The weighted logarithmic mean

From MaRDI portal





Based on the integral formula \(L(x, y)= \int^ 1_ 0 x^ t y^{1- t} dt\) for the logarithmic mean in two variables, the author generalizes it to several variables by defining \[ L(\mu; x_ 1, x_ 2,\dots, x_ n)= \int x^{t_ 1}_ 1 x^{t_ 2}_ 2\cdots x^{t_ n}_ n d\mu(t). \] He shows that this is homogeneous of degree 1 in \((x_ 1,\dots, x_ n)\) and lies between the weighted geometric and arithmetic means with weights \[ w_ k= \int t_ k d\mu(t)\qquad (k= 1,\dots, n). \] Here \(t_ 1+ t_ 2+\cdots+ t_ n= 1\), \(0\leq t_ k< 1\), the integration is done with respect to \(t= (t_ 2,\dots, t_ n)\) and thus on the simplex defined by \(t_ k\geq 0\) \((k= 2,\dots, n)\), \(t_ 2+\cdots+ t_ n\leq 1\), while \(\mu\) is a probability measure on that simplex. He then offers confluent hypergeometric functions as particular cases and derives, mainly by using numerical quadrature formulae with great virtuosity, inequalities and limiting formulas for these means. The latter state that the limits of \(L(cu_ 1,\dots, cu_ n; x_ 1,\dots, x_ n)\) as \(c\) tends to 0 or \(\infty\) are the weighted geometric and arithmetic means, respectively, of \(x_ 1,\dots, x_ n\) with the weights \(u_ 1,\dots, u_ n(c, u_ 1,\dots, u_ n> 0, u_ 1+\cdots+ u_ n= 1)\).




Cited in
(30)








This page was built for publication: The weighted logarithmic mean

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1343975)