Determinants

One Number That Judges the Whole Matrix

The determinant compresses a square matrix into a single number that answers the question: does this matrix destroy information?

The 2×2 case — cross-multiply and subtract:

$$\det\begin{pmatrix} a & b \\ c & d \end{pmatrix} = ad - bc$$

The $3\times 3$ case — the rule of Sarrus: three southeast diagonals minus three northeast diagonals:

$$\det A = aei + bfg + cdh - ceg - bdi - afh$$

What the number tells you:

• $\det A \ne 0$: the matrix is regular (invertible) — an inverse $A^{-1}$ exists, and the system $Ax = b$ has exactly one solution for any $b$.
• $\det A = 0$: the matrix is singular — no inverse; its rows (or columns) are linearly dependent: one row is redundant, a disguised copy of the others.

$$\det A \ne 0 \iff A^{-1} \text{ exists} \iff \text{unique solution}$$

Useful properties (each one a shortcut in real computations):

• Swap two rows → determinant changes sign.
• Multiply one row by $k$ → determinant multiplies by $k$.
• A row of zeros, or two equal rows → determinant is 0.
• $\det(AB) = \det A \cdot \det B$, and $\det A^T = \det A$.

Business reading: your model links choices to outcomes through $A$. If $\det A = 0$, some outcomes are simply unreachable and others have infinitely many explanations — two 'different' policy levers are secretly the same lever. The determinant is the cheapest diagnostic in linear algebra: one number, computed before you waste a week solving an unsolvable model.

What the determinant verdict means

Tip: Before any inverse computation, evaluate the determinant first. If it is zero you have saved yourself the whole calculation; if it is tiny, treat results with suspicion — near-singular systems amplify small data errors enormously.

The Determinant Is an Area

Here is the picture behind the formula. Take the 2×2 matrix's two column vectors and draw the parallelogram they span. Then:

$$|\det A| = \text{area of that parallelogram}$$

The matrix $\begin{pmatrix} 3 & 1 \\ 0 & 2 \end{pmatrix}$ sends the unit square to a parallelogram of area $3\cdot 2 - 1\cdot 0 = 6$: the transformation scales all areas by 6. In three dimensions, the determinant scales volumes. The sign records orientation — a negative determinant means the transformation flips the plane like a page.

Now watch singularity happen. Slide one column toward a multiple of the other. The parallelogram gets thinner… thinner… and at the exact moment the columns align, it collapses to a line segment: area zero.

$$\text{columns aligned} \iff \text{area} = 0 \iff \det = 0$$

That collapse is what 'linearly dependent' means geometrically — and it shows why singular matrices cannot be inverted: the transformation squashed a 2-D plane onto a 1-D line. Two different input points now land on the same output. You cannot un-squash: asked to reverse, the matrix wouldn't know which of the infinitely many preimages to return. Information was destroyed, and $A^{-1}$ would need to recreate it from nothing.

This is also why $\det(AB) = \det A\cdot\det B$ stops being mysterious: apply two transformations in a row and the area scalings simply multiply.

When an economic model's matrix drifts toward singularity, its solution doesn't politely fail — it first becomes hypersensitive (near-zero area, wild rescaling to invert), then non-unique. Watching the parallelogram thin out is watching a model lose its grip, one degree of alignment at a time.

Tip: Carry the picture, not just the formula: the determinant is the area-scaling factor of the transformation. $|\det| = 6$ means every region comes out 6 times larger; $\det < 0$ means the plane was flipped; $\det = 0$ means everything got flattened onto a line — information destroyed.

Common pitfall: Forgetting the absolute value. The parallelogram's area is $|\det A|$; the sign is separate information about orientation, not a negative area.