An augmented Lagrangian method for parameter identifications in parabolic systems

DOI10.1006/JMAA.2001.7593MaRDI QIDQ5955572zbMATH OpenOpenAlexFDO

Publication date 13 October 2002

Published in Journal of Mathematical Analysis and Applications (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1006/jmaa.2001.7593

augmented Lagrangian functional conductivity coefficient minimization of constrained functionals saddle point

Initial-boundary value problems for second-order parabolic equations (35K20) Inverse problems for PDEs (35R30) Optimality conditions for problems involving partial differential equations (49K20) Optimality conditions for solutions belonging to restricted classes (Lipschitz controls, bang-bang controls, etc.) (49K30)

Review text

Let \(\Omega \subset {\mathbb R}^n\), \(n=1,2,3\), be a bounded domain satisfying one of the following properties: i) \(\Omega\) is convex; ii) \(\partial \Omega\) is of class \(C^{1,1}\); iii) \(\partial \Omega\) is piecewise of class \(C^2\) and for any corner \(x_0\in \partial \Omega\) there exists \(r_0>0\) such that \(B(x_0,r)\cap \Omega\) is convex. Moreover, denote \[ K=\{ q\in H^2(\Omega): \gamma_1\leq q\leq \gamma_2 {\text{ for\;a.e.\;}} x\in \Omega\}, \qquad 0 < \gamma_1\leq \gamma_2, \] and \[ W(T_1,T_2) = L^2((T_1,T_2);H^2(\Omega))\cap L^\infty((T_1,T_2);H^1_0(\Omega))\cap H^1((T_1,T_2);L^2(\Omega)), \qquad 0<T_1<T_2. \] The authors consider the problem of recovering the conductivity coefficient \(q\in K\) in the parabolic problem: \[ (1)\quad \left\{ \begin{align*}{ & D_tv - \nabla \cdot (q\nabla v) = f\quad {\text{ in }} \Omega \times (0,T],\cr & v(0,\cdot)=u_0\quad {\text{in }} \Omega,\qquad v=0 \quad {\text{in }} \partial \Omega \times (0,T].\cr}\end{align*}\right. \] The data \(f\) and \(u_0\) are assumed to belong to \(L^2((0,T);L^2(\Omega))\) and \(H^1_0(\Omega)\), respectively. The problem of recovering \(q\) is actually formulated as the problem of minimizing in \(K\times W(0,T)\) the following (non-convex) quadratic functional: \[ J(q,v) = \int_{T-\sigma}^T dt\int_\Omega \Big|D_t(v-z)- \nabla \cdot [q\nabla (v-z)]\Big|^2 dx + \beta \int_\Omega \sum_{i,j=1}^n |D_{x_i}D_{x_j}q|^2 dx,\quad \beta>0, \] under the constraint \((1)\), where \(z\in W(T-\sigma,T)\) denotes some interpolation of (finite) measurements of \(v\) in \(\Omega \times [T-\sigma,T]\) for some (small) \(\sigma\). \noindent First the authors show that the minimization problem admits (at least) a solution \((q^*,v^*)\in K\times W(0,T)\). Then they introduce the so-called augmented Lagrangian functional \[ {\mathcal L}_r(q,v,\mu) = J(q,v) + \int_0^T dt\int_\Omega \mu[D_te(q,v)- \Delta e(q,v)] dx + {r\over 2}\int_0^T dt\int_\Omega |\nabla e(q,v)|^2 dx,\quad r>0, \] where \(e(q,v)\) denotes the solution in \(W(0,T)\) to the the following parabolic problem related to a pair \((q,v)\in K\times W(0,T)\): \[ \left\{ \begin{align*}{ & D_te - \Delta e = D_tv - \nabla \cdot (q\nabla v) - f\quad \text{ in } \Omega \times (0,T],\cr & e(0,\cdot)=v(0,\cdot)-u_0\quad \text{ in } \Omega,\qquad e=0 \quad \text{ in } \partial \Omega \times (0,T].}\end{align*} \right. \] Finally, the authors show that a point \((q^*,v^*)\in K\times W(0,T)\) is a minimizer of the functional \(J\) under the constraint \((1)\) if and only if there exists \(\mu^*\in L^2((0,T);L^2(\Omega))\) such that \({\mathcal L}_r\) admits \((q^*,v^*,\mu^*)\) as a saddle point.

Recommendations

Cites work

Cited in

(16)

This page was built for publication: An augmented Lagrangian method for parameter identifications in parabolic systems

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5955572)