I can see how it works in one of the links you can see that there's 2 faceted almost half cylinders attached together with a 90' twist so you get quantized (in space, not color value) sampling on each x/y axis.
I was thinking just that - I’d imagine it’d need to filter light to excite something that’d emit light at the wavelengths you want. It’s, as you noticed, a much more interesting problem.
Or emit white light when excited by a filtered wavelength which you’d filter again to get the color you want. With this second one you could do arbitrary palettes such as the Commodore 64 one.
Not exactly a crystal, but a quick google tells me trichroic prisms exist (https://en.wikipedia.org/wiki/Dichroic_prism). Once you've separated out RGB, could you quantize them and recombine them?
Well, “color” is fundamentally an electromagnetic spectrum. You can’t describe it by 3 values, as it might have an arbitrary shape. What it does is separate out the whole spectrum/graph into 3 graphs that overlap, and summing them up would roughly be the original. But neither of them is quantized this way, the R is not a single value, but still is a spectrum.
It’s a human sensory “feature” that certain graphs can be substituted by more standard ones that can be represented by a single number, e.g. a red LED’s spectrum between a certain range might combine well with a green and blue’s given spectrum, so that the combined graph looks similar to the original to us. Nonetheless, we have “compressed” the original and lost data.
I have to assume we're aiming for art, not science, here. If someone figures out how to get step changes to a light source with passive components, I'm gonna be impressed.
Splitting the light three ways first would just be the icing on the cake.
