Strona główna » ZA » Seminars » Seminars in 2026 »
prof. dr hab. inż. Jan Sokołowski (IBS PAN i IECL) GFDS Method in Shape Optimization
Does the objective functional reach the minimum in the allowable shape class? This question concerns a problem in a functional space, even before it is discretized. A separate proof is required for the convergence of the computational method to the local minimizer. I will consider this problem for the GFDS ( gradient flow dynamical system) method, in which the negative gradient of the functional determines the normal velocity of the edge.
In the monograph written together with P. I. Plotnikov [1], we study the flow of an obstacle through a compressible fluid described by the Navier–Stokes equations. When the shape of the obstacle changes, the area of flow also changes. The convergence of the corresponding Sobolev spaces in the sense of Mosco allows us to approximate the test functions in the boundary area with the functions specified in subsequent areas. Together with the assumptions of the compactness of the class of acceptable areas, we use this property to prove the existence of the optimal shape of the obstacle.
I will present new results on GFDS on the basis of our work published in "Russian Mathematical Surveys" [2]. In it, we consider the identification of the insertion shape based on a single boundary measurement and the Bernoulli problem with free edge. In both of these planar problems, circumferential regularization is sufficient to demonstrate the existence of an optimal curve, i.e. adding a component proportional to the length of this curve to the objective functional.
The proof of the convergence of gradient flow in an infinitely dimensional space of shapes requires stronger regularization in these models. In the identification problem, we add to the Kohn–Vogelius functional a component proportional to the Willmore energy , where it denotes the curvature of the insertion edge . In Bernoulli's problem, additional regularization is provided by the Möbius energy.