Galerkin approach for estimating boundary data in Poisson equation on annular domain with application to heat transfer coefficient estimation in coiled tubes (Q2414698)

An inverse problem for the Poisson equation on an annular region in polar coordinates is considered, \begin{align*} & \lambda_w \frac{1}{r} \frac{\partial}{\partial r} \Big( r\frac{\partial T}{\partial r} \Big) +\lambda_w \frac{1}{r^2} \frac{\partial^2 T}{\partial \theta^2} + q_g = 0, \quad r_1 < r < r_\mathrm{E}, \ 0 \le \theta \le 2 \pi, \tag{a}\\ & \lambda_w \frac{\partial T}{\partial r}(r_\mathrm{E},\theta) = \alpha (T_{\mathrm{env}} -T(r_\mathrm{E},\theta)), \quad 0 \le \theta \le 2 \pi, \tag{b}\\ & -\lambda_w \frac{\partial T}{\partial r}(r_1,\theta) = Q(\theta), \quad 0 \le \theta \le 2 \pi, \tag{c} \end{align*} where \( \lambda_w \) denotes the wall thermal conductivity, and \( q_g \) is a source function. In addition, the numbers \( \alpha, \, T_{\mathrm{env}} \) and \( Q(\theta) \) denote the reciprocal heat transfer resistance, the environment temperature and the heat flux distribution along the inner boundary, respectively. The inverse problem related to (a)--(c) is to recover the Neumann data \( Q \) at the inner boundary from measured noisy temperature data at the outer boundary. For the numerical solution, the problem is decomposed as \( T = V + u \), where \( V \) solves (a)--(c) with \( q_g = 0 \) and \( T_{\mathrm{env}} = 0 \), and the function \( u \) solves (a)--(c) with \( Q = 0 \). As a preparation, in Section 2, trigonometric collocation for the forward problem related with the Laplace equation for the function \( V \) is studied. More precisely, from given data \( Q_j = Q(\theta_j) \) for \( j = 0,1,\dots,N-1 \), where \( \theta_j = 2\pi j/N \) denote the collocation points, an approximation of the form \( V_N(r,\theta) = v_0(r) + \sum_{k=1}^N v_k(r) \cos k \theta + w_k(r) \sin k\theta \) is obtained, with appropriate coefficient functions \( v_k(r), w_k(r) \) specified in the paper. Section~3 then is devoted to the inverse problem for the Laplace equation corresponding to the function \( V \), assuming that the values of \( u \) are exactly available. The discretized problem is formulated as a system of linear equations \( A_N \mathsf{Q}=\mathsf{G} \), where \( \mathsf{Q}= (Q_j)_{j=0,\dots,N-1} \) and \( \mathsf{G}= (G_j)_{j=0,\dots,N-1} \). Here, \( G_j = V_N(r_\mathrm{E},\theta_j) \) are measured data, and \( Q_j \approx Q(\theta_j) \) for \( j = 0,1,\ldots,N-1 \). A singular value decomposition of the matrix \( A_N \in \mathbb{R}^{N \times N} \) is presented, and truncated SVD and Tikhonov regularization for the stable solution of \(A_N \mathsf{Q} = G \) are considered, with the discrepancy principle as parameter choice strategy. The concluding section offers some numerical illustrations.