Convexity properties of the cone of nonnegative polynomials (Q705127): Difference between revisions

In the linear space \(P_{n,2k}\) of all homogeneous polynomials of degree \(2k\) in \(n\) real variables the author considers the cone \(C\) consisting of non-negative polynomials and its subcones \(Sq\) consisting of sums of squares of polynomials of degree \(k\) and \(Lf\) consisting of sums of \(2k\)-powers of linear forms. Given a number \(m\geq k\) he also considers the convex cone \(K(2m)=\{f\in P_{n,2k}:(x_1^2+\dots+x_n^2)^{m-k}f\in Lf_{n,2m}\}\) of all polynomials whose restriction to the unit sphere is the sum of \(2m\)th powers of linear forms. Intersecting the cones with the hyperplane \(M=\{f\in P_{n,2k}:\int_{S^{n-1}}f\, d\sigma=1\}\) one obtains compact convex sets \(\overline{C}=C\cap M\), \(\overline{Sq}=Sq\cap M\) and \(\overline{Lf}=Lf\cap M\), \(\overline{K}(2m)=K(2m)\cap M\) that are bases of the cones \(C\), \(Sq\), \(Lf\), \(K(2m)\) respectively. In Theorem 7.8 the author gives a lower bound for the fraction of volumes \({\text{vol}\,\overline{K}(2m)}/{\text{vol}\,{\overline C}}\) implying that this fraction approaches 1 as \(m\) goes to infinity. This implies that all strictly positive polynomials lie in some \(K(2m)\), a result proved earlier by \textit{B. Reznik} [Contemp. Math. 253, 251--272 (2000; Zbl 0972.11021)]. In Theorem 5.3 it is shown that the John ellipsoid of \(\overline{C}\) is a ball of radius \((C^{2k}_{n+2k-1}-1)^{-1/2}\) centered at \((x_1^2+\dots+x_n^2)^k\). On the other hand, in Theorem 7.6 the author explicitly describes the Loewner ellipsoid of \(\overline{Lf}\). As a by-product of the proofs some interesting integral inequalities are discovered. For example, it is proved in Theorem 6.3 that \(\| f\| _\infty\leq C^k_{n+k-1}\cdot\| f\| _1\) for any non-negative polynomial \(f\in P_{n,2k}\).