Sharp maximal inequalities for the moments of martingales and non-negative submartingales (Q654410)

The author proves two sharp maximal inequalities for differentially subordinated martingales and non-negative submartingales, respectively. An adapted and integrable real-valued process \((g_n:n \in \mathbb{N})\) is called differentially subordinate to another adapted and integrable real-valued process \((f_n:n \in \mathbb{N})\) if a.s. \(|g_n-g_{n-1}|\leq |f_n - f_{n-1}|\). If \(\alpha \in [0,1]\), then \(g\) is called \(\alpha\)-strongly subordinate to \(f\) if it is differentially subordinate to \(f\) and if further \[ |\operatorname{E}(g_{n+1}-g_n|\mathcal{F}_n)| \leq \alpha |\operatorname{E}(f_{n+1}-f_n|\mathcal{F}_n)|. \] The first main result of the paper is that for martingales \(f\) and \(g\), where \(g\) is differentially subordinate to \(f\), it holds for all \(p\geq 2\) \[ \| \max_{k\leq n} |g_k|\|_{L^p} \leq p \| f \|_{L^p}, \] and the constant \(p\) is optimal. This statement follows from a previous result by Burkholder in combination with Doob's maximal inequality; however, the optimality of the constant is new. The second main result is that, if \(f\) is a non-negative submartingale and if \(g\) is \(\alpha\)-strongly subordinate to \(f\), then, for all \(p\geq 2\), \[ \| \max_{k\leq n} |g_k|\|_{L^p} \leq (\alpha+1)p \| f \|_{L^p} \] holds, and the constant \((\alpha+1)p\) is optimal. This is then applied to stochastic integrals and to Itô processes.