Determinants — Area, Volume, and Invertibility

The previous chapter asked: does a system of equations have a solution? A key ingredient in that answer was whether the matrix is invertible — does it have full rank? This chapter gives you a single number that answers both questions at once. That number is the determinant.

By the end of this chapter you will be able to:

• Compute the determinant of a and a matrix.
• Interpret the determinant geometrically as a scaling factor for area or volume.
• Connect to linear dependence and non-invertibility.
• Compute the cross product using a determinant, and use it to find surface normals.

The Stretch Factor Puzzle

Imagine your physics engine applies a transformation matrix to an entire region of space — say, a unit square representing the bounding area around a character's hitbox. After the transformation, how large is that region? Twice the original area? Half? Squished to zero?

You could compute the new corners, measure the resulting shape by hand, and divide. But there is a far faster route: compute the determinant of the transformation matrix, and you get that scaling factor directly. One number, telling you exactly how much the transformation stretches or squishes area.

That is the geometric soul of the determinant: it measures how a matrix scales space.

The 2×2 Determinant — Area of a Parallelogram

Start with the simplest case: a matrix

The determinant of is:

The formula looks arbitrary until you see where it comes from.

From Unit Square to Parallelogram

The columns of are two vectors: and .

These two vectors define a parallelogram — take as one side and as the adjacent side:

        v2
       /
      /
     /______
    /      /
   /______/   <- parallelogram spanned by v1 and v2
  v1

The area of that parallelogram is .[^1]

You can see why from a geometric argument. Place the parallelogram inside a rectangle of width and height . The area of the rectangle is . Subtract the triangles and rectangles that surround the parallelogram but are not part of it. After careful accounting, exactly survives.[^2]

A quick check

Apply this to a rotation matrix .

Rotation does not change area — it merely spins the parallelogram. A determinant of confirms that. ✓

The Sign: Orientation

The formula gives a signed area. When is it negative?

Take a reflection matrix (flip across the -axis). .

The absolute area is — the unit square maps to itself. But the sign flipped because the transformation reverses orientation: what was counterclockwise before is now clockwise.[^2]

Before (CCW):    After reflection (CW):
  v2               v1
  ^                ^
  |                |
  +---> v1         +---> v2 (reflected)

A positive determinant means the transformation preserves the "handedness" of space. A negative determinant means it flips it — like holding a page of text up to a mirror. The magnitude of the determinant tells you the size change; the sign tells you whether the flip happened.

INFO

In 2D, orientation is just "which way is counterclockwise?" In 3D, it is the difference between a right-handed coordinate system (the standard axis setup) and a left-handed one. Your game engine almost certainly uses a specific handedness convention and will break if you accidentally flip it.

A Worked 2×2 Example

Find the area of the parallelogram spanned by and .

Build the matrix with these as columns:

Area square units.

(The sign is positive, confirming that is counterclockwise from .)

The 3×3 Determinant — Volume of a Parallelepiped

Now move to three dimensions. Three vectors span a parallelepiped — a 3D box with parallelogram-shaped faces. The determinant of the matrix whose columns are those vectors equals the signed volume of that shape.[^2]

         +----------+
        /|          /|
       / |         / |
      +----------+   |
      |  |       |   |
      |  +-------|---+
      | /        |  /
      |/         | /
      +----------+
   parallelepiped spanned by three vectors

For a matrix

the determinant is computed by cofactor expansion along the first row:[^3]

Each is the determinant of the submatrix left after deleting row and column . The alternating , , signs follow the checkerboard pattern:[^3]

Why the alternating signs?

Each cofactor . The factor gives the checkerboard. Expanding along any row or any column yields the same result — a property that can be proved by induction but is more usefully accepted as a computational convenience.

Worked 3×3 Example

Let

Expand along the first row:

Compute the three determinants:

Assemble:

The three columns of span a parallelepiped of volume cubic units.

Sarrus's Rule (a memory trick for 3×3)

Write the matrix, copy columns 1 and 2 to the right, then sum three "down-right" diagonals and subtract three "down-left" diagonals. This gives the same result as cofactor expansion but is easier to execute mentally. It only works for .

Determinant = 0 Means Collapse

Here is where determinants and the Chapter 6 ideas of linear independence and rank converge.

Suppose two of your three vectors are parallel — say, the third column is just twice the second. Then your "parallelepiped" is completely flat: a 2D region trapped in 3D space. Its volume is zero. And indeed, when two columns are linearly dependent, the determinant is zero.[^2]

Genuine volume:            Collapsed (coplanar):
    /|                         /
   / |                        /___
  /  |       vs.             /   /
 /___|                      /___/   <- flat, zero volume

More generally:

These three statements are equivalent.[^4] When the columns are linearly dependent, at least one column can be written as a combination of the others — meaning the transformation smashes at least one dimension to zero. Information is destroyed. No inverse can undo that destruction.

This bridges directly to the rank discussion from last chapter: a matrix has precisely when its rank is less than (less than full rank).

