Patching, creating and applying individual PEFs to sections of the model, and multigridding, incrementally decreasing the bin size to capture finer and finer features, may prove to be useful in filling in the missing data. Ultimately, a combination of the two may be even better.

If the calculated does prove to work sufficiently, it may be a good idea to implement a vector which could apply different values to different parts of the model.

Finally, since the helix is already being used to apply the regularization, preconditioning would reduce convergence time.

7/5/1998