We warmly congratulate Nils-Christian Kempke on the successful defense of his dissertation titled "Structure- and Hardware-Aware Algorithms for Large-Scale Linear and Mixed-Integer Linear Optimization" on September 17, 2026, at Technische Universität Berlin!
Unprecedented Opportunities for Complex Energy System Models
With the ongoing energy transition and sector coupling, the demands on mathematical modeling and optimization are rising rapidly. Energy system models of the future are characterized by ever-increasing complexity, massive dimensions, and high spatial and temporal resolutions. Traditional optimization approaches are increasingly reaching their performance limits.
In his dissertation, Nils-Christian Kempke focuses on the development of structure- and hardware-aware algorithms for large-scale linear and mixed-integer linear optimization problems. His research results open up unprecedented possibilities:
- Massively parallel and structure-aware solvers: Through distributed, parallel presolving methods and massively parallel interior-point methods specifically for LP problems with arrowhead structure, highly structured optimization tasks can be solved scalably on modern high-performance computers.
- Hardware acceleration (GPU): Using GPU-accelerated variants of specialized algorithms (such as for the market-split problem), his research efficiently exploits the parallel computing power of modern graphics processors.
- Efficient heuristics for large-scale models: By using low-precision first-order (FO) LP solutions within fix-and-propagate heuristics as well as specialized heuristic techniques for unit commitment models, high-quality solutions for extremely large optimization problems can be found in a fraction of the usual computing time. Since the FO part is ideally suited for GPU acceleration, this approach combines CPU and GPU architectures into a future-proof, hybrid hardware strategy.
The methods developed by Nils-Christian thus establish the mathematical and technological foundation to make even highly complex, next-generation energy system models computationally feasible.
The research work was funded within the framework of the UNSEEN project by the Federal Ministry for Economic Affairs and Energy (BMWE / former BMWK). Part of the work was also conducted at the Research Campus MODAL, funded by the BMFTR. In the EnergyLab, researchers at the Zuse Institute Berlin (ZIB) work alongside industry partners at the intersection of applied mathematics, algorithmic optimization, and real-world energy industry challenges to actively drive the decarbonization of infrastructure. The EnergyLab warmly congratulates Nils-Christian on this outstanding scientific achievement! The EnergyLab will continue along the path paved by Nils toward specialized solvers that are optimally adapted to state-of-the-art hardware.
