Using differential equations to obtain joint moments of first-passage times of increasing Lévy processes (Q962030)

Consider a subordinator \(D\) and the associated first-passage time process \(E\): \[ E(t) = \inf\{s: D(s) > t \} \] also called the inverse subordinator. The authors characterize the finite-dimensional distributions of this process by showing that its Laplace-transform solves a PDE. More precisely, define \[ H^{(n)} (\{s_i, t_i\}_{i=1}^n) = H^{(n)}(s_1,\dots,s_n, t_1, \dots, t_n) = P(E(t_i) > s_i, i = 1, \dots, n) \] and \[ \widetilde{H^{(n)}}(\{s_i, \lambda_i\}_{i=1}^n) = \int_0^\infty \dots \int_0^\infty e^{-\lambda_1 t_1- \dots - \lambda_n t_n} H^{(n)}(\{s_i, t_i\}_{i=1}^n)dt_1 \dots dt_n \] Then \(\widetilde{H^{(n)}}\) solves the PDE \[ \left( \partial_{s_1} + \dots + \partial_{s_n}\right) \widetilde{H^{(n)}} = - \Phi(\lambda_1 + \dots + \lambda_n)\widetilde{H^{(n)}} \] with boundary conditions \[ H^{(1)}(0,\lambda) = \frac{1}{\lambda} \] respectively \[ H^{(n)}(\{s_i=0\}) = \frac{1}{\lambda_i} \widetilde{H^{(n-1)}}(\{s_k, \lambda_k \}) \] Here \(\Phi\) is the Laplace exponent of the subordinator \(D\). This result is used to obtain a recursive formula for the joint moments of \(E\): Let \[ U(\{t_i,m_i\}_{i=1}^n) = U(t_1, \dots, t_n; m_1, \dots, m_n) = E( E(t_t)^{m_1} \dots E(t_n)^{m_n}) \] Then the Laplace transform of U satisfies the recursion \[ \tilde{U}(\{t_i,m_i\}_{i=1}^n) = \frac{\sum_{i=1}^n m_i \tilde{U}(\lambda_1, \dots, \lambda_n; m_1, \dots, m_i - 1,\dots, m_n)}{\Phi(\lambda_1 + \dots + \lambda_n)} \] The authors develop a method to invert this Laplace transform, to obtain a recursion for the joint moments themselves: \[ U(\{t_i,m_i\}_{i=1}^n) = \int_0^{t_{\min}} \sum_{i=1}^n m_i U(t_1 - \tau, \dots, t_n - \tau, m_1, \dots, m_i-1, \dots, m_n) dU(\tau) \] here \(dU(\tau)\) denotes the renewal measure, whose distribution function is given by \(U(t) = E(E(t))\). Finally these results are applied to local times of Markov processes, which can be considered as inverse subordinators.