Spectral gap and rate of convergence to equilibrium for a class of conditioned Brownian motions (Q2485855)

The Brownian motion on the line can be restricted to a finite interval \([0,\gamma ]\) in different ways. Besides the known methods a new one has been introduced by \textit{I. Grigorescu} and \textit{M. Kang} [J. Theor. Probab. 15, No. 3, 817--844 (2002; Zbl 1016.60074)]: Starting from some point in \((0,\gamma)\) the process is catapulted to the point \(p\gamma\), where \(p\in (0,{1\over 2}]\), as soon as the process hits the boundary. This process is shown to be an ergodic Markov process whose transition probabilities converge to an invariant measure at an exponential rate independent of \(p\). This process is labelled Brownian motion physically returned to the point \(p\gamma\). Intrigued by the independence of the convergence rate of \(p\) a process \((X_{t}^{\text{cond},p})\) called Brownian motion returned to the point \(p\gamma\) by conditioning is constructed: Being in \((0,p\gamma)\) (or \((p\gamma,\gamma)\)) the process looks like a Brownian motion conditioned to hit \(p\gamma\) before \(0\) (or \(\gamma\)). The process is the solution of the martingale problem of the operator \[ L_{p}=\left\{ \begin{matrix} {1\over 2}{d^2\over dx^2} + {1\over x}{1\over dx} & \text{for } x\in (0,p\gamma)\cr {1\over 2}{d^2\over dx^2} - {1\over \gamma - x}{1\over dx} & \text{for } x\in (p\gamma,\gamma).\end{matrix}\right. \] Conditioning the Brownian motion on the event that the time of the \(n\)th upcrossing of the interval \([p\gamma,p\gamma +\delta]\) is smaller than the hitting time of the boundary of \([0,\gamma]\) the author shows that the process \((X_{t}^{\text{cond},p})\) can be obtained as a weak limit letting first \(n\to\infty\) and then \(\delta \to 0\). Furthermore, the corresponding distribution \(P_{x}^{\text{cond},p}\) is described in terms of its Radon-Nikodym derivative with respect to the Wiener measure \(P_{x}\). Eventually, the convergence of the transition probabilities to the invariant measure is proven and the \(p\)-dependence of the rate is established.