Determinants — cofactor expansion along any row - column: Visual walkthrough

We will not assume you know what a determinant "is". We start by looking at a grid of numbers and asking one honest question.

Step 1 — What is this number we are hunting?

WHAT. Take a square block of numbers — same count across as down. We call it a matrix. A $2 \times 2$ matrix has 2 rows and 2 columns:

A = (a c b d) .

Here $a, b, c, d$ are just numbers, and the rows run left-to-right, the columns run top-to-bottom.

WHY. A matrix is a machine that stretches and shears flat space: it sends the little unit square (the square with corners at $(0, 0), (1, 0), (0, 1), (1, 1)$ ) to a slanted parallelogram. The determinant is a single number that measures the signed area of that parallelogram — how much the machine grows or shrinks area. If the parallelogram gets squashed to a line, its area is $0$ , and $det A = 0$ .

PICTURE. The blue unit square becomes the orange parallelogram. Its area is the determinant.

Figure — Determinants — cofactor expansion along any row - column

Everything larger is going to be reduced back to this $2 \times 2$ base case. That is the whole plan.

Step 2 — The only true definition: pick one from each row AND each column

WHAT. To go beyond $2 \times 2$ we need the honest, from-scratch definition of the determinant, the Leibniz formula. It says: form every possible way of choosing one entry from each row so that no two chosen entries share a column, multiply the chosen entries, attach a sign, and add up all such products.

WHY. The parallelogram-area idea (Step 1) is the meaning, but to compute it for any size we need a formula in the raw numbers. "One per row, one per column" is exactly the algebra of area-under-a-linear-map — no entry gets used twice, no line gets counted twice.

PICTURE. For a $3 \times 3$ grid there are $3! = 6$ legal choices. Each is a set of three cells, one per row, one per column — like three non-attacking rooks on a chessboard. Three of these patterns carry a $+$ , three carry a $-$ .

Check it against Step 1: for $2 \times 2$ there are exactly two permutations. "Row 1→col 1, row 2→col 2" gives $+ a d$ . "Row 1→col 2, row 2→col 1" is one swap, so $sgn = - 1$ , giving $- b c$ . Total $a d - b c$ . ✓ The base case is Leibniz.

Step 3 — Where does the minus sign really come from?

WHAT. We just claimed the anti-diagonal choice gets a $-$ . Let's see why, so the checkerboard later feels inevitable.

WHY. The sign is the parity (even/odd swap count) of the permutation. One swap = one $- 1$ . Two swaps cancel to $+ 1$ . This is the single mechanism that produces every sign in every determinant — no exceptions.

PICTURE. Start with both rooks on the main diagonal (the identity, sign $+$ ). Slide them to the anti-diagonal: that is one swap of the two columns. One swap → parity flips → sign becomes $-$ .

Step 4 — Sort the sum by "where does row 1 point?"

WHAT. Now we chop the big Leibniz sum for an $n \times n$ matrix into buckets, according to which column the row-1 entry sits in. Call that column $j = σ (1)$ .

det A = j = 1 \sum n a_{1 j} σ \in S_{n} σ (1) = j \sum sgn (σ) k = 2 \prod n a_{k, σ (k)} .

Reading the symbols:

$\sum_{j = 1}^{n}$ :: outer sum — one bucket for each possible landing column $j$ of row 1.
$a_{1 j}$ :: the row-1 entry, pulled out front (it is common to every term in that bucket).
inner $\sum_{σ (1) = j}$ :: all the ways rows $2, \dots, n$ can still be legally filled, given row 1 is locked to column $j$ .
$\prod_{k = 2}^{n} a_{k, σ (k)}$ :: the product of those remaining rows' entries.

WHY. No term is created or destroyed — we just filed the $n!$ products into $n$ labelled folders. Every original term lands in exactly one folder (the one matching its $j$ ).

PICTURE. Row 1's rook must go somewhere. Each choice of its column paints one folder; the other rooks fill the rest of the board avoiding that column and row 1.

Step 5 — Each folder is a smaller determinant: the minor appears

WHAT. Freeze the row-1 rook in column $j$ . Now delete row 1 and column $j$ . The rooks left over sit in an $(n - 1) \times (n - 1)$ board, filling one per row and one per column of that smaller board — which is exactly the Leibniz sum of the smaller matrix. That smaller determinant is the minor $M_{1 j}$ .

WHY. Once row 1 and column $j$ are gone, the inner sum stops being about the big matrix and becomes a genuine smaller determinant — the recursion we wanted. This is how an $n \times n$ problem shrinks to $(n - 1) \times (n - 1)$ problems.

PICTURE. Grey out row 1 and column $j$ ; the surviving $2 \times 2$ block below-right is the minor.

