Rank of a Matrix and Its Applications
On this page
Direct answer
Rank is the number of linearly independent rows (equivalently columns) of a matrix — operationally, the count of nonzero rows left after reduction to echelon form, or the order of the largest nonsingular minor. For an m×n matrix, rank ≤ min(m, n), and a square matrix is invertible exactly when its rank is full. The payoff is consistency theory: AX = B has a solution iff rank(A) equals rank([A | B]); that solution is unique when the common rank equals the number of unknowns, and infinitely many solutions with (n − r) free parameters when it is smaller.
What you must remember
- Computation: row-reduce to echelon form and count pivots; elementary row operations never change the rank.
- Bounds: rank(A) ≤ min(m, n); for square A of order n, rank n ⟺ |A| ≠ 0; rank 0 means A is the zero matrix.
- Minor view: rank is the size of the largest square submatrix with nonzero determinant — for [[1, 2, 3], [2, 4, 6], [3, 6, 9]] all 2×2 minors vanish, so the rank is 1 (every row is a multiple of the first).
- Consistency: rank(A) = rank([A | B]) ⇒ consistent; common rank r = n gives a unique solution, r < n gives infinitely many parametrised by n − r free variables.
- Inconsistent signature: rank(A) < rank([A | B]) — elimination produces a row 0 = k with k ≠ 0.
- Inequalities: rank(AB) ≤ min(rank A, rank B) and rank(A + B) ≤ rank A + rank B; multiplying can only destroy rank, never create it.
- Homogeneous systems: AX = 0 always has the trivial solution; nontrivial solutions exist iff rank(A) < n — for square A, iff |A| = 0.
Three equations, three verdicts
Consider x + y + z = 6, x + 2y + 3z = 14, x + 4y + 9z = 36. Row-reduce the augmented matrix: R₂ − R₁ gives (0, 1, 2 | 8) and R₃ − R₁ gives (0, 3, 8 | 30); then R₃ − 3R₂ gives (0, 0, 2 | 6). Three pivots, ranks equal at 3 — unique solution, and back-substitution produces z = 3, y = 2, x = 1 (verify: 1 + 8 + 27 = 36). Now replace the third equation with 2x + 3y + 4z = 20, whose coefficient row is the sum of the first two. Elimination yields a zero pivot row in A and, because 20 = 6 + 14, a matching zero in the augmented column: rank(A) = rank([A | B]) = 2 < 3, so infinitely many solutions — one free parameter. Change the constant to 21 and the same elimination produces 0 = 1: ranks differ, no solution. One template, three verdicts — this is exactly how Main frames its "number of solutions" numericals, and the determinant alone (zero in both singular cases) cannot distinguish the last two; only the augmented rank can.
Where students slip
Reading inconsistency off a zero determinant is the fundamental error: |A| = 0 guarantees only that the solution is not unique — consistency is an augmented-matrix question. The second slip is confusing the rank condition for uniqueness: the common rank must equal the number of unknowns n, not merely the rank of A; a 2×3 system with rank 2 is "full rank" yet still has infinitely many solutions because 2 < 3. Third, the product inequality gets trusted in the reverse direction — rank(AB) ≤ min(rank A, rank B) bounds from above, and students who expect rank(AB) ≥ rank A for nonsingular A forget that multiplying by a singular matrix collapses everything to that matrix's rank. Verify solutions by substitution after back-substitution; arithmetic slips in elimination are the largest single source of lost marks in this topic.
Frequently asked questions
How do I find the rank of a matrix quickly?
Row-reduce to echelon form and count nonzero rows — elementary operations preserve rank, and each pivot is one unit of it.
When does AX = B have infinitely many solutions?
When rank(A) = rank([A | B]) = r < n; the solution set then carries n − r free parameters.
Is rank(AB) ever larger than rank(A)?
No — rank(AB) ≤ min(rank A, rank B); a product cannot have rank exceeding either factor.
What does rank say about homogeneous systems?
AX = 0 has only the trivial solution when rank(A) = n; nontrivial solutions appear exactly when rank drops below the number of unknowns.
Do column operations preserve rank?
Yes, column operations preserve rank — but for solving AX = B only row operations are legal, since they alone preserve the solution set.