Chapter 7 — Key Takeaways
The one idea
A matrix is a linear transformation written down in coordinates. Multiplying a vector by a matrix applies that transformation — rotate, scale, shear, reflect, or project. The numbers inside a matrix are not arbitrary: they record where the transformation sends the standard basis vectors. Once you see a matrix as a verb (a motion of space) rather than a noun (a grid of numbers), the rest of linear algebra reorganizes itself around that picture.
The big ideas, in order
- Linear means "grid stays a grid." A transformation is linear iff it preserves vector addition and scalar multiplication (superposition), iff it keeps grid lines straight, parallel, and evenly spaced with the origin pinned. Consequence: a linear map must fix the origin, so translation is not linear (it's affine).
- A linear map is determined by where the basis vectors go. By superposition, $T(\mathbf{v}) = T(x\mathbf{e}_1 + y\mathbf{e}_2) = xT(\mathbf{e}_1) + yT(\mathbf{e}_2)$. Knowing $T(\mathbf{e}_1)$ and $T(\mathbf{e}_2)$ tells you $T$ everywhere.
- The columns of a matrix are the images of the basis vectors. To build a matrix, ask "where do $\mathbf{e}_1, \mathbf{e}_2$ go?" and stack the answers as columns. To read a matrix, look at its columns.
- Matrix × vector = weighted sum of the columns. $A\mathbf{v} = x(\text{col }1) + y(\text{col }2)$, derived from linearity — not a memorized row-times-column rule (that framing waits for Chapter 8's composition story). A consequence you'll use forever: $A\mathbf{v}$ always lands in the span of the columns.
- Linear ⇔ representable by a matrix. Proven both ways: every matrix gives a linear map (distributivity), and every linear map equals multiplication by the matrix of its basis-vector images (and that matrix is unique in a fixed basis).
- The transpose $A^{\mathsf{T}}$ flips rows and columns ($(A^{\mathsf{T}})_{ij} = a_{ji}$). Generally a different transformation from $A$; equal to $A$ only for symmetric matrices. Its full geometric meaning waits for Part IV.
The transformation zoo (build each by "where do the basis vectors go?")
| Transformation | Matrix | $\det$ | What it does |
|---|---|---|---|
| Identity | $\begin{bmatrix}1&0\\0&1\end{bmatrix}$ | $1$ | nothing |
| Scaling $(s_x, s_y)$ | $\begin{bmatrix}s_x&0\\0&s_y\end{bmatrix}$ | $s_x s_y$ | stretch/squash along axes |
| Rotation by $\theta$ | $\begin{bmatrix}\cos\theta&-\sin\theta\\\sin\theta&\cos\theta\end{bmatrix}$ | $1$ | turn (lengths & angles preserved) |
| Horizontal shear $k$ | $\begin{bmatrix}1&k\\0&1\end{bmatrix}$ | $1$ | slant (area preserved) |
| Reflection (across $x$) | $\begin{bmatrix}1&0\\0&-1\end{bmatrix}$ | $-1$ | flip (orientation reversed) |
| Projection (onto $x$) | $\begin{bmatrix}1&0\\0&0\end{bmatrix}$ | $0$ | flatten (irreversible) |
Reading the determinant (made rigorous in Chapter 11): magnitude = area-scaling factor; sign = orientation ($+$ preserved, $-$ flipped); $0$ = space flattened, so the matrix is singular and cannot be undone.
Skills you gained
- Build the matrix of any 2D transformation you can picture by tracking $\mathbf{e}_1, \mathbf{e}_2$.
- Derive the rotation matrix from the unit circle (never memorize it again).
- Read a mystery matrix into a geometric description from its columns and determinant.
- Compute a matrix-vector product as a weighted sum of columns, by hand and in numpy.
- Combine two transformations by tracking the basis vectors through both steps (a geometric preview of matrix multiplication).
- Compute a transpose; recognize when $A^{\mathsf{T}} \neq A$.
Terms to know
linear transformation (linear map), standard basis vector, matrix of a transformation, columns as images, matrix–vector product, weighted sum of columns, identity matrix, scaling, rotation matrix, shear, reflection, projection, singular matrix, orientation, transpose, symmetric matrix.
How this connects to the recurring themes
- Theme 1 (transformations are the point). This chapter is the thesis: a matrix is how we represent a linear transformation. Change coordinates (Chapter 16) and the matrix changes while the transformation stays the same.
- Theme 2 (geometry = algebra). The deforming unit square and the matrix entries are two views of one object; "build the matrix" and "draw the picture" are the same act.
- Theme 3 (computation validates theory). Every numeric result matched numpy, and your from-scratch
toolkit/matrices.py(apply,transpose) now backs the ideas with code.
Toolkit contribution
toolkit/matrices.py — apply(A, v) (matrix–vector as a weighted sum of columns) and transpose(A), both pure Python, verified against numpy's @ and .T. Chapter 8 adds matmul (matrix–matrix as composition) to the same module.
Forward references
- Chapter 8 — Matrix operations; multiplication as composition (where the row-times-column rule finally earns its place) and why $AB \neq BA$.
- Chapter 9 — The inverse: undoing a transformation; invertible iff no information is lost (det $\neq 0$).
- Chapter 11 — The determinant as signed area/volume scaling; the visualizer's title number made rigorous.
- Chapter 13 — The column space $C(A)$: the set of all reachable outputs $A\mathbf{v}$ — already glimpsed as "the span of the columns."
- Chapter 16 — Change of basis: same transformation, different matrix.
- Chapter 21 — Orthogonal matrices and rotations, generalizing §7.5.3.
- Chapter 23 — Eigenvectors: the directions a transformation does not knock off their own line — which you can already spot in a diagonal matrix's axes.