Row Reduction

Precise definition

Row reduction uses three reversible operations: interchange two rows, multiply a row by a nonzero scalar, and add a multiple of one row to another. These operations create a row-equivalent matrix, usually REF or RREF.

Notation and mathematical language

Row equivalence preserves the solution set of $Ax=b$, the null space of a coefficient matrix, row space, and rank. It does not preserve the literal column space, determinant value without tracking, eigenvalues, or the represented linear transformation.

Conceptual picture

Each operation replaces equations by equivalent linear combinations. RREF exposes pivot variables and free variables, while the nonzero rows of RREF form a basis of the row space.

Conditions and key results

Scaling by zero is not reversible. When extracting a column-space basis, record pivot indices and return to original columns. Numerical implementations need pivoting and tolerance choices; exact symbolic zeros and floating approximations differ.

A reliable strategy

1. Write the matrix or augmented system and choose a pivot in the leftmost available nonzero column.
2. Swap if necessary, scale if convenient, and clear entries below; continue to REF.
3. For RREF, make pivots one and clear entries above them.
4. Use pivot positions for rank and variables, then verify any requested subspace basis in the original matrix.

Fully worked example

Interpretation and application

Row reduction solves systems and computes rank, inverses, kernels, and bases. It is exact as an algebraic procedure; with measured or floating data, conditioning and tolerance determine whether a near-zero pivot should be treated as zero.

Common mistakes

Verification and reasonableness

• Reverse operations conceptually or multiply by elementary matrices.
• Check RREF structural conditions and unique pivot locations.
• Compare rank with determinant for a square matrix or with an independent factorization.

Practice

1. Does row replacement change a system's solutions?
2. Do row operations preserve column space?
3. Where are row-space basis vectors found after reduction?

Further deduction

RREF is a canonical representative of a row-equivalence class: two matrices are row-equivalent exactly when they have the same RREF. This uniqueness makes it a reliable comparison tool even though the sequence of row operations is far from unique. REF alone does not have this canonical property.

Row reduction also solves matrix equations. To find all $X$ with $AX=B$, treat each column of $B$ as a right-hand side; the same row operations or factorization apply to every column. Consistency can differ by column, and the homogeneous null-space freedom appears independently in each solution column.

Pivot choice is irrelevant to the exact reduced row-echelon form but important in floating-point computation. Partial pivoting swaps a large available entry into the pivot position to avoid division by a tiny number and control rounding growth. Exact hand problems can choose pivots for simple arithmetic; numerical solvers track stability as well as algebraic equivalence. Row swaps alter a determinant's sign but do not change the solution set of the associated linear system.

Related topics

• Echelon Form
• Gaussian Elimination
• Inverse Matrix