HiDALGO2 took part at the ICIAM 2023 conference in Tokyo in August 2023, where Zoltán Horváth from our partner Széchenyi István Egyetem (SZE) was an invited speaker in the mini-symposium “Physics-based and data-driven modeling for digital twins” with joint work of Mátyás Constans. He presented the RedSIM solver created in HiDALGO2.
RedSIM is a CFD solver for compressible fluids of unstructured meshes and runs on multi-GPU hardware very efficiently. Its further novelty is that RedSIM has a model order reduction module, which runs in productional mode on systems with a degree of freedom of 10^8 for modelling urban air pollution. It collects data from full-order simulations, extracts the most valuable information from the snapshots by SVD, and creates a much faster simulation for the same problem with limited loss of accuracy. No other code for model reduction that works for such a large problem was presented at the conference. The audience and the discussion group acknowledged the proposed theorem, which gives sufficient conditions under which the SVD-based low-rank matrix approximations ensure the physical interpretability of the numerical procedure. Further information about the lectures of the mini-symposium can be found here: https://iciam2023.org/registered_data?id=01099#04925 n