On Friday, 24 June 2016 at 21:11:52 UTC, Timon Gehr wrote:
By treating 0^0 consistently as 1, you never run into this kind of problem. Doesn't this demonstrate that they are doing it wrong? How would you design the notation?

You really need to look at the context. It is not uncommon that solvers favour total functions where even division is defined for "x / 0", for very pragmatic reasons. So even a total function for power could be useful in some contexts (i.e. considered defined not only for "0^0", but also "0^-1").

However, IEEE754-2008 is more principled. It separates pow into power over natural numbers "pown" (where 0^0==1) and power over inexact reals "powr". This is rather clear from the exception thrown at "powr(1.0,inf)", as the inexact nature of a "1.0" floating point could take it towards 0, 1 or inf.

And you cannot really know if inexact reals of "0.0^0.0" should go towards 0.0 or 1.0. It would make most sense to return the interval of potential values: [0,1]

Reply via email to