Elliptical/Oval Bokeh Explained

[Brian Caldwell]

Apologies in advance if this is widely known.  Personally, I've never found a really good explanation of why front anamorphs produce oval bokeh and rear anamorphs don't, despite reading my fair share of patents, technical papers, internet gossip and the like.  Feeling that my own understanding needed some firming up I finally set up some paraxial models and went through the math in gory detail.  It all boils down to how front and rear converters alter (or don't alter) the f/#, and basic DOF type circle of confusion calculations.  It has nothing to do with higher order aberrations, or the shape of the front lens, or various mechanical aspects of the lens.

Briefly:

1) A front anamorph is just a special case of a front afocal attachment, and as a result it preserves the f/# of the lens its attached to.  With an anamorphic front lens the focal length is shorter in the powered axis than in the non-powered axis.  For example, consider a 2:1 anamorph attached to a 100mm f/2 spherical lens.  In this case the net focal length is 50mm in the powered axis and 100mm in the non-powered axis, but in both cases the aperture remains f/2.  If you venture into the weeds to do circle of confusion calculations for a given object-space defocus you discover that a de-focused point source evaluated at the image plane is an ellipse with an aspect ratio of 4:1.  However, you only need to de-squeeze the image by 2x to correct the in-focus geometry, so you are left with de-focused ellipses with an aspect ratio of 2:1.

2) A rear anamorph is just a special case of a rear-mounted teleconverter, and as a result it *does not* preserve the f/# of the lens its attached to.  In particular, in the powered axis the aperture becomes slower.  For example, consider an 50mm f/2 spherical lens with a 2x rear anamorph.  Here the net focal length is 100mm in the powered axis, but the aperture has dropped to f/4, and is still 50mm f/2 in the non-powered axis.  When you do the circle of confusion calculations with object-space defocus you find the on-sensor defocused image to be an ellipse with a 2:1 aspect ratio.  When you desqueeze by 2x this defocus ellipse becomes a perfect circle.

Bottom Line:  Rear anamorphs have circular bokeh because they *don't* preserve the f/# of the spherical lens in both axes, while front anamorphs have elliptical bokeh because they *do* preserve the f/# of the spherical lens in both axes.

[Caleb Genheimer]

I've always been surprised at people that swear by 1.33X or 1.5X anamorphics, because it seems that they don't understand this stuff. It is a HUGE part of the look! If you're going to go less than 2X (or if you're going to use a rear anamorph), you almost may as well just use a streak filter and crop with a barrel-distorted spherical prime. 

One of my better ideas would be for someone to do a range of 1.5X lenses where the aperture itself is a 1.5X oval. DSO already technically does this with their TRUMP lenses (as individual disks), but a nice 11-bladed aperture mechanism would not be difficult to redesign to close as an oval rather than as a circle. IMO you still need at minimum 1.5X to get decent flares that aren't thin and wimpy.

[Brian Caldwell]

19 minutes ago, Caleb Genheimer said:

I've always been surprised at people that swear by 1.33X or 1.5X anamorphics, because it seems that they don't understand this stuff. It is a HUGE part of the look! If you're going to go less than 2X (or if you're going to use a rear anamorph), you almost may as well just use a streak filter and crop with a barrel-distorted spherical prime. 

. . . . .

I agree for the most part, although 1.79x is a good alternative to 2x, since it is more efficient and in a good implementation actually looks more anamorphic than many 2x lenses.

[ken]

1 hour ago, Caleb Genheimer said:

I've always been surprised at people that swear by 1.33X or 1.5X anamorphics, because it seems that they don't understand this stuff. It is a HUGE part of the look! If you're going to go less than 2X (or if you're going to use a rear anamorph), you almost may as well just use a streak filter and crop with a barrel-distorted spherical prime. 

One of my better ideas would be for someone to do a range of 1.5X lenses where the aperture itself is a 1.5X oval. DSO already technically does this with their TRUMP lenses (as individual disks), but a nice 11-bladed aperture mechanism would not be difficult to redesign to close as an oval rather than as a circle. IMO you still need at minimum 1.5X to get decent flares that aren't thin and wimpy.

IMHO, flare does not related anamorphic ratio.  It mainly depends on coating.   Many 1.33x lenses (ie. Century's) have much better flare then 2x lenes (golden projection lenses).  But for better oval bokeh, 2x lenses are the best.  1.7-1.9x lenses are just rare than 2x lenses.  So they are more expensive.  1.5x lens is more close 1.33x lens than 2x lens.  It is also quite expensive too.