Common misconception

A zero determinant does not mean the matrix is "broken" or useless — it just means it is not invertible. A projection matrix intentionally collapses one dimension and has a zero determinant by design. Knowing the determinant is zero tells you exactly what kind of transformation you have.

Row Operations and the Determinant

The three row operations from Gaussian elimination each affect the determinant in a predictable way:[^2]

Row operation	Effect on
Swap two rows	Multiply by
Multiply a row by scalar	Multiply by
Add a multiple of one row to another	No change

The third rule is the geometric key: shearing a parallelogram (sliding one side parallel to the base) does not change its area. The base and height stay the same.

These rules give you a fast route to the determinant for special matrices. If is triangular (all zeros above or below the diagonal), then is just the product of the diagonal entries — no expansion needed.

The Cross Product as a Determinant

The cross product of two 3D vectors and is defined as:[^5]

where , , are the standard basis unit vectors. Expanding along the first row:

This notation is slightly unusual because the first row contains vectors, not scalars — but the mechanical expansion is identical. Treat , , as symbols and collect the coefficients.[^5]

What the Cross Product Gives You

The result has two remarkable properties:[^5]

1. Direction: is perpendicular to both and . The direction follows the right-hand rule: curl your right-hand fingers from toward , and your thumb points in the direction of .

2. Magnitude: , where is the angle between them. This equals the area of the parallelogram spanned by and .

         ^ n = a x b  (perpendicular to the face)
         |
         |
  b _____|
   /    /
  /    /
 /____/
   a

Surface Normals in 3D Graphics

Every triangle in a 3D mesh has a surface normal — a unit vector pointing straight out from the face. Lighting calculations, backface culling, and collision response all need this normal. The cross product computes it directly.[^5]

Given three vertices , , of a triangle, form two edge vectors:

Then:

Normalize to get a unit normal: .

The sign matters here: if your vertices are listed counterclockwise (from the viewer's perspective), points toward the viewer. If they are listed clockwise, it points away. This is why consistent vertex winding order is critical in 3D engines — and why the orientation-flipping property of the determinant is not just a mathematical curiosity but a real source of rendering bugs.

Winding order and backface culling

When a GPU performs backface culling — skipping triangles facing away from the camera — it uses the sign of to decide. Flip your vertices accidentally, flip the cross product sign, and that face disappears. The determinant's orientation property is the root cause.

The Determinant as a Stretch Factor

There is one more property that ties everything together.

When you apply a transformation matrix to any region of space, the volume of that region scales by exactly .[^2]

This holds for any shape — a sphere, a polygon, a cloud of particles. The determinant is a universal stretch factor for volume.

A few instructive values:

| | Meaning | |---|---| | | Collapse — the transformation destroys at least one dimension | | | Space shrinks (area/volume decreases) | | | Area/volume is preserved (e.g., rotations, reflections) | | | Space expands |

Rotations always have (or for improper rotations that include a reflection), which is why they preserve both size and shape. Uniform scaling by factor in dimensions has determinant — doubling lengths in 3D multiplies volume by .

Composition:

When you compose two transformations, their stretch factors multiply:[^1]

If doubles area and triples it, then multiplies area by . This also confirms the inverse rule: must undo 's stretch, so

And if , there is no way to define — which is exactly why a zero-determinant matrix has no inverse.

Chapter Recap

Four ideas to carry forward:

1. The determinant is a signed scaling factor. For a matrix, is the area of the parallelogram formed by the columns. For a matrix, it is the volume of the parallelepiped. The sign encodes orientation.

2. means collapse. Zero determinant, linear dependence, and non-invertibility are three ways of saying the same thing: the transformation crushes at least one dimension to zero, destroying information.

3. The cross product is a determinant. The symbolic determinant with basis vectors in the first row computes a vector perpendicular to two given vectors — the foundation of surface normals in 3D graphics.

4. Determinants compose multiplicatively. — the stretch factors multiply when transformations are composed. Rotations preserve volume (). The identity transformation stretches by exactly .

The next chapter introduces eigenvalues and eigenvectors — the special vectors that a matrix only stretches, never rotates. The eigenvalue of a vector turns out to be intimately connected to the determinant: you find eigenvalues by setting a related determinant to zero.

References

[^1]: "Determinants and Volumes." Interactive Linear Algebra, Georgia Tech. https://textbooks.math.gatech.edu/ila/determinants-volumes.html

[^2]: "Determinants and Geometry." Linear Algebra: A First Course, De Anza College, Mathematics LibreTexts. https://math.libretexts.org/Courses/De_Anza_College/Linear_Algebra:_A_First_Course/03:_Determinants/3.04:_Determinants_and_Geometry

[^3]: "Cofactor Expansions." Interactive Linear Algebra, Georgia Tech. https://textbooks.math.gatech.edu/ila/determinants-cofactors.html

[^4]: "Invertible Matrix." Wikipedia. https://en.wikipedia.org/wiki/Invertible_matrix

[^5]: "Cross Product." Wikipedia. https://en.wikipedia.org/wiki/Cross_product