Talk:Sequence of non-squares

I chose this as a way to show how easy it is to investigate functions in a programming language. --Paddy3118 08:42, 24 August 2008 (UTC)

Stability, Accuracy

The formula need to be investigated for numeric stability. Calculation of sqrt is inexact, it need to be shown that for all n in question, floor(1/2 + sqrt(n)) yields the exact result. A minimal requirement for this (though insufficient, I guess) is that sqrt has an error below 0.5. Note that the addition following to squaring will normalize for big n. Also conversion of those to floating point becomes quickly inexact when 32-bit floats are used. --Dmitry-kazakov 09:51, 24 August 2008 (UTC)

I suggest they choose/use the precision that gives correct results over the stated range, or reduce the range to suit the precision they have and add a comment. I think its true that most languages on Rosetta supporting floating point also support more than 32 bit FP values. --Paddy3118 11:28, 24 August 2008 (UTC)