![\"Siegel:](\"https:\/\/www.adlershof.de\/fileadmin\/user_upload\/bilder\/logos\/HU_Siegel_lang.svg\")
<\/a>Scalable Wave Solvers in Large-Scale Scientific Computing: From Convergence to Energy Efficiency\u00a0Prof. Vandana Dwarka<\/p>\nDelft Institute of Applied Mathematics, Delft University of Technology, The Netherlands<\/p>\n
Simulating complex wave phenomena requires solving massive systems of partial differential equations (PDEs). As model fidelity and grid resolutions increase, standard iterative numerical methods stall. Systems become severely ill-conditioned, creating prohibitive bottlenecks in both computational time and raw energy consumption. While recent advancements in Artificial Intelligence and Physics-Informed Neural Networks (PINNs) offer new pathways for surrogate modeling, these methods frequently lack the robustness required for high-fidelity physics and often struggle to capture the high-frequency oscillatory behavior inherent in these PDEs.<\/p>\n
This talk examines robust scalable solver paradigms designed to address these fundamental limitations. We focus on the mechanics of multilevel solvers\u00a0and advanced preconditioning strategies, such as deflation and shifted operators. These techniques are essential for achieving scalable or grid-independent convergence, ensuring that solver performance remains stable as problem sizes scale toward the exascale regime. Finally, we discuss the relationship between algorithmic efficiency and hardware implementation, demonstrating how minimizing communication overhead and iteration counts directly translates into reduced energy costs and improved sustainability in large-scale scientific computing.<\/p>\n