skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Nonlinear Smoothing and the EM Algorithm for Positive Integral Equations of the First Kind

Abstract

We study a modification of the EMS algorithm in which each step of the EMS algorithm is preceded by a nonlinear smoothing step of the form Nf-exp(S*log f) , where S is the smoothing operator of the EMS algorithm. In the context of positive integral equations (a la positron emission tomography) the resulting algorithm is related to a convex minimization problem which always admits a unique smooth solution, in contrast to the unmodified maximum likelihood setup. The new algorithm has slightly stronger monotonicity properties than the original EM algorithm. This suggests that the modified EMS algorithm is actually an EM algorithm for the modified problem. The existence of a smooth solution to the modified maximum likelihood problem and the monotonicity together imply the strong convergence of the new algorithm. We also present some simulation results for the integral equation of stereology, which suggests that the new algorithm behaves roughly like the EMS algorithm.

Authors:
 [1]
  1. Department of Mathematical Sciences, University of Delaware, Newark, DE 19716 (United States)
Publication Date:
OSTI Identifier:
21064290
Resource Type:
Journal Article
Resource Relation:
Journal Name: Applied Mathematics and Optimization; Journal Volume: 39; Journal Issue: 1; Other Information: DOI: 10.1007/s002459900099; Copyright (c) 1999 Springer-Verlag New York Inc.; Article Copyright (c) Inc. 1999 Springer-Verlag New York; Country of input: International Atomic Energy Agency (IAEA)
Country of Publication:
United States
Language:
English
Subject:
71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSICS; ALGORITHMS; CONVERGENCE; INTEGRAL EQUATIONS; MAXIMUM-LIKELIHOOD FIT; MINIMIZATION; NONLINEAR PROBLEMS; POSITRON COMPUTED TOMOGRAPHY; SIMULATION

Citation Formats

Eggermont, P. P. B. Nonlinear Smoothing and the EM Algorithm for Positive Integral Equations of the First Kind. United States: N. p., 1999. Web. doi:10.1007/S002459900099.
Eggermont, P. P. B. Nonlinear Smoothing and the EM Algorithm for Positive Integral Equations of the First Kind. United States. doi:10.1007/S002459900099.
Eggermont, P. P. B. Fri . "Nonlinear Smoothing and the EM Algorithm for Positive Integral Equations of the First Kind". United States. doi:10.1007/S002459900099.
@article{osti_21064290,
title = {Nonlinear Smoothing and the EM Algorithm for Positive Integral Equations of the First Kind},
author = {Eggermont, P. P. B.},
abstractNote = {We study a modification of the EMS algorithm in which each step of the EMS algorithm is preceded by a nonlinear smoothing step of the form Nf-exp(S*log f) , where S is the smoothing operator of the EMS algorithm. In the context of positive integral equations (a la positron emission tomography) the resulting algorithm is related to a convex minimization problem which always admits a unique smooth solution, in contrast to the unmodified maximum likelihood setup. The new algorithm has slightly stronger monotonicity properties than the original EM algorithm. This suggests that the modified EMS algorithm is actually an EM algorithm for the modified problem. The existence of a smooth solution to the modified maximum likelihood problem and the monotonicity together imply the strong convergence of the new algorithm. We also present some simulation results for the integral equation of stereology, which suggests that the new algorithm behaves roughly like the EMS algorithm.},
doi = {10.1007/S002459900099},
journal = {Applied Mathematics and Optimization},
number = 1,
volume = 39,
place = {United States},
year = {Fri Jan 15 00:00:00 EST 1999},
month = {Fri Jan 15 00:00:00 EST 1999}
}
  • In (1) the problem of the conditioning of matrices arising in the numerical solution of integral equations of the first kind by Galerkin and collocation schemes was investigated. A lower bound on the matrix condition number was found and its behavior as a function of the smoothness of the kernel of the original equation was studied. Some numerical results presented in (1) and much more extensive studies reported in (2) served to demonstrate the validity and usefulness of the theory. In these studies all basis functions were required to be orthonormal. Current computational methods frequently use non-orthonormal bases (for example,more » splines), and the question naturally arises as to whether such bases yield better conditions. Actually, in both (1) and (2) piecewise constant splines were examined as a special orthonormal set. They, of course, conformed to the general analytical theory. Computationally they usually proved more satisfactory than more classical orthonormal bases, providing both better conditioning and more satisfactory numerical results. In this paper the authors shall examine much more general bases including some that are spline-like. They show that in general little change is made in the analytical results found in (1). They also provide several corroborating numerical examples.« less
  • Recently effrots have been made to quantify the difficulties inherent in numerically sovling linear Fredholm integral equations of the first kind (J. Integral Equations, to appear.). In particular, the classical quadrature approach, collocation methods, and Galerkin schemes that make use of various orthonormal basis functions have been shown to lead to matrices with high condition numbers. In fact, it has been possible to obtain explicit lower bounds on these condition numbers as a function of the smoothness of the kernel, essentially independent of the choice of orthonormal basis. These bounds all approach infinity as the number of basis functions increases.more » In this article we present a numerical study of condition numbers arising from collocation and Galerkin methods with step-function and Legendre polynomial bases. The condition number for each kernel and basis set studied is exhibited as a function of the number of basis functions used. The effect that these ill-conditioned matrices have on the accuracy of solutions is demonstrated computationally. The information obtained gives an indication of the efficacy: and the dangers: of the collocation and Galerkin schemes in practical situations.« less
  • We construct and analyze for convergence a quadrature-iteration method for Volterra integral equations of the second kind and a quadrature-splitting method for linear equations. The iteration processes producing an approximate solution are accelerated, because the integral operator is approximated by a quadrature operator with an arbitrarily small residual operator.