Symmetric Successive Over-Relaxation¶
A preconditioner built from matched forward and backward SOR triangular factors around an inverse diagonal, with optional relaxation parameter, for iterative linear-system solution.
Core Idea¶
SSOR symmetrizes the logic of successive over-relaxation. A matrix split into diagonal, lower, and transposed upper parts yields a factorized preconditioner applied through forward and backward triangular solves rather than dense inversion.
The diagonal must be suitable and the splitting convention explicit. The relaxation parameter changes scaling and performance; SSOR can improve convergence but its sequential sweeps, ordering sensitivity, and spectral effect must be evaluated within the chosen iterative solver.
Scope of Application¶
- Numerical linear algebra. Preconditions sparse systems.
- Conjugate-gradient methods. Provides a symmetric positive-definite-compatible form under conditions.
- PDE solvers. Uses matrix locality and triangular sweeps.
- Iterative-method analysis. Studies ordering and spectral effects.
Clarity¶
State sign and matrix splitting, symmetry/definiteness, scalar or block D, ordering, ω convention and range, left/right/split preconditioning, triangular-solve accuracy, solver, stopping rule, and convergence comparison. Inclusion test: Require a declared symmetric matrix splitting, invertible diagonal/block diagonal, forward and backward triangular factors, valid relaxation parameter, and use as a preconditioner or symmetric sweep. Exclusion test: Exclude ordinary SOR with only a forward sweep, Jacobi diagonal scaling, incomplete Cholesky assumed identical, and the exact system matrix confused with M. Nearest boundary: Gauss–Seidel uses one triangular direction; symmetric Gauss–Seidel/SSOR combines forward and backward directions, with SSOR conventions adding relaxation scaling. Exit condition: The method leaves the class if the paired transpose sweep is absent or the splitting/parameter makes the factorization ill-defined. Common misclassifications: It is not one forward SOR sweep. It is not Jacobi scaling. It is not generally the exact inverse of A. It is not automatically identical to incomplete Cholesky. Nearest named distinctions: SOR iteration: Is a stationary iteration, not necessarily a symmetric preconditioner. Symmetric Gauss–Seidel: Is closely related and often the ω=1 convention. Incomplete Cholesky: Builds approximate factors through elimination. Jacobi preconditioner: Uses only the diagonal.
Manages Complexity¶
The method turns a stationary relaxation sweep into a symmetric factorized operator whose usefulness depends on matrix structure, ordering, tuning, and hardware execution.
Abstract Reasoning¶
- Verify matrix assumptions and choose ordering.
- Form D, L, and matching transpose under one convention.
- Select a valid relaxation parameter.
- Apply factors as triangular solves.
- Measure preconditioned convergence and computational cost.
Knowledge Transfer¶
SSOR settings transfer only with comparable matrix scaling, ordering, sparsity, definiteness, diagonal blocks, solver, and architecture; ω and iteration gains are problem-dependent.
Relationships to Other Abstractions¶
Current abstraction Symmetric Successive Over-Relaxation Domain-specific
Parents (1) — more general patterns this builds on
-
Symmetric Successive Over-Relaxation presupposes Iterative method Domain-specific
Symmetric Successive Over-Relaxation presupposes Iterative method because its preconditioner is built from the forward and backward factors of an iterative SOR scheme.
Hierarchy path (1) — routes to 1 parentless root
- Symmetric Successive Over-Relaxation → Iterative method → Iteration
Neighborhood in Abstraction Space¶
Symmetric Successive Over-Relaxation sits in a crowded region of the domain-specific corpus (32nd percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.
Family — Physical & Geometric Dynamical Quantities (29 abstractions)
Nearest neighbors
- Matrix Multiplication — 0.90
- Symbolic Cholesky Decomposition — 0.89
- Low-rank matrix approximations — 0.88
- Parabolic Cylindrical Coordinates — 0.88
- Matrix equivalence — 0.88
Computed from structural-signature embeddings · 2026-10-08