Matrix operations — addition, subtraction (conditions): Visual walkthrough

This page rebuilds matrix addition from nothing. No symbol appears before we draw it. By the end you will see why two matrices can only be added when they have the same shape — it will feel obvious, not like a rule someone handed you.

We build on Matrix Notation and Terminology for the words "row", "column", and "order", and we point to Zero Matrix and Identity Matrix, Scalar Multiplication of Matrices, Transpose of a Matrix and the parent along the way.

Step 1 — What a matrix actually is: a labelled grid

WHAT. A matrix is just a rectangle of numbers arranged in rows (the horizontal lines) and columns (the vertical lines). Nothing more.

WHY. Before we combine two grids we must agree on what one grid means. Every number sits in a box, and the box's position carries meaning — "row 1, column 2" is a different slot from "row 2, column 1", even if the numbers happen to be equal.

PICTURE. Look at the grid below. The box highlighted in magenta is at row 2, column 3. We write its address as $a_{23}$ — read "a-two-three". The little " $2$ " says which row, the " $3$ " says which column. The small letter $a$ just names which matrix this number belongs to.

Step 2 — The "order" is the shape of the grid

WHAT. The order of a matrix is written $m \times n$ , said " $m$ by $n$ ". Here $m$ is the number of rows and $n$ is the number of columns.

WHY. Two grids can hold the same count of numbers yet be shaped completely differently. To compare grids we must talk about their shape, not their total count. The order is that shape.

PICTURE. The left grid is $2 \times 3$ — 2 rows, 3 columns, 6 boxes. The right grid is $3 \times 2$ — 3 rows, 2 columns, also 6 boxes. Same number of boxes, different shape. Trace the arrows: box "row 1, column 3" exists on the left but there is no column 3 on the right.

Step 3 — Adding means pairing boxes at the same address

WHAT. To add two matrices $A$ and $B$ we walk through every address $(i, j)$ and add the two numbers sitting there: $c_{ij} = box (i, j) of A a_{ij} + box (i, j) of B b_{ij}$

Read it term by term: the output box $c_{ij}$ gets the number from $A$ 's box at that address plus the number from $B$ 's box at the exact same address. Same row, same column, both times.

WHY. Each box means something specific ("Monday–dishes minutes"). Adding box $(1, 1)$ of one chart to box $(1, 1)$ of the other keeps the meaning intact: we are combining the same fact from two sources. This is why the operation is called element-wise — box meets its twin.

PICTURE. Two transparent grids stacked on top of each other. Each box glows where its partner sits directly beneath it; we drop the sum into that slot. Every add is strictly vertical — box $(i, j)$ only ever meets box $(i, j)$ .

