A random continued fraction in \(\mathbb{R}^{d+1}\) with an inverse Gaussian distribution (Q1907521)

The authors generalize gamma (resp. inverse Gaussian) distribution, noticed \(\gamma_{\lambda, b}\) (resp. \(\mu_{\lambda, a,b})\), to \(]0,\infty [\times \mathbb{R}^d\): the density of \(\gamma_{\lambda, b}\), resp. \(\mu_{\lambda, a, b}\), with respect to \(dx_0 d\vec x\), is \[ \begin{aligned} f(x_0, \vec x) &= c_{\lambda, b} x_0^{\lambda- d/2 -1} \exp \Biggl\{ - {{\vec x\cdot \vec x} \over {2x_0}} - {{b_0 x_0} \over 2}+ \vec b\cdot \vec x\Biggr\}, \qquad \lambda>0,\\ \text{resp.} f(x_0, \vec x) &= c_{\lambda, a,b} f(x_0, \vec x) \exp (-a/ 2x_0\}, \qquad \lambda\in \mathbb{R}, \end{aligned} \] where \(a>0\) and \(b^*= b_0- \vec b\cdot \vec b>0\). These distributions satisfy: \(\gamma_{\lambda, b} * \mu_{-\lambda, a,b}= \mu_{\lambda, a,b}\), \(\lambda\) being positive. Moreover if \(\lambda\in \mathbb{R}\), the image of \(\mu_{\lambda, b^*,c}\) by \(h(\cdot; \vec b, \vec c)\) is \(\mu_{-\lambda, c^*,b}\), \(h(x_0, \vec x; \vec b, \vec c)= ({1\over x_0}, -\vec b+ {{\vec x+ \vec c} \over x_0})\), \(c^* >0\). Let \((Y_n)\) be a sequence of independent r.v. such that \({\mathcal L} (Y_{2n+1})= \gamma_{\lambda, b}, {\mathcal L} (Y_{2n})= \gamma_{\lambda,c}\). The authors consider the right random walk on \(]0,\infty [\times \mathbb{R}^d: Z_n (x)= F_1\circ F_2\circ \dots\circ F_n (x)\), where \(F_n(x)= h(Y_{2n}+ h(Y_{2n-1}+ x; \vec c, \vec b); \vec b, \vec c)\). Then the main result is the following: \(Z_n(x)\) converges a.s., the limit does not depend on \(x\) and its distribution is \(\mu_{-\lambda, c^*,b}\). As a result, the authors prove that if \(X\), \(Y_1\) and \(Y_2\) are independent r.v. on \(]0, \infty]\times \mathbb{R}^d\), and \({\mathcal L} (X)= {\mathcal L} (F_1 (X))\), then \({\mathcal L} (X)= \mu_{-\lambda, c^*, b}\).