next up previous print clean
Next: Diagonal weighting preconditioning Up: Practical implementation of ICO Previous: Log-stretch Fourier transform

Iterative solution for the pseudo-inverse

For each frequency component (or bandwidth) the systems of linear equations to be solved is still very large and one has to resort to iterative methods. This solves a huge set of simultaneous equations without the need to write down the matrix of coefficients. I use an iterative scheme based on the conjugate gradient solver, which produces a good result at a reasonable cost. Experience has shown that a satisfactory solution is achieved in less than 10 iterations. The conjugate gradient algorithm generates a sequence of approximate solutions whose computations each involve the application of the adjoint followed by the forward operator. The inner-products are essentially AMO transformations and their computation is therefore reasonably cheap.


next up previous print clean
Next: Diagonal weighting preconditioning Up: Practical implementation of ICO Previous: Log-stretch Fourier transform
Stanford Exploration Project
1/18/2001