Self-characteristic distributions (Q5926850): Difference between revisions

The author looks for probability densities that are, except for a multiplicative constant, equal to their characteristic functions. Well-known examples are the characteristic functions \(\exp[-u^{2}/2]\) and \(\cosh u\), corresponding to the standard normal and hyperbolic cosine distributions, respectively. He proves the validity of the following construction. Let \(f\) be a (real) square-integrable positive definite function, and let \(F\) be its symmetric Fourier transform. Then the function \(p\) defined by \(p(x)=f^{2}(x)+F\star F(x),\) with \(\star\) denoting convolution, is a self-Fourier transform. He conjectures that this construction yields all self characteristic functions, but does not have a proof. A number of examples is given. The conjecture would need the \(\sqrt{p}\) to be a characteristic function if \(p\) is a self-characteristic function, which seems hard to prove. For the two examples given above this property is evident, since the normal and the hyperbolic cosine distributions are infinitely divisible. Reviewer's remark: The author might have remarked that the set of self-characteristic functions is closed under (normed) positive linear combination. In the construction above one could also take \(f\) nonnegative and define \(p=f^{k}+F^{\star k}\) with \(k\) prime.