Dev Notes: 2017_01_16

Rational Numbers And Floating Point

Rational numbers can represent any of the floating point numbers that we can represent in C++ (single, or double), although they may be a little large. The advantage is that rationals don't have to lose precision in subsequent calculations, but there are a couple of problems.

The first problem is that rational numbers will typically be larger than floating point, but this isn't a major concern right now. The second, more significant, problem is that floating point numbers are often not particularly good approximations to more interesting rational numbers.

If we consider, say, the rational 1/10 (0.1) then there's no way to accurately represent it with a conventional power-of-2-based floating point number. We end up with a representation that is close, but not quite the same. The IEEE754 double precision form of 0.1 is actually closer to 0.1000000000000000055511151231257827021181583404541015625, encoded as 0x3fb999999999999a. When we convert this to a rational we get 0x1999999999999a/0x100000000000000, and this gets normalized to 0xccccccccccccd/0x80000000000000.

See Also

• Dev Notes: 2017_01_15