Investigation methods for inverse problems (Q1411137)

The author is an outstanding mathematician from the Novosibirsk School and is an expert in the theory of both direct and inverse problems concerning hyperbolic differential equations as well as an expert in integral geometry. A fundamental feature in his research works consists, when dealing with linear hyperbolic equations involving distributional right-hand sides, of a skillful representation of the solution into a singular part and a regular one in the neighbourhood of the characteristic cone. In such a way the author can reduce, and solve via classical methods, inverse distributional problems to classical ones for the \textit{regular} part of the solution. The present book collects many of the author's investigations as well as others published in shortened versions and, in some sense, it can be considered as a continuation of his book ``Inverse Problems of Mathematical Physics'' [Moscow: Nauka (1984; Zbl 0576.35001)]. It should be stressed that all the problems have a physical meaning and are suggested by applications to both scientific and real life questions. \noindent As far as multidimensional inverse problems dealt with in the book are concerned, determination of solutions usually means uniqueness and continuous dependence on the data, but \textit{not} existence due to the over-determination of the additional information as well as to the difficulty of characterizing admissible data. However, some exceptions to this ``rule'' are allowed in special function classes. At any rate, also in the one-dimensional case existence usually means local existence. The book is divided into five chapters dealing with inverse problems. The first chapter has an introductory character and is devoted to exhibiting some examples of one-dimensional problems. They consist in the determination of the unknown propagation velocity in a kinematic problem, of a coefficient in a lower-order term in the string equation and of the velocity and the density in a layered medium. In Chapter 2 the basic inverse problem dealt with consists in determining the possible uniqueness, in a bounded domain \(\Omega\), of the coefficients - depending on space variables only -- in the wave -- equation \(D^2_tu-Lu=\delta(x-y)\delta(t)\) depending on the parameter \(y\in \Omega\), \(L\) being a general linear elliptic operator in divergence form. The additional information consists in prescribing some coefficients of the singular part of \(u(t,x,y)\) on the characteristic cone \(t=\tau(x,y)\) with vertex at \((0,y)\), \(\tau\) being given, as well as the values of \(u(t,x,y)\) when \((x,y)\) and \(t\) vary, respectively, on \(\partial \Omega\times \partial \Omega\) and in the interval \([0,\tau(x,y)+\varepsilon]\), \(\varepsilon\) being a positive real number. A uniqueness result holds for the principal part of \(L\) when \(L=\text{div}[a(x)\nabla]\), while for the coefficients of the lower order terms uniqueness holds up to a potential function. A similar problem is studied when in the previous hyperbolic equation a friction term \(\sigma(x)D_tu\), with an unknown coefficient \(\sigma(x)\), is added. Finally, for the equation \(D^2_tu-\Delta u + \sigma(x)D_tu +q(x)u=\delta(x-y)\delta(t)\) the so-called heterogeneity problem is studied consisting of recovering the coefficients \(\sigma\) and \(q\) assumed to be known, and constant, outside a compact domain \(\Omega\). In Chapter 3 the author is concerned with recovering the coefficient \(q(x,z)\) in the wave equation \[ D^2_tu-D_z^2u-\Delta_xu -q(x,z)u = 0,\quad (x,z,t)\in {\mathbb R}^n\times {\mathbb R}_+\times {\mathbb R} \] subject to the following conditions: \textit{(i)\ } \(u=0\) in \({\mathbb R}^n\times {\mathbb R}_+\times {\mathbb R}_-\); \textit{(ii)\ } \(D_zu(x,0,t)=-g(x)\delta'(t)\) in \((x,t)\in {\mathbb R}^n\times {\mathbb R}\); \textit{(iii)\ } \(D_zu(x,0,t)=F(x,t)\) in \((x,t)\in {\mathbb R}^n\times {\mathbb R}\). Splitting the solution \(u\) in its singular and regular parts and taking advantage of the d'Alembert formula related to the one-dimensional operator \(D^2_t-D_z^2\), the author reformulates the inverse problem as an integro-differential equation of the second kind, where the operator \(\Delta_x\) appears under the integral sign. In order to show a local existence result the author introduces a special class of functions, which are analytic with respect to \(x\) and are smooth with respect to \(z\). A similar technique works also for analogous problems related to the equations \[ D^2_tu-D_z^2u-\Delta_xu -q(x,z)D_tu = 0 \quad\text{and}\quad D^2_tu-q(x,y)(D_x^2u+D_y^2u) = 0. \] Finally, a regularization procedure is given to solve the previous problems numerically. Chapter 4 deals with identifying unknown coefficients depending on (all) space variables in multidimensional hyperbolic equations when a single information on the solution of the direct problem is available. The first differential equation involved is \[ D^2_tu-\Delta u -q(x)u = \delta(x_1)\delta(t) \quad\text{in } {\mathbb R}^n_+\times {\mathbb R}, \] where \({\mathbb R}^n_+={\mathbb R}_+\times {\mathbb R}^{n-1}\). The prescribed conditions are: \(u=f\) and \(D_nu=g\), on \(S_T=\{(x,t)\in \partial \Omega\times {\mathbb R}: x_1<t<x_1+T\}\), \(D_n\) denoting the outer normal derivative. Using energy estimates, the author can prove, for \(T\) large enough, uniqueness in \(L^2(\Omega)\) and conditional continuous dependence results for the unknown coefficient \(q\). Similar inverse problems can be solved for the equations \[ D^2_tu + \sigma(x)D_tu -\Delta u - q(x)\cdot \nabla u = \delta(x\cdot \gamma)\delta(t) \text{ and }D^2_tu - q(x)^2\Delta u = \delta(x_1)\delta(t). \] In the first problem \(\gamma\) stands for a unit vector of \({\mathbb R}^n\) and \(S_T\) is defined by \(S_T=\{(x,t)\in \partial \Omega\times {\mathbb R}: \gamma \cdot x<t<\gamma \cdot x+T\}\). In the latter problem one assumes that \(S_T=\{(x,t)\in \partial \Omega\times {\mathbb R}: \tau(x)<t<x_1+\tau(x)\}\), \(\tau\) being a suitable solution to the equation \(| \nabla \tau| =1/q\), and one has to show that the recovered function \(q\) is strictly positive, too. In this latter case the conditional continuous dependence results require the knowledge of \(\tau\) and \(D_\tau\) on a part of the boundary of \(\partial \Omega\). We stress that all the previous equations are related to wave fields generated by planar source. However, also conditional continuous dependence results for inverse problems related to equations with point sources, such as \(D^2_tu-\Delta u -q(x)u = \delta(x,t)\), can be established. In this case the two pieces of additional information must be given on \(S_T=\{(x,t)\in \partial \Omega\times {\mathbb R}: | x| <t<| x| +T\}\). Finally, in Chapter 5 the author deals with inverse problems related to the transport equation \[ \nu \cdot \nabla u(x,\nu) + \sigma(x) u(x,\nu) + \int_{S^{n-1}}K(x,\nu,\nu')u(x,\nu')\,d\nu'=f(x,\nu). \] The first result is related to the uniqueness of the pair \((\sigma,h)\) when \(F=h\) depends on \(x\) only, and some suitable additional information is available, while conditional continuous dependence results are obtained only when one of the functions \(\sigma\) or \(h\) is unknown. Estimates of the same type can also derived, in 2D and 3D, when the relaxation \(\sigma\) and the dispersion index \(K\) are unknown.