EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Accelerating Conjugate Gradient Solvers for Homogenization Problems With Unitary Neural Operators

Julius Herb, Felix Fritzen

International Journal for Numerical Methods in Engineering · 15.03.2026 · 127 (5)

DOI: 10.1002/nme.70277 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal for Numerical Methods in Engineering, online erschienen am 06.03.2026. Verfasst von Julius Herb und 1 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 68 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Deutsche Forschungsgemeinschaft.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Rapid and reliable solvers for parametric partial differential equations (PDEs) are needed in many scientific and engineering disciplines. For example, there is a growing demand for composites and architected materials with heterogeneous microstructures. Designing such materials and predicting their behavior in practical applications requires solving homogenization problems typically governed by PDEs for a wide range of material parameters and microstructures. While classical numerical solvers offer reliable and accurate solutions supported by a solid theoretical foundation, their high computational costs and slow convergence remain limiting factors. As a result, scientific machine learning (ML) is emerging as a promising alternative, aiming to rapidly approximate solutions using surrogate models. However, such approaches often lack guaranteed accuracy and physical consistency. This raises the question of whether it is possible to develop hybrid approaches that combine the advantages of both data‐driven methods and classical solvers. To address this, we introduce UNO‐CG, a hybrid solver that accelerates conjugate gradient (CG) solvers using specially designed machine‐learned preconditioners, while ensuring convergence by construction. As a preconditioner, we propose Unitary Neural Operators (UNOs) as a modification of the established Fourier neural operators (FNOs). Our method can be interpreted as a data‐driven discovery of Green's functions, which are then used much like expert knowledge to accelerate iterative solvers. We evaluate UNO‐CG on various homogenization problems involving materials with heterogeneous microstructures and millions of degrees of freedom (DOF). Our results demonstrate that UNO‐CG enables a substantial reduction in the number of CG iterations and is competitive with handcrafted preconditioners for homogenization problems that involve expert knowledge. Moreover, UNO‐CG maintains strong performance across a variety of boundary conditions, where many specialized solvers are not applicable, highlighting its versatility and robustness, which is supported by our extensive numerical study.

Förderung

  • Deutsche Forschungsgemeinschaft

Zitieren

APA

Herb, J., & Fritzen, F. (2026). Accelerating Conjugate Gradient Solvers for Homogenization Problems With Unitary Neural Operators. International Journal for Numerical Methods in Engineering, 127(5). https://doi.org/10.1002/nme.70277

Vancouver

Herb J, Fritzen F. Accelerating Conjugate Gradient Solvers for Homogenization Problems With Unitary Neural Operators. International Journal for Numerical Methods in Engineering. 2026;127(5). doi:10.1002/nme.70277

BibTeX
@article{Herb2026,
  author = {Herb, Julius and Fritzen, Felix},
  title = {Accelerating Conjugate Gradient Solvers for Homogenization Problems With Unitary Neural Operators},
  journal = {International Journal for Numerical Methods in Engineering},
  year = {2026},
  volume = {127},
  number = {5},
  doi = {10.1002/nme.70277}
}
Kostenlos starten →