1. Setup

A system with as many equations as unknowns, viewed as a known matrix transforming an unknown input onto a known output:

Running example.

We assume : each output has exactly one input.


2. A Wrong but Useful First Idea

Observation

and . If dot products survived the transformation, we could get and by dotting the output with the transformed basis vectors (the columns).

Why it fails

Most transformations don’t preserve dot products: vectors with a positive dot product can be pulled apart until it’s negative; perpendicular vectors usually stop being perpendicular; stretching inflates dot products.

Definition 2.1 – Orthonormal transformation

A transformation that preserves dot products: basis vectors stay perpendicular with unit length — rigid motions like rotations, no stretching or squishing.

For orthonormal the shortcut does work:

Example. Rotation by 90°: , → , . Check: ✓.

Lesson: look for a description of coordinates that does behave predictably under the transformation.


3. Coordinates as Signed Areas

Lemma 3.1

  • Parallelogram spanned by and : base 1, height → signed area .
  • Parallelogram spanned by and : signed area .

(Signed in the determinant sense, so negative coordinates give negative areas; the order of the two vectors matters.)

In 3D

The z-coordinate = signed volume of the parallelepiped spanned by (base area 1, height ). Similarly for and : replace the matching basis vector by .


4. Deriving Cramer’s Rule

Key fact

After the transformation, every area scales by the same factor, .

For

The parallelogram becomes . So

For

The parallelogram becomes :

Theorem 4.1 – Cramer’s rule

where is with its -th column replaced by the output vector . Everything uses only known data: the columns of and .


5. Worked Example

UnknownReplaced matrixDeterminantValue

Check: ✓

⚠️ Pitfall: Replace the column, not the row — and only works when .


6. Higher Dimensions

For a 3×3 system: the -coordinate is the volume of the parallelepiped . After transformation it becomes with volume scaled by :

and likewise for and .

MethodSpeedInsight
Gaussian eliminationFastMechanical
Cramer’s ruleSlow for large Connects determinants ↔ systems geometrically

Exercises: Work out the full 3D version of the argument yourself, then solve with Cramer’s rule (lesson page).