I should probably also note that xdiv(take(sc(-(k+pn)),p),q,XMFLR) is not exactly normal division.
In some cases, it's taking the floor after dividing by xbase^x (for some x). But in other cases (and this seems to be important), it's adding leading "zeros" which presumably extend the precision of the result somewhat like multiplying by xbase^x (otherwise 1%q would produce a zero result). Anyways, it seems like this should be simple, but I'm still struggling with it. Thanks, -- Raul On Thu, Oct 13, 2022 at 1:19 PM Raul Miller <[email protected]> wrote: > > On Thu, Oct 13, 2022 at 12:29 PM Henry Rich <[email protected]> wrote: > > I don't understand the question: what are you trying to compute? Doesn't > > gmp support division? > > Sure, and mpq_get_d() documented at > https://gmplib.org/manual/Rational-Conversions fails J's accuracy > tests, with off-by-least-significant-bit errors. > > I've got most of those worked out, but I still fail on this test. > > > Knuth has code for division IIRC. > > It's not just division, it's the stability of the algorithm when > discarding precision. > > Probably I just need to find a relevant reference on numerical analysis. > > Thanks, > > -- > Raul ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
