Graph splitting algorithms for inconsistent feasibility problems

In this work, we study graph splitting algorithms for feasibility problems involving closed and convex sets whose intersection may be empty. We characterize their minimal displacement vector, which is known to be the limit of the differences between successive iterates, in terms of a least-squares problem measuring disagreement between the sets according to the graph structure. We then specialize to closed affine subspaces, proving strong convergence of the shadow sequence to a solution of such a least-squares problem whenever one exists. Our analysis extends classical Douglas-Rachford results on best approximation pairs to graph splitting algorithms.