Skip to content
LearnMathora

Linear algebra · 03 · Verbs for space · 9 min

Matrices as transformations

A matrix looks like a grid of numbers. It is — but the grid is a disguise. A matrix is an action: a machine that moves every point in space at once. Rotate, stretch, flip, shear — each is a matrix; applying it is multiplication.

Build the intuition

Read the columns: where do the axes land?

A 2×2 matrix's first column says where the x-arrow (1,0) lands; the second, where the y-arrow (0,1) lands. That's the entire decoding manual. The matrix [[0,−1],[1,0]] sends (1,0)→(0,1) and (0,1)→(−1,0): a quarter turn. Read columns, see the verb.

(abcd)(xy)=(ax+bycx+dy)\begin{pmatrix} a & b \\ c & d \end{pmatrix}\begin{pmatrix} x \\ y \end{pmatrix} = \begin{pmatrix} ax + by \\ cx + dy \end{pmatrix}

Composition: verbs chain

Rotate, then stretch? Multiply the two matrices once, and the product is a single matrix that does both. A game engine composes camera, object, and perspective transforms into one matrix per frame — then applies it to millions of points cheaply. Chaining is the superpower.

The determinant: the area receipt

Each transformation scales areas by a fixed factor — the determinant. Det 2: areas double. Det 1: areas preserved (pure rotation). Det 0: space is flattened onto a line, information destroyed, no undo. The determinant is the one-number health report of a matrix.

See it move

InteractiveA matrix is a verb
Rotate 45°: The whole plane turns together. Every game camera move is a cousin of this matrix. Matrix: [0.71, -0.71; 0.71, 0.71]

Six verbs, one grid. Watch what each matrix does to space itself — and read its columns to see why.

A worked example

Decode a matrix by its columns

  1. What does M = [[2, 0], [0, 1]] do?

  2. Column 1: (1,0) lands on (2,0) — the x-direction doubles. Column 2: (0,1) stays put.

  3. Verdict: a horizontal stretch ×2. Check the determinant: 2·1 − 0·0 = 2 — areas double, consistent.

  4. No formula sheet needed: columns tell the story every time.

Out in the world

Neural networks are matrix stacks

Each layer of a neural network multiplies its input vector by a learned matrix (then bends the result slightly). “Training” means nudging those matrix entries until the chain of transformations turns pixels into “cat.” GPT-class models are mostly very large matrices, multiplied very fast.

Common confusion, cleared

Matrix multiplication should work like number multiplication.

It composes actions, so order matters: rotate-then-stretch ≠ stretch-then-rotate. AB ≠ BA isn't a defect — it's faithfulness to how actions combine.

A matrix is just a table for storing numbers.

Storage is incidental. The mathematical content is the transformation it performs — the same matrix is a verb, not a spreadsheet.

Recap

  • A matrix moves all of space; its columns show where the axes land.
  • Multiplying matrices chains transformations into one.
  • The determinant reports the area-scaling factor; zero means collapse.