Onsager-Machlup functionals and maximum a posteriori estimation for a class of non-Gaussian random fields (Q805065)

Suppose that \(\{\) u(x), \(x\in D\}\) is a random field on a bounded domain D of \(R^ d\), \(d\geq 2\), which satisfies a nonlinear stochastic partial differential equation of the form \[ Pu(x)+F(u(x))=n(x),\quad x\in D,\quad P_{\partial}u(x)=0,\quad x\in \partial D, \] where n(x) is white noise, P is a strongly elliptic operator and \(P_{\partial}\) is a boundary operator. The authors show the existence of an Onsager-Machlup functional for the random field u(x), under some smoothness conditions on the function F. The basic tool in proving this result is a general theorem of Ramer and Kusuoka which allows to compute the Radon-Nikodym derivative of a nonlinear transformation of the Wiener measure. As an application the authors study the problem of the maximum a posteriori (MAP) estimation of the solution u(x) given nonlinear observations y(x) perturbed by an independent white noise. - This paper is an extension of a previous work of the authors [ibid. 35, No.2, 151-167 (1990; Zbl 0716.62101)], in the case where F is linear and u(x) is Gaussian.