But there is a leftover twist in the sign, and that twist is the checkerboard. Step 6.

Step 6 — Untwisting the sign gives the checkerboard $(- 1)^{1 + j}$

WHAT. Inside a folder, the sign of the full permutation $σ$ is not simply the sign of the small permutation on the leftover board. To read the small board naturally (its columns in order $1, 2, \dots$ ), we must first slide column $j$ out of the way, hopping it past the $j - 1$ columns to its left. Each hop is one swap → one factor of $- 1$ . Together with row 1's position this bundles into $(- 1)^{1 + j}$ .

one folder σ (1) = j \sum sgn (σ) k \geq 2 \prod a_{k, σ (k)} = (- 1)^{1 + j} M_{1 j} = C_{1 j} .

$(- 1)^{1 + j}$ :: the swap-count to shove entry $(1, j)$ into the top-left corner.
$M_{1 j}$ :: the small determinant from Step 5.
$C_{1 j}$ :: their product — the cofactor, i.e. the signed minor.

WHY. Sliding a column past $k$ other columns is $k$ adjacent swaps, and each swap flips the sign (Step 3). Row 1 is already at the top, contributing the " $1$ " in the exponent. That is the complete origin of the checkerboard $+, -, +, \dots$

PICTURE. The rook in $(1, j)$ marches left, one column-swap at a time, each hop stamping a $\times (- 1)$ , until it reaches the corner. Count the hops = $j - 1$ ; add row 1 = exponent $1 + j$ .

Substituting the folder result back into Step 4:

Step 7 — Why ANY row and ANY column also work

WHAT. We only expanded along row 1. Why is row 3, or column 2, guaranteed to give the same number?

WHY (two moves).

Any row. The Leibniz sum treats all rows symmetrically — nothing marked row 1 as special except our choice to bucket by it first. Bucket by row $i$ instead and the identical argument yields $det A = \sum_{j} a_{ij} C_{ij}$ . The extra swaps to slide row $i$ up to the top add $i - 1$ hops, which is exactly why the exponent is $i + j$ , not $1 + j$ .
Any column. Transposing (flipping rows↔columns) leaves the Leibniz sum untouched, so $det A = det A^{T}$ . A column expansion of $A$ is a row expansion of $A^{T}$ — already proven. Hence every column works too.

PICTURE. The same board, expanded three different ways (row 1, row 3, column 2), all funnel to the identical single number.

Step 8 — Degenerate and edge cases (nothing left uncovered)

WHAT & WHY. A derivation isn't finished until the weird inputs behave.

A zero entry $a_{ij} = 0$ . Its whole term $a_{ij} C_{ij} = 0$ drops out. This is the 80/20 rule: expand along the line with the most zeros to kill terms before you compute them.
A zero row (or zero column). Every $a_{ij}$ on that line is $0$ , so expanding along it gives $det A = 0$ . Geometrically the parallelogram is squashed flat → area $0$ → matrix is singular.
$1 \times 1$ matrix. $det (a) = a$ . The recursion bottoms out here; the $2 \times 2$ formula is really two $1 \times 1$ minors.
Two equal rows. Swapping them changes nothing, yet swapping must flip the sign (Step 3): $det A = - det A \Rightarrow det A = 0$ . Consistent with the collapse picture.

PICTURE. Left: a zero column flattens the parallelogram to a segment (area 0). Right: two equal rows force $det = - det = 0$ .

The one-picture summary

Everything on this page collapses into one flow: big determinant → sort by row 1 → freeze → delete → sign-hop → smaller determinants → recurse.

Recall Feynman retelling (say it to a 12-year-old)

A determinant is one number that says how much a grid stretches a shape. To find it, place "rooks" on the grid so no two share a row or column — every legal placement gives a product of numbers, and each gets a $+$ or $-$ depending on how many times you had to swap columns to tidy them up (even swaps $= +$ , odd $= -$ ). Add all those signed products: that's the determinant. The clever shortcut: fix where the top row's rook goes, then cover that row and column with your hand — what's left is a smaller grid you solve the same way. Covering a column that's, say, 2 steps from the left costs 2 swaps, which is where the checkerboard $+, -, +$ comes from. And because the rooks don't care which row or column you started sorting by, you can sweep along any line — pick the one with the most zeros, do less work, get the same number every time.

Connections

Leibniz formula for determinants — the ground truth we expanded (Steps 2–3).
Properties of determinants (multilinearity, alternation) — the swap-flips-sign fact (Step 3) is alternation.
Determinant via row reduction — the fast cousin that manufactures zeros first.
Adjugate matrix and inverse — packages every cofactor into one matrix.
Cramer's rule — solves $A x = b$ from these same cofactors.
Invertibility and singular matrices — the $det A = 0$ collapse of Step 8.