4.1 What is linear algebra?

Linear algebra is the mathematics of two operations on a collection of objects called vectors: you can add two vectors to get a vector, and you can multiply a vector by a number to get a vector. That is the whole definition. Anything else linear algebra studies — matrices, eigenvalues, inner products, the spectral theorem — is built on those two operations.

The reason this minimal-looking structure is so powerful is that an enormous range of physical and mathematical objects satisfy these two rules:

Position vectors in space. Add two displacements; scale by a constant. The result is another displacement.
Forces on a body. Add two forces (parallelogram law); scale by a number.
Solutions of a linear differential equation. Add two solutions of $\ddot x + \omega_0^2 x = 0$ , get another solution. Scale by a constant, get another. This is the superposition principle that makes Fourier analysis work.
Probability distributions over a finite set, signal samples, the components of a quantum state, the modes of a cavity, RGB colours, deep-learning gradients.

Anything you can add and scale is, at root, a vector. Linear algebra is the unified language for all of it.

This first lesson skips the algebra and shows the geometric picture that everything else in the chapter will live inside.

Vectors as arrows; matrices as transformations

A two-dimensional vector $\mathbf{v} = (x, y)$ is a column of two numbers, or equivalently an arrow from the origin to the point $(x, y)$ in the plane. Adding two vectors corresponds to stacking arrows tip-to-tail; scaling stretches an arrow without rotating it. That is the entire content of “vectors form a vector space.”

A matrix is the next concept. A $2 \times 2$ matrix

A \;=\; \begin{pmatrix} a_{11} & a_{12} \\ a_{21} & a_{22} \end{pmatrix}

is a rule for sending one vector to another. The recipe is matrix–vector multiplication: given an input $\mathbf{v} = (x, y)$ , the output is

A \mathbf{v} \;=\; \begin{pmatrix} a_{11} x + a_{12} y \\ a_{21} x + a_{22} y \end{pmatrix}.

We will work through this formula carefully in 4.2. For now, take the geometric view: a matrix is a transformation of the plane that respects addition and scaling. If you input $\mathbf{u} + \mathbf{v}$ you get $A\mathbf{u} + A\mathbf{v}$ ; if you input $c \mathbf{v}$ you get $c \cdot A\mathbf{v}$ . That linearity property is so restrictive that the matrix’s behaviour is completely determined by what it does to just two vectors — the standard basis $\mathbf{e}_1 = (1, 0)$ and $\mathbf{e}_2 = (0, 1)$ — because every other vector is $x \mathbf{e}_1 + y \mathbf{e}_2$ .

The picture

preset:

a₁₁ = 1.00 a₁₂ = 0.00

a₂₁ = 0.00 a₂₂ = 1.00

The pale grid is the original plane; the blue grid is its image under A. The two coloured arrows are the images of the standard basis vectors: A e₁ is the first column of A, A e₂ is the second. Every vector v = (x, y) ends up at x · A e₁ + y · A e₂ — multiplying by A means "express v in the e₁, e₂ basis, then walk the columns." The determinant counts oriented area: positive det preserves orientation, negative det flips it, zero det collapses the plane onto a line or a point.

The interactive shows a unit grid (faint, in the original plane) being mapped to a parallelogram grid (blue) by a $2 \times 2$ matrix $A$ . Pick a preset or slide the four matrix entries directly. Things to absorb:

The two coloured arrows are the columns of $A$ . Specifically, $A \mathbf{e}_1$ is the first column $(a_{11}, a_{21})$ and $A \mathbf{e}_2$ is the second column $(a_{12}, a_{22})$ . Every other vector’s image is determined by these two: $A \mathbf{v} = x \cdot A\mathbf{e}_1 + y \cdot A\mathbf{e}_2$ . The columns of a matrix are where it sends the basis vectors.
The grid stays parallel. A linear transformation always maps the unit grid to a parallelogram grid; parallel lines stay parallel, ratios of parallel lengths are preserved. What changes is the angle between the grid lines (set by the relationship of the two columns) and the area of each cell (set by the determinant).
The determinant is the area factor. Each cell of the original unit grid has area 1; each cell of the transformed grid has area $|\det A|$ . The sign of $\det A$ tells you whether the transformation preserves orientation (positive) or flips it (negative, like a mirror reflection). When $\det A = 0$ the plane collapses onto a single line or a point — the matrix has destroyed a dimension’s worth of information.
Named transformations have geometric meaning. Rotation rotates the grid rigidly. Scaling stretches axes by different amounts. Shear tilts vertical lines while keeping horizontals fixed. Reflection mirrors across an axis. Projection flattens onto a line. Every $2 \times 2$ matrix is some combination of these basic moves, in the precise sense that any matrix can be factored as rotation × scaling × rotation — the singular-value decomposition, which we will not develop here but which is worth knowing exists.

Why this matters beyond pictures

The geometric picture in two dimensions generalises directly to higher dimensions, even when “picture” stops being literally possible. A $3 \times 3$ matrix is a linear transformation of 3-D space; an $n \times n$ matrix is a linear transformation of $n$ -dimensional space; an infinite-dimensional linear operator (the time-evolution operator of a quantum state, the Laplacian acting on functions in a cavity) is a linear transformation of an infinite-dimensional function space.

The structure carries over: the operator has columns (in some basis), an “area” generalisation (a determinant or a related volume form), eigenvalues (directions only stretched, not rotated), and a notion of singularity (loss of information). Every one of those concepts appears repeatedly across Sound, Hearing, and Cavitation, because every linear PDE on the bookshelf is a linear operator whose eigenvalue structure determines its physics.

The next lesson defines the operations carefully — what matrix–vector multiplication does, how matrices compose, what the inverse means.