A Fourier truncated regularization method for a Cauchy problem of a semi-linear elliptic equation (Q2443301)

In this paper, a regularization approach for the stable approximate identification of \(u(y,\cdot)\) from data of \(\varphi\) is suggested and studied, where \(u(y,x)\) satisfies for a bounded domain \(\Omega \subset {\mathbb R}^n,\;n>1,\) with smooth boundary \(\partial \Omega\) the system \[ \begin{cases} u_{yy}(y,x)-L_x u(y,x)=f(y,x,u(y,x)), & x \in \Omega, \;0<y<T,\\ u(y,x) =0, & x \in \partial \Omega, \;0 \leq y \leq T,\\ u(0,x)=\varphi(x), & x \in \Omega, \\ u_y(0,x)=0, & x \in \Omega. \end{cases} \] In this context, \(L_x\) denotes a linear densely defined self-adjoint and positive definite operator with respect to \(x\) mapping in a Hilbert space. Moreover \(f\) is a prescribed function with \[ \|f(y,x,w)-f(y,x,v)\| \leq k \|w-v\|. \] It is well-known that this inverse problem is severely ill-posed. For obtaining regularized solutions, Zhang and Wei exploit the Fourier truncated regularization method, which, as mentioned by the authors, had been used before in some other papers to deal with related ill-posed Cauchy problems. For a noise level \(\delta>0\) and noisy data \(\varphi^\delta\) with \(\|\varphi-\varphi^\delta\|\leq \delta\) the approximate solution with regularization parameter \(\lambda_\delta>0\) solves the integral equation \[ u^\delta_{\lambda_\delta}(y,x)=\sum _{\lambda_ \leq \lambda_\delta} \left(\cosh(\sqrt{\lambda_n}y)\varphi_n^\delta+\int _0^y \frac{\sinh(\sqrt{\lambda_n}(y-\tau))}{\sqrt{\lambda_n}}f_n(u^\delta_{\lambda_\delta})(\tau)d\tau \right)X_n \] with \[ \varphi_n^\delta=\langle \varphi^\delta,X_n\rangle \] and \[ f_n(u^\delta_{\lambda_\delta})(y)=\langle f(y,x,u^\delta_{\lambda_\delta}(y,x)),X_n \rangle. \] Here, \(\lambda_n\) and \(X_n\) are eigenvalues and eigenelements, respectively, of the operator \(L_x\) with homogeneous boundary conditions. Stabiliy of the regularized solutions is shown in the sense \[ \|u^\delta_{\lambda_{1,\delta}}(y,\cdot)-u^\delta_{\lambda_{2,\delta}}(y,\cdot)\| \leq C_1\|\varphi_1^\delta-\varphi_2^\delta\| \] with a constant \(C_1=C_1(y,k,T)\). Moreover, logarithmic convergence rates are derived if the solution smoothness is expressed by the fact that positive constants \(r\) and \(E\) exist such that the exact solution \(u(y,\cdot)\) fulfills the inequality \[ \sum _{n=1}^\infty \exp(2(T+r)\sqrt{\lambda_n})|\langle u(y,\cdot),X_n\rangle|^2 \leq E^2. \] Numerical case studies are presented at the end of the paper, but it is not clear how to find the constant \(r>0\) which is used explicitly in the a priori choice of the regularization parameter \(\lambda_\delta\) under consideration.