In finite element simulations for computational mechanics, the solution of large, sparse linear systems represents the primary computational bottleneck. Direct solvers provide robust, black-box execution but scale poorly in memory for large three-dimensional problems. Conversely, scalable iterative methods are often hindered by the need for complex, problem-specific parameter tuning when dealing with heterogeneous domains, irregular discretizations, and problems involving contact, fracture, or kinematic constraints. To address these limitations, we present a black-box iterative solver framework built upon the Chronos package. By evaluating the algebraic properties and block structure of the system matrix, the framework automatically selects an appropriate combination of Krylov subspace methods and preconditioning strategies, such as Algebraic Multigrid (AMG) and the Reverse Augmented Constrained Preconditioner (RACP), without user intervention. The proposed framework integrates automated symmetry detection, block partitioning for Lagrange multipliers, and adaptive preconditioner recycling. Several benchmarks drawn from real-world applications are presented, together with scalability tests on the Leonardo supercomputer and integration within the Code_Aster suite. The results demonstrate that the framework provides a direct-solver-like interface while delivering substantial memory savings and significant speedups over state-of-the-art solvers.

An automated black-box iterative solver framework for large-scale computational mechanics

Lupi M.
;
Franceschini A.;Janna C.
2027

Abstract

In finite element simulations for computational mechanics, the solution of large, sparse linear systems represents the primary computational bottleneck. Direct solvers provide robust, black-box execution but scale poorly in memory for large three-dimensional problems. Conversely, scalable iterative methods are often hindered by the need for complex, problem-specific parameter tuning when dealing with heterogeneous domains, irregular discretizations, and problems involving contact, fracture, or kinematic constraints. To address these limitations, we present a black-box iterative solver framework built upon the Chronos package. By evaluating the algebraic properties and block structure of the system matrix, the framework automatically selects an appropriate combination of Krylov subspace methods and preconditioning strategies, such as Algebraic Multigrid (AMG) and the Reverse Augmented Constrained Preconditioner (RACP), without user intervention. The proposed framework integrates automated symmetry detection, block partitioning for Lagrange multipliers, and adaptive preconditioner recycling. Several benchmarks drawn from real-world applications are presented, together with scalability tests on the Leonardo supercomputer and integration within the Code_Aster suite. The results demonstrate that the framework provides a direct-solver-like interface while delivering substantial memory savings and significant speedups over state-of-the-art solvers.
File in questo prodotto:
File Dimensione Formato  
1-s2.0-S0045782526007358-main.pdf

accesso aperto

Tipologia: Published (Publisher's Version of Record)
Licenza: Creative commons
Dimensione 2.27 MB
Formato Adobe PDF
2.27 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11577/3622938
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? 0
  • OpenAlex ND
social impact