Re: [boost] [rfc] Inexact quantities arithmetics

From: Marat Khalili (0x8.0p15_at_[hidden])
Date: 2007-07-09 01:02:38

Andreas Pokorny wrote:
> This looks like interval arithmetic:
> 
> http://www.boost.org/libs/numeric/interval/doc/interval.htm
> and:
> http://www.open-std.org/JTC1/SC22/WG21/docs/papers/2006/n2067.pdf

Yes, this point needs to be explained. I did know this library exists, 
but it was not directly useful, there were differences:

1) First of all if one uses intervals instead if inexact quantities 
result for value (what's it? middle of the interval?) is affected. One 
might not want to estimate error this way.

2) Then, it only supports 'totally ordered' base types, no complex, 
vectors and so on. Adding support for these is most part of the work.

3) And it doesn't use derivatives, but simply calculates each function 
twice. It can be faster or slower, but the result is different.

4) Moreover, dyadic functions, mostly pow(inexact, inexact), are not 
implemented there yet. Implementing them up to the second order is the 
most part of the fun. :)

5) On the other hand, there are some interval class features, like 
rounding, I really don't need.

What is true is interval can be treated as a distribution (uniformly 
between specified values), so results have to be similar sometimes, 
especially at the first order.

So,

Could it be wrapper for the interval class? Probably it could, but it 
looked like headache for me.

Will inexact class support converting to and from intervals? Most 
likely, but to confidence intervals with all this quantile machinery.

With Best Regards,
Marat