A modification of the branching process (Q788402)

Suppose \(\eta_ t\), \(t\geq 0\), is the usual Markov age-dependent branching process (M.b.p.) with transition intensities (\(h\to 0): P\{\eta_{t+h}=k+r-1| \eta_ t=k\}=\delta_{1r}+k\alpha_ rh+o(h)\) where \(\alpha_ 1<0\), \(\alpha_ i\geq 0\), \(i\neq 1\), \(\sum \alpha_ i=0\). (Authors use the name: ''time-continuous Galton-Watson process''.) The ''upside-down'' modification of M.b.p. \(\eta_ t\) is a Markov process \(\xi_ t\), described by: \(P\{\xi_{t+h}=k+1-r| \xi_ t=k\}=\delta_{1r}+k\alpha_ rh+o(h),\) 0\(\leq r\leq k\), \(P\{\xi_{t+h}=0| \xi_ t=k\}=\delta_{0k}+k{\hat \alpha}_ kh+o(h),\) \(k\geq 0\), where \({\hat \alpha}{}_ k=\sum_{i\geq k+1}\alpha_ i\). The generating function (g.f.) of \((\alpha_ i)\) is defined as \(\alpha(\theta)=\sum \alpha_ k\theta^ k\). Thus \(\xi_ t\) has a tendency towards increasing when \(\eta_ t\) tends to decrease, etc. Authors consider the extinction probabilities of \(\xi_ t\). In this respect, \(\eta_ t\) and \(\xi_ t\) are ''dual'' in the following sense (Theorem 1): \[ (1)\quad \alpha_ 0P\{\xi_ t>0| \xi_ 0=1\}P\{\eta_ t>0| \eta_ 0=1\}=\alpha(G(t)) \] where G(t) is the extinction probability at t, for \(\eta_ t\). Proof of (1), as well as of the other results is based upon the fact that \(P\{\xi_ t>0| \xi_ 0=1\}=\alpha(G(t))/(\alpha_ 0(1-G(t))).\) The well known properties of G(t) lead then to analogous properties of \(\Phi_ k(t)=P\{\xi_ t>0| \xi_ 0=k\}\). For the critical case \((\alpha '(1)=0\), \(\alpha\) ''(1)\(\neq 0\), \(\alpha ''(1)<\infty)\), we have (Theorem 2): \(\Phi_ 1(t)=(\alpha_ 0(t)^{-1}(1+o(1))\), \(t\to \infty\); and more generally (Theorem 3): \[ \Phi_ n(t)=(\sum^{n}_{k=1}k\nu_{n-k})t^{- 1}(1+o(1)),\quad t\to \infty, \] where the g.f. of \((\nu_ k)\) is \((1- \theta)^ 2(\alpha(\theta))^{-1}\). In the supercritical case \((\alpha '(1)<0\); for this data, \(\eta_ t\) is subcritical), Theorem 4 gives: \[ \Phi_ n(t)\to -\alpha '(1)\sum^{n}_{k=0}b_ k,\quad t\to \infty, \] where the g.f. of \((b_ k)\) is \((1-\theta)(\alpha(\theta))^{-1}\). Finally, in the subcritical case \((\alpha '(1)>0\), \(\alpha ''(1)<\infty)\), Theorem 5 gives: \[ \Phi_ k(t)=\gamma_ ke^{- at}(1+o(1)),\quad t\to \infty, \] where \(a=-\alpha '(\rho)>0\), \(\rho\) is the smallest positive root of \(\alpha\) (u) and the g.f. of \((\gamma_ k)\) is \(\theta(\alpha(\theta)\cdot \exp \{a\int^{\theta}_{0}(\alpha(u))^{-1}du\})^{-1}.\) So, the extinction probabilities of \(\xi_ t\) behave asymptotically in a similar way as those of \(\eta_ t\), in respective cases.