EQUIVALENCE OF NON-NEGATIVE RANDOM TRANSLATES OF AN IID RANDOM SEQUENCE

Abstract

Let $\bf X=\{X_k\}$ be an IID random sequence and $\bf Y=\{Y_k\}$ be an independent random sequence also independent of $\bf X$. Denote by $\mu_{\bf X}$ and $\mu_{\bf X+\bf Y}$ the probability measures on the sequence space induced by $\bf X$ and $\bf X+\bf Y =\{X_k+Y_k\}$, respectively. The problem is to characterize $\mu_{\bf X+\bf Y}\sim\mu_{\bf X}$ in terms of $\mu_{\bf Y}$ in the case where $\bf X$ is non-negative. Sato and Tamashiro [6] first discussed this problem assuming the existence of ${f_{\bf X}(x)=\frac{d\mu_{X_1}}{dx}(x)}$. They gave several necessary or sufficient conditions for $\mu_{\bf X+\bf Y}\sim\mu_{\bf X}$ under some additional assumptions on $f_{\bf X}$ or on $\bf Y$. The authors precisely improve these results. First they rationalize the assumption of the existence of $f_{\bf X}$. Then they prove that the condition (C.6) is necessary for wider classes of $f_\bf X$ with local regularities. They also prove if the $p$-integral $I_p^0(\bf X)\lt \infty$ and $\bf Y\in \ell_p^+$ a.s., then (C.6) is necessary and sufficient. Furthermore, in the typical case where $\bf X$ is exponentially distributed, they prove an explicit necessary and sufficient condition for $\mu_{\bf X+\bf Y}\sim \mu_\bf X$.