Algebraic multigrid methods (AMG) offer scalable solvers for large-scale simulations, typically with complexity for unknowns. However, AMG efficiency depends on choices of — smoothers, cycling strategies, intergrid operators, etc. — which are highly problem-dependent. In this talk, we present AMG preconditioners automatically designed using grammar-guided genetic programming (G3P). We formulate grammar rules that generate AMG methods with flexible cycles, i.e., smoothers, relaxation weights, and intergrid operators are chosen independently for each step within the cycle. Our approach extends beyond standard V-, W-, or F-cycles, providing a larger search space with potential superior solver choices [1]. These flexible AMG methods are implemented within the hypre software library. Here, we design automated AMG for a physically realistic laser beam welding process, where the thermomechanical behavior is modeled with a nonlinear coupled system of time-dependent thermoelasticity equations [2]. Flexible AMG preconditioners, optimized with G3P, are used to accelerate the convergence of the GMRES method applied to solve the linearized systems (via Newton’s method). We compare the BoomerAMG from hypre, available in PETSc, with our flexible G3P-designed AMG preconditioner. Our numerical results show significant improvements in terms of solving times and convergence rates. Moreover, our results show that flexible AMG is robustly scalable in the sense that efficient flexible AMG cycles are constructed with a small example problem and then efficiently used for larger and realistic problems. REFERENCES [1] Jonas Schmitt, Sebastian Kuckuk, and Harald Köstler. 2021. EvoStencils: a grammar-based genetic programming approach for constructing efficient geometric multigrid methods. Genetic Programming and Evolvable Machines 22, 4 (Dec 2021), 511–537. https://doi.org/10.1007/s10710-021-09412-w [2] Tommaso Bevilacqua, Axel Klawonn, Martin Lanser, & Adam Wasiak. (2024). Monolithic overlapping Schwarz preconditioners for nonlinear finite element simulations of laser beam welding processes. https://doi.org/10.48550/arXiv.2407.03230