Solving Systems in Linear Algebra: Elimination vs. Matrix Inverse

In matrix form, $A = \begin{bmatrix}2 & 3 \\ 4 & 6\end{bmatrix}$, $x = \begin{bmatrix}x \\ y\end{bmatrix}$, and $b = \begin{bmatrix}7 \\ 14\end{bmatrix}$: the coefficients fill $A$, the unknowns stack into $x$, the constants stack into $b$.

A general $2 \times 2$ matrix $\begin{bmatrix}p & q \\ r & s\end{bmatrix}$ has an inverse exactly when $ps - qr \neq 0$, and that inverse is $\frac{1}{ps - qr}\begin{bmatrix}s & -q \\ -r & p\end{bmatrix}$; when $ps - qr = 0$ there is none. Everything hangs on that determinant. Here $\det(A) = 2 \times 6 - 3 \times 4 = 0$, and there is no such number as $\frac{1}{0}$. The method does not give a wrong answer. It gives nothing.

Before reading on, say in one sentence why the calculation stopped. In two classroom experiments with school algebra students working through a tutoring program, those who explained worked examples while practising understood more than those who practised alone, and the researchers cautiously suggested — their own reading — that examples containing a wrong answer may help understanding most^[1].

Elimination uses no inverse. Multiply the first equation by $2$ and it becomes $4x + 6y = 14$ — the second equation, exactly. The two are one equation written twice, so nothing pins down a point. Solve the first for $x$:

Every pair $\left(\frac{7 - 3y}{2},\, y\right)$ with real $y$ is a solution, since $2 \cdot \frac{7 - 3y}{2} + 3y = 7 - 3y + 3y = 7$ and the second equation is twice the first. A complete answer here is not a number but that line, plus the check that it satisfies both equations.

The theorem is often quoted circularly: A is invertible when it has an inverse. Its real content is that several different conditions on a square matrix $A$ are one condition.

• $A$ has an inverse.
• $\det(A) \neq 0$.
• The rows of $A$ are linearly independent — no row is a linear combination of the others.
• $Ax = 0$ has only the solution $x = 0$.

Our matrix fails all four for one reason: its second row is twice its first. The rows are dependent, the determinant is zero, and the matrix is singular — $Ax = b$ then has either no solutions or infinitely many, never exactly one.

With more unknowns than equations the matrix is not even square, so $A^{-1}$ is undefined and the solution set can be a plane or larger. Elimination has no such precondition.

"Two equations, two unknowns, a square block of coefficients" is a surface feature. "Is this matrix invertible" is the principle underneath. A classic set of experiments on physics problems found that beginners sorted and approached problems by their surface look, while experts worked from the principle involved^[2]. Those experiments were about physics, not matrices — a general picture of how a beginner picks a method, not evidence about linear algebra.

A widely used framework for describing mathematical reasoning — built from earlier classroom studies, not a new experiment of its own — argues that students most often reproduce a remembered procedure instead of reasoning about the mathematics in front of them^[3]. A remembered procedure never checks its own precondition. The formula $x = A^{-1}b$ gets run like a recipe.

One study looked directly at this method. Among 116 practising mathematics teachers taking a first linear algebra course at a university in Zimbabwe, only a small number had a settled sense of what a determinant and an inverse are, as opposed to how to compute them; that gap held them back on the inverse method^[4].

The shape of this post is deliberate. In one experiment, 70 seventh-graders were split in two: one group studied two ways of solving the same equation side by side, the other studied the same methods one after the other. The side-by-side group ended up better at solving equations and at picking a sensible method. On grasp of the underlying ideas, the two groups came out about the same^[5].

That last part matters: side-by-side study did not build more understanding than studying the methods one at a time^[5]. And those were seventh-graders solving linear equations, not students doing linear algebra. What carries over is the arrangement — both methods on one system — not the finding.

For a small square system with a nonzero determinant, the inverse is quick and exact. Take $x + y = 3$ and $2x - y = 0$, where $\det(A) = (1)(-1) - (1)(2) = -3$, so $A^{-1}b = \frac{1}{-3}\begin{bmatrix}-1 & -1 \\ -2 & 1\end{bmatrix}\begin{bmatrix}3 \\ 0\end{bmatrix} = \frac{1}{-3}\begin{bmatrix}-3 \\ -6\end{bmatrix} = \begin{bmatrix}1 \\ 2\end{bmatrix}$. Substituting into both originals: $1 + 2 = 3$ and $2(1) - 2 = 0$. The inverse also pays off when the same $A$ meets several different $b$: build it once, reuse it. Otherwise elimination is more general: it takes non-square systems, and separates no solution from infinitely many rather than stalling.

Attempt it yourself first. In one study comparing two kinds of practice, students who worked out their own path through the tasks did better on the test afterwards than students handed a ready-made procedure^[6]. It came from the same research group as the framework above — one line of work, not two confirmations.

Build $A$, $x$ and $b$ as above, write $Ax = b$, and solve it both ways.

Here $\det(A) = 2 \times 3 - (-1) \times 5 = 6 + 5 = 11$, which is not zero, so both routes are open:

By elimination, the first equation gives $y = 2x - 1$, so $5x + 3(2x - 1) = 19$, then $11x = 22$, $x = 2$ and $y = 3$. Substituting into both originals: $2(2) - 3 = 1$ and $5(2) + 3(3) = 19$. Both routes agree.

Almost none of this evidence is about the mistake above. The one study here that watched anyone use the inverse method followed 116 practising mathematics teachers in a first linear algebra course at a university in Zimbabwe — not high-school students, not first-year undergraduates^[4]. It reported that a weak grip on determinants and inverses held them back, not that they could not finish^[4].

The rest is background from other subjects and other ages: sorting physics problems^[2], seventh-graders solving linear equations^[5], school algebra students inside a tutoring program^[1], and a reasoning framework with its companion practice study^[3][6]. Not one of those studies watched a person work with a matrix.

On the reflex this post is about — reaching for $A^{-1}$ when the determinant is zero — I could find no study measuring how often it happens or what stops it. The comparison study is also more modest than it sounds: that group gained more on procedure and method choice, but no more than the other group on understanding the underlying ideas^[5]. Take it as a reason to arrange your practice this way, not as proof that the arrangement teaches linear algebra.