Closed mfkasim1 closed 11 months ago
If the solution exists, can we proof that the convergence always exists?
The perturbation theory only leads to fixed-point iteration of y = g[y] = L0_inv[f] - L0_inv_L[sigma(y)]
. mode=1
only leads to the naive fixed-point iteration which only works if the absolute eigenvalue of the Jacobian of g[y]
less than 1. mode=2
leads to linear mixing, which is a bit better, but not so much. With linear mixing, you can only converge if the largest eigenvalue of the Jacobian satisfies 1 < largest_eival < -(1 + eps) / (1 - eps)
. So, the convergence does not always exists. Unless maybe if we do some nonlinear preconditioning on the equation.
So many things happened after this and this issue becomes unrelated.