An approximation to infinitely divisible laws (Q1897518): Difference between revisions

Let \(F_{\theta, \sigma, L, R}(x)\) be the infinitely divisible distribution function whose characteristic function is given by (Lévy formula): \[ \int^\infty_{- \infty} e^{itx} dF_{\theta, \sigma, L,R}(x)= \exp\Biggl\{i\theta t- {\sigma^2\over 2} t^2+ \int^0_{- \infty} K(t, x) dL(x)+ \int^\infty_0 K(t, x) dR(x)\Biggr\}, \] where \(K(t, x)= e^{itx}- 1- itx/(1+ x^2)\), \(\theta\in \mathbb{R}\), \(\sigma\geq 0\), the functions \(L(.)\) and \(R(.)\) satisfy some known conditions and are taken so that \(L(- \infty)= 0\), \(R(\infty)= 0\). Let \[ \psi_L(u):= \inf\{x< 0: L(x)> u\},\quad \psi_R(u):= \inf\{x< 0: -R(x)> u\} \] be the inverse functions, where the infimum of the empty set is taken to be zero. Let \(Z\) be a standard normal random variable and \(S^{(M)}_n= \sum^n_{k= 1} Y^M_k\), \(M= L\), \(R\), where all the random variables \(Z\), \(Y^{(M)}_k\) are independent and \(Y^{(M)}_k\) are all exponentially distributed on \((0, \infty)\). The author considers the problem of approximating \(F_{\theta, \sigma, L, R}(.)\) by the distribution functions \[ F^{n, m}_{\theta, \sigma, L, R}(x)= P\{V^{(L)}_n+ \sigma Z- V^{(R)}_m+ \theta- \theta_L+ \theta_R< x\}, \] where \[ V_n(M):= \sum^n_{j= 1} \psi_M(S^{(M)}_j)- \int^{S^{(M)}_n}_1 \psi_M(u) du, \] \[ \theta_M:= \int^1_0 {\psi_M(s)\over 1+ \psi^2_M(s)} ds \int^\infty_1 {\psi^3_M(s)\over 1+ \psi^2_M(s)} ds,\qquad M= L, R. \] The main result estimates the Lévy distance \(D_{n, m}(L, R)\) between \(F^{n, m}_{\theta, \sigma, L, R}(.)\) and \(F_{\theta, \sigma, L, R}(.)\). This distance depends upon how fast the functions \(\psi_L(u)\) and \(\psi_R(u)\) approach zero as \(u\to \infty\). The result looks particularly nice when the spectral (Lévy) measure in the Lévy formula is finite: \(L(0-)< \infty\) and \(R(0+)> - \infty\): \[ D_{n, m}(L, R)\leq {[L(0-)]^n\over n!}+ {[- R(0+)]^m\over m!}\qquad\text{for all} m,n\in \mathbb{R}. \] The paper is concluded by a number of examples.