An iterative linear solver that constructs mutually conjugate search directions to minimize the quadratic form associated with a symmetric positive-definite matrix and its right-hand side.

Principle

Build a sequence of search directions that are conjugate with respect to the system matrix so that error components are eliminated in orthogonal subspaces, achieving rapid convergence for well-conditioned problems.

Demonstration

Applied to a sparse symmetric positive-definite system A x = b, the method produces iterates that minimize the A-weighted error and, in exact arithmetic, reaches the exact solution in at most n steps for n unknowns.

Misapplication

Running the method on a nonsymmetric or indefinite matrix without modification often yields breakdowns or incorrect convergence; specialized variants are required for those cases.

Consequence

Provides a memory-efficient solver with fast convergence for large sparse problems when the matrix meets the method's symmetry and definiteness requirements.

Reversal

Direct factorization methods (e.g., Gaussian elimination) compute the solution via matrix decomposition rather than iterative subspace minimization and trade memory for predictable arithmetic steps.

Boundary

Requires the system matrix to be symmetric and positive-definite; preconditioning is typically necessary for ill-conditioned systems and different algorithms are used outside this scope.

Semantic Tension

Versus steepest-descent: steepest-descent chooses gradients as directions and can converge slowly, while the conjugate directions here accelerate convergence by avoiding repeated reduction of the same error components.

Synthesis

An iterative algorithm that, by building conjugate directions and minimizing the associated quadratic form, efficiently solves large sparse symmetric positive-definite linear systems under appropriate preconditioning.