Positive definite metric spaces (Q369652): Difference between revisions

The initial point of the paper is the notion of \textit{magnitude} of a metric space introduced by \textit{T. Leinster} in the very interesting paper [``Metric spaces. Post at \textit{The n-Category Café}'', (2008). {\texttt{http://golem.ph.utexas.edu/category/2008/02/metric\_spaces.html}}], see also [``The magnitude of metric spaces'', \url{arXiv:1012.5857}], which is a numerical invariant of metric spaces, analogous to the cardinality of finite sets or the Euler characteristic of topological spaces. From the author's abstract: ``This paper develops the theory of a class of metric spaces, positive definite metric spaces, for which magnitude is more tractable than in general. Positive definiteness is a generalization of the classical property of negative type for a metric space, which is known to hold for many interesting classes of spaces. It is proved that all the proposed definitions of magnitude coincide for compact positive definite metric spaces and further results are proved about the behavior of magnitude as a function of such spaces. Finally, some facts about the magnitude of compact subsets of \(\ell^n_p\) for \(p\leq 2\) are proved, generalizing results of Leinster for \(p=1,2\) using properties of these spaces which are somewhat stronger than positive definiteness.'' Let \((X,d)\) be a metric space. A \textit{weight measure} for \((X,d)\) is a finite signed measure \(\mu\) on \(X\) such that \[ \int_X e^{-d(x,y)}d\mu(y)=1 \tag{1} \] for every \(x\in X\). If \(X\) possesses a weight measure \(\mu\), then the \textit{magnitude} of \(X\) is defined as \(\mu(X)\). A metric space \(X\) is called positive definite if, for every finite \(A\subseteq X\), the matrix \(\xi_A\in\mathbb R^A\) with \(\xi_A(x,y)=e^{-d(x,y)}\) is positive definite. For the case of finite \(X\), equality (1) can be written as \[ \sum_{y\in X} e^{-d(x,y)}w(y)=1. \tag{2} \] A vector \(w\in\mathbb R^X\) is called a \textit{weighting} for \(A\) if (2) holds for every \(x\in X\). Denote by \(M(X)\) the space of finite signed Borel measures on \(X\) and define \[ Z_X(\mu,\nu)=\int_X\int_X e^{-d(x,y)}d\mu(x)d\nu(y) \] for \(\mu,\nu\in M(X)\). { Theorem 1.} Let \((X,d)\) be a compact positive definite metric space. The supremum \[ |X|:=\sup\Big\{ \frac{\mu(X)^2}{Z_X(\mu,\mu)}: \mu\in M(X),\;Z_X(\mu,\mu)\neq0 \Big\} \] is achieved for a measure \(\mu\) if and only if \(\mu\) is a nonzero scalar multiple of a weight measure for \(X\). If \(\mu\) is a weight measure for \(X\), then \(|A|\) equals the magnitude of \(X\). { Theorem 2.} For every compact positive defined metric space \(X\) the equality \[ |X|=\sup\{|A|:A\subseteq X\text{ and } card A<\infty\} \] holds. It is also proved that the function \(A\mapsto|A|\) is lower semicontinuous with respect to Gromov-Hausdorff convergence on the class of compact positive definite metric spaces. Some conditions for positive definiteness of metric spaces are found. { Theorem 3.} The following conditions are equivalent for every metric space \((X,d)\) (i) The metric space \((X,td)\) is positive definite for every \(t>0\). (ii) \((X,d)\) is of negative type, i.e. \((X,d^{\frac12})\) is isometric to a subset of a Hilbert space.