Inverse Heat Conduction Problems (IHCPs) have been extensively studied over the last 50 years. They have numerous applications in many branches of science and technology. The problem consists in determining the temperature and heat flux at inaccessible parts of the boundary of a 2-or 3-dimensional body from corresponding data -called 'Cauchy data' -from accessible parts of the boundary. It is well-known that IHCPs are illposed which means that small perturbations in the data may cause large errors in the solution.In this contribution we give a short overview over our contributions to multidimensional IHCP's and indicate what computational results for which sort of problems we have obtained.
Statement of the problemIn [4] we have established the theoretical background for multidimensional inverse heat conduction problems. In this contribution, we outline and refer to several of our 2-d. calculations.The importance of inverse heat conduction problems and appropriate solution algorithms are established in numerous works (see, e.g. In a 2-d. IHCP we want to determine the temperature and the heat flux on an inaccessible part of the boundary of a 2-d. domain from corresponding data on another (accessible) part (s. Fig. 1). The data on the latter are called Cauchy data. The underlying problem is the heat equation which can have the more general form studied in [4]. The solutions of the associated direct problem as well as the inverse problem are understood in weak sense (see [4]). This allows the Cauchy data to be only from L 2 . The initial condition can be given or not. In the latter case, our method is also able to identify the initial temperature distribution.The idea of our method is very simple: since the initial condition v 0 and the heat flux v 1 = ∂u/∂N | S 2 at the inaccessible part Γ 2 of the boundary are not known, we consider them as a control v = (v 0 , v 1 ) to minimize the defecton the accessible part of Γ 1 of the boundary; here, we set S i = Γ i × (0, T ], i = 1, 2, where T > 0 is the final time. In [4] we have proved the existence of the optimal control, and also obtained the gradient of the defect functional by means of an appropriate adjoint problem. Since the optimal control problem is still unstable, we have to use a regularization method for it.
RegularizationWe solve the underlying inverse problem by discretization in combination with Tikhonov's regularization using a zeroth order penalty term as well as iterative regularization via an appropriate stopping rule.As the underlying operator we choose the Neumann-to-Dirichlet mapping A :which maps the (unknown) initial function v 0 and the heat flux v 1 = ∂u/∂N |S 2 to u| S 1 where u is the solution of the heat equation in weak form.The minimizing functional to determine v = (v 0 , v 1 ),is differentiable with gradient