On Tue, 09 Feb 1999 21:30:39 -0500, Nathan Kopp <Nat### [at] Koppcom>
wrote:

>Seriously... doing this kind of diffraction would be very difficult.  First,
>as someone else mentioned, it would require backwards ray tracing.
>Secondly, it would require modeling light as waves instead of particles.
>The ray-tracing model can be visualized as lots of tiny particles flying
>through space.  I'm not sure how you'd change this model to incorporate
>expanding wavefronts.

>If anyone has any ideas, I'd be interested to hear them.  Keep in mind
>the very small size of these objects/slits that will be causing
>diffraction, and how dense you would need to trace rays in order to
>guarantee enough rays going through the slit or getting close enough
>to the strand of hair.
>
>Also keep in mind the (amount of work):(amount it affects an image)
>ratio.  ;-)
>
>-Nathan

You don't need to model waves (id est, trace spheres) and you can't. A
spherical wave has infinitely many intersection with a plane, the ones
cophasal forming ellipses. OTOH, if you trace rays and remember that
not only x,y and z, but also phi (phase shift) are functions of t. So
for interference, once you've told the program where the slit is, it
will trace it's bounding box with a number of rays and then treat the
rays hitting the edges of the slit (in an epsilon vicinity) as
secondary sources. Now if you look at the screen (i.e. send a camera
ray to it) you'll see that is is lit by these secondary light sources
(or photons if you wish :) ), but if you keep in mind phase shifts you
can correctly calculate the total light intensity at a point, i.e. get
minima and maxima. Same counts for diffraction (what are 100000
photons for Pete's sake? :) ), if the program knows where the grating
is and how dense, it could send a ray per aperture quite precisely.
Hair, too. In summary, you don't want diffraction etc. patterns
throrought the whole pic, so concentrate the rays where you know you
need them.

Another problem is wavelength. While one can specify it implicitly for
the aforementioned purposes, it is needed for dispersion, Rayleigh
scattering, etc. I have given thought to this and even tried to find
away to Fourier-analyse rgb components :) (I was young and naive then,
now I'm only young), but gave up and concluded that simple sampling
will be easier. Someone else found this before I did (Greetings Mr.
Wilson ), it seems. I thought that, instead of rgbf colours, light
spectrum could be used (like a color_map, but shows the intensity and
absorption of light for different wavelenghts), but then again,
nobody's *that* crazy to use this things.

Well that's my 35lv on the topic.

Peter

Post a reply to this message