Drift estimation for stochastic reaction-diffusion systems (Q2293717)

This paper deals with an estimation problem for a model defined by a non-linear stochastic differential equation. A brief description of the paper follows. Remember some concepts about (linear) operators for Hilbert's spaces. Let \(H\) be a complex separable Hilbert space, \(A\) an operator defined over \(H\) with range in \(H\), \(\lambda\in\mathbb{C}\) is a regular value of \(A\) if \((A-\lambda\mathrm{Id})\) is invertible and bounded. \[ \rho(A)=\{\lambda\in\mathbb{C}\vert \lambda\text{ is a regular value of }A\}. \] \[ \sigma(A)=\text{complement of }\rho(A)\text{ in }\mathbb{C}. \] \(\rho(A)\) is the \(A\)-resolvent and \(\sigma(A)\) is the \(A\)-spectrum. The following mathematical model is considered in this paper. \(A\) is bounded, negative and self-adjoint; \(\rho(A)\) is a compact set; \((\Omega,F,P)\) is a complete probability space; \((F_t)_{t\geq 0}\) is a filtration on \((\Omega,F,P)\); \(T>0\); for every \(0\leq t\leq T\), \(X_{t}\) is a random function defined over \(\Omega\) with values in \(H\), and \(X=(X_t)_{T\geq t\geq 0}\), is a stochastic process such that \(X_t\) is \(F_t\)-measurable for all \(T\geq t\geq 0\) (\(X\) is \((F_t)_{t\geq 0}\)-adapted). The following stochastic differential equation is satisfied \[ dX_{t}=(\theta AX_t+F(t,X_t))dt+BdW_t \] with initial condition \(X_0\in H\), where \(\theta\) is an unknown parameter, \(F:[0,T]\times H\to H\) such that \(\omega\in\Omega\mapsto F(t,X_t(\omega))\in H\) is a measurable function, \(B=(-A)^{-\gamma}\), \(\gamma>0\), and \(W=(W_t)_{T\geq t\geq 0}\) is a cylindrical Wiener process with values in \(H\). Let \(\lambda_1\leq\lambda_1\leq\dots\) the eigenvalues of \(-A\) and \((\Phi_{n})_{n\geq 1}\) be the corresponding orthogonal base of eigenvectors. The orthogonal projection for each \(N=1,2,\dots\) is \(P_N:H\to\mathrm{Span}(\{\Phi_1,\dots,\Phi_N\})\). The authors study the asymptotic behavior of three estimators of the \(\theta\) parameter which are based on \(X_N=P_N(X)\). One of them was partially studied in a previous research by other authors. The main result of this paper shows the consistency and asymptotic normality of the three estimators given certain conditions regarding the possible non-linearity of \(F\). The paper illustrates the theory of the subject by means of some examples of interest in a variety of applications. Furthermore, the robustness of the estimators to perturbations in the model defined above is shown using these examples. In a simulation study of a particular case of the general model, some numerical results of the calculation of the studied estimators are shown. The proof of the main theorem of the paper is done over some pages where necessary results that are of interest in themselves are proved. Finally, the authors show how the proposed techniques can be extended to a particular case of a system of two stochastic differential equations. As a simplification of this review we assume that all operators have domain equal to \(H\). For the purpose of this paper the authors work with operators whose domain is a subspace of \(H\). This paper is well written, rigorous in concepts and formal results and with an extensive list of references that allows you to read it thoroughly. However, extensive knowledge of functional analysis is necessary, in particular Hilbert spaces and stochastic differential equations.