Identification of parameters in ill-posed linear parabolic equations (Q1882462)

The aim of this paper is to study the identification of conductivities \(q\) in the linear parabolic equation \[ \begin{cases} {\partial y\over\partial t}- \nabla(q(x)\nabla y)= f(x,t)\quad &\text{in }Q= \Omega\times (0,T),\\ y(x,t)= 0\quad &\text{on }\Sigma= \partial\Omega\times (0,T),\end{cases}\tag{1} \] where \(f\in L^2(Q)\) is a given function, and \(q\) is a function taken from the set \[ K= \{q\in w^{1,\infty}(\Omega)\mid 0< \alpha_1\leq q(x)\leq \alpha_2\text{ for a.e. }x\in\Omega\}. \] Let \(\omega\subset \Omega\) be an open subset and \(Q^\omega= \omega\times (0,T)\). For any \(\delta> 0\) let \(Q_\delta= \Omega\times (\delta, T)\) and \[ Y= \{y\in L^2_{\text{loc}}(0, T; L^2(\Omega))\mid y\in H^{2,1}(Q_\delta)\text{ for each }\delta\in (0,T)\}, \] where \[ H^{2,1}(Q_\delta)= \Biggl\{y\mid y,\, {\partial y\over\partial t},\, {\partial y\over\partial x_i},\;{\partial^2 y\over\partial x_i\partial x_j}\in L^2(Q_\delta),\, i,j= 1,\dots, n\Biggr\}. \] The main interest is to identify \(q\) from the set \(K\) via the observation \(z(x,t)\in L^2(Q^\omega)\) on the subset \(\omega\subset\Omega\). More precisely, the authors study the following identification problem \[ \text{Min\,}J(y,q)= {1\over 2} \int_{Q^\omega}| y- z|^2 \,dx\,dt+{1\over 2}\| q\|_{W^{1,\infty}(\Omega)}\tag{P} \] over all \((y,q)\in Y\times K\) satisfying (1). It is shown that there exists at least one solution to problem (P). The problem (P) is an inverse problem, while the direct problem is to determine the values of the solutions to (1) for a given \(q\). Both the inverse and direct problems are ill-posed because of the absence of initial data. An approximation scheme for problem (P) is given as follows. Let \(Y_0= \{y\in Y\mid y= 0\) on \(\partial\Omega\times (0,T)\}\) and define \(J_\varepsilon: Y_0\to [0,+\infty)\) by \[ J_\varepsilon(y,q)= J(y,q)+ {1\over\varepsilon} \int_Q\Biggl({\partial y\over\partial t}- \nabla(q\nabla y)- f\Biggr)^2\,dx\,dt. \] The following approximation problem is considered \[ \text{Min\,}J_\varepsilon(Y,q)\quad\text{over all }(y,q)\in Y_0\times K,\quad \varepsilon> 0.\tag{P\(_\varepsilon\)} \] If \((y_\varepsilon,q_\varepsilon)\) is a solution to problem \((\text{P}_\varepsilon)\), then there is a sequence \(\{(y_{\varepsilon_n}, q_{\varepsilon_n})\}^\infty_{n=1}\) with \(\lim_{n\to\infty} \varepsilon_n= 0\), such that, as \(n\to\infty\), \[ \begin{gathered} y_{\varepsilon_n}\to y^*\quad\text{weakly in }L^2(Q^\omega)\cap H^{2,1}(Q_\delta),\\ \text{strongly in }C([\delta, T]; H_0'(\Omega)),\quad\forall\delta\in (0,T),\\ q_{\varepsilon_n}\to q^*\quad\text{weakly star in }W^{1,\infty}(\Omega),\end{gathered} \] and \[ \lim_{n\to\infty}\,J_{\varepsilon_n}(y_{\varepsilon_n}, q_{\varepsilon_n})= J(y^*, q^*). \] Moreover, \((y^*, q^*)\) is a solution to problem (P).