Linear Algebra: Kernel and Range

At the heart of understanding linear transformations—whether modeling forces in a structure, processing signals, or compressing data—lie two fundamental subspaces: the kernel and the range. These are not abstract curiosities but essential tools for diagnosing the solutions to systems of equations, the invertibility of transformations, and the flow of information through a linear model. Mastering their interplay is key to analyzing stability, controllability, and dimensionality in virtually every engineering field.

Defining the Kernel and Range

Let $T:V\to W$ be a linear transformation between vector spaces. The kernel (or null space) of $T$ is the set of all vectors in the domain $V$ that map to the zero vector in $W$. Formally, $\text{ker}(T)=\{\mathbf{v}\in V\mid T(\mathbf{v})=0_W\}$. Crucially, the kernel is always a subspace of the domain $V$. Intuitively, it captures all the "blind spots" or inputs that the transformation completely annihilates.

The range (or image) of $T$ is the set of all possible outputs in $W$. Formally, $\text{range}(T)=\{T(\mathbf{v})\mid \mathbf{v}\in V\}$. The range is always a subspace of the codomain $W$. It represents the "footprint" or the actual reach of the transformation within the target space. For a transformation represented by a matrix $A$, the range is precisely the column space of $A$, as the output $A\mathbf{x}$ is a linear combination of $A$'s columns.

Computing the Kernel and Range for a Matrix

For a linear transformation $T(\mathbf{x})=A\mathbf{x}$ where $A$ is an $m\times n$ matrix, computation becomes a matter of solving linear systems.

Kernel (Null Space): To find $\text{ker}(A)=\{\mathbf{x}\mid A\mathbf{x}=0\}$, you solve the homogeneous system $A\mathbf{x}=0$.

1. Row reduce $A$ to its reduced row echelon form (RREF).
2. Express the leading variables in terms of the free variables.
3. The solution set is spanned by the vectors you get by setting each free variable to 1 (and the others to 0). These spanning vectors form a basis for the kernel.

Range (Column Space): To find a basis for $\text{range}(A)$ (the column space), identify the pivot columns in the original matrix $A$.

1. Row reduce $A$ to find the pivot positions.
2. The columns in the original matrix $A$ that correspond to these pivot positions are linearly independent and span the column space. They form a basis for the range.

Example: Let $A=\left[\begin{array}{ccc}1& 2& -1\\ 2& 4& -2\end{array}\right]$.

• Kernel: Solving $A\mathbf{x}=0$ gives RREF of $\left[\begin{array}{ccc}1& 2& -1\\ 0& 0& 0\end{array}\right]$. With $x_2=s$ and $x_3=t$ free, $x_1=-2s+t$. The kernel is spanned by $\left[\begin{array}{c}-2\\ 1\\ 0\end{array}\right]$ and $\left[\begin{array}{c}1\\ 0\\ 1\end{array}\right]$.
• Range: The only pivot is in column 1. Thus, the range is the line spanned by the first column: $\text{span}\left(\left[\begin{array}{c}1\\ 2\end{array}\right]\right)$.

The Rank-Nullity Theorem

This fundamental theorem connects the dimensions of the kernel and range. For a linear transformation $T:V\to W$, where $V$ is finite-dimensional:

$$\dim (\text{ker}(T))+\dim (\text{range}(T))=\dim (V)$$

The dimension of the kernel is called the nullity of $T$. The dimension of the range is called the rank of $T$. For an $m\times n$ matrix $A$, the theorem states: $\text{nullity}(A)+\text{rank}(A)=n$ (the number of columns).

Proof Sketch: The elegance of the proof lies in building a basis for $V$. Start with a basis $\{{\mathbf{u}}_1,...,{\mathbf{u}}_k\}$ for the kernel. Extend this to a basis $\{{\mathbf{u}}_1,...,{\mathbf{u}}_k,{\mathbf{v}}_1,...,{\mathbf{v}}_r\}$ for all of $V$. You can then show that $\{T({\mathbf{v}}_1),...,T({\mathbf{v}}_r)\}$ forms a basis for the range. Since $k=\text{nullity}(T)$ and $r=\text{rank}(T)$, and $k+r=\dim (V)$, the theorem holds. This proof is constructive, showing how vectors that are independent "modulo the kernel" map to an independent set in the range.

