[stephencanon]

https://en.wikipedia.org/wiki/Remez_algorithm

It’s a very simple iterative algorithm, essentially the dumbest thing that could possibly work (like most good algorithms). It fails to converge for functions that have poles nearby unless you have a very good initial guess (the Chebyshev or Carathéodory-Fejér approximants are ~always good starting points and easily computed). In practice you want to optimize a weighted L-inf norm rather than absolute, because floating-point errors are measured in a relative norm.