Faster fast evaluation of thin plate splines in two dimensions

A new method for fast evaluation of thin plate splines in two dimensions is presented. The paper first develops exponential approximations to thin plate splines. These are the analytical basis for an improved fast multipole evaluator. Analytic error bounds are supplemented by offline parallel numerical computation of the underlying error constants. These error constants enable adaptive selection of series lengths as a function of the weights associated with a source panel, and the desired accuracy. Numerical comparisons with a competing algorithm show that the new method is significantly faster when moderate to high precision is required.