A recent CEEC-related research contribution received the Best Paper Award at HPCAsia 2026 for the work titled “A Matrix-Free Algebraic hp-Multigrid Method for Computational Fluid Dynamics Applications.” Overall, 143 papers were submitted to the conference, sixty-nine were accepted, and only three were selected as best paper finalists, of which CEEC’s contribution was deamed best. This recognition highlights a successful international collaboration between the RIKEN Center for Computational Science (Japan), Sandia National Laboratories (USA), and KTH Royal Institute of Technology (Sweden).
Efficiently solving large sparse linear systems remains a significant bottleneck in extreme-scale computational fluid dynamics (CFD), particularly for incompressible flows where pressure Poisson equations often dominate computational cost. Although hp-multigrid methods provide convergence independent of mesh resolution and polynomial order, most existing approaches depend on geometric coarsening or explicitly assembled matrices. These assumptions are incompatible with modern matrix-free, GPU-accelerated solvers.
The awarded paper presents a fully matrix-free algebraic hp-multigrid method that eliminates the need for matrix assembly and geometric rediscretization. The multigrid hierarchy is constructed exclusively from mesh adjacency information, making the approach suitable for high-order solvers with limited geometric data. This method integrates p-multigrid with algebraic h-coarsening and employs lightweight index mappings for restriction and prolongation, thereby maintaining efficiency without increasing memory or communication overhead.
Implemented within the Neko high-performance CFD framework, the solver maintains the matrix-free paradigm across all multigrid levels and remains compatible with accelerator architectures. Performance evaluations on the Dardel and LUMI supercomputers demonstrate robust scalability on modern GPU-based systems, achieving competitive results compared to established hybrid multigrid methods and offering improved portability.
This work highlights CEEC’s contribution, through international collaboration, to the development of robust and portable solver technologies for extreme-scale CFD. The Best Paper Award at HPCAsia 2026 underscores both the scientific quality and the broader impact of this research on next-generation high-performance computing.
For more, read the paper on the ACM website: https://dl.acm.org/doi/10.1145/3773656.3773686