next up previous print clean
Next: Noise and (f,x) interpolation Up: Convergence versus accuracy Previous: Testing


What is the best thing to do? Preconditioning is cheap via the helix, and gives the best results, if you twiddle the parameters enough. Damping helps reduce the sensitivity to number of iterations, in favor of the somewhat less sensitive $\epsilon$ parameter. Generally I skip the damping, but that choice is somewhat data dependent.

Stanford Exploration Project