Some inequalities with local times in zero of a Brownian motion (Q1198557): Difference between revisions

Consider \(B=(B_ t, t\geq 0)\), a real-valued Brownian motion started at 0, and let \(L=(L_ t, t\geq 0)\) be its local time process at the level 0. This long and interesting article focusses on a number of inequalities involving \(B\) and \(L\) evaluated at certain stopping times. First, consider \(\mu\), a centered probability measure on \(\mathbb{R}\) (i.e. \(\langle\mu,| x|\rangle<\infty\), \(\langle\mu,x\rangle=0\)) with \(\mu(\{0\})=0\). Skorokhod's problem consists of searching for a stopping time \(T\) in the natural filtration of \(B\), such that the stopped Brownian motion \((B_{T\land t}, t\geq 0)\) is a uniformly integrable martingale and \(B_ T\overset{(d)}=\mu\). Denote by \({\mathcal T}_ \mu\) the set of stopping times that solve Skorokhod's problem for \(\mu\). In a previous paper, the author [Sémin. de probabilités XVII, Proc. 1981/82, Lect. Notes Math. 986, 227-239 (1983; Zbl 0512.60071)] obtained the following solution. There exist \(h_ \mu\) and \(k_ \mu\), two nonnegative nondecreasing functions on \([0,\infty)\) such that \(T_ \mu=T(h_ \mu,k_ \mu)\in{\mathcal T}_ \mu\), where \[ T(h,k)=\inf\{t: B^ +_ t=h(L_ t)\quad\text{or}\quad B^ -_ t=k(L_ t)\}. \] Here, the author shows that \(T_ \mu\) is the unique solution such that, for all convex functions \(\varphi\): \[ E(\varphi(L_ T))\leq E(\varphi(L_{T_ \mu})),\quad\text{for all}\quad T\in{\mathcal T}_ \mu.\tag{*} \] Similarly, there exist \(\hat h_ \mu\) and \(\hat k_ \mu\) nonincreasing functions on \([0,\infty)\) such that \(\hat T_ \mu=T(\hat h_ \mu,\hat k_ \mu)\) is the unique solution of Skorokhod's problem which fulfills \[ E(\varphi(L_{\hat T_ \mu}))\leq E(\varphi(L_ T)),\quad\text{for all}\quad T\in{\mathcal T}_ \mu,\leqno(*)^ \land \] for every convex function \(\varphi\). The author specifies also the convex functions \(\varphi\) for which \((*)\) (respectively \((*)^ \land\)) characterizes \(T_ \mu\) (respectively \(\hat T_ \mu\)). All this can be extended to the case when \(\mu(\{0\})\) is positive. The extremal solutions \(T_ \mu\) and \(\hat T_ \mu\) are then used to deduce inequalities between \(L_ T\) and \(B_ T\), where \(T\) is any stopping time such that \(\{B_{T\land t}, t\geq 0)\) is a uniformly integrable martingale. The idea is to consider the law \(\mu\) of \(B_ T\) and the corresponding stopping times \(T_ \mu\) and \(\hat T_ \mu\), and to apply \((*)\) and \((*)^ \land\). Hardy's transform and Hardy- Littlewood's inequality plays an important part in this approach. One gets \[ E((L_ T)^ p)\leq p^ p E(| B_ T|^ p)\quad\text{when}\quad p>1 \] [this was first proven by \textit{J. Azéma} and \textit{M. Yor} in ``Temps locaux'', Astérisque 52-53, 3-16 (1978; Zbl 0385.60063)] and \[ E(| B_ T|^ p)\leq p^{-p} E((L_ T)^ p)\quad\text{when}\quad 0<p<1. \] Moreover, it is shown that \(p^ p\) and \(p^{-p}\) are the best possible constants in the above inequalities. Finally, the author obtains inequalities between \(L_ \infty(M)\), the local time at level zero taken at time \(\infty\) of a continuous local martingale \(M\) started at 0, \(M^*_ \infty=\sup\{| M_ s|, s\geq 0\}\) and their ratios.