support for unicode in identifiers

[Vlad Levenfeld via Digitalmars-d-learn]

I was pretty happy to find that I could use mu and sigma when 
writing statistical routines, but I've found that for more 
obscure non-ascii characters the support is hit or miss. For 
example, none of the subscripts are valid characters, but I can 
use superscript n as well as dot-notation for derivatives.
I'm using dmd 2.065. What's the story behind the scenes? Is there 
a rationale behind the supported/unsupported or is it 
happenstance? Is there anywhere I can find a list of supported 
characters?