Search papers, labs, and topics across Lattice.
This paper explores automated design of algebraic multigrid (AMG) methods using evolutionary algorithms, specifically genetic programming (GP) guided by context-free grammars. The approach optimizes flexible cycling strategies, including level-specific smoothing and non-recursive cycling patterns, which are traditionally difficult to design manually. Experiments using the hypre library show that GP-generated AMG cycles can improve performance as both solvers and preconditioners.
Genetic programming can discover unconventional multigrid cycles that outperform hand-tuned methods, suggesting automated algorithm design can unlock untapped performance in classical numerical solvers.
Although multigrid is asymptotically optimal for solving many important partial differential equations, its efficiency relies heavily on the careful selection of the individual algorithmic components. In contrast to recent approaches that can optimize certain multigrid components using deep learning techniques, we adopt a complementary strategy, employing evolutionary algorithms to construct efficient multigrid cycles from proven algorithmic building blocks. Here, we will present its application to generate efficient algebraic multigrid methods with so-called \emph{flexible cycling}, that is, level-specific smoothing sequences and non-recursive cycling patterns. The search space with such non-standard cycles is intractable to navigate manually, and is generated using genetic programming (GP) guided by context-free grammars. Numerical experiments with the linear algebra library, \emph{hypre}, demonstrate the potential of these non-standard GP cycles to improve multigrid performance both as a solver and a preconditioner.
