This has been another one of those really interesting threads. Good discussion, experiments, and url's, folks!
Bill D wrote:Can I just pretend that pixels and RGB are Silver Halide Crystals and Dye Layers? LOL
Wait...you mean they're not?!
Those latest crops tell the story, don't they? From wide open down to around f/11, the diffraction blur is smaller than a pixel, so you gain DOF without losing sharpness. Around your f/22 setting, the blur goes to a couple of pixels wide, and you start to lose sharpness. At f/32, it's several pixels wide, and the blur really gets noticeable.
Because of differences in how the lenses focus, at 1:1 your Canon's "f/22" setting is probably about like my Sigma 105's "f/16" setting. As shown in
my article, the Sigma is noticeably fuzzy at 1:1 and f/22.
That Canon lens looks very nice, by the way. Not only is it pretty sharp all the way out to f/2.8, but I don't see nearly as much focus shift as the Sigma has. With the Sigma, its wide-open focus point is really pretty close to the front, not the center, of its f/11 focus range. I have to take this into account when choosing focus point for a single shot.
DaveW wrote:However this does not explain why it happens more at smaller effective apertures that at their actual physical size.
There are a couple of easy ways to think about it. The more accurate one is to think that diffraction depends on the
angular diameter of the aperture. Suppose you fix the aperture at so many mm in diameter, focus at infinity, and measure the size of the Airy disk. Then add extension so that the lens focuses at 1:1. How big is the disk now? Well, the distance from sensor to lens has doubled. So, the fixed size aperture now subtends half the angle that it used to, and diffraction is twice as bad -- the Airy disk doubles in size. The other way is to just imagine that the diffraction blur scales linearly with distance to the lens. Doubling the distance doubles the size of the blur.
Charles Krebs wrote:.... wow... just think...
an f.2 50mm lens would need a front lens diameter of 250mm. I'd need one of my sons to carry that baby!
Not only that, but it wouldn't help much with the diffraction problem, even ignoring the, um, rather formidable problem of aberrations.
What diffraction depends on is actually the angle subtended by the aperture. (More precisely, it depends on "numerical aperture", symbol NA, which is the sine of half the angle.) Make the lens as wide as you like, it's not going to subtend more than 180 degrees!
I like Charlie's description of how resolution varies with aperture. If you think of a graph, resolution (height) versus aperture (width), it usually looks like a hill with its highest point someplace in the middle. Like most hills made of dirt, you can move a fair distance laterally from the high spot before you lose much elevation. I usually see one click-stop that's measurably best, sometimes two about the same (with the best aperture between clicks), but going one stop on either side of best is seldom enough to attract attention.
However, the preceding paragraph only applies if you're looking at just resolution, ignoring DOF.
If you're trying to optimize aperture to get maximum DOF at a specified resolution, then the story is different. In that case, what you end up doing is stopping down until diffraction by itself consumes most of your tolerance. From that optimum point, making the aperture one f/stop smaller is like falling off a cliff -- your useful DOF ("depth of detail") drops to zero because the image is too fuzzy everywhere! (Think of going from f/22 to f/32 in Bill's examples.)
On the other hand, opening up one f/stop from the point of optimum DOF doesn't make much difference. I forget the exact number, and I'm away from my references at the moment, but as I recall, the DOF at one f/stop wider than optimum is still like 80-90% of DOF at optimum.
--Rik