A study of regularization by monotonicity (Q2366294): Difference between revisions

In order to overcome the non-uniqueness of solutions \(p(t)\) of integral equations of the type \(\int_ 0^ 1 K(s,p(t))dt=r(s)\), \(0\leq s\leq 1\), the authors propose to use a ``decreasing rearrangement'' approach developed by \textit{H. W. Engl}, the second author and \textit{H. Zeisel} [A decreasing rearrangement approach for a class of ill-posed integral equations. Institutsbericht 447, Joh.-Kepler-Univ. Linz (Austria), Inst. f. Math. (1991)], amounting to determine instead of \(p(t)\) the ``distribution function'' \(x(t)\) equal to the Lebesgue measure of the set of points \(\tau\) with \(p(\tau)>t\). This function \(x(t)\) is monotonically decreasing and obeys (if it exists) under certain conditions a linear integral equation of the first kind. The authors describe an approximation method of combining the midpoint quadrature rule with quadratic optimization (the constraints being given by the requirement of monotonicity), and they prove a theorem on its convergence for refinement of step-size. By displaying results of a case study they illustrate the effects of discretization in case of exact data. Finally, they discuss, for the case of randomly perturbed data, some variants of the Tikhonov regularization with monotoncity constraint and again give visual demonstration of applicability.