Implicit Taylor methods for parabolic problems with nonsmooth data and applications to optimal heat control (Q817472)

The parabolic boundary value problem and its numerical discretization is considered. The authors focus on the linear parabolic heat equation of a two space dimensional domain with natural Robin boundary condition but with nonsmooth data. In this case, where the jump effects of high-frequency oscillations are observable over a long time horizon, widely used time integration methods like backward Euler or trapezoidal rule do not give the appropriate results. The authors present implicit Taylor methods (ITM) as an effective tool for problems with non-smooth data. The problem is solved by semidiscretization via the finite element method in space using standard techniques and the resulting \(N\)-dimensional linear initial value problem of ordinary differential equations solved by ITM, especially they are focused on values \(q=1,2,3\) for ITM. For each ITM-\(q\) step for solving large-scale linear algebraic systems they use preconditioned conjugate gradient methods, which are very effective in this case. For the case \(q=2,3\) they derive the optimal preconditioning parameters. Finally the application of the ITM method to an optimal heat control problem is investigated in this article. They use again the ITM-\(q\) technique to the semidiscrete optimal control problem. For the numerical solution of the discretized optimization problem quasi-Newton methods like Broyden's symmetric update or DFP techniques are very efficient. The advantage of these methods is that they require gradient evaluations and the implementation of the adapted step size rules. In case of boundary controls for simplicity a specific penalty approach has been used. The efficiency of the proposed algorithm is confirmed on several test examples which create the final part of the article.