On Fri, Aug 4, 2017 at 11:44 AM, Steve D'Aprano <steve+pyt...@pearwood.info> wrote: > On Sat, 5 Aug 2017 01:47 am, Ian Kelly wrote: > >> Here's a much smaller upper bound: >> >>>>> n = 2 ** 53 >>>>> s = isqrt_newton(n) >>>>> n >> 9007199254740992 >>>>> s >> 94906265 >>>>> m = (s+1)**2 - 1 >>>>> m >> 9007199326062755 >>>>> isqrt_newton(m) == isqrt_float(m) >> False > > Oooh, interesting. How did you get that? By luck, or do you have some reason > for > picking (s+1)**2 - 1?
Partly luck. I reasoned that a mismatch would be due to incorrect rounding, which would most likely happen just before a perfect square. So I tried the value one less than the first perfect square after 2 ** 53 and it happened to work. I also tried the value two less, which did not. In any case, Chris's finding beats mine by a factor of two. -- https://mail.python.org/mailman/listinfo/python-list
