Optimal trapping for Brownian motion: a nonlinear analogue of the torsion function (Q2660176)

The torsion problem was originally studied by Pólya in his celebrated paper on Torsional rigidity (1948). In its original form it stands for the effect of the shape of a planar domain on the resistance, under twisting, of an elastic cylinder whose cross section is the given domain. According to the interpretation suggested by Pólya, this torsional rigidity \(P\) of a domain \(\Omega\subset\mathbb{R}^2\) is given by \[ P(\Omega) := \sup_{u\in \mathbb{H}_0(\Omega)}\frac{\left( \int_\Omega u\right)^2}{\left(\int_\Omega |\nabla u|^2\right)^2} \ . \] Using the method of Steiner Symmetrization he proved, between other things, that the torsional rigidity is maximized for a disk, among all domains of the same area \(|\Omega|\). The stress function \(u\) is given by a constant multiple of the solution to the Dirichlet problem \[ \Delta u=-1 , \ u|_{\partial\Omega}=0 \tag{\(*\)} \] and Pólya's result corresponds to the statement that the maximal torsion \(P(\Omega)\) within all domains of a prescribed area \(|\Omega|\) is the disc (or the ball in general dimension \(d\)). This is a manifestation of the optimality of the ball for various variational optimization problems, discussed by Pólya himself and many subsequent researchers. There are other possible interpretation for this solution to the problem (\(*\)), e.g. \(u(x)\) is the mean time of escape form \(\Omega\) of a canonical Brownian particle originated at \(x\in\Omega\). In this sense, the integral of \(u\) indicates the average escape time, so, in this sense, the ball is the domain for which the average escape time of is maximal. In this paper the authors show that the ball is also the optimal shape to a related problem: find the vector field \(\vec{b}\) constraint by \(\|\vec{b}\|_\infty:= b\) and domain \(\Omega\) which maximize the escape time of a Brownian motion combined with the drift \(\vec{b}\). This corresponds to the maximization of some norm of \(u\) (e.g. \(\|u\|_p\), \(p\geq 1\) or \(\|\nabla u\|_1\)) on \(\vec{b}, \Omega\) subject to \(|\Omega|\),\( \|\vec{b}\|_\infty\) given, for the solution of \[ -\Delta u + \vec{b}\cdot \nabla u = -1, \ u|_{\partial\Omega}=0. \tag{\(**\)} \] The authors show, at the first step, that the maximizer of \(\|\nabla u\|_1\) is obtained on {\em any} domain \(\Omega\) by the vector field \(\vec{b}=-b\frac{\nabla u}{|\nabla u|}\), that is, the optimal vector field corresponds to the solution itself and is pointing in the opposite direction of the ''heat flux'' \(\nabla u\). This implies the reduction of this equation (\(**\)) to a nonlinear one: \[ -\Delta u - b| \nabla u |= -1, \ u|_{\partial\Omega}=0 . \tag{\(***\)} \] Then they analyze (\(***\)) and show, by contour integration, co area formula and using the level domains \(\Omega_t:= \{ x\in \Omega, u(x)\geq t\}\) that the function \(f(c):= b\sup_{|\Omega|=c}\int_\Omega |\nabla u|\) satisfies a differential inequality \(f^{'}(c)\leq b\frac{f(c)+c}{c_d c^{d-1/d}}\) where \(c_d\) is the isoperimetric constant in \(\mathbb{R}^d\) (\(|\partial \Omega|\geq c_d|\Omega|^{(d-1)/d}\)), and that this inequality turns into an equality only if \(\Omega\) is a ball. Related inequalities are shown also for \(h_p(c):= \sup_{|\Omega|=c} \|u\|^p_{p, \Omega}\) for \(\infty\geq p\geq 1\). This implies their main result.