Hardy's inequality and ultrametric matrices (Q1970431)

The inequality of G.H. Hardy referred to is the following. Let \(p>1\) and let \(a_{1},a_{2},\dots{}\) be nonnegative numbers, not all \(0\). Then, for all \(n\geq 1\): \[ \sum_{k=1}^{n}\left( \frac{a_{1}+a_{2}+\dots{}+a_{k}}{k}\right) ^{p}<\left( \frac{p}{p-1}\right) ^{p}\sum_{k=1}^{n}a_{k}^{p} \] where the constant \(( \frac{p}{p-1}) ^{p}\) is best possible. An \(n\times n\) symmetric ``ultrametric'' matrix is a real symmetric matrix which can be represented in the form \(A=\sum\tau_{k}u_{k}u_{k}^{\top}\) as a sum of at most \(2n-1\) terms where each \(\tau_{k}>0\) and where \(u_{1},u_{2},\dots{}\) are \(\{ 0,1\} \) (column) vectors which together span \(\mathbb{R} ^{n}.\) Using properties of the spectral radii of ultrametric matrices, the authors prove a generalization of Hardy's inequality when \(p=2.\) Namely, for all sequences \(\{ r_{k}\} \) of positive real numbers which satisfy an appropriate convergence criterion, there is an associated constant \(\gamma>0\) such that for all nonnegative \(a_{1},a_{2},\dots{}\) which are not all \(0\): \[ \sum_{k=1}^{n}\frac{(a_{1}+a_{2}+\dots{}+a_{k})^{2}}{r_{k}}<\gamma\sum_{k=1} ^{n}a_{k}^{2}. \] Their methods do not appear to apply in the cases where \(p\neq 2\).