Injective, Surjective, and Bijective Transformations

The kernel and range directly characterize fundamental properties of a linear map.

A transformation $T$ is injective (or one-to-one) if distinct inputs map to distinct outputs: $T(\mathbf{u})=T(\mathbf{v})$ implies $\mathbf{u}=\mathbf{v}$. A linear transformation is injective if and only if its kernel is trivial (contains only the zero vector). If $\text{ker}(T)=\{0\}$, then the nullity is 0, and by the Rank-Nullity Theorem, the rank equals $\dim (V)$. This means all information from the domain is preserved in the mapping.

A transformation $T$ is surjective (or onto) if its range is all of the codomain $W$: $\text{range}(T)=W$. Surjectivity is determined entirely by the range. For a matrix transformation $A\mathbf{x}$, surjectivity means every vector in ${\mathbb{R}}^m$ can be formed from the columns of $A$, which requires $\text{rank}(A)=m$.

A transformation that is both injective and surjective is bijective (an isomorphism). For a matrix $A$, this implies it must be square ($n=m$) and have full rank ($\text{rank}(A)=n$), which is equivalent to $A$ being invertible.

Connecting to System Solution Properties

These concepts provide a complete picture for solving the linear system $A\mathbf{x}=\mathbf{b}$.

1. Existence (Is $\mathbf{b}$ in the range?): A solution exists if and only if $\mathbf{b}$ is in the column space (range) of $A$. For a surjective transformation (rank = $m$), a solution exists for every $\mathbf{b}$.

1. Uniqueness (Is the kernel trivial?): If a solution exists, it is unique if and only if the kernel of $A$ is trivial (nullity = 0). If the kernel is non-trivial, then any particular solution ${\mathbf{x}}_p$ can be added to any vector ${\mathbf{x}}_n$ in the kernel to get another solution: $A({\mathbf{x}}_p+{\mathbf{x}}_n)=A{\mathbf{x}}_p+A{\mathbf{x}}_n=\mathbf{b}+0=\mathbf{b}$. Thus, the complete solution set is ${\mathbf{x}}_p+\text{ker}(A)$.

This framework tells you that the structure of the solution space to $A\mathbf{x}=\mathbf{b}$ is an affine space (a translated subspace) parallel to the kernel of $A$. The Rank-Nullity Theorem governs the dimensions: for an $m\times n$ system, $\text{rank}(A)$ tells you the "degrees of freedom" constrained by the equations, while $\text{nullity}(A)$ tells you the dimensions of the solution space when it exists.

Common Pitfalls

• Confusing the kernel with the zero vector: The kernel is a set (a subspace), not a single vector. While it always contains the zero vector, its dimension may be greater than zero. Saying "the kernel is zero" is incorrect; say "the kernel is trivial" or "the nullity is zero."
• Using RREF columns for the range basis: When finding a basis for the column space (range), you must return to the pivot columns of the original matrix $A$, not its RREF. The RREF columns do not generally have the same span as the original columns.
• Misapplying Rank-Nullity: The theorem applies to the domain dimension. For an $m\times n$ matrix, the sum of rank and nullity equals $n$ (number of columns/input dimension), not $m$. Confusing this leads to incorrect conclusions about surjectivity.
• Overlooking injectivity for non-square matrices: A transformation can be injective even if $A$ is not square (e.g., a tall, full-column-rank matrix). Injectivity only requires a trivial kernel, which is possible whenever $\text{rank}(A)=n$, regardless of $m$.

Summary

• The kernel of a linear transformation is the subspace of all inputs that map to zero, defining its "blind spots." The range is the subspace of all possible outputs, defining its "reach."
• The Rank-Nullity Theorem, $\text{rank}(T)+\text{nullity}(T)=\dim (\text{domain})$, is a cardinal rule linking the dimensions of these subspaces.
• A transformation is injective (one-to-one) if and only if its kernel is trivial, and surjective (onto) if and only if its range equals the entire codomain.
• For a matrix equation $A\mathbf{x}=\mathbf{b}$, the range determines if a solution exists ($\mathbf{b}$ must be in the column space), while the kernel determines the uniqueness or structure of the solution set.
• Computationally, the kernel is found by solving $A\mathbf{x}=0$, and a basis for the range is given by the pivot columns of the original matrix $A$.