sjl said:Well, if light hits an interface between elements at a 90 degree angle, it will go straight ahead. At a lesser angle than that, it will be bent at an angle that depends upon the refraction index of the materials at the interface. eg: when light enters glass from air, it will be bent towards the direction of the interface; if it enters air from glass, it will be bent away from the direction of the interface.
So I'd say that the key is the interface as the light moves out of the first element it hits. If it's angled just right relative to the angle of incidence where the light comes in, I can definitely see a fisheye being able to see 'behind' itself. You'd be talking about a severe curve on both sides of the front element, with the rear curve being more severe curve than the forward curve, but I don't see it as utterly impossible.
Have a look at this PDF for the construction of the original 220 degree FOV fisheye.
I'm WELL aware of the principals of optics. But you're only discussing geometrical optics, let me introduce some quantum mechanics. Specifically probability of reflection. Incident light to glass reflects ~4% of light, actually it is 0%-16% but it depends on thickness. Since we cannot yet make glass a perfect even thickness enough to make it reflect either of those, we pretty much always observe 4% for relatively small angles. Reflection skyrockets if the angle approaches 80-90º Thus the front element of lenses try to keep the angle of incidence as low as possible to avoid this. But there is no way around it if you plan on bringing light from slightly behind the front of the lens.
I can see how it could happen, but I just cannot see how such a design would not have severe vignetting. Especially since light from any given angle has to enter at a very specific part of the lens.