WP1: Discretization
WP1 focuses on geometric domain representations (meshes), physics-based models, and space-time discretization methods optimized for exascale computing.
1. Recent Highlights (2024-2025)
CGAL 6.0 Release
Feature-preserving alpha wrapping integrated in CGAL 6.0
City-Scale Meshing
Ktirio-geom pipeline for city-scale watertight meshes
EuroHPC Benchmarks
Thermal-bridge and ParMMG benchmarks on EuroHPC systems
Multiphysics Ready
Domain-decomposition meshing for multiphysics applications
On the Road to Exascale
Thermomechanical simulations successfully scaled to over 4000 CPU cores
2. Objectives
The key objectives are twofold:
Geometric Domain Representations
Meshes are the main starting point for simulations. WP1 develops adaptive, multiresolution, robust-to-defects, and efficient parallel representations of large-scale models supporting:
-
Valid large-scale mesh generation from CAD or measurement data
-
Adaptive mesh refinement (AMR) for structured/unstructured grids
-
Nonconforming methods for parallel efficiency
-
All-hexahedral block grids for high-order methods
Physics-Based Models
Multiphysics coupling at multiple scales in space and time requires:
-
High-order methods to increase computational intensity and reduce communications
-
Nonconforming methods designed to avoid/reduce/minimize communications
-
Parallel-in-time strategies for evolution equations
-
Adaptive coupling techniques for multiphysics/multiscale problems
4. Key Tasks
T1.1: Mesh Generation
-
Valid large-scale mesh generation from defect-laden CAD/measurement data
-
Two-level mesh generation for nonconforming domain decomposition
-
All-hexahedral block structured meshes
T1.2-T1.3: Adaptive Mesh Refinement
-
Unstructured grids: Mmg/ParMmg scalability (billions of cells)
-
Cartesian/block grids: Patch-based, cell-based, wavelet multiresolution
-
Load-balancing algorithms for dynamic adaptation
T1.4: Finite Element Exascale Framework (FEEF)
-
High-order/spectral FEM on unstructured grids
-
Full de Rham complex support
-
Nonconforming methods with optimized kernels
T1.5: Nonconforming Methods
-
Trefftz, HDG, HHO, mortar element methods
-
Improved conditioning for iterative solvers
-
Integration with domain decomposition (WP3)
T1.6: Time-Strategy
-
Parallel-in-time (parareal) for parabolic PDEs and Hamiltonian systems
-
Error control for multiscale problems with relaxation sources
-
Adaptive operator splitting and multiresolution schemes
T1.7: Multiphysics Coupling
-
Partitioned discretization with field projections and conservations
-
Adaptive high-order coupling for heterogeneous physics
-
Multiscale coupling via discontinuous Galerkin and neural networks
5. Leads & Partners
Lead Institution |
CEA |
Co-Leaders |
UNISTRA, INRIA, École Polytechnique, Sorbonne Université |
Duration |
Months 1-60 |
6. Addressed Exascale Bottlenecks
WP1 targets bottlenecks B2, B6, B7, B9, B10, B11-B13:
-
Communication reduction through high-order and nonconforming methods
-
Data movement efficiency via adaptive representations
-
Parallel scalability with AMR and domain decomposition
-
Resilience through robust mesh generation algorithms
7. Deliverables
| ID | Title | Due Dates |
|---|---|---|
D1.1-S |
Software toolboxes for mesh generation/adaptation, space-time discretization and coupling |
M24, M36, M48, M60 |
D1.2-MR |
Activity reports (included in annual report D0.2-TR) |
M12, M24, M36, M48, M60 |
D1.3-B |
Benchmarking analysis report (bottlenecks and breakthroughs) |
M12, M24, M36, M48, M60 |