General Perturbation Resilient Dynamic String-Averaging for Inconsistent Problems with Superiorization

Kay Barshad and Yair Censor Department of Mathematics, University of Haifa, Mt. Carmel, Haifa 3498838, Israel ✉ [email protected] ; ✉ [email protected] · 2025

Read General Perturbation Resilient Dynamic String-Averaging for Inconsistent Problems with Superiorization by Kay Barshad and Yair Censor Department of Mathematics, University of Haifa, Mt. Carmel, Haifa 3498838, Israel ✉ [email protected] ; ✉ [email protected] (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-00717.txt.

Introduction

Kay Barshad and Yair Censor extend an iterative method for inconsistent problems in a 2025 paper. String-averaging methods project onto sets in a flexible order. When the sets share no common point, older theory falls silent.

Their general scheme handles that inconsistent case directly. Convergence is proven with a notion called strong coherence. The method also tolerates bounded noise at each step. A steering idea called superiorization can ride on top. That pairing can push solutions toward extra goals, like smoothness. Readers in imaging and optimization get guarantees for problems that have no exact answer.

Catalog record

CREATORKay Barshad and Yair Censor Department of Mathematics, University of Haifa, Mt. Carmel, Haifa 3498838, Israel ✉ [email protected] ; ✉ [email protected]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE78 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-00717

