Basis selection for SOS programs via facial reduction and polyhedral approximations

We develop a monomial basis selection procedure for sum-of-squares (SOS) programs based on facial reduction. Using linear programming and polyhedral approximations, the proposed technique finds a face of the SOS cone containing the feasible set of a given SOS program. The identified face in turn identifies a set of monomials that can be used to convert the SOS program into a semidefinite program (SDP). The technique can be viewed as a generalization of standard parsing algorithms for monomial basis selection. As we illustrate with examples, the proposed method can lead to smaller SDPs that are simpler to solve.