Struct HybridMultigridSolver 

pub struct HybridMultigridSolver { /* private fields */ }

Hybrid 2-level multigrid solver with a learned smoother at the fine level.

The solver performs V-cycles:

1. Pre-smooth (learned smoother)
2. Restrict residual to coarse grid
3. Solve coarse system (direct solve)
4. Prolongate correction to fine grid
5. Post-smooth (learned smoother)

If the learned smoother increases the residual, the solver falls back to classical weighted Jacobi.

Implementations

impl HybridMultigridSolver

pub fn new( a_values: Vec<f64>, a_row_ptr: Vec<usize>, a_col_idx: Vec<usize>, smoother: Box<dyn Smoother>, config: LearnedSmootherConfig, ) -> SparseResult<Self>

Create a new hybrid multigrid solver.

Arguments

• a_values, a_row_ptr, a_col_idx: fine-grid operator in CSR
• smoother: a trained (or untrained) learned smoother
• config: solver configuration

pub fn solve(&self, b: &[f64]) -> SparseResult<Vec<f64>>

Solve A x = b using iterative V-cycles.

Returns the solution vector. Falls back to Jacobi smoothing if the learned smoother causes divergence.

pub fn compare_with_jacobi( &self, b: &[f64], n_iterations: usize, ) -> SparseResult<SmootherMetrics>

Compare learned smoother convergence with classical Jacobi.

Runs both for n_iterations V-cycles and returns metrics.

pub fn dim(&self) -> usize

Problem dimension.

pub fn coarse_dim(&self) -> usize

Coarse grid dimension.