Time-changed spectrally positive Lévy processes started from infinity (Q2040064)

Consider a spectrally positive Lévy process \(Z\) with log-Laplace exponent \(\Psi\), i.e.,\ such that \(\mathbb{E}[e^{-\lambda Z_1}] = \exp(\Psi(\lambda))\), and a positive continuous function \(R\) on \((0, \infty)\). Conditional to the path \((Z_t)_{t\geq 0}\) stopped upon reaching \(0\), one defines a time-changed path \((X_t)_{t\geq 0}=(Z_{\theta(t)})_{t\geq 0}\) with \(\theta(t)\) defined as the inverse of \(t\mapsto \int_0^t \frac{1}{R(Z_s)}\, ds\). With this definition, \(X\) is a Markov process that the authors call a \textit{nonlinear CSBP}, in relation to the fact that any continuous state branching process can be written as such a time-change of a spectrally positive Lévy process with \(R(x)=x\). The main result is that when the process \(X\) is (sub)critical, i.e., when \(\Psi'(0+)\geq 0\), then \(\infty\) is an entrance boundary for it if and only if \[ \int^\infty\frac{1}{x\Psi(1/x)R(x)}\,dx < \infty. \] The authors further give the expectation of the hitting time of any fixed \(b>0\) under \(\mathbb{P}_\infty\), in terms of the function \(R\) and the so-called scale function \(W\) of the Lévy process \(Z\). Under natural regularity conditions for \(R\) and in the subcritical case (\(\Psi'(0+)>0\)), it is proven that \(X\) comes down to infinity with speed \(\varphi^{-1}(x)\), where \[ \varphi(x) := \frac{1}{\Psi'(0+)}\int_x^{\infty}\frac{dx}{R(x)}, \] meaning that \(X_t/\varphi^{-1}(t)\) goes to \(1\) as \(t\to 0\), in \(\mathbb{P}_\infty\)-probability. This result is strengthened to an almost-sure convergence if \(R\) is regularly varying at \(\infty\) with index \(\theta > 1\), under some further technical conditions. In the critical case (\(\Psi'(0+)=0\)), if \(\Psi(\lambda)\sim_0 c\lambda^{\alpha}\) for \(c> 0\) and \(\alpha\in(1,2]\), and if \(R\) is regularly varying at \(\infty\) with index \(\theta > \alpha\), then \(X\) comes down from infinity and oscillates widely in the sense that \(\limsup_{t\to 0}X_t/\underline{X}_t=\infty\) almost surely, where \(\underline{X}_t\) is the running infimum of \(X\); it is also shown that \(\underline{X}_t\) can be normalized to converge in distribution to a nonconstant random variable as \(t\to 0\). On the other hand, it is shown that if \(R\sim g(x)e^{\theta x}\), then the \(X\) comes down from infinity without oscillating, i.e.,\ it behaves similarly as in the subcritical case. The proofs rely mainly on studying the behavior, as \(b\to\infty\), of the hitting times of \(b\) by the process \(X\) started from infinity; key to the analysis is the fact that these hitting times can be expressed as functionals of the Lévy process \(Z\), namely weighted occupation times.