The paper investigates a mathematical property called infinite divisibility for certain random variables. A random variable is infinitely divisible if it can always be written as the sum of any number of independent, identically distributed random variables. The central objects of study are positive stable random variables, a well-known family of probability distributions with heavy tails, raised to arbitrary real-number powers. The question is: for which powers does the resulting random variable remain infinitely divisible?
The main result provides a complete and clean answer. If you take a positive stable random variable and raise it to a power p, the result is infinitely divisible if and only if p is either positive or sufficiently negative (specifically, more negative than a threshold that depends on the stability parameter of the original distribution). For powers that fall in between, the property fails. This completes a picture that was only half known before: the case of negative powers had been characterized by earlier researchers, and this paper handles the positive powers and assembles both halves into one unified statement.
The proof strategy combines several tools from probability theory. A key ingredient is a factorization result due to Kanter, which breaks a stable random variable into simpler components. The authors also establish a useful general result: if you multiply the sum of a random variable with a constant by an independent random variable drawn from a broad class of distributions, the product is guaranteed to be infinitely divisible. This product property is interesting in its own right and does the heavy lifting in the proof, connecting the structure of stable distributions to the theory of infinitely divisible laws through a family of distributions called generalized gamma convolutions.