Strongly polynomial-time algorithm for over-constraint resolution: efficient debugging of timing constraint violations

A system of binary linear constraints or difference constraints (SDC) contains a set of variables that are constrained by a set of unary or binary linear inequalities. In such diverse applications as scheduling, interface timing verification, real-time systems, multimedia systems, layout compaction, and constraint satisfaction, SDCs have successfully been used to model systems of both temporal and spatial constraints. Formally, SDCs are modeled by weighted, directed (constraint) graphs. The consistency of an SDC means that there is at least one instantiation of its variables that satisfies all its constraints. It is well known that the absence of positive cycles in a graph implies the consistency of the corresponding SDC, so the consistency can be decided in strongly polynomial time. If a SDC is found to be inconsistent, it has to be repaired to make it consistent. This task is equivalent to removing positive cycles from the corresponding graph. All the previous algorithms for this task take time proportional to the number of positive cycles in the graph, which can grow exponentially. In this paper, we propose a strongly polynomial-time algorithm, i.e., an algorithm whose time complexity is polynomial in the size of the graph. Our algorithm takes in a graph and returns a list of edges and the changes in their weights to remove all the positive cycles from the graph. We experimentally quantify the length of the edge list and the running time of the algorithm on large benchmark graphs. We show that both are very small, so our algorithm is practical.