Exact Moment Representation in Polynomial Optimization
Résumé
We investigate the problem of representing moment sequences by measures in the context of
Polynomial Optimization Problems. This consists in finding the infimum of a real polynomial on
a real semialgebraic set defined by polynomial inequalities. We analyze the exactness of Moment
Matrix (MoM) relaxations, dual to the Sum of Squares (SoS) relaxations, which are hierarchies
of convex cones introduced by Lasserre to approximate measures and positive polynomials. We
investigate in particular flat truncation properties, which allow testing effectively when MoM
exactness holds.
We consider the quadratic module Q generated by the inequalities. We show that the dual
of the MoM relaxation coincides with the SoS relaxation extended with the real radical of
the support of Q, and focus on the zero-dimensional case, generalizing results for equations
defining a finite real variety. We deduce sufficient and necessary conditions for flat truncation,
under the finite convergence assumption: flat truncation happens if and only if the support of
the quadratic module associated with the minimizers is of dimension zero. We also bound the
order of the relaxation at which flat truncation holds.
As corollaries, we conclude that flat truncation holds:
• when regularity conditions, known as Boundary Hessian Conditions, hold: this result
implies that flat truncation and MoM exactness holds generically;
• when the support of the quadratic module Q is zero-dimensional;
• in singular cases, flat truncation holds for the MoM relaxation extended with the polar
constraints when the real variety of polar points is finite.
Effective numerical computations illustrate these flat truncation properties.
Origine : Fichiers produits par l'(les) auteur(s)