Projecting onto a Capped Rotated Second-Order Cone

Abstract: This paper establishes a closed-form expression for projecting onto a capped rotated second-order cone. This convex set arises in the perspective relaxation of mixed-integer nonlinear programs (MINLP) with binary indicator variables. The rapid computation of the projection onto this set is expected to enable the development of effective methods for solving the continuous relaxation of MINLPs whose feasible region may involve a Cartesian product of a large number of such sets. The closed-form established herein consists of seven cases, one of which is a solution of a cubic equation and another is a solution of a quartic equation. Although quartic equations possess closed-form solutions, numerical solutions are typically used in practice. Based on bounds that we prove using additional case analysis, we develop a specialized bisection-based method to solve the resulting quartic equation. In experiments we first demonstrate that the projection problem is solved faster and more accurately with our closed-form, together with a standard polynomial equation solver, compared with a general state-of-the-art interior-point solver and compared with a state-of-the art conic first-order method solver. We also demonstrate the efficacy of our bisection-based specialized numerical method for solving the quartic equation.