question

To do this "physically accurate" the light rays would have to have a polar attribute attached to them somehow, and to fully simulate, you'd start running some quantum weirdness that definitely goes beyond the scope of what Arnold (or even Maxwell) is designed for.

Is there something specific you're trying to do?

Thanks for the anwer, actually how can I use (or where I can find docs/examples) for simulating seing my obj throught a polarised lens ? Do you also think it is possible to simulate something similar of what we have when we cross two linear polariser ?

My take is that I don't think it would make more realistic renderings. Light is generally unpolarized if you ain't trying to simulate laser beams or the final phase of a sun eclipse. After some kind of refractions or reflections it could end-up being polarized however light transport in rendering applications is then generally computed over diffuse interreflections so in a general case where caustics are not computed it would still be virtually unpolarized. That said linear polarization with lens filters for example is generally used to tweak reflections but for that we have already ray flags to ctrl independently reflections from eye/camera rays etc and to simulate some kind of polarization.