Chudnovsky's formula adds a constant number of digits per iteration, or linear increase in precision. Gauss-Legendre doubles the number of digits per iteration, or quadratic convergence. Borwein's quadruples the number of digits per iteration, or quartic convergence.
Thus it's interesting that Chudnovsky is competitive, if not better, than Gauss-Legendre or Borwein in terms of CPU time in actual implementations.
I think binary splitting has something to do with it.
No comments :
Post a Comment