[Haskell-cafe] Re: rounding errors with real numbers.

Henning Thielemann wrote:

Maybe you should use a kind of convex combination, that is

(x-oldLower)*a + (oldUpper-x)*b

Maybe lower*(1-z) + upper*z where z = (x-oldLower) / (oldUpper-oldLower). I think this will always map oldLower and oldUpper to lower and upper exactly, but I'm not sure it's monotonic. -- Ben