A bivariate local limit theorem (Q1173953): Difference between revisions

A random 2-vector in \(X\subset\mathbb{R}^{(2)}\) is said to have a stable distribution if there exist a sequence of i.i.d. vectors \(Y_ n=(Y_ n^{(1)},Y_ n^{(2)})\) and associated 2-vector sequences \(\{b_ n\}\), and \(\{a_ n\}\) with \(b_ n=\{b_ n^{(1)},b_ n^{(2)}\}\), \(a_ n=\{a_ n^{(1)},a_ n^{(2)}\}\), \(a_ n^{(i)}>0\), \(i=1,2\), with \(\{S_ n^{(1)}/a_ n^{(1)},S_ n^{(2)}/a_ n^{(2)}\}- b_ n\to X\) in law as \(n\to\infty\), where \(S_ n^{(i)}=\sum_{i=1}^ n Y_ n^{(i)}\), \(i=1,2\). Write \(S_ n=(S_ n^{(1)},S_ n^{(2)})\); if \(X\) is a non-degenerate stable random 2- vector, then its characteristic function is integrable over \(\mathbb{R}^{(2)}\). Let \(g\) be the p.d.f. of \(X\) and \(f(t)\) be the characteristic function of \(Y_ 1\). The following local limit theorems have been proved: (a) If \(Y_ 1\) has a 1-lattice distribution [i.e. \(f(t)\) has period \(2\pi\) in each variable, but \(| f(t)|<1\) for \(t\not\in\{(2m\pi,2n\pi):\;m,n=0,1,2,\dots\}]\), then, for every point \(x\) having integer coordinates, \(a_ n^{(1)}a_ n^{(2)}{\mathcal P}(S_ n=x)-g(x_ n)\to 0\), as \(n\to\infty\), uniformly in \(x\), where \(x_ n=(x^{(1)}/a_ n^{(1)},x_ n^{(2)}/a_ n^{(2)})\). (b) If \(Y_ 1\) has a non-lattice distribution, then \[ a_ n^{(1)}a_ n^{(2)}{\mathcal P}(x^{(i)}\leq S_ n^{(i)}\leq x^{(i)}+h^{(i)},\;(i=1,2)-h^{(1)}h^{(2)}g(x_ n)\to 0, \] as \(n\to\infty\), uniformly for \(x\in\mathbb{R}^{(2)}\) and \(h\in\) compact subsets of \(\mathbb{R}^ +\times\mathbb{R}^ +\). (c) If \(Y_ 1\) has a jointly 1-lattice and non-lattice [i.e. \(f(2m\pi,0)=1\) for \(m=0,\pm1,\pm2,\dots\) but \(| f(t)|<1\), otherwise] distribution, then \[ a_ n^{(1)}a_ n^{(2)}{\mathcal P}(S_ n^{(1)}=x^{(1)},x^{(2)}\leq S_ n^{(2)}\leq x^{(2)}+h)-hg(x_ n)\to 0, \] as \(n\to\infty\), uniformly for \(\mathbb{Z}\times\mathbb{R}\) and \(h\in\) compact subsets of \(\mathbb{R}^ +\).