Sharp inequalities for the distribution of a stochastic integral in which the integrator is a bounded submartingale (Q1897182)

Let \(X\) be an adapted submartingale with respect to a filtration satisfying the usual conditions, with \(|X|_\infty \leq 1\) and let \(H\) be a predictable process with values in the closed unit ball of \(\mathbb{R}^\nu\). Define \[ Y_t = H_0 X_0 + \int_{(0,t]} H_s dX_s\quad \text{and} \quad Y^* = \sup_{t> 0} |Y_t|. \] The following problem is addressed: For each \(\lambda > 0\), find a function \(U_\lambda : [-1,1 ]\times \mathbb{R}^\nu \to [0,1]\) such that for any \(H\) as above \[ P\{Y^* \geq \lambda \} \leq E[U_\lambda (X_0, Y_0)], \] and this estimate is sharp in the sense that \(\forall \lambda > 0\), \(\forall x \in [-1, 1]\), \(\forall y \in \mathbb{R}^\nu\), \(|y|\leq |x|\Rightarrow \forall \beta < U_\lambda(x,y)\), \(\exists H, X\) s.t. \(X_0 = x\), \(X_0 H_0 = y\), \(P\{Y^* \geq \lambda\} > \beta\). Such function \(U_\lambda\) is found explicitly. The inequality and its sharpness is proved through a similar inequality for discrete time submartingales, extended then to stochastic integrals using stopping times. As a consequence, the following sharp inequality is found: for each \(\lambda > 4\), \[ P\{Y^* \geq \lambda\} \leq \gamma e^{-\lambda/4}, \] where \(\gamma = (8 + \sqrt{2}) e/12\).