The dispersion (red, green, blue) seems working under normal light conditions by using this methode Video.
But in the extreme case of using a prism it does not, the prism stays dark, no matter how high the samples are (100000).
So I simplified the problem from red, blue, green to only white light to reduce the complexity. The scene consists now only of two reflective walls in the background, one floor, one prism (material in photo below), one spot light and the camera.
One can see that one light ray goes straight ahead and the other is refract to the right. Is there a way to get only one retracted light ray through the prism and again retracted when leaving the prism?



