The central limit theorem for Chébli-Trimèche hypergroups (Q1822824): Difference between revisions

Let A be a suitable \(C^ 2\)-function on \({\mathbb{R}}_+\) with \(A'/A\) decreasing and \(A'/A\to 2\rho\) for \(x\to \infty\). A defines a hypergroup structure on \({\mathbb{R}}_+\) (Chébli-Trimèche-hypergroup), such that the dual hypergroup is given by the solutions \(\phi_{\lambda}\) of the second-order differential equations \[ \phi ''_{\lambda}+(A'/A)\phi '_{\lambda}+(\rho^ 2+\lambda^ 2)\phi_{\lambda}=0,\quad \phi_{\lambda}(0)=1,\quad \phi '_{\lambda}(0)=0, \] with \(\lambda\in \hat K:={\mathbb{R}}_+\cup i[0,\rho]\). There exists a ``Gaussian'' semigroup of probabilities \((\alpha_ t)\) which is defined by the (generalized) Fourier transform. Let \((X_ n)\) be a sequence of independent \({\mathbb{R}}_+\) valued random variables. Then the convolution products of the corresponding probability distributions define a generalized random walk. (Convolution with respect to the hypergroup structure). Expectation and variance is defined according to the convolution structure. Then for an i.i.d. sequence \((X_ n)_{n\geq 1}\) with finite variance the generalized random walks with normalization \((N^{-1/2}X_ n)_{n=1...N}\) converge to a (generalized) Gaussian distribution \(\alpha_ t\). On the other hand, let \(S_ n\) be the generalized random walk defined by \((X_ n)\) and assume \(\rho >0\), then \(n^{-1/2}S_ n\) converges with a suitable centering to a normal distribution N(0,\(\sigma)\) (independent of the special hypergroup structure). If \(\rho =0\) a similar result holds if normal distributions are replaced by Raleigh distributions.