On Tue, Jan 19, 2010 at 4:50 PM, DIETER ENSSLEN <[email protected]> wrote:
> the regular 'pi' by two significant digits?!   ..5862   vs
>  ..5898 , the last figures to 14 places. A
> difference of  -3.6e-12.  J gives   ..5897 931.

http://www.eveandersson.com/pi/digits/1000000
claims this should be 5897 9328

Meanwhile, J uses 64 bit IEEE floating point
numbers, which are described at
http://en.wikipedia.org/wiki/IEEE_754-2008

In the context of pi, this means that J has 51
bits to represent the fractional part of pi.

(I think it is important to realize that J's numbers do not use
decimal fractions but use binary fractions instead.  Thus, for
example, 0.2 can only be represented approximately, even
though 0.0009765625 can be represented exactly.)

Anyways,

   10^.2^51
15.3525

We should expect pi in J to be accurate to 15 digits to the
right of the decimal point, and its sixteenth digit should not
be too inaccurate.

   x:<.o.10^17
314159265358979328

Ironically, J itself is more accurate in this case than the routine it
uses to display fractional numbers.  This might be a coincidence
(since those last two digits are somewhat spurious) but probably
also has something to do with the architecture of J's implementation..

FYI,

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

Reply via email to