Penalizing null recurrent diffusions (Q456219)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Penalizing null recurrent diffusions |
scientific article |
Statements
Penalizing null recurrent diffusions (English)
0 references
23 October 2012
0 references
Let \(X= (X_t,\operatorname{P}_x)_{t\geq 0}\) be a linear regular null-recurrent diffusion in \(\mathbb{R}_+\) with 0 as instantaneously reflecting and \(+\infty\) as natural boundary, \(({\mathfrak F}_t)_{t\geq 0}\) the corresponding natural filtration, \(m\) the speed measure, \((L_a(t))_{t\geq 0}\) the normalized local time of \(X\) at a \[ L_a(t):= \lim_{\varepsilon\to\infty} \int^1_0 \mathbf{1}_{[a,a+\varepsilon[} (Xs)\,ds/m([a, a+\varepsilon[), \] and \((\tau_a(s))_{s\geq 0}\) the right-continuous inverse of \((L_a(t))_{t\geq 0}\). When \(X_a=a\), \((\tau_a(s))_{s\geq 0}\) is a subordinator. Denote its Lévy-measure by \(\nu^{(a)}\). Let \({\mathfrak L}\), respectively \({\mathfrak R}\), denote the class of probability measures \(\mu\) satisfying \(\lim_{t\to\infty}\mu([t+ x,\infty[)/\mu([, t\infty[)= 1\) for all \(x\in\mathbb{R}\), respectively, \(\mu([t,\infty[)\lim_{t\to\infty} t^{-\beta}\eta(t)\), \(\beta\geq 0\), \(\mu\) slowly varying. For \(a\geq 0\), \(g_a(t):= \sup\{u\leq t: X_u= a\}\), and let \((F_t)_{t\geq 0}\) be a positive, predictable process such that \[ 0<\operatorname{E}_x \int^t_0 F_u dL_a(u)<\infty. \] The author derives asymptotic expressions for \[ \int^t_0 \nu^{(a)}([s,\infty[)\,ds \] and \[ \operatorname{E}_x \int^t_0 F(g_a(s))\,ds, \] when \(\nu^{(0)}\in{\mathfrak L}\), and for \(\nu^{(a)}([t,\infty[)\) and \(\operatorname{E}_xF(g_a(t))\), when even \(\nu^{(0)}\in{\mathfrak R}\). If \(\nu^{(0)}\in{\mathfrak L}\) and \[ \Gamma_t:= \int^t_0 F(g_a(s)\,ds, \] or \(\nu^{(0)}\in{\mathfrak R}\), \(F\) decreasing, and \(\Gamma_t:= F(g_a(t))\), then the penalizing principle is satisfied by \(\Gamma= (\Gamma_t)_{t\geq 0}\), i.e., there exists a probability measure \(\mathbb{Q}^{(F)}_x\) (which is the same in both cases) such that for all \(s\geq 0\) and all \(\Lambda(s)\in{\mathfrak F}_s\), \[ \lim_{t\to\infty} \operatorname{E}_x\mathrm{\mathbf 1}_{\Lambda(s)}\Gamma_t/\operatorname{E}_x \Gamma_t= \mathbb{Q}^{(F)}_x(\Lambda(s)). \] The measure \(\mathbb{Q}^{(F)}_x\) is shown to be absolutely continuous respective \(\operatorname{P}_x\), the density given explicitly. Furthermore, defining \(g(a):= \sup\{s\geq 0: X_s= a\}\), \(g(a)\) is finite a.s. \([\mathbb{Q}^{(F)}_x]\), conditional to \(g(a)\), \((X_t)_{t\leq g(a)}\) and \((X_{g(a)+t})_{t\geq 0}\) are independent, \((X_{g(a)+u})_{u\geq 0}\) is transient, tends to \(+\infty\), as \(u\to\infty\), and its law does not depend on \(\Gamma\). Two examples are given and preceding penalization results by \textit{J. Najnudel}, \textit{B. Roynette} and \textit{M. Yor} [A global view of Brownian penalisations. Tokyo: Mathematical Society of Japan (2009; Zbl 1180.60004)] recovered as special case.
0 references
penalization
0 references
null recurrent diffusions
0 references
last passage times
0 references
inverse local time
0 references