Mixed dimensional problems on non-matching grids at the exascale: solutions, challenges, and perspectives

This presentation addresses the challenges and solutions associated with mixed-dimensional problems on non-matching grids, with a perspective towards exascale computing. Many physical phenomena involve coupled partial differential equations (PDEs) defined on domains of heterogeneous dimensions, leading to non-matching coupling at the interfaces. To tackle these problems, the reduced Lagrange multiplier method [2] provides a general mathematical framework for analysis and approximation. This approach sheds light on the stability, well-posedness, and error associated with dimensionality reduction, but requires the efficient solution of saddle point problems involving non-matching discretizations.
For the efficient numerical solution of the resulting large-scale systems, particularly on complex geometries like vascular networks embedded in biological tissues [1], we borrow a technique that was developed for non-nested multigrid methods, which offer the required flexibility by allowing the exchange of information between arbitrarily overlapping and distributed grid hierarchies with a matrix-free implementation [3], and we developed augmented Lagrangian-based preconditioners that offer optimal and scalable solutions [4].
The continued development of such robust and scalable algorithms within high-performance finite element libraries like deal.II is crucial for tackling these challenging problems at the exascale.

[1] Luca Heltai, Alfonso Caiazzo, and Lucas O. Müller, Multiscale coupling of one-dimensional vascular models and elastic tissues, Annals of Biomedical Engineering 49 (2021), 3243–3254.

[2] Luca Heltai and Paolo Zunino, Reduced lagrange multiplier approach for non-matching coupling of mixed-dimensional domains, Mathematical Models and Methods in Applied Sciences 33 (2023), no. 12, 2425–2462.

[3] Marco Feder, Luca Heltai, Martin Kronbichler, and Peter Munch, Matrix-free implementation of the non-nested multigrid method, Arxiv, 2024.

[4] Michele Benzi, Marco Feder, Luca Heltai, and Federica Mugnaioni, Optimal and scalable augmented Lagrangian preconditioners for fictitious domain problems, Arxiv, 2025.

Contatto:
paola.antonietti@polimi.it