Linear Algebra › Section D · Vector spaces › Topic D.2
Null spaces, column spaces, and kernels
Every matrix carries two subspaces: the vectors it sends to zero and the vectors it can produce. For a general linear transformation they become the kernel and the range, and together they decide whether the map is one-to-one or onto.
Two subspaces from every matrix
"Subspaces, column space, and null space" introduced two subspaces attached to an m x n matrix \(A\). This topic looks at them again inside the general setting of "Vector spaces and subspaces", and asks two practical questions. How do you write down the vectors of each space explicitly? And how do you test whether a given vector belongs? The answers turn out to be opposite for the two spaces. The same ideas then carry over to any linear transformation between vector spaces, such as differentiation of polynomials, where they are called the kernel and the range.
The null space
The null space of \(A\) is the set of all solutions of the homogeneous equation:
It is a subspace of \(\mathbb{R}^n\), where n is the number of columns. The proof is the subspace test. \(A\mathbf{0} = \mathbf{0}\), so the zero vector is in it. If \(A\mathbf{u} = \mathbf{0}\) and \(A\mathbf{v} = \mathbf{0}\), then \(A(\mathbf{u} + \mathbf{v}) = \mathbf{0} + \mathbf{0} = \mathbf{0}\). And \(A(c\mathbf{u}) = c(A\mathbf{u}) = \mathbf{0}\). The null space is defined implicitly, by a condition, so finding its vectors takes work.
Worked example: an explicit description of Nul A
Find a spanning set for \(\operatorname{Nul} A\), where
1. Row reduce \(A\) to reduced echelon form: the rows become (1, 2, 0, 1), (0, 0, 1, -2) and (0, 0, 0, 0). 2. The pivots are in columns 1 and 3, so \(x_2\) and \(x_4\) are free. 3. Solve for the basic variables: \(x_1 = -2x_2 - x_4\) and \(x_3 = 2x_4\). 4. Write the solution in parametric vector form:
So \(\operatorname{Nul} A\) is the span of \((-2, 1, 0, 0)\) and \((-1, 0, 2, 1)\), a subspace of \(\mathbb{R}^4\).
The column space
The column space of \(A\) is the set of all linear combinations of its columns: \(\operatorname{Col} A = \operatorname{Span}\{\mathbf{a}_1, \ldots, \mathbf{a}_n\}\). Since \(A\mathbf{x} = x_1\mathbf{a}_1 + \cdots + x_n\mathbf{a}_n\), the same set can be written
It is a subspace of \(\mathbb{R}^m\), where m is the number of rows, because every span is a subspace. The column space is defined explicitly: you can produce as many of its vectors as you like just by choosing \(\mathbf{x}\). It is all of \(\mathbb{R}^m\) exactly when \(A\mathbf{x} = \mathbf{b}\) has a solution for every \(\mathbf{b}\) in \(\mathbb{R}^m\).
Worked example: testing membership
Use the same 3 x 4 matrix \(A\). 1. Is \(\mathbf{u} = (-3, 1, 2, 1)\) in \(\operatorname{Nul} A\)? Just multiply: row 1 gives \(-3 + 2 - 2 + 3 = 0\), row 2 gives \(-6 + 4 + 0 + 2 = 0\), row 3 gives \(-9 + 6 - 2 + 5 = 0\). So \(A\mathbf{u} = \mathbf{0}\) and yes. 2. Is \(\mathbf{b} = (2, 3, 5)\) in \(\operatorname{Col} A\)? Notice that row 3 of \(A\) is row 1 plus row 2, so every output \(A\mathbf{x}\) has third entry equal to the sum of the first two. Row reducing \([A \mid \mathbf{b}]\) confirms that the system is consistent exactly when \(b_3 = b_1 + b_2\). Here \(5 = 2 + 3\), so yes. 3. For \(\mathbf{b} = (1, 1, 1)\), \(1 \ne 2\), so it is not in \(\operatorname{Col} A\).
Contrasting the two spaces
For an m x n matrix, \(\operatorname{Nul} A\) lives in \(\mathbb{R}^n\) and \(\operatorname{Col} A\) lives in \(\mathbb{R}^m\); in the example they sit in \(\mathbb{R}^4\) and \(\mathbb{R}^3\). Their strengths are opposite. Testing whether \(\mathbf{u}\) is in \(\operatorname{Nul} A\) is quick, one multiplication, but listing its vectors needs row reduction. Listing vectors of \(\operatorname{Col} A\) is free, since the columns themselves are in it, but testing \(\mathbf{b}\) needs a row reduction of \([A \mid \mathbf{b}]\). Two more links: \(\operatorname{Nul} A = \{\mathbf{0}\}\) exactly when \(A\mathbf{x} = \mathbf{0}\) has only the trivial solution, and \(\operatorname{Col} A = \mathbb{R}^m\) exactly when \(A\) has a pivot in every row.
