On circumcentered direct methods for monotone variational inequality problems

Abstract: The variational inequality problem (VIP) plays a central role in the theory and applications in continuous optimization. In particular, minimization problems and KKT systems can be regard as VIPs. In this work, we present the first methods using circumcenter iterations for solving VIPs. The circumcentered-reflection method (CRM) is a tool based on projections developed with the aim of finding a point in the intersection of finitely many closed convex sets. CRM has gone through enhancements and adaptations over the last few years and was proven to be faster in many settings than competitors such as alternating projections and the Douglas-Rachford method. One of the nice features of CRM is that it is able to deal with approximate projections, which is exactly what we enforced in this article theoretically and numerically. We present both a circumcenter method for the VIP with paramonotone operator and monotone operator, the first employing exact projections and the second approximate ones. Convergence results showing their convergence are established. Numerically, our experiments indicate good performance, including advantages over the well-know extragradient method.