Reducing a System to Its Skeleton
Gaussian elimination is the algorithm behind solving linear systems, and it also reveals a matrix's deepest numerical feature. The goal is to reach row echelon form: a staircase pattern where each row's leading nonzero entry (its pivot) sits strictly to the right of the pivot above, and all-zero rows sink to the bottom. Reduced row echelon form goes further — each pivot is 1 and is the only nonzero entry in its column, giving a unique canonical form.
The process uses only the three row operations (swap, scale, add a multiple), applied to clear out entries below each pivot in turn. Because these operations preserve the solution set, the echelon form has the same solutions as the original system but is trivial to solve by back-substitution.
The number of pivots is the matrix's rank — the count of independent rows (equivalently, independent columns). Rank measures how much genuine information the matrix carries:
- Rank tells you the number of independent equations in a system.
- For an matrix, full rank () means the matrix is invertible and the system has a unique solution.
- Rank deficiency (rank ) means dependent rows, hence free variables — no unique solution.
Rank also governs solution counts precisely (the Rank–Nullity picture): variables without pivots are free, and the number of free variables is (number of unknowns) minus rank. Zero free variables gives a unique solution; more give infinitely many (or, if an inconsistent row appears, none).
Common pitfall: confusing the presence of a zero row with the system having no solution. A row of all zeros in the coefficient part just signals a dependent equation (a free variable, hence infinitely many solutions). "No solution" comes instead from an inconsistent row — zeros in the coefficients but a nonzero constant, i.e. . A zero row means redundancy; a row means contradiction.