Quasilimiting behavior for one-dimensional diffusions with killing (Q662429): Difference between revisions

Let \(k,b: (0,\infty )\rightarrow (0,\infty )\) be functions, \(b\) continuous and integrable on every \([0,x]\). For \(\alpha \in [0,\infty ]\), let \[ (L^{k,\alpha }\psi )(x)=-(1/2)\psi ''(x)-b(x)\psi '(x)+k(x)\psi (x) \] be a positive self-adjoint operator on \(L^{2}(\Gamma)\) (\(\Gamma\) the measure of speed); its domain depends on \(\alpha\) via the Dirichlet form \(q^{k,\alpha }\) corresponding to \[ L^{k}=-\frac{1}{2\gamma}\frac{d}{dx}\gamma \frac{d}{dx}+k, \] where \(\gamma (x)=e^{2\int \chi_{[0,x]}(s)b(s)ds}\). Let \(\varphi(\lambda,\cdot)\) be the solution of \[ (L^{k}-\lambda )\varphi (\lambda ,\cdot )=0,\;\varphi (\lambda ,0)=\frac{1}{1+\alpha},\;\varphi '(\lambda ,0)= \frac{\alpha}{1+\alpha}. \] Denote by \(\lambda_{0}^{k}\) the infimum of the spectrum of \(L^{k}\). A process \(X_{t}\) is said to converge from the initial distribution \(\nu\) to the quasistationary distribution \(\varphi\) on compacta if, for \(z>0\), \(A\subset [0,z]\), we have \[ \lim_{t\rightarrow\infty }\operatorname{P}_{\nu }(X_{t}\in A\mid X_{t}\leq z)= \frac{\int_{_{A}}\varphi(y)\gamma (y)dy}{\int_{_{0}}^{^{z}}\varphi (y)\gamma (y)dy}. \] If \[ \lim_{t\rightarrow \infty }\operatorname{P}_{\nu }(X_{t}\in A\mid \tau_{\partial }>t)=\frac{\int_{_{A}}\varphi (y)\gamma (y)dy}{\int_{_{0}}^{^{\infty }}\varphi (y)\gamma (y)dy} \] for all \(A\), the denominator being finite, then the above expression without the term ``on compacta'' is used. The authors prove the following theorems. The killed diffusion \(X_{t}\) (i.e., corresponding to \(L^{k}\)) converges on compacta to the quasisationary distribution with density proportional to \(\varphi(\lambda_{0}^{k},\cdot)\) from any initial distribution with compact support in \((0,\infty)\). This result is preceded by \[ \lim_{s\rightarrow \infty }\left(\frac{p(t+s,x,y)}{p(s,a,a)}\right)=\frac{e^{-\lambda_{0}^{k}t} \varphi (\lambda_{0}^{k},x)\varphi (\lambda_{0}^{k},y)}{\varphi (\lambda_{0}^{k},a)^{2}}. \] If \(\infty\) is an entrance boundary, then the spectrum of \(L^{k}\) is discrete. If \(\liminf_{x\rightarrow \infty }k(x)>\lambda_{0}^{k}\) then, for every \(U\), \[ \lim_{t\rightarrow \infty }e^{\lambda_{0}^{k}t}\operatorname{P}^{x}(X_{t}\in U\mid \tau_{\partial }>t)=u_{\lambda_{0}^{k}}(x)\int_{_{U}}u_{\lambda_{0}^{k}}(y)\gamma (y)dy, \] where \(u_{\lambda_{0}^{k}}\in L^{2}(\Gamma )\) is the \(\lambda_{0}^{k}\)-eigenfunction and \(X_{t}\) corresponds to the Dirichlet form \(q^{k}\) of \(L^{k}\), converges to the quasistationary distribution \(u_{\lambda_{0}^{k}}\). Suppose that \(K=\lim_{x\rightarrow \infty }k(x)\) exists. If \(\infty\) is a natural boundary, \(K<\lambda_{0}^{k}\) and \(\int \gamma (x)^{-1}dx=\infty\), then \(X_{t}\), started from an initial distribution with compact support in \((0,\infty )\), converges to the quasistationary distribution with \(\Gamma\) density \(\varphi (\lambda_{0}^{k},\cdot )\), \(\lim_{t\rightarrow \infty }\operatorname{P}_{\nu }(\tau_{\partial }>t+r\mid \tau_{\partial }>t)\) exists and is represented as \(e^{-\lambda_{0}^{k}r}\). If \(K <\lambda_{0}^{k}\) and \(\int\gamma(x)^{-1}dx<\infty\), then \(X_{t}\) escapes to infinity and \[ \limsup_{t\rightarrow \infty }t^{-1}\log \operatorname{P}_{t}(X_{t}\leq z\mid \tau_{\partial }>t)=-(\lambda_{0}^{k}-K). \] If \(0\) is regular and \(\infty\) is an entrance boundary, then \(\lambda_{0}^{k}\) is isolated in the spectrum and \(u_{_{\lambda_{0}}k}\geq 0\) and for every initial distribution \(\nu\) with compact support \(\subset(0,\infty)\), \(X_{t}\) converges to the distribution having \(\Gamma\) density \(u_{\lambda_{0}{k}}/\int u_{\lambda_{0}{k}}(x)\gamma (x)dx\). If \(0\) is regular and \(\infty\) inaccessible, \(\alpha \in (0,\infty ]\), \(k=0\), \(\operatorname{P}_{x}(T_{0}<\infty )=1\), then there exists a quasistationary distribution if and only if, for some \(\varepsilon >0\), \(\operatorname{E}_{x}(e^{\varepsilon T_0})<\infty\). There exists a unique such distribution if and only if, for every \(a>0\), \(\sup \operatorname{E}_{x}(e^{aT_b})<\infty\) for some \(b=b(a)\); this is equivalent to \(\infty\) being an entrance boundary. If \(0\) is regular and \(\infty\) is not accessible, then \(X_{t}\) escapes to \(\infty\) if \(\int\varphi (\lambda_{0}^{k},y)d\Gamma (y)=\infty\) and \(X_{t}\) converges to the quasilimiting distribution \[ \varphi(\lambda_{0}^{k},y)d\Gamma (y)/\int \varphi (\lambda_{0}^{k},y)d\Gamma (y) \] when the integral converges. The paper contains many corollaries and illuminating comments. The authors point out that they eliminated some restricting hypotheses from previous results. The contents and proofs of the paper rely on many quotations. There are many preliminary results occupying more than half of the paper.