On Wed, Jan 31, 2018 at 4:36 PM, William Tanksley, Jr
<[email protected]> wrote:
> Raul Miller <[email protected]> wrote:
>> But which approximation holds the most error?
>
> The naive Stern-Brocot traversal will take you only to fractions which
> minimize certain error properties _relative to the size of the numerator
> and denominator_. A bigger numerator and denominator will naturally allow
> for less error, but obviously they have their own expenses.

Sure, but what was interesting here was that this traversal brought me
to a numerator and denominator which were smaller than the x: result
and yet at the same time had less error than the x: result.

   }.0.1,5419351r1725033 1285290289249r409120605684-pi
2.21448e_14 _1.49116e_13

That's an error value about 15% of the magnitude of the x: error
value, using approximately 60% of the digits of the x: representation.

That looks very nice, if it's not just a coincidence.

Thanks,

-- 
Raul
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to