WML: Numerical Methods and Fast Algorithms of Computational Electromagnetics

Presentation Abstract
Numerical methods for computational electromagnetics (CEM) are ubiquitous in design of today’s microwave and THz electronics, wireless communication links, high-speed digital interconnects and various other applied areas driving modern information and communication technologies to their new frontiers. Acceleration of these methods with fast algorithms and their deployment on heterogeneous high-performance computing platforms featuring farms of CPUs and GPUs enables the shrinking of simulation times from days to seconds, ensuring rapid virtual prototyping and drastically shrinking the time to market for today’s industrial, consumer, and defence products. Depending on the applications, sophistication of the geometric and material properties, as well as required accuracy of the simulations, differential equation-based methods such as FEM and FDTD, integral equation methods such as MoM and LCN, or high-frequency asymptotic methods such as SBR are commonly used. To ensure minimum simulation time and memory use, these methods are typically not implemented in their stand-alone form, but are used in conjunction with sophisticated sparse matrix algorithms, hierarchical compression schemes, and tensor train decompositions, and are often deployed on hybrid shared and distributed memory multiprocessors augmented with GPUs. The workshop will consist of two parts (half-day each): Part I will introduce microwave engineers and active users of commercial tools in a step-by-step manner to the underlying electromagnetic theory and algorithmic background of popular computational tools by means of a comprehensive coverage on the most popular numerical schemes such as FEM, FDTD, MoM, High-Frequency asymptotic methods and their hybridization through domain decomposition strategies. Hands-on exercises delivered through Slido platform will make Part I of the workshop interactive and engaging for the participants. It will conclude with a unified outlook at the discussed numerical methods. Part II of the workshop will target an advanced audience and introduce iterative fast algorithms in CEM, including FFT based methods and Fast Multipole Method as well as emerging fast direct algorithms based on hierarchical matrices (H- and H2-matrices) and tensor train decompositions. The relation of the material characterization to CEM modeling will be discussed in this part also. Part II will conclude with an expert panel discussion on recent advances in the use of machine-learning methods in CEM.