I am working on a diamond cut.  A while ago, in comparing ior.inc 
dispersion with Lilysoft's SpectralRender ("Native dispersion v. 
spectral render" in p.b.i), I noticed a discrepancy in the apparent 
index of refraction of BK7 glass between the two renders.  So, just to 
make sure that wasn't a fluke, I tested both methods with a diamond prism.

The images are of a glowing rod viewed through the prism.  I used black 
body 6504 for the SpectralRender runs because the black body curve is 
smoother in the violet area than D65; but I used rgb 1 (equivalent to 
D65) for the native POV-Ray runs, because POV-Ray's dispersion algorithm 
alters the hues of wavelengths for colored lights.  The two whites are 
very similar, but D65 has less violet.  The SpectralRender output is 
gamut-mapped to preserve hues.  The environment for all renders was 
centralized in an include file to guarantee consistency.

Image thru-pbi-diamond1.png shows my initial set of test runs.  The 
leftmost spectrum uses the dispersion value from consts.inc, and the 
rightmost uses the dispersion value from ior.inc.  I have not shown a 
render of the IOR from consts.inc, because I've long considered that 
value a typo, but if SpectralRender's detailed wavelength data are the 
benchmark, it is clear that const.inc's dispersion value is not accurate 
either.

But there is still a difference between ior.inc and SpectralRender in 
the overall amount of refraction.

I took a look at their respective values for n.  They are:

   ior.inc   SpectralRender
   -------   --------------
   2.4170    2.4176

The SpectralRender value is that of spline S_IOR_Diamond at wavelength D.

That difference is too small to account for the discrepancy between the 
renders.  My next thought was, what if POV-Ray's dispersion algorithm 
assumes that n is at a more central wavelength, say 550 mn?  I tried ior 
S_IOR_Diamond(550), and the result is the leftmost spectrum of image 
thru-pbi-diamond2.png.  But I had to take it down even further, to ior 
S_IOR_Diamond(515), for a comparable match.  That render is the 
rightmost spectrum.

It didn't take long browsing ior.inc to find in the notes that n is 
assumed to be at wavelength D.  I am no expert at any of this, and my 
ability to concentrate on *anything* (e.g., math, physics, SDL code) has 
waned sharply since my days as a young tech nerd, so I was at a loss as 
to what was going on.

But then I belatedly got the idea to render an image with no dispersion. 
  The result is the center panel of image thru-pbi-diamond3.png. The rod 
does indeed match the center of the ior.inc spectrum (to the left) and 
the D wavelength of the SpectralRender spectrum (to the right).  It 
appears the dispersion algorithm does assume that the IOR is for a 
central wavelength, rather than for D.  As for why I needed to go 
shorter than 550 for a good visual match, my guess is the non-linear 
correlation between refraction and wavelength.

Post a reply to this message

Attachments:
Download 'thru-pbi-diamond1.png' (47 KB) Download 'thru-pbi-diamond2.png' (42 KB) Download 'thru-pbi-diamond3.png' (32 KB)

Preview of image 'thru-pbi-diamond1.png'

Preview of image 'thru-pbi-diamond2.png'

Preview of image 'thru-pbi-diamond3.png'