Entropies of sums of independent gamma random variables (Q6161612): Difference between revisions

Let \(X_1,X_2,\ldots\) be independent and identically distributed gamma random variables, with parameter \(\gamma>0\) and density function \(\Gamma(\gamma)^{-1}x^{\gamma-1}e^{-x}\) for \(x>0\). The authors establish conditions under which the function \[ (a_1,\ldots,a_n)\mapsto\mathbb{E}\Phi\left(\sum_{j=1}^n\sqrt{a_j}(X_j-\gamma)\right) \] is either Schur-convex or Schur-concave on \(\mathbb{R}^n_+\), for a function \(\Phi\). Results are also established for similar functions without the centering of the \(X_j\). These are applied to give bounds on the Shannon and Rényi entropies of \(\sum_{j=1}^n\sqrt{a_j}X_j\) for \(a_j\) such that \(a_1+\cdots+a_n=1\), showing that these are maximized with the choice \(a_j=1/n\). The authors also establish upper and lower bounds on the density function of \(\sum_{j=1}^n\sqrt{a_j}X_j\), considering separately the cases \(1/2\leq\gamma<1\) and \(\gamma\geq1\).