A sequence of elementary row (and optionally column) operations applied to a matrix to transform it into row-echelon or reduced row-echelon form, used to solve linear systems, compute rank, find inverses, and obtain bases for solution spaces by systematic elimination of variables.

Principle

Elementary row operations (swap rows, scale a row by a nonzero scalar, add a scalar multiple of one row to another) preserve the solution set of the corresponding linear system; applying them systematically removes unknowns step by step to triangularize the coefficient matrix.

Demonstration

Solve a 3×3 linear system: form the augmented matrix, use row operations to produce an upper triangular echelon form, perform back-substitution to obtain the unique solution when pivots are nonzero; in the singular case the procedure reveals dependencies and parametrizes the solution space.

Misapplication

Dividing by a pivot that is zero without first permuting rows, or treating floating-point pivoting carelessly (omitting partial/complete pivoting) can produce division-by-zero errors or severe numerical instability and incorrect numerical solutions.

Consequence

Correct application yields row-echelon (or reduced) form, from which one reads rank, determines solvability, computes inverses when full rank, and obtains explicit parametric descriptions of solution spaces; it underlies LU factorization and many direct linear solvers.

Reversal

The converse is to perform arbitrary non-invertible row operations (scaling by zero, nonreversible transforms) which destroy solution equivalence; undoing elimination requires recording row operations or storing factors (LU) because the transformation is not uniquely recoverable from the triangular form alone.

Boundary

Applies to linear systems over fields (or division rings with care); over rings without division (like integers) direct application fails without adaptations (Smith normal form). In numerical contexts stability and pivoting strategy are essential limits on direct use.

Semantic Tension

Tension exists with iterative methods: elimination is a direct, finite procedure ideal for small-to-moderate dense systems, while iterative solvers (Conjugate Gradient, GMRES) are preferable for very large sparse or ill-conditioned systems due to memory and complexity tradeoffs.

Synthesis

Gaussian elimination is the elementary-row-operations procedure that reduces a linear system to triangular (or reduced) form, exposing rank and solvability and enabling direct solution, matrix inversion, and decomposition into LU factors when applicable.