Step 4 — A worked add, box by box

WHAT. Take

B = \begin{bmatrix} 4 & 1 & -1 \\ -3 & 7 & 6 \end{bmatrix}.$$ Both are $2\times 3$, so every box has a partner. Add each pair: $$A+B = \begin{bmatrix} 2+4 & -3+1 & 5+(-1) \\ 1+(-3) & 0+7 & -2+6 \end{bmatrix} = \begin{bmatrix} 6 & -2 & 4 \\ -2 & 7 & 4 \end{bmatrix}.$$ **WHY.** We first *checked the shape* ($2\times3 = 2\times3$ ✓), because only then does every box have a twin. Then we did six independent additions — one per address — no box left lonely. **PICTURE.** Each output box is coloured, with its two source boxes traced back by arrows. Notice box $(1,2)$: it took $-3$ from $A$ and $+1$ from $B$ to make $-2$. Nothing crosses columns or rows. Reveal-yourself checks: - Box $(2,1)$ of $A+B$ ::: $1+(-3) = -2$ - Box $(1,3)$ of $A+B$ ::: $5+(-1) = 4$ --- ## Step 5 — Subtraction is the same picture, with a flip **WHAT.** Subtraction just adds the **negative** of every box: $$d_{ij} = a_{ij} - b_{ij} = a_{ij} + (-b_{ij}).$$ Here $-b_{ij}$ is the same number as $b_{ij}$ but with its sign flipped. This ties to [[Scalar Multiplication of Matrices]]: $-B$ is $B$ scaled by $-1$. **WHY.** By turning "minus" into "add the negative", subtraction inherits the *exact same* twin-pairing machinery — so it needs the *exact same* matching shapes. We do not invent a new rule. **PICTURE.** Same stacked grids as Step 3, but $B$'s layer is shown "negated" (signs flipped, drawn in violet). We then add as before. Quick worked case with $$P=\begin{bmatrix}5&-1\\3&2\\0&4\end{bmatrix},\quad Q=\begin{bmatrix}2&3\\-1&1\\5&-2\end{bmatrix} \;\Rightarrow\; P-Q=\begin{bmatrix}5-2&-1-3\\3-(-1)&2-1\\0-5&4-(-2)\end{bmatrix} =\begin{bmatrix}3&-4\\4&1\\-5&6\end{bmatrix}.$$ Both are $3\times2$ — twins exist everywhere, so it's defined. --- ## Step 6 — The degenerate case: shapes disagree **WHAT.** Try to add $$M = \begin{bmatrix} 1 & 2 & 3 \\ 4 & 5 & 6 \end{bmatrix}\;(2\times 3),\qquad N = \begin{bmatrix} 7 & 8 \\ 9 & 10 \\ 11 & 12 \end{bmatrix}\;(3\times 2).$$ **WHY.** Walk to address $(1,3)$: $M$ has a box there (the number $3$), but $N$ has **no column 3** — its box is missing. There is nothing to add $3$ to. Walk to address $(3,1)$: $N$ has a box ($11$), but $M$ has **no row 3**. Two orphans, no twins. **PICTURE.** The orphaned boxes are ringed in orange with a "?" — they float with no partner beneath. Because algebra must be deterministic, we may **not** invent a value, so the whole operation is declared ==undefined==. > [!mistake] "Same total count is enough" > $M$ and $N$ each hold 6 numbers, so it *feels* addable. But addition pairs by **address**, not by count. The tempting fix "just transpose $N$" gives a *different* operation $M + N^{\mathsf T}$ (see [[Transpose of a Matrix]]) — it changes what each box means, so it is not the sum of $M$ and $N$. > [!recall]- Every shape case at a glance > Same order → defined ::: yes, add every twin > Different order (even same total count) → defined ::: no, undefined > Adding the zero matrix $O$ of the same order → result ::: unchanged, $A+O=A$ (see [[Zero Matrix and Identity Matrix]]) --- ## The one-picture summary Everything above collapses into one flow: **agree on shape → pair twin boxes → add → drop into the same address.** If any box lacks a twin, stop — undefined. > [!recall]- Feynman retelling > You and a friend each keep a chore chart: days along the top, chores down the side, minutes inside each square. To find your *combined* effort you go square by square: "Monday–dishes: I did 15, you did 20, together 35." You never mix Monday–dishes with Tuesday–laundry — only the *same square* meets the *same square*. This only works if both charts have the identical layout: same days across, same chores down. If your chart has an extra day your friend doesn't have, that column has no partner square, so the whole combining is impossible — undefined. Subtraction is the same walk, you just flip your friend's numbers to negative first ("you did 5 more minutes than me"). That is the entire story of matrix addition and subtraction: same shape, pair the twins, done. --- ## Connections - [[Matrix Operations — Addition, Subtraction (Conditions)]] — the parent rule this page derives. - [[Matrix Notation and Terminology]] — where "row", "column", "order", $a_{ij}$ come from. - [[Zero Matrix and Identity Matrix]] — the identity $O$ for addition. - [[Scalar Multiplication of Matrices]] — $-B$ is $B$ scaled by $-1$; powers subtraction. - [[Transpose of a Matrix]] — why $A+B^{\mathsf T}$ is a *different* operation. - [[Matrix Multiplication]] — contrast: multiplication needs $n$ (of $A$) $=$ rows (of $B$), not equal shapes. - [[System of Linear Equations]] — combining systems as combining matrices.