not a "law" that it must exactly match commercial tools, especially if we have different conventions and it's documented well.

Unfortunately I have very little experience with the commercial tools, maybe we can discuss what we expect to happen somewhere else (sorry I let the discourse invite expire 😅 )?

And on the stop issue, I agree. If there’s no stop, most (if not all) calculations should fail. update_paraxial relies on marginal and chief rays, so without a stop that should definitely raise an error. Making things stricter sounds like a good move. Defaulting to the first surface could be an option, but probably best not to assume that automatically.

I'll open a PR 👍

[HarrisonKramer]

I'll follow up there and tag you.

Perfect, thanks.

I sent you the discord server link again. We can discuss more nuanced details there if it's easier.

I'll open a PR 👍

Sounds good!!

[rjmoerland]

For reference:

• Point 1: PR Paraxial focal length: fix wrong sign for negative lenses #382 fixes point 1.
• Point 2: PR flip(): don't flip sign of paraxial surfaces #381 fixes point 2.
• Point 4: Behavior is actually totally valid: I've used it for an immersion lens with cover slip, where the last physical surface is made of glass and the imaging medium is water. Optiland gives the correct focal length, because the media are different. ✅