On stochastic elliptic boundary value problems associated with Gaussian Markov random fields

From MaRDI portal





[For the entire collection see Zbl 0754.00020.] Let \(\varphi=\varphi(k,\beta)=\{\varphi(x),\;x\in\mathbb{R}^ 2\}\) be a real stationary mean zero Gaussian field which satisfies the stochastic Laplace equation \((\Delta-k^ 2)\varphi(x)=\beta{\overset\circ W}(x)\), \(\beta\), \(k\) be real nonzero constants and \(\{{\overset\circ W}(x),x\in \mathbb{R}^ 2\}\) be a real Gaussian white noise field on the plane. In Section 1 the authors present some preliminary properties of the field \(\varphi\), its sample functions, computer simulations of \(\varphi(1,2\pi^{1/2})\) on the unit square and normal derivatives of \(\varphi\) on the smooth curves \(\Gamma\). They reformulate the least square prediction problem for \(\varphi\) as a stochastic boundary value problem. In Section 2 they list some basic questions of sample design, none having complete answers. Namely, let \({\mathcal D}\subset \mathbb{R}^ 2\) be a bounded domain, \(\varphi\) is observable on \(\mathbb{R}^ 2\backslash {\mathcal D}\) and it is necessary to estimate \(\varphi\) on \({\mathcal D}\) and to study the properties of the prediction errors. Section 3 contains the main results about prediction errors. For example, let \[ \varphi_{\mathcal D}(x)=E\{\varphi(x)/\varphi(y),\;y\in {\mathcal D}\},\qquad e_{\mathcal D}(x)=\varphi(x)-\varphi_{\mathcal D}(x),\qquad \sigma^ 2_{\mathcal D}(x)=E[e_{\mathcal D}(x)]^ 2, \] \[ \varphi=\varphi(1,2\pi^{1/2}),\qquad (f,f)=1/(4\pi)\int_{\mathcal D}|(\Delta-I)f(x)|^ 2dx,\;f\in C_ 0^ \infty({\mathcal D}). \] Then \[ \sigma^ 2_{\mathcal D}(x)=\sup\{f(x)^ 2:(f,f)\leq 1\}. \] Section 4 contains the proofs.











This page was built for publication: On stochastic elliptic boundary value problems associated with Gaussian Markov random fields

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4029030)