A probabilistic approach to the solution of the Neumann problem for nonlinear parabolic equations (Q4787677)

Consider the Neumann problem for the quasilinear parabolic equation \[ \frac{\partial u}{\partial t} + \frac{1}{2} \sum^d_{i,k=1} a_{i,k} (t,x,u) \frac{\partial^2 u}{\partial x_i \partial x_k} + \sum^d_{i=1} b_i(t,x,u) \frac{\partial u}{\partial x_i} + g(t,x,u) = 0 \] for \((t,x) \in Q = [t_0,T] \times D \) cylinder in \(R^{d+1}\) with bounded domain \(D\in R^d\), with terminal conditions \[ u(T,x) = \varphi (x) \] for \(x \in \partial D\), and boundary conditions \[ \frac{\partial u}{\partial \nu} = \psi (t,x,u) \] for \(t\in [t_0,T]\), \(x \in \partial D\), where \(\nu\) is the direction of the internal normal to the boundary \(\partial D\) at the point \(x \in \partial D\). The authors follow the probabilistic approach to constructing layer methods for solving this Neumann problem as originally proposed in \textit{G. N. Milstein} [Numer. Methods Partial Differ. Equations 18, 490-522 (2002; Zbl 1014.65005)]. That is they construct numerical approximations \(\bar{u}(t_k,x)\) of solution \(u(t_k,x)\) using the well-known probabilistic representation \[ u(t,x) = E [ \varphi(X_{t,x}(T))+Z_{t,x,0}(T)] \] where \(X\) and \(Z\) solve an initial value problem for certain Itô SDEs [see \textit{M. Freidlin}, Functional integration and partial differential equations (1985l Zbl 0568.60057) or \textit{I. I. Gichman} and \textit{A. W. Skorochod}, Stochastische Differentialgleichungen (1971; Zbl 0287.34063)]. Therefore, they can exploit numerical integration techniques in the weak sense as described in [\textit{G. N. Milstein}, Theory Probab. Appl. 40, No. 3, 556-563 (1995); translation from Teor. Veryatn. Primen. 40, No. 3, 657-665 (1993; Zbl 0859.60055) or Theory Probab. Theory 41, 210-218 (1996)]. Their layer methods are constructed along equidistant grids of the interval \([t_0,T]\). Here, a spacediscretization is not necessary, as commonly met while applying difference methods. Thus, layer methods may be preferrable, see also numerical tests [in \textit{G. N. Milstein} and \textit{M. V. Tretyakov}, IMA J. Numer. Anal. 21, 887-917 (2001; Zbl 0993.65014)]. A convergence theorem with convergence rate \(O(h^{1/2})\) is proven. This rate can be improved to \(O(h)\) when a space-discretization is additionally taken into account. The construction and algorithms for several layer methods are given. Numerical tests are conducted for the Burgers equation. Other probabilistic approaches may be found by \textit{H. J. Kushner} [Probability methods for approximations in stochastic control and for elliptic equations, Math. Sci. Eng. 129, (1977; Zbl 0547.93076)] and \textit{D. Talay} and \textit{L. Tubaro} [in: C. Graham, et al. (eds.), Probabilistic models for nonlinear partial differential equations, Lect. Notes Math. 1627 (1996; Zbl 0844.00022)]. All in all, this paper belongs to a series of papers of Milstein and coauthors which present some essential cornerstones in the probabilistic approach to numerical integration of nonrandom boundary value problems for parabolic partial differential equations.