Improving Inter-Block Backtracking with Interval Newton

Abstract : Inter-block backtracking (IBB) computes all the solutions of sparse systems of nonlinear equations over the reals. This algorithm, introduced by Bliek et al. (1998) handles a system of equations previously decomposed into a set of (small) k × k sub-systems, called blocks. Partial solutions are computed in the different blocks in a certain order and combined together to obtain the set of global solutions. When solutions inside blocks are computed with interval-based techniques, IBB can be viewed as a new interval-based algorithm for solving decomposed systems of non-linear equations. Previous implementations used Ilog Solver and its IlcInterval library as a black box, which implied several strong limitations. New versions come from the integration of IBB with the interval-based library Ibex. IBB is now reliable (no solution is lost) while still gaining at least one order of magnitude w.r.t. solving the entire system. On a sample of benchmarks, we have compared several variants of IBB that differ in the way the contraction/filtering is performed inside blocks and is shared between blocks. We have observed that the use of interval Newton inside blocks has the most positive impact on the robustness and performance of IBB. This modifies the influence of other features, such as intelligent backtracking. Also, an incremental variant of inter-block filtering makes this feature more often fruitful.
Type de document :
Article dans une revue
Constraints, Springer Verlag, 2010, 15 (1), pp.93-116. 〈10.1007/s10601-009-9072-5〉
Liste complète des métadonnées

Littérature citée [26 références]  Voir  Masquer  Télécharger
Contributeur : Gilles Chabert <>
Soumis le : mardi 8 octobre 2013 - 11:05:56
Dernière modification le : mardi 16 janvier 2018 - 14:36:38
Document(s) archivé(s) le : jeudi 9 janvier 2014 - 02:20:12


Fichiers éditeurs autorisés sur une archive ouverte



Bertrand Neveu, Gilles Trombettoni, Gilles Chabert. Improving Inter-Block Backtracking with Interval Newton. Constraints, Springer Verlag, 2010, 15 (1), pp.93-116. 〈10.1007/s10601-009-9072-5〉. 〈hal-00481324〉



Consultations de la notice


Téléchargements de fichiers