A study in lens diffraction f2.8 - f 22

I recall an article (though I can't find the reference) where it was demonstrated how careful sharpening can undo the effects of diffraction in images taken with small apertures to the point that it would take pixel peeping to see the softening effects of diffraction. Those weren't shots of test charts but real-life images (landscapes, as I recall). It was demonstrated that the softening by diffraction was gradual with decreasing aperture and could be counteracted over quite a wide range of aperture with sharpening. Only when stopping down too far did one hit the "brick wall" and the remedy no longer worked. Though I generally try to avoid stopping down beyond f/11 - my take-away from the demo was not to worry too much about diffraction.

f11 for what format? what would you suggest with 4x5?

I recall an article (though I can't find the reference) where it was demonstrated how careful sharpening can undo the effects of diffraction in images taken with small apertures to the point that it would take pixel peeping to see the softening effects of diffraction. Those weren't shots of test charts but real-life images (landscapes, as I recall). It was demonstrated that the softening by diffraction was gradual with decreasing aperture and could be counteracted over quite a wide range of aperture with sharpening. Only when stopping down too far did one hit the "brick wall" and the remedy no longer worked. Though I generally try to avoid stopping down beyond f/11 - my take-away from the demo was not to worry too much about diffraction.

Sharpening needs detail to sharpen. The resolution limiting effect of diffraction is due to Airy discs growing in size and overlapping such that detail is no longer resolved.

There is nothing to bring back by sharpening. You can only sharpen the coarser detail that you have not yet lost.

You cannot undo or counter diffraction. Detail lost is detail lost. Gone forever.

There is that phenomenon, though, that a high edge contrast low resolution image appears more pleasing than a low contrast but higher resolution iimage. They must have fallen in that trap.

f11 for what format? what would you suggest with 4x5?

Think equivalence, so the longer focal lengths will have less diffraction at the same f/stop and if you stop down to the same aperture size the diffraction will be the same but less magnified because you don't have to enlarge the larger format as much as the smaller format.

Would people use different standards based on whether you can tilt the lens or not?

The reason I've been getting so much heat here is that I say you should not be overly concerned about some arbitrary standard that most of the time can be compensated for with additional sharpening (thus, I say f/11 works for me because it's still almost completely recoverable).

I have a couple of tilt lenses. One of them is a Rokinon 24mm f/3.5 that is pretty much a dog at f/5.6 (not surprising considering it's a relatively cheap lens and f/5.6 is less than two stops from wide open), but by f/11 (yeah, f/11, again) it's well behaved even fully shifted. Shifting is my primary usage for this lens, but when I'm close to the ground and want the foreground to be in-focus right up to the nearest foreground then it's time to tilt the focus plane, and at wider apertures that focus plane will be so narrow near the camera that I usually have a hard time not having the focus plane drift away from the ground because I often do want something that's above ground level to also be in-focus. Case in point:

I wanted all the grass in the foreground in-focus, and I wanted the two rocks in the foreground in-focus, so with the lens wide open using Live View on my D500 I ended up tilting the focus plane about 1° and with the near focus being just in front of the nearest part of the foreground and the farthest part of the focus that wasn't in the sky being just in front of the top of the two rocks, that way when I stopped down everything from the top of the two rocks to every blade of grass in front of them is now within an acceptable DOF. I shifted the lens up for the second tile, used Photoshop's Photomerge to combine them, applied selective curves to the sky and separately to the rocks and foreground, cropped it to 1:1 aspect ratio, and sharpened for the output. Here are 100% crops of the left foreground, two portions of the rocks in front, and the rock in the background (which shows the DOF is not infinite):

One last point here. I have used f/8 with some success on this lens, and since I was using the D500 that was actually tempting for me, and I actually tried it on another day in this same spot. However, and I think this is an important point, I didn't care for the extra level of blur on the background rock; I actually don't want the background rock in-focus either, but I didn't want it overly blurred and found f/11 better suited my own personal aesthetic (YMMV).

Sharpening needs detail to sharpen. The resolution limiting effect of diffraction is due to Airy discs growing in size and overlapping such that detail is no longer resolved.

There is nothing to bring back by sharpening. You can only sharpen the coarser detail that you have not yet lost.

 

You cannot undo or counter diffraction. Detail lost is detail lost. Gone forever.

