The BlurFit system models laboratory measurements of blur images obtained with a focal plane imaging array using a series of Zernike polynomials to represent the wavefront distortion in the exit pupil of an optical system. This is a difficult analysis to perform correctly and the use of functional programming together with libraries written to handle multidimensional arrays and vectorized arithmetic has proven to be effective in framing the problem correctly and succinctly. The modeling uses a variant of the Levenberg-Marquardt minimization procedure, combined with monitoring of the Eigenvalue spectrum of the local curvature matrix to overcome near degeneracies in the model specification.

BlurFit exploits functional programming in two ways. First, the use of a functional technique limits the spread of nasty side effects that are so prevalent in a Fortran-like language that use large COMMON blocks to share data. Second, the elegant data structures available in an ML language makes the process of data entry and reporting much easier. Modelers can incorporate new fitting terms by simply adding equations for their terms to an input specification, no code changes are necessary. The vectorized math routines are predominantly imperative for speed of execution, but this is kept beneath the surface in a well modularized library. External glue routines were written in C to incorporate a high-performance 2-D FFT routine, and for numerous Math Recipes routines for matrix inversion and Eigen-analysis. Visualization of the fitting process is provided by means of an RPC connection to RSI/IDL. The speed of the system is roughly 70% of an equivalent written entirely in C, and twice as fast as one written in RSI/IDL, but the code is inestimably more maintainable in its functional form.

Also see the homepage: