Error bounds in maximum entropy approximations

From MaRDI portal





Given \(N+1\) power moments of an unknown distribution \(P\), \(\mu_ n= \int_ 0^ 1 x^ n P(x)dx\), \(n=0,\dots, N\), one is often interested in averages of some function over the distribution, \(\langle F\rangle= \int_ 0^ 1 F(x)P(x) dx\). In the paper, a bound is derived on \(\delta F_ N= |\langle F\rangle- \langle F\rangle_ N|= | \int_ 0^ 1 F(x) (P(x)- P_ N(x)) dx|\), \(P_ N(x)\) being the approximate solution. The bound is given by an expression of the type \(\delta F_ N= | \sum_{n=N+1}^ \infty a_ n (\mu_ n- \widetilde{\mu}_ n)|\), where \(\widetilde{\mu}_ n\) are higher moments of the approximate solution. It is argued and documented by examples that often it is sufficient to compute a small number of higher order terms to get a very tight bound \(\delta F_ N\). Thus the method is advantageous in the cases where more moments are initially known than can be used in the solution algorithm due to numerical errors.











This page was built for publication: Error bounds in maximum entropy approximations

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4276813)