I would add two caveats here. First, and this is what I've been arguing in this thread from the beginning, that detail needs to actually be recorded to be relevant, so f/11 generally records a blurrier version of detail than f/5.6, but both versions are usually near enough to each other for the detail to be recorded by a 45+ MP 135 format sensor, and my experience is that if I am displaying them for the purposes of pixel peeping then I need to sharpen both versions, albeit less so for the f/5.6 version. My second caveat is that you can only see what your display size allows, so if you're displaying on a typical desktop monitor you won't be able to see very much fine detail so brilliantly captured at f/5.6 on my 21 MP D500 or with a 45 MP D850, and even with a 20x30 inch print on the wall looked at as closely as I can comfortably see with my old eyes (it's time for me to consider trifocals) I don't see all the detail I can get from my 36 MP D800.

For example:

Here are two differently sized crops (100% with sharpening on the left; 33% without sharpening on the right):

I have a 20x30 inch print on the wall, unframed (yeah, just tacked up on the wall), and I can go right up to it and look at it, and what I see approximates the right crop.

BTW, that shot was taken back in 2014 with my newly acquired D800 on a well secured Gitzo tripod using MUP and a remote release with an old AF-NIKKOR 50mm f/1.8D (a lens I especially like for its sunstars) at f/5.6.

f11 for what format? what would you suggest with 4x5?

DX and FX, I have no experience with large format.

Under what conditions is it visible? When pixel peeping? I'm much more concerned with what is noticeable viewing a print from a normal distance than with what can be detected with a magnifing lens.

It's visible under magnified LiveView or EVF viewing before image capture, but that wasn't the argument.

It seemed to be being claimed that diffraction wasn't visible at all unless an aperture like f/16 was used, and that just isn't the case.

Information, or knowledge, exists outside of and apart from its application. What one does with knowledge is entirely up to the individual - use it or ignore it. But denying it is an entirely different matter.

It seemed to be being claimed that diffraction wasn't visible at all unless an aperture like f/16 was used, and that just isn't the case.

I know I didn't actually say you can't see diffraction. I showed the effects of diffraction at f/11 before and after post-processing.

One of the epiphanies of mirrorless digital cameras is that many lenses are sharper wide open than at f/11. We demand and get corners nearly as sharp as the center. We also have tools which can be used to focus very precisely (revealing that DOF is a useful fiction).

 

The falloff in resolution at smaller apertures is due not only to diffraction, but residual errors in element curvature, spacing and centering. The uncertainty of these factors add roughly as the root-mean-square of the individual values. This causes the net resolution (contrast) to be rounded, rather than a sharp peak or break in performance. Even though diffraction is calculated as a singular value, it is represented in nature as a set of concentric Airey rings. Rather than binary parameters, I prefer to think of these phenomena as uncertainties. This disqualifies me from ever being a politician or lawyer.

Both diffraction and aberration try to put an exact number on something that isn't so exact.

There is the Rayleigh criterion for diffraction, which is reasonably but surely not exact to two or three significant digits.

Chromatic aberration and diffraction are both wavelength dependent.

A lens might be diffraction limited for red light, and not for blue. Then what do you say?

The ability to separate peaks, as in the Rayleigh criterion, depends on the signal/noise ratio.

One can apply deconvolution algorithms to separate peaks if there is enough signal.

(Note the early results from the Hubble telescope are of brighter objects.)

So, even though it shouldn't be, it is likely that the results are different for digital and film,

for a variety of reasons.

You can apply a mathematical tool called Discrete Fourier Analysis to analyze interactions in digital systems. Since diffraction is a quantum phenomena, the net product with a digital sensor is affected.

This is or more academic than practical interest, since in the real world, there are many overlapping source of light, hence diffraction, producing effects which are reasonably described as a Gaussian blur.

So, even though it shouldn't be, it is likely that the results are different for digital and film,

for a variety of reasons.

Mainly because film is crap - for a variety of reasons, but mainly because its emulsion has a very measurable depth. Especially true for colour film.

This effectively extends depth-of-field (while disallowing exact focus) and covers up some degree of spherical aberration and diffraction.

Also, the way that silver is ejected from a halide crystal during development, and the diffusion of developer oxidation products towards a coupler globule, means that the resulting image point is produced with a random spatial offset from where the initiating photons hit the halide crystal.