Matrix Multiplication: Why Order Matters More Than the Numbers

Stretch, Then Shear. Or Shear, Then Stretch.

Take a square on a screen. First stretch it horizontally, then shear it to the right. Now reset and do those same two operations in the opposite order.

Same starting shape. Same two instructions. Different result.

That is matrix multiplication in one sentence. Matrices don't just hold numbers. They encode actions. And when you stack actions, order becomes part of the answer.

Each Entry Is One Dot Product

Forget the scary brackets for a second. Matrix multiplication is just repeated row-by-column dot products.

Suppose

The top-left entry of $AB$ comes from the first row of $A$ and the first column of $B$:

Do that for every row-column pairing and you get

That's the whole engine. No mystery. Just lots of little dot products done in a disciplined way.

The Inner Dimensions Are the Gatekeeper

Students memorize "rows times columns" and still get stuck on when multiplication is allowed. The cleaner rule is this: the inside numbers have to match.

A $2 \times 3$ matrix can multiply a $3 \times 4$ matrix because the 3s match. The result is $2 \times 4$. But a $2 \times 3$ matrix cannot multiply a $2 \times 2$ matrix in that order. The middle numbers disagree, so the row-column pairings don't line up.

I think of it as a lock and key. The first matrix provides rows. The second provides columns. If the row length and column length don't match, there is nothing to multiply.

Why Order Matters More Than The Numbers

This is where matrix multiplication stops feeling like arithmetic and starts feeling like choreography.

Let

Then

Same matrices. Different products. Which means matrix multiplication is not commutative. $AB \ne BA$ in general.

That's not some annoying algebra exception. It's the point. In graphics, "rotate then scale" is different from "scale then rotate." In robotics, one movement changes the coordinate frame for the next. In machine learning, each layer transforms the data before handing it to the next layer. Order is part of the meaning.

The Quiet Engine Behind Search, Graphics, and AI

Matrix multiplication shows up anywhere lots of inputs need to be mixed into lots of outputs.

• Computer graphics: rotating, scaling, and projecting 3D objects onto a 2D screen
• Neural networks: every dense layer is a matrix multiply plus a nonlinearity
• Economics and statistics: input-output models, regression, covariance transforms
• Linear systems: row operations, inverses, and solution structure all live nearby

That's why matrix multiplication keeps pointing toward other ideas. Once you're comfortable with products, the next natural stops are determinants and reduced forms. Our guide on matrix determinants explains how a matrix scales area or volume, and the RREF guide shows how matrices expose whether a system actually has a solution.

Quick Questions