Dynkin Games via Dirichlet Forms and Singular Control of One-Dimensional Diffusions (Q4785654)

Let \(X\) be a locally compact separable metric space and \(m\) a positive Radon measure on \(X\) with full support. If \(({\mathcal E}, {\mathcal F})\) is a regular Dirichlet form on \(L^2(X, m)\), then there is an \(m\)-symmetric Hunt process on \(X\) associated with this form. Given \(\alpha>0\) , \(H\in L^2(X, m)\), and \(f_1, f_2\in {\mathcal F}\) with \(-f_1\leq f_2\), we put \(K=\{u\in {\mathcal F}: -f_1\leq u\leq f_2\}\). It is well known that the inequality \[ {\mathcal E}_{\alpha}(V, u-V)\geq (H, u-V), \quad \forall u\in K, \] has a unique solution \(V\in K\). In the paper ``Stopping games for symmetric Markov processes'' [Probab. Math. Stat. 4, 185--196 (1984; Zbl 0561.60050)], \textit{J. Zabczyk} showed that the function \(V\) above is the value function of the Dynkin game \[ V(x)=\inf_{\tau}\sup_{\sigma}J_x(\tau, \sigma)=\sup_{\sigma}\inf_{\tau}J_x(\tau, \sigma), \] where \[ J_x(\tau,\sigma)\colon =E_x \int_{0}^{\tau\wedge\sigma}e^{-\alpha t}H(X_t) dt+E_{x} e^{-(\tau\wedge \sigma)}( -1_{\sigma \leq \tau} f_1 (X_\sigma +1_{\tau<\sigma}f_2 (x_\tau))) \] and \(\tau, \sigma\) are stopping times. In this paper, the authors apply this result to the special case when \(X_t\) is a regular diffusion on \(R\) with both boundaries natural, and \(H,f_1,f_2\) satisfy some regularity conditions. Restricting first to a finite interval \((-A,A)\), with reflecting endpoints, the authors find a subinterval \((a,b)\) on which \(-f_{1}<V < f_2\) and \((\alpha-\frac{d}{ds} \frac{d}{dm})V=H\), where \(s\) and \(m\) are the natural scale and the speed measure. In this case, \(V(x)=J_x(\widehat \tau, \widehat \sigma)\), where \(\widehat \tau\) and \(\widehat \sigma\) are the hitting times of \([-A,a]\) and \([b,A]\), respectively. Then they show that an integral function \(W\) of \(V\) with respect to \(ds\) is a solution of a certain free boundary value problem involving the operator \(\frac{d}{ds} \frac{d}{dm}\). In the last section of this paper, the authors identify \(W\) with the optimal return function of a singular stochastic control of the diffusion \(X_t\).