Anonymous user
Talk:Sequence of non-squares: Difference between revisions
→Stability, Accuracy
(→algorithm: new section) |
|||
Line 34:
::: Well, using that (k+1)^2 = k^2 + 2k + 1, you can even completely eliminate all multiplications. However the stated goal of the task (see top of discussion page) was "to show how easy it is to investigate functions in a programming language." So while this algorithm is certainly a solution of the task (well, printing out some values is missing) avoiding the numeric problems, I'm not sure if it isn't against the idea of the task. Maybe a clarification of the task would be a good idea. --[[User:Ce|Ce]] 07:36, 3 September 2008 (UTC)
:::: Try and implement something '''close to''' the given function that is being investigated, at least at first. --[[User:Paddy3118|Paddy3118]] 18:03, 3 September 2008 (UTC)
:IEEE754 guarantees almost 7 decimal digits (6.9something iirc) for normal 32-bit floats and the last computer architecture I've ever heard of that used non-IEEE754 floats was the DEC VAX. Certainly anything x87-based should abide by that. Since log(2) is about 0.3, this means square roots should be stable and OK for this task up to about 10^6.6 or there abouts. Since we're only supposed to test a million numbers, I see no problem here anywhere (if someone feels like testing it, it should break down a little over 4 million for 32-bit floats, a little over 2e15 for 64-bit floats). Or am I missing something here somewhere?
[[User:Sgeier|Sgeier]] 21:02, 28 September 2009 (UTC)
== Zero ==
| |||