Kernel and range of a linear transformation
A linear transformation \(T\) from a vector space \(V\) to a vector space \(W\) assigns to each \(\mathbf{x}\) in \(V\) a vector \(T(\mathbf{x})\) in \(W\) such that \(T(\mathbf{u} + \mathbf{v}) = T(\mathbf{u}) + T(\mathbf{v})\) and \(T(c\mathbf{u}) = cT(\mathbf{u})\). Its kernel is the set of \(\mathbf{x}\) in \(V\) with \(T(\mathbf{x}) = \mathbf{0}\), a subspace of \(V\). Its range is the set of all outputs \(T(\mathbf{x})\), a subspace of \(W\). For \(T(\mathbf{x}) = A\mathbf{x}\), the kernel is \(\operatorname{Nul} A\) and the range is \(\operatorname{Col} A\). For the map sending a twice-differentiable function \(y\) to \(y'' + y\), the kernel is the solution set of \(y'' + y = 0\), which is why that set is a subspace.
Worked example: differentiation
Let \(D\) send each polynomial in \(\mathbb{P}_3\) to its derivative in \(\mathbb{P}_2\):
1. Linearity: the derivative of a sum is the sum of the derivatives, and constants factor out, so \(D\) is linear. 2. Kernel: \(D(p) = 0\) forces \(b = c = d = 0\), so the kernel is the constant polynomials. 3. Range: any \(q = e + ft + gt^2\) is the derivative of \(et + \frac{f}{2}t^2 + \frac{g}{3}t^3\), so the range is all of \(\mathbb{P}_2\). 4. Conclusion: \(D\) reaches everything in \(\mathbb{P}_2\), but different polynomials, such as \(t^2\) and \(t^2 + 7\), share a derivative.
One-to-one, onto and isomorphisms
\(T\) is onto when its range is all of \(W\). \(T\) is one-to-one when different inputs always give different outputs, and for a linear map this happens exactly when the kernel is \(\{\mathbf{0}\}\). The reason: if \(T(\mathbf{u}) = T(\mathbf{v})\), then \(T(\mathbf{u} - \mathbf{v}) = \mathbf{0}\), so \(\mathbf{u} - \mathbf{v}\) is in the kernel, and a trivial kernel forces \(\mathbf{u} = \mathbf{v}\). For a matrix map, one-to-one means a pivot in every column, and onto means a pivot in every row. A linear map that is both is an isomorphism, and its inverse is linear too. "Rank and the rank-nullity theorem" shows how the sizes of kernel and range always balance.
Common mistakes to avoid
Mixing up where the spaces live is the most common error: for a 3 x 4 matrix, null space vectors have 4 entries and column space vectors have 3. The second is reading the null space from the original matrix instead of from the reduced echelon form. The third is saying a map is one-to-one because the kernel "contains zero"; every kernel contains zero, and one-to-one needs the kernel to be only zero. Finally, a map can be onto without being one-to-one, as differentiation shows, so check each property separately.
Key takeaways
\(\operatorname{Nul} A\), the solutions of \(A\mathbf{x} = \mathbf{0}\), is a subspace of \(\mathbb{R}^n\); describe it explicitly by solving in parametric vector form. \(\operatorname{Col} A\), the span of the columns and the set of all \(A\mathbf{x}\), is a subspace of \(\mathbb{R}^m\); test membership by row reducing \([A \mid \mathbf{b}]\). For a linear map, the kernel and range generalise these two spaces. A linear map is one-to-one exactly when its kernel is \(\{\mathbf{0}\}\), onto when its range is the whole codomain, and an isomorphism when it is both.
Flashcards
Flashcard
All x with Ax = 0
Test your understanding
Question 1 of 5
A is a 5 x 7 matrix. Where do Nul A and Col A live?
Question 2 of 5
A has rows (1, -2, 1) and (2, 0, -6). Which vector is in Nul A?
Question 3 of 5
What is the kernel of the differentiation map from P₃ to P₂?
Question 4 of 5
A linear transformation T has kernel {0}. What can you conclude?
Question 5 of 5
A is the 3 x 4 matrix whose third row is the sum of its first two rows, and its column space is exactly the vectors with b₃ = b₁ + b₂. Which vector is in Col A?
