Column space and null space
Every matrix carries two fundamental subspaces: the column space of outputs it can produce and the null space of inputs it sends to zero. Bases for these spaces describe the matrix completely.
The column space
The column space Col A of an m x n matrix is the span of its columns, a subspace of \(\mathbb{R}^m\). It is exactly the set of vectors \(\mathbf{b}\) for which \(A\mathbf{x} = \mathbf{b}\) is solvable. If Col A is all of \(\mathbb{R}^m\), every system with that coefficient matrix has a solution.
The null space
The null space Nul A is the set of all solutions of \(A\mathbf{x} = \mathbf{0}\), a subspace of \(\mathbb{R}^n\). It always contains \(\mathbf{0}\) and is closed under addition and scaling, since \(A(\mathbf{u} + \mathbf{v}) = A\mathbf{u} + A\mathbf{v} = \mathbf{0}\). Nul A is \(\{\mathbf{0}\}\) exactly when the columns of A are independent.
A basis for the null space
Row reduce A to reduced echelon form, solve for basic variables in terms of free variables, and write the general solution as a combination of vectors, one per free variable. Those vectors form a basis of Nul A, and their number, the number of free variables, is the dimension of the null space, also called the nullity.
A basis for the column space
The pivot columns of A, the original columns of A in the positions where the echelon form has pivots, form a basis for Col A. Use the columns of A itself, not of its echelon form: row operations change the column space, though they preserve linear relationships among the columns.
Why pivot columns work
Row operations do not change which combinations of columns equal zero, since they preserve solutions of Ax = 0. So the dependencies among A's columns match those in the echelon form, where non-pivot columns are clearly combinations of pivot columns. Hence the pivot columns of A are independent and span Col A.
The row space
The row space Row A is the span of the rows of A, a subspace of R^n. Row operations do not change it, so the non-zero rows of any echelon form of A are a basis for Row A. The row space is the column space of A^T, and in R^n it is orthogonal to the null space, a connection explored in the inner-product unit.
Comparing the spaces
For an m x n matrix, Col A lives in R^m, while Nul A and Row A live in R^n. Col A is described explicitly, as combinations of known vectors, but testing membership requires solving a system. Nul A is described implicitly, by equations, so checking membership is easy (just multiply by A) but finding vectors in it requires solving.
Worked example
Let A have rows (1, 2, 0, 1), (2, 4, 1, 4), (3, 6, 1, 5). Its reduced echelon form has rows (1, 2, 0, 1), (0, 0, 1, 2), (0, 0, 0, 0), with pivots in columns 1 and 3. A basis for Col A is columns 1 and 3 of A: (1, 2, 3) and (0, 1, 1). The free variables are \(x_2\) and \(x_4\): \(x_1 = -2x_2 - x_4\), \(x_3 = -2x_4\), giving the null space basis \((-2, 1, 0, 0)\) and \((-1, 0, -2, 1)\).
Key takeaways
Col A, the span of A's columns in R^m, is the set of b for which Ax = b is solvable. Nul A, the solution set of Ax = 0 in R^n, is a subspace, trivial exactly when columns are independent. A null space basis comes from the parametric solution, one vector per free variable; its dimension is the nullity. The pivot columns of the original A form a basis of Col A, because row operations preserve column dependencies but change the column space. Non-zero rows of an echelon form give a basis of Row A, which is orthogonal to Nul A. Col A is explicit; Nul A is implicit.
Flashcard
Span of A's columns
Check your understanding
Why must a basis for Col A use columns of A itself rather than columns of its echelon form?
