Box consistency through Adaptive Shaving

Alexandre Goldsztejn (1) , Frédéric Goualard (1, 1)

1 LINA - Laboratoire d'Informatique de Nantes Atlantique

The canonical algorithm to enforce box consistency over a constraint relies on a dichotomic process to isolate the leftmost and rightmost solutions. We identify some weaknesses of the standard implementations of this approach and review the existing body of work to tackle them; we then present an adaptive shaving process to achieve box consistency by tightening a domain from both bounds inward. Experimental results show a significant improvement over existing approaches in terms of robustness for difficult problems.

Mots clés: local consistency; constraint programming; Newton method

Domaines: Analyse numérique [cs.NA]

Alexandre Goldsztejn, Frédéric Goualard. Box consistency through Adaptive Shaving. 25th Symposium On Applied Computing, Mar 2010, Sierre, Switzerland. pp.2049-2054. ⟨hal-00428678⟩