Interactive Essay

The Determinant: How Much of the Plane Survived?

I spent more time than I'd like to admit computing ad - bc without knowing what the number meant. It's the signed area of a parallelogram. Once I saw that, everything else in this topic stopped feeling like a list of rules to memorize.

I like starting with shear because it looks dramatic and still keeps determinant 1.

The whole grid slants. The square becomes a parallelogram. But each little patch still takes up the same amount of room. Then watch what happens when orientation flips.

sheardet = 1 — the grid leans, area unchanged· click any handle to take control

Determinant view

Drag the red and blue handles. The whole grid moves with them.

det = 1

Area scale

1×

Orientation

preserved

State

invertible

area preserved, orientation preserved

original grid

transformed grid

u basis vector

v basis vector

orientation triangle

Matrix A

A =

Presets

If you have not gone through the matrix essay yet, start with What Does a Matrix Actually Do?. This post picks up from there. A matrix moves the basis vectors and the entire grid follows, so once you have that picture the question of how much of the plane survived becomes something you can actually ask.

The answer is $\left| \det(A) \right| = \text{area scale factor}$ . The magnitude is how much area changed. The sign records whether the transformed square kept its original orientation or ended up flipped through a mirror somewhere in the process.

Four behaviors, one number

ad - bc is a perfectly good formula, but it tells you how to compute the determinant, not what it is. I would rather start with the square sitting inside a visible grid. Apply a matrix and the square leans, or spreads, or comes back mirrored, or gets crushed flat. Every other little patch in the plane does the same thing at the same time.

From that picture, ad - bc reads as the area of the parallelogram the two column vectors span. The algebra and the geometry are describing the same thing.

Determinant view

The gray grid is the old plane. The dark grid is the plane after A.

det = 1

Area scale

1×

Orientation

preserved

State

invertible

area preserved, orientation preserved

original grid

transformed grid

u basis vector

v basis vector

orientation triangle

Shear

det = 1.00

Shear slants the whole grid. The square leans, but every small patch keeps the same area.

Try This

Find a matrix with determinant 1 that looks nothing like the identity. Then find one with determinant -1 that changes very little except the orientation.

When the determinant hits zero

Nonzero determinants, positive or negative, still leave you with a full plane — just scaled and possibly flipped. Zero is a different case entirely. When the two column vectors become parallel, the parallelogram collapses into a line segment and all the area is gone. At that point the transformation is squashing the entire plane into a single line, which means multiple input points land on the same output and there is no inverse.

Move the slider and watch the columns align. The parallelogram thins out gradually and the inverse equation starts blowing up at the same rate. They fail together for the same reason.

\det(A) = 0 \iff A \text{ is singular}

Determinant view

The gray grid is the old plane. The dark grid is the plane after A.

det = 0.58

Area scale

0.58×

Orientation

preserved

State

invertible

area survives, orientation preserved

original grid

transformed grid

u basis vector

v basis vector

orientation triangle

Align The Columns

invertible

determinant = 0.576. As the columns line up, the square gets thinner. At zero, different inputs land on the same output line.

Why the formula is ad - bc

Algebraically, the determinant of a 2×2 matrix is simply ad - bc. Take a concrete example:

\det\!\left(\begin{bmatrix}3 & 1 \\ 2 & 4\end{bmatrix}\right) = (3 \cdot 4) - (1 \cdot 2) = 10

That is the whole calculation. Where it gets interesting is understanding what that number is actually counting.

The columns of the matrix are the two edge vectors of the transformed unit square. Call them u = (a, c) and v = (b, d). Together they form a parallelogram, and the area of that parallelogram is exactly ad - bc. The derivation comes straight from subtracting the surrounding triangles and rectangles from a big bounding box:

\underbrace{(a+b)(c+d)}_{\text{big rectangle}} - \underbrace{2 \cdot \tfrac{1}{2}ac}_{\text{top/bottom triangles}} - \underbrace{2 \cdot \tfrac{1}{2}bd}_{\text{side triangles}} - \underbrace{2bc}_{\text{corner rectangles}} = ad - bc

Everything cancels except ad - bc. The formula is the area of the parallelogram that the two column vectors span, computed by subtracting the surrounding pieces. The interactive below lets you drag those edge vectors and watch it update.

Area Derivation

Drag the handles. The bounding box minus the six colored pieces leaves the parallelogram.

big rectangle

(a+b)(c+d)

8.12

ac triangles ×2

2.10

bd triangles ×2

1.33

bc rectangles ×2

2bc

1.40

8.12−2.10−1.33−1.40=3.29= ad − bc

u = (2.1, 1.0)v = (0.7, 1.9)

Same calculation, but from the algebra side. You can drag the vectors and watch the formula fill in with actual numbers.

