The recovering of a probability density function f(x) from its Mellin trans
form M(s) is considered. The approximate f(M)(x) is chosen resorting to max
imum entropy technique constrained by the first derivatives of M(s) evaluat
ed at s = 1. So the basic properties of a probability density are saved.
Existence conditions of the approximate f(M)(x), entropy-convergence and th
en L-1-norm convergence are proved. Some numerical examples are reported. R
esorting to the Mellin transform is an alternative to Laplace one, as the r
ecovered probability distribution is heavy-tailed, or equivalently its prob
ability density function has abscissa convergence Laplace transform equal t
o 0. (C) 2001 Elsevier Science Inc. All rights reserved.