ratinterpol always rats?



xefe wrote:
> El vie, 07-11-2008 a las 10:05 -0500, Raymond Toy escribi?:
>> I notice that ratinterpol always calls rat on its arguments.  Is there
>> a particular reason for this?
>>
>> I was just playing around with ratinterpol with float and bigfloat
>> points, and the result is always rational.  I could just float or
>> bigfloat the result, but I was wondering why ratinterpol always
>> rationalizes floats.  I guess in practice it may not matter
>>
>> (Perhaps the documentation for ratinterpol should mention this, too.)
> 
> 
> Hello,
> 
> I introduced the rat call in the lagrange function to force rational
> arithmetic due to the inestability of floating point calculations in
> high degree polynomials.

I don't think using rat fixes the stability problem. :-)
> 
> We could add a boolean option, say force_rat, both in lagrange and
> ratinterpol, so that users can control whether to force rational
> arithmetic or not.

That's ok.  I don't really need it; I was just curious.

I was just playing around with minimax approximations (via Remez
exchange algorithm) and was looking at how ratinterpol works.

Thanks,

Ray