Time-X project: TIME parallelization for eXascale computing and beyond

Nowadays digital twins are commonly designed and investigated in industry before physical products are manufactured. To construct such virtual prototypes multiobjective optimization of several key performance indicators from different physical domains is performed. For a wide range of applications, e.g., electric motors, high-fidelity transient simulations are necessary, as they allow for a precise description of the system’s evolution in time. These time-domain calculations are typically very time consuming, which becomes especially critical when a large number of scenarios or operating modes have to be simulated. This urges the need for novel efficient algorithms, suitable for modern and future high-performance computing (HPC) systems.

Figure: Sketch of parallel-in-time method Parareal.

The growth in computational power of the HPC architectures aroused attention for parallel-in-time (PinT) methods, when the parallelization based on the traditional decomposition of the spatial domain reached its saturation limit. Exploiting the immense potential of PinT integration to unlock the performance of exascale systems, the recently initiated (1 April 2021) project Time-X focuses on massively parallel HPC solution of real-life demands in electromagnetics, medicine, weather and climate. Time-X aims at the development of software and novel algorithmic concepts for PinT simulation of the multi-scale phenomena, optimization and uncertainty quantification problems in the chosen diverse applications fields.

Contact:

Iryna Kulchytska-Ruchka (Computational Electromagnetics Group, Technische Universität Darmstadt)

%d bloggers like this: