Eigenvalues of euclidean distance matrices (Q1183169): Difference between revisions

Let \(x_ 1,x_ 2,\dots,x_ n\) be \(n\) points (\(n>1\)) in \(R^ d\) with \(\| x_ i-x_ j\|\geq\varepsilon\) for \(i\neq j\). It is proved that the matrix \(A=(\| x_ i-x_ j\|)_{i,j=1,\dots,n}\) has an inverse \(A^{-1}\), whose norm (as an operator on \(\ell^ n_ 2\)) verifies the inequality \(\| A^{-1}\|\leq c\sqrt{d}/\varepsilon\), where \(c\) is an absolute constant. It follows that \(\| A^{- 1}\|\| A\|\lesssim(D\varepsilon)^{d+1}\), where the sign \(\lesssim\) indicates inequality up to a constant depending upon \(d\) and \(D\) is an upper bound for the diameter of the set \(x_ 1\dots,x_ n\). From the obtained results one can deduce that the functions of the form (1) \(f(x)=\delta+\sum_ 1^ n \alpha_ j\| x-x_ j\|\), \(\delta\) a constant and \(\sum_ 1^ n \alpha_ j=0\), can be used in the interpolation with arbitrary data. One shows that the norm of the interpolation operator \(T: \ell_ \infty^ n\to L_ \infty(R^ d)\) by functions of the form (1) depends heavily on the number of points and/or the distribution of these points in \(R^ d\).