Determinant view

Drag the red and blue handles. The whole grid moves with them.

det = 1.42

Area scale

1.42×

Orientation

preserved

State

invertible

area survives, orientation preserved

original grid

transformed grid

u basis vector

v basis vector

orientation triangle

Matrix A

A =

\det(A) = ad - bc

Start with the columns. They are the two edges of the transformed unit square.

What else the determinant tells you

Once you have the geometric picture, the algebraic rules mostly follow. Multiplying two matrices multiplies their area scales, which is why det(AB) = det(A) · det(B). Inverses need a nonzero determinant because you cannot undo a transformation that already flattened the plane. In ML and probability, the same idea shows up in change of variables: if a transformation stretches the space, the same amount of probability is now spread over a larger area, so the density goes down.

\det(AB) = \det(A)\det(B)

The dots in the panel below move with the grid. That is essentially what a Jacobian tracks when you change variables in an integral.

Matrix A

Shear

Matrix B

Stretch

Compose A with B

det(A)

det(B)

1.85

det(active)

density scale

One dimension up

In three dimensions you are measuring volume instead of area, but the reading is the same. A 3×3 matrix maps a unit cube to a parallelepiped, and the determinant is the signed volume of that parallelepiped. The sign still flips when orientation reverses, and zero still means the transformation lost a dimension.

The formula you will usually compute by hand is the cofactor expansion along the first row:

\det\!\begin{pmatrix}a&b&c\\d&e&f\\g&h&i\end{pmatrix} = a(ei-fh) - b(di-fg) + c(dh-eg)

Each term picks one entry from the first row, multiplies it by the 2×2 determinant of the remaining rows and columns after removing that entry's row and column, then alternates sign. The pattern + − + comes from orientation: each minor is measuring a sub-face of the parallelepiped.

There is a more compact way to write the same thing using the triple product of the three column vectors:

\det(A) = u \cdot (v \times w)

This works because v × w gives a vector perpendicular to the base face with magnitude equal to the base area. Dotting with u then projects the height onto that direction, so you get base area times height — which is volume. The sign comes from whether u points the same direction as v × w (positive) or opposite (negative). Both formulas compute the same number; the triple product just makes the geometry explicit.

You can also read the result as $\left| \det(A) \right| = \text{volume scale factor}$ . Try dragging one column until it falls into the span of the other two. The box does not need every edge to shrink to zero — it only needs to lose one genuine 3D direction.

3D Determinant View

Drag the basis vectors and watch the cube lose volume.

The pale lattice is the old space. The darker lattice is the transformed space. Orbit the view, then drag one colored tip into the plane of the other two and watch the box lose thickness.

det = +1

xyz

xy spanxz spanyz span

Drag tips · Orbit camera

Volume Scale

1×

absolute determinant

Orientation

preserved

same handedness

State

invertible

volume still survives

Matrix Editor

These columns are the transformed basis vectors. Edit the entries directly or use the presets to jump to a useful case.

A =

What to try

Make the determinant negative without changing the cube size too much.
Keep all three columns nonzero, but drag one into the span of the other two until the box goes flat.
Find two very different-looking boxes that still have determinant 1.

Determinant laboratory

The last panel is a scratchpad. Try building three matrices with determinant 1 that look nothing alike, then see how close to zero you can get without the plane actually collapsing.

Determinant is a useful number but it is a narrow one. It says nothing about which direction the shear went, or what the eigenvalues are, or whether the matrix is close to orthogonal. It is one measurement out of several that matter, which is why inverses, eigenvectors, and SVD all end up in the same course.

Determinant view

Drag the red and blue handles. The whole grid moves with them.

det = 1

Area scale

1×

Orientation

preserved

State

invertible

area preserved, orientation preserved

original grid

transformed grid

u basis vector

v basis vector

orientation triangle

Matrix A

A =

Presets

Apply Another Matrix

Apply B after A

B =

Things to try

Make three very different matrices with determinant 1.
Get the determinant very close to 0 without actually hitting 0.
Build a matrix with negative determinant and area scale 2.

Current Readout

determinant = 1

Save the interesting cases here. It helps to compare matrices that share the same determinant.

Summary

The determinant is the signed area scale factor of a linear transformation. Magnitude tells you how much area changed, the sign tells you whether orientation flipped, and zero means the plane collapsed into something lower-dimensional. That covers a lot of ground for one number.

In 3D the same idea applies to volume. The determinant of a 3×3 matrix measures how much a unit cube changes in the process. ad - bc and its 3D analogue are formulas for those volumes, which is why they are worth computing rather than just memorizing.

Quick check

Five questions.

1.A transformation has determinant −3. What does that tell you?

2.The two column vectors of a matrix become parallel. What happens?

3.det(A) = 3 and det(B) = 4. What is det(B · A)?

4.The columns of a matrix are u = (2, 1) and v = (0, 3). What is the area of the parallelogram they span?

5.A 3×3 matrix has determinant 0. What does that mean geometrically?