Non-linear Neumann's condition for the heat equation: a probabilistic representation using catalytic super-Brownian motion (Q2567139)

The purpose of this paper is to give a probabilistic representation for solutions of the mixed Dirichlet nonlinear Neumann boundary value problem (DNP) \[ \left\{ \begin{aligned} \varDelta u & = 0 \quad \text{\,in} \quad D, \\ u & = \varphi \quad \text{on} \quad F_2, \\ \partial_n u + 2 u^2 & = 0 \quad \text{\,on} \quad F_1. \end{aligned} \right. \tag{1} \] Here \(D\) is a bounded domain in \({\mathbb R}^d\) with smooth boundary \(\partial D\), \(( F_1, F_2)\) is a nontrivial partition of \(\partial D\), \(\varphi\) is a nonnegative bounded and continuous function defined on \(F_2\), and \(\partial_n\) denotes the outward normal derivative on the boundary \(\partial D\). One of the most remarkable features in this paper consists in the fact that, to solve the DNP, the authors employ a catalytic super-Brownian motion with underlying motion \(B = (B_t)\) being a Brownian motion reflected on \(\partial D\), killed when it reaches \(F_2\) and catalysed by the set \(F_1\), and also that, instead of building the catalytic super-Brownian motion as a limit of branching particle systems, they use the construction introduced by \textit{P. Mörters} and \textit{P. Vogt} [Stochastic Processes Appl.\ 115, No. 1, 77--90 (2005; Zbl 1072.60070)] based on collision local time. Here is the first main result of this paper. Theorem A. The function \( w(x) = - \log {\mathbb E}_{\delta_x}^Z [ \exp \{ - \langle Z^{\text{Dir}}, \varphi \rangle ] \) \(( \varphi \in {\mathcal B}_+ ( \bar{F}_2)\) continuous) is a nonnegative weak solution of the DNP (1). \({\mathcal B}_+( \bar{F}_2)\) denotes the set of nonnegative measurable functions defined on \(\bar{F}_2\), and \(\delta_x\) is the Dirac mass at \(x\). Essentially, the most important point is to show how to construct such a random measure \(Z^{\text{Dir}}\). Let \(( H^x; x \in F_1)\) be the family of \(\sigma\)-finite measures which describe the law of the excursion of a reflected Brownian motion in \(D\) started from \(x \in F_1\). It is shown that the associated capacitary local time \(L\) on \(F_1\) has a density \(\rho\) with respect to the local time of \(B\) on \(F_1\). When \(( \ell_t^*)\) is the local time of \(B\) on \(F_1\) before it hits \(F_2\) and \(\ell^{*, -1}\) is its right-continuous inverse, then \(\xi_t\) is defined as \(( \ell_t^{*, -1}, B_{ \ell_t^{*, -1}})\). Then let us consider, under \({\mathbb P}_{\nu}^X\), a superprocess \(( X_t )\) started at the initial measure \(\nu\), with quadratic branching mechanism and underlying motion \(\xi\). In particular, \(X_t\) takes values in the set \({\mathcal M}_f( {\mathbb R}_+ \times F_1)\) of finite measures on \({\mathbb R}_+ \times F_1\). By making use of the total occupation measure \(\Gamma(dr, dx) = \int_0^{\infty} ds X_s(dr, dx),\) the random measure \(Z^{\text{Dir}}\) on \(F_2\) is defined for any \(\varphi \in {\mathcal B}_+(F_2)\) by \[ \langle Z^{\text{Dir}}, \varphi \rangle = \iint \Gamma(dr, dx) \rho(x) H^x[ \varphi(e_{\tau}) 1_{ \{ \tau < \infty \} }], \] where \(\tau\) is the hitting time of \(F_2\) for the excursion \(e\) under \(H^x\). Intuitively, the measure \(Z^{\text{Dir}}\) describes the death positions of infinitesimal particles released from the catalyst at time \(dr\) and position \(dx\) according to the random measure \(\rho(x) \Gamma(dr, dx)\), performing independent Brownian excursions outside \(F_1\) killed when they first reach \(F_2\). \({\mathbb E}_{\delta_x}^Z\) denotes the expectation in terms of the law \({\mathbb P}_{\delta_x}^X\) of the exit measure \(Z^{\text{Dir}}\) when \(\eta = \delta_x\) is taken in the initial measure \(\nu = \delta_0 \otimes \eta \in {\mathcal M}_+({\mathbb R}_+ \times F_1)\). In the second part, using techniques developed by \textit{R. Abraham} and \textit{J.-F. Delmas} [Probab.\ Theory Relat.\ Fields 128, No. 4, 475--516 (2004; Zbl 1054.60081)], the authors replace the Dirichlet condition on \(F_2\) by a Neumann condition. Actually, a similar probabilistic representation formula for solutions to the problem \[ \left\{ \begin{aligned} \varDelta u & = 0 \quad \text{in} \quad D, \\ \partial_n u - 2 \varphi & = 0 \quad \text{on} \quad F_2, \\ \partial_n u + 2 u^2 & = 0 \quad \text{on} \quad F_1, \end{aligned} \right. \tag{2} \] is derived as well.