pyamg.relaxation#
Relaxation methods.
The multigrid cycle is formed by two complementary procedures: relaxation and coarse-grid correction. The role of relaxation is to rapidly damp oscillatory (high-frequency) errors out of the approximate solution. When the error is smooth, it can then be accurately represented on the coarser grid, where a solution, or approximate solution, can be computed.
Iterative methods for linear systems that have an error smoothing property are valid relaxation methods. Since the purpose of a relaxation method is to smooth oscillatory errors, its effectiveness on non-oscillatory errors is not important. This point explains why simple iterative methods like Gauss-Seidel iteration are effective relaxation methods while being very slow to converge to the solution of Ax=b.
- PyAMG implements relaxation methods of the following varieties:
Jacobi iteration
Gauss-Seidel iteration
Successive Over-Relaxation
Polynomial smoothing (e.g. Chebyshev)
Jacobi and Gauss-Seidel on the normal equations (A.H A and A A.H)
Krylov methods: gmres, cg, cgnr, cgne
No pre- or postsmoother
Refer to the docstrings of the individual methods for additional information.