Floating point is only useful to the extent that it approximates real arithmetic.