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

General Perturbation Resilient Dynamic String-Averaging for Inconsistent Problems with Superiorization Kay Barshad and Yair Censor Department of Mathematics, University of Haifa, Mt. Carmel, Haifa 3498838, Israel ✉ [email protected] ; ✉ [email protected] Abstract In this paper we introduce a General Dynamic String-Averaging (GDSA) iterative scheme and investigate its convergence properties in the inconsistent case, that is, when the input operators don’t have a common fixed point. The Dynamic String-Averaging Projection (DSAP) algorithm itself was introduced in an 2013 paper, where its strong convergence and bounded perturbation resilience were studied in the consistent case (that is, when the sets under consideration had a nonempty intersection). Results involving combination of the DSAP method with superiorization, were presented in 2015. The proof of the weak convergence of our GDSA method is based on the notion of “strong coherence” of sequences of operators that was introduced in 2019. This is an improvement of the property of “coherence” of sequences of operators introduced in 2001 by Bauschke and Combetts. Strong coherence provides a more convenient sufficient convergence condition for methods that employ infinite sequences of operators and it turns out to be a useful general tool when applied to proving the convergence of many iterative methods. In this paper we combine the ideas of both dynamic string-averaging and strong coherence, in order to analyze our GDSA method for a general class of operators and its bounded perturbation resilience in the inconsistent case with weak and strong convergence. We then discuss an application of the GDSA method to the Superiorization Methodology, developing results on the behavior of its superiorized version. Communicated by Fabian Flores-Bazàn 2010 Mathematics Subject Classification: 46N10, 46N40, 47H09,47H10, 47J25, 47N10, 65F10, 65J99. Keywords and phrases: Approximately shrinking operator, bounded perturbation resilience, bounded regularity, coherence, coherent sequence of operators, common fixed point problem, convex feasibility problem, dynamic string-averaging, Fejér monotonicity, metric projection, nonexpansive operator, strong coherence, strongly coherent sequence of operators, superiorization, weak convergence, weak regularity. 1 Introduction A strongly convergent Dynamic String-Averaging Projection (DSAP) algorithm, based on convex combinations and compositions of operators, was introduced, along with its bounded perturbation resilience, by Censor and Zaslavski in [ 22 ] for a family of nonempty, closed and convex sets \left\{C_{i}\right\}_{i=1}^{m} , where the considered operators were given by a sequence of metric projections \left\{P_{C_{i}}\right\}_{i=1}^{m} in the consistent case, that is, when \cap_{i=1}^{m}C_{i}\not=\emptyset . A superiorized version of this algorithm appeared in [ 23 ] , where it was proved that any sequence generated by it either converges to a constrained minimum point of the objective function employed in the superiorization process, or that it is strictly Fejér monotone with respect to a subset of the solution set of the constrained minimization problem. String-averaging projection (SAP) methods form a general algorithmic framework introduced in [ 19 ] . Subsequently, they were developed in a variety of situations such as for convex feasibility with infinitely many sets [ 27 ] , for incremental stochastic subgradient algorithms [ 24 ] and for proton computed tomography image reconstruction [ 6 ] , to name but a few. See also [ 2 ] , where perturbation resilience of such methods was further studied and [ 1 ] , where linear convergence rates for a certain class of extrapolated fixed point algorithms which are based on dynamic string-averaging methods were established. In the present paper we consider more general ideas, focusing on more general algorithmic structures which are applied to the inconsistent case (that is, when the input operators don’t have a common fixed point), and when the objective function, involved in the superiorization, is only required to be convex and continuous. For a positive integer m and a given family of nonexpansive operators \left\{U_{i}\right\}_{i=1}^{m} , U_{i}:\mathcal{H}\rightarrow\mathcal{H} for each i=1,2,\dots,m , without a common fixed point, we construct a General Dynamic String-Averaging (GDSA) algorithm (Algorithm 4.5 below) based on convex combinations, compositions and relaxations of the operators of the aforementioned family. In Theorem 4.8 below we investigate weak and strong convergence properties of our GDSA algorithm and its bounded perturbation resilience. To this end we introduce a “ \limsup -admissible sequence of operators” (Definition 4.2 below) and show that if, after a certain GDSA procedure, the sequence of output operators is \limsup -admissible, then the algorithm converges weakly (and under additional assumptions – strongly) to a point in a certain (assumed nonempty) set, even though the given input operators don’t have a common fixed point. We further apply our GDSA algorithm to the Superiorization Methodology (SM). The SM is not aiming to solve the constrained minimization problem under consideration, but to find a feasible point for the original feasibility-seeking problem which is “superior”, i.e., has smaller or equal objective function value than that of a point returned by the feasibility-seeking only algorithm that is employed by the SM. More details about the SM appear below in Subsection 5.1. The weak convergence of our GDSA algorithm and its superiorized version is based, inter alia, on the theory of coherence, which was introduced by Bauschke and Combetts in [ 9 ] and further developed by Barshad, Reich and Zalas in [ 5 ] , see also [ 3 ] and [ 4 ] . All the results below, concerning our algorithmic schemes, remain valid in the consistent case as well. However, since it is generally desirable to assume certain properties of the input operators, which is possible in this case as it was shown in [ 29 ] , such results are of limited interest in the consistent case due to the assumptions placed on the output operators in our GDSA procedure. It is worth noting the significance of the inconsistent case at this point. The convex feasibility problem (CFP) is to find a feasible point in the intersection of a family of convex and closed sets. If the intersection is empty, then the CFP is inconsistent and a feasible point does not exist. However, algorithmic research on inconsistent CFPs does exist and is mainly focused on two directions. One is oriented toward defining other solution concepts that will apply, such as proximity function minimization, wherein a proximity function measures, in some way, the total violation of all constraints (see, for instance, Example 4.10 below). The second direction investigates the behavior of algorithms that are designed to solve a consistent CFP when applied to inconsistent problems. This direction is fueled by situations wherein one lacks a priori information about the consistency or inconsistency of the CFP or does not wish to invest computational resources to get hold of such knowledge prior to running his algorithm. A telegraphical review of some recent works on inconsistent CFPs appears in [ 21 ] . The rest of the paper is organized as follows. In Section 2 we provide some background which is needed to establish our results. In Section 3 we introduce the general bounded regularity with its properties and discuss the well-known notion of approximate shrinking of operators. In Section 4 we present the properties of our GDSA algorithm. Notations and initial tools appear in Subsection 4.1 and lemmata leading to the main result are in Subsection 4.2 . In Section 5 , we present applications of this algorithm to the Superiorization Methodology, followed by

Read the complete text