| Ben Rudiak-Gould 2007-08-15, 7:11 pm |
| Jon Harrop wrote:
> patrick.crenshaw@gmail.com wrote:
>
> This is an excellent idea: Mathematica already did it.
Based on your followup message, I think you're talking about something else.
The OP appears to be asking about a system in which a max-error demand on
the output is propagated backward through the computation. This is what's
usually known as exact real arithmetic, and there are lots of libraries for
it in various languages, including functional languages. Here's one in Haskell:
http://www.augustsson.net/Darcs/CReal/CRealI.hs
You seem to be talking about interval arithmetic, where a max-error
specification on the input is propagated forward through the computation.
I think Mathematica also supports exact real arithmetic with the N function,
though the documentation is a bit unclear. I think the output is guaranteed
accurate to the specified number of digits, even though the documentation
only says (quite meaninglessly) that it's precise to that many digits.
-- Ben
|