Approach: Use the existing programs COMP (which does the ray tracing), VSIM (which provides a convenient way of specifying the optical parameters), and NPSOLVE (a general program for nonlinear minimization), which run on ordinary computers, and convert them for parallel architectures, with special code for computing partial derivatives in parallel. Most of the time, with these problems, is spent in computing the partial derivatives needed for the iterative solution of the nonlinear adjustment. When numerical partial derivatives are used, they are computed by perturbing each parameter a small amount and computing the change in the objective function. Therefore, the adopted approach uses one processor for each parameter (or a few parameters if their number exceeds the number of processors) and computes the corresponding partial derivatives on this processor. Since typical optical problems have a large number of parameters (roughly equal to the typical number of processors), this approach produces a good fit of the problem to the available hardware.
Accomplishments: Portions of the program have been converted to run on the T3D. However, some Fortran features that were used in the existing program do not exist on the T3D. Therefore, C routines have been written to perform these computations.
Significance: The solution for optical parameters is important in the process of determining the source of aberrations of an optical system from an analysis of images taken through it. For example, the existing program was used in diagnosing the problem with the Hubble Space Telescope. However, the program involves a large amount of computing. Therefore, it runs very slowly on ordinary computers, and parallel computation is desired. Since the program NPSOLVE used in this computation does a general minimization, the results of this effort may be useful in many other fields in addition to optics.
Status/Plans: The needed C routines have been written. The conversion of the remaining Fortran code is proceeding. The parallel implementation of the numerical partial derivative computation remains to be done.
Point of Contact:
Donald B. Gennery
Jet Propulsion Laboratory
gennery@bryce.jpl.nasa.gov
(818)354-4315