Nonlinear Krylov Acceleration (NKA) is a method for accelerating the convergence of fixed-point (Picard) iterations. Many Newton-like and inexact Newton methods are fixed point iterations. The NKA project provides the canonical implementation of the method for several programming languages. The black-box accelerator is simple to integrate into existing code. Placed in the iteration loop, it observes the sequence of solution updates and replaces them with improved updates using information it has gleaned from previous solution iterates.
It was only recently recognized (2011) that NLK is essentially equivalent to Anderson Acceleration for a specific choice of mixing parameter. NLK was independently devised in the late 1980's using a very different approach, and though it leads to the same algebraic method, NLK's organization is somewhat different, and arguably superior. The NLK approach also provides clear rationale for the proper choice of Anderson's arbitrary mixing parameter.
Be the first to post a review of NKA!