Representation of nonnegative convex polynomials (Q944212)

A main topic in real algebraic geometry is the representation of polynomials positive on a basic semialgebraic set \[ K := \{x \in {\mathbb R}^n \, | \, g_1 (x) \geq 0, \ldots, g_m (x) \geq 0 \} \subset {\mathbb R}^n \] where \(g_j \in {\mathbb R}[X]\). A celebrated result in this area is the Schmüdgen Positivstellensatz which states that if \(K\) is compact and \(f\) is a polynomial positive in \(K\) then \(f\) belongs to the preordering \(P(g)\) generated by the \(g_j\)'s. Another important result, due to Putinar, states that under some additional assumption the polynomial \(f\) belongs to the quadratic module \(Q(g)\). In this paper the author specialize these representation results when \(f\) is a convex polynomial and the \(g_j\)'s are concave, so that \(K\) is a (not necessarily compact) convex basic semialgebraic set, which is an important case in optimization. To be more precise, let us define the subset \[ Q_c(g) := \left\{ \sigma + \sum_1^m \lambda_j g_j \, | \, \lambda \in {\mathbb R}^m_+, \, \sigma \in \Sigma^2, \, \sigma \text{ convex} \right\} \subset Q(g), \] where \(\Sigma^2\) denotes the cone of polynomials which are sums of squares of polynomials. The set \(Q_c(g)\) is a specialization of the quadratic module \(Q(g)\) to the convex case, as the weights associated to the \(g_j\)'s are nonnegative scalars and the polynomial \(\sigma\) is convex. The main result is the following: let \(F \subset {\mathbb R}[X]\) be the convex cone of convex polynomials nonnegative on \(K\) and suppose that there exists \(z \in K\) such that \(g_j(z) > 0\), \(\forall j = 1, \ldots, m\). Then \(Q_c(g) \cap F\) is dense in \(F\) for the \(l_1\)-norm of coefficients. In particular, if \(K = {\mathbb R}^n\) then \(\Sigma^2 \cap F\) is dense in \(F\).