Weak drifts of infinitely divisible distributions and their applications (Q376248)

Let \(ID\), resp. \(ID_0\) denote the convolution semigroups of infinitely divisible laws on \(\mathbb{R}^d\), resp. the infinitely divisible laws without Gaussian component. The second characteristic function, i.e., the logarithm of the Fourier transform of \(\mu\in ID_0\) is given in the form \[ L_\mu: y\mapsto \int_{\mathbb{R}^d\backslash \{0\}}\left(\mathrm{e}^{i\langle y,x\rangle}-1-i\langle y,x\rangle 1_{B_1}(x)\right) d\nu_\mu(x) + i\langle y,\gamma_\mu\rangle, \] where \(B_1\) denotes the closed unit ball and \(\nu_\mu\) the Lévy measure. If \(\int_{B_1}|x|d\nu_\mu(x)<\infty\) (*), then \(L_\mu\) is representable as \[ L_\mu(y)=\int_{\mathbb{R}^d\backslash \{0\}}\left(\mathrm{e}^{i\langle y,x\rangle}-1\right) d\nu_\mu(x) + i\langle y,\gamma_\mu^0\rangle \] (\(\gamma_\mu^0\) denoting the drift ). And analogously, if \(\int_{\complement B_1}|x|d\nu_\mu(x)<\infty \) (**), then \[ L_\mu(y)= \int_{\mathbb{R}^d\backslash \{0\}}\left(\mathrm{e}^{i\langle y,x\rangle}-1-i\langle y,x\rangle \right) d\nu_\mu(x) + i\langle y,m_\mu\rangle \] (\(m_\mu\) denoting the mean). There exists a mapping \(\mu\mapsto\mu'\) on \(ID_0\), called `inversion' (in fact, a mapping between infinitesimal generators of the corresponding convolution semigroups), roughly spoken, exchanging the masses of Lévy measures at \(0\) and at \(\infty\) by the inversion at the sphere \(\{|x|=1\}\). Precisely, \(\nu_{\mu'}(B):=\int 1_B(|x|^{-2} x)|x|^2 d\nu_\mu(x)\) and \(\gamma_{\mu'} := -\gamma_\mu +\int_{|x|=1} x d\nu_\mu(x)\). For properties of this inversion and applications the reader is referred e.g. to the first named author's previous investigations [Lect. Notes Math. 2001, 1--91 (2010; Zbl 1222.60019); ALEA Lat. Am. J. Probab. Math. Stat. 8, 1--17 (2011; Zbl 1276.60016); J. Theor. Probab. 26, No. 4, 901--931 (2013; Zbl 1301.60018)] and the references mentioned there. If the integrability conditions (*), resp. (**) are not supposed to be satisfied, the authors define more generally `weak means' by \(m_\mu:=\gamma_\mu + \lim_{a\to\infty}\int_{1<|x|\leq a} x d\nu_\mu(x)\) (assuming that this limit exists), and analogously `weak drifts' \(\gamma_\mu^0\) if \(\lim_{\epsilon \to 0}\int_{\epsilon<|x|\leq 1}x d\nu_\mu(x)\) exists. Furthermore, the authors define `absolute' versions of weak means and weak drifts in a natural way. A key result for the following (Theorem 2.4) says that the inversion maps probabilities \(\mu\) with (absolute) weak mean to probabilities \(\mu'\) with (absolute) weak drift and conversely. Section 3 is concerned with stochastic integral maps and their conjugates, defined in the afore mentioned investigations, \(\Phi_{f_h}: \rho\mapsto \mathcal{L}\left(\int_0^c f_h dX_s^{(\rho)}\right)\), for suitable functions \(f_h\) and Lévy processes \(\left(X_s^{(\rho)}\right)\) with distribution \(\rho\) at time \(s=1\). (A precise definition is too technical to be given in a review.) The ranges of these maps \(\Phi_{f_h}\) and of their `conjugates' are closely related to probabilities with weak means and weak drifts. Section 4 is concerned with ranges of iterates of such stochastic integral mappings, and Section 5 with weak laws of large numbers and Shtatland's theorem for Lévy processes \(\left(X_s^{(\mu)}\right)\), comparing the existence of limits \(\lim_{\epsilon\to 0}\mathcal{L}\left((\epsilon^{-1} X_\epsilon^{(\mu)}\right)=\delta_c\) and \(\lim_{t\to \infty}\mathcal{L}\left((t^{-1} X_t^{(\mu)}\right)=\delta_{-c}\), and the relations with the existence of weak means and weak drifts.