Home›Courses›Linear Algebra›Week 6 · Lecture 21
← Lecture 20
On this pageMatrix from Standard Basis·Non-Standard Bases·Worked: T(5,2)·Kernel & Range — Definitions·The Dimension Theorem·Worked: T : R³→R⁴·Why Study Ker & Range?·Shortcut via Rank·Worked: T on M₂₂·Beyond ℝⁿ
Course Home →
MATH-120 · Linear Algebra · Lecture 2115 July 2026

The Matrix of a Transformation, Kernel, Range & the Dimension Theorem

How to read off a transformation’s matrix from where it sends a basis, and the two subspaces — kernel and range — that measure everything a transformation destroys and everything it can reach

Week 6 · Wednesday, 15 July 2026
Part A
The Matrix of a Transformation

§1Finding A From Where the Standard Basis Lands

Lecture 20 told us every linear transformation $T:\mathbb{R}^2\to\mathbb{R}^2$ has some matrix $A$ with $T(v)=Av$. Here is the practical question: given a transformation, how do you actually find $A$?

The key method

The columns of $A$ are exactly the images of the standard basis vectors: $$A = \begin{pmatrix} | & | & & | \\ T(\mathbf{e}_1) & T(\mathbf{e}_2) & \cdots & T(\mathbf{e}_n) \\ | & | & & | \end{pmatrix}.$$ Nothing more is needed — knowing where $T$ sends $\mathbf{e}_1,\ldots,\mathbf{e}_n$ pins down $A$ completely, because those columns are the matrix.

Example 1Building A from T(e1) and T(e2)

Suppose $T(1,0)=(2,1)$ and $T(0,1)=(0,1)$. Reading these off as columns:

$$A = \begin{pmatrix}2&0\\1&1\end{pmatrix}.$$

Check: $A\binom{1}{0}=\binom{2}{1}$ ✓ and $A\binom{0}{1}=\binom{0}{1}$ ✓ — both match exactly.

§2Part B — Generalizing to Non-Standard Bases

The standard basis is convenient, but nothing about the underlying idea requires it. The real principle is broader:

The general principle

If we know the image of every vector in a basis — any basis, not just the standard one — then we can find the image of any vector at all. Write the target vector as a combination of the basis, then apply linearity term by term.

§3Fully Worked Example — Using a Non-Standard Basis

★ 2T:R²→R² with T(1,1)=(2,1), T(0,1)=(2,3). Find T(5,2).
Parts C–H
Kernel, Range & the Dimension Theorem

§4Kernel and Range — Definitions

Every linear transformation $T:V\to W$ carries two natural subspaces along with it — one living inside $V$, one living inside $W$.

Kernel

$$\operatorname{Ker}(T) := \{\,v\in V : T(v)=\mathbf{0}\,\} \subseteq V$$ — every vector $T$ crushes down to zero. Ker(T) is a subspace of $V$.

Range (Image)

$$\operatorname{Range}(T) := \{\,w\in W : \exists\, v \text{ with } T(v)=w\,\} \subseteq W$$ — every output $T$ can actually produce. Range(T) is a subspace of $W$.

Exercise C.1Convince yourself — Ker(T) and Range(T) really are subspaces

For both sets, check the three subspace criteria: contains the zero vector, closed under addition, closed under scalar multiplication.

§5The Dimension (Rank–Nullity) Theorem

The Dimension Theorem

$$\dim(\operatorname{Ker}(T)) + \dim(\operatorname{Range}(T)) = \dim(V).$$

Equivalent, in matrix language

$$\operatorname{Nullity}(A) + \operatorname{Rank}(A) = \#\,\text{columns of } A.$$

🧠
Why this deserves to be believed
Every dimension of $V$ has to go somewhere once you apply $T$: it either gets crushed to zero (contributing to $\dim\operatorname{Ker}(T)$) or it survives as a genuinely new independent direction in the output (contributing to $\dim\operatorname{Range}(T)$). There's no third option and no double-counting, so the two pieces add up to exactly the dimension you started with. This is the single most useful bookkeeping tool for the rest of this lecture — every example below ends with this equation checking out.

§6Fully Worked Example — T : ℝ³ → ℝ⁴

★ 3T(x,y,z) = (x-y+2z, x+y-z, 2x+z, 2y-3z)

§7Why Do We Study Ker(T) and Range(T)?

Q: Why do we study Ker(T) and Range(T)? A:

$$\dim(\operatorname{Ker}(T)) = 0 \iff T \text{ is one-to-one (injective)}.$$

$$\dim(\operatorname{Range}(T)) = \dim(W) \iff T \text{ is onto (surjective)}.$$

🔑
What these mean in practice
A trivial kernel ($\operatorname{Ker}(T)=\{\mathbf{0}\}$) means $T$ never collapses two different inputs onto the same output — nothing gets lost, so $T$ is injective. A full-dimensional range ($\operatorname{Range}(T)=W$) means $T$ can reach every point of the codomain — nothing is unreachable, so $T$ is surjective. Together, if both hold (and $\dim V=\dim W$), $T$ is a bijection — an isomorphism — meaning $V$ and $W$ are, for all structural purposes, the same space wearing different labels. Part I closes the lecture with exactly this kind of example.

§8The General Shortcut

Boxed procedure

Given the matrix $A$ of $T:V\to W$, to find $\dim(\operatorname{Ker}\,T)$ and $\dim(\operatorname{Range}\,T)$: just find $\operatorname{rank}(A)$. If $\operatorname{rank}(A)=r$, then by the dimension theorem:

$$\dim(\operatorname{Ker}(T)) = \dim(V) - r.$$

One row reduction answers both questions at once — $\operatorname{rank}(A)$ is $\dim(\operatorname{Range}(T))$ directly, and subtracting from $\dim(V)$ hands you $\dim(\operatorname{Ker}(T))$ for free.

§9Fully Worked Example — T : M₂₂ → M₂₂

★ 4T([a,b;c,d]) = [a+b, b+c; c+d, d+a]
Part I
Beyond the Textbook — Transformations Between Non-ℝⁿ Spaces

§10Two Examples Not in the Book

Everything above used $\mathbb{R}^n$ or a matrix space as domain and codomain. The Ker/Range machinery does not care — it works exactly the same way when the vectors are polynomials.

★ 5(a) T : P₃ → R⁴, the coordinate map

Define $T:P_3\to\mathbb{R}^4$ by $$T(a+bx+cx^2+dx^3) = (a,b,c,d).$$ This is exactly the coordinate-vector isomorphism — it says $P_3$ and $\mathbb{R}^4$ are structurally identical, differing only in notation.

★ 6(b) T : M₂₂ → P₃, a deliberate mirror of the M₂₂ example

Define $T:M_{22}\to P_3$ by

$$T\begin{pmatrix}a&b\\c&d\end{pmatrix} = (a+b) + (b+c)x + (c+d)x^2 + (d+a)x^3.$$

This uses exactly the same coefficient pattern as the $M_{22}\to M_{22}$ example in §9 — only now the four output numbers are read off as polynomial coefficients instead of matrix entries. So, relative to the standard bases $\{E_{11},E_{12},E_{21},E_{22}\}$ of $M_{22}$ and $\{1,x,x^2,x^3\}$ of $P_3$, the matrix of $T$ is identical to the one in §9:

$$A = \begin{pmatrix}1&1&0&0\\0&1&1&0\\0&0&1&1\\1&0&0&1\end{pmatrix}, \qquad \operatorname{rank}(A)=3 \;\text{ (identical row reduction as before).}$$

🔭
Where this leads next
We now have a complete toolkit: build the matrix of any transformation from basis images, and read off injectivity, surjectivity, and dimension all from a single rank computation. The natural next question — one row reduction was doing a lot of work in every example above — is how the matrix of a transformation itself changes when you switch to a different basis, which is where change-of-basis and similarity pick the story back up.
Looking back

Where a transformation sends a basis tells you everything; one rank computation tells you the rest.

The matrix of a transformation is built from basis images, its kernel is a null space, its range is a column space, and the dimension theorem ties the two together with no room for anything to go missing or get double-counted. Every worked example today — $\mathbb{R}^2$, $\mathbb{R}^3\to\mathbb{R}^4$, $M_{22}$, $P_3$ — was the exact same three-step recipe, just wearing different notation.

📌 Course Administration — not examinable, informational only

For the upcoming Quiz and Exam: sections §7.1, §7.2.

There is a double quiz on Monday, covering §8.1, §8.2, §7.1, and §7.2.

A policy referred to in passing as the "n−2 policy" was also mentioned — see the official course policy document for the exact details rather than relying on this summary.

Lecture 21 — complete
MATH-120 · Shoaib Khan · LUMS · July 2026
← Lecture 20Course Home →