Total internal reflection in a right-angle prism
Light inside glass that meets a face at 45° cannot leave: a right-angle prism turns a beam by 90° with one such reflection, and sends it back with two.
Open in the canvas →Background
Light passing from glass to air bends away from the normal of the surface. As the angle of incidence inside the glass grows, the refracted ray leans further over, until at the critical angle, arcsin(1/n) for glass of index n in air, it runs along the surface. Beyond that angle there is no refracted ray: the light is completely reflected back into the glass[1].
For common optical glass the critical angle is close to 42°, just under 45°, and that makes a prism with angles of 45°, 90° and 45° a mirror with no coating. A beam entering one of the short faces meets the long face at 45° and is turned by 90°. A beam entering the long face is reflected by both short faces in turn and leaves through the long face again, turned by 180° and shifted sideways: the Porro prism, named after Ignazio Porro, and used in pairs to turn the image upright in binoculars[1,2].
The reflection is total, but the light is not confined to the glass. An evanescent wave runs along the outside of the face, its amplitude falling off exponentially with distance; a second piece of glass brought close enough picks it up, and the reflection is no longer total (frustrated total internal reflection)[1]. The reflection also shifts the phase of the two polarizations by different amounts, which the Fresnel rhomb uses to do the work of a quarter-wave plate[1].
What this setup demonstrates
Both prisms are drawn with Freeform glass, a polygon of N-BK7 with short faces of 25.4 mm. (The Prism element is an isosceles prism whose apex angle stops at 80°.) At 633 nm the glass catalogue gives n = 1.515, so the critical angle is 41.3°.
In the upper setup the beam enters a short face head-on, meets the long face at 45° and leaves through the other short face, turned by 90°. In the lower one it enters the long face, 9 mm above the middle, is reflected by both short faces and returns parallel to itself, 18 mm lower.
Each power meter reads 922 µW of the laser's 1 mW. None of the loss is at the reflections. Freeform glass has a setting for the transmission of each face a ray crosses, here 96%, the value for uncoated glass of this index met head-on (the reflected part is ((n−1)/(n+1))², 4.2%). Each beam crosses two faces: 0.96² = 0.9216.
Things to try. Set the per-surface transmission of a prism to 100%: its meter reads 1.00 mW, all of the light. Select the upper prism, choose Constant index as its glass model and lower the index: at 1.42 the critical angle is 44.8° and the beam is still reflected; at 1.41 it is 45.2°, more than the 45° at which the beam arrives, and the beam leaves through the long face instead. The limit is n = 1/sin 45° = 1.414. Turn the lower prism by up to 5.5° either way: the beam still comes back parallel to the way it went in, which is what makes two mirrors at 90° a retroreflector in their plane. Past 5.6° one of the two reflections falls below the critical angle and the light leaves the prism there.
Below the critical angle the tracer sends all of the light through the face, less the per-surface setting. Real glass reflects a part that depends on the angle and on the polarization, and that part rises steeply to 100% as the critical angle is approached; here the change at the critical angle is a step. The faces the beam enters and leaves by keep their set transmission at any angle, with no Fresnel calculation.
There is no evanescent wave, so a second prism brought up to the reflecting face does not frustrate the reflection, however close. The reflection changes the polarization as an ideal mirror does, keeping p and s and swapping the two diagonals, but the phase shift total internal reflection adds between the two polarizations is not applied, so a Fresnel rhomb drawn this way leaves linear light linear.