English

Semidefinite Representation of Convex Sets

Optimization and Control 2008-07-21 v5 Algebraic Geometry

Abstract

Let S={x\ren:g1(x)0,...,gm(x)0}S =\{x\in \re^n: g_1(x)\geq 0, ..., g_m(x)\geq 0\} be a semialgebraic set defined by multivariate polynomials gi(x)g_i(x). Assume SS is convex, compact and has nonempty interior. Let Si={x\ren:gi(x)0}S_i =\{x\in \re^n: g_i(x)\geq 0\}, and \bdS\bdS (resp. \bdSi\bdS_i) be the boundary of SS (resp. SiS_i). This paper discusses whether SS can be represented as the projection of some LMI representable set. Such SS is called semidefinite representable or SDP representable. The contributions of this paper: {\bf (i)} Assume gi(x)g_i(x) are all concave on SS. If the positive definite Lagrange Hessian (PDLH) condition holds, i.e., the Hessian of the Lagrange function for optimization problem of minimizing any nonzero linear function Tx\ell^Tx on SS is positive definite at the minimizer, then SS is SDP representable. {\bf (ii)} If each gi(x)g_i(x) is either sos-concave (2gi(x)=W(x)TW(x)-\nabla^2g_i(x)=W(x)^TW(x) for some possibly nonsquare matrix polynomial W(x)W(x)) or strictly quasi-concave on SS, then SS is SDP representable. {\bf (iii)} If each SiS_i is either sos-convex or poscurv-convex (SiS_i is compact convex, whose boundary has positive curvature and is nonsingular, i.e. gi(x)0\nabla g_i(x) \not = 0 on \bdSiS\bdS_i \cap S), then SS is SDP representable. This also holds for SiS_i for which \bdSiS\bdS_i \cap S extends smoothly to the boundary of a poscurv-convex set containing SS. {\bf (iv)} We give the complexity of Schm\"{u}dgen and Putinar's matrix Positivstellensatz, which are critical to the proofs of (i)-(iii).

Keywords

Cite

@article{arxiv.0705.4068,
  title  = {Semidefinite Representation of Convex Sets},
  author = {J. William Helton and Jiawang Nie},
  journal= {arXiv preprint arXiv:0705.4068},
  year   = {2008}
}
R2 v1 2026-06-21T08:32:41.608Z