Range reduction using fixed points

National audienceThe efficiency of sBB depends on many parameters, among which the width of the variable ranges at each node. The fastest range reduction algorithm is called Feasibility-Based Bounds Tightening (FBBT) : it is an iterative procedure that propagates bounds up and down the expression trees [1] representing the constraints in (1), tightening them by using the constraint bounds (-?, 0]. Depending on the instance, and even limited to linear constraints only, FBBT may not converge finitely to its limit point. Tolerance-based termination criteria yield finite termination but, in general, in unbounded time (for every time bound, there is an instance exceeding it). So, although the FBBT is practically fast, its theoretical worst-case complexity status is far from satisfactory