Tight bounds on the exponential approximation of some aging distributions (Q1099522)

In certain applications, the exponential distribution (Exp) is used to approximate, in a suitable metric, some unknown distributions for nonnegative random variables, describing ``age''. Let F be the d.f. of \(X\geq 0\), where \(E(X)=1\), \(E(x^ 2)<\infty\). If \[ E(X-y| X>y)\leq (\geq)E(X)=1, \] then X or its d.f. F is termed new better (new worse) than used in expectation, denoted NBUE (NWUE). Letting \(\rho =| E(x^ 2/2)-1|\), the following tight bounds are obtained for the metric: \(\Delta (F_ 1,F_ 2)=\sup_{x}| F_ 1(x)-F_ 2(x)|.\) Theorem: If F is NBUE, then \(\Delta (F,Exp)\leq 1-\exp (- (2\rho)^{1/2}),\) and if F is NWUE, then \(\Delta (F,Exp)\leq (\rho^ 2+2\rho)^{1/2}- \rho.\) Only real variable methods are used in getting the extremal distributions. Similar methods establish the corresponding results for harmonic NBUE and NWUE classes of d.f.'s.