Quasistationary distributions for one-dimensional diffusions with singular boundary points (Q1999918): Difference between revisions

This paper treats the characterization of the existence of quasi-stationary distribution for diffusions on \((0, \infty)\) allowing singular behavior at 0 and \(\infty\). It is shown that, for the sure case of absorption at 0, there exists a quasi-stationary distribution as soon as the spectrum of the generator is strictly positive. This result complements results obtained by \textit{P. Cattiaux} et al. [Ann. Probab. 37, No. 5, 1926--1969 (2009; Zbl 1176.92041)] and \textit{M. Kolb} and \textit{D. Steinsaltz} [Ann. Probab. 40, No. 1, 162--212 (2012; Zbl 1278.60121)] for 0 being a regular boundary point. Moreover, this result also extends results obtained by Cattiaux et al. [loc. cit.] in the above, on singular diffusions. More precisely, the authors consider the one-dimensional diffusion \(( X_t )\), \(t \geq 0\), that is the solution to the stochastic differential equation \[ d X_t = d B_t - b(X_t) dt, \] where \(( B_t)\), \(t \geq 0\), is a standard one-dimensional Brownian motion and \(b\) : \((0, \infty) \to\mathbb{R}\) is the drift. Assume that the drift \(b\) is continuously differentiable on \((0, \infty)\), namely, \(b \in C^1(0, \infty)\). The first hitting time of 0 by \(( X_t )\), \(t \geq 0\), defined by \[ T_0 := \inf \{ \, t \geq 0: \,\, X_t = 0 \, \}. \] The density of the speed measure of (1) is given by \[ \rho(x) :=\exp \left\{ - \int_1^x \, 2 b(s) \, ds \, \right\}, \quad x \in (0, \infty). \] When \(\nu\) is a probability measure on \((0, \infty)\), then the Yaglom limit of \(( X_t )\), \(t \geq 0\), is defined by \[ \lim_{ t \to \infty} \, {\mathbb P}^{\nu} ( \, X_t \in ( \cdot ) \, \vert \, T_0 > t \, ). \] Assumption 1. Let \(z \in (0, \infty)\) and \(t > 0\). Then \[ e^{- t L} \mathbf{1}_{[0,z]} \in L^2( (0, \infty), \rho) \] and \[ \lim_{ \varepsilon \to 0} \, e^{-t L} \,\mathbf{1}_{ [ \varepsilon, z] } = e^{- t L} \, \mathbf{1}_{ [0,z]} \] in \[ L^2( (0, \infty), \rho) := \{ \, f : (0, \infty) \to {\mathbb R}; \,\, f \,\, \text{is measurable and satisfies} \,\, \int_0^{\infty} \, \vert f(x) \vert^2 \, \rho( dx) \, < \infty\, \}. \] The followings are the main results of this paper. Theorem 2. Suppose that 0 is an exit boundary, and \(\infty\) is a natural boundary. Under Assumption 1 we have the following classification: (i) If absorption is not certain and \(\lambda_0 > 0\), then \(( X_t ), t \geq 0\) escapes to infinity exponentially fast. (ii) If absorption is certain and \(\lambda_0 > 0\), then \(( X_t ), t \geq 0\) converges to a quasi-stationary distribution. This quasi-stationary distribution is the unique Yaglom limit and it attracts all the compactly supported initial distributions of \(X_0\). Theorem 3. If 0 is an exit boundary and \(\infty\) is an entrance boundary, then \(\lambda_0 > 0\) and \(( X_t )\), \(t \geq 0\), converges to the unique quasi-stationary distribution.