On Friday 18 September 2026 at 2.30 pm,
Lorenz Gillner, University of Wismar,
will give a lecture as part of his planned PhD thesis, entitled
Reliable Methods on Graphics Processing Units in the Context of Energy Systems
The lecture will take place in a hybrid format.
IQON, Industriestr. 11, Room 1-Meet and https://us02web.zoom.us/j/2972318676?pwd=dmVIZjBOM2ltYjRNYkduS2JXYUJaUT09
The lecture will be delivered in German.
Abstract:
Uncertainties are ubiquitous in scientific computing. From measurement inaccuracies and
model uncertainty to discretisation errors arising from the use of computers with finite
precision, uncertainty affects the reliability of computer-aided calculations. Greater
reliability can be achieved by treating uncertainties as an integral part of the system.
Set-theoretic techniques, such as interval methods, enable calculations involving quantities subject to uncertainty
that are nevertheless verified. At the same time, growing demands on computer systems
are driving a transformation of traditional computer architecture, thereby enabling new approaches to
problem-solving. Graphics processing units (GPUs), for example, have long since ceased to be used exclusively in the
field of computer graphics. Thanks to their ability to provide hardware-based acceleration
of parallelisable operations, they have evolved into an integral part of modern heterogeneous
high-performance computing systems. However, computations on co-processors such as
GPUs remain largely unverified. Established software and algorithms must therefore be reconsidered from a new
perspective in order to achieve a computational advantage—both in terms of the
problem complexity and the efficiency of the solution. The aim of this PhD project is the application
of GPU-accelerated interval methods for the reliable simulation and optimisation of dynamic
systems. This talk focuses on the further development of conventional interval software for
heterogeneous computing systems and demonstrates how the use of the GPU can counteract the practical hurdles associated with
interval methods. The methods developed are tested for their practical suitability using the example of
parameter estimation for a battery cell.
Supervisors: Prof. Dr.-Ing. habil. Andreas Rauh (UOL) and Prof. Dr. rer. nat. habil. Ekaterina Auer (Wismar University of Applied Sciences)