Re: [boost] Re: RFC: Differential Arithmetic Library

From: j.adelman_at_[hidden]
Date: 2002-02-12 15:13:15

Quoting Peter Schmitteckert 
<boost_at_[hidden]>:

> Salut,
> 
> On Tuesday 12 February 2002 20:20, 
j.adelman_at_[hidden] wrote:
> 
> > Does anyone else see a need for a template-
based
> > library for these kind of calculations?  If it
> > works, it won't be pretty, but it'll save some
> 
> Why won't it be pretty ? Have I missed 
something ?
> I like my aproach because it's simple.
> 
> > anguish from when the paper-and-pencil method
> > fails you.  The obvious question is whether 
the
> 
> That's one benefit: reducing typos. On does not 
need to
> type the derivative of a function.
> 
> > speed benefit is worth it.  I think it is when
> > the calculation will occur in the inner loop 
of a
> > program which takes 4 days to run, and that's 
the
> > cut-down version I ran when I realised how 
long
> > the full version would take to run.
> 
> Well, a typical application would be an inner 
loop of a
> differential equation solver. These tend to run 
very long
> if you have to apply them on a grid.
> And yes, I'd be happy if my application would 
finish it 4 days :).
> 
> Best wishes,
> Peter

Sorry, I was unclear.  I was suggesting something 
different from what your code does -- a tool for 
differentiating expressions into closed form at 
compile time.  Something that would look like:

typedef 
reciprocal<sub<constant<1>,exp<negate<var<0> > > 
> > logistic; // 1/(1-exp(-x))
//...
deriv<logistic,1>(x);

I think that qualifies at messy before I even 
think about implementing it.

James

-------------------------------------------------
This mail sent through UK Online webmail