Public Thesis Defense of Pierre BALTY - IMMC
sst |
Development and assessment of an adaptive multiresolution Vortex Particle-Mesh method for massively parallel simulations of unbounded incompressible flows
Monday September 29, 2025 - 4:15pm - Room BARB93 - Place Sainte-Barbe 1, 1348 Louvain-la-Neuve
Vortical flows arise in many areas, from aircraft wakes to environmental mixing processes. Understanding their dynamics is essential for improving industrial efficiency and deepening knowledge of natural phenomena.
In such a context, high-fidelity simulations provide a powerful tool for capturing these flows with unprecedented accuracy. Among the existing approaches, the Vortex Particle-Mesh (VPM) method is particularly well-suited, as it has low numerical errors and natural treatment of unbounded domains while maintaining a moderate computational cost. However, the wide disparity in spatial scales, combined with the intrinsic incompressibility of these flows, still leads to a significant computational cost that often limits the use of simulations for large-scale problems.
This thesis aims to reduce this cost by working on two distinct fronts: minimizing the algorithmic overhead on large supercomputers and reducing the need for computational resources. First, enforcing incompressibility requires solving a Poisson equation, which entails global communications across all the computational units. This leads to significant communication overhead, especially when scaling to thousands of cores. This work hence introduces new implementations of a Fast Fourier Transform–based Poisson solver that substantially reduce its time-to-solution compared to existing frameworks. Second, the computational efficiency can be further enhanced by focusing the resources only where needed. To that end, this thesis develops a VPM framework that integrates a wavelet-based adaptive multiresolution approach, which dynamically adjusts the number of computational elements to the relevant physical scales. The framework is then validated through a series of test cases, demonstrating both its accuracy and efficiency.
Jury Members :
Prof. Philippe Chatelain (UCLouvain)(Promoteur)
Prof. Renaud Ronsse (UCLouvain) (Président)
Prof. Jean-François Remacle (UCLouvain) (Secrétaire)
Prof. Grégoire Winckelmans (UCLouvain)
Prof. Wim van Rees (Massachusetts Institute of Technology)
Prof. Philippe Poncet (Université de Pau et des Pays de l’Adour)