Michael Hinze

Michael Hinze, Martin Kunkel

We consider integrated circuits with semiconductors modeled by modified nodal analysis and drift-diffusion equations. The drift-diffusion equations are discretized in space using mixed finite element method. This discretization yields a high dimensional differential-algebraic equation. We show how proper orthogonal decomposition (POD) can be used to reduce the dimension of the model. We compare reduced and fine models and give numerical results for a basic network with one diode. Furthermore we discuss an adaptive approach to construct POD models which are valid over certain parameter ranges. Finally, numerical investigations for the reduction of a 4-diode rectifier network are presented, which clearly indicate that POD model reduction delivers surrogate models for the diodes involved, which depend on the position of the semiconductor in the network.

Ahmad Ahmad Ali, Klaus Deckelnick, Michael Hinze

An optimal control problem subject to an elliptic obstacle problem is studied. We obtain a numerical approximation of this problem by discretising the PDE obtained via a Moreau--Yosida type penalisation. For the resulting discrete control problem we provide a condition that allows to decide whether a solution of the necessary first order conditions is a global minimum. In addition we show that the corresponding result can be transferred to the limit problem provided that the above condition holds uniformly in the penalisation and discretisation parameters. Numerical examples with unique global solutions are presented.

Michael Hinze, Tran Nhan Tam Quyen

In this paper we study the inverse problem of identifying the diffusion matrix in an elliptic PDE from measurements. The convex energy functional method with Tikhonov regularization is applied to tackle this problem. For the discretization we use the variational discretization concept, where the PDE is discretized with piecewise linear, continuous finite elements. We show the convergence of approximations. Using a suitable source condition, we prove an error bound for discrete solutions. For the numerical solution we propose a gradient-projection algorithm and prove the strong convergence of its iterates to a solution of the identification problem. Finally, we present a numerical experiment which illustrates our theoretical results.

Michael Hintermüller, Michael Hinze, Denis Korolev

In this work, we propose a novel layerwise adaptive construction method for neural network architectures. Our approach is based on a goal--oriented dual-weighted residual technique for the optimal control of neural differential equations. This leads to an ordinary differential equation constrained optimization problem with controls acting as coefficients and a specific loss function. We implement our approach on the basis of a DG(0) Galerkin discretization of the neural ODE, leading to an explicit Euler time marching scheme. For the optimization we use steepest descent. Finally, we apply our method to the construction of neural networks for the classification of data sets, where we present results for a selection of well known examples from the literature.

Harald Garcke, Michael Hinze, Christian Kahle

We propose to model physical effects at the sharp density interface between atmosphere and ocean with the help of diffuse interface approaches for multiphase flows with variable densities. We use the variable-density model proposed in \cite{m6:AbelsGarckeGruen_CHNSmodell}. This results in a Cahn-Hilliard/Navier-Stokes type system which we complement with tangential Dirichlet boundary conditions to incorporate the effect of wind in the atmosphere. Wind is responsible for waves at the surface of the ocean, whose dynamics have an important impact on the $CO_2-$exchange between ocean and atmosphere. We tackle this mathematical model numerically with fully adaptive and integrated numerical schemes tailored to the simulation of variable density multiphase flows governed by diffuse interface models. Here, {\it fully adaptive, integrated, efficient, and reliable} means that the mesh resolution is chosen by the numerical algorithm according to a prescribed error tolerance in the {\it a posteriori} error control on the basis of residual-based error indicators, which allow to estimate the true error from below (efficient) and from above (reliable). Our approach is based on the work of \cite{m6:HintermuellerHinzeKahle_adaptiveCHNS,m6:GarckeHinzeKahle_CHNS_AGG_linearStableTimeDisc}, where a fully adaptive efficient and reliable numerical method for the simulation of two-dimensional multiphase flows with variable densities is developed. We incorporate the stimulation of surface waves via appropriate boundary conditions.