An update-and-stabilize framework for the minimum-norm-point problem

We consider the minimum-norm-point (MNP) problem of polyhedra, a well-studied problem that encompasses linear programming. Inspired by Wolfe’s classical MNP algorithm, we present a general algorithmic framework that performs first order update steps, combined with iterations that aim to ‘stabilize’ the current iterate with additional projections, i.e., finding a locally optimal solution whilst keeping the current tight inequalities. We bound the number of iterations polynomially in the dimension and in the associated circuit imbalance measure. In particular, the algorithm is strongly polynomial for network flow instances. The conic version of Wolfe’s algorithm is a special instantiation of our framework; as a consequence, we obtain convergence bounds for this algorithm. Our preliminary computational experiments show a significant improvement over standard first-order methods.