Algrbra Package

[kmcollins]

I wrote a virtual CPU in Java a few years ago. You can do exactly correct math with a couple of registers for the number and one for the exponent. With 64 bits you can get 20 significant digits and the exponent scales it. Using addition and subtraction for multiplication and division provides exactly correct results all of the time. A 32-bit float, which is what Python uses, can have at most 10 significant digits, and multiplication and division, as implemented in the usual programming language or FPU, always produces error in the lowest order digit(s). Approximation of roots only requires an iterative refinement algorithm, like Newton's square root algorithm, to get to the level of accuracy desired.
You can and should rewrite it if you don't trust the author.