search

roystgnr / MetaPhysicL

Metaprogramming and operator-overloaded classes for numerical simulations
Other
22 stars 12 forks source link

Fix std::pow(0,b>1) derivatives #49

Closed roystgnr closed 5 years ago

roystgnr commented 5 years ago

We were previously trying to avoid an extra pow call by simply dividing the result of the original call by the base, but this gives 0/0 => NaN when the base is 0, even for the b > 1 case where the derivative should be well-defined.

Hopefully this fixes the aside @dschwen pointed out in #35.

This ends up forcing an extra std::pow call in the DualNumber case, which isn't great for efficiency, but isn't the yet-another-if_else efficiency disaster I was fearing.

roystgnr commented 5 years ago

Oh, hell, even this isn't enough, is it? When a == 0, b == 1, we'll be evaluating pow(0,0) and getting NaN instead of 1. Maybe we'll be stuck with another if_else after all.

Still, this seems to be a solid upgrade over the previous code, so let's move to it for now.