Row Echelon Form and Reduction
On this page
Direct answer
Row reduction turns any matrix into a staircase shape using three reversible moves — swap two rows, scale a row by a non-zero constant, add a multiple of one row to another. A matrix is in echelon form when every leading entry sits strictly to the right of the one above and all entries below each leading entry are zero; counting the non-zero rows then gives the rank. For a system Ax = b, reducing the augmented matrix decides everything: rank(A) = rank([A|b]) = n gives a unique solution, equal ranks below n give infinitely many solutions, and a mismatch gives none.
What you must remember
- The three row operations: Ri ↔ Rj; Ri → kRi (k ≠ 0); Ri → Ri + kRj. They preserve the solution set of any linear system and never change the rank — column operations do not enjoy this protection, so systems are row-only.
- Echelon conditions: zeros below each leading entry, leading entries marching right, zero rows at the bottom; the reduced form (RREF) adds leading 1s and zeros above and below every pivot.
- Rank: the number of non-zero rows in any echelon form, equivalently the number of pivots — the same however you choose your operations.
- Consistency test: rank(A) = rank([A|b]) = n → unique solution; the ranks equal but below n → infinitely many solutions with (n − r) free parameters; rank(A) < rank([A|b]) → inconsistent, signalled by a row of the form [0 0 0 | c] with c ≠ 0.
- Inverse by reduction: run [A | I] → [I | A⁻¹]; legitimate only for square A with det ≠ 0, i.e. full rank.
- Homogeneous systems: always consistent (the zero solution always exists); non-trivial solutions exist iff rank < n, and m < n equations in n unknowns guarantee infinitely many.
- Determinant bookkeeping: a row swap multiplies det by −1 and scaling a row by k multiplies det by k, so the same triangularisation evaluates determinants.
Reduction walk-through
Take the system x + y + z = 6; x + 2y + 3z = 14; x + 4y + 7z = 30. Clear the first column: R2 → R2 − R1 gives y + 2z = 8, and R3 → R3 − R1 gives 3y + 6z = 24. One more move, R3 → R3 − 3R2, produces 0 = 0 — the third equation has dissolved. So rank(A) = rank(augmented) = 2 < 3: infinitely many solutions with one parameter. Back-substitute: z = t, y = 8 − 2t, x = 6 − y − z = t − 2, giving the solution line (t − 2, 8 − 2t, t); checking t = 2 gives (0, 4, 2), which satisfies all three equations.
Now change one digit: make the last equation x + 4y + 7z = 31. The same steps produce 3y + 6z = 25 in the third row, and after R3 → R3 − 3R2 the row reads 0 = 1 — the inconsistency row [0 0 0 | 1]. One digit moved the system from a line of solutions to no solution, and reduction exposes this instantly, which is exactly why the exam format favours it over elimination by hand.
Where the marks leak
Main asks rank as a numerical answer — count pivots, but choose operations that avoid fractions: swap rows to place a convenient 1 in the pivot position before eliminating. Advanced prefers parameterised consistency: "for which values of λ does the system have no solution?" The trap there is dividing by an expression that can itself be zero; reduce completely first and only then split cases on λ. The other recurring leak is using column operations on an augmented matrix — they permute variables or mix coefficients across equations and silently destroy the system, even though they do preserve rank. Keep columns untouched whenever an augmented matrix is on the table.
Frequently asked questions
Does the rank depend on which row operations I choose?
No — every reduction path ends with the same number of non-zero rows, since rank equals the order of the largest non-zero minor and row operations preserve it.
When does a system have infinitely many solutions?
When rank(A) = rank(augmented) is strictly less than the number of unknowns; each missing pivot contributes one free parameter to the general solution.
Can I use column operations on an augmented matrix?
No — row operations preserve the solution set, while column operations mix coefficients of different variables and change the system being solved.
How do I find A⁻¹ by row reduction?
Adjoin the identity to form [A | I] and row-reduce until the left block becomes I; the right block is then A⁻¹, valid only when A is square and invertible.
What does a row [0 0 0 | 5] in the reduced augmented matrix mean?
It asserts 0 = 5, so the system is inconsistent and has no solution, whatever the other rows say.