Two special-purpose iterative algorithms for maximization of Burg's entropy function subject to linear inequalities are presented. Both are "row-action" methods which use in each iteration the information contained in only one constraint. One is an underrelaxed Bregman algorithm which requires, at each iterative step, the solution of a system of equations. In contrast with this, the second algorithm employs a closed-form formula for the iterative step. Complete analyses of the convergence for both algorithms are given.
Bibliographical noteFunding Information:
Part of this work was done during a visit of Y. Censor to the Instituto de Matematica Pura e Aplicada (IMPA) in Rio de Janeiro, Brazil. We thank Professor Lindolpho de Carvalho Dias, Director of IMPA, for making this visit possible. The research of Y. Censor has been also supported by NIH grant HL-28438 while visiting the Medical Image Processing Group (MIPG) at the Department of Radiology. Hospital of the University of Pennsylvania, Philadelphia, PA.
ASJC Scopus subject areas
- Numerical Analysis
- Computational Mathematics
- Applied Mathematics