I can justify data interpolation in both human and mathematical terms. In human terms, the solution to a problem often follows from the adjoint operator, where the data space has enough known values. With a good display of data space, people often apply the adjoint operator in their minds. Filling the data space prevents distraction and confusion. The mathematical justification is that inversion methods are notorious for slow convergence. Consider that matrix-inversion costs are proportional to the cube of the number of unknowns. Computers balk when the number of unknowns goes above one thousand; and our images generally have millions. By extending the operator (which relates the model to the data) to include missing data, we can hope for a far more rapid convergence to the solution. On the extended data, perhaps the adjoint alone will be enough. Finally, we are not falsely influenced by the ``data not paid for'' if we adjust it so that there is no residual between it and the final model.

10/21/1998