Back to Volume
Paper: The MATPHOT Algorithm for Digital Point Spread Function CCD Stellar Photometry
Volume: 281, Astronomical Data Analysis Software and Systems XI
Page: 387
Authors: Mighell, Kenneth J.
Abstract: Most CCD stellar photometric reduction packages use analytical functions to represent the stellar Point Spread Function (PSF). These PSF-fitting programs generally compute all the major partial derivatives of the observational model by differentiating the volume integral of the PSF over a pixel. Real-world PSFs are frequently very complicated and may not be exactly representable with any combination of analytical functions. Deviations of the real-world PSF from the analytical PSF are then generally stored in a residual matrix. Diffraction rings and spikes can provide a great deal of information about the position of a star, yet information about such common observational effects generally resides only in the residual matrix. Such useful information is generally not used in the PSF-fitting process except for the final step involving the determination of the chi-square goodness-of-fit between the CCD observation and the model where the intensity-scaled residual matrix is added to the mathematical model of the observation just before the goodness-of-fit is computed. I describe some of the key features of my MATPHOT algorithm for digital PSF-fitting CCD stellar photometry where the PSF is represented by a matrix of numbers. The mathematics of determining the partial derivatives of the observational model with respect to the x and y direction vectors is exactly the same with analytical or digital PSFs. The implementation methodology, however, is quite different. In the case of digital PSFs, the partial derivatives can be determined using numerical differentiation techniques on the digital PSFs. I compare the advantages and disadvantages with respect to traditional PSF-fitting algorithms based on analytical representations of the PSF. The MATPHOT algorithm is an ideal candidate for parallel processing. Instead of operating in the traditional single-processor mode of analyzing one pixel at a time, the MATPHOT algorithm can be written to operate on an image-plane basis which lends itself quite naturally to parallel processing. The increase in computational speed can be enormous given sufficient computational power. The first public release of the C source code of the single-processor version of the MATPHOT algorithm will be available in the fourth quarter of 2001 at the official MATPHOT web site of MATPHO . Plans for the future development of the parallel-processing version of the MATPHOT algorithm using Beowulf clusters are described.
Back to Volume