Header Logo

Junior Research Groups

To strengthen the support for excellent young researchers and to give them a head start in building their own research profile within the supportive environment of a large center, ECMath is funding three positions for heads of Junior research Groups in application driven mathematical research. The group leaders are also part of the BMS Postdoctoral Faculty.

Optimization under Uncertainty

PI: Dr. Guillaume Sagnol
Technische Universität Berlin
Institute for Mathematics
Straße des 17. Juni 136
D-10623 Berlin, Germany
Phone: +49 (0) 314 78846
Fax.: +49 (030) 314 21117
email: sagnol(at)math.tu-berlin.de

The management of an operating theatre in hospitals, or of a supply chain network, raises extremely challenging mathematical decision problems. First, such planning and scheduling problems typically have a combinatorial nature (which resource should be allocated to which activity ?) which leads to an explosion of the number of solutions. Second, long-term decisions are impacted by future events which are often hard to foresee or to quantify. This is the case in the health care sector, where decision makers must deal with emergency arrivals or uncertain treatment durations. The sources of uncertainty are also many and varied in the industrial sector, e.g. the future demands of goods must be predicted, and service disruptions can affect a distribution network.

It is therefore essential to compute planning strategies that cope with the uncertainty. Decision makers now have a very large amount of data available, but there is a need for intelligent systems that extract the useful information to handle the uncertainty and to make sound planning decisions. To achieve this goal, we develop new data-driven robust optimization frameworks. We are particularly interested in solution strategies which allow a certain amount of flexibility, in order to react adequately to unexpected events, and which take the correlations between those events into account.

Numerics and Optimization of Robust Equilibria

PI: Dr. Carlos Rautenberg
Humboldt-Universität zu Berlin
Institute for Mathematics
Unter den Linden 6
D-10099 Berlin, Germany
Phone: +49 (0)30 2093 5497
Fax: +49 (0)30 2093 5859
email: carlos.rautenberg(at)math.hu-berlin.de

In particular, Quasi-variational Inequalities (QVIs) stand out as an specific type of these problems with a profound impact on the formulation of physical and economical phenomena. They represent a generalization of the usual constrained optimization problem with the addition that the constraint also depends on the variable of interest. Although some success has been obtained in the theoretical description of solutions, there are still many open questions in analytical aspects and there is a scarcity of numerical methods that guarantee approximation of solutions due to the highly nonlinear nature of QVIs.

Generally, as controllable variables are available, optimization problems for QVIs arise naturally as certain criteria is required to be minimized or satisfied. A successful development of solution algorithms, optimization techniques and control approximation schemes would provide tools to solve problems that go from optimally denoising biomedical images to large scale problems as the prevention of floods in sensitive regions.

As a specific problem, we highlight the optimal placement of sensors for the estimation of the temperature distribution in buildings, which also possesses multi-scale features and where stochastic perturbations are present. The goal here is to properly locate sensors within certain regions to reliably estimate the temperature distributions in other areas. A successful resolution to this problem would substantially improve the energy efficiency in the control of the of the temperature in buildings.

Artificial Photosynthesis / Scientific Computing

PI: Dr. Sebastian Matera
Freie Unversität Berlin
Institute for Mathematics
Arnimallee 6 (Raum 013)
D-14195 Berlin, Germany
Phone: +49 (0)30 838 75335
Fax: +49 (0)30 838 75412
email: matera(at)math.fu-berlin.de

On a technical scale, the central interest of chemistry is the macroscopic yield of one or more chemical compounds. On the other hand, chemical bonding and reaction are best understood on a sub-nanometer scale.

In order to utilize this understanding on a macroscopic scale, we are developing and improving electronic structure based multi-scale modeling strategies. The focus is on stochastic mesoscopic models of surface reactivity, which link microscopic and macroscopic theories, and their integration into the numerical solution of the partial differential equations governing macroscopic, reactive flow.

Different numerical techniques are employed and developed/improved like stochastic simulation or high dimensional interpolation/integration. Special emphasis is put into the development of user-friendly software. With this, we address a number of problems, e. g. the error propagation during coarse-graining or the identification of those microscopic aspects controlling macroscopic behavior.