Matrix Algebra: Determinant and Trace of a Matrix Explained

If you’ve ever delved into economics, data science, or engineering, you’ve likely encountered matrices. They are the spreadsheets of linear algebra, the mathematical language we use to organize and manipulate complex systems of information. But a matrix is more than just a box of numbers. It has a story to tell. Two of the most important numbers that reveal its character are the determinant and the trace. These two values, each calculated in its own way, give us profound insights into the matrix’s power, its stability, and its fundamental properties.

Understanding the determinant: A matrix’s “quality score” 🔗

Imagine you have a system of linear equations. In econometrics, this could be a model describing the relationship between interest rates, inflation, and GDP growth. The coefficients of this model can be neatly arranged into a square matrix. Now, you need to know: does this system have a clear, unique solution? Is it a reliable model?

The determinant is a single scalar value (just a plain number) that answers this question. Think of it as a “quality score” for a square matrix. Its value tells you whether the matrix is well-behaved or if it’s hiding a critical problem. A non-zero determinant is a green light, while a zero determinant is a flashing red warning sign. But before we get into what it means, we have to figure out how to find this special number.

Calculating the determinant using cofactors 🔗

For a tiny 2×2 matrix, the calculation is simple. For a matrix A:

 A = | a b | | c d | 

The determinant is `det(A) = ad – bc`. Easy enough.

But for a 3×3, 4×4, or larger matrix, we need a more systematic approach. This is where cofactor expansion comes in. It’s a recursive method, which means we break a big problem down into smaller, identical problems until we get to a 2×2 matrix we know how to solve. This process has three key steps: finding minors, converting them to cofactors, and then expanding.

Step 1: Understanding minors 🔗

A “minor” is the determinant of a smaller matrix inside your main matrix. For any element in your matrix, let’s call it `a_ij` (the element in row `i` and column `j`), its minor, `M_ij`, is found by:

1. Deleting row `i`.
2. Deleting column `j`.
3. Calculating the determinant of the smaller matrix that remains.

Let’s use a 3×3 example:

 | 1 2 3 | A = | 0 4 5 | | 1 0 6 | 

Let’s find the minor of the element `a_11` (which is 1). We delete the 1st row and 1st column:

 | - - - | | - 4 5 | | - 0 6 | 

The remaining 2×2 matrix is `| 4 5 |`. Its determinant is `(4 * 6) – (5 * 0) = 24`. So, `M_11 = 24`.

What about the minor of `a_12` (which is 2)? We delete the 1st row and 2nd column:

 | - - - | | 0 - 5 | | 1 - 6 | 

The remaining 2×2 matrix is `| 0 5 |`. Its determinant is `(0 * 6) – (5 * 1) = -5`. So, `M_12 = -5`.

Step 2: From minors to cofactors 🔗

A “cofactor” is just a minor with a sign change. The sign depends on the position of the element. The formula for the cofactor `C_ij` is:

`C_ij = (-1)^(i+j) * M_ij`

This `(-1)^(i+j)` part just creates a “checkerboard” of positive and negative signs across the matrix:

 | + - + | | - + - | | + - + | ...and so on. 

Using our example from before:

• `C_11` = (-1)^(1+1) * M_11 = (+1) * 24 = 24
• `C_12` = (-1)^(1+2) * M_12 = (-1) * (-5) = 5

See? The cofactor `C_12` is 5, while its minor `M_12` was -5. The sign flipped.

Step 3: The cofactor expansion 🔗

Now you have all the pieces. To find the determinant of the entire matrix, you just do this:

1. Pick one row or one column. Any single one will do.
2. Multiply each element in that row/column by its own cofactor.
3. Add up the results.

The formula for expansion along row `i` is: `det(A) = a_i1*C_i1 + a_i2*C_i2 + … + a_in*C_in`

Let’s finish our example by expanding along the 1st row:

`det(A) = a_11*C_11 + a_12*C_12 + a_13*C_13`

We already have `C_11 = 24` and `C_12 = 5`. We just need `C_13`:

• Minor `M_13`: Delete row 1, col 3. We are left with `| 0 4 |`. Determinant = `(0 * 0) – (4 * 1) = -4`.
• Cofactor `C_13`: (-1)^(1+3) * M_13 = (+1) * (-4) = -4.

Now, plug everything in:

det(A) = (1 * 24) + (2 * 5) + (3 * -4) det(A) = 24 + 10 - 12 det(A) = 22

The determinant of our matrix is 22. And because 22 is not zero, we already know something vital: this matrix is “non-singular.”

The big reveal: Singular vs. Non-Singular matrices 🔗

This is the most important practical use of the determinant. It’s a simple binary test: is the determinant zero or not?

Non-singular matrices: The heroes of linear algebra 🔗

A matrix is non-singular if `det(A) ≠ 0`.

This is the “green light” we were talking about. It means the matrix is invertible, which means a matrix `A⁻¹` (A-inverse) exists. Why do we care? In an economic model represented by `Ax = b` (where `A` is your coefficient matrix, `x` is your vector of variables, and `b` is your vector of outcomes), a non-singular `A` means you can find a single, unique solution by doing `x = A⁻¹b`.

This tells you that your model is well-defined. Your variables are linearly independent, meaning one variable isn’t just a simple combination of another (like including “height in feet” and “height in inches” as two separate variables). A non-singular matrix means your puzzle has an answer.

Singular matrices: The plot twist 🔗

A matrix is singular if `det(A) = 0`.

This is the “red flag.” It means the matrix is not invertible (it has no inverse). If your model’s matrix `A` is singular, you cannot find a unique solution. Your system of equations is broken. It either has *no solutions* or *infinitely many solutions*, but never just one.

In econometrics, this is the mathematical basis for the problem of perfect multicollinearity. If you run a regression and the `X’X` matrix (a core part of the calculation) is singular, its determinant is zero. This means at least two of your independent variables are perfectly correlated. Your model can’t tell them apart. For example, if `Variable_1 = 2 * Variable_2`, the model breaks. The determinant being zero is the system’s way of telling you, “I can’t solve this; your inputs are redundant.”

Key properties of the determinant 🔗

Calculating determinants with cofactor expansion can be tedious. Luckily, there are several “rules of the road” or properties that help us understand determinants better and, in some cases, save us a lot of work.

Property 1: The transpose doesn’t change a thing 🔗

The “transpose” of a matrix, written `A’` or `Aᵀ`, is the matrix flipped over its main diagonal (rows become columns and columns become rows). A key property is that this operation has no effect on the determinant.

`det(A) = det(A’)`

This is why we can use cofactor expansion along any *column* just as easily as any *row*. The answer will always be the same.

Property 2: The scalar multiplication rule 🔗

This one is a bit tricky and often trips people up. If you multiply one single row (or one single column) of a matrix `A` by a scalar `k`, the new determinant is `k * det(A)`.

But what if you multiply the *entire* matrix by `k`? If `A` is an `n x n` matrix, you are effectively multiplying `n` different rows by `k`. Therefore, the `k` gets pulled out `n` times:

`det(kA) = kⁿ * det(A)`

For a 3×3 matrix, `det(2A) = 2³ * det(A) = 8 * det(A)`. It’s a common mistake to just pull out a single `k`.

Property 3: The multiplicative property 🔗

This is one of the most elegant properties. The determinant of a product of matrices is simply the product of their individual determinants.

`det(AB) = det(A) * det(B)`

This product rule is incredibly useful. For example, it gives us a simple way to find the determinant of an inverse matrix. We know that `A * A⁻¹ = I` (the identity matrix, which has a determinant of 1). Using the product rule: `det(A * A⁻¹) = det(I)` `det(A) * det(A⁻¹) = 1` `det(A⁻¹) = 1 / det(A)`

This clearly shows, once again, why `det(A)` can’t be zero. If it were, you’d be dividing by zero, which is impossible!

Shifting gears: The simple but powerful trace 🔗

If the determinant is the complex, sometimes tricky-to-calculate “quality score,” the trace is its humble, straightforward cousin. It’s much easier to find, but it’s just as important in many fields, especially in statistics and econometrics.

What is the trace? 🔗

The trace of a square matrix, written `tr(A)`, is simply the sum of the elements on the main diagonal (the one from top-left to bottom-right).

`tr(A) = a₁₁ + a₂₂ + … + aₙₙ`

That’s it. No cofactors, no multiplication, just addition.

For our 3×3 matrix from before:

 | 1 2 3 | A = | 0 4 5 | | 1 0 6 | 

The trace is: `tr(A) = 1 + 4 + 6 = 11`.

Key properties of the trace 🔗

The trace’s power comes from its simple and reliable properties. The most important one is its linearity. This means the trace operation “plays nicely” with addition and scalar multiplication.

• `tr(A + B) = tr(A) + tr(B)`
• `tr(cA) = c * tr(A)`

These properties make it incredibly easy to manipulate equations involving the trace.

Furthermore, like the determinant, the trace doesn’t care about transposition:

• `tr(A) = tr(A’)` (or `tr(Aᵀ)`)

This is obvious, because the main diagonal elements don’t move when you transpose the matrix. These linearity and transpose properties make the trace a fundamental tool. In advanced econometrics and statistics, the trace has a deep connection to the eigenvalues of a matrix (just as the determinant is their product, the trace is their sum). It’s used in goodness-of-fit tests, variance calculations, and analyzing the “degrees of freedom” in a model.

What do you think?

After seeing the ‘red flag’ of a zero determinant, what steps do you think an econometrician or data scientist might take to fix their model? And how does the simplicity of the trace compared to the complexity of the determinant make you think about their different roles in analyzing data?

References

1. https://ms.mcmaster.ca/~thomam17/courses/1b03-1zc3/ws19/1zc3/lectures/monday_part2.pdf
2. https://medium.com/@_monitsharma/computational-linear-algebra-trace-and-determinant-d4c01523cb6c
3. https://www.geeksforgeeks.org/engineering-mathematics/properties-of-determinants-of-matrices/
4. https://www.statlect.com/matrix-algebra/determinant-properties
5. https://en.wikipedia.org/wiki/Trace_(linear_algebra)