An NGSolve addon providing algebraic multigrid preconditioning via AMGCL.

• AMG preconditioner for NGSolve sparse matrices
• Shared memory parallelism (OpenMP)
• Multiple coarsening strategies: smoothed aggregation, Ruge-Stuben
• Multiple relaxation schemes: SPAI-0, Gauss-Seidel, ILU0, damped Jacobi
• Seamless integration with NGSolve iterative solvers

pip install git+https://github.com/cerbsim/ngs_amgcl.git

git clone --recursive https://github.com/cerbsim/ngs_amgcl.git
cd ngs_amgcl
pip install --no-build-isolation .

git clone --recursive https://github.com/cerbsim/ngs_amgcl.git
cd ngs_amgcl
mkdir build && cd build
cmake ..
make -j4 install

from ngsolve import *
import ngsolve_amgcl

# ... set up mesh, FESpace, BilinearForm a ...

# Configure AMGCL options
opts = ngsolve_amgcl.AMGCLOptions()
opts.coarsening = "smoothed_aggregation"
opts.relaxation = "spai0"

# Create preconditioner
pre = ngsolve_amgcl.AMGCLPreconditioner(a.mat, fes.FreeDofs(), opts)

# Use in a CG solver
from ngsolve.krylovspace import CGSolver
inv = CGSolver(mat=a.mat, pre=pre, printrates=True, tol=1e-8)
gfu.vec.data = inv * f.vec

• "smoothed_aggregation" (default)
• "ruge_stuben"

• "spai0" (default) - Sparse Approximate Inverse
• "gauss_seidel" - Gauss-Seidel
• "ilu0" - Incomplete LU factorization
• "damped_jacobi" - Damped Jacobi

MIT