
Boost : 
From: Martin Bonner (martin.bonner_at_[hidden])
Date: 20050914 03:54:26
Original Message
From: Daryle Walker [mailto:darylew_at_[hidden]]
Sent: 14 September 2005 07:12
To: Boost mailing list
Subject: Re: [boost] Interest in an arbitrary precision library?
> On 9/13/05 7:28 AM, "Andy Little" <andy_at_[hidden]>
> wrote:
>
>> "Daryle Walker" <darylew_at_[hidden]> wrote
>>
>>> OK. When you say "arbitrary precision," you mean that a precision
>>> limit must be set (at runtime) before an operation. Most people
>>> use "arbitrary precision" to mean unlimited precision, not your
>>> "runtime cutoff" precision.
>>
>> Are there really libraries that have unlimited precision?
>> What happens when the result of a computation is irrational?
>
> You can't store it conventionally, since such numbers would need an
> infinite amount of memory. You would just give up and have to define
> some sort of rounding/cutoff philosophy. As another poster said,
> you could store the irrational components of a number with some sort
> of formula (but only for algebraically irrational numbers, not
> transcendentally irrational numbers).
Why only for algebraiclly irrational numbers? There is no reason (in
principle) that the representation of an irrational shouldn't include trig /
hyperbolic / bessel functions of irrationals, or even definite integrals.
Of course, at this point you get close to rewriting a chunk of Matlab or
Mathematica. On the other hand, simplifying
(sqrt(5)sqrt(3))*(sqrt(5)+sqrt(3)) to exactly 2 is quite a challenge too
(and if you don't then the unlimited precision representation doesn't really
buy you anything).
 Martin Bonner Martin.Bonner_at_[hidden] Pi Technology, Milton Hall, Ely Road, Milton, Cambridge, CB4 6WZ, ENGLAND Tel: +44 (0)1223 441434
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk