Upper and Lower Bounds in Exponential Tauberian Theorems

Abstract

In this text we study, for positive random variables, the relation between the behaviour of the Laplace transform near infinity and the distribution near zero. A result of De Bruijn shows that $E(e^{-\lambda X})\sim \exp (r{\lambda }^{\alpha })$ for $\lambda \to \infty$ and $P(X\le \epsilon )\sim \exp (s/{\epsilon }^{\beta })$ for $\epsilon \downarrow 0$ are in some sense equivalent (for $1/\alpha =1/\beta +1$) and gives a relation between the constants $r$ and $s$. We illustrate how this result can be used to obtain simple large deviation results. For use in more complex situations we also give a generalisation of De Bruijn's result to the case when the upper and lower limits are different from each other.

slightly more general result, add some more references