PDE and applications seminar | Nancy

Abstract

As part of control theory, stabilization consists in finding a way to make stable a trajectory of a system on which one has some means of action. In this talk, we will discuss recent advances in stabilization of PDEs, starting with one of the most natural approaches for nonlinear systems, quadratic Lyapunov functions, to more complex approaches such as Fredholm backstepping. Backstepping consists in finding a control operator such that the PDE system can be invertibly mapped to a simpler PDE system for which stability is known. Surprisingly powerful, this approach offers the possibility to deal with very general classes of systems. We will review the origin of the method and present new results that resolve a question opened in 2017 and illustrate it on the rapid stabilization of the linearized water-wave equations. Finally, if time allows we will talk about a completely different subject: teaching mathematics to an AI and we will consider two questions, can we train an AI to predict the solution of a mathematical problem? can we train an AI to prove a statement?

I will present some recent work in collaboration with Elisa Davoli (TU Wien) and Katerina Nik (University of Vienna) on a three-dimensional quasistatic morpholelastic model. The mechanical response of the body and its growth are modeled by the interplay of hyperelastic energy minimization and growth dynamics. An existence result is obtained by regularization and time-discretization, also taking advantage of an exponential-update scheme. Then, we allow the growth dynamics to depend on an additional scalar field modeling nutrient concentration, and formulate an optimal control problem. Eventually, we tackle the existence of coupled morphoelastic and nutrient solutions, when the latter is allowed to diffuse and interact with the growing body.

In this talk I will provide an overview on the problem of controllability of parameter dependent systems. I will explore different control notions successfully developed through the last decade.
The aim of the control function is to steer the system to a state satisfying some properties prescribed either at some time instant T>0 or during a given time interval. These properties may be separated with respect to parameter values and can refer just to a single system itself (e.g. greedy control), or may consider solutions corresponding to the whole parameter range (e.g. ensemble control, averaged control). In the latter case control functions are designed as parameter invariant, implying a same control is to be applied to the system independently of a particular realization of the parameter, while in the first case controls vary along with the parameter. Beside the positive theoretical results, for each notion we provide a computational algorithm.

Dans cet exposé on va s’intéresser à une inégalité de Faber-Krahn inverse pour la valeur propre fondamentale $\mu_1(\Omega)$ de l’opérateur complètement nonlinéaire

\[ \mathcal{P}_N^+ u := \lambda_N(D^2 u), \]

où $\Omega \subset \mathbb{R}^N$ est un ouvert borné et convexe, et $\lambda_N(D^2 u)$ est la plus grande valeur propre de la matrice hessienne de $u$. On verra que le résultat découle de l’inégalité isopérimétrique

\[ \mu_1(\Omega) \leq \frac{\pi^2}{\text{diam}(\Omega)^2}. \]

De plus, on va discuter de la minimisation de $\mu_1$ sous différents types de contraintes. Les résultats ont été obtenus en collaboration avec Julio D. Rossi et Ariel Salort (Buenos Aires).

On considère un système de réaction-diffusion non locale décrivant l’adaptation d’un pathogène à $H$ hôtes, chacun étant associé à un différent optimum phénotypique dans $\mathbb R^n$. Le comportement en temps grand (persistance vs extinction) du problème de Cauchy associé est donné par le signe d’une valeur propre principale. Une grande partie de l’étude se concentre sur le cas $H=3$ (qui est très riche!). On compare notamment avec le cas $H=2$ et montre que la présence d’un troisième hôte peut favoriser ou entraver l’adaptation…