Discretized Tikhonov-Phillips regularization for a naturally linearized parameter identification problem (Q2577533)

The author studies the problem of recovering the diffusion coefficient \(a=a(x)\), \(x \in \Omega\) from noisy measurement \(u^{\delta}\) of a solution \(u\) to the boundary value problem \(-\nabla(a \nabla u)=f\) in \(\Omega\), \(u=g\) on \(\partial \Omega\). Let \(a_0\) be an initial guess for \(a\). The original problem is reformulated as the linear ill-posed operator equation \(As=r^{\delta}\), where \(A\) takes \(s=a-a_0\) to the solution \(z\) of \(-\nabla(a_0 \nabla z)=\nabla(s \nabla u_{sm}^{\delta})\) in \(\Omega\), \(z=0\) on \(\partial \Omega\); \(r^{\delta}=u_{sm}^{\delta}-u_0\), \(u_{sm}^{\delta}\) is a smoothed version of \(u^{\delta}\) and \(u_0\) solves the problem \(-\nabla(a_0 \nabla u_0)=f\) in \(\Omega\), \(u_0=g\) on \(\partial \Omega\). To the linear equation a discrete version of Tikhonov-Phillips regularization method is applied.