Inverse problems and Carleman estimates. Global uniqueness, global convergence and experimental data (Q2232660)

The book is mainly devoted to Carleman estimates and their applications. The contents of the book is described in the first chapter. The second chapter deals with constructing Carleman functions \(\psi(x)\) for elliptic, parabolic, and hyperbolic operators and the corresponding stability estimates are proven. Let \(L\) be a second order differential operator defined in a domain \(G\subset \mathbb{R}^{n}\) and let \(\psi(x)\) be a sufficiently smooth function such that \(|\nabla \psi(x)|\neq 0\) (\(x\in G)\). Pointwise Carleman estimates are inequalities of the form \[ \varphi^{2}(Lu)^{2}\geq C (\lambda^{3} |u|^{2}\varphi^{2}(x)+\lambda|\nabla u|^{2}\varphi^{2}(x))-\operatorname{div}U,\ \forall x\in \Omega_{h},\ \forall u\in C^{2}(\overline{\Omega_{h}}), \] where \(x\in \Omega_{h}=\{x\in G: \psi(x)\geq h\geq 0\}\), \(\varphi(x)=e^{\lambda \psi(x)}\), \(|U|\leq C_{1}(|u|^{2}+|\nabla u|^{2})\), and \(\lambda \geq \lambda_{0}\in \mathbb{R}\). The above function \(\psi\) is called a Carleman function. These estimates actually do not depend on the form of a differential operator and the lower order terms in this operator and allow to prove global uniqueness theorems in a unified manner. Let \(u\) be a solution to the differential inequality \(|Lu|\leq c_{1}|u|+c_{2}|\nabla u|+|f|\) satisfying the boundary conditions \(u|_{\Gamma_{h}}=g_{0}\), \(u_{\nu}|_{\Gamma_{h}}=g_{1}\), where \(\Gamma_{h}=\{x\in \partial G:\psi\geq h\}\) and \(\nu\) is a normal to the boundary. The estimates of the form (1) ensure stability estimates of the form \[ \|u\|_{W_{2}^{1}(\Omega_{h+3\varepsilon})}\leq C_{1} (1+\|u\|_{W_{2}^{1}(\Omega_{h+\varepsilon})}\delta^{\beta},\ \ \delta=\|f\|_{L_{2}(\Omega_{h})}+\|g_{1}\|_{L_{2}(\Gamma_{h})}+\|g_{0}\|_{W_{2}^{1}(\Gamma_{h})},\ \beta\in (0,1). \] The results of Chapter 2 are employed in Chapter 3, where uniqueness theorems for coefficient inverse problems are established. In addition to the initial and boundary conditions, the Cauchy data are given on the boundary in elliptic case and on the lateral boundary of the cylinder (\((x,t)\in G\times (0,T)\)) in the parabolic and hyperbolic cases. The problem is to find a solution to the equation and to recover a coefficient of the equation (depending only on \(x\) in the parabolic and hyperbolic case). Moreover, in this chapter in the parabolic case uniqueness of recovering a coefficient and the initial data is proven when the final overdetermination condition is imposed. The quasi-reversibility numerical method by Lions and Lattes for ill-posed Cauchy problems and linear PDEs is treated in Chapter 4. The QRM is stated for the problem of the minimization of a Tikhonov-like functional with an unbounded partial differential operator in it. A typical form of the Tikhonov-like functional with a Carleman function \(\psi\) is as follows: \(J(w)=e^{-\lambda \delta}\int_{\Omega}(Lw(x))^{2}\varphi^{2}\,dx+\gamma \|w\|_{W_{2}^{s}(\Omega)}^{2}\), where \(\varphi=e^{\lambda \psi}\), \(\gamma>0\), is the regularization parameter, and the multiplier \(e^{-\lambda \delta}\) here allows to preserve some balance between the first and second terms in the right-hand side. The existence and uniqueness of the minimizer to this functional is established, the convergence analysis is performed, and the convergence rate of minimizers to the exact solution as the noise level in the data tends to zero, is found. The proof is obtained via an application of Carleman estimates. It is demonstrated that the QRM can be applied to a wide class of ill-posed Cauchy problems for linear PDEs, that is, the QRM is a universal regularization method. It works for those PDEs, for which Carleman estimates hold. Chapters 5--11 are devoted to the convexification method introduced by Klibanov M. In contrast to the previous chapters, it is often the case that equations under consideration are quasilinear. In the 5th chapter, it is shown that in the presence of Carleman estimates the construction approximate solutions to all ill-posed problems of Chapter 2 can be reduced to the minimization of some strictly convex functionals. It is demonstrated that the gradient projection method applied converges and the rate of convergence is established. As is known, conventional numerical methods for nonlinear ill-posed problems based on the minimization procedure of the least squares cost functionals typically suffer from the phenomenon of local minima and ravines. In this case, the situation is quite different and uniqueness of the minimizer holds. At the end of the chapter, some numerical examples are presented in the one-dimensional case. The problem of recovering the lower order coefficient in a second order differential operator is examined in Chapter 6. The additional data are the restricted Dirichlet-to-Neumann map which is a collection \((u|_{\Gamma}, \frac{\partial u}{\partial \nu}|_{\Gamma}\)) of the Dirichlet and Neumann data depending on one parameter \(x_{0}\in [0,1]\) and defined on some non-characteristic surface \(\Gamma\) lying inside the domain which can coincide with \(\mathbb{R}^{n}\) or be a bounded domain (in the latter case a solution satisfies the Dirichlet boundary conditions on the boundary). A special orthonormal basis is constructed for solving this problem. Stability estimates are obtained, and the uniqueness theorem is proven under certain conditions on the data. A weighted Tikhonov-like strictly convex functional is constructed to solve the problem numerically. A sequence of function converging to a minimizer is constructed and the rate of convergence is estimated. Examples in the parabolic and hyperbolic cases are displayed. The exposition in the next five chapters follows the same scheme. They are devoted to solving some classical problems of mathematical physics. Chapter 7 deals with electrical impedance tomography problem of recovering the conductivity distribution with the restricted Dirichlet-to-Neumann map (as in the previous chapter). Theorems of uniqueness and stability estimates are obtained, the weighted Tikhonov-like strictly convex functionals are constructed and their properties are described. The results of numerical experiments are presented. The same results are obtained in the case of a coefficient inverse problem for a hyperbolic equation with a single location of the point source in Chapter 8. The simultaneous recovering of a coefficient and the initial data are treated in Chapter 9, with the Cauchy data on the lateral boundary of the cylinder and the values of a solution \(u(x,t_{0})\) at some moment of time \(t_{0}\) given. The recovering of the function \(c(x)\) in the Helmholtz equation \(-\Delta +c(x)k^{2}=\delta(x-x_{0})\) in \(\mathbb{R}^{3}\) with the Sommerfeld radiation condition at infinity is considered in Chapter 10. The overdetermination condition is of the form \(u(x,x_{0})=F(x,x_{0})\), with \(x_{0}\) lying on some straight line and \(x\in \Gamma\), where \(\Gamma=\partial \Omega\) and \(\sigma(x)\equiv 1\) on \(\mathbb{R}^{n}\setminus \Omega\). Relying on construction of a Tikhonov-like functional, the authors propose a globally convergent numerical method for the challenging 3D travel time tomography problem with formally determined incomplete data in Chapter 11. The incompleteness means that the data are measured only on a part of the boundary of the domain of interest. Some uniqueness and stability results are also presented. Numerical algorithm and the results of numerical experiments are given in Chapter 12 in the case of the linearized problem.