On the inverse Klain map (Q364190)

This paper is an important contribution to the theory of valuations on convex bodies. Let \(\mathrm{Val}\) denote the space of continuous translation-invariant valuations on the family \(\mathcal{K}(\mathbb{R}^n)\) of convex bodies in euclidean \(n\)-space. Thus, \(\phi \in \mathrm{Val}\) satisfies the valuation property \[ \phi(K \cup L) + \phi(K \cap L) = \phi(K) + \phi(L) \] whenever \(K,L \in \mathcal{K}(\mathbb{R}^n)\) are such that \(K \cup L \in \mathcal{K}(\mathbb{R}^n)\) also, \(\phi(K + t) = \phi(K)\) for \(K \in \mathcal{K}(\mathbb{R}^n)\) and \(t \in \mathbb{R}^n\), and continuity is with respect to the Hausdorff metric on \(\mathcal{K}(\mathbb{R}^n)\). In [Proc. Lond. Math. Soc., III. Ser. 35, 113--135 (1977; Zbl 0353.52001)] the reviewer showed that \(\phi\) admits an expression \[ \phi = \phi_0 + \cdots + \phi_n, \] where \(\phi_i \in \mathrm{Val}_i\), the subspace of those valuations which are homogeneous of degree \(i\), so that \(\phi_i(\lambda K) = \lambda^i\phi(K)\) for \(\lambda \geq 0\); this is called by the authors the McMullen decomposition. Since monotone valuations are continuous, this decomposition holds for them as well; moreover, \textit{A. Bernig} and \textit{J. H. G. Fu} [Ann. Math. (2) 173, No. 2, 907--945 (2011; Zbl 1230.52014)] proved that \(\phi\) is monotone if and only if each \(\phi_i\) is. The core result of this paper is that the \(\phi_i\) are not necessarily positive if \(\phi\) itself is. If \(\mathrm{Val}_i^+\) is the subspace of even valuations in \(\mathrm{Val}_i\), so that \(\phi(-K) = \phi(K)\) for \(\phi \in \mathrm{Val}_i^+\) and \(K \in \mathcal{K}(\mathbb{R}^n)\), denote by \(\phi_E\) the restriction of \(\phi\) to \(E \in \mathrm{Gr}_i\), the Grassmannian of \(i\)-dimensional subspaces. Then there is a proportionality factor \(\mathrm{Kl}_\phi(E)\) such that \[ \phi_E = \mathrm{Kl}_\phi(E)\mathrm{vol}_i, \] with \(\mathrm{vol}_i\) the usual \(i\)-dimensional volume on \(E\). Then \(\mathrm{Kl}_\phi: \mathbb{C}\) is the Klain function of \(\phi\), and \(\mathrm{Kl}: \mathrm{Val}_i^+ \to C(\mathrm{Gr}_i)\) is called the Klain map. In [\textit{D. A. Klain}, Trans. Am. Math. Soc. 352, No. 1, 71--93 (2000; Zbl 0940.52002)] it is proved to be injective. Continuity and monotonicity of the inverse Klain map are closely related. For given \(K \in \mathcal{K}(\mathbb{R}^n)\), the former of the two conditions \[ \mathrm{Kl}_\phi \geq 0 \implies \phi(K) \geq 0, \quad |\phi(K)| \leq C\|\mathrm{Kl}_\phi\|, \] with \(C\) depending on \(K\), implies the latter for all \(\phi \in \mathrm{Val}_i^+\). The latter condition holds for generalized zonoids, but if \(0 < i < n-1\) then there are centrally symmetric polytopes for which it fails. Indeed, if \(0 < i < n\) and \(\mathcal{G}(i)\) consists of those centrally symmetric \(K \in \mathcal{K}(\mathbb{R}^n)\) for which there is a signed Borel measure \(\mu_K\) on \(\mathrm{Gr}_i\) such that \[ \mathrm{vol}_i(K \mid E) = \int_{\mathrm{Gr}_i} \, \cos(E,F)\,d\mu_K(F), \] with \(K \mid E\) the orthogonal projection of \(K\) on \(E \in \mathrm{Gr}_i\), then this condition holds exactly when \(K \in \mathcal{G}(i)\). It is shown here that a polytope \(P \in \mathcal{G}(i)\) only when its \((i+1)\)-faces are centrally symmetric; in particular, the condition can fail. The former condition holds exactly for the subset \(\mathcal{K}(i)\) of \(\mathcal{G}(i)\) for which \(\mu_K\) is positive. Note that \(\mathcal{K}(1)\) consists of the zonoids, while \(\mathcal{K}(n-1) = \mathcal{G}(n-1)\) consists of all centrally symmetric convex bodies. Putting all this together, a consequence is that there exists a positive, even, translation-invariant, continuous valuation \(\phi\) on \(\mathcal{K}(\mathbb{R}^n)\) not all of whose homogeneous components are positive. Therefore, the Minkowski valuation \(\Phi : \mathcal{K}(\mathbb{R}^n) \to \mathcal{K}(\mathbb{R}^n)\) defined by \[ \Phi K = \phi(K)B^n, \] with \(B^n\) the unit ball, cannot be expressed as a sum